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/timer.h> 20 #include <linux/string.h> 21 #include <linux/slab.h> 22 #include <linux/kernel.h> 23 #include <linux/kthread.h> 24 #include <linux/interrupt.h> 25 #include <linux/blkdev.h> 26 #include <linux/delay.h> 27 28 #include <scsi/scsi.h> 29 #include <scsi/scsi_cmnd.h> 30 #include <scsi/scsi_dbg.h> 31 #include <scsi/scsi_device.h> 32 #include <scsi/scsi_eh.h> 33 #include <scsi/scsi_transport.h> 34 #include <scsi/scsi_host.h> 35 #include <scsi/scsi_ioctl.h> 36 37 #include "scsi_priv.h" 38 #include "scsi_logging.h" 39 40 #define SENSE_TIMEOUT (10*HZ) 41 #define START_UNIT_TIMEOUT (30*HZ) 42 43 /* 44 * These should *probably* be handled by the host itself. 45 * Since it is allowed to sleep, it probably should. 46 */ 47 #define BUS_RESET_SETTLE_TIME (10) 48 #define HOST_RESET_SETTLE_TIME (10) 49 50 /* called with shost->host_lock held */ 51 void scsi_eh_wakeup(struct Scsi_Host *shost) 52 { 53 if (shost->host_busy == shost->host_failed) { 54 wake_up_process(shost->ehandler); 55 SCSI_LOG_ERROR_RECOVERY(5, 56 printk("Waking error handler thread\n")); 57 } 58 } 59 60 /** 61 * scsi_schedule_eh - schedule EH for SCSI host 62 * @shost: SCSI host to invoke error handling on. 63 * 64 * Schedule SCSI EH without scmd. 65 **/ 66 void scsi_schedule_eh(struct Scsi_Host *shost) 67 { 68 unsigned long flags; 69 70 spin_lock_irqsave(shost->host_lock, flags); 71 72 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || 73 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { 74 shost->host_eh_scheduled++; 75 scsi_eh_wakeup(shost); 76 } 77 78 spin_unlock_irqrestore(shost->host_lock, flags); 79 } 80 EXPORT_SYMBOL_GPL(scsi_schedule_eh); 81 82 /** 83 * scsi_eh_scmd_add - add scsi cmd to error handling. 84 * @scmd: scmd to run eh on. 85 * @eh_flag: optional SCSI_EH flag. 86 * 87 * Return value: 88 * 0 on failure. 89 **/ 90 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag) 91 { 92 struct Scsi_Host *shost = scmd->device->host; 93 unsigned long flags; 94 int ret = 0; 95 96 if (!shost->ehandler) 97 return 0; 98 99 spin_lock_irqsave(shost->host_lock, flags); 100 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 101 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)) 102 goto out_unlock; 103 104 ret = 1; 105 scmd->eh_eflags |= eh_flag; 106 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 107 shost->host_failed++; 108 scsi_eh_wakeup(shost); 109 out_unlock: 110 spin_unlock_irqrestore(shost->host_lock, flags); 111 return ret; 112 } 113 114 /** 115 * scsi_add_timer - Start timeout timer for a single scsi command. 116 * @scmd: scsi command that is about to start running. 117 * @timeout: amount of time to allow this command to run. 118 * @complete: timeout function to call if timer isn't canceled. 119 * 120 * Notes: 121 * This should be turned into an inline function. Each scsi command 122 * has its own timer, and as it is added to the queue, we set up the 123 * timer. When the command completes, we cancel the timer. 124 **/ 125 void scsi_add_timer(struct scsi_cmnd *scmd, int timeout, 126 void (*complete)(struct scsi_cmnd *)) 127 { 128 129 /* 130 * If the clock was already running for this command, then 131 * first delete the timer. The timer handling code gets rather 132 * confused if we don't do this. 133 */ 134 if (scmd->eh_timeout.function) 135 del_timer(&scmd->eh_timeout); 136 137 scmd->eh_timeout.data = (unsigned long)scmd; 138 scmd->eh_timeout.expires = jiffies + timeout; 139 scmd->eh_timeout.function = (void (*)(unsigned long)) complete; 140 141 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:" 142 " %d, (%p)\n", __FUNCTION__, 143 scmd, timeout, complete)); 144 145 add_timer(&scmd->eh_timeout); 146 } 147 148 /** 149 * scsi_delete_timer - Delete/cancel timer for a given function. 150 * @scmd: Cmd that we are canceling timer for 151 * 152 * Notes: 153 * This should be turned into an inline function. 154 * 155 * Return value: 156 * 1 if we were able to detach the timer. 0 if we blew it, and the 157 * timer function has already started to run. 158 **/ 159 int scsi_delete_timer(struct scsi_cmnd *scmd) 160 { 161 int rtn; 162 163 rtn = del_timer(&scmd->eh_timeout); 164 165 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p," 166 " rtn: %d\n", __FUNCTION__, 167 scmd, rtn)); 168 169 scmd->eh_timeout.data = (unsigned long)NULL; 170 scmd->eh_timeout.function = NULL; 171 172 return rtn; 173 } 174 175 /** 176 * scsi_times_out - Timeout function for normal scsi commands. 177 * @scmd: Cmd that is timing out. 178 * 179 * Notes: 180 * We do not need to lock this. There is the potential for a race 181 * only in that the normal completion handling might run, but if the 182 * normal completion function determines that the timer has already 183 * fired, then it mustn't do anything. 184 **/ 185 void scsi_times_out(struct scsi_cmnd *scmd) 186 { 187 scsi_log_completion(scmd, TIMEOUT_ERROR); 188 189 if (scmd->device->host->transportt->eh_timed_out) 190 switch (scmd->device->host->transportt->eh_timed_out(scmd)) { 191 case EH_HANDLED: 192 __scsi_done(scmd); 193 return; 194 case EH_RESET_TIMER: 195 scsi_add_timer(scmd, scmd->timeout_per_command, 196 scsi_times_out); 197 return; 198 case EH_NOT_HANDLED: 199 break; 200 } 201 202 if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) { 203 scmd->result |= DID_TIME_OUT << 16; 204 __scsi_done(scmd); 205 } 206 } 207 208 /** 209 * scsi_block_when_processing_errors - Prevent cmds from being queued. 210 * @sdev: Device on which we are performing recovery. 211 * 212 * Description: 213 * We block until the host is out of error recovery, and then check to 214 * see whether the host or the device is offline. 215 * 216 * Return value: 217 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 218 **/ 219 int scsi_block_when_processing_errors(struct scsi_device *sdev) 220 { 221 int online; 222 223 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 224 225 online = scsi_device_online(sdev); 226 227 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__, 228 online)); 229 230 return online; 231 } 232 EXPORT_SYMBOL(scsi_block_when_processing_errors); 233 234 #ifdef CONFIG_SCSI_LOGGING 235 /** 236 * scsi_eh_prt_fail_stats - Log info on failures. 237 * @shost: scsi host being recovered. 238 * @work_q: Queue of scsi cmds to process. 239 **/ 240 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 241 struct list_head *work_q) 242 { 243 struct scsi_cmnd *scmd; 244 struct scsi_device *sdev; 245 int total_failures = 0; 246 int cmd_failed = 0; 247 int cmd_cancel = 0; 248 int devices_failed = 0; 249 250 shost_for_each_device(sdev, shost) { 251 list_for_each_entry(scmd, work_q, eh_entry) { 252 if (scmd->device == sdev) { 253 ++total_failures; 254 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) 255 ++cmd_cancel; 256 else 257 ++cmd_failed; 258 } 259 } 260 261 if (cmd_cancel || cmd_failed) { 262 SCSI_LOG_ERROR_RECOVERY(3, 263 sdev_printk(KERN_INFO, sdev, 264 "%s: cmds failed: %d, cancel: %d\n", 265 __FUNCTION__, cmd_failed, 266 cmd_cancel)); 267 cmd_cancel = 0; 268 cmd_failed = 0; 269 ++devices_failed; 270 } 271 } 272 273 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d" 274 " devices require eh work\n", 275 total_failures, devices_failed)); 276 } 277 #endif 278 279 /** 280 * scsi_check_sense - Examine scsi cmd sense 281 * @scmd: Cmd to have sense checked. 