Linux Audio

Check our new training course

Loading...
v5.14.15
   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
  53static void scsi_eh_done(struct scsi_cmnd *scmd);
  54
  55/*
  56 * These should *probably* be handled by the host itself.
  57 * Since it is allowed to sleep, it probably should.
  58 */
  59#define BUS_RESET_SETTLE_TIME   (10)
  60#define HOST_RESET_SETTLE_TIME  (10)
  61
  62static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
  63static enum scsi_disposition scsi_try_to_abort_cmd(struct scsi_host_template *,
  64						   struct scsi_cmnd *);
  65
 
  66void scsi_eh_wakeup(struct Scsi_Host *shost)
  67{
  68	lockdep_assert_held(shost->host_lock);
  69
  70	if (scsi_host_busy(shost) == shost->host_failed) {
  71		trace_scsi_eh_wakeup(shost);
  72		wake_up_process(shost->ehandler);
  73		SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
  74			"Waking error handler thread\n"));
  75	}
  76}
  77
  78/**
  79 * scsi_schedule_eh - schedule EH for SCSI host
  80 * @shost:	SCSI host to invoke error handling on.
  81 *
  82 * Schedule SCSI EH without scmd.
  83 */
  84void scsi_schedule_eh(struct Scsi_Host *shost)
  85{
  86	unsigned long flags;
  87
  88	spin_lock_irqsave(shost->host_lock, flags);
  89
  90	if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
  91	    scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
  92		shost->host_eh_scheduled++;
  93		scsi_eh_wakeup(shost);
  94	}
  95
  96	spin_unlock_irqrestore(shost->host_lock, flags);
  97}
  98EXPORT_SYMBOL_GPL(scsi_schedule_eh);
  99
 100static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
 101{
 102	if (!shost->last_reset || shost->eh_deadline == -1)
 103		return 0;
 104
 105	/*
 106	 * 32bit accesses are guaranteed to be atomic
 107	 * (on all supported architectures), so instead
 108	 * of using a spinlock we can as well double check
 109	 * if eh_deadline has been set to 'off' during the
 110	 * time_before call.
 111	 */
 112	if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
 113	    shost->eh_deadline > -1)
 114		return 0;
 115
 116	return 1;
 117}
 118
 119static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd)
 120{
 121	if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
 122		return true;
 123
 124	return ++cmd->retries <= cmd->allowed;
 125}
 126
 127static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd)
 128{
 129	struct scsi_device *sdev = cmd->device;
 130	struct Scsi_Host *host = sdev->host;
 131
 132	if (host->hostt->eh_should_retry_cmd)
 133		return  host->hostt->eh_should_retry_cmd(cmd);
 134
 135	return true;
 136}
 137
 138/**
 139 * scmd_eh_abort_handler - Handle command aborts
 140 * @work:	command to be aborted.
 141 *
 142 * Note: this function must be called only for a command that has timed out.
 143 * Because the block layer marks a request as complete before it calls
 144 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
 145 * timed out do not have any effect. Hence it is safe to call
 146 * scsi_finish_command() from this function.
 147 */
 148void
 149scmd_eh_abort_handler(struct work_struct *work)
 150{
 151	struct scsi_cmnd *scmd =
 152		container_of(work, struct scsi_cmnd, abort_work.work);
 153	struct scsi_device *sdev = scmd->device;
 154	enum scsi_disposition rtn;
 155
 156	if (scsi_host_eh_past_deadline(sdev->host)) {
 157		SCSI_LOG_ERROR_RECOVERY(3,
 158			scmd_printk(KERN_INFO, scmd,
 159				    "eh timeout, not aborting\n"));
 
 160	} else {
 161		SCSI_LOG_ERROR_RECOVERY(3,
 162			scmd_printk(KERN_INFO, scmd,
 163				    "aborting command\n"));
 164		rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
 165		if (rtn == SUCCESS) {
 166			set_host_byte(scmd, DID_TIME_OUT);
 167			if (scsi_host_eh_past_deadline(sdev->host)) {
 168				SCSI_LOG_ERROR_RECOVERY(3,
 169					scmd_printk(KERN_INFO, scmd,
 170						    "eh timeout, not retrying "
 171						    "aborted command\n"));
 
 172			} else if (!scsi_noretry_cmd(scmd) &&
 173				   scsi_cmd_retry_allowed(scmd) &&
 174				scsi_eh_should_retry_cmd(scmd)) {
 175				SCSI_LOG_ERROR_RECOVERY(3,
 176					scmd_printk(KERN_WARNING, scmd,
 177						    "retry aborted command\n"));
 
 178				scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
 179				return;
 180			} else {
 181				SCSI_LOG_ERROR_RECOVERY(3,
 182					scmd_printk(KERN_WARNING, scmd,
 183						    "finish aborted command\n"));
 
 184				scsi_finish_command(scmd);
 185				return;
 186			}
 187		} else {
 188			SCSI_LOG_ERROR_RECOVERY(3,
 189				scmd_printk(KERN_INFO, scmd,
 190					    "cmd abort %s\n",
 191					    (rtn == FAST_IO_FAIL) ?
 192					    "not send" : "failed"));
 193		}
 194	}
 195
 196	scsi_eh_scmd_add(scmd);
 
 
 
 
 
 
 
 197}
 198
 199/**
 200 * scsi_abort_command - schedule a command abort
 201 * @scmd:	scmd to abort.
 202 *
 203 * We only need to abort commands after a command timeout
 204 */
 205static int
 206scsi_abort_command(struct scsi_cmnd *scmd)
 207{
 208	struct scsi_device *sdev = scmd->device;
 209	struct Scsi_Host *shost = sdev->host;
 210	unsigned long flags;
 211
 212	if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
 213		/*
 214		 * Retry after abort failed, escalate to next level.
 215		 */
 
 216		SCSI_LOG_ERROR_RECOVERY(3,
 217			scmd_printk(KERN_INFO, scmd,
 218				    "previous abort failed\n"));
 219		BUG_ON(delayed_work_pending(&scmd->abort_work));
 220		return FAILED;
 221	}
 222
 
 
 
 
 223	spin_lock_irqsave(shost->host_lock, flags);
 
 
 
 
 
 
 
 
 
 224	if (shost->eh_deadline != -1 && !shost->last_reset)
 225		shost->last_reset = jiffies;
 226	spin_unlock_irqrestore(shost->host_lock, flags);
 227
 228	scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
 229	SCSI_LOG_ERROR_RECOVERY(3,
 230		scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
 
 231	queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
 232	return SUCCESS;
 233}
 234
 235/**
 236 * scsi_eh_reset - call into ->eh_action to reset internal counters
 237 * @scmd:	scmd to run eh on.
 238 *
 239 * The scsi driver might be carrying internal state about the
 240 * devices, so we need to call into the driver to reset the
 241 * internal state once the error handler is started.
 242 */
 243static void scsi_eh_reset(struct scsi_cmnd *scmd)
 244{
 245	if (!blk_rq_is_passthrough(scmd->request)) {
 246		struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
 247		if (sdrv->eh_reset)
 248			sdrv->eh_reset(scmd);
 249	}
 250}
 251
 252static void scsi_eh_inc_host_failed(struct rcu_head *head)
 253{
 254	struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
 255	struct Scsi_Host *shost = scmd->device->host;
 256	unsigned long flags;
 257
 258	spin_lock_irqsave(shost->host_lock, flags);
 259	shost->host_failed++;
 260	scsi_eh_wakeup(shost);
 261	spin_unlock_irqrestore(shost->host_lock, flags);
 262}
 263
 264/**
 265 * scsi_eh_scmd_add - add scsi cmd to error handling.
 266 * @scmd:	scmd to run eh on.
 
 
 
 
 267 */
 268void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
 269{
 270	struct Scsi_Host *shost = scmd->device->host;
 271	unsigned long flags;
 272	int ret;
 273
 274	WARN_ON_ONCE(!shost->ehandler);
 
 275
 276	spin_lock_irqsave(shost->host_lock, flags);
 277	if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
 278		ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
 279		WARN_ON_ONCE(ret);
 280	}
 281	if (shost->eh_deadline != -1 && !shost->last_reset)
 282		shost->last_reset = jiffies;
 283
 284	scsi_eh_reset(scmd);
 
 
 
 285	list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
 
 
 
 286	spin_unlock_irqrestore(shost->host_lock, flags);
 287	/*
 288	 * Ensure that all tasks observe the host state change before the
 289	 * host_failed change.
 290	 */
 291	call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
 292}
 293
 294/**
 295 * scsi_times_out - Timeout function for normal scsi commands.
 296 * @req:	request that is timing out.
 297 *
 298 * Notes:
 299 *     We do not need to lock this.  There is the potential for a race
 300 *     only in that the normal completion handling might run, but if the
 301 *     normal completion function determines that the timer has already
 302 *     fired, then it mustn't do anything.
 303 */
 304enum blk_eh_timer_return scsi_times_out(struct request *req)
 305{
 306	struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
 307	enum blk_eh_timer_return rtn = BLK_EH_DONE;
 308	struct Scsi_Host *host = scmd->device->host;
 309
 310	trace_scsi_dispatch_cmd_timeout(scmd);
 311	scsi_log_completion(scmd, TIMEOUT_ERROR);
 312
 313	if (host->eh_deadline != -1 && !host->last_reset)
 314		host->last_reset = jiffies;
 315
 316	if (host->hostt->eh_timed_out)
 
 
 317		rtn = host->hostt->eh_timed_out(scmd);
 318
 319	if (rtn == BLK_EH_DONE) {
 320		/*
 321		 * Set the command to complete first in order to prevent a real
 322		 * completion from releasing the command while error handling
 323		 * is using it. If the command was already completed, then the
 324		 * lower level driver beat the timeout handler, and it is safe
 325		 * to return without escalating error recovery.
 326		 *
 327		 * If timeout handling lost the race to a real completion, the
 328		 * block layer may ignore that due to a fake timeout injection,
 329		 * so return RESET_TIMER to allow error handling another shot
 330		 * at this command.
 331		 */
 332		if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
 333			return BLK_EH_RESET_TIMER;
 334		if (scsi_abort_command(scmd) != SUCCESS) {
 335			set_host_byte(scmd, DID_TIME_OUT);
 336			scsi_eh_scmd_add(scmd);
 337		}
 338	}
 339
 340	return rtn;
 341}
 342
 343/**
 344 * scsi_block_when_processing_errors - Prevent cmds from being queued.
 345 * @sdev:	Device on which we are performing recovery.
 346 *
 347 * Description:
 348 *     We block until the host is out of error recovery, and then check to
 349 *     see whether the host or the device is offline.
 350 *
 351 * Return value:
 352 *     0 when dev was taken offline by error recovery. 1 OK to proceed.
 353 */
 354int scsi_block_when_processing_errors(struct scsi_device *sdev)
 355{
 356	int online;
 357
 358	wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
 359
 360	online = scsi_device_online(sdev);
 361
 
 
 
 362	return online;
 363}
 364EXPORT_SYMBOL(scsi_block_when_processing_errors);
 365
 366#ifdef CONFIG_SCSI_LOGGING
 367/**
 368 * scsi_eh_prt_fail_stats - Log info on failures.
 369 * @shost:	scsi host being recovered.
 370 * @work_q:	Queue of scsi cmds to process.
 371 */
 372static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
 373					  struct list_head *work_q)
 374{
 375	struct scsi_cmnd *scmd;
 376	struct scsi_device *sdev;
 377	int total_failures = 0;
 378	int cmd_failed = 0;
 379	int cmd_cancel = 0;
 380	int devices_failed = 0;
 381
 382	shost_for_each_device(sdev, shost) {
 383		list_for_each_entry(scmd, work_q, eh_entry) {
 384			if (scmd->device == sdev) {
 385				++total_failures;
 386				if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
 387					++cmd_cancel;
 388				else
 389					++cmd_failed;
 390			}
 391		}
 392
 393		if (cmd_cancel || cmd_failed) {
 394			SCSI_LOG_ERROR_RECOVERY(3,
 395				shost_printk(KERN_INFO, shost,
 396					    "%s: cmds failed: %d, cancel: %d\n",
 397					    __func__, cmd_failed,
 398					    cmd_cancel));
 399			cmd_cancel = 0;
 400			cmd_failed = 0;
 401			++devices_failed;
 402		}
 403	}
 404
 405	SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
 406				   "Total of %d commands on %d"
 407				   " devices require eh work\n",
 408				   total_failures, devices_failed));
 409}
 410#endif
 411
 412 /**
 413 * scsi_report_lun_change - Set flag on all *other* devices on the same target
 414 *                          to indicate that a UNIT ATTENTION is expected.
 415 * @sdev:	Device reporting the UNIT ATTENTION
 416 */
 417static void scsi_report_lun_change(struct scsi_device *sdev)
 418{
 419	sdev->sdev_target->expecting_lun_change = 1;
 420}
 421
 422/**
 423 * scsi_report_sense - Examine scsi sense information and log messages for
 424 *		       certain conditions, also issue uevents for some of them.
 425 * @sdev:	Device reporting the sense code
 426 * @sshdr:	sshdr to be examined
 427 */
 428static void scsi_report_sense(struct scsi_device *sdev,
 429			      struct scsi_sense_hdr *sshdr)
 430{
 431	enum scsi_device_event evt_type = SDEV_EVT_MAXBITS;	/* i.e. none */
 432
 433	if (sshdr->sense_key == UNIT_ATTENTION) {
 434		if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
 435			evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
 436			sdev_printk(KERN_WARNING, sdev,
 437				    "Inquiry data has changed");
 438		} else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
 439			evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
 440			scsi_report_lun_change(sdev);
 441			sdev_printk(KERN_WARNING, sdev,
 442				    "Warning! Received an indication that the "
 443				    "LUN assignments on this target have "
 444				    "changed. The Linux SCSI layer does not "
 445				    "automatically remap LUN assignments.\n");
 446		} else if (sshdr->asc == 0x3f)
 447			sdev_printk(KERN_WARNING, sdev,
 448				    "Warning! Received an indication that the "
 449				    "operating parameters on this target have "
 450				    "changed. The Linux SCSI layer does not "
 451				    "automatically adjust these parameters.\n");
 452
 453		if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
 454			evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
 455			sdev_printk(KERN_WARNING, sdev,
 456				    "Warning! Received an indication that the "
 457				    "LUN reached a thin provisioning soft "
 458				    "threshold.\n");
 459		}
 460
 461		if (sshdr->asc == 0x29) {
 462			evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
 463			sdev_printk(KERN_WARNING, sdev,
 464				    "Power-on or device reset occurred\n");
 465		}
 466
 467		if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
 468			evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
 469			sdev_printk(KERN_WARNING, sdev,
 470				    "Mode parameters changed");
 471		} else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
 472			evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
 473			sdev_printk(KERN_WARNING, sdev,
 474				    "Asymmetric access state changed");
 475		} else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
 476			evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
 477			sdev_printk(KERN_WARNING, sdev,
 478				    "Capacity data has changed");
 479		} else if (sshdr->asc == 0x2a)
 480			sdev_printk(KERN_WARNING, sdev,
 481				    "Parameters changed");
 482	}
 483
 484	if (evt_type != SDEV_EVT_MAXBITS) {
 485		set_bit(evt_type, sdev->pending_events);
 486		schedule_work(&sdev->event_work);
 487	}
 488}
 489
 490/**
 491 * scsi_check_sense - Examine scsi cmd sense
 492 * @scmd:	Cmd to have sense checked.
 493 *
 494 * Return value:
 495 *	SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
 496 *
 497 * Notes:
 498 *	When a deferred error is detected the current command has
 499 *	not been executed and needs retrying.
 500 */
 501enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd)
 502{
 503	struct scsi_device *sdev = scmd->device;
 504	struct scsi_sense_hdr sshdr;
 505
 506	if (! scsi_command_normalize_sense(scmd, &sshdr))
 507		return FAILED;	/* no valid sense data */
 508
 
 
 
 
 
 
 
 
 509	scsi_report_sense(sdev, &sshdr);
 510
 511	if (scsi_sense_is_deferred(&sshdr))
 512		return NEEDS_RETRY;
 513
 514	if (sdev->handler && sdev->handler->check_sense) {
 515		enum scsi_disposition rc;
 
 516
 517		rc = sdev->handler->check_sense(sdev, &sshdr);
 518		if (rc != SCSI_RETURN_NOT_HANDLED)
 519			return rc;
 520		/* handler does not care. Drop down to default handling */
 521	}
 522
 523	if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
 524		/*
 525		 * nasty: for mid-layer issued TURs, we need to return the
 526		 * actual sense data without any recovery attempt.  For eh
 527		 * issued ones, we need to try to recover and interpret
 528		 */
 529		return SUCCESS;
 530
 531	/*
 532	 * Previous logic looked for FILEMARK, EOM or ILI which are
 533	 * mainly associated with tapes and returned SUCCESS.
 534	 */
 535	if (sshdr.response_code == 0x70) {
 536		/* fixed format */
 537		if (scmd->sense_buffer[2] & 0xe0)
 538			return SUCCESS;
 539	} else {
 540		/*
 541		 * descriptor format: look for "stream commands sense data
 542		 * descriptor" (see SSC-3). Assume single sense data
 543		 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
 544		 */
 545		if ((sshdr.additional_length > 3) &&
 546		    (scmd->sense_buffer[8] == 0x4) &&
 547		    (scmd->sense_buffer[11] & 0xe0))
 548			return SUCCESS;
 549	}
 550
 551	switch (sshdr.sense_key) {
 552	case NO_SENSE:
 553		return SUCCESS;
 554	case RECOVERED_ERROR:
 555		return /* soft_error */ SUCCESS;
 556
 557	case ABORTED_COMMAND:
 558		if (sshdr.asc == 0x10) /* DIF */
 559			return SUCCESS;
 560
 561		if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
 562			return ADD_TO_MLQUEUE;
 563		if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
 564		    sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
 565			return ADD_TO_MLQUEUE;
 566
 567		return NEEDS_RETRY;
 568	case NOT_READY:
 569	case UNIT_ATTENTION:
 570		/*
 571		 * if we are expecting a cc/ua because of a bus reset that we
 572		 * performed, treat this just as a retry.  otherwise this is
 573		 * information that we should pass up to the upper-level driver
 574		 * so that we can deal with it there.
 575		 */
 576		if (scmd->device->expecting_cc_ua) {
 577			/*
 578			 * Because some device does not queue unit
 579			 * attentions correctly, we carefully check
 580			 * additional sense code and qualifier so as
 581			 * not to squash media change unit attention.
 582			 */
 583			if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
 584				scmd->device->expecting_cc_ua = 0;
 585				return NEEDS_RETRY;
 586			}
 587		}
 588		/*
 589		 * we might also expect a cc/ua if another LUN on the target
 590		 * reported a UA with an ASC/ASCQ of 3F 0E -
 591		 * REPORTED LUNS DATA HAS CHANGED.
 592		 */
 593		if (scmd->device->sdev_target->expecting_lun_change &&
 594		    sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
 595			return NEEDS_RETRY;
 596		/*
 597		 * if the device is in the process of becoming ready, we
 598		 * should retry.
 599		 */
 600		if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
 601			return NEEDS_RETRY;
 602		/*
 603		 * if the device is not started, we need to wake
 604		 * the error handler to start the motor
 605		 */
 606		if (scmd->device->allow_restart &&
 607		    (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
 608			return FAILED;
 609		/*
 610		 * Pass the UA upwards for a determination in the completion
 611		 * functions.
 612		 */
 613		return SUCCESS;
 614
 615		/* these are not supported */
 616	case DATA_PROTECT:
 617		if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
 618			/* Thin provisioning hard threshold reached */
 619			set_host_byte(scmd, DID_ALLOC_FAILURE);
 620			return SUCCESS;
 621		}
 622		fallthrough;
 623	case COPY_ABORTED:
 624	case VOLUME_OVERFLOW:
 625	case MISCOMPARE:
 626	case BLANK_CHECK:
 627		set_host_byte(scmd, DID_TARGET_FAILURE);
 628		return SUCCESS;
 629
 630	case MEDIUM_ERROR:
 631		if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
 632		    sshdr.asc == 0x13 || /* AMNF DATA FIELD */
 633		    sshdr.asc == 0x14) { /* RECORD NOT FOUND */
 634			set_host_byte(scmd, DID_MEDIUM_ERROR);
 635			return SUCCESS;
 636		}
 637		return NEEDS_RETRY;
 638
 639	case HARDWARE_ERROR:
 640		if (scmd->device->retry_hwerror)
 641			return ADD_TO_MLQUEUE;
 642		else
 643			set_host_byte(scmd, DID_TARGET_FAILURE);
 644		fallthrough;
 645
 646	case ILLEGAL_REQUEST:
 647		if (sshdr.asc == 0x20 || /* Invalid command operation code */
 648		    sshdr.asc == 0x21 || /* Logical block address out of range */
 649		    sshdr.asc == 0x22 || /* Invalid function */
 650		    sshdr.asc == 0x24 || /* Invalid field in cdb */
 651		    sshdr.asc == 0x26 || /* Parameter value invalid */
 652		    sshdr.asc == 0x27) { /* Write protected */
 653			set_host_byte(scmd, DID_TARGET_FAILURE);
 654		}
 655		return SUCCESS;
 656
 657	default:
 658		return SUCCESS;
 659	}
 660}
 661EXPORT_SYMBOL_GPL(scsi_check_sense);
 662
 663static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
 664{
 665	struct scsi_host_template *sht = sdev->host->hostt;
 666	struct scsi_device *tmp_sdev;
 667
 668	if (!sht->track_queue_depth ||
 669	    sdev->queue_depth >= sdev->max_queue_depth)
 670		return;
 671
 672	if (time_before(jiffies,
 673	    sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
 674		return;
 675
 676	if (time_before(jiffies,
 677	    sdev->last_queue_full_time + sdev->queue_ramp_up_period))
 678		return;
 679
 680	/*
 681	 * Walk all devices of a target and do
 682	 * ramp up on them.
 683	 */
 684	shost_for_each_device(tmp_sdev, sdev->host) {
 685		if (tmp_sdev->channel != sdev->channel ||
 686		    tmp_sdev->id != sdev->id ||
 687		    tmp_sdev->queue_depth == sdev->max_queue_depth)
 688			continue;
 689
 690		scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
 
