xref: /openbmc/linux/drivers/scsi/scsi_error.c (revision aceb4ab9)
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(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 
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 	atomic_inc(&scmd->device->iodone_cnt);
364 	if (scsi_abort_command(scmd) != SUCCESS) {
365 		set_host_byte(scmd, DID_TIME_OUT);
366 		scsi_eh_scmd_add(scmd);
367 	}
368 
369 	return BLK_EH_DONE;
370 }
371 
372 /**
373  * scsi_block_when_processing_errors - Prevent cmds from being queued.
374  * @sdev:	Device on which we are performing recovery.
375  *
376  * Description:
377  *     We block until the host is out of error recovery, and then check to
378  *     see whether the host or the device is offline.
379  *
380  * Return value:
381  *     0 when dev was taken offline by error recovery. 1 OK to proceed.
382  */
scsi_block_when_processing_errors(struct scsi_device * sdev)383 int scsi_block_when_processing_errors(struct scsi_device *sdev)
384 {
385 	int online;
386 
387 	wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
388 
389 	online = scsi_device_online(sdev);
390 
391 	return online;
392 }
393 EXPORT_SYMBOL(scsi_block_when_processing_errors);
394 
395 #ifdef CONFIG_SCSI_LOGGING
396 /**
397  * scsi_eh_prt_fail_stats - Log info on failures.
398  * @shost:	scsi host being recovered.
399  * @work_q:	Queue of scsi cmds to process.
400  */
scsi_eh_prt_fail_stats(struct Scsi_Host * shost,struct list_head * work_q)401 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
402 					  struct list_head *work_q)
403 {
404 	struct scsi_cmnd *scmd;
405 	struct scsi_device *sdev;
406 	int total_failures = 0;
407 	int cmd_failed = 0;
408 	int cmd_cancel = 0;
409 	int devices_failed = 0;
410 
411 	shost_for_each_device(sdev, shost) {
412 		list_for_each_entry(scmd, work_q, eh_entry) {
413 			if (scmd->device == sdev) {
414 				++total_failures;
415 				if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
416 					++cmd_cancel;
417 				else
418 					++cmd_failed;
419 			}
420 		}
421 
422 		if (cmd_cancel || cmd_failed) {
423 			SCSI_LOG_ERROR_RECOVERY(3,
424 				shost_printk(KERN_INFO, shost,
425 					    "%s: cmds failed: %d, cancel: %d\n",
426 					    __func__, cmd_failed,
427 					    cmd_cancel));
428 			cmd_cancel = 0;
429 			cmd_failed = 0;
430 			++devices_failed;
431 		}
432 	}
433 
434 	SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
435 				   "Total of %d commands on %d"
436 				   " devices require eh work\n",
437 				   total_failures, devices_failed));
438 }
439 #endif
440 
441  /**
442  * scsi_report_lun_change - Set flag on all *other* devices on the same target
443  *                          to indicate that a UNIT ATTENTION is expected.
444  * @sdev:	Device reporting the UNIT ATTENTION
445  */
scsi_report_lun_change(struct scsi_device * sdev)446 static void scsi_report_lun_change(struct scsi_device *sdev)
447 {
448 	sdev->sdev_target->expecting_lun_change = 1;
449 }
450 
451 /**
452  * scsi_report_sense - Examine scsi sense information and log messages for
453  *		       certain conditions, also issue uevents for some of them.
454  * @sdev:	Device reporting the sense code
455  * @sshdr:	sshdr to be examined
456  */
scsi_report_sense(struct scsi_device * sdev,struct scsi_sense_hdr * sshdr)457 static void scsi_report_sense(struct scsi_device *sdev,
458 			      struct scsi_sense_hdr *sshdr)
459 {
460 	enum scsi_device_event evt_type = SDEV_EVT_MAXBITS;	/* i.e. none */
461 
462 	if (sshdr->sense_key == UNIT_ATTENTION) {
463 		if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
464 			evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
465 			sdev_printk(KERN_WARNING, sdev,
466 				    "Inquiry data has changed");
467 		} else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
468 			evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
469 			scsi_report_lun_change(sdev);
470 			sdev_printk(KERN_WARNING, sdev,
471 				    "LUN assignments on this target have "
472 				    "changed. The Linux SCSI layer does not "
473 				    "automatically remap LUN assignments.\n");
474 		} else if (sshdr->asc == 0x3f)
475 			sdev_printk(KERN_WARNING, sdev,
476 				    "Operating parameters on this target have "
477 				    "changed. The Linux SCSI layer does not "
478 				    "automatically adjust these parameters.\n");
479 
480 		if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
481 			evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
482 			sdev_printk(KERN_WARNING, sdev,
483 				    "Warning! Received an indication that the "
484 				    "LUN reached a thin provisioning soft "
485 				    "threshold.\n");
486 		}
487 
488 		if (sshdr->asc == 0x29) {
489 			evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
490 			/*
491 			 * Do not print message if it is an expected side-effect
492 			 * of runtime PM.
493 			 */
494 			if (!sdev->silence_suspend)
495 				sdev_printk(KERN_WARNING, sdev,
496 					    "Power-on or device reset occurred\n");
497 		}
498 
499 		if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
500 			evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
501 			sdev_printk(KERN_WARNING, sdev,
502 				    "Mode parameters changed");
503 		} else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
504 			evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
505 			sdev_printk(KERN_WARNING, sdev,
506 				    "Asymmetric access state changed");
507 		} else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
508 			evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
509 			sdev_printk(KERN_WARNING, sdev,
510 				    "Capacity data has changed");
511 		} else if (sshdr->asc == 0x2a)
512 			sdev_printk(KERN_WARNING, sdev,
513 				    "Parameters changed");
514 	}
515 
516 	if (evt_type != SDEV_EVT_MAXBITS) {
517 		set_bit(evt_type, sdev->pending_events);
518 		schedule_work(&sdev->event_work);
519 	}
520 }
521 
set_scsi_ml_byte(struct scsi_cmnd * cmd,u8 status)522 static inline void set_scsi_ml_byte(struct scsi_cmnd *cmd, u8 status)
523 {
524 	cmd->result = (cmd->result & 0xffff00ff) | (status << 8);
525 }
526 
527 /**
528  * scsi_check_sense - Examine scsi cmd sense
529  * @scmd:	Cmd to have sense checked.
530  *
531  * Return value:
532  *	SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
533  *
534  * Notes:
535  *	When a deferred error is detected the current command has
536  *	not been executed and needs retrying.
537  */
scsi_check_sense(struct scsi_cmnd * scmd)538 enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd)
539 {
540 	struct request *req = scsi_cmd_to_rq(scmd);
541 	struct scsi_device *sdev = scmd->device;
542 	struct scsi_sense_hdr sshdr;
543 
544 	if (! scsi_command_normalize_sense(scmd, &sshdr))
545 		return FAILED;	/* no valid sense data */
546 
547 	scsi_report_sense(sdev, &sshdr);
548 
549 	if (scsi_sense_is_deferred(&sshdr))
550 		return NEEDS_RETRY;
551 
552 	if (sdev->handler && sdev->handler->check_sense) {
553 		enum scsi_disposition rc;
554 
555 		rc = sdev->handler->check_sense(sdev, &sshdr);
556 		if (rc != SCSI_RETURN_NOT_HANDLED)
557 			return rc;
558 		/* handler does not care. Drop down to default handling */
559 	}
560 
561 	if (scmd->cmnd[0] == TEST_UNIT_READY &&
562 	    scmd->submitter != SUBMITTED_BY_SCSI_ERROR_HANDLER)
563 		/*
564 		 * nasty: for mid-layer issued TURs, we need to return the
565 		 * actual sense data without any recovery attempt.  For eh
566 		 * issued ones, we need to try to recover and interpret
567 		 */
568 		return SUCCESS;
569 
570 	/*
571 	 * Previous logic looked for FILEMARK, EOM or ILI which are
572 	 * mainly associated with tapes and returned SUCCESS.
573 	 */
574 	if (sshdr.response_code == 0x70) {
575 		/* fixed format */
576 		if (scmd->sense_buffer[2] & 0xe0)
577 			return SUCCESS;
578 	} else {
579 		/*
580 		 * descriptor format: look for "stream commands sense data
581 		 * descriptor" (see SSC-3). Assume single sense data
582 		 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
583 		 */
584 		if ((sshdr.additional_length > 3) &&
585 		    (scmd->sense_buffer[8] == 0x4) &&
586 		    (scmd->sense_buffer[11] & 0xe0))
587 			return SUCCESS;
588 	}
589 
590 	switch (sshdr.sense_key) {
591 	case NO_SENSE:
592 		return SUCCESS;
593 	case RECOVERED_ERROR:
594 		return /* soft_error */ SUCCESS;
595 
596 	case ABORTED_COMMAND:
597 		if (sshdr.asc == 0x10) /* DIF */
598 			return SUCCESS;
599 
600 		/*
601 		 * Check aborts due to command duration limit policy:
602 		 * ABORTED COMMAND additional sense code with the
603 		 * COMMAND TIMEOUT BEFORE PROCESSING or
604 		 * COMMAND TIMEOUT DURING PROCESSING or
605 		 * COMMAND TIMEOUT DURING PROCESSING DUE TO ERROR RECOVERY
606 		 * additional sense code qualifiers.
607 		 */
608 		if (sshdr.asc == 0x2e &&
609 		    sshdr.ascq >= 0x01 && sshdr.ascq <= 0x03) {
610 			set_scsi_ml_byte(scmd, SCSIML_STAT_DL_TIMEOUT);
611 			req->cmd_flags |= REQ_FAILFAST_DEV;
612 			req->rq_flags |= RQF_QUIET;
613 			return SUCCESS;
614 		}
615 
616 		if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
617 			return ADD_TO_MLQUEUE;
618 		if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
619 		    sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
620 			return ADD_TO_MLQUEUE;
621 
622 		return NEEDS_RETRY;
623 	case NOT_READY:
624 	case UNIT_ATTENTION:
625 		/*
626 		 * if we are expecting a cc/ua because of a bus reset that we
627 		 * performed, treat this just as a retry.  otherwise this is
628 		 * information that we should pass up to the upper-level driver
629 		 * so that we can deal with it there.
630 		 */
631 		if (scmd->device->expecting_cc_ua) {
632 			/*
633 			 * Because some device does not queue unit
634 			 * attentions correctly, we carefully check
635 			 * additional sense code and qualifier so as
636 			 * not to squash media change unit attention.
