scsi_error.c (71675B)
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 60static int scsi_eh_try_stu(struct scsi_cmnd *scmd); 61static enum scsi_disposition scsi_try_to_abort_cmd(struct scsi_host_template *, 62 struct scsi_cmnd *); 63 64void 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 */ 82void 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} 96EXPORT_SYMBOL_GPL(scsi_schedule_eh); 97 98static 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 117static 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 125static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd) 126{ 127 struct scsi_device *sdev = cmd->device; 128 struct Scsi_Host *host = sdev->host; 129 130 if (host->hostt->eh_should_retry_cmd) 131 return host->hostt->eh_should_retry_cmd(cmd); 132 133 return true; 134} 135 136/** 137 * scmd_eh_abort_handler - Handle command aborts 138 * @work: command to be aborted. 139 * 140 * Note: this function must be called only for a command that has timed out. 141 * Because the block layer marks a request as complete before it calls 142 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has 143 * timed out do not have any effect. Hence it is safe to call 144 * scsi_finish_command() from this function. 145 */ 146void 147scmd_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 213out: 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 */ 227static int 228scsi_abort_command(struct scsi_cmnd *scmd) 229{ 230 struct scsi_device *sdev = scmd->device; 231 struct Scsi_Host *shost = sdev->host; 232 unsigned long flags; 233 234 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 235 /* 236 * Retry after abort failed, escalate to next level. 237 */ 238 SCSI_LOG_ERROR_RECOVERY(3, 239 scmd_printk(KERN_INFO, scmd, 240 "previous abort failed\n")); 241 BUG_ON(delayed_work_pending(&scmd->abort_work)); 242 return FAILED; 243 } 244 245 spin_lock_irqsave(shost->host_lock, flags); 246 if (shost->eh_deadline != -1 && !shost->last_reset) 247 shost->last_reset = jiffies; 248 BUG_ON(!list_empty(&scmd->eh_entry)); 249 list_add_tail(&scmd->eh_entry, &shost->eh_abort_list); 250 spin_unlock_irqrestore(shost->host_lock, flags); 251 252 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; 253 SCSI_LOG_ERROR_RECOVERY(3, 254 scmd_printk(KERN_INFO, scmd, "abort scheduled\n")); 255 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100); 256 return SUCCESS; 257} 258 259/** 260 * scsi_eh_reset - call into ->eh_action to reset internal counters 261 * @scmd: scmd to run eh on. 262 * 263 * The scsi driver might be carrying internal state about the 264 * devices, so we need to call into the driver to reset the 265 * internal state once the error handler is started. 266 */ 267static void scsi_eh_reset(struct scsi_cmnd *scmd) 268{ 269 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) { 270 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 271 if (sdrv->eh_reset) 272 sdrv->eh_reset(scmd); 273 } 274} 275 276static void scsi_eh_inc_host_failed(struct rcu_head *head) 277{ 278 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu); 279 struct Scsi_Host *shost = scmd->device->host; 280 unsigned long flags; 281 282 spin_lock_irqsave(shost->host_lock, flags); 283 shost->host_failed++; 284 scsi_eh_wakeup(shost); 285 spin_unlock_irqrestore(shost->host_lock, flags); 286} 287 288/** 289 * scsi_eh_scmd_add - add scsi cmd to error handling. 290 * @scmd: scmd to run eh on. 291 */ 292void scsi_eh_scmd_add(struct scsi_cmnd *scmd) 293{ 294 struct Scsi_Host *shost = scmd->device->host; 295 unsigned long flags; 296 int ret; 297 298 WARN_ON_ONCE(!shost->ehandler); 299 300 spin_lock_irqsave(shost->host_lock, flags); 301 if (scsi_host_set_state(shost, SHOST_RECOVERY)) { 302 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY); 303 WARN_ON_ONCE(ret); 304 } 305 if (shost->eh_deadline != -1 && !shost->last_reset) 306 shost->last_reset = jiffies; 307 308 scsi_eh_reset(scmd); 309 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); 310 spin_unlock_irqrestore(shost->host_lock, flags); 311 /* 312 * Ensure that all tasks observe the host state change before the 313 * host_failed change. 314 */ 315 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed); 316} 317 318/** 319 * scsi_times_out - Timeout function for normal scsi commands. 320 * @req: request that is timing out. 321 * 322 * Notes: 323 * We do not need to lock this. There is the potential for a race 324 * only in that the normal completion handling might run, but if the 325 * normal completion function determines that the timer has already 326 * fired, then it mustn't do anything. 327 */ 328enum blk_eh_timer_return scsi_times_out(struct request *req) 329{ 330 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req); 331 enum blk_eh_timer_return rtn = BLK_EH_DONE; 332 struct Scsi_Host *host = scmd->device->host; 333 334 trace_scsi_dispatch_cmd_timeout(scmd); 335 scsi_log_completion(scmd, TIMEOUT_ERROR); 336 337 if (host->eh_deadline != -1 && !host->last_reset) 338 host->last_reset = jiffies; 339 340 if (host->hostt->eh_timed_out) 341 rtn = host->hostt->eh_timed_out(scmd); 342 343 if (rtn == BLK_EH_DONE) { 344 /* 345 * Set the command to complete first in order to prevent a real 346 * completion from releasing the command while error handling 347 * is using it. If the command was already completed, then the 348 * lower level driver beat the timeout handler, and it is safe 349 * to return without escalating error recovery. 350 * 351 * If timeout handling lost the race to a real completion, the 352 * block layer may ignore that due to a fake timeout injection, 353 * so return RESET_TIMER to allow error handling another shot 354 * at this command. 355 */ 356 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state)) 357 return BLK_EH_RESET_TIMER; 358 if (scsi_abort_command(scmd) != SUCCESS) { 359 set_host_byte(scmd, DID_TIME_OUT); 360 scsi_eh_scmd_add(scmd); 361 } 362 } 363 364 return rtn; 365} 366 367/** 368 * scsi_block_when_processing_errors - Prevent cmds from being queued. 369 * @sdev: Device on which we are performing recovery. 370 * 371 * Description: 372 * We block until the host is out of error recovery, and then check to 373 * see whether the host or the device is offline. 374 * 375 * Return value: 376 * 0 when dev was taken offline by error recovery. 1 OK to proceed. 377 */ 378int scsi_block_when_processing_errors(struct scsi_device *sdev) 379{ 380 int online; 381 382 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); 383 384 online = scsi_device_online(sdev); 385 386 return online; 387} 388EXPORT_SYMBOL(scsi_block_when_processing_errors); 389 390#ifdef CONFIG_SCSI_LOGGING 391/** 392 * scsi_eh_prt_fail_stats - Log info on failures. 393 * @shost: scsi host being recovered. 394 * @work_q: Queue of scsi cmds to process. 395 */ 396static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, 397 struct list_head *work_q) 398{ 399 struct scsi_cmnd *scmd; 400 struct scsi_device *sdev; 401 int total_failures = 0; 402 int cmd_failed = 0; 403 int cmd_cancel = 0; 404 int devices_failed = 0; 405 406 shost_for_each_device(sdev, shost) { 407 list_for_each_entry(scmd, work_q, eh_entry) { 408 if (scmd->device == sdev) { 409 ++total_failures; 410 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) 411 ++cmd_cancel; 412 else 413 ++cmd_failed; 414 } 415 } 416 417 if (cmd_cancel || cmd_failed) { 418 SCSI_LOG_ERROR_RECOVERY(3, 419 shost_printk(KERN_INFO, shost, 420 "%s: cmds failed: %d, cancel: %d\n", 421 __func__, cmd_failed, 422 cmd_cancel)); 423 cmd_cancel = 0; 424 cmd_failed = 0; 425 ++devices_failed; 426 } 427 } 428 429 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, 430 "Total of %d commands on %d" 431 " devices require eh work\n", 432 total_failures, devices_failed)); 433} 434#endif 435 436 /** 437 * scsi_report_lun_change - Set flag on all *other* devices on the same target 438 * to indicate that a UNIT ATTENTION is expected. 439 * @sdev: Device reporting the UNIT ATTENTION 440 */ 441static void scsi_report_lun_change(struct scsi_device *sdev) 442{ 443 sdev->sdev_target->expecting_lun_change = 1; 444} 445 446/** 447 * scsi_report_sense - Examine scsi sense information and log messages for 448 * certain conditions, also issue uevents for some of them. 449 * @sdev: Device reporting the sense code 450 * @sshdr: sshdr to be examined 451 */ 452static void scsi_report_sense(struct scsi_device *sdev, 453 struct scsi_sense_hdr *sshdr) 454{ 455 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ 456 457 if (sshdr->sense_key == UNIT_ATTENTION) { 458 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { 459 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; 460 sdev_printk(KERN_WARNING, sdev, 461 "Inquiry data has changed"); 462 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { 463 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; 464 scsi_report_lun_change(sdev); 465 sdev_printk(KERN_WARNING, sdev, 466 "Warning! Received an indication that the " 467 "LUN assignments on this target have " 468 "changed. The Linux SCSI layer does not " 469 "automatically remap LUN assignments.