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 void scsi_rcu_eh_wakeup(struct work_struct *work) 77 { 78 struct Scsi_Host *shost = container_of(work, struct Scsi_Host, eh_work); 79 unsigned long flags; 80 unsigned int busy; 81 82 /* 83 * Ensure any running scsi_dec_host_busy has completed its rcu section 84 * so changes to host state and host_eh_scheduled are visible to all 85 * future calls of scsi_dec_host_busy 86 */ 87 synchronize_rcu(); 88 89 busy = scsi_host_busy(shost); 90 91 spin_lock_irqsave(shost->host_lock, flags); 92 scsi_eh_wakeup(shost, busy); 93 spin_unlock_irqrestore(shost->host_lock, flags); 94 } 95 96 /** 97 * scsi_schedule_eh - schedule EH for SCSI host 98 * @shost: SCSI host to invoke error handling on. 99 * 100 * Schedule SCSI EH without scmd. 101 */ 102 void scsi_schedule_eh(struct Scsi_Host *shost) 103 { 104 unsigned long flags; 105 106 spin_lock_irqsave(shost->host_lock, flags); 107 108 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || 109 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { 110 shost->host_eh_scheduled++; 111 queue_work(shost->tmf_work_q, &shost->eh_work); 112 } 113 114 spin_unlock_irqrestore(shost->host_lock, flags); 115 } 116 EXPORT_SYMBOL_GPL(scsi_schedule_eh); 117 118 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost) 119 { 120 if (!shost->last_reset || shost->eh_deadline == -1) 121 return 0; 122 123 /* 124 * 32bit accesses are guaranteed to be atomic 125 * (on all supported architectures), so instead 126 * of using a spinlock we can as well double check 127 * if eh_deadline has been set to 'off' during the 128 * time_before call. 129 */ 130 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) && 131 shost->eh_deadline > -1) 132 return 0; 133 134 return 1; 135 } 136 137 static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd) 138 { 139 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) 140 return true; 141 142 return ++cmd->retries <= cmd->allowed; 143 } 144 145 static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd) 146 { 147 struct scsi_device *sdev = cmd->device; 148 struct Scsi_Host *host = sdev->host; 149 150 if (host->hostt->eh_should_retry_cmd) 151 return host->hostt->eh_should_retry_cmd(cmd); 152 153 return true; 154 } 155 156 /** 157 * scmd_eh_abort_handler - Handle command aborts 158 * @work: command to be aborted. 159 * 160 * Note: this function must be called only for a command that has timed out. 161 * Because the block layer marks a request as complete before it calls 162 * scsi_timeout(), a .scsi_done() call from the LLD for a command that has 163 * timed out do not have any effect. Hence it is safe to call 164 * scsi_finish_command() from this function. 165 */ 166 void 167 scmd_eh_abort_handler(struct work_struct *work) 168 { 169 struct scsi_cmnd *scmd = 170 container_of(work, struct scsi_cmnd, abort_work.work); 171 struct scsi_device *sdev = scmd->device; 172 struct Scsi_Host *shost = sdev->host; 173 enum scsi_disposition rtn; 174 unsigned long flags; 175 176 if (scsi_host_eh_past_deadline(shost)) { 177 SCSI_LOG_ERROR_RECOVERY(3, 178 scmd_printk(KERN_INFO, scmd, 179 "eh timeout, not aborting\n")); 180 goto out; 181 } 182 183 SCSI_LOG_ERROR_RECOVERY(3, 184 scmd_printk(KERN_INFO, scmd, 185 "aborting command\n")); 186 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd); 187 if (rtn != SUCCESS) { 188 SCSI_LOG_ERROR_RECOVERY(3, 189 scmd_printk(KERN_INFO, scmd, 190 "cmd abort %s\n", 191 (rtn == FAST_IO_FAIL) ? 192 "not send" : "failed")); 193 goto out; 194 } 195 set_host_byte(scmd, DID_TIME_OUT); 196 if (scsi_host_eh_past_deadline(shost)) { 197 SCSI_LOG_ERROR_RECOVERY(3, 198 scmd_printk(KERN_INFO, scmd, 199 "eh timeout, not retrying " 200 "aborted command\n")); 201 goto out; 202 } 203 204 spin_lock_irqsave(shost->host_lock, flags); 205 list_del_init(&scmd->eh_entry); 206 207 /* 208 * If the abort succeeds, and there is no further 209 * EH action, clear the ->last_reset time. 210 */ 211 if (list_empty(&shost->eh_abort_list) && 212 list_empty(&shost->eh_cmd_q)) 213 if (shost->eh_deadline != -1) 214 shost->last_reset = 0; 215 216 spin_unlock_irqrestore(shost->host_lock, flags); 217 218 if (!scsi_noretry_cmd(scmd) && 219 scsi_cmd_retry_allowed(scmd) && 220 scsi_eh_should_retry_cmd(scmd)) { 221 SCSI_LOG_ERROR_RECOVERY(3, 222 scmd_printk(KERN_WARNING, scmd, 223 "retry aborted command\n")); 224 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 225 } else { 226 SCSI_LOG_ERROR_RECOVERY(3, 227 scmd_printk(KERN_WARNING, scmd, 228 "finish aborted command\n")); 229 scsi_finish_command(scmd); 230 } 231 return; 232 233 out: 234 spin_lock_irqsave(shost->host_lock, flags); 235 list_del_init(&scmd->eh_entry); 236 spin_unlock_irqrestore(shost->host_lock, flags); 237 238 scsi_eh_scmd_add(scmd); 239 } 240 241 /** 242 * scsi_abort_command - schedule a command abort 243 * @scmd: scmd to abort. 244 * 245 * We only need to abort commands after a command timeout 246 */ 247 static int 248 scsi_abort_command(struct scsi_cmnd *scmd) 249 { 250 struct scsi_device *sdev = scmd->device; 251 struct Scsi_Host *shost = sdev->host; 252 unsigned long flags; 253 254 if (!shost->hostt->eh_abort_handler) { 255 /* No abort handler, fail command directly */ 256 return FAILED; 257 } 258 259 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 260 /* 261 * Retry after abort failed, escalate to next level. 262 */ 263 SCSI_LOG_ERROR_RECOVERY(3, 264 scmd_printk(KERN_INFO, scmd, 265 "previous abort failed\n")); 266 BUG_ON(delayed_work_pending(&scmd->abort_work)); 267 return FAILED; 268 } 269 270 spin_lock_irqsave(shost->host_lock, flags); 271 if (shost->eh_deadline != -1 && !shost->last_reset) 272 shost->last_reset = jiffies; 273 BUG_ON(!list_empty(&scmd->eh_entry)); 274 list_add_tail(&scmd->eh_entry, &shost->eh_abort_list); 275 spin_unlock_irqrestore(shost->host_lock, flags); 276 277 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; 278 SCSI_LOG_ERROR_RECOVERY(3, 279 scmd_printk(KERN_INFO, scmd, "abort scheduled\n")); 280 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100); 281 return SUCCESS; 282 } 283 284 /** 285 * scsi_eh_reset - call into ->eh_action to reset internal counters 286 * @scmd: scmd to run eh on. 287 * 288 * The scsi driver might be carrying internal state about the 289 * devices, so we need to call into the driver to reset the 290 * internal state once the error handler is started. 291 */ 292 static void scsi_eh_reset(struct scsi_cmnd *scmd) 293 { 294 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) { 295 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 296 if (sdrv->eh_reset) 297 sdrv->eh_reset(scmd); 298 } 299 } 300 301 static void scsi_eh_inc_host_failed(struct rcu_head *head) 302 { 303 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu); 304 struct Scsi_Host *shost = scmd->device->host; 305 unsigned int busy; 306 unsigned long flags; 307 308 spin_lock_irqsave(shost->host_lock, flags); 309 shost->host_failed++; 310 spin_unlock_irqrestore(shost->host_lock, flags); 311 /* 312 * The counting of busy requests needs to occur after adding to 313 * host_failed or after the lock acquire for adding to host_failed 314 * to prevent a race with host unbusy and missing an eh wakeup. 315 */ 316 busy = scsi_host_busy(shost); 317 318 spin_lock_irqsave(shost->host_lock, flags); 319 scsi_eh_wakeup(shost, busy); 320 spin_unlock_irqrestore(shost->host_lock, flags); 321 } 322 323 /** 324 * scsi_eh_scmd_add - add scsi cmd to error handling. 325 * @scmd: scmd to run eh on. 326 */ 327 void scsi_eh_scmd_add(struct scsi_cmnd *scmd) 328 { 329 struct Scsi_Host *shost = scmd->device->host; 330 unsigned long flags; 331 int ret; 332 333 WARN_ON_ONCE(!shost->ehandler); 334 WARN_ON_ONCE(!test_bit(SCMD_STATE_INFLIGHT, &scmd->state)); 335 336 spin_lock_irqsave(shost->host_lock, flags); 337 if (scsi_host_set_state(shost, SHOST_RECOVERY)) { 338 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY); 339 WARN_ON_ONCE(ret); 340 } 341 if (shost->eh_deadline != -1 && !shost->last_reset) 342 shost->last_reset = jiffies; 343 344 scsi_eh_reset(scmd); 345 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 346 spin_unlock_irqrestore(shost->host_lock, flags); 347 /* 348 * Ensure that all tasks observe the host state change before the 349 * host_failed change. 350 */ 351 call_rcu_hurry(&scmd->rcu, scsi_eh_inc_host_failed); 352 } 353 354 /** 355 * scsi_timeout - Timeout function for normal scsi commands. 356 * @req: request that is timing out. 357 * 358 * Notes: 359 * We do not need to lock this. There is the potential for a race 360 * only in that the normal completion handling might run, but if the 361 * normal completion function determines that the timer has already 362 * fired, then it mustn't do anything. 363 */ 364 enum blk_eh_timer_return scsi_timeout(struct request *req) 365 { 366 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req); 367 struct Scsi_Host *host = scmd->device->host; 368 369 trace_scsi_dispatch_cmd_timeout(scmd); 370 scsi_log_completion(scmd, TIMEOUT_ERROR); 371 372 atomic_inc(&scmd->device->iotmo_cnt); 373 if (host->eh_deadline != -1 && !host->last_reset) 374 host->last_reset = jiffies; 375 376 if (host->hostt->eh_timed_out) { 377 switch (host->hostt->eh_timed_out(scmd)) { 378 case SCSI_EH_DONE: 379 return BLK_EH_DONE; 380 case SCSI_EH_RESET_TIMER: 381 return BLK_EH_RESET_TIMER; 382 case SCSI_EH_NOT_HANDLED: 383 break; 384 } 385 } 386 387 /* 388 * If scsi_done() has already set SCMD_STATE_COMPLETE, do not modify 389 * *scmd. 390 */ 391 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state)) 392 return BLK_EH_DONE; 393 atomic_inc(&scmd->device->iodone_cnt); 394 if (scsi_abort_command(scmd) != SUCCESS) { 395 set_host_byte(scmd, DID_TIME_OUT); 396 scsi_eh_scmd_add(scmd); 397 } 398 399 return BLK_EH_DONE; 400 } 401 402 /** 403 * scsi_block_when_processing_errors - Prevent cmds from being queued. 404 * @sdev: Device on which we are performing recovery. 