637 			 */
638 			if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
639 				scmd->device->expecting_cc_ua = 0;
640 				return NEEDS_RETRY;
641 			}
642 		}
643 		/*
644 		 * we might also expect a cc/ua if another LUN on the target
645 		 * reported a UA with an ASC/ASCQ of 3F 0E -
646 		 * REPORTED LUNS DATA HAS CHANGED.
647 		 */
648 		if (scmd->device->sdev_target->expecting_lun_change &&
649 		    sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
650 			return NEEDS_RETRY;
651 		/*
652 		 * if the device is in the process of becoming ready, we
653 		 * should retry.
654 		 */
655 		if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
656 			return NEEDS_RETRY;
657 		/*
658 		 * if the device is not started, we need to wake
659 		 * the error handler to start the motor
660 		 */
661 		if (scmd->device->allow_restart &&
662 		    (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
663 			return FAILED;
664 		/*
665 		 * Pass the UA upwards for a determination in the completion
666 		 * functions.
667 		 */
668 		return SUCCESS;
669 
670 		/* these are not supported */
671 	case DATA_PROTECT:
672 		if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
673 			/* Thin provisioning hard threshold reached */
674 			set_scsi_ml_byte(scmd, SCSIML_STAT_NOSPC);
675 			return SUCCESS;
676 		}
677 		fallthrough;
678 	case COPY_ABORTED:
679 	case VOLUME_OVERFLOW:
680 	case MISCOMPARE:
681 	case BLANK_CHECK:
682 		set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE);
683 		return SUCCESS;
684 
685 	case MEDIUM_ERROR:
686 		if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
687 		    sshdr.asc == 0x13 || /* AMNF DATA FIELD */
688 		    sshdr.asc == 0x14) { /* RECORD NOT FOUND */
689 			set_scsi_ml_byte(scmd, SCSIML_STAT_MED_ERROR);
690 			return SUCCESS;
691 		}
692 		return NEEDS_RETRY;
693 
694 	case HARDWARE_ERROR:
695 		if (scmd->device->retry_hwerror)
696 			return ADD_TO_MLQUEUE;
697 		else
698 			set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE);
699 		fallthrough;
700 
701 	case ILLEGAL_REQUEST:
702 		if (sshdr.asc == 0x20 || /* Invalid command operation code */
703 		    sshdr.asc == 0x21 || /* Logical block address out of range */
704 		    sshdr.asc == 0x22 || /* Invalid function */
705 		    sshdr.asc == 0x24 || /* Invalid field in cdb */
706 		    sshdr.asc == 0x26 || /* Parameter value invalid */
707 		    sshdr.asc == 0x27) { /* Write protected */
708 			set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE);
709 		}
710 		return SUCCESS;
711 
712 	case COMPLETED:
713 		if (sshdr.asc == 0x55 && sshdr.ascq == 0x0a) {
714 			set_scsi_ml_byte(scmd, SCSIML_STAT_DL_TIMEOUT);
715 			req->cmd_flags |= REQ_FAILFAST_DEV;
716 			req->rq_flags |= RQF_QUIET;
717 		}
718 		return SUCCESS;
719 
720 	default:
721 		return SUCCESS;
722 	}
723 }
724 EXPORT_SYMBOL_GPL(scsi_check_sense);
725 
scsi_handle_queue_ramp_up(struct scsi_device * sdev)726 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
727 {
728 	const struct scsi_host_template *sht = sdev->host->hostt;
729 	struct scsi_device *tmp_sdev;
730 
731 	if (!sht->track_queue_depth ||
732 	    sdev->queue_depth >= sdev->max_queue_depth)
733 		return;
734 
735 	if (time_before(jiffies,
736 	    sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
737 		return;
738 
739 	if (time_before(jiffies,
740 	    sdev->last_queue_full_time + sdev->queue_ramp_up_period))
741 		return;
742 
743 	/*
744 	 * Walk all devices of a target and do
745 	 * ramp up on them.
746 	 */
747 	shost_for_each_device(tmp_sdev, sdev->host) {
748 		if (tmp_sdev->channel != sdev->channel ||
749 		    tmp_sdev->id != sdev->id ||
750 		    tmp_sdev->queue_depth == sdev->max_queue_depth)
751 			continue;
752 
753 		scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
754 		sdev->last_queue_ramp_up = jiffies;
755 	}
756 }
757 
scsi_handle_queue_full(struct scsi_device * sdev)758 static void scsi_handle_queue_full(struct scsi_device *sdev)
759 {
760 	const struct scsi_host_template *sht = sdev->host->hostt;
761 	struct scsi_device *tmp_sdev;
762 
763 	if (!sht->track_queue_depth)
764 		return;
765 
766 	shost_for_each_device(tmp_sdev, sdev->host) {
767 		if (tmp_sdev->channel != sdev->channel ||
768 		    tmp_sdev->id != sdev->id)
769 			continue;
770 		/*
771 		 * We do not know the number of commands that were at
772 		 * the device when we got the queue full so we start
773 		 * from the highest possible value and work our way down.
774 		 */
775 		scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
776 	}
777 }
778 
779 /**
780  * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
781  * @scmd:	SCSI cmd to examine.
782  *
783  * Notes:
784  *    This is *only* called when we are examining the status of commands
785  *    queued during error recovery.  the main difference here is that we
786  *    don't allow for the possibility of retries here, and we are a lot
787  *    more restrictive about what we consider acceptable.
788  */
scsi_eh_completed_normally(struct scsi_cmnd * scmd)789 static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd)
790 {
791 	/*
792 	 * first check the host byte, to see if there is anything in there
793 	 * that would indicate what we need to do.
794 	 */
795 	if (host_byte(scmd->result) == DID_RESET) {
796 		/*
797 		 * rats.  we are already in the error handler, so we now
798 		 * get to try and figure out what to do next.  if the sense
799 		 * is valid, we have a pretty good idea of what to do.
800 		 * if not, we mark it as FAILED.
801 		 */
802 		return scsi_check_sense(scmd);
803 	}
804 	if (host_byte(scmd->result) != DID_OK)
805 		return FAILED;
806 
807 	/*
808 	 * now, check the status byte to see if this indicates
809 	 * anything special.
810 	 */
811 	switch (get_status_byte(scmd)) {
812 	case SAM_STAT_GOOD:
813 		scsi_handle_queue_ramp_up(scmd->device);
814 		if (scmd->sense_buffer && SCSI_SENSE_VALID(scmd))
815 			/*
816 			 * If we have sense data, call scsi_check_sense() in
817 			 * order to set the correct SCSI ML byte (if any).
818 			 * No point in checking the return value, since the
819 			 * command has already completed successfully.
820 			 */
821 			scsi_check_sense(scmd);
822 		fallthrough;
823 	case SAM_STAT_COMMAND_TERMINATED:
824 		return SUCCESS;
825 	case SAM_STAT_CHECK_CONDITION:
826 		return scsi_check_sense(scmd);
827 	case SAM_STAT_CONDITION_MET:
828 	case SAM_STAT_INTERMEDIATE:
829 	case SAM_STAT_INTERMEDIATE_CONDITION_MET:
830 		/*
831 		 * who knows?  FIXME(eric)
832 		 */
833 		return SUCCESS;
834 	case SAM_STAT_RESERVATION_CONFLICT:
835 		if (scmd->cmnd[0] == TEST_UNIT_READY)
836 			/* it is a success, we probed the device and
837 			 * found it */
838 			return SUCCESS;
839 		/* otherwise, we failed to send the command */
840 		return FAILED;
841 	case SAM_STAT_TASK_SET_FULL:
842 		scsi_handle_queue_full(scmd->device);
843 		fallthrough;
844 	case SAM_STAT_BUSY:
845 		return NEEDS_RETRY;
846 	default:
847 		return FAILED;
848 	}
849 	return FAILED;
850 }
851 
852 /**
853  * scsi_eh_done - Completion function for error handling.
854  * @scmd:	Cmd that is done.
855  */
scsi_eh_done(struct scsi_cmnd * scmd)856 void scsi_eh_done(struct scsi_cmnd *scmd)
857 {
858 	struct completion *eh_action;
859 
860 	SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
861 			"%s result: %x\n", __func__, scmd->result));
862 
863 	eh_action = scmd->device->host->eh_action;
864 	if (eh_action)
865 		complete(eh_action);
866 }
867 
868 /**
869  * scsi_try_host_reset - ask host adapter to reset itself
870  * @scmd:	SCSI cmd to send host reset.
871  */
scsi_try_host_reset(struct scsi_cmnd * scmd)872 static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
873 {
874 	unsigned long flags;
875 	enum scsi_disposition rtn;
876 	struct Scsi_Host *host = scmd->device->host;
877 	const struct scsi_host_template *hostt = host->hostt;
878 
879 	SCSI_LOG_ERROR_RECOVERY(3,
880 		shost_printk(KERN_INFO, host, "Snd Host RST\n"));
881 
882 	if (!hostt->eh_host_reset_handler)
883 		return FAILED;
884 
885 	rtn = hostt->eh_host_reset_handler(scmd);
886 
887 	if (rtn == SUCCESS) {
888 		if (!hostt->skip_settle_delay)
889 			ssleep(HOST_RESET_SETTLE_TIME);
890 		spin_lock_irqsave(host->host_lock, flags);
891 		scsi_report_bus_reset(host, scmd_channel(scmd));
892 		spin_unlock_irqrestore(host->host_lock, flags);
893 	}
894 
895 	return rtn;
896 }
897 
898 /**
899  * scsi_try_bus_reset - ask host to perform a bus reset
900  * @scmd:	SCSI cmd to send bus reset.
901  */
scsi_try_bus_reset(struct scsi_cmnd * scmd)902 static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
903 {
904 	unsigned long flags;
905 	enum scsi_disposition rtn;
906 	struct Scsi_Host *host = scmd->device->host;
907 	const struct scsi_host_template *hostt = host->hostt;
908 
909 	SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
910 		"%s: Snd Bus RST\n", __func__));
911 
912 	if (!hostt->eh_bus_reset_handler)
913 		return FAILED;
914 
915 	rtn = hostt->eh_bus_reset_handler(scmd);
916 
917 	if (rtn == SUCCESS) {
918 		if (!hostt->skip_settle_delay)
919 			ssleep(BUS_RESET_SETTLE_TIME);
920 		spin_lock_irqsave(host->host_lock, flags);
921 		scsi_report_bus_reset(host, scmd_channel(scmd));
922 		spin_unlock_irqrestore(host->host_lock, flags);
923 	}
924 
925 	return rtn;
926 }
927 
__scsi_report_device_reset(struct scsi_device * sdev,void * data)928 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
929 {
930 	sdev->was_reset = 1;
931 	sdev->expecting_cc_ua = 1;
932 }
933 
934 /**
935  * scsi_try_target_reset - Ask host to perform a target reset
936  * @scmd:	SCSI cmd used to send a target reset
937  *
938  * Notes:
939  *    There is no timeout for this operation.  if this operation is
940  *    unreliable for a given host, then the host itself needs to put a
941  *    timer on it, and set the host back to a consistent state prior to
942  *    returning.
