xref: /freebsd/sys/kern/kern_shutdown.c (revision f9218d3d4fd34f082473b3a021c6d4d109fb47cf)
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_ddb_trace.h"
44 #include "opt_ddb_unattended.h"
45 #include "opt_hw_wdog.h"
46 #include "opt_mac.h"
47 #include "opt_panic.h"
48 #include "opt_show_busybufs.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/bio.h>
53 #include <sys/buf.h>
54 #include <sys/conf.h>
55 #include <sys/cons.h>
56 #include <sys/disklabel.h>
57 #include <sys/eventhandler.h>
58 #include <sys/kernel.h>
59 #include <sys/kthread.h>
60 #include <sys/mac.h>
61 #include <sys/malloc.h>
62 #include <sys/mount.h>
63 #include <sys/proc.h>
64 #include <sys/reboot.h>
65 #include <sys/resourcevar.h>
66 #include <sys/smp.h>		/* smp_active */
67 #include <sys/sysctl.h>
68 #include <sys/sysproto.h>
69 #include <sys/vnode.h>
70 
71 #include <machine/pcb.h>
72 #include <machine/md_var.h>
73 #include <machine/smp.h>
74 
75 #include <sys/signalvar.h>
76 #ifdef DDB
77 #include <ddb/ddb.h>
78 #endif
79 
80 #ifndef PANIC_REBOOT_WAIT_TIME
81 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
82 #endif
83 
84 /*
85  * Note that stdarg.h and the ANSI style va_start macro is used for both
86  * ANSI and traditional C compilers.
87  */
88 #include <machine/stdarg.h>
89 
90 #ifdef DDB
91 #ifdef DDB_UNATTENDED
92 int debugger_on_panic = 0;
93 #else
94 int debugger_on_panic = 1;
95 #endif
96 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW,
97 	&debugger_on_panic, 0, "Run debugger on kernel panic");
98 
99 #ifdef DDB_TRACE
100 int trace_on_panic = 1;
101 #else
102 int trace_on_panic = 0;
103 #endif
104 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW,
105 	&trace_on_panic, 0, "Print stack trace on kernel panic");
106 #endif
107 
108 int sync_on_panic = 1;
109 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW,
110 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
111 
112 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment");
113 
114 #ifdef	HW_WDOG
115 /*
116  * If there is a hardware watchdog, point this at the function needed to
117  * hold it off.
118  * It's needed when the kernel needs to do some lengthy operations.
119  * e.g. in wd.c when dumping core.. It's most annoying to have
120  * your precious core-dump only half written because the wdog kicked in.
121  */
122 watchdog_tickle_fn wdog_tickler = NULL;
123 #endif	/* HW_WDOG */
124 
125 /*
126  * Variable panicstr contains argument to first call to panic; used as flag
127  * to indicate that the kernel has already called panic.
128  */
129 const char *panicstr;
130 
131 int dumping;				/* system is dumping */
132 static struct dumperinfo dumper;	/* our selected dumper */
133 static struct pcb dumppcb;		/* "You Are Here" sign for dump-debuggers */
134 
135 static void boot(int) __dead2;
136 static void poweroff_wait(void *, int);
137 static void shutdown_halt(void *junk, int howto);
138 static void shutdown_panic(void *junk, int howto);
139 static void shutdown_reset(void *junk, int howto);
140 
141 /* register various local shutdown events */
142 static void
143 shutdown_conf(void *unused)
144 {
145 
146 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
147 	    SHUTDOWN_PRI_FIRST);
148 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
149 	    SHUTDOWN_PRI_LAST + 100);
150 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
151 	    SHUTDOWN_PRI_LAST + 100);
152 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
153 	    SHUTDOWN_PRI_LAST + 200);
154 }
155 
156 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL)
157 
158 /*
159  * The system call that results in a reboot
160  *
161  * MPSAFE
162  */
163 /* ARGSUSED */
164 int
165 reboot(struct thread *td, struct reboot_args *uap)
166 {
167 	int error;
168 
169 	error = 0;
170 #ifdef MAC
171 	error = mac_check_system_reboot(td->td_ucred, uap->opt);
172 #endif
173 	if (error == 0)
174 		error = suser(td);
175 	if (error == 0) {
176 		mtx_lock(&Giant);
177 		boot(uap->opt);
178 		mtx_unlock(&Giant);
179 	}
180 	return (error);
181 }
182 
183 /*
184  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
185  */
186 static int shutdown_howto = 0;
