xref: /freebsd/sys/kern/kern_shutdown.c (revision e0c27215058b5786c78fcfb3963eebe61a989511)
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  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_ddb.h"
45 #include "opt_ddb_trace.h"
46 #include "opt_ddb_unattended.h"
47 #include "opt_hw_wdog.h"
48 #include "opt_mac.h"
49 #include "opt_panic.h"
50 #include "opt_show_busybufs.h"
51 
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/bio.h>
55 #include <sys/buf.h>
56 #include <sys/conf.h>
57 #include <sys/cons.h>
58 #include <sys/eventhandler.h>
59 #include <sys/kernel.h>
60 #include <sys/kthread.h>
61 #include <sys/mac.h>
62 #include <sys/malloc.h>
63 #include <sys/mount.h>
64 #include <sys/proc.h>
65 #include <sys/reboot.h>
66 #include <sys/resourcevar.h>
67 #include <sys/smp.h>		/* smp_active */
68 #include <sys/sysctl.h>
69 #include <sys/sysproto.h>
70 #include <sys/vnode.h>
71 
72 #include <machine/pcb.h>
73 #include <machine/md_var.h>
74 #include <machine/smp.h>
75 
76 #include <sys/signalvar.h>
77 #ifdef DDB
78 #include <ddb/ddb.h>
79 #endif
80 
81 #ifndef PANIC_REBOOT_WAIT_TIME
82 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
83 #endif
84 
85 /*
86  * Note that stdarg.h and the ANSI style va_start macro is used for both
87  * ANSI and traditional C compilers.
88  */
89 #include <machine/stdarg.h>
90 
91 #ifdef DDB
92 #ifdef DDB_UNATTENDED
93 int debugger_on_panic = 0;
94 #else
95 int debugger_on_panic = 1;
96 #endif
97 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW,
98 	&debugger_on_panic, 0, "Run debugger on kernel panic");
99 
100 #ifdef DDB_TRACE
101 int trace_on_panic = 1;
102 #else
103 int trace_on_panic = 0;
104 #endif
105 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW,
106 	&trace_on_panic, 0, "Print stack trace on kernel panic");
107 #endif
108 
109 int sync_on_panic = 1;
110 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW,
111 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
112 
113 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment");
114 
115 #ifdef	HW_WDOG
116 /*
117  * If there is a hardware watchdog, point this at the function needed to
118  * hold it off.
119  * It's needed when the kernel needs to do some lengthy operations.
120  * e.g. in wd.c when dumping core.. It's most annoying to have
121  * your precious core-dump only half written because the wdog kicked in.
122  */
123 watchdog_tickle_fn wdog_tickler = NULL;
124 #endif	/* HW_WDOG */
125 
126 /*
127  * Variable panicstr contains argument to first call to panic; used as flag
128  * to indicate that the kernel has already called panic.
129  */
130 const char *panicstr;
131 
132 int dumping;				/* system is dumping */
133 static struct dumperinfo dumper;	/* our selected dumper */
134 static struct pcb dumppcb;		/* "You Are Here" sign for dump-debuggers */
135 
136 static void boot(int) __dead2;
137 static void poweroff_wait(void *, int);
138 static void shutdown_halt(void *junk, int howto);
139 static void shutdown_panic(void *junk, int howto);
140 static void shutdown_reset(void *junk, int howto);
141 
142 /* register various local shutdown events */
143 static void
144 shutdown_conf(void *unused)
145 {
146 
147 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
148 	    SHUTDOWN_PRI_FIRST);
149 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
150 	    SHUTDOWN_PRI_LAST + 100);
151 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
152 	    SHUTDOWN_PRI_LAST + 100);
153 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
154 	    SHUTDOWN_PRI_LAST + 200);
155 }
156 
157 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL)
158 
159 /*
160  * The system call that results in a reboot
161  *
162  * MPSAFE
163  */
164 /* ARGSUSED */
165 int
166 reboot(struct thread *td, struct reboot_args *uap)
167 {
168 	int error;
169 
170 	error = 0;
171 #ifdef MAC
172 	error = mac_check_system_reboot(td->td_ucred, uap->opt);
173 #endif
174 	if (error == 0)
175 		error = suser(td);
176 	if (error == 0) {
177 		mtx_lock(&Giant);
178 		boot(uap->opt);
179 		mtx_unlock(&Giant);
180 	}
181 	return (error);
182 }
183 
184 /*
185  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
186  */
