xref: /freebsd/sys/kern/kern_shutdown.c (revision 97cb52fa9aefd90fad38790fded50905aeeb9b9e)
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_gzio.h"
45 #include "opt_kdb.h"
46 #include "opt_panic.h"
47 #include "opt_sched.h"
48 #include "opt_watchdog.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/bio.h>
53 #include <sys/buf.h>
54 #include <sys/conf.h>
55 #include <sys/cons.h>
56 #include <sys/eventhandler.h>
57 #include <sys/filedesc.h>
58 #include <sys/gzio.h>
59 #include <sys/jail.h>
60 #include <sys/kdb.h>
61 #include <sys/kernel.h>
62 #include <sys/kerneldump.h>
63 #include <sys/kthread.h>
64 #include <sys/ktr.h>
65 #include <sys/malloc.h>
66 #include <sys/mount.h>
67 #include <sys/priv.h>
68 #include <sys/proc.h>
69 #include <sys/reboot.h>
70 #include <sys/resourcevar.h>
71 #include <sys/rwlock.h>
72 #include <sys/sched.h>
73 #include <sys/smp.h>
74 #include <sys/sysctl.h>
75 #include <sys/sysproto.h>
76 #include <sys/vnode.h>
77 #include <sys/watchdog.h>
78 
79 #include <crypto/rijndael/rijndael-api-fst.h>
80 #include <crypto/sha2/sha256.h>
81 
82 #include <ddb/ddb.h>
83 
84 #include <machine/cpu.h>
85 #include <machine/dump.h>
86 #include <machine/pcb.h>
87 #include <machine/smp.h>
88 
89 #include <security/mac/mac_framework.h>
90 
91 #include <vm/vm.h>
92 #include <vm/vm_object.h>
93 #include <vm/vm_page.h>
94 #include <vm/vm_pager.h>
95 #include <vm/swap_pager.h>
96 
97 #include <sys/signalvar.h>
98 
99 static MALLOC_DEFINE(M_DUMPER, "dumper", "dumper block buffer");
100 
101 #ifndef PANIC_REBOOT_WAIT_TIME
102 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
103 #endif
104 static int panic_reboot_wait_time = PANIC_REBOOT_WAIT_TIME;
105 SYSCTL_INT(_kern, OID_AUTO, panic_reboot_wait_time, CTLFLAG_RWTUN,
106     &panic_reboot_wait_time, 0,
107     "Seconds to wait before rebooting after a panic");
108 
109 /*
110  * Note that stdarg.h and the ANSI style va_start macro is used for both
111  * ANSI and traditional C compilers.
112  */
113 #include <machine/stdarg.h>
114 
115 #ifdef KDB
116 #ifdef KDB_UNATTENDED
117 int debugger_on_panic = 0;
118 #else
119 int debugger_on_panic = 1;
120 #endif
121 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic,
122     CTLFLAG_RWTUN | CTLFLAG_SECURE,
123     &debugger_on_panic, 0, "Run debugger on kernel panic");
124 
125 #ifdef KDB_TRACE
126 static int trace_on_panic = 1;
127 #else
128 static int trace_on_panic = 0;
129 #endif
130 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
131     CTLFLAG_RWTUN | CTLFLAG_SECURE,
132     &trace_on_panic, 0, "Print stack trace on kernel panic");
133 #endif /* KDB */
134 
135 static int sync_on_panic = 0;
136 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN,
137 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
138 
139 static bool poweroff_on_panic = 0;
140 SYSCTL_BOOL(_kern, OID_AUTO, poweroff_on_panic, CTLFLAG_RWTUN,
141 	&poweroff_on_panic, 0, "Do a power off instead of a reboot on a panic");
142 
143 static bool powercycle_on_panic = 0;
144 SYSCTL_BOOL(_kern, OID_AUTO, powercycle_on_panic, CTLFLAG_RWTUN,
145 	&powercycle_on_panic, 0, "Do a power cycle instead of a reboot on a panic");
146 
147 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0,
148     "Shutdown environment");
149 
150 #ifndef DIAGNOSTIC
151 static int show_busybufs;
152 #else
153 static int show_busybufs = 1;
154 #endif
155 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
156 	&show_busybufs, 0, "");
157 
158 int suspend_blocked = 0;
159 SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW,
160 	&suspend_blocked, 0, "Block suspend due to a pending shutdown");
161 
162 #ifdef EKCD
163 FEATURE(ekcd, "Encrypted kernel crash dumps support");
164 
165 MALLOC_DEFINE(M_EKCD, "ekcd", "Encrypted kernel crash dumps data");
166 
167 struct kerneldumpcrypto {
168 	uint8_t			kdc_encryption;
169 	uint8_t			kdc_iv[KERNELDUMP_IV_MAX_SIZE];
170 	keyInstance		kdc_ki;
171 	cipherInstance		kdc_ci;
