xref: /freebsd/sys/kern/kern_shutdown.c (revision d37ea99837e6ad50837fd9fe1771ddf1c3ba6002)
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  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_shutdown.c	8.3 (Berkeley) 1/21/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_ddb.h"
41 #include "opt_ddb_trace.h"
42 #include "opt_ddb_unattended.h"
43 #include "opt_hw_wdog.h"
44 #include "opt_mac.h"
45 #include "opt_panic.h"
46 #include "opt_show_busybufs.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/conf.h>
53 #include <sys/cons.h>
54 #include <sys/eventhandler.h>
55 #include <sys/kernel.h>
56 #include <sys/kthread.h>
57 #include <sys/mac.h>
58 #include <sys/malloc.h>
59 #include <sys/mount.h>
60 #include <sys/proc.h>
61 #include <sys/reboot.h>
62 #include <sys/resourcevar.h>
63 #include <sys/smp.h>		/* smp_active */
64 #include <sys/sysctl.h>
65 #include <sys/sysproto.h>
66 #include <sys/vnode.h>
67 
68 #include <machine/cpu.h>
69 #include <machine/pcb.h>
70 #include <machine/smp.h>
71 
72 #include <sys/signalvar.h>
73 #ifdef DDB
74 #include <ddb/ddb.h>
75 #endif
76 
77 #ifndef PANIC_REBOOT_WAIT_TIME
78 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
79 #endif
80 
81 /*
82  * Note that stdarg.h and the ANSI style va_start macro is used for both
83  * ANSI and traditional C compilers.
84  */
85 #include <machine/stdarg.h>
86 
87 #ifdef DDB
88 #ifdef DDB_UNATTENDED
89 int debugger_on_panic = 0;
90 #else
91 int debugger_on_panic = 1;
92 #endif
93 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW,
94 	&debugger_on_panic, 0, "Run debugger on kernel panic");
95 
96 #ifdef DDB_TRACE
97 int trace_on_panic = 1;
98 #else
99 int trace_on_panic = 0;
100 #endif
101 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW,
102 	&trace_on_panic, 0, "Print stack trace on kernel panic");
103 #endif
104 
105 int sync_on_panic = 1;
106 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW,
107 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
108 
109 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment");
110 
111 #ifdef	HW_WDOG
112 /*
113  * If there is a hardware watchdog, point this at the function needed to
114  * hold it off.
115  * It's needed when the kernel needs to do some lengthy operations.
116  * e.g. in wd.c when dumping core.. It's most annoying to have
117  * your precious core-dump only half written because the wdog kicked in.
118  */
119 watchdog_tickle_fn wdog_tickler = NULL;
120 #endif	/* HW_WDOG */
121 
122 /*
123  * Variable panicstr contains argument to first call to panic; used as flag
124  * to indicate that the kernel has already called panic.
125  */
126 const char *panicstr;
127 
128 int dumping;				/* system is dumping */
129 static struct dumperinfo dumper;	/* our selected dumper */
130 static struct pcb dumppcb;		/* "You Are Here" sign for dump-debuggers */
131 
132 static void boot(int) __dead2;
133 static void poweroff_wait(void *, int);
134 static void shutdown_halt(void *junk, int howto);
135 static void shutdown_panic(void *junk, int howto);
136 static void shutdown_reset(void *junk, int howto);
137 
138 /* register various local shutdown events */
139 static void
140 shutdown_conf(void *unused)
141 {
142 
143 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
144 	    SHUTDOWN_PRI_FIRST);
145 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
146 	    SHUTDOWN_PRI_LAST + 100);
147 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
148 	    SHUTDOWN_PRI_LAST + 100);
149 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
150 	    SHUTDOWN_PRI_LAST + 200);
151 }
152 
153 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL)
154 
155 /*
156  * The system call that results in a reboot
157  *
158  * MPSAFE
159  */
160 /* ARGSUSED */
161 int
162 reboot(struct thread *td, struct reboot_args *uap)
163 {
164 	int error;
165 
166 	error = 0;
167 #ifdef MAC
168 	error = mac_check_system_reboot(td->td_ucred, uap->opt);
169 #endif
170 	if (error == 0)
171 		error = suser(td);
172 	if (error == 0) {
173 		mtx_lock(&Giant);
174 		boot(uap->opt);
175 		mtx_unlock(&Giant);
176 	}
177 	return (error);
178 }
179 
180 /*
181  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
