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