xref: /freebsd/sys/kern/kern_shutdown.c (revision 7431dfd4580e850375fe5478d92ec770344db098)
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 	int error;
253 
254 	error = 0;
255 	if (dumping)
256 		return (EBUSY);
257 	if (dumper.dumper == NULL)
258 		return (ENXIO);
259 
260 	savectx(&dumppcb);
261 	dumptid = curthread->td_tid;
262 	dumping++;
263 
264 	coredump = TRUE;
265 #ifdef DDB
266 	if (textdump && textdump_pending) {
267 		coredump = FALSE;
268 		textdump_dumpsys(&dumper);
269 	}
270 #endif
271 	if (coredump)
272 		error = dumpsys(&dumper);
273 
274 	dumping--;
275 	return (error);
276 }
277 
278 static int
279 isbufbusy(struct buf *bp)
280 {
281 	if (((bp->b_flags & (B_INVAL | B_PERSISTENT)) == 0 &&
282 	    BUF_ISLOCKED(bp)) ||
283 	    ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI))
284 		return (1);
285 	return (0);
286 }
287 
288 /*
289  * Shutdown the system cleanly to prepare for reboot, halt, or power off.
290  */
291 void
292 kern_reboot(int howto)
293 {
294 	static int first_buf_printf = 1;
295 	static int waittime = -1;
296 
297 #if defined(SMP)
298 	/*
299 	 * Bind us to CPU 0 so that all shutdown code runs there.  Some
300 	 * systems don't shutdown properly (i.e., ACPI power off) if we
301 	 * run on another processor.
302 	 */
303 	if (!SCHEDULER_STOPPED()) {
304 		thread_lock(curthread);
305 		sched_bind(curthread, 0);
306 		thread_unlock(curthread);
307 		KASSERT(PCPU_GET(cpuid) == 0, ("boot: not running on cpu 0"));
308 	}
309 #endif
310 	/* We're in the process of rebooting. */
311 	rebooting = 1;
312 
313 	/* We are out of the debugger now. */
314 	kdb_active = 0;
315 
316 	/*
317 	 * Do any callouts that should be done BEFORE syncing the filesystems.
318 	 */
319 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
320 
321 	/*
322 	 * Now sync filesystems
323 	 */
324 	if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
325 		register struct buf *bp;
326 		int iter, nbusy, pbusy;
327 #ifndef PREEMPTION
328 		int subiter;
329 #endif
330 
331 		waittime = 0;
332 
333 		wdog_kern_pat(WD_LASTVAL);
334 		sys_sync(curthread, NULL);
335 
336 		/*
337 		 * With soft updates, some buffers that are
338 		 * written will be remarked as dirty until other
339 		 * buffers are written.
340 		 */
341 		for (iter = pbusy = 0; iter < 20; iter++) {
342 			nbusy = 0;
343 			for (bp = &buf[nbuf]; --bp >= buf; )
344 				if (isbufbusy(bp))
345 					nbusy++;
346 			if (nbusy == 0) {
347 				if (first_buf_printf)
348 					printf("All buffers synced.");
349 				break;
350 			}
351 			if (first_buf_printf) {
352 				printf("Syncing disks, buffers remaining... ");
353 				first_buf_printf = 0;
354 			}
355 			printf("%d ", nbusy);
356 			if (nbusy < pbusy)
357 				iter = 0;
358 			pbusy = nbusy;
359 
360 			wdog_kern_pat(WD_LASTVAL);
361 			sys_sync(curthread, NULL);
362 
363 #ifdef PREEMPTION
364 			/*
365 			 * Drop Giant and spin for a while to allow
366 			 * interrupt threads to run.
367 			 */
368 			DROP_GIANT();
369 			DELAY(50000 * iter);
370 			PICKUP_GIANT();
371 #else
372 			/*
373 			 * Drop Giant and context switch several times to
374 			 * allow interrupt threads to run.
