xref: /freebsd/sys/kern/kern_shutdown.c (revision 298f5fdc242b760e70cd3494e3a4f1f50b20664d)
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_kdb.h"
42 #include "opt_panic.h"
43 #include "opt_sched.h"
44 #include "opt_watchdog.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/jail.h>
54 #include <sys/kdb.h>
55 #include <sys/kernel.h>
56 #include <sys/kerneldump.h>
57 #include <sys/kthread.h>
58 #include <sys/ktr.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/priv.h>
62 #include <sys/proc.h>
63 #include <sys/reboot.h>
64 #include <sys/resourcevar.h>
65 #include <sys/rwlock.h>
66 #include <sys/sched.h>
67 #include <sys/smp.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysproto.h>
70 #include <sys/vnode.h>
71 #include <sys/watchdog.h>
72 
73 #include <ddb/ddb.h>
74 
75 #include <machine/cpu.h>
76 #include <machine/pcb.h>
77 #include <machine/smp.h>
78 
79 #include <security/mac/mac_framework.h>
80 
81 #include <vm/vm.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_pager.h>
85 #include <vm/swap_pager.h>
86 
87 #include <sys/signalvar.h>
88 
89 #ifndef PANIC_REBOOT_WAIT_TIME
90 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
91 #endif
92 static int panic_reboot_wait_time = PANIC_REBOOT_WAIT_TIME;
93 SYSCTL_INT(_kern, OID_AUTO, panic_reboot_wait_time, CTLFLAG_RWTUN,
94     &panic_reboot_wait_time, 0,
95     "Seconds to wait before rebooting after a panic");
96 
97 /*
98  * Note that stdarg.h and the ANSI style va_start macro is used for both
99  * ANSI and traditional C compilers.
100  */
101 #include <machine/stdarg.h>
102 
103 #ifdef KDB
104 #ifdef KDB_UNATTENDED
105 int debugger_on_panic = 0;
106 #else
107 int debugger_on_panic = 1;
108 #endif
109 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic,
110     CTLFLAG_RWTUN | CTLFLAG_SECURE,
111     &debugger_on_panic, 0, "Run debugger on kernel panic");
112 
113 #ifdef KDB_TRACE
114 static int trace_on_panic = 1;
115 #else
116 static int trace_on_panic = 0;
117 #endif
118 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
119     CTLFLAG_RWTUN | CTLFLAG_SECURE,
120     &trace_on_panic, 0, "Print stack trace on kernel panic");
121 #endif /* KDB */
122 
123 static int sync_on_panic = 0;
124 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN,
125 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
126 
127 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0,
128     "Shutdown environment");
129 
130 #ifndef DIAGNOSTIC
131 static int show_busybufs;
132 #else
133 static int show_busybufs = 1;
134 #endif
135 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
136 	&show_busybufs, 0, "");
137 
138 /*
139  * Variable panicstr contains argument to first call to panic; used as flag
140  * to indicate that the kernel has already called panic.
141  */
142 const char *panicstr;
143 
144 int dumping;				/* system is dumping */
145 int rebooting;				/* system is rebooting */
146 static struct dumperinfo dumper;	/* our selected dumper */
147 
148 /* Context information for dump-debuggers. */
149 static struct pcb dumppcb;		/* Registers. */
150 lwpid_t dumptid;			/* Thread ID. */
151 
152 static void poweroff_wait(void *, int);
153 static void shutdown_halt(void *junk, int howto);
154 static void shutdown_panic(void *junk, int howto);
155 static void shutdown_reset(void *junk, int howto);
156 static void vpanic(const char *fmt, va_list ap) __dead2;
157 
158 /* register various local shutdown events */
159 static void
160 shutdown_conf(void *unused)
161 {
162 
163 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
164 	    SHUTDOWN_PRI_FIRST);
165 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
166 	    SHUTDOWN_PRI_LAST + 100);
167 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
168 	    SHUTDOWN_PRI_LAST + 100);
169 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
170 	    SHUTDOWN_PRI_LAST + 200);
171 }
172 
173 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
174 
175 /*
176  * The system call that results in a reboot.
