1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * scsi_error.c Copyright (C) 1997 Eric Youngdale 4 * 5 * SCSI error/timeout handling 6 * Initial versions: Eric Youngdale. Based upon conversations with 7 * Leonard Zubkoff and David Miller at Linux Expo, 8 * ideas originating from all over the place. 9 * 10 * Restructured scsi_unjam_host and associated functions. 11 * September 04, 2002 Mike Anderson (andmike@us.ibm.com) 12 * 13 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and 14 * minor cleanups. 15 * September 30, 2002 Mike Anderson (andmike@us.ibm.com) 16 */ 17 18 #include <linux/module.h> 19 #include <linux/sched.h> 20 #include <linux/gfp.h> 21 #include <linux/timer.h> 22 #include <linux/string.h> 23 #include <linux/kernel.h> 24 #include <linux/freezer.h> 25 #include <linux/kthread.h> 26 #include <linux/interrupt.h> 27 #include <linux/blkdev.h> 28 #include <linux/delay.h> 29 #include <linux/jiffies.h> 30 31 #include <scsi/scsi.h> 32 #include <scsi/scsi_cmnd.h> 33 #include <scsi/scsi_dbg.h> 34 #include <scsi/scsi_device.h> 35 #include <scsi/scsi_driver.h> 36 #include <scsi/scsi_eh.h> 37 #include <scsi/scsi_common.h> 38 #include <scsi/scsi_transport.h> 39 #include <scsi/scsi_host.h> 40 #include <scsi/scsi_ioctl.h> 41 #include <scsi/scsi_dh.h> 42 #include <scsi/scsi_devinfo.h> 43 #include <scsi/sg.h> 44 45 #include "scsi_priv.h" 46 #include "scsi_logging.h" 47 #include "scsi_transport_api.h" 48 49 #include <trace/events/scsi.h> 50 51 #include <asm/unaligned.h> 52 53 /* 54 * These should *probably* be handled by the host itself. 55 * Since it is allowed to sleep, it probably should. 56 */ 57 #define BUS_RESET_SETTLE_TIME (10) 58 #define HOST_RESET_SETTLE_TIME (10) 59 60 static int scsi_eh_try_stu(struct scsi_cmnd *scmd); 61 static enum scsi_disposition scsi_try_to_abort_cmd(struct scsi_host_template *, 62 struct scsi_cmnd *); 63 64 void scsi_eh_wakeup(struct Scsi_Host *shost) 65 { 66 lockdep_assert_held(shost->host_lock); 67 68 if (scsi_host_busy(shost) == shost->host_failed) { 69 trace_scsi_eh_wakeup(shost); 70 wake_up_process(shost->ehandler); 71 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost, 72 "Waking error handler thread\n")); 73 } 74 } 75 76 /** 77 * scsi_schedule_eh - schedule EH for SCSI host 78 * @shost: SCSI host to invoke error handling on. 79 * 80 * Schedule SCSI EH without scmd. 81 */ 82 void scsi_schedule_eh(struct Scsi_Host *shost) 83 { 84 unsigned long flags; 85 86 spin_lock_irqsave(shost->host_lock, flags); 87 88 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || 89 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { 90 shost->host_eh_scheduled++; 91 scsi_eh_wakeup(shost); 92 } 93 94 spin_unlock_irqrestore(shost->host_lock, flags); 95 } 96 EXPORT_SYMBOL_GPL(scsi_schedule_eh); 97 98 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost) 99 { 100 if (!shost->last_reset || shost->eh_deadline == -1) 101 return 0; 102 103 /* 104 * 32bit accesses are guaranteed to be atomic 105 * (on all supported architectures), so instead 106 * of using a spinlock we can as well double check 107 * if eh_deadline has been set to 'off' during the 108 * time_before call. 109 */ 110 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) && 111 shost->eh_deadline > -1) 112 return 0; 113 114 return 1; 115 } 116 117 static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd) 118 { 119 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) 120 return true; 121 122 return ++cmd->retries <= cmd->allowed; 123 } 124 125 static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd) 126 { 127 struct scsi_device *sdev = cmd->device; 128 struct Scsi_Host *host = sdev->host; 129 130 if (host->hostt->eh_should_retry_cmd) 131 return host->hostt->eh_should_retry_cmd(cmd); 132 133 return true; 134 } 135 136 /** 137 * scmd_eh_abort_handler - Handle command aborts 138 * @work: command to be aborted. 139 * 140 * Note: this function must be called only for a command that has timed out. 141 * Because the block layer marks a request as complete before it calls 142 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has 143 * timed out do not have any effect. Hence it is safe to call 144 * scsi_finish_command() from this function. 145 */ 146 void 147 scmd_eh_abort_handler(struct work_struct *work) 148 { 149 struct scsi_cmnd *scmd = 150 container_of(work, struct scsi_cmnd, abort_work.work); 151 struct scsi_device *sdev = scmd->device; 152 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 (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 235 /* 236 * Retry after abort failed, escalate to next level. 237 */ 238 SCSI_LOG_ERROR_RECOVERY(3, 239 scmd_printk(KERN_INFO, scmd, 240 "previous abort failed\n")); 241 BUG_ON(delayed_work_pending(&scmd->abort_work)); 242 return FAILED; 243 } 244 245 spin_lock_irqsave(shost->host_lock, flags); 246 if (shost->eh_deadline != -1 && !shost->last_reset) 247 shost->last_reset = jiffies; 248 BUG_ON(!list_empty(&scmd->eh_entry)); 249 list_add_tail(&scmd->eh_entry, &shost->eh_abort_list); 250 spin_unlock_irqrestore(shost->host_lock, flags); 251 252 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; 253 SCSI_LOG_ERROR_RECOVERY(3, 254 scmd_printk(KERN_INFO, scmd, "abort scheduled\n")); 255 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100); 256 return SUCCESS; 257 } 258 259 /** 260 * scsi_eh_reset - call into ->eh_action to reset internal counters 261 * @scmd: scmd to run eh on. 262 * 263 * The scsi driver might be carrying internal state about the 264 * devices, so we need to call into the driver to reset the 265 * internal state once the error handler is started. 266 */ 267 static void scsi_eh_reset(struct scsi_cmnd *scmd) 268 { 269 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) { 270 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 271 if (sdrv->eh_reset) 272 sdrv->eh_reset(scmd); 273 } 274 } 275 276 static void scsi_eh_inc_host_failed(struct rcu_head *head) 277 { 278 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu); 279 struct Scsi_Host *shost = scmd->device->host; 280 unsigned long flags; 281 282 spin_lock_irqsave(shost->host_lock, flags); 283 shost->host_failed++; 284 scsi_eh_wakeup(shost); 285 spin_unlock_irqrestore(shost->host_lock, flags); 286 } 287 288 /** 289 * scsi_eh_scmd_add - add scsi cmd to error handling. 290 * @scmd: scmd to run eh on. 291 */ 292 void scsi_eh_scmd_add(struct scsi_cmnd *scmd) 293 { 294 struct Scsi_Host *shost = scmd->device->host; 295 unsigned long flags; 296 int ret; 297 298 WARN_ON_ONCE(!shost->ehandler); 299 300 spin_lock_irqsave(shost->host_lock, flags); 301 if (scsi_host_set_state(shost, SHOST_RECOVERY)) { 302 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY); 303 WARN_ON_ONCE(ret); 304 } 305 if (shost->eh_deadline != -1 && !shost->last_reset) 306 shost->last_reset = jiffies; 307 308 scsi_eh_reset(scmd); 309 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 310 spin_unlock_irqrestore(shost->host_lock, flags); 311 /* 312 * Ensure that all tasks observe the host state change before the 313 * host_failed change. 314 */ 315 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed); 316 } 317 318 /** 319 * scsi_times_out - Timeout function for normal scsi commands. 320 * @req: request that is timing out. 321 * 322 * Notes: 323 * We do not need to lock this. There is the potential for a race 324 * only in that the normal completion handling might run, but if the 325 * normal completion function determines that the timer has already 326 * fired, then it mustn't do anything. 327 */ 328 enum blk_eh_timer_return scsi_times_out(struct request *req) 329 { 330 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req); 331 enum blk_eh_timer_return rtn = BLK_EH_DONE; 332 struct Scsi_Host *host = scmd->device->host; 333 334 trace_scsi_dispatch_cmd_timeout(scmd); 335 scsi_log_completion(scmd, TIMEOUT_ERROR); 336 337 if (host->eh_deadline != -1 && !host->last_reset) 338 host->last_reset = jiffies; 339 340 if (host->hostt->eh_timed_out) 341 rtn = host->hostt->eh_timed_out(scmd); 342 343 if (rtn == BLK_EH_DONE) { 344 /* 345 * Set the command to complete first in order to prevent a real 346 * completion from releasing the command while error handling 347 * is using it. If the command was already completed, then the 348 * lower level driver beat the timeout handler, and it is safe 349 * to return without escalating error recovery. 350 * 351 * If timeout handling lost the race to a real completion, the 352 * block layer may ignore that due to a fake timeout injection, 353 * so return RESET_TIMER to allow error handling another shot 354 * at this command. 355 */ 356 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state)) 357 return BLK_EH_RESET_TIMER; 358 if (scsi_abort_command(scmd) != SUCCESS) { 359 set_host_byte(scmd, DID_TIME_OUT); 360 scsi_eh_scmd_add(scmd); 361 } 362 } 363 364 return rtn; 365 } 366 367 /** 368 * scsi_block_when_processing_errors - Prevent cmds from being queued. 369 * @sdev: Device on which we are performing recovery. 370 * 371 * Description: 372 * We block until the host is out of error recovery, and then check to 373 * see whether the host or the device is offline. 