172 	uint32_t		kdc_dumpkeysize;
173 	struct kerneldumpkey	kdc_dumpkey[];
174 };
175 #endif
176 
177 #ifdef GZIO
178 struct kerneldumpgz {
179 	struct gzio_stream	*kdgz_stream;
180 	uint8_t			*kdgz_buf;
181 	size_t			kdgz_resid;
182 };
183 
184 static struct kerneldumpgz *kerneldumpgz_create(struct dumperinfo *di,
185 		    uint8_t compression);
186 static void	kerneldumpgz_destroy(struct dumperinfo *di);
187 static int	kerneldumpgz_write_cb(void *cb, size_t len, off_t off, void *arg);
188 
189 static int kerneldump_gzlevel = 6;
190 SYSCTL_INT(_kern, OID_AUTO, kerneldump_gzlevel, CTLFLAG_RWTUN,
191     &kerneldump_gzlevel, 0,
192     "Kernel crash dump gzip compression level");
193 #endif /* GZIO */
194 
195 /*
196  * Variable panicstr contains argument to first call to panic; used as flag
197  * to indicate that the kernel has already called panic.
198  */
199 const char *panicstr;
200 
201 int dumping;				/* system is dumping */
202 int rebooting;				/* system is rebooting */
203 static struct dumperinfo dumper;	/* our selected dumper */
204 
205 /* Context information for dump-debuggers. */
206 static struct pcb dumppcb;		/* Registers. */
207 lwpid_t dumptid;			/* Thread ID. */
208 
209 static struct cdevsw reroot_cdevsw = {
210      .d_version = D_VERSION,
211      .d_name    = "reroot",
212 };
213 
214 static void poweroff_wait(void *, int);
215 static void shutdown_halt(void *junk, int howto);
216 static void shutdown_panic(void *junk, int howto);
217 static void shutdown_reset(void *junk, int howto);
218 static int kern_reroot(void);
219 
220 /* register various local shutdown events */
221 static void
222 shutdown_conf(void *unused)
223 {
224 
225 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
226 	    SHUTDOWN_PRI_FIRST);
227 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
228 	    SHUTDOWN_PRI_LAST + 100);
229 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
230 	    SHUTDOWN_PRI_LAST + 100);
231 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
232 	    SHUTDOWN_PRI_LAST + 200);
233 }
234 
235 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
236 
237 /*
238  * The only reason this exists is to create the /dev/reroot/ directory,
239  * used by reroot code in init(8) as a mountpoint for tmpfs.
240  */
241 static void
242 reroot_conf(void *unused)
243 {
244 	int error;
245 	struct cdev *cdev;
246 
247 	error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev,
248 	    &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot");
249 	if (error != 0) {
250 		printf("%s: failed to create device node, error %d",
251 		    __func__, error);
252 	}
253 }
254 
255 SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL);
256 
257 /*
258  * The system call that results in a reboot.
259  */
260 /* ARGSUSED */
261 int
262 sys_reboot(struct thread *td, struct reboot_args *uap)
263 {
264 	int error;
265 
266 	error = 0;
267 #ifdef MAC
268 	error = mac_system_check_reboot(td->td_ucred, uap->opt);
269 #endif
270 	if (error == 0)
271 		error = priv_check(td, PRIV_REBOOT);
272 	if (error == 0) {
273 		if (uap->opt & RB_REROOT) {
274 			error = kern_reroot();
275 		} else {
276 			mtx_lock(&Giant);
277 			kern_reboot(uap->opt);
278 			mtx_unlock(&Giant);
279 		}
280 	}
281 	return (error);
282 }
283 
284 /*
285  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
286  */
287 void
288 shutdown_nice(int howto)
289 {
290 
291 	if (initproc != NULL) {
292 		/* Send a signal to init(8) and have it shutdown the world. */
293 		PROC_LOCK(initproc);
294 		if (howto & RB_POWEROFF)
295 			kern_psignal(initproc, SIGUSR2);
296 		else if (howto & RB_POWERCYCLE)
297 			kern_psignal(initproc, SIGWINCH);
298 		else if (howto & RB_HALT)
299 			kern_psignal(initproc, SIGUSR1);
300 		else
301 			kern_psignal(initproc, SIGINT);
302 		PROC_UNLOCK(initproc);
303 	} else {
304 		/* No init(8) running, so simply reboot. */
305 		kern_reboot(howto | RB_NOSYNC);
306 	}
307 }
308 
309 static void
