xref: /freebsd/sys/kern/kern_shutdown.c (revision 6026dcd7ca888f3433f4df34ede69a77c1eb7701)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
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.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
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.
23  *
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
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_shutdown.c	8.3 (Berkeley) 1/21/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_ddb.h"
43 #include "opt_ekcd.h"
44 #include "opt_kdb.h"
45 #include "opt_panic.h"
46 #include "opt_sched.h"
47 #include "opt_watchdog.h"
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/bio.h>
52 #include <sys/buf.h>
53 #include <sys/conf.h>
54 #include <sys/compressor.h>
55 #include <sys/cons.h>
56 #include <sys/eventhandler.h>
57 #include <sys/filedesc.h>
58 #include <sys/jail.h>
59 #include <sys/kdb.h>
60 #include <sys/kernel.h>
61 #include <sys/kerneldump.h>
62 #include <sys/kthread.h>
63 #include <sys/ktr.h>
64 #include <sys/malloc.h>
65 #include <sys/mount.h>
66 #include <sys/priv.h>
67 #include <sys/proc.h>
68 #include <sys/reboot.h>
69 #include <sys/resourcevar.h>
70 #include <sys/rwlock.h>
71 #include <sys/sched.h>
72 #include <sys/smp.h>
73 #include <sys/sysctl.h>
74 #include <sys/sysproto.h>
75 #include <sys/vnode.h>
76 #include <sys/watchdog.h>
77 
78 #include <crypto/rijndael/rijndael-api-fst.h>
79 #include <crypto/sha2/sha256.h>
80 
81 #include <ddb/ddb.h>
82 
83 #include <machine/cpu.h>
84 #include <machine/dump.h>
85 #include <machine/pcb.h>
86 #include <machine/smp.h>
87 
88 #include <security/mac/mac_framework.h>
89 
90 #include <vm/vm.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_page.h>
93 #include <vm/vm_pager.h>
94 #include <vm/swap_pager.h>
95 
96 #include <sys/signalvar.h>
97 
98 static MALLOC_DEFINE(M_DUMPER, "dumper", "dumper block buffer");
99 
100 #ifndef PANIC_REBOOT_WAIT_TIME
101 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
102 #endif
103 static int panic_reboot_wait_time = PANIC_REBOOT_WAIT_TIME;
104 SYSCTL_INT(_kern, OID_AUTO, panic_reboot_wait_time, CTLFLAG_RWTUN,
105     &panic_reboot_wait_time, 0,
106     "Seconds to wait before rebooting after a panic");
107 
108 /*
109  * Note that stdarg.h and the ANSI style va_start macro is used for both
110  * ANSI and traditional C compilers.
111  */
112 #include <machine/stdarg.h>
113 
114 #ifdef KDB
115 #ifdef KDB_UNATTENDED
116 int debugger_on_panic = 0;
117 #else
118 int debugger_on_panic = 1;
119 #endif
120 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic,
121     CTLFLAG_RWTUN | CTLFLAG_SECURE,
122     &debugger_on_panic, 0, "Run debugger on kernel panic");
123 
124 #ifdef KDB_TRACE
125 static int trace_on_panic = 1;
126 #else
127 static int trace_on_panic = 0;
128 #endif
129 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
130     CTLFLAG_RWTUN | CTLFLAG_SECURE,
131     &trace_on_panic, 0, "Print stack trace on kernel panic");
132 #endif /* KDB */
133 
134 static int sync_on_panic = 0;
135 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN,
136 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
137 
138 static bool poweroff_on_panic = 0;
139 SYSCTL_BOOL(_kern, OID_AUTO, poweroff_on_panic, CTLFLAG_RWTUN,
140 	&poweroff_on_panic, 0, "Do a power off instead of a reboot on a panic");
141 
142 static bool powercycle_on_panic = 0;
143 SYSCTL_BOOL(_kern, OID_AUTO, powercycle_on_panic, CTLFLAG_RWTUN,
144 	&powercycle_on_panic, 0, "Do a power cycle instead of a reboot on a panic");
145 
146 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0,
147     "Shutdown environment");
148 
149 #ifndef DIAGNOSTIC
150 static int show_busybufs;
151 #else
152 static int show_busybufs = 1;
153 #endif
154 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
155 	&show_busybufs, 0, "");
156 
157 int suspend_blocked = 0;
158 SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW,
159 	&suspend_blocked, 0, "Block suspend due to a pending shutdown");
160 
161 #ifdef EKCD
162 FEATURE(ekcd, "Encrypted kernel crash dumps support");
163 
164 MALLOC_DEFINE(M_EKCD, "ekcd", "Encrypted kernel crash dumps data");
165 
166 struct kerneldumpcrypto {
167 	uint8_t			kdc_encryption;
168 	uint8_t			kdc_iv[KERNELDUMP_IV_MAX_SIZE];
169 	keyInstance		kdc_ki;
170 	cipherInstance		kdc_ci;
171 	uint32_t		kdc_dumpkeysize;
172 	struct kerneldumpkey	kdc_dumpkey[];
173 };
174 #endif
175 
176 struct kerneldumpcomp {
177 	uint8_t			kdc_format;
178 	struct compressor	*kdc_stream;
179 	uint8_t			*kdc_buf;
180 	size_t			kdc_resid;
181 };
182 
183 static struct kerneldumpcomp *kerneldumpcomp_create(struct dumperinfo *di,
184 		    uint8_t compression);
185 static void	kerneldumpcomp_destroy(struct dumperinfo *di);
186 static int	kerneldumpcomp_write_cb(void *base, size_t len, off_t off, void *arg);
187 
188 static int kerneldump_gzlevel = 6;
189 SYSCTL_INT(_kern, OID_AUTO, kerneldump_gzlevel, CTLFLAG_RWTUN,
190     &kerneldump_gzlevel, 0,
191     "Kernel crash dump compression level");
192 
193 /*
194  * Variable panicstr contains argument to first call to panic; used as flag
195  * to indicate that the kernel has already called panic.
