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/rijndael/rijndael-api-fst.h>
84 #include <crypto/sha2/sha256.h>
88 #include <machine/cpu.h>
89 #include <machine/dump.h>
90 #include <machine/pcb.h>
91 #include <machine/smp.h>
93 #include <security/mac/mac_framework.h>
96 #include <vm/vm_object.h>
97 #include <vm/vm_page.h>
98 #include <vm/vm_pager.h>
99 #include <vm/swap_pager.h>
101 #include <sys/signalvar.h>
103 static MALLOC_DEFINE(M_DUMPER, "dumper", "dumper block buffer");
105 #ifndef PANIC_REBOOT_WAIT_TIME
106 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
108 static int panic_reboot_wait_time = PANIC_REBOOT_WAIT_TIME;
109 SYSCTL_INT(_kern, OID_AUTO, panic_reboot_wait_time, CTLFLAG_RWTUN,
110 &panic_reboot_wait_time, 0,
111 "Seconds to wait before rebooting after a panic");
114 * Note that stdarg.h and the ANSI style va_start macro is used for both
115 * ANSI and traditional C compilers.
117 #include <machine/stdarg.h>
120 #ifdef KDB_UNATTENDED
121 static int debugger_on_panic = 0;
123 static int debugger_on_panic = 1;
125 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic,
126 CTLFLAG_RWTUN | CTLFLAG_SECURE,
127 &debugger_on_panic, 0, "Run debugger on kernel panic");
129 int debugger_on_trap = 0;
130 SYSCTL_INT(_debug, OID_AUTO, debugger_on_trap,
131 CTLFLAG_RWTUN | CTLFLAG_SECURE,
132 &debugger_on_trap, 0, "Run debugger on kernel trap before panic");
135 static int trace_on_panic = 1;
136 static bool trace_all_panics = true;
138 static int trace_on_panic = 0;
139 static bool trace_all_panics = false;
141 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
142 CTLFLAG_RWTUN | CTLFLAG_SECURE,
143 &trace_on_panic, 0, "Print stack trace on kernel panic");
144 SYSCTL_BOOL(_debug, OID_AUTO, trace_all_panics, CTLFLAG_RWTUN,
145 &trace_all_panics, 0, "Print stack traces on secondary kernel panics");
148 static int sync_on_panic = 0;
149 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN,
150 &sync_on_panic, 0, "Do a sync before rebooting from a panic");
152 static bool poweroff_on_panic = 0;
153 SYSCTL_BOOL(_kern, OID_AUTO, poweroff_on_panic, CTLFLAG_RWTUN,
154 &poweroff_on_panic, 0, "Do a power off instead of a reboot on a panic");
156 static bool powercycle_on_panic = 0;
157 SYSCTL_BOOL(_kern, OID_AUTO, powercycle_on_panic, CTLFLAG_RWTUN,
158 &powercycle_on_panic, 0, "Do a power cycle instead of a reboot on a panic");
160 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0,
161 "Shutdown environment");
164 static int show_busybufs;
166 static int show_busybufs = 1;
168 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
169 &show_busybufs, 0, "");
171 int suspend_blocked = 0;
172 SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW,
173 &suspend_blocked, 0, "Block suspend due to a pending shutdown");
176 FEATURE(ekcd, "Encrypted kernel crash dumps support");
178 MALLOC_DEFINE(M_EKCD, "ekcd", "Encrypted kernel crash dumps data");
180 struct kerneldumpcrypto {
181 uint8_t kdc_encryption;
182 uint8_t kdc_iv[KERNELDUMP_IV_MAX_SIZE];
184 cipherInstance kdc_ci;
185 uint32_t kdc_dumpkeysize;
186 struct kerneldumpkey kdc_dumpkey[];
190 struct kerneldumpcomp {
192 struct compressor *kdc_stream;
197 static struct kerneldumpcomp *kerneldumpcomp_create(struct dumperinfo *di,
198 uint8_t compression);
199 static void kerneldumpcomp_destroy(struct dumperinfo *di);
200 static int kerneldumpcomp_write_cb(void *base, size_t len, off_t off, void *arg);
202 static int kerneldump_gzlevel = 6;
203 SYSCTL_INT(_kern, OID_AUTO, kerneldump_gzlevel, CTLFLAG_RWTUN,
204 &kerneldump_gzlevel, 0,
205 "Kernel crash dump compression level");
208 * Variable panicstr contains argument to first call to panic; used as flag
209 * to indicate that the kernel has already called panic.
211 const char *panicstr;
213 int dumping; /* system is dumping */
214 int rebooting; /* system is rebooting */
216 * Used to serialize between sysctl kern.shutdown.dumpdevname and list
217 * modifications via ioctl.