375 			 */
376 			DROP_GIANT();
377 			for (subiter = 0; subiter < 50 * iter; subiter++) {
378 				thread_lock(curthread);
379 				mi_switch(SW_VOL, NULL);
380 				thread_unlock(curthread);
381 				DELAY(1000);
382 			}
383 			PICKUP_GIANT();
384 #endif
385 		}
386 		printf("\n");
387 		/*
388 		 * Count only busy local buffers to prevent forcing
389 		 * a fsck if we're just a client of a wedged NFS server
390 		 */
391 		nbusy = 0;
392 		for (bp = &buf[nbuf]; --bp >= buf; ) {
393 			if (isbufbusy(bp)) {
394 #if 0
395 /* XXX: This is bogus.  We should probably have a BO_REMOTE flag instead */
396 				if (bp->b_dev == NULL) {
397 					TAILQ_REMOVE(&mountlist,
398 					    bp->b_vp->v_mount, mnt_list);
399 					continue;
400 				}
401 #endif
402 				nbusy++;
403 				if (show_busybufs > 0) {
404 					printf(
405 	    "%d: buf:%p, vnode:%p, flags:%0x, blkno:%jd, lblkno:%jd, buflock:",
406 					    nbusy, bp, bp->b_vp, bp->b_flags,
407 					    (intmax_t)bp->b_blkno,
408 					    (intmax_t)bp->b_lblkno);
409 					BUF_LOCKPRINTINFO(bp);
410 					if (show_busybufs > 1)
411 						vn_printf(bp->b_vp,
412 						    "vnode content: ");
413 				}
414 			}
415 		}
416 		if (nbusy) {
417 			/*
418 			 * Failed to sync all blocks. Indicate this and don't
419 			 * unmount filesystems (thus forcing an fsck on reboot).
420 			 */
421 			printf("Giving up on %d buffers\n", nbusy);
422 			DELAY(5000000);	/* 5 seconds */
423 		} else {
424 			if (!first_buf_printf)
425 				printf("Final sync complete\n");
426 			/*
427 			 * Unmount filesystems
428 			 */
429 			if (panicstr == 0)
430 				vfs_unmountall();
431 		}
432 		swapoff_all();
433 		DELAY(100000);		/* wait for console output to finish */
434 	}
435 
436 	print_uptime();
437 
438 	cngrab();
439 
440 	/*
441 	 * Ok, now do things that assume all filesystem activity has
442 	 * been completed.
443 	 */
444 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
445 
446 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
447 		doadump(TRUE);
448 
449 	/* Now that we're going to really halt the system... */
450 	EVENTHANDLER_INVOKE(shutdown_final, howto);
451 
452 	for(;;) ;	/* safety against shutdown_reset not working */
453 	/* NOTREACHED */
454 }
455 
456 /*
457  * If the shutdown was a clean halt, behave accordingly.
458  */
459 static void
460 shutdown_halt(void *junk, int howto)
461 {
462 
463 	if (howto & RB_HALT) {
464 		printf("\n");
465 		printf("The operating system has halted.\n");
466 		printf("Please press any key to reboot.\n\n");
467 		switch (cngetc()) {
468 		case -1:		/* No console, just die */
469 			cpu_halt();
470 			/* NOTREACHED */
471 		default:
472 			howto &= ~RB_HALT;
473 			break;
474 		}
475 	}
476 }
477 
478 /*
479  * Check to see if the system paniced, pause and then reboot
480  * according to the specified delay.
