2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1986, 1988, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)kern_shutdown.c 8.3 (Berkeley) 1/21/94
39 #include <sys/cdefs.h>
43 #include "opt_panic.h"
44 #include "opt_printf.h"
45 #include "opt_sched.h"
46 #include "opt_watchdog.h"
48 #include <sys/param.h>
49 #include <sys/systm.h>
53 #include <sys/compressor.h>
56 #include <sys/eventhandler.h>
57 #include <sys/filedesc.h>
60 #include <sys/kernel.h>
61 #include <sys/kerneldump.h>
62 #include <sys/kthread.h>
64 #include <sys/malloc.h>
66 #include <sys/mount.h>
69 #include <sys/reboot.h>
70 #include <sys/resourcevar.h>
71 #include <sys/rwlock.h>
73 #include <sys/sched.h>
75 #include <sys/sysctl.h>
76 #include <sys/sysproto.h>
77 #include <sys/taskqueue.h>
78 #include <sys/vnode.h>
79 #include <sys/watchdog.h>
81 #include <crypto/chacha20/chacha.h>
82 #include <crypto/rijndael/rijndael-api-fst.h>
83 #include <crypto/sha2/sha256.h>
87 #include <machine/cpu.h>
88 #include <machine/dump.h>
89 #include <machine/pcb.h>
90 #include <machine/smp.h>
92 #include <security/mac/mac_framework.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_page.h>
97 #include <vm/vm_pager.h>
98 #include <vm/swap_pager.h>
100 #include <sys/signalvar.h>
102 static MALLOC_DEFINE(M_DUMPER, "dumper", "dumper block buffer");
104 #ifndef PANIC_REBOOT_WAIT_TIME
105 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
107 static int panic_reboot_wait_time = PANIC_REBOOT_WAIT_TIME;
108 SYSCTL_INT(_kern, OID_AUTO, panic_reboot_wait_time, CTLFLAG_RWTUN,
109 &panic_reboot_wait_time, 0,
110 "Seconds to wait before rebooting after a panic");
113 * Note that stdarg.h and the ANSI style va_start macro is used for both
114 * ANSI and traditional C compilers.
116 #include <machine/stdarg.h>
119 #ifdef KDB_UNATTENDED
120 int debugger_on_panic = 0;
122 int debugger_on_panic = 1;
124 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic,
125 CTLFLAG_RWTUN | CTLFLAG_SECURE,
126 &debugger_on_panic, 0, "Run debugger on kernel panic");
128 static bool debugger_on_recursive_panic = false;
129 SYSCTL_BOOL(_debug, OID_AUTO, debugger_on_recursive_panic,
130 CTLFLAG_RWTUN | CTLFLAG_SECURE,
131 &debugger_on_recursive_panic, 0, "Run debugger on recursive kernel panic");
133 int debugger_on_trap = 0;
134 SYSCTL_INT(_debug, OID_AUTO, debugger_on_trap,
135 CTLFLAG_RWTUN | CTLFLAG_SECURE,
136 &debugger_on_trap, 0, "Run debugger on kernel trap before panic");
139 static int trace_on_panic = 1;
140 static bool trace_all_panics = true;
142 static int trace_on_panic = 0;
143 static bool trace_all_panics = false;
145 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
146 CTLFLAG_RWTUN | CTLFLAG_SECURE,
147 &trace_on_panic, 0, "Print stack trace on kernel panic");
148 SYSCTL_BOOL(_debug, OID_AUTO, trace_all_panics, CTLFLAG_RWTUN,
149 &trace_all_panics, 0, "Print stack traces on secondary kernel panics");
152 static int sync_on_panic = 0;
153 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN,
154 &sync_on_panic, 0, "Do a sync before rebooting from a panic");
156 static bool poweroff_on_panic = 0;
157 SYSCTL_BOOL(_kern, OID_AUTO, poweroff_on_panic, CTLFLAG_RWTUN,
158 &poweroff_on_panic, 0, "Do a power off instead of a reboot on a panic");
160 static bool powercycle_on_panic = 0;
161 SYSCTL_BOOL(_kern, OID_AUTO, powercycle_on_panic, CTLFLAG_RWTUN,
162 &powercycle_on_panic, 0, "Do a power cycle instead of a reboot on a panic");
164 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
165 "Shutdown environment");
168 static int show_busybufs;
170 static int show_busybufs = 1;
172 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
174 "Show busy buffers during shutdown");
176 int suspend_blocked = 0;
177 SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW,
178 &suspend_blocked, 0, "Block suspend due to a pending shutdown");
181 FEATURE(ekcd, "Encrypted kernel crash dumps support");
183 MALLOC_DEFINE(M_EKCD, "ekcd", "Encrypted kernel crash dumps data");
185 struct kerneldumpcrypto {
186 uint8_t kdc_encryption;
187 uint8_t kdc_iv[KERNELDUMP_IV_MAX_SIZE];
191 cipherInstance aes_ci;
193 struct chacha_ctx u_chacha;
195 #define kdc_ki u.u_aes.aes_ki
196 #define kdc_ci u.u_aes.aes_ci
197 #define kdc_chacha u.u_chacha
198 uint32_t kdc_dumpkeysize;
199 struct kerneldumpkey kdc_dumpkey[];
203 struct kerneldumpcomp {
205 struct compressor *kdc_stream;
210 static struct kerneldumpcomp *kerneldumpcomp_create(struct dumperinfo *di,
211 uint8_t compression);
212 static void kerneldumpcomp_destroy(struct dumperinfo *di);
213 static int kerneldumpcomp_write_cb(void *base, size_t len, off_t off, void *arg);
215 static int kerneldump_gzlevel = 6;
216 SYSCTL_INT(_kern, OID_AUTO, kerneldump_gzlevel, CTLFLAG_RWTUN,
217 &kerneldump_gzlevel, 0,
218 "Kernel crash dump compression level");
221 * Variable panicstr contains argument to first call to panic; used as flag
222 * to indicate that the kernel has already called panic.