187 
188 void
189 shutdown_nice(int howto)
190 {
191 
192 	shutdown_howto = howto;
193 
194 	/* Send a signal to init(8) and have it shutdown the world */
195 	if (initproc != NULL) {
196 		PROC_LOCK(initproc);
197 		psignal(initproc, SIGINT);
198 		PROC_UNLOCK(initproc);
199 	} else {
200 		/* No init(8) running, so simply reboot */
201 		boot(RB_NOSYNC);
202 	}
203 	return;
204 }
205 static int	waittime = -1;
206 
207 static void
208 print_uptime(void)
209 {
210 	int f;
211 	struct timespec ts;
212 
213 	getnanouptime(&ts);
214 	printf("Uptime: ");
215 	f = 0;
216 	if (ts.tv_sec >= 86400) {
217 		printf("%ldd", (long)ts.tv_sec / 86400);
218 		ts.tv_sec %= 86400;
219 		f = 1;
220 	}
221 	if (f || ts.tv_sec >= 3600) {
222 		printf("%ldh", (long)ts.tv_sec / 3600);
223 		ts.tv_sec %= 3600;
224 		f = 1;
225 	}
226 	if (f || ts.tv_sec >= 60) {
227 		printf("%ldm", (long)ts.tv_sec / 60);
228 		ts.tv_sec %= 60;
229 		f = 1;
230 	}
231 	printf("%lds\n", (long)ts.tv_sec);
232 }
233 
234 static void
235 doadump(void)
236 {
237 
238 	savectx(&dumppcb);
239 	dumping++;
240 	dumpsys(&dumper);
241 }
242 
243 /*
244  *  Go through the rigmarole of shutting down..
245  * this used to be in machdep.c but I'll be dammned if I could see
246  * anything machine dependant in it.
247  */
248 static void
249 boot(int howto)
250 {
251 
252 	/* collect extra flags that shutdown_nice might have set */
253 	howto |= shutdown_howto;
254 
255 #ifdef DDB
256 	/* We are out of the debugger now. */
257 	db_active = 0;
258 #endif
259 
260 #ifdef SMP
261 	if (smp_active)
262 		printf("boot() called on cpu#%d\n", PCPU_GET(cpuid));
263 #endif
264 	/*
265 	 * Do any callouts that should be done BEFORE syncing the filesystems.
266 	 */
267 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
268 
269 	/*
270 	 * Now sync filesystems
271 	 */
272 	if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
273 		register struct buf *bp;
274 		int iter, nbusy, pbusy;
275 		int subiter;
276 
277 		waittime = 0;
278 		printf("\nsyncing disks, buffers remaining... ");
279 
280 		sync(&thread0, NULL);
281 
282 		/*
283 		 * With soft updates, some buffers that are
284 		 * written will be remarked as dirty until other
285 		 * buffers are written.
286 		 */
287 		for (iter = pbusy = 0; iter < 20; iter++) {
288 			nbusy = 0;
289 			for (bp = &buf[nbuf]; --bp >= buf; ) {
290 				if ((bp->b_flags & B_INVAL) == 0 &&
291 				    BUF_REFCNT(bp) > 0) {
292 					nbusy++;
293 				} else if ((bp->b_flags & (B_DELWRI | B_INVAL))
294 						== B_DELWRI) {
295 					/* bawrite(bp);*/
296 					nbusy++;
297 				}
298 			}
299 			if (nbusy == 0)
300 				break;
301 			printf("%d ", nbusy);
302 			if (nbusy < pbusy)
303 				iter = 0;
304 			pbusy = nbusy;
305 			sync(&thread0, NULL);
306  			if (curthread != NULL) {
307 				DROP_GIANT();
308    				for (subiter = 0; subiter < 50 * iter; subiter++) {
309      					mtx_lock_spin(&sched_lock);
310 					curthread->td_proc->p_stats->p_ru.ru_nvcsw++;
311      					mi_switch(); /* Allow interrupt threads to run */
312      					mtx_unlock_spin(&sched_lock);
313      					DELAY(1000);
314    				}
315 				PICKUP_GIANT();
316  			} else
317 			DELAY(50000 * iter);
318 		}
319 		printf("\n");
320 		/*
321 		 * Count only busy local buffers to prevent forcing
322 		 * a fsck if we're just a client of a wedged NFS server
323 		 */
324 		nbusy = 0;
325 		for (bp = &buf[nbuf]; --bp >= buf; ) {
326 			if (((bp->b_flags&B_INVAL) == 0 && BUF_REFCNT(bp)) ||
327 			    ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) {
328 				if (bp->b_dev == NODEV) {
329 					TAILQ_REMOVE(&mountlist,
330 					    bp->b_vp->v_mount, mnt_list);
331 					continue;
332 				}
333 				nbusy++;
334 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC)
335 				printf(
336 			    "%d: dev:%s, flags:%0x, blkno:%ld, lblkno:%ld\n",
337 				    nbusy, devtoname(bp->b_dev),
338 				    bp->b_flags, (long)bp->b_blkno,
339 				    (long)bp->b_lblkno);
340 #endif
341 			}
342 		}
343 		if (nbusy) {
344 			/*
345 			 * Failed to sync all blocks. Indicate this and don't
346 			 * unmount filesystems (thus forcing an fsck on reboot).
