]> CyberLeo.Net >> Repos - FreeBSD/stable/10.git/blob - sys/kern/kern_shutdown.c
Copy head (r256279) to stable/10 as part of the 10.0-RELEASE cycle.
[FreeBSD/stable/10.git] / sys / kern / kern_shutdown.c
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
93 /*
94  * Note that stdarg.h and the ANSI style va_start macro is used for both
95  * ANSI and traditional C compilers.
96  */
97 #include <machine/stdarg.h>
98
99 #ifdef KDB
100 #ifdef KDB_UNATTENDED
101 int debugger_on_panic = 0;
102 #else
103 int debugger_on_panic = 1;
104 #endif
105 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic,
106     CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_TUN,
107     &debugger_on_panic, 0, "Run debugger on kernel panic");
108 TUNABLE_INT("debug.debugger_on_panic", &debugger_on_panic);
109
110 #ifdef KDB_TRACE
111 static int trace_on_panic = 1;
112 #else
113 static int trace_on_panic = 0;
114 #endif
115 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
116     CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_TUN,
117     &trace_on_panic, 0, "Print stack trace on kernel panic");
118 TUNABLE_INT("debug.trace_on_panic", &trace_on_panic);
119 #endif /* KDB */
120
121 static int sync_on_panic = 0;
122 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW | CTLFLAG_TUN,
123         &sync_on_panic, 0, "Do a sync before rebooting from a panic");
124 TUNABLE_INT("kern.sync_on_panic", &sync_on_panic);
125
126 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0,
127     "Shutdown environment");
128
129 #ifndef DIAGNOSTIC
130 static int show_busybufs;
131 #else
132 static int show_busybufs = 1;
133 #endif
134 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
135         &show_busybufs, 0, "");
136
137 /*
138  * Variable panicstr contains argument to first call to panic; used as flag
139  * to indicate that the kernel has already called panic.
140  */
141 const char *panicstr;
142
143 int dumping;                            /* system is dumping */
144 int rebooting;                          /* system is rebooting */
145 static struct dumperinfo dumper;        /* our selected dumper */
146
147 /* Context information for dump-debuggers. */
148 static struct pcb dumppcb;              /* Registers. */
149 lwpid_t dumptid;                        /* Thread ID. */
150
151 static void poweroff_wait(void *, int);
152 static void shutdown_halt(void *junk, int howto);
153 static void shutdown_panic(void *junk, int howto);
154 static void shutdown_reset(void *junk, int howto);
155 static void vpanic(const char *fmt, va_list ap) __dead2;
156
157 /* register various local shutdown events */
158 static void
159 shutdown_conf(void *unused)
160 {
161
162         EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
163             SHUTDOWN_PRI_FIRST);
164         EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
165             SHUTDOWN_PRI_LAST + 100);
166         EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
167             SHUTDOWN_PRI_LAST + 100);
168         EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
169             SHUTDOWN_PRI_LAST + 200);
170 }
171
172 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
173
174 /*
175  * The system call that results in a reboot.
176  */
177 /* ARGSUSED */
178 int
179 sys_reboot(struct thread *td, struct reboot_args *uap)
180 {
181         int error;
182
183         error = 0;
184 #ifdef MAC
185         error = mac_system_check_reboot(td->td_ucred, uap->opt);
186 #endif
187         if (error == 0)
188                 error = priv_check(td, PRIV_REBOOT);
189         if (error == 0) {
190                 mtx_lock(&Giant);
191                 kern_reboot(uap->opt);
192                 mtx_unlock(&Giant);
193         }
194         return (error);
195 }
196
197 /*
198  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
199  */
200 static int shutdown_howto = 0;
201
202 void
203 shutdown_nice(int howto)
204 {
205
206         shutdown_howto = howto;
207
208         /* Send a signal to init(8) and have it shutdown the world */
209         if (initproc != NULL) {
210                 PROC_LOCK(initproc);
211                 kern_psignal(initproc, SIGINT);
212                 PROC_UNLOCK(initproc);
213         } else {
214                 /* No init(8) running, so simply reboot */
215                 kern_reboot(RB_NOSYNC);
216         }
217         return;
218 }
219 static int      waittime = -1;
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
294 #if defined(SMP)
295         /*
296          * Bind us to CPU 0 so that all shutdown code runs there.  Some
297          * systems don't shutdown properly (i.e., ACPI power off) if we
298          * run on another processor.