196  */
197 const char *panicstr;
198 
199 int dumping;				/* system is dumping */
200 int rebooting;				/* system is rebooting */
201 static struct dumperinfo dumper;	/* our selected dumper */
202 
203 /* Context information for dump-debuggers. */
204 static struct pcb dumppcb;		/* Registers. */
205 lwpid_t dumptid;			/* Thread ID. */
206 
207 static struct cdevsw reroot_cdevsw = {
208      .d_version = D_VERSION,
209      .d_name    = "reroot",
210 };
211 
212 static void poweroff_wait(void *, int);
213 static void shutdown_halt(void *junk, int howto);
214 static void shutdown_panic(void *junk, int howto);
215 static void shutdown_reset(void *junk, int howto);
216 static int kern_reroot(void);
217 
218 /* register various local shutdown events */
219 static void
220 shutdown_conf(void *unused)
221 {
222 
223 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
224 	    SHUTDOWN_PRI_FIRST);
225 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
226 	    SHUTDOWN_PRI_LAST + 100);
227 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
228 	    SHUTDOWN_PRI_LAST + 100);
229 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
230 	    SHUTDOWN_PRI_LAST + 200);
231 }
232 
233 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
234 
235 /*
236  * The only reason this exists is to create the /dev/reroot/ directory,
237  * used by reroot code in init(8) as a mountpoint for tmpfs.
238  */
239 static void
240 reroot_conf(void *unused)
241 {
242 	int error;
243 	struct cdev *cdev;
244 
245 	error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev,
246 	    &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot");
247 	if (error != 0) {
248 		printf("%s: failed to create device node, error %d",
249 		    __func__, error);
250 	}
251 }
252 
253 SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL);
254 
255 /*
256  * The system call that results in a reboot.
257  */
258 /* ARGSUSED */
259 int
260 sys_reboot(struct thread *td, struct reboot_args *uap)
261 {
262 	int error;
263 
264 	error = 0;
265 #ifdef MAC
266 	error = mac_system_check_reboot(td->td_ucred, uap->opt);
267 #endif
268 	if (error == 0)
269 		error = priv_check(td, PRIV_REBOOT);
270 	if (error == 0) {
271 		if (uap->opt & RB_REROOT) {
272 			error = kern_reroot();
273 		} else {
274 			mtx_lock(&Giant);
275 			kern_reboot(uap->opt);
276 			mtx_unlock(&Giant);
277 		}
278 	}
279 	return (error);
280 }
281 
282 /*
283  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
284  */
285 void
286 shutdown_nice(int howto)
287 {
288 
289 	if (initproc != NULL) {
290 		/* Send a signal to init(8) and have it shutdown the world. */
291 		PROC_LOCK(initproc);
292 		if (howto & RB_POWEROFF)
293 			kern_psignal(initproc, SIGUSR2);
294 		else if (howto & RB_POWERCYCLE)
295 			kern_psignal(initproc, SIGWINCH);
296 		else if (howto & RB_HALT)
297 			kern_psignal(initproc, SIGUSR1);
298 		else
299 			kern_psignal(initproc, SIGINT);
300 		PROC_UNLOCK(initproc);
301 	} else {
302 		/* No init(8) running, so simply reboot. */
303 		kern_reboot(howto | RB_NOSYNC);
304 	}
305 }
306 
307 static void
308 print_uptime(void)
309 {
310 	int f;
311 	struct timespec ts;
312 
313 	getnanouptime(&ts);
314 	printf("Uptime: ");
315 	f = 0;
316 	if (ts.tv_sec >= 86400) {
317 		printf("%ldd", (long)ts.tv_sec / 86400);
318 		ts.tv_sec %= 86400;
319 		f = 1;
320 	}
321 	if (f || ts.tv_sec >= 3600) {
322 		printf("%ldh", (long)ts.tv_sec / 3600);
323 		ts.tv_sec %= 3600;
324 		f = 1;
325 	}
326 	if (f || ts.tv_sec >= 60) {
327 		printf("%ldm", (long)ts.tv_sec / 60);
328 		ts.tv_sec %= 60;
329 		f = 1;
330 	}
331 	printf("%lds\n", (long)ts.tv_sec);
332 }
333 
334 int
335 doadump(boolean_t textdump)
336 {
337 	boolean_t coredump;
338 	int error;
339 
340 	error = 0;
341 	if (dumping)
342 		return (EBUSY);
343 	if (dumper.dumper == NULL)
344 		return (ENXIO);
345 
346 	savectx(&dumppcb);
347 	dumptid = curthread->td_tid;
348 	dumping++;
349 
350 	coredump = TRUE;
351 #ifdef DDB
352 	if (textdump && textdump_pending) {
353 		coredump = FALSE;
354 		textdump_dumpsys(&dumper);
355 	}
356 #endif
357 	if (coredump)
358 		error = dumpsys(&dumper);
359 
360 	dumping--;
361 	return (error);
362 }
363 
364 /*
365  * Shutdown the system cleanly to prepare for reboot, halt, or power off.
366  */
367 void
368 kern_reboot(int howto)
369 {
370 	static int once = 0;
371 
372 #if defined(SMP)
373 	/*
374 	 * Bind us to the first CPU so that all shutdown code runs there.  Some
375 	 * systems don't shutdown properly (i.e., ACPI power off) if we
376 	 * run on another processor.
377 	 */
378 	if (!SCHEDULER_STOPPED()) {
379 		thread_lock(curthread);
380 		sched_bind(curthread, CPU_FIRST());
381 		thread_unlock(curthread);
382 		KASSERT(PCPU_GET(cpuid) == CPU_FIRST(),
383 		    ("boot: not running on cpu 0"));
384 	}
385 #endif
386 	/* We're in the process of rebooting. */
387 	rebooting = 1;
388 
389 	/* We are out of the debugger now. */
390 	kdb_active = 0;
391 
392 	/*
393 	 * Do any callouts that should be done BEFORE syncing the filesystems.
