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 <linux/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(const struct scsi_host_template *, 62 struct scsi_cmnd *); 63 64 void scsi_eh_wakeup(struct Scsi_Host *shost, unsigned int busy) 65 { 66 lockdep_assert_held(shost->host_lock); 67 68 if (busy == 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, scsi_host_busy(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_timeout(), 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 struct Scsi_Host *shost = sdev->host; 153 enum scsi_disposition rtn; 154 unsigned long flags; 155 156 if (scsi_host_eh_past_deadline(shost)) { 157 SCSI_LOG_ERROR_RECOVERY(3, 158 scmd_printk(KERN_INFO, scmd, 159 "eh timeout, not aborting\n")); 160 goto out; 161 } 162 163 SCSI_LOG_ERROR_RECOVERY(3, 164 scmd_printk(KERN_INFO, scmd, 165 "aborting command\n")); 166 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd); 167 if (rtn != SUCCESS) { 168 SCSI_LOG_ERROR_RECOVERY(3, 169 scmd_printk(KERN_INFO, scmd, 170 "cmd abort %s\n", 171 (rtn == FAST_IO_FAIL) ? 172 "not send" : "failed")); 173 goto out; 174 } 175 set_host_byte(scmd, DID_TIME_OUT); 176 if (scsi_host_eh_past_deadline(shost)) { 177 SCSI_LOG_ERROR_RECOVERY(3, 178 scmd_printk(KERN_INFO, scmd, 179 "eh timeout, not retrying " 180 "aborted command\n")); 181 goto out; 182 } 183 184 spin_lock_irqsave(shost->host_lock, flags); 185 list_del_init(&scmd->eh_entry); 186 187 /* 188 * If the abort succeeds, and there is no further 189 * EH action, clear the ->last_reset time. 190 */ 191 if (list_empty(&shost->eh_abort_list) && 192 list_empty(&shost->eh_cmd_q)) 193 if (shost->eh_deadline != -1) 194 shost->last_reset = 0; 195 196 spin_unlock_irqrestore(shost->host_lock, flags); 197 198 if (!scsi_noretry_cmd(scmd) && 199 scsi_cmd_retry_allowed(scmd) && 200 scsi_eh_should_retry_cmd(scmd)) { 201 SCSI_LOG_ERROR_RECOVERY(3, 202 scmd_printk(KERN_WARNING, scmd, 203 "retry aborted command\n")); 204 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 205 } else { 206 SCSI_LOG_ERROR_RECOVERY(3, 207 scmd_printk(KERN_WARNING, scmd, 208 "finish aborted command\n")); 209 scsi_finish_command(scmd); 210 } 211 return; 212 213 out: 214 spin_lock_irqsave(shost->host_lock, flags); 215 list_del_init(&scmd->eh_entry); 216 spin_unlock_irqrestore(shost->host_lock, flags); 217 218 scsi_eh_scmd_add(scmd); 219 } 220 221 /** 222 * scsi_abort_command - schedule a command abort 223 * @scmd: scmd to abort. 224 * 225 * We only need to abort commands after a command timeout 226 */ 227 static int 228 scsi_abort_command(struct scsi_cmnd *scmd) 229 { 230 struct scsi_device *sdev = scmd->device; 231 struct Scsi_Host *shost = sdev->host; 232 unsigned long flags; 233 234 if (!shost->hostt->eh_abort_handler) { 235 /* No abort handler, fail command directly */ 236 return FAILED; 237 } 238 239 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 240 /* 241 * Retry after abort failed, escalate to next level. 242 */ 243 SCSI_LOG_ERROR_RECOVERY(3, 244 scmd_printk(KERN_INFO, scmd, 245 "previous abort failed\n")); 246 BUG_ON(delayed_work_pending(&scmd->abort_work)); 247 return FAILED; 248 } 249 250 spin_lock_irqsave(shost->host_lock, flags); 251 if (shost->eh_deadline != -1 && !shost->last_reset) 252 shost->last_reset = jiffies; 253 BUG_ON(!list_empty(&scmd->eh_entry)); 254 list_add_tail(&scmd->eh_entry, &shost->eh_abort_list); 255 spin_unlock_irqrestore(shost->host_lock, flags); 256 257 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; 258 SCSI_LOG_ERROR_RECOVERY(3, 259 scmd_printk(KERN_INFO, scmd, "abort scheduled\n")); 260 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100); 261 return SUCCESS; 262 } 263 264 /** 265 * scsi_eh_reset - call into ->eh_action to reset internal counters 266 * @scmd: scmd to run eh on. 267 * 268 * The scsi driver might be carrying internal state about the 269 * devices, so we need to call into the driver to reset the 270 * internal state once the error handler is started. 271 */ 272 static void scsi_eh_reset(struct scsi_cmnd *scmd) 273 { 274 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) { 275 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 276 if (sdrv->eh_reset) 277 sdrv->eh_reset(scmd); 278 } 279 } 280 281 static void scsi_eh_inc_host_failed(struct rcu_head *head) 282 { 283 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu); 284 struct Scsi_Host *shost = scmd->device->host; 285 unsigned int busy = scsi_host_busy(shost); 286 unsigned long flags; 287 288 spin_lock_irqsave(shost->host_lock, flags); 289 shost->host_failed++; 290 scsi_eh_wakeup(shost, busy); 291 spin_unlock_irqrestore(shost->host_lock, flags); 292 } 293 294 /** 295 * scsi_eh_scmd_add - add scsi cmd to error handling. 296 * @scmd: scmd to run eh on. 297 */ 298 void scsi_eh_scmd_add(struct scsi_cmnd *scmd) 299 { 300 struct Scsi_Host *shost = scmd->device->host; 301 unsigned long flags; 302 int ret; 303 304 WARN_ON_ONCE(!shost->ehandler); 305 WARN_ON_ONCE(!test_bit(SCMD_STATE_INFLIGHT, &scmd->state)); 306 307 spin_lock_irqsave(shost->host_lock, flags); 308 if (scsi_host_set_state(shost, SHOST_RECOVERY)) { 309 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY); 310 WARN_ON_ONCE(ret); 311 } 312 if (shost->eh_deadline != -1 && !shost->last_reset) 313 shost->last_reset = jiffies; 314 315 scsi_eh_reset(scmd); 316 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 317 spin_unlock_irqrestore(shost->host_lock, flags); 318 /* 319 * Ensure that all tasks observe the host state change before the 320 * host_failed change. 321 */ 322 call_rcu_hurry(&scmd->rcu, scsi_eh_inc_host_failed); 323 } 324 325 /** 326 * scsi_timeout - Timeout function for normal scsi commands. 327 * @req: request that is timing out. 328 * 329 * Notes: 330 * We do not need to lock this. There is the potential for a race 331 * only in that the normal completion handling might run, but if the 332 * normal completion function determines that the timer has already 333 * fired, then it mustn't do anything. 334 */ 335 enum blk_eh_timer_return scsi_timeout(struct request *req) 336 { 337 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req); 338 struct Scsi_Host *host = scmd->device->host; 339 340 trace_scsi_dispatch_cmd_timeout(scmd); 341 scsi_log_completion(scmd, TIMEOUT_ERROR); 342 343 atomic_inc(&scmd->device->iotmo_cnt); 344 if (host->eh_deadline != -1 && !host->last_reset) 345 host->last_reset = jiffies; 346 347 if (host->hostt->eh_timed_out) { 348 switch (host->hostt->eh_timed_out(scmd)) { 349 case SCSI_EH_DONE: 350 return BLK_EH_DONE; 351 case SCSI_EH_RESET_TIMER: 352 return BLK_EH_RESET_TIMER; 353 case SCSI_EH_NOT_HANDLED: 354 break; 355 } 356 } 357 358 /* 359 * If scsi_done() has already set SCMD_STATE_COMPLETE, do not modify 360 * *scmd. 361 */ 362 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state)) 363 return BLK_EH_DONE; 364 atomic_inc(&scmd->device->iodone_cnt); 365 if (scsi_abort_command(scmd) != SUCCESS) { 366 set_host_byte(scmd, DID_TIME_OUT); 367 scsi_eh_scmd_add(scmd); 368 } 369 370 return BLK_EH_DONE; 371 } 372 373 /** 374 * scsi_block_when_processing_errors - Prevent cmds from being queued. 375 * @sdev: Device on which we are performing recovery. 376 * 377 * Description: 378 * We block until the host is out of error recovery, and then check to 379 * see whether the host or the device is offline. 380 * 381 * Return value: 382 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 383 */ 384 int scsi_block_when_processing_errors(struct scsi_device *sdev) 385 { 386 int online; 387 388 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 389 390 online = scsi_device_online(sdev); 391 392 return online; 393 } 394 EXPORT_SYMBOL(scsi_block_when_processing_errors); 395 396 #ifdef CONFIG_SCSI_LOGGING 397 /** 398 * scsi_eh_prt_fail_stats - Log info on failures. 399 * @shost: scsi host being recovered. 400 * @work_q: Queue of scsi cmds to process. 401 */ 402 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 403 struct list_head *work_q) 404 { 405 struct scsi_cmnd *scmd; 406 struct scsi_device *sdev; 407 int total_failures = 0; 408 int cmd_failed = 0; 409 int cmd_cancel = 0; 410 int devices_failed = 0; 411 412 shost_for_each_device(sdev, shost) { 413 list_for_each_entry(scmd, work_q, eh_entry) { 414 if (scmd->device == sdev) { 415 ++total_failures; 416 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) 417 ++cmd_cancel; 418 else 419 ++cmd_failed; 420 } 421 } 422 423 if (cmd_cancel || cmd_failed) { 424 SCSI_LOG_ERROR_RECOVERY(3, 425 shost_printk(KERN_INFO, shost, 426 "%s: cmds failed: %d, cancel: %d\n", 427 __func__, cmd_failed, 428 cmd_cancel)); 429 cmd_cancel = 0; 430 cmd_failed = 0; 431 ++devices_failed; 432 } 433 } 434 435 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, 436 "Total of %d commands on %d" 437 " devices require eh work\n", 438 total_failures, devices_failed)); 439 } 440 #endif 441 442 /** 443 * scsi_report_lun_change - Set flag on all *other* devices on the same target 444 * to indicate that a UNIT ATTENTION is expected. 