xref: /dflybsd-src/sys/kern/kern_shutdown.c (revision 32af04f740541c3e943edbdb70f13dff978147be)
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 1.62 2008/01/05 13:23:48 corecode 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/disk.h>
54 #include <sys/diskslice.h>
55 #include <sys/reboot.h>
56 #include <sys/proc.h>
57 #include <sys/priv.h>
58 #include <sys/fcntl.h>		/* FREAD	*/
59 #include <sys/stat.h>		/* S_IFCHR	*/
60 #include <sys/vnode.h>
61 #include <sys/kernel.h>
62 #include <sys/kerneldump.h>
63 #include <sys/kthread.h>
64 #include <sys/malloc.h>
65 #include <sys/mount.h>
66 #include <sys/queue.h>
67 #include <sys/sysctl.h>
68 #include <sys/vkernel.h>
69 #include <sys/conf.h>
70 #include <sys/sysproto.h>
71 #include <sys/device.h>
72 #include <sys/cons.h>
73 #include <sys/shm.h>
74 #include <sys/kerneldump.h>
75 #include <sys/kern_syscall.h>
76 #include <vm/vm_map.h>
77 #include <vm/pmap.h>
78 
79 #include <sys/thread2.h>
80 #include <sys/buf2.h>
81 
82 #include <machine/clock.h>
83 #include <machine/md_var.h>
84 #include <machine/smp.h>		/* smp_active_mask, cpuid */
85 #include <machine/vmparam.h>
86 #include <machine/thread.h>
87 
88 #include <sys/signalvar.h>
89 
90 #include <sys/wdog.h>
91 #include <dev/misc/gpio/gpio.h>
92 
93 #ifndef PANIC_REBOOT_WAIT_TIME
94 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
95 #endif
96 
97 /*
98  * Note that stdarg.h and the ANSI style va_start macro is used for both
99  * ANSI and traditional C compilers.  We use the machine version to stay
100  * within the confines of the kernel header files.
101  */
102 #include <machine/stdarg.h>
103 
104 #ifdef DDB
105 #include <ddb/ddb.h>
106 #ifdef DDB_UNATTENDED
107 int debugger_on_panic = 0;
108 #else
109 int debugger_on_panic = 1;
110 #endif
111 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW,
112 	&debugger_on_panic, 0, "Run debugger on kernel panic");
113 
114 #ifdef DDB_TRACE
115 int trace_on_panic = 1;
116 #else
117 int trace_on_panic = 0;
118 #endif
119 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW,
120 	&trace_on_panic, 0, "Print stack trace on kernel panic");
121 #endif
122 
123 static int sync_on_panic = 0;
124 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW,
125 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
126 
127 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment");
128 
129 #ifdef	HW_WDOG
130 /*
131  * If there is a hardware watchdog, point this at the function needed to
132  * hold it off.
133  * It's needed when the kernel needs to do some lengthy operations.
134  * e.g. in wd.c when dumping core.. It's most annoying to have
135  * your precious core-dump only half written because the wdog kicked in.
136  */
137 watchdog_tickle_fn wdog_tickler = NULL;
138 #endif	/* HW_WDOG */
139 
140 /*
141  * Variable panicstr contains argument to first call to panic; used as flag
142  * to indicate that the kernel has already called panic.