481  */
482 static void
483 shutdown_panic(void *junk, int howto)
484 {
485 	int loop;
486 
487 	if (howto & RB_DUMP) {
488 		if (panic_reboot_wait_time != 0) {
489 			if (panic_reboot_wait_time != -1) {
490 				printf("Automatic reboot in %d seconds - "
491 				       "press a key on the console to abort\n",
492 					panic_reboot_wait_time);
493 				for (loop = panic_reboot_wait_time * 10;
494 				     loop > 0; --loop) {
495 					DELAY(1000 * 100); /* 1/10th second */
496 					/* Did user type a key? */
497 					if (cncheckc() != -1)
498 						break;
499 				}
500 				if (!loop)
501 					return;
502 			}
503 		} else { /* zero time specified - reboot NOW */
504 			return;
505 		}
506 		printf("--> Press a key on the console to reboot,\n");
507 		printf("--> or switch off the system now.\n");
508 		cngetc();
509 	}
510 }
511 
512 /*
513  * Everything done, now reset
514  */
515 static void
516 shutdown_reset(void *junk, int howto)
517 {
518 
519 	printf("Rebooting...\n");
520 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
521 
522 	/*
523 	 * Acquiring smp_ipi_mtx here has a double effect:
524 	 * - it disables interrupts avoiding CPU0 preemption
525 	 *   by fast handlers (thus deadlocking  against other CPUs)
526 	 * - it avoids deadlocks against smp_rendezvous() or, more
527 	 *   generally, threads busy-waiting, with this spinlock held,
528 	 *   and waiting for responses by threads on other CPUs
529 	 *   (ie. smp_tlb_shootdown()).
530 	 *
531 	 * For the !SMP case it just needs to handle the former problem.
532 	 */
533 #ifdef SMP
534 	mtx_lock_spin(&smp_ipi_mtx);
535 #else
536 	spinlock_enter();
537 #endif
538 
539 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
540 	cpu_reset();
541 	/* NOTREACHED */ /* assuming reset worked */
542 }
543 
544 #if defined(WITNESS) || defined(INVARIANTS)
545 static int kassert_warn_only = 0;
546 #ifdef KDB
547 static int kassert_do_kdb = 0;
548 #endif
549 #ifdef KTR
550 static int kassert_do_ktr = 0;
551 #endif
552 static int kassert_do_log = 1;
553 static int kassert_log_pps_limit = 4;
554 static int kassert_log_mute_at = 0;
555 static int kassert_log_panic_at = 0;
556 static int kassert_warnings = 0;
557 
558 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW, NULL, "kassert options");
559 
560 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, CTLFLAG_RWTUN,
561     &kassert_warn_only, 0,
562     "KASSERT triggers a panic (1) or just a warning (0)");
563 
564 #ifdef KDB
565 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, CTLFLAG_RWTUN,
566     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
567 #endif
568 
569 #ifdef KTR
570 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, CTLFLAG_RWTUN,
571     &kassert_do_ktr, 0,
572     "KASSERT does a KTR, set this to the KTRMASK you want");
573 #endif
574 
575 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, CTLFLAG_RWTUN,
576     &kassert_do_log, 0, "KASSERT triggers a panic (1) or just a warning (0)");
577 
578 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RWTUN,
579     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
580 
581 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, CTLFLAG_RWTUN,
582     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
583 
584 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, CTLFLAG_RWTUN,
585     &kassert_log_pps_limit, 0, "limit number of log messages per second");
586 
587 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, CTLFLAG_RWTUN,
588     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
589 
590 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
591 
592 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
593     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE, NULL, 0,
594     kassert_sysctl_kassert, "I", "set to trigger a test kassert");
595 
596 static int
597 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
598 {
599 	int error, i;
600 
601 	error = sysctl_wire_old_buffer(req, sizeof(int));
602 	if (error == 0) {
603 		i = 0;
604 		error = sysctl_handle_int(oidp, &i, 0, req);
605 	}
606 	if (error != 0 || req->newptr == NULL)
607 		return (error);
608 	KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
609 	return (0);
610 }
611 
612 /*
613  * Called by KASSERT, this decides if we will panic
614  * or if we will log via printf and/or ktr.
615  */
616 void
617 kassert_panic(const char *fmt, ...)
618 {
619 	static char buf[256];
620 	va_list ap;
621 
622 	va_start(ap, fmt);
623 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
624 	va_end(ap);
625 
626 	/*
627 	 * panic if we're not just warning, or if we've exceeded
628 	 * kassert_log_panic_at warnings.
