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_done - Completion function for error handling. 421 * @scmd: Cmd that is done. 422 **/ 423 static void scsi_eh_done(struct scsi_cmnd *scmd) 424 { 425 struct completion *eh_action; 426 427 SCSI_LOG_ERROR_RECOVERY(3, 428 printk("%s scmd: %p result: %x\n", 429 __FUNCTION__, scmd, scmd->result)); 430 431 eh_action = scmd->device->host->eh_action; 432 if (eh_action) 433 complete(eh_action); 434 } 435 436 /** 437 * scsi_send_eh_cmnd - send a cmd to a device as part of error recovery. 438 * @scmd: SCSI Cmd to send. 439 * @timeout: Timeout for cmd. 440 * 441 * Return value: 442 * SUCCESS or FAILED or NEEDS_RETRY 443 **/ 444 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, int timeout) 445 { 446 struct scsi_device *sdev = scmd->device; 447 struct Scsi_Host *shost = sdev->host; 448 DECLARE_COMPLETION(done); 449 unsigned long timeleft; 450 unsigned long flags; 451 int rtn; 452 453 if (sdev->scsi_level <= SCSI_2) 454 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 455 (sdev->lun << 5 & 0xe0); 456 457 shost->eh_action = &done; 458 scmd->request->rq_status = RQ_SCSI_BUSY; 459 460 spin_lock_irqsave(shost->host_lock, flags); 461 scsi_log_send(scmd); 462 shost->hostt->queuecommand(scmd, scsi_eh_done); 463 spin_unlock_irqrestore(shost->host_lock, flags); 464 465 timeleft = wait_for_completion_timeout(&done, timeout); 466 467 scmd->request->rq_status = RQ_SCSI_DONE; 468 shost->eh_action = NULL; 469 470 scsi_log_completion(scmd, SUCCESS); 471 472 SCSI_LOG_ERROR_RECOVERY(3, 473 printk("%s: scmd: %p, timeleft: %ld\n", 474 __FUNCTION__, scmd, timeleft)); 475 476 /* 477 * If there is time left scsi_eh_done got called, and we will 478 * examine the actual status codes to see whether the command 479 * actually did complete normally, else tell the host to forget 480 * about this command. 481 */ 482 if (timeleft) { 483 rtn = scsi_eh_completed_normally(scmd); 484 SCSI_LOG_ERROR_RECOVERY(3, 485 printk("%s: scsi_eh_completed_normally %x\n", 486 __FUNCTION__, rtn)); 487 488 switch (rtn) { 489 case SUCCESS: 490 case NEEDS_RETRY: 491 case FAILED: 492 break; 493 default: 494 rtn = FAILED; 495 break; 496 } 497 } else { 498 /* 499 * FIXME(eric) - we are not tracking whether we could 500 * abort a timed out command or not. not sure how 501 * we should treat them differently anyways. 502 */ 503 if (shost->hostt->eh_abort_handler) 504 shost->hostt->eh_abort_handler(scmd); 505 rtn = FAILED; 506 } 507 508 return rtn; 509 } 510 511 /** 512 * scsi_request_sense - Request sense data from a particular target. 513 * @scmd: SCSI cmd for request sense. 514 * 515 * Notes: 516 * Some hosts automatically obtain this information, others require 517 * that we obtain it on our own. This function will *not* return until 518 * the command either times out, or it completes. 519 **/ 520 static int scsi_request_sense(struct scsi_cmnd *scmd) 521 { 522 static unsigned char generic_sense[6] = 523 {REQUEST_SENSE, 0, 0, 0, 252, 0}; 524 unsigned char *scsi_result; 525 int saved_result; 526 int rtn; 527 528 memcpy(scmd->cmnd, generic_sense, sizeof(generic_sense)); 529 530 scsi_result = kmalloc(252, GFP_ATOMIC | ((scmd->device->host->hostt->unchecked_isa_dma) ? __GFP_DMA : 0)); 531 532 533 if (unlikely(!scsi_result)) { 534 printk(KERN_ERR "%s: cannot allocate scsi_result.\n", 535 __FUNCTION__); 536 return FAILED; 537 } 538 539 /* 540 * zero the sense buffer. some host adapters automatically always 541 * request sense, so it is not a good idea that 542 * scmd->request_buffer and scmd->sense_buffer point to the same 543 * address (db). 0 is not a valid sense code. 544 */ 545 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); 546 memset(scsi_result, 0, 252); 547 548 saved_result = scmd->result; 549 scmd->request_buffer = scsi_result; 550 scmd->request_bufflen = 252; 551 scmd->use_sg = 0; 552 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 553 scmd->sc_data_direction = DMA_FROM_DEVICE; 554 scmd->underflow = 0; 555 556 rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT); 557 558 /* last chance to have valid sense data */ 559 if(!SCSI_SENSE_VALID(scmd)) { 560 memcpy(scmd->sense_buffer, scmd->request_buffer, 561 sizeof(scmd->sense_buffer)); 562 } 563 564 kfree(scsi_result); 565 566 /* 567 * when we eventually call scsi_finish, we really wish to complete 568 * the original request, so let's restore the original data. (db) 569 */ 570 scsi_setup_cmd_retry(scmd); 571 scmd->result = saved_result; 572 return rtn; 573 } 574 575 /** 576 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 577 * @scmd: Original SCSI cmd that eh has finished. 578 * @done_q: Queue for processed commands. 579 * 580 * Notes: 581 * We don't want to use the normal command completion while we are are 582 * still handling errors - it may cause other commands to be queued, 583 * and that would disturb what we are doing. thus we really want to 584 * keep a list of pending commands for final completion, and once we 585 * are ready to leave error handling we handle completion for real. 586 **/ 587 static void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, 588 struct list_head *done_q) 589 { 590 scmd->device->host->host_failed--; 591 scmd->eh_eflags = 0; 592 593 /* 594 * set this back so that the upper level can correctly free up 595 * things. 