1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  *	watchdog_dev.c
4  *
5  *	(c) Copyright 2008-2011 Alan Cox <alan@lxorguk.ukuu.org.uk>,
6  *						All Rights Reserved.
7  *
8  *	(c) Copyright 2008-2011 Wim Van Sebroeck <wim@iguana.be>.
9  *
10  *	(c) Copyright 2021 Hewlett Packard Enterprise Development LP.
11  *
12  *	This source code is part of the generic code that can be used
13  *	by all the watchdog timer drivers.
14  *
15  *	This part of the generic code takes care of the following
16  *	misc device: /dev/watchdog.
17  *
18  *	Based on source code of the following authors:
19  *	  Matt Domsch <Matt_Domsch@dell.com>,
20  *	  Rob Radez <rob@osinvestor.com>,
21  *	  Rusty Lynch <rusty@linux.co.intel.com>
22  *	  Satyam Sharma <satyam@infradead.org>
23  *	  Randy Dunlap <randy.dunlap@oracle.com>
24  *
25  *	Neither Alan Cox, CymruNet Ltd., Wim Van Sebroeck nor Iguana vzw.
26  *	admit liability nor provide warranty for any of this software.
27  *	This material is provided "AS-IS" and at no charge.
28  */
29 
30 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 
32 #include <linux/cdev.h>		/* For character device */
33 #include <linux/errno.h>	/* For the -ENODEV/... values */
34 #include <linux/fs.h>		/* For file operations */
35 #include <linux/init.h>		/* For __init/__exit/... */
36 #include <linux/hrtimer.h>	/* For hrtimers */
37 #include <linux/kernel.h>	/* For printk/panic/... */
38 #include <linux/kthread.h>	/* For kthread_work */
39 #include <linux/miscdevice.h>	/* For handling misc devices */
40 #include <linux/module.h>	/* For module stuff/... */
41 #include <linux/mutex.h>	/* For mutexes */
42 #include <linux/slab.h>		/* For memory functions */
43 #include <linux/types.h>	/* For standard types (like size_t) */
44 #include <linux/watchdog.h>	/* For watchdog specific items */
45 #include <linux/uaccess.h>	/* For copy_to_user/put_user/... */
46 
47 #include "watchdog_core.h"
48 #include "watchdog_pretimeout.h"
49 
50 /* the dev_t structure to store the dynamically allocated watchdog devices */
51 static dev_t watchdog_devt;
52 /* Reference to watchdog device behind /dev/watchdog */
53 static struct watchdog_core_data *old_wd_data;
54 
55 static struct kthread_worker *watchdog_kworker;
56 
57 static bool handle_boot_enabled =
58 	IS_ENABLED(CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED);
59 
60 static unsigned open_timeout = CONFIG_WATCHDOG_OPEN_TIMEOUT;
61 
62 static bool watchdog_past_open_deadline(struct watchdog_core_data *data)
63 {
64 	return ktime_after(ktime_get(), data->open_deadline);
65 }
66 
67 static void watchdog_set_open_deadline(struct watchdog_core_data *data)
68 {
69 	data->open_deadline = open_timeout ?
70 		ktime_get() + ktime_set(open_timeout, 0) : KTIME_MAX;
71 }
72 
73 static inline bool watchdog_need_worker(struct watchdog_device *wdd)
74 {
75 	/* All variables in milli-seconds */
76 	unsigned int hm = wdd->max_hw_heartbeat_ms;
77 	unsigned int t = wdd->timeout * 1000;
78 
79 	/*
80 	 * A worker to generate heartbeat requests is needed if all of the
81 	 * following conditions are true.
82 	 * - Userspace activated the watchdog.
83 	 * - The driver provided a value for the maximum hardware timeout, and
84 	 *   thus is aware that the framework supports generating heartbeat
85 	 *   requests.
86 	 * - Userspace requests a longer timeout than the hardware can handle.
87 	 *
88 	 * Alternatively, if userspace has not opened the watchdog
89 	 * device, we take care of feeding the watchdog if it is
90 	 * running.
91 	 */
92 	return (hm && watchdog_active(wdd) && t > hm) ||
93 		(t && !watchdog_active(wdd) && watchdog_hw_running(wdd));
94 }
95 
96 static ktime_t watchdog_next_keepalive(struct watchdog_device *wdd)
97 {
98 	struct watchdog_core_data *wd_data = wdd->wd_data;
99 	unsigned int timeout_ms = wdd->timeout * 1000;
100 	ktime_t keepalive_interval;
101 	ktime_t last_heartbeat, latest_heartbeat;
102 	ktime_t virt_timeout;
103 	unsigned int hw_heartbeat_ms;
104 
105 	if (watchdog_active(wdd))
106 		virt_timeout = ktime_add(wd_data->last_keepalive,
107 					 ms_to_ktime(timeout_ms));
108 	else
109 		virt_timeout = wd_data->open_deadline;
110 
111 	hw_heartbeat_ms = min_not_zero(timeout_ms, wdd->max_hw_heartbeat_ms);
112 	keepalive_interval = ms_to_ktime(hw_heartbeat_ms / 2);
113 
114 	/*
115 	 * To ensure that the watchdog times out wdd->timeout seconds
116 	 * after the most recent ping from userspace, the last
117 	 * worker ping has to come in hw_heartbeat_ms before this timeout.
