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_dbg.h> 30 #include <scsi/scsi_device.h> 31 #include <scsi/scsi_eh.h> 32 #include <scsi/scsi_host.h> 33 #include <scsi/scsi_ioctl.h> 34 #include <scsi/scsi_request.h> 35 36 #include "scsi_priv.h" 37 #include "scsi_logging.h" 38 39 #define SENSE_TIMEOUT (10*HZ) 40 #define START_UNIT_TIMEOUT (30*HZ) 41 42 /* 43 * These should *probably* be handled by the host itself. 44 * Since it is allowed to sleep, it probably should. 45 */ 46 #define BUS_RESET_SETTLE_TIME (10) 47 #define HOST_RESET_SETTLE_TIME (10) 48 49 /* called with shost->host_lock held */ 50 void scsi_eh_wakeup(struct Scsi_Host *shost) 51 { 52 if (shost->host_busy == shost->host_failed) { 53 wake_up_process(shost->ehandler); 54 SCSI_LOG_ERROR_RECOVERY(5, 55 printk("Waking error handler thread\n")); 56 } 57 } 58 59 /** 60 * scsi_eh_scmd_add - add scsi cmd to error handling. 61 * @scmd: scmd to run eh on. 62 * @eh_flag: optional SCSI_EH flag. 63 * 64 * Return value: 65 * 0 on failure. 66 **/ 67 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag) 68 { 69 struct Scsi_Host *shost = scmd->device->host; 70 unsigned long flags; 71 int ret = 0; 72 73 if (!shost->ehandler) 74 return 0; 75 76 spin_lock_irqsave(shost->host_lock, flags); 77 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 78 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)) 79 goto out_unlock; 80 81 ret = 1; 82 scmd->eh_eflags |= eh_flag; 83 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 84 shost->host_failed++; 85 scsi_eh_wakeup(shost); 86 out_unlock: 87 spin_unlock_irqrestore(shost->host_lock, flags); 88 return ret; 89 } 90 91 /** 92 * scsi_add_timer - Start timeout timer for a single scsi command. 93 * @scmd: scsi command that is about to start running. 94 * @timeout: amount of time to allow this command to run. 95 * @complete: timeout function to call if timer isn't canceled. 96 * 97 * Notes: 98 * This should be turned into an inline function. Each scsi command 99 * has its own timer, and as it is added to the queue, we set up the 100 * timer. When the command completes, we cancel the timer. 101 **/ 102 void scsi_add_timer(struct scsi_cmnd *scmd, int timeout, 103 void (*complete)(struct scsi_cmnd *)) 104 { 105 106 /* 107 * If the clock was already running for this command, then 108 * first delete the timer. The timer handling code gets rather 109 * confused if we don't do this. 110 */ 111 if (scmd->eh_timeout.function) 112 del_timer(&scmd->eh_timeout); 113 114 scmd->eh_timeout.data = (unsigned long)scmd; 115 scmd->eh_timeout.expires = jiffies + timeout; 116 scmd->eh_timeout.function = (void (*)(unsigned long)) complete; 117 118 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:" 119 " %d, (%p)\n", __FUNCTION__, 120 scmd, timeout, complete)); 121 122 add_timer(&scmd->eh_timeout); 123 } 124 125 /** 126 * scsi_delete_timer - Delete/cancel timer for a given function. 127 * @scmd: Cmd that we are canceling timer for 128 * 129 * Notes: 130 * This should be turned into an inline function. 131 * 132 * Return value: 133 * 1 if we were able to detach the timer. 0 if we blew it, and the 134 * timer function has already started to run. 135 **/ 136 int scsi_delete_timer(struct scsi_cmnd *scmd) 137 { 138 int rtn; 139 140 rtn = del_timer(&scmd->eh_timeout); 141 142 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p," 143 " rtn: %d\n", __FUNCTION__, 144 scmd, rtn)); 145 146 scmd->eh_timeout.data = (unsigned long)NULL; 147 scmd->eh_timeout.function = NULL; 148 149 return rtn; 150 } 151 152 /** 153 * scsi_times_out - Timeout function for normal scsi commands. 154 * @scmd: Cmd that is timing out. 155 * 156 * Notes: 157 * We do not need to lock this. There is the potential for a race 158 * only in that the normal completion handling might run, but if the 159 * normal completion function determines that the timer has already 160 * fired, then it mustn't do anything. 161 **/ 162 void scsi_times_out(struct scsi_cmnd *scmd) 163 { 164 scsi_log_completion(scmd, TIMEOUT_ERROR); 165 166 if (scmd->device->host->hostt->eh_timed_out) 167 switch (scmd->device->host->hostt->eh_timed_out(scmd)) { 168 case EH_HANDLED: 169 __scsi_done(scmd); 170 return; 171 case EH_RESET_TIMER: 172 /* This allows a single retry even of a command 173 * with allowed == 0 */ 174 if (scmd->retries++ > scmd->allowed) 175 break; 176 scsi_add_timer(scmd, scmd->timeout_per_command, 177 scsi_times_out); 178 return; 179 case EH_NOT_HANDLED: 180 break; 181 } 182 183 if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) { 184 scmd->result |= DID_TIME_OUT << 16; 185 __scsi_done(scmd); 186 } 187 } 188 189 /** 190 * scsi_block_when_processing_errors - Prevent cmds from being queued. 191 * @sdev: Device on which we are performing recovery. 192 * 193 * Description: 194 * We block until the host is out of error recovery, and then check to 195 * see whether the host or the device is offline. 196 * 197 * Return value: 198 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 199 **/ 200 int scsi_block_when_processing_errors(struct scsi_device *sdev) 201 { 202 int online; 203 204 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 205 206 online = scsi_device_online(sdev); 207 208 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__, 209 online)); 210 211 return online; 212 } 213 EXPORT_SYMBOL(scsi_block_when_processing_errors); 214 215 #ifdef CONFIG_SCSI_LOGGING 216 /** 217 * scsi_eh_prt_fail_stats - Log info on failures. 218 * @shost: scsi host being recovered. 219 * @work_q: Queue of scsi cmds to process. 220 **/ 221 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 222 struct list_head *work_q) 223 { 224 struct scsi_cmnd *scmd; 225 struct scsi_device *sdev; 226 int total_failures = 0; 227 int cmd_failed = 0; 228 int cmd_cancel = 0; 229 int devices_failed = 0; 230 231 shost_for_each_device(sdev, shost) { 232 list_for_each_entry(scmd, work_q, eh_entry) { 233 if (scmd->device == sdev) { 234 ++total_failures; 235 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) 236 ++cmd_cancel; 237 else 238 ++cmd_failed; 239 } 240 } 241 242 if (cmd_cancel || cmd_failed) { 243 SCSI_LOG_ERROR_RECOVERY(3, 244 sdev_printk(KERN_INFO, sdev, 245 "%s: cmds failed: %d, cancel: %d\n", 246 __FUNCTION__, cmd_failed, 247 cmd_cancel)); 248 cmd_cancel = 0; 249 cmd_failed = 0; 250 ++devices_failed; 251 } 252 } 253 254 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d" 255 " devices require eh work\n", 256 total_failures, devices_failed)); 257 } 258 #endif 259 260 /** 261 * scsi_check_sense - Examine scsi cmd sense 262 * @scmd: Cmd to have sense checked. 263 * 264 * Return value: 265 * SUCCESS or FAILED or NEEDS_RETRY 266 * 267 * Notes: 268 * When a deferred error is detected the current command has 269 * not been executed and needs retrying. 270 **/ 271 static int scsi_check_sense(struct scsi_cmnd *scmd) 272 { 273 struct scsi_sense_hdr sshdr; 274 275 if (! scsi_command_normalize_sense(scmd, &sshdr)) 276 return FAILED; /* no valid sense data */ 277 278 if (scsi_sense_is_deferred(&sshdr)) 279 return NEEDS_RETRY; 280 281 /* 282 * Previous logic looked for FILEMARK, EOM or ILI which are 283 * mainly associated with tapes and returned SUCCESS. 