 
 
 
 691		sdev->last_queue_ramp_up = jiffies;
 692	}
 693}
 694
 695static void scsi_handle_queue_full(struct scsi_device *sdev)
 696{
 697	struct scsi_host_template *sht = sdev->host->hostt;
 698	struct scsi_device *tmp_sdev;
 699
 700	if (!sht->track_queue_depth)
 701		return;
 702
 703	shost_for_each_device(tmp_sdev, sdev->host) {
 704		if (tmp_sdev->channel != sdev->channel ||
 705		    tmp_sdev->id != sdev->id)
 706			continue;
 707		/*
 708		 * We do not know the number of commands that were at
 709		 * the device when we got the queue full so we start
 710		 * from the highest possible value and work our way down.
 711		 */
 712		scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
 
 713	}
 714}
 715
 716/**
 717 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
 718 * @scmd:	SCSI cmd to examine.
 719 *
 720 * Notes:
 721 *    This is *only* called when we are examining the status of commands
 722 *    queued during error recovery.  the main difference here is that we
 723 *    don't allow for the possibility of retries here, and we are a lot
 724 *    more restrictive about what we consider acceptable.
 725 */
 726static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd)
 727{
 728	/*
 729	 * first check the host byte, to see if there is anything in there
 730	 * that would indicate what we need to do.
 731	 */
 732	if (host_byte(scmd->result) == DID_RESET) {
 733		/*
 734		 * rats.  we are already in the error handler, so we now
 735		 * get to try and figure out what to do next.  if the sense
 736		 * is valid, we have a pretty good idea of what to do.
 737		 * if not, we mark it as FAILED.
 738		 */
 739		return scsi_check_sense(scmd);
 740	}
 741	if (host_byte(scmd->result) != DID_OK)
 742		return FAILED;
 743
 744	/*
 
 
 
 
 
 
 745	 * now, check the status byte to see if this indicates
 746	 * anything special.
 747	 */
 748	switch (get_status_byte(scmd)) {
 749	case SAM_STAT_GOOD:
 750		scsi_handle_queue_ramp_up(scmd->device);
 751		fallthrough;
 752	case SAM_STAT_COMMAND_TERMINATED:
 753		return SUCCESS;
 754	case SAM_STAT_CHECK_CONDITION:
 755		return scsi_check_sense(scmd);
 756	case SAM_STAT_CONDITION_MET:
 757	case SAM_STAT_INTERMEDIATE:
 758	case SAM_STAT_INTERMEDIATE_CONDITION_MET:
 759		/*
 760		 * who knows?  FIXME(eric)
 761		 */
 762		return SUCCESS;
 763	case SAM_STAT_RESERVATION_CONFLICT:
 764		if (scmd->cmnd[0] == TEST_UNIT_READY)
 765			/* it is a success, we probed the device and
 766			 * found it */
 767			return SUCCESS;
 768		/* otherwise, we failed to send the command */
 769		return FAILED;
 770	case SAM_STAT_TASK_SET_FULL:
 771		scsi_handle_queue_full(scmd->device);
 772		fallthrough;
 773	case SAM_STAT_BUSY:
 774		return NEEDS_RETRY;
 775	default:
 776		return FAILED;
 777	}
 778	return FAILED;
 779}
 780
 781/**
 782 * scsi_eh_done - Completion function for error handling.
 783 * @scmd:	Cmd that is done.
 784 */
 785static void scsi_eh_done(struct scsi_cmnd *scmd)
 786{
 787	struct completion *eh_action;
 788
 789	SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
 790			"%s result: %x\n", __func__, scmd->result));
 
 791
 792	eh_action = scmd->device->host->eh_action;
 793	if (eh_action)
 794		complete(eh_action);
 795}
 796
 797/**
 798 * scsi_try_host_reset - ask host adapter to reset itself
 799 * @scmd:	SCSI cmd to send host reset.
 800 */
 801static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
 802{
 803	unsigned long flags;
 804	enum scsi_disposition rtn;
 805	struct Scsi_Host *host = scmd->device->host;
 806	struct scsi_host_template *hostt = host->hostt;
 807
 808	SCSI_LOG_ERROR_RECOVERY(3,
 809		shost_printk(KERN_INFO, host, "Snd Host RST\n"));
 810
 811	if (!hostt->eh_host_reset_handler)
 812		return FAILED;
 813
 814	rtn = hostt->eh_host_reset_handler(scmd);
 815
 816	if (rtn == SUCCESS) {
 817		if (!hostt->skip_settle_delay)
 818			ssleep(HOST_RESET_SETTLE_TIME);
 819		spin_lock_irqsave(host->host_lock, flags);
 820		scsi_report_bus_reset(host, scmd_channel(scmd));
 821		spin_unlock_irqrestore(host->host_lock, flags);
 822	}
 823
 824	return rtn;
 825}
 826
 827/**
 828 * scsi_try_bus_reset - ask host to perform a bus reset
 829 * @scmd:	SCSI cmd to send bus reset.
 830 */
 831static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
 832{
 833	unsigned long flags;
 834	enum scsi_disposition rtn;
 835	struct Scsi_Host *host = scmd->device->host;
 836	struct scsi_host_template *hostt = host->hostt;
 837
 838	SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
 839		"%s: Snd Bus RST\n", __func__));
 840
 841	if (!hostt->eh_bus_reset_handler)
 842		return FAILED;
 843
 844	rtn = hostt->eh_bus_reset_handler(scmd);
 845
 846	if (rtn == SUCCESS) {
 847		if (!hostt->skip_settle_delay)
 848			ssleep(BUS_RESET_SETTLE_TIME);
 849		spin_lock_irqsave(host->host_lock, flags);
 850		scsi_report_bus_reset(host, scmd_channel(scmd));
 851		spin_unlock_irqrestore(host->host_lock, flags);
 852	}
 853
 854	return rtn;
 855}
 856
 857static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
 858{
 859	sdev->was_reset = 1;
 860	sdev->expecting_cc_ua = 1;
 861}
 862
 863/**
 864 * scsi_try_target_reset - Ask host to perform a target reset
 865 * @scmd:	SCSI cmd used to send a target reset
 866 *
 867 * Notes:
 868 *    There is no timeout for this operation.  if this operation is
 869 *    unreliable for a given host, then the host itself needs to put a
 870 *    timer on it, and set the host back to a consistent state prior to
 871 *    returning.
 872 */
 873static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd)
 874{
 875	unsigned long flags;
 876	enum scsi_disposition rtn;
 877	struct Scsi_Host *host = scmd->device->host;
 878	struct scsi_host_template *hostt = host->hostt;
 879
 880	if (!hostt->eh_target_reset_handler)
 881		return FAILED;
 882
 883	rtn = hostt->eh_target_reset_handler(scmd);
 884	if (rtn == SUCCESS) {
 885		spin_lock_irqsave(host->host_lock, flags);
 886		__starget_for_each_device(scsi_target(scmd->device), NULL,
 887					  __scsi_report_device_reset);
 888		spin_unlock_irqrestore(host->host_lock, flags);
 889	}
 890
 891	return rtn;
 892}
 893
 894/**
 895 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
 896 * @scmd:	SCSI cmd used to send BDR
 897 *
 898 * Notes:
 899 *    There is no timeout for this operation.  if this operation is
 900 *    unreliable for a given host, then the host itself needs to put a
 901 *    timer on it, and set the host back to a consistent state prior to
 902 *    returning.
 903 */
 904static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
 905{
 906	enum scsi_disposition rtn;
 907	struct scsi_host_template *hostt = scmd->device->host->hostt;
 908
 909	if (!hostt->eh_device_reset_handler)
 910		return FAILED;
 911
 912	rtn = hostt->eh_device_reset_handler(scmd);
 913	if (rtn == SUCCESS)
 914		__scsi_report_device_reset(scmd->device, NULL);
 915	return rtn;
 916}
 917
 918/**
 919 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
 920 * @hostt:	SCSI driver host template
 921 * @scmd:	SCSI cmd used to send a target reset
 922 *
 923 * Return value:
 924 *	SUCCESS, FAILED, or FAST_IO_FAIL
 925 *
 926 * Notes:
 927 *    SUCCESS does not necessarily indicate that the command
 928 *    has been aborted; it only indicates that the LLDDs
 929 *    has cleared all references to that command.
 930 *    LLDDs should return FAILED only if an abort was required
 931 *    but could not be executed. LLDDs should return FAST_IO_FAIL
 932 *    if the device is temporarily unavailable (eg due to a
 933 *    link down on FibreChannel)
 934 */
 935static enum scsi_disposition
 936scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
 937{
 938	if (!hostt->eh_abort_handler)
 939		return FAILED;
 940
 941	return hostt->eh_abort_handler(scmd);
 942}
 943
 944static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
 945{
 946	if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
 947		if (scsi_try_bus_device_reset(scmd) != SUCCESS)
 948			if (scsi_try_target_reset(scmd) != SUCCESS)
 949				if (scsi_try_bus_reset(scmd) != SUCCESS)
 950					scsi_try_host_reset(scmd);
 951}
 952
 953/**
 954 * scsi_eh_prep_cmnd  - Save a scsi command info as part of error recovery
 955 * @scmd:       SCSI command structure to hijack
 956 * @ses:        structure to save restore information
 957 * @cmnd:       CDB to send. Can be NULL if no new cmnd is needed
 958 * @cmnd_size:  size in bytes of @cmnd (must be <= BLK_MAX_CDB)
 959 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
 960 *
 961 * This function is used to save a scsi command information before re-execution
 962 * as part of the error recovery process.  If @sense_bytes is 0 the command
 963 * sent must be one that does not transfer any data.  If @sense_bytes != 0
 964 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
 965 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
 966 */
 967void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
 968			unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
 969{
 970	struct scsi_device *sdev = scmd->device;
 971
 972	/*
 973	 * We need saved copies of a number of fields - this is because
 974	 * error handling may need to overwrite these with different values
 975	 * to run different commands, and once error handling is complete,
 976	 * we will need to restore these values prior to running the actual
 977	 * command.
 978	 */
 979	ses->cmd_len = scmd->cmd_len;
 980	ses->cmnd = scmd->cmnd;
 981	ses->data_direction = scmd->sc_data_direction;
 982	ses->sdb = scmd->sdb;
 
 983	ses->result = scmd->result;
 984	ses->resid_len = scmd->req.resid_len;
 985	ses->underflow = scmd->underflow;
 986	ses->prot_op = scmd->prot_op;
 987	ses->eh_eflags = scmd->eh_eflags;
 988
 989	scmd->prot_op = SCSI_PROT_NORMAL;
 990	scmd->eh_eflags = 0;
 991	scmd->cmnd = ses->eh_cmnd;
 992	memset(scmd->cmnd, 0, BLK_MAX_CDB);
 993	memset(&scmd->sdb, 0, sizeof(scmd->sdb));
 
 994	scmd->result = 0;
 995	scmd->req.resid_len = 0;
 996
 997	if (sense_bytes) {
 998		scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
 999					 sense_bytes);
1000		sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
1001			    scmd->sdb.length);
1002		scmd->sdb.table.sgl = &ses->sense_sgl;
1003		scmd->sc_data_direction = DMA_FROM_DEVICE;
1004		scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
1005		scmd->cmnd[0] = REQUEST_SENSE;
1006		scmd->cmnd[4] = scmd->sdb.length;
1007		scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1008	} else {
1009		scmd->sc_data_direction = DMA_NONE;
1010		if (cmnd) {
1011			BUG_ON(cmnd_size > BLK_MAX_CDB);
1012			memcpy(scmd->cmnd, cmnd, cmnd_size);
1013			scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1014		}
1015	}
1016
1017	scmd->underflow = 0;
1018
1019	if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1020		scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
1021			(sdev->lun << 5 & 0xe0);
1022
1023	/*
1024	 * Zero the sense buffer.  The scsi spec mandates that any
1025	 * untransferred sense data should be interpreted as being zero.
1026	 */
1027	memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1028}
1029EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1030
1031/**
1032 * scsi_eh_restore_cmnd  - Restore a scsi command info as part of error recovery
1033 * @scmd:       SCSI command structure to restore
1034 * @ses:        saved information from a coresponding call to scsi_eh_prep_cmnd
1035 *
1036 * Undo any damage done by above scsi_eh_prep_cmnd().
1037 */
1038void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1039{
1040	/*
1041	 * Restore original data
1042	 */
1043	scmd->cmd_len = ses->cmd_len;
1044	scmd->cmnd = ses->cmnd;
1045	scmd->sc_data_direction = ses->data_direction;
1046	scmd->sdb = ses->sdb;
 
1047	scmd->result = ses->result;
1048	scmd->req.resid_len = ses->resid_len;
1049	scmd->underflow = ses->underflow;
1050	scmd->prot_op = ses->prot_op;
1051	scmd->eh_eflags = ses->eh_eflags;
1052}
1053EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1054
1055/**
1056 * scsi_send_eh_cmnd  - submit a scsi command as part of error recovery
1057 * @scmd:       SCSI command structure to hijack
1058 * @cmnd:       CDB to send
1059 * @cmnd_size:  size in bytes of @cmnd
1060 * @timeout:    timeout for this request
1061 * @sense_bytes: size of sense data to copy or 0
1062 *
1063 * This function is used to send a scsi command down to a target device
1064 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1065 *
1066 * Return value:
1067 *    SUCCESS or FAILED or NEEDS_RETRY
1068 */
1069static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
1070	unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes)
1071{
1072	struct scsi_device *sdev = scmd->device;
1073	struct Scsi_Host *shost = sdev->host;
1074	DECLARE_COMPLETION_ONSTACK(done);
1075	unsigned long timeleft = timeout, delay;
1076	struct scsi_eh_save ses;
1077	const unsigned long stall_for = msecs_to_jiffies(100);
1078	int rtn;
1079
1080retry:
1081	scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1082	shost->eh_action = &done;
1083
1084	scsi_log_send(scmd);
1085	scmd->scsi_done = scsi_eh_done;
1086
1087	/*
1088	 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1089	 * change the SCSI device state after we have examined it and before
1090	 * .queuecommand() is called.
1091	 */
1092	mutex_lock(&sdev->state_mutex);
1093	while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1094		mutex_unlock(&sdev->state_mutex);
1095		SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1096			"%s: state %d <> %d\n", __func__, sdev->sdev_state,
1097			SDEV_BLOCK));
1098		delay = min(timeleft, stall_for);
1099		timeleft -= delay;
1100		msleep(jiffies_to_msecs(delay));
1101		mutex_lock(&sdev->state_mutex);
1102	}
1103	if (sdev->sdev_state != SDEV_BLOCK)
1104		rtn = shost->hostt->queuecommand(shost, scmd);
1105	else
1106		rtn = FAILED;
1107	mutex_unlock(&sdev->state_mutex);
1108
1109	if (rtn) {
1110		if (timeleft > stall_for) {
1111			scsi_eh_restore_cmnd(scmd, &ses);
1112			timeleft -= stall_for;
1113			msleep(jiffies_to_msecs(stall_for));
1114			goto retry;
1115		}
1116		/* signal not to enter either branch of the if () below */
1117		timeleft = 0;
1118		rtn = FAILED;
1119	} else {
1120		timeleft = wait_for_completion_timeout(&done, timeout);
1121		rtn = SUCCESS;
1122	}
1123
1124	shost->eh_action = NULL;
1125
1126	scsi_log_completion(scmd, rtn);
1127
1128	SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1129			"%s timeleft: %ld\n",
1130			__func__, timeleft));
1131
1132	/*
1133	 * If there is time left scsi_eh_done got called, and we will examine
1134	 * the actual status codes to see whether the command actually did
1135	 * complete normally, else if we have a zero return and no time left,
1136	 * the command must still be pending, so abort it and return FAILED.
1137	 * If we never actually managed to issue the command, because
1138	 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1139	 * value above (so don't execute either branch of the if)
1140	 */
1141	if (timeleft) {
1142		rtn = scsi_eh_completed_normally(scmd);
1143		SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1144			"%s: scsi_eh_completed_normally %x\n", __func__, rtn));
 
1145
1146		switch (rtn) {
1147		case SUCCESS:
1148		case NEEDS_RETRY:
1149		case FAILED:
1150			break;
1151		case ADD_TO_MLQUEUE:
1152			rtn = NEEDS_RETRY;
1153			break;
1154		default:
1155			rtn = FAILED;
1156			break;
1157		}
1158	} else if (rtn != FAILED) {
1159		scsi_abort_eh_cmnd(scmd);
1160		rtn = FAILED;
1161	}
1162
1163	scsi_eh_restore_cmnd(scmd, &ses);
1164
1165	return rtn;
1166}
1167
1168/**
1169 * scsi_request_sense - Request sense data from a particular target.
1170 * @scmd:	SCSI cmd for request sense.
1171 *
1172 * Notes:
1173 *    Some hosts automatically obtain this information, others require
1174 *    that we obtain it on our own. This function will *not* return until
1175 *    the command either times out, or it completes.
1176 */
1177static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd)
1178{
1179	return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1180}
1181
1182static enum scsi_disposition
1183scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn)
1184{
1185	if (!blk_rq_is_passthrough(scmd->request)) {
1186		struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1187		if (sdrv->eh_action)
1188			rtn = sdrv->eh_action(scmd, rtn);
1189	}
1190	return rtn;
1191}
1192
1193/**
1194 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1195 * @scmd:	Original SCSI cmd that eh has finished.
1196 * @done_q:	Queue for processed commands.
1197 *
1198 * Notes:
1199 *    We don't want to use the normal command completion while we are are
1200 *    still handling errors - it may cause other commands to be queued,
1201 *    and that would disturb what we are doing.  Thus we really want to
1202 *    keep a list of pending commands for final completion, and once we
1203 *    are ready to leave error handling we handle completion for real.
1204 */
1205void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1206{
 
 
1207	list_move_tail(&scmd->eh_entry, done_q);
1208}
1209EXPORT_SYMBOL(scsi_eh_finish_cmd);
1210
1211/**
1212 * scsi_eh_get_sense - Get device sense data.
1213 * @work_q:	Queue of commands to process.
1214 * @done_q:	Queue of processed commands.
1215 *
1216 * Description:
1217 *    See if we need to request sense information.  if so, then get it
1218 *    now, so we have a better idea of what to do.
1219 *
1220 * Notes:
1221 *    This has the unfortunate side effect that if a shost adapter does
1222 *    not automatically request sense information, we end up shutting
1223 *    it down before we request it.
1224 *
1225 *    All drivers should request sense information internally these days,
1226 *    so for now all I have to say is tough noogies if you end up in here.
1227 *
1228 *    XXX: Long term this code should go away, but that needs an audit of
1229 *         all LLDDs first.
1230 */
1231int scsi_eh_get_sense(struct list_head *work_q,
1232		      struct list_head *done_q)
1233{
1234	struct scsi_cmnd *scmd, *next;
1235	struct Scsi_Host *shost;
1236	enum scsi_disposition rtn;
1237
1238	/*
1239	 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1240	 * should not get sense.
1241	 */
1242	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1243		if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1244		    SCSI_SENSE_VALID(scmd))
1245			continue;
1246
1247		shost = scmd->device->host;
1248		if (scsi_host_eh_past_deadline(shost)) {
1249			SCSI_LOG_ERROR_RECOVERY(3,
1250				scmd_printk(KERN_INFO, scmd,
1251					    "%s: skip request sense, past eh deadline\n",
1252					     current->comm));
1253			break;
1254		}
1255		if (!scsi_status_is_check_condition(scmd->result))
1256			/*
1257			 * don't request sense if there's no check condition
1258			 * status because the error we're processing isn't one
1259			 * that has a sense code (and some devices get
1260			 * confused by sense requests out of the blue)
1261			 */
1262			continue;
1263
1264		SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1265						  "%s: requesting sense\n",
1266						  current->comm));
1267		rtn = scsi_request_sense(scmd);
1268		if (rtn != SUCCESS)
1269			continue;
1270
1271		SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1272			"sense requested, result %x\n", scmd->result));
1273		SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
 
