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. 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 * $FreeBSD: src/sys/kern/kern_shutdown.c,v 1.72.2.12 2002/02/21 19:15:10 dillon Exp $
36 */
37
38 #include "opt_ddb.h"
39 #include "opt_ddb_trace.h"
40 #include "opt_panic.h"
41 #include "use_gpio.h"
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/eventhandler.h>
46 #include <sys/buf.h>
47 #include <sys/disk.h>
48 #include <sys/diskslice.h>
49 #include <sys/reboot.h>
50 #include <sys/proc.h>
51 #include <sys/caps.h>
52 #include <sys/fcntl.h> /* FREAD */
53 #include <sys/stat.h> /* S_IFCHR */
54 #include <sys/vnode.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/queue.h>
61 #include <sys/sysctl.h>
62 #include <sys/vkernel.h>
63 #include <sys/conf.h>
64 #include <sys/sysmsg.h>
65 #include <sys/device.h>
66 #include <sys/cons.h>
67 #include <sys/kbio.h>
68 #include <sys/shm.h>
69 #include <sys/kern_syscall.h>
70 #include <vm/vm_map.h>
71 #include <vm/pmap.h>
72
73 #include <sys/thread2.h>
74 #include <sys/buf2.h>
75 #include <sys/mplock2.h>
76
77 #include <machine/cpu.h>
78 #include <machine/clock.h>
79 #include <machine/md_var.h>
80 #include <machine/smp.h> /* smp_active_mask, cpuid */
81 #include <machine/vmparam.h>
82 #include <machine/thread.h>
83
84 #include <sys/signalvar.h>
85
86 #if defined(WDOG_DISABLE_ON_PANIC)
87 #include <sys/wdog.h>
88 #endif
89 #include <dev/acpica/acpi_pvpanic/panic_notifier.h>
90 #if (NGPIO > 0) && defined(ERROR_LED_ON_PANIC)
91 #include <dev/misc/gpio/gpio.h>
92 #endif
93
94 #ifndef PANIC_REBOOT_WAIT_TIME
95 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
96 #endif
97
98 /*
99 * Note that stdarg.h and the ANSI style va_start macro is used for both
100 * ANSI and traditional C compilers. We use the machine version to stay
101 * within the confines of the kernel header files.
102 */
103 #include <machine/stdarg.h>
104
105 #ifdef DDB
106 #include <ddb/ddb.h>
107 #ifdef DDB_UNATTENDED
108 int debugger_on_panic = 0;
109 #else
110 int debugger_on_panic = 1;
111 #endif
112 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RW,
113 &debugger_on_panic, 0, "Run debugger on kernel panic");
114
115 #ifdef DDB_TRACE
116 int trace_on_panic = 1;
117 #else
118 int trace_on_panic = 0;
119 #endif
120 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RW,
121 &trace_on_panic, 0, "Print stack trace on kernel panic");
122 #endif
123
124 static int sync_on_panic = 0;
125 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RW,
126 &sync_on_panic, 0, "Do a sync before rebooting from a panic");
127
128 SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment");
129
130 /*
131 * Variable panicstr contains argument to first call to panic; used as flag
132 * to indicate that the kernel has already called panic.
