xref: /linux/drivers/scsi/scsi_error.c (revision 2e3fcbcc3b0eb9b96d2912cdac920f0ae8d1c8f2)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  scsi_error.c Copyright (C) 1997 Eric Youngdale
4  *
5  *  SCSI error/timeout handling
6  *      Initial versions: Eric Youngdale.  Based upon conversations with
7  *                        Leonard Zubkoff and David Miller at Linux Expo,
8  *                        ideas originating from all over the place.
9  *
10  *	Restructured scsi_unjam_host and associated functions.
11  *	September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12  *
13  *	Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
14  *	minor cleanups.
15  *	September 30, 2002 Mike Anderson (andmike@us.ibm.com)
16  */
17 
18 #include <linux/module.h>
19 #include <linux/sched.h>
20 #include <linux/gfp.h>
21 #include <linux/timer.h>
22 #include <linux/string.h>
23 #include <linux/kernel.h>
24 #include <linux/freezer.h>
25 #include <linux/kthread.h>
26 #include <linux/interrupt.h>
27 #include <linux/blkdev.h>
28 #include <linux/delay.h>
29 #include <linux/jiffies.h>
30 
31 #include <scsi/scsi.h>
32 #include <scsi/scsi_cmnd.h>
33 #include <scsi/scsi_dbg.h>
34 #include <scsi/scsi_device.h>
35 #include <scsi/scsi_driver.h>
36 #include <scsi/scsi_eh.h>
37 #include <scsi/scsi_common.h>
38 #include <scsi/scsi_transport.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_ioctl.h>
41 #include <scsi/scsi_dh.h>
42 #include <scsi/scsi_devinfo.h>
43 #include <scsi/sg.h>
44 
45 #include "scsi_priv.h"
46 #include "scsi_logging.h"
47 #include "scsi_transport_api.h"
48 
49 #include <trace/events/scsi.h>
50 
51 #include <linux/unaligned.h>
52 
53 /*
54  * These should *probably* be handled by the host itself.
55  * Since it is allowed to sleep, it probably should.
56  */
57 #define BUS_RESET_SETTLE_TIME   (10)
58 #define HOST_RESET_SETTLE_TIME  (10)
59 
60 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
61 static enum scsi_disposition scsi_try_to_abort_cmd(const struct scsi_host_template *,
62 						   struct scsi_cmnd *);
63 
scsi_eh_wakeup(struct Scsi_Host * shost,unsigned int busy)64 void scsi_eh_wakeup(struct Scsi_Host *shost, unsigned int busy)
65 {
66 	lockdep_assert_held(shost->host_lock);
67 
68 	if (busy == shost->host_failed) {
69 		trace_scsi_eh_wakeup(shost);
70 		wake_up_process(shost->ehandler);
71 		SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
72 			"Waking error handler thread\n"));
73 	}
74 }
75 
76 /**
77  * scsi_schedule_eh - schedule EH for SCSI host
78  * @shost:	SCSI host to invoke error handling on.
79  *
80  * Schedule SCSI EH without scmd.
81  */
scsi_schedule_eh(struct Scsi_Host * shost)82 void scsi_schedule_eh(struct Scsi_Host *shost)
83 {
84 	unsigned long flags;
85 
86 	spin_lock_irqsave(shost->host_lock, flags);
87 
88 	if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
89 	    scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
90 		shost->host_eh_scheduled++;
91 		scsi_eh_wakeup(shost, scsi_host_busy(shost));
92 	}
93 
94 	spin_unlock_irqrestore(shost->host_lock, flags);
95 }
96 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
97 
scsi_host_eh_past_deadline(struct Scsi_Host * shost)98 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
99 {
100 	if (!shost->last_reset || shost->eh_deadline == -1)
101 		return 0;
102 
103 	/*
104 	 * 32bit accesses are guaranteed to be atomic
105 	 * (on all supported architectures), so instead
106 	 * of using a spinlock we can as well double check
107 	 * if eh_deadline has been set to 'off' during the
108 	 * time_before call.
109 	 */
110 	if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
111 	    shost->eh_deadline > -1)
112 		return 0;
113 
114 	return 1;
115 }
116 
scsi_cmd_retry_allowed(struct scsi_cmnd * cmd)117 static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd)
118 {
119 	if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
120 		return true;
121 
122 	return ++cmd->retries <= cmd->allowed;
123 }
124 
scsi_eh_should_retry_cmd(struct scsi_cmnd * cmd)125 static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd)
126 {
127 	struct scsi_device *sdev = cmd->device;
128 	struct Scsi_Host *host = sdev->host;
129 
130 	if (host->hostt->eh_should_retry_cmd)
131 		return  host->hostt->eh_should_retry_cmd(cmd);
132 
133 	return true;
134 }
135 
136 /**
137  * scmd_eh_abort_handler - Handle command aborts
138  * @work:	command to be aborted.
139  *
140  * Note: this function must be called only for a command that has timed out.
141  * Because the block layer marks a request as complete before it calls
142  * scsi_timeout(), a .scsi_done() call from the LLD for a command that has
143  * timed out do not have any effect. Hence it is safe to call
144  * scsi_finish_command() from this function.
145  */
146 void
scmd_eh_abort_handler(struct work_struct * work)147 scmd_eh_abort_handler(struct work_struct *work)
148 {
149 	struct scsi_cmnd *scmd =
150 		container_of(work, struct scsi_cmnd, abort_work.work);
151 	struct scsi_device *sdev = scmd->device;
152 	struct Scsi_Host *shost = sdev->host;
153 	enum scsi_disposition rtn;
154 	unsigned long flags;
155 
156 	if (scsi_host_eh_past_deadline(shost)) {
157 		SCSI_LOG_ERROR_RECOVERY(3,
158 			scmd_printk(KERN_INFO, scmd,
159 				    "eh timeout, not aborting\n"));
160 		goto out;
161 	}
162 
163 	SCSI_LOG_ERROR_RECOVERY(3,
164 			scmd_printk(KERN_INFO, scmd,
165 				    "aborting command\n"));
166 	rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
167 	if (rtn != SUCCESS) {
168 		SCSI_LOG_ERROR_RECOVERY(3,
169 			scmd_printk(KERN_INFO, scmd,
170 				    "cmd abort %s\n",
171 				    (rtn == FAST_IO_FAIL) ?
172 				    "not send" : "failed"));
173 		goto out;
174 	}
175 	set_host_byte(scmd, DID_TIME_OUT);
176 	if (scsi_host_eh_past_deadline(shost)) {
177 		SCSI_LOG_ERROR_RECOVERY(3,
178 			scmd_printk(KERN_INFO, scmd,
179 				    "eh timeout, not retrying "
180 				    "aborted command\n"));
181 		goto out;
182 	}
183 
184 	spin_lock_irqsave(shost->host_lock, flags);
185 	list_del_init(&scmd->eh_entry);
186 
187 	/*
188 	 * If the abort succeeds, and there is no further
189 	 * EH action, clear the ->last_reset time.
190 	 */
191 	if (list_empty(&shost->eh_abort_list) &&
192 	    list_empty(&shost->eh_cmd_q))
193 		if (shost->eh_deadline != -1)
194 			shost->last_reset = 0;
195 
196 	spin_unlock_irqrestore(shost->host_lock, flags);
197 
198 	if (!scsi_noretry_cmd(scmd) &&
199 	    scsi_cmd_retry_allowed(scmd) &&
200 	    scsi_eh_should_retry_cmd(scmd)) {
201 		SCSI_LOG_ERROR_RECOVERY(3,
202 			scmd_printk(KERN_WARNING, scmd,
203 				    "retry aborted command\n"));
204 		scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
205 	} else {
206 		SCSI_LOG_ERROR_RECOVERY(3,
207 			scmd_printk(KERN_WARNING, scmd,
208 				    "finish aborted command\n"));
209 		scsi_finish_command(scmd);
210 	}
211 	return;
212 
213 out:
214 	spin_lock_irqsave(shost->host_lock, flags);
215 	list_del_init(&scmd->eh_entry);
216 	spin_unlock_irqrestore(shost->host_lock, flags);
217 
218 	scsi_eh_scmd_add(scmd);
219 }
220 
221 /**
222  * scsi_abort_command - schedule a command abort
223  * @scmd:	scmd to abort.
224  *
225  * We only need to abort commands after a command timeout
226  */
227 static int
scsi_abort_command(struct scsi_cmnd * scmd)228 scsi_abort_command(struct scsi_cmnd *scmd)
229 {
230 	struct scsi_device *sdev = scmd->device;
231 	struct Scsi_Host *shost = sdev->host;
232 	unsigned long flags;
233 
234 	if (!shost->hostt->eh_abort_handler) {
235 		/* No abort handler, fail command directly */
236 		return FAILED;
237 	}
238 
239 	if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
240 		/*
241 		 * Retry after abort failed, escalate to next level.
242 		 */
243 		SCSI_LOG_ERROR_RECOVERY(3,
244 			scmd_printk(KERN_INFO, scmd,
245 				    "previous abort failed\n"));
246 		BUG_ON(delayed_work_pending(&scmd->abort_work));
247 		return FAILED;
248 	}
249 
250 	spin_lock_irqsave(shost->host_lock, flags);
251 	if (shost->eh_deadline != -1 && !shost->last_reset)
252 		shost->last_reset = jiffies;
253 	BUG_ON(!list_empty(&scmd->eh_entry));
254 	list_add_tail(&scmd->eh_entry, &shost->eh_abort_list);
255 	spin_unlock_irqrestore(shost->host_lock, flags);
256 
257 	scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
258 	SCSI_LOG_ERROR_RECOVERY(3,
259 		scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
260 	queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
261 	return SUCCESS;
262 }
263 
264 /**
265  * scsi_eh_reset - call into ->eh_action to reset internal counters
266  * @scmd:	scmd to run eh on.
267  *
268  * The scsi driver might be carrying internal state about the
269  * devices, so we need to call into the driver to reset the
270  * internal state once the error handler is started.
271  */
scsi_eh_reset(struct scsi_cmnd * scmd)272 static void scsi_eh_reset(struct scsi_cmnd *scmd)
273 {
274 	if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
275 		struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
276 		if (sdrv->eh_reset)
277 			sdrv->eh_reset(scmd);
278 	}
279 }
280 
scsi_eh_inc_host_failed(struct rcu_head * head)281 static void scsi_eh_inc_host_failed(struct rcu_head *head)
282 {
283 	struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
284 	struct Scsi_Host *shost = scmd->device->host;
285 	unsigned int busy = scsi_host_busy(shost);
286 	unsigned long flags;
287 
288 	spin_lock_irqsave(shost->host_lock, flags);
289 	shost->host_failed++;
290 	scsi_eh_wakeup(shost, busy);
291 	spin_unlock_irqrestore(shost->host_lock, flags);
292 }
293 
294 /**
295  * scsi_eh_scmd_add - add scsi cmd to error handling.
296  * @scmd:	scmd to run eh on.
297  */
scsi_eh_scmd_add(struct scsi_cmnd * scmd)298 void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
299 {
300 	struct Scsi_Host *shost = scmd->device->host;
301 	unsigned long flags;
302 	int ret;
303 
304 	WARN_ON_ONCE(!shost->ehandler);
305 	WARN_ON_ONCE(!test_bit(SCMD_STATE_INFLIGHT, &scmd->state));
306 
307 	spin_lock_irqsave(shost->host_lock, flags);
308 	if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
309 		ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
310 		WARN_ON_ONCE(ret);
311 	}
312 	if (shost->eh_deadline != -1 && !shost->last_reset)
313 		shost->last_reset = jiffies;
314 
315 	scsi_eh_reset(scmd);
316 	list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
317 	spin_unlock_irqrestore(shost->host_lock, flags);
318 	/*
319 	 * Ensure that all tasks observe the host state change before the
320 	 * host_failed change.
321 	 */
322 	call_rcu_hurry(&scmd->rcu, scsi_eh_inc_host_failed);
323 }
324 
325 /**
326  * scsi_timeout - Timeout function for normal scsi commands.
327  * @req:	request that is timing out.
328  *
329  * Notes:
330  *     We do not need to lock this.  There is the potential for a race
331  *     only in that the normal completion handling might run, but if the
332  *     normal completion function determines that the timer has already
333  *     fired, then it mustn't do anything.
