xref: /freebsd/sys/kern/kern_shutdown.c (revision 77a0943ded95b9e6438f7db70c4a28e4d93946d4)
1 /*-
2  * Copyright (c) 1986, 1988, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_shutdown.c	8.3 (Berkeley) 1/21/94
39  * $FreeBSD$
40  */
41 
42 #include "opt_ddb.h"
43 #include "opt_hw_wdog.h"
44 #include "opt_panic.h"
45 #include "opt_show_busybufs.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/eventhandler.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/reboot.h>
53 #include <sys/proc.h>
54 #include <sys/vnode.h>
55 #include <sys/kernel.h>
56 #include <sys/kthread.h>
57 #include <sys/mount.h>
58 #include <sys/mutex.h>
59 #include <sys/queue.h>
60 #include <sys/sysctl.h>
61 #include <sys/conf.h>
62 #include <sys/sysproto.h>
63 #include <sys/cons.h>
64 
65 #include <machine/pcb.h>
66 #include <machine/lock.h>
67 #include <machine/md_var.h>
68 #include <machine/smp.h>		/* smp_active, cpuid */
69 
70 #include <sys/signalvar.h>
71 
72 #ifndef PANIC_REBOOT_WAIT_TIME
73 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
74 #endif
75 
76 /*
77  * Note that stdarg.h and the ANSI style va_start macro is used for both
78  * ANSI and traditional C compilers.
79  */
80 #include <machine/stdarg.h>
81 
82 #ifdef DDB
83 #ifdef DDB_UNATTENDED
84 int debugger_on_panic = 0;
85 #else
86 int debugger_on_panic = 1;
87 #endif
88 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW,
89 	&debugger_on_panic, 0, "Run debugger on kernel panic");
90 #endif
91 
92 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment");
93 
94 #ifdef	HW_WDOG
95 /*
96  * If there is a hardware watchdog, point this at the function needed to
97  * hold it off.
98  * It's needed when the kernel needs to do some lengthy operations.
99  * e.g. in wd.c when dumping core.. It's most annoying to have
100  * your precious core-dump only half written because the wdog kicked in.
101  */
102 watchdog_tickle_fn wdog_tickler = NULL;
103 #endif	/* HW_WDOG */
104 
105 /*
106  * Variable panicstr contains argument to first call to panic; used as flag
107  * to indicate that the kernel has already called panic.
108  */
109 const char *panicstr;
110 
111 int dumping;				 /* system is dumping */
112 
113 static void boot(int) __dead2;
114 static void dumpsys(void);
115 static void poweroff_wait(void *, int);
116 static void shutdown_halt(void *junk, int howto);
117 static void shutdown_panic(void *junk, int howto);
118 static void shutdown_reset(void *junk, int howto);
119 
120 /* register various local shutdown events */
121 static void
122 shutdown_conf(void *unused)
123 {
124 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, SHUTDOWN_PRI_FIRST);
125 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, SHUTDOWN_PRI_LAST + 100);
126 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, SHUTDOWN_PRI_LAST + 100);
127 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL, SHUTDOWN_PRI_LAST + 200);
128 }
129 
130 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL)
131 
132 /* ARGSUSED */
133 
134 /*
135  * The system call that results in a reboot
136  */
137 int
138 reboot(struct proc *p, struct reboot_args *uap)
139 {
140 	int error;
141 
142 	if ((error = suser(p)))
143 		return (error);
144 
145 	boot(uap->opt);
146 	return (0);
147 }
148 
149 /*
150  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
151  */
152 static int shutdown_howto = 0;
153 
154 void
155 shutdown_nice(int howto)
156 {
157 	shutdown_howto = howto;
158 
159 	/* Send a signal to init(8) and have it shutdown the world */
160 	if (initproc != NULL) {
161 		psignal(initproc, SIGINT);
162 	} else {
163 		/* No init(8) running, so simply reboot */
164 		boot(RB_NOSYNC);
165 	}
166 	return;
167 }
168 static int	waittime = -1;
169 static struct pcb dumppcb;
170 
171 static void
172 print_uptime(void)
173 {
174 	int f;
175 	struct timespec ts;
176 
177 	getnanouptime(&ts);
178 	printf("Uptime: ");
179 	f = 0;
180 	if (ts.tv_sec >= 86400) {
181 		printf("%ldd", (long)ts.tv_sec / 86400);
182 		ts.tv_sec %= 86400;
183 		f = 1;
184 	}
185 	if (f || ts.tv_sec >= 3600) {
186 		printf("%ldh", (long)ts.tv_sec / 3600);
187 		ts.tv_sec %= 3600;
188 		f = 1;
189 	}
190 	if (f || ts.tv_sec >= 60) {
191 		printf("%ldm", (long)ts.tv_sec / 60);
192 		ts.tv_sec %= 60;
193 		f = 1;
194 	}
195 	printf("%lds\n", (long)ts.tv_sec);
196 }
197 
198 /*
199  *  Go through the rigmarole of shutting down..