445 * @sdev: Device reporting the UNIT ATTENTION 446 */ 447 static void scsi_report_lun_change(struct scsi_device *sdev) 448 { 449 sdev->sdev_target->expecting_lun_change = 1; 450 } 451 452 /** 453 * scsi_report_sense - Examine scsi sense information and log messages for 454 * certain conditions, also issue uevents for some of them. 455 * @sdev: Device reporting the sense code 456 * @sshdr: sshdr to be examined 457 */ 458 static void scsi_report_sense(struct scsi_device *sdev, 459 struct scsi_sense_hdr *sshdr) 460 { 461 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ 462 463 if (sshdr->sense_key == UNIT_ATTENTION) { 464 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { 465 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; 466 sdev_printk(KERN_WARNING, sdev, 467 "Inquiry data has changed"); 468 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { 469 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; 470 scsi_report_lun_change(sdev); 471 sdev_printk(KERN_WARNING, sdev, 472 "LUN assignments on this target have " 473 "changed. The Linux SCSI layer does not " 474 "automatically remap LUN assignments.\n"); 475 } else if (sshdr->asc == 0x3f) 476 sdev_printk(KERN_WARNING, sdev, 477 "Operating parameters on this target have " 478 "changed. The Linux SCSI layer does not " 479 "automatically adjust these parameters.\n"); 480 481 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { 482 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; 483 sdev_printk(KERN_WARNING, sdev, 484 "Warning! Received an indication that the " 485 "LUN reached a thin provisioning soft " 486 "threshold.\n"); 487 } 488 489 if (sshdr->asc == 0x29) { 490 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED; 491 /* 492 * Do not print message if it is an expected side-effect 493 * of runtime PM. 494 */ 495 if (!sdev->silence_suspend) 496 sdev_printk(KERN_WARNING, sdev, 497 "Power-on or device reset occurred\n"); 498 } 499 500 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { 501 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; 502 sdev_printk(KERN_WARNING, sdev, 503 "Mode parameters changed"); 504 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { 505 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; 506 sdev_printk(KERN_WARNING, sdev, 507 "Asymmetric access state changed"); 508 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { 509 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; 510 sdev_printk(KERN_WARNING, sdev, 511 "Capacity data has changed"); 512 } else if (sshdr->asc == 0x2a) 513 sdev_printk(KERN_WARNING, sdev, 514 "Parameters changed"); 515 } 516 517 if (evt_type != SDEV_EVT_MAXBITS) { 518 set_bit(evt_type, sdev->pending_events); 519 schedule_work(&sdev->event_work); 520 } 521 } 522 523 static inline void set_scsi_ml_byte(struct scsi_cmnd *cmd, u8 status) 524 { 525 cmd->result = (cmd->result & 0xffff00ff) | (status << 8); 526 } 527 528 /** 529 * scsi_check_sense - Examine scsi cmd sense 530 * @scmd: Cmd to have sense checked. 531 * 532 * Return value: 533 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE 534 * 535 * Notes: 536 * When a deferred error is detected the current command has 537 * not been executed and needs retrying. 538 */ 539 enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd) 540 { 541 struct request *req = scsi_cmd_to_rq(scmd); 542 struct scsi_device *sdev = scmd->device; 543 struct scsi_sense_hdr sshdr; 544 545 if (! scsi_command_normalize_sense(scmd, &sshdr)) 546 return FAILED; /* no valid sense data */ 547 548 scsi_report_sense(sdev, &sshdr); 549 550 if (sshdr.sense_key == UNIT_ATTENTION) { 551 /* 552 * Increment the counters for Power on/Reset or New Media so 553 * that all ULDs interested in these can see that those have 554 * happened, even if someone else gets the sense data. 555 */ 556 if (sshdr.asc == 0x28) 557 scmd->device->ua_new_media_ctr++; 558 else if (sshdr.asc == 0x29) 559 scmd->device->ua_por_ctr++; 560 } 561 562 if (scsi_sense_is_deferred(&sshdr)) 563 return NEEDS_RETRY; 564 565 if (sdev->handler && sdev->handler->check_sense) { 566 enum scsi_disposition rc; 567 568 rc = sdev->handler->check_sense(sdev, &sshdr); 569 if (rc != SCSI_RETURN_NOT_HANDLED) 570 return rc; 571 /* handler does not care. Drop down to default handling */ 572 } 573 574 if (scmd->cmnd[0] == TEST_UNIT_READY && 575 scmd->submitter != SUBMITTED_BY_SCSI_ERROR_HANDLER) 576 /* 577 * nasty: for mid-layer issued TURs, we need to return the 578 * actual sense data without any recovery attempt. For eh 579 * issued ones, we need to try to recover and interpret 580 */ 581 return SUCCESS; 582 583 /* 584 * Previous logic looked for FILEMARK, EOM or ILI which are 585 * mainly associated with tapes and returned SUCCESS. 586 */ 587 if (sshdr.response_code == 0x70) { 588 /* fixed format */ 589 if (scmd->sense_buffer[2] & 0xe0) 590 return SUCCESS; 591 } else { 592 /* 593 * descriptor format: look for "stream commands sense data 594 * descriptor" (see SSC-3). Assume single sense data 595 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 596 */ 597 if ((sshdr.additional_length > 3) && 598 (scmd->sense_buffer[8] == 0x4) && 599 (scmd->sense_buffer[11] & 0xe0)) 600 return SUCCESS; 601 } 602 603 switch (sshdr.sense_key) { 604 case NO_SENSE: 605 return SUCCESS; 606 case RECOVERED_ERROR: 607 return /* soft_error */ SUCCESS; 608 609 case ABORTED_COMMAND: 610 if (sshdr.asc == 0x10) /* DIF */ 611 return SUCCESS; 612 613 /* 614 * Check aborts due to command duration limit policy: 615 * ABORTED COMMAND additional sense code with the 616 * COMMAND TIMEOUT BEFORE PROCESSING or 617 * COMMAND TIMEOUT DURING PROCESSING or 618 * COMMAND TIMEOUT DURING PROCESSING DUE TO ERROR RECOVERY 619 * additional sense code qualifiers. 620 */ 621 if (sshdr.asc == 0x2e && 622 sshdr.ascq >= 0x01 && sshdr.ascq <= 0x03) { 623 set_scsi_ml_byte(scmd, SCSIML_STAT_DL_TIMEOUT); 624 req->cmd_flags |= REQ_FAILFAST_DEV; 625 req->rq_flags |= RQF_QUIET; 626 return SUCCESS; 627 } 628 629 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF) 630 return ADD_TO_MLQUEUE; 631 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 && 632 sdev->sdev_bflags & BLIST_RETRY_ASC_C1) 633 return ADD_TO_MLQUEUE; 634 635 return NEEDS_RETRY; 636 case NOT_READY: 637 case UNIT_ATTENTION: 638 /* 639 * if we are expecting a cc/ua because of a bus reset that we 640 * performed, treat this just as a retry. otherwise this is 641 * information that we should pass up to the upper-level driver 642 * so that we can deal with it there. 643 */ 644 if (scmd->device->expecting_cc_ua) { 645 /* 646 * Because some device does not queue unit 647 * attentions correctly, we carefully check 648 * additional sense code and qualifier so as 649 * not to squash media change unit attention. 650 */ 651 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { 652 scmd->device->expecting_cc_ua = 0; 653 return NEEDS_RETRY; 654 } 655 } 656 /* 657 * we might also expect a cc/ua if another LUN on the target 658 * reported a UA with an ASC/ASCQ of 3F 0E - 659 * REPORTED LUNS DATA HAS CHANGED. 660 */ 661 if (scmd->device->sdev_target->expecting_lun_change && 662 sshdr.asc == 0x3f && sshdr.ascq == 0x0e) 663 return NEEDS_RETRY; 664 /* 665 * if the device is in the process of becoming ready, we 666 * should retry. 667 */ 668 if ((sshdr.asc == 0x04) && 669 (sshdr.ascq == 0x01 || sshdr.ascq == 0x0a)) 670 return NEEDS_RETRY; 671 /* 672 * if the device is not started, we need to wake 673 * the error handler to start the motor 674 */ 675 if (scmd->device->allow_restart && 676 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 677 return FAILED; 678 /* 679 * Pass the UA upwards for a determination in the completion 680 * functions. 681 */ 682 return SUCCESS; 683 684 /* these are not supported */ 685 case DATA_PROTECT: 686 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { 687 /* Thin provisioning hard threshold reached */ 688 set_scsi_ml_byte(scmd, SCSIML_STAT_NOSPC); 689 return SUCCESS; 690 } 691 fallthrough; 692 case COPY_ABORTED: 693 case VOLUME_OVERFLOW: 694 case MISCOMPARE: 695 case BLANK_CHECK: 696 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE); 697 return SUCCESS; 698 699 case MEDIUM_ERROR: 700 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 701 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 702 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 703 set_scsi_ml_byte(scmd, SCSIML_STAT_MED_ERROR); 704 return SUCCESS; 705 } 706 return NEEDS_RETRY; 707 708 case HARDWARE_ERROR: 709 if (scmd->device->retry_hwerror) 710 return ADD_TO_MLQUEUE; 711 else 712 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE); 713 fallthrough; 714 715 case ILLEGAL_REQUEST: 716 if (sshdr.asc == 0x20 || /* Invalid command operation code */ 717 sshdr.asc == 0x21 || /* Logical block address out of range */ 718 sshdr.asc == 0x22 || /* Invalid function */ 719 sshdr.asc == 0x24 || /* Invalid field in cdb */ 720 sshdr.asc == 0x26 || /* Parameter value invalid */ 721 sshdr.asc == 0x27) { /* Write protected */ 722 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE); 723 } 724 return SUCCESS; 725 726 case COMPLETED: 727 /* 728 * A command using command duration limits (CDL) with a 729 * descriptor set with policy 0xD may be completed with success 730 * and the sense data DATA CURRENTLY UNAVAILABLE, indicating 731 * that the command was in fact aborted because it exceeded its 732 * duration limit. Never retry these commands. 