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 #include <sys/watchdog.h>
  70
  71 #include <ddb/ddb.h>
  72
  73 #include <machine/cpu.h>
  74 #include <machine/pcb.h>
  75 #include <machine/smp.h>
  76
  77 #include <security/mac/mac_framework.h>
  78
  79 #include <vm/vm.h>
  80 #include <vm/vm_object.h>
  81 #include <vm/vm_page.h>
  82 #include <vm/vm_pager.h>
  83 #include <vm/swap_pager.h>
  84
  85 #include <sys/signalvar.h>
  86
  87 #ifndef PANIC_REBOOT_WAIT_TIME
  88 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
  89 #endif
  90
  91 /*
  92  * Note that stdarg.h and the ANSI style va_start macro is used for both
  93  * ANSI and traditional C compilers.
  94  */
  95 #include <machine/stdarg.h>
  96
  97 #ifdef KDB
  98 #ifdef KDB_UNATTENDED
  99 int debugger_on_panic = 0;
 100 #else
 101 int debugger_on_panic = 1;
 102 #endif
 103 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic,
 104     CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_TUN,
 105     &debugger_on_panic, 0, "Run debugger on kernel panic");
 106 TUNABLE_INT("debug.debugger_on_panic", &debugger_on_panic);
 107
 108 #ifdef KDB_TRACE
 109 static int trace_on_panic = 1;
 110 #else
 111 static int trace_on_panic = 0;
 112 #endif
 113 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
 114     CTLFLAG_RW | CTLFLAG_SECURE | 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 dumping;                            /* system is dumping */
 146 int rebooting;                          /* system is rebooting */
 147 static struct dumperinfo dumper;        /* our selected dumper */
 148
 149 /* Context information for dump-debuggers. */
 150 static struct pcb dumppcb;              /* Registers. */
 151 static lwpid_t dumptid;                 /* Thread ID. */
 152
 153 static void poweroff_wait(void *, int);
 154 static void shutdown_halt(void *junk, int howto);
 155 static void shutdown_panic(void *junk, int howto);
 156 static void shutdown_reset(void *junk, int howto);
 157
 158 /* register various local shutdown events */
 159 static void
 160 shutdown_conf(void *unused)
 161 {
 162
 163         EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
 164             SHUTDOWN_PRI_FIRST);
 165         EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
 166             SHUTDOWN_PRI_LAST + 100);
 167         EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
 168             SHUTDOWN_PRI_LAST + 100);
 169         EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
 170             SHUTDOWN_PRI_LAST + 200);
 171 }
 172
 173 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
 174
 175 /*
 176  * The system call that results in a reboot.
 177  */
 178 /* ARGSUSED */
 179 int
 180 sys_reboot(struct thread *td, struct reboot_args *uap)
 181 {
 182         int error;
 183
 184         error = 0;
 185 #ifdef MAC
 186         error = mac_system_check_reboot(td->td_ucred, uap->opt);
 187 #endif
 188         if (error == 0)
 189                 error = priv_check(td, PRIV_REBOOT);
 190         if (error == 0) {
 191                 mtx_lock(&Giant);
 192                 kern_reboot(uap->opt);
 193                 mtx_unlock(&Giant);
 194         }
 195         return (error);
 196 }
 197
 198 /*
 199  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
 200  */
 201 static int shutdown_howto = 0;
 202
 203 void
 204 shutdown_nice(int howto)
 205 {
 206
 207         shutdown_howto = howto;
 208
 209         /* Send a signal to init(8) and have it shutdown the world */
 210         if (initproc != NULL) {
 211                 PROC_LOCK(initproc);
 212                 kern_psignal(initproc, SIGINT);
 213                 PROC_UNLOCK(initproc);
 214         } else {
 215                 /* No init(8) running, so simply reboot */
 216                 kern_reboot(RB_NOSYNC);
 217         }
 218         return;
 219 }
 220 static int      waittime = -1;
 221
 222 static void
 223 print_uptime(void)
 224 {
 225         int f;
 226         struct timespec ts;
 227
 228         getnanouptime(&ts);
 229         printf("Uptime: ");
 230         f = 0;
 231         if (ts.tv_sec >= 86400) {
 232                 printf("%ldd", (long)ts.tv_sec / 86400);
 233                 ts.tv_sec %= 86400;
 234                 f = 1;
 235         }
 236         if (f || ts.tv_sec >= 3600) {
 237                 printf("%ldh", (long)ts.tv_sec / 3600);
 238                 ts.tv_sec %= 3600;
 239                 f = 1;
 240         }
 241         if (f || ts.tv_sec >= 60) {
 242                 printf("%ldm", (long)ts.tv_sec / 60);
 243                 ts.tv_sec %= 60;
