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 * now, check the status byte to see if this indicates 746 * anything special. 747 */ 748 switch (get_status_byte(scmd)) { 749 case SAM_STAT_GOOD: 750 scsi_handle_queue_ramp_up(scmd->device); 751 fallthrough; 752 case SAM_STAT_COMMAND_TERMINATED: 753 return SUCCESS; 754 case SAM_STAT_CHECK_CONDITION: 755 return scsi_check_sense(scmd); 756 case SAM_STAT_CONDITION_MET: 757 case SAM_STAT_INTERMEDIATE: 758 case SAM_STAT_INTERMEDIATE_CONDITION_MET: 759 /* 760 * who knows? FIXME(eric) 761 */ 762 return SUCCESS; 763 case SAM_STAT_RESERVATION_CONFLICT: 764 if (scmd->cmnd[0] == TEST_UNIT_READY) 765 /* it is a success, we probed the device and 766 * found it */ 767 return SUCCESS; 768 /* otherwise, we failed to send the command */ 769 return FAILED; 770 case SAM_STAT_TASK_SET_FULL: 771 scsi_handle_queue_full(scmd->device); 772 fallthrough; 773 case SAM_STAT_BUSY: 774 return NEEDS_RETRY; 775 default: 776 return FAILED; 777 } 778 return FAILED; 779 } 780 781 /** 782 * scsi_eh_done - Completion function for error handling. 783 * @scmd: Cmd that is done. 784 */ 785 static void scsi_eh_done(struct scsi_cmnd *scmd) 786 { 787 struct completion *eh_action; 788 789 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 790 "%s result: %x\n", __func__, scmd->result)); 791 792 eh_action = scmd->device->host->eh_action; 793 if (eh_action) 794 complete(eh_action); 795 } 796 797 /** 798 * scsi_try_host_reset - ask host adapter to reset itself 799 * @scmd: SCSI cmd to send host reset. 800 */ 801 static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd) 802 { 803 unsigned long flags; 804 enum scsi_disposition rtn; 805 struct Scsi_Host *host = scmd->device->host; 806 struct scsi_host_template *hostt = host->hostt; 807 808 SCSI_LOG_ERROR_RECOVERY(3, 809 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 810 811 if (!hostt->eh_host_reset_handler) 812 return FAILED; 813 814 rtn = hostt->eh_host_reset_handler(scmd); 815 816 if (rtn == SUCCESS) { 817 if (!hostt->skip_settle_delay) 818 ssleep(HOST_RESET_SETTLE_TIME); 819 spin_lock_irqsave(host->host_lock, flags); 820 scsi_report_bus_reset(host, scmd_channel(scmd)); 821 spin_unlock_irqrestore(host->host_lock, flags); 822 } 823 824 return rtn; 825 } 826 827 /** 828 * scsi_try_bus_reset - ask host to perform a bus reset 829 * @scmd: SCSI cmd to send bus reset. 830 */ 831 static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd) 832 { 833 unsigned long flags; 834 enum scsi_disposition rtn; 835 struct Scsi_Host *host = scmd->device->host; 836 struct scsi_host_template *hostt = host->hostt; 837 838 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 839 "%s: Snd Bus RST\n", __func__)); 840 841 if (!hostt->eh_bus_reset_handler) 842 return FAILED; 843 844 rtn = hostt->eh_bus_reset_handler(scmd); 845 846 if (rtn == SUCCESS) { 847 if (!hostt->skip_settle_delay) 848 ssleep(BUS_RESET_SETTLE_TIME); 849 spin_lock_irqsave(host->host_lock, flags); 850 scsi_report_bus_reset(host, scmd_channel(scmd)); 851 spin_unlock_irqrestore(host->host_lock, flags); 852 } 853 854 return rtn; 855 } 856 857 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 858 { 859 sdev->was_reset = 1; 860 sdev->expecting_cc_ua = 1; 861 } 862 863 /** 864 * scsi_try_target_reset - Ask host to perform a target reset 865 * @scmd: SCSI cmd used to send a target reset 866 * 867 * Notes: 868 * There is no timeout for this operation. if this operation is 869 * unreliable for a given host, then the host itself needs to put a 870 * timer on it, and set the host back to a consistent state prior to 871 * returning. 872 */ 873 static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd) 874 { 875 unsigned long flags; 876 enum scsi_disposition rtn; 877 struct Scsi_Host *host = scmd->device->host; 878 struct scsi_host_template *hostt = host->hostt; 879 880 if (!hostt->eh_target_reset_handler) 881 return FAILED; 882 883 rtn = hostt->eh_target_reset_handler(scmd); 884 if (rtn == SUCCESS) { 885 spin_lock_irqsave(host->host_lock, flags); 886 __starget_for_each_device(scsi_target(scmd->device), NULL, 887 __scsi_report_device_reset); 888 spin_unlock_irqrestore(host->host_lock, flags); 889 } 890 891 return rtn; 892 } 893 894 /** 895 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 896 * @scmd: SCSI cmd used to send BDR 897 * 898 * Notes: 899 * There is no timeout for this operation. if this operation is 900 * unreliable for a given host, then the host itself needs to put a 901 * timer on it, and set the host back to a consistent state prior to 902 * returning. 903 */ 904 static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 905 { 906 enum scsi_disposition rtn; 907 struct scsi_host_template *hostt = scmd->device->host->hostt; 908 909 if (!hostt->eh_device_reset_handler) 910 return FAILED; 911 912 rtn = hostt->eh_device_reset_handler(scmd); 913 if (rtn == SUCCESS) 914 __scsi_report_device_reset(scmd->device, NULL); 915 return rtn; 916 } 917 918 /** 919 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command 920 * @hostt: SCSI driver host template 921 * @scmd: SCSI cmd used to send a target reset 922 * 923 * Return value: 924 * SUCCESS, FAILED, or FAST_IO_FAIL 925 * 926 * Notes: 927 * SUCCESS does not necessarily indicate that the command 928 * has been aborted; it only indicates that the LLDDs 929 * has cleared all references to that command. 930 * LLDDs should return FAILED only if an abort was required 931 * but could not be executed. LLDDs should return FAST_IO_FAIL 932 * if the device is temporarily unavailable (eg due to a 933 * link down on FibreChannel) 934 */ 935 static enum scsi_disposition 936 scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd) 937 { 938 if (!hostt->eh_abort_handler) 939 return FAILED; 940 941 return hostt->eh_abort_handler(scmd); 942 } 943 944 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 945 { 946 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 947 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 948 if (scsi_try_target_reset(scmd) != SUCCESS) 949 if (scsi_try_bus_reset(scmd) != SUCCESS) 950 scsi_try_host_reset(scmd); 951 } 952 953 /** 954 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 955 * @scmd: SCSI command structure to hijack 956 * @ses: structure to save restore information 957 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 958 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) 959 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 960 * 961 * This function is used to save a scsi command information before re-execution 962 * as part of the error recovery process. If @sense_bytes is 0 the command 963 * sent must be one that does not transfer any data. If @sense_bytes != 0 964 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 965 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 966 */ 967 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 968 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 969 { 970 struct scsi_device *sdev = scmd->device; 971 972 /* 973 * We need saved copies of a number of fields - this is because 974 * error handling may need to overwrite these with different values 975 * to run different commands, and once error handling is complete, 976 * we will need to restore these values prior to running the actual 977 * command. 