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 static void scsi_eh_done(struct scsi_cmnd *scmd); 54 55 /* 56 * These should *probably* be handled by the host itself. 57 * Since it is allowed to sleep, it probably should. 58 */ 59 #define BUS_RESET_SETTLE_TIME (10) 60 #define HOST_RESET_SETTLE_TIME (10) 61 62 static int scsi_eh_try_stu(struct scsi_cmnd *scmd); 63 static enum scsi_disposition scsi_try_to_abort_cmd(struct scsi_host_template *, 64 struct scsi_cmnd *); 65 66 void scsi_eh_wakeup(struct Scsi_Host *shost) 67 { 68 lockdep_assert_held(shost->host_lock); 69 70 if (scsi_host_busy(shost) == shost->host_failed) { 71 trace_scsi_eh_wakeup(shost); 72 wake_up_process(shost->ehandler); 73 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost, 74 "Waking error handler thread\n")); 75 } 76 } 77 78 /** 79 * scsi_schedule_eh - schedule EH for SCSI host 80 * @shost: SCSI host to invoke error handling on. 81 * 82 * Schedule SCSI EH without scmd. 83 */ 84 void scsi_schedule_eh(struct Scsi_Host *shost) 85 { 86 unsigned long flags; 87 88 spin_lock_irqsave(shost->host_lock, flags); 89 90 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || 91 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { 92 shost->host_eh_scheduled++; 93 scsi_eh_wakeup(shost); 94 } 95 96 spin_unlock_irqrestore(shost->host_lock, flags); 97 } 98 EXPORT_SYMBOL_GPL(scsi_schedule_eh); 99 100 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost) 101 { 102 if (!shost->last_reset || shost->eh_deadline == -1) 103 return 0; 104 105 /* 106 * 32bit accesses are guaranteed to be atomic 107 * (on all supported architectures), so instead 108 * of using a spinlock we can as well double check 109 * if eh_deadline has been set to 'off' during the 110 * time_before call. 111 */ 112 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) && 113 shost->eh_deadline > -1) 114 return 0; 115 116 return 1; 117 } 118 119 static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd) 120 { 121 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) 122 return true; 123 124 return ++cmd->retries <= cmd->allowed; 125 } 126 127 static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd) 128 { 129 struct scsi_device *sdev = cmd->device; 130 struct Scsi_Host *host = sdev->host; 131 132 if (host->hostt->eh_should_retry_cmd) 133 return host->hostt->eh_should_retry_cmd(cmd); 134 135 return true; 136 } 137 138 /** 139 * scmd_eh_abort_handler - Handle command aborts 140 * @work: command to be aborted. 141 * 142 * Note: this function must be called only for a command that has timed out. 143 * Because the block layer marks a request as complete before it calls 144 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has 145 * timed out do not have any effect. Hence it is safe to call 146 * scsi_finish_command() from this function. 147 */ 148 void 149 scmd_eh_abort_handler(struct work_struct *work) 150 { 151 struct scsi_cmnd *scmd = 152 container_of(work, struct scsi_cmnd, abort_work.work); 153 struct scsi_device *sdev = scmd->device; 154 enum scsi_disposition rtn; 155 156 if (scsi_host_eh_past_deadline(sdev->host)) { 157 SCSI_LOG_ERROR_RECOVERY(3, 158 scmd_printk(KERN_INFO, scmd, 159 "eh timeout, not aborting\n")); 160 } else { 161 SCSI_LOG_ERROR_RECOVERY(3, 162 scmd_printk(KERN_INFO, scmd, 163 "aborting command\n")); 164 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd); 165 if (rtn == SUCCESS) { 166 set_host_byte(scmd, DID_TIME_OUT); 167 if (scsi_host_eh_past_deadline(sdev->host)) { 168 SCSI_LOG_ERROR_RECOVERY(3, 169 scmd_printk(KERN_INFO, scmd, 170 "eh timeout, not retrying " 171 "aborted command\n")); 172 } else if (!scsi_noretry_cmd(scmd) && 173 scsi_cmd_retry_allowed(scmd) && 174 scsi_eh_should_retry_cmd(scmd)) { 175 SCSI_LOG_ERROR_RECOVERY(3, 176 scmd_printk(KERN_WARNING, scmd, 177 "retry aborted command\n")); 178 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 179 return; 180 } else { 181 SCSI_LOG_ERROR_RECOVERY(3, 182 scmd_printk(KERN_WARNING, scmd, 183 "finish aborted command\n")); 184 scsi_finish_command(scmd); 185 return; 186 } 187 } else { 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 } 194 } 195 196 scsi_eh_scmd_add(scmd); 197 } 198 199 /** 200 * scsi_abort_command - schedule a command abort 201 * @scmd: scmd to abort. 202 * 203 * We only need to abort commands after a command timeout 204 */ 205 static int 206 scsi_abort_command(struct scsi_cmnd *scmd) 207 { 208 struct scsi_device *sdev = scmd->device; 209 struct Scsi_Host *shost = sdev->host; 210 unsigned long flags; 211 212 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 213 /* 214 * Retry after abort failed, escalate to next level. 215 */ 216 SCSI_LOG_ERROR_RECOVERY(3, 217 scmd_printk(KERN_INFO, scmd, 218 "previous abort failed\n")); 219 BUG_ON(delayed_work_pending(&scmd->abort_work)); 220 return FAILED; 221 } 222 223 spin_lock_irqsave(shost->host_lock, flags); 224 if (shost->eh_deadline != -1 && !shost->last_reset) 225 shost->last_reset = jiffies; 226 spin_unlock_irqrestore(shost->host_lock, flags); 227 228 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; 229 SCSI_LOG_ERROR_RECOVERY(3, 230 scmd_printk(KERN_INFO, scmd, "abort scheduled\n")); 231 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100); 232 return SUCCESS; 233 } 234 235 /** 236 * scsi_eh_reset - call into ->eh_action to reset internal counters 237 * @scmd: scmd to run eh on. 238 * 239 * The scsi driver might be carrying internal state about the 240 * devices, so we need to call into the driver to reset the 241 * internal state once the error handler is started. 242 */ 243 static void scsi_eh_reset(struct scsi_cmnd *scmd) 244 { 245 if (!blk_rq_is_passthrough(scmd->request)) { 246 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 247 if (sdrv->eh_reset) 248 sdrv->eh_reset(scmd); 249 } 250 } 251 252 static void scsi_eh_inc_host_failed(struct rcu_head *head) 253 { 254 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu); 255 struct Scsi_Host *shost = scmd->device->host; 256 unsigned long flags; 257 258 spin_lock_irqsave(shost->host_lock, flags); 259 shost->host_failed++; 260 scsi_eh_wakeup(shost); 261 spin_unlock_irqrestore(shost->host_lock, flags); 262 } 263 264 /** 265 * scsi_eh_scmd_add - add scsi cmd to error handling. 266 * @scmd: scmd to run eh on. 267 */ 268 void scsi_eh_scmd_add(struct scsi_cmnd *scmd) 269 { 270 struct Scsi_Host *shost = scmd->device->host; 271 unsigned long flags; 272 int ret; 273 274 WARN_ON_ONCE(!shost->ehandler); 275 276 spin_lock_irqsave(shost->host_lock, flags); 277 if (scsi_host_set_state(shost, SHOST_RECOVERY)) { 278 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY); 279 WARN_ON_ONCE(ret); 280 } 281 if (shost->eh_deadline != -1 && !shost->last_reset) 282 shost->last_reset = jiffies; 283 284 scsi_eh_reset(scmd); 285 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 286 spin_unlock_irqrestore(shost->host_lock, flags); 287 /* 288 * Ensure that all tasks observe the host state change before the 289 * host_failed change. 290 */ 291 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed); 292 } 293 294 /** 295 * scsi_times_out - Timeout function for normal scsi commands. 296 * @req: request that is timing out. 297 * 298 * Notes: 299 * We do not need to lock this. There is the potential for a race 300 * only in that the normal completion handling might run, but if the 301 * normal completion function determines that the timer has already 302 * fired, then it mustn't do anything. 303 */ 304 enum blk_eh_timer_return scsi_times_out(struct request *req) 305 { 306 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req); 307 enum blk_eh_timer_return rtn = BLK_EH_DONE; 308 struct Scsi_Host *host = scmd->device->host; 309 310 trace_scsi_dispatch_cmd_timeout(scmd); 311 scsi_log_completion(scmd, TIMEOUT_ERROR); 312 313 if (host->eh_deadline != -1 && !host->last_reset) 314 host->last_reset = jiffies; 315 316 if (host->hostt->eh_timed_out) 317 rtn = host->hostt->eh_timed_out(scmd); 318 319 if (rtn == BLK_EH_DONE) { 320 /* 321 * Set the command to complete first in order to prevent a real 322 * completion from releasing the command while error handling 323 * is using it. If the command was already completed, then the 324 * lower level driver beat the timeout handler, and it is safe 325 * to return without escalating error recovery. 326 * 327 * If timeout handling lost the race to a real completion, the 328 * block layer may ignore that due to a fake timeout injection, 329 * so return RESET_TIMER to allow error handling another shot 330 * at this command. 331 */ 332 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state)) 333 return BLK_EH_RESET_TIMER; 334 if (scsi_abort_command(scmd) != SUCCESS) { 335 set_host_byte(scmd, DID_TIME_OUT); 336 scsi_eh_scmd_add(scmd); 337 } 338 } 339 340 return rtn; 341 } 342 343 /** 344 * scsi_block_when_processing_errors - Prevent cmds from being queued. 345 * @sdev: Device on which we are performing recovery. 346 * 347 * Description: 348 * We block until the host is out of error recovery, and then check to 349 * see whether the host or the device is offline. 350 * 351 * Return value: 352 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 353 */ 354 int scsi_block_when_processing_errors(struct scsi_device *sdev) 355 { 356 int online; 357 358 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 359 360 online = scsi_device_online(sdev); 361 362 return online; 363 } 364 EXPORT_SYMBOL(scsi_block_when_processing_errors); 365 366 #ifdef CONFIG_SCSI_LOGGING 367 /** 368 * scsi_eh_prt_fail_stats - Log info on failures. 