224 const char *panicstr;
225 bool __read_frequently panicked;
227 int __read_mostly dumping; /* system is dumping */
228 int rebooting; /* system is rebooting */
230 * Used to serialize between sysctl kern.shutdown.dumpdevname and list
231 * modifications via ioctl.
233 static struct mtx dumpconf_list_lk;
234 MTX_SYSINIT(dumper_configs, &dumpconf_list_lk, "dumper config list", MTX_DEF);
236 /* Our selected dumper(s). */
237 static TAILQ_HEAD(dumpconflist, dumperinfo) dumper_configs =
238 TAILQ_HEAD_INITIALIZER(dumper_configs);
240 /* Context information for dump-debuggers, saved by the dump_savectx() macro. */
241 struct pcb dumppcb; /* Registers. */
242 lwpid_t dumptid; /* Thread ID. */
244 static struct cdevsw reroot_cdevsw = {
245 .d_version = D_VERSION,
249 static void poweroff_wait(void *, int);
250 static void shutdown_halt(void *junk, int howto);
251 static void shutdown_panic(void *junk, int howto);
252 static void shutdown_reset(void *junk, int howto);
253 static int kern_reroot(void);
255 /* register various local shutdown events */
257 shutdown_conf(void *unused)
260 EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
262 EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
263 SHUTDOWN_PRI_LAST + 100);
264 EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
265 SHUTDOWN_PRI_LAST + 100);
268 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
271 * The only reason this exists is to create the /dev/reroot/ directory,
272 * used by reroot code in init(8) as a mountpoint for tmpfs.
275 reroot_conf(void *unused)
280 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev,
281 &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot");
283 printf("%s: failed to create device node, error %d",
288 SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL);
291 * The system call that results in a reboot.
295 sys_reboot(struct thread *td, struct reboot_args *uap)
301 error = mac_system_check_reboot(td->td_ucred, uap->opt);
304 error = priv_check(td, PRIV_REBOOT);
306 if (uap->opt & RB_REROOT)
307 error = kern_reroot();
309 kern_reboot(uap->opt);
315 shutdown_nice_task_fn(void *arg, int pending __unused)
319 howto = (uintptr_t)arg;
320 /* Send a signal to init(8) and have it shutdown the world. */
322 if (howto & RB_POWEROFF)
323 kern_psignal(initproc, SIGUSR2);
324 else if (howto & RB_POWERCYCLE)
325 kern_psignal(initproc, SIGWINCH);
326 else if (howto & RB_HALT)
327 kern_psignal(initproc, SIGUSR1);
329 kern_psignal(initproc, SIGINT);
330 PROC_UNLOCK(initproc);
333 static struct task shutdown_nice_task = TASK_INITIALIZER(0,
334 &shutdown_nice_task_fn, NULL);
337 * Called by events that want to shut down.. e.g <CTL><ALT><DEL> on a PC
340 shutdown_nice(int howto)
343 if (initproc != NULL && !SCHEDULER_STOPPED()) {
344 shutdown_nice_task.ta_context = (void *)(uintptr_t)howto;
345 taskqueue_enqueue(taskqueue_fast, &shutdown_nice_task);
348 * No init(8) running, or scheduler would not allow it
349 * to run, so simply reboot.
351 kern_reboot(howto | RB_NOSYNC);
364 if (ts.tv_sec >= 86400) {
365 printf("%ldd", (long)ts.tv_sec / 86400);
369 if (f || ts.tv_sec >= 3600) {
370 printf("%ldh", (long)ts.tv_sec / 3600);
374 if (f || ts.tv_sec >= 60) {
375 printf("%ldm", (long)ts.tv_sec / 60);
379 printf("%lds\n", (long)ts.tv_sec);
383 doadump(boolean_t textdump)
391 if (TAILQ_EMPTY(&dumper_configs))
399 if (textdump && textdump_pending) {
401 textdump_dumpsys(TAILQ_FIRST(&dumper_configs));
405 struct dumperinfo *di;
407 TAILQ_FOREACH(di, &dumper_configs, di_next) {
419 * kern_reboot(9): Shut down the system cleanly to prepare for reboot, halt, or
423 kern_reboot(int howto)
428 * Normal paths here don't hold Giant, but we can wind up here
429 * unexpectedly with it held. Drop it now so we don't have to
430 * drop and pick it up elsewhere. The paths it is locking will
431 * never be returned to, and it is preferable to preclude
432 * deadlock than to lock against code that won't ever
435 while (mtx_owned(&Giant))
440 * Bind us to the first CPU so that all shutdown code runs there. Some
441 * systems don't shutdown properly (i.e., ACPI power off) if we
442 * run on another processor.
