1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * scsi_error.c Copyright (C) 1997 Eric Youngdale 4 * 5 * SCSI error/timeout handling 6 * Initial versions: Eric Youngdale. Based upon conversations with 7 * Leonard Zubkoff and David Miller at Linux Expo, 8 * ideas originating from all over the place. 9 * 10 * Restructured scsi_unjam_host and associated functions. 11 * September 04, 2002 Mike Anderson (andmike@us.ibm.com) 12 * 13 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and 14 * minor cleanups. 15 * September 30, 2002 Mike Anderson (andmike@us.ibm.com) 16 */ 17 18 #include <linux/module.h> 19 #include <linux/sched.h> 20 #include <linux/gfp.h> 21 #include <linux/timer.h> 22 #include <linux/string.h> 23 #include <linux/kernel.h> 24 #include <linux/freezer.h> 25 #include <linux/kthread.h> 26 #include <linux/interrupt.h> 27 #include <linux/blkdev.h> 28 #include <linux/delay.h> 29 #include <linux/jiffies.h> 30 31 #include <scsi/scsi.h> 32 #include <scsi/scsi_cmnd.h> 33 #include <scsi/scsi_dbg.h> 34 #include <scsi/scsi_device.h> 35 #include <scsi/scsi_driver.h> 36 #include <scsi/scsi_eh.h> 37 #include <scsi/scsi_common.h> 38 #include <scsi/scsi_transport.h> 39 #include <scsi/scsi_host.h> 40 #include <scsi/scsi_ioctl.h> 41 #include <scsi/scsi_dh.h> 42 #include <scsi/scsi_devinfo.h> 43 #include <scsi/sg.h> 44 45 #include "scsi_priv.h" 46 #include "scsi_logging.h" 47 #include "scsi_transport_api.h" 48 49 #include <trace/events/scsi.h> 50 51 #include <asm/unaligned.h> 52 53 /* 54 * These should *probably* be handled by the host itself. 55 * Since it is allowed to sleep, it probably should. 56 */ 57 #define BUS_RESET_SETTLE_TIME (10) 58 #define HOST_RESET_SETTLE_TIME (10) 59 60 static int scsi_eh_try_stu(struct scsi_cmnd *scmd); 61 static enum scsi_disposition scsi_try_to_abort_cmd(struct scsi_host_template *, 62 struct scsi_cmnd *); 63 64 void scsi_eh_wakeup(struct Scsi_Host *shost) 65 { 66 lockdep_assert_held(shost->host_lock); 67 68 if (scsi_host_busy(shost) == shost->host_failed) { 69 trace_scsi_eh_wakeup(shost); 70 wake_up_process(shost->ehandler); 71 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost, 72 "Waking error handler thread\n")); 73 } 74 } 75 76 /** 77 * scsi_schedule_eh - schedule EH for SCSI host 78 * @shost: SCSI host to invoke error handling on. 79 * 80 * Schedule SCSI EH without scmd. 81 */ 82 void scsi_schedule_eh(struct Scsi_Host *shost) 83 { 84 unsigned long flags; 85 86 spin_lock_irqsave(shost->host_lock, flags); 87 88 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || 89 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { 90 shost->host_eh_scheduled++; 91 scsi_eh_wakeup(shost); 92 } 93 94 spin_unlock_irqrestore(shost->host_lock, flags); 95 } 96 EXPORT_SYMBOL_GPL(scsi_schedule_eh); 97 98 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost) 99 { 100 if (!shost->last_reset || shost->eh_deadline == -1) 101 return 0; 102 103 /* 104 * 32bit accesses are guaranteed to be atomic 105 * (on all supported architectures), so instead 106 * of using a spinlock we can as well double check 107 * if eh_deadline has been set to 'off' during the 108 * time_before call. 109 */ 110 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) && 111 shost->eh_deadline > -1) 112 return 0; 113 114 return 1; 115 } 116 117 static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd) 118 { 119 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) 120 return true; 121 122 return ++cmd->retries <= cmd->allowed; 123 } 124 125 static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd) 126 { 127 struct scsi_device *sdev = cmd->device; 128 struct Scsi_Host *host = sdev->host; 129 130 if (host->hostt->eh_should_retry_cmd) 131 return host->hostt->eh_should_retry_cmd(cmd); 132 133 return true; 134 } 135 136 /** 137 * scmd_eh_abort_handler - Handle command aborts 138 * @work: command to be aborted. 139 * 140 * Note: this function must be called only for a command that has timed out. 141 * Because the block layer marks a request as complete before it calls 142 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has 143 * timed out do not have any effect. Hence it is safe to call 144 * scsi_finish_command() from this function. 145 */ 146 void 147 scmd_eh_abort_handler(struct work_struct *work) 148 { 149 struct scsi_cmnd *scmd = 150 container_of(work, struct scsi_cmnd, abort_work.work); 151 struct scsi_device *sdev = scmd->device; 152 enum scsi_disposition rtn; 153 154 if (scsi_host_eh_past_deadline(sdev->host)) { 155 SCSI_LOG_ERROR_RECOVERY(3, 156 scmd_printk(KERN_INFO, scmd, 157 "eh timeout, not aborting\n")); 158 } else { 159 SCSI_LOG_ERROR_RECOVERY(3, 160 scmd_printk(KERN_INFO, scmd, 161 "aborting command\n")); 162 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd); 163 if (rtn == SUCCESS) { 164 set_host_byte(scmd, DID_TIME_OUT); 165 if (scsi_host_eh_past_deadline(sdev->host)) { 166 SCSI_LOG_ERROR_RECOVERY(3, 167 scmd_printk(KERN_INFO, scmd, 168 "eh timeout, not retrying " 169 "aborted command\n")); 170 } else if (!scsi_noretry_cmd(scmd) && 171 scsi_cmd_retry_allowed(scmd) && 172 scsi_eh_should_retry_cmd(scmd)) { 173 SCSI_LOG_ERROR_RECOVERY(3, 174 scmd_printk(KERN_WARNING, scmd, 175 "retry aborted command\n")); 176 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 177 return; 178 } else { 179 SCSI_LOG_ERROR_RECOVERY(3, 180 scmd_printk(KERN_WARNING, scmd, 181 "finish aborted command\n")); 182 scsi_finish_command(scmd); 183 return; 184 } 185 } else { 186 SCSI_LOG_ERROR_RECOVERY(3, 187 scmd_printk(KERN_INFO, scmd, 188 "cmd abort %s\n", 189 (rtn == FAST_IO_FAIL) ? 190 "not send" : "failed")); 191 } 192 } 193 194 scsi_eh_scmd_add(scmd); 195 } 196 197 /** 198 * scsi_abort_command - schedule a command abort 199 * @scmd: scmd to abort. 200 * 201 * We only need to abort commands after a command timeout 202 */ 203 static int 204 scsi_abort_command(struct scsi_cmnd *scmd) 205 { 206 struct scsi_device *sdev = scmd->device; 207 struct Scsi_Host *shost = sdev->host; 208 unsigned long flags; 209 210 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 211 /* 212 * Retry after abort failed, escalate to next level. 213 */ 214 SCSI_LOG_ERROR_RECOVERY(3, 215 scmd_printk(KERN_INFO, scmd, 216 "previous abort failed\n")); 217 BUG_ON(delayed_work_pending(&scmd->abort_work)); 218 return FAILED; 219 } 220 221 spin_lock_irqsave(shost->host_lock, flags); 222 if (shost->eh_deadline != -1 && !shost->last_reset) 223 shost->last_reset = jiffies; 224 spin_unlock_irqrestore(shost->host_lock, flags); 225 226 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; 227 SCSI_LOG_ERROR_RECOVERY(3, 228 scmd_printk(KERN_INFO, scmd, "abort scheduled\n")); 229 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100); 230 return SUCCESS; 231 } 232 233 /** 234 * scsi_eh_reset - call into ->eh_action to reset internal counters 235 * @scmd: scmd to run eh on. 236 * 237 * The scsi driver might be carrying internal state about the 238 * devices, so we need to call into the driver to reset the 239 * internal state once the error handler is started. 240 */ 241 static void scsi_eh_reset(struct scsi_cmnd *scmd) 242 { 243 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) { 244 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 245 if (sdrv->eh_reset) 246 sdrv->eh_reset(scmd); 247 } 248 } 249 250 static void scsi_eh_inc_host_failed(struct rcu_head *head) 251 { 252 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu); 253 struct Scsi_Host *shost = scmd->device->host; 254 unsigned long flags; 255 256 spin_lock_irqsave(shost->host_lock, flags); 257 shost->host_failed++; 258 scsi_eh_wakeup(shost); 259 spin_unlock_irqrestore(shost->host_lock, flags); 260 } 261 262 /** 263 * scsi_eh_scmd_add - add scsi cmd to error handling. 264 * @scmd: scmd to run eh on. 265 */ 266 void scsi_eh_scmd_add(struct scsi_cmnd *scmd) 267 { 268 struct Scsi_Host *shost = scmd->device->host; 269 unsigned long flags; 270 int ret; 271 272 WARN_ON_ONCE(!shost->ehandler); 273 274 spin_lock_irqsave(shost->host_lock, flags); 275 if (scsi_host_set_state(shost, SHOST_RECOVERY)) { 276 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY); 277 WARN_ON_ONCE(ret); 278 } 279 if (shost->eh_deadline != -1 && !shost->last_reset) 280 shost->last_reset = jiffies; 281 282 scsi_eh_reset(scmd); 283 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 284 spin_unlock_irqrestore(shost->host_lock, flags); 285 /* 286 * Ensure that all tasks observe the host state change before the 287 * host_failed change. 288 */ 289 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed); 290 } 291 292 /** 293 * scsi_times_out - Timeout function for normal scsi commands. 294 * @req: request that is timing out. 295 * 296 * Notes: 297 * We do not need to lock this. There is the potential for a race 298 * only in that the normal completion handling might run, but if the 299 * normal completion function determines that the timer has already 300 * fired, then it mustn't do anything. 301 */ 302 enum blk_eh_timer_return scsi_times_out(struct request *req) 303 { 304 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req); 305 enum blk_eh_timer_return rtn = BLK_EH_DONE; 306 struct Scsi_Host *host = scmd->device->host; 307 308 trace_scsi_dispatch_cmd_timeout(scmd); 309 scsi_log_completion(scmd, TIMEOUT_ERROR); 310 311 if (host->eh_deadline != -1 && !host->last_reset) 312 host->last_reset = jiffies; 313 314 if (host->hostt->eh_timed_out) 315 rtn = host->hostt->eh_timed_out(scmd); 316 317 if (rtn == BLK_EH_DONE) { 318 /* 319 * Set the command to complete first in order to prevent a real 320 * completion from releasing the command while error handling 321 * is using it. If the command was already completed, then the 322 * lower level driver beat the timeout handler, and it is safe 323 * to return without escalating error recovery. 