 244                 f = 1;
 245         }
 246         printf("%lds\n", (long)ts.tv_sec);
 247 }
 248
 249 int
 250 doadump(boolean_t textdump)
 251 {
 252         boolean_t coredump;
 253
 254         if (dumping)
 255                 return (EBUSY);
 256         if (dumper.dumper == NULL)
 257                 return (ENXIO);
 258
 259         savectx(&dumppcb);
 260         dumptid = curthread->td_tid;
 261         dumping++;
 262
 263         coredump = TRUE;
 264 #ifdef DDB
 265         if (textdump && textdump_pending) {
 266                 coredump = FALSE;
 267                 textdump_dumpsys(&dumper);
 268         }
 269 #endif
 270         if (coredump)
 271                 dumpsys(&dumper);
 272
 273         dumping--;
 274         return (0);
 275 }
 276
 277 static int
 278 isbufbusy(struct buf *bp)
 279 {
 280         if (((bp->b_flags & (B_INVAL | B_PERSISTENT)) == 0 &&
 281             BUF_ISLOCKED(bp)) ||
 282             ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI))
 283                 return (1);
 284         return (0);
 285 }
 286
 287 /*
 288  * Shutdown the system cleanly to prepare for reboot, halt, or power off.
 289  */
 290 void
 291 kern_reboot(int howto)
 292 {
 293         static int first_buf_printf = 1;
 294
 295 #if defined(SMP)
 296         /*
 297          * Bind us to CPU 0 so that all shutdown code runs there.  Some
 298          * systems don't shutdown properly (i.e., ACPI power off) if we
 299          * run on another processor.
 300          */
 301         if (!SCHEDULER_STOPPED()) {
 302                 thread_lock(curthread);
 303                 sched_bind(curthread, 0);
 304                 thread_unlock(curthread);
 305                 KASSERT(PCPU_GET(cpuid) == 0, ("boot: not running on cpu 0"));
 306         }
 307 #endif
 308         /* We're in the process of rebooting. */
 309         rebooting = 1;
 310
 311         /* collect extra flags that shutdown_nice might have set */
 312         howto |= shutdown_howto;
 313
 314         /* We are out of the debugger now. */
 315         kdb_active = 0;
 316
 317         /*
 318          * Do any callouts that should be done BEFORE syncing the filesystems.
 319          */
 320         EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
 321
 322         /*
 323          * Now sync filesystems
 324          */
 325         if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
 326                 register struct buf *bp;
 327                 int iter, nbusy, pbusy;
 328 #ifndef PREEMPTION
 329                 int subiter;
 330 #endif
 331
 332                 waittime = 0;
 333
 334                 wdog_kern_pat(WD_LASTVAL);
 335                 sys_sync(curthread, NULL);
 336
 337                 /*
 338                  * With soft updates, some buffers that are
 339                  * written will be remarked as dirty until other
 340                  * buffers are written.
 341                  */
 342                 for (iter = pbusy = 0; iter < 20; iter++) {
 343                         nbusy = 0;
 344                         for (bp = &buf[nbuf]; --bp >= buf; )
 345                                 if (isbufbusy(bp))
 346                                         nbusy++;
 347                         if (nbusy == 0) {
 348                                 if (first_buf_printf)
 349                                         printf("All buffers synced.");
 350                                 break;
 351                         }
 352                         if (first_buf_printf) {
 353                                 printf("Syncing disks, buffers remaining... ");
 354                                 first_buf_printf = 0;
 355                         }
 356                         printf("%d ", nbusy);
 357                         if (nbusy < pbusy)
 358                                 iter = 0;
 359                         pbusy = nbusy;
 360
 361                         wdog_kern_pat(WD_LASTVAL);
 362                         sys_sync(curthread, NULL);
 363
 364 #ifdef PREEMPTION
 365                         /*
 366                          * Drop Giant and spin for a while to allow
 367                          * interrupt threads to run.
 368                          */
 369                         DROP_GIANT();
 370                         DELAY(50000 * iter);
 371                         PICKUP_GIANT();
 372 #else
 373                         /*
 374                          * Drop Giant and context switch several times to
 375                          * allow interrupt threads to run.
