Linux Audio

Check our new training course

Loading...
v3.1
   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
  29#include <scsi/scsi.h>
  30#include <scsi/scsi_cmnd.h>
  31#include <scsi/scsi_dbg.h>
  32#include <scsi/scsi_device.h>
 
  33#include <scsi/scsi_eh.h>
  34#include <scsi/scsi_transport.h>
  35#include <scsi/scsi_host.h>
  36#include <scsi/scsi_ioctl.h>
  37
  38#include "scsi_priv.h"
  39#include "scsi_logging.h"
  40#include "scsi_transport_api.h"
  41
  42#include <trace/events/scsi.h>
  43
 
 
  44#define SENSE_TIMEOUT		(10*HZ)
  45
  46/*
  47 * These should *probably* be handled by the host itself.
  48 * Since it is allowed to sleep, it probably should.
  49 */
  50#define BUS_RESET_SETTLE_TIME   (10)
  51#define HOST_RESET_SETTLE_TIME  (10)
  52
  53static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
  54
  55/* called with shost->host_lock held */
  56void scsi_eh_wakeup(struct Scsi_Host *shost)
  57{
  58	if (shost->host_busy == shost->host_failed) {
  59		trace_scsi_eh_wakeup(shost);
  60		wake_up_process(shost->ehandler);
  61		SCSI_LOG_ERROR_RECOVERY(5,
  62				printk("Waking error handler thread\n"));
  63	}
  64}
  65
  66/**
  67 * scsi_schedule_eh - schedule EH for SCSI host
  68 * @shost:	SCSI host to invoke error handling on.
  69 *
  70 * Schedule SCSI EH without scmd.
  71 */
  72void scsi_schedule_eh(struct Scsi_Host *shost)
  73{
  74	unsigned long flags;
  75
  76	spin_lock_irqsave(shost->host_lock, flags);
  77
  78	if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
  79	    scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
  80		shost->host_eh_scheduled++;
  81		scsi_eh_wakeup(shost);
  82	}
  83
  84	spin_unlock_irqrestore(shost->host_lock, flags);
  85}
  86EXPORT_SYMBOL_GPL(scsi_schedule_eh);
  87
  88/**
  89 * scsi_eh_scmd_add - add scsi cmd to error handling.
  90 * @scmd:	scmd to run eh on.
  91 * @eh_flag:	optional SCSI_EH flag.
  92 *
  93 * Return value:
  94 *	0 on failure.
  95 */
  96int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
  97{
  98	struct Scsi_Host *shost = scmd->device->host;
  99	unsigned long flags;
 100	int ret = 0;
 101
 102	if (!shost->ehandler)
 103		return 0;
 104
 105	spin_lock_irqsave(shost->host_lock, flags);
 106	if (scsi_host_set_state(shost, SHOST_RECOVERY))
 107		if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
 108			goto out_unlock;
 109
 110	ret = 1;
 111	scmd->eh_eflags |= eh_flag;
 112	list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
 113	shost->host_failed++;
 114	scsi_eh_wakeup(shost);
 115 out_unlock:
 116	spin_unlock_irqrestore(shost->host_lock, flags);
 117	return ret;
 118}
 119
 120/**
 121 * scsi_times_out - Timeout function for normal scsi commands.
 122 * @req:	request that is timing out.
 123 *
 124 * Notes:
 125 *     We do not need to lock this.  There is the potential for a race
 126 *     only in that the normal completion handling might run, but if the
 127 *     normal completion function determines that the timer has already
 128 *     fired, then it mustn't do anything.
 129 */
 130enum blk_eh_timer_return scsi_times_out(struct request *req)
 131{
 132	struct scsi_cmnd *scmd = req->special;
 133	enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
 134	struct Scsi_Host *host = scmd->device->host;
 135
 136	trace_scsi_dispatch_cmd_timeout(scmd);
 137	scsi_log_completion(scmd, TIMEOUT_ERROR);
 138
 139	if (host->transportt->eh_timed_out)
 140		rtn = host->transportt->eh_timed_out(scmd);
 141	else if (host->hostt->eh_timed_out)
 142		rtn = host->hostt->eh_timed_out(scmd);
 143
 
 
 144	if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
 145		     !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
 146		scmd->result |= DID_TIME_OUT << 16;
 147		rtn = BLK_EH_HANDLED;
 148	}
 149
 150	return rtn;
 151}
 152
 153/**
 154 * scsi_block_when_processing_errors - Prevent cmds from being queued.
 155 * @sdev:	Device on which we are performing recovery.
 156 *
 157 * Description:
 158 *     We block until the host is out of error recovery, and then check to
 159 *     see whether the host or the device is offline.
 160 *
 161 * Return value:
 162 *     0 when dev was taken offline by error recovery. 1 OK to proceed.
 163 */
 164int scsi_block_when_processing_errors(struct scsi_device *sdev)
 165{
 166	int online;
 167
 168	wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
 169
 170	online = scsi_device_online(sdev);
 171
 172	SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
 173					  online));
 174
 175	return online;
 176}
 177EXPORT_SYMBOL(scsi_block_when_processing_errors);
 178
 179#ifdef CONFIG_SCSI_LOGGING
 180/**
 181 * scsi_eh_prt_fail_stats - Log info on failures.
 182 * @shost:	scsi host being recovered.
 183 * @work_q:	Queue of scsi cmds to process.
 184 */
 185static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
 186					  struct list_head *work_q)
 187{
 188	struct scsi_cmnd *scmd;
 189	struct scsi_device *sdev;
 190	int total_failures = 0;
 191	int cmd_failed = 0;
 192	int cmd_cancel = 0;
 193	int devices_failed = 0;
 194
 195	shost_for_each_device(sdev, shost) {
 196		list_for_each_entry(scmd, work_q, eh_entry) {
 197			if (scmd->device == sdev) {
 198				++total_failures;
 199				if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
 200					++cmd_cancel;
 201				else
 202					++cmd_failed;
 203			}
 204		}
 205
 206		if (cmd_cancel || cmd_failed) {
 207			SCSI_LOG_ERROR_RECOVERY(3,
 208				sdev_printk(KERN_INFO, sdev,
 209					    "%s: cmds failed: %d, cancel: %d\n",
 210					    __func__, cmd_failed,
 211					    cmd_cancel));
 212			cmd_cancel = 0;
 213			cmd_failed = 0;
 214			++devices_failed;
 215		}
 216	}
 217
 218	SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
 219					  " devices require eh work\n",
 220				   total_failures, devices_failed));
 221}
 222#endif
 223
 224/**
 225 * scsi_check_sense - Examine scsi cmd sense
 226 * @scmd:	Cmd to have sense checked.
 227 *
 228 * Return value:
 229 *	SUCCESS or FAILED or NEEDS_RETRY or TARGET_ERROR
 230 *
 231 * Notes:
 232 *	When a deferred error is detected the current command has
 233 *	not been executed and needs retrying.
 234 */
 235static int scsi_check_sense(struct scsi_cmnd *scmd)
 236{
 237	struct scsi_device *sdev = scmd->device;
 238	struct scsi_sense_hdr sshdr;
 239
 240	if (! scsi_command_normalize_sense(scmd, &sshdr))
 241		return FAILED;	/* no valid sense data */
 242
 
 
 
 
 
 
 
 
 243	if (scsi_sense_is_deferred(&sshdr))
 244		return NEEDS_RETRY;
 245
 246	if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
 247			sdev->scsi_dh_data->scsi_dh->check_sense) {
 248		int rc;
 249
 250		rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
 251		if (rc != SCSI_RETURN_NOT_HANDLED)
 252			return rc;
 253		/* handler does not care. Drop down to default handling */
 254	}
 255
 256	/*
 257	 * Previous logic looked for FILEMARK, EOM or ILI which are
 258	 * mainly associated with tapes and returned SUCCESS.
 259	 */
 260	if (sshdr.response_code == 0x70) {
 261		/* fixed format */
 262		if (scmd->sense_buffer[2] & 0xe0)
 263			return SUCCESS;
 264	} else {
 265		/*
 266		 * descriptor format: look for "stream commands sense data
 267		 * descriptor" (see SSC-3). Assume single sense data
 268		 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
 269		 */
 270		if ((sshdr.additional_length > 3) &&
 271		    (scmd->sense_buffer[8] == 0x4) &&
 272		    (scmd->sense_buffer[11] & 0xe0))
 273			return SUCCESS;
 274	}
 275
 276	switch (sshdr.sense_key) {
 277	case NO_SENSE:
 278		return SUCCESS;
 279	case RECOVERED_ERROR:
 280		return /* soft_error */ SUCCESS;
 281
 282	case ABORTED_COMMAND:
 283		if (sshdr.asc == 0x10) /* DIF */
 284			return SUCCESS;
 285
 286		return NEEDS_RETRY;
 287	case NOT_READY:
 288	case UNIT_ATTENTION:
 289		/*
 290		 * if we are expecting a cc/ua because of a bus reset that we
 291		 * performed, treat this just as a retry.  otherwise this is
 292		 * information that we should pass up to the upper-level driver
 293		 * so that we can deal with it there.
 294		 */
 295		if (scmd->device->expecting_cc_ua) {
 296			scmd->device->expecting_cc_ua = 0;
 297			return NEEDS_RETRY;
 
 
 
 
 
 
 
 
 298		}
 299		/*
 300		 * if the device is in the process of becoming ready, we
 301		 * should retry.
 302		 */
 303		if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
 304			return NEEDS_RETRY;
 305		/*
 306		 * if the device is not started, we need to wake
 307		 * the error handler to start the motor
 308		 */
 309		if (scmd->device->allow_restart &&
 310		    (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
 311			return FAILED;
 312
 313		if (sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
 314			scmd_printk(KERN_WARNING, scmd,
 315				    "Warning! Received an indication that the "
 316				    "LUN assignments on this target have "
 317				    "changed. The Linux SCSI layer does not "
 318				    "automatically remap LUN assignments.\n");
 319		else if (sshdr.asc == 0x3f)
 320			scmd_printk(KERN_WARNING, scmd,
 321				    "Warning! Received an indication that the "
 322				    "operating parameters on this target have "
 323				    "changed. The Linux SCSI layer does not "
 324				    "automatically adjust these parameters.\n");
 325
 326		if (sshdr.asc == 0x38 && sshdr.ascq == 0x07)
 327			scmd_printk(KERN_WARNING, scmd,
 328				    "Warning! Received an indication that the "
 329				    "LUN reached a thin provisioning soft "
 330				    "threshold.\n");
 331
 332		/*
 333		 * Pass the UA upwards for a determination in the completion
 334		 * functions.
 335		 */
 336		return SUCCESS;
 337
 338		/* these are not supported */
 339	case COPY_ABORTED:
 340	case VOLUME_OVERFLOW:
 341	case MISCOMPARE:
 342	case BLANK_CHECK:
 343	case DATA_PROTECT:
 344		return TARGET_ERROR;
 345
 346	case MEDIUM_ERROR:
 347		if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
 348		    sshdr.asc == 0x13 || /* AMNF DATA FIELD */
 349		    sshdr.asc == 0x14) { /* RECORD NOT FOUND */
 350			return TARGET_ERROR;
 351		}
 352		return NEEDS_RETRY;
 353
 354	case HARDWARE_ERROR:
 355		if (scmd->device->retry_hwerror)
 356			return ADD_TO_MLQUEUE;
 357		else
 358			return TARGET_ERROR;
 359
 360	case ILLEGAL_REQUEST:
 
 
 
 
 
 
 