943  */
scsi_try_target_reset(struct scsi_cmnd * scmd)944 static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd)
945 {
946 	unsigned long flags;
947 	enum scsi_disposition rtn;
948 	struct Scsi_Host *host = scmd->device->host;
949 	const struct scsi_host_template *hostt = host->hostt;
950 
951 	if (!hostt->eh_target_reset_handler)
952 		return FAILED;
953 
954 	rtn = hostt->eh_target_reset_handler(scmd);
955 	if (rtn == SUCCESS) {
956 		spin_lock_irqsave(host->host_lock, flags);
957 		__starget_for_each_device(scsi_target(scmd->device), NULL,
958 					  __scsi_report_device_reset);
959 		spin_unlock_irqrestore(host->host_lock, flags);
960 	}
961 
962 	return rtn;
963 }
964 
965 /**
966  * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
967  * @scmd:	SCSI cmd used to send BDR
968  *
969  * Notes:
970  *    There is no timeout for this operation.  if this operation is
971  *    unreliable for a given host, then the host itself needs to put a
972  *    timer on it, and set the host back to a consistent state prior to
973  *    returning.
974  */
scsi_try_bus_device_reset(struct scsi_cmnd * scmd)975 static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
976 {
977 	enum scsi_disposition rtn;
978 	const struct scsi_host_template *hostt = scmd->device->host->hostt;
979 
980 	if (!hostt->eh_device_reset_handler)
981 		return FAILED;
982 
983 	rtn = hostt->eh_device_reset_handler(scmd);
984 	if (rtn == SUCCESS)
985 		__scsi_report_device_reset(scmd->device, NULL);
986 	return rtn;
987 }
988 
989 /**
990  * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
991  * @hostt:	SCSI driver host template
992  * @scmd:	SCSI cmd used to send a target reset
993  *
994  * Return value:
995  *	SUCCESS, FAILED, or FAST_IO_FAIL
996  *
997  * Notes:
998  *    SUCCESS does not necessarily indicate that the command
999  *    has been aborted; it only indicates that the LLDDs
1000  *    has cleared all references to that command.
1001  *    LLDDs should return FAILED only if an abort was required
1002  *    but could not be executed. LLDDs should return FAST_IO_FAIL
1003  *    if the device is temporarily unavailable (eg due to a
1004  *    link down on FibreChannel)
1005  */
1006 static enum scsi_disposition
scsi_try_to_abort_cmd(const struct scsi_host_template * hostt,struct scsi_cmnd * scmd)1007 scsi_try_to_abort_cmd(const struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
1008 {
1009 	if (!hostt->eh_abort_handler)
1010 		return FAILED;
1011 
1012 	return hostt->eh_abort_handler(scmd);
1013 }
1014 
scsi_abort_eh_cmnd(struct scsi_cmnd * scmd)1015 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
1016 {
1017 	if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
1018 		if (scsi_try_bus_device_reset(scmd) != SUCCESS)
1019 			if (scsi_try_target_reset(scmd) != SUCCESS)
1020 				if (scsi_try_bus_reset(scmd) != SUCCESS)
1021 					scsi_try_host_reset(scmd);
1022 }
1023 
1024 /**
1025  * scsi_eh_prep_cmnd  - Save a scsi command info as part of error recovery
1026  * @scmd:       SCSI command structure to hijack
1027  * @ses:        structure to save restore information
1028  * @cmnd:       CDB to send. Can be NULL if no new cmnd is needed
1029  * @cmnd_size:  size in bytes of @cmnd (must be <= MAX_COMMAND_SIZE)
1030  * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
1031  *
1032  * This function is used to save a scsi command information before re-execution
1033  * as part of the error recovery process.  If @sense_bytes is 0 the command
1034  * sent must be one that does not transfer any data.  If @sense_bytes != 0
1035  * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
1036  * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
1037  */
scsi_eh_prep_cmnd(struct scsi_cmnd * scmd,struct scsi_eh_save * ses,unsigned char * cmnd,int cmnd_size,unsigned sense_bytes)1038 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
1039 			unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
1040 {
1041 	struct scsi_device *sdev = scmd->device;
1042 
1043 	/*
1044 	 * We need saved copies of a number of fields - this is because
1045 	 * error handling may need to overwrite these with different values
1046 	 * to run different commands, and once error handling is complete,
1047 	 * we will need to restore these values prior to running the actual
1048 	 * command.
1049 	 */
1050 	ses->cmd_len = scmd->cmd_len;
1051 	ses->data_direction = scmd->sc_data_direction;
1052 	ses->sdb = scmd->sdb;
1053 	ses->result = scmd->result;
1054 	ses->resid_len = scmd->resid_len;
1055 	ses->underflow = scmd->underflow;
1056 	ses->prot_op = scmd->prot_op;
1057 	ses->eh_eflags = scmd->eh_eflags;
1058 
1059 	scmd->prot_op = SCSI_PROT_NORMAL;
1060 	scmd->eh_eflags = 0;
1061 	memcpy(ses->cmnd, scmd->cmnd, sizeof(ses->cmnd));
1062 	memset(scmd->cmnd, 0, sizeof(scmd->cmnd));
1063 	memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1064 	scmd->result = 0;
1065 	scmd->resid_len = 0;
1066 
1067 	if (sense_bytes) {
1068 		scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
1069 					 sense_bytes);
1070 		sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
1071 			    scmd->sdb.length);
1072 		scmd->sdb.table.sgl = &ses->sense_sgl;
1073 		scmd->sc_data_direction = DMA_FROM_DEVICE;
1074 		scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
1075 		scmd->cmnd[0] = REQUEST_SENSE;
1076 		scmd->cmnd[4] = scmd->sdb.length;
1077 		scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1078 	} else {
1079 		scmd->sc_data_direction = DMA_NONE;
1080 		if (cmnd) {
1081 			BUG_ON(cmnd_size > sizeof(scmd->cmnd));
1082 			memcpy(scmd->cmnd, cmnd, cmnd_size);
1083 			scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1084 		}
1085 	}
1086 
1087 	scmd->underflow = 0;
1088 
1089 	if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1090 		scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
1091 			(sdev->lun << 5 & 0xe0);
1092 
1093 	/*
1094 	 * Zero the sense buffer.  The scsi spec mandates that any
1095 	 * untransferred sense data should be interpreted as being zero.
1096 	 */
1097 	memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1098 }
1099 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1100 
1101 /**
1102  * scsi_eh_restore_cmnd  - Restore a scsi command info as part of error recovery
1103  * @scmd:       SCSI command structure to restore
1104  * @ses:        saved information from a coresponding call to scsi_eh_prep_cmnd
1105  *
1106  * Undo any damage done by above scsi_eh_prep_cmnd().
1107  */
scsi_eh_restore_cmnd(struct scsi_cmnd * scmd,struct scsi_eh_save * ses)1108 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1109 {
1110 	/*
1111 	 * Restore original data
1112 	 */
1113 	scmd->cmd_len = ses->cmd_len;
1114 	memcpy(scmd->cmnd, ses->cmnd, sizeof(ses->cmnd));
1115 	scmd->sc_data_direction = ses->data_direction;
1116 	scmd->sdb = ses->sdb;
1117 	scmd->result = ses->result;
1118 	scmd->resid_len = ses->resid_len;
1119 	scmd->underflow = ses->underflow;
1120 	scmd->prot_op = ses->prot_op;
1121 	scmd->eh_eflags = ses->eh_eflags;
1122 }
1123 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1124 
1125 /**
1126  * scsi_send_eh_cmnd  - submit a scsi command as part of error recovery
1127  * @scmd:       SCSI command structure to hijack
1128  * @cmnd:       CDB to send
1129  * @cmnd_size:  size in bytes of @cmnd
1130  * @timeout:    timeout for this request
1131  * @sense_bytes: size of sense data to copy or 0
1132  *
1133  * This function is used to send a scsi command down to a target device
1134  * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1135  *
1136  * Return value:
1137  *    SUCCESS or FAILED or NEEDS_RETRY
1138  */
scsi_send_eh_cmnd(struct scsi_cmnd * scmd,unsigned char * cmnd,int cmnd_size,int timeout,unsigned sense_bytes)1139 static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
1140 	unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes)
1141 {
1142 	struct scsi_device *sdev = scmd->device;
1143 	struct Scsi_Host *shost = sdev->host;
1144 	DECLARE_COMPLETION_ONSTACK(done);
1145 	unsigned long timeleft = timeout, delay;
1146 	struct scsi_eh_save ses;
1147 	const unsigned long stall_for = msecs_to_jiffies(100);
1148 	int rtn;
1149 
1150 retry:
1151 	scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1152 	shost->eh_action = &done;
1153 
1154 	scsi_log_send(scmd);
1155 	scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER;
1156 	scmd->flags |= SCMD_LAST;
1157 
1158 	/*
1159 	 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1160 	 * change the SCSI device state after we have examined it and before
1161 	 * .queuecommand() is called.
1162 	 */
1163 	mutex_lock(&sdev->state_mutex);
1164 	while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1165 		mutex_unlock(&sdev->state_mutex);
1166 		SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1167 			"%s: state %d <> %d\n", __func__, sdev->sdev_state,
1168 			SDEV_BLOCK));
1169 		delay = min(timeleft, stall_for);
1170 		timeleft -= delay;
1171 		msleep(jiffies_to_msecs(delay));
1172 		mutex_lock(&sdev->state_mutex);
1173 	}
1174 	if (sdev->sdev_state != SDEV_BLOCK)
1175 		rtn = shost->hostt->queuecommand(shost, scmd);
1176 	else
1177 		rtn = FAILED;
1178 	mutex_unlock(&sdev->state_mutex);
1179 
1180 	if (rtn) {
1181 		if (timeleft > stall_for) {
1182 			scsi_eh_restore_cmnd(scmd, &ses);
1183 
1184 			timeleft -= stall_for;
1185 			msleep(jiffies_to_msecs(stall_for));
1186 			goto retry;
1187 		}
1188 		/* signal not to enter either branch of the if () below */
1189 		timeleft = 0;
1190 		rtn = FAILED;
1191 	} else {
1192 		timeleft = wait_for_completion_timeout(&done, timeout);
1193 		rtn = SUCCESS;
1194 	}
1195 
1196 	shost->eh_action = NULL;
1197 
1198 	scsi_log_completion(scmd, rtn);
1199 
1200 	SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1201 			"%s timeleft: %ld\n",
1202 			__func__, timeleft));
1203 
1204 	/*
1205 	 * If there is time left scsi_eh_done got called, and we will examine
1206 	 * the actual status codes to see whether the command actually did
1207 	 * complete normally, else if we have a zero return and no time left,
1208 	 * the command must still be pending, so abort it and return FAILED.