347 			 */
348 			printf("giving up on %d buffers\n", nbusy);
349 			DELAY(5000000);	/* 5 seconds */
350 		} else {
351 			printf("done\n");
352 			/*
353 			 * Unmount filesystems
354 			 */
355 			if (panicstr == 0)
356 				vfs_unmountall();
357 		}
358 		DELAY(100000);		/* wait for console output to finish */
359 	}
360 
361 	print_uptime();
362 
363 	/*
364 	 * Ok, now do things that assume all filesystem activity has
365 	 * been completed.
366 	 */
367 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
368 	splhigh();
369 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP &&
370 	    !cold && dumper.dumper != NULL && !dumping)
371 		doadump();
372 
373 	/* Now that we're going to really halt the system... */
374 	EVENTHANDLER_INVOKE(shutdown_final, howto);
375 
376 	for(;;) ;	/* safety against shutdown_reset not working */
377 	/* NOTREACHED */
378 }
379 
380 /*
381  * If the shutdown was a clean halt, behave accordingly.
382  */
383 static void
384 shutdown_halt(void *junk, int howto)
385 {
386 
387 	if (howto & RB_HALT) {
388 		printf("\n");
389 		printf("The operating system has halted.\n");
390 		printf("Please press any key to reboot.\n\n");
391 		switch (cngetc()) {
392 		case -1:		/* No console, just die */
393 			cpu_halt();
394 			/* NOTREACHED */
395 		default:
396 			howto &= ~RB_HALT;
397 			break;
398 		}
399 	}
400 }
401 
402 /*
403  * Check to see if the system paniced, pause and then reboot
404  * according to the specified delay.
405  */
406 static void
407 shutdown_panic(void *junk, int howto)
408 {
409 	int loop;
410 
411 	if (howto & RB_DUMP) {
412 		if (PANIC_REBOOT_WAIT_TIME != 0) {
413 			if (PANIC_REBOOT_WAIT_TIME != -1) {
414 				printf("Automatic reboot in %d seconds - "
415 				       "press a key on the console to abort\n",
416 					PANIC_REBOOT_WAIT_TIME);
417 				for (loop = PANIC_REBOOT_WAIT_TIME * 10;
418 				     loop > 0; --loop) {
419 					DELAY(1000 * 100); /* 1/10th second */
420 					/* Did user type a key? */
421 					if (cncheckc() != -1)
422 						break;
423 				}
424 				if (!loop)
425 					return;
426 			}
427 		} else { /* zero time specified - reboot NOW */
428 			return;
429 		}
430 		printf("--> Press a key on the console to reboot,\n");
431 		printf("--> or switch off the system now.\n");
432 		cngetc();
433 	}
434 }
435 
436 /*
437  * Everything done, now reset
438  */
439 static void
440 shutdown_reset(void *junk, int howto)
441 {
442 
443 	printf("Rebooting...\n");
444 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
445 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
446 	cpu_reset();
447 	/* NOTREACHED */ /* assuming reset worked */
448 }
449 
450 /*
451  * Print a backtrace if we can.
452  */
453 
454 void
455 backtrace(void)
456 {
457 
458 #ifdef DDB
459 	printf("Stack backtrace:\n");
460 	db_print_backtrace();
461 #else
462 	printf("Sorry, need DDB option to print backtrace");
463 #endif
464 }
465 
466 #ifdef SMP
467 static u_int panic_cpu = NOCPU;
468 #endif
469 
470 /*
471  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
472  * and then reboots.  If we are called twice, then we avoid trying to sync
473  * the disks as this often leads to recursive panics.
474  *
475  * MPSAFE
476  */
477 void
478 panic(const char *fmt, ...)
