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 static int panic_reboot_wait_time = PANIC_REBOOT_WAIT_TIME;
  91 SYSCTL_INT(_kern, OID_AUTO, panic_reboot_wait_time, CTLFLAG_RW | CTLFLAG_TUN,
  92     &panic_reboot_wait_time, 0,
  93     "Seconds to wait before rebooting after a panic");
  94 TUNABLE_INT("kern.panic_reboot_wait_time", &panic_reboot_wait_time);
  95
  96 /*
  97  * Note that stdarg.h and the ANSI style va_start macro is used for both
  98  * ANSI and traditional C compilers.
  99  */
 100 #include <machine/stdarg.h>
 101
 102 #ifdef KDB
 103 #ifdef KDB_UNATTENDED
 104 int debugger_on_panic = 0;
 105 #else
 106 int debugger_on_panic = 1;
 107 #endif
 108 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic,
 109     CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_TUN,
 110     &debugger_on_panic, 0, "Run debugger on kernel panic");
 111 TUNABLE_INT("debug.debugger_on_panic", &debugger_on_panic);
 112
 113 #ifdef KDB_TRACE
 114 static int trace_on_panic = 1;
 115 #else
 116 static int trace_on_panic = 0;
 117 #endif
 118 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
 119     CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_TUN,
 120     &trace_on_panic, 0, "Print stack trace on kernel panic");
 121 TUNABLE_INT("debug.trace_on_panic", &trace_on_panic);
 122 #endif /* KDB */
 123
 124 static int sync_on_panic = 0;
 125 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW | CTLFLAG_TUN,
 126         &sync_on_panic, 0, "Do a sync before rebooting from a panic");
 127 TUNABLE_INT("kern.sync_on_panic", &sync_on_panic);
 128
 129 static int stop_scheduler_on_panic = 1;
 130 SYSCTL_INT(_kern, OID_AUTO, stop_scheduler_on_panic, CTLFLAG_RW | CTLFLAG_TUN,
 131     &stop_scheduler_on_panic, 0, "stop scheduler upon entering panic");
 132 TUNABLE_INT("kern.stop_scheduler_on_panic", &stop_scheduler_on_panic);
 133
 134 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0,
 135     "Shutdown environment");
 136
 137 #ifndef DIAGNOSTIC
 138 static int show_busybufs;
 139 #else
 140 static int show_busybufs = 1;
 141 #endif
 142 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
 143         &show_busybufs, 0, "");
 144
 145 /*
 146  * Variable panicstr contains argument to first call to panic; used as flag
 147  * to indicate that the kernel has already called panic.
 148  */
 149 const char *panicstr;
 150
 151 int dumping;                            /* system is dumping */
 152 int rebooting;                          /* system is rebooting */
 153 static struct dumperinfo dumper;        /* our selected dumper */
 154
 155 /* Context information for dump-debuggers. */
 156 static struct pcb dumppcb;              /* Registers. */
 157 lwpid_t dumptid;                        /* Thread ID. */
 158
 159 static void poweroff_wait(void *, int);
 160 static void shutdown_halt(void *junk, int howto);
 161 static void shutdown_panic(void *junk, int howto);
 162 static void shutdown_reset(void *junk, int howto);
 163
 164 /* register various local shutdown events */
 165 static void
 166 shutdown_conf(void *unused)
 167 {
 168
 169         EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
 170             SHUTDOWN_PRI_FIRST);
 171         EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
 172             SHUTDOWN_PRI_LAST + 100);
 173         EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
 174             SHUTDOWN_PRI_LAST + 100);
 175         EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
 176             SHUTDOWN_PRI_LAST + 200);
 177 }
 178
 179 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
 180
 181 /*
 182  * The system call that results in a reboot.