334  */
scsi_timeout(struct request * req)335 enum blk_eh_timer_return scsi_timeout(struct request *req)
336 {
337 	struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
338 	struct Scsi_Host *host = scmd->device->host;
339 
340 	trace_scsi_dispatch_cmd_timeout(scmd);
341 	scsi_log_completion(scmd, TIMEOUT_ERROR);
342 
343 	atomic_inc(&scmd->device->iotmo_cnt);
344 	if (host->eh_deadline != -1 && !host->last_reset)
345 		host->last_reset = jiffies;
346 
347 	if (host->hostt->eh_timed_out) {
348 		switch (host->hostt->eh_timed_out(scmd)) {
349 		case SCSI_EH_DONE:
350 			return BLK_EH_DONE;
351 		case SCSI_EH_RESET_TIMER:
352 			return BLK_EH_RESET_TIMER;
353 		case SCSI_EH_NOT_HANDLED:
354 			break;
355 		}
356 	}
357 
358 	/*
359 	 * If scsi_done() has already set SCMD_STATE_COMPLETE, do not modify
360 	 * *scmd.
361 	 */
362 	if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
363 		return BLK_EH_DONE;
364 	atomic_inc(&scmd->device->iodone_cnt);
365 	if (scsi_abort_command(scmd) != SUCCESS) {
366 		set_host_byte(scmd, DID_TIME_OUT);
367 		scsi_eh_scmd_add(scmd);
368 	}
369 
370 	return BLK_EH_DONE;
371 }
372 
373 /**
374  * scsi_block_when_processing_errors - Prevent cmds from being queued.
375  * @sdev:	Device on which we are performing recovery.
376  *
377  * Description:
378  *     We block until the host is out of error recovery, and then check to
379  *     see whether the host or the device is offline.
380  *
381  * Return value:
382  *     0 when dev was taken offline by error recovery. 1 OK to proceed.
383  */
scsi_block_when_processing_errors(struct scsi_device * sdev)384 int scsi_block_when_processing_errors(struct scsi_device *sdev)
385 {
386 	int online;
387 
388 	wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
389 
390 	online = scsi_device_online(sdev);
391 
392 	return online;
393 }
394 EXPORT_SYMBOL(scsi_block_when_processing_errors);
395 
396 #ifdef CONFIG_SCSI_LOGGING
397 /**
398  * scsi_eh_prt_fail_stats - Log info on failures.
399  * @shost:	scsi host being recovered.
400  * @work_q:	Queue of scsi cmds to process.
401  */
scsi_eh_prt_fail_stats(struct Scsi_Host * shost,struct list_head * work_q)402 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
403 					  struct list_head *work_q)
404 {
405 	struct scsi_cmnd *scmd;
406 	struct scsi_device *sdev;
407 	int total_failures = 0;
408 	int cmd_failed = 0;
409 	int cmd_cancel = 0;
410 	int devices_failed = 0;
411 
412 	shost_for_each_device(sdev, shost) {
413 		list_for_each_entry(scmd, work_q, eh_entry) {
414 			if (scmd->device == sdev) {
415 				++total_failures;
416 				if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
417 					++cmd_cancel;
418 				else
419 					++cmd_failed;
420 			}
421 		}
422 
423 		if (cmd_cancel || cmd_failed) {
424 			SCSI_LOG_ERROR_RECOVERY(3,
425 				shost_printk(KERN_INFO, shost,
426 					    "%s: cmds failed: %d, cancel: %d\n",
427 					    __func__, cmd_failed,
428 					    cmd_cancel));
429 			cmd_cancel = 0;
430 			cmd_failed = 0;
431 			++devices_failed;
432 		}
433 	}
434 
435 	SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
436 				   "Total of %d commands on %d"
437 				   " devices require eh work\n",
438 				   total_failures, devices_failed));
439 }
440 #endif
441 
442  /**
443  * scsi_report_lun_change - Set flag on all *other* devices on the same target
444  *                          to indicate that a UNIT ATTENTION is expected.
445  * @sdev:	Device reporting the UNIT ATTENTION
446  */
scsi_report_lun_change(struct scsi_device * sdev)447 static void scsi_report_lun_change(struct scsi_device *sdev)
448 {
449 	sdev->sdev_target->expecting_lun_change = 1;
450 }
451 
452 /**
453  * scsi_report_sense - Examine scsi sense information and log messages for
454  *		       certain conditions, also issue uevents for some of them.
455  * @sdev:	Device reporting the sense code
456  * @sshdr:	sshdr to be examined
457  */
scsi_report_sense(struct scsi_device * sdev,struct scsi_sense_hdr * sshdr)458 static void scsi_report_sense(struct scsi_device *sdev,
459 			      struct scsi_sense_hdr *sshdr)
460 {
461 	enum scsi_device_event evt_type = SDEV_EVT_MAXBITS;	/* i.e. none */
462 
463 	if (sshdr->sense_key == UNIT_ATTENTION) {
464 		if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
465 			evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
466 			sdev_printk(KERN_WARNING, sdev,
467 				    "Inquiry data has changed");
468 		} else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
469 			evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
470 			scsi_report_lun_change(sdev);
471 			sdev_printk(KERN_WARNING, sdev,
472 				    "LUN assignments on this target have "
473 				    "changed. The Linux SCSI layer does not "
474 				    "automatically remap LUN assignments.\n");
475 		} else if (sshdr->asc == 0x3f)
476 			sdev_printk(KERN_WARNING, sdev,
477 				    "Operating parameters on this target have "
478 				    "changed. The Linux SCSI layer does not "
479 				    "automatically adjust these parameters.\n");
480 
481 		if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
482 			evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
483 			sdev_printk(KERN_WARNING, sdev,
484 				    "Warning! Received an indication that the "
485 				    "LUN reached a thin provisioning soft "
486 				    "threshold.\n");
487 		}
488 
489 		if (sshdr->asc == 0x29) {
490 			evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
491 			/*
492 			 * Do not print message if it is an expected side-effect
493 			 * of runtime PM.
494 			 */
495 			if (!sdev->silence_suspend)
496 				sdev_printk(KERN_WARNING, sdev,
497 					    "Power-on or device reset occurred\n");
498 		}
499 
500 		if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
501 			evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
502 			sdev_printk(KERN_WARNING, sdev,
503 				    "Mode parameters changed");
504 		} else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
505 			evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
506 			sdev_printk(KERN_WARNING, sdev,
507 				    "Asymmetric access state changed");
508 		} else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
509 			evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
510 			sdev_printk(KERN_WARNING, sdev,
511 				    "Capacity data has changed");
512 		} else if (sshdr->asc == 0x2a)
513 			sdev_printk(KERN_WARNING, sdev,
514 				    "Parameters changed");
515 	}
516 
517 	if (evt_type != SDEV_EVT_MAXBITS) {
518 		set_bit(evt_type, sdev->pending_events);
519 		schedule_work(&sdev->event_work);
520 	}
521 }
522 
set_scsi_ml_byte(struct scsi_cmnd * cmd,u8 status)523 static inline void set_scsi_ml_byte(struct scsi_cmnd *cmd, u8 status)
524 {
525 	cmd->result = (cmd->result & 0xffff00ff) | (status << 8);
526 }
527 
528 /**
529  * scsi_check_sense - Examine scsi cmd sense
530  * @scmd:	Cmd to have sense checked.
531  *
532  * Return value:
533  *	SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
534  *
535  * Notes:
536  *	When a deferred error is detected the current command has
537  *	not been executed and needs retrying.
538  */
scsi_check_sense(struct scsi_cmnd * scmd)539 enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd)
540 {
541 	struct request *req = scsi_cmd_to_rq(scmd);
542 	struct scsi_device *sdev = scmd->device;
543 	struct scsi_sense_hdr sshdr;
544 
545 	if (! scsi_command_normalize_sense(scmd, &sshdr))
546 		return FAILED;	/* no valid sense data */
547 
548 	scsi_report_sense(sdev, &sshdr);
549 
550 	if (sshdr.sense_key == UNIT_ATTENTION) {
551 		/*
552 		 * Increment the counters for Power on/Reset or New Media so
553 		 * that all ULDs interested in these can see that those have
554 		 * happened, even if someone else gets the sense data.
555 		 */
556 		if (sshdr.asc == 0x28)
557 			scmd->device->ua_new_media_ctr++;
558 		else if (sshdr.asc == 0x29)
559 			scmd->device->ua_por_ctr++;
560 	}
561 
562 	if (scsi_sense_is_deferred(&sshdr))
563 		return NEEDS_RETRY;
564 
565 	if (sdev->handler && sdev->handler->check_sense) {
566 		enum scsi_disposition rc;
567 
568 		rc = sdev->handler->check_sense(sdev, &sshdr);
569 		if (rc != SCSI_RETURN_NOT_HANDLED)
570 			return rc;
571 		/* handler does not care. Drop down to default handling */
572 	}
573 
574 	if (scmd->cmnd[0] == TEST_UNIT_READY &&
575 	    scmd->submitter != SUBMITTED_BY_SCSI_ERROR_HANDLER)
576 		/*
577 		 * nasty: for mid-layer issued TURs, we need to return the
578 		 * actual sense data without any recovery attempt.  For eh
579 		 * issued ones, we need to try to recover and interpret
580 		 */
581 		return SUCCESS;
582 
583 	/*
584 	 * Previous logic looked for FILEMARK, EOM or ILI which are
585 	 * mainly associated with tapes and returned SUCCESS.
586 	 */
587 	if (sshdr.response_code == 0x70) {
588 		/* fixed format */
589 		if (scmd->sense_buffer[2] & 0xe0)
590 			return SUCCESS;
591 	} else {
592 		/*
593 		 * descriptor format: look for "stream commands sense data
594 		 * descriptor" (see SSC-3). Assume single sense data
595 		 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
596 		 */
597 		if ((sshdr.additional_length > 3) &&
598 		    (scmd->sense_buffer[8] == 0x4) &&
599 		    (scmd->sense_buffer[11] & 0xe0))
600 			return SUCCESS;
601 	}
602 
603 	switch (sshdr.sense_key) {
604 	case NO_SENSE:
605 		return SUCCESS;
606 	case RECOVERED_ERROR:
607 		return /* soft_error */ SUCCESS;
608 
609 	case ABORTED_COMMAND:
610 		if (sshdr.asc == 0x10) /* DIF */
611 			return SUCCESS;
612 
613 		/*
614 		 * Check aborts due to command duration limit policy:
615 		 * ABORTED COMMAND additional sense code with the
616 		 * COMMAND TIMEOUT BEFORE PROCESSING or
617 		 * COMMAND TIMEOUT DURING PROCESSING or
618 		 * COMMAND TIMEOUT DURING PROCESSING DUE TO ERROR RECOVERY
619 		 * additional sense code qualifiers.
620 		 */
621 		if (sshdr.asc == 0x2e &&
622 		    sshdr.ascq >= 0x01 && sshdr.ascq <= 0x03) {
623 			set_scsi_ml_byte(scmd, SCSIML_STAT_DL_TIMEOUT);
624 			req->cmd_flags |= REQ_FAILFAST_DEV;
625 			req->rq_flags |= RQF_QUIET;
626 			return SUCCESS;
627 		}
628 
629 		if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
630 			return ADD_TO_MLQUEUE;
631 		if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
632 		    sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
633 			return ADD_TO_MLQUEUE;
634 
635 		return NEEDS_RETRY;
636 	case NOT_READY:
637 	case UNIT_ATTENTION:
638 		/*
639 		 * if we are expecting a cc/ua because of a bus reset that we
640 		 * performed, treat this just as a retry.  otherwise this is
641 		 * information that we should pass up to the upper-level driver
642 		 * so that we can deal with it there.
643 		 */
644 		if (scmd->device->expecting_cc_ua) {
645 			/*
646 			 * Because some device does not queue unit
647 			 * attentions correctly, we carefully check
648 			 * additional sense code and qualifier so as
649 			 * not to squash media change unit attention.