 376                          */
 377                         DROP_GIANT();
 378                         for (subiter = 0; subiter < 50 * iter; subiter++) {
 379                                 thread_lock(curthread);
 380                                 mi_switch(SW_VOL, NULL);
 381                                 thread_unlock(curthread);
 382                                 DELAY(1000);
 383                         }
 384                         PICKUP_GIANT();
 385 #endif
 386                 }
 387                 printf("\n");
 388                 /*
 389                  * Count only busy local buffers to prevent forcing
 390                  * a fsck if we're just a client of a wedged NFS server
 391                  */
 392                 nbusy = 0;
 393                 for (bp = &buf[nbuf]; --bp >= buf; ) {
 394                         if (isbufbusy(bp)) {
 395 #if 0
 396 /* XXX: This is bogus.  We should probably have a BO_REMOTE flag instead */
 397                                 if (bp->b_dev == NULL) {
 398                                         TAILQ_REMOVE(&mountlist,
 399                                             bp->b_vp->v_mount, mnt_list);
 400                                         continue;
 401                                 }
 402 #endif
 403                                 nbusy++;
 404                                 if (show_busybufs > 0) {
 405                                         printf(
 406             "%d: buf:%p, vnode:%p, flags:%0x, blkno:%jd, lblkno:%jd, buflock:",
 407                                             nbusy, bp, bp->b_vp, bp->b_flags,
 408                                             (intmax_t)bp->b_blkno,
 409                                             (intmax_t)bp->b_lblkno);
 410                                         BUF_LOCKPRINTINFO(bp);
 411                                         if (show_busybufs > 1)
 412                                                 vn_printf(bp->b_vp,
 413                                                     "vnode content: ");
 414                                 }
 415                         }
 416                 }
 417                 if (nbusy) {
 418                         /*
 419                          * Failed to sync all blocks. Indicate this and don't
 420                          * unmount filesystems (thus forcing an fsck on reboot).
 421                          */
 422                         printf("Giving up on %d buffers\n", nbusy);
 423                         DELAY(5000000); /* 5 seconds */
 424                 } else {
 425                         if (!first_buf_printf)
 426                                 printf("Final sync complete\n");
 427                         /*
 428                          * Unmount filesystems
 429                          */
 430                         if (panicstr == 0)
 431                                 vfs_unmountall();
 432                 }
 433                 swapoff_all();
 434                 DELAY(100000);          /* wait for console output to finish */
 435         }
 436
 437         print_uptime();
 438
 439         cngrab();
 440
 441         /*
 442          * Ok, now do things that assume all filesystem activity has
 443          * been completed.
 444          */
 445         EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
 446
 447         if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
 448                 doadump(TRUE);
 449
 450         /* Now that we're going to really halt the system... */
 451         EVENTHANDLER_INVOKE(shutdown_final, howto);
 452
 453         for(;;) ;       /* safety against shutdown_reset not working */
 454         /* NOTREACHED */
 455 }
 456
 457 /*
 458  * If the shutdown was a clean halt, behave accordingly.
 459  */
 460 static void
 461 shutdown_halt(void *junk, int howto)
 462 {
 463
 464         if (howto & RB_HALT) {
 465                 printf("\n");
 466                 printf("The operating system has halted.\n");
 467                 printf("Please press any key to reboot.\n\n");
 468                 switch (cngetc()) {
 469                 case -1:                /* No console, just die */
 470                         cpu_halt();
 471                         /* NOTREACHED */
 472                 default:
 473                         howto &= ~RB_HALT;
 474                         break;
 475                 }
 476         }
 477 }
 478
 479 /*
 480  * Check to see if the system paniced, pause and then reboot
 481  * according to the specified delay.
 482  */
 483 static void
 484 shutdown_panic(void *junk, int howto)
 485 {
 486         int loop;
 487
 488         if (howto & RB_DUMP) {
 489                 if (PANIC_REBOOT_WAIT_TIME != 0) {
 490                         if (PANIC_REBOOT_WAIT_TIME != -1) {
 491                                 printf("Automatic reboot in %d seconds - "
 492                                        "press a key on the console to abort\n",
 493                                         PANIC_REBOOT_WAIT_TIME);
 494                                 for (loop = PANIC_REBOOT_WAIT_TIME * 10;
 495                                      loop > 0; --loop) {
 496                                         DELAY(1000 * 100); /* 1/10th second */
 497                                         /* Did user type a key? */
 498                                         if (cncheckc() != -1)
 499                                                 break;
 500                                 }
 501                                 if (!loop)
 502                                         return;
 503                         }
 504                 } else { /* zero time specified - reboot NOW */
 505                         return;
 506                 }
 507                 printf("--> Press a key on the console to reboot,\n");
 508                 printf("--> or switch off the system now.\n");
 509                 cngetc();
 510         }
 511 }
 512
 513 /*
 514  * Everything done, now reset
 515  */
 516 static void
 517 shutdown_reset(void *junk, int howto)
 518 {
 519
 520         printf("Rebooting...\n");
 521         DELAY(1000000); /* wait 1 sec for printf's to complete and be read */
 522
 523         /*
 524          * Acquiring smp_ipi_mtx here has a double effect:
 525          * - it disables interrupts avoiding CPU0 preemption
 526          *   by fast handlers (thus deadlocking  against other CPUs)
 527          * - it avoids deadlocks against smp_rendezvous() or, more
 528          *   generally, threads busy-waiting, with this spinlock held,
 529          *   and waiting for responses by threads on other CPUs
 530          *   (ie. smp_tlb_shootdown()).
