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/kernel.h> 22 #include <linux/freezer.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 #include "scsi_transport_api.h" 40 41 #define SENSE_TIMEOUT (10*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_times_out - Timeout function for normal scsi commands. 116 * @req: request that is timing out. 117 * 118 * Notes: 119 * We do not need to lock this. There is the potential for a race 120 * only in that the normal completion handling might run, but if the 121 * normal completion function determines that the timer has already 122 * fired, then it mustn't do anything. 123 */ 124 enum blk_eh_timer_return scsi_times_out(struct request *req) 125 { 126 struct scsi_cmnd *scmd = req->special; 127 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED; 128 129 scsi_log_completion(scmd, TIMEOUT_ERROR); 130 131 if (scmd->device->host->transportt->eh_timed_out) 132 rtn = scmd->device->host->transportt->eh_timed_out(scmd); 133 else if (scmd->device->host->hostt->eh_timed_out) 134 rtn = scmd->device->host->hostt->eh_timed_out(scmd); 135 136 if (unlikely(rtn == BLK_EH_NOT_HANDLED && 137 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) { 138 scmd->result |= DID_TIME_OUT << 16; 139 rtn = BLK_EH_HANDLED; 140 } 141 142 return rtn; 143 } 144 145 /** 146 * scsi_block_when_processing_errors - Prevent cmds from being queued. 147 * @sdev: Device on which we are performing recovery. 148 * 149 * Description: 150 * We block until the host is out of error recovery, and then check to 151 * see whether the host or the device is offline. 152 * 153 * Return value: 154 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 155 */ 156 int scsi_block_when_processing_errors(struct scsi_device *sdev) 157 { 158 int online; 159 160 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 161 162 online = scsi_device_online(sdev); 163 164 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__, 165 online)); 166 167 return online; 168 } 169 EXPORT_SYMBOL(scsi_block_when_processing_errors); 170 171 #ifdef CONFIG_SCSI_LOGGING 172 /** 173 * scsi_eh_prt_fail_stats - Log info on failures. 174 * @shost: scsi host being recovered. 175 * @work_q: Queue of scsi cmds to process. 176 */ 177 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 178 struct list_head *work_q) 179 { 180 struct scsi_cmnd *scmd; 181 struct scsi_device *sdev; 182 int total_failures = 0; 183 int cmd_failed = 0; 184 int cmd_cancel = 0; 185 int devices_failed = 0; 186 187 shost_for_each_device(sdev, shost) { 188 list_for_each_entry(scmd, work_q, eh_entry) { 189 if (scmd->device == sdev) { 190 ++total_failures; 191 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) 192 ++cmd_cancel; 193 else 194 ++cmd_failed; 195 } 196 } 197 198 if (cmd_cancel || cmd_failed) { 199 SCSI_LOG_ERROR_RECOVERY(3, 200 sdev_printk(KERN_INFO, sdev, 201 "%s: cmds failed: %d, cancel: %d\n", 202 __func__, cmd_failed, 203 cmd_cancel)); 204 cmd_cancel = 0; 205 cmd_failed = 0; 206 ++devices_failed; 207 } 208 } 209 210 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d" 211 " devices require eh work\n", 212 total_failures, devices_failed)); 213 } 214 #endif 215 216 /** 217 * scsi_check_sense - Examine scsi cmd sense 218 * @scmd: Cmd to have sense checked. 219 * 220 * Return value: 221 * SUCCESS or FAILED or NEEDS_RETRY 222 * 223 * Notes: 224 * When a deferred error is detected the current command has 225 * not been executed and needs retrying. 226 */ 227 static int scsi_check_sense(struct scsi_cmnd *scmd) 228 { 229 struct scsi_device *sdev = scmd->device; 230 struct scsi_sense_hdr sshdr; 231 232 if (! scsi_command_normalize_sense(scmd, &sshdr)) 233 return FAILED; /* no valid sense data */ 234 235 if (scsi_sense_is_deferred(&sshdr)) 236 return NEEDS_RETRY; 237 238 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh && 239 sdev->scsi_dh_data->scsi_dh->check_sense) { 240 int rc; 241 242 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr); 243 if (rc != SCSI_RETURN_NOT_HANDLED) 244 return rc; 245 /* handler does not care. Drop down to default handling */ 246 } 247 248 /* 249 * Previous logic looked for FILEMARK, EOM or ILI which are 250 * mainly associated with tapes and returned SUCCESS. 251 */ 252 if (sshdr.response_code == 0x70) { 253 /* fixed format */ 254 if (scmd->sense_buffer[2] & 0xe0) 255 return SUCCESS; 256 } else { 257 /* 258 * descriptor format: look for "stream commands sense data 259 * descriptor" (see SSC-3). Assume single sense data 260 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 261 */ 262 if ((sshdr.additional_length > 3) && 263 (scmd->sense_buffer[8] == 0x4) && 264 (scmd->sense_buffer[11] & 0xe0)) 265 return SUCCESS; 266 } 267 268 switch (sshdr.sense_key) { 269 case NO_SENSE: 270 return SUCCESS; 271 case RECOVERED_ERROR: 272 return /* soft_error */ SUCCESS; 273 274 case ABORTED_COMMAND: 275 if (sshdr.asc == 0x10) /* DIF */ 276 return SUCCESS; 277 278 return NEEDS_RETRY; 279 case NOT_READY: 280 case UNIT_ATTENTION: 281 /* 282 * if we are expecting a cc/ua because of a bus reset that we 283 * performed, treat this just as a retry. otherwise this is 284 * information that we should pass up to the upper-level driver 285 * so that we can deal with it there. 286 */ 287 if (scmd->device->expecting_cc_ua) { 288 scmd->device->expecting_cc_ua = 0; 289 return NEEDS_RETRY; 290 } 291 /* 292 * if the device is in the process of becoming ready, we 293 * should retry. 294 */ 295 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 296 return NEEDS_RETRY; 297 /* 298 * if the device is not started, we need to wake 299 * the error handler to start the motor 300 */ 301 if (scmd->device->allow_restart && 302 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 303 return FAILED; 304 return SUCCESS; 305 306 /* these three are not supported */ 307 case COPY_ABORTED: 308 case VOLUME_OVERFLOW: 309 case MISCOMPARE: 310 return SUCCESS; 311 312 case MEDIUM_ERROR: 313 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 314 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 315 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 316 return SUCCESS; 317 } 318 return NEEDS_RETRY; 319 320 case HARDWARE_ERROR: 321 if (scmd->device->retry_hwerror) 322 return ADD_TO_MLQUEUE; 323 else 324 return SUCCESS; 325 326 case ILLEGAL_REQUEST: 327 case BLANK_CHECK: 328 case DATA_PROTECT: 329 default: 330 return SUCCESS; 331 } 332 } 333 334 /** 335 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 336 * @scmd: SCSI cmd to examine. 337 * 338 * Notes: 339 * This is *only* called when we are examining the status of commands 340 * queued during error recovery. the main difference here is that we 341 * don't allow for the possibility of retries here, and we are a lot 342 * more restrictive about what we consider acceptable. 343 */ 344 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) 345 { 346 /* 347 * first check the host byte, to see if there is anything in there 348 * that would indicate what we need to do. 349 */ 350 if (host_byte(scmd->result) == DID_RESET) { 351 /* 352 * rats. we are already in the error handler, so we now 353 * get to try and figure out what to do next. if the sense 354 * is valid, we have a pretty good idea of what to do. 355 * if not, we mark it as FAILED. 