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