394 	 */
395 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
396 
397 	/*
398 	 * Now sync filesystems
399 	 */
400 	if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
401 		once = 1;
402 		bufshutdown(show_busybufs);
403 	}
404 
405 	print_uptime();
406 
407 	cngrab();
408 
409 	/*
410 	 * Ok, now do things that assume all filesystem activity has
411 	 * been completed.
412 	 */
413 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
414 
415 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
416 		doadump(TRUE);
417 
418 	/* Now that we're going to really halt the system... */
419 	EVENTHANDLER_INVOKE(shutdown_final, howto);
420 
421 	for(;;) ;	/* safety against shutdown_reset not working */
422 	/* NOTREACHED */
423 }
424 
425 /*
426  * The system call that results in changing the rootfs.
427  */
428 static int
429 kern_reroot(void)
430 {
431 	struct vnode *oldrootvnode, *vp;
432 	struct mount *mp, *devmp;
433 	int error;
434 
435 	if (curproc != initproc)
436 		return (EPERM);
437 
438 	/*
439 	 * Mark the filesystem containing currently-running executable
440 	 * (the temporary copy of init(8)) busy.
441 	 */
442 	vp = curproc->p_textvp;
443 	error = vn_lock(vp, LK_SHARED);
444 	if (error != 0)
445 		return (error);
446 	mp = vp->v_mount;
447 	error = vfs_busy(mp, MBF_NOWAIT);
448 	if (error != 0) {
449 		vfs_ref(mp);
450 		VOP_UNLOCK(vp, 0);
451 		error = vfs_busy(mp, 0);
452 		vn_lock(vp, LK_SHARED | LK_RETRY);
453 		vfs_rel(mp);
454 		if (error != 0) {
455 			VOP_UNLOCK(vp, 0);
456 			return (ENOENT);
457 		}
458 		if (vp->v_iflag & VI_DOOMED) {
459 			VOP_UNLOCK(vp, 0);
460 			vfs_unbusy(mp);
461 			return (ENOENT);
462 		}
463 	}
464 	VOP_UNLOCK(vp, 0);
465 
466 	/*
467 	 * Remove the filesystem containing currently-running executable
468 	 * from the mount list, to prevent it from being unmounted
469 	 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
470 	 *
471 	 * Also preserve /dev - forcibly unmounting it could cause driver
472 	 * reinitialization.
473 	 */
474 
475 	vfs_ref(rootdevmp);
476 	devmp = rootdevmp;
477 	rootdevmp = NULL;
478 
479 	mtx_lock(&mountlist_mtx);
480 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
481 	TAILQ_REMOVE(&mountlist, devmp, mnt_list);
482 	mtx_unlock(&mountlist_mtx);
483 
484 	oldrootvnode = rootvnode;
485 
486 	/*
487 	 * Unmount everything except for the two filesystems preserved above.
488 	 */
489 	vfs_unmountall();
490 
491 	/*
492 	 * Add /dev back; vfs_mountroot() will move it into its new place.
493 	 */
494 	mtx_lock(&mountlist_mtx);
495 	TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
496 	mtx_unlock(&mountlist_mtx);
497 	rootdevmp = devmp;
498 	vfs_rel(rootdevmp);
499 
500 	/*
501 	 * Mount the new rootfs.
502 	 */
503 	vfs_mountroot();
504 
505 	/*
506 	 * Update all references to the old rootvnode.
507 	 */
508 	mountcheckdirs(oldrootvnode, rootvnode);
509 
510 	/*
511 	 * Add the temporary filesystem back and unbusy it.
512 	 */
513 	mtx_lock(&mountlist_mtx);
514 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
515 	mtx_unlock(&mountlist_mtx);
516 	vfs_unbusy(mp);
517 
518 	return (0);
519 }
520 
521 /*
522  * If the shutdown was a clean halt, behave accordingly.
523  */
524 static void
525 shutdown_halt(void *junk, int howto)
526 {
527 
528 	if (howto & RB_HALT) {
529 		printf("\n");
530 		printf("The operating system has halted.\n");
531 		printf("Please press any key to reboot.\n\n");
532 		switch (cngetc()) {
533 		case -1:		/* No console, just die */
534 			cpu_halt();
535 			/* NOTREACHED */
536 		default:
537 			break;
538 		}
539 	}
540 }
541 
542 /*
543  * Check to see if the system paniced, pause and then reboot
544  * according to the specified delay.
545  */
546 static void
547 shutdown_panic(void *junk, int howto)
548 {
549 	int loop;
550 
551 	if (howto & RB_DUMP) {
552 		if (panic_reboot_wait_time != 0) {
553 			if (panic_reboot_wait_time != -1) {
554 				printf("Automatic reboot in %d seconds - "
555 				       "press a key on the console to abort\n",
556 					panic_reboot_wait_time);
557 				for (loop = panic_reboot_wait_time * 10;
558 				     loop > 0; --loop) {
559 					DELAY(1000 * 100); /* 1/10th second */
560 					/* Did user type a key? */
561 					if (cncheckc() != -1)
562 						break;
563 				}
564 				if (!loop)
565 					return;
566 			}
567 		} else { /* zero time specified - reboot NOW */
568 			return;
569 		}
570 		printf("--> Press a key on the console to reboot,\n");
571 		printf("--> or switch off the system now.\n");
572 		cngetc();
573 	}
574 }
575 
576 /*
577  * Everything done, now reset
578  */
579 static void
580 shutdown_reset(void *junk, int howto)
581 {
582 
583 	printf("Rebooting...\n");
584 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
585 
586 	/*
587 	 * Acquiring smp_ipi_mtx here has a double effect:
588 	 * - it disables interrupts avoiding CPU0 preemption
589 	 *   by fast handlers (thus deadlocking  against other CPUs)
590 	 * - it avoids deadlocks against smp_rendezvous() or, more
591 	 *   generally, threads busy-waiting, with this spinlock held,
592 	 *   and waiting for responses by threads on other CPUs
593 	 *   (ie. smp_tlb_shootdown()).