 
 361	default:
 362		return SUCCESS;
 363	}
 364}
 365
 366static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
 367{
 368	struct scsi_host_template *sht = sdev->host->hostt;
 369	struct scsi_device *tmp_sdev;
 370
 371	if (!sht->change_queue_depth ||
 372	    sdev->queue_depth >= sdev->max_queue_depth)
 373		return;
 374
 375	if (time_before(jiffies,
 376	    sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
 377		return;
 378
 379	if (time_before(jiffies,
 380	    sdev->last_queue_full_time + sdev->queue_ramp_up_period))
 381		return;
 382
 383	/*
 384	 * Walk all devices of a target and do
 385	 * ramp up on them.
 386	 */
 387	shost_for_each_device(tmp_sdev, sdev->host) {
 388		if (tmp_sdev->channel != sdev->channel ||
 389		    tmp_sdev->id != sdev->id ||
 390		    tmp_sdev->queue_depth == sdev->max_queue_depth)
 391			continue;
 392		/*
 393		 * call back into LLD to increase queue_depth by one
 394		 * with ramp up reason code.
 395		 */
 396		sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
 397					SCSI_QDEPTH_RAMP_UP);
 398		sdev->last_queue_ramp_up = jiffies;
 399	}
 400}
 401
 402static void scsi_handle_queue_full(struct scsi_device *sdev)
 403{
 404	struct scsi_host_template *sht = sdev->host->hostt;
 405	struct scsi_device *tmp_sdev;
 406
 407	if (!sht->change_queue_depth)
 408		return;
 409
 410	shost_for_each_device(tmp_sdev, sdev->host) {
 411		if (tmp_sdev->channel != sdev->channel ||
 412		    tmp_sdev->id != sdev->id)
 413			continue;
 414		/*
 415		 * We do not know the number of commands that were at
 416		 * the device when we got the queue full so we start
 417		 * from the highest possible value and work our way down.
 418		 */
 419		sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
 420					SCSI_QDEPTH_QFULL);
 421	}
 422}
 423
 424/**
 425 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
 426 * @scmd:	SCSI cmd to examine.
 427 *
 428 * Notes:
 429 *    This is *only* called when we are examining the status of commands
 430 *    queued during error recovery.  the main difference here is that we
 431 *    don't allow for the possibility of retries here, and we are a lot
 432 *    more restrictive about what we consider acceptable.
 433 */
 434static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
 435{
 436	/*
 437	 * first check the host byte, to see if there is anything in there
 438	 * that would indicate what we need to do.
 439	 */
 440	if (host_byte(scmd->result) == DID_RESET) {
 441		/*
 442		 * rats.  we are already in the error handler, so we now
 443		 * get to try and figure out what to do next.  if the sense
 444		 * is valid, we have a pretty good idea of what to do.
 445		 * if not, we mark it as FAILED.
 446		 */
 447		return scsi_check_sense(scmd);
 448	}
 449	if (host_byte(scmd->result) != DID_OK)
 450		return FAILED;
 451
 452	/*
 453	 * next, check the message byte.
 454	 */
 455	if (msg_byte(scmd->result) != COMMAND_COMPLETE)
 456		return FAILED;
 457
 458	/*
 459	 * now, check the status byte to see if this indicates
 460	 * anything special.
 461	 */
 462	switch (status_byte(scmd->result)) {
 463	case GOOD:
 464		scsi_handle_queue_ramp_up(scmd->device);
 465	case COMMAND_TERMINATED:
 466		return SUCCESS;
 467	case CHECK_CONDITION:
 468		return scsi_check_sense(scmd);
 469	case CONDITION_GOOD:
 470	case INTERMEDIATE_GOOD:
 471	case INTERMEDIATE_C_GOOD:
 472		/*
 473		 * who knows?  FIXME(eric)
 474		 */
 475		return SUCCESS;
 476	case RESERVATION_CONFLICT:
 477		if (scmd->cmnd[0] == TEST_UNIT_READY)
 478			/* it is a success, we probed the device and
 479			 * found it */
 480			return SUCCESS;
 481		/* otherwise, we failed to send the command */
 482		return FAILED;
 483	case QUEUE_FULL:
 484		scsi_handle_queue_full(scmd->device);
 485		/* fall through */
 486	case BUSY:
 487		return NEEDS_RETRY;
 488	default:
 489		return FAILED;
 490	}
 491	return FAILED;
 492}
 493
 494/**
 495 * scsi_eh_done - Completion function for error handling.
 496 * @scmd:	Cmd that is done.
 497 */
 498static void scsi_eh_done(struct scsi_cmnd *scmd)
 499{
 500	struct completion *eh_action;
 501
 502	SCSI_LOG_ERROR_RECOVERY(3,
 503		printk("%s scmd: %p result: %x\n",
 504			__func__, scmd, scmd->result));
 505
 506	eh_action = scmd->device->host->eh_action;
 507	if (eh_action)
 508		complete(eh_action);
 509}
 510
 511/**
 512 * scsi_try_host_reset - ask host adapter to reset itself
 513 * @scmd:	SCSI cmd to send hsot reset.
 514 */
 515static int scsi_try_host_reset(struct scsi_cmnd *scmd)
 516{
 517	unsigned long flags;
 518	int rtn;
 519	struct Scsi_Host *host = scmd->device->host;
 520	struct scsi_host_template *hostt = host->hostt;
 521
 522	SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
 523					  __func__));
 524
 525	if (!hostt->eh_host_reset_handler)
 526		return FAILED;
 527
 528	rtn = hostt->eh_host_reset_handler(scmd);
 529
 530	if (rtn == SUCCESS) {
 531		if (!hostt->skip_settle_delay)
 532			ssleep(HOST_RESET_SETTLE_TIME);
 533		spin_lock_irqsave(host->host_lock, flags);
 534		scsi_report_bus_reset(host, scmd_channel(scmd));
 535		spin_unlock_irqrestore(host->host_lock, flags);
 536	}
 537
 538	return rtn;
 539}
 540
 541/**
 542 * scsi_try_bus_reset - ask host to perform a bus reset
 543 * @scmd:	SCSI cmd to send bus reset.
 544 */
 545static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
 546{
 547	unsigned long flags;
 548	int rtn;
 549	struct Scsi_Host *host = scmd->device->host;
 550	struct scsi_host_template *hostt = host->hostt;
 551
 552	SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
 553					  __func__));
 554
 555	if (!hostt->eh_bus_reset_handler)
 556		return FAILED;
 557
 558	rtn = hostt->eh_bus_reset_handler(scmd);
 559
 560	if (rtn == SUCCESS) {
 561		if (!hostt->skip_settle_delay)
 562			ssleep(BUS_RESET_SETTLE_TIME);
 563		spin_lock_irqsave(host->host_lock, flags);
 564		scsi_report_bus_reset(host, scmd_channel(scmd));
 565		spin_unlock_irqrestore(host->host_lock, flags);
 566	}
 567
 568	return rtn;
 569}
 570
 571static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
 572{
 573	sdev->was_reset = 1;
 574	sdev->expecting_cc_ua = 1;
 575}
 576
 577/**
 578 * scsi_try_target_reset - Ask host to perform a target reset
 579 * @scmd:	SCSI cmd used to send a target reset
 580 *
 581 * Notes:
 582 *    There is no timeout for this operation.  if this operation is
 583 *    unreliable for a given host, then the host itself needs to put a
 584 *    timer on it, and set the host back to a consistent state prior to
 585 *    returning.
 586 */
 587static int scsi_try_target_reset(struct scsi_cmnd *scmd)
 588{
 589	unsigned long flags;
 590	int rtn;
 591	struct Scsi_Host *host = scmd->device->host;
 592	struct scsi_host_template *hostt = host->hostt;
 593
 594	if (!hostt->eh_target_reset_handler)
 595		return FAILED;
 596
 597	rtn = hostt->eh_target_reset_handler(scmd);
 598	if (rtn == SUCCESS) {
 599		spin_lock_irqsave(host->host_lock, flags);
 600		__starget_for_each_device(scsi_target(scmd->device), NULL,
 601					  __scsi_report_device_reset);
 602		spin_unlock_irqrestore(host->host_lock, flags);
 603	}
 604
 605	return rtn;
 606}
 607
 608/**
 609 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
 610 * @scmd:	SCSI cmd used to send BDR
 611 *
 612 * Notes:
 613 *    There is no timeout for this operation.  if this operation is
 614 *    unreliable for a given host, then the host itself needs to put a
 615 *    timer on it, and set the host back to a consistent state prior to
 616 *    returning.
 617 */
 618static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
 619{
 620	int rtn;
 621	struct scsi_host_template *hostt = scmd->device->host->hostt;
 622
 623	if (!hostt->eh_device_reset_handler)
 624		return FAILED;
 625
 626	rtn = hostt->eh_device_reset_handler(scmd);
 627	if (rtn == SUCCESS)
 628		__scsi_report_device_reset(scmd->device, NULL);
 629	return rtn;
 630}
 631
 632static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
 633{
 634	if (!hostt->eh_abort_handler)
 635		return FAILED;
 636
 637	return hostt->eh_abort_handler(scmd);
 638}
 639
 640static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
 641{
 642	if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
 643		if (scsi_try_bus_device_reset(scmd) != SUCCESS)
 644			if (scsi_try_target_reset(scmd) != SUCCESS)
 645				if (scsi_try_bus_reset(scmd) != SUCCESS)
 646					scsi_try_host_reset(scmd);
 647}
 648
 649/**
 650 * scsi_eh_prep_cmnd  - Save a scsi command info as part of error recory
 651 * @scmd:       SCSI command structure to hijack
 652 * @ses:        structure to save restore information
 653 * @cmnd:       CDB to send. Can be NULL if no new cmnd is needed
 654 * @cmnd_size:  size in bytes of @cmnd (must be <= BLK_MAX_CDB)
 655 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
 656 *
 657 * This function is used to save a scsi command information before re-execution
 658 * as part of the error recovery process.  If @sense_bytes is 0 the command
 659 * sent must be one that does not transfer any data.  If @sense_bytes != 0
 660 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
 661 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
 662 */
 663void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
 664			unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
 665{
 666	struct scsi_device *sdev = scmd->device;
 667
 668	/*
 669	 * We need saved copies of a number of fields - this is because
 670	 * error handling may need to overwrite these with different values
 671	 * to run different commands, and once error handling is complete,
 672	 * we will need to restore these values prior to running the actual
 673	 * command.
 674	 */
 675	ses->cmd_len = scmd->cmd_len;
 676	ses->cmnd = scmd->cmnd;
 677	ses->data_direction = scmd->sc_data_direction;
 678	ses->sdb = scmd->sdb;
 679	ses->next_rq = scmd->request->next_rq;
 680	ses->result = scmd->result;
 681	ses->underflow = scmd->underflow;
 682	ses->prot_op = scmd->prot_op;
 683
 684	scmd->prot_op = SCSI_PROT_NORMAL;
 685	scmd->cmnd = ses->eh_cmnd;
 686	memset(scmd->cmnd, 0, BLK_MAX_CDB);
 687	memset(&scmd->sdb, 0, sizeof(scmd->sdb));
 688	scmd->request->next_rq = NULL;
 689
 690	if (sense_bytes) {
 691		scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
 692					 sense_bytes);
 693		sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
 694			    scmd->sdb.length);
 695		scmd->sdb.table.sgl = &ses->sense_sgl;
 696		scmd->sc_data_direction = DMA_FROM_DEVICE;
 697		scmd->sdb.table.nents = 1;
 698		scmd->cmnd[0] = REQUEST_SENSE;
 699		scmd->cmnd[4] = scmd->sdb.length;
 700		scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
 701	} else {
 702		scmd->sc_data_direction = DMA_NONE;
 703		if (cmnd) {
 704			BUG_ON(cmnd_size > BLK_MAX_CDB);
 705			memcpy(scmd->cmnd, cmnd, cmnd_size);
 706			scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
 707		}
 708	}
 709
 710	scmd->underflow = 0;
 711
 712	if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
 713		scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
 714			(sdev->lun << 5 & 0xe0);
 715
 716	/*
 717	 * Zero the sense buffer.  The scsi spec mandates that any
 718	 * untransferred sense data should be interpreted as being zero.
 719	 */
 720	memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
 721}
 722EXPORT_SYMBOL(scsi_eh_prep_cmnd);
 723
 724/**
 725 * scsi_eh_restore_cmnd  - Restore a scsi command info as part of error recory
 726 * @scmd:       SCSI command structure to restore
 727 * @ses:        saved information from a coresponding call to scsi_eh_prep_cmnd
 728 *
 729 * Undo any damage done by above scsi_eh_prep_cmnd().
 730 */
 731void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
 732{
 733	/*
 734	 * Restore original data
 735	 */
 736	scmd->cmd_len = ses->cmd_len;
 737	scmd->cmnd = ses->cmnd;
 738	scmd->sc_data_direction = ses->data_direction;
 739	scmd->sdb = ses->sdb;
 740	scmd->request->next_rq = ses->next_rq;
 741	scmd->result = ses->result;
 742	scmd->underflow = ses->underflow;
 743	scmd->prot_op = ses->prot_op;
 744}
 745EXPORT_SYMBOL(scsi_eh_restore_cmnd);
 746
 747/**
 748 * scsi_send_eh_cmnd  - submit a scsi command as part of error recory
 749 * @scmd:       SCSI command structure to hijack
 750 * @cmnd:       CDB to send
 751 * @cmnd_size:  size in bytes of @cmnd
 752 * @timeout:    timeout for this request
 753 * @sense_bytes: size of sense data to copy or 0
 754 *
 755 * This function is used to send a scsi command down to a target device
 756 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
 757 *
 758 * Return value:
 759 *    SUCCESS or FAILED or NEEDS_RETRY
 760 */
 761static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
 762			     int cmnd_size, int timeout, unsigned sense_bytes)
 763{
 764	struct scsi_device *sdev = scmd->device;
 
 765	struct Scsi_Host *shost = sdev->host;
 766	DECLARE_COMPLETION_ONSTACK(done);
 767	unsigned long timeleft;
 768	struct scsi_eh_save ses;
 769	int rtn;
 770
 771	scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
 772	shost->eh_action = &done;
 773
 774	scsi_log_send(scmd);
 775	scmd->scsi_done = scsi_eh_done;
 776	shost->hostt->queuecommand(shost, scmd);
 777
 778	timeleft = wait_for_completion_timeout(&done, timeout);
 779
 780	shost->eh_action = NULL;
 781
 782	scsi_log_completion(scmd, SUCCESS);
 783
 784	SCSI_LOG_ERROR_RECOVERY(3,
 785		printk("%s: scmd: %p, timeleft: %ld\n",
 786			__func__, scmd, timeleft));
 787
 788	/*
 789	 * If there is time left scsi_eh_done got called, and we will
 790	 * examine the actual status codes to see whether the command
 791	 * actually did complete normally, else tell the host to forget
 792	 * about this command.
 793	 */
 794	if (timeleft) {
 795		rtn = scsi_eh_completed_normally(scmd);
 796		SCSI_LOG_ERROR_RECOVERY(3,
 797			printk("%s: scsi_eh_completed_normally %x\n",
 798			       __func__, rtn));
 799
 800		switch (rtn) {
 801		case SUCCESS:
 802		case NEEDS_RETRY:
 803		case FAILED:
 804		case TARGET_ERROR:
 805			break;
 806		case ADD_TO_MLQUEUE:
 807			rtn = NEEDS_RETRY;
 808			break;
 809		default:
 810			rtn = FAILED;
 811			break;
 812		}
 813	} else {
 814		scsi_abort_eh_cmnd(scmd);
 815		rtn = FAILED;
 816	}
 817
 818	scsi_eh_restore_cmnd(scmd, &ses);
 
 
 