356 */ 357 return scsi_check_sense(scmd); 358 } 359 if (host_byte(scmd->result) != DID_OK) 360 return FAILED; 361 362 /* 363 * next, check the message byte. 364 */ 365 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 366 return FAILED; 367 368 /* 369 * now, check the status byte to see if this indicates 370 * anything special. 371 */ 372 switch (status_byte(scmd->result)) { 373 case GOOD: 374 case COMMAND_TERMINATED: 375 return SUCCESS; 376 case CHECK_CONDITION: 377 return scsi_check_sense(scmd); 378 case CONDITION_GOOD: 379 case INTERMEDIATE_GOOD: 380 case INTERMEDIATE_C_GOOD: 381 /* 382 * who knows? FIXME(eric) 383 */ 384 return SUCCESS; 385 case BUSY: 386 case QUEUE_FULL: 387 case RESERVATION_CONFLICT: 388 default: 389 return FAILED; 390 } 391 return FAILED; 392 } 393 394 /** 395 * scsi_eh_done - Completion function for error handling. 396 * @scmd: Cmd that is done. 397 */ 398 static void scsi_eh_done(struct scsi_cmnd *scmd) 399 { 400 struct completion *eh_action; 401 402 SCSI_LOG_ERROR_RECOVERY(3, 403 printk("%s scmd: %p result: %x\n", 404 __func__, scmd, scmd->result)); 405 406 eh_action = scmd->device->host->eh_action; 407 if (eh_action) 408 complete(eh_action); 409 } 410 411 /** 412 * scsi_try_host_reset - ask host adapter to reset itself 413 * @scmd: SCSI cmd to send hsot reset. 414 */ 415 static int scsi_try_host_reset(struct scsi_cmnd *scmd) 416 { 417 unsigned long flags; 418 int rtn; 419 420 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n", 421 __func__)); 422 423 if (!scmd->device->host->hostt->eh_host_reset_handler) 424 return FAILED; 425 426 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd); 427 428 if (rtn == SUCCESS) { 429 if (!scmd->device->host->hostt->skip_settle_delay) 430 ssleep(HOST_RESET_SETTLE_TIME); 431 spin_lock_irqsave(scmd->device->host->host_lock, flags); 432 scsi_report_bus_reset(scmd->device->host, 433 scmd_channel(scmd)); 434 spin_unlock_irqrestore(scmd->device->host->host_lock, flags); 435 } 436 437 return rtn; 438 } 439 440 /** 441 * scsi_try_bus_reset - ask host to perform a bus reset 442 * @scmd: SCSI cmd to send bus reset. 443 */ 444 static int scsi_try_bus_reset(struct scsi_cmnd *scmd) 445 { 446 unsigned long flags; 447 int rtn; 448 449 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n", 450 __func__)); 451 452 if (!scmd->device->host->hostt->eh_bus_reset_handler) 453 return FAILED; 454 455 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd); 456 457 if (rtn == SUCCESS) { 458 if (!scmd->device->host->hostt->skip_settle_delay) 459 ssleep(BUS_RESET_SETTLE_TIME); 460 spin_lock_irqsave(scmd->device->host->host_lock, flags); 461 scsi_report_bus_reset(scmd->device->host, 462 scmd_channel(scmd)); 463 spin_unlock_irqrestore(scmd->device->host->host_lock, flags); 464 } 465 466 return rtn; 467 } 468 469 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 470 { 471 sdev->was_reset = 1; 472 sdev->expecting_cc_ua = 1; 473 } 474 475 /** 476 * scsi_try_target_reset - Ask host to perform a target reset 477 * @scmd: SCSI cmd used to send a target reset 478 * 479 * Notes: 480 * There is no timeout for this operation. if this operation is 481 * unreliable for a given host, then the host itself needs to put a 482 * timer on it, and set the host back to a consistent state prior to 483 * returning. 484 */ 485 static int scsi_try_target_reset(struct scsi_cmnd *scmd) 486 { 487 unsigned long flags; 488 int rtn; 489 490 if (!scmd->device->host->hostt->eh_target_reset_handler) 491 return FAILED; 492 493 rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd); 494 if (rtn == SUCCESS) { 495 spin_lock_irqsave(scmd->device->host->host_lock, flags); 496 __starget_for_each_device(scsi_target(scmd->device), NULL, 497 __scsi_report_device_reset); 498 spin_unlock_irqrestore(scmd->device->host->host_lock, flags); 499 } 500 501 return rtn; 502 } 503 504 /** 505 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 506 * @scmd: SCSI cmd used to send BDR 507 * 508 * Notes: 509 * There is no timeout for this operation. if this operation is 510 * unreliable for a given host, then the host itself needs to put a 511 * timer on it, and set the host back to a consistent state prior to 512 * returning. 513 */ 514 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 515 { 516 int rtn; 517 518 if (!scmd->device->host->hostt->eh_device_reset_handler) 519 return FAILED; 520 521 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd); 522 if (rtn == SUCCESS) 523 __scsi_report_device_reset(scmd->device, NULL); 524 return rtn; 525 } 526 527 static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd) 528 { 529 if (!scmd->device->host->hostt->eh_abort_handler) 530 return FAILED; 531 532 return scmd->device->host->hostt->eh_abort_handler(scmd); 533 } 534 535 /** 536 * scsi_try_to_abort_cmd - Ask host to abort a running command. 537 * @scmd: SCSI cmd to abort from Lower Level. 538 * 539 * Notes: 540 * This function will not return until the user's completion function 541 * has been called. there is no timeout on this operation. if the 542 * author of the low-level driver wishes this operation to be timed, 543 * they can provide this facility themselves. helper functions in 544 * scsi_error.c can be supplied to make this easier to do. 545 */ 546 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd) 547 { 548 /* 549 * scsi_done was called just after the command timed out and before 550 * we had a chance to process it. (db) 551 */ 552 if (scmd->serial_number == 0) 553 return SUCCESS; 554 return __scsi_try_to_abort_cmd(scmd); 555 } 556 557 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 558 { 559 if (__scsi_try_to_abort_cmd(scmd) != SUCCESS) 560 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 561 if (scsi_try_target_reset(scmd) != SUCCESS) 562 if (scsi_try_bus_reset(scmd) != SUCCESS) 563 scsi_try_host_reset(scmd); 564 } 565 566 /** 567 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory 568 * @scmd: SCSI command structure to hijack 569 * @ses: structure to save restore information 570 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 571 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) 572 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 573 * 574 * This function is used to save a scsi command information before re-execution 575 * as part of the error recovery process. If @sense_bytes is 0 the command 576 * sent must be one that does not transfer any data. If @sense_bytes != 0 577 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 578 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 579 */ 580 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 581 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 582 { 583 struct scsi_device *sdev = scmd->device; 584 585 /* 586 * We need saved copies of a number of fields - this is because 587 * error handling may need to overwrite these with different values 588 * to run different commands, and once error handling is complete, 589 * we will need to restore these values prior to running the actual 590 * command. 