444 if (!SCHEDULER_STOPPED()) {
445 thread_lock(curthread);
446 sched_bind(curthread, CPU_FIRST());
447 thread_unlock(curthread);
448 KASSERT(PCPU_GET(cpuid) == CPU_FIRST(),
449 ("%s: not running on cpu 0", __func__));
452 /* We're in the process of rebooting. */
455 /* We are out of the debugger now. */
459 * Do any callouts that should be done BEFORE syncing the filesystems.
461 EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
464 * Now sync filesystems
466 if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
468 bufshutdown(show_busybufs);
476 * Ok, now do things that assume all filesystem activity has
479 EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
481 if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
484 /* Now that we're going to really halt the system... */
485 EVENTHANDLER_INVOKE(shutdown_final, howto);
488 * Call this directly so that reset is attempted even if shutdown
489 * handlers are not yet registered.
491 shutdown_reset(NULL, howto);
493 for(;;) ; /* safety against shutdown_reset not working */
498 * The system call that results in changing the rootfs.
503 struct vnode *oldrootvnode, *vp;
504 struct mount *mp, *devmp;
507 if (curproc != initproc)
511 * Mark the filesystem containing currently-running executable
512 * (the temporary copy of init(8)) busy.
514 vp = curproc->p_textvp;
515 error = vn_lock(vp, LK_SHARED);
519 error = vfs_busy(mp, MBF_NOWAIT);
523 error = vfs_busy(mp, 0);
524 vn_lock(vp, LK_SHARED | LK_RETRY);
530 if (VN_IS_DOOMED(vp)) {
539 * Remove the filesystem containing currently-running executable
540 * from the mount list, to prevent it from being unmounted
541 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
543 * Also preserve /dev - forcibly unmounting it could cause driver
551 mtx_lock(&mountlist_mtx);
552 TAILQ_REMOVE(&mountlist, mp, mnt_list);
553 TAILQ_REMOVE(&mountlist, devmp, mnt_list);
554 mtx_unlock(&mountlist_mtx);
556 oldrootvnode = rootvnode;
559 * Unmount everything except for the two filesystems preserved above.
564 * Add /dev back; vfs_mountroot() will move it into its new place.
566 mtx_lock(&mountlist_mtx);
567 TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
568 mtx_unlock(&mountlist_mtx);
573 * Mount the new rootfs.
578 * Update all references to the old rootvnode.
580 mountcheckdirs(oldrootvnode, rootvnode);
583 * Add the temporary filesystem back and unbusy it.
585 mtx_lock(&mountlist_mtx);
586 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
587 mtx_unlock(&mountlist_mtx);
594 * If the shutdown was a clean halt, behave accordingly.
597 shutdown_halt(void *junk, int howto)
600 if (howto & RB_HALT) {
602 printf("The operating system has halted.\n");
603 printf("Please press any key to reboot.\n\n");
605 wdog_kern_pat(WD_TO_NEVER);
608 case -1: /* No console, just die */
618 * Check to see if the system panicked, pause and then reboot
619 * according to the specified delay.
622 shutdown_panic(void *junk, int howto)
626 if (howto & RB_DUMP) {
627 if (panic_reboot_wait_time != 0) {
628 if (panic_reboot_wait_time != -1) {
629 printf("Automatic reboot in %d seconds - "
630 "press a key on the console to abort\n",
631 panic_reboot_wait_time);
632 for (loop = panic_reboot_wait_time * 10;
634 DELAY(1000 * 100); /* 1/10th second */
635 /* Did user type a key? */
636 if (cncheckc() != -1)
642 } else { /* zero time specified - reboot NOW */
645 printf("--> Press a key on the console to reboot,\n");
646 printf("--> or switch off the system now.\n");
652 * Everything done, now reset
655 shutdown_reset(void *junk, int howto)
658 printf("Rebooting...\n");
659 DELAY(1000000); /* wait 1 sec for printf's to complete and be read */
662 * Acquiring smp_ipi_mtx here has a double effect:
663 * - it disables interrupts avoiding CPU0 preemption
664 * by fast handlers (thus deadlocking against other CPUs)
665 * - it avoids deadlocks against smp_rendezvous() or, more
666 * generally, threads busy-waiting, with this spinlock held,
667 * and waiting for responses by threads on other CPUs
668 * (ie. smp_tlb_shootdown()).
670 * For the !SMP case it just needs to handle the former problem.