 531          *
 532          * For the !SMP case it just needs to handle the former problem.
 533          */
 534 #ifdef SMP
 535         mtx_lock_spin(&smp_ipi_mtx);
 536 #else
 537         spinlock_enter();
 538 #endif
 539
 540         /* cpu_boot(howto); */ /* doesn't do anything at the moment */
 541         cpu_reset();
 542         /* NOTREACHED */ /* assuming reset worked */
 543 }
 544
 545 /*
 546  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
 547  * and then reboots.  If we are called twice, then we avoid trying to sync
 548  * the disks as this often leads to recursive panics.
 549  */
 550 void
 551 panic(const char *fmt, ...)
 552 {
 553 #ifdef SMP
 554         static volatile u_int panic_cpu = NOCPU;
 555         cpuset_t other_cpus;
 556 #endif
 557         struct thread *td = curthread;
 558         int bootopt, newpanic;
 559         va_list ap;
 560         static char buf[256];
 561
 562         if (stop_scheduler_on_panic)
 563                 spinlock_enter();
 564         else
 565                 critical_enter();
 566
 567 #ifdef SMP
 568         /*
 569          * We don't want multiple CPU's to panic at the same time, so we
 570          * use panic_cpu as a simple spinlock.  We have to keep checking
 571          * panic_cpu if we are spinning in case the panic on the first
 572          * CPU is canceled.
 573          */
 574         if (panic_cpu != PCPU_GET(cpuid))
 575                 while (atomic_cmpset_int(&panic_cpu, NOCPU,
 576                     PCPU_GET(cpuid)) == 0)
 577                         while (panic_cpu != NOCPU)
 578                                 ; /* nothing */
 579
 580         if (stop_scheduler_on_panic) {
 581                 if (panicstr == NULL && !kdb_active) {
 582                         other_cpus = all_cpus;
 583                         CPU_CLR(PCPU_GET(cpuid), &other_cpus);
 584                         stop_cpus_hard(other_cpus);
 585                 }
 586
 587                 /*
 588                  * We set stop_scheduler here and not in the block above,
 589                  * because we want to ensure that if panic has been called and
 590                  * stop_scheduler_on_panic is true, then stop_scheduler will
 591                  * always be set.  Even if panic has been entered from kdb.
 592                  */
 593                 td->td_stopsched = 1;
 594         }
 595 #endif
 596
 597         bootopt = RB_AUTOBOOT;
 598         newpanic = 0;
 599         if (panicstr)
 600                 bootopt |= RB_NOSYNC;
 601         else {
 602                 bootopt |= RB_DUMP;
 603                 panicstr = fmt;
 604                 newpanic = 1;
 605         }
 606
 607         va_start(ap, fmt);
 608         if (newpanic) {
 609                 (void)vsnprintf(buf, sizeof(buf), fmt, ap);
 610                 panicstr = buf;
 611                 cngrab();
 612                 printf("panic: %s\n", buf);
 613         } else {
 614                 printf("panic: ");
 615                 vprintf(fmt, ap);
 616                 printf("\n");
 617         }
 618         va_end(ap);
 619 #ifdef SMP
 620         printf("cpuid = %d\n", PCPU_GET(cpuid));
 621 #endif
 622
 623 #ifdef KDB
 624         if (newpanic && trace_on_panic)
 625                 kdb_backtrace();
 626         if (debugger_on_panic)
 627                 kdb_enter(KDB_WHY_PANIC, "panic");
 628 #endif
 629         /*thread_lock(td); */
 630         td->td_flags |= TDF_INPANIC;
 631         /* thread_unlock(td); */
 632         if (!sync_on_panic)
 633                 bootopt |= RB_NOSYNC;
 634         if (!stop_scheduler_on_panic)
 635                 critical_exit();
 636         kern_reboot(bootopt);
 637 }
 638
 639 /*
 640  * Support for poweroff delay.