324 * 325 * If timeout handling lost the race to a real completion, the 326 * block layer may ignore that due to a fake timeout injection, 327 * so return RESET_TIMER to allow error handling another shot 328 * at this command. 329 */ 330 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state)) 331 return BLK_EH_RESET_TIMER; 332 if (scsi_abort_command(scmd) != SUCCESS) { 333 set_host_byte(scmd, DID_TIME_OUT); 334 scsi_eh_scmd_add(scmd); 335 } 336 } 337 338 return rtn; 339 } 340 341 /** 342 * scsi_block_when_processing_errors - Prevent cmds from being queued. 343 * @sdev: Device on which we are performing recovery. 344 * 345 * Description: 346 * We block until the host is out of error recovery, and then check to 347 * see whether the host or the device is offline. 348 * 349 * Return value: 350 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 351 */ 352 int scsi_block_when_processing_errors(struct scsi_device *sdev) 353 { 354 int online; 355 356 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 357 358 online = scsi_device_online(sdev); 359 360 return online; 361 } 362 EXPORT_SYMBOL(scsi_block_when_processing_errors); 363 364 #ifdef CONFIG_SCSI_LOGGING 365 /** 366 * scsi_eh_prt_fail_stats - Log info on failures. 367 * @shost: scsi host being recovered. 368 * @work_q: Queue of scsi cmds to process. 369 */ 370 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 371 struct list_head *work_q) 372 { 373 struct scsi_cmnd *scmd; 374 struct scsi_device *sdev; 375 int total_failures = 0; 376 int cmd_failed = 0; 377 int cmd_cancel = 0; 378 int devices_failed = 0; 379 380 shost_for_each_device(sdev, shost) { 381 list_for_each_entry(scmd, work_q, eh_entry) { 382 if (scmd->device == sdev) { 383 ++total_failures; 384 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) 385 ++cmd_cancel; 386 else 387 ++cmd_failed; 388 } 389 } 390 391 if (cmd_cancel || cmd_failed) { 392 SCSI_LOG_ERROR_RECOVERY(3, 393 shost_printk(KERN_INFO, shost, 394 "%s: cmds failed: %d, cancel: %d\n", 395 __func__, cmd_failed, 396 cmd_cancel)); 397 cmd_cancel = 0; 398 cmd_failed = 0; 399 ++devices_failed; 400 } 401 } 402 403 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, 404 "Total of %d commands on %d" 405 " devices require eh work\n", 406 total_failures, devices_failed)); 407 } 408 #endif 409 410 /** 411 * scsi_report_lun_change - Set flag on all *other* devices on the same target 412 * to indicate that a UNIT ATTENTION is expected. 413 * @sdev: Device reporting the UNIT ATTENTION 414 */ 415 static void scsi_report_lun_change(struct scsi_device *sdev) 416 { 417 sdev->sdev_target->expecting_lun_change = 1; 418 } 419 420 /** 421 * scsi_report_sense - Examine scsi sense information and log messages for 422 * certain conditions, also issue uevents for some of them. 423 * @sdev: Device reporting the sense code 424 * @sshdr: sshdr to be examined 425 */ 426 static void scsi_report_sense(struct scsi_device *sdev, 427 struct scsi_sense_hdr *sshdr) 428 { 429 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ 430 431 if (sshdr->sense_key == UNIT_ATTENTION) { 432 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { 433 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; 434 sdev_printk(KERN_WARNING, sdev, 435 "Inquiry data has changed"); 436 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { 437 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; 438 scsi_report_lun_change(sdev); 439 sdev_printk(KERN_WARNING, sdev, 440 "Warning! Received an indication that the " 441 "LUN assignments on this target have " 442 "changed. The Linux SCSI layer does not " 443 "automatically remap LUN assignments.\n"); 444 } else if (sshdr->asc == 0x3f) 445 sdev_printk(KERN_WARNING, sdev, 446 "Warning! Received an indication that the " 447 "operating parameters on this target have " 448 "changed. The Linux SCSI layer does not " 449 "automatically adjust these parameters.\n"); 450 451 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { 452 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; 453 sdev_printk(KERN_WARNING, sdev, 454 "Warning! Received an indication that the " 455 "LUN reached a thin provisioning soft " 456 "threshold.\n"); 457 } 458 459 if (sshdr->asc == 0x29) { 460 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED; 461 sdev_printk(KERN_WARNING, sdev, 462 "Power-on or device reset occurred\n"); 463 } 464 465 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { 466 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; 467 sdev_printk(KERN_WARNING, sdev, 468 "Mode parameters changed"); 469 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { 470 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; 471 sdev_printk(KERN_WARNING, sdev, 472 "Asymmetric access state changed"); 473 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { 474 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; 475 sdev_printk(KERN_WARNING, sdev, 476 "Capacity data has changed"); 477 } else if (sshdr->asc == 0x2a) 478 sdev_printk(KERN_WARNING, sdev, 479 "Parameters changed"); 480 } 481 482 if (evt_type != SDEV_EVT_MAXBITS) { 483 set_bit(evt_type, sdev->pending_events); 484 schedule_work(&sdev->event_work); 485 } 486 } 487 488 /** 489 * scsi_check_sense - Examine scsi cmd sense 490 * @scmd: Cmd to have sense checked. 491 * 492 * Return value: 493 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE 494 * 495 * Notes: 496 * When a deferred error is detected the current command has 497 * not been executed and needs retrying. 498 */ 499 enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd) 500 { 501 struct scsi_device *sdev = scmd->device; 502 struct scsi_sense_hdr sshdr; 503 504 if (! scsi_command_normalize_sense(scmd, &sshdr)) 505 return FAILED; /* no valid sense data */ 506 507 scsi_report_sense(sdev, &sshdr); 508 509 if (scsi_sense_is_deferred(&sshdr)) 510 return NEEDS_RETRY; 511 512 if (sdev->handler && sdev->handler->check_sense) { 513 enum scsi_disposition rc; 514 515 rc = sdev->handler->check_sense(sdev, &sshdr); 516 if (rc != SCSI_RETURN_NOT_HANDLED) 517 return rc; 518 /* handler does not care. Drop down to default handling */ 519 } 520 521 if (scmd->cmnd[0] == TEST_UNIT_READY && 522 scmd->submitter != SUBMITTED_BY_SCSI_ERROR_HANDLER) 523 /* 524 * nasty: for mid-layer issued TURs, we need to return the 525 * actual sense data without any recovery attempt. For eh 526 * issued ones, we need to try to recover and interpret 527 */ 528 return SUCCESS; 529 530 /* 531 * Previous logic looked for FILEMARK, EOM or ILI which are 532 * mainly associated with tapes and returned SUCCESS. 533 */ 534 if (sshdr.response_code == 0x70) { 535 /* fixed format */ 536 if (scmd->sense_buffer[2] & 0xe0) 537 return SUCCESS; 538 } else { 539 /* 540 * descriptor format: look for "stream commands sense data 541 * descriptor" (see SSC-3). Assume single sense data 542 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 543 */ 544 if ((sshdr.additional_length > 3) && 545 (scmd->sense_buffer[8] == 0x4) && 546 (scmd->sense_buffer[11] & 0xe0)) 547 return SUCCESS; 548 } 549 550 switch (sshdr.sense_key) { 551 case NO_SENSE: 552 return SUCCESS; 553 case RECOVERED_ERROR: 554 return /* soft_error */ SUCCESS; 555 556 case ABORTED_COMMAND: 557 if (sshdr.asc == 0x10) /* DIF */ 558 return SUCCESS; 559 560 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF) 561 return ADD_TO_MLQUEUE; 562 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 && 563 sdev->sdev_bflags & BLIST_RETRY_ASC_C1) 564 return ADD_TO_MLQUEUE; 565 566 return NEEDS_RETRY; 567 case NOT_READY: 568 case UNIT_ATTENTION: 569 /* 570 * if we are expecting a cc/ua because of a bus reset that we 571 * performed, treat this just as a retry. otherwise this is 572 * information that we should pass up to the upper-level driver 573 * so that we can deal with it there. 574 */ 575 if (scmd->device->expecting_cc_ua) { 576 /* 577 * Because some device does not queue unit 578 * attentions correctly, we carefully check 579 * additional sense code and qualifier so as 580 * not to squash media change unit attention. 581 */ 582 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { 583 scmd->device->expecting_cc_ua = 0; 584 return NEEDS_RETRY; 585 } 586 } 587 /* 588 * we might also expect a cc/ua if another LUN on the target 589 * reported a UA with an ASC/ASCQ of 3F 0E - 590 * REPORTED LUNS DATA HAS CHANGED. 591 */ 592 if (scmd->device->sdev_target->expecting_lun_change && 593 sshdr.asc == 0x3f && sshdr.ascq == 0x0e) 594 return NEEDS_RETRY; 595 /* 596 * if the device is in the process of becoming ready, we 597 * should retry. 598 */ 599 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 600 return NEEDS_RETRY; 601 /* 602 * if the device is not started, we need to wake 603 * the error handler to start the motor 604 */ 605 if (scmd->device->allow_restart && 606 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 607 return FAILED; 608 /* 609 * Pass the UA upwards for a determination in the completion 610 * functions. 