xref: /freebsd/sys/kern/kern_shutdown.c (revision daf1cffce2e07931f27c6c6998652e90df6ba87e)
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/buf.h>
51 #include <sys/reboot.h>
52 #include <sys/proc.h>
53 #include <sys/vnode.h>
54 #include <sys/kernel.h>
55 #include <sys/kthread.h>
56 #include <sys/mount.h>
57 #include <sys/queue.h>
58 #include <sys/sysctl.h>
59 #include <sys/conf.h>
60 #include <sys/sysproto.h>
61 #include <sys/cons.h>
62 
63 #include <machine/pcb.h>
64 #include <machine/clock.h>
65 #include <machine/md_var.h>
66 #ifdef SMP
67 #include <machine/smp.h>		/* smp_active, cpuid */
68 #endif
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 static void boot __P((int)) __dead2;
112 static void dumpsys __P((void));
113 static int setdumpdev __P((dev_t dev));
114 static void poweroff_wait __P((void *, int));
115 static void print_uptime __P((void));
116 static void shutdown_halt __P((void *junk, int howto));
117 static void shutdown_panic __P((void *junk, int howto));
118 static void shutdown_reset __P((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(p, uap)
139 	struct proc *p;
140 	struct reboot_args *uap;
141 {
142 	int error;
143 
144 	if ((error = suser(p)))
145 		return (error);
146 
147 	boot(uap->opt);
148 	return (0);
149 }
150 
151 /*
152  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
153  */
154 void
155 shutdown_nice()
156 {
157 	/* Send a signal to init(8) and have it shutdown the world */
158 	if (initproc != NULL) {
159 		psignal(initproc, SIGINT);
160 	} else {
161 		/* No init(8) running, so simply reboot */
162 		boot(RB_NOSYNC);
163 	}
164 	return;
165 }
166 static int	waittime = -1;
167 static struct pcb dumppcb;
168 
169 static void
170 print_uptime()
171 {
172 	int f;
173 	struct timespec ts;
174 
175 	getnanouptime(&ts);
176 	printf("Uptime: ");
177 	f = 0;
178 	if (ts.tv_sec >= 86400) {
179 		printf("%ldd", ts.tv_sec / 86400);
180 		ts.tv_sec %= 86400;
181 		f = 1;
182 	}
183 	if (f || ts.tv_sec >= 3600) {
184 		printf("%ldh", ts.tv_sec / 3600);
185 		ts.tv_sec %= 3600;
186 		f = 1;
187 	}
188 	if (f || ts.tv_sec >= 60) {
189 		printf("%ldm", ts.tv_sec / 60);
190 		ts.tv_sec %= 60;
191 		f = 1;
192 	}
193 	printf("%lds\n", ts.tv_sec);
194 }
195 
196 /*
197  *  Go through the rigmarole of shutting down..
198  * this used to be in machdep.c but I'll be dammned if I could see
199  * anything machine dependant in it.
200  */
201 static void
202 boot(howto)
203 	int howto;
204 {
205 
206 #ifdef SMP
207 	if (smp_active) {
208 		printf("boot() called on cpu#%d\n", cpuid);
209 	}
210 #endif
211 	/*
212 	 * Do any callouts that should be done BEFORE syncing the filesystems.
213 	 */
214 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
215 
216 	/*
217 	 * Now sync filesystems
218 	 */
219 	if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
220 		register struct buf *bp;
221 		int iter, nbusy;
222 
223 		waittime = 0;
224 		printf("\nsyncing disks... ");
225 
226 		sync(&proc0, NULL);
227 
228 		/*
229 		 * With soft updates, some buffers that are
230 		 * written will be remarked as dirty until other
231 		 * buffers are written.
232 		 */
233 		for (iter = 0; iter < 20; iter++) {
234 			nbusy = 0;
235 			for (bp = &buf[nbuf]; --bp >= buf; ) {
236 				if ((bp->b_flags & B_INVAL) == 0 &&
237 				    BUF_REFCNT(bp) > 0) {
238 					nbusy++;
239 				} else if ((bp->b_flags & (B_DELWRI | B_INVAL))
240 						== B_DELWRI) {
241 					/* bawrite(bp);*/
242 					nbusy++;
243 				}
244 			}
245 			if (nbusy == 0)
246 				break;
247 			printf("%d ", nbusy);
248 			sync(&proc0, NULL);
249 			DELAY(50000 * iter);
250 		}
251 		printf("\n");
252 		/*
253 		 * Count only busy local buffers to prevent forcing
254 		 * a fsck if we're just a client of a wedged NFS server
255 		 */
256 		nbusy = 0;
257 		for (bp = &buf[nbuf]; --bp >= buf; ) {
258 			if (((bp->b_flags&B_INVAL) == 0 && BUF_REFCNT(bp)) ||
259 			    ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) {
260 				if (bp->b_dev == NODEV) {
261 					TAILQ_REMOVE(&mountlist,
262 					    bp->b_vp->v_mount, mnt_list);
263 					continue;
264 				}
265 				nbusy++;
266 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC)
267 				printf(
268 			    "%d: dev:%s, flags:%08lx, blkno:%ld, lblkno:%ld\n",
269 				    nbusy, devtoname(bp->b_dev),
270 				    bp->b_flags, (long)bp->b_blkno,
271 				    (long)bp->b_lblkno);
272 #endif
273 			}
274 		}
275 		if (nbusy) {
276 			/*
277 			 * Failed to sync all blocks. Indicate this and don't
278 			 * unmount filesystems (thus forcing an fsck on reboot).
