]> CyberLeo.Net >> Repos - FreeBSD/stable/8.git/blob - sys/kern/kern_shutdown.c
Copy head to stable/8 as part of 8.0 Release cycle.
[FreeBSD/stable/8.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_show_busybufs.h"
44 #include "opt_sched.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/malloc.h>
59 #include <sys/mount.h>
60 #include <sys/priv.h>
61 #include <sys/proc.h>
62 #include <sys/reboot.h>
63 #include <sys/resourcevar.h>
64 #include <sys/sched.h>
65 #include <sys/smp.h>            /* smp_active */
66 #include <sys/sysctl.h>
67 #include <sys/sysproto.h>
68
69 #include <ddb/ddb.h>
70
71 #include <machine/cpu.h>
72 #include <machine/pcb.h>
73 #include <machine/smp.h>
74
75 #include <security/mac/mac_framework.h>
76
77 #include <vm/vm.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_pager.h>
81 #include <vm/swap_pager.h>
82
83 #include <sys/signalvar.h>
84
85 #ifndef PANIC_REBOOT_WAIT_TIME
86 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
87 #endif
88
89 /*
90  * Note that stdarg.h and the ANSI style va_start macro is used for both
91  * ANSI and traditional C compilers.
92  */
93 #include <machine/stdarg.h>
94
95 #ifdef KDB
96 #ifdef KDB_UNATTENDED
97 int debugger_on_panic = 0;
98 #else
99 int debugger_on_panic = 1;
100 #endif
101 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW,
102         &debugger_on_panic, 0, "Run debugger on kernel panic");
103
104 #ifdef KDB_TRACE
105 int trace_on_panic = 1;
106 #else
107 int trace_on_panic = 0;
108 #endif
109 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW,
110         &trace_on_panic, 0, "Print stack trace on kernel panic");
111 #endif /* KDB */
112
113 int sync_on_panic = 0;
114 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW,
115         &sync_on_panic, 0, "Do a sync before rebooting from a panic");
116
117 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment");
118
119 /*
120  * Variable panicstr contains argument to first call to panic; used as flag
121  * to indicate that the kernel has already called panic.
122  */
123 const char *panicstr;
124
125 int dumping;                            /* system is dumping */
126 int rebooting;                          /* system is rebooting */
127 static struct dumperinfo dumper;        /* our selected dumper */
128
129 /* Context information for dump-debuggers. */
130 static struct pcb dumppcb;              /* Registers. */
131 static lwpid_t dumptid;                 /* Thread ID. */
132
133 static void boot(int) __dead2;
134 static void poweroff_wait(void *, int);
135 static void shutdown_halt(void *junk, int howto);
136 static void shutdown_panic(void *junk, int howto);
137 static void shutdown_reset(void *junk, int howto);
138
139 /* register various local shutdown events */
140 static void
141 shutdown_conf(void *unused)
142 {
143
144         EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
145             SHUTDOWN_PRI_FIRST);
146         EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
147             SHUTDOWN_PRI_LAST + 100);
148         EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
149             SHUTDOWN_PRI_LAST + 100);
150         EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
151             SHUTDOWN_PRI_LAST + 200);
152 }
153
154 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
155
156 /*
157  * The system call that results in a reboot.