611 */ 612 return SUCCESS; 613 614 /* these are not supported */ 615 case DATA_PROTECT: 616 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { 617 /* Thin provisioning hard threshold reached */ 618 set_host_byte(scmd, DID_ALLOC_FAILURE); 619 return SUCCESS; 620 } 621 fallthrough; 622 case COPY_ABORTED: 623 case VOLUME_OVERFLOW: 624 case MISCOMPARE: 625 case BLANK_CHECK: 626 set_host_byte(scmd, DID_TARGET_FAILURE); 627 return SUCCESS; 628 629 case MEDIUM_ERROR: 630 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 631 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 632 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 633 set_host_byte(scmd, DID_MEDIUM_ERROR); 634 return SUCCESS; 635 } 636 return NEEDS_RETRY; 637 638 case HARDWARE_ERROR: 639 if (scmd->device->retry_hwerror) 640 return ADD_TO_MLQUEUE; 641 else 642 set_host_byte(scmd, DID_TARGET_FAILURE); 643 fallthrough; 644 645 case ILLEGAL_REQUEST: 646 if (sshdr.asc == 0x20 || /* Invalid command operation code */ 647 sshdr.asc == 0x21 || /* Logical block address out of range */ 648 sshdr.asc == 0x22 || /* Invalid function */ 649 sshdr.asc == 0x24 || /* Invalid field in cdb */ 650 sshdr.asc == 0x26 || /* Parameter value invalid */ 651 sshdr.asc == 0x27) { /* Write protected */ 652 set_host_byte(scmd, DID_TARGET_FAILURE); 653 } 654 return SUCCESS; 655 656 default: 657 return SUCCESS; 658 } 659 } 660 EXPORT_SYMBOL_GPL(scsi_check_sense); 661 662 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) 663 { 664 struct scsi_host_template *sht = sdev->host->hostt; 665 struct scsi_device *tmp_sdev; 666 667 if (!sht->track_queue_depth || 668 sdev->queue_depth >= sdev->max_queue_depth) 669 return; 670 671 if (time_before(jiffies, 672 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) 673 return; 674 675 if (time_before(jiffies, 676 sdev->last_queue_full_time + sdev->queue_ramp_up_period)) 677 return; 678 679 /* 680 * Walk all devices of a target and do 681 * ramp up on them. 682 */ 683 shost_for_each_device(tmp_sdev, sdev->host) { 684 if (tmp_sdev->channel != sdev->channel || 685 tmp_sdev->id != sdev->id || 686 tmp_sdev->queue_depth == sdev->max_queue_depth) 687 continue; 688 689 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); 690 sdev->last_queue_ramp_up = jiffies; 691 } 692 } 693 694 static void scsi_handle_queue_full(struct scsi_device *sdev) 695 { 696 struct scsi_host_template *sht = sdev->host->hostt; 697 struct scsi_device *tmp_sdev; 698 699 if (!sht->track_queue_depth) 700 return; 701 702 shost_for_each_device(tmp_sdev, sdev->host) { 703 if (tmp_sdev->channel != sdev->channel || 704 tmp_sdev->id != sdev->id) 705 continue; 706 /* 707 * We do not know the number of commands that were at 708 * the device when we got the queue full so we start 709 * from the highest possible value and work our way down. 710 */ 711 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); 712 } 713 } 714 715 /** 716 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 717 * @scmd: SCSI cmd to examine. 718 * 719 * Notes: 720 * This is *only* called when we are examining the status of commands 721 * queued during error recovery. the main difference here is that we 722 * don't allow for the possibility of retries here, and we are a lot 723 * more restrictive about what we consider acceptable. 724 */ 725 static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd) 726 { 727 /* 728 * first check the host byte, to see if there is anything in there 729 * that would indicate what we need to do. 730 */ 731 if (host_byte(scmd->result) == DID_RESET) { 732 /* 733 * rats. we are already in the error handler, so we now 734 * get to try and figure out what to do next. if the sense 735 * is valid, we have a pretty good idea of what to do. 736 * if not, we mark it as FAILED. 737 */ 738 return scsi_check_sense(scmd); 739 } 740 if (host_byte(scmd->result) != DID_OK) 741 return FAILED; 742 743 /* 744 * now, check the status byte to see if this indicates 745 * anything special. 746 */ 747 switch (get_status_byte(scmd)) { 748 case SAM_STAT_GOOD: 749 scsi_handle_queue_ramp_up(scmd->device); 750 fallthrough; 751 case SAM_STAT_COMMAND_TERMINATED: 752 return SUCCESS; 753 case SAM_STAT_CHECK_CONDITION: 754 return scsi_check_sense(scmd); 755 case SAM_STAT_CONDITION_MET: 756 case SAM_STAT_INTERMEDIATE: 757 case SAM_STAT_INTERMEDIATE_CONDITION_MET: 758 /* 759 * who knows? FIXME(eric) 760 */ 761 return SUCCESS; 762 case SAM_STAT_RESERVATION_CONFLICT: 763 if (scmd->cmnd[0] == TEST_UNIT_READY) 764 /* it is a success, we probed the device and 765 * found it */ 766 return SUCCESS; 767 /* otherwise, we failed to send the command */ 768 return FAILED; 769 case SAM_STAT_TASK_SET_FULL: 770 scsi_handle_queue_full(scmd->device); 771 fallthrough; 772 case SAM_STAT_BUSY: 773 return NEEDS_RETRY; 774 default: 775 return FAILED; 776 } 777 return FAILED; 778 } 779 780 /** 781 * scsi_eh_done - Completion function for error handling. 782 * @scmd: Cmd that is done. 783 */ 784 void scsi_eh_done(struct scsi_cmnd *scmd) 785 { 786 struct completion *eh_action; 787 788 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 789 "%s result: %x\n", __func__, scmd->result)); 790 791 eh_action = scmd->device->host->eh_action; 792 if (eh_action) 793 complete(eh_action); 794 } 795 796 /** 797 * scsi_try_host_reset - ask host adapter to reset itself 798 * @scmd: SCSI cmd to send host reset. 799 */ 800 static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd) 801 { 802 unsigned long flags; 803 enum scsi_disposition rtn; 804 struct Scsi_Host *host = scmd->device->host; 805 struct scsi_host_template *hostt = host->hostt; 806 807 SCSI_LOG_ERROR_RECOVERY(3, 808 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 809 810 if (!hostt->eh_host_reset_handler) 811 return FAILED; 812 813 rtn = hostt->eh_host_reset_handler(scmd); 814 815 if (rtn == SUCCESS) { 816 if (!hostt->skip_settle_delay) 817 ssleep(HOST_RESET_SETTLE_TIME); 818 spin_lock_irqsave(host->host_lock, flags); 819 scsi_report_bus_reset(host, scmd_channel(scmd)); 820 spin_unlock_irqrestore(host->host_lock, flags); 821 } 822 823 return rtn; 824 } 825 826 /** 827 * scsi_try_bus_reset - ask host to perform a bus reset 828 * @scmd: SCSI cmd to send bus reset. 829 */ 830 static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd) 831 { 832 unsigned long flags; 833 enum scsi_disposition rtn; 834 struct Scsi_Host *host = scmd->device->host; 835 struct scsi_host_template *hostt = host->hostt; 836 837 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 838 "%s: Snd Bus RST\n", __func__)); 839 840 if (!hostt->eh_bus_reset_handler) 841 return FAILED; 842 843 rtn = hostt->eh_bus_reset_handler(scmd); 844 845 if (rtn == SUCCESS) { 846 if (!hostt->skip_settle_delay) 847 ssleep(BUS_RESET_SETTLE_TIME); 848 spin_lock_irqsave(host->host_lock, flags); 849 scsi_report_bus_reset(host, scmd_channel(scmd)); 850 spin_unlock_irqrestore(host->host_lock, flags); 851 } 852 853 return rtn; 854 } 855 856 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 857 { 858 sdev->was_reset = 1; 859 sdev->expecting_cc_ua = 1; 860 } 861 862 /** 863 * scsi_try_target_reset - Ask host to perform a target reset 864 * @scmd: SCSI cmd used to send a target reset 865 * 866 * Notes: 867 * There is no timeout for this operation. if this operation is 868 * unreliable for a given host, then the host itself needs to put a 869 * timer on it, and set the host back to a consistent state prior to 870 * returning. 871 */ 872 static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd) 873 { 874 unsigned long flags; 875 enum scsi_disposition rtn; 876 struct Scsi_Host *host = scmd->device->host; 877 struct scsi_host_template *hostt = host->hostt; 878 879 if (!hostt->eh_target_reset_handler) 880 return FAILED; 881 882 rtn = hostt->eh_target_reset_handler(scmd); 883 if (rtn == SUCCESS) { 884 spin_lock_irqsave(host->host_lock, flags); 885 __starget_for_each_device(scsi_target(scmd->device), NULL, 886 __scsi_report_device_reset); 887 spin_unlock_irqrestore(host->host_lock, flags); 888 } 889 890 return rtn; 891 } 892 893 /** 894 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 895 * @scmd: SCSI cmd used to send BDR 896 * 897 * Notes: 898 * There is no timeout for this operation. if this operation is 899 * unreliable for a given host, then the host itself needs to put a 900 * timer on it, and set the host back to a consistent state prior to 901 * returning. 902 */ 903 static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 904 { 905 enum scsi_disposition rtn; 906 struct scsi_host_template *hostt = scmd->device->host->hostt; 907 908 if (!hostt->eh_device_reset_handler) 909 return FAILED; 910 911 rtn = hostt->eh_device_reset_handler(scmd); 912 if (rtn == SUCCESS) 913 __scsi_report_device_reset(scmd->device, NULL); 914 return rtn; 915 } 916 917 /** 918 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command 919 * @hostt: SCSI driver host template 920 * @scmd: SCSI cmd used to send a target reset 921 * 922 * Return value: 923 * SUCCESS, FAILED, or FAST_IO_FAIL 924 * 925 * Notes: 926 * SUCCESS does not necessarily indicate that the command 927 * has been aborted; it only indicates that the LLDDs 928 * has cleared all references to that command. 