133 */
134 const char *panicstr;
135
136 __read_mostly int dumping; /* system is dumping */
137 static struct dumperinfo dumper; /* selected dumper */
138
139 __read_frequently globaldata_t panic_cpu_gd; /* used in lock assertion */
140 struct lwkt_tokref panic_tokens[LWKT_MAXTOKENS];
141 int panic_tokens_count;
142
143 __read_mostly int bootverbose = 0; /* note: assignment to force non-bss */
144 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW,
145 &bootverbose, 0, "Verbose kernel messages");
146
147 int cold = 1; /* note: assignment to force non-bss */
148 int dumplo; /* OBSOLETE - savecore compat */
149 u_int64_t dumplo64;
150
151 static void boot (int) __dead2;
152 static int setdumpdev (cdev_t dev);
153 static void poweroff_wait (void *, int);
154 static void print_uptime (void);
155 static void shutdown_halt (void *junk, int howto);
156 static void shutdown_panic (void *junk, int howto);
157 static void shutdown_reset (void *junk, int howto);
158 static int shutdown_busycount1(struct buf *bp, void *info);
159 static int shutdown_busycount2(struct buf *bp, void *info);
160 static void shutdown_cleanup_proc(struct proc *p);
161
162 /* register various local shutdown events */
163 static void
shutdown_conf(void * unused)164 shutdown_conf(void *unused)
165 {
166 EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, SHUTDOWN_PRI_FIRST);
167 EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, SHUTDOWN_PRI_LAST + 100);
168 EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, SHUTDOWN_PRI_LAST + 100);
169 EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL, SHUTDOWN_PRI_LAST + 200);
170 }
171
172 SYSINIT(shutdown_conf, SI_BOOT2_MACHDEP, SI_ORDER_ANY, shutdown_conf, NULL);
173
174 /* ARGSUSED */
175
176 /*
177 * The system call that results in a reboot
178 *
179 * MPALMOSTSAFE
180 */
181 int
sys_reboot(struct sysmsg * sysmsg,const struct reboot_args * uap)182 sys_reboot(struct sysmsg *sysmsg, const struct reboot_args *uap)
183 {
184 int error;
185
186 if ((error = caps_priv_check_self(SYSCAP_NOREBOOT)))
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
shutdown_nice(int howto)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
print_uptime(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
boot(int howto)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 (!CPUMASK_ISUP(smp_active_mask)) {
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 int zcount;
309
310 waittime = 0;
311 zcount = 0;
312 kprintf("\nsyncing disks... ");
313
314 sys_sync(NULL, NULL);
315
316 /*
317 * With soft updates, some buffers that are written will be
318 * remarked as dirty until other buffers are written.
319 *
320 * sys_sync() usually runs asynchronously, to give us a
321 * better chance of syncing the rest of the filesystems when
322 * one or more of them are stuck.
323 */
324 for (iter = pbusy = 0; iter < 20 + zcount; iter++) {
325 if (iter <= 10)
326 nbusy = scan_all_buffers(shutdown_busycount1,
327 &iter);
328 else
329 nbusy = scan_all_buffers(shutdown_busycount2,
330 &iter);
331 kprintf("%d ", nbusy);
332 if (nbusy == 0) {
333 if (++zcount == 3)
334 break;
335 } else {
336 zcount = 0;
337 }
338
339 /*
340 * There could be a lot to sync, only allow iter to
341 * proceed while there is progress.
342 */
343 if (nbusy < pbusy) {
344 if (iter > 10)
345 iter = 10;
346 else
347 iter = 0;
348 }
349 pbusy = nbusy;
350
351 /*
352 * XXX:
353 * Process soft update work queue if buffers don't sync
354 * after 6 iterations by permitting the syncer to run.
355 */
356 if (iter > 5)
357 bio_ops_sync(NULL);
358
359 sys_sync(NULL, NULL);
360 tsleep(boot, 0, "shutdn", hz * iter / 20 + 1);
361 }
362 kprintf("\n");
363
364 if (zcount < 3) {
365 /*
366 * Failed to sync all blocks. Indicate this and don't
367 * unmount filesystems (thus forcing an fsck on reboot).
368 */
369 kprintf("giving up on %d buffers\n", nbusy);
370 #ifdef DDB
371 if (debugger_on_panic)
372 Debugger("busy buffer problem");
373 #endif /* DDB */
374 tsleep(boot, 0, "shutdn", hz * 5 + 1);
375 } else {
376 kprintf("done\n");
377
378 /*
379 * Unmount filesystems
380 */
381 if (panicstr == NULL)
382 vfs_unmountall(1);
383 }
384 tsleep(boot, 0, "shutdn", hz / 10 + 1);
385 }
386
387 print_uptime();
388
389 /*
390 * Dump before doing post_sync shutdown ops
391 */
392 crit_enter();
393 if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold) {
394 dumpsys();
395 }
396
397 /*
398 * Ok, now do things that assume all filesystem activity has
399 * been completed. This will also call the device shutdown
400 * methods.