158  */
159 /* ARGSUSED */
160 int
161 reboot(struct thread *td, struct reboot_args *uap)
162 {
163         int error;
164
165         error = 0;
166 #ifdef MAC
167         error = mac_system_check_reboot(td->td_ucred, uap->opt);
168 #endif
169         if (error == 0)
170                 error = priv_check(td, PRIV_REBOOT);
171         if (error == 0) {
172                 mtx_lock(&Giant);
173                 boot(uap->opt);
174                 mtx_unlock(&Giant);
175         }
176         return (error);
177 }
178
179 /*
180  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
181  */
182 static int shutdown_howto = 0;
183
184 void
185 shutdown_nice(int howto)
186 {
187
188         shutdown_howto = howto;
189
190         /* Send a signal to init(8) and have it shutdown the world */
191         if (initproc != NULL) {
192                 PROC_LOCK(initproc);
193                 psignal(initproc, SIGINT);
194                 PROC_UNLOCK(initproc);
195         } else {
196                 /* No init(8) running, so simply reboot */
197                 boot(RB_NOSYNC);
198         }
199         return;
200 }
201 static int      waittime = -1;
202
203 static void
204 print_uptime(void)
205 {
206         int f;
207         struct timespec ts;
208
209         getnanouptime(&ts);
210         printf("Uptime: ");
211         f = 0;
212         if (ts.tv_sec >= 86400) {
213                 printf("%ldd", (long)ts.tv_sec / 86400);
214                 ts.tv_sec %= 86400;
215                 f = 1;
216         }
217         if (f || ts.tv_sec >= 3600) {
218                 printf("%ldh", (long)ts.tv_sec / 3600);
219                 ts.tv_sec %= 3600;
220                 f = 1;
221         }
222         if (f || ts.tv_sec >= 60) {
223                 printf("%ldm", (long)ts.tv_sec / 60);
224                 ts.tv_sec %= 60;
225                 f = 1;
226         }
227         printf("%lds\n", (long)ts.tv_sec);
228 }
229
230 static void
231 doadump(void)
232 {
233
234         /*
235          * Sometimes people have to call this from the kernel debugger. 
236          * (if 'panic' can not dump)
237          * Give them a clue as to why they can't dump.
238          */
239         if (dumper.dumper == NULL) {
240                 printf("Cannot dump. Device not defined or unavailable.\n");
241                 return;
242         }
243
244         savectx(&dumppcb);
245         dumptid = curthread->td_tid;
246         dumping++;
247 #ifdef DDB
248         if (textdump_pending)
249                 textdump_dumpsys(&dumper);
250         else
251 #endif
252                 dumpsys(&dumper);
253         dumping--;
254 }
255
256 static int
257 isbufbusy(struct buf *bp)
258 {
259         if (((bp->b_flags & (B_INVAL | B_PERSISTENT)) == 0 &&
260             BUF_ISLOCKED(bp)) ||
261             ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI))
262                 return (1);
263         return (0);
264 }
265
266 /*
267  * Shutdown the system cleanly to prepare for reboot, halt, or power off.
268  */
269 static void
270 boot(int howto)
271 {
272         static int first_buf_printf = 1;
273
274 #if defined(SMP)
275         /*
276          * Bind us to CPU 0 so that all shutdown code runs there.  Some
277          * systems don't shutdown properly (i.e., ACPI power off) if we
278          * run on another processor.
279          */
280         thread_lock(curthread);
281         sched_bind(curthread, 0);
282         thread_unlock(curthread);
283         KASSERT(PCPU_GET(cpuid) == 0, ("boot: not running on cpu 0"));
284 #endif
285         /* We're in the process of rebooting. */
286         rebooting = 1;
287
288         /* collect extra flags that shutdown_nice might have set */
289         howto |= shutdown_howto;
290
291         /* We are out of the debugger now. */
292         kdb_active = 0;
293
294         /*
295          * Do any callouts that should be done BEFORE syncing the filesystems.
296          */
297         EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
298
299         /* 
300          * Now sync filesystems
301          */
302         if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
303                 register struct buf *bp;
304                 int iter, nbusy, pbusy;
305 #ifndef PREEMPTION
306                 int subiter;
307 #endif
308
309                 waittime = 0;
310
311                 sync(curthread, NULL);
312
313                 /*
314                  * With soft updates, some buffers that are
315                  * written will be remarked as dirty until other
316                  * buffers are written.