929 * LLDDs should return FAILED only if an abort was required 930 * but could not be executed. LLDDs should return FAST_IO_FAIL 931 * if the device is temporarily unavailable (eg due to a 932 * link down on FibreChannel) 933 */ 934 static enum scsi_disposition 935 scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd) 936 { 937 if (!hostt->eh_abort_handler) 938 return FAILED; 939 940 return hostt->eh_abort_handler(scmd); 941 } 942 943 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 944 { 945 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 946 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 947 if (scsi_try_target_reset(scmd) != SUCCESS) 948 if (scsi_try_bus_reset(scmd) != SUCCESS) 949 scsi_try_host_reset(scmd); 950 } 951 952 /** 953 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 954 * @scmd: SCSI command structure to hijack 955 * @ses: structure to save restore information 956 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 957 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) 958 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 959 * 960 * This function is used to save a scsi command information before re-execution 961 * as part of the error recovery process. If @sense_bytes is 0 the command 962 * sent must be one that does not transfer any data. If @sense_bytes != 0 963 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 964 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 965 */ 966 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 967 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 968 { 969 struct scsi_device *sdev = scmd->device; 970 971 /* 972 * We need saved copies of a number of fields - this is because 973 * error handling may need to overwrite these with different values 974 * to run different commands, and once error handling is complete, 975 * we will need to restore these values prior to running the actual 976 * command. 977 */ 978 ses->cmd_len = scmd->cmd_len; 979 ses->cmnd = scmd->cmnd; 980 ses->data_direction = scmd->sc_data_direction; 981 ses->sdb = scmd->sdb; 982 ses->result = scmd->result; 983 ses->resid_len = scmd->req.resid_len; 984 ses->underflow = scmd->underflow; 985 ses->prot_op = scmd->prot_op; 986 ses->eh_eflags = scmd->eh_eflags; 987 988 scmd->prot_op = SCSI_PROT_NORMAL; 989 scmd->eh_eflags = 0; 990 scmd->cmnd = ses->eh_cmnd; 991 memset(scmd->cmnd, 0, BLK_MAX_CDB); 992 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 993 scmd->result = 0; 994 scmd->req.resid_len = 0; 995 996 if (sense_bytes) { 997 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 998 sense_bytes); 999 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 1000 scmd->sdb.length); 1001 scmd->sdb.table.sgl = &ses->sense_sgl; 1002 scmd->sc_data_direction = DMA_FROM_DEVICE; 1003 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; 1004 scmd->cmnd[0] = REQUEST_SENSE; 1005 scmd->cmnd[4] = scmd->sdb.length; 1006 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1007 } else { 1008 scmd->sc_data_direction = DMA_NONE; 1009 if (cmnd) { 1010 BUG_ON(cmnd_size > BLK_MAX_CDB); 1011 memcpy(scmd->cmnd, cmnd, cmnd_size); 1012 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1013 } 1014 } 1015 1016 scmd->underflow = 0; 1017 1018 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 1019 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 1020 (sdev->lun << 5 & 0xe0); 1021 1022 /* 1023 * Zero the sense buffer. The scsi spec mandates that any 1024 * untransferred sense data should be interpreted as being zero. 1025 */ 1026 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 1027 } 1028 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 1029 1030 /** 1031 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 1032 * @scmd: SCSI command structure to restore 1033 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 1034 * 1035 * Undo any damage done by above scsi_eh_prep_cmnd(). 1036 */ 1037 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 1038 { 1039 /* 1040 * Restore original data 1041 */ 1042 scmd->cmd_len = ses->cmd_len; 1043 scmd->cmnd = ses->cmnd; 1044 scmd->sc_data_direction = ses->data_direction; 1045 scmd->sdb = ses->sdb; 1046 scmd->result = ses->result; 1047 scmd->req.resid_len = ses->resid_len; 1048 scmd->underflow = ses->underflow; 1049 scmd->prot_op = ses->prot_op; 1050 scmd->eh_eflags = ses->eh_eflags; 1051 } 1052 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 1053 1054 /** 1055 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 1056 * @scmd: SCSI command structure to hijack 1057 * @cmnd: CDB to send 1058 * @cmnd_size: size in bytes of @cmnd 1059 * @timeout: timeout for this request 1060 * @sense_bytes: size of sense data to copy or 0 1061 * 1062 * This function is used to send a scsi command down to a target device 1063 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 1064 * 1065 * Return value: 1066 * SUCCESS or FAILED or NEEDS_RETRY 1067 */ 1068 static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd, 1069 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes) 1070 { 1071 struct scsi_device *sdev = scmd->device; 1072 struct Scsi_Host *shost = sdev->host; 1073 DECLARE_COMPLETION_ONSTACK(done); 1074 unsigned long timeleft = timeout, delay; 1075 struct scsi_eh_save ses; 1076 const unsigned long stall_for = msecs_to_jiffies(100); 1077 int rtn; 1078 1079 retry: 1080 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1081 shost->eh_action = &done; 1082 1083 scsi_log_send(scmd); 1084 scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER; 1085 1086 /* 1087 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can 1088 * change the SCSI device state after we have examined it and before 1089 * .queuecommand() is called. 1090 */ 1091 mutex_lock(&sdev->state_mutex); 1092 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) { 1093 mutex_unlock(&sdev->state_mutex); 1094 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev, 1095 "%s: state %d <> %d\n", __func__, sdev->sdev_state, 1096 SDEV_BLOCK)); 1097 delay = min(timeleft, stall_for); 1098 timeleft -= delay; 1099 msleep(jiffies_to_msecs(delay)); 1100 mutex_lock(&sdev->state_mutex); 1101 } 1102 if (sdev->sdev_state != SDEV_BLOCK) 1103 rtn = shost->hostt->queuecommand(shost, scmd); 1104 else 1105 rtn = FAILED; 1106 mutex_unlock(&sdev->state_mutex); 1107 1108 if (rtn) { 1109 if (timeleft > stall_for) { 1110 scsi_eh_restore_cmnd(scmd, &ses); 1111 1112 timeleft -= stall_for; 1113 msleep(jiffies_to_msecs(stall_for)); 1114 goto retry; 1115 } 1116 /* signal not to enter either branch of the if () below */ 1117 timeleft = 0; 1118 rtn = FAILED; 1119 } else { 1120 timeleft = wait_for_completion_timeout(&done, timeout); 1121 rtn = SUCCESS; 1122 } 1123 1124 shost->eh_action = NULL; 1125 1126 scsi_log_completion(scmd, rtn); 1127 1128 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1129 "%s timeleft: %ld\n", 1130 __func__, timeleft)); 1131 1132 /* 1133 * If there is time left scsi_eh_done got called, and we will examine 1134 * the actual status codes to see whether the command actually did 1135 * complete normally, else if we have a zero return and no time left, 1136 * the command must still be pending, so abort it and return FAILED. 1137 * If we never actually managed to issue the command, because 1138 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1139 * value above (so don't execute either branch of the if) 1140 */ 1141 if (timeleft) { 1142 rtn = scsi_eh_completed_normally(scmd); 1143 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1144 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1145 1146 switch (rtn) { 1147 case SUCCESS: 1148 case NEEDS_RETRY: 1149 case FAILED: 1150 break; 1151 case ADD_TO_MLQUEUE: 1152 rtn = NEEDS_RETRY; 1153 break; 1154 default: 1155 rtn = FAILED; 1156 break; 1157 } 1158 } else if (rtn != FAILED) { 1159 scsi_abort_eh_cmnd(scmd); 1160 rtn = FAILED; 1161 } 1162 1163 scsi_eh_restore_cmnd(scmd, &ses); 1164 1165 return rtn; 1166 } 1167 1168 /** 1169 * scsi_request_sense - Request sense data from a particular target. 1170 * @scmd: SCSI cmd for request sense. 1171 * 1172 * Notes: 1173 * Some hosts automatically obtain this information, others require 1174 * that we obtain it on our own. This function will *not* return until 1175 * the command either times out, or it completes. 1176 */ 1177 static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd) 1178 { 1179 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1180 } 1181 1182 static enum scsi_disposition 1183 scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn) 1184 { 1185 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) { 1186 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1187 if (sdrv->eh_action) 1188 rtn = sdrv->eh_action(scmd, rtn); 1189 } 1190 return rtn; 1191 } 1192 1193 /** 1194 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1195 * @scmd: Original SCSI cmd that eh has finished. 1196 * @done_q: Queue for processed commands. 1197 * 1198 * Notes: 1199 * We don't want to use the normal command completion while we are are 1200 * still handling errors - it may cause other commands to be queued, 1201 * and that would disturb what we are doing. Thus we really want to 1202 * keep a list of pending commands for final completion, and once we 1203 * are ready to leave error handling we handle completion for real. 1204 */ 1205 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1206 { 1207 list_move_tail(&scmd->eh_entry, done_q); 1208 } 1209 EXPORT_SYMBOL(scsi_eh_finish_cmd); 1210 1211 /** 1212 * scsi_eh_get_sense - Get device sense data. 1213 * @work_q: Queue of commands to process. 1214 * @done_q: Queue of processed commands. 1215 * 1216 * Description: 1217 * See if we need to request sense information. if so, then get it 1218 * now, so we have a better idea of what to do. 1219 * 1220 * Notes: 1221 * This has the unfortunate side effect that if a shost adapter does 1222 * not automatically request sense information, we end up shutting 1223 * it down before we request it. 1224 * 1225 * All drivers should request sense information internally these days, 1226 * so for now all I have to say is tough noogies if you end up in here. 1227 * 1228 * XXX: Long term this code should go away, but that needs an audit of 1229 * all LLDDs first. 