374 * 375 * Return value: 376 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 377 */ 378 int scsi_block_when_processing_errors(struct scsi_device *sdev) 379 { 380 int online; 381 382 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 383 384 online = scsi_device_online(sdev); 385 386 return online; 387 } 388 EXPORT_SYMBOL(scsi_block_when_processing_errors); 389 390 #ifdef CONFIG_SCSI_LOGGING 391 /** 392 * scsi_eh_prt_fail_stats - Log info on failures. 393 * @shost: scsi host being recovered. 394 * @work_q: Queue of scsi cmds to process. 395 */ 396 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 397 struct list_head *work_q) 398 { 399 struct scsi_cmnd *scmd; 400 struct scsi_device *sdev; 401 int total_failures = 0; 402 int cmd_failed = 0; 403 int cmd_cancel = 0; 404 int devices_failed = 0; 405 406 shost_for_each_device(sdev, shost) { 407 list_for_each_entry(scmd, work_q, eh_entry) { 408 if (scmd->device == sdev) { 409 ++total_failures; 410 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) 411 ++cmd_cancel; 412 else 413 ++cmd_failed; 414 } 415 } 416 417 if (cmd_cancel || cmd_failed) { 418 SCSI_LOG_ERROR_RECOVERY(3, 419 shost_printk(KERN_INFO, shost, 420 "%s: cmds failed: %d, cancel: %d\n", 421 __func__, cmd_failed, 422 cmd_cancel)); 423 cmd_cancel = 0; 424 cmd_failed = 0; 425 ++devices_failed; 426 } 427 } 428 429 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, 430 "Total of %d commands on %d" 431 " devices require eh work\n", 432 total_failures, devices_failed)); 433 } 434 #endif 435 436 /** 437 * scsi_report_lun_change - Set flag on all *other* devices on the same target 438 * to indicate that a UNIT ATTENTION is expected. 439 * @sdev: Device reporting the UNIT ATTENTION 440 */ 441 static void scsi_report_lun_change(struct scsi_device *sdev) 442 { 443 sdev->sdev_target->expecting_lun_change = 1; 444 } 445 446 /** 447 * scsi_report_sense - Examine scsi sense information and log messages for 448 * certain conditions, also issue uevents for some of them. 449 * @sdev: Device reporting the sense code 450 * @sshdr: sshdr to be examined 451 */ 452 static void scsi_report_sense(struct scsi_device *sdev, 453 struct scsi_sense_hdr *sshdr) 454 { 455 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ 456 457 if (sshdr->sense_key == UNIT_ATTENTION) { 458 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { 459 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; 460 sdev_printk(KERN_WARNING, sdev, 461 "Inquiry data has changed"); 462 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { 463 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; 464 scsi_report_lun_change(sdev); 465 sdev_printk(KERN_WARNING, sdev, 466 "Warning! Received an indication that the " 467 "LUN assignments on this target have " 468 "changed. The Linux SCSI layer does not " 469 "automatically remap LUN assignments.\n"); 470 } else if (sshdr->asc == 0x3f) 471 sdev_printk(KERN_WARNING, sdev, 472 "Warning! Received an indication that the " 473 "operating parameters on this target have " 474 "changed. The Linux SCSI layer does not " 475 "automatically adjust these parameters.\n"); 476 477 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { 478 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; 479 sdev_printk(KERN_WARNING, sdev, 480 "Warning! Received an indication that the " 481 "LUN reached a thin provisioning soft " 482 "threshold.\n"); 483 } 484 485 if (sshdr->asc == 0x29) { 486 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED; 487 sdev_printk(KERN_WARNING, sdev, 488 "Power-on or device reset occurred\n"); 489 } 490 491 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { 492 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; 493 sdev_printk(KERN_WARNING, sdev, 494 "Mode parameters changed"); 495 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { 496 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; 497 sdev_printk(KERN_WARNING, sdev, 498 "Asymmetric access state changed"); 499 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { 500 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; 501 sdev_printk(KERN_WARNING, sdev, 502 "Capacity data has changed"); 503 } else if (sshdr->asc == 0x2a) 504 sdev_printk(KERN_WARNING, sdev, 505 "Parameters changed"); 506 } 507 508 if (evt_type != SDEV_EVT_MAXBITS) { 509 set_bit(evt_type, sdev->pending_events); 510 schedule_work(&sdev->event_work); 511 } 512 } 513 514 /** 515 * scsi_check_sense - Examine scsi cmd sense 516 * @scmd: Cmd to have sense checked. 517 * 518 * Return value: 519 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE 520 * 521 * Notes: 522 * When a deferred error is detected the current command has 523 * not been executed and needs retrying. 524 */ 525 enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd) 526 { 527 struct scsi_device *sdev = scmd->device; 528 struct scsi_sense_hdr sshdr; 529 530 if (! scsi_command_normalize_sense(scmd, &sshdr)) 531 return FAILED; /* no valid sense data */ 532 533 scsi_report_sense(sdev, &sshdr); 534 535 if (scsi_sense_is_deferred(&sshdr)) 536 return NEEDS_RETRY; 537 538 if (sdev->handler && sdev->handler->check_sense) { 539 enum scsi_disposition rc; 540 541 rc = sdev->handler->check_sense(sdev, &sshdr); 542 if (rc != SCSI_RETURN_NOT_HANDLED) 543 return rc; 544 /* handler does not care. Drop down to default handling */ 545 } 546 547 if (scmd->cmnd[0] == TEST_UNIT_READY && 548 scmd->submitter != SUBMITTED_BY_SCSI_ERROR_HANDLER) 549 /* 550 * nasty: for mid-layer issued TURs, we need to return the 551 * actual sense data without any recovery attempt. For eh 552 * issued ones, we need to try to recover and interpret 553 */ 554 return SUCCESS; 555 556 /* 557 * Previous logic looked for FILEMARK, EOM or ILI which are 558 * mainly associated with tapes and returned SUCCESS. 559 */ 560 if (sshdr.response_code == 0x70) { 561 /* fixed format */ 562 if (scmd->sense_buffer[2] & 0xe0) 563 return SUCCESS; 564 } else { 565 /* 566 * descriptor format: look for "stream commands sense data 567 * descriptor" (see SSC-3). Assume single sense data 568 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 569 */ 570 if ((sshdr.additional_length > 3) && 571 (scmd->sense_buffer[8] == 0x4) && 572 (scmd->sense_buffer[11] & 0xe0)) 573 return SUCCESS; 574 } 575 576 switch (sshdr.sense_key) { 577 case NO_SENSE: 578 return SUCCESS; 579 case RECOVERED_ERROR: 580 return /* soft_error */ SUCCESS; 581 582 case ABORTED_COMMAND: 583 if (sshdr.asc == 0x10) /* DIF */ 584 return SUCCESS; 585 586 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF) 587 return ADD_TO_MLQUEUE; 588 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 && 589 sdev->sdev_bflags & BLIST_RETRY_ASC_C1) 590 return ADD_TO_MLQUEUE; 591 592 return NEEDS_RETRY; 593 case NOT_READY: 594 case UNIT_ATTENTION: 595 /* 596 * if we are expecting a cc/ua because of a bus reset that we 597 * performed, treat this just as a retry. otherwise this is 598 * information that we should pass up to the upper-level driver 599 * so that we can deal with it there. 600 */ 601 if (scmd->device->expecting_cc_ua) { 602 /* 603 * Because some device does not queue unit 604 * attentions correctly, we carefully check 605 * additional sense code and qualifier so as 606 * not to squash media change unit attention. 607 */ 608 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { 609 scmd->device->expecting_cc_ua = 0; 610 return NEEDS_RETRY; 611 } 612 } 613 /* 614 * we might also expect a cc/ua if another LUN on the target 615 * reported a UA with an ASC/ASCQ of 3F 0E - 616 * REPORTED LUNS DATA HAS CHANGED. 617 */ 618 if (scmd->device->sdev_target->expecting_lun_change && 619 sshdr.asc == 0x3f && sshdr.ascq == 0x0e) 620 return NEEDS_RETRY; 621 /* 622 * if the device is in the process of becoming ready, we 623 * should retry. 624 */ 625 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 626 return NEEDS_RETRY; 627 /* 628 * if the device is not started, we need to wake 629 * the error handler to start the motor 630 */ 631 if (scmd->device->allow_restart && 632 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 633 return FAILED; 634 /* 635 * Pass the UA upwards for a determination in the completion 636 * functions. 