177  */
178 /* ARGSUSED */
179 int
180 sys_reboot(struct thread *td, struct reboot_args *uap)
181 {
182 	int error;
183 
184 	error = 0;
185 #ifdef MAC
186 	error = mac_system_check_reboot(td->td_ucred, uap->opt);
187 #endif
188 	if (error == 0)
189 		error = priv_check(td, PRIV_REBOOT);
190 	if (error == 0) {
191 		mtx_lock(&Giant);
192 		kern_reboot(uap->opt);
193 		mtx_unlock(&Giant);
194 	}
195 	return (error);
196 }
197 
198 /*
199  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
200  */
201 void
202 shutdown_nice(int howto)
203 {
204 
205 	if (initproc != NULL) {
206 		/* Send a signal to init(8) and have it shutdown the world. */
207 		PROC_LOCK(initproc);
208 		if (howto & RB_POWEROFF)
209 			kern_psignal(initproc, SIGUSR2);
210 		else if (howto & RB_HALT)
211 			kern_psignal(initproc, SIGUSR1);
212 		else
213 			kern_psignal(initproc, SIGINT);
214 		PROC_UNLOCK(initproc);
215 	} else {
216 		/* No init(8) running, so simply reboot. */
217 		kern_reboot(howto | RB_NOSYNC);
218 	}
219 }
220 
221 static void
222 print_uptime(void)
223 {
224 	int f;
225 	struct timespec ts;
226 
227 	getnanouptime(&ts);
228 	printf("Uptime: ");
229 	f = 0;
230 	if (ts.tv_sec >= 86400) {
231 		printf("%ldd", (long)ts.tv_sec / 86400);
232 		ts.tv_sec %= 86400;
233 		f = 1;
234 	}
235 	if (f || ts.tv_sec >= 3600) {
236 		printf("%ldh", (long)ts.tv_sec / 3600);
237 		ts.tv_sec %= 3600;
238 		f = 1;
239 	}
240 	if (f || ts.tv_sec >= 60) {
241 		printf("%ldm", (long)ts.tv_sec / 60);
242 		ts.tv_sec %= 60;
243 		f = 1;
244 	}
245 	printf("%lds\n", (long)ts.tv_sec);
246 }
247 
248 int
249 doadump(boolean_t textdump)
250 {
251 	boolean_t coredump;
252 
253 	if (dumping)
254 		return (EBUSY);
255 	if (dumper.dumper == NULL)
256 		return (ENXIO);
257 
258 	savectx(&dumppcb);
259 	dumptid = curthread->td_tid;
260 	dumping++;
261 
262 	coredump = TRUE;
263 #ifdef DDB
264 	if (textdump && textdump_pending) {
265 		coredump = FALSE;
266 		textdump_dumpsys(&dumper);
267 	}
268 #endif
269 	if (coredump)
270 		dumpsys(&dumper);
271 
272 	dumping--;
273 	return (0);
274 }
275 
276 static int
277 isbufbusy(struct buf *bp)
278 {
279 	if (((bp->b_flags & (B_INVAL | B_PERSISTENT)) == 0 &&
280 	    BUF_ISLOCKED(bp)) ||
281 	    ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI))
282 		return (1);
283 	return (0);
284 }
285 
286 /*
287  * Shutdown the system cleanly to prepare for reboot, halt, or power off.
288  */
289 void
290 kern_reboot(int howto)
291 {
292 	static int first_buf_printf = 1;
293 	static int waittime = -1;
294 
295 #if defined(SMP)
296 	/*
297 	 * Bind us to CPU 0 so that all shutdown code runs there.  Some
298 	 * systems don't shutdown properly (i.e., ACPI power off) if we
299 	 * run on another processor.
300 	 */
301 	if (!SCHEDULER_STOPPED()) {
302 		thread_lock(curthread);
303 		sched_bind(curthread, 0);
304 		thread_unlock(curthread);
305 		KASSERT(PCPU_GET(cpuid) == 0, ("boot: not running on cpu 0"));
306 	}
307 #endif
308 	/* We're in the process of rebooting. */
309 	rebooting = 1;
310 
311 	/* We are out of the debugger now. */
312 	kdb_active = 0;
313 
314 	/*
315 	 * Do any callouts that should be done BEFORE syncing the filesystems.
316 	 */
317 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
318 
319 	/*
320 	 * Now sync filesystems
321 	 */
322 	if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
323 		register struct buf *bp;
324 		int iter, nbusy, pbusy;
325 #ifndef PREEMPTION
326 		int subiter;
327 #endif
328 
329 		waittime = 0;
330 
331 		wdog_kern_pat(WD_LASTVAL);
332 		sys_sync(curthread, NULL);
333 
334 		/*
335 		 * With soft updates, some buffers that are
336 		 * written will be remarked as dirty until other
337 		 * buffers are written.
