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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)kern_shutdown.c 8.3 (Berkeley) 1/21/94 39 * $FreeBSD: src/sys/kern/kern_shutdown.c,v 1.72.2.12 2002/02/21 19:15:10 dillon Exp $ 40 */ 41 42 #include "opt_ddb.h" 43 #include "opt_ddb_trace.h" 44 #include "opt_panic.h" 45 #include "opt_show_busybufs.h" 46 #include "use_gpio.h" 47 48 #include <sys/param.h> 49 #include <sys/systm.h> 50 #include <sys/eventhandler.h> 51 #include <sys/buf.h> 52 #include <sys/disk.h> 53 #include <sys/diskslice.h> 54 #include <sys/reboot.h> 55 #include <sys/proc.h> 56 #include <sys/priv.h> 57 #include <sys/fcntl.h> /* FREAD */ 58 #include <sys/stat.h> /* S_IFCHR */ 59 #include <sys/vnode.h> 60 #include <sys/kernel.h> 61 #include <sys/kerneldump.h> 62 #include <sys/kthread.h> 63 #include <sys/malloc.h> 64 #include <sys/mount.h> 65 #include <sys/queue.h> 66 #include <sys/sysctl.h> 67 #include <sys/vkernel.h> 68 #include <sys/conf.h> 69 #include <sys/sysproto.h> 70 #include <sys/device.h> 71 #include <sys/cons.h> 72 #include <sys/shm.h> 73 #include <sys/kern_syscall.h> 74 #include <vm/vm_map.h> 75 #include <vm/pmap.h> 76 77 #include <sys/thread2.h> 78 #include <sys/buf2.h> 79 #include <sys/mplock2.h> 80 81 #include <machine/cpu.h> 82 #include <machine/clock.h> 83 #include <machine/md_var.h> 84 #include <machine/smp.h> /* smp_active_mask, cpuid */ 85 #include <machine/vmparam.h> 86 #include <machine/thread.h> 87 88 #include <sys/signalvar.h> 89 90 #include <sys/wdog.h> 91 #include <dev/misc/gpio/gpio.h> 92 93 #ifndef PANIC_REBOOT_WAIT_TIME 94 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */ 95 #endif 96 97 /* 98 * Note that stdarg.h and the ANSI style va_start macro is used for both 99 * ANSI and traditional C compilers. We use the machine version to stay 100 * within the confines of the kernel header files. 101 */ 102 #include <machine/stdarg.h> 103 104 #ifdef DDB 105 #include <ddb/ddb.h> 106 #ifdef DDB_UNATTENDED 107 int debugger_on_panic = 0; 108 #else 109 int debugger_on_panic = 1; 110 #endif 111 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW, 112 &debugger_on_panic, 0, "Run debugger on kernel panic"); 113 114 #ifdef DDB_TRACE 115 int trace_on_panic = 1; 116 #else 117 int trace_on_panic = 0; 118 #endif 119 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW, 120 &trace_on_panic, 0, "Print stack trace on kernel panic"); 121 #endif 122 123 static int sync_on_panic = 0; 124 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW, 125 &sync_on_panic, 0, "Do a sync before rebooting from a panic"); 126 127 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment"); 128 129 /* 130 * Variable panicstr contains argument to first call to panic; used as flag 131 * to indicate that the kernel has already called panic. 132 */ 133 const char *panicstr; 134 135 int dumping; /* system is dumping */ 136 static struct dumperinfo dumper; /* selected dumper */ 137 138 globaldata_t panic_cpu_gd; /* which cpu took the panic */ 139 struct lwkt_tokref panic_tokens[LWKT_MAXTOKENS]; 140 int panic_tokens_count; 141 142 int bootverbose = 0; /* note: assignment to force non-bss */ 143 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW, 144 &bootverbose, 0, "Verbose kernel messages"); 145 146 int cold = 1; /* note: assignment to force