 
 819	return rtn;
 820}
 821
 822/**
 823 * scsi_request_sense - Request sense data from a particular target.
 824 * @scmd:	SCSI cmd for request sense.
 825 *
 826 * Notes:
 827 *    Some hosts automatically obtain this information, others require
 828 *    that we obtain it on our own. This function will *not* return until
 829 *    the command either times out, or it completes.
 830 */
 831static int scsi_request_sense(struct scsi_cmnd *scmd)
 832{
 833	return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
 834}
 835
 836/**
 837 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
 838 * @scmd:	Original SCSI cmd that eh has finished.
 839 * @done_q:	Queue for processed commands.
 840 *
 841 * Notes:
 842 *    We don't want to use the normal command completion while we are are
 843 *    still handling errors - it may cause other commands to be queued,
 844 *    and that would disturb what we are doing.  Thus we really want to
 845 *    keep a list of pending commands for final completion, and once we
 846 *    are ready to leave error handling we handle completion for real.
 847 */
 848void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
 849{
 850	scmd->device->host->host_failed--;
 851	scmd->eh_eflags = 0;
 852	list_move_tail(&scmd->eh_entry, done_q);
 853}
 854EXPORT_SYMBOL(scsi_eh_finish_cmd);
 855
 856/**
 857 * scsi_eh_get_sense - Get device sense data.
 858 * @work_q:	Queue of commands to process.
 859 * @done_q:	Queue of processed commands.
 860 *
 861 * Description:
 862 *    See if we need to request sense information.  if so, then get it
 863 *    now, so we have a better idea of what to do.
 864 *
 865 * Notes:
 866 *    This has the unfortunate side effect that if a shost adapter does
 867 *    not automatically request sense information, we end up shutting
 868 *    it down before we request it.
 869 *
 870 *    All drivers should request sense information internally these days,
 871 *    so for now all I have to say is tough noogies if you end up in here.
 872 *
 873 *    XXX: Long term this code should go away, but that needs an audit of
 874 *         all LLDDs first.
 875 */
 876int scsi_eh_get_sense(struct list_head *work_q,
 877		      struct list_head *done_q)
 878{
 879	struct scsi_cmnd *scmd, *next;
 880	int rtn;
 881
 882	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
 883		if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
 884		    SCSI_SENSE_VALID(scmd))
 885			continue;
 886
 887		SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
 888						  "%s: requesting sense\n",
 889						  current->comm));
 890		rtn = scsi_request_sense(scmd);
 891		if (rtn != SUCCESS)
 892			continue;
 893
 894		SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
 895						  " result %x\n", scmd,
 896						  scmd->result));
 897		SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
 898
 899		rtn = scsi_decide_disposition(scmd);
 900
 901		/*
 902		 * if the result was normal, then just pass it along to the
 903		 * upper level.
 904		 */
 905		if (rtn == SUCCESS)
 906			/* we don't want this command reissued, just
 907			 * finished with the sense data, so set
 908			 * retries to the max allowed to ensure it
 909			 * won't get reissued */
 910			scmd->retries = scmd->allowed;
 911		else if (rtn != NEEDS_RETRY)
 912			continue;
 913
 914		scsi_eh_finish_cmd(scmd, done_q);
 915	}
 916
 917	return list_empty(work_q);
 918}
 919EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
 920
 921/**
 922 * scsi_eh_tur - Send TUR to device.
 923 * @scmd:	&scsi_cmnd to send TUR
 924 *
 925 * Return value:
 926 *    0 - Device is ready. 1 - Device NOT ready.
 927 */
 928static int scsi_eh_tur(struct scsi_cmnd *scmd)
 929{
 930	static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
 931	int retry_cnt = 1, rtn;
 932
 933retry_tur:
 934	rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
 935
 936	SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
 937		__func__, scmd, rtn));
 938
 939	switch (rtn) {
 940	case NEEDS_RETRY:
 941		if (retry_cnt--)
 942			goto retry_tur;
 943		/*FALLTHRU*/
 944	case SUCCESS:
 945		return 0;
 946	default:
 947		return 1;
 948	}
 949}
 950
 951/**
 952 * scsi_eh_test_devices - check if devices are responding from error recovery.
 953 * @cmd_list:	scsi commands in error recovery.
 954 * @work_q:     queue for commands which still need more error recovery
 955 * @done_q:     queue for commands which are finished
 956 * @try_stu:    boolean on if a STU command should be tried in addition to TUR.
 957 *
 958 * Decription:
 959 *    Tests if devices are in a working state.  Commands to devices now in
 960 *    a working state are sent to the done_q while commands to devices which
 961 *    are still failing to respond are returned to the work_q for more
 962 *    processing.
 963 **/
 964static int scsi_eh_test_devices(struct list_head *cmd_list,
 965				struct list_head *work_q,
 966				struct list_head *done_q, int try_stu)
 967{
 968	struct scsi_cmnd *scmd, *next;
 969	struct scsi_device *sdev;
 970	int finish_cmds;
 971
 972	while (!list_empty(cmd_list)) {
 973		scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
 974		sdev = scmd->device;
 975
 976		finish_cmds = !scsi_device_online(scmd->device) ||
 977			(try_stu && !scsi_eh_try_stu(scmd) &&
 978			 !scsi_eh_tur(scmd)) ||
 979			!scsi_eh_tur(scmd);
 980
 981		list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
 982			if (scmd->device == sdev) {
 983				if (finish_cmds)
 984					scsi_eh_finish_cmd(scmd, done_q);
 985				else
 986					list_move_tail(&scmd->eh_entry, work_q);
 987			}
 988	}
 989	return list_empty(work_q);
 990}
 991
 992
 993/**
 994 * scsi_eh_abort_cmds - abort pending commands.
 995 * @work_q:	&list_head for pending commands.
 996 * @done_q:	&list_head for processed commands.
 997 *
 998 * Decription:
 999 *    Try and see whether or not it makes sense to try and abort the
1000 *    running command.  This only works out to be the case if we have one
1001 *    command that has timed out.  If the command simply failed, it makes
1002 *    no sense to try and abort the command, since as far as the shost
1003 *    adapter is concerned, it isn't running.
1004 */
1005static int scsi_eh_abort_cmds(struct list_head *work_q,
1006			      struct list_head *done_q)
1007{
1008	struct scsi_cmnd *scmd, *next;
1009	LIST_HEAD(check_list);
1010	int rtn;
1011
1012	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1013		if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1014			continue;
1015		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1016						  "0x%p\n", current->comm,
1017						  scmd));
1018		rtn = scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd);
1019		if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1020			scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1021			if (rtn == FAST_IO_FAIL)
1022				scsi_eh_finish_cmd(scmd, done_q);
1023			else
1024				list_move_tail(&scmd->eh_entry, &check_list);
1025		} else
1026			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1027							  " cmd failed:"
1028							  "0x%p\n",
1029							  current->comm,
1030							  scmd));
1031	}
1032
1033	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1034}
1035
1036/**
1037 * scsi_eh_try_stu - Send START_UNIT to device.
1038 * @scmd:	&scsi_cmnd to send START_UNIT
1039 *
1040 * Return value:
1041 *    0 - Device is ready. 1 - Device NOT ready.
1042 */
1043static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1044{
1045	static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1046
1047	if (scmd->device->allow_restart) {
1048		int i, rtn = NEEDS_RETRY;
1049
1050		for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1051			rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1052
1053		if (rtn == SUCCESS)
1054			return 0;
1055	}
1056
1057	return 1;
1058}
1059
1060 /**
1061 * scsi_eh_stu - send START_UNIT if needed
1062 * @shost:	&scsi host being recovered.
1063 * @work_q:     &list_head for pending commands.
1064 * @done_q:	&list_head for processed commands.
1065 *
1066 * Notes:
1067 *    If commands are failing due to not ready, initializing command required,
1068 *	try revalidating the device, which will end up sending a start unit.
1069 */
1070static int scsi_eh_stu(struct Scsi_Host *shost,
1071			      struct list_head *work_q,
1072			      struct list_head *done_q)
1073{
1074	struct scsi_cmnd *scmd, *stu_scmd, *next;
1075	struct scsi_device *sdev;
1076
1077	shost_for_each_device(sdev, shost) {
1078		stu_scmd = NULL;
1079		list_for_each_entry(scmd, work_q, eh_entry)
1080			if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1081			    scsi_check_sense(scmd) == FAILED ) {
1082				stu_scmd = scmd;
1083				break;
1084			}
1085
1086		if (!stu_scmd)
1087			continue;
1088
1089		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1090						  " 0x%p\n", current->comm, sdev));
1091
1092		if (!scsi_eh_try_stu(stu_scmd)) {
1093			if (!scsi_device_online(sdev) ||
1094			    !scsi_eh_tur(stu_scmd)) {
1095				list_for_each_entry_safe(scmd, next,
1096							  work_q, eh_entry) {
1097					if (scmd->device == sdev)
1098						scsi_eh_finish_cmd(scmd, done_q);
1099				}
1100			}
1101		} else {
1102			SCSI_LOG_ERROR_RECOVERY(3,
1103						printk("%s: START_UNIT failed to sdev:"
1104						       " 0x%p\n", current->comm, sdev));
1105		}
1106	}
1107
1108	return list_empty(work_q);
1109}
1110
1111
1112/**
1113 * scsi_eh_bus_device_reset - send bdr if needed
1114 * @shost:	scsi host being recovered.
1115 * @work_q:     &list_head for pending commands.
1116 * @done_q:	&list_head for processed commands.
1117 *
1118 * Notes:
1119 *    Try a bus device reset.  Still, look to see whether we have multiple
1120 *    devices that are jammed or not - if we have multiple devices, it
1121 *    makes no sense to try bus_device_reset - we really would need to try
1122 *    a bus_reset instead.
1123 */
1124static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1125				    struct list_head *work_q,
1126				    struct list_head *done_q)
1127{
1128	struct scsi_cmnd *scmd, *bdr_scmd, *next;
1129	struct scsi_device *sdev;
1130	int rtn;
1131
1132	shost_for_each_device(sdev, shost) {
1133		bdr_scmd = NULL;
1134		list_for_each_entry(scmd, work_q, eh_entry)
1135			if (scmd->device == sdev) {
1136				bdr_scmd = scmd;
1137				break;
1138			}
1139
1140		if (!bdr_scmd)
1141			continue;
1142
1143		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1144						  " 0x%p\n", current->comm,
1145						  sdev));
1146		rtn = scsi_try_bus_device_reset(bdr_scmd);
1147		if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1148			if (!scsi_device_online(sdev) ||
1149			    rtn == FAST_IO_FAIL ||
1150			    !scsi_eh_tur(bdr_scmd)) {
1151				list_for_each_entry_safe(scmd, next,
1152							 work_q, eh_entry) {
1153					if (scmd->device == sdev)
1154						scsi_eh_finish_cmd(scmd,
1155								   done_q);
1156				}
1157			}
1158		} else {
1159			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1160							  " failed sdev:"
1161							  "0x%p\n",
1162							  current->comm,
1163							   sdev));
1164		}
1165	}
1166
1167	return list_empty(work_q);
1168}
1169
1170/**
1171 * scsi_eh_target_reset - send target reset if needed
1172 * @shost:	scsi host being recovered.
1173 * @work_q:     &list_head for pending commands.
1174 * @done_q:	&list_head for processed commands.
1175 *
1176 * Notes:
1177 *    Try a target reset.
1178 */
1179static int scsi_eh_target_reset(struct Scsi_Host *shost,
1180				struct list_head *work_q,
1181				struct list_head *done_q)
1182{
1183	LIST_HEAD(tmp_list);
1184	LIST_HEAD(check_list);
1185
1186	list_splice_init(work_q, &tmp_list);
1187
1188	while (!list_empty(&tmp_list)) {
1189		struct scsi_cmnd *next, *scmd;
1190		int rtn;
1191		unsigned int id;
1192
1193		scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1194		id = scmd_id(scmd);
1195
1196		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1197						  "to target %d\n",
1198						  current->comm, id));
1199		rtn = scsi_try_target_reset(scmd);
1200		if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1201			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1202							  " failed target: "
1203							  "%d\n",
1204							  current->comm, id));
1205		list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1206			if (scmd_id(scmd) != id)
1207				continue;
1208
1209			if (rtn == SUCCESS)
1210				list_move_tail(&scmd->eh_entry, &check_list);
1211			else if (rtn == FAST_IO_FAIL)
1212				scsi_eh_finish_cmd(scmd, done_q);
1213			else
1214				/* push back on work queue for further processing */
1215				list_move(&scmd->eh_entry, work_q);
1216		}
1217	}
1218
1219	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1220}
1221
1222/**
1223 * scsi_eh_bus_reset - send a bus reset
1224 * @shost:	&scsi host being recovered.
1225 * @work_q:     &list_head for pending commands.
1226 * @done_q:	&list_head for processed commands.
1227 */
1228static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1229			     struct list_head *work_q,
1230			     struct list_head *done_q)
1231{
1232	struct scsi_cmnd *scmd, *chan_scmd, *next;
1233	LIST_HEAD(check_list);
1234	unsigned int channel;
1235	int rtn;
1236
1237	/*
1238	 * we really want to loop over the various channels, and do this on
1239	 * a channel by channel basis.  we should also check to see if any
1240	 * of the failed commands are on soft_reset devices, and if so, skip
1241	 * the reset.
1242	 */
1243
1244	for (channel = 0; channel <= shost->max_channel; channel++) {
1245		chan_scmd = NULL;
1246		list_for_each_entry(scmd, work_q, eh_entry) {
1247			if (channel == scmd_channel(scmd)) {
1248				chan_scmd = scmd;
1249				break;
1250				/*
1251				 * FIXME add back in some support for
1252				 * soft_reset devices.
1253				 */
1254			}
1255		}
1256
1257		if (!chan_scmd)
1258			continue;
1259		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1260						  " %d\n", current->comm,
1261						  channel));
1262		rtn = scsi_try_bus_reset(chan_scmd);
1263		if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1264			list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1265				if (channel == scmd_channel(scmd)) {
1266					if (rtn == FAST_IO_FAIL)
1267						scsi_eh_finish_cmd(scmd,
1268								   done_q);
1269					else
1270						list_move_tail(&scmd->eh_entry,
1271							       &check_list);
1272				}
1273			}
1274		} else {
1275			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1276							  " failed chan: %d\n",
1277							  current->comm,
1278							  channel));
1279		}
1280	}
1281	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1282}
1283
1284/**
1285 * scsi_eh_host_reset - send a host reset
1286 * @work_q:	list_head for processed commands.
1287 * @done_q:	list_head for processed commands.
1288 */
1289static int scsi_eh_host_reset(struct list_head *work_q,
1290			      struct list_head *done_q)
1291{
1292	struct scsi_cmnd *scmd, *next;
1293	LIST_HEAD(check_list);
1294	int rtn;
1295
1296	if (!list_empty(work_q)) {
1297		scmd = list_entry(work_q->next,
1298				  struct scsi_cmnd, eh_entry);
1299
1300		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1301						  , current->comm));
1302
1303		rtn = scsi_try_host_reset(scmd);
1304		if (rtn == SUCCESS) {
1305			list_splice_init(work_q, &check_list);
1306		} else if (rtn == FAST_IO_FAIL) {
1307			list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1308					scsi_eh_finish_cmd(scmd, done_q);
1309			}
1310		} else {
1311			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1312							  " failed\n",
1313							  current->comm));
1314		}
1315	}
1316	return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1317}
1318
1319/**
1320 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1321 * @work_q:	list_head for processed commands.
1322 * @done_q:	list_head for processed commands.
1323 */
1324static void scsi_eh_offline_sdevs(struct list_head *work_q,
1325				  struct list_head *done_q)
1326{
1327	struct scsi_cmnd *scmd, *next;
1328
1329	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1330		sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1331			    "not ready after error recovery\n");
1332		scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1333		if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1334			/*
1335			 * FIXME: Handle lost cmds.
1336			 */
1337		}
1338		scsi_eh_finish_cmd(scmd, done_q);
1339	}
1340	return;
1341}
1342
1343/**
1344 * scsi_noretry_cmd - determinte if command should be failed fast
1345 * @scmd:	SCSI cmd to examine.
1346 */
1347int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1348{
1349	switch (host_byte(scmd->result)) {
1350	case DID_OK:
1351		break;
1352	case DID_BUS_BUSY:
1353		return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1354	case DID_PARITY:
1355		return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1356	case DID_ERROR:
1357		if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1358		    status_byte(scmd->result) == RESERVATION_CONFLICT)
1359			return 0;
1360		/* fall through */
1361	case DID_SOFT_ERROR:
1362		return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1363	}
1364
1365	switch (status_byte(scmd->result)) {
1366	case CHECK_CONDITION:
1367		/*
1368		 * assume caller has checked sense and determinted
1369		 * the check condition was retryable.
1370		 */
1371		if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1372		    scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1373			return 1;
1374	}
1375
1376	return 0;
1377}
1378
1379/**
1380 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1381 * @scmd:	SCSI cmd to examine.
1382 *
1383 * Notes:
1384 *    This is *only* called when we are examining the status after sending
1385 *    out the actual data command.  any commands that are queued for error
1386 *    recovery (e.g. test_unit_ready) do *not* come through here.
1387 *
1388 *    When this routine returns failed, it means the error handler thread
1389 *    is woken.  In cases where the error code indicates an error that
1390 *    doesn't require the error handler read (i.e. we don't need to
1391 *    abort/reset), this function should return SUCCESS.
1392 */
1393int scsi_decide_disposition(struct scsi_cmnd *scmd)
1394{
1395	int rtn;
1396
1397	/*
1398	 * if the device is offline, then we clearly just pass the result back
1399	 * up to the top level.
1400	 */
1401	if (!scsi_device_online(scmd->device)) {
1402		SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1403						  " as SUCCESS\n",
1404						  __func__));
1405		return SUCCESS;
1406	}
1407
1408	/*
1409	 * first check the host byte, to see if there is anything in there
1410	 * that would indicate what we need to do.
1411	 */
1412	switch (host_byte(scmd->result)) {
1413	case DID_PASSTHROUGH:
1414		/*
1415		 * no matter what, pass this through to the upper layer.
1416		 * nuke this special code so that it looks like we are saying
1417		 * did_ok.
1418		 */
1419		scmd->result &= 0xff00ffff;
1420		return SUCCESS;
1421	case DID_OK:
1422		/*
1423		 * looks good.  drop through, and check the next byte.
1424		 */
1425		break;
1426	case DID_NO_CONNECT:
1427	case DID_BAD_TARGET:
1428	case DID_ABORT:
1429		/*
1430		 * note - this means that we just report the status back
1431		 * to the top level driver, not that we actually think
1432		 * that it indicates SUCCESS.
1433		 */
1434		return SUCCESS;
1435		/*
1436		 * when the low level driver returns did_soft_error,
1437		 * it is responsible for keeping an internal retry counter
1438		 * in order to avoid endless loops (db)
1439		 *
1440		 * actually this is a bug in this function here.  we should
1441		 * be mindful of the maximum number of retries specified
1442		 * and not get stuck in a loop.
1443		 */
1444	case DID_SOFT_ERROR:
1445		goto maybe_retry;
1446	case DID_IMM_RETRY:
1447		return NEEDS_RETRY;
1448
1449	case DID_REQUEUE:
1450		return ADD_TO_MLQUEUE;
1451	case DID_TRANSPORT_DISRUPTED:
1452		/*
1453		 * LLD/transport was disrupted during processing of the IO.
1454		 * The transport class is now blocked/blocking,
1455		 * and the transport will decide what to do with the IO
1456		 * based on its timers and recovery capablilities if
1457		 * there are enough retries.
1458		 */
1459		goto maybe_retry;
1460	case DID_TRANSPORT_FAILFAST:
1461		/*
1462		 * The transport decided to failfast the IO (most likely
1463		 * the fast io fail tmo fired), so send IO directly upwards.
1464		 */
1465		return SUCCESS;
1466	case DID_ERROR:
1467		if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1468		    status_byte(scmd->result) == RESERVATION_CONFLICT)
1469			/*
1470			 * execute reservation conflict processing code
1471			 * lower down
1472			 */
1473			break;
1474		/* fallthrough */
1475	case DID_BUS_BUSY:
1476	case DID_PARITY:
1477		goto maybe_retry;
1478	case DID_TIME_OUT:
1479		/*
1480		 * when we scan the bus, we get timeout messages for
1481		 * these commands if there is no device available.
1482		 * other hosts report did_no_connect for the same thing.
1483		 */
1484		if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1485		     scmd->cmnd[0] == INQUIRY)) {
1486			return SUCCESS;
1487		} else {
1488			return FAILED;
1489		}
1490	case DID_RESET:
1491		return SUCCESS;
1492	default:
1493		return FAILED;
1494	}
1495
1496	/*
1497	 * next, check the message byte.
1498	 */
1499	if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1500		return FAILED;
1501
1502	/*
1503	 * check the status byte to see if this indicates anything special.
1504	 */
1505	switch (status_byte(scmd->result)) {
1506	case QUEUE_FULL:
1507		scsi_handle_queue_full(scmd->device);
1508		/*
1509		 * the case of trying to send too many commands to a
1510		 * tagged queueing device.
1511		 */
1512	case BUSY:
1513		/*
1514		 * device can't talk to us at the moment.  Should only
1515		 * occur (SAM-3) when the task queue is empty, so will cause
1516		 * the empty queue handling to trigger a stall in the
1517		 * device.
1518		 */
1519		return ADD_TO_MLQUEUE;
1520	case GOOD:
1521		scsi_handle_queue_ramp_up(scmd->device);
1522	case COMMAND_TERMINATED:
1523		return SUCCESS;
1524	case TASK_ABORTED:
1525		goto maybe_retry;
1526	case CHECK_CONDITION:
1527		rtn = scsi_check_sense(scmd);
1528		if (rtn == NEEDS_RETRY)
1529			goto maybe_retry;
1530		else if (rtn == TARGET_ERROR) {
1531			/*
1532			 * Need to modify host byte to signal a
1533			 * permanent target failure
1534			 */
1535			scmd->result |= (DID_TARGET_FAILURE << 16);
1536			rtn = SUCCESS;
1537		}
1538		/* if rtn == FAILED, we have no sense information;
1539		 * returning FAILED will wake the error handler thread
1540		 * to collect the sense and redo the decide
1541		 * disposition */
1542		return rtn;
1543	case CONDITION_GOOD:
1544	case INTERMEDIATE_GOOD:
1545	case INTERMEDIATE_C_GOOD:
1546	case ACA_ACTIVE:
1547		/*
1548		 * who knows?  FIXME(eric)
1549		 */
1550		return SUCCESS;
1551
1552	case RESERVATION_CONFLICT:
1553		sdev_printk(KERN_INFO, scmd->device,
1554			    "reservation conflict\n");
1555		scmd->result |= (DID_NEXUS_FAILURE << 16);
1556		return SUCCESS; /* causes immediate i/o error */
1557	default:
1558		return FAILED;
1559	}
1560	return FAILED;
1561
1562      maybe_retry:
1563
1564	/* we requeue for retry because the error was retryable, and
1565	 * the request was not marked fast fail.  Note that above,
1566	 * even if the request is marked fast fail, we still requeue
1567	 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1568	if ((++scmd->retries) <= scmd->allowed
1569	    && !scsi_noretry_cmd(scmd)) {
1570		return NEEDS_RETRY;
1571	} else {
1572		/*
1573		 * no more retries - report this one back to upper level.
1574		 */
1575		return SUCCESS;
1576	}
1577}
1578
1579static void eh_lock_door_done(struct request *req, int uptodate)
1580{
1581	__blk_put_request(req->q, req);
1582}
1583
1584/**
1585 * scsi_eh_lock_door - Prevent medium removal for the specified device
1586 * @sdev:	SCSI device to prevent medium removal
1587 *
1588 * Locking:
1589 * 	We must be called from process context.
1590 *
1591 * Notes:
1592 * 	We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1593 * 	head of the devices request queue, and continue.
1594 */
1595static void scsi_eh_lock_door(struct scsi_device *sdev)
1596{
1597	struct request *req;
1598
1599	/*
1600	 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1601	 * request becomes available
1602	 */
1603	req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1604
1605	req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1606	req->cmd[1] = 0;
1607	req->cmd[2] = 0;
1608	req->cmd[3] = 0;
1609	req->cmd[4] = SCSI_REMOVAL_PREVENT;
1610	req->cmd[5] = 0;
1611
1612	req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1613
1614	req->cmd_type = REQ_TYPE_BLOCK_PC;
1615	req->cmd_flags |= REQ_QUIET;
1616	req->timeout = 10 * HZ;
1617	req->retries = 5;
1618
1619	blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1620}
1621
1622/**
1623 * scsi_restart_operations - restart io operations to the specified host.
1624 * @shost:	Host we are restarting.
1625 *
1626 * Notes:
1627 *    When we entered the error handler, we blocked all further i/o to
1628 *    this device.  we need to 'reverse' this process.
1629 */
1630static void scsi_restart_operations(struct Scsi_Host *shost)
1631{
1632	struct scsi_device *sdev;
1633	unsigned long flags;
1634
1635	/*
1636	 * If the door was locked, we need to insert a door lock request
1637	 * onto the head of the SCSI request queue for the device.  There
1638	 * is no point trying to lock the door of an off-line device.
1639	 */
1640	shost_for_each_device(sdev, shost) {
1641		if (scsi_device_online(sdev) && sdev->locked)
1642			scsi_eh_lock_door(sdev);
1643	}
1644
1645	/*
1646	 * next free up anything directly waiting upon the host.  this
1647	 * will be requests for character device operations, and also for
1648	 * ioctls to queued block devices.
1649	 */
1650	SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1651					  __func__));
1652
1653	spin_lock_irqsave(shost->host_lock, flags);
1654	if (scsi_host_set_state(shost, SHOST_RUNNING))
1655		if (scsi_host_set_state(shost, SHOST_CANCEL))
1656			BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1657	spin_unlock_irqrestore(shost->host_lock, flags);
1658
1659	wake_up(&shost->host_wait);
1660
1661	/*
1662	 * finally we need to re-initiate requests that may be pending.  we will
1663	 * have had everything blocked while error handling is taking place, and
1664	 * now that error recovery is done, we will need to ensure that these
1665	 * requests are started.
1666	 */
1667	scsi_run_host_queues(shost);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1668}
1669
1670/**
1671 * scsi_eh_ready_devs - check device ready state and recover if not.
1672 * @shost: 	host to be recovered.
1673 * @work_q:     &list_head for pending commands.
1674 * @done_q:	&list_head for processed commands.
1675 */
1676void scsi_eh_ready_devs(struct Scsi_Host *shost,
1677			struct list_head *work_q,
1678			struct list_head *done_q)
1679{
1680	if (!scsi_eh_stu(shost, work_q, done_q))
1681		if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1682			if (!scsi_eh_target_reset(shost, work_q, done_q))
1683				if (!scsi_eh_bus_reset(shost, work_q, done_q))
1684					if (!scsi_eh_host_reset(work_q, done_q))
1685						scsi_eh_offline_sdevs(work_q,
1686								      done_q);
1687}
1688EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1689
1690/**
1691 * scsi_eh_flush_done_q - finish processed commands or retry them.
1692 * @done_q:	list_head of processed commands.
1693 */
1694void scsi_eh_flush_done_q(struct list_head *done_q)
1695{
1696	struct scsi_cmnd *scmd, *next;
1697
1698	list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1699		list_del_init(&scmd->eh_entry);
1700		if (scsi_device_online(scmd->device) &&
1701		    !scsi_noretry_cmd(scmd) &&
1702		    (++scmd->retries <= scmd->allowed)) {
1703			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1704							  " retry cmd: %p\n",
1705							  current->comm,
1706							  scmd));
1707				scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1708		} else {
1709			/*
1710			 * If just we got sense for the device (called
1711			 * scsi_eh_get_sense), scmd->result is already
1712			 * set, do not set DRIVER_TIMEOUT.
1713			 */
1714			if (!scmd->result)
1715				scmd->result |= (DRIVER_TIMEOUT << 24);
1716			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1717							" cmd: %p\n",
1718							current->comm, scmd));
1719			scsi_finish_command(scmd);
1720		}
1721	}
1722}
1723EXPORT_SYMBOL(scsi_eh_flush_done_q);
1724
1725/**
1726 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1727 * @shost:	Host to unjam.
1728 *
1729 * Notes:
1730 *    When we come in here, we *know* that all commands on the bus have
1731 *    either completed, failed or timed out.  we also know that no further
1732 *    commands are being sent to the host, so things are relatively quiet
1733 *    and we have freedom to fiddle with things as we wish.
1734 *
1735 *    This is only the *default* implementation.  it is possible for
1736 *    individual drivers to supply their own version of this function, and
1737 *    if the maintainer wishes to do this, it is strongly suggested that
1738 *    this function be taken as a template and modified.  this function
1739 *    was designed to correctly handle problems for about 95% of the
1740 *    different cases out there, and it should always provide at least a
1741 *    reasonable amount of error recovery.
1742 *
1743 *    Any command marked 'failed' or 'timeout' must eventually have
1744 *    scsi_finish_cmd() called for it.  we do all of the retry stuff
1745 *    here, so when we restart the host after we return it should have an
1746 *    empty queue.
1747 */
1748static void scsi_unjam_host(struct Scsi_Host *shost)
1749{
1750	unsigned long flags;
1751	LIST_HEAD(eh_work_q);
1752	LIST_HEAD(eh_done_q);
1753
1754	spin_lock_irqsave(shost->host_lock, flags);
1755	list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1756	spin_unlock_irqrestore(shost->host_lock, flags);
1757
1758	SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1759
1760	if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1761		if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1762			scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1763
1764	scsi_eh_flush_done_q(&eh_done_q);
1765}
1766
1767/**
1768 * scsi_error_handler - SCSI error handler thread
1769 * @data:	Host for which we are running.
1770 *
1771 * Notes:
1772 *    This is the main error handling loop.  This is run as a kernel thread
1773 *    for every SCSI host and handles all error handling activity.
1774 */
1775int scsi_error_handler(void *data)
1776{
1777	struct Scsi_Host *shost = data;
1778
1779	/*
1780	 * We use TASK_INTERRUPTIBLE so that the thread is not
1781	 * counted against the load average as a running process.
1782	 * We never actually get interrupted because kthread_run
1783	 * disables signal delivery for the created thread.
1784	 */
1785	set_current_state(TASK_INTERRUPTIBLE);
1786	while (!kthread_should_stop()) {
1787		if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1788		    shost->host_failed != shost->host_busy) {
1789			SCSI_LOG_ERROR_RECOVERY(1,
1790				printk("Error handler scsi_eh_%d sleeping\n",
1791					shost->host_no));
1792			schedule();
1793			set_current_state(TASK_INTERRUPTIBLE);
1794			continue;
1795		}
1796
1797		__set_current_state(TASK_RUNNING);
1798		SCSI_LOG_ERROR_RECOVERY(1,
1799			printk("Error handler scsi_eh_%d waking up\n",
1800				shost->host_no));
1801
1802		/*
1803		 * We have a host that is failing for some reason.  Figure out
1804		 * what we need to do to get it up and online again (if we can).
1805		 * If we fail, we end up taking the thing offline.
1806		 */
1807		if (scsi_autopm_get_host(shost) != 0) {
1808			SCSI_LOG_ERROR_RECOVERY(1,
1809				printk(KERN_ERR "Error handler scsi_eh_%d "
1810						"unable to autoresume\n",
1811						shost->host_no));
1812			continue;
1813		}
1814
1815		if (shost->transportt->eh_strategy_handler)
1816			shost->transportt->eh_strategy_handler(shost);
1817		else
1818			scsi_unjam_host(shost);
1819
1820		/*
1821		 * Note - if the above fails completely, the action is to take
1822		 * individual devices offline and flush the queue of any
1823		 * outstanding requests that may have been pending.  When we
1824		 * restart, we restart any I/O to any other devices on the bus
1825		 * which are still online.
1826		 */
1827		scsi_restart_operations(shost);
1828		scsi_autopm_put_host(shost);
 