1230 */ 1231 int scsi_eh_get_sense(struct list_head *work_q, 1232 struct list_head *done_q) 1233 { 1234 struct scsi_cmnd *scmd, *next; 1235 struct Scsi_Host *shost; 1236 enum scsi_disposition rtn; 1237 1238 /* 1239 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, 1240 * should not get sense. 1241 */ 1242 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1243 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || 1244 SCSI_SENSE_VALID(scmd)) 1245 continue; 1246 1247 shost = scmd->device->host; 1248 if (scsi_host_eh_past_deadline(shost)) { 1249 SCSI_LOG_ERROR_RECOVERY(3, 1250 scmd_printk(KERN_INFO, scmd, 1251 "%s: skip request sense, past eh deadline\n", 1252 current->comm)); 1253 break; 1254 } 1255 if (!scsi_status_is_check_condition(scmd->result)) 1256 /* 1257 * don't request sense if there's no check condition 1258 * status because the error we're processing isn't one 1259 * that has a sense code (and some devices get 1260 * confused by sense requests out of the blue) 1261 */ 1262 continue; 1263 1264 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1265 "%s: requesting sense\n", 1266 current->comm)); 1267 rtn = scsi_request_sense(scmd); 1268 if (rtn != SUCCESS) 1269 continue; 1270 1271 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1272 "sense requested, result %x\n", scmd->result)); 1273 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); 1274 1275 rtn = scsi_decide_disposition(scmd); 1276 1277 /* 1278 * if the result was normal, then just pass it along to the 1279 * upper level. 1280 */ 1281 if (rtn == SUCCESS) 1282 /* 1283 * We don't want this command reissued, just finished 1284 * with the sense data, so set retries to the max 1285 * allowed to ensure it won't get reissued. If the user 1286 * has requested infinite retries, we also want to 1287 * finish this command, so force completion by setting 1288 * retries and allowed to the same value. 1289 */ 1290 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) 1291 scmd->retries = scmd->allowed = 1; 1292 else 1293 scmd->retries = scmd->allowed; 1294 else if (rtn != NEEDS_RETRY) 1295 continue; 1296 1297 scsi_eh_finish_cmd(scmd, done_q); 1298 } 1299 1300 return list_empty(work_q); 1301 } 1302 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1303 1304 /** 1305 * scsi_eh_tur - Send TUR to device. 1306 * @scmd: &scsi_cmnd to send TUR 1307 * 1308 * Return value: 1309 * 0 - Device is ready. 1 - Device NOT ready. 1310 */ 1311 static int scsi_eh_tur(struct scsi_cmnd *scmd) 1312 { 1313 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1314 int retry_cnt = 1; 1315 enum scsi_disposition rtn; 1316 1317 retry_tur: 1318 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1319 scmd->device->eh_timeout, 0); 1320 1321 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1322 "%s return: %x\n", __func__, rtn)); 1323 1324 switch (rtn) { 1325 case NEEDS_RETRY: 1326 if (retry_cnt--) 1327 goto retry_tur; 1328 fallthrough; 1329 case SUCCESS: 1330 return 0; 1331 default: 1332 return 1; 1333 } 1334 } 1335 1336 /** 1337 * scsi_eh_test_devices - check if devices are responding from error recovery. 1338 * @cmd_list: scsi commands in error recovery. 1339 * @work_q: queue for commands which still need more error recovery 1340 * @done_q: queue for commands which are finished 1341 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1342 * 1343 * Decription: 1344 * Tests if devices are in a working state. Commands to devices now in 1345 * a working state are sent to the done_q while commands to devices which 1346 * are still failing to respond are returned to the work_q for more 1347 * processing. 1348 **/ 1349 static int scsi_eh_test_devices(struct list_head *cmd_list, 1350 struct list_head *work_q, 1351 struct list_head *done_q, int try_stu) 1352 { 1353 struct scsi_cmnd *scmd, *next; 1354 struct scsi_device *sdev; 1355 int finish_cmds; 1356 1357 while (!list_empty(cmd_list)) { 1358 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1359 sdev = scmd->device; 1360 1361 if (!try_stu) { 1362 if (scsi_host_eh_past_deadline(sdev->host)) { 1363 /* Push items back onto work_q */ 1364 list_splice_init(cmd_list, work_q); 1365 SCSI_LOG_ERROR_RECOVERY(3, 1366 sdev_printk(KERN_INFO, sdev, 1367 "%s: skip test device, past eh deadline", 1368 current->comm)); 1369 break; 1370 } 1371 } 1372 1373 finish_cmds = !scsi_device_online(scmd->device) || 1374 (try_stu && !scsi_eh_try_stu(scmd) && 1375 !scsi_eh_tur(scmd)) || 1376 !scsi_eh_tur(scmd); 1377 1378 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1379 if (scmd->device == sdev) { 1380 if (finish_cmds && 1381 (try_stu || 1382 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1383 scsi_eh_finish_cmd(scmd, done_q); 1384 else 1385 list_move_tail(&scmd->eh_entry, work_q); 1386 } 1387 } 1388 return list_empty(work_q); 1389 } 1390 1391 /** 1392 * scsi_eh_try_stu - Send START_UNIT to device. 1393 * @scmd: &scsi_cmnd to send START_UNIT 1394 * 1395 * Return value: 1396 * 0 - Device is ready. 1 - Device NOT ready. 1397 */ 1398 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1399 { 1400 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1401 1402 if (scmd->device->allow_restart) { 1403 int i; 1404 enum scsi_disposition rtn = NEEDS_RETRY; 1405 1406 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1407 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0); 1408 1409 if (rtn == SUCCESS) 1410 return 0; 1411 } 1412 1413 return 1; 1414 } 1415 1416 /** 1417 * scsi_eh_stu - send START_UNIT if needed 1418 * @shost: &scsi host being recovered. 1419 * @work_q: &list_head for pending commands. 1420 * @done_q: &list_head for processed commands. 1421 * 1422 * Notes: 1423 * If commands are failing due to not ready, initializing command required, 1424 * try revalidating the device, which will end up sending a start unit. 1425 */ 1426 static int scsi_eh_stu(struct Scsi_Host *shost, 1427 struct list_head *work_q, 1428 struct list_head *done_q) 1429 { 1430 struct scsi_cmnd *scmd, *stu_scmd, *next; 1431 struct scsi_device *sdev; 1432 1433 shost_for_each_device(sdev, shost) { 1434 if (scsi_host_eh_past_deadline(shost)) { 1435 SCSI_LOG_ERROR_RECOVERY(3, 1436 sdev_printk(KERN_INFO, sdev, 1437 "%s: skip START_UNIT, past eh deadline\n", 1438 current->comm)); 1439 scsi_device_put(sdev); 1440 break; 1441 } 1442 stu_scmd = NULL; 1443 list_for_each_entry(scmd, work_q, eh_entry) 1444 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1445 scsi_check_sense(scmd) == FAILED ) { 1446 stu_scmd = scmd; 1447 break; 1448 } 1449 1450 if (!stu_scmd) 1451 continue; 1452 1453 SCSI_LOG_ERROR_RECOVERY(3, 1454 sdev_printk(KERN_INFO, sdev, 1455 "%s: Sending START_UNIT\n", 1456 current->comm)); 1457 1458 if (!scsi_eh_try_stu(stu_scmd)) { 1459 if (!scsi_device_online(sdev) || 1460 !scsi_eh_tur(stu_scmd)) { 1461 list_for_each_entry_safe(scmd, next, 1462 work_q, eh_entry) { 1463 if (scmd->device == sdev && 1464 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1465 scsi_eh_finish_cmd(scmd, done_q); 1466 } 1467 } 1468 } else { 1469 SCSI_LOG_ERROR_RECOVERY(3, 1470 sdev_printk(KERN_INFO, sdev, 1471 "%s: START_UNIT failed\n", 1472 current->comm)); 1473 } 1474 } 1475 1476 return list_empty(work_q); 1477 } 1478 1479 1480 /** 1481 * scsi_eh_bus_device_reset - send bdr if needed 1482 * @shost: scsi host being recovered. 1483 * @work_q: &list_head for pending commands. 1484 * @done_q: &list_head for processed commands. 1485 * 1486 * Notes: 1487 * Try a bus device reset. Still, look to see whether we have multiple 1488 * devices that are jammed or not - if we have multiple devices, it 1489 * makes no sense to try bus_device_reset - we really would need to try 1490 * a bus_reset instead. 1491 */ 1492 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1493 struct list_head *work_q, 1494 struct list_head *done_q) 1495 { 1496 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1497 struct scsi_device *sdev; 1498 enum scsi_disposition rtn; 1499 1500 shost_for_each_device(sdev, shost) { 1501 if (scsi_host_eh_past_deadline(shost)) { 1502 SCSI_LOG_ERROR_RECOVERY(3, 1503 sdev_printk(KERN_INFO, sdev, 1504 "%s: skip BDR, past eh deadline\n", 1505 current->comm)); 1506 scsi_device_put(sdev); 1507 break; 1508 } 1509 bdr_scmd = NULL; 1510 list_for_each_entry(scmd, work_q, eh_entry) 1511 if (scmd->device == sdev) { 1512 bdr_scmd = scmd; 1513 break; 1514 } 1515 1516 if (!bdr_scmd) 1517 continue; 1518 1519 SCSI_LOG_ERROR_RECOVERY(3, 1520 sdev_printk(KERN_INFO, sdev, 1521 "%s: Sending BDR\n", current->comm)); 1522 rtn = scsi_try_bus_device_reset(bdr_scmd); 1523 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1524 if (!scsi_device_online(sdev) || 1525 rtn == FAST_IO_FAIL || 1526 !scsi_eh_tur(bdr_scmd)) { 1527 list_for_each_entry_safe(scmd, next, 1528 work_q, eh_entry) { 1529 if (scmd->device == sdev && 1530 scsi_eh_action(scmd, rtn) != FAILED) 1531 scsi_eh_finish_cmd(scmd, 1532 done_q); 1533 } 1534 } 1535 } else { 1536 SCSI_LOG_ERROR_RECOVERY(3, 1537 sdev_printk(KERN_INFO, sdev, 1538 "%s: BDR failed\n", current->comm)); 1539 } 1540 } 1541 1542 return list_empty(work_q); 1543 } 1544 1545 /** 1546 * scsi_eh_target_reset - send target reset if needed 1547 * @shost: scsi host being recovered. 1548 * @work_q: &list_head for pending commands. 1549 * @done_q: &list_head for processed commands. 1550 * 1551 * Notes: 1552 * Try a target reset. 