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. */
243 static 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);
268 EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
269 SHUTDOWN_PRI_LAST + 200);
272 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
275 * The only reason this exists is to create the /dev/reroot/ directory,
276 * used by reroot code in init(8) as a mountpoint for tmpfs.
279 reroot_conf(void *unused)
284 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev,
285 &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot");
287 printf("%s: failed to create device node, error %d",
292 SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL);
295 * The system call that results in a reboot.
299 sys_reboot(struct thread *td, struct reboot_args *uap)
305 error = mac_system_check_reboot(td->td_ucred, uap->opt);
308 error = priv_check(td, PRIV_REBOOT);
310 if (uap->opt & RB_REROOT)
311 error = kern_reroot();
313 kern_reboot(uap->opt);
319 shutdown_nice_task_fn(void *arg, int pending __unused)
323 howto = (uintptr_t)arg;
324 /* Send a signal to init(8) and have it shutdown the world. */
326 if (howto & RB_POWEROFF)
327 kern_psignal(initproc, SIGUSR2);
328 else if (howto & RB_POWERCYCLE)
329 kern_psignal(initproc, SIGWINCH);
330 else if (howto & RB_HALT)
331 kern_psignal(initproc, SIGUSR1);
333 kern_psignal(initproc, SIGINT);
334 PROC_UNLOCK(initproc);
337 static struct task shutdown_nice_task = TASK_INITIALIZER(0,
338 &shutdown_nice_task_fn, NULL);
341 * Called by events that want to shut down.. e.g <CTL><ALT><DEL> on a PC
344 shutdown_nice(int howto)
347 if (initproc != NULL && !SCHEDULER_STOPPED()) {
348 shutdown_nice_task.ta_context = (void *)(uintptr_t)howto;
349 taskqueue_enqueue(taskqueue_fast, &shutdown_nice_task);
352 * No init(8) running, or scheduler would not allow it
353 * to run, so simply reboot.
355 kern_reboot(howto | RB_NOSYNC);
368 if (ts.tv_sec >= 86400) {
369 printf("%ldd", (long)ts.tv_sec / 86400);
373 if (f || ts.tv_sec >= 3600) {
374 printf("%ldh", (long)ts.tv_sec / 3600);
378 if (f || ts.tv_sec >= 60) {
379 printf("%ldm", (long)ts.tv_sec / 60);
383 printf("%lds\n", (long)ts.tv_sec);
387 doadump(boolean_t textdump)
395 if (TAILQ_EMPTY(&dumper_configs))
399 dumptid = curthread->td_tid;
404 if (textdump && textdump_pending) {
406 textdump_dumpsys(TAILQ_FIRST(&dumper_configs));
410 struct dumperinfo *di;
412 TAILQ_FOREACH(di, &dumper_configs, di_next) {
424 * kern_reboot(9): Shut down the system cleanly to prepare for reboot, halt, or
428 kern_reboot(int howto)
433 * Normal paths here don't hold Giant, but we can wind up here
434 * unexpectedly with it held. Drop it now so we don't have to
435 * drop and pick it up elsewhere. The paths it is locking will
436 * never be returned to, and it is preferable to preclude
437 * deadlock than to lock against code that won't ever
440 while (mtx_owned(&Giant))
445 * Bind us to the first CPU so that all shutdown code runs there. Some
446 * systems don't shutdown properly (i.e., ACPI power off) if we
447 * run on another processor.
449 if (!SCHEDULER_STOPPED()) {
450 thread_lock(curthread);
451 sched_bind(curthread, CPU_FIRST());
452 thread_unlock(curthread);
453 KASSERT(PCPU_GET(cpuid) == CPU_FIRST(),
454 ("%s: not running on cpu 0", __func__));
457 /* We're in the process of rebooting. */
460 /* We are out of the debugger now. */
464 * Do any callouts that should be done BEFORE syncing the filesystems.
466 EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
469 * Now sync filesystems
471 if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
473 bufshutdown(show_busybufs);
481 * Ok, now do things that assume all filesystem activity has
484 EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
486 if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
489 /* Now that we're going to really halt the system... */
490 EVENTHANDLER_INVOKE(shutdown_final, howto);
492 for(;;) ; /* safety against shutdown_reset not working */
497 * The system call that results in changing the rootfs.