594 	 *
595 	 * For the !SMP case it just needs to handle the former problem.
596 	 */
597 #ifdef SMP
598 	mtx_lock_spin(&smp_ipi_mtx);
599 #else
600 	spinlock_enter();
601 #endif
602 
603 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
604 	cpu_reset();
605 	/* NOTREACHED */ /* assuming reset worked */
606 }
607 
608 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
609 static int kassert_warn_only = 0;
610 #ifdef KDB
611 static int kassert_do_kdb = 0;
612 #endif
613 #ifdef KTR
614 static int kassert_do_ktr = 0;
615 #endif
616 static int kassert_do_log = 1;
617 static int kassert_log_pps_limit = 4;
618 static int kassert_log_mute_at = 0;
619 static int kassert_log_panic_at = 0;
620 static int kassert_warnings = 0;
621 
622 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW, NULL, "kassert options");
623 
624 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, CTLFLAG_RWTUN,
625     &kassert_warn_only, 0,
626     "KASSERT triggers a panic (1) or just a warning (0)");
627 
628 #ifdef KDB
629 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, CTLFLAG_RWTUN,
630     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
631 #endif
632 
633 #ifdef KTR
634 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, CTLFLAG_RWTUN,
635     &kassert_do_ktr, 0,
636     "KASSERT does a KTR, set this to the KTRMASK you want");
637 #endif
638 
639 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, CTLFLAG_RWTUN,
640     &kassert_do_log, 0, "KASSERT triggers a panic (1) or just a warning (0)");
641 
642 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RWTUN,
643     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
644 
645 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, CTLFLAG_RWTUN,
646     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
647 
648 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, CTLFLAG_RWTUN,
649     &kassert_log_pps_limit, 0, "limit number of log messages per second");
650 
651 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, CTLFLAG_RWTUN,
652     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
653 
654 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
655 
656 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
657     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE, NULL, 0,
658     kassert_sysctl_kassert, "I", "set to trigger a test kassert");
659 
660 static int
661 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
662 {
663 	int error, i;
664 
665 	error = sysctl_wire_old_buffer(req, sizeof(int));
666 	if (error == 0) {
667 		i = 0;
668 		error = sysctl_handle_int(oidp, &i, 0, req);
669 	}
670 	if (error != 0 || req->newptr == NULL)
671 		return (error);
672 	KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
673 	return (0);
674 }
675 
676 /*
677  * Called by KASSERT, this decides if we will panic
678  * or if we will log via printf and/or ktr.
679  */
680 void
681 kassert_panic(const char *fmt, ...)
682 {
683 	static char buf[256];
684 	va_list ap;
685 
686 	va_start(ap, fmt);
687 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
688 	va_end(ap);
689 
690 	/*
691 	 * panic if we're not just warning, or if we've exceeded
692 	 * kassert_log_panic_at warnings.
693 	 */
694 	if (!kassert_warn_only ||
695 	    (kassert_log_panic_at > 0 &&
696 	     kassert_warnings >= kassert_log_panic_at)) {
697 		va_start(ap, fmt);
698 		vpanic(fmt, ap);
699 		/* NORETURN */
700 	}
701 #ifdef KTR
702 	if (kassert_do_ktr)
703 		CTR0(ktr_mask, buf);
704 #endif /* KTR */
705 	/*
706 	 * log if we've not yet met the mute limit.
707 	 */
708 	if (kassert_do_log &&
709 	    (kassert_log_mute_at == 0 ||
710 	     kassert_warnings < kassert_log_mute_at)) {
711 		static  struct timeval lasterr;
712 		static  int curerr;
713 
714 		if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
715 			printf("KASSERT failed: %s\n", buf);
716 			kdb_backtrace();
717 		}
718 	}
719 #ifdef KDB
720 	if (kassert_do_kdb) {
721 		kdb_enter(KDB_WHY_KASSERT, buf);
722 	}
723 #endif
724 	atomic_add_int(&kassert_warnings, 1);
725 }
726 #endif
727 
728 /*
729  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
730  * and then reboots.  If we are called twice, then we avoid trying to sync
731  * the disks as this often leads to recursive panics.
732  */
733 void
734 panic(const char *fmt, ...)
735 {
736 	va_list ap;
737 
738 	va_start(ap, fmt);
739 	vpanic(fmt, ap);
740 }
741 
742 void
743 vpanic(const char *fmt, va_list ap)
744 {
745 #ifdef SMP
746 	cpuset_t other_cpus;
747 #endif
748 	struct thread *td = curthread;
749 	int bootopt, newpanic;
750 	static char buf[256];
751 
752 	spinlock_enter();
753 
754 #ifdef SMP
755 	/*
756 	 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
757 	 * concurrently entering panic.  Only the winner will proceed
758 	 * further.
759 	 */
760 	if (panicstr == NULL && !kdb_active) {
761 		other_cpus = all_cpus;
762 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
763 		stop_cpus_hard(other_cpus);
764 	}
765 #endif
766 
767 	/*
768 	 * Ensure that the scheduler is stopped while panicking, even if panic
769 	 * has been entered from kdb.