673 mtx_lock_spin(&smp_ipi_mtx);
679 /* NOTREACHED */ /* assuming reset worked */
682 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
683 static int kassert_warn_only = 0;
685 static int kassert_do_kdb = 0;
688 static int kassert_do_ktr = 0;
690 static int kassert_do_log = 1;
691 static int kassert_log_pps_limit = 4;
692 static int kassert_log_mute_at = 0;
693 static int kassert_log_panic_at = 0;
694 static int kassert_suppress_in_panic = 0;
695 static int kassert_warnings = 0;
697 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
700 #ifdef KASSERT_PANIC_OPTIONAL
701 #define KASSERT_RWTUN CTLFLAG_RWTUN
703 #define KASSERT_RWTUN CTLFLAG_RDTUN
706 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, KASSERT_RWTUN,
707 &kassert_warn_only, 0,
708 "KASSERT triggers a panic (0) or just a warning (1)");
711 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, KASSERT_RWTUN,
712 &kassert_do_kdb, 0, "KASSERT will enter the debugger");
716 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, KASSERT_RWTUN,
718 "KASSERT does a KTR, set this to the KTRMASK you want");
721 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, KASSERT_RWTUN,
723 "If warn_only is enabled, log (1) or do not log (0) assertion violations");
725 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RD | CTLFLAG_STATS,
726 &kassert_warnings, 0, "number of KASSERTs that have been triggered");
728 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, KASSERT_RWTUN,
729 &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
731 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, KASSERT_RWTUN,
732 &kassert_log_pps_limit, 0, "limit number of log messages per second");
734 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, KASSERT_RWTUN,
735 &kassert_log_mute_at, 0, "max number of KASSERTS to log");
737 SYSCTL_INT(_debug_kassert, OID_AUTO, suppress_in_panic, KASSERT_RWTUN,
738 &kassert_suppress_in_panic, 0,
739 "KASSERTs will be suppressed while handling a panic");
742 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
744 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
745 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_MPSAFE, NULL, 0,
746 kassert_sysctl_kassert, "I",
747 "set to trigger a test kassert");
750 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
754 error = sysctl_wire_old_buffer(req, sizeof(int));
757 error = sysctl_handle_int(oidp, &i, 0, req);
759 if (error != 0 || req->newptr == NULL)
761 KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
765 #ifdef KASSERT_PANIC_OPTIONAL
767 * Called by KASSERT, this decides if we will panic
768 * or if we will log via printf and/or ktr.
771 kassert_panic(const char *fmt, ...)
773 static char buf[256];
777 (void)vsnprintf(buf, sizeof(buf), fmt, ap);
781 * If we are suppressing secondary panics, log the warning but do not
782 * re-enter panic/kdb.
784 if (KERNEL_PANICKED() && kassert_suppress_in_panic) {
785 if (kassert_do_log) {
786 printf("KASSERT failed: %s\n", buf);
788 if (trace_all_panics && trace_on_panic)
796 * panic if we're not just warning, or if we've exceeded
797 * kassert_log_panic_at warnings.
799 if (!kassert_warn_only ||
800 (kassert_log_panic_at > 0 &&
801 kassert_warnings >= kassert_log_panic_at)) {
811 * log if we've not yet met the mute limit.
813 if (kassert_do_log &&
814 (kassert_log_mute_at == 0 ||
815 kassert_warnings < kassert_log_mute_at)) {
816 static struct timeval lasterr;
819 if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
820 printf("KASSERT failed: %s\n", buf);
825 if (kassert_do_kdb) {
826 kdb_enter(KDB_WHY_KASSERT, buf);
829 atomic_add_int(&kassert_warnings, 1);
831 #endif /* KASSERT_PANIC_OPTIONAL */
835 * Panic is called on unresolvable fatal errors. It prints "panic: mesg",
836 * and then reboots. If we are called twice, then we avoid trying to sync
837 * the disks as this often leads to recursive panics.
840 panic(const char *fmt, ...)
849 vpanic(const char *fmt, va_list ap)
854 struct thread *td = curthread;
855 int bootopt, newpanic;
856 static char buf[256];
862 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
863 * concurrently entering panic. Only the winner will proceed
866 if (panicstr == NULL && !kdb_active) {
867 other_cpus = all_cpus;
868 CPU_CLR(PCPU_GET(cpuid), &other_cpus);
869 stop_cpus_hard(other_cpus);
874 * Ensure that the scheduler is stopped while panicking, even if panic
875 * has been entered from kdb.
877 td->td_stopsched = 1;
879 bootopt = RB_AUTOBOOT;
881 if (KERNEL_PANICKED())
882 bootopt |= RB_NOSYNC;
891 (void)vsnprintf(buf, sizeof(buf), fmt, ap);
894 printf("panic: %s\n", buf);
901 printf("cpuid = %d\n", PCPU_GET(cpuid));
903 printf("time = %jd\n", (intmax_t )time_second);
905 if ((newpanic || trace_all_panics) && trace_on_panic)
907 if (debugger_on_panic)
908 kdb_enter(KDB_WHY_PANIC, "panic");
909 else if (!newpanic && debugger_on_recursive_panic)
910 kdb_enter(KDB_WHY_PANIC, "re-panic");
912 /*thread_lock(td); */
913 td->td_flags |= TDF_INPANIC;
914 /* thread_unlock(td); */
916 bootopt |= RB_NOSYNC;
917 if (poweroff_on_panic)
918 bootopt |= RB_POWEROFF;
919 if (powercycle_on_panic)
920 bootopt |= RB_POWERCYCLE;
921 kern_reboot(bootopt);
925 * Support for poweroff delay.