 641  *
 642  * Please note that setting this delay too short might power off your machine
 643  * before the write cache on your hard disk has been flushed, leading to
 644  * soft-updates inconsistencies.
 645  */
 646 #ifndef POWEROFF_DELAY
 647 # define POWEROFF_DELAY 5000
 648 #endif
 649 static int poweroff_delay = POWEROFF_DELAY;
 650
 651 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
 652         &poweroff_delay, 0, "");
 653
 654 static void
 655 poweroff_wait(void *junk, int howto)
 656 {
 657
 658         if (!(howto & RB_POWEROFF) || poweroff_delay <= 0)
 659                 return;
 660         DELAY(poweroff_delay * 1000);
 661 }
 662
 663 /*
 664  * Some system processes (e.g. syncer) need to be stopped at appropriate
 665  * points in their main loops prior to a system shutdown, so that they
 666  * won't interfere with the shutdown process (e.g. by holding a disk buf
 667  * to cause sync to fail).  For each of these system processes, register
 668  * shutdown_kproc() as a handler for one of shutdown events.
 669  */
 670 static int kproc_shutdown_wait = 60;
 671 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
 672     &kproc_shutdown_wait, 0, "");
 673
 674 void
 675 kproc_shutdown(void *arg, int howto)
 676 {
 677         struct proc *p;
 678         int error;
 679
 680         if (panicstr)
 681                 return;
 682
 683         p = (struct proc *)arg;
 684         printf("Waiting (max %d seconds) for system process `%s' to stop...",
 685             kproc_shutdown_wait, p->p_comm);
 686         error = kproc_suspend(p, kproc_shutdown_wait * hz);
 687
 688         if (error == EWOULDBLOCK)
 689                 printf("timed out\n");
 690         else
 691                 printf("done\n");
 692 }
 693
 694 void
 695 kthread_shutdown(void *arg, int howto)
 696 {
 697         struct thread *td;
 698         int error;
 699
 700         if (panicstr)
 701                 return;
 702
 703         td = (struct thread *)arg;
 704         printf("Waiting (max %d seconds) for system thread `%s' to stop...",
 705             kproc_shutdown_wait, td->td_name);
 706         error = kthread_suspend(td, kproc_shutdown_wait * hz);
 707
 708         if (error == EWOULDBLOCK)
 709                 printf("timed out\n");
 710         else
 711                 printf("done\n");
 712 }
 713
 714 /* Registration of dumpers */
 715 int
 716 set_dumper(struct dumperinfo *di)
 717 {
 718
 719         if (di == NULL) {
 720                 bzero(&dumper, sizeof dumper);
 721                 return (0);
 722         }
 723         if (dumper.dumper != NULL)
 724                 return (EBUSY);
 725         dumper = *di;
 726         return (0);
 727 }
 728
 729 /* Call dumper with bounds checking. */
 730 int
 731 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
 732     off_t offset, size_t length)
 733 {
 734
 735         if (length != 0 && (offset < di->mediaoffset ||
 736             offset - di->mediaoffset + length > di->mediasize)) {
 737                 printf("Attempt to write outside dump device boundaries.\n"
 738             "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
 739                     (intmax_t)offset, (intmax_t)di->mediaoffset,
 740                     (uintmax_t)length, (intmax_t)di->mediasize);
 741                 return (ENOSPC);
 742         }
 743         return (di->dumper(di->priv, virtual, physical, offset, length));
 744 }
 745
 746 void
 747 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver,
 748     uint64_t dumplen, uint32_t blksz)
 749 {
 750
 751         bzero(kdh, sizeof(*kdh));
 752         strncpy(kdh->magic, magic, sizeof(kdh->magic));
 753         strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
 754         kdh->version = htod32(KERNELDUMPVERSION);
 755         kdh->architectureversion = htod32(archver);
 756         kdh->dumplength = htod64(dumplen);
 757         kdh->dumptime = htod64(time_second);
 758         kdh->blocksize = htod32(blksz);
 759         strncpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
 760         strncpy(kdh->versionstring, version, sizeof(kdh->versionstring));
 761         if (panicstr != NULL)
 762                 strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
 763         kdh->parity = kerneldump_parity(kdh);
 764 }