1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * scsi_error.c Copyright (C) 1997 Eric Youngdale 4 * 5 * SCSI error/timeout handling 6 * Initial versions: Eric Youngdale. Based upon conversations with 7 * Leonard Zubkoff and David Miller at Linux Expo, 8 * ideas originating from all over the place. 9 * 10 * Restructured scsi_unjam_host and associated functions. 11 * September 04, 2002 Mike Anderson (andmike@us.ibm.com) 12 * 13 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and 14 * minor cleanups. 15 * September 30, 2002 Mike Anderson (andmike@us.ibm.com) 16 */ 17 18 #include <linux/module.h> 19 #include <linux/sched.h> 20 #include <linux/gfp.h> 21 #include <linux/timer.h> 22 #include <linux/string.h> 23 #include <linux/kernel.h> 24 #include <linux/freezer.h> 25 #include <linux/kthread.h> 26 #include <linux/interrupt.h> 27 #include <linux/blkdev.h> 28 #include <linux/delay.h> 29 #include <linux/jiffies.h> 30 31 #include <scsi/scsi.h> 32 #include <scsi/scsi_cmnd.h> 33 #include <scsi/scsi_dbg.h> 34 #include <scsi/scsi_device.h> 35 #include <scsi/scsi_driver.h> 36 #include <scsi/scsi_eh.h> 37 #include <scsi/scsi_common.h> 38 #include <scsi/scsi_transport.h> 39 #include <scsi/scsi_host.h> 40 #include <scsi/scsi_ioctl.h> 41 #include <scsi/scsi_dh.h> 42 #include <scsi/scsi_devinfo.h> 43 #include <scsi/sg.h> 44 45 #include "scsi_priv.h" 46 #include "scsi_logging.h" 47 #include "scsi_transport_api.h" 48 49 #include <trace/events/scsi.h> 50 51 #include <asm/unaligned.h> 52 53 /* 54 * These should *probably* be handled by the host itself. 55 * Since it is allowed to sleep, it probably should. 56 */ 57 #define BUS_RESET_SETTLE_TIME (10) 58 #define HOST_RESET_SETTLE_TIME (10) 59 60 static int scsi_eh_try_stu(struct scsi_cmnd *scmd); 61 static enum scsi_disposition scsi_try_to_abort_cmd(struct scsi_host_template *, 62 struct scsi_cmnd *); 63 64 void scsi_eh_wakeup(struct Scsi_Host *shost) 65 { 66 lockdep_assert_held(shost->host_lock); 67 68 if (scsi_host_busy(shost) == shost->host_failed) { 69 trace_scsi_eh_wakeup(shost); 70 wake_up_process(shost->ehandler); 71 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost, 72 "Waking error handler thread\n")); 73 } 74 } 75 76 /** 77 * scsi_schedule_eh - schedule EH for SCSI host 78 * @shost: SCSI host to invoke error handling on. 79 * 80 * Schedule SCSI EH without scmd. 81 */ 82 void scsi_schedule_eh(struct Scsi_Host *shost) 83 { 84 unsigned long flags; 85 86 spin_lock_irqsave(shost->host_lock, flags); 87 88 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || 89 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { 90 shost->host_eh_scheduled++; 91 scsi_eh_wakeup(shost); 92 } 93 94 spin_unlock_irqrestore(shost->host_lock, flags); 95 } 96 EXPORT_SYMBOL_GPL(scsi_schedule_eh); 97 98 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost) 99 { 100 if (!shost->last_reset || shost->eh_deadline == -1) 101 return 0; 102 103 /* 104 * 32bit accesses are guaranteed to be atomic 105 * (on all supported architectures), so instead 106 * of using a spinlock we can as well double check 107 * if eh_deadline has been set to 'off' during the 108 * time_before call. 109 */ 110 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) && 111 shost->eh_deadline > -1) 112 return 0; 113 114 return 1; 115 } 116 117 static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd) 118 { 119 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) 120 return true; 121 122 return ++cmd->retries <= cmd->allowed; 123 } 124 125 static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd) 126 { 127 struct scsi_device *sdev = cmd->device; 128 struct Scsi_Host *host = sdev->host; 129 130 if (host->hostt->eh_should_retry_cmd) 131 return host->hostt->eh_should_retry_cmd(cmd); 132 133 return true; 134 } 135 136 /** 137 * scmd_eh_abort_handler - Handle command aborts 138 * @work: command to be aborted. 139 * 140 * Note: this function must be called only for a command that has timed out. 141 * Because the block layer marks a request as complete before it calls 142 * scsi_timeout(), a .scsi_done() call from the LLD for a command that has 143 * timed out do not have any effect. Hence it is safe to call 144 * scsi_finish_command() from this function. 145 */ 146 void 147 scmd_eh_abort_handler(struct work_struct *work) 148 { 149 struct scsi_cmnd *scmd = 150 container_of(work, struct scsi_cmnd, abort_work.work); 151 struct scsi_device *sdev = scmd->device; 152 struct Scsi_Host *shost = sdev->host; 153 enum scsi_disposition rtn; 154 unsigned long flags; 155 156 if (scsi_host_eh_past_deadline(shost)) { 157 SCSI_LOG_ERROR_RECOVERY(3, 158 scmd_printk(KERN_INFO, scmd, 159 "eh timeout, not aborting\n")); 160 goto out; 161 } 162 163 SCSI_LOG_ERROR_RECOVERY(3, 164 scmd_printk(KERN_INFO, scmd, 165 "aborting command\n")); 166 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd); 167 if (rtn != SUCCESS) { 168 SCSI_LOG_ERROR_RECOVERY(3, 169 scmd_printk(KERN_INFO, scmd, 170 "cmd abort %s\n", 171 (rtn == FAST_IO_FAIL) ? 172 "not send" : "failed")); 173 goto out; 174 } 175 set_host_byte(scmd, DID_TIME_OUT); 176 if (scsi_host_eh_past_deadline(shost)) { 177 SCSI_LOG_ERROR_RECOVERY(3, 178 scmd_printk(KERN_INFO, scmd, 179 "eh timeout, not retrying " 180 "aborted command\n")); 181 goto out; 182 } 183 184 spin_lock_irqsave(shost->host_lock, flags); 185 list_del_init(&scmd->eh_entry); 186 187 /* 188 * If the abort succeeds, and there is no further 189 * EH action, clear the ->last_reset time. 190 */ 191 if (list_empty(&shost->eh_abort_list) && 192 list_empty(&shost->eh_cmd_q)) 193 if (shost->eh_deadline != -1) 194 shost->last_reset = 0; 195 196 spin_unlock_irqrestore(shost->host_lock, flags); 197 198 if (!scsi_noretry_cmd(scmd) && 199 scsi_cmd_retry_allowed(scmd) && 200 scsi_eh_should_retry_cmd(scmd)) { 201 SCSI_LOG_ERROR_RECOVERY(3, 202 scmd_printk(KERN_WARNING, scmd, 203 "retry aborted command\n")); 204 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 205 } else { 206 SCSI_LOG_ERROR_RECOVERY(3, 207 scmd_printk(KERN_WARNING, scmd, 208 "finish aborted command\n")); 209 scsi_finish_command(scmd); 210 } 211 return; 212 213 out: 214 spin_lock_irqsave(shost->host_lock, flags); 215 list_del_init(&scmd->eh_entry); 216 spin_unlock_irqrestore(shost->host_lock, flags); 217 218 scsi_eh_scmd_add(scmd); 219 } 220 221 /** 222 * scsi_abort_command - schedule a command abort 223 * @scmd: scmd to abort. 224 * 225 * We only need to abort commands after a command timeout 226 */ 227 static int 228 scsi_abort_command(struct scsi_cmnd *scmd) 229 { 230 struct scsi_device *sdev = scmd->device; 231 struct Scsi_Host *shost = sdev->host; 232 unsigned long flags; 233 234 if (!shost->hostt->eh_abort_handler) { 235 /* No abort handler, fail command directly */ 236 return FAILED; 237 } 238 239 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 240 /* 241 * Retry after abort failed, escalate to next level. 242 */ 243 SCSI_LOG_ERROR_RECOVERY(3, 244 scmd_printk(KERN_INFO, scmd, 245 "previous abort failed\n")); 246 BUG_ON(delayed_work_pending(&scmd->abort_work)); 247 return FAILED; 248 } 249 250 spin_lock_irqsave(shost->host_lock, flags); 251 if (shost->eh_deadline != -1 && !shost->last_reset) 252 shost->last_reset = jiffies; 253 BUG_ON(!list_empty(&scmd->eh_entry)); 254 list_add_tail(&scmd->eh_entry, &shost->eh_abort_list); 255 spin_unlock_irqrestore(shost->host_lock, flags); 256 257 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; 258 SCSI_LOG_ERROR_RECOVERY(3, 259 scmd_printk(KERN_INFO, scmd, "abort scheduled\n")); 260 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100); 261 return SUCCESS; 262 } 263 264 /** 265 * scsi_eh_reset - call into ->eh_action to reset internal counters 266 * @scmd: scmd to run eh on. 267 * 268 * The scsi driver might be carrying internal state about the 269 * devices, so we need to call into the driver to reset the 270 * internal state once the error handler is started. 271 */ 272 static void scsi_eh_reset(struct scsi_cmnd *scmd) 273 { 274 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) { 275 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 276 if (sdrv->eh_reset) 277 sdrv->eh_reset(scmd); 278 } 279 } 280 281 static void scsi_eh_inc_host_failed(struct rcu_head *head) 282 { 283 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu); 284 struct Scsi_Host *shost = scmd->device->host; 285 unsigned long flags; 286 287 spin_lock_irqsave(shost->host_lock, flags); 288 shost->host_failed++; 289 scsi_eh_wakeup(shost); 290 spin_unlock_irqrestore(shost->host_lock, flags); 291 } 292 293 /** 294 * scsi_eh_scmd_add - add scsi cmd to error handling. 295 * @scmd: scmd to run eh on. 296 */ 297 void scsi_eh_scmd_add(struct scsi_cmnd *scmd) 298 { 299 struct Scsi_Host *shost = scmd->device->host; 300 unsigned long flags; 301 int ret; 302 303 WARN_ON_ONCE(!shost->ehandler); 304 305 spin_lock_irqsave(shost->host_lock, flags); 306 if (scsi_host_set_state(shost, SHOST_RECOVERY)) { 307 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY); 308 WARN_ON_ONCE(ret); 309 } 310 if (shost->eh_deadline != -1 && !shost->last_reset) 311 shost->last_reset = jiffies; 312 313 scsi_eh_reset(scmd); 314 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 315 spin_unlock_irqrestore(shost->host_lock, flags); 316 /* 317 * Ensure that all tasks observe the host state change before the 318 * host_failed change. 319 */ 320 call_rcu_hurry(&scmd->rcu, scsi_eh_inc_host_failed); 321 } 322 323 /** 324 * scsi_timeout - Timeout function for normal scsi commands. 