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_transport.h> 37 #include <scsi/scsi_host.h> 38 #include <scsi/scsi_ioctl.h> 39 40 #include "scsi_priv.h" 41 #include "scsi_logging.h" 42 #include "scsi_transport_api.h" 43 44 #include <trace/events/scsi.h> 45 46 static void scsi_eh_done(struct scsi_cmnd *scmd); 47 48 /* 49 * These should *probably* be handled by the host itself. 50 * Since it is allowed to sleep, it probably should. 51 */ 52 #define BUS_RESET_SETTLE_TIME (10) 53 #define HOST_RESET_SETTLE_TIME (10) 54 55 static int scsi_eh_try_stu(struct scsi_cmnd *scmd); 56 static int scsi_try_to_abort_cmd(struct scsi_host_template *, 57 struct scsi_cmnd *); 58 59 /* called with shost->host_lock held */ 60 void scsi_eh_wakeup(struct Scsi_Host *shost) 61 { 62 if (atomic_read(&shost->host_busy) == shost->host_failed) { 63 trace_scsi_eh_wakeup(shost); 64 wake_up_process(shost->ehandler); 65 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost, 66 "Waking error handler thread\n")); 67 } 68 } 69 70 /** 71 * scsi_schedule_eh - schedule EH for SCSI host 72 * @shost: SCSI host to invoke error handling on. 73 * 74 * Schedule SCSI EH without scmd. 75 */ 76 void scsi_schedule_eh(struct Scsi_Host *shost) 77 { 78 unsigned long flags; 79 80 spin_lock_irqsave(shost->host_lock, flags); 81 82 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || 83 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { 84 shost->host_eh_scheduled++; 85 scsi_eh_wakeup(shost); 86 } 87 88 spin_unlock_irqrestore(shost->host_lock, flags); 89 } 90 EXPORT_SYMBOL_GPL(scsi_schedule_eh); 91 92 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost) 93 { 94 if (!shost->last_reset || shost->eh_deadline == -1) 95 return 0; 96 97 /* 98 * 32bit accesses are guaranteed to be atomic 99 * (on all supported architectures), so instead 100 * of using a spinlock we can as well double check 101 * if eh_deadline has been set to 'off' during the 102 * time_before call. 103 */ 104 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) && 105 shost->eh_deadline > -1) 106 return 0; 107 108 return 1; 109 } 110 111 /** 112 * scmd_eh_abort_handler - Handle command aborts 113 * @work: command to be aborted. 114 */ 115 void 116 scmd_eh_abort_handler(struct work_struct *work) 117 { 118 struct scsi_cmnd *scmd = 119 container_of(work, struct scsi_cmnd, abort_work.work); 120 struct scsi_device *sdev = scmd->device; 121 int rtn; 122 123 if (scsi_host_eh_past_deadline(sdev->host)) { 124 SCSI_LOG_ERROR_RECOVERY(3, 125 scmd_printk(KERN_INFO, scmd, 126 "scmd %p eh timeout, not aborting\n", 127 scmd)); 128 } else { 129 SCSI_LOG_ERROR_RECOVERY(3, 130 scmd_printk(KERN_INFO, scmd, 131 "aborting command %p\n", scmd)); 132 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd); 133 if (rtn == SUCCESS) { 134 set_host_byte(scmd, DID_TIME_OUT); 135 if (scsi_host_eh_past_deadline(sdev->host)) { 136 SCSI_LOG_ERROR_RECOVERY(3, 137 scmd_printk(KERN_INFO, scmd, 138 "scmd %p eh timeout, " 139 "not retrying aborted " 140 "command\n", scmd)); 141 } else if (!scsi_noretry_cmd(scmd) && 142 (++scmd->retries <= scmd->allowed)) { 143 SCSI_LOG_ERROR_RECOVERY(3, 144 scmd_printk(KERN_WARNING, scmd, 145 "scmd %p retry " 146 "aborted command\n", scmd)); 147 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 148 return; 149 } else { 150 SCSI_LOG_ERROR_RECOVERY(3, 151 scmd_printk(KERN_WARNING, scmd, 152 "scmd %p finish " 153 "aborted command\n", scmd)); 154 scsi_finish_command(scmd); 155 return; 156 } 157 } else { 158 SCSI_LOG_ERROR_RECOVERY(3, 159 scmd_printk(KERN_INFO, scmd, 160 "scmd %p abort failed, rtn %d\n", 161 scmd, rtn)); 162 } 163 } 164 165 if (!scsi_eh_scmd_add(scmd, 0)) { 166 SCSI_LOG_ERROR_RECOVERY(3, 167 scmd_printk(KERN_WARNING, scmd, 168 "scmd %p terminate " 169 "aborted command\n", scmd)); 170 set_host_byte(scmd, DID_TIME_OUT); 171 scsi_finish_command(scmd); 172 } 173 } 174 175 /** 176 * scsi_abort_command - schedule a command abort 177 * @scmd: scmd to abort. 178 * 179 * We only need to abort commands after a command timeout 180 */ 181 static int 182 scsi_abort_command(struct scsi_cmnd *scmd) 183 { 184 struct scsi_device *sdev = scmd->device; 185 struct Scsi_Host *shost = sdev->host; 186 unsigned long flags; 187 188 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 189 /* 190 * Retry after abort failed, escalate to next level. 191 */ 192 scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED; 193 SCSI_LOG_ERROR_RECOVERY(3, 194 scmd_printk(KERN_INFO, scmd, 195 "scmd %p previous abort failed\n", scmd)); 196 BUG_ON(delayed_work_pending(&scmd->abort_work)); 197 return FAILED; 198 } 199 200 /* 201 * Do not try a command abort if 202 * SCSI EH has already started. 203 */ 204 spin_lock_irqsave(shost->host_lock, flags); 205 if (scsi_host_in_recovery(shost)) { 206 spin_unlock_irqrestore(shost->host_lock, flags); 207 SCSI_LOG_ERROR_RECOVERY(3, 208 scmd_printk(KERN_INFO, scmd, 209 "scmd %p not aborting, host in recovery\n", 210 scmd)); 211 return FAILED; 212 } 213 214 if (shost->eh_deadline != -1 && !shost->last_reset) 215 shost->last_reset = jiffies; 216 spin_unlock_irqrestore(shost->host_lock, flags); 217 218 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; 219 SCSI_LOG_ERROR_RECOVERY(3, 220 scmd_printk(KERN_INFO, scmd, 221 "scmd %p abort scheduled\n", scmd)); 222 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100); 223 return SUCCESS; 224 } 225 226 /** 227 * scsi_eh_scmd_add - add scsi cmd to error handling. 228 * @scmd: scmd to run eh on. 229 * @eh_flag: optional SCSI_EH flag. 230 * 231 * Return value: 232 * 0 on failure. 233 */ 234 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag) 235 { 236 struct Scsi_Host *shost = scmd->device->host; 237 unsigned long flags; 238 int ret = 0; 239 240 if (!shost->ehandler) 241 return 0; 242 243 spin_lock_irqsave(shost->host_lock, flags); 244 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 245 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)) 246 goto out_unlock; 247 248 if (shost->eh_deadline != -1 && !shost->last_reset) 249 shost->last_reset = jiffies; 250 251 ret = 1; 252 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) 253 eh_flag &= ~SCSI_EH_CANCEL_CMD; 254 scmd->eh_eflags |= eh_flag; 255 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 256 shost->host_failed++; 257 scsi_eh_wakeup(shost); 258 out_unlock: 259 spin_unlock_irqrestore(shost->host_lock, flags); 260 return ret; 261 } 262 263 /** 264 * scsi_times_out - Timeout function for normal scsi commands. 265 * @req: request that is timing out. 266 * 267 * Notes: 268 * We do not need to lock this. There is the potential for a race 269 * only in that the normal completion handling might run, but if the 270 * normal completion function determines that the timer has already 271 * fired, then it mustn't do anything. 272 */ 273 enum blk_eh_timer_return scsi_times_out(struct request *req) 274 { 275 struct scsi_cmnd *scmd = req->special; 276 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED; 277 struct Scsi_Host *host = scmd->device->host; 278 279 trace_scsi_dispatch_cmd_timeout(scmd); 280 scsi_log_completion(scmd, TIMEOUT_ERROR); 281 282 if (host->eh_deadline != -1 && !host->last_reset) 283 host->last_reset = jiffies; 284 285 if (host->transportt->eh_timed_out) 286 rtn = host->transportt->eh_timed_out(scmd); 287 else if (host->hostt->eh_timed_out) 288 rtn = host->hostt->eh_timed_out(scmd); 289 290 if (rtn == BLK_EH_NOT_HANDLED) { 291 if (!host->hostt->no_async_abort && 292 scsi_abort_command(scmd) == SUCCESS) 293 return BLK_EH_NOT_HANDLED; 294 295 set_host_byte(scmd, DID_TIME_OUT); 296 if (!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)) 297 rtn = BLK_EH_HANDLED; 298 } 299 300 return rtn; 301 } 302 303 /** 304 * scsi_block_when_processing_errors - Prevent cmds from being queued. 305 * @sdev: Device on which we are performing recovery. 306 * 307 * Description: 308 * We block until the host is out of error recovery, and then check to 309 * see whether the host or the device is offline. 310 * 311 * Return value: 312 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 313 */ 314 int scsi_block_when_processing_errors(struct scsi_device *sdev) 315 { 316 int online; 317 318 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 319 320 online = scsi_device_online(sdev); 321 322 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev, 323 "%s: rtn: %d\n", __func__, online)); 324 325 return online; 326 } 327 EXPORT_SYMBOL(scsi_block_when_processing_errors); 328 329 #ifdef CONFIG_SCSI_LOGGING 330 /** 331 * scsi_eh_prt_fail_stats - Log info on failures. 332 * @shost: scsi host being recovered. 333 * @work_q: Queue of scsi cmds to process. 334 */ 335 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 336 struct list_head *work_q) 337 { 338 struct scsi_cmnd *scmd; 339 struct scsi_device *sdev; 340 int total_failures = 0; 341 int cmd_failed = 0; 342 int cmd_cancel = 0; 343 int devices_failed = 0; 344 345 shost_for_each_device(sdev, shost) { 346 list_for_each_entry(scmd, work_q, eh_entry) { 347 if (scmd->device == sdev) { 348 ++total_failures; 349 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) 350 ++cmd_cancel; 351 else 352 ++cmd_failed; 353 } 354 } 355 356 if (cmd_cancel || cmd_failed) { 357 SCSI_LOG_ERROR_RECOVERY(3, 358 sdev_printk(KERN_INFO, sdev, 359 "%s: cmds failed: %d, cancel: %d\n", 360 __func__, cmd_failed, 361 cmd_cancel)); 362 cmd_cancel = 0; 363 cmd_failed = 0; 364 ++devices_failed; 365 } 366 } 367 368 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, 369 "Total of %d commands on %d" 370 " devices require eh work\n", 371 total_failures, devices_failed)); 372 } 373 #endif 374 375 /** 376 * scsi_report_lun_change - Set flag on all *other* devices on the same target 377 * to indicate that a UNIT ATTENTION is expected. 