182  */
183 static int shutdown_howto = 0;
184 
185 void
186 shutdown_nice(int howto)
187 {
188 
189 	shutdown_howto = howto;
190 
191 	/* Send a signal to init(8) and have it shutdown the world */
192 	if (initproc != NULL) {
193 		PROC_LOCK(initproc);
194 		psignal(initproc, SIGINT);
195 		PROC_UNLOCK(initproc);
196 	} else {
197 		/* No init(8) running, so simply reboot */
198 		boot(RB_NOSYNC);
199 	}
200 	return;
201 }
202 static int	waittime = -1;
203 
204 static void
205 print_uptime(void)
206 {
207 	int f;
208 	struct timespec ts;
209 
210 	getnanouptime(&ts);
211 	printf("Uptime: ");
212 	f = 0;
213 	if (ts.tv_sec >= 86400) {
214 		printf("%ldd", (long)ts.tv_sec / 86400);
215 		ts.tv_sec %= 86400;
216 		f = 1;
217 	}
218 	if (f || ts.tv_sec >= 3600) {
219 		printf("%ldh", (long)ts.tv_sec / 3600);
220 		ts.tv_sec %= 3600;
221 		f = 1;
222 	}
223 	if (f || ts.tv_sec >= 60) {
224 		printf("%ldm", (long)ts.tv_sec / 60);
225 		ts.tv_sec %= 60;
226 		f = 1;
227 	}
228 	printf("%lds\n", (long)ts.tv_sec);
229 }
230 
231 static void
232 doadump(void)
233 {
234 
235 	savectx(&dumppcb);
236 	dumping++;
237 	dumpsys(&dumper);
238 }
239 
240 /*
241  *  Go through the rigmarole of shutting down..
242  * this used to be in machdep.c but I'll be dammned if I could see
243  * anything machine dependant in it.
244  */
245 static void
246 boot(int howto)
247 {
248 
249 	/* collect extra flags that shutdown_nice might have set */
250 	howto |= shutdown_howto;
251 
252 #ifdef DDB
253 	/* We are out of the debugger now. */
254 	db_active = 0;
255 #endif
256 
257 #ifdef SMP
258 	if (smp_active)
259 		printf("boot() called on cpu#%d\n", PCPU_GET(cpuid));
260 #endif
261 	/*
262 	 * Do any callouts that should be done BEFORE syncing the filesystems.
263 	 */
264 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
265 
266 	/*
267 	 * Now sync filesystems
268 	 */
269 	if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
270 		register struct buf *bp;
271 		int iter, nbusy, pbusy;
272 		int subiter;
273 
274 		waittime = 0;
275 		printf("\nsyncing disks, buffers remaining... ");
276 
277 		sync(&thread0, NULL);
278 
279 		/*
280 		 * With soft updates, some buffers that are
281 		 * written will be remarked as dirty until other
282 		 * buffers are written.
283 		 */
284 		for (iter = pbusy = 0; iter < 20; iter++) {
285 			nbusy = 0;
286 			for (bp = &buf[nbuf]; --bp >= buf; ) {
287 				if ((bp->b_flags & B_INVAL) == 0 &&
288 				    BUF_REFCNT(bp) > 0) {
289 					nbusy++;
290 				} else if ((bp->b_flags & (B_DELWRI | B_INVAL))
291 						== B_DELWRI) {
292 					/* bawrite(bp);*/
293 					nbusy++;
294 				}
295 			}
296 			if (nbusy == 0)
297 				break;
298 			printf("%d ", nbusy);
299 			if (nbusy < pbusy)
300 				iter = 0;
301 			pbusy = nbusy;
302 			sync(&thread0, NULL);
303  			if (curthread != NULL) {
304 				DROP_GIANT();
305    				for (subiter = 0; subiter < 50 * iter; subiter++) {
306      					mtx_lock_spin(&sched_lock);
307 					/*
308 					 * Allow interrupt threads to run
309 					 */
310      					mi_switch(SW_VOL);
311      					mtx_unlock_spin(&sched_lock);
312      					DELAY(1000);
313    				}
314 				PICKUP_GIANT();
315  			} else
316 			DELAY(50000 * iter);
317 		}
318 		printf("\n");
319 		/*
320 		 * Count only busy local buffers to prevent forcing
321 		 * a fsck if we're just a client of a wedged NFS server
322 		 */
323 		nbusy = 0;
324 		for (bp = &buf[nbuf]; --bp >= buf; ) {
325 			if (((bp->b_flags&B_INVAL) == 0 && BUF_REFCNT(bp)) ||
326 			    ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) {
327 				if (bp->b_dev == NULL) {
328 					TAILQ_REMOVE(&mountlist,
329 					    bp->b_vp->v_mount, mnt_list);
330 					continue;
331 				}
332 				nbusy++;
333 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC)
334 				printf(
335 			    "%d: dev:%s, flags:%0x, blkno:%ld, lblkno:%ld\n",
336 				    nbusy, devtoname(bp->b_dev),
337 				    bp->b_flags, (long)bp->b_blkno,
338 				    (long)bp->b_lblkno);
339 #endif
340 			}
341 		}
342 		if (nbusy) {
343 			/*
344 			 * Failed to sync all blocks. Indicate this and don't
345 			 * unmount filesystems (thus forcing an fsck on reboot).