770 	 */
771 	td->td_stopsched = 1;
772 
773 	bootopt = RB_AUTOBOOT;
774 	newpanic = 0;
775 	if (panicstr)
776 		bootopt |= RB_NOSYNC;
777 	else {
778 		bootopt |= RB_DUMP;
779 		panicstr = fmt;
780 		newpanic = 1;
781 	}
782 
783 	if (newpanic) {
784 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
785 		panicstr = buf;
786 		cngrab();
787 		printf("panic: %s\n", buf);
788 	} else {
789 		printf("panic: ");
790 		vprintf(fmt, ap);
791 		printf("\n");
792 	}
793 #ifdef SMP
794 	printf("cpuid = %d\n", PCPU_GET(cpuid));
795 #endif
796 	printf("time = %jd\n", (intmax_t )time_second);
797 #ifdef KDB
798 	if (newpanic && trace_on_panic)
799 		kdb_backtrace();
800 	if (debugger_on_panic)
801 		kdb_enter(KDB_WHY_PANIC, "panic");
802 #endif
803 	/*thread_lock(td); */
804 	td->td_flags |= TDF_INPANIC;
805 	/* thread_unlock(td); */
806 	if (!sync_on_panic)
807 		bootopt |= RB_NOSYNC;
808 	if (poweroff_on_panic)
809 		bootopt |= RB_POWEROFF;
810 	if (powercycle_on_panic)
811 		bootopt |= RB_POWERCYCLE;
812 	kern_reboot(bootopt);
813 }
814 
815 /*
816  * Support for poweroff delay.
817  *
818  * Please note that setting this delay too short might power off your machine
819  * before the write cache on your hard disk has been flushed, leading to
820  * soft-updates inconsistencies.
821  */
822 #ifndef POWEROFF_DELAY
823 # define POWEROFF_DELAY 5000
824 #endif
825 static int poweroff_delay = POWEROFF_DELAY;
826 
827 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
828     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
829 
830 static void
831 poweroff_wait(void *junk, int howto)
832 {
833 
834 	if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
835 		return;
836 	DELAY(poweroff_delay * 1000);
837 }
838 
839 /*
840  * Some system processes (e.g. syncer) need to be stopped at appropriate
841  * points in their main loops prior to a system shutdown, so that they
842  * won't interfere with the shutdown process (e.g. by holding a disk buf
843  * to cause sync to fail).  For each of these system processes, register
844  * shutdown_kproc() as a handler for one of shutdown events.
845  */
846 static int kproc_shutdown_wait = 60;
847 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
848     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
849 
850 void
851 kproc_shutdown(void *arg, int howto)
852 {
853 	struct proc *p;
854 	int error;
855 
856 	if (panicstr)
857 		return;
858 
859 	p = (struct proc *)arg;
860 	printf("Waiting (max %d seconds) for system process `%s' to stop... ",
861 	    kproc_shutdown_wait, p->p_comm);
862 	error = kproc_suspend(p, kproc_shutdown_wait * hz);
863 
864 	if (error == EWOULDBLOCK)
865 		printf("timed out\n");
866 	else
867 		printf("done\n");
868 }
869 
870 void
871 kthread_shutdown(void *arg, int howto)
872 {
873 	struct thread *td;
874 	int error;
875 
876 	if (panicstr)
877 		return;
878 
879 	td = (struct thread *)arg;
880 	printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
881 	    kproc_shutdown_wait, td->td_name);
882 	error = kthread_suspend(td, kproc_shutdown_wait * hz);
883 
884 	if (error == EWOULDBLOCK)
885 		printf("timed out\n");
886 	else
887 		printf("done\n");
888 }
889 
890 static char dumpdevname[sizeof(((struct cdev*)NULL)->si_name)];
891 SYSCTL_STRING(_kern_shutdown, OID_AUTO, dumpdevname, CTLFLAG_RD,
892     dumpdevname, 0, "Device for kernel dumps");
893 
894 static int	_dump_append(struct dumperinfo *di, void *virtual,
895 		    vm_offset_t physical, size_t length);
896 
897 #ifdef EKCD
898 static struct kerneldumpcrypto *
899 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
900     const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
901 {
902 	struct kerneldumpcrypto *kdc;
903 	struct kerneldumpkey *kdk;
904 	uint32_t dumpkeysize;
905 
906 	dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
907 	kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
908 
909 	arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
910 
911 	kdc->kdc_encryption = encryption;
912 	switch (kdc->kdc_encryption) {
913 	case KERNELDUMP_ENC_AES_256_CBC:
914 		if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
915 			goto failed;
916 		break;
917 	default:
918 		goto failed;
919 	}
920 
921 	kdc->kdc_dumpkeysize = dumpkeysize;
922 	kdk = kdc->kdc_dumpkey;
923 	kdk->kdk_encryption = kdc->kdc_encryption;
924 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
925 	kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
926 	memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
927 
928 	return (kdc);
929 failed:
930 	explicit_bzero(kdc, sizeof(*kdc) + dumpkeysize);
931 	free(kdc, M_EKCD);
932 	return (NULL);
933 }
934 
935 static int
936 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
937 {
938 	uint8_t hash[SHA256_DIGEST_LENGTH];
939 	SHA256_CTX ctx;
940 	struct kerneldumpkey *kdk;
941 	int error;
942 
943 	error = 0;
944 
945 	if (kdc == NULL)
946 		return (0);
947 
948 	/*
949 	 * When a user enters ddb it can write a crash dump multiple times.
950 	 * Each time it should be encrypted using a different IV.