1274
1275		rtn = scsi_decide_disposition(scmd);
1276
1277		/*
1278		 * if the result was normal, then just pass it along to the
1279		 * upper level.
1280		 */
1281		if (rtn == SUCCESS)
1282			/*
1283			 * We don't want this command reissued, just finished
1284			 * with the sense data, so set retries to the max
1285			 * allowed to ensure it won't get reissued. If the user
1286			 * has requested infinite retries, we also want to
1287			 * finish this command, so force completion by setting
1288			 * retries and allowed to the same value.
1289			 */
1290			if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
1291				scmd->retries = scmd->allowed = 1;
1292			else
1293				scmd->retries = scmd->allowed;
1294		else if (rtn != NEEDS_RETRY)
1295			continue;
1296
1297		scsi_eh_finish_cmd(scmd, done_q);
1298	}
1299
1300	return list_empty(work_q);
1301}
1302EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1303
1304/**
1305 * scsi_eh_tur - Send TUR to device.
1306 * @scmd:	&scsi_cmnd to send TUR
1307 *
1308 * Return value:
1309 *    0 - Device is ready. 1 - Device NOT ready.
1310 */
1311static int scsi_eh_tur(struct scsi_cmnd *scmd)
1312{
1313	static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1314	int retry_cnt = 1;
1315	enum scsi_disposition rtn;
1316
1317retry_tur:
1318	rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1319				scmd->device->eh_timeout, 0);
1320
1321	SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1322		"%s return: %x\n", __func__, rtn));
1323
1324	switch (rtn) {
1325	case NEEDS_RETRY:
1326		if (retry_cnt--)
1327			goto retry_tur;
1328		fallthrough;
1329	case SUCCESS:
1330		return 0;
1331	default:
1332		return 1;
1333	}
1334}
1335
1336/**
1337 * scsi_eh_test_devices - check if devices are responding from error recovery.
1338 * @cmd_list:	scsi commands in error recovery.
1339 * @work_q:	queue for commands which still need more error recovery
1340 * @done_q:	queue for commands which are finished
1341 * @try_stu:	boolean on if a STU command should be tried in addition to TUR.
1342 *
1343 * Decription:
1344 *    Tests if devices are in a working state.  Commands to devices now in
1345 *    a working state are sent to the done_q while commands to devices which
1346 *    are still failing to respond are returned to the work_q for more
1347 *    processing.
1348 **/
1349static int scsi_eh_test_devices(struct list_head *cmd_list,
1350				struct list_head *work_q,
1351				struct list_head *done_q, int try_stu)
1352{
1353	struct scsi_cmnd *scmd, *next;
1354	struct scsi_device *sdev;
1355	int finish_cmds;
1356
1357	while (!list_empty(cmd_list)) {
1358		scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1359		sdev = scmd->device;
1360
1361		if (!try_stu) {
1362			if (scsi_host_eh_past_deadline(sdev->host)) {
1363				/* Push items back onto work_q */
1364				list_splice_init(cmd_list, work_q);
1365				SCSI_LOG_ERROR_RECOVERY(3,
1366					sdev_printk(KERN_INFO, sdev,
1367						    "%s: skip test device, past eh deadline",
1368						    current->comm));
1369				break;
1370			}
1371		}
1372
1373		finish_cmds = !scsi_device_online(scmd->device) ||
1374			(try_stu && !scsi_eh_try_stu(scmd) &&
1375			 !scsi_eh_tur(scmd)) ||
1376			!scsi_eh_tur(scmd);
1377
1378		list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1379			if (scmd->device == sdev) {
1380				if (finish_cmds &&
1381				    (try_stu ||
1382				     scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1383					scsi_eh_finish_cmd(scmd, done_q);
1384				else
1385					list_move_tail(&scmd->eh_entry, work_q);
1386			}
1387	}
1388	return list_empty(work_q);
1389}
1390
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1391/**
1392 * scsi_eh_try_stu - Send START_UNIT to device.
1393 * @scmd:	&scsi_cmnd to send START_UNIT
1394 *
1395 * Return value:
1396 *    0 - Device is ready. 1 - Device NOT ready.
1397 */
1398static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1399{
1400	static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1401
1402	if (scmd->device->allow_restart) {
1403		int i;
1404		enum scsi_disposition rtn = NEEDS_RETRY;
1405
1406		for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1407			rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1408
1409		if (rtn == SUCCESS)
1410			return 0;
1411	}
1412
1413	return 1;
1414}
1415
1416 /**
1417 * scsi_eh_stu - send START_UNIT if needed
1418 * @shost:	&scsi host being recovered.
1419 * @work_q:	&list_head for pending commands.
1420 * @done_q:	&list_head for processed commands.
1421 *
1422 * Notes:
1423 *    If commands are failing due to not ready, initializing command required,
1424 *	try revalidating the device, which will end up sending a start unit.
1425 */
1426static int scsi_eh_stu(struct Scsi_Host *shost,
1427			      struct list_head *work_q,
1428			      struct list_head *done_q)
1429{
1430	struct scsi_cmnd *scmd, *stu_scmd, *next;
1431	struct scsi_device *sdev;
1432
1433	shost_for_each_device(sdev, shost) {
1434		if (scsi_host_eh_past_deadline(shost)) {
1435			SCSI_LOG_ERROR_RECOVERY(3,
1436				sdev_printk(KERN_INFO, sdev,
1437					    "%s: skip START_UNIT, past eh deadline\n",
1438					    current->comm));
1439			scsi_device_put(sdev);
1440			break;
1441		}
1442		stu_scmd = NULL;
1443		list_for_each_entry(scmd, work_q, eh_entry)
1444			if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1445			    scsi_check_sense(scmd) == FAILED ) {
1446				stu_scmd = scmd;
1447				break;
1448			}
1449
1450		if (!stu_scmd)
1451			continue;
1452
1453		SCSI_LOG_ERROR_RECOVERY(3,
1454			sdev_printk(KERN_INFO, sdev,
1455				     "%s: Sending START_UNIT\n",
1456				    current->comm));
1457
1458		if (!scsi_eh_try_stu(stu_scmd)) {
1459			if (!scsi_device_online(sdev) ||
1460			    !scsi_eh_tur(stu_scmd)) {
1461				list_for_each_entry_safe(scmd, next,
1462							  work_q, eh_entry) {
1463					if (scmd->device == sdev &&
1464					    scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1465						scsi_eh_finish_cmd(scmd, done_q);
1466				}
1467			}
1468		} else {
1469			SCSI_LOG_ERROR_RECOVERY(3,
1470				sdev_printk(KERN_INFO, sdev,
1471					    "%s: START_UNIT failed\n",
1472					    current->comm));
1473		}
1474	}
1475
1476	return list_empty(work_q);
1477}
1478
1479
1480/**
1481 * scsi_eh_bus_device_reset - send bdr if needed
1482 * @shost:	scsi host being recovered.
1483 * @work_q:	&list_head for pending commands.
1484 * @done_q:	&list_head for processed commands.
1485 *
1486 * Notes:
1487 *    Try a bus device reset.  Still, look to see whether we have multiple
1488 *    devices that are jammed or not - if we have multiple devices, it
1489 *    makes no sense to try bus_device_reset - we really would need to try
1490 *    a bus_reset instead.
1491 */
1492static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1493				    struct list_head *work_q,
1494				    struct list_head *done_q)
1495{
1496	struct scsi_cmnd *scmd, *bdr_scmd, *next;
1497	struct scsi_device *sdev;
1498	enum scsi_disposition rtn;
1499
1500	shost_for_each_device(sdev, shost) {
1501		if (scsi_host_eh_past_deadline(shost)) {
1502			SCSI_LOG_ERROR_RECOVERY(3,
1503				sdev_printk(KERN_INFO, sdev,
1504					    "%s: skip BDR, past eh deadline\n",
1505					     current->comm));
1506			scsi_device_put(sdev);
1507			break;
1508		}
1509		bdr_scmd = NULL;
1510		list_for_each_entry(scmd, work_q, eh_entry)
1511			if (scmd->device == sdev) {
1512				bdr_scmd = scmd;
1513				break;
1514			}
1515
1516		if (!bdr_scmd)
1517			continue;
1518
1519		SCSI_LOG_ERROR_RECOVERY(3,
1520			sdev_printk(KERN_INFO, sdev,
1521				     "%s: Sending BDR\n", current->comm));
1522		rtn = scsi_try_bus_device_reset(bdr_scmd);
1523		if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1524			if (!scsi_device_online(sdev) ||
1525			    rtn == FAST_IO_FAIL ||
1526			    !scsi_eh_tur(bdr_scmd)) {
1527				list_for_each_entry_safe(scmd, next,
1528							 work_q, eh_entry) {
1529					if (scmd->device == sdev &&
1530					    scsi_eh_action(scmd, rtn) != FAILED)
1531						scsi_eh_finish_cmd(scmd,
1532								   done_q);
1533				}
1534			}
1535		} else {
1536			SCSI_LOG_ERROR_RECOVERY(3,
1537				sdev_printk(KERN_INFO, sdev,
1538					    "%s: BDR failed\n", current->comm));
 
 
1539		}
1540	}
1541
1542	return list_empty(work_q);
1543}
1544
1545/**
1546 * scsi_eh_target_reset - send target reset if needed
1547 * @shost:	scsi host being recovered.
1548 * @work_q:	&list_head for pending commands.
1549 * @done_q:	&list_head for processed commands.
1550 *
1551 * Notes:
1552 *    Try a target reset.
1553 */
1554static int scsi_eh_target_reset(struct Scsi_Host *shost,
1555				struct list_head *work_q,
1556				struct list_head *done_q)
1557{
1558	LIST_HEAD(tmp_list);
1559	LIST_HEAD(check_list);
1560
1561	list_splice_init(work_q, &tmp_list);
1562
1563	while (!list_empty(&tmp_list)) {
1564		struct scsi_cmnd *next, *scmd;
1565		enum scsi_disposition rtn;
1566		unsigned int id;
1567
1568		if (scsi_host_eh_past_deadline(shost)) {
1569			/* push back on work queue for further processing */
1570			list_splice_init(&check_list, work_q);
1571			list_splice_init(&tmp_list, work_q);
1572			SCSI_LOG_ERROR_RECOVERY(3,
1573				shost_printk(KERN_INFO, shost,
1574					    "%s: Skip target reset, past eh deadline\n",
1575					     current->comm));
1576			return list_empty(work_q);
1577		}
1578
1579		scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1580		id = scmd_id(scmd);
1581
1582		SCSI_LOG_ERROR_RECOVERY(3,
1583			shost_printk(KERN_INFO, shost,
1584				     "%s: Sending target reset to target %d\n",
1585				     current->comm, id));
1586		rtn = scsi_try_target_reset(scmd);
1587		if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1588			SCSI_LOG_ERROR_RECOVERY(3,
1589				shost_printk(KERN_INFO, shost,
1590					     "%s: Target reset failed"
1591					     " target: %d\n",
1592					     current->comm, id));
1593		list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1594			if (scmd_id(scmd) != id)
1595				continue;
1596
1597			if (rtn == SUCCESS)
1598				list_move_tail(&scmd->eh_entry, &check_list);
1599			else if (rtn == FAST_IO_FAIL)
1600				scsi_eh_finish_cmd(scmd, done_q);
1601			else
1602				/* push back on work queue for further processing */
1603				list_move(&scmd->eh_entry, work_q);
1604		}
1605	}
1606
1607	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1608}
1609
1610/**
1611 * scsi_eh_bus_reset - send a bus reset
1612 * @shost:	&scsi host being recovered.
1613 * @work_q:	&list_head for pending commands.
1614 * @done_q:	&list_head for processed commands.
1615 */
1616static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1617			     struct list_head *work_q,
1618			     struct list_head *done_q)
1619{
1620	struct scsi_cmnd *scmd, *chan_scmd, *next;
1621	LIST_HEAD(check_list);
1622	unsigned int channel;
1623	enum scsi_disposition rtn;
1624
1625	/*
1626	 * we really want to loop over the various channels, and do this on
1627	 * a channel by channel basis.  we should also check to see if any
1628	 * of the failed commands are on soft_reset devices, and if so, skip
1629	 * the reset.
1630	 */
1631
1632	for (channel = 0; channel <= shost->max_channel; channel++) {
1633		if (scsi_host_eh_past_deadline(shost)) {
1634			list_splice_init(&check_list, work_q);
1635			SCSI_LOG_ERROR_RECOVERY(3,
1636				shost_printk(KERN_INFO, shost,
1637					    "%s: skip BRST, past eh deadline\n",
1638					     current->comm));
1639			return list_empty(work_q);
1640		}
1641
1642		chan_scmd = NULL;
1643		list_for_each_entry(scmd, work_q, eh_entry) {
1644			if (channel == scmd_channel(scmd)) {
1645				chan_scmd = scmd;
1646				break;
1647				/*
1648				 * FIXME add back in some support for
1649				 * soft_reset devices.
1650				 */
1651			}
1652		}
1653
1654		if (!chan_scmd)
1655			continue;
1656		SCSI_LOG_ERROR_RECOVERY(3,
1657			shost_printk(KERN_INFO, shost,
1658				     "%s: Sending BRST chan: %d\n",
1659				     current->comm, channel));
1660		rtn = scsi_try_bus_reset(chan_scmd);
1661		if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1662			list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1663				if (channel == scmd_channel(scmd)) {
1664					if (rtn == FAST_IO_FAIL)
1665						scsi_eh_finish_cmd(scmd,
1666								   done_q);
1667					else
1668						list_move_tail(&scmd->eh_entry,
1669							       &check_list);
1670				}
1671			}
1672		} else {
1673			SCSI_LOG_ERROR_RECOVERY(3,
1674				shost_printk(KERN_INFO, shost,
1675					     "%s: BRST failed chan: %d\n",
1676					     current->comm, channel));
1677		}
1678	}
1679	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1680}
1681
1682/**
1683 * scsi_eh_host_reset - send a host reset
1684 * @shost:	host to be reset.
1685 * @work_q:	&list_head for pending commands.
1686 * @done_q:	&list_head for processed commands.
1687 */
1688static int scsi_eh_host_reset(struct Scsi_Host *shost,
1689			      struct list_head *work_q,
1690			      struct list_head *done_q)
1691{
1692	struct scsi_cmnd *scmd, *next;
1693	LIST_HEAD(check_list);
1694	enum scsi_disposition rtn;
1695
1696	if (!list_empty(work_q)) {
1697		scmd = list_entry(work_q->next,
1698				  struct scsi_cmnd, eh_entry);
1699
1700		SCSI_LOG_ERROR_RECOVERY(3,
1701			shost_printk(KERN_INFO, shost,
1702				     "%s: Sending HRST\n",
1703				     current->comm));
1704
1705		rtn = scsi_try_host_reset(scmd);
1706		if (rtn == SUCCESS) {
1707			list_splice_init(work_q, &check_list);
1708		} else if (rtn == FAST_IO_FAIL) {
1709			list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1710					scsi_eh_finish_cmd(scmd, done_q);
1711			}
1712		} else {
1713			SCSI_LOG_ERROR_RECOVERY(3,
1714				shost_printk(KERN_INFO, shost,
1715					     "%s: HRST failed\n",
1716					     current->comm));
1717		}
1718	}
1719	return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1720}
1721
1722/**
1723 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1724 * @work_q:	&list_head for pending commands.
1725 * @done_q:	&list_head for processed commands.
1726 */
1727static void scsi_eh_offline_sdevs(struct list_head *work_q,
1728				  struct list_head *done_q)
1729{
1730	struct scsi_cmnd *scmd, *next;
1731	struct scsi_device *sdev;
1732
1733	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1734		sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1735			    "not ready after error recovery\n");
1736		sdev = scmd->device;
1737
1738		mutex_lock(&sdev->state_mutex);
1739		scsi_device_set_state(sdev, SDEV_OFFLINE);
1740		mutex_unlock(&sdev->state_mutex);
1741
1742		scsi_eh_finish_cmd(scmd, done_q);
1743	}
1744	return;
1745}
1746
1747/**
1748 * scsi_noretry_cmd - determine if command should be failed fast
1749 * @scmd:	SCSI cmd to examine.
1750 */
1751int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1752{
1753	switch (host_byte(scmd->result)) {
1754	case DID_OK:
1755		break;
1756	case DID_TIME_OUT:
1757		goto check_type;
1758	case DID_BUS_BUSY:
1759		return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1760	case DID_PARITY:
1761		return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1762	case DID_ERROR:
1763		if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
 
1764			return 0;
1765		fallthrough;
1766	case DID_SOFT_ERROR:
1767		return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1768	}
1769
1770	if (!scsi_status_is_check_condition(scmd->result))
1771		return 0;
1772
1773check_type:
1774	/*
1775	 * assume caller has checked sense and determined
1776	 * the check condition was retryable.
1777	 */
1778	if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1779	    blk_rq_is_passthrough(scmd->request))
1780		return 1;
1781
1782	return 0;
1783}
1784
1785/**
1786 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1787 * @scmd:	SCSI cmd to examine.
1788 *
1789 * Notes:
1790 *    This is *only* called when we are examining the status after sending
1791 *    out the actual data command.  any commands that are queued for error
1792 *    recovery (e.g. test_unit_ready) do *not* come through here.
1793 *
1794 *    When this routine returns failed, it means the error handler thread
1795 *    is woken.  In cases where the error code indicates an error that
1796 *    doesn't require the error handler read (i.e. we don't need to
1797 *    abort/reset), this function should return SUCCESS.
1798 */
1799enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd)
1800{
1801	enum scsi_disposition rtn;
1802
1803	/*
1804	 * if the device is offline, then we clearly just pass the result back
1805	 * up to the top level.
1806	 */
1807	if (!scsi_device_online(scmd->device)) {
1808		SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1809			"%s: device offline - report as SUCCESS\n", __func__));
 
1810		return SUCCESS;
1811	}
1812
1813	/*
1814	 * first check the host byte, to see if there is anything in there
1815	 * that would indicate what we need to do.
1816	 */
1817	switch (host_byte(scmd->result)) {
1818	case DID_PASSTHROUGH:
1819		/*
1820		 * no matter what, pass this through to the upper layer.
1821		 * nuke this special code so that it looks like we are saying
1822		 * did_ok.
1823		 */
1824		scmd->result &= 0xff00ffff;
1825		return SUCCESS;
1826	case DID_OK:
1827		/*
1828		 * looks good.  drop through, and check the next byte.
1829		 */
1830		break;
1831	case DID_ABORT:
1832		if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1833			set_host_byte(scmd, DID_TIME_OUT);
1834			return SUCCESS;
1835		}
1836		fallthrough;
1837	case DID_NO_CONNECT:
1838	case DID_BAD_TARGET:
1839		/*
1840		 * note - this means that we just report the status back
1841		 * to the top level driver, not that we actually think
1842		 * that it indicates SUCCESS.
1843		 */
1844		return SUCCESS;
1845	case DID_SOFT_ERROR:
1846		/*
1847		 * when the low level driver returns did_soft_error,
1848		 * it is responsible for keeping an internal retry counter
1849		 * in order to avoid endless loops (db)
 
 
 
 
1850		 */
 
1851		goto maybe_retry;
1852	case DID_IMM_RETRY:
1853		return NEEDS_RETRY;
1854
1855	case DID_REQUEUE:
1856		return ADD_TO_MLQUEUE;
1857	case DID_TRANSPORT_DISRUPTED:
1858		/*
1859		 * LLD/transport was disrupted during processing of the IO.
1860		 * The transport class is now blocked/blocking,
1861		 * and the transport will decide what to do with the IO
1862		 * based on its timers and recovery capablilities if
1863		 * there are enough retries.
1864		 */
1865		goto maybe_retry;
1866	case DID_TRANSPORT_FAILFAST:
1867		/*
1868		 * The transport decided to failfast the IO (most likely
1869		 * the fast io fail tmo fired), so send IO directly upwards.
1870		 */
1871		return SUCCESS;
1872	case DID_TRANSPORT_MARGINAL:
1873		/*
1874		 * caller has decided not to do retries on
1875		 * abort success, so send IO directly upwards
1876		 */
1877		return SUCCESS;
1878	case DID_ERROR:
1879		if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
 
1880			/*
1881			 * execute reservation conflict processing code
1882			 * lower down
1883			 */
1884			break;
1885		fallthrough;
1886	case DID_BUS_BUSY:
1887	case DID_PARITY:
1888		goto maybe_retry;
1889	case DID_TIME_OUT:
1890		/*
1891		 * when we scan the bus, we get timeout messages for
1892		 * these commands if there is no device available.
1893		 * other hosts report did_no_connect for the same thing.
1894		 */
1895		if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1896		     scmd->cmnd[0] == INQUIRY)) {
1897			return SUCCESS;
1898		} else {
1899			return FAILED;
1900		}
1901	case DID_RESET:
1902		return SUCCESS;
1903	default:
1904		return FAILED;
1905	}
1906
1907	/*
 
 
 
 
 