118 	 */
119 	last_heartbeat = ktime_sub(virt_timeout, ms_to_ktime(hw_heartbeat_ms));
120 	latest_heartbeat = ktime_sub(last_heartbeat, ktime_get());
121 	if (ktime_before(latest_heartbeat, keepalive_interval))
122 		return latest_heartbeat;
123 	return keepalive_interval;
124 }
125 
126 static inline void watchdog_update_worker(struct watchdog_device *wdd)
127 {
128 	struct watchdog_core_data *wd_data = wdd->wd_data;
129 
130 	if (watchdog_need_worker(wdd)) {
131 		ktime_t t = watchdog_next_keepalive(wdd);
132 
133 		if (t > 0)
134 			hrtimer_start(&wd_data->timer, t,
135 				      HRTIMER_MODE_REL_HARD);
136 	} else {
137 		hrtimer_cancel(&wd_data->timer);
138 	}
139 }
140 
141 static int __watchdog_ping(struct watchdog_device *wdd)
142 {
143 	struct watchdog_core_data *wd_data = wdd->wd_data;
144 	ktime_t earliest_keepalive, now;
145 	int err;
146 
147 	earliest_keepalive = ktime_add(wd_data->last_hw_keepalive,
148 				       ms_to_ktime(wdd->min_hw_heartbeat_ms));
149 	now = ktime_get();
150 
151 	if (ktime_after(earliest_keepalive, now)) {
152 		hrtimer_start(&wd_data->timer,
153 			      ktime_sub(earliest_keepalive, now),
154 			      HRTIMER_MODE_REL_HARD);
155 		return 0;
156 	}
157 
158 	wd_data->last_hw_keepalive = now;
159 
160 	if (wdd->ops->ping)
161 		err = wdd->ops->ping(wdd);  /* ping the watchdog */
162 	else
163 		err = wdd->ops->start(wdd); /* restart watchdog */
164 
165 	if (err == 0)
166 		watchdog_hrtimer_pretimeout_start(wdd);
167 
168 	watchdog_update_worker(wdd);
169 
170 	return err;
171 }
172 
173 /*
174  *	watchdog_ping: ping the watchdog.
175  *	@wdd: the watchdog device to ping
176  *
177  *	The caller must hold wd_data->lock.
178  *
179  *	If the watchdog has no own ping operation then it needs to be
180  *	restarted via the start operation. This wrapper function does
181  *	exactly that.
182  *	We only ping when the watchdog device is running.
183  */
184 
185 static int watchdog_ping(struct watchdog_device *wdd)
186 {
187 	struct watchdog_core_data *wd_data = wdd->wd_data;
188 
189 	if (!watchdog_active(wdd) && !watchdog_hw_running(wdd))
190 		return 0;
191 
192 	set_bit(_WDOG_KEEPALIVE, &wd_data->status);
193 
194 	wd_data->last_keepalive = ktime_get();
195 	return __watchdog_ping(wdd);
196 }
197 
198 static bool watchdog_worker_should_ping(struct watchdog_core_data *wd_data)
199 {
200 	struct watchdog_device *wdd = wd_data->wdd;
201 
202 	if (!wdd)
203 		return false;
204 
205 	if (watchdog_active(wdd))
206 		return true;
207 
208 	return watchdog_hw_running(wdd) && !watchdog_past_open_deadline(wd_data);
209 }
210 
211 static void watchdog_ping_work(struct kthread_work *work)
212 {
213 	struct watchdog_core_data *wd_data;
214 
215 	wd_data = container_of(work, struct watchdog_core_data, work);
216 
217 	mutex_lock(&wd_data->lock);
218 	if (watchdog_worker_should_ping(wd_data))
219 		__watchdog_ping(wd_data->wdd);
220 	mutex_unlock(&wd_data->lock);
221 }
222 
223 static enum hrtimer_restart watchdog_timer_expired(struct hrtimer *timer)
224 {
225 	struct watchdog_core_data *wd_data;
226 
227 	wd_data = container_of(timer, struct watchdog_core_data, timer);
228 
229 	kthread_queue_work(watchdog_kworker, &wd_data->work);
230 	return HRTIMER_NORESTART;
231 }
232 
233 /*
234  *	watchdog_start: wrapper to start the watchdog.
235  *	@wdd: the watchdog device to start
236  *
237  *	The caller must hold wd_data->lock.
238  *
239  *	Start the watchdog if it is not active and mark it active.
240  *	This function returns zero on success or a negative errno code for
241  *	failure.
242  */
243 
244 static int watchdog_start(struct watchdog_device *wdd)
245 {
246 	struct watchdog_core_data *wd_data = wdd->wd_data;
247 	ktime_t started_at;
248 	int err;
249 
250 	if (watchdog_active(wdd))
251 		return 0;
252 
253 	set_bit(_WDOG_KEEPALIVE, &wd_data->status);
254 
255 	started_at = ktime_get();
256 	if (watchdog_hw_running(wdd) && wdd->ops->ping) {
257 		err = __watchdog_ping(wdd);
258 		if (err == 0) {
259 			set_bit(WDOG_ACTIVE, &wdd->status);
260 			watchdog_hrtimer_pretimeout_start(wdd);
261 		}
262 	} else {
263 		err = wdd->ops->start(wdd);
264 		if (err == 0) {
265 			set_bit(WDOG_ACTIVE, &wdd->status);
266 			wd_data->last_keepalive = started_at;
267 			wd_data->last_hw_keepalive = started_at;
268 			watchdog_update_worker(wdd);
269 			watchdog_hrtimer_pretimeout_start(wdd);
270 		}
271 	}
272 
273 	return err;
274 }
275 
276 /*
277  *	watchdog_stop: wrapper to stop the watchdog.
278  *	@wdd: the watchdog device to stop
279  *
280  *	The caller must hold wd_data->lock.
281  *
282  *	Stop the watchdog if it is still active and unmark it active.
283  *	This function returns zero on success or a negative errno code for
284  *	failure.
285  *	If the 'nowayout' feature was set, the watchdog cannot be stopped.
