1 /* 2 * scsi_error.c Copyright (C) 1997 Eric Youngdale 3 * 4 * SCSI error/timeout handling 5 * Initial versions: Eric Youngdale. Based upon conversations with 6 * Leonard Zubkoff and David Miller at Linux Expo, 7 * ideas originating from all over the place. 8 * 9 * Restructured scsi_unjam_host and associated functions. 10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com) 11 * 12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and 13 * minor cleanups. 14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com) 15 */ 16 17 #include <linux/module.h> 18 #include <linux/sched.h> 19 #include <linux/timer.h> 20 #include <linux/string.h> 21 #include <linux/slab.h> 22 #include <linux/kernel.h> 23 #include <linux/kthread.h> 24 #include <linux/interrupt.h> 25 #include <linux/blkdev.h> 26 #include <linux/delay.h> 27 28 #include <scsi/scsi.h> 29 #include <scsi/scsi_cmnd.h> 30 #include <scsi/scsi_dbg.h> 31 #include <scsi/scsi_device.h> 32 #include <scsi/scsi_eh.h> 33 #include <scsi/scsi_transport.h> 34 #include <scsi/scsi_host.h> 35 #include <scsi/scsi_ioctl.h> 36 37 #include "scsi_priv.h" 38 #include "scsi_logging.h" 39 40 #define SENSE_TIMEOUT (10*HZ) 41 #define START_UNIT_TIMEOUT (30*HZ) 42 43 /* 44 * These should *probably* be handled by the host itself. 45 * Since it is allowed to sleep, it probably should. 46 */ 47 #define BUS_RESET_SETTLE_TIME (10) 48 #define HOST_RESET_SETTLE_TIME (10) 49 50 /* called with shost->host_lock held */ 51 void scsi_eh_wakeup(struct Scsi_Host *shost) 52 { 53 if (shost->host_busy == shost->host_failed) { 54 wake_up_process(shost->ehandler); 55 SCSI_LOG_ERROR_RECOVERY(5, 56 printk("Waking error handler thread\n")); 57 } 58 } 59 60 /** 61 * scsi_eh_scmd_add - add scsi cmd to error handling. 62 * @scmd: scmd to run eh on. 63 * @eh_flag: optional SCSI_EH flag. 64 * 65 * Return value: 66 * 0 on failure. 67 **/ 68 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag) 69 { 70 struct Scsi_Host *shost = scmd->device->host; 71 unsigned long flags; 72 int ret = 0; 73 74 if (!shost->ehandler) 75 return 0; 76 77 spin_lock_irqsave(shost->host_lock, flags); 78 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 79 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)) 80 goto out_unlock; 81 82 ret = 1; 83 scmd->eh_eflags |= eh_flag; 84 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 85 shost->host_failed++; 86 scsi_eh_wakeup(shost); 87 out_unlock: 88 spin_unlock_irqrestore(shost->host_lock, flags); 89 return ret; 90 } 91 92 /** 93 * scsi_add_timer - Start timeout timer for a single scsi command. 94 * @scmd: scsi command that is about to start running. 95 * @timeout: amount of time to allow this command to run. 96 * @complete: timeout function to call if timer isn't canceled. 97 * 98 * Notes: 99 * This should be turned into an inline function. Each scsi command 100 * has its own timer, and as it is added to the queue, we set up the 101 * timer. When the command completes, we cancel the timer. 102 **/ 103 void scsi_add_timer(struct scsi_cmnd *scmd, int timeout, 104 void (*complete)(struct scsi_cmnd *)) 105 { 106 107 /* 108 * If the clock was already running for this command, then 109 * first delete the timer. The timer handling code gets rather 110 * confused if we don't do this. 111 */ 112 if (scmd->eh_timeout.function) 113 del_timer(&scmd->eh_timeout); 114 115 scmd->eh_timeout.data = (unsigned long)scmd; 116 scmd->eh_timeout.expires = jiffies + timeout; 117 scmd->eh_timeout.function = (void (*)(unsigned long)) complete; 118 119 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:" 120 " %d, (%p)\n", __FUNCTION__, 121 scmd, timeout, complete)); 122 123 add_timer(&scmd->eh_timeout); 124 } 125 126 /** 127 * scsi_delete_timer - Delete/cancel timer for a given function. 128 * @scmd: Cmd that we are canceling timer for 129 * 130 * Notes: 131 * This should be turned into an inline function. 132 * 133 * Return value: 134 * 1 if we were able to detach the timer. 0 if we blew it, and the 135 * timer function has already started to run. 136 **/ 137 int scsi_delete_timer(struct scsi_cmnd *scmd) 138 { 139 int rtn; 140 141 rtn = del_timer(&scmd->eh_timeout); 142 143 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p," 144 " rtn: %d\n", __FUNCTION__, 145 scmd, rtn)); 146 147 scmd->eh_timeout.data = (unsigned long)NULL; 148 scmd->eh_timeout.function = NULL; 149 150 return rtn; 151 } 152 153 /** 154 * scsi_times_out - Timeout function for normal scsi commands. 155 * @scmd: Cmd that is timing out. 156 * 157 * Notes: 158 * We do not need to lock this. There is the potential for a race 159 * only in that the normal completion handling might run, but if the 160 * normal completion function determines that the timer has already 161 * fired, then it mustn't do anything. 162 **/ 163 void scsi_times_out(struct scsi_cmnd *scmd) 164 { 165 scsi_log_completion(scmd, TIMEOUT_ERROR); 166 167 if (scmd->device->host->transportt->eh_timed_out) 168 switch (scmd->device->host->transportt->eh_timed_out(scmd)) { 169 case EH_HANDLED: 170 __scsi_done(scmd); 171 return; 172 case EH_RESET_TIMER: 173 scsi_add_timer(scmd, scmd->timeout_per_command, 174 scsi_times_out); 175 return; 176 case EH_NOT_HANDLED: 177 break; 178 } 179 180 if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) { 181 scmd->result |= DID_TIME_OUT << 16; 182 __scsi_done(scmd); 183 } 184 } 185 186 /** 187 * scsi_block_when_processing_errors - Prevent cmds from being queued. 188 * @sdev: Device on which we are performing recovery. 189 * 190 * Description: 191 * We block until the host is out of error recovery, and then check to 192 * see whether the host or the device is offline. 193 * 194 * Return value: 195 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 196 **/ 197 int scsi_block_when_processing_errors(struct scsi_device *sdev) 198 { 199 int online; 200 201 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 202 203 online = scsi_device_online(sdev); 204 205 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__, 206 online)); 207 208 return online; 209 } 210 EXPORT_SYMBOL(scsi_block_when_processing_errors); 211 212 #ifdef CONFIG_SCSI_LOGGING 213 /** 214 * scsi_eh_prt_fail_stats - Log info on failures. 215 * @shost: scsi host being recovered. 216 * @work_q: Queue of scsi cmds to process. 217 **/ 218 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 219 struct list_head *work_q) 220 { 221 struct scsi_cmnd *scmd; 222 struct scsi_device *sdev; 223 int total_failures = 0; 224 int cmd_failed = 0; 225 int cmd_cancel = 0; 226 int devices_failed = 0; 227 228 shost_for_each_device(sdev, shost) { 229 list_for_each_entry(scmd, work_q, eh_entry) { 230 if (scmd->device == sdev) { 231 ++total_failures; 232 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) 233 ++cmd_cancel; 234 else 235 ++cmd_failed; 236 } 237 } 238 239 if (cmd_cancel || cmd_failed) { 240 SCSI_LOG_ERROR_RECOVERY(3, 241 sdev_printk(KERN_INFO, sdev, 242 "%s: cmds failed: %d, cancel: %d\n", 243 __FUNCTION__, cmd_failed, 244 cmd_cancel)); 245 cmd_cancel = 0; 246 cmd_failed = 0; 247 ++devices_failed; 248 } 249 } 250 251 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d" 252 " devices require eh work\n", 253 total_failures, devices_failed)); 254 } 255 #endif 256 257 /** 258 * scsi_check_sense - Examine scsi cmd sense 259 * @scmd: Cmd to have sense checked. 