1829		set_current_state(TASK_INTERRUPTIBLE);
1830	}
1831	__set_current_state(TASK_RUNNING);
1832
1833	SCSI_LOG_ERROR_RECOVERY(1,
1834		printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1835	shost->ehandler = NULL;
1836	return 0;
1837}
1838
1839/*
1840 * Function:    scsi_report_bus_reset()
1841 *
1842 * Purpose:     Utility function used by low-level drivers to report that
1843 *		they have observed a bus reset on the bus being handled.
1844 *
1845 * Arguments:   shost       - Host in question
1846 *		channel     - channel on which reset was observed.
1847 *
1848 * Returns:     Nothing
1849 *
1850 * Lock status: Host lock must be held.
1851 *
1852 * Notes:       This only needs to be called if the reset is one which
1853 *		originates from an unknown location.  Resets originated
1854 *		by the mid-level itself don't need to call this, but there
1855 *		should be no harm.
1856 *
1857 *		The main purpose of this is to make sure that a CHECK_CONDITION
1858 *		is properly treated.
1859 */
1860void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1861{
1862	struct scsi_device *sdev;
1863
1864	__shost_for_each_device(sdev, shost) {
1865		if (channel == sdev_channel(sdev))
1866			__scsi_report_device_reset(sdev, NULL);
1867	}
1868}
1869EXPORT_SYMBOL(scsi_report_bus_reset);
1870
1871/*
1872 * Function:    scsi_report_device_reset()
1873 *
1874 * Purpose:     Utility function used by low-level drivers to report that
1875 *		they have observed a device reset on the device being handled.
1876 *
1877 * Arguments:   shost       - Host in question
1878 *		channel     - channel on which reset was observed
1879 *		target	    - target on which reset was observed
1880 *
1881 * Returns:     Nothing
1882 *
1883 * Lock status: Host lock must be held
1884 *
1885 * Notes:       This only needs to be called if the reset is one which
1886 *		originates from an unknown location.  Resets originated
1887 *		by the mid-level itself don't need to call this, but there
1888 *		should be no harm.
1889 *
1890 *		The main purpose of this is to make sure that a CHECK_CONDITION
1891 *		is properly treated.
1892 */
1893void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1894{
1895	struct scsi_device *sdev;
1896
1897	__shost_for_each_device(sdev, shost) {
1898		if (channel == sdev_channel(sdev) &&
1899		    target == sdev_id(sdev))
1900			__scsi_report_device_reset(sdev, NULL);
1901	}
1902}
1903EXPORT_SYMBOL(scsi_report_device_reset);
1904
1905static void
1906scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1907{
1908}
1909
1910/*
1911 * Function:	scsi_reset_provider
1912 *
1913 * Purpose:	Send requested reset to a bus or device at any phase.
1914 *
1915 * Arguments:	device	- device to send reset to
1916 *		flag - reset type (see scsi.h)
1917 *
1918 * Returns:	SUCCESS/FAILURE.
1919 *
1920 * Notes:	This is used by the SCSI Generic driver to provide
1921 *		Bus/Device reset capability.
1922 */
1923int
1924scsi_reset_provider(struct scsi_device *dev, int flag)
1925{
1926	struct scsi_cmnd *scmd;
1927	struct Scsi_Host *shost = dev->host;
1928	struct request req;
1929	unsigned long flags;
1930	int rtn;
1931
1932	if (scsi_autopm_get_host(shost) < 0)
1933		return FAILED;
1934
1935	scmd = scsi_get_command(dev, GFP_KERNEL);
1936	blk_rq_init(NULL, &req);
1937	scmd->request = &req;
1938
1939	scmd->cmnd = req.cmd;
1940
1941	scmd->scsi_done		= scsi_reset_provider_done_command;
1942	memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1943
1944	scmd->cmd_len			= 0;
1945
1946	scmd->sc_data_direction		= DMA_BIDIRECTIONAL;
1947
1948	spin_lock_irqsave(shost->host_lock, flags);
1949	shost->tmf_in_progress = 1;
1950	spin_unlock_irqrestore(shost->host_lock, flags);
1951
1952	switch (flag) {
1953	case SCSI_TRY_RESET_DEVICE:
1954		rtn = scsi_try_bus_device_reset(scmd);
1955		if (rtn == SUCCESS)
1956			break;
1957		/* FALLTHROUGH */
1958	case SCSI_TRY_RESET_TARGET:
1959		rtn = scsi_try_target_reset(scmd);
1960		if (rtn == SUCCESS)
1961			break;
1962		/* FALLTHROUGH */
1963	case SCSI_TRY_RESET_BUS:
1964		rtn = scsi_try_bus_reset(scmd);
1965		if (rtn == SUCCESS)
1966			break;
1967		/* FALLTHROUGH */
1968	case SCSI_TRY_RESET_HOST:
1969		rtn = scsi_try_host_reset(scmd);
1970		break;
1971	default:
1972		rtn = FAILED;
1973	}
1974
1975	spin_lock_irqsave(shost->host_lock, flags);
1976	shost->tmf_in_progress = 0;
1977	spin_unlock_irqrestore(shost->host_lock, flags);
1978
1979	/*
1980	 * be sure to wake up anyone who was sleeping or had their queue
1981	 * suspended while we performed the TMF.
1982	 */
1983	SCSI_LOG_ERROR_RECOVERY(3,
1984		printk("%s: waking up host to restart after TMF\n",
1985		__func__));
1986
1987	wake_up(&shost->host_wait);
1988
1989	scsi_run_host_queues(shost);
1990
1991	scsi_next_command(scmd);
1992	scsi_autopm_put_host(shost);
1993	return rtn;
1994}
1995EXPORT_SYMBOL(scsi_reset_provider);
1996
1997/**
1998 * scsi_normalize_sense - normalize main elements from either fixed or
1999 *			descriptor sense data format into a common format.
2000 *
2001 * @sense_buffer:	byte array containing sense data returned by device
2002 * @sb_len:		number of valid bytes in sense_buffer
2003 * @sshdr:		pointer to instance of structure that common
2004 *			elements are written to.
2005 *
2006 * Notes:
2007 *	The "main elements" from sense data are: response_code, sense_key,
2008 *	asc, ascq and additional_length (only for descriptor format).
2009 *
2010 *	Typically this function can be called after a device has
2011 *	responded to a SCSI command with the CHECK_CONDITION status.
2012 *
2013 * Return value:
2014 *	1 if valid sense data information found, else 0;
2015 */
2016int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2017                         struct scsi_sense_hdr *sshdr)
2018{
2019	if (!sense_buffer || !sb_len)
2020		return 0;
2021
2022	memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2023
2024	sshdr->response_code = (sense_buffer[0] & 0x7f);
2025
2026	if (!scsi_sense_valid(sshdr))
2027		return 0;
2028
2029	if (sshdr->response_code >= 0x72) {
2030		/*
2031		 * descriptor format
2032		 */
2033		if (sb_len > 1)
2034			sshdr->sense_key = (sense_buffer[1] & 0xf);
2035		if (sb_len > 2)
2036			sshdr->asc = sense_buffer[2];
2037		if (sb_len > 3)
2038			sshdr->ascq = sense_buffer[3];
2039		if (sb_len > 7)
2040			sshdr->additional_length = sense_buffer[7];
2041	} else {
2042		/*
2043		 * fixed format
2044		 */
2045		if (sb_len > 2)
2046			sshdr->sense_key = (sense_buffer[2] & 0xf);
2047		if (sb_len > 7) {
2048			sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2049					 sb_len : (sense_buffer[7] + 8);
2050			if (sb_len > 12)
2051				sshdr->asc = sense_buffer[12];
2052			if (sb_len > 13)
2053				sshdr->ascq = sense_buffer[13];
2054		}
2055	}
2056
2057	return 1;
2058}
2059EXPORT_SYMBOL(scsi_normalize_sense);
2060
2061int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2062				 struct scsi_sense_hdr *sshdr)
2063{
2064	return scsi_normalize_sense(cmd->sense_buffer,
2065			SCSI_SENSE_BUFFERSIZE, sshdr);
2066}
2067EXPORT_SYMBOL(scsi_command_normalize_sense);
2068
2069/**
2070 * scsi_sense_desc_find - search for a given descriptor type in	descriptor sense data format.
2071 * @sense_buffer:	byte array of descriptor format sense data
2072 * @sb_len:		number of valid bytes in sense_buffer
2073 * @desc_type:		value of descriptor type to find
2074 *			(e.g. 0 -> information)
2075 *
2076 * Notes:
2077 *	only valid when sense data is in descriptor format
2078 *
2079 * Return value:
2080 *	pointer to start of (first) descriptor if found else NULL
2081 */
2082const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2083				int desc_type)
2084{
2085	int add_sen_len, add_len, desc_len, k;
2086	const u8 * descp;
2087
2088	if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2089		return NULL;
2090	if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2091		return NULL;
2092	add_sen_len = (add_sen_len < (sb_len - 8)) ?
2093			add_sen_len : (sb_len - 8);
2094	descp = &sense_buffer[8];
2095	for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2096		descp += desc_len;
2097		add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2098		desc_len = add_len + 2;
2099		if (descp[0] == desc_type)
2100			return descp;
2101		if (add_len < 0) // short descriptor ??
2102			break;
2103	}
2104	return NULL;
2105}
2106EXPORT_SYMBOL(scsi_sense_desc_find);
2107
2108/**
2109 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2110 * @sense_buffer:	byte array of sense data
2111 * @sb_len:		number of valid bytes in sense_buffer
2112 * @info_out:		pointer to 64 integer where 8 or 4 byte information
2113 *			field will be placed if found.
2114 *
2115 * Return value:
2116 *	1 if information field found, 0 if not found.
2117 */
2118int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2119			    u64 * info_out)
2120{
2121	int j;
2122	const u8 * ucp;
2123	u64 ull;
2124
2125	if (sb_len < 7)
2126		return 0;
2127	switch (sense_buffer[0] & 0x7f) {
2128	case 0x70:
2129	case 0x71:
2130		if (sense_buffer[0] & 0x80) {
2131			*info_out = (sense_buffer[3] << 24) +
2132				    (sense_buffer[4] << 16) +
2133				    (sense_buffer[5] << 8) + sense_buffer[6];
2134			return 1;
2135		} else
2136			return 0;
2137	case 0x72:
2138	case 0x73:
2139		ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2140					   0 /* info desc */);
2141		if (ucp && (0xa == ucp[1])) {
2142			ull = 0;
2143			for (j = 0; j < 8; ++j) {
2144				if (j > 0)
2145					ull <<= 8;
2146				ull |= ucp[4 + j];
2147			}
2148			*info_out = ull;
2149			return 1;
2150		} else
2151			return 0;
2152	default:
2153		return 0;
2154	}
2155}
2156EXPORT_SYMBOL(scsi_get_sense_info_fld);
2157
2158/**
2159 * scsi_build_sense_buffer - build sense data in a buffer
2160 * @desc:	Sense format (non zero == descriptor format,
2161 * 		0 == fixed format)
2162 * @buf:	Where to build sense data
2163 * @key:	Sense key
2164 * @asc:	Additional sense code
2165 * @ascq:	Additional sense code qualifier
2166 *
2167 **/
2168void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2169{
2170	if (desc) {
2171		buf[0] = 0x72;	/* descriptor, current */
2172		buf[1] = key;
2173		buf[2] = asc;
2174		buf[3] = ascq;
2175		buf[7] = 0;
2176	} else {
2177		buf[0] = 0x70;	/* fixed, current */
2178		buf[2] = key;
2179		buf[7] = 0xa;
2180		buf[12] = asc;
2181		buf[13] = ascq;
2182	}
2183}
2184EXPORT_SYMBOL(scsi_build_sense_buffer);
v3.5.6
   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
  29#include <scsi/scsi.h>
  30#include <scsi/scsi_cmnd.h>
  31#include <scsi/scsi_dbg.h>
  32#include <scsi/scsi_device.h>
  33#include <scsi/scsi_driver.h>
  34#include <scsi/scsi_eh.h>
  35#include <scsi/scsi_transport.h>
  36#include <scsi/scsi_host.h>
  37#include <scsi/scsi_ioctl.h>
  38
  39#include "scsi_priv.h"
  40#include "scsi_logging.h"
  41#include "scsi_transport_api.h"
  42
  43#include <trace/events/scsi.h>
  44
  45static void scsi_eh_done(struct scsi_cmnd *scmd);
  46
  47#define SENSE_TIMEOUT		(10*HZ)
  48
  49/*
  50 * These should *probably* be handled by the host itself.
  51 * Since it is allowed to sleep, it probably should.
  52 */
  53#define BUS_RESET_SETTLE_TIME   (10)
  54#define HOST_RESET_SETTLE_TIME  (10)
  55
  56static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
  57
  58/* called with shost->host_lock held */
  59void scsi_eh_wakeup(struct Scsi_Host *shost)
  60{
  61	if (shost->host_busy == shost->host_failed) {
  62		trace_scsi_eh_wakeup(shost);
  63		wake_up_process(shost->ehandler);
  64		SCSI_LOG_ERROR_RECOVERY(5,
  65				printk("Waking error handler thread\n"));
  66	}
  67}
  68
  69/**
  70 * scsi_schedule_eh - schedule EH for SCSI host
  71 * @shost:	SCSI host to invoke error handling on.
  72 *
  73 * Schedule SCSI EH without scmd.
  74 */
  75void scsi_schedule_eh(struct Scsi_Host *shost)
  76{
  77	unsigned long flags;
  78
  79	spin_lock_irqsave(shost->host_lock, flags);
  80
  81	if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
  82	    scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
  83		shost->host_eh_scheduled++;
  84		scsi_eh_wakeup(shost);
  85	}
  86
  87	spin_unlock_irqrestore(shost->host_lock, flags);
  88}
  89EXPORT_SYMBOL_GPL(scsi_schedule_eh);
  90
  91/**
  92 * scsi_eh_scmd_add - add scsi cmd to error handling.
  93 * @scmd:	scmd to run eh on.
  94 * @eh_flag:	optional SCSI_EH flag.
  95 *
  96 * Return value:
  97 *	0 on failure.
  98 */
  99int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
 100{
 101	struct Scsi_Host *shost = scmd->device->host;
 102	unsigned long flags;
 103	int ret = 0;
 104
 105	if (!shost->ehandler)
 106		return 0;
 107
 108	spin_lock_irqsave(shost->host_lock, flags);
 109	if (scsi_host_set_state(shost, SHOST_RECOVERY))
 110		if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
 111			goto out_unlock;
 112
 113	ret = 1;
 114	scmd->eh_eflags |= eh_flag;
 115	list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
 116	shost->host_failed++;
 117	scsi_eh_wakeup(shost);
 118 out_unlock:
 119	spin_unlock_irqrestore(shost->host_lock, flags);
 120	return ret;
 121}
 122
 123/**
 124 * scsi_times_out - Timeout function for normal scsi commands.
 125 * @req:	request that is timing out.
 126 *
 127 * Notes:
 128 *     We do not need to lock this.  There is the potential for a race
 129 *     only in that the normal completion handling might run, but if the
 130 *     normal completion function determines that the timer has already
 131 *     fired, then it mustn't do anything.
 132 */
 133enum blk_eh_timer_return scsi_times_out(struct request *req)
 134{
 135	struct scsi_cmnd *scmd = req->special;
 136	enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
 137	struct Scsi_Host *host = scmd->device->host;
 138
 139	trace_scsi_dispatch_cmd_timeout(scmd);
 140	scsi_log_completion(scmd, TIMEOUT_ERROR);
 141
 142	if (host->transportt->eh_timed_out)
 143		rtn = host->transportt->eh_timed_out(scmd);
 144	else if (host->hostt->eh_timed_out)
 145		rtn = host->hostt->eh_timed_out(scmd);
 146
 147	scmd->result |= DID_TIME_OUT << 16;
 148
 149	if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
 150		     !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)))
 