286  */
287 
288 static int watchdog_stop(struct watchdog_device *wdd)
289 {
290 	int err = 0;
291 
292 	if (!watchdog_active(wdd))
293 		return 0;
294 
295 	if (test_bit(WDOG_NO_WAY_OUT, &wdd->status)) {
296 		pr_info("watchdog%d: nowayout prevents watchdog being stopped!\n",
297 			wdd->id);
298 		return -EBUSY;
299 	}
300 
301 	if (wdd->ops->stop) {
302 		clear_bit(WDOG_HW_RUNNING, &wdd->status);
303 		err = wdd->ops->stop(wdd);
304 	} else {
305 		set_bit(WDOG_HW_RUNNING, &wdd->status);
306 	}
307 
308 	if (err == 0) {
309 		clear_bit(WDOG_ACTIVE, &wdd->status);
310 		watchdog_update_worker(wdd);
311 		watchdog_hrtimer_pretimeout_stop(wdd);
312 	}
313 
314 	return err;
315 }
316 
317 /*
318  *	watchdog_get_status: wrapper to get the watchdog status
319  *	@wdd: the watchdog device to get the status from
320  *
321  *	The caller must hold wd_data->lock.
322  *
323  *	Get the watchdog's status flags.
324  */
325 
326 static unsigned int watchdog_get_status(struct watchdog_device *wdd)
327 {
328 	struct watchdog_core_data *wd_data = wdd->wd_data;
329 	unsigned int status;
330 
331 	if (wdd->ops->status)
332 		status = wdd->ops->status(wdd);
333 	else
334 		status = wdd->bootstatus & (WDIOF_CARDRESET |
335 					    WDIOF_OVERHEAT |
336 					    WDIOF_FANFAULT |
337 					    WDIOF_EXTERN1 |
338 					    WDIOF_EXTERN2 |
339 					    WDIOF_POWERUNDER |
340 					    WDIOF_POWEROVER);
341 
342 	if (test_bit(_WDOG_ALLOW_RELEASE, &wd_data->status))
343 		status |= WDIOF_MAGICCLOSE;
344 
345 	if (test_and_clear_bit(_WDOG_KEEPALIVE, &wd_data->status))
346 		status |= WDIOF_KEEPALIVEPING;
347 
348 	if (IS_ENABLED(CONFIG_WATCHDOG_HRTIMER_PRETIMEOUT))
349 		status |= WDIOF_PRETIMEOUT;
350 
351 	return status;
352 }
353 
354 /*
355  *	watchdog_set_timeout: set the watchdog timer timeout
356  *	@wdd: the watchdog device to set the timeout for
357  *	@timeout: timeout to set in seconds
358  *
359  *	The caller must hold wd_data->lock.
360  */
361 
362 static int watchdog_set_timeout(struct watchdog_device *wdd,
363 							unsigned int timeout)
364 {
365 	int err = 0;
366 
367 	if (!(wdd->info->options & WDIOF_SETTIMEOUT))
368 		return -EOPNOTSUPP;
369 
370 	if (watchdog_timeout_invalid(wdd, timeout))
371 		return -EINVAL;
372 
373 	if (wdd->ops->set_timeout) {
374 		err = wdd->ops->set_timeout(wdd, timeout);
375 	} else {
376 		wdd->timeout = timeout;
377 		/* Disable pretimeout if it doesn't fit the new timeout */
378 		if (wdd->pretimeout >= wdd->timeout)
379 			wdd->pretimeout = 0;
380 	}
381 
382 	watchdog_update_worker(wdd);
383 
384 	return err;
385 }
386 
387 /*
388  *	watchdog_set_pretimeout: set the watchdog timer pretimeout
389  *	@wdd: the watchdog device to set the timeout for
390  *	@timeout: pretimeout to set in seconds
391  */
392 
393 static int watchdog_set_pretimeout(struct watchdog_device *wdd,
394 				   unsigned int timeout)
395 {
396 	int err = 0;
397 
398 	if (!watchdog_have_pretimeout(wdd))
399 		return -EOPNOTSUPP;
400 
401 	if (watchdog_pretimeout_invalid(wdd, timeout))
402 		return -EINVAL;
403 
404 	if (wdd->ops->set_pretimeout)
405 		err = wdd->ops->set_pretimeout(wdd, timeout);
406 	else
407 		wdd->pretimeout = timeout;
408 
409 	return err;
410 }
411 
412 /*
413  *	watchdog_get_timeleft: wrapper to get the time left before a reboot
414  *	@wdd: the watchdog device to get the remaining time from
415  *	@timeleft: the time that's left
416  *
417  *	The caller must hold wd_data->lock.
418  *
419  *	Get the time before a watchdog will reboot (if not pinged).