317                  */
318                 for (iter = pbusy = 0; iter < 20; iter++) {
319                         nbusy = 0;
320                         for (bp = &buf[nbuf]; --bp >= buf; )
321                                 if (isbufbusy(bp))
322                                         nbusy++;
323                         if (nbusy == 0) {
324                                 if (first_buf_printf)
325                                         printf("All buffers synced.");
326                                 break;
327                         }
328                         if (first_buf_printf) {
329                                 printf("Syncing disks, buffers remaining... ");
330                                 first_buf_printf = 0;
331                         }
332                         printf("%d ", nbusy);
333                         if (nbusy < pbusy)
334                                 iter = 0;
335                         pbusy = nbusy;
336                         sync(curthread, NULL);
337
338 #ifdef PREEMPTION
339                         /*
340                          * Drop Giant and spin for a while to allow
341                          * interrupt threads to run.
342                          */
343                         DROP_GIANT();
344                         DELAY(50000 * iter);
345                         PICKUP_GIANT();
346 #else
347                         /*
348                          * Drop Giant and context switch several times to
349                          * allow interrupt threads to run.
350                          */
351                         DROP_GIANT();
352                         for (subiter = 0; subiter < 50 * iter; subiter++) {
353                                 thread_lock(curthread);
354                                 mi_switch(SW_VOL, NULL);
355                                 thread_unlock(curthread);
356                                 DELAY(1000);
357                         }
358                         PICKUP_GIANT();
359 #endif
360                 }
361                 printf("\n");
362                 /*
363                  * Count only busy local buffers to prevent forcing 
364                  * a fsck if we're just a client of a wedged NFS server
365                  */
366                 nbusy = 0;
367                 for (bp = &buf[nbuf]; --bp >= buf; ) {
368                         if (isbufbusy(bp)) {
369 #if 0
370 /* XXX: This is bogus.  We should probably have a BO_REMOTE flag instead */
371                                 if (bp->b_dev == NULL) {
372                                         TAILQ_REMOVE(&mountlist,
373                                             bp->b_vp->v_mount, mnt_list);
374                                         continue;
375                                 }
376 #endif
377                                 nbusy++;
378 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC)
379                                 printf(
380                             "%d: bufobj:%p, flags:%0x, blkno:%ld, lblkno:%ld\n",
381                                     nbusy, bp->b_bufobj,
382                                     bp->b_flags, (long)bp->b_blkno,
383                                     (long)bp->b_lblkno);
384 #endif
385                         }
386                 }
387                 if (nbusy) {
388                         /*
389                          * Failed to sync all blocks. Indicate this and don't
390                          * unmount filesystems (thus forcing an fsck on reboot).
391                          */
392                         printf("Giving up on %d buffers\n", nbusy);
393                         DELAY(5000000); /* 5 seconds */
394                 } else {
395                         if (!first_buf_printf)
396                                 printf("Final sync complete\n");
397                         /*
398                          * Unmount filesystems
399                          */
400                         if (panicstr == 0)
401                                 vfs_unmountall();
402                 }
403                 swapoff_all();
404                 DELAY(100000);          /* wait for console output to finish */
405         }
406
407         print_uptime();
408
409         /*
410          * Ok, now do things that assume all filesystem activity has
411          * been completed.
412          */
413         EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
414
415         /* XXX This doesn't disable interrupts any more.  Reconsider? */
416         splhigh();
417
418         if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping) 
419                 doadump();
420
421         /* Now that we're going to really halt the system... */
422         EVENTHANDLER_INVOKE(shutdown_final, howto);
423
424         for(;;) ;       /* safety against shutdown_reset not working */
425         /* NOTREACHED */
426 }
427
428 /*
429  * If the shutdown was a clean halt, behave accordingly.
430  */
431 static void
432 shutdown_halt(void *junk, int howto)
433 {
434
435         if (howto & RB_HALT) {
436                 printf("\n");
437                 printf("The operating system has halted.\n");
438                 printf("Please press any key to reboot.\n\n");
439                 switch (cngetc()) {
440                 case -1:                /* No console, just die */
441                         cpu_halt();
442                         /* NOTREACHED */
443                 default:
444                         howto &= ~RB_HALT;
445                         break;
446                 }
447         }
448 }
449
450 /*
451  * Check to see if the system paniced, pause and then reboot
452  * according to the specified delay.