637 */ 638 return SUCCESS; 639 640 /* these are not supported */ 641 case DATA_PROTECT: 642 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { 643 /* Thin provisioning hard threshold reached */ 644 set_host_byte(scmd, DID_ALLOC_FAILURE); 645 return SUCCESS; 646 } 647 fallthrough; 648 case COPY_ABORTED: 649 case VOLUME_OVERFLOW: 650 case MISCOMPARE: 651 case BLANK_CHECK: 652 set_host_byte(scmd, DID_TARGET_FAILURE); 653 return SUCCESS; 654 655 case MEDIUM_ERROR: 656 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 657 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 658 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 659 set_host_byte(scmd, DID_MEDIUM_ERROR); 660 return SUCCESS; 661 } 662 return NEEDS_RETRY; 663 664 case HARDWARE_ERROR: 665 if (scmd->device->retry_hwerror) 666 return ADD_TO_MLQUEUE; 667 else 668 set_host_byte(scmd, DID_TARGET_FAILURE); 669 fallthrough; 670 671 case ILLEGAL_REQUEST: 672 if (sshdr.asc == 0x20 || /* Invalid command operation code */ 673 sshdr.asc == 0x21 || /* Logical block address out of range */ 674 sshdr.asc == 0x22 || /* Invalid function */ 675 sshdr.asc == 0x24 || /* Invalid field in cdb */ 676 sshdr.asc == 0x26 || /* Parameter value invalid */ 677 sshdr.asc == 0x27) { /* Write protected */ 678 set_host_byte(scmd, DID_TARGET_FAILURE); 679 } 680 return SUCCESS; 681 682 default: 683 return SUCCESS; 684 } 685 } 686 EXPORT_SYMBOL_GPL(scsi_check_sense); 687 688 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) 689 { 690 struct scsi_host_template *sht = sdev->host->hostt; 691 struct scsi_device *tmp_sdev; 692 693 if (!sht->track_queue_depth || 694 sdev->queue_depth >= sdev->max_queue_depth) 695 return; 696 697 if (time_before(jiffies, 698 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) 699 return; 700 701 if (time_before(jiffies, 702 sdev->last_queue_full_time + sdev->queue_ramp_up_period)) 703 return; 704 705 /* 706 * Walk all devices of a target and do 707 * ramp up on them. 708 */ 709 shost_for_each_device(tmp_sdev, sdev->host) { 710 if (tmp_sdev->channel != sdev->channel || 711 tmp_sdev->id != sdev->id || 712 tmp_sdev->queue_depth == sdev->max_queue_depth) 713 continue; 714 715 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); 716 sdev->last_queue_ramp_up = jiffies; 717 } 718 } 719 720 static void scsi_handle_queue_full(struct scsi_device *sdev) 721 { 722 struct scsi_host_template *sht = sdev->host->hostt; 723 struct scsi_device *tmp_sdev; 724 725 if (!sht->track_queue_depth) 726 return; 727 728 shost_for_each_device(tmp_sdev, sdev->host) { 729 if (tmp_sdev->channel != sdev->channel || 730 tmp_sdev->id != sdev->id) 731 continue; 732 /* 733 * We do not know the number of commands that were at 734 * the device when we got the queue full so we start 735 * from the highest possible value and work our way down. 736 */ 737 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); 738 } 739 } 740 741 /** 742 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 743 * @scmd: SCSI cmd to examine. 744 * 745 * Notes: 746 * This is *only* called when we are examining the status of commands 747 * queued during error recovery. the main difference here is that we 748 * don't allow for the possibility of retries here, and we are a lot 749 * more restrictive about what we consider acceptable. 750 */ 751 static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd) 752 { 753 /* 754 * first check the host byte, to see if there is anything in there 755 * that would indicate what we need to do. 756 */ 757 if (host_byte(scmd->result) == DID_RESET) { 758 /* 759 * rats. we are already in the error handler, so we now 760 * get to try and figure out what to do next. if the sense 761 * is valid, we have a pretty good idea of what to do. 762 * if not, we mark it as FAILED. 763 */ 764 return scsi_check_sense(scmd); 765 } 766 if (host_byte(scmd->result) != DID_OK) 767 return FAILED; 768 769 /* 770 * now, check the status byte to see if this indicates 771 * anything special. 772 */ 773 switch (get_status_byte(scmd)) { 774 case SAM_STAT_GOOD: 775 scsi_handle_queue_ramp_up(scmd->device); 776 fallthrough; 777 case SAM_STAT_COMMAND_TERMINATED: 778 return SUCCESS; 779 case SAM_STAT_CHECK_CONDITION: 780 return scsi_check_sense(scmd); 781 case SAM_STAT_CONDITION_MET: 782 case SAM_STAT_INTERMEDIATE: 783 case SAM_STAT_INTERMEDIATE_CONDITION_MET: 784 /* 785 * who knows? FIXME(eric) 786 */ 787 return SUCCESS; 788 case SAM_STAT_RESERVATION_CONFLICT: 789 if (scmd->cmnd[0] == TEST_UNIT_READY) 790 /* it is a success, we probed the device and 791 * found it */ 792 return SUCCESS; 793 /* otherwise, we failed to send the command */ 794 return FAILED; 795 case SAM_STAT_TASK_SET_FULL: 796 scsi_handle_queue_full(scmd->device); 797 fallthrough; 798 case SAM_STAT_BUSY: 799 return NEEDS_RETRY; 800 default: 801 return FAILED; 802 } 803 return FAILED; 804 } 805 806 /** 807 * scsi_eh_done - Completion function for error handling. 808 * @scmd: Cmd that is done. 809 */ 810 void scsi_eh_done(struct scsi_cmnd *scmd) 811 { 812 struct completion *eh_action; 813 814 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 815 "%s result: %x\n", __func__, scmd->result)); 816 817 eh_action = scmd->device->host->eh_action; 818 if (eh_action) 819 complete(eh_action); 820 } 821 822 /** 823 * scsi_try_host_reset - ask host adapter to reset itself 824 * @scmd: SCSI cmd to send host reset. 825 */ 826 static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd) 827 { 828 unsigned long flags; 829 enum scsi_disposition rtn; 830 struct Scsi_Host *host = scmd->device->host; 831 struct scsi_host_template *hostt = host->hostt; 832 833 SCSI_LOG_ERROR_RECOVERY(3, 834 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 835 836 if (!hostt->eh_host_reset_handler) 837 return FAILED; 838 839 rtn = hostt->eh_host_reset_handler(scmd); 840 841 if (rtn == SUCCESS) { 842 if (!hostt->skip_settle_delay) 843 ssleep(HOST_RESET_SETTLE_TIME); 844 spin_lock_irqsave(host->host_lock, flags); 845 scsi_report_bus_reset(host, scmd_channel(scmd)); 846 spin_unlock_irqrestore(host->host_lock, flags); 847 } 848 849 return rtn; 850 } 851 852 /** 853 * scsi_try_bus_reset - ask host to perform a bus reset 854 * @scmd: SCSI cmd to send bus reset. 855 */ 856 static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd) 857 { 858 unsigned long flags; 859 enum scsi_disposition rtn; 860 struct Scsi_Host *host = scmd->device->host; 861 struct scsi_host_template *hostt = host->hostt; 862 863 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 864 "%s: Snd Bus RST\n", __func__)); 865 866 if (!hostt->eh_bus_reset_handler) 867 return FAILED; 868 869 rtn = hostt->eh_bus_reset_handler(scmd); 870 871 if (rtn == SUCCESS) { 872 if (!hostt->skip_settle_delay) 873 ssleep(BUS_RESET_SETTLE_TIME); 874 spin_lock_irqsave(host->host_lock, flags); 875 scsi_report_bus_reset(host, scmd_channel(scmd)); 876 spin_unlock_irqrestore(host->host_lock, flags); 877 } 878 879 return rtn; 880 } 881 882 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 883 { 884 sdev->was_reset = 1; 885 sdev->expecting_cc_ua = 1; 886 } 887 888 /** 889 * scsi_try_target_reset - Ask host to perform a target reset 890 * @scmd: SCSI cmd used to send a target reset 891 * 892 * Notes: 893 * There is no timeout for this operation. if this operation is 894 * unreliable for a given host, then the host itself needs to put a 895 * timer on it, and set the host back to a consistent state prior to 896 * returning. 897 */ 898 static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd) 899 { 900 unsigned long flags; 901 enum scsi_disposition rtn; 902 struct Scsi_Host *host = scmd->device->host; 903 struct scsi_host_template *hostt = host->hostt; 904 905 if (!hostt->eh_target_reset_handler) 906 return FAILED; 907 908 rtn = hostt->eh_target_reset_handler(scmd); 909 if (rtn == SUCCESS) { 910 spin_lock_irqsave(host->host_lock, flags); 911 __starget_for_each_device(scsi_target(scmd->device), NULL, 912 __scsi_report_device_reset); 913 spin_unlock_irqrestore(host->host_lock, flags); 914 } 915 916 return rtn; 917 } 918 919 /** 920 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 921 * @scmd: SCSI cmd used to send BDR 922 * 923 * Notes: 924 * There is no timeout for this operation. if this operation is 925 * unreliable for a given host, then the host itself needs to put a 926 * timer on it, and set the host back to a consistent state prior to 927 * returning. 928 */ 929 static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 930 { 931 enum scsi_disposition rtn; 932 struct scsi_host_template *hostt = scmd->device->host->hostt; 933 934 if (!hostt->eh_device_reset_handler) 935 return FAILED; 936 937 rtn = hostt->eh_device_reset_handler(scmd); 938 if (rtn == SUCCESS) 939 __scsi_report_device_reset(scmd->device, NULL); 940 return rtn; 941 } 942 943 /** 944 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command 945 * @hostt: SCSI driver host template 946 * @scmd: SCSI cmd used to send a target reset 947 * 948 * Return value: 949 * SUCCESS, FAILED, or FAST_IO_FAIL 950 * 951 * Notes: 952 * SUCCESS does not necessarily indicate that the command 953 * has been aborted; it only indicates that the LLDDs 954 * has cleared all references to that command. 955 * LLDDs should return FAILED only if an abort was required 956 * but could not be executed. LLDDs should return FAST_IO_FAIL 957 * if the device is temporarily unavailable (eg due to a 958 * link down on FibreChannel) 959 */ 960 static enum scsi_disposition 961 scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd) 962 { 963 if (!hostt->eh_abort_handler) 964 return FAILED; 965 966 return hostt->eh_abort_handler(scmd); 967 } 968 969 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 970 { 971 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 972 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 973 if (scsi_try_target_reset(scmd) != SUCCESS) 974 if (scsi_try_bus_reset(scmd) != SUCCESS) 975 scsi_try_host_reset(scmd); 976 } 977 978 /** 979 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 980 * @scmd: SCSI command structure to hijack 981 * @ses: structure to save restore information 982 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 983 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) 984 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 985 * 986 * This function is used to save a scsi command information before re-execution 987 * as part of the error recovery process. If @sense_bytes is 0 the command 988 * sent must be one that does not transfer any data. If @sense_bytes != 0 989 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 990 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 991 */ 992 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 993 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 994 { 995 struct scsi_device *sdev = scmd->device; 996 997 /* 998 * We need saved copies of a number of fields - this is because 999 * error handling may need to overwrite these with different values 1000 * to run different commands, and once error handling is complete, 1001 * we will need to restore these values prior to running the actual 1002 * command. 