502 struct vnode *oldrootvnode, *vp;
503 struct mount *mp, *devmp;
506 if (curproc != initproc)
510 * Mark the filesystem containing currently-running executable
511 * (the temporary copy of init(8)) busy.
513 vp = curproc->p_textvp;
514 error = vn_lock(vp, LK_SHARED);
518 error = vfs_busy(mp, MBF_NOWAIT);
522 error = vfs_busy(mp, 0);
523 vn_lock(vp, LK_SHARED | LK_RETRY);
529 if (VN_IS_DOOMED(vp)) {
538 * Remove the filesystem containing currently-running executable
539 * from the mount list, to prevent it from being unmounted
540 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
542 * Also preserve /dev - forcibly unmounting it could cause driver
550 mtx_lock(&mountlist_mtx);
551 TAILQ_REMOVE(&mountlist, mp, mnt_list);
552 TAILQ_REMOVE(&mountlist, devmp, mnt_list);
553 mtx_unlock(&mountlist_mtx);
555 oldrootvnode = rootvnode;
558 * Unmount everything except for the two filesystems preserved above.
563 * Add /dev back; vfs_mountroot() will move it into its new place.
565 mtx_lock(&mountlist_mtx);
566 TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
567 mtx_unlock(&mountlist_mtx);
572 * Mount the new rootfs.
577 * Update all references to the old rootvnode.
579 mountcheckdirs(oldrootvnode, rootvnode);
582 * Add the temporary filesystem back and unbusy it.
584 mtx_lock(&mountlist_mtx);
585 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
586 mtx_unlock(&mountlist_mtx);
593 * If the shutdown was a clean halt, behave accordingly.
596 shutdown_halt(void *junk, int howto)
599 if (howto & RB_HALT) {
601 printf("The operating system has halted.\n");
602 printf("Please press any key to reboot.\n\n");
604 wdog_kern_pat(WD_TO_NEVER);
607 case -1: /* No console, just die */
617 * Check to see if the system panicked, pause and then reboot
618 * according to the specified delay.
621 shutdown_panic(void *junk, int howto)
625 if (howto & RB_DUMP) {
626 if (panic_reboot_wait_time != 0) {
627 if (panic_reboot_wait_time != -1) {
628 printf("Automatic reboot in %d seconds - "
629 "press a key on the console to abort\n",
630 panic_reboot_wait_time);
631 for (loop = panic_reboot_wait_time * 10;
633 DELAY(1000 * 100); /* 1/10th second */
634 /* Did user type a key? */
635 if (cncheckc() != -1)
641 } else { /* zero time specified - reboot NOW */
644 printf("--> Press a key on the console to reboot,\n");
645 printf("--> or switch off the system now.\n");
651 * Everything done, now reset
654 shutdown_reset(void *junk, int howto)
657 printf("Rebooting...\n");
658 DELAY(1000000); /* wait 1 sec for printf's to complete and be read */
661 * Acquiring smp_ipi_mtx here has a double effect:
662 * - it disables interrupts avoiding CPU0 preemption
663 * by fast handlers (thus deadlocking against other CPUs)
664 * - it avoids deadlocks against smp_rendezvous() or, more
665 * generally, threads busy-waiting, with this spinlock held,
666 * and waiting for responses by threads on other CPUs
667 * (ie. smp_tlb_shootdown()).
669 * For the !SMP case it just needs to handle the former problem.