325 * @req: request that is timing out. 326 * 327 * Notes: 328 * We do not need to lock this. There is the potential for a race 329 * only in that the normal completion handling might run, but if the 330 * normal completion function determines that the timer has already 331 * fired, then it mustn't do anything. 332 */ 333 enum blk_eh_timer_return scsi_timeout(struct request *req) 334 { 335 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req); 336 struct Scsi_Host *host = scmd->device->host; 337 338 trace_scsi_dispatch_cmd_timeout(scmd); 339 scsi_log_completion(scmd, TIMEOUT_ERROR); 340 341 atomic_inc(&scmd->device->iotmo_cnt); 342 if (host->eh_deadline != -1 && !host->last_reset) 343 host->last_reset = jiffies; 344 345 if (host->hostt->eh_timed_out) { 346 switch (host->hostt->eh_timed_out(scmd)) { 347 case SCSI_EH_DONE: 348 return BLK_EH_DONE; 349 case SCSI_EH_RESET_TIMER: 350 return BLK_EH_RESET_TIMER; 351 case SCSI_EH_NOT_HANDLED: 352 break; 353 } 354 } 355 356 /* 357 * If scsi_done() has already set SCMD_STATE_COMPLETE, do not modify 358 * *scmd. 359 */ 360 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state)) 361 return BLK_EH_DONE; 362 atomic_inc(&scmd->device->iodone_cnt); 363 if (scsi_abort_command(scmd) != SUCCESS) { 364 set_host_byte(scmd, DID_TIME_OUT); 365 scsi_eh_scmd_add(scmd); 366 } 367 368 return BLK_EH_DONE; 369 } 370 371 /** 372 * scsi_block_when_processing_errors - Prevent cmds from being queued. 373 * @sdev: Device on which we are performing recovery. 374 * 375 * Description: 376 * We block until the host is out of error recovery, and then check to 377 * see whether the host or the device is offline. 378 * 379 * Return value: 380 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 381 */ 382 int scsi_block_when_processing_errors(struct scsi_device *sdev) 383 { 384 int online; 385 386 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 387 388 online = scsi_device_online(sdev); 389 390 return online; 391 } 392 EXPORT_SYMBOL(scsi_block_when_processing_errors); 393 394 #ifdef CONFIG_SCSI_LOGGING 395 /** 396 * scsi_eh_prt_fail_stats - Log info on failures. 397 * @shost: scsi host being recovered. 398 * @work_q: Queue of scsi cmds to process. 399 */ 400 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 401 struct list_head *work_q) 402 { 403 struct scsi_cmnd *scmd; 404 struct scsi_device *sdev; 405 int total_failures = 0; 406 int cmd_failed = 0; 407 int cmd_cancel = 0; 408 int devices_failed = 0; 409 410 shost_for_each_device(sdev, shost) { 411 list_for_each_entry(scmd, work_q, eh_entry) { 412 if (scmd->device == sdev) { 413 ++total_failures; 414 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) 415 ++cmd_cancel; 416 else 417 ++cmd_failed; 418 } 419 } 420 421 if (cmd_cancel || cmd_failed) { 422 SCSI_LOG_ERROR_RECOVERY(3, 423 shost_printk(KERN_INFO, shost, 424 "%s: cmds failed: %d, cancel: %d\n", 425 __func__, cmd_failed, 426 cmd_cancel)); 427 cmd_cancel = 0; 428 cmd_failed = 0; 429 ++devices_failed; 430 } 431 } 432 433 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, 434 "Total of %d commands on %d" 435 " devices require eh work\n", 436 total_failures, devices_failed)); 437 } 438 #endif 439 440 /** 441 * scsi_report_lun_change - Set flag on all *other* devices on the same target 442 * to indicate that a UNIT ATTENTION is expected. 443 * @sdev: Device reporting the UNIT ATTENTION 444 */ 445 static void scsi_report_lun_change(struct scsi_device *sdev) 446 { 447 sdev->sdev_target->expecting_lun_change = 1; 448 } 449 450 /** 451 * scsi_report_sense - Examine scsi sense information and log messages for 452 * certain conditions, also issue uevents for some of them. 453 * @sdev: Device reporting the sense code 454 * @sshdr: sshdr to be examined 455 */ 456 static void scsi_report_sense(struct scsi_device *sdev, 457 struct scsi_sense_hdr *sshdr) 458 { 459 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ 460 461 if (sshdr->sense_key == UNIT_ATTENTION) { 462 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { 463 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; 464 sdev_printk(KERN_WARNING, sdev, 465 "Inquiry data has changed"); 466 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { 467 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; 468 scsi_report_lun_change(sdev); 469 sdev_printk(KERN_WARNING, sdev, 470 "LUN assignments on this target have " 471 "changed. The Linux SCSI layer does not " 472 "automatically remap LUN assignments.\n"); 473 } else if (sshdr->asc == 0x3f) 474 sdev_printk(KERN_WARNING, sdev, 475 "Operating parameters on this target have " 476 "changed. The Linux SCSI layer does not " 477 "automatically adjust these parameters.\n"); 478 479 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { 480 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; 481 sdev_printk(KERN_WARNING, sdev, 482 "Warning! Received an indication that the " 483 "LUN reached a thin provisioning soft " 484 "threshold.\n"); 485 } 486 487 if (sshdr->asc == 0x29) { 488 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED; 489 /* 490 * Do not print message if it is an expected side-effect 491 * of runtime PM. 492 */ 493 if (!sdev->silence_suspend) 494 sdev_printk(KERN_WARNING, sdev, 495 "Power-on or device reset occurred\n"); 496 } 497 498 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { 499 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; 500 sdev_printk(KERN_WARNING, sdev, 501 "Mode parameters changed"); 502 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { 503 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; 504 sdev_printk(KERN_WARNING, sdev, 505 "Asymmetric access state changed"); 506 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { 507 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; 508 sdev_printk(KERN_WARNING, sdev, 509 "Capacity data has changed"); 510 } else if (sshdr->asc == 0x2a) 511 sdev_printk(KERN_WARNING, sdev, 512 "Parameters changed"); 513 } 514 515 if (evt_type != SDEV_EVT_MAXBITS) { 516 set_bit(evt_type, sdev->pending_events); 517 schedule_work(&sdev->event_work); 518 } 519 } 520 521 static inline void set_scsi_ml_byte(struct scsi_cmnd *cmd, u8 status) 522 { 523 cmd->result = (cmd->result & 0xffff00ff) | (status << 8); 524 } 525 526 /** 527 * scsi_check_sense - Examine scsi cmd sense 528 * @scmd: Cmd to have sense checked. 529 * 530 * Return value: 531 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE 532 * 533 * Notes: 534 * When a deferred error is detected the current command has 535 * not been executed and needs retrying. 536 */ 537 enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd) 538 { 539 struct scsi_device *sdev = scmd->device; 540 struct scsi_sense_hdr sshdr; 541 542 if (! scsi_command_normalize_sense(scmd, &sshdr)) 543 return FAILED; /* no valid sense data */ 544 545 scsi_report_sense(sdev, &sshdr); 546 547 if (scsi_sense_is_deferred(&sshdr)) 548 return NEEDS_RETRY; 549 550 if (sdev->handler && sdev->handler->check_sense) { 551 enum scsi_disposition rc; 552 553 rc = sdev->handler->check_sense(sdev, &sshdr); 554 if (rc != SCSI_RETURN_NOT_HANDLED) 555 return rc; 556 /* handler does not care. Drop down to default handling */ 557 } 558 559 if (scmd->cmnd[0] == TEST_UNIT_READY && 560 scmd->submitter != SUBMITTED_BY_SCSI_ERROR_HANDLER) 561 /* 562 * nasty: for mid-layer issued TURs, we need to return the 563 * actual sense data without any recovery attempt. For eh 564 * issued ones, we need to try to recover and interpret 565 */ 566 return SUCCESS; 567 568 /* 569 * Previous logic looked for FILEMARK, EOM or ILI which are 570 * mainly associated with tapes and returned SUCCESS. 571 */ 572 if (sshdr.response_code == 0x70) { 573 /* fixed format */ 574 if (scmd->sense_buffer[2] & 0xe0) 575 return SUCCESS; 576 } else { 577 /* 578 * descriptor format: look for "stream commands sense data 579 * descriptor" (see SSC-3). Assume single sense data 580 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 581 */ 582 if ((sshdr.additional_length > 3) && 583 (scmd->sense_buffer[8] == 0x4) && 584 (scmd->sense_buffer[11] & 0xe0)) 585 return SUCCESS; 586 } 587 588 switch (sshdr.sense_key) { 589 case NO_SENSE: 590 return SUCCESS; 591 case RECOVERED_ERROR: 592 return /* soft_error */ SUCCESS; 593 594 case ABORTED_COMMAND: 595 if (sshdr.asc == 0x10) /* DIF */ 596 return SUCCESS; 597 598 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF) 599 return ADD_TO_MLQUEUE; 600 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 && 601 sdev->sdev_bflags & BLIST_RETRY_ASC_C1) 602 return ADD_TO_MLQUEUE; 603 604 return NEEDS_RETRY; 605 case NOT_READY: 606 case UNIT_ATTENTION: 607 /* 608 * if we are expecting a cc/ua because of a bus reset that we 609 * performed, treat this just as a retry. otherwise this is 610 * information that we should pass up to the upper-level driver 611 * so that we can deal with it there. 612 */ 613 if (scmd->device->expecting_cc_ua) { 614 /* 615 * Because some device does not queue unit 616 * attentions correctly, we carefully check 617 * additional sense code and qualifier so as 618 * not to squash media change unit attention. 619 */ 620 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { 621 scmd->device->expecting_cc_ua = 0; 622 return NEEDS_RETRY; 623 } 624 } 625 /* 626 * we might also expect a cc/ua if another LUN on the target 627 * reported a UA with an ASC/ASCQ of 3F 0E - 628 * REPORTED LUNS DATA HAS CHANGED. 629 */ 630 if (scmd->device->sdev_target->expecting_lun_change && 631 sshdr.asc == 0x3f && sshdr.ascq == 0x0e) 632 return NEEDS_RETRY; 633 /* 634 * if the device is in the process of becoming ready, we 635 * should retry. 