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/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>
  66 #include <sys/sysctl.h>
  67 #include <sys/sysproto.h>
  68 #include <sys/vnode.h>
  69 #ifdef SW_WATCHDOG
  70 #include <sys/watchdog.h>
  71 #endif
  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, CTLFLAG_RW | CTLFLAG_TUN,
 106         &debugger_on_panic, 0, "Run debugger on kernel panic");
 107 TUNABLE_INT("debug.debugger_on_panic", &debugger_on_panic);
 108
 109 #ifdef KDB_TRACE
 110 static int trace_on_panic = 1;
 111 #else
 112 static int trace_on_panic = 0;
 113 #endif
 114 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW | CTLFLAG_TUN,
 115         &trace_on_panic, 0, "Print stack trace on kernel panic");
 116 TUNABLE_INT("debug.trace_on_panic", &trace_on_panic);
 117 #endif /* KDB */
 118
 119 static int sync_on_panic = 0;
 120 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW | CTLFLAG_TUN,
 121         &sync_on_panic, 0, "Do a sync before rebooting from a panic");
 122 TUNABLE_INT("kern.sync_on_panic", &sync_on_panic);
 123
 124 static int stop_scheduler_on_panic = 1;
 125 SYSCTL_INT(_kern, OID_AUTO, stop_scheduler_on_panic, CTLFLAG_RW | CTLFLAG_TUN,
 126     &stop_scheduler_on_panic, 0, "stop scheduler upon entering panic");
 127 TUNABLE_INT("kern.stop_scheduler_on_panic", &stop_scheduler_on_panic);
 128
 129 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment");
 130
 131 #ifndef DIAGNOSTIC
 132 static int show_busybufs;
 133 #else
 134 static int show_busybufs = 1;
 135 #endif
 136 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
 137         &show_busybufs, 0, "");
 138
 139 /*
 140  * Variable panicstr contains argument to first call to panic; used as flag
 141  * to indicate that the kernel has already called panic.
 142  */
 143 const char *panicstr;
 144
 145 int stop_scheduler;                     /* system stopped CPUs for panic */
 146 int dumping;                            /* system is dumping */
 147 int rebooting;                          /* system is rebooting */
 148 static struct dumperinfo dumper;        /* our selected dumper */
 149
 150 /* Context information for dump-debuggers. */
 151 static struct pcb dumppcb;              /* Registers. */
 152 static lwpid_t dumptid;                 /* Thread ID. */
 153
 154 static void poweroff_wait(void *, int);
 155 static void shutdown_halt(void *junk, int howto);
 156 static void shutdown_panic(void *junk, int howto);
 157 static void shutdown_reset(void *junk, int howto);
 158
 159 /* register various local shutdown events */
 160 static void
 161 shutdown_conf(void *unused)
 162 {
 163
 164         EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
 165             SHUTDOWN_PRI_FIRST);
 166         EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
 167             SHUTDOWN_PRI_LAST + 100);
 168         EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
 169             SHUTDOWN_PRI_LAST + 100);
 170         EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
 171             SHUTDOWN_PRI_LAST + 200);
 172 }
 173
 174 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
 175
 176 /*
 177  * The system call that results in a reboot.