284 */ 285 if (sshdr.response_code == 0x70) { 286 /* fixed format */ 287 if (scmd->sense_buffer[2] & 0xe0) 288 return SUCCESS; 289 } else { 290 /* 291 * descriptor format: look for "stream commands sense data 292 * descriptor" (see SSC-3). Assume single sense data 293 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 294 */ 295 if ((sshdr.additional_length > 3) && 296 (scmd->sense_buffer[8] == 0x4) && 297 (scmd->sense_buffer[11] & 0xe0)) 298 return SUCCESS; 299 } 300 301 switch (sshdr.sense_key) { 302 case NO_SENSE: 303 return SUCCESS; 304 case RECOVERED_ERROR: 305 return /* soft_error */ SUCCESS; 306 307 case ABORTED_COMMAND: 308 return NEEDS_RETRY; 309 case NOT_READY: 310 case UNIT_ATTENTION: 311 /* 312 * if we are expecting a cc/ua because of a bus reset that we 313 * performed, treat this just as a retry. otherwise this is 314 * information that we should pass up to the upper-level driver 315 * so that we can deal with it there. 316 */ 317 if (scmd->device->expecting_cc_ua) { 318 scmd->device->expecting_cc_ua = 0; 319 return NEEDS_RETRY; 320 } 321 /* 322 * if the device is in the process of becoming ready, we 323 * should retry. 324 */ 325 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 326 return NEEDS_RETRY; 327 /* 328 * if the device is not started, we need to wake 329 * the error handler to start the motor 330 */ 331 if (scmd->device->allow_restart && 332 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 333 return FAILED; 334 return SUCCESS; 335 336 /* these three are not supported */ 337 case COPY_ABORTED: 338 case VOLUME_OVERFLOW: 339 case MISCOMPARE: 340 return SUCCESS; 341 342 case MEDIUM_ERROR: 343 return NEEDS_RETRY; 344 345 case HARDWARE_ERROR: 346 if (scmd->device->retry_hwerror) 347 return NEEDS_RETRY; 348 else 349 return SUCCESS; 350 351 case ILLEGAL_REQUEST: 352 case BLANK_CHECK: 353 case DATA_PROTECT: 354 default: 355 return SUCCESS; 356 } 357 } 358 359 /** 360 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 361 * @scmd: SCSI cmd to examine. 362 * 363 * Notes: 364 * This is *only* called when we are examining the status of commands 365 * queued during error recovery. the main difference here is that we 366 * don't allow for the possibility of retries here, and we are a lot 367 * more restrictive about what we consider acceptable. 368 **/ 369 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) 370 { 371 /* 372 * first check the host byte, to see if there is anything in there 373 * that would indicate what we need to do. 374 */ 375 if (host_byte(scmd->result) == DID_RESET) { 376 /* 377 * rats. we are already in the error handler, so we now 378 * get to try and figure out what to do next. if the sense 379 * is valid, we have a pretty good idea of what to do. 380 * if not, we mark it as FAILED. 381 */ 382 return scsi_check_sense(scmd); 383 } 384 if (host_byte(scmd->result) != DID_OK) 385 return FAILED; 386 387 /* 388 * next, check the message byte. 389 */ 390 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 391 return FAILED; 392 393 /* 394 * now, check the status byte to see if this indicates 395 * anything special. 396 */ 397 switch (status_byte(scmd->result)) { 398 case GOOD: 399 case COMMAND_TERMINATED: 400 return SUCCESS; 401 case CHECK_CONDITION: 402 return scsi_check_sense(scmd); 403 case CONDITION_GOOD: 404 case INTERMEDIATE_GOOD: 405 case INTERMEDIATE_C_GOOD: 406 /* 407 * who knows? FIXME(eric) 408 */ 409 return SUCCESS; 410 case BUSY: 411 case QUEUE_FULL: 412 case RESERVATION_CONFLICT: 413 default: 414 return FAILED; 415 } 416 return FAILED; 417 } 418 419 /** 420 * scsi_eh_times_out - timeout function for error handling. 421 * @scmd: Cmd that is timing out. 422 * 423 * Notes: 424 * During error handling, the kernel thread will be sleeping waiting 425 * for some action to complete on the device. our only job is to 426 * record that it timed out, and to wake up the thread. 427 **/ 428 static void scsi_eh_times_out(struct scsi_cmnd *scmd) 429 { 430 scmd->eh_eflags |= SCSI_EH_REC_TIMEOUT; 431 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd:%p\n", __FUNCTION__, 432 scmd)); 433 434 up(scmd->device->host->eh_action); 435 } 436 437 /** 438 * scsi_eh_done - Completion function for error handling. 439 * @scmd: Cmd that is done. 440 **/ 441 static void scsi_eh_done(struct scsi_cmnd *scmd) 442 { 443 /* 444 * if the timeout handler is already running, then just set the 445 * flag which says we finished late, and return. we have no 446 * way of stopping the timeout handler from running, so we must 447 * always defer to it. 448 */ 449 if (del_timer(&scmd->eh_timeout)) { 450 scmd->request->rq_status = RQ_SCSI_DONE; 451 452 SCSI_LOG_ERROR_RECOVERY(3, printk("%s scmd: %p result: %x\n", 453 __FUNCTION__, scmd, scmd->result)); 454 455 up(scmd->device->host->eh_action); 456 } 457 } 458 459 /** 460 * scsi_send_eh_cmnd - send a cmd to a device as part of error recovery. 461 * @scmd: SCSI Cmd to send. 462 * @timeout: Timeout for cmd. 463 * 464 * Notes: 465 * The initialization of the structures is quite a bit different in 466 * this case, and furthermore, there is a different completion handler 467 * vs scsi_dispatch_cmd. 468 * Return value: 469 * SUCCESS or FAILED or NEEDS_RETRY 470 **/ 471 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, int timeout) 472 { 473 struct scsi_device *sdev = scmd->device; 474 struct Scsi_Host *shost = sdev->host; 475 DECLARE_MUTEX_LOCKED(sem); 476 unsigned long flags; 477 int rtn = SUCCESS; 478 479 /* 480 * we will use a queued command if possible, otherwise we will 481 * emulate the queuing and calling of completion function ourselves. 482 */ 483 if (sdev->scsi_level <= SCSI_2) 484 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 485 (sdev->lun << 5 & 0xe0); 486 487 scsi_add_timer(scmd, timeout, scsi_eh_times_out); 488 489 /* 490 * set up the semaphore so we wait for the command to complete. 491 */ 492 shost->eh_action = &sem; 493 scmd->request->rq_status = RQ_SCSI_BUSY; 494 495 spin_lock_irqsave(shost->host_lock, flags); 496 scsi_log_send(scmd); 497 shost->hostt->queuecommand(scmd, scsi_eh_done); 498 spin_unlock_irqrestore(shost->host_lock, flags); 499 500 down(&sem); 501 scsi_log_completion(scmd, SUCCESS); 502 503 shost->eh_action = NULL; 504 505 /* 506 * see if timeout. if so, tell the host to forget about it. 507 * in other words, we don't want a callback any more. 508 */ 509 if (scmd->eh_eflags & SCSI_EH_REC_TIMEOUT) { 510 scmd->eh_eflags &= ~SCSI_EH_REC_TIMEOUT; 511 512 /* 513 * as far as the low level driver is 514 * concerned, this command is still active, so 515 * we must give the low level driver a chance 516 * to abort it. (db) 517 * 518 * FIXME(eric) - we are not tracking whether we could 519 * abort a timed out command or not. not sure how 520 * we should treat them differently anyways. 521 */ 522 if (shost->hostt->eh_abort_handler) 523 shost->hostt->eh_abort_handler(scmd); 524 525 scmd->request->rq_status = RQ_SCSI_DONE; 526 rtn = FAILED; 527 } 528 529 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd: %p, rtn:%x\n", 530 __FUNCTION__, scmd, rtn)); 531 532 /* 533 * now examine the actual status codes to see whether the command 534 * actually did complete normally. 