927 * Please note that setting this delay too short might power off your machine
928 * before the write cache on your hard disk has been flushed, leading to
929 * soft-updates inconsistencies.
931 #ifndef POWEROFF_DELAY
932 # define POWEROFF_DELAY 5000
934 static int poweroff_delay = POWEROFF_DELAY;
936 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
937 &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
940 poweroff_wait(void *junk, int howto)
943 if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
945 DELAY(poweroff_delay * 1000);
949 * Some system processes (e.g. syncer) need to be stopped at appropriate
950 * points in their main loops prior to a system shutdown, so that they
951 * won't interfere with the shutdown process (e.g. by holding a disk buf
952 * to cause sync to fail). For each of these system processes, register
953 * shutdown_kproc() as a handler for one of shutdown events.
955 static int kproc_shutdown_wait = 60;
956 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
957 &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
960 kproc_shutdown(void *arg, int howto)
965 if (KERNEL_PANICKED())
968 p = (struct proc *)arg;
969 printf("Waiting (max %d seconds) for system process `%s' to stop... ",
970 kproc_shutdown_wait, p->p_comm);
971 error = kproc_suspend(p, kproc_shutdown_wait * hz);
973 if (error == EWOULDBLOCK)
974 printf("timed out\n");
980 kthread_shutdown(void *arg, int howto)
985 if (KERNEL_PANICKED())
988 td = (struct thread *)arg;
989 printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
990 kproc_shutdown_wait, td->td_name);
991 error = kthread_suspend(td, kproc_shutdown_wait * hz);
993 if (error == EWOULDBLOCK)
994 printf("timed out\n");
1000 dumpdevname_sysctl_handler(SYSCTL_HANDLER_ARGS)
1003 struct dumperinfo *di;
1007 error = sysctl_wire_old_buffer(req, 0);
1011 sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1013 mtx_lock(&dumpconf_list_lk);
1014 TAILQ_FOREACH(di, &dumper_configs, di_next) {
1015 if (di != TAILQ_FIRST(&dumper_configs))
1016 sbuf_putc(&sb, ',');
1017 sbuf_cat(&sb, di->di_devname);
1019 mtx_unlock(&dumpconf_list_lk);
1021 error = sbuf_finish(&sb);
1025 SYSCTL_PROC(_kern_shutdown, OID_AUTO, dumpdevname,
1026 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, &dumper_configs, 0,
1027 dumpdevname_sysctl_handler, "A",
1028 "Device(s) for kernel dumps");
1030 static int _dump_append(struct dumperinfo *di, void *virtual, size_t length);
1033 static struct kerneldumpcrypto *
1034 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
1035 const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1037 struct kerneldumpcrypto *kdc;
1038 struct kerneldumpkey *kdk;
1039 uint32_t dumpkeysize;
1041 dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
1042 kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
1044 arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
1046 kdc->kdc_encryption = encryption;
1047 switch (kdc->kdc_encryption) {
1048 case KERNELDUMP_ENC_AES_256_CBC:
1049 if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
1052 case KERNELDUMP_ENC_CHACHA20:
1053 chacha_keysetup(&kdc->kdc_chacha, key, 256);
1059 kdc->kdc_dumpkeysize = dumpkeysize;
1060 kdk = kdc->kdc_dumpkey;
1061 kdk->kdk_encryption = kdc->kdc_encryption;
1062 memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1063 kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
1064 memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
1073 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
1075 uint8_t hash[SHA256_DIGEST_LENGTH];
1077 struct kerneldumpkey *kdk;
1086 * When a user enters ddb it can write a crash dump multiple times.
1087 * Each time it should be encrypted using a different IV.
1090 SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1091 SHA256_Final(hash, &ctx);
1092 bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1094 switch (kdc->kdc_encryption) {
1095 case KERNELDUMP_ENC_AES_256_CBC:
1096 if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1097 kdc->kdc_iv) <= 0) {
1102 case KERNELDUMP_ENC_CHACHA20:
1103 chacha_ivsetup(&kdc->kdc_chacha, kdc->kdc_iv, NULL);
1110 kdk = kdc->kdc_dumpkey;
1111 memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1113 explicit_bzero(hash, sizeof(hash));
1118 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
1123 return (kdc->kdc_dumpkeysize);
1127 static struct kerneldumpcomp *
1128 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression)
1130 struct kerneldumpcomp *kdcomp;
1133 switch (compression) {
1134 case KERNELDUMP_COMP_GZIP:
1135 format = COMPRESS_GZIP;
1137 case KERNELDUMP_COMP_ZSTD:
1138 format = COMPRESS_ZSTD;
1144 kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO);
1145 kdcomp->kdc_format = compression;
1146 kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb,
1147 format, di->maxiosize, kerneldump_gzlevel, di);
1148 if (kdcomp->kdc_stream == NULL) {
1149 free(kdcomp, M_DUMPER);
1152 kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1157 kerneldumpcomp_destroy(struct dumperinfo *di)
1159 struct kerneldumpcomp *kdcomp;
1161 kdcomp = di->kdcomp;
1164 compressor_fini(kdcomp->kdc_stream);
1165 zfree(kdcomp->kdc_buf, M_DUMPER);
1166 free(kdcomp, M_DUMPER);
1170 * Free a dumper. Must not be present on global list.
