1 /* 2 * scsi_error.c Copyright (C) 1997 Eric Youngdale 3 * 4 * SCSI error/timeout handling 5 * Initial versions: Eric Youngdale. Based upon conversations with 6 * Leonard Zubkoff and David Miller at Linux Expo, 7 * ideas originating from all over the place. 8 * 9 * Restructured scsi_unjam_host and associated functions. 10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com) 11 * 12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and 13 * minor cleanups. 14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com) 15 */ 16 17 #include <linux/module.h> 18 #include <linux/sched.h> 19 #include <linux/timer.h> 20 #include <linux/string.h> 21 #include <linux/slab.h> 22 #include <linux/kernel.h> 23 #include <linux/kthread.h> 24 #include <linux/interrupt.h> 25 #include <linux/blkdev.h> 26 #include <linux/delay.h> 27 28 #include <scsi/scsi.h> 29 #include <scsi/scsi_cmnd.h> 30 #include <scsi/scsi_dbg.h> 31 #include <scsi/scsi_device.h> 32 #include <scsi/scsi_eh.h> 33 #include <scsi/scsi_transport.h> 34 #include <scsi/scsi_host.h> 35 #include <scsi/scsi_ioctl.h> 36 37 #include "scsi_priv.h" 38 #include "scsi_logging.h" 39 40 #define SENSE_TIMEOUT (10*HZ) 41 #define START_UNIT_TIMEOUT (30*HZ) 42 43 /* 44 * These should *probably* be handled by the host itself. 45 * Since it is allowed to sleep, it probably should. 46 */ 47 #define BUS_RESET_SETTLE_TIME (10) 48 #define HOST_RESET_SETTLE_TIME (10) 49 50 /* called with shost->host_lock held */ 51 void scsi_eh_wakeup(struct Scsi_Host *shost) 52 { 53 if (shost->host_busy == shost->host_failed) { 54 wake_up_process(shost->ehandler); 55 SCSI_LOG_ERROR_RECOVERY(5, 56 printk("Waking error handler thread\n")); 57 } 58 } 59 60 /** 61 * scsi_schedule_eh - schedule EH for SCSI host 62 * @shost: SCSI host to invoke error handling on. 63 * 64 * Schedule SCSI EH without scmd. 65 **/ 66 void scsi_schedule_eh(struct Scsi_Host *shost) 67 { 68 unsigned long flags; 69 70 spin_lock_irqsave(shost->host_lock, flags); 71 72 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || 73 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { 74 shost->host_eh_scheduled++; 75 scsi_eh_wakeup(shost); 76 } 77 78 spin_unlock_irqrestore(shost->host_lock, flags); 79 } 80 EXPORT_SYMBOL_GPL(scsi_schedule_eh); 81 82 /** 83 * scsi_eh_scmd_add - add scsi cmd to error handling. 84 * @scmd: scmd to run eh on. 85 * @eh_flag: optional SCSI_EH flag. 86 * 87 * Return value: 88 * 0 on failure. 89 **/ 90 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag) 91 { 92 struct Scsi_Host *shost = scmd->device->host; 93 unsigned long flags; 94 int ret = 0; 95 96 if (!shost->ehandler) 97 return 0; 98 99 spin_lock_irqsave(shost->host_lock, flags); 100 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 101 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)) 102 goto out_unlock; 103 104 ret = 1; 105 scmd->eh_eflags |= eh_flag; 106 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 107 shost->host_failed++; 108 scsi_eh_wakeup(shost); 109 out_unlock: 110 spin_unlock_irqrestore(shost->host_lock, flags); 111 return ret; 112 } 113 114 /** 115 * scsi_add_timer - Start timeout timer for a single scsi command. 116 * @scmd: scsi command that is about to start running. 117 * @timeout: amount of time to allow this command to run. 118 * @complete: timeout function to call if timer isn't canceled. 119 * 120 * Notes: 121 * This should be turned into an inline function. Each scsi command 122 * has its own timer, and as it is added to the queue, we set up the 123 * timer. When the command completes, we cancel the timer. 124 **/ 125 void scsi_add_timer(struct scsi_cmnd *scmd, int timeout, 126 void (*complete)(struct scsi_cmnd *)) 127 { 128 129 /* 130 * If the clock was already running for this command, then 131 * first delete the timer. The timer handling code gets rather 132 * confused if we don't do this. 133 */ 134 if (scmd->eh_timeout.function) 135 del_timer(&scmd->eh_timeout); 136 137 scmd->eh_timeout.data = (unsigned long)scmd; 138 scmd->eh_timeout.expires = jiffies + timeout; 139 scmd->eh_timeout.function = (void (*)(unsigned long)) complete; 140 141 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:" 142 " %d, (%p)\n", __FUNCTION__, 143 scmd, timeout, complete)); 144 145 add_timer(&scmd->eh_timeout); 146 } 147 148 /** 149 * scsi_delete_timer - Delete/cancel timer for a given function. 150 * @scmd: Cmd that we are canceling timer for 151 * 152 * Notes: 153 * This should be turned into an inline function. 154 * 155 * Return value: 156 * 1 if we were able to detach the timer. 0 if we blew it, and the 157 * timer function has already started to run. 158 **/ 159 int scsi_delete_timer(struct scsi_cmnd *scmd) 160 { 161 int rtn; 162 163 rtn = del_timer(&scmd->eh_timeout); 164 165 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p," 166 " rtn: %d\n", __FUNCTION__, 167 scmd, rtn)); 168 169 scmd->eh_timeout.data = (unsigned long)NULL; 170 scmd->eh_timeout.function = NULL; 171 172 return rtn; 173 } 174 175 /** 176 * scsi_times_out - Timeout function for normal scsi commands. 177 * @scmd: Cmd that is timing out. 178 * 179 * Notes: 180 * We do not need to lock this. There is the potential for a race 181 * only in that the normal completion handling might run, but if the 182 * normal completion function determines that the timer has already 183 * fired, then it mustn't do anything. 184 **/ 185 void scsi_times_out(struct scsi_cmnd *scmd) 186 { 187 enum scsi_eh_timer_return (* eh_timed_out)(struct scsi_cmnd *); 188 189 scsi_log_completion(scmd, TIMEOUT_ERROR); 190 191 if (scmd->device->host->transportt->eh_timed_out) 192 eh_timed_out = scmd->device->host->transportt->eh_timed_out; 193 else if (scmd->device->host->hostt->eh_timed_out) 194 eh_timed_out = scmd->device->host->hostt->eh_timed_out; 195 else 196 eh_timed_out = NULL; 197 198 if (eh_timed_out) 199 switch (eh_timed_out(scmd)) { 200 case EH_HANDLED: 201 __scsi_done(scmd); 202 return; 203 case EH_RESET_TIMER: 204 scsi_add_timer(scmd, scmd->timeout_per_command, 205 scsi_times_out); 206 return; 207 case EH_NOT_HANDLED: 208 break; 209 } 210 211 if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) { 212 scmd->result |= DID_TIME_OUT << 16; 213 __scsi_done(scmd); 214 } 215 } 216 217 /** 218 * scsi_block_when_processing_errors - Prevent cmds from being queued. 219 * @sdev: Device on which we are performing recovery. 220 * 221 * Description: 222 * We block until the host is out of error recovery, and then check to 223 * see whether the host or the device is offline. 224 * 225 * Return value: 226 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 227 **/ 228 int scsi_block_when_processing_errors(struct scsi_device *sdev) 229 { 230 int online; 231 232 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 233 234 online = scsi_device_online(sdev); 235 236 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__, 237 online)); 238 239 return online; 240 } 241 EXPORT_SYMBOL(scsi_block_when_processing_errors); 242 243 #ifdef CONFIG_SCSI_LOGGING 244 /** 245 * scsi_eh_prt_fail_stats - Log info on failures. 