650 			 */
651 			if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
652 				scmd->device->expecting_cc_ua = 0;
653 				return NEEDS_RETRY;
654 			}
655 		}
656 		/*
657 		 * we might also expect a cc/ua if another LUN on the target
658 		 * reported a UA with an ASC/ASCQ of 3F 0E -
659 		 * REPORTED LUNS DATA HAS CHANGED.
660 		 */
661 		if (scmd->device->sdev_target->expecting_lun_change &&
662 		    sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
663 			return NEEDS_RETRY;
664 		/*
665 		 * if the device is in the process of becoming ready, we
666 		 * should retry.
667 		 */
668 		if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
669 			return NEEDS_RETRY;
670 		/*
671 		 * if the device is not started, we need to wake
672 		 * the error handler to start the motor
673 		 */
674 		if (scmd->device->allow_restart &&
675 		    (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
676 			return FAILED;
677 		/*
678 		 * Pass the UA upwards for a determination in the completion
679 		 * functions.
680 		 */
681 		return SUCCESS;
682 
683 		/* these are not supported */
684 	case DATA_PROTECT:
685 		if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
686 			/* Thin provisioning hard threshold reached */
687 			set_scsi_ml_byte(scmd, SCSIML_STAT_NOSPC);
688 			return SUCCESS;
689 		}
690 		fallthrough;
691 	case COPY_ABORTED:
692 	case VOLUME_OVERFLOW:
693 	case MISCOMPARE:
694 	case BLANK_CHECK:
695 		set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE);
696 		return SUCCESS;
697 
698 	case MEDIUM_ERROR:
699 		if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
700 		    sshdr.asc == 0x13 || /* AMNF DATA FIELD */
701 		    sshdr.asc == 0x14) { /* RECORD NOT FOUND */
702 			set_scsi_ml_byte(scmd, SCSIML_STAT_MED_ERROR);
703 			return SUCCESS;
704 		}
705 		return NEEDS_RETRY;
706 
707 	case HARDWARE_ERROR:
708 		if (scmd->device->retry_hwerror)
709 			return ADD_TO_MLQUEUE;
710 		else
711 			set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE);
712 		fallthrough;
713 
714 	case ILLEGAL_REQUEST:
715 		if (sshdr.asc == 0x20 || /* Invalid command operation code */
716 		    sshdr.asc == 0x21 || /* Logical block address out of range */
717 		    sshdr.asc == 0x22 || /* Invalid function */
718 		    sshdr.asc == 0x24 || /* Invalid field in cdb */
719 		    sshdr.asc == 0x26 || /* Parameter value invalid */
720 		    sshdr.asc == 0x27) { /* Write protected */
721 			set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE);
722 		}
723 		return SUCCESS;
724 
725 	case COMPLETED:
726 		/*
727 		 * A command using command duration limits (CDL) with a
728 		 * descriptor set with policy 0xD may be completed with success
729 		 * and the sense data DATA CURRENTLY UNAVAILABLE, indicating
730 		 * that the command was in fact aborted because it exceeded its
731 		 * duration limit. Never retry these commands.
732 		 */
733 		if (sshdr.asc == 0x55 && sshdr.ascq == 0x0a) {
734 			set_scsi_ml_byte(scmd, SCSIML_STAT_DL_TIMEOUT);
735 			req->cmd_flags |= REQ_FAILFAST_DEV;
736 			req->rq_flags |= RQF_QUIET;
737 		}
738 		return SUCCESS;
739 
740 	default:
741 		return SUCCESS;
742 	}
743 }
744 EXPORT_SYMBOL_GPL(scsi_check_sense);
745 
scsi_handle_queue_ramp_up(struct scsi_device * sdev)746 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
747 {
748 	const struct scsi_host_template *sht = sdev->host->hostt;
749 	struct scsi_device *tmp_sdev;
750 
751 	if (!sht->track_queue_depth ||
752 	    sdev->queue_depth >= sdev->max_queue_depth)
753 		return;
754 
755 	if (time_before(jiffies,
756 	    sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
757 		return;
758 
759 	if (time_before(jiffies,
760 	    sdev->last_queue_full_time + sdev->queue_ramp_up_period))
761 		return;
762 
763 	/*
764 	 * Walk all devices of a target and do
765 	 * ramp up on them.
766 	 */
767 	shost_for_each_device(tmp_sdev, sdev->host) {
768 		if (tmp_sdev->channel != sdev->channel ||
769 		    tmp_sdev->id != sdev->id ||
770 		    tmp_sdev->queue_depth == sdev->max_queue_depth)
771 			continue;
772 
773 		scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
774 		sdev->last_queue_ramp_up = jiffies;
775 	}
776 }
777 
scsi_handle_queue_full(struct scsi_device * sdev)778 static void scsi_handle_queue_full(struct scsi_device *sdev)
779 {
780 	const struct scsi_host_template *sht = sdev->host->hostt;
781 	struct scsi_device *tmp_sdev;
782 
783 	if (!sht->track_queue_depth)
784 		return;
785 
786 	shost_for_each_device(tmp_sdev, sdev->host) {
787 		if (tmp_sdev->channel != sdev->channel ||
788 		    tmp_sdev->id != sdev->id)
789 			continue;
790 		/*
791 		 * We do not know the number of commands that were at
792 		 * the device when we got the queue full so we start
793 		 * from the highest possible value and work our way down.
794 		 */
795 		scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
796 	}
797 }
798 
799 /**
800  * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
801  * @scmd:	SCSI cmd to examine.
802  *
803  * Notes:
804  *    This is *only* called when we are examining the status of commands
805  *    queued during error recovery.  the main difference here is that we
806  *    don't allow for the possibility of retries here, and we are a lot
807  *    more restrictive about what we consider acceptable.
808  */
scsi_eh_completed_normally(struct scsi_cmnd * scmd)809 static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd)
810 {
811 	/*
812 	 * first check the host byte, to see if there is anything in there
813 	 * that would indicate what we need to do.
814 	 */
815 	if (host_byte(scmd->result) == DID_RESET) {
816 		/*
817 		 * rats.  we are already in the error handler, so we now
818 		 * get to try and figure out what to do next.  if the sense
819 		 * is valid, we have a pretty good idea of what to do.
820 		 * if not, we mark it as FAILED.
821 		 */
822 		return scsi_check_sense(scmd);
823 	}
824 	if (host_byte(scmd->result) != DID_OK)
825 		return FAILED;
826 
827 	/*
828 	 * now, check the status byte to see if this indicates
829 	 * anything special.
830 	 */
831 	switch (get_status_byte(scmd)) {
832 	case SAM_STAT_GOOD:
833 		scsi_handle_queue_ramp_up(scmd->device);
834 		if (scmd->sense_buffer && SCSI_SENSE_VALID(scmd))
835 			/*
836 			 * If we have sense data, call scsi_check_sense() in
837 			 * order to set the correct SCSI ML byte (if any).
838 			 * No point in checking the return value, since the
839 			 * command has already completed successfully.
840 			 */
841 			scsi_check_sense(scmd);
842 		fallthrough;
843 	case SAM_STAT_COMMAND_TERMINATED:
844 		return SUCCESS;
845 	case SAM_STAT_CHECK_CONDITION:
846 		return scsi_check_sense(scmd);
847 	case SAM_STAT_CONDITION_MET:
848 	case SAM_STAT_INTERMEDIATE:
849 	case SAM_STAT_INTERMEDIATE_CONDITION_MET:
850 		/*
851 		 * who knows?  FIXME(eric)
852 		 */
853 		return SUCCESS;
854 	case SAM_STAT_RESERVATION_CONFLICT:
855 		if (scmd->cmnd[0] == TEST_UNIT_READY)
856 			/* it is a success, we probed the device and
857 			 * found it */
858 			return SUCCESS;
859 		/* otherwise, we failed to send the command */
860 		return FAILED;
861 	case SAM_STAT_TASK_SET_FULL:
862 		scsi_handle_queue_full(scmd->device);
863 		fallthrough;
864 	case SAM_STAT_BUSY:
865 		return NEEDS_RETRY;
866 	default:
867 		return FAILED;
868 	}
869 	return FAILED;
870 }
871 
872 /**
873  * scsi_eh_done - Completion function for error handling.
874  * @scmd:	Cmd that is done.
875  */
scsi_eh_done(struct scsi_cmnd * scmd)876 void scsi_eh_done(struct scsi_cmnd *scmd)
877 {
878 	struct completion *eh_action;
879 
880 	SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
881 			"%s result: %x\n", __func__, scmd->result));
882 
883 	eh_action = scmd->device->host->eh_action;
884 	if (eh_action)
885 		complete(eh_action);
886 }
887 
888 /**
889  * scsi_try_host_reset - ask host adapter to reset itself
890  * @scmd:	SCSI cmd to send host reset.
891  */
scsi_try_host_reset(struct scsi_cmnd * scmd)892 static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
893 {
894 	unsigned long flags;
895 	enum scsi_disposition rtn;
896 	struct Scsi_Host *host = scmd->device->host;
897 	const struct scsi_host_template *hostt = host->hostt;
898 
899 	SCSI_LOG_ERROR_RECOVERY(3,
900 		shost_printk(KERN_INFO, host, "Snd Host RST\n"));
901 
902 	if (!hostt->eh_host_reset_handler)
903 		return FAILED;
904 
905 	rtn = hostt->eh_host_reset_handler(scmd);
906 
907 	if (rtn == SUCCESS) {
908 		if (!hostt->skip_settle_delay)
909 			ssleep(HOST_RESET_SETTLE_TIME);
910 		spin_lock_irqsave(host->host_lock, flags);
911 		scsi_report_bus_reset(host, scmd_channel(scmd));
912 		spin_unlock_irqrestore(host->host_lock, flags);
913 	}
914 
915 	return rtn;
916 }
917 
918 /**
919  * scsi_try_bus_reset - ask host to perform a bus reset
920  * @scmd:	SCSI cmd to send bus reset.
921  */
scsi_try_bus_reset(struct scsi_cmnd * scmd)922 static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
923 {
924 	unsigned long flags;
925 	enum scsi_disposition rtn;
926 	struct Scsi_Host *host = scmd->device->host;
927 	const struct scsi_host_template *hostt = host->hostt;
928 
929 	SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
930 		"%s: Snd Bus RST\n", __func__));
931 
932 	if (!hostt->eh_bus_reset_handler)
933 		return FAILED;
934 
935 	rtn = hostt->eh_bus_reset_handler(scmd);
936 
937 	if (rtn == SUCCESS) {
938 		if (!hostt->skip_settle_delay)
939 			ssleep(BUS_RESET_SETTLE_TIME);
940 		spin_lock_irqsave(host->host_lock, flags);
941 		scsi_report_bus_reset(host, scmd_channel(scmd));
942 		spin_unlock_irqrestore(host->host_lock, flags);
943 	}
944 
945 	return rtn;
946 }
947 
__scsi_report_device_reset(struct scsi_device * sdev,void * data)948 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
949 {
950 	sdev->was_reset = 1;
951 	sdev->expecting_cc_ua = 1;
952 }
953 
954 /**
955  * scsi_try_target_reset - Ask host to perform a target reset
956  * @scmd:	SCSI cmd used to send a target reset
957  *
958  * Notes:
959  *    There is no timeout for this operation.  if this operation is
960  *    unreliable for a given host, then the host itself needs to put a
961  *    timer on it, and set the host back to a consistent state prior to
962  *    returning.
963  */
scsi_try_target_reset(struct scsi_cmnd * scmd)964 static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd)
965 {
966 	unsigned long flags;
967 	enum scsi_disposition rtn;
968 	struct Scsi_Host *host = scmd->device->host;
969 	const struct scsi_host_template *hostt = host->hostt;
970 
971 	if (!hostt->eh_target_reset_handler)
972 		return FAILED;
973 
974 	rtn = hostt->eh_target_reset_handler(scmd);
975 	if (rtn == SUCCESS) {
976 		spin_lock_irqsave(host->host_lock, flags);
977 		__starget_for_each_device(scsi_target(scmd->device), NULL,
978 					  __scsi_report_device_reset);
979 		spin_unlock_irqrestore(host->host_lock, flags);
980 	}
981 
982 	return rtn;
983 }
984 
985 /**
986  * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
987  * @scmd:	SCSI cmd used to send BDR
988  *
989  * Notes:
990  *    There is no timeout for this operation.  if this operation is
991  *    unreliable for a given host, then the host itself needs to put a
992  *    timer on it, and set the host back to a consistent state prior to
993  *    returning.