420  */
421 
422 static int watchdog_get_timeleft(struct watchdog_device *wdd,
423 							unsigned int *timeleft)
424 {
425 	*timeleft = 0;
426 
427 	if (!wdd->ops->get_timeleft)
428 		return -EOPNOTSUPP;
429 
430 	*timeleft = wdd->ops->get_timeleft(wdd);
431 
432 	return 0;
433 }
434 
435 #ifdef CONFIG_WATCHDOG_SYSFS
436 static ssize_t nowayout_show(struct device *dev, struct device_attribute *attr,
437 				char *buf)
438 {
439 	struct watchdog_device *wdd = dev_get_drvdata(dev);
440 
441 	return sysfs_emit(buf, "%d\n", !!test_bit(WDOG_NO_WAY_OUT,
442 						  &wdd->status));
443 }
444 
445 static ssize_t nowayout_store(struct device *dev, struct device_attribute *attr,
446 				const char *buf, size_t len)
447 {
448 	struct watchdog_device *wdd = dev_get_drvdata(dev);
449 	unsigned int value;
450 	int ret;
451 
452 	ret = kstrtouint(buf, 0, &value);
453 	if (ret)
454 		return ret;
455 	if (value > 1)
456 		return -EINVAL;
457 	/* nowayout cannot be disabled once set */
458 	if (test_bit(WDOG_NO_WAY_OUT, &wdd->status) && !value)
459 		return -EPERM;
460 	watchdog_set_nowayout(wdd, value);
461 	return len;
462 }
463 static DEVICE_ATTR_RW(nowayout);
464 
465 static ssize_t status_show(struct device *dev, struct device_attribute *attr,
466 				char *buf)
467 {
468 	struct watchdog_device *wdd = dev_get_drvdata(dev);
469 	struct watchdog_core_data *wd_data = wdd->wd_data;
470 	unsigned int status;
471 
472 	mutex_lock(&wd_data->lock);
473 	status = watchdog_get_status(wdd);
474 	mutex_unlock(&wd_data->lock);
475 
476 	return sysfs_emit(buf, "0x%x\n", status);
477 }
478 static DEVICE_ATTR_RO(status);
479 
480 static ssize_t bootstatus_show(struct device *dev,
481 				struct device_attribute *attr, char *buf)
482 {
483 	struct watchdog_device *wdd = dev_get_drvdata(dev);
484 
485 	return sysfs_emit(buf, "%u\n", wdd->bootstatus);
486 }
487 static DEVICE_ATTR_RO(bootstatus);
488 
489 static ssize_t timeleft_show(struct device *dev, struct device_attribute *attr,
490 				char *buf)
491 {
492 	struct watchdog_device *wdd = dev_get_drvdata(dev);
493 	struct watchdog_core_data *wd_data = wdd->wd_data;
494 	ssize_t status;
495 	unsigned int val;
496 
497 	mutex_lock(&wd_data->lock);
498 	status = watchdog_get_timeleft(wdd, &val);
499 	mutex_unlock(&wd_data->lock);
500 	if (!status)
501 		status = sysfs_emit(buf, "%u\n", val);
502 
503 	return status;
504 }
505 static DEVICE_ATTR_RO(timeleft);
506 
507 static ssize_t timeout_show(struct device *dev, struct device_attribute *attr,
508 				char *buf)
509 {
510 	struct watchdog_device *wdd = dev_get_drvdata(dev);
511 
512 	return sysfs_emit(buf, "%u\n", wdd->timeout);
513 }
514 static DEVICE_ATTR_RO(timeout);
515 
516 static ssize_t min_timeout_show(struct device *dev,
517 				struct device_attribute *attr, char *buf)
518 {
519 	struct watchdog_device *wdd = dev_get_drvdata(dev);
520 
521 	return sysfs_emit(buf, "%u\n", wdd->min_timeout);
522 }
523 static DEVICE_ATTR_RO(min_timeout);
524 
525 static ssize_t max_timeout_show(struct device *dev,
526 				struct device_attribute *attr, char *buf)
527 {
528 	struct watchdog_device *wdd = dev_get_drvdata(dev);
529 
530 	return sysfs_emit(buf, "%u\n", wdd->max_timeout);
531 }
532 static DEVICE_ATTR_RO(max_timeout);
533 
534 static ssize_t pretimeout_show(struct device *dev,
535 			       struct device_attribute *attr, char *buf)
536 {
537 	struct watchdog_device *wdd = dev_get_drvdata(dev);
538 
539 	return sysfs_emit(buf, "%u\n", wdd->pretimeout);
540 }
541 static DEVICE_ATTR_RO(pretimeout);
542 
543 static ssize_t identity_show(struct device *dev, struct device_attribute *attr,
544 				char *buf)
545 {
546 	struct watchdog_device *wdd = dev_get_drvdata(dev);
547 
548 	return sysfs_emit(buf, "%s\n", wdd->info->identity);
549 }
550 static DEVICE_ATTR_RO(identity);
551 
552 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
553 				char *buf)
554 {
555 	struct watchdog_device *wdd = dev_get_drvdata(dev);
556 
557 	if (watchdog_active(wdd))
558 		return sysfs_emit(buf, "active\n");
559 
560 	return sysfs_emit(buf, "inactive\n");
561 }
562 static DEVICE_ATTR_RO(state);
563 
564 static ssize_t pretimeout_available_governors_show(struct device *dev,
565 				   struct device_attribute *attr, char *buf)
566 {
567 	return watchdog_pretimeout_available_governors_get(buf);
568 }
569 static DEVICE_ATTR_RO(pretimeout_available_governors);
570 
571 static ssize_t pretimeout_governor_show(struct device *dev,
572 					struct device_attribute *attr,
573 					char *buf)
574 {
575 	struct watchdog_device *wdd = dev_get_drvdata(dev);
576 
577 	return watchdog_pretimeout_governor_get(wdd, buf);
578 }
579 
580 static ssize_t pretimeout_governor_store(struct device *dev,
581 					 struct device_attribute *attr,
582 					 const char *buf, size_t count)
583 {
584 	struct watchdog_device *wdd = dev_get_drvdata(dev);
585 	int ret = watchdog_pretimeout_governor_set(wdd, buf);
586 
587 	if (!ret)
588 		ret = count;
589 
590 	return ret;
591 }
592 static DEVICE_ATTR_RW(pretimeout_governor);
593 
594 static umode_t wdt_is_visible(struct kobject *kobj, struct attribute *attr,
595 				int n)
596 {
597 	struct device *dev = kobj_to_dev(kobj);
598 	struct watchdog_device *wdd = dev_get_drvdata(dev);
599 	umode_t mode = attr->mode;
600 
601 	if (attr == &dev_attr_timeleft.attr && !wdd->ops->get_timeleft)
602 		mode = 0;
603 	else if (attr == &dev_attr_pretimeout.attr && !watchdog_have_pretimeout(wdd))
604 		mode = 0;
605 	else if ((attr == &dev_attr_pretimeout_governor.attr ||
606 		  attr == &dev_attr_pretimeout_available_governors.attr) &&
607 		 (!watchdog_have_pretimeout(wdd) || !IS_ENABLED(CONFIG_WATCHDOG_PRETIMEOUT_GOV)))
608 		mode = 0;
609 
610 	return mode;
611 }
612 static struct attribute *wdt_attrs[] = {
613 	&dev_attr_state.attr,
614 	&dev_attr_identity.attr,
615 	&dev_attr_timeout.attr,
616 	&dev_attr_min_timeout.attr,
617 	&dev_attr_max_timeout.attr,
618 	&dev_attr_pretimeout.attr,
619 	&dev_attr_timeleft.attr,
620 	&dev_attr_bootstatus.attr,
621 	&dev_attr_status.attr,
622 	&dev_attr_nowayout.attr,
623 	&dev_attr_pretimeout_governor.attr,
624 	&dev_attr_pretimeout_available_governors.attr,
625 	NULL,
626 };
627 
628 static const struct attribute_group wdt_group = {
629 	.attrs = wdt_attrs,
630 	.is_visible = wdt_is_visible,
631 };
632 __ATTRIBUTE_GROUPS(wdt);
633 #else
634 #define wdt_groups	NULL
635 #endif
636 
637 /*
638  *	watchdog_ioctl_op: call the watchdog drivers ioctl op if defined
639  *	@wdd: the watchdog device to do the ioctl on
640  *	@cmd: watchdog command
641  *	@arg: argument pointer
642  *
643  *	The caller must hold wd_data->lock.