 151		rtn = BLK_EH_HANDLED;
 
 152
 153	return rtn;
 154}
 155
 156/**
 157 * scsi_block_when_processing_errors - Prevent cmds from being queued.
 158 * @sdev:	Device on which we are performing recovery.
 159 *
 160 * Description:
 161 *     We block until the host is out of error recovery, and then check to
 162 *     see whether the host or the device is offline.
 163 *
 164 * Return value:
 165 *     0 when dev was taken offline by error recovery. 1 OK to proceed.
 166 */
 167int scsi_block_when_processing_errors(struct scsi_device *sdev)
 168{
 169	int online;
 170
 171	wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
 172
 173	online = scsi_device_online(sdev);
 174
 175	SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
 176					  online));
 177
 178	return online;
 179}
 180EXPORT_SYMBOL(scsi_block_when_processing_errors);
 181
 182#ifdef CONFIG_SCSI_LOGGING
 183/**
 184 * scsi_eh_prt_fail_stats - Log info on failures.
 185 * @shost:	scsi host being recovered.
 186 * @work_q:	Queue of scsi cmds to process.
 187 */
 188static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
 189					  struct list_head *work_q)
 190{
 191	struct scsi_cmnd *scmd;
 192	struct scsi_device *sdev;
 193	int total_failures = 0;
 194	int cmd_failed = 0;
 195	int cmd_cancel = 0;
 196	int devices_failed = 0;
 197
 198	shost_for_each_device(sdev, shost) {
 199		list_for_each_entry(scmd, work_q, eh_entry) {
 200			if (scmd->device == sdev) {
 201				++total_failures;
 202				if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
 203					++cmd_cancel;
 204				else
 205					++cmd_failed;
 206			}
 207		}
 208
 209		if (cmd_cancel || cmd_failed) {
 210			SCSI_LOG_ERROR_RECOVERY(3,
 211				sdev_printk(KERN_INFO, sdev,
 212					    "%s: cmds failed: %d, cancel: %d\n",
 213					    __func__, cmd_failed,
 214					    cmd_cancel));
 215			cmd_cancel = 0;
 216			cmd_failed = 0;
 217			++devices_failed;
 218		}
 219	}
 220
 221	SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
 222					  " devices require eh work\n",
 223				   total_failures, devices_failed));
 224}
 225#endif
 226
 227/**
 228 * scsi_check_sense - Examine scsi cmd sense
 229 * @scmd:	Cmd to have sense checked.
 230 *
 231 * Return value:
 232 *	SUCCESS or FAILED or NEEDS_RETRY or TARGET_ERROR
 233 *
 234 * Notes:
 235 *	When a deferred error is detected the current command has
 236 *	not been executed and needs retrying.
 237 */
 238static int scsi_check_sense(struct scsi_cmnd *scmd)
 239{
 240	struct scsi_device *sdev = scmd->device;
 241	struct scsi_sense_hdr sshdr;
 242
 243	if (! scsi_command_normalize_sense(scmd, &sshdr))
 244		return FAILED;	/* no valid sense data */
 245
 246	if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
 247		/*
 248		 * nasty: for mid-layer issued TURs, we need to return the
 249		 * actual sense data without any recovery attempt.  For eh
 250		 * issued ones, we need to try to recover and interpret
 251		 */
 252		return SUCCESS;
 253
 254	if (scsi_sense_is_deferred(&sshdr))
 255		return NEEDS_RETRY;
 256
 257	if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
 258			sdev->scsi_dh_data->scsi_dh->check_sense) {
 259		int rc;
 260
 261		rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
 262		if (rc != SCSI_RETURN_NOT_HANDLED)
 263			return rc;
 264		/* handler does not care. Drop down to default handling */
 265	}
 266
 267	/*
 268	 * Previous logic looked for FILEMARK, EOM or ILI which are
 269	 * mainly associated with tapes and returned SUCCESS.
 270	 */
 271	if (sshdr.response_code == 0x70) {
 272		/* fixed format */
 273		if (scmd->sense_buffer[2] & 0xe0)
 274			return SUCCESS;
 275	} else {
 276		/*
 277		 * descriptor format: look for "stream commands sense data
 278		 * descriptor" (see SSC-3). Assume single sense data
 279		 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
 280		 */
 281		if ((sshdr.additional_length > 3) &&
 282		    (scmd->sense_buffer[8] == 0x4) &&
 283		    (scmd->sense_buffer[11] & 0xe0))
 284			return SUCCESS;
 285	}
 286
 287	switch (sshdr.sense_key) {
 288	case NO_SENSE:
 289		return SUCCESS;
 290	case RECOVERED_ERROR:
 291		return /* soft_error */ SUCCESS;
 292
 293	case ABORTED_COMMAND:
 294		if (sshdr.asc == 0x10) /* DIF */
 295			return SUCCESS;
 296
 297		return NEEDS_RETRY;
 298	case NOT_READY:
 299	case UNIT_ATTENTION:
 300		/*
 301		 * if we are expecting a cc/ua because of a bus reset that we
 302		 * performed, treat this just as a retry.  otherwise this is
 303		 * information that we should pass up to the upper-level driver
 304		 * so that we can deal with it there.
 305		 */
 306		if (scmd->device->expecting_cc_ua) {
 307			/*
 308			 * Because some device does not queue unit
 309			 * attentions correctly, we carefully check
 310			 * additional sense code and qualifier so as
 311			 * not to squash media change unit attention.
 312			 */
 313			if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
 314				scmd->device->expecting_cc_ua = 0;
 315				return NEEDS_RETRY;
 316			}
 317		}
 318		/*
 319		 * if the device is in the process of becoming ready, we
 320		 * should retry.
 321		 */
 322		if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
 323			return NEEDS_RETRY;
 324		/*
 325		 * if the device is not started, we need to wake
 326		 * the error handler to start the motor
 327		 */
 328		if (scmd->device->allow_restart &&
 329		    (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
 330			return FAILED;
 331
 332		if (sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
 333			scmd_printk(KERN_WARNING, scmd,
 334				    "Warning! Received an indication that the "
 335				    "LUN assignments on this target have "
 336				    "changed. The Linux SCSI layer does not "
 337				    "automatically remap LUN assignments.\n");
 338		else if (sshdr.asc == 0x3f)
 339			scmd_printk(KERN_WARNING, scmd,
 340				    "Warning! Received an indication that the "
 341				    "operating parameters on this target have "
 342				    "changed. The Linux SCSI layer does not "
 343				    "automatically adjust these parameters.\n");
 344
 345		if (sshdr.asc == 0x38 && sshdr.ascq == 0x07)
 346			scmd_printk(KERN_WARNING, scmd,
 347				    "Warning! Received an indication that the "
 348				    "LUN reached a thin provisioning soft "
 349				    "threshold.\n");
 350
 351		/*
 352		 * Pass the UA upwards for a determination in the completion
 353		 * functions.
 354		 */
 355		return SUCCESS;
 356
 357		/* these are not supported */
 358	case COPY_ABORTED:
 359	case VOLUME_OVERFLOW:
 360	case MISCOMPARE:
 361	case BLANK_CHECK:
 362	case DATA_PROTECT:
 363		return TARGET_ERROR;
 364
 365	case MEDIUM_ERROR:
 366		if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
 367		    sshdr.asc == 0x13 || /* AMNF DATA FIELD */
 368		    sshdr.asc == 0x14) { /* RECORD NOT FOUND */
 369			return TARGET_ERROR;
 370		}
 371		return NEEDS_RETRY;
 372
 373	case HARDWARE_ERROR:
 374		if (scmd->device->retry_hwerror)
 375			return ADD_TO_MLQUEUE;
 376		else
 377			return TARGET_ERROR;
 378
 379	case ILLEGAL_REQUEST:
 380		if (sshdr.asc == 0x20 || /* Invalid command operation code */
 381		    sshdr.asc == 0x21 || /* Logical block address out of range */
 382		    sshdr.asc == 0x24 || /* Invalid field in cdb */
 383		    sshdr.asc == 0x26) { /* Parameter value invalid */
 384			return TARGET_ERROR;
 385		}
 386		return SUCCESS;
 387
 388	default:
 389		return SUCCESS;
 390	}
 391}
 392
 393static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
 394{
 395	struct scsi_host_template *sht = sdev->host->hostt;
 396	struct scsi_device *tmp_sdev;
 397
 398	if (!sht->change_queue_depth ||
 399	    sdev->queue_depth >= sdev->max_queue_depth)
 400		return;
 401
 402	if (time_before(jiffies,
 403	    sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
 404		return;
 405
 406	if (time_before(jiffies,
 407	    sdev->last_queue_full_time + sdev->queue_ramp_up_period))
 408		return;
 409
 410	/*
 411	 * Walk all devices of a target and do
 412	 * ramp up on them.
 413	 */
 414	shost_for_each_device(tmp_sdev, sdev->host) {
 415		if (tmp_sdev->channel != sdev->channel ||
 416		    tmp_sdev->id != sdev->id ||
 417		    tmp_sdev->queue_depth == sdev->max_queue_depth)
 418			continue;
 419		/*
 420		 * call back into LLD to increase queue_depth by one
 421		 * with ramp up reason code.
 422		 */
 423		sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
 424					SCSI_QDEPTH_RAMP_UP);
 425		sdev->last_queue_ramp_up = jiffies;
 426	}
 427}
 428
 429static void scsi_handle_queue_full(struct scsi_device *sdev)
 430{
 431	struct scsi_host_template *sht = sdev->host->hostt;
 432	struct scsi_device *tmp_sdev;
 433
 434	if (!sht->change_queue_depth)
 435		return;
 436
 437	shost_for_each_device(tmp_sdev, sdev->host) {
 438		if (tmp_sdev->channel != sdev->channel ||
 439		    tmp_sdev->id != sdev->id)
 440			continue;
 441		/*
 442		 * We do not know the number of commands that were at
 443		 * the device when we got the queue full so we start
 444		 * from the highest possible value and work our way down.
 445		 */
 446		sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
 447					SCSI_QDEPTH_QFULL);
 448	}
 449}
 450
 451/**
 452 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
 453 * @scmd:	SCSI cmd to examine.
 454 *
 455 * Notes:
 456 *    This is *only* called when we are examining the status of commands
 457 *    queued during error recovery.  the main difference here is that we
 458 *    don't allow for the possibility of retries here, and we are a lot
 459 *    more restrictive about what we consider acceptable.
 460 */
 461static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
 462{
 463	/*
 464	 * first check the host byte, to see if there is anything in there
 465	 * that would indicate what we need to do.
 466	 */
 467	if (host_byte(scmd->result) == DID_RESET) {
 468		/*
 469		 * rats.  we are already in the error handler, so we now
 470		 * get to try and figure out what to do next.  if the sense
 471		 * is valid, we have a pretty good idea of what to do.
 472		 * if not, we mark it as FAILED.
 473		 */
 474		return scsi_check_sense(scmd);
 475	}
 476	if (host_byte(scmd->result) != DID_OK)
 477		return FAILED;
 478
 479	/*
 480	 * next, check the message byte.
 481	 */
 482	if (msg_byte(scmd->result) != COMMAND_COMPLETE)
 483		return FAILED;
 484
 485	/*
 486	 * now, check the status byte to see if this indicates
 487	 * anything special.
 488	 */
 489	switch (status_byte(scmd->result)) {
 490	case GOOD:
 491		scsi_handle_queue_ramp_up(scmd->device);
 492	case COMMAND_TERMINATED:
 493		return SUCCESS;
 494	case CHECK_CONDITION:
 495		return scsi_check_sense(scmd);
 496	case CONDITION_GOOD:
 497	case INTERMEDIATE_GOOD:
 498	case INTERMEDIATE_C_GOOD:
 499		/*
 500		 * who knows?  FIXME(eric)
 501		 */
 502		return SUCCESS;
 503	case RESERVATION_CONFLICT:
 504		if (scmd->cmnd[0] == TEST_UNIT_READY)
 505			/* it is a success, we probed the device and
 506			 * found it */
 507			return SUCCESS;
 508		/* otherwise, we failed to send the command */
 509		return FAILED;
 510	case QUEUE_FULL:
 511		scsi_handle_queue_full(scmd->device);
 512		/* fall through */
 513	case BUSY:
 514		return NEEDS_RETRY;
 515	default:
 516		return FAILED;
 517	}
 518	return FAILED;
 519}
 520
 521/**
 522 * scsi_eh_done - Completion function for error handling.
 523 * @scmd:	Cmd that is done.
 524 */
 525static void scsi_eh_done(struct scsi_cmnd *scmd)
 526{
 527	struct completion *eh_action;
 528
 529	SCSI_LOG_ERROR_RECOVERY(3,
 530		printk("%s scmd: %p result: %x\n",
 531			__func__, scmd, scmd->result));
 532
 533	eh_action = scmd->device->host->eh_action;
 534	if (eh_action)
 535		complete(eh_action);
 536}
 537
 538/**
 539 * scsi_try_host_reset - ask host adapter to reset itself
 540 * @scmd:	SCSI cmd to send hsot reset.
 541 */
 542static int scsi_try_host_reset(struct scsi_cmnd *scmd)
 543{
 544	unsigned long flags;
 545	int rtn;
 546	struct Scsi_Host *host = scmd->device->host;
 547	struct scsi_host_template *hostt = host->hostt;
 548
 549	SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
 550					  __func__));
 551
 552	if (!hostt->eh_host_reset_handler)
 553		return FAILED;
 554
 555	rtn = hostt->eh_host_reset_handler(scmd);
 556
 557	if (rtn == SUCCESS) {
 558		if (!hostt->skip_settle_delay)
 559			ssleep(HOST_RESET_SETTLE_TIME);
 560		spin_lock_irqsave(host->host_lock, flags);
 561		scsi_report_bus_reset(host, scmd_channel(scmd));
 562		spin_unlock_irqrestore(host->host_lock, flags);
 563	}
 564
 565	return rtn;
 566}
 567
 568/**
 569 * scsi_try_bus_reset - ask host to perform a bus reset
 570 * @scmd:	SCSI cmd to send bus reset.
 571 */
 572static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
 573{
 574	unsigned long flags;
 575	int rtn;
 576	struct Scsi_Host *host = scmd->device->host;
 577	struct scsi_host_template *hostt = host->hostt;
 578
 579	SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
 580					  __func__));
 581
 582	if (!hostt->eh_bus_reset_handler)
 583		return FAILED;
 584
 585	rtn = hostt->eh_bus_reset_handler(scmd);
 586
 587	if (rtn == SUCCESS) {
 588		if (!hostt->skip_settle_delay)
 589			ssleep(BUS_RESET_SETTLE_TIME);
 590		spin_lock_irqsave(host->host_lock, flags);
 591		scsi_report_bus_reset(host, scmd_channel(scmd));
 592		spin_unlock_irqrestore(host->host_lock, flags);
 593	}
 594
 595	return rtn;
 596}
 597
 598static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
 599{
 600	sdev->was_reset = 1;
 601	sdev->expecting_cc_ua = 1;
 602}
 603
 604/**
 605 * scsi_try_target_reset - Ask host to perform a target reset
 606 * @scmd:	SCSI cmd used to send a target reset
 607 *
 608 * Notes:
 609 *    There is no timeout for this operation.  if this operation is
 610 *    unreliable for a given host, then the host itself needs to put a
 611 *    timer on it, and set the host back to a consistent state prior to
 612 *    returning.
 613 */
 614static int scsi_try_target_reset(struct scsi_cmnd *scmd)
 615{
 616	unsigned long flags;
 617	int rtn;
 618	struct Scsi_Host *host = scmd->device->host;
 619	struct scsi_host_template *hostt = host->hostt;
 620
 621	if (!hostt->eh_target_reset_handler)
 622		return FAILED;
 623
 624	rtn = hostt->eh_target_reset_handler(scmd);
 625	if (rtn == SUCCESS) {
 626		spin_lock_irqsave(host->host_lock, flags);
 627		__starget_for_each_device(scsi_target(scmd->device), NULL,
 628					  __scsi_report_device_reset);
 629		spin_unlock_irqrestore(host->host_lock, flags);
 630	}
 631
 632	return rtn;
 633}
 634
 635/**
 636 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
 637 * @scmd:	SCSI cmd used to send BDR
 638 *
 639 * Notes:
 640 *    There is no timeout for this operation.  if this operation is
 641 *    unreliable for a given host, then the host itself needs to put a
 642 *    timer on it, and set the host back to a consistent state prior to
 643 *    returning.
 644 */
 645static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
 646{
 647	int rtn;
 648	struct scsi_host_template *hostt = scmd->device->host->hostt;
 649
 650	if (!hostt->eh_device_reset_handler)
 651		return FAILED;
 652
 653	rtn = hostt->eh_device_reset_handler(scmd);
 654	if (rtn == SUCCESS)
 655		__scsi_report_device_reset(scmd->device, NULL);
 656	return rtn;
 657}
 658
 659static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
 660{
 661	if (!hostt->eh_abort_handler)
 662		return FAILED;
 663
 664	return hostt->eh_abort_handler(scmd);
 665}
 666
 667static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
 668{
 669	if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
 670		if (scsi_try_bus_device_reset(scmd) != SUCCESS)
 671			if (scsi_try_target_reset(scmd) != SUCCESS)
 672				if (scsi_try_bus_reset(scmd) != SUCCESS)
 673					scsi_try_host_reset(scmd);
 674}
 675
 676/**
 677 * scsi_eh_prep_cmnd  - Save a scsi command info as part of error recovery
 678 * @scmd:       SCSI command structure to hijack
 679 * @ses:        structure to save restore information
 680 * @cmnd:       CDB to send. Can be NULL if no new cmnd is needed
 681 * @cmnd_size:  size in bytes of @cmnd (must be <= BLK_MAX_CDB)
 682 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
 683 *
 684 * This function is used to save a scsi command information before re-execution