596 */ 597 scsi_setup_cmd_retry(scmd); 598 list_move_tail(&scmd->eh_entry, done_q); 599 } 600 601 /** 602 * scsi_eh_get_sense - Get device sense data. 603 * @work_q: Queue of commands to process. 604 * @done_q: Queue of proccessed commands.. 605 * 606 * Description: 607 * See if we need to request sense information. if so, then get it 608 * now, so we have a better idea of what to do. 609 * 610 * Notes: 611 * This has the unfortunate side effect that if a shost adapter does 612 * not automatically request sense information, that we end up shutting 613 * it down before we request it. 614 * 615 * All drivers should request sense information internally these days, 616 * so for now all I have to say is tough noogies if you end up in here. 617 * 618 * XXX: Long term this code should go away, but that needs an audit of 619 * all LLDDs first. 620 **/ 621 static int scsi_eh_get_sense(struct list_head *work_q, 622 struct list_head *done_q) 623 { 624 struct scsi_cmnd *scmd, *next; 625 int rtn; 626 627 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 628 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) || 629 SCSI_SENSE_VALID(scmd)) 630 continue; 631 632 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 633 "%s: requesting sense\n", 634 current->comm)); 635 rtn = scsi_request_sense(scmd); 636 if (rtn != SUCCESS) 637 continue; 638 639 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p" 640 " result %x\n", scmd, 641 scmd->result)); 642 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd)); 643 644 rtn = scsi_decide_disposition(scmd); 645 646 /* 647 * if the result was normal, then just pass it along to the 648 * upper level. 649 */ 650 if (rtn == SUCCESS) 651 /* we don't want this command reissued, just 652 * finished with the sense data, so set 653 * retries to the max allowed to ensure it 654 * won't get reissued */ 655 scmd->retries = scmd->allowed; 656 else if (rtn != NEEDS_RETRY) 657 continue; 658 659 scsi_eh_finish_cmd(scmd, done_q); 660 } 661 662 return list_empty(work_q); 663 } 664 665 /** 666 * scsi_try_to_abort_cmd - Ask host to abort a running command. 667 * @scmd: SCSI cmd to abort from Lower Level. 668 * 669 * Notes: 670 * This function will not return until the user's completion function 671 * has been called. there is no timeout on this operation. if the 672 * author of the low-level driver wishes this operation to be timed, 673 * they can provide this facility themselves. helper functions in 674 * scsi_error.c can be supplied to make this easier to do. 675 **/ 676 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd) 677 { 678 if (!scmd->device->host->hostt->eh_abort_handler) 679 return FAILED; 680 681 /* 682 * scsi_done was called just after the command timed out and before 683 * we had a chance to process it. (db) 684 */ 685 if (scmd->serial_number == 0) 686 return SUCCESS; 687 return scmd->device->host->hostt->eh_abort_handler(scmd); 688 } 689 690 /** 691 * scsi_eh_tur - Send TUR to device. 692 * @scmd: Scsi cmd to send TUR 693 * 694 * Return value: 695 * 0 - Device is ready. 1 - Device NOT ready. 696 **/ 697 static int scsi_eh_tur(struct scsi_cmnd *scmd) 698 { 699 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 700 int retry_cnt = 1, rtn; 701 int saved_result; 702 703 retry_tur: 704 memcpy(scmd->cmnd, tur_command, sizeof(tur_command)); 705 706 /* 707 * zero the sense buffer. the scsi spec mandates that any 708 * untransferred sense data should be interpreted as being zero. 709 */ 710 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); 711 712 saved_result = scmd->result; 713 scmd->request_buffer = NULL; 714 scmd->request_bufflen = 0; 715 scmd->use_sg = 0; 716 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 717 scmd->underflow = 0; 718 scmd->sc_data_direction = DMA_NONE; 719 720 rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT); 721 722 /* 723 * when we eventually call scsi_finish, we really wish to complete 724 * the original request, so let's restore the original data. (db) 725 */ 726 scsi_setup_cmd_retry(scmd); 727 scmd->result = saved_result; 728 729 /* 730 * hey, we are done. let's look to see what happened. 731 */ 732 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", 733 __FUNCTION__, scmd, rtn)); 734 if (rtn == SUCCESS) 735 return 0; 736 else if (rtn == NEEDS_RETRY) { 737 if (retry_cnt--) 738 goto retry_tur; 739 return 0; 740 } 741 return 1; 742 } 743 744 /** 745 * scsi_eh_abort_cmds - abort canceled commands. 746 * @shost: scsi host being recovered. 747 * @eh_done_q: list_head for processed commands. 748 * 749 * Decription: 750 * Try and see whether or not it makes sense to try and abort the 751 * running command. this only works out to be the case if we have one 752 * command that has timed out. if the command simply failed, it makes 753 * no sense to try and abort the command, since as far as the shost 754 * adapter is concerned, it isn't running. 