non-bss */ 147 int dumplo; /* OBSOLETE - savecore compat */ 148 u_int64_t dumplo64; 149 150 static void boot (int) __dead2; 151 static int setdumpdev (cdev_t dev); 152 static void poweroff_wait (void *, int); 153 static void print_uptime (void); 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 static int shutdown_busycount1(struct buf *bp, void *info); 158 static int shutdown_busycount2(struct buf *bp, void *info); 159 static void shutdown_cleanup_proc(struct proc *p); 160 161 /* register various local shutdown events */ 162 static void 163 shutdown_conf(void *unused) 164 { 165 EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, SHUTDOWN_PRI_FIRST); 166 EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, SHUTDOWN_PRI_LAST + 100); 167 EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, SHUTDOWN_PRI_LAST + 100); 168 EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL, SHUTDOWN_PRI_LAST + 200); 169 } 170 171 SYSINIT(shutdown_conf, SI_BOOT2_MACHDEP, SI_ORDER_ANY, shutdown_conf, NULL) 172 173 /* ARGSUSED */ 174 175 /* 176 * The system call that results in a reboot 177 * 178 * MPALMOSTSAFE 179 */ 180 int 181 sys_reboot(struct reboot_args *uap) 182 { 183 struct thread *td = curthread; 184 int error; 185 186 if ((error = priv_check(td, PRIV_REBOOT))) 187 return (error); 188 189 get_mplock(); 190 boot(uap->opt); 191 rel_mplock(); 192 return (0); 193 } 194 195 /* 196 * Called by events that want to shut down.. e.g <CTL><ALT><DEL> on a PC 197 */ 198 static int shutdown_howto = 0; 199 200 void 201 shutdown_nice(int howto) 202 { 203 shutdown_howto = howto; 204 205 /* Send a signal to init(8) and have it shutdown the world */ 206 if (initproc != NULL) { 207 ksignal(initproc, SIGINT); 208 } else { 209 /* No init(8) running, so simply reboot */ 210 boot(RB_NOSYNC); 211 } 212 return; 213 } 214 static int waittime = -1; 215 struct pcb dumppcb; 216 struct thread *dumpthread; 217 218 static void 219 print_uptime(void) 220 { 221 int f; 222 struct timespec ts; 223 224 getnanouptime(&ts); 225 kprintf("Uptime: "); 226 f = 0; 227 if (ts.tv_sec >= 86400) { 228 kprintf("%ldd", ts.tv_sec / 86400); 229 ts.tv_sec %= 86400; 230 f = 1; 231 } 232 if (f || ts.tv_sec >= 3600) { 233 kprintf("%ldh", ts.tv_sec / 3600); 234 ts.tv_sec %= 3600; 235 f = 1; 236 } 237 if (f || ts.tv_sec >= 60) { 238 kprintf("%ldm", ts.tv_sec / 60); 239 ts.tv_sec %= 60; 240 f = 1; 241 } 242 kprintf("%lds\n", ts.tv_sec); 243 } 244 245 /* 246 * Go through the rigmarole of shutting down.. 247 * this used to be in machdep.c but I'll be dammned if I could see 248 * anything machine dependant in it. 249 */ 250 static void 251 boot(int howto) 252 { 253 /* 254 * Get rid of any user scheduler baggage and then give 255 * us a high priority. 256 */ 257 if (curthread->td_release) 258 curthread->td_release(curthread); 259 lwkt_setpri_self(TDPRI_MAX); 260 261 /* collect extra flags that shutdown_nice might have set */ 262 howto |= shutdown_howto; 263 264 /* 265 * We really want to shutdown on the BSP. Subsystems such as ACPI 266 * can't power-down the box otherwise. 267 */ 268 if (smp_active_mask > 1) { 269 kprintf("boot() called on cpu#%d\n", mycpu->gd_cpuid); 270 } 271 if (panicstr == NULL && mycpu->gd_cpuid != 0) { 272 kprintf("Switching to cpu #0 for shutdown\n"); 273 lwkt_setcpu_self(globaldata_find(0)); 274 } 275 /* 276 * Do any callouts that should be done BEFORE syncing the filesystems. 