xref: /freebsd/sys/kern/kern_shutdown.c (revision 5521ff5a4d1929056e7ffc982fac3341ca54df7c)
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/bio.h>
50 #include <sys/buf.h>
51 #include <sys/conf.h>
52 #include <sys/cons.h>
53 #include <sys/eventhandler.h>
54 #include <sys/kernel.h>
55 #include <sys/kthread.h>
56 #include <sys/mount.h>
57 #include <sys/proc.h>
58 #include <sys/reboot.h>
59 #include <sys/resourcevar.h>
60 #include <sys/smp.h>		/* smp_active */
61 #include <sys/sysctl.h>
62 #include <sys/sysproto.h>
63 #include <sys/vnode.h>
64 
65 #include <machine/pcb.h>
66 #include <machine/md_var.h>
67 
68 #include <sys/signalvar.h>
69 
70 #ifndef PANIC_REBOOT_WAIT_TIME
71 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
72 #endif
73 
74 /*
75  * Note that stdarg.h and the ANSI style va_start macro is used for both
76  * ANSI and traditional C compilers.
77  */
78 #include <machine/stdarg.h>
79 
80 #ifdef DDB
81 #ifdef DDB_UNATTENDED
82 int debugger_on_panic = 0;
83 #else
84 int debugger_on_panic = 1;
85 #endif
86 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW,
87 	&debugger_on_panic, 0, "Run debugger on kernel panic");
88 #endif
89 
90 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment");
91 
92 #ifdef	HW_WDOG
93 /*
94  * If there is a hardware watchdog, point this at the function needed to
95  * hold it off.
96  * It's needed when the kernel needs to do some lengthy operations.
97  * e.g. in wd.c when dumping core.. It's most annoying to have
98  * your precious core-dump only half written because the wdog kicked in.
99  */
100 watchdog_tickle_fn wdog_tickler = NULL;
101 #endif	/* HW_WDOG */
102 
103 /*
104  * Variable panicstr contains argument to first call to panic; used as flag
105  * to indicate that the kernel has already called panic.
106  */
107 const char *panicstr;
108 
109 int dumping;				 /* system is dumping */
110 
111 static void boot(int) __dead2;
112 static void dumpsys(void);
113 static void poweroff_wait(void *, int);
114 static void shutdown_halt(void *junk, int howto);
115 static void shutdown_panic(void *junk, int howto);
116 static void shutdown_reset(void *junk, int howto);
117 
118 /* register various local shutdown events */
119 static void
120 shutdown_conf(void *unused)
121 {
122 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, SHUTDOWN_PRI_FIRST);
123 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, SHUTDOWN_PRI_LAST + 100);
124 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, SHUTDOWN_PRI_LAST + 100);
125 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL, SHUTDOWN_PRI_LAST + 200);
126 }
127 
128 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL)
129 
130 /* ARGSUSED */
131 
132 /*
133  * The system call that results in a reboot
134  */
135 int
136 reboot(struct proc *p, struct reboot_args *uap)
137 {
138 	int error;
139 
140 	if ((error = suser(p)))
141 		return (error);
142 
143 	boot(uap->opt);
144 	return (0);
145 }
146 
147 /*
148  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
149  */
150 static int shutdown_howto = 0;
151 