405 * 406 * Description: 407 * We block until the host is out of error recovery, and then check to 408 * see whether the host or the device is offline. 409 * 410 * Return value: 411 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 412 */ 413 int scsi_block_when_processing_errors(struct scsi_device *sdev) 414 { 415 int online; 416 417 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 418 419 online = scsi_device_online(sdev); 420 421 return online; 422 } 423 EXPORT_SYMBOL(scsi_block_when_processing_errors); 424 425 #ifdef CONFIG_SCSI_LOGGING 426 /** 427 * scsi_eh_prt_fail_stats - Log info on failures. 428 * @shost: scsi host being recovered. 429 * @work_q: Queue of scsi cmds to process. 430 */ 431 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 432 struct list_head *work_q) 433 { 434 struct scsi_cmnd *scmd; 435 struct scsi_device *sdev; 436 int total_failures = 0; 437 int cmd_failed = 0; 438 int cmd_cancel = 0; 439 int devices_failed = 0; 440 441 shost_for_each_device(sdev, shost) { 442 list_for_each_entry(scmd, work_q, eh_entry) { 443 if (scmd->device == sdev) { 444 ++total_failures; 445 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) 446 ++cmd_cancel; 447 else 448 ++cmd_failed; 449 } 450 } 451 452 if (cmd_cancel || cmd_failed) { 453 SCSI_LOG_ERROR_RECOVERY(3, 454 shost_printk(KERN_INFO, shost, 455 "%s: cmds failed: %d, cancel: %d\n", 456 __func__, cmd_failed, 457 cmd_cancel)); 458 cmd_cancel = 0; 459 cmd_failed = 0; 460 ++devices_failed; 461 } 462 } 463 464 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, 465 "Total of %d commands on %d" 466 " devices require eh work\n", 467 total_failures, devices_failed)); 468 } 469 #endif 470 471 /** 472 * scsi_report_lun_change - Set flag on all *other* devices on the same target 473 * to indicate that a UNIT ATTENTION is expected. 474 * @sdev: Device reporting the UNIT ATTENTION 475 */ 476 static void scsi_report_lun_change(struct scsi_device *sdev) 477 { 478 sdev->sdev_target->expecting_lun_change = 1; 479 } 480 481 /** 482 * scsi_report_sense - Examine scsi sense information and log messages for 483 * certain conditions, also issue uevents for some of them. 484 * @sdev: Device reporting the sense code 485 * @sshdr: sshdr to be examined 486 */ 487 static void scsi_report_sense(struct scsi_device *sdev, 488 struct scsi_sense_hdr *sshdr) 489 { 490 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ 491 492 if (sshdr->sense_key == UNIT_ATTENTION) { 493 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { 494 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; 495 sdev_printk(KERN_WARNING, sdev, 496 "Inquiry data has changed"); 497 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { 498 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; 499 scsi_report_lun_change(sdev); 500 sdev_printk(KERN_WARNING, sdev, 501 "LUN assignments on this target have " 502 "changed. The Linux SCSI layer does not " 503 "automatically remap LUN assignments.\n"); 504 } else if (sshdr->asc == 0x3f) 505 sdev_printk(KERN_WARNING, sdev, 506 "Operating parameters on this target have " 507 "changed. The Linux SCSI layer does not " 508 "automatically adjust these parameters.\n"); 509 510 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { 511 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; 512 sdev_printk(KERN_WARNING, sdev, 513 "Warning! Received an indication that the " 514 "LUN reached a thin provisioning soft " 515 "threshold.\n"); 516 } 517 518 if (sshdr->asc == 0x29) { 519 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED; 520 /* 521 * Do not print message if it is an expected side-effect 522 * of runtime PM. 523 */ 524 if (!sdev->silence_suspend) 525 sdev_printk(KERN_WARNING, sdev, 526 "Power-on or device reset occurred\n"); 527 } 528 529 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { 530 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; 531 sdev_printk(KERN_WARNING, sdev, 532 "Mode parameters changed"); 533 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { 534 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; 535 sdev_printk(KERN_WARNING, sdev, 536 "Asymmetric access state changed"); 537 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { 538 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; 539 sdev_printk(KERN_WARNING, sdev, 540 "Capacity data has changed"); 541 } else if (sshdr->asc == 0x2a) 542 sdev_printk(KERN_WARNING, sdev, 543 "Parameters changed"); 544 } 545 546 if (evt_type != SDEV_EVT_MAXBITS) { 547 set_bit(evt_type, sdev->pending_events); 548 schedule_work(&sdev->event_work); 549 } 550 } 551 552 static inline void set_scsi_ml_byte(struct scsi_cmnd *cmd, u8 status) 553 { 554 cmd->result = (cmd->result & 0xffff00ff) | (status << 8); 555 } 556 557 /** 558 * scsi_check_sense - Examine scsi cmd sense 559 * @scmd: Cmd to have sense checked. 560 * 561 * Return value: 562 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE 563 * 564 * Notes: 565 * When a deferred error is detected the current command has 566 * not been executed and needs retrying. 567 */ 568 enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd) 569 { 570 struct request *req = scsi_cmd_to_rq(scmd); 571 struct scsi_device *sdev = scmd->device; 572 struct scsi_sense_hdr sshdr; 573 574 if (! scsi_command_normalize_sense(scmd, &sshdr)) 575 return FAILED; /* no valid sense data */ 576 577 scsi_report_sense(sdev, &sshdr); 578 579 if (sshdr.sense_key == UNIT_ATTENTION) { 580 /* 581 * Increment the counters for Power on/Reset or New Media so 582 * that all ULDs interested in these can see that those have 583 * happened, even if someone else gets the sense data. 584 */ 585 if (sshdr.asc == 0x28) 586 atomic_inc(&sdev->ua_new_media_ctr); 587 else if (sshdr.asc == 0x29) 588 atomic_inc(&sdev->ua_por_ctr); 589 } 590 591 if (scsi_sense_is_deferred(&sshdr)) 592 return NEEDS_RETRY; 593 594 if (sdev->handler && sdev->handler->check_sense) { 595 enum scsi_disposition rc; 596 597 rc = sdev->handler->check_sense(sdev, &sshdr); 598 if (rc != SCSI_RETURN_NOT_HANDLED) 599 return rc; 600 /* handler does not care. Drop down to default handling */ 601 } 602 603 if (scmd->cmnd[0] == TEST_UNIT_READY && 604 scmd->submitter != SUBMITTED_BY_SCSI_ERROR_HANDLER) 605 /* 606 * nasty: for mid-layer issued TURs, we need to return the 607 * actual sense data without any recovery attempt. For eh 608 * issued ones, we need to try to recover and interpret 609 */ 610 return SUCCESS; 611 612 /* 613 * Previous logic looked for FILEMARK, EOM or ILI which are 614 * mainly associated with tapes and returned SUCCESS. 615 */ 616 if (sshdr.response_code == 0x70) { 617 /* fixed format */ 618 if (scmd->sense_buffer[2] & 0xe0) 619 return SUCCESS; 620 } else { 621 /* 622 * descriptor format: look for "stream commands sense data 623 * descriptor" (see SSC-3). Assume single sense data 624 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 625 */ 626 if ((sshdr.additional_length > 3) && 627 (scmd->sense_buffer[8] == 0x4) && 628 (scmd->sense_buffer[11] & 0xe0)) 629 return SUCCESS; 630 } 631 632 switch (sshdr.sense_key) { 633 case NO_SENSE: 634 return SUCCESS; 635 case RECOVERED_ERROR: 636 return /* soft_error */ SUCCESS; 637 638 case ABORTED_COMMAND: 639 if (sshdr.asc == 0x10) /* DIF */ 640 return SUCCESS; 641 642 /* 643 * Check aborts due to command duration limit policy: 644 * ABORTED COMMAND additional sense code with the 645 * COMMAND TIMEOUT BEFORE PROCESSING or 646 * COMMAND TIMEOUT DURING PROCESSING or 647 * COMMAND TIMEOUT DURING PROCESSING DUE TO ERROR RECOVERY 648 * additional sense code qualifiers. 649 */ 650 if (sshdr.asc == 0x2e && 651 sshdr.ascq >= 0x01 && sshdr.ascq <= 0x03) { 652 set_scsi_ml_byte(scmd, SCSIML_STAT_DL_TIMEOUT); 653 req->cmd_flags |= REQ_FAILFAST_DEV; 654 req->rq_flags |= RQF_QUIET; 655 return SUCCESS; 656 } 657 658 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF) 659 return ADD_TO_MLQUEUE; 660 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 && 661 sdev->sdev_bflags & BLIST_RETRY_ASC_C1) 662 return ADD_TO_MLQUEUE; 663 664 return NEEDS_RETRY; 665 case NOT_READY: 666 case UNIT_ATTENTION: 667 /* 668 * if we are expecting a cc/ua because of a bus reset that we 669 * performed, treat this just as a retry. otherwise this is 670 * information that we should pass up to the upper-level driver 671 * so that we can deal with it there. 672 */ 673 if (scmd->device->expecting_cc_ua) { 674 /* 675 * Because some device does not queue unit 676 * attentions correctly, we carefully check 677 * additional sense code and qualifier so as 678 * not to squash media change unit attention. 679 */ 680 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { 681 scmd->device->expecting_cc_ua = 0; 682 return NEEDS_RETRY; 683 } 684 } 685 /* 686 * we might also expect a cc/ua if another LUN on the target 687 * reported a UA with an ASC/ASCQ of 3F 0E - 688 * REPORTED LUNS DATA HAS CHANGED. 689 */ 690 if (scmd->device->sdev_target->expecting_lun_change && 691 sshdr.asc == 0x3f && sshdr.ascq == 0x0e) 692 return NEEDS_RETRY; 693 /* 694 * if the device is in the process of becoming ready, we 695 * should retry. 696 */ 697 if ((sshdr.asc == 0x04) && 698 (sshdr.ascq == 0x01 || sshdr.ascq == 0x0a)) 699 return NEEDS_RETRY; 700 /* 701 * if the device is not started, we need to wake 702 * the error handler to start the motor 703 */ 704 if (scmd->device->allow_restart && 705 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 706 return FAILED; 707 /* 708 * Pass the UA upwards for a determination in the completion 709 * functions. 