 
1908	 * check the status byte to see if this indicates anything special.
1909	 */
1910	switch (get_status_byte(scmd)) {
1911	case SAM_STAT_TASK_SET_FULL:
1912		scsi_handle_queue_full(scmd->device);
1913		/*
1914		 * the case of trying to send too many commands to a
1915		 * tagged queueing device.
1916		 */
1917		fallthrough;
1918	case SAM_STAT_BUSY:
1919		/*
1920		 * device can't talk to us at the moment.  Should only
1921		 * occur (SAM-3) when the task queue is empty, so will cause
1922		 * the empty queue handling to trigger a stall in the
1923		 * device.
1924		 */
1925		return ADD_TO_MLQUEUE;
1926	case SAM_STAT_GOOD:
1927		if (scmd->cmnd[0] == REPORT_LUNS)
1928			scmd->device->sdev_target->expecting_lun_change = 0;
1929		scsi_handle_queue_ramp_up(scmd->device);
1930		fallthrough;
1931	case SAM_STAT_COMMAND_TERMINATED:
1932		return SUCCESS;
1933	case SAM_STAT_TASK_ABORTED:
1934		goto maybe_retry;
1935	case SAM_STAT_CHECK_CONDITION:
1936		rtn = scsi_check_sense(scmd);
1937		if (rtn == NEEDS_RETRY)
1938			goto maybe_retry;
1939		/* if rtn == FAILED, we have no sense information;
1940		 * returning FAILED will wake the error handler thread
1941		 * to collect the sense and redo the decide
1942		 * disposition */
1943		return rtn;
1944	case SAM_STAT_CONDITION_MET:
1945	case SAM_STAT_INTERMEDIATE:
1946	case SAM_STAT_INTERMEDIATE_CONDITION_MET:
1947	case SAM_STAT_ACA_ACTIVE:
1948		/*
1949		 * who knows?  FIXME(eric)
1950		 */
1951		return SUCCESS;
1952
1953	case SAM_STAT_RESERVATION_CONFLICT:
1954		sdev_printk(KERN_INFO, scmd->device,
1955			    "reservation conflict\n");
1956		set_host_byte(scmd, DID_NEXUS_FAILURE);
1957		return SUCCESS; /* causes immediate i/o error */
1958	default:
1959		return FAILED;
1960	}
1961	return FAILED;
1962
1963maybe_retry:
1964
1965	/* we requeue for retry because the error was retryable, and
1966	 * the request was not marked fast fail.  Note that above,
1967	 * even if the request is marked fast fail, we still requeue
1968	 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1969	if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) {
 
1970		return NEEDS_RETRY;
1971	} else {
1972		/*
1973		 * no more retries - report this one back to upper level.
1974		 */
1975		return SUCCESS;
1976	}
1977}
1978
1979static void eh_lock_door_done(struct request *req, blk_status_t status)
1980{
1981	blk_put_request(req);
1982}
1983
1984/**
1985 * scsi_eh_lock_door - Prevent medium removal for the specified device
1986 * @sdev:	SCSI device to prevent medium removal
1987 *
1988 * Locking:
1989 * 	We must be called from process context.
1990 *
1991 * Notes:
1992 * 	We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1993 * 	head of the devices request queue, and continue.
1994 */
1995static void scsi_eh_lock_door(struct scsi_device *sdev)
1996{
1997	struct request *req;
1998	struct scsi_request *rq;
1999
2000	req = blk_get_request(sdev->request_queue, REQ_OP_DRV_IN, 0);
2001	if (IS_ERR(req))
2002		return;
2003	rq = scsi_req(req);
 
 
 
 
 
 
 
 
2004
2005	rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
2006	rq->cmd[1] = 0;
2007	rq->cmd[2] = 0;
2008	rq->cmd[3] = 0;
2009	rq->cmd[4] = SCSI_REMOVAL_PREVENT;
2010	rq->cmd[5] = 0;
2011	rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
2012
2013	req->rq_flags |= RQF_QUIET;
 
2014	req->timeout = 10 * HZ;
2015	rq->retries = 5;
2016
2017	blk_execute_rq_nowait(NULL, req, 1, eh_lock_door_done);
2018}
2019
2020/**
2021 * scsi_restart_operations - restart io operations to the specified host.
2022 * @shost:	Host we are restarting.
2023 *
2024 * Notes:
2025 *    When we entered the error handler, we blocked all further i/o to
2026 *    this device.  we need to 'reverse' this process.
2027 */
2028static void scsi_restart_operations(struct Scsi_Host *shost)
2029{
2030	struct scsi_device *sdev;
2031	unsigned long flags;
2032
2033	/*
2034	 * If the door was locked, we need to insert a door lock request
2035	 * onto the head of the SCSI request queue for the device.  There
2036	 * is no point trying to lock the door of an off-line device.
2037	 */
2038	shost_for_each_device(sdev, shost) {
2039		if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2040			scsi_eh_lock_door(sdev);
2041			sdev->was_reset = 0;
2042		}
2043	}
2044
2045	/*
2046	 * next free up anything directly waiting upon the host.  this
2047	 * will be requests for character device operations, and also for
2048	 * ioctls to queued block devices.
2049	 */
2050	SCSI_LOG_ERROR_RECOVERY(3,
2051		shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
 
2052
2053	spin_lock_irqsave(shost->host_lock, flags);
2054	if (scsi_host_set_state(shost, SHOST_RUNNING))
2055		if (scsi_host_set_state(shost, SHOST_CANCEL))
2056			BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2057	spin_unlock_irqrestore(shost->host_lock, flags);
2058
2059	wake_up(&shost->host_wait);
2060
2061	/*
2062	 * finally we need to re-initiate requests that may be pending.  we will
2063	 * have had everything blocked while error handling is taking place, and
2064	 * now that error recovery is done, we will need to ensure that these
2065	 * requests are started.
2066	 */
2067	scsi_run_host_queues(shost);
2068
2069	/*
2070	 * if eh is active and host_eh_scheduled is pending we need to re-run
2071	 * recovery.  we do this check after scsi_run_host_queues() to allow
2072	 * everything pent up since the last eh run a chance to make forward
2073	 * progress before we sync again.  Either we'll immediately re-run
2074	 * recovery or scsi_device_unbusy() will wake us again when these
2075	 * pending commands complete.
2076	 */
2077	spin_lock_irqsave(shost->host_lock, flags);
2078	if (shost->host_eh_scheduled)
2079		if (scsi_host_set_state(shost, SHOST_RECOVERY))
2080			WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2081	spin_unlock_irqrestore(shost->host_lock, flags);
2082}
2083
2084/**
2085 * scsi_eh_ready_devs - check device ready state and recover if not.
2086 * @shost:	host to be recovered.
2087 * @work_q:	&list_head for pending commands.
2088 * @done_q:	&list_head for processed commands.
2089 */
2090void scsi_eh_ready_devs(struct Scsi_Host *shost,
2091			struct list_head *work_q,
2092			struct list_head *done_q)
2093{
2094	if (!scsi_eh_stu(shost, work_q, done_q))
2095		if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2096			if (!scsi_eh_target_reset(shost, work_q, done_q))
2097				if (!scsi_eh_bus_reset(shost, work_q, done_q))
2098					if (!scsi_eh_host_reset(shost, work_q, done_q))
2099						scsi_eh_offline_sdevs(work_q,
2100								      done_q);
2101}
2102EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2103
2104/**
2105 * scsi_eh_flush_done_q - finish processed commands or retry them.
2106 * @done_q:	list_head of processed commands.
2107 */
2108void scsi_eh_flush_done_q(struct list_head *done_q)
2109{
2110	struct scsi_cmnd *scmd, *next;
2111
2112	list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2113		list_del_init(&scmd->eh_entry);
2114		if (scsi_device_online(scmd->device) &&
2115		    !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd) &&
2116			scsi_eh_should_retry_cmd(scmd)) {
2117			SCSI_LOG_ERROR_RECOVERY(3,
2118				scmd_printk(KERN_INFO, scmd,
2119					     "%s: flush retry cmd\n",
2120					     current->comm));
2121				scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2122		} else {
2123			/*
2124			 * If just we got sense for the device (called
2125			 * scsi_eh_get_sense), scmd->result is already
2126			 * set, do not set DID_TIME_OUT.
2127			 */
2128			if (!scmd->result)
2129				scmd->result |= (DID_TIME_OUT << 16);
2130			SCSI_LOG_ERROR_RECOVERY(3,
2131				scmd_printk(KERN_INFO, scmd,
2132					     "%s: flush finish cmd\n",
2133					     current->comm));
2134			scsi_finish_command(scmd);
2135		}
2136	}
2137}
2138EXPORT_SYMBOL(scsi_eh_flush_done_q);
2139
2140/**
2141 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2142 * @shost:	Host to unjam.
2143 *
2144 * Notes:
2145 *    When we come in here, we *know* that all commands on the bus have
2146 *    either completed, failed or timed out.  we also know that no further
2147 *    commands are being sent to the host, so things are relatively quiet
2148 *    and we have freedom to fiddle with things as we wish.
2149 *
2150 *    This is only the *default* implementation.  it is possible for
2151 *    individual drivers to supply their own version of this function, and
2152 *    if the maintainer wishes to do this, it is strongly suggested that
2153 *    this function be taken as a template and modified.  this function
2154 *    was designed to correctly handle problems for about 95% of the
2155 *    different cases out there, and it should always provide at least a
2156 *    reasonable amount of error recovery.
2157 *
2158 *    Any command marked 'failed' or 'timeout' must eventually have
2159 *    scsi_finish_cmd() called for it.  we do all of the retry stuff
2160 *    here, so when we restart the host after we return it should have an
2161 *    empty queue.
2162 */
2163static void scsi_unjam_host(struct Scsi_Host *shost)
2164{
2165	unsigned long flags;
2166	LIST_HEAD(eh_work_q);
2167	LIST_HEAD(eh_done_q);
2168
2169	spin_lock_irqsave(shost->host_lock, flags);
2170	list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2171	spin_unlock_irqrestore(shost->host_lock, flags);
2172
2173	SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2174
2175	if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2176		scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
 
2177
2178	spin_lock_irqsave(shost->host_lock, flags);
2179	if (shost->eh_deadline != -1)
2180		shost->last_reset = 0;
2181	spin_unlock_irqrestore(shost->host_lock, flags);
2182	scsi_eh_flush_done_q(&eh_done_q);
2183}
2184
2185/**
2186 * scsi_error_handler - SCSI error handler thread
2187 * @data:	Host for which we are running.
2188 *
2189 * Notes:
2190 *    This is the main error handling loop.  This is run as a kernel thread
2191 *    for every SCSI host and handles all error handling activity.
2192 */
2193int scsi_error_handler(void *data)
2194{
2195	struct Scsi_Host *shost = data;
2196
2197	/*
2198	 * We use TASK_INTERRUPTIBLE so that the thread is not
2199	 * counted against the load average as a running process.
2200	 * We never actually get interrupted because kthread_run
2201	 * disables signal delivery for the created thread.
2202	 */
2203	while (true) {
2204		/*
2205		 * The sequence in kthread_stop() sets the stop flag first
2206		 * then wakes the process.  To avoid missed wakeups, the task
2207		 * should always be in a non running state before the stop
2208		 * flag is checked
2209		 */
2210		set_current_state(TASK_INTERRUPTIBLE);
2211		if (kthread_should_stop())
2212			break;
2213
2214		if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2215		    shost->host_failed != scsi_host_busy(shost)) {
2216			SCSI_LOG_ERROR_RECOVERY(1,
2217				shost_printk(KERN_INFO, shost,
2218					     "scsi_eh_%d: sleeping\n",
2219					     shost->host_no));
2220			schedule();
2221			continue;
2222		}
2223
2224		__set_current_state(TASK_RUNNING);
2225		SCSI_LOG_ERROR_RECOVERY(1,
2226			shost_printk(KERN_INFO, shost,
2227				     "scsi_eh_%d: waking up %d/%d/%d\n",
2228				     shost->host_no, shost->host_eh_scheduled,
2229				     shost->host_failed,
2230				     scsi_host_busy(shost)));
2231
2232		/*
2233		 * We have a host that is failing for some reason.  Figure out
2234		 * what we need to do to get it up and online again (if we can).
2235		 * If we fail, we end up taking the thing offline.
2236		 */
2237		if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2238			SCSI_LOG_ERROR_RECOVERY(1,
2239				shost_printk(KERN_ERR, shost,
2240					     "scsi_eh_%d: unable to autoresume\n",
2241					     shost->host_no));
2242			continue;
2243		}
2244
2245		if (shost->transportt->eh_strategy_handler)
2246			shost->transportt->eh_strategy_handler(shost);
2247		else
2248			scsi_unjam_host(shost);
2249
2250		/* All scmds have been handled */
2251		shost->host_failed = 0;
2252
2253		/*
2254		 * Note - if the above fails completely, the action is to take
2255		 * individual devices offline and flush the queue of any
2256		 * outstanding requests that may have been pending.  When we
2257		 * restart, we restart any I/O to any other devices on the bus
2258		 * which are still online.
2259		 */
2260		scsi_restart_operations(shost);
2261		if (!shost->eh_noresume)
2262			scsi_autopm_put_host(shost);
2263	}
2264	__set_current_state(TASK_RUNNING);
2265
2266	SCSI_LOG_ERROR_RECOVERY(1,
2267		shost_printk(KERN_INFO, shost,
2268			     "Error handler scsi_eh_%d exiting\n",
2269			     shost->host_no));
2270	shost->ehandler = NULL;
2271	return 0;
2272}
2273
2274/*
2275 * Function:    scsi_report_bus_reset()
2276 *
2277 * Purpose:     Utility function used by low-level drivers to report that
2278 *		they have observed a bus reset on the bus being handled.
2279 *
2280 * Arguments:   shost       - Host in question
2281 *		channel     - channel on which reset was observed.
2282 *
2283 * Returns:     Nothing
2284 *
2285 * Lock status: Host lock must be held.
2286 *
2287 * Notes:       This only needs to be called if the reset is one which
2288 *		originates from an unknown location.  Resets originated
2289 *		by the mid-level itself don't need to call this, but there
2290 *		should be no harm.
2291 *
2292 *		The main purpose of this is to make sure that a CHECK_CONDITION
2293 *		is properly treated.
2294 */
2295void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2296{
2297	struct scsi_device *sdev;
2298
2299	__shost_for_each_device(sdev, shost) {
2300		if (channel == sdev_channel(sdev))
2301			__scsi_report_device_reset(sdev, NULL);
2302	}
2303}
2304EXPORT_SYMBOL(scsi_report_bus_reset);
2305
2306/*
2307 * Function:    scsi_report_device_reset()
2308 *
2309 * Purpose:     Utility function used by low-level drivers to report that
2310 *		they have observed a device reset on the device being handled.
2311 *
2312 * Arguments:   shost       - Host in question
2313 *		channel     - channel on which reset was observed
2314 *		target	    - target on which reset was observed
2315 *
2316 * Returns:     Nothing
2317 *
2318 * Lock status: Host lock must be held
2319 *
2320 * Notes:       This only needs to be called if the reset is one which
2321 *		originates from an unknown location.  Resets originated
2322 *		by the mid-level itself don't need to call this, but there
2323 *		should be no harm.
2324 *
2325 *		The main purpose of this is to make sure that a CHECK_CONDITION
2326 *		is properly treated.
2327 */
2328void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2329{
2330	struct scsi_device *sdev;
2331
2332	__shost_for_each_device(sdev, shost) {
2333		if (channel == sdev_channel(sdev) &&
2334		    target == sdev_id(sdev))
2335			__scsi_report_device_reset(sdev, NULL);
2336	}
2337}
2338EXPORT_SYMBOL(scsi_report_device_reset);
2339
2340static void
2341scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2342{
2343}
2344
2345/**
2346 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2347 * @dev:	scsi_device to operate on
2348 * @arg:	reset type (see sg.h)
 
 
 
 
 
 
 
 
2349 */
2350int
2351scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2352{
2353	struct scsi_cmnd *scmd;
2354	struct Scsi_Host *shost = dev->host;
2355	struct request *rq;
2356	unsigned long flags;
2357	int error = 0, val;
2358	enum scsi_disposition rtn;
2359
2360	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2361		return -EACCES;
2362
2363	error = get_user(val, arg);
2364	if (error)
2365		return error;
2366
2367	if (scsi_autopm_get_host(shost) < 0)
2368		return -EIO;
2369
2370	error = -EIO;
2371	rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2372			shost->hostt->cmd_size, GFP_KERNEL);
2373	if (!rq)
2374		goto out_put_autopm_host;
2375	blk_rq_init(NULL, rq);
 
 
 
 
2376
2377	scmd = (struct scsi_cmnd *)(rq + 1);
2378	scsi_init_command(dev, scmd);
2379	scmd->request = rq;
2380	scmd->cmnd = scsi_req(rq)->cmd;
2381
2382	scmd->scsi_done		= scsi_reset_provider_done_command;
2383	memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2384
2385	scmd->cmd_len			= 0;
2386
2387	scmd->sc_data_direction		= DMA_BIDIRECTIONAL;
2388
2389	spin_lock_irqsave(shost->host_lock, flags);
2390	shost->tmf_in_progress = 1;
2391	spin_unlock_irqrestore(shost->host_lock, flags);
2392
2393	switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2394	case SG_SCSI_RESET_NOTHING:
2395		rtn = SUCCESS;
2396		break;
2397	case SG_SCSI_RESET_DEVICE:
2398		rtn = scsi_try_bus_device_reset(scmd);
2399		if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2400			break;
2401		fallthrough;
2402	case SG_SCSI_RESET_TARGET:
2403		rtn = scsi_try_target_reset(scmd);
2404		if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2405			break;
2406		fallthrough;
2407	case SG_SCSI_RESET_BUS:
2408		rtn = scsi_try_bus_reset(scmd);
2409		if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2410			break;
2411		fallthrough;
2412	case SG_SCSI_RESET_HOST:
2413		rtn = scsi_try_host_reset(scmd);
2414		if (rtn == SUCCESS)
2415			break;
2416		fallthrough;
 
 
 
2417	default:
2418		rtn = FAILED;
2419		break;
2420	}
2421
2422	error = (rtn == SUCCESS) ? 0 : -EIO;
2423
2424	spin_lock_irqsave(shost->host_lock, flags);
2425	shost->tmf_in_progress = 0;
2426	spin_unlock_irqrestore(shost->host_lock, flags);
2427
2428	/*
2429	 * be sure to wake up anyone who was sleeping or had their queue
2430	 * suspended while we performed the TMF.
2431	 */
2432	SCSI_LOG_ERROR_RECOVERY(3,
2433		shost_printk(KERN_INFO, shost,
2434			     "waking up host to restart after TMF\n"));
2435
2436	wake_up(&shost->host_wait);
2437	scsi_run_host_queues(shost);
2438
2439	kfree(rq);
2440
 
2441out_put_autopm_host:
2442	scsi_autopm_put_host(shost);
2443	return error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2444}
 
2445
2446bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2447				  struct scsi_sense_hdr *sshdr)
2448{
2449	return scsi_normalize_sense(cmd->sense_buffer,
2450			SCSI_SENSE_BUFFERSIZE, sshdr);
2451}
2452EXPORT_SYMBOL(scsi_command_normalize_sense);
2453
2454/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2455 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2456 * @sense_buffer:	byte array of sense data
2457 * @sb_len:		number of valid bytes in sense_buffer
2458 * @info_out:		pointer to 64 integer where 8 or 4 byte information
2459 *			field will be placed if found.
2460 *
2461 * Return value:
2462 *	true if information field found, false if not found.
2463 */
2464bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2465			     u64 *info_out)
2466{
 
2467	const u8 * ucp;
 
2468
2469	if (sb_len < 7)
2470		return false;
2471	switch (sense_buffer[0] & 0x7f) {
2472	case 0x70:
2473	case 0x71:
2474		if (sense_buffer[0] & 0x80) {
2475			*info_out = get_unaligned_be32(&sense_buffer[3]);
2476			return true;
2477		}
2478		return false;
 
 
2479	case 0x72:
2480	case 0x73:
2481		ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2482					   0 /* info desc */);
2483		if (ucp && (0xa == ucp[1])) {
2484			*info_out = get_unaligned_be64(&ucp[4]);
2485			return true;
2486		}
2487		return false;
 
 
 
 
 
 
2488	default:
2489		return false;
2490	}
2491}
2492EXPORT_SYMBOL(scsi_get_sense_info_fld);
v3.15
 
   1/*
   2 *  scsi_error.c Copyright (C) 1997 Eric Youngdale
   3 *
   4 *  SCSI error/timeout handling
   5 *      Initial versions: Eric Youngdale.  Based upon conversations with
   6 *                        Leonard Zubkoff and David Miller at Linux Expo,
   7 *                        ideas originating from all over the place.
   8 *
   9 *	Restructured scsi_unjam_host and associated functions.
  10 *	September 04, 2002 Mike Anderson (andmike@us.ibm.com)
  11 *
  12 *	Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
  13 *	minor cleanups.
  14 *	September 30, 2002 Mike Anderson (andmike@us.ibm.com)
  15 */
  16
  17#include <linux/module.h>
  18#include <linux/sched.h>
  19#include <linux/gfp.h>
  20#include <linux/timer.h>
  21#include <linux/string.h>
  22#include <linux/kernel.h>
  23#include <linux/freezer.h>
  24#include <linux/kthread.h>
  25#include <linux/interrupt.h>
  26#include <linux/blkdev.h>
  27#include <linux/delay.h>
  28#include <linux/jiffies.h>
  29
  30#include <scsi/scsi.h>
  31#include <scsi/scsi_cmnd.h>
  32#include <scsi/scsi_dbg.h>
  33#include <scsi/scsi_device.h>
  34#include <scsi/scsi_driver.h>
  35#include <scsi/scsi_eh.h>
 
  36#include <scsi/scsi_transport.h>
  37#include <scsi/scsi_host.h>
  38#include <scsi/scsi_ioctl.h>
 
 
 
  39
  40#include "scsi_priv.h"
  41#include "scsi_logging.h"
  42#include "scsi_transport_api.h"
  43
  44#include <trace/events/scsi.h>
  45
 
 
  46static void scsi_eh_done(struct scsi_cmnd *scmd);
  47
  48/*
  49 * These should *probably* be handled by the host itself.
  50 * Since it is allowed to sleep, it probably should.
  51 */
  52#define BUS_RESET_SETTLE_TIME   (10)
  53#define HOST_RESET_SETTLE_TIME  (10)
  54
  55static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
  56static int scsi_try_to_abort_cmd(struct scsi_host_template *,
  57				 struct scsi_cmnd *);
  58
  59/* called with shost->host_lock held */
  60void scsi_eh_wakeup(struct Scsi_Host *shost)
  61{
  62	if (shost->host_busy == shost->host_failed) {
 
 
  63		trace_scsi_eh_wakeup(shost);
  64		wake_up_process(shost->ehandler);
  65		SCSI_LOG_ERROR_RECOVERY(5,
  66				printk("Waking error handler thread\n"));
  67	}
  68}
  69
  70/**
  71 * scsi_schedule_eh - schedule EH for SCSI host
  72 * @shost:	SCSI host to invoke error handling on.
  73 *
  74 * Schedule SCSI EH without scmd.
  75 */
  76void scsi_schedule_eh(struct Scsi_Host *shost)
  77{
  78	unsigned long flags;
  79
  80	spin_lock_irqsave(shost->host_lock, flags);
  81
  82	if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
  83	    scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
  84		shost->host_eh_scheduled++;
  85		scsi_eh_wakeup(shost);
  86	}
  87
  88	spin_unlock_irqrestore(shost->host_lock, flags);
  89}
  90EXPORT_SYMBOL_GPL(scsi_schedule_eh);
  91
  92static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
  93{
  94	if (!shost->last_reset || shost->eh_deadline == -1)
  95		return 0;
  96
  97	/*
  98	 * 32bit accesses are guaranteed to be atomic
  99	 * (on all supported architectures), so instead
 100	 * of using a spinlock we can as well double check
 101	 * if eh_deadline has been set to 'off' during the
 102	 * time_before call.
 103	 */
 104	if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
 105	    shost->eh_deadline > -1)
 106		return 0;
 107
 108	return 1;
 109}
 110
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 111/**
 112 * scmd_eh_abort_handler - Handle command aborts
 113 * @work:	command to be aborted.
 