277 */ 278 EVENTHANDLER_INVOKE(shutdown_pre_sync, howto); 279 280 /* 281 * Try to get rid of any remaining FS references. The calling 282 * process, proc0, and init may still hold references. The 283 * VFS cache subsystem may still hold a root reference to root. 284 * 285 * XXX this needs work. We really need to SIGSTOP all remaining 286 * processes in order to avoid blowups due to proc0's filesystem 287 * references going away. For now just make sure that the init 288 * process is stopped. 289 */ 290 if (panicstr == NULL) { 291 shutdown_cleanup_proc(curproc); 292 shutdown_cleanup_proc(&proc0); 293 if (initproc) { 294 if (initproc != curproc) { 295 ksignal(initproc, SIGSTOP); 296 tsleep(boot, 0, "shutdn", hz / 20); 297 } 298 shutdown_cleanup_proc(initproc); 299 } 300 vfs_cache_setroot(NULL, NULL); 301 } 302 303 /* 304 * Now sync filesystems 305 */ 306 if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) { 307 int iter, nbusy, pbusy; 308 309 waittime = 0; 310 kprintf("\nsyncing disks... "); 311 312 sys_sync(NULL); /* YYY was sync(&proc0, NULL). why proc0 ? */ 313 314 /* 315 * With soft updates, some buffers that are 316 * written will be remarked as dirty until other 317 * buffers are written. 318 */ 319 for (iter = pbusy = 0; iter < 20; iter++) { 320 nbusy = scan_all_buffers(shutdown_busycount1, NULL); 321 if (nbusy == 0) 322 break; 323 kprintf("%d ", nbusy); 324 if (nbusy < pbusy) 325 iter = 0; 326 pbusy = nbusy; 327 /* 328 * XXX: 329 * Process soft update work queue if buffers don't sync 330 * after 6 iterations by permitting the syncer to run. 331 */ 332 if (iter > 5) 333 bio_ops_sync(NULL); 334 335 sys_sync(NULL); /* YYY was sync(&proc0, NULL). why proc0 ? */ 336 tsleep(boot, 0, "shutdn", hz * iter / 20 + 1); 337 } 338 kprintf("\n"); 339 /* 340 * Count only busy local buffers to prevent forcing 341 * a fsck if we're just a client of a wedged NFS server 342 */ 343 nbusy = scan_all_buffers(shutdown_busycount2, NULL); 344 if (nbusy) { 345 /* 346 * Failed to sync all blocks. Indicate this and don't 347 * unmount filesystems (thus forcing an fsck on reboot). 348 */ 349 kprintf("giving up on %d buffers\n", nbusy); 350 #ifdef DDB 351 if (debugger_on_panic) 352 Debugger("busy buffer problem"); 353 #endif /* DDB */ 354 tsleep(boot, 0, "shutdn", hz * 5 + 1); 355 } else { 356 kprintf("done\n"); 357 /* 358 * Unmount filesystems 359 */ 360 if (panicstr == NULL) 361 vfs_unmountall(); 362 } 363 tsleep(boot, 0, "shutdn", hz / 10 + 1); 364 } 365 366 print_uptime(); 367 368 /* 369 * Dump before doing post_sync shutdown ops 370 */ 371 crit_enter(); 372 if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold) { 373 dumpsys(); 374 } 375 376 /* 377 * Ok, now do things that assume all filesystem activity has 378 * been completed. This will also call the device shutdown 379 * methods. 380 */ 381 EVENTHANDLER_INVOKE(shutdown_post_sync, howto); 382 383 /* Now that we're going to really halt the system... */ 384 EVENTHANDLER_INVOKE(shutdown_final, howto); 385 386 for(;;) ; /* safety against shutdown_reset not working */ 387 /* NOTREACHED */ 388 } 389 390 /* 391 * Pass 1 - Figure out if there are any busy or dirty buffers still present. 