143  */
144 const char *panicstr;
145 
146 int dumping;				/* system is dumping */
147 static struct dumperinfo dumper;	/* selected dumper */
148 
149 #ifdef SMP
150 u_int panic_cpu_interlock;		/* panic interlock */
151 globaldata_t panic_cpu_gd;		/* which cpu took the panic */
152 #endif
153 
154 int bootverbose = 0;			/* note: assignment to force non-bss */
155 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW,
156 	   &bootverbose, 0, "Verbose kernel messages");
157 
158 int cold = 1;				/* note: assignment to force non-bss */
159 int dumplo;				/* OBSOLETE - savecore compat */
160 u_int64_t dumplo64;
161 
162 static void boot (int) __dead2;
163 static int setdumpdev (cdev_t dev);
164 static void poweroff_wait (void *, int);
165 static void print_uptime (void);
166 static void shutdown_halt (void *junk, int howto);
167 static void shutdown_panic (void *junk, int howto);
168 static void shutdown_reset (void *junk, int howto);
169 static int shutdown_busycount1(struct buf *bp, void *info);
170 static int shutdown_busycount2(struct buf *bp, void *info);
171 static void shutdown_cleanup_proc(struct proc *p);
172 
173 /* register various local shutdown events */
174 static void
175 shutdown_conf(void *unused)
176 {
177 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, SHUTDOWN_PRI_FIRST);
178 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, SHUTDOWN_PRI_LAST + 100);
179 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, SHUTDOWN_PRI_LAST + 100);
180 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL, SHUTDOWN_PRI_LAST + 200);
181 }
182 
183 SYSINIT(shutdown_conf, SI_BOOT2_MACHDEP, SI_ORDER_ANY, shutdown_conf, NULL)
184 
185 /* ARGSUSED */
186 
187 /*
188  * The system call that results in a reboot
189  */
190 int
191 sys_reboot(struct reboot_args *uap)
192 {
193 	struct thread *td = curthread;
194 	int error;
195 
196 	if ((error = priv_check(td, PRIV_REBOOT)))
197 		return (error);
198 
199 	boot(uap->opt);
200 	return (0);
201 }
202 
203 /*
204  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
205  */
206 static int shutdown_howto = 0;
207 
208 void
209 shutdown_nice(int howto)
210 {
211 	shutdown_howto = howto;
212 
213 	/* Send a signal to init(8) and have it shutdown the world */
214 	if (initproc != NULL) {
215 		ksignal(initproc, SIGINT);
216 	} else {
217 		/* No init(8) running, so simply reboot */
218 		boot(RB_NOSYNC);
219 	}
220 	return;
221 }
222 static int	waittime = -1;
223 struct pcb dumppcb;
224 struct thread *dumpthread;
225 
226 static void
227 print_uptime(void)
228 {
229 	int f;
230 	struct timespec ts;
231 
232 	getnanouptime(&ts);
233 	kprintf("Uptime: ");
234 	f = 0;
235 	if (ts.tv_sec >= 86400) {
236 		kprintf("%ldd", ts.tv_sec / 86400);
237 		ts.tv_sec %= 86400;
238 		f = 1;
239 	}
240 	if (f || ts.tv_sec >= 3600) {
241 		kprintf("%ldh", ts.tv_sec / 3600);
242 		ts.tv_sec %= 3600;
243 		f = 1;
244 	}
245 	if (f || ts.tv_sec >= 60) {
246 		kprintf("%ldm", ts.tv_sec / 60);
247 		ts.tv_sec %= 60;
248 		f = 1;
249 	}
250 	kprintf("%lds\n", ts.tv_sec);
251 }
252 
253 /*
254  *  Go through the rigmarole of shutting down..
255  * this used to be in machdep.c but I'll be dammned if I could see
256  * anything machine dependant in it.
257  */
258 static void
259 boot(int howto)
260 {
261 	/*
262 	 * Get rid of any user scheduler baggage and then give
263 	 * us a high priority.
264 	 */
265 	if (curthread->td_release)
266 		curthread->td_release(curthread);
267 	lwkt_setpri_self(TDPRI_MAX);
268 
269 	/* collect extra flags that shutdown_nice might have set */
270 	howto |= shutdown_howto;
271 
272 #ifdef SMP
273 	/*
274 	 * We really want to shutdown on the BSP.  Subsystems such as ACPI
275 	 * can't power-down the box otherwise.
276 	 */
277 	if (smp_active_mask > 1) {
278 		kprintf("boot() called on cpu#%d\n", mycpu->gd_cpuid);
279 	}
280 	if (panicstr == NULL && mycpu->gd_cpuid != 0) {
281 		kprintf("Switching to cpu #0 for shutdown\n");
282 		lwkt_setcpu_self(globaldata_find(0));
283 	}
284 #endif
285 	/*
286 	 * Do any callouts that should be done BEFORE syncing the filesystems.
287 	 */
288 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
289 
290 	/*
291 	 * Try to get rid of any remaining FS references.  The calling
292 	 * process, proc0, and init may still hold references.  The
293 	 * VFS cache subsystem may still hold a root reference to root.
294 	 *
295 	 * XXX this needs work.  We really need to SIGSTOP all remaining
296 	 * processes in order to avoid blowups due to proc0's filesystem
297 	 * references going away.  For now just make sure that the init
298 	 * process is stopped.