200  * this used to be in machdep.c but I'll be dammned if I could see
201  * anything machine dependant in it.
202  */
203 static void
204 boot(int howto)
205 {
206 
207 	/* collect extra flags that shutdown_nice might have set */
208 	howto |= shutdown_howto;
209 
210 #ifdef SMP
211 	if (smp_active)
212 		printf("boot() called on cpu#%d\n", cpuid);
213 #endif
214 	/*
215 	 * Do any callouts that should be done BEFORE syncing the filesystems.
216 	 */
217 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
218 
219 	/*
220 	 * Now sync filesystems
221 	 */
222 	if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
223 		register struct buf *bp;
224 		int iter, nbusy, pbusy;
225 		int subiter;
226 
227 		waittime = 0;
228 		printf("\nsyncing disks... ");
229 
230 		sync(&proc0, NULL);
231 
232 		/*
233 		 * With soft updates, some buffers that are
234 		 * written will be remarked as dirty until other
235 		 * buffers are written.
236 		 */
237 		for (iter = pbusy = 0; iter < 20; iter++) {
238 			nbusy = 0;
239 			for (bp = &buf[nbuf]; --bp >= buf; ) {
240 				if ((bp->b_flags & B_INVAL) == 0 &&
241 				    BUF_REFCNT(bp) > 0) {
242 					nbusy++;
243 				} else if ((bp->b_flags & (B_DELWRI | B_INVAL))
244 						== B_DELWRI) {
245 					/* bawrite(bp);*/
246 					nbusy++;
247 				}
248 			}
249 			if (nbusy == 0)
250 				break;
251 			printf("%d ", nbusy);
252 			if (nbusy < pbusy)
253 				iter = 0;
254 			pbusy = nbusy;
255 			sync(&proc0, NULL);
256  			if (curproc != NULL) {
257 				DROP_GIANT_NOSWITCH();
258    				for (subiter = 0; subiter < 50 * iter; subiter++) {
259      					mtx_enter(&sched_lock, MTX_SPIN);
260      					setrunqueue(curproc);
261      					mi_switch(); /* Allow interrupt threads to run */
262      					mtx_exit(&sched_lock, MTX_SPIN);
263      					DELAY(1000);
264    				}
265 				PICKUP_GIANT();
266  			} else
267 			DELAY(50000 * iter);
268 		}
269 		printf("\n");
270 		/*
271 		 * Count only busy local buffers to prevent forcing
272 		 * a fsck if we're just a client of a wedged NFS server
273 		 */
274 		nbusy = 0;
275 		for (bp = &buf[nbuf]; --bp >= buf; ) {
276 			if (((bp->b_flags&B_INVAL) == 0 && BUF_REFCNT(bp)) ||
277 			    ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) {
278 				if (bp->b_dev == NODEV) {
279 					TAILQ_REMOVE(&mountlist,
280 					    bp->b_vp->v_mount, mnt_list);
281 					continue;
282 				}
283 				nbusy++;
284 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC)
285 				printf(
286 			    "%d: dev:%s, flags:%08lx, blkno:%ld, lblkno:%ld\n",
287 				    nbusy, devtoname(bp->b_dev),
288 				    bp->b_flags, (long)bp->b_blkno,
289 				    (long)bp->b_lblkno);
290 #endif
291 			}
292 		}
293 		if (nbusy) {
294 			/*
295 			 * Failed to sync all blocks. Indicate this and don't
296 			 * unmount filesystems (thus forcing an fsck on reboot).
297 			 */
298 			printf("giving up on %d buffers\n", nbusy);
299 			DELAY(5000000);	/* 5 seconds */
300 		} else {
301 			printf("done\n");
302 			/*
303 			 * Unmount filesystems
304 			 */
305 			if (panicstr == 0)
306 				vfs_unmountall();
307 		}
308 		DELAY(100000);		/* wait for console output to finish */
309 	}
310 
311 	print_uptime();
312 
313 	/*
314 	 * Ok, now do things that assume all filesystem activity has
315 	 * been completed.