535 */ 536 if (rtn == SUCCESS) { 537 rtn = scsi_eh_completed_normally(scmd); 538 SCSI_LOG_ERROR_RECOVERY(3, 539 printk("%s: scsi_eh_completed_normally %x\n", 540 __FUNCTION__, rtn)); 541 switch (rtn) { 542 case SUCCESS: 543 case NEEDS_RETRY: 544 case FAILED: 545 break; 546 default: 547 rtn = FAILED; 548 break; 549 } 550 } 551 552 return rtn; 553 } 554 555 /** 556 * scsi_request_sense - Request sense data from a particular target. 557 * @scmd: SCSI cmd for request sense. 558 * 559 * Notes: 560 * Some hosts automatically obtain this information, others require 561 * that we obtain it on our own. This function will *not* return until 562 * the command either times out, or it completes. 563 **/ 564 static int scsi_request_sense(struct scsi_cmnd *scmd) 565 { 566 static unsigned char generic_sense[6] = 567 {REQUEST_SENSE, 0, 0, 0, 252, 0}; 568 unsigned char *scsi_result; 569 int saved_result; 570 int rtn; 571 572 memcpy(scmd->cmnd, generic_sense, sizeof(generic_sense)); 573 574 scsi_result = kmalloc(252, GFP_ATOMIC | ((scmd->device->host->hostt->unchecked_isa_dma) ? __GFP_DMA : 0)); 575 576 577 if (unlikely(!scsi_result)) { 578 printk(KERN_ERR "%s: cannot allocate scsi_result.\n", 579 __FUNCTION__); 580 return FAILED; 581 } 582 583 /* 584 * zero the sense buffer. some host adapters automatically always 585 * request sense, so it is not a good idea that 586 * scmd->request_buffer and scmd->sense_buffer point to the same 587 * address (db). 0 is not a valid sense code. 588 */ 589 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); 590 memset(scsi_result, 0, 252); 591 592 saved_result = scmd->result; 593 scmd->request_buffer = scsi_result; 594 scmd->request_bufflen = 252; 595 scmd->use_sg = 0; 596 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 597 scmd->sc_data_direction = DMA_FROM_DEVICE; 598 scmd->underflow = 0; 599 600 rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT); 601 602 /* last chance to have valid sense data */ 603 if(!SCSI_SENSE_VALID(scmd)) { 604 memcpy(scmd->sense_buffer, scmd->request_buffer, 605 sizeof(scmd->sense_buffer)); 606 } 607 608 kfree(scsi_result); 609 610 /* 611 * when we eventually call scsi_finish, we really wish to complete 612 * the original request, so let's restore the original data. (db) 613 */ 614 scsi_setup_cmd_retry(scmd); 615 scmd->result = saved_result; 616 return rtn; 617 } 618 619 /** 620 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 621 * @scmd: Original SCSI cmd that eh has finished. 622 * @done_q: Queue for processed commands. 623 * 624 * Notes: 625 * We don't want to use the normal command completion while we are are 626 * still handling errors - it may cause other commands to be queued, 627 * and that would disturb what we are doing. thus we really want to 628 * keep a list of pending commands for final completion, and once we 629 * are ready to leave error handling we handle completion for real. 630 **/ 631 static void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, 632 struct list_head *done_q) 633 { 634 scmd->device->host->host_failed--; 635 scmd->eh_eflags = 0; 636 637 /* 638 * set this back so that the upper level can correctly free up 639 * things. 640 */ 641 scsi_setup_cmd_retry(scmd); 642 list_move_tail(&scmd->eh_entry, done_q); 643 } 644 645 /** 646 * scsi_eh_get_sense - Get device sense data. 647 * @work_q: Queue of commands to process. 648 * @done_q: Queue of proccessed commands.. 649 * 650 * Description: 651 * See if we need to request sense information. if so, then get it 652 * now, so we have a better idea of what to do. 653 * 654 * Notes: 655 * This has the unfortunate side effect that if a shost adapter does 656 * not automatically request sense information, that we end up shutting 657 * it down before we request it. 658 * 659 * All drivers should request sense information internally these days, 660 * so for now all I have to say is tough noogies if you end up in here. 661 * 662 * XXX: Long term this code should go away, but that needs an audit of 663 * all LLDDs first. 664 **/ 665 static int scsi_eh_get_sense(struct list_head *work_q, 666 struct list_head *done_q) 667 { 668 struct scsi_cmnd *scmd, *next; 669 int rtn; 670 671 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 672 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) || 673 SCSI_SENSE_VALID(scmd)) 674 continue; 675 676 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 677 "%s: requesting sense\n", 678 current->comm)); 679 rtn = scsi_request_sense(scmd); 680 if (rtn != SUCCESS) 681 continue; 682 683 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p" 684 " result %x\n", scmd, 685 scmd->result)); 686 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd)); 687 688 rtn = scsi_decide_disposition(scmd); 689 690 /* 691 * if the result was normal, then just pass it along to the 692 * upper level. 693 */ 694 if (rtn == SUCCESS) 695 /* we don't want this command reissued, just 696 * finished with the sense data, so set 697 * retries to the max allowed to ensure it 698 * won't get reissued */ 699 scmd->retries = scmd->allowed; 700 else if (rtn != NEEDS_RETRY) 701 continue; 702 703 scsi_eh_finish_cmd(scmd, done_q); 704 } 705 706 return list_empty(work_q); 707 } 708 709 /** 710 * scsi_try_to_abort_cmd - Ask host to abort a running command. 711 * @scmd: SCSI cmd to abort from Lower Level. 712 * 713 * Notes: 714 * This function will not return until the user's completion function 715 * has been called. there is no timeout on this operation. if the 716 * author of the low-level driver wishes this operation to be timed, 717 * they can provide this facility themselves. helper functions in 718 * scsi_error.c can be supplied to make this easier to do. 719 **/ 720 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd) 721 { 722 if (!scmd->device->host->hostt->eh_abort_handler) 723 return FAILED; 724 725 /* 726 * scsi_done was called just after the command timed out and before 727 * we had a chance to process it. (db) 728 */ 729 if (scmd->serial_number == 0) 730 return SUCCESS; 731 return scmd->device->host->hostt->eh_abort_handler(scmd); 732 } 733 734 /** 735 * scsi_eh_tur - Send TUR to device. 736 * @scmd: Scsi cmd to send TUR 737 * 738 * Return value: 739 * 0 - Device is ready. 1 - Device NOT ready. 740 **/ 741 static int scsi_eh_tur(struct scsi_cmnd *scmd) 742 { 743 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 744 int retry_cnt = 1, rtn; 745 int saved_result; 746 747 retry_tur: 748 memcpy(scmd->cmnd, tur_command, sizeof(tur_command)); 749 750 /* 751 * zero the sense buffer. the scsi spec mandates that any 752 * untransferred sense data should be interpreted as being zero. 753 */ 754 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); 755 756 saved_result = scmd->result; 757 scmd->request_buffer = NULL; 758 scmd->request_bufflen = 0; 759 scmd->use_sg = 0; 760 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 761 scmd->underflow = 0; 762 scmd->sc_data_direction = DMA_NONE; 763 764 rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT); 765 766 /* 767 * when we eventually call scsi_finish, we really wish to complete 768 * the original request, so let's restore the original data. (db) 769 */ 770 scsi_setup_cmd_retry(scmd); 771 scmd->result = saved_result; 772 773 /* 774 * hey, we are done. let's look to see what happened. 