978 */ 979 ses->cmd_len = scmd->cmd_len; 980 ses->cmnd = scmd->cmnd; 981 ses->data_direction = scmd->sc_data_direction; 982 ses->sdb = scmd->sdb; 983 ses->result = scmd->result; 984 ses->resid_len = scmd->req.resid_len; 985 ses->underflow = scmd->underflow; 986 ses->prot_op = scmd->prot_op; 987 ses->eh_eflags = scmd->eh_eflags; 988 989 scmd->prot_op = SCSI_PROT_NORMAL; 990 scmd->eh_eflags = 0; 991 scmd->cmnd = ses->eh_cmnd; 992 memset(scmd->cmnd, 0, BLK_MAX_CDB); 993 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 994 scmd->result = 0; 995 scmd->req.resid_len = 0; 996 997 if (sense_bytes) { 998 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 999 sense_bytes); 1000 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 1001 scmd->sdb.length); 1002 scmd->sdb.table.sgl = &ses->sense_sgl; 1003 scmd->sc_data_direction = DMA_FROM_DEVICE; 1004 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; 1005 scmd->cmnd[0] = REQUEST_SENSE; 1006 scmd->cmnd[4] = scmd->sdb.length; 1007 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1008 } else { 1009 scmd->sc_data_direction = DMA_NONE; 1010 if (cmnd) { 1011 BUG_ON(cmnd_size > BLK_MAX_CDB); 1012 memcpy(scmd->cmnd, cmnd, cmnd_size); 1013 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1014 } 1015 } 1016 1017 scmd->underflow = 0; 1018 1019 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 1020 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 1021 (sdev->lun << 5 & 0xe0); 1022 1023 /* 1024 * Zero the sense buffer. The scsi spec mandates that any 1025 * untransferred sense data should be interpreted as being zero. 1026 */ 1027 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 1028 } 1029 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 1030 1031 /** 1032 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 1033 * @scmd: SCSI command structure to restore 1034 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 1035 * 1036 * Undo any damage done by above scsi_eh_prep_cmnd(). 1037 */ 1038 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 1039 { 1040 /* 1041 * Restore original data 1042 */ 1043 scmd->cmd_len = ses->cmd_len; 1044 scmd->cmnd = ses->cmnd; 1045 scmd->sc_data_direction = ses->data_direction; 1046 scmd->sdb = ses->sdb; 1047 scmd->result = ses->result; 1048 scmd->req.resid_len = ses->resid_len; 1049 scmd->underflow = ses->underflow; 1050 scmd->prot_op = ses->prot_op; 1051 scmd->eh_eflags = ses->eh_eflags; 1052 } 1053 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 1054 1055 /** 1056 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 1057 * @scmd: SCSI command structure to hijack 1058 * @cmnd: CDB to send 1059 * @cmnd_size: size in bytes of @cmnd 1060 * @timeout: timeout for this request 1061 * @sense_bytes: size of sense data to copy or 0 1062 * 1063 * This function is used to send a scsi command down to a target device 1064 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 1065 * 1066 * Return value: 1067 * SUCCESS or FAILED or NEEDS_RETRY 1068 */ 1069 static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd, 1070 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes) 1071 { 1072 struct scsi_device *sdev = scmd->device; 1073 struct Scsi_Host *shost = sdev->host; 1074 DECLARE_COMPLETION_ONSTACK(done); 1075 unsigned long timeleft = timeout, delay; 1076 struct scsi_eh_save ses; 1077 const unsigned long stall_for = msecs_to_jiffies(100); 1078 int rtn; 1079 1080 retry: 1081 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1082 shost->eh_action = &done; 1083 1084 scsi_log_send(scmd); 1085 scmd->scsi_done = scsi_eh_done; 1086 1087 /* 1088 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can 1089 * change the SCSI device state after we have examined it and before 1090 * .queuecommand() is called. 1091 */ 1092 mutex_lock(&sdev->state_mutex); 1093 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) { 1094 mutex_unlock(&sdev->state_mutex); 1095 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev, 1096 "%s: state %d <> %d\n", __func__, sdev->sdev_state, 1097 SDEV_BLOCK)); 1098 delay = min(timeleft, stall_for); 1099 timeleft -= delay; 1100 msleep(jiffies_to_msecs(delay)); 1101 mutex_lock(&sdev->state_mutex); 1102 } 1103 if (sdev->sdev_state != SDEV_BLOCK) 1104 rtn = shost->hostt->queuecommand(shost, scmd); 1105 else 1106 rtn = FAILED; 1107 mutex_unlock(&sdev->state_mutex); 1108 1109 if (rtn) { 1110 if (timeleft > stall_for) { 1111 scsi_eh_restore_cmnd(scmd, &ses); 1112 timeleft -= stall_for; 1113 msleep(jiffies_to_msecs(stall_for)); 1114 goto retry; 1115 } 1116 /* signal not to enter either branch of the if () below */ 1117 timeleft = 0; 1118 rtn = FAILED; 1119 } else { 1120 timeleft = wait_for_completion_timeout(&done, timeout); 1121 rtn = SUCCESS; 1122 } 1123 1124 shost->eh_action = NULL; 1125 1126 scsi_log_completion(scmd, rtn); 1127 1128 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1129 "%s timeleft: %ld\n", 1130 __func__, timeleft)); 1131 1132 /* 1133 * If there is time left scsi_eh_done got called, and we will examine 1134 * the actual status codes to see whether the command actually did 1135 * complete normally, else if we have a zero return and no time left, 1136 * the command must still be pending, so abort it and return FAILED. 1137 * If we never actually managed to issue the command, because 1138 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1139 * value above (so don't execute either branch of the if) 1140 */ 1141 if (timeleft) { 1142 rtn = scsi_eh_completed_normally(scmd); 1143 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1144 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1145 1146 switch (rtn) { 1147 case SUCCESS: 1148 case NEEDS_RETRY: 1149 case FAILED: 1150 break; 1151 case ADD_TO_MLQUEUE: 1152 rtn = NEEDS_RETRY; 1153 break; 1154 default: 1155 rtn = FAILED; 1156 break; 1157 } 1158 } else if (rtn != FAILED) { 1159 scsi_abort_eh_cmnd(scmd); 1160 rtn = FAILED; 1161 } 1162 1163 scsi_eh_restore_cmnd(scmd, &ses); 1164 1165 return rtn; 1166 } 1167 1168 /** 1169 * scsi_request_sense - Request sense data from a particular target. 1170 * @scmd: SCSI cmd for request sense. 