401 */
402 EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
403
404 /* Now that we're going to really halt the system... */
405 EVENTHANDLER_INVOKE(shutdown_final, howto);
406
407 for(;;) ; /* safety against shutdown_reset not working */
408 /* NOTREACHED */
409 }
410
411 /*
412 * Pass 1 - Figure out if there are any busy or dirty buffers still present.
413 *
414 * We ignore TMPFS mounts in this pass.
415 */
416 static int
shutdown_busycount1(struct buf * bp,void * info __unused)417 shutdown_busycount1(struct buf *bp, void *info __unused)
418 {
419 struct vnode *vp;
420
421 if ((vp = bp->b_vp) != NULL && vp->v_tag == VT_TMPFS)
422 return (0);
423 if ((bp->b_flags & B_INVAL) == 0 && BUF_LOCKINUSE(bp))
424 return(1);
425 if ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI)
426 return (1);
427 return (0);
428 }
429
430 /*
431 * Pass 2 - only run after pass 1 has completed or has given up
432 *
433 * We ignore TMPFS, NFS, MFS, and SMBFS mounts in this pass.
434 */
435 static int
shutdown_busycount2(struct buf * bp,void * info)436 shutdown_busycount2(struct buf *bp, void *info)
437 {
438 struct vnode *vp;
439 int *iterp = info;
440 const char *mpath;
441
442 /*
443 * Ignore tmpfs and nfs mounts
444 */
445 if ((vp = bp->b_vp) != NULL) {
446 if (vp->v_tag == VT_TMPFS)
447 return (0);
448 if (vp->v_tag == VT_NFS)
449 return (0);
450 if (vp->v_tag == VT_MFS)
451 return (0);
452 if (vp->v_tag == VT_SMBFS)
453 return (0);
454 }
455
456 /*
457 * Only count buffers stuck on I/O, ignore everything else
458 */
459 if (((bp->b_flags & B_INVAL) == 0 && BUF_LOCKINUSE(bp)) ||
460 ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI)) {
461 /*
462 * Only count buffers undergoing write I/O
463 * on the related vnode.
464 */
465 if (bp->b_vp == NULL ||
466 bio_track_active(&bp->b_vp->v_track_write) == 0) {
467 return (0);
468 }
469 if (*iterp > 15) {
470 mpath = "?";
471 if (bp->b_vp->v_mount)
472 mpath = bp->b_vp->v_mount->mnt_stat.f_mntonname;
473
474 kprintf("%p on %s, flags:%08x, loffset:%jd, "
475 "doffset:%jd\n",
476 bp,
477 mpath,
478 bp->b_flags,
479 (intmax_t)bp->b_loffset,
480 (intmax_t)bp->b_bio2.bio_offset);
481 }
482 return(1);
483 }
484 return(0);
485 }
486
487 /*
488 * If the shutdown was a clean halt, behave accordingly.
489 */
490 static void
shutdown_halt(void * junk,int howto)491 shutdown_halt(void *junk, int howto)
492 {
493 if (howto & RB_HALT) {
494 kprintf("\n");
495 kprintf("The operating system has halted.\n");
496 #ifdef _KERNEL_VIRTUAL
497 cpu_halt();
498 #else
499 kprintf("Please press any key to reboot.\n\n");
500 cnpoll(TRUE);
501 switch (cngetc()) {
502 case -1: /* No console, just die */
503 cpu_halt();
504 /* NOTREACHED */
505 default:
506 howto &= ~RB_HALT;
507 break;
508 }
509 #endif
510 }
511 }
512
513 /*
514 * Check to see if the system paniced, pause and then reboot
515 * according to the specified delay.