951 	 */
952 	SHA256_Init(&ctx);
953 	SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
954 	SHA256_Final(hash, &ctx);
955 	bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
956 
957 	switch (kdc->kdc_encryption) {
958 	case KERNELDUMP_ENC_AES_256_CBC:
959 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
960 		    kdc->kdc_iv) <= 0) {
961 			error = EINVAL;
962 			goto out;
963 		}
964 		break;
965 	default:
966 		error = EINVAL;
967 		goto out;
968 	}
969 
970 	kdk = kdc->kdc_dumpkey;
971 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
972 out:
973 	explicit_bzero(hash, sizeof(hash));
974 	return (error);
975 }
976 
977 static uint32_t
978 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
979 {
980 
981 	if (kdc == NULL)
982 		return (0);
983 	return (kdc->kdc_dumpkeysize);
984 }
985 #endif /* EKCD */
986 
987 static struct kerneldumpcomp *
988 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression)
989 {
990 	struct kerneldumpcomp *kdcomp;
991 	int format;
992 
993 	switch (compression) {
994 	case KERNELDUMP_COMP_GZIP:
995 		format = COMPRESS_GZIP;
996 		break;
997 	case KERNELDUMP_COMP_ZSTD:
998 		format = COMPRESS_ZSTD;
999 		break;
1000 	default:
1001 		return (NULL);
1002 	}
1003 
1004 	kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO);
1005 	kdcomp->kdc_format = compression;
1006 	kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb,
1007 	    format, di->maxiosize, kerneldump_gzlevel, di);
1008 	if (kdcomp->kdc_stream == NULL) {
1009 		free(kdcomp, M_DUMPER);
1010 		return (NULL);
1011 	}
1012 	kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1013 	return (kdcomp);
1014 }
1015 
1016 static void
1017 kerneldumpcomp_destroy(struct dumperinfo *di)
1018 {
1019 	struct kerneldumpcomp *kdcomp;
1020 
1021 	kdcomp = di->kdcomp;
1022 	if (kdcomp == NULL)
1023 		return;
1024 	compressor_fini(kdcomp->kdc_stream);
1025 	explicit_bzero(kdcomp->kdc_buf, di->maxiosize);
1026 	free(kdcomp->kdc_buf, M_DUMPER);
1027 	free(kdcomp, M_DUMPER);
1028 }
1029 
1030 /* Registration of dumpers */
1031 int
1032 set_dumper(struct dumperinfo *di, const char *devname, struct thread *td,
1033     uint8_t compression, uint8_t encryption, const uint8_t *key,
1034     uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1035 {
1036 	size_t wantcopy;
1037 	int error;
1038 
1039 	error = priv_check(td, PRIV_SETDUMPER);
1040 	if (error != 0)
1041 		return (error);
1042 
1043 	if (di == NULL) {
1044 		error = 0;
1045 		goto cleanup;
1046 	}
1047 	if (dumper.dumper != NULL)
1048 		return (EBUSY);
1049 	dumper = *di;
1050 	dumper.blockbuf = NULL;
1051 	dumper.kdcrypto = NULL;
1052 	dumper.kdcomp = NULL;
1053 
1054 	if (encryption != KERNELDUMP_ENC_NONE) {
1055 #ifdef EKCD
1056 		dumper.kdcrypto = kerneldumpcrypto_create(di->blocksize,
1057 		    encryption, key, encryptedkeysize, encryptedkey);
1058 		if (dumper.kdcrypto == NULL) {
1059 			error = EINVAL;
1060 			goto cleanup;
1061 		}
1062 #else
1063 		error = EOPNOTSUPP;
1064 		goto cleanup;
1065 #endif
1066 	}
1067 
1068 	wantcopy = strlcpy(dumpdevname, devname, sizeof(dumpdevname));
1069 	if (wantcopy >= sizeof(dumpdevname)) {
1070 		printf("set_dumper: device name truncated from '%s' -> '%s'\n",
1071 		    devname, dumpdevname);
1072 	}
1073 
1074 	if (compression != KERNELDUMP_COMP_NONE) {
1075 		/*
1076 		 * We currently can't support simultaneous encryption and
1077 		 * compression.
1078 		 */
1079 		if (encryption != KERNELDUMP_ENC_NONE) {
1080 			error = EOPNOTSUPP;
1081 			goto cleanup;
1082 		}
1083 		dumper.kdcomp = kerneldumpcomp_create(&dumper, compression);
1084 		if (dumper.kdcomp == NULL) {
1085 			error = EINVAL;
1086 			goto cleanup;
1087 		}
1088 	}
1089 
1090 	dumper.blockbuf = malloc(di->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1091 	return (0);
1092 cleanup:
1093 #ifdef EKCD
1094 	if (dumper.kdcrypto != NULL) {
1095 		explicit_bzero(dumper.kdcrypto, sizeof(*dumper.kdcrypto) +
1096 		    dumper.kdcrypto->kdc_dumpkeysize);
1097 		free(dumper.kdcrypto, M_EKCD);
1098 	}
1099 #endif
1100 
1101 	kerneldumpcomp_destroy(&dumper);
1102 
1103 	if (dumper.blockbuf != NULL) {
1104 		explicit_bzero(dumper.blockbuf, dumper.blocksize);
1105 		free(dumper.blockbuf, M_DUMPER);
1106 	}
1107 	explicit_bzero(&dumper, sizeof(dumper));
1108 	dumpdevname[0] = '\0';
1109 	return (error);
1110 }
1111 
1112 static int
1113 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1114 {
1115 
1116 	if (length != 0 && (offset < di->mediaoffset ||
1117 	    offset - di->mediaoffset + length > di->mediasize)) {
1118 		printf("Attempt to write outside dump device boundaries.\n"
1119 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1120 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
1121 		    (uintmax_t)length, (intmax_t)di->mediasize);
1122 		return (ENOSPC);
1123 	}
1124 	if (length % di->blocksize != 0) {
1125 		printf("Attempt to write partial block of length %ju.\n",
1126 		    (uintmax_t)length);
1127 		return (EINVAL);
1128 	}
1129 	if (offset % di->blocksize != 0) {
1130 		printf("Attempt to write at unaligned offset %jd.\n",
1131 		    (intmax_t)offset);
1132 		return (EINVAL);
1133 	}
1134 
1135 	return (0);
1136 }
1137 
1138 #ifdef EKCD
1139 static int
1140 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1141 {