152 void
153 shutdown_nice(int howto)
154 {
155 	shutdown_howto = howto;
156 
157 	/* Send a signal to init(8) and have it shutdown the world */
158 	if (initproc != NULL) {
159 		PROC_LOCK(initproc);
160 		psignal(initproc, SIGINT);
161 		PROC_UNLOCK(initproc);
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", PCPU_GET(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_lock_spin(&sched_lock);
260      					setrunqueue(curproc);
261 					curproc->p_stats->p_ru.ru_nvcsw++;
262      					mi_switch(); /* Allow interrupt threads to run */
263      					mtx_unlock_spin(&sched_lock);
264      					DELAY(1000);
265    				}
266 				PICKUP_GIANT();
267  			} else
268 			DELAY(50000 * iter);
269 		}
270 		printf("\n");
271 		/*
272 		 * Count only busy local buffers to prevent forcing
273 		 * a fsck if we're just a client of a wedged NFS server
274 		 */
275 		nbusy = 0;
276 		for (bp = &buf[nbuf]; --bp >= buf; ) {
277 			if (((bp->b_flags&B_INVAL) == 0 && BUF_REFCNT(bp)) ||
278 			    ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) {
279 				if (bp->b_dev == NODEV) {
280 					TAILQ_REMOVE(&mountlist,
281 					    bp->b_vp->v_mount, mnt_list);
282 					continue;
283 				}
284 				nbusy++;
285 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC)
286 				printf(
287 			    "%d: dev:%s, flags:%08lx, blkno:%ld, lblkno:%ld\n",
288 				    nbusy, devtoname(bp->b_dev),
289 				    bp->b_flags, (long)bp->b_blkno,
290 				    (long)bp->b_lblkno);
291 #endif
292 			}
293 		}
294 		if (nbusy) {
295 			/*
296 			 * Failed to sync all blocks. Indicate this and don't
297 			 * unmount filesystems (thus forcing an fsck on reboot).
298 			 */
299 			printf("giving up on %d buffers\n", nbusy);
300 			DELAY(5000000);	/* 5 seconds */
301 		} else {
302 			printf("done\n");
303 			/*
304 			 * Unmount filesystems
305 			 */
306 			if (panicstr == 0)
307 				vfs_unmountall();
308 		}
309 		DELAY(100000);		/* wait for console output to finish */
310 	}
311 
312 	print_uptime();
313 
314 	/*
315 	 * Ok, now do things that assume all filesystem activity has
316 	 * been completed.
317 	 */
318 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
319 	splhigh();
320 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold)
321 		dumpsys();
322 
323 	/* Now that we're going to really halt the system... */
324 	EVENTHANDLER_INVOKE(shutdown_final, howto);
325 
326 	for(;;) ;	/* safety against shutdown_reset not working */
327 	/* NOTREACHED */
328 }
329 
330 /*
331  * If the shutdown was a clean halt, behave accordingly.
332  */
333 static void
334 shutdown_halt(void *junk, int howto)
335 {
336 	if (howto & RB_HALT) {
337 		printf("\n");
338 		printf("The operating system has halted.\n");
339 		printf("Please press any key to reboot.\n\n");
340 		switch (cngetc()) {
341 		case -1:		/* No console, just die */
342 			cpu_halt();
343 			/* NOTREACHED */
344 		default:
345 			howto &= ~RB_HALT;
346 			break;
347 		}
348 	}
349 }
350 
351 /*
352  * Check to see if the system paniced, pause and then reboot
353  * according to the specified delay.