755 **/ 756 static int scsi_eh_abort_cmds(struct list_head *work_q, 757 struct list_head *done_q) 758 { 759 struct scsi_cmnd *scmd, *next; 760 int rtn; 761 762 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 763 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD)) 764 continue; 765 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:" 766 "0x%p\n", current->comm, 767 scmd)); 768 rtn = scsi_try_to_abort_cmd(scmd); 769 if (rtn == SUCCESS) { 770 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD; 771 if (!scsi_device_online(scmd->device) || 772 !scsi_eh_tur(scmd)) { 773 scsi_eh_finish_cmd(scmd, done_q); 774 } 775 776 } else 777 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting" 778 " cmd failed:" 779 "0x%p\n", 780 current->comm, 781 scmd)); 782 } 783 784 return list_empty(work_q); 785 } 786 787 /** 788 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 789 * @scmd: SCSI cmd used to send BDR 790 * 791 * Notes: 792 * There is no timeout for this operation. if this operation is 793 * unreliable for a given host, then the host itself needs to put a 794 * timer on it, and set the host back to a consistent state prior to 795 * returning. 796 **/ 797 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 798 { 799 int rtn; 800 801 if (!scmd->device->host->hostt->eh_device_reset_handler) 802 return FAILED; 803 804 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd); 805 if (rtn == SUCCESS) { 806 scmd->device->was_reset = 1; 807 scmd->device->expecting_cc_ua = 1; 808 } 809 810 return rtn; 811 } 812 813 /** 814 * scsi_eh_try_stu - Send START_UNIT to device. 815 * @scmd: Scsi cmd to send START_UNIT 816 * 817 * Return value: 818 * 0 - Device is ready. 1 - Device NOT ready. 819 **/ 820 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 821 { 822 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 823 int rtn; 824 int saved_result; 825 826 if (!scmd->device->allow_restart) 827 return 1; 828 829 memcpy(scmd->cmnd, stu_command, sizeof(stu_command)); 830 831 /* 832 * zero the sense buffer. the scsi spec mandates that any 833 * untransferred sense data should be interpreted as being zero. 834 */ 835 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); 836 837 saved_result = scmd->result; 838 scmd->request_buffer = NULL; 839 scmd->request_bufflen = 0; 840 scmd->use_sg = 0; 841 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 842 scmd->underflow = 0; 843 scmd->sc_data_direction = DMA_NONE; 844 845 rtn = scsi_send_eh_cmnd(scmd, START_UNIT_TIMEOUT); 846 847 /* 848 * when we eventually call scsi_finish, we really wish to complete 849 * the original request, so let's restore the original data. (db) 850 */ 851 scsi_setup_cmd_retry(scmd); 852 scmd->result = saved_result; 853 854 /* 855 * hey, we are done. let's look to see what happened. 856 */ 857 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", 858 __FUNCTION__, scmd, rtn)); 859 if (rtn == SUCCESS) 860 return 0; 861 return 1; 862 } 863 864 /** 865 * scsi_eh_stu - send START_UNIT if needed 866 * @shost: scsi host being recovered. 867 * @eh_done_q: list_head for processed commands. 868 * 869 * Notes: 870 * If commands are failing due to not ready, initializing command required, 871 * try revalidating the device, which will end up sending a start unit. 872 **/ 873 static int scsi_eh_stu(struct Scsi_Host *shost, 874 struct list_head *work_q, 875 struct list_head *done_q) 876 { 877 struct scsi_cmnd *scmd, *stu_scmd, *next; 878 struct scsi_device *sdev; 879 880 shost_for_each_device(sdev, shost) { 881 stu_scmd = NULL; 882 list_for_each_entry(scmd, work_q, eh_entry) 883 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 884 scsi_check_sense(scmd) == FAILED ) { 885 stu_scmd = scmd; 886 break; 887 } 888 889 if (!stu_scmd) 890 continue; 891 892 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:" 893 " 0x%p\n", current->comm, sdev)); 894 895 if (!scsi_eh_try_stu(stu_scmd)) { 896 if (!scsi_device_online(sdev) || 897 !scsi_eh_tur(stu_scmd)) { 898 list_for_each_entry_safe(scmd, next, 899 work_q, eh_entry) { 900 if (scmd->device == sdev) 901 scsi_eh_finish_cmd(scmd, done_q); 902 } 903 } 904 } else { 905 SCSI_LOG_ERROR_RECOVERY(3, 906 printk("%s: START_UNIT failed to sdev:" 907 " 0x%p\n", current->comm, sdev)); 908 } 909 } 910 911 return list_empty(work_q); 912 } 913 914 915 /** 916 * scsi_eh_bus_device_reset - send bdr if needed 917 * @shost: scsi host being recovered. 918 * @eh_done_q: list_head for processed commands. 919 * 920 * Notes: 921 * Try a bus device reset. still, look to see whether we have multiple 922 * devices that are jammed or not - if we have multiple devices, it 923 * makes no sense to try bus_device_reset - we really would need to try 924 * a bus_reset instead. 