xref: /freebsd/sys/kern/kern_shutdown.c (revision 6fd05b64b5b65dd4ba9b86482a0634a5f0b96c29)
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_kdb.h"
41 #include "opt_hw_wdog.h"
42 #include "opt_mac.h"
43 #include "opt_panic.h"
44 #include "opt_show_busybufs.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/bio.h>
49 #include <sys/buf.h>
50 #include <sys/conf.h>
51 #include <sys/cons.h>
52 #include <sys/eventhandler.h>
53 #include <sys/kdb.h>
54 #include <sys/kernel.h>
55 #include <sys/kthread.h>
56 #include <sys/mac.h>
57 #include <sys/malloc.h>
58 #include <sys/mount.h>
59 #include <sys/proc.h>
60 #include <sys/reboot.h>
61 #include <sys/resourcevar.h>
62 #include <sys/smp.h>		/* smp_active */
63 #include <sys/sysctl.h>
64 #include <sys/sysproto.h>
65 #include <sys/vnode.h>
66 
67 #include <machine/cpu.h>
68 #include <machine/pcb.h>
69 #include <machine/smp.h>
70 
71 #include <sys/signalvar.h>
72 
73 #ifndef PANIC_REBOOT_WAIT_TIME
74 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
75 #endif
76 
77 /*
78  * Note that stdarg.h and the ANSI style va_start macro is used for both
79  * ANSI and traditional C compilers.
80  */
81 #include <machine/stdarg.h>
82 
83 #ifdef KDB
84 #ifdef KDB_UNATTENDED
85 int debugger_on_panic = 0;
86 #else
87 int debugger_on_panic = 1;
88 #endif
89 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW,
90 	&debugger_on_panic, 0, "Run debugger on kernel panic");
91 
92 #ifdef KDB_TRACE
93 int trace_on_panic = 1;
94 #else
95 int trace_on_panic = 0;
96 #endif
97 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW,
98 	&trace_on_panic, 0, "Print stack trace on kernel panic");
99 #endif /* KDB */
100 
101 int sync_on_panic = 1;
102 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW,
103 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
104 
105 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment");
106 
107 #ifdef	HW_WDOG
108 /*
109  * If there is a hardware watchdog, point this at the function needed to
110  * hold it off.
111  * It's needed when the kernel needs to do some lengthy operations.
112  * e.g. in wd.c when dumping core.. It's most annoying to have
113  * your precious core-dump only half written because the wdog kicked in.
114  */
115 watchdog_tickle_fn wdog_tickler = NULL;
116 #endif	/* HW_WDOG */
117 
118 /*
119  * Variable panicstr contains argument to first call to panic; used as flag
120  * to indicate that the kernel has already called panic.
121  */
122 const char *panicstr;
123 
124 int dumping;				/* system is dumping */
125 static struct dumperinfo dumper;	/* our selected dumper */
126 
127 /* Context information for dump-debuggers. */
128 static struct pcb dumppcb;		/* Registers. */
129 static lwpid_t dumptid;			/* Thread ID. */
130 
131 static void boot(int) __dead2;
132 static void poweroff_wait(void *, int);
133 static void shutdown_halt(void *junk, int howto);
134 static void shutdown_panic(void *junk, int howto);
135 static void shutdown_reset(void *junk, int howto);
136 
137 /* register various local shutdown events */
138 static void
139 shutdown_conf(void *unused)
140 {
141 
142 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
143 	    SHUTDOWN_PRI_FIRST);
144 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
145 	    SHUTDOWN_PRI_LAST + 100);
146 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
147 	    SHUTDOWN_PRI_LAST + 100);
148 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
149 	    SHUTDOWN_PRI_LAST + 200);
150 }
151 
152 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL)
153 
154 /*
155  * The system call that results in a reboot
156  *
157  * MPSAFE
158  */
159 /* ARGSUSED */
160 int
161 reboot(struct thread *td, struct reboot_args *uap)
162 {
163 	int error;
164 
165 	error = 0;
166 #ifdef MAC
167 	error = mac_check_system_reboot(td->td_ucred, uap->opt);
168 #endif
169 	if (error == 0)
170 		error = suser(td);
171 	if (error == 0) {