591 */ 592 ses->cmd_len = scmd->cmd_len; 593 ses->cmnd = scmd->cmnd; 594 ses->data_direction = scmd->sc_data_direction; 595 ses->sdb = scmd->sdb; 596 ses->next_rq = scmd->request->next_rq; 597 ses->result = scmd->result; 598 ses->underflow = scmd->underflow; 599 ses->prot_op = scmd->prot_op; 600 601 scmd->prot_op = SCSI_PROT_NORMAL; 602 scmd->cmnd = ses->eh_cmnd; 603 memset(scmd->cmnd, 0, BLK_MAX_CDB); 604 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 605 scmd->request->next_rq = NULL; 606 607 if (sense_bytes) { 608 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 609 sense_bytes); 610 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 611 scmd->sdb.length); 612 scmd->sdb.table.sgl = &ses->sense_sgl; 613 scmd->sc_data_direction = DMA_FROM_DEVICE; 614 scmd->sdb.table.nents = 1; 615 scmd->cmnd[0] = REQUEST_SENSE; 616 scmd->cmnd[4] = scmd->sdb.length; 617 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 618 } else { 619 scmd->sc_data_direction = DMA_NONE; 620 if (cmnd) { 621 BUG_ON(cmnd_size > BLK_MAX_CDB); 622 memcpy(scmd->cmnd, cmnd, cmnd_size); 623 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 624 } 625 } 626 627 scmd->underflow = 0; 628 629 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 630 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 631 (sdev->lun << 5 & 0xe0); 632 633 /* 634 * Zero the sense buffer. The scsi spec mandates that any 635 * untransferred sense data should be interpreted as being zero. 636 */ 637 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 638 } 639 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 640 641 /** 642 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory 643 * @scmd: SCSI command structure to restore 644 * @ses: saved information from a coresponding call to scsi_prep_eh_cmnd 645 * 646 * Undo any damage done by above scsi_prep_eh_cmnd(). 647 */ 648 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 649 { 650 /* 651 * Restore original data 652 */ 653 scmd->cmd_len = ses->cmd_len; 654 scmd->cmnd = ses->cmnd; 655 scmd->sc_data_direction = ses->data_direction; 656 scmd->sdb = ses->sdb; 657 scmd->request->next_rq = ses->next_rq; 658 scmd->result = ses->result; 659 scmd->underflow = ses->underflow; 660 scmd->prot_op = ses->prot_op; 661 } 662 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 663 664 /** 665 * scsi_send_eh_cmnd - submit a scsi command as part of error recory 666 * @scmd: SCSI command structure to hijack 667 * @cmnd: CDB to send 668 * @cmnd_size: size in bytes of @cmnd 669 * @timeout: timeout for this request 670 * @sense_bytes: size of sense data to copy or 0 671 * 672 * This function is used to send a scsi command down to a target device 673 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 674 * 675 * Return value: 676 * SUCCESS or FAILED or NEEDS_RETRY 677 */ 678 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd, 679 int cmnd_size, int timeout, unsigned sense_bytes) 680 { 681 struct scsi_device *sdev = scmd->device; 682 struct Scsi_Host *shost = sdev->host; 683 DECLARE_COMPLETION_ONSTACK(done); 684 unsigned long timeleft; 685 unsigned long flags; 686 struct scsi_eh_save ses; 687 int rtn; 688 689 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 690 shost->eh_action = &done; 691 692 spin_lock_irqsave(shost->host_lock, flags); 693 scsi_log_send(scmd); 694 shost->hostt->queuecommand(scmd, scsi_eh_done); 695 spin_unlock_irqrestore(shost->host_lock, flags); 696 697 timeleft = wait_for_completion_timeout(&done, timeout); 698 699 shost->eh_action = NULL; 700 701 scsi_log_completion(scmd, SUCCESS); 702 703 SCSI_LOG_ERROR_RECOVERY(3, 704 printk("%s: scmd: %p, timeleft: %ld\n", 705 __func__, scmd, timeleft)); 706 707 /* 708 * If there is time left scsi_eh_done got called, and we will 709 * examine the actual status codes to see whether the command 710 * actually did complete normally, else tell the host to forget 711 * about this command. 712 */ 713 if (timeleft) { 714 rtn = scsi_eh_completed_normally(scmd); 715 SCSI_LOG_ERROR_RECOVERY(3, 716 printk("%s: scsi_eh_completed_normally %x\n", 717 __func__, rtn)); 718 719 switch (rtn) { 720 case SUCCESS: 721 case NEEDS_RETRY: 722 case FAILED: 723 break; 724 default: 725 rtn = FAILED; 726 break; 727 } 728 } else { 729 scsi_abort_eh_cmnd(scmd); 730 rtn = FAILED; 731 } 732 733 scsi_eh_restore_cmnd(scmd, &ses); 734 return rtn; 735 } 736 737 /** 738 * scsi_request_sense - Request sense data from a particular target. 739 * @scmd: SCSI cmd for request sense. 740 * 741 * Notes: 742 * Some hosts automatically obtain this information, others require 743 * that we obtain it on our own. This function will *not* return until 744 * the command either times out, or it completes. 745 */ 746 static int scsi_request_sense(struct scsi_cmnd *scmd) 747 { 748 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0); 749 } 750 751 /** 752 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 753 * @scmd: Original SCSI cmd that eh has finished. 754 * @done_q: Queue for processed commands. 755 * 756 * Notes: 757 * We don't want to use the normal command completion while we are are 758 * still handling errors - it may cause other commands to be queued, 759 * and that would disturb what we are doing. Thus we really want to 760 * keep a list of pending commands for final completion, and once we 761 * are ready to leave error handling we handle completion for real. 762 */ 763 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 764 { 765 scmd->device->host->host_failed--; 766 scmd->eh_eflags = 0; 767 list_move_tail(&scmd->eh_entry, done_q); 768 } 769 EXPORT_SYMBOL(scsi_eh_finish_cmd); 770 771 /** 772 * scsi_eh_get_sense - Get device sense data. 773 * @work_q: Queue of commands to process. 774 * @done_q: Queue of processed commands. 775 * 776 * Description: 777 * See if we need to request sense information. if so, then get it 778 * now, so we have a better idea of what to do. 779 * 780 * Notes: 781 * This has the unfortunate side effect that if a shost adapter does 782 * not automatically request sense information, we end up shutting 783 * it down before we request it. 784 * 785 * All drivers should request sense information internally these days, 786 * so for now all I have to say is tough noogies if you end up in here. 787 * 788 * XXX: Long term this code should go away, but that needs an audit of 789 * all LLDDs first. 790 */ 791 int scsi_eh_get_sense(struct list_head *work_q, 792 struct list_head *done_q) 793 { 794 struct scsi_cmnd *scmd, *next; 795 int rtn; 796 797 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 798 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) || 799 SCSI_SENSE_VALID(scmd)) 800 continue; 801 802 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 803 "%s: requesting sense\n", 804 current->comm)); 805 rtn = scsi_request_sense(scmd); 806 if (rtn != SUCCESS) 807 continue; 808 809 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p" 810 " result %x\n", scmd, 811 scmd->result)); 812 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd)); 813 814 rtn = scsi_decide_disposition(scmd); 815 816 /* 817 * if the result was normal, then just pass it along to the 818 * upper level. 