925 **/ 926 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 927 struct list_head *work_q, 928 struct list_head *done_q) 929 { 930 struct scsi_cmnd *scmd, *bdr_scmd, *next; 931 struct scsi_device *sdev; 932 int rtn; 933 934 shost_for_each_device(sdev, shost) { 935 bdr_scmd = NULL; 936 list_for_each_entry(scmd, work_q, eh_entry) 937 if (scmd->device == sdev) { 938 bdr_scmd = scmd; 939 break; 940 } 941 942 if (!bdr_scmd) 943 continue; 944 945 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:" 946 " 0x%p\n", current->comm, 947 sdev)); 948 rtn = scsi_try_bus_device_reset(bdr_scmd); 949 if (rtn == SUCCESS) { 950 if (!scsi_device_online(sdev) || 951 !scsi_eh_tur(bdr_scmd)) { 952 list_for_each_entry_safe(scmd, next, 953 work_q, eh_entry) { 954 if (scmd->device == sdev) 955 scsi_eh_finish_cmd(scmd, 956 done_q); 957 } 958 } 959 } else { 960 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR" 961 " failed sdev:" 962 "0x%p\n", 963 current->comm, 964 sdev)); 965 } 966 } 967 968 return list_empty(work_q); 969 } 970 971 /** 972 * scsi_try_bus_reset - ask host to perform a bus reset 973 * @scmd: SCSI cmd to send bus reset. 974 **/ 975 static int scsi_try_bus_reset(struct scsi_cmnd *scmd) 976 { 977 unsigned long flags; 978 int rtn; 979 980 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n", 981 __FUNCTION__)); 982 983 if (!scmd->device->host->hostt->eh_bus_reset_handler) 984 return FAILED; 985 986 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd); 987 988 if (rtn == SUCCESS) { 989 if (!scmd->device->host->hostt->skip_settle_delay) 990 ssleep(BUS_RESET_SETTLE_TIME); 991 spin_lock_irqsave(scmd->device->host->host_lock, flags); 992 scsi_report_bus_reset(scmd->device->host, 993 scmd_channel(scmd)); 994 spin_unlock_irqrestore(scmd->device->host->host_lock, flags); 995 } 996 997 return rtn; 998 } 999 1000 /** 1001 * scsi_try_host_reset - ask host adapter to reset itself 1002 * @scmd: SCSI cmd to send hsot reset. 1003 **/ 1004 static int scsi_try_host_reset(struct scsi_cmnd *scmd) 1005 { 1006 unsigned long flags; 1007 int rtn; 1008 1009 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n", 1010 __FUNCTION__)); 1011 1012 if (!scmd->device->host->hostt->eh_host_reset_handler) 1013 return FAILED; 1014 1015 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd); 1016 1017 if (rtn == SUCCESS) { 1018 if (!scmd->device->host->hostt->skip_settle_delay) 1019 ssleep(HOST_RESET_SETTLE_TIME); 1020 spin_lock_irqsave(scmd->device->host->host_lock, flags); 1021 scsi_report_bus_reset(scmd->device->host, 1022 scmd_channel(scmd)); 1023 spin_unlock_irqrestore(scmd->device->host->host_lock, flags); 1024 } 1025 1026 return rtn; 1027 } 1028 1029 /** 1030 * scsi_eh_bus_reset - send a bus reset 1031 * @shost: scsi host being recovered. 1032 * @eh_done_q: list_head for processed commands. 1033 **/ 1034 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1035 struct list_head *work_q, 1036 struct list_head *done_q) 1037 { 1038 struct scsi_cmnd *scmd, *chan_scmd, *next; 1039 unsigned int channel; 1040 int rtn; 1041 1042 /* 1043 * we really want to loop over the various channels, and do this on 1044 * a channel by channel basis. we should also check to see if any 1045 * of the failed commands are on soft_reset devices, and if so, skip 1046 * the reset. 1047 */ 1048 1049 for (channel = 0; channel <= shost->max_channel; channel++) { 1050 chan_scmd = NULL; 1051 list_for_each_entry(scmd, work_q, eh_entry) { 1052 if (channel == scmd_channel(scmd)) { 1053 chan_scmd = scmd; 1054 break; 1055 /* 1056 * FIXME add back in some support for 1057 * soft_reset devices. 1058 */ 1059 } 1060 } 1061 1062 if (!chan_scmd) 1063 continue; 1064 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:" 1065 " %d\n", current->comm, 1066 channel)); 1067 rtn = scsi_try_bus_reset(chan_scmd); 1068 if (rtn == SUCCESS) { 1069 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1070 if (channel == scmd_channel(scmd)) 1071 if (!scsi_device_online(scmd->device) || 1072 !scsi_eh_tur(scmd)) 1073 scsi_eh_finish_cmd(scmd, 1074 done_q); 1075 } 1076 } else { 1077 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST" 1078 " failed chan: %d\n", 1079 current->comm, 1080 channel)); 1081 } 1082 } 1083 return list_empty(work_q); 1084 } 1085 1086 /** 1087 * scsi_eh_host_reset - send a host reset 1088 * @work_q: list_head for processed commands. 1089 * @done_q: list_head for processed commands. 1090 **/ 1091 static int scsi_eh_host_reset(struct list_head *work_q, 1092 struct list_head *done_q) 1093 { 1094 struct scsi_cmnd *scmd, *next; 1095 int rtn; 1096 1097 if (!list_empty(work_q)) { 1098 scmd = list_entry(work_q->next, 1099 struct scsi_cmnd, eh_entry); 1100 1101 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n" 1102 , current->comm)); 1103 1104 rtn = scsi_try_host_reset(scmd); 1105 if (rtn == SUCCESS) { 1106 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1107 if (!scsi_device_online(scmd->device) || 1108 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) || 1109 !scsi_eh_tur(scmd)) 1110 scsi_eh_finish_cmd(scmd, done_q); 1111 } 1112 } else { 1113 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST" 1114 " failed\n", 1115 current->comm)); 1116 } 1117 } 1118 return list_empty(work_q); 1119 } 1120 1121 /** 1122 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1123 * @work_q: list_head for processed commands. 1124 * @done_q: list_head for processed commands. 