282 * 283 * Return value: 284 * SUCCESS or FAILED or NEEDS_RETRY 285 * 286 * Notes: 287 * When a deferred error is detected the current command has 288 * not been executed and needs retrying. 289 **/ 290 static int scsi_check_sense(struct scsi_cmnd *scmd) 291 { 292 struct scsi_sense_hdr sshdr; 293 294 if (! scsi_command_normalize_sense(scmd, &sshdr)) 295 return FAILED; /* no valid sense data */ 296 297 if (scsi_sense_is_deferred(&sshdr)) 298 return NEEDS_RETRY; 299 300 /* 301 * Previous logic looked for FILEMARK, EOM or ILI which are 302 * mainly associated with tapes and returned SUCCESS. 303 */ 304 if (sshdr.response_code == 0x70) { 305 /* fixed format */ 306 if (scmd->sense_buffer[2] & 0xe0) 307 return SUCCESS; 308 } else { 309 /* 310 * descriptor format: look for "stream commands sense data 311 * descriptor" (see SSC-3). Assume single sense data 312 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 313 */ 314 if ((sshdr.additional_length > 3) && 315 (scmd->sense_buffer[8] == 0x4) && 316 (scmd->sense_buffer[11] & 0xe0)) 317 return SUCCESS; 318 } 319 320 switch (sshdr.sense_key) { 321 case NO_SENSE: 322 return SUCCESS; 323 case RECOVERED_ERROR: 324 return /* soft_error */ SUCCESS; 325 326 case ABORTED_COMMAND: 327 return NEEDS_RETRY; 328 case NOT_READY: 329 case UNIT_ATTENTION: 330 /* 331 * if we are expecting a cc/ua because of a bus reset that we 332 * performed, treat this just as a retry. otherwise this is 333 * information that we should pass up to the upper-level driver 334 * so that we can deal with it there. 335 */ 336 if (scmd->device->expecting_cc_ua) { 337 scmd->device->expecting_cc_ua = 0; 338 return NEEDS_RETRY; 339 } 340 /* 341 * if the device is in the process of becoming ready, we 342 * should retry. 343 */ 344 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 345 return NEEDS_RETRY; 346 /* 347 * if the device is not started, we need to wake 348 * the error handler to start the motor 349 */ 350 if (scmd->device->allow_restart && 351 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 352 return FAILED; 353 return SUCCESS; 354 355 /* these three are not supported */ 356 case COPY_ABORTED: 357 case VOLUME_OVERFLOW: 358 case MISCOMPARE: 359 return SUCCESS; 360 361 case MEDIUM_ERROR: 362 return NEEDS_RETRY; 363 364 case HARDWARE_ERROR: 365 if (scmd->device->retry_hwerror) 366 return NEEDS_RETRY; 367 else 368 return SUCCESS; 369 370 case ILLEGAL_REQUEST: 371 case BLANK_CHECK: 372 case DATA_PROTECT: 373 default: 374 return SUCCESS; 375 } 376 } 377 378 /** 379 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 380 * @scmd: SCSI cmd to examine. 381 * 382 * Notes: 383 * This is *only* called when we are examining the status of commands 384 * queued during error recovery. the main difference here is that we 385 * don't allow for the possibility of retries here, and we are a lot 386 * more restrictive about what we consider acceptable. 387 **/ 388 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) 389 { 390 /* 391 * first check the host byte, to see if there is anything in there 392 * that would indicate what we need to do. 393 */ 394 if (host_byte(scmd->result) == DID_RESET) { 395 /* 396 * rats. we are already in the error handler, so we now 397 * get to try and figure out what to do next. if the sense 398 * is valid, we have a pretty good idea of what to do. 399 * if not, we mark it as FAILED. 400 */ 401 return scsi_check_sense(scmd); 402 } 403 if (host_byte(scmd->result) != DID_OK) 404 return FAILED; 405 406 /* 407 * next, check the message byte. 408 */ 409 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 410 return FAILED; 411 412 /* 413 * now, check the status byte to see if this indicates 414 * anything special. 415 */ 416 switch (status_byte(scmd->result)) { 417 case GOOD: 418 case COMMAND_TERMINATED: 419 return SUCCESS; 420 case CHECK_CONDITION: 421 return scsi_check_sense(scmd); 422 case CONDITION_GOOD: 423 case INTERMEDIATE_GOOD: 424 case INTERMEDIATE_C_GOOD: 425 /* 426 * who knows? FIXME(eric) 427 */ 428 return SUCCESS; 429 case BUSY: 430 case QUEUE_FULL: 431 case RESERVATION_CONFLICT: 432 default: 433 return FAILED; 434 } 435 return FAILED; 436 } 437 438 /** 439 * scsi_eh_done - Completion function for error handling. 440 * @scmd: Cmd that is done. 441 **/ 442 static void scsi_eh_done(struct scsi_cmnd *scmd) 443 { 444 struct completion *eh_action; 445 446 SCSI_LOG_ERROR_RECOVERY(3, 447 printk("%s scmd: %p result: %x\n", 448 __FUNCTION__, scmd, scmd->result)); 449 450 eh_action = scmd->device->host->eh_action; 451 if (eh_action) 452 complete(eh_action); 453 } 454 455 /** 456 * scsi_send_eh_cmnd - send a cmd to a device as part of error recovery. 457 * @scmd: SCSI Cmd to send. 458 * @timeout: Timeout for cmd. 459 * 460 * Return value: 461 * SUCCESS or FAILED or NEEDS_RETRY 462 **/ 463 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, int timeout) 464 { 465 struct scsi_device *sdev = scmd->device; 466 struct Scsi_Host *shost = sdev->host; 467 DECLARE_COMPLETION(done); 468 unsigned long timeleft; 469 unsigned long flags; 470 int rtn; 471 472 if (sdev->scsi_level <= SCSI_2) 473 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 474 (sdev->lun << 5 & 0xe0); 475 476 shost->eh_action = &done; 477 478 spin_lock_irqsave(shost->host_lock, flags); 479 scsi_log_send(scmd); 480 shost->hostt->queuecommand(scmd, scsi_eh_done); 481 spin_unlock_irqrestore(shost->host_lock, flags); 482 483 timeleft = wait_for_completion_timeout(&done, timeout); 484 485 shost->eh_action = NULL; 486 487 scsi_log_completion(scmd, SUCCESS); 488 489 SCSI_LOG_ERROR_RECOVERY(3, 490 printk("%s: scmd: %p, timeleft: %ld\n", 491 __FUNCTION__, scmd, timeleft)); 492 493 /* 494 * If there is time left scsi_eh_done got called, and we will 495 * examine the actual status codes to see whether the command 496 * actually did complete normally, else tell the host to forget 497 * about this command. 498 */ 499 if (timeleft) { 500 rtn = scsi_eh_completed_normally(scmd); 501 SCSI_LOG_ERROR_RECOVERY(3, 502 printk("%s: scsi_eh_completed_normally %x\n", 503 __FUNCTION__, rtn)); 504 505 switch (rtn) { 506 case SUCCESS: 507 case NEEDS_RETRY: 508 case FAILED: 509 break; 510 default: 511 rtn = FAILED; 512 break; 513 } 514 } else { 515 /* 516 * FIXME(eric) - we are not tracking whether we could 517 * abort a timed out command or not. not sure how 518 * we should treat them differently anyways. 