310 print_uptime(void)
311 {
312 	int f;
313 	struct timespec ts;
314 
315 	getnanouptime(&ts);
316 	printf("Uptime: ");
317 	f = 0;
318 	if (ts.tv_sec >= 86400) {
319 		printf("%ldd", (long)ts.tv_sec / 86400);
320 		ts.tv_sec %= 86400;
321 		f = 1;
322 	}
323 	if (f || ts.tv_sec >= 3600) {
324 		printf("%ldh", (long)ts.tv_sec / 3600);
325 		ts.tv_sec %= 3600;
326 		f = 1;
327 	}
328 	if (f || ts.tv_sec >= 60) {
329 		printf("%ldm", (long)ts.tv_sec / 60);
330 		ts.tv_sec %= 60;
331 		f = 1;
332 	}
333 	printf("%lds\n", (long)ts.tv_sec);
334 }
335 
336 int
337 doadump(boolean_t textdump)
338 {
339 	boolean_t coredump;
340 	int error;
341 
342 	error = 0;
343 	if (dumping)
344 		return (EBUSY);
345 	if (dumper.dumper == NULL)
346 		return (ENXIO);
347 
348 	savectx(&dumppcb);
349 	dumptid = curthread->td_tid;
350 	dumping++;
351 
352 	coredump = TRUE;
353 #ifdef DDB
354 	if (textdump && textdump_pending) {
355 		coredump = FALSE;
356 		textdump_dumpsys(&dumper);
357 	}
358 #endif
359 	if (coredump)
360 		error = dumpsys(&dumper);
361 
362 	dumping--;
363 	return (error);
364 }
365 
366 /*
367  * Shutdown the system cleanly to prepare for reboot, halt, or power off.
368  */
369 void
370 kern_reboot(int howto)
371 {
372 	static int once = 0;
373 
374 #if defined(SMP)
375 	/*
376 	 * Bind us to the first CPU so that all shutdown code runs there.  Some
377 	 * systems don't shutdown properly (i.e., ACPI power off) if we
378 	 * run on another processor.
379 	 */
380 	if (!SCHEDULER_STOPPED()) {
381 		thread_lock(curthread);
382 		sched_bind(curthread, CPU_FIRST());
383 		thread_unlock(curthread);
384 		KASSERT(PCPU_GET(cpuid) == CPU_FIRST(),
385 		    ("boot: not running on cpu 0"));
386 	}
387 #endif
388 	/* We're in the process of rebooting. */
389 	rebooting = 1;
390 
391 	/* We are out of the debugger now. */
392 	kdb_active = 0;
393 
394 	/*
395 	 * Do any callouts that should be done BEFORE syncing the filesystems.
396 	 */
397 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
398 
399 	/*
400 	 * Now sync filesystems
401 	 */
402 	if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
403 		once = 1;
404 		bufshutdown(show_busybufs);
405 	}
406 
407 	print_uptime();
408 
409 	cngrab();
410 
411 	/*
412 	 * Ok, now do things that assume all filesystem activity has
413 	 * been completed.
414 	 */
415 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
416 
417 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
418 		doadump(TRUE);
419 
420 	/* Now that we're going to really halt the system... */
421 	EVENTHANDLER_INVOKE(shutdown_final, howto);
422 
423 	for(;;) ;	/* safety against shutdown_reset not working */
424 	/* NOTREACHED */
425 }
426 
427 /*
428  * The system call that results in changing the rootfs.
429  */
430 static int
431 kern_reroot(void)
432 {
433 	struct vnode *oldrootvnode, *vp;
434 	struct mount *mp, *devmp;
435 	int error;
436 
437 	if (curproc != initproc)
438 		return (EPERM);
439 
440 	/*
441 	 * Mark the filesystem containing currently-running executable
442 	 * (the temporary copy of init(8)) busy.
443 	 */
444 	vp = curproc->p_textvp;
445 	error = vn_lock(vp, LK_SHARED);
446 	if (error != 0)
447 		return (error);
448 	mp = vp->v_mount;
449 	error = vfs_busy(mp, MBF_NOWAIT);
450 	if (error != 0) {
451 		vfs_ref(mp);
452 		VOP_UNLOCK(vp, 0);
453 		error = vfs_busy(mp, 0);
454 		vn_lock(vp, LK_SHARED | LK_RETRY);
455 		vfs_rel(mp);
456 		if (error != 0) {
457 			VOP_UNLOCK(vp, 0);
458 			return (ENOENT);
459 		}
460 		if (vp->v_iflag & VI_DOOMED) {
461 			VOP_UNLOCK(vp, 0);
462 			vfs_unbusy(mp);
463 			return (ENOENT);
464 		}
465 	}
466 	VOP_UNLOCK(vp, 0);
467 
468 	/*
469 	 * Remove the filesystem containing currently-running executable
470 	 * from the mount list, to prevent it from being unmounted
471 	 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
472 	 *
473 	 * Also preserve /dev - forcibly unmounting it could cause driver
474 	 * reinitialization.