 178  */
 179 /* ARGSUSED */
 180 int
 181 sys_reboot(struct thread *td, struct reboot_args *uap)
 182 {
 183         int error;
 184
 185         error = 0;
 186 #ifdef MAC
 187         error = mac_system_check_reboot(td->td_ucred, uap->opt);
 188 #endif
 189         if (error == 0)
 190                 error = priv_check(td, PRIV_REBOOT);
 191         if (error == 0) {
 192                 mtx_lock(&Giant);
 193                 kern_reboot(uap->opt);
 194                 mtx_unlock(&Giant);
 195         }
 196         return (error);
 197 }
 198
 199 /*
 200  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
 201  */
 202 static int shutdown_howto = 0;
 203
 204 void
 205 shutdown_nice(int howto)
 206 {
 207
 208         shutdown_howto = howto;
 209
 210         /* Send a signal to init(8) and have it shutdown the world */
 211         if (initproc != NULL) {
 212                 PROC_LOCK(initproc);
 213                 kern_psignal(initproc, SIGINT);
 214                 PROC_UNLOCK(initproc);
 215         } else {
 216                 /* No init(8) running, so simply reboot */
 217                 kern_reboot(RB_NOSYNC);
 218         }
 219         return;
 220 }
 221 static int      waittime = -1;
 222
 223 static void
 224 print_uptime(void)
 225 {
 226         int f;
 227         struct timespec ts;
 228
 229         getnanouptime(&ts);
 230         printf("Uptime: ");
 231         f = 0;
 232         if (ts.tv_sec >= 86400) {
 233                 printf("%ldd", (long)ts.tv_sec / 86400);
 234                 ts.tv_sec %= 86400;
 235                 f = 1;
 236         }
 237         if (f || ts.tv_sec >= 3600) {
 238                 printf("%ldh", (long)ts.tv_sec / 3600);
 239                 ts.tv_sec %= 3600;
 240                 f = 1;
 241         }
 242         if (f || ts.tv_sec >= 60) {
 243                 printf("%ldm", (long)ts.tv_sec / 60);
 244                 ts.tv_sec %= 60;
 245                 f = 1;
 246         }
 247         printf("%lds\n", (long)ts.tv_sec);
 248 }
 249
 250 int
 251 doadump(boolean_t textdump)
 252 {
 253         boolean_t coredump;
 254
 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                 dumpsys(&dumper);
 273
 274         dumping--;
 275         return (0);
 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
 296 #if defined(SMP)
 297         /*
 298          * Bind us to CPU 0 so that all shutdown code runs there.  Some
 299          * systems don't shutdown properly (i.e., ACPI power off) if we
 300          * run on another processor.
 301          */
 302         if (!SCHEDULER_STOPPED()) {
 303                 thread_lock(curthread);
 304                 sched_bind(curthread, 0);
 305                 thread_unlock(curthread);
 306                 KASSERT(PCPU_GET(cpuid) == 0, ("boot: not running on cpu 0"));
 307         }
 308 #endif
 309         /* We're in the process of rebooting. */
 310         rebooting = 1;
 311
 312         /* collect extra flags that shutdown_nice might have set */
 313         howto |= shutdown_howto;
 314
 315         /* We are out of the debugger now. */
 316         kdb_active = 0;
 317
 318         /*
 319          * Do any callouts that should be done BEFORE syncing the filesystems.
 320          */
 321         EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
 322
 323         /*
 324          * Now sync filesystems
 325          */
 326         if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
 327                 register struct buf *bp;
 328                 int iter, nbusy, pbusy;
 329 #ifndef PREEMPTION
 330                 int subiter;
 331 #endif
 332
 333                 waittime = 0;
 334
 335 #ifdef SW_WATCHDOG
 336                 wdog_kern_pat(WD_LASTVAL);
 337 #endif
 338                 sys_sync(curthread, NULL);
 339
 340                 /*
 341                  * With soft updates, some buffers that are
 342                  * written will be remarked as dirty until other
 343                  * buffers are written.
 344                  */
 345                 for (iter = pbusy = 0; iter < 20; iter++) {
 346                         nbusy = 0;
 347                         for (bp = &buf[nbuf]; --bp >= buf; )
 348                                 if (isbufbusy(bp))
 349                                         nbusy++;
 350                         if (nbusy == 0) {
 351                                 if (first_buf_printf)
 352                                         printf("All buffers synced.");
 353                                 break;
 354                         }
 355                         if (first_buf_printf) {
 356                                 printf("Syncing disks, buffers remaining... ");
 357                                 first_buf_printf = 0;
 358                         }
 359                         printf("%d ", nbusy);
 360                         if (nbusy < pbusy)
 361                                 iter = 0;
 362                         pbusy = nbusy;
 363 #ifdef SW_WATCHDOG
 364                         wdog_kern_pat(WD_LASTVAL);
 365 #endif
 366                         sys_sync(curthread, NULL);
 367
 368 #ifdef PREEMPTION
 369                         /*
 370                          * Drop Giant and spin for a while to allow
 371                          * interrupt threads to run.
 372                          */
 373                         DROP_GIANT();
 374                         DELAY(50000 * iter);
 375                         PICKUP_GIANT();
 376 #else
 377                         /*
 378                          * Drop Giant and context switch several times to
 379                          * allow interrupt threads to run.