260 * 261 * Return value: 262 * SUCCESS or FAILED or NEEDS_RETRY 263 * 264 * Notes: 265 * When a deferred error is detected the current command has 266 * not been executed and needs retrying. 267 **/ 268 static int scsi_check_sense(struct scsi_cmnd *scmd) 269 { 270 struct scsi_sense_hdr sshdr; 271 272 if (! scsi_command_normalize_sense(scmd, &sshdr)) 273 return FAILED; /* no valid sense data */ 274 275 if (scsi_sense_is_deferred(&sshdr)) 276 return NEEDS_RETRY; 277 278 /* 279 * Previous logic looked for FILEMARK, EOM or ILI which are 280 * mainly associated with tapes and returned SUCCESS. 281 */ 282 if (sshdr.response_code == 0x70) { 283 /* fixed format */ 284 if (scmd->sense_buffer[2] & 0xe0) 285 return SUCCESS; 286 } else { 287 /* 288 * descriptor format: look for "stream commands sense data 289 * descriptor" (see SSC-3). Assume single sense data 290 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 291 */ 292 if ((sshdr.additional_length > 3) && 293 (scmd->sense_buffer[8] == 0x4) && 294 (scmd->sense_buffer[11] & 0xe0)) 295 return SUCCESS; 296 } 297 298 switch (sshdr.sense_key) { 299 case NO_SENSE: 300 return SUCCESS; 301 case RECOVERED_ERROR: 302 return /* soft_error */ SUCCESS; 303 304 case ABORTED_COMMAND: 305 return NEEDS_RETRY; 306 case NOT_READY: 307 case UNIT_ATTENTION: 308 /* 309 * if we are expecting a cc/ua because of a bus reset that we 310 * performed, treat this just as a retry. otherwise this is 311 * information that we should pass up to the upper-level driver 312 * so that we can deal with it there. 313 */ 314 if (scmd->device->expecting_cc_ua) { 315 scmd->device->expecting_cc_ua = 0; 316 return NEEDS_RETRY; 317 } 318 /* 319 * if the device is in the process of becoming ready, we 320 * should retry. 321 */ 322 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 323 return NEEDS_RETRY; 324 /* 325 * if the device is not started, we need to wake 326 * the error handler to start the motor 327 */ 328 if (scmd->device->allow_restart && 329 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 330 return FAILED; 331 return SUCCESS; 332 333 /* these three are not supported */ 334 case COPY_ABORTED: 335 case VOLUME_OVERFLOW: 336 case MISCOMPARE: 337 return SUCCESS; 338 339 case MEDIUM_ERROR: 340 return NEEDS_RETRY; 341 342 case HARDWARE_ERROR: 343 if (scmd->device->retry_hwerror) 344 return NEEDS_RETRY; 345 else 346 return SUCCESS; 347 348 case ILLEGAL_REQUEST: 349 case BLANK_CHECK: 350 case DATA_PROTECT: 351 default: 352 return SUCCESS; 353 } 354 } 355 356 /** 357 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 358 * @scmd: SCSI cmd to examine. 359 * 360 * Notes: 361 * This is *only* called when we are examining the status of commands 362 * queued during error recovery. the main difference here is that we 363 * don't allow for the possibility of retries here, and we are a lot 364 * more restrictive about what we consider acceptable. 365 **/ 366 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) 367 { 368 /* 369 * first check the host byte, to see if there is anything in there 370 * that would indicate what we need to do. 371 */ 372 if (host_byte(scmd->result) == DID_RESET) { 373 /* 374 * rats. we are already in the error handler, so we now 375 * get to try and figure out what to do next. if the sense 376 * is valid, we have a pretty good idea of what to do. 377 * if not, we mark it as FAILED. 378 */ 379 return scsi_check_sense(scmd); 380 } 381 if (host_byte(scmd->result) != DID_OK) 382 return FAILED; 383 384 /* 385 * next, check the message byte. 386 */ 387 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 388 return FAILED; 389 390 /* 391 * now, check the status byte to see if this indicates 392 * anything special. 393 */ 394 switch (status_byte(scmd->result)) { 395 case GOOD: 396 case COMMAND_TERMINATED: 397 return SUCCESS; 398 case CHECK_CONDITION: 399 return scsi_check_sense(scmd); 400 case CONDITION_GOOD: 401 case INTERMEDIATE_GOOD: 402 case INTERMEDIATE_C_GOOD: 403 /* 404 * who knows? FIXME(eric) 405 */ 406 return SUCCESS; 407 case BUSY: 408 case QUEUE_FULL: 409 case RESERVATION_CONFLICT: 410 default: 411 return FAILED; 412 } 413 return FAILED; 414 } 415 416 /** 417 * scsi_eh_done - Completion function for error handling. 418 * @scmd: Cmd that is done. 419 **/ 420 static void scsi_eh_done(struct scsi_cmnd *scmd) 421 { 422 struct completion *eh_action; 423 424 SCSI_LOG_ERROR_RECOVERY(3, 425 printk("%s scmd: %p result: %x\n", 426 __FUNCTION__, scmd, scmd->result)); 427 428 eh_action = scmd->device->host->eh_action; 429 if (eh_action) 430 complete(eh_action); 431 } 432 433 /** 434 * scsi_send_eh_cmnd - send a cmd to a device as part of error recovery. 435 * @scmd: SCSI Cmd to send. 436 * @timeout: Timeout for cmd. 437 * 438 * Return value: 439 * SUCCESS or FAILED or NEEDS_RETRY 440 **/ 441 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, int timeout) 442 { 443 struct scsi_device *sdev = scmd->device; 444 struct Scsi_Host *shost = sdev->host; 445 DECLARE_COMPLETION(done); 446 unsigned long timeleft; 447 unsigned long flags; 448 int rtn; 449 450 if (sdev->scsi_level <= SCSI_2) 451 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 452 (sdev->lun << 5 & 0xe0); 453 454 shost->eh_action = &done; 455 456 spin_lock_irqsave(shost->host_lock, flags); 457 scsi_log_send(scmd); 458 shost->hostt->queuecommand(scmd, scsi_eh_done); 459 spin_unlock_irqrestore(shost->host_lock, flags); 460 461 timeleft = wait_for_completion_timeout(&done, timeout); 462 463 shost->eh_action = NULL; 464 465 scsi_log_completion(scmd, SUCCESS); 466 467 SCSI_LOG_ERROR_RECOVERY(3, 468 printk("%s: scmd: %p, timeleft: %ld\n", 469 __FUNCTION__, scmd, timeleft)); 470 471 /* 472 * If there is time left scsi_eh_done got called, and we will 473 * examine the actual status codes to see whether the command 474 * actually did complete normally, else tell the host to forget 475 * about this command. 476 */ 477 if (timeleft) { 478 rtn = scsi_eh_completed_normally(scmd); 479 SCSI_LOG_ERROR_RECOVERY(3, 480 printk("%s: scsi_eh_completed_normally %x\n", 481 __FUNCTION__, rtn)); 482 483 switch (rtn) { 484 case SUCCESS: 485 case NEEDS_RETRY: 486 case FAILED: 487 break; 488 default: 489 rtn = FAILED; 490 break; 491 } 492 } else { 493 /* 494 * FIXME(eric) - we are not tracking whether we could 495 * abort a timed out command or not. not sure how 496 * we should treat them differently anyways. 