672 mtx_lock_spin(&smp_ipi_mtx);
678 /* NOTREACHED */ /* assuming reset worked */
681 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
682 static int kassert_warn_only = 0;
684 static int kassert_do_kdb = 0;
687 static int kassert_do_ktr = 0;
689 static int kassert_do_log = 1;
690 static int kassert_log_pps_limit = 4;
691 static int kassert_log_mute_at = 0;
692 static int kassert_log_panic_at = 0;
693 static int kassert_suppress_in_panic = 0;
694 static int kassert_warnings = 0;
696 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
699 #ifdef KASSERT_PANIC_OPTIONAL
700 #define KASSERT_RWTUN CTLFLAG_RWTUN
702 #define KASSERT_RWTUN CTLFLAG_RDTUN
705 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, KASSERT_RWTUN,
706 &kassert_warn_only, 0,
707 "KASSERT triggers a panic (0) or just a warning (1)");
710 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, KASSERT_RWTUN,
711 &kassert_do_kdb, 0, "KASSERT will enter the debugger");
715 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, KASSERT_RWTUN,
717 "KASSERT does a KTR, set this to the KTRMASK you want");
720 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, KASSERT_RWTUN,
722 "If warn_only is enabled, log (1) or do not log (0) assertion violations");
724 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RD | CTLFLAG_STATS,
725 &kassert_warnings, 0, "number of KASSERTs that have been triggered");
727 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, KASSERT_RWTUN,
728 &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
730 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, KASSERT_RWTUN,
731 &kassert_log_pps_limit, 0, "limit number of log messages per second");
733 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, KASSERT_RWTUN,
734 &kassert_log_mute_at, 0, "max number of KASSERTS to log");
736 SYSCTL_INT(_debug_kassert, OID_AUTO, suppress_in_panic, KASSERT_RWTUN,
737 &kassert_suppress_in_panic, 0,
738 "KASSERTs will be suppressed while handling a panic");
741 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
743 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
744 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_MPSAFE, NULL, 0,
745 kassert_sysctl_kassert, "I",
746 "set to trigger a test kassert");
749 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
753 error = sysctl_wire_old_buffer(req, sizeof(int));
756 error = sysctl_handle_int(oidp, &i, 0, req);
758 if (error != 0 || req->newptr == NULL)
760 KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
764 #ifdef KASSERT_PANIC_OPTIONAL
766 * Called by KASSERT, this decides if we will panic
767 * or if we will log via printf and/or ktr.
770 kassert_panic(const char *fmt, ...)
772 static char buf[256];
776 (void)vsnprintf(buf, sizeof(buf), fmt, ap);
780 * If we are suppressing secondary panics, log the warning but do not
781 * re-enter panic/kdb.
783 if (panicstr != NULL && kassert_suppress_in_panic) {
784 if (kassert_do_log) {
785 printf("KASSERT failed: %s\n", buf);
787 if (trace_all_panics && trace_on_panic)
795 * panic if we're not just warning, or if we've exceeded
796 * kassert_log_panic_at warnings.
798 if (!kassert_warn_only ||
799 (kassert_log_panic_at > 0 &&
800 kassert_warnings >= kassert_log_panic_at)) {
810 * log if we've not yet met the mute limit.
812 if (kassert_do_log &&
813 (kassert_log_mute_at == 0 ||
814 kassert_warnings < kassert_log_mute_at)) {
815 static struct timeval lasterr;
818 if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
819 printf("KASSERT failed: %s\n", buf);
824 if (kassert_do_kdb) {
825 kdb_enter(KDB_WHY_KASSERT, buf);
828 atomic_add_int(&kassert_warnings, 1);
830 #endif /* KASSERT_PANIC_OPTIONAL */
834 * Panic is called on unresolvable fatal errors. It prints "panic: mesg",
835 * and then reboots. If we are called twice, then we avoid trying to sync
836 * the disks as this often leads to recursive panics.
839 panic(const char *fmt, ...)
848 vpanic(const char *fmt, va_list ap)
853 struct thread *td = curthread;
854 int bootopt, newpanic;
855 static char buf[256];
861 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
862 * concurrently entering panic. Only the winner will proceed
865 if (panicstr == NULL && !kdb_active) {
866 other_cpus = all_cpus;
867 CPU_CLR(PCPU_GET(cpuid), &other_cpus);
868 stop_cpus_hard(other_cpus);
873 * Ensure that the scheduler is stopped while panicking, even if panic
874 * has been entered from kdb.