299 	 */
300 	if (panicstr == NULL) {
301 		shutdown_cleanup_proc(curproc);
302 		shutdown_cleanup_proc(&proc0);
303 		if (initproc) {
304 			if (initproc != curproc) {
305 				ksignal(initproc, SIGSTOP);
306 				tsleep(boot, 0, "shutdn", hz / 20);
307 			}
308 			shutdown_cleanup_proc(initproc);
309 		}
310 		vfs_cache_setroot(NULL, NULL);
311 	}
312 
313 	/*
314 	 * Now sync filesystems
315 	 */
316 	if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
317 		int iter, nbusy, pbusy;
318 
319 		waittime = 0;
320 		kprintf("\nsyncing disks... ");
321 
322 		sys_sync(NULL);	/* YYY was sync(&proc0, NULL). why proc0 ? */
323 
324 		/*
325 		 * With soft updates, some buffers that are
326 		 * written will be remarked as dirty until other
327 		 * buffers are written.
328 		 */
329 		for (iter = pbusy = 0; iter < 20; iter++) {
330 			nbusy = scan_all_buffers(shutdown_busycount1, NULL);
331 			if (nbusy == 0)
332 				break;
333 			kprintf("%d ", nbusy);
334 			if (nbusy < pbusy)
335 				iter = 0;
336 			pbusy = nbusy;
337 			/*
338 			 * XXX:
339 			 * Process soft update work queue if buffers don't sync
340 			 * after 6 iterations by permitting the syncer to run.
341 			 */
342 			if (iter > 5)
343 				bio_ops_sync(NULL);
344 
345 			sys_sync(NULL); /* YYY was sync(&proc0, NULL). why proc0 ? */
346 			tsleep(boot, 0, "shutdn", hz * iter / 20 + 1);
347 		}
348 		kprintf("\n");
349 		/*
350 		 * Count only busy local buffers to prevent forcing
351 		 * a fsck if we're just a client of a wedged NFS server
352 		 */
353 		nbusy = scan_all_buffers(shutdown_busycount2, NULL);
354 		if (nbusy) {
355 			/*
356 			 * Failed to sync all blocks. Indicate this and don't
357 			 * unmount filesystems (thus forcing an fsck on reboot).
358 			 */
359 			kprintf("giving up on %d buffers\n", nbusy);
360 #ifdef DDB
361 			Debugger("busy buffer problem");
362 #endif /* DDB */
363 			tsleep(boot, 0, "shutdn", hz * 5 + 1);
364 		} else {
365 			kprintf("done\n");
366 			/*
367 			 * Unmount filesystems
368 			 */
369 			if (panicstr == NULL)
370 				vfs_unmountall();
371 		}
372 		tsleep(boot, 0, "shutdn", hz / 10 + 1);
373 	}
374 
375 	print_uptime();
376 
377 	/*
378 	 * Dump before doing post_sync shutdown ops
379 	 */
380 	crit_enter();
381 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold &&
382 	    dumper.dumper != NULL && !dumping) {
383 		dumping++;
384 		dumpsys(&dumper);
385 	}
386 
387 	/*
388 	 * Ok, now do things that assume all filesystem activity has
389 	 * been completed.  This will also call the device shutdown
390 	 * methods.
391 	 */
392 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
393 
394 	/* Now that we're going to really halt the system... */
395 	EVENTHANDLER_INVOKE(shutdown_final, howto);
396 
397 	for(;;) ;	/* safety against shutdown_reset not working */
398 	/* NOTREACHED */
399 }
400 
401 static int
402 shutdown_busycount1(struct buf *bp, void *info)
403 {
404 	if ((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp) > 0)
405 		return(1);
406 	if ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI)
407 		return (1);
408 	return (0);
409 }
410 
411 static int
412 shutdown_busycount2(struct buf *bp, void *info)
413 {
414 	if (((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp)) ||
415 	    ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) {
416 		/*
417 		 * Only count buffers undergoing write I/O
418 		 * on the related vnode.
419 		 */
420 		if (bp->b_vp == NULL ||
421 		    bio_track_active(&bp->b_vp->v_track_write) == 0) {
422 			return (0);
423 		}
424 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC)
425 		kprintf(
426 	    "%p dev:?, flags:%08x, loffset:%lld, doffset:%lld\n",
427 		    bp,
428 		    bp->b_flags, bp->b_loffset,
429 		    bp->b_bio2.bio_offset);
430 #endif
431 		return(1);
432 	}
433 	return(0);
434 }
435 
436 /*
437  * If the shutdown was a clean halt, behave accordingly.