1171 * 1172 * Notes: 1173 * Some hosts automatically obtain this information, others require 1174 * that we obtain it on our own. This function will *not* return until 1175 * the command either times out, or it completes. 1176 */ 1177 static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd) 1178 { 1179 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1180 } 1181 1182 static enum scsi_disposition 1183 scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn) 1184 { 1185 if (!blk_rq_is_passthrough(scmd->request)) { 1186 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1187 if (sdrv->eh_action) 1188 rtn = sdrv->eh_action(scmd, rtn); 1189 } 1190 return rtn; 1191 } 1192 1193 /** 1194 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1195 * @scmd: Original SCSI cmd that eh has finished. 1196 * @done_q: Queue for processed commands. 1197 * 1198 * Notes: 1199 * We don't want to use the normal command completion while we are are 1200 * still handling errors - it may cause other commands to be queued, 1201 * and that would disturb what we are doing. Thus we really want to 1202 * keep a list of pending commands for final completion, and once we 1203 * are ready to leave error handling we handle completion for real. 1204 */ 1205 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1206 { 1207 list_move_tail(&scmd->eh_entry, done_q); 1208 } 1209 EXPORT_SYMBOL(scsi_eh_finish_cmd); 1210 1211 /** 1212 * scsi_eh_get_sense - Get device sense data. 1213 * @work_q: Queue of commands to process. 1214 * @done_q: Queue of processed commands. 1215 * 1216 * Description: 1217 * See if we need to request sense information. if so, then get it 1218 * now, so we have a better idea of what to do. 1219 * 1220 * Notes: 1221 * This has the unfortunate side effect that if a shost adapter does 1222 * not automatically request sense information, we end up shutting 1223 * it down before we request it. 1224 * 1225 * All drivers should request sense information internally these days, 1226 * so for now all I have to say is tough noogies if you end up in here. 1227 * 1228 * XXX: Long term this code should go away, but that needs an audit of 1229 * all LLDDs first. 1230 */ 1231 int scsi_eh_get_sense(struct list_head *work_q, 1232 struct list_head *done_q) 1233 { 1234 struct scsi_cmnd *scmd, *next; 1235 struct Scsi_Host *shost; 1236 enum scsi_disposition rtn; 1237 1238 /* 1239 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, 1240 * should not get sense. 1241 */ 1242 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1243 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || 1244 SCSI_SENSE_VALID(scmd)) 1245 continue; 1246 1247 shost = scmd->device->host; 1248 if (scsi_host_eh_past_deadline(shost)) { 1249 SCSI_LOG_ERROR_RECOVERY(3, 1250 scmd_printk(KERN_INFO, scmd, 1251 "%s: skip request sense, past eh deadline\n", 1252 current->comm)); 1253 break; 1254 } 1255 if (!scsi_status_is_check_condition(scmd->result)) 1256 /* 1257 * don't request sense if there's no check condition 1258 * status because the error we're processing isn't one 1259 * that has a sense code (and some devices get 1260 * confused by sense requests out of the blue) 1261 */ 1262 continue; 1263 1264 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1265 "%s: requesting sense\n", 1266 current->comm)); 1267 rtn = scsi_request_sense(scmd); 1268 if (rtn != SUCCESS) 1269 continue; 1270 1271 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1272 "sense requested, result %x\n", scmd->result)); 1273 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); 1274 1275 rtn = scsi_decide_disposition(scmd); 1276 1277 /* 1278 * if the result was normal, then just pass it along to the 1279 * upper level. 1280 */ 1281 if (rtn == SUCCESS) 1282 /* 1283 * We don't want this command reissued, just finished 1284 * with the sense data, so set retries to the max 1285 * allowed to ensure it won't get reissued. If the user 1286 * has requested infinite retries, we also want to 1287 * finish this command, so force completion by setting 1288 * retries and allowed to the same value. 1289 */ 1290 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) 1291 scmd->retries = scmd->allowed = 1; 1292 else 1293 scmd->retries = scmd->allowed; 1294 else if (rtn != NEEDS_RETRY) 1295 continue; 1296 1297 scsi_eh_finish_cmd(scmd, done_q); 1298 } 1299 1300 return list_empty(work_q); 1301 } 1302 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1303 1304 /** 1305 * scsi_eh_tur - Send TUR to device. 1306 * @scmd: &scsi_cmnd to send TUR 1307 * 1308 * Return value: 1309 * 0 - Device is ready. 1 - Device NOT ready. 1310 */ 1311 static int scsi_eh_tur(struct scsi_cmnd *scmd) 1312 { 1313 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1314 int retry_cnt = 1; 1315 enum scsi_disposition rtn; 1316 1317 retry_tur: 1318 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1319 scmd->device->eh_timeout, 0); 1320 1321 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1322 "%s return: %x\n", __func__, rtn)); 1323 1324 switch (rtn) { 1325 case NEEDS_RETRY: 1326 if (retry_cnt--) 1327 goto retry_tur; 1328 fallthrough; 1329 case SUCCESS: 1330 return 0; 1331 default: 1332 return 1; 1333 } 1334 } 1335 1336 /** 1337 * scsi_eh_test_devices - check if devices are responding from error recovery. 1338 * @cmd_list: scsi commands in error recovery. 1339 * @work_q: queue for commands which still need more error recovery 1340 * @done_q: queue for commands which are finished 1341 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1342 * 1343 * Decription: 1344 * Tests if devices are in a working state. Commands to devices now in 1345 * a working state are sent to the done_q while commands to devices which 1346 * are still failing to respond are returned to the work_q for more 1347 * processing. 1348 **/ 1349 static int scsi_eh_test_devices(struct list_head *cmd_list, 1350 struct list_head *work_q, 1351 struct list_head *done_q, int try_stu) 1352 { 1353 struct scsi_cmnd *scmd, *next; 1354 struct scsi_device *sdev; 1355 int finish_cmds; 1356 1357 while (!list_empty(cmd_list)) { 1358 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1359 sdev = scmd->device; 1360 1361 if (!try_stu) { 1362 if (scsi_host_eh_past_deadline(sdev->host)) { 1363 /* Push items back onto work_q */ 1364 list_splice_init(cmd_list, work_q); 1365 SCSI_LOG_ERROR_RECOVERY(3, 1366 sdev_printk(KERN_INFO, sdev, 1367 "%s: skip test device, past eh deadline", 1368 current->comm)); 1369 break; 1370 } 1371 } 1372 1373 finish_cmds = !scsi_device_online(scmd->device) || 1374 (try_stu && !scsi_eh_try_stu(scmd) && 1375 !scsi_eh_tur(scmd)) || 1376 !scsi_eh_tur(scmd); 1377 1378 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1379 if (scmd->device == sdev) { 1380 if (finish_cmds && 1381 (try_stu || 1382 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1383 scsi_eh_finish_cmd(scmd, done_q); 1384 else 1385 list_move_tail(&scmd->eh_entry, work_q); 1386 } 1387 } 1388 return list_empty(work_q); 1389 } 1390 1391 /** 1392 * scsi_eh_try_stu - Send START_UNIT to device. 