1173 dumper_destroy(struct dumperinfo *di)
1179 zfree(di->blockbuf, M_DUMPER);
1180 kerneldumpcomp_destroy(di);
1182 zfree(di->kdcrypto, M_EKCD);
1184 zfree(di, M_DUMPER);
1188 * Allocate and set up a new dumper from the provided template.
1191 dumper_create(const struct dumperinfo *di_template, const char *devname,
1192 const struct diocskerneldump_arg *kda, struct dumperinfo **dip)
1194 struct dumperinfo *newdi;
1200 /* Allocate a new dumper */
1201 newdi = malloc(sizeof(*newdi) + strlen(devname) + 1, M_DUMPER,
1203 memcpy(newdi, di_template, sizeof(*newdi));
1204 newdi->blockbuf = NULL;
1205 newdi->kdcrypto = NULL;
1206 newdi->kdcomp = NULL;
1207 strcpy(newdi->di_devname, devname);
1209 if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
1211 newdi->kdcrypto = kerneldumpcrypto_create(newdi->blocksize,
1212 kda->kda_encryption, kda->kda_key,
1213 kda->kda_encryptedkeysize, kda->kda_encryptedkey);
1214 if (newdi->kdcrypto == NULL) {
1223 if (kda->kda_compression != KERNELDUMP_COMP_NONE) {
1226 * We can't support simultaneous unpadded block cipher
1227 * encryption and compression because there is no guarantee the
1228 * length of the compressed result is exactly a multiple of the
1229 * cipher block size.
1231 if (kda->kda_encryption == KERNELDUMP_ENC_AES_256_CBC) {
1236 newdi->kdcomp = kerneldumpcomp_create(newdi,
1237 kda->kda_compression);
1238 if (newdi->kdcomp == NULL) {
1243 newdi->blockbuf = malloc(newdi->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1248 dumper_destroy(newdi);
1253 * Create a new dumper and register it in the global list.
1256 dumper_insert(const struct dumperinfo *di_template, const char *devname,
1257 const struct diocskerneldump_arg *kda)
1259 struct dumperinfo *newdi, *listdi;
1264 index = kda->kda_index;
1265 MPASS(index != KDA_REMOVE && index != KDA_REMOVE_DEV &&
1266 index != KDA_REMOVE_ALL);
1268 error = priv_check(curthread, PRIV_SETDUMPER);
1272 error = dumper_create(di_template, devname, kda, &newdi);
1276 /* Add the new configuration to the queue */
1277 mtx_lock(&dumpconf_list_lk);
1279 TAILQ_FOREACH(listdi, &dumper_configs, di_next) {
1281 TAILQ_INSERT_BEFORE(listdi, newdi, di_next);
1288 TAILQ_INSERT_TAIL(&dumper_configs, newdi, di_next);
1289 mtx_unlock(&dumpconf_list_lk);
1296 dumper_ddb_insert(struct dumperinfo *newdi)
1298 TAILQ_INSERT_HEAD(&dumper_configs, newdi, di_next);
1302 dumper_ddb_remove(struct dumperinfo *di)
1304 TAILQ_REMOVE(&dumper_configs, di, di_next);
1309 dumper_config_match(const struct dumperinfo *di, const char *devname,
1310 const struct diocskerneldump_arg *kda)
1312 if (kda->kda_index == KDA_REMOVE_ALL)
1315 if (strcmp(di->di_devname, devname) != 0)
1319 * Allow wildcard removal of configs matching a device on g_dev_orphan.
1321 if (kda->kda_index == KDA_REMOVE_DEV)
1324 if (di->kdcomp != NULL) {
1325 if (di->kdcomp->kdc_format != kda->kda_compression)
1327 } else if (kda->kda_compression != KERNELDUMP_COMP_NONE)
1330 if (di->kdcrypto != NULL) {
1331 if (di->kdcrypto->kdc_encryption != kda->kda_encryption)
1334 * Do we care to verify keys match to delete? It seems weird
1335 * to expect multiple fallback dump configurations on the same
1336 * device that only differ in crypto key.
1340 if (kda->kda_encryption != KERNELDUMP_ENC_NONE)
1347 * Remove and free the requested dumper(s) from the global list.