1209 	 * If we never actually managed to issue the command, because
1210 	 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1211 	 * value above (so don't execute either branch of the if)
1212 	 */
1213 	if (timeleft) {
1214 		rtn = scsi_eh_completed_normally(scmd);
1215 		SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1216 			"%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1217 
1218 		switch (rtn) {
1219 		case SUCCESS:
1220 		case NEEDS_RETRY:
1221 		case FAILED:
1222 			break;
1223 		case ADD_TO_MLQUEUE:
1224 			rtn = NEEDS_RETRY;
1225 			break;
1226 		default:
1227 			rtn = FAILED;
1228 			break;
1229 		}
1230 	} else if (rtn != FAILED) {
1231 		scsi_abort_eh_cmnd(scmd);
1232 		rtn = FAILED;
1233 	}
1234 
1235 	scsi_eh_restore_cmnd(scmd, &ses);
1236 
1237 	return rtn;
1238 }
1239 
1240 /**
1241  * scsi_request_sense - Request sense data from a particular target.
1242  * @scmd:	SCSI cmd for request sense.
1243  *
1244  * Notes:
1245  *    Some hosts automatically obtain this information, others require
1246  *    that we obtain it on our own. This function will *not* return until
1247  *    the command either times out, or it completes.
1248  */
scsi_request_sense(struct scsi_cmnd * scmd)1249 static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd)
1250 {
1251 	return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1252 }
1253 
1254 static enum scsi_disposition
scsi_eh_action(struct scsi_cmnd * scmd,enum scsi_disposition rtn)1255 scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn)
1256 {
1257 	if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
1258 		struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1259 		if (sdrv->eh_action)
1260 			rtn = sdrv->eh_action(scmd, rtn);
1261 	}
1262 	return rtn;
1263 }
1264 
1265 /**
1266  * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1267  * @scmd:	Original SCSI cmd that eh has finished.
1268  * @done_q:	Queue for processed commands.
1269  *
1270  * Notes:
1271  *    We don't want to use the normal command completion while we are are
1272  *    still handling errors - it may cause other commands to be queued,
1273  *    and that would disturb what we are doing.  Thus we really want to
1274  *    keep a list of pending commands for final completion, and once we
1275  *    are ready to leave error handling we handle completion for real.
1276  */
scsi_eh_finish_cmd(struct scsi_cmnd * scmd,struct list_head * done_q)1277 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1278 {
1279 	list_move_tail(&scmd->eh_entry, done_q);
1280 }
1281 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1282 
1283 /**
1284  * scsi_eh_get_sense - Get device sense data.
1285  * @work_q:	Queue of commands to process.
1286  * @done_q:	Queue of processed commands.
1287  *
1288  * Description:
1289  *    See if we need to request sense information.  if so, then get it
1290  *    now, so we have a better idea of what to do.
1291  *
1292  * Notes:
1293  *    This has the unfortunate side effect that if a shost adapter does
1294  *    not automatically request sense information, we end up shutting
1295  *    it down before we request it.
1296  *
1297  *    All drivers should request sense information internally these days,
1298  *    so for now all I have to say is tough noogies if you end up in here.
1299  *
1300  *    XXX: Long term this code should go away, but that needs an audit of
1301  *         all LLDDs first.
1302  */
scsi_eh_get_sense(struct list_head * work_q,struct list_head * done_q)1303 int scsi_eh_get_sense(struct list_head *work_q,
1304 		      struct list_head *done_q)
1305 {
1306 	struct scsi_cmnd *scmd, *next;
1307 	struct Scsi_Host *shost;
1308 	enum scsi_disposition rtn;
1309 
1310 	/*
1311 	 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1312 	 * should not get sense.
1313 	 */
1314 	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1315 		if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1316 		    SCSI_SENSE_VALID(scmd))
1317 			continue;
1318 
1319 		shost = scmd->device->host;
1320 		if (scsi_host_eh_past_deadline(shost)) {
1321 			SCSI_LOG_ERROR_RECOVERY(3,
1322 				scmd_printk(KERN_INFO, scmd,
1323 					    "%s: skip request sense, past eh deadline\n",
1324 					     current->comm));
1325 			break;
1326 		}
1327 		if (!scsi_status_is_check_condition(scmd->result))
1328 			/*
1329 			 * don't request sense if there's no check condition
1330 			 * status because the error we're processing isn't one
1331 			 * that has a sense code (and some devices get
1332 			 * confused by sense requests out of the blue)
1333 			 */
1334 			continue;
1335 
1336 		SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1337 						  "%s: requesting sense\n",
1338 						  current->comm));
1339 		rtn = scsi_request_sense(scmd);
1340 		if (rtn != SUCCESS)
1341 			continue;
1342 
1343 		SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1344 			"sense requested, result %x\n", scmd->result));
1345 		SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1346 
1347 		rtn = scsi_decide_disposition(scmd);
1348 
1349 		/*
1350 		 * if the result was normal, then just pass it along to the
1351 		 * upper level.
1352 		 */
1353 		if (rtn == SUCCESS)
1354 			/*
1355 			 * We don't want this command reissued, just finished
1356 			 * with the sense data, so set retries to the max
1357 			 * allowed to ensure it won't get reissued. If the user
1358 			 * has requested infinite retries, we also want to
1359 			 * finish this command, so force completion by setting
1360 			 * retries and allowed to the same value.
1361 			 */
1362 			if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
1363 				scmd->retries = scmd->allowed = 1;
1364 			else
1365 				scmd->retries = scmd->allowed;
1366 		else if (rtn != NEEDS_RETRY)
1367 			continue;
1368 
1369 		scsi_eh_finish_cmd(scmd, done_q);
1370 	}
1371 
1372 	return list_empty(work_q);
1373 }
1374 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1375 
1376 /**
1377  * scsi_eh_tur - Send TUR to device.
1378  * @scmd:	&scsi_cmnd to send TUR
1379  *
1380  * Return value:
1381  *    0 - Device is ready. 1 - Device NOT ready.
1382  */
scsi_eh_tur(struct scsi_cmnd * scmd)1383 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1384 {
1385 	static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1386 	int retry_cnt = 1;
1387 	enum scsi_disposition rtn;
1388 
1389 retry_tur:
1390 	rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1391 				scmd->device->eh_timeout, 0);
1392 
1393 	SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1394 		"%s return: %x\n", __func__, rtn));
1395 
1396 	switch (rtn) {
1397 	case NEEDS_RETRY:
1398 		if (retry_cnt--)
1399 			goto retry_tur;
1400 		fallthrough;
1401 	case SUCCESS:
1402 		return 0;
1403 	default:
1404 		return 1;
1405 	}
1406 }
1407 
1408 /**
1409  * scsi_eh_test_devices - check if devices are responding from error recovery.
1410  * @cmd_list:	scsi commands in error recovery.
1411  * @work_q:	queue for commands which still need more error recovery
1412  * @done_q:	queue for commands which are finished
1413  * @try_stu:	boolean on if a STU command should be tried in addition to TUR.
1414  *
1415  * Decription:
1416  *    Tests if devices are in a working state.  Commands to devices now in
1417  *    a working state are sent to the done_q while commands to devices which
1418  *    are still failing to respond are returned to the work_q for more
1419  *    processing.
1420  **/
scsi_eh_test_devices(struct list_head * cmd_list,struct list_head * work_q,struct list_head * done_q,int try_stu)1421 static int scsi_eh_test_devices(struct list_head *cmd_list,
1422 				struct list_head *work_q,
1423 				struct list_head *done_q, int try_stu)
1424 {
1425 	struct scsi_cmnd *scmd, *next;
1426 	struct scsi_device *sdev;
1427 	int finish_cmds;
1428 
1429 	while (!list_empty(cmd_list)) {
1430 		scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1431 		sdev = scmd->device;
1432 
1433 		if (!try_stu) {
1434 			if (scsi_host_eh_past_deadline(sdev->host)) {
1435 				/* Push items back onto work_q */
1436 				list_splice_init(cmd_list, work_q);
1437 				SCSI_LOG_ERROR_RECOVERY(3,
1438 					sdev_printk(KERN_INFO, sdev,
1439 						    "%s: skip test device, past eh deadline",
1440 						    current->comm));
1441 				break;
1442 			}
1443 		}
1444 
1445 		finish_cmds = !scsi_device_online(scmd->device) ||
1446 			(try_stu && !scsi_eh_try_stu(scmd) &&
1447 			 !scsi_eh_tur(scmd)) ||
1448 			!scsi_eh_tur(scmd);
1449 
1450 		list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1451 			if (scmd->device == sdev) {
1452 				if (finish_cmds &&
1453 				    (try_stu ||
1454 				     scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1455 					scsi_eh_finish_cmd(scmd, done_q);
1456 				else
1457 					list_move_tail(&scmd->eh_entry, work_q);
1458 			}
1459 	}
1460 	return list_empty(work_q);
1461 }
1462 
1463 /**
1464  * scsi_eh_try_stu - Send START_UNIT to device.
1465  * @scmd:	&scsi_cmnd to send START_UNIT
1466  *
1467  * Return value:
1468  *    0 - Device is ready. 1 - Device NOT ready.
1469  */
scsi_eh_try_stu(struct scsi_cmnd * scmd)1470 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1471 {
1472 	static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1473 
1474 	if (scmd->device->allow_restart) {
1475 		int i;
1476 		enum scsi_disposition rtn = NEEDS_RETRY;
1477 
1478 		for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1479 			rtn = scsi_send_eh_cmnd(scmd, stu_command, 6,
1480 						scmd->device->eh_timeout, 0);
1481 
1482 		if (rtn == SUCCESS)
1483 			return 0;
1484 	}
1485 
1486 	return 1;
1487 }
1488 
1489  /**
1490  * scsi_eh_stu - send START_UNIT if needed
1491  * @shost:	&scsi host being recovered.