475 	 */
476 
477 	vfs_ref(rootdevmp);
478 	devmp = rootdevmp;
479 	rootdevmp = NULL;
480 
481 	mtx_lock(&mountlist_mtx);
482 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
483 	TAILQ_REMOVE(&mountlist, devmp, mnt_list);
484 	mtx_unlock(&mountlist_mtx);
485 
486 	oldrootvnode = rootvnode;
487 
488 	/*
489 	 * Unmount everything except for the two filesystems preserved above.
490 	 */
491 	vfs_unmountall();
492 
493 	/*
494 	 * Add /dev back; vfs_mountroot() will move it into its new place.
495 	 */
496 	mtx_lock(&mountlist_mtx);
497 	TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
498 	mtx_unlock(&mountlist_mtx);
499 	rootdevmp = devmp;
500 	vfs_rel(rootdevmp);
501 
502 	/*
503 	 * Mount the new rootfs.
504 	 */
505 	vfs_mountroot();
506 
507 	/*
508 	 * Update all references to the old rootvnode.
509 	 */
510 	mountcheckdirs(oldrootvnode, rootvnode);
511 
512 	/*
513 	 * Add the temporary filesystem back and unbusy it.
514 	 */
515 	mtx_lock(&mountlist_mtx);
516 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
517 	mtx_unlock(&mountlist_mtx);
518 	vfs_unbusy(mp);
519 
520 	return (0);
521 }
522 
523 /*
524  * If the shutdown was a clean halt, behave accordingly.
525  */
526 static void
527 shutdown_halt(void *junk, int howto)
528 {
529 
530 	if (howto & RB_HALT) {
531 		printf("\n");
532 		printf("The operating system has halted.\n");
533 		printf("Please press any key to reboot.\n\n");
534 		switch (cngetc()) {
535 		case -1:		/* No console, just die */
536 			cpu_halt();
537 			/* NOTREACHED */
538 		default:
539 			howto &= ~RB_HALT;
540 			break;
541 		}
542 	}
543 }
544 
545 /*
546  * Check to see if the system paniced, pause and then reboot
547  * according to the specified delay.
548  */
549 static void
550 shutdown_panic(void *junk, int howto)
551 {
552 	int loop;
553 
554 	if (howto & RB_DUMP) {
555 		if (panic_reboot_wait_time != 0) {
556 			if (panic_reboot_wait_time != -1) {
557 				printf("Automatic reboot in %d seconds - "
558 				       "press a key on the console to abort\n",
559 					panic_reboot_wait_time);
560 				for (loop = panic_reboot_wait_time * 10;
561 				     loop > 0; --loop) {
562 					DELAY(1000 * 100); /* 1/10th second */
563 					/* Did user type a key? */
564 					if (cncheckc() != -1)
565 						break;
566 				}
567 				if (!loop)
568 					return;
569 			}
570 		} else { /* zero time specified - reboot NOW */
571 			return;
572 		}
573 		printf("--> Press a key on the console to reboot,\n");
574 		printf("--> or switch off the system now.\n");
575 		cngetc();
576 	}
577 }
578 
579 /*
580  * Everything done, now reset
581  */
582 static void
583 shutdown_reset(void *junk, int howto)
584 {
585 
586 	printf("Rebooting...\n");
587 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
588 
589 	/*
590 	 * Acquiring smp_ipi_mtx here has a double effect:
591 	 * - it disables interrupts avoiding CPU0 preemption
592 	 *   by fast handlers (thus deadlocking  against other CPUs)
593 	 * - it avoids deadlocks against smp_rendezvous() or, more
594 	 *   generally, threads busy-waiting, with this spinlock held,
595 	 *   and waiting for responses by threads on other CPUs
596 	 *   (ie. smp_tlb_shootdown()).
597 	 *
598 	 * For the !SMP case it just needs to handle the former problem.