1350 dumper_remove(const char *devname, const struct diocskerneldump_arg *kda)
1352 struct dumperinfo *di, *sdi;
1356 error = priv_check(curthread, PRIV_SETDUMPER);
1361 * Try to find a matching configuration, and kill it.
1363 * NULL 'kda' indicates remove any configuration matching 'devname',
1364 * which may remove multiple configurations in atypical configurations.
1367 mtx_lock(&dumpconf_list_lk);
1368 TAILQ_FOREACH_SAFE(di, &dumper_configs, di_next, sdi) {
1369 if (dumper_config_match(di, devname, kda)) {
1371 TAILQ_REMOVE(&dumper_configs, di, di_next);
1375 mtx_unlock(&dumpconf_list_lk);
1377 /* Only produce ENOENT if a more targeted match didn't match. */
1378 if (!found && kda->kda_index == KDA_REMOVE)
1384 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1387 if (di->mediasize > 0 && length != 0 && (offset < di->mediaoffset ||
1388 offset - di->mediaoffset + length > di->mediasize)) {
1389 if (di->kdcomp != NULL && offset >= di->mediaoffset) {
1391 "Compressed dump failed to fit in device boundaries.\n");
1395 printf("Attempt to write outside dump device boundaries.\n"
1396 "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1397 (intmax_t)offset, (intmax_t)di->mediaoffset,
1398 (uintmax_t)length, (intmax_t)di->mediasize);
1401 if (length % di->blocksize != 0) {
1402 printf("Attempt to write partial block of length %ju.\n",
1406 if (offset % di->blocksize != 0) {
1407 printf("Attempt to write at unaligned offset %jd.\n",
1417 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1420 switch (kdc->kdc_encryption) {
1421 case KERNELDUMP_ENC_AES_256_CBC:
1422 if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1423 8 * size, buf) <= 0) {
1426 if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1427 buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1431 case KERNELDUMP_ENC_CHACHA20:
1432 chacha_encrypt_bytes(&kdc->kdc_chacha, buf, buf, size);
1441 /* Encrypt data and call dumper. */
1443 dump_encrypted_write(struct dumperinfo *di, void *virtual, off_t offset,
1446 static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1447 struct kerneldumpcrypto *kdc;
1453 while (length > 0) {
1454 nbytes = MIN(length, sizeof(buf));
1455 bcopy(virtual, buf, nbytes);
1457 if (dump_encrypt(kdc, buf, nbytes) != 0)
1460 error = dump_write(di, buf, offset, nbytes);
1465 virtual = (void *)((uint8_t *)virtual + nbytes);
1474 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1476 struct dumperinfo *di;
1477 size_t resid, rlength;
1482 if (length % di->blocksize != 0) {
1484 * This must be the final write after flushing the compression
1485 * stream. Write as many full blocks as possible and stash the
1486 * residual data in the dumper's block buffer. It will be
1487 * padded and written in dump_finish().
1489 rlength = rounddown(length, di->blocksize);
1491 error = _dump_append(di, base, rlength);
1495 resid = length - rlength;
1496 memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1497 bzero((uint8_t *)di->blockbuf + resid, di->blocksize - resid);
1498 di->kdcomp->kdc_resid = resid;
1501 return (_dump_append(di, base, length));
1505 * Write kernel dump headers at the beginning and end of the dump extent.
1506 * Write the kernel dump encryption key after the leading header if we were
1507 * configured to do so.
1510 dump_write_headers(struct dumperinfo *di, struct kerneldumpheader *kdh)
1513 struct kerneldumpcrypto *kdc;
1521 hdrsz = sizeof(*kdh);
1522 if (hdrsz > di->blocksize)
1527 keysize = kerneldumpcrypto_dumpkeysize(kdc);
1533 * If the dump device has special handling for headers, let it take care
1534 * of writing them out.
1536 if (di->dumper_hdr != NULL)
1537 return (di->dumper_hdr(di, kdh));
1539 if (hdrsz == di->blocksize)
1543 memset(buf, 0, di->blocksize);
1544 memcpy(buf, kdh, hdrsz);
1547 extent = dtoh64(kdh->dumpextent);
1550 error = dump_write(di, kdc->kdc_dumpkey,
1551 di->mediaoffset + di->mediasize - di->blocksize - extent -
1558 error = dump_write(di, buf,
1559 di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1560 keysize, di->blocksize);
1562 error = dump_write(di, buf, di->mediaoffset + di->mediasize -
1563 di->blocksize, di->blocksize);
1568 * Don't touch the first SIZEOF_METADATA bytes on the dump device. This is to
1569 * protect us from metadata and metadata from us.
1571 #define SIZEOF_METADATA (64 * 1024)
1574 * Do some preliminary setup for a kernel dump: initialize state for encryption,
1575 * if requested, and make sure that we have enough space on the dump device.
1577 * We set things up so that the dump ends before the last sector of the dump
1578 * device, at which the trailing header is written.
1580 * +-----------+------+-----+----------------------------+------+
1581 * | | lhdr | key | ... kernel dump ... | thdr |
1582 * +-----------+------+-----+----------------------------+------+
1583 * 1 blk opt <------- dump extent --------> 1 blk
1585 * Dumps written using dump_append() start at the beginning of the extent.