369 * @shost: scsi host being recovered. 370 * @work_q: Queue of scsi cmds to process. 371 */ 372 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 373 struct list_head *work_q) 374 { 375 struct scsi_cmnd *scmd; 376 struct scsi_device *sdev; 377 int total_failures = 0; 378 int cmd_failed = 0; 379 int cmd_cancel = 0; 380 int devices_failed = 0; 381 382 shost_for_each_device(sdev, shost) { 383 list_for_each_entry(scmd, work_q, eh_entry) { 384 if (scmd->device == sdev) { 385 ++total_failures; 386 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) 387 ++cmd_cancel; 388 else 389 ++cmd_failed; 390 } 391 } 392 393 if (cmd_cancel || cmd_failed) { 394 SCSI_LOG_ERROR_RECOVERY(3, 395 shost_printk(KERN_INFO, shost, 396 "%s: cmds failed: %d, cancel: %d\n", 397 __func__, cmd_failed, 398 cmd_cancel)); 399 cmd_cancel = 0; 400 cmd_failed = 0; 401 ++devices_failed; 402 } 403 } 404 405 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, 406 "Total of %d commands on %d" 407 " devices require eh work\n", 408 total_failures, devices_failed)); 409 } 410 #endif 411 412 /** 413 * scsi_report_lun_change - Set flag on all *other* devices on the same target 414 * to indicate that a UNIT ATTENTION is expected. 415 * @sdev: Device reporting the UNIT ATTENTION 416 */ 417 static void scsi_report_lun_change(struct scsi_device *sdev) 418 { 419 sdev->sdev_target->expecting_lun_change = 1; 420 } 421 422 /** 423 * scsi_report_sense - Examine scsi sense information and log messages for 424 * certain conditions, also issue uevents for some of them. 425 * @sdev: Device reporting the sense code 426 * @sshdr: sshdr to be examined 427 */ 428 static void scsi_report_sense(struct scsi_device *sdev, 429 struct scsi_sense_hdr *sshdr) 430 { 431 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ 432 433 if (sshdr->sense_key == UNIT_ATTENTION) { 434 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { 435 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; 436 sdev_printk(KERN_WARNING, sdev, 437 "Inquiry data has changed"); 438 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { 439 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; 440 scsi_report_lun_change(sdev); 441 sdev_printk(KERN_WARNING, sdev, 442 "Warning! Received an indication that the " 443 "LUN assignments on this target have " 444 "changed. The Linux SCSI layer does not " 445 "automatically remap LUN assignments.\n"); 446 } else if (sshdr->asc == 0x3f) 447 sdev_printk(KERN_WARNING, sdev, 448 "Warning! Received an indication that the " 449 "operating parameters on this target have " 450 "changed. The Linux SCSI layer does not " 451 "automatically adjust these parameters.\n"); 452 453 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { 454 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; 455 sdev_printk(KERN_WARNING, sdev, 456 "Warning! Received an indication that the " 457 "LUN reached a thin provisioning soft " 458 "threshold.\n"); 459 } 460 461 if (sshdr->asc == 0x29) { 462 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED; 463 sdev_printk(KERN_WARNING, sdev, 464 "Power-on or device reset occurred\n"); 465 } 466 467 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { 468 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; 469 sdev_printk(KERN_WARNING, sdev, 470 "Mode parameters changed"); 471 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { 472 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; 473 sdev_printk(KERN_WARNING, sdev, 474 "Asymmetric access state changed"); 475 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { 476 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; 477 sdev_printk(KERN_WARNING, sdev, 478 "Capacity data has changed"); 479 } else if (sshdr->asc == 0x2a) 480 sdev_printk(KERN_WARNING, sdev, 481 "Parameters changed"); 482 } 483 484 if (evt_type != SDEV_EVT_MAXBITS) { 485 set_bit(evt_type, sdev->pending_events); 486 schedule_work(&sdev->event_work); 487 } 488 } 489 490 /** 491 * scsi_check_sense - Examine scsi cmd sense 492 * @scmd: Cmd to have sense checked. 493 * 494 * Return value: 495 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE 496 * 497 * Notes: 498 * When a deferred error is detected the current command has 499 * not been executed and needs retrying. 500 */ 501 enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd) 502 { 503 struct scsi_device *sdev = scmd->device; 504 struct scsi_sense_hdr sshdr; 505 506 if (! scsi_command_normalize_sense(scmd, &sshdr)) 507 return FAILED; /* no valid sense data */ 508 509 scsi_report_sense(sdev, &sshdr); 510 511 if (scsi_sense_is_deferred(&sshdr)) 512 return NEEDS_RETRY; 513 514 if (sdev->handler && sdev->handler->check_sense) { 515 enum scsi_disposition rc; 516 517 rc = sdev->handler->check_sense(sdev, &sshdr); 518 if (rc != SCSI_RETURN_NOT_HANDLED) 519 return rc; 520 /* handler does not care. Drop down to default handling */ 521 } 522 523 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done) 524 /* 525 * nasty: for mid-layer issued TURs, we need to return the 526 * actual sense data without any recovery attempt. For eh 527 * issued ones, we need to try to recover and interpret 528 */ 529 return SUCCESS; 530 531 /* 532 * Previous logic looked for FILEMARK, EOM or ILI which are 533 * mainly associated with tapes and returned SUCCESS. 534 */ 535 if (sshdr.response_code == 0x70) { 536 /* fixed format */ 537 if (scmd->sense_buffer[2] & 0xe0) 538 return SUCCESS; 539 } else { 540 /* 541 * descriptor format: look for "stream commands sense data 542 * descriptor" (see SSC-3). Assume single sense data 543 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 544 */ 545 if ((sshdr.additional_length > 3) && 546 (scmd->sense_buffer[8] == 0x4) && 547 (scmd->sense_buffer[11] & 0xe0)) 548 return SUCCESS; 549 } 550 551 switch (sshdr.sense_key) { 552 case NO_SENSE: 553 return SUCCESS; 554 case RECOVERED_ERROR: 555 return /* soft_error */ SUCCESS; 556 557 case ABORTED_COMMAND: 558 if (sshdr.asc == 0x10) /* DIF */ 559 return SUCCESS; 560 561 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF) 562 return ADD_TO_MLQUEUE; 563 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 && 564 sdev->sdev_bflags & BLIST_RETRY_ASC_C1) 565 return ADD_TO_MLQUEUE; 566 567 return NEEDS_RETRY; 568 case NOT_READY: 569 case UNIT_ATTENTION: 570 /* 571 * if we are expecting a cc/ua because of a bus reset that we 572 * performed, treat this just as a retry. otherwise this is 573 * information that we should pass up to the upper-level driver 574 * so that we can deal with it there. 575 */ 576 if (scmd->device->expecting_cc_ua) { 577 /* 578 * Because some device does not queue unit 579 * attentions correctly, we carefully check 580 * additional sense code and qualifier so as 581 * not to squash media change unit attention. 582 */ 583 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { 584 scmd->device->expecting_cc_ua = 0; 585 return NEEDS_RETRY; 586 } 587 } 588 /* 589 * we might also expect a cc/ua if another LUN on the target 590 * reported a UA with an ASC/ASCQ of 3F 0E - 591 * REPORTED LUNS DATA HAS CHANGED. 592 */ 593 if (scmd->device->sdev_target->expecting_lun_change && 594 sshdr.asc == 0x3f && sshdr.ascq == 0x0e) 595 return NEEDS_RETRY; 596 /* 597 * if the device is in the process of becoming ready, we 598 * should retry. 599 */ 600 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 601 return NEEDS_RETRY; 602 /* 603 * if the device is not started, we need to wake 604 * the error handler to start the motor 605 */ 606 if (scmd->device->allow_restart && 607 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 608 return FAILED; 609 /* 610 * Pass the UA upwards for a determination in the completion 611 * functions. 612 */ 613 return SUCCESS; 614 615 /* these are not supported */ 616 case DATA_PROTECT: 617 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { 618 /* Thin provisioning hard threshold reached */ 619 set_host_byte(scmd, DID_ALLOC_FAILURE); 620 return SUCCESS; 621 } 622 fallthrough; 623 case COPY_ABORTED: 624 case VOLUME_OVERFLOW: 625 case MISCOMPARE: 626 case BLANK_CHECK: 627 set_host_byte(scmd, DID_TARGET_FAILURE); 628 return SUCCESS; 629 630 case MEDIUM_ERROR: 631 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 632 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 633 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 634 set_host_byte(scmd, DID_MEDIUM_ERROR); 635 return SUCCESS; 636 } 637 return NEEDS_RETRY; 638 639 case HARDWARE_ERROR: 640 if (scmd->device->retry_hwerror) 641 return ADD_TO_MLQUEUE; 642 else 643 set_host_byte(scmd, DID_TARGET_FAILURE); 644 fallthrough; 645 646 case ILLEGAL_REQUEST: 647 if (sshdr.asc == 0x20 || /* Invalid command operation code */ 648 sshdr.asc == 0x21 || /* Logical block address out of range */ 649 sshdr.asc == 0x22 || /* Invalid function */ 650 sshdr.asc == 0x24 || /* Invalid field in cdb */ 651 sshdr.asc == 0x26 || /* Parameter value invalid */ 652 sshdr.asc == 0x27) { /* Write protected */ 653 set_host_byte(scmd, DID_TARGET_FAILURE); 654 } 655 return SUCCESS; 656 657 default: 658 return SUCCESS; 659 } 660 } 661 EXPORT_SYMBOL_GPL(scsi_check_sense); 662 663 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) 664 { 665 struct scsi_host_template *sht = sdev->host->hostt; 666 struct scsi_device *tmp_sdev; 667 668 if (!sht->track_queue_depth || 669 sdev->queue_depth >= sdev->max_queue_depth) 670 return; 671 672 if (time_before(jiffies, 673 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) 674 return; 675 676 if (time_before(jiffies, 677 sdev->last_queue_full_time + sdev->queue_ramp_up_period)) 678 return; 679 680 /* 681 * Walk all devices of a target and do 682 * ramp up on them. 683 */ 684 shost_for_each_device(tmp_sdev, sdev->host) { 685 if (tmp_sdev->channel != sdev->channel || 686 tmp_sdev->id != sdev->id || 687 tmp_sdev->queue_depth == sdev->max_queue_depth) 688 continue; 689 690 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); 691 sdev->last_queue_ramp_up = jiffies; 692 } 693 } 694 695 static void scsi_handle_queue_full(struct scsi_device *sdev) 696 { 697 struct scsi_host_template *sht = sdev->host->hostt; 698 struct scsi_device *tmp_sdev; 699 700 if (!sht->track_queue_depth) 701 return; 702 703 shost_for_each_device(tmp_sdev, sdev->host) { 704 if (tmp_sdev->channel != sdev->channel || 705 tmp_sdev->id != sdev->id) 706 continue; 707 /* 708 * We do not know the number of commands that were at 709 * the device when we got the queue full so we start 710 * from the highest possible value and work our way down. 711 */ 712 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); 713 } 714 } 715 716 /** 717 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 718 * @scmd: SCSI cmd to examine. 719 * 720 * Notes: 721 * This is *only* called when we are examining the status of commands 722 * queued during error recovery. the main difference here is that we 723 * don't allow for the possibility of retries here, and we are a lot 724 * more restrictive about what we consider acceptable. 725 */ 726 static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd) 727 { 728 /* 729 * first check the host byte, to see if there is anything in there 730 * that would indicate what we need to do. 731 */ 732 if (host_byte(scmd->result) == DID_RESET) { 733 /* 734 * rats. we are already in the error handler, so we now 735 * get to try and figure out what to do next. if the sense 736 * is valid, we have a pretty good idea of what to do. 737 * if not, we mark it as FAILED. 738 */ 739 return scsi_check_sense(scmd); 740 } 741 if (host_byte(scmd->result) != DID_OK) 742 return FAILED; 743 744 /* 745 * next, check the message byte. 746 */ 747 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 748 return FAILED; 749 750 /* 751 * now, check the status byte to see if this indicates 752 * anything special. 753 */ 754 switch (status_byte(scmd->result)) { 755 case GOOD: 756 scsi_handle_queue_ramp_up(scmd->device); 757 fallthrough; 758 case COMMAND_TERMINATED: 759 return SUCCESS; 760 case CHECK_CONDITION: 761 return scsi_check_sense(scmd); 762 case CONDITION_GOOD: 763 case INTERMEDIATE_GOOD: 764 case INTERMEDIATE_C_GOOD: 765 /* 766 * who knows? FIXME(eric) 767 */ 768 return SUCCESS; 769 case RESERVATION_CONFLICT: 770 if (scmd->cmnd[0] == TEST_UNIT_READY) 771 /* it is a success, we probed the device and 772 * found it */ 773 return SUCCESS; 774 /* otherwise, we failed to send the command */ 775 return FAILED; 776 case QUEUE_FULL: 777 scsi_handle_queue_full(scmd->device); 778 fallthrough; 779 case BUSY: 780 return NEEDS_RETRY; 781 default: 782 return FAILED; 783 } 784 return FAILED; 785 } 786 787 /** 788 * scsi_eh_done - Completion function for error handling. 789 * @scmd: Cmd that is done. 790 */ 791 static void scsi_eh_done(struct scsi_cmnd *scmd) 792 { 793 struct completion *eh_action; 794 795 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 796 "%s result: %x\n", __func__, scmd->result)); 797 798 eh_action = scmd->device->host->eh_action; 799 if (eh_action) 800 complete(eh_action); 801 } 802 803 /** 804 * scsi_try_host_reset - ask host adapter to reset itself 805 * @scmd: SCSI cmd to send host reset. 806 */ 807 static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd) 808 { 809 unsigned long flags; 810 enum scsi_disposition rtn; 811 struct Scsi_Host *host = scmd->device->host; 812 struct scsi_host_template *hostt = host->hostt; 813 814 SCSI_LOG_ERROR_RECOVERY(3, 815 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 816 817 if (!hostt->eh_host_reset_handler) 818 return FAILED; 819 820 rtn = hostt->eh_host_reset_handler(scmd); 821 822 if (rtn == SUCCESS) { 823 if (!hostt->skip_settle_delay) 824 ssleep(HOST_RESET_SETTLE_TIME); 825 spin_lock_irqsave(host->host_lock, flags); 826 scsi_report_bus_reset(host, scmd_channel(scmd)); 827 spin_unlock_irqrestore(host->host_lock, flags); 828 } 829 830 return rtn; 831 } 832 833 /** 834 * scsi_try_bus_reset - ask host to perform a bus reset 835 * @scmd: SCSI cmd to send bus reset. 836 */ 837 static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd) 838 { 839 unsigned long flags; 840 enum scsi_disposition rtn; 841 struct Scsi_Host *host = scmd->device->host; 842 struct scsi_host_template *hostt = host->hostt; 843 844 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 845 "%s: Snd Bus RST\n", __func__)); 846 847 if (!hostt->eh_bus_reset_handler) 848 return FAILED; 849 850 rtn = hostt->eh_bus_reset_handler(scmd); 851 852 if (rtn == SUCCESS) { 853 if (!hostt->skip_settle_delay) 854 ssleep(BUS_RESET_SETTLE_TIME); 855 spin_lock_irqsave(host->host_lock, flags); 856 scsi_report_bus_reset(host, scmd_channel(scmd)); 857 spin_unlock_irqrestore(host->host_lock, flags); 858 } 859 860 return rtn; 861 } 862 863 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 864 { 865 sdev->was_reset = 1; 866 sdev->expecting_cc_ua = 1; 867 } 868 869 /** 870 * scsi_try_target_reset - Ask host to perform a target reset 871 * @scmd: SCSI cmd used to send a target reset 872 * 873 * Notes: 874 * There is no timeout for this operation. if this operation is 875 * unreliable for a given host, then the host itself needs to put a 876 * timer on it, and set the host back to a consistent state prior to 877 * returning. 878 */ 879 static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd) 880 { 881 unsigned long flags; 882 enum scsi_disposition rtn; 883 struct Scsi_Host *host = scmd->device->host; 884 struct scsi_host_template *hostt = host->hostt; 885 886 if (!hostt->eh_target_reset_handler) 887 return FAILED; 888 889 rtn = hostt->eh_target_reset_handler(scmd); 890 if (rtn == SUCCESS) { 891 spin_lock_irqsave(host->host_lock, flags); 892 __starget_for_each_device(scsi_target(scmd->device), NULL, 893 __scsi_report_device_reset); 894 spin_unlock_irqrestore(host->host_lock, flags); 895 } 896 897 return rtn; 898 } 899 900 /** 901 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 902 * @scmd: SCSI cmd used to send BDR 903 * 904 * Notes: 905 * There is no timeout for this operation. if this operation is 906 * unreliable for a given host, then the host itself needs to put a 907 * timer on it, and set the host back to a consistent state prior to 908 * returning. 909 */ 910 static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 911 { 912 enum scsi_disposition rtn; 913 struct scsi_host_template *hostt = scmd->device->host->hostt; 914 915 if (!hostt->eh_device_reset_handler) 916 return FAILED; 917 918 rtn = hostt->eh_device_reset_handler(scmd); 919 if (rtn == SUCCESS) 920 __scsi_report_device_reset(scmd->device, NULL); 921 return rtn; 922 } 923 924 /** 925 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command 926 * @hostt: SCSI driver host template 927 * @scmd: SCSI cmd used to send a target reset 928 * 929 * Return value: 930 * SUCCESS, FAILED, or FAST_IO_FAIL 931 * 932 * Notes: 933 * SUCCESS does not necessarily indicate that the command 934 * has been aborted; it only indicates that the LLDDs 935 * has cleared all references to that command. 936 * LLDDs should return FAILED only if an abort was required 937 * but could not be executed. LLDDs should return FAST_IO_FAIL 938 * if the device is temporarily unavailable (eg due to a 939 * link down on FibreChannel) 940 */ 941 static enum scsi_disposition 942 scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd) 943 { 944 if (!hostt->eh_abort_handler) 945 return FAILED; 946 947 return hostt->eh_abort_handler(scmd); 948 } 949 950 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 951 { 952 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 953 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 954 if (scsi_try_target_reset(scmd) != SUCCESS) 955 if (scsi_try_bus_reset(scmd) != SUCCESS) 956 scsi_try_host_reset(scmd); 957 } 958 959 /** 960 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 961 * @scmd: SCSI command structure to hijack 962 * @ses: structure to save restore information 963 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 964 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) 965 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 966 * 967 * This function is used to save a scsi command information before re-execution 968 * as part of the error recovery process. If @sense_bytes is 0 the command 969 * sent must be one that does not transfer any data. If @sense_bytes != 0 970 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 971 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 972 */ 973 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 974 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 975 { 976 struct scsi_device *sdev = scmd->device; 977 978 /* 979 * We need saved copies of a number of fields - this is because 980 * error handling may need to overwrite these with different values 981 * to run different commands, and once error handling is complete, 982 * we will need to restore these values prior to running the actual 983 * command. 