219 static struct mtx dumpconf_list_lk;
220 MTX_SYSINIT(dumper_configs, &dumpconf_list_lk, "dumper config list", MTX_DEF);
222 /* Our selected dumper(s). */
223 static TAILQ_HEAD(dumpconflist, dumperinfo) dumper_configs =
224 TAILQ_HEAD_INITIALIZER(dumper_configs);
226 /* Context information for dump-debuggers. */
227 static struct pcb dumppcb; /* Registers. */
228 lwpid_t dumptid; /* Thread ID. */
230 static struct cdevsw reroot_cdevsw = {
231 .d_version = D_VERSION,
235 static void poweroff_wait(void *, int);
236 static void shutdown_halt(void *junk, int howto);
237 static void shutdown_panic(void *junk, int howto);
238 static void shutdown_reset(void *junk, int howto);
239 static int kern_reroot(void);
241 /* register various local shutdown events */
243 shutdown_conf(void *unused)
246 EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
248 EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
249 SHUTDOWN_PRI_LAST + 100);
250 EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
251 SHUTDOWN_PRI_LAST + 100);
252 EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
253 SHUTDOWN_PRI_LAST + 200);
256 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
259 * The only reason this exists is to create the /dev/reroot/ directory,
260 * used by reroot code in init(8) as a mountpoint for tmpfs.
263 reroot_conf(void *unused)
268 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev,
269 &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot");
271 printf("%s: failed to create device node, error %d",
276 SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL);
279 * The system call that results in a reboot.
283 sys_reboot(struct thread *td, struct reboot_args *uap)
289 error = mac_system_check_reboot(td->td_ucred, uap->opt);
292 error = priv_check(td, PRIV_REBOOT);
294 if (uap->opt & RB_REROOT)
295 error = kern_reroot();
297 kern_reboot(uap->opt);
303 shutdown_nice_task_fn(void *arg, int pending __unused)
307 howto = (uintptr_t)arg;
308 /* Send a signal to init(8) and have it shutdown the world. */
310 if (howto & RB_POWEROFF)
311 kern_psignal(initproc, SIGUSR2);
312 else if (howto & RB_POWERCYCLE)
313 kern_psignal(initproc, SIGWINCH);
314 else if (howto & RB_HALT)
315 kern_psignal(initproc, SIGUSR1);
317 kern_psignal(initproc, SIGINT);
318 PROC_UNLOCK(initproc);
321 static struct task shutdown_nice_task = TASK_INITIALIZER(0,
322 &shutdown_nice_task_fn, NULL);
325 * Called by events that want to shut down.. e.g <CTL><ALT><DEL> on a PC
328 shutdown_nice(int howto)
331 if (initproc != NULL && !SCHEDULER_STOPPED()) {
332 shutdown_nice_task.ta_context = (void *)(uintptr_t)howto;
333 taskqueue_enqueue(taskqueue_fast, &shutdown_nice_task);
336 * No init(8) running, or scheduler would not allow it
337 * to run, so simply reboot.
339 kern_reboot(howto | RB_NOSYNC);
352 if (ts.tv_sec >= 86400) {
353 printf("%ldd", (long)ts.tv_sec / 86400);
357 if (f || ts.tv_sec >= 3600) {
358 printf("%ldh", (long)ts.tv_sec / 3600);
362 if (f || ts.tv_sec >= 60) {
363 printf("%ldm", (long)ts.tv_sec / 60);
367 printf("%lds\n", (long)ts.tv_sec);
371 doadump(boolean_t textdump)
379 if (TAILQ_EMPTY(&dumper_configs))
383 dumptid = curthread->td_tid;
388 if (textdump && textdump_pending) {
390 textdump_dumpsys(TAILQ_FIRST(&dumper_configs));
394 struct dumperinfo *di;
396 TAILQ_FOREACH(di, &dumper_configs, di_next) {
408 * Shutdown the system cleanly to prepare for reboot, halt, or power off.
411 kern_reboot(int howto)
416 * Normal paths here don't hold Giant, but we can wind up here
417 * unexpectedly with it held. Drop it now so we don't have to
418 * drop and pick it up elsewhere. The paths it is locking will
419 * never be returned to, and it is preferable to preclude
420 * deadlock than to lock against code that won't ever
423 while (mtx_owned(&Giant))
428 * Bind us to the first CPU so that all shutdown code runs there. Some
429 * systems don't shutdown properly (i.e., ACPI power off) if we
430 * run on another processor.