299          */
300         if (!SCHEDULER_STOPPED()) {
301                 thread_lock(curthread);
302                 sched_bind(curthread, 0);
303                 thread_unlock(curthread);
304                 KASSERT(PCPU_GET(cpuid) == 0, ("boot: not running on cpu 0"));
305         }
306 #endif
307         /* We're in the process of rebooting. */
308         rebooting = 1;
309
310         /* collect extra flags that shutdown_nice might have set */
311         howto |= shutdown_howto;
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_RW | CTLFLAG_TUN,
561     &kassert_warn_only, 0,
562     "KASSERT triggers a panic (1) or just a warning (0)");
563 TUNABLE_INT("debug.kassert.warn_only", &kassert_warn_only);
564
565 #ifdef KDB
566 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, CTLFLAG_RW | CTLFLAG_TUN,
567     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
568 TUNABLE_INT("debug.kassert.do_kdb", &kassert_do_kdb);
569 #endif
570
571 #ifdef KTR
572 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, CTLFLAG_RW | CTLFLAG_TUN,
573     &kassert_do_ktr, 0,
574     "KASSERT does a KTR, set this to the KTRMASK you want");
575 TUNABLE_INT("debug.kassert.do_ktr", &kassert_do_ktr);
576 #endif
577
578 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, CTLFLAG_RW | CTLFLAG_TUN,
579     &kassert_do_log, 0, "KASSERT triggers a panic (1) or just a warning (0)");
580 TUNABLE_INT("debug.kassert.do_log", &kassert_do_log);
581
582 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RW | CTLFLAG_TUN,
583     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
584 TUNABLE_INT("debug.kassert.warnings", &kassert_warnings);
585
586 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, CTLFLAG_RW | CTLFLAG_TUN,
587     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
588 TUNABLE_INT("debug.kassert.log_panic_at", &kassert_log_panic_at);
589
590 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, CTLFLAG_RW | CTLFLAG_TUN,
591     &kassert_log_pps_limit, 0, "limit number of log messages per second");
592 TUNABLE_INT("debug.kassert.log_pps_limit", &kassert_log_pps_limit);
593
594 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, CTLFLAG_RW | CTLFLAG_TUN,
595     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
596 TUNABLE_INT("debug.kassert.log_mute_at", &kassert_log_mute_at);
597
598 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
599
600 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
601     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE, NULL, 0,
602     kassert_sysctl_kassert, "I", "set to trigger a test kassert");
603
604 static int
605 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
606 {
607         int error, i;
608
609         error = sysctl_wire_old_buffer(req, sizeof(int));
610         if (error == 0) {
611                 i = 0;
612                 error = sysctl_handle_int(oidp, &i, 0, req);
613         }
614         if (error != 0 || req->newptr == NULL)
615                 return (error);
616         KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
617         return (0);
618 }
619
620 /*
621  * Called by KASSERT, this decides if we will panic
622  * or if we will log via printf and/or ktr.
623  */
624 void
625 kassert_panic(const char *fmt, ...)
626 {
627         static char buf[256];
628         va_list ap;
629
630         va_start(ap, fmt);
631         (void)vsnprintf(buf, sizeof(buf), fmt, ap);
632         va_end(ap);
633
634         /*
635          * panic if we're not just warning, or if we've exceeded
636          * kassert_log_panic_at warnings.