453  */
454 static void
455 shutdown_panic(void *junk, int howto)
456 {
457         int loop;
458
459         if (howto & RB_DUMP) {
460                 if (PANIC_REBOOT_WAIT_TIME != 0) {
461                         if (PANIC_REBOOT_WAIT_TIME != -1) {
462                                 printf("Automatic reboot in %d seconds - "
463                                        "press a key on the console to abort\n",
464                                         PANIC_REBOOT_WAIT_TIME);
465                                 for (loop = PANIC_REBOOT_WAIT_TIME * 10;
466                                      loop > 0; --loop) {
467                                         DELAY(1000 * 100); /* 1/10th second */
468                                         /* Did user type a key? */
469                                         if (cncheckc() != -1)
470                                                 break;
471                                 }
472                                 if (!loop)
473                                         return;
474                         }
475                 } else { /* zero time specified - reboot NOW */
476                         return;
477                 }
478                 printf("--> Press a key on the console to reboot,\n");
479                 printf("--> or switch off the system now.\n");
480                 cngetc();
481         }
482 }
483
484 /*
485  * Everything done, now reset
486  */
487 static void
488 shutdown_reset(void *junk, int howto)
489 {
490
491         printf("Rebooting...\n");
492         DELAY(1000000); /* wait 1 sec for printf's to complete and be read */
493         /* cpu_boot(howto); */ /* doesn't do anything at the moment */
494         cpu_reset();
495         /* NOTREACHED */ /* assuming reset worked */
496 }
497
498 #ifdef SMP
499 static u_int panic_cpu = NOCPU;
500 #endif
501
502 /*
503  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
504  * and then reboots.  If we are called twice, then we avoid trying to sync
505  * the disks as this often leads to recursive panics.
506  */
507 void
508 panic(const char *fmt, ...)
509 {
510         struct thread *td = curthread;
511         int bootopt, newpanic;
512         va_list ap;
513         static char buf[256];
514
515         critical_enter();
516 #ifdef SMP
517         /*
518          * We don't want multiple CPU's to panic at the same time, so we
519          * use panic_cpu as a simple spinlock.  We have to keep checking
520          * panic_cpu if we are spinning in case the panic on the first
521          * CPU is canceled.
522          */
523         if (panic_cpu != PCPU_GET(cpuid))
524                 while (atomic_cmpset_int(&panic_cpu, NOCPU,
525                     PCPU_GET(cpuid)) == 0)
526                         while (panic_cpu != NOCPU)
527                                 ; /* nothing */
528 #endif
529
530         bootopt = RB_AUTOBOOT | RB_DUMP;
531         newpanic = 0;
532         if (panicstr)
533                 bootopt |= RB_NOSYNC;
534         else {
535                 panicstr = fmt;
536                 newpanic = 1;
537         }
538
539         va_start(ap, fmt);
540         if (newpanic) {
541                 (void)vsnprintf(buf, sizeof(buf), fmt, ap);
542                 panicstr = buf;
543                 printf("panic: %s\n", buf);
544         } else {
545                 printf("panic: ");
546                 vprintf(fmt, ap);
547                 printf("\n");
548         }
549         va_end(ap);
550 #ifdef SMP
551         printf("cpuid = %d\n", PCPU_GET(cpuid));
552 #endif
553
554 #ifdef KDB
555         if (newpanic && trace_on_panic)
556                 kdb_backtrace();
557         if (debugger_on_panic)
558                 kdb_enter(KDB_WHY_PANIC, "panic");
559 #ifdef RESTARTABLE_PANICS
560         /* See if the user aborted the panic, in which case we continue. */
561         if (panicstr == NULL) {
562 #ifdef SMP
563                 atomic_store_rel_int(&panic_cpu, NOCPU);
564 #endif
565                 return;
566         }
567 #endif
568 #endif
569         /*thread_lock(td); */
570         td->td_flags |= TDF_INPANIC;
571         /* thread_unlock(td); */
572         if (!sync_on_panic)
573                 bootopt |= RB_NOSYNC;
574         critical_exit();
575         boot(bootopt);
576 }
577
578 /*
579  * Support for poweroff delay.