994  */
scsi_try_bus_device_reset(struct scsi_cmnd * scmd)995 static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
996 {
997 	enum scsi_disposition rtn;
998 	const struct scsi_host_template *hostt = scmd->device->host->hostt;
999 
1000 	if (!hostt->eh_device_reset_handler)
1001 		return FAILED;
1002 
1003 	rtn = hostt->eh_device_reset_handler(scmd);
1004 	if (rtn == SUCCESS)
1005 		__scsi_report_device_reset(scmd->device, NULL);
1006 	return rtn;
1007 }
1008 
1009 /**
1010  * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
1011  * @hostt:	SCSI driver host template
1012  * @scmd:	SCSI cmd used to send a target reset
1013  *
1014  * Return value:
1015  *	SUCCESS, FAILED, or FAST_IO_FAIL
1016  *
1017  * Notes:
1018  *    SUCCESS does not necessarily indicate that the command
1019  *    has been aborted; it only indicates that the LLDDs
1020  *    has cleared all references to that command.
1021  *    LLDDs should return FAILED only if an abort was required
1022  *    but could not be executed. LLDDs should return FAST_IO_FAIL
1023  *    if the device is temporarily unavailable (eg due to a
1024  *    link down on FibreChannel)
1025  */
1026 static enum scsi_disposition
scsi_try_to_abort_cmd(const struct scsi_host_template * hostt,struct scsi_cmnd * scmd)1027 scsi_try_to_abort_cmd(const struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
1028 {
1029 	if (!hostt->eh_abort_handler)
1030 		return FAILED;
1031 
1032 	return hostt->eh_abort_handler(scmd);
1033 }
1034 
scsi_abort_eh_cmnd(struct scsi_cmnd * scmd)1035 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
1036 {
1037 	if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
1038 		if (scsi_try_bus_device_reset(scmd) != SUCCESS)
1039 			if (scsi_try_target_reset(scmd) != SUCCESS)
1040 				if (scsi_try_bus_reset(scmd) != SUCCESS)
1041 					scsi_try_host_reset(scmd);
1042 }
1043 
1044 /**
1045  * scsi_eh_prep_cmnd  - Save a scsi command info as part of error recovery
1046  * @scmd:       SCSI command structure to hijack
1047  * @ses:        structure to save restore information
1048  * @cmnd:       CDB to send. Can be NULL if no new cmnd is needed
1049  * @cmnd_size:  size in bytes of @cmnd (must be <= MAX_COMMAND_SIZE)
1050  * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
1051  *
1052  * This function is used to save a scsi command information before re-execution
1053  * as part of the error recovery process.  If @sense_bytes is 0 the command
1054  * sent must be one that does not transfer any data.  If @sense_bytes != 0
1055  * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
1056  * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
1057  */
scsi_eh_prep_cmnd(struct scsi_cmnd * scmd,struct scsi_eh_save * ses,unsigned char * cmnd,int cmnd_size,unsigned sense_bytes)1058 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
1059 			unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
1060 {
1061 	struct scsi_device *sdev = scmd->device;
1062 
1063 	/*
1064 	 * We need saved copies of a number of fields - this is because
1065 	 * error handling may need to overwrite these with different values
1066 	 * to run different commands, and once error handling is complete,
1067 	 * we will need to restore these values prior to running the actual
1068 	 * command.
1069 	 */
1070 	ses->cmd_len = scmd->cmd_len;
1071 	ses->data_direction = scmd->sc_data_direction;
1072 	ses->sdb = scmd->sdb;
1073 	ses->result = scmd->result;
1074 	ses->resid_len = scmd->resid_len;
1075 	ses->underflow = scmd->underflow;
1076 	ses->prot_op = scmd->prot_op;
1077 	ses->eh_eflags = scmd->eh_eflags;
1078 
1079 	scmd->prot_op = SCSI_PROT_NORMAL;
1080 	scmd->eh_eflags = 0;
1081 	memcpy(ses->cmnd, scmd->cmnd, sizeof(ses->cmnd));
1082 	memset(scmd->cmnd, 0, sizeof(scmd->cmnd));
1083 	memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1084 	scmd->result = 0;
1085 	scmd->resid_len = 0;
1086 
1087 	if (sense_bytes) {
1088 		scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
1089 					 sense_bytes);
1090 		sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
1091 			    scmd->sdb.length);
1092 		scmd->sdb.table.sgl = &ses->sense_sgl;
1093 		scmd->sc_data_direction = DMA_FROM_DEVICE;
1094 		scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
1095 		scmd->cmnd[0] = REQUEST_SENSE;
1096 		scmd->cmnd[4] = scmd->sdb.length;
1097 		scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1098 	} else {
1099 		scmd->sc_data_direction = DMA_NONE;
1100 		if (cmnd) {
1101 			BUG_ON(cmnd_size > sizeof(scmd->cmnd));
1102 			memcpy(scmd->cmnd, cmnd, cmnd_size);
1103 			scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1104 		}
1105 	}
1106 
1107 	scmd->underflow = 0;
1108 
1109 	if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1110 		scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
1111 			(sdev->lun << 5 & 0xe0);
1112 
1113 	/*
1114 	 * Zero the sense buffer.  The scsi spec mandates that any
1115 	 * untransferred sense data should be interpreted as being zero.
1116 	 */
1117 	memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1118 }
1119 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1120 
1121 /**
1122  * scsi_eh_restore_cmnd  - Restore a scsi command info as part of error recovery
1123  * @scmd:       SCSI command structure to restore
1124  * @ses:        saved information from a coresponding call to scsi_eh_prep_cmnd
1125  *
1126  * Undo any damage done by above scsi_eh_prep_cmnd().
1127  */
scsi_eh_restore_cmnd(struct scsi_cmnd * scmd,struct scsi_eh_save * ses)1128 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1129 {
1130 	/*
1131 	 * Restore original data
1132 	 */
1133 	scmd->cmd_len = ses->cmd_len;
1134 	memcpy(scmd->cmnd, ses->cmnd, sizeof(ses->cmnd));
1135 	scmd->sc_data_direction = ses->data_direction;
1136 	scmd->sdb = ses->sdb;
1137 	scmd->result = ses->result;
1138 	scmd->resid_len = ses->resid_len;
1139 	scmd->underflow = ses->underflow;
1140 	scmd->prot_op = ses->prot_op;
1141 	scmd->eh_eflags = ses->eh_eflags;
1142 }
1143 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1144 
1145 /**
1146  * scsi_send_eh_cmnd  - submit a scsi command as part of error recovery
1147  * @scmd:       SCSI command structure to hijack
1148  * @cmnd:       CDB to send
1149  * @cmnd_size:  size in bytes of @cmnd
1150  * @timeout:    timeout for this request
1151  * @sense_bytes: size of sense data to copy or 0
1152  *
1153  * This function is used to send a scsi command down to a target device
1154  * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1155  *
1156  * Return value:
1157  *    SUCCESS or FAILED or NEEDS_RETRY
1158  */
scsi_send_eh_cmnd(struct scsi_cmnd * scmd,unsigned char * cmnd,int cmnd_size,int timeout,unsigned sense_bytes)1159 static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
1160 	unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes)
1161 {
1162 	struct scsi_device *sdev = scmd->device;
1163 	struct Scsi_Host *shost = sdev->host;
1164 	DECLARE_COMPLETION_ONSTACK(done);
1165 	unsigned long timeleft = timeout, delay;
1166 	struct scsi_eh_save ses;
1167 	const unsigned long stall_for = msecs_to_jiffies(100);
1168 	int rtn;
1169 
1170 retry:
1171 	scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1172 	shost->eh_action = &done;
1173 
1174 	scsi_log_send(scmd);
1175 	scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER;
1176 	scmd->flags |= SCMD_LAST;
1177 
1178 	/*
1179 	 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1180 	 * change the SCSI device state after we have examined it and before
1181 	 * .queuecommand() is called.
1182 	 */
1183 	mutex_lock(&sdev->state_mutex);
1184 	while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1185 		mutex_unlock(&sdev->state_mutex);
1186 		SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1187 			"%s: state %d <> %d\n", __func__, sdev->sdev_state,
1188 			SDEV_BLOCK));
1189 		delay = min(timeleft, stall_for);
1190 		timeleft -= delay;
1191 		msleep(jiffies_to_msecs(delay));
1192 		mutex_lock(&sdev->state_mutex);
1193 	}
1194 	if (sdev->sdev_state != SDEV_BLOCK)
1195 		rtn = shost->hostt->queuecommand(shost, scmd);
1196 	else
1197 		rtn = FAILED;
1198 	mutex_unlock(&sdev->state_mutex);
1199 
1200 	if (rtn) {
1201 		if (timeleft > stall_for) {
1202 			scsi_eh_restore_cmnd(scmd, &ses);
1203 
1204 			timeleft -= stall_for;
1205 			msleep(jiffies_to_msecs(stall_for));
1206 			goto retry;
1207 		}
1208 		/* signal not to enter either branch of the if () below */
1209 		timeleft = 0;
1210 		rtn = FAILED;
1211 	} else {
1212 		timeleft = wait_for_completion_timeout(&done, timeout);
1213 		rtn = SUCCESS;
1214 	}
1215 
1216 	shost->eh_action = NULL;
1217 
1218 	scsi_log_completion(scmd, rtn);
1219 
1220 	SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1221 			"%s timeleft: %ld\n",
1222 			__func__, timeleft));
1223 
1224 	/*
1225 	 * If there is time left scsi_eh_done got called, and we will examine
1226 	 * the actual status codes to see whether the command actually did
1227 	 * complete normally, else if we have a zero return and no time left,
1228 	 * the command must still be pending, so abort it and return FAILED.
1229 	 * If we never actually managed to issue the command, because
1230 	 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1231 	 * value above (so don't execute either branch of the if)
1232 	 */
1233 	if (timeleft) {
1234 		rtn = scsi_eh_completed_normally(scmd);
1235 		SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1236 			"%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1237 
1238 		switch (rtn) {
1239 		case SUCCESS:
1240 		case NEEDS_RETRY:
1241 		case FAILED:
1242 			break;
1243 		case ADD_TO_MLQUEUE:
1244 			rtn = NEEDS_RETRY;
1245 			break;
1246 		default:
1247 			rtn = FAILED;
1248 			break;
1249 		}
1250 	} else if (rtn != FAILED) {
1251 		scsi_abort_eh_cmnd(scmd);
1252 		rtn = FAILED;
1253 	}
1254 
1255 	scsi_eh_restore_cmnd(scmd, &ses);
1256 
1257 	return rtn;
1258 }
1259 
1260 /**
1261  * scsi_request_sense - Request sense data from a particular target.
1262  * @scmd:	SCSI cmd for request sense.
1263  *
1264  * Notes:
1265  *    Some hosts automatically obtain this information, others require
1266  *    that we obtain it on our own. This function will *not* return until
1267  *    the command either times out, or it completes.
1268  */
scsi_request_sense(struct scsi_cmnd * scmd)1269 static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd)
1270 {
1271 	return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1272 }
1273 
1274 static enum scsi_disposition
scsi_eh_action(struct scsi_cmnd * scmd,enum scsi_disposition rtn)1275 scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn)
1276 {
1277 	if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
1278 		struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1279 		if (sdrv->eh_action)
1280 			rtn = sdrv->eh_action(scmd, rtn);
1281 	}
1282 	return rtn;
1283 }
1284 
1285 /**
1286  * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1287  * @scmd:	Original SCSI cmd that eh has finished.
1288  * @done_q:	Queue for processed commands.
1289  *
1290  * Notes:
1291  *    We don't want to use the normal command completion while we are are
1292  *    still handling errors - it may cause other commands to be queued,
1293  *    and that would disturb what we are doing.  Thus we really want to
1294  *    keep a list of pending commands for final completion, and once we
1295  *    are ready to leave error handling we handle completion for real.
1296  */
scsi_eh_finish_cmd(struct scsi_cmnd * scmd,struct list_head * done_q)1297 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1298 {
1299 	list_move_tail(&scmd->eh_entry, done_q);
1300 }
1301 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1302 
1303 /**
1304  * scsi_eh_get_sense - Get device sense data.