775 */ 776 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", 777 __FUNCTION__, scmd, rtn)); 778 if (rtn == SUCCESS) 779 return 0; 780 else if (rtn == NEEDS_RETRY) { 781 if (retry_cnt--) 782 goto retry_tur; 783 return 0; 784 } 785 return 1; 786 } 787 788 /** 789 * scsi_eh_abort_cmds - abort canceled commands. 790 * @shost: scsi host being recovered. 791 * @eh_done_q: list_head for processed commands. 792 * 793 * Decription: 794 * Try and see whether or not it makes sense to try and abort the 795 * running command. this only works out to be the case if we have one 796 * command that has timed out. if the command simply failed, it makes 797 * no sense to try and abort the command, since as far as the shost 798 * adapter is concerned, it isn't running. 799 **/ 800 static int scsi_eh_abort_cmds(struct list_head *work_q, 801 struct list_head *done_q) 802 { 803 struct scsi_cmnd *scmd, *next; 804 int rtn; 805 806 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 807 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD)) 808 continue; 809 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:" 810 "0x%p\n", current->comm, 811 scmd)); 812 rtn = scsi_try_to_abort_cmd(scmd); 813 if (rtn == SUCCESS) { 814 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD; 815 if (!scsi_device_online(scmd->device) || 816 !scsi_eh_tur(scmd)) { 817 scsi_eh_finish_cmd(scmd, done_q); 818 } 819 820 } else 821 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting" 822 " cmd failed:" 823 "0x%p\n", 824 current->comm, 825 scmd)); 826 } 827 828 return list_empty(work_q); 829 } 830 831 /** 832 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 833 * @scmd: SCSI cmd used to send BDR 834 * 835 * Notes: 836 * There is no timeout for this operation. if this operation is 837 * unreliable for a given host, then the host itself needs to put a 838 * timer on it, and set the host back to a consistent state prior to 839 * returning. 840 **/ 841 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 842 { 843 int rtn; 844 845 if (!scmd->device->host->hostt->eh_device_reset_handler) 846 return FAILED; 847 848 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd); 849 if (rtn == SUCCESS) { 850 scmd->device->was_reset = 1; 851 scmd->device->expecting_cc_ua = 1; 852 } 853 854 return rtn; 855 } 856 857 /** 858 * scsi_eh_try_stu - Send START_UNIT to device. 859 * @scmd: Scsi cmd to send START_UNIT 860 * 861 * Return value: 862 * 0 - Device is ready. 1 - Device NOT ready. 863 **/ 864 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 865 { 866 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 867 int rtn; 868 int saved_result; 869 870 if (!scmd->device->allow_restart) 871 return 1; 872 873 memcpy(scmd->cmnd, stu_command, sizeof(stu_command)); 874 875 /* 876 * zero the sense buffer. the scsi spec mandates that any 877 * untransferred sense data should be interpreted as being zero. 878 */ 879 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); 880 881 saved_result = scmd->result; 882 scmd->request_buffer = NULL; 883 scmd->request_bufflen = 0; 884 scmd->use_sg = 0; 885 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 886 scmd->underflow = 0; 887 scmd->sc_data_direction = DMA_NONE; 888 889 rtn = scsi_send_eh_cmnd(scmd, START_UNIT_TIMEOUT); 890 891 /* 892 * when we eventually call scsi_finish, we really wish to complete 893 * the original request, so let's restore the original data. (db) 894 */ 895 scsi_setup_cmd_retry(scmd); 896 scmd->result = saved_result; 897 898 /* 899 * hey, we are done. let's look to see what happened. 900 */ 901 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", 902 __FUNCTION__, scmd, rtn)); 903 if (rtn == SUCCESS) 904 return 0; 905 return 1; 906 } 907 908 /** 909 * scsi_eh_stu - send START_UNIT if needed 910 * @shost: scsi host being recovered. 911 * @eh_done_q: list_head for processed commands. 912 * 913 * Notes: 914 * If commands are failing due to not ready, initializing command required, 915 * try revalidating the device, which will end up sending a start unit. 916 **/ 917 static int scsi_eh_stu(struct Scsi_Host *shost, 918 struct list_head *work_q, 919 struct list_head *done_q) 920 { 921 struct scsi_cmnd *scmd, *stu_scmd, *next; 922 struct scsi_device *sdev; 923 924 shost_for_each_device(sdev, shost) { 925 stu_scmd = NULL; 926 list_for_each_entry(scmd, work_q, eh_entry) 927 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 928 scsi_check_sense(scmd) == FAILED ) { 929 stu_scmd = scmd; 930 break; 931 } 932 933 if (!stu_scmd) 934 continue; 935 936 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:" 937 " 0x%p\n", current->comm, sdev)); 938 939 if (!scsi_eh_try_stu(stu_scmd)) { 940 if (!scsi_device_online(sdev) || 941 !scsi_eh_tur(stu_scmd)) { 942 list_for_each_entry_safe(scmd, next, 943 work_q, eh_entry) { 944 if (scmd->device == sdev) 945 scsi_eh_finish_cmd(scmd, done_q); 946 } 947 } 948 } else { 949 SCSI_LOG_ERROR_RECOVERY(3, 950 printk("%s: START_UNIT failed to sdev:" 951 " 0x%p\n", current->comm, sdev)); 952 } 953 } 954 955 return list_empty(work_q); 956 } 957 958 959 /** 960 * scsi_eh_bus_device_reset - send bdr if needed 961 * @shost: scsi host being recovered. 962 * @eh_done_q: list_head for processed commands. 963 * 964 * Notes: 965 * Try a bus device reset. still, look to see whether we have multiple 966 * devices that are jammed or not - if we have multiple devices, it 967 * makes no sense to try bus_device_reset - we really would need to try 968 * a bus_reset instead. 969 **/ 970 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 971 struct list_head *work_q, 972 struct list_head *done_q) 973 { 974 struct scsi_cmnd *scmd, *bdr_scmd, *next; 975 struct scsi_device *sdev; 976 int rtn; 977 978 shost_for_each_device(sdev, shost) { 979 bdr_scmd = NULL; 980 list_for_each_entry(scmd, work_q, eh_entry) 981 if (scmd->device == sdev) { 982 bdr_scmd = scmd; 983 break; 984 } 985 986 if (!bdr_scmd) 987 continue; 988 989 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:" 990 " 0x%p\n", current->comm, 991 sdev)); 992 rtn = scsi_try_bus_device_reset(bdr_scmd); 993 if (rtn == SUCCESS) { 994 if (!scsi_device_online(sdev) || 995 !scsi_eh_tur(bdr_scmd)) { 996 list_for_each_entry_safe(scmd, next, 997 work_q, eh_entry) { 998 if (scmd->device == sdev) 999 scsi_eh_finish_cmd(scmd, 1000 done_q); 1001 } 1002 } 1003 } else { 1004 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR" 1005 " failed sdev:" 1006 "0x%p\n", 1007 current->comm, 1008 sdev)); 1009 } 1010 } 1011 1012 return list_empty(work_q); 1013 } 1014 1015 /** 1016 * scsi_try_bus_reset - ask host to perform a bus reset 1017 * @scmd: SCSI cmd to send bus reset. 