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