378 * @sdev: Device reporting the UNIT ATTENTION 379 */ 380 static void scsi_report_lun_change(struct scsi_device *sdev) 381 { 382 sdev->sdev_target->expecting_lun_change = 1; 383 } 384 385 /** 386 * scsi_report_sense - Examine scsi sense information and log messages for 387 * certain conditions, also issue uevents for some of them. 388 * @sdev: Device reporting the sense code 389 * @sshdr: sshdr to be examined 390 */ 391 static void scsi_report_sense(struct scsi_device *sdev, 392 struct scsi_sense_hdr *sshdr) 393 { 394 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ 395 396 if (sshdr->sense_key == UNIT_ATTENTION) { 397 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { 398 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; 399 sdev_printk(KERN_WARNING, sdev, 400 "Inquiry data has changed"); 401 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { 402 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; 403 scsi_report_lun_change(sdev); 404 sdev_printk(KERN_WARNING, sdev, 405 "Warning! Received an indication that the " 406 "LUN assignments on this target have " 407 "changed. The Linux SCSI layer does not " 408 "automatically remap LUN assignments.\n"); 409 } else if (sshdr->asc == 0x3f) 410 sdev_printk(KERN_WARNING, sdev, 411 "Warning! Received an indication that the " 412 "operating parameters on this target have " 413 "changed. The Linux SCSI layer does not " 414 "automatically adjust these parameters.\n"); 415 416 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { 417 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; 418 sdev_printk(KERN_WARNING, sdev, 419 "Warning! Received an indication that the " 420 "LUN reached a thin provisioning soft " 421 "threshold.\n"); 422 } 423 424 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { 425 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; 426 sdev_printk(KERN_WARNING, sdev, 427 "Mode parameters changed"); 428 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { 429 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; 430 sdev_printk(KERN_WARNING, sdev, 431 "Capacity data has changed"); 432 } else if (sshdr->asc == 0x2a) 433 sdev_printk(KERN_WARNING, sdev, 434 "Parameters changed"); 435 } 436 437 if (evt_type != SDEV_EVT_MAXBITS) { 438 set_bit(evt_type, sdev->pending_events); 439 schedule_work(&sdev->event_work); 440 } 441 } 442 443 /** 444 * scsi_check_sense - Examine scsi cmd sense 445 * @scmd: Cmd to have sense checked. 446 * 447 * Return value: 448 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE 449 * 450 * Notes: 451 * When a deferred error is detected the current command has 452 * not been executed and needs retrying. 453 */ 454 static int scsi_check_sense(struct scsi_cmnd *scmd) 455 { 456 struct scsi_device *sdev = scmd->device; 457 struct scsi_sense_hdr sshdr; 458 459 if (! scsi_command_normalize_sense(scmd, &sshdr)) 460 return FAILED; /* no valid sense data */ 461 462 scsi_report_sense(sdev, &sshdr); 463 464 if (scsi_sense_is_deferred(&sshdr)) 465 return NEEDS_RETRY; 466 467 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh && 468 sdev->scsi_dh_data->scsi_dh->check_sense) { 469 int rc; 470 471 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr); 472 if (rc != SCSI_RETURN_NOT_HANDLED) 473 return rc; 474 /* handler does not care. Drop down to default handling */ 475 } 476 477 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done) 478 /* 479 * nasty: for mid-layer issued TURs, we need to return the 480 * actual sense data without any recovery attempt. For eh 481 * issued ones, we need to try to recover and interpret 482 */ 483 return SUCCESS; 484 485 /* 486 * Previous logic looked for FILEMARK, EOM or ILI which are 487 * mainly associated with tapes and returned SUCCESS. 488 */ 489 if (sshdr.response_code == 0x70) { 490 /* fixed format */ 491 if (scmd->sense_buffer[2] & 0xe0) 492 return SUCCESS; 493 } else { 494 /* 495 * descriptor format: look for "stream commands sense data 496 * descriptor" (see SSC-3). Assume single sense data 497 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 498 */ 499 if ((sshdr.additional_length > 3) && 500 (scmd->sense_buffer[8] == 0x4) && 501 (scmd->sense_buffer[11] & 0xe0)) 502 return SUCCESS; 503 } 504 505 switch (sshdr.sense_key) { 506 case NO_SENSE: 507 return SUCCESS; 508 case RECOVERED_ERROR: 509 return /* soft_error */ SUCCESS; 510 511 case ABORTED_COMMAND: 512 if (sshdr.asc == 0x10) /* DIF */ 513 return SUCCESS; 514 515 return NEEDS_RETRY; 516 case NOT_READY: 517 case UNIT_ATTENTION: 518 /* 519 * if we are expecting a cc/ua because of a bus reset that we 520 * performed, treat this just as a retry. otherwise this is 521 * information that we should pass up to the upper-level driver 522 * so that we can deal with it there. 523 */ 524 if (scmd->device->expecting_cc_ua) { 525 /* 526 * Because some device does not queue unit 527 * attentions correctly, we carefully check 528 * additional sense code and qualifier so as 529 * not to squash media change unit attention. 530 */ 531 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { 532 scmd->device->expecting_cc_ua = 0; 533 return NEEDS_RETRY; 534 } 535 } 536 /* 537 * we might also expect a cc/ua if another LUN on the target 538 * reported a UA with an ASC/ASCQ of 3F 0E - 539 * REPORTED LUNS DATA HAS CHANGED. 540 */ 541 if (scmd->device->sdev_target->expecting_lun_change && 542 sshdr.asc == 0x3f && sshdr.ascq == 0x0e) 543 return NEEDS_RETRY; 544 /* 545 * if the device is in the process of becoming ready, we 546 * should retry. 547 */ 548 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 549 return NEEDS_RETRY; 550 /* 551 * if the device is not started, we need to wake 552 * the error handler to start the motor 553 */ 554 if (scmd->device->allow_restart && 555 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 556 return FAILED; 557 /* 558 * Pass the UA upwards for a determination in the completion 559 * functions. 560 */ 561 return SUCCESS; 562 563 /* these are not supported */ 564 case DATA_PROTECT: 565 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { 566 /* Thin provisioning hard threshold reached */ 567 set_host_byte(scmd, DID_ALLOC_FAILURE); 568 return SUCCESS; 569 } 570 case COPY_ABORTED: 571 case VOLUME_OVERFLOW: 572 case MISCOMPARE: 573 case BLANK_CHECK: 574 set_host_byte(scmd, DID_TARGET_FAILURE); 575 return SUCCESS; 576 577 case MEDIUM_ERROR: 578 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 579 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 580 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 581 set_host_byte(scmd, DID_MEDIUM_ERROR); 582 return SUCCESS; 583 } 584 return NEEDS_RETRY; 585 586 case HARDWARE_ERROR: 587 if (scmd->device->retry_hwerror) 588 return ADD_TO_MLQUEUE; 589 else 590 set_host_byte(scmd, DID_TARGET_FAILURE); 591 592 case ILLEGAL_REQUEST: 593 if (sshdr.asc == 0x20 || /* Invalid command operation code */ 594 sshdr.asc == 0x21 || /* Logical block address out of range */ 595 sshdr.asc == 0x24 || /* Invalid field in cdb */ 596 sshdr.asc == 0x26) { /* Parameter value invalid */ 597 set_host_byte(scmd, DID_TARGET_FAILURE); 598 } 599 return SUCCESS; 600 601 default: 602 return SUCCESS; 603 } 604 } 605 606 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) 607 { 608 struct scsi_host_template *sht = sdev->host->hostt; 609 struct scsi_device *tmp_sdev; 610 611 if (!sht->change_queue_depth || 612 sdev->queue_depth >= sdev->max_queue_depth) 613 return; 614 615 if (time_before(jiffies, 616 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) 617 return; 618 619 if (time_before(jiffies, 620 sdev->last_queue_full_time + sdev->queue_ramp_up_period)) 621 return; 622 623 /* 624 * Walk all devices of a target and do 625 * ramp up on them. 626 */ 627 shost_for_each_device(tmp_sdev, sdev->host) { 628 if (tmp_sdev->channel != sdev->channel || 629 tmp_sdev->id != sdev->id || 630 tmp_sdev->queue_depth == sdev->max_queue_depth) 631 continue; 632 /* 633 * call back into LLD to increase queue_depth by one 634 * with ramp up reason code. 635 */ 636 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1, 637 SCSI_QDEPTH_RAMP_UP); 638 sdev->last_queue_ramp_up = jiffies; 639 } 640 } 641 642 static void scsi_handle_queue_full(struct scsi_device *sdev) 643 { 644 struct scsi_host_template *sht = sdev->host->hostt; 645 struct scsi_device *tmp_sdev; 646 647 if (!sht->change_queue_depth) 648 return; 649 650 shost_for_each_device(tmp_sdev, sdev->host) { 651 if (tmp_sdev->channel != sdev->channel || 652 tmp_sdev->id != sdev->id) 653 continue; 654 /* 655 * We do not know the number of commands that were at 656 * the device when we got the queue full so we start 657 * from the highest possible value and work our way down. 658 */ 659 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1, 660 SCSI_QDEPTH_QFULL); 661 } 662 } 663 664 /** 665 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 666 * @scmd: SCSI cmd to examine. 667 * 668 * Notes: 669 * This is *only* called when we are examining the status of commands 670 * queued during error recovery. the main difference here is that we 671 * don't allow for the possibility of retries here, and we are a lot 672 * more restrictive about what we consider acceptable. 