629 	 */
630 	if (!kassert_warn_only ||
631 	    (kassert_log_panic_at > 0 &&
632 	     kassert_warnings >= kassert_log_panic_at)) {
633 		va_start(ap, fmt);
634 		vpanic(fmt, ap);
635 		/* NORETURN */
636 	}
637 #ifdef KTR
638 	if (kassert_do_ktr)
639 		CTR0(ktr_mask, buf);
640 #endif /* KTR */
641 	/*
642 	 * log if we've not yet met the mute limit.
643 	 */
644 	if (kassert_do_log &&
645 	    (kassert_log_mute_at == 0 ||
646 	     kassert_warnings < kassert_log_mute_at)) {
647 		static  struct timeval lasterr;
648 		static  int curerr;
649 
650 		if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
651 			printf("KASSERT failed: %s\n", buf);
652 			kdb_backtrace();
653 		}
654 	}
655 #ifdef KDB
656 	if (kassert_do_kdb) {
657 		kdb_enter(KDB_WHY_KASSERT, buf);
658 	}
659 #endif
660 	atomic_add_int(&kassert_warnings, 1);
661 }
662 #endif
663 
664 /*
665  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
666  * and then reboots.  If we are called twice, then we avoid trying to sync
667  * the disks as this often leads to recursive panics.
668  */
669 void
670 panic(const char *fmt, ...)
671 {
672 	va_list ap;
673 
674 	va_start(ap, fmt);
675 	vpanic(fmt, ap);
676 }
677 
678 static void
679 vpanic(const char *fmt, va_list ap)
680 {
681 #ifdef SMP
682 	cpuset_t other_cpus;
683 #endif
684 	struct thread *td = curthread;
685 	int bootopt, newpanic;
686 	static char buf[256];
687 
688 	spinlock_enter();
689 
690 #ifdef SMP
691 	/*
692 	 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
693 	 * concurrently entering panic.  Only the winner will proceed
694 	 * further.
695 	 */
696 	if (panicstr == NULL && !kdb_active) {
697 		other_cpus = all_cpus;
698 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
699 		stop_cpus_hard(other_cpus);
700 	}
701 
702 	/*
703 	 * We set stop_scheduler here and not in the block above,
704 	 * because we want to ensure that if panic has been called and
705 	 * stop_scheduler_on_panic is true, then stop_scheduler will
706 	 * always be set.  Even if panic has been entered from kdb.
707 	 */
708 	td->td_stopsched = 1;
709 #endif
710 
711 	bootopt = RB_AUTOBOOT;
712 	newpanic = 0;
713 	if (panicstr)
714 		bootopt |= RB_NOSYNC;
715 	else {
716 		bootopt |= RB_DUMP;
717 		panicstr = fmt;
718 		newpanic = 1;
719 	}
720 
721 	if (newpanic) {
722 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
723 		panicstr = buf;
724 		cngrab();
725 		printf("panic: %s\n", buf);
726 	} else {
727 		printf("panic: ");
728 		vprintf(fmt, ap);
729 		printf("\n");
730 	}
731 #ifdef SMP
732 	printf("cpuid = %d\n", PCPU_GET(cpuid));
733 #endif
734 
735 #ifdef KDB
736 	if (newpanic && trace_on_panic)
737 		kdb_backtrace();
738 	if (debugger_on_panic)
739 		kdb_enter(KDB_WHY_PANIC, "panic");
740 #endif
741 	/*thread_lock(td); */
742 	td->td_flags |= TDF_INPANIC;
743 	/* thread_unlock(td); */
744 	if (!sync_on_panic)
745 		bootopt |= RB_NOSYNC;
746 	kern_reboot(bootopt);
747 }
748 
749 /*
750  * Support for poweroff delay.
751  *
752  * Please note that setting this delay too short might power off your machine
753  * before the write cache on your hard disk has been flushed, leading to
754  * soft-updates inconsistencies.