1305  * @work_q:	Queue of commands to process.
1306  * @done_q:	Queue of processed commands.
1307  *
1308  * Description:
1309  *    See if we need to request sense information.  if so, then get it
1310  *    now, so we have a better idea of what to do.
1311  *
1312  * Notes:
1313  *    This has the unfortunate side effect that if a shost adapter does
1314  *    not automatically request sense information, we end up shutting
1315  *    it down before we request it.
1316  *
1317  *    All drivers should request sense information internally these days,
1318  *    so for now all I have to say is tough noogies if you end up in here.
1319  *
1320  *    XXX: Long term this code should go away, but that needs an audit of
1321  *         all LLDDs first.
1322  */
scsi_eh_get_sense(struct list_head * work_q,struct list_head * done_q)1323 int scsi_eh_get_sense(struct list_head *work_q,
1324 		      struct list_head *done_q)
1325 {
1326 	struct scsi_cmnd *scmd, *next;
1327 	struct Scsi_Host *shost;
1328 	enum scsi_disposition rtn;
1329 
1330 	/*
1331 	 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1332 	 * should not get sense.
1333 	 */
1334 	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1335 		if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1336 		    SCSI_SENSE_VALID(scmd))
1337 			continue;
1338 
1339 		shost = scmd->device->host;
1340 		if (scsi_host_eh_past_deadline(shost)) {
1341 			SCSI_LOG_ERROR_RECOVERY(3,
1342 				scmd_printk(KERN_INFO, scmd,
1343 					    "%s: skip request sense, past eh deadline\n",
1344 					     current->comm));
1345 			break;
1346 		}
1347 		if (!scsi_status_is_check_condition(scmd->result))
1348 			/*
1349 			 * don't request sense if there's no check condition
1350 			 * status because the error we're processing isn't one
1351 			 * that has a sense code (and some devices get
1352 			 * confused by sense requests out of the blue)
1353 			 */
1354 			continue;
1355 
1356 		SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1357 						  "%s: requesting sense\n",
1358 						  current->comm));
1359 		rtn = scsi_request_sense(scmd);
1360 		if (rtn != SUCCESS)
1361 			continue;
1362 
1363 		SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1364 			"sense requested, result %x\n", scmd->result));
1365 		SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1366 
1367 		rtn = scsi_decide_disposition(scmd);
1368 
1369 		/*
1370 		 * if the result was normal, then just pass it along to the
1371 		 * upper level.
1372 		 */
1373 		if (rtn == SUCCESS)
1374 			/*
1375 			 * We don't want this command reissued, just finished
1376 			 * with the sense data, so set retries to the max
1377 			 * allowed to ensure it won't get reissued. If the user
1378 			 * has requested infinite retries, we also want to
1379 			 * finish this command, so force completion by setting
1380 			 * retries and allowed to the same value.
1381 			 */
1382 			if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
1383 				scmd->retries = scmd->allowed = 1;
1384 			else
1385 				scmd->retries = scmd->allowed;
1386 		else if (rtn != NEEDS_RETRY)
1387 			continue;
1388 
1389 		scsi_eh_finish_cmd(scmd, done_q);
1390 	}
1391 
1392 	return list_empty(work_q);
1393 }
1394 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1395 
1396 /**
1397  * scsi_eh_tur - Send TUR to device.
1398  * @scmd:	&scsi_cmnd to send TUR
1399  *
1400  * Return value:
1401  *    0 - Device is ready. 1 - Device NOT ready.
1402  */
scsi_eh_tur(struct scsi_cmnd * scmd)1403 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1404 {
1405 	static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1406 	int retry_cnt = 1;
1407 	enum scsi_disposition rtn;
1408 
1409 retry_tur:
1410 	rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1411 				scmd->device->eh_timeout, 0);
1412 
1413 	SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1414 		"%s return: %x\n", __func__, rtn));
1415 
1416 	switch (rtn) {
1417 	case NEEDS_RETRY:
1418 		if (retry_cnt--)
1419 			goto retry_tur;
1420 		fallthrough;
1421 	case SUCCESS:
1422 		return 0;
1423 	default:
1424 		return 1;
1425 	}
1426 }
1427 
1428 /**
1429  * scsi_eh_test_devices - check if devices are responding from error recovery.
1430  * @cmd_list:	scsi commands in error recovery.
1431  * @work_q:	queue for commands which still need more error recovery
1432  * @done_q:	queue for commands which are finished
1433  * @try_stu:	boolean on if a STU command should be tried in addition to TUR.
1434  *
1435  * Decription:
1436  *    Tests if devices are in a working state.  Commands to devices now in
1437  *    a working state are sent to the done_q while commands to devices which
1438  *    are still failing to respond are returned to the work_q for more
1439  *    processing.
1440  **/
scsi_eh_test_devices(struct list_head * cmd_list,struct list_head * work_q,struct list_head * done_q,int try_stu)1441 static int scsi_eh_test_devices(struct list_head *cmd_list,
1442 				struct list_head *work_q,
1443 				struct list_head *done_q, int try_stu)
1444 {
1445 	struct scsi_cmnd *scmd, *next;
1446 	struct scsi_device *sdev;
1447 	int finish_cmds;
1448 
1449 	while (!list_empty(cmd_list)) {
1450 		scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1451 		sdev = scmd->device;
1452 
1453 		if (!try_stu) {
1454 			if (scsi_host_eh_past_deadline(sdev->host)) {
1455 				/* Push items back onto work_q */
1456 				list_splice_init(cmd_list, work_q);
1457 				SCSI_LOG_ERROR_RECOVERY(3,
1458 					sdev_printk(KERN_INFO, sdev,
1459 						    "%s: skip test device, past eh deadline",
1460 						    current->comm));
1461 				break;
1462 			}
1463 		}
1464 
1465 		finish_cmds = !scsi_device_online(scmd->device) ||
1466 			(try_stu && !scsi_eh_try_stu(scmd) &&
1467 			 !scsi_eh_tur(scmd)) ||
1468 			!scsi_eh_tur(scmd);
1469 
1470 		list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1471 			if (scmd->device == sdev) {
1472 				if (finish_cmds &&
1473 				    (try_stu ||
1474 				     scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1475 					scsi_eh_finish_cmd(scmd, done_q);
1476 				else
1477 					list_move_tail(&scmd->eh_entry, work_q);
1478 			}
1479 	}
1480 	return list_empty(work_q);
1481 }
1482 
1483 /**
1484  * scsi_eh_try_stu - Send START_UNIT to device.
1485  * @scmd:	&scsi_cmnd to send START_UNIT
1486  *
1487  * Return value:
1488  *    0 - Device is ready. 1 - Device NOT ready.
1489  */
scsi_eh_try_stu(struct scsi_cmnd * scmd)1490 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1491 {
1492 	static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1493 
1494 	if (scmd->device->allow_restart) {
1495 		int i;
1496 		enum scsi_disposition rtn = NEEDS_RETRY;
1497 
1498 		for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1499 			rtn = scsi_send_eh_cmnd(scmd, stu_command, 6,
1500 						scmd->device->eh_timeout, 0);
1501 
1502 		if (rtn == SUCCESS)
1503 			return 0;
1504 	}
1505 
1506 	return 1;
1507 }
1508 
1509  /**
1510  * scsi_eh_stu - send START_UNIT if needed
1511  * @shost:	&scsi host being recovered.
1512  * @work_q:	&list_head for pending commands.
1513  * @done_q:	&list_head for processed commands.
1514  *
1515  * Notes:
1516  *    If commands are failing due to not ready, initializing command required,
1517  *	try revalidating the device, which will end up sending a start unit.
1518  */
scsi_eh_stu(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1519 static int scsi_eh_stu(struct Scsi_Host *shost,
1520 			      struct list_head *work_q,
1521 			      struct list_head *done_q)
1522 {
1523 	struct scsi_cmnd *scmd, *stu_scmd, *next;
1524 	struct scsi_device *sdev;
1525 
1526 	shost_for_each_device(sdev, shost) {
1527 		if (scsi_host_eh_past_deadline(shost)) {
1528 			SCSI_LOG_ERROR_RECOVERY(3,
1529 				sdev_printk(KERN_INFO, sdev,
1530 					    "%s: skip START_UNIT, past eh deadline\n",
1531 					    current->comm));
1532 			scsi_device_put(sdev);
1533 			break;
1534 		}
1535 		stu_scmd = NULL;
1536 		list_for_each_entry(scmd, work_q, eh_entry)
1537 			if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1538 			    scsi_check_sense(scmd) == FAILED ) {
1539 				stu_scmd = scmd;
1540 				break;
1541 			}
1542 
1543 		if (!stu_scmd)
1544 			continue;
1545 
1546 		SCSI_LOG_ERROR_RECOVERY(3,
1547 			sdev_printk(KERN_INFO, sdev,
1548 				     "%s: Sending START_UNIT\n",
1549 				    current->comm));
1550 
1551 		if (!scsi_eh_try_stu(stu_scmd)) {
1552 			if (!scsi_device_online(sdev) ||
1553 			    !scsi_eh_tur(stu_scmd)) {
1554 				list_for_each_entry_safe(scmd, next,
1555 							  work_q, eh_entry) {
1556 					if (scmd->device == sdev &&
1557 					    scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1558 						scsi_eh_finish_cmd(scmd, done_q);
1559 				}
1560 			}
1561 		} else {
1562 			SCSI_LOG_ERROR_RECOVERY(3,
1563 				sdev_printk(KERN_INFO, sdev,
1564 					    "%s: START_UNIT failed\n",
1565 					    current->comm));
1566 		}
1567 	}
1568 
1569 	return list_empty(work_q);
1570 }
1571 
1572 
1573 /**
1574  * scsi_eh_bus_device_reset - send bdr if needed
1575  * @shost:	scsi host being recovered.
1576  * @work_q:	&list_head for pending commands.
1577  * @done_q:	&list_head for processed commands.
1578  *
1579  * Notes:
1580  *    Try a bus device reset.  Still, look to see whether we have multiple
1581  *    devices that are jammed or not - if we have multiple devices, it
1582  *    makes no sense to try bus_device_reset - we really would need to try
1583  *    a bus_reset instead.
1584  */
scsi_eh_bus_device_reset(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1585 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1586 				    struct list_head *work_q,
1587 				    struct list_head *done_q)
1588 {
1589 	struct scsi_cmnd *scmd, *bdr_scmd, *next;
1590 	struct scsi_device *sdev;
1591 	enum scsi_disposition rtn;
1592 
1593 	shost_for_each_device(sdev, shost) {
1594 		if (scsi_host_eh_past_deadline(shost)) {
1595 			SCSI_LOG_ERROR_RECOVERY(3,
1596 				sdev_printk(KERN_INFO, sdev,
1597 					    "%s: skip BDR, past eh deadline\n",
1598 					     current->comm));
1599 			scsi_device_put(sdev);
1600 			break;
1601 		}
1602 		bdr_scmd = NULL;
1603 		list_for_each_entry(scmd, work_q, eh_entry)
1604 			if (scmd->device == sdev) {
1605 				bdr_scmd = scmd;
1606 				break;
1607 			}
1608 
1609 		if (!bdr_scmd)
1610 			continue;
1611 
1612 		SCSI_LOG_ERROR_RECOVERY(3,
1613 			sdev_printk(KERN_INFO, sdev,
1614 				     "%s: Sending BDR\n", current->comm));
1615 		rtn = scsi_try_bus_device_reset(bdr_scmd);
1616 		if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1617 			if (!scsi_device_online(sdev) ||
1618 			    rtn == FAST_IO_FAIL ||
1619 			    !scsi_eh_tur(bdr_scmd)) {
1620 				list_for_each_entry_safe(scmd, next,
1621 							 work_q, eh_entry) {
1622 					if (scmd->device == sdev &&
1623 					    scsi_eh_action(scmd, rtn) != FAILED)
1624 						scsi_eh_finish_cmd(scmd,
1625 								   done_q);
1626 				}
1627 			}
1628 		} else {
1629 			SCSI_LOG_ERROR_RECOVERY(3,
1630 				sdev_printk(KERN_INFO, sdev,
1631 					    "%s: BDR failed\n", current->comm));
1632 		}
1633 	}
1634 
1635 	return list_empty(work_q);
1636 }
1637 
1638 /**
1639  * scsi_eh_target_reset - send target reset if needed
1640  * @shost:	scsi host being recovered.