338 		 */
339 		for (iter = pbusy = 0; iter < 20; iter++) {
340 			nbusy = 0;
341 			for (bp = &buf[nbuf]; --bp >= buf; )
342 				if (isbufbusy(bp))
343 					nbusy++;
344 			if (nbusy == 0) {
345 				if (first_buf_printf)
346 					printf("All buffers synced.");
347 				break;
348 			}
349 			if (first_buf_printf) {
350 				printf("Syncing disks, buffers remaining... ");
351 				first_buf_printf = 0;
352 			}
353 			printf("%d ", nbusy);
354 			if (nbusy < pbusy)
355 				iter = 0;
356 			pbusy = nbusy;
357 
358 			wdog_kern_pat(WD_LASTVAL);
359 			sys_sync(curthread, NULL);
360 
361 #ifdef PREEMPTION
362 			/*
363 			 * Drop Giant and spin for a while to allow
364 			 * interrupt threads to run.
365 			 */
366 			DROP_GIANT();
367 			DELAY(50000 * iter);
368 			PICKUP_GIANT();
369 #else
370 			/*
371 			 * Drop Giant and context switch several times to
372 			 * allow interrupt threads to run.
373 			 */
374 			DROP_GIANT();
375 			for (subiter = 0; subiter < 50 * iter; subiter++) {
376 				thread_lock(curthread);
377 				mi_switch(SW_VOL, NULL);
378 				thread_unlock(curthread);
379 				DELAY(1000);
380 			}
381 			PICKUP_GIANT();
382 #endif
383 		}
384 		printf("\n");
385 		/*
386 		 * Count only busy local buffers to prevent forcing
387 		 * a fsck if we're just a client of a wedged NFS server
388 		 */
389 		nbusy = 0;
390 		for (bp = &buf[nbuf]; --bp >= buf; ) {
391 			if (isbufbusy(bp)) {
392 #if 0
393 /* XXX: This is bogus.  We should probably have a BO_REMOTE flag instead */
394 				if (bp->b_dev == NULL) {
395 					TAILQ_REMOVE(&mountlist,
396 					    bp->b_vp->v_mount, mnt_list);
397 					continue;
398 				}
399 #endif
400 				nbusy++;
401 				if (show_busybufs > 0) {
402 					printf(
403 	    "%d: buf:%p, vnode:%p, flags:%0x, blkno:%jd, lblkno:%jd, buflock:",
404 					    nbusy, bp, bp->b_vp, bp->b_flags,
405 					    (intmax_t)bp->b_blkno,
406 					    (intmax_t)bp->b_lblkno);
407 					BUF_LOCKPRINTINFO(bp);
408 					if (show_busybufs > 1)
409 						vn_printf(bp->b_vp,
410 						    "vnode content: ");
411 				}
412 			}
413 		}
414 		if (nbusy) {
415 			/*
416 			 * Failed to sync all blocks. Indicate this and don't
417 			 * unmount filesystems (thus forcing an fsck on reboot).
418 			 */
419 			printf("Giving up on %d buffers\n", nbusy);
420 			DELAY(5000000);	/* 5 seconds */
421 		} else {
422 			if (!first_buf_printf)
423 				printf("Final sync complete\n");
424 			/*
425 			 * Unmount filesystems
426 			 */
427 			if (panicstr == 0)
428 				vfs_unmountall();
429 		}
430 		swapoff_all();
431 		DELAY(100000);		/* wait for console output to finish */
432 	}
433 
434 	print_uptime();
435 
436 	cngrab();
437 
438 	/*
439 	 * Ok, now do things that assume all filesystem activity has
440 	 * been completed.
441 	 */
442 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
443 
444 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
445 		doadump(TRUE);
446 
447 	/* Now that we're going to really halt the system... */
448 	EVENTHANDLER_INVOKE(shutdown_final, howto);
449 
450 	for(;;) ;	/* safety against shutdown_reset not working */
451 	/* NOTREACHED */
452 }
453 
454 /*
455  * If the shutdown was a clean halt, behave accordingly.