479 {
480 	struct thread *td = curthread;
481 	int bootopt, newpanic;
482 	va_list ap;
483 	static char buf[256];
484 
485 #ifdef SMP
486 	/*
487 	 * We don't want multiple CPU's to panic at the same time, so we
488 	 * use panic_cpu as a simple spinlock.  We have to keep checking
489 	 * panic_cpu if we are spinning in case the panic on the first
490 	 * CPU is canceled.
491 	 */
492 	if (panic_cpu != PCPU_GET(cpuid))
493 		while (atomic_cmpset_int(&panic_cpu, NOCPU,
494 		    PCPU_GET(cpuid)) == 0)
495 			while (panic_cpu != NOCPU)
496 				; /* nothing */
497 #endif
498 
499 	bootopt = RB_AUTOBOOT | RB_DUMP;
500 	newpanic = 0;
501 	if (panicstr)
502 		bootopt |= RB_NOSYNC;
503 	else {
504 		panicstr = fmt;
505 		newpanic = 1;
506 	}
507 
508 	va_start(ap, fmt);
509 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
510 	if (panicstr == fmt)
511 		panicstr = buf;
512 	va_end(ap);
513 	printf("panic: %s\n", buf);
514 #ifdef SMP
515 	/* two separate prints in case of an unmapped page and trap */
516 	printf("cpuid = %d; ", PCPU_GET(cpuid));
517 #ifdef APIC_IO
518 	printf("lapic.id = %08x\n", lapic.id);
519 #else
520 	printf("\n");
521 #endif
522 #endif
523 
524 #if defined(DDB)
525 	if (newpanic && trace_on_panic)
526 		backtrace();
527 	if (debugger_on_panic)
528 		Debugger ("panic");
529 #ifdef RESTARTABLE_PANICS
530 	/* See if the user aborted the panic, in which case we continue. */
531 	if (panicstr == NULL) {
532 #ifdef SMP
533 		atomic_store_rel_int(&panic_cpu, NOCPU);
534 #endif
535 		return;
536 	}
537 #endif
538 #endif
539 	td->td_flags |= TDF_INPANIC;
540 	if (!sync_on_panic)
541 		bootopt |= RB_NOSYNC;
542 	boot(bootopt);
543 }
544 
545 /*
546  * Support for poweroff delay.
547  */
548 #ifndef POWEROFF_DELAY
549 # define POWEROFF_DELAY 5000
550 #endif
551 static int poweroff_delay = POWEROFF_DELAY;
552 
553 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
554 	&poweroff_delay, 0, "");
555 
556 static void
557 poweroff_wait(void *junk, int howto)
558 {
559 
560 	if (!(howto & RB_POWEROFF) || poweroff_delay <= 0)
561 		return;
562 	DELAY(poweroff_delay * 1000);
563 }
564 
565 /*
566  * Some system processes (e.g. syncer) need to be stopped at appropriate
567  * points in their main loops prior to a system shutdown, so that they
568  * won't interfere with the shutdown process (e.g. by holding a disk buf
569  * to cause sync to fail).  For each of these system processes, register
570  * shutdown_kproc() as a handler for one of shutdown events.
571  */
572 static int kproc_shutdown_wait = 60;
573 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
574     &kproc_shutdown_wait, 0, "");
575 
576 void
577 kproc_shutdown(void *arg, int howto)
578 {
579 	struct proc *p;
580 	int error;
581 
582 	if (panicstr)
583 		return;
584 
585 	p = (struct proc *)arg;
586 	printf("Waiting (max %d seconds) for system process `%s' to stop...",
587 	    kproc_shutdown_wait, p->p_comm);
588 	error = kthread_suspend(p, kproc_shutdown_wait * hz);
589 
590 	if (error == EWOULDBLOCK)
591 		printf("timed out\n");
592 	else
593 		printf("stopped\n");
594 }
595 
596 /* Registration of dumpers */
597 int
598 set_dumper(struct dumperinfo *di)
599 {
600 
601 	if (di == NULL) {
602 		bzero(&dumper, sizeof dumper);
603 		return (0);
604 	}
605 	if (dumper.dumper != NULL)
606 		return (EBUSY);
607 	dumper = *di;
608 	return (0);
609 }
610 
611 #if defined(__powerpc__)
612 void
613 dumpsys(struct dumperinfo *di __unused)
614 {
615 
616 	printf("Kernel dumps not implemented on this architecture\n");
617 }
618 #endif
619