1125 * 1126 **/ 1127 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1128 struct list_head *done_q) 1129 { 1130 struct scsi_cmnd *scmd, *next; 1131 1132 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1133 sdev_printk(KERN_INFO, scmd->device, 1134 "scsi: Device offlined - not" 1135 " ready after error recovery\n"); 1136 scsi_device_set_state(scmd->device, SDEV_OFFLINE); 1137 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) { 1138 /* 1139 * FIXME: Handle lost cmds. 1140 */ 1141 } 1142 scsi_eh_finish_cmd(scmd, done_q); 1143 } 1144 return; 1145 } 1146 1147 /** 1148 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1149 * @scmd: SCSI cmd to examine. 1150 * 1151 * Notes: 1152 * This is *only* called when we are examining the status after sending 1153 * out the actual data command. any commands that are queued for error 1154 * recovery (e.g. test_unit_ready) do *not* come through here. 1155 * 1156 * When this routine returns failed, it means the error handler thread 1157 * is woken. In cases where the error code indicates an error that 1158 * doesn't require the error handler read (i.e. we don't need to 1159 * abort/reset), this function should return SUCCESS. 1160 **/ 1161 int scsi_decide_disposition(struct scsi_cmnd *scmd) 1162 { 1163 int rtn; 1164 1165 /* 1166 * if the device is offline, then we clearly just pass the result back 1167 * up to the top level. 1168 */ 1169 if (!scsi_device_online(scmd->device)) { 1170 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report" 1171 " as SUCCESS\n", 1172 __FUNCTION__)); 1173 return SUCCESS; 1174 } 1175 1176 /* 1177 * first check the host byte, to see if there is anything in there 1178 * that would indicate what we need to do. 1179 */ 1180 switch (host_byte(scmd->result)) { 1181 case DID_PASSTHROUGH: 1182 /* 1183 * no matter what, pass this through to the upper layer. 1184 * nuke this special code so that it looks like we are saying 1185 * did_ok. 1186 */ 1187 scmd->result &= 0xff00ffff; 1188 return SUCCESS; 1189 case DID_OK: 1190 /* 1191 * looks good. drop through, and check the next byte. 1192 */ 1193 break; 1194 case DID_NO_CONNECT: 1195 case DID_BAD_TARGET: 1196 case DID_ABORT: 1197 /* 1198 * note - this means that we just report the status back 1199 * to the top level driver, not that we actually think 1200 * that it indicates SUCCESS. 1201 */ 1202 return SUCCESS; 1203 /* 1204 * when the low level driver returns did_soft_error, 1205 * it is responsible for keeping an internal retry counter 1206 * in order to avoid endless loops (db) 1207 * 1208 * actually this is a bug in this function here. we should 1209 * be mindful of the maximum number of retries specified 1210 * and not get stuck in a loop. 1211 */ 1212 case DID_SOFT_ERROR: 1213 goto maybe_retry; 1214 case DID_IMM_RETRY: 1215 return NEEDS_RETRY; 1216 1217 case DID_REQUEUE: 1218 return ADD_TO_MLQUEUE; 1219 1220 case DID_ERROR: 1221 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1222 status_byte(scmd->result) == RESERVATION_CONFLICT) 1223 /* 1224 * execute reservation conflict processing code 1225 * lower down 1226 */ 1227 break; 1228 /* fallthrough */ 1229 1230 case DID_BUS_BUSY: 1231 case DID_PARITY: 1232 goto maybe_retry; 1233 case DID_TIME_OUT: 1234 /* 1235 * when we scan the bus, we get timeout messages for 1236 * these commands if there is no device available. 1237 * other hosts report did_no_connect for the same thing. 1238 */ 1239 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1240 scmd->cmnd[0] == INQUIRY)) { 1241 return SUCCESS; 1242 } else { 1243 return FAILED; 1244 } 1245 case DID_RESET: 1246 return SUCCESS; 1247 default: 1248 return FAILED; 1249 } 1250 1251 /* 1252 * next, check the message byte. 1253 */ 1254 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 1255 return FAILED; 1256 1257 /* 1258 * check the status byte to see if this indicates anything special. 1259 */ 1260 switch (status_byte(scmd->result)) { 1261 case QUEUE_FULL: 1262 /* 1263 * the case of trying to send too many commands to a 1264 * tagged queueing device. 1265 */ 1266 case BUSY: 1267 /* 1268 * device can't talk to us at the moment. Should only 1269 * occur (SAM-3) when the task queue is empty, so will cause 1270 * the empty queue handling to trigger a stall in the 1271 * device. 1272 */ 1273 return ADD_TO_MLQUEUE; 1274 case GOOD: 1275 case COMMAND_TERMINATED: 1276 case TASK_ABORTED: 1277 return SUCCESS; 1278 case CHECK_CONDITION: 1279 rtn = scsi_check_sense(scmd); 1280 if (rtn == NEEDS_RETRY) 1281 goto maybe_retry; 1282 /* if rtn == FAILED, we have no sense information; 1283 * returning FAILED will wake the error handler thread 1284 * to collect the sense and redo the decide 1285 * disposition */ 1286 return rtn; 1287 case CONDITION_GOOD: 1288 case INTERMEDIATE_GOOD: 1289 case INTERMEDIATE_C_GOOD: 1290 case ACA_ACTIVE: 1291 /* 1292 * who knows? FIXME(eric) 1293 */ 1294 return SUCCESS; 1295 1296 case RESERVATION_CONFLICT: 1297 sdev_printk(KERN_INFO, scmd->device, 1298 "reservation conflict\n"); 1299 return SUCCESS; /* causes immediate i/o error */ 1300 default: 1301 return FAILED; 1302 } 1303 return FAILED; 1304 1305 maybe_retry: 1306 1307 /* we requeue for retry because the error was retryable, and 1308 * the request was not marked fast fail. Note that above, 1309 * even if the request is marked fast fail, we still requeue 1310 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1311 if ((++scmd->retries) < scmd->allowed 1312 && !blk_noretry_request(scmd->request)) { 1313 return NEEDS_RETRY; 1314 } else { 1315 /* 1316 * no more retries - report this one back to upper level. 