519 */ 520 if (shost->hostt->eh_abort_handler) 521 shost->hostt->eh_abort_handler(scmd); 522 rtn = FAILED; 523 } 524 525 return rtn; 526 } 527 528 /** 529 * scsi_request_sense - Request sense data from a particular target. 530 * @scmd: SCSI cmd for request sense. 531 * 532 * Notes: 533 * Some hosts automatically obtain this information, others require 534 * that we obtain it on our own. This function will *not* return until 535 * the command either times out, or it completes. 536 **/ 537 static int scsi_request_sense(struct scsi_cmnd *scmd) 538 { 539 static unsigned char generic_sense[6] = 540 {REQUEST_SENSE, 0, 0, 0, 252, 0}; 541 unsigned char *scsi_result; 542 int saved_result; 543 int rtn; 544 545 memcpy(scmd->cmnd, generic_sense, sizeof(generic_sense)); 546 547 scsi_result = kmalloc(252, GFP_ATOMIC | ((scmd->device->host->hostt->unchecked_isa_dma) ? __GFP_DMA : 0)); 548 549 550 if (unlikely(!scsi_result)) { 551 printk(KERN_ERR "%s: cannot allocate scsi_result.\n", 552 __FUNCTION__); 553 return FAILED; 554 } 555 556 /* 557 * zero the sense buffer. some host adapters automatically always 558 * request sense, so it is not a good idea that 559 * scmd->request_buffer and scmd->sense_buffer point to the same 560 * address (db). 0 is not a valid sense code. 561 */ 562 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); 563 memset(scsi_result, 0, 252); 564 565 saved_result = scmd->result; 566 scmd->request_buffer = scsi_result; 567 scmd->request_bufflen = 252; 568 scmd->use_sg = 0; 569 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 570 scmd->sc_data_direction = DMA_FROM_DEVICE; 571 scmd->underflow = 0; 572 573 rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT); 574 575 /* last chance to have valid sense data */ 576 if(!SCSI_SENSE_VALID(scmd)) { 577 memcpy(scmd->sense_buffer, scmd->request_buffer, 578 sizeof(scmd->sense_buffer)); 579 } 580 581 kfree(scsi_result); 582 583 /* 584 * when we eventually call scsi_finish, we really wish to complete 585 * the original request, so let's restore the original data. (db) 586 */ 587 scsi_setup_cmd_retry(scmd); 588 scmd->result = saved_result; 589 return rtn; 590 } 591 592 /** 593 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 594 * @scmd: Original SCSI cmd that eh has finished. 595 * @done_q: Queue for processed commands. 596 * 597 * Notes: 598 * We don't want to use the normal command completion while we are are 599 * still handling errors - it may cause other commands to be queued, 600 * and that would disturb what we are doing. thus we really want to 601 * keep a list of pending commands for final completion, and once we 602 * are ready to leave error handling we handle completion for real. 603 **/ 604 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 605 { 606 scmd->device->host->host_failed--; 607 scmd->eh_eflags = 0; 608 609 /* 610 * set this back so that the upper level can correctly free up 611 * things. 612 */ 613 scsi_setup_cmd_retry(scmd); 614 list_move_tail(&scmd->eh_entry, done_q); 615 } 616 EXPORT_SYMBOL(scsi_eh_finish_cmd); 617 618 /** 619 * scsi_eh_get_sense - Get device sense data. 620 * @work_q: Queue of commands to process. 621 * @done_q: Queue of proccessed commands.. 622 * 623 * Description: 624 * See if we need to request sense information. if so, then get it 625 * now, so we have a better idea of what to do. 626 * 627 * Notes: 628 * This has the unfortunate side effect that if a shost adapter does 629 * not automatically request sense information, that we end up shutting 630 * it down before we request it. 631 * 632 * All drivers should request sense information internally these days, 633 * so for now all I have to say is tough noogies if you end up in here. 634 * 635 * XXX: Long term this code should go away, but that needs an audit of 636 * all LLDDs first. 637 **/ 638 static int scsi_eh_get_sense(struct list_head *work_q, 639 struct list_head *done_q) 640 { 641 struct scsi_cmnd *scmd, *next; 642 int rtn; 643 644 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 645 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) || 646 SCSI_SENSE_VALID(scmd)) 647 continue; 648 649 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 650 "%s: requesting sense\n", 651 current->comm)); 652 rtn = scsi_request_sense(scmd); 653 if (rtn != SUCCESS) 654 continue; 655 656 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p" 657 " result %x\n", scmd, 658 scmd->result)); 659 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd)); 660 661 rtn = scsi_decide_disposition(scmd); 662 663 /* 664 * if the result was normal, then just pass it along to the 665 * upper level. 666 */ 667 if (rtn == SUCCESS) 668 /* we don't want this command reissued, just 669 * finished with the sense data, so set 670 * retries to the max allowed to ensure it 671 * won't get reissued */ 672 scmd->retries = scmd->allowed; 673 else if (rtn != NEEDS_RETRY) 674 continue; 675 676 scsi_eh_finish_cmd(scmd, done_q); 677 } 678 679 return list_empty(work_q); 680 } 681 682 /** 683 * scsi_try_to_abort_cmd - Ask host to abort a running command. 684 * @scmd: SCSI cmd to abort from Lower Level. 685 * 686 * Notes: 687 * This function will not return until the user's completion function 688 * has been called. there is no timeout on this operation. if the 689 * author of the low-level driver wishes this operation to be timed, 690 * they can provide this facility themselves. helper functions in 691 * scsi_error.c can be supplied to make this easier to do. 692 **/ 693 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd) 694 { 695 if (!scmd->device->host->hostt->eh_abort_handler) 696 return FAILED; 697 698 /* 699 * scsi_done was called just after the command timed out and before 700 * we had a chance to process it. (db) 701 */ 702 if (scmd->serial_number == 0) 703 return SUCCESS; 704 return scmd->device->host->hostt->eh_abort_handler(scmd); 705 } 706 707 /** 708 * scsi_eh_tur - Send TUR to device. 709 * @scmd: Scsi cmd to send TUR 710 * 711 * Return value: 712 * 0 - Device is ready. 1 - Device NOT ready. 713 **/ 714 static int scsi_eh_tur(struct scsi_cmnd *scmd) 715 { 716 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 717 int retry_cnt = 1, rtn; 718 int saved_result; 719 720 retry_tur: 721 memcpy(scmd->cmnd, tur_command, sizeof(tur_command)); 722 723 /* 724 * zero the sense buffer. the scsi spec mandates that any 725 * untransferred sense data should be interpreted as being zero. 726 */ 727 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); 728 729 saved_result = scmd->result; 730 scmd->request_buffer = NULL; 731 scmd->request_bufflen = 0; 732 scmd->use_sg = 0; 733 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 734 scmd->underflow = 0; 735 scmd->sc_data_direction = DMA_NONE; 736 737 rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT); 738 739 /* 740 * when we eventually call scsi_finish, we really wish to complete 741 * the original request, so let's restore the original data. (db) 742 */ 743 scsi_setup_cmd_retry(scmd); 744 scmd->result = saved_result; 745 746 /* 747 * hey, we are done. let's look to see what happened. 