\n"); 470 } else if (sshdr->asc == 0x3f) 471 sdev_printk(KERN_WARNING, sdev, 472 "Warning! Received an indication that the " 473 "operating parameters on this target have " 474 "changed. The Linux SCSI layer does not " 475 "automatically adjust these parameters.\n"); 476 477 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { 478 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; 479 sdev_printk(KERN_WARNING, sdev, 480 "Warning! Received an indication that the " 481 "LUN reached a thin provisioning soft " 482 "threshold.\n"); 483 } 484 485 if (sshdr->asc == 0x29) { 486 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED; 487 /* 488 * Do not print message if it is an expected side-effect 489 * of runtime PM. 490 */ 491 if (!sdev->silence_suspend) 492 sdev_printk(KERN_WARNING, sdev, 493 "Power-on or device reset occurred\n"); 494 } 495 496 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { 497 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; 498 sdev_printk(KERN_WARNING, sdev, 499 "Mode parameters changed"); 500 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { 501 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; 502 sdev_printk(KERN_WARNING, sdev, 503 "Asymmetric access state changed"); 504 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { 505 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; 506 sdev_printk(KERN_WARNING, sdev, 507 "Capacity data has changed"); 508 } else if (sshdr->asc == 0x2a) 509 sdev_printk(KERN_WARNING, sdev, 510 "Parameters changed"); 511 } 512 513 if (evt_type != SDEV_EVT_MAXBITS) { 514 set_bit(evt_type, sdev->pending_events); 515 schedule_work(&sdev->event_work); 516 } 517} 518 519/** 520 * scsi_check_sense - Examine scsi cmd sense 521 * @scmd: Cmd to have sense checked. 522 * 523 * Return value: 524 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE 525 * 526 * Notes: 527 * When a deferred error is detected the current command has 528 * not been executed and needs retrying. 529 */ 530enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd) 531{ 532 struct scsi_device *sdev = scmd->device; 533 struct scsi_sense_hdr sshdr; 534 535 if (! scsi_command_normalize_sense(scmd, &sshdr)) 536 return FAILED; /* no valid sense data */ 537 538 scsi_report_sense(sdev, &sshdr); 539 540 if (scsi_sense_is_deferred(&sshdr)) 541 return NEEDS_RETRY; 542 543 if (sdev->handler && sdev->handler->check_sense) { 544 enum scsi_disposition rc; 545 546 rc = sdev->handler->check_sense(sdev, &sshdr); 547 if (rc != SCSI_RETURN_NOT_HANDLED) 548 return rc; 549 /* handler does not care. Drop down to default handling */ 550 } 551 552 if (scmd->cmnd[0] == TEST_UNIT_READY && 553 scmd->submitter != SUBMITTED_BY_SCSI_ERROR_HANDLER) 554 /* 555 * nasty: for mid-layer issued TURs, we need to return the 556 * actual sense data without any recovery attempt. For eh 557 * issued ones, we need to try to recover and interpret 558 */ 559 return SUCCESS; 560 561 /* 562 * Previous logic looked for FILEMARK, EOM or ILI which are 563 * mainly associated with tapes and returned SUCCESS. 564 */ 565 if (sshdr.response_code == 0x70) { 566 /* fixed format */ 567 if (scmd->sense_buffer[2] & 0xe0) 568 return SUCCESS; 569 } else { 570 /* 571 * descriptor format: look for "stream commands sense data 572 * descriptor" (see SSC-3). Assume single sense data 573 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. 574 */ 575 if ((sshdr.additional_length > 3) && 576 (scmd->sense_buffer[8] == 0x4) && 577 (scmd->sense_buffer[11] & 0xe0)) 578 return SUCCESS; 579 } 580 581 switch (sshdr.sense_key) { 582 case NO_SENSE: 583 return SUCCESS; 584 case RECOVERED_ERROR: 585 return /* soft_error */ SUCCESS; 586 587 case ABORTED_COMMAND: 588 if (sshdr.asc == 0x10) /* DIF */ 589 return SUCCESS; 590 591 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF) 592 return ADD_TO_MLQUEUE; 593 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 && 594 sdev->sdev_bflags & BLIST_RETRY_ASC_C1) 595 return ADD_TO_MLQUEUE; 596 597 return NEEDS_RETRY; 598 case NOT_READY: 599 case UNIT_ATTENTION: 600 /* 601 * if we are expecting a cc/ua because of a bus reset that we 602 * performed, treat this just as a retry. otherwise this is 603 * information that we should pass up to the upper-level driver 604 * so that we can deal with it there. 605 */ 606 if (scmd->device->expecting_cc_ua) { 607 /* 608 * Because some device does not queue unit 609 * attentions correctly, we carefully check 610 * additional sense code and qualifier so as 611 * not to squash media change unit attention. 612 */ 613 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { 614 scmd->device->expecting_cc_ua = 0; 615 return NEEDS_RETRY; 616 } 617 } 618 /* 619 * we might also expect a cc/ua if another LUN on the target 620 * reported a UA with an ASC/ASCQ of 3F 0E - 621 * REPORTED LUNS DATA HAS CHANGED. 622 */ 623 if (scmd->device->sdev_target->expecting_lun_change && 624 sshdr.asc == 0x3f && sshdr.ascq == 0x0e) 625 return NEEDS_RETRY; 626 /* 627 * if the device is in the process of becoming ready, we 628 * should retry. 629 */ 630 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) 631 return NEEDS_RETRY; 632 /* 633 * if the device is not started, we need to wake 634 * the error handler to start the motor 635 */ 636 if (scmd->device->allow_restart && 637 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) 638 return FAILED; 639 /* 640 * Pass the UA upwards for a determination in the completion 641 * functions. 642 */ 643 return SUCCESS; 644 645 /* these are not supported */ 646 case DATA_PROTECT: 647 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { 648 /* Thin provisioning hard threshold reached */ 649 set_host_byte(scmd, DID_ALLOC_FAILURE); 650 return SUCCESS; 651 } 652 fallthrough; 653 case COPY_ABORTED: 654 case VOLUME_OVERFLOW: 655 case MISCOMPARE: 656 case BLANK_CHECK: 657 set_host_byte(scmd, DID_TARGET_FAILURE); 658 return SUCCESS; 659 660 case MEDIUM_ERROR: 661 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ 662 sshdr.asc == 0x13 || /* AMNF DATA FIELD */ 663 sshdr.asc == 0x14) { /* RECORD NOT FOUND */ 664 set_host_byte(scmd, DID_MEDIUM_ERROR); 665 return SUCCESS; 666 } 667 return NEEDS_RETRY; 668 669 case HARDWARE_ERROR: 670 if (scmd->device->retry_hwerror) 671 return ADD_TO_MLQUEUE; 672 else 673 set_host_byte(scmd, DID_TARGET_FAILURE); 674 fallthrough; 675 676 case ILLEGAL_REQUEST: 677 if (sshdr.asc == 0x20 || /* Invalid command operation code */ 678 sshdr.asc == 0x21 || /* Logical block address out of range */ 679 sshdr.asc == 0x22 || /* Invalid function */ 680 sshdr.asc == 0x24 || /* Invalid field in cdb */ 681 sshdr.asc == 0x26 || /* Parameter value invalid */ 682 sshdr.asc == 0x27) { /* Write protected */ 683 set_host_byte(scmd, DID_TARGET_FAILURE); 684 } 685 return SUCCESS; 686 687 default: 688 return SUCCESS; 689 } 690} 691EXPORT_SYMBOL_GPL(scsi_check_sense); 692 693static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) 694{ 695 struct scsi_host_template *sht = sdev->host->hostt; 696 struct scsi_device *tmp_sdev; 697 698 if (!sht->track_queue_depth || 699 sdev->queue_depth >= sdev->max_queue_depth) 700 return; 701 702 if (time_before(jiffies, 703 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) 704 return; 705 706 if (time_before(jiffies, 707 sdev->last_queue_full_time + sdev->queue_ramp_up_period)) 708 return; 709 710 /* 711 * Walk all devices of a target and do 712 * ramp up on them. 713 */ 714 shost_for_each_device(tmp_sdev, sdev->host) { 715 if (tmp_sdev->channel != sdev->channel || 716 tmp_sdev->id != sdev->id || 717 tmp_sdev->queue_depth == sdev->max_queue_depth) 718 continue; 719 720 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); 721 sdev->last_queue_ramp_up = jiffies; 722 } 723} 724 725static void scsi_handle_queue_full(struct scsi_device *sdev) 726{ 727 struct scsi_host_template *sht = sdev->host->hostt; 728 struct scsi_device *tmp_sdev; 729 730 if (!sht->track_queue_depth) 731 return; 732 733 shost_for_each_device(tmp_sdev, sdev->host) { 734 if (tmp_sdev->channel != sdev->channel || 735 tmp_sdev->id != sdev->id) 736 continue; 737 /* 738 * We do not know the number of commands that were at 739 * the device when we got the queue full so we start 740 * from the highest possible value and work our way down. 741 */ 742 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); 743 } 744} 745 746/** 747 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. 748 * @scmd: SCSI cmd to examine. 749 * 750 * Notes: 751 * This is *only* called when we are examining the status of commands 752 * queued during error recovery. the main difference here is that we 753 * don't allow for the possibility of retries here, and we are a lot 754 * more restrictive about what we consider acceptable. 