1003 */ 1004 ses->cmd_len = scmd->cmd_len; 1005 ses->cmnd = scmd->cmnd; 1006 ses->data_direction = scmd->sc_data_direction; 1007 ses->sdb = scmd->sdb; 1008 ses->result = scmd->result; 1009 ses->resid_len = scmd->req.resid_len; 1010 ses->underflow = scmd->underflow; 1011 ses->prot_op = scmd->prot_op; 1012 ses->eh_eflags = scmd->eh_eflags; 1013 1014 scmd->prot_op = SCSI_PROT_NORMAL; 1015 scmd->eh_eflags = 0; 1016 scmd->cmnd = ses->eh_cmnd; 1017 memset(scmd->cmnd, 0, BLK_MAX_CDB); 1018 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 1019 scmd->result = 0; 1020 scmd->req.resid_len = 0; 1021 1022 if (sense_bytes) { 1023 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 1024 sense_bytes); 1025 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 1026 scmd->sdb.length); 1027 scmd->sdb.table.sgl = &ses->sense_sgl; 1028 scmd->sc_data_direction = DMA_FROM_DEVICE; 1029 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; 1030 scmd->cmnd[0] = REQUEST_SENSE; 1031 scmd->cmnd[4] = scmd->sdb.length; 1032 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1033 } else { 1034 scmd->sc_data_direction = DMA_NONE; 1035 if (cmnd) { 1036 BUG_ON(cmnd_size > BLK_MAX_CDB); 1037 memcpy(scmd->cmnd, cmnd, cmnd_size); 1038 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1039 } 1040 } 1041 1042 scmd->underflow = 0; 1043 1044 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 1045 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 1046 (sdev->lun << 5 & 0xe0); 1047 1048 /* 1049 * Zero the sense buffer. The scsi spec mandates that any 1050 * untransferred sense data should be interpreted as being zero. 1051 */ 1052 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 1053 } 1054 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 1055 1056 /** 1057 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 1058 * @scmd: SCSI command structure to restore 1059 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 1060 * 1061 * Undo any damage done by above scsi_eh_prep_cmnd(). 1062 */ 1063 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 1064 { 1065 /* 1066 * Restore original data 1067 */ 1068 scmd->cmd_len = ses->cmd_len; 1069 scmd->cmnd = ses->cmnd; 1070 scmd->sc_data_direction = ses->data_direction; 1071 scmd->sdb = ses->sdb; 1072 scmd->result = ses->result; 1073 scmd->req.resid_len = ses->resid_len; 1074 scmd->underflow = ses->underflow; 1075 scmd->prot_op = ses->prot_op; 1076 scmd->eh_eflags = ses->eh_eflags; 1077 } 1078 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 1079 1080 /** 1081 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 1082 * @scmd: SCSI command structure to hijack 1083 * @cmnd: CDB to send 1084 * @cmnd_size: size in bytes of @cmnd 1085 * @timeout: timeout for this request 1086 * @sense_bytes: size of sense data to copy or 0 1087 * 1088 * This function is used to send a scsi command down to a target device 1089 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 1090 * 1091 * Return value: 1092 * SUCCESS or FAILED or NEEDS_RETRY 1093 */ 1094 static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd, 1095 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes) 1096 { 1097 struct scsi_device *sdev = scmd->device; 1098 struct Scsi_Host *shost = sdev->host; 1099 DECLARE_COMPLETION_ONSTACK(done); 1100 unsigned long timeleft = timeout, delay; 1101 struct scsi_eh_save ses; 1102 const unsigned long stall_for = msecs_to_jiffies(100); 1103 int rtn; 1104 1105 retry: 1106 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1107 shost->eh_action = &done; 1108 1109 scsi_log_send(scmd); 1110 scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER; 1111 1112 /* 1113 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can 1114 * change the SCSI device state after we have examined it and before 1115 * .queuecommand() is called. 1116 */ 1117 mutex_lock(&sdev->state_mutex); 1118 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) { 1119 mutex_unlock(&sdev->state_mutex); 1120 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev, 1121 "%s: state %d <> %d\n", __func__, sdev->sdev_state, 1122 SDEV_BLOCK)); 1123 delay = min(timeleft, stall_for); 1124 timeleft -= delay; 1125 msleep(jiffies_to_msecs(delay)); 1126 mutex_lock(&sdev->state_mutex); 1127 } 1128 if (sdev->sdev_state != SDEV_BLOCK) 1129 rtn = shost->hostt->queuecommand(shost, scmd); 1130 else 1131 rtn = FAILED; 1132 mutex_unlock(&sdev->state_mutex); 1133 1134 if (rtn) { 1135 if (timeleft > stall_for) { 1136 scsi_eh_restore_cmnd(scmd, &ses); 1137 1138 timeleft -= stall_for; 1139 msleep(jiffies_to_msecs(stall_for)); 1140 goto retry; 1141 } 1142 /* signal not to enter either branch of the if () below */ 1143 timeleft = 0; 1144 rtn = FAILED; 1145 } else { 1146 timeleft = wait_for_completion_timeout(&done, timeout); 1147 rtn = SUCCESS; 1148 } 1149 1150 shost->eh_action = NULL; 1151 1152 scsi_log_completion(scmd, rtn); 1153 1154 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1155 "%s timeleft: %ld\n", 1156 __func__, timeleft)); 1157 1158 /* 1159 * If there is time left scsi_eh_done got called, and we will examine 1160 * the actual status codes to see whether the command actually did 1161 * complete normally, else if we have a zero return and no time left, 1162 * the command must still be pending, so abort it and return FAILED. 1163 * If we never actually managed to issue the command, because 1164 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1165 * value above (so don't execute either branch of the if) 1166 */ 1167 if (timeleft) { 1168 rtn = scsi_eh_completed_normally(scmd); 1169 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1170 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1171 1172 switch (rtn) { 1173 case SUCCESS: 1174 case NEEDS_RETRY: 1175 case FAILED: 1176 break; 1177 case ADD_TO_MLQUEUE: 1178 rtn = NEEDS_RETRY; 1179 break; 1180 default: 1181 rtn = FAILED; 1182 break; 1183 } 1184 } else if (rtn != FAILED) { 1185 scsi_abort_eh_cmnd(scmd); 1186 rtn = FAILED; 1187 } 1188 1189 scsi_eh_restore_cmnd(scmd, &ses); 1190 1191 return rtn; 1192 } 1193 1194 /** 1195 * scsi_request_sense - Request sense data from a particular target. 1196 * @scmd: SCSI cmd for request sense. 1197 * 1198 * Notes: 1199 * Some hosts automatically obtain this information, others require 1200 * that we obtain it on our own. This function will *not* return until 1201 * the command either times out, or it completes. 1202 */ 1203 static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd) 1204 { 1205 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1206 } 1207 1208 static enum scsi_disposition 1209 scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn) 1210 { 1211 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) { 1212 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1213 if (sdrv->eh_action) 1214 rtn = sdrv->eh_action(scmd, rtn); 1215 } 1216 return rtn; 1217 } 1218 1219 /** 1220 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1221 * @scmd: Original SCSI cmd that eh has finished. 1222 * @done_q: Queue for processed commands. 1223 * 1224 * Notes: 1225 * We don't want to use the normal command completion while we are are 1226 * still handling errors - it may cause other commands to be queued, 1227 * and that would disturb what we are doing. Thus we really want to 1228 * keep a list of pending commands for final completion, and once we 1229 * are ready to leave error handling we handle completion for real. 1230 */ 1231 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1232 { 1233 list_move_tail(&scmd->eh_entry, done_q); 1234 } 1235 EXPORT_SYMBOL(scsi_eh_finish_cmd); 1236 1237 /** 1238 * scsi_eh_get_sense - Get device sense data. 1239 * @work_q: Queue of commands to process. 1240 * @done_q: Queue of processed commands. 1241 * 1242 * Description: 1243 * See if we need to request sense information. if so, then get it 1244 * now, so we have a better idea of what to do. 1245 * 1246 * Notes: 1247 * This has the unfortunate side effect that if a shost adapter does 1248 * not automatically request sense information, we end up shutting 1249 * it down before we request it. 1250 * 1251 * All drivers should request sense information internally these days, 1252 * so for now all I have to say is tough noogies if you end up in here. 1253 * 1254 * XXX: Long term this code should go away, but that needs an audit of 1255 * all LLDDs first. 1256 */ 1257 int scsi_eh_get_sense(struct list_head *work_q, 1258 struct list_head *done_q) 1259 { 1260 struct scsi_cmnd *scmd, *next; 1261 struct Scsi_Host *shost; 1262 enum scsi_disposition rtn; 1263 1264 /* 1265 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, 1266 * should not get sense. 