644  */
645 
646 static int watchdog_ioctl_op(struct watchdog_device *wdd, unsigned int cmd,
647 							unsigned long arg)
648 {
649 	if (!wdd->ops->ioctl)
650 		return -ENOIOCTLCMD;
651 
652 	return wdd->ops->ioctl(wdd, cmd, arg);
653 }
654 
655 /*
656  *	watchdog_write: writes to the watchdog.
657  *	@file: file from VFS
658  *	@data: user address of data
659  *	@len: length of data
660  *	@ppos: pointer to the file offset
661  *
662  *	A write to a watchdog device is defined as a keepalive ping.
663  *	Writing the magic 'V' sequence allows the next close to turn
664  *	off the watchdog (if 'nowayout' is not set).
665  */
666 
667 static ssize_t watchdog_write(struct file *file, const char __user *data,
668 						size_t len, loff_t *ppos)
669 {
670 	struct watchdog_core_data *wd_data = file->private_data;
671 	struct watchdog_device *wdd;
672 	int err;
673 	size_t i;
674 	char c;
675 
676 	if (len == 0)
677 		return 0;
678 
679 	/*
680 	 * Note: just in case someone wrote the magic character
681 	 * five months ago...
682 	 */
683 	clear_bit(_WDOG_ALLOW_RELEASE, &wd_data->status);
684 
685 	/* scan to see whether or not we got the magic character */
686 	for (i = 0; i != len; i++) {
687 		if (get_user(c, data + i))
688 			return -EFAULT;
689 		if (c == 'V')
690 			set_bit(_WDOG_ALLOW_RELEASE, &wd_data->status);
691 	}
692 
693 	/* someone wrote to us, so we send the watchdog a keepalive ping */
694 
695 	err = -ENODEV;
696 	mutex_lock(&wd_data->lock);
697 	wdd = wd_data->wdd;
698 	if (wdd)
699 		err = watchdog_ping(wdd);
700 	mutex_unlock(&wd_data->lock);
701 
702 	if (err < 0)
703 		return err;
704 
705 	return len;
706 }
707 
708 /*
709  *	watchdog_ioctl: handle the different ioctl's for the watchdog device.
710  *	@file: file handle to the device
711  *	@cmd: watchdog command
712  *	@arg: argument pointer
713  *
714  *	The watchdog API defines a common set of functions for all watchdogs
715  *	according to their available features.
716  */
717 
718 static long watchdog_ioctl(struct file *file, unsigned int cmd,
719 							unsigned long arg)
720 {
721 	struct watchdog_core_data *wd_data = file->private_data;
722 	void __user *argp = (void __user *)arg;
723 	struct watchdog_device *wdd;
724 	int __user *p = argp;
725 	unsigned int val;
726 	int err;
727 
728 	mutex_lock(&wd_data->lock);
729 
730 	wdd = wd_data->wdd;
731 	if (!wdd) {
732 		err = -ENODEV;
733 		goto out_ioctl;
734 	}
735 
736 	err = watchdog_ioctl_op(wdd, cmd, arg);
737 	if (err != -ENOIOCTLCMD)
738 		goto out_ioctl;
739 
740 	switch (cmd) {
741 	case WDIOC_GETSUPPORT:
742 		err = copy_to_user(argp, wdd->info,
743 			sizeof(struct watchdog_info)) ? -EFAULT : 0;
744 		break;
745 	case WDIOC_GETSTATUS:
746 		val = watchdog_get_status(wdd);
747 		err = put_user(val, p);
748 		break;
749 	case WDIOC_GETBOOTSTATUS:
750 		err = put_user(wdd->bootstatus, p);
751 		break;
752 	case WDIOC_SETOPTIONS:
753 		if (get_user(val, p)) {
754 			err = -EFAULT;
755 			break;
756 		}
757 		if (val & WDIOS_DISABLECARD) {
758 			err = watchdog_stop(wdd);
759 			if (err < 0)
760 				break;
761 		}
762 		if (val & WDIOS_ENABLECARD)
763 			err = watchdog_start(wdd);
764 		break;
765 	case WDIOC_KEEPALIVE:
766 		if (!(wdd->info->options & WDIOF_KEEPALIVEPING)) {
767 			err = -EOPNOTSUPP;
768 			break;
769 		}
770 		err = watchdog_ping(wdd);
771 		break;
772 	case WDIOC_SETTIMEOUT:
773 		if (get_user(val, p)) {
774 			err = -EFAULT;
775 			break;
776 		}
777 		err = watchdog_set_timeout(wdd, val);
778 		if (err < 0)
779 			break;
780 		/* If the watchdog is active then we send a keepalive ping
781 		 * to make sure that the watchdog keep's running (and if
782 		 * possible that it takes the new timeout) */
783 		err = watchdog_ping(wdd);
784 		if (err < 0)
785 			break;
786 		fallthrough;
787 	case WDIOC_GETTIMEOUT:
788 		/* timeout == 0 means that we don't know the timeout */
789 		if (wdd->timeout == 0) {
790 			err = -EOPNOTSUPP;
791 			break;
792 		}
793 		err = put_user(wdd->timeout, p);
794 		break;
795 	case WDIOC_GETTIMELEFT:
796 		err = watchdog_get_timeleft(wdd, &val);
797 		if (err < 0)
798 			break;
799 		err = put_user(val, p);
800 		break;
801 	case WDIOC_SETPRETIMEOUT:
802 		if (get_user(val, p)) {
803 			err = -EFAULT;
804 			break;
805 		}
806 		err = watchdog_set_pretimeout(wdd, val);
807 		break;
808 	case WDIOC_GETPRETIMEOUT:
809 		err = put_user(wdd->pretimeout, p);
810 		break;
811 	default:
812 		err = -ENOTTY;
813 		break;
814 	}
815 
816 out_ioctl:
817 	mutex_unlock(&wd_data->lock);
818 	return err;
819 }
820 
821 /*
822  *	watchdog_open: open the /dev/watchdog* devices.