819 */ 820 if (rtn == SUCCESS) 821 /* we don't want this command reissued, just 822 * finished with the sense data, so set 823 * retries to the max allowed to ensure it 824 * won't get reissued */ 825 scmd->retries = scmd->allowed; 826 else if (rtn != NEEDS_RETRY) 827 continue; 828 829 scsi_eh_finish_cmd(scmd, done_q); 830 } 831 832 return list_empty(work_q); 833 } 834 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 835 836 /** 837 * scsi_eh_tur - Send TUR to device. 838 * @scmd: &scsi_cmnd to send TUR 839 * 840 * Return value: 841 * 0 - Device is ready. 1 - Device NOT ready. 842 */ 843 static int scsi_eh_tur(struct scsi_cmnd *scmd) 844 { 845 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 846 int retry_cnt = 1, rtn; 847 848 retry_tur: 849 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0); 850 851 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", 852 __func__, scmd, rtn)); 853 854 switch (rtn) { 855 case NEEDS_RETRY: 856 if (retry_cnt--) 857 goto retry_tur; 858 /*FALLTHRU*/ 859 case SUCCESS: 860 return 0; 861 default: 862 return 1; 863 } 864 } 865 866 /** 867 * scsi_eh_abort_cmds - abort pending commands. 868 * @work_q: &list_head for pending commands. 869 * @done_q: &list_head for processed commands. 870 * 871 * Decription: 872 * Try and see whether or not it makes sense to try and abort the 873 * running command. This only works out to be the case if we have one 874 * command that has timed out. If the command simply failed, it makes 875 * no sense to try and abort the command, since as far as the shost 876 * adapter is concerned, it isn't running. 877 */ 878 static int scsi_eh_abort_cmds(struct list_head *work_q, 879 struct list_head *done_q) 880 { 881 struct scsi_cmnd *scmd, *next; 882 int rtn; 883 884 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 885 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD)) 886 continue; 887 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:" 888 "0x%p\n", current->comm, 889 scmd)); 890 rtn = scsi_try_to_abort_cmd(scmd); 891 if (rtn == SUCCESS) { 892 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD; 893 if (!scsi_device_online(scmd->device) || 894 !scsi_eh_tur(scmd)) { 895 scsi_eh_finish_cmd(scmd, done_q); 896 } 897 898 } else 899 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting" 900 " cmd failed:" 901 "0x%p\n", 902 current->comm, 903 scmd)); 904 } 905 906 return list_empty(work_q); 907 } 908 909 /** 910 * scsi_eh_try_stu - Send START_UNIT to device. 911 * @scmd: &scsi_cmnd to send START_UNIT 912 * 913 * Return value: 914 * 0 - Device is ready. 1 - Device NOT ready. 915 */ 916 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 917 { 918 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 919 920 if (scmd->device->allow_restart) { 921 int i, rtn = NEEDS_RETRY; 922 923 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 924 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0); 925 926 if (rtn == SUCCESS) 927 return 0; 928 } 929 930 return 1; 931 } 932 933 /** 934 * scsi_eh_stu - send START_UNIT if needed 935 * @shost: &scsi host being recovered. 936 * @work_q: &list_head for pending commands. 937 * @done_q: &list_head for processed commands. 938 * 939 * Notes: 940 * If commands are failing due to not ready, initializing command required, 941 * try revalidating the device, which will end up sending a start unit. 942 */ 943 static int scsi_eh_stu(struct Scsi_Host *shost, 944 struct list_head *work_q, 945 struct list_head *done_q) 946 { 947 struct scsi_cmnd *scmd, *stu_scmd, *next; 948 struct scsi_device *sdev; 949 950 shost_for_each_device(sdev, shost) { 951 stu_scmd = NULL; 952 list_for_each_entry(scmd, work_q, eh_entry) 953 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 954 scsi_check_sense(scmd) == FAILED ) { 955 stu_scmd = scmd; 956 break; 957 } 958 959 if (!stu_scmd) 960 continue; 961 962 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:" 963 " 0x%p\n", current->comm, sdev)); 964 965 if (!scsi_eh_try_stu(stu_scmd)) { 966 if (!scsi_device_online(sdev) || 967 !scsi_eh_tur(stu_scmd)) { 968 list_for_each_entry_safe(scmd, next, 969 work_q, eh_entry) { 970 if (scmd->device == sdev) 971 scsi_eh_finish_cmd(scmd, done_q); 972 } 973 } 974 } else { 975 SCSI_LOG_ERROR_RECOVERY(3, 976 printk("%s: START_UNIT failed to sdev:" 977 " 0x%p\n", current->comm, sdev)); 978 } 979 } 980 981 return list_empty(work_q); 982 } 983 984 985 /** 986 * scsi_eh_bus_device_reset - send bdr if needed 987 * @shost: scsi host being recovered. 988 * @work_q: &list_head for pending commands. 989 * @done_q: &list_head for processed commands. 990 * 991 * Notes: 992 * Try a bus device reset. Still, look to see whether we have multiple 993 * devices that are jammed or not - if we have multiple devices, it 994 * makes no sense to try bus_device_reset - we really would need to try 995 * a bus_reset instead. 996 */ 997 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 998 struct list_head *work_q, 999 struct list_head *done_q) 1000 { 1001 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1002 struct scsi_device *sdev; 1003 int rtn; 1004 1005 shost_for_each_device(sdev, shost) { 1006 bdr_scmd = NULL; 1007 list_for_each_entry(scmd, work_q, eh_entry) 1008 if (scmd->device == sdev) { 1009 bdr_scmd = scmd; 1010 break; 1011 } 1012 1013 if (!bdr_scmd) 1014 continue; 1015 1016 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:" 1017 " 0x%p\n", current->comm, 1018 sdev)); 1019 rtn = scsi_try_bus_device_reset(bdr_scmd); 1020 if (rtn == SUCCESS) { 1021 if (!scsi_device_online(sdev) || 1022 !scsi_eh_tur(bdr_scmd)) { 1023 list_for_each_entry_safe(scmd, next, 1024 work_q, eh_entry) { 1025 if (scmd->device == sdev) 1026 scsi_eh_finish_cmd(scmd, 1027 done_q); 1028 } 1029 } 1030 } else { 1031 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR" 1032 " failed sdev:" 1033 "0x%p\n", 1034 current->comm, 1035 sdev)); 1036 } 1037 } 1038 1039 return list_empty(work_q); 1040 } 1041 1042 /** 1043 * scsi_eh_target_reset - send target reset if needed 1044 * @shost: scsi host being recovered. 1045 * @work_q: &list_head for pending commands. 1046 * @done_q: &list_head for processed commands. 1047 * 1048 * Notes: 1049 * Try a target reset. 1050 */ 1051 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1052 struct list_head *work_q, 1053 struct list_head *done_q) 1054 { 1055 struct scsi_cmnd *scmd, *tgtr_scmd, *next; 1056 unsigned int id = 0; 1057 int rtn; 1058 1059 do { 1060 tgtr_scmd = NULL; 1061 list_for_each_entry(scmd, work_q, eh_entry) { 1062 if (id == scmd_id(scmd)) { 1063 tgtr_scmd = scmd; 1064 break; 1065 } 1066 } 1067 if (!tgtr_scmd) { 1068 /* not one exactly equal; find the next highest */ 1069 list_for_each_entry(scmd, work_q, eh_entry) { 1070 if (scmd_id(scmd) > id && 1071 (!tgtr_scmd || 1072 scmd_id(tgtr_scmd) > scmd_id(scmd))) 1073 tgtr_scmd = scmd; 1074 } 1075 } 1076 if (!tgtr_scmd) 1077 /* no more commands, that's it */ 1078 break; 1079 1080 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset " 1081 "to target %d\n", 1082 current->comm, id)); 1083 rtn = scsi_try_target_reset(tgtr_scmd); 1084 if (rtn == SUCCESS) { 1085 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1086 if (id == scmd_id(scmd)) 1087 if (!scsi_device_online(scmd->device) || 1088 !scsi_eh_tur(tgtr_scmd)) 1089 scsi_eh_finish_cmd(scmd, 1090 done_q); 1091 } 1092 } else 1093 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset" 1094 " failed target: " 1095 "%d\n", 1096 current->comm, id)); 1097 id++; 1098 } while(id != 0); 1099 1100 return list_empty(work_q); 1101 } 1102 1103 /** 1104 * scsi_eh_bus_reset - send a bus reset 1105 * @shost: &scsi host being recovered. 