246 * @shost: scsi host being recovered. 247 * @work_q: Queue of scsi cmds to process. 248 **/ 249 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 250 struct list_head *work_q) 251 { 252 struct scsi_cmnd *scmd; 253 struct scsi_device *sdev; 254 int total_failures = 0; 255 int cmd_failed = 0; 256 int cmd_cancel = 0; 257 int devices_failed = 0; 258 259 shost_for_each_device(sdev, shost) { 260 list_for_each_entry(scmd, work_q, eh_entry) { 261 if (scmd->device == sdev) { 262 ++total_failures; 263 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) 264 ++cmd_cancel; 265 else 266 ++cmd_failed; 267 } 268 } 269 270 if (cmd_cancel || cmd_failed) { 271 SCSI_LOG_ERROR_RECOVERY(3, 272 sdev_printk(KERN_INFO, sdev, 273 "%s: cmds failed: %d, cancel: %d\n", 274 __FUNCTION__, cmd_failed, 275 cmd_cancel)); 276 cmd_cancel = 0; 277 cmd_failed = 0; 278 ++devices_failed; 279 } 280 } 281 282 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d" 283 " devices require eh work\n", 284 total_failures, devices_failed)); 285 } 286 #endif 287 288 /** 289 * scsi_check_sense - Examine scsi cmd sense 290 * @scmd: Cmd to have sense checked. 291 * 292 * Return value: 293 * SUCCESS or FAILED or NEEDS_RETRY 294 * 295 * Notes: 296 * When a deferred error is detected the current command has 297 * not been executed and needs retrying. 298 **/ 299 static int scsi_check_sense(struct scsi_cmnd *scmd) 300 { 301 struct scsi_sense_hdr sshdr; 302 303 if (! scsi_command_normalize_sense(scmd, &sshdr)) 304 return FAILED; /* no valid sense data */ 305 306 if (scsi_sense_is_deferred(&sshdr)) 307 return NEEDS_RETRY; 308 309 /* 310 * Previous logic looked for FILEMARK, EOM or ILI which are 311 * mainly associated with tapes and returned SUCCESS. 312 */ 313 if (sshdr.response_code == 0x70) { 314 /* fixed format */ 315 if (scmd->sense_buffer[2] & 0xe0) 316 return SUCCESS; 317 } else { 318 /* 319 * descriptor format: look for "stream commands sense data 320 * descriptor" (see SSC-3). Assume single sense data 321 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 322 */ 323 if ((sshdr.additional_length > 3) && 324 (scmd->sense_buffer[8] == 0x4) && 325 (scmd->sense_buffer[11] & 0xe0)) 326 return SUCCESS; 327 } 328 329 switch (sshdr.sense_key) { 330 case NO_SENSE: 331 return SUCCESS; 332 case RECOVERED_ERROR: 333 return /* soft_error */ SUCCESS; 334 335 case ABORTED_COMMAND: 336 return NEEDS_RETRY; 337 case NOT_READY: 338 case UNIT_ATTENTION: 339 /* 340 * if we are expecting a cc/ua because of a bus reset that we 341 * performed, treat this just as a retry. otherwise this is 342 * information that we should pass up to the upper-level driver 343 * so that we can deal with it there. 344 */ 345 if (scmd->device->expecting_cc_ua) { 346 scmd->device->expecting_cc_ua = 0; 347 return NEEDS_RETRY; 348 } 349 /* 350 * if the device is in the process of becoming ready, we 351 * should retry. 352 */ 353 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 354 return NEEDS_RETRY; 355 /* 356 * if the device is not started, we need to wake 357 * the error handler to start the motor 358 */ 359 if (scmd->device->allow_restart && 360 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 361 return FAILED; 362 return SUCCESS; 363 364 /* these three are not supported */ 365 case COPY_ABORTED: 366 case VOLUME_OVERFLOW: 367 case MISCOMPARE: 368 return SUCCESS; 369 370 case MEDIUM_ERROR: 371 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 372 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 373 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 374 return SUCCESS; 375 } 376 return NEEDS_RETRY; 377 378 case HARDWARE_ERROR: 379 if (scmd->device->retry_hwerror) 380 return NEEDS_RETRY; 381 else 382 return SUCCESS; 383 384 case ILLEGAL_REQUEST: 385 case BLANK_CHECK: 386 case DATA_PROTECT: 387 default: 388 return SUCCESS; 389 } 390 } 391 392 /** 393 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 394 * @scmd: SCSI cmd to examine. 395 * 396 * Notes: 397 * This is *only* called when we are examining the status of commands 398 * queued during error recovery. the main difference here is that we 399 * don't allow for the possibility of retries here, and we are a lot 400 * more restrictive about what we consider acceptable. 401 **/ 402 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) 403 { 404 /* 405 * first check the host byte, to see if there is anything in there 406 * that would indicate what we need to do. 407 */ 408 if (host_byte(scmd->result) == DID_RESET) { 409 /* 410 * rats. we are already in the error handler, so we now 411 * get to try and figure out what to do next. if the sense 412 * is valid, we have a pretty good idea of what to do. 413 * if not, we mark it as FAILED. 414 */ 415 return scsi_check_sense(scmd); 416 } 417 if (host_byte(scmd->result) != DID_OK) 418 return FAILED; 419 420 /* 421 * next, check the message byte. 422 */ 423 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 424 return FAILED; 425 426 /* 427 * now, check the status byte to see if this indicates 428 * anything special. 429 */ 430 switch (status_byte(scmd->result)) { 431 case GOOD: 432 case COMMAND_TERMINATED: 433 return SUCCESS; 434 case CHECK_CONDITION: 435 return scsi_check_sense(scmd); 436 case CONDITION_GOOD: 437 case INTERMEDIATE_GOOD: 438 case INTERMEDIATE_C_GOOD: 439 /* 440 * who knows? FIXME(eric) 441 */ 442 return SUCCESS; 443 case BUSY: 444 case QUEUE_FULL: 445 case RESERVATION_CONFLICT: 446 default: 447 return FAILED; 448 } 449 return FAILED; 450 } 451 452 /** 453 * scsi_eh_done - Completion function for error handling. 454 * @scmd: Cmd that is done. 455 **/ 456 static void scsi_eh_done(struct scsi_cmnd *scmd) 457 { 458 struct completion *eh_action; 459 460 SCSI_LOG_ERROR_RECOVERY(3, 461 printk("%s scmd: %p result: %x\n", 462 __FUNCTION__, scmd, scmd->result)); 463 464 eh_action = scmd->device->host->eh_action; 465 if (eh_action) 466 complete(eh_action); 467 } 468 469 /** 470 * scsi_try_host_reset - ask host adapter to reset itself 471 * @scmd: SCSI cmd to send hsot reset. 472 **/ 473 static int scsi_try_host_reset(struct scsi_cmnd *scmd) 474 { 475 unsigned long flags; 476 int rtn; 477 478 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n", 479 __FUNCTION__)); 480 481 if (!scmd->device->host->hostt->eh_host_reset_handler) 482 return FAILED; 483 484 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd); 485 486 if (rtn == SUCCESS) { 487 if (!scmd->device->host->hostt->skip_settle_delay) 488 ssleep(HOST_RESET_SETTLE_TIME); 489 spin_lock_irqsave(scmd->device->host->host_lock, flags); 490 scsi_report_bus_reset(scmd->device->host, 491 scmd_channel(scmd)); 492 spin_unlock_irqrestore(scmd->device->host->host_lock, flags); 493 } 494 495 return rtn; 496 } 497 498 /** 499 * scsi_try_bus_reset - ask host to perform a bus reset 500 * @scmd: SCSI cmd to send bus reset. 