1553 */ 1554 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1555 struct list_head *work_q, 1556 struct list_head *done_q) 1557 { 1558 LIST_HEAD(tmp_list); 1559 LIST_HEAD(check_list); 1560 1561 list_splice_init(work_q, &tmp_list); 1562 1563 while (!list_empty(&tmp_list)) { 1564 struct scsi_cmnd *next, *scmd; 1565 enum scsi_disposition rtn; 1566 unsigned int id; 1567 1568 if (scsi_host_eh_past_deadline(shost)) { 1569 /* push back on work queue for further processing */ 1570 list_splice_init(&check_list, work_q); 1571 list_splice_init(&tmp_list, work_q); 1572 SCSI_LOG_ERROR_RECOVERY(3, 1573 shost_printk(KERN_INFO, shost, 1574 "%s: Skip target reset, past eh deadline\n", 1575 current->comm)); 1576 return list_empty(work_q); 1577 } 1578 1579 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1580 id = scmd_id(scmd); 1581 1582 SCSI_LOG_ERROR_RECOVERY(3, 1583 shost_printk(KERN_INFO, shost, 1584 "%s: Sending target reset to target %d\n", 1585 current->comm, id)); 1586 rtn = scsi_try_target_reset(scmd); 1587 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1588 SCSI_LOG_ERROR_RECOVERY(3, 1589 shost_printk(KERN_INFO, shost, 1590 "%s: Target reset failed" 1591 " target: %d\n", 1592 current->comm, id)); 1593 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1594 if (scmd_id(scmd) != id) 1595 continue; 1596 1597 if (rtn == SUCCESS) 1598 list_move_tail(&scmd->eh_entry, &check_list); 1599 else if (rtn == FAST_IO_FAIL) 1600 scsi_eh_finish_cmd(scmd, done_q); 1601 else 1602 /* push back on work queue for further processing */ 1603 list_move(&scmd->eh_entry, work_q); 1604 } 1605 } 1606 1607 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1608 } 1609 1610 /** 1611 * scsi_eh_bus_reset - send a bus reset 1612 * @shost: &scsi host being recovered. 1613 * @work_q: &list_head for pending commands. 1614 * @done_q: &list_head for processed commands. 1615 */ 1616 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1617 struct list_head *work_q, 1618 struct list_head *done_q) 1619 { 1620 struct scsi_cmnd *scmd, *chan_scmd, *next; 1621 LIST_HEAD(check_list); 1622 unsigned int channel; 1623 enum scsi_disposition rtn; 1624 1625 /* 1626 * we really want to loop over the various channels, and do this on 1627 * a channel by channel basis. we should also check to see if any 1628 * of the failed commands are on soft_reset devices, and if so, skip 1629 * the reset. 1630 */ 1631 1632 for (channel = 0; channel <= shost->max_channel; channel++) { 1633 if (scsi_host_eh_past_deadline(shost)) { 1634 list_splice_init(&check_list, work_q); 1635 SCSI_LOG_ERROR_RECOVERY(3, 1636 shost_printk(KERN_INFO, shost, 1637 "%s: skip BRST, past eh deadline\n", 1638 current->comm)); 1639 return list_empty(work_q); 1640 } 1641 1642 chan_scmd = NULL; 1643 list_for_each_entry(scmd, work_q, eh_entry) { 1644 if (channel == scmd_channel(scmd)) { 1645 chan_scmd = scmd; 1646 break; 1647 /* 1648 * FIXME add back in some support for 1649 * soft_reset devices. 1650 */ 1651 } 1652 } 1653 1654 if (!chan_scmd) 1655 continue; 1656 SCSI_LOG_ERROR_RECOVERY(3, 1657 shost_printk(KERN_INFO, shost, 1658 "%s: Sending BRST chan: %d\n", 1659 current->comm, channel)); 1660 rtn = scsi_try_bus_reset(chan_scmd); 1661 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1662 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1663 if (channel == scmd_channel(scmd)) { 1664 if (rtn == FAST_IO_FAIL) 1665 scsi_eh_finish_cmd(scmd, 1666 done_q); 1667 else 1668 list_move_tail(&scmd->eh_entry, 1669 &check_list); 1670 } 1671 } 1672 } else { 1673 SCSI_LOG_ERROR_RECOVERY(3, 1674 shost_printk(KERN_INFO, shost, 1675 "%s: BRST failed chan: %d\n", 1676 current->comm, channel)); 1677 } 1678 } 1679 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1680 } 1681 1682 /** 1683 * scsi_eh_host_reset - send a host reset 1684 * @shost: host to be reset. 1685 * @work_q: &list_head for pending commands. 1686 * @done_q: &list_head for processed commands. 1687 */ 1688 static int scsi_eh_host_reset(struct Scsi_Host *shost, 1689 struct list_head *work_q, 1690 struct list_head *done_q) 1691 { 1692 struct scsi_cmnd *scmd, *next; 1693 LIST_HEAD(check_list); 1694 enum scsi_disposition rtn; 1695 1696 if (!list_empty(work_q)) { 1697 scmd = list_entry(work_q->next, 1698 struct scsi_cmnd, eh_entry); 1699 1700 SCSI_LOG_ERROR_RECOVERY(3, 1701 shost_printk(KERN_INFO, shost, 1702 "%s: Sending HRST\n", 1703 current->comm)); 1704 1705 rtn = scsi_try_host_reset(scmd); 1706 if (rtn == SUCCESS) { 1707 list_splice_init(work_q, &check_list); 1708 } else if (rtn == FAST_IO_FAIL) { 1709 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1710 scsi_eh_finish_cmd(scmd, done_q); 1711 } 1712 } else { 1713 SCSI_LOG_ERROR_RECOVERY(3, 1714 shost_printk(KERN_INFO, shost, 1715 "%s: HRST failed\n", 1716 current->comm)); 1717 } 1718 } 1719 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1720 } 1721 1722 /** 1723 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1724 * @work_q: &list_head for pending commands. 1725 * @done_q: &list_head for processed commands. 1726 */ 1727 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1728 struct list_head *done_q) 1729 { 1730 struct scsi_cmnd *scmd, *next; 1731 struct scsi_device *sdev; 1732 1733 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1734 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1735 "not ready after error recovery\n"); 1736 sdev = scmd->device; 1737 1738 mutex_lock(&sdev->state_mutex); 1739 scsi_device_set_state(sdev, SDEV_OFFLINE); 1740 mutex_unlock(&sdev->state_mutex); 1741 1742 scsi_eh_finish_cmd(scmd, done_q); 1743 } 1744 return; 1745 } 1746 1747 /** 1748 * scsi_noretry_cmd - determine if command should be failed fast 1749 * @scmd: SCSI cmd to examine. 1750 */ 1751 int scsi_noretry_cmd(struct scsi_cmnd *scmd) 1752 { 1753 struct request *req = scsi_cmd_to_rq(scmd); 1754 1755 switch (host_byte(scmd->result)) { 1756 case DID_OK: 1757 break; 1758 case DID_TIME_OUT: 1759 goto check_type; 1760 case DID_BUS_BUSY: 1761 return req->cmd_flags & REQ_FAILFAST_TRANSPORT; 1762 case DID_PARITY: 1763 return req->cmd_flags & REQ_FAILFAST_DEV; 1764 case DID_ERROR: 1765 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT) 1766 return 0; 1767 fallthrough; 1768 case DID_SOFT_ERROR: 1769 return req->cmd_flags & REQ_FAILFAST_DRIVER; 1770 } 1771 1772 if (!scsi_status_is_check_condition(scmd->result)) 1773 return 0; 1774 1775 check_type: 1776 /* 1777 * assume caller has checked sense and determined 1778 * the check condition was retryable. 1779 */ 1780 if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(req)) 1781 return 1; 1782 1783 return 0; 1784 } 1785 1786 /** 1787 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1788 * @scmd: SCSI cmd to examine. 1789 * 1790 * Notes: 1791 * This is *only* called when we are examining the status after sending 1792 * out the actual data command. any commands that are queued for error 1793 * recovery (e.g. test_unit_ready) do *not* come through here. 1794 * 1795 * When this routine returns failed, it means the error handler thread 1796 * is woken. In cases where the error code indicates an error that 1797 * doesn't require the error handler read (i.e. we don't need to 1798 * abort/reset), this function should return SUCCESS. 1799 */ 1800 enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd) 1801 { 1802 enum scsi_disposition rtn; 1803 1804 /* 1805 * if the device is offline, then we clearly just pass the result back 1806 * up to the top level. 1807 */ 1808 if (!scsi_device_online(scmd->device)) { 1809 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1810 "%s: device offline - report as SUCCESS\n", __func__)); 1811 return SUCCESS; 1812 } 1813 1814 /* 1815 * first check the host byte, to see if there is anything in there 1816 * that would indicate what we need to do. 1817 */ 1818 switch (host_byte(scmd->result)) { 1819 case DID_PASSTHROUGH: 1820 /* 1821 * no matter what, pass this through to the upper layer. 1822 * nuke this special code so that it looks like we are saying 1823 * did_ok. 1824 */ 1825 scmd->result &= 0xff00ffff; 1826 return SUCCESS; 1827 case DID_OK: 1828 /* 1829 * looks good. drop through, and check the next byte. 1830 */ 1831 break; 1832 case DID_ABORT: 1833 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1834 set_host_byte(scmd, DID_TIME_OUT); 1835 return SUCCESS; 1836 } 1837 fallthrough; 1838 case DID_NO_CONNECT: 1839 case DID_BAD_TARGET: 1840 /* 1841 * note - this means that we just report the status back 1842 * to the top level driver, not that we actually think 1843 * that it indicates SUCCESS. 1844 */ 1845 return SUCCESS; 1846 case DID_SOFT_ERROR: 1847 /* 1848 * when the low level driver returns did_soft_error, 1849 * it is responsible for keeping an internal retry counter 1850 * in order to avoid endless loops (db) 1851 */ 1852 goto maybe_retry; 1853 case DID_IMM_RETRY: 1854 return NEEDS_RETRY; 1855 1856 case DID_REQUEUE: 1857 return ADD_TO_MLQUEUE; 1858 case DID_TRANSPORT_DISRUPTED: 1859 /* 1860 * LLD/transport was disrupted during processing of the IO. 1861 * The transport class is now blocked/blocking, 1862 * and the transport will decide what to do with the IO 1863 * based on its timers and recovery capablilities if 1864 * there are enough retries. 1865 */ 1866 goto maybe_retry; 1867 case DID_TRANSPORT_FAILFAST: 1868 /* 1869 * The transport decided to failfast the IO (most likely 1870 * the fast io fail tmo fired), so send IO directly upwards. 