748 */ 749 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", 750 __FUNCTION__, scmd, rtn)); 751 if (rtn == SUCCESS) 752 return 0; 753 else if (rtn == NEEDS_RETRY) { 754 if (retry_cnt--) 755 goto retry_tur; 756 return 0; 757 } 758 return 1; 759 } 760 761 /** 762 * scsi_eh_abort_cmds - abort canceled commands. 763 * @shost: scsi host being recovered. 764 * @eh_done_q: list_head for processed commands. 765 * 766 * Decription: 767 * Try and see whether or not it makes sense to try and abort the 768 * running command. this only works out to be the case if we have one 769 * command that has timed out. if the command simply failed, it makes 770 * no sense to try and abort the command, since as far as the shost 771 * adapter is concerned, it isn't running. 772 **/ 773 static int scsi_eh_abort_cmds(struct list_head *work_q, 774 struct list_head *done_q) 775 { 776 struct scsi_cmnd *scmd, *next; 777 int rtn; 778 779 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 780 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD)) 781 continue; 782 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:" 783 "0x%p\n", current->comm, 784 scmd)); 785 rtn = scsi_try_to_abort_cmd(scmd); 786 if (rtn == SUCCESS) { 787 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD; 788 if (!scsi_device_online(scmd->device) || 789 !scsi_eh_tur(scmd)) { 790 scsi_eh_finish_cmd(scmd, done_q); 791 } 792 793 } else 794 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting" 795 " cmd failed:" 796 "0x%p\n", 797 current->comm, 798 scmd)); 799 } 800 801 return list_empty(work_q); 802 } 803 804 /** 805 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 806 * @scmd: SCSI cmd used to send BDR 807 * 808 * Notes: 809 * There is no timeout for this operation. if this operation is 810 * unreliable for a given host, then the host itself needs to put a 811 * timer on it, and set the host back to a consistent state prior to 812 * returning. 813 **/ 814 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 815 { 816 int rtn; 817 818 if (!scmd->device->host->hostt->eh_device_reset_handler) 819 return FAILED; 820 821 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd); 822 if (rtn == SUCCESS) { 823 scmd->device->was_reset = 1; 824 scmd->device->expecting_cc_ua = 1; 825 } 826 827 return rtn; 828 } 829 830 /** 831 * scsi_eh_try_stu - Send START_UNIT to device. 832 * @scmd: Scsi cmd to send START_UNIT 833 * 834 * Return value: 835 * 0 - Device is ready. 1 - Device NOT ready. 836 **/ 837 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 838 { 839 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 840 int rtn; 841 int saved_result; 842 843 if (!scmd->device->allow_restart) 844 return 1; 845 846 memcpy(scmd->cmnd, stu_command, sizeof(stu_command)); 847 848 /* 849 * zero the sense buffer. the scsi spec mandates that any 850 * untransferred sense data should be interpreted as being zero. 851 */ 852 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); 853 854 saved_result = scmd->result; 855 scmd->request_buffer = NULL; 856 scmd->request_bufflen = 0; 857 scmd->use_sg = 0; 858 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 859 scmd->underflow = 0; 860 scmd->sc_data_direction = DMA_NONE; 861 862 rtn = scsi_send_eh_cmnd(scmd, START_UNIT_TIMEOUT); 863 864 /* 865 * when we eventually call scsi_finish, we really wish to complete 866 * the original request, so let's restore the original data. (db) 867 */ 868 scsi_setup_cmd_retry(scmd); 869 scmd->result = saved_result; 870 871 /* 872 * hey, we are done. let's look to see what happened. 873 */ 874 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", 875 __FUNCTION__, scmd, rtn)); 876 if (rtn == SUCCESS) 877 return 0; 878 return 1; 879 } 880 881 /** 882 * scsi_eh_stu - send START_UNIT if needed 883 * @shost: scsi host being recovered. 884 * @eh_done_q: list_head for processed commands. 885 * 886 * Notes: 887 * If commands are failing due to not ready, initializing command required, 888 * try revalidating the device, which will end up sending a start unit. 889 **/ 890 static int scsi_eh_stu(struct Scsi_Host *shost, 891 struct list_head *work_q, 892 struct list_head *done_q) 893 { 894 struct scsi_cmnd *scmd, *stu_scmd, *next; 895 struct scsi_device *sdev; 896 897 shost_for_each_device(sdev, shost) { 898 stu_scmd = NULL; 899 list_for_each_entry(scmd, work_q, eh_entry) 900 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 901 scsi_check_sense(scmd) == FAILED ) { 902 stu_scmd = scmd; 903 break; 904 } 905 906 if (!stu_scmd) 907 continue; 908 909 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:" 910 " 0x%p\n", current->comm, sdev)); 911 912 if (!scsi_eh_try_stu(stu_scmd)) { 913 if (!scsi_device_online(sdev) || 914 !scsi_eh_tur(stu_scmd)) { 915 list_for_each_entry_safe(scmd, next, 916 work_q, eh_entry) { 917 if (scmd->device == sdev) 918 scsi_eh_finish_cmd(scmd, done_q); 919 } 920 } 921 } else { 922 SCSI_LOG_ERROR_RECOVERY(3, 923 printk("%s: START_UNIT failed to sdev:" 924 " 0x%p\n", current->comm, sdev)); 925 } 926 } 927 928 return list_empty(work_q); 929 } 930 931 932 /** 933 * scsi_eh_bus_device_reset - send bdr if needed 934 * @shost: scsi host being recovered. 935 * @eh_done_q: list_head for processed commands. 936 * 937 * Notes: 938 * Try a bus device reset. still, look to see whether we have multiple 939 * devices that are jammed or not - if we have multiple devices, it 940 * makes no sense to try bus_device_reset - we really would need to try 941 * a bus_reset instead. 942 **/ 943 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 944 struct list_head *work_q, 945 struct list_head *done_q) 946 { 947 struct scsi_cmnd *scmd, *bdr_scmd, *next; 948 struct scsi_device *sdev; 949 int rtn; 950 951 shost_for_each_device(sdev, shost) { 952 bdr_scmd = NULL; 953 list_for_each_entry(scmd, work_q, eh_entry) 954 if (scmd->device == sdev) { 955 bdr_scmd = scmd; 956 break; 957 } 958 959 if (!bdr_scmd) 960 continue; 961 962 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:" 963 " 0x%p\n", current->comm, 964 sdev)); 965 rtn = scsi_try_bus_device_reset(bdr_scmd); 966 if (rtn == SUCCESS) { 967 if (!scsi_device_online(sdev) || 968 !scsi_eh_tur(bdr_scmd)) { 969 list_for_each_entry_safe(scmd, next, 970 work_q, eh_entry) { 971 if (scmd->device == sdev) 972 scsi_eh_finish_cmd(scmd, 973 done_q); 974 } 975 } 976 } else { 977 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR" 978 " failed sdev:" 979 "0x%p\n", 980 current->comm, 981 sdev)); 982 } 983 } 984 985 return list_empty(work_q); 986 } 987 988 /** 989 * scsi_try_bus_reset - ask host to perform a bus reset 990 * @scmd: SCSI cmd to send bus reset. 