 685 * as part of the error recovery process.  If @sense_bytes is 0 the command
 686 * sent must be one that does not transfer any data.  If @sense_bytes != 0
 687 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
 688 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
 689 */
 690void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
 691			unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
 692{
 693	struct scsi_device *sdev = scmd->device;
 694
 695	/*
 696	 * We need saved copies of a number of fields - this is because
 697	 * error handling may need to overwrite these with different values
 698	 * to run different commands, and once error handling is complete,
 699	 * we will need to restore these values prior to running the actual
 700	 * command.
 701	 */
 702	ses->cmd_len = scmd->cmd_len;
 703	ses->cmnd = scmd->cmnd;
 704	ses->data_direction = scmd->sc_data_direction;
 705	ses->sdb = scmd->sdb;
 706	ses->next_rq = scmd->request->next_rq;
 707	ses->result = scmd->result;
 708	ses->underflow = scmd->underflow;
 709	ses->prot_op = scmd->prot_op;
 710
 711	scmd->prot_op = SCSI_PROT_NORMAL;
 712	scmd->cmnd = ses->eh_cmnd;
 713	memset(scmd->cmnd, 0, BLK_MAX_CDB);
 714	memset(&scmd->sdb, 0, sizeof(scmd->sdb));
 715	scmd->request->next_rq = NULL;
 716
 717	if (sense_bytes) {
 718		scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
 719					 sense_bytes);
 720		sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
 721			    scmd->sdb.length);
 722		scmd->sdb.table.sgl = &ses->sense_sgl;
 723		scmd->sc_data_direction = DMA_FROM_DEVICE;
 724		scmd->sdb.table.nents = 1;
 725		scmd->cmnd[0] = REQUEST_SENSE;
 726		scmd->cmnd[4] = scmd->sdb.length;
 727		scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
 728	} else {
 729		scmd->sc_data_direction = DMA_NONE;
 730		if (cmnd) {
 731			BUG_ON(cmnd_size > BLK_MAX_CDB);
 732			memcpy(scmd->cmnd, cmnd, cmnd_size);
 733			scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
 734		}
 735	}
 736
 737	scmd->underflow = 0;
 738
 739	if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
 740		scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
 741			(sdev->lun << 5 & 0xe0);
 742
 743	/*
 744	 * Zero the sense buffer.  The scsi spec mandates that any
 745	 * untransferred sense data should be interpreted as being zero.
 746	 */
 747	memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
 748}
 749EXPORT_SYMBOL(scsi_eh_prep_cmnd);
 750
 751/**
 752 * scsi_eh_restore_cmnd  - Restore a scsi command info as part of error recovery
 753 * @scmd:       SCSI command structure to restore
 754 * @ses:        saved information from a coresponding call to scsi_eh_prep_cmnd
 755 *
 756 * Undo any damage done by above scsi_eh_prep_cmnd().
 757 */
 758void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
 759{
 760	/*
 761	 * Restore original data
 762	 */
 763	scmd->cmd_len = ses->cmd_len;
 764	scmd->cmnd = ses->cmnd;
 765	scmd->sc_data_direction = ses->data_direction;
 766	scmd->sdb = ses->sdb;
 767	scmd->request->next_rq = ses->next_rq;
 768	scmd->result = ses->result;
 769	scmd->underflow = ses->underflow;
 770	scmd->prot_op = ses->prot_op;
 771}
 772EXPORT_SYMBOL(scsi_eh_restore_cmnd);
 773
 774/**
 775 * scsi_send_eh_cmnd  - submit a scsi command as part of error recovery
 776 * @scmd:       SCSI command structure to hijack
 777 * @cmnd:       CDB to send
 778 * @cmnd_size:  size in bytes of @cmnd
 779 * @timeout:    timeout for this request
 780 * @sense_bytes: size of sense data to copy or 0
 781 *
 782 * This function is used to send a scsi command down to a target device
 783 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
 784 *
 785 * Return value:
 786 *    SUCCESS or FAILED or NEEDS_RETRY
 787 */
 788static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
 789			     int cmnd_size, int timeout, unsigned sense_bytes)
 790{
 791	struct scsi_device *sdev = scmd->device;
 792	struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
 793	struct Scsi_Host *shost = sdev->host;
 794	DECLARE_COMPLETION_ONSTACK(done);
 795	unsigned long timeleft;
 796	struct scsi_eh_save ses;
 797	int rtn;
 798
 799	scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
 800	shost->eh_action = &done;
 801
 802	scsi_log_send(scmd);
 803	scmd->scsi_done = scsi_eh_done;
 804	shost->hostt->queuecommand(shost, scmd);
 805
 806	timeleft = wait_for_completion_timeout(&done, timeout);
 807
 808	shost->eh_action = NULL;
 809
 810	scsi_log_completion(scmd, SUCCESS);
 811
 812	SCSI_LOG_ERROR_RECOVERY(3,
 813		printk("%s: scmd: %p, timeleft: %ld\n",
 814			__func__, scmd, timeleft));
 815
 816	/*
 817	 * If there is time left scsi_eh_done got called, and we will
 818	 * examine the actual status codes to see whether the command
 819	 * actually did complete normally, else tell the host to forget
 820	 * about this command.
 821	 */
 822	if (timeleft) {
 823		rtn = scsi_eh_completed_normally(scmd);
 824		SCSI_LOG_ERROR_RECOVERY(3,
 825			printk("%s: scsi_eh_completed_normally %x\n",
 826			       __func__, rtn));
 827
 828		switch (rtn) {
 829		case SUCCESS:
 830		case NEEDS_RETRY:
 831		case FAILED:
 832		case TARGET_ERROR:
 833			break;
 834		case ADD_TO_MLQUEUE:
 835			rtn = NEEDS_RETRY;
 836			break;
 837		default:
 838			rtn = FAILED;
 839			break;
 840		}
 841	} else {
 842		scsi_abort_eh_cmnd(scmd);
 843		rtn = FAILED;
 844	}
 845
 846	scsi_eh_restore_cmnd(scmd, &ses);
 847
 848	if (sdrv && sdrv->eh_action)
 849		rtn = sdrv->eh_action(scmd, cmnd, cmnd_size, rtn);
 850
 851	return rtn;
 852}
 853
 854/**
 855 * scsi_request_sense - Request sense data from a particular target.
 856 * @scmd:	SCSI cmd for request sense.
 857 *
 858 * Notes:
 859 *    Some hosts automatically obtain this information, others require
 860 *    that we obtain it on our own. This function will *not* return until
 861 *    the command either times out, or it completes.
 862 */
 863static int scsi_request_sense(struct scsi_cmnd *scmd)
 864{
 865	return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
 866}
 867
 868/**
 869 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
 870 * @scmd:	Original SCSI cmd that eh has finished.
 871 * @done_q:	Queue for processed commands.
 872 *
 873 * Notes:
 874 *    We don't want to use the normal command completion while we are are
 875 *    still handling errors - it may cause other commands to be queued,
 876 *    and that would disturb what we are doing.  Thus we really want to
 877 *    keep a list of pending commands for final completion, and once we
 878 *    are ready to leave error handling we handle completion for real.
 879 */
 880void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
 881{
 882	scmd->device->host->host_failed--;
 883	scmd->eh_eflags = 0;
 884	list_move_tail(&scmd->eh_entry, done_q);
 885}
 886EXPORT_SYMBOL(scsi_eh_finish_cmd);
 887
 888/**
 889 * scsi_eh_get_sense - Get device sense data.
 890 * @work_q:	Queue of commands to process.
 891 * @done_q:	Queue of processed commands.
 892 *
 893 * Description:
 894 *    See if we need to request sense information.  if so, then get it
 895 *    now, so we have a better idea of what to do.
 896 *
 897 * Notes:
 898 *    This has the unfortunate side effect that if a shost adapter does
 899 *    not automatically request sense information, we end up shutting
 900 *    it down before we request it.
 901 *
 902 *    All drivers should request sense information internally these days,
 903 *    so for now all I have to say is tough noogies if you end up in here.
 904 *
 905 *    XXX: Long term this code should go away, but that needs an audit of
 906 *         all LLDDs first.
 907 */
 908int scsi_eh_get_sense(struct list_head *work_q,
 909		      struct list_head *done_q)
 910{
 911	struct scsi_cmnd *scmd, *next;
 912	int rtn;
 913
 914	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
 915		if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
 916		    SCSI_SENSE_VALID(scmd))
 917			continue;
 918
 919		SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
 920						  "%s: requesting sense\n",
 921						  current->comm));
 922		rtn = scsi_request_sense(scmd);
 923		if (rtn != SUCCESS)
 924			continue;
 925
 926		SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
 927						  " result %x\n", scmd,
 928						  scmd->result));
 929		SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
 930
 931		rtn = scsi_decide_disposition(scmd);
 932
 933		/*
 934		 * if the result was normal, then just pass it along to the
 935		 * upper level.
 936		 */
 937		if (rtn == SUCCESS)
 938			/* we don't want this command reissued, just
 939			 * finished with the sense data, so set
 940			 * retries to the max allowed to ensure it
 941			 * won't get reissued */
 942			scmd->retries = scmd->allowed;
 943		else if (rtn != NEEDS_RETRY)
 944			continue;
 945
 946		scsi_eh_finish_cmd(scmd, done_q);
 947	}
 948
 949	return list_empty(work_q);
 950}
 951EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
 952
 953/**
 954 * scsi_eh_tur - Send TUR to device.
 955 * @scmd:	&scsi_cmnd to send TUR
 956 *
 957 * Return value:
 958 *    0 - Device is ready. 1 - Device NOT ready.
 959 */
 960static int scsi_eh_tur(struct scsi_cmnd *scmd)
 961{
 962	static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
 963	int retry_cnt = 1, rtn;
 964
 965retry_tur:
 966	rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
 967
 968	SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
 969		__func__, scmd, rtn));
 970
 971	switch (rtn) {
 972	case NEEDS_RETRY:
 973		if (retry_cnt--)
 974			goto retry_tur;
 975		/*FALLTHRU*/
 976	case SUCCESS:
 977		return 0;
 978	default:
 979		return 1;
 980	}
 981}
 982
 983/**
 984 * scsi_eh_test_devices - check if devices are responding from error recovery.
 985 * @cmd_list:	scsi commands in error recovery.
 986 * @work_q:     queue for commands which still need more error recovery
 987 * @done_q:     queue for commands which are finished
 988 * @try_stu:    boolean on if a STU command should be tried in addition to TUR.
 989 *
 990 * Decription:
 991 *    Tests if devices are in a working state.  Commands to devices now in
 992 *    a working state are sent to the done_q while commands to devices which
 993 *    are still failing to respond are returned to the work_q for more
 994 *    processing.
 995 **/
 996static int scsi_eh_test_devices(struct list_head *cmd_list,
 997				struct list_head *work_q,
 998				struct list_head *done_q, int try_stu)
 999{
1000	struct scsi_cmnd *scmd, *next;
1001	struct scsi_device *sdev;
1002	int finish_cmds;
1003
1004	while (!list_empty(cmd_list)) {
1005		scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1006		sdev = scmd->device;
1007
1008		finish_cmds = !scsi_device_online(scmd->device) ||
1009			(try_stu && !scsi_eh_try_stu(scmd) &&
1010			 !scsi_eh_tur(scmd)) ||
1011			!scsi_eh_tur(scmd);
1012
1013		list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1014			if (scmd->device == sdev) {
1015				if (finish_cmds)
1016					scsi_eh_finish_cmd(scmd, done_q);
1017				else
1018					list_move_tail(&scmd->eh_entry, work_q);
1019			}
1020	}
1021	return list_empty(work_q);
1022}
1023
1024
1025/**
1026 * scsi_eh_abort_cmds - abort pending commands.
1027 * @work_q:	&list_head for pending commands.
1028 * @done_q:	&list_head for processed commands.
1029 *
1030 * Decription:
1031 *    Try and see whether or not it makes sense to try and abort the
1032 *    running command.  This only works out to be the case if we have one
1033 *    command that has timed out.  If the command simply failed, it makes
1034 *    no sense to try and abort the command, since as far as the shost
1035 *    adapter is concerned, it isn't running.
1036 */
1037static int scsi_eh_abort_cmds(struct list_head *work_q,
1038			      struct list_head *done_q)
1039{
1040	struct scsi_cmnd *scmd, *next;
1041	LIST_HEAD(check_list);
1042	int rtn;
1043
1044	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1045		if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1046			continue;
1047		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1048						  "0x%p\n", current->comm,
1049						  scmd));
1050		rtn = scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd);
1051		if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1052			scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1053			if (rtn == FAST_IO_FAIL)
1054				scsi_eh_finish_cmd(scmd, done_q);
1055			else
1056				list_move_tail(&scmd->eh_entry, &check_list);
1057		} else
1058			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1059							  " cmd failed:"
1060							  "0x%p\n",
1061							  current->comm,
1062							  scmd));
1063	}
1064
1065	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1066}
1067
1068/**
1069 * scsi_eh_try_stu - Send START_UNIT to device.
1070 * @scmd:	&scsi_cmnd to send START_UNIT
1071 *
1072 * Return value:
1073 *    0 - Device is ready. 1 - Device NOT ready.
1074 */
1075static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1076{
1077	static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1078
1079	if (scmd->device->allow_restart) {
1080		int i, rtn = NEEDS_RETRY;
1081
1082		for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1083			rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1084
1085		if (rtn == SUCCESS)
1086			return 0;
1087	}
1088
1089	return 1;
1090}
1091
1092 /**
1093 * scsi_eh_stu - send START_UNIT if needed
1094 * @shost:	&scsi host being recovered.
1095 * @work_q:     &list_head for pending commands.
1096 * @done_q:	&list_head for processed commands.
1097 *
1098 * Notes:
1099 *    If commands are failing due to not ready, initializing command required,
1100 *	try revalidating the device, which will end up sending a start unit.
1101 */
1102static int scsi_eh_stu(struct Scsi_Host *shost,
1103			      struct list_head *work_q,
1104			      struct list_head *done_q)
1105{
1106	struct scsi_cmnd *scmd, *stu_scmd, *next;
1107	struct scsi_device *sdev;
1108
1109	shost_for_each_device(sdev, shost) {
1110		stu_scmd = NULL;
1111		list_for_each_entry(scmd, work_q, eh_entry)
1112			if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1113			    scsi_check_sense(scmd) == FAILED ) {
1114				stu_scmd = scmd;
1115				break;
1116			}
1117
1118		if (!stu_scmd)
1119			continue;
1120
1121		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1122						  " 0x%p\n", current->comm, sdev));
1123
1124		if (!scsi_eh_try_stu(stu_scmd)) {
1125			if (!scsi_device_online(sdev) ||
1126			    !scsi_eh_tur(stu_scmd)) {
1127				list_for_each_entry_safe(scmd, next,
1128							  work_q, eh_entry) {
1129					if (scmd->device == sdev)
1130						scsi_eh_finish_cmd(scmd, done_q);
1131				}
1132			}
1133		} else {
1134			SCSI_LOG_ERROR_RECOVERY(3,
1135						printk("%s: START_UNIT failed to sdev:"
1136						       " 0x%p\n", current->comm, sdev));
1137		}
1138	}
1139
1140	return list_empty(work_q);
1141}
1142
1143
1144/**
1145 * scsi_eh_bus_device_reset - send bdr if needed
1146 * @shost:	scsi host being recovered.
1147 * @work_q:     &list_head for pending commands.
1148 * @done_q:	&list_head for processed commands.
1149 *
1150 * Notes:
1151 *    Try a bus device reset.  Still, look to see whether we have multiple
1152 *    devices that are jammed or not - if we have multiple devices, it
1153 *    makes no sense to try bus_device_reset - we really would need to try
1154 *    a bus_reset instead.
1155 */
1156static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1157				    struct list_head *work_q,
1158				    struct list_head *done_q)
1159{
1160	struct scsi_cmnd *scmd, *bdr_scmd, *next;
1161	struct scsi_device *sdev;
1162	int rtn;
1163
1164	shost_for_each_device(sdev, shost) {
1165		bdr_scmd = NULL;
1166		list_for_each_entry(scmd, work_q, eh_entry)
1167			if (scmd->device == sdev) {
1168				bdr_scmd = scmd;
1169				break;
1170			}
1171
1172		if (!bdr_scmd)
1173			continue;
1174
1175		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1176						  " 0x%p\n", current->comm,
1177						  sdev));
1178		rtn = scsi_try_bus_device_reset(bdr_scmd);
1179		if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1180			if (!scsi_device_online(sdev) ||
1181			    rtn == FAST_IO_FAIL ||
1182			    !scsi_eh_tur(bdr_scmd)) {
1183				list_for_each_entry_safe(scmd, next,
1184							 work_q, eh_entry) {
1185					if (scmd->device == sdev)
1186						scsi_eh_finish_cmd(scmd,