636 */ 637 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 638 return NEEDS_RETRY; 639 /* 640 * if the device is not started, we need to wake 641 * the error handler to start the motor 642 */ 643 if (scmd->device->allow_restart && 644 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 645 return FAILED; 646 /* 647 * Pass the UA upwards for a determination in the completion 648 * functions. 649 */ 650 return SUCCESS; 651 652 /* these are not supported */ 653 case DATA_PROTECT: 654 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { 655 /* Thin provisioning hard threshold reached */ 656 set_scsi_ml_byte(scmd, SCSIML_STAT_NOSPC); 657 return SUCCESS; 658 } 659 fallthrough; 660 case COPY_ABORTED: 661 case VOLUME_OVERFLOW: 662 case MISCOMPARE: 663 case BLANK_CHECK: 664 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE); 665 return SUCCESS; 666 667 case MEDIUM_ERROR: 668 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 669 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 670 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 671 set_scsi_ml_byte(scmd, SCSIML_STAT_MED_ERROR); 672 return SUCCESS; 673 } 674 return NEEDS_RETRY; 675 676 case HARDWARE_ERROR: 677 if (scmd->device->retry_hwerror) 678 return ADD_TO_MLQUEUE; 679 else 680 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE); 681 fallthrough; 682 683 case ILLEGAL_REQUEST: 684 if (sshdr.asc == 0x20 || /* Invalid command operation code */ 685 sshdr.asc == 0x21 || /* Logical block address out of range */ 686 sshdr.asc == 0x22 || /* Invalid function */ 687 sshdr.asc == 0x24 || /* Invalid field in cdb */ 688 sshdr.asc == 0x26 || /* Parameter value invalid */ 689 sshdr.asc == 0x27) { /* Write protected */ 690 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE); 691 } 692 return SUCCESS; 693 694 default: 695 return SUCCESS; 696 } 697 } 698 EXPORT_SYMBOL_GPL(scsi_check_sense); 699 700 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) 701 { 702 struct scsi_host_template *sht = sdev->host->hostt; 703 struct scsi_device *tmp_sdev; 704 705 if (!sht->track_queue_depth || 706 sdev->queue_depth >= sdev->max_queue_depth) 707 return; 708 709 if (time_before(jiffies, 710 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) 711 return; 712 713 if (time_before(jiffies, 714 sdev->last_queue_full_time + sdev->queue_ramp_up_period)) 715 return; 716 717 /* 718 * Walk all devices of a target and do 719 * ramp up on them. 720 */ 721 shost_for_each_device(tmp_sdev, sdev->host) { 722 if (tmp_sdev->channel != sdev->channel || 723 tmp_sdev->id != sdev->id || 724 tmp_sdev->queue_depth == sdev->max_queue_depth) 725 continue; 726 727 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); 728 sdev->last_queue_ramp_up = jiffies; 729 } 730 } 731 732 static void scsi_handle_queue_full(struct scsi_device *sdev) 733 { 734 struct scsi_host_template *sht = sdev->host->hostt; 735 struct scsi_device *tmp_sdev; 736 737 if (!sht->track_queue_depth) 738 return; 739 740 shost_for_each_device(tmp_sdev, sdev->host) { 741 if (tmp_sdev->channel != sdev->channel || 742 tmp_sdev->id != sdev->id) 743 continue; 744 /* 745 * We do not know the number of commands that were at 746 * the device when we got the queue full so we start 747 * from the highest possible value and work our way down. 748 */ 749 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); 750 } 751 } 752 753 /** 754 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 755 * @scmd: SCSI cmd to examine. 756 * 757 * Notes: 758 * This is *only* called when we are examining the status of commands 759 * queued during error recovery. the main difference here is that we 760 * don't allow for the possibility of retries here, and we are a lot 761 * more restrictive about what we consider acceptable. 762 */ 763 static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd) 764 { 765 /* 766 * first check the host byte, to see if there is anything in there 767 * that would indicate what we need to do. 768 */ 769 if (host_byte(scmd->result) == DID_RESET) { 770 /* 771 * rats. we are already in the error handler, so we now 772 * get to try and figure out what to do next. if the sense 773 * is valid, we have a pretty good idea of what to do. 774 * if not, we mark it as FAILED. 775 */ 776 return scsi_check_sense(scmd); 777 } 778 if (host_byte(scmd->result) != DID_OK) 779 return FAILED; 780 781 /* 782 * now, check the status byte to see if this indicates 783 * anything special. 784 */ 785 switch (get_status_byte(scmd)) { 786 case SAM_STAT_GOOD: 787 scsi_handle_queue_ramp_up(scmd->device); 788 fallthrough; 789 case SAM_STAT_COMMAND_TERMINATED: 790 return SUCCESS; 791 case SAM_STAT_CHECK_CONDITION: 792 return scsi_check_sense(scmd); 793 case SAM_STAT_CONDITION_MET: 794 case SAM_STAT_INTERMEDIATE: 795 case SAM_STAT_INTERMEDIATE_CONDITION_MET: 796 /* 797 * who knows? FIXME(eric) 798 */ 799 return SUCCESS; 800 case SAM_STAT_RESERVATION_CONFLICT: 801 if (scmd->cmnd[0] == TEST_UNIT_READY) 802 /* it is a success, we probed the device and 803 * found it */ 804 return SUCCESS; 805 /* otherwise, we failed to send the command */ 806 return FAILED; 807 case SAM_STAT_TASK_SET_FULL: 808 scsi_handle_queue_full(scmd->device); 809 fallthrough; 810 case SAM_STAT_BUSY: 811 return NEEDS_RETRY; 812 default: 813 return FAILED; 814 } 815 return FAILED; 816 } 817 818 /** 819 * scsi_eh_done - Completion function for error handling. 820 * @scmd: Cmd that is done. 821 */ 822 void scsi_eh_done(struct scsi_cmnd *scmd) 823 { 824 struct completion *eh_action; 825 826 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 827 "%s result: %x\n", __func__, scmd->result)); 828 829 eh_action = scmd->device->host->eh_action; 830 if (eh_action) 831 complete(eh_action); 832 } 833 834 /** 835 * scsi_try_host_reset - ask host adapter to reset itself 836 * @scmd: SCSI cmd to send host reset. 837 */ 838 static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd) 839 { 840 unsigned long flags; 841 enum scsi_disposition rtn; 842 struct Scsi_Host *host = scmd->device->host; 843 struct scsi_host_template *hostt = host->hostt; 844 845 SCSI_LOG_ERROR_RECOVERY(3, 846 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 847 848 if (!hostt->eh_host_reset_handler) 849 return FAILED; 850 851 rtn = hostt->eh_host_reset_handler(scmd); 852 853 if (rtn == SUCCESS) { 854 if (!hostt->skip_settle_delay) 855 ssleep(HOST_RESET_SETTLE_TIME); 856 spin_lock_irqsave(host->host_lock, flags); 857 scsi_report_bus_reset(host, scmd_channel(scmd)); 858 spin_unlock_irqrestore(host->host_lock, flags); 859 } 860 861 return rtn; 862 } 863 864 /** 865 * scsi_try_bus_reset - ask host to perform a bus reset 866 * @scmd: SCSI cmd to send bus reset. 867 */ 868 static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd) 869 { 870 unsigned long flags; 871 enum scsi_disposition rtn; 872 struct Scsi_Host *host = scmd->device->host; 873 struct scsi_host_template *hostt = host->hostt; 874 875 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 876 "%s: Snd Bus RST\n", __func__)); 877 878 if (!hostt->eh_bus_reset_handler) 879 return FAILED; 880 881 rtn = hostt->eh_bus_reset_handler(scmd); 882 883 if (rtn == SUCCESS) { 884 if (!hostt->skip_settle_delay) 885 ssleep(BUS_RESET_SETTLE_TIME); 886 spin_lock_irqsave(host->host_lock, flags); 887 scsi_report_bus_reset(host, scmd_channel(scmd)); 888 spin_unlock_irqrestore(host->host_lock, flags); 889 } 890 891 return rtn; 892 } 893 894 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 895 { 896 sdev->was_reset = 1; 897 sdev->expecting_cc_ua = 1; 898 } 899 900 /** 901 * scsi_try_target_reset - Ask host to perform a target reset 902 * @scmd: SCSI cmd used to send a target reset 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_target_reset(struct scsi_cmnd *scmd) 911 { 912 unsigned long flags; 913 enum scsi_disposition rtn; 914 struct Scsi_Host *host = scmd->device->host; 915 struct scsi_host_template *hostt = host->hostt; 916 917 if (!hostt->eh_target_reset_handler) 918 return FAILED; 919 920 rtn = hostt->eh_target_reset_handler(scmd); 921 if (rtn == SUCCESS) { 922 spin_lock_irqsave(host->host_lock, flags); 923 __starget_for_each_device(scsi_target(scmd->device), NULL, 924 __scsi_report_device_reset); 925 spin_unlock_irqrestore(host->host_lock, flags); 926 } 927 928 return rtn; 929 } 930 931 /** 932 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 933 * @scmd: SCSI cmd used to send BDR 934 * 935 * Notes: 936 * There is no timeout for this operation. if this operation is 937 * unreliable for a given host, then the host itself needs to put a 938 * timer on it, and set the host back to a consistent state prior to 939 * returning. 940 */ 941 static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 942 { 943 enum scsi_disposition rtn; 944 struct scsi_host_template *hostt = scmd->device->host->hostt; 945 946 if (!hostt->eh_device_reset_handler) 947 return FAILED; 948 949 rtn = hostt->eh_device_reset_handler(scmd); 950 if (rtn == SUCCESS) 951 __scsi_report_device_reset(scmd->device, NULL); 952 return rtn; 953 } 954 955 /** 956 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command 957 * @hostt: SCSI driver host template 958 * @scmd: SCSI cmd used to send a target reset 959 * 960 * Return value: 961 * SUCCESS, FAILED, or FAST_IO_FAIL 962 * 963 * Notes: 964 * SUCCESS does not necessarily indicate that the command 965 * has been aborted; it only indicates that the LLDDs 966 * has cleared all references to that command. 