1871 */ 1872 return SUCCESS; 1873 case DID_TRANSPORT_MARGINAL: 1874 /* 1875 * caller has decided not to do retries on 1876 * abort success, so send IO directly upwards 1877 */ 1878 return SUCCESS; 1879 case DID_ERROR: 1880 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT) 1881 /* 1882 * execute reservation conflict processing code 1883 * lower down 1884 */ 1885 break; 1886 fallthrough; 1887 case DID_BUS_BUSY: 1888 case DID_PARITY: 1889 goto maybe_retry; 1890 case DID_TIME_OUT: 1891 /* 1892 * when we scan the bus, we get timeout messages for 1893 * these commands if there is no device available. 1894 * other hosts report did_no_connect for the same thing. 1895 */ 1896 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1897 scmd->cmnd[0] == INQUIRY)) { 1898 return SUCCESS; 1899 } else { 1900 return FAILED; 1901 } 1902 case DID_RESET: 1903 return SUCCESS; 1904 default: 1905 return FAILED; 1906 } 1907 1908 /* 1909 * check the status byte to see if this indicates anything special. 1910 */ 1911 switch (get_status_byte(scmd)) { 1912 case SAM_STAT_TASK_SET_FULL: 1913 scsi_handle_queue_full(scmd->device); 1914 /* 1915 * the case of trying to send too many commands to a 1916 * tagged queueing device. 1917 */ 1918 fallthrough; 1919 case SAM_STAT_BUSY: 1920 /* 1921 * device can't talk to us at the moment. Should only 1922 * occur (SAM-3) when the task queue is empty, so will cause 1923 * the empty queue handling to trigger a stall in the 1924 * device. 1925 */ 1926 return ADD_TO_MLQUEUE; 1927 case SAM_STAT_GOOD: 1928 if (scmd->cmnd[0] == REPORT_LUNS) 1929 scmd->device->sdev_target->expecting_lun_change = 0; 1930 scsi_handle_queue_ramp_up(scmd->device); 1931 fallthrough; 1932 case SAM_STAT_COMMAND_TERMINATED: 1933 return SUCCESS; 1934 case SAM_STAT_TASK_ABORTED: 1935 goto maybe_retry; 1936 case SAM_STAT_CHECK_CONDITION: 1937 rtn = scsi_check_sense(scmd); 1938 if (rtn == NEEDS_RETRY) 1939 goto maybe_retry; 1940 /* if rtn == FAILED, we have no sense information; 1941 * returning FAILED will wake the error handler thread 1942 * to collect the sense and redo the decide 1943 * disposition */ 1944 return rtn; 1945 case SAM_STAT_CONDITION_MET: 1946 case SAM_STAT_INTERMEDIATE: 1947 case SAM_STAT_INTERMEDIATE_CONDITION_MET: 1948 case SAM_STAT_ACA_ACTIVE: 1949 /* 1950 * who knows? FIXME(eric) 1951 */ 1952 return SUCCESS; 1953 1954 case SAM_STAT_RESERVATION_CONFLICT: 1955 sdev_printk(KERN_INFO, scmd->device, 1956 "reservation conflict\n"); 1957 set_host_byte(scmd, DID_NEXUS_FAILURE); 1958 return SUCCESS; /* causes immediate i/o error */ 1959 default: 1960 return FAILED; 1961 } 1962 return FAILED; 1963 1964 maybe_retry: 1965 1966 /* we requeue for retry because the error was retryable, and 1967 * the request was not marked fast fail. Note that above, 1968 * even if the request is marked fast fail, we still requeue 1969 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1970 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) { 1971 return NEEDS_RETRY; 1972 } else { 1973 /* 1974 * no more retries - report this one back to upper level. 1975 */ 1976 return SUCCESS; 1977 } 1978 } 1979 1980 static void eh_lock_door_done(struct request *req, blk_status_t status) 1981 { 1982 blk_put_request(req); 1983 } 1984 1985 /** 1986 * scsi_eh_lock_door - Prevent medium removal for the specified device 1987 * @sdev: SCSI device to prevent medium removal 1988 * 1989 * Locking: 1990 * We must be called from process context. 1991 * 1992 * Notes: 1993 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 1994 * head of the devices request queue, and continue. 1995 */ 1996 static void scsi_eh_lock_door(struct scsi_device *sdev) 1997 { 1998 struct request *req; 1999 struct scsi_request *rq; 2000 2001 req = blk_get_request(sdev->request_queue, REQ_OP_DRV_IN, 0); 2002 if (IS_ERR(req)) 2003 return; 2004 rq = scsi_req(req); 2005 2006 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL; 2007 rq->cmd[1] = 0; 2008 rq->cmd[2] = 0; 2009 rq->cmd[3] = 0; 2010 rq->cmd[4] = SCSI_REMOVAL_PREVENT; 2011 rq->cmd[5] = 0; 2012 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]); 2013 2014 req->rq_flags |= RQF_QUIET; 2015 req->timeout = 10 * HZ; 2016 rq->retries = 5; 2017 2018 blk_execute_rq_nowait(NULL, req, 1, eh_lock_door_done); 2019 } 2020 2021 /** 2022 * scsi_restart_operations - restart io operations to the specified host. 2023 * @shost: Host we are restarting. 2024 * 2025 * Notes: 2026 * When we entered the error handler, we blocked all further i/o to 2027 * this device. we need to 'reverse' this process. 2028 */ 2029 static void scsi_restart_operations(struct Scsi_Host *shost) 2030 { 2031 struct scsi_device *sdev; 2032 unsigned long flags; 2033 2034 /* 2035 * If the door was locked, we need to insert a door lock request 2036 * onto the head of the SCSI request queue for the device. There 2037 * is no point trying to lock the door of an off-line device. 2038 */ 2039 shost_for_each_device(sdev, shost) { 2040 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 2041 scsi_eh_lock_door(sdev); 2042 sdev->was_reset = 0; 2043 } 2044 } 2045 2046 /* 2047 * next free up anything directly waiting upon the host. this 2048 * will be requests for character device operations, and also for 2049 * ioctls to queued block devices. 2050 */ 2051 SCSI_LOG_ERROR_RECOVERY(3, 2052 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 2053 2054 spin_lock_irqsave(shost->host_lock, flags); 2055 if (scsi_host_set_state(shost, SHOST_RUNNING)) 2056 if (scsi_host_set_state(shost, SHOST_CANCEL)) 2057 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 2058 spin_unlock_irqrestore(shost->host_lock, flags); 2059 2060 wake_up(&shost->host_wait); 2061 2062 /* 2063 * finally we need to re-initiate requests that may be pending. we will 2064 * have had everything blocked while error handling is taking place, and 2065 * now that error recovery is done, we will need to ensure that these 2066 * requests are started. 2067 */ 2068 scsi_run_host_queues(shost); 2069 2070 /* 2071 * if eh is active and host_eh_scheduled is pending we need to re-run 2072 * recovery. we do this check after scsi_run_host_queues() to allow 2073 * everything pent up since the last eh run a chance to make forward 2074 * progress before we sync again. Either we'll immediately re-run 2075 * recovery or scsi_device_unbusy() will wake us again when these 2076 * pending commands complete. 2077 */ 2078 spin_lock_irqsave(shost->host_lock, flags); 2079 if (shost->host_eh_scheduled) 2080 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 2081 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 2082 spin_unlock_irqrestore(shost->host_lock, flags); 2083 } 2084 2085 /** 2086 * scsi_eh_ready_devs - check device ready state and recover if not. 2087 * @shost: host to be recovered. 2088 * @work_q: &list_head for pending commands. 2089 * @done_q: &list_head for processed commands. 2090 */ 2091 void scsi_eh_ready_devs(struct Scsi_Host *shost, 2092 struct list_head *work_q, 2093 struct list_head *done_q) 2094 { 2095 if (!scsi_eh_stu(shost, work_q, done_q)) 2096 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2097 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2098 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2099 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2100 scsi_eh_offline_sdevs(work_q, 2101 done_q); 2102 } 2103 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2104 2105 /** 2106 * scsi_eh_flush_done_q - finish processed commands or retry them. 2107 * @done_q: list_head of processed commands. 2108 */ 2109 void scsi_eh_flush_done_q(struct list_head *done_q) 2110 { 2111 struct scsi_cmnd *scmd, *next; 2112 2113 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2114 list_del_init(&scmd->eh_entry); 2115 if (scsi_device_online(scmd->device) && 2116 !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd) && 2117 scsi_eh_should_retry_cmd(scmd)) { 2118 SCSI_LOG_ERROR_RECOVERY(3, 2119 scmd_printk(KERN_INFO, scmd, 2120 "%s: flush retry cmd\n", 2121 current->comm)); 2122 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2123 } else { 2124 /* 2125 * If just we got sense for the device (called 2126 * scsi_eh_get_sense), scmd->result is already 2127 * set, do not set DID_TIME_OUT. 2128 */ 2129 if (!scmd->result) 2130 scmd->result |= (DID_TIME_OUT << 16); 2131 SCSI_LOG_ERROR_RECOVERY(3, 2132 scmd_printk(KERN_INFO, scmd, 2133 "%s: flush finish cmd\n", 2134 current->comm)); 2135 scsi_finish_command(scmd); 2136 } 2137 } 2138 } 2139 EXPORT_SYMBOL(scsi_eh_flush_done_q); 2140 2141 /** 2142 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2143 * @shost: Host to unjam. 2144 * 2145 * Notes: 2146 * When we come in here, we *know* that all commands on the bus have 2147 * either completed, failed or timed out. we also know that no further 2148 * commands are being sent to the host, so things are relatively quiet 2149 * and we have freedom to fiddle with things as we wish. 2150 * 2151 * This is only the *default* implementation. it is possible for 2152 * individual drivers to supply their own version of this function, and 2153 * if the maintainer wishes to do this, it is strongly suggested that 2154 * this function be taken as a template and modified. this function 2155 * was designed to correctly handle problems for about 95% of the 2156 * different cases out there, and it should always provide at least a 2157 * reasonable amount of error recovery. 2158 * 2159 * Any command marked 'failed' or 'timeout' must eventually have 2160 * scsi_finish_cmd() called for it. we do all of the retry stuff 2161 * here, so when we restart the host after we return it should have an 2162 * empty queue. 