1586 * Uncompressed dumps will use the entire extent, but compressed dumps typically
1587 * will not. The true length of the dump is recorded in the leading and trailing
1588 * headers once the dump has been completed.
1590 * The dump device may provide a callback, in which case it will initialize
1591 * dumpoff and take care of laying out the headers.
1594 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1597 struct kerneldumpcrypto *kdc;
1600 uint64_t dumpextent, span;
1605 /* Send the key before the dump so a partial dump is still usable. */
1607 error = kerneldumpcrypto_init(kdc);
1610 keysize = kerneldumpcrypto_dumpkeysize(kdc);
1611 key = keysize > 0 ? kdc->kdc_dumpkey : NULL;
1618 if (di->dumper_start != NULL) {
1619 error = di->dumper_start(di, key, keysize);
1621 dumpextent = dtoh64(kdh->dumpextent);
1622 span = SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1624 if (di->mediasize < span) {
1625 if (di->kdcomp == NULL)
1629 * We don't yet know how much space the compressed dump
1630 * will occupy, so try to use the whole swap partition
1631 * (minus the first 64KB) in the hope that the
1632 * compressed dump will fit. If that doesn't turn out to
1633 * be enough, the bounds checking in dump_write()
1634 * will catch us and cause the dump to fail.
1636 dumpextent = di->mediasize - span + dumpextent;
1637 kdh->dumpextent = htod64(dumpextent);
1641 * The offset at which to begin writing the dump.
1643 di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1646 di->origdumpoff = di->dumpoff;
1651 _dump_append(struct dumperinfo *di, void *virtual, size_t length)
1656 if (di->kdcrypto != NULL)
1657 error = dump_encrypted_write(di, virtual, di->dumpoff, length);
1660 error = dump_write(di, virtual, di->dumpoff, length);
1662 di->dumpoff += length;
1667 * Write to the dump device starting at dumpoff. When compression is enabled,
1668 * writes to the device will be performed using a callback that gets invoked
1669 * when the compression stream's output buffer is full.
1672 dump_append(struct dumperinfo *di, void *virtual, size_t length)
1676 if (di->kdcomp != NULL) {
1677 /* Bounce through a buffer to avoid CRC errors. */
1678 if (length > di->maxiosize)
1680 buf = di->kdcomp->kdc_buf;
1681 memmove(buf, virtual, length);
1682 return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1684 return (_dump_append(di, virtual, length));
1688 * Write to the dump device at the specified offset.
1691 dump_write(struct dumperinfo *di, void *virtual, off_t offset, size_t length)
1695 error = dump_check_bounds(di, offset, length);
1698 return (di->dumper(di->priv, virtual, offset, length));
1702 * Perform kernel dump finalization: flush the compression stream, if necessary,
1703 * write the leading and trailing kernel dump headers now that we know the true
1704 * length of the dump, and optionally write the encryption key following the
1708 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1712 if (di->kdcomp != NULL) {
1713 error = compressor_flush(di->kdcomp->kdc_stream);
1714 if (error == EAGAIN) {
1715 /* We have residual data in di->blockbuf. */
1716 error = _dump_append(di, di->blockbuf, di->blocksize);
1718 /* Compensate for _dump_append()'s adjustment. */
1719 di->dumpoff -= di->blocksize - di->kdcomp->kdc_resid;
1720 di->kdcomp->kdc_resid = 0;
1726 * We now know the size of the compressed dump, so update the
1727 * header accordingly and recompute parity.
1729 kdh->dumplength = htod64(di->dumpoff - di->origdumpoff);
1731 kdh->parity = kerneldump_parity(kdh);
1733 compressor_reset(di->kdcomp->kdc_stream);
1736 error = dump_write_headers(di, kdh);
1740 (void)dump_write(di, NULL, 0, 0);
1745 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1746 const char *magic, uint32_t archver, uint64_t dumplen)
1750 bzero(kdh, sizeof(*kdh));
1751 strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1752 strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1753 kdh->version = htod32(KERNELDUMPVERSION);
1754 kdh->architectureversion = htod32(archver);
1755 kdh->dumplength = htod64(dumplen);
1756 kdh->dumpextent = kdh->dumplength;
1757 kdh->dumptime = htod64(time_second);
1759 kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1761 kdh->dumpkeysize = 0;
1763 kdh->blocksize = htod32(di->blocksize);
1764 strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1765 dstsize = sizeof(kdh->versionstring);
1766 if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1767 kdh->versionstring[dstsize - 2] = '\n';
1768 if (panicstr != NULL)
1769 strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1770 if (di->kdcomp != NULL)
1771 kdh->compression = di->kdcomp->kdc_format;
1772 kdh->parity = kerneldump_parity(kdh);
1776 DB_SHOW_COMMAND(panic, db_show_panic)
1779 if (panicstr == NULL)
1780 db_printf("panicstr not set\n");
1782 db_printf("panic: %s\n", panicstr);