733 */ 734 if (sshdr.asc == 0x55 && sshdr.ascq == 0x0a) { 735 set_scsi_ml_byte(scmd, SCSIML_STAT_DL_TIMEOUT); 736 req->cmd_flags |= REQ_FAILFAST_DEV; 737 req->rq_flags |= RQF_QUIET; 738 } 739 return SUCCESS; 740 741 default: 742 return SUCCESS; 743 } 744 } 745 EXPORT_SYMBOL_GPL(scsi_check_sense); 746 747 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) 748 { 749 const struct scsi_host_template *sht = sdev->host->hostt; 750 struct scsi_device *tmp_sdev; 751 752 if (!sht->track_queue_depth || 753 sdev->queue_depth >= sdev->max_queue_depth) 754 return; 755 756 if (time_before(jiffies, 757 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) 758 return; 759 760 if (time_before(jiffies, 761 sdev->last_queue_full_time + sdev->queue_ramp_up_period)) 762 return; 763 764 /* 765 * Walk all devices of a target and do 766 * ramp up on them. 767 */ 768 shost_for_each_device(tmp_sdev, sdev->host) { 769 if (tmp_sdev->channel != sdev->channel || 770 tmp_sdev->id != sdev->id || 771 tmp_sdev->queue_depth == sdev->max_queue_depth) 772 continue; 773 774 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); 775 sdev->last_queue_ramp_up = jiffies; 776 } 777 } 778 779 static void scsi_handle_queue_full(struct scsi_device *sdev) 780 { 781 const struct scsi_host_template *sht = sdev->host->hostt; 782 struct scsi_device *tmp_sdev; 783 784 if (!sht->track_queue_depth) 785 return; 786 787 shost_for_each_device(tmp_sdev, sdev->host) { 788 if (tmp_sdev->channel != sdev->channel || 789 tmp_sdev->id != sdev->id) 790 continue; 791 /* 792 * We do not know the number of commands that were at 793 * the device when we got the queue full so we start 794 * from the highest possible value and work our way down. 795 */ 796 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); 797 } 798 } 799 800 /** 801 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 802 * @scmd: SCSI cmd to examine. 803 * 804 * Notes: 805 * This is *only* called when we are examining the status of commands 806 * queued during error recovery. the main difference here is that we 807 * don't allow for the possibility of retries here, and we are a lot 808 * more restrictive about what we consider acceptable. 809 */ 810 static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd) 811 { 812 /* 813 * first check the host byte, to see if there is anything in there 814 * that would indicate what we need to do. 815 */ 816 if (host_byte(scmd->result) == DID_RESET) { 817 /* 818 * rats. we are already in the error handler, so we now 819 * get to try and figure out what to do next. if the sense 820 * is valid, we have a pretty good idea of what to do. 821 * if not, we mark it as FAILED. 822 */ 823 return scsi_check_sense(scmd); 824 } 825 if (host_byte(scmd->result) != DID_OK) 826 return FAILED; 827 828 /* 829 * now, check the status byte to see if this indicates 830 * anything special. 831 */ 832 switch (get_status_byte(scmd)) { 833 case SAM_STAT_GOOD: 834 scsi_handle_queue_ramp_up(scmd->device); 835 if (scmd->sense_buffer && SCSI_SENSE_VALID(scmd)) 836 /* 837 * If we have sense data, call scsi_check_sense() in 838 * order to set the correct SCSI ML byte (if any). 839 * No point in checking the return value, since the 840 * command has already completed successfully. 841 */ 842 scsi_check_sense(scmd); 843 fallthrough; 844 case SAM_STAT_COMMAND_TERMINATED: 845 return SUCCESS; 846 case SAM_STAT_CHECK_CONDITION: 847 return scsi_check_sense(scmd); 848 case SAM_STAT_CONDITION_MET: 849 case SAM_STAT_INTERMEDIATE: 850 case SAM_STAT_INTERMEDIATE_CONDITION_MET: 851 /* 852 * who knows? FIXME(eric) 853 */ 854 return SUCCESS; 855 case SAM_STAT_RESERVATION_CONFLICT: 856 if (scmd->cmnd[0] == TEST_UNIT_READY) 857 /* it is a success, we probed the device and 858 * found it */ 859 return SUCCESS; 860 /* otherwise, we failed to send the command */ 861 return FAILED; 862 case SAM_STAT_TASK_SET_FULL: 863 scsi_handle_queue_full(scmd->device); 864 fallthrough; 865 case SAM_STAT_BUSY: 866 return NEEDS_RETRY; 867 default: 868 return FAILED; 869 } 870 return FAILED; 871 } 872 873 /** 874 * scsi_eh_done - Completion function for error handling. 875 * @scmd: Cmd that is done. 876 */ 877 void scsi_eh_done(struct scsi_cmnd *scmd) 878 { 879 struct completion *eh_action; 880 881 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 882 "%s result: %x\n", __func__, scmd->result)); 883 884 eh_action = scmd->device->host->eh_action; 885 if (eh_action) 886 complete(eh_action); 887 } 888 889 /** 890 * scsi_try_host_reset - ask host adapter to reset itself 891 * @scmd: SCSI cmd to send host reset. 892 */ 893 static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd) 894 { 895 unsigned long flags; 896 enum scsi_disposition rtn; 897 struct Scsi_Host *host = scmd->device->host; 898 const struct scsi_host_template *hostt = host->hostt; 899 900 SCSI_LOG_ERROR_RECOVERY(3, 901 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 902 903 if (!hostt->eh_host_reset_handler) 904 return FAILED; 905 906 rtn = hostt->eh_host_reset_handler(scmd); 907 908 if (rtn == SUCCESS) { 909 if (!hostt->skip_settle_delay) 910 ssleep(HOST_RESET_SETTLE_TIME); 911 spin_lock_irqsave(host->host_lock, flags); 912 scsi_report_bus_reset(host, scmd_channel(scmd)); 913 spin_unlock_irqrestore(host->host_lock, flags); 914 } 915 916 return rtn; 917 } 918 919 /** 920 * scsi_try_bus_reset - ask host to perform a bus reset 921 * @scmd: SCSI cmd to send bus reset. 922 */ 923 static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd) 924 { 925 unsigned long flags; 926 enum scsi_disposition rtn; 927 struct Scsi_Host *host = scmd->device->host; 928 const struct scsi_host_template *hostt = host->hostt; 929 930 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 931 "%s: Snd Bus RST\n", __func__)); 932 933 if (!hostt->eh_bus_reset_handler) 934 return FAILED; 935 936 rtn = hostt->eh_bus_reset_handler(scmd); 937 938 if (rtn == SUCCESS) { 939 if (!hostt->skip_settle_delay) 940 ssleep(BUS_RESET_SETTLE_TIME); 941 spin_lock_irqsave(host->host_lock, flags); 942 scsi_report_bus_reset(host, scmd_channel(scmd)); 943 spin_unlock_irqrestore(host->host_lock, flags); 944 } 945 946 return rtn; 947 } 948 949 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 950 { 951 sdev->was_reset = 1; 952 sdev->expecting_cc_ua = 1; 953 } 954 955 /** 956 * scsi_try_target_reset - Ask host to perform a target reset 957 * @scmd: SCSI cmd used to send a target reset 958 * 959 * Notes: 960 * There is no timeout for this operation. if this operation is 961 * unreliable for a given host, then the host itself needs to put a 962 * timer on it, and set the host back to a consistent state prior to 963 * returning. 964 */ 965 static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd) 966 { 967 unsigned long flags; 968 enum scsi_disposition rtn; 969 struct Scsi_Host *host = scmd->device->host; 970 const struct scsi_host_template *hostt = host->hostt; 971 972 if (!hostt->eh_target_reset_handler) 973 return FAILED; 974 975 rtn = hostt->eh_target_reset_handler(scmd); 976 if (rtn == SUCCESS) { 977 spin_lock_irqsave(host->host_lock, flags); 978 __starget_for_each_device(scsi_target(scmd->device), NULL, 979 __scsi_report_device_reset); 980 spin_unlock_irqrestore(host->host_lock, flags); 981 } 982 983 return rtn; 984 } 985 986 /** 987 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 988 * @scmd: SCSI cmd used to send BDR 989 * 990 * Notes: 991 * There is no timeout for this operation. if this operation is 992 * unreliable for a given host, then the host itself needs to put a 993 * timer on it, and set the host back to a consistent state prior to 994 * returning. 995 */ 996 static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 997 { 998 enum scsi_disposition rtn; 999 const struct scsi_host_template *hostt = scmd->device->host->hostt; 1000 1001 if (!hostt->eh_device_reset_handler) 1002 return FAILED; 1003 1004 rtn = hostt->eh_device_reset_handler(scmd); 1005 if (rtn == SUCCESS) 1006 __scsi_report_device_reset(scmd->device, NULL); 1007 return rtn; 1008 } 1009 1010 /** 1011 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command 1012 * @hostt: SCSI driver host template 1013 * @scmd: SCSI cmd used to send a target reset 1014 * 1015 * Return value: 1016 * SUCCESS, FAILED, or FAST_IO_FAIL 1017 * 1018 * Notes: 1019 * SUCCESS does not necessarily indicate that the command 1020 * has been aborted; it only indicates that the LLDDs 1021 * has cleared all references to that command. 1022 * LLDDs should return FAILED only if an abort was required 1023 * but could not be executed. LLDDs should return FAST_IO_FAIL 1024 * if the device is temporarily unavailable (eg due to a 1025 * link down on FibreChannel) 1026 */ 1027 static enum scsi_disposition 1028 scsi_try_to_abort_cmd(const struct scsi_host_template *hostt, struct scsi_cmnd *scmd) 1029 { 1030 if (!hostt->eh_abort_handler) 1031 return FAILED; 1032 1033 return hostt->eh_abort_handler(scmd); 1034 } 1035 1036 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 1037 { 1038 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 1039 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 1040 if (scsi_try_target_reset(scmd) != SUCCESS) 1041 if (scsi_try_bus_reset(scmd) != SUCCESS) 1042 scsi_try_host_reset(scmd); 1043 } 1044 1045 /** 1046 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 1047 * @scmd: SCSI command structure to hijack 1048 * @ses: structure to save restore information 1049 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 1050 * @cmnd_size: size in bytes of @cmnd (must be <= MAX_COMMAND_SIZE) 1051 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 1052 * 1053 * This function is used to save a scsi command information before re-execution 1054 * as part of the error recovery process. If @sense_bytes is 0 the command 1055 * sent must be one that does not transfer any data. If @sense_bytes != 0 1056 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 1057 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 1058 */ 1059 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 1060 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 1061 { 1062 struct scsi_device *sdev = scmd->device; 1063 1064 /* 1065 * We need saved copies of a number of fields - this is because 1066 * error handling may need to overwrite these with different values 1067 * to run different commands, and once error handling is complete, 1068 * we will need to restore these values prior to running the actual 1069 * command. 