 183  */
 184 /* ARGSUSED */
 185 int
 186 sys_reboot(struct thread *td, struct reboot_args *uap)
 187 {
 188         int error;
 189
 190         error = 0;
 191 #ifdef MAC
 192         error = mac_system_check_reboot(td->td_ucred, uap->opt);
 193 #endif
 194         if (error == 0)
 195                 error = priv_check(td, PRIV_REBOOT);
 196         if (error == 0) {
 197                 mtx_lock(&Giant);
 198                 kern_reboot(uap->opt);
 199                 mtx_unlock(&Giant);
 200         }
 201         return (error);
 202 }
 203
 204 /*
 205  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
 206  */
 207 static int shutdown_howto = 0;
 208
 209 void
 210 shutdown_nice(int howto)
 211 {
 212
 213         shutdown_howto = howto;
 214
 215         /* Send a signal to init(8) and have it shutdown the world */
 216         if (initproc != NULL) {
 217                 PROC_LOCK(initproc);
 218                 kern_psignal(initproc, SIGINT);
 219                 PROC_UNLOCK(initproc);
 220         } else {
 221                 /* No init(8) running, so simply reboot */
 222                 kern_reboot(RB_NOSYNC);
 223         }
 224         return;
 225 }
 226 static int      waittime = -1;
 227
 228 static void
 229 print_uptime(void)
 230 {
 231         int f;
 232         struct timespec ts;
 233
 234         getnanouptime(&ts);
 235         printf("Uptime: ");
 236         f = 0;
 237         if (ts.tv_sec >= 86400) {
 238                 printf("%ldd", (long)ts.tv_sec / 86400);
 239                 ts.tv_sec %= 86400;
 240                 f = 1;
 241         }
 242         if (f || ts.tv_sec >= 3600) {
 243                 printf("%ldh", (long)ts.tv_sec / 3600);
 244                 ts.tv_sec %= 3600;
 245                 f = 1;
 246         }
 247         if (f || ts.tv_sec >= 60) {
 248                 printf("%ldm", (long)ts.tv_sec / 60);
 249                 ts.tv_sec %= 60;
 250                 f = 1;
 251         }
 252         printf("%lds\n", (long)ts.tv_sec);
 253 }
 254
 255 int
 256 doadump(boolean_t textdump)
 257 {
 258         boolean_t coredump;
 259
 260         if (dumping)
 261                 return (EBUSY);
 262         if (dumper.dumper == NULL)
 263                 return (ENXIO);
 264
 265         savectx(&dumppcb);
 266         dumptid = curthread->td_tid;
 267         dumping++;
 268
 269         coredump = TRUE;
 270 #ifdef DDB
 271         if (textdump && textdump_pending) {
 272                 coredump = FALSE;
 273                 textdump_dumpsys(&dumper);
 274         }
 275 #endif
 276         if (coredump)
 277                 dumpsys(&dumper);
 278
 279         dumping--;
 280         return (0);
 281 }
 282
 283 static int
 284 isbufbusy(struct buf *bp)
 285 {
 286         if (((bp->b_flags & (B_INVAL | B_PERSISTENT)) == 0 &&
 287             BUF_ISLOCKED(bp)) ||
 288             ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI))
 289                 return (1);
 290         return (0);
 291 }
 292
 293 /*
 294  * Shutdown the system cleanly to prepare for reboot, halt, or power off.
 295  */
 296 void
 297 kern_reboot(int howto)
 298 {
 299         static int first_buf_printf = 1;
 300
 301 #if defined(SMP)
 302         /*
 303          * Bind us to CPU 0 so that all shutdown code runs there.  Some
 304          * systems don't shutdown properly (i.e., ACPI power off) if we
 305          * run on another processor.
 306          */
 307         if (!SCHEDULER_STOPPED()) {
 308                 thread_lock(curthread);
 309                 sched_bind(curthread, 0);
 310                 thread_unlock(curthread);
 311                 KASSERT(PCPU_GET(cpuid) == 0, ("boot: not running on cpu 0"));
 312         }
 313 #endif
 314         /* We're in the process of rebooting. */
 315         rebooting = 1;
 316
 317         /* collect extra flags that shutdown_nice might have set */
 318         howto |= shutdown_howto;
 319
 320         /* We are out of the debugger now. */
 321         kdb_active = 0;
 322
 323         /*
 324          * Do any callouts that should be done BEFORE syncing the filesystems.