967 * LLDDs should return FAILED only if an abort was required 968 * but could not be executed. LLDDs should return FAST_IO_FAIL 969 * if the device is temporarily unavailable (eg due to a 970 * link down on FibreChannel) 971 */ 972 static enum scsi_disposition 973 scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd) 974 { 975 if (!hostt->eh_abort_handler) 976 return FAILED; 977 978 return hostt->eh_abort_handler(scmd); 979 } 980 981 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 982 { 983 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 984 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 985 if (scsi_try_target_reset(scmd) != SUCCESS) 986 if (scsi_try_bus_reset(scmd) != SUCCESS) 987 scsi_try_host_reset(scmd); 988 } 989 990 /** 991 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 992 * @scmd: SCSI command structure to hijack 993 * @ses: structure to save restore information 994 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 995 * @cmnd_size: size in bytes of @cmnd (must be <= MAX_COMMAND_SIZE) 996 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 997 * 998 * This function is used to save a scsi command information before re-execution 999 * as part of the error recovery process. If @sense_bytes is 0 the command 1000 * sent must be one that does not transfer any data. If @sense_bytes != 0 1001 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 1002 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 1003 */ 1004 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 1005 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 1006 { 1007 struct scsi_device *sdev = scmd->device; 1008 1009 /* 1010 * We need saved copies of a number of fields - this is because 1011 * error handling may need to overwrite these with different values 1012 * to run different commands, and once error handling is complete, 1013 * we will need to restore these values prior to running the actual 1014 * command. 1015 */ 1016 ses->cmd_len = scmd->cmd_len; 1017 ses->data_direction = scmd->sc_data_direction; 1018 ses->sdb = scmd->sdb; 1019 ses->result = scmd->result; 1020 ses->resid_len = scmd->resid_len; 1021 ses->underflow = scmd->underflow; 1022 ses->prot_op = scmd->prot_op; 1023 ses->eh_eflags = scmd->eh_eflags; 1024 1025 scmd->prot_op = SCSI_PROT_NORMAL; 1026 scmd->eh_eflags = 0; 1027 memcpy(ses->cmnd, scmd->cmnd, sizeof(ses->cmnd)); 1028 memset(scmd->cmnd, 0, sizeof(scmd->cmnd)); 1029 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 1030 scmd->result = 0; 1031 scmd->resid_len = 0; 1032 1033 if (sense_bytes) { 1034 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 1035 sense_bytes); 1036 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 1037 scmd->sdb.length); 1038 scmd->sdb.table.sgl = &ses->sense_sgl; 1039 scmd->sc_data_direction = DMA_FROM_DEVICE; 1040 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; 1041 scmd->cmnd[0] = REQUEST_SENSE; 1042 scmd->cmnd[4] = scmd->sdb.length; 1043 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1044 } else { 1045 scmd->sc_data_direction = DMA_NONE; 1046 if (cmnd) { 1047 BUG_ON(cmnd_size > sizeof(scmd->cmnd)); 1048 memcpy(scmd->cmnd, cmnd, cmnd_size); 1049 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1050 } 1051 } 1052 1053 scmd->underflow = 0; 1054 1055 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 1056 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 1057 (sdev->lun << 5 & 0xe0); 1058 1059 /* 1060 * Zero the sense buffer. The scsi spec mandates that any 1061 * untransferred sense data should be interpreted as being zero. 1062 */ 1063 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 1064 } 1065 EXPORT_SYMBOL(scsi_eh_prep_cmnd); 1066 1067 /** 1068 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 1069 * @scmd: SCSI command structure to restore 1070 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 1071 * 1072 * Undo any damage done by above scsi_eh_prep_cmnd(). 1073 */ 1074 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 1075 { 1076 /* 1077 * Restore original data 1078 */ 1079 scmd->cmd_len = ses->cmd_len; 1080 memcpy(scmd->cmnd, ses->cmnd, sizeof(ses->cmnd)); 1081 scmd->sc_data_direction = ses->data_direction; 1082 scmd->sdb = ses->sdb; 1083 scmd->result = ses->result; 1084 scmd->resid_len = ses->resid_len; 1085 scmd->underflow = ses->underflow; 1086 scmd->prot_op = ses->prot_op; 1087 scmd->eh_eflags = ses->eh_eflags; 1088 } 1089 EXPORT_SYMBOL(scsi_eh_restore_cmnd); 1090 1091 /** 1092 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 1093 * @scmd: SCSI command structure to hijack 1094 * @cmnd: CDB to send 1095 * @cmnd_size: size in bytes of @cmnd 1096 * @timeout: timeout for this request 1097 * @sense_bytes: size of sense data to copy or 0 1098 * 1099 * This function is used to send a scsi command down to a target device 1100 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 1101 * 1102 * Return value: 1103 * SUCCESS or FAILED or NEEDS_RETRY 1104 */ 1105 static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd, 1106 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes) 1107 { 1108 struct scsi_device *sdev = scmd->device; 1109 struct Scsi_Host *shost = sdev->host; 1110 DECLARE_COMPLETION_ONSTACK(done); 1111 unsigned long timeleft = timeout, delay; 1112 struct scsi_eh_save ses; 1113 const unsigned long stall_for = msecs_to_jiffies(100); 1114 int rtn; 1115 1116 retry: 1117 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1118 shost->eh_action = &done; 1119 1120 scsi_log_send(scmd); 1121 scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER; 1122 1123 /* 1124 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can 1125 * change the SCSI device state after we have examined it and before 1126 * .queuecommand() is called. 1127 */ 1128 mutex_lock(&sdev->state_mutex); 1129 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) { 1130 mutex_unlock(&sdev->state_mutex); 1131 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev, 1132 "%s: state %d <> %d\n", __func__, sdev->sdev_state, 1133 SDEV_BLOCK)); 1134 delay = min(timeleft, stall_for); 1135 timeleft -= delay; 1136 msleep(jiffies_to_msecs(delay)); 1137 mutex_lock(&sdev->state_mutex); 1138 } 1139 if (sdev->sdev_state != SDEV_BLOCK) 1140 rtn = shost->hostt->queuecommand(shost, scmd); 1141 else 1142 rtn = FAILED; 1143 mutex_unlock(&sdev->state_mutex); 1144 1145 if (rtn) { 1146 if (timeleft > stall_for) { 1147 scsi_eh_restore_cmnd(scmd, &ses); 1148 1149 timeleft -= stall_for; 1150 msleep(jiffies_to_msecs(stall_for)); 1151 goto retry; 1152 } 1153 /* signal not to enter either branch of the if () below */ 1154 timeleft = 0; 1155 rtn = FAILED; 1156 } else { 1157 timeleft = wait_for_completion_timeout(&done, timeout); 1158 rtn = SUCCESS; 1159 } 1160 1161 shost->eh_action = NULL; 1162 1163 scsi_log_completion(scmd, rtn); 1164 1165 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1166 "%s timeleft: %ld\n", 1167 __func__, timeleft)); 1168 1169 /* 1170 * If there is time left scsi_eh_done got called, and we will examine 1171 * the actual status codes to see whether the command actually did 1172 * complete normally, else if we have a zero return and no time left, 1173 * the command must still be pending, so abort it and return FAILED. 1174 * If we never actually managed to issue the command, because 1175 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1176 * value above (so don't execute either branch of the if) 1177 */ 1178 if (timeleft) { 1179 rtn = scsi_eh_completed_normally(scmd); 1180 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1181 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1182 1183 switch (rtn) { 1184 case SUCCESS: 1185 case NEEDS_RETRY: 1186 case FAILED: 1187 break; 1188 case ADD_TO_MLQUEUE: 1189 rtn = NEEDS_RETRY; 1190 break; 1191 default: 1192 rtn = FAILED; 1193 break; 1194 } 1195 } else if (rtn != FAILED) { 1196 scsi_abort_eh_cmnd(scmd); 1197 rtn = FAILED; 1198 } 1199 1200 scsi_eh_restore_cmnd(scmd, &ses); 1201 1202 return rtn; 1203 } 1204 1205 /** 1206 * scsi_request_sense - Request sense data from a particular target. 1207 * @scmd: SCSI cmd for request sense. 1208 * 1209 * Notes: 1210 * Some hosts automatically obtain this information, others require 1211 * that we obtain it on our own. This function will *not* return until 1212 * the command either times out, or it completes. 1213 */ 1214 static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd) 1215 { 1216 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1217 } 1218 1219 static enum scsi_disposition 1220 scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn) 1221 { 1222 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) { 1223 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1224 if (sdrv->eh_action) 1225 rtn = sdrv->eh_action(scmd, rtn); 1226 } 1227 return rtn; 1228 } 1229 1230 /** 1231 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1232 * @scmd: Original SCSI cmd that eh has finished. 1233 * @done_q: Queue for processed commands. 1234 * 1235 * Notes: 1236 * We don't want to use the normal command completion while we are are 1237 * still handling errors - it may cause other commands to be queued, 1238 * and that would disturb what we are doing. Thus we really want to 1239 * keep a list of pending commands for final completion, and once we 1240 * are ready to leave error handling we handle completion for real. 1241 */ 1242 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1243 { 1244 list_move_tail(&scmd->eh_entry, done_q); 1245 } 1246 EXPORT_SYMBOL(scsi_eh_finish_cmd); 1247 1248 /** 1249 * scsi_eh_get_sense - Get device sense data. 1250 * @work_q: Queue of commands to process. 1251 * @done_q: Queue of processed commands. 