501 **/ 502 static int scsi_try_bus_reset(struct scsi_cmnd *scmd) 503 { 504 unsigned long flags; 505 int rtn; 506 507 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n", 508 __FUNCTION__)); 509 510 if (!scmd->device->host->hostt->eh_bus_reset_handler) 511 return FAILED; 512 513 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd); 514 515 if (rtn == SUCCESS) { 516 if (!scmd->device->host->hostt->skip_settle_delay) 517 ssleep(BUS_RESET_SETTLE_TIME); 518 spin_lock_irqsave(scmd->device->host->host_lock, flags); 519 scsi_report_bus_reset(scmd->device->host, 520 scmd_channel(scmd)); 521 spin_unlock_irqrestore(scmd->device->host->host_lock, flags); 522 } 523 524 return rtn; 525 } 526 527 /** 528 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 529 * @scmd: SCSI cmd used to send BDR 530 * 531 * Notes: 532 * There is no timeout for this operation. if this operation is 533 * unreliable for a given host, then the host itself needs to put a 534 * timer on it, and set the host back to a consistent state prior to 535 * returning. 536 **/ 537 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 538 { 539 int rtn; 540 541 if (!scmd->device->host->hostt->eh_device_reset_handler) 542 return FAILED; 543 544 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd); 545 if (rtn == SUCCESS) { 546 scmd->device->was_reset = 1; 547 scmd->device->expecting_cc_ua = 1; 548 } 549 550 return rtn; 551 } 552 553 static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd) 554 { 555 if (!scmd->device->host->hostt->eh_abort_handler) 556 return FAILED; 557 558 return scmd->device->host->hostt->eh_abort_handler(scmd); 559 } 560 561 /** 562 * scsi_try_to_abort_cmd - Ask host to abort a running command. 563 * @scmd: SCSI cmd to abort from Lower Level. 564 * 565 * Notes: 566 * This function will not return until the user's completion function 567 * has been called. there is no timeout on this operation. if the 568 * author of the low-level driver wishes this operation to be timed, 569 * they can provide this facility themselves. helper functions in 570 * scsi_error.c can be supplied to make this easier to do. 571 **/ 572 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd) 573 { 574 /* 575 * scsi_done was called just after the command timed out and before 576 * we had a chance to process it. (db) 577 */ 578 if (scmd->serial_number == 0) 579 return SUCCESS; 580 return __scsi_try_to_abort_cmd(scmd); 581 } 582 583 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 584 { 585 if (__scsi_try_to_abort_cmd(scmd) != SUCCESS) 586 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 587 if (scsi_try_bus_reset(scmd) != SUCCESS) 588 scsi_try_host_reset(scmd); 589 } 590 591 /** 592 * scsi_send_eh_cmnd - submit a scsi command as part of error recory 593 * @scmd: SCSI command structure to hijack 594 * @cmnd: CDB to send 595 * @cmnd_size: size in bytes of @cmnd 596 * @timeout: timeout for this request 597 * @copy_sense: request sense data if set to 1 598 * 599 * This function is used to send a scsi command down to a target device 600 * as part of the error recovery process. If @copy_sense is 0 the command 601 * sent must be one that does not transfer any data. If @copy_sense is 1 602 * the command must be REQUEST_SENSE and this functions copies out the 603 * sense buffer it got into @scmd->sense_buffer. 604 * 605 * Return value: 606 * SUCCESS or FAILED or NEEDS_RETRY 607 **/ 608 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd, 609 int cmnd_size, int timeout, int copy_sense) 610 { 611 struct scsi_device *sdev = scmd->device; 612 struct Scsi_Host *shost = sdev->host; 613 int old_result = scmd->result; 614 DECLARE_COMPLETION_ONSTACK(done); 615 unsigned long timeleft; 616 unsigned long flags; 617 struct scatterlist sgl; 618 unsigned char old_cmnd[MAX_COMMAND_SIZE]; 619 enum dma_data_direction old_data_direction; 620 unsigned short old_use_sg; 621 unsigned char old_cmd_len; 622 unsigned old_bufflen; 623 void *old_buffer; 624 int rtn; 625 626 /* 627 * We need saved copies of a number of fields - this is because 628 * error handling may need to overwrite these with different values 629 * to run different commands, and once error handling is complete, 630 * we will need to restore these values prior to running the actual 631 * command. 632 */ 633 old_buffer = scmd->request_buffer; 634 old_bufflen = scmd->request_bufflen; 635 memcpy(old_cmnd, scmd->cmnd, sizeof(scmd->cmnd)); 636 old_data_direction = scmd->sc_data_direction; 637 old_cmd_len = scmd->cmd_len; 638 old_use_sg = scmd->use_sg; 639 640 memset(scmd->cmnd, 0, sizeof(scmd->cmnd)); 641 memcpy(scmd->cmnd, cmnd, cmnd_size); 642 643 if (copy_sense) { 644 gfp_t gfp_mask = GFP_ATOMIC; 645 646 if (shost->hostt->unchecked_isa_dma) 647 gfp_mask |= __GFP_DMA; 648 649 sgl.page = alloc_page(gfp_mask); 650 if (!sgl.page) 651 return FAILED; 652 sgl.offset = 0; 653 sgl.length = 252; 654 655 scmd->sc_data_direction = DMA_FROM_DEVICE; 656 scmd->request_bufflen = sgl.length; 657 scmd->request_buffer = &sgl; 658 scmd->use_sg = 1; 659 } else { 660 scmd->request_buffer = NULL; 661 scmd->request_bufflen = 0; 662 scmd->sc_data_direction = DMA_NONE; 663 scmd->use_sg = 0; 664 } 665 666 scmd->underflow = 0; 667 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 668 669 if (sdev->scsi_level <= SCSI_2) 670 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 671 (sdev->lun << 5 & 0xe0); 672 673 /* 674 * Zero the sense buffer. The scsi spec mandates that any 675 * untransferred sense data should be interpreted as being zero. 676 */ 677 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); 678 679 shost->eh_action = &done; 680 681 spin_lock_irqsave(shost->host_lock, flags); 682 scsi_log_send(scmd); 683 shost->hostt->queuecommand(scmd, scsi_eh_done); 684 spin_unlock_irqrestore(shost->host_lock, flags); 685 686 timeleft = wait_for_completion_timeout(&done, timeout); 687 688 shost->eh_action = NULL; 689 690 scsi_log_completion(scmd, SUCCESS); 691 692 SCSI_LOG_ERROR_RECOVERY(3, 693 printk("%s: scmd: %p, timeleft: %ld\n", 694 __FUNCTION__, scmd, timeleft)); 695 696 /* 697 * If there is time left scsi_eh_done got called, and we will 698 * examine the actual status codes to see whether the command 699 * actually did complete normally, else tell the host to forget 700 * about this command. 701 */ 702 if (timeleft) { 703 rtn = scsi_eh_completed_normally(scmd); 704 SCSI_LOG_ERROR_RECOVERY(3, 705 printk("%s: scsi_eh_completed_normally %x\n", 706 __FUNCTION__, rtn)); 707 708 switch (rtn) { 709 case SUCCESS: 710 case NEEDS_RETRY: 711 case FAILED: 712 break; 713 default: 714 rtn = FAILED; 715 break; 716 } 717 } else { 718 scsi_abort_eh_cmnd(scmd); 719 rtn = FAILED; 720 } 721 722 723 /* 724 * Last chance to have valid sense data. 725 */ 726 if (copy_sense) { 727 if (!SCSI_SENSE_VALID(scmd)) { 728 memcpy(scmd->sense_buffer, page_address(sgl.page), 729 sizeof(scmd->sense_buffer)); 730 } 731 __free_page(sgl.page); 732 } 733 734 735 /* 736 * Restore original data 737 */ 738 scmd->request_buffer = old_buffer; 739 scmd->request_bufflen = old_bufflen; 740 memcpy(scmd->cmnd, old_cmnd, sizeof(scmd->cmnd)); 741 scmd->sc_data_direction = old_data_direction; 742 scmd->cmd_len = old_cmd_len; 743 scmd->use_sg = old_use_sg; 744 scmd->result = old_result; 745 return rtn; 746 } 747 748 /** 749 * scsi_request_sense - Request sense data from a particular target. 