432 if (!SCHEDULER_STOPPED()) {
433 thread_lock(curthread);
434 sched_bind(curthread, CPU_FIRST());
435 thread_unlock(curthread);
436 KASSERT(PCPU_GET(cpuid) == CPU_FIRST(),
437 ("boot: not running on cpu 0"));
440 /* We're in the process of rebooting. */
443 /* We are out of the debugger now. */
447 * Do any callouts that should be done BEFORE syncing the filesystems.
449 EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
452 * Now sync filesystems
454 if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
456 bufshutdown(show_busybufs);
464 * Ok, now do things that assume all filesystem activity has
467 EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
469 if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
472 /* Now that we're going to really halt the system... */
473 EVENTHANDLER_INVOKE(shutdown_final, howto);
475 for(;;) ; /* safety against shutdown_reset not working */
480 * The system call that results in changing the rootfs.
485 struct vnode *oldrootvnode, *vp;
486 struct mount *mp, *devmp;
489 if (curproc != initproc)
493 * Mark the filesystem containing currently-running executable
494 * (the temporary copy of init(8)) busy.
496 vp = curproc->p_textvp;
497 error = vn_lock(vp, LK_SHARED);
501 error = vfs_busy(mp, MBF_NOWAIT);
505 error = vfs_busy(mp, 0);
506 vn_lock(vp, LK_SHARED | LK_RETRY);
512 if (vp->v_iflag & VI_DOOMED) {
521 * Remove the filesystem containing currently-running executable
522 * from the mount list, to prevent it from being unmounted
523 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
525 * Also preserve /dev - forcibly unmounting it could cause driver
533 mtx_lock(&mountlist_mtx);
534 TAILQ_REMOVE(&mountlist, mp, mnt_list);
535 TAILQ_REMOVE(&mountlist, devmp, mnt_list);
536 mtx_unlock(&mountlist_mtx);
538 oldrootvnode = rootvnode;
541 * Unmount everything except for the two filesystems preserved above.
546 * Add /dev back; vfs_mountroot() will move it into its new place.
548 mtx_lock(&mountlist_mtx);
549 TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
550 mtx_unlock(&mountlist_mtx);
555 * Mount the new rootfs.
560 * Update all references to the old rootvnode.
562 mountcheckdirs(oldrootvnode, rootvnode);
565 * Add the temporary filesystem back and unbusy it.
567 mtx_lock(&mountlist_mtx);
568 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
569 mtx_unlock(&mountlist_mtx);
576 * If the shutdown was a clean halt, behave accordingly.
579 shutdown_halt(void *junk, int howto)
582 if (howto & RB_HALT) {
584 printf("The operating system has halted.\n");
585 printf("Please press any key to reboot.\n\n");
587 case -1: /* No console, just die */
597 * Check to see if the system paniced, pause and then reboot
598 * according to the specified delay.
601 shutdown_panic(void *junk, int howto)
605 if (howto & RB_DUMP) {
606 if (panic_reboot_wait_time != 0) {
607 if (panic_reboot_wait_time != -1) {
608 printf("Automatic reboot in %d seconds - "
609 "press a key on the console to abort\n",
610 panic_reboot_wait_time);
611 for (loop = panic_reboot_wait_time * 10;
613 DELAY(1000 * 100); /* 1/10th second */
614 /* Did user type a key? */
615 if (cncheckc() != -1)
621 } else { /* zero time specified - reboot NOW */
624 printf("--> Press a key on the console to reboot,\n");
625 printf("--> or switch off the system now.\n");
631 * Everything done, now reset
634 shutdown_reset(void *junk, int howto)
637 printf("Rebooting...\n");
638 DELAY(1000000); /* wait 1 sec for printf's to complete and be read */
641 * Acquiring smp_ipi_mtx here has a double effect:
642 * - it disables interrupts avoiding CPU0 preemption
643 * by fast handlers (thus deadlocking against other CPUs)
644 * - it avoids deadlocks against smp_rendezvous() or, more
645 * generally, threads busy-waiting, with this spinlock held,
646 * and waiting for responses by threads on other CPUs
647 * (ie. smp_tlb_shootdown()).
649 * For the !SMP case it just needs to handle the former problem.