1492  * @work_q:	&list_head for pending commands.
1493  * @done_q:	&list_head for processed commands.
1494  *
1495  * Notes:
1496  *    If commands are failing due to not ready, initializing command required,
1497  *	try revalidating the device, which will end up sending a start unit.
1498  */
scsi_eh_stu(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1499 static int scsi_eh_stu(struct Scsi_Host *shost,
1500 			      struct list_head *work_q,
1501 			      struct list_head *done_q)
1502 {
1503 	struct scsi_cmnd *scmd, *stu_scmd, *next;
1504 	struct scsi_device *sdev;
1505 
1506 	shost_for_each_device(sdev, shost) {
1507 		if (scsi_host_eh_past_deadline(shost)) {
1508 			SCSI_LOG_ERROR_RECOVERY(3,
1509 				sdev_printk(KERN_INFO, sdev,
1510 					    "%s: skip START_UNIT, past eh deadline\n",
1511 					    current->comm));
1512 			scsi_device_put(sdev);
1513 			break;
1514 		}
1515 		stu_scmd = NULL;
1516 		list_for_each_entry(scmd, work_q, eh_entry)
1517 			if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1518 			    scsi_check_sense(scmd) == FAILED ) {
1519 				stu_scmd = scmd;
1520 				break;
1521 			}
1522 
1523 		if (!stu_scmd)
1524 			continue;
1525 
1526 		SCSI_LOG_ERROR_RECOVERY(3,
1527 			sdev_printk(KERN_INFO, sdev,
1528 				     "%s: Sending START_UNIT\n",
1529 				    current->comm));
1530 
1531 		if (!scsi_eh_try_stu(stu_scmd)) {
1532 			if (!scsi_device_online(sdev) ||
1533 			    !scsi_eh_tur(stu_scmd)) {
1534 				list_for_each_entry_safe(scmd, next,
1535 							  work_q, eh_entry) {
1536 					if (scmd->device == sdev &&
1537 					    scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1538 						scsi_eh_finish_cmd(scmd, done_q);
1539 				}
1540 			}
1541 		} else {
1542 			SCSI_LOG_ERROR_RECOVERY(3,
1543 				sdev_printk(KERN_INFO, sdev,
1544 					    "%s: START_UNIT failed\n",
1545 					    current->comm));
1546 		}
1547 	}
1548 
1549 	return list_empty(work_q);
1550 }
1551 
1552 
1553 /**
1554  * scsi_eh_bus_device_reset - send bdr if needed
1555  * @shost:	scsi host being recovered.
1556  * @work_q:	&list_head for pending commands.
1557  * @done_q:	&list_head for processed commands.
1558  *
1559  * Notes:
1560  *    Try a bus device reset.  Still, look to see whether we have multiple
1561  *    devices that are jammed or not - if we have multiple devices, it
1562  *    makes no sense to try bus_device_reset - we really would need to try
1563  *    a bus_reset instead.
1564  */
scsi_eh_bus_device_reset(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1565 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1566 				    struct list_head *work_q,
1567 				    struct list_head *done_q)
1568 {
1569 	struct scsi_cmnd *scmd, *bdr_scmd, *next;
1570 	struct scsi_device *sdev;
1571 	enum scsi_disposition rtn;
1572 
1573 	shost_for_each_device(sdev, shost) {
1574 		if (scsi_host_eh_past_deadline(shost)) {
1575 			SCSI_LOG_ERROR_RECOVERY(3,
1576 				sdev_printk(KERN_INFO, sdev,
1577 					    "%s: skip BDR, past eh deadline\n",
1578 					     current->comm));
1579 			scsi_device_put(sdev);
1580 			break;
1581 		}
1582 		bdr_scmd = NULL;
1583 		list_for_each_entry(scmd, work_q, eh_entry)
1584 			if (scmd->device == sdev) {
1585 				bdr_scmd = scmd;
1586 				break;
1587 			}
1588 
1589 		if (!bdr_scmd)
1590 			continue;
1591 
1592 		SCSI_LOG_ERROR_RECOVERY(3,
1593 			sdev_printk(KERN_INFO, sdev,
1594 				     "%s: Sending BDR\n", current->comm));
1595 		rtn = scsi_try_bus_device_reset(bdr_scmd);
1596 		if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1597 			if (!scsi_device_online(sdev) ||
1598 			    rtn == FAST_IO_FAIL ||
1599 			    !scsi_eh_tur(bdr_scmd)) {
1600 				list_for_each_entry_safe(scmd, next,
1601 							 work_q, eh_entry) {
1602 					if (scmd->device == sdev &&
1603 					    scsi_eh_action(scmd, rtn) != FAILED)
1604 						scsi_eh_finish_cmd(scmd,
1605 								   done_q);
1606 				}
1607 			}
1608 		} else {
1609 			SCSI_LOG_ERROR_RECOVERY(3,
1610 				sdev_printk(KERN_INFO, sdev,
1611 					    "%s: BDR failed\n", current->comm));
1612 		}
1613 	}
1614 
1615 	return list_empty(work_q);
1616 }
1617 
1618 /**
1619  * scsi_eh_target_reset - send target reset if needed
1620  * @shost:	scsi host being recovered.
1621  * @work_q:	&list_head for pending commands.
1622  * @done_q:	&list_head for processed commands.
1623  *
1624  * Notes:
1625  *    Try a target reset.
1626  */
scsi_eh_target_reset(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1627 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1628 				struct list_head *work_q,
1629 				struct list_head *done_q)
1630 {
1631 	LIST_HEAD(tmp_list);
1632 	LIST_HEAD(check_list);
1633 
1634 	list_splice_init(work_q, &tmp_list);
1635 
1636 	while (!list_empty(&tmp_list)) {
1637 		struct scsi_cmnd *next, *scmd;
1638 		enum scsi_disposition rtn;
1639 		unsigned int id;
1640 
1641 		if (scsi_host_eh_past_deadline(shost)) {
1642 			/* push back on work queue for further processing */
1643 			list_splice_init(&check_list, work_q);
1644 			list_splice_init(&tmp_list, work_q);
1645 			SCSI_LOG_ERROR_RECOVERY(3,
1646 				shost_printk(KERN_INFO, shost,
1647 					    "%s: Skip target reset, past eh deadline\n",
1648 					     current->comm));
1649 			return list_empty(work_q);
1650 		}
1651 
1652 		scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1653 		id = scmd_id(scmd);
1654 
1655 		SCSI_LOG_ERROR_RECOVERY(3,
1656 			shost_printk(KERN_INFO, shost,
1657 				     "%s: Sending target reset to target %d\n",
1658 				     current->comm, id));
1659 		rtn = scsi_try_target_reset(scmd);
1660 		if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1661 			SCSI_LOG_ERROR_RECOVERY(3,
1662 				shost_printk(KERN_INFO, shost,
1663 					     "%s: Target reset failed"
1664 					     " target: %d\n",
1665 					     current->comm, id));
1666 		list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1667 			if (scmd_id(scmd) != id)
1668 				continue;
1669 
1670 			if (rtn == SUCCESS)
1671 				list_move_tail(&scmd->eh_entry, &check_list);
1672 			else if (rtn == FAST_IO_FAIL)
1673 				scsi_eh_finish_cmd(scmd, done_q);
1674 			else
1675 				/* push back on work queue for further processing */
1676 				list_move(&scmd->eh_entry, work_q);
1677 		}
1678 	}
1679 
1680 	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1681 }
1682 
1683 /**
1684  * scsi_eh_bus_reset - send a bus reset
1685  * @shost:	&scsi host being recovered.
1686  * @work_q:	&list_head for pending commands.
1687  * @done_q:	&list_head for processed commands.
1688  */
scsi_eh_bus_reset(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1689 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1690 			     struct list_head *work_q,
1691 			     struct list_head *done_q)
1692 {
1693 	struct scsi_cmnd *scmd, *chan_scmd, *next;
1694 	LIST_HEAD(check_list);
1695 	unsigned int channel;
1696 	enum scsi_disposition rtn;
1697 
1698 	/*
1699 	 * we really want to loop over the various channels, and do this on
1700 	 * a channel by channel basis.  we should also check to see if any
1701 	 * of the failed commands are on soft_reset devices, and if so, skip
1702 	 * the reset.
1703 	 */
1704 
1705 	for (channel = 0; channel <= shost->max_channel; channel++) {
1706 		if (scsi_host_eh_past_deadline(shost)) {
1707 			list_splice_init(&check_list, work_q);
1708 			SCSI_LOG_ERROR_RECOVERY(3,
1709 				shost_printk(KERN_INFO, shost,
1710 					    "%s: skip BRST, past eh deadline\n",
1711 					     current->comm));
1712 			return list_empty(work_q);
1713 		}
1714 
1715 		chan_scmd = NULL;
1716 		list_for_each_entry(scmd, work_q, eh_entry) {
1717 			if (channel == scmd_channel(scmd)) {
1718 				chan_scmd = scmd;
1719 				break;
1720 				/*
1721 				 * FIXME add back in some support for
1722 				 * soft_reset devices.
1723 				 */
1724 			}
1725 		}
1726 
1727 		if (!chan_scmd)
1728 			continue;
1729 		SCSI_LOG_ERROR_RECOVERY(3,
1730 			shost_printk(KERN_INFO, shost,
1731 				     "%s: Sending BRST chan: %d\n",
1732 				     current->comm, channel));
1733 		rtn = scsi_try_bus_reset(chan_scmd);
1734 		if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1735 			list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1736 				if (channel == scmd_channel(scmd)) {
1737 					if (rtn == FAST_IO_FAIL)
1738 						scsi_eh_finish_cmd(scmd,
1739 								   done_q);
1740 					else
1741 						list_move_tail(&scmd->eh_entry,
1742 							       &check_list);
1743 				}
1744 			}
1745 		} else {
1746 			SCSI_LOG_ERROR_RECOVERY(3,
1747 				shost_printk(KERN_INFO, shost,
1748 					     "%s: BRST failed chan: %d\n",
1749 					     current->comm, channel));
1750 		}
1751 	}
1752 	return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1753 }
1754 
1755 /**
1756  * scsi_eh_host_reset - send a host reset
1757  * @shost:	host to be reset.
1758  * @work_q:	&list_head for pending commands.