710 */ 711 return SUCCESS; 712 713 /* these are not supported */ 714 case DATA_PROTECT: 715 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { 716 /* Thin provisioning hard threshold reached */ 717 set_scsi_ml_byte(scmd, SCSIML_STAT_NOSPC); 718 return SUCCESS; 719 } 720 fallthrough; 721 case COPY_ABORTED: 722 case VOLUME_OVERFLOW: 723 case MISCOMPARE: 724 case BLANK_CHECK: 725 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE); 726 return SUCCESS; 727 728 case MEDIUM_ERROR: 729 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 730 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 731 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 732 set_scsi_ml_byte(scmd, SCSIML_STAT_MED_ERROR); 733 return SUCCESS; 734 } 735 return NEEDS_RETRY; 736 737 case HARDWARE_ERROR: 738 if (scmd->device->retry_hwerror) 739 return ADD_TO_MLQUEUE; 740 else 741 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE); 742 fallthrough; 743 744 case ILLEGAL_REQUEST: 745 if (sshdr.asc == 0x20 || /* Invalid command operation code */ 746 sshdr.asc == 0x21 || /* Logical block address out of range */ 747 sshdr.asc == 0x22 || /* Invalid function */ 748 sshdr.asc == 0x24 || /* Invalid field in cdb */ 749 sshdr.asc == 0x26 || /* Parameter value invalid */ 750 sshdr.asc == 0x27) { /* Write protected */ 751 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE); 752 } 753 return SUCCESS; 754 755 case COMPLETED: 756 /* 757 * A command using command duration limits (CDL) with a 758 * descriptor set with policy 0xD may be completed with success 759 * and the sense data DATA CURRENTLY UNAVAILABLE, indicating 760 * that the command was in fact aborted because it exceeded its 761 * duration limit. Never retry these commands. 762 */ 763 if (sshdr.asc == 0x55 && sshdr.ascq == 0x0a) { 764 set_scsi_ml_byte(scmd, SCSIML_STAT_DL_TIMEOUT); 765 req->cmd_flags |= REQ_FAILFAST_DEV; 766 req->rq_flags |= RQF_QUIET; 767 } 768 return SUCCESS; 769 770 default: 771 return SUCCESS; 772 } 773 } 774 EXPORT_SYMBOL_GPL(scsi_check_sense); 775 776 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) 777 { 778 const struct scsi_host_template *sht = sdev->host->hostt; 779 struct scsi_device *tmp_sdev; 780 781 if (!sdev->budget_map.map) 782 return; 783 784 if (!sht->track_queue_depth || 785 sdev->queue_depth >= sdev->max_queue_depth) 786 return; 787 788 if (time_before(jiffies, 789 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) 790 return; 791 792 if (time_before(jiffies, 793 sdev->last_queue_full_time + sdev->queue_ramp_up_period)) 794 return; 795 796 /* 797 * Walk all devices of a target and do 798 * ramp up on them. 799 */ 800 shost_for_each_device(tmp_sdev, sdev->host) { 801 if (tmp_sdev->channel != sdev->channel || 802 tmp_sdev->id != sdev->id || 803 tmp_sdev->queue_depth == sdev->max_queue_depth) 804 continue; 805 806 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); 807 sdev->last_queue_ramp_up = jiffies; 808 } 809 } 810 811 static void scsi_handle_queue_full(struct scsi_device *sdev) 812 { 813 const struct scsi_host_template *sht = sdev->host->hostt; 814 struct scsi_device *tmp_sdev; 815 816 if (!sht->track_queue_depth) 817 return; 818 819 shost_for_each_device(tmp_sdev, sdev->host) { 820 if (tmp_sdev->channel != sdev->channel || 821 tmp_sdev->id != sdev->id) 822 continue; 823 /* 824 * We do not know the number of commands that were at 825 * the device when we got the queue full so we start 826 * from the highest possible value and work our way down. 827 */ 828 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); 829 } 830 } 831 832 /** 833 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 834 * @scmd: SCSI cmd to examine. 835 * 836 * Notes: 837 * This is *only* called when we are examining the status of commands 838 * queued during error recovery. the main difference here is that we 839 * don't allow for the possibility of retries here, and we are a lot 840 * more restrictive about what we consider acceptable. 841 */ 842 static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd) 843 { 844 /* 845 * first check the host byte, to see if there is anything in there 846 * that would indicate what we need to do. 847 */ 848 if (host_byte(scmd->result) == DID_RESET) { 849 /* 850 * rats. we are already in the error handler, so we now 851 * get to try and figure out what to do next. if the sense 852 * is valid, we have a pretty good idea of what to do. 853 * if not, we mark it as FAILED. 854 */ 855 return scsi_check_sense(scmd); 856 } 857 if (host_byte(scmd->result) != DID_OK) 858 return FAILED; 859 860 /* 861 * now, check the status byte to see if this indicates 862 * anything special. 863 */ 864 switch (get_status_byte(scmd)) { 865 case SAM_STAT_GOOD: 866 scsi_handle_queue_ramp_up(scmd->device); 867 if (scmd->sense_buffer && SCSI_SENSE_VALID(scmd)) 868 /* 869 * If we have sense data, call scsi_check_sense() in 870 * order to set the correct SCSI ML byte (if any). 871 * No point in checking the return value, since the 872 * command has already completed successfully. 873 */ 874 scsi_check_sense(scmd); 875 fallthrough; 876 case SAM_STAT_COMMAND_TERMINATED: 877 return SUCCESS; 878 case SAM_STAT_CHECK_CONDITION: 879 return scsi_check_sense(scmd); 880 case SAM_STAT_CONDITION_MET: 881 case SAM_STAT_INTERMEDIATE: 882 case SAM_STAT_INTERMEDIATE_CONDITION_MET: 883 /* 884 * who knows? FIXME(eric) 885 */ 886 return SUCCESS; 887 case SAM_STAT_RESERVATION_CONFLICT: 888 if (scmd->cmnd[0] == TEST_UNIT_READY) 889 /* it is a success, we probed the device and 890 * found it */ 891 return SUCCESS; 892 /* otherwise, we failed to send the command */ 893 return FAILED; 894 case SAM_STAT_TASK_SET_FULL: 895 scsi_handle_queue_full(scmd->device); 896 fallthrough; 897 case SAM_STAT_BUSY: 898 return NEEDS_RETRY; 899 default: 900 return FAILED; 901 } 902 return FAILED; 903 } 904 905 /** 906 * scsi_eh_done - Completion function for error handling. 907 * @scmd: Cmd that is done. 908 */ 909 void scsi_eh_done(struct scsi_cmnd *scmd) 910 { 911 struct completion *eh_action; 912 913 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 914 "%s result: %x\n", __func__, scmd->result)); 915 916 eh_action = scmd->device->host->eh_action; 917 if (eh_action) 918 complete(eh_action); 919 } 920 921 /** 922 * scsi_try_host_reset - ask host adapter to reset itself 923 * @scmd: SCSI cmd to send host reset. 924 */ 925 static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd) 926 { 927 unsigned long flags; 928 enum scsi_disposition rtn; 929 struct Scsi_Host *host = scmd->device->host; 930 const struct scsi_host_template *hostt = host->hostt; 931 932 SCSI_LOG_ERROR_RECOVERY(3, 933 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 934 935 if (!hostt->eh_host_reset_handler) 936 return FAILED; 937 938 rtn = hostt->eh_host_reset_handler(scmd); 939 940 if (rtn == SUCCESS) { 941 if (!hostt->skip_settle_delay) 942 ssleep(HOST_RESET_SETTLE_TIME); 943 spin_lock_irqsave(host->host_lock, flags); 944 scsi_report_bus_reset(host, scmd_channel(scmd)); 945 spin_unlock_irqrestore(host->host_lock, flags); 946 } 947 948 return rtn; 949 } 950 951 /** 952 * scsi_try_bus_reset - ask host to perform a bus reset 953 * @scmd: SCSI cmd to send bus reset. 954 */ 955 static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd) 956 { 957 unsigned long flags; 958 enum scsi_disposition rtn; 959 struct Scsi_Host *host = scmd->device->host; 960 const struct scsi_host_template *hostt = host->hostt; 961 962 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 963 "%s: Snd Bus RST\n", __func__)); 964 965 if (!hostt->eh_bus_reset_handler) 966 return FAILED; 967 968 rtn = hostt->eh_bus_reset_handler(scmd); 969 970 if (rtn == SUCCESS) { 971 if (!hostt->skip_settle_delay) 972 ssleep(BUS_RESET_SETTLE_TIME); 973 spin_lock_irqsave(host->host_lock, flags); 974 scsi_report_bus_reset(host, scmd_channel(scmd)); 975 spin_unlock_irqrestore(host->host_lock, flags); 976 } 977 978 return rtn; 979 } 980 981 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 982 { 983 sdev->was_reset = 1; 984 sdev->expecting_cc_ua = 1; 985 } 986 987 /** 988 * scsi_try_target_reset - Ask host to perform a target reset 989 * @scmd: SCSI cmd used to send a target reset 990 * 991 * Notes: 992 * There is no timeout for this operation. if this operation is 993 * unreliable for a given host, then the host itself needs to put a 994 * timer on it, and set the host back to a consistent state prior to 995 * returning. 996 */ 997 static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd) 998 { 999 unsigned long flags; 1000 enum scsi_disposition rtn; 1001 struct Scsi_Host *host = scmd->device->host; 1002 const struct scsi_host_template *hostt = host->hostt; 1003 1004 if (!hostt->eh_target_reset_handler) 1005 return FAILED; 1006 1007 rtn = hostt->eh_target_reset_handler(scmd); 1008 if (rtn == SUCCESS) { 1009 spin_lock_irqsave(host->host_lock, flags); 1010 __starget_for_each_device(scsi_target(scmd->device), NULL, 1011 __scsi_report_device_reset); 1012 spin_unlock_irqrestore(host->host_lock, flags); 1013 } 1014 1015 return rtn; 1016 } 1017 1018 /** 1019 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 1020 * @scmd: SCSI cmd used to send BDR 1021 * 1022 * Notes: 1023 * There is no timeout for this operation. if this operation is 1024 * unreliable for a given host, then the host itself needs to put a 1025 * timer on it, and set the host back to a consistent state prior to 1026 * returning. 1027 */ 1028 static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 1029 { 1030 enum scsi_disposition rtn; 1031 const struct scsi_host_template *hostt = scmd->device->host->hostt; 1032 1033 if (!hostt->eh_device_reset_handler) 1034 return FAILED; 1035 1036 rtn = hostt->eh_device_reset_handler(scmd); 1037 if (rtn == SUCCESS) 1038 __scsi_report_device_reset(scmd->device, NULL); 1039 return rtn; 1040 } 1041 1042 /** 1043 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command 1044 * @hostt: SCSI driver host template 1045 * @scmd: SCSI cmd used to send a target reset 1046 * 1047 * Return value: 1048 * SUCCESS, FAILED, or FAST_IO_FAIL 1049 * 1050 * Notes: 1051 * SUCCESS does not necessarily indicate that the command 1052 * has been aborted; it only indicates that the LLDDs 1053 * has cleared all references to that command. 