1018 **/ 1019 static int scsi_try_bus_reset(struct scsi_cmnd *scmd) 1020 { 1021 unsigned long flags; 1022 int rtn; 1023 1024 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n", 1025 __FUNCTION__)); 1026 1027 if (!scmd->device->host->hostt->eh_bus_reset_handler) 1028 return FAILED; 1029 1030 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd); 1031 1032 if (rtn == SUCCESS) { 1033 if (!scmd->device->host->hostt->skip_settle_delay) 1034 ssleep(BUS_RESET_SETTLE_TIME); 1035 spin_lock_irqsave(scmd->device->host->host_lock, flags); 1036 scsi_report_bus_reset(scmd->device->host, 1037 scmd_channel(scmd)); 1038 spin_unlock_irqrestore(scmd->device->host->host_lock, flags); 1039 } 1040 1041 return rtn; 1042 } 1043 1044 /** 1045 * scsi_try_host_reset - ask host adapter to reset itself 1046 * @scmd: SCSI cmd to send hsot reset. 1047 **/ 1048 static int scsi_try_host_reset(struct scsi_cmnd *scmd) 1049 { 1050 unsigned long flags; 1051 int rtn; 1052 1053 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n", 1054 __FUNCTION__)); 1055 1056 if (!scmd->device->host->hostt->eh_host_reset_handler) 1057 return FAILED; 1058 1059 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd); 1060 1061 if (rtn == SUCCESS) { 1062 if (!scmd->device->host->hostt->skip_settle_delay) 1063 ssleep(HOST_RESET_SETTLE_TIME); 1064 spin_lock_irqsave(scmd->device->host->host_lock, flags); 1065 scsi_report_bus_reset(scmd->device->host, 1066 scmd_channel(scmd)); 1067 spin_unlock_irqrestore(scmd->device->host->host_lock, flags); 1068 } 1069 1070 return rtn; 1071 } 1072 1073 /** 1074 * scsi_eh_bus_reset - send a bus reset 1075 * @shost: scsi host being recovered. 1076 * @eh_done_q: list_head for processed commands. 1077 **/ 1078 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1079 struct list_head *work_q, 1080 struct list_head *done_q) 1081 { 1082 struct scsi_cmnd *scmd, *chan_scmd, *next; 1083 unsigned int channel; 1084 int rtn; 1085 1086 /* 1087 * we really want to loop over the various channels, and do this on 1088 * a channel by channel basis. we should also check to see if any 1089 * of the failed commands are on soft_reset devices, and if so, skip 1090 * the reset. 1091 */ 1092 1093 for (channel = 0; channel <= shost->max_channel; channel++) { 1094 chan_scmd = NULL; 1095 list_for_each_entry(scmd, work_q, eh_entry) { 1096 if (channel == scmd_channel(scmd)) { 1097 chan_scmd = scmd; 1098 break; 1099 /* 1100 * FIXME add back in some support for 1101 * soft_reset devices. 1102 */ 1103 } 1104 } 1105 1106 if (!chan_scmd) 1107 continue; 1108 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:" 1109 " %d\n", current->comm, 1110 channel)); 1111 rtn = scsi_try_bus_reset(chan_scmd); 1112 if (rtn == SUCCESS) { 1113 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1114 if (channel == scmd_channel(scmd)) 1115 if (!scsi_device_online(scmd->device) || 1116 !scsi_eh_tur(scmd)) 1117 scsi_eh_finish_cmd(scmd, 1118 done_q); 1119 } 1120 } else { 1121 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST" 1122 " failed chan: %d\n", 1123 current->comm, 1124 channel)); 1125 } 1126 } 1127 return list_empty(work_q); 1128 } 1129 1130 /** 1131 * scsi_eh_host_reset - send a host reset 1132 * @work_q: list_head for processed commands. 1133 * @done_q: list_head for processed commands. 1134 **/ 1135 static int scsi_eh_host_reset(struct list_head *work_q, 1136 struct list_head *done_q) 1137 { 1138 struct scsi_cmnd *scmd, *next; 1139 int rtn; 1140 1141 if (!list_empty(work_q)) { 1142 scmd = list_entry(work_q->next, 1143 struct scsi_cmnd, eh_entry); 1144 1145 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n" 1146 , current->comm)); 1147 1148 rtn = scsi_try_host_reset(scmd); 1149 if (rtn == SUCCESS) { 1150 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1151 if (!scsi_device_online(scmd->device) || 1152 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) || 1153 !scsi_eh_tur(scmd)) 1154 scsi_eh_finish_cmd(scmd, done_q); 1155 } 1156 } else { 1157 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST" 1158 " failed\n", 1159 current->comm)); 1160 } 1161 } 1162 return list_empty(work_q); 1163 } 1164 1165 /** 1166 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1167 * @work_q: list_head for processed commands. 1168 * @done_q: list_head for processed commands. 1169 * 1170 **/ 1171 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1172 struct list_head *done_q) 1173 { 1174 struct scsi_cmnd *scmd, *next; 1175 1176 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1177 sdev_printk(KERN_INFO, scmd->device, 1178 "scsi: Device offlined - not" 1179 " ready after error recovery\n"); 1180 scsi_device_set_state(scmd->device, SDEV_OFFLINE); 1181 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) { 1182 /* 1183 * FIXME: Handle lost cmds. 1184 */ 1185 } 1186 scsi_eh_finish_cmd(scmd, done_q); 1187 } 1188 return; 1189 } 1190 1191 /** 1192 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1193 * @scmd: SCSI cmd to examine. 1194 * 1195 * Notes: 1196 * This is *only* called when we are examining the status after sending 1197 * out the actual data command. any commands that are queued for error 1198 * recovery (e.g. test_unit_ready) do *not* come through here. 1199 * 1200 * When this routine returns failed, it means the error handler thread 1201 * is woken. In cases where the error code indicates an error that 1202 * doesn't require the error handler read (i.e. we don't need to 1203 * abort/reset), this function should return SUCCESS. 1204 **/ 1205 int scsi_decide_disposition(struct scsi_cmnd *scmd) 1206 { 1207 int rtn; 1208 1209 /* 1210 * if the device is offline, then we clearly just pass the result back 1211 * up to the top level. 1212 */ 1213 if (!scsi_device_online(scmd->device)) { 1214 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report" 1215 " as SUCCESS\n", 1216 __FUNCTION__)); 1217 return SUCCESS; 1218 } 1219 1220 /* 1221 * first check the host byte, to see if there is anything in there 1222 * that would indicate what we need to do. 1223 */ 1224 switch (host_byte(scmd->result)) { 1225 case DID_PASSTHROUGH: 1226 /* 1227 * no matter what, pass this through to the upper layer. 1228 * nuke this special code so that it looks like we are saying 1229 * did_ok. 1230 */ 1231 scmd->result &= 0xff00ffff; 1232 return SUCCESS; 1233 case DID_OK: 1234 /* 1235 * looks good. drop through, and check the next byte. 1236 */ 1237 break; 1238 case DID_NO_CONNECT: 1239 case DID_BAD_TARGET: 1240 case DID_ABORT: 1241 /* 1242 * note - this means that we just report the status back 1243 * to the top level driver, not that we actually think 1244 * that it indicates SUCCESS. 1245 */ 1246 return SUCCESS; 1247 /* 1248 * when the low level driver returns did_soft_error, 1249 * it is responsible for keeping an internal retry counter 1250 * in order to avoid endless loops (db) 1251 * 1252 * actually this is a bug in this function here. we should 1253 * be mindful of the maximum number of retries specified 1254 * and not get stuck in a loop. 1255 */ 1256 case DID_SOFT_ERROR: 1257 goto maybe_retry; 1258 case DID_IMM_RETRY: 1259 return NEEDS_RETRY; 1260 1261 case DID_REQUEUE: 1262 return ADD_TO_MLQUEUE; 1263 1264 case DID_ERROR: 1265 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1266 status_byte(scmd->result) == RESERVATION_CONFLICT) 1267 /* 1268 * execute reservation conflict processing code 1269 * lower down 1270 */ 1271 break; 1272 /* fallthrough */ 1273 1274 case DID_BUS_BUSY: 1275 case DID_PARITY: 1276 goto maybe_retry; 1277 case DID_TIME_OUT: 1278 /* 1279 * when we scan the bus, we get timeout messages for 1280 * these commands if there is no device available. 