392 * 393 * We ignore TMPFS mounts in this pass. 394 */ 395 static int 396 shutdown_busycount1(struct buf *bp, void *info) 397 { 398 struct vnode *vp; 399 400 if ((vp = bp->b_vp) != NULL && vp->v_tag == VT_TMPFS) 401 return (0); 402 if ((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp) > 0) 403 return(1); 404 if ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) 405 return (1); 406 return (0); 407 } 408 409 /* 410 * Pass 2 - only run after pass 1 has completed or has given up 411 * 412 * We ignore TMPFS, NFS, MFS, and SMBFS mounts in this pass. 413 */ 414 static int 415 shutdown_busycount2(struct buf *bp, void *info) 416 { 417 struct vnode *vp; 418 419 /* 420 * Ignore tmpfs and nfs mounts 421 */ 422 if ((vp = bp->b_vp) != NULL) { 423 if (vp->v_tag == VT_TMPFS) 424 return (0); 425 if (vp->v_tag == VT_NFS) 426 return (0); 427 if (vp->v_tag == VT_MFS) 428 return (0); 429 if (vp->v_tag == VT_SMBFS) 430 return (0); 431 } 432 433 /* 434 * Only count buffers stuck on I/O, ignore everything else 435 */ 436 if (((bp->b_flags & B_INVAL) == 0 && BUF_REFCNT(bp)) || 437 ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) { 438 /* 439 * Only count buffers undergoing write I/O 440 * on the related vnode. 441 */ 442 if (bp->b_vp == NULL || 443 bio_track_active(&bp->b_vp->v_track_write) == 0) { 444 return (0); 445 } 446 #if defined(SHOW_BUSYBUFS) || defined(DIAGNOSTIC) 447 kprintf( 448 "%p dev:?, flags:%08x, loffset:%jd, doffset:%jd\n", 449 bp, 450 bp->b_flags, (intmax_t)bp->b_loffset, 451 (intmax_t)bp->b_bio2.bio_offset); 452 #endif 453 return(1); 454 } 455 return(0); 456 } 457 458 /* 459 * If the shutdown was a clean halt, behave accordingly. 460 */ 461 static void 462 shutdown_halt(void *junk, int howto) 463 { 464 if (howto & RB_HALT) { 465 kprintf("\n"); 466 kprintf("The operating system has halted.\n"); 467 #ifdef _KERNEL_VIRTUAL 468 cpu_halt(); 469 #else 470 kprintf("Please press any key to reboot.\n\n"); 471 switch (cngetc()) { 472 case -1: /* No console, just die */ 473 cpu_halt(); 474 /* NOTREACHED */ 475 default: 476 howto &= ~RB_HALT; 477 break; 478 } 479 #endif 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 kprintf("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 kprintf("--> Press a key on the console to reboot,\n"); 512 kprintf("--> 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 kprintf("Rebooting...\n"); 524 DELAY(1000000); /* wait 1 sec for kprintf's to complete and be read */ 525 /* cpu_boot(howto); */ /* doesn't do anything at the moment */ 526 cpu_reset(); 527 /* NOTREACHED */ /* assuming reset worked */ 528 } 529 530 /* 531 * Try to remove FS references in the specified process. This function 532 * is used during shutdown 533 */ 534 static 535 void 536 shutdown_cleanup_proc(struct proc *p) 537 { 538 struct filedesc *fdp; 539 struct vmspace *vm; 540 541 if (p == NULL) 542 return; 543 if ((fdp = p->p_fd) != NULL) { 544 kern_closefrom(0); 545 if (fdp->fd_cdir) { 546 cache_drop(&fdp->fd_ncdir); 547 vrele(fdp->fd_cdir); 548 fdp->fd_cdir = NULL; 549 } 550 if (fdp->fd_rdir) { 551 cache_drop(&fdp->fd_nrdir); 552 vrele(fdp->fd_rdir); 553 fdp->fd_rdir = NULL; 554 } 555 if (fdp->fd_jdir) { 556 cache_drop(&fdp->fd_njdir); 557 vrele(fdp->fd_jdir); 558 fdp->fd_jdir = NULL; 559 } 560 } 561 if (p->p_vkernel) 562 vkernel_exit(p); 563 if (p->p_textvp) { 564 vrele(p->p_textvp); 565 p->p_textvp = NULL; 566 } 567 vm = p->p_vmspace; 568 if (vm != NULL) { 569 pmap_remove_pages(vmspace_pmap(vm), 570 VM_MIN_USER_ADDRESS, 571 VM_MAX_USER_ADDRESS); 572 vm_map_remove(&vm->vm_map, 573 VM_MIN_USER_ADDRESS, 574 VM_MAX_USER_ADDRESS); 575 } 576 } 577 578 /* 579 * Magic number for savecore 580 * 581 * exported (symorder) and used at least by savecore(8) 582 * 583 * Mark it as used so that gcc doesn't optimize it away. 584 */ 585 __attribute__((__used__)) 586 static u_long const dumpmag = 0x8fca0101UL; 587 588 __attribute__((__used__)) 589 static int dumpsize = 0; /* also for savecore */ 590 591 static int dodump = 1; 592 593 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0, 594 "Try to perform coredump on kernel panic"); 595 596 void 597 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver, 598 uint64_t dumplen, uint32_t blksz) 599 { 600 bzero(kdh, sizeof(*kdh)); 601 strncpy(kdh->magic, magic, sizeof(kdh->magic)); 602 strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture)); 603 kdh->version = htod32(KERNELDUMPVERSION); 604 kdh->architectureversion = htod32(archver); 605 kdh->dumplength = htod64(dumplen); 606 kdh->dumptime = htod64(time_second); 607 kdh->blocksize = htod32(blksz); 608 strncpy(kdh->hostname, hostname, sizeof(kdh->hostname)); 609 strncpy(kdh->versionstring, version, sizeof(kdh->versionstring)); 610 if (panicstr != NULL) 611 strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring)); 612 kdh->parity = kerneldump_parity(kdh); 613 } 614 615 static int 616 setdumpdev(cdev_t dev) 617 { 618 int error; 619 int doopen; 620 621 if (dev == NULL) { 622 disk_dumpconf(NULL, 0/*off*/); 623 return (0); 624 } 625 626 /* 627 * We have to open the device before we can perform ioctls on it, 628 * or the slice/label data may not be present. Device opens are 629 * usually tracked by specfs, but the dump device can be set in 630 * early boot and may not be open so this is somewhat of a hack. 631 */ 632 doopen = (dev->si_sysref.refcnt == 1); 633 if (doopen) { 634 error = dev_dopen(dev, FREAD, S_IFCHR, proc0.p_ucred); 635 if (error) 636 return (error); 637 } 638 error = disk_dumpconf(dev, 1/*on*/); 639 640 return error; 641 } 642 643 /* ARGSUSED */ 644 static void dump_conf (void *dummy); 645 static void 646 dump_conf(void *dummy) 647 { 648 char *path; 649 cdev_t dev; 650 int _dummy; 651 652 path = kmalloc(MNAMELEN, M_TEMP, M_WAITOK); 653 if (TUNABLE_STR_FETCH("dumpdev", path, MNAMELEN) != 0) { 654 /* 655 * Make sure all disk devices created so far have also been 656 * probed, and also make sure that the newly created device 657 * nodes for probed disks are ready, too. 658 * 659 * XXX - Delay an additional 2 seconds to help drivers which 660 * pickup devices asynchronously and are not caught by 661 * CAM's initial probe. 662 */ 663 sync_devs(); 664 tsleep(&_dummy, 0, "syncer", hz*2); 665 666 dev = kgetdiskbyname(path); 667 if (dev != NULL) 668 dumpdev = dev; 669 } 670 kfree(path, M_TEMP); 671 if (setdumpdev(dumpdev) != 0) 672 dumpdev = NULL; 673 } 674 675 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL) 676 677 static int 678 sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS) 679 { 680 int error; 681 udev_t ndumpdev; 682 683 ndumpdev = dev2udev(dumpdev); 684 error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req); 685 if (error == 0 && req->newptr != NULL) 686 error = setdumpdev(udev2dev(ndumpdev, 0)); 687 return (error); 688 } 689 690 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW, 691 0, sizeof dumpdev, sysctl_kern_dumpdev, "T,udev_t", ""); 692 693 /* 694 * Panic is called on unresolvable fatal errors. It prints "panic: mesg", 695 * and then reboots. If we are called twice, then we avoid trying to sync 696 * the disks as this often leads to recursive panics. 