497 */ 498 if (shost->hostt->eh_abort_handler) 499 shost->hostt->eh_abort_handler(scmd); 500 rtn = FAILED; 501 } 502 503 return rtn; 504 } 505 506 /** 507 * scsi_request_sense - Request sense data from a particular target. 508 * @scmd: SCSI cmd for request sense. 509 * 510 * Notes: 511 * Some hosts automatically obtain this information, others require 512 * that we obtain it on our own. This function will *not* return until 513 * the command either times out, or it completes. 514 **/ 515 static int scsi_request_sense(struct scsi_cmnd *scmd) 516 { 517 static unsigned char generic_sense[6] = 518 {REQUEST_SENSE, 0, 0, 0, 252, 0}; 519 unsigned char *scsi_result; 520 int saved_result; 521 int rtn; 522 523 memcpy(scmd->cmnd, generic_sense, sizeof(generic_sense)); 524 525 scsi_result = kmalloc(252, GFP_ATOMIC | ((scmd->device->host->hostt->unchecked_isa_dma) ? __GFP_DMA : 0)); 526 527 528 if (unlikely(!scsi_result)) { 529 printk(KERN_ERR "%s: cannot allocate scsi_result.\n", 530 __FUNCTION__); 531 return FAILED; 532 } 533 534 /* 535 * zero the sense buffer. some host adapters automatically always 536 * request sense, so it is not a good idea that 537 * scmd->request_buffer and scmd->sense_buffer point to the same 538 * address (db). 0 is not a valid sense code. 539 */ 540 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); 541 memset(scsi_result, 0, 252); 542 543 saved_result = scmd->result; 544 scmd->request_buffer = scsi_result; 545 scmd->request_bufflen = 252; 546 scmd->use_sg = 0; 547 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 548 scmd->sc_data_direction = DMA_FROM_DEVICE; 549 scmd->underflow = 0; 550 551 rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT); 552 553 /* last chance to have valid sense data */ 554 if(!SCSI_SENSE_VALID(scmd)) { 555 memcpy(scmd->sense_buffer, scmd->request_buffer, 556 sizeof(scmd->sense_buffer)); 557 } 558 559 kfree(scsi_result); 560 561 /* 562 * when we eventually call scsi_finish, we really wish to complete 563 * the original request, so let's restore the original data. (db) 564 */ 565 scsi_setup_cmd_retry(scmd); 566 scmd->result = saved_result; 567 return rtn; 568 } 569 570 /** 571 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 572 * @scmd: Original SCSI cmd that eh has finished. 573 * @done_q: Queue for processed commands. 574 * 575 * Notes: 576 * We don't want to use the normal command completion while we are are 577 * still handling errors - it may cause other commands to be queued, 578 * and that would disturb what we are doing. thus we really want to 579 * keep a list of pending commands for final completion, and once we 580 * are ready to leave error handling we handle completion for real. 581 **/ 582 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 583 { 584 scmd->device->host->host_failed--; 585 scmd->eh_eflags = 0; 586 587 /* 588 * set this back so that the upper level can correctly free up 589 * things. 590 */ 591 scsi_setup_cmd_retry(scmd); 592 list_move_tail(&scmd->eh_entry, done_q); 593 } 594 EXPORT_SYMBOL(scsi_eh_finish_cmd); 595 596 /** 597 * scsi_eh_get_sense - Get device sense data. 598 * @work_q: Queue of commands to process. 599 * @done_q: Queue of proccessed commands.. 600 * 601 * Description: 602 * See if we need to request sense information. if so, then get it 603 * now, so we have a better idea of what to do. 604 * 605 * Notes: 606 * This has the unfortunate side effect that if a shost adapter does 607 * not automatically request sense information, that we end up shutting 608 * it down before we request it. 609 * 610 * All drivers should request sense information internally these days, 611 * so for now all I have to say is tough noogies if you end up in here. 612 * 613 * XXX: Long term this code should go away, but that needs an audit of 614 * all LLDDs first. 615 **/ 616 static int scsi_eh_get_sense(struct list_head *work_q, 617 struct list_head *done_q) 618 { 619 struct scsi_cmnd *scmd, *next; 620 int rtn; 621 622 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 623 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) || 624 SCSI_SENSE_VALID(scmd)) 625 continue; 626 627 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 628 "%s: requesting sense\n", 629 current->comm)); 630 rtn = scsi_request_sense(scmd); 631 if (rtn != SUCCESS) 632 continue; 633 634 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p" 635 " result %x\n", scmd, 636 scmd->result)); 637 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd)); 638 639 rtn = scsi_decide_disposition(scmd); 640 641 /* 642 * if the result was normal, then just pass it along to the 643 * upper level. 644 */ 645 if (rtn == SUCCESS) 646 /* we don't want this command reissued, just 647 * finished with the sense data, so set 648 * retries to the max allowed to ensure it 649 * won't get reissued */ 650 scmd->retries = scmd->allowed; 651 else if (rtn != NEEDS_RETRY) 652 continue; 653 654 scsi_eh_finish_cmd(scmd, done_q); 655 } 656 657 return list_empty(work_q); 658 } 659 660 /** 661 * scsi_try_to_abort_cmd - Ask host to abort a running command. 662 * @scmd: SCSI cmd to abort from Lower Level. 663 * 664 * Notes: 665 * This function will not return until the user's completion function 666 * has been called. there is no timeout on this operation. if the 667 * author of the low-level driver wishes this operation to be timed, 668 * they can provide this facility themselves. helper functions in 669 * scsi_error.c can be supplied to make this easier to do. 670 **/ 671 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd) 672 { 673 if (!scmd->device->host->hostt->eh_abort_handler) 674 return FAILED; 675 676 /* 677 * scsi_done was called just after the command timed out and before 678 * we had a chance to process it. (db) 679 */ 680 if (scmd->serial_number == 0) 681 return SUCCESS; 682 return scmd->device->host->hostt->eh_abort_handler(scmd); 683 } 684 685 /** 686 * scsi_eh_tur - Send TUR to device. 687 * @scmd: Scsi cmd to send TUR 688 * 689 * Return value: 690 * 0 - Device is ready. 1 - Device NOT ready. 691 **/ 692 static int scsi_eh_tur(struct scsi_cmnd *scmd) 693 { 694 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 695 int retry_cnt = 1, rtn; 696 int saved_result; 697 698 retry_tur: 699 memcpy(scmd->cmnd, tur_command, sizeof(tur_command)); 700 701 /* 702 * zero the sense buffer. the scsi spec mandates that any 703 * untransferred sense data should be interpreted as being zero. 704 */ 705 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); 706 707 saved_result = scmd->result; 708 scmd->request_buffer = NULL; 709 scmd->request_bufflen = 0; 710 scmd->use_sg = 0; 711 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 712 scmd->underflow = 0; 713 scmd->sc_data_direction = DMA_NONE; 714 715 rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT); 716 717 /* 718 * when we eventually call scsi_finish, we really wish to complete 719 * the original request, so let's restore the original data. (db) 720 */ 721 scsi_setup_cmd_retry(scmd); 722 scmd->result = saved_result; 723 724 /* 725 * hey, we are done. let's look to see what happened. 