580  */
581 #ifndef POWEROFF_DELAY
582 # define POWEROFF_DELAY 5000
583 #endif
584 static int poweroff_delay = POWEROFF_DELAY;
585
586 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
587         &poweroff_delay, 0, "");
588
589 static void
590 poweroff_wait(void *junk, int howto)
591 {
592
593         if (!(howto & RB_POWEROFF) || poweroff_delay <= 0)
594                 return;
595         DELAY(poweroff_delay * 1000);
596 }
597
598 /*
599  * Some system processes (e.g. syncer) need to be stopped at appropriate
600  * points in their main loops prior to a system shutdown, so that they
601  * won't interfere with the shutdown process (e.g. by holding a disk buf
602  * to cause sync to fail).  For each of these system processes, register
603  * shutdown_kproc() as a handler for one of shutdown events.
604  */
605 static int kproc_shutdown_wait = 60;
606 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
607     &kproc_shutdown_wait, 0, "");
608
609 void
610 kproc_shutdown(void *arg, int howto)
611 {
612         struct proc *p;
613         char procname[MAXCOMLEN + 1];
614         int error;
615
616         if (panicstr)
617                 return;
618
619         p = (struct proc *)arg;
620         strlcpy(procname, p->p_comm, sizeof(procname));
621         printf("Waiting (max %d seconds) for system process `%s' to stop...",
622             kproc_shutdown_wait, procname);
623         error = kproc_suspend(p, kproc_shutdown_wait * hz);
624
625         if (error == EWOULDBLOCK)
626                 printf("timed out\n");
627         else
628                 printf("done\n");
629 }
630
631 void
632 kthread_shutdown(void *arg, int howto)
633 {
634         struct thread *td;
635         char procname[MAXCOMLEN + 1];
636         int error;
637
638         if (panicstr)
639                 return;
640
641         td = (struct thread *)arg;
642         strlcpy(procname, td->td_name, sizeof(procname));
643         printf("Waiting (max %d seconds) for system thread `%s' to stop...",
644             kproc_shutdown_wait, procname);
645         error = kthread_suspend(td, kproc_shutdown_wait * hz);
646
647         if (error == EWOULDBLOCK)
648                 printf("timed out\n");
649         else
650                 printf("done\n");
651 }
652
653 /* Registration of dumpers */
654 int
655 set_dumper(struct dumperinfo *di)
656 {
657
658         if (di == NULL) {
659                 bzero(&dumper, sizeof dumper);
660                 return (0);
661         }
662         if (dumper.dumper != NULL)
663                 return (EBUSY);
664         dumper = *di;
665         return (0);
666 }
667
668 /* Call dumper with bounds checking. */
669 int
670 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
671     off_t offset, size_t length)
672 {
673
674         if (length != 0 && (offset < di->mediaoffset ||
675             offset - di->mediaoffset + length > di->mediasize)) {
676                 printf("Attempt to write outside dump device boundaries.\n");
677                 return (ENXIO);
678         }
679         return (di->dumper(di->priv, virtual, physical, offset, length));
680 }
681
682 void
683 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver,
684     uint64_t dumplen, uint32_t blksz)
685 {
686
687         bzero(kdh, sizeof(*kdh));
688         strncpy(kdh->magic, magic, sizeof(kdh->magic));
689         strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
690         kdh->version = htod32(KERNELDUMPVERSION);
691         kdh->architectureversion = htod32(archver);
692         kdh->dumplength = htod64(dumplen);
693         kdh->dumptime = htod64(time_second);
694         kdh->blocksize = htod32(blksz);
695         strncpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
696         strncpy(kdh->versionstring, version, sizeof(kdh->versionstring));
697         if (panicstr != NULL)
698                 strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
699         kdh->parity = kerneldump_parity(kdh);
700 }