823  *	@inode: inode of device
824  *	@file: file handle to device
825  *
826  *	When the /dev/watchdog* device gets opened, we start the watchdog.
827  *	Watch out: the /dev/watchdog device is single open, so we make sure
828  *	it can only be opened once.
829  */
830 
831 static int watchdog_open(struct inode *inode, struct file *file)
832 {
833 	struct watchdog_core_data *wd_data;
834 	struct watchdog_device *wdd;
835 	bool hw_running;
836 	int err;
837 
838 	/* Get the corresponding watchdog device */
839 	if (imajor(inode) == MISC_MAJOR)
840 		wd_data = old_wd_data;
841 	else
842 		wd_data = container_of(inode->i_cdev, struct watchdog_core_data,
843 				       cdev);
844 
845 	/* the watchdog is single open! */
846 	if (test_and_set_bit(_WDOG_DEV_OPEN, &wd_data->status))
847 		return -EBUSY;
848 
849 	wdd = wd_data->wdd;
850 
851 	/*
852 	 * If the /dev/watchdog device is open, we don't want the module
853 	 * to be unloaded.
854 	 */
855 	hw_running = watchdog_hw_running(wdd);
856 	if (!hw_running && !try_module_get(wdd->ops->owner)) {
857 		err = -EBUSY;
858 		goto out_clear;
859 	}
860 
861 	err = watchdog_start(wdd);
862 	if (err < 0)
863 		goto out_mod;
864 
865 	file->private_data = wd_data;
866 
867 	if (!hw_running)
868 		get_device(&wd_data->dev);
869 
870 	/*
871 	 * open_timeout only applies for the first open from
872 	 * userspace. Set open_deadline to infinity so that the kernel
873 	 * will take care of an always-running hardware watchdog in
874 	 * case the device gets magic-closed or WDIOS_DISABLECARD is
875 	 * applied.
876 	 */
877 	wd_data->open_deadline = KTIME_MAX;
878 
879 	/* dev/watchdog is a virtual (and thus non-seekable) filesystem */
880 	return stream_open(inode, file);
881 
882 out_mod:
883 	module_put(wd_data->wdd->ops->owner);
884 out_clear:
885 	clear_bit(_WDOG_DEV_OPEN, &wd_data->status);
886 	return err;
887 }
888 
889 static void watchdog_core_data_release(struct device *dev)
890 {
891 	struct watchdog_core_data *wd_data;
892 
893 	wd_data = container_of(dev, struct watchdog_core_data, dev);
894 
895 	kfree(wd_data);
896 }
897 
898 /*
899  *	watchdog_release: release the watchdog device.
900  *	@inode: inode of device
901  *	@file: file handle to device
902  *
903  *	This is the code for when /dev/watchdog gets closed. We will only
904  *	stop the watchdog when we have received the magic char (and nowayout
905  *	was not set), else the watchdog will keep running.
906  */
907 
908 static int watchdog_release(struct inode *inode, struct file *file)
909 {
910 	struct watchdog_core_data *wd_data = file->private_data;
911 	struct watchdog_device *wdd;
912 	int err = -EBUSY;
913 	bool running;
914 
915 	mutex_lock(&wd_data->lock);
916 
917 	wdd = wd_data->wdd;
918 	if (!wdd)
919 		goto done;
920 
921 	/*
922 	 * We only stop the watchdog if we received the magic character
923 	 * or if WDIOF_MAGICCLOSE is not set. If nowayout was set then
924 	 * watchdog_stop will fail.
925 	 */
926 	if (!watchdog_active(wdd))
927 		err = 0;
928 	else if (test_and_clear_bit(_WDOG_ALLOW_RELEASE, &wd_data->status) ||
929 		 !(wdd->info->options & WDIOF_MAGICCLOSE))
930 		err = watchdog_stop(wdd);
931 
932 	/* If the watchdog was not stopped, send a keepalive ping */
933 	if (err < 0) {
934 		pr_crit("watchdog%d: watchdog did not stop!\n", wdd->id);
935 		watchdog_ping(wdd);
936 	}
937 
938 	watchdog_update_worker(wdd);
939 
940 	/* make sure that /dev/watchdog can be re-opened */
941 	clear_bit(_WDOG_DEV_OPEN, &wd_data->status);
942 
943 done:
944 	running = wdd && watchdog_hw_running(wdd);
945 	mutex_unlock(&wd_data->lock);
946 	/*
947 	 * Allow the owner module to be unloaded again unless the watchdog
948 	 * is still running. If the watchdog is still running, it can not
949 	 * be stopped, and its driver must not be unloaded.