1106 * @work_q: &list_head for pending commands. 1107 * @done_q: &list_head for processed commands. 1108 */ 1109 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1110 struct list_head *work_q, 1111 struct list_head *done_q) 1112 { 1113 struct scsi_cmnd *scmd, *chan_scmd, *next; 1114 unsigned int channel; 1115 int rtn; 1116 1117 /* 1118 * we really want to loop over the various channels, and do this on 1119 * a channel by channel basis. we should also check to see if any 1120 * of the failed commands are on soft_reset devices, and if so, skip 1121 * the reset. 1122 */ 1123 1124 for (channel = 0; channel <= shost->max_channel; channel++) { 1125 chan_scmd = NULL; 1126 list_for_each_entry(scmd, work_q, eh_entry) { 1127 if (channel == scmd_channel(scmd)) { 1128 chan_scmd = scmd; 1129 break; 1130 /* 1131 * FIXME add back in some support for 1132 * soft_reset devices. 1133 */ 1134 } 1135 } 1136 1137 if (!chan_scmd) 1138 continue; 1139 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:" 1140 " %d\n", current->comm, 1141 channel)); 1142 rtn = scsi_try_bus_reset(chan_scmd); 1143 if (rtn == SUCCESS) { 1144 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1145 if (channel == scmd_channel(scmd)) 1146 if (!scsi_device_online(scmd->device) || 1147 !scsi_eh_tur(scmd)) 1148 scsi_eh_finish_cmd(scmd, 1149 done_q); 1150 } 1151 } else { 1152 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST" 1153 " failed chan: %d\n", 1154 current->comm, 1155 channel)); 1156 } 1157 } 1158 return list_empty(work_q); 1159 } 1160 1161 /** 1162 * scsi_eh_host_reset - send a host reset 1163 * @work_q: list_head for processed commands. 1164 * @done_q: list_head for processed commands. 1165 */ 1166 static int scsi_eh_host_reset(struct list_head *work_q, 1167 struct list_head *done_q) 1168 { 1169 struct scsi_cmnd *scmd, *next; 1170 int rtn; 1171 1172 if (!list_empty(work_q)) { 1173 scmd = list_entry(work_q->next, 1174 struct scsi_cmnd, eh_entry); 1175 1176 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n" 1177 , current->comm)); 1178 1179 rtn = scsi_try_host_reset(scmd); 1180 if (rtn == SUCCESS) { 1181 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1182 if (!scsi_device_online(scmd->device) || 1183 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) || 1184 !scsi_eh_tur(scmd)) 1185 scsi_eh_finish_cmd(scmd, done_q); 1186 } 1187 } else { 1188 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST" 1189 " failed\n", 1190 current->comm)); 1191 } 1192 } 1193 return list_empty(work_q); 1194 } 1195 1196 /** 1197 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1198 * @work_q: list_head for processed commands. 1199 * @done_q: list_head for processed commands. 1200 */ 1201 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1202 struct list_head *done_q) 1203 { 1204 struct scsi_cmnd *scmd, *next; 1205 1206 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1207 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1208 "not ready after error recovery\n"); 1209 scsi_device_set_state(scmd->device, SDEV_OFFLINE); 1210 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) { 1211 /* 1212 * FIXME: Handle lost cmds. 1213 */ 1214 } 1215 scsi_eh_finish_cmd(scmd, done_q); 1216 } 1217 return; 1218 } 1219 1220 /** 1221 * scsi_noretry_cmd - determinte if command should be failed fast 1222 * @scmd: SCSI cmd to examine. 1223 */ 1224 int scsi_noretry_cmd(struct scsi_cmnd *scmd) 1225 { 1226 switch (host_byte(scmd->result)) { 1227 case DID_OK: 1228 break; 1229 case DID_BUS_BUSY: 1230 return blk_failfast_transport(scmd->request); 1231 case DID_PARITY: 1232 return blk_failfast_dev(scmd->request); 1233 case DID_ERROR: 1234 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1235 status_byte(scmd->result) == RESERVATION_CONFLICT) 1236 return 0; 1237 /* fall through */ 1238 case DID_SOFT_ERROR: 1239 return blk_failfast_driver(scmd->request); 1240 } 1241 1242 switch (status_byte(scmd->result)) { 1243 case CHECK_CONDITION: 1244 /* 1245 * assume caller has checked sense and determinted 1246 * the check condition was retryable. 1247 */ 1248 return blk_failfast_dev(scmd->request); 1249 } 1250 1251 return 0; 1252 } 1253 1254 /** 1255 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1256 * @scmd: SCSI cmd to examine. 1257 * 1258 * Notes: 1259 * This is *only* called when we are examining the status after sending 1260 * out the actual data command. any commands that are queued for error 1261 * recovery (e.g. test_unit_ready) do *not* come through here. 1262 * 1263 * When this routine returns failed, it means the error handler thread 1264 * is woken. In cases where the error code indicates an error that 1265 * doesn't require the error handler read (i.e. we don't need to 1266 * abort/reset), this function should return SUCCESS. 1267 */ 1268 int scsi_decide_disposition(struct scsi_cmnd *scmd) 1269 { 1270 int rtn; 1271 1272 /* 1273 * if the device is offline, then we clearly just pass the result back 1274 * up to the top level. 1275 */ 1276 if (!scsi_device_online(scmd->device)) { 1277 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report" 1278 " as SUCCESS\n", 1279 __func__)); 1280 return SUCCESS; 1281 } 1282 1283 /* 1284 * first check the host byte, to see if there is anything in there 1285 * that would indicate what we need to do. 1286 */ 1287 switch (host_byte(scmd->result)) { 1288 case DID_PASSTHROUGH: 1289 /* 1290 * no matter what, pass this through to the upper layer. 1291 * nuke this special code so that it looks like we are saying 1292 * did_ok. 1293 */ 1294 scmd->result &= 0xff00ffff; 1295 return SUCCESS; 1296 case DID_OK: 1297 /* 1298 * looks good. drop through, and check the next byte. 1299 */ 1300 break; 1301 case DID_NO_CONNECT: 1302 case DID_BAD_TARGET: 1303 case DID_ABORT: 1304 /* 1305 * note - this means that we just report the status back 1306 * to the top level driver, not that we actually think 1307 * that it indicates SUCCESS. 1308 */ 1309 return SUCCESS; 1310 /* 1311 * when the low level driver returns did_soft_error, 1312 * it is responsible for keeping an internal retry counter 1313 * in order to avoid endless loops (db) 1314 * 1315 * actually this is a bug in this function here. we should 1316 * be mindful of the maximum number of retries specified 1317 * and not get stuck in a loop. 1318 */ 1319 case DID_SOFT_ERROR: 1320 goto maybe_retry; 1321 case DID_IMM_RETRY: 1322 return NEEDS_RETRY; 1323 1324 case DID_REQUEUE: 1325 return ADD_TO_MLQUEUE; 1326 case DID_TRANSPORT_DISRUPTED: 1327 /* 1328 * LLD/transport was disrupted during processing of the IO. 1329 * The transport class is now blocked/blocking, 1330 * and the transport will decide what to do with the IO 1331 * based on its timers and recovery capablilities if 1332 * there are enough retries. 