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>
40 __FBSDID("$FreeBSD$");
45 #include "opt_panic.h"
46 #include "opt_printf.h"
47 #include "opt_sched.h"
48 #include "opt_watchdog.h"
50 #include <sys/param.h>
51 #include <sys/systm.h>
55 #include <sys/compressor.h>
58 #include <sys/eventhandler.h>
59 #include <sys/filedesc.h>
62 #include <sys/kernel.h>
63 #include <sys/kerneldump.h>
64 #include <sys/kthread.h>
66 #include <sys/malloc.h>
68 #include <sys/mount.h>
71 #include <sys/reboot.h>
72 #include <sys/resourcevar.h>
73 #include <sys/rwlock.h>
75 #include <sys/sched.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysproto.h>
79 #include <sys/taskqueue.h>
80 #include <sys/vnode.h>
81 #include <sys/watchdog.h>
83 #include <crypto/chacha20/chacha.h>
84 #include <crypto/rijndael/rijndael-api-fst.h>
85 #include <crypto/sha2/sha256.h>
89 #include <machine/cpu.h>
90 #include <machine/dump.h>
91 #include <machine/pcb.h>
92 #include <machine/smp.h>
94 #include <security/mac/mac_framework.h>
97 #include <vm/vm_object.h>
98 #include <vm/vm_page.h>
99 #include <vm/vm_pager.h>
100 #include <vm/swap_pager.h>
102 #include <sys/signalvar.h>
104 static MALLOC_DEFINE(M_DUMPER, "dumper", "dumper block buffer");
106 #ifndef PANIC_REBOOT_WAIT_TIME
107 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
109 static int panic_reboot_wait_time = PANIC_REBOOT_WAIT_TIME;
110 SYSCTL_INT(_kern, OID_AUTO, panic_reboot_wait_time, CTLFLAG_RWTUN,
111 &panic_reboot_wait_time, 0,
112 "Seconds to wait before rebooting after a panic");
115 * Note that stdarg.h and the ANSI style va_start macro is used for both
116 * ANSI and traditional C compilers.
118 #include <machine/stdarg.h>
121 #ifdef KDB_UNATTENDED
122 int debugger_on_panic = 0;
124 int debugger_on_panic = 1;
126 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic,
127 CTLFLAG_RWTUN | CTLFLAG_SECURE,
128 &debugger_on_panic, 0, "Run debugger on kernel panic");
130 static bool debugger_on_recursive_panic = false;
131 SYSCTL_BOOL(_debug, OID_AUTO, debugger_on_recursive_panic,
132 CTLFLAG_RWTUN | CTLFLAG_SECURE,
133 &debugger_on_recursive_panic, 0, "Run debugger on recursive kernel panic");
135 int debugger_on_trap = 0;
136 SYSCTL_INT(_debug, OID_AUTO, debugger_on_trap,
137 CTLFLAG_RWTUN | CTLFLAG_SECURE,
138 &debugger_on_trap, 0, "Run debugger on kernel trap before panic");
141 static int trace_on_panic = 1;
142 static bool trace_all_panics = true;
144 static int trace_on_panic = 0;
145 static bool trace_all_panics = false;
147 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
148 CTLFLAG_RWTUN | CTLFLAG_SECURE,
149 &trace_on_panic, 0, "Print stack trace on kernel panic");
150 SYSCTL_BOOL(_debug, OID_AUTO, trace_all_panics, CTLFLAG_RWTUN,
151 &trace_all_panics, 0, "Print stack traces on secondary kernel panics");
154 static int sync_on_panic = 0;
155 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN,
156 &sync_on_panic, 0, "Do a sync before rebooting from a panic");
158 static bool poweroff_on_panic = 0;
159 SYSCTL_BOOL(_kern, OID_AUTO, poweroff_on_panic, CTLFLAG_RWTUN,
160 &poweroff_on_panic, 0, "Do a power off instead of a reboot on a panic");
162 static bool powercycle_on_panic = 0;
163 SYSCTL_BOOL(_kern, OID_AUTO, powercycle_on_panic, CTLFLAG_RWTUN,
164 &powercycle_on_panic, 0, "Do a power cycle instead of a reboot on a panic");
166 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
167 "Shutdown environment");
170 static int show_busybufs;
172 static int show_busybufs = 1;
174 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
176 "Show busy buffers during shutdown");
178 int suspend_blocked = 0;
179 SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW,
180 &suspend_blocked, 0, "Block suspend due to a pending shutdown");
183 FEATURE(ekcd, "Encrypted kernel crash dumps support");
185 MALLOC_DEFINE(M_EKCD, "ekcd", "Encrypted kernel crash dumps data");
187 struct kerneldumpcrypto {
188 uint8_t kdc_encryption;
189 uint8_t kdc_iv[KERNELDUMP_IV_MAX_SIZE];
193 cipherInstance aes_ci;
195 struct chacha_ctx u_chacha;
197 #define kdc_ki u.u_aes.aes_ki
198 #define kdc_ci u.u_aes.aes_ci
199 #define kdc_chacha u.u_chacha
200 uint32_t kdc_dumpkeysize;
201 struct kerneldumpkey kdc_dumpkey[];
205 struct kerneldumpcomp {
207 struct compressor *kdc_stream;
212 static struct kerneldumpcomp *kerneldumpcomp_create(struct dumperinfo *di,
213 uint8_t compression);
214 static void kerneldumpcomp_destroy(struct dumperinfo *di);
215 static int kerneldumpcomp_write_cb(void *base, size_t len, off_t off, void *arg);
217 static int kerneldump_gzlevel = 6;
218 SYSCTL_INT(_kern, OID_AUTO, kerneldump_gzlevel, CTLFLAG_RWTUN,
219 &kerneldump_gzlevel, 0,
220 "Kernel crash dump compression level");
223 * Variable panicstr contains argument to first call to panic; used as flag
224 * to indicate that the kernel has already called panic.