1641  * @work_q:	&list_head for pending commands.
1642  * @done_q:	&list_head for processed commands.
1643  *
1644  * Notes:
1645  *    Try a target reset.
1646  */
scsi_eh_target_reset(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1647 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1648 				struct list_head *work_q,
1649 				struct list_head *done_q)
1650 {
1651 	LIST_HEAD(tmp_list);
1652 	LIST_HEAD(check_list);
1653 
1654 	list_splice_init(work_q, &tmp_list);
1655 
1656 	while (!list_empty(&tmp_list)) {
1657 		struct scsi_cmnd *next, *scmd;
1658 		enum scsi_disposition rtn;
1659 		unsigned int id;
1660 
1661 		if (scsi_host_eh_past_deadline(shost)) {
1662 			/* push back on work queue for further processing */
1663 			list_splice_init(&check_list, work_q);
1664 			list_splice_init(&tmp_list, work_q);
1665 			SCSI_LOG_ERROR_RECOVERY(3,
1666 				shost_printk(KERN_INFO, shost,
1667 					    "%s: Skip target reset, past eh deadline\n",
1668 					     current->comm));
1669 			return list_empty(work_q);
1670 		}
1671 
1672 		scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1673 		id = scmd_id(scmd);
1674 
1675 		SCSI_LOG_ERROR_RECOVERY(3,
1676 			shost_printk(KERN_INFO, shost,
1677 				     "%s: Sending target reset to target %d\n",
1678 				     current->comm, id));
1679 		rtn = scsi_try_target_reset(scmd);
1680 		if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1681 			SCSI_LOG_ERROR_RECOVERY(3,
1682 				shost_printk(KERN_INFO, shost,
1683 					     "%s: Target reset failed"
1684 					     " target: %d\n",
1685 					     current->comm, id));
1686 		list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1687 			if (scmd_id(scmd) != id)
1688 				continue;
1689 
1690 			if (rtn == SUCCESS)
1691 				list_move_tail(&scmd->eh_entry, &check_list);
1692 			else if (rtn == FAST_IO_FAIL)
1693 				scsi_eh_finish_cmd(scmd, done_q);
1694 			else
1695 				/* push back on work queue for further processing */
1696 				list_move(&scmd->eh_entry, work_q);
1697 		}
1698 	}
1699 
1700 	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1701 }
1702 
1703 /**
1704  * scsi_eh_bus_reset - send a bus reset
1705  * @shost:	&scsi host being recovered.
1706  * @work_q:	&list_head for pending commands.
1707  * @done_q:	&list_head for processed commands.
1708  */
scsi_eh_bus_reset(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1709 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1710 			     struct list_head *work_q,
1711 			     struct list_head *done_q)
1712 {
1713 	struct scsi_cmnd *scmd, *chan_scmd, *next;
1714 	LIST_HEAD(check_list);
1715 	unsigned int channel;
1716 	enum scsi_disposition rtn;
1717 
1718 	/*
1719 	 * we really want to loop over the various channels, and do this on
1720 	 * a channel by channel basis.  we should also check to see if any
1721 	 * of the failed commands are on soft_reset devices, and if so, skip
1722 	 * the reset.
1723 	 */
1724 
1725 	for (channel = 0; channel <= shost->max_channel; channel++) {
1726 		if (scsi_host_eh_past_deadline(shost)) {
1727 			list_splice_init(&check_list, work_q);
1728 			SCSI_LOG_ERROR_RECOVERY(3,
1729 				shost_printk(KERN_INFO, shost,
1730 					    "%s: skip BRST, past eh deadline\n",
1731 					     current->comm));
1732 			return list_empty(work_q);
1733 		}
1734 
1735 		chan_scmd = NULL;
1736 		list_for_each_entry(scmd, work_q, eh_entry) {
1737 			if (channel == scmd_channel(scmd)) {
1738 				chan_scmd = scmd;
1739 				break;
1740 				/*
1741 				 * FIXME add back in some support for
1742 				 * soft_reset devices.
1743 				 */
1744 			}
1745 		}
1746 
1747 		if (!chan_scmd)
1748 			continue;
1749 		SCSI_LOG_ERROR_RECOVERY(3,
1750 			shost_printk(KERN_INFO, shost,
1751 				     "%s: Sending BRST chan: %d\n",
1752 				     current->comm, channel));
1753 		rtn = scsi_try_bus_reset(chan_scmd);
1754 		if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1755 			list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1756 				if (channel == scmd_channel(scmd)) {
1757 					if (rtn == FAST_IO_FAIL)
1758 						scsi_eh_finish_cmd(scmd,
1759 								   done_q);
1760 					else
1761 						list_move_tail(&scmd->eh_entry,
1762 							       &check_list);
1763 				}
1764 			}
1765 		} else {
1766 			SCSI_LOG_ERROR_RECOVERY(3,
1767 				shost_printk(KERN_INFO, shost,
1768 					     "%s: BRST failed chan: %d\n",
1769 					     current->comm, channel));
1770 		}
1771 	}
1772 	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1773 }
1774 
1775 /**
1776  * scsi_eh_host_reset - send a host reset
1777  * @shost:	host to be reset.
1778  * @work_q:	&list_head for pending commands.
1779  * @done_q:	&list_head for processed commands.
1780  */
scsi_eh_host_reset(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1781 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1782 			      struct list_head *work_q,
1783 			      struct list_head *done_q)
1784 {
1785 	struct scsi_cmnd *scmd, *next;
1786 	LIST_HEAD(check_list);
1787 	enum scsi_disposition rtn;
1788 
1789 	if (!list_empty(work_q)) {
1790 		scmd = list_entry(work_q->next,
1791 				  struct scsi_cmnd, eh_entry);
1792 
1793 		SCSI_LOG_ERROR_RECOVERY(3,
1794 			shost_printk(KERN_INFO, shost,
1795 				     "%s: Sending HRST\n",
1796 				     current->comm));
1797 
1798 		rtn = scsi_try_host_reset(scmd);
1799 		if (rtn == SUCCESS) {
1800 			list_splice_init(work_q, &check_list);
1801 		} else if (rtn == FAST_IO_FAIL) {
1802 			list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1803 					scsi_eh_finish_cmd(scmd, done_q);
1804 			}
1805 		} else {
1806 			SCSI_LOG_ERROR_RECOVERY(3,
1807 				shost_printk(KERN_INFO, shost,
1808 					     "%s: HRST failed\n",
1809 					     current->comm));
1810 		}
1811 	}
1812 	return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1813 }
1814 
1815 /**
1816  * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1817  * @work_q:	&list_head for pending commands.
1818  * @done_q:	&list_head for processed commands.
1819  */
scsi_eh_offline_sdevs(struct list_head * work_q,struct list_head * done_q)1820 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1821 				  struct list_head *done_q)
1822 {
1823 	struct scsi_cmnd *scmd, *next;
1824 	struct scsi_device *sdev;
1825 
1826 	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1827 		sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1828 			    "not ready after error recovery\n");
1829 		sdev = scmd->device;
1830 
1831 		mutex_lock(&sdev->state_mutex);
1832 		scsi_device_set_state(sdev, SDEV_OFFLINE);
1833 		mutex_unlock(&sdev->state_mutex);
1834 
1835 		scsi_eh_finish_cmd(scmd, done_q);
1836 	}
1837 	return;
1838 }
1839 
1840 /**
1841  * scsi_noretry_cmd - determine if command should be failed fast
1842  * @scmd:	SCSI cmd to examine.
1843  */
scsi_noretry_cmd(struct scsi_cmnd * scmd)1844 bool scsi_noretry_cmd(struct scsi_cmnd *scmd)
1845 {
1846 	struct request *req = scsi_cmd_to_rq(scmd);
1847 
1848 	switch (host_byte(scmd->result)) {
1849 	case DID_OK:
1850 		break;
1851 	case DID_TIME_OUT:
1852 		goto check_type;
1853 	case DID_BUS_BUSY:
1854 		return !!(req->cmd_flags & REQ_FAILFAST_TRANSPORT);
1855 	case DID_PARITY:
1856 		return !!(req->cmd_flags & REQ_FAILFAST_DEV);
1857 	case DID_ERROR:
1858 		if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
1859 			return false;
1860 		fallthrough;
1861 	case DID_SOFT_ERROR:
1862 		return !!(req->cmd_flags & REQ_FAILFAST_DRIVER);
1863 	}
1864 
1865 	/* Never retry commands aborted due to a duration limit timeout */
1866 	if (scsi_ml_byte(scmd->result) == SCSIML_STAT_DL_TIMEOUT)
1867 		return true;
1868 
1869 	if (!scsi_status_is_check_condition(scmd->result))
1870 		return false;
1871 
1872 check_type:
1873 	/*
1874 	 * assume caller has checked sense and determined
1875 	 * the check condition was retryable.
1876 	 */
1877 	if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(req))
1878 		return true;
1879 
1880 	return false;
1881 }
1882 
1883 /**
1884  * scsi_decide_disposition - Disposition a cmd on return from LLD.
1885  * @scmd:	SCSI cmd to examine.
1886  *
1887  * Notes:
1888  *    This is *only* called when we are examining the status after sending
1889  *    out the actual data command.  any commands that are queued for error
1890  *    recovery (e.g. test_unit_ready) do *not* come through here.
1891  *
1892  *    When this routine returns failed, it means the error handler thread
1893  *    is woken.  In cases where the error code indicates an error that
1894  *    doesn't require the error handler read (i.e. we don't need to
1895  *    abort/reset), this function should return SUCCESS.
1896  */
scsi_decide_disposition(struct scsi_cmnd * scmd)1897 enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd)
1898 {
1899 	enum scsi_disposition rtn;
1900 
1901 	/*
1902 	 * if the device is offline, then we clearly just pass the result back
1903 	 * up to the top level.
1904 	 */
1905 	if (!scsi_device_online(scmd->device)) {
1906 		SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1907 			"%s: device offline - report as SUCCESS\n", __func__));
1908 		return SUCCESS;
1909 	}
1910 
1911 	/*
1912 	 * first check the host byte, to see if there is anything in there
1913 	 * that would indicate what we need to do.
1914 	 */
1915 	switch (host_byte(scmd->result)) {
1916 	case DID_PASSTHROUGH:
1917 		/*
1918 		 * no matter what, pass this through to the upper layer.
1919 		 * nuke this special code so that it looks like we are saying
1920 		 * did_ok.
1921 		 */
1922 		scmd->result &= 0xff00ffff;
1923 		return SUCCESS;
1924 	case DID_OK:
1925 		/*
1926 		 * looks good.  drop through, and check the next byte.
1927 		 */
1928 		break;
1929 	case DID_ABORT:
1930 		if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1931 			set_host_byte(scmd, DID_TIME_OUT);
1932 			return SUCCESS;
1933 		}
1934 		fallthrough;
1935 	case DID_NO_CONNECT:
1936 	case DID_BAD_TARGET:
1937 		/*
1938 		 * note - this means that we just report the status back
1939 		 * to the top level driver, not that we actually think
1940 		 * that it indicates SUCCESS.
1941 		 */
1942 		return SUCCESS;
1943 	case DID_SOFT_ERROR:
1944 		/*
1945 		 * when the low level driver returns did_soft_error,
1946 		 * it is responsible for keeping an internal retry counter
1947 		 * in order to avoid endless loops (db)
1948 		 */
1949 		goto maybe_retry;
1950 	case DID_IMM_RETRY:
1951 		return NEEDS_RETRY;
1952 
1953 	case DID_REQUEUE:
1954 		return ADD_TO_MLQUEUE;
1955 	case DID_TRANSPORT_DISRUPTED:
1956 		/*
1957 		 * LLD/transport was disrupted during processing of the IO.
1958 		 * The transport class is now blocked/blocking,
1959 		 * and the transport will decide what to do with the IO
1960 		 * based on its timers and recovery capablilities if
1961 		 * there are enough retries.