1267 */ 1268 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1269 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || 1270 SCSI_SENSE_VALID(scmd)) 1271 continue; 1272 1273 shost = scmd->device->host; 1274 if (scsi_host_eh_past_deadline(shost)) { 1275 SCSI_LOG_ERROR_RECOVERY(3, 1276 scmd_printk(KERN_INFO, scmd, 1277 "%s: skip request sense, past eh deadline\n", 1278 current->comm)); 1279 break; 1280 } 1281 if (!scsi_status_is_check_condition(scmd->result)) 1282 /* 1283 * don't request sense if there's no check condition 1284 * status because the error we're processing isn't one 1285 * that has a sense code (and some devices get 1286 * confused by sense requests out of the blue) 1287 */ 1288 continue; 1289 1290 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1291 "%s: requesting sense\n", 1292 current->comm)); 1293 rtn = scsi_request_sense(scmd); 1294 if (rtn != SUCCESS) 1295 continue; 1296 1297 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1298 "sense requested, result %x\n", scmd->result)); 1299 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); 1300 1301 rtn = scsi_decide_disposition(scmd); 1302 1303 /* 1304 * if the result was normal, then just pass it along to the 1305 * upper level. 1306 */ 1307 if (rtn == SUCCESS) 1308 /* 1309 * We don't want this command reissued, just finished 1310 * with the sense data, so set retries to the max 1311 * allowed to ensure it won't get reissued. If the user 1312 * has requested infinite retries, we also want to 1313 * finish this command, so force completion by setting 1314 * retries and allowed to the same value. 1315 */ 1316 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) 1317 scmd->retries = scmd->allowed = 1; 1318 else 1319 scmd->retries = scmd->allowed; 1320 else if (rtn != NEEDS_RETRY) 1321 continue; 1322 1323 scsi_eh_finish_cmd(scmd, done_q); 1324 } 1325 1326 return list_empty(work_q); 1327 } 1328 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1329 1330 /** 1331 * scsi_eh_tur - Send TUR to device. 1332 * @scmd: &scsi_cmnd to send TUR 1333 * 1334 * Return value: 1335 * 0 - Device is ready. 1 - Device NOT ready. 1336 */ 1337 static int scsi_eh_tur(struct scsi_cmnd *scmd) 1338 { 1339 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1340 int retry_cnt = 1; 1341 enum scsi_disposition rtn; 1342 1343 retry_tur: 1344 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1345 scmd->device->eh_timeout, 0); 1346 1347 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1348 "%s return: %x\n", __func__, rtn)); 1349 1350 switch (rtn) { 1351 case NEEDS_RETRY: 1352 if (retry_cnt--) 1353 goto retry_tur; 1354 fallthrough; 1355 case SUCCESS: 1356 return 0; 1357 default: 1358 return 1; 1359 } 1360 } 1361 1362 /** 1363 * scsi_eh_test_devices - check if devices are responding from error recovery. 1364 * @cmd_list: scsi commands in error recovery. 1365 * @work_q: queue for commands which still need more error recovery 1366 * @done_q: queue for commands which are finished 1367 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1368 * 1369 * Decription: 1370 * Tests if devices are in a working state. Commands to devices now in 1371 * a working state are sent to the done_q while commands to devices which 1372 * are still failing to respond are returned to the work_q for more 1373 * processing. 1374 **/ 1375 static int scsi_eh_test_devices(struct list_head *cmd_list, 1376 struct list_head *work_q, 1377 struct list_head *done_q, int try_stu) 1378 { 1379 struct scsi_cmnd *scmd, *next; 1380 struct scsi_device *sdev; 1381 int finish_cmds; 1382 1383 while (!list_empty(cmd_list)) { 1384 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1385 sdev = scmd->device; 1386 1387 if (!try_stu) { 1388 if (scsi_host_eh_past_deadline(sdev->host)) { 1389 /* Push items back onto work_q */ 1390 list_splice_init(cmd_list, work_q); 1391 SCSI_LOG_ERROR_RECOVERY(3, 1392 sdev_printk(KERN_INFO, sdev, 1393 "%s: skip test device, past eh deadline", 1394 current->comm)); 1395 break; 1396 } 1397 } 1398 1399 finish_cmds = !scsi_device_online(scmd->device) || 1400 (try_stu && !scsi_eh_try_stu(scmd) && 1401 !scsi_eh_tur(scmd)) || 1402 !scsi_eh_tur(scmd); 1403 1404 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1405 if (scmd->device == sdev) { 1406 if (finish_cmds && 1407 (try_stu || 1408 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1409 scsi_eh_finish_cmd(scmd, done_q); 1410 else 1411 list_move_tail(&scmd->eh_entry, work_q); 1412 } 1413 } 1414 return list_empty(work_q); 1415 } 1416 1417 /** 1418 * scsi_eh_try_stu - Send START_UNIT to device. 1419 * @scmd: &scsi_cmnd to send START_UNIT 1420 * 1421 * Return value: 1422 * 0 - Device is ready. 1 - Device NOT ready. 1423 */ 1424 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1425 { 1426 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1427 1428 if (scmd->device->allow_restart) { 1429 int i; 1430 enum scsi_disposition rtn = NEEDS_RETRY; 1431 1432 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1433 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, 1434 scmd->device->eh_timeout, 0); 1435 1436 if (rtn == SUCCESS) 1437 return 0; 1438 } 1439 1440 return 1; 1441 } 1442 1443 /** 1444 * scsi_eh_stu - send START_UNIT if needed 1445 * @shost: &scsi host being recovered. 1446 * @work_q: &list_head for pending commands. 1447 * @done_q: &list_head for processed commands. 1448 * 1449 * Notes: 1450 * If commands are failing due to not ready, initializing command required, 1451 * try revalidating the device, which will end up sending a start unit. 1452 */ 1453 static int scsi_eh_stu(struct Scsi_Host *shost, 1454 struct list_head *work_q, 1455 struct list_head *done_q) 1456 { 1457 struct scsi_cmnd *scmd, *stu_scmd, *next; 1458 struct scsi_device *sdev; 1459 1460 shost_for_each_device(sdev, shost) { 1461 if (scsi_host_eh_past_deadline(shost)) { 1462 SCSI_LOG_ERROR_RECOVERY(3, 1463 sdev_printk(KERN_INFO, sdev, 1464 "%s: skip START_UNIT, past eh deadline\n", 1465 current->comm)); 1466 scsi_device_put(sdev); 1467 break; 1468 } 1469 stu_scmd = NULL; 1470 list_for_each_entry(scmd, work_q, eh_entry) 1471 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1472 scsi_check_sense(scmd) == FAILED ) { 1473 stu_scmd = scmd; 1474 break; 1475 } 1476 1477 if (!stu_scmd) 1478 continue; 1479 1480 SCSI_LOG_ERROR_RECOVERY(3, 1481 sdev_printk(KERN_INFO, sdev, 1482 "%s: Sending START_UNIT\n", 1483 current->comm)); 1484 1485 if (!scsi_eh_try_stu(stu_scmd)) { 1486 if (!scsi_device_online(sdev) || 1487 !scsi_eh_tur(stu_scmd)) { 1488 list_for_each_entry_safe(scmd, next, 1489 work_q, eh_entry) { 1490 if (scmd->device == sdev && 1491 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1492 scsi_eh_finish_cmd(scmd, done_q); 1493 } 1494 } 1495 } else { 1496 SCSI_LOG_ERROR_RECOVERY(3, 1497 sdev_printk(KERN_INFO, sdev, 1498 "%s: START_UNIT failed\n", 1499 current->comm)); 1500 } 1501 } 1502 1503 return list_empty(work_q); 1504 } 1505 1506 1507 /** 1508 * scsi_eh_bus_device_reset - send bdr if needed 1509 * @shost: scsi host being recovered. 1510 * @work_q: &list_head for pending commands. 1511 * @done_q: &list_head for processed commands. 1512 * 1513 * Notes: 1514 * Try a bus device reset. Still, look to see whether we have multiple 1515 * devices that are jammed or not - if we have multiple devices, it 1516 * makes no sense to try bus_device_reset - we really would need to try 1517 * a bus_reset instead. 1518 */ 1519 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1520 struct list_head *work_q, 1521 struct list_head *done_q) 1522 { 1523 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1524 struct scsi_device *sdev; 1525 enum scsi_disposition rtn; 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 BDR, past eh deadline\n", 1532 current->comm)); 1533 scsi_device_put(sdev); 1534 break; 1535 } 1536 bdr_scmd = NULL; 1537 list_for_each_entry(scmd, work_q, eh_entry) 1538 if (scmd->device == sdev) { 1539 bdr_scmd = scmd; 1540 break; 1541 } 1542 1543 if (!bdr_scmd) 1544 continue; 1545 1546 SCSI_LOG_ERROR_RECOVERY(3, 1547 sdev_printk(KERN_INFO, sdev, 1548 "%s: Sending BDR\n", current->comm)); 1549 rtn = scsi_try_bus_device_reset(bdr_scmd); 1550 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1551 if (!scsi_device_online(sdev) || 1552 rtn == FAST_IO_FAIL || 1553 !scsi_eh_tur(bdr_scmd)) { 1554 list_for_each_entry_safe(scmd, next, 1555 work_q, eh_entry) { 1556 if (scmd->device == sdev && 1557 scsi_eh_action(scmd, rtn) != FAILED) 1558 scsi_eh_finish_cmd(scmd, 1559 done_q); 1560 } 1561 } 1562 } else { 1563 SCSI_LOG_ERROR_RECOVERY(3, 1564 sdev_printk(KERN_INFO, sdev, 1565 "%s: BDR failed\n", current->comm)); 1566 } 1567 } 1568 1569 return list_empty(work_q); 1570 } 1571 1572 /** 1573 * scsi_eh_target_reset - send target reset if needed 1574 * @shost: scsi host being recovered. 1575 * @work_q: &list_head for pending commands. 1576 * @done_q: &list_head for processed commands. 1577 * 1578 * Notes: 1579 * Try a target reset. 