226 const char *panicstr;
227 bool __read_frequently panicked;
229 int __read_mostly dumping; /* system is dumping */
230 int rebooting; /* system is rebooting */
232 * Used to serialize between sysctl kern.shutdown.dumpdevname and list
233 * modifications via ioctl.
235 static struct mtx dumpconf_list_lk;
236 MTX_SYSINIT(dumper_configs, &dumpconf_list_lk, "dumper config list", MTX_DEF);
238 /* Our selected dumper(s). */
239 static TAILQ_HEAD(dumpconflist, dumperinfo) dumper_configs =
240 TAILQ_HEAD_INITIALIZER(dumper_configs);
242 /* Context information for dump-debuggers, saved by the dump_savectx() macro. */
243 struct pcb dumppcb; /* Registers. */
244 lwpid_t dumptid; /* Thread ID. */
246 static struct cdevsw reroot_cdevsw = {
247 .d_version = D_VERSION,
251 static void poweroff_wait(void *, int);
252 static void shutdown_halt(void *junk, int howto);
253 static void shutdown_panic(void *junk, int howto);
254 static void shutdown_reset(void *junk, int howto);
255 static int kern_reroot(void);
257 /* register various local shutdown events */
259 shutdown_conf(void *unused)
262 EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
264 EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
265 SHUTDOWN_PRI_LAST + 100);
266 EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
267 SHUTDOWN_PRI_LAST + 100);
270 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
273 * The only reason this exists is to create the /dev/reroot/ directory,
274 * used by reroot code in init(8) as a mountpoint for tmpfs.
277 reroot_conf(void *unused)
282 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev,
283 &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot");
285 printf("%s: failed to create device node, error %d",
290 SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL);
293 * The system call that results in a reboot.
297 sys_reboot(struct thread *td, struct reboot_args *uap)
303 error = mac_system_check_reboot(td->td_ucred, uap->opt);
306 error = priv_check(td, PRIV_REBOOT);
308 if (uap->opt & RB_REROOT)
309 error = kern_reroot();
311 kern_reboot(uap->opt);
317 shutdown_nice_task_fn(void *arg, int pending __unused)
321 howto = (uintptr_t)arg;
322 /* Send a signal to init(8) and have it shutdown the world. */
324 if (howto & RB_POWEROFF)
325 kern_psignal(initproc, SIGUSR2);
326 else if (howto & RB_POWERCYCLE)
327 kern_psignal(initproc, SIGWINCH);
328 else if (howto & RB_HALT)
329 kern_psignal(initproc, SIGUSR1);
331 kern_psignal(initproc, SIGINT);
332 PROC_UNLOCK(initproc);
335 static struct task shutdown_nice_task = TASK_INITIALIZER(0,
336 &shutdown_nice_task_fn, NULL);
339 * Called by events that want to shut down.. e.g <CTL><ALT><DEL> on a PC
342 shutdown_nice(int howto)
345 if (initproc != NULL && !SCHEDULER_STOPPED()) {
346 shutdown_nice_task.ta_context = (void *)(uintptr_t)howto;
347 taskqueue_enqueue(taskqueue_fast, &shutdown_nice_task);
350 * No init(8) running, or scheduler would not allow it
351 * to run, so simply reboot.
353 kern_reboot(howto | RB_NOSYNC);
366 if (ts.tv_sec >= 86400) {
367 printf("%ldd", (long)ts.tv_sec / 86400);
371 if (f || ts.tv_sec >= 3600) {
372 printf("%ldh", (long)ts.tv_sec / 3600);
376 if (f || ts.tv_sec >= 60) {
377 printf("%ldm", (long)ts.tv_sec / 60);
381 printf("%lds\n", (long)ts.tv_sec);
385 doadump(boolean_t textdump)
393 if (TAILQ_EMPTY(&dumper_configs))
401 if (textdump && textdump_pending) {
403 textdump_dumpsys(TAILQ_FIRST(&dumper_configs));
407 struct dumperinfo *di;
409 TAILQ_FOREACH(di, &dumper_configs, di_next) {
421 * kern_reboot(9): Shut down the system cleanly to prepare for reboot, halt, or
425 kern_reboot(int howto)
430 * Normal paths here don't hold Giant, but we can wind up here
431 * unexpectedly with it held. Drop it now so we don't have to
432 * drop and pick it up elsewhere. The paths it is locking will
433 * never be returned to, and it is preferable to preclude
434 * deadlock than to lock against code that won't ever
437 while (mtx_owned(&Giant))
442 * Bind us to the first CPU so that all shutdown code runs there. Some
443 * systems don't shutdown properly (i.e., ACPI power off) if we
444 * run on another processor.