697 */ 698 void 699 panic(const char *fmt, ...) 700 { 701 int bootopt, newpanic; 702 globaldata_t gd = mycpu; 703 thread_t td = gd->gd_curthread; 704 __va_list ap; 705 static char buf[256]; 706 707 /* 708 * If a panic occurs on multiple cpus before the first is able to 709 * halt the other cpus, only one cpu is allowed to take the panic. 710 * Attempt to be verbose about this situation but if the kprintf() 711 * itself panics don't let us overrun the kernel stack. 712 * 713 * Be very nasty about descheduling our thread at the lowest 714 * level possible in an attempt to freeze the thread without 715 * inducing further panics. 716 * 717 * Bumping gd_trap_nesting_level will also bypass assertions in 718 * lwkt_switch() and allow us to switch away even if we are a 719 * FAST interrupt or IPI. 720 * 721 * The setting of panic_cpu_gd also determines how kprintf() 722 * spin-locks itself. DDB can set panic_cpu_gd as well. 723 */ 724 for (;;) { 725 globaldata_t xgd = panic_cpu_gd; 726 727 /* 728 * Someone else got the panic cpu 729 */ 730 if (xgd && xgd != gd) { 731 crit_enter(); 732 ++mycpu->gd_trap_nesting_level; 733 if (mycpu->gd_trap_nesting_level < 25) { 734 kprintf("SECONDARY PANIC ON CPU %d THREAD %p\n", 735 mycpu->gd_cpuid, td); 736 } 737 td->td_release = NULL; /* be a grinch */ 738 for (;;) { 739 lwkt_deschedule_self(td); 740 lwkt_switch(); 741 } 742 /* NOT REACHED */ 743 /* --mycpu->gd_trap_nesting_level */ 744 /* crit_exit() */ 745 } 746 747 /* 748 * Reentrant panic 749 */ 750 if (xgd && xgd == gd) 751 break; 752 753 /* 754 * We got it 755 */ 756 if (atomic_cmpset_ptr(&panic_cpu_gd, NULL, gd)) 757 break; 758 } 759 /* 760 * Try to get the system into a working state. Save information 761 * we are about to destroy. 762 */ 763 kvcreinitspin(); 764 if (panicstr == NULL) { 765 bcopy(td->td_toks_array, panic_tokens, sizeof(panic_tokens)); 766 panic_tokens_count = td->td_toks_stop - &td->td_toks_base; 767 } 768 lwkt_relalltokens(td); 769 td->td_toks_stop = &td->td_toks_base; 770 771 /* 772 * Setup 773 */ 774 bootopt = RB_AUTOBOOT | RB_DUMP; 775 if (sync_on_panic == 0) 776 bootopt |= RB_NOSYNC; 777 newpanic = 0; 778 if (panicstr) { 779 bootopt |= RB_NOSYNC; 780 } else { 781 panicstr = fmt; 782 newpanic = 1; 783 } 784 785 /* 786 * Format the panic string. 787 */ 788 __va_start(ap, fmt); 789 kvsnprintf(buf, sizeof(buf), fmt, ap); 790 if (panicstr == fmt) 791 panicstr = buf; 792 __va_end(ap); 793 kprintf("panic: %s\n", buf); 794 /* two separate prints in case of an unmapped page and trap */ 795 kprintf("cpuid = %d\n", mycpu->gd_cpuid); 796 797 #if (NGPIO > 0) && defined(ERROR_LED_ON_PANIC) 798 led_switch("error", 1); 799 #endif 800 801 #if defined(WDOG_DISABLE_ON_PANIC) 802 wdog_disable(); 803 #endif 804 805 /* 806 * Enter the debugger or fall through & dump. Entering the 807 * debugger will stop cpus. If not entering the debugger stop 808 * cpus here. 809 */ 810 #if defined(DDB) 811 if (newpanic && trace_on_panic) 812 print_backtrace(-1); 813 if (debugger_on_panic) 814 Debugger("panic"); 815 else 816 #endif 817 if (newpanic) 818 stop_cpus(mycpu->gd_other_cpus); 819 boot(bootopt); 820 } 821 822 /* 823 * Support for poweroff delay. 