1054 * LLDDs should return FAILED only if an abort was required 1055 * but could not be executed. LLDDs should return FAST_IO_FAIL 1056 * if the device is temporarily unavailable (eg due to a 1057 * link down on FibreChannel) 1058 */ 1059 static enum scsi_disposition 1060 scsi_try_to_abort_cmd(const struct scsi_host_template *hostt, struct scsi_cmnd *scmd) 1061 { 1062 if (!hostt->eh_abort_handler) 1063 return FAILED; 1064 1065 return hostt->eh_abort_handler(scmd); 1066 } 1067 1068 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 1069 { 1070 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 1071 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 1072 if (scsi_try_target_reset(scmd) != SUCCESS) 1073 if (scsi_try_bus_reset(scmd) != SUCCESS) 1074 scsi_try_host_reset(scmd); 1075 } 1076 1077 /** 1078 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 1079 * @scmd: SCSI command structure to hijack 1080 * @ses: structure to save restore information 1081 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 1082 * @cmnd_size: size in bytes of @cmnd (must be <= MAX_COMMAND_SIZE) 1083 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 1084 * 1085 * This function is used to save a scsi command information before re-execution 1086 * as part of the error recovery process. If @sense_bytes is 0 the command 1087 * sent must be one that does not transfer any data. If @sense_bytes != 0 1088 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 1089 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 1090 */ 1091 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 1092 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 1093 { 1094 struct scsi_device *sdev = scmd->device; 1095 #ifdef CONFIG_BLK_INLINE_ENCRYPTION 1096 struct request *rq = scsi_cmd_to_rq(scmd); 1097 #endif 1098 1099 /* 1100 * We need saved copies of a number of fields - this is because 1101 * error handling may need to overwrite these with different values 1102 * to run different commands, and once error handling is complete, 1103 * we will need to restore these values prior to running the actual 1104 * command. 1105 */ 1106 ses->cmd_len = scmd->cmd_len; 1107 ses->data_direction = scmd->sc_data_direction; 1108 ses->sdb = scmd->sdb; 1109 ses->result = scmd->result; 1110 ses->resid_len = scmd->resid_len; 1111 ses->underflow = scmd->underflow; 1112 ses->prot_op = scmd->prot_op; 1113 ses->eh_eflags = scmd->eh_eflags; 1114 1115 scmd->prot_op = SCSI_PROT_NORMAL; 1116 scmd->eh_eflags = 0; 1117 memcpy(ses->cmnd, scmd->cmnd, sizeof(ses->cmnd)); 1118 memset(scmd->cmnd, 0, sizeof(scmd->cmnd)); 1119 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 1120 scmd->result = 0; 1121 scmd->resid_len = 0; 1122 1123 if (sense_bytes) { 1124 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 1125 sense_bytes); 1126 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 1127 scmd->sdb.length); 1128 scmd->sdb.table.sgl = &ses->sense_sgl; 1129 scmd->sc_data_direction = DMA_FROM_DEVICE; 1130 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; 1131 scmd->cmnd[0] = REQUEST_SENSE; 1132 scmd->cmnd[4] = scmd->sdb.length; 1133 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1134 } else { 1135 scmd->sc_data_direction = DMA_NONE; 1136 if (cmnd) { 1137 BUG_ON(cmnd_size > sizeof(scmd->cmnd)); 1138 memcpy(scmd->cmnd, cmnd, cmnd_size); 1139 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1140 } 1141 } 1142 1143 scmd->underflow = 0; 1144 1145 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 1146 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 1147 (sdev->lun << 5 & 0xe0); 1148 1149 /* 1150 * Encryption must be disabled for the commands submitted by the error handler. 1151 * Hence, clear the encryption context information. 1152 */ 1153 #ifdef CONFIG_BLK_INLINE_ENCRYPTION 1154 ses->rq_crypt_keyslot = rq->crypt_keyslot; 1155 ses->rq_crypt_ctx = rq->crypt_ctx; 1156 1157 rq->crypt_keyslot = NULL; 1158 rq->crypt_ctx = NULL; 1159 #endif 1160 1161 /* 1162 * Zero the sense buffer. The scsi spec mandates that any 1163 * untransferred sense data should be interpreted as being zero. 1164 */ 1165 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 1166 } 1167 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 1168 1169 /** 1170 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 1171 * @scmd: SCSI command structure to restore 1172 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 1173 * 1174 * Undo any damage done by above scsi_eh_prep_cmnd(). 1175 */ 1176 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 1177 { 1178 #ifdef CONFIG_BLK_INLINE_ENCRYPTION 1179 struct request *rq = scsi_cmd_to_rq(scmd); 1180 #endif 1181 1182 /* 1183 * Restore original data 1184 */ 1185 scmd->cmd_len = ses->cmd_len; 1186 memcpy(scmd->cmnd, ses->cmnd, sizeof(ses->cmnd)); 1187 scmd->sc_data_direction = ses->data_direction; 1188 scmd->sdb = ses->sdb; 1189 scmd->result = ses->result; 1190 scmd->resid_len = ses->resid_len; 1191 scmd->underflow = ses->underflow; 1192 scmd->prot_op = ses->prot_op; 1193 scmd->eh_eflags = ses->eh_eflags; 1194 1195 #ifdef CONFIG_BLK_INLINE_ENCRYPTION 1196 rq->crypt_keyslot = ses->rq_crypt_keyslot; 1197 rq->crypt_ctx = ses->rq_crypt_ctx; 1198 #endif 1199 } 1200 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 1201 1202 /** 1203 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 1204 * @scmd: SCSI command structure to hijack 1205 * @cmnd: CDB to send 1206 * @cmnd_size: size in bytes of @cmnd 1207 * @timeout: timeout for this request 1208 * @sense_bytes: size of sense data to copy or 0 1209 * 1210 * This function is used to send a scsi command down to a target device 1211 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 1212 * 1213 * Return value: 1214 * SUCCESS or FAILED or NEEDS_RETRY 1215 */ 1216 static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd, 1217 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes) 1218 { 1219 struct scsi_device *sdev = scmd->device; 1220 struct Scsi_Host *shost = sdev->host; 1221 DECLARE_COMPLETION_ONSTACK(done); 1222 unsigned long timeleft = timeout, delay; 1223 struct scsi_eh_save ses; 1224 const unsigned long stall_for = msecs_to_jiffies(100); 1225 int rtn; 1226 1227 retry: 1228 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1229 shost->eh_action = &done; 1230 1231 scsi_log_send(scmd); 1232 scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER; 1233 scmd->flags |= SCMD_LAST; 1234 1235 /* 1236 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can 1237 * change the SCSI device state after we have examined it and before 1238 * .queuecommand() is called. 1239 */ 1240 mutex_lock(&sdev->state_mutex); 1241 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) { 1242 mutex_unlock(&sdev->state_mutex); 1243 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev, 1244 "%s: state %d <> %d\n", __func__, sdev->sdev_state, 1245 SDEV_BLOCK)); 1246 delay = min(timeleft, stall_for); 1247 timeleft -= delay; 1248 msleep(jiffies_to_msecs(delay)); 1249 mutex_lock(&sdev->state_mutex); 1250 } 1251 if (sdev->sdev_state != SDEV_BLOCK) 1252 rtn = shost->hostt->queuecommand(shost, scmd); 1253 else 1254 rtn = FAILED; 1255 mutex_unlock(&sdev->state_mutex); 1256 1257 if (rtn) { 1258 if (timeleft > stall_for) { 1259 scsi_eh_restore_cmnd(scmd, &ses); 1260 1261 timeleft -= stall_for; 1262 msleep(jiffies_to_msecs(stall_for)); 1263 goto retry; 1264 } 1265 /* signal not to enter either branch of the if () below */ 1266 timeleft = 0; 1267 rtn = FAILED; 1268 } else { 1269 timeleft = wait_for_completion_timeout(&done, timeout); 1270 rtn = SUCCESS; 1271 } 1272 1273 shost->eh_action = NULL; 1274 1275 scsi_log_completion(scmd, rtn); 1276 1277 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1278 "%s timeleft: %ld\n", 1279 __func__, timeleft)); 1280 1281 /* 1282 * If there is time left scsi_eh_done got called, and we will examine 1283 * the actual status codes to see whether the command actually did 1284 * complete normally, else if we have a zero return and no time left, 1285 * the command must still be pending, so abort it and return FAILED. 1286 * If we never actually managed to issue the command, because 1287 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1288 * value above (so don't execute either branch of the if) 1289 */ 1290 if (timeleft) { 1291 rtn = scsi_eh_completed_normally(scmd); 1292 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1293 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1294 1295 switch (rtn) { 1296 case SUCCESS: 1297 case NEEDS_RETRY: 1298 case FAILED: 1299 break; 1300 case ADD_TO_MLQUEUE: 1301 rtn = NEEDS_RETRY; 1302 break; 1303 default: 1304 rtn = FAILED; 1305 break; 1306 } 1307 } else if (rtn != FAILED) { 1308 scsi_abort_eh_cmnd(scmd); 1309 rtn = FAILED; 1310 } 1311 1312 scsi_eh_restore_cmnd(scmd, &ses); 1313 1314 return rtn; 1315 } 1316 1317 /** 1318 * scsi_request_sense - Request sense data from a particular target. 1319 * @scmd: SCSI cmd for request sense. 1320 * 1321 * Notes: 1322 * Some hosts automatically obtain this information, others require 1323 * that we obtain it on our own. This function will *not* return until 1324 * the command either times out, or it completes. 1325 */ 1326 static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd) 1327 { 1328 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1329 } 1330 1331 static enum scsi_disposition 1332 scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn) 1333 { 1334 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) { 1335 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1336 if (sdrv->eh_action) 1337 rtn = sdrv->eh_action(scmd, rtn); 1338 } 1339 return rtn; 1340 } 1341 1342 /** 1343 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1344 * @scmd: Original SCSI cmd that eh has finished. 1345 * @done_q: Queue for processed commands. 1346 * 1347 * Notes: 1348 * We don't want to use the normal command completion while we are are 1349 * still handling errors - it may cause other commands to be queued, 1350 * and that would disturb what we are doing. Thus we really want to 1351 * keep a list of pending commands for final completion, and once we 1352 * are ready to leave error handling we handle completion for real. 1353 */ 1354 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1355 { 1356 list_move_tail(&scmd->eh_entry, done_q); 1357 } 1358 EXPORT_SYMBOL(scsi_eh_finish_cmd); 1359 1360 /** 1361 * scsi_eh_get_sense - Get device sense data. 1362 * @work_q: Queue of commands to process. 