1333 */ 1334 goto maybe_retry; 1335 case DID_TRANSPORT_FAILFAST: 1336 /* 1337 * The transport decided to failfast the IO (most likely 1338 * the fast io fail tmo fired), so send IO directly upwards. 1339 */ 1340 return SUCCESS; 1341 case DID_ERROR: 1342 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1343 status_byte(scmd->result) == RESERVATION_CONFLICT) 1344 /* 1345 * execute reservation conflict processing code 1346 * lower down 1347 */ 1348 break; 1349 /* fallthrough */ 1350 1351 case DID_BUS_BUSY: 1352 case DID_PARITY: 1353 goto maybe_retry; 1354 case DID_TIME_OUT: 1355 /* 1356 * when we scan the bus, we get timeout messages for 1357 * these commands if there is no device available. 1358 * other hosts report did_no_connect for the same thing. 1359 */ 1360 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1361 scmd->cmnd[0] == INQUIRY)) { 1362 return SUCCESS; 1363 } else { 1364 return FAILED; 1365 } 1366 case DID_RESET: 1367 return SUCCESS; 1368 default: 1369 return FAILED; 1370 } 1371 1372 /* 1373 * next, check the message byte. 1374 */ 1375 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 1376 return FAILED; 1377 1378 /* 1379 * check the status byte to see if this indicates anything special. 1380 */ 1381 switch (status_byte(scmd->result)) { 1382 case QUEUE_FULL: 1383 /* 1384 * the case of trying to send too many commands to a 1385 * tagged queueing device. 1386 */ 1387 case BUSY: 1388 /* 1389 * device can't talk to us at the moment. Should only 1390 * occur (SAM-3) when the task queue is empty, so will cause 1391 * the empty queue handling to trigger a stall in the 1392 * device. 1393 */ 1394 return ADD_TO_MLQUEUE; 1395 case GOOD: 1396 case COMMAND_TERMINATED: 1397 return SUCCESS; 1398 case TASK_ABORTED: 1399 goto maybe_retry; 1400 case CHECK_CONDITION: 1401 rtn = scsi_check_sense(scmd); 1402 if (rtn == NEEDS_RETRY) 1403 goto maybe_retry; 1404 /* if rtn == FAILED, we have no sense information; 1405 * returning FAILED will wake the error handler thread 1406 * to collect the sense and redo the decide 1407 * disposition */ 1408 return rtn; 1409 case CONDITION_GOOD: 1410 case INTERMEDIATE_GOOD: 1411 case INTERMEDIATE_C_GOOD: 1412 case ACA_ACTIVE: 1413 /* 1414 * who knows? FIXME(eric) 1415 */ 1416 return SUCCESS; 1417 1418 case RESERVATION_CONFLICT: 1419 sdev_printk(KERN_INFO, scmd->device, 1420 "reservation conflict\n"); 1421 return SUCCESS; /* causes immediate i/o error */ 1422 default: 1423 return FAILED; 1424 } 1425 return FAILED; 1426 1427 maybe_retry: 1428 1429 /* we requeue for retry because the error was retryable, and 1430 * the request was not marked fast fail. Note that above, 1431 * even if the request is marked fast fail, we still requeue 1432 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1433 if ((++scmd->retries) <= scmd->allowed 1434 && !scsi_noretry_cmd(scmd)) { 1435 return NEEDS_RETRY; 1436 } else { 1437 /* 1438 * no more retries - report this one back to upper level. 1439 */ 1440 return SUCCESS; 1441 } 1442 } 1443 1444 /** 1445 * scsi_eh_lock_door - Prevent medium removal for the specified device 1446 * @sdev: SCSI device to prevent medium removal 1447 * 1448 * Locking: 1449 * We must be called from process context; scsi_allocate_request() 1450 * may sleep. 1451 * 1452 * Notes: 1453 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 1454 * head of the devices request queue, and continue. 1455 * 1456 * Bugs: 1457 * scsi_allocate_request() may sleep waiting for existing requests to 1458 * be processed. However, since we haven't kicked off any request 1459 * processing for this host, this may deadlock. 1460 * 1461 * If scsi_allocate_request() fails for what ever reason, we 1462 * completely forget to lock the door. 1463 */ 1464 static void scsi_eh_lock_door(struct scsi_device *sdev) 1465 { 1466 unsigned char cmnd[MAX_COMMAND_SIZE]; 1467 1468 cmnd[0] = ALLOW_MEDIUM_REMOVAL; 1469 cmnd[1] = 0; 1470 cmnd[2] = 0; 1471 cmnd[3] = 0; 1472 cmnd[4] = SCSI_REMOVAL_PREVENT; 1473 cmnd[5] = 0; 1474 1475 scsi_execute_async(sdev, cmnd, 6, DMA_NONE, NULL, 0, 0, 10 * HZ, 1476 5, NULL, NULL, GFP_KERNEL); 1477 } 1478 1479 1480 /** 1481 * scsi_restart_operations - restart io operations to the specified host. 1482 * @shost: Host we are restarting. 1483 * 1484 * Notes: 1485 * When we entered the error handler, we blocked all further i/o to 1486 * this device. we need to 'reverse' this process. 1487 */ 1488 static void scsi_restart_operations(struct Scsi_Host *shost) 1489 { 1490 struct scsi_device *sdev; 1491 unsigned long flags; 1492 1493 /* 1494 * If the door was locked, we need to insert a door lock request 1495 * onto the head of the SCSI request queue for the device. There 1496 * is no point trying to lock the door of an off-line device. 1497 */ 1498 shost_for_each_device(sdev, shost) { 1499 if (scsi_device_online(sdev) && sdev->locked) 1500 scsi_eh_lock_door(sdev); 1501 } 1502 1503 /* 1504 * next free up anything directly waiting upon the host. this 1505 * will be requests for character device operations, and also for 1506 * ioctls to queued block devices. 1507 */ 1508 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n", 1509 __func__)); 1510 1511 spin_lock_irqsave(shost->host_lock, flags); 1512 if (scsi_host_set_state(shost, SHOST_RUNNING)) 1513 if (scsi_host_set_state(shost, SHOST_CANCEL)) 1514 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 1515 spin_unlock_irqrestore(shost->host_lock, flags); 1516 1517 wake_up(&shost->host_wait); 1518 1519 /* 1520 * finally we need to re-initiate requests that may be pending. we will 1521 * have had everything blocked while error handling is taking place, and 1522 * now that error recovery is done, we will need to ensure that these 1523 * requests are started. 1524 */ 1525 scsi_run_host_queues(shost); 1526 } 1527 1528 /** 1529 * scsi_eh_ready_devs - check device ready state and recover if not. 1530 * @shost: host to be recovered. 1531 * @work_q: &list_head for pending commands. 1532 * @done_q: &list_head for processed commands. 1533 */ 1534 void scsi_eh_ready_devs(struct Scsi_Host *shost, 1535 struct list_head *work_q, 1536 struct list_head *done_q) 1537 { 1538 if (!scsi_eh_stu(shost, work_q, done_q)) 1539 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 1540 if (!scsi_eh_target_reset(shost, work_q, done_q)) 1541 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 1542 if (!scsi_eh_host_reset(work_q, done_q)) 1543 scsi_eh_offline_sdevs(work_q, 1544 done_q); 1545 } 1546 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 1547 1548 /** 1549 * scsi_eh_flush_done_q - finish processed commands or retry them. 1550 * @done_q: list_head of processed commands. 