xref: /dflybsd-src/sys/kern/kern_shutdown.c (revision 53e987cee557d989dbf172d8a3c2ade9ea6fc46f)
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: src/sys/kern/kern_shutdown.c,v 1.72.2.12 2002/02/21 19:15:10 dillon Exp $
40  * $DragonFly: src/sys/kern/kern_shutdown.c,v 1.19 2005/04/19 17:54:42 dillon Exp $
41  */
42 
43 #include "opt_ddb.h"
44 #include "opt_ddb_trace.h"
45 #include "opt_hw_wdog.h"
46 #include "opt_panic.h"
47 #include "opt_show_busybufs.h"
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/eventhandler.h>
52 #include <sys/buf.h>
53 #include <sys/disklabel.h>
54 #include <sys/reboot.h>
55 #include <sys/proc.h>
56 #include <sys/vnode.h>
57 #include <sys/kernel.h>
58 #include <sys/kthread.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/queue.h>
62 #include <sys/sysctl.h>
63 #include <sys/conf.h>
64 #include <sys/sysproto.h>
65 #include <sys/device.h>
66 #include <sys/cons.h>
67 #include <sys/buf2.h>
68 
69 #include <machine/pcb.h>
70 #include <machine/clock.h>
71 #include <machine/md_var.h>
72 #include <machine/smp.h>		/* smp_active_mask, cpuid */
73 
74 #include <sys/signalvar.h>
75 
76 #ifndef PANIC_REBOOT_WAIT_TIME
77 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
78 #endif
79 
80 /*
81  * Note that stdarg.h and the ANSI style va_start macro is used for both
82  * ANSI and traditional C compilers.  We use the machine version to stay
83  * within the confines of the kernel header files.
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 extern void db_print_backtrace(void);
97 
98 #ifdef DDB_TRACE
99 int trace_on_panic = 1;
100 #else
101 int trace_on_panic = 0;
102 #endif
103 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW,
104 	&trace_on_panic, 0, "Print stack trace on kernel panic");
105 #endif
106 
107 static int sync_on_panic = 1;
108 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW,
109 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
110 
111 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment");
112 
113 #ifdef	HW_WDOG
114 /*
115  * If there is a hardware watchdog, point this at the function needed to
116  * hold it off.
117  * It's needed when the kernel needs to do some lengthy operations.
118  * e.g. in wd.c when dumping core.. It's most annoying to have
119  * your precious core-dump only half written because the wdog kicked in.
120  */
121 watchdog_tickle_fn wdog_tickler = NULL;
122 #endif	/* HW_WDOG */
123 
124 /*
125  * Variable panicstr contains argument to first call to panic; used as flag
126  * to indicate that the kernel has already called panic.
127  */
128 const char *panicstr;
129 
130 int dumping;				/* system is dumping */
131 
132 static void boot (int) __dead2;
133 static void dumpsys (void);
134 static int setdumpdev (dev_t dev);
135 static void poweroff_wait (void *, int);
136 static void print_uptime (void);
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 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, SHUTDOWN_PRI_FIRST);
146 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, SHUTDOWN_PRI_LAST + 100);
147 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, SHUTDOWN_PRI_LAST + 100);
148 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL, SHUTDOWN_PRI_LAST + 200);
149 }
150 
151 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL)
152 
153 /* ARGSUSED */
154 
155 /*
156  * The system call that results in a reboot
157  */
158 int
159 reboot(struct reboot_args *uap)
160 {
161 	struct thread *td = curthread;
162 	int error;
163 
164 	if ((error = suser(td)))
165 		return (error);
166 
167 	boot(uap->opt);
168 	return (0);
169 }
170 
171 /*
172  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
173  */
174 static int shutdown_howto = 0;
175 
176 void
177 shutdown_nice(int howto)
178 {
179 	shutdown_howto = howto;
180 
181 	/* Send a signal to init(8) and have it shutdown the world */
182 	if (initproc != NULL) {
183 		psignal(initproc, SIGINT);
184 	} else {
185 		/* No init(8) running, so simply reboot */
186 		boot(RB_NOSYNC);
187 	}
188 	return;
189 }
190 static int	waittime = -1;
191 static struct pcb dumppcb;
192 
193 static void
194 print_uptime()
195 {
196 	int f;
197 	struct timespec ts;
198 
199 	getnanouptime(&ts);
200 	printf("Uptime: ");
201 	f = 0;
202 	if (ts.tv_sec >= 86400) {
203 		printf("%ldd", ts.tv_sec / 86400);
204 		ts.tv_sec %= 86400;
205 		f = 1;
206 	}
207 	if (f || ts.tv_sec >= 3600) {
208 		printf("%ldh", ts.tv_sec / 3600);
209 		ts.tv_sec %= 3600;
210 		f = 1;
211 	}
212 	if (f || ts.tv_sec >= 60) {
213 		printf("%ldm", ts.tv_sec / 60);
214 		ts.tv_sec %= 60;
215 		f = 1;
216 	}
217 	printf("%lds\n", ts.tv_sec);
218 }
219 
220 /*
221  *  Go through the rigmarole of shutting down..