456  */
457 static void
458 shutdown_halt(void *junk, int howto)
459 {
460 
461 	if (howto & RB_HALT) {
462 		printf("\n");
463 		printf("The operating system has halted.\n");
464 		printf("Please press any key to reboot.\n\n");
465 		switch (cngetc()) {
466 		case -1:		/* No console, just die */
467 			cpu_halt();
468 			/* NOTREACHED */
469 		default:
470 			howto &= ~RB_HALT;
471 			break;
472 		}
473 	}
474 }
475 
476 /*
477  * Check to see if the system paniced, pause and then reboot
478  * according to the specified delay.
479  */
480 static void
481 shutdown_panic(void *junk, int howto)
482 {
483 	int loop;
484 
485 	if (howto & RB_DUMP) {
486 		if (panic_reboot_wait_time != 0) {
487 			if (panic_reboot_wait_time != -1) {
488 				printf("Automatic reboot in %d seconds - "
489 				       "press a key on the console to abort\n",
490 					panic_reboot_wait_time);
491 				for (loop = panic_reboot_wait_time * 10;
492 				     loop > 0; --loop) {
493 					DELAY(1000 * 100); /* 1/10th second */
494 					/* Did user type a key? */
495 					if (cncheckc() != -1)
496 						break;
497 				}
498 				if (!loop)
499 					return;
500 			}
501 		} else { /* zero time specified - reboot NOW */
502 			return;
503 		}
504 		printf("--> Press a key on the console to reboot,\n");
505 		printf("--> or switch off the system now.\n");
506 		cngetc();
507 	}
508 }
509 
510 /*
511  * Everything done, now reset
512  */
513 static void
514 shutdown_reset(void *junk, int howto)
515 {
516 
517 	printf("Rebooting...\n");
518 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
519 
520 	/*
521 	 * Acquiring smp_ipi_mtx here has a double effect:
522 	 * - it disables interrupts avoiding CPU0 preemption
523 	 *   by fast handlers (thus deadlocking  against other CPUs)
524 	 * - it avoids deadlocks against smp_rendezvous() or, more
525 	 *   generally, threads busy-waiting, with this spinlock held,
526 	 *   and waiting for responses by threads on other CPUs
527 	 *   (ie. smp_tlb_shootdown()).
528 	 *
529 	 * For the !SMP case it just needs to handle the former problem.
530 	 */
531 #ifdef SMP
532 	mtx_lock_spin(&smp_ipi_mtx);
533 #else
534 	spinlock_enter();
535 #endif
536 
537 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
538 	cpu_reset();
539 	/* NOTREACHED */ /* assuming reset worked */
540 }
541 
542 #if defined(WITNESS) || defined(INVARIANTS)
543 static int kassert_warn_only = 0;
544 #ifdef KDB
545 static int kassert_do_kdb = 0;
546 #endif
547 #ifdef KTR
548 static int kassert_do_ktr = 0;
549 #endif
550 static int kassert_do_log = 1;
551 static int kassert_log_pps_limit = 4;
552 static int kassert_log_mute_at = 0;
553 static int kassert_log_panic_at = 0;
554 static int kassert_warnings = 0;
555 
556 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW, NULL, "kassert options");
557 
558 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, CTLFLAG_RWTUN,
559     &kassert_warn_only, 0,
560     "KASSERT triggers a panic (1) or just a warning (0)");
561 
562 #ifdef KDB
563 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, CTLFLAG_RWTUN,
564     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
565 #endif
566 
567 #ifdef KTR
568 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, CTLFLAG_RWTUN,
569     &kassert_do_ktr, 0,
570     "KASSERT does a KTR, set this to the KTRMASK you want");
571 #endif
572 
573 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, CTLFLAG_RWTUN,
574     &kassert_do_log, 0, "KASSERT triggers a panic (1) or just a warning (0)");
575 
576 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RWTUN,
577     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
578 
579 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, CTLFLAG_RWTUN,
580     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
581 
582 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, CTLFLAG_RWTUN,
583     &kassert_log_pps_limit, 0, "limit number of log messages per second");
584 
585 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, CTLFLAG_RWTUN,
586     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
587 
588 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
589 
590 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
591     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE, NULL, 0,
592     kassert_sysctl_kassert, "I", "set to trigger a test kassert");
593 
594 static int
595 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
596 {
597 	int error, i;
598 
599 	error = sysctl_wire_old_buffer(req, sizeof(int));
600 	if (error == 0) {
601 		i = 0;
602 		error = sysctl_handle_int(oidp, &i, 0, req);
603 	}
604 	if (error != 0 || req->newptr == NULL)
605 		return (error);
606 	KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
607 	return (0);
608 }
609 
610 /*
611  * Called by KASSERT, this decides if we will panic
612  * or if we will log via printf and/or ktr.