438  */
439 static void
440 shutdown_halt(void *junk, int howto)
441 {
442 	if (howto & RB_HALT) {
443 		kprintf("\n");
444 		kprintf("The operating system has halted.\n");
445 #ifdef _KERNEL_VIRTUAL
446 		cpu_halt();
447 #else
448 		kprintf("Please press any key to reboot.\n\n");
449 		switch (cngetc()) {
450 		case -1:		/* No console, just die */
451 			cpu_halt();
452 			/* NOTREACHED */
453 		default:
454 			howto &= ~RB_HALT;
455 			break;
456 		}
457 #endif
458 	}
459 }
460 
461 /*
462  * Check to see if the system paniced, pause and then reboot
463  * according to the specified delay.
464  */
465 static void
466 shutdown_panic(void *junk, int howto)
467 {
468 	int loop;
469 
470 	if (howto & RB_DUMP) {
471 		if (PANIC_REBOOT_WAIT_TIME != 0) {
472 			if (PANIC_REBOOT_WAIT_TIME != -1) {
473 				kprintf("Automatic reboot in %d seconds - "
474 				       "press a key on the console to abort\n",
475 					PANIC_REBOOT_WAIT_TIME);
476 				for (loop = PANIC_REBOOT_WAIT_TIME * 10;
477 				     loop > 0; --loop) {
478 					DELAY(1000 * 100); /* 1/10th second */
479 					/* Did user type a key? */
480 					if (cncheckc() != -1)
481 						break;
482 				}
483 				if (!loop)
484 					return;
485 			}
486 		} else { /* zero time specified - reboot NOW */
487 			return;
488 		}
489 		kprintf("--> Press a key on the console to reboot,\n");
490 		kprintf("--> or switch off the system now.\n");
491 		cngetc();
492 	}
493 }
494 
495 /*
496  * Everything done, now reset
497  */
498 static void
499 shutdown_reset(void *junk, int howto)
500 {
501 	kprintf("Rebooting...\n");
502 	DELAY(1000000);	/* wait 1 sec for kprintf's to complete and be read */
503 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
504 	cpu_reset();
505 	/* NOTREACHED */ /* assuming reset worked */
506 }
507 
508 /*
509  * Try to remove FS references in the specified process.  This function
510  * is used during shutdown
511  */
512 static
513 void
514 shutdown_cleanup_proc(struct proc *p)
515 {
516 	struct filedesc *fdp;
517 	struct vmspace *vm;
518 
519 	if (p == NULL)
520 		return;
521 	if ((fdp = p->p_fd) != NULL) {
522 		kern_closefrom(0);
523 		if (fdp->fd_cdir) {
524 			cache_drop(&fdp->fd_ncdir);
525 			vrele(fdp->fd_cdir);
526 			fdp->fd_cdir = NULL;
527 		}
528 		if (fdp->fd_rdir) {
529 			cache_drop(&fdp->fd_nrdir);
530 			vrele(fdp->fd_rdir);
531 			fdp->fd_rdir = NULL;
532 		}
533 		if (fdp->fd_jdir) {
534 			cache_drop(&fdp->fd_njdir);
535 			vrele(fdp->fd_jdir);
536 			fdp->fd_jdir = NULL;
537 		}
538 	}
539 	if (p->p_vkernel)
540 		vkernel_exit(p);
541 	if (p->p_textvp) {
542 		vrele(p->p_textvp);
543 		p->p_textvp = NULL;
544 	}
545 	vm = p->p_vmspace;
546 	if (vm != NULL) {
547 		pmap_remove_pages(vmspace_pmap(vm),
548 				  VM_MIN_USER_ADDRESS,
549 				  VM_MAX_USER_ADDRESS);
550 		vm_map_remove(&vm->vm_map,
551 			      VM_MIN_USER_ADDRESS,
552 			      VM_MAX_USER_ADDRESS);
553 	}
554 }
555 
556 /*
557  * Magic number for savecore
558  *
559  * exported (symorder) and used at least by savecore(8)
560  *
561  * Mark it as used so that gcc doesn't optimize it away.