446 if (!SCHEDULER_STOPPED()) {
447 thread_lock(curthread);
448 sched_bind(curthread, CPU_FIRST());
449 thread_unlock(curthread);
450 KASSERT(PCPU_GET(cpuid) == CPU_FIRST(),
451 ("%s: not running on cpu 0", __func__));
454 /* We're in the process of rebooting. */
457 /* We are out of the debugger now. */
461 * Do any callouts that should be done BEFORE syncing the filesystems.
463 EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
466 * Now sync filesystems
468 if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
470 bufshutdown(show_busybufs);
478 * Ok, now do things that assume all filesystem activity has
481 EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
483 if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
486 /* Now that we're going to really halt the system... */
487 EVENTHANDLER_INVOKE(shutdown_final, howto);
490 * Call this directly so that reset is attempted even if shutdown
491 * handlers are not yet registered.
493 shutdown_reset(NULL, howto);
495 for(;;) ; /* safety against shutdown_reset not working */
500 * The system call that results in changing the rootfs.
505 struct vnode *oldrootvnode, *vp;
506 struct mount *mp, *devmp;
509 if (curproc != initproc)
513 * Mark the filesystem containing currently-running executable
514 * (the temporary copy of init(8)) busy.
516 vp = curproc->p_textvp;
517 error = vn_lock(vp, LK_SHARED);
521 error = vfs_busy(mp, MBF_NOWAIT);
525 error = vfs_busy(mp, 0);
526 vn_lock(vp, LK_SHARED | LK_RETRY);
532 if (VN_IS_DOOMED(vp)) {
541 * Remove the filesystem containing currently-running executable
542 * from the mount list, to prevent it from being unmounted
543 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
545 * Also preserve /dev - forcibly unmounting it could cause driver
553 mtx_lock(&mountlist_mtx);
554 TAILQ_REMOVE(&mountlist, mp, mnt_list);
555 TAILQ_REMOVE(&mountlist, devmp, mnt_list);
556 mtx_unlock(&mountlist_mtx);
558 oldrootvnode = rootvnode;
561 * Unmount everything except for the two filesystems preserved above.
566 * Add /dev back; vfs_mountroot() will move it into its new place.
568 mtx_lock(&mountlist_mtx);
569 TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
570 mtx_unlock(&mountlist_mtx);
575 * Mount the new rootfs.
580 * Update all references to the old rootvnode.
582 mountcheckdirs(oldrootvnode, rootvnode);
585 * Add the temporary filesystem back and unbusy it.
587 mtx_lock(&mountlist_mtx);
588 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
589 mtx_unlock(&mountlist_mtx);
596 * If the shutdown was a clean halt, behave accordingly.
599 shutdown_halt(void *junk, int howto)
602 if (howto & RB_HALT) {
604 printf("The operating system has halted.\n");
605 printf("Please press any key to reboot.\n\n");
607 wdog_kern_pat(WD_TO_NEVER);
610 case -1: /* No console, just die */
620 * Check to see if the system panicked, pause and then reboot
621 * according to the specified delay.
624 shutdown_panic(void *junk, int howto)
628 if (howto & RB_DUMP) {
629 if (panic_reboot_wait_time != 0) {
630 if (panic_reboot_wait_time != -1) {
631 printf("Automatic reboot in %d seconds - "
632 "press a key on the console to abort\n",
633 panic_reboot_wait_time);
634 for (loop = panic_reboot_wait_time * 10;
636 DELAY(1000 * 100); /* 1/10th second */
637 /* Did user type a key? */
638 if (cncheckc() != -1)
644 } else { /* zero time specified - reboot NOW */
647 printf("--> Press a key on the console to reboot,\n");
648 printf("--> or switch off the system now.\n");
654 * Everything done, now reset
657 shutdown_reset(void *junk, int howto)
660 printf("Rebooting...\n");
661 DELAY(1000000); /* wait 1 sec for printf's to complete and be read */
664 * Acquiring smp_ipi_mtx here has a double effect:
665 * - it disables interrupts avoiding CPU0 preemption
666 * by fast handlers (thus deadlocking against other CPUs)
667 * - it avoids deadlocks against smp_rendezvous() or, more
668 * generally, threads busy-waiting, with this spinlock held,
669 * and waiting for responses by threads on other CPUs
670 * (ie. smp_tlb_shootdown()).
672 * For the !SMP case it just needs to handle the former problem.