637          */
638         if (!kassert_warn_only ||
639             (kassert_log_panic_at > 0 &&
640              kassert_warnings >= kassert_log_panic_at)) {
641                 va_start(ap, fmt);
642                 vpanic(fmt, ap);
643                 /* NORETURN */
644         }
645 #ifdef KTR
646         if (kassert_do_ktr)
647                 CTR0(ktr_mask, buf);
648 #endif /* KTR */
649         /*
650          * log if we've not yet met the mute limit.
651          */
652         if (kassert_do_log &&
653             (kassert_log_mute_at == 0 ||
654              kassert_warnings < kassert_log_mute_at)) {
655                 static  struct timeval lasterr;
656                 static  int curerr;
657
658                 if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
659                         printf("KASSERT failed: %s\n", buf);
660                         kdb_backtrace();
661                 }
662         }
663 #ifdef KDB
664         if (kassert_do_kdb) {
665                 kdb_enter(KDB_WHY_KASSERT, buf);
666         }
667 #endif
668         atomic_add_int(&kassert_warnings, 1);
669 }
670 #endif
671
672 /*
673  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
674  * and then reboots.  If we are called twice, then we avoid trying to sync
675  * the disks as this often leads to recursive panics.
676  */
677 void
678 panic(const char *fmt, ...)
679 {
680         va_list ap;
681
682         va_start(ap, fmt);
683         vpanic(fmt, ap);
684 }
685
686 static void
687 vpanic(const char *fmt, va_list ap)
688 {
689 #ifdef SMP
690         cpuset_t other_cpus;
691 #endif
692         struct thread *td = curthread;
693         int bootopt, newpanic;
694         static char buf[256];
695
696         spinlock_enter();
697
698 #ifdef SMP
699         /*
700          * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
701          * concurrently entering panic.  Only the winner will proceed
702          * further.
703          */
704         if (panicstr == NULL && !kdb_active) {
705                 other_cpus = all_cpus;
706                 CPU_CLR(PCPU_GET(cpuid), &other_cpus);
707                 stop_cpus_hard(other_cpus);
708         }
709
710         /*
711          * We set stop_scheduler here and not in the block above,
712          * because we want to ensure that if panic has been called and
713          * stop_scheduler_on_panic is true, then stop_scheduler will
714          * always be set.  Even if panic has been entered from kdb.
715          */
716         td->td_stopsched = 1;
717 #endif
718
719         bootopt = RB_AUTOBOOT;
720         newpanic = 0;
721         if (panicstr)
722                 bootopt |= RB_NOSYNC;
723         else {
724                 bootopt |= RB_DUMP;
725                 panicstr = fmt;
726                 newpanic = 1;
727         }
728
729         if (newpanic) {
730                 (void)vsnprintf(buf, sizeof(buf), fmt, ap);
731                 panicstr = buf;
732                 cngrab();
733                 printf("panic: %s\n", buf);
734         } else {
735                 printf("panic: ");
736                 vprintf(fmt, ap);
737                 printf("\n");
738         }
739 #ifdef SMP
740         printf("cpuid = %d\n", PCPU_GET(cpuid));
741 #endif
742
743 #ifdef KDB
744         if (newpanic && trace_on_panic)
745                 kdb_backtrace();
746         if (debugger_on_panic)
747                 kdb_enter(KDB_WHY_PANIC, "panic");
748 #endif
749         /*thread_lock(td); */
750         td->td_flags |= TDF_INPANIC;
751         /* thread_unlock(td); */
752         if (!sync_on_panic)
753                 bootopt |= RB_NOSYNC;
754         kern_reboot(bootopt);
755 }
756
757 /*
758  * Support for poweroff delay.
759  *
760  * Please note that setting this delay too short might power off your machine
761  * before the write cache on your hard disk has been flushed, leading to
762  * soft-updates inconsistencies.