279 			 */
280 			printf("giving up on %d buffers\n", nbusy);
281 			DELAY(5000000);	/* 5 seconds */
282 		} else {
283 			printf("done\n");
284 			/*
285 			 * Unmount filesystems
286 			 */
287 			if (panicstr == 0)
288 				vfs_unmountall();
289 		}
290 		DELAY(100000);		/* wait for console output to finish */
291 	}
292 
293 	print_uptime();
294 
295 	/*
296 	 * Ok, now do things that assume all filesystem activity has
297 	 * been completed.
298 	 */
299 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
300 	splhigh();
301 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold) {
302 		savectx(&dumppcb);
303 #ifdef __i386__
304 		dumppcb.pcb_cr3 = rcr3();
305 #endif
306 		dumpsys();
307 	}
308 
309 	/* Now that we're going to really halt the system... */
310 	EVENTHANDLER_INVOKE(shutdown_final, howto);
311 
312 	for(;;) ;	/* safety against shutdown_reset not working */
313 	/* NOTREACHED */
314 }
315 
316 /*
317  * If the shutdown was a clean halt, behave accordingly.
318  */
319 static void
320 shutdown_halt(void *junk, int howto)
321 {
322 	if (howto & RB_HALT) {
323 		printf("\n");
324 		printf("The operating system has halted.\n");
325 		printf("Please press any key to reboot.\n\n");
326 		switch (cngetc()) {
327 		case -1:		/* No console, just die */
328 			cpu_halt();
329 			/* NOTREACHED */
330 		default:
331 			howto &= ~RB_HALT;
332 			break;
333 		}
334 	}
335 }
336 
337 /*
338  * Check to see if the system paniced, pause and then reboot
339  * according to the specified delay.
340  */
341 static void
342 shutdown_panic(void *junk, int howto)
343 {
344 	int loop;
345 
346 	if (howto & RB_DUMP) {
347 		if (PANIC_REBOOT_WAIT_TIME != 0) {
348 			if (PANIC_REBOOT_WAIT_TIME != -1) {
349 				printf("Automatic reboot in %d seconds - "
350 				       "press a key on the console to abort\n",
351 					PANIC_REBOOT_WAIT_TIME);
352 				for (loop = PANIC_REBOOT_WAIT_TIME * 10;
353 				     loop > 0; --loop) {
354 					DELAY(1000 * 100); /* 1/10th second */
355 					/* Did user type a key? */
356 					if (cncheckc() != -1)
357 						break;
358 				}
359 				if (!loop)
360 					return;
361 			}
362 		} else { /* zero time specified - reboot NOW */
363 			return;
364 		}
365 		printf("--> Press a key on the console to reboot <--\n");
366 		cngetc();
367 	}
368 }
369 
370 /*
371  * Everything done, now reset
372  */
373 static void
374 shutdown_reset(void *junk, int howto)
375 {
376 	printf("Rebooting...\n");
377 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
378 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
379 	cpu_reset();
380 	/* NOTREACHED */ /* assuming reset worked */
381 }
382 
383 /*
384  * Magic number for savecore
385  *
386  * exported (symorder) and used at least by savecore(8)
387  *
388  */
389 static u_long const	dumpmag = 0x8fca0101UL;
390 
391 static int	dumpsize = 0;		/* also for savecore */
392 
393 static int	dodump = 1;
394 
395 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0,
396     "Try to perform coredump on kernel panic");
397 
398 static int
399 setdumpdev(dev)
400 	dev_t dev;
401 {
402 	int psize;
403 	long newdumplo;
404 
405 	if (dev == NODEV) {
406 		dumpdev = dev;
407 		return (0);
408 	}
409 	if (devsw(dev) == NULL)
410 		return (ENXIO);		/* XXX is this right? */
411 	if (devsw(dev)->d_psize == NULL)
412 		return (ENXIO);		/* XXX should be ENODEV ? */
413 	psize = devsw(dev)->d_psize(dev);
414 	if (psize == -1)
415 		return (ENXIO);		/* XXX should be ENODEV ? */
416 	/*
417 	 * XXX should clean up checking in dumpsys() to be more like this.