726 */ 727 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", 728 __FUNCTION__, scmd, rtn)); 729 if (rtn == SUCCESS) 730 return 0; 731 else if (rtn == NEEDS_RETRY) { 732 if (retry_cnt--) 733 goto retry_tur; 734 return 0; 735 } 736 return 1; 737 } 738 739 /** 740 * scsi_eh_abort_cmds - abort canceled commands. 741 * @shost: scsi host being recovered. 742 * @eh_done_q: list_head for processed commands. 743 * 744 * Decription: 745 * Try and see whether or not it makes sense to try and abort the 746 * running command. this only works out to be the case if we have one 747 * command that has timed out. if the command simply failed, it makes 748 * no sense to try and abort the command, since as far as the shost 749 * adapter is concerned, it isn't running. 750 **/ 751 static int scsi_eh_abort_cmds(struct list_head *work_q, 752 struct list_head *done_q) 753 { 754 struct scsi_cmnd *scmd, *next; 755 int rtn; 756 757 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 758 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD)) 759 continue; 760 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:" 761 "0x%p\n", current->comm, 762 scmd)); 763 rtn = scsi_try_to_abort_cmd(scmd); 764 if (rtn == SUCCESS) { 765 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD; 766 if (!scsi_device_online(scmd->device) || 767 !scsi_eh_tur(scmd)) { 768 scsi_eh_finish_cmd(scmd, done_q); 769 } 770 771 } else 772 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting" 773 " cmd failed:" 774 "0x%p\n", 775 current->comm, 776 scmd)); 777 } 778 779 return list_empty(work_q); 780 } 781 782 /** 783 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 784 * @scmd: SCSI cmd used to send BDR 785 * 786 * Notes: 787 * There is no timeout for this operation. if this operation is 788 * unreliable for a given host, then the host itself needs to put a 789 * timer on it, and set the host back to a consistent state prior to 790 * returning. 791 **/ 792 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 793 { 794 int rtn; 795 796 if (!scmd->device->host->hostt->eh_device_reset_handler) 797 return FAILED; 798 799 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd); 800 if (rtn == SUCCESS) { 801 scmd->device->was_reset = 1; 802 scmd->device->expecting_cc_ua = 1; 803 } 804 805 return rtn; 806 } 807 808 /** 809 * scsi_eh_try_stu - Send START_UNIT to device. 810 * @scmd: Scsi cmd to send START_UNIT 811 * 812 * Return value: 813 * 0 - Device is ready. 1 - Device NOT ready. 814 **/ 815 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 816 { 817 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 818 int rtn; 819 int saved_result; 820 821 if (!scmd->device->allow_restart) 822 return 1; 823 824 memcpy(scmd->cmnd, stu_command, sizeof(stu_command)); 825 826 /* 827 * zero the sense buffer. the scsi spec mandates that any 828 * untransferred sense data should be interpreted as being zero. 829 */ 830 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); 831 832 saved_result = scmd->result; 833 scmd->request_buffer = NULL; 834 scmd->request_bufflen = 0; 835 scmd->use_sg = 0; 836 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 837 scmd->underflow = 0; 838 scmd->sc_data_direction = DMA_NONE; 839 840 rtn = scsi_send_eh_cmnd(scmd, START_UNIT_TIMEOUT); 841 842 /* 843 * when we eventually call scsi_finish, we really wish to complete 844 * the original request, so let's restore the original data. (db) 845 */ 846 scsi_setup_cmd_retry(scmd); 847 scmd->result = saved_result; 848 849 /* 850 * hey, we are done. let's look to see what happened. 851 */ 852 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", 853 __FUNCTION__, scmd, rtn)); 854 if (rtn == SUCCESS) 855 return 0; 856 return 1; 857 } 858 859 /** 860 * scsi_eh_stu - send START_UNIT if needed 861 * @shost: scsi host being recovered. 862 * @eh_done_q: list_head for processed commands. 863 * 864 * Notes: 865 * If commands are failing due to not ready, initializing command required, 866 * try revalidating the device, which will end up sending a start unit. 867 **/ 868 static int scsi_eh_stu(struct Scsi_Host *shost, 869 struct list_head *work_q, 870 struct list_head *done_q) 871 { 872 struct scsi_cmnd *scmd, *stu_scmd, *next; 873 struct scsi_device *sdev; 874 875 shost_for_each_device(sdev, shost) { 876 stu_scmd = NULL; 877 list_for_each_entry(scmd, work_q, eh_entry) 878 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 879 scsi_check_sense(scmd) == FAILED ) { 880 stu_scmd = scmd; 881 break; 882 } 883 884 if (!stu_scmd) 885 continue; 886 887 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:" 888 " 0x%p\n", current->comm, sdev)); 889 890 if (!scsi_eh_try_stu(stu_scmd)) { 891 if (!scsi_device_online(sdev) || 892 !scsi_eh_tur(stu_scmd)) { 893 list_for_each_entry_safe(scmd, next, 894 work_q, eh_entry) { 895 if (scmd->device == sdev) 896 scsi_eh_finish_cmd(scmd, done_q); 897 } 898 } 899 } else { 900 SCSI_LOG_ERROR_RECOVERY(3, 901 printk("%s: START_UNIT failed to sdev:" 902 " 0x%p\n", current->comm, sdev)); 903 } 904 } 905 906 return list_empty(work_q); 907 } 908 909 910 /** 911 * scsi_eh_bus_device_reset - send bdr if needed 912 * @shost: scsi host being recovered. 913 * @eh_done_q: list_head for processed commands. 914 * 915 * Notes: 916 * Try a bus device reset. still, look to see whether we have multiple 917 * devices that are jammed or not - if we have multiple devices, it 918 * makes no sense to try bus_device_reset - we really would need to try 919 * a bus_reset instead. 920 **/ 921 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 922 struct list_head *work_q, 923 struct list_head *done_q) 924 { 925 struct scsi_cmnd *scmd, *bdr_scmd, *next; 926 struct scsi_device *sdev; 927 int rtn; 928 929 shost_for_each_device(sdev, shost) { 930 bdr_scmd = NULL; 931 list_for_each_entry(scmd, work_q, eh_entry) 932 if (scmd->device == sdev) { 933 bdr_scmd = scmd; 934 break; 935 } 936 937 if (!bdr_scmd) 938 continue; 939 940 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:" 941 " 0x%p\n", current->comm, 942 sdev)); 943 rtn = scsi_try_bus_device_reset(bdr_scmd); 944 if (rtn == SUCCESS) { 945 if (!scsi_device_online(sdev) || 946 !scsi_eh_tur(bdr_scmd)) { 947 list_for_each_entry_safe(scmd, next, 948 work_q, eh_entry) { 949 if (scmd->device == sdev) 950 scsi_eh_finish_cmd(scmd, 951 done_q); 952 } 953 } 954 } else { 955 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR" 956 " failed sdev:" 957 "0x%p\n", 958 current->comm, 959 sdev)); 960 } 961 } 962 963 return list_empty(work_q); 964 } 965 966 /** 967 * scsi_try_bus_reset - ask host to perform a bus reset 968 * @scmd: SCSI cmd to send bus reset. 