673 */ 674 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) 675 { 676 /* 677 * first check the host byte, to see if there is anything in there 678 * that would indicate what we need to do. 679 */ 680 if (host_byte(scmd->result) == DID_RESET) { 681 /* 682 * rats. we are already in the error handler, so we now 683 * get to try and figure out what to do next. if the sense 684 * is valid, we have a pretty good idea of what to do. 685 * if not, we mark it as FAILED. 686 */ 687 return scsi_check_sense(scmd); 688 } 689 if (host_byte(scmd->result) != DID_OK) 690 return FAILED; 691 692 /* 693 * next, check the message byte. 694 */ 695 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 696 return FAILED; 697 698 /* 699 * now, check the status byte to see if this indicates 700 * anything special. 701 */ 702 switch (status_byte(scmd->result)) { 703 case GOOD: 704 scsi_handle_queue_ramp_up(scmd->device); 705 case COMMAND_TERMINATED: 706 return SUCCESS; 707 case CHECK_CONDITION: 708 return scsi_check_sense(scmd); 709 case CONDITION_GOOD: 710 case INTERMEDIATE_GOOD: 711 case INTERMEDIATE_C_GOOD: 712 /* 713 * who knows? FIXME(eric) 714 */ 715 return SUCCESS; 716 case RESERVATION_CONFLICT: 717 if (scmd->cmnd[0] == TEST_UNIT_READY) 718 /* it is a success, we probed the device and 719 * found it */ 720 return SUCCESS; 721 /* otherwise, we failed to send the command */ 722 return FAILED; 723 case QUEUE_FULL: 724 scsi_handle_queue_full(scmd->device); 725 /* fall through */ 726 case BUSY: 727 return NEEDS_RETRY; 728 default: 729 return FAILED; 730 } 731 return FAILED; 732 } 733 734 /** 735 * scsi_eh_done - Completion function for error handling. 736 * @scmd: Cmd that is done. 737 */ 738 static void scsi_eh_done(struct scsi_cmnd *scmd) 739 { 740 struct completion *eh_action; 741 742 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 743 "%s scmd: %p result: %x\n", 744 __func__, scmd, scmd->result)); 745 746 eh_action = scmd->device->host->eh_action; 747 if (eh_action) 748 complete(eh_action); 749 } 750 751 /** 752 * scsi_try_host_reset - ask host adapter to reset itself 753 * @scmd: SCSI cmd to send host reset. 754 */ 755 static int scsi_try_host_reset(struct scsi_cmnd *scmd) 756 { 757 unsigned long flags; 758 int rtn; 759 struct Scsi_Host *host = scmd->device->host; 760 struct scsi_host_template *hostt = host->hostt; 761 762 SCSI_LOG_ERROR_RECOVERY(3, 763 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 764 765 if (!hostt->eh_host_reset_handler) 766 return FAILED; 767 768 rtn = hostt->eh_host_reset_handler(scmd); 769 770 if (rtn == SUCCESS) { 771 if (!hostt->skip_settle_delay) 772 ssleep(HOST_RESET_SETTLE_TIME); 773 spin_lock_irqsave(host->host_lock, flags); 774 scsi_report_bus_reset(host, scmd_channel(scmd)); 775 spin_unlock_irqrestore(host->host_lock, flags); 776 } 777 778 return rtn; 779 } 780 781 /** 782 * scsi_try_bus_reset - ask host to perform a bus reset 783 * @scmd: SCSI cmd to send bus reset. 784 */ 785 static int scsi_try_bus_reset(struct scsi_cmnd *scmd) 786 { 787 unsigned long flags; 788 int rtn; 789 struct Scsi_Host *host = scmd->device->host; 790 struct scsi_host_template *hostt = host->hostt; 791 792 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 793 "%s: Snd Bus RST\n", __func__)); 794 795 if (!hostt->eh_bus_reset_handler) 796 return FAILED; 797 798 rtn = hostt->eh_bus_reset_handler(scmd); 799 800 if (rtn == SUCCESS) { 801 if (!hostt->skip_settle_delay) 802 ssleep(BUS_RESET_SETTLE_TIME); 803 spin_lock_irqsave(host->host_lock, flags); 804 scsi_report_bus_reset(host, scmd_channel(scmd)); 805 spin_unlock_irqrestore(host->host_lock, flags); 806 } 807 808 return rtn; 809 } 810 811 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 812 { 813 sdev->was_reset = 1; 814 sdev->expecting_cc_ua = 1; 815 } 816 817 /** 818 * scsi_try_target_reset - Ask host to perform a target reset 819 * @scmd: SCSI cmd used to send a target reset 820 * 821 * Notes: 822 * There is no timeout for this operation. if this operation is 823 * unreliable for a given host, then the host itself needs to put a 824 * timer on it, and set the host back to a consistent state prior to 825 * returning. 826 */ 827 static int scsi_try_target_reset(struct scsi_cmnd *scmd) 828 { 829 unsigned long flags; 830 int rtn; 831 struct Scsi_Host *host = scmd->device->host; 832 struct scsi_host_template *hostt = host->hostt; 833 834 if (!hostt->eh_target_reset_handler) 835 return FAILED; 836 837 rtn = hostt->eh_target_reset_handler(scmd); 838 if (rtn == SUCCESS) { 839 spin_lock_irqsave(host->host_lock, flags); 840 __starget_for_each_device(scsi_target(scmd->device), NULL, 841 __scsi_report_device_reset); 842 spin_unlock_irqrestore(host->host_lock, flags); 843 } 844 845 return rtn; 846 } 847 848 /** 849 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 850 * @scmd: SCSI cmd used to send BDR 851 * 852 * Notes: 853 * There is no timeout for this operation. if this operation is 854 * unreliable for a given host, then the host itself needs to put a 855 * timer on it, and set the host back to a consistent state prior to 856 * returning. 857 */ 858 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 859 { 860 int rtn; 861 struct scsi_host_template *hostt = scmd->device->host->hostt; 862 863 if (!hostt->eh_device_reset_handler) 864 return FAILED; 865 866 rtn = hostt->eh_device_reset_handler(scmd); 867 if (rtn == SUCCESS) 868 __scsi_report_device_reset(scmd->device, NULL); 869 return rtn; 870 } 871 872 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd) 873 { 874 if (!hostt->eh_abort_handler) 875 return FAILED; 876 877 return hostt->eh_abort_handler(scmd); 878 } 879 880 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 881 { 882 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 883 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 884 if (scsi_try_target_reset(scmd) != SUCCESS) 885 if (scsi_try_bus_reset(scmd) != SUCCESS) 886 scsi_try_host_reset(scmd); 887 } 888 889 /** 890 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 891 * @scmd: SCSI command structure to hijack 892 * @ses: structure to save restore information 893 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 894 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) 895 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 896 * 897 * This function is used to save a scsi command information before re-execution 898 * as part of the error recovery process. If @sense_bytes is 0 the command 899 * sent must be one that does not transfer any data. If @sense_bytes != 0 900 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 901 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 902 */ 903 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 904 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 905 { 906 struct scsi_device *sdev = scmd->device; 907 908 /* 909 * We need saved copies of a number of fields - this is because 910 * error handling may need to overwrite these with different values 911 * to run different commands, and once error handling is complete, 912 * we will need to restore these values prior to running the actual 913 * command. 914 */ 915 ses->cmd_len = scmd->cmd_len; 916 ses->cmnd = scmd->cmnd; 917 ses->data_direction = scmd->sc_data_direction; 918 ses->sdb = scmd->sdb; 919 ses->next_rq = scmd->request->next_rq; 920 ses->result = scmd->result; 921 ses->underflow = scmd->underflow; 922 ses->prot_op = scmd->prot_op; 923 924 scmd->prot_op = SCSI_PROT_NORMAL; 925 scmd->eh_eflags = 0; 926 scmd->cmnd = ses->eh_cmnd; 927 memset(scmd->cmnd, 0, BLK_MAX_CDB); 928 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 929 scmd->request->next_rq = NULL; 930 scmd->result = 0; 931 932 if (sense_bytes) { 933 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 934 sense_bytes); 935 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 936 scmd->sdb.length); 937 scmd->sdb.table.sgl = &ses->sense_sgl; 938 scmd->sc_data_direction = DMA_FROM_DEVICE; 939 scmd->sdb.table.nents = 1; 940 scmd->cmnd[0] = REQUEST_SENSE; 941 scmd->cmnd[4] = scmd->sdb.length; 942 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 943 } else { 944 scmd->sc_data_direction = DMA_NONE; 945 if (cmnd) { 946 BUG_ON(cmnd_size > BLK_MAX_CDB); 947 memcpy(scmd->cmnd, cmnd, cmnd_size); 948 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 949 } 950 } 951 952 scmd->underflow = 0; 953 954 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 955 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 956 (sdev->lun << 5 & 0xe0); 957 958 /* 959 * Zero the sense buffer. The scsi spec mandates that any 960 * untransferred sense data should be interpreted as being zero. 961 */ 962 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 963 } 964 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 965 966 /** 967 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 968 * @scmd: SCSI command structure to restore 969 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 970 * 971 * Undo any damage done by above scsi_eh_prep_cmnd(). 