346 			 */
347 			printf("giving up on %d buffers\n", nbusy);
348 			DELAY(5000000);	/* 5 seconds */
349 		} else {
350 			printf("done\n");
351 			/*
352 			 * Unmount filesystems
353 			 */
354 			if (panicstr == 0)
355 				vfs_unmountall();
356 		}
357 		DELAY(100000);		/* wait for console output to finish */
358 	}
359 
360 	print_uptime();
361 
362 	/*
363 	 * Ok, now do things that assume all filesystem activity has
364 	 * been completed.
365 	 */
366 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
367 	splhigh();
368 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP &&
369 	    !cold && dumper.dumper != NULL && !dumping)
370 		doadump();
371 
372 	/* Now that we're going to really halt the system... */
373 	EVENTHANDLER_INVOKE(shutdown_final, howto);
374 
375 	for(;;) ;	/* safety against shutdown_reset not working */
376 	/* NOTREACHED */
377 }
378 
379 /*
380  * If the shutdown was a clean halt, behave accordingly.
381  */
382 static void
383 shutdown_halt(void *junk, int howto)
384 {
385 
386 	if (howto & RB_HALT) {
387 		printf("\n");
388 		printf("The operating system has halted.\n");
389 		printf("Please press any key to reboot.\n\n");
390 		switch (cngetc()) {
391 		case -1:		/* No console, just die */
392 			cpu_halt();
393 			/* NOTREACHED */
394 		default:
395 			howto &= ~RB_HALT;
396 			break;
397 		}
398 	}
399 }
400 
401 /*
402  * Check to see if the system paniced, pause and then reboot
403  * according to the specified delay.
404  */
405 static void
406 shutdown_panic(void *junk, int howto)
407 {
408 	int loop;
409 
410 	if (howto & RB_DUMP) {
411 		if (PANIC_REBOOT_WAIT_TIME != 0) {
412 			if (PANIC_REBOOT_WAIT_TIME != -1) {
413 				printf("Automatic reboot in %d seconds - "
414 				       "press a key on the console to abort\n",
415 					PANIC_REBOOT_WAIT_TIME);
416 				for (loop = PANIC_REBOOT_WAIT_TIME * 10;
417 				     loop > 0; --loop) {
418 					DELAY(1000 * 100); /* 1/10th second */
419 					/* Did user type a key? */
420 					if (cncheckc() != -1)
421 						break;
422 				}
423 				if (!loop)
424 					return;
425 			}
426 		} else { /* zero time specified - reboot NOW */
427 			return;
428 		}
429 		printf("--> Press a key on the console to reboot,\n");
430 		printf("--> or switch off the system now.\n");
431 		cngetc();
432 	}
433 }
434 
435 /*
436  * Everything done, now reset
437  */
438 static void
439 shutdown_reset(void *junk, int howto)
440 {
441 
442 	printf("Rebooting...\n");
443 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
444 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
445 	cpu_reset();
446 	/* NOTREACHED */ /* assuming reset worked */
447 }
448 
449 /*
450  * Print a backtrace if we can.
451  */
452 
453 void
454 backtrace(void)
455 {
456 
457 #ifdef DDB
458 	printf("Stack backtrace:\n");
459 	db_print_backtrace();
460 #else
461 	printf("Sorry, need DDB option to print backtrace");
462 #endif
463 }
464 
465 #ifdef SMP
466 static u_int panic_cpu = NOCPU;
467 #endif
468 
469 /*
470  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
471  * and then reboots.  If we are called twice, then we avoid trying to sync
472  * the disks as this often leads to recursive panics.
473  *
474  * MPSAFE
475  */
476 void
477 panic(const char *fmt, ...)