187 static int shutdown_howto = 0;
188 
189 void
190 shutdown_nice(int howto)
191 {
192 
193 	shutdown_howto = howto;
194 
195 	/* Send a signal to init(8) and have it shutdown the world */
196 	if (initproc != NULL) {
197 		PROC_LOCK(initproc);
198 		psignal(initproc, SIGINT);
199 		PROC_UNLOCK(initproc);
200 	} else {
201 		/* No init(8) running, so simply reboot */
202 		boot(RB_NOSYNC);
203 	}
204 	return;
205 }
206 static int	waittime = -1;
207 
208 static void
209 print_uptime(void)
210 {
211 	int f;
212 	struct timespec ts;
213 
214 	getnanouptime(&ts);
215 	printf("Uptime: ");
216 	f = 0;
217 	if (ts.tv_sec >= 86400) {
218 		printf("%ldd", (long)ts.tv_sec / 86400);
219 		ts.tv_sec %= 86400;
220 		f = 1;
221 	}
222 	if (f || ts.tv_sec >= 3600) {
223 		printf("%ldh", (long)ts.tv_sec / 3600);
224 		ts.tv_sec %= 3600;
225 		f = 1;
226 	}
227 	if (f || ts.tv_sec >= 60) {
228 		printf("%ldm", (long)ts.tv_sec / 60);
229 		ts.tv_sec %= 60;
230 		f = 1;
231 	}
232 	printf("%lds\n", (long)ts.tv_sec);
233 }
234 
235 static void
236 doadump(void)
237 {
238 
239 	savectx(&dumppcb);
240 	dumping++;
241 	dumpsys(&dumper);
242 }
243 
244 /*
245  *  Go through the rigmarole of shutting down..
246  * this used to be in machdep.c but I'll be dammned if I could see
247  * anything machine dependant in it.
248  */
249 static void
250 boot(int howto)
251 {
252 
253 	/* collect extra flags that shutdown_nice might have set */
254 	howto |= shutdown_howto;
255 
256 #ifdef DDB
257 	/* We are out of the debugger now. */
258 	db_active = 0;
259 #endif
260 
261 #ifdef SMP
262 	if (smp_active)
263 		printf("boot() called on cpu#%d\n", PCPU_GET(cpuid));
264 #endif
265 	/*
266 	 * Do any callouts that should be done BEFORE syncing the filesystems.
267 	 */
268 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
269 
270 	/*
271 	 * Now sync filesystems
272 	 */
273 	if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
274 		register struct buf *bp;
275 		int iter, nbusy, pbusy;
276 		int subiter;
277 
278 		waittime = 0;
279 		printf("\nsyncing disks, buffers remaining... ");
280 
281 		sync(&thread0, NULL);
282 
283 		/*
284 		 * With soft updates, some buffers that are
285 		 * written will be remarked as dirty until other
286 		 * buffers are written.
287 		 */
288 		for (iter = pbusy = 0; iter < 20; iter++) {
289 			nbusy = 0;
290 			for (bp = &buf[nbuf]; --bp >= buf; ) {
291 				if ((bp->b_flags & B_INVAL) == 0 &&
292 				    BUF_REFCNT(bp) > 0) {
293 					nbusy++;
294 				} else if ((bp->b_flags & (B_DELWRI | B_INVAL))
295 						== B_DELWRI) {
296 					/* bawrite(bp);*/
297 					nbusy++;
298 				}
299 			}
300 			if (nbusy == 0)
301 				break;
302 			printf("%d ", nbusy);
303 			if (nbusy < pbusy)
304 				iter = 0;
305 			pbusy = nbusy;
306 			sync(&thread0, NULL);
307  			if (curthread != NULL) {
308 				DROP_GIANT();
309    				for (subiter = 0; subiter < 50 * iter; subiter++) {
310      					mtx_lock_spin(&sched_lock);
311 					curthread->td_proc->p_stats->p_ru.ru_nvcsw++;
312      					mi_switch(); /* Allow interrupt threads to run */
313      					mtx_unlock_spin(&sched_lock);
314      					DELAY(1000);
315    				}
316 				PICKUP_GIANT();
317  			} else
318 			DELAY(50000 * iter);
319 		}
320 		printf("\n");
321 		/*
322 		 * Count only busy local buffers to prevent forcing
323 		 * a fsck if we're just a client of a wedged NFS server
324 		 */
325 		nbusy = 0;
326 		for (bp = &buf[nbuf]; --bp >= buf; ) {
327 			if (((bp->b_flags&B_INVAL) == 0 && BUF_REFCNT(bp)) ||
328 			    ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) {
329 				if (bp->b_dev == NODEV) {
330 					TAILQ_REMOVE(&mountlist,
331 					    bp->b_vp->v_mount, mnt_list);
332 					continue;
333 				}
334 				nbusy++;
335 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC)
336 				printf(
337 			    "%d: dev:%s, flags:%0x, blkno:%ld, lblkno:%ld\n",
338 				    nbusy, devtoname(bp->b_dev),
339 				    bp->b_flags, (long)bp->b_blkno,
340 				    (long)bp->b_lblkno);
341 #endif
342 			}
343 		}
344 		if (nbusy) {
345 			/*
346 			 * Failed to sync all blocks. Indicate this and don't
347 			 * unmount filesystems (thus forcing an fsck on reboot).