755 */ 756static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd) 757{ 758 /* 759 * first check the host byte, to see if there is anything in there 760 * that would indicate what we need to do. 761 */ 762 if (host_byte(scmd->result) == DID_RESET) { 763 /* 764 * rats. we are already in the error handler, so we now 765 * get to try and figure out what to do next. if the sense 766 * is valid, we have a pretty good idea of what to do. 767 * if not, we mark it as FAILED. 768 */ 769 return scsi_check_sense(scmd); 770 } 771 if (host_byte(scmd->result) != DID_OK) 772 return FAILED; 773 774 /* 775 * now, check the status byte to see if this indicates 776 * anything special. 777 */ 778 switch (get_status_byte(scmd)) { 779 case SAM_STAT_GOOD: 780 scsi_handle_queue_ramp_up(scmd->device); 781 fallthrough; 782 case SAM_STAT_COMMAND_TERMINATED: 783 return SUCCESS; 784 case SAM_STAT_CHECK_CONDITION: 785 return scsi_check_sense(scmd); 786 case SAM_STAT_CONDITION_MET: 787 case SAM_STAT_INTERMEDIATE: 788 case SAM_STAT_INTERMEDIATE_CONDITION_MET: 789 /* 790 * who knows? FIXME(eric) 791 */ 792 return SUCCESS; 793 case SAM_STAT_RESERVATION_CONFLICT: 794 if (scmd->cmnd[0] == TEST_UNIT_READY) 795 /* it is a success, we probed the device and 796 * found it */ 797 return SUCCESS; 798 /* otherwise, we failed to send the command */ 799 return FAILED; 800 case SAM_STAT_TASK_SET_FULL: 801 scsi_handle_queue_full(scmd->device); 802 fallthrough; 803 case SAM_STAT_BUSY: 804 return NEEDS_RETRY; 805 default: 806 return FAILED; 807 } 808 return FAILED; 809} 810 811/** 812 * scsi_eh_done - Completion function for error handling. 813 * @scmd: Cmd that is done. 814 */ 815void scsi_eh_done(struct scsi_cmnd *scmd) 816{ 817 struct completion *eh_action; 818 819 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 820 "%s result: %x\n", __func__, scmd->result)); 821 822 eh_action = scmd->device->host->eh_action; 823 if (eh_action) 824 complete(eh_action); 825} 826 827/** 828 * scsi_try_host_reset - ask host adapter to reset itself 829 * @scmd: SCSI cmd to send host reset. 830 */ 831static enum scsi_disposition scsi_try_host_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, 839 shost_printk(KERN_INFO, host, "Snd Host RST\n")); 840 841 if (!hostt->eh_host_reset_handler) 842 return FAILED; 843 844 rtn = hostt->eh_host_reset_handler(scmd); 845 846 if (rtn == SUCCESS) { 847 if (!hostt->skip_settle_delay) 848 ssleep(HOST_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/** 858 * scsi_try_bus_reset - ask host to perform a bus reset 859 * @scmd: SCSI cmd to send bus reset. 860 */ 861static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd) 862{ 863 unsigned long flags; 864 enum scsi_disposition rtn; 865 struct Scsi_Host *host = scmd->device->host; 866 struct scsi_host_template *hostt = host->hostt; 867 868 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 869 "%s: Snd Bus RST\n", __func__)); 870 871 if (!hostt->eh_bus_reset_handler) 872 return FAILED; 873 874 rtn = hostt->eh_bus_reset_handler(scmd); 875 876 if (rtn == SUCCESS) { 877 if (!hostt->skip_settle_delay) 878 ssleep(BUS_RESET_SETTLE_TIME); 879 spin_lock_irqsave(host->host_lock, flags); 880 scsi_report_bus_reset(host, scmd_channel(scmd)); 881 spin_unlock_irqrestore(host->host_lock, flags); 882 } 883 884 return rtn; 885} 886 887static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) 888{ 889 sdev->was_reset = 1; 890 sdev->expecting_cc_ua = 1; 891} 892 893/** 894 * scsi_try_target_reset - Ask host to perform a target reset 895 * @scmd: SCSI cmd used to send a target reset 896 * 897 * Notes: 898 * There is no timeout for this operation. if this operation is 899 * unreliable for a given host, then the host itself needs to put a 900 * timer on it, and set the host back to a consistent state prior to 901 * returning. 902 */ 903static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd) 904{ 905 unsigned long flags; 906 enum scsi_disposition rtn; 907 struct Scsi_Host *host = scmd->device->host; 908 struct scsi_host_template *hostt = host->hostt; 909 910 if (!hostt->eh_target_reset_handler) 911 return FAILED; 912 913 rtn = hostt->eh_target_reset_handler(scmd); 914 if (rtn == SUCCESS) { 915 spin_lock_irqsave(host->host_lock, flags); 916 __starget_for_each_device(scsi_target(scmd->device), NULL, 917 __scsi_report_device_reset); 918 spin_unlock_irqrestore(host->host_lock, flags); 919 } 920 921 return rtn; 922} 923 924/** 925 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev 926 * @scmd: SCSI cmd used to send BDR 927 * 928 * Notes: 929 * There is no timeout for this operation. if this operation is 930 * unreliable for a given host, then the host itself needs to put a 931 * timer on it, and set the host back to a consistent state prior to 932 * returning. 933 */ 934static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd) 935{ 936 enum scsi_disposition rtn; 937 struct scsi_host_template *hostt = scmd->device->host->hostt; 938 939 if (!hostt->eh_device_reset_handler) 940 return FAILED; 941 942 rtn = hostt->eh_device_reset_handler(scmd); 943 if (rtn == SUCCESS) 944 __scsi_report_device_reset(scmd->device, NULL); 945 return rtn; 946} 947 948/** 949 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command 950 * @hostt: SCSI driver host template 951 * @scmd: SCSI cmd used to send a target reset 952 * 953 * Return value: 954 * SUCCESS, FAILED, or FAST_IO_FAIL 955 * 956 * Notes: 957 * SUCCESS does not necessarily indicate that the command 958 * has been aborted; it only indicates that the LLDDs 959 * has cleared all references to that command. 960 * LLDDs should return FAILED only if an abort was required 961 * but could not be executed. LLDDs should return FAST_IO_FAIL 962 * if the device is temporarily unavailable (eg due to a 963 * link down on FibreChannel) 964 */ 965static enum scsi_disposition 966scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd) 967{ 968 if (!hostt->eh_abort_handler) 969 return FAILED; 970 971 return hostt->eh_abort_handler(scmd); 972} 973 974static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) 975{ 976 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) 977 if (scsi_try_bus_device_reset(scmd) != SUCCESS) 978 if (scsi_try_target_reset(scmd) != SUCCESS) 979 if (scsi_try_bus_reset(scmd) != SUCCESS) 980 scsi_try_host_reset(scmd); 981} 982 983/** 984 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery 985 * @scmd: SCSI command structure to hijack 986 * @ses: structure to save restore information 987 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed 988 * @cmnd_size: size in bytes of @cmnd (must be <= MAX_COMMAND_SIZE) 989 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) 990 * 991 * This function is used to save a scsi command information before re-execution 992 * as part of the error recovery process. If @sense_bytes is 0 the command 993 * sent must be one that does not transfer any data. If @sense_bytes != 0 994 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command 995 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. 996 */ 997void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, 998 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) 999{ 1000 struct scsi_device *sdev = scmd->device; 1001 1002 /* 1003 * We need saved copies of a number of fields - this is because 1004 * error handling may need to overwrite these with different values 1005 * to run different commands, and once error handling is complete, 1006 * we will need to restore these values prior to running the actual 1007 * command. 1008 */ 1009 ses->cmd_len = scmd->cmd_len; 1010 ses->data_direction = scmd->sc_data_direction; 1011 ses->sdb = scmd->sdb; 1012 ses->result = scmd->result; 1013 ses->resid_len = scmd->resid_len; 1014 ses->underflow = scmd->underflow; 1015 ses->prot_op = scmd->prot_op; 1016 ses->eh_eflags = scmd->eh_eflags; 1017 1018 scmd->prot_op = SCSI_PROT_NORMAL; 1019 scmd->eh_eflags = 0; 1020 memcpy(ses->cmnd, scmd->cmnd, sizeof(ses->cmnd)); 1021 memset(scmd->cmnd, 0, sizeof(scmd->cmnd)); 1022 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 1023 scmd->result = 0; 1024 scmd->resid_len = 0; 1025 1026 if (sense_bytes) { 1027 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, 1028 sense_bytes); 1029 sg_init_one(&ses->sense_sgl, scmd->sense_buffer, 1030 scmd->sdb.length); 1031 scmd->sdb.table.sgl = &ses->sense_sgl; 1032 scmd->sc_data_direction = DMA_FROM_DEVICE; 1033 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; 1034 scmd->cmnd[0] = REQUEST_SENSE; 1035 scmd->cmnd[4] = scmd->sdb.length; 1036 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1037 } else { 1038 scmd->sc_data_direction = DMA_NONE; 1039 if (cmnd) { 1040 BUG_ON(cmnd_size > sizeof(scmd->cmnd)); 1041 memcpy(scmd->cmnd, cmnd, cmnd_size); 1042 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 1043 } 1044 } 1045 1046 scmd->underflow = 0; 1047 1048 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) 1049 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | 1050 (sdev->lun << 5 & 0xe0); 1051 1052 /* 1053 * Zero the sense buffer. The scsi spec mandates that any 1054 * untransferred sense data should be interpreted as being zero. 1055 */ 1056 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 1057} 1058EXPORT_SYMBOL(scsi_eh_prep_cmnd); 1059 1060/** 1061 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery 1062 * @scmd: SCSI command structure to restore 1063 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd 1064 * 1065 * Undo any damage done by above scsi_eh_prep_cmnd(). 