172 		mtx_lock(&Giant);
173 		boot(uap->opt);
174 		mtx_unlock(&Giant);
175 	}
176 	return (error);
177 }
178 
179 /*
180  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
181  */
182 static int shutdown_howto = 0;
183 
184 void
185 shutdown_nice(int howto)
186 {
187 
188 	shutdown_howto = howto;
189 
190 	/* Send a signal to init(8) and have it shutdown the world */
191 	if (initproc != NULL) {
192 		PROC_LOCK(initproc);
193 		psignal(initproc, SIGINT);
194 		PROC_UNLOCK(initproc);
195 	} else {
196 		/* No init(8) running, so simply reboot */
197 		boot(RB_NOSYNC);
198 	}
199 	return;
200 }
201 static int	waittime = -1;
202 
203 static void
204 print_uptime(void)
205 {
206 	int f;
207 	struct timespec ts;
208 
209 	getnanouptime(&ts);
210 	printf("Uptime: ");
211 	f = 0;
212 	if (ts.tv_sec >= 86400) {
213 		printf("%ldd", (long)ts.tv_sec / 86400);
214 		ts.tv_sec %= 86400;
215 		f = 1;
216 	}
217 	if (f || ts.tv_sec >= 3600) {
218 		printf("%ldh", (long)ts.tv_sec / 3600);
219 		ts.tv_sec %= 3600;
220 		f = 1;
221 	}
222 	if (f || ts.tv_sec >= 60) {
223 		printf("%ldm", (long)ts.tv_sec / 60);
224 		ts.tv_sec %= 60;
225 		f = 1;
226 	}
227 	printf("%lds\n", (long)ts.tv_sec);
228 }
229 
230 static void
231 doadump(void)
232 {
233 
234 	savectx(&dumppcb);
235 	dumptid = curthread->td_tid;
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 	static int first_buf_printf = 1;
249 
250 	/* collect extra flags that shutdown_nice might have set */
251 	howto |= shutdown_howto;
252 
253 	/* We are out of the debugger now. */
254 	kdb_active = 0;
255 
256 #ifdef SMP
257 	if (smp_active)
258 		printf("boot() called on cpu#%d\n", PCPU_GET(cpuid));
259 #endif
260 	/*
261 	 * Do any callouts that should be done BEFORE syncing the filesystems.
262 	 */
263 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
264 
265 	/*
266 	 * Now sync filesystems
267 	 */
268 	if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
269 		register struct buf *bp;
270 		int iter, nbusy, pbusy;
271 #ifndef PREEMPTION
272 		int subiter;
273 #endif
274 
275 		waittime = 0;
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 			if (first_buf_printf) {
299 				printf("syncing disks, buffers remaining... ");
300 				first_buf_printf = 0;
301 			}
302 			printf("%d ", nbusy);
303 			if (nbusy < pbusy)
304 				iter = 0;
305 			pbusy = nbusy;
306 			sync(&thread0, NULL);
307 
308 #ifdef PREEMPTION
309 			/*
310 			 * Drop Giant and spin for a while to allow
311 			 * interrupt threads to run.
312 			 */
313 			DROP_GIANT();
314 			DELAY(50000 * iter);
315 			PICKUP_GIANT();
316 #else
317 			/*
318 			 * Drop Giant and context switch several times to
319 			 * allow interrupt threads to run.
320 			 */
321 			DROP_GIANT();
322 			for (subiter = 0; subiter < 50 * iter; subiter++) {
323 				mtx_lock_spin(&sched_lock);
324 				mi_switch(SW_VOL, NULL);
325 				mtx_unlock_spin(&sched_lock);
326 				DELAY(1000);
327 			}
328 			PICKUP_GIANT();
329 #endif
330 		}
331 
332 		/*
333 		 * Count only busy local buffers to prevent forcing
334 		 * a fsck if we're just a client of a wedged NFS server
335 		 */
336 		nbusy = 0;
337 		for (bp = &buf[nbuf]; --bp >= buf; ) {
338 			if (((bp->b_flags&B_INVAL) == 0 && BUF_REFCNT(bp)) ||
339 			    ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) {
340 				if (bp->b_dev == NULL) {
341 					TAILQ_REMOVE(&mountlist,
342 					    bp->b_vp->v_mount, mnt_list);
343 					continue;
344 				}
345 				nbusy++;
346 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC)
347 				printf(
348 			    "%d: dev:%s, flags:%0x, blkno:%ld, lblkno:%ld\n",
349 				    nbusy, devtoname(bp->b_dev),
350 				    bp->b_flags, (long)bp->b_blkno,
351 				    (long)bp->b_lblkno);
352 #endif
353 			}
354 		}
355 		if (nbusy) {
356 			/*
357 			 * Failed to sync all blocks. Indicate this and don't
358 			 * unmount filesystems (thus forcing an fsck on reboot).