972 */ 973 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 974 { 975 /* 976 * Restore original data 977 */ 978 scmd->cmd_len = ses->cmd_len; 979 scmd->cmnd = ses->cmnd; 980 scmd->sc_data_direction = ses->data_direction; 981 scmd->sdb = ses->sdb; 982 scmd->request->next_rq = ses->next_rq; 983 scmd->result = ses->result; 984 scmd->underflow = ses->underflow; 985 scmd->prot_op = ses->prot_op; 986 } 987 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 988 989 /** 990 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 991 * @scmd: SCSI command structure to hijack 992 * @cmnd: CDB to send 993 * @cmnd_size: size in bytes of @cmnd 994 * @timeout: timeout for this request 995 * @sense_bytes: size of sense data to copy or 0 996 * 997 * This function is used to send a scsi command down to a target device 998 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 999 * 1000 * Return value: 1001 * SUCCESS or FAILED or NEEDS_RETRY 1002 */ 1003 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd, 1004 int cmnd_size, int timeout, unsigned sense_bytes) 1005 { 1006 struct scsi_device *sdev = scmd->device; 1007 struct Scsi_Host *shost = sdev->host; 1008 DECLARE_COMPLETION_ONSTACK(done); 1009 unsigned long timeleft = timeout; 1010 struct scsi_eh_save ses; 1011 const unsigned long stall_for = msecs_to_jiffies(100); 1012 int rtn; 1013 1014 retry: 1015 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1016 shost->eh_action = &done; 1017 1018 scsi_log_send(scmd); 1019 scmd->scsi_done = scsi_eh_done; 1020 rtn = shost->hostt->queuecommand(shost, scmd); 1021 if (rtn) { 1022 if (timeleft > stall_for) { 1023 scsi_eh_restore_cmnd(scmd, &ses); 1024 timeleft -= stall_for; 1025 msleep(jiffies_to_msecs(stall_for)); 1026 goto retry; 1027 } 1028 /* signal not to enter either branch of the if () below */ 1029 timeleft = 0; 1030 rtn = NEEDS_RETRY; 1031 } else { 1032 timeleft = wait_for_completion_timeout(&done, timeout); 1033 rtn = SUCCESS; 1034 } 1035 1036 shost->eh_action = NULL; 1037 1038 scsi_log_completion(scmd, rtn); 1039 1040 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1041 "%s: scmd: %p, timeleft: %ld\n", 1042 __func__, scmd, timeleft)); 1043 1044 /* 1045 * If there is time left scsi_eh_done got called, and we will examine 1046 * the actual status codes to see whether the command actually did 1047 * complete normally, else if we have a zero return and no time left, 1048 * the command must still be pending, so abort it and return FAILED. 1049 * If we never actually managed to issue the command, because 1050 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1051 * value above (so don't execute either branch of the if) 1052 */ 1053 if (timeleft) { 1054 rtn = scsi_eh_completed_normally(scmd); 1055 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1056 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1057 1058 switch (rtn) { 1059 case SUCCESS: 1060 case NEEDS_RETRY: 1061 case FAILED: 1062 break; 1063 case ADD_TO_MLQUEUE: 1064 rtn = NEEDS_RETRY; 1065 break; 1066 default: 1067 rtn = FAILED; 1068 break; 1069 } 1070 } else if (!rtn) { 1071 scsi_abort_eh_cmnd(scmd); 1072 rtn = FAILED; 1073 } 1074 1075 scsi_eh_restore_cmnd(scmd, &ses); 1076 1077 return rtn; 1078 } 1079 1080 /** 1081 * scsi_request_sense - Request sense data from a particular target. 1082 * @scmd: SCSI cmd for request sense. 1083 * 1084 * Notes: 1085 * Some hosts automatically obtain this information, others require 1086 * that we obtain it on our own. This function will *not* return until 1087 * the command either times out, or it completes. 1088 */ 1089 static int scsi_request_sense(struct scsi_cmnd *scmd) 1090 { 1091 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1092 } 1093 1094 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn) 1095 { 1096 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) { 1097 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1098 if (sdrv->eh_action) 1099 rtn = sdrv->eh_action(scmd, rtn); 1100 } 1101 return rtn; 1102 } 1103 1104 /** 1105 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1106 * @scmd: Original SCSI cmd that eh has finished. 1107 * @done_q: Queue for processed commands. 1108 * 1109 * Notes: 1110 * We don't want to use the normal command completion while we are are 1111 * still handling errors - it may cause other commands to be queued, 1112 * and that would disturb what we are doing. Thus we really want to 1113 * keep a list of pending commands for final completion, and once we 1114 * are ready to leave error handling we handle completion for real. 1115 */ 1116 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1117 { 1118 scmd->device->host->host_failed--; 1119 scmd->eh_eflags = 0; 1120 list_move_tail(&scmd->eh_entry, done_q); 1121 } 1122 EXPORT_SYMBOL(scsi_eh_finish_cmd); 1123 1124 /** 1125 * scsi_eh_get_sense - Get device sense data. 1126 * @work_q: Queue of commands to process. 1127 * @done_q: Queue of processed commands. 1128 * 1129 * Description: 1130 * See if we need to request sense information. if so, then get it 1131 * now, so we have a better idea of what to do. 1132 * 1133 * Notes: 1134 * This has the unfortunate side effect that if a shost adapter does 1135 * not automatically request sense information, we end up shutting 1136 * it down before we request it. 1137 * 1138 * All drivers should request sense information internally these days, 1139 * so for now all I have to say is tough noogies if you end up in here. 1140 * 1141 * XXX: Long term this code should go away, but that needs an audit of 1142 * all LLDDs first. 1143 */ 1144 int scsi_eh_get_sense(struct list_head *work_q, 1145 struct list_head *done_q) 1146 { 1147 struct scsi_cmnd *scmd, *next; 1148 struct Scsi_Host *shost; 1149 int rtn; 1150 1151 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1152 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) || 1153 SCSI_SENSE_VALID(scmd)) 1154 continue; 1155 1156 shost = scmd->device->host; 1157 if (scsi_host_eh_past_deadline(shost)) { 1158 SCSI_LOG_ERROR_RECOVERY(3, 1159 shost_printk(KERN_INFO, shost, 1160 "skip %s, past eh deadline\n", 1161 __func__)); 1162 break; 1163 } 1164 if (status_byte(scmd->result) != CHECK_CONDITION) 1165 /* 1166 * don't request sense if there's no check condition 1167 * status because the error we're processing isn't one 1168 * that has a sense code (and some devices get 1169 * confused by sense requests out of the blue) 1170 */ 1171 continue; 1172 1173 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1174 "%s: requesting sense\n", 1175 current->comm)); 1176 rtn = scsi_request_sense(scmd); 1177 if (rtn != SUCCESS) 1178 continue; 1179 1180 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1181 "sense requested for %p result %x\n", 1182 scmd, scmd->result)); 1183 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd)); 1184 1185 rtn = scsi_decide_disposition(scmd); 1186 1187 /* 1188 * if the result was normal, then just pass it along to the 1189 * upper level. 1190 */ 1191 if (rtn == SUCCESS) 1192 /* we don't want this command reissued, just 1193 * finished with the sense data, so set 1194 * retries to the max allowed to ensure it 1195 * won't get reissued */ 1196 scmd->retries = scmd->allowed; 1197 else if (rtn != NEEDS_RETRY) 1198 continue; 1199 1200 scsi_eh_finish_cmd(scmd, done_q); 1201 } 1202 1203 return list_empty(work_q); 1204 } 1205 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1206 1207 /** 1208 * scsi_eh_tur - Send TUR to device. 1209 * @scmd: &scsi_cmnd to send TUR 1210 * 1211 * Return value: 1212 * 0 - Device is ready. 1 - Device NOT ready. 1213 */ 1214 static int scsi_eh_tur(struct scsi_cmnd *scmd) 1215 { 1216 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1217 int retry_cnt = 1, rtn; 1218 1219 retry_tur: 1220 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1221 scmd->device->eh_timeout, 0); 1222 1223 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1224 "%s: scmd %p rtn %x\n", __func__, scmd, rtn)); 1225 1226 switch (rtn) { 1227 case NEEDS_RETRY: 1228 if (retry_cnt--) 1229 goto retry_tur; 1230 /*FALLTHRU*/ 1231 case SUCCESS: 1232 return 0; 1233 default: 1234 return 1; 1235 } 1236 } 1237 1238 /** 1239 * scsi_eh_test_devices - check if devices are responding from error recovery. 1240 * @cmd_list: scsi commands in error recovery. 1241 * @work_q: queue for commands which still need more error recovery 1242 * @done_q: queue for commands which are finished 1243 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1244 * 1245 * Decription: 1246 * Tests if devices are in a working state. Commands to devices now in 1247 * a working state are sent to the done_q while commands to devices which 1248 * are still failing to respond are returned to the work_q for more 1249 * processing. 1250 **/ 1251 static int scsi_eh_test_devices(struct list_head *cmd_list, 1252 struct list_head *work_q, 1253 struct list_head *done_q, int try_stu) 1254 { 1255 struct scsi_cmnd *scmd, *next; 1256 struct scsi_device *sdev; 1257 int finish_cmds; 1258 1259 while (!list_empty(cmd_list)) { 1260 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1261 sdev = scmd->device; 1262 1263 if (!try_stu) { 1264 if (scsi_host_eh_past_deadline(sdev->host)) { 1265 /* Push items back onto work_q */ 1266 list_splice_init(cmd_list, work_q); 1267 SCSI_LOG_ERROR_RECOVERY(3, 1268 shost_printk(KERN_INFO, sdev->host, 1269 "skip %s, past eh deadline", 1270 __func__)); 1271 break; 1272 } 1273 } 1274 1275 finish_cmds = !scsi_device_online(scmd->device) || 1276 (try_stu && !scsi_eh_try_stu(scmd) && 1277 !scsi_eh_tur(scmd)) || 1278 !scsi_eh_tur(scmd); 1279 1280 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1281 if (scmd->device == sdev) { 1282 if (finish_cmds && 1283 (try_stu || 1284 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1285 scsi_eh_finish_cmd(scmd, done_q); 1286 else 1287 list_move_tail(&scmd->eh_entry, work_q); 1288 } 1289 } 1290 return list_empty(work_q); 1291 } 1292 1293 1294 /** 1295 * scsi_eh_abort_cmds - abort pending commands. 1296 * @work_q: &list_head for pending commands. 1297 * @done_q: &list_head for processed commands. 1298 * 1299 * Decription: 1300 * Try and see whether or not it makes sense to try and abort the 1301 * running command. This only works out to be the case if we have one 1302 * command that has timed out. If the command simply failed, it makes 1303 * no sense to try and abort the command, since as far as the shost 1304 * adapter is concerned, it isn't running. 