750 * @scmd: SCSI cmd for request sense. 751 * 752 * Notes: 753 * Some hosts automatically obtain this information, others require 754 * that we obtain it on our own. This function will *not* return until 755 * the command either times out, or it completes. 756 **/ 757 static int scsi_request_sense(struct scsi_cmnd *scmd) 758 { 759 static unsigned char generic_sense[6] = 760 {REQUEST_SENSE, 0, 0, 0, 252, 0}; 761 762 return scsi_send_eh_cmnd(scmd, generic_sense, 6, SENSE_TIMEOUT, 1); 763 } 764 765 /** 766 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 767 * @scmd: Original SCSI cmd that eh has finished. 768 * @done_q: Queue for processed commands. 769 * 770 * Notes: 771 * We don't want to use the normal command completion while we are are 772 * still handling errors - it may cause other commands to be queued, 773 * and that would disturb what we are doing. thus we really want to 774 * keep a list of pending commands for final completion, and once we 775 * are ready to leave error handling we handle completion for real. 776 **/ 777 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 778 { 779 scmd->device->host->host_failed--; 780 scmd->eh_eflags = 0; 781 list_move_tail(&scmd->eh_entry, done_q); 782 } 783 EXPORT_SYMBOL(scsi_eh_finish_cmd); 784 785 /** 786 * scsi_eh_get_sense - Get device sense data. 787 * @work_q: Queue of commands to process. 788 * @done_q: Queue of proccessed commands.. 789 * 790 * Description: 791 * See if we need to request sense information. if so, then get it 792 * now, so we have a better idea of what to do. 793 * 794 * Notes: 795 * This has the unfortunate side effect that if a shost adapter does 796 * not automatically request sense information, that we end up shutting 797 * it down before we request it. 798 * 799 * All drivers should request sense information internally these days, 800 * so for now all I have to say is tough noogies if you end up in here. 801 * 802 * XXX: Long term this code should go away, but that needs an audit of 803 * all LLDDs first. 804 **/ 805 int scsi_eh_get_sense(struct list_head *work_q, 806 struct list_head *done_q) 807 { 808 struct scsi_cmnd *scmd, *next; 809 int rtn; 810 811 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 812 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) || 813 SCSI_SENSE_VALID(scmd)) 814 continue; 815 816 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 817 "%s: requesting sense\n", 818 current->comm)); 819 rtn = scsi_request_sense(scmd); 820 if (rtn != SUCCESS) 821 continue; 822 823 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p" 824 " result %x\n", scmd, 825 scmd->result)); 826 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd)); 827 828 rtn = scsi_decide_disposition(scmd); 829 830 /* 831 * if the result was normal, then just pass it along to the 832 * upper level. 833 */ 834 if (rtn == SUCCESS) 835 /* we don't want this command reissued, just 836 * finished with the sense data, so set 837 * retries to the max allowed to ensure it 838 * won't get reissued */ 839 scmd->retries = scmd->allowed; 840 else if (rtn != NEEDS_RETRY) 841 continue; 842 843 scsi_eh_finish_cmd(scmd, done_q); 844 } 845 846 return list_empty(work_q); 847 } 848 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 849 850 /** 851 * scsi_eh_tur - Send TUR to device. 852 * @scmd: Scsi cmd to send TUR 853 * 854 * Return value: 855 * 0 - Device is ready. 1 - Device NOT ready. 856 **/ 857 static int scsi_eh_tur(struct scsi_cmnd *scmd) 858 { 859 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 860 int retry_cnt = 1, rtn; 861 862 retry_tur: 863 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0); 864 865 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", 866 __FUNCTION__, scmd, rtn)); 867 868 switch (rtn) { 869 case NEEDS_RETRY: 870 if (retry_cnt--) 871 goto retry_tur; 872 /*FALLTHRU*/ 873 case SUCCESS: 874 return 0; 875 default: 876 return 1; 877 } 878 } 879 880 /** 881 * scsi_eh_abort_cmds - abort canceled commands. 882 * @shost: scsi host being recovered. 883 * @eh_done_q: list_head for processed commands. 884 * 885 * Decription: 886 * Try and see whether or not it makes sense to try and abort the 887 * running command. this only works out to be the case if we have one 888 * command that has timed out. if the command simply failed, it makes 889 * no sense to try and abort the command, since as far as the shost 890 * adapter is concerned, it isn't running. 891 **/ 892 static int scsi_eh_abort_cmds(struct list_head *work_q, 893 struct list_head *done_q) 894 { 895 struct scsi_cmnd *scmd, *next; 896 int rtn; 897 898 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 899 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD)) 900 continue; 901 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:" 902 "0x%p\n", current->comm, 903 scmd)); 904 rtn = scsi_try_to_abort_cmd(scmd); 905 if (rtn == SUCCESS) { 906 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD; 907 if (!scsi_device_online(scmd->device) || 908 !scsi_eh_tur(scmd)) { 909 scsi_eh_finish_cmd(scmd, done_q); 910 } 911 912 } else 913 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting" 914 " cmd failed:" 915 "0x%p\n", 916 current->comm, 917 scmd)); 918 } 919 920 return list_empty(work_q); 921 } 922 923 /** 924 * scsi_eh_try_stu - Send START_UNIT to device. 925 * @scmd: Scsi cmd to send START_UNIT 926 * 927 * Return value: 928 * 0 - Device is ready. 1 - Device NOT ready. 929 **/ 930 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 931 { 932 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 933 934 if (scmd->device->allow_restart) { 935 int i, rtn = NEEDS_RETRY; 936 937 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 938 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, 939 START_UNIT_TIMEOUT, 0); 940 941 if (rtn == SUCCESS) 942 return 0; 943 } 944 945 return 1; 946 } 947 948 /** 949 * scsi_eh_stu - send START_UNIT if needed 950 * @shost: scsi host being recovered. 951 * @eh_done_q: list_head for processed commands. 