652 mtx_lock_spin(&smp_ipi_mtx);
657 /* cpu_boot(howto); */ /* doesn't do anything at the moment */
659 /* NOTREACHED */ /* assuming reset worked */
662 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
663 static int kassert_warn_only = 0;
665 static int kassert_do_kdb = 0;
668 static int kassert_do_ktr = 0;
670 static int kassert_do_log = 1;
671 static int kassert_log_pps_limit = 4;
672 static int kassert_log_mute_at = 0;
673 static int kassert_log_panic_at = 0;
674 static int kassert_suppress_in_panic = 0;
675 static int kassert_warnings = 0;
677 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW, NULL, "kassert options");
679 #ifdef KASSERT_PANIC_OPTIONAL
680 #define KASSERT_RWTUN CTLFLAG_RWTUN
682 #define KASSERT_RWTUN CTLFLAG_RDTUN
685 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, KASSERT_RWTUN,
686 &kassert_warn_only, 0,
687 "KASSERT triggers a panic (0) or just a warning (1)");
690 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, KASSERT_RWTUN,
691 &kassert_do_kdb, 0, "KASSERT will enter the debugger");
695 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, KASSERT_RWTUN,
697 "KASSERT does a KTR, set this to the KTRMASK you want");
700 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, KASSERT_RWTUN,
702 "If warn_only is enabled, log (1) or do not log (0) assertion violations");
704 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, KASSERT_RWTUN,
705 &kassert_warnings, 0, "number of KASSERTs that have been triggered");
707 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, KASSERT_RWTUN,
708 &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
710 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, KASSERT_RWTUN,
711 &kassert_log_pps_limit, 0, "limit number of log messages per second");
713 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, KASSERT_RWTUN,
714 &kassert_log_mute_at, 0, "max number of KASSERTS to log");
716 SYSCTL_INT(_debug_kassert, OID_AUTO, suppress_in_panic, KASSERT_RWTUN,
717 &kassert_suppress_in_panic, 0,
718 "KASSERTs will be suppressed while handling a panic");
721 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
723 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
724 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE, NULL, 0,
725 kassert_sysctl_kassert, "I", "set to trigger a test kassert");
728 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
732 error = sysctl_wire_old_buffer(req, sizeof(int));
735 error = sysctl_handle_int(oidp, &i, 0, req);
737 if (error != 0 || req->newptr == NULL)
739 KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
743 #ifdef KASSERT_PANIC_OPTIONAL
745 * Called by KASSERT, this decides if we will panic
746 * or if we will log via printf and/or ktr.
749 kassert_panic(const char *fmt, ...)
751 static char buf[256];
755 (void)vsnprintf(buf, sizeof(buf), fmt, ap);
759 * If we are suppressing secondary panics, log the warning but do not
760 * re-enter panic/kdb.
762 if (panicstr != NULL && kassert_suppress_in_panic) {
763 if (kassert_do_log) {
764 printf("KASSERT failed: %s\n", buf);
766 if (trace_all_panics && trace_on_panic)
774 * panic if we're not just warning, or if we've exceeded
775 * kassert_log_panic_at warnings.
777 if (!kassert_warn_only ||
778 (kassert_log_panic_at > 0 &&
779 kassert_warnings >= kassert_log_panic_at)) {
789 * log if we've not yet met the mute limit.
791 if (kassert_do_log &&
792 (kassert_log_mute_at == 0 ||
793 kassert_warnings < kassert_log_mute_at)) {
794 static struct timeval lasterr;
797 if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
798 printf("KASSERT failed: %s\n", buf);
803 if (kassert_do_kdb) {
804 kdb_enter(KDB_WHY_KASSERT, buf);
807 atomic_add_int(&kassert_warnings, 1);
809 #endif /* KASSERT_PANIC_OPTIONAL */
813 * Panic is called on unresolvable fatal errors. It prints "panic: mesg",
814 * and then reboots. If we are called twice, then we avoid trying to sync
815 * the disks as this often leads to recursive panics.
818 panic(const char *fmt, ...)
827 vpanic(const char *fmt, va_list ap)
832 struct thread *td = curthread;
833 int bootopt, newpanic;
834 static char buf[256];
840 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
841 * concurrently entering panic. Only the winner will proceed
844 if (panicstr == NULL && !kdb_active) {
845 other_cpus = all_cpus;
846 CPU_CLR(PCPU_GET(cpuid), &other_cpus);
847 stop_cpus_hard(other_cpus);
852 * Ensure that the scheduler is stopped while panicking, even if panic
853 * has been entered from kdb.
855 td->td_stopsched = 1;
857 bootopt = RB_AUTOBOOT;
860 bootopt |= RB_NOSYNC;
868 (void)vsnprintf(buf, sizeof(buf), fmt, ap);
871 printf("panic: %s\n", buf);
878 printf("cpuid = %d\n", PCPU_GET(cpuid));
880 printf("time = %jd\n", (intmax_t )time_second);
882 if ((newpanic || trace_all_panics) && trace_on_panic)
884 if (debugger_on_panic)
885 kdb_enter(KDB_WHY_PANIC, "panic");
887 /*thread_lock(td); */
888 td->td_flags |= TDF_INPANIC;
889 /* thread_unlock(td); */
891 bootopt |= RB_NOSYNC;
892 if (poweroff_on_panic)
893 bootopt |= RB_POWEROFF;
894 if (powercycle_on_panic)
895 bootopt |= RB_POWERCYCLE;
896 kern_reboot(bootopt);
900 * Support for poweroff delay.