1393 * @scmd: &scsi_cmnd to send START_UNIT 1394 * 1395 * Return value: 1396 * 0 - Device is ready. 1 - Device NOT ready. 1397 */ 1398 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1399 { 1400 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1401 1402 if (scmd->device->allow_restart) { 1403 int i; 1404 enum scsi_disposition rtn = NEEDS_RETRY; 1405 1406 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1407 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0); 1408 1409 if (rtn == SUCCESS) 1410 return 0; 1411 } 1412 1413 return 1; 1414 } 1415 1416 /** 1417 * scsi_eh_stu - send START_UNIT if needed 1418 * @shost: &scsi host being recovered. 1419 * @work_q: &list_head for pending commands. 1420 * @done_q: &list_head for processed commands. 1421 * 1422 * Notes: 1423 * If commands are failing due to not ready, initializing command required, 1424 * try revalidating the device, which will end up sending a start unit. 1425 */ 1426 static int scsi_eh_stu(struct Scsi_Host *shost, 1427 struct list_head *work_q, 1428 struct list_head *done_q) 1429 { 1430 struct scsi_cmnd *scmd, *stu_scmd, *next; 1431 struct scsi_device *sdev; 1432 1433 shost_for_each_device(sdev, shost) { 1434 if (scsi_host_eh_past_deadline(shost)) { 1435 SCSI_LOG_ERROR_RECOVERY(3, 1436 sdev_printk(KERN_INFO, sdev, 1437 "%s: skip START_UNIT, past eh deadline\n", 1438 current->comm)); 1439 scsi_device_put(sdev); 1440 break; 1441 } 1442 stu_scmd = NULL; 1443 list_for_each_entry(scmd, work_q, eh_entry) 1444 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1445 scsi_check_sense(scmd) == FAILED ) { 1446 stu_scmd = scmd; 1447 break; 1448 } 1449 1450 if (!stu_scmd) 1451 continue; 1452 1453 SCSI_LOG_ERROR_RECOVERY(3, 1454 sdev_printk(KERN_INFO, sdev, 1455 "%s: Sending START_UNIT\n", 1456 current->comm)); 1457 1458 if (!scsi_eh_try_stu(stu_scmd)) { 1459 if (!scsi_device_online(sdev) || 1460 !scsi_eh_tur(stu_scmd)) { 1461 list_for_each_entry_safe(scmd, next, 1462 work_q, eh_entry) { 1463 if (scmd->device == sdev && 1464 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1465 scsi_eh_finish_cmd(scmd, done_q); 1466 } 1467 } 1468 } else { 1469 SCSI_LOG_ERROR_RECOVERY(3, 1470 sdev_printk(KERN_INFO, sdev, 1471 "%s: START_UNIT failed\n", 1472 current->comm)); 1473 } 1474 } 1475 1476 return list_empty(work_q); 1477 } 1478 1479 1480 /** 1481 * scsi_eh_bus_device_reset - send bdr if needed 1482 * @shost: scsi host being recovered. 1483 * @work_q: &list_head for pending commands. 1484 * @done_q: &list_head for processed commands. 1485 * 1486 * Notes: 1487 * Try a bus device reset. Still, look to see whether we have multiple 1488 * devices that are jammed or not - if we have multiple devices, it 1489 * makes no sense to try bus_device_reset - we really would need to try 1490 * a bus_reset instead. 1491 */ 1492 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1493 struct list_head *work_q, 1494 struct list_head *done_q) 1495 { 1496 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1497 struct scsi_device *sdev; 1498 enum scsi_disposition rtn; 1499 1500 shost_for_each_device(sdev, shost) { 1501 if (scsi_host_eh_past_deadline(shost)) { 1502 SCSI_LOG_ERROR_RECOVERY(3, 1503 sdev_printk(KERN_INFO, sdev, 1504 "%s: skip BDR, past eh deadline\n", 1505 current->comm)); 1506 scsi_device_put(sdev); 1507 break; 1508 } 1509 bdr_scmd = NULL; 1510 list_for_each_entry(scmd, work_q, eh_entry) 1511 if (scmd->device == sdev) { 1512 bdr_scmd = scmd; 1513 break; 1514 } 1515 1516 if (!bdr_scmd) 1517 continue; 1518 1519 SCSI_LOG_ERROR_RECOVERY(3, 1520 sdev_printk(KERN_INFO, sdev, 1521 "%s: Sending BDR\n", current->comm)); 1522 rtn = scsi_try_bus_device_reset(bdr_scmd); 1523 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1524 if (!scsi_device_online(sdev) || 1525 rtn == FAST_IO_FAIL || 1526 !scsi_eh_tur(bdr_scmd)) { 1527 list_for_each_entry_safe(scmd, next, 1528 work_q, eh_entry) { 1529 if (scmd->device == sdev && 1530 scsi_eh_action(scmd, rtn) != FAILED) 1531 scsi_eh_finish_cmd(scmd, 1532 done_q); 1533 } 1534 } 1535 } else { 1536 SCSI_LOG_ERROR_RECOVERY(3, 1537 sdev_printk(KERN_INFO, sdev, 1538 "%s: BDR failed\n", current->comm)); 1539 } 1540 } 1541 1542 return list_empty(work_q); 1543 } 1544 1545 /** 1546 * scsi_eh_target_reset - send target reset if needed 1547 * @shost: scsi host being recovered. 1548 * @work_q: &list_head for pending commands. 1549 * @done_q: &list_head for processed commands. 1550 * 1551 * Notes: 1552 * Try a target reset. 1553 */ 1554 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1555 struct list_head *work_q, 1556 struct list_head *done_q) 1557 { 1558 LIST_HEAD(tmp_list); 1559 LIST_HEAD(check_list); 1560 1561 list_splice_init(work_q, &tmp_list); 1562 1563 while (!list_empty(&tmp_list)) { 1564 struct scsi_cmnd *next, *scmd; 1565 enum scsi_disposition rtn; 1566 unsigned int id; 1567 1568 if (scsi_host_eh_past_deadline(shost)) { 1569 /* push back on work queue for further processing */ 1570 list_splice_init(&check_list, work_q); 1571 list_splice_init(&tmp_list, work_q); 1572 SCSI_LOG_ERROR_RECOVERY(3, 1573 shost_printk(KERN_INFO, shost, 1574 "%s: Skip target reset, past eh deadline\n", 1575 current->comm)); 1576 return list_empty(work_q); 1577 } 1578 1579 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1580 id = scmd_id(scmd); 1581 1582 SCSI_LOG_ERROR_RECOVERY(3, 1583 shost_printk(KERN_INFO, shost, 1584 "%s: Sending target reset to target %d\n", 1585 current->comm, id)); 1586 rtn = scsi_try_target_reset(scmd); 1587 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1588 SCSI_LOG_ERROR_RECOVERY(3, 1589 shost_printk(KERN_INFO, shost, 1590 "%s: Target reset failed" 1591 " target: %d\n", 1592 current->comm, id)); 1593 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1594 if (scmd_id(scmd) != id) 1595 continue; 1596 1597 if (rtn == SUCCESS) 1598 list_move_tail(&scmd->eh_entry, &check_list); 1599 else if (rtn == FAST_IO_FAIL) 1600 scsi_eh_finish_cmd(scmd, done_q); 1601 else 1602 /* push back on work queue for further processing */ 1603 list_move(&scmd->eh_entry, work_q); 1604 } 1605 } 1606 1607 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1608 } 1609 1610 /** 1611 * scsi_eh_bus_reset - send a bus reset 1612 * @shost: &scsi host being recovered. 1613 * @work_q: &list_head for pending commands. 1614 * @done_q: &list_head for processed commands. 