599 	 */
600 #ifdef SMP
601 	mtx_lock_spin(&smp_ipi_mtx);
602 #else
603 	spinlock_enter();
604 #endif
605 
606 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
607 	cpu_reset();
608 	/* NOTREACHED */ /* assuming reset worked */
609 }
610 
611 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
612 static int kassert_warn_only = 0;
613 #ifdef KDB
614 static int kassert_do_kdb = 0;
615 #endif
616 #ifdef KTR
617 static int kassert_do_ktr = 0;
618 #endif
619 static int kassert_do_log = 1;
620 static int kassert_log_pps_limit = 4;
621 static int kassert_log_mute_at = 0;
622 static int kassert_log_panic_at = 0;
623 static int kassert_warnings = 0;
624 
625 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW, NULL, "kassert options");
626 
627 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, CTLFLAG_RWTUN,
628     &kassert_warn_only, 0,
629     "KASSERT triggers a panic (1) or just a warning (0)");
630 
631 #ifdef KDB
632 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, CTLFLAG_RWTUN,
633     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
634 #endif
635 
636 #ifdef KTR
637 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, CTLFLAG_RWTUN,
638     &kassert_do_ktr, 0,
639     "KASSERT does a KTR, set this to the KTRMASK you want");
640 #endif
641 
642 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, CTLFLAG_RWTUN,
643     &kassert_do_log, 0, "KASSERT triggers a panic (1) or just a warning (0)");
644 
645 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RWTUN,
646     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
647 
648 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, CTLFLAG_RWTUN,
649     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
650 
651 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, CTLFLAG_RWTUN,
652     &kassert_log_pps_limit, 0, "limit number of log messages per second");
653 
654 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, CTLFLAG_RWTUN,
655     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
656 
657 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
658 
659 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
660     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE, NULL, 0,
661     kassert_sysctl_kassert, "I", "set to trigger a test kassert");
662 
663 static int
664 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
665 {
666 	int error, i;
667 
668 	error = sysctl_wire_old_buffer(req, sizeof(int));
669 	if (error == 0) {
670 		i = 0;
671 		error = sysctl_handle_int(oidp, &i, 0, req);
672 	}
673 	if (error != 0 || req->newptr == NULL)
674 		return (error);
675 	KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
676 	return (0);
677 }
678 
679 /*
680  * Called by KASSERT, this decides if we will panic
681  * or if we will log via printf and/or ktr.
682  */
683 void
684 kassert_panic(const char *fmt, ...)
685 {
686 	static char buf[256];
687 	va_list ap;
688 
689 	va_start(ap, fmt);
690 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
691 	va_end(ap);
692 
693 	/*
694 	 * panic if we're not just warning, or if we've exceeded
695 	 * kassert_log_panic_at warnings.
696 	 */
697 	if (!kassert_warn_only ||
698 	    (kassert_log_panic_at > 0 &&
699 	     kassert_warnings >= kassert_log_panic_at)) {
700 		va_start(ap, fmt);
701 		vpanic(fmt, ap);
702 		/* NORETURN */
703 	}
704 #ifdef KTR
705 	if (kassert_do_ktr)
706 		CTR0(ktr_mask, buf);
707 #endif /* KTR */
708 	/*
709 	 * log if we've not yet met the mute limit.
710 	 */
711 	if (kassert_do_log &&
712 	    (kassert_log_mute_at == 0 ||
713 	     kassert_warnings < kassert_log_mute_at)) {
714 		static  struct timeval lasterr;
715 		static  int curerr;
716 
717 		if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
718 			printf("KASSERT failed: %s\n", buf);
719 			kdb_backtrace();
720 		}
721 	}
722 #ifdef KDB
723 	if (kassert_do_kdb) {
724 		kdb_enter(KDB_WHY_KASSERT, buf);
725 	}
726 #endif
727 	atomic_add_int(&kassert_warnings, 1);
728 }
729 #endif
730 
731 /*
732  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
733  * and then reboots.  If we are called twice, then we avoid trying to sync
734  * the disks as this often leads to recursive panics.
735  */
736 void
737 panic(const char *fmt, ...)
738 {
739 	va_list ap;
740 
741 	va_start(ap, fmt);
742 	vpanic(fmt, ap);
743 }
744 
745 void
746 vpanic(const char *fmt, va_list ap)
747 {
748 #ifdef SMP
749 	cpuset_t other_cpus;
750 #endif
751 	struct thread *td = curthread;
752 	int bootopt, newpanic;
753 	static char buf[256];
754 
755 	spinlock_enter();
756 
757 #ifdef SMP
758 	/*
759 	 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
760 	 * concurrently entering panic.  Only the winner will proceed
761 	 * further.