1066 */ 1067void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) 1068{ 1069 /* 1070 * Restore original data 1071 */ 1072 scmd->cmd_len = ses->cmd_len; 1073 memcpy(scmd->cmnd, ses->cmnd, sizeof(ses->cmnd)); 1074 scmd->sc_data_direction = ses->data_direction; 1075 scmd->sdb = ses->sdb; 1076 scmd->result = ses->result; 1077 scmd->resid_len = ses->resid_len; 1078 scmd->underflow = ses->underflow; 1079 scmd->prot_op = ses->prot_op; 1080 scmd->eh_eflags = ses->eh_eflags; 1081} 1082EXPORT_SYMBOL(scsi_eh_restore_cmnd); 1083 1084/** 1085 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery 1086 * @scmd: SCSI command structure to hijack 1087 * @cmnd: CDB to send 1088 * @cmnd_size: size in bytes of @cmnd 1089 * @timeout: timeout for this request 1090 * @sense_bytes: size of sense data to copy or 0 1091 * 1092 * This function is used to send a scsi command down to a target device 1093 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. 1094 * 1095 * Return value: 1096 * SUCCESS or FAILED or NEEDS_RETRY 1097 */ 1098static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd, 1099 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes) 1100{ 1101 struct scsi_device *sdev = scmd->device; 1102 struct Scsi_Host *shost = sdev->host; 1103 DECLARE_COMPLETION_ONSTACK(done); 1104 unsigned long timeleft = timeout, delay; 1105 struct scsi_eh_save ses; 1106 const unsigned long stall_for = msecs_to_jiffies(100); 1107 int rtn; 1108 1109retry: 1110 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); 1111 shost->eh_action = &done; 1112 1113 scsi_log_send(scmd); 1114 scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER; 1115 1116 /* 1117 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can 1118 * change the SCSI device state after we have examined it and before 1119 * .queuecommand() is called. 1120 */ 1121 mutex_lock(&sdev->state_mutex); 1122 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) { 1123 mutex_unlock(&sdev->state_mutex); 1124 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev, 1125 "%s: state %d <> %d\n", __func__, sdev->sdev_state, 1126 SDEV_BLOCK)); 1127 delay = min(timeleft, stall_for); 1128 timeleft -= delay; 1129 msleep(jiffies_to_msecs(delay)); 1130 mutex_lock(&sdev->state_mutex); 1131 } 1132 if (sdev->sdev_state != SDEV_BLOCK) 1133 rtn = shost->hostt->queuecommand(shost, scmd); 1134 else 1135 rtn = FAILED; 1136 mutex_unlock(&sdev->state_mutex); 1137 1138 if (rtn) { 1139 if (timeleft > stall_for) { 1140 scsi_eh_restore_cmnd(scmd, &ses); 1141 1142 timeleft -= stall_for; 1143 msleep(jiffies_to_msecs(stall_for)); 1144 goto retry; 1145 } 1146 /* signal not to enter either branch of the if () below */ 1147 timeleft = 0; 1148 rtn = FAILED; 1149 } else { 1150 timeleft = wait_for_completion_timeout(&done, timeout); 1151 rtn = SUCCESS; 1152 } 1153 1154 shost->eh_action = NULL; 1155 1156 scsi_log_completion(scmd, rtn); 1157 1158 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1159 "%s timeleft: %ld\n", 1160 __func__, timeleft)); 1161 1162 /* 1163 * If there is time left scsi_eh_done got called, and we will examine 1164 * the actual status codes to see whether the command actually did 1165 * complete normally, else if we have a zero return and no time left, 1166 * the command must still be pending, so abort it and return FAILED. 1167 * If we never actually managed to issue the command, because 1168 * ->queuecommand() kept returning non zero, use the rtn = FAILED 1169 * value above (so don't execute either branch of the if) 1170 */ 1171 if (timeleft) { 1172 rtn = scsi_eh_completed_normally(scmd); 1173 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1174 "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); 1175 1176 switch (rtn) { 1177 case SUCCESS: 1178 case NEEDS_RETRY: 1179 case FAILED: 1180 break; 1181 case ADD_TO_MLQUEUE: 1182 rtn = NEEDS_RETRY; 1183 break; 1184 default: 1185 rtn = FAILED; 1186 break; 1187 } 1188 } else if (rtn != FAILED) { 1189 scsi_abort_eh_cmnd(scmd); 1190 rtn = FAILED; 1191 } 1192 1193 scsi_eh_restore_cmnd(scmd, &ses); 1194 1195 return rtn; 1196} 1197 1198/** 1199 * scsi_request_sense - Request sense data from a particular target. 1200 * @scmd: SCSI cmd for request sense. 1201 * 1202 * Notes: 1203 * Some hosts automatically obtain this information, others require 1204 * that we obtain it on our own. This function will *not* return until 1205 * the command either times out, or it completes. 1206 */ 1207static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd) 1208{ 1209 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); 1210} 1211 1212static enum scsi_disposition 1213scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn) 1214{ 1215 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) { 1216 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); 1217 if (sdrv->eh_action) 1218 rtn = sdrv->eh_action(scmd, rtn); 1219 } 1220 return rtn; 1221} 1222 1223/** 1224 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. 1225 * @scmd: Original SCSI cmd that eh has finished. 1226 * @done_q: Queue for processed commands. 1227 * 1228 * Notes: 1229 * We don't want to use the normal command completion while we are are 1230 * still handling errors - it may cause other commands to be queued, 1231 * and that would disturb what we are doing. Thus we really want to 1232 * keep a list of pending commands for final completion, and once we 1233 * are ready to leave error handling we handle completion for real. 1234 */ 1235void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) 1236{ 1237 list_move_tail(&scmd->eh_entry, done_q); 1238} 1239EXPORT_SYMBOL(scsi_eh_finish_cmd); 1240 1241/** 1242 * scsi_eh_get_sense - Get device sense data. 1243 * @work_q: Queue of commands to process. 1244 * @done_q: Queue of processed commands. 1245 * 1246 * Description: 1247 * See if we need to request sense information. if so, then get it 1248 * now, so we have a better idea of what to do. 1249 * 1250 * Notes: 1251 * This has the unfortunate side effect that if a shost adapter does 1252 * not automatically request sense information, we end up shutting 1253 * it down before we request it. 1254 * 1255 * All drivers should request sense information internally these days, 1256 * so for now all I have to say is tough noogies if you end up in here. 1257 * 1258 * XXX: Long term this code should go away, but that needs an audit of 1259 * all LLDDs first. 1260 */ 1261int scsi_eh_get_sense(struct list_head *work_q, 1262 struct list_head *done_q) 1263{ 1264 struct scsi_cmnd *scmd, *next; 1265 struct Scsi_Host *shost; 1266 enum scsi_disposition rtn; 1267 1268 /* 1269 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, 1270 * should not get sense. 1271 */ 1272 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1273 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || 1274 SCSI_SENSE_VALID(scmd)) 1275 continue; 1276 1277 shost = scmd->device->host; 1278 if (scsi_host_eh_past_deadline(shost)) { 1279 SCSI_LOG_ERROR_RECOVERY(3, 1280 scmd_printk(KERN_INFO, scmd, 1281 "%s: skip request sense, past eh deadline\n", 1282 current->comm)); 1283 break; 1284 } 1285 if (!scsi_status_is_check_condition(scmd->result)) 1286 /* 1287 * don't request sense if there's no check condition 1288 * status because the error we're processing isn't one 1289 * that has a sense code (and some devices get 1290 * confused by sense requests out of the blue) 1291 */ 1292 continue; 1293 1294 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, 1295 "%s: requesting sense\n", 1296 current->comm)); 1297 rtn = scsi_request_sense(scmd); 1298 if (rtn != SUCCESS) 1299 continue; 1300 1301 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1302 "sense requested, result %x\n", scmd->result)); 1303 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); 1304 1305 rtn = scsi_decide_disposition(scmd); 1306 1307 /* 1308 * if the result was normal, then just pass it along to the 1309 * upper level. 1310 */ 1311 if (rtn == SUCCESS) 1312 /* 1313 * We don't want this command reissued, just finished 1314 * with the sense data, so set retries to the max 1315 * allowed to ensure it won't get reissued. If the user 1316 * has requested infinite retries, we also want to 1317 * finish this command, so force completion by setting 1318 * retries and allowed to the same value. 1319 */ 1320 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) 1321 scmd->retries = scmd->allowed = 1; 1322 else 1323 scmd->retries = scmd->allowed; 1324 else if (rtn != NEEDS_RETRY) 1325 continue; 1326 1327 scsi_eh_finish_cmd(scmd, done_q); 1328 } 1329 1330 return list_empty(work_q); 1331} 1332EXPORT_SYMBOL_GPL(scsi_eh_get_sense); 1333 1334/** 1335 * scsi_eh_tur - Send TUR to device. 