516 */
517 static void
shutdown_panic(void * junk,int howto)518 shutdown_panic(void *junk, int howto)
519 {
520 int loop;
521 int c;
522
523 if (howto & RB_DUMP) {
524 if (PANIC_REBOOT_WAIT_TIME != 0) {
525 if (PANIC_REBOOT_WAIT_TIME != -1) {
526 kprintf("Automatic reboot in %d seconds - "
527 "press a key on the console to abort\n",
528 PANIC_REBOOT_WAIT_TIME);
529 for (loop = PANIC_REBOOT_WAIT_TIME * 10;
530 loop > 0; --loop) {
531 DELAY(1000 * 100); /* 1/10th second */
532 /* Did user type a key? */
533 c = cncheckc();
534 if (c != -1 && c != NOKEY)
535 break;
536 }
537 if (!loop)
538 return;
539 }
540 } else { /* zero time specified - reboot NOW */
541 return;
542 }
543 kprintf("--> Press a key on the console to reboot,\n");
544 kprintf("--> or switch off the system now.\n");
545 cngetc();
546 }
547 }
548
549 /*
550 * Everything done, now reset
551 */
552 static void
shutdown_reset(void * junk,int howto)553 shutdown_reset(void *junk, int howto)
554 {
555 kprintf("Rebooting...\n");
556 DELAY(1000000); /* wait 1 sec for kprintf's to complete and be read */
557 /* cpu_boot(howto); */ /* doesn't do anything at the moment */
558 cpu_reset();
559 /* NOTREACHED */ /* assuming reset worked */
560 }
561
562 /*
563 * Try to remove FS references in the specified process. This function
564 * is used during shutdown
565 */
566 static
567 void
shutdown_cleanup_proc(struct proc * p)568 shutdown_cleanup_proc(struct proc *p)
569 {
570 struct filedesc *fdp;
571 struct vmspace *vm;
572
573 if (p == NULL)
574 return;
575 if ((fdp = p->p_fd) != NULL) {
576 kern_closefrom(0);
577 if (fdp->fd_cdir) {
578 cache_drop(&fdp->fd_ncdir);
579 vrele(fdp->fd_cdir);
580 fdp->fd_cdir = NULL;
581 }
582 if (fdp->fd_rdir) {
583 cache_drop(&fdp->fd_nrdir);
584 vrele(fdp->fd_rdir);
585 fdp->fd_rdir = NULL;
586 }
587 if (fdp->fd_jdir) {
588 cache_drop(&fdp->fd_njdir);
589 vrele(fdp->fd_jdir);
590 fdp->fd_jdir = NULL;
591 }
592 }
593 if (p->p_vkernel)
594 vkernel_exit(p);
595 if (p->p_textvp) {
596 vrele(p->p_textvp);
597 p->p_textvp = NULL;
598 }
599 if (p->p_textnch.ncp)
600 cache_drop(&p->p_textnch);
601 vm = p->p_vmspace;
602 if (vm != NULL) {
603 pmap_remove_pages(vmspace_pmap(vm),
604 VM_MIN_USER_ADDRESS,
605 VM_MAX_USER_ADDRESS);
606 vm_map_remove(&vm->vm_map,
607 VM_MIN_USER_ADDRESS,
608 VM_MAX_USER_ADDRESS);
609 }
610 }
611
612 /*
613 * Magic number for savecore
614 *
615 * exported (symorder) and used at least by savecore(8)
616 *
617 * Mark it as used so that gcc doesn't optimize it away.
618 */
619 __attribute__((__used__))
620 static u_long const dumpmag = 0x8fca0101UL;
621
622 __attribute__((__used__))
623 static int dumpsize = 0; /* also for savecore */
624
625 static int dodump = 1;
626
627 SYSCTL_INT(_machdep, OID_AUTO, do_dump, CTLFLAG_RW, &dodump, 0,
628 "Try to perform coredump on kernel panic");
629
630 void
mkdumpheader(struct kerneldumpheader * kdh,char * magic,uint32_t archver,uint64_t dumplen,uint32_t blksz)631 mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver,
632 uint64_t dumplen, uint32_t blksz)
633 {
634 bzero(kdh, sizeof(*kdh));
635 strncpy(kdh->magic, magic, sizeof(kdh->magic));
636 strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
637 kdh->version = htod32(KERNELDUMPVERSION);
638 kdh->architectureversion = htod32(archver);
639 kdh->dumplength = htod64(dumplen);
640 kdh->dumptime = htod64(time_second);
641 kdh->blocksize = htod32(blksz);
642 strncpy(kdh->hostname, hostname, sizeof(kdh->hostname));
643 strncpy(kdh->versionstring, version, sizeof(kdh->versionstring));
644 if (panicstr != NULL)
645 strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
646 kdh->parity = kerneldump_parity(kdh);
647 }
648
649 static int
setdumpdev(cdev_t dev)650 setdumpdev(cdev_t dev)
651 {
652 int error;
653 int doopen;
654
655 if (dev == NULL) {
656 disk_dumpconf(NULL, 0/*off*/);
657 dumpdev = NULL;
658 return (0);
659 }
660
661 /*
662 * We have to open the device before we can perform ioctls on it,
663 * or the slice/label data may not be present. Device opens are
664 * usually tracked by specfs, but the dump device can be set in
665 * early boot and may not be open so this is somewhat of a hack.