 380                          */
 381                         DROP_GIANT();
 382                         for (subiter = 0; subiter < 50 * iter; subiter++) {
 383                                 thread_lock(curthread);
 384                                 mi_switch(SW_VOL, NULL);
 385                                 thread_unlock(curthread);
 386                                 DELAY(1000);
 387                         }
 388                         PICKUP_GIANT();
 389 #endif
 390                 }
 391                 printf("\n");
 392                 /*
 393                  * Count only busy local buffers to prevent forcing
 394                  * a fsck if we're just a client of a wedged NFS server
 395                  */
 396                 nbusy = 0;
 397                 for (bp = &buf[nbuf]; --bp >= buf; ) {
 398                         if (isbufbusy(bp)) {
 399 #if 0
 400 /* XXX: This is bogus.  We should probably have a BO_REMOTE flag instead */
 401                                 if (bp->b_dev == NULL) {
 402                                         TAILQ_REMOVE(&mountlist,
 403                                             bp->b_vp->v_mount, mnt_list);
 404                                         continue;
 405                                 }
 406 #endif
 407                                 nbusy++;
 408                                 if (show_busybufs > 0) {
 409                                         printf(
 410             "%d: buf:%p, vnode:%p, flags:%0x, blkno:%jd, lblkno:%jd, buflock:",
 411                                             nbusy, bp, bp->b_vp, bp->b_flags,
 412                                             (intmax_t)bp->b_blkno,
 413                                             (intmax_t)bp->b_lblkno);
 414                                         BUF_LOCKPRINTINFO(bp);
 415                                         if (show_busybufs > 1)
 416                                                 vn_printf(bp->b_vp,
 417                                                     "vnode content: ");
 418                                 }
 419                         }
 420                 }
 421                 if (nbusy) {
 422                         /*
 423                          * Failed to sync all blocks. Indicate this and don't
 424                          * unmount filesystems (thus forcing an fsck on reboot).
 425                          */
 426                         printf("Giving up on %d buffers\n", nbusy);
 427                         DELAY(5000000); /* 5 seconds */
 428                 } else {
 429                         if (!first_buf_printf)
 430                                 printf("Final sync complete\n");
 431                         /*
 432                          * Unmount filesystems
 433                          */
 434                         if (panicstr == 0)
 435                                 vfs_unmountall();
 436                 }
 437                 swapoff_all();
 438                 DELAY(100000);          /* wait for console output to finish */
 439         }
 440
 441         print_uptime();
 442
 443         cngrab();
 444
 445         /*
 446          * Ok, now do things that assume all filesystem activity has
 447          * been completed.
 448          */
 449         EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
 450
 451         if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
 452                 doadump(TRUE);
 453
 454         /* Now that we're going to really halt the system... */
 455         EVENTHANDLER_INVOKE(shutdown_final, howto);
 456
 457         for(;;) ;       /* safety against shutdown_reset not working */
 458         /* NOTREACHED */
 459 }
 460
 461 /*
 462  * If the shutdown was a clean halt, behave accordingly.
 463  */
 464 static void
 465 shutdown_halt(void *junk, int howto)
 466 {
 467
 468         if (howto & RB_HALT) {
 469                 printf("\n");
 470                 printf("The operating system has halted.\n");
 471                 printf("Please press any key to reboot.\n\n");
 472                 switch (cngetc()) {
 473                 case -1:                /* No console, just die */
 474                         cpu_halt();
 475                         /* NOTREACHED */
 476                 default:
 477                         howto &= ~RB_HALT;
 478                         break;
 479                 }
 480         }
 481 }
 482
 483 /*
 484  * Check to see if the system paniced, pause and then reboot
 485  * according to the specified delay.
 486  */
 487 static void
 488 shutdown_panic(void *junk, int howto)
 489 {
 490         int loop;
 491
 492         if (howto & RB_DUMP) {
 493                 if (PANIC_REBOOT_WAIT_TIME != 0) {
 494                         if (PANIC_REBOOT_WAIT_TIME != -1) {
 495                                 printf("Automatic reboot in %d seconds - "
 496                                        "press a key on the console to abort\n",
 497                                         PANIC_REBOOT_WAIT_TIME);
 498                                 for (loop = PANIC_REBOOT_WAIT_TIME * 10;
 499                                      loop > 0; --loop) {
 500                                         DELAY(1000 * 100); /* 1/10th second */
 501                                         /* Did user type a key? */
 502                                         if (cncheckc() != -1)
 503                                                 break;
 504                                 }
 505                                 if (!loop)
 506                                         return;
 507                         }
 508                 } else { /* zero time specified - reboot NOW */
 509                         return;
 510                 }
 511                 printf("--> Press a key on the console to reboot,\n");
 512                 printf("--> or switch off the system now.\n");
 513                 cngetc();
 514         }
 515 }
 516
 517 /*
 518  * Everything done, now reset
 519  */
 520 static void
 521 shutdown_reset(void *junk, int howto)
 522 {
 523
 524         printf("Rebooting...\n");
 525         DELAY(1000000); /* wait 1 sec for printf's to complete and be read */
 526
 527         /*
 528          * Acquiring smp_ipi_mtx here has a double effect:
 529          * - it disables interrupts avoiding CPU0 preemption
 530          *   by fast handlers (thus deadlocking  against other CPUs)
 531          * - it avoids deadlocks against smp_rendezvous() or, more
 532          *   generally, threads busy-waiting, with this spinlock held,
 533          *   and waiting for responses by threads on other CPUs
 534          *   (ie. smp_tlb_shootdown()).