984 */ 985 ses->cmd_len = scmd->cmd_len; 986 ses->cmnd = scmd->cmnd; 987 ses->data_direction = scmd->sc_data_direction; 988 ses->sdb = scmd->sdb; 989 ses->result = scmd->result; 990 ses->resid_len = scmd->req.resid_len; 991 ses->underflow = scmd->underflow; 992 ses->prot_op = scmd->prot_op; 993 ses->eh_eflags = scmd->eh_eflags; 994 995 scmd->prot_op = SCSI_PROT_NORMAL; 996 scmd->eh_eflags = 0; 997 scmd->cmnd = ses->eh_cmnd; 998 memset(scmd->cmnd, 0, BLK_MAX_CDB); 999 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 1000 scmd->result = 0; 1001 scmd->req.resid_len = 0; 1002 1003 if (sense_bytes) { 1004 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 1005 sense_bytes); 1006 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 1007 scmd->sdb.length); 1008 scmd->sdb.table.sgl = &ses->sense_sgl; 1009 scmd->sc_data_direction = DMA_FROM_DEVICE; 1010 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; 1011 scmd->cmnd[0] = REQUEST_SENSE; 1012 scmd->cmnd[4] = scmd->sdb.length; 1013 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1014 } else { 1015 scmd->sc_data_direction = DMA_NONE; 1016 if (cmnd) { 1017 BUG_ON(cmnd_size > BLK_MAX_CDB); 1018 memcpy(scmd->cmnd, cmnd, cmnd_size); 1019 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1020 } 1021 } 1022 1023 scmd->underflow = 0; 1024 1025 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 1026 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 1027 (sdev->lun << 5 & 0xe0); 1028 1029 /* 1030 * Zero the sense buffer. The scsi spec mandates that any 1031 * untransferred sense data should be interpreted as being zero. 1032 */ 1033 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 1034 } 1035 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 1036 1037 /** 1038 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 1039 * @scmd: SCSI command structure to restore 1040 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 1041 * 1042 * Undo any damage done by above scsi_eh_prep_cmnd(). 1043 */ 1044 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 1045 { 1046 /* 1047 * Restore original data 1048 */ 1049 scmd->cmd_len = ses->cmd_len; 1050 scmd->cmnd = ses->cmnd; 1051 scmd->sc_data_direction = ses->data_direction; 1052 scmd->sdb = ses->sdb; 1053 scmd->result = ses->result; 1054 scmd->req.resid_len = ses->resid_len; 1055 scmd->underflow = ses->underflow; 1056 scmd->prot_op = ses->prot_op; 1057 scmd->eh_eflags = ses->eh_eflags; 1058 } 1059 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 1060 1061 /** 1062 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 1063 * @scmd: SCSI command structure to hijack 1064 * @cmnd: CDB to send 1065 * @cmnd_size: size in bytes of @cmnd 1066 * @timeout: timeout for this request 1067 * @sense_bytes: size of sense data to copy or 0 1068 * 1069 * This function is used to send a scsi command down to a target device 1070 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 1071 * 1072 * Return value: 1073 * SUCCESS or FAILED or NEEDS_RETRY 1074 */ 1075 static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd, 1076 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes) 1077 { 1078 struct scsi_device *sdev = scmd->device; 1079 struct Scsi_Host *shost = sdev->host; 1080 DECLARE_COMPLETION_ONSTACK(done); 1081 unsigned long timeleft = timeout, delay; 1082 struct scsi_eh_save ses; 1083 const unsigned long stall_for = msecs_to_jiffies(100); 1084 int rtn; 1085 1086 retry: 1087 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1088 shost->eh_action = &done; 1089 1090 scsi_log_send(scmd); 1091 scmd->scsi_done = scsi_eh_done; 1092 1093 /* 1094 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can 1095 * change the SCSI device state after we have examined it and before 1096 * .queuecommand() is called. 1097 */ 1098 mutex_lock(&sdev->state_mutex); 1099 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) { 1100 mutex_unlock(&sdev->state_mutex); 1101 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev, 1102 "%s: state %d <> %d\n", __func__, sdev->sdev_state, 1103 SDEV_BLOCK)); 1104 delay = min(timeleft, stall_for); 1105 timeleft -= delay; 1106 msleep(jiffies_to_msecs(delay)); 1107 mutex_lock(&sdev->state_mutex); 1108 } 1109 if (sdev->sdev_state != SDEV_BLOCK) 1110 rtn = shost->hostt->queuecommand(shost, scmd); 1111 else 1112 rtn = FAILED; 1113 mutex_unlock(&sdev->state_mutex); 1114 1115 if (rtn) { 1116 if (timeleft > stall_for) { 1117 scsi_eh_restore_cmnd(scmd, &ses); 1118 timeleft -= stall_for; 1119 msleep(jiffies_to_msecs(stall_for)); 1120 goto retry; 1121 } 1122 /* signal not to enter either branch of the if () below */ 1123 timeleft = 0; 1124 rtn = FAILED; 1125 } else { 1126 timeleft = wait_for_completion_timeout(&done, timeout); 1127 rtn = SUCCESS; 1128 } 1129 1130 shost->eh_action = NULL; 1131 1132 scsi_log_completion(scmd, rtn); 1133 1134 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1135 "%s timeleft: %ld\n", 1136 __func__, timeleft)); 1137 1138 /* 1139 * If there is time left scsi_eh_done got called, and we will examine 1140 * the actual status codes to see whether the command actually did 1141 * complete normally, else if we have a zero return and no time left, 1142 * the command must still be pending, so abort it and return FAILED. 1143 * If we never actually managed to issue the command, because 1144 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1145 * value above (so don't execute either branch of the if) 1146 */ 1147 if (timeleft) { 1148 rtn = scsi_eh_completed_normally(scmd); 1149 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1150 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1151 1152 switch (rtn) { 1153 case SUCCESS: 1154 case NEEDS_RETRY: 1155 case FAILED: 1156 break; 1157 case ADD_TO_MLQUEUE: 1158 rtn = NEEDS_RETRY; 1159 break; 1160 default: 1161 rtn = FAILED; 1162 break; 1163 } 1164 } else if (rtn != FAILED) { 1165 scsi_abort_eh_cmnd(scmd); 1166 rtn = FAILED; 1167 } 1168 1169 scsi_eh_restore_cmnd(scmd, &ses); 1170 1171 return rtn; 1172 } 1173 1174 /** 1175 * scsi_request_sense - Request sense data from a particular target. 1176 * @scmd: SCSI cmd for request sense. 1177 * 1178 * Notes: 1179 * Some hosts automatically obtain this information, others require 1180 * that we obtain it on our own. This function will *not* return until 1181 * the command either times out, or it completes. 1182 */ 1183 static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd) 1184 { 1185 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1186 } 1187 1188 static enum scsi_disposition 1189 scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn) 1190 { 1191 if (!blk_rq_is_passthrough(scmd->request)) { 1192 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1193 if (sdrv->eh_action) 1194 rtn = sdrv->eh_action(scmd, rtn); 1195 } 1196 return rtn; 1197 } 1198 1199 /** 1200 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1201 * @scmd: Original SCSI cmd that eh has finished. 1202 * @done_q: Queue for processed commands. 1203 * 1204 * Notes: 1205 * We don't want to use the normal command completion while we are are 1206 * still handling errors - it may cause other commands to be queued, 1207 * and that would disturb what we are doing. Thus we really want to 1208 * keep a list of pending commands for final completion, and once we 1209 * are ready to leave error handling we handle completion for real. 1210 */ 1211 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1212 { 1213 list_move_tail(&scmd->eh_entry, done_q); 1214 } 1215 EXPORT_SYMBOL(scsi_eh_finish_cmd); 1216 1217 /** 1218 * scsi_eh_get_sense - Get device sense data. 1219 * @work_q: Queue of commands to process. 1220 * @done_q: Queue of processed commands. 1221 * 1222 * Description: 1223 * See if we need to request sense information. if so, then get it 1224 * now, so we have a better idea of what to do. 1225 * 1226 * Notes: 1227 * This has the unfortunate side effect that if a shost adapter does 1228 * not automatically request sense information, we end up shutting 1229 * it down before we request it. 1230 * 1231 * All drivers should request sense information internally these days, 1232 * so for now all I have to say is tough noogies if you end up in here. 1233 * 1234 * XXX: Long term this code should go away, but that needs an audit of 1235 * all LLDDs first. 1236 */ 1237 int scsi_eh_get_sense(struct list_head *work_q, 1238 struct list_head *done_q) 1239 { 1240 struct scsi_cmnd *scmd, *next; 1241 struct Scsi_Host *shost; 1242 enum scsi_disposition rtn; 1243 1244 /* 1245 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, 1246 * should not get sense. 