562  */
563 __attribute__((__used__))
564 	static u_long const dumpmag = 0x8fca0101UL;
565 
566 __attribute__((__used__))
567 	static int	dumpsize = 0;		/* also for savecore */
568 
569 static int	dodump = 1;
570 
571 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0,
572     "Try to perform coredump on kernel panic");
573 
574 void
575 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver,
576     uint64_t dumplen, uint32_t blksz)
577 {
578 	bzero(kdh, sizeof(*kdh));
579 	strncpy(kdh->magic, magic, sizeof(kdh->magic));
580 	strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
581 	kdh->version = htod32(KERNELDUMPVERSION);
582 	kdh->architectureversion = htod32(archver);
583 	kdh->dumplength = htod64(dumplen);
584 	kdh->dumptime = htod64(time_second);
585 	kdh->blocksize = htod32(blksz);
586 	strncpy(kdh->hostname, hostname, sizeof(kdh->hostname));
587 	strncpy(kdh->versionstring, version, sizeof(kdh->versionstring));
588 	if (panicstr != NULL)
589 		strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
590 	kdh->parity = kerneldump_parity(kdh);
591 }
592 
593 static int
594 setdumpdev(cdev_t dev)
595 {
596 	int error;
597 	int doopen;
598 
599 	if (dev == NULL) {
600 		disk_dumpconf(NULL, 0/*off*/);
601 		return (0);
602 	}
603 
604 	/*
605 	 * We have to open the device before we can perform ioctls on it,
606 	 * or the slice/label data may not be present.  Device opens are
607 	 * usually tracked by specfs, but the dump device can be set in
608 	 * early boot and may not be open so this is somewhat of a hack.
609 	 */
610 	doopen = (dev->si_sysref.refcnt == 1);
611 	if (doopen) {
612 		error = dev_dopen(dev, FREAD, S_IFCHR, proc0.p_ucred);
613 		if (error)
614 			return (error);
615 	}
616 	error = disk_dumpconf(dev, 1/*on*/);
617 
618 	return error;
619 }
620 
621 /* ARGSUSED */
622 static void dump_conf (void *dummy);
623 static void
624 dump_conf(void *dummy)
625 {
626 	char *path;
627 	cdev_t dev;
628 
629 	path = kmalloc(MNAMELEN, M_TEMP, M_WAITOK);
630 	if (TUNABLE_STR_FETCH("dumpdev", path, MNAMELEN) != 0) {
631 		dev = kgetdiskbyname(path);
632 		if (dev != NULL)
633 			dumpdev = dev;
634 	}
635 	kfree(path, M_TEMP);
636 	if (setdumpdev(dumpdev) != 0)
637 		dumpdev = NULL;
638 }
639 
640 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL)
641 
642 static int
643 sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS)
644 {
645 	int error;
646 	udev_t ndumpdev;
647 
648 	ndumpdev = dev2udev(dumpdev);
649 	error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req);
650 	if (error == 0 && req->newptr != NULL)
651 		error = setdumpdev(udev2dev(ndumpdev, 0));
652 	return (error);
653 }
654 
655 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW,
656 	0, sizeof dumpdev, sysctl_kern_dumpdev, "T,udev_t", "");
657 
658 /*
659  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
660  * and then reboots.  If we are called twice, then we avoid trying to sync
661  * the disks as this often leads to recursive panics.
662  */
663 void
664 panic(const char *fmt, ...)
665 {
666 	int bootopt, newpanic;
667 	__va_list ap;
668 	static char buf[256];
669 
670 #ifdef SMP
671 	/*
672 	 * If a panic occurs on multiple cpus before the first is able to
673 	 * halt the other cpus, only one cpu is allowed to take the panic.
674 	 * Attempt to be verbose about this situation but if the kprintf()
675 	 * itself panics don't let us overrun the kernel stack.
676 	 *
677 	 * Be very nasty about descheduling our thread at the lowest
678 	 * level possible in an attempt to freeze the thread without
679 	 * inducing further panics.
680 	 *
681 	 * Bumping gd_trap_nesting_level will also bypass assertions in
682 	 * lwkt_switch() and allow us to switch away even if we are a
683 	 * FAST interrupt or IPI.