952 * 953 * Notes: 954 * If commands are failing due to not ready, initializing command required, 955 * try revalidating the device, which will end up sending a start unit. 956 **/ 957 static int scsi_eh_stu(struct Scsi_Host *shost, 958 struct list_head *work_q, 959 struct list_head *done_q) 960 { 961 struct scsi_cmnd *scmd, *stu_scmd, *next; 962 struct scsi_device *sdev; 963 964 shost_for_each_device(sdev, shost) { 965 stu_scmd = NULL; 966 list_for_each_entry(scmd, work_q, eh_entry) 967 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 968 scsi_check_sense(scmd) == FAILED ) { 969 stu_scmd = scmd; 970 break; 971 } 972 973 if (!stu_scmd) 974 continue; 975 976 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:" 977 " 0x%p\n", current->comm, sdev)); 978 979 if (!scsi_eh_try_stu(stu_scmd)) { 980 if (!scsi_device_online(sdev) || 981 !scsi_eh_tur(stu_scmd)) { 982 list_for_each_entry_safe(scmd, next, 983 work_q, eh_entry) { 984 if (scmd->device == sdev) 985 scsi_eh_finish_cmd(scmd, done_q); 986 } 987 } 988 } else { 989 SCSI_LOG_ERROR_RECOVERY(3, 990 printk("%s: START_UNIT failed to sdev:" 991 " 0x%p\n", current->comm, sdev)); 992 } 993 } 994 995 return list_empty(work_q); 996 } 997 998 999 /** 1000 * scsi_eh_bus_device_reset - send bdr if needed 1001 * @shost: scsi host being recovered. 1002 * @eh_done_q: list_head for processed commands. 1003 * 1004 * Notes: 1005 * Try a bus device reset. still, look to see whether we have multiple 1006 * devices that are jammed or not - if we have multiple devices, it 1007 * makes no sense to try bus_device_reset - we really would need to try 1008 * a bus_reset instead. 1009 **/ 1010 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1011 struct list_head *work_q, 1012 struct list_head *done_q) 1013 { 1014 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1015 struct scsi_device *sdev; 1016 int rtn; 1017 1018 shost_for_each_device(sdev, shost) { 1019 bdr_scmd = NULL; 1020 list_for_each_entry(scmd, work_q, eh_entry) 1021 if (scmd->device == sdev) { 1022 bdr_scmd = scmd; 1023 break; 1024 } 1025 1026 if (!bdr_scmd) 1027 continue; 1028 1029 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:" 1030 " 0x%p\n", current->comm, 1031 sdev)); 1032 rtn = scsi_try_bus_device_reset(bdr_scmd); 1033 if (rtn == SUCCESS) { 1034 if (!scsi_device_online(sdev) || 1035 !scsi_eh_tur(bdr_scmd)) { 1036 list_for_each_entry_safe(scmd, next, 1037 work_q, eh_entry) { 1038 if (scmd->device == sdev) 1039 scsi_eh_finish_cmd(scmd, 1040 done_q); 1041 } 1042 } 1043 } else { 1044 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR" 1045 " failed sdev:" 1046 "0x%p\n", 1047 current->comm, 1048 sdev)); 1049 } 1050 } 1051 1052 return list_empty(work_q); 1053 } 1054 1055 /** 1056 * scsi_eh_bus_reset - send a bus reset 1057 * @shost: scsi host being recovered. 1058 * @eh_done_q: list_head for processed commands. 1059 **/ 1060 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1061 struct list_head *work_q, 1062 struct list_head *done_q) 1063 { 1064 struct scsi_cmnd *scmd, *chan_scmd, *next; 1065 unsigned int channel; 1066 int rtn; 1067 1068 /* 1069 * we really want to loop over the various channels, and do this on 1070 * a channel by channel basis. we should also check to see if any 1071 * of the failed commands are on soft_reset devices, and if so, skip 1072 * the reset. 1073 */ 1074 1075 for (channel = 0; channel <= shost->max_channel; channel++) { 1076 chan_scmd = NULL; 1077 list_for_each_entry(scmd, work_q, eh_entry) { 1078 if (channel == scmd_channel(scmd)) { 1079 chan_scmd = scmd; 1080 break; 1081 /* 1082 * FIXME add back in some support for 1083 * soft_reset devices. 1084 */ 1085 } 1086 } 1087 1088 if (!chan_scmd) 1089 continue; 1090 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:" 1091 " %d\n", current->comm, 1092 channel)); 1093 rtn = scsi_try_bus_reset(chan_scmd); 1094 if (rtn == SUCCESS) { 1095 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1096 if (channel == scmd_channel(scmd)) 1097 if (!scsi_device_online(scmd->device) || 1098 !scsi_eh_tur(scmd)) 1099 scsi_eh_finish_cmd(scmd, 1100 done_q); 1101 } 1102 } else { 1103 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST" 1104 " failed chan: %d\n", 1105 current->comm, 1106 channel)); 1107 } 1108 } 1109 return list_empty(work_q); 1110 } 1111 1112 /** 1113 * scsi_eh_host_reset - send a host reset 1114 * @work_q: list_head for processed commands. 1115 * @done_q: list_head for processed commands. 1116 **/ 1117 static int scsi_eh_host_reset(struct list_head *work_q, 1118 struct list_head *done_q) 1119 { 1120 struct scsi_cmnd *scmd, *next; 1121 int rtn; 1122 1123 if (!list_empty(work_q)) { 1124 scmd = list_entry(work_q->next, 1125 struct scsi_cmnd, eh_entry); 1126 1127 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n" 1128 , current->comm)); 1129 1130 rtn = scsi_try_host_reset(scmd); 1131 if (rtn == SUCCESS) { 1132 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1133 if (!scsi_device_online(scmd->device) || 1134 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) || 1135 !scsi_eh_tur(scmd)) 1136 scsi_eh_finish_cmd(scmd, done_q); 1137 } 1138 } else { 1139 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST" 1140 " failed\n", 1141 current->comm)); 1142 } 1143 } 1144 return list_empty(work_q); 1145 } 1146 1147 /** 1148 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1149 * @work_q: list_head for processed commands. 1150 * @done_q: list_head for processed commands. 1151 * 1152 **/ 1153 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1154 struct list_head *done_q) 1155 { 1156 struct scsi_cmnd *scmd, *next; 1157 1158 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1159 sdev_printk(KERN_INFO, scmd->device, 1160 "scsi: Device offlined - not" 1161 " ready after error recovery\n"); 1162 scsi_device_set_state(scmd->device, SDEV_OFFLINE); 1163 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) { 1164 /* 1165 * FIXME: Handle lost cmds. 1166 */ 1167 } 1168 scsi_eh_finish_cmd(scmd, done_q); 1169 } 1170 return; 1171 } 1172 1173 /** 1174 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1175 * @scmd: SCSI cmd to examine. 1176 * 1177 * Notes: 1178 * This is *only* called when we are examining the status after sending 1179 * out the actual data command. any commands that are queued for error 1180 * recovery (e.g. test_unit_ready) do *not* come through here. 1181 * 1182 * When this routine returns failed, it means the error handler thread 1183 * is woken. In cases where the error code indicates an error that 1184 * doesn't require the error handler read (i.e. we don't need to 1185 * abort/reset), this function should return SUCCESS. 1186 **/ 1187 int scsi_decide_disposition(struct scsi_cmnd *scmd) 1188 { 1189 int rtn; 1190 1191 /* 1192 * if the device is offline, then we clearly just pass the result back 1193 * up to the top level. 1194 */ 1195 if (!scsi_device_online(scmd->device)) { 1196 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report" 1197 " as SUCCESS\n", 1198 __FUNCTION__)); 1199 return SUCCESS; 1200 } 1201 1202 /* 1203 * first check the host byte, to see if there is anything in there 1204 * that would indicate what we need to do. 1205 */ 1206 switch (host_byte(scmd->result)) { 1207 case DID_PASSTHROUGH: 1208 /* 1209 * no matter what, pass this through to the upper layer. 