902 * Please note that setting this delay too short might power off your machine
903 * before the write cache on your hard disk has been flushed, leading to
904 * soft-updates inconsistencies.
906 #ifndef POWEROFF_DELAY
907 # define POWEROFF_DELAY 5000
909 static int poweroff_delay = POWEROFF_DELAY;
911 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
912 &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
915 poweroff_wait(void *junk, int howto)
918 if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
920 DELAY(poweroff_delay * 1000);
924 * Some system processes (e.g. syncer) need to be stopped at appropriate
925 * points in their main loops prior to a system shutdown, so that they
926 * won't interfere with the shutdown process (e.g. by holding a disk buf
927 * to cause sync to fail). For each of these system processes, register
928 * shutdown_kproc() as a handler for one of shutdown events.
930 static int kproc_shutdown_wait = 60;
931 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
932 &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
935 kproc_shutdown(void *arg, int howto)
943 p = (struct proc *)arg;
944 printf("Waiting (max %d seconds) for system process `%s' to stop... ",
945 kproc_shutdown_wait, p->p_comm);
946 error = kproc_suspend(p, kproc_shutdown_wait * hz);
948 if (error == EWOULDBLOCK)
949 printf("timed out\n");
955 kthread_shutdown(void *arg, int howto)
963 td = (struct thread *)arg;
964 printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
965 kproc_shutdown_wait, td->td_name);
966 error = kthread_suspend(td, kproc_shutdown_wait * hz);
968 if (error == EWOULDBLOCK)
969 printf("timed out\n");
975 dumpdevname_sysctl_handler(SYSCTL_HANDLER_ARGS)
978 struct dumperinfo *di;
982 error = sysctl_wire_old_buffer(req, 0);
986 sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
988 mtx_lock(&dumpconf_list_lk);
989 TAILQ_FOREACH(di, &dumper_configs, di_next) {
990 if (di != TAILQ_FIRST(&dumper_configs))
992 sbuf_cat(&sb, di->di_devname);
994 mtx_unlock(&dumpconf_list_lk);
996 error = sbuf_finish(&sb);
1000 SYSCTL_PROC(_kern_shutdown, OID_AUTO, dumpdevname, CTLTYPE_STRING | CTLFLAG_RD,
1001 &dumper_configs, 0, dumpdevname_sysctl_handler, "A",
1002 "Device(s) for kernel dumps");
1004 static int _dump_append(struct dumperinfo *di, void *virtual,
1005 vm_offset_t physical, size_t length);
1008 static struct kerneldumpcrypto *
1009 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
1010 const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1012 struct kerneldumpcrypto *kdc;
1013 struct kerneldumpkey *kdk;
1014 uint32_t dumpkeysize;
1016 dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
1017 kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
1019 arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
1021 kdc->kdc_encryption = encryption;
1022 switch (kdc->kdc_encryption) {
1023 case KERNELDUMP_ENC_AES_256_CBC:
1024 if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
1031 kdc->kdc_dumpkeysize = dumpkeysize;
1032 kdk = kdc->kdc_dumpkey;
1033 kdk->kdk_encryption = kdc->kdc_encryption;
1034 memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1035 kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
1036 memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
1040 explicit_bzero(kdc, sizeof(*kdc) + dumpkeysize);
1046 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
1048 uint8_t hash[SHA256_DIGEST_LENGTH];
1050 struct kerneldumpkey *kdk;
1059 * When a user enters ddb it can write a crash dump multiple times.
1060 * Each time it should be encrypted using a different IV.
1063 SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1064 SHA256_Final(hash, &ctx);
1065 bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1067 switch (kdc->kdc_encryption) {
1068 case KERNELDUMP_ENC_AES_256_CBC:
1069 if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1070 kdc->kdc_iv) <= 0) {
1080 kdk = kdc->kdc_dumpkey;
1081 memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1083 explicit_bzero(hash, sizeof(hash));
1088 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
1093 return (kdc->kdc_dumpkeysize);
1097 static struct kerneldumpcomp *
1098 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression)
1100 struct kerneldumpcomp *kdcomp;
1103 switch (compression) {
1104 case KERNELDUMP_COMP_GZIP:
1105 format = COMPRESS_GZIP;
1107 case KERNELDUMP_COMP_ZSTD:
1108 format = COMPRESS_ZSTD;
1114 kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO);
1115 kdcomp->kdc_format = compression;
1116 kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb,
1117 format, di->maxiosize, kerneldump_gzlevel, di);
1118 if (kdcomp->kdc_stream == NULL) {
1119 free(kdcomp, M_DUMPER);
1122 kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1127 kerneldumpcomp_destroy(struct dumperinfo *di)
1129 struct kerneldumpcomp *kdcomp;
1131 kdcomp = di->kdcomp;
1134 compressor_fini(kdcomp->kdc_stream);
1135 explicit_bzero(kdcomp->kdc_buf, di->maxiosize);
1136 free(kdcomp->kdc_buf, M_DUMPER);
1137 free(kdcomp, M_DUMPER);
1141 * Must not be present on global list.