1551 */ 1552 void scsi_eh_flush_done_q(struct list_head *done_q) 1553 { 1554 struct scsi_cmnd *scmd, *next; 1555 1556 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 1557 list_del_init(&scmd->eh_entry); 1558 if (scsi_device_online(scmd->device) && 1559 !scsi_noretry_cmd(scmd) && 1560 (++scmd->retries <= scmd->allowed)) { 1561 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush" 1562 " retry cmd: %p\n", 1563 current->comm, 1564 scmd)); 1565 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 1566 } else { 1567 /* 1568 * If just we got sense for the device (called 1569 * scsi_eh_get_sense), scmd->result is already 1570 * set, do not set DRIVER_TIMEOUT. 1571 */ 1572 if (!scmd->result) 1573 scmd->result |= (DRIVER_TIMEOUT << 24); 1574 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish" 1575 " cmd: %p\n", 1576 current->comm, scmd)); 1577 scsi_finish_command(scmd); 1578 } 1579 } 1580 } 1581 EXPORT_SYMBOL(scsi_eh_flush_done_q); 1582 1583 /** 1584 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 1585 * @shost: Host to unjam. 1586 * 1587 * Notes: 1588 * When we come in here, we *know* that all commands on the bus have 1589 * either completed, failed or timed out. we also know that no further 1590 * commands are being sent to the host, so things are relatively quiet 1591 * and we have freedom to fiddle with things as we wish. 1592 * 1593 * This is only the *default* implementation. it is possible for 1594 * individual drivers to supply their own version of this function, and 1595 * if the maintainer wishes to do this, it is strongly suggested that 1596 * this function be taken as a template and modified. this function 1597 * was designed to correctly handle problems for about 95% of the 1598 * different cases out there, and it should always provide at least a 1599 * reasonable amount of error recovery. 1600 * 1601 * Any command marked 'failed' or 'timeout' must eventually have 1602 * scsi_finish_cmd() called for it. we do all of the retry stuff 1603 * here, so when we restart the host after we return it should have an 1604 * empty queue. 1605 */ 1606 static void scsi_unjam_host(struct Scsi_Host *shost) 1607 { 1608 unsigned long flags; 1609 LIST_HEAD(eh_work_q); 1610 LIST_HEAD(eh_done_q); 1611 1612 spin_lock_irqsave(shost->host_lock, flags); 1613 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 1614 spin_unlock_irqrestore(shost->host_lock, flags); 1615 1616 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 1617 1618 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 1619 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q)) 1620 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 1621 1622 scsi_eh_flush_done_q(&eh_done_q); 1623 } 1624 1625 /** 1626 * scsi_error_handler - SCSI error handler thread 1627 * @data: Host for which we are running. 1628 * 1629 * Notes: 1630 * This is the main error handling loop. This is run as a kernel thread 1631 * for every SCSI host and handles all error handling activity. 1632 */ 1633 int scsi_error_handler(void *data) 1634 { 1635 struct Scsi_Host *shost = data; 1636 1637 /* 1638 * We use TASK_INTERRUPTIBLE so that the thread is not 1639 * counted against the load average as a running process. 1640 * We never actually get interrupted because kthread_run 1641 * disables signal delivery for the created thread. 1642 */ 1643 set_current_state(TASK_INTERRUPTIBLE); 1644 while (!kthread_should_stop()) { 1645 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 1646 shost->host_failed != shost->host_busy) { 1647 SCSI_LOG_ERROR_RECOVERY(1, 1648 printk("Error handler scsi_eh_%d sleeping\n", 1649 shost->host_no)); 1650 schedule(); 1651 set_current_state(TASK_INTERRUPTIBLE); 1652 continue; 1653 } 1654 1655 __set_current_state(TASK_RUNNING); 1656 SCSI_LOG_ERROR_RECOVERY(1, 1657 printk("Error handler scsi_eh_%d waking up\n", 1658 shost->host_no)); 1659 1660 /* 1661 * We have a host that is failing for some reason. Figure out 1662 * what we need to do to get it up and online again (if we can). 1663 * If we fail, we end up taking the thing offline. 1664 */ 1665 if (shost->transportt->eh_strategy_handler) 1666 shost->transportt->eh_strategy_handler(shost); 1667 else 1668 scsi_unjam_host(shost); 1669 1670 /* 1671 * Note - if the above fails completely, the action is to take 1672 * individual devices offline and flush the queue of any 1673 * outstanding requests that may have been pending. When we 1674 * restart, we restart any I/O to any other devices on the bus 1675 * which are still online. 1676 */ 1677 scsi_restart_operations(shost); 1678 set_current_state(TASK_INTERRUPTIBLE); 1679 } 1680 __set_current_state(TASK_RUNNING); 1681 1682 SCSI_LOG_ERROR_RECOVERY(1, 1683 printk("Error handler scsi_eh_%d exiting\n", shost->host_no)); 1684 shost->ehandler = NULL; 1685 return 0; 1686 } 1687 1688 /* 1689 * Function: scsi_report_bus_reset() 1690 * 1691 * Purpose: Utility function used by low-level drivers to report that 1692 * they have observed a bus reset on the bus being handled. 1693 * 1694 * Arguments: shost - Host in question 1695 * channel - channel on which reset was observed. 1696 * 1697 * Returns: Nothing 1698 * 1699 * Lock status: Host lock must be held. 1700 * 1701 * Notes: This only needs to be called if the reset is one which 1702 * originates from an unknown location. Resets originated 1703 * by the mid-level itself don't need to call this, but there 1704 * should be no harm. 1705 * 1706 * The main purpose of this is to make sure that a CHECK_CONDITION 1707 * is properly treated. 1708 */ 1709 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 1710 { 1711 struct scsi_device *sdev; 1712 1713 __shost_for_each_device(sdev, shost) { 1714 if (channel == sdev_channel(sdev)) 1715 __scsi_report_device_reset(sdev, NULL); 1716 } 1717 } 1718 EXPORT_SYMBOL(scsi_report_bus_reset); 1719 1720 /* 1721 * Function: scsi_report_device_reset() 1722 * 1723 * Purpose: Utility function used by low-level drivers to report that 1724 * they have observed a device reset on the device being handled. 1725 * 1726 * Arguments: shost - Host in question 1727 * channel - channel on which reset was observed 1728 * target - target on which reset was observed 1729 * 1730 * Returns: Nothing 1731 * 1732 * Lock status: Host lock must be held 1733 * 1734 * Notes: This only needs to be called if the reset is one which 1735 * originates from an unknown location. Resets originated 1736 * by the mid-level itself don't need to call this, but there 1737 * should be no harm. 1738 * 1739 * The main purpose of this is to make sure that a CHECK_CONDITION 1740 * is properly treated. 1741 */ 1742 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 1743 { 1744 struct scsi_device *sdev; 1745 1746 __shost_for_each_device(sdev, shost) { 1747 if (channel == sdev_channel(sdev) && 1748 target == sdev_id(sdev)) 1749 __scsi_report_device_reset(sdev, NULL); 1750 } 1751 } 1752 EXPORT_SYMBOL(scsi_report_device_reset); 1753 1754 static void 1755 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 1756 { 1757 } 1758 1759 /* 1760 * Function: scsi_reset_provider 1761 * 1762 * Purpose: Send requested reset to a bus or device at any phase. 1763 * 1764 * Arguments: device - device to send reset to 1765 * flag - reset type (see scsi.h) 1766 * 1767 * Returns: SUCCESS/FAILURE. 1768 * 1769 * Notes: This is used by the SCSI Generic driver to provide 1770 * Bus/Device reset capability. 