1336 * @scmd: &scsi_cmnd to send TUR 1337 * 1338 * Return value: 1339 * 0 - Device is ready. 1 - Device NOT ready. 1340 */ 1341static int scsi_eh_tur(struct scsi_cmnd *scmd) 1342{ 1343 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; 1344 int retry_cnt = 1; 1345 enum scsi_disposition rtn; 1346 1347retry_tur: 1348 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, 1349 scmd->device->eh_timeout, 0); 1350 1351 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, 1352 "%s return: %x\n", __func__, rtn)); 1353 1354 switch (rtn) { 1355 case NEEDS_RETRY: 1356 if (retry_cnt--) 1357 goto retry_tur; 1358 fallthrough; 1359 case SUCCESS: 1360 return 0; 1361 default: 1362 return 1; 1363 } 1364} 1365 1366/** 1367 * scsi_eh_test_devices - check if devices are responding from error recovery. 1368 * @cmd_list: scsi commands in error recovery. 1369 * @work_q: queue for commands which still need more error recovery 1370 * @done_q: queue for commands which are finished 1371 * @try_stu: boolean on if a STU command should be tried in addition to TUR. 1372 * 1373 * Decription: 1374 * Tests if devices are in a working state. Commands to devices now in 1375 * a working state are sent to the done_q while commands to devices which 1376 * are still failing to respond are returned to the work_q for more 1377 * processing. 1378 **/ 1379static int scsi_eh_test_devices(struct list_head *cmd_list, 1380 struct list_head *work_q, 1381 struct list_head *done_q, int try_stu) 1382{ 1383 struct scsi_cmnd *scmd, *next; 1384 struct scsi_device *sdev; 1385 int finish_cmds; 1386 1387 while (!list_empty(cmd_list)) { 1388 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); 1389 sdev = scmd->device; 1390 1391 if (!try_stu) { 1392 if (scsi_host_eh_past_deadline(sdev->host)) { 1393 /* Push items back onto work_q */ 1394 list_splice_init(cmd_list, work_q); 1395 SCSI_LOG_ERROR_RECOVERY(3, 1396 sdev_printk(KERN_INFO, sdev, 1397 "%s: skip test device, past eh deadline", 1398 current->comm)); 1399 break; 1400 } 1401 } 1402 1403 finish_cmds = !scsi_device_online(scmd->device) || 1404 (try_stu && !scsi_eh_try_stu(scmd) && 1405 !scsi_eh_tur(scmd)) || 1406 !scsi_eh_tur(scmd); 1407 1408 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) 1409 if (scmd->device == sdev) { 1410 if (finish_cmds && 1411 (try_stu || 1412 scsi_eh_action(scmd, SUCCESS) == SUCCESS)) 1413 scsi_eh_finish_cmd(scmd, done_q); 1414 else 1415 list_move_tail(&scmd->eh_entry, work_q); 1416 } 1417 } 1418 return list_empty(work_q); 1419} 1420 1421/** 1422 * scsi_eh_try_stu - Send START_UNIT to device. 1423 * @scmd: &scsi_cmnd to send START_UNIT 1424 * 1425 * Return value: 1426 * 0 - Device is ready. 1 - Device NOT ready. 1427 */ 1428static int scsi_eh_try_stu(struct scsi_cmnd *scmd) 1429{ 1430 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; 1431 1432 if (scmd->device->allow_restart) { 1433 int i; 1434 enum scsi_disposition rtn = NEEDS_RETRY; 1435 1436 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) 1437 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, 1438 scmd->device->eh_timeout, 0); 1439 1440 if (rtn == SUCCESS) 1441 return 0; 1442 } 1443 1444 return 1; 1445} 1446 1447 /** 1448 * scsi_eh_stu - send START_UNIT if needed 1449 * @shost: &scsi host being recovered. 1450 * @work_q: &list_head for pending commands. 1451 * @done_q: &list_head for processed commands. 1452 * 1453 * Notes: 1454 * If commands are failing due to not ready, initializing command required, 1455 * try revalidating the device, which will end up sending a start unit. 1456 */ 1457static int scsi_eh_stu(struct Scsi_Host *shost, 1458 struct list_head *work_q, 1459 struct list_head *done_q) 1460{ 1461 struct scsi_cmnd *scmd, *stu_scmd, *next; 1462 struct scsi_device *sdev; 1463 1464 shost_for_each_device(sdev, shost) { 1465 if (scsi_host_eh_past_deadline(shost)) { 1466 SCSI_LOG_ERROR_RECOVERY(3, 1467 sdev_printk(KERN_INFO, sdev, 1468 "%s: skip START_UNIT, past eh deadline\n", 1469 current->comm)); 1470 scsi_device_put(sdev); 1471 break; 1472 } 1473 stu_scmd = NULL; 1474 list_for_each_entry(scmd, work_q, eh_entry) 1475 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && 1476 scsi_check_sense(scmd) == FAILED ) { 1477 stu_scmd = scmd; 1478 break; 1479 } 1480 1481 if (!stu_scmd) 1482 continue; 1483 1484 SCSI_LOG_ERROR_RECOVERY(3, 1485 sdev_printk(KERN_INFO, sdev, 1486 "%s: Sending START_UNIT\n", 1487 current->comm)); 1488 1489 if (!scsi_eh_try_stu(stu_scmd)) { 1490 if (!scsi_device_online(sdev) || 1491 !scsi_eh_tur(stu_scmd)) { 1492 list_for_each_entry_safe(scmd, next, 1493 work_q, eh_entry) { 1494 if (scmd->device == sdev && 1495 scsi_eh_action(scmd, SUCCESS) == SUCCESS) 1496 scsi_eh_finish_cmd(scmd, done_q); 1497 } 1498 } 1499 } else { 1500 SCSI_LOG_ERROR_RECOVERY(3, 1501 sdev_printk(KERN_INFO, sdev, 1502 "%s: START_UNIT failed\n", 1503 current->comm)); 1504 } 1505 } 1506 1507 return list_empty(work_q); 1508} 1509 1510 1511/** 1512 * scsi_eh_bus_device_reset - send bdr if needed 1513 * @shost: scsi host being recovered. 1514 * @work_q: &list_head for pending commands. 1515 * @done_q: &list_head for processed commands. 1516 * 1517 * Notes: 1518 * Try a bus device reset. Still, look to see whether we have multiple 1519 * devices that are jammed or not - if we have multiple devices, it 1520 * makes no sense to try bus_device_reset - we really would need to try 1521 * a bus_reset instead. 1522 */ 1523static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, 1524 struct list_head *work_q, 1525 struct list_head *done_q) 1526{ 1527 struct scsi_cmnd *scmd, *bdr_scmd, *next; 1528 struct scsi_device *sdev; 1529 enum scsi_disposition rtn; 1530 1531 shost_for_each_device(sdev, shost) { 1532 if (scsi_host_eh_past_deadline(shost)) { 1533 SCSI_LOG_ERROR_RECOVERY(3, 1534 sdev_printk(KERN_INFO, sdev, 1535 "%s: skip BDR, past eh deadline\n", 1536 current->comm)); 1537 scsi_device_put(sdev); 1538 break; 1539 } 1540 bdr_scmd = NULL; 1541 list_for_each_entry(scmd, work_q, eh_entry) 1542 if (scmd->device == sdev) { 1543 bdr_scmd = scmd; 1544 break; 1545 } 1546 1547 if (!bdr_scmd) 1548 continue; 1549 1550 SCSI_LOG_ERROR_RECOVERY(3, 1551 sdev_printk(KERN_INFO, sdev, 1552 "%s: Sending BDR\n", current->comm)); 1553 rtn = scsi_try_bus_device_reset(bdr_scmd); 1554 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1555 if (!scsi_device_online(sdev) || 1556 rtn == FAST_IO_FAIL || 1557 !scsi_eh_tur(bdr_scmd)) { 1558 list_for_each_entry_safe(scmd, next, 1559 work_q, eh_entry) { 1560 if (scmd->device == sdev && 1561 scsi_eh_action(scmd, rtn) != FAILED) 1562 scsi_eh_finish_cmd(scmd, 1563 done_q); 1564 } 1565 } 1566 } else { 1567 SCSI_LOG_ERROR_RECOVERY(3, 1568 sdev_printk(KERN_INFO, sdev, 1569 "%s: BDR failed\n", current->comm)); 1570 } 1571 } 1572 1573 return list_empty(work_q); 1574} 1575 1576/** 1577 * scsi_eh_target_reset - send target reset if needed 1578 * @shost: scsi host being recovered. 1579 * @work_q: &list_head for pending commands. 1580 * @done_q: &list_head for processed commands. 1581 * 1582 * Notes: 1583 * Try a target reset. 1584 */ 1585static int scsi_eh_target_reset(struct Scsi_Host *shost, 1586 struct list_head *work_q, 1587 struct list_head *done_q) 1588{ 1589 LIST_HEAD(tmp_list); 1590 LIST_HEAD(check_list); 1591 1592 list_splice_init(work_q, &tmp_list); 1593 1594 while (!list_empty(&tmp_list)) { 1595 struct scsi_cmnd *next, *scmd; 1596 enum scsi_disposition rtn; 1597 unsigned int id; 1598 1599 if (scsi_host_eh_past_deadline(shost)) { 1600 /* push back on work queue for further processing */ 1601 list_splice_init(&check_list, work_q); 1602 list_splice_init(&tmp_list, work_q); 1603 SCSI_LOG_ERROR_RECOVERY(3, 1604 shost_printk(KERN_INFO, shost, 1605 "%s: Skip target reset, past eh deadline\n", 1606 current->comm)); 1607 return list_empty(work_q); 1608 } 1609 1610 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); 1611 id = scmd_id(scmd); 1612 1613 SCSI_LOG_ERROR_RECOVERY(3, 1614 shost_printk(KERN_INFO, shost, 1615 "%s: Sending target reset to target %d\n", 1616 current->comm, id)); 1617 rtn = scsi_try_target_reset(scmd); 1618 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) 1619 SCSI_LOG_ERROR_RECOVERY(3, 1620 shost_printk(KERN_INFO, shost, 1621 "%s: Target reset failed" 1622 " target: %d\n", 1623 current->comm, id)); 1624 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { 1625 if (scmd_id(scmd) != id) 1626 continue; 1627 1628 if (rtn == SUCCESS) 1629 list_move_tail(&scmd->eh_entry, &check_list); 1630 else if (rtn == FAST_IO_FAIL) 1631 scsi_eh_finish_cmd(scmd, done_q); 1632 else 1633 /* push back on work queue for further processing */ 1634 list_move(&scmd->eh_entry, work_q); 1635 } 1636 } 1637 1638 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1639} 1640 1641/** 1642 * scsi_eh_bus_reset - send a bus reset 1643 * @shost: &scsi host being recovered. 