1210 * nuke this special code so that it looks like we are saying 1211 * did_ok. 1212 */ 1213 scmd->result &= 0xff00ffff; 1214 return SUCCESS; 1215 case DID_OK: 1216 /* 1217 * looks good. drop through, and check the next byte. 1218 */ 1219 break; 1220 case DID_NO_CONNECT: 1221 case DID_BAD_TARGET: 1222 case DID_ABORT: 1223 /* 1224 * note - this means that we just report the status back 1225 * to the top level driver, not that we actually think 1226 * that it indicates SUCCESS. 1227 */ 1228 return SUCCESS; 1229 /* 1230 * when the low level driver returns did_soft_error, 1231 * it is responsible for keeping an internal retry counter 1232 * in order to avoid endless loops (db) 1233 * 1234 * actually this is a bug in this function here. we should 1235 * be mindful of the maximum number of retries specified 1236 * and not get stuck in a loop. 1237 */ 1238 case DID_SOFT_ERROR: 1239 goto maybe_retry; 1240 case DID_IMM_RETRY: 1241 return NEEDS_RETRY; 1242 1243 case DID_REQUEUE: 1244 return ADD_TO_MLQUEUE; 1245 1246 case DID_ERROR: 1247 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1248 status_byte(scmd->result) == RESERVATION_CONFLICT) 1249 /* 1250 * execute reservation conflict processing code 1251 * lower down 1252 */ 1253 break; 1254 /* fallthrough */ 1255 1256 case DID_BUS_BUSY: 1257 case DID_PARITY: 1258 goto maybe_retry; 1259 case DID_TIME_OUT: 1260 /* 1261 * when we scan the bus, we get timeout messages for 1262 * these commands if there is no device available. 1263 * other hosts report did_no_connect for the same thing. 1264 */ 1265 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1266 scmd->cmnd[0] == INQUIRY)) { 1267 return SUCCESS; 1268 } else { 1269 return FAILED; 1270 } 1271 case DID_RESET: 1272 return SUCCESS; 1273 default: 1274 return FAILED; 1275 } 1276 1277 /* 1278 * next, check the message byte. 1279 */ 1280 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 1281 return FAILED; 1282 1283 /* 1284 * check the status byte to see if this indicates anything special. 1285 */ 1286 switch (status_byte(scmd->result)) { 1287 case QUEUE_FULL: 1288 /* 1289 * the case of trying to send too many commands to a 1290 * tagged queueing device. 1291 */ 1292 case BUSY: 1293 /* 1294 * device can't talk to us at the moment. Should only 1295 * occur (SAM-3) when the task queue is empty, so will cause 1296 * the empty queue handling to trigger a stall in the 1297 * device. 1298 */ 1299 return ADD_TO_MLQUEUE; 1300 case GOOD: 1301 case COMMAND_TERMINATED: 1302 case TASK_ABORTED: 1303 return SUCCESS; 1304 case CHECK_CONDITION: 1305 rtn = scsi_check_sense(scmd); 1306 if (rtn == NEEDS_RETRY) 1307 goto maybe_retry; 1308 /* if rtn == FAILED, we have no sense information; 1309 * returning FAILED will wake the error handler thread 1310 * to collect the sense and redo the decide 1311 * disposition */ 1312 return rtn; 1313 case CONDITION_GOOD: 1314 case INTERMEDIATE_GOOD: 1315 case INTERMEDIATE_C_GOOD: 1316 case ACA_ACTIVE: 1317 /* 1318 * who knows? FIXME(eric) 1319 */ 1320 return SUCCESS; 1321 1322 case RESERVATION_CONFLICT: 1323 sdev_printk(KERN_INFO, scmd->device, 1324 "reservation conflict\n"); 1325 return SUCCESS; /* causes immediate i/o error */ 1326 default: 1327 return FAILED; 1328 } 1329 return FAILED; 1330 1331 maybe_retry: 1332 1333 /* we requeue for retry because the error was retryable, and 1334 * the request was not marked fast fail. Note that above, 1335 * even if the request is marked fast fail, we still requeue 1336 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1337 if ((++scmd->retries) <= scmd->allowed 1338 && !blk_noretry_request(scmd->request)) { 1339 return NEEDS_RETRY; 1340 } else { 1341 /* 1342 * no more retries - report this one back to upper level. 1343 */ 1344 return SUCCESS; 1345 } 1346 } 1347 1348 /** 1349 * scsi_eh_lock_door - Prevent medium removal for the specified device 1350 * @sdev: SCSI device to prevent medium removal 1351 * 1352 * Locking: 1353 * We must be called from process context; scsi_allocate_request() 1354 * may sleep. 1355 * 1356 * Notes: 1357 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 1358 * head of the devices request queue, and continue. 1359 * 1360 * Bugs: 1361 * scsi_allocate_request() may sleep waiting for existing requests to 1362 * be processed. However, since we haven't kicked off any request 1363 * processing for this host, this may deadlock. 1364 * 1365 * If scsi_allocate_request() fails for what ever reason, we 1366 * completely forget to lock the door. 1367 **/ 1368 static void scsi_eh_lock_door(struct scsi_device *sdev) 1369 { 1370 unsigned char cmnd[MAX_COMMAND_SIZE]; 1371 1372 cmnd[0] = ALLOW_MEDIUM_REMOVAL; 1373 cmnd[1] = 0; 1374 cmnd[2] = 0; 1375 cmnd[3] = 0; 1376 cmnd[4] = SCSI_REMOVAL_PREVENT; 1377 cmnd[5] = 0; 1378 1379 scsi_execute_async(sdev, cmnd, 6, DMA_NONE, NULL, 0, 0, 10 * HZ, 1380 5, NULL, NULL, GFP_KERNEL); 1381 } 1382 1383 1384 /** 1385 * scsi_restart_operations - restart io operations to the specified host. 1386 * @shost: Host we are restarting. 1387 * 1388 * Notes: 1389 * When we entered the error handler, we blocked all further i/o to 1390 * this device. we need to 'reverse' this process. 1391 **/ 1392 static void scsi_restart_operations(struct Scsi_Host *shost) 1393 { 1394 struct scsi_device *sdev; 1395 unsigned long flags; 1396 1397 /* 1398 * If the door was locked, we need to insert a door lock request 1399 * onto the head of the SCSI request queue for the device. There 1400 * is no point trying to lock the door of an off-line device. 1401 */ 1402 shost_for_each_device(sdev, shost) { 1403 if (scsi_device_online(sdev) && sdev->locked) 1404 scsi_eh_lock_door(sdev); 1405 } 1406 1407 /* 1408 * next free up anything directly waiting upon the host. this 1409 * will be requests for character device operations, and also for 1410 * ioctls to queued block devices. 1411 */ 1412 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n", 1413 __FUNCTION__)); 1414 1415 spin_lock_irqsave(shost->host_lock, flags); 1416 if (scsi_host_set_state(shost, SHOST_RUNNING)) 1417 if (scsi_host_set_state(shost, SHOST_CANCEL)) 1418 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 1419 spin_unlock_irqrestore(shost->host_lock, flags); 1420 1421 wake_up(&shost->host_wait); 1422 1423 /* 1424 * finally we need to re-initiate requests that may be pending. we will 1425 * have had everything blocked while error handling is taking place, and 1426 * now that error recovery is done, we will need to ensure that these 1427 * requests are started. 1428 */ 1429 scsi_run_host_queues(shost); 1430 } 1431 1432 /** 1433 * scsi_eh_ready_devs - check device ready state and recover if not. 1434 * @shost: host to be recovered. 1435 * @eh_done_q: list_head for processed commands. 1436 * 1437 **/ 1438 void scsi_eh_ready_devs(struct Scsi_Host *shost, 1439 struct list_head *work_q, 1440 struct list_head *done_q) 1441 { 1442 if (!scsi_eh_stu(shost, work_q, done_q)) 1443 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 1444 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 1445 if (!scsi_eh_host_reset(work_q, done_q)) 1446 scsi_eh_offline_sdevs(work_q, done_q); 1447 } 1448 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 1449 1450 /** 1451 * scsi_eh_flush_done_q - finish processed commands or retry them. 1452 * @done_q: list_head of processed commands. 