666 */
667 doopen = (dev->si_sysref.refcnt == 1);
668 if (doopen) {
669 error = dev_dopen(dev, FREAD, S_IFCHR,
670 proc0.p_ucred, NULL, NULL);
671 if (error)
672 return (error);
673 }
674 error = disk_dumpconf(dev, 1/*on*/);
675 if (error == 0)
676 dumpdev = dev;
677
678 return error;
679 }
680
681 /* ARGSUSED */
682 static void dump_conf (void *dummy);
683 static void
dump_conf(void * dummy)684 dump_conf(void *dummy)
685 {
686 char *path;
687 cdev_t dev;
688 int _dummy;
689
690 path = kmalloc(MNAMELEN, M_TEMP, M_WAITOK);
691 if (TUNABLE_STR_FETCH("dumpdev", path, MNAMELEN) != 0) {
692 /*
693 * Make sure all disk devices created so far have also been
694 * probed, and also make sure that the newly created device
695 * nodes for probed disks are ready, too.
696 *
697 * XXX - Delay an additional 2 seconds to help drivers which
698 * pickup devices asynchronously and are not caught by
699 * CAM's initial probe.
700 */
701 sync_devs();
702 tsleep(&_dummy, 0, "syncer", hz*2);
703
704 dev = kgetdiskbyname(path);
705 if (dev != NULL)
706 dumpdev = dev;
707 }
708 kfree(path, M_TEMP);
709 if (setdumpdev(dumpdev) != 0)
710 dumpdev = NULL;
711 }
712
713 SYSINIT(dump_conf, SI_SUB_DUMP_CONF, SI_ORDER_FIRST, dump_conf, NULL);
714
715 static int
sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS)716 sysctl_kern_dumpdev(SYSCTL_HANDLER_ARGS)
717 {
718 int error;
719 dev_t ndumpdev;
720
721 ndumpdev = devid_from_dev(dumpdev);
722 error = sysctl_handle_opaque(oidp, &ndumpdev, sizeof ndumpdev, req);
723 if (error == 0 && req->newptr != NULL)
724 error = setdumpdev(dev_from_devid(ndumpdev, 0));
725 return (error);
726 }
727
728 SYSCTL_PROC(_kern, KERN_DUMPDEV, dumpdev, CTLTYPE_OPAQUE|CTLFLAG_RW,
729 0, sizeof dumpdev, sysctl_kern_dumpdev, "T,udev_t", "");
730
731 static struct panicerinfo *panic_notifier;
732
733 int
set_panic_notifier(struct panicerinfo * info)734 set_panic_notifier(struct panicerinfo *info)
735 {
736 if (info == NULL)
737 panic_notifier = NULL;
738 else if (panic_notifier != NULL)
739 return 1;
740 else
741 panic_notifier = info;
742
743 return 0;
744 }
745
746 /*
747 * Panic is called on unresolvable fatal errors. It prints "panic: mesg",
748 * and then reboots. If we are called twice, then we avoid trying to sync
749 * the disks as this often leads to recursive panics.
750 */
751 void
panic(const char * fmt,...)752 panic(const char *fmt, ...)
753 {
754 int bootopt, newpanic;
755 globaldata_t gd = mycpu;
756 thread_t td = gd->gd_curthread;
757 __va_list ap;
758 static char buf[256];
759
760 /*
761 * If a panic occurs on multiple cpus before the first is able to
762 * halt the other cpus, only one cpu is allowed to take the panic.