 535          *
 536          * For the !SMP case it just needs to handle the former problem.
 537          */
 538 #ifdef SMP
 539         mtx_lock_spin(&smp_ipi_mtx);
 540 #else
 541         spinlock_enter();
 542 #endif
 543
 544         /* cpu_boot(howto); */ /* doesn't do anything at the moment */
 545         cpu_reset();
 546         /* NOTREACHED */ /* assuming reset worked */
 547 }
 548
 549 /*
 550  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
 551  * and then reboots.  If we are called twice, then we avoid trying to sync
 552  * the disks as this often leads to recursive panics.
 553  */
 554 void
 555 panic(const char *fmt, ...)
 556 {
 557 #ifdef SMP
 558         static volatile u_int panic_cpu = NOCPU;
 559         cpuset_t other_cpus;
 560 #endif
 561         struct thread *td = curthread;
 562         int bootopt, newpanic;
 563         va_list ap;
 564         static char buf[256];
 565
 566         if (stop_scheduler_on_panic)
 567                 spinlock_enter();
 568         else
 569                 critical_enter();
 570
 571 #ifdef SMP
 572         /*
 573          * We don't want multiple CPU's to panic at the same time, so we
 574          * use panic_cpu as a simple spinlock.  We have to keep checking
 575          * panic_cpu if we are spinning in case the panic on the first
 576          * CPU is canceled.
 577          */
 578         if (panic_cpu != PCPU_GET(cpuid))
 579                 while (atomic_cmpset_int(&panic_cpu, NOCPU,
 580                     PCPU_GET(cpuid)) == 0)
 581                         while (panic_cpu != NOCPU)
 582                                 ; /* nothing */
 583
 584         if (stop_scheduler_on_panic) {
 585                 if (panicstr == NULL && !kdb_active) {
 586                         other_cpus = all_cpus;
 587                         CPU_CLR(PCPU_GET(cpuid), &other_cpus);
 588                         stop_cpus_hard(other_cpus);
 589                 }
 590
 591                 /*
 592                  * We set stop_scheduler here and not in the block above,
 593                  * because we want to ensure that if panic has been called and
 594                  * stop_scheduler_on_panic is true, then stop_scheduler will
 595                  * always be set.  Even if panic has been entered from kdb.
 596                  */
 597                 stop_scheduler = 1;
 598         }
 599 #endif
 600
 601         bootopt = RB_AUTOBOOT;
 602         newpanic = 0;
 603         if (panicstr)
 604                 bootopt |= RB_NOSYNC;
 605         else {
 606                 bootopt |= RB_DUMP;
 607                 panicstr = fmt;
 608                 newpanic = 1;
 609         }
 610
 611         va_start(ap, fmt);
 612         if (newpanic) {
 613                 (void)vsnprintf(buf, sizeof(buf), fmt, ap);
 614                 panicstr = buf;
 615                 cngrab();
 616                 printf("panic: %s\n", buf);
 617         } else {
 618                 printf("panic: ");
 619                 vprintf(fmt, ap);
 620                 printf("\n");
 621         }
 622         va_end(ap);
 623 #ifdef SMP
 624         printf("cpuid = %d\n", PCPU_GET(cpuid));
 625 #endif
 626
 627 #ifdef KDB
 628         if (newpanic && trace_on_panic)
 629                 kdb_backtrace();
 630         if (debugger_on_panic)
 631                 kdb_enter(KDB_WHY_PANIC, "panic");
 632 #endif
 633         /*thread_lock(td); */
 634         td->td_flags |= TDF_INPANIC;
 635         /* thread_unlock(td); */
 636         if (!sync_on_panic)
 637                 bootopt |= RB_NOSYNC;
 638         if (!stop_scheduler_on_panic)
 639                 critical_exit();
 640         kern_reboot(bootopt);
 641 }
 642
 643 /*
 644  * Support for poweroff delay.