1281 * other hosts report did_no_connect for the same thing. 1282 */ 1283 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1284 scmd->cmnd[0] == INQUIRY)) { 1285 return SUCCESS; 1286 } else { 1287 return FAILED; 1288 } 1289 case DID_RESET: 1290 return SUCCESS; 1291 default: 1292 return FAILED; 1293 } 1294 1295 /* 1296 * next, check the message byte. 1297 */ 1298 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 1299 return FAILED; 1300 1301 /* 1302 * check the status byte to see if this indicates anything special. 1303 */ 1304 switch (status_byte(scmd->result)) { 1305 case QUEUE_FULL: 1306 /* 1307 * the case of trying to send too many commands to a 1308 * tagged queueing device. 1309 */ 1310 case BUSY: 1311 /* 1312 * device can't talk to us at the moment. Should only 1313 * occur (SAM-3) when the task queue is empty, so will cause 1314 * the empty queue handling to trigger a stall in the 1315 * device. 1316 */ 1317 return ADD_TO_MLQUEUE; 1318 case GOOD: 1319 case COMMAND_TERMINATED: 1320 case TASK_ABORTED: 1321 return SUCCESS; 1322 case CHECK_CONDITION: 1323 rtn = scsi_check_sense(scmd); 1324 if (rtn == NEEDS_RETRY) 1325 goto maybe_retry; 1326 /* if rtn == FAILED, we have no sense information; 1327 * returning FAILED will wake the error handler thread 1328 * to collect the sense and redo the decide 1329 * disposition */ 1330 return rtn; 1331 case CONDITION_GOOD: 1332 case INTERMEDIATE_GOOD: 1333 case INTERMEDIATE_C_GOOD: 1334 case ACA_ACTIVE: 1335 /* 1336 * who knows? FIXME(eric) 1337 */ 1338 return SUCCESS; 1339 1340 case RESERVATION_CONFLICT: 1341 sdev_printk(KERN_INFO, scmd->device, 1342 "reservation conflict\n"); 1343 return SUCCESS; /* causes immediate i/o error */ 1344 default: 1345 return FAILED; 1346 } 1347 return FAILED; 1348 1349 maybe_retry: 1350 1351 /* we requeue for retry because the error was retryable, and 1352 * the request was not marked fast fail. Note that above, 1353 * even if the request is marked fast fail, we still requeue 1354 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1355 if ((++scmd->retries) < scmd->allowed 1356 && !blk_noretry_request(scmd->request)) { 1357 return NEEDS_RETRY; 1358 } else { 1359 /* 1360 * no more retries - report this one back to upper level. 1361 */ 1362 return SUCCESS; 1363 } 1364 } 1365 1366 /** 1367 * scsi_eh_lock_done - done function for eh door lock request 1368 * @scmd: SCSI command block for the door lock request 1369 * 1370 * Notes: 1371 * We completed the asynchronous door lock request, and it has either 1372 * locked the door or failed. We must free the command structures 1373 * associated with this request. 1374 **/ 1375 static void scsi_eh_lock_done(struct scsi_cmnd *scmd) 1376 { 1377 struct scsi_request *sreq = scmd->sc_request; 1378 1379 scsi_release_request(sreq); 1380 } 1381 1382 1383 /** 1384 * scsi_eh_lock_door - Prevent medium removal for the specified device 1385 * @sdev: SCSI device to prevent medium removal 1386 * 1387 * Locking: 1388 * We must be called from process context; scsi_allocate_request() 1389 * may sleep. 1390 * 1391 * Notes: 1392 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 1393 * head of the devices request queue, and continue. 1394 * 1395 * Bugs: 1396 * scsi_allocate_request() may sleep waiting for existing requests to 1397 * be processed. However, since we haven't kicked off any request 1398 * processing for this host, this may deadlock. 1399 * 1400 * If scsi_allocate_request() fails for what ever reason, we 1401 * completely forget to lock the door. 1402 **/ 1403 static void scsi_eh_lock_door(struct scsi_device *sdev) 1404 { 1405 struct scsi_request *sreq = scsi_allocate_request(sdev, GFP_KERNEL); 1406 1407 if (unlikely(!sreq)) { 1408 printk(KERN_ERR "%s: request allocate failed," 1409 "prevent media removal cmd not sent\n", __FUNCTION__); 1410 return; 1411 } 1412 1413 sreq->sr_cmnd[0] = ALLOW_MEDIUM_REMOVAL; 1414 sreq->sr_cmnd[1] = 0; 1415 sreq->sr_cmnd[2] = 0; 1416 sreq->sr_cmnd[3] = 0; 1417 sreq->sr_cmnd[4] = SCSI_REMOVAL_PREVENT; 1418 sreq->sr_cmnd[5] = 0; 1419 sreq->sr_data_direction = DMA_NONE; 1420 sreq->sr_bufflen = 0; 1421 sreq->sr_buffer = NULL; 1422 sreq->sr_allowed = 5; 1423 sreq->sr_done = scsi_eh_lock_done; 1424 sreq->sr_timeout_per_command = 10 * HZ; 1425 sreq->sr_cmd_len = COMMAND_SIZE(sreq->sr_cmnd[0]); 1426 1427 scsi_insert_special_req(sreq, 1); 1428 } 1429 1430 1431 /** 1432 * scsi_restart_operations - restart io operations to the specified host. 1433 * @shost: Host we are restarting. 1434 * 1435 * Notes: 1436 * When we entered the error handler, we blocked all further i/o to 1437 * this device. we need to 'reverse' this process. 1438 **/ 1439 static void scsi_restart_operations(struct Scsi_Host *shost) 1440 { 1441 struct scsi_device *sdev; 1442 unsigned long flags; 1443 1444 /* 1445 * If the door was locked, we need to insert a door lock request 1446 * onto the head of the SCSI request queue for the device. There 1447 * is no point trying to lock the door of an off-line device. 1448 */ 1449 shost_for_each_device(sdev, shost) { 1450 if (scsi_device_online(sdev) && sdev->locked) 1451 scsi_eh_lock_door(sdev); 1452 } 1453 1454 /* 1455 * next free up anything directly waiting upon the host. this 1456 * will be requests for character device operations, and also for 1457 * ioctls to queued block devices. 1458 */ 1459 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n", 1460 __FUNCTION__)); 1461 1462 spin_lock_irqsave(shost->host_lock, flags); 1463 if (scsi_host_set_state(shost, SHOST_RUNNING)) 1464 if (scsi_host_set_state(shost, SHOST_CANCEL)) 1465 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 1466 spin_unlock_irqrestore(shost->host_lock, flags); 1467 1468 wake_up(&shost->host_wait); 1469 1470 /* 1471 * finally we need to re-initiate requests that may be pending. we will 1472 * have had everything blocked while error handling is taking place, and 1473 * now that error recovery is done, we will need to ensure that these 1474 * requests are started. 1475 */ 1476 scsi_run_host_queues(shost); 1477 } 1478 1479 /** 1480 * scsi_eh_ready_devs - check device ready state and recover if not. 1481 * @shost: host to be recovered. 1482 * @eh_done_q: list_head for processed commands. 1483 * 1484 **/ 1485 static void scsi_eh_ready_devs(struct Scsi_Host *shost, 1486 struct list_head *work_q, 1487 struct list_head *done_q) 1488 { 1489 if (!scsi_eh_stu(shost, work_q, done_q)) 1490 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 1491 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 1492 if (!scsi_eh_host_reset(work_q, done_q)) 1493 scsi_eh_offline_sdevs(work_q, done_q); 1494 } 1495 1496 /** 1497 * scsi_eh_flush_done_q - finish processed commands or retry them. 1498 * @done_q: list_head of processed commands. 