763 * Attempt to be verbose about this situation but if the kprintf()
764 * itself panics don't let us overrun the kernel stack.
765 *
766 * Be very nasty about descheduling our thread at the lowest
767 * level possible in an attempt to freeze the thread without
768 * inducing further panics.
769 *
770 * Bumping gd_trap_nesting_level will also bypass assertions in
771 * lwkt_switch() and allow us to switch away even if we are a
772 * FAST interrupt or IPI.
773 *
774 * The setting of panic_cpu_gd also determines how kprintf()
775 * spin-locks itself. DDB can set panic_cpu_gd as well.
776 */
777 for (;;) {
778 globaldata_t xgd = panic_cpu_gd;
779
780 /*
781 * Someone else got the panic cpu
782 */
783 if (xgd && xgd != gd) {
784 crit_enter();
785 ++mycpu->gd_trap_nesting_level;
786 if (mycpu->gd_trap_nesting_level < 25) {
787 kprintf("SECONDARY PANIC ON CPU %d THREAD %p\n",
788 mycpu->gd_cpuid, td);
789 }
790 td->td_release = NULL; /* be a grinch */
791 for (;;) {
792 lwkt_deschedule_self(td);
793 lwkt_switch();
794 }
795 /* NOT REACHED */
796 /* --mycpu->gd_trap_nesting_level */
797 /* crit_exit() */
798 }
799
800 /*
801 * Reentrant panic
802 */
803 if (xgd && xgd == gd)
804 break;
805
806 /*
807 * We got it
808 */
809 if (atomic_cmpset_ptr(&panic_cpu_gd, NULL, gd))
810 break;
811 }
812 /*
813 * Try to get the system into a working state. Save information
814 * we are about to destroy.
815 */
816 kvcreinitspin();
817 if (panicstr == NULL) {
818 bcopy(td->td_toks_array, panic_tokens, sizeof(panic_tokens));
819 panic_tokens_count = td->td_toks_stop - &td->td_toks_base;
820 }
821 lwkt_relalltokens(td);
822 td->td_toks_stop = &td->td_toks_base;
823 if (gd->gd_spinlocks)
824 kprintf("panic with %d spinlocks held\n", gd->gd_spinlocks);
825 gd->gd_spinlocks = 0;
826
827 /*
828 * Setup
829 */
830 bootopt = RB_AUTOBOOT | RB_DUMP;
831 if (sync_on_panic == 0)
832 bootopt |= RB_NOSYNC;
833 newpanic = 0;
834 if (panicstr) {
835 bootopt |= RB_NOSYNC;
836 } else {
837 panicstr = fmt;
838 newpanic = 1;
839 }
840
841 /*
842 * Format the panic string.
843 */
844 __va_start(ap, fmt);
845 kvsnprintf(buf, sizeof(buf), fmt, ap);
846 if (panicstr == fmt)
847 panicstr = buf;
848 __va_end(ap);
849 if (panic_notifier != NULL)
850 panic_notifier->notifier(panic_notifier->arg);
851 kprintf("panic: %s\n", buf);
852 /* two separate prints in case of an unmapped page and trap */
853 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
854
855 #if (NGPIO > 0) && defined(ERROR_LED_ON_PANIC)
856 led_switch("error", 1);
857 #endif
858
859 #if defined(WDOG_DISABLE_ON_PANIC)
860 wdog_disable();
861 #endif
862
863 /*
864 * Make sure kgdb knows who we are, there won't be a stoppcbs[]
865 * entry since our cpu wasn't stopped.
866 */
867 savectx(&dumppcb);
868 dumpthread = curthread;
869
870 /*
871 * Enter the debugger or fall through & dump. Entering the
872 * debugger will stop cpus. If not entering the debugger stop
873 * cpus here.
874 *
875 * Limit the trace history to leave more panic data on a
876 * potentially row-limited console.