 325          */
 326         EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
 327
 328         /*
 329          * Now sync filesystems
 330          */
 331         if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
 332                 register struct buf *bp;
 333                 int iter, nbusy, pbusy;
 334 #ifndef PREEMPTION
 335                 int subiter;
 336 #endif
 337
 338                 waittime = 0;
 339
 340                 wdog_kern_pat(WD_LASTVAL);
 341                 sys_sync(curthread, NULL);
 342
 343                 /*
 344                  * With soft updates, some buffers that are
 345                  * written will be remarked as dirty until other
 346                  * buffers are written.
 347                  */
 348                 for (iter = pbusy = 0; iter < 20; iter++) {
 349                         nbusy = 0;
 350                         for (bp = &buf[nbuf]; --bp >= buf; )
 351                                 if (isbufbusy(bp))
 352                                         nbusy++;
 353                         if (nbusy == 0) {
 354                                 if (first_buf_printf)
 355                                         printf("All buffers synced.");
 356                                 break;
 357                         }
 358                         if (first_buf_printf) {
 359                                 printf("Syncing disks, buffers remaining... ");
 360                                 first_buf_printf = 0;
 361                         }
 362                         printf("%d ", nbusy);
 363                         if (nbusy < pbusy)
 364                                 iter = 0;
 365                         pbusy = nbusy;
 366
 367                         wdog_kern_pat(WD_LASTVAL);
 368                         sys_sync(curthread, NULL);
 369
 370 #ifdef PREEMPTION
 371                         /*
 372                          * Drop Giant and spin for a while to allow
 373                          * interrupt threads to run.
 374                          */
 375                         DROP_GIANT();
 376                         DELAY(50000 * iter);
 377                         PICKUP_GIANT();
 378 #else
 379                         /*
 380                          * Drop Giant and context switch several times to
 381                          * allow interrupt threads to run.
 382                          */
 383                         DROP_GIANT();
 384                         for (subiter = 0; subiter < 50 * iter; subiter++) {
 385                                 thread_lock(curthread);
 386                                 mi_switch(SW_VOL, NULL);
 387                                 thread_unlock(curthread);
 388                                 DELAY(1000);
 389                         }
 390                         PICKUP_GIANT();
 391 #endif
 392                 }
 393                 printf("\n");
 394                 /*
 395                  * Count only busy local buffers to prevent forcing
 396                  * a fsck if we're just a client of a wedged NFS server
 397                  */
 398                 nbusy = 0;
 399                 for (bp = &buf[nbuf]; --bp >= buf; ) {
 400                         if (isbufbusy(bp)) {
 401 #if 0
 402 /* XXX: This is bogus.  We should probably have a BO_REMOTE flag instead */
 403                                 if (bp->b_dev == NULL) {
 404                                         TAILQ_REMOVE(&mountlist,
 405                                             bp->b_vp->v_mount, mnt_list);
 406                                         continue;
 407                                 }
 408 #endif
 409                                 nbusy++;
 410                                 if (show_busybufs > 0) {
 411                                         printf(
 412             "%d: buf:%p, vnode:%p, flags:%0x, blkno:%jd, lblkno:%jd, buflock:",
 413                                             nbusy, bp, bp->b_vp, bp->b_flags,
 414                                             (intmax_t)bp->b_blkno,
 415                                             (intmax_t)bp->b_lblkno);
 416                                         BUF_LOCKPRINTINFO(bp);
 417                                         if (show_busybufs > 1)
 418                                                 vn_printf(bp->b_vp,
 419                                                     "vnode content: ");
 420                                 }
 421                         }
 422                 }
 423                 if (nbusy) {
 424                         /*
 425                          * Failed to sync all blocks. Indicate this and don't
 426                          * unmount filesystems (thus forcing an fsck on reboot).