969 **/ 970 static int scsi_try_bus_reset(struct scsi_cmnd *scmd) 971 { 972 unsigned long flags; 973 int rtn; 974 975 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n", 976 __FUNCTION__)); 977 978 if (!scmd->device->host->hostt->eh_bus_reset_handler) 979 return FAILED; 980 981 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd); 982 983 if (rtn == SUCCESS) { 984 if (!scmd->device->host->hostt->skip_settle_delay) 985 ssleep(BUS_RESET_SETTLE_TIME); 986 spin_lock_irqsave(scmd->device->host->host_lock, flags); 987 scsi_report_bus_reset(scmd->device->host, 988 scmd_channel(scmd)); 989 spin_unlock_irqrestore(scmd->device->host->host_lock, flags); 990 } 991 992 return rtn; 993 } 994 995 /** 996 * scsi_try_host_reset - ask host adapter to reset itself 997 * @scmd: SCSI cmd to send hsot reset. 998 **/ 999 static int scsi_try_host_reset(struct scsi_cmnd *scmd) 1000 { 1001 unsigned long flags; 1002 int rtn; 1003 1004 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n", 1005 __FUNCTION__)); 1006 1007 if (!scmd->device->host->hostt->eh_host_reset_handler) 1008 return FAILED; 1009 1010 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd); 1011 1012 if (rtn == SUCCESS) { 1013 if (!scmd->device->host->hostt->skip_settle_delay) 1014 ssleep(HOST_RESET_SETTLE_TIME); 1015 spin_lock_irqsave(scmd->device->host->host_lock, flags); 1016 scsi_report_bus_reset(scmd->device->host, 1017 scmd_channel(scmd)); 1018 spin_unlock_irqrestore(scmd->device->host->host_lock, flags); 1019 } 1020 1021 return rtn; 1022 } 1023 1024 /** 1025 * scsi_eh_bus_reset - send a bus reset 1026 * @shost: scsi host being recovered. 1027 * @eh_done_q: list_head for processed commands. 1028 **/ 1029 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1030 struct list_head *work_q, 1031 struct list_head *done_q) 1032 { 1033 struct scsi_cmnd *scmd, *chan_scmd, *next; 1034 unsigned int channel; 1035 int rtn; 1036 1037 /* 1038 * we really want to loop over the various channels, and do this on 1039 * a channel by channel basis. we should also check to see if any 1040 * of the failed commands are on soft_reset devices, and if so, skip 1041 * the reset. 1042 */ 1043 1044 for (channel = 0; channel <= shost->max_channel; channel++) { 1045 chan_scmd = NULL; 1046 list_for_each_entry(scmd, work_q, eh_entry) { 1047 if (channel == scmd_channel(scmd)) { 1048 chan_scmd = scmd; 1049 break; 1050 /* 1051 * FIXME add back in some support for 1052 * soft_reset devices. 1053 */ 1054 } 1055 } 1056 1057 if (!chan_scmd) 1058 continue; 1059 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:" 1060 " %d\n", current->comm, 1061 channel)); 1062 rtn = scsi_try_bus_reset(chan_scmd); 1063 if (rtn == SUCCESS) { 1064 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1065 if (channel == scmd_channel(scmd)) 1066 if (!scsi_device_online(scmd->device) || 1067 !scsi_eh_tur(scmd)) 1068 scsi_eh_finish_cmd(scmd, 1069 done_q); 1070 } 1071 } else { 1072 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST" 1073 " failed chan: %d\n", 1074 current->comm, 1075 channel)); 1076 } 1077 } 1078 return list_empty(work_q); 1079 } 1080 1081 /** 1082 * scsi_eh_host_reset - send a host reset 1083 * @work_q: list_head for processed commands. 1084 * @done_q: list_head for processed commands. 1085 **/ 1086 static int scsi_eh_host_reset(struct list_head *work_q, 1087 struct list_head *done_q) 1088 { 1089 struct scsi_cmnd *scmd, *next; 1090 int rtn; 1091 1092 if (!list_empty(work_q)) { 1093 scmd = list_entry(work_q->next, 1094 struct scsi_cmnd, eh_entry); 1095 1096 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n" 1097 , current->comm)); 1098 1099 rtn = scsi_try_host_reset(scmd); 1100 if (rtn == SUCCESS) { 1101 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1102 if (!scsi_device_online(scmd->device) || 1103 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) || 1104 !scsi_eh_tur(scmd)) 1105 scsi_eh_finish_cmd(scmd, done_q); 1106 } 1107 } else { 1108 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST" 1109 " failed\n", 1110 current->comm)); 1111 } 1112 } 1113 return list_empty(work_q); 1114 } 1115 1116 /** 1117 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1118 * @work_q: list_head for processed commands. 1119 * @done_q: list_head for processed commands. 1120 * 1121 **/ 1122 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1123 struct list_head *done_q) 1124 { 1125 struct scsi_cmnd *scmd, *next; 1126 1127 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1128 sdev_printk(KERN_INFO, scmd->device, 1129 "scsi: Device offlined - not" 1130 " ready after error recovery\n"); 1131 scsi_device_set_state(scmd->device, SDEV_OFFLINE); 1132 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) { 1133 /* 1134 * FIXME: Handle lost cmds. 1135 */ 1136 } 1137 scsi_eh_finish_cmd(scmd, done_q); 1138 } 1139 return; 1140 } 1141 1142 /** 1143 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1144 * @scmd: SCSI cmd to examine. 1145 * 1146 * Notes: 1147 * This is *only* called when we are examining the status after sending 1148 * out the actual data command. any commands that are queued for error 1149 * recovery (e.g. test_unit_ready) do *not* come through here. 1150 * 1151 * When this routine returns failed, it means the error handler thread 1152 * is woken. In cases where the error code indicates an error that 1153 * doesn't require the error handler read (i.e. we don't need to 1154 * abort/reset), this function should return SUCCESS. 1155 **/ 1156 int scsi_decide_disposition(struct scsi_cmnd *scmd) 1157 { 1158 int rtn; 1159 1160 /* 1161 * if the device is offline, then we clearly just pass the result back 1162 * up to the top level. 1163 */ 1164 if (!scsi_device_online(scmd->device)) { 1165 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report" 1166 " as SUCCESS\n", 1167 __FUNCTION__)); 1168 return SUCCESS; 1169 } 1170 1171 /* 1172 * first check the host byte, to see if there is anything in there 1173 * that would indicate what we need to do. 1174 */ 1175 switch (host_byte(scmd->result)) { 1176 case DID_PASSTHROUGH: 1177 /* 1178 * no matter what, pass this through to the upper layer. 1179 * nuke this special code so that it looks like we are saying 1180 * did_ok. 1181 */ 1182 scmd->result &= 0xff00ffff; 1183 return SUCCESS; 1184 case DID_OK: 1185 /* 1186 * looks good. drop through, and check the next byte. 1187 */ 1188 break; 1189 case DID_NO_CONNECT: 1190 case DID_BAD_TARGET: 1191 case DID_ABORT: 1192 /* 1193 * note - this means that we just report the status back 1194 * to the top level driver, not that we actually think 1195 * that it indicates SUCCESS. 