1615 */ 1616 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1617 struct list_head *work_q, 1618 struct list_head *done_q) 1619 { 1620 struct scsi_cmnd *scmd, *chan_scmd, *next; 1621 LIST_HEAD(check_list); 1622 unsigned int channel; 1623 enum scsi_disposition rtn; 1624 1625 /* 1626 * we really want to loop over the various channels, and do this on 1627 * a channel by channel basis. we should also check to see if any 1628 * of the failed commands are on soft_reset devices, and if so, skip 1629 * the reset. 1630 */ 1631 1632 for (channel = 0; channel <= shost->max_channel; channel++) { 1633 if (scsi_host_eh_past_deadline(shost)) { 1634 list_splice_init(&check_list, work_q); 1635 SCSI_LOG_ERROR_RECOVERY(3, 1636 shost_printk(KERN_INFO, shost, 1637 "%s: skip BRST, past eh deadline\n", 1638 current->comm)); 1639 return list_empty(work_q); 1640 } 1641 1642 chan_scmd = NULL; 1643 list_for_each_entry(scmd, work_q, eh_entry) { 1644 if (channel == scmd_channel(scmd)) { 1645 chan_scmd = scmd; 1646 break; 1647 /* 1648 * FIXME add back in some support for 1649 * soft_reset devices. 1650 */ 1651 } 1652 } 1653 1654 if (!chan_scmd) 1655 continue; 1656 SCSI_LOG_ERROR_RECOVERY(3, 1657 shost_printk(KERN_INFO, shost, 1658 "%s: Sending BRST chan: %d\n", 1659 current->comm, channel)); 1660 rtn = scsi_try_bus_reset(chan_scmd); 1661 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1662 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1663 if (channel == scmd_channel(scmd)) { 1664 if (rtn == FAST_IO_FAIL) 1665 scsi_eh_finish_cmd(scmd, 1666 done_q); 1667 else 1668 list_move_tail(&scmd->eh_entry, 1669 &check_list); 1670 } 1671 } 1672 } else { 1673 SCSI_LOG_ERROR_RECOVERY(3, 1674 shost_printk(KERN_INFO, shost, 1675 "%s: BRST failed chan: %d\n", 1676 current->comm, channel)); 1677 } 1678 } 1679 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1680 } 1681 1682 /** 1683 * scsi_eh_host_reset - send a host reset 1684 * @shost: host to be reset. 1685 * @work_q: &list_head for pending commands. 1686 * @done_q: &list_head for processed commands. 1687 */ 1688 static int scsi_eh_host_reset(struct Scsi_Host *shost, 1689 struct list_head *work_q, 1690 struct list_head *done_q) 1691 { 1692 struct scsi_cmnd *scmd, *next; 1693 LIST_HEAD(check_list); 1694 enum scsi_disposition rtn; 1695 1696 if (!list_empty(work_q)) { 1697 scmd = list_entry(work_q->next, 1698 struct scsi_cmnd, eh_entry); 1699 1700 SCSI_LOG_ERROR_RECOVERY(3, 1701 shost_printk(KERN_INFO, shost, 1702 "%s: Sending HRST\n", 1703 current->comm)); 1704 1705 rtn = scsi_try_host_reset(scmd); 1706 if (rtn == SUCCESS) { 1707 list_splice_init(work_q, &check_list); 1708 } else if (rtn == FAST_IO_FAIL) { 1709 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1710 scsi_eh_finish_cmd(scmd, done_q); 1711 } 1712 } else { 1713 SCSI_LOG_ERROR_RECOVERY(3, 1714 shost_printk(KERN_INFO, shost, 1715 "%s: HRST failed\n", 1716 current->comm)); 1717 } 1718 } 1719 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1720 } 1721 1722 /** 1723 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1724 * @work_q: &list_head for pending commands. 1725 * @done_q: &list_head for processed commands. 1726 */ 1727 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1728 struct list_head *done_q) 1729 { 1730 struct scsi_cmnd *scmd, *next; 1731 struct scsi_device *sdev; 1732 1733 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1734 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1735 "not ready after error recovery\n"); 1736 sdev = scmd->device; 1737 1738 mutex_lock(&sdev->state_mutex); 1739 scsi_device_set_state(sdev, SDEV_OFFLINE); 1740 mutex_unlock(&sdev->state_mutex); 1741 1742 scsi_eh_finish_cmd(scmd, done_q); 1743 } 1744 return; 1745 } 1746 1747 /** 1748 * scsi_noretry_cmd - determine if command should be failed fast 1749 * @scmd: SCSI cmd to examine. 1750 */ 1751 int scsi_noretry_cmd(struct scsi_cmnd *scmd) 1752 { 1753 switch (host_byte(scmd->result)) { 1754 case DID_OK: 1755 break; 1756 case DID_TIME_OUT: 1757 goto check_type; 1758 case DID_BUS_BUSY: 1759 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT); 1760 case DID_PARITY: 1761 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV); 1762 case DID_ERROR: 1763 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT) 1764 return 0; 1765 fallthrough; 1766 case DID_SOFT_ERROR: 1767 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER); 1768 } 1769 1770 if (!scsi_status_is_check_condition(scmd->result)) 1771 return 0; 1772 1773 check_type: 1774 /* 1775 * assume caller has checked sense and determined 1776 * the check condition was retryable. 1777 */ 1778 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV || 1779 blk_rq_is_passthrough(scmd->request)) 1780 return 1; 1781 1782 return 0; 1783 } 1784 1785 /** 1786 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1787 * @scmd: SCSI cmd to examine. 1788 * 1789 * Notes: 1790 * This is *only* called when we are examining the status after sending 1791 * out the actual data command. any commands that are queued for error 1792 * recovery (e.g. test_unit_ready) do *not* come through here. 1793 * 1794 * When this routine returns failed, it means the error handler thread 1795 * is woken. In cases where the error code indicates an error that 1796 * doesn't require the error handler read (i.e. we don't need to 1797 * abort/reset), this function should return SUCCESS. 1798 */ 1799 enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd) 1800 { 1801 enum scsi_disposition rtn; 1802 1803 /* 1804 * if the device is offline, then we clearly just pass the result back 1805 * up to the top level. 1806 */ 1807 if (!scsi_device_online(scmd->device)) { 1808 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1809 "%s: device offline - report as SUCCESS\n", __func__)); 1810 return SUCCESS; 1811 } 1812 1813 /* 1814 * first check the host byte, to see if there is anything in there 1815 * that would indicate what we need to do. 1816 */ 1817 switch (host_byte(scmd->result)) { 1818 case DID_PASSTHROUGH: 1819 /* 1820 * no matter what, pass this through to the upper layer. 1821 * nuke this special code so that it looks like we are saying 1822 * did_ok. 1823 */ 1824 scmd->result &= 0xff00ffff; 1825 return SUCCESS; 1826 case DID_OK: 1827 /* 1828 * looks good. drop through, and check the next byte. 1829 */ 1830 break; 1831 case DID_ABORT: 1832 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1833 set_host_byte(scmd, DID_TIME_OUT); 1834 return SUCCESS; 1835 } 1836 fallthrough; 1837 case DID_NO_CONNECT: 1838 case DID_BAD_TARGET: 1839 /* 1840 * note - this means that we just report the status back 1841 * to the top level driver, not that we actually think 1842 * that it indicates SUCCESS. 