2163 */ 2164 static void scsi_unjam_host(struct Scsi_Host *shost) 2165 { 2166 unsigned long flags; 2167 LIST_HEAD(eh_work_q); 2168 LIST_HEAD(eh_done_q); 2169 2170 spin_lock_irqsave(shost->host_lock, flags); 2171 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2172 spin_unlock_irqrestore(shost->host_lock, flags); 2173 2174 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2175 2176 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2177 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2178 2179 spin_lock_irqsave(shost->host_lock, flags); 2180 if (shost->eh_deadline != -1) 2181 shost->last_reset = 0; 2182 spin_unlock_irqrestore(shost->host_lock, flags); 2183 scsi_eh_flush_done_q(&eh_done_q); 2184 } 2185 2186 /** 2187 * scsi_error_handler - SCSI error handler thread 2188 * @data: Host for which we are running. 2189 * 2190 * Notes: 2191 * This is the main error handling loop. This is run as a kernel thread 2192 * for every SCSI host and handles all error handling activity. 2193 */ 2194 int scsi_error_handler(void *data) 2195 { 2196 struct Scsi_Host *shost = data; 2197 2198 /* 2199 * We use TASK_INTERRUPTIBLE so that the thread is not 2200 * counted against the load average as a running process. 2201 * We never actually get interrupted because kthread_run 2202 * disables signal delivery for the created thread. 2203 */ 2204 while (true) { 2205 /* 2206 * The sequence in kthread_stop() sets the stop flag first 2207 * then wakes the process. To avoid missed wakeups, the task 2208 * should always be in a non running state before the stop 2209 * flag is checked 2210 */ 2211 set_current_state(TASK_INTERRUPTIBLE); 2212 if (kthread_should_stop()) 2213 break; 2214 2215 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2216 shost->host_failed != scsi_host_busy(shost)) { 2217 SCSI_LOG_ERROR_RECOVERY(1, 2218 shost_printk(KERN_INFO, shost, 2219 "scsi_eh_%d: sleeping\n", 2220 shost->host_no)); 2221 schedule(); 2222 continue; 2223 } 2224 2225 __set_current_state(TASK_RUNNING); 2226 SCSI_LOG_ERROR_RECOVERY(1, 2227 shost_printk(KERN_INFO, shost, 2228 "scsi_eh_%d: waking up %d/%d/%d\n", 2229 shost->host_no, shost->host_eh_scheduled, 2230 shost->host_failed, 2231 scsi_host_busy(shost))); 2232 2233 /* 2234 * We have a host that is failing for some reason. Figure out 2235 * what we need to do to get it up and online again (if we can). 2236 * If we fail, we end up taking the thing offline. 2237 */ 2238 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { 2239 SCSI_LOG_ERROR_RECOVERY(1, 2240 shost_printk(KERN_ERR, shost, 2241 "scsi_eh_%d: unable to autoresume\n", 2242 shost->host_no)); 2243 continue; 2244 } 2245 2246 if (shost->transportt->eh_strategy_handler) 2247 shost->transportt->eh_strategy_handler(shost); 2248 else 2249 scsi_unjam_host(shost); 2250 2251 /* All scmds have been handled */ 2252 shost->host_failed = 0; 2253 2254 /* 2255 * Note - if the above fails completely, the action is to take 2256 * individual devices offline and flush the queue of any 2257 * outstanding requests that may have been pending. When we 2258 * restart, we restart any I/O to any other devices on the bus 2259 * which are still online. 2260 */ 2261 scsi_restart_operations(shost); 2262 if (!shost->eh_noresume) 2263 scsi_autopm_put_host(shost); 2264 } 2265 __set_current_state(TASK_RUNNING); 2266 2267 SCSI_LOG_ERROR_RECOVERY(1, 2268 shost_printk(KERN_INFO, shost, 2269 "Error handler scsi_eh_%d exiting\n", 2270 shost->host_no)); 2271 shost->ehandler = NULL; 2272 return 0; 2273 } 2274 2275 /* 2276 * Function: scsi_report_bus_reset() 2277 * 2278 * Purpose: Utility function used by low-level drivers to report that 2279 * they have observed a bus reset on the bus being handled. 2280 * 2281 * Arguments: shost - Host in question 2282 * channel - channel on which reset was observed. 2283 * 2284 * Returns: Nothing 2285 * 2286 * Lock status: Host lock must be held. 2287 * 2288 * Notes: This only needs to be called if the reset is one which 2289 * originates from an unknown location. Resets originated 2290 * by the mid-level itself don't need to call this, but there 2291 * should be no harm. 2292 * 2293 * The main purpose of this is to make sure that a CHECK_CONDITION 2294 * is properly treated. 2295 */ 2296 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2297 { 2298 struct scsi_device *sdev; 2299 2300 __shost_for_each_device(sdev, shost) { 2301 if (channel == sdev_channel(sdev)) 2302 __scsi_report_device_reset(sdev, NULL); 2303 } 2304 } 2305 EXPORT_SYMBOL(scsi_report_bus_reset); 2306 2307 /* 2308 * Function: scsi_report_device_reset() 2309 * 2310 * Purpose: Utility function used by low-level drivers to report that 2311 * they have observed a device reset on the device being handled. 2312 * 2313 * Arguments: shost - Host in question 2314 * channel - channel on which reset was observed 2315 * target - target on which reset was observed 2316 * 2317 * Returns: Nothing 2318 * 2319 * Lock status: Host lock must be held 2320 * 2321 * Notes: This only needs to be called if the reset is one which 2322 * originates from an unknown location. Resets originated 2323 * by the mid-level itself don't need to call this, but there 2324 * should be no harm. 2325 * 2326 * The main purpose of this is to make sure that a CHECK_CONDITION 2327 * is properly treated. 2328 */ 2329 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2330 { 2331 struct scsi_device *sdev; 2332 2333 __shost_for_each_device(sdev, shost) { 2334 if (channel == sdev_channel(sdev) && 2335 target == sdev_id(sdev)) 2336 __scsi_report_device_reset(sdev, NULL); 2337 } 2338 } 2339 EXPORT_SYMBOL(scsi_report_device_reset); 2340 2341 /** 2342 * scsi_ioctl_reset: explicitly reset a host/bus/target/device 2343 * @dev: scsi_device to operate on 2344 * @arg: reset type (see sg.h) 2345 */ 2346 int 2347 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) 2348 { 2349 struct scsi_cmnd *scmd; 2350 struct Scsi_Host *shost = dev->host; 2351 struct request *rq; 2352 unsigned long flags; 2353 int error = 0, val; 2354 enum scsi_disposition rtn; 2355 2356 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) 2357 return -EACCES; 2358 2359 error = get_user(val, arg); 2360 if (error) 2361 return error; 2362 2363 if (scsi_autopm_get_host(shost) < 0) 2364 return -EIO; 2365 2366 error = -EIO; 2367 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) + 2368 shost->hostt->cmd_size, GFP_KERNEL); 2369 if (!rq) 2370 goto out_put_autopm_host; 2371 blk_rq_init(NULL, rq); 2372 2373 scmd = (struct scsi_cmnd *)(rq + 1); 2374 scsi_init_command(dev, scmd); 2375 scmd->cmnd = scsi_req(rq)->cmd; 2376 2377 scmd->submitter = SUBMITTED_BY_SCSI_RESET_IOCTL; 2378 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 2379 2380 scmd->cmd_len = 0; 2381 2382 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 2383 2384 spin_lock_irqsave(shost->host_lock, flags); 2385 shost->tmf_in_progress = 1; 2386 spin_unlock_irqrestore(shost->host_lock, flags); 2387 2388 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { 2389 case SG_SCSI_RESET_NOTHING: 2390 rtn = SUCCESS; 2391 break; 2392 case SG_SCSI_RESET_DEVICE: 2393 rtn = scsi_try_bus_device_reset(scmd); 2394 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2395 break; 2396 fallthrough; 2397 case SG_SCSI_RESET_TARGET: 2398 rtn = scsi_try_target_reset(scmd); 2399 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2400 break; 2401 fallthrough; 2402 case SG_SCSI_RESET_BUS: 2403 rtn = scsi_try_bus_reset(scmd); 2404 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2405 break; 2406 fallthrough; 2407 case SG_SCSI_RESET_HOST: 2408 rtn = scsi_try_host_reset(scmd); 2409 if (rtn == SUCCESS) 2410 break; 2411 fallthrough; 2412 default: 2413 rtn = FAILED; 2414 break; 2415 } 2416 2417 error = (rtn == SUCCESS) ? 0 : -EIO; 2418 2419 spin_lock_irqsave(shost->host_lock, flags); 2420 shost->tmf_in_progress = 0; 2421 spin_unlock_irqrestore(shost->host_lock, flags); 2422 2423 /* 2424 * be sure to wake up anyone who was sleeping or had their queue 2425 * suspended while we performed the TMF. 2426 */ 2427 SCSI_LOG_ERROR_RECOVERY(3, 2428 shost_printk(KERN_INFO, shost, 2429 "waking up host to restart after TMF\n")); 2430 2431 wake_up(&shost->host_wait); 2432 scsi_run_host_queues(shost); 2433 2434 kfree(rq); 2435 2436 out_put_autopm_host: 2437 scsi_autopm_put_host(shost); 2438 return error; 2439 } 2440 2441 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, 2442 struct scsi_sense_hdr *sshdr) 2443 { 2444 return scsi_normalize_sense(cmd->sense_buffer, 2445 SCSI_SENSE_BUFFERSIZE, sshdr); 2446 } 2447 EXPORT_SYMBOL(scsi_command_normalize_sense); 2448 2449 /** 2450 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2451 * @sense_buffer: byte array of sense data 2452 * @sb_len: number of valid bytes in sense_buffer 2453 * @info_out: pointer to 64 integer where 8 or 4 byte information 2454 * field will be placed if found. 2455 * 2456 * Return value: 2457 * true if information field found, false if not found. 2458 */ 2459 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len, 2460 u64 *info_out) 2461 { 2462 const u8 * ucp; 2463 2464 if (sb_len < 7) 2465 return false; 2466 switch (sense_buffer[0] & 0x7f) { 2467 case 0x70: 2468 case 0x71: 2469 if (sense_buffer[0] & 0x80) { 2470 *info_out = get_unaligned_be32(&sense_buffer[3]); 2471 return true; 2472 } 2473 return false; 2474 case 0x72: 2475 case 0x73: 2476 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2477 0 /* info desc */); 2478 if (ucp && (0xa == ucp[1])) { 2479 *info_out = get_unaligned_be64(&ucp[4]); 2480 return true; 2481 } 2482 return false; 2483 default: 2484 return false; 2485 } 2486 } 2487 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2488