1196 */ 1197 return SUCCESS; 1198 /* 1199 * when the low level driver returns did_soft_error, 1200 * it is responsible for keeping an internal retry counter 1201 * in order to avoid endless loops (db) 1202 * 1203 * actually this is a bug in this function here. we should 1204 * be mindful of the maximum number of retries specified 1205 * and not get stuck in a loop. 1206 */ 1207 case DID_SOFT_ERROR: 1208 goto maybe_retry; 1209 case DID_IMM_RETRY: 1210 return NEEDS_RETRY; 1211 1212 case DID_REQUEUE: 1213 return ADD_TO_MLQUEUE; 1214 1215 case DID_ERROR: 1216 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1217 status_byte(scmd->result) == RESERVATION_CONFLICT) 1218 /* 1219 * execute reservation conflict processing code 1220 * lower down 1221 */ 1222 break; 1223 /* fallthrough */ 1224 1225 case DID_BUS_BUSY: 1226 case DID_PARITY: 1227 goto maybe_retry; 1228 case DID_TIME_OUT: 1229 /* 1230 * when we scan the bus, we get timeout messages for 1231 * these commands if there is no device available. 1232 * other hosts report did_no_connect for the same thing. 1233 */ 1234 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1235 scmd->cmnd[0] == INQUIRY)) { 1236 return SUCCESS; 1237 } else { 1238 return FAILED; 1239 } 1240 case DID_RESET: 1241 return SUCCESS; 1242 default: 1243 return FAILED; 1244 } 1245 1246 /* 1247 * next, check the message byte. 1248 */ 1249 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 1250 return FAILED; 1251 1252 /* 1253 * check the status byte to see if this indicates anything special. 1254 */ 1255 switch (status_byte(scmd->result)) { 1256 case QUEUE_FULL: 1257 /* 1258 * the case of trying to send too many commands to a 1259 * tagged queueing device. 1260 */ 1261 case BUSY: 1262 /* 1263 * device can't talk to us at the moment. Should only 1264 * occur (SAM-3) when the task queue is empty, so will cause 1265 * the empty queue handling to trigger a stall in the 1266 * device. 1267 */ 1268 return ADD_TO_MLQUEUE; 1269 case GOOD: 1270 case COMMAND_TERMINATED: 1271 case TASK_ABORTED: 1272 return SUCCESS; 1273 case CHECK_CONDITION: 1274 rtn = scsi_check_sense(scmd); 1275 if (rtn == NEEDS_RETRY) 1276 goto maybe_retry; 1277 /* if rtn == FAILED, we have no sense information; 1278 * returning FAILED will wake the error handler thread 1279 * to collect the sense and redo the decide 1280 * disposition */ 1281 return rtn; 1282 case CONDITION_GOOD: 1283 case INTERMEDIATE_GOOD: 1284 case INTERMEDIATE_C_GOOD: 1285 case ACA_ACTIVE: 1286 /* 1287 * who knows? FIXME(eric) 1288 */ 1289 return SUCCESS; 1290 1291 case RESERVATION_CONFLICT: 1292 sdev_printk(KERN_INFO, scmd->device, 1293 "reservation conflict\n"); 1294 return SUCCESS; /* causes immediate i/o error */ 1295 default: 1296 return FAILED; 1297 } 1298 return FAILED; 1299 1300 maybe_retry: 1301 1302 /* we requeue for retry because the error was retryable, and 1303 * the request was not marked fast fail. Note that above, 1304 * even if the request is marked fast fail, we still requeue 1305 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1306 if ((++scmd->retries) <= scmd->allowed 1307 && !blk_noretry_request(scmd->request)) { 1308 return NEEDS_RETRY; 1309 } else { 1310 /* 1311 * no more retries - report this one back to upper level. 1312 */ 1313 return SUCCESS; 1314 } 1315 } 1316 1317 /** 1318 * scsi_eh_lock_door - Prevent medium removal for the specified device 1319 * @sdev: SCSI device to prevent medium removal 1320 * 1321 * Locking: 1322 * We must be called from process context; scsi_allocate_request() 1323 * may sleep. 1324 * 1325 * Notes: 1326 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 1327 * head of the devices request queue, and continue. 1328 * 1329 * Bugs: 1330 * scsi_allocate_request() may sleep waiting for existing requests to 1331 * be processed. However, since we haven't kicked off any request 1332 * processing for this host, this may deadlock. 1333 * 1334 * If scsi_allocate_request() fails for what ever reason, we 1335 * completely forget to lock the door. 1336 **/ 1337 static void scsi_eh_lock_door(struct scsi_device *sdev) 1338 { 1339 unsigned char cmnd[MAX_COMMAND_SIZE]; 1340 1341 cmnd[0] = ALLOW_MEDIUM_REMOVAL; 1342 cmnd[1] = 0; 1343 cmnd[2] = 0; 1344 cmnd[3] = 0; 1345 cmnd[4] = SCSI_REMOVAL_PREVENT; 1346 cmnd[5] = 0; 1347 1348 scsi_execute_async(sdev, cmnd, 6, DMA_NONE, NULL, 0, 0, 10 * HZ, 1349 5, NULL, NULL, GFP_KERNEL); 1350 } 1351 1352 1353 /** 1354 * scsi_restart_operations - restart io operations to the specified host. 1355 * @shost: Host we are restarting. 1356 * 1357 * Notes: 1358 * When we entered the error handler, we blocked all further i/o to 1359 * this device. we need to 'reverse' this process. 1360 **/ 1361 static void scsi_restart_operations(struct Scsi_Host *shost) 1362 { 1363 struct scsi_device *sdev; 1364 unsigned long flags; 1365 1366 /* 1367 * If the door was locked, we need to insert a door lock request 1368 * onto the head of the SCSI request queue for the device. There 1369 * is no point trying to lock the door of an off-line device. 1370 */ 1371 shost_for_each_device(sdev, shost) { 1372 if (scsi_device_online(sdev) && sdev->locked) 1373 scsi_eh_lock_door(sdev); 1374 } 1375 1376 /* 1377 * next free up anything directly waiting upon the host. this 1378 * will be requests for character device operations, and also for 1379 * ioctls to queued block devices. 1380 */ 1381 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n", 1382 __FUNCTION__)); 1383 1384 spin_lock_irqsave(shost->host_lock, flags); 1385 if (scsi_host_set_state(shost, SHOST_RUNNING)) 1386 if (scsi_host_set_state(shost, SHOST_CANCEL)) 1387 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 1388 spin_unlock_irqrestore(shost->host_lock, flags); 1389 1390 wake_up(&shost->host_wait); 1391 1392 /* 1393 * finally we need to re-initiate requests that may be pending. we will 1394 * have had everything blocked while error handling is taking place, and 1395 * now that error recovery is done, we will need to ensure that these 1396 * requests are started. 1397 */ 1398 scsi_run_host_queues(shost); 1399 } 1400 1401 /** 1402 * scsi_eh_ready_devs - check device ready state and recover if not. 1403 * @shost: host to be recovered. 1404 * @eh_done_q: list_head for processed commands. 1405 * 1406 **/ 1407 static void scsi_eh_ready_devs(struct Scsi_Host *shost, 1408 struct list_head *work_q, 1409 struct list_head *done_q) 1410 { 1411 if (!scsi_eh_stu(shost, work_q, done_q)) 1412 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 1413 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 1414 if (!scsi_eh_host_reset(work_q, done_q)) 1415 scsi_eh_offline_sdevs(work_q, done_q); 1416 } 1417 1418 /** 1419 * scsi_eh_flush_done_q - finish processed commands or retry them. 1420 * @done_q: list_head of processed commands. 