1843 */ 1844 return SUCCESS; 1845 case DID_SOFT_ERROR: 1846 /* 1847 * when the low level driver returns did_soft_error, 1848 * it is responsible for keeping an internal retry counter 1849 * in order to avoid endless loops (db) 1850 */ 1851 goto maybe_retry; 1852 case DID_IMM_RETRY: 1853 return NEEDS_RETRY; 1854 1855 case DID_REQUEUE: 1856 return ADD_TO_MLQUEUE; 1857 case DID_TRANSPORT_DISRUPTED: 1858 /* 1859 * LLD/transport was disrupted during processing of the IO. 1860 * The transport class is now blocked/blocking, 1861 * and the transport will decide what to do with the IO 1862 * based on its timers and recovery capablilities if 1863 * there are enough retries. 1864 */ 1865 goto maybe_retry; 1866 case DID_TRANSPORT_FAILFAST: 1867 /* 1868 * The transport decided to failfast the IO (most likely 1869 * the fast io fail tmo fired), so send IO directly upwards. 1870 */ 1871 return SUCCESS; 1872 case DID_TRANSPORT_MARGINAL: 1873 /* 1874 * caller has decided not to do retries on 1875 * abort success, so send IO directly upwards 1876 */ 1877 return SUCCESS; 1878 case DID_ERROR: 1879 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT) 1880 /* 1881 * execute reservation conflict processing code 1882 * lower down 1883 */ 1884 break; 1885 fallthrough; 1886 case DID_BUS_BUSY: 1887 case DID_PARITY: 1888 goto maybe_retry; 1889 case DID_TIME_OUT: 1890 /* 1891 * when we scan the bus, we get timeout messages for 1892 * these commands if there is no device available. 1893 * other hosts report did_no_connect for the same thing. 1894 */ 1895 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1896 scmd->cmnd[0] == INQUIRY)) { 1897 return SUCCESS; 1898 } else { 1899 return FAILED; 1900 } 1901 case DID_RESET: 1902 return SUCCESS; 1903 default: 1904 return FAILED; 1905 } 1906 1907 /* 1908 * check the status byte to see if this indicates anything special. 1909 */ 1910 switch (get_status_byte(scmd)) { 1911 case SAM_STAT_TASK_SET_FULL: 1912 scsi_handle_queue_full(scmd->device); 1913 /* 1914 * the case of trying to send too many commands to a 1915 * tagged queueing device. 1916 */ 1917 fallthrough; 1918 case SAM_STAT_BUSY: 1919 /* 1920 * device can't talk to us at the moment. Should only 1921 * occur (SAM-3) when the task queue is empty, so will cause 1922 * the empty queue handling to trigger a stall in the 1923 * device. 1924 */ 1925 return ADD_TO_MLQUEUE; 1926 case SAM_STAT_GOOD: 1927 if (scmd->cmnd[0] == REPORT_LUNS) 1928 scmd->device->sdev_target->expecting_lun_change = 0; 1929 scsi_handle_queue_ramp_up(scmd->device); 1930 fallthrough; 1931 case SAM_STAT_COMMAND_TERMINATED: 1932 return SUCCESS; 1933 case SAM_STAT_TASK_ABORTED: 1934 goto maybe_retry; 1935 case SAM_STAT_CHECK_CONDITION: 1936 rtn = scsi_check_sense(scmd); 1937 if (rtn == NEEDS_RETRY) 1938 goto maybe_retry; 1939 /* if rtn == FAILED, we have no sense information; 1940 * returning FAILED will wake the error handler thread 1941 * to collect the sense and redo the decide 1942 * disposition */ 1943 return rtn; 1944 case SAM_STAT_CONDITION_MET: 1945 case SAM_STAT_INTERMEDIATE: 1946 case SAM_STAT_INTERMEDIATE_CONDITION_MET: 1947 case SAM_STAT_ACA_ACTIVE: 1948 /* 1949 * who knows? FIXME(eric) 1950 */ 1951 return SUCCESS; 1952 1953 case SAM_STAT_RESERVATION_CONFLICT: 1954 sdev_printk(KERN_INFO, scmd->device, 1955 "reservation conflict\n"); 1956 set_host_byte(scmd, DID_NEXUS_FAILURE); 1957 return SUCCESS; /* causes immediate i/o error */ 1958 default: 1959 return FAILED; 1960 } 1961 return FAILED; 1962 1963 maybe_retry: 1964 1965 /* we requeue for retry because the error was retryable, and 1966 * the request was not marked fast fail. Note that above, 1967 * even if the request is marked fast fail, we still requeue 1968 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1969 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) { 1970 return NEEDS_RETRY; 1971 } else { 1972 /* 1973 * no more retries - report this one back to upper level. 1974 */ 1975 return SUCCESS; 1976 } 1977 } 1978 1979 static void eh_lock_door_done(struct request *req, blk_status_t status) 1980 { 1981 blk_put_request(req); 1982 } 1983 1984 /** 1985 * scsi_eh_lock_door - Prevent medium removal for the specified device 1986 * @sdev: SCSI device to prevent medium removal 1987 * 1988 * Locking: 1989 * We must be called from process context. 1990 * 1991 * Notes: 1992 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 1993 * head of the devices request queue, and continue. 1994 */ 1995 static void scsi_eh_lock_door(struct scsi_device *sdev) 1996 { 1997 struct request *req; 1998 struct scsi_request *rq; 1999 2000 req = blk_get_request(sdev->request_queue, REQ_OP_DRV_IN, 0); 2001 if (IS_ERR(req)) 2002 return; 2003 rq = scsi_req(req); 2004 2005 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL; 2006 rq->cmd[1] = 0; 2007 rq->cmd[2] = 0; 2008 rq->cmd[3] = 0; 2009 rq->cmd[4] = SCSI_REMOVAL_PREVENT; 2010 rq->cmd[5] = 0; 2011 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]); 2012 2013 req->rq_flags |= RQF_QUIET; 2014 req->timeout = 10 * HZ; 2015 rq->retries = 5; 2016 2017 blk_execute_rq_nowait(NULL, req, 1, eh_lock_door_done); 2018 } 2019 2020 /** 2021 * scsi_restart_operations - restart io operations to the specified host. 2022 * @shost: Host we are restarting. 2023 * 2024 * Notes: 2025 * When we entered the error handler, we blocked all further i/o to 2026 * this device. we need to 'reverse' this process. 2027 */ 2028 static void scsi_restart_operations(struct Scsi_Host *shost) 2029 { 2030 struct scsi_device *sdev; 2031 unsigned long flags; 2032 2033 /* 2034 * If the door was locked, we need to insert a door lock request 2035 * onto the head of the SCSI request queue for the device. There 2036 * is no point trying to lock the door of an off-line device. 2037 */ 2038 shost_for_each_device(sdev, shost) { 2039 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 2040 scsi_eh_lock_door(sdev); 2041 sdev->was_reset = 0; 2042 } 2043 } 2044 2045 /* 2046 * next free up anything directly waiting upon the host. this 2047 * will be requests for character device operations, and also for 2048 * ioctls to queued block devices. 2049 */ 2050 SCSI_LOG_ERROR_RECOVERY(3, 2051 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 2052 2053 spin_lock_irqsave(shost->host_lock, flags); 2054 if (scsi_host_set_state(shost, SHOST_RUNNING)) 2055 if (scsi_host_set_state(shost, SHOST_CANCEL)) 2056 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 2057 spin_unlock_irqrestore(shost->host_lock, flags); 2058 2059 wake_up(&shost->host_wait); 2060 2061 /* 2062 * finally we need to re-initiate requests that may be pending. we will 2063 * have had everything blocked while error handling is taking place, and 2064 * now that error recovery is done, we will need to ensure that these 2065 * requests are started. 