1247 */ 1248 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1249 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || 1250 SCSI_SENSE_VALID(scmd)) 1251 continue; 1252 1253 shost = scmd->device->host; 1254 if (scsi_host_eh_past_deadline(shost)) { 1255 SCSI_LOG_ERROR_RECOVERY(3, 1256 scmd_printk(KERN_INFO, scmd, 1257 "%s: skip request sense, past eh deadline\n", 1258 current->comm)); 1259 break; 1260 } 1261 if (status_byte(scmd->result) != CHECK_CONDITION) 1262 /* 1263 * don't request sense if there's no check condition 1264 * status because the error we're processing isn't one 1265 * that has a sense code (and some devices get 1266 * confused by sense requests out of the blue) 1267 */ 1268 continue; 1269 1270 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1271 "%s: requesting sense\n", 1272 current->comm)); 1273 rtn = scsi_request_sense(scmd); 1274 if (rtn != SUCCESS) 1275 continue; 1276 1277 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1278 "sense requested, result %x\n", scmd->result)); 1279 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); 1280 1281 rtn = scsi_decide_disposition(scmd); 1282 1283 /* 1284 * if the result was normal, then just pass it along to the 1285 * upper level. 1286 */ 1287 if (rtn == SUCCESS) 1288 /* 1289 * We don't want this command reissued, just finished 1290 * with the sense data, so set retries to the max 1291 * allowed to ensure it won't get reissued. If the user 1292 * has requested infinite retries, we also want to 1293 * finish this command, so force completion by setting 1294 * retries and allowed to the same value. 1295 */ 1296 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) 1297 scmd->retries = scmd->allowed = 1; 1298 else 1299 scmd->retries = scmd->allowed; 1300 else if (rtn != NEEDS_RETRY) 1301 continue; 1302 1303 scsi_eh_finish_cmd(scmd, done_q); 1304 } 1305 1306 return list_empty(work_q); 1307 } 1308 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1309 1310 /** 1311 * scsi_eh_tur - Send TUR to device. 1312 * @scmd: &scsi_cmnd to send TUR 1313 * 1314 * Return value: 1315 * 0 - Device is ready. 1 - Device NOT ready. 1316 */ 1317 static int scsi_eh_tur(struct scsi_cmnd *scmd) 1318 { 1319 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1320 int retry_cnt = 1; 1321 enum scsi_disposition rtn; 1322 1323 retry_tur: 1324 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1325 scmd->device->eh_timeout, 0); 1326 1327 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1328 "%s return: %x\n", __func__, rtn)); 1329 1330 switch (rtn) { 1331 case NEEDS_RETRY: 1332 if (retry_cnt--) 1333 goto retry_tur; 1334 fallthrough; 1335 case SUCCESS: 1336 return 0; 1337 default: 1338 return 1; 1339 } 1340 } 1341 1342 /** 1343 * scsi_eh_test_devices - check if devices are responding from error recovery. 1344 * @cmd_list: scsi commands in error recovery. 1345 * @work_q: queue for commands which still need more error recovery 1346 * @done_q: queue for commands which are finished 1347 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1348 * 1349 * Decription: 1350 * Tests if devices are in a working state. Commands to devices now in 1351 * a working state are sent to the done_q while commands to devices which 1352 * are still failing to respond are returned to the work_q for more 1353 * processing. 1354 **/ 1355 static int scsi_eh_test_devices(struct list_head *cmd_list, 1356 struct list_head *work_q, 1357 struct list_head *done_q, int try_stu) 1358 { 1359 struct scsi_cmnd *scmd, *next; 1360 struct scsi_device *sdev; 1361 int finish_cmds; 1362 1363 while (!list_empty(cmd_list)) { 1364 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1365 sdev = scmd->device; 1366 1367 if (!try_stu) { 1368 if (scsi_host_eh_past_deadline(sdev->host)) { 1369 /* Push items back onto work_q */ 1370 list_splice_init(cmd_list, work_q); 1371 SCSI_LOG_ERROR_RECOVERY(3, 1372 sdev_printk(KERN_INFO, sdev, 1373 "%s: skip test device, past eh deadline", 1374 current->comm)); 1375 break; 1376 } 1377 } 1378 1379 finish_cmds = !scsi_device_online(scmd->device) || 1380 (try_stu && !scsi_eh_try_stu(scmd) && 1381 !scsi_eh_tur(scmd)) || 1382 !scsi_eh_tur(scmd); 1383 1384 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1385 if (scmd->device == sdev) { 1386 if (finish_cmds && 1387 (try_stu || 1388 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1389 scsi_eh_finish_cmd(scmd, done_q); 1390 else 1391 list_move_tail(&scmd->eh_entry, work_q); 1392 } 1393 } 1394 return list_empty(work_q); 1395 } 1396 1397 /** 1398 * scsi_eh_try_stu - Send START_UNIT to device. 1399 * @scmd: &scsi_cmnd to send START_UNIT 1400 * 1401 * Return value: 1402 * 0 - Device is ready. 1 - Device NOT ready. 1403 */ 1404 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1405 { 1406 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1407 1408 if (scmd->device->allow_restart) { 1409 int i; 1410 enum scsi_disposition rtn = NEEDS_RETRY; 1411 1412 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1413 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0); 1414 1415 if (rtn == SUCCESS) 1416 return 0; 1417 } 1418 1419 return 1; 1420 } 1421 1422 /** 1423 * scsi_eh_stu - send START_UNIT if needed 1424 * @shost: &scsi host being recovered. 1425 * @work_q: &list_head for pending commands. 1426 * @done_q: &list_head for processed commands. 1427 * 1428 * Notes: 1429 * If commands are failing due to not ready, initializing command required, 1430 * try revalidating the device, which will end up sending a start unit. 1431 */ 1432 static int scsi_eh_stu(struct Scsi_Host *shost, 1433 struct list_head *work_q, 1434 struct list_head *done_q) 1435 { 1436 struct scsi_cmnd *scmd, *stu_scmd, *next; 1437 struct scsi_device *sdev; 1438 1439 shost_for_each_device(sdev, shost) { 1440 if (scsi_host_eh_past_deadline(shost)) { 1441 SCSI_LOG_ERROR_RECOVERY(3, 1442 sdev_printk(KERN_INFO, sdev, 1443 "%s: skip START_UNIT, past eh deadline\n", 1444 current->comm)); 1445 scsi_device_put(sdev); 1446 break; 1447 } 1448 stu_scmd = NULL; 1449 list_for_each_entry(scmd, work_q, eh_entry) 1450 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1451 scsi_check_sense(scmd) == FAILED ) { 1452 stu_scmd = scmd; 1453 break; 1454 } 1455 1456 if (!stu_scmd) 1457 continue; 1458 1459 SCSI_LOG_ERROR_RECOVERY(3, 1460 sdev_printk(KERN_INFO, sdev, 1461 "%s: Sending START_UNIT\n", 1462 current->comm)); 1463 1464 if (!scsi_eh_try_stu(stu_scmd)) { 1465 if (!scsi_device_online(sdev) || 1466 !scsi_eh_tur(stu_scmd)) { 1467 list_for_each_entry_safe(scmd, next, 1468 work_q, eh_entry) { 1469 if (scmd->device == sdev && 1470 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1471 scsi_eh_finish_cmd(scmd, done_q); 1472 } 1473 } 1474 } else { 1475 SCSI_LOG_ERROR_RECOVERY(3, 1476 sdev_printk(KERN_INFO, sdev, 1477 "%s: START_UNIT failed\n", 1478 current->comm)); 1479 } 1480 } 1481 1482 return list_empty(work_q); 1483 } 1484 1485 1486 /** 1487 * scsi_eh_bus_device_reset - send bdr if needed 1488 * @shost: scsi host being recovered. 1489 * @work_q: &list_head for pending commands. 1490 * @done_q: &list_head for processed commands. 1491 * 1492 * Notes: 1493 * Try a bus device reset. Still, look to see whether we have multiple 1494 * devices that are jammed or not - if we have multiple devices, it 1495 * makes no sense to try bus_device_reset - we really would need to try 1496 * a bus_reset instead. 1497 */ 1498 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1499 struct list_head *work_q, 1500 struct list_head *done_q) 1501 { 1502 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1503 struct scsi_device *sdev; 1504 enum scsi_disposition rtn; 1505 1506 shost_for_each_device(sdev, shost) { 1507 if (scsi_host_eh_past_deadline(shost)) { 1508 SCSI_LOG_ERROR_RECOVERY(3, 1509 sdev_printk(KERN_INFO, sdev, 1510 "%s: skip BDR, past eh deadline\n", 1511 current->comm)); 1512 scsi_device_put(sdev); 1513 break; 1514 } 1515 bdr_scmd = NULL; 1516 list_for_each_entry(scmd, work_q, eh_entry) 1517 if (scmd->device == sdev) { 1518 bdr_scmd = scmd; 1519 break; 1520 } 1521 1522 if (!bdr_scmd) 1523 continue; 1524 1525 SCSI_LOG_ERROR_RECOVERY(3, 1526 sdev_printk(KERN_INFO, sdev, 1527 "%s: Sending BDR\n", current->comm)); 1528 rtn = scsi_try_bus_device_reset(bdr_scmd); 1529 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1530 if (!scsi_device_online(sdev) || 1531 rtn == FAST_IO_FAIL || 1532 !scsi_eh_tur(bdr_scmd)) { 1533 list_for_each_entry_safe(scmd, next, 1534 work_q, eh_entry) { 1535 if (scmd->device == sdev && 1536 scsi_eh_action(scmd, rtn) != FAILED) 1537 scsi_eh_finish_cmd(scmd, 1538 done_q); 1539 } 1540 } 1541 } else { 1542 SCSI_LOG_ERROR_RECOVERY(3, 1543 sdev_printk(KERN_INFO, sdev, 1544 "%s: BDR failed\n", current->comm)); 1545 } 1546 } 1547 1548 return list_empty(work_q); 1549 } 1550 1551 /** 1552 * scsi_eh_target_reset - send target reset if needed 1553 * @shost: scsi host being recovered. 1554 * @work_q: &list_head for pending commands. 1555 * @done_q: &list_head for processed commands. 1556 * 1557 * Notes: 1558 * Try a target reset. 