 427                          */
 428                         printf("Giving up on %d buffers\n", nbusy);
 429                         DELAY(5000000); /* 5 seconds */
 430                 } else {
 431                         if (!first_buf_printf)
 432                                 printf("Final sync complete\n");
 433                         /*
 434                          * Unmount filesystems
 435                          */
 436                         if (panicstr == 0)
 437                                 vfs_unmountall();
 438                 }
 439                 swapoff_all();
 440                 DELAY(100000);          /* wait for console output to finish */
 441         }
 442
 443         print_uptime();
 444
 445         cngrab();
 446
 447         /*
 448          * Ok, now do things that assume all filesystem activity has
 449          * been completed.
 450          */
 451         EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
 452
 453         if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
 454                 doadump(TRUE);
 455
 456         /* Now that we're going to really halt the system... */
 457         EVENTHANDLER_INVOKE(shutdown_final, howto);
 458
 459         for(;;) ;       /* safety against shutdown_reset not working */
 460         /* NOTREACHED */
 461 }
 462
 463 /*
 464  * If the shutdown was a clean halt, behave accordingly.
 465  */
 466 static void
 467 shutdown_halt(void *junk, int howto)
 468 {
 469
 470         if (howto & RB_HALT) {
 471                 printf("\n");
 472                 printf("The operating system has halted.\n");
 473                 printf("Please press any key to reboot.\n\n");
 474                 switch (cngetc()) {
 475                 case -1:                /* No console, just die */
 476                         cpu_halt();
 477                         /* NOTREACHED */
 478                 default:
 479                         howto &= ~RB_HALT;
 480                         break;
 481                 }
 482         }
 483 }
 484
 485 /*
 486  * Check to see if the system paniced, pause and then reboot
 487  * according to the specified delay.
 488  */
 489 static void
 490 shutdown_panic(void *junk, int howto)
 491 {
 492         int loop;
 493
 494         if (howto & RB_DUMP) {
 495                 if (panic_reboot_wait_time != 0) {
 496                         if (panic_reboot_wait_time != -1) {
 497                                 printf("Automatic reboot in %d seconds - "
 498                                        "press a key on the console to abort\n",
 499                                         panic_reboot_wait_time);
 500                                 for (loop = panic_reboot_wait_time * 10;
 501                                      loop > 0; --loop) {
 502                                         DELAY(1000 * 100); /* 1/10th second */
 503                                         /* Did user type a key? */
 504                                         if (cncheckc() != -1)
 505                                                 break;
 506                                 }
 507                                 if (!loop)
 508                                         return;
 509                         }
 510                 } else { /* zero time specified - reboot NOW */
 511                         return;
 512                 }
 513                 printf("--> Press a key on the console to reboot,\n");
 514                 printf("--> or switch off the system now.\n");
 515                 cngetc();
 516         }
 517 }
 518
 519 /*
 520  * Everything done, now reset
 521  */
 522 static void
 523 shutdown_reset(void *junk, int howto)
 524 {
 525
 526         printf("Rebooting...\n");
 527         DELAY(1000000); /* wait 1 sec for printf's to complete and be read */
 528
 529         /*
 530          * Acquiring smp_ipi_mtx here has a double effect:
 531          * - it disables interrupts avoiding CPU0 preemption
 532          *   by fast handlers (thus deadlocking  against other CPUs)
 533          * - it avoids deadlocks against smp_rendezvous() or, more
 534          *   generally, threads busy-waiting, with this spinlock held,
 535          *   and waiting for responses by threads on other CPUs
 536          *   (ie. smp_tlb_shootdown()).
 537          *
 538          * For the !SMP case it just needs to handle the former problem.