991 **/ 992 static int scsi_try_bus_reset(struct scsi_cmnd *scmd) 993 { 994 unsigned long flags; 995 int rtn; 996 997 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n", 998 __FUNCTION__)); 999 1000 if (!scmd->device->host->hostt->eh_bus_reset_handler) 1001 return FAILED; 1002 1003 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd); 1004 1005 if (rtn == SUCCESS) { 1006 if (!scmd->device->host->hostt->skip_settle_delay) 1007 ssleep(BUS_RESET_SETTLE_TIME); 1008 spin_lock_irqsave(scmd->device->host->host_lock, flags); 1009 scsi_report_bus_reset(scmd->device->host, 1010 scmd_channel(scmd)); 1011 spin_unlock_irqrestore(scmd->device->host->host_lock, flags); 1012 } 1013 1014 return rtn; 1015 } 1016 1017 /** 1018 * scsi_try_host_reset - ask host adapter to reset itself 1019 * @scmd: SCSI cmd to send hsot reset. 1020 **/ 1021 static int scsi_try_host_reset(struct scsi_cmnd *scmd) 1022 { 1023 unsigned long flags; 1024 int rtn; 1025 1026 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n", 1027 __FUNCTION__)); 1028 1029 if (!scmd->device->host->hostt->eh_host_reset_handler) 1030 return FAILED; 1031 1032 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd); 1033 1034 if (rtn == SUCCESS) { 1035 if (!scmd->device->host->hostt->skip_settle_delay) 1036 ssleep(HOST_RESET_SETTLE_TIME); 1037 spin_lock_irqsave(scmd->device->host->host_lock, flags); 1038 scsi_report_bus_reset(scmd->device->host, 1039 scmd_channel(scmd)); 1040 spin_unlock_irqrestore(scmd->device->host->host_lock, flags); 1041 } 1042 1043 return rtn; 1044 } 1045 1046 /** 1047 * scsi_eh_bus_reset - send a bus reset 1048 * @shost: scsi host being recovered. 1049 * @eh_done_q: list_head for processed commands. 1050 **/ 1051 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1052 struct list_head *work_q, 1053 struct list_head *done_q) 1054 { 1055 struct scsi_cmnd *scmd, *chan_scmd, *next; 1056 unsigned int channel; 1057 int rtn; 1058 1059 /* 1060 * we really want to loop over the various channels, and do this on 1061 * a channel by channel basis. we should also check to see if any 1062 * of the failed commands are on soft_reset devices, and if so, skip 1063 * the reset. 1064 */ 1065 1066 for (channel = 0; channel <= shost->max_channel; channel++) { 1067 chan_scmd = NULL; 1068 list_for_each_entry(scmd, work_q, eh_entry) { 1069 if (channel == scmd_channel(scmd)) { 1070 chan_scmd = scmd; 1071 break; 1072 /* 1073 * FIXME add back in some support for 1074 * soft_reset devices. 1075 */ 1076 } 1077 } 1078 1079 if (!chan_scmd) 1080 continue; 1081 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:" 1082 " %d\n", current->comm, 1083 channel)); 1084 rtn = scsi_try_bus_reset(chan_scmd); 1085 if (rtn == SUCCESS) { 1086 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1087 if (channel == scmd_channel(scmd)) 1088 if (!scsi_device_online(scmd->device) || 1089 !scsi_eh_tur(scmd)) 1090 scsi_eh_finish_cmd(scmd, 1091 done_q); 1092 } 1093 } else { 1094 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST" 1095 " failed chan: %d\n", 1096 current->comm, 1097 channel)); 1098 } 1099 } 1100 return list_empty(work_q); 1101 } 1102 1103 /** 1104 * scsi_eh_host_reset - send a host reset 1105 * @work_q: list_head for processed commands. 1106 * @done_q: list_head for processed commands. 1107 **/ 1108 static int scsi_eh_host_reset(struct list_head *work_q, 1109 struct list_head *done_q) 1110 { 1111 struct scsi_cmnd *scmd, *next; 1112 int rtn; 1113 1114 if (!list_empty(work_q)) { 1115 scmd = list_entry(work_q->next, 1116 struct scsi_cmnd, eh_entry); 1117 1118 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n" 1119 , current->comm)); 1120 1121 rtn = scsi_try_host_reset(scmd); 1122 if (rtn == SUCCESS) { 1123 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1124 if (!scsi_device_online(scmd->device) || 1125 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) || 1126 !scsi_eh_tur(scmd)) 1127 scsi_eh_finish_cmd(scmd, done_q); 1128 } 1129 } else { 1130 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST" 1131 " failed\n", 1132 current->comm)); 1133 } 1134 } 1135 return list_empty(work_q); 1136 } 1137 1138 /** 1139 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1140 * @work_q: list_head for processed commands. 1141 * @done_q: list_head for processed commands. 1142 * 1143 **/ 1144 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1145 struct list_head *done_q) 1146 { 1147 struct scsi_cmnd *scmd, *next; 1148 1149 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1150 sdev_printk(KERN_INFO, scmd->device, 1151 "scsi: Device offlined - not" 1152 " ready after error recovery\n"); 1153 scsi_device_set_state(scmd->device, SDEV_OFFLINE); 1154 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) { 1155 /* 1156 * FIXME: Handle lost cmds. 1157 */ 1158 } 1159 scsi_eh_finish_cmd(scmd, done_q); 1160 } 1161 return; 1162 } 1163 1164 /** 1165 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1166 * @scmd: SCSI cmd to examine. 1167 * 1168 * Notes: 1169 * This is *only* called when we are examining the status after sending 1170 * out the actual data command. any commands that are queued for error 1171 * recovery (e.g. test_unit_ready) do *not* come through here. 1172 * 1173 * When this routine returns failed, it means the error handler thread 1174 * is woken. In cases where the error code indicates an error that 1175 * doesn't require the error handler read (i.e. we don't need to 1176 * abort/reset), this function should return SUCCESS. 1177 **/ 1178 int scsi_decide_disposition(struct scsi_cmnd *scmd) 1179 { 1180 int rtn; 1181 1182 /* 1183 * if the device is offline, then we clearly just pass the result back 1184 * up to the top level. 1185 */ 1186 if (!scsi_device_online(scmd->device)) { 1187 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report" 1188 " as SUCCESS\n", 1189 __FUNCTION__)); 1190 return SUCCESS; 1191 } 1192 1193 /* 1194 * first check the host byte, to see if there is anything in there 1195 * that would indicate what we need to do. 1196 */ 1197 switch (host_byte(scmd->result)) { 1198 case DID_PASSTHROUGH: 1199 /* 1200 * no matter what, pass this through to the upper layer. 1201 * nuke this special code so that it looks like we are saying 1202 * did_ok. 1203 */ 1204 scmd->result &= 0xff00ffff; 1205 return SUCCESS; 1206 case DID_OK: 1207 /* 1208 * looks good. drop through, and check the next byte. 1209 */ 1210 break; 1211 case DID_NO_CONNECT: 1212 case DID_BAD_TARGET: 1213 case DID_ABORT: 1214 /* 1215 * note - this means that we just report the status back 1216 * to the top level driver, not that we actually think 1217 * that it indicates SUCCESS. 