876 td->td_stopsched = 1;
878 bootopt = RB_AUTOBOOT;
881 bootopt |= RB_NOSYNC;
890 (void)vsnprintf(buf, sizeof(buf), fmt, ap);
893 printf("panic: %s\n", buf);
900 printf("cpuid = %d\n", PCPU_GET(cpuid));
902 printf("time = %jd\n", (intmax_t )time_second);
904 if ((newpanic || trace_all_panics) && trace_on_panic)
906 if (debugger_on_panic)
907 kdb_enter(KDB_WHY_PANIC, "panic");
908 else if (!newpanic && debugger_on_recursive_panic)
909 kdb_enter(KDB_WHY_PANIC, "re-panic");
911 /*thread_lock(td); */
912 td->td_flags |= TDF_INPANIC;
913 /* thread_unlock(td); */
915 bootopt |= RB_NOSYNC;
916 if (poweroff_on_panic)
917 bootopt |= RB_POWEROFF;
918 if (powercycle_on_panic)
919 bootopt |= RB_POWERCYCLE;
920 kern_reboot(bootopt);
924 * Support for poweroff delay.
926 * Please note that setting this delay too short might power off your machine
927 * before the write cache on your hard disk has been flushed, leading to
928 * soft-updates inconsistencies.
930 #ifndef POWEROFF_DELAY
931 # define POWEROFF_DELAY 5000
933 static int poweroff_delay = POWEROFF_DELAY;
935 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
936 &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
939 poweroff_wait(void *junk, int howto)
942 if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
944 DELAY(poweroff_delay * 1000);
948 * Some system processes (e.g. syncer) need to be stopped at appropriate
949 * points in their main loops prior to a system shutdown, so that they
950 * won't interfere with the shutdown process (e.g. by holding a disk buf
951 * to cause sync to fail). For each of these system processes, register
952 * shutdown_kproc() as a handler for one of shutdown events.
954 static int kproc_shutdown_wait = 60;
955 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
956 &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
959 kproc_shutdown(void *arg, int howto)
967 p = (struct proc *)arg;
968 printf("Waiting (max %d seconds) for system process `%s' to stop... ",
969 kproc_shutdown_wait, p->p_comm);
970 error = kproc_suspend(p, kproc_shutdown_wait * hz);
972 if (error == EWOULDBLOCK)
973 printf("timed out\n");
979 kthread_shutdown(void *arg, int howto)
987 td = (struct thread *)arg;
988 printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
989 kproc_shutdown_wait, td->td_name);
990 error = kthread_suspend(td, kproc_shutdown_wait * hz);
992 if (error == EWOULDBLOCK)
993 printf("timed out\n");
999 dumpdevname_sysctl_handler(SYSCTL_HANDLER_ARGS)
1002 struct dumperinfo *di;
1006 error = sysctl_wire_old_buffer(req, 0);
1010 sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1012 mtx_lock(&dumpconf_list_lk);
1013 TAILQ_FOREACH(di, &dumper_configs, di_next) {
1014 if (di != TAILQ_FIRST(&dumper_configs))
1015 sbuf_putc(&sb, ',');
1016 sbuf_cat(&sb, di->di_devname);
1018 mtx_unlock(&dumpconf_list_lk);
1020 error = sbuf_finish(&sb);
1024 SYSCTL_PROC(_kern_shutdown, OID_AUTO, dumpdevname,
1025 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, &dumper_configs, 0,
1026 dumpdevname_sysctl_handler, "A",
1027 "Device(s) for kernel dumps");
1029 static int _dump_append(struct dumperinfo *di, void *virtual,
1030 vm_offset_t physical, 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 * Must not be present on global list.
1173 free_single_dumper(struct dumperinfo *di)
1179 zfree(di->blockbuf, M_DUMPER);
1181 kerneldumpcomp_destroy(di);
1184 zfree(di->kdcrypto, M_EKCD);
1186 zfree(di, M_DUMPER);
1189 /* Registration of dumpers */
1191 dumper_insert(const struct dumperinfo *di_template, const char *devname,
1192 const struct diocskerneldump_arg *kda)
1194 struct dumperinfo *newdi, *listdi;
1199 index = kda->kda_index;
1200 MPASS(index != KDA_REMOVE && index != KDA_REMOVE_DEV &&
1201 index != KDA_REMOVE_ALL);
1203 error = priv_check(curthread, PRIV_SETDUMPER);
1207 newdi = malloc(sizeof(*newdi) + strlen(devname) + 1, M_DUMPER, M_WAITOK
1209 memcpy(newdi, di_template, sizeof(*newdi));
1210 newdi->blockbuf = NULL;
1211 newdi->kdcrypto = NULL;
1212 newdi->kdcomp = NULL;
1213 strcpy(newdi->di_devname, devname);
1215 if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
1217 newdi->kdcrypto = kerneldumpcrypto_create(di_template->blocksize,
1218 kda->kda_encryption, kda->kda_key,
1219 kda->kda_encryptedkeysize, kda->kda_encryptedkey);
1220 if (newdi->kdcrypto == NULL) {
1229 if (kda->kda_compression != KERNELDUMP_COMP_NONE) {
1232 * We can't support simultaneous unpadded block cipher
1233 * encryption and compression because there is no guarantee the
1234 * length of the compressed result is exactly a multiple of the
1235 * cipher block size.