1252 * 1253 * Description: 1254 * See if we need to request sense information. if so, then get it 1255 * now, so we have a better idea of what to do. 1256 * 1257 * Notes: 1258 * This has the unfortunate side effect that if a shost adapter does 1259 * not automatically request sense information, we end up shutting 1260 * it down before we request it. 1261 * 1262 * All drivers should request sense information internally these days, 1263 * so for now all I have to say is tough noogies if you end up in here. 1264 * 1265 * XXX: Long term this code should go away, but that needs an audit of 1266 * all LLDDs first. 1267 */ 1268 int scsi_eh_get_sense(struct list_head *work_q, 1269 struct list_head *done_q) 1270 { 1271 struct scsi_cmnd *scmd, *next; 1272 struct Scsi_Host *shost; 1273 enum scsi_disposition rtn; 1274 1275 /* 1276 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, 1277 * should not get sense. 1278 */ 1279 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1280 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || 1281 SCSI_SENSE_VALID(scmd)) 1282 continue; 1283 1284 shost = scmd->device->host; 1285 if (scsi_host_eh_past_deadline(shost)) { 1286 SCSI_LOG_ERROR_RECOVERY(3, 1287 scmd_printk(KERN_INFO, scmd, 1288 "%s: skip request sense, past eh deadline\n", 1289 current->comm)); 1290 break; 1291 } 1292 if (!scsi_status_is_check_condition(scmd->result)) 1293 /* 1294 * don't request sense if there's no check condition 1295 * status because the error we're processing isn't one 1296 * that has a sense code (and some devices get 1297 * confused by sense requests out of the blue) 1298 */ 1299 continue; 1300 1301 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1302 "%s: requesting sense\n", 1303 current->comm)); 1304 rtn = scsi_request_sense(scmd); 1305 if (rtn != SUCCESS) 1306 continue; 1307 1308 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1309 "sense requested, result %x\n", scmd->result)); 1310 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); 1311 1312 rtn = scsi_decide_disposition(scmd); 1313 1314 /* 1315 * if the result was normal, then just pass it along to the 1316 * upper level. 1317 */ 1318 if (rtn == SUCCESS) 1319 /* 1320 * We don't want this command reissued, just finished 1321 * with the sense data, so set retries to the max 1322 * allowed to ensure it won't get reissued. If the user 1323 * has requested infinite retries, we also want to 1324 * finish this command, so force completion by setting 1325 * retries and allowed to the same value. 1326 */ 1327 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) 1328 scmd->retries = scmd->allowed = 1; 1329 else 1330 scmd->retries = scmd->allowed; 1331 else if (rtn != NEEDS_RETRY) 1332 continue; 1333 1334 scsi_eh_finish_cmd(scmd, done_q); 1335 } 1336 1337 return list_empty(work_q); 1338 } 1339 EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1340 1341 /** 1342 * scsi_eh_tur - Send TUR to device. 1343 * @scmd: &scsi_cmnd to send TUR 1344 * 1345 * Return value: 1346 * 0 - Device is ready. 1 - Device NOT ready. 1347 */ 1348 static int scsi_eh_tur(struct scsi_cmnd *scmd) 1349 { 1350 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1351 int retry_cnt = 1; 1352 enum scsi_disposition rtn; 1353 1354 retry_tur: 1355 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1356 scmd->device->eh_timeout, 0); 1357 1358 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1359 "%s return: %x\n", __func__, rtn)); 1360 1361 switch (rtn) { 1362 case NEEDS_RETRY: 1363 if (retry_cnt--) 1364 goto retry_tur; 1365 fallthrough; 1366 case SUCCESS: 1367 return 0; 1368 default: 1369 return 1; 1370 } 1371 } 1372 1373 /** 1374 * scsi_eh_test_devices - check if devices are responding from error recovery. 1375 * @cmd_list: scsi commands in error recovery. 1376 * @work_q: queue for commands which still need more error recovery 1377 * @done_q: queue for commands which are finished 1378 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1379 * 1380 * Decription: 1381 * Tests if devices are in a working state. Commands to devices now in 1382 * a working state are sent to the done_q while commands to devices which 1383 * are still failing to respond are returned to the work_q for more 1384 * processing. 1385 **/ 1386 static int scsi_eh_test_devices(struct list_head *cmd_list, 1387 struct list_head *work_q, 1388 struct list_head *done_q, int try_stu) 1389 { 1390 struct scsi_cmnd *scmd, *next; 1391 struct scsi_device *sdev; 1392 int finish_cmds; 1393 1394 while (!list_empty(cmd_list)) { 1395 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1396 sdev = scmd->device; 1397 1398 if (!try_stu) { 1399 if (scsi_host_eh_past_deadline(sdev->host)) { 1400 /* Push items back onto work_q */ 1401 list_splice_init(cmd_list, work_q); 1402 SCSI_LOG_ERROR_RECOVERY(3, 1403 sdev_printk(KERN_INFO, sdev, 1404 "%s: skip test device, past eh deadline", 1405 current->comm)); 1406 break; 1407 } 1408 } 1409 1410 finish_cmds = !scsi_device_online(scmd->device) || 1411 (try_stu && !scsi_eh_try_stu(scmd) && 1412 !scsi_eh_tur(scmd)) || 1413 !scsi_eh_tur(scmd); 1414 1415 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1416 if (scmd->device == sdev) { 1417 if (finish_cmds && 1418 (try_stu || 1419 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1420 scsi_eh_finish_cmd(scmd, done_q); 1421 else 1422 list_move_tail(&scmd->eh_entry, work_q); 1423 } 1424 } 1425 return list_empty(work_q); 1426 } 1427 1428 /** 1429 * scsi_eh_try_stu - Send START_UNIT to device. 1430 * @scmd: &scsi_cmnd to send START_UNIT 1431 * 1432 * Return value: 1433 * 0 - Device is ready. 1 - Device NOT ready. 1434 */ 1435 static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1436 { 1437 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1438 1439 if (scmd->device->allow_restart) { 1440 int i; 1441 enum scsi_disposition rtn = NEEDS_RETRY; 1442 1443 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1444 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, 1445 scmd->device->eh_timeout, 0); 1446 1447 if (rtn == SUCCESS) 1448 return 0; 1449 } 1450 1451 return 1; 1452 } 1453 1454 /** 1455 * scsi_eh_stu - send START_UNIT if needed 1456 * @shost: &scsi host being recovered. 1457 * @work_q: &list_head for pending commands. 1458 * @done_q: &list_head for processed commands. 1459 * 1460 * Notes: 1461 * If commands are failing due to not ready, initializing command required, 1462 * try revalidating the device, which will end up sending a start unit. 1463 */ 1464 static int scsi_eh_stu(struct Scsi_Host *shost, 1465 struct list_head *work_q, 1466 struct list_head *done_q) 1467 { 1468 struct scsi_cmnd *scmd, *stu_scmd, *next; 1469 struct scsi_device *sdev; 1470 1471 shost_for_each_device(sdev, shost) { 1472 if (scsi_host_eh_past_deadline(shost)) { 1473 SCSI_LOG_ERROR_RECOVERY(3, 1474 sdev_printk(KERN_INFO, sdev, 1475 "%s: skip START_UNIT, past eh deadline\n", 1476 current->comm)); 1477 scsi_device_put(sdev); 1478 break; 1479 } 1480 stu_scmd = NULL; 1481 list_for_each_entry(scmd, work_q, eh_entry) 1482 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1483 scsi_check_sense(scmd) == FAILED ) { 1484 stu_scmd = scmd; 1485 break; 1486 } 1487 1488 if (!stu_scmd) 1489 continue; 1490 1491 SCSI_LOG_ERROR_RECOVERY(3, 1492 sdev_printk(KERN_INFO, sdev, 1493 "%s: Sending START_UNIT\n", 1494 current->comm)); 1495 1496 if (!scsi_eh_try_stu(stu_scmd)) { 1497 if (!scsi_device_online(sdev) || 1498 !scsi_eh_tur(stu_scmd)) { 1499 list_for_each_entry_safe(scmd, next, 1500 work_q, eh_entry) { 1501 if (scmd->device == sdev && 1502 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1503 scsi_eh_finish_cmd(scmd, done_q); 1504 } 1505 } 1506 } else { 1507 SCSI_LOG_ERROR_RECOVERY(3, 1508 sdev_printk(KERN_INFO, sdev, 1509 "%s: START_UNIT failed\n", 1510 current->comm)); 1511 } 1512 } 1513 1514 return list_empty(work_q); 1515 } 1516 1517 1518 /** 1519 * scsi_eh_bus_device_reset - send bdr if needed 1520 * @shost: scsi host being recovered. 1521 * @work_q: &list_head for pending commands. 1522 * @done_q: &list_head for processed commands. 1523 * 1524 * Notes: 1525 * Try a bus device reset. Still, look to see whether we have multiple 1526 * devices that are jammed or not - if we have multiple devices, it 1527 * makes no sense to try bus_device_reset - we really would need to try 1528 * a bus_reset instead. 1529 */ 1530 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1531 struct list_head *work_q, 1532 struct list_head *done_q) 1533 { 1534 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1535 struct scsi_device *sdev; 1536 enum scsi_disposition rtn; 1537 1538 shost_for_each_device(sdev, shost) { 1539 if (scsi_host_eh_past_deadline(shost)) { 1540 SCSI_LOG_ERROR_RECOVERY(3, 1541 sdev_printk(KERN_INFO, sdev, 1542 "%s: skip BDR, past eh deadline\n", 1543 current->comm)); 1544 scsi_device_put(sdev); 1545 break; 1546 } 1547 bdr_scmd = NULL; 1548 list_for_each_entry(scmd, work_q, eh_entry) 1549 if (scmd->device == sdev) { 1550 bdr_scmd = scmd; 1551 break; 1552 } 1553 1554 if (!bdr_scmd) 1555 continue; 1556 1557 SCSI_LOG_ERROR_RECOVERY(3, 1558 sdev_printk(KERN_INFO, sdev, 1559 "%s: Sending BDR\n", current->comm)); 1560 rtn = scsi_try_bus_device_reset(bdr_scmd); 1561 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1562 if (!scsi_device_online(sdev) || 1563 rtn == FAST_IO_FAIL || 1564 !scsi_eh_tur(bdr_scmd)) { 1565 list_for_each_entry_safe(scmd, next, 1566 work_q, eh_entry) { 1567 if (scmd->device == sdev && 1568 scsi_eh_action(scmd, rtn) != FAILED) 1569 scsi_eh_finish_cmd(scmd, 1570 done_q); 1571 } 1572 } 1573 } else { 1574 SCSI_LOG_ERROR_RECOVERY(3, 1575 sdev_printk(KERN_INFO, sdev, 1576 "%s: BDR failed\n", current->comm)); 1577 } 1578 } 1579 1580 return list_empty(work_q); 1581 } 1582 1583 /** 1584 * scsi_eh_target_reset - send target reset if needed 1585 * @shost: scsi host being recovered. 1586 * @work_q: &list_head for pending commands. 