1363 * @done_q: Queue of processed commands. 1364 * 1365 * Description: 1366 * See if we need to request sense information. if so, then get it 1367 * now, so we have a better idea of what to do. 1368 * 1369 * Notes: 1370 * This has the unfortunate side effect that if a shost adapter does 1371 * not automatically request sense information, we end up shutting 1372 * it down before we request it. 1373 * 1374 * All drivers should request sense information internally these days, 1375 * so for now all I have to say is tough noogies if you end up in here. 1376 * 1377 * XXX: Long term this code should go away, but that needs an audit of 1378 * all LLDDs first. 1379 */ 1380 int scsi_eh_get_sense(struct list_head *work_q, 1381 struct list_head *done_q) 1382 { 1383 struct scsi_cmnd *scmd, *next; 1384 struct Scsi_Host *shost; 1385 enum scsi_disposition rtn; 1386 1387 /* 1388 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, 1389 * should not get sense. 1390 */ 1391 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1392 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || 1393 SCSI_SENSE_VALID(scmd)) 1394 continue; 1395 1396 shost = scmd->device->host; 1397 if (scsi_host_eh_past_deadline(shost)) { 1398 SCSI_LOG_ERROR_RECOVERY(3, 1399 scmd_printk(KERN_INFO, scmd, 1400 "%s: skip request sense, past eh deadline\n", 1401 current->comm)); 1402 break; 1403 } 1404 if (!scsi_status_is_check_condition(scmd->result)) 1405 /* 1406 * don't request sense if there's no check condition 1407 * status because the error we're processing isn't one 1408 * that has a sense code (and some devices get 1409 * confused by sense requests out of the blue) 1410 */ 1411 continue; 1412 1413 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1414 "%s: requesting sense\n", 1415 current->comm)); 1416 rtn = scsi_request_sense(scmd); 1417 if (rtn != SUCCESS) 1418 continue; 1419 1420 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1421 "sense requested, result %x\n", scmd->result)); 1422 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); 1423 1424 rtn = scsi_decide_disposition(scmd); 1425 1426 /* 1427 * if the result was normal, then just pass it along to the 1428 * upper level. 1429 */ 1430 if (rtn == SUCCESS) 1431 /* 1432 * We don't want this command reissued, just finished 1433 * with the sense data, so set retries to the max 1434 * allowed to ensure it won't get reissued. If the user 1435 * has requested infinite retries, we also want to 1436 * finish this command, so force completion by setting 1437 * retries and allowed to the same value. 1438 */ 1439 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) 1440 scmd->retries = scmd->allowed = 1; 1441 else 1442 scmd->retries = scmd->allowed; 1443 else if (rtn != NEEDS_RETRY) 1444 continue; 1445 1446 scsi_eh_finish_cmd(scmd, done_q); 1447 } 1448 1449 return list_empty(work_q); 1450 } 1451 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1452 1453 /** 1454 * scsi_eh_tur - Send TUR to device. 1455 * @scmd: &scsi_cmnd to send TUR 1456 * 1457 * Return value: 1458 * 0 - Device is ready. 1 - Device NOT ready. 1459 */ 1460 static int scsi_eh_tur(struct scsi_cmnd *scmd) 1461 { 1462 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1463 int retry_cnt = 1; 1464 enum scsi_disposition rtn; 1465 1466 retry_tur: 1467 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1468 scmd->device->eh_timeout, 0); 1469 1470 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1471 "%s return: %x\n", __func__, rtn)); 1472 1473 switch (rtn) { 1474 case NEEDS_RETRY: 1475 if (retry_cnt--) 1476 goto retry_tur; 1477 fallthrough; 1478 case SUCCESS: 1479 return 0; 1480 default: 1481 return 1; 1482 } 1483 } 1484 1485 /** 1486 * scsi_eh_test_devices - check if devices are responding from error recovery. 1487 * @cmd_list: scsi commands in error recovery. 1488 * @work_q: queue for commands which still need more error recovery 1489 * @done_q: queue for commands which are finished 1490 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1491 * 1492 * Decription: 1493 * Tests if devices are in a working state. Commands to devices now in 1494 * a working state are sent to the done_q while commands to devices which 1495 * are still failing to respond are returned to the work_q for more 1496 * processing. 1497 **/ 1498 static int scsi_eh_test_devices(struct list_head *cmd_list, 1499 struct list_head *work_q, 1500 struct list_head *done_q, int try_stu) 1501 { 1502 struct scsi_cmnd *scmd, *next; 1503 struct scsi_device *sdev; 1504 int finish_cmds; 1505 1506 while (!list_empty(cmd_list)) { 1507 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1508 sdev = scmd->device; 1509 1510 if (!try_stu) { 1511 if (scsi_host_eh_past_deadline(sdev->host)) { 1512 /* Push items back onto work_q */ 1513 list_splice_init(cmd_list, work_q); 1514 SCSI_LOG_ERROR_RECOVERY(3, 1515 sdev_printk(KERN_INFO, sdev, 1516 "%s: skip test device, past eh deadline", 1517 current->comm)); 1518 break; 1519 } 1520 } 1521 1522 finish_cmds = !scsi_device_online(scmd->device) || 1523 (try_stu && !scsi_eh_try_stu(scmd) && 1524 !scsi_eh_tur(scmd)) || 1525 !scsi_eh_tur(scmd); 1526 1527 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1528 if (scmd->device == sdev) { 1529 if (finish_cmds && 1530 (try_stu || 1531 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1532 scsi_eh_finish_cmd(scmd, done_q); 1533 else 1534 list_move_tail(&scmd->eh_entry, work_q); 1535 } 1536 } 1537 return list_empty(work_q); 1538 } 1539 1540 /** 1541 * scsi_eh_try_stu - Send START_UNIT to device. 1542 * @scmd: &scsi_cmnd to send START_UNIT 1543 * 1544 * Return value: 1545 * 0 - Device is ready. 1 - Device NOT ready. 1546 */ 1547 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1548 { 1549 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1550 1551 if (scmd->device->allow_restart) { 1552 int i; 1553 enum scsi_disposition rtn = NEEDS_RETRY; 1554 1555 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1556 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, 1557 scmd->device->eh_timeout, 0); 1558 1559 if (rtn == SUCCESS) 1560 return 0; 1561 } 1562 1563 return 1; 1564 } 1565 1566 /** 1567 * scsi_eh_stu - send START_UNIT if needed 1568 * @shost: &scsi host being recovered. 1569 * @work_q: &list_head for pending commands. 1570 * @done_q: &list_head for processed commands. 1571 * 1572 * Notes: 1573 * If commands are failing due to not ready, initializing command required, 1574 * try revalidating the device, which will end up sending a start unit. 1575 */ 1576 static int scsi_eh_stu(struct Scsi_Host *shost, 1577 struct list_head *work_q, 1578 struct list_head *done_q) 1579 { 1580 struct scsi_cmnd *scmd, *stu_scmd, *next; 1581 struct scsi_device *sdev; 1582 1583 shost_for_each_device(sdev, shost) { 1584 if (scsi_host_eh_past_deadline(shost)) { 1585 SCSI_LOG_ERROR_RECOVERY(3, 1586 sdev_printk(KERN_INFO, sdev, 1587 "%s: skip START_UNIT, past eh deadline\n", 1588 current->comm)); 1589 scsi_device_put(sdev); 1590 break; 1591 } 1592 stu_scmd = NULL; 1593 list_for_each_entry(scmd, work_q, eh_entry) 1594 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1595 scsi_check_sense(scmd) == FAILED ) { 1596 stu_scmd = scmd; 1597 break; 1598 } 1599 1600 if (!stu_scmd) 1601 continue; 1602 1603 SCSI_LOG_ERROR_RECOVERY(3, 1604 sdev_printk(KERN_INFO, sdev, 1605 "%s: Sending START_UNIT\n", 1606 current->comm)); 1607 1608 if (!scsi_eh_try_stu(stu_scmd)) { 1609 if (!scsi_device_online(sdev) || 1610 !scsi_eh_tur(stu_scmd)) { 1611 list_for_each_entry_safe(scmd, next, 1612 work_q, eh_entry) { 1613 if (scmd->device == sdev && 1614 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1615 scsi_eh_finish_cmd(scmd, done_q); 1616 } 1617 } 1618 } else { 1619 SCSI_LOG_ERROR_RECOVERY(3, 1620 sdev_printk(KERN_INFO, sdev, 1621 "%s: START_UNIT failed\n", 1622 current->comm)); 1623 } 1624 } 1625 1626 return list_empty(work_q); 1627 } 1628 1629 1630 /** 1631 * scsi_eh_bus_device_reset - send bdr if needed 1632 * @shost: scsi host being recovered. 1633 * @work_q: &list_head for pending commands. 1634 * @done_q: &list_head for processed commands. 1635 * 1636 * Notes: 1637 * Try a bus device reset. Still, look to see whether we have multiple 1638 * devices that are jammed or not - if we have multiple devices, it 1639 * makes no sense to try bus_device_reset - we really would need to try 1640 * a bus_reset instead. 1641 */ 1642 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1643 struct list_head *work_q, 1644 struct list_head *done_q) 1645 { 1646 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1647 struct scsi_device *sdev; 1648 enum scsi_disposition rtn; 1649 1650 shost_for_each_device(sdev, shost) { 1651 if (scsi_host_eh_past_deadline(shost)) { 1652 SCSI_LOG_ERROR_RECOVERY(3, 1653 sdev_printk(KERN_INFO, sdev, 1654 "%s: skip BDR, past eh deadline\n", 1655 current->comm)); 1656 scsi_device_put(sdev); 1657 break; 1658 } 1659 bdr_scmd = NULL; 1660 list_for_each_entry(scmd, work_q, eh_entry) 1661 if (scmd->device == sdev) { 1662 bdr_scmd = scmd; 1663 break; 1664 } 1665 1666 if (!bdr_scmd) 1667 continue; 1668 1669 SCSI_LOG_ERROR_RECOVERY(3, 1670 sdev_printk(KERN_INFO, sdev, 1671 "%s: Sending BDR\n", current->comm)); 1672 rtn = scsi_try_bus_device_reset(bdr_scmd); 1673 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1674 if (!scsi_device_online(sdev) || 1675 rtn == FAST_IO_FAIL || 1676 !scsi_eh_tur(bdr_scmd)) { 1677 list_for_each_entry_safe(scmd, next, 1678 work_q, eh_entry) { 1679 if (scmd->device == sdev && 1680 scsi_eh_action(scmd, rtn) != FAILED) 1681 scsi_eh_finish_cmd(scmd, 1682 done_q); 1683 } 1684 } 1685 } else { 1686 SCSI_LOG_ERROR_RECOVERY(3, 1687 sdev_printk(KERN_INFO, sdev, 1688 "%s: BDR failed\n", current->comm)); 1689 } 1690 } 1691 1692 return list_empty(work_q); 1693 } 1694 1695 /** 1696 * scsi_eh_target_reset - send target reset if needed 1697 * @shost: scsi host being recovered. 