1559 */ 1560 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1561 struct list_head *work_q, 1562 struct list_head *done_q) 1563 { 1564 LIST_HEAD(tmp_list); 1565 LIST_HEAD(check_list); 1566 1567 list_splice_init(work_q, &tmp_list); 1568 1569 while (!list_empty(&tmp_list)) { 1570 struct scsi_cmnd *next, *scmd; 1571 enum scsi_disposition rtn; 1572 unsigned int id; 1573 1574 if (scsi_host_eh_past_deadline(shost)) { 1575 /* push back on work queue for further processing */ 1576 list_splice_init(&check_list, work_q); 1577 list_splice_init(&tmp_list, work_q); 1578 SCSI_LOG_ERROR_RECOVERY(3, 1579 shost_printk(KERN_INFO, shost, 1580 "%s: Skip target reset, past eh deadline\n", 1581 current->comm)); 1582 return list_empty(work_q); 1583 } 1584 1585 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1586 id = scmd_id(scmd); 1587 1588 SCSI_LOG_ERROR_RECOVERY(3, 1589 shost_printk(KERN_INFO, shost, 1590 "%s: Sending target reset to target %d\n", 1591 current->comm, id)); 1592 rtn = scsi_try_target_reset(scmd); 1593 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1594 SCSI_LOG_ERROR_RECOVERY(3, 1595 shost_printk(KERN_INFO, shost, 1596 "%s: Target reset failed" 1597 " target: %d\n", 1598 current->comm, id)); 1599 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1600 if (scmd_id(scmd) != id) 1601 continue; 1602 1603 if (rtn == SUCCESS) 1604 list_move_tail(&scmd->eh_entry, &check_list); 1605 else if (rtn == FAST_IO_FAIL) 1606 scsi_eh_finish_cmd(scmd, done_q); 1607 else 1608 /* push back on work queue for further processing */ 1609 list_move(&scmd->eh_entry, work_q); 1610 } 1611 } 1612 1613 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1614 } 1615 1616 /** 1617 * scsi_eh_bus_reset - send a bus reset 1618 * @shost: &scsi host being recovered. 1619 * @work_q: &list_head for pending commands. 1620 * @done_q: &list_head for processed commands. 1621 */ 1622 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1623 struct list_head *work_q, 1624 struct list_head *done_q) 1625 { 1626 struct scsi_cmnd *scmd, *chan_scmd, *next; 1627 LIST_HEAD(check_list); 1628 unsigned int channel; 1629 enum scsi_disposition rtn; 1630 1631 /* 1632 * we really want to loop over the various channels, and do this on 1633 * a channel by channel basis. we should also check to see if any 1634 * of the failed commands are on soft_reset devices, and if so, skip 1635 * the reset. 1636 */ 1637 1638 for (channel = 0; channel <= shost->max_channel; channel++) { 1639 if (scsi_host_eh_past_deadline(shost)) { 1640 list_splice_init(&check_list, work_q); 1641 SCSI_LOG_ERROR_RECOVERY(3, 1642 shost_printk(KERN_INFO, shost, 1643 "%s: skip BRST, past eh deadline\n", 1644 current->comm)); 1645 return list_empty(work_q); 1646 } 1647 1648 chan_scmd = NULL; 1649 list_for_each_entry(scmd, work_q, eh_entry) { 1650 if (channel == scmd_channel(scmd)) { 1651 chan_scmd = scmd; 1652 break; 1653 /* 1654 * FIXME add back in some support for 1655 * soft_reset devices. 1656 */ 1657 } 1658 } 1659 1660 if (!chan_scmd) 1661 continue; 1662 SCSI_LOG_ERROR_RECOVERY(3, 1663 shost_printk(KERN_INFO, shost, 1664 "%s: Sending BRST chan: %d\n", 1665 current->comm, channel)); 1666 rtn = scsi_try_bus_reset(chan_scmd); 1667 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1668 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1669 if (channel == scmd_channel(scmd)) { 1670 if (rtn == FAST_IO_FAIL) 1671 scsi_eh_finish_cmd(scmd, 1672 done_q); 1673 else 1674 list_move_tail(&scmd->eh_entry, 1675 &check_list); 1676 } 1677 } 1678 } else { 1679 SCSI_LOG_ERROR_RECOVERY(3, 1680 shost_printk(KERN_INFO, shost, 1681 "%s: BRST failed chan: %d\n", 1682 current->comm, channel)); 1683 } 1684 } 1685 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1686 } 1687 1688 /** 1689 * scsi_eh_host_reset - send a host reset 1690 * @shost: host to be reset. 1691 * @work_q: &list_head for pending commands. 1692 * @done_q: &list_head for processed commands. 1693 */ 1694 static int scsi_eh_host_reset(struct Scsi_Host *shost, 1695 struct list_head *work_q, 1696 struct list_head *done_q) 1697 { 1698 struct scsi_cmnd *scmd, *next; 1699 LIST_HEAD(check_list); 1700 enum scsi_disposition rtn; 1701 1702 if (!list_empty(work_q)) { 1703 scmd = list_entry(work_q->next, 1704 struct scsi_cmnd, eh_entry); 1705 1706 SCSI_LOG_ERROR_RECOVERY(3, 1707 shost_printk(KERN_INFO, shost, 1708 "%s: Sending HRST\n", 1709 current->comm)); 1710 1711 rtn = scsi_try_host_reset(scmd); 1712 if (rtn == SUCCESS) { 1713 list_splice_init(work_q, &check_list); 1714 } else if (rtn == FAST_IO_FAIL) { 1715 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1716 scsi_eh_finish_cmd(scmd, done_q); 1717 } 1718 } else { 1719 SCSI_LOG_ERROR_RECOVERY(3, 1720 shost_printk(KERN_INFO, shost, 1721 "%s: HRST failed\n", 1722 current->comm)); 1723 } 1724 } 1725 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1726 } 1727 1728 /** 1729 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1730 * @work_q: &list_head for pending commands. 1731 * @done_q: &list_head for processed commands. 1732 */ 1733 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1734 struct list_head *done_q) 1735 { 1736 struct scsi_cmnd *scmd, *next; 1737 struct scsi_device *sdev; 1738 1739 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1740 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1741 "not ready after error recovery\n"); 1742 sdev = scmd->device; 1743 1744 mutex_lock(&sdev->state_mutex); 1745 scsi_device_set_state(sdev, SDEV_OFFLINE); 1746 mutex_unlock(&sdev->state_mutex); 1747 1748 scsi_eh_finish_cmd(scmd, done_q); 1749 } 1750 return; 1751 } 1752 1753 /** 1754 * scsi_noretry_cmd - determine if command should be failed fast 1755 * @scmd: SCSI cmd to examine. 1756 */ 1757 int scsi_noretry_cmd(struct scsi_cmnd *scmd) 1758 { 1759 switch (host_byte(scmd->result)) { 1760 case DID_OK: 1761 break; 1762 case DID_TIME_OUT: 1763 goto check_type; 1764 case DID_BUS_BUSY: 1765 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT); 1766 case DID_PARITY: 1767 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV); 1768 case DID_ERROR: 1769 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1770 status_byte(scmd->result) == RESERVATION_CONFLICT) 1771 return 0; 1772 fallthrough; 1773 case DID_SOFT_ERROR: 1774 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER); 1775 } 1776 1777 if (status_byte(scmd->result) != CHECK_CONDITION) 1778 return 0; 1779 1780 check_type: 1781 /* 1782 * assume caller has checked sense and determined 1783 * the check condition was retryable. 1784 */ 1785 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV || 1786 blk_rq_is_passthrough(scmd->request)) 1787 return 1; 1788 1789 return 0; 1790 } 1791 1792 /** 1793 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1794 * @scmd: SCSI cmd to examine. 1795 * 1796 * Notes: 1797 * This is *only* called when we are examining the status after sending 1798 * out the actual data command. any commands that are queued for error 1799 * recovery (e.g. test_unit_ready) do *not* come through here. 1800 * 1801 * When this routine returns failed, it means the error handler thread 1802 * is woken. In cases where the error code indicates an error that 1803 * doesn't require the error handler read (i.e. we don't need to 1804 * abort/reset), this function should return SUCCESS. 1805 */ 1806 enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd) 1807 { 1808 enum scsi_disposition rtn; 1809 1810 /* 1811 * if the device is offline, then we clearly just pass the result back 1812 * up to the top level. 1813 */ 1814 if (!scsi_device_online(scmd->device)) { 1815 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1816 "%s: device offline - report as SUCCESS\n", __func__)); 1817 return SUCCESS; 1818 } 1819 1820 /* 1821 * first check the host byte, to see if there is anything in there 1822 * that would indicate what we need to do. 1823 */ 1824 switch (host_byte(scmd->result)) { 1825 case DID_PASSTHROUGH: 1826 /* 1827 * no matter what, pass this through to the upper layer. 1828 * nuke this special code so that it looks like we are saying 1829 * did_ok. 1830 */ 1831 scmd->result &= 0xff00ffff; 1832 return SUCCESS; 1833 case DID_OK: 1834 /* 1835 * looks good. drop through, and check the next byte. 1836 */ 1837 break; 1838 case DID_ABORT: 1839 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1840 set_host_byte(scmd, DID_TIME_OUT); 1841 return SUCCESS; 1842 } 1843 fallthrough; 1844 case DID_NO_CONNECT: 1845 case DID_BAD_TARGET: 1846 /* 1847 * note - this means that we just report the status back 1848 * to the top level driver, not that we actually think 1849 * that it indicates SUCCESS. 1850 */ 1851 return SUCCESS; 1852 case DID_SOFT_ERROR: 1853 /* 1854 * when the low level driver returns did_soft_error, 1855 * it is responsible for keeping an internal retry counter 1856 * in order to avoid endless loops (db) 1857 */ 1858 goto maybe_retry; 1859 case DID_IMM_RETRY: 1860 return NEEDS_RETRY; 1861 1862 case DID_REQUEUE: 1863 return ADD_TO_MLQUEUE; 1864 case DID_TRANSPORT_DISRUPTED: 1865 /* 1866 * LLD/transport was disrupted during processing of the IO. 1867 * The transport class is now blocked/blocking, 1868 * and the transport will decide what to do with the IO 1869 * based on its timers and recovery capablilities if 1870 * there are enough retries. 1871 */ 1872 goto maybe_retry; 1873 case DID_TRANSPORT_FAILFAST: 1874 /* 1875 * The transport decided to failfast the IO (most likely 1876 * the fast io fail tmo fired), so send IO directly upwards. 1877 */ 1878 return SUCCESS; 1879 case DID_TRANSPORT_MARGINAL: 1880 /* 1881 * caller has decided not to do retries on 1882 * abort success, so send IO directly upwards 1883 */ 1884 return SUCCESS; 1885 case DID_ERROR: 1886 if (msg_byte(scmd->result) == COMMAND_COMPLETE && 1887 status_byte(scmd->result) == RESERVATION_CONFLICT) 1888 /* 1889 * execute reservation conflict processing code 1890 * lower down 1891 */ 1892 break; 1893 fallthrough; 1894 case DID_BUS_BUSY: 1895 case DID_PARITY: 1896 goto maybe_retry; 1897 case DID_TIME_OUT: 1898 /* 1899 * when we scan the bus, we get timeout messages for 1900 * these commands if there is no device available. 1901 * other hosts report did_no_connect for the same thing. 