316 	 */
317 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
318 	splhigh();
319 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold)
320 		dumpsys();
321 
322 	/* Now that we're going to really halt the system... */
323 	EVENTHANDLER_INVOKE(shutdown_final, howto);
324 
325 	for(;;) ;	/* safety against shutdown_reset not working */
326 	/* NOTREACHED */
327 }
328 
329 /*
330  * If the shutdown was a clean halt, behave accordingly.
331  */
332 static void
333 shutdown_halt(void *junk, int howto)
334 {
335 	if (howto & RB_HALT) {
336 		printf("\n");
337 		printf("The operating system has halted.\n");
338 		printf("Please press any key to reboot.\n\n");
339 		switch (cngetc()) {
340 		case -1:		/* No console, just die */
341 			cpu_halt();
342 			/* NOTREACHED */
343 		default:
344 			howto &= ~RB_HALT;
345 			break;
346 		}
347 	}
348 }
349 
350 /*
351  * Check to see if the system paniced, pause and then reboot
352  * according to the specified delay.
353  */
354 static void
355 shutdown_panic(void *junk, int howto)
356 {
357 	int loop;
358 
359 	if (howto & RB_DUMP) {
360 		if (PANIC_REBOOT_WAIT_TIME != 0) {
361 			if (PANIC_REBOOT_WAIT_TIME != -1) {
362 				printf("Automatic reboot in %d seconds - "
363 				       "press a key on the console to abort\n",
364 					PANIC_REBOOT_WAIT_TIME);
365 				for (loop = PANIC_REBOOT_WAIT_TIME * 10;
366 				     loop > 0; --loop) {
367 					DELAY(1000 * 100); /* 1/10th second */
368 					/* Did user type a key? */
369 					if (cncheckc() != -1)
370 						break;
371 				}
372 				if (!loop)
373 					return;
374 			}
375 		} else { /* zero time specified - reboot NOW */
376 			return;
377 		}
378 		printf("--> Press a key on the console to reboot <--\n");
379 		cngetc();
380 	}
381 }
382 
383 /*
384  * Everything done, now reset
385  */
386 static void
387 shutdown_reset(void *junk, int howto)
388 {
389 	printf("Rebooting...\n");
390 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
391 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
392 	cpu_reset();
393 	/* NOTREACHED */ /* assuming reset worked */
394 }
395 
396 /*
397  * Magic number for savecore
398  *
399  * exported (symorder) and used at least by savecore(8)
400  *
401  */
402 static u_long const	dumpmag = 0x8fca0101UL;
403 
404 static int	dumpsize = 0;		/* also for savecore */
405 
406 static int	dodump = 1;
407 
408 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0,
409     "Try to perform coredump on kernel panic");
410 
411 static int
412 setdumpdev(dev_t dev)
413 {
414 	int psize;
415 	long newdumplo;
416 
417 	if (dev == NODEV) {
418 		dumpdev = dev;
419 		return (0);
420 	}
421 	if (devsw(dev) == NULL)
422 		return (ENXIO);		/* XXX is this right? */
423 	if (devsw(dev)->d_psize == NULL)
424 		return (ENXIO);		/* XXX should be ENODEV ? */
425 	psize = devsw(dev)->d_psize(dev);
426 	if (psize == -1)
427 		return (ENXIO);		/* XXX should be ENODEV ? */
428 	/*
429 	 * XXX should clean up checking in dumpsys() to be more like this.
430 	 */
431 	newdumplo = psize - Maxmem * PAGE_SIZE / DEV_BSIZE;
432 	if (newdumplo < 0)
433 		return (ENOSPC);
434 	dumpdev = dev;
435 	dumplo = newdumplo;
436 	return (0);
437 }
438 
439 
440 /* ARGSUSED */
441 static void
442 dump_conf(void *dummy)
443 {
444 	if (setdumpdev(dumpdev) != 0)
445 		dumpdev = NODEV;
446 }
447 
448 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL)
449 
450 static int
451 sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS)
452 {
453 	int error;
454 	udev_t ndumpdev;
455 
456 	ndumpdev = dev2udev(dumpdev);
457 	error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req);
458 	if (error == 0 && req->newptr != NULL)
459 		error = setdumpdev(udev2dev(ndumpdev, 0));
460 	return (error);
461 }
462 
463 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW,
464 	0, sizeof dumpdev, sysctl_kern_dumpdev, "T,dev_t", "");
465 
466 /*
467  * Doadump comes here after turning off memory management and
468  * getting on the dump stack, either when called above, or by
469  * the auto-restart code.