1698 * @work_q: &list_head for pending commands. 1699 * @done_q: &list_head for processed commands. 1700 * 1701 * Notes: 1702 * Try a target reset. 1703 */ 1704 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1705 struct list_head *work_q, 1706 struct list_head *done_q) 1707 { 1708 LIST_HEAD(tmp_list); 1709 LIST_HEAD(check_list); 1710 1711 list_splice_init(work_q, &tmp_list); 1712 1713 while (!list_empty(&tmp_list)) { 1714 struct scsi_cmnd *next, *scmd; 1715 enum scsi_disposition rtn; 1716 unsigned int id; 1717 1718 if (scsi_host_eh_past_deadline(shost)) { 1719 /* push back on work queue for further processing */ 1720 list_splice_init(&check_list, work_q); 1721 list_splice_init(&tmp_list, work_q); 1722 SCSI_LOG_ERROR_RECOVERY(3, 1723 shost_printk(KERN_INFO, shost, 1724 "%s: Skip target reset, past eh deadline\n", 1725 current->comm)); 1726 return list_empty(work_q); 1727 } 1728 1729 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1730 id = scmd_id(scmd); 1731 1732 SCSI_LOG_ERROR_RECOVERY(3, 1733 shost_printk(KERN_INFO, shost, 1734 "%s: Sending target reset to target %d\n", 1735 current->comm, id)); 1736 rtn = scsi_try_target_reset(scmd); 1737 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1738 SCSI_LOG_ERROR_RECOVERY(3, 1739 shost_printk(KERN_INFO, shost, 1740 "%s: Target reset failed" 1741 " target: %d\n", 1742 current->comm, id)); 1743 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1744 if (scmd_id(scmd) != id) 1745 continue; 1746 1747 if (rtn == SUCCESS) 1748 list_move_tail(&scmd->eh_entry, &check_list); 1749 else if (rtn == FAST_IO_FAIL) 1750 scsi_eh_finish_cmd(scmd, done_q); 1751 else 1752 /* push back on work queue for further processing */ 1753 list_move(&scmd->eh_entry, work_q); 1754 } 1755 } 1756 1757 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1758 } 1759 1760 /** 1761 * scsi_eh_bus_reset - send a bus reset 1762 * @shost: &scsi host being recovered. 1763 * @work_q: &list_head for pending commands. 1764 * @done_q: &list_head for processed commands. 1765 */ 1766 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1767 struct list_head *work_q, 1768 struct list_head *done_q) 1769 { 1770 struct scsi_cmnd *scmd, *chan_scmd, *next; 1771 LIST_HEAD(check_list); 1772 unsigned int channel; 1773 enum scsi_disposition rtn; 1774 1775 /* 1776 * we really want to loop over the various channels, and do this on 1777 * a channel by channel basis. we should also check to see if any 1778 * of the failed commands are on soft_reset devices, and if so, skip 1779 * the reset. 1780 */ 1781 1782 for (channel = 0; channel <= shost->max_channel; channel++) { 1783 if (scsi_host_eh_past_deadline(shost)) { 1784 list_splice_init(&check_list, work_q); 1785 SCSI_LOG_ERROR_RECOVERY(3, 1786 shost_printk(KERN_INFO, shost, 1787 "%s: skip BRST, past eh deadline\n", 1788 current->comm)); 1789 return list_empty(work_q); 1790 } 1791 1792 chan_scmd = NULL; 1793 list_for_each_entry(scmd, work_q, eh_entry) { 1794 if (channel == scmd_channel(scmd)) { 1795 chan_scmd = scmd; 1796 break; 1797 /* 1798 * FIXME add back in some support for 1799 * soft_reset devices. 1800 */ 1801 } 1802 } 1803 1804 if (!chan_scmd) 1805 continue; 1806 SCSI_LOG_ERROR_RECOVERY(3, 1807 shost_printk(KERN_INFO, shost, 1808 "%s: Sending BRST chan: %d\n", 1809 current->comm, channel)); 1810 rtn = scsi_try_bus_reset(chan_scmd); 1811 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1812 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1813 if (channel == scmd_channel(scmd)) { 1814 if (rtn == FAST_IO_FAIL) 1815 scsi_eh_finish_cmd(scmd, 1816 done_q); 1817 else 1818 list_move_tail(&scmd->eh_entry, 1819 &check_list); 1820 } 1821 } 1822 } else { 1823 SCSI_LOG_ERROR_RECOVERY(3, 1824 shost_printk(KERN_INFO, shost, 1825 "%s: BRST failed chan: %d\n", 1826 current->comm, channel)); 1827 } 1828 } 1829 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1830 } 1831 1832 /** 1833 * scsi_eh_host_reset - send a host reset 1834 * @shost: host to be reset. 1835 * @work_q: &list_head for pending commands. 1836 * @done_q: &list_head for processed commands. 1837 */ 1838 static int scsi_eh_host_reset(struct Scsi_Host *shost, 1839 struct list_head *work_q, 1840 struct list_head *done_q) 1841 { 1842 struct scsi_cmnd *scmd, *next; 1843 LIST_HEAD(check_list); 1844 enum scsi_disposition rtn; 1845 1846 if (!list_empty(work_q)) { 1847 scmd = list_entry(work_q->next, 1848 struct scsi_cmnd, eh_entry); 1849 1850 SCSI_LOG_ERROR_RECOVERY(3, 1851 shost_printk(KERN_INFO, shost, 1852 "%s: Sending HRST\n", 1853 current->comm)); 1854 1855 rtn = scsi_try_host_reset(scmd); 1856 if (rtn == SUCCESS) { 1857 list_splice_init(work_q, &check_list); 1858 } else if (rtn == FAST_IO_FAIL) { 1859 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1860 scsi_eh_finish_cmd(scmd, done_q); 1861 } 1862 } else { 1863 SCSI_LOG_ERROR_RECOVERY(3, 1864 shost_printk(KERN_INFO, shost, 1865 "%s: HRST failed\n", 1866 current->comm)); 1867 } 1868 } 1869 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1870 } 1871 1872 /** 1873 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1874 * @work_q: &list_head for pending commands. 1875 * @done_q: &list_head for processed commands. 1876 */ 1877 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1878 struct list_head *done_q) 1879 { 1880 struct scsi_cmnd *scmd, *next; 1881 struct scsi_device *sdev; 1882 1883 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1884 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1885 "not ready after error recovery\n"); 1886 sdev = scmd->device; 1887 1888 mutex_lock(&sdev->state_mutex); 1889 scsi_device_set_state(sdev, SDEV_OFFLINE); 1890 mutex_unlock(&sdev->state_mutex); 1891 1892 scsi_eh_finish_cmd(scmd, done_q); 1893 } 1894 return; 1895 } 1896 1897 /** 1898 * scsi_noretry_cmd - determine if command should be failed fast 1899 * @scmd: SCSI cmd to examine. 1900 */ 1901 bool scsi_noretry_cmd(struct scsi_cmnd *scmd) 1902 { 1903 struct request *req = scsi_cmd_to_rq(scmd); 1904 1905 switch (host_byte(scmd->result)) { 1906 case DID_OK: 1907 break; 1908 case DID_TIME_OUT: 1909 goto check_type; 1910 case DID_BUS_BUSY: 1911 return !!(req->cmd_flags & REQ_FAILFAST_TRANSPORT); 1912 case DID_PARITY: 1913 return !!(req->cmd_flags & REQ_FAILFAST_DEV); 1914 case DID_ERROR: 1915 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT) 1916 return false; 1917 fallthrough; 1918 case DID_SOFT_ERROR: 1919 return !!(req->cmd_flags & REQ_FAILFAST_DRIVER); 1920 } 1921 1922 /* Never retry commands aborted due to a duration limit timeout */ 1923 if (scsi_ml_byte(scmd->result) == SCSIML_STAT_DL_TIMEOUT) 1924 return true; 1925 1926 if (!scsi_status_is_check_condition(scmd->result)) 1927 return false; 1928 1929 check_type: 1930 /* 1931 * assume caller has checked sense and determined 1932 * the check condition was retryable. 1933 */ 1934 if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(req)) 1935 return true; 1936 1937 return false; 1938 } 1939 1940 /** 1941 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1942 * @scmd: SCSI cmd to examine. 1943 * 1944 * Notes: 1945 * This is *only* called when we are examining the status after sending 1946 * out the actual data command. any commands that are queued for error 1947 * recovery (e.g. test_unit_ready) do *not* come through here. 1948 * 1949 * When this routine returns failed, it means the error handler thread 1950 * is woken. In cases where the error code indicates an error that 1951 * doesn't require the error handler read (i.e. we don't need to 1952 * abort/reset), this function should return SUCCESS. 1953 */ 1954 enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd) 1955 { 1956 enum scsi_disposition rtn; 1957 1958 /* 1959 * if the device is offline, then we clearly just pass the result back 1960 * up to the top level. 1961 */ 1962 if (!scsi_device_online(scmd->device)) { 1963 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1964 "%s: device offline - report as SUCCESS\n", __func__)); 1965 return SUCCESS; 1966 } 1967 1968 /* 1969 * first check the host byte, to see if there is anything in there 1970 * that would indicate what we need to do. 1971 */ 1972 switch (host_byte(scmd->result)) { 1973 case DID_PASSTHROUGH: 1974 /* 1975 * no matter what, pass this through to the upper layer. 1976 * nuke this special code so that it looks like we are saying 1977 * did_ok. 1978 */ 1979 scmd->result &= 0xff00ffff; 1980 return SUCCESS; 1981 case DID_OK: 1982 /* 1983 * looks good. drop through, and check the next byte. 1984 */ 1985 break; 1986 case DID_ABORT: 1987 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1988 set_host_byte(scmd, DID_TIME_OUT); 1989 return SUCCESS; 1990 } 1991 fallthrough; 1992 case DID_NO_CONNECT: 1993 case DID_BAD_TARGET: 1994 /* 1995 * note - this means that we just report the status back 1996 * to the top level driver, not that we actually think 1997 * that it indicates SUCCESS. 1998 */ 1999 return SUCCESS; 2000 case DID_SOFT_ERROR: 2001 /* 2002 * when the low level driver returns did_soft_error, 2003 * it is responsible for keeping an internal retry counter 2004 * in order to avoid endless loops (db) 2005 */ 2006 goto maybe_retry; 2007 case DID_IMM_RETRY: 2008 return NEEDS_RETRY; 2009 2010 case DID_REQUEUE: 2011 return ADD_TO_MLQUEUE; 2012 case DID_TRANSPORT_DISRUPTED: 2013 /* 2014 * LLD/transport was disrupted during processing of the IO. 2015 * The transport class is now blocked/blocking, 2016 * and the transport will decide what to do with the IO 2017 * based on its timers and recovery capablilities if 2018 * there are enough retries. 2019 */ 2020 goto maybe_retry; 2021 case DID_TRANSPORT_FAILFAST: 2022 /* 2023 * The transport decided to failfast the IO (most likely 2024 * the fast io fail tmo fired), so send IO directly upwards. 