675 mtx_lock_spin(&smp_ipi_mtx);
681 /* NOTREACHED */ /* assuming reset worked */
684 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
685 static int kassert_warn_only = 0;
687 static int kassert_do_kdb = 0;
690 static int kassert_do_ktr = 0;
692 static int kassert_do_log = 1;
693 static int kassert_log_pps_limit = 4;
694 static int kassert_log_mute_at = 0;
695 static int kassert_log_panic_at = 0;
696 static int kassert_suppress_in_panic = 0;
697 static int kassert_warnings = 0;
699 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
702 #ifdef KASSERT_PANIC_OPTIONAL
703 #define KASSERT_RWTUN CTLFLAG_RWTUN
705 #define KASSERT_RWTUN CTLFLAG_RDTUN
708 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, KASSERT_RWTUN,
709 &kassert_warn_only, 0,
710 "KASSERT triggers a panic (0) or just a warning (1)");
713 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, KASSERT_RWTUN,
714 &kassert_do_kdb, 0, "KASSERT will enter the debugger");
718 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, KASSERT_RWTUN,
720 "KASSERT does a KTR, set this to the KTRMASK you want");
723 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, KASSERT_RWTUN,
725 "If warn_only is enabled, log (1) or do not log (0) assertion violations");
727 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RD | CTLFLAG_STATS,
728 &kassert_warnings, 0, "number of KASSERTs that have been triggered");
730 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, KASSERT_RWTUN,
731 &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
733 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, KASSERT_RWTUN,
734 &kassert_log_pps_limit, 0, "limit number of log messages per second");
736 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, KASSERT_RWTUN,
737 &kassert_log_mute_at, 0, "max number of KASSERTS to log");
739 SYSCTL_INT(_debug_kassert, OID_AUTO, suppress_in_panic, KASSERT_RWTUN,
740 &kassert_suppress_in_panic, 0,
741 "KASSERTs will be suppressed while handling a panic");
744 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
746 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
747 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_MPSAFE, NULL, 0,
748 kassert_sysctl_kassert, "I",
749 "set to trigger a test kassert");
752 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
756 error = sysctl_wire_old_buffer(req, sizeof(int));
759 error = sysctl_handle_int(oidp, &i, 0, req);
761 if (error != 0 || req->newptr == NULL)
763 KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
767 #ifdef KASSERT_PANIC_OPTIONAL
769 * Called by KASSERT, this decides if we will panic
770 * or if we will log via printf and/or ktr.
773 kassert_panic(const char *fmt, ...)
775 static char buf[256];
779 (void)vsnprintf(buf, sizeof(buf), fmt, ap);
783 * If we are suppressing secondary panics, log the warning but do not
784 * re-enter panic/kdb.
786 if (KERNEL_PANICKED() && kassert_suppress_in_panic) {
787 if (kassert_do_log) {
788 printf("KASSERT failed: %s\n", buf);
790 if (trace_all_panics && trace_on_panic)
798 * panic if we're not just warning, or if we've exceeded
799 * kassert_log_panic_at warnings.
801 if (!kassert_warn_only ||
802 (kassert_log_panic_at > 0 &&
803 kassert_warnings >= kassert_log_panic_at)) {
813 * log if we've not yet met the mute limit.
815 if (kassert_do_log &&
816 (kassert_log_mute_at == 0 ||
817 kassert_warnings < kassert_log_mute_at)) {
818 static struct timeval lasterr;
821 if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
822 printf("KASSERT failed: %s\n", buf);
827 if (kassert_do_kdb) {
828 kdb_enter(KDB_WHY_KASSERT, buf);
831 atomic_add_int(&kassert_warnings, 1);
833 #endif /* KASSERT_PANIC_OPTIONAL */
837 * Panic is called on unresolvable fatal errors. It prints "panic: mesg",
838 * and then reboots. If we are called twice, then we avoid trying to sync
839 * the disks as this often leads to recursive panics.
842 panic(const char *fmt, ...)
851 vpanic(const char *fmt, va_list ap)
856 struct thread *td = curthread;
857 int bootopt, newpanic;
858 static char buf[256];
864 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
865 * concurrently entering panic. Only the winner will proceed
868 if (panicstr == NULL && !kdb_active) {
869 other_cpus = all_cpus;
870 CPU_CLR(PCPU_GET(cpuid), &other_cpus);
871 stop_cpus_hard(other_cpus);
876 * Ensure that the scheduler is stopped while panicking, even if panic
877 * has been entered from kdb.
879 td->td_stopsched = 1;
881 bootopt = RB_AUTOBOOT;
883 if (KERNEL_PANICKED())
884 bootopt |= RB_NOSYNC;
893 (void)vsnprintf(buf, sizeof(buf), fmt, ap);
896 printf("panic: %s\n", buf);
903 printf("cpuid = %d\n", PCPU_GET(cpuid));
905 printf("time = %jd\n", (intmax_t )time_second);
907 if ((newpanic || trace_all_panics) && trace_on_panic)
909 if (debugger_on_panic)
910 kdb_enter(KDB_WHY_PANIC, "panic");
911 else if (!newpanic && debugger_on_recursive_panic)
912 kdb_enter(KDB_WHY_PANIC, "re-panic");
914 /*thread_lock(td); */
915 td->td_flags |= TDF_INPANIC;
916 /* thread_unlock(td); */
918 bootopt |= RB_NOSYNC;
919 if (poweroff_on_panic)
920 bootopt |= RB_POWEROFF;
921 if (powercycle_on_panic)
922 bootopt |= RB_POWERCYCLE;
923 kern_reboot(bootopt);
927 * Support for poweroff delay.