1237 if (kda->kda_encryption == KERNELDUMP_ENC_AES_256_CBC) {
1242 newdi->kdcomp = kerneldumpcomp_create(newdi,
1243 kda->kda_compression);
1244 if (newdi->kdcomp == NULL) {
1250 newdi->blockbuf = malloc(newdi->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1252 /* Add the new configuration to the queue */
1253 mtx_lock(&dumpconf_list_lk);
1255 TAILQ_FOREACH(listdi, &dumper_configs, di_next) {
1257 TAILQ_INSERT_BEFORE(listdi, newdi, di_next);
1264 TAILQ_INSERT_TAIL(&dumper_configs, newdi, di_next);
1265 mtx_unlock(&dumpconf_list_lk);
1270 free_single_dumper(newdi);
1276 dumper_ddb_insert(struct dumperinfo *newdi)
1278 TAILQ_INSERT_HEAD(&dumper_configs, newdi, di_next);
1282 dumper_ddb_remove(struct dumperinfo *di)
1284 TAILQ_REMOVE(&dumper_configs, di, di_next);
1289 dumper_config_match(const struct dumperinfo *di, const char *devname,
1290 const struct diocskerneldump_arg *kda)
1292 if (kda->kda_index == KDA_REMOVE_ALL)
1295 if (strcmp(di->di_devname, devname) != 0)
1299 * Allow wildcard removal of configs matching a device on g_dev_orphan.
1301 if (kda->kda_index == KDA_REMOVE_DEV)
1304 if (di->kdcomp != NULL) {
1305 if (di->kdcomp->kdc_format != kda->kda_compression)
1307 } else if (kda->kda_compression != KERNELDUMP_COMP_NONE)
1310 if (di->kdcrypto != NULL) {
1311 if (di->kdcrypto->kdc_encryption != kda->kda_encryption)
1314 * Do we care to verify keys match to delete? It seems weird
1315 * to expect multiple fallback dump configurations on the same
1316 * device that only differ in crypto key.
1320 if (kda->kda_encryption != KERNELDUMP_ENC_NONE)
1327 dumper_remove(const char *devname, const struct diocskerneldump_arg *kda)
1329 struct dumperinfo *di, *sdi;
1333 error = priv_check(curthread, PRIV_SETDUMPER);
1338 * Try to find a matching configuration, and kill it.
1340 * NULL 'kda' indicates remove any configuration matching 'devname',
1341 * which may remove multiple configurations in atypical configurations.