1305 */ 1306 static int scsi_eh_abort_cmds(struct list_head *work_q, 1307 struct list_head *done_q) 1308 { 1309 struct scsi_cmnd *scmd, *next; 1310 LIST_HEAD(check_list); 1311 int rtn; 1312 struct Scsi_Host *shost; 1313 1314 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1315 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD)) 1316 continue; 1317 shost = scmd->device->host; 1318 if (scsi_host_eh_past_deadline(shost)) { 1319 list_splice_init(&check_list, work_q); 1320 SCSI_LOG_ERROR_RECOVERY(3, 1321 shost_printk(KERN_INFO, shost, 1322 "skip %s, past eh deadline\n", 1323 __func__)); 1324 return list_empty(work_q); 1325 } 1326 SCSI_LOG_ERROR_RECOVERY(3, 1327 shost_printk(KERN_INFO, shost, 1328 "%s: aborting cmd: 0x%p\n", 1329 current->comm, scmd)); 1330 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd); 1331 if (rtn == FAILED) { 1332 SCSI_LOG_ERROR_RECOVERY(3, 1333 shost_printk(KERN_INFO, shost, 1334 "%s: aborting cmd failed: 0x%p\n", 1335 current->comm, scmd)); 1336 list_splice_init(&check_list, work_q); 1337 return list_empty(work_q); 1338 } 1339 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD; 1340 if (rtn == FAST_IO_FAIL) 1341 scsi_eh_finish_cmd(scmd, done_q); 1342 else 1343 list_move_tail(&scmd->eh_entry, &check_list); 1344 } 1345 1346 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1347 } 1348 1349 /** 1350 * scsi_eh_try_stu - Send START_UNIT to device. 1351 * @scmd: &scsi_cmnd to send START_UNIT 1352 * 1353 * Return value: 1354 * 0 - Device is ready. 1 - Device NOT ready. 1355 */ 1356 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1357 { 1358 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1359 1360 if (scmd->device->allow_restart) { 1361 int i, rtn = NEEDS_RETRY; 1362 1363 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1364 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0); 1365 1366 if (rtn == SUCCESS) 1367 return 0; 1368 } 1369 1370 return 1; 1371 } 1372 1373 /** 1374 * scsi_eh_stu - send START_UNIT if needed 1375 * @shost: &scsi host being recovered. 1376 * @work_q: &list_head for pending commands. 1377 * @done_q: &list_head for processed commands. 1378 * 1379 * Notes: 1380 * If commands are failing due to not ready, initializing command required, 1381 * try revalidating the device, which will end up sending a start unit. 1382 */ 1383 static int scsi_eh_stu(struct Scsi_Host *shost, 1384 struct list_head *work_q, 1385 struct list_head *done_q) 1386 { 1387 struct scsi_cmnd *scmd, *stu_scmd, *next; 1388 struct scsi_device *sdev; 1389 1390 shost_for_each_device(sdev, shost) { 1391 if (scsi_host_eh_past_deadline(shost)) { 1392 SCSI_LOG_ERROR_RECOVERY(3, 1393 shost_printk(KERN_INFO, shost, 1394 "skip %s, past eh deadline\n", 1395 __func__)); 1396 break; 1397 } 1398 stu_scmd = NULL; 1399 list_for_each_entry(scmd, work_q, eh_entry) 1400 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1401 scsi_check_sense(scmd) == FAILED ) { 1402 stu_scmd = scmd; 1403 break; 1404 } 1405 1406 if (!stu_scmd) 1407 continue; 1408 1409 SCSI_LOG_ERROR_RECOVERY(3, 1410 shost_printk(KERN_INFO, shost, 1411 "%s: Sending START_UNIT to sdev: 0x%p\n", 1412 current->comm, sdev)); 1413 1414 if (!scsi_eh_try_stu(stu_scmd)) { 1415 if (!scsi_device_online(sdev) || 1416 !scsi_eh_tur(stu_scmd)) { 1417 list_for_each_entry_safe(scmd, next, 1418 work_q, eh_entry) { 1419 if (scmd->device == sdev && 1420 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1421 scsi_eh_finish_cmd(scmd, done_q); 1422 } 1423 } 1424 } else { 1425 SCSI_LOG_ERROR_RECOVERY(3, 1426 shost_printk(KERN_INFO, shost, 1427 "%s: START_UNIT failed to sdev:" 1428 " 0x%p\n", current->comm, sdev)); 1429 } 1430 } 1431 1432 return list_empty(work_q); 1433 } 1434 1435 1436 /** 1437 * scsi_eh_bus_device_reset - send bdr if needed 1438 * @shost: scsi host being recovered. 1439 * @work_q: &list_head for pending commands. 1440 * @done_q: &list_head for processed commands. 1441 * 1442 * Notes: 1443 * Try a bus device reset. Still, look to see whether we have multiple 1444 * devices that are jammed or not - if we have multiple devices, it 1445 * makes no sense to try bus_device_reset - we really would need to try 1446 * a bus_reset instead. 1447 */ 1448 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1449 struct list_head *work_q, 1450 struct list_head *done_q) 1451 { 1452 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1453 struct scsi_device *sdev; 1454 int rtn; 1455 1456 shost_for_each_device(sdev, shost) { 1457 if (scsi_host_eh_past_deadline(shost)) { 1458 SCSI_LOG_ERROR_RECOVERY(3, 1459 shost_printk(KERN_INFO, shost, 1460 "skip %s, past eh deadline\n", 1461 __func__)); 1462 break; 1463 } 1464 bdr_scmd = NULL; 1465 list_for_each_entry(scmd, work_q, eh_entry) 1466 if (scmd->device == sdev) { 1467 bdr_scmd = scmd; 1468 break; 1469 } 1470 1471 if (!bdr_scmd) 1472 continue; 1473 1474 SCSI_LOG_ERROR_RECOVERY(3, 1475 shost_printk(KERN_INFO, shost, 1476 "%s: Sending BDR sdev: 0x%p\n", 1477 current->comm, sdev)); 1478 rtn = scsi_try_bus_device_reset(bdr_scmd); 1479 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1480 if (!scsi_device_online(sdev) || 1481 rtn == FAST_IO_FAIL || 1482 !scsi_eh_tur(bdr_scmd)) { 1483 list_for_each_entry_safe(scmd, next, 1484 work_q, eh_entry) { 1485 if (scmd->device == sdev && 1486 scsi_eh_action(scmd, rtn) != FAILED) 1487 scsi_eh_finish_cmd(scmd, 1488 done_q); 1489 } 1490 } 1491 } else { 1492 SCSI_LOG_ERROR_RECOVERY(3, 1493 shost_printk(KERN_INFO, shost, 1494 "%s: BDR failed sdev: 0x%p\n", 1495 current->comm, sdev)); 1496 } 1497 } 1498 1499 return list_empty(work_q); 1500 } 1501 1502 /** 1503 * scsi_eh_target_reset - send target reset if needed 1504 * @shost: scsi host being recovered. 1505 * @work_q: &list_head for pending commands. 1506 * @done_q: &list_head for processed commands. 1507 * 1508 * Notes: 1509 * Try a target reset. 1510 */ 1511 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1512 struct list_head *work_q, 1513 struct list_head *done_q) 1514 { 1515 LIST_HEAD(tmp_list); 1516 LIST_HEAD(check_list); 1517 1518 list_splice_init(work_q, &tmp_list); 1519 1520 while (!list_empty(&tmp_list)) { 1521 struct scsi_cmnd *next, *scmd; 1522 int rtn; 1523 unsigned int id; 1524 1525 if (scsi_host_eh_past_deadline(shost)) { 1526 /* push back on work queue for further processing */ 1527 list_splice_init(&check_list, work_q); 1528 list_splice_init(&tmp_list, work_q); 1529 SCSI_LOG_ERROR_RECOVERY(3, 1530 shost_printk(KERN_INFO, shost, 1531 "skip %s, past eh deadline\n", 1532 __func__)); 1533 return list_empty(work_q); 1534 } 1535 1536 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1537 id = scmd_id(scmd); 1538 1539 SCSI_LOG_ERROR_RECOVERY(3, 1540 shost_printk(KERN_INFO, shost, 1541 "%s: Sending target reset to target %d\n", 1542 current->comm, id)); 1543 rtn = scsi_try_target_reset(scmd); 1544 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1545 SCSI_LOG_ERROR_RECOVERY(3, 1546 shost_printk(KERN_INFO, shost, 1547 "%s: Target reset failed" 1548 " target: %d\n", 1549 current->comm, id)); 1550 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1551 if (scmd_id(scmd) != id) 1552 continue; 1553 1554 if (rtn == SUCCESS) 1555 list_move_tail(&scmd->eh_entry, &check_list); 1556 else if (rtn == FAST_IO_FAIL) 1557 scsi_eh_finish_cmd(scmd, done_q); 1558 else 1559 /* push back on work queue for further processing */ 1560 list_move(&scmd->eh_entry, work_q); 1561 } 1562 } 1563 1564 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1565 } 1566 1567 /** 1568 * scsi_eh_bus_reset - send a bus reset 1569 * @shost: &scsi host being recovered. 1570 * @work_q: &list_head for pending commands. 1571 * @done_q: &list_head for processed commands. 1572 */ 1573 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1574 struct list_head *work_q, 1575 struct list_head *done_q) 1576 { 1577 struct scsi_cmnd *scmd, *chan_scmd, *next; 1578 LIST_HEAD(check_list); 1579 unsigned int channel; 1580 int rtn; 1581 1582 /* 1583 * we really want to loop over the various channels, and do this on 1584 * a channel by channel basis. we should also check to see if any 1585 * of the failed commands are on soft_reset devices, and if so, skip 1586 * the reset. 1587 */ 1588 1589 for (channel = 0; channel <= shost->max_channel; channel++) { 1590 if (scsi_host_eh_past_deadline(shost)) { 1591 list_splice_init(&check_list, work_q); 1592 SCSI_LOG_ERROR_RECOVERY(3, 1593 shost_printk(KERN_INFO, shost, 1594 "skip %s, past eh deadline\n", 1595 __func__)); 1596 return list_empty(work_q); 1597 } 1598 1599 chan_scmd = NULL; 1600 list_for_each_entry(scmd, work_q, eh_entry) { 1601 if (channel == scmd_channel(scmd)) { 1602 chan_scmd = scmd; 1603 break; 1604 /* 1605 * FIXME add back in some support for 1606 * soft_reset devices. 1607 */ 1608 } 1609 } 1610 1611 if (!chan_scmd) 1612 continue; 1613 SCSI_LOG_ERROR_RECOVERY(3, 1614 shost_printk(KERN_INFO, shost, 1615 "%s: Sending BRST chan: %d\n", 1616 current->comm, channel)); 1617 rtn = scsi_try_bus_reset(chan_scmd); 1618 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1619 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1620 if (channel == scmd_channel(scmd)) { 1621 if (rtn == FAST_IO_FAIL) 1622 scsi_eh_finish_cmd(scmd, 1623 done_q); 1624 else 1625 list_move_tail(&scmd->eh_entry, 1626 &check_list); 1627 } 1628 } 1629 } else { 1630 SCSI_LOG_ERROR_RECOVERY(3, 1631 shost_printk(KERN_INFO, shost, 1632 "%s: BRST failed chan: %d\n", 1633 current->comm, channel)); 1634 } 1635 } 1636 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1637 } 1638 1639 /** 1640 * scsi_eh_host_reset - send a host reset 1641 * @shost: host to be reset. 