 
 
 
 
 
 114 */
 115void
 116scmd_eh_abort_handler(struct work_struct *work)
 117{
 118	struct scsi_cmnd *scmd =
 119		container_of(work, struct scsi_cmnd, abort_work.work);
 120	struct scsi_device *sdev = scmd->device;
 121	int rtn;
 122
 123	if (scsi_host_eh_past_deadline(sdev->host)) {
 124		SCSI_LOG_ERROR_RECOVERY(3,
 125			scmd_printk(KERN_INFO, scmd,
 126				    "scmd %p eh timeout, not aborting\n",
 127				    scmd));
 128	} else {
 129		SCSI_LOG_ERROR_RECOVERY(3,
 130			scmd_printk(KERN_INFO, scmd,
 131				    "aborting command %p\n", scmd));
 132		rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
 133		if (rtn == SUCCESS) {
 134			scmd->result |= DID_TIME_OUT << 16;
 135			if (scsi_host_eh_past_deadline(sdev->host)) {
 136				SCSI_LOG_ERROR_RECOVERY(3,
 137					scmd_printk(KERN_INFO, scmd,
 138						    "scmd %p eh timeout, "
 139						    "not retrying aborted "
 140						    "command\n", scmd));
 141			} else if (!scsi_noretry_cmd(scmd) &&
 142			    (++scmd->retries <= scmd->allowed)) {
 
 143				SCSI_LOG_ERROR_RECOVERY(3,
 144					scmd_printk(KERN_WARNING, scmd,
 145						    "scmd %p retry "
 146						    "aborted command\n", scmd));
 147				scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
 148				return;
 149			} else {
 150				SCSI_LOG_ERROR_RECOVERY(3,
 151					scmd_printk(KERN_WARNING, scmd,
 152						    "scmd %p finish "
 153						    "aborted command\n", scmd));
 154				scsi_finish_command(scmd);
 155				return;
 156			}
 157		} else {
 158			SCSI_LOG_ERROR_RECOVERY(3,
 159				scmd_printk(KERN_INFO, scmd,
 160					    "scmd %p abort failed, rtn %d\n",
 161					    scmd, rtn));
 
 162		}
 163	}
 164
 165	if (!scsi_eh_scmd_add(scmd, 0)) {
 166		SCSI_LOG_ERROR_RECOVERY(3,
 167			scmd_printk(KERN_WARNING, scmd,
 168				    "scmd %p terminate "
 169				    "aborted command\n", scmd));
 170		scmd->result |= DID_TIME_OUT << 16;
 171		scsi_finish_command(scmd);
 172	}
 173}
 174
 175/**
 176 * scsi_abort_command - schedule a command abort
 177 * @scmd:	scmd to abort.
 178 *
 179 * We only need to abort commands after a command timeout
 180 */
 181static int
 182scsi_abort_command(struct scsi_cmnd *scmd)
 183{
 184	struct scsi_device *sdev = scmd->device;
 185	struct Scsi_Host *shost = sdev->host;
 186	unsigned long flags;
 187
 188	if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
 189		/*
 190		 * Retry after abort failed, escalate to next level.
 191		 */
 192		scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED;
 193		SCSI_LOG_ERROR_RECOVERY(3,
 194			scmd_printk(KERN_INFO, scmd,
 195				    "scmd %p previous abort failed\n", scmd));
 196		cancel_delayed_work(&scmd->abort_work);
 197		return FAILED;
 198	}
 199
 200	/*
 201	 * Do not try a command abort if
 202	 * SCSI EH has already started.
 203	 */
 204	spin_lock_irqsave(shost->host_lock, flags);
 205	if (scsi_host_in_recovery(shost)) {
 206		spin_unlock_irqrestore(shost->host_lock, flags);
 207		SCSI_LOG_ERROR_RECOVERY(3,
 208			scmd_printk(KERN_INFO, scmd,
 209				    "scmd %p not aborting, host in recovery\n",
 210				    scmd));
 211		return FAILED;
 212	}
 213
 214	if (shost->eh_deadline != -1 && !shost->last_reset)
 215		shost->last_reset = jiffies;
 216	spin_unlock_irqrestore(shost->host_lock, flags);
 217
 218	scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
 219	SCSI_LOG_ERROR_RECOVERY(3,
 220		scmd_printk(KERN_INFO, scmd,
 221			    "scmd %p abort scheduled\n", scmd));
 222	queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
 223	return SUCCESS;
 224}
 225
 226/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 227 * scsi_eh_scmd_add - add scsi cmd to error handling.
 228 * @scmd:	scmd to run eh on.
 229 * @eh_flag:	optional SCSI_EH flag.
 230 *
 231 * Return value:
 232 *	0 on failure.
 233 */
 234int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
 235{
 236	struct Scsi_Host *shost = scmd->device->host;
 237	unsigned long flags;
 238	int ret = 0;
 239
 240	if (!shost->ehandler)
 241		return 0;
 242
 243	spin_lock_irqsave(shost->host_lock, flags);
 244	if (scsi_host_set_state(shost, SHOST_RECOVERY))
 245		if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
 246			goto out_unlock;
 247
 248	if (shost->eh_deadline != -1 && !shost->last_reset)
 249		shost->last_reset = jiffies;
 250
 251	ret = 1;
 252	if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
 253		eh_flag &= ~SCSI_EH_CANCEL_CMD;
 254	scmd->eh_eflags |= eh_flag;
 255	list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
 256	shost->host_failed++;
 257	scsi_eh_wakeup(shost);
 258 out_unlock:
 259	spin_unlock_irqrestore(shost->host_lock, flags);
 260	return ret;
 
 
 
 
 261}
 262
 263/**
 264 * scsi_times_out - Timeout function for normal scsi commands.
 265 * @req:	request that is timing out.
 266 *
 267 * Notes:
 268 *     We do not need to lock this.  There is the potential for a race
 269 *     only in that the normal completion handling might run, but if the
 270 *     normal completion function determines that the timer has already
 271 *     fired, then it mustn't do anything.
 272 */
 273enum blk_eh_timer_return scsi_times_out(struct request *req)
 274{
 275	struct scsi_cmnd *scmd = req->special;
 276	enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
 277	struct Scsi_Host *host = scmd->device->host;
 278
 279	trace_scsi_dispatch_cmd_timeout(scmd);
 280	scsi_log_completion(scmd, TIMEOUT_ERROR);
 281
 282	if (host->eh_deadline != -1 && !host->last_reset)
 283		host->last_reset = jiffies;
 284
 285	if (host->transportt->eh_timed_out)
 286		rtn = host->transportt->eh_timed_out(scmd);
 287	else if (host->hostt->eh_timed_out)
 288		rtn = host->hostt->eh_timed_out(scmd);
 289
 290	if (rtn == BLK_EH_NOT_HANDLED && !host->hostt->no_async_abort)
 291		if (scsi_abort_command(scmd) == SUCCESS)
 292			return BLK_EH_NOT_HANDLED;
 293
 294	scmd->result |= DID_TIME_OUT << 16;
 295
 296	if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
 297		     !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)))
 298		rtn = BLK_EH_HANDLED;
 
 
 
 
 
 
 
 
 
 
 
 299
 300	return rtn;
 301}
 302
 303/**
 304 * scsi_block_when_processing_errors - Prevent cmds from being queued.
 305 * @sdev:	Device on which we are performing recovery.
 306 *
 307 * Description:
 308 *     We block until the host is out of error recovery, and then check to
 309 *     see whether the host or the device is offline.
 310 *
 311 * Return value:
 312 *     0 when dev was taken offline by error recovery. 1 OK to proceed.
 313 */
 314int scsi_block_when_processing_errors(struct scsi_device *sdev)
 315{
 316	int online;
 317
 318	wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
 319
 320	online = scsi_device_online(sdev);
 321
 322	SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
 323					  online));
 324
 325	return online;
 326}
 327EXPORT_SYMBOL(scsi_block_when_processing_errors);
 328
 329#ifdef CONFIG_SCSI_LOGGING
 330/**
 331 * scsi_eh_prt_fail_stats - Log info on failures.
 332 * @shost:	scsi host being recovered.
 333 * @work_q:	Queue of scsi cmds to process.
 334 */
 335static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
 336					  struct list_head *work_q)
 337{
 338	struct scsi_cmnd *scmd;
 339	struct scsi_device *sdev;
 340	int total_failures = 0;
 341	int cmd_failed = 0;
 342	int cmd_cancel = 0;
 343	int devices_failed = 0;
 344
 345	shost_for_each_device(sdev, shost) {
 346		list_for_each_entry(scmd, work_q, eh_entry) {
 347			if (scmd->device == sdev) {
 348				++total_failures;
 349				if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
 350					++cmd_cancel;
 351				else
 352					++cmd_failed;
 353			}
 354		}
 355
 356		if (cmd_cancel || cmd_failed) {
 357			SCSI_LOG_ERROR_RECOVERY(3,
 358				sdev_printk(KERN_INFO, sdev,
 359					    "%s: cmds failed: %d, cancel: %d\n",
 360					    __func__, cmd_failed,
 361					    cmd_cancel));
 362			cmd_cancel = 0;
 363			cmd_failed = 0;
 364			++devices_failed;
 365		}
 366	}
 367
 368	SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
 369					  " devices require eh work\n",
 
 370				   total_failures, devices_failed));
 371}
 372#endif
 373
 374 /**
 375 * scsi_report_lun_change - Set flag on all *other* devices on the same target
 376 *                          to indicate that a UNIT ATTENTION is expected.
 377 * @sdev:	Device reporting the UNIT ATTENTION
 378 */
 379static void scsi_report_lun_change(struct scsi_device *sdev)
 380{
 381	sdev->sdev_target->expecting_lun_change = 1;
 382}
 383
 384/**
 385 * scsi_report_sense - Examine scsi sense information and log messages for
 386 *		       certain conditions, also issue uevents for some of them.
 387 * @sdev:	Device reporting the sense code
 388 * @sshdr:	sshdr to be examined
 389 */
 390static void scsi_report_sense(struct scsi_device *sdev,
 391			      struct scsi_sense_hdr *sshdr)
 392{
 393	enum scsi_device_event evt_type = SDEV_EVT_MAXBITS;	/* i.e. none */
 394
 395	if (sshdr->sense_key == UNIT_ATTENTION) {
 396		if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
 397			evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
 398			sdev_printk(KERN_WARNING, sdev,
 399				    "Inquiry data has changed");
 400		} else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
 401			evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
 402			scsi_report_lun_change(sdev);
 403			sdev_printk(KERN_WARNING, sdev,
 404				    "Warning! Received an indication that the "
 405				    "LUN assignments on this target have "
 406				    "changed. The Linux SCSI layer does not "
 407				    "automatically remap LUN assignments.\n");
 408		} else if (sshdr->asc == 0x3f)
 409			sdev_printk(KERN_WARNING, sdev,
 410				    "Warning! Received an indication that the "
 411				    "operating parameters on this target have "
 412				    "changed. The Linux SCSI layer does not "
 413				    "automatically adjust these parameters.\n");
 414
 415		if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
 416			evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
 417			sdev_printk(KERN_WARNING, sdev,
 418				    "Warning! Received an indication that the "
 419				    "LUN reached a thin provisioning soft "
 420				    "threshold.\n");
 421		}
 422
 
 
 
 
 
 
 423		if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
 424			evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
 425			sdev_printk(KERN_WARNING, sdev,
 426				    "Mode parameters changed");
 
 
 
 
 427		} else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
 428			evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
 429			sdev_printk(KERN_WARNING, sdev,
 430				    "Capacity data has changed");
 431		} else if (sshdr->asc == 0x2a)
 432			sdev_printk(KERN_WARNING, sdev,
 433				    "Parameters changed");
 434	}
 435
 436	if (evt_type != SDEV_EVT_MAXBITS) {
 437		set_bit(evt_type, sdev->pending_events);
 438		schedule_work(&sdev->event_work);
 439	}
 440}
 441
 442/**
 443 * scsi_check_sense - Examine scsi cmd sense
 444 * @scmd:	Cmd to have sense checked.
 445 *
 446 * Return value:
 447 *	SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
 448 *
 449 * Notes:
 450 *	When a deferred error is detected the current command has
 451 *	not been executed and needs retrying.
 452 */
 453static int scsi_check_sense(struct scsi_cmnd *scmd)
 454{
 455	struct scsi_device *sdev = scmd->device;
 456	struct scsi_sense_hdr sshdr;
 457
 458	if (! scsi_command_normalize_sense(scmd, &sshdr))
 459		return FAILED;	/* no valid sense data */
 460
 461	if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
 462		/*
 463		 * nasty: for mid-layer issued TURs, we need to return the
 464		 * actual sense data without any recovery attempt.  For eh
 465		 * issued ones, we need to try to recover and interpret
 466		 */
 467		return SUCCESS;
 468
 469	scsi_report_sense(sdev, &sshdr);
 470
 471	if (scsi_sense_is_deferred(&sshdr))
 472		return NEEDS_RETRY;
 473
 474	if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
 475			sdev->scsi_dh_data->scsi_dh->check_sense) {
 476		int rc;
 477
 478		rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
 479		if (rc != SCSI_RETURN_NOT_HANDLED)
 480			return rc;
 481		/* handler does not care. Drop down to default handling */
 482	}
 483
 
 
 
 
 
 
 
 
 484	/*
 485	 * Previous logic looked for FILEMARK, EOM or ILI which are
 486	 * mainly associated with tapes and returned SUCCESS.
 487	 */
 488	if (sshdr.response_code == 0x70) {
 489		/* fixed format */
 490		if (scmd->sense_buffer[2] & 0xe0)
 491			return SUCCESS;
 492	} else {
 493		/*
 494		 * descriptor format: look for "stream commands sense data
 495		 * descriptor" (see SSC-3). Assume single sense data
 496		 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
 497		 */
 498		if ((sshdr.additional_length > 3) &&
 499		    (scmd->sense_buffer[8] == 0x4) &&
 500		    (scmd->sense_buffer[11] & 0xe0))
 501			return SUCCESS;
 502	}
 503
 504	switch (sshdr.sense_key) {
 505	case NO_SENSE:
 506		return SUCCESS;
 507	case RECOVERED_ERROR:
 508		return /* soft_error */ SUCCESS;
 509
 510	case ABORTED_COMMAND:
 511		if (sshdr.asc == 0x10) /* DIF */
 512			return SUCCESS;
 513
 
 
 
 
 
 
 514		return NEEDS_RETRY;
 515	case NOT_READY:
 516	case UNIT_ATTENTION:
 517		/*
 518		 * if we are expecting a cc/ua because of a bus reset that we
 519		 * performed, treat this just as a retry.  otherwise this is
 520		 * information that we should pass up to the upper-level driver
 521		 * so that we can deal with it there.
 522		 */
 523		if (scmd->device->expecting_cc_ua) {
 524			/*
 525			 * Because some device does not queue unit
 526			 * attentions correctly, we carefully check
 527			 * additional sense code and qualifier so as
 528			 * not to squash media change unit attention.
 529			 */
 530			if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
 531				scmd->device->expecting_cc_ua = 0;
 532				return NEEDS_RETRY;
 533			}
 534		}
 535		/*
 536		 * we might also expect a cc/ua if another LUN on the target
 537		 * reported a UA with an ASC/ASCQ of 3F 0E -
 538		 * REPORTED LUNS DATA HAS CHANGED.
 539		 */
 540		if (scmd->device->sdev_target->expecting_lun_change &&
 541		    sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
 542			return NEEDS_RETRY;
 543		/*
 544		 * if the device is in the process of becoming ready, we
 545		 * should retry.
 546		 */
 547		if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
 548			return NEEDS_RETRY;
 549		/*
 550		 * if the device is not started, we need to wake
 551		 * the error handler to start the motor
 552		 */
 553		if (scmd->device->allow_restart &&
 554		    (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
 555			return FAILED;
 556		/*
 557		 * Pass the UA upwards for a determination in the completion
 558		 * functions.
 559		 */
 560		return SUCCESS;
 561
 562		/* these are not supported */
 563	case DATA_PROTECT:
 564		if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
 565			/* Thin provisioning hard threshold reached */
 566			set_host_byte(scmd, DID_ALLOC_FAILURE);
 567			return SUCCESS;
 568		}
 
 569	case COPY_ABORTED:
 570	case VOLUME_OVERFLOW:
 571	case MISCOMPARE:
 572	case BLANK_CHECK:
 573		set_host_byte(scmd, DID_TARGET_FAILURE);
 574		return SUCCESS;
 575
 576	case MEDIUM_ERROR:
 577		if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
 578		    sshdr.asc == 0x13 || /* AMNF DATA FIELD */
 579		    sshdr.asc == 0x14) { /* RECORD NOT FOUND */
 580			set_host_byte(scmd, DID_MEDIUM_ERROR);
 581			return SUCCESS;
 582		}
 583		return NEEDS_RETRY;
 584
 585	case HARDWARE_ERROR:
 586		if (scmd->device->retry_hwerror)
 587			return ADD_TO_MLQUEUE;
 588		else
 589			set_host_byte(scmd, DID_TARGET_FAILURE);
 
 590
 591	case ILLEGAL_REQUEST:
 592		if (sshdr.asc == 0x20 || /* Invalid command operation code */
 593		    sshdr.asc == 0x21 || /* Logical block address out of range */
 
 594		    sshdr.asc == 0x24 || /* Invalid field in cdb */
 595		    sshdr.asc == 0x26) { /* Parameter value invalid */
 
 596			set_host_byte(scmd, DID_TARGET_FAILURE);
 597		}
 598		return SUCCESS;
 599
 600	default:
 601		return SUCCESS;
 602	}
 603}
 