1499 * 1500 **/ 1501 static void scsi_eh_flush_done_q(struct list_head *done_q) 1502 { 1503 struct scsi_cmnd *scmd, *next; 1504 1505 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 1506 list_del_init(&scmd->eh_entry); 1507 if (scsi_device_online(scmd->device) && 1508 !blk_noretry_request(scmd->request) && 1509 (++scmd->retries < scmd->allowed)) { 1510 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush" 1511 " retry cmd: %p\n", 1512 current->comm, 1513 scmd)); 1514 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 1515 } else { 1516 /* 1517 * If just we got sense for the device (called 1518 * scsi_eh_get_sense), scmd->result is already 1519 * set, do not set DRIVER_TIMEOUT. 1520 */ 1521 if (!scmd->result) 1522 scmd->result |= (DRIVER_TIMEOUT << 24); 1523 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish" 1524 " cmd: %p\n", 1525 current->comm, scmd)); 1526 scsi_finish_command(scmd); 1527 } 1528 } 1529 } 1530 1531 /** 1532 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 1533 * @shost: Host to unjam. 1534 * 1535 * Notes: 1536 * When we come in here, we *know* that all commands on the bus have 1537 * either completed, failed or timed out. we also know that no further 1538 * commands are being sent to the host, so things are relatively quiet 1539 * and we have freedom to fiddle with things as we wish. 1540 * 1541 * This is only the *default* implementation. it is possible for 1542 * individual drivers to supply their own version of this function, and 1543 * if the maintainer wishes to do this, it is strongly suggested that 1544 * this function be taken as a template and modified. this function 1545 * was designed to correctly handle problems for about 95% of the 1546 * different cases out there, and it should always provide at least a 1547 * reasonable amount of error recovery. 1548 * 1549 * Any command marked 'failed' or 'timeout' must eventually have 1550 * scsi_finish_cmd() called for it. we do all of the retry stuff 1551 * here, so when we restart the host after we return it should have an 1552 * empty queue. 1553 **/ 1554 static void scsi_unjam_host(struct Scsi_Host *shost) 1555 { 1556 unsigned long flags; 1557 LIST_HEAD(eh_work_q); 1558 LIST_HEAD(eh_done_q); 1559 1560 spin_lock_irqsave(shost->host_lock, flags); 1561 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 1562 spin_unlock_irqrestore(shost->host_lock, flags); 1563 1564 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 1565 1566 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 1567 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q)) 1568 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 1569 1570 scsi_eh_flush_done_q(&eh_done_q); 1571 } 1572 1573 /** 1574 * scsi_error_handler - Handle errors/timeouts of SCSI cmds. 1575 * @data: Host for which we are running. 1576 * 1577 * Notes: 1578 * This is always run in the context of a kernel thread. The idea is 1579 * that we start this thing up when the kernel starts up (one per host 1580 * that we detect), and it immediately goes to sleep and waits for some 1581 * event (i.e. failure). When this takes place, we have the job of 1582 * trying to unjam the bus and restarting things. 1583 **/ 1584 int scsi_error_handler(void *data) 1585 { 1586 struct Scsi_Host *shost = (struct Scsi_Host *) data; 1587 int rtn; 1588 1589 current->flags |= PF_NOFREEZE; 1590 1591 1592 /* 1593 * Note - we always use TASK_INTERRUPTIBLE even if the module 1594 * was loaded as part of the kernel. The reason is that 1595 * UNINTERRUPTIBLE would cause this thread to be counted in 1596 * the load average as a running process, and an interruptible 1597 * wait doesn't. 1598 */ 1599 set_current_state(TASK_INTERRUPTIBLE); 1600 while (!kthread_should_stop()) { 1601 if (shost->host_failed == 0 || 1602 shost->host_failed != shost->host_busy) { 1603 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler" 1604 " scsi_eh_%d" 1605 " sleeping\n", 1606 shost->host_no)); 1607 schedule(); 1608 set_current_state(TASK_INTERRUPTIBLE); 1609 continue; 1610 } 1611 1612 __set_current_state(TASK_RUNNING); 1613 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler" 1614 " scsi_eh_%d waking" 1615 " up\n",shost->host_no)); 1616 1617 shost->eh_active = 1; 1618 1619 /* 1620 * We have a host that is failing for some reason. Figure out 1621 * what we need to do to get it up and online again (if we can). 1622 * If we fail, we end up taking the thing offline. 1623 */ 1624 if (shost->hostt->eh_strategy_handler) 1625 rtn = shost->hostt->eh_strategy_handler(shost); 1626 else 1627 scsi_unjam_host(shost); 1628 1629 shost->eh_active = 0; 1630 1631 /* 1632 * Note - if the above fails completely, the action is to take 1633 * individual devices offline and flush the queue of any 1634 * outstanding requests that may have been pending. When we 1635 * restart, we restart any I/O to any other devices on the bus 1636 * which are still online. 1637 */ 1638 scsi_restart_operations(shost); 1639 set_current_state(TASK_INTERRUPTIBLE); 1640 } 1641 1642 __set_current_state(TASK_RUNNING); 1643 1644 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler scsi_eh_%d" 1645 " exiting\n",shost->host_no)); 1646 1647 /* 1648 * Make sure that nobody tries to wake us up again. 1649 */ 1650 shost->ehandler = NULL; 1651 return 0; 1652 } 1653 1654 /* 1655 * Function: scsi_report_bus_reset() 1656 * 1657 * Purpose: Utility function used by low-level drivers to report that 1658 * they have observed a bus reset on the bus being handled. 1659 * 1660 * Arguments: shost - Host in question 1661 * channel - channel on which reset was observed. 1662 * 1663 * Returns: Nothing 1664 * 1665 * Lock status: Host lock must be held. 1666 * 1667 * Notes: This only needs to be called if the reset is one which 1668 * originates from an unknown location. Resets originated 1669 * by the mid-level itself don't need to call this, but there 1670 * should be no harm. 1671 * 1672 * The main purpose of this is to make sure that a CHECK_CONDITION 1673 * is properly treated. 1674 */ 1675 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 1676 { 1677 struct scsi_device *sdev; 1678 1679 __shost_for_each_device(sdev, shost) { 1680 if (channel == sdev_channel(sdev)) { 1681 sdev->was_reset = 1; 1682 sdev->expecting_cc_ua = 1; 1683 } 1684 } 1685 } 1686 EXPORT_SYMBOL(scsi_report_bus_reset); 1687 1688 /* 1689 * Function: scsi_report_device_reset() 1690 * 1691 * Purpose: Utility function used by low-level drivers to report that 1692 * they have observed a device reset on the device being handled. 1693 * 1694 * Arguments: shost - Host in question 1695 * channel - channel on which reset was observed 1696 * target - target on which reset was observed 1697 * 1698 * Returns: Nothing 1699 * 1700 * Lock status: Host lock must be held 1701 * 1702 * Notes: This only needs to be called if the reset is one which 1703 * originates from an unknown location. Resets originated 1704 * by the mid-level itself don't need to call this, but there 1705 * should be no harm. 1706 * 1707 * The main purpose of this is to make sure that a CHECK_CONDITION 1708 * is properly treated. 