1962 		 */
1963 		goto maybe_retry;
1964 	case DID_TRANSPORT_FAILFAST:
1965 		/*
1966 		 * The transport decided to failfast the IO (most likely
1967 		 * the fast io fail tmo fired), so send IO directly upwards.
1968 		 */
1969 		return SUCCESS;
1970 	case DID_TRANSPORT_MARGINAL:
1971 		/*
1972 		 * caller has decided not to do retries on
1973 		 * abort success, so send IO directly upwards
1974 		 */
1975 		return SUCCESS;
1976 	case DID_ERROR:
1977 		if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
1978 			/*
1979 			 * execute reservation conflict processing code
1980 			 * lower down
1981 			 */
1982 			break;
1983 		fallthrough;
1984 	case DID_BUS_BUSY:
1985 	case DID_PARITY:
1986 		goto maybe_retry;
1987 	case DID_TIME_OUT:
1988 		/*
1989 		 * when we scan the bus, we get timeout messages for
1990 		 * these commands if there is no device available.
1991 		 * other hosts report did_no_connect for the same thing.
1992 		 */
1993 		if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1994 		     scmd->cmnd[0] == INQUIRY)) {
1995 			return SUCCESS;
1996 		} else {
1997 			return FAILED;
1998 		}
1999 	case DID_RESET:
2000 		return SUCCESS;
2001 	default:
2002 		return FAILED;
2003 	}
2004 
2005 	/*
2006 	 * check the status byte to see if this indicates anything special.
2007 	 */
2008 	switch (get_status_byte(scmd)) {
2009 	case SAM_STAT_TASK_SET_FULL:
2010 		scsi_handle_queue_full(scmd->device);
2011 		/*
2012 		 * the case of trying to send too many commands to a
2013 		 * tagged queueing device.
2014 		 */
2015 		fallthrough;
2016 	case SAM_STAT_BUSY:
2017 		/*
2018 		 * device can't talk to us at the moment.  Should only
2019 		 * occur (SAM-3) when the task queue is empty, so will cause
2020 		 * the empty queue handling to trigger a stall in the
2021 		 * device.
2022 		 */
2023 		return ADD_TO_MLQUEUE;
2024 	case SAM_STAT_GOOD:
2025 		if (scmd->cmnd[0] == REPORT_LUNS)
2026 			scmd->device->sdev_target->expecting_lun_change = 0;
2027 		scsi_handle_queue_ramp_up(scmd->device);
2028 		if (scmd->sense_buffer && SCSI_SENSE_VALID(scmd))
2029 			/*
2030 			 * If we have sense data, call scsi_check_sense() in
2031 			 * order to set the correct SCSI ML byte (if any).
2032 			 * No point in checking the return value, since the
2033 			 * command has already completed successfully.
2034 			 */
2035 			scsi_check_sense(scmd);
2036 		fallthrough;
2037 	case SAM_STAT_COMMAND_TERMINATED:
2038 		return SUCCESS;
2039 	case SAM_STAT_TASK_ABORTED:
2040 		goto maybe_retry;
2041 	case SAM_STAT_CHECK_CONDITION:
2042 		rtn = scsi_check_sense(scmd);
2043 		if (rtn == NEEDS_RETRY)
2044 			goto maybe_retry;
2045 		/* if rtn == FAILED, we have no sense information;
2046 		 * returning FAILED will wake the error handler thread
2047 		 * to collect the sense and redo the decide
2048 		 * disposition */
2049 		return rtn;
2050 	case SAM_STAT_CONDITION_MET:
2051 	case SAM_STAT_INTERMEDIATE:
2052 	case SAM_STAT_INTERMEDIATE_CONDITION_MET:
2053 	case SAM_STAT_ACA_ACTIVE:
2054 		/*
2055 		 * who knows?  FIXME(eric)
2056 		 */
2057 		return SUCCESS;
2058 
2059 	case SAM_STAT_RESERVATION_CONFLICT:
2060 		sdev_printk(KERN_INFO, scmd->device,
2061 			    "reservation conflict\n");
2062 		set_scsi_ml_byte(scmd, SCSIML_STAT_RESV_CONFLICT);
2063 		return SUCCESS; /* causes immediate i/o error */
2064 	}
2065 	return FAILED;
2066 
2067 maybe_retry:
2068 
2069 	/* we requeue for retry because the error was retryable, and
2070 	 * the request was not marked fast fail.  Note that above,
2071 	 * even if the request is marked fast fail, we still requeue
2072 	 * for queue congestion conditions (QUEUE_FULL or BUSY) */
2073 	if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) {
2074 		return NEEDS_RETRY;
2075 	} else {
2076 		/*
2077 		 * no more retries - report this one back to upper level.
2078 		 */
2079 		return SUCCESS;
2080 	}
2081 }
2082 
eh_lock_door_done(struct request * req,blk_status_t status)2083 static enum rq_end_io_ret eh_lock_door_done(struct request *req,
2084 					    blk_status_t status)
2085 {
2086 	blk_mq_free_request(req);
2087 	return RQ_END_IO_NONE;
2088 }
2089 
2090 /**
2091  * scsi_eh_lock_door - Prevent medium removal for the specified device
2092  * @sdev:	SCSI device to prevent medium removal
2093  *
2094  * Locking:
2095  * 	We must be called from process context.
2096  *
2097  * Notes:
2098  * 	We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
2099  * 	head of the devices request queue, and continue.
2100  */
scsi_eh_lock_door(struct scsi_device * sdev)2101 static void scsi_eh_lock_door(struct scsi_device *sdev)
2102 {
2103 	struct scsi_cmnd *scmd;
2104 	struct request *req;
2105 
2106 	req = scsi_alloc_request(sdev->request_queue, REQ_OP_DRV_IN, 0);
2107 	if (IS_ERR(req))
2108 		return;
2109 	scmd = blk_mq_rq_to_pdu(req);
2110 
2111 	scmd->cmnd[0] = ALLOW_MEDIUM_REMOVAL;
2112 	scmd->cmnd[1] = 0;
2113 	scmd->cmnd[2] = 0;
2114 	scmd->cmnd[3] = 0;
2115 	scmd->cmnd[4] = SCSI_REMOVAL_PREVENT;
2116 	scmd->cmnd[5] = 0;
2117 	scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
2118 	scmd->allowed = 5;
2119 
2120 	req->rq_flags |= RQF_QUIET;
2121 	req->timeout = 10 * HZ;
2122 	req->end_io = eh_lock_door_done;
2123 
2124 	blk_execute_rq_nowait(req, true);
2125 }
2126 
2127 /**
2128  * scsi_restart_operations - restart io operations to the specified host.
2129  * @shost:	Host we are restarting.
2130  *
2131  * Notes:
2132  *    When we entered the error handler, we blocked all further i/o to
2133  *    this device.  we need to 'reverse' this process.
2134  */
scsi_restart_operations(struct Scsi_Host * shost)2135 static void scsi_restart_operations(struct Scsi_Host *shost)
2136 {
2137 	struct scsi_device *sdev;
2138 	unsigned long flags;
2139 
2140 	/*
2141 	 * If the door was locked, we need to insert a door lock request
2142 	 * onto the head of the SCSI request queue for the device.  There
2143 	 * is no point trying to lock the door of an off-line device.
2144 	 */
2145 	shost_for_each_device(sdev, shost) {
2146 		if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2147 			scsi_eh_lock_door(sdev);
2148 			sdev->was_reset = 0;
2149 		}
2150 	}
2151 
2152 	/*
2153 	 * next free up anything directly waiting upon the host.  this
2154 	 * will be requests for character device operations, and also for
2155 	 * ioctls to queued block devices.
2156 	 */
2157 	SCSI_LOG_ERROR_RECOVERY(3,
2158 		shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2159 
2160 	spin_lock_irqsave(shost->host_lock, flags);
2161 	if (scsi_host_set_state(shost, SHOST_RUNNING))
2162 		if (scsi_host_set_state(shost, SHOST_CANCEL))
2163 			BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2164 	spin_unlock_irqrestore(shost->host_lock, flags);
2165 
2166 	wake_up(&shost->host_wait);
2167 
2168 	/*
2169 	 * finally we need to re-initiate requests that may be pending.  we will
2170 	 * have had everything blocked while error handling is taking place, and
2171 	 * now that error recovery is done, we will need to ensure that these
2172 	 * requests are started.
2173 	 */
2174 	scsi_run_host_queues(shost);
2175 
2176 	/*
2177 	 * if eh is active and host_eh_scheduled is pending we need to re-run
2178 	 * recovery.  we do this check after scsi_run_host_queues() to allow
2179 	 * everything pent up since the last eh run a chance to make forward
2180 	 * progress before we sync again.  Either we'll immediately re-run
2181 	 * recovery or scsi_device_unbusy() will wake us again when these
2182 	 * pending commands complete.
2183 	 */
2184 	spin_lock_irqsave(shost->host_lock, flags);
2185 	if (shost->host_eh_scheduled)
2186 		if (scsi_host_set_state(shost, SHOST_RECOVERY))
2187 			WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2188 	spin_unlock_irqrestore(shost->host_lock, flags);
2189 }
2190 
2191 /**
2192  * scsi_eh_ready_devs - check device ready state and recover if not.
2193  * @shost:	host to be recovered.
2194  * @work_q:	&list_head for pending commands.
2195  * @done_q:	&list_head for processed commands.
2196  */
scsi_eh_ready_devs(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)2197 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2198 			struct list_head *work_q,
2199 			struct list_head *done_q)
2200 {
2201 	if (!scsi_eh_stu(shost, work_q, done_q))
2202 		if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2203 			if (!scsi_eh_target_reset(shost, work_q, done_q))
2204 				if (!scsi_eh_bus_reset(shost, work_q, done_q))
2205 					if (!scsi_eh_host_reset(shost, work_q, done_q))
2206 						scsi_eh_offline_sdevs(work_q,
2207 								      done_q);
2208 }
2209 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2210 
2211 /**
2212  * scsi_eh_flush_done_q - finish processed commands or retry them.
2213  * @done_q:	list_head of processed commands.
2214  */
scsi_eh_flush_done_q(struct list_head * done_q)2215 void scsi_eh_flush_done_q(struct list_head *done_q)
2216 {
2217 	struct scsi_cmnd *scmd, *next;
2218 
2219 	list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2220 		struct scsi_device *sdev = scmd->device;
2221 
2222 		list_del_init(&scmd->eh_entry);
2223 		if (scsi_device_online(sdev) && !scsi_noretry_cmd(scmd) &&
2224 		    scsi_cmd_retry_allowed(scmd) &&
2225 		    scsi_eh_should_retry_cmd(scmd)) {
2226 			SCSI_LOG_ERROR_RECOVERY(3,
2227 				scmd_printk(KERN_INFO, scmd,
2228 					     "%s: flush retry cmd\n",
2229 					     current->comm));
2230 				scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2231 				blk_mq_kick_requeue_list(sdev->request_queue);
2232 		} else {
2233 			/*
2234 			 * If just we got sense for the device (called
2235 			 * scsi_eh_get_sense), scmd->result is already
2236 			 * set, do not set DID_TIME_OUT.
2237 			 */
2238 			if (!scmd->result &&
2239 			    !(scmd->flags & SCMD_FORCE_EH_SUCCESS))
2240 				scmd->result |= (DID_TIME_OUT << 16);
2241 			SCSI_LOG_ERROR_RECOVERY(3,
2242 				scmd_printk(KERN_INFO, scmd,
2243 					     "%s: flush finish cmd\n",
2244 					     current->comm));
2245 			scsi_finish_command(scmd);
2246 		}
2247 	}
2248 }
2249 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2250 
2251 /**
2252  * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2253  * @shost:	Host to unjam.
2254  *
2255  * Notes:
2256  *    When we come in here, we *know* that all commands on the bus have
2257  *    either completed, failed or timed out.  we also know that no further
2258  *    commands are being sent to the host, so things are relatively quiet
2259  *    and we have freedom to fiddle with things as we wish.
2260  *
2261  *    This is only the *default* implementation.  it is possible for
2262  *    individual drivers to supply their own version of this function, and
2263  *    if the maintainer wishes to do this, it is strongly suggested that
2264  *    this function be taken as a template and modified.  this function
2265  *    was designed to correctly handle problems for about 95% of the
2266  *    different cases out there, and it should always provide at least a
2267  *    reasonable amount of error recovery.