1902 */ 1903 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1904 scmd->cmnd[0] == INQUIRY)) { 1905 return SUCCESS; 1906 } else { 1907 return FAILED; 1908 } 1909 case DID_RESET: 1910 return SUCCESS; 1911 default: 1912 return FAILED; 1913 } 1914 1915 /* 1916 * next, check the message byte. 1917 */ 1918 if (msg_byte(scmd->result) != COMMAND_COMPLETE) 1919 return FAILED; 1920 1921 /* 1922 * check the status byte to see if this indicates anything special. 1923 */ 1924 switch (status_byte(scmd->result)) { 1925 case QUEUE_FULL: 1926 scsi_handle_queue_full(scmd->device); 1927 /* 1928 * the case of trying to send too many commands to a 1929 * tagged queueing device. 1930 */ 1931 fallthrough; 1932 case BUSY: 1933 /* 1934 * device can't talk to us at the moment. Should only 1935 * occur (SAM-3) when the task queue is empty, so will cause 1936 * the empty queue handling to trigger a stall in the 1937 * device. 1938 */ 1939 return ADD_TO_MLQUEUE; 1940 case GOOD: 1941 if (scmd->cmnd[0] == REPORT_LUNS) 1942 scmd->device->sdev_target->expecting_lun_change = 0; 1943 scsi_handle_queue_ramp_up(scmd->device); 1944 fallthrough; 1945 case COMMAND_TERMINATED: 1946 return SUCCESS; 1947 case TASK_ABORTED: 1948 goto maybe_retry; 1949 case CHECK_CONDITION: 1950 rtn = scsi_check_sense(scmd); 1951 if (rtn == NEEDS_RETRY) 1952 goto maybe_retry; 1953 /* if rtn == FAILED, we have no sense information; 1954 * returning FAILED will wake the error handler thread 1955 * to collect the sense and redo the decide 1956 * disposition */ 1957 return rtn; 1958 case CONDITION_GOOD: 1959 case INTERMEDIATE_GOOD: 1960 case INTERMEDIATE_C_GOOD: 1961 case ACA_ACTIVE: 1962 /* 1963 * who knows? FIXME(eric) 1964 */ 1965 return SUCCESS; 1966 1967 case RESERVATION_CONFLICT: 1968 sdev_printk(KERN_INFO, scmd->device, 1969 "reservation conflict\n"); 1970 set_host_byte(scmd, DID_NEXUS_FAILURE); 1971 return SUCCESS; /* causes immediate i/o error */ 1972 default: 1973 return FAILED; 1974 } 1975 return FAILED; 1976 1977 maybe_retry: 1978 1979 /* we requeue for retry because the error was retryable, and 1980 * the request was not marked fast fail. Note that above, 1981 * even if the request is marked fast fail, we still requeue 1982 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1983 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) { 1984 return NEEDS_RETRY; 1985 } else { 1986 /* 1987 * no more retries - report this one back to upper level. 1988 */ 1989 return SUCCESS; 1990 } 1991 } 1992 1993 static void eh_lock_door_done(struct request *req, blk_status_t status) 1994 { 1995 blk_put_request(req); 1996 } 1997 1998 /** 1999 * scsi_eh_lock_door - Prevent medium removal for the specified device 2000 * @sdev: SCSI device to prevent medium removal 2001 * 2002 * Locking: 2003 * We must be called from process context. 2004 * 2005 * Notes: 2006 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 2007 * head of the devices request queue, and continue. 2008 */ 2009 static void scsi_eh_lock_door(struct scsi_device *sdev) 2010 { 2011 struct request *req; 2012 struct scsi_request *rq; 2013 2014 req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, 0); 2015 if (IS_ERR(req)) 2016 return; 2017 rq = scsi_req(req); 2018 2019 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL; 2020 rq->cmd[1] = 0; 2021 rq->cmd[2] = 0; 2022 rq->cmd[3] = 0; 2023 rq->cmd[4] = SCSI_REMOVAL_PREVENT; 2024 rq->cmd[5] = 0; 2025 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]); 2026 2027 req->rq_flags |= RQF_QUIET; 2028 req->timeout = 10 * HZ; 2029 rq->retries = 5; 2030 2031 blk_execute_rq_nowait(NULL, req, 1, eh_lock_door_done); 2032 } 2033 2034 /** 2035 * scsi_restart_operations - restart io operations to the specified host. 2036 * @shost: Host we are restarting. 2037 * 2038 * Notes: 2039 * When we entered the error handler, we blocked all further i/o to 2040 * this device. we need to 'reverse' this process. 2041 */ 2042 static void scsi_restart_operations(struct Scsi_Host *shost) 2043 { 2044 struct scsi_device *sdev; 2045 unsigned long flags; 2046 2047 /* 2048 * If the door was locked, we need to insert a door lock request 2049 * onto the head of the SCSI request queue for the device. There 2050 * is no point trying to lock the door of an off-line device. 2051 */ 2052 shost_for_each_device(sdev, shost) { 2053 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 2054 scsi_eh_lock_door(sdev); 2055 sdev->was_reset = 0; 2056 } 2057 } 2058 2059 /* 2060 * next free up anything directly waiting upon the host. this 2061 * will be requests for character device operations, and also for 2062 * ioctls to queued block devices. 2063 */ 2064 SCSI_LOG_ERROR_RECOVERY(3, 2065 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 2066 2067 spin_lock_irqsave(shost->host_lock, flags); 2068 if (scsi_host_set_state(shost, SHOST_RUNNING)) 2069 if (scsi_host_set_state(shost, SHOST_CANCEL)) 2070 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 2071 spin_unlock_irqrestore(shost->host_lock, flags); 2072 2073 wake_up(&shost->host_wait); 2074 2075 /* 2076 * finally we need to re-initiate requests that may be pending. we will 2077 * have had everything blocked while error handling is taking place, and 2078 * now that error recovery is done, we will need to ensure that these 2079 * requests are started. 2080 */ 2081 scsi_run_host_queues(shost); 2082 2083 /* 2084 * if eh is active and host_eh_scheduled is pending we need to re-run 2085 * recovery. we do this check after scsi_run_host_queues() to allow 2086 * everything pent up since the last eh run a chance to make forward 2087 * progress before we sync again. Either we'll immediately re-run 2088 * recovery or scsi_device_unbusy() will wake us again when these 2089 * pending commands complete. 2090 */ 2091 spin_lock_irqsave(shost->host_lock, flags); 2092 if (shost->host_eh_scheduled) 2093 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 2094 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 2095 spin_unlock_irqrestore(shost->host_lock, flags); 2096 } 2097 2098 /** 2099 * scsi_eh_ready_devs - check device ready state and recover if not. 2100 * @shost: host to be recovered. 2101 * @work_q: &list_head for pending commands. 2102 * @done_q: &list_head for processed commands. 2103 */ 2104 void scsi_eh_ready_devs(struct Scsi_Host *shost, 2105 struct list_head *work_q, 2106 struct list_head *done_q) 2107 { 2108 if (!scsi_eh_stu(shost, work_q, done_q)) 2109 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2110 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2111 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2112 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2113 scsi_eh_offline_sdevs(work_q, 2114 done_q); 2115 } 2116 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2117 2118 /** 2119 * scsi_eh_flush_done_q - finish processed commands or retry them. 2120 * @done_q: list_head of processed commands. 2121 */ 2122 void scsi_eh_flush_done_q(struct list_head *done_q) 2123 { 2124 struct scsi_cmnd *scmd, *next; 2125 2126 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2127 list_del_init(&scmd->eh_entry); 2128 if (scsi_device_online(scmd->device) && 2129 !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd) && 2130 scsi_eh_should_retry_cmd(scmd)) { 2131 SCSI_LOG_ERROR_RECOVERY(3, 2132 scmd_printk(KERN_INFO, scmd, 2133 "%s: flush retry cmd\n", 2134 current->comm)); 2135 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2136 } else { 2137 /* 2138 * If just we got sense for the device (called 2139 * scsi_eh_get_sense), scmd->result is already 2140 * set, do not set DRIVER_TIMEOUT. 2141 */ 2142 if (!scmd->result) 2143 scmd->result |= (DRIVER_TIMEOUT << 24); 2144 SCSI_LOG_ERROR_RECOVERY(3, 2145 scmd_printk(KERN_INFO, scmd, 2146 "%s: flush finish cmd\n", 2147 current->comm)); 2148 scsi_finish_command(scmd); 2149 } 2150 } 2151 } 2152 EXPORT_SYMBOL(scsi_eh_flush_done_q); 2153 2154 /** 2155 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2156 * @shost: Host to unjam. 2157 * 2158 * Notes: 2159 * When we come in here, we *know* that all commands on the bus have 2160 * either completed, failed or timed out. we also know that no further 2161 * commands are being sent to the host, so things are relatively quiet 2162 * and we have freedom to fiddle with things as we wish. 2163 * 2164 * This is only the *default* implementation. it is possible for 2165 * individual drivers to supply their own version of this function, and 2166 * if the maintainer wishes to do this, it is strongly suggested that 2167 * this function be taken as a template and modified. this function 2168 * was designed to correctly handle problems for about 95% of the 2169 * different cases out there, and it should always provide at least a 2170 * reasonable amount of error recovery. 2171 * 2172 * Any command marked 'failed' or 'timeout' must eventually have 2173 * scsi_finish_cmd() called for it. we do all of the retry stuff 2174 * here, so when we restart the host after we return it should have an 2175 * empty queue. 2176 */ 2177 static void scsi_unjam_host(struct Scsi_Host *shost) 2178 { 2179 unsigned long flags; 2180 LIST_HEAD(eh_work_q); 2181 LIST_HEAD(eh_done_q); 2182 2183 spin_lock_irqsave(shost->host_lock, flags); 2184 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2185 spin_unlock_irqrestore(shost->host_lock, flags); 2186 2187 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2188 2189 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2190 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2191 2192 spin_lock_irqsave(shost->host_lock, flags); 2193 if (shost->eh_deadline != -1) 2194 shost->last_reset = 0; 2195 spin_unlock_irqrestore(shost->host_lock, flags); 2196 scsi_eh_flush_done_q(&eh_done_q); 2197 } 2198 2199 /** 2200 * scsi_error_handler - SCSI error handler thread 2201 * @data: Host for which we are running. 