1709 */ 1710 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 1711 { 1712 struct scsi_device *sdev; 1713 1714 __shost_for_each_device(sdev, shost) { 1715 if (channel == sdev_channel(sdev) && 1716 target == sdev_id(sdev)) { 1717 sdev->was_reset = 1; 1718 sdev->expecting_cc_ua = 1; 1719 } 1720 } 1721 } 1722 EXPORT_SYMBOL(scsi_report_device_reset); 1723 1724 static void 1725 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 1726 { 1727 } 1728 1729 /* 1730 * Function: scsi_reset_provider 1731 * 1732 * Purpose: Send requested reset to a bus or device at any phase. 1733 * 1734 * Arguments: device - device to send reset to 1735 * flag - reset type (see scsi.h) 1736 * 1737 * Returns: SUCCESS/FAILURE. 1738 * 1739 * Notes: This is used by the SCSI Generic driver to provide 1740 * Bus/Device reset capability. 1741 */ 1742 int 1743 scsi_reset_provider(struct scsi_device *dev, int flag) 1744 { 1745 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL); 1746 struct request req; 1747 int rtn; 1748 1749 scmd->request = &req; 1750 memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout)); 1751 scmd->request->rq_status = RQ_SCSI_BUSY; 1752 1753 memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd)); 1754 1755 scmd->scsi_done = scsi_reset_provider_done_command; 1756 scmd->done = NULL; 1757 scmd->buffer = NULL; 1758 scmd->bufflen = 0; 1759 scmd->request_buffer = NULL; 1760 scmd->request_bufflen = 0; 1761 1762 scmd->cmd_len = 0; 1763 1764 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 1765 scmd->sc_request = NULL; 1766 scmd->sc_magic = SCSI_CMND_MAGIC; 1767 1768 init_timer(&scmd->eh_timeout); 1769 1770 /* 1771 * Sometimes the command can get back into the timer chain, 1772 * so use the pid as an identifier. 1773 */ 1774 scmd->pid = 0; 1775 1776 switch (flag) { 1777 case SCSI_TRY_RESET_DEVICE: 1778 rtn = scsi_try_bus_device_reset(scmd); 1779 if (rtn == SUCCESS) 1780 break; 1781 /* FALLTHROUGH */ 1782 case SCSI_TRY_RESET_BUS: 1783 rtn = scsi_try_bus_reset(scmd); 1784 if (rtn == SUCCESS) 1785 break; 1786 /* FALLTHROUGH */ 1787 case SCSI_TRY_RESET_HOST: 1788 rtn = scsi_try_host_reset(scmd); 1789 break; 1790 default: 1791 rtn = FAILED; 1792 } 1793 1794 scsi_next_command(scmd); 1795 return rtn; 1796 } 1797 EXPORT_SYMBOL(scsi_reset_provider); 1798 1799 /** 1800 * scsi_normalize_sense - normalize main elements from either fixed or 1801 * descriptor sense data format into a common format. 1802 * 1803 * @sense_buffer: byte array containing sense data returned by device 1804 * @sb_len: number of valid bytes in sense_buffer 1805 * @sshdr: pointer to instance of structure that common 1806 * elements are written to. 1807 * 1808 * Notes: 1809 * The "main elements" from sense data are: response_code, sense_key, 1810 * asc, ascq and additional_length (only for descriptor format). 1811 * 1812 * Typically this function can be called after a device has 1813 * responded to a SCSI command with the CHECK_CONDITION status. 1814 * 1815 * Return value: 1816 * 1 if valid sense data information found, else 0; 1817 **/ 1818 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len, 1819 struct scsi_sense_hdr *sshdr) 1820 { 1821 if (!sense_buffer || !sb_len) 1822 return 0; 1823 1824 memset(sshdr, 0, sizeof(struct scsi_sense_hdr)); 1825 1826 sshdr->response_code = (sense_buffer[0] & 0x7f); 1827 1828 if (!scsi_sense_valid(sshdr)) 1829 return 0; 1830 1831 if (sshdr->response_code >= 0x72) { 1832 /* 1833 * descriptor format 1834 */ 1835 if (sb_len > 1) 1836 sshdr->sense_key = (sense_buffer[1] & 0xf); 1837 if (sb_len > 2) 1838 sshdr->asc = sense_buffer[2]; 1839 if (sb_len > 3) 1840 sshdr->ascq = sense_buffer[3]; 1841 if (sb_len > 7) 1842 sshdr->additional_length = sense_buffer[7]; 1843 } else { 1844 /* 1845 * fixed format 1846 */ 1847 if (sb_len > 2) 1848 sshdr->sense_key = (sense_buffer[2] & 0xf); 1849 if (sb_len > 7) { 1850 sb_len = (sb_len < (sense_buffer[7] + 8)) ? 1851 sb_len : (sense_buffer[7] + 8); 1852 if (sb_len > 12) 1853 sshdr->asc = sense_buffer[12]; 1854 if (sb_len > 13) 1855 sshdr->ascq = sense_buffer[13]; 1856 } 1857 } 1858 1859 return 1; 1860 } 1861 EXPORT_SYMBOL(scsi_normalize_sense); 1862 1863 int scsi_request_normalize_sense(struct scsi_request *sreq, 1864 struct scsi_sense_hdr *sshdr) 1865 { 1866 return scsi_normalize_sense(sreq->sr_sense_buffer, 1867 sizeof(sreq->sr_sense_buffer), sshdr); 1868 } 1869 EXPORT_SYMBOL(scsi_request_normalize_sense); 1870 1871 int scsi_command_normalize_sense(struct scsi_cmnd *cmd, 1872 struct scsi_sense_hdr *sshdr) 1873 { 1874 return scsi_normalize_sense(cmd->sense_buffer, 1875 sizeof(cmd->sense_buffer), sshdr); 1876 } 1877 EXPORT_SYMBOL(scsi_command_normalize_sense); 1878 1879 /** 1880 * scsi_sense_desc_find - search for a given descriptor type in 1881 * descriptor sense data format. 1882 * 1883 * @sense_buffer: byte array of descriptor format sense data 1884 * @sb_len: number of valid bytes in sense_buffer 1885 * @desc_type: value of descriptor type to find 1886 * (e.g. 0 -> information) 1887 * 1888 * Notes: 1889 * only valid when sense data is in descriptor format 1890 * 1891 * Return value: 1892 * pointer to start of (first) descriptor if found else NULL 1893 **/ 1894 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len, 1895 int desc_type) 1896 { 1897 int add_sen_len, add_len, desc_len, k; 1898 const u8 * descp; 1899 1900 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7]))) 1901 return NULL; 1902 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73)) 1903 return NULL; 1904 add_sen_len = (add_sen_len < (sb_len - 8)) ? 1905 add_sen_len : (sb_len - 8); 1906 descp = &sense_buffer[8]; 1907 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) { 1908 descp += desc_len; 1909 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1; 1910 desc_len = add_len + 2; 1911 if (descp[0] == desc_type) 1912 return descp; 1913 if (add_len < 0) // short descriptor ?? 1914 break; 1915 } 1916 return NULL; 1917 } 1918 EXPORT_SYMBOL(scsi_sense_desc_find); 1919 1920 /** 1921 * scsi_get_sense_info_fld - attempts to get information field from 1922 * sense data (either fixed or descriptor format) 1923 * 1924 * @sense_buffer: byte array of sense data 1925 * @sb_len: number of valid bytes in sense_buffer 1926 * @info_out: pointer to 64 integer where 8 or 4 byte information 1927 * field will be placed if found. 1928 * 1929 * Return value: 1930 * 1 if information field found, 0 if not found. 1931 **/ 1932 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len, 1933 u64 * info_out) 1934 { 1935 int j; 1936 const u8 * ucp; 1937 u64 ull; 1938 1939 if (sb_len < 7) 1940 return 0; 1941 switch (sense_buffer[0] & 0x7f) { 1942 case 0x70: 1943 case 0x71: 1944 if (sense_buffer[0] & 0x80) { 1945 *info_out = (sense_buffer[3] << 24) + 1946 (sense_buffer[4] << 16) + 1947 (sense_buffer[5] << 8) + sense_buffer[6]; 1948 return 1; 1949 } else 1950 return 0; 1951 case 0x72: 1952 case 0x73: 1953 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 1954 0 /* info desc */); 1955 if (ucp && (0xa == ucp[1])) { 1956 ull = 0; 1957 for (j = 0; j < 8; ++j) { 1958 if (j > 0) 1959 ull <<= 8; 1960 ull |= ucp[4 + j]; 1961 } 1962 *info_out = ull; 1963 return 1; 1964 } else 1965 return 0; 1966 default: 1967 return 0; 1968 } 1969 } 1970 EXPORT_SYMBOL(scsi_get_sense_info_fld); 1971