1187								   done_q);
1188				}
1189			}
1190		} else {
1191			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1192							  " failed sdev:"
1193							  "0x%p\n",
1194							  current->comm,
1195							   sdev));
1196		}
1197	}
1198
1199	return list_empty(work_q);
1200}
1201
1202/**
1203 * scsi_eh_target_reset - send target reset if needed
1204 * @shost:	scsi host being recovered.
1205 * @work_q:     &list_head for pending commands.
1206 * @done_q:	&list_head for processed commands.
1207 *
1208 * Notes:
1209 *    Try a target reset.
1210 */
1211static int scsi_eh_target_reset(struct Scsi_Host *shost,
1212				struct list_head *work_q,
1213				struct list_head *done_q)
1214{
1215	LIST_HEAD(tmp_list);
1216	LIST_HEAD(check_list);
1217
1218	list_splice_init(work_q, &tmp_list);
1219
1220	while (!list_empty(&tmp_list)) {
1221		struct scsi_cmnd *next, *scmd;
1222		int rtn;
1223		unsigned int id;
1224
1225		scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1226		id = scmd_id(scmd);
1227
1228		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1229						  "to target %d\n",
1230						  current->comm, id));
1231		rtn = scsi_try_target_reset(scmd);
1232		if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1233			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1234							  " failed target: "
1235							  "%d\n",
1236							  current->comm, id));
1237		list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1238			if (scmd_id(scmd) != id)
1239				continue;
1240
1241			if (rtn == SUCCESS)
1242				list_move_tail(&scmd->eh_entry, &check_list);
1243			else if (rtn == FAST_IO_FAIL)
1244				scsi_eh_finish_cmd(scmd, done_q);
1245			else
1246				/* push back on work queue for further processing */
1247				list_move(&scmd->eh_entry, work_q);
1248		}
1249	}
1250
1251	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1252}
1253
1254/**
1255 * scsi_eh_bus_reset - send a bus reset
1256 * @shost:	&scsi host being recovered.
1257 * @work_q:     &list_head for pending commands.
1258 * @done_q:	&list_head for processed commands.
1259 */
1260static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1261			     struct list_head *work_q,
1262			     struct list_head *done_q)
1263{
1264	struct scsi_cmnd *scmd, *chan_scmd, *next;
1265	LIST_HEAD(check_list);
1266	unsigned int channel;
1267	int rtn;
1268
1269	/*
1270	 * we really want to loop over the various channels, and do this on
1271	 * a channel by channel basis.  we should also check to see if any
1272	 * of the failed commands are on soft_reset devices, and if so, skip
1273	 * the reset.
1274	 */
1275
1276	for (channel = 0; channel <= shost->max_channel; channel++) {
1277		chan_scmd = NULL;
1278		list_for_each_entry(scmd, work_q, eh_entry) {
1279			if (channel == scmd_channel(scmd)) {
1280				chan_scmd = scmd;
1281				break;
1282				/*
1283				 * FIXME add back in some support for
1284				 * soft_reset devices.
1285				 */
1286			}
1287		}
1288
1289		if (!chan_scmd)
1290			continue;
1291		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1292						  " %d\n", current->comm,
1293						  channel));
1294		rtn = scsi_try_bus_reset(chan_scmd);
1295		if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1296			list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1297				if (channel == scmd_channel(scmd)) {
1298					if (rtn == FAST_IO_FAIL)
1299						scsi_eh_finish_cmd(scmd,
1300								   done_q);
1301					else
1302						list_move_tail(&scmd->eh_entry,
1303							       &check_list);
1304				}
1305			}
1306		} else {
1307			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1308							  " failed chan: %d\n",
1309							  current->comm,
1310							  channel));
1311		}
1312	}
1313	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1314}
1315
1316/**
1317 * scsi_eh_host_reset - send a host reset
1318 * @work_q:	list_head for processed commands.
1319 * @done_q:	list_head for processed commands.
1320 */
1321static int scsi_eh_host_reset(struct list_head *work_q,
1322			      struct list_head *done_q)
1323{
1324	struct scsi_cmnd *scmd, *next;
1325	LIST_HEAD(check_list);
1326	int rtn;
1327
1328	if (!list_empty(work_q)) {
1329		scmd = list_entry(work_q->next,
1330				  struct scsi_cmnd, eh_entry);
1331
1332		SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1333						  , current->comm));
1334
1335		rtn = scsi_try_host_reset(scmd);
1336		if (rtn == SUCCESS) {
1337			list_splice_init(work_q, &check_list);
1338		} else if (rtn == FAST_IO_FAIL) {
1339			list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1340					scsi_eh_finish_cmd(scmd, done_q);
1341			}
1342		} else {
1343			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1344							  " failed\n",
1345							  current->comm));
1346		}
1347	}
1348	return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1349}
1350
1351/**
1352 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1353 * @work_q:	list_head for processed commands.
1354 * @done_q:	list_head for processed commands.
1355 */
1356static void scsi_eh_offline_sdevs(struct list_head *work_q,
1357				  struct list_head *done_q)
1358{
1359	struct scsi_cmnd *scmd, *next;
1360
1361	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1362		sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1363			    "not ready after error recovery\n");
1364		scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1365		if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1366			/*
1367			 * FIXME: Handle lost cmds.
1368			 */
1369		}
1370		scsi_eh_finish_cmd(scmd, done_q);
1371	}
1372	return;
1373}
1374
1375/**
1376 * scsi_noretry_cmd - determinte if command should be failed fast
1377 * @scmd:	SCSI cmd to examine.
1378 */
1379int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1380{
1381	switch (host_byte(scmd->result)) {
1382	case DID_OK:
1383		break;
1384	case DID_BUS_BUSY:
1385		return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1386	case DID_PARITY:
1387		return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1388	case DID_ERROR:
1389		if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1390		    status_byte(scmd->result) == RESERVATION_CONFLICT)
1391			return 0;
1392		/* fall through */
1393	case DID_SOFT_ERROR:
1394		return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1395	}
1396
1397	switch (status_byte(scmd->result)) {
1398	case CHECK_CONDITION:
1399		/*
1400		 * assume caller has checked sense and determinted
1401		 * the check condition was retryable.
1402		 */
1403		if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1404		    scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1405			return 1;
1406	}
1407
1408	return 0;
1409}
1410
1411/**
1412 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1413 * @scmd:	SCSI cmd to examine.
1414 *
1415 * Notes:
1416 *    This is *only* called when we are examining the status after sending
1417 *    out the actual data command.  any commands that are queued for error
1418 *    recovery (e.g. test_unit_ready) do *not* come through here.
1419 *
1420 *    When this routine returns failed, it means the error handler thread
1421 *    is woken.  In cases where the error code indicates an error that
1422 *    doesn't require the error handler read (i.e. we don't need to
1423 *    abort/reset), this function should return SUCCESS.
1424 */
1425int scsi_decide_disposition(struct scsi_cmnd *scmd)
1426{
1427	int rtn;
1428
1429	/*
1430	 * if the device is offline, then we clearly just pass the result back
1431	 * up to the top level.
1432	 */
1433	if (!scsi_device_online(scmd->device)) {
1434		SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1435						  " as SUCCESS\n",
1436						  __func__));
1437		return SUCCESS;
1438	}
1439
1440	/*
1441	 * first check the host byte, to see if there is anything in there
1442	 * that would indicate what we need to do.
1443	 */
1444	switch (host_byte(scmd->result)) {
1445	case DID_PASSTHROUGH:
1446		/*
1447		 * no matter what, pass this through to the upper layer.
1448		 * nuke this special code so that it looks like we are saying
1449		 * did_ok.
1450		 */
1451		scmd->result &= 0xff00ffff;
1452		return SUCCESS;
1453	case DID_OK:
1454		/*
1455		 * looks good.  drop through, and check the next byte.
1456		 */
1457		break;
1458	case DID_NO_CONNECT:
1459	case DID_BAD_TARGET:
1460	case DID_ABORT:
1461		/*
1462		 * note - this means that we just report the status back
1463		 * to the top level driver, not that we actually think
1464		 * that it indicates SUCCESS.
1465		 */
1466		return SUCCESS;
1467		/*
1468		 * when the low level driver returns did_soft_error,
1469		 * it is responsible for keeping an internal retry counter
1470		 * in order to avoid endless loops (db)
1471		 *
1472		 * actually this is a bug in this function here.  we should
1473		 * be mindful of the maximum number of retries specified
1474		 * and not get stuck in a loop.
1475		 */
1476	case DID_SOFT_ERROR:
1477		goto maybe_retry;
1478	case DID_IMM_RETRY:
1479		return NEEDS_RETRY;
1480
1481	case DID_REQUEUE:
1482		return ADD_TO_MLQUEUE;
1483	case DID_TRANSPORT_DISRUPTED:
1484		/*
1485		 * LLD/transport was disrupted during processing of the IO.
1486		 * The transport class is now blocked/blocking,
1487		 * and the transport will decide what to do with the IO
1488		 * based on its timers and recovery capablilities if
1489		 * there are enough retries.
1490		 */
1491		goto maybe_retry;
1492	case DID_TRANSPORT_FAILFAST:
1493		/*
1494		 * The transport decided to failfast the IO (most likely
1495		 * the fast io fail tmo fired), so send IO directly upwards.
1496		 */
1497		return SUCCESS;
1498	case DID_ERROR:
1499		if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1500		    status_byte(scmd->result) == RESERVATION_CONFLICT)
1501			/*
1502			 * execute reservation conflict processing code
1503			 * lower down
1504			 */
1505			break;
1506		/* fallthrough */
1507	case DID_BUS_BUSY:
1508	case DID_PARITY:
1509		goto maybe_retry;
1510	case DID_TIME_OUT:
1511		/*
1512		 * when we scan the bus, we get timeout messages for
1513		 * these commands if there is no device available.
1514		 * other hosts report did_no_connect for the same thing.
1515		 */
1516		if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1517		     scmd->cmnd[0] == INQUIRY)) {
1518			return SUCCESS;
1519		} else {
1520			return FAILED;
1521		}
1522	case DID_RESET:
1523		return SUCCESS;
1524	default:
1525		return FAILED;
1526	}
1527
1528	/*
1529	 * next, check the message byte.
1530	 */
1531	if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1532		return FAILED;
1533
1534	/*
1535	 * check the status byte to see if this indicates anything special.
1536	 */
1537	switch (status_byte(scmd->result)) {
1538	case QUEUE_FULL:
1539		scsi_handle_queue_full(scmd->device);
1540		/*
1541		 * the case of trying to send too many commands to a
1542		 * tagged queueing device.
1543		 */
1544	case BUSY:
1545		/*
1546		 * device can't talk to us at the moment.  Should only
1547		 * occur (SAM-3) when the task queue is empty, so will cause
1548		 * the empty queue handling to trigger a stall in the
1549		 * device.
1550		 */
1551		return ADD_TO_MLQUEUE;
1552	case GOOD:
1553		scsi_handle_queue_ramp_up(scmd->device);
1554	case COMMAND_TERMINATED:
1555		return SUCCESS;
1556	case TASK_ABORTED:
1557		goto maybe_retry;
1558	case CHECK_CONDITION:
1559		rtn = scsi_check_sense(scmd);
1560		if (rtn == NEEDS_RETRY)
1561			goto maybe_retry;
1562		else if (rtn == TARGET_ERROR) {
1563			/*
1564			 * Need to modify host byte to signal a
1565			 * permanent target failure
1566			 */
1567			set_host_byte(scmd, DID_TARGET_FAILURE);
1568			rtn = SUCCESS;
1569		}
1570		/* if rtn == FAILED, we have no sense information;
1571		 * returning FAILED will wake the error handler thread
1572		 * to collect the sense and redo the decide
1573		 * disposition */
1574		return rtn;
1575	case CONDITION_GOOD:
1576	case INTERMEDIATE_GOOD:
1577	case INTERMEDIATE_C_GOOD:
1578	case ACA_ACTIVE:
1579		/*
1580		 * who knows?  FIXME(eric)
1581		 */
1582		return SUCCESS;
1583
1584	case RESERVATION_CONFLICT:
1585		sdev_printk(KERN_INFO, scmd->device,
1586			    "reservation conflict\n");
1587		set_host_byte(scmd, DID_NEXUS_FAILURE);
1588		return SUCCESS; /* causes immediate i/o error */
1589	default:
1590		return FAILED;
1591	}
1592	return FAILED;
1593
1594      maybe_retry:
1595
1596	/* we requeue for retry because the error was retryable, and
1597	 * the request was not marked fast fail.  Note that above,
1598	 * even if the request is marked fast fail, we still requeue
1599	 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1600	if ((++scmd->retries) <= scmd->allowed
1601	    && !scsi_noretry_cmd(scmd)) {
1602		return NEEDS_RETRY;
1603	} else {
1604		/*
1605		 * no more retries - report this one back to upper level.
1606		 */
1607		return SUCCESS;
1608	}
1609}
1610
1611static void eh_lock_door_done(struct request *req, int uptodate)
1612{
1613	__blk_put_request(req->q, req);
1614}
1615
1616/**
1617 * scsi_eh_lock_door - Prevent medium removal for the specified device
1618 * @sdev:	SCSI device to prevent medium removal
1619 *
1620 * Locking:
1621 * 	We must be called from process context.
1622 *
1623 * Notes:
1624 * 	We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1625 * 	head of the devices request queue, and continue.
1626 */
1627static void scsi_eh_lock_door(struct scsi_device *sdev)
1628{
1629	struct request *req;
1630
1631	/*
1632	 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1633	 * request becomes available
1634	 */
1635	req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1636
1637	req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1638	req->cmd[1] = 0;
1639	req->cmd[2] = 0;
1640	req->cmd[3] = 0;
1641	req->cmd[4] = SCSI_REMOVAL_PREVENT;
1642	req->cmd[5] = 0;
1643
1644	req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1645
1646	req->cmd_type = REQ_TYPE_BLOCK_PC;
1647	req->cmd_flags |= REQ_QUIET;
1648	req->timeout = 10 * HZ;
1649	req->retries = 5;
1650
1651	blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1652}
1653
1654/**
1655 * scsi_restart_operations - restart io operations to the specified host.
1656 * @shost:	Host we are restarting.
1657 *
1658 * Notes:
1659 *    When we entered the error handler, we blocked all further i/o to
1660 *    this device.  we need to 'reverse' this process.
1661 */
1662static void scsi_restart_operations(struct Scsi_Host *shost)
1663{
1664	struct scsi_device *sdev;
1665	unsigned long flags;
1666
1667	/*
1668	 * If the door was locked, we need to insert a door lock request
1669	 * onto the head of the SCSI request queue for the device.  There
1670	 * is no point trying to lock the door of an off-line device.
1671	 */
1672	shost_for_each_device(sdev, shost) {
1673		if (scsi_device_online(sdev) && sdev->locked)
1674			scsi_eh_lock_door(sdev);
1675	}
1676
1677	/*
1678	 * next free up anything directly waiting upon the host.  this
1679	 * will be requests for character device operations, and also for
1680	 * ioctls to queued block devices.
1681	 */
1682	SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1683					  __func__));
1684
1685	spin_lock_irqsave(shost->host_lock, flags);
1686	if (scsi_host_set_state(shost, SHOST_RUNNING))
1687		if (scsi_host_set_state(shost, SHOST_CANCEL))
1688			BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1689	spin_unlock_irqrestore(shost->host_lock, flags);
1690
1691	wake_up(&shost->host_wait);
1692
1693	/*
1694	 * finally we need to re-initiate requests that may be pending.  we will
1695	 * have had everything blocked while error handling is taking place, and
1696	 * now that error recovery is done, we will need to ensure that these
1697	 * requests are started.
1698	 */
1699	scsi_run_host_queues(shost);
1700
1701	/*
1702	 * if eh is active and host_eh_scheduled is pending we need to re-run
1703	 * recovery.  we do this check after scsi_run_host_queues() to allow
1704	 * everything pent up since the last eh run a chance to make forward
1705	 * progress before we sync again.  Either we'll immediately re-run
1706	 * recovery or scsi_device_unbusy() will wake us again when these
1707	 * pending commands complete.
1708	 */
1709	spin_lock_irqsave(shost->host_lock, flags);
1710	if (shost->host_eh_scheduled)
1711		if (scsi_host_set_state(shost, SHOST_RECOVERY))
1712			WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