1317 */ 1318 return SUCCESS; 1319 } 1320 } 1321 1322 /** 1323 * scsi_eh_lock_done - done function for eh door lock request 1324 * @scmd: SCSI command block for the door lock request 1325 * 1326 * Notes: 1327 * We completed the asynchronous door lock request, and it has either 1328 * locked the door or failed. We must free the command structures 1329 * associated with this request. 1330 **/ 1331 static void scsi_eh_lock_done(struct scsi_cmnd *scmd) 1332 { 1333 struct scsi_request *sreq = scmd->sc_request; 1334 1335 scsi_release_request(sreq); 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 struct scsi_request *sreq = scsi_allocate_request(sdev, GFP_KERNEL); 1362 1363 if (unlikely(!sreq)) { 1364 printk(KERN_ERR "%s: request allocate failed," 1365 "prevent media removal cmd not sent\n", __FUNCTION__); 1366 return; 1367 } 1368 1369 sreq->sr_cmnd[0] = ALLOW_MEDIUM_REMOVAL; 1370 sreq->sr_cmnd[1] = 0; 1371 sreq->sr_cmnd[2] = 0; 1372 sreq->sr_cmnd[3] = 0; 1373 sreq->sr_cmnd[4] = SCSI_REMOVAL_PREVENT; 1374 sreq->sr_cmnd[5] = 0; 1375 sreq->sr_data_direction = DMA_NONE; 1376 sreq->sr_bufflen = 0; 1377 sreq->sr_buffer = NULL; 1378 sreq->sr_allowed = 5; 1379 sreq->sr_done = scsi_eh_lock_done; 1380 sreq->sr_timeout_per_command = 10 * HZ; 1381 sreq->sr_cmd_len = COMMAND_SIZE(sreq->sr_cmnd[0]); 1382 1383 scsi_insert_special_req(sreq, 1); 1384 } 1385 1386 1387 /** 1388 * scsi_restart_operations - restart io operations to the specified host. 1389 * @shost: Host we are restarting. 1390 * 1391 * Notes: 1392 * When we entered the error handler, we blocked all further i/o to 1393 * this device. we need to 'reverse' this process. 1394 **/ 1395 static void scsi_restart_operations(struct Scsi_Host *shost) 1396 { 1397 struct scsi_device *sdev; 1398 unsigned long flags; 1399 1400 /* 1401 * If the door was locked, we need to insert a door lock request 1402 * onto the head of the SCSI request queue for the device. There 1403 * is no point trying to lock the door of an off-line device. 1404 */ 1405 shost_for_each_device(sdev, shost) { 1406 if (scsi_device_online(sdev) && sdev->locked) 1407 scsi_eh_lock_door(sdev); 1408 } 1409 1410 /* 1411 * next free up anything directly waiting upon the host. this 1412 * will be requests for character device operations, and also for 1413 * ioctls to queued block devices. 1414 */ 1415 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n", 1416 __FUNCTION__)); 1417 1418 spin_lock_irqsave(shost->host_lock, flags); 1419 if (scsi_host_set_state(shost, SHOST_RUNNING)) 1420 if (scsi_host_set_state(shost, SHOST_CANCEL)) 1421 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 1422 spin_unlock_irqrestore(shost->host_lock, flags); 1423 1424 wake_up(&shost->host_wait); 1425 1426 /* 1427 * finally we need to re-initiate requests that may be pending. we will 1428 * have had everything blocked while error handling is taking place, and 1429 * now that error recovery is done, we will need to ensure that these 1430 * requests are started. 1431 */ 1432 scsi_run_host_queues(shost); 1433 } 1434 1435 /** 1436 * scsi_eh_ready_devs - check device ready state and recover if not. 1437 * @shost: host to be recovered. 1438 * @eh_done_q: list_head for processed commands. 1439 * 1440 **/ 1441 static void scsi_eh_ready_devs(struct Scsi_Host *shost, 1442 struct list_head *work_q, 1443 struct list_head *done_q) 1444 { 1445 if (!scsi_eh_stu(shost, work_q, done_q)) 1446 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 1447 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 1448 if (!scsi_eh_host_reset(work_q, done_q)) 1449 scsi_eh_offline_sdevs(work_q, done_q); 1450 } 1451 1452 /** 1453 * scsi_eh_flush_done_q - finish processed commands or retry them. 1454 * @done_q: list_head of processed commands. 1455 * 1456 **/ 1457 static void scsi_eh_flush_done_q(struct list_head *done_q) 1458 { 1459 struct scsi_cmnd *scmd, *next; 1460 1461 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 1462 list_del_init(&scmd->eh_entry); 1463 if (scsi_device_online(scmd->device) && 1464 !blk_noretry_request(scmd->request) && 1465 (++scmd->retries < scmd->allowed)) { 1466 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush" 1467 " retry cmd: %p\n", 1468 current->comm, 1469 scmd)); 1470 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 1471 } else { 1472 /* 1473 * If just we got sense for the device (called 1474 * scsi_eh_get_sense), scmd->result is already 1475 * set, do not set DRIVER_TIMEOUT. 1476 */ 1477 if (!scmd->result) 1478 scmd->result |= (DRIVER_TIMEOUT << 24); 1479 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish" 1480 " cmd: %p\n", 1481 current->comm, scmd)); 1482 scsi_finish_command(scmd); 1483 } 1484 } 1485 } 1486 1487 /** 1488 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 1489 * @shost: Host to unjam. 