824 */ 825 #ifndef POWEROFF_DELAY 826 # define POWEROFF_DELAY 5000 827 #endif 828 static int poweroff_delay = POWEROFF_DELAY; 829 830 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW, 831 &poweroff_delay, 0, ""); 832 833 static void 834 poweroff_wait(void *junk, int howto) 835 { 836 if(!(howto & RB_POWEROFF) || poweroff_delay <= 0) 837 return; 838 DELAY(poweroff_delay * 1000); 839 } 840 841 /* 842 * Some system processes (e.g. syncer) need to be stopped at appropriate 843 * points in their main loops prior to a system shutdown, so that they 844 * won't interfere with the shutdown process (e.g. by holding a disk buf 845 * to cause sync to fail). For each of these system processes, register 846 * shutdown_kproc() as a handler for one of shutdown events. 847 */ 848 static int kproc_shutdown_wait = 60; 849 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW, 850 &kproc_shutdown_wait, 0, ""); 851 852 void 853 shutdown_kproc(void *arg, int howto) 854 { 855 struct thread *td; 856 struct proc *p; 857 int error; 858 859 if (panicstr) 860 return; 861 862 td = (struct thread *)arg; 863 if ((p = td->td_proc) != NULL) { 864 kprintf("Waiting (max %d seconds) for system process `%s' to stop...", 865 kproc_shutdown_wait, p->p_comm); 866 } else { 867 kprintf("Waiting (max %d seconds) for system thread %s to stop...", 868 kproc_shutdown_wait, td->td_comm); 869 } 870 error = suspend_kproc(td, kproc_shutdown_wait * hz); 871 872 if (error == EWOULDBLOCK) 873 kprintf("timed out\n"); 874 else 875 kprintf("stopped\n"); 876 } 877 878 /* Registration of dumpers */ 879 int 880 set_dumper(struct dumperinfo *di) 881 { 882 if (di == NULL) { 883 bzero(&dumper, sizeof(dumper)); 884 return 0; 885 } 886 887 if (dumper.dumper != NULL) 888 return (EBUSY); 889 890 dumper = *di; 891 return 0; 892 } 893 894 void 895 dumpsys(void) 896 { 897 #if defined (_KERNEL_VIRTUAL) 898 /* VKERNELs don't support dumps */ 899 kprintf("VKERNEL doesn't support dumps\n"); 900 return; 901 #endif 902 /* 903 * If there is a dumper registered and we aren't dumping already, call 904 * the machine dependent dumpsys (md_dumpsys) to do the hard work. 905 * 906 * XXX: while right now the md_dumpsys() of x86 and x86_64 could be 907 * factored out completely into here, I rather keep them machine 908 * dependent in case we ever add a platform which does not share 909 * the same dumpsys() code, such as arm. 910 */ 911 if (dumper.dumper != NULL && !dumping) { 912 dumping++; 913 md_dumpsys(&dumper); 914 } 915 } 916 917 int dump_stop_usertds = 0; 918 919 static 920 void 921 need_user_resched_remote(void *dummy) 922 { 923 need_user_resched(); 924 } 925 926 void 927 dump_reactivate_cpus(void) 928 { 929 globaldata_t gd; 930 int cpu, seq; 931 932 dump_stop_usertds = 1; 933 934 need_user_resched(); 935 936 for (cpu = 0; cpu < ncpus; cpu++) { 937 gd = globaldata_find(cpu); 938 seq = lwkt_send_ipiq(gd, need_user_resched_remote, NULL); 939 lwkt_wait_ipiq(gd, seq); 940 } 941 942 restart_cpus(stopped_cpus); 943 } 944