1644 * @work_q: &list_head for pending commands. 1645 * @done_q: &list_head for processed commands. 1646 */ 1647static int scsi_eh_bus_reset(struct Scsi_Host *shost, 1648 struct list_head *work_q, 1649 struct list_head *done_q) 1650{ 1651 struct scsi_cmnd *scmd, *chan_scmd, *next; 1652 LIST_HEAD(check_list); 1653 unsigned int channel; 1654 enum scsi_disposition rtn; 1655 1656 /* 1657 * we really want to loop over the various channels, and do this on 1658 * a channel by channel basis. we should also check to see if any 1659 * of the failed commands are on soft_reset devices, and if so, skip 1660 * the reset. 1661 */ 1662 1663 for (channel = 0; channel <= shost->max_channel; channel++) { 1664 if (scsi_host_eh_past_deadline(shost)) { 1665 list_splice_init(&check_list, work_q); 1666 SCSI_LOG_ERROR_RECOVERY(3, 1667 shost_printk(KERN_INFO, shost, 1668 "%s: skip BRST, past eh deadline\n", 1669 current->comm)); 1670 return list_empty(work_q); 1671 } 1672 1673 chan_scmd = NULL; 1674 list_for_each_entry(scmd, work_q, eh_entry) { 1675 if (channel == scmd_channel(scmd)) { 1676 chan_scmd = scmd; 1677 break; 1678 /* 1679 * FIXME add back in some support for 1680 * soft_reset devices. 1681 */ 1682 } 1683 } 1684 1685 if (!chan_scmd) 1686 continue; 1687 SCSI_LOG_ERROR_RECOVERY(3, 1688 shost_printk(KERN_INFO, shost, 1689 "%s: Sending BRST chan: %d\n", 1690 current->comm, channel)); 1691 rtn = scsi_try_bus_reset(chan_scmd); 1692 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { 1693 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1694 if (channel == scmd_channel(scmd)) { 1695 if (rtn == FAST_IO_FAIL) 1696 scsi_eh_finish_cmd(scmd, 1697 done_q); 1698 else 1699 list_move_tail(&scmd->eh_entry, 1700 &check_list); 1701 } 1702 } 1703 } else { 1704 SCSI_LOG_ERROR_RECOVERY(3, 1705 shost_printk(KERN_INFO, shost, 1706 "%s: BRST failed chan: %d\n", 1707 current->comm, channel)); 1708 } 1709 } 1710 return scsi_eh_test_devices(&check_list, work_q, done_q, 0); 1711} 1712 1713/** 1714 * scsi_eh_host_reset - send a host reset 1715 * @shost: host to be reset. 1716 * @work_q: &list_head for pending commands. 1717 * @done_q: &list_head for processed commands. 1718 */ 1719static int scsi_eh_host_reset(struct Scsi_Host *shost, 1720 struct list_head *work_q, 1721 struct list_head *done_q) 1722{ 1723 struct scsi_cmnd *scmd, *next; 1724 LIST_HEAD(check_list); 1725 enum scsi_disposition rtn; 1726 1727 if (!list_empty(work_q)) { 1728 scmd = list_entry(work_q->next, 1729 struct scsi_cmnd, eh_entry); 1730 1731 SCSI_LOG_ERROR_RECOVERY(3, 1732 shost_printk(KERN_INFO, shost, 1733 "%s: Sending HRST\n", 1734 current->comm)); 1735 1736 rtn = scsi_try_host_reset(scmd); 1737 if (rtn == SUCCESS) { 1738 list_splice_init(work_q, &check_list); 1739 } else if (rtn == FAST_IO_FAIL) { 1740 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1741 scsi_eh_finish_cmd(scmd, done_q); 1742 } 1743 } else { 1744 SCSI_LOG_ERROR_RECOVERY(3, 1745 shost_printk(KERN_INFO, shost, 1746 "%s: HRST failed\n", 1747 current->comm)); 1748 } 1749 } 1750 return scsi_eh_test_devices(&check_list, work_q, done_q, 1); 1751} 1752 1753/** 1754 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover 1755 * @work_q: &list_head for pending commands. 1756 * @done_q: &list_head for processed commands. 1757 */ 1758static void scsi_eh_offline_sdevs(struct list_head *work_q, 1759 struct list_head *done_q) 1760{ 1761 struct scsi_cmnd *scmd, *next; 1762 struct scsi_device *sdev; 1763 1764 list_for_each_entry_safe(scmd, next, work_q, eh_entry) { 1765 sdev_printk(KERN_INFO, scmd->device, "Device offlined - " 1766 "not ready after error recovery\n"); 1767 sdev = scmd->device; 1768 1769 mutex_lock(&sdev->state_mutex); 1770 scsi_device_set_state(sdev, SDEV_OFFLINE); 1771 mutex_unlock(&sdev->state_mutex); 1772 1773 scsi_eh_finish_cmd(scmd, done_q); 1774 } 1775 return; 1776} 1777 1778/** 1779 * scsi_noretry_cmd - determine if command should be failed fast 1780 * @scmd: SCSI cmd to examine. 1781 */ 1782int scsi_noretry_cmd(struct scsi_cmnd *scmd) 1783{ 1784 struct request *req = scsi_cmd_to_rq(scmd); 1785 1786 switch (host_byte(scmd->result)) { 1787 case DID_OK: 1788 break; 1789 case DID_TIME_OUT: 1790 goto check_type; 1791 case DID_BUS_BUSY: 1792 return req->cmd_flags & REQ_FAILFAST_TRANSPORT; 1793 case DID_PARITY: 1794 return req->cmd_flags & REQ_FAILFAST_DEV; 1795 case DID_ERROR: 1796 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT) 1797 return 0; 1798 fallthrough; 1799 case DID_SOFT_ERROR: 1800 return req->cmd_flags & REQ_FAILFAST_DRIVER; 1801 } 1802 1803 if (!scsi_status_is_check_condition(scmd->result)) 1804 return 0; 1805 1806check_type: 1807 /* 1808 * assume caller has checked sense and determined 1809 * the check condition was retryable. 1810 */ 1811 if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(req)) 1812 return 1; 1813 1814 return 0; 1815} 1816 1817/** 1818 * scsi_decide_disposition - Disposition a cmd on return from LLD. 1819 * @scmd: SCSI cmd to examine. 1820 * 1821 * Notes: 1822 * This is *only* called when we are examining the status after sending 1823 * out the actual data command. any commands that are queued for error 1824 * recovery (e.g. test_unit_ready) do *not* come through here. 1825 * 1826 * When this routine returns failed, it means the error handler thread 1827 * is woken. In cases where the error code indicates an error that 1828 * doesn't require the error handler read (i.e. we don't need to 1829 * abort/reset), this function should return SUCCESS. 1830 */ 1831enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd) 1832{ 1833 enum scsi_disposition rtn; 1834 1835 /* 1836 * if the device is offline, then we clearly just pass the result back 1837 * up to the top level. 1838 */ 1839 if (!scsi_device_online(scmd->device)) { 1840 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, 1841 "%s: device offline - report as SUCCESS\n", __func__)); 1842 return SUCCESS; 1843 } 1844 1845 /* 1846 * first check the host byte, to see if there is anything in there 1847 * that would indicate what we need to do. 1848 */ 1849 switch (host_byte(scmd->result)) { 1850 case DID_PASSTHROUGH: 1851 /* 1852 * no matter what, pass this through to the upper layer. 1853 * nuke this special code so that it looks like we are saying 1854 * did_ok. 1855 */ 1856 scmd->result &= 0xff00ffff; 1857 return SUCCESS; 1858 case DID_OK: 1859 /* 1860 * looks good. drop through, and check the next byte. 1861 */ 1862 break; 1863 case DID_ABORT: 1864 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { 1865 set_host_byte(scmd, DID_TIME_OUT); 1866 return SUCCESS; 1867 } 1868 fallthrough; 1869 case DID_NO_CONNECT: 1870 case DID_BAD_TARGET: 1871 /* 1872 * note - this means that we just report the status back 1873 * to the top level driver, not that we actually think 1874 * that it indicates SUCCESS. 1875 */ 1876 return SUCCESS; 1877 case DID_SOFT_ERROR: 1878 /* 1879 * when the low level driver returns did_soft_error, 1880 * it is responsible for keeping an internal retry counter 1881 * in order to avoid endless loops (db) 1882 */ 1883 goto maybe_retry; 1884 case DID_IMM_RETRY: 1885 return NEEDS_RETRY; 1886 1887 case DID_REQUEUE: 1888 return ADD_TO_MLQUEUE; 1889 case DID_TRANSPORT_DISRUPTED: 1890 /* 1891 * LLD/transport was disrupted during processing of the IO. 1892 * The transport class is now blocked/blocking, 1893 * and the transport will decide what to do with the IO 1894 * based on its timers and recovery capablilities if 1895 * there are enough retries. 1896 */ 1897 goto maybe_retry; 1898 case DID_TRANSPORT_FAILFAST: 1899 /* 1900 * The transport decided to failfast the IO (most likely 1901 * the fast io fail tmo fired), so send IO directly upwards. 1902 */ 1903 return SUCCESS; 1904 case DID_TRANSPORT_MARGINAL: 1905 /* 1906 * caller has decided not to do retries on 1907 * abort success, so send IO directly upwards 1908 */ 1909 return SUCCESS; 1910 case DID_ERROR: 1911 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT) 1912 /* 1913 * execute reservation conflict processing code 1914 * lower down 1915 */ 1916 break; 1917 fallthrough; 1918 case DID_BUS_BUSY: 1919 case DID_PARITY: 1920 goto maybe_retry; 1921 case DID_TIME_OUT: 1922 /* 1923 * when we scan the bus, we get timeout messages for 1924 * these commands if there is no device available. 1925 * other hosts report did_no_connect for the same thing. 1926 */ 1927 if ((scmd->cmnd[0] == TEST_UNIT_READY || 1928 scmd->cmnd[0] == INQUIRY)) { 1929 return SUCCESS; 1930 } else { 1931 return FAILED; 1932 } 1933 case DID_RESET: 1934 return SUCCESS; 1935 default: 1936 return FAILED; 1937 } 1938 1939 /* 1940 * check the status byte to see if this indicates anything special. 1941 */ 1942 switch (get_status_byte(scmd)) { 1943 case SAM_STAT_TASK_SET_FULL: 1944 scsi_handle_queue_full(scmd->device); 1945 /* 1946 * the case of trying to send too many commands to a 1947 * tagged queueing device. 