1070 */ 1071 ses->cmd_len = scmd->cmd_len; 1072 ses->data_direction = scmd->sc_data_direction; 1073 ses->sdb = scmd->sdb; 1074 ses->result = scmd->result; 1075 ses->resid_len = scmd->resid_len; 1076 ses->underflow = scmd->underflow; 1077 ses->prot_op = scmd->prot_op; 1078 ses->eh_eflags = scmd->eh_eflags; 1079 1080 scmd->prot_op = SCSI_PROT_NORMAL; 1081 scmd->eh_eflags = 0; 1082 memcpy(ses->cmnd, scmd->cmnd, sizeof(ses->cmnd)); 1083 memset(scmd->cmnd, 0, sizeof(scmd->cmnd)); 1084 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 1085 scmd->result = 0; 1086 scmd->resid_len = 0; 1087 1088 if (sense_bytes) { 1089 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 1090 sense_bytes); 1091 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 1092 scmd->sdb.length); 1093 scmd->sdb.table.sgl = &ses->sense_sgl; 1094 scmd->sc_data_direction = DMA_FROM_DEVICE; 1095 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; 1096 scmd->cmnd[0] = REQUEST_SENSE; 1097 scmd->cmnd[4] = scmd->sdb.length; 1098 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1099 } else { 1100 scmd->sc_data_direction = DMA_NONE; 1101 if (cmnd) { 1102 BUG_ON(cmnd_size > sizeof(scmd->cmnd)); 1103 memcpy(scmd->cmnd, cmnd, cmnd_size); 1104 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1105 } 1106 } 1107 1108 scmd->underflow = 0; 1109 1110 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 1111 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 1112 (sdev->lun << 5 & 0xe0); 1113 1114 /* 1115 * Zero the sense buffer. The scsi spec mandates that any 1116 * untransferred sense data should be interpreted as being zero. 1117 */ 1118 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 1119 } 1120 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 1121 1122 /** 1123 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 1124 * @scmd: SCSI command structure to restore 1125 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 1126 * 1127 * Undo any damage done by above scsi_eh_prep_cmnd(). 1128 */ 1129 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 1130 { 1131 /* 1132 * Restore original data 1133 */ 1134 scmd->cmd_len = ses->cmd_len; 1135 memcpy(scmd->cmnd, ses->cmnd, sizeof(ses->cmnd)); 1136 scmd->sc_data_direction = ses->data_direction; 1137 scmd->sdb = ses->sdb; 1138 scmd->result = ses->result; 1139 scmd->resid_len = ses->resid_len; 1140 scmd->underflow = ses->underflow; 1141 scmd->prot_op = ses->prot_op; 1142 scmd->eh_eflags = ses->eh_eflags; 1143 } 1144 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 1145 1146 /** 1147 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 1148 * @scmd: SCSI command structure to hijack 1149 * @cmnd: CDB to send 1150 * @cmnd_size: size in bytes of @cmnd 1151 * @timeout: timeout for this request 1152 * @sense_bytes: size of sense data to copy or 0 1153 * 1154 * This function is used to send a scsi command down to a target device 1155 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 1156 * 1157 * Return value: 1158 * SUCCESS or FAILED or NEEDS_RETRY 1159 */ 1160 static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd, 1161 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes) 1162 { 1163 struct scsi_device *sdev = scmd->device; 1164 struct Scsi_Host *shost = sdev->host; 1165 DECLARE_COMPLETION_ONSTACK(done); 1166 unsigned long timeleft = timeout, delay; 1167 struct scsi_eh_save ses; 1168 const unsigned long stall_for = msecs_to_jiffies(100); 1169 int rtn; 1170 1171 retry: 1172 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1173 shost->eh_action = &done; 1174 1175 scsi_log_send(scmd); 1176 scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER; 1177 scmd->flags |= SCMD_LAST; 1178 1179 /* 1180 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can 1181 * change the SCSI device state after we have examined it and before 1182 * .queuecommand() is called. 1183 */ 1184 mutex_lock(&sdev->state_mutex); 1185 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) { 1186 mutex_unlock(&sdev->state_mutex); 1187 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev, 1188 "%s: state %d <> %d\n", __func__, sdev->sdev_state, 1189 SDEV_BLOCK)); 1190 delay = min(timeleft, stall_for); 1191 timeleft -= delay; 1192 msleep(jiffies_to_msecs(delay)); 1193 mutex_lock(&sdev->state_mutex); 1194 } 1195 if (sdev->sdev_state != SDEV_BLOCK) 1196 rtn = shost->hostt->queuecommand(shost, scmd); 1197 else 1198 rtn = FAILED; 1199 mutex_unlock(&sdev->state_mutex); 1200 1201 if (rtn) { 1202 if (timeleft > stall_for) { 1203 scsi_eh_restore_cmnd(scmd, &ses); 1204 1205 timeleft -= stall_for; 1206 msleep(jiffies_to_msecs(stall_for)); 1207 goto retry; 1208 } 1209 /* signal not to enter either branch of the if () below */ 1210 timeleft = 0; 1211 rtn = FAILED; 1212 } else { 1213 timeleft = wait_for_completion_timeout(&done, timeout); 1214 rtn = SUCCESS; 1215 } 1216 1217 shost->eh_action = NULL; 1218 1219 scsi_log_completion(scmd, rtn); 1220 1221 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1222 "%s timeleft: %ld\n", 1223 __func__, timeleft)); 1224 1225 /* 1226 * If there is time left scsi_eh_done got called, and we will examine 1227 * the actual status codes to see whether the command actually did 1228 * complete normally, else if we have a zero return and no time left, 1229 * the command must still be pending, so abort it and return FAILED. 1230 * If we never actually managed to issue the command, because 1231 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1232 * value above (so don't execute either branch of the if) 1233 */ 1234 if (timeleft) { 1235 rtn = scsi_eh_completed_normally(scmd); 1236 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1237 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1238 1239 switch (rtn) { 1240 case SUCCESS: 1241 case NEEDS_RETRY: 1242 case FAILED: 1243 break; 1244 case ADD_TO_MLQUEUE: 1245 rtn = NEEDS_RETRY; 1246 break; 1247 default: 1248 rtn = FAILED; 1249 break; 1250 } 1251 } else if (rtn != FAILED) { 1252 scsi_abort_eh_cmnd(scmd); 1253 rtn = FAILED; 1254 } 1255 1256 scsi_eh_restore_cmnd(scmd, &ses); 1257 1258 return rtn; 1259 } 1260 1261 /** 1262 * scsi_request_sense - Request sense data from a particular target. 1263 * @scmd: SCSI cmd for request sense. 1264 * 1265 * Notes: 1266 * Some hosts automatically obtain this information, others require 1267 * that we obtain it on our own. This function will *not* return until 1268 * the command either times out, or it completes. 1269 */ 1270 static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd) 1271 { 1272 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1273 } 1274 1275 static enum scsi_disposition 1276 scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn) 1277 { 1278 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) { 1279 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1280 if (sdrv->eh_action) 1281 rtn = sdrv->eh_action(scmd, rtn); 1282 } 1283 return rtn; 1284 } 1285 1286 /** 1287 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1288 * @scmd: Original SCSI cmd that eh has finished. 1289 * @done_q: Queue for processed commands. 1290 * 1291 * Notes: 1292 * We don't want to use the normal command completion while we are are 1293 * still handling errors - it may cause other commands to be queued, 1294 * and that would disturb what we are doing. Thus we really want to 1295 * keep a list of pending commands for final completion, and once we 1296 * are ready to leave error handling we handle completion for real. 1297 */ 1298 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1299 { 1300 list_move_tail(&scmd->eh_entry, done_q); 1301 } 1302 EXPORT_SYMBOL(scsi_eh_finish_cmd); 1303 1304 /** 1305 * scsi_eh_get_sense - Get device sense data. 1306 * @work_q: Queue of commands to process. 1307 * @done_q: Queue of processed commands. 1308 * 1309 * Description: 1310 * See if we need to request sense information. if so, then get it 1311 * now, so we have a better idea of what to do. 1312 * 1313 * Notes: 1314 * This has the unfortunate side effect that if a shost adapter does 1315 * not automatically request sense information, we end up shutting 1316 * it down before we request it. 1317 * 1318 * All drivers should request sense information internally these days, 1319 * so for now all I have to say is tough noogies if you end up in here. 1320 * 1321 * XXX: Long term this code should go away, but that needs an audit of 1322 * all LLDDs first. 1323 */ 1324 int scsi_eh_get_sense(struct list_head *work_q, 1325 struct list_head *done_q) 1326 { 1327 struct scsi_cmnd *scmd, *next; 1328 struct Scsi_Host *shost; 1329 enum scsi_disposition rtn; 1330 1331 /* 1332 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, 1333 * should not get sense. 