359 			 */
360 			printf("giving up on %d buffers\n", nbusy);
361 			DELAY(5000000);	/* 5 seconds */
362 		} else {
363 			printf("done\n");
364 			/*
365 			 * Unmount filesystems
366 			 */
367 			if (panicstr == 0)
368 				vfs_unmountall();
369 		}
370 		DELAY(100000);		/* wait for console output to finish */
371 	}
372 
373 	print_uptime();
374 
375 	/*
376 	 * Ok, now do things that assume all filesystem activity has
377 	 * been completed.
378 	 */
379 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
380 	splhigh();
381 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP &&
382 	    !cold && dumper.dumper != NULL && !dumping)
383 		doadump();
384 
385 	/* Now that we're going to really halt the system... */
386 	EVENTHANDLER_INVOKE(shutdown_final, howto);
387 
388 	for(;;) ;	/* safety against shutdown_reset not working */
389 	/* NOTREACHED */
390 }
391 
392 /*
393  * If the shutdown was a clean halt, behave accordingly.
394  */
395 static void
396 shutdown_halt(void *junk, int howto)
397 {
398 
399 	if (howto & RB_HALT) {
400 		printf("\n");
401 		printf("The operating system has halted.\n");
402 		printf("Please press any key to reboot.\n\n");
403 		switch (cngetc()) {
404 		case -1:		/* No console, just die */
405 			cpu_halt();
406 			/* NOTREACHED */
407 		default:
408 			howto &= ~RB_HALT;
409 			break;
410 		}
411 	}
412 }
413 
414 /*
415  * Check to see if the system paniced, pause and then reboot
416  * according to the specified delay.
417  */
418 static void
419 shutdown_panic(void *junk, int howto)
420 {
421 	int loop;
422 
423 	if (howto & RB_DUMP) {
424 		if (PANIC_REBOOT_WAIT_TIME != 0) {
425 			if (PANIC_REBOOT_WAIT_TIME != -1) {
426 				printf("Automatic reboot in %d seconds - "
427 				       "press a key on the console to abort\n",
428 					PANIC_REBOOT_WAIT_TIME);
429 				for (loop = PANIC_REBOOT_WAIT_TIME * 10;
430 				     loop > 0; --loop) {
431 					DELAY(1000 * 100); /* 1/10th second */
432 					/* Did user type a key? */
433 					if (cncheckc() != -1)
434 						break;
435 				}
436 				if (!loop)
437 					return;
438 			}
439 		} else { /* zero time specified - reboot NOW */
440 			return;
441 		}
442 		printf("--> Press a key on the console to reboot,\n");
443 		printf("--> or switch off the system now.\n");
444 		cngetc();
445 	}
446 }
447 
448 /*
449  * Everything done, now reset
450  */
451 static void
452 shutdown_reset(void *junk, int howto)
453 {
454 
455 	printf("Rebooting...\n");
456 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
457 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
458 	cpu_reset();
459 	/* NOTREACHED */ /* assuming reset worked */
460 }
461 
462 #ifdef SMP
463 static u_int panic_cpu = NOCPU;
464 #endif
465 
466 /*
467  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
468  * and then reboots.  If we are called twice, then we avoid trying to sync
469  * the disks as this often leads to recursive panics.
470  *
471  * MPSAFE
472  */
473 void
474 panic(const char *fmt, ...)
475 {
476 	struct thread *td = curthread;
477 	int bootopt, newpanic;
478 	va_list ap;
479 	static char buf[256];
480 
481 #ifdef SMP
482 	/*
483 	 * We don't want multiple CPU's to panic at the same time, so we
484 	 * use panic_cpu as a simple spinlock.  We have to keep checking
485 	 * panic_cpu if we are spinning in case the panic on the first
486 	 * CPU is canceled.