929 * Please note that setting this delay too short might power off your machine
930 * before the write cache on your hard disk has been flushed, leading to
931 * soft-updates inconsistencies.
933 #ifndef POWEROFF_DELAY
934 # define POWEROFF_DELAY 5000
936 static int poweroff_delay = POWEROFF_DELAY;
938 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
939 &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
942 poweroff_wait(void *junk, int howto)
945 if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
947 DELAY(poweroff_delay * 1000);
951 * Some system processes (e.g. syncer) need to be stopped at appropriate
952 * points in their main loops prior to a system shutdown, so that they
953 * won't interfere with the shutdown process (e.g. by holding a disk buf
954 * to cause sync to fail). For each of these system processes, register
955 * shutdown_kproc() as a handler for one of shutdown events.
957 static int kproc_shutdown_wait = 60;
958 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
959 &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
962 kproc_shutdown(void *arg, int howto)
967 if (KERNEL_PANICKED())
970 p = (struct proc *)arg;
971 printf("Waiting (max %d seconds) for system process `%s' to stop... ",
972 kproc_shutdown_wait, p->p_comm);
973 error = kproc_suspend(p, kproc_shutdown_wait * hz);
975 if (error == EWOULDBLOCK)
976 printf("timed out\n");
982 kthread_shutdown(void *arg, int howto)
987 if (KERNEL_PANICKED())
990 td = (struct thread *)arg;
991 printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
992 kproc_shutdown_wait, td->td_name);
993 error = kthread_suspend(td, kproc_shutdown_wait * hz);
995 if (error == EWOULDBLOCK)
996 printf("timed out\n");
1002 dumpdevname_sysctl_handler(SYSCTL_HANDLER_ARGS)
1005 struct dumperinfo *di;
1009 error = sysctl_wire_old_buffer(req, 0);
1013 sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1015 mtx_lock(&dumpconf_list_lk);
1016 TAILQ_FOREACH(di, &dumper_configs, di_next) {
1017 if (di != TAILQ_FIRST(&dumper_configs))
1018 sbuf_putc(&sb, ',');
1019 sbuf_cat(&sb, di->di_devname);
1021 mtx_unlock(&dumpconf_list_lk);
1023 error = sbuf_finish(&sb);
1027 SYSCTL_PROC(_kern_shutdown, OID_AUTO, dumpdevname,
1028 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, &dumper_configs, 0,
1029 dumpdevname_sysctl_handler, "A",
1030 "Device(s) for kernel dumps");
1032 static int _dump_append(struct dumperinfo *di, void *virtual, size_t length);
1035 static struct kerneldumpcrypto *
1036 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
1037 const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1039 struct kerneldumpcrypto *kdc;
1040 struct kerneldumpkey *kdk;
1041 uint32_t dumpkeysize;
1043 dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
1044 kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
1046 arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
1048 kdc->kdc_encryption = encryption;
1049 switch (kdc->kdc_encryption) {
1050 case KERNELDUMP_ENC_AES_256_CBC:
1051 if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
1054 case KERNELDUMP_ENC_CHACHA20:
1055 chacha_keysetup(&kdc->kdc_chacha, key, 256);
1061 kdc->kdc_dumpkeysize = dumpkeysize;
1062 kdk = kdc->kdc_dumpkey;
1063 kdk->kdk_encryption = kdc->kdc_encryption;
1064 memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1065 kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
1066 memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
1075 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
1077 uint8_t hash[SHA256_DIGEST_LENGTH];
1079 struct kerneldumpkey *kdk;
1088 * When a user enters ddb it can write a crash dump multiple times.
1089 * Each time it should be encrypted using a different IV.
1092 SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1093 SHA256_Final(hash, &ctx);
1094 bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1096 switch (kdc->kdc_encryption) {
1097 case KERNELDUMP_ENC_AES_256_CBC:
1098 if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1099 kdc->kdc_iv) <= 0) {
1104 case KERNELDUMP_ENC_CHACHA20:
1105 chacha_ivsetup(&kdc->kdc_chacha, kdc->kdc_iv, NULL);
1112 kdk = kdc->kdc_dumpkey;
1113 memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1115 explicit_bzero(hash, sizeof(hash));
1120 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
1125 return (kdc->kdc_dumpkeysize);
1129 static struct kerneldumpcomp *
1130 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression)
1132 struct kerneldumpcomp *kdcomp;
1135 switch (compression) {
1136 case KERNELDUMP_COMP_GZIP:
1137 format = COMPRESS_GZIP;
1139 case KERNELDUMP_COMP_ZSTD:
1140 format = COMPRESS_ZSTD;
1146 kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO);
1147 kdcomp->kdc_format = compression;
1148 kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb,
1149 format, di->maxiosize, kerneldump_gzlevel, di);
1150 if (kdcomp->kdc_stream == NULL) {
1151 free(kdcomp, M_DUMPER);
1154 kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1159 kerneldumpcomp_destroy(struct dumperinfo *di)
1161 struct kerneldumpcomp *kdcomp;
1163 kdcomp = di->kdcomp;
1166 compressor_fini(kdcomp->kdc_stream);
1167 zfree(kdcomp->kdc_buf, M_DUMPER);
1168 free(kdcomp, M_DUMPER);
1172 * Free a dumper. Must not be present on global list.