1334 */ 1335 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1336 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || 1337 SCSI_SENSE_VALID(scmd)) 1338 continue; 1339 1340 shost = scmd->device->host; 1341 if (scsi_host_eh_past_deadline(shost)) { 1342 SCSI_LOG_ERROR_RECOVERY(3, 1343 scmd_printk(KERN_INFO, scmd, 1344 "%s: skip request sense, past eh deadline\n", 1345 current->comm)); 1346 break; 1347 } 1348 if (!scsi_status_is_check_condition(scmd->result)) 1349 /* 1350 * don't request sense if there's no check condition 1351 * status because the error we're processing isn't one 1352 * that has a sense code (and some devices get 1353 * confused by sense requests out of the blue) 1354 */ 1355 continue; 1356 1357 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1358 "%s: requesting sense\n", 1359 current->comm)); 1360 rtn = scsi_request_sense(scmd); 1361 if (rtn != SUCCESS) 1362 continue; 1363 1364 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1365 "sense requested, result %x\n", scmd->result)); 1366 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); 1367 1368 rtn = scsi_decide_disposition(scmd); 1369 1370 /* 1371 * if the result was normal, then just pass it along to the 1372 * upper level. 1373 */ 1374 if (rtn == SUCCESS) 1375 /* 1376 * We don't want this command reissued, just finished 1377 * with the sense data, so set retries to the max 1378 * allowed to ensure it won't get reissued. If the user 1379 * has requested infinite retries, we also want to 1380 * finish this command, so force completion by setting 1381 * retries and allowed to the same value. 1382 */ 1383 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) 1384 scmd->retries = scmd->allowed = 1; 1385 else 1386 scmd->retries = scmd->allowed; 1387 else if (rtn != NEEDS_RETRY) 1388 continue; 1389 1390 scsi_eh_finish_cmd(scmd, done_q); 1391 } 1392 1393 return list_empty(work_q); 1394 } 1395 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1396 1397 /** 1398 * scsi_eh_tur - Send TUR to device. 1399 * @scmd: &scsi_cmnd to send TUR 1400 * 1401 * Return value: 1402 * 0 - Device is ready. 1 - Device NOT ready. 1403 */ 1404 static int scsi_eh_tur(struct scsi_cmnd *scmd) 1405 { 1406 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1407 int retry_cnt = 1; 1408 enum scsi_disposition rtn; 1409 1410 retry_tur: 1411 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1412 scmd->device->eh_timeout, 0); 1413 1414 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1415 "%s return: %x\n", __func__, rtn)); 1416 1417 switch (rtn) { 1418 case NEEDS_RETRY: 1419 if (retry_cnt--) 1420 goto retry_tur; 1421 fallthrough; 1422 case SUCCESS: 1423 return 0; 1424 default: 1425 return 1; 1426 } 1427 } 1428 1429 /** 1430 * scsi_eh_test_devices - check if devices are responding from error recovery. 1431 * @cmd_list: scsi commands in error recovery. 1432 * @work_q: queue for commands which still need more error recovery 1433 * @done_q: queue for commands which are finished 1434 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1435 * 1436 * Decription: 1437 * Tests if devices are in a working state. Commands to devices now in 1438 * a working state are sent to the done_q while commands to devices which 1439 * are still failing to respond are returned to the work_q for more 1440 * processing. 1441 **/ 1442 static int scsi_eh_test_devices(struct list_head *cmd_list, 1443 struct list_head *work_q, 1444 struct list_head *done_q, int try_stu) 1445 { 1446 struct scsi_cmnd *scmd, *next; 1447 struct scsi_device *sdev; 1448 int finish_cmds; 1449 1450 while (!list_empty(cmd_list)) { 1451 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1452 sdev = scmd->device; 1453 1454 if (!try_stu) { 1455 if (scsi_host_eh_past_deadline(sdev->host)) { 1456 /* Push items back onto work_q */ 1457 list_splice_init(cmd_list, work_q); 1458 SCSI_LOG_ERROR_RECOVERY(3, 1459 sdev_printk(KERN_INFO, sdev, 1460 "%s: skip test device, past eh deadline", 1461 current->comm)); 1462 break; 1463 } 1464 } 1465 1466 finish_cmds = !scsi_device_online(scmd->device) || 1467 (try_stu && !scsi_eh_try_stu(scmd) && 1468 !scsi_eh_tur(scmd)) || 1469 !scsi_eh_tur(scmd); 1470 1471 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1472 if (scmd->device == sdev) { 1473 if (finish_cmds && 1474 (try_stu || 1475 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1476 scsi_eh_finish_cmd(scmd, done_q); 1477 else 1478 list_move_tail(&scmd->eh_entry, work_q); 1479 } 1480 } 1481 return list_empty(work_q); 1482 } 1483 1484 /** 1485 * scsi_eh_try_stu - Send START_UNIT to device. 1486 * @scmd: &scsi_cmnd to send START_UNIT 1487 * 1488 * Return value: 1489 * 0 - Device is ready. 1 - Device NOT ready. 1490 */ 1491 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1492 { 1493 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1494 1495 if (scmd->device->allow_restart) { 1496 int i; 1497 enum scsi_disposition rtn = NEEDS_RETRY; 1498 1499 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1500 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, 1501 scmd->device->eh_timeout, 0); 1502 1503 if (rtn == SUCCESS) 1504 return 0; 1505 } 1506 1507 return 1; 1508 } 1509 1510 /** 1511 * scsi_eh_stu - send START_UNIT if needed 1512 * @shost: &scsi host being recovered. 1513 * @work_q: &list_head for pending commands. 1514 * @done_q: &list_head for processed commands. 1515 * 1516 * Notes: 1517 * If commands are failing due to not ready, initializing command required, 1518 * try revalidating the device, which will end up sending a start unit. 1519 */ 1520 static int scsi_eh_stu(struct Scsi_Host *shost, 1521 struct list_head *work_q, 1522 struct list_head *done_q) 1523 { 1524 struct scsi_cmnd *scmd, *stu_scmd, *next; 1525 struct scsi_device *sdev; 1526 1527 shost_for_each_device(sdev, shost) { 1528 if (scsi_host_eh_past_deadline(shost)) { 1529 SCSI_LOG_ERROR_RECOVERY(3, 1530 sdev_printk(KERN_INFO, sdev, 1531 "%s: skip START_UNIT, past eh deadline\n", 1532 current->comm)); 1533 scsi_device_put(sdev); 1534 break; 1535 } 1536 stu_scmd = NULL; 1537 list_for_each_entry(scmd, work_q, eh_entry) 1538 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1539 scsi_check_sense(scmd) == FAILED ) { 1540 stu_scmd = scmd; 1541 break; 1542 } 1543 1544 if (!stu_scmd) 1545 continue; 1546 1547 SCSI_LOG_ERROR_RECOVERY(3, 1548 sdev_printk(KERN_INFO, sdev, 1549 "%s: Sending START_UNIT\n", 1550 current->comm)); 1551 1552 if (!scsi_eh_try_stu(stu_scmd)) { 1553 if (!scsi_device_online(sdev) || 1554 !scsi_eh_tur(stu_scmd)) { 1555 list_for_each_entry_safe(scmd, next, 1556 work_q, eh_entry) { 1557 if (scmd->device == sdev && 1558 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1559 scsi_eh_finish_cmd(scmd, done_q); 1560 } 1561 } 1562 } else { 1563 SCSI_LOG_ERROR_RECOVERY(3, 1564 sdev_printk(KERN_INFO, sdev, 1565 "%s: START_UNIT failed\n", 1566 current->comm)); 1567 } 1568 } 1569 1570 return list_empty(work_q); 1571 } 1572 1573 1574 /** 1575 * scsi_eh_bus_device_reset - send bdr if needed 1576 * @shost: scsi host being recovered. 1577 * @work_q: &list_head for pending commands. 1578 * @done_q: &list_head for processed commands. 1579 * 1580 * Notes: 1581 * Try a bus device reset. Still, look to see whether we have multiple 1582 * devices that are jammed or not - if we have multiple devices, it 1583 * makes no sense to try bus_device_reset - we really would need to try 1584 * a bus_reset instead. 1585 */ 1586 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1587 struct list_head *work_q, 1588 struct list_head *done_q) 1589 { 1590 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1591 struct scsi_device *sdev; 1592 enum scsi_disposition rtn; 1593 1594 shost_for_each_device(sdev, shost) { 1595 if (scsi_host_eh_past_deadline(shost)) { 1596 SCSI_LOG_ERROR_RECOVERY(3, 1597 sdev_printk(KERN_INFO, sdev, 1598 "%s: skip BDR, past eh deadline\n", 1599 current->comm)); 1600 scsi_device_put(sdev); 1601 break; 1602 } 1603 bdr_scmd = NULL; 1604 list_for_each_entry(scmd, work_q, eh_entry) 1605 if (scmd->device == sdev) { 1606 bdr_scmd = scmd; 1607 break; 1608 } 1609 1610 if (!bdr_scmd) 1611 continue; 1612 1613 SCSI_LOG_ERROR_RECOVERY(3, 1614 sdev_printk(KERN_INFO, sdev, 1615 "%s: Sending BDR\n", current->comm)); 1616 rtn = scsi_try_bus_device_reset(bdr_scmd); 1617 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1618 if (!scsi_device_online(sdev) || 1619 rtn == FAST_IO_FAIL || 1620 !scsi_eh_tur(bdr_scmd)) { 1621 list_for_each_entry_safe(scmd, next, 1622 work_q, eh_entry) { 1623 if (scmd->device == sdev && 1624 scsi_eh_action(scmd, rtn) != FAILED) 1625 scsi_eh_finish_cmd(scmd, 1626 done_q); 1627 } 1628 } 1629 } else { 1630 SCSI_LOG_ERROR_RECOVERY(3, 1631 sdev_printk(KERN_INFO, sdev, 1632 "%s: BDR failed\n", current->comm)); 1633 } 1634 } 1635 1636 return list_empty(work_q); 1637 } 1638 1639 /** 1640 * scsi_eh_target_reset - send target reset if needed 1641 * @shost: scsi host being recovered. 1642 * @work_q: &list_head for pending commands. 1643 * @done_q: &list_head for processed commands. 1644 * 1645 * Notes: 1646 * Try a target reset. 1647 */ 1648 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1649 struct list_head *work_q, 1650 struct list_head *done_q) 1651 { 1652 LIST_HEAD(tmp_list); 1653 LIST_HEAD(check_list); 1654 1655 list_splice_init(work_q, &tmp_list); 1656 1657 while (!list_empty(&tmp_list)) { 1658 struct scsi_cmnd *next, *scmd; 1659 enum scsi_disposition rtn; 1660 unsigned int id; 1661 1662 if (scsi_host_eh_past_deadline(shost)) { 1663 /* push back on work queue for further processing */ 1664 list_splice_init(&check_list, work_q); 1665 list_splice_init(&tmp_list, work_q); 1666 SCSI_LOG_ERROR_RECOVERY(3, 1667 shost_printk(KERN_INFO, shost, 1668 "%s: Skip target reset, past eh deadline\n", 1669 current->comm)); 1670 return list_empty(work_q); 1671 } 1672 1673 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1674 id = scmd_id(scmd); 1675 1676 SCSI_LOG_ERROR_RECOVERY(3, 1677 shost_printk(KERN_INFO, shost, 1678 "%s: Sending target reset to target %d\n", 1679 current->comm, id)); 1680 rtn = scsi_try_target_reset(scmd); 1681 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1682 SCSI_LOG_ERROR_RECOVERY(3, 1683 shost_printk(KERN_INFO, shost, 1684 "%s: Target reset failed" 1685 " target: %d\n", 1686 current->comm, id)); 1687 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1688 if (scmd_id(scmd) != id) 1689 continue; 1690 1691 if (rtn == SUCCESS) 1692 list_move_tail(&scmd->eh_entry, &check_list); 1693 else if (rtn == FAST_IO_FAIL) 1694 scsi_eh_finish_cmd(scmd, done_q); 1695 else 1696 /* push back on work queue for further processing */ 1697 list_move(&scmd->eh_entry, work_q); 1698 } 1699 } 1700 1701 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1702 } 1703 1704 /** 1705 * scsi_eh_bus_reset - send a bus reset 1706 * @shost: &scsi host being recovered. 