613  */
614 void
615 kassert_panic(const char *fmt, ...)
616 {
617 	static char buf[256];
618 	va_list ap;
619 
620 	va_start(ap, fmt);
621 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
622 	va_end(ap);
623 
624 	/*
625 	 * panic if we're not just warning, or if we've exceeded
626 	 * kassert_log_panic_at warnings.
627 	 */
628 	if (!kassert_warn_only ||
629 	    (kassert_log_panic_at > 0 &&
630 	     kassert_warnings >= kassert_log_panic_at)) {
631 		va_start(ap, fmt);
632 		vpanic(fmt, ap);
633 		/* NORETURN */
634 	}
635 #ifdef KTR
636 	if (kassert_do_ktr)
637 		CTR0(ktr_mask, buf);
638 #endif /* KTR */
639 	/*
640 	 * log if we've not yet met the mute limit.
641 	 */
642 	if (kassert_do_log &&
643 	    (kassert_log_mute_at == 0 ||
644 	     kassert_warnings < kassert_log_mute_at)) {
645 		static  struct timeval lasterr;
646 		static  int curerr;
647 
648 		if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
649 			printf("KASSERT failed: %s\n", buf);
650 			kdb_backtrace();
651 		}
652 	}
653 #ifdef KDB
654 	if (kassert_do_kdb) {
655 		kdb_enter(KDB_WHY_KASSERT, buf);
656 	}
657 #endif
658 	atomic_add_int(&kassert_warnings, 1);
659 }
660 #endif
661 
662 /*
663  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
664  * and then reboots.  If we are called twice, then we avoid trying to sync
665  * the disks as this often leads to recursive panics.
666  */
667 void
668 panic(const char *fmt, ...)
669 {
670 	va_list ap;
671 
672 	va_start(ap, fmt);
673 	vpanic(fmt, ap);
674 }
675 
676 static void
677 vpanic(const char *fmt, va_list ap)
678 {
679 #ifdef SMP
680 	cpuset_t other_cpus;
681 #endif
682 	struct thread *td = curthread;
683 	int bootopt, newpanic;
684 	static char buf[256];
685 
686 	spinlock_enter();
687 
688 #ifdef SMP
689 	/*
690 	 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
691 	 * concurrently entering panic.  Only the winner will proceed
692 	 * further.
693 	 */
694 	if (panicstr == NULL && !kdb_active) {
695 		other_cpus = all_cpus;
696 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
697 		stop_cpus_hard(other_cpus);
698 	}
699 
700 	/*
701 	 * We set stop_scheduler here and not in the block above,
702 	 * because we want to ensure that if panic has been called and
703 	 * stop_scheduler_on_panic is true, then stop_scheduler will
704 	 * always be set.  Even if panic has been entered from kdb.
705 	 */
706 	td->td_stopsched = 1;
707 #endif
708 
709 	bootopt = RB_AUTOBOOT;
710 	newpanic = 0;
711 	if (panicstr)
712 		bootopt |= RB_NOSYNC;
713 	else {
714 		bootopt |= RB_DUMP;
715 		panicstr = fmt;
716 		newpanic = 1;
717 	}
718 
719 	if (newpanic) {
720 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
721 		panicstr = buf;
722 		cngrab();
723 		printf("panic: %s\n", buf);
724 	} else {
725 		printf("panic: ");
726 		vprintf(fmt, ap);
727 		printf("\n");
728 	}
729 #ifdef SMP
730 	printf("cpuid = %d\n", PCPU_GET(cpuid));
731 #endif
732 
733 #ifdef KDB
734 	if (newpanic && trace_on_panic)
735 		kdb_backtrace();
736 	if (debugger_on_panic)
737 		kdb_enter(KDB_WHY_PANIC, "panic");
738 #endif
739 	/*thread_lock(td); */
740 	td->td_flags |= TDF_INPANIC;
741 	/* thread_unlock(td); */
742 	if (!sync_on_panic)
743 		bootopt |= RB_NOSYNC;
744 	kern_reboot(bootopt);
745 }
746 
747 /*
748  * Support for poweroff delay.
749  *
750  * Please note that setting this delay too short might power off your machine
751  * before the write cache on your hard disk has been flushed, leading to
752  * soft-updates inconsistencies.