1344 mtx_lock(&dumpconf_list_lk);
1345 TAILQ_FOREACH_SAFE(di, &dumper_configs, di_next, sdi) {
1346 if (dumper_config_match(di, devname, kda)) {
1348 TAILQ_REMOVE(&dumper_configs, di, di_next);
1349 free_single_dumper(di);
1352 mtx_unlock(&dumpconf_list_lk);
1354 /* Only produce ENOENT if a more targeted match didn't match. */
1355 if (!found && kda->kda_index == KDA_REMOVE)
1361 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1364 if (di->mediasize > 0 && length != 0 && (offset < di->mediaoffset ||
1365 offset - di->mediaoffset + length > di->mediasize)) {
1366 if (di->kdcomp != NULL && offset >= di->mediaoffset) {
1368 "Compressed dump failed to fit in device boundaries.\n");
1372 printf("Attempt to write outside dump device boundaries.\n"
1373 "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1374 (intmax_t)offset, (intmax_t)di->mediaoffset,
1375 (uintmax_t)length, (intmax_t)di->mediasize);
1378 if (length % di->blocksize != 0) {
1379 printf("Attempt to write partial block of length %ju.\n",
1383 if (offset % di->blocksize != 0) {
1384 printf("Attempt to write at unaligned offset %jd.\n",
1394 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1397 switch (kdc->kdc_encryption) {
1398 case KERNELDUMP_ENC_AES_256_CBC:
1399 if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1400 8 * size, buf) <= 0) {
1403 if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1404 buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1408 case KERNELDUMP_ENC_CHACHA20:
1409 chacha_encrypt_bytes(&kdc->kdc_chacha, buf, buf, size);
1418 /* Encrypt data and call dumper. */
1420 dump_encrypted_write(struct dumperinfo *di, void *virtual,
1421 vm_offset_t physical, off_t offset, size_t length)
1423 static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1424 struct kerneldumpcrypto *kdc;
1430 while (length > 0) {
1431 nbytes = MIN(length, sizeof(buf));
1432 bcopy(virtual, buf, nbytes);
1434 if (dump_encrypt(kdc, buf, nbytes) != 0)
1437 error = dump_write(di, buf, physical, offset, nbytes);
1442 virtual = (void *)((uint8_t *)virtual + nbytes);
1451 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1453 struct dumperinfo *di;
1454 size_t resid, rlength;
1459 if (length % di->blocksize != 0) {
1461 * This must be the final write after flushing the compression
1462 * stream. Write as many full blocks as possible and stash the
1463 * residual data in the dumper's block buffer. It will be
1464 * padded and written in dump_finish().
1466 rlength = rounddown(length, di->blocksize);
1468 error = _dump_append(di, base, 0, rlength);
1472 resid = length - rlength;
1473 memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1474 bzero((uint8_t *)di->blockbuf + resid, di->blocksize - resid);
1475 di->kdcomp->kdc_resid = resid;
1478 return (_dump_append(di, base, 0, length));
1482 * Write kernel dump headers at the beginning and end of the dump extent.
1483 * Write the kernel dump encryption key after the leading header if we were
1484 * configured to do so.
1487 dump_write_headers(struct dumperinfo *di, struct kerneldumpheader *kdh)
1490 struct kerneldumpcrypto *kdc;
1498 hdrsz = sizeof(*kdh);
1499 if (hdrsz > di->blocksize)
1504 key = kdc->kdc_dumpkey;
1505 keysize = kerneldumpcrypto_dumpkeysize(kdc);
1512 * If the dump device has special handling for headers, let it take care
1513 * of writing them out.
1515 if (di->dumper_hdr != NULL)
1516 return (di->dumper_hdr(di, kdh, key, keysize));
1518 if (hdrsz == di->blocksize)
1522 memset(buf, 0, di->blocksize);
1523 memcpy(buf, kdh, hdrsz);
1526 extent = dtoh64(kdh->dumpextent);
1529 error = dump_write(di, kdc->kdc_dumpkey, 0,
1530 di->mediaoffset + di->mediasize - di->blocksize - extent -
1537 error = dump_write(di, buf, 0,
1538 di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1539 keysize, di->blocksize);
1541 error = dump_write(di, buf, 0, di->mediaoffset + di->mediasize -
1542 di->blocksize, di->blocksize);
1547 * Don't touch the first SIZEOF_METADATA bytes on the dump device. This is to
1548 * protect us from metadata and metadata from us.
1550 #define SIZEOF_METADATA (64 * 1024)
1553 * Do some preliminary setup for a kernel dump: initialize state for encryption,
1554 * if requested, and make sure that we have enough space on the dump device.
1556 * We set things up so that the dump ends before the last sector of the dump
1557 * device, at which the trailing header is written.
1559 * +-----------+------+-----+----------------------------+------+
1560 * | | lhdr | key | ... kernel dump ... | thdr |
1561 * +-----------+------+-----+----------------------------+------+
1562 * 1 blk opt <------- dump extent --------> 1 blk
1564 * Dumps written using dump_append() start at the beginning of the extent.
1565 * Uncompressed dumps will use the entire extent, but compressed dumps typically
1566 * will not. The true length of the dump is recorded in the leading and trailing
1567 * headers once the dump has been completed.
1569 * The dump device may provide a callback, in which case it will initialize
1570 * dumpoff and take care of laying out the headers.