2025 */ 2026 return SUCCESS; 2027 case DID_TRANSPORT_MARGINAL: 2028 /* 2029 * caller has decided not to do retries on 2030 * abort success, so send IO directly upwards 2031 */ 2032 return SUCCESS; 2033 case DID_ERROR: 2034 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT) 2035 /* 2036 * execute reservation conflict processing code 2037 * lower down 2038 */ 2039 break; 2040 fallthrough; 2041 case DID_BUS_BUSY: 2042 case DID_PARITY: 2043 goto maybe_retry; 2044 case DID_TIME_OUT: 2045 /* 2046 * when we scan the bus, we get timeout messages for 2047 * these commands if there is no device available. 2048 * other hosts report did_no_connect for the same thing. 2049 */ 2050 if ((scmd->cmnd[0] == TEST_UNIT_READY || 2051 scmd->cmnd[0] == INQUIRY)) { 2052 return SUCCESS; 2053 } else { 2054 return FAILED; 2055 } 2056 case DID_RESET: 2057 return SUCCESS; 2058 default: 2059 return FAILED; 2060 } 2061 2062 /* 2063 * check the status byte to see if this indicates anything special. 2064 */ 2065 switch (get_status_byte(scmd)) { 2066 case SAM_STAT_TASK_SET_FULL: 2067 scsi_handle_queue_full(scmd->device); 2068 /* 2069 * the case of trying to send too many commands to a 2070 * tagged queueing device. 2071 */ 2072 fallthrough; 2073 case SAM_STAT_BUSY: 2074 /* 2075 * device can't talk to us at the moment. Should only 2076 * occur (SAM-3) when the task queue is empty, so will cause 2077 * the empty queue handling to trigger a stall in the 2078 * device. 2079 */ 2080 return ADD_TO_MLQUEUE; 2081 case SAM_STAT_GOOD: 2082 if (scmd->cmnd[0] == REPORT_LUNS) 2083 scmd->device->sdev_target->expecting_lun_change = 0; 2084 scsi_handle_queue_ramp_up(scmd->device); 2085 if (scmd->sense_buffer && SCSI_SENSE_VALID(scmd)) 2086 /* 2087 * If we have sense data, call scsi_check_sense() in 2088 * order to set the correct SCSI ML byte (if any). 2089 * No point in checking the return value, since the 2090 * command has already completed successfully. 2091 */ 2092 scsi_check_sense(scmd); 2093 fallthrough; 2094 case SAM_STAT_COMMAND_TERMINATED: 2095 return SUCCESS; 2096 case SAM_STAT_TASK_ABORTED: 2097 goto maybe_retry; 2098 case SAM_STAT_CHECK_CONDITION: 2099 rtn = scsi_check_sense(scmd); 2100 if (rtn == NEEDS_RETRY) 2101 goto maybe_retry; 2102 /* if rtn == FAILED, we have no sense information; 2103 * returning FAILED will wake the error handler thread 2104 * to collect the sense and redo the decide 2105 * disposition */ 2106 return rtn; 2107 case SAM_STAT_CONDITION_MET: 2108 case SAM_STAT_INTERMEDIATE: 2109 case SAM_STAT_INTERMEDIATE_CONDITION_MET: 2110 case SAM_STAT_ACA_ACTIVE: 2111 /* 2112 * who knows? FIXME(eric) 2113 */ 2114 return SUCCESS; 2115 2116 case SAM_STAT_RESERVATION_CONFLICT: 2117 sdev_printk(KERN_INFO, scmd->device, 2118 "reservation conflict\n"); 2119 set_scsi_ml_byte(scmd, SCSIML_STAT_RESV_CONFLICT); 2120 return SUCCESS; /* causes immediate i/o error */ 2121 } 2122 return FAILED; 2123 2124 maybe_retry: 2125 2126 /* we requeue for retry because the error was retryable, and 2127 * the request was not marked fast fail. Note that above, 2128 * even if the request is marked fast fail, we still requeue 2129 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 2130 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) { 2131 return NEEDS_RETRY; 2132 } else { 2133 /* 2134 * no more retries - report this one back to upper level. 2135 */ 2136 return SUCCESS; 2137 } 2138 } 2139 2140 static enum rq_end_io_ret eh_lock_door_done(struct request *req, 2141 blk_status_t status, 2142 const struct io_comp_batch *iob) 2143 { 2144 blk_mq_free_request(req); 2145 return RQ_END_IO_NONE; 2146 } 2147 2148 /** 2149 * scsi_eh_lock_door - Prevent medium removal for the specified device 2150 * @sdev: SCSI device to prevent medium removal 2151 * 2152 * Locking: 2153 * We must be called from process context. 2154 * 2155 * Notes: 2156 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 2157 * head of the devices request queue, and continue. 2158 */ 2159 static void scsi_eh_lock_door(struct scsi_device *sdev) 2160 { 2161 struct scsi_cmnd *scmd; 2162 struct request *req; 2163 2164 req = scsi_alloc_request(sdev->request_queue, REQ_OP_DRV_IN, BLK_MQ_REQ_NOWAIT); 2165 if (IS_ERR(req)) 2166 return; 2167 scmd = blk_mq_rq_to_pdu(req); 2168 2169 scmd->cmnd[0] = ALLOW_MEDIUM_REMOVAL; 2170 scmd->cmnd[1] = 0; 2171 scmd->cmnd[2] = 0; 2172 scmd->cmnd[3] = 0; 2173 scmd->cmnd[4] = SCSI_REMOVAL_PREVENT; 2174 scmd->cmnd[5] = 0; 2175 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 2176 scmd->allowed = 5; 2177 2178 req->rq_flags |= RQF_QUIET; 2179 req->timeout = 10 * HZ; 2180 req->end_io = eh_lock_door_done; 2181 2182 blk_execute_rq_nowait(req, true); 2183 } 2184 2185 /** 2186 * scsi_restart_operations - restart io operations to the specified host. 2187 * @shost: Host we are restarting. 2188 * 2189 * Notes: 2190 * When we entered the error handler, we blocked all further i/o to 2191 * this device. we need to 'reverse' this process. 2192 */ 2193 static void scsi_restart_operations(struct Scsi_Host *shost) 2194 { 2195 struct scsi_device *sdev; 2196 unsigned long flags; 2197 2198 /* 2199 * If the door was locked, we need to insert a door lock request 2200 * onto the head of the SCSI request queue for the device. There 2201 * is no point trying to lock the door of an off-line device. 2202 */ 2203 shost_for_each_device(sdev, shost) { 2204 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 2205 scsi_eh_lock_door(sdev); 2206 sdev->was_reset = 0; 2207 } 2208 } 2209 2210 /* 2211 * next free up anything directly waiting upon the host. this 2212 * will be requests for character device operations, and also for 2213 * ioctls to queued block devices. 2214 */ 2215 SCSI_LOG_ERROR_RECOVERY(3, 2216 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 2217 2218 spin_lock_irqsave(shost->host_lock, flags); 2219 if (scsi_host_set_state(shost, SHOST_RUNNING)) 2220 if (scsi_host_set_state(shost, SHOST_CANCEL)) 2221 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 2222 spin_unlock_irqrestore(shost->host_lock, flags); 2223 2224 wake_up(&shost->host_wait); 2225 2226 /* 2227 * finally we need to re-initiate requests that may be pending. we will 2228 * have had everything blocked while error handling is taking place, and 2229 * now that error recovery is done, we will need to ensure that these 2230 * requests are started. 2231 */ 2232 scsi_run_host_queues(shost); 2233 2234 /* 2235 * if eh is active and host_eh_scheduled is pending we need to re-run 2236 * recovery. we do this check after scsi_run_host_queues() to allow 2237 * everything pent up since the last eh run a chance to make forward 2238 * progress before we sync again. Either we'll immediately re-run 2239 * recovery or scsi_device_unbusy() will wake us again when these 2240 * pending commands complete. 2241 */ 2242 spin_lock_irqsave(shost->host_lock, flags); 2243 if (shost->host_eh_scheduled) 2244 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 2245 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 2246 spin_unlock_irqrestore(shost->host_lock, flags); 2247 } 2248 2249 /** 2250 * scsi_eh_ready_devs - check device ready state and recover if not. 2251 * @shost: host to be recovered. 2252 * @work_q: &list_head for pending commands. 2253 * @done_q: &list_head for processed commands. 2254 */ 2255 void scsi_eh_ready_devs(struct Scsi_Host *shost, 2256 struct list_head *work_q, 2257 struct list_head *done_q) 2258 { 2259 if (!scsi_eh_stu(shost, work_q, done_q)) 2260 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2261 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2262 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2263 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2264 scsi_eh_offline_sdevs(work_q, 2265 done_q); 2266 } 2267 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2268 2269 /** 2270 * scsi_eh_flush_done_q - finish processed commands or retry them. 2271 * @done_q: list_head of processed commands. 2272 */ 2273 void scsi_eh_flush_done_q(struct list_head *done_q) 2274 { 2275 struct scsi_cmnd *scmd, *next; 2276 2277 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2278 struct scsi_device *sdev = scmd->device; 2279 2280 list_del_init(&scmd->eh_entry); 2281 if (scsi_device_online(sdev) && !scsi_noretry_cmd(scmd) && 2282 scsi_cmd_retry_allowed(scmd) && 2283 scsi_eh_should_retry_cmd(scmd)) { 2284 SCSI_LOG_ERROR_RECOVERY(3, 2285 scmd_printk(KERN_INFO, scmd, 2286 "%s: flush retry cmd\n", 2287 current->comm)); 2288 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2289 blk_mq_kick_requeue_list(sdev->request_queue); 2290 } else { 2291 /* 2292 * If just we got sense for the device (called 2293 * scsi_eh_get_sense), scmd->result is already 2294 * set, do not set DID_TIME_OUT. 2295 */ 2296 if (!scmd->result && 2297 !(scmd->flags & SCMD_FORCE_EH_SUCCESS)) 2298 scmd->result |= (DID_TIME_OUT << 16); 2299 SCSI_LOG_ERROR_RECOVERY(3, 2300 scmd_printk(KERN_INFO, scmd, 2301 "%s: flush finish cmd\n", 2302 current->comm)); 2303 scsi_finish_command(scmd); 2304 } 2305 } 2306 } 2307 EXPORT_SYMBOL(scsi_eh_flush_done_q); 2308 2309 /** 2310 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2311 * @shost: Host to unjam. 2312 * 2313 * Notes: 2314 * When we come in here, we *know* that all commands on the bus have 2315 * either completed, failed or timed out. we also know that no further 2316 * commands are being sent to the host, so things are relatively quiet 2317 * and we have freedom to fiddle with things as we wish. 2318 * 2319 * This is only the *default* implementation. it is possible for 2320 * individual drivers to supply their own version of this function, and 2321 * if the maintainer wishes to do this, it is strongly suggested that 2322 * this function be taken as a template and modified. this function 2323 * was designed to correctly handle problems for about 95% of the 2324 * different cases out there, and it should always provide at least a 2325 * reasonable amount of error recovery. 