1490 * 1491 * Notes: 1492 * When we come in here, we *know* that all commands on the bus have 1493 * either completed, failed or timed out. we also know that no further 1494 * commands are being sent to the host, so things are relatively quiet 1495 * and we have freedom to fiddle with things as we wish. 1496 * 1497 * This is only the *default* implementation. it is possible for 1498 * individual drivers to supply their own version of this function, and 1499 * if the maintainer wishes to do this, it is strongly suggested that 1500 * this function be taken as a template and modified. this function 1501 * was designed to correctly handle problems for about 95% of the 1502 * different cases out there, and it should always provide at least a 1503 * reasonable amount of error recovery. 1504 * 1505 * Any command marked 'failed' or 'timeout' must eventually have 1506 * scsi_finish_cmd() called for it. we do all of the retry stuff 1507 * here, so when we restart the host after we return it should have an 1508 * empty queue. 1509 **/ 1510 static void scsi_unjam_host(struct Scsi_Host *shost) 1511 { 1512 unsigned long flags; 1513 LIST_HEAD(eh_work_q); 1514 LIST_HEAD(eh_done_q); 1515 1516 spin_lock_irqsave(shost->host_lock, flags); 1517 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 1518 spin_unlock_irqrestore(shost->host_lock, flags); 1519 1520 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 1521 1522 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 1523 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q)) 1524 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 1525 1526 scsi_eh_flush_done_q(&eh_done_q); 1527 } 1528 1529 /** 1530 * scsi_error_handler - SCSI error handler thread 1531 * @data: Host for which we are running. 1532 * 1533 * Notes: 1534 * This is the main error handling loop. This is run as a kernel thread 1535 * for every SCSI host and handles all error handling activity. 1536 **/ 1537 int scsi_error_handler(void *data) 1538 { 1539 struct Scsi_Host *shost = data; 1540 1541 current->flags |= PF_NOFREEZE; 1542 1543 /* 1544 * We use TASK_INTERRUPTIBLE so that the thread is not 1545 * counted against the load average as a running process. 1546 * We never actually get interrupted because kthread_run 1547 * disables singal delivery for the created thread. 1548 */ 1549 set_current_state(TASK_INTERRUPTIBLE); 1550 while (!kthread_should_stop()) { 1551 if (shost->host_failed == 0 || 1552 shost->host_failed != shost->host_busy) { 1553 SCSI_LOG_ERROR_RECOVERY(1, 1554 printk("Error handler scsi_eh_%d sleeping\n", 1555 shost->host_no)); 1556 schedule(); 1557 set_current_state(TASK_INTERRUPTIBLE); 1558 continue; 1559 } 1560 1561 __set_current_state(TASK_RUNNING); 1562 SCSI_LOG_ERROR_RECOVERY(1, 1563 printk("Error handler scsi_eh_%d waking up\n", 1564 shost->host_no)); 1565 1566 /* 1567 * We have a host that is failing for some reason. Figure out 1568 * what we need to do to get it up and online again (if we can). 1569 * If we fail, we end up taking the thing offline. 1570 */ 1571 if (shost->hostt->eh_strategy_handler) 1572 shost->hostt->eh_strategy_handler(shost); 1573 else 1574 scsi_unjam_host(shost); 1575 1576 /* 1577 * Note - if the above fails completely, the action is to take 1578 * individual devices offline and flush the queue of any 1579 * outstanding requests that may have been pending. When we 1580 * restart, we restart any I/O to any other devices on the bus 1581 * which are still online. 1582 */ 1583 scsi_restart_operations(shost); 1584 set_current_state(TASK_INTERRUPTIBLE); 1585 } 1586 __set_current_state(TASK_RUNNING); 1587 1588 SCSI_LOG_ERROR_RECOVERY(1, 1589 printk("Error handler scsi_eh_%d exiting\n", shost->host_no)); 1590 shost->ehandler = NULL; 1591 return 0; 1592 } 1593 1594 /* 1595 * Function: scsi_report_bus_reset() 1596 * 1597 * Purpose: Utility function used by low-level drivers to report that 1598 * they have observed a bus reset on the bus being handled. 1599 * 1600 * Arguments: shost - Host in question 1601 * channel - channel on which reset was observed. 1602 * 1603 * Returns: Nothing 1604 * 1605 * Lock status: Host lock must be held. 1606 * 1607 * Notes: This only needs to be called if the reset is one which 1608 * originates from an unknown location. Resets originated 1609 * by the mid-level itself don't need to call this, but there 1610 * should be no harm. 1611 * 1612 * The main purpose of this is to make sure that a CHECK_CONDITION 1613 * is properly treated. 1614 */ 1615 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 1616 { 1617 struct scsi_device *sdev; 1618 1619 __shost_for_each_device(sdev, shost) { 1620 if (channel == sdev_channel(sdev)) { 1621 sdev->was_reset = 1; 1622 sdev->expecting_cc_ua = 1; 1623 } 1624 } 1625 } 1626 EXPORT_SYMBOL(scsi_report_bus_reset); 1627 1628 /* 1629 * Function: scsi_report_device_reset() 1630 * 1631 * Purpose: Utility function used by low-level drivers to report that 1632 * they have observed a device reset on the device being handled. 