1759  * @done_q:	&list_head for processed commands.
1760  */
scsi_eh_host_reset(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)1761 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1762 			      struct list_head *work_q,
1763 			      struct list_head *done_q)
1764 {
1765 	struct scsi_cmnd *scmd, *next;
1766 	LIST_HEAD(check_list);
1767 	enum scsi_disposition rtn;
1768 
1769 	if (!list_empty(work_q)) {
1770 		scmd = list_entry(work_q->next,
1771 				  struct scsi_cmnd, eh_entry);
1772 
1773 		SCSI_LOG_ERROR_RECOVERY(3,
1774 			shost_printk(KERN_INFO, shost,
1775 				     "%s: Sending HRST\n",
1776 				     current->comm));
1777 
1778 		rtn = scsi_try_host_reset(scmd);
1779 		if (rtn == SUCCESS) {
1780 			list_splice_init(work_q, &check_list);
1781 		} else if (rtn == FAST_IO_FAIL) {
1782 			list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1783 					scsi_eh_finish_cmd(scmd, done_q);
1784 			}
1785 		} else {
1786 			SCSI_LOG_ERROR_RECOVERY(3,
1787 				shost_printk(KERN_INFO, shost,
1788 					     "%s: HRST failed\n",
1789 					     current->comm));
1790 		}
1791 	}
1792 	return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1793 }
1794 
1795 /**
1796  * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1797  * @work_q:	&list_head for pending commands.
1798  * @done_q:	&list_head for processed commands.
1799  */
scsi_eh_offline_sdevs(struct list_head * work_q,struct list_head * done_q)1800 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1801 				  struct list_head *done_q)
1802 {
1803 	struct scsi_cmnd *scmd, *next;
1804 	struct scsi_device *sdev;
1805 
1806 	list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1807 		sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1808 			    "not ready after error recovery\n");
1809 		sdev = scmd->device;
1810 
1811 		mutex_lock(&sdev->state_mutex);
1812 		scsi_device_set_state(sdev, SDEV_OFFLINE);
1813 		mutex_unlock(&sdev->state_mutex);
1814 
1815 		scsi_eh_finish_cmd(scmd, done_q);
1816 	}
1817 	return;
1818 }
1819 
1820 /**
1821  * scsi_noretry_cmd - determine if command should be failed fast
1822  * @scmd:	SCSI cmd to examine.
1823  */
scsi_noretry_cmd(struct scsi_cmnd * scmd)1824 bool scsi_noretry_cmd(struct scsi_cmnd *scmd)
1825 {
1826 	struct request *req = scsi_cmd_to_rq(scmd);
1827 
1828 	switch (host_byte(scmd->result)) {
1829 	case DID_OK:
1830 		break;
1831 	case DID_TIME_OUT:
1832 		goto check_type;
1833 	case DID_BUS_BUSY:
1834 		return !!(req->cmd_flags & REQ_FAILFAST_TRANSPORT);
1835 	case DID_PARITY:
1836 		return !!(req->cmd_flags & REQ_FAILFAST_DEV);
1837 	case DID_ERROR:
1838 		if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
1839 			return false;
1840 		fallthrough;
1841 	case DID_SOFT_ERROR:
1842 		return !!(req->cmd_flags & REQ_FAILFAST_DRIVER);
1843 	}
1844 
1845 	/* Never retry commands aborted due to a duration limit timeout */
1846 	if (scsi_ml_byte(scmd->result) == SCSIML_STAT_DL_TIMEOUT)
1847 		return true;
1848 
1849 	if (!scsi_status_is_check_condition(scmd->result))
1850 		return false;
1851 
1852 check_type:
1853 	/*
1854 	 * assume caller has checked sense and determined
1855 	 * the check condition was retryable.
1856 	 */
1857 	if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(req))
1858 		return true;
1859 
1860 	return false;
1861 }
1862 
1863 /**
1864  * scsi_decide_disposition - Disposition a cmd on return from LLD.
1865  * @scmd:	SCSI cmd to examine.
1866  *
1867  * Notes:
1868  *    This is *only* called when we are examining the status after sending
1869  *    out the actual data command.  any commands that are queued for error
1870  *    recovery (e.g. test_unit_ready) do *not* come through here.
1871  *
1872  *    When this routine returns failed, it means the error handler thread
1873  *    is woken.  In cases where the error code indicates an error that
1874  *    doesn't require the error handler read (i.e. we don't need to
1875  *    abort/reset), this function should return SUCCESS.
1876  */
scsi_decide_disposition(struct scsi_cmnd * scmd)1877 enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd)
1878 {
1879 	enum scsi_disposition rtn;
1880 
1881 	/*
1882 	 * if the device is offline, then we clearly just pass the result back
1883 	 * up to the top level.
1884 	 */
1885 	if (!scsi_device_online(scmd->device)) {
1886 		SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1887 			"%s: device offline - report as SUCCESS\n", __func__));
1888 		return SUCCESS;
1889 	}
1890 
1891 	/*
1892 	 * first check the host byte, to see if there is anything in there
1893 	 * that would indicate what we need to do.
1894 	 */
1895 	switch (host_byte(scmd->result)) {
1896 	case DID_PASSTHROUGH:
1897 		/*
1898 		 * no matter what, pass this through to the upper layer.
1899 		 * nuke this special code so that it looks like we are saying
1900 		 * did_ok.
1901 		 */
1902 		scmd->result &= 0xff00ffff;
1903 		return SUCCESS;
1904 	case DID_OK:
1905 		/*
1906 		 * looks good.  drop through, and check the next byte.
1907 		 */
1908 		break;
1909 	case DID_ABORT:
1910 		if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1911 			set_host_byte(scmd, DID_TIME_OUT);
1912 			return SUCCESS;
1913 		}
1914 		fallthrough;
1915 	case DID_NO_CONNECT:
1916 	case DID_BAD_TARGET:
1917 		/*
1918 		 * note - this means that we just report the status back
1919 		 * to the top level driver, not that we actually think
1920 		 * that it indicates SUCCESS.
1921 		 */
1922 		return SUCCESS;
1923 	case DID_SOFT_ERROR:
1924 		/*
1925 		 * when the low level driver returns did_soft_error,
1926 		 * it is responsible for keeping an internal retry counter
1927 		 * in order to avoid endless loops (db)
1928 		 */
1929 		goto maybe_retry;
1930 	case DID_IMM_RETRY:
1931 		return NEEDS_RETRY;
1932 
1933 	case DID_REQUEUE:
1934 		return ADD_TO_MLQUEUE;
1935 	case DID_TRANSPORT_DISRUPTED:
1936 		/*
1937 		 * LLD/transport was disrupted during processing of the IO.
1938 		 * The transport class is now blocked/blocking,
1939 		 * and the transport will decide what to do with the IO
1940 		 * based on its timers and recovery capablilities if
1941 		 * there are enough retries.
1942 		 */
1943 		goto maybe_retry;
1944 	case DID_TRANSPORT_FAILFAST:
1945 		/*
1946 		 * The transport decided to failfast the IO (most likely
1947 		 * the fast io fail tmo fired), so send IO directly upwards.
1948 		 */
1949 		return SUCCESS;
1950 	case DID_TRANSPORT_MARGINAL:
1951 		/*
1952 		 * caller has decided not to do retries on
1953 		 * abort success, so send IO directly upwards
1954 		 */
1955 		return SUCCESS;
1956 	case DID_ERROR:
1957 		if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
1958 			/*
1959 			 * execute reservation conflict processing code
1960 			 * lower down
1961 			 */
1962 			break;
1963 		fallthrough;
1964 	case DID_BUS_BUSY:
1965 	case DID_PARITY:
1966 		goto maybe_retry;
1967 	case DID_TIME_OUT:
1968 		/*
1969 		 * when we scan the bus, we get timeout messages for
1970 		 * these commands if there is no device available.
1971 		 * other hosts report did_no_connect for the same thing.
1972 		 */
1973 		if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1974 		     scmd->cmnd[0] == INQUIRY)) {
1975 			return SUCCESS;
1976 		} else {
1977 			return FAILED;
1978 		}
1979 	case DID_RESET:
1980 		return SUCCESS;
1981 	default:
1982 		return FAILED;
1983 	}
1984 
1985 	/*
1986 	 * check the status byte to see if this indicates anything special.
1987 	 */
1988 	switch (get_status_byte(scmd)) {
1989 	case SAM_STAT_TASK_SET_FULL:
1990 		scsi_handle_queue_full(scmd->device);
1991 		/*
1992 		 * the case of trying to send too many commands to a
1993 		 * tagged queueing device.
1994 		 */
1995 		fallthrough;
1996 	case SAM_STAT_BUSY:
1997 		/*
1998 		 * device can't talk to us at the moment.  Should only
1999 		 * occur (SAM-3) when the task queue is empty, so will cause
2000 		 * the empty queue handling to trigger a stall in the
2001 		 * device.
2002 		 */
2003 		return ADD_TO_MLQUEUE;
2004 	case SAM_STAT_GOOD:
2005 		if (scmd->cmnd[0] == REPORT_LUNS)
2006 			scmd->device->sdev_target->expecting_lun_change = 0;
2007 		scsi_handle_queue_ramp_up(scmd->device);
2008 		if (scmd->sense_buffer && SCSI_SENSE_VALID(scmd))
2009 			/*
2010 			 * If we have sense data, call scsi_check_sense() in
2011 			 * order to set the correct SCSI ML byte (if any).
2012 			 * No point in checking the return value, since the
2013 			 * command has already completed successfully.
2014 			 */
2015 			scsi_check_sense(scmd);
2016 		fallthrough;
2017 	case SAM_STAT_COMMAND_TERMINATED:
2018 		return SUCCESS;
2019 	case SAM_STAT_TASK_ABORTED:
2020 		goto maybe_retry;
2021 	case SAM_STAT_CHECK_CONDITION:
2022 		rtn = scsi_check_sense(scmd);
2023 		if (rtn == NEEDS_RETRY)
2024 			goto maybe_retry;
2025 		/* if rtn == FAILED, we have no sense information;
2026 		 * returning FAILED will wake the error handler thread
2027 		 * to collect the sense and redo the decide
2028 		 * disposition */
2029 		return rtn;
2030 	case SAM_STAT_CONDITION_MET:
2031 	case SAM_STAT_INTERMEDIATE:
2032 	case SAM_STAT_INTERMEDIATE_CONDITION_MET:
2033 	case SAM_STAT_ACA_ACTIVE:
2034 		/*
2035 		 * who knows?  FIXME(eric)
2036 		 */
2037 		return SUCCESS;
2038 
2039 	case SAM_STAT_RESERVATION_CONFLICT:
2040 		sdev_printk(KERN_INFO, scmd->device,
2041 			    "reservation conflict\n");
2042 		set_scsi_ml_byte(scmd, SCSIML_STAT_RESV_CONFLICT);
2043 		return SUCCESS; /* causes immediate i/o error */
2044 	}
2045 	return FAILED;
2046 
2047 maybe_retry:
2048 
2049 	/* we requeue for retry because the error was retryable, and
2050 	 * the request was not marked fast fail.  Note that above,
2051 	 * even if the request is marked fast fail, we still requeue
2052 	 * for queue congestion conditions (QUEUE_FULL or BUSY) */
2053 	if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) {
2054 		return NEEDS_RETRY;
2055 	} else {
2056 		/*
2057 		 * no more retries - report this one back to upper level.