222  * this used to be in machdep.c but I'll be dammned if I could see
223  * anything machine dependant in it.
224  */
225 static void
226 boot(int howto)
227 {
228 
229 	/* collect extra flags that shutdown_nice might have set */
230 	howto |= shutdown_howto;
231 
232 #ifdef SMP
233 	if (smp_active_mask > 1) {
234 		printf("boot() called on cpu#%d\n", mycpu->gd_cpuid);
235 	}
236 #endif
237 	/*
238 	 * Do any callouts that should be done BEFORE syncing the filesystems.
239 	 */
240 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
241 
242 	/*
243 	 * Now sync filesystems
244 	 */
245 	if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
246 		struct buf *bp;
247 		int iter, nbusy, pbusy;
248 
249 		waittime = 0;
250 		printf("\nsyncing disks... ");
251 
252 		sync(NULL);	/* YYY was sync(&proc0, NULL). why proc0 ? */
253 
254 		/*
255 		 * With soft updates, some buffers that are
256 		 * written will be remarked as dirty until other
257 		 * buffers are written.
258 		 */
259 		for (iter = pbusy = 0; iter < 20; iter++) {
260 			nbusy = 0;
261 			for (bp = &buf[nbuf]; --bp >= buf; ) {
262 				if ((bp->b_flags & B_INVAL) == 0 &&
263 				    BUF_REFCNT(bp) > 0) {
264 					nbusy++;
265 				} else if ((bp->b_flags & (B_DELWRI | B_INVAL))
266 						== B_DELWRI) {
267 					/* bawrite(bp);*/
268 					nbusy++;
269 				}
270 			}
271 			if (nbusy == 0)
272 				break;
273 			printf("%d ", nbusy);
274 			if (nbusy < pbusy)
275 				iter = 0;
276 			pbusy = nbusy;
277 			/*
278 			 * XXX:
279 			 * Process soft update work queue if buffers don't sync
280 			 * after 6 iterations by permitting the syncer to run.
281 			 */
282 			if (iter > 5 && bioops.io_sync)
283 				(*bioops.io_sync)(NULL);
284 			sync(NULL); /* YYY was sync(&proc0, NULL). why proc0 ? */
285 			DELAY(50000 * iter);
286 		}
287 		printf("\n");
288 		/*
289 		 * Count only busy local buffers to prevent forcing
290 		 * a fsck if we're just a client of a wedged NFS server
291 		 */
292 		nbusy = 0;
293 		for (bp = &buf[nbuf]; --bp >= buf; ) {
294 			if (((bp->b_flags&B_INVAL) == 0 && BUF_REFCNT(bp)) ||
295 			    ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) {
296 				if (bp->b_dev == NODEV) {
297 					mountlist_remove(bp->b_vp->v_mount);
298 					continue;
299 				}
300 				nbusy++;
301 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC)
302 				printf(
303 			    "%p %d: dev:%s, flags:%08lx, blkno:%ld, lblkno:%ld\n",
304 				    bp, nbusy, devtoname(bp->b_dev),
305 				    bp->b_flags, (long)bp->b_blkno,
306 				    (long)bp->b_lblkno);
307 #endif
308 			}
309 		}
310 		if (nbusy) {
311 			/*
312 			 * Failed to sync all blocks. Indicate this and don't
313 			 * unmount filesystems (thus forcing an fsck on reboot).
314 			 */
315 			printf("giving up on %d buffers\n", nbusy);
316 #ifdef DDB
317 			Debugger("busy buffer problem");
318 #endif /* DDB */
319 			DELAY(5000000);	/* 5 seconds */
320 		} else {
321 			printf("done\n");
322 			/*
323 			 * Unmount filesystems
324 			 */
325 			if (panicstr == 0)
326 				vfs_unmountall();
327 		}
328 		DELAY(100000);		/* wait for console output to finish */
329 	}
330 
331 	print_uptime();
332 
333 	/*
334 	 * Ok, now do things that assume all filesystem activity has
335 	 * been completed.
336 	 */
337 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
338 	splhigh();
339 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold)
340 		dumpsys();
341 
342 	/* Now that we're going to really halt the system... */
343 	EVENTHANDLER_INVOKE(shutdown_final, howto);
344 
345 	for(;;) ;	/* safety against shutdown_reset not working */
346 	/* NOTREACHED */
347 }
348 
349 /*
350  * If the shutdown was a clean halt, behave accordingly.