1175 dumper_destroy(struct dumperinfo *di)
1181 zfree(di->blockbuf, M_DUMPER);
1182 kerneldumpcomp_destroy(di);
1184 zfree(di->kdcrypto, M_EKCD);
1186 zfree(di, M_DUMPER);
1190 * Allocate and set up a new dumper from the provided template.
1193 dumper_create(const struct dumperinfo *di_template, const char *devname,
1194 const struct diocskerneldump_arg *kda, struct dumperinfo **dip)
1196 struct dumperinfo *newdi;
1202 /* Allocate a new dumper */
1203 newdi = malloc(sizeof(*newdi) + strlen(devname) + 1, M_DUMPER,
1205 memcpy(newdi, di_template, sizeof(*newdi));
1206 newdi->blockbuf = NULL;
1207 newdi->kdcrypto = NULL;
1208 newdi->kdcomp = NULL;
1209 strcpy(newdi->di_devname, devname);
1211 if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
1213 newdi->kdcrypto = kerneldumpcrypto_create(newdi->blocksize,
1214 kda->kda_encryption, kda->kda_key,
1215 kda->kda_encryptedkeysize, kda->kda_encryptedkey);
1216 if (newdi->kdcrypto == NULL) {
1225 if (kda->kda_compression != KERNELDUMP_COMP_NONE) {
1228 * We can't support simultaneous unpadded block cipher
1229 * encryption and compression because there is no guarantee the
1230 * length of the compressed result is exactly a multiple of the
1231 * cipher block size.
1233 if (kda->kda_encryption == KERNELDUMP_ENC_AES_256_CBC) {
1238 newdi->kdcomp = kerneldumpcomp_create(newdi,
1239 kda->kda_compression);
1240 if (newdi->kdcomp == NULL) {
1245 newdi->blockbuf = malloc(newdi->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1250 dumper_destroy(newdi);
1255 * Create a new dumper and register it in the global list.
1258 dumper_insert(const struct dumperinfo *di_template, const char *devname,
1259 const struct diocskerneldump_arg *kda)
1261 struct dumperinfo *newdi, *listdi;
1266 index = kda->kda_index;
1267 MPASS(index != KDA_REMOVE && index != KDA_REMOVE_DEV &&
1268 index != KDA_REMOVE_ALL);
1270 error = priv_check(curthread, PRIV_SETDUMPER);
1274 error = dumper_create(di_template, devname, kda, &newdi);
1278 /* Add the new configuration to the queue */
1279 mtx_lock(&dumpconf_list_lk);
1281 TAILQ_FOREACH(listdi, &dumper_configs, di_next) {
1283 TAILQ_INSERT_BEFORE(listdi, newdi, di_next);
1290 TAILQ_INSERT_TAIL(&dumper_configs, newdi, di_next);
1291 mtx_unlock(&dumpconf_list_lk);
1298 dumper_ddb_insert(struct dumperinfo *newdi)
1300 TAILQ_INSERT_HEAD(&dumper_configs, newdi, di_next);
1304 dumper_ddb_remove(struct dumperinfo *di)
1306 TAILQ_REMOVE(&dumper_configs, di, di_next);
1311 dumper_config_match(const struct dumperinfo *di, const char *devname,
1312 const struct diocskerneldump_arg *kda)
1314 if (kda->kda_index == KDA_REMOVE_ALL)
1317 if (strcmp(di->di_devname, devname) != 0)
1321 * Allow wildcard removal of configs matching a device on g_dev_orphan.
1323 if (kda->kda_index == KDA_REMOVE_DEV)
1326 if (di->kdcomp != NULL) {
1327 if (di->kdcomp->kdc_format != kda->kda_compression)
1329 } else if (kda->kda_compression != KERNELDUMP_COMP_NONE)
1332 if (di->kdcrypto != NULL) {
1333 if (di->kdcrypto->kdc_encryption != kda->kda_encryption)
1336 * Do we care to verify keys match to delete? It seems weird
1337 * to expect multiple fallback dump configurations on the same
1338 * device that only differ in crypto key.
1342 if (kda->kda_encryption != KERNELDUMP_ENC_NONE)
1349 * Remove and free the requested dumper(s) from the global list.