 604
 605static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
 606{
 607	struct scsi_host_template *sht = sdev->host->hostt;
 608	struct scsi_device *tmp_sdev;
 609
 610	if (!sht->change_queue_depth ||
 611	    sdev->queue_depth >= sdev->max_queue_depth)
 612		return;
 613
 614	if (time_before(jiffies,
 615	    sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
 616		return;
 617
 618	if (time_before(jiffies,
 619	    sdev->last_queue_full_time + sdev->queue_ramp_up_period))
 620		return;
 621
 622	/*
 623	 * Walk all devices of a target and do
 624	 * ramp up on them.
 625	 */
 626	shost_for_each_device(tmp_sdev, sdev->host) {
 627		if (tmp_sdev->channel != sdev->channel ||
 628		    tmp_sdev->id != sdev->id ||
 629		    tmp_sdev->queue_depth == sdev->max_queue_depth)
 630			continue;
 631		/*
 632		 * call back into LLD to increase queue_depth by one
 633		 * with ramp up reason code.
 634		 */
 635		sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
 636					SCSI_QDEPTH_RAMP_UP);
 637		sdev->last_queue_ramp_up = jiffies;
 638	}
 639}
 640
 641static void scsi_handle_queue_full(struct scsi_device *sdev)
 642{
 643	struct scsi_host_template *sht = sdev->host->hostt;
 644	struct scsi_device *tmp_sdev;
 645
 646	if (!sht->change_queue_depth)
 647		return;
 648
 649	shost_for_each_device(tmp_sdev, sdev->host) {
 650		if (tmp_sdev->channel != sdev->channel ||
 651		    tmp_sdev->id != sdev->id)
 652			continue;
 653		/*
 654		 * We do not know the number of commands that were at
 655		 * the device when we got the queue full so we start
 656		 * from the highest possible value and work our way down.
 657		 */
 658		sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
 659					SCSI_QDEPTH_QFULL);
 660	}
 661}
 662
 663/**
 664 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
 665 * @scmd:	SCSI cmd to examine.
 666 *
 667 * Notes:
 668 *    This is *only* called when we are examining the status of commands
 669 *    queued during error recovery.  the main difference here is that we
 670 *    don't allow for the possibility of retries here, and we are a lot
 671 *    more restrictive about what we consider acceptable.
 672 */
 673static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
 674{
 675	/*
 676	 * first check the host byte, to see if there is anything in there
 677	 * that would indicate what we need to do.
 678	 */
 679	if (host_byte(scmd->result) == DID_RESET) {
 680		/*
 681		 * rats.  we are already in the error handler, so we now
 682		 * get to try and figure out what to do next.  if the sense
 683		 * is valid, we have a pretty good idea of what to do.
 684		 * if not, we mark it as FAILED.
 685		 */
 686		return scsi_check_sense(scmd);
 687	}
 688	if (host_byte(scmd->result) != DID_OK)
 689		return FAILED;
 690
 691	/*
 692	 * next, check the message byte.
 693	 */
 694	if (msg_byte(scmd->result) != COMMAND_COMPLETE)
 695		return FAILED;
 696
 697	/*
 698	 * now, check the status byte to see if this indicates
 699	 * anything special.
 700	 */
 701	switch (status_byte(scmd->result)) {
 702	case GOOD:
 703		scsi_handle_queue_ramp_up(scmd->device);
 704	case COMMAND_TERMINATED:
 
 705		return SUCCESS;
 706	case CHECK_CONDITION:
 707		return scsi_check_sense(scmd);
 708	case CONDITION_GOOD:
 709	case INTERMEDIATE_GOOD:
 710	case INTERMEDIATE_C_GOOD:
 711		/*
 712		 * who knows?  FIXME(eric)
 713		 */
 714		return SUCCESS;
 715	case RESERVATION_CONFLICT:
 716		if (scmd->cmnd[0] == TEST_UNIT_READY)
 717			/* it is a success, we probed the device and
 718			 * found it */
 719			return SUCCESS;
 720		/* otherwise, we failed to send the command */
 721		return FAILED;
 722	case QUEUE_FULL:
 723		scsi_handle_queue_full(scmd->device);
 724		/* fall through */
 725	case BUSY:
 726		return NEEDS_RETRY;
 727	default:
 728		return FAILED;
 729	}
 730	return FAILED;
 731}
 732
 733/**
 734 * scsi_eh_done - Completion function for error handling.
 735 * @scmd:	Cmd that is done.
 736 */
 737static void scsi_eh_done(struct scsi_cmnd *scmd)
 738{
 739	struct completion *eh_action;
 740
 741	SCSI_LOG_ERROR_RECOVERY(3,
 742		printk("%s scmd: %p result: %x\n",
 743			__func__, scmd, scmd->result));
 744
 745	eh_action = scmd->device->host->eh_action;
 746	if (eh_action)
 747		complete(eh_action);
 748}
 749
 750/**
 751 * scsi_try_host_reset - ask host adapter to reset itself
 752 * @scmd:	SCSI cmd to send host reset.
 753 */
 754static int scsi_try_host_reset(struct scsi_cmnd *scmd)
 755{
 756	unsigned long flags;
 757	int rtn;
 758	struct Scsi_Host *host = scmd->device->host;
 759	struct scsi_host_template *hostt = host->hostt;
 760
 761	SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
 762					  __func__));
 763
 764	if (!hostt->eh_host_reset_handler)
 765		return FAILED;
 766
 767	rtn = hostt->eh_host_reset_handler(scmd);
 768
 769	if (rtn == SUCCESS) {
 770		if (!hostt->skip_settle_delay)
 771			ssleep(HOST_RESET_SETTLE_TIME);
 772		spin_lock_irqsave(host->host_lock, flags);
 773		scsi_report_bus_reset(host, scmd_channel(scmd));
 774		spin_unlock_irqrestore(host->host_lock, flags);
 775	}
 776
 777	return rtn;
 778}
 779
 780/**
 781 * scsi_try_bus_reset - ask host to perform a bus reset
 782 * @scmd:	SCSI cmd to send bus reset.
 783 */
 784static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
 785{
 786	unsigned long flags;
 787	int rtn;
 788	struct Scsi_Host *host = scmd->device->host;
 789	struct scsi_host_template *hostt = host->hostt;
 790
 791	SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
 792					  __func__));
 793
 794	if (!hostt->eh_bus_reset_handler)
 795		return FAILED;
 796
 797	rtn = hostt->eh_bus_reset_handler(scmd);
 798
 799	if (rtn == SUCCESS) {
 800		if (!hostt->skip_settle_delay)
 801			ssleep(BUS_RESET_SETTLE_TIME);
 802		spin_lock_irqsave(host->host_lock, flags);
 803		scsi_report_bus_reset(host, scmd_channel(scmd));
 804		spin_unlock_irqrestore(host->host_lock, flags);
 805	}
 806
 807	return rtn;
 808}
 809
 810static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
 811{
 812	sdev->was_reset = 1;
 813	sdev->expecting_cc_ua = 1;
 814}
 815
 816/**
 817 * scsi_try_target_reset - Ask host to perform a target reset
 818 * @scmd:	SCSI cmd used to send a target reset
 819 *
 820 * Notes:
 821 *    There is no timeout for this operation.  if this operation is
 822 *    unreliable for a given host, then the host itself needs to put a
 823 *    timer on it, and set the host back to a consistent state prior to
 824 *    returning.
 825 */
 826static int scsi_try_target_reset(struct scsi_cmnd *scmd)
 827{
 828	unsigned long flags;
 829	int rtn;
 830	struct Scsi_Host *host = scmd->device->host;
 831	struct scsi_host_template *hostt = host->hostt;
 832
 833	if (!hostt->eh_target_reset_handler)
 834		return FAILED;
 835
 836	rtn = hostt->eh_target_reset_handler(scmd);
 837	if (rtn == SUCCESS) {
 838		spin_lock_irqsave(host->host_lock, flags);
 839		__starget_for_each_device(scsi_target(scmd->device), NULL,
 840					  __scsi_report_device_reset);
 841		spin_unlock_irqrestore(host->host_lock, flags);
 842	}
 843
 844	return rtn;
 845}
 846
 847/**
 848 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
 849 * @scmd:	SCSI cmd used to send BDR
 850 *
 851 * Notes:
 852 *    There is no timeout for this operation.  if this operation is
 853 *    unreliable for a given host, then the host itself needs to put a
 854 *    timer on it, and set the host back to a consistent state prior to
 855 *    returning.
 856 */
 857static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
 858{
 859	int rtn;
 860	struct scsi_host_template *hostt = scmd->device->host->hostt;
 861
 862	if (!hostt->eh_device_reset_handler)
 863		return FAILED;
 864
 865	rtn = hostt->eh_device_reset_handler(scmd);
 866	if (rtn == SUCCESS)
 867		__scsi_report_device_reset(scmd->device, NULL);
 868	return rtn;
 869}
 870
 871static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 872{
 873	if (!hostt->eh_abort_handler)
 874		return FAILED;
 875
 876	return hostt->eh_abort_handler(scmd);
 877}
 878
 879static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
 880{
 881	if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
 882		if (scsi_try_bus_device_reset(scmd) != SUCCESS)
 883			if (scsi_try_target_reset(scmd) != SUCCESS)
 884				if (scsi_try_bus_reset(scmd) != SUCCESS)
 885					scsi_try_host_reset(scmd);
 886}
 887
 888/**
 889 * scsi_eh_prep_cmnd  - Save a scsi command info as part of error recovery
 890 * @scmd:       SCSI command structure to hijack
 891 * @ses:        structure to save restore information
 892 * @cmnd:       CDB to send. Can be NULL if no new cmnd is needed
 893 * @cmnd_size:  size in bytes of @cmnd (must be <= BLK_MAX_CDB)
 894 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
 895 *
 896 * This function is used to save a scsi command information before re-execution
 897 * as part of the error recovery process.  If @sense_bytes is 0 the command
 898 * sent must be one that does not transfer any data.  If @sense_bytes != 0
 899 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
 900 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
 901 */
 902void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
 903			unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
 904{
 905	struct scsi_device *sdev = scmd->device;
 906
 907	/*
 908	 * We need saved copies of a number of fields - this is because
 909	 * error handling may need to overwrite these with different values
 910	 * to run different commands, and once error handling is complete,
 911	 * we will need to restore these values prior to running the actual
 912	 * command.
 913	 */
 914	ses->cmd_len = scmd->cmd_len;
 915	ses->cmnd = scmd->cmnd;
 916	ses->data_direction = scmd->sc_data_direction;
 917	ses->sdb = scmd->sdb;
 918	ses->next_rq = scmd->request->next_rq;
 919	ses->result = scmd->result;
 
 920	ses->underflow = scmd->underflow;
 921	ses->prot_op = scmd->prot_op;
 
 922
 923	scmd->prot_op = SCSI_PROT_NORMAL;
 924	scmd->eh_eflags = 0;
 925	scmd->cmnd = ses->eh_cmnd;
 926	memset(scmd->cmnd, 0, BLK_MAX_CDB);
 927	memset(&scmd->sdb, 0, sizeof(scmd->sdb));
 928	scmd->request->next_rq = NULL;
 929	scmd->result = 0;
 
 930
 931	if (sense_bytes) {
 932		scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
 933					 sense_bytes);
 934		sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
 935			    scmd->sdb.length);
 936		scmd->sdb.table.sgl = &ses->sense_sgl;
 937		scmd->sc_data_direction = DMA_FROM_DEVICE;
 938		scmd->sdb.table.nents = 1;
 939		scmd->cmnd[0] = REQUEST_SENSE;
 940		scmd->cmnd[4] = scmd->sdb.length;
 941		scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
 942	} else {
 943		scmd->sc_data_direction = DMA_NONE;
 944		if (cmnd) {
 945			BUG_ON(cmnd_size > BLK_MAX_CDB);
 946			memcpy(scmd->cmnd, cmnd, cmnd_size);
 947			scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
 948		}
 949	}
 950
 951	scmd->underflow = 0;
 952
 953	if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
 954		scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
 955			(sdev->lun << 5 & 0xe0);
 956
 957	/*
 958	 * Zero the sense buffer.  The scsi spec mandates that any
 959	 * untransferred sense data should be interpreted as being zero.
 960	 */
 961	memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
 962}
 963EXPORT_SYMBOL(scsi_eh_prep_cmnd);
 964
 965/**
 966 * scsi_eh_restore_cmnd  - Restore a scsi command info as part of error recovery
 967 * @scmd:       SCSI command structure to restore
 968 * @ses:        saved information from a coresponding call to scsi_eh_prep_cmnd
 969 *
 970 * Undo any damage done by above scsi_eh_prep_cmnd().
 971 */
 972void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
 973{
 974	/*
 975	 * Restore original data
 976	 */
 977	scmd->cmd_len = ses->cmd_len;
 978	scmd->cmnd = ses->cmnd;
 979	scmd->sc_data_direction = ses->data_direction;
 980	scmd->sdb = ses->sdb;
 981	scmd->request->next_rq = ses->next_rq;
 982	scmd->result = ses->result;
 
 983	scmd->underflow = ses->underflow;
 984	scmd->prot_op = ses->prot_op;
 
 985}
 986EXPORT_SYMBOL(scsi_eh_restore_cmnd);
 987
 988/**
 989 * scsi_send_eh_cmnd  - submit a scsi command as part of error recovery
 990 * @scmd:       SCSI command structure to hijack
 991 * @cmnd:       CDB to send
 992 * @cmnd_size:  size in bytes of @cmnd
 993 * @timeout:    timeout for this request
 994 * @sense_bytes: size of sense data to copy or 0
 995 *
 996 * This function is used to send a scsi command down to a target device
 997 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
 998 *
 999 * Return value:
1000 *    SUCCESS or FAILED or NEEDS_RETRY
1001 */
1002static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1003			     int cmnd_size, int timeout, unsigned sense_bytes)
1004{
1005	struct scsi_device *sdev = scmd->device;
1006	struct Scsi_Host *shost = sdev->host;
1007	DECLARE_COMPLETION_ONSTACK(done);
1008	unsigned long timeleft = timeout;
1009	struct scsi_eh_save ses;
1010	const unsigned long stall_for = msecs_to_jiffies(100);
1011	int rtn;
1012
1013retry:
1014	scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1015	shost->eh_action = &done;
1016
1017	scsi_log_send(scmd);
1018	scmd->scsi_done = scsi_eh_done;
1019	rtn = shost->hostt->queuecommand(shost, scmd);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1020	if (rtn) {
1021		if (timeleft > stall_for) {
1022			scsi_eh_restore_cmnd(scmd, &ses);
1023			timeleft -= stall_for;
1024			msleep(jiffies_to_msecs(stall_for));
1025			goto retry;
1026		}
1027		/* signal not to enter either branch of the if () below */
1028		timeleft = 0;
1029		rtn = NEEDS_RETRY;
1030	} else {
1031		timeleft = wait_for_completion_timeout(&done, timeout);
 
1032	}
1033
1034	shost->eh_action = NULL;
1035
1036	scsi_log_completion(scmd, rtn);
1037
1038	SCSI_LOG_ERROR_RECOVERY(3,
1039		printk("%s: scmd: %p, timeleft: %ld\n",
1040			__func__, scmd, timeleft));
1041
1042	/*
1043	 * If there is time left scsi_eh_done got called, and we will examine
1044	 * the actual status codes to see whether the command actually did
1045	 * complete normally, else if we have a zero return and no time left,
1046	 * the command must still be pending, so abort it and return FAILED.
1047	 * If we never actually managed to issue the command, because
1048	 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1049	 * value above (so don't execute either branch of the if)
1050	 */
1051	if (timeleft) {
1052		rtn = scsi_eh_completed_normally(scmd);
1053		SCSI_LOG_ERROR_RECOVERY(3,
1054			printk("%s: scsi_eh_completed_normally %x\n",
1055			       __func__, rtn));
1056
1057		switch (rtn) {
1058		case SUCCESS:
1059		case NEEDS_RETRY:
1060		case FAILED:
1061			break;
1062		case ADD_TO_MLQUEUE:
1063			rtn = NEEDS_RETRY;
1064			break;
1065		default:
1066			rtn = FAILED;
1067			break;
1068		}
1069	} else if (!rtn) {
1070		scsi_abort_eh_cmnd(scmd);
1071		rtn = FAILED;
1072	}
1073
1074	scsi_eh_restore_cmnd(scmd, &ses);
1075
1076	return rtn;
1077}
1078
1079/**
1080 * scsi_request_sense - Request sense data from a particular target.
1081 * @scmd:	SCSI cmd for request sense.
1082 *
1083 * Notes:
1084 *    Some hosts automatically obtain this information, others require
1085 *    that we obtain it on our own. This function will *not* return until
1086 *    the command either times out, or it completes.
1087 */
1088static int scsi_request_sense(struct scsi_cmnd *scmd)
1089{
1090	return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1091}
1092
1093static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
 
1094{
1095	if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
1096		struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1097		if (sdrv->eh_action)
1098			rtn = sdrv->eh_action(scmd, rtn);
1099	}
1100	return rtn;
1101}
1102
1103/**
1104 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1105 * @scmd:	Original SCSI cmd that eh has finished.
1106 * @done_q:	Queue for processed commands.
1107 *
1108 * Notes:
1109 *    We don't want to use the normal command completion while we are are
1110 *    still handling errors - it may cause other commands to be queued,
1111 *    and that would disturb what we are doing.  Thus we really want to
1112 *    keep a list of pending commands for final completion, and once we
1113 *    are ready to leave error handling we handle completion for real.
1114 */
1115void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1116{
1117	scmd->device->host->host_failed--;
1118	scmd->eh_eflags = 0;
1119	list_move_tail(&scmd->eh_entry, done_q);
1120}
1121EXPORT_SYMBOL(scsi_eh_finish_cmd);
1122
1123/**
1124 * scsi_eh_get_sense - Get device sense data.
1125 * @work_q:	Queue of commands to process.
1126 * @done_q:	Queue of processed commands.
1127 *
1128 * Description:
1129 *    See if we need to request sense information.  if so, then get it
1130 *    now, so we have a better idea of what to do.
1131 *
1132 * Notes:
1133 *    This has the unfortunate side effect that if a shost adapter does
1134 *    not automatically request sense information, we end up shutting
1135 *    it down before we request it.
1136 *
1137 *    All drivers should request sense information internally these days,
1138 *    so for now all I have to say is tough noogies if you end up in here.
1139 *
1140 *    XXX: Long term this code should go away, but that needs an audit of
1141 *         all LLDDs first.
1142 */
1143int scsi_eh_get_sense(struct list_head *work_q,
1144		      struct list_head *done_q)
1145{
1146	struct scsi_cmnd *scmd, *next;
1147	struct Scsi_Host *shost;
1148	int rtn;
1149
 
 
 
 
1150	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1151		if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
1152		    SCSI_SENSE_VALID(scmd))
1153			continue;
1154
1155		shost = scmd->device->host;
1156		if (scsi_host_eh_past_deadline(shost)) {
1157			SCSI_LOG_ERROR_RECOVERY(3,
1158				shost_printk(KERN_INFO, shost,
1159					    "skip %s, past eh deadline\n",
1160					     __func__));
1161			break;
1162		}
1163		if (status_byte(scmd->result) != CHECK_CONDITION)
1164			/*
1165			 * don't request sense if there's no check condition
1166			 * status because the error we're processing isn't one
1167			 * that has a sense code (and some devices get
1168			 * confused by sense requests out of the blue)
1169			 */
1170			continue;
1171
1172		SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1173						  "%s: requesting sense\n",
1174						  current->comm));
1175		rtn = scsi_request_sense(scmd);
1176		if (rtn != SUCCESS)
1177			continue;
1178
1179		SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
1180						  " result %x\n", scmd,
1181						  scmd->result));
1182		SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
1183
1184		rtn = scsi_decide_disposition(scmd);
1185
1186		/*
1187		 * if the result was normal, then just pass it along to the
1188		 * upper level.
1189		 */
1190		if (rtn == SUCCESS)
1191			/* we don't want this command reissued, just
1192			 * finished with the sense data, so set
1193			 * retries to the max allowed to ensure it
1194			 * won't get reissued */
1195			scmd->retries = scmd->allowed;
 
 
 
 
 
 
 
1196		else if (rtn != NEEDS_RETRY)
1197			continue;
1198
1199		scsi_eh_finish_cmd(scmd, done_q);
1200	}
1201
1202	return list_empty(work_q);
1203}
1204EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1205
1206/**
1207 * scsi_eh_tur - Send TUR to device.
1208 * @scmd:	&scsi_cmnd to send TUR
1209 *
1210 * Return value:
1211 *    0 - Device is ready. 1 - Device NOT ready.
1212 */
1213static int scsi_eh_tur(struct scsi_cmnd *scmd)
1214{
1215	static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1216	int retry_cnt = 1, rtn;
 