2326 * 2327 * Any command marked 'failed' or 'timeout' must eventually have 2328 * scsi_finish_cmd() called for it. we do all of the retry stuff 2329 * here, so when we restart the host after we return it should have an 2330 * empty queue. 2331 */ 2332 static void scsi_unjam_host(struct Scsi_Host *shost) 2333 { 2334 unsigned long flags; 2335 LIST_HEAD(eh_work_q); 2336 LIST_HEAD(eh_done_q); 2337 2338 spin_lock_irqsave(shost->host_lock, flags); 2339 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2340 spin_unlock_irqrestore(shost->host_lock, flags); 2341 2342 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2343 2344 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2345 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2346 2347 spin_lock_irqsave(shost->host_lock, flags); 2348 if (shost->eh_deadline != -1) 2349 shost->last_reset = 0; 2350 spin_unlock_irqrestore(shost->host_lock, flags); 2351 scsi_eh_flush_done_q(&eh_done_q); 2352 } 2353 2354 /** 2355 * scsi_error_handler - SCSI error handler thread 2356 * @data: Host for which we are running. 2357 * 2358 * Notes: 2359 * This is the main error handling loop. This is run as a kernel thread 2360 * for every SCSI host and handles all error handling activity. 2361 */ 2362 int scsi_error_handler(void *data) 2363 { 2364 struct Scsi_Host *shost = data; 2365 bool eh_noresume; 2366 2367 /* 2368 * We use TASK_INTERRUPTIBLE so that the thread is not 2369 * counted against the load average as a running process. 2370 * We never actually get interrupted because kthread_run 2371 * disables signal delivery for the created thread. 2372 */ 2373 while (true) { 2374 /* 2375 * The sequence in kthread_stop() sets the stop flag first 2376 * then wakes the process. To avoid missed wakeups, the task 2377 * should always be in a non running state before the stop 2378 * flag is checked 2379 */ 2380 set_current_state(TASK_INTERRUPTIBLE); 2381 if (kthread_should_stop()) 2382 break; 2383 2384 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2385 shost->host_failed != scsi_host_busy(shost)) { 2386 SCSI_LOG_ERROR_RECOVERY(1, 2387 shost_printk(KERN_INFO, shost, 2388 "scsi_eh_%d: sleeping\n", 2389 shost->host_no)); 2390 schedule(); 2391 continue; 2392 } 2393 2394 __set_current_state(TASK_RUNNING); 2395 SCSI_LOG_ERROR_RECOVERY(1, 2396 shost_printk(KERN_INFO, shost, 2397 "scsi_eh_%d: waking up %d/%d/%d\n", 2398 shost->host_no, shost->host_eh_scheduled, 2399 shost->host_failed, 2400 scsi_host_busy(shost))); 2401 2402 /* 2403 * We have a host that is failing for some reason. Figure out 2404 * what we need to do to get it up and online again (if we can). 2405 * If we fail, we end up taking the thing offline. 2406 */ 2407 eh_noresume = READ_ONCE(shost->eh_noresume); 2408 if (!eh_noresume && scsi_autopm_get_host(shost) != 0) { 2409 SCSI_LOG_ERROR_RECOVERY(1, 2410 shost_printk(KERN_ERR, shost, 2411 "scsi_eh_%d: unable to autoresume\n", 2412 shost->host_no)); 2413 continue; 2414 } 2415 2416 if (shost->transportt->eh_strategy_handler) 2417 shost->transportt->eh_strategy_handler(shost); 2418 else 2419 scsi_unjam_host(shost); 2420 2421 /* All scmds have been handled */ 2422 shost->host_failed = 0; 2423 2424 /* 2425 * Note - if the above fails completely, the action is to take 2426 * individual devices offline and flush the queue of any 2427 * outstanding requests that may have been pending. When we 2428 * restart, we restart any I/O to any other devices on the bus 2429 * which are still online. 2430 */ 2431 scsi_restart_operations(shost); 2432 if (!eh_noresume) 2433 scsi_autopm_put_host(shost); 2434 } 2435 __set_current_state(TASK_RUNNING); 2436 2437 SCSI_LOG_ERROR_RECOVERY(1, 2438 shost_printk(KERN_INFO, shost, 2439 "Error handler scsi_eh_%d exiting\n", 2440 shost->host_no)); 2441 shost->ehandler = NULL; 2442 return 0; 2443 } 2444 2445 /** 2446 * scsi_report_bus_reset() - report bus reset observed 2447 * 2448 * Utility function used by low-level drivers to report that 2449 * they have observed a bus reset on the bus being handled. 2450 * 2451 * @shost: Host in question 2452 * @channel: channel on which reset was observed. 2453 * 2454 * Returns: Nothing 2455 * 2456 * Lock status: Host lock must be held. 2457 * 2458 * Notes: This only needs to be called if the reset is one which 2459 * originates from an unknown location. Resets originated 2460 * by the mid-level itself don't need to call this, but there 2461 * should be no harm. 2462 * 2463 * The main purpose of this is to make sure that a CHECK_CONDITION 2464 * is properly treated. 2465 */ 2466 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2467 { 2468 struct scsi_device *sdev; 2469 2470 __shost_for_each_device(sdev, shost) { 2471 if (channel == sdev_channel(sdev)) 2472 __scsi_report_device_reset(sdev, NULL); 2473 } 2474 } 2475 EXPORT_SYMBOL(scsi_report_bus_reset); 2476 2477 /** 2478 * scsi_report_device_reset() - report device reset observed 2479 * 2480 * Utility function used by low-level drivers to report that 2481 * they have observed a device reset on the device being handled. 2482 * 2483 * @shost: Host in question 2484 * @channel: channel on which reset was observed 2485 * @target: target on which reset was observed 2486 * 2487 * Returns: Nothing 2488 * 2489 * Lock status: Host lock must be held 2490 * 2491 * Notes: This only needs to be called if the reset is one which 2492 * originates from an unknown location. Resets originated 2493 * by the mid-level itself don't need to call this, but there 2494 * should be no harm. 2495 * 2496 * The main purpose of this is to make sure that a CHECK_CONDITION 2497 * is properly treated. 2498 */ 2499 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2500 { 2501 struct scsi_device *sdev; 2502 2503 __shost_for_each_device(sdev, shost) { 2504 if (channel == sdev_channel(sdev) && 2505 target == sdev_id(sdev)) 2506 __scsi_report_device_reset(sdev, NULL); 2507 } 2508 } 2509 EXPORT_SYMBOL(scsi_report_device_reset); 2510 2511 /** 2512 * scsi_ioctl_reset: explicitly reset a host/bus/target/device 2513 * @dev: scsi_device to operate on 2514 * @arg: reset type (see sg.h) 2515 */ 2516 int 2517 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) 2518 { 2519 struct scsi_cmnd *scmd; 2520 struct Scsi_Host *shost = dev->host; 2521 struct request *rq; 2522 unsigned long flags; 2523 int error = 0, val; 2524 enum scsi_disposition rtn; 2525 2526 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) 2527 return -EACCES; 2528 2529 error = get_user(val, arg); 2530 if (error) 2531 return error; 2532 2533 if (scsi_autopm_get_host(shost) < 0) 2534 return -EIO; 2535 2536 error = -EIO; 2537 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) + 2538 shost->hostt->cmd_size, GFP_KERNEL); 2539 if (!rq) 2540 goto out_put_autopm_host; 2541 blk_rq_init(NULL, rq); 2542 2543 scmd = (struct scsi_cmnd *)(rq + 1); 2544 scsi_init_command(dev, scmd); 2545 2546 scmd->submitter = SUBMITTED_BY_SCSI_RESET_IOCTL; 2547 scmd->flags |= SCMD_LAST; 2548 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 2549 2550 scmd->cmd_len = 0; 2551 2552 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 2553 2554 spin_lock_irqsave(shost->host_lock, flags); 2555 shost->tmf_in_progress = 1; 2556 spin_unlock_irqrestore(shost->host_lock, flags); 2557 2558 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { 2559 case SG_SCSI_RESET_NOTHING: 2560 rtn = SUCCESS; 2561 break; 2562 case SG_SCSI_RESET_DEVICE: 2563 rtn = scsi_try_bus_device_reset(scmd); 2564 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2565 break; 2566 fallthrough; 2567 case SG_SCSI_RESET_TARGET: 2568 rtn = scsi_try_target_reset(scmd); 2569 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2570 break; 2571 fallthrough; 2572 case SG_SCSI_RESET_BUS: 2573 rtn = scsi_try_bus_reset(scmd); 2574 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2575 break; 2576 fallthrough; 2577 case SG_SCSI_RESET_HOST: 2578 rtn = scsi_try_host_reset(scmd); 2579 if (rtn == SUCCESS) 2580 break; 2581 fallthrough; 2582 default: 2583 rtn = FAILED; 2584 break; 2585 } 2586 2587 error = (rtn == SUCCESS) ? 0 : -EIO; 2588 2589 spin_lock_irqsave(shost->host_lock, flags); 2590 shost->tmf_in_progress = 0; 2591 spin_unlock_irqrestore(shost->host_lock, flags); 2592 2593 /* 2594 * be sure to wake up anyone who was sleeping or had their queue 2595 * suspended while we performed the TMF. 2596 */ 2597 SCSI_LOG_ERROR_RECOVERY(3, 2598 shost_printk(KERN_INFO, shost, 2599 "waking up host to restart after TMF\n")); 2600 2601 wake_up(&shost->host_wait); 2602 scsi_run_host_queues(shost); 2603 2604 kfree(rq); 2605 2606 out_put_autopm_host: 2607 scsi_autopm_put_host(shost); 2608 return error; 2609 } 2610 2611 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, 2612 struct scsi_sense_hdr *sshdr) 2613 { 2614 return scsi_normalize_sense(cmd->sense_buffer, 2615 SCSI_SENSE_BUFFERSIZE, sshdr); 2616 } 2617 EXPORT_SYMBOL(scsi_command_normalize_sense); 2618 2619 /** 2620 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2621 * @sense_buffer: byte array of sense data 2622 * @sb_len: number of valid bytes in sense_buffer 2623 * @info_out: pointer to 64 integer where 8 or 4 byte information 2624 * field will be placed if found. 2625 * 2626 * Return value: 2627 * true if information field found, false if not found. 2628 */ 2629 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len, 2630 u64 *info_out) 2631 { 2632 const u8 * ucp; 2633 2634 if (sb_len < 7) 2635 return false; 2636 switch (sense_buffer[0] & 0x7f) { 2637 case 0x70: 2638 case 0x71: 2639 if (sense_buffer[0] & 0x80) { 2640 *info_out = get_unaligned_be32(&sense_buffer[3]); 2641 return true; 2642 } 2643 return false; 2644 case 0x72: 2645 case 0x73: 2646 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2647 0 /* info desc */); 2648 if (ucp && (0xa == ucp[1])) { 2649 *info_out = get_unaligned_be64(&ucp[4]); 2650 return true; 2651 } 2652 return false; 2653 default: 2654 return false; 2655 } 2656 } 2657 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2658