1453 * 1454 **/ 1455 void scsi_eh_flush_done_q(struct list_head *done_q) 1456 { 1457 struct scsi_cmnd *scmd, *next; 1458 1459 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 1460 list_del_init(&scmd->eh_entry); 1461 if (scsi_device_online(scmd->device) && 1462 !blk_noretry_request(scmd->request) && 1463 (++scmd->retries <= scmd->allowed)) { 1464 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush" 1465 " retry cmd: %p\n", 1466 current->comm, 1467 scmd)); 1468 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 1469 } else { 1470 /* 1471 * If just we got sense for the device (called 1472 * scsi_eh_get_sense), scmd->result is already 1473 * set, do not set DRIVER_TIMEOUT. 1474 */ 1475 if (!scmd->result) 1476 scmd->result |= (DRIVER_TIMEOUT << 24); 1477 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish" 1478 " cmd: %p\n", 1479 current->comm, scmd)); 1480 scsi_finish_command(scmd); 1481 } 1482 } 1483 } 1484 EXPORT_SYMBOL(scsi_eh_flush_done_q); 1485 1486 /** 1487 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 1488 * @shost: Host to unjam. 1489 * 1490 * Notes: 1491 * When we come in here, we *know* that all commands on the bus have 1492 * either completed, failed or timed out. we also know that no further 1493 * commands are being sent to the host, so things are relatively quiet 1494 * and we have freedom to fiddle with things as we wish. 1495 * 1496 * This is only the *default* implementation. it is possible for 1497 * individual drivers to supply their own version of this function, and 1498 * if the maintainer wishes to do this, it is strongly suggested that 1499 * this function be taken as a template and modified. this function 1500 * was designed to correctly handle problems for about 95% of the 1501 * different cases out there, and it should always provide at least a 1502 * reasonable amount of error recovery. 1503 * 1504 * Any command marked 'failed' or 'timeout' must eventually have 1505 * scsi_finish_cmd() called for it. we do all of the retry stuff 1506 * here, so when we restart the host after we return it should have an 1507 * empty queue. 1508 **/ 1509 static void scsi_unjam_host(struct Scsi_Host *shost) 1510 { 1511 unsigned long flags; 1512 LIST_HEAD(eh_work_q); 1513 LIST_HEAD(eh_done_q); 1514 1515 spin_lock_irqsave(shost->host_lock, flags); 1516 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 1517 spin_unlock_irqrestore(shost->host_lock, flags); 1518 1519 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 1520 1521 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 1522 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q)) 1523 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 1524 1525 scsi_eh_flush_done_q(&eh_done_q); 1526 } 1527 1528 /** 1529 * scsi_error_handler - SCSI error handler thread 1530 * @data: Host for which we are running. 1531 * 1532 * Notes: 1533 * This is the main error handling loop. This is run as a kernel thread 1534 * for every SCSI host and handles all error handling activity. 1535 **/ 1536 int scsi_error_handler(void *data) 1537 { 1538 struct Scsi_Host *shost = data; 1539 1540 current->flags |= PF_NOFREEZE; 1541 1542 /* 1543 * We use TASK_INTERRUPTIBLE so that the thread is not 1544 * counted against the load average as a running process. 1545 * We never actually get interrupted because kthread_run 1546 * disables singal delivery for the created thread. 1547 */ 1548 set_current_state(TASK_INTERRUPTIBLE); 1549 while (!kthread_should_stop()) { 1550 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 1551 shost->host_failed != shost->host_busy) { 1552 SCSI_LOG_ERROR_RECOVERY(1, 1553 printk("Error handler scsi_eh_%d sleeping\n", 1554 shost->host_no)); 1555 schedule(); 1556 set_current_state(TASK_INTERRUPTIBLE); 1557 continue; 1558 } 1559 1560 __set_current_state(TASK_RUNNING); 1561 SCSI_LOG_ERROR_RECOVERY(1, 1562 printk("Error handler scsi_eh_%d waking up\n", 1563 shost->host_no)); 1564 1565 /* 1566 * We have a host that is failing for some reason. Figure out 1567 * what we need to do to get it up and online again (if we can). 1568 * If we fail, we end up taking the thing offline. 1569 */ 1570 if (shost->transportt->eh_strategy_handler) 1571 shost->transportt->eh_strategy_handler(shost); 1572 else 1573 scsi_unjam_host(shost); 1574 1575 /* 1576 * Note - if the above fails completely, the action is to take 1577 * individual devices offline and flush the queue of any 1578 * outstanding requests that may have been pending. When we 1579 * restart, we restart any I/O to any other devices on the bus 1580 * which are still online. 1581 */ 1582 scsi_restart_operations(shost); 1583 set_current_state(TASK_INTERRUPTIBLE); 1584 } 1585 __set_current_state(TASK_RUNNING); 1586 1587 SCSI_LOG_ERROR_RECOVERY(1, 1588 printk("Error handler scsi_eh_%d exiting\n", shost->host_no)); 1589 shost->ehandler = NULL; 1590 return 0; 1591 } 1592 1593 /* 1594 * Function: scsi_report_bus_reset() 1595 * 1596 * Purpose: Utility function used by low-level drivers to report that 1597 * they have observed a bus reset on the bus being handled. 1598 * 1599 * Arguments: shost - Host in question 1600 * channel - channel on which reset was observed. 1601 * 1602 * Returns: Nothing 1603 * 1604 * Lock status: Host lock must be held. 1605 * 1606 * Notes: This only needs to be called if the reset is one which 1607 * originates from an unknown location. Resets originated 1608 * by the mid-level itself don't need to call this, but there 1609 * should be no harm. 1610 * 1611 * The main purpose of this is to make sure that a CHECK_CONDITION 1612 * is properly treated. 1613 */ 1614 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 1615 { 1616 struct scsi_device *sdev; 1617 1618 __shost_for_each_device(sdev, shost) { 1619 if (channel == sdev_channel(sdev)) { 1620 sdev->was_reset = 1; 1621 sdev->expecting_cc_ua = 1; 1622 } 1623 } 1624 } 1625 EXPORT_SYMBOL(scsi_report_bus_reset); 1626 1627 /* 1628 * Function: scsi_report_device_reset() 1629 * 1630 * Purpose: Utility function used by low-level drivers to report that 1631 * they have observed a device reset on the device being handled. 1632 * 1633 * Arguments: shost - Host in question 1634 * channel - channel on which reset was observed 1635 * target - target on which reset was observed 1636 * 1637 * Returns: Nothing 1638 * 1639 * Lock status: Host lock must be held 1640 * 1641 * Notes: This only needs to be called if the reset is one which 1642 * originates from an unknown location. Resets originated 1643 * by the mid-level itself don't need to call this, but there 1644 * should be no harm. 1645 * 1646 * The main purpose of this is to make sure that a CHECK_CONDITION 1647 * is properly treated. 