348 			 */
349 			printf("giving up on %d buffers\n", nbusy);
350 			DELAY(5000000);	/* 5 seconds */
351 		} else {
352 			printf("done\n");
353 			/*
354 			 * Unmount filesystems
355 			 */
356 			if (panicstr == 0)
357 				vfs_unmountall();
358 		}
359 		DELAY(100000);		/* wait for console output to finish */
360 	}
361 
362 	print_uptime();
363 
364 	/*
365 	 * Ok, now do things that assume all filesystem activity has
366 	 * been completed.
367 	 */
368 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
369 	splhigh();
370 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP &&
371 	    !cold && dumper.dumper != NULL && !dumping)
372 		doadump();
373 
374 	/* Now that we're going to really halt the system... */
375 	EVENTHANDLER_INVOKE(shutdown_final, howto);
376 
377 	for(;;) ;	/* safety against shutdown_reset not working */
378 	/* NOTREACHED */
379 }
380 
381 /*
382  * If the shutdown was a clean halt, behave accordingly.
383  */
384 static void
385 shutdown_halt(void *junk, int howto)
386 {
387 
388 	if (howto & RB_HALT) {
389 		printf("\n");
390 		printf("The operating system has halted.\n");
391 		printf("Please press any key to reboot.\n\n");
392 		switch (cngetc()) {
393 		case -1:		/* No console, just die */
394 			cpu_halt();
395 			/* NOTREACHED */
396 		default:
397 			howto &= ~RB_HALT;
398 			break;
399 		}
400 	}
401 }
402 
403 /*
404  * Check to see if the system paniced, pause and then reboot
405  * according to the specified delay.
406  */
407 static void
408 shutdown_panic(void *junk, int howto)
409 {
410 	int loop;
411 
412 	if (howto & RB_DUMP) {
413 		if (PANIC_REBOOT_WAIT_TIME != 0) {
414 			if (PANIC_REBOOT_WAIT_TIME != -1) {
415 				printf("Automatic reboot in %d seconds - "
416 				       "press a key on the console to abort\n",
417 					PANIC_REBOOT_WAIT_TIME);
418 				for (loop = PANIC_REBOOT_WAIT_TIME * 10;
419 				     loop > 0; --loop) {
420 					DELAY(1000 * 100); /* 1/10th second */
421 					/* Did user type a key? */
422 					if (cncheckc() != -1)
423 						break;
424 				}
425 				if (!loop)
426 					return;
427 			}
428 		} else { /* zero time specified - reboot NOW */
429 			return;
430 		}
431 		printf("--> Press a key on the console to reboot,\n");
432 		printf("--> or switch off the system now.\n");
433 		cngetc();
434 	}
435 }
436 
437 /*
438  * Everything done, now reset
439  */
440 static void
441 shutdown_reset(void *junk, int howto)
442 {
443 
444 	printf("Rebooting...\n");
445 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
446 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
447 	cpu_reset();
448 	/* NOTREACHED */ /* assuming reset worked */
449 }
450 
451 /*
452  * Print a backtrace if we can.
453  */
454 
455 void
456 backtrace(void)
457 {
458 
459 #ifdef DDB
460 	printf("Stack backtrace:\n");
461 	db_print_backtrace();
462 #else
463 	printf("Sorry, need DDB option to print backtrace");
464 #endif
465 }
466 
467 #ifdef SMP
468 static u_int panic_cpu = NOCPU;
469 #endif
470 
471 /*
472  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
473  * and then reboots.  If we are called twice, then we avoid trying to sync
474  * the disks as this often leads to recursive panics.
475  *
476  * MPSAFE
477  */
478 void
479 panic(const char *fmt, ...)