1948 */ 1949 fallthrough; 1950 case SAM_STAT_BUSY: 1951 /* 1952 * device can't talk to us at the moment. Should only 1953 * occur (SAM-3) when the task queue is empty, so will cause 1954 * the empty queue handling to trigger a stall in the 1955 * device. 1956 */ 1957 return ADD_TO_MLQUEUE; 1958 case SAM_STAT_GOOD: 1959 if (scmd->cmnd[0] == REPORT_LUNS) 1960 scmd->device->sdev_target->expecting_lun_change = 0; 1961 scsi_handle_queue_ramp_up(scmd->device); 1962 fallthrough; 1963 case SAM_STAT_COMMAND_TERMINATED: 1964 return SUCCESS; 1965 case SAM_STAT_TASK_ABORTED: 1966 goto maybe_retry; 1967 case SAM_STAT_CHECK_CONDITION: 1968 rtn = scsi_check_sense(scmd); 1969 if (rtn == NEEDS_RETRY) 1970 goto maybe_retry; 1971 /* if rtn == FAILED, we have no sense information; 1972 * returning FAILED will wake the error handler thread 1973 * to collect the sense and redo the decide 1974 * disposition */ 1975 return rtn; 1976 case SAM_STAT_CONDITION_MET: 1977 case SAM_STAT_INTERMEDIATE: 1978 case SAM_STAT_INTERMEDIATE_CONDITION_MET: 1979 case SAM_STAT_ACA_ACTIVE: 1980 /* 1981 * who knows? FIXME(eric) 1982 */ 1983 return SUCCESS; 1984 1985 case SAM_STAT_RESERVATION_CONFLICT: 1986 sdev_printk(KERN_INFO, scmd->device, 1987 "reservation conflict\n"); 1988 set_host_byte(scmd, DID_NEXUS_FAILURE); 1989 return SUCCESS; /* causes immediate i/o error */ 1990 } 1991 return FAILED; 1992 1993maybe_retry: 1994 1995 /* we requeue for retry because the error was retryable, and 1996 * the request was not marked fast fail. Note that above, 1997 * even if the request is marked fast fail, we still requeue 1998 * for queue congestion conditions (QUEUE_FULL or BUSY) */ 1999 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) { 2000 return NEEDS_RETRY; 2001 } else { 2002 /* 2003 * no more retries - report this one back to upper level. 2004 */ 2005 return SUCCESS; 2006 } 2007} 2008 2009static void eh_lock_door_done(struct request *req, blk_status_t status) 2010{ 2011 blk_mq_free_request(req); 2012} 2013 2014/** 2015 * scsi_eh_lock_door - Prevent medium removal for the specified device 2016 * @sdev: SCSI device to prevent medium removal 2017 * 2018 * Locking: 2019 * We must be called from process context. 2020 * 2021 * Notes: 2022 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the 2023 * head of the devices request queue, and continue. 2024 */ 2025static void scsi_eh_lock_door(struct scsi_device *sdev) 2026{ 2027 struct scsi_cmnd *scmd; 2028 struct request *req; 2029 2030 req = scsi_alloc_request(sdev->request_queue, REQ_OP_DRV_IN, 0); 2031 if (IS_ERR(req)) 2032 return; 2033 scmd = blk_mq_rq_to_pdu(req); 2034 2035 scmd->cmnd[0] = ALLOW_MEDIUM_REMOVAL; 2036 scmd->cmnd[1] = 0; 2037 scmd->cmnd[2] = 0; 2038 scmd->cmnd[3] = 0; 2039 scmd->cmnd[4] = SCSI_REMOVAL_PREVENT; 2040 scmd->cmnd[5] = 0; 2041 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); 2042 scmd->allowed = 5; 2043 2044 req->rq_flags |= RQF_QUIET; 2045 req->timeout = 10 * HZ; 2046 req->end_io = eh_lock_door_done; 2047 2048 blk_execute_rq_nowait(req, true); 2049} 2050 2051/** 2052 * scsi_restart_operations - restart io operations to the specified host. 2053 * @shost: Host we are restarting. 2054 * 2055 * Notes: 2056 * When we entered the error handler, we blocked all further i/o to 2057 * this device. we need to 'reverse' this process. 2058 */ 2059static void scsi_restart_operations(struct Scsi_Host *shost) 2060{ 2061 struct scsi_device *sdev; 2062 unsigned long flags; 2063 2064 /* 2065 * If the door was locked, we need to insert a door lock request 2066 * onto the head of the SCSI request queue for the device. There 2067 * is no point trying to lock the door of an off-line device. 2068 */ 2069 shost_for_each_device(sdev, shost) { 2070 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { 2071 scsi_eh_lock_door(sdev); 2072 sdev->was_reset = 0; 2073 } 2074 } 2075 2076 /* 2077 * next free up anything directly waiting upon the host. this 2078 * will be requests for character device operations, and also for 2079 * ioctls to queued block devices. 2080 */ 2081 SCSI_LOG_ERROR_RECOVERY(3, 2082 shost_printk(KERN_INFO, shost, "waking up host to restart\n")); 2083 2084 spin_lock_irqsave(shost->host_lock, flags); 2085 if (scsi_host_set_state(shost, SHOST_RUNNING)) 2086 if (scsi_host_set_state(shost, SHOST_CANCEL)) 2087 BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); 2088 spin_unlock_irqrestore(shost->host_lock, flags); 2089 2090 wake_up(&shost->host_wait); 2091 2092 /* 2093 * finally we need to re-initiate requests that may be pending. we will 2094 * have had everything blocked while error handling is taking place, and 2095 * now that error recovery is done, we will need to ensure that these 2096 * requests are started. 2097 */ 2098 scsi_run_host_queues(shost); 2099 2100 /* 2101 * if eh is active and host_eh_scheduled is pending we need to re-run 2102 * recovery. we do this check after scsi_run_host_queues() to allow 2103 * everything pent up since the last eh run a chance to make forward 2104 * progress before we sync again. Either we'll immediately re-run 2105 * recovery or scsi_device_unbusy() will wake us again when these 2106 * pending commands complete. 2107 */ 2108 spin_lock_irqsave(shost->host_lock, flags); 2109 if (shost->host_eh_scheduled) 2110 if (scsi_host_set_state(shost, SHOST_RECOVERY)) 2111 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); 2112 spin_unlock_irqrestore(shost->host_lock, flags); 2113} 2114 2115/** 2116 * scsi_eh_ready_devs - check device ready state and recover if not. 2117 * @shost: host to be recovered. 2118 * @work_q: &list_head for pending commands. 2119 * @done_q: &list_head for processed commands. 2120 */ 2121void scsi_eh_ready_devs(struct Scsi_Host *shost, 2122 struct list_head *work_q, 2123 struct list_head *done_q) 2124{ 2125 if (!scsi_eh_stu(shost, work_q, done_q)) 2126 if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) 2127 if (!scsi_eh_target_reset(shost, work_q, done_q)) 2128 if (!scsi_eh_bus_reset(shost, work_q, done_q)) 2129 if (!scsi_eh_host_reset(shost, work_q, done_q)) 2130 scsi_eh_offline_sdevs(work_q, 2131 done_q); 2132} 2133EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); 2134 2135/** 2136 * scsi_eh_flush_done_q - finish processed commands or retry them. 2137 * @done_q: list_head of processed commands. 2138 */ 2139void scsi_eh_flush_done_q(struct list_head *done_q) 2140{ 2141 struct scsi_cmnd *scmd, *next; 2142 2143 list_for_each_entry_safe(scmd, next, done_q, eh_entry) { 2144 list_del_init(&scmd->eh_entry); 2145 if (scsi_device_online(scmd->device) && 2146 !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd) && 2147 scsi_eh_should_retry_cmd(scmd)) { 2148 SCSI_LOG_ERROR_RECOVERY(3, 2149 scmd_printk(KERN_INFO, scmd, 2150 "%s: flush retry cmd\n", 2151 current->comm)); 2152 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); 2153 } else { 2154 /* 2155 * If just we got sense for the device (called 2156 * scsi_eh_get_sense), scmd->result is already 2157 * set, do not set DID_TIME_OUT. 2158 */ 2159 if (!scmd->result) 2160 scmd->result |= (DID_TIME_OUT << 16); 2161 SCSI_LOG_ERROR_RECOVERY(3, 2162 scmd_printk(KERN_INFO, scmd, 2163 "%s: flush finish cmd\n", 2164 current->comm)); 2165 scsi_finish_command(scmd); 2166 } 2167 } 2168} 2169EXPORT_SYMBOL(scsi_eh_flush_done_q); 2170 2171/** 2172 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. 2173 * @shost: Host to unjam. 2174 * 2175 * Notes: 2176 * When we come in here, we *know* that all commands on the bus have 2177 * either completed, failed or timed out. we also know that no further 2178 * commands are being sent to the host, so things are relatively quiet 2179 * and we have freedom to fiddle with things as we wish. 2180 * 2181 * This is only the *default* implementation. it is possible for 2182 * individual drivers to supply their own version of this function, and 2183 * if the maintainer wishes to do this, it is strongly suggested that 2184 * this function be taken as a template and modified. this function 2185 * was designed to correctly handle problems for about 95% of the 2186 * different cases out there, and it should always provide at least a 2187 * reasonable amount of error recovery. 2188 * 2189 * Any command marked 'failed' or 'timeout' must eventually have 2190 * scsi_finish_cmd() called for it. we do all of the retry stuff 2191 * here, so when we restart the host after we return it should have an 2192 * empty queue. 2193 */ 2194static void scsi_unjam_host(struct Scsi_Host *shost) 2195{ 2196 unsigned long flags; 2197 LIST_HEAD(eh_work_q); 2198 LIST_HEAD(eh_done_q); 2199 2200 spin_lock_irqsave(shost->host_lock, flags); 2201 list_splice_init(&shost->eh_cmd_q, &eh_work_q); 2202 spin_unlock_irqrestore(shost->host_lock, flags); 2203 2204 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); 2205 2206 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) 2207 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); 2208 2209 spin_lock_irqsave(shost->host_lock, flags); 2210 if (shost->eh_deadline != -1) 2211 shost->last_reset = 0; 2212 spin_unlock_irqrestore(shost->host_lock, flags); 2213 scsi_eh_flush_done_q(&eh_done_q); 2214} 2215 2216/** 2217 * scsi_error_handler - SCSI error handler thread 2218 * @data: Host for which we are running. 