1142 
1143 	switch (kdc->kdc_encryption) {
1144 	case KERNELDUMP_ENC_AES_256_CBC:
1145 		if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1146 		    8 * size, buf) <= 0) {
1147 			return (EIO);
1148 		}
1149 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1150 		    buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1151 			return (EIO);
1152 		}
1153 		break;
1154 	default:
1155 		return (EINVAL);
1156 	}
1157 
1158 	return (0);
1159 }
1160 
1161 /* Encrypt data and call dumper. */
1162 static int
1163 dump_encrypted_write(struct dumperinfo *di, void *virtual,
1164     vm_offset_t physical, off_t offset, size_t length)
1165 {
1166 	static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1167 	struct kerneldumpcrypto *kdc;
1168 	int error;
1169 	size_t nbytes;
1170 
1171 	kdc = di->kdcrypto;
1172 
1173 	while (length > 0) {
1174 		nbytes = MIN(length, sizeof(buf));
1175 		bcopy(virtual, buf, nbytes);
1176 
1177 		if (dump_encrypt(kdc, buf, nbytes) != 0)
1178 			return (EIO);
1179 
1180 		error = dump_write(di, buf, physical, offset, nbytes);
1181 		if (error != 0)
1182 			return (error);
1183 
1184 		offset += nbytes;
1185 		virtual = (void *)((uint8_t *)virtual + nbytes);
1186 		length -= nbytes;
1187 	}
1188 
1189 	return (0);
1190 }
1191 
1192 static int
1193 dump_write_key(struct dumperinfo *di, off_t offset)
1194 {
1195 	struct kerneldumpcrypto *kdc;
1196 
1197 	kdc = di->kdcrypto;
1198 	if (kdc == NULL)
1199 		return (0);
1200 	return (dump_write(di, kdc->kdc_dumpkey, 0, offset,
1201 	    kdc->kdc_dumpkeysize));
1202 }
1203 #endif /* EKCD */
1204 
1205 static int
1206 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1207 {
1208 	struct dumperinfo *di;
1209 	size_t resid, rlength;
1210 	int error;
1211 
1212 	di = arg;
1213 
1214 	if (length % di->blocksize != 0) {
1215 		/*
1216 		 * This must be the final write after flushing the compression
1217 		 * stream. Write as many full blocks as possible and stash the
1218 		 * residual data in the dumper's block buffer. It will be
1219 		 * padded and written in dump_finish().
1220 		 */
1221 		rlength = rounddown(length, di->blocksize);
1222 		if (rlength != 0) {
1223 			error = _dump_append(di, base, 0, rlength);
1224 			if (error != 0)
1225 				return (error);
1226 		}
1227 		resid = length - rlength;
1228 		memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1229 		di->kdcomp->kdc_resid = resid;
1230 		return (EAGAIN);
1231 	}
1232 	return (_dump_append(di, base, 0, length));
1233 }
1234 
1235 /*
1236  * Write a kerneldumpheader at the specified offset. The header structure is 512
1237  * bytes in size, but we must pad to the device sector size.
1238  */
1239 static int
1240 dump_write_header(struct dumperinfo *di, struct kerneldumpheader *kdh,
1241     off_t offset)
1242 {
1243 	void *buf;
1244 	size_t hdrsz;
1245 
1246 	hdrsz = sizeof(*kdh);
1247 	if (hdrsz > di->blocksize)
1248 		return (ENOMEM);
1249 
1250 	if (hdrsz == di->blocksize)
1251 		buf = kdh;
1252 	else {
1253 		buf = di->blockbuf;
1254 		memset(buf, 0, di->blocksize);
1255 		memcpy(buf, kdh, hdrsz);
1256 	}
1257 
1258 	return (dump_write(di, buf, 0, offset, di->blocksize));
1259 }
1260 
1261 /*
1262  * Don't touch the first SIZEOF_METADATA bytes on the dump device.  This is to
1263  * protect us from metadata and metadata from us.
1264  */
1265 #define	SIZEOF_METADATA		(64 * 1024)
1266 
1267 /*
1268  * Do some preliminary setup for a kernel dump: initialize state for encryption,
1269  * if requested, and make sure that we have enough space on the dump device.
1270  *
1271  * We set things up so that the dump ends before the last sector of the dump
1272  * device, at which the trailing header is written.
1273  *
1274  *     +-----------+------+-----+----------------------------+------+
1275  *     |           | lhdr | key |    ... kernel dump ...     | thdr |
1276  *     +-----------+------+-----+----------------------------+------+
1277  *                   1 blk  opt <------- dump extent --------> 1 blk
1278  *
1279  * Dumps written using dump_append() start at the beginning of the extent.
1280  * Uncompressed dumps will use the entire extent, but compressed dumps typically
1281  * will not. The true length of the dump is recorded in the leading and trailing
1282  * headers once the dump has been completed.
1283  */
1284 int
1285 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1286 {
1287 	uint64_t dumpextent;
1288 	uint32_t keysize;
1289 
1290 #ifdef EKCD
1291 	int error = kerneldumpcrypto_init(di->kdcrypto);
1292 	if (error != 0)
1293 		return (error);
1294 	keysize = kerneldumpcrypto_dumpkeysize(di->kdcrypto);
1295 #else
1296 	keysize = 0;
1297 #endif
1298 
1299 	dumpextent = dtoh64(kdh->dumpextent);
1300 	if (di->mediasize < SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1301 	    keysize) {
1302 		if (di->kdcomp != NULL) {
1303 			/*
1304 			 * We don't yet know how much space the compressed dump
1305 			 * will occupy, so try to use the whole swap partition
1306 			 * (minus the first 64KB) in the hope that the
1307 			 * compressed dump will fit. If that doesn't turn out to
1308 			 * be enough, the bounds checking in dump_write()
1309 			 * will catch us and cause the dump to fail.