1648 */ 1649 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 1650 { 1651 struct scsi_device *sdev; 1652 1653 __shost_for_each_device(sdev, shost) { 1654 if (channel == sdev_channel(sdev) && 1655 target == sdev_id(sdev)) { 1656 sdev->was_reset = 1; 1657 sdev->expecting_cc_ua = 1; 1658 } 1659 } 1660 } 1661 EXPORT_SYMBOL(scsi_report_device_reset); 1662 1663 static void 1664 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 1665 { 1666 } 1667 1668 /* 1669 * Function: scsi_reset_provider 1670 * 1671 * Purpose: Send requested reset to a bus or device at any phase. 1672 * 1673 * Arguments: device - device to send reset to 1674 * flag - reset type (see scsi.h) 1675 * 1676 * Returns: SUCCESS/FAILURE. 1677 * 1678 * Notes: This is used by the SCSI Generic driver to provide 1679 * Bus/Device reset capability. 1680 */ 1681 int 1682 scsi_reset_provider(struct scsi_device *dev, int flag) 1683 { 1684 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL); 1685 struct Scsi_Host *shost = dev->host; 1686 struct request req; 1687 unsigned long flags; 1688 int rtn; 1689 1690 scmd->request = &req; 1691 memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout)); 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->request_buffer = NULL; 1698 scmd->request_bufflen = 0; 1699 1700 scmd->cmd_len = 0; 1701 1702 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 1703 1704 init_timer(&scmd->eh_timeout); 1705 1706 /* 1707 * Sometimes the command can get back into the timer chain, 1708 * so use the pid as an identifier. 1709 */ 1710 scmd->pid = 0; 1711 1712 spin_lock_irqsave(shost->host_lock, flags); 1713 shost->tmf_in_progress = 1; 1714 spin_unlock_irqrestore(shost->host_lock, flags); 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 spin_lock_irqsave(shost->host_lock, flags); 1735 shost->tmf_in_progress = 0; 1736 spin_unlock_irqrestore(shost->host_lock, flags); 1737 1738 /* 1739 * be sure to wake up anyone who was sleeping or had their queue 1740 * suspended while we performed the TMF. 1741 */ 1742 SCSI_LOG_ERROR_RECOVERY(3, 1743 printk("%s: waking up host to restart after TMF\n", 1744 __FUNCTION__)); 1745 1746 wake_up(&shost->host_wait); 1747 1748 scsi_run_host_queues(shost); 1749 1750 scsi_next_command(scmd); 1751 return rtn; 1752 } 1753 EXPORT_SYMBOL(scsi_reset_provider); 1754 1755 /** 1756 * scsi_normalize_sense - normalize main elements from either fixed or 1757 * descriptor sense data format into a common format. 1758 * 1759 * @sense_buffer: byte array containing sense data returned by device 1760 * @sb_len: number of valid bytes in sense_buffer 1761 * @sshdr: pointer to instance of structure that common 1762 * elements are written to. 1763 * 1764 * Notes: 1765 * The "main elements" from sense data are: response_code, sense_key, 1766 * asc, ascq and additional_length (only for descriptor format). 1767 * 1768 * Typically this function can be called after a device has 1769 * responded to a SCSI command with the CHECK_CONDITION status. 1770 * 1771 * Return value: 1772 * 1 if valid sense data information found, else 0; 1773 **/ 1774 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len, 1775 struct scsi_sense_hdr *sshdr) 1776 { 1777 if (!sense_buffer || !sb_len) 1778 return 0; 1779 1780 memset(sshdr, 0, sizeof(struct scsi_sense_hdr)); 1781 1782 sshdr->response_code = (sense_buffer[0] & 0x7f); 1783 1784 if (!scsi_sense_valid(sshdr)) 1785 return 0; 1786 1787 if (sshdr->response_code >= 0x72) { 1788 /* 1789 * descriptor format 1790 */ 1791 if (sb_len > 1) 1792 sshdr->sense_key = (sense_buffer[1] & 0xf); 1793 if (sb_len > 2) 1794 sshdr->asc = sense_buffer[2]; 1795 if (sb_len > 3) 1796 sshdr->ascq = sense_buffer[3]; 1797 if (sb_len > 7) 1798 sshdr->additional_length = sense_buffer[7]; 1799 } else { 1800 /* 1801 * fixed format 1802 */ 1803 if (sb_len > 2) 1804 sshdr->sense_key = (sense_buffer[2] & 0xf); 1805 if (sb_len > 7) { 1806 sb_len = (sb_len < (sense_buffer[7] + 8)) ? 1807 sb_len : (sense_buffer[7] + 8); 1808 if (sb_len > 12) 1809 sshdr->asc = sense_buffer[12]; 1810 if (sb_len > 13) 1811 sshdr->ascq = sense_buffer[13]; 1812 } 1813 } 1814 1815 return 1; 1816 } 1817 EXPORT_SYMBOL(scsi_normalize_sense); 1818 1819 int scsi_command_normalize_sense(struct scsi_cmnd *cmd, 1820 struct scsi_sense_hdr *sshdr) 1821 { 1822 return scsi_normalize_sense(cmd->sense_buffer, 1823 sizeof(cmd->sense_buffer), sshdr); 1824 } 1825 EXPORT_SYMBOL(scsi_command_normalize_sense); 1826 1827 /** 1828 * scsi_sense_desc_find - search for a given descriptor type in 1829 * descriptor sense data format. 1830 * 1831 * @sense_buffer: byte array of descriptor format sense data 1832 * @sb_len: number of valid bytes in sense_buffer 1833 * @desc_type: value of descriptor type to find 1834 * (e.g. 0 -> information) 1835 * 1836 * Notes: 1837 * only valid when sense data is in descriptor format 1838 * 1839 * Return value: 1840 * pointer to start of (first) descriptor if found else NULL 1841 **/ 1842 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len, 1843 int desc_type) 1844 { 1845 int add_sen_len, add_len, desc_len, k; 1846 const u8 * descp; 1847 1848 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7]))) 1849 return NULL; 1850 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73)) 1851 return NULL; 1852 add_sen_len = (add_sen_len < (sb_len - 8)) ? 1853 add_sen_len : (sb_len - 8); 1854 descp = &sense_buffer[8]; 1855 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) { 1856 descp += desc_len; 1857 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1; 1858 desc_len = add_len + 2; 1859 if (descp[0] == desc_type) 1860 return descp; 1861 if (add_len < 0) // short descriptor ?? 1862 break; 1863 } 1864 return NULL; 1865 } 1866 EXPORT_SYMBOL(scsi_sense_desc_find); 1867 1868 /** 1869 * scsi_get_sense_info_fld - attempts to get information field from 1870 * sense data (either fixed or descriptor format) 1871 * 1872 * @sense_buffer: byte array of sense data 1873 * @sb_len: number of valid bytes in sense_buffer 1874 * @info_out: pointer to 64 integer where 8 or 4 byte information 1875 * field will be placed if found. 1876 * 1877 * Return value: 1878 * 1 if information field found, 0 if not found. 1879 **/ 1880 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len, 1881 u64 * info_out) 1882 { 1883 int j; 1884 const u8 * ucp; 1885 u64 ull; 1886 1887 if (sb_len < 7) 1888 return 0; 1889 switch (sense_buffer[0] & 0x7f) { 1890 case 0x70: 1891 case 0x71: 1892 if (sense_buffer[0] & 0x80) { 1893 *info_out = (sense_buffer[3] << 24) + 1894 (sense_buffer[4] << 16) + 1895 (sense_buffer[5] << 8) + sense_buffer[6]; 1896 return 1; 1897 } else 1898 return 0; 1899 case 0x72: 1900 case 0x73: 1901 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 1902 0 /* info desc */); 1903 if (ucp && (0xa == ucp[1])) { 1904 ull = 0; 1905 for (j = 0; j < 8; ++j) { 1906 if (j > 0) 1907 ull <<= 8; 1908 ull |= ucp[4 + j]; 1909 } 1910 *info_out = ull; 1911 return 1; 1912 } else 1913 return 0; 1914 default: 1915 return 0; 1916 } 1917 } 1918 EXPORT_SYMBOL(scsi_get_sense_info_fld); 1919