950 	 */
951 	if (!running) {
952 		module_put(wd_data->cdev.owner);
953 		put_device(&wd_data->dev);
954 	}
955 	return 0;
956 }
957 
958 static const struct file_operations watchdog_fops = {
959 	.owner		= THIS_MODULE,
960 	.write		= watchdog_write,
961 	.unlocked_ioctl	= watchdog_ioctl,
962 	.compat_ioctl	= compat_ptr_ioctl,
963 	.open		= watchdog_open,
964 	.release	= watchdog_release,
965 };
966 
967 static struct miscdevice watchdog_miscdev = {
968 	.minor		= WATCHDOG_MINOR,
969 	.name		= "watchdog",
970 	.fops		= &watchdog_fops,
971 };
972 
973 static struct class watchdog_class = {
974 	.name =		"watchdog",
975 	.owner =	THIS_MODULE,
976 	.dev_groups =	wdt_groups,
977 };
978 
979 /*
980  *	watchdog_cdev_register: register watchdog character device
981  *	@wdd: watchdog device
982  *
983  *	Register a watchdog character device including handling the legacy
984  *	/dev/watchdog node. /dev/watchdog is actually a miscdevice and
985  *	thus we set it up like that.
986  */
987 
988 static int watchdog_cdev_register(struct watchdog_device *wdd)
989 {
990 	struct watchdog_core_data *wd_data;
991 	int err;
992 
993 	wd_data = kzalloc(sizeof(struct watchdog_core_data), GFP_KERNEL);
994 	if (!wd_data)
995 		return -ENOMEM;
996 	mutex_init(&wd_data->lock);
997 
998 	wd_data->wdd = wdd;
999 	wdd->wd_data = wd_data;
1000 
1001 	if (IS_ERR_OR_NULL(watchdog_kworker)) {
1002 		kfree(wd_data);
1003 		return -ENODEV;
1004 	}
1005 
1006 	device_initialize(&wd_data->dev);
1007 	wd_data->dev.devt = MKDEV(MAJOR(watchdog_devt), wdd->id);
1008 	wd_data->dev.class = &watchdog_class;
1009 	wd_data->dev.parent = wdd->parent;
1010 	wd_data->dev.groups = wdd->groups;
1011 	wd_data->dev.release = watchdog_core_data_release;
1012 	dev_set_drvdata(&wd_data->dev, wdd);
1013 	dev_set_name(&wd_data->dev, "watchdog%d", wdd->id);
1014 
1015 	kthread_init_work(&wd_data->work, watchdog_ping_work);
1016 	hrtimer_init(&wd_data->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
1017 	wd_data->timer.function = watchdog_timer_expired;
1018 	watchdog_hrtimer_pretimeout_init(wdd);
1019 
1020 	if (wdd->id == 0) {
1021 		old_wd_data = wd_data;
1022 		watchdog_miscdev.parent = wdd->parent;
1023 		err = misc_register(&watchdog_miscdev);
1024 		if (err != 0) {
1025 			pr_err("%s: cannot register miscdev on minor=%d (err=%d).\n",
1026 				wdd->info->identity, WATCHDOG_MINOR, err);
1027 			if (err == -EBUSY)
1028 				pr_err("%s: a legacy watchdog module is probably present.\n",
1029 					wdd->info->identity);
1030 			old_wd_data = NULL;
1031 			put_device(&wd_data->dev);
1032 			return err;
1033 		}
1034 	}
1035 
1036 	/* Fill in the data structures */
1037 	cdev_init(&wd_data->cdev, &watchdog_fops);
1038 
1039 	/* Add the device */
1040 	err = cdev_device_add(&wd_data->cdev, &wd_data->dev);
1041 	if (err) {
1042 		pr_err("watchdog%d unable to add device %d:%d\n",
1043 			wdd->id,  MAJOR(watchdog_devt), wdd->id);
1044 		if (wdd->id == 0) {
1045 			misc_deregister(&watchdog_miscdev);
1046 			old_wd_data = NULL;
1047 			put_device(&wd_data->dev);
1048 		}
1049 		return err;
1050 	}
1051 
1052 	wd_data->cdev.owner = wdd->ops->owner;
1053 
1054 	/* Record time of most recent heartbeat as 'just before now'. */
1055 	wd_data->last_hw_keepalive = ktime_sub(ktime_get(), 1);
1056 	watchdog_set_open_deadline(wd_data);
1057 
1058 	/*
1059 	 * If the watchdog is running, prevent its driver from being unloaded,
1060 	 * and schedule an immediate ping.
1061 	 */
1062 	if (watchdog_hw_running(wdd)) {
1063 		__module_get(wdd->ops->owner);
1064 		get_device(&wd_data->dev);
1065 		if (handle_boot_enabled)
1066 			hrtimer_start(&wd_data->timer, 0,
1067 				      HRTIMER_MODE_REL_HARD);
1068 		else
1069 			pr_info("watchdog%d running and kernel based pre-userspace handler disabled\n",
1070 				wdd->id);
1071 	}
1072 
1073 	return 0;
1074 }
1075 
1076 /*
1077  *	watchdog_cdev_unregister: unregister watchdog character device
1078  *	@watchdog: watchdog device
1079  *
1080  *	Unregister watchdog character device and if needed the legacy
1081  *	/dev/watchdog device.