763  */
764 #ifndef POWEROFF_DELAY
765 # define POWEROFF_DELAY 5000
766 #endif
767 static int poweroff_delay = POWEROFF_DELAY;
768
769 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
770     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
771
772 static void
773 poweroff_wait(void *junk, int howto)
774 {
775
776         if (!(howto & RB_POWEROFF) || poweroff_delay <= 0)
777                 return;
778         DELAY(poweroff_delay * 1000);
779 }
780
781 /*
782  * Some system processes (e.g. syncer) need to be stopped at appropriate
783  * points in their main loops prior to a system shutdown, so that they
784  * won't interfere with the shutdown process (e.g. by holding a disk buf
785  * to cause sync to fail).  For each of these system processes, register
786  * shutdown_kproc() as a handler for one of shutdown events.
787  */
788 static int kproc_shutdown_wait = 60;
789 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
790     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
791
792 void
793 kproc_shutdown(void *arg, int howto)
794 {
795         struct proc *p;
796         int error;
797
798         if (panicstr)
799                 return;
800
801         p = (struct proc *)arg;
802         printf("Waiting (max %d seconds) for system process `%s' to stop...",
803             kproc_shutdown_wait, p->p_comm);
804         error = kproc_suspend(p, kproc_shutdown_wait * hz);
805
806         if (error == EWOULDBLOCK)
807                 printf("timed out\n");
808         else
809                 printf("done\n");
810 }
811
812 void
813 kthread_shutdown(void *arg, int howto)
814 {
815         struct thread *td;
816         int error;
817
818         if (panicstr)
819                 return;
820
821         td = (struct thread *)arg;
822         printf("Waiting (max %d seconds) for system thread `%s' to stop...",
823             kproc_shutdown_wait, td->td_name);
824         error = kthread_suspend(td, kproc_shutdown_wait * hz);
825
826         if (error == EWOULDBLOCK)
827                 printf("timed out\n");
828         else
829                 printf("done\n");
830 }
831
832 static char dumpdevname[sizeof(((struct cdev*)NULL)->si_name)];
833 SYSCTL_STRING(_kern_shutdown, OID_AUTO, dumpdevname, CTLFLAG_RD,
834     dumpdevname, 0, "Device for kernel dumps");
835
836 /* Registration of dumpers */
837 int
838 set_dumper(struct dumperinfo *di, const char *devname)
839 {
840         size_t wantcopy;
841
842         if (di == NULL) {
843                 bzero(&dumper, sizeof dumper);
844                 dumpdevname[0] = '\0';
845                 return (0);
846         }
847         if (dumper.dumper != NULL)
848                 return (EBUSY);
849         dumper = *di;
850         wantcopy = strlcpy(dumpdevname, devname, sizeof(dumpdevname));
851         if (wantcopy >= sizeof(dumpdevname)) {
852                 printf("set_dumper: device name truncated from '%s' -> '%s'\n",
853                         devname, dumpdevname);
854         }
855         return (0);
856 }
857
858 /* Call dumper with bounds checking. */
859 int
860 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
861     off_t offset, size_t length)
862 {
863
864         if (length != 0 && (offset < di->mediaoffset ||
865             offset - di->mediaoffset + length > di->mediasize)) {
866                 printf("Attempt to write outside dump device boundaries.\n"
867             "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
868                     (intmax_t)offset, (intmax_t)di->mediaoffset,
869                     (uintmax_t)length, (intmax_t)di->mediasize);
870                 return (ENOSPC);
871         }
872         return (di->dumper(di->priv, virtual, physical, offset, length));
873 }
874
875 void
876 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver,
877     uint64_t dumplen, uint32_t blksz)
878 {
879
880         bzero(kdh, sizeof(*kdh));
881         strncpy(kdh->magic, magic, sizeof(kdh->magic));
882         strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
883         kdh->version = htod32(KERNELDUMPVERSION);
884         kdh->architectureversion = htod32(archver);
885         kdh->dumplength = htod64(dumplen);
886         kdh->dumptime = htod64(time_second);
887         kdh->blocksize = htod32(blksz);
888         strncpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
889         strncpy(kdh->versionstring, version, sizeof(kdh->versionstring));
890         if (panicstr != NULL)
891                 strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
892         kdh->parity = kerneldump_parity(kdh);
893 }