1218 */ 1219 return SUCCESS; 1220 /* 1221 * when the low level driver returns did_soft_error, 1222 * it is responsible for keeping an internal retry counter 1223 * in order to avoid endless loops (db) 1224 * 1225 * actually this is a bug in this function here. we should 1226 * be mindful of the maximum number of retries specified 1227 * and not get stuck in a loop. 1228 */ 1229 case DID_SOFT_ERROR: 1230 goto maybe_retry; 1231 case DID_IMM_RETRY: 1232 return NEEDS_RETRY; 1233 1234 case DID_REQUEUE: 1235 return ADD_TO_MLQUEUE; 1236 1237 case DID_ERROR: 1238 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1239 status_byte(scmd->result) == RESERVATION_CONFLICT) 1240 /* 1241 * execute reservation conflict processing code 1242 * lower down 1243 */ 1244 break; 1245 /* fallthrough */ 1246 1247 case DID_BUS_BUSY: 1248 case DID_PARITY: 1249 goto maybe_retry; 1250 case DID_TIME_OUT: 1251 /* 1252 * when we scan the bus, we get timeout messages for 1253 * these commands if there is no device available. 1254 * other hosts report did_no_connect for the same thing. 1255 */ 1256 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1257 scmd->cmnd[0] == INQUIRY)) { 1258 return SUCCESS; 1259 } else { 1260 return FAILED; 1261 } 1262 case DID_RESET: 1263 return SUCCESS; 1264 default: 1265 return FAILED; 1266 } 1267 1268 /* 1269 * next, check the message byte. 1270 */ 1271 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 1272 return FAILED; 1273 1274 /* 1275 * check the status byte to see if this indicates anything special. 1276 */ 1277 switch (status_byte(scmd->result)) { 1278 case QUEUE_FULL: 1279 /* 1280 * the case of trying to send too many commands to a 1281 * tagged queueing device. 1282 */ 1283 case BUSY: 1284 /* 1285 * device can't talk to us at the moment. Should only 1286 * occur (SAM-3) when the task queue is empty, so will cause 1287 * the empty queue handling to trigger a stall in the 1288 * device. 1289 */ 1290 return ADD_TO_MLQUEUE; 1291 case GOOD: 1292 case COMMAND_TERMINATED: 1293 case TASK_ABORTED: 1294 return SUCCESS; 1295 case CHECK_CONDITION: 1296 rtn = scsi_check_sense(scmd); 1297 if (rtn == NEEDS_RETRY) 1298 goto maybe_retry; 1299 /* if rtn == FAILED, we have no sense information; 1300 * returning FAILED will wake the error handler thread 1301 * to collect the sense and redo the decide 1302 * disposition */ 1303 return rtn; 1304 case CONDITION_GOOD: 1305 case INTERMEDIATE_GOOD: 1306 case INTERMEDIATE_C_GOOD: 1307 case ACA_ACTIVE: 1308 /* 1309 * who knows? FIXME(eric) 1310 */ 1311 return SUCCESS; 1312 1313 case RESERVATION_CONFLICT: 1314 sdev_printk(KERN_INFO, scmd->device, 1315 "reservation conflict\n"); 1316 return SUCCESS; /* causes immediate i/o error */ 1317 default: 1318 return FAILED; 1319 } 1320 return FAILED; 1321 1322 maybe_retry: 1323 1324 /* we requeue for retry because the error was retryable, and 1325 * the request was not marked fast fail. Note that above, 1326 * even if the request is marked fast fail, we still requeue 1327 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1328 if ((++scmd->retries) <= scmd->allowed 1329 && !blk_noretry_request(scmd->request)) { 1330 return NEEDS_RETRY; 1331 } else { 1332 /* 1333 * no more retries - report this one back to upper level. 1334 */ 1335 return SUCCESS; 1336 } 1337 } 1338 1339 /** 1340 * scsi_eh_lock_door - Prevent medium removal for the specified device 1341 * @sdev: SCSI device to prevent medium removal 1342 * 1343 * Locking: 1344 * We must be called from process context; scsi_allocate_request() 1345 * may sleep. 1346 * 1347 * Notes: 1348 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 1349 * head of the devices request queue, and continue. 1350 * 1351 * Bugs: 1352 * scsi_allocate_request() may sleep waiting for existing requests to 1353 * be processed. However, since we haven't kicked off any request 1354 * processing for this host, this may deadlock. 1355 * 1356 * If scsi_allocate_request() fails for what ever reason, we 1357 * completely forget to lock the door. 1358 **/ 1359 static void scsi_eh_lock_door(struct scsi_device *sdev) 1360 { 1361 unsigned char cmnd[MAX_COMMAND_SIZE]; 1362 1363 cmnd[0] = ALLOW_MEDIUM_REMOVAL; 1364 cmnd[1] = 0; 1365 cmnd[2] = 0; 1366 cmnd[3] = 0; 1367 cmnd[4] = SCSI_REMOVAL_PREVENT; 1368 cmnd[5] = 0; 1369 1370 scsi_execute_async(sdev, cmnd, 6, DMA_NONE, NULL, 0, 0, 10 * HZ, 1371 5, NULL, NULL, GFP_KERNEL); 1372 } 1373 1374 1375 /** 1376 * scsi_restart_operations - restart io operations to the specified host. 1377 * @shost: Host we are restarting. 1378 * 1379 * Notes: 1380 * When we entered the error handler, we blocked all further i/o to 1381 * this device. we need to 'reverse' this process. 1382 **/ 1383 static void scsi_restart_operations(struct Scsi_Host *shost) 1384 { 1385 struct scsi_device *sdev; 1386 unsigned long flags; 1387 1388 /* 1389 * If the door was locked, we need to insert a door lock request 1390 * onto the head of the SCSI request queue for the device. There 1391 * is no point trying to lock the door of an off-line device. 1392 */ 1393 shost_for_each_device(sdev, shost) { 1394 if (scsi_device_online(sdev) && sdev->locked) 1395 scsi_eh_lock_door(sdev); 1396 } 1397 1398 /* 1399 * next free up anything directly waiting upon the host. this 1400 * will be requests for character device operations, and also for 1401 * ioctls to queued block devices. 1402 */ 1403 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n", 1404 __FUNCTION__)); 1405 1406 spin_lock_irqsave(shost->host_lock, flags); 1407 if (scsi_host_set_state(shost, SHOST_RUNNING)) 1408 if (scsi_host_set_state(shost, SHOST_CANCEL)) 1409 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 1410 spin_unlock_irqrestore(shost->host_lock, flags); 1411 1412 wake_up(&shost->host_wait); 1413 1414 /* 1415 * finally we need to re-initiate requests that may be pending. we will 1416 * have had everything blocked while error handling is taking place, and 1417 * now that error recovery is done, we will need to ensure that these 1418 * requests are started. 1419 */ 1420 scsi_run_host_queues(shost); 1421 } 1422 1423 /** 1424 * scsi_eh_ready_devs - check device ready state and recover if not. 1425 * @shost: host to be recovered. 1426 * @eh_done_q: list_head for processed commands. 1427 * 1428 **/ 1429 static void scsi_eh_ready_devs(struct Scsi_Host *shost, 1430 struct list_head *work_q, 1431 struct list_head *done_q) 1432 { 1433 if (!scsi_eh_stu(shost, work_q, done_q)) 1434 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 1435 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 1436 if (!scsi_eh_host_reset(work_q, done_q)) 1437 scsi_eh_offline_sdevs(work_q, done_q); 1438 } 1439 1440 /** 1441 * scsi_eh_flush_done_q - finish processed commands or retry them. 1442 * @done_q: list_head of processed commands. 1443 * 1444 **/ 1445 void scsi_eh_flush_done_q(struct list_head *done_q) 1446 { 1447 struct scsi_cmnd *scmd, *next; 1448 1449 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 1450 list_del_init(&scmd->eh_entry); 1451 if (scsi_device_online(scmd->device) && 1452 !blk_noretry_request(scmd->request) && 1453 (++scmd->retries <= scmd->allowed)) { 1454 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush" 1455 " retry cmd: %p\n", 1456 current->comm, 1457 scmd)); 1458 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 1459 } else { 1460 /* 1461 * If just we got sense for the device (called 1462 * scsi_eh_get_sense), scmd->result is already 1463 * set, do not set DRIVER_TIMEOUT. 