470  */
471 static void
472 dumpsys(void)
473 {
474 	int	error;
475 
476 	savectx(&dumppcb);
477 	if (dumping++) {
478 		printf("Dump already in progress, bailing...\n");
479 		return;
480 	}
481 	if (!dodump)
482 		return;
483 	if (dumpdev == NODEV)
484 		return;
485 	if (!(devsw(dumpdev)))
486 		return;
487 	if (!(devsw(dumpdev)->d_dump))
488 		return;
489 	dumpsize = Maxmem;
490 	printf("\ndumping to dev %s, offset %ld\n", devtoname(dumpdev), dumplo);
491 	printf("dump ");
492 	error = (*devsw(dumpdev)->d_dump)(dumpdev);
493 	if (error == 0) {
494 		printf("succeeded\n");
495 		return;
496 	}
497 	printf("failed, reason: ");
498 	switch (error) {
499 	case ENODEV:
500 		printf("device doesn't support a dump routine\n");
501 		break;
502 
503 	case ENXIO:
504 		printf("device bad\n");
505 		break;
506 
507 	case EFAULT:
508 		printf("device not ready\n");
509 		break;
510 
511 	case EINVAL:
512 		printf("area improper\n");
513 		break;
514 
515 	case EIO:
516 		printf("i/o error\n");
517 		break;
518 
519 	case EINTR:
520 		printf("aborted from console\n");
521 		break;
522 
523 	default:
524 		printf("unknown, error = %d\n", error);
525 		break;
526 	}
527 }
528 
529 /*
530  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
531  * and then reboots.  If we are called twice, then we avoid trying to sync
532  * the disks as this often leads to recursive panics.
533  */
534 void
535 panic(const char *fmt, ...)
536 {
537 	int bootopt;
538 	va_list ap;
539 	static char buf[256];
540 
541 #ifdef SMP
542 	/* Only 1 CPU can panic at a time */
543 	s_lock(&panic_lock);
544 #endif
545 
546 	bootopt = RB_AUTOBOOT | RB_DUMP;
547 	if (panicstr)
548 		bootopt |= RB_NOSYNC;
549 	else
550 		panicstr = fmt;
551 
552 	va_start(ap, fmt);
553 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
554 	if (panicstr == fmt)
555 		panicstr = buf;
556 	va_end(ap);
557 	printf("panic: %s\n", buf);
558 #ifdef SMP
559 	/* two seperate prints in case of an unmapped page and trap */
560 	printf("cpuid = %d; ", cpuid);
561 	printf("lapic.id = %08x\n", lapic.id);
562 #endif
563 
564 #if defined(DDB)
565 	if (debugger_on_panic)
566 		Debugger ("panic");
567 #endif
568 	boot(bootopt);
569 }
570 
571 /*
572  * Support for poweroff delay.
573  */
574 #ifndef POWEROFF_DELAY
575 # define POWEROFF_DELAY 5000
576 #endif
577 static int poweroff_delay = POWEROFF_DELAY;
578 
579 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
580 	&poweroff_delay, 0, "");
581 
582 static void
583 poweroff_wait(void *junk, int howto)
584 {
585 	if(!(howto & RB_POWEROFF) || poweroff_delay <= 0)
586 		return;
587 	DELAY(poweroff_delay * 1000);
588 }
589 
590 /*
591  * Some system processes (e.g. syncer) need to be stopped at appropriate
592  * points in their main loops prior to a system shutdown, so that they
593  * won't interfere with the shutdown process (e.g. by holding a disk buf
594  * to cause sync to fail).  For each of these system processes, register
595  * shutdown_kproc() as a handler for one of shutdown events.
596  */
597 static int kproc_shutdown_wait = 60;
598 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
599     &kproc_shutdown_wait, 0, "");
600 
601 void
602 shutdown_kproc(void *arg, int howto)
603 {
604 	struct proc *p;
605 	int error;
606 
607 	if (panicstr)
608 		return;
609 
610 	p = (struct proc *)arg;
611 	printf("Waiting (max %d seconds) for system process `%s' to stop...",
612 	    kproc_shutdown_wait, p->p_comm);
613 	error = suspend_kproc(p, kproc_shutdown_wait * hz);
614 
615 	if (error == EWOULDBLOCK)
616 		printf("timed out\n");
617 	else
618 		printf("stopped\n");
619 }
620