1144 free_single_dumper(struct dumperinfo *di)
1150 if (di->blockbuf != NULL) {
1151 explicit_bzero(di->blockbuf, di->blocksize);
1152 free(di->blockbuf, M_DUMPER);
1155 kerneldumpcomp_destroy(di);
1158 if (di->kdcrypto != NULL) {
1159 explicit_bzero(di->kdcrypto, sizeof(*di->kdcrypto) +
1160 di->kdcrypto->kdc_dumpkeysize);
1161 free(di->kdcrypto, M_EKCD);
1165 explicit_bzero(di, sizeof(*di));
1169 /* Registration of dumpers */
1171 dumper_insert(const struct dumperinfo *di_template, const char *devname,
1172 const struct diocskerneldump_arg *kda)
1174 struct dumperinfo *newdi, *listdi;
1179 index = kda->kda_index;
1180 MPASS(index != KDA_REMOVE && index != KDA_REMOVE_DEV &&
1181 index != KDA_REMOVE_ALL);
1183 error = priv_check(curthread, PRIV_SETDUMPER);
1187 newdi = malloc(sizeof(*newdi) + strlen(devname) + 1, M_DUMPER, M_WAITOK
1189 memcpy(newdi, di_template, sizeof(*newdi));
1190 newdi->blockbuf = NULL;
1191 newdi->kdcrypto = NULL;
1192 newdi->kdcomp = NULL;
1193 strcpy(newdi->di_devname, devname);
1195 if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
1197 newdi->kdcrypto = kerneldumpcrypto_create(di_template->blocksize,
1198 kda->kda_encryption, kda->kda_key,
1199 kda->kda_encryptedkeysize, kda->kda_encryptedkey);
1200 if (newdi->kdcrypto == NULL) {
1209 if (kda->kda_compression != KERNELDUMP_COMP_NONE) {
1211 * We currently can't support simultaneous encryption and
1212 * compression because our only encryption mode is an unpadded
1213 * block cipher, go figure. This is low hanging fruit to fix.
1215 if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
1219 newdi->kdcomp = kerneldumpcomp_create(newdi,
1220 kda->kda_compression);
1221 if (newdi->kdcomp == NULL) {
1227 newdi->blockbuf = malloc(newdi->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1229 /* Add the new configuration to the queue */
1230 mtx_lock(&dumpconf_list_lk);
1232 TAILQ_FOREACH(listdi, &dumper_configs, di_next) {
1234 TAILQ_INSERT_BEFORE(listdi, newdi, di_next);
1241 TAILQ_INSERT_TAIL(&dumper_configs, newdi, di_next);
1242 mtx_unlock(&dumpconf_list_lk);
1247 free_single_dumper(newdi);
1252 dumper_config_match(const struct dumperinfo *di, const char *devname,
1253 const struct diocskerneldump_arg *kda)
1255 if (kda->kda_index == KDA_REMOVE_ALL)
1258 if (strcmp(di->di_devname, devname) != 0)
1262 * Allow wildcard removal of configs matching a device on g_dev_orphan.
1264 if (kda->kda_index == KDA_REMOVE_DEV)
1267 if (di->kdcomp != NULL) {
1268 if (di->kdcomp->kdc_format != kda->kda_compression)
1270 } else if (kda->kda_compression != KERNELDUMP_COMP_NONE)
1273 if (di->kdcrypto != NULL) {
1274 if (di->kdcrypto->kdc_encryption != kda->kda_encryption)
1277 * Do we care to verify keys match to delete? It seems weird
1278 * to expect multiple fallback dump configurations on the same
1279 * device that only differ in crypto key.
1283 if (kda->kda_encryption != KERNELDUMP_ENC_NONE)
1290 dumper_remove(const char *devname, const struct diocskerneldump_arg *kda)
1292 struct dumperinfo *di, *sdi;
1296 error = priv_check(curthread, PRIV_SETDUMPER);
1301 * Try to find a matching configuration, and kill it.
1303 * NULL 'kda' indicates remove any configuration matching 'devname',
1304 * which may remove multiple configurations in atypical configurations.