351  */
352 static void
353 shutdown_halt(void *junk, int howto)
354 {
355 	if (howto & RB_HALT) {
356 		printf("\n");
357 		printf("The operating system has halted.\n");
358 		printf("Please press any key to reboot.\n\n");
359 		switch (cngetc()) {
360 		case -1:		/* No console, just die */
361 			cpu_halt();
362 			/* NOTREACHED */
363 		default:
364 			howto &= ~RB_HALT;
365 			break;
366 		}
367 	}
368 }
369 
370 /*
371  * Check to see if the system paniced, pause and then reboot
372  * according to the specified delay.
373  */
374 static void
375 shutdown_panic(void *junk, int howto)
376 {
377 	int loop;
378 
379 	if (howto & RB_DUMP) {
380 		if (PANIC_REBOOT_WAIT_TIME != 0) {
381 			if (PANIC_REBOOT_WAIT_TIME != -1) {
382 				printf("Automatic reboot in %d seconds - "
383 				       "press a key on the console to abort\n",
384 					PANIC_REBOOT_WAIT_TIME);
385 				for (loop = PANIC_REBOOT_WAIT_TIME * 10;
386 				     loop > 0; --loop) {
387 					DELAY(1000 * 100); /* 1/10th second */
388 					/* Did user type a key? */
389 					if (cncheckc() != -1)
390 						break;
391 				}
392 				if (!loop)
393 					return;
394 			}
395 		} else { /* zero time specified - reboot NOW */
396 			return;
397 		}
398 		printf("--> Press a key on the console to reboot,\n");
399 		printf("--> or switch off the system now.\n");
400 		cngetc();
401 	}
402 }
403 
404 /*
405  * Everything done, now reset
406  */
407 static void
408 shutdown_reset(void *junk, int howto)
409 {
410 	printf("Rebooting...\n");
411 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
412 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
413 	cpu_reset();
414 	/* NOTREACHED */ /* assuming reset worked */
415 }
416 
417 /*
418  * Magic number for savecore
419  *
420  * exported (symorder) and used at least by savecore(8)
421  *
422  */
423 static u_long const	dumpmag = 0x8fca0101UL;
424 
425 static int	dumpsize = 0;		/* also for savecore */
426 
427 static int	dodump = 1;
428 
429 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0,
430     "Try to perform coredump on kernel panic");
431 
432 static int
433 setdumpdev(dev)
434 	dev_t dev;
435 {
436 	int psize;
437 	long newdumplo;
438 
439 	if (dev == NODEV) {
440 		dumpdev = dev;
441 		return (0);
442 	}
443 	psize = dev_dpsize(dev);
444 	if (psize == -1)
445 		return (ENXIO);
446 	/*
447 	 * XXX should clean up checking in dumpsys() to be more like this.
448 	 */
449 	newdumplo = psize - Maxmem * (PAGE_SIZE / DEV_BSIZE);
450 	if (newdumplo <= LABELSECTOR)
451 		return (ENOSPC);
452 	dumpdev = dev;
453 	dumplo = newdumplo;
454 	return (0);
455 }
456 
457 
458 /* ARGSUSED */
459 static void dump_conf (void *dummy);
460 static void
461 dump_conf(dummy)
462 	void *dummy;
463 {
464 	char *path;
465 	dev_t dev;
466 
467 	path = malloc(MNAMELEN, M_TEMP, M_WAITOK);
468 	if (TUNABLE_STR_FETCH("dumpdev", path, MNAMELEN) != 0) {
469 		dev = getdiskbyname(path);
470 		if (dev != NODEV)
471 			dumpdev = dev;
472 	}
473 	free(path, M_TEMP);
474 	if (setdumpdev(dumpdev) != 0)
475 		dumpdev = NODEV;
476 }
477 
478 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL)
479 
480 static int
481 sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS)
482 {
483 	int error;
484 	udev_t ndumpdev;
485 
486 	ndumpdev = dev2udev(dumpdev);
487 	error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req);
488 	if (error == 0 && req->newptr != NULL)
489 		error = setdumpdev(udev2dev(ndumpdev, 0));
490 	return (error);
491 }
492 
493 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW,
494 	0, sizeof dumpdev, sysctl_kern_dumpdev, "T,dev_t", "");
495 
496 /*
497  * Doadump comes here after turning off memory management and
498  * getting on the dump stack, either when called above, or by
499  * the auto-restart code.