2066 */ 2067 scsi_run_host_queues(shost); 2068 2069 /* 2070 * if eh is active and host_eh_scheduled is pending we need to re-run 2071 * recovery. we do this check after scsi_run_host_queues() to allow 2072 * everything pent up since the last eh run a chance to make forward 2073 * progress before we sync again. Either we'll immediately re-run 2074 * recovery or scsi_device_unbusy() will wake us again when these 2075 * pending commands complete. 2076 */ 2077 spin_lock_irqsave(shost->host_lock, flags); 2078 if (shost->host_eh_scheduled) 2079 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 2080 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 2081 spin_unlock_irqrestore(shost->host_lock, flags); 2082 } 2083 2084 /** 2085 * scsi_eh_ready_devs - check device ready state and recover if not. 2086 * @shost: host to be recovered. 2087 * @work_q: &list_head for pending commands. 2088 * @done_q: &list_head for processed commands. 2089 */ 2090 void scsi_eh_ready_devs(struct Scsi_Host *shost, 2091 struct list_head *work_q, 2092 struct list_head *done_q) 2093 { 2094 if (!scsi_eh_stu(shost, work_q, done_q)) 2095 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2096 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2097 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2098 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2099 scsi_eh_offline_sdevs(work_q, 2100 done_q); 2101 } 2102 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2103 2104 /** 2105 * scsi_eh_flush_done_q - finish processed commands or retry them. 2106 * @done_q: list_head of processed commands. 2107 */ 2108 void scsi_eh_flush_done_q(struct list_head *done_q) 2109 { 2110 struct scsi_cmnd *scmd, *next; 2111 2112 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2113 list_del_init(&scmd->eh_entry); 2114 if (scsi_device_online(scmd->device) && 2115 !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd) && 2116 scsi_eh_should_retry_cmd(scmd)) { 2117 SCSI_LOG_ERROR_RECOVERY(3, 2118 scmd_printk(KERN_INFO, scmd, 2119 "%s: flush retry cmd\n", 2120 current->comm)); 2121 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2122 } else { 2123 /* 2124 * If just we got sense for the device (called 2125 * scsi_eh_get_sense), scmd->result is already 2126 * set, do not set DID_TIME_OUT. 2127 */ 2128 if (!scmd->result) 2129 scmd->result |= (DID_TIME_OUT << 16); 2130 SCSI_LOG_ERROR_RECOVERY(3, 2131 scmd_printk(KERN_INFO, scmd, 2132 "%s: flush finish cmd\n", 2133 current->comm)); 2134 scsi_finish_command(scmd); 2135 } 2136 } 2137 } 2138 EXPORT_SYMBOL(scsi_eh_flush_done_q); 2139 2140 /** 2141 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2142 * @shost: Host to unjam. 2143 * 2144 * Notes: 2145 * When we come in here, we *know* that all commands on the bus have 2146 * either completed, failed or timed out. we also know that no further 2147 * commands are being sent to the host, so things are relatively quiet 2148 * and we have freedom to fiddle with things as we wish. 2149 * 2150 * This is only the *default* implementation. it is possible for 2151 * individual drivers to supply their own version of this function, and 2152 * if the maintainer wishes to do this, it is strongly suggested that 2153 * this function be taken as a template and modified. this function 2154 * was designed to correctly handle problems for about 95% of the 2155 * different cases out there, and it should always provide at least a 2156 * reasonable amount of error recovery. 2157 * 2158 * Any command marked 'failed' or 'timeout' must eventually have 2159 * scsi_finish_cmd() called for it. we do all of the retry stuff 2160 * here, so when we restart the host after we return it should have an 2161 * empty queue. 2162 */ 2163 static void scsi_unjam_host(struct Scsi_Host *shost) 2164 { 2165 unsigned long flags; 2166 LIST_HEAD(eh_work_q); 2167 LIST_HEAD(eh_done_q); 2168 2169 spin_lock_irqsave(shost->host_lock, flags); 2170 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2171 spin_unlock_irqrestore(shost->host_lock, flags); 2172 2173 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2174 2175 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2176 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2177 2178 spin_lock_irqsave(shost->host_lock, flags); 2179 if (shost->eh_deadline != -1) 2180 shost->last_reset = 0; 2181 spin_unlock_irqrestore(shost->host_lock, flags); 2182 scsi_eh_flush_done_q(&eh_done_q); 2183 } 2184 2185 /** 2186 * scsi_error_handler - SCSI error handler thread 2187 * @data: Host for which we are running. 2188 * 2189 * Notes: 2190 * This is the main error handling loop. This is run as a kernel thread 2191 * for every SCSI host and handles all error handling activity. 2192 */ 2193 int scsi_error_handler(void *data) 2194 { 2195 struct Scsi_Host *shost = data; 2196 2197 /* 2198 * We use TASK_INTERRUPTIBLE so that the thread is not 2199 * counted against the load average as a running process. 2200 * We never actually get interrupted because kthread_run 2201 * disables signal delivery for the created thread. 2202 */ 2203 while (true) { 2204 /* 2205 * The sequence in kthread_stop() sets the stop flag first 2206 * then wakes the process. To avoid missed wakeups, the task 2207 * should always be in a non running state before the stop 2208 * flag is checked 2209 */ 2210 set_current_state(TASK_INTERRUPTIBLE); 2211 if (kthread_should_stop()) 2212 break; 2213 2214 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2215 shost->host_failed != scsi_host_busy(shost)) { 2216 SCSI_LOG_ERROR_RECOVERY(1, 2217 shost_printk(KERN_INFO, shost, 2218 "scsi_eh_%d: sleeping\n", 2219 shost->host_no)); 2220 schedule(); 2221 continue; 2222 } 2223 2224 __set_current_state(TASK_RUNNING); 2225 SCSI_LOG_ERROR_RECOVERY(1, 2226 shost_printk(KERN_INFO, shost, 2227 "scsi_eh_%d: waking up %d/%d/%d\n", 2228 shost->host_no, shost->host_eh_scheduled, 2229 shost->host_failed, 2230 scsi_host_busy(shost))); 2231 2232 /* 2233 * We have a host that is failing for some reason. Figure out 2234 * what we need to do to get it up and online again (if we can). 2235 * If we fail, we end up taking the thing offline. 2236 */ 2237 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { 2238 SCSI_LOG_ERROR_RECOVERY(1, 2239 shost_printk(KERN_ERR, shost, 2240 "scsi_eh_%d: unable to autoresume\n", 2241 shost->host_no)); 2242 continue; 2243 } 2244 2245 if (shost->transportt->eh_strategy_handler) 2246 shost->transportt->eh_strategy_handler(shost); 2247 else 2248 scsi_unjam_host(shost); 2249 2250 /* All scmds have been handled */ 2251 shost->host_failed = 0; 2252 2253 /* 2254 * Note - if the above fails completely, the action is to take 2255 * individual devices offline and flush the queue of any 2256 * outstanding requests that may have been pending. When we 2257 * restart, we restart any I/O to any other devices on the bus 2258 * which are still online. 