487 	 */
488 	if (panic_cpu != PCPU_GET(cpuid))
489 		while (atomic_cmpset_int(&panic_cpu, NOCPU,
490 		    PCPU_GET(cpuid)) == 0)
491 			while (panic_cpu != NOCPU)
492 				; /* nothing */
493 #endif
494 
495 	bootopt = RB_AUTOBOOT | RB_DUMP;
496 	newpanic = 0;
497 	if (panicstr)
498 		bootopt |= RB_NOSYNC;
499 	else {
500 		panicstr = fmt;
501 		newpanic = 1;
502 	}
503 
504 	va_start(ap, fmt);
505 	if (newpanic) {
506 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
507 		panicstr = buf;
508 		printf("panic: %s\n", buf);
509 	} else {
510 		printf("panic: ");
511 		vprintf(fmt, ap);
512 		printf("\n");
513 	}
514 	va_end(ap);
515 #ifdef SMP
516 	/* two separate prints in case of an unmapped page and trap */
517 	printf("cpuid = %d; ", PCPU_GET(cpuid));
518 #ifdef APIC_IO
519 	printf("lapic.id = %08x\n", lapic.id);
520 #else
521 	printf("\n");
522 #endif
523 #endif
524 
525 #ifdef KDB
526 	if (newpanic && trace_on_panic)
527 		kdb_backtrace();
528 	if (debugger_on_panic)
529 		kdb_enter("panic");
530 #ifdef RESTARTABLE_PANICS
531 	/* See if the user aborted the panic, in which case we continue. */
532 	if (panicstr == NULL) {
533 #ifdef SMP
534 		atomic_store_rel_int(&panic_cpu, NOCPU);
535 #endif
536 		return;
537 	}
538 #endif
539 #endif
540 	mtx_lock_spin(&sched_lock);
541 	td->td_flags |= TDF_INPANIC;
542 	mtx_unlock_spin(&sched_lock);
543 	if (!sync_on_panic)
544 		bootopt |= RB_NOSYNC;
545 	boot(bootopt);
546 }
547 
548 /*
549  * Support for poweroff delay.
550  */
551 #ifndef POWEROFF_DELAY
552 # define POWEROFF_DELAY 5000
553 #endif
554 static int poweroff_delay = POWEROFF_DELAY;
555 
556 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
557 	&poweroff_delay, 0, "");
558 
559 static void
560 poweroff_wait(void *junk, int howto)
561 {
562 
563 	if (!(howto & RB_POWEROFF) || poweroff_delay <= 0)
564 		return;
565 	DELAY(poweroff_delay * 1000);
566 }
567 
568 /*
569  * Some system processes (e.g. syncer) need to be stopped at appropriate
570  * points in their main loops prior to a system shutdown, so that they
571  * won't interfere with the shutdown process (e.g. by holding a disk buf
572  * to cause sync to fail).  For each of these system processes, register
573  * shutdown_kproc() as a handler for one of shutdown events.
574  */
575 static int kproc_shutdown_wait = 60;
576 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
577     &kproc_shutdown_wait, 0, "");
578 
579 void
580 kproc_shutdown(void *arg, int howto)
581 {
582 	struct proc *p;
583 	char procname[MAXCOMLEN + 1];
584 	int error;
585 
586 	if (panicstr)
587 		return;
588 
589 	p = (struct proc *)arg;
590 	strlcpy(procname, p->p_comm, sizeof(procname));
591 	printf("Waiting (max %d seconds) for system process `%s' to stop...\n",
592 	    kproc_shutdown_wait, procname);
593 	error = kthread_suspend(p, kproc_shutdown_wait * hz);
594 
595 	if (error == EWOULDBLOCK)
596 		printf("Stop of '%s' timed out.\n", procname);
597 	else
598 		printf("Process '%s' stopped.\n", procname);
599 }
600 
601 /* Registration of dumpers */
602 int
603 set_dumper(struct dumperinfo *di)
604 {
605 
606 	if (di == NULL) {
607 		bzero(&dumper, sizeof dumper);
608 		return (0);
609 	}
610 	if (dumper.dumper != NULL)
611 		return (EBUSY);
612 	dumper = *di;
613 	return (0);
614 }
615 
616 #if defined(__powerpc__)
617 void
618 dumpsys(struct dumperinfo *di __unused)
619 {
620 
621 	printf("Kernel dumps not implemented on this architecture\n");
622 }
623 #endif
624