1310 			 */
1311 			dumpextent = di->mediasize - SIZEOF_METADATA -
1312 			    2 * di->blocksize - keysize;
1313 			kdh->dumpextent = htod64(dumpextent);
1314 		} else
1315 			return (E2BIG);
1316 	}
1317 
1318 	/* The offset at which to begin writing the dump. */
1319 	di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1320 	    dumpextent;
1321 
1322 	return (0);
1323 }
1324 
1325 static int
1326 _dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1327     size_t length)
1328 {
1329 	int error;
1330 
1331 #ifdef EKCD
1332 	if (di->kdcrypto != NULL)
1333 		error = dump_encrypted_write(di, virtual, physical, di->dumpoff,
1334 		    length);
1335 	else
1336 #endif
1337 		error = dump_write(di, virtual, physical, di->dumpoff, length);
1338 	if (error == 0)
1339 		di->dumpoff += length;
1340 	return (error);
1341 }
1342 
1343 /*
1344  * Write to the dump device starting at dumpoff. When compression is enabled,
1345  * writes to the device will be performed using a callback that gets invoked
1346  * when the compression stream's output buffer is full.
1347  */
1348 int
1349 dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1350     size_t length)
1351 {
1352 	void *buf;
1353 
1354 	if (di->kdcomp != NULL) {
1355 		/* Bounce through a buffer to avoid CRC errors. */
1356 		if (length > di->maxiosize)
1357 			return (EINVAL);
1358 		buf = di->kdcomp->kdc_buf;
1359 		memmove(buf, virtual, length);
1360 		return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1361 	}
1362 	return (_dump_append(di, virtual, physical, length));
1363 }
1364 
1365 /*
1366  * Write to the dump device at the specified offset.
1367  */
1368 int
1369 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1370     off_t offset, size_t length)
1371 {
1372 	int error;
1373 
1374 	error = dump_check_bounds(di, offset, length);
1375 	if (error != 0)
1376 		return (error);
1377 	return (di->dumper(di->priv, virtual, physical, offset, length));
1378 }
1379 
1380 /*
1381  * Perform kernel dump finalization: flush the compression stream, if necessary,
1382  * write the leading and trailing kernel dump headers now that we know the true
1383  * length of the dump, and optionally write the encryption key following the
1384  * leading header.
1385  */
1386 int
1387 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1388 {
1389 	uint64_t extent;
1390 	uint32_t keysize;
1391 	int error;
1392 
1393 	extent = dtoh64(kdh->dumpextent);
1394 
1395 #ifdef EKCD
1396 	keysize = kerneldumpcrypto_dumpkeysize(di->kdcrypto);
1397 #else
1398 	keysize = 0;
1399 #endif
1400 
1401 	if (di->kdcomp != NULL) {
1402 		error = compressor_flush(di->kdcomp->kdc_stream);
1403 		if (error == EAGAIN) {
1404 			/* We have residual data in di->blockbuf. */
1405 			error = dump_write(di, di->blockbuf, 0, di->dumpoff,
1406 			    di->blocksize);
1407 			di->dumpoff += di->kdcomp->kdc_resid;
1408 			di->kdcomp->kdc_resid = 0;
1409 		}
1410 		if (error != 0)
1411 			return (error);
1412 
1413 		/*
1414 		 * We now know the size of the compressed dump, so update the
1415 		 * header accordingly and recompute parity.
1416 		 */
1417 		kdh->dumplength = htod64(di->dumpoff -
1418 		    (di->mediaoffset + di->mediasize - di->blocksize - extent));
1419 		kdh->parity = 0;
1420 		kdh->parity = kerneldump_parity(kdh);
1421 
1422 		compressor_reset(di->kdcomp->kdc_stream);
1423 	}
1424 
1425 	/*
1426 	 * Write kerneldump headers at the beginning and end of the dump extent.
1427 	 * Write the key after the leading header.
1428 	 */
1429 	error = dump_write_header(di, kdh,
1430 	    di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1431 	    keysize);
1432 	if (error != 0)
1433 		return (error);
1434 
1435 #ifdef EKCD
1436 	error = dump_write_key(di,
1437 	    di->mediaoffset + di->mediasize - di->blocksize - extent - keysize);
1438 	if (error != 0)
1439 		return (error);
1440 #endif
1441 
1442 	error = dump_write_header(di, kdh,
1443 	    di->mediaoffset + di->mediasize - di->blocksize);
1444 	if (error != 0)
1445 		return (error);
1446 
1447 	(void)dump_write(di, NULL, 0, 0, 0);
1448 	return (0);
1449 }
1450 
1451 void
1452 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1453     char *magic, uint32_t archver, uint64_t dumplen)
1454 {
1455 	size_t dstsize;
1456 
1457 	bzero(kdh, sizeof(*kdh));
1458 	strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1459 	strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1460 	kdh->version = htod32(KERNELDUMPVERSION);
1461 	kdh->architectureversion = htod32(archver);
1462 	kdh->dumplength = htod64(dumplen);
1463 	kdh->dumpextent = kdh->dumplength;
1464 	kdh->dumptime = htod64(time_second);
1465 #ifdef EKCD
1466 	kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1467 #else
1468 	kdh->dumpkeysize = 0;
1469 #endif
1470 	kdh->blocksize = htod32(di->blocksize);
1471 	strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1472 	dstsize = sizeof(kdh->versionstring);
1473 	if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1474 		kdh->versionstring[dstsize - 2] = '\n';
1475 	if (panicstr != NULL)
1476 		strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1477 	if (di->kdcomp != NULL)
1478 		kdh->compression = di->kdcomp->kdc_format;
1479 	kdh->parity = kerneldump_parity(kdh);
1480 }
1481 
1482 #ifdef DDB
1483 DB_SHOW_COMMAND(panic, db_show_panic)
1484 {
1485 
1486 	if (panicstr == NULL)
1487 		db_printf("panicstr not set\n");
1488 	else
1489 		db_printf("panic: %s\n", panicstr);
1490 }
1491 #endif
1492