1580 */ 1581 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1582 struct list_head *work_q, 1583 struct list_head *done_q) 1584 { 1585 LIST_HEAD(tmp_list); 1586 LIST_HEAD(check_list); 1587 1588 list_splice_init(work_q, &tmp_list); 1589 1590 while (!list_empty(&tmp_list)) { 1591 struct scsi_cmnd *next, *scmd; 1592 enum scsi_disposition rtn; 1593 unsigned int id; 1594 1595 if (scsi_host_eh_past_deadline(shost)) { 1596 /* push back on work queue for further processing */ 1597 list_splice_init(&check_list, work_q); 1598 list_splice_init(&tmp_list, work_q); 1599 SCSI_LOG_ERROR_RECOVERY(3, 1600 shost_printk(KERN_INFO, shost, 1601 "%s: Skip target reset, past eh deadline\n", 1602 current->comm)); 1603 return list_empty(work_q); 1604 } 1605 1606 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1607 id = scmd_id(scmd); 1608 1609 SCSI_LOG_ERROR_RECOVERY(3, 1610 shost_printk(KERN_INFO, shost, 1611 "%s: Sending target reset to target %d\n", 1612 current->comm, id)); 1613 rtn = scsi_try_target_reset(scmd); 1614 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1615 SCSI_LOG_ERROR_RECOVERY(3, 1616 shost_printk(KERN_INFO, shost, 1617 "%s: Target reset failed" 1618 " target: %d\n", 1619 current->comm, id)); 1620 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1621 if (scmd_id(scmd) != id) 1622 continue; 1623 1624 if (rtn == SUCCESS) 1625 list_move_tail(&scmd->eh_entry, &check_list); 1626 else if (rtn == FAST_IO_FAIL) 1627 scsi_eh_finish_cmd(scmd, done_q); 1628 else 1629 /* push back on work queue for further processing */ 1630 list_move(&scmd->eh_entry, work_q); 1631 } 1632 } 1633 1634 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1635 } 1636 1637 /** 1638 * scsi_eh_bus_reset - send a bus reset 1639 * @shost: &scsi host being recovered. 1640 * @work_q: &list_head for pending commands. 1641 * @done_q: &list_head for processed commands. 1642 */ 1643 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1644 struct list_head *work_q, 1645 struct list_head *done_q) 1646 { 1647 struct scsi_cmnd *scmd, *chan_scmd, *next; 1648 LIST_HEAD(check_list); 1649 unsigned int channel; 1650 enum scsi_disposition rtn; 1651 1652 /* 1653 * we really want to loop over the various channels, and do this on 1654 * a channel by channel basis. we should also check to see if any 1655 * of the failed commands are on soft_reset devices, and if so, skip 1656 * the reset. 1657 */ 1658 1659 for (channel = 0; channel <= shost->max_channel; channel++) { 1660 if (scsi_host_eh_past_deadline(shost)) { 1661 list_splice_init(&check_list, work_q); 1662 SCSI_LOG_ERROR_RECOVERY(3, 1663 shost_printk(KERN_INFO, shost, 1664 "%s: skip BRST, past eh deadline\n", 1665 current->comm)); 1666 return list_empty(work_q); 1667 } 1668 1669 chan_scmd = NULL; 1670 list_for_each_entry(scmd, work_q, eh_entry) { 1671 if (channel == scmd_channel(scmd)) { 1672 chan_scmd = scmd; 1673 break; 1674 /* 1675 * FIXME add back in some support for 1676 * soft_reset devices. 1677 */ 1678 } 1679 } 1680 1681 if (!chan_scmd) 1682 continue; 1683 SCSI_LOG_ERROR_RECOVERY(3, 1684 shost_printk(KERN_INFO, shost, 1685 "%s: Sending BRST chan: %d\n", 1686 current->comm, channel)); 1687 rtn = scsi_try_bus_reset(chan_scmd); 1688 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1689 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1690 if (channel == scmd_channel(scmd)) { 1691 if (rtn == FAST_IO_FAIL) 1692 scsi_eh_finish_cmd(scmd, 1693 done_q); 1694 else 1695 list_move_tail(&scmd->eh_entry, 1696 &check_list); 1697 } 1698 } 1699 } else { 1700 SCSI_LOG_ERROR_RECOVERY(3, 1701 shost_printk(KERN_INFO, shost, 1702 "%s: BRST failed chan: %d\n", 1703 current->comm, channel)); 1704 } 1705 } 1706 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1707 } 1708 1709 /** 1710 * scsi_eh_host_reset - send a host reset 1711 * @shost: host to be reset. 1712 * @work_q: &list_head for pending commands. 1713 * @done_q: &list_head for processed commands. 1714 */ 1715 static int scsi_eh_host_reset(struct Scsi_Host *shost, 1716 struct list_head *work_q, 1717 struct list_head *done_q) 1718 { 1719 struct scsi_cmnd *scmd, *next; 1720 LIST_HEAD(check_list); 1721 enum scsi_disposition rtn; 1722 1723 if (!list_empty(work_q)) { 1724 scmd = list_entry(work_q->next, 1725 struct scsi_cmnd, eh_entry); 1726 1727 SCSI_LOG_ERROR_RECOVERY(3, 1728 shost_printk(KERN_INFO, shost, 1729 "%s: Sending HRST\n", 1730 current->comm)); 1731 1732 rtn = scsi_try_host_reset(scmd); 1733 if (rtn == SUCCESS) { 1734 list_splice_init(work_q, &check_list); 1735 } else if (rtn == FAST_IO_FAIL) { 1736 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1737 scsi_eh_finish_cmd(scmd, done_q); 1738 } 1739 } else { 1740 SCSI_LOG_ERROR_RECOVERY(3, 1741 shost_printk(KERN_INFO, shost, 1742 "%s: HRST failed\n", 1743 current->comm)); 1744 } 1745 } 1746 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1747 } 1748 1749 /** 1750 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1751 * @work_q: &list_head for pending commands. 1752 * @done_q: &list_head for processed commands. 1753 */ 1754 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1755 struct list_head *done_q) 1756 { 1757 struct scsi_cmnd *scmd, *next; 1758 struct scsi_device *sdev; 1759 1760 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1761 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1762 "not ready after error recovery\n"); 1763 sdev = scmd->device; 1764 1765 mutex_lock(&sdev->state_mutex); 1766 scsi_device_set_state(sdev, SDEV_OFFLINE); 1767 mutex_unlock(&sdev->state_mutex); 1768 1769 scsi_eh_finish_cmd(scmd, done_q); 1770 } 1771 return; 1772 } 1773 1774 /** 1775 * scsi_noretry_cmd - determine if command should be failed fast 1776 * @scmd: SCSI cmd to examine. 1777 */ 1778 int scsi_noretry_cmd(struct scsi_cmnd *scmd) 1779 { 1780 struct request *req = scsi_cmd_to_rq(scmd); 1781 1782 switch (host_byte(scmd->result)) { 1783 case DID_OK: 1784 break; 1785 case DID_TIME_OUT: 1786 goto check_type; 1787 case DID_BUS_BUSY: 1788 return req->cmd_flags & REQ_FAILFAST_TRANSPORT; 1789 case DID_PARITY: 1790 return req->cmd_flags & REQ_FAILFAST_DEV; 1791 case DID_ERROR: 1792 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT) 1793 return 0; 1794 fallthrough; 1795 case DID_SOFT_ERROR: 1796 return req->cmd_flags & REQ_FAILFAST_DRIVER; 1797 } 1798 1799 if (!scsi_status_is_check_condition(scmd->result)) 1800 return 0; 1801 1802 check_type: 1803 /* 1804 * assume caller has checked sense and determined 1805 * the check condition was retryable. 1806 */ 1807 if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(req)) 1808 return 1; 1809 1810 return 0; 1811 } 1812 1813 /** 1814 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1815 * @scmd: SCSI cmd to examine. 1816 * 1817 * Notes: 1818 * This is *only* called when we are examining the status after sending 1819 * out the actual data command. any commands that are queued for error 1820 * recovery (e.g. test_unit_ready) do *not* come through here. 1821 * 1822 * When this routine returns failed, it means the error handler thread 1823 * is woken. In cases where the error code indicates an error that 1824 * doesn't require the error handler read (i.e. we don't need to 1825 * abort/reset), this function should return SUCCESS. 1826 */ 1827 enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd) 1828 { 1829 enum scsi_disposition rtn; 1830 1831 /* 1832 * if the device is offline, then we clearly just pass the result back 1833 * up to the top level. 1834 */ 1835 if (!scsi_device_online(scmd->device)) { 1836 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1837 "%s: device offline - report as SUCCESS\n", __func__)); 1838 return SUCCESS; 1839 } 1840 1841 /* 1842 * first check the host byte, to see if there is anything in there 1843 * that would indicate what we need to do. 1844 */ 1845 switch (host_byte(scmd->result)) { 1846 case DID_PASSTHROUGH: 1847 /* 1848 * no matter what, pass this through to the upper layer. 1849 * nuke this special code so that it looks like we are saying 1850 * did_ok. 1851 */ 1852 scmd->result &= 0xff00ffff; 1853 return SUCCESS; 1854 case DID_OK: 1855 /* 1856 * looks good. drop through, and check the next byte. 1857 */ 1858 break; 1859 case DID_ABORT: 1860 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1861 set_host_byte(scmd, DID_TIME_OUT); 1862 return SUCCESS; 1863 } 1864 fallthrough; 1865 case DID_NO_CONNECT: 1866 case DID_BAD_TARGET: 1867 /* 1868 * note - this means that we just report the status back 1869 * to the top level driver, not that we actually think 1870 * that it indicates SUCCESS. 1871 */ 1872 return SUCCESS; 1873 case DID_SOFT_ERROR: 1874 /* 1875 * when the low level driver returns did_soft_error, 1876 * it is responsible for keeping an internal retry counter 1877 * in order to avoid endless loops (db) 1878 */ 1879 goto maybe_retry; 1880 case DID_IMM_RETRY: 1881 return NEEDS_RETRY; 1882 1883 case DID_REQUEUE: 1884 return ADD_TO_MLQUEUE; 1885 case DID_TRANSPORT_DISRUPTED: 1886 /* 1887 * LLD/transport was disrupted during processing of the IO. 1888 * The transport class is now blocked/blocking, 1889 * and the transport will decide what to do with the IO 1890 * based on its timers and recovery capablilities if 1891 * there are enough retries. 1892 */ 1893 goto maybe_retry; 1894 case DID_TRANSPORT_FAILFAST: 1895 /* 1896 * The transport decided to failfast the IO (most likely 1897 * the fast io fail tmo fired), so send IO directly upwards. 1898 */ 1899 return SUCCESS; 1900 case DID_TRANSPORT_MARGINAL: 1901 /* 1902 * caller has decided not to do retries on 1903 * abort success, so send IO directly upwards 1904 */ 1905 return SUCCESS; 1906 case DID_ERROR: 1907 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT) 1908 /* 1909 * execute reservation conflict processing code 1910 * lower down 1911 */ 1912 break; 1913 fallthrough; 1914 case DID_BUS_BUSY: 1915 case DID_PARITY: 1916 goto maybe_retry; 1917 case DID_TIME_OUT: 1918 /* 1919 * when we scan the bus, we get timeout messages for 1920 * these commands if there is no device available. 