 539          */
 540 #ifdef SMP
 541         mtx_lock_spin(&smp_ipi_mtx);
 542 #else
 543         spinlock_enter();
 544 #endif
 545
 546         /* cpu_boot(howto); */ /* doesn't do anything at the moment */
 547         cpu_reset();
 548         /* NOTREACHED */ /* assuming reset worked */
 549 }
 550
 551 /*
 552  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
 553  * and then reboots.  If we are called twice, then we avoid trying to sync
 554  * the disks as this often leads to recursive panics.
 555  */
 556 void
 557 panic(const char *fmt, ...)
 558 {
 559 #ifdef SMP
 560         static volatile u_int panic_cpu = NOCPU;
 561         cpuset_t other_cpus;
 562 #endif
 563         struct thread *td = curthread;
 564         int bootopt, newpanic;
 565         va_list ap;
 566         static char buf[256];
 567
 568         if (stop_scheduler_on_panic)
 569                 spinlock_enter();
 570         else
 571                 critical_enter();
 572
 573 #ifdef SMP
 574         /*
 575          * We don't want multiple CPU's to panic at the same time, so we
 576          * use panic_cpu as a simple spinlock.  We have to keep checking
 577          * panic_cpu if we are spinning in case the panic on the first
 578          * CPU is canceled.
 579          */
 580         if (panic_cpu != PCPU_GET(cpuid))
 581                 while (atomic_cmpset_int(&panic_cpu, NOCPU,
 582                     PCPU_GET(cpuid)) == 0)
 583                         while (panic_cpu != NOCPU)
 584                                 ; /* nothing */
 585
 586         if (stop_scheduler_on_panic) {
 587                 if (panicstr == NULL && !kdb_active) {
 588                         other_cpus = all_cpus;
 589                         CPU_CLR(PCPU_GET(cpuid), &other_cpus);
 590                         stop_cpus_hard(other_cpus);
 591                 }
 592
 593                 /*
 594                  * We set stop_scheduler here and not in the block above,
 595                  * because we want to ensure that if panic has been called and
 596                  * stop_scheduler_on_panic is true, then stop_scheduler will
 597                  * always be set.  Even if panic has been entered from kdb.
 598                  */
 599                 td->td_stopsched = 1;
 600         }
 601 #endif
 602
 603         bootopt = RB_AUTOBOOT;
 604         newpanic = 0;
 605         if (panicstr)
 606                 bootopt |= RB_NOSYNC;
 607         else {
 608                 bootopt |= RB_DUMP;
 609                 panicstr = fmt;
 610                 newpanic = 1;
 611         }
 612
 613         va_start(ap, fmt);
 614         if (newpanic) {
 615                 (void)vsnprintf(buf, sizeof(buf), fmt, ap);
 616                 panicstr = buf;
 617                 cngrab();
 618                 printf("panic: %s\n", buf);
 619         } else {
 620                 printf("panic: ");
 621                 vprintf(fmt, ap);
 622                 printf("\n");
 623         }
 624         va_end(ap);
 625 #ifdef SMP
 626         printf("cpuid = %d\n", PCPU_GET(cpuid));
 627 #endif
 628
 629 #ifdef KDB
 630         if (newpanic && trace_on_panic)
 631                 kdb_backtrace();
 632         if (debugger_on_panic)
 633                 kdb_enter(KDB_WHY_PANIC, "panic");
 634 #endif
 635         /*thread_lock(td); */
 636         td->td_flags |= TDF_INPANIC;
 637         /* thread_unlock(td); */
 638         if (!sync_on_panic)
 639                 bootopt |= RB_NOSYNC;
 640         if (!stop_scheduler_on_panic)
 641                 critical_exit();
 642         kern_reboot(bootopt);
 643 }
 644
 645 /*
 646  * Support for poweroff delay.
 647  *
 648  * Please note that setting this delay too short might power off your machine
 649  * before the write cache on your hard disk has been flushed, leading to
 650  * soft-updates inconsistencies.