1307 mtx_lock(&dumpconf_list_lk);
1308 TAILQ_FOREACH_SAFE(di, &dumper_configs, di_next, sdi) {
1309 if (dumper_config_match(di, devname, kda)) {
1311 TAILQ_REMOVE(&dumper_configs, di, di_next);
1312 free_single_dumper(di);
1315 mtx_unlock(&dumpconf_list_lk);
1317 /* Only produce ENOENT if a more targeted match didn't match. */
1318 if (!found && kda->kda_index == KDA_REMOVE)
1324 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1327 if (di->mediasize > 0 && length != 0 && (offset < di->mediaoffset ||
1328 offset - di->mediaoffset + length > di->mediasize)) {
1329 if (di->kdcomp != NULL && offset >= di->mediaoffset) {
1331 "Compressed dump failed to fit in device boundaries.\n");
1335 printf("Attempt to write outside dump device boundaries.\n"
1336 "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1337 (intmax_t)offset, (intmax_t)di->mediaoffset,
1338 (uintmax_t)length, (intmax_t)di->mediasize);
1341 if (length % di->blocksize != 0) {
1342 printf("Attempt to write partial block of length %ju.\n",
1346 if (offset % di->blocksize != 0) {
1347 printf("Attempt to write at unaligned offset %jd.\n",
1357 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1360 switch (kdc->kdc_encryption) {
1361 case KERNELDUMP_ENC_AES_256_CBC:
1362 if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1363 8 * size, buf) <= 0) {
1366 if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1367 buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1378 /* Encrypt data and call dumper. */
1380 dump_encrypted_write(struct dumperinfo *di, void *virtual,
1381 vm_offset_t physical, off_t offset, size_t length)
1383 static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1384 struct kerneldumpcrypto *kdc;
1390 while (length > 0) {
1391 nbytes = MIN(length, sizeof(buf));
1392 bcopy(virtual, buf, nbytes);
1394 if (dump_encrypt(kdc, buf, nbytes) != 0)
1397 error = dump_write(di, buf, physical, offset, nbytes);
1402 virtual = (void *)((uint8_t *)virtual + nbytes);
1411 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1413 struct dumperinfo *di;
1414 size_t resid, rlength;
1419 if (length % di->blocksize != 0) {
1421 * This must be the final write after flushing the compression
1422 * stream. Write as many full blocks as possible and stash the
1423 * residual data in the dumper's block buffer. It will be
1424 * padded and written in dump_finish().
1426 rlength = rounddown(length, di->blocksize);
1428 error = _dump_append(di, base, 0, rlength);
1432 resid = length - rlength;
1433 memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1434 di->kdcomp->kdc_resid = resid;
1437 return (_dump_append(di, base, 0, length));
1441 * Write kernel dump headers at the beginning and end of the dump extent.
1442 * Write the kernel dump encryption key after the leading header if we were
1443 * configured to do so.
1446 dump_write_headers(struct dumperinfo *di, struct kerneldumpheader *kdh)
1449 struct kerneldumpcrypto *kdc;
1457 hdrsz = sizeof(*kdh);
1458 if (hdrsz > di->blocksize)
1463 key = kdc->kdc_dumpkey;
1464 keysize = kerneldumpcrypto_dumpkeysize(kdc);
1471 * If the dump device has special handling for headers, let it take care
1472 * of writing them out.
1474 if (di->dumper_hdr != NULL)
1475 return (di->dumper_hdr(di, kdh, key, keysize));
1477 if (hdrsz == di->blocksize)
1481 memset(buf, 0, di->blocksize);
1482 memcpy(buf, kdh, hdrsz);
1485 extent = dtoh64(kdh->dumpextent);
1488 error = dump_write(di, kdc->kdc_dumpkey, 0,
1489 di->mediaoffset + di->mediasize - di->blocksize - extent -
1496 error = dump_write(di, buf, 0,
1497 di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1498 keysize, di->blocksize);
1500 error = dump_write(di, buf, 0, di->mediaoffset + di->mediasize -
1501 di->blocksize, di->blocksize);
1506 * Don't touch the first SIZEOF_METADATA bytes on the dump device. This is to
1507 * protect us from metadata and metadata from us.
1509 #define SIZEOF_METADATA (64 * 1024)
1512 * Do some preliminary setup for a kernel dump: initialize state for encryption,
1513 * if requested, and make sure that we have enough space on the dump device.
1515 * We set things up so that the dump ends before the last sector of the dump
1516 * device, at which the trailing header is written.