1587 * @done_q: &list_head for processed commands. 1588 * 1589 * Notes: 1590 * Try a target reset. 1591 */ 1592 static int scsi_eh_target_reset(struct Scsi_Host *shost, 1593 struct list_head *work_q, 1594 struct list_head *done_q) 1595 { 1596 LIST_HEAD(tmp_list); 1597 LIST_HEAD(check_list); 1598 1599 list_splice_init(work_q, &tmp_list); 1600 1601 while (!list_empty(&tmp_list)) { 1602 struct scsi_cmnd *next, *scmd; 1603 enum scsi_disposition rtn; 1604 unsigned int id; 1605 1606 if (scsi_host_eh_past_deadline(shost)) { 1607 /* push back on work queue for further processing */ 1608 list_splice_init(&check_list, work_q); 1609 list_splice_init(&tmp_list, work_q); 1610 SCSI_LOG_ERROR_RECOVERY(3, 1611 shost_printk(KERN_INFO, shost, 1612 "%s: Skip target reset, past eh deadline\n", 1613 current->comm)); 1614 return list_empty(work_q); 1615 } 1616 1617 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1618 id = scmd_id(scmd); 1619 1620 SCSI_LOG_ERROR_RECOVERY(3, 1621 shost_printk(KERN_INFO, shost, 1622 "%s: Sending target reset to target %d\n", 1623 current->comm, id)); 1624 rtn = scsi_try_target_reset(scmd); 1625 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1626 SCSI_LOG_ERROR_RECOVERY(3, 1627 shost_printk(KERN_INFO, shost, 1628 "%s: Target reset failed" 1629 " target: %d\n", 1630 current->comm, id)); 1631 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1632 if (scmd_id(scmd) != id) 1633 continue; 1634 1635 if (rtn == SUCCESS) 1636 list_move_tail(&scmd->eh_entry, &check_list); 1637 else if (rtn == FAST_IO_FAIL) 1638 scsi_eh_finish_cmd(scmd, done_q); 1639 else 1640 /* push back on work queue for further processing */ 1641 list_move(&scmd->eh_entry, work_q); 1642 } 1643 } 1644 1645 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1646 } 1647 1648 /** 1649 * scsi_eh_bus_reset - send a bus reset 1650 * @shost: &scsi host being recovered. 1651 * @work_q: &list_head for pending commands. 1652 * @done_q: &list_head for processed commands. 1653 */ 1654 static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1655 struct list_head *work_q, 1656 struct list_head *done_q) 1657 { 1658 struct scsi_cmnd *scmd, *chan_scmd, *next; 1659 LIST_HEAD(check_list); 1660 unsigned int channel; 1661 enum scsi_disposition rtn; 1662 1663 /* 1664 * we really want to loop over the various channels, and do this on 1665 * a channel by channel basis. we should also check to see if any 1666 * of the failed commands are on soft_reset devices, and if so, skip 1667 * the reset. 1668 */ 1669 1670 for (channel = 0; channel <= shost->max_channel; channel++) { 1671 if (scsi_host_eh_past_deadline(shost)) { 1672 list_splice_init(&check_list, work_q); 1673 SCSI_LOG_ERROR_RECOVERY(3, 1674 shost_printk(KERN_INFO, shost, 1675 "%s: skip BRST, past eh deadline\n", 1676 current->comm)); 1677 return list_empty(work_q); 1678 } 1679 1680 chan_scmd = NULL; 1681 list_for_each_entry(scmd, work_q, eh_entry) { 1682 if (channel == scmd_channel(scmd)) { 1683 chan_scmd = scmd; 1684 break; 1685 /* 1686 * FIXME add back in some support for 1687 * soft_reset devices. 1688 */ 1689 } 1690 } 1691 1692 if (!chan_scmd) 1693 continue; 1694 SCSI_LOG_ERROR_RECOVERY(3, 1695 shost_printk(KERN_INFO, shost, 1696 "%s: Sending BRST chan: %d\n", 1697 current->comm, channel)); 1698 rtn = scsi_try_bus_reset(chan_scmd); 1699 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1700 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1701 if (channel == scmd_channel(scmd)) { 1702 if (rtn == FAST_IO_FAIL) 1703 scsi_eh_finish_cmd(scmd, 1704 done_q); 1705 else 1706 list_move_tail(&scmd->eh_entry, 1707 &check_list); 1708 } 1709 } 1710 } else { 1711 SCSI_LOG_ERROR_RECOVERY(3, 1712 shost_printk(KERN_INFO, shost, 1713 "%s: BRST failed chan: %d\n", 1714 current->comm, channel)); 1715 } 1716 } 1717 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1718 } 1719 1720 /** 1721 * scsi_eh_host_reset - send a host reset 1722 * @shost: host to be reset. 1723 * @work_q: &list_head for pending commands. 1724 * @done_q: &list_head for processed commands. 1725 */ 1726 static int scsi_eh_host_reset(struct Scsi_Host *shost, 1727 struct list_head *work_q, 1728 struct list_head *done_q) 1729 { 1730 struct scsi_cmnd *scmd, *next; 1731 LIST_HEAD(check_list); 1732 enum scsi_disposition rtn; 1733 1734 if (!list_empty(work_q)) { 1735 scmd = list_entry(work_q->next, 1736 struct scsi_cmnd, eh_entry); 1737 1738 SCSI_LOG_ERROR_RECOVERY(3, 1739 shost_printk(KERN_INFO, shost, 1740 "%s: Sending HRST\n", 1741 current->comm)); 1742 1743 rtn = scsi_try_host_reset(scmd); 1744 if (rtn == SUCCESS) { 1745 list_splice_init(work_q, &check_list); 1746 } else if (rtn == FAST_IO_FAIL) { 1747 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1748 scsi_eh_finish_cmd(scmd, done_q); 1749 } 1750 } else { 1751 SCSI_LOG_ERROR_RECOVERY(3, 1752 shost_printk(KERN_INFO, shost, 1753 "%s: HRST failed\n", 1754 current->comm)); 1755 } 1756 } 1757 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1758 } 1759 1760 /** 1761 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1762 * @work_q: &list_head for pending commands. 1763 * @done_q: &list_head for processed commands. 1764 */ 1765 static void scsi_eh_offline_sdevs(struct list_head *work_q, 1766 struct list_head *done_q) 1767 { 1768 struct scsi_cmnd *scmd, *next; 1769 struct scsi_device *sdev; 1770 1771 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1772 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1773 "not ready after error recovery\n"); 1774 sdev = scmd->device; 1775 1776 mutex_lock(&sdev->state_mutex); 1777 scsi_device_set_state(sdev, SDEV_OFFLINE); 1778 mutex_unlock(&sdev->state_mutex); 1779 1780 scsi_eh_finish_cmd(scmd, done_q); 1781 } 1782 return; 1783 } 1784 1785 /** 1786 * scsi_noretry_cmd - determine if command should be failed fast 1787 * @scmd: SCSI cmd to examine. 1788 */ 1789 bool scsi_noretry_cmd(struct scsi_cmnd *scmd) 1790 { 1791 struct request *req = scsi_cmd_to_rq(scmd); 1792 1793 switch (host_byte(scmd->result)) { 1794 case DID_OK: 1795 break; 1796 case DID_TIME_OUT: 1797 goto check_type; 1798 case DID_BUS_BUSY: 1799 return !!(req->cmd_flags & REQ_FAILFAST_TRANSPORT); 1800 case DID_PARITY: 1801 return !!(req->cmd_flags & REQ_FAILFAST_DEV); 1802 case DID_ERROR: 1803 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT) 1804 return false; 1805 fallthrough; 1806 case DID_SOFT_ERROR: 1807 return !!(req->cmd_flags & REQ_FAILFAST_DRIVER); 1808 } 1809 1810 if (!scsi_status_is_check_condition(scmd->result)) 1811 return false; 1812 1813 check_type: 1814 /* 1815 * assume caller has checked sense and determined 1816 * the check condition was retryable. 1817 */ 1818 if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(req)) 1819 return true; 1820 1821 return false; 1822 } 1823 1824 /** 1825 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1826 * @scmd: SCSI cmd to examine. 1827 * 1828 * Notes: 1829 * This is *only* called when we are examining the status after sending 1830 * out the actual data command. any commands that are queued for error 1831 * recovery (e.g. test_unit_ready) do *not* come through here. 1832 * 1833 * When this routine returns failed, it means the error handler thread 1834 * is woken. In cases where the error code indicates an error that 1835 * doesn't require the error handler read (i.e. we don't need to 1836 * abort/reset), this function should return SUCCESS. 1837 */ 1838 enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd) 1839 { 1840 enum scsi_disposition rtn; 1841 1842 /* 1843 * if the device is offline, then we clearly just pass the result back 1844 * up to the top level. 1845 */ 1846 if (!scsi_device_online(scmd->device)) { 1847 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1848 "%s: device offline - report as SUCCESS\n", __func__)); 1849 return SUCCESS; 1850 } 1851 1852 /* 1853 * first check the host byte, to see if there is anything in there 1854 * that would indicate what we need to do. 1855 */ 1856 switch (host_byte(scmd->result)) { 1857 case DID_PASSTHROUGH: 1858 /* 1859 * no matter what, pass this through to the upper layer. 1860 * nuke this special code so that it looks like we are saying 1861 * did_ok. 1862 */ 1863 scmd->result &= 0xff00ffff; 1864 return SUCCESS; 1865 case DID_OK: 1866 /* 1867 * looks good. drop through, and check the next byte. 1868 */ 1869 break; 1870 case DID_ABORT: 1871 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1872 set_host_byte(scmd, DID_TIME_OUT); 1873 return SUCCESS; 1874 } 1875 fallthrough; 1876 case DID_NO_CONNECT: 1877 case DID_BAD_TARGET: 1878 /* 1879 * note - this means that we just report the status back 1880 * to the top level driver, not that we actually think 1881 * that it indicates SUCCESS. 1882 */ 1883 return SUCCESS; 1884 case DID_SOFT_ERROR: 1885 /* 1886 * when the low level driver returns did_soft_error, 1887 * it is responsible for keeping an internal retry counter 1888 * in order to avoid endless loops (db) 1889 */ 1890 goto maybe_retry; 1891 case DID_IMM_RETRY: 1892 return NEEDS_RETRY; 1893 1894 case DID_REQUEUE: 1895 return ADD_TO_MLQUEUE; 1896 case DID_TRANSPORT_DISRUPTED: 1897 /* 1898 * LLD/transport was disrupted during processing of the IO. 1899 * The transport class is now blocked/blocking, 1900 * and the transport will decide what to do with the IO 1901 * based on its timers and recovery capablilities if 1902 * there are enough retries. 1903 */ 1904 goto maybe_retry; 1905 case DID_TRANSPORT_FAILFAST: 1906 /* 1907 * The transport decided to failfast the IO (most likely 1908 * the fast io fail tmo fired), so send IO directly upwards. 