1464 */ 1465 if (!scmd->result) 1466 scmd->result |= (DRIVER_TIMEOUT << 24); 1467 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish" 1468 " cmd: %p\n", 1469 current->comm, scmd)); 1470 scsi_finish_command(scmd); 1471 } 1472 } 1473 } 1474 EXPORT_SYMBOL(scsi_eh_flush_done_q); 1475 1476 /** 1477 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 1478 * @shost: Host to unjam. 1479 * 1480 * Notes: 1481 * When we come in here, we *know* that all commands on the bus have 1482 * either completed, failed or timed out. we also know that no further 1483 * commands are being sent to the host, so things are relatively quiet 1484 * and we have freedom to fiddle with things as we wish. 1485 * 1486 * This is only the *default* implementation. it is possible for 1487 * individual drivers to supply their own version of this function, and 1488 * if the maintainer wishes to do this, it is strongly suggested that 1489 * this function be taken as a template and modified. this function 1490 * was designed to correctly handle problems for about 95% of the 1491 * different cases out there, and it should always provide at least a 1492 * reasonable amount of error recovery. 1493 * 1494 * Any command marked 'failed' or 'timeout' must eventually have 1495 * scsi_finish_cmd() called for it. we do all of the retry stuff 1496 * here, so when we restart the host after we return it should have an 1497 * empty queue. 1498 **/ 1499 static void scsi_unjam_host(struct Scsi_Host *shost) 1500 { 1501 unsigned long flags; 1502 LIST_HEAD(eh_work_q); 1503 LIST_HEAD(eh_done_q); 1504 1505 spin_lock_irqsave(shost->host_lock, flags); 1506 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 1507 spin_unlock_irqrestore(shost->host_lock, flags); 1508 1509 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 1510 1511 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 1512 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q)) 1513 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 1514 1515 scsi_eh_flush_done_q(&eh_done_q); 1516 } 1517 1518 /** 1519 * scsi_error_handler - SCSI error handler thread 1520 * @data: Host for which we are running. 1521 * 1522 * Notes: 1523 * This is the main error handling loop. This is run as a kernel thread 1524 * for every SCSI host and handles all error handling activity. 1525 **/ 1526 int scsi_error_handler(void *data) 1527 { 1528 struct Scsi_Host *shost = data; 1529 1530 current->flags |= PF_NOFREEZE; 1531 1532 /* 1533 * We use TASK_INTERRUPTIBLE so that the thread is not 1534 * counted against the load average as a running process. 1535 * We never actually get interrupted because kthread_run 1536 * disables singal delivery for the created thread. 1537 */ 1538 set_current_state(TASK_INTERRUPTIBLE); 1539 while (!kthread_should_stop()) { 1540 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 1541 shost->host_failed != shost->host_busy) { 1542 SCSI_LOG_ERROR_RECOVERY(1, 1543 printk("Error handler scsi_eh_%d sleeping\n", 1544 shost->host_no)); 1545 schedule(); 1546 set_current_state(TASK_INTERRUPTIBLE); 1547 continue; 1548 } 1549 1550 __set_current_state(TASK_RUNNING); 1551 SCSI_LOG_ERROR_RECOVERY(1, 1552 printk("Error handler scsi_eh_%d waking up\n", 1553 shost->host_no)); 1554 1555 /* 1556 * We have a host that is failing for some reason. Figure out 1557 * what we need to do to get it up and online again (if we can). 1558 * If we fail, we end up taking the thing offline. 1559 */ 1560 if (shost->transportt->eh_strategy_handler) 1561 shost->transportt->eh_strategy_handler(shost); 1562 else 1563 scsi_unjam_host(shost); 1564 1565 /* 1566 * Note - if the above fails completely, the action is to take 1567 * individual devices offline and flush the queue of any 1568 * outstanding requests that may have been pending. When we 1569 * restart, we restart any I/O to any other devices on the bus 1570 * which are still online. 1571 */ 1572 scsi_restart_operations(shost); 1573 set_current_state(TASK_INTERRUPTIBLE); 1574 } 1575 __set_current_state(TASK_RUNNING); 1576 1577 SCSI_LOG_ERROR_RECOVERY(1, 1578 printk("Error handler scsi_eh_%d exiting\n", shost->host_no)); 1579 shost->ehandler = NULL; 1580 return 0; 1581 } 1582 1583 /* 1584 * Function: scsi_report_bus_reset() 1585 * 1586 * Purpose: Utility function used by low-level drivers to report that 1587 * they have observed a bus reset on the bus being handled. 1588 * 1589 * Arguments: shost - Host in question 1590 * channel - channel on which reset was observed. 1591 * 1592 * Returns: Nothing 1593 * 1594 * Lock status: Host lock must be held. 1595 * 1596 * Notes: This only needs to be called if the reset is one which 1597 * originates from an unknown location. Resets originated 1598 * by the mid-level itself don't need to call this, but there 1599 * should be no harm. 1600 * 1601 * The main purpose of this is to make sure that a CHECK_CONDITION 1602 * is properly treated. 1603 */ 1604 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 1605 { 1606 struct scsi_device *sdev; 1607 1608 __shost_for_each_device(sdev, shost) { 1609 if (channel == sdev_channel(sdev)) { 1610 sdev->was_reset = 1; 1611 sdev->expecting_cc_ua = 1; 1612 } 1613 } 1614 } 1615 EXPORT_SYMBOL(scsi_report_bus_reset); 1616 1617 /* 1618 * Function: scsi_report_device_reset() 1619 * 1620 * Purpose: Utility function used by low-level drivers to report that 1621 * they have observed a device reset on the device being handled. 1622 * 1623 * Arguments: shost - Host in question 1624 * channel - channel on which reset was observed 1625 * target - target on which reset was observed 1626 * 1627 * Returns: Nothing 1628 * 1629 * Lock status: Host lock must be held 1630 * 1631 * Notes: This only needs to be called if the reset is one which 1632 * originates from an unknown location. Resets originated 1633 * by the mid-level itself don't need to call this, but there 1634 * should be no harm. 1635 * 1636 * The main purpose of this is to make sure that a CHECK_CONDITION 1637 * is properly treated. 1638 */ 1639 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 1640 { 1641 struct scsi_device *sdev; 1642 1643 __shost_for_each_device(sdev, shost) { 1644 if (channel == sdev_channel(sdev) && 1645 target == sdev_id(sdev)) { 1646 sdev->was_reset = 1; 1647 sdev->expecting_cc_ua = 1; 1648 } 1649 } 1650 } 1651 EXPORT_SYMBOL(scsi_report_device_reset); 1652 1653 static void 1654 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 1655 { 1656 } 1657 1658 /* 1659 * Function: scsi_reset_provider 1660 * 1661 * Purpose: Send requested reset to a bus or device at any phase. 1662 * 1663 * Arguments: device - device to send reset to 1664 * flag - reset type (see scsi.h) 1665 * 1666 * Returns: SUCCESS/FAILURE. 