753  */
754 #ifndef POWEROFF_DELAY
755 # define POWEROFF_DELAY 5000
756 #endif
757 static int poweroff_delay = POWEROFF_DELAY;
758 
759 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
760     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
761 
762 static void
763 poweroff_wait(void *junk, int howto)
764 {
765 
766 	if (!(howto & RB_POWEROFF) || poweroff_delay <= 0)
767 		return;
768 	DELAY(poweroff_delay * 1000);
769 }
770 
771 /*
772  * Some system processes (e.g. syncer) need to be stopped at appropriate
773  * points in their main loops prior to a system shutdown, so that they
774  * won't interfere with the shutdown process (e.g. by holding a disk buf
775  * to cause sync to fail).  For each of these system processes, register
776  * shutdown_kproc() as a handler for one of shutdown events.
777  */
778 static int kproc_shutdown_wait = 60;
779 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
780     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
781 
782 void
783 kproc_shutdown(void *arg, int howto)
784 {
785 	struct proc *p;
786 	int error;
787 
788 	if (panicstr)
789 		return;
790 
791 	p = (struct proc *)arg;
792 	printf("Waiting (max %d seconds) for system process `%s' to stop...",
793 	    kproc_shutdown_wait, p->p_comm);
794 	error = kproc_suspend(p, kproc_shutdown_wait * hz);
795 
796 	if (error == EWOULDBLOCK)
797 		printf("timed out\n");
798 	else
799 		printf("done\n");
800 }
801 
802 void
803 kthread_shutdown(void *arg, int howto)
804 {
805 	struct thread *td;
806 	int error;
807 
808 	if (panicstr)
809 		return;
810 
811 	td = (struct thread *)arg;
812 	printf("Waiting (max %d seconds) for system thread `%s' to stop...",
813 	    kproc_shutdown_wait, td->td_name);
814 	error = kthread_suspend(td, kproc_shutdown_wait * hz);
815 
816 	if (error == EWOULDBLOCK)
817 		printf("timed out\n");
818 	else
819 		printf("done\n");
820 }
821 
822 static char dumpdevname[sizeof(((struct cdev*)NULL)->si_name)];
823 SYSCTL_STRING(_kern_shutdown, OID_AUTO, dumpdevname, CTLFLAG_RD,
824     dumpdevname, 0, "Device for kernel dumps");
825 
826 /* Registration of dumpers */
827 int
828 set_dumper(struct dumperinfo *di, const char *devname)
829 {
830 	size_t wantcopy;
831 
832 	if (di == NULL) {
833 		bzero(&dumper, sizeof dumper);
834 		dumpdevname[0] = '\0';
835 		return (0);
836 	}
837 	if (dumper.dumper != NULL)
838 		return (EBUSY);
839 	dumper = *di;
840 	wantcopy = strlcpy(dumpdevname, devname, sizeof(dumpdevname));
841 	if (wantcopy >= sizeof(dumpdevname)) {
842 		printf("set_dumper: device name truncated from '%s' -> '%s'\n",
843 			devname, dumpdevname);
844 	}
845 	return (0);
846 }
847 
848 /* Call dumper with bounds checking. */
849 int
850 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
851     off_t offset, size_t length)
852 {
853 
854 	if (length != 0 && (offset < di->mediaoffset ||
855 	    offset - di->mediaoffset + length > di->mediasize)) {
856 		printf("Attempt to write outside dump device boundaries.\n"
857 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
858 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
859 		    (uintmax_t)length, (intmax_t)di->mediasize);
860 		return (ENOSPC);
861 	}
862 	return (di->dumper(di->priv, virtual, physical, offset, length));
863 }
864 
865 void
866 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver,
867     uint64_t dumplen, uint32_t blksz)
868 {
869 
870 	bzero(kdh, sizeof(*kdh));
871 	strncpy(kdh->magic, magic, sizeof(kdh->magic));
872 	strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
873 	kdh->version = htod32(KERNELDUMPVERSION);
874 	kdh->architectureversion = htod32(archver);
875 	kdh->dumplength = htod64(dumplen);
876 	kdh->dumptime = htod64(time_second);
877 	kdh->blocksize = htod32(blksz);
878 	strncpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
879 	strncpy(kdh->versionstring, version, sizeof(kdh->versionstring));
880 	if (panicstr != NULL)
881 		strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
882 	kdh->parity = kerneldump_parity(kdh);
883 }
884