1573 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1575 uint64_t dumpextent, span;
1580 error = kerneldumpcrypto_init(di->kdcrypto);
1583 keysize = kerneldumpcrypto_dumpkeysize(di->kdcrypto);
1589 if (di->dumper_start != NULL) {
1590 error = di->dumper_start(di);
1592 dumpextent = dtoh64(kdh->dumpextent);
1593 span = SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1595 if (di->mediasize < span) {
1596 if (di->kdcomp == NULL)
1600 * We don't yet know how much space the compressed dump
1601 * will occupy, so try to use the whole swap partition
1602 * (minus the first 64KB) in the hope that the
1603 * compressed dump will fit. If that doesn't turn out to
1604 * be enough, the bounds checking in dump_write()
1605 * will catch us and cause the dump to fail.
1607 dumpextent = di->mediasize - span + dumpextent;
1608 kdh->dumpextent = htod64(dumpextent);
1612 * The offset at which to begin writing the dump.
1614 di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1617 di->origdumpoff = di->dumpoff;
1622 _dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1628 if (di->kdcrypto != NULL)
1629 error = dump_encrypted_write(di, virtual, physical, di->dumpoff,
1633 error = dump_write(di, virtual, physical, di->dumpoff, length);
1635 di->dumpoff += length;
1640 * Write to the dump device starting at dumpoff. When compression is enabled,
1641 * writes to the device will be performed using a callback that gets invoked
1642 * when the compression stream's output buffer is full.
1645 dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1650 if (di->kdcomp != NULL) {
1651 /* Bounce through a buffer to avoid CRC errors. */
1652 if (length > di->maxiosize)
1654 buf = di->kdcomp->kdc_buf;
1655 memmove(buf, virtual, length);
1656 return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1658 return (_dump_append(di, virtual, physical, length));
1662 * Write to the dump device at the specified offset.
1665 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1666 off_t offset, size_t length)
1670 error = dump_check_bounds(di, offset, length);
1673 return (di->dumper(di->priv, virtual, physical, offset, length));
1677 * Perform kernel dump finalization: flush the compression stream, if necessary,
1678 * write the leading and trailing kernel dump headers now that we know the true
1679 * length of the dump, and optionally write the encryption key following the
1683 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1687 if (di->kdcomp != NULL) {
1688 error = compressor_flush(di->kdcomp->kdc_stream);
1689 if (error == EAGAIN) {
1690 /* We have residual data in di->blockbuf. */
1691 error = _dump_append(di, di->blockbuf, 0, di->blocksize);
1693 /* Compensate for _dump_append()'s adjustment. */
1694 di->dumpoff -= di->blocksize - di->kdcomp->kdc_resid;
1695 di->kdcomp->kdc_resid = 0;
1701 * We now know the size of the compressed dump, so update the
1702 * header accordingly and recompute parity.
1704 kdh->dumplength = htod64(di->dumpoff - di->origdumpoff);
1706 kdh->parity = kerneldump_parity(kdh);
1708 compressor_reset(di->kdcomp->kdc_stream);
1711 error = dump_write_headers(di, kdh);
1715 (void)dump_write(di, NULL, 0, 0, 0);
1720 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1721 const char *magic, uint32_t archver, uint64_t dumplen)
1725 bzero(kdh, sizeof(*kdh));
1726 strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1727 strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1728 kdh->version = htod32(KERNELDUMPVERSION);
1729 kdh->architectureversion = htod32(archver);
1730 kdh->dumplength = htod64(dumplen);
1731 kdh->dumpextent = kdh->dumplength;
1732 kdh->dumptime = htod64(time_second);
1734 kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1736 kdh->dumpkeysize = 0;
1738 kdh->blocksize = htod32(di->blocksize);
1739 strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1740 dstsize = sizeof(kdh->versionstring);
1741 if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1742 kdh->versionstring[dstsize - 2] = '\n';
1743 if (panicstr != NULL)
1744 strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1745 if (di->kdcomp != NULL)
1746 kdh->compression = di->kdcomp->kdc_format;
1747 kdh->parity = kerneldump_parity(kdh);
1751 DB_SHOW_COMMAND(panic, db_show_panic)
1754 if (panicstr == NULL)
1755 db_printf("panicstr not set\n");
1757 db_printf("panic: %s\n", panicstr);