2219 * 2220 * Notes: 2221 * This is the main error handling loop. This is run as a kernel thread 2222 * for every SCSI host and handles all error handling activity. 2223 */ 2224int scsi_error_handler(void *data) 2225{ 2226 struct Scsi_Host *shost = data; 2227 2228 /* 2229 * We use TASK_INTERRUPTIBLE so that the thread is not 2230 * counted against the load average as a running process. 2231 * We never actually get interrupted because kthread_run 2232 * disables signal delivery for the created thread. 2233 */ 2234 while (true) { 2235 /* 2236 * The sequence in kthread_stop() sets the stop flag first 2237 * then wakes the process. To avoid missed wakeups, the task 2238 * should always be in a non running state before the stop 2239 * flag is checked 2240 */ 2241 set_current_state(TASK_INTERRUPTIBLE); 2242 if (kthread_should_stop()) 2243 break; 2244 2245 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || 2246 shost->host_failed != scsi_host_busy(shost)) { 2247 SCSI_LOG_ERROR_RECOVERY(1, 2248 shost_printk(KERN_INFO, shost, 2249 "scsi_eh_%d: sleeping\n", 2250 shost->host_no)); 2251 schedule(); 2252 continue; 2253 } 2254 2255 __set_current_state(TASK_RUNNING); 2256 SCSI_LOG_ERROR_RECOVERY(1, 2257 shost_printk(KERN_INFO, shost, 2258 "scsi_eh_%d: waking up %d/%d/%d\n", 2259 shost->host_no, shost->host_eh_scheduled, 2260 shost->host_failed, 2261 scsi_host_busy(shost))); 2262 2263 /* 2264 * We have a host that is failing for some reason. Figure out 2265 * what we need to do to get it up and online again (if we can). 2266 * If we fail, we end up taking the thing offline. 2267 */ 2268 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { 2269 SCSI_LOG_ERROR_RECOVERY(1, 2270 shost_printk(KERN_ERR, shost, 2271 "scsi_eh_%d: unable to autoresume\n", 2272 shost->host_no)); 2273 continue; 2274 } 2275 2276 if (shost->transportt->eh_strategy_handler) 2277 shost->transportt->eh_strategy_handler(shost); 2278 else 2279 scsi_unjam_host(shost); 2280 2281 /* All scmds have been handled */ 2282 shost->host_failed = 0; 2283 2284 /* 2285 * Note - if the above fails completely, the action is to take 2286 * individual devices offline and flush the queue of any 2287 * outstanding requests that may have been pending. When we 2288 * restart, we restart any I/O to any other devices on the bus 2289 * which are still online. 2290 */ 2291 scsi_restart_operations(shost); 2292 if (!shost->eh_noresume) 2293 scsi_autopm_put_host(shost); 2294 } 2295 __set_current_state(TASK_RUNNING); 2296 2297 SCSI_LOG_ERROR_RECOVERY(1, 2298 shost_printk(KERN_INFO, shost, 2299 "Error handler scsi_eh_%d exiting\n", 2300 shost->host_no)); 2301 shost->ehandler = NULL; 2302 return 0; 2303} 2304 2305/* 2306 * Function: scsi_report_bus_reset() 2307 * 2308 * Purpose: Utility function used by low-level drivers to report that 2309 * they have observed a bus reset on the bus being handled. 2310 * 2311 * Arguments: shost - Host in question 2312 * channel - channel on which reset was observed. 2313 * 2314 * Returns: Nothing 2315 * 2316 * Lock status: Host lock must be held. 2317 * 2318 * Notes: This only needs to be called if the reset is one which 2319 * originates from an unknown location. Resets originated 2320 * by the mid-level itself don't need to call this, but there 2321 * should be no harm. 2322 * 2323 * The main purpose of this is to make sure that a CHECK_CONDITION 2324 * is properly treated. 2325 */ 2326void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) 2327{ 2328 struct scsi_device *sdev; 2329 2330 __shost_for_each_device(sdev, shost) { 2331 if (channel == sdev_channel(sdev)) 2332 __scsi_report_device_reset(sdev, NULL); 2333 } 2334} 2335EXPORT_SYMBOL(scsi_report_bus_reset); 2336 2337/* 2338 * Function: scsi_report_device_reset() 2339 * 2340 * Purpose: Utility function used by low-level drivers to report that 2341 * they have observed a device reset on the device being handled. 2342 * 2343 * Arguments: shost - Host in question 2344 * channel - channel on which reset was observed 2345 * target - target on which reset was observed 2346 * 2347 * Returns: Nothing 2348 * 2349 * Lock status: Host lock must be held 2350 * 2351 * Notes: This only needs to be called if the reset is one which 2352 * originates from an unknown location. Resets originated 2353 * by the mid-level itself don't need to call this, but there 2354 * should be no harm. 2355 * 2356 * The main purpose of this is to make sure that a CHECK_CONDITION 2357 * is properly treated. 2358 */ 2359void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) 2360{ 2361 struct scsi_device *sdev; 2362 2363 __shost_for_each_device(sdev, shost) { 2364 if (channel == sdev_channel(sdev) && 2365 target == sdev_id(sdev)) 2366 __scsi_report_device_reset(sdev, NULL); 2367 } 2368} 2369EXPORT_SYMBOL(scsi_report_device_reset); 2370 2371/** 2372 * scsi_ioctl_reset: explicitly reset a host/bus/target/device 2373 * @dev: scsi_device to operate on 2374 * @arg: reset type (see sg.h) 2375 */ 2376int 2377scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) 2378{ 2379 struct scsi_cmnd *scmd; 2380 struct Scsi_Host *shost = dev->host; 2381 struct request *rq; 2382 unsigned long flags; 2383 int error = 0, val; 2384 enum scsi_disposition rtn; 2385 2386 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) 2387 return -EACCES; 2388 2389 error = get_user(val, arg); 2390 if (error) 2391 return error; 2392 2393 if (scsi_autopm_get_host(shost) < 0) 2394 return -EIO; 2395 2396 error = -EIO; 2397 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) + 2398 shost->hostt->cmd_size, GFP_KERNEL); 2399 if (!rq) 2400 goto out_put_autopm_host; 2401 blk_rq_init(NULL, rq); 2402 2403 scmd = (struct scsi_cmnd *)(rq + 1); 2404 scsi_init_command(dev, scmd); 2405 2406 scmd->submitter = SUBMITTED_BY_SCSI_RESET_IOCTL; 2407 memset(&scmd->sdb, 0, sizeof(scmd->sdb)); 2408 2409 scmd->cmd_len = 0; 2410 2411 scmd->sc_data_direction = DMA_BIDIRECTIONAL; 2412 2413 spin_lock_irqsave(shost->host_lock, flags); 2414 shost->tmf_in_progress = 1; 2415 spin_unlock_irqrestore(shost->host_lock, flags); 2416 2417 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { 2418 case SG_SCSI_RESET_NOTHING: 2419 rtn = SUCCESS; 2420 break; 2421 case SG_SCSI_RESET_DEVICE: 2422 rtn = scsi_try_bus_device_reset(scmd); 2423 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2424 break; 2425 fallthrough; 2426 case SG_SCSI_RESET_TARGET: 2427 rtn = scsi_try_target_reset(scmd); 2428 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2429 break; 2430 fallthrough; 2431 case SG_SCSI_RESET_BUS: 2432 rtn = scsi_try_bus_reset(scmd); 2433 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) 2434 break; 2435 fallthrough; 2436 case SG_SCSI_RESET_HOST: 2437 rtn = scsi_try_host_reset(scmd); 2438 if (rtn == SUCCESS) 2439 break; 2440 fallthrough; 2441 default: 2442 rtn = FAILED; 2443 break; 2444 } 2445 2446 error = (rtn == SUCCESS) ? 0 : -EIO; 2447 2448 spin_lock_irqsave(shost->host_lock, flags); 2449 shost->tmf_in_progress = 0; 2450 spin_unlock_irqrestore(shost->host_lock, flags); 2451 2452 /* 2453 * be sure to wake up anyone who was sleeping or had their queue 2454 * suspended while we performed the TMF. 2455 */ 2456 SCSI_LOG_ERROR_RECOVERY(3, 2457 shost_printk(KERN_INFO, shost, 2458 "waking up host to restart after TMF\n")); 2459 2460 wake_up(&shost->host_wait); 2461 scsi_run_host_queues(shost); 2462 2463 kfree(rq); 2464 2465out_put_autopm_host: 2466 scsi_autopm_put_host(shost); 2467 return error; 2468} 2469 2470bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, 2471 struct scsi_sense_hdr *sshdr) 2472{ 2473 return scsi_normalize_sense(cmd->sense_buffer, 2474 SCSI_SENSE_BUFFERSIZE, sshdr); 2475} 2476EXPORT_SYMBOL(scsi_command_normalize_sense); 2477 2478/** 2479 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) 2480 * @sense_buffer: byte array of sense data 2481 * @sb_len: number of valid bytes in sense_buffer 2482 * @info_out: pointer to 64 integer where 8 or 4 byte information 2483 * field will be placed if found. 2484 * 2485 * Return value: 2486 * true if information field found, false if not found. 2487 */ 2488bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len, 2489 u64 *info_out) 2490{ 2491 const u8 * ucp; 2492 2493 if (sb_len < 7) 2494 return false; 2495 switch (sense_buffer[0] & 0x7f) { 2496 case 0x70: 2497 case 0x71: 2498 if (sense_buffer[0] & 0x80) { 2499 *info_out = get_unaligned_be32(&sense_buffer[3]); 2500 return true; 2501 } 2502 return false; 2503 case 0x72: 2504 case 0x73: 2505 ucp = scsi_sense_desc_find(sense_buffer, sb_len, 2506 0 /* info desc */); 2507 if (ucp && (0xa == ucp[1])) { 2508 *info_out = get_unaligned_be64(&ucp[4]); 2509 return true; 2510 } 2511 return false; 2512 default: 2513 return false; 2514 } 2515} 2516EXPORT_SYMBOL(scsi_get_sense_info_fld);