 645  *
 646  * Please note that setting this delay too short might power off your machine
 647  * before the write cache on your hard disk has been flushed, leading to
 648  * soft-updates inconsistencies.
 649  */
 650 #ifndef POWEROFF_DELAY
 651 # define POWEROFF_DELAY 5000
 652 #endif
 653 static int poweroff_delay = POWEROFF_DELAY;
 654
 655 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
 656         &poweroff_delay, 0, "");
 657
 658 static void
 659 poweroff_wait(void *junk, int howto)
 660 {
 661
 662         if (!(howto & RB_POWEROFF) || poweroff_delay <= 0)
 663                 return;
 664         DELAY(poweroff_delay * 1000);
 665 }
 666
 667 /*
 668  * Some system processes (e.g. syncer) need to be stopped at appropriate
 669  * points in their main loops prior to a system shutdown, so that they
 670  * won't interfere with the shutdown process (e.g. by holding a disk buf
 671  * to cause sync to fail).  For each of these system processes, register
 672  * shutdown_kproc() as a handler for one of shutdown events.
 673  */
 674 static int kproc_shutdown_wait = 60;
 675 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
 676     &kproc_shutdown_wait, 0, "");
 677
 678 void
 679 kproc_shutdown(void *arg, int howto)
 680 {
 681         struct proc *p;
 682         int error;
 683
 684         if (panicstr)
 685                 return;
 686
 687         p = (struct proc *)arg;
 688         printf("Waiting (max %d seconds) for system process `%s' to stop...",
 689             kproc_shutdown_wait, p->p_comm);
 690         error = kproc_suspend(p, kproc_shutdown_wait * hz);
 691
 692         if (error == EWOULDBLOCK)
 693                 printf("timed out\n");
 694         else
 695                 printf("done\n");
 696 }
 697
 698 void
 699 kthread_shutdown(void *arg, int howto)
 700 {
 701         struct thread *td;
 702         int error;
 703
 704         if (panicstr)
 705                 return;
 706
 707         td = (struct thread *)arg;
 708         printf("Waiting (max %d seconds) for system thread `%s' to stop...",
 709             kproc_shutdown_wait, td->td_name);
 710         error = kthread_suspend(td, kproc_shutdown_wait * hz);
 711
 712         if (error == EWOULDBLOCK)
 713                 printf("timed out\n");
 714         else
 715                 printf("done\n");
 716 }
 717
 718 /* Registration of dumpers */
 719 int
 720 set_dumper(struct dumperinfo *di)
 721 {
 722
 723         if (di == NULL) {
 724                 bzero(&dumper, sizeof dumper);
 725                 return (0);
 726         }
 727         if (dumper.dumper != NULL)
 728                 return (EBUSY);
 729         dumper = *di;
 730         return (0);
 731 }
 732
 733 /* Call dumper with bounds checking. */
 734 int
 735 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
 736     off_t offset, size_t length)
 737 {
 738
 739         if (length != 0 && (offset < di->mediaoffset ||
 740             offset - di->mediaoffset + length > di->mediasize)) {
 741                 printf("Attempt to write outside dump device boundaries.\n"
 742             "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
 743                     (intmax_t)offset, (intmax_t)di->mediaoffset,
 744                     (uintmax_t)length, (intmax_t)di->mediasize);
 745                 return (ENOSPC);
 746         }
 747         return (di->dumper(di->priv, virtual, physical, offset, length));
 748 }
 749
 750 void
 751 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver,
 752     uint64_t dumplen, uint32_t blksz)
 753 {
 754
 755         bzero(kdh, sizeof(*kdh));
 756         strncpy(kdh->magic, magic, sizeof(kdh->magic));
 757         strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
 758         kdh->version = htod32(KERNELDUMPVERSION);
 759         kdh->architectureversion = htod32(archver);
 760         kdh->dumplength = htod64(dumplen);
 761         kdh->dumptime = htod64(time_second);
 762         kdh->blocksize = htod32(blksz);
 763         strncpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
 764         strncpy(kdh->versionstring, version, sizeof(kdh->versionstring));
 765         if (panicstr != NULL)
 766                 strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
 767         kdh->parity = kerneldump_parity(kdh);
 768 }