1713	spin_unlock_irqrestore(shost->host_lock, flags);
1714}
1715
1716/**
1717 * scsi_eh_ready_devs - check device ready state and recover if not.
1718 * @shost: 	host to be recovered.
1719 * @work_q:     &list_head for pending commands.
1720 * @done_q:	&list_head for processed commands.
1721 */
1722void scsi_eh_ready_devs(struct Scsi_Host *shost,
1723			struct list_head *work_q,
1724			struct list_head *done_q)
1725{
1726	if (!scsi_eh_stu(shost, work_q, done_q))
1727		if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1728			if (!scsi_eh_target_reset(shost, work_q, done_q))
1729				if (!scsi_eh_bus_reset(shost, work_q, done_q))
1730					if (!scsi_eh_host_reset(work_q, done_q))
1731						scsi_eh_offline_sdevs(work_q,
1732								      done_q);
1733}
1734EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1735
1736/**
1737 * scsi_eh_flush_done_q - finish processed commands or retry them.
1738 * @done_q:	list_head of processed commands.
1739 */
1740void scsi_eh_flush_done_q(struct list_head *done_q)
1741{
1742	struct scsi_cmnd *scmd, *next;
1743
1744	list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1745		list_del_init(&scmd->eh_entry);
1746		if (scsi_device_online(scmd->device) &&
1747		    !scsi_noretry_cmd(scmd) &&
1748		    (++scmd->retries <= scmd->allowed)) {
1749			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1750							  " retry cmd: %p\n",
1751							  current->comm,
1752							  scmd));
1753				scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1754		} else {
1755			/*
1756			 * If just we got sense for the device (called
1757			 * scsi_eh_get_sense), scmd->result is already
1758			 * set, do not set DRIVER_TIMEOUT.
1759			 */
1760			if (!scmd->result)
1761				scmd->result |= (DRIVER_TIMEOUT << 24);
1762			SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1763							" cmd: %p\n",
1764							current->comm, scmd));
1765			scsi_finish_command(scmd);
1766		}
1767	}
1768}
1769EXPORT_SYMBOL(scsi_eh_flush_done_q);
1770
1771/**
1772 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1773 * @shost:	Host to unjam.
1774 *
1775 * Notes:
1776 *    When we come in here, we *know* that all commands on the bus have
1777 *    either completed, failed or timed out.  we also know that no further
1778 *    commands are being sent to the host, so things are relatively quiet
1779 *    and we have freedom to fiddle with things as we wish.
1780 *
1781 *    This is only the *default* implementation.  it is possible for
1782 *    individual drivers to supply their own version of this function, and
1783 *    if the maintainer wishes to do this, it is strongly suggested that
1784 *    this function be taken as a template and modified.  this function
1785 *    was designed to correctly handle problems for about 95% of the
1786 *    different cases out there, and it should always provide at least a
1787 *    reasonable amount of error recovery.
1788 *
1789 *    Any command marked 'failed' or 'timeout' must eventually have
1790 *    scsi_finish_cmd() called for it.  we do all of the retry stuff
1791 *    here, so when we restart the host after we return it should have an
1792 *    empty queue.
1793 */
1794static void scsi_unjam_host(struct Scsi_Host *shost)
1795{
1796	unsigned long flags;
1797	LIST_HEAD(eh_work_q);
1798	LIST_HEAD(eh_done_q);
1799
1800	spin_lock_irqsave(shost->host_lock, flags);
1801	list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1802	spin_unlock_irqrestore(shost->host_lock, flags);
1803
1804	SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1805
1806	if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1807		if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1808			scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1809
1810	scsi_eh_flush_done_q(&eh_done_q);
1811}
1812
1813/**
1814 * scsi_error_handler - SCSI error handler thread
1815 * @data:	Host for which we are running.
1816 *
1817 * Notes:
1818 *    This is the main error handling loop.  This is run as a kernel thread
1819 *    for every SCSI host and handles all error handling activity.
1820 */
1821int scsi_error_handler(void *data)
1822{
1823	struct Scsi_Host *shost = data;
1824
1825	/*
1826	 * We use TASK_INTERRUPTIBLE so that the thread is not
1827	 * counted against the load average as a running process.
1828	 * We never actually get interrupted because kthread_run
1829	 * disables signal delivery for the created thread.
1830	 */
1831	set_current_state(TASK_INTERRUPTIBLE);
1832	while (!kthread_should_stop()) {
1833		if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1834		    shost->host_failed != shost->host_busy) {
1835			SCSI_LOG_ERROR_RECOVERY(1,
1836				printk("Error handler scsi_eh_%d sleeping\n",
1837					shost->host_no));
1838			schedule();
1839			set_current_state(TASK_INTERRUPTIBLE);
1840			continue;
1841		}
1842
1843		__set_current_state(TASK_RUNNING);
1844		SCSI_LOG_ERROR_RECOVERY(1,
1845			printk("Error handler scsi_eh_%d waking up\n",
1846				shost->host_no));
1847
1848		/*
1849		 * We have a host that is failing for some reason.  Figure out
1850		 * what we need to do to get it up and online again (if we can).
1851		 * If we fail, we end up taking the thing offline.
1852		 */
1853		if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
1854			SCSI_LOG_ERROR_RECOVERY(1,
1855				printk(KERN_ERR "Error handler scsi_eh_%d "
1856						"unable to autoresume\n",
1857						shost->host_no));
1858			continue;
1859		}
1860
1861		if (shost->transportt->eh_strategy_handler)
1862			shost->transportt->eh_strategy_handler(shost);
1863		else
1864			scsi_unjam_host(shost);
1865
1866		/*
1867		 * Note - if the above fails completely, the action is to take
1868		 * individual devices offline and flush the queue of any
1869		 * outstanding requests that may have been pending.  When we
1870		 * restart, we restart any I/O to any other devices on the bus
1871		 * which are still online.
1872		 */
1873		scsi_restart_operations(shost);
1874		if (!shost->eh_noresume)
1875			scsi_autopm_put_host(shost);
1876		set_current_state(TASK_INTERRUPTIBLE);
1877	}
1878	__set_current_state(TASK_RUNNING);
1879
1880	SCSI_LOG_ERROR_RECOVERY(1,
1881		printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1882	shost->ehandler = NULL;
1883	return 0;
1884}
1885
1886/*
1887 * Function:    scsi_report_bus_reset()
1888 *
1889 * Purpose:     Utility function used by low-level drivers to report that
1890 *		they have observed a bus reset on the bus being handled.
1891 *
1892 * Arguments:   shost       - Host in question
1893 *		channel     - channel on which reset was observed.
1894 *
1895 * Returns:     Nothing
1896 *
1897 * Lock status: Host lock must be held.
1898 *
1899 * Notes:       This only needs to be called if the reset is one which
1900 *		originates from an unknown location.  Resets originated
1901 *		by the mid-level itself don't need to call this, but there
1902 *		should be no harm.
1903 *
1904 *		The main purpose of this is to make sure that a CHECK_CONDITION
1905 *		is properly treated.
1906 */
1907void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1908{
1909	struct scsi_device *sdev;
1910
1911	__shost_for_each_device(sdev, shost) {
1912		if (channel == sdev_channel(sdev))
1913			__scsi_report_device_reset(sdev, NULL);
1914	}
1915}
1916EXPORT_SYMBOL(scsi_report_bus_reset);
1917
1918/*
1919 * Function:    scsi_report_device_reset()
1920 *
1921 * Purpose:     Utility function used by low-level drivers to report that
1922 *		they have observed a device reset on the device being handled.
1923 *
1924 * Arguments:   shost       - Host in question
1925 *		channel     - channel on which reset was observed
1926 *		target	    - target on which reset was observed
1927 *
1928 * Returns:     Nothing
1929 *
1930 * Lock status: Host lock must be held
1931 *
1932 * Notes:       This only needs to be called if the reset is one which
1933 *		originates from an unknown location.  Resets originated
1934 *		by the mid-level itself don't need to call this, but there
1935 *		should be no harm.
1936 *
1937 *		The main purpose of this is to make sure that a CHECK_CONDITION
1938 *		is properly treated.
1939 */
1940void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1941{
1942	struct scsi_device *sdev;
1943
1944	__shost_for_each_device(sdev, shost) {
1945		if (channel == sdev_channel(sdev) &&
1946		    target == sdev_id(sdev))
1947			__scsi_report_device_reset(sdev, NULL);
1948	}
1949}
1950EXPORT_SYMBOL(scsi_report_device_reset);
1951
1952static void
1953scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1954{
1955}
1956
1957/*
1958 * Function:	scsi_reset_provider
1959 *
1960 * Purpose:	Send requested reset to a bus or device at any phase.
1961 *
1962 * Arguments:	device	- device to send reset to
1963 *		flag - reset type (see scsi.h)
1964 *
1965 * Returns:	SUCCESS/FAILURE.
1966 *
1967 * Notes:	This is used by the SCSI Generic driver to provide
1968 *		Bus/Device reset capability.
1969 */
1970int
1971scsi_reset_provider(struct scsi_device *dev, int flag)
1972{
1973	struct scsi_cmnd *scmd;
1974	struct Scsi_Host *shost = dev->host;
1975	struct request req;
1976	unsigned long flags;
1977	int rtn;
1978
1979	if (scsi_autopm_get_host(shost) < 0)
1980		return FAILED;
1981
1982	scmd = scsi_get_command(dev, GFP_KERNEL);
1983	blk_rq_init(NULL, &req);
1984	scmd->request = &req;
1985
1986	scmd->cmnd = req.cmd;
1987
1988	scmd->scsi_done		= scsi_reset_provider_done_command;
1989	memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1990
1991	scmd->cmd_len			= 0;
1992
1993	scmd->sc_data_direction		= DMA_BIDIRECTIONAL;
1994
1995	spin_lock_irqsave(shost->host_lock, flags);
1996	shost->tmf_in_progress = 1;
1997	spin_unlock_irqrestore(shost->host_lock, flags);
1998
1999	switch (flag) {
2000	case SCSI_TRY_RESET_DEVICE:
2001		rtn = scsi_try_bus_device_reset(scmd);
2002		if (rtn == SUCCESS)
2003			break;
2004		/* FALLTHROUGH */
2005	case SCSI_TRY_RESET_TARGET:
2006		rtn = scsi_try_target_reset(scmd);
2007		if (rtn == SUCCESS)
2008			break;
2009		/* FALLTHROUGH */
2010	case SCSI_TRY_RESET_BUS:
2011		rtn = scsi_try_bus_reset(scmd);
2012		if (rtn == SUCCESS)
2013			break;
2014		/* FALLTHROUGH */
2015	case SCSI_TRY_RESET_HOST:
2016		rtn = scsi_try_host_reset(scmd);
2017		break;
2018	default:
2019		rtn = FAILED;
2020	}
2021
2022	spin_lock_irqsave(shost->host_lock, flags);
2023	shost->tmf_in_progress = 0;
2024	spin_unlock_irqrestore(shost->host_lock, flags);
2025
2026	/*
2027	 * be sure to wake up anyone who was sleeping or had their queue
2028	 * suspended while we performed the TMF.
2029	 */
2030	SCSI_LOG_ERROR_RECOVERY(3,
2031		printk("%s: waking up host to restart after TMF\n",
2032		__func__));
2033
2034	wake_up(&shost->host_wait);
2035
2036	scsi_run_host_queues(shost);
2037
2038	scsi_next_command(scmd);
2039	scsi_autopm_put_host(shost);
2040	return rtn;
2041}
2042EXPORT_SYMBOL(scsi_reset_provider);
2043
2044/**
2045 * scsi_normalize_sense - normalize main elements from either fixed or
2046 *			descriptor sense data format into a common format.
2047 *
2048 * @sense_buffer:	byte array containing sense data returned by device
2049 * @sb_len:		number of valid bytes in sense_buffer
2050 * @sshdr:		pointer to instance of structure that common
2051 *			elements are written to.
2052 *
2053 * Notes:
2054 *	The "main elements" from sense data are: response_code, sense_key,
2055 *	asc, ascq and additional_length (only for descriptor format).
2056 *
2057 *	Typically this function can be called after a device has
2058 *	responded to a SCSI command with the CHECK_CONDITION status.
2059 *
2060 * Return value:
2061 *	1 if valid sense data information found, else 0;
2062 */
2063int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2064                         struct scsi_sense_hdr *sshdr)
2065{
2066	if (!sense_buffer || !sb_len)
2067		return 0;
2068
2069	memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2070
2071	sshdr->response_code = (sense_buffer[0] & 0x7f);
2072
2073	if (!scsi_sense_valid(sshdr))
2074		return 0;
2075
2076	if (sshdr->response_code >= 0x72) {
2077		/*
2078		 * descriptor format
2079		 */
2080		if (sb_len > 1)
2081			sshdr->sense_key = (sense_buffer[1] & 0xf);
2082		if (sb_len > 2)
2083			sshdr->asc = sense_buffer[2];
2084		if (sb_len > 3)
2085			sshdr->ascq = sense_buffer[3];
2086		if (sb_len > 7)
2087			sshdr->additional_length = sense_buffer[7];
2088	} else {
2089		/*
2090		 * fixed format
2091		 */
2092		if (sb_len > 2)
2093			sshdr->sense_key = (sense_buffer[2] & 0xf);
2094		if (sb_len > 7) {
2095			sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2096					 sb_len : (sense_buffer[7] + 8);
2097			if (sb_len > 12)
2098				sshdr->asc = sense_buffer[12];
2099			if (sb_len > 13)
2100				sshdr->ascq = sense_buffer[13];
2101		}
2102	}
2103
2104	return 1;
2105}
2106EXPORT_SYMBOL(scsi_normalize_sense);
2107
2108int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2109				 struct scsi_sense_hdr *sshdr)
2110{
2111	return scsi_normalize_sense(cmd->sense_buffer,
2112			SCSI_SENSE_BUFFERSIZE, sshdr);
2113}
2114EXPORT_SYMBOL(scsi_command_normalize_sense);
2115
2116/**
2117 * scsi_sense_desc_find - search for a given descriptor type in	descriptor sense data format.
2118 * @sense_buffer:	byte array of descriptor format sense data
2119 * @sb_len:		number of valid bytes in sense_buffer
2120 * @desc_type:		value of descriptor type to find
2121 *			(e.g. 0 -> information)
2122 *
2123 * Notes:
2124 *	only valid when sense data is in descriptor format
2125 *
2126 * Return value:
2127 *	pointer to start of (first) descriptor if found else NULL
2128 */
2129const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2130				int desc_type)
2131{
2132	int add_sen_len, add_len, desc_len, k;
2133	const u8 * descp;
2134
2135	if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2136		return NULL;
2137	if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2138		return NULL;
2139	add_sen_len = (add_sen_len < (sb_len - 8)) ?
2140			add_sen_len : (sb_len - 8);
2141	descp = &sense_buffer[8];
2142	for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2143		descp += desc_len;
2144		add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2145		desc_len = add_len + 2;
2146		if (descp[0] == desc_type)
2147			return descp;
2148		if (add_len < 0) // short descriptor ??
2149			break;
2150	}
2151	return NULL;
2152}
2153EXPORT_SYMBOL(scsi_sense_desc_find);
2154
2155/**
2156 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2157 * @sense_buffer:	byte array of sense data
2158 * @sb_len:		number of valid bytes in sense_buffer
2159 * @info_out:		pointer to 64 integer where 8 or 4 byte information
2160 *			field will be placed if found.
2161 *
2162 * Return value:
2163 *	1 if information field found, 0 if not found.
2164 */
2165int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2166			    u64 * info_out)
2167{
2168	int j;
2169	const u8 * ucp;
2170	u64 ull;
2171
2172	if (sb_len < 7)
2173		return 0;
2174	switch (sense_buffer[0] & 0x7f) {
2175	case 0x70:
2176	case 0x71:
2177		if (sense_buffer[0] & 0x80) {
2178			*info_out = (sense_buffer[3] << 24) +
2179				    (sense_buffer[4] << 16) +
2180				    (sense_buffer[5] << 8) + sense_buffer[6];
2181			return 1;
2182		} else
2183			return 0;
2184	case 0x72:
2185	case 0x73:
2186		ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2187					   0 /* info desc */);
2188		if (ucp && (0xa == ucp[1])) {
2189			ull = 0;
2190			for (j = 0; j < 8; ++j) {
2191				if (j > 0)
2192					ull <<= 8;
2193				ull |= ucp[4 + j];
2194			}
2195			*info_out = ull;
2196			return 1;
2197		} else
2198			return 0;
2199	default:
2200		return 0;
2201	}
2202}
2203EXPORT_SYMBOL(scsi_get_sense_info_fld);
2204
2205/**
2206 * scsi_build_sense_buffer - build sense data in a buffer
2207 * @desc:	Sense format (non zero == descriptor format,
2208 * 		0 == fixed format)
2209 * @buf:	Where to build sense data
2210 * @key:	Sense key
2211 * @asc:	Additional sense code
2212 * @ascq:	Additional sense code qualifier
2213 *
2214 **/
2215void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2216{
2217	if (desc) {
2218		buf[0] = 0x72;	/* descriptor, current */
2219		buf[1] = key;
2220		buf[2] = asc;
2221		buf[3] = ascq;
2222		buf[7] = 0;
2223	} else {
2224		buf[0] = 0x70;	/* fixed, current */
2225		buf[2] = key;
2226		buf[7] = 0xa;
2227		buf[12] = asc;
2228		buf[13] = ascq;
2229	}
2230}
2231EXPORT_SYMBOL(scsi_build_sense_buffer);