1633 * 1634 * Arguments: shost - Host in question 1635 * channel - channel on which reset was observed 1636 * target - target on which reset was observed 1637 * 1638 * Returns: Nothing 1639 * 1640 * Lock status: Host lock must be held 1641 * 1642 * Notes: This only needs to be called if the reset is one which 1643 * originates from an unknown location. Resets originated 1644 * by the mid-level itself don't need to call this, but there 1645 * should be no harm. 1646 * 1647 * The main purpose of this is to make sure that a CHECK_CONDITION 1648 * is properly treated. 1649 */ 1650 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 1651 { 1652 struct scsi_device *sdev; 1653 1654 __shost_for_each_device(sdev, shost) { 1655 if (channel == sdev_channel(sdev) && 1656 target == sdev_id(sdev)) { 1657 sdev->was_reset = 1; 1658 sdev->expecting_cc_ua = 1; 1659 } 1660 } 1661 } 1662 EXPORT_SYMBOL(scsi_report_device_reset); 1663 1664 static void 1665 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 1666 { 1667 } 1668 1669 /* 1670 * Function: scsi_reset_provider 1671 * 1672 * Purpose: Send requested reset to a bus or device at any phase. 1673 * 1674 * Arguments: device - device to send reset to 1675 * flag - reset type (see scsi.h) 1676 * 1677 * Returns: SUCCESS/FAILURE. 1678 * 1679 * Notes: This is used by the SCSI Generic driver to provide 1680 * Bus/Device reset capability. 1681 */ 1682 int 1683 scsi_reset_provider(struct scsi_device *dev, int flag) 1684 { 1685 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL); 1686 struct request req; 1687 int rtn; 1688 1689 scmd->request = &req; 1690 memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout)); 1691 scmd->request->rq_status = RQ_SCSI_BUSY; 1692 1693 memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd)); 1694 1695 scmd->scsi_done = scsi_reset_provider_done_command; 1696 scmd->done = NULL; 1697 scmd->buffer = NULL; 1698 scmd->bufflen = 0; 1699 scmd->request_buffer = NULL; 1700 scmd->request_bufflen = 0; 1701 1702 scmd->cmd_len = 0; 1703 1704 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 1705 scmd->sc_request = NULL; 1706 scmd->sc_magic = SCSI_CMND_MAGIC; 1707 1708 init_timer(&scmd->eh_timeout); 1709 1710 /* 1711 * Sometimes the command can get back into the timer chain, 1712 * so use the pid as an identifier. 1713 */ 1714 scmd->pid = 0; 1715 1716 switch (flag) { 1717 case SCSI_TRY_RESET_DEVICE: 1718 rtn = scsi_try_bus_device_reset(scmd); 1719 if (rtn == SUCCESS) 1720 break; 1721 /* FALLTHROUGH */ 1722 case SCSI_TRY_RESET_BUS: 1723 rtn = scsi_try_bus_reset(scmd); 1724 if (rtn == SUCCESS) 1725 break; 1726 /* FALLTHROUGH */ 1727 case SCSI_TRY_RESET_HOST: 1728 rtn = scsi_try_host_reset(scmd); 1729 break; 1730 default: 1731 rtn = FAILED; 1732 } 1733 1734 scsi_next_command(scmd); 1735 return rtn; 1736 } 1737 EXPORT_SYMBOL(scsi_reset_provider); 1738 1739 /** 1740 * scsi_normalize_sense - normalize main elements from either fixed or 1741 * descriptor sense data format into a common format. 1742 * 1743 * @sense_buffer: byte array containing sense data returned by device 1744 * @sb_len: number of valid bytes in sense_buffer 1745 * @sshdr: pointer to instance of structure that common 1746 * elements are written to. 1747 * 1748 * Notes: 1749 * The "main elements" from sense data are: response_code, sense_key, 1750 * asc, ascq and additional_length (only for descriptor format). 1751 * 1752 * Typically this function can be called after a device has 1753 * responded to a SCSI command with the CHECK_CONDITION status. 1754 * 1755 * Return value: 1756 * 1 if valid sense data information found, else 0; 1757 **/ 1758 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len, 1759 struct scsi_sense_hdr *sshdr) 1760 { 1761 if (!sense_buffer || !sb_len) 1762 return 0; 1763 1764 memset(sshdr, 0, sizeof(struct scsi_sense_hdr)); 1765 1766 sshdr->response_code = (sense_buffer[0] & 0x7f); 1767 1768 if (!scsi_sense_valid(sshdr)) 1769 return 0; 1770 1771 if (sshdr->response_code >= 0x72) { 1772 /* 1773 * descriptor format 1774 */ 1775 if (sb_len > 1) 1776 sshdr->sense_key = (sense_buffer[1] & 0xf); 1777 if (sb_len > 2) 1778 sshdr->asc = sense_buffer[2]; 1779 if (sb_len > 3) 1780 sshdr->ascq = sense_buffer[3]; 1781 if (sb_len > 7) 1782 sshdr->additional_length = sense_buffer[7]; 1783 } else { 1784 /* 1785 * fixed format 1786 */ 1787 if (sb_len > 2) 1788 sshdr->sense_key = (sense_buffer[2] & 0xf); 1789 if (sb_len > 7) { 1790 sb_len = (sb_len < (sense_buffer[7] + 8)) ? 1791 sb_len : (sense_buffer[7] + 8); 1792 if (sb_len > 12) 1793 sshdr->asc = sense_buffer[12]; 1794 if (sb_len > 13) 1795 sshdr->ascq = sense_buffer[13]; 1796 } 1797 } 1798 1799 return 1; 1800 } 1801 EXPORT_SYMBOL(scsi_normalize_sense); 1802 1803 int scsi_request_normalize_sense(struct scsi_request *sreq, 1804 struct scsi_sense_hdr *sshdr) 1805 { 1806 return scsi_normalize_sense(sreq->sr_sense_buffer, 1807 sizeof(sreq->sr_sense_buffer), sshdr); 1808 } 1809 EXPORT_SYMBOL(scsi_request_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