 651  */
 652 #ifndef POWEROFF_DELAY
 653 # define POWEROFF_DELAY 5000
 654 #endif
 655 static int poweroff_delay = POWEROFF_DELAY;
 656
 657 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
 658         &poweroff_delay, 0, "");
 659
 660 static void
 661 poweroff_wait(void *junk, int howto)
 662 {
 663
 664         if (!(howto & RB_POWEROFF) || poweroff_delay <= 0)
 665                 return;
 666         DELAY(poweroff_delay * 1000);
 667 }
 668
 669 /*
 670  * Some system processes (e.g. syncer) need to be stopped at appropriate
 671  * points in their main loops prior to a system shutdown, so that they
 672  * won't interfere with the shutdown process (e.g. by holding a disk buf
 673  * to cause sync to fail).  For each of these system processes, register
 674  * shutdown_kproc() as a handler for one of shutdown events.
 675  */
 676 static int kproc_shutdown_wait = 60;
 677 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
 678     &kproc_shutdown_wait, 0, "");
 679
 680 void
 681 kproc_shutdown(void *arg, int howto)
 682 {
 683         struct proc *p;
 684         int error;
 685
 686         if (panicstr)
 687                 return;
 688
 689         p = (struct proc *)arg;
 690         printf("Waiting (max %d seconds) for system process `%s' to stop...",
 691             kproc_shutdown_wait, p->p_comm);
 692         error = kproc_suspend(p, kproc_shutdown_wait * hz);
 693
 694         if (error == EWOULDBLOCK)
 695                 printf("timed out\n");
 696         else
 697                 printf("done\n");
 698 }
 699
 700 void
 701 kthread_shutdown(void *arg, int howto)
 702 {
 703         struct thread *td;
 704         int error;
 705
 706         if (panicstr)
 707                 return;
 708
 709         td = (struct thread *)arg;
 710         printf("Waiting (max %d seconds) for system thread `%s' to stop...",
 711             kproc_shutdown_wait, td->td_name);
 712         error = kthread_suspend(td, kproc_shutdown_wait * hz);
 713
 714         if (error == EWOULDBLOCK)
 715                 printf("timed out\n");
 716         else
 717                 printf("done\n");
 718 }
 719
 720 /* Registration of dumpers */
 721 int
 722 set_dumper(struct dumperinfo *di)
 723 {
 724
 725         if (di == NULL) {
 726                 bzero(&dumper, sizeof dumper);
 727                 return (0);
 728         }
 729         if (dumper.dumper != NULL)
 730                 return (EBUSY);
 731         dumper = *di;
 732         return (0);
 733 }
 734
 735 /* Call dumper with bounds checking. */
 736 int
 737 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
 738     off_t offset, size_t length)
 739 {
 740
 741         if (length != 0 && (offset < di->mediaoffset ||
 742             offset - di->mediaoffset + length > di->mediasize)) {
 743                 printf("Attempt to write outside dump device boundaries.\n"
 744             "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
 745                     (intmax_t)offset, (intmax_t)di->mediaoffset,
 746                     (uintmax_t)length, (intmax_t)di->mediasize);
 747                 return (ENOSPC);
 748         }
 749         return (di->dumper(di->priv, virtual, physical, offset, length));
 750 }
 751
 752 void
 753 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver,
 754     uint64_t dumplen, uint32_t blksz)
 755 {
 756
 757         bzero(kdh, sizeof(*kdh));
 758         strncpy(kdh->magic, magic, sizeof(kdh->magic));
 759         strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
 760         kdh->version = htod32(KERNELDUMPVERSION);
 761         kdh->architectureversion = htod32(archver);
 762         kdh->dumplength = htod64(dumplen);
 763         kdh->dumptime = htod64(time_second);
 764         kdh->blocksize = htod32(blksz);
 765         strncpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
 766         strncpy(kdh->versionstring, version, sizeof(kdh->versionstring));
 767         if (panicstr != NULL)
 768                 strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
 769         kdh->parity = kerneldump_parity(kdh);
 770 }