1909 */ 1910 return SUCCESS; 1911 case DID_TRANSPORT_MARGINAL: 1912 /* 1913 * caller has decided not to do retries on 1914 * abort success, so send IO directly upwards 1915 */ 1916 return SUCCESS; 1917 case DID_ERROR: 1918 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT) 1919 /* 1920 * execute reservation conflict processing code 1921 * lower down 1922 */ 1923 break; 1924 fallthrough; 1925 case DID_BUS_BUSY: 1926 case DID_PARITY: 1927 goto maybe_retry; 1928 case DID_TIME_OUT: 1929 /* 1930 * when we scan the bus, we get timeout messages for 1931 * these commands if there is no device available. 1932 * other hosts report did_no_connect for the same thing. 1933 */ 1934 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1935 scmd->cmnd[0] == INQUIRY)) { 1936 return SUCCESS; 1937 } else { 1938 return FAILED; 1939 } 1940 case DID_RESET: 1941 return SUCCESS; 1942 default: 1943 return FAILED; 1944 } 1945 1946 /* 1947 * check the status byte to see if this indicates anything special. 1948 */ 1949 switch (get_status_byte(scmd)) { 1950 case SAM_STAT_TASK_SET_FULL: 1951 scsi_handle_queue_full(scmd->device); 1952 /* 1953 * the case of trying to send too many commands to a 1954 * tagged queueing device. 1955 */ 1956 fallthrough; 1957 case SAM_STAT_BUSY: 1958 /* 1959 * device can't talk to us at the moment. Should only 1960 * occur (SAM-3) when the task queue is empty, so will cause 1961 * the empty queue handling to trigger a stall in the 1962 * device. 1963 */ 1964 return ADD_TO_MLQUEUE; 1965 case SAM_STAT_GOOD: 1966 if (scmd->cmnd[0] == REPORT_LUNS) 1967 scmd->device->sdev_target->expecting_lun_change = 0; 1968 scsi_handle_queue_ramp_up(scmd->device); 1969 fallthrough; 1970 case SAM_STAT_COMMAND_TERMINATED: 1971 return SUCCESS; 1972 case SAM_STAT_TASK_ABORTED: 1973 goto maybe_retry; 1974 case SAM_STAT_CHECK_CONDITION: 1975 rtn = scsi_check_sense(scmd); 1976 if (rtn == NEEDS_RETRY) 1977 goto maybe_retry; 1978 /* if rtn == FAILED, we have no sense information; 1979 * returning FAILED will wake the error handler thread 1980 * to collect the sense and redo the decide 1981 * disposition */ 1982 return rtn; 1983 case SAM_STAT_CONDITION_MET: 1984 case SAM_STAT_INTERMEDIATE: 1985 case SAM_STAT_INTERMEDIATE_CONDITION_MET: 1986 case SAM_STAT_ACA_ACTIVE: 1987 /* 1988 * who knows? FIXME(eric) 1989 */ 1990 return SUCCESS; 1991 1992 case SAM_STAT_RESERVATION_CONFLICT: 1993 sdev_printk(KERN_INFO, scmd->device, 1994 "reservation conflict\n"); 1995 set_scsi_ml_byte(scmd, SCSIML_STAT_RESV_CONFLICT); 1996 return SUCCESS; /* causes immediate i/o error */ 1997 } 1998 return FAILED; 1999 2000 maybe_retry: 2001 2002 /* we requeue for retry because the error was retryable, and 2003 * the request was not marked fast fail. Note that above, 2004 * even if the request is marked fast fail, we still requeue 2005 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 2006 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) { 2007 return NEEDS_RETRY; 2008 } else { 2009 /* 2010 * no more retries - report this one back to upper level. 2011 */ 2012 return SUCCESS; 2013 } 2014 } 2015 2016 static enum rq_end_io_ret eh_lock_door_done(struct request *req, 2017 blk_status_t status) 2018 { 2019 blk_mq_free_request(req); 2020 return RQ_END_IO_NONE; 2021 } 2022 2023 /** 2024 * scsi_eh_lock_door - Prevent medium removal for the specified device 2025 * @sdev: SCSI device to prevent medium removal 2026 * 2027 * Locking: 2028 * We must be called from process context. 2029 * 2030 * Notes: 2031 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 2032 * head of the devices request queue, and continue. 2033 */ 2034 static void scsi_eh_lock_door(struct scsi_device *sdev) 2035 { 2036 struct scsi_cmnd *scmd; 2037 struct request *req; 2038 2039 req = scsi_alloc_request(sdev->request_queue, REQ_OP_DRV_IN, 0); 2040 if (IS_ERR(req)) 2041 return; 2042 scmd = blk_mq_rq_to_pdu(req); 2043 2044 scmd->cmnd[0] = ALLOW_MEDIUM_REMOVAL; 2045 scmd->cmnd[1] = 0; 2046 scmd->cmnd[2] = 0; 2047 scmd->cmnd[3] = 0; 2048 scmd->cmnd[4] = SCSI_REMOVAL_PREVENT; 2049 scmd->cmnd[5] = 0; 2050 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 2051 scmd->allowed = 5; 2052 2053 req->rq_flags |= RQF_QUIET; 2054 req->timeout = 10 * HZ; 2055 req->end_io = eh_lock_door_done; 2056 2057 blk_execute_rq_nowait(req, true); 2058 } 2059 2060 /** 2061 * scsi_restart_operations - restart io operations to the specified host. 2062 * @shost: Host we are restarting. 2063 * 2064 * Notes: 2065 * When we entered the error handler, we blocked all further i/o to 2066 * this device. we need to 'reverse' this process. 2067 */ 2068 static void scsi_restart_operations(struct Scsi_Host *shost) 2069 { 2070 struct scsi_device *sdev; 2071 unsigned long flags; 2072 2073 /* 2074 * If the door was locked, we need to insert a door lock request 2075 * onto the head of the SCSI request queue for the device. There 2076 * is no point trying to lock the door of an off-line device. 2077 */ 2078 shost_for_each_device(sdev, shost) { 2079 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 2080 scsi_eh_lock_door(sdev); 2081 sdev->was_reset = 0; 2082 } 2083 } 2084 2085 /* 2086 * next free up anything directly waiting upon the host. this 2087 * will be requests for character device operations, and also for 2088 * ioctls to queued block devices. 2089 */ 2090 SCSI_LOG_ERROR_RECOVERY(3, 2091 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 2092 2093 spin_lock_irqsave(shost->host_lock, flags); 2094 if (scsi_host_set_state(shost, SHOST_RUNNING)) 2095 if (scsi_host_set_state(shost, SHOST_CANCEL)) 2096 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 2097 spin_unlock_irqrestore(shost->host_lock, flags); 2098 2099 wake_up(&shost->host_wait); 2100 2101 /* 2102 * finally we need to re-initiate requests that may be pending. we will 2103 * have had everything blocked while error handling is taking place, and 2104 * now that error recovery is done, we will need to ensure that these 2105 * requests are started. 2106 */ 2107 scsi_run_host_queues(shost); 2108 2109 /* 2110 * if eh is active and host_eh_scheduled is pending we need to re-run 2111 * recovery. we do this check after scsi_run_host_queues() to allow 2112 * everything pent up since the last eh run a chance to make forward 2113 * progress before we sync again. Either we'll immediately re-run 2114 * recovery or scsi_device_unbusy() will wake us again when these 2115 * pending commands complete. 2116 */ 2117 spin_lock_irqsave(shost->host_lock, flags); 2118 if (shost->host_eh_scheduled) 2119 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 2120 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 2121 spin_unlock_irqrestore(shost->host_lock, flags); 2122 } 2123 2124 /** 2125 * scsi_eh_ready_devs - check device ready state and recover if not. 2126 * @shost: host to be recovered. 2127 * @work_q: &list_head for pending commands. 2128 * @done_q: &list_head for processed commands. 2129 */ 2130 void scsi_eh_ready_devs(struct Scsi_Host *shost, 2131 struct list_head *work_q, 2132 struct list_head *done_q) 2133 { 2134 if (!scsi_eh_stu(shost, work_q, done_q)) 2135 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2136 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2137 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2138 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2139 scsi_eh_offline_sdevs(work_q, 2140 done_q); 2141 } 2142 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2143 2144 /** 2145 * scsi_eh_flush_done_q - finish processed commands or retry them. 2146 * @done_q: list_head of processed commands. 2147 */ 2148 void scsi_eh_flush_done_q(struct list_head *done_q) 2149 { 2150 struct scsi_cmnd *scmd, *next; 2151 2152 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2153 list_del_init(&scmd->eh_entry); 2154 if (scsi_device_online(scmd->device) && 2155 !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd) && 2156 scsi_eh_should_retry_cmd(scmd)) { 2157 SCSI_LOG_ERROR_RECOVERY(3, 2158 scmd_printk(KERN_INFO, scmd, 2159 "%s: flush retry cmd\n", 2160 current->comm)); 2161 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2162 } else { 2163 /* 2164 * If just we got sense for the device (called 2165 * scsi_eh_get_sense), scmd->result is already 2166 * set, do not set DID_TIME_OUT. 2167 */ 2168 if (!scmd->result) 2169 scmd->result |= (DID_TIME_OUT << 16); 2170 SCSI_LOG_ERROR_RECOVERY(3, 2171 scmd_printk(KERN_INFO, scmd, 2172 "%s: flush finish cmd\n", 2173 current->comm)); 2174 scsi_finish_command(scmd); 2175 } 2176 } 2177 } 2178 EXPORT_SYMBOL(scsi_eh_flush_done_q); 2179 2180 /** 2181 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2182 * @shost: Host to unjam. 2183 * 2184 * Notes: 2185 * When we come in here, we *know* that all commands on the bus have 2186 * either completed, failed or timed out. we also know that no further 2187 * commands are being sent to the host, so things are relatively quiet 2188 * and we have freedom to fiddle with things as we wish. 2189 * 2190 * This is only the *default* implementation. it is possible for 2191 * individual drivers to supply their own version of this function, and 2192 * if the maintainer wishes to do this, it is strongly suggested that 2193 * this function be taken as a template and modified. this function 2194 * was designed to correctly handle problems for about 95% of the 2195 * different cases out there, and it should always provide at least a 2196 * reasonable amount of error recovery. 2197 * 2198 * Any command marked 'failed' or 'timeout' must eventually have 2199 * scsi_finish_cmd() called for it. we do all of the retry stuff 2200 * here, so when we restart the host after we return it should have an 2201 * empty queue. 