354  */
355 static void
356 shutdown_panic(void *junk, int howto)
357 {
358 	int loop;
359 
360 	if (howto & RB_DUMP) {
361 		if (PANIC_REBOOT_WAIT_TIME != 0) {
362 			if (PANIC_REBOOT_WAIT_TIME != -1) {
363 				printf("Automatic reboot in %d seconds - "
364 				       "press a key on the console to abort\n",
365 					PANIC_REBOOT_WAIT_TIME);
366 				for (loop = PANIC_REBOOT_WAIT_TIME * 10;
367 				     loop > 0; --loop) {
368 					DELAY(1000 * 100); /* 1/10th second */
369 					/* Did user type a key? */
370 					if (cncheckc() != -1)
371 						break;
372 				}
373 				if (!loop)
374 					return;
375 			}
376 		} else { /* zero time specified - reboot NOW */
377 			return;
378 		}
379 		printf("--> Press a key on the console to reboot <--\n");
380 		cngetc();
381 	}
382 }
383 
384 /*
385  * Everything done, now reset
386  */
387 static void
388 shutdown_reset(void *junk, int howto)
389 {
390 	printf("Rebooting...\n");
391 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
392 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
393 	cpu_reset();
394 	/* NOTREACHED */ /* assuming reset worked */
395 }
396 
397 /*
398  * Magic number for savecore
399  *
400  * exported (symorder) and used at least by savecore(8)
401  *
402  */
403 static u_long const	dumpmag = 0x8fca0101UL;
404 
405 static int	dumpsize = 0;		/* also for savecore */
406 
407 static int	dodump = 1;
408 
409 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0,
410     "Try to perform coredump on kernel panic");
411 
412 static int
413 setdumpdev(dev_t dev)
414 {
415 	int psize;
416 	long newdumplo;
417 
418 	if (dev == NODEV) {
419 		dumpdev = dev;
420 		return (0);
421 	}
422 	if (devsw(dev) == NULL)
423 		return (ENXIO);		/* XXX is this right? */
424 	if (devsw(dev)->d_psize == NULL)
425 		return (ENXIO);		/* XXX should be ENODEV ? */
426 	psize = devsw(dev)->d_psize(dev);
427 	if (psize == -1)
428 		return (ENXIO);		/* XXX should be ENODEV ? */
429 	/*
430 	 * XXX should clean up checking in dumpsys() to be more like this.
431 	 */
432 	newdumplo = psize - Maxmem * PAGE_SIZE / DEV_BSIZE;
433 	if (newdumplo < 0)
434 		return (ENOSPC);
435 	dumpdev = dev;
436 	dumplo = newdumplo;
437 	return (0);
438 }
439 
440 
441 /* ARGSUSED */
442 static void
443 dump_conf(void *dummy)
444 {
445 	if (setdumpdev(dumpdev) != 0)
446 		dumpdev = NODEV;
447 }
448 
449 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL)
450 
451 static int
452 sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS)
453 {
454 	int error;
455 	udev_t ndumpdev;
456 
457 	ndumpdev = dev2udev(dumpdev);
458 	error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req);
459 	if (error == 0 && req->newptr != NULL)
460 		error = setdumpdev(udev2dev(ndumpdev, 0));
461 	return (error);
462 }
463 
464 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW,
465 	0, sizeof dumpdev, sysctl_kern_dumpdev, "T,dev_t", "");
466 
467 /*
468  * Doadump comes here after turning off memory management and
469  * getting on the dump stack, either when called above, or by
470  * the auto-restart code.
471  */
472 static void
473 dumpsys(void)
474 {
475 	int	error;
476 
477 	savectx(&dumppcb);
478 	if (dumping++) {
479 		printf("Dump already in progress, bailing...\n");
480 		return;
481 	}
482 	if (!dodump)
483 		return;
484 	if (dumpdev == NODEV)
485 		return;
486 	if (!(devsw(dumpdev)))
487 		return;
488 	if (!(devsw(dumpdev)->d_dump))
489 		return;
490 	dumpsize = Maxmem;
491 	printf("\ndumping to dev %s, offset %ld\n", devtoname(dumpdev), dumplo);
492 	printf("dump ");
493 	error = (*devsw(dumpdev)->d_dump)(dumpdev);
494 	if (error == 0) {
495 		printf("succeeded\n");
496 		return;
497 	}
498 	printf("failed, reason: ");
499 	switch (error) {
500 	case ENODEV:
501 		printf("device doesn't support a dump routine\n");
502 		break;
503 
504 	case ENXIO:
505 		printf("device bad\n");
506 		break;
507 
508 	case EFAULT:
509 		printf("device not ready\n");
510 		break;
511 
512 	case EINVAL:
513 		printf("area improper\n");
514 		break;
515 
516 	case EIO:
517 		printf("i/o error\n");
518 		break;
519 
520 	case EINTR:
521 		printf("aborted from console\n");
522 		break;
523 
524 	default:
525 		printf("unknown, error = %d\n", error);
526 		break;
527 	}
528 }
529 
530 int
531 dumpstatus(vm_offset_t addr, long count)
532 {
533 	int c;
534 
535 	if (addr % (1024 * 1024) == 0) {
536 #ifdef HW_WDOG
537 		if (wdog_tickler)
538 			(*wdog_tickler)();
539 #endif
540 		printf("%ld ", count / (1024 * 1024));
541 	}
542 
543 	if ((c = cncheckc()) == 0x03)
544 		return -1;
545 	else if (c != -1)
546 		printf("[CTRL-C to abort] ");
547 
548 	return 0;
549 }
550 
551 #ifdef SMP
552 static u_int panic_cpu = NOCPU;
553 #endif
554 
555 /*
556  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
557  * and then reboots.  If we are called twice, then we avoid trying to sync
558  * the disks as this often leads to recursive panics.