1082  */
1083 
1084 static void watchdog_cdev_unregister(struct watchdog_device *wdd)
1085 {
1086 	struct watchdog_core_data *wd_data = wdd->wd_data;
1087 
1088 	cdev_device_del(&wd_data->cdev, &wd_data->dev);
1089 	if (wdd->id == 0) {
1090 		misc_deregister(&watchdog_miscdev);
1091 		old_wd_data = NULL;
1092 	}
1093 
1094 	if (watchdog_active(wdd) &&
1095 	    test_bit(WDOG_STOP_ON_UNREGISTER, &wdd->status)) {
1096 		watchdog_stop(wdd);
1097 	}
1098 
1099 	mutex_lock(&wd_data->lock);
1100 	wd_data->wdd = NULL;
1101 	wdd->wd_data = NULL;
1102 	mutex_unlock(&wd_data->lock);
1103 
1104 	hrtimer_cancel(&wd_data->timer);
1105 	kthread_cancel_work_sync(&wd_data->work);
1106 	watchdog_hrtimer_pretimeout_stop(wdd);
1107 
1108 	put_device(&wd_data->dev);
1109 }
1110 
1111 /*
1112  *	watchdog_dev_register: register a watchdog device
1113  *	@wdd: watchdog device
1114  *
1115  *	Register a watchdog device including handling the legacy
1116  *	/dev/watchdog node. /dev/watchdog is actually a miscdevice and
1117  *	thus we set it up like that.
1118  */
1119 
1120 int watchdog_dev_register(struct watchdog_device *wdd)
1121 {
1122 	int ret;
1123 
1124 	ret = watchdog_cdev_register(wdd);
1125 	if (ret)
1126 		return ret;
1127 
1128 	ret = watchdog_register_pretimeout(wdd);
1129 	if (ret)
1130 		watchdog_cdev_unregister(wdd);
1131 
1132 	return ret;
1133 }
1134 
1135 /*
1136  *	watchdog_dev_unregister: unregister a watchdog device
1137  *	@watchdog: watchdog device
1138  *
1139  *	Unregister watchdog device and if needed the legacy
1140  *	/dev/watchdog device.
1141  */
1142 
1143 void watchdog_dev_unregister(struct watchdog_device *wdd)
1144 {
1145 	watchdog_unregister_pretimeout(wdd);
1146 	watchdog_cdev_unregister(wdd);
1147 }
1148 
1149 /*
1150  *	watchdog_set_last_hw_keepalive: set last HW keepalive time for watchdog
1151  *	@wdd: watchdog device
1152  *	@last_ping_ms: time since last HW heartbeat
1153  *
1154  *	Adjusts the last known HW keepalive time for a watchdog timer.
1155  *	This is needed if the watchdog is already running when the probe
1156  *	function is called, and it can't be pinged immediately. This
1157  *	function must be called immediately after watchdog registration,
1158  *	and min_hw_heartbeat_ms must be set for this to be useful.
1159  */
1160 int watchdog_set_last_hw_keepalive(struct watchdog_device *wdd,
1161 				   unsigned int last_ping_ms)
1162 {
1163 	struct watchdog_core_data *wd_data;
1164 	ktime_t now;
1165 
1166 	if (!wdd)
1167 		return -EINVAL;
1168 
1169 	wd_data = wdd->wd_data;
1170 
1171 	now = ktime_get();
1172 
1173 	wd_data->last_hw_keepalive = ktime_sub(now, ms_to_ktime(last_ping_ms));
1174 
1175 	return __watchdog_ping(wdd);
1176 }
1177 EXPORT_SYMBOL_GPL(watchdog_set_last_hw_keepalive);
1178 
1179 /*
1180  *	watchdog_dev_init: init dev part of watchdog core
1181  *
1182  *	Allocate a range of chardev nodes to use for watchdog devices
1183  */
1184 
1185 int __init watchdog_dev_init(void)
1186 {
1187 	int err;
1188 
1189 	watchdog_kworker = kthread_create_worker(0, "watchdogd");
1190 	if (IS_ERR(watchdog_kworker)) {
1191 		pr_err("Failed to create watchdog kworker\n");
1192 		return PTR_ERR(watchdog_kworker);
1193 	}
1194 	sched_set_fifo(watchdog_kworker->task);
1195 
1196 	err = class_register(&watchdog_class);
1197 	if (err < 0) {
1198 		pr_err("couldn't register class\n");
1199 		goto err_register;
1200 	}
1201 
1202 	err = alloc_chrdev_region(&watchdog_devt, 0, MAX_DOGS, "watchdog");
1203 	if (err < 0) {
1204 		pr_err("watchdog: unable to allocate char dev region\n");
1205 		goto err_alloc;
1206 	}
1207 
1208 	return 0;
1209 
1210 err_alloc:
1211 	class_unregister(&watchdog_class);
1212 err_register:
1213 	kthread_destroy_worker(watchdog_kworker);
1214 	return err;
1215 }
1216 
1217 /*
1218  *	watchdog_dev_exit: exit dev part of watchdog core
1219  *
1220  *	Release the range of chardev nodes used for watchdog devices
1221  */
1222 
1223 void __exit watchdog_dev_exit(void)
1224 {
1225 	unregister_chrdev_region(watchdog_devt, MAX_DOGS);
1226 	class_unregister(&watchdog_class);
1227 	kthread_destroy_worker(watchdog_kworker);
1228 }
1229 
1230 module_param(handle_boot_enabled, bool, 0444);
1231 MODULE_PARM_DESC(handle_boot_enabled,
1232 	"Watchdog core auto-updates boot enabled watchdogs before userspace takes over (default="
1233 	__MODULE_STRING(IS_ENABLED(CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED)) ")");
1234 
1235 module_param(open_timeout, uint, 0644);
1236 MODULE_PARM_DESC(open_timeout,
1237 	"Maximum time (in seconds, 0 means infinity) for userspace to take over a running watchdog (default="
1238 	__MODULE_STRING(CONFIG_WATCHDOG_OPEN_TIMEOUT) ")");
1239