1771 */ 1772 int 1773 scsi_reset_provider(struct scsi_device *dev, int flag) 1774 { 1775 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL); 1776 struct Scsi_Host *shost = dev->host; 1777 struct request req; 1778 unsigned long flags; 1779 int rtn; 1780 1781 blk_rq_init(NULL, &req); 1782 scmd->request = &req; 1783 1784 scmd->cmnd = req.cmd; 1785 1786 scmd->scsi_done = scsi_reset_provider_done_command; 1787 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 1788 1789 scmd->cmd_len = 0; 1790 1791 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 1792 1793 spin_lock_irqsave(shost->host_lock, flags); 1794 shost->tmf_in_progress = 1; 1795 spin_unlock_irqrestore(shost->host_lock, flags); 1796 1797 switch (flag) { 1798 case SCSI_TRY_RESET_DEVICE: 1799 rtn = scsi_try_bus_device_reset(scmd); 1800 if (rtn == SUCCESS) 1801 break; 1802 /* FALLTHROUGH */ 1803 case SCSI_TRY_RESET_TARGET: 1804 rtn = scsi_try_target_reset(scmd); 1805 if (rtn == SUCCESS) 1806 break; 1807 /* FALLTHROUGH */ 1808 case SCSI_TRY_RESET_BUS: 1809 rtn = scsi_try_bus_reset(scmd); 1810 if (rtn == SUCCESS) 1811 break; 1812 /* FALLTHROUGH */ 1813 case SCSI_TRY_RESET_HOST: 1814 rtn = scsi_try_host_reset(scmd); 1815 break; 1816 default: 1817 rtn = FAILED; 1818 } 1819 1820 spin_lock_irqsave(shost->host_lock, flags); 1821 shost->tmf_in_progress = 0; 1822 spin_unlock_irqrestore(shost->host_lock, flags); 1823 1824 /* 1825 * be sure to wake up anyone who was sleeping or had their queue 1826 * suspended while we performed the TMF. 1827 */ 1828 SCSI_LOG_ERROR_RECOVERY(3, 1829 printk("%s: waking up host to restart after TMF\n", 1830 __func__)); 1831 1832 wake_up(&shost->host_wait); 1833 1834 scsi_run_host_queues(shost); 1835 1836 scsi_next_command(scmd); 1837 return rtn; 1838 } 1839 EXPORT_SYMBOL(scsi_reset_provider); 1840 1841 /** 1842 * scsi_normalize_sense - normalize main elements from either fixed or 1843 * descriptor sense data format into a common format. 1844 * 1845 * @sense_buffer: byte array containing sense data returned by device 1846 * @sb_len: number of valid bytes in sense_buffer 1847 * @sshdr: pointer to instance of structure that common 1848 * elements are written to. 1849 * 1850 * Notes: 1851 * The "main elements" from sense data are: response_code, sense_key, 1852 * asc, ascq and additional_length (only for descriptor format). 1853 * 1854 * Typically this function can be called after a device has 1855 * responded to a SCSI command with the CHECK_CONDITION status. 1856 * 1857 * Return value: 1858 * 1 if valid sense data information found, else 0; 1859 */ 1860 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len, 1861 struct scsi_sense_hdr *sshdr) 1862 { 1863 if (!sense_buffer || !sb_len) 1864 return 0; 1865 1866 memset(sshdr, 0, sizeof(struct scsi_sense_hdr)); 1867 1868 sshdr->response_code = (sense_buffer[0] & 0x7f); 1869 1870 if (!scsi_sense_valid(sshdr)) 1871 return 0; 1872 1873 if (sshdr->response_code >= 0x72) { 1874 /* 1875 * descriptor format 1876 */ 1877 if (sb_len > 1) 1878 sshdr->sense_key = (sense_buffer[1] & 0xf); 1879 if (sb_len > 2) 1880 sshdr->asc = sense_buffer[2]; 1881 if (sb_len > 3) 1882 sshdr->ascq = sense_buffer[3]; 1883 if (sb_len > 7) 1884 sshdr->additional_length = sense_buffer[7]; 1885 } else { 1886 /* 1887 * fixed format 1888 */ 1889 if (sb_len > 2) 1890 sshdr->sense_key = (sense_buffer[2] & 0xf); 1891 if (sb_len > 7) { 1892 sb_len = (sb_len < (sense_buffer[7] + 8)) ? 1893 sb_len : (sense_buffer[7] + 8); 1894 if (sb_len > 12) 1895 sshdr->asc = sense_buffer[12]; 1896 if (sb_len > 13) 1897 sshdr->ascq = sense_buffer[13]; 1898 } 1899 } 1900 1901 return 1; 1902 } 1903 EXPORT_SYMBOL(scsi_normalize_sense); 1904 1905 int scsi_command_normalize_sense(struct scsi_cmnd *cmd, 1906 struct scsi_sense_hdr *sshdr) 1907 { 1908 return scsi_normalize_sense(cmd->sense_buffer, 1909 SCSI_SENSE_BUFFERSIZE, sshdr); 1910 } 1911 EXPORT_SYMBOL(scsi_command_normalize_sense); 1912 1913 /** 1914 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format. 1915 * @sense_buffer: byte array of descriptor format sense data 1916 * @sb_len: number of valid bytes in sense_buffer 1917 * @desc_type: value of descriptor type to find 1918 * (e.g. 0 -> information) 1919 * 1920 * Notes: 1921 * only valid when sense data is in descriptor format 1922 * 1923 * Return value: 1924 * pointer to start of (first) descriptor if found else NULL 1925 */ 1926 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len, 1927 int desc_type) 1928 { 1929 int add_sen_len, add_len, desc_len, k; 1930 const u8 * descp; 1931 1932 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7]))) 1933 return NULL; 1934 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73)) 1935 return NULL; 1936 add_sen_len = (add_sen_len < (sb_len - 8)) ? 1937 add_sen_len : (sb_len - 8); 1938 descp = &sense_buffer[8]; 1939 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) { 1940 descp += desc_len; 1941 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1; 1942 desc_len = add_len + 2; 1943 if (descp[0] == desc_type) 1944 return descp; 1945 if (add_len < 0) // short descriptor ?? 1946 break; 1947 } 1948 return NULL; 1949 } 1950 EXPORT_SYMBOL(scsi_sense_desc_find); 1951 1952 /** 1953 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 1954 * @sense_buffer: byte array of sense data 1955 * @sb_len: number of valid bytes in sense_buffer 1956 * @info_out: pointer to 64 integer where 8 or 4 byte information 1957 * field will be placed if found. 1958 * 1959 * Return value: 1960 * 1 if information field found, 0 if not found. 1961 */ 1962 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len, 1963 u64 * info_out) 1964 { 1965 int j; 1966 const u8 * ucp; 1967 u64 ull; 1968 1969 if (sb_len < 7) 1970 return 0; 1971 switch (sense_buffer[0] & 0x7f) { 1972 case 0x70: 1973 case 0x71: 1974 if (sense_buffer[0] & 0x80) { 1975 *info_out = (sense_buffer[3] << 24) + 1976 (sense_buffer[4] << 16) + 1977 (sense_buffer[5] << 8) + sense_buffer[6]; 1978 return 1; 1979 } else 1980 return 0; 1981 case 0x72: 1982 case 0x73: 1983 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 1984 0 /* info desc */); 1985 if (ucp && (0xa == ucp[1])) { 1986 ull = 0; 1987 for (j = 0; j < 8; ++j) { 1988 if (j > 0) 1989 ull <<= 8; 1990 ull |= ucp[4 + j]; 1991 } 1992 *info_out = ull; 1993 return 1; 1994 } else 1995 return 0; 1996 default: 1997 return 0; 1998 } 1999 } 2000 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2001 2002 /** 2003 * scsi_build_sense_buffer - build sense data in a buffer 2004 * @desc: Sense format (non zero == descriptor format, 2005 * 0 == fixed format) 2006 * @buf: Where to build sense data 2007 * @key: Sense key 2008 * @asc: Additional sense code 2009 * @ascq: Additional sense code qualifier 2010 * 2011 **/ 2012 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq) 2013 { 2014 if (desc) { 2015 buf[0] = 0x72; /* descriptor, current */ 2016 buf[1] = key; 2017 buf[2] = asc; 2018 buf[3] = ascq; 2019 buf[7] = 0; 2020 } else { 2021 buf[0] = 0x70; /* fixed, current */ 2022 buf[2] = key; 2023 buf[7] = 0xa; 2024 buf[12] = asc; 2025 buf[13] = ascq; 2026 } 2027 } 2028 EXPORT_SYMBOL(scsi_build_sense_buffer); 2029