500  */
501 static void
502 dumpsys(void)
503 {
504 	int	error;
505 
506 	savectx(&dumppcb);
507 	if (dumping++) {
508 		printf("Dump already in progress, bailing...\n");
509 		return;
510 	}
511 	if (!dodump)
512 		return;
513 	if (dumpdev == NODEV)
514 		return;
515 	dumpsize = Maxmem;
516 	printf("\ndumping to dev %s, offset %ld\n", devtoname(dumpdev), dumplo);
517 	printf("dump ");
518 	error = dev_ddump(dumpdev);
519 	if (error == 0) {
520 		printf("succeeded\n");
521 		return;
522 	}
523 	printf("failed, reason: ");
524 	switch (error) {
525 	case ENOSYS:
526 	case ENODEV:
527 		printf("device doesn't support a dump routine\n");
528 		break;
529 
530 	case ENXIO:
531 		printf("device bad\n");
532 		break;
533 
534 	case EFAULT:
535 		printf("device not ready\n");
536 		break;
537 
538 	case EINVAL:
539 		printf("area improper\n");
540 		break;
541 
542 	case EIO:
543 		printf("i/o error\n");
544 		break;
545 
546 	case EINTR:
547 		printf("aborted from console\n");
548 		break;
549 
550 	default:
551 		printf("unknown, error = %d\n", error);
552 		break;
553 	}
554 }
555 
556 int
557 dumpstatus(vm_offset_t addr, off_t count)
558 {
559 	int c;
560 
561 	if (addr % (1024 * 1024) == 0) {
562 #ifdef HW_WDOG
563 		if (wdog_tickler)
564 			(*wdog_tickler)();
565 #endif
566 		printf("%ld ", (long)(count / (1024 * 1024)));
567 	}
568 
569 	if ((c = cncheckc()) == 0x03)
570 		return -1;
571 	else if (c != -1)
572 		printf("[CTRL-C to abort] ");
573 
574 	return 0;
575 }
576 
577 /*
578  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
579  * and then reboots.  If we are called twice, then we avoid trying to sync
580  * the disks as this often leads to recursive panics.
581  */
582 void
583 panic(const char *fmt, ...)
584 {
585 	int bootopt, newpanic;
586 	__va_list ap;
587 	static char buf[256];
588 
589 	bootopt = RB_AUTOBOOT | RB_DUMP;
590 	if (sync_on_panic == 0)
591 		bootopt |= RB_NOSYNC;
592 	newpanic = 0;
593 	if (panicstr)
594 		bootopt |= RB_NOSYNC;
595 	else {
596 		panicstr = fmt;
597 		newpanic = 1;
598 	}
599 
600 	__va_start(ap, fmt);
601 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
602 	if (panicstr == fmt)
603 		panicstr = buf;
604 	__va_end(ap);
605 	printf("panic: %s\n", buf);
606 #ifdef SMP
607 	/* three separate prints in case of an unmapped page and trap */
608 	printf("mp_lock = %08x; ", mp_lock);
609 	printf("cpuid = %d; ", mycpu->gd_cpuid);
610 	printf("lapic.id = %08x\n", lapic.id);
611 #endif
612 
613 #if defined(DDB)
614 	if (newpanic && trace_on_panic)
615 		db_print_backtrace();
616 	if (debugger_on_panic)
617 		Debugger ("panic");
618 #endif
619 	boot(bootopt);
620 }
621 
622 /*
623  * Support for poweroff delay.
624  */
625 #ifndef POWEROFF_DELAY
626 # define POWEROFF_DELAY 5000
627 #endif
628 static int poweroff_delay = POWEROFF_DELAY;
629 
630 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
631 	&poweroff_delay, 0, "");
632 
633 static void
634 poweroff_wait(void *junk, int howto)
635 {
636 	if(!(howto & RB_POWEROFF) || poweroff_delay <= 0)
637 		return;
638 	DELAY(poweroff_delay * 1000);
639 }
640 
641 /*
642  * Some system processes (e.g. syncer) need to be stopped at appropriate
643  * points in their main loops prior to a system shutdown, so that they
644  * won't interfere with the shutdown process (e.g. by holding a disk buf
645  * to cause sync to fail).  For each of these system processes, register
646  * shutdown_kproc() as a handler for one of shutdown events.
647  */
648 static int kproc_shutdown_wait = 60;
649 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
650     &kproc_shutdown_wait, 0, "");
651 
652 void
653 shutdown_kproc(void *arg, int howto)
654 {
655 	struct thread *td;
656 	struct proc *p;
657 	int error;
658 
659 	if (panicstr)
660 		return;
661 
662 	td = (struct thread *)arg;
663 	if ((p = td->td_proc) != NULL) {
664 	    printf("Waiting (max %d seconds) for system process `%s' to stop...",
665 		kproc_shutdown_wait, p->p_comm);
666 	} else {
667 	    printf("Waiting (max %d seconds) for system thread %s to stop...",
668 		kproc_shutdown_wait, td->td_comm);
669 	}
670 	error = suspend_kproc(td, kproc_shutdown_wait * hz);
671 
672 	if (error == EWOULDBLOCK)
673 		printf("timed out\n");
674 	else
675 		printf("stopped\n");
676 }
677