1921 * other hosts report did_no_connect for the same thing. 1922 */ 1923 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1924 scmd->cmnd[0] == INQUIRY)) { 1925 return SUCCESS; 1926 } else { 1927 return FAILED; 1928 } 1929 case DID_RESET: 1930 return SUCCESS; 1931 default: 1932 return FAILED; 1933 } 1934 1935 /* 1936 * check the status byte to see if this indicates anything special. 1937 */ 1938 switch (get_status_byte(scmd)) { 1939 case SAM_STAT_TASK_SET_FULL: 1940 scsi_handle_queue_full(scmd->device); 1941 /* 1942 * the case of trying to send too many commands to a 1943 * tagged queueing device. 1944 */ 1945 fallthrough; 1946 case SAM_STAT_BUSY: 1947 /* 1948 * device can't talk to us at the moment. Should only 1949 * occur (SAM-3) when the task queue is empty, so will cause 1950 * the empty queue handling to trigger a stall in the 1951 * device. 1952 */ 1953 return ADD_TO_MLQUEUE; 1954 case SAM_STAT_GOOD: 1955 if (scmd->cmnd[0] == REPORT_LUNS) 1956 scmd->device->sdev_target->expecting_lun_change = 0; 1957 scsi_handle_queue_ramp_up(scmd->device); 1958 fallthrough; 1959 case SAM_STAT_COMMAND_TERMINATED: 1960 return SUCCESS; 1961 case SAM_STAT_TASK_ABORTED: 1962 goto maybe_retry; 1963 case SAM_STAT_CHECK_CONDITION: 1964 rtn = scsi_check_sense(scmd); 1965 if (rtn == NEEDS_RETRY) 1966 goto maybe_retry; 1967 /* if rtn == FAILED, we have no sense information; 1968 * returning FAILED will wake the error handler thread 1969 * to collect the sense and redo the decide 1970 * disposition */ 1971 return rtn; 1972 case SAM_STAT_CONDITION_MET: 1973 case SAM_STAT_INTERMEDIATE: 1974 case SAM_STAT_INTERMEDIATE_CONDITION_MET: 1975 case SAM_STAT_ACA_ACTIVE: 1976 /* 1977 * who knows? FIXME(eric) 1978 */ 1979 return SUCCESS; 1980 1981 case SAM_STAT_RESERVATION_CONFLICT: 1982 sdev_printk(KERN_INFO, scmd->device, 1983 "reservation conflict\n"); 1984 set_host_byte(scmd, DID_NEXUS_FAILURE); 1985 return SUCCESS; /* causes immediate i/o error */ 1986 default: 1987 return FAILED; 1988 } 1989 return FAILED; 1990 1991 maybe_retry: 1992 1993 /* we requeue for retry because the error was retryable, and 1994 * the request was not marked fast fail. Note that above, 1995 * even if the request is marked fast fail, we still requeue 1996 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1997 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) { 1998 return NEEDS_RETRY; 1999 } else { 2000 /* 2001 * no more retries - report this one back to upper level. 2002 */ 2003 return SUCCESS; 2004 } 2005 } 2006 2007 static void eh_lock_door_done(struct request *req, blk_status_t status) 2008 { 2009 blk_mq_free_request(req); 2010 } 2011 2012 /** 2013 * scsi_eh_lock_door - Prevent medium removal for the specified device 2014 * @sdev: SCSI device to prevent medium removal 2015 * 2016 * Locking: 2017 * We must be called from process context. 2018 * 2019 * Notes: 2020 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 2021 * head of the devices request queue, and continue. 2022 */ 2023 static void scsi_eh_lock_door(struct scsi_device *sdev) 2024 { 2025 struct request *req; 2026 struct scsi_request *rq; 2027 2028 req = scsi_alloc_request(sdev->request_queue, REQ_OP_DRV_IN, 0); 2029 if (IS_ERR(req)) 2030 return; 2031 rq = scsi_req(req); 2032 2033 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL; 2034 rq->cmd[1] = 0; 2035 rq->cmd[2] = 0; 2036 rq->cmd[3] = 0; 2037 rq->cmd[4] = SCSI_REMOVAL_PREVENT; 2038 rq->cmd[5] = 0; 2039 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]); 2040 2041 req->rq_flags |= RQF_QUIET; 2042 req->timeout = 10 * HZ; 2043 rq->retries = 5; 2044 2045 blk_execute_rq_nowait(req, true, eh_lock_door_done); 2046 } 2047 2048 /** 2049 * scsi_restart_operations - restart io operations to the specified host. 2050 * @shost: Host we are restarting. 2051 * 2052 * Notes: 2053 * When we entered the error handler, we blocked all further i/o to 2054 * this device. we need to 'reverse' this process. 2055 */ 2056 static void scsi_restart_operations(struct Scsi_Host *shost) 2057 { 2058 struct scsi_device *sdev; 2059 unsigned long flags; 2060 2061 /* 2062 * If the door was locked, we need to insert a door lock request 2063 * onto the head of the SCSI request queue for the device. There 2064 * is no point trying to lock the door of an off-line device. 2065 */ 2066 shost_for_each_device(sdev, shost) { 2067 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 2068 scsi_eh_lock_door(sdev); 2069 sdev->was_reset = 0; 2070 } 2071 } 2072 2073 /* 2074 * next free up anything directly waiting upon the host. this 2075 * will be requests for character device operations, and also for 2076 * ioctls to queued block devices. 2077 */ 2078 SCSI_LOG_ERROR_RECOVERY(3, 2079 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 2080 2081 spin_lock_irqsave(shost->host_lock, flags); 2082 if (scsi_host_set_state(shost, SHOST_RUNNING)) 2083 if (scsi_host_set_state(shost, SHOST_CANCEL)) 2084 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 2085 spin_unlock_irqrestore(shost->host_lock, flags); 2086 2087 wake_up(&shost->host_wait); 2088 2089 /* 2090 * finally we need to re-initiate requests that may be pending. we will 2091 * have had everything blocked while error handling is taking place, and 2092 * now that error recovery is done, we will need to ensure that these 2093 * requests are started. 2094 */ 2095 scsi_run_host_queues(shost); 2096 2097 /* 2098 * if eh is active and host_eh_scheduled is pending we need to re-run 2099 * recovery. we do this check after scsi_run_host_queues() to allow 2100 * everything pent up since the last eh run a chance to make forward 2101 * progress before we sync again. Either we'll immediately re-run 2102 * recovery or scsi_device_unbusy() will wake us again when these 2103 * pending commands complete. 2104 */ 2105 spin_lock_irqsave(shost->host_lock, flags); 2106 if (shost->host_eh_scheduled) 2107 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 2108 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 2109 spin_unlock_irqrestore(shost->host_lock, flags); 2110 } 2111 2112 /** 2113 * scsi_eh_ready_devs - check device ready state and recover if not. 2114 * @shost: host to be recovered. 2115 * @work_q: &list_head for pending commands. 2116 * @done_q: &list_head for processed commands. 2117 */ 2118 void scsi_eh_ready_devs(struct Scsi_Host *shost, 2119 struct list_head *work_q, 2120 struct list_head *done_q) 2121 { 2122 if (!scsi_eh_stu(shost, work_q, done_q)) 2123 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2124 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2125 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2126 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2127 scsi_eh_offline_sdevs(work_q, 2128 done_q); 2129 } 2130 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2131 2132 /** 2133 * scsi_eh_flush_done_q - finish processed commands or retry them. 2134 * @done_q: list_head of processed commands. 2135 */ 2136 void scsi_eh_flush_done_q(struct list_head *done_q) 2137 { 2138 struct scsi_cmnd *scmd, *next; 2139 2140 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2141 list_del_init(&scmd->eh_entry); 2142 if (scsi_device_online(scmd->device) && 2143 !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd) && 2144 scsi_eh_should_retry_cmd(scmd)) { 2145 SCSI_LOG_ERROR_RECOVERY(3, 2146 scmd_printk(KERN_INFO, scmd, 2147 "%s: flush retry cmd\n", 2148 current->comm)); 2149 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2150 } else { 2151 /* 2152 * If just we got sense for the device (called 2153 * scsi_eh_get_sense), scmd->result is already 2154 * set, do not set DID_TIME_OUT. 2155 */ 2156 if (!scmd->result) 2157 scmd->result |= (DID_TIME_OUT << 16); 2158 SCSI_LOG_ERROR_RECOVERY(3, 2159 scmd_printk(KERN_INFO, scmd, 2160 "%s: flush finish cmd\n", 2161 current->comm)); 2162 scsi_finish_command(scmd); 2163 } 2164 } 2165 } 2166 EXPORT_SYMBOL(scsi_eh_flush_done_q); 2167 2168 /** 2169 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2170 * @shost: Host to unjam. 2171 * 2172 * Notes: 2173 * When we come in here, we *know* that all commands on the bus have 2174 * either completed, failed or timed out. we also know that no further 2175 * commands are being sent to the host, so things are relatively quiet 2176 * and we have freedom to fiddle with things as we wish. 2177 * 2178 * This is only the *default* implementation. it is possible for 2179 * individual drivers to supply their own version of this function, and 2180 * if the maintainer wishes to do this, it is strongly suggested that 2181 * this function be taken as a template and modified. this function 2182 * was designed to correctly handle problems for about 95% of the 2183 * different cases out there, and it should always provide at least a 2184 * reasonable amount of error recovery. 2185 * 2186 * Any command marked 'failed' or 'timeout' must eventually have 2187 * scsi_finish_cmd() called for it. we do all of the retry stuff 2188 * here, so when we restart the host after we return it should have an 2189 * empty queue. 