2202 */ 2203 static void scsi_unjam_host(struct Scsi_Host *shost) 2204 { 2205 unsigned long flags; 2206 LIST_HEAD(eh_work_q); 2207 LIST_HEAD(eh_done_q); 2208 2209 spin_lock_irqsave(shost->host_lock, flags); 2210 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2211 spin_unlock_irqrestore(shost->host_lock, flags); 2212 2213 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2214 2215 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2216 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2217 2218 spin_lock_irqsave(shost->host_lock, flags); 2219 if (shost->eh_deadline != -1) 2220 shost->last_reset = 0; 2221 spin_unlock_irqrestore(shost->host_lock, flags); 2222 scsi_eh_flush_done_q(&eh_done_q); 2223 } 2224 2225 /** 2226 * scsi_error_handler - SCSI error handler thread 2227 * @data: Host for which we are running. 2228 * 2229 * Notes: 2230 * This is the main error handling loop. This is run as a kernel thread 2231 * for every SCSI host and handles all error handling activity. 2232 */ 2233 int scsi_error_handler(void *data) 2234 { 2235 struct Scsi_Host *shost = data; 2236 2237 /* 2238 * We use TASK_INTERRUPTIBLE so that the thread is not 2239 * counted against the load average as a running process. 2240 * We never actually get interrupted because kthread_run 2241 * disables signal delivery for the created thread. 2242 */ 2243 while (true) { 2244 /* 2245 * The sequence in kthread_stop() sets the stop flag first 2246 * then wakes the process. To avoid missed wakeups, the task 2247 * should always be in a non running state before the stop 2248 * flag is checked 2249 */ 2250 set_current_state(TASK_INTERRUPTIBLE); 2251 if (kthread_should_stop()) 2252 break; 2253 2254 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2255 shost->host_failed != scsi_host_busy(shost)) { 2256 SCSI_LOG_ERROR_RECOVERY(1, 2257 shost_printk(KERN_INFO, shost, 2258 "scsi_eh_%d: sleeping\n", 2259 shost->host_no)); 2260 schedule(); 2261 continue; 2262 } 2263 2264 __set_current_state(TASK_RUNNING); 2265 SCSI_LOG_ERROR_RECOVERY(1, 2266 shost_printk(KERN_INFO, shost, 2267 "scsi_eh_%d: waking up %d/%d/%d\n", 2268 shost->host_no, shost->host_eh_scheduled, 2269 shost->host_failed, 2270 scsi_host_busy(shost))); 2271 2272 /* 2273 * We have a host that is failing for some reason. Figure out 2274 * what we need to do to get it up and online again (if we can). 2275 * If we fail, we end up taking the thing offline. 2276 */ 2277 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { 2278 SCSI_LOG_ERROR_RECOVERY(1, 2279 shost_printk(KERN_ERR, shost, 2280 "scsi_eh_%d: unable to autoresume\n", 2281 shost->host_no)); 2282 continue; 2283 } 2284 2285 if (shost->transportt->eh_strategy_handler) 2286 shost->transportt->eh_strategy_handler(shost); 2287 else 2288 scsi_unjam_host(shost); 2289 2290 /* All scmds have been handled */ 2291 shost->host_failed = 0; 2292 2293 /* 2294 * Note - if the above fails completely, the action is to take 2295 * individual devices offline and flush the queue of any 2296 * outstanding requests that may have been pending. When we 2297 * restart, we restart any I/O to any other devices on the bus 2298 * which are still online. 2299 */ 2300 scsi_restart_operations(shost); 2301 if (!shost->eh_noresume) 2302 scsi_autopm_put_host(shost); 2303 } 2304 __set_current_state(TASK_RUNNING); 2305 2306 SCSI_LOG_ERROR_RECOVERY(1, 2307 shost_printk(KERN_INFO, shost, 2308 "Error handler scsi_eh_%d exiting\n", 2309 shost->host_no)); 2310 shost->ehandler = NULL; 2311 return 0; 2312 } 2313 2314 /* 2315 * Function: scsi_report_bus_reset() 2316 * 2317 * Purpose: Utility function used by low-level drivers to report that 2318 * they have observed a bus reset on the bus being handled. 2319 * 2320 * Arguments: shost - Host in question 2321 * channel - channel on which reset was observed. 2322 * 2323 * Returns: Nothing 2324 * 2325 * Lock status: Host lock must be held. 2326 * 2327 * Notes: This only needs to be called if the reset is one which 2328 * originates from an unknown location. Resets originated 2329 * by the mid-level itself don't need to call this, but there 2330 * should be no harm. 2331 * 2332 * The main purpose of this is to make sure that a CHECK_CONDITION 2333 * is properly treated. 2334 */ 2335 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2336 { 2337 struct scsi_device *sdev; 2338 2339 __shost_for_each_device(sdev, shost) { 2340 if (channel == sdev_channel(sdev)) 2341 __scsi_report_device_reset(sdev, NULL); 2342 } 2343 } 2344 EXPORT_SYMBOL(scsi_report_bus_reset); 2345 2346 /* 2347 * Function: scsi_report_device_reset() 2348 * 2349 * Purpose: Utility function used by low-level drivers to report that 2350 * they have observed a device reset on the device being handled. 2351 * 2352 * Arguments: shost - Host in question 2353 * channel - channel on which reset was observed 2354 * target - target on which reset was observed 2355 * 2356 * Returns: Nothing 2357 * 2358 * Lock status: Host lock must be held 2359 * 2360 * Notes: This only needs to be called if the reset is one which 2361 * originates from an unknown location. Resets originated 2362 * by the mid-level itself don't need to call this, but there 2363 * should be no harm. 2364 * 2365 * The main purpose of this is to make sure that a CHECK_CONDITION 2366 * is properly treated. 2367 */ 2368 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2369 { 2370 struct scsi_device *sdev; 2371 2372 __shost_for_each_device(sdev, shost) { 2373 if (channel == sdev_channel(sdev) && 2374 target == sdev_id(sdev)) 2375 __scsi_report_device_reset(sdev, NULL); 2376 } 2377 } 2378 EXPORT_SYMBOL(scsi_report_device_reset); 2379 2380 /** 2381 * scsi_ioctl_reset: explicitly reset a host/bus/target/device 2382 * @dev: scsi_device to operate on 2383 * @arg: reset type (see sg.h) 2384 */ 2385 int 2386 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) 2387 { 2388 struct scsi_cmnd *scmd; 2389 struct Scsi_Host *shost = dev->host; 2390 struct request *rq; 2391 unsigned long flags; 2392 int error = 0, val; 2393 enum scsi_disposition rtn; 2394 2395 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) 2396 return -EACCES; 2397 2398 error = get_user(val, arg); 2399 if (error) 2400 return error; 2401 2402 if (scsi_autopm_get_host(shost) < 0) 2403 return -EIO; 2404 2405 error = -EIO; 2406 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) + 2407 shost->hostt->cmd_size, GFP_KERNEL); 2408 if (!rq) 2409 goto out_put_autopm_host; 2410 blk_rq_init(NULL, rq); 2411 2412 scmd = (struct scsi_cmnd *)(rq + 1); 2413 scsi_init_command(dev, scmd); 2414 2415 scmd->submitter = SUBMITTED_BY_SCSI_RESET_IOCTL; 2416 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 2417 2418 scmd->cmd_len = 0; 2419 2420 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 2421 2422 spin_lock_irqsave(shost->host_lock, flags); 2423 shost->tmf_in_progress = 1; 2424 spin_unlock_irqrestore(shost->host_lock, flags); 2425 2426 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { 2427 case SG_SCSI_RESET_NOTHING: 2428 rtn = SUCCESS; 2429 break; 2430 case SG_SCSI_RESET_DEVICE: 2431 rtn = scsi_try_bus_device_reset(scmd); 2432 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2433 break; 2434 fallthrough; 2435 case SG_SCSI_RESET_TARGET: 2436 rtn = scsi_try_target_reset(scmd); 2437 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2438 break; 2439 fallthrough; 2440 case SG_SCSI_RESET_BUS: 2441 rtn = scsi_try_bus_reset(scmd); 2442 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2443 break; 2444 fallthrough; 2445 case SG_SCSI_RESET_HOST: 2446 rtn = scsi_try_host_reset(scmd); 2447 if (rtn == SUCCESS) 2448 break; 2449 fallthrough; 2450 default: 2451 rtn = FAILED; 2452 break; 2453 } 2454 2455 error = (rtn == SUCCESS) ? 0 : -EIO; 2456 2457 spin_lock_irqsave(shost->host_lock, flags); 2458 shost->tmf_in_progress = 0; 2459 spin_unlock_irqrestore(shost->host_lock, flags); 2460 2461 /* 2462 * be sure to wake up anyone who was sleeping or had their queue 2463 * suspended while we performed the TMF. 2464 */ 2465 SCSI_LOG_ERROR_RECOVERY(3, 2466 shost_printk(KERN_INFO, shost, 2467 "waking up host to restart after TMF\n")); 2468 2469 wake_up(&shost->host_wait); 2470 scsi_run_host_queues(shost); 2471 2472 kfree(rq); 2473 2474 out_put_autopm_host: 2475 scsi_autopm_put_host(shost); 2476 return error; 2477 } 2478 2479 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, 2480 struct scsi_sense_hdr *sshdr) 2481 { 2482 return scsi_normalize_sense(cmd->sense_buffer, 2483 SCSI_SENSE_BUFFERSIZE, sshdr); 2484 } 2485 EXPORT_SYMBOL(scsi_command_normalize_sense); 2486 2487 /** 2488 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2489 * @sense_buffer: byte array of sense data 2490 * @sb_len: number of valid bytes in sense_buffer 2491 * @info_out: pointer to 64 integer where 8 or 4 byte information 2492 * field will be placed if found. 2493 * 2494 * Return value: 2495 * true if information field found, false if not found. 2496 */ 2497 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len, 2498 u64 *info_out) 2499 { 2500 const u8 * ucp; 2501 2502 if (sb_len < 7) 2503 return false; 2504 switch (sense_buffer[0] & 0x7f) { 2505 case 0x70: 2506 case 0x71: 2507 if (sense_buffer[0] & 0x80) { 2508 *info_out = get_unaligned_be32(&sense_buffer[3]); 2509 return true; 2510 } 2511 return false; 2512 case 0x72: 2513 case 0x73: 2514 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2515 0 /* info desc */); 2516 if (ucp && (0xa == ucp[1])) { 2517 *info_out = get_unaligned_be64(&ucp[4]); 2518 return true; 2519 } 2520 return false; 2521 default: 2522 return false; 2523 } 2524 } 2525 EXPORT_SYMBOL(scsi_get_sense_info_fld); 2526