1707 * @work_q: &list_head for pending commands. 1708 * @done_q: &list_head for processed commands. 1709 */ 1710 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1711 struct list_head *work_q, 1712 struct list_head *done_q) 1713 { 1714 struct scsi_cmnd *scmd, *chan_scmd, *next; 1715 LIST_HEAD(check_list); 1716 unsigned int channel; 1717 enum scsi_disposition rtn; 1718 1719 /* 1720 * we really want to loop over the various channels, and do this on 1721 * a channel by channel basis. we should also check to see if any 1722 * of the failed commands are on soft_reset devices, and if so, skip 1723 * the reset. 1724 */ 1725 1726 for (channel = 0; channel <= shost->max_channel; channel++) { 1727 if (scsi_host_eh_past_deadline(shost)) { 1728 list_splice_init(&check_list, work_q); 1729 SCSI_LOG_ERROR_RECOVERY(3, 1730 shost_printk(KERN_INFO, shost, 1731 "%s: skip BRST, past eh deadline\n", 1732 current->comm)); 1733 return list_empty(work_q); 1734 } 1735 1736 chan_scmd = NULL; 1737 list_for_each_entry(scmd, work_q, eh_entry) { 1738 if (channel == scmd_channel(scmd)) { 1739 chan_scmd = scmd; 1740 break; 1741 /* 1742 * FIXME add back in some support for 1743 * soft_reset devices. 1744 */ 1745 } 1746 } 1747 1748 if (!chan_scmd) 1749 continue; 1750 SCSI_LOG_ERROR_RECOVERY(3, 1751 shost_printk(KERN_INFO, shost, 1752 "%s: Sending BRST chan: %d\n", 1753 current->comm, channel)); 1754 rtn = scsi_try_bus_reset(chan_scmd); 1755 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1756 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1757 if (channel == scmd_channel(scmd)) { 1758 if (rtn == FAST_IO_FAIL) 1759 scsi_eh_finish_cmd(scmd, 1760 done_q); 1761 else 1762 list_move_tail(&scmd->eh_entry, 1763 &check_list); 1764 } 1765 } 1766 } else { 1767 SCSI_LOG_ERROR_RECOVERY(3, 1768 shost_printk(KERN_INFO, shost, 1769 "%s: BRST failed chan: %d\n", 1770 current->comm, channel)); 1771 } 1772 } 1773 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1774 } 1775 1776 /** 1777 * scsi_eh_host_reset - send a host reset 1778 * @shost: host to be reset. 1779 * @work_q: &list_head for pending commands. 1780 * @done_q: &list_head for processed commands. 1781 */ 1782 static int scsi_eh_host_reset(struct Scsi_Host *shost, 1783 struct list_head *work_q, 1784 struct list_head *done_q) 1785 { 1786 struct scsi_cmnd *scmd, *next; 1787 LIST_HEAD(check_list); 1788 enum scsi_disposition rtn; 1789 1790 if (!list_empty(work_q)) { 1791 scmd = list_entry(work_q->next, 1792 struct scsi_cmnd, eh_entry); 1793 1794 SCSI_LOG_ERROR_RECOVERY(3, 1795 shost_printk(KERN_INFO, shost, 1796 "%s: Sending HRST\n", 1797 current->comm)); 1798 1799 rtn = scsi_try_host_reset(scmd); 1800 if (rtn == SUCCESS) { 1801 list_splice_init(work_q, &check_list); 1802 } else if (rtn == FAST_IO_FAIL) { 1803 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1804 scsi_eh_finish_cmd(scmd, done_q); 1805 } 1806 } else { 1807 SCSI_LOG_ERROR_RECOVERY(3, 1808 shost_printk(KERN_INFO, shost, 1809 "%s: HRST failed\n", 1810 current->comm)); 1811 } 1812 } 1813 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1814 } 1815 1816 /** 1817 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1818 * @work_q: &list_head for pending commands. 1819 * @done_q: &list_head for processed commands. 1820 */ 1821 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1822 struct list_head *done_q) 1823 { 1824 struct scsi_cmnd *scmd, *next; 1825 struct scsi_device *sdev; 1826 1827 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1828 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1829 "not ready after error recovery\n"); 1830 sdev = scmd->device; 1831 1832 mutex_lock(&sdev->state_mutex); 1833 scsi_device_set_state(sdev, SDEV_OFFLINE); 1834 mutex_unlock(&sdev->state_mutex); 1835 1836 scsi_eh_finish_cmd(scmd, done_q); 1837 } 1838 return; 1839 } 1840 1841 /** 1842 * scsi_noretry_cmd - determine if command should be failed fast 1843 * @scmd: SCSI cmd to examine. 1844 */ 1845 bool scsi_noretry_cmd(struct scsi_cmnd *scmd) 1846 { 1847 struct request *req = scsi_cmd_to_rq(scmd); 1848 1849 switch (host_byte(scmd->result)) { 1850 case DID_OK: 1851 break; 1852 case DID_TIME_OUT: 1853 goto check_type; 1854 case DID_BUS_BUSY: 1855 return !!(req->cmd_flags & REQ_FAILFAST_TRANSPORT); 1856 case DID_PARITY: 1857 return !!(req->cmd_flags & REQ_FAILFAST_DEV); 1858 case DID_ERROR: 1859 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT) 1860 return false; 1861 fallthrough; 1862 case DID_SOFT_ERROR: 1863 return !!(req->cmd_flags & REQ_FAILFAST_DRIVER); 1864 } 1865 1866 /* Never retry commands aborted due to a duration limit timeout */ 1867 if (scsi_ml_byte(scmd->result) == SCSIML_STAT_DL_TIMEOUT) 1868 return true; 1869 1870 if (!scsi_status_is_check_condition(scmd->result)) 1871 return false; 1872 1873 check_type: 1874 /* 1875 * assume caller has checked sense and determined 1876 * the check condition was retryable. 1877 */ 1878 if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(req)) 1879 return true; 1880 1881 return false; 1882 } 1883 1884 /** 1885 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1886 * @scmd: SCSI cmd to examine. 1887 * 1888 * Notes: 1889 * This is *only* called when we are examining the status after sending 1890 * out the actual data command. any commands that are queued for error 1891 * recovery (e.g. test_unit_ready) do *not* come through here. 1892 * 1893 * When this routine returns failed, it means the error handler thread 1894 * is woken. In cases where the error code indicates an error that 1895 * doesn't require the error handler read (i.e. we don't need to 1896 * abort/reset), this function should return SUCCESS. 1897 */ 1898 enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd) 1899 { 1900 enum scsi_disposition rtn; 1901 1902 /* 1903 * if the device is offline, then we clearly just pass the result back 1904 * up to the top level. 1905 */ 1906 if (!scsi_device_online(scmd->device)) { 1907 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1908 "%s: device offline - report as SUCCESS\n", __func__)); 1909 return SUCCESS; 1910 } 1911 1912 /* 1913 * first check the host byte, to see if there is anything in there 1914 * that would indicate what we need to do. 1915 */ 1916 switch (host_byte(scmd->result)) { 1917 case DID_PASSTHROUGH: 1918 /* 1919 * no matter what, pass this through to the upper layer. 1920 * nuke this special code so that it looks like we are saying 1921 * did_ok. 1922 */ 1923 scmd->result &= 0xff00ffff; 1924 return SUCCESS; 1925 case DID_OK: 1926 /* 1927 * looks good. drop through, and check the next byte. 1928 */ 1929 break; 1930 case DID_ABORT: 1931 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1932 set_host_byte(scmd, DID_TIME_OUT); 1933 return SUCCESS; 1934 } 1935 fallthrough; 1936 case DID_NO_CONNECT: 1937 case DID_BAD_TARGET: 1938 /* 1939 * note - this means that we just report the status back 1940 * to the top level driver, not that we actually think 1941 * that it indicates SUCCESS. 1942 */ 1943 return SUCCESS; 1944 case DID_SOFT_ERROR: 1945 /* 1946 * when the low level driver returns did_soft_error, 1947 * it is responsible for keeping an internal retry counter 1948 * in order to avoid endless loops (db) 1949 */ 1950 goto maybe_retry; 1951 case DID_IMM_RETRY: 1952 return NEEDS_RETRY; 1953 1954 case DID_REQUEUE: 1955 return ADD_TO_MLQUEUE; 1956 case DID_TRANSPORT_DISRUPTED: 1957 /* 1958 * LLD/transport was disrupted during processing of the IO. 1959 * The transport class is now blocked/blocking, 1960 * and the transport will decide what to do with the IO 1961 * based on its timers and recovery capablilities if 1962 * there are enough retries. 1963 */ 1964 goto maybe_retry; 1965 case DID_TRANSPORT_FAILFAST: 1966 /* 1967 * The transport decided to failfast the IO (most likely 1968 * the fast io fail tmo fired), so send IO directly upwards. 1969 */ 1970 return SUCCESS; 1971 case DID_TRANSPORT_MARGINAL: 1972 /* 1973 * caller has decided not to do retries on 1974 * abort success, so send IO directly upwards 1975 */ 1976 return SUCCESS; 1977 case DID_ERROR: 1978 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT) 1979 /* 1980 * execute reservation conflict processing code 1981 * lower down 1982 */ 1983 break; 1984 fallthrough; 1985 case DID_BUS_BUSY: 1986 case DID_PARITY: 1987 goto maybe_retry; 1988 case DID_TIME_OUT: 1989 /* 1990 * when we scan the bus, we get timeout messages for 1991 * these commands if there is no device available. 1992 * other hosts report did_no_connect for the same thing. 