1642 * @work_q: &list_head for pending commands. 1643 * @done_q: &list_head for processed commands. 1644 */ 1645 static int scsi_eh_host_reset(struct Scsi_Host *shost, 1646 struct list_head *work_q, 1647 struct list_head *done_q) 1648 { 1649 struct scsi_cmnd *scmd, *next; 1650 LIST_HEAD(check_list); 1651 int rtn; 1652 1653 if (!list_empty(work_q)) { 1654 scmd = list_entry(work_q->next, 1655 struct scsi_cmnd, eh_entry); 1656 1657 SCSI_LOG_ERROR_RECOVERY(3, 1658 shost_printk(KERN_INFO, shost, 1659 "%s: Sending HRST\n", 1660 current->comm)); 1661 1662 rtn = scsi_try_host_reset(scmd); 1663 if (rtn == SUCCESS) { 1664 list_splice_init(work_q, &check_list); 1665 } else if (rtn == FAST_IO_FAIL) { 1666 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1667 scsi_eh_finish_cmd(scmd, done_q); 1668 } 1669 } else { 1670 SCSI_LOG_ERROR_RECOVERY(3, 1671 shost_printk(KERN_INFO, shost, 1672 "%s: HRST failed\n", 1673 current->comm)); 1674 } 1675 } 1676 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1677 } 1678 1679 /** 1680 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1681 * @work_q: &list_head for pending commands. 1682 * @done_q: &list_head for processed commands. 1683 */ 1684 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1685 struct list_head *done_q) 1686 { 1687 struct scsi_cmnd *scmd, *next; 1688 1689 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1690 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1691 "not ready after error recovery\n"); 1692 scsi_device_set_state(scmd->device, SDEV_OFFLINE); 1693 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) { 1694 /* 1695 * FIXME: Handle lost cmds. 1696 */ 1697 } 1698 scsi_eh_finish_cmd(scmd, done_q); 1699 } 1700 return; 1701 } 1702 1703 /** 1704 * scsi_noretry_cmd - determine if command should be failed fast 1705 * @scmd: SCSI cmd to examine. 1706 */ 1707 int scsi_noretry_cmd(struct scsi_cmnd *scmd) 1708 { 1709 switch (host_byte(scmd->result)) { 1710 case DID_OK: 1711 break; 1712 case DID_TIME_OUT: 1713 goto check_type; 1714 case DID_BUS_BUSY: 1715 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT); 1716 case DID_PARITY: 1717 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV); 1718 case DID_ERROR: 1719 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1720 status_byte(scmd->result) == RESERVATION_CONFLICT) 1721 return 0; 1722 /* fall through */ 1723 case DID_SOFT_ERROR: 1724 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER); 1725 } 1726 1727 if (status_byte(scmd->result) != CHECK_CONDITION) 1728 return 0; 1729 1730 check_type: 1731 /* 1732 * assume caller has checked sense and determined 1733 * the check condition was retryable. 1734 */ 1735 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV || 1736 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC) 1737 return 1; 1738 else 1739 return 0; 1740 } 1741 1742 /** 1743 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1744 * @scmd: SCSI cmd to examine. 1745 * 1746 * Notes: 1747 * This is *only* called when we are examining the status after sending 1748 * out the actual data command. any commands that are queued for error 1749 * recovery (e.g. test_unit_ready) do *not* come through here. 1750 * 1751 * When this routine returns failed, it means the error handler thread 1752 * is woken. In cases where the error code indicates an error that 1753 * doesn't require the error handler read (i.e. we don't need to 1754 * abort/reset), this function should return SUCCESS. 1755 */ 1756 int scsi_decide_disposition(struct scsi_cmnd *scmd) 1757 { 1758 int rtn; 1759 1760 /* 1761 * if the device is offline, then we clearly just pass the result back 1762 * up to the top level. 1763 */ 1764 if (!scsi_device_online(scmd->device)) { 1765 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1766 "%s: device offline - report as SUCCESS\n", __func__)); 1767 return SUCCESS; 1768 } 1769 1770 /* 1771 * first check the host byte, to see if there is anything in there 1772 * that would indicate what we need to do. 1773 */ 1774 switch (host_byte(scmd->result)) { 1775 case DID_PASSTHROUGH: 1776 /* 1777 * no matter what, pass this through to the upper layer. 1778 * nuke this special code so that it looks like we are saying 1779 * did_ok. 1780 */ 1781 scmd->result &= 0xff00ffff; 1782 return SUCCESS; 1783 case DID_OK: 1784 /* 1785 * looks good. drop through, and check the next byte. 1786 */ 1787 break; 1788 case DID_ABORT: 1789 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1790 set_host_byte(scmd, DID_TIME_OUT); 1791 return SUCCESS; 1792 } 1793 case DID_NO_CONNECT: 1794 case DID_BAD_TARGET: 1795 /* 1796 * note - this means that we just report the status back 1797 * to the top level driver, not that we actually think 1798 * that it indicates SUCCESS. 1799 */ 1800 return SUCCESS; 1801 /* 1802 * when the low level driver returns did_soft_error, 1803 * it is responsible for keeping an internal retry counter 1804 * in order to avoid endless loops (db) 1805 * 1806 * actually this is a bug in this function here. we should 1807 * be mindful of the maximum number of retries specified 1808 * and not get stuck in a loop. 1809 */ 1810 case DID_SOFT_ERROR: 1811 goto maybe_retry; 1812 case DID_IMM_RETRY: 1813 return NEEDS_RETRY; 1814 1815 case DID_REQUEUE: 1816 return ADD_TO_MLQUEUE; 1817 case DID_TRANSPORT_DISRUPTED: 1818 /* 1819 * LLD/transport was disrupted during processing of the IO. 1820 * The transport class is now blocked/blocking, 1821 * and the transport will decide what to do with the IO 1822 * based on its timers and recovery capablilities if 1823 * there are enough retries. 1824 */ 1825 goto maybe_retry; 1826 case DID_TRANSPORT_FAILFAST: 1827 /* 1828 * The transport decided to failfast the IO (most likely 1829 * the fast io fail tmo fired), so send IO directly upwards. 1830 */ 1831 return SUCCESS; 1832 case DID_ERROR: 1833 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1834 status_byte(scmd->result) == RESERVATION_CONFLICT) 1835 /* 1836 * execute reservation conflict processing code 1837 * lower down 1838 */ 1839 break; 1840 /* fallthrough */ 1841 case DID_BUS_BUSY: 1842 case DID_PARITY: 1843 goto maybe_retry; 1844 case DID_TIME_OUT: 1845 /* 1846 * when we scan the bus, we get timeout messages for 1847 * these commands if there is no device available. 1848 * other hosts report did_no_connect for the same thing. 1849 */ 1850 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1851 scmd->cmnd[0] == INQUIRY)) { 1852 return SUCCESS; 1853 } else { 1854 return FAILED; 1855 } 1856 case DID_RESET: 1857 return SUCCESS; 1858 default: 1859 return FAILED; 1860 } 1861 1862 /* 1863 * next, check the message byte. 1864 */ 1865 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 1866 return FAILED; 1867 1868 /* 1869 * check the status byte to see if this indicates anything special. 1870 */ 1871 switch (status_byte(scmd->result)) { 1872 case QUEUE_FULL: 1873 scsi_handle_queue_full(scmd->device); 1874 /* 1875 * the case of trying to send too many commands to a 1876 * tagged queueing device. 1877 */ 1878 case BUSY: 1879 /* 1880 * device can't talk to us at the moment. Should only 1881 * occur (SAM-3) when the task queue is empty, so will cause 1882 * the empty queue handling to trigger a stall in the 1883 * device. 1884 */ 1885 return ADD_TO_MLQUEUE; 1886 case GOOD: 1887 if (scmd->cmnd[0] == REPORT_LUNS) 1888 scmd->device->sdev_target->expecting_lun_change = 0; 1889 scsi_handle_queue_ramp_up(scmd->device); 1890 case COMMAND_TERMINATED: 1891 return SUCCESS; 1892 case TASK_ABORTED: 1893 goto maybe_retry; 1894 case CHECK_CONDITION: 1895 rtn = scsi_check_sense(scmd); 1896 if (rtn == NEEDS_RETRY) 1897 goto maybe_retry; 1898 /* if rtn == FAILED, we have no sense information; 1899 * returning FAILED will wake the error handler thread 1900 * to collect the sense and redo the decide 1901 * disposition */ 1902 return rtn; 1903 case CONDITION_GOOD: 1904 case INTERMEDIATE_GOOD: 1905 case INTERMEDIATE_C_GOOD: 1906 case ACA_ACTIVE: 1907 /* 1908 * who knows? FIXME(eric) 1909 */ 1910 return SUCCESS; 1911 1912 case RESERVATION_CONFLICT: 1913 sdev_printk(KERN_INFO, scmd->device, 1914 "reservation conflict\n"); 1915 set_host_byte(scmd, DID_NEXUS_FAILURE); 1916 return SUCCESS; /* causes immediate i/o error */ 1917 default: 1918 return FAILED; 1919 } 1920 return FAILED; 1921 1922 maybe_retry: 1923 1924 /* we requeue for retry because the error was retryable, and 1925 * the request was not marked fast fail. Note that above, 1926 * even if the request is marked fast fail, we still requeue 1927 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1928 if ((++scmd->retries) <= scmd->allowed 1929 && !scsi_noretry_cmd(scmd)) { 1930 return NEEDS_RETRY; 1931 } else { 1932 /* 1933 * no more retries - report this one back to upper level. 1934 */ 1935 return SUCCESS; 1936 } 1937 } 1938 1939 static void eh_lock_door_done(struct request *req, int uptodate) 1940 { 1941 __blk_put_request(req->q, req); 1942 } 1943 1944 /** 1945 * scsi_eh_lock_door - Prevent medium removal for the specified device 1946 * @sdev: SCSI device to prevent medium removal 1947 * 1948 * Locking: 1949 * We must be called from process context. 1950 * 1951 * Notes: 1952 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 1953 * head of the devices request queue, and continue. 