1217
1218retry_tur:
1219	rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1220				scmd->device->eh_timeout, 0);
1221
1222	SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
1223		__func__, scmd, rtn));
1224
1225	switch (rtn) {
1226	case NEEDS_RETRY:
1227		if (retry_cnt--)
1228			goto retry_tur;
1229		/*FALLTHRU*/
1230	case SUCCESS:
1231		return 0;
1232	default:
1233		return 1;
1234	}
1235}
1236
1237/**
1238 * scsi_eh_test_devices - check if devices are responding from error recovery.
1239 * @cmd_list:	scsi commands in error recovery.
1240 * @work_q:     queue for commands which still need more error recovery
1241 * @done_q:     queue for commands which are finished
1242 * @try_stu:    boolean on if a STU command should be tried in addition to TUR.
1243 *
1244 * Decription:
1245 *    Tests if devices are in a working state.  Commands to devices now in
1246 *    a working state are sent to the done_q while commands to devices which
1247 *    are still failing to respond are returned to the work_q for more
1248 *    processing.
1249 **/
1250static int scsi_eh_test_devices(struct list_head *cmd_list,
1251				struct list_head *work_q,
1252				struct list_head *done_q, int try_stu)
1253{
1254	struct scsi_cmnd *scmd, *next;
1255	struct scsi_device *sdev;
1256	int finish_cmds;
1257
1258	while (!list_empty(cmd_list)) {
1259		scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1260		sdev = scmd->device;
1261
1262		if (!try_stu) {
1263			if (scsi_host_eh_past_deadline(sdev->host)) {
1264				/* Push items back onto work_q */
1265				list_splice_init(cmd_list, work_q);
1266				SCSI_LOG_ERROR_RECOVERY(3,
1267					shost_printk(KERN_INFO, sdev->host,
1268						     "skip %s, past eh deadline",
1269						     __func__));
1270				break;
1271			}
1272		}
1273
1274		finish_cmds = !scsi_device_online(scmd->device) ||
1275			(try_stu && !scsi_eh_try_stu(scmd) &&
1276			 !scsi_eh_tur(scmd)) ||
1277			!scsi_eh_tur(scmd);
1278
1279		list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1280			if (scmd->device == sdev) {
1281				if (finish_cmds &&
1282				    (try_stu ||
1283				     scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1284					scsi_eh_finish_cmd(scmd, done_q);
1285				else
1286					list_move_tail(&scmd->eh_entry, work_q);
1287			}
1288	}
1289	return list_empty(work_q);
1290}
1291
1292
1293/**
1294 * scsi_eh_abort_cmds - abort pending commands.
1295 * @work_q:	&list_head for pending commands.
1296 * @done_q:	&list_head for processed commands.
1297 *
1298 * Decription:
1299 *    Try and see whether or not it makes sense to try and abort the
1300 *    running command.  This only works out to be the case if we have one
1301 *    command that has timed out.  If the command simply failed, it makes
1302 *    no sense to try and abort the command, since as far as the shost
1303 *    adapter is concerned, it isn't running.
1304 */
1305static int scsi_eh_abort_cmds(struct list_head *work_q,
1306			      struct list_head *done_q)
1307{
1308	struct scsi_cmnd *scmd, *next;
1309	LIST_HEAD(check_list);
1310	int rtn;
1311	struct Scsi_Host *shost;
1312
1313	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1314		if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1315			continue;
1316		shost = scmd->device->host;
1317		if (scsi_host_eh_past_deadline(shost)) {
1318			list_splice_init(&check_list, work_q);
1319			SCSI_LOG_ERROR_RECOVERY(3,
1320				shost_printk(KERN_INFO, shost,
1321					    "skip %s, past eh deadline\n",
1322					     __func__));
1323			return list_empty(work_q);
1324		}
1325		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1326						  "0x%p\n", current->comm,
1327						  scmd));
1328		rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
1329		if (rtn == FAILED) {
1330			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1331							  " cmd failed:"
1332							  "0x%p\n",
1333							  current->comm,
1334							  scmd));
1335			list_splice_init(&check_list, work_q);
1336			return list_empty(work_q);
1337		}
1338		scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1339		if (rtn == FAST_IO_FAIL)
1340			scsi_eh_finish_cmd(scmd, done_q);
1341		else
1342			list_move_tail(&scmd->eh_entry, &check_list);
1343	}
1344
1345	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1346}
1347
1348/**
1349 * scsi_eh_try_stu - Send START_UNIT to device.
1350 * @scmd:	&scsi_cmnd to send START_UNIT
1351 *
1352 * Return value:
1353 *    0 - Device is ready. 1 - Device NOT ready.
1354 */
1355static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1356{
1357	static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1358
1359	if (scmd->device->allow_restart) {
1360		int i, rtn = NEEDS_RETRY;
 
1361
1362		for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1363			rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1364
1365		if (rtn == SUCCESS)
1366			return 0;
1367	}
1368
1369	return 1;
1370}
1371
1372 /**
1373 * scsi_eh_stu - send START_UNIT if needed
1374 * @shost:	&scsi host being recovered.
1375 * @work_q:     &list_head for pending commands.
1376 * @done_q:	&list_head for processed commands.
1377 *
1378 * Notes:
1379 *    If commands are failing due to not ready, initializing command required,
1380 *	try revalidating the device, which will end up sending a start unit.
1381 */
1382static int scsi_eh_stu(struct Scsi_Host *shost,
1383			      struct list_head *work_q,
1384			      struct list_head *done_q)
1385{
1386	struct scsi_cmnd *scmd, *stu_scmd, *next;
1387	struct scsi_device *sdev;
1388
1389	shost_for_each_device(sdev, shost) {
1390		if (scsi_host_eh_past_deadline(shost)) {
1391			SCSI_LOG_ERROR_RECOVERY(3,
1392				shost_printk(KERN_INFO, shost,
1393					    "skip %s, past eh deadline\n",
1394					     __func__));
 
1395			break;
1396		}
1397		stu_scmd = NULL;
1398		list_for_each_entry(scmd, work_q, eh_entry)
1399			if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1400			    scsi_check_sense(scmd) == FAILED ) {
1401				stu_scmd = scmd;
1402				break;
1403			}
1404
1405		if (!stu_scmd)
1406			continue;
1407
1408		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1409						  " 0x%p\n", current->comm, sdev));
 
 
1410
1411		if (!scsi_eh_try_stu(stu_scmd)) {
1412			if (!scsi_device_online(sdev) ||
1413			    !scsi_eh_tur(stu_scmd)) {
1414				list_for_each_entry_safe(scmd, next,
1415							  work_q, eh_entry) {
1416					if (scmd->device == sdev &&
1417					    scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1418						scsi_eh_finish_cmd(scmd, done_q);
1419				}
1420			}
1421		} else {
1422			SCSI_LOG_ERROR_RECOVERY(3,
1423						printk("%s: START_UNIT failed to sdev:"
1424						       " 0x%p\n", current->comm, sdev));
 
1425		}
1426	}
1427
1428	return list_empty(work_q);
1429}
1430
1431
1432/**
1433 * scsi_eh_bus_device_reset - send bdr if needed
1434 * @shost:	scsi host being recovered.
1435 * @work_q:     &list_head for pending commands.
1436 * @done_q:	&list_head for processed commands.
1437 *
1438 * Notes:
1439 *    Try a bus device reset.  Still, look to see whether we have multiple
1440 *    devices that are jammed or not - if we have multiple devices, it
1441 *    makes no sense to try bus_device_reset - we really would need to try
1442 *    a bus_reset instead.
1443 */
1444static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1445				    struct list_head *work_q,
1446				    struct list_head *done_q)
1447{
1448	struct scsi_cmnd *scmd, *bdr_scmd, *next;
1449	struct scsi_device *sdev;
1450	int rtn;
1451
1452	shost_for_each_device(sdev, shost) {
1453		if (scsi_host_eh_past_deadline(shost)) {
1454			SCSI_LOG_ERROR_RECOVERY(3,
1455				shost_printk(KERN_INFO, shost,
1456					    "skip %s, past eh deadline\n",
1457					     __func__));
 
1458			break;
1459		}
1460		bdr_scmd = NULL;
1461		list_for_each_entry(scmd, work_q, eh_entry)
1462			if (scmd->device == sdev) {
1463				bdr_scmd = scmd;
1464				break;
1465			}
1466
1467		if (!bdr_scmd)
1468			continue;
1469
1470		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1471						  " 0x%p\n", current->comm,
1472						  sdev));
1473		rtn = scsi_try_bus_device_reset(bdr_scmd);
1474		if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1475			if (!scsi_device_online(sdev) ||
1476			    rtn == FAST_IO_FAIL ||
1477			    !scsi_eh_tur(bdr_scmd)) {
1478				list_for_each_entry_safe(scmd, next,
1479							 work_q, eh_entry) {
1480					if (scmd->device == sdev &&
1481					    scsi_eh_action(scmd, rtn) != FAILED)
1482						scsi_eh_finish_cmd(scmd,
1483								   done_q);
1484				}
1485			}
1486		} else {
1487			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1488							  " failed sdev:"
1489							  "0x%p\n",
1490							  current->comm,
1491							   sdev));
1492		}
1493	}
1494
1495	return list_empty(work_q);
1496}
1497
1498/**
1499 * scsi_eh_target_reset - send target reset if needed
1500 * @shost:	scsi host being recovered.
1501 * @work_q:     &list_head for pending commands.
1502 * @done_q:	&list_head for processed commands.
1503 *
1504 * Notes:
1505 *    Try a target reset.
1506 */
1507static int scsi_eh_target_reset(struct Scsi_Host *shost,
1508				struct list_head *work_q,
1509				struct list_head *done_q)
1510{
1511	LIST_HEAD(tmp_list);
1512	LIST_HEAD(check_list);
1513
1514	list_splice_init(work_q, &tmp_list);
1515
1516	while (!list_empty(&tmp_list)) {
1517		struct scsi_cmnd *next, *scmd;
1518		int rtn;
1519		unsigned int id;
1520
1521		if (scsi_host_eh_past_deadline(shost)) {
1522			/* push back on work queue for further processing */
1523			list_splice_init(&check_list, work_q);
1524			list_splice_init(&tmp_list, work_q);
1525			SCSI_LOG_ERROR_RECOVERY(3,
1526				shost_printk(KERN_INFO, shost,
1527					    "skip %s, past eh deadline\n",
1528					     __func__));
1529			return list_empty(work_q);
1530		}
1531
1532		scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1533		id = scmd_id(scmd);
1534
1535		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1536						  "to target %d\n",
1537						  current->comm, id));
 
1538		rtn = scsi_try_target_reset(scmd);
1539		if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1540			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1541							  " failed target: "
1542							  "%d\n",
1543							  current->comm, id));
 
1544		list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1545			if (scmd_id(scmd) != id)
1546				continue;
1547
1548			if (rtn == SUCCESS)
1549				list_move_tail(&scmd->eh_entry, &check_list);
1550			else if (rtn == FAST_IO_FAIL)
1551				scsi_eh_finish_cmd(scmd, done_q);
1552			else
1553				/* push back on work queue for further processing */
1554				list_move(&scmd->eh_entry, work_q);
1555		}
1556	}
1557
1558	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1559}
1560
1561/**
1562 * scsi_eh_bus_reset - send a bus reset
1563 * @shost:	&scsi host being recovered.
1564 * @work_q:     &list_head for pending commands.
1565 * @done_q:	&list_head for processed commands.
1566 */
1567static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1568			     struct list_head *work_q,
1569			     struct list_head *done_q)
1570{
1571	struct scsi_cmnd *scmd, *chan_scmd, *next;
1572	LIST_HEAD(check_list);
1573	unsigned int channel;
1574	int rtn;
1575
1576	/*
1577	 * we really want to loop over the various channels, and do this on
1578	 * a channel by channel basis.  we should also check to see if any
1579	 * of the failed commands are on soft_reset devices, and if so, skip
1580	 * the reset.
1581	 */
1582
1583	for (channel = 0; channel <= shost->max_channel; channel++) {
1584		if (scsi_host_eh_past_deadline(shost)) {
1585			list_splice_init(&check_list, work_q);
1586			SCSI_LOG_ERROR_RECOVERY(3,
1587				shost_printk(KERN_INFO, shost,
1588					    "skip %s, past eh deadline\n",
1589					     __func__));
1590			return list_empty(work_q);
1591		}
1592
1593		chan_scmd = NULL;
1594		list_for_each_entry(scmd, work_q, eh_entry) {
1595			if (channel == scmd_channel(scmd)) {
1596				chan_scmd = scmd;
1597				break;
1598				/*
1599				 * FIXME add back in some support for
1600				 * soft_reset devices.
1601				 */
1602			}
1603		}
1604
1605		if (!chan_scmd)
1606			continue;
1607		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1608						  " %d\n", current->comm,
1609						  channel));
 
1610		rtn = scsi_try_bus_reset(chan_scmd);
1611		if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1612			list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1613				if (channel == scmd_channel(scmd)) {
1614					if (rtn == FAST_IO_FAIL)
1615						scsi_eh_finish_cmd(scmd,
1616								   done_q);
1617					else
1618						list_move_tail(&scmd->eh_entry,
1619							       &check_list);
1620				}
1621			}
1622		} else {
1623			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1624							  " failed chan: %d\n",
1625							  current->comm,
1626							  channel));
1627		}
1628	}
1629	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1630}
1631
1632/**
1633 * scsi_eh_host_reset - send a host reset
1634 * @work_q:	list_head for processed commands.
1635 * @done_q:	list_head for processed commands.
 
1636 */
1637static int scsi_eh_host_reset(struct list_head *work_q,
 
1638			      struct list_head *done_q)
1639{
1640	struct scsi_cmnd *scmd, *next;
1641	LIST_HEAD(check_list);
1642	int rtn;
1643
1644	if (!list_empty(work_q)) {
1645		scmd = list_entry(work_q->next,
1646				  struct scsi_cmnd, eh_entry);
1647
1648		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1649						  , current->comm));
 
 
1650
1651		rtn = scsi_try_host_reset(scmd);
1652		if (rtn == SUCCESS) {
1653			list_splice_init(work_q, &check_list);
1654		} else if (rtn == FAST_IO_FAIL) {
1655			list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1656					scsi_eh_finish_cmd(scmd, done_q);
1657			}
1658		} else {
1659			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1660							  " failed\n",
1661							  current->comm));
 
1662		}
1663	}
1664	return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1665}
1666
1667/**
1668 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1669 * @work_q:	list_head for processed commands.
1670 * @done_q:	list_head for processed commands.
1671 */
1672static void scsi_eh_offline_sdevs(struct list_head *work_q,
1673				  struct list_head *done_q)
1674{
1675	struct scsi_cmnd *scmd, *next;
 
1676
1677	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1678		sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1679			    "not ready after error recovery\n");
1680		scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1681		if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1682			/*
1683			 * FIXME: Handle lost cmds.
1684			 */
1685		}
1686		scsi_eh_finish_cmd(scmd, done_q);
1687	}
1688	return;
1689}
1690
1691/**
1692 * scsi_noretry_cmd - determine if command should be failed fast
1693 * @scmd:	SCSI cmd to examine.
1694 */
1695int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1696{
1697	switch (host_byte(scmd->result)) {
1698	case DID_OK:
1699		break;
1700	case DID_TIME_OUT:
1701		goto check_type;
1702	case DID_BUS_BUSY:
1703		return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1704	case DID_PARITY:
1705		return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1706	case DID_ERROR:
1707		if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1708		    status_byte(scmd->result) == RESERVATION_CONFLICT)
1709			return 0;
1710		/* fall through */
1711	case DID_SOFT_ERROR:
1712		return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1713	}
1714
1715	if (status_byte(scmd->result) != CHECK_CONDITION)
1716		return 0;
1717
1718check_type:
1719	/*
1720	 * assume caller has checked sense and determined
1721	 * the check condition was retryable.
1722	 */
1723	if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1724	    scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1725		return 1;
1726	else
1727		return 0;
1728}
1729
1730/**
1731 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1732 * @scmd:	SCSI cmd to examine.
1733 *
1734 * Notes:
1735 *    This is *only* called when we are examining the status after sending
1736 *    out the actual data command.  any commands that are queued for error
1737 *    recovery (e.g. test_unit_ready) do *not* come through here.
1738 *
1739 *    When this routine returns failed, it means the error handler thread
1740 *    is woken.  In cases where the error code indicates an error that
1741 *    doesn't require the error handler read (i.e. we don't need to
1742 *    abort/reset), this function should return SUCCESS.
1743 */
1744int scsi_decide_disposition(struct scsi_cmnd *scmd)
1745{
1746	int rtn;
1747
1748	/*
1749	 * if the device is offline, then we clearly just pass the result back
1750	 * up to the top level.
1751	 */
1752	if (!scsi_device_online(scmd->device)) {
1753		SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1754						  " as SUCCESS\n",
1755						  __func__));
1756		return SUCCESS;
1757	}
1758
1759	/*
1760	 * first check the host byte, to see if there is anything in there
1761	 * that would indicate what we need to do.
1762	 */
1763	switch (host_byte(scmd->result)) {
1764	case DID_PASSTHROUGH:
1765		/*
1766		 * no matter what, pass this through to the upper layer.
1767		 * nuke this special code so that it looks like we are saying
1768		 * did_ok.
1769		 */
1770		scmd->result &= 0xff00ffff;
1771		return SUCCESS;
1772	case DID_OK:
1773		/*
1774		 * looks good.  drop through, and check the next byte.
1775		 */
1776		break;
1777	case DID_ABORT:
1778		if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1779			scmd->result |= DID_TIME_OUT << 16;
1780			return SUCCESS;
1781		}
 
1782	case DID_NO_CONNECT:
1783	case DID_BAD_TARGET:
1784		/*
1785		 * note - this means that we just report the status back
1786		 * to the top level driver, not that we actually think
1787		 * that it indicates SUCCESS.
1788		 */
1789		return SUCCESS;
 
1790		/*
1791		 * when the low level driver returns did_soft_error,
1792		 * it is responsible for keeping an internal retry counter
1793		 * in order to avoid endless loops (db)
1794		 *
1795		 * actually this is a bug in this function here.  we should
1796		 * be mindful of the maximum number of retries specified
1797		 * and not get stuck in a loop.
1798		 */
1799	case DID_SOFT_ERROR:
1800		goto maybe_retry;
1801	case DID_IMM_RETRY:
1802		return NEEDS_RETRY;
1803
1804	case DID_REQUEUE:
1805		return ADD_TO_MLQUEUE;
1806	case DID_TRANSPORT_DISRUPTED:
1807		/*
1808		 * LLD/transport was disrupted during processing of the IO.
1809		 * The transport class is now blocked/blocking,
1810		 * and the transport will decide what to do with the IO
1811		 * based on its timers and recovery capablilities if
1812		 * there are enough retries.
1813		 */
1814		goto maybe_retry;
1815	case DID_TRANSPORT_FAILFAST:
1816		/*
1817		 * The transport decided to failfast the IO (most likely
1818		 * the fast io fail tmo fired), so send IO directly upwards.
1819		 */
1820		return SUCCESS;
 
 
 
 
 
 
1821	case DID_ERROR:
1822		if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1823		    status_byte(scmd->result) == RESERVATION_CONFLICT)
1824			/*
1825			 * execute reservation conflict processing code
1826			 * lower down
1827			 */
1828			break;
1829		/* fallthrough */
1830	case DID_BUS_BUSY:
1831	case DID_PARITY:
1832		goto maybe_retry;
1833	case DID_TIME_OUT:
1834		/*
1835		 * when we scan the bus, we get timeout messages for
1836		 * these commands if there is no device available.
1837		 * other hosts report did_no_connect for the same thing.
1838		 */
1839		if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1840		     scmd->cmnd[0] == INQUIRY)) {
1841			return SUCCESS;
1842		} else {
1843			return FAILED;
1844		}
1845	case DID_RESET:
1846		return SUCCESS;
1847	default:
1848		return FAILED;
1849	}
1850
1851	/*
1852	 * next, check the message byte.
1853	 */
1854	if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1855		return FAILED;
1856
1857	/*
1858	 * check the status byte to see if this indicates anything special.
1859	 */
1860	switch (status_byte(scmd->result)) {
1861	case QUEUE_FULL:
1862		scsi_handle_queue_full(scmd->device);
1863		/*
1864		 * the case of trying to send too many commands to a
1865		 * tagged queueing device.
1866		 */
1867	case BUSY:
 
1868		/*
1869		 * device can't talk to us at the moment.  Should only
1870		 * occur (SAM-3) when the task queue is empty, so will cause
1871		 * the empty queue handling to trigger a stall in the
1872		 * device.
1873		 */
1874		return ADD_TO_MLQUEUE;
1875	case GOOD:
1876		if (scmd->cmnd[0] == REPORT_LUNS)
1877			scmd->device->sdev_target->expecting_lun_change = 0;
1878		scsi_handle_queue_ramp_up(scmd->device);
1879	case COMMAND_TERMINATED:
 
1880		return SUCCESS;
1881	case TASK_ABORTED:
1882		goto maybe_retry;
1883	case CHECK_CONDITION:
1884		rtn = scsi_check_sense(scmd);
1885		if (rtn == NEEDS_RETRY)
1886			goto maybe_retry;
1887		/* if rtn == FAILED, we have no sense information;
1888		 * returning FAILED will wake the error handler thread
1889		 * to collect the sense and redo the decide
1890		 * disposition */
1891		return rtn;
1892	case CONDITION_GOOD:
1893	case INTERMEDIATE_GOOD:
1894	case INTERMEDIATE_C_GOOD:
1895	case ACA_ACTIVE:
1896		/*
1897		 * who knows?  FIXME(eric)
1898		 */
1899		return SUCCESS;
1900
1901	case RESERVATION_CONFLICT:
1902		sdev_printk(KERN_INFO, scmd->device,
1903			    "reservation conflict\n");
1904		set_host_byte(scmd, DID_NEXUS_FAILURE);
1905		return SUCCESS; /* causes immediate i/o error */
1906	default:
1907		return FAILED;
1908	}
1909	return FAILED;
1910
1911      maybe_retry:
1912
1913	/* we requeue for retry because the error was retryable, and
1914	 * the request was not marked fast fail.  Note that above,
1915	 * even if the request is marked fast fail, we still requeue
1916	 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1917	if ((++scmd->retries) <= scmd->allowed
1918	    && !scsi_noretry_cmd(scmd)) {
1919		return NEEDS_RETRY;
1920	} else {
1921		/*
1922		 * no more retries - report this one back to upper level.
1923		 */
1924		return SUCCESS;
1925	}
1926}
1927
1928static void eh_lock_door_done(struct request *req, int uptodate)
1929{
1930	__blk_put_request(req->q, req);
1931}
1932
1933/**
1934 * scsi_eh_lock_door - Prevent medium removal for the specified device
1935 * @sdev:	SCSI device to prevent medium removal
1936 *
1937 * Locking:
1938 * 	We must be called from process context.
1939 *
1940 * Notes:
1941 * 	We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1942 * 	head of the devices request queue, and continue.
1943 */
1944static void scsi_eh_lock_door(struct scsi_device *sdev)
1945{
1946	struct request *req;
 