877 */
878
879 #if defined(DDB)
880 if (newpanic && trace_on_panic)
881 print_backtrace(6);
882 if (debugger_on_panic)
883 Debugger("panic");
884 else
885 #endif
886 if (newpanic)
887 stop_cpus(mycpu->gd_other_cpus);
888 boot(bootopt);
889 }
890
891 /*
892 * Support for poweroff delay.
893 */
894 #ifndef POWEROFF_DELAY
895 # define POWEROFF_DELAY 5000
896 #endif
897 static int poweroff_delay = POWEROFF_DELAY;
898
899 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
900 &poweroff_delay, 0, "");
901
902 static void
poweroff_wait(void * junk,int howto)903 poweroff_wait(void *junk, int howto)
904 {
905 if(!(howto & RB_POWEROFF) || poweroff_delay <= 0)
906 return;
907 DELAY(poweroff_delay * 1000);
908 }
909
910 /*
911 * Some system processes (e.g. syncer) need to be stopped at appropriate
912 * points in their main loops prior to a system shutdown, so that they
913 * won't interfere with the shutdown process (e.g. by holding a disk buf
914 * to cause sync to fail). For each of these system processes, register
915 * shutdown_kproc() as a handler for one of shutdown events.
916 */
917 static int kproc_shutdown_wait = 60;
918 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
919 &kproc_shutdown_wait, 0, "");
920
921 void
shutdown_kproc(void * arg,int howto)922 shutdown_kproc(void *arg, int howto)
923 {
924 struct thread *td;
925 struct proc *p;
926 int error;
927
928 if (panicstr)
929 return;
930
931 td = (struct thread *)arg;
932 if ((p = td->td_proc) != NULL) {
933 kprintf("Waiting (max %d seconds) for system process `%s' to stop...",
934 kproc_shutdown_wait, p->p_comm);
935 } else {
936 kprintf("Waiting (max %d seconds) for system thread %s to stop...",
937 kproc_shutdown_wait, td->td_comm);
938 }
939 error = suspend_kproc(td, kproc_shutdown_wait * hz);
940
941 if (error == EWOULDBLOCK)
942 kprintf("timed out\n");
943 else
944 kprintf("stopped\n");
945 }
946
947 /* Registration of dumpers */
948 int
set_dumper(struct dumperinfo * di)949 set_dumper(struct dumperinfo *di)
950 {
951 if (di == NULL) {
952 bzero(&dumper, sizeof(dumper));
953 return 0;
954 }
955
956 if (dumper.dumper != NULL)
957 return (EBUSY);
958
959 dumper = *di;
960 return 0;
961 }
962
963 void
dumpsys(void)964 dumpsys(void)
965 {
966 #if defined (_KERNEL_VIRTUAL)
967 /* vkernels don't support dumps */
968 kprintf("vkernels don't support dumps\n");
969 return;
970 #endif
971 /*
972 * If there is a dumper registered and we aren't dumping already, call
973 * the machine dependent dumpsys (md_dumpsys) to do the hard work.
974 *
975 * XXX: while right now the md_dumpsys() of x86 and x86_64 could be
976 * factored out completely into here, I rather keep them machine
977 * dependent in case we ever add a platform which does not share
978 * the same dumpsys() code, such as arm.
979 */
980 if (dumper.dumper != NULL && !dumping) {
981 dumping++;
982 md_dumpsys(&dumper);
983 }
984 }
985
986 __read_frequently int dump_stop_usertds = 0;
987
988 static
989 void
need_user_resched_remote(void * dummy)990 need_user_resched_remote(void *dummy)
991 {
992 need_user_resched();
993 }
994
995 void
dump_reactivate_cpus(void)996 dump_reactivate_cpus(void)
997 {
998 globaldata_t gd;
999 int cpu, seq;
1000
1001 dump_stop_usertds = 1;
1002
1003 need_user_resched();
1004
1005 for (cpu = 0; cpu < ncpus; cpu++) {
1006 gd = globaldata_find(cpu);
1007 seq = lwkt_send_ipiq(gd, need_user_resched_remote, NULL);
1008 lwkt_wait_ipiq(gd, seq);
1009 }
1010
1011 restart_cpus(stopped_cpus);
1012 }
1013