1421 * 1422 **/ 1423 void scsi_eh_flush_done_q(struct list_head *done_q) 1424 { 1425 struct scsi_cmnd *scmd, *next; 1426 1427 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 1428 list_del_init(&scmd->eh_entry); 1429 if (scsi_device_online(scmd->device) && 1430 !blk_noretry_request(scmd->request) && 1431 (++scmd->retries <= scmd->allowed)) { 1432 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush" 1433 " retry cmd: %p\n", 1434 current->comm, 1435 scmd)); 1436 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 1437 } else { 1438 /* 1439 * If just we got sense for the device (called 1440 * scsi_eh_get_sense), scmd->result is already 1441 * set, do not set DRIVER_TIMEOUT. 1442 */ 1443 if (!scmd->result) 1444 scmd->result |= (DRIVER_TIMEOUT << 24); 1445 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish" 1446 " cmd: %p\n", 1447 current->comm, scmd)); 1448 scsi_finish_command(scmd); 1449 } 1450 } 1451 } 1452 EXPORT_SYMBOL(scsi_eh_flush_done_q); 1453 1454 /** 1455 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 1456 * @shost: Host to unjam. 1457 * 1458 * Notes: 1459 * When we come in here, we *know* that all commands on the bus have 1460 * either completed, failed or timed out. we also know that no further 1461 * commands are being sent to the host, so things are relatively quiet 1462 * and we have freedom to fiddle with things as we wish. 1463 * 1464 * This is only the *default* implementation. it is possible for 1465 * individual drivers to supply their own version of this function, and 1466 * if the maintainer wishes to do this, it is strongly suggested that 1467 * this function be taken as a template and modified. this function 1468 * was designed to correctly handle problems for about 95% of the 1469 * different cases out there, and it should always provide at least a 1470 * reasonable amount of error recovery. 1471 * 1472 * Any command marked 'failed' or 'timeout' must eventually have 1473 * scsi_finish_cmd() called for it. we do all of the retry stuff 1474 * here, so when we restart the host after we return it should have an 1475 * empty queue. 1476 **/ 1477 static void scsi_unjam_host(struct Scsi_Host *shost) 1478 { 1479 unsigned long flags; 1480 LIST_HEAD(eh_work_q); 1481 LIST_HEAD(eh_done_q); 1482 1483 spin_lock_irqsave(shost->host_lock, flags); 1484 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 1485 spin_unlock_irqrestore(shost->host_lock, flags); 1486 1487 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 1488 1489 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 1490 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q)) 1491 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 1492 1493 scsi_eh_flush_done_q(&eh_done_q); 1494 } 1495 1496 /** 1497 * scsi_error_handler - SCSI error handler thread 1498 * @data: Host for which we are running. 1499 * 1500 * Notes: 1501 * This is the main error handling loop. This is run as a kernel thread 1502 * for every SCSI host and handles all error handling activity. 1503 **/ 1504 int scsi_error_handler(void *data) 1505 { 1506 struct Scsi_Host *shost = data; 1507 1508 current->flags |= PF_NOFREEZE; 1509 1510 /* 1511 * We use TASK_INTERRUPTIBLE so that the thread is not 1512 * counted against the load average as a running process. 1513 * We never actually get interrupted because kthread_run 1514 * disables singal delivery for the created thread. 1515 */ 1516 set_current_state(TASK_INTERRUPTIBLE); 1517 while (!kthread_should_stop()) { 1518 if (shost->host_failed == 0 || 1519 shost->host_failed != shost->host_busy) { 1520 SCSI_LOG_ERROR_RECOVERY(1, 1521 printk("Error handler scsi_eh_%d sleeping\n", 1522 shost->host_no)); 1523 schedule(); 1524 set_current_state(TASK_INTERRUPTIBLE); 1525 continue; 1526 } 1527 1528 __set_current_state(TASK_RUNNING); 1529 SCSI_LOG_ERROR_RECOVERY(1, 1530 printk("Error handler scsi_eh_%d waking up\n", 1531 shost->host_no)); 1532 1533 /* 1534 * We have a host that is failing for some reason. Figure out 1535 * what we need to do to get it up and online again (if we can). 1536 * If we fail, we end up taking the thing offline. 1537 */ 1538 if (shost->transportt->eh_strategy_handler) 1539 shost->transportt->eh_strategy_handler(shost); 1540 else 1541 scsi_unjam_host(shost); 1542 1543 /* 1544 * Note - if the above fails completely, the action is to take 1545 * individual devices offline and flush the queue of any 1546 * outstanding requests that may have been pending. When we 1547 * restart, we restart any I/O to any other devices on the bus 1548 * which are still online. 1549 */ 1550 scsi_restart_operations(shost); 1551 set_current_state(TASK_INTERRUPTIBLE); 1552 } 1553 __set_current_state(TASK_RUNNING); 1554 1555 SCSI_LOG_ERROR_RECOVERY(1, 1556 printk("Error handler scsi_eh_%d exiting\n", shost->host_no)); 1557 shost->ehandler = NULL; 1558 return 0; 1559 } 1560 1561 /* 1562 * Function: scsi_report_bus_reset() 1563 * 1564 * Purpose: Utility function used by low-level drivers to report that 1565 * they have observed a bus reset on the bus being handled. 1566 * 1567 * Arguments: shost - Host in question 1568 * channel - channel on which reset was observed. 1569 * 1570 * Returns: Nothing 1571 * 1572 * Lock status: Host lock must be held. 1573 * 1574 * Notes: This only needs to be called if the reset is one which 1575 * originates from an unknown location. Resets originated 1576 * by the mid-level itself don't need to call this, but there 1577 * should be no harm. 1578 * 1579 * The main purpose of this is to make sure that a CHECK_CONDITION 1580 * is properly treated. 1581 */ 1582 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 1583 { 1584 struct scsi_device *sdev; 1585 1586 __shost_for_each_device(sdev, shost) { 1587 if (channel == sdev_channel(sdev)) { 1588 sdev->was_reset = 1; 1589 sdev->expecting_cc_ua = 1; 1590 } 1591 } 1592 } 1593 EXPORT_SYMBOL(scsi_report_bus_reset); 1594 1595 /* 1596 * Function: scsi_report_device_reset() 1597 * 1598 * Purpose: Utility function used by low-level drivers to report that 1599 * they have observed a device reset on the device being handled. 1600 * 1601 * Arguments: shost - Host in question 1602 * channel - channel on which reset was observed 1603 * target - target on which reset was observed 1604 * 1605 * Returns: Nothing 1606 * 1607 * Lock status: Host lock must be held 1608 * 1609 * Notes: This only needs to be called if the reset is one which 1610 * originates from an unknown location. Resets originated 1611 * by the mid-level itself don't need to call this, but there 1612 * should be no harm. 1613 * 1614 * The main purpose of this is to make sure that a CHECK_CONDITION 1615 * is properly treated. 