1954 */ 1955 static void scsi_eh_lock_door(struct scsi_device *sdev) 1956 { 1957 struct request *req; 1958 1959 /* 1960 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a 1961 * request becomes available 1962 */ 1963 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL); 1964 if (IS_ERR(req)) 1965 return; 1966 1967 blk_rq_set_block_pc(req); 1968 1969 req->cmd[0] = ALLOW_MEDIUM_REMOVAL; 1970 req->cmd[1] = 0; 1971 req->cmd[2] = 0; 1972 req->cmd[3] = 0; 1973 req->cmd[4] = SCSI_REMOVAL_PREVENT; 1974 req->cmd[5] = 0; 1975 1976 req->cmd_len = COMMAND_SIZE(req->cmd[0]); 1977 1978 req->cmd_flags |= REQ_QUIET; 1979 req->timeout = 10 * HZ; 1980 req->retries = 5; 1981 1982 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done); 1983 } 1984 1985 /** 1986 * scsi_restart_operations - restart io operations to the specified host. 1987 * @shost: Host we are restarting. 1988 * 1989 * Notes: 1990 * When we entered the error handler, we blocked all further i/o to 1991 * this device. we need to 'reverse' this process. 1992 */ 1993 static void scsi_restart_operations(struct Scsi_Host *shost) 1994 { 1995 struct scsi_device *sdev; 1996 unsigned long flags; 1997 1998 /* 1999 * If the door was locked, we need to insert a door lock request 2000 * onto the head of the SCSI request queue for the device. There 2001 * is no point trying to lock the door of an off-line device. 2002 */ 2003 shost_for_each_device(sdev, shost) { 2004 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 2005 scsi_eh_lock_door(sdev); 2006 sdev->was_reset = 0; 2007 } 2008 } 2009 2010 /* 2011 * next free up anything directly waiting upon the host. this 2012 * will be requests for character device operations, and also for 2013 * ioctls to queued block devices. 2014 */ 2015 SCSI_LOG_ERROR_RECOVERY(3, 2016 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 2017 2018 spin_lock_irqsave(shost->host_lock, flags); 2019 if (scsi_host_set_state(shost, SHOST_RUNNING)) 2020 if (scsi_host_set_state(shost, SHOST_CANCEL)) 2021 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 2022 spin_unlock_irqrestore(shost->host_lock, flags); 2023 2024 wake_up(&shost->host_wait); 2025 2026 /* 2027 * finally we need to re-initiate requests that may be pending. we will 2028 * have had everything blocked while error handling is taking place, and 2029 * now that error recovery is done, we will need to ensure that these 2030 * requests are started. 2031 */ 2032 scsi_run_host_queues(shost); 2033 2034 /* 2035 * if eh is active and host_eh_scheduled is pending we need to re-run 2036 * recovery. we do this check after scsi_run_host_queues() to allow 2037 * everything pent up since the last eh run a chance to make forward 2038 * progress before we sync again. Either we'll immediately re-run 2039 * recovery or scsi_device_unbusy() will wake us again when these 2040 * pending commands complete. 2041 */ 2042 spin_lock_irqsave(shost->host_lock, flags); 2043 if (shost->host_eh_scheduled) 2044 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 2045 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 2046 spin_unlock_irqrestore(shost->host_lock, flags); 2047 } 2048 2049 /** 2050 * scsi_eh_ready_devs - check device ready state and recover if not. 2051 * @shost: host to be recovered. 2052 * @work_q: &list_head for pending commands. 2053 * @done_q: &list_head for processed commands. 2054 */ 2055 void scsi_eh_ready_devs(struct Scsi_Host *shost, 2056 struct list_head *work_q, 2057 struct list_head *done_q) 2058 { 2059 if (!scsi_eh_stu(shost, work_q, done_q)) 2060 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2061 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2062 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2063 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2064 scsi_eh_offline_sdevs(work_q, 2065 done_q); 2066 } 2067 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2068 2069 /** 2070 * scsi_eh_flush_done_q - finish processed commands or retry them. 2071 * @done_q: list_head of processed commands. 2072 */ 2073 void scsi_eh_flush_done_q(struct list_head *done_q) 2074 { 2075 struct scsi_cmnd *scmd, *next; 2076 2077 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2078 list_del_init(&scmd->eh_entry); 2079 if (scsi_device_online(scmd->device) && 2080 !scsi_noretry_cmd(scmd) && 2081 (++scmd->retries <= scmd->allowed)) { 2082 SCSI_LOG_ERROR_RECOVERY(3, 2083 scmd_printk(KERN_INFO, scmd, 2084 "%s: flush retry cmd: %p\n", 2085 current->comm, scmd)); 2086 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2087 } else { 2088 /* 2089 * If just we got sense for the device (called 2090 * scsi_eh_get_sense), scmd->result is already 2091 * set, do not set DRIVER_TIMEOUT. 2092 */ 2093 if (!scmd->result) 2094 scmd->result |= (DRIVER_TIMEOUT << 24); 2095 SCSI_LOG_ERROR_RECOVERY(3, 2096 scmd_printk(KERN_INFO, scmd, 2097 "%s: flush finish cmd: %p\n", 2098 current->comm, scmd)); 2099 scsi_finish_command(scmd); 2100 } 2101 } 2102 } 2103 EXPORT_SYMBOL(scsi_eh_flush_done_q); 2104 2105 /** 2106 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2107 * @shost: Host to unjam. 2108 * 2109 * Notes: 2110 * When we come in here, we *know* that all commands on the bus have 2111 * either completed, failed or timed out. we also know that no further 2112 * commands are being sent to the host, so things are relatively quiet 2113 * and we have freedom to fiddle with things as we wish. 2114 * 2115 * This is only the *default* implementation. it is possible for 2116 * individual drivers to supply their own version of this function, and 2117 * if the maintainer wishes to do this, it is strongly suggested that 2118 * this function be taken as a template and modified. this function 2119 * was designed to correctly handle problems for about 95% of the 2120 * different cases out there, and it should always provide at least a 2121 * reasonable amount of error recovery. 2122 * 2123 * Any command marked 'failed' or 'timeout' must eventually have 2124 * scsi_finish_cmd() called for it. we do all of the retry stuff 2125 * here, so when we restart the host after we return it should have an 2126 * empty queue. 2127 */ 2128 static void scsi_unjam_host(struct Scsi_Host *shost) 2129 { 2130 unsigned long flags; 2131 LIST_HEAD(eh_work_q); 2132 LIST_HEAD(eh_done_q); 2133 2134 spin_lock_irqsave(shost->host_lock, flags); 2135 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2136 spin_unlock_irqrestore(shost->host_lock, flags); 2137 2138 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2139 2140 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2141 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q)) 2142 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2143 2144 spin_lock_irqsave(shost->host_lock, flags); 2145 if (shost->eh_deadline != -1) 2146 shost->last_reset = 0; 2147 spin_unlock_irqrestore(shost->host_lock, flags); 2148 scsi_eh_flush_done_q(&eh_done_q); 2149 } 2150 2151 /** 2152 * scsi_error_handler - SCSI error handler thread 2153 * @data: Host for which we are running. 2154 * 2155 * Notes: 2156 * This is the main error handling loop. This is run as a kernel thread 2157 * for every SCSI host and handles all error handling activity. 2158 */ 2159 int scsi_error_handler(void *data) 2160 { 2161 struct Scsi_Host *shost = data; 2162 2163 /* 2164 * We use TASK_INTERRUPTIBLE so that the thread is not 2165 * counted against the load average as a running process. 2166 * We never actually get interrupted because kthread_run 2167 * disables signal delivery for the created thread. 2168 */ 2169 while (!kthread_should_stop()) { 2170 set_current_state(TASK_INTERRUPTIBLE); 2171 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2172 shost->host_failed != atomic_read(&shost->host_busy)) { 2173 SCSI_LOG_ERROR_RECOVERY(1, 2174 shost_printk(KERN_INFO, shost, 2175 "scsi_eh_%d: sleeping\n", 2176 shost->host_no)); 2177 schedule(); 2178 continue; 2179 } 2180 2181 __set_current_state(TASK_RUNNING); 2182 SCSI_LOG_ERROR_RECOVERY(1, 2183 shost_printk(KERN_INFO, shost, 2184 "scsi_eh_%d: waking up %d/%d/%d\n", 2185 shost->host_no, shost->host_eh_scheduled, 2186 shost->host_failed, 2187 atomic_read(&shost->host_busy))); 2188 2189 /* 2190 * We have a host that is failing for some reason. Figure out 2191 * what we need to do to get it up and online again (if we can). 2192 * If we fail, we end up taking the thing offline. 2193 */ 2194 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { 2195 SCSI_LOG_ERROR_RECOVERY(1, 2196 printk(KERN_ERR "Error handler scsi_eh_%d " 2197 "unable to autoresume\n", 2198 shost->host_no)); 2199 continue; 2200 } 2201 2202 if (shost->transportt->eh_strategy_handler) 2203 shost->transportt->eh_strategy_handler(shost); 2204 else 2205 scsi_unjam_host(shost); 2206 2207 /* 2208 * Note - if the above fails completely, the action is to take 2209 * individual devices offline and flush the queue of any 2210 * outstanding requests that may have been pending. When we 2211 * restart, we restart any I/O to any other devices on the bus 2212 * which are still online. 2213 */ 2214 scsi_restart_operations(shost); 2215 if (!shost->eh_noresume) 2216 scsi_autopm_put_host(shost); 2217 } 2218 __set_current_state(TASK_RUNNING); 2219 2220 SCSI_LOG_ERROR_RECOVERY(1, 2221 shost_printk(KERN_INFO, shost, 2222 "Error handler scsi_eh_%d exiting\n", 2223 shost->host_no)); 2224 shost->ehandler = NULL; 2225 return 0; 2226 } 2227 2228 /* 2229 * Function: scsi_report_bus_reset() 2230 * 2231 * Purpose: Utility function used by low-level drivers to report that 2232 * they have observed a bus reset on the bus being handled. 2233 * 2234 * Arguments: shost - Host in question 2235 * channel - channel on which reset was observed. 