559  */
560 void
561 panic(const char *fmt, ...)
562 {
563 	int bootopt;
564 	va_list ap;
565 	static char buf[256];
566 
567 #ifdef SMP
568 	/* Only 1 CPU can panic at a time */
569 	if (panic_cpu != PCPU_GET(cpuid) &&
570 	    atomic_cmpset_int(&panic_cpu, NOCPU, PCPU_GET(cpuid)) == 0) {
571 		for (;;)
572 			; /* nothing */
573 	}
574 #endif
575 
576 	bootopt = RB_AUTOBOOT | RB_DUMP;
577 	if (panicstr)
578 		bootopt |= RB_NOSYNC;
579 	else
580 		panicstr = fmt;
581 
582 	va_start(ap, fmt);
583 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
584 	if (panicstr == fmt)
585 		panicstr = buf;
586 	va_end(ap);
587 	printf("panic: %s\n", buf);
588 #ifdef SMP
589 	/* two separate prints in case of an unmapped page and trap */
590 	printf("cpuid = %d; ", PCPU_GET(cpuid));
591 #ifdef APIC_IO
592 	printf("lapic.id = %08x\n", lapic.id);
593 #endif
594 #endif
595 
596 #if defined(DDB)
597 	if (debugger_on_panic)
598 		Debugger ("panic");
599 #endif
600 	boot(bootopt);
601 }
602 
603 /*
604  * Support for poweroff delay.
605  */
606 #ifndef POWEROFF_DELAY
607 # define POWEROFF_DELAY 5000
608 #endif
609 static int poweroff_delay = POWEROFF_DELAY;
610 
611 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
612 	&poweroff_delay, 0, "");
613 
614 static void
615 poweroff_wait(void *junk, int howto)
616 {
617 	if(!(howto & RB_POWEROFF) || poweroff_delay <= 0)
618 		return;
619 	DELAY(poweroff_delay * 1000);
620 }
621 
622 /*
623  * Some system processes (e.g. syncer) need to be stopped at appropriate
624  * points in their main loops prior to a system shutdown, so that they
625  * won't interfere with the shutdown process (e.g. by holding a disk buf
626  * to cause sync to fail).  For each of these system processes, register
627  * shutdown_kproc() as a handler for one of shutdown events.
628  */
629 static int kproc_shutdown_wait = 60;
630 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
631     &kproc_shutdown_wait, 0, "");
632 
633 void
634 kproc_shutdown(void *arg, int howto)
635 {
636 	struct proc *p;
637 	int error;
638 
639 	if (panicstr)
640 		return;
641 
642 	p = (struct proc *)arg;
643 	printf("Waiting (max %d seconds) for system process `%s' to stop...",
644 	    kproc_shutdown_wait, p->p_comm);
645 	error = kthread_suspend(p, kproc_shutdown_wait * hz);
646 
647 	if (error == EWOULDBLOCK)
648 		printf("timed out\n");
649 	else
650 		printf("stopped\n");
651 }
652