2058 		 */
2059 		return SUCCESS;
2060 	}
2061 }
2062 
eh_lock_door_done(struct request * req,blk_status_t status)2063 static enum rq_end_io_ret eh_lock_door_done(struct request *req,
2064 					    blk_status_t status)
2065 {
2066 	blk_mq_free_request(req);
2067 	return RQ_END_IO_NONE;
2068 }
2069 
2070 /**
2071  * scsi_eh_lock_door - Prevent medium removal for the specified device
2072  * @sdev:	SCSI device to prevent medium removal
2073  *
2074  * Locking:
2075  * 	We must be called from process context.
2076  *
2077  * Notes:
2078  * 	We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
2079  * 	head of the devices request queue, and continue.
2080  */
scsi_eh_lock_door(struct scsi_device * sdev)2081 static void scsi_eh_lock_door(struct scsi_device *sdev)
2082 {
2083 	struct scsi_cmnd *scmd;
2084 	struct request *req;
2085 
2086 	req = scsi_alloc_request(sdev->request_queue, REQ_OP_DRV_IN, 0);
2087 	if (IS_ERR(req))
2088 		return;
2089 	scmd = blk_mq_rq_to_pdu(req);
2090 
2091 	scmd->cmnd[0] = ALLOW_MEDIUM_REMOVAL;
2092 	scmd->cmnd[1] = 0;
2093 	scmd->cmnd[2] = 0;
2094 	scmd->cmnd[3] = 0;
2095 	scmd->cmnd[4] = SCSI_REMOVAL_PREVENT;
2096 	scmd->cmnd[5] = 0;
2097 	scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
2098 	scmd->allowed = 5;
2099 
2100 	req->rq_flags |= RQF_QUIET;
2101 	req->timeout = 10 * HZ;
2102 	req->end_io = eh_lock_door_done;
2103 
2104 	blk_execute_rq_nowait(req, true);
2105 }
2106 
2107 /**
2108  * scsi_restart_operations - restart io operations to the specified host.
2109  * @shost:	Host we are restarting.
2110  *
2111  * Notes:
2112  *    When we entered the error handler, we blocked all further i/o to
2113  *    this device.  we need to 'reverse' this process.
2114  */
scsi_restart_operations(struct Scsi_Host * shost)2115 static void scsi_restart_operations(struct Scsi_Host *shost)
2116 {
2117 	struct scsi_device *sdev;
2118 	unsigned long flags;
2119 
2120 	/*
2121 	 * If the door was locked, we need to insert a door lock request
2122 	 * onto the head of the SCSI request queue for the device.  There
2123 	 * is no point trying to lock the door of an off-line device.
2124 	 */
2125 	shost_for_each_device(sdev, shost) {
2126 		if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2127 			scsi_eh_lock_door(sdev);
2128 			sdev->was_reset = 0;
2129 		}
2130 	}
2131 
2132 	/*
2133 	 * next free up anything directly waiting upon the host.  this
2134 	 * will be requests for character device operations, and also for
2135 	 * ioctls to queued block devices.
2136 	 */
2137 	SCSI_LOG_ERROR_RECOVERY(3,
2138 		shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2139 
2140 	spin_lock_irqsave(shost->host_lock, flags);
2141 	if (scsi_host_set_state(shost, SHOST_RUNNING))
2142 		if (scsi_host_set_state(shost, SHOST_CANCEL))
2143 			BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2144 	spin_unlock_irqrestore(shost->host_lock, flags);
2145 
2146 	wake_up(&shost->host_wait);
2147 
2148 	/*
2149 	 * finally we need to re-initiate requests that may be pending.  we will
2150 	 * have had everything blocked while error handling is taking place, and
2151 	 * now that error recovery is done, we will need to ensure that these
2152 	 * requests are started.
2153 	 */
2154 	scsi_run_host_queues(shost);
2155 
2156 	/*
2157 	 * if eh is active and host_eh_scheduled is pending we need to re-run
2158 	 * recovery.  we do this check after scsi_run_host_queues() to allow
2159 	 * everything pent up since the last eh run a chance to make forward
2160 	 * progress before we sync again.  Either we'll immediately re-run
2161 	 * recovery or scsi_device_unbusy() will wake us again when these
2162 	 * pending commands complete.
2163 	 */
2164 	spin_lock_irqsave(shost->host_lock, flags);
2165 	if (shost->host_eh_scheduled)
2166 		if (scsi_host_set_state(shost, SHOST_RECOVERY))
2167 			WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2168 	spin_unlock_irqrestore(shost->host_lock, flags);
2169 }
2170 
2171 /**
2172  * scsi_eh_ready_devs - check device ready state and recover if not.
2173  * @shost:	host to be recovered.
2174  * @work_q:	&list_head for pending commands.
2175  * @done_q:	&list_head for processed commands.
2176  */
scsi_eh_ready_devs(struct Scsi_Host * shost,struct list_head * work_q,struct list_head * done_q)2177 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2178 			struct list_head *work_q,
2179 			struct list_head *done_q)
2180 {
2181 	if (!scsi_eh_stu(shost, work_q, done_q))
2182 		if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2183 			if (!scsi_eh_target_reset(shost, work_q, done_q))
2184 				if (!scsi_eh_bus_reset(shost, work_q, done_q))
2185 					if (!scsi_eh_host_reset(shost, work_q, done_q))
2186 						scsi_eh_offline_sdevs(work_q,
2187 								      done_q);
2188 }
2189 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2190 
2191 /**
2192  * scsi_eh_flush_done_q - finish processed commands or retry them.
2193  * @done_q:	list_head of processed commands.
2194  */
scsi_eh_flush_done_q(struct list_head * done_q)2195 void scsi_eh_flush_done_q(struct list_head *done_q)
2196 {
2197 	struct scsi_cmnd *scmd, *next;
2198 
2199 	list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2200 		struct scsi_device *sdev = scmd->device;
2201 
2202 		list_del_init(&scmd->eh_entry);
2203 		if (scsi_device_online(sdev) && !scsi_noretry_cmd(scmd) &&
2204 		    scsi_cmd_retry_allowed(scmd) &&
2205 		    scsi_eh_should_retry_cmd(scmd)) {
2206 			SCSI_LOG_ERROR_RECOVERY(3,
2207 				scmd_printk(KERN_INFO, scmd,
2208 					     "%s: flush retry cmd\n",
2209 					     current->comm));
2210 				scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2211 				blk_mq_kick_requeue_list(sdev->request_queue);
2212 		} else {
2213 			/*
2214 			 * If just we got sense for the device (called
2215 			 * scsi_eh_get_sense), scmd->result is already
2216 			 * set, do not set DID_TIME_OUT.
2217 			 */
2218 			if (!scmd->result &&
2219 			    !(scmd->flags & SCMD_FORCE_EH_SUCCESS))
2220 				scmd->result |= (DID_TIME_OUT << 16);
2221 			SCSI_LOG_ERROR_RECOVERY(3,
2222 				scmd_printk(KERN_INFO, scmd,
2223 					     "%s: flush finish cmd\n",
2224 					     current->comm));
2225 			scsi_finish_command(scmd);
2226 		}
2227 	}
2228 }
2229 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2230 
2231 /**
2232  * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2233  * @shost:	Host to unjam.
2234  *
2235  * Notes:
2236  *    When we come in here, we *know* that all commands on the bus have
2237  *    either completed, failed or timed out.  we also know that no further
2238  *    commands are being sent to the host, so things are relatively quiet
2239  *    and we have freedom to fiddle with things as we wish.
2240  *
2241  *    This is only the *default* implementation.  it is possible for
2242  *    individual drivers to supply their own version of this function, and
2243  *    if the maintainer wishes to do this, it is strongly suggested that
2244  *    this function be taken as a template and modified.  this function
2245  *    was designed to correctly handle problems for about 95% of the
2246  *    different cases out there, and it should always provide at least a
2247  *    reasonable amount of error recovery.
2248  *
2249  *    Any command marked 'failed' or 'timeout' must eventually have
2250  *    scsi_finish_cmd() called for it.  we do all of the retry stuff
2251  *    here, so when we restart the host after we return it should have an
2252  *    empty queue.
2253  */
scsi_unjam_host(struct Scsi_Host * shost)2254 static void scsi_unjam_host(struct Scsi_Host *shost)
2255 {
2256 	unsigned long flags;
2257 	LIST_HEAD(eh_work_q);
2258 	LIST_HEAD(eh_done_q);
2259 
2260 	spin_lock_irqsave(shost->host_lock, flags);
2261 	list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2262 	spin_unlock_irqrestore(shost->host_lock, flags);
2263 
2264 	SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2265 
2266 	if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2267 		scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2268 
2269 	spin_lock_irqsave(shost->host_lock, flags);
2270 	if (shost->eh_deadline != -1)
2271 		shost->last_reset = 0;
2272 	spin_unlock_irqrestore(shost->host_lock, flags);
2273 	scsi_eh_flush_done_q(&eh_done_q);
2274 }
2275 
2276 /**
2277  * scsi_error_handler - SCSI error handler thread
2278  * @data:	Host for which we are running.
2279  *
2280  * Notes:
2281  *    This is the main error handling loop.  This is run as a kernel thread
2282  *    for every SCSI host and handles all error handling activity.
2283  */
scsi_error_handler(void * data)2284 int scsi_error_handler(void *data)
2285 {
2286 	struct Scsi_Host *shost = data;
2287 
2288 	/*
2289 	 * We use TASK_INTERRUPTIBLE so that the thread is not
2290 	 * counted against the load average as a running process.