1616 */ 1617 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 1618 { 1619 struct scsi_device *sdev; 1620 1621 __shost_for_each_device(sdev, shost) { 1622 if (channel == sdev_channel(sdev) && 1623 target == sdev_id(sdev)) { 1624 sdev->was_reset = 1; 1625 sdev->expecting_cc_ua = 1; 1626 } 1627 } 1628 } 1629 EXPORT_SYMBOL(scsi_report_device_reset); 1630 1631 static void 1632 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 1633 { 1634 } 1635 1636 /* 1637 * Function: scsi_reset_provider 1638 * 1639 * Purpose: Send requested reset to a bus or device at any phase. 1640 * 1641 * Arguments: device - device to send reset to 1642 * flag - reset type (see scsi.h) 1643 * 1644 * Returns: SUCCESS/FAILURE. 1645 * 1646 * Notes: This is used by the SCSI Generic driver to provide 1647 * Bus/Device reset capability. 1648 */ 1649 int 1650 scsi_reset_provider(struct scsi_device *dev, int flag) 1651 { 1652 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL); 1653 struct request req; 1654 int rtn; 1655 1656 scmd->request = &req; 1657 memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout)); 1658 1659 memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd)); 1660 1661 scmd->scsi_done = scsi_reset_provider_done_command; 1662 scmd->done = NULL; 1663 scmd->buffer = NULL; 1664 scmd->bufflen = 0; 1665 scmd->request_buffer = NULL; 1666 scmd->request_bufflen = 0; 1667 1668 scmd->cmd_len = 0; 1669 1670 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 1671 1672 init_timer(&scmd->eh_timeout); 1673 1674 /* 1675 * Sometimes the command can get back into the timer chain, 1676 * so use the pid as an identifier. 1677 */ 1678 scmd->pid = 0; 1679 1680 switch (flag) { 1681 case SCSI_TRY_RESET_DEVICE: 1682 rtn = scsi_try_bus_device_reset(scmd); 1683 if (rtn == SUCCESS) 1684 break; 1685 /* FALLTHROUGH */ 1686 case SCSI_TRY_RESET_BUS: 1687 rtn = scsi_try_bus_reset(scmd); 1688 if (rtn == SUCCESS) 1689 break; 1690 /* FALLTHROUGH */ 1691 case SCSI_TRY_RESET_HOST: 1692 rtn = scsi_try_host_reset(scmd); 1693 break; 1694 default: 1695 rtn = FAILED; 1696 } 1697 1698 scsi_next_command(scmd); 1699 return rtn; 1700 } 1701 EXPORT_SYMBOL(scsi_reset_provider); 1702 1703 /** 1704 * scsi_normalize_sense - normalize main elements from either fixed or 1705 * descriptor sense data format into a common format. 1706 * 1707 * @sense_buffer: byte array containing sense data returned by device 1708 * @sb_len: number of valid bytes in sense_buffer 1709 * @sshdr: pointer to instance of structure that common 1710 * elements are written to. 1711 * 1712 * Notes: 1713 * The "main elements" from sense data are: response_code, sense_key, 1714 * asc, ascq and additional_length (only for descriptor format). 1715 * 1716 * Typically this function can be called after a device has 1717 * responded to a SCSI command with the CHECK_CONDITION status. 1718 * 1719 * Return value: 1720 * 1 if valid sense data information found, else 0; 1721 **/ 1722 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len, 1723 struct scsi_sense_hdr *sshdr) 1724 { 1725 if (!sense_buffer || !sb_len) 1726 return 0; 1727 1728 memset(sshdr, 0, sizeof(struct scsi_sense_hdr)); 1729 1730 sshdr->response_code = (sense_buffer[0] & 0x7f); 1731 1732 if (!scsi_sense_valid(sshdr)) 1733 return 0; 1734 1735 if (sshdr->response_code >= 0x72) { 1736 /* 1737 * descriptor format 1738 */ 1739 if (sb_len > 1) 1740 sshdr->sense_key = (sense_buffer[1] & 0xf); 1741 if (sb_len > 2) 1742 sshdr->asc = sense_buffer[2]; 1743 if (sb_len > 3) 1744 sshdr->ascq = sense_buffer[3]; 1745 if (sb_len > 7) 1746 sshdr->additional_length = sense_buffer[7]; 1747 } else { 1748 /* 1749 * fixed format 1750 */ 1751 if (sb_len > 2) 1752 sshdr->sense_key = (sense_buffer[2] & 0xf); 1753 if (sb_len > 7) { 1754 sb_len = (sb_len < (sense_buffer[7] + 8)) ? 1755 sb_len : (sense_buffer[7] + 8); 1756 if (sb_len > 12) 1757 sshdr->asc = sense_buffer[12]; 1758 if (sb_len > 13) 1759 sshdr->ascq = sense_buffer[13]; 1760 } 1761 } 1762 1763 return 1; 1764 } 1765 EXPORT_SYMBOL(scsi_normalize_sense); 1766 1767 int scsi_command_normalize_sense(struct scsi_cmnd *cmd, 1768 struct scsi_sense_hdr *sshdr) 1769 { 1770 return scsi_normalize_sense(cmd->sense_buffer, 1771 sizeof(cmd->sense_buffer), sshdr); 1772 } 1773 EXPORT_SYMBOL(scsi_command_normalize_sense); 1774 1775 /** 1776 * scsi_sense_desc_find - search for a given descriptor type in 1777 * descriptor sense data format. 1778 * 1779 * @sense_buffer: byte array of descriptor format sense data 1780 * @sb_len: number of valid bytes in sense_buffer 1781 * @desc_type: value of descriptor type to find 1782 * (e.g. 0 -> information) 1783 * 1784 * Notes: 1785 * only valid when sense data is in descriptor format 1786 * 1787 * Return value: 1788 * pointer to start of (first) descriptor if found else NULL 1789 **/ 1790 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len, 1791 int desc_type) 1792 { 1793 int add_sen_len, add_len, desc_len, k; 1794 const u8 * descp; 1795 1796 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7]))) 1797 return NULL; 1798 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73)) 1799 return NULL; 1800 add_sen_len = (add_sen_len < (sb_len - 8)) ? 1801 add_sen_len : (sb_len - 8); 1802 descp = &sense_buffer[8]; 1803 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) { 1804 descp += desc_len; 1805 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1; 1806 desc_len = add_len + 2; 1807 if (descp[0] == desc_type) 1808 return descp; 1809 if (add_len < 0) // short descriptor ?? 1810 break; 1811 } 1812 return NULL; 1813 } 1814 EXPORT_SYMBOL(scsi_sense_desc_find); 1815 1816 /** 1817 * scsi_get_sense_info_fld - attempts to get information field from 1818 * sense data (either fixed or descriptor format) 1819 * 1820 * @sense_buffer: byte array of sense data 1821 * @sb_len: number of valid bytes in sense_buffer 1822 * @info_out: pointer to 64 integer where 8 or 4 byte information 1823 * field will be placed if found. 1824 * 1825 * Return value: 1826 * 1 if information field found, 0 if not found. 1827 **/ 1828 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len, 1829 u64 * info_out) 1830 { 1831 int j; 1832 const u8 * ucp; 1833 u64 ull; 1834 1835 if (sb_len < 7) 1836 return 0; 1837 switch (sense_buffer[0] & 0x7f) { 1838 case 0x70: 1839 case 0x71: 1840 if (sense_buffer[0] & 0x80) { 1841 *info_out = (sense_buffer[3] << 24) + 1842 (sense_buffer[4] << 16) + 1843 (sense_buffer[5] << 8) + sense_buffer[6]; 1844 return 1; 1845 } else 1846 return 0; 1847 case 0x72: 1848 case 0x73: 1849 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 1850 0 /* info desc */); 1851 if (ucp && (0xa == ucp[1])) { 1852 ull = 0; 1853 for (j = 0; j < 8; ++j) { 1854 if (j > 0) 1855 ull <<= 8; 1856 ull |= ucp[4 + j]; 1857 } 1858 *info_out = ull; 1859 return 1; 1860 } else 1861 return 0; 1862 default: 1863 return 0; 1864 } 1865 } 1866 EXPORT_SYMBOL(scsi_get_sense_info_fld); 1867