1667 * 1668 * Notes: This is used by the SCSI Generic driver to provide 1669 * Bus/Device reset capability. 1670 */ 1671 int 1672 scsi_reset_provider(struct scsi_device *dev, int flag) 1673 { 1674 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL); 1675 struct Scsi_Host *shost = dev->host; 1676 struct request req; 1677 unsigned long flags; 1678 int rtn; 1679 1680 scmd->request = &req; 1681 memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout)); 1682 1683 memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd)); 1684 1685 scmd->scsi_done = scsi_reset_provider_done_command; 1686 scmd->done = NULL; 1687 scmd->buffer = NULL; 1688 scmd->bufflen = 0; 1689 scmd->request_buffer = NULL; 1690 scmd->request_bufflen = 0; 1691 1692 scmd->cmd_len = 0; 1693 1694 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 1695 1696 init_timer(&scmd->eh_timeout); 1697 1698 /* 1699 * Sometimes the command can get back into the timer chain, 1700 * so use the pid as an identifier. 1701 */ 1702 scmd->pid = 0; 1703 1704 spin_lock_irqsave(shost->host_lock, flags); 1705 shost->tmf_in_progress = 1; 1706 spin_unlock_irqrestore(shost->host_lock, flags); 1707 1708 switch (flag) { 1709 case SCSI_TRY_RESET_DEVICE: 1710 rtn = scsi_try_bus_device_reset(scmd); 1711 if (rtn == SUCCESS) 1712 break; 1713 /* FALLTHROUGH */ 1714 case SCSI_TRY_RESET_BUS: 1715 rtn = scsi_try_bus_reset(scmd); 1716 if (rtn == SUCCESS) 1717 break; 1718 /* FALLTHROUGH */ 1719 case SCSI_TRY_RESET_HOST: 1720 rtn = scsi_try_host_reset(scmd); 1721 break; 1722 default: 1723 rtn = FAILED; 1724 } 1725 1726 spin_lock_irqsave(shost->host_lock, flags); 1727 shost->tmf_in_progress = 0; 1728 spin_unlock_irqrestore(shost->host_lock, flags); 1729 1730 /* 1731 * be sure to wake up anyone who was sleeping or had their queue 1732 * suspended while we performed the TMF. 1733 */ 1734 SCSI_LOG_ERROR_RECOVERY(3, 1735 printk("%s: waking up host to restart after TMF\n", 1736 __FUNCTION__)); 1737 1738 wake_up(&shost->host_wait); 1739 1740 scsi_run_host_queues(shost); 1741 1742 scsi_next_command(scmd); 1743 return rtn; 1744 } 1745 EXPORT_SYMBOL(scsi_reset_provider); 1746 1747 /** 1748 * scsi_normalize_sense - normalize main elements from either fixed or 1749 * descriptor sense data format into a common format. 1750 * 1751 * @sense_buffer: byte array containing sense data returned by device 1752 * @sb_len: number of valid bytes in sense_buffer 1753 * @sshdr: pointer to instance of structure that common 1754 * elements are written to. 1755 * 1756 * Notes: 1757 * The "main elements" from sense data are: response_code, sense_key, 1758 * asc, ascq and additional_length (only for descriptor format). 1759 * 1760 * Typically this function can be called after a device has 1761 * responded to a SCSI command with the CHECK_CONDITION status. 1762 * 1763 * Return value: 1764 * 1 if valid sense data information found, else 0; 1765 **/ 1766 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len, 1767 struct scsi_sense_hdr *sshdr) 1768 { 1769 if (!sense_buffer || !sb_len) 1770 return 0; 1771 1772 memset(sshdr, 0, sizeof(struct scsi_sense_hdr)); 1773 1774 sshdr->response_code = (sense_buffer[0] & 0x7f); 1775 1776 if (!scsi_sense_valid(sshdr)) 1777 return 0; 1778 1779 if (sshdr->response_code >= 0x72) { 1780 /* 1781 * descriptor format 1782 */ 1783 if (sb_len > 1) 1784 sshdr->sense_key = (sense_buffer[1] & 0xf); 1785 if (sb_len > 2) 1786 sshdr->asc = sense_buffer[2]; 1787 if (sb_len > 3) 1788 sshdr->ascq = sense_buffer[3]; 1789 if (sb_len > 7) 1790 sshdr->additional_length = sense_buffer[7]; 1791 } else { 1792 /* 1793 * fixed format 1794 */ 1795 if (sb_len > 2) 1796 sshdr->sense_key = (sense_buffer[2] & 0xf); 1797 if (sb_len > 7) { 1798 sb_len = (sb_len < (sense_buffer[7] + 8)) ? 1799 sb_len : (sense_buffer[7] + 8); 1800 if (sb_len > 12) 1801 sshdr->asc = sense_buffer[12]; 1802 if (sb_len > 13) 1803 sshdr->ascq = sense_buffer[13]; 1804 } 1805 } 1806 1807 return 1; 1808 } 1809 EXPORT_SYMBOL(scsi_normalize_sense); 1810 1811 int scsi_command_normalize_sense(struct scsi_cmnd *cmd, 1812 struct scsi_sense_hdr *sshdr) 1813 { 1814 return scsi_normalize_sense(cmd->sense_buffer, 1815 sizeof(cmd->sense_buffer), sshdr); 1816 } 1817 EXPORT_SYMBOL(scsi_command_normalize_sense); 1818 1819 /** 1820 * scsi_sense_desc_find - search for a given descriptor type in 1821 * descriptor sense data format. 1822 * 1823 * @sense_buffer: byte array of descriptor format sense data 1824 * @sb_len: number of valid bytes in sense_buffer 1825 * @desc_type: value of descriptor type to find 1826 * (e.g. 0 -> information) 1827 * 1828 * Notes: 1829 * only valid when sense data is in descriptor format 1830 * 1831 * Return value: 1832 * pointer to start of (first) descriptor if found else NULL 1833 **/ 1834 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len, 1835 int desc_type) 1836 { 1837 int add_sen_len, add_len, desc_len, k; 1838 const u8 * descp; 1839 1840 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7]))) 1841 return NULL; 1842 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73)) 1843 return NULL; 1844 add_sen_len = (add_sen_len < (sb_len - 8)) ? 1845 add_sen_len : (sb_len - 8); 1846 descp = &sense_buffer[8]; 1847 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) { 1848 descp += desc_len; 1849 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1; 1850 desc_len = add_len + 2; 1851 if (descp[0] == desc_type) 1852 return descp; 1853 if (add_len < 0) // short descriptor ?? 1854 break; 1855 } 1856 return NULL; 1857 } 1858 EXPORT_SYMBOL(scsi_sense_desc_find); 1859 1860 /** 1861 * scsi_get_sense_info_fld - attempts to get information field from 1862 * sense data (either fixed or descriptor format) 1863 * 1864 * @sense_buffer: byte array of sense data 1865 * @sb_len: number of valid bytes in sense_buffer 1866 * @info_out: pointer to 64 integer where 8 or 4 byte information 1867 * field will be placed if found. 1868 * 1869 * Return value: 1870 * 1 if information field found, 0 if not found. 1871 **/ 1872 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len, 1873 u64 * info_out) 1874 { 1875 int j; 1876 const u8 * ucp; 1877 u64 ull; 1878 1879 if (sb_len < 7) 1880 return 0; 1881 switch (sense_buffer[0] & 0x7f) { 1882 case 0x70: 1883 case 0x71: 1884 if (sense_buffer[0] & 0x80) { 1885 *info_out = (sense_buffer[3] << 24) + 1886 (sense_buffer[4] << 16) + 1887 (sense_buffer[5] << 8) + sense_buffer[6]; 1888 return 1; 1889 } else 1890 return 0; 1891 case 0x72: 1892 case 0x73: 1893 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 1894 0 /* info desc */); 1895 if (ucp && (0xa == ucp[1])) { 1896 ull = 0; 1897 for (j = 0; j < 8; ++j) { 1898 if (j > 0) 1899 ull <<= 8; 1900 ull |= ucp[4 + j]; 1901 } 1902 *info_out = ull; 1903 return 1; 1904 } else 1905 return 0; 1906 default: 1907 return 0; 1908 } 1909 } 1910 EXPORT_SYMBOL(scsi_get_sense_info_fld); 1911