755  */
756 #ifndef POWEROFF_DELAY
757 # define POWEROFF_DELAY 5000
758 #endif
759 static int poweroff_delay = POWEROFF_DELAY;
760 
761 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
762     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
763 
764 static void
765 poweroff_wait(void *junk, int howto)
766 {
767 
768 	if (!(howto & RB_POWEROFF) || poweroff_delay <= 0)
769 		return;
770 	DELAY(poweroff_delay * 1000);
771 }
772 
773 /*
774  * Some system processes (e.g. syncer) need to be stopped at appropriate
775  * points in their main loops prior to a system shutdown, so that they
776  * won't interfere with the shutdown process (e.g. by holding a disk buf
777  * to cause sync to fail).  For each of these system processes, register
778  * shutdown_kproc() as a handler for one of shutdown events.
779  */
780 static int kproc_shutdown_wait = 60;
781 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
782     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
783 
784 void
785 kproc_shutdown(void *arg, int howto)
786 {
787 	struct proc *p;
788 	int error;
789 
790 	if (panicstr)
791 		return;
792 
793 	p = (struct proc *)arg;
794 	printf("Waiting (max %d seconds) for system process `%s' to stop...",
795 	    kproc_shutdown_wait, p->p_comm);
796 	error = kproc_suspend(p, kproc_shutdown_wait * hz);
797 
798 	if (error == EWOULDBLOCK)
799 		printf("timed out\n");
800 	else
801 		printf("done\n");
802 }
803 
804 void
805 kthread_shutdown(void *arg, int howto)
806 {
807 	struct thread *td;
808 	int error;
809 
810 	if (panicstr)
811 		return;
812 
813 	td = (struct thread *)arg;
814 	printf("Waiting (max %d seconds) for system thread `%s' to stop...",
815 	    kproc_shutdown_wait, td->td_name);
816 	error = kthread_suspend(td, kproc_shutdown_wait * hz);
817 
818 	if (error == EWOULDBLOCK)
819 		printf("timed out\n");
820 	else
821 		printf("done\n");
822 }
823 
824 static char dumpdevname[sizeof(((struct cdev*)NULL)->si_name)];
825 SYSCTL_STRING(_kern_shutdown, OID_AUTO, dumpdevname, CTLFLAG_RD,
826     dumpdevname, 0, "Device for kernel dumps");
827 
828 /* Registration of dumpers */
829 int
830 set_dumper(struct dumperinfo *di, const char *devname)
831 {
832 	size_t wantcopy;
833 
834 	if (di == NULL) {
835 		bzero(&dumper, sizeof dumper);
836 		dumpdevname[0] = '\0';
837 		return (0);
838 	}
839 	if (dumper.dumper != NULL)
840 		return (EBUSY);
841 	dumper = *di;
842 	wantcopy = strlcpy(dumpdevname, devname, sizeof(dumpdevname));
843 	if (wantcopy >= sizeof(dumpdevname)) {
844 		printf("set_dumper: device name truncated from '%s' -> '%s'\n",
845 			devname, dumpdevname);
846 	}
847 	return (0);
848 }
849 
850 /* Call dumper with bounds checking. */
851 int
852 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
853     off_t offset, size_t length)
854 {
855 
856 	if (length != 0 && (offset < di->mediaoffset ||
857 	    offset - di->mediaoffset + length > di->mediasize)) {
858 		printf("Attempt to write outside dump device boundaries.\n"
859 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
860 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
861 		    (uintmax_t)length, (intmax_t)di->mediasize);
862 		return (ENOSPC);
863 	}
864 	return (di->dumper(di->priv, virtual, physical, offset, length));
865 }
866 
867 void
868 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver,
869     uint64_t dumplen, uint32_t blksz)
870 {
871 
872 	bzero(kdh, sizeof(*kdh));
873 	strncpy(kdh->magic, magic, sizeof(kdh->magic));
874 	strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
875 	kdh->version = htod32(KERNELDUMPVERSION);
876 	kdh->architectureversion = htod32(archver);
877 	kdh->dumplength = htod64(dumplen);
878 	kdh->dumptime = htod64(time_second);
879 	kdh->blocksize = htod32(blksz);
880 	strncpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
881 	strncpy(kdh->versionstring, version, sizeof(kdh->versionstring));
882 	if (panicstr != NULL)
883 		strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
884 	kdh->parity = kerneldump_parity(kdh);
885 }
886