1947
1948	/*
1949	 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1950	 * request becomes available
1951	 */
1952	req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1953
1954	req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1955	req->cmd[1] = 0;
1956	req->cmd[2] = 0;
1957	req->cmd[3] = 0;
1958	req->cmd[4] = SCSI_REMOVAL_PREVENT;
1959	req->cmd[5] = 0;
1960
1961	req->cmd_len = COMMAND_SIZE(req->cmd[0]);
 
 
 
 
 
 
1962
1963	req->cmd_type = REQ_TYPE_BLOCK_PC;
1964	req->cmd_flags |= REQ_QUIET;
1965	req->timeout = 10 * HZ;
1966	req->retries = 5;
1967
1968	blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1969}
1970
1971/**
1972 * scsi_restart_operations - restart io operations to the specified host.
1973 * @shost:	Host we are restarting.
1974 *
1975 * Notes:
1976 *    When we entered the error handler, we blocked all further i/o to
1977 *    this device.  we need to 'reverse' this process.
1978 */
1979static void scsi_restart_operations(struct Scsi_Host *shost)
1980{
1981	struct scsi_device *sdev;
1982	unsigned long flags;
1983
1984	/*
1985	 * If the door was locked, we need to insert a door lock request
1986	 * onto the head of the SCSI request queue for the device.  There
1987	 * is no point trying to lock the door of an off-line device.
1988	 */
1989	shost_for_each_device(sdev, shost) {
1990		if (scsi_device_online(sdev) && sdev->locked)
1991			scsi_eh_lock_door(sdev);
 
 
1992	}
1993
1994	/*
1995	 * next free up anything directly waiting upon the host.  this
1996	 * will be requests for character device operations, and also for
1997	 * ioctls to queued block devices.
1998	 */
1999	SCSI_LOG_ERROR_RECOVERY(3,
2000		printk("scsi_eh_%d waking up host to restart\n",
2001		       shost->host_no));
2002
2003	spin_lock_irqsave(shost->host_lock, flags);
2004	if (scsi_host_set_state(shost, SHOST_RUNNING))
2005		if (scsi_host_set_state(shost, SHOST_CANCEL))
2006			BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2007	spin_unlock_irqrestore(shost->host_lock, flags);
2008
2009	wake_up(&shost->host_wait);
2010
2011	/*
2012	 * finally we need to re-initiate requests that may be pending.  we will
2013	 * have had everything blocked while error handling is taking place, and
2014	 * now that error recovery is done, we will need to ensure that these
2015	 * requests are started.
2016	 */
2017	scsi_run_host_queues(shost);
2018
2019	/*
2020	 * if eh is active and host_eh_scheduled is pending we need to re-run
2021	 * recovery.  we do this check after scsi_run_host_queues() to allow
2022	 * everything pent up since the last eh run a chance to make forward
2023	 * progress before we sync again.  Either we'll immediately re-run
2024	 * recovery or scsi_device_unbusy() will wake us again when these
2025	 * pending commands complete.
2026	 */
2027	spin_lock_irqsave(shost->host_lock, flags);
2028	if (shost->host_eh_scheduled)
2029		if (scsi_host_set_state(shost, SHOST_RECOVERY))
2030			WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2031	spin_unlock_irqrestore(shost->host_lock, flags);
2032}
2033
2034/**
2035 * scsi_eh_ready_devs - check device ready state and recover if not.
2036 * @shost: 	host to be recovered.
2037 * @work_q:     &list_head for pending commands.
2038 * @done_q:	&list_head for processed commands.
2039 */
2040void scsi_eh_ready_devs(struct Scsi_Host *shost,
2041			struct list_head *work_q,
2042			struct list_head *done_q)
2043{
2044	if (!scsi_eh_stu(shost, work_q, done_q))
2045		if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2046			if (!scsi_eh_target_reset(shost, work_q, done_q))
2047				if (!scsi_eh_bus_reset(shost, work_q, done_q))
2048					if (!scsi_eh_host_reset(work_q, done_q))
2049						scsi_eh_offline_sdevs(work_q,
2050								      done_q);
2051}
2052EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2053
2054/**
2055 * scsi_eh_flush_done_q - finish processed commands or retry them.
2056 * @done_q:	list_head of processed commands.
2057 */
2058void scsi_eh_flush_done_q(struct list_head *done_q)
2059{
2060	struct scsi_cmnd *scmd, *next;
2061
2062	list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2063		list_del_init(&scmd->eh_entry);
2064		if (scsi_device_online(scmd->device) &&
2065		    !scsi_noretry_cmd(scmd) &&
2066		    (++scmd->retries <= scmd->allowed)) {
2067			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
2068							  " retry cmd: %p\n",
2069							  current->comm,
2070							  scmd));
2071				scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2072		} else {
2073			/*
2074			 * If just we got sense for the device (called
2075			 * scsi_eh_get_sense), scmd->result is already
2076			 * set, do not set DRIVER_TIMEOUT.
2077			 */
2078			if (!scmd->result)
2079				scmd->result |= (DRIVER_TIMEOUT << 24);
2080			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
2081							" cmd: %p\n",
2082							current->comm, scmd));
 
2083			scsi_finish_command(scmd);
2084		}
2085	}
2086}
2087EXPORT_SYMBOL(scsi_eh_flush_done_q);
2088
2089/**
2090 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2091 * @shost:	Host to unjam.
2092 *
2093 * Notes:
2094 *    When we come in here, we *know* that all commands on the bus have
2095 *    either completed, failed or timed out.  we also know that no further
2096 *    commands are being sent to the host, so things are relatively quiet
2097 *    and we have freedom to fiddle with things as we wish.
2098 *
2099 *    This is only the *default* implementation.  it is possible for
2100 *    individual drivers to supply their own version of this function, and
2101 *    if the maintainer wishes to do this, it is strongly suggested that
2102 *    this function be taken as a template and modified.  this function
2103 *    was designed to correctly handle problems for about 95% of the
2104 *    different cases out there, and it should always provide at least a
2105 *    reasonable amount of error recovery.
2106 *
2107 *    Any command marked 'failed' or 'timeout' must eventually have
2108 *    scsi_finish_cmd() called for it.  we do all of the retry stuff
2109 *    here, so when we restart the host after we return it should have an
2110 *    empty queue.
2111 */
2112static void scsi_unjam_host(struct Scsi_Host *shost)
2113{
2114	unsigned long flags;
2115	LIST_HEAD(eh_work_q);
2116	LIST_HEAD(eh_done_q);
2117
2118	spin_lock_irqsave(shost->host_lock, flags);
2119	list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2120	spin_unlock_irqrestore(shost->host_lock, flags);
2121
2122	SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2123
2124	if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2125		if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2126			scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2127
2128	spin_lock_irqsave(shost->host_lock, flags);
2129	if (shost->eh_deadline != -1)
2130		shost->last_reset = 0;
2131	spin_unlock_irqrestore(shost->host_lock, flags);
2132	scsi_eh_flush_done_q(&eh_done_q);
2133}
2134
2135/**
2136 * scsi_error_handler - SCSI error handler thread
2137 * @data:	Host for which we are running.
2138 *
2139 * Notes:
2140 *    This is the main error handling loop.  This is run as a kernel thread
2141 *    for every SCSI host and handles all error handling activity.
2142 */
2143int scsi_error_handler(void *data)
2144{
2145	struct Scsi_Host *shost = data;
2146
2147	/*
2148	 * We use TASK_INTERRUPTIBLE so that the thread is not
2149	 * counted against the load average as a running process.
2150	 * We never actually get interrupted because kthread_run
2151	 * disables signal delivery for the created thread.
2152	 */
2153	while (!kthread_should_stop()) {
 
 
 
 
 
 
2154		set_current_state(TASK_INTERRUPTIBLE);
 
 
 
2155		if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2156		    shost->host_failed != shost->host_busy) {
2157			SCSI_LOG_ERROR_RECOVERY(1,
2158				printk("scsi_eh_%d: sleeping\n",
2159					shost->host_no));
 
2160			schedule();
2161			continue;
2162		}
2163
2164		__set_current_state(TASK_RUNNING);
2165		SCSI_LOG_ERROR_RECOVERY(1,
2166			printk("scsi_eh_%d: waking up %d/%d/%d\n",
2167			       shost->host_no, shost->host_eh_scheduled,
2168			       shost->host_failed, shost->host_busy));
 
 
2169
2170		/*
2171		 * We have a host that is failing for some reason.  Figure out
2172		 * what we need to do to get it up and online again (if we can).
2173		 * If we fail, we end up taking the thing offline.
2174		 */
2175		if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2176			SCSI_LOG_ERROR_RECOVERY(1,
2177				printk(KERN_ERR "Error handler scsi_eh_%d "
2178						"unable to autoresume\n",
2179						shost->host_no));
2180			continue;
2181		}
2182
2183		if (shost->transportt->eh_strategy_handler)
2184			shost->transportt->eh_strategy_handler(shost);
2185		else
2186			scsi_unjam_host(shost);
2187
 
 
 
2188		/*
2189		 * Note - if the above fails completely, the action is to take
2190		 * individual devices offline and flush the queue of any
2191		 * outstanding requests that may have been pending.  When we
2192		 * restart, we restart any I/O to any other devices on the bus
2193		 * which are still online.
2194		 */
2195		scsi_restart_operations(shost);
2196		if (!shost->eh_noresume)
2197			scsi_autopm_put_host(shost);
2198	}
2199	__set_current_state(TASK_RUNNING);
2200
2201	SCSI_LOG_ERROR_RECOVERY(1,
2202		printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
 
 
2203	shost->ehandler = NULL;
2204	return 0;
2205}
2206
2207/*
2208 * Function:    scsi_report_bus_reset()
2209 *
2210 * Purpose:     Utility function used by low-level drivers to report that
2211 *		they have observed a bus reset on the bus being handled.
2212 *
2213 * Arguments:   shost       - Host in question
2214 *		channel     - channel on which reset was observed.
2215 *
2216 * Returns:     Nothing
2217 *
2218 * Lock status: Host lock must be held.
2219 *
2220 * Notes:       This only needs to be called if the reset is one which
2221 *		originates from an unknown location.  Resets originated
2222 *		by the mid-level itself don't need to call this, but there
2223 *		should be no harm.
2224 *
2225 *		The main purpose of this is to make sure that a CHECK_CONDITION
2226 *		is properly treated.
2227 */
2228void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2229{
2230	struct scsi_device *sdev;
2231
2232	__shost_for_each_device(sdev, shost) {
2233		if (channel == sdev_channel(sdev))
2234			__scsi_report_device_reset(sdev, NULL);
2235	}
2236}
2237EXPORT_SYMBOL(scsi_report_bus_reset);
2238
2239/*
2240 * Function:    scsi_report_device_reset()
2241 *
2242 * Purpose:     Utility function used by low-level drivers to report that
2243 *		they have observed a device reset on the device being handled.
2244 *
2245 * Arguments:   shost       - Host in question
2246 *		channel     - channel on which reset was observed
2247 *		target	    - target on which reset was observed
2248 *
2249 * Returns:     Nothing
2250 *
2251 * Lock status: Host lock must be held
2252 *
2253 * Notes:       This only needs to be called if the reset is one which
2254 *		originates from an unknown location.  Resets originated
2255 *		by the mid-level itself don't need to call this, but there
2256 *		should be no harm.
2257 *
2258 *		The main purpose of this is to make sure that a CHECK_CONDITION
2259 *		is properly treated.
2260 */
2261void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2262{
2263	struct scsi_device *sdev;
2264
2265	__shost_for_each_device(sdev, shost) {
2266		if (channel == sdev_channel(sdev) &&
2267		    target == sdev_id(sdev))
2268			__scsi_report_device_reset(sdev, NULL);
2269	}
2270}
2271EXPORT_SYMBOL(scsi_report_device_reset);
2272
2273static void
2274scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2275{
2276}
2277
2278/*
2279 * Function:	scsi_reset_provider
2280 *
2281 * Purpose:	Send requested reset to a bus or device at any phase.
2282 *
2283 * Arguments:	device	- device to send reset to
2284 *		flag - reset type (see scsi.h)
2285 *
2286 * Returns:	SUCCESS/FAILURE.
2287 *
2288 * Notes:	This is used by the SCSI Generic driver to provide
2289 *		Bus/Device reset capability.
2290 */
2291int
2292scsi_reset_provider(struct scsi_device *dev, int flag)
2293{
2294	struct scsi_cmnd *scmd;
2295	struct Scsi_Host *shost = dev->host;
2296	struct request req;
2297	unsigned long flags;
2298	int rtn;
 
 
 
 
 
 
 
 
2299
2300	if (scsi_autopm_get_host(shost) < 0)
2301		return FAILED;
2302
2303	if (!get_device(&dev->sdev_gendev)) {
2304		rtn = FAILED;
 
 
2305		goto out_put_autopm_host;
2306	}
2307
2308	scmd = scsi_get_command(dev, GFP_KERNEL);
2309	blk_rq_init(NULL, &req);
2310	scmd->request = &req;
2311
2312	scmd->cmnd = req.cmd;
 
 
 
2313
2314	scmd->scsi_done		= scsi_reset_provider_done_command;
2315	memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2316
2317	scmd->cmd_len			= 0;
2318
2319	scmd->sc_data_direction		= DMA_BIDIRECTIONAL;
2320
2321	spin_lock_irqsave(shost->host_lock, flags);
2322	shost->tmf_in_progress = 1;
2323	spin_unlock_irqrestore(shost->host_lock, flags);
2324
2325	switch (flag) {
2326	case SCSI_TRY_RESET_DEVICE:
 
 
 
2327		rtn = scsi_try_bus_device_reset(scmd);
2328		if (rtn == SUCCESS)
2329			break;
2330		/* FALLTHROUGH */
2331	case SCSI_TRY_RESET_TARGET:
2332		rtn = scsi_try_target_reset(scmd);
2333		if (rtn == SUCCESS)
2334			break;
2335		/* FALLTHROUGH */
2336	case SCSI_TRY_RESET_BUS:
2337		rtn = scsi_try_bus_reset(scmd);
 
 
 
 
 
2338		if (rtn == SUCCESS)
2339			break;
2340		/* FALLTHROUGH */
2341	case SCSI_TRY_RESET_HOST:
2342		rtn = scsi_try_host_reset(scmd);
2343		break;
2344	default:
2345		rtn = FAILED;
 
2346	}
2347
 
 
2348	spin_lock_irqsave(shost->host_lock, flags);
2349	shost->tmf_in_progress = 0;
2350	spin_unlock_irqrestore(shost->host_lock, flags);
2351
2352	/*
2353	 * be sure to wake up anyone who was sleeping or had their queue
2354	 * suspended while we performed the TMF.
2355	 */
2356	SCSI_LOG_ERROR_RECOVERY(3,
2357		printk("%s: waking up host to restart after TMF\n",
2358		__func__));
2359
2360	wake_up(&shost->host_wait);
 
2361
2362	scsi_run_host_queues(shost);
2363
2364	scsi_next_command(scmd);
2365out_put_autopm_host:
2366	scsi_autopm_put_host(shost);
2367	return rtn;
2368}
2369EXPORT_SYMBOL(scsi_reset_provider);
2370
2371/**
2372 * scsi_normalize_sense - normalize main elements from either fixed or
2373 *			descriptor sense data format into a common format.
2374 *
2375 * @sense_buffer:	byte array containing sense data returned by device
2376 * @sb_len:		number of valid bytes in sense_buffer
2377 * @sshdr:		pointer to instance of structure that common
2378 *			elements are written to.
2379 *
2380 * Notes:
2381 *	The "main elements" from sense data are: response_code, sense_key,
2382 *	asc, ascq and additional_length (only for descriptor format).
2383 *
2384 *	Typically this function can be called after a device has
2385 *	responded to a SCSI command with the CHECK_CONDITION status.
2386 *
2387 * Return value:
2388 *	1 if valid sense data information found, else 0;
2389 */
2390int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2391                         struct scsi_sense_hdr *sshdr)
2392{
2393	if (!sense_buffer || !sb_len)
2394		return 0;
2395
2396	memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2397
2398	sshdr->response_code = (sense_buffer[0] & 0x7f);
2399
2400	if (!scsi_sense_valid(sshdr))
2401		return 0;
2402
2403	if (sshdr->response_code >= 0x72) {
2404		/*
2405		 * descriptor format
2406		 */
2407		if (sb_len > 1)
2408			sshdr->sense_key = (sense_buffer[1] & 0xf);
2409		if (sb_len > 2)
2410			sshdr->asc = sense_buffer[2];
2411		if (sb_len > 3)
2412			sshdr->ascq = sense_buffer[3];
2413		if (sb_len > 7)
2414			sshdr->additional_length = sense_buffer[7];
2415	} else {
2416		/*
2417		 * fixed format
2418		 */
2419		if (sb_len > 2)
2420			sshdr->sense_key = (sense_buffer[2] & 0xf);
2421		if (sb_len > 7) {
2422			sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2423					 sb_len : (sense_buffer[7] + 8);
2424			if (sb_len > 12)
2425				sshdr->asc = sense_buffer[12];
2426			if (sb_len > 13)
2427				sshdr->ascq = sense_buffer[13];
2428		}
2429	}
2430
2431	return 1;
2432}
2433EXPORT_SYMBOL(scsi_normalize_sense);
2434
2435int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2436				 struct scsi_sense_hdr *sshdr)
2437{
2438	return scsi_normalize_sense(cmd->sense_buffer,
2439			SCSI_SENSE_BUFFERSIZE, sshdr);
2440}
2441EXPORT_SYMBOL(scsi_command_normalize_sense);
2442
2443/**
2444 * scsi_sense_desc_find - search for a given descriptor type in	descriptor sense data format.
2445 * @sense_buffer:	byte array of descriptor format sense data
2446 * @sb_len:		number of valid bytes in sense_buffer
2447 * @desc_type:		value of descriptor type to find
2448 *			(e.g. 0 -> information)
2449 *
2450 * Notes:
2451 *	only valid when sense data is in descriptor format
2452 *
2453 * Return value:
2454 *	pointer to start of (first) descriptor if found else NULL
2455 */
2456const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2457				int desc_type)
2458{
2459	int add_sen_len, add_len, desc_len, k;
2460	const u8 * descp;
2461
2462	if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2463		return NULL;
2464	if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2465		return NULL;
2466	add_sen_len = (add_sen_len < (sb_len - 8)) ?
2467			add_sen_len : (sb_len - 8);
2468	descp = &sense_buffer[8];
2469	for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2470		descp += desc_len;
2471		add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2472		desc_len = add_len + 2;
2473		if (descp[0] == desc_type)
2474			return descp;
2475		if (add_len < 0) // short descriptor ??
2476			break;
2477	}
2478	return NULL;
2479}
2480EXPORT_SYMBOL(scsi_sense_desc_find);
2481
2482/**
2483 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2484 * @sense_buffer:	byte array of sense data
2485 * @sb_len:		number of valid bytes in sense_buffer
2486 * @info_out:		pointer to 64 integer where 8 or 4 byte information
2487 *			field will be placed if found.
2488 *
2489 * Return value:
2490 *	1 if information field found, 0 if not found.
2491 */
2492int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2493			    u64 * info_out)
2494{
2495	int j;
2496	const u8 * ucp;
2497	u64 ull;
2498
2499	if (sb_len < 7)
2500		return 0;
2501	switch (sense_buffer[0] & 0x7f) {
2502	case 0x70:
2503	case 0x71:
2504		if (sense_buffer[0] & 0x80) {
2505			*info_out = (sense_buffer[3] << 24) +
2506				    (sense_buffer[4] << 16) +
2507				    (sense_buffer[5] << 8) + sense_buffer[6];
2508			return 1;
2509		} else
2510			return 0;
2511	case 0x72:
2512	case 0x73:
2513		ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2514					   0 /* info desc */);
2515		if (ucp && (0xa == ucp[1])) {
2516			ull = 0;
2517			for (j = 0; j < 8; ++j) {
2518				if (j > 0)
2519					ull <<= 8;
2520				ull |= ucp[4 + j];
2521			}
2522			*info_out = ull;
2523			return 1;
2524		} else
2525			return 0;
2526	default:
2527		return 0;
2528	}
2529}
2530EXPORT_SYMBOL(scsi_get_sense_info_fld);
2531
2532/**
2533 * scsi_build_sense_buffer - build sense data in a buffer
2534 * @desc:	Sense format (non zero == descriptor format,
2535 * 		0 == fixed format)
2536 * @buf:	Where to build sense data
2537 * @key:	Sense key
2538 * @asc:	Additional sense code
2539 * @ascq:	Additional sense code qualifier
2540 *
2541 **/
2542void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2543{
2544	if (desc) {
2545		buf[0] = 0x72;	/* descriptor, current */
2546		buf[1] = key;
2547		buf[2] = asc;
2548		buf[3] = ascq;
2549		buf[7] = 0;
2550	} else {
2551		buf[0] = 0x70;	/* fixed, current */
2552		buf[2] = key;
2553		buf[7] = 0xa;
2554		buf[12] = asc;
2555		buf[13] = ascq;
2556	}
2557}
2558EXPORT_SYMBOL(scsi_build_sense_buffer);