418 	 */
419 	newdumplo = psize - Maxmem * PAGE_SIZE / DEV_BSIZE;
420 	if (newdumplo < 0)
421 		return (ENOSPC);
422 	dumpdev = dev;
423 	dumplo = newdumplo;
424 	return (0);
425 }
426 
427 
428 /* ARGSUSED */
429 static void dump_conf __P((void *dummy));
430 static void
431 dump_conf(dummy)
432 	void *dummy;
433 {
434 	if (setdumpdev(dumpdev) != 0)
435 		dumpdev = NODEV;
436 }
437 
438 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL)
439 
440 static int
441 sysctl_kern_dumpdev SYSCTL_HANDLER_ARGS
442 {
443 	int error;
444 	udev_t ndumpdev;
445 
446 	ndumpdev = dev2udev(dumpdev);
447 	error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req);
448 	if (error == 0 && req->newptr != NULL)
449 		error = setdumpdev(udev2dev(ndumpdev, 0));
450 	return (error);
451 }
452 
453 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW,
454 	0, sizeof dumpdev, sysctl_kern_dumpdev, "T,dev_t", "");
455 
456 /*
457  * Doadump comes here after turning off memory management and
458  * getting on the dump stack, either when called above, or by
459  * the auto-restart code.
460  */
461 static void
462 dumpsys(void)
463 {
464 	int	error;
465 	static int dumping;
466 
467 	if (dumping++) {
468 		printf("Dump already in progress, bailing...\n");
469 		return;
470 	}
471 	if (!dodump)
472 		return;
473 	if (dumpdev == NODEV)
474 		return;
475 	if (!(devsw(dumpdev)))
476 		return;
477 	if (!(devsw(dumpdev)->d_dump))
478 		return;
479 	dumpsize = Maxmem;
480 	printf("\ndumping to dev %s, offset %ld\n", devtoname(dumpdev), dumplo);
481 	printf("dump ");
482 	error = (*devsw(dumpdev)->d_dump)(dumpdev);
483 	if (error == 0) {
484 		printf("succeeded\n");
485 		return;
486 	}
487 	printf("failed, reason: ");
488 	switch (error) {
489 	case ENODEV:
490 		printf("device doesn't support a dump routine\n");
491 		break;
492 
493 	case ENXIO:
494 		printf("device bad\n");
495 		break;
496 
497 	case EFAULT:
498 		printf("device not ready\n");
499 		break;
500 
501 	case EINVAL:
502 		printf("area improper\n");
503 		break;
504 
505 	case EIO:
506 		printf("i/o error\n");
507 		break;
508 
509 	case EINTR:
510 		printf("aborted from console\n");
511 		break;
512 
513 	default:
514 		printf("unknown, error = %d\n", error);
515 		break;
516 	}
517 }
518 
519 /*
520  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
521  * and then reboots.  If we are called twice, then we avoid trying to sync
522  * the disks as this often leads to recursive panics.
523  */
524 void
525 panic(const char *fmt, ...)
526 {
527 	int bootopt;
528 	va_list ap;
529 	static char buf[256];
530 
531 	bootopt = RB_AUTOBOOT | RB_DUMP;
532 	if (panicstr)
533 		bootopt |= RB_NOSYNC;
534 	else
535 		panicstr = fmt;
536 
537 	va_start(ap, fmt);
538 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
539 	if (panicstr == fmt)
540 		panicstr = buf;
541 	va_end(ap);
542 	printf("panic: %s\n", buf);
543 #ifdef SMP
544 	/* three seperate prints in case of an unmapped page and trap */
545 	printf("mp_lock = %08x; ", mp_lock);
546 	printf("cpuid = %d; ", cpuid);
547 	printf("lapic.id = %08x\n", lapic.id);
548 #endif
549 
550 #if defined(DDB)
551 	if (debugger_on_panic)
552 		Debugger ("panic");
553 #endif
554 	boot(bootopt);
555 }
556 
557 /*
558  * Support for poweroff delay.
559  */
560 #ifndef POWEROFF_DELAY
561 # define POWEROFF_DELAY 5000
562 #endif
563 static int poweroff_delay = POWEROFF_DELAY;
564 
565 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
566 	&poweroff_delay, 0, "");
567 
568 static void
569 poweroff_wait(void *junk, int howto)
570 {
571 	if(!(howto & RB_POWEROFF) || poweroff_delay <= 0)
572 		return;
573 	DELAY(poweroff_delay * 1000);
574 }
575 
576 /*
577  * Some system processes (e.g. syncer) need to be stopped at appropriate
578  * points in their main loops prior to a system shutdown, so that they
579  * won't interfere with the shutdown process (e.g. by holding a disk buf
580  * to cause sync to fail).  For each of these system processes, register
581  * shutdown_kproc() as a handler for one of shutdown events.
582  */
583 static int kproc_shutdown_wait = 60;
584 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
585     &kproc_shutdown_wait, 0, "");
586 
587 void
588 shutdown_kproc(void *arg, int howto)
589 {
590 	struct proc *p;
591 	int error;
592 
593 	if (panicstr)
594 		return;
595 
596 	p = (struct proc *)arg;
597 	printf("Waiting (max %d seconds) for system process `%s' to stop...",
598 	    kproc_shutdown_wait, p->p_comm);
599 	error = suspend_kproc(p, kproc_shutdown_wait * hz);
600 
601 	if (error == EWOULDBLOCK)
602 		printf("timed out\n");
603 	else
604 		printf("stopped\n");
605 }
606