1352 dumper_remove(const char *devname, const struct diocskerneldump_arg *kda)
1354 struct dumperinfo *di, *sdi;
1358 error = priv_check(curthread, PRIV_SETDUMPER);
1363 * Try to find a matching configuration, and kill it.
1365 * NULL 'kda' indicates remove any configuration matching 'devname',
1366 * which may remove multiple configurations in atypical configurations.
1369 mtx_lock(&dumpconf_list_lk);
1370 TAILQ_FOREACH_SAFE(di, &dumper_configs, di_next, sdi) {
1371 if (dumper_config_match(di, devname, kda)) {
1373 TAILQ_REMOVE(&dumper_configs, di, di_next);
1377 mtx_unlock(&dumpconf_list_lk);
1379 /* Only produce ENOENT if a more targeted match didn't match. */
1380 if (!found && kda->kda_index == KDA_REMOVE)
1386 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1389 if (di->mediasize > 0 && length != 0 && (offset < di->mediaoffset ||
1390 offset - di->mediaoffset + length > di->mediasize)) {
1391 if (di->kdcomp != NULL && offset >= di->mediaoffset) {
1393 "Compressed dump failed to fit in device boundaries.\n");
1397 printf("Attempt to write outside dump device boundaries.\n"
1398 "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1399 (intmax_t)offset, (intmax_t)di->mediaoffset,
1400 (uintmax_t)length, (intmax_t)di->mediasize);
1403 if (length % di->blocksize != 0) {
1404 printf("Attempt to write partial block of length %ju.\n",
1408 if (offset % di->blocksize != 0) {
1409 printf("Attempt to write at unaligned offset %jd.\n",
1419 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1422 switch (kdc->kdc_encryption) {
1423 case KERNELDUMP_ENC_AES_256_CBC:
1424 if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1425 8 * size, buf) <= 0) {
1428 if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1429 buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1433 case KERNELDUMP_ENC_CHACHA20:
1434 chacha_encrypt_bytes(&kdc->kdc_chacha, buf, buf, size);
1443 /* Encrypt data and call dumper. */
1445 dump_encrypted_write(struct dumperinfo *di, void *virtual, off_t offset,
1448 static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1449 struct kerneldumpcrypto *kdc;
1455 while (length > 0) {
1456 nbytes = MIN(length, sizeof(buf));
1457 bcopy(virtual, buf, nbytes);
1459 if (dump_encrypt(kdc, buf, nbytes) != 0)
1462 error = dump_write(di, buf, offset, nbytes);
1467 virtual = (void *)((uint8_t *)virtual + nbytes);
1476 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1478 struct dumperinfo *di;
1479 size_t resid, rlength;
1484 if (length % di->blocksize != 0) {
1486 * This must be the final write after flushing the compression
1487 * stream. Write as many full blocks as possible and stash the
1488 * residual data in the dumper's block buffer. It will be
1489 * padded and written in dump_finish().
1491 rlength = rounddown(length, di->blocksize);
1493 error = _dump_append(di, base, rlength);
1497 resid = length - rlength;
1498 memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1499 bzero((uint8_t *)di->blockbuf + resid, di->blocksize - resid);
1500 di->kdcomp->kdc_resid = resid;
1503 return (_dump_append(di, base, length));
1507 * Write kernel dump headers at the beginning and end of the dump extent.
1508 * Write the kernel dump encryption key after the leading header if we were
1509 * configured to do so.
1512 dump_write_headers(struct dumperinfo *di, struct kerneldumpheader *kdh)
1515 struct kerneldumpcrypto *kdc;
1523 hdrsz = sizeof(*kdh);
1524 if (hdrsz > di->blocksize)
1529 keysize = kerneldumpcrypto_dumpkeysize(kdc);
1535 * If the dump device has special handling for headers, let it take care
1536 * of writing them out.
1538 if (di->dumper_hdr != NULL)
1539 return (di->dumper_hdr(di, kdh));
1541 if (hdrsz == di->blocksize)
1545 memset(buf, 0, di->blocksize);
1546 memcpy(buf, kdh, hdrsz);
1549 extent = dtoh64(kdh->dumpextent);
1552 error = dump_write(di, kdc->kdc_dumpkey,
1553 di->mediaoffset + di->mediasize - di->blocksize - extent -
1560 error = dump_write(di, buf,
1561 di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1562 keysize, di->blocksize);
1564 error = dump_write(di, buf, di->mediaoffset + di->mediasize -
1565 di->blocksize, di->blocksize);
1570 * Don't touch the first SIZEOF_METADATA bytes on the dump device. This is to
1571 * protect us from metadata and metadata from us.
1573 #define SIZEOF_METADATA (64 * 1024)
1576 * Do some preliminary setup for a kernel dump: initialize state for encryption,
1577 * if requested, and make sure that we have enough space on the dump device.
1579 * We set things up so that the dump ends before the last sector of the dump
1580 * device, at which the trailing header is written.
1582 * +-----------+------+-----+----------------------------+------+
1583 * | | lhdr | key | ... kernel dump ... | thdr |
1584 * +-----------+------+-----+----------------------------+------+
1585 * 1 blk opt <------- dump extent --------> 1 blk
1587 * Dumps written using dump_append() start at the beginning of the extent.