478 {
479 	struct thread *td = curthread;
480 	int bootopt, newpanic;
481 	va_list ap;
482 	static char buf[256];
483 
484 #ifdef SMP
485 	/*
486 	 * We don't want multiple CPU's to panic at the same time, so we
487 	 * use panic_cpu as a simple spinlock.  We have to keep checking
488 	 * panic_cpu if we are spinning in case the panic on the first
489 	 * CPU is canceled.
490 	 */
491 	if (panic_cpu != PCPU_GET(cpuid))
492 		while (atomic_cmpset_int(&panic_cpu, NOCPU,
493 		    PCPU_GET(cpuid)) == 0)
494 			while (panic_cpu != NOCPU)
495 				; /* nothing */
496 #endif
497 
498 	bootopt = RB_AUTOBOOT | RB_DUMP;
499 	newpanic = 0;
500 	if (panicstr)
501 		bootopt |= RB_NOSYNC;
502 	else {
503 		panicstr = fmt;
504 		newpanic = 1;
505 	}
506 
507 	va_start(ap, fmt);
508 	if (newpanic) {
509 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
510 		panicstr = buf;
511 		printf("panic: %s\n", buf);
512 	} else {
513 		printf("panic: ");
514 		vprintf(fmt, ap);
515 		printf("\n");
516 	}
517 	va_end(ap);
518 #ifdef SMP
519 	/* two separate prints in case of an unmapped page and trap */
520 	printf("cpuid = %d; ", PCPU_GET(cpuid));
521 #ifdef APIC_IO
522 	printf("lapic.id = %08x\n", lapic.id);
523 #else
524 	printf("\n");
525 #endif
526 #endif
527 
528 #if defined(DDB)
529 	if (newpanic && trace_on_panic)
530 		backtrace();
531 	if (debugger_on_panic)
532 		Debugger ("panic");
533 #ifdef RESTARTABLE_PANICS
534 	/* See if the user aborted the panic, in which case we continue. */
535 	if (panicstr == NULL) {
536 #ifdef SMP
537 		atomic_store_rel_int(&panic_cpu, NOCPU);
538 #endif
539 		return;
540 	}
541 #endif
542 #endif
543 	mtx_lock_spin(&sched_lock);
544 	td->td_flags |= TDF_INPANIC;
545 	mtx_unlock_spin(&sched_lock);
546 	if (!sync_on_panic)
547 		bootopt |= RB_NOSYNC;
548 	boot(bootopt);
549 }
550 
551 /*
552  * Support for poweroff delay.
553  */
554 #ifndef POWEROFF_DELAY
555 # define POWEROFF_DELAY 5000
556 #endif
557 static int poweroff_delay = POWEROFF_DELAY;
558 
559 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
560 	&poweroff_delay, 0, "");
561 
562 static void
563 poweroff_wait(void *junk, int howto)
564 {
565 
566 	if (!(howto & RB_POWEROFF) || poweroff_delay <= 0)
567 		return;
568 	DELAY(poweroff_delay * 1000);
569 }
570 
571 /*
572  * Some system processes (e.g. syncer) need to be stopped at appropriate
573  * points in their main loops prior to a system shutdown, so that they
574  * won't interfere with the shutdown process (e.g. by holding a disk buf
575  * to cause sync to fail).  For each of these system processes, register
576  * shutdown_kproc() as a handler for one of shutdown events.
577  */
578 static int kproc_shutdown_wait = 60;
579 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
580     &kproc_shutdown_wait, 0, "");
581 
582 void
583 kproc_shutdown(void *arg, int howto)
584 {
585 	struct proc *p;
586 	int error;
587 
588 	if (panicstr)
589 		return;
590 
591 	p = (struct proc *)arg;
592 	printf("Waiting (max %d seconds) for system process `%s' to stop...",
593 	    kproc_shutdown_wait, p->p_comm);
594 	error = kthread_suspend(p, kproc_shutdown_wait * hz);
595 
596 	if (error == EWOULDBLOCK)
597 		printf("timed out\n");
598 	else
599 		printf("stopped\n");
600 }
601 
602 /* Registration of dumpers */
603 int
604 set_dumper(struct dumperinfo *di)
605 {
606 
607 	if (di == NULL) {
608 		bzero(&dumper, sizeof dumper);
609 		return (0);
610 	}
611 	if (dumper.dumper != NULL)
612 		return (EBUSY);
613 	dumper = *di;
614 	return (0);
615 }
616 
617 #if defined(__powerpc__)
618 void
619 dumpsys(struct dumperinfo *di __unused)
620 {
621 
622 	printf("Kernel dumps not implemented on this architecture\n");
623 }
624 #endif
625