2202 * 2203 * Notes: 2204 * This is the main error handling loop. This is run as a kernel thread 2205 * for every SCSI host and handles all error handling activity. 2206 */ 2207 int scsi_error_handler(void *data) 2208 { 2209 struct Scsi_Host *shost = data; 2210 2211 /* 2212 * We use TASK_INTERRUPTIBLE so that the thread is not 2213 * counted against the load average as a running process. 2214 * We never actually get interrupted because kthread_run 2215 * disables signal delivery for the created thread. 2216 */ 2217 while (true) { 2218 /* 2219 * The sequence in kthread_stop() sets the stop flag first 2220 * then wakes the process. To avoid missed wakeups, the task 2221 * should always be in a non running state before the stop 2222 * flag is checked 2223 */ 2224 set_current_state(TASK_INTERRUPTIBLE); 2225 if (kthread_should_stop()) 2226 break; 2227 2228 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2229 shost->host_failed != scsi_host_busy(shost)) { 2230 SCSI_LOG_ERROR_RECOVERY(1, 2231 shost_printk(KERN_INFO, shost, 2232 "scsi_eh_%d: sleeping\n", 2233 shost->host_no)); 2234 schedule(); 2235 continue; 2236 } 2237 2238 __set_current_state(TASK_RUNNING); 2239 SCSI_LOG_ERROR_RECOVERY(1, 2240 shost_printk(KERN_INFO, shost, 2241 "scsi_eh_%d: waking up %d/%d/%d\n", 2242 shost->host_no, shost->host_eh_scheduled, 2243 shost->host_failed, 2244 scsi_host_busy(shost))); 2245 2246 /* 2247 * We have a host that is failing for some reason. Figure out 2248 * what we need to do to get it up and online again (if we can). 2249 * If we fail, we end up taking the thing offline. 2250 */ 2251 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { 2252 SCSI_LOG_ERROR_RECOVERY(1, 2253 shost_printk(KERN_ERR, shost, 2254 "scsi_eh_%d: unable to autoresume\n", 2255 shost->host_no)); 2256 continue; 2257 } 2258 2259 if (shost->transportt->eh_strategy_handler) 2260 shost->transportt->eh_strategy_handler(shost); 2261 else 2262 scsi_unjam_host(shost); 2263 2264 /* All scmds have been handled */ 2265 shost->host_failed = 0; 2266 2267 /* 2268 * Note - if the above fails completely, the action is to take 2269 * individual devices offline and flush the queue of any 2270 * outstanding requests that may have been pending. When we 2271 * restart, we restart any I/O to any other devices on the bus 2272 * which are still online. 2273 */ 2274 scsi_restart_operations(shost); 2275 if (!shost->eh_noresume) 2276 scsi_autopm_put_host(shost); 2277 } 2278 __set_current_state(TASK_RUNNING); 2279 2280 SCSI_LOG_ERROR_RECOVERY(1, 2281 shost_printk(KERN_INFO, shost, 2282 "Error handler scsi_eh_%d exiting\n", 2283 shost->host_no)); 2284 shost->ehandler = NULL; 2285 return 0; 2286 } 2287 2288 /* 2289 * Function: scsi_report_bus_reset() 2290 * 2291 * Purpose: Utility function used by low-level drivers to report that 2292 * they have observed a bus reset on the bus being handled. 2293 * 2294 * Arguments: shost - Host in question 2295 * channel - channel on which reset was observed. 2296 * 2297 * Returns: Nothing 2298 * 2299 * Lock status: Host lock must be held. 2300 * 2301 * Notes: This only needs to be called if the reset is one which 2302 * originates from an unknown location. Resets originated 2303 * by the mid-level itself don't need to call this, but there 2304 * should be no harm. 2305 * 2306 * The main purpose of this is to make sure that a CHECK_CONDITION 2307 * is properly treated. 2308 */ 2309 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2310 { 2311 struct scsi_device *sdev; 2312 2313 __shost_for_each_device(sdev, shost) { 2314 if (channel == sdev_channel(sdev)) 2315 __scsi_report_device_reset(sdev, NULL); 2316 } 2317 } 2318 EXPORT_SYMBOL(scsi_report_bus_reset); 2319 2320 /* 2321 * Function: scsi_report_device_reset() 2322 * 2323 * Purpose: Utility function used by low-level drivers to report that 2324 * they have observed a device reset on the device being handled. 2325 * 2326 * Arguments: shost - Host in question 2327 * channel - channel on which reset was observed 2328 * target - target on which reset was observed 2329 * 2330 * Returns: Nothing 2331 * 2332 * Lock status: Host lock must be held 2333 * 2334 * Notes: This only needs to be called if the reset is one which 2335 * originates from an unknown location. Resets originated 2336 * by the mid-level itself don't need to call this, but there 2337 * should be no harm. 2338 * 2339 * The main purpose of this is to make sure that a CHECK_CONDITION 2340 * is properly treated. 2341 */ 2342 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2343 { 2344 struct scsi_device *sdev; 2345 2346 __shost_for_each_device(sdev, shost) { 2347 if (channel == sdev_channel(sdev) && 2348 target == sdev_id(sdev)) 2349 __scsi_report_device_reset(sdev, NULL); 2350 } 2351 } 2352 EXPORT_SYMBOL(scsi_report_device_reset); 2353 2354 static void 2355 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 2356 { 2357 } 2358 2359 /** 2360 * scsi_ioctl_reset: explicitly reset a host/bus/target/device 2361 * @dev: scsi_device to operate on 2362 * @arg: reset type (see sg.h) 2363 */ 2364 int 2365 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) 2366 { 2367 struct scsi_cmnd *scmd; 2368 struct Scsi_Host *shost = dev->host; 2369 struct request *rq; 2370 unsigned long flags; 2371 int error = 0, val; 2372 enum scsi_disposition rtn; 2373 2374 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) 2375 return -EACCES; 2376 2377 error = get_user(val, arg); 2378 if (error) 2379 return error; 2380 2381 if (scsi_autopm_get_host(shost) < 0) 2382 return -EIO; 2383 2384 error = -EIO; 2385 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) + 2386 shost->hostt->cmd_size, GFP_KERNEL); 2387 if (!rq) 2388 goto out_put_autopm_host; 2389 blk_rq_init(NULL, rq); 2390 2391 scmd = (struct scsi_cmnd *)(rq + 1); 2392 scsi_init_command(dev, scmd); 2393 scmd->request = rq; 2394 scmd->cmnd = scsi_req(rq)->cmd; 2395 2396 scmd->scsi_done = scsi_reset_provider_done_command; 2397 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 2398 2399 scmd->cmd_len = 0; 2400 2401 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 2402 2403 spin_lock_irqsave(shost->host_lock, flags); 2404 shost->tmf_in_progress = 1; 2405 spin_unlock_irqrestore(shost->host_lock, flags); 2406 2407 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { 2408 case SG_SCSI_RESET_NOTHING: 2409 rtn = SUCCESS; 2410 break; 2411 case SG_SCSI_RESET_DEVICE: 2412 rtn = scsi_try_bus_device_reset(scmd); 2413 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2414 break; 2415 fallthrough; 2416 case SG_SCSI_RESET_TARGET: 2417 rtn = scsi_try_target_reset(scmd); 2418 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2419 break; 2420 fallthrough; 2421 case SG_SCSI_RESET_BUS: 2422 rtn = scsi_try_bus_reset(scmd); 2423 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2424 break; 2425 fallthrough; 2426 case SG_SCSI_RESET_HOST: 2427 rtn = scsi_try_host_reset(scmd); 2428 if (rtn == SUCCESS) 2429 break; 2430 fallthrough; 2431 default: 2432 rtn = FAILED; 2433 break; 2434 } 2435 2436 error = (rtn == SUCCESS) ? 0 : -EIO; 2437 2438 spin_lock_irqsave(shost->host_lock, flags); 2439 shost->tmf_in_progress = 0; 2440 spin_unlock_irqrestore(shost->host_lock, flags); 2441 2442 /* 2443 * be sure to wake up anyone who was sleeping or had their queue 2444 * suspended while we performed the TMF. 2445 */ 2446 SCSI_LOG_ERROR_RECOVERY(3, 2447 shost_printk(KERN_INFO, shost, 2448 "waking up host to restart after TMF\n")); 2449 2450 wake_up(&shost->host_wait); 2451 scsi_run_host_queues(shost); 2452 2453 kfree(rq); 2454 2455 out_put_autopm_host: 2456 scsi_autopm_put_host(shost); 2457 return error; 2458 } 2459 2460 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, 2461 struct scsi_sense_hdr *sshdr) 2462 { 2463 return scsi_normalize_sense(cmd->sense_buffer, 2464 SCSI_SENSE_BUFFERSIZE, sshdr); 2465 } 2466 EXPORT_SYMBOL(scsi_command_normalize_sense); 2467 2468 /** 2469 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2470 * @sense_buffer: byte array of sense data 2471 * @sb_len: number of valid bytes in sense_buffer 2472 * @info_out: pointer to 64 integer where 8 or 4 byte information 2473 * field will be placed if found. 2474 * 2475 * Return value: 2476 * true if information field found, false if not found. 2477 */ 2478 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len, 2479 u64 *info_out) 2480 { 2481 const u8 * ucp; 2482 2483 if (sb_len < 7) 2484 return false; 2485 switch (sense_buffer[0] & 0x7f) { 2486 case 0x70: 2487 case 0x71: 2488 if (sense_buffer[0] & 0x80) { 2489 *info_out = get_unaligned_be32(&sense_buffer[3]); 2490 return true; 2491 } 2492 return false; 2493 case 0x72: 2494 case 0x73: 2495 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2496 0 /* info desc */); 2497 if (ucp && (0xa == ucp[1])) { 2498 *info_out = get_unaligned_be64(&ucp[4]); 2499 return true; 2500 } 2501 return false; 2502 default: 2503 return false; 2504 } 2505 } 2506 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2507