2236 * 2237 * Returns: Nothing 2238 * 2239 * Lock status: Host lock must be held. 2240 * 2241 * Notes: This only needs to be called if the reset is one which 2242 * originates from an unknown location. Resets originated 2243 * by the mid-level itself don't need to call this, but there 2244 * should be no harm. 2245 * 2246 * The main purpose of this is to make sure that a CHECK_CONDITION 2247 * is properly treated. 2248 */ 2249 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2250 { 2251 struct scsi_device *sdev; 2252 2253 __shost_for_each_device(sdev, shost) { 2254 if (channel == sdev_channel(sdev)) 2255 __scsi_report_device_reset(sdev, NULL); 2256 } 2257 } 2258 EXPORT_SYMBOL(scsi_report_bus_reset); 2259 2260 /* 2261 * Function: scsi_report_device_reset() 2262 * 2263 * Purpose: Utility function used by low-level drivers to report that 2264 * they have observed a device reset on the device being handled. 2265 * 2266 * Arguments: shost - Host in question 2267 * channel - channel on which reset was observed 2268 * target - target on which reset was observed 2269 * 2270 * Returns: Nothing 2271 * 2272 * Lock status: Host lock must be held 2273 * 2274 * Notes: This only needs to be called if the reset is one which 2275 * originates from an unknown location. Resets originated 2276 * by the mid-level itself don't need to call this, but there 2277 * should be no harm. 2278 * 2279 * The main purpose of this is to make sure that a CHECK_CONDITION 2280 * is properly treated. 2281 */ 2282 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2283 { 2284 struct scsi_device *sdev; 2285 2286 __shost_for_each_device(sdev, shost) { 2287 if (channel == sdev_channel(sdev) && 2288 target == sdev_id(sdev)) 2289 __scsi_report_device_reset(sdev, NULL); 2290 } 2291 } 2292 EXPORT_SYMBOL(scsi_report_device_reset); 2293 2294 static void 2295 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 2296 { 2297 } 2298 2299 /* 2300 * Function: scsi_reset_provider 2301 * 2302 * Purpose: Send requested reset to a bus or device at any phase. 2303 * 2304 * Arguments: device - device to send reset to 2305 * flag - reset type (see scsi.h) 2306 * 2307 * Returns: SUCCESS/FAILURE. 2308 * 2309 * Notes: This is used by the SCSI Generic driver to provide 2310 * Bus/Device reset capability. 2311 */ 2312 int 2313 scsi_reset_provider(struct scsi_device *dev, int flag) 2314 { 2315 struct scsi_cmnd *scmd; 2316 struct Scsi_Host *shost = dev->host; 2317 struct request req; 2318 unsigned long flags; 2319 int rtn; 2320 2321 if (scsi_autopm_get_host(shost) < 0) 2322 return FAILED; 2323 2324 if (!get_device(&dev->sdev_gendev)) { 2325 rtn = FAILED; 2326 goto out_put_autopm_host; 2327 } 2328 2329 scmd = scsi_get_command(dev, GFP_KERNEL); 2330 if (!scmd) { 2331 rtn = FAILED; 2332 put_device(&dev->sdev_gendev); 2333 goto out_put_autopm_host; 2334 } 2335 2336 blk_rq_init(NULL, &req); 2337 scmd->request = &req; 2338 2339 scmd->cmnd = req.cmd; 2340 2341 scmd->scsi_done = scsi_reset_provider_done_command; 2342 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 2343 2344 scmd->cmd_len = 0; 2345 2346 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 2347 2348 spin_lock_irqsave(shost->host_lock, flags); 2349 shost->tmf_in_progress = 1; 2350 spin_unlock_irqrestore(shost->host_lock, flags); 2351 2352 switch (flag) { 2353 case SCSI_TRY_RESET_DEVICE: 2354 rtn = scsi_try_bus_device_reset(scmd); 2355 if (rtn == SUCCESS) 2356 break; 2357 /* FALLTHROUGH */ 2358 case SCSI_TRY_RESET_TARGET: 2359 rtn = scsi_try_target_reset(scmd); 2360 if (rtn == SUCCESS) 2361 break; 2362 /* FALLTHROUGH */ 2363 case SCSI_TRY_RESET_BUS: 2364 rtn = scsi_try_bus_reset(scmd); 2365 if (rtn == SUCCESS) 2366 break; 2367 /* FALLTHROUGH */ 2368 case SCSI_TRY_RESET_HOST: 2369 rtn = scsi_try_host_reset(scmd); 2370 break; 2371 default: 2372 rtn = FAILED; 2373 } 2374 2375 spin_lock_irqsave(shost->host_lock, flags); 2376 shost->tmf_in_progress = 0; 2377 spin_unlock_irqrestore(shost->host_lock, flags); 2378 2379 /* 2380 * be sure to wake up anyone who was sleeping or had their queue 2381 * suspended while we performed the TMF. 2382 */ 2383 SCSI_LOG_ERROR_RECOVERY(3, 2384 shost_printk(KERN_INFO, shost, 2385 "waking up host to restart after TMF\n")); 2386 2387 wake_up(&shost->host_wait); 2388 2389 scsi_run_host_queues(shost); 2390 2391 scsi_next_command(scmd); 2392 out_put_autopm_host: 2393 scsi_autopm_put_host(shost); 2394 return rtn; 2395 } 2396 EXPORT_SYMBOL(scsi_reset_provider); 2397 2398 /** 2399 * scsi_normalize_sense - normalize main elements from either fixed or 2400 * descriptor sense data format into a common format. 2401 * 2402 * @sense_buffer: byte array containing sense data returned by device 2403 * @sb_len: number of valid bytes in sense_buffer 2404 * @sshdr: pointer to instance of structure that common 2405 * elements are written to. 2406 * 2407 * Notes: 2408 * The "main elements" from sense data are: response_code, sense_key, 2409 * asc, ascq and additional_length (only for descriptor format). 2410 * 2411 * Typically this function can be called after a device has 2412 * responded to a SCSI command with the CHECK_CONDITION status. 2413 * 2414 * Return value: 2415 * 1 if valid sense data information found, else 0; 2416 */ 2417 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len, 2418 struct scsi_sense_hdr *sshdr) 2419 { 2420 if (!sense_buffer || !sb_len) 2421 return 0; 2422 2423 memset(sshdr, 0, sizeof(struct scsi_sense_hdr)); 2424 2425 sshdr->response_code = (sense_buffer[0] & 0x7f); 2426 2427 if (!scsi_sense_valid(sshdr)) 2428 return 0; 2429 2430 if (sshdr->response_code >= 0x72) { 2431 /* 2432 * descriptor format 2433 */ 2434 if (sb_len > 1) 2435 sshdr->sense_key = (sense_buffer[1] & 0xf); 2436 if (sb_len > 2) 2437 sshdr->asc = sense_buffer[2]; 2438 if (sb_len > 3) 2439 sshdr->ascq = sense_buffer[3]; 2440 if (sb_len > 7) 2441 sshdr->additional_length = sense_buffer[7]; 2442 } else { 2443 /* 2444 * fixed format 2445 */ 2446 if (sb_len > 2) 2447 sshdr->sense_key = (sense_buffer[2] & 0xf); 2448 if (sb_len > 7) { 2449 sb_len = (sb_len < (sense_buffer[7] + 8)) ? 2450 sb_len : (sense_buffer[7] + 8); 2451 if (sb_len > 12) 2452 sshdr->asc = sense_buffer[12]; 2453 if (sb_len > 13) 2454 sshdr->ascq = sense_buffer[13]; 2455 } 2456 } 2457 2458 return 1; 2459 } 2460 EXPORT_SYMBOL(scsi_normalize_sense); 2461 2462 int scsi_command_normalize_sense(struct scsi_cmnd *cmd, 2463 struct scsi_sense_hdr *sshdr) 2464 { 2465 return scsi_normalize_sense(cmd->sense_buffer, 2466 SCSI_SENSE_BUFFERSIZE, sshdr); 2467 } 2468 EXPORT_SYMBOL(scsi_command_normalize_sense); 2469 2470 /** 2471 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format. 2472 * @sense_buffer: byte array of descriptor format sense data 2473 * @sb_len: number of valid bytes in sense_buffer 2474 * @desc_type: value of descriptor type to find 2475 * (e.g. 0 -> information) 2476 * 2477 * Notes: 2478 * only valid when sense data is in descriptor format 2479 * 2480 * Return value: 2481 * pointer to start of (first) descriptor if found else NULL 2482 */ 2483 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len, 2484 int desc_type) 2485 { 2486 int add_sen_len, add_len, desc_len, k; 2487 const u8 * descp; 2488 2489 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7]))) 2490 return NULL; 2491 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73)) 2492 return NULL; 2493 add_sen_len = (add_sen_len < (sb_len - 8)) ? 2494 add_sen_len : (sb_len - 8); 2495 descp = &sense_buffer[8]; 2496 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) { 2497 descp += desc_len; 2498 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1; 2499 desc_len = add_len + 2; 2500 if (descp[0] == desc_type) 2501 return descp; 2502 if (add_len < 0) // short descriptor ?? 2503 break; 2504 } 2505 return NULL; 2506 } 2507 EXPORT_SYMBOL(scsi_sense_desc_find); 2508 2509 /** 2510 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2511 * @sense_buffer: byte array of sense data 2512 * @sb_len: number of valid bytes in sense_buffer 2513 * @info_out: pointer to 64 integer where 8 or 4 byte information 2514 * field will be placed if found. 2515 * 2516 * Return value: 2517 * 1 if information field found, 0 if not found. 2518 */ 2519 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len, 2520 u64 * info_out) 2521 { 2522 int j; 2523 const u8 * ucp; 2524 u64 ull; 2525 2526 if (sb_len < 7) 2527 return 0; 2528 switch (sense_buffer[0] & 0x7f) { 2529 case 0x70: 2530 case 0x71: 2531 if (sense_buffer[0] & 0x80) { 2532 *info_out = (sense_buffer[3] << 24) + 2533 (sense_buffer[4] << 16) + 2534 (sense_buffer[5] << 8) + sense_buffer[6]; 2535 return 1; 2536 } else 2537 return 0; 2538 case 0x72: 2539 case 0x73: 2540 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2541 0 /* info desc */); 2542 if (ucp && (0xa == ucp[1])) { 2543 ull = 0; 2544 for (j = 0; j < 8; ++j) { 2545 if (j > 0) 2546 ull <<= 8; 2547 ull |= ucp[4 + j]; 2548 } 2549 *info_out = ull; 2550 return 1; 2551 } else 2552 return 0; 2553 default: 2554 return 0; 2555 } 2556 } 2557 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2558 2559 /** 2560 * scsi_build_sense_buffer - build sense data in a buffer 2561 * @desc: Sense format (non zero == descriptor format, 2562 * 0 == fixed format) 2563 * @buf: Where to build sense data 2564 * @key: Sense key 2565 * @asc: Additional sense code 2566 * @ascq: Additional sense code qualifier 2567 * 2568 **/ 2569 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq) 2570 { 2571 if (desc) { 2572 buf[0] = 0x72; /* descriptor, current */ 2573 buf[1] = key; 2574 buf[2] = asc; 2575 buf[3] = ascq; 2576 buf[7] = 0; 2577 } else { 2578 buf[0] = 0x70; /* fixed, current */ 2579 buf[2] = key; 2580 buf[7] = 0xa; 2581 buf[12] = asc; 2582 buf[13] = ascq; 2583 } 2584 } 2585 EXPORT_SYMBOL(scsi_build_sense_buffer); 2586