2259 */ 2260 scsi_restart_operations(shost); 2261 if (!shost->eh_noresume) 2262 scsi_autopm_put_host(shost); 2263 } 2264 __set_current_state(TASK_RUNNING); 2265 2266 SCSI_LOG_ERROR_RECOVERY(1, 2267 shost_printk(KERN_INFO, shost, 2268 "Error handler scsi_eh_%d exiting\n", 2269 shost->host_no)); 2270 shost->ehandler = NULL; 2271 return 0; 2272 } 2273 2274 /* 2275 * Function: scsi_report_bus_reset() 2276 * 2277 * Purpose: Utility function used by low-level drivers to report that 2278 * they have observed a bus reset on the bus being handled. 2279 * 2280 * Arguments: shost - Host in question 2281 * channel - channel on which reset was observed. 2282 * 2283 * Returns: Nothing 2284 * 2285 * Lock status: Host lock must be held. 2286 * 2287 * Notes: This only needs to be called if the reset is one which 2288 * originates from an unknown location. Resets originated 2289 * by the mid-level itself don't need to call this, but there 2290 * should be no harm. 2291 * 2292 * The main purpose of this is to make sure that a CHECK_CONDITION 2293 * is properly treated. 2294 */ 2295 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2296 { 2297 struct scsi_device *sdev; 2298 2299 __shost_for_each_device(sdev, shost) { 2300 if (channel == sdev_channel(sdev)) 2301 __scsi_report_device_reset(sdev, NULL); 2302 } 2303 } 2304 EXPORT_SYMBOL(scsi_report_bus_reset); 2305 2306 /* 2307 * Function: scsi_report_device_reset() 2308 * 2309 * Purpose: Utility function used by low-level drivers to report that 2310 * they have observed a device reset on the device being handled. 2311 * 2312 * Arguments: shost - Host in question 2313 * channel - channel on which reset was observed 2314 * target - target on which reset was observed 2315 * 2316 * Returns: Nothing 2317 * 2318 * Lock status: Host lock must be held 2319 * 2320 * Notes: This only needs to be called if the reset is one which 2321 * originates from an unknown location. Resets originated 2322 * by the mid-level itself don't need to call this, but there 2323 * should be no harm. 2324 * 2325 * The main purpose of this is to make sure that a CHECK_CONDITION 2326 * is properly treated. 2327 */ 2328 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2329 { 2330 struct scsi_device *sdev; 2331 2332 __shost_for_each_device(sdev, shost) { 2333 if (channel == sdev_channel(sdev) && 2334 target == sdev_id(sdev)) 2335 __scsi_report_device_reset(sdev, NULL); 2336 } 2337 } 2338 EXPORT_SYMBOL(scsi_report_device_reset); 2339 2340 static void 2341 scsi_reset_provider_done_command(struct scsi_cmnd *scmd) 2342 { 2343 } 2344 2345 /** 2346 * scsi_ioctl_reset: explicitly reset a host/bus/target/device 2347 * @dev: scsi_device to operate on 2348 * @arg: reset type (see sg.h) 2349 */ 2350 int 2351 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) 2352 { 2353 struct scsi_cmnd *scmd; 2354 struct Scsi_Host *shost = dev->host; 2355 struct request *rq; 2356 unsigned long flags; 2357 int error = 0, val; 2358 enum scsi_disposition rtn; 2359 2360 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) 2361 return -EACCES; 2362 2363 error = get_user(val, arg); 2364 if (error) 2365 return error; 2366 2367 if (scsi_autopm_get_host(shost) < 0) 2368 return -EIO; 2369 2370 error = -EIO; 2371 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) + 2372 shost->hostt->cmd_size, GFP_KERNEL); 2373 if (!rq) 2374 goto out_put_autopm_host; 2375 blk_rq_init(NULL, rq); 2376 2377 scmd = (struct scsi_cmnd *)(rq + 1); 2378 scsi_init_command(dev, scmd); 2379 scmd->request = rq; 2380 scmd->cmnd = scsi_req(rq)->cmd; 2381 2382 scmd->scsi_done = scsi_reset_provider_done_command; 2383 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 2384 2385 scmd->cmd_len = 0; 2386 2387 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 2388 2389 spin_lock_irqsave(shost->host_lock, flags); 2390 shost->tmf_in_progress = 1; 2391 spin_unlock_irqrestore(shost->host_lock, flags); 2392 2393 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { 2394 case SG_SCSI_RESET_NOTHING: 2395 rtn = SUCCESS; 2396 break; 2397 case SG_SCSI_RESET_DEVICE: 2398 rtn = scsi_try_bus_device_reset(scmd); 2399 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2400 break; 2401 fallthrough; 2402 case SG_SCSI_RESET_TARGET: 2403 rtn = scsi_try_target_reset(scmd); 2404 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2405 break; 2406 fallthrough; 2407 case SG_SCSI_RESET_BUS: 2408 rtn = scsi_try_bus_reset(scmd); 2409 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2410 break; 2411 fallthrough; 2412 case SG_SCSI_RESET_HOST: 2413 rtn = scsi_try_host_reset(scmd); 2414 if (rtn == SUCCESS) 2415 break; 2416 fallthrough; 2417 default: 2418 rtn = FAILED; 2419 break; 2420 } 2421 2422 error = (rtn == SUCCESS) ? 0 : -EIO; 2423 2424 spin_lock_irqsave(shost->host_lock, flags); 2425 shost->tmf_in_progress = 0; 2426 spin_unlock_irqrestore(shost->host_lock, flags); 2427 2428 /* 2429 * be sure to wake up anyone who was sleeping or had their queue 2430 * suspended while we performed the TMF. 2431 */ 2432 SCSI_LOG_ERROR_RECOVERY(3, 2433 shost_printk(KERN_INFO, shost, 2434 "waking up host to restart after TMF\n")); 2435 2436 wake_up(&shost->host_wait); 2437 scsi_run_host_queues(shost); 2438 2439 kfree(rq); 2440 2441 out_put_autopm_host: 2442 scsi_autopm_put_host(shost); 2443 return error; 2444 } 2445 2446 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, 2447 struct scsi_sense_hdr *sshdr) 2448 { 2449 return scsi_normalize_sense(cmd->sense_buffer, 2450 SCSI_SENSE_BUFFERSIZE, sshdr); 2451 } 2452 EXPORT_SYMBOL(scsi_command_normalize_sense); 2453 2454 /** 2455 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2456 * @sense_buffer: byte array of sense data 2457 * @sb_len: number of valid bytes in sense_buffer 2458 * @info_out: pointer to 64 integer where 8 or 4 byte information 2459 * field will be placed if found. 2460 * 2461 * Return value: 2462 * true if information field found, false if not found. 2463 */ 2464 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len, 2465 u64 *info_out) 2466 { 2467 const u8 * ucp; 2468 2469 if (sb_len < 7) 2470 return false; 2471 switch (sense_buffer[0] & 0x7f) { 2472 case 0x70: 2473 case 0x71: 2474 if (sense_buffer[0] & 0x80) { 2475 *info_out = get_unaligned_be32(&sense_buffer[3]); 2476 return true; 2477 } 2478 return false; 2479 case 0x72: 2480 case 0x73: 2481 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2482 0 /* info desc */); 2483 if (ucp && (0xa == ucp[1])) { 2484 *info_out = get_unaligned_be64(&ucp[4]); 2485 return true; 2486 } 2487 return false; 2488 default: 2489 return false; 2490 } 2491 } 2492 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2493