1518 * +-----------+------+-----+----------------------------+------+
1519 * | | lhdr | key | ... kernel dump ... | thdr |
1520 * +-----------+------+-----+----------------------------+------+
1521 * 1 blk opt <------- dump extent --------> 1 blk
1523 * Dumps written using dump_append() start at the beginning of the extent.
1524 * Uncompressed dumps will use the entire extent, but compressed dumps typically
1525 * will not. The true length of the dump is recorded in the leading and trailing
1526 * headers once the dump has been completed.
1528 * The dump device may provide a callback, in which case it will initialize
1529 * dumpoff and take care of laying out the headers.
1532 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1534 uint64_t dumpextent, span;
1539 error = kerneldumpcrypto_init(di->kdcrypto);
1542 keysize = kerneldumpcrypto_dumpkeysize(di->kdcrypto);
1548 if (di->dumper_start != NULL) {
1549 error = di->dumper_start(di);
1551 dumpextent = dtoh64(kdh->dumpextent);
1552 span = SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1554 if (di->mediasize < span) {
1555 if (di->kdcomp == NULL)
1559 * We don't yet know how much space the compressed dump
1560 * will occupy, so try to use the whole swap partition
1561 * (minus the first 64KB) in the hope that the
1562 * compressed dump will fit. If that doesn't turn out to
1563 * be enough, the bounds checking in dump_write()
1564 * will catch us and cause the dump to fail.
1566 dumpextent = di->mediasize - span + dumpextent;
1567 kdh->dumpextent = htod64(dumpextent);
1571 * The offset at which to begin writing the dump.
1573 di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1576 di->origdumpoff = di->dumpoff;
1581 _dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1587 if (di->kdcrypto != NULL)
1588 error = dump_encrypted_write(di, virtual, physical, di->dumpoff,
1592 error = dump_write(di, virtual, physical, di->dumpoff, length);
1594 di->dumpoff += length;
1599 * Write to the dump device starting at dumpoff. When compression is enabled,
1600 * writes to the device will be performed using a callback that gets invoked
1601 * when the compression stream's output buffer is full.
1604 dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1609 if (di->kdcomp != NULL) {
1610 /* Bounce through a buffer to avoid CRC errors. */
1611 if (length > di->maxiosize)
1613 buf = di->kdcomp->kdc_buf;
1614 memmove(buf, virtual, length);
1615 return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1617 return (_dump_append(di, virtual, physical, length));
1621 * Write to the dump device at the specified offset.
1624 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1625 off_t offset, size_t length)
1629 error = dump_check_bounds(di, offset, length);
1632 return (di->dumper(di->priv, virtual, physical, offset, length));
1636 * Perform kernel dump finalization: flush the compression stream, if necessary,
1637 * write the leading and trailing kernel dump headers now that we know the true
1638 * length of the dump, and optionally write the encryption key following the
1642 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1646 if (di->kdcomp != NULL) {
1647 error = compressor_flush(di->kdcomp->kdc_stream);
1648 if (error == EAGAIN) {
1649 /* We have residual data in di->blockbuf. */
1650 error = dump_write(di, di->blockbuf, 0, di->dumpoff,
1652 di->dumpoff += di->kdcomp->kdc_resid;
1653 di->kdcomp->kdc_resid = 0;
1659 * We now know the size of the compressed dump, so update the
1660 * header accordingly and recompute parity.
1662 kdh->dumplength = htod64(di->dumpoff - di->origdumpoff);
1664 kdh->parity = kerneldump_parity(kdh);
1666 compressor_reset(di->kdcomp->kdc_stream);
1669 error = dump_write_headers(di, kdh);
1673 (void)dump_write(di, NULL, 0, 0, 0);
1678 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1679 char *magic, uint32_t archver, uint64_t dumplen)
1683 bzero(kdh, sizeof(*kdh));
1684 strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1685 strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1686 kdh->version = htod32(KERNELDUMPVERSION);
1687 kdh->architectureversion = htod32(archver);
1688 kdh->dumplength = htod64(dumplen);
1689 kdh->dumpextent = kdh->dumplength;
1690 kdh->dumptime = htod64(time_second);
1692 kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1694 kdh->dumpkeysize = 0;
1696 kdh->blocksize = htod32(di->blocksize);
1697 strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1698 dstsize = sizeof(kdh->versionstring);
1699 if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1700 kdh->versionstring[dstsize - 2] = '\n';
1701 if (panicstr != NULL)
1702 strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1703 if (di->kdcomp != NULL)
1704 kdh->compression = di->kdcomp->kdc_format;
1705 kdh->parity = kerneldump_parity(kdh);
1709 DB_SHOW_COMMAND(panic, db_show_panic)
1712 if (panicstr == NULL)
1713 db_printf("panicstr not set\n");
1715 db_printf("panic: %s\n", panicstr);