1993 */ 1994 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1995 scmd->cmnd[0] == INQUIRY)) { 1996 return SUCCESS; 1997 } else { 1998 return FAILED; 1999 } 2000 case DID_RESET: 2001 return SUCCESS; 2002 default: 2003 return FAILED; 2004 } 2005 2006 /* 2007 * check the status byte to see if this indicates anything special. 2008 */ 2009 switch (get_status_byte(scmd)) { 2010 case SAM_STAT_TASK_SET_FULL: 2011 scsi_handle_queue_full(scmd->device); 2012 /* 2013 * the case of trying to send too many commands to a 2014 * tagged queueing device. 2015 */ 2016 fallthrough; 2017 case SAM_STAT_BUSY: 2018 /* 2019 * device can't talk to us at the moment. Should only 2020 * occur (SAM-3) when the task queue is empty, so will cause 2021 * the empty queue handling to trigger a stall in the 2022 * device. 2023 */ 2024 return ADD_TO_MLQUEUE; 2025 case SAM_STAT_GOOD: 2026 if (scmd->cmnd[0] == REPORT_LUNS) 2027 scmd->device->sdev_target->expecting_lun_change = 0; 2028 scsi_handle_queue_ramp_up(scmd->device); 2029 if (scmd->sense_buffer && SCSI_SENSE_VALID(scmd)) 2030 /* 2031 * If we have sense data, call scsi_check_sense() in 2032 * order to set the correct SCSI ML byte (if any). 2033 * No point in checking the return value, since the 2034 * command has already completed successfully. 2035 */ 2036 scsi_check_sense(scmd); 2037 fallthrough; 2038 case SAM_STAT_COMMAND_TERMINATED: 2039 return SUCCESS; 2040 case SAM_STAT_TASK_ABORTED: 2041 goto maybe_retry; 2042 case SAM_STAT_CHECK_CONDITION: 2043 rtn = scsi_check_sense(scmd); 2044 if (rtn == NEEDS_RETRY) 2045 goto maybe_retry; 2046 /* if rtn == FAILED, we have no sense information; 2047 * returning FAILED will wake the error handler thread 2048 * to collect the sense and redo the decide 2049 * disposition */ 2050 return rtn; 2051 case SAM_STAT_CONDITION_MET: 2052 case SAM_STAT_INTERMEDIATE: 2053 case SAM_STAT_INTERMEDIATE_CONDITION_MET: 2054 case SAM_STAT_ACA_ACTIVE: 2055 /* 2056 * who knows? FIXME(eric) 2057 */ 2058 return SUCCESS; 2059 2060 case SAM_STAT_RESERVATION_CONFLICT: 2061 sdev_printk(KERN_INFO, scmd->device, 2062 "reservation conflict\n"); 2063 set_scsi_ml_byte(scmd, SCSIML_STAT_RESV_CONFLICT); 2064 return SUCCESS; /* causes immediate i/o error */ 2065 } 2066 return FAILED; 2067 2068 maybe_retry: 2069 2070 /* we requeue for retry because the error was retryable, and 2071 * the request was not marked fast fail. Note that above, 2072 * even if the request is marked fast fail, we still requeue 2073 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 2074 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) { 2075 return NEEDS_RETRY; 2076 } else { 2077 /* 2078 * no more retries - report this one back to upper level. 2079 */ 2080 return SUCCESS; 2081 } 2082 } 2083 2084 static enum rq_end_io_ret eh_lock_door_done(struct request *req, 2085 blk_status_t status) 2086 { 2087 blk_mq_free_request(req); 2088 return RQ_END_IO_NONE; 2089 } 2090 2091 /** 2092 * scsi_eh_lock_door - Prevent medium removal for the specified device 2093 * @sdev: SCSI device to prevent medium removal 2094 * 2095 * Locking: 2096 * We must be called from process context. 2097 * 2098 * Notes: 2099 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 2100 * head of the devices request queue, and continue. 2101 */ 2102 static void scsi_eh_lock_door(struct scsi_device *sdev) 2103 { 2104 struct scsi_cmnd *scmd; 2105 struct request *req; 2106 2107 req = scsi_alloc_request(sdev->request_queue, REQ_OP_DRV_IN, 0); 2108 if (IS_ERR(req)) 2109 return; 2110 scmd = blk_mq_rq_to_pdu(req); 2111 2112 scmd->cmnd[0] = ALLOW_MEDIUM_REMOVAL; 2113 scmd->cmnd[1] = 0; 2114 scmd->cmnd[2] = 0; 2115 scmd->cmnd[3] = 0; 2116 scmd->cmnd[4] = SCSI_REMOVAL_PREVENT; 2117 scmd->cmnd[5] = 0; 2118 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 2119 scmd->allowed = 5; 2120 2121 req->rq_flags |= RQF_QUIET; 2122 req->timeout = 10 * HZ; 2123 req->end_io = eh_lock_door_done; 2124 2125 blk_execute_rq_nowait(req, true); 2126 } 2127 2128 /** 2129 * scsi_restart_operations - restart io operations to the specified host. 2130 * @shost: Host we are restarting. 2131 * 2132 * Notes: 2133 * When we entered the error handler, we blocked all further i/o to 2134 * this device. we need to 'reverse' this process. 2135 */ 2136 static void scsi_restart_operations(struct Scsi_Host *shost) 2137 { 2138 struct scsi_device *sdev; 2139 unsigned long flags; 2140 2141 /* 2142 * If the door was locked, we need to insert a door lock request 2143 * onto the head of the SCSI request queue for the device. There 2144 * is no point trying to lock the door of an off-line device. 2145 */ 2146 shost_for_each_device(sdev, shost) { 2147 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 2148 scsi_eh_lock_door(sdev); 2149 sdev->was_reset = 0; 2150 } 2151 } 2152 2153 /* 2154 * next free up anything directly waiting upon the host. this 2155 * will be requests for character device operations, and also for 2156 * ioctls to queued block devices. 2157 */ 2158 SCSI_LOG_ERROR_RECOVERY(3, 2159 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 2160 2161 spin_lock_irqsave(shost->host_lock, flags); 2162 if (scsi_host_set_state(shost, SHOST_RUNNING)) 2163 if (scsi_host_set_state(shost, SHOST_CANCEL)) 2164 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 2165 spin_unlock_irqrestore(shost->host_lock, flags); 2166 2167 wake_up(&shost->host_wait); 2168 2169 /* 2170 * finally we need to re-initiate requests that may be pending. we will 2171 * have had everything blocked while error handling is taking place, and 2172 * now that error recovery is done, we will need to ensure that these 2173 * requests are started. 2174 */ 2175 scsi_run_host_queues(shost); 2176 2177 /* 2178 * if eh is active and host_eh_scheduled is pending we need to re-run 2179 * recovery. we do this check after scsi_run_host_queues() to allow 2180 * everything pent up since the last eh run a chance to make forward 2181 * progress before we sync again. Either we'll immediately re-run 2182 * recovery or scsi_device_unbusy() will wake us again when these 2183 * pending commands complete. 2184 */ 2185 spin_lock_irqsave(shost->host_lock, flags); 2186 if (shost->host_eh_scheduled) 2187 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 2188 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 2189 spin_unlock_irqrestore(shost->host_lock, flags); 2190 } 2191 2192 /** 2193 * scsi_eh_ready_devs - check device ready state and recover if not. 2194 * @shost: host to be recovered. 2195 * @work_q: &list_head for pending commands. 2196 * @done_q: &list_head for processed commands. 2197 */ 2198 void scsi_eh_ready_devs(struct Scsi_Host *shost, 2199 struct list_head *work_q, 2200 struct list_head *done_q) 2201 { 2202 if (!scsi_eh_stu(shost, work_q, done_q)) 2203 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2204 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2205 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2206 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2207 scsi_eh_offline_sdevs(work_q, 2208 done_q); 2209 } 2210 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2211 2212 /** 2213 * scsi_eh_flush_done_q - finish processed commands or retry them. 2214 * @done_q: list_head of processed commands. 2215 */ 2216 void scsi_eh_flush_done_q(struct list_head *done_q) 2217 { 2218 struct scsi_cmnd *scmd, *next; 2219 2220 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2221 struct scsi_device *sdev = scmd->device; 2222 2223 list_del_init(&scmd->eh_entry); 2224 if (scsi_device_online(sdev) && !scsi_noretry_cmd(scmd) && 2225 scsi_cmd_retry_allowed(scmd) && 2226 scsi_eh_should_retry_cmd(scmd)) { 2227 SCSI_LOG_ERROR_RECOVERY(3, 2228 scmd_printk(KERN_INFO, scmd, 2229 "%s: flush retry cmd\n", 2230 current->comm)); 2231 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2232 blk_mq_kick_requeue_list(sdev->request_queue); 2233 } else { 2234 /* 2235 * If just we got sense for the device (called 2236 * scsi_eh_get_sense), scmd->result is already 2237 * set, do not set DID_TIME_OUT. 2238 */ 2239 if (!scmd->result && 2240 !(scmd->flags & SCMD_FORCE_EH_SUCCESS)) 2241 scmd->result |= (DID_TIME_OUT << 16); 2242 SCSI_LOG_ERROR_RECOVERY(3, 2243 scmd_printk(KERN_INFO, scmd, 2244 "%s: flush finish cmd\n", 2245 current->comm)); 2246 scsi_finish_command(scmd); 2247 } 2248 } 2249 } 2250 EXPORT_SYMBOL(scsi_eh_flush_done_q); 2251 2252 /** 2253 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2254 * @shost: Host to unjam. 2255 * 2256 * Notes: 2257 * When we come in here, we *know* that all commands on the bus have 2258 * either completed, failed or timed out. we also know that no further 2259 * commands are being sent to the host, so things are relatively quiet 2260 * and we have freedom to fiddle with things as we wish. 2261 * 2262 * This is only the *default* implementation. it is possible for 2263 * individual drivers to supply their own version of this function, and 2264 * if the maintainer wishes to do this, it is strongly suggested that 2265 * this function be taken as a template and modified. this function 2266 * was designed to correctly handle problems for about 95% of the 2267 * different cases out there, and it should always provide at least a 2268 * reasonable amount of error recovery. 2269 * 2270 * Any command marked 'failed' or 'timeout' must eventually have 2271 * scsi_finish_cmd() called for it. we do all of the retry stuff 2272 * here, so when we restart the host after we return it should have an 2273 * empty queue. 