2268  *
2269  *    Any command marked 'failed' or 'timeout' must eventually have
2270  *    scsi_finish_cmd() called for it.  we do all of the retry stuff
2271  *    here, so when we restart the host after we return it should have an
2272  *    empty queue.
2273  */
scsi_unjam_host(struct Scsi_Host * shost)2274 static void scsi_unjam_host(struct Scsi_Host *shost)
2275 {
2276 	unsigned long flags;
2277 	LIST_HEAD(eh_work_q);
2278 	LIST_HEAD(eh_done_q);
2279 
2280 	spin_lock_irqsave(shost->host_lock, flags);
2281 	list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2282 	spin_unlock_irqrestore(shost->host_lock, flags);
2283 
2284 	SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2285 
2286 	if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2287 		scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2288 
2289 	spin_lock_irqsave(shost->host_lock, flags);
2290 	if (shost->eh_deadline != -1)
2291 		shost->last_reset = 0;
2292 	spin_unlock_irqrestore(shost->host_lock, flags);
2293 	scsi_eh_flush_done_q(&eh_done_q);
2294 }
2295 
2296 /**
2297  * scsi_error_handler - SCSI error handler thread
2298  * @data:	Host for which we are running.
2299  *
2300  * Notes:
2301  *    This is the main error handling loop.  This is run as a kernel thread
2302  *    for every SCSI host and handles all error handling activity.
2303  */
scsi_error_handler(void * data)2304 int scsi_error_handler(void *data)
2305 {
2306 	struct Scsi_Host *shost = data;
2307 
2308 	/*
2309 	 * We use TASK_INTERRUPTIBLE so that the thread is not
2310 	 * counted against the load average as a running process.
2311 	 * We never actually get interrupted because kthread_run
2312 	 * disables signal delivery for the created thread.
2313 	 */
2314 	while (true) {
2315 		/*
2316 		 * The sequence in kthread_stop() sets the stop flag first
2317 		 * then wakes the process.  To avoid missed wakeups, the task
2318 		 * should always be in a non running state before the stop
2319 		 * flag is checked
2320 		 */
2321 		set_current_state(TASK_INTERRUPTIBLE);
2322 		if (kthread_should_stop())
2323 			break;
2324 
2325 		if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2326 		    shost->host_failed != scsi_host_busy(shost)) {
2327 			SCSI_LOG_ERROR_RECOVERY(1,
2328 				shost_printk(KERN_INFO, shost,
2329 					     "scsi_eh_%d: sleeping\n",
2330 					     shost->host_no));
2331 			schedule();
2332 			continue;
2333 		}
2334 
2335 		__set_current_state(TASK_RUNNING);
2336 		SCSI_LOG_ERROR_RECOVERY(1,
2337 			shost_printk(KERN_INFO, shost,
2338 				     "scsi_eh_%d: waking up %d/%d/%d\n",
2339 				     shost->host_no, shost->host_eh_scheduled,
2340 				     shost->host_failed,
2341 				     scsi_host_busy(shost)));
2342 
2343 		/*
2344 		 * We have a host that is failing for some reason.  Figure out
2345 		 * what we need to do to get it up and online again (if we can).
2346 		 * If we fail, we end up taking the thing offline.
2347 		 */
2348 		if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2349 			SCSI_LOG_ERROR_RECOVERY(1,
2350 				shost_printk(KERN_ERR, shost,
2351 					     "scsi_eh_%d: unable to autoresume\n",
2352 					     shost->host_no));
2353 			continue;
2354 		}
2355 
2356 		if (shost->transportt->eh_strategy_handler)
2357 			shost->transportt->eh_strategy_handler(shost);
2358 		else
2359 			scsi_unjam_host(shost);
2360 
2361 		/* All scmds have been handled */
2362 		shost->host_failed = 0;
2363 
2364 		/*
2365 		 * Note - if the above fails completely, the action is to take
2366 		 * individual devices offline and flush the queue of any
2367 		 * outstanding requests that may have been pending.  When we
2368 		 * restart, we restart any I/O to any other devices on the bus
2369 		 * which are still online.
2370 		 */
2371 		scsi_restart_operations(shost);
2372 		if (!shost->eh_noresume)
2373 			scsi_autopm_put_host(shost);
2374 	}
2375 	__set_current_state(TASK_RUNNING);
2376 
2377 	SCSI_LOG_ERROR_RECOVERY(1,
2378 		shost_printk(KERN_INFO, shost,
2379 			     "Error handler scsi_eh_%d exiting\n",
2380 			     shost->host_no));
2381 	shost->ehandler = NULL;
2382 	return 0;
2383 }
2384 
2385 /**
2386  * scsi_report_bus_reset() - report bus reset observed
2387  *
2388  * Utility function used by low-level drivers to report that
2389  * they have observed a bus reset on the bus being handled.
2390  *
2391  * @shost:      Host in question
2392  * @channel:    channel on which reset was observed.
2393  *
2394  * Returns:     Nothing
2395  *
2396  * Lock status: Host lock must be held.
2397  *
2398  * Notes:       This only needs to be called if the reset is one which
2399  *		originates from an unknown location.  Resets originated
2400  *		by the mid-level itself don't need to call this, but there
2401  *		should be no harm.
2402  *
2403  *		The main purpose of this is to make sure that a CHECK_CONDITION
2404  *		is properly treated.
2405  */
scsi_report_bus_reset(struct Scsi_Host * shost,int channel)2406 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2407 {
2408 	struct scsi_device *sdev;
2409 
2410 	__shost_for_each_device(sdev, shost) {
2411 		if (channel == sdev_channel(sdev))
2412 			__scsi_report_device_reset(sdev, NULL);
2413 	}
2414 }
2415 EXPORT_SYMBOL(scsi_report_bus_reset);
2416 
2417 /**
2418  * scsi_report_device_reset() - report device reset observed
2419  *
2420  * Utility function used by low-level drivers to report that
2421  * they have observed a device reset on the device being handled.
2422  *
2423  * @shost:      Host in question
2424  * @channel:    channel on which reset was observed
2425  * @target:     target on which reset was observed
2426  *
2427  * Returns:     Nothing
2428  *
2429  * Lock status: Host lock must be held
2430  *
2431  * Notes:       This only needs to be called if the reset is one which
2432  *		originates from an unknown location.  Resets originated
2433  *		by the mid-level itself don't need to call this, but there
2434  *		should be no harm.
2435  *
2436  *		The main purpose of this is to make sure that a CHECK_CONDITION
2437  *		is properly treated.
2438  */
scsi_report_device_reset(struct Scsi_Host * shost,int channel,int target)2439 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2440 {
2441 	struct scsi_device *sdev;
2442 
2443 	__shost_for_each_device(sdev, shost) {
2444 		if (channel == sdev_channel(sdev) &&
2445 		    target == sdev_id(sdev))
2446 			__scsi_report_device_reset(sdev, NULL);
2447 	}
2448 }
2449 EXPORT_SYMBOL(scsi_report_device_reset);
2450 
2451 /**
2452  * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2453  * @dev:	scsi_device to operate on
2454  * @arg:	reset type (see sg.h)
2455  */
2456 int
scsi_ioctl_reset(struct scsi_device * dev,int __user * arg)2457 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2458 {
2459 	struct scsi_cmnd *scmd;
2460 	struct Scsi_Host *shost = dev->host;
2461 	struct request *rq;
2462 	unsigned long flags;
2463 	int error = 0, val;
2464 	enum scsi_disposition rtn;
2465 
2466 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2467 		return -EACCES;
2468 
2469 	error = get_user(val, arg);
2470 	if (error)
2471 		return error;
2472 
2473 	if (scsi_autopm_get_host(shost) < 0)
2474 		return -EIO;
2475 
2476 	error = -EIO;
2477 	rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2478 			shost->hostt->cmd_size, GFP_KERNEL);
2479 	if (!rq)
2480 		goto out_put_autopm_host;
2481 	blk_rq_init(NULL, rq);
2482 
2483 	scmd = (struct scsi_cmnd *)(rq + 1);
2484 	scsi_init_command(dev, scmd);
2485 
2486 	scmd->submitter = SUBMITTED_BY_SCSI_RESET_IOCTL;
2487 	scmd->flags |= SCMD_LAST;
2488 	memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2489 
2490 	scmd->cmd_len			= 0;
2491 
2492 	scmd->sc_data_direction		= DMA_BIDIRECTIONAL;
2493 
2494 	spin_lock_irqsave(shost->host_lock, flags);
2495 	shost->tmf_in_progress = 1;
2496 	spin_unlock_irqrestore(shost->host_lock, flags);
2497 
2498 	switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2499 	case SG_SCSI_RESET_NOTHING:
2500 		rtn = SUCCESS;
2501 		break;
2502 	case SG_SCSI_RESET_DEVICE:
2503 		rtn = scsi_try_bus_device_reset(scmd);
2504 		if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2505 			break;
2506 		fallthrough;
2507 	case SG_SCSI_RESET_TARGET:
2508 		rtn = scsi_try_target_reset(scmd);
2509 		if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2510 			break;
2511 		fallthrough;
2512 	case SG_SCSI_RESET_BUS:
2513 		rtn = scsi_try_bus_reset(scmd);
2514 		if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2515 			break;
2516 		fallthrough;
2517 	case SG_SCSI_RESET_HOST:
2518 		rtn = scsi_try_host_reset(scmd);
2519 		if (rtn == SUCCESS)
2520 			break;
2521 		fallthrough;
2522 	default:
2523 		rtn = FAILED;
2524 		break;
2525 	}
2526 
2527 	error = (rtn == SUCCESS) ? 0 : -EIO;
2528 
2529 	spin_lock_irqsave(shost->host_lock, flags);
2530 	shost->tmf_in_progress = 0;
2531 	spin_unlock_irqrestore(shost->host_lock, flags);
2532 
2533 	/*
2534 	 * be sure to wake up anyone who was sleeping or had their queue
2535 	 * suspended while we performed the TMF.
2536 	 */
2537 	SCSI_LOG_ERROR_RECOVERY(3,
2538 		shost_printk(KERN_INFO, shost,
2539 			     "waking up host to restart after TMF\n"));
2540 
2541 	wake_up(&shost->host_wait);
2542 	scsi_run_host_queues(shost);
2543 
2544 	kfree(rq);
2545 
2546 out_put_autopm_host:
2547 	scsi_autopm_put_host(shost);
2548 	return error;
2549 }
2550 
scsi_command_normalize_sense(const struct scsi_cmnd * cmd,struct scsi_sense_hdr * sshdr)2551 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2552 				  struct scsi_sense_hdr *sshdr)
2553 {
2554 	return scsi_normalize_sense(cmd->sense_buffer,
2555 			SCSI_SENSE_BUFFERSIZE, sshdr);
2556 }
2557 EXPORT_SYMBOL(scsi_command_normalize_sense);
2558 
2559 /**
2560  * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2561  * @sense_buffer:	byte array of sense data
2562  * @sb_len:		number of valid bytes in sense_buffer
2563  * @info_out:		pointer to 64 integer where 8 or 4 byte information
2564  *			field will be placed if found.
2565  *
2566  * Return value:
2567  *	true if information field found, false if not found.
2568  */
scsi_get_sense_info_fld(const u8 * sense_buffer,int sb_len,u64 * info_out)2569 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2570 			     u64 *info_out)
2571 {
2572 	const u8 * ucp;
2573 
2574 	if (sb_len < 7)
2575 		return false;
2576 	switch (sense_buffer[0] & 0x7f) {
2577 	case 0x70:
2578 	case 0x71:
2579 		if (sense_buffer[0] & 0x80) {
2580 			*info_out = get_unaligned_be32(&sense_buffer[3]);
2581 			return true;
2582 		}
2583 		return false;
2584 	case 0x72:
2585 	case 0x73:
2586 		ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2587 					   0 /* info desc */);
2588 		if (ucp && (0xa == ucp[1])) {
2589 			*info_out = get_unaligned_be64(&ucp[4]);
2590 			return true;
2591 		}
2592 		return false;
2593 	default:
2594 		return false;
2595 	}
2596 }
2597 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2598