2190 */ 2191 static void scsi_unjam_host(struct Scsi_Host *shost) 2192 { 2193 unsigned long flags; 2194 LIST_HEAD(eh_work_q); 2195 LIST_HEAD(eh_done_q); 2196 2197 spin_lock_irqsave(shost->host_lock, flags); 2198 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2199 spin_unlock_irqrestore(shost->host_lock, flags); 2200 2201 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2202 2203 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2204 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2205 2206 spin_lock_irqsave(shost->host_lock, flags); 2207 if (shost->eh_deadline != -1) 2208 shost->last_reset = 0; 2209 spin_unlock_irqrestore(shost->host_lock, flags); 2210 scsi_eh_flush_done_q(&eh_done_q); 2211 } 2212 2213 /** 2214 * scsi_error_handler - SCSI error handler thread 2215 * @data: Host for which we are running. 2216 * 2217 * Notes: 2218 * This is the main error handling loop. This is run as a kernel thread 2219 * for every SCSI host and handles all error handling activity. 2220 */ 2221 int scsi_error_handler(void *data) 2222 { 2223 struct Scsi_Host *shost = data; 2224 2225 /* 2226 * We use TASK_INTERRUPTIBLE so that the thread is not 2227 * counted against the load average as a running process. 2228 * We never actually get interrupted because kthread_run 2229 * disables signal delivery for the created thread. 2230 */ 2231 while (true) { 2232 /* 2233 * The sequence in kthread_stop() sets the stop flag first 2234 * then wakes the process. To avoid missed wakeups, the task 2235 * should always be in a non running state before the stop 2236 * flag is checked 2237 */ 2238 set_current_state(TASK_INTERRUPTIBLE); 2239 if (kthread_should_stop()) 2240 break; 2241 2242 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2243 shost->host_failed != scsi_host_busy(shost)) { 2244 SCSI_LOG_ERROR_RECOVERY(1, 2245 shost_printk(KERN_INFO, shost, 2246 "scsi_eh_%d: sleeping\n", 2247 shost->host_no)); 2248 schedule(); 2249 continue; 2250 } 2251 2252 __set_current_state(TASK_RUNNING); 2253 SCSI_LOG_ERROR_RECOVERY(1, 2254 shost_printk(KERN_INFO, shost, 2255 "scsi_eh_%d: waking up %d/%d/%d\n", 2256 shost->host_no, shost->host_eh_scheduled, 2257 shost->host_failed, 2258 scsi_host_busy(shost))); 2259 2260 /* 2261 * We have a host that is failing for some reason. Figure out 2262 * what we need to do to get it up and online again (if we can). 2263 * If we fail, we end up taking the thing offline. 2264 */ 2265 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { 2266 SCSI_LOG_ERROR_RECOVERY(1, 2267 shost_printk(KERN_ERR, shost, 2268 "scsi_eh_%d: unable to autoresume\n", 2269 shost->host_no)); 2270 continue; 2271 } 2272 2273 if (shost->transportt->eh_strategy_handler) 2274 shost->transportt->eh_strategy_handler(shost); 2275 else 2276 scsi_unjam_host(shost); 2277 2278 /* All scmds have been handled */ 2279 shost->host_failed = 0; 2280 2281 /* 2282 * Note - if the above fails completely, the action is to take 2283 * individual devices offline and flush the queue of any 2284 * outstanding requests that may have been pending. When we 2285 * restart, we restart any I/O to any other devices on the bus 2286 * which are still online. 2287 */ 2288 scsi_restart_operations(shost); 2289 if (!shost->eh_noresume) 2290 scsi_autopm_put_host(shost); 2291 } 2292 __set_current_state(TASK_RUNNING); 2293 2294 SCSI_LOG_ERROR_RECOVERY(1, 2295 shost_printk(KERN_INFO, shost, 2296 "Error handler scsi_eh_%d exiting\n", 2297 shost->host_no)); 2298 shost->ehandler = NULL; 2299 return 0; 2300 } 2301 2302 /* 2303 * Function: scsi_report_bus_reset() 2304 * 2305 * Purpose: Utility function used by low-level drivers to report that 2306 * they have observed a bus reset on the bus being handled. 2307 * 2308 * Arguments: shost - Host in question 2309 * channel - channel on which reset was observed. 2310 * 2311 * Returns: Nothing 2312 * 2313 * Lock status: Host lock must be held. 2314 * 2315 * Notes: This only needs to be called if the reset is one which 2316 * originates from an unknown location. Resets originated 2317 * by the mid-level itself don't need to call this, but there 2318 * should be no harm. 2319 * 2320 * The main purpose of this is to make sure that a CHECK_CONDITION 2321 * is properly treated. 2322 */ 2323 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2324 { 2325 struct scsi_device *sdev; 2326 2327 __shost_for_each_device(sdev, shost) { 2328 if (channel == sdev_channel(sdev)) 2329 __scsi_report_device_reset(sdev, NULL); 2330 } 2331 } 2332 EXPORT_SYMBOL(scsi_report_bus_reset); 2333 2334 /* 2335 * Function: scsi_report_device_reset() 2336 * 2337 * Purpose: Utility function used by low-level drivers to report that 2338 * they have observed a device reset on the device being handled. 2339 * 2340 * Arguments: shost - Host in question 2341 * channel - channel on which reset was observed 2342 * target - target on which reset was observed 2343 * 2344 * Returns: Nothing 2345 * 2346 * Lock status: Host lock must be held 2347 * 2348 * Notes: This only needs to be called if the reset is one which 2349 * originates from an unknown location. Resets originated 2350 * by the mid-level itself don't need to call this, but there 2351 * should be no harm. 2352 * 2353 * The main purpose of this is to make sure that a CHECK_CONDITION 2354 * is properly treated. 2355 */ 2356 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2357 { 2358 struct scsi_device *sdev; 2359 2360 __shost_for_each_device(sdev, shost) { 2361 if (channel == sdev_channel(sdev) && 2362 target == sdev_id(sdev)) 2363 __scsi_report_device_reset(sdev, NULL); 2364 } 2365 } 2366 EXPORT_SYMBOL(scsi_report_device_reset); 2367 2368 /** 2369 * scsi_ioctl_reset: explicitly reset a host/bus/target/device 2370 * @dev: scsi_device to operate on 2371 * @arg: reset type (see sg.h) 2372 */ 2373 int 2374 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) 2375 { 2376 struct scsi_cmnd *scmd; 2377 struct Scsi_Host *shost = dev->host; 2378 struct request *rq; 2379 unsigned long flags; 2380 int error = 0, val; 2381 enum scsi_disposition rtn; 2382 2383 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) 2384 return -EACCES; 2385 2386 error = get_user(val, arg); 2387 if (error) 2388 return error; 2389 2390 if (scsi_autopm_get_host(shost) < 0) 2391 return -EIO; 2392 2393 error = -EIO; 2394 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) + 2395 shost->hostt->cmd_size, GFP_KERNEL); 2396 if (!rq) 2397 goto out_put_autopm_host; 2398 blk_rq_init(NULL, rq); 2399 2400 scmd = (struct scsi_cmnd *)(rq + 1); 2401 scsi_init_command(dev, scmd); 2402 scmd->cmnd = scsi_req(rq)->cmd; 2403 2404 scmd->submitter = SUBMITTED_BY_SCSI_RESET_IOCTL; 2405 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 2406 2407 scmd->cmd_len = 0; 2408 2409 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 2410 2411 spin_lock_irqsave(shost->host_lock, flags); 2412 shost->tmf_in_progress = 1; 2413 spin_unlock_irqrestore(shost->host_lock, flags); 2414 2415 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { 2416 case SG_SCSI_RESET_NOTHING: 2417 rtn = SUCCESS; 2418 break; 2419 case SG_SCSI_RESET_DEVICE: 2420 rtn = scsi_try_bus_device_reset(scmd); 2421 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2422 break; 2423 fallthrough; 2424 case SG_SCSI_RESET_TARGET: 2425 rtn = scsi_try_target_reset(scmd); 2426 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2427 break; 2428 fallthrough; 2429 case SG_SCSI_RESET_BUS: 2430 rtn = scsi_try_bus_reset(scmd); 2431 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2432 break; 2433 fallthrough; 2434 case SG_SCSI_RESET_HOST: 2435 rtn = scsi_try_host_reset(scmd); 2436 if (rtn == SUCCESS) 2437 break; 2438 fallthrough; 2439 default: 2440 rtn = FAILED; 2441 break; 2442 } 2443 2444 error = (rtn == SUCCESS) ? 0 : -EIO; 2445 2446 spin_lock_irqsave(shost->host_lock, flags); 2447 shost->tmf_in_progress = 0; 2448 spin_unlock_irqrestore(shost->host_lock, flags); 2449 2450 /* 2451 * be sure to wake up anyone who was sleeping or had their queue 2452 * suspended while we performed the TMF. 2453 */ 2454 SCSI_LOG_ERROR_RECOVERY(3, 2455 shost_printk(KERN_INFO, shost, 2456 "waking up host to restart after TMF\n")); 2457 2458 wake_up(&shost->host_wait); 2459 scsi_run_host_queues(shost); 2460 2461 kfree(rq); 2462 2463 out_put_autopm_host: 2464 scsi_autopm_put_host(shost); 2465 return error; 2466 } 2467 2468 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, 2469 struct scsi_sense_hdr *sshdr) 2470 { 2471 return scsi_normalize_sense(cmd->sense_buffer, 2472 SCSI_SENSE_BUFFERSIZE, sshdr); 2473 } 2474 EXPORT_SYMBOL(scsi_command_normalize_sense); 2475 2476 /** 2477 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2478 * @sense_buffer: byte array of sense data 2479 * @sb_len: number of valid bytes in sense_buffer 2480 * @info_out: pointer to 64 integer where 8 or 4 byte information 2481 * field will be placed if found. 2482 * 2483 * Return value: 2484 * true if information field found, false if not found. 2485 */ 2486 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len, 2487 u64 *info_out) 2488 { 2489 const u8 * ucp; 2490 2491 if (sb_len < 7) 2492 return false; 2493 switch (sense_buffer[0] & 0x7f) { 2494 case 0x70: 2495 case 0x71: 2496 if (sense_buffer[0] & 0x80) { 2497 *info_out = get_unaligned_be32(&sense_buffer[3]); 2498 return true; 2499 } 2500 return false; 2501 case 0x72: 2502 case 0x73: 2503 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2504 0 /* info desc */); 2505 if (ucp && (0xa == ucp[1])) { 2506 *info_out = get_unaligned_be64(&ucp[4]); 2507 return true; 2508 } 2509 return false; 2510 default: 2511 return false; 2512 } 2513 } 2514 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2515