2274 */ 2275 static void scsi_unjam_host(struct Scsi_Host *shost) 2276 { 2277 unsigned long flags; 2278 LIST_HEAD(eh_work_q); 2279 LIST_HEAD(eh_done_q); 2280 2281 spin_lock_irqsave(shost->host_lock, flags); 2282 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2283 spin_unlock_irqrestore(shost->host_lock, flags); 2284 2285 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2286 2287 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2288 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2289 2290 spin_lock_irqsave(shost->host_lock, flags); 2291 if (shost->eh_deadline != -1) 2292 shost->last_reset = 0; 2293 spin_unlock_irqrestore(shost->host_lock, flags); 2294 scsi_eh_flush_done_q(&eh_done_q); 2295 } 2296 2297 /** 2298 * scsi_error_handler - SCSI error handler thread 2299 * @data: Host for which we are running. 2300 * 2301 * Notes: 2302 * This is the main error handling loop. This is run as a kernel thread 2303 * for every SCSI host and handles all error handling activity. 2304 */ 2305 int scsi_error_handler(void *data) 2306 { 2307 struct Scsi_Host *shost = data; 2308 2309 /* 2310 * We use TASK_INTERRUPTIBLE so that the thread is not 2311 * counted against the load average as a running process. 2312 * We never actually get interrupted because kthread_run 2313 * disables signal delivery for the created thread. 2314 */ 2315 while (true) { 2316 /* 2317 * The sequence in kthread_stop() sets the stop flag first 2318 * then wakes the process. To avoid missed wakeups, the task 2319 * should always be in a non running state before the stop 2320 * flag is checked 2321 */ 2322 set_current_state(TASK_INTERRUPTIBLE); 2323 if (kthread_should_stop()) 2324 break; 2325 2326 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2327 shost->host_failed != scsi_host_busy(shost)) { 2328 SCSI_LOG_ERROR_RECOVERY(1, 2329 shost_printk(KERN_INFO, shost, 2330 "scsi_eh_%d: sleeping\n", 2331 shost->host_no)); 2332 schedule(); 2333 continue; 2334 } 2335 2336 __set_current_state(TASK_RUNNING); 2337 SCSI_LOG_ERROR_RECOVERY(1, 2338 shost_printk(KERN_INFO, shost, 2339 "scsi_eh_%d: waking up %d/%d/%d\n", 2340 shost->host_no, shost->host_eh_scheduled, 2341 shost->host_failed, 2342 scsi_host_busy(shost))); 2343 2344 /* 2345 * We have a host that is failing for some reason. Figure out 2346 * what we need to do to get it up and online again (if we can). 2347 * If we fail, we end up taking the thing offline. 2348 */ 2349 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { 2350 SCSI_LOG_ERROR_RECOVERY(1, 2351 shost_printk(KERN_ERR, shost, 2352 "scsi_eh_%d: unable to autoresume\n", 2353 shost->host_no)); 2354 continue; 2355 } 2356 2357 if (shost->transportt->eh_strategy_handler) 2358 shost->transportt->eh_strategy_handler(shost); 2359 else 2360 scsi_unjam_host(shost); 2361 2362 /* All scmds have been handled */ 2363 shost->host_failed = 0; 2364 2365 /* 2366 * Note - if the above fails completely, the action is to take 2367 * individual devices offline and flush the queue of any 2368 * outstanding requests that may have been pending. When we 2369 * restart, we restart any I/O to any other devices on the bus 2370 * which are still online. 2371 */ 2372 scsi_restart_operations(shost); 2373 if (!shost->eh_noresume) 2374 scsi_autopm_put_host(shost); 2375 } 2376 __set_current_state(TASK_RUNNING); 2377 2378 SCSI_LOG_ERROR_RECOVERY(1, 2379 shost_printk(KERN_INFO, shost, 2380 "Error handler scsi_eh_%d exiting\n", 2381 shost->host_no)); 2382 shost->ehandler = NULL; 2383 return 0; 2384 } 2385 2386 /** 2387 * scsi_report_bus_reset() - report bus reset observed 2388 * 2389 * Utility function used by low-level drivers to report that 2390 * they have observed a bus reset on the bus being handled. 2391 * 2392 * @shost: Host in question 2393 * @channel: channel on which reset was observed. 2394 * 2395 * Returns: Nothing 2396 * 2397 * Lock status: Host lock must be held. 2398 * 2399 * Notes: This only needs to be called if the reset is one which 2400 * originates from an unknown location. Resets originated 2401 * by the mid-level itself don't need to call this, but there 2402 * should be no harm. 2403 * 2404 * The main purpose of this is to make sure that a CHECK_CONDITION 2405 * is properly treated. 2406 */ 2407 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2408 { 2409 struct scsi_device *sdev; 2410 2411 __shost_for_each_device(sdev, shost) { 2412 if (channel == sdev_channel(sdev)) 2413 __scsi_report_device_reset(sdev, NULL); 2414 } 2415 } 2416 EXPORT_SYMBOL(scsi_report_bus_reset); 2417 2418 /** 2419 * scsi_report_device_reset() - report device reset observed 2420 * 2421 * Utility function used by low-level drivers to report that 2422 * they have observed a device reset on the device being handled. 2423 * 2424 * @shost: Host in question 2425 * @channel: channel on which reset was observed 2426 * @target: target on which reset was observed 2427 * 2428 * Returns: Nothing 2429 * 2430 * Lock status: Host lock must be held 2431 * 2432 * Notes: This only needs to be called if the reset is one which 2433 * originates from an unknown location. Resets originated 2434 * by the mid-level itself don't need to call this, but there 2435 * should be no harm. 2436 * 2437 * The main purpose of this is to make sure that a CHECK_CONDITION 2438 * is properly treated. 2439 */ 2440 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2441 { 2442 struct scsi_device *sdev; 2443 2444 __shost_for_each_device(sdev, shost) { 2445 if (channel == sdev_channel(sdev) && 2446 target == sdev_id(sdev)) 2447 __scsi_report_device_reset(sdev, NULL); 2448 } 2449 } 2450 EXPORT_SYMBOL(scsi_report_device_reset); 2451 2452 /** 2453 * scsi_ioctl_reset: explicitly reset a host/bus/target/device 2454 * @dev: scsi_device to operate on 2455 * @arg: reset type (see sg.h) 2456 */ 2457 int 2458 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) 2459 { 2460 struct scsi_cmnd *scmd; 2461 struct Scsi_Host *shost = dev->host; 2462 struct request *rq; 2463 unsigned long flags; 2464 int error = 0, val; 2465 enum scsi_disposition rtn; 2466 2467 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) 2468 return -EACCES; 2469 2470 error = get_user(val, arg); 2471 if (error) 2472 return error; 2473 2474 if (scsi_autopm_get_host(shost) < 0) 2475 return -EIO; 2476 2477 error = -EIO; 2478 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) + 2479 shost->hostt->cmd_size, GFP_KERNEL); 2480 if (!rq) 2481 goto out_put_autopm_host; 2482 blk_rq_init(NULL, rq); 2483 2484 scmd = (struct scsi_cmnd *)(rq + 1); 2485 scsi_init_command(dev, scmd); 2486 2487 scmd->submitter = SUBMITTED_BY_SCSI_RESET_IOCTL; 2488 scmd->flags |= SCMD_LAST; 2489 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 2490 2491 scmd->cmd_len = 0; 2492 2493 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 2494 2495 spin_lock_irqsave(shost->host_lock, flags); 2496 shost->tmf_in_progress = 1; 2497 spin_unlock_irqrestore(shost->host_lock, flags); 2498 2499 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { 2500 case SG_SCSI_RESET_NOTHING: 2501 rtn = SUCCESS; 2502 break; 2503 case SG_SCSI_RESET_DEVICE: 2504 rtn = scsi_try_bus_device_reset(scmd); 2505 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2506 break; 2507 fallthrough; 2508 case SG_SCSI_RESET_TARGET: 2509 rtn = scsi_try_target_reset(scmd); 2510 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2511 break; 2512 fallthrough; 2513 case SG_SCSI_RESET_BUS: 2514 rtn = scsi_try_bus_reset(scmd); 2515 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2516 break; 2517 fallthrough; 2518 case SG_SCSI_RESET_HOST: 2519 rtn = scsi_try_host_reset(scmd); 2520 if (rtn == SUCCESS) 2521 break; 2522 fallthrough; 2523 default: 2524 rtn = FAILED; 2525 break; 2526 } 2527 2528 error = (rtn == SUCCESS) ? 0 : -EIO; 2529 2530 spin_lock_irqsave(shost->host_lock, flags); 2531 shost->tmf_in_progress = 0; 2532 spin_unlock_irqrestore(shost->host_lock, flags); 2533 2534 /* 2535 * be sure to wake up anyone who was sleeping or had their queue 2536 * suspended while we performed the TMF. 2537 */ 2538 SCSI_LOG_ERROR_RECOVERY(3, 2539 shost_printk(KERN_INFO, shost, 2540 "waking up host to restart after TMF\n")); 2541 2542 wake_up(&shost->host_wait); 2543 scsi_run_host_queues(shost); 2544 2545 kfree(rq); 2546 2547 out_put_autopm_host: 2548 scsi_autopm_put_host(shost); 2549 return error; 2550 } 2551 2552 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, 2553 struct scsi_sense_hdr *sshdr) 2554 { 2555 return scsi_normalize_sense(cmd->sense_buffer, 2556 SCSI_SENSE_BUFFERSIZE, sshdr); 2557 } 2558 EXPORT_SYMBOL(scsi_command_normalize_sense); 2559 2560 /** 2561 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2562 * @sense_buffer: byte array of sense data 2563 * @sb_len: number of valid bytes in sense_buffer 2564 * @info_out: pointer to 64 integer where 8 or 4 byte information 2565 * field will be placed if found. 2566 * 2567 * Return value: 2568 * true if information field found, false if not found. 2569 */ 2570 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len, 2571 u64 *info_out) 2572 { 2573 const u8 * ucp; 2574 2575 if (sb_len < 7) 2576 return false; 2577 switch (sense_buffer[0] & 0x7f) { 2578 case 0x70: 2579 case 0x71: 2580 if (sense_buffer[0] & 0x80) { 2581 *info_out = get_unaligned_be32(&sense_buffer[3]); 2582 return true; 2583 } 2584 return false; 2585 case 0x72: 2586 case 0x73: 2587 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2588 0 /* info desc */); 2589 if (ucp && (0xa == ucp[1])) { 2590 *info_out = get_unaligned_be64(&ucp[4]); 2591 return true; 2592 } 2593 return false; 2594 default: 2595 return false; 2596 } 2597 } 2598 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2599