684 	 */
685 	if (atomic_poll_acquire_int(&panic_cpu_interlock)) {
686 		panic_cpu_gd = mycpu;
687 	} else if (panic_cpu_gd != mycpu) {
688 		crit_enter();
689 		++mycpu->gd_trap_nesting_level;
690 		if (mycpu->gd_trap_nesting_level < 25) {
691 			kprintf("SECONDARY PANIC ON CPU %d THREAD %p\n",
692 				mycpu->gd_cpuid, curthread);
693 		}
694 		curthread->td_release = NULL;	/* be a grinch */
695 		for (;;) {
696 			lwkt_deschedule_self(curthread);
697 			lwkt_switch();
698 		}
699 		/* NOT REACHED */
700 		/* --mycpu->gd_trap_nesting_level */
701 		/* crit_exit() */
702 	}
703 #endif
704 	bootopt = RB_AUTOBOOT | RB_DUMP;
705 	if (sync_on_panic == 0)
706 		bootopt |= RB_NOSYNC;
707 	newpanic = 0;
708 	if (panicstr)
709 		bootopt |= RB_NOSYNC;
710 	else {
711 		panicstr = fmt;
712 		newpanic = 1;
713 	}
714 
715 	__va_start(ap, fmt);
716 	kvsnprintf(buf, sizeof(buf), fmt, ap);
717 	if (panicstr == fmt)
718 		panicstr = buf;
719 	__va_end(ap);
720 	kprintf("panic: %s\n", buf);
721 #ifdef SMP
722 	/* two separate prints in case of an unmapped page and trap */
723 	kprintf("mp_lock = %08x; ", mp_lock);
724 	kprintf("cpuid = %d\n", mycpu->gd_cpuid);
725 #endif
726 
727 #if (NGPIO > 0) && defined(ERROR_LED_ON_PANIC)
728 	led_switch("error", 1);
729 #endif
730 
731 #if defined(WDOG_DISABLE_ON_PANIC) && defined(WATCHDOG_ENABLE)
732 	wdog_disable();
733 #endif
734 
735 #if defined(DDB)
736 	if (newpanic && trace_on_panic)
737 		print_backtrace();
738 	if (debugger_on_panic)
739 		Debugger("panic");
740 #endif
741 	boot(bootopt);
742 }
743 
744 /*
745  * Support for poweroff delay.
746  */
747 #ifndef POWEROFF_DELAY
748 # define POWEROFF_DELAY 5000
749 #endif
750 static int poweroff_delay = POWEROFF_DELAY;
751 
752 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
753 	&poweroff_delay, 0, "");
754 
755 static void
756 poweroff_wait(void *junk, int howto)
757 {
758 	if(!(howto & RB_POWEROFF) || poweroff_delay <= 0)
759 		return;
760 	DELAY(poweroff_delay * 1000);
761 }
762 
763 /*
764  * Some system processes (e.g. syncer) need to be stopped at appropriate
765  * points in their main loops prior to a system shutdown, so that they
766  * won't interfere with the shutdown process (e.g. by holding a disk buf
767  * to cause sync to fail).  For each of these system processes, register
768  * shutdown_kproc() as a handler for one of shutdown events.
769  */
770 static int kproc_shutdown_wait = 60;
771 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
772     &kproc_shutdown_wait, 0, "");
773 
774 void
775 shutdown_kproc(void *arg, int howto)
776 {
777 	struct thread *td;
778 	struct proc *p;
779 	int error;
780 
781 	if (panicstr)
782 		return;
783 
784 	td = (struct thread *)arg;
785 	if ((p = td->td_proc) != NULL) {
786 	    kprintf("Waiting (max %d seconds) for system process `%s' to stop...",
787 		kproc_shutdown_wait, p->p_comm);
788 	} else {
789 	    kprintf("Waiting (max %d seconds) for system thread %s to stop...",
790 		kproc_shutdown_wait, td->td_comm);
791 	}
792 	error = suspend_kproc(td, kproc_shutdown_wait * hz);
793 
794 	if (error == EWOULDBLOCK)
795 		kprintf("timed out\n");
796 	else
797 		kprintf("stopped\n");
798 }
799 
800 /* Registration of dumpers */
801 int
802 set_dumper(struct dumperinfo *di)
803 {
804 	if (di == NULL) {
805 		bzero(&dumper, sizeof(dumper));
806 		return 0;
807 	}
808 
809 	if (dumper.dumper != NULL)
810 		return (EBUSY);
811 
812 	dumper = *di;
813 	return 0;
814 }
815 
816 #if defined (_KERNEL_VIRTUAL)
817 /* VKERNELs don't support dumps */
818 void
819 dumpsys(struct dumperinfo *di __unused)
820 {
821 	kprintf("VKERNEL doesn't support dumps\n");
822 }
823 #endif
824