480 {
481 	struct thread *td = curthread;
482 	int bootopt, newpanic;
483 	va_list ap;
484 	static char buf[256];
485 
486 #ifdef SMP
487 	/*
488 	 * We don't want multiple CPU's to panic at the same time, so we
489 	 * use panic_cpu as a simple spinlock.  We have to keep checking
490 	 * panic_cpu if we are spinning in case the panic on the first
491 	 * CPU is canceled.
492 	 */
493 	if (panic_cpu != PCPU_GET(cpuid))
494 		while (atomic_cmpset_int(&panic_cpu, NOCPU,
495 		    PCPU_GET(cpuid)) == 0)
496 			while (panic_cpu != NOCPU)
497 				; /* nothing */
498 #endif
499 
500 	bootopt = RB_AUTOBOOT | RB_DUMP;
501 	newpanic = 0;
502 	if (panicstr)
503 		bootopt |= RB_NOSYNC;
504 	else {
505 		panicstr = fmt;
506 		newpanic = 1;
507 	}
508 
509 	va_start(ap, fmt);
510 	if (newpanic) {
511 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
512 		panicstr = buf;
513 		printf("panic: %s\n", buf);
514 	} else {
515 		printf("panic: ");
516 		vprintf(fmt, ap);
517 		printf("\n");
518 	}
519 	va_end(ap);
520 #ifdef SMP
521 	/* two separate prints in case of an unmapped page and trap */
522 	printf("cpuid = %d; ", PCPU_GET(cpuid));
523 #ifdef APIC_IO
524 	printf("lapic.id = %08x\n", lapic.id);
525 #else
526 	printf("\n");
527 #endif
528 #endif
529 
530 #if defined(DDB)
531 	if (newpanic && trace_on_panic)
532 		backtrace();
533 	if (debugger_on_panic)
534 		Debugger ("panic");
535 #ifdef RESTARTABLE_PANICS
536 	/* See if the user aborted the panic, in which case we continue. */
537 	if (panicstr == NULL) {
538 #ifdef SMP
539 		atomic_store_rel_int(&panic_cpu, NOCPU);
540 #endif
541 		return;
542 	}
543 #endif
544 #endif
545 	mtx_lock_spin(&sched_lock);
546 	td->td_flags |= TDF_INPANIC;
547 	mtx_unlock_spin(&sched_lock);
548 	if (!sync_on_panic)
549 		bootopt |= RB_NOSYNC;
550 	boot(bootopt);
551 }
552 
553 /*
554  * Support for poweroff delay.
555  */
556 #ifndef POWEROFF_DELAY
557 # define POWEROFF_DELAY 5000
558 #endif
559 static int poweroff_delay = POWEROFF_DELAY;
560 
561 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
562 	&poweroff_delay, 0, "");
563 
564 static void
565 poweroff_wait(void *junk, int howto)
566 {
567 
568 	if (!(howto & RB_POWEROFF) || poweroff_delay <= 0)
569 		return;
570 	DELAY(poweroff_delay * 1000);
571 }
572 
573 /*
574  * Some system processes (e.g. syncer) need to be stopped at appropriate
575  * points in their main loops prior to a system shutdown, so that they
576  * won't interfere with the shutdown process (e.g. by holding a disk buf
577  * to cause sync to fail).  For each of these system processes, register
578  * shutdown_kproc() as a handler for one of shutdown events.
579  */
580 static int kproc_shutdown_wait = 60;
581 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
582     &kproc_shutdown_wait, 0, "");
583 
584 void
585 kproc_shutdown(void *arg, int howto)
586 {
587 	struct proc *p;
588 	int error;
589 
590 	if (panicstr)
591 		return;
592 
593 	p = (struct proc *)arg;
594 	printf("Waiting (max %d seconds) for system process `%s' to stop...",
595 	    kproc_shutdown_wait, p->p_comm);
596 	error = kthread_suspend(p, kproc_shutdown_wait * hz);
597 
598 	if (error == EWOULDBLOCK)
599 		printf("timed out\n");
600 	else
601 		printf("stopped\n");
602 }
603 
604 /* Registration of dumpers */
605 int
606 set_dumper(struct dumperinfo *di)
607 {
608 
609 	if (di == NULL) {
610 		bzero(&dumper, sizeof dumper);
611 		return (0);
612 	}
613 	if (dumper.dumper != NULL)
614 		return (EBUSY);
615 	dumper = *di;
616 	return (0);
617 }
618 
619 #if defined(__powerpc__)
620 void
621 dumpsys(struct dumperinfo *di __unused)
622 {
623 
624 	printf("Kernel dumps not implemented on this architecture\n");
625 }
626 #endif
627