762 	 */
763 	if (panicstr == NULL && !kdb_active) {
764 		other_cpus = all_cpus;
765 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
766 		stop_cpus_hard(other_cpus);
767 	}
768 #endif
769 
770 	/*
771 	 * Ensure that the scheduler is stopped while panicking, even if panic
772 	 * has been entered from kdb.
773 	 */
774 	td->td_stopsched = 1;
775 
776 	bootopt = RB_AUTOBOOT;
777 	newpanic = 0;
778 	if (panicstr)
779 		bootopt |= RB_NOSYNC;
780 	else {
781 		bootopt |= RB_DUMP;
782 		panicstr = fmt;
783 		newpanic = 1;
784 	}
785 
786 	if (newpanic) {
787 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
788 		panicstr = buf;
789 		cngrab();
790 		printf("panic: %s\n", buf);
791 	} else {
792 		printf("panic: ");
793 		vprintf(fmt, ap);
794 		printf("\n");
795 	}
796 #ifdef SMP
797 	printf("cpuid = %d\n", PCPU_GET(cpuid));
798 #endif
799 	printf("time = %jd\n", (intmax_t )time_second);
800 #ifdef KDB
801 	if (newpanic && trace_on_panic)
802 		kdb_backtrace();
803 	if (debugger_on_panic)
804 		kdb_enter(KDB_WHY_PANIC, "panic");
805 #endif
806 	/*thread_lock(td); */
807 	td->td_flags |= TDF_INPANIC;
808 	/* thread_unlock(td); */
809 	if (!sync_on_panic)
810 		bootopt |= RB_NOSYNC;
811 	if (poweroff_on_panic)
812 		bootopt |= RB_POWEROFF;
813 	if (powercycle_on_panic)
814 		bootopt |= RB_POWERCYCLE;
815 	kern_reboot(bootopt);
816 }
817 
818 /*
819  * Support for poweroff delay.
820  *
821  * Please note that setting this delay too short might power off your machine
822  * before the write cache on your hard disk has been flushed, leading to
823  * soft-updates inconsistencies.
824  */
825 #ifndef POWEROFF_DELAY
826 # define POWEROFF_DELAY 5000
827 #endif
828 static int poweroff_delay = POWEROFF_DELAY;
829 
830 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
831     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
832 
833 static void
834 poweroff_wait(void *junk, int howto)
835 {
836 
837 	if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
838 		return;
839 	DELAY(poweroff_delay * 1000);
840 }
841 
842 /*
843  * Some system processes (e.g. syncer) need to be stopped at appropriate
844  * points in their main loops prior to a system shutdown, so that they
845  * won't interfere with the shutdown process (e.g. by holding a disk buf
846  * to cause sync to fail).  For each of these system processes, register
847  * shutdown_kproc() as a handler for one of shutdown events.
848  */
849 static int kproc_shutdown_wait = 60;
850 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
851     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
852 
853 void
854 kproc_shutdown(void *arg, int howto)
855 {
856 	struct proc *p;
857 	int error;
858 
859 	if (panicstr)
860 		return;
861 
862 	p = (struct proc *)arg;
863 	printf("Waiting (max %d seconds) for system process `%s' to stop... ",
864 	    kproc_shutdown_wait, p->p_comm);
865 	error = kproc_suspend(p, kproc_shutdown_wait * hz);
866 
867 	if (error == EWOULDBLOCK)
868 		printf("timed out\n");
869 	else
870 		printf("done\n");
871 }
872 
873 void
874 kthread_shutdown(void *arg, int howto)
875 {
876 	struct thread *td;
877 	int error;
878 
879 	if (panicstr)
880 		return;
881 
882 	td = (struct thread *)arg;
883 	printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
884 	    kproc_shutdown_wait, td->td_name);
885 	error = kthread_suspend(td, kproc_shutdown_wait * hz);
886 
887 	if (error == EWOULDBLOCK)
888 		printf("timed out\n");
889 	else
890 		printf("done\n");
891 }
892 
893 static char dumpdevname[sizeof(((struct cdev*)NULL)->si_name)];
894 SYSCTL_STRING(_kern_shutdown, OID_AUTO, dumpdevname, CTLFLAG_RD,
895     dumpdevname, 0, "Device for kernel dumps");
896 
897 static int	_dump_append(struct dumperinfo *di, void *virtual,
898 		    vm_offset_t physical, size_t length);
899 
900 #ifdef EKCD
901 static struct kerneldumpcrypto *
902 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
903     const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
904 {
905 	struct kerneldumpcrypto *kdc;
906 	struct kerneldumpkey *kdk;
907 	uint32_t dumpkeysize;
908 
909 	dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
910 	kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
911 
912 	arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
913 
914 	kdc->kdc_encryption = encryption;
915 	switch (kdc->kdc_encryption) {
916 	case KERNELDUMP_ENC_AES_256_CBC:
917 		if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
918 			goto failed;
919 		break;
920 	default:
921 		goto failed;
922 	}
923 
924 	kdc->kdc_dumpkeysize = dumpkeysize;
925 	kdk = kdc->kdc_dumpkey;
926 	kdk->kdk_encryption = kdc->kdc_encryption;
927 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
928 	kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
929 	memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
930 
931 	return (kdc);
932 failed:
933 	explicit_bzero(kdc, sizeof(*kdc) + dumpkeysize);
934 	free(kdc, M_EKCD);
935 	return (NULL);
936 }
937 
938 static int
939 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
940 {
941 	uint8_t hash[SHA256_DIGEST_LENGTH];
942 	SHA256_CTX ctx;
943 	struct kerneldumpkey *kdk;
944 	int error;
945 
946 	error = 0;
947 
948 	if (kdc == NULL)
949 		return (0);
950 
951 	/*
952 	 * When a user enters ddb it can write a crash dump multiple times.