2291 	 * We never actually get interrupted because kthread_run
2292 	 * disables signal delivery for the created thread.
2293 	 */
2294 	while (true) {
2295 		/*
2296 		 * The sequence in kthread_stop() sets the stop flag first
2297 		 * then wakes the process.  To avoid missed wakeups, the task
2298 		 * should always be in a non running state before the stop
2299 		 * flag is checked
2300 		 */
2301 		set_current_state(TASK_INTERRUPTIBLE);
2302 		if (kthread_should_stop())
2303 			break;
2304 
2305 		if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2306 		    shost->host_failed != scsi_host_busy(shost)) {
2307 			SCSI_LOG_ERROR_RECOVERY(1,
2308 				shost_printk(KERN_INFO, shost,
2309 					     "scsi_eh_%d: sleeping\n",
2310 					     shost->host_no));
2311 			schedule();
2312 			continue;
2313 		}
2314 
2315 		__set_current_state(TASK_RUNNING);
2316 		SCSI_LOG_ERROR_RECOVERY(1,
2317 			shost_printk(KERN_INFO, shost,
2318 				     "scsi_eh_%d: waking up %d/%d/%d\n",
2319 				     shost->host_no, shost->host_eh_scheduled,
2320 				     shost->host_failed,
2321 				     scsi_host_busy(shost)));
2322 
2323 		/*
2324 		 * We have a host that is failing for some reason.  Figure out
2325 		 * what we need to do to get it up and online again (if we can).
2326 		 * If we fail, we end up taking the thing offline.
2327 		 */
2328 		if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2329 			SCSI_LOG_ERROR_RECOVERY(1,
2330 				shost_printk(KERN_ERR, shost,
2331 					     "scsi_eh_%d: unable to autoresume\n",
2332 					     shost->host_no));
2333 			continue;
2334 		}
2335 
2336 		if (shost->transportt->eh_strategy_handler)
2337 			shost->transportt->eh_strategy_handler(shost);
2338 		else
2339 			scsi_unjam_host(shost);
2340 
2341 		/* All scmds have been handled */
2342 		shost->host_failed = 0;
2343 
2344 		/*
2345 		 * Note - if the above fails completely, the action is to take
2346 		 * individual devices offline and flush the queue of any
2347 		 * outstanding requests that may have been pending.  When we
2348 		 * restart, we restart any I/O to any other devices on the bus
2349 		 * which are still online.
2350 		 */
2351 		scsi_restart_operations(shost);
2352 		if (!shost->eh_noresume)
2353 			scsi_autopm_put_host(shost);
2354 	}
2355 	__set_current_state(TASK_RUNNING);
2356 
2357 	SCSI_LOG_ERROR_RECOVERY(1,
2358 		shost_printk(KERN_INFO, shost,
2359 			     "Error handler scsi_eh_%d exiting\n",
2360 			     shost->host_no));
2361 	shost->ehandler = NULL;
2362 	return 0;
2363 }
2364 
2365 /*
2366  * Function:    scsi_report_bus_reset()
2367  *
2368  * Purpose:     Utility function used by low-level drivers to report that
2369  *		they have observed a bus reset on the bus being handled.
2370  *
2371  * Arguments:   shost       - Host in question
2372  *		channel     - channel on which reset was observed.
2373  *
2374  * Returns:     Nothing
2375  *
2376  * Lock status: Host lock must be held.
2377  *
2378  * Notes:       This only needs to be called if the reset is one which
2379  *		originates from an unknown location.  Resets originated
2380  *		by the mid-level itself don't need to call this, but there
2381  *		should be no harm.
2382  *
2383  *		The main purpose of this is to make sure that a CHECK_CONDITION
2384  *		is properly treated.
2385  */
scsi_report_bus_reset(struct Scsi_Host * shost,int channel)2386 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2387 {
2388 	struct scsi_device *sdev;
2389 
2390 	__shost_for_each_device(sdev, shost) {
2391 		if (channel == sdev_channel(sdev))
2392 			__scsi_report_device_reset(sdev, NULL);
2393 	}
2394 }
2395 EXPORT_SYMBOL(scsi_report_bus_reset);
2396 
2397 /*
2398  * Function:    scsi_report_device_reset()
2399  *
2400  * Purpose:     Utility function used by low-level drivers to report that
2401  *		they have observed a device reset on the device being handled.
2402  *
2403  * Arguments:   shost       - Host in question
2404  *		channel     - channel on which reset was observed
2405  *		target	    - target on which reset was observed
2406  *
2407  * Returns:     Nothing
2408  *
2409  * Lock status: Host lock must be held
2410  *
2411  * Notes:       This only needs to be called if the reset is one which
2412  *		originates from an unknown location.  Resets originated
2413  *		by the mid-level itself don't need to call this, but there
2414  *		should be no harm.
2415  *
2416  *		The main purpose of this is to make sure that a CHECK_CONDITION
2417  *		is properly treated.
2418  */
scsi_report_device_reset(struct Scsi_Host * shost,int channel,int target)2419 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2420 {
2421 	struct scsi_device *sdev;
2422 
2423 	__shost_for_each_device(sdev, shost) {
2424 		if (channel == sdev_channel(sdev) &&
2425 		    target == sdev_id(sdev))
2426 			__scsi_report_device_reset(sdev, NULL);
2427 	}
2428 }
2429 EXPORT_SYMBOL(scsi_report_device_reset);
2430 
2431 /**
2432  * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2433  * @dev:	scsi_device to operate on
2434  * @arg:	reset type (see sg.h)
2435  */
2436 int
scsi_ioctl_reset(struct scsi_device * dev,int __user * arg)2437 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2438 {
2439 	struct scsi_cmnd *scmd;
2440 	struct Scsi_Host *shost = dev->host;
2441 	struct request *rq;
2442 	unsigned long flags;
2443 	int error = 0, val;
2444 	enum scsi_disposition rtn;
2445 
2446 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2447 		return -EACCES;
2448 
2449 	error = get_user(val, arg);
2450 	if (error)
2451 		return error;
2452 
2453 	if (scsi_autopm_get_host(shost) < 0)
2454 		return -EIO;
2455 
2456 	error = -EIO;
2457 	rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2458 			shost->hostt->cmd_size, GFP_KERNEL);
2459 	if (!rq)
2460 		goto out_put_autopm_host;
2461 	blk_rq_init(NULL, rq);
2462 
2463 	scmd = (struct scsi_cmnd *)(rq + 1);
2464 	scsi_init_command(dev, scmd);
2465 
2466 	scmd->submitter = SUBMITTED_BY_SCSI_RESET_IOCTL;
2467 	scmd->flags |= SCMD_LAST;
2468 	memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2469 
2470 	scmd->cmd_len			= 0;
2471 
2472 	scmd->sc_data_direction		= DMA_BIDIRECTIONAL;
2473 
2474 	spin_lock_irqsave(shost->host_lock, flags);
2475 	shost->tmf_in_progress = 1;
2476 	spin_unlock_irqrestore(shost->host_lock, flags);
2477 
2478 	switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2479 	case SG_SCSI_RESET_NOTHING:
2480 		rtn = SUCCESS;
2481 		break;
2482 	case SG_SCSI_RESET_DEVICE:
2483 		rtn = scsi_try_bus_device_reset(scmd);
2484 		if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2485 			break;
2486 		fallthrough;
2487 	case SG_SCSI_RESET_TARGET:
2488 		rtn = scsi_try_target_reset(scmd);
2489 		if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2490 			break;
2491 		fallthrough;
2492 	case SG_SCSI_RESET_BUS:
2493 		rtn = scsi_try_bus_reset(scmd);
2494 		if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2495 			break;
2496 		fallthrough;
2497 	case SG_SCSI_RESET_HOST:
2498 		rtn = scsi_try_host_reset(scmd);
2499 		if (rtn == SUCCESS)
2500 			break;
2501 		fallthrough;
2502 	default:
2503 		rtn = FAILED;
2504 		break;
2505 	}
2506 
2507 	error = (rtn == SUCCESS) ? 0 : -EIO;
2508 
2509 	spin_lock_irqsave(shost->host_lock, flags);
2510 	shost->tmf_in_progress = 0;
2511 	spin_unlock_irqrestore(shost->host_lock, flags);
2512 
2513 	/*
2514 	 * be sure to wake up anyone who was sleeping or had their queue
2515 	 * suspended while we performed the TMF.
2516 	 */
2517 	SCSI_LOG_ERROR_RECOVERY(3,
2518 		shost_printk(KERN_INFO, shost,
2519 			     "waking up host to restart after TMF\n"));
2520 
2521 	wake_up(&shost->host_wait);
2522 	scsi_run_host_queues(shost);
2523 
2524 	kfree(rq);
2525 
2526 out_put_autopm_host:
2527 	scsi_autopm_put_host(shost);
2528 	return error;
2529 }
2530 
scsi_command_normalize_sense(const struct scsi_cmnd * cmd,struct scsi_sense_hdr * sshdr)2531 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2532 				  struct scsi_sense_hdr *sshdr)
2533 {
2534 	return scsi_normalize_sense(cmd->sense_buffer,
2535 			SCSI_SENSE_BUFFERSIZE, sshdr);
2536 }
2537 EXPORT_SYMBOL(scsi_command_normalize_sense);
2538 
2539 /**
2540  * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2541  * @sense_buffer:	byte array of sense data
2542  * @sb_len:		number of valid bytes in sense_buffer
2543  * @info_out:		pointer to 64 integer where 8 or 4 byte information
2544  *			field will be placed if found.
2545  *
2546  * Return value:
2547  *	true if information field found, false if not found.
2548  */
scsi_get_sense_info_fld(const u8 * sense_buffer,int sb_len,u64 * info_out)2549 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2550 			     u64 *info_out)
2551 {
2552 	const u8 * ucp;
2553 
2554 	if (sb_len < 7)
2555 		return false;
2556 	switch (sense_buffer[0] & 0x7f) {
2557 	case 0x70:
2558 	case 0x71:
2559 		if (sense_buffer[0] & 0x80) {
2560 			*info_out = get_unaligned_be32(&sense_buffer[3]);
2561 			return true;
2562 		}
2563 		return false;
2564 	case 0x72:
2565 	case 0x73:
2566 		ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2567 					   0 /* info desc */);
2568 		if (ucp && (0xa == ucp[1])) {
2569 			*info_out = get_unaligned_be64(&ucp[4]);
2570 			return true;
2571 		}
2572 		return false;
2573 	default:
2574 		return false;
2575 	}
2576 }
2577 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2578