1588 * Uncompressed dumps will use the entire extent, but compressed dumps typically
1589 * will not. The true length of the dump is recorded in the leading and trailing
1590 * headers once the dump has been completed.
1592 * The dump device may provide a callback, in which case it will initialize
1593 * dumpoff and take care of laying out the headers.
1596 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1599 struct kerneldumpcrypto *kdc;
1602 uint64_t dumpextent, span;
1607 /* Send the key before the dump so a partial dump is still usable. */
1609 error = kerneldumpcrypto_init(kdc);
1612 keysize = kerneldumpcrypto_dumpkeysize(kdc);
1613 key = keysize > 0 ? kdc->kdc_dumpkey : NULL;
1620 if (di->dumper_start != NULL) {
1621 error = di->dumper_start(di, key, keysize);
1623 dumpextent = dtoh64(kdh->dumpextent);
1624 span = SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1626 if (di->mediasize < span) {
1627 if (di->kdcomp == NULL)
1631 * We don't yet know how much space the compressed dump
1632 * will occupy, so try to use the whole swap partition
1633 * (minus the first 64KB) in the hope that the
1634 * compressed dump will fit. If that doesn't turn out to
1635 * be enough, the bounds checking in dump_write()
1636 * will catch us and cause the dump to fail.
1638 dumpextent = di->mediasize - span + dumpextent;
1639 kdh->dumpextent = htod64(dumpextent);
1643 * The offset at which to begin writing the dump.
1645 di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1648 di->origdumpoff = di->dumpoff;
1653 _dump_append(struct dumperinfo *di, void *virtual, size_t length)
1658 if (di->kdcrypto != NULL)
1659 error = dump_encrypted_write(di, virtual, di->dumpoff, length);
1662 error = dump_write(di, virtual, di->dumpoff, length);
1664 di->dumpoff += length;
1669 * Write to the dump device starting at dumpoff. When compression is enabled,
1670 * writes to the device will be performed using a callback that gets invoked
1671 * when the compression stream's output buffer is full.
1674 dump_append(struct dumperinfo *di, void *virtual, size_t length)
1678 if (di->kdcomp != NULL) {
1679 /* Bounce through a buffer to avoid CRC errors. */
1680 if (length > di->maxiosize)
1682 buf = di->kdcomp->kdc_buf;
1683 memmove(buf, virtual, length);
1684 return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1686 return (_dump_append(di, virtual, length));
1690 * Write to the dump device at the specified offset.
1693 dump_write(struct dumperinfo *di, void *virtual, off_t offset, size_t length)
1697 error = dump_check_bounds(di, offset, length);
1700 return (di->dumper(di->priv, virtual, offset, length));
1704 * Perform kernel dump finalization: flush the compression stream, if necessary,
1705 * write the leading and trailing kernel dump headers now that we know the true
1706 * length of the dump, and optionally write the encryption key following the
1710 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1714 if (di->kdcomp != NULL) {
1715 error = compressor_flush(di->kdcomp->kdc_stream);
1716 if (error == EAGAIN) {
1717 /* We have residual data in di->blockbuf. */
1718 error = _dump_append(di, di->blockbuf, di->blocksize);
1720 /* Compensate for _dump_append()'s adjustment. */
1721 di->dumpoff -= di->blocksize - di->kdcomp->kdc_resid;
1722 di->kdcomp->kdc_resid = 0;
1728 * We now know the size of the compressed dump, so update the
1729 * header accordingly and recompute parity.
1731 kdh->dumplength = htod64(di->dumpoff - di->origdumpoff);
1733 kdh->parity = kerneldump_parity(kdh);
1735 compressor_reset(di->kdcomp->kdc_stream);
1738 error = dump_write_headers(di, kdh);
1742 (void)dump_write(di, NULL, 0, 0);
1747 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1748 const char *magic, uint32_t archver, uint64_t dumplen)
1752 bzero(kdh, sizeof(*kdh));
1753 strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1754 strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1755 kdh->version = htod32(KERNELDUMPVERSION);
1756 kdh->architectureversion = htod32(archver);
1757 kdh->dumplength = htod64(dumplen);
1758 kdh->dumpextent = kdh->dumplength;
1759 kdh->dumptime = htod64(time_second);
1761 kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1763 kdh->dumpkeysize = 0;
1765 kdh->blocksize = htod32(di->blocksize);
1766 strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1767 dstsize = sizeof(kdh->versionstring);
1768 if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1769 kdh->versionstring[dstsize - 2] = '\n';
1770 if (panicstr != NULL)
1771 strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1772 if (di->kdcomp != NULL)
1773 kdh->compression = di->kdcomp->kdc_format;
1774 kdh->parity = kerneldump_parity(kdh);
1778 DB_SHOW_COMMAND(panic, db_show_panic)
1781 if (panicstr == NULL)
1782 db_printf("panicstr not set\n");
1784 db_printf("panic: %s\n", panicstr);