953 	 * Each time it should be encrypted using a different IV.
954 	 */
955 	SHA256_Init(&ctx);
956 	SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
957 	SHA256_Final(hash, &ctx);
958 	bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
959 
960 	switch (kdc->kdc_encryption) {
961 	case KERNELDUMP_ENC_AES_256_CBC:
962 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
963 		    kdc->kdc_iv) <= 0) {
964 			error = EINVAL;
965 			goto out;
966 		}
967 		break;
968 	default:
969 		error = EINVAL;
970 		goto out;
971 	}
972 
973 	kdk = kdc->kdc_dumpkey;
974 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
975 out:
976 	explicit_bzero(hash, sizeof(hash));
977 	return (error);
978 }
979 
980 static uint32_t
981 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
982 {
983 
984 	if (kdc == NULL)
985 		return (0);
986 	return (kdc->kdc_dumpkeysize);
987 }
988 #endif /* EKCD */
989 
990 #ifdef GZIO
991 static struct kerneldumpgz *
992 kerneldumpgz_create(struct dumperinfo *di, uint8_t compression)
993 {
994 	struct kerneldumpgz *kdgz;
995 
996 	if (compression != KERNELDUMP_COMP_GZIP)
997 		return (NULL);
998 	kdgz = malloc(sizeof(*kdgz), M_DUMPER, M_WAITOK | M_ZERO);
999 	kdgz->kdgz_stream = gzio_init(kerneldumpgz_write_cb, GZIO_DEFLATE,
1000 	    di->maxiosize, kerneldump_gzlevel, di);
1001 	if (kdgz->kdgz_stream == NULL) {
1002 		free(kdgz, M_DUMPER);
1003 		return (NULL);
1004 	}
1005 	kdgz->kdgz_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1006 	return (kdgz);
1007 }
1008 
1009 static void
1010 kerneldumpgz_destroy(struct dumperinfo *di)
1011 {
1012 	struct kerneldumpgz *kdgz;
1013 
1014 	kdgz = di->kdgz;
1015 	if (kdgz == NULL)
1016 		return;
1017 	gzio_fini(kdgz->kdgz_stream);
1018 	explicit_bzero(kdgz->kdgz_buf, di->maxiosize);
1019 	free(kdgz->kdgz_buf, M_DUMPER);
1020 	free(kdgz, M_DUMPER);
1021 }
1022 #endif /* GZIO */
1023 
1024 /* Registration of dumpers */
1025 int
1026 set_dumper(struct dumperinfo *di, const char *devname, struct thread *td,
1027     uint8_t compression, uint8_t encryption, const uint8_t *key,
1028     uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1029 {
1030 	size_t wantcopy;
1031 	int error;
1032 
1033 	error = priv_check(td, PRIV_SETDUMPER);
1034 	if (error != 0)
1035 		return (error);
1036 
1037 	if (di == NULL) {
1038 		error = 0;
1039 		goto cleanup;
1040 	}
1041 	if (dumper.dumper != NULL)
1042 		return (EBUSY);
1043 	dumper = *di;
1044 	dumper.blockbuf = NULL;
1045 	dumper.kdc = NULL;
1046 	dumper.kdgz = NULL;
1047 
1048 	if (encryption != KERNELDUMP_ENC_NONE) {
1049 #ifdef EKCD
1050 		dumper.kdc = kerneldumpcrypto_create(di->blocksize, encryption,
1051 		    key, encryptedkeysize, encryptedkey);
1052 		if (dumper.kdc == NULL) {
1053 			error = EINVAL;
1054 			goto cleanup;
1055 		}
1056 #else
1057 		error = EOPNOTSUPP;
1058 		goto cleanup;
1059 #endif
1060 	}
1061 
1062 	wantcopy = strlcpy(dumpdevname, devname, sizeof(dumpdevname));
1063 	if (wantcopy >= sizeof(dumpdevname)) {
1064 		printf("set_dumper: device name truncated from '%s' -> '%s'\n",
1065 		    devname, dumpdevname);
1066 	}
1067 
1068 	if (compression != KERNELDUMP_COMP_NONE) {
1069 #ifdef GZIO
1070 		/*
1071 		 * We currently can't support simultaneous encryption and
1072 		 * compression.
1073 		 */
1074 		if (encryption != KERNELDUMP_ENC_NONE) {
1075 			error = EOPNOTSUPP;
1076 			goto cleanup;
1077 		}
1078 		dumper.kdgz = kerneldumpgz_create(&dumper, compression);
1079 		if (dumper.kdgz == NULL) {
1080 			error = EINVAL;
1081 			goto cleanup;
1082 		}
1083 #else
1084 		error = EOPNOTSUPP;
1085 		goto cleanup;
1086 #endif
1087 	}
1088 
1089 	dumper.blockbuf = malloc(di->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1090 	return (0);
1091 cleanup:
1092 #ifdef EKCD
1093 	if (dumper.kdc != NULL) {
1094 		explicit_bzero(dumper.kdc, sizeof(*dumper.kdc) +
1095 		    dumper.kdc->kdc_dumpkeysize);
1096 		free(dumper.kdc, M_EKCD);
1097 	}
1098 #endif
1099 
1100 #ifdef GZIO
1101 	kerneldumpgz_destroy(&dumper);
1102 #endif
1103 
1104 	if (dumper.blockbuf != NULL) {
1105 		explicit_bzero(dumper.blockbuf, dumper.blocksize);
1106 		free(dumper.blockbuf, M_DUMPER);
1107 	}
1108 	explicit_bzero(&dumper, sizeof(dumper));
1109 	dumpdevname[0] = '\0';
1110 	return (error);
1111 }
1112 
1113 static int
1114 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1115 {
1116 
1117 	if (length != 0 && (offset < di->mediaoffset ||
1118 	    offset - di->mediaoffset + length > di->mediasize)) {
1119 		printf("Attempt to write outside dump device boundaries.\n"
1120 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1121 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
1122 		    (uintmax_t)length, (intmax_t)di->mediasize);
1123 		return (ENOSPC);
1124 	}
1125 	if (length % di->blocksize != 0) {
1126 		printf("Attempt to write partial block of length %ju.\n",
1127 		    (uintmax_t)length);
1128 		return (EINVAL);
1129 	}
1130 	if (offset % di->blocksize != 0) {
1131 		printf("Attempt to write at unaligned offset %jd.\n",
1132 		    (intmax_t)offset);
1133 		return (EINVAL);
1134 	}
1135 
1136 	return (0);
1137 }
1138 
1139 #ifdef EKCD
1140 static int
1141 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1142 {
1143 
1144 	switch (kdc->kdc_encryption) {
1145 	case KERNELDUMP_ENC_AES_256_CBC:
1146 		if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1147 		    8 * size, buf) <= 0) {
1148 			return (EIO);
1149 		}
1150 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1151 		    buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1152 			return (EIO);
1153 		}
1154 		break;
1155 	default:
1156 		return (EINVAL);
1157 	}
1158 
1159 	return (0);
1160 }
1161 
1162 /* Encrypt data and call dumper. */
1163 static int
1164 dump_encrypted_write(struct dumperinfo *di, void *virtual,
1165     vm_offset_t physical, off_t offset, size_t length)
1166 {
1167 	static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1168 	struct kerneldumpcrypto *kdc;
1169 	int error;
1170 	size_t nbytes;
1171 
1172 	kdc = di->kdc;
1173 
1174 	while (length > 0) {
1175 		nbytes = MIN(length, sizeof(buf));
1176 		bcopy(virtual, buf, nbytes);
1177 
1178 		if (dump_encrypt(kdc, buf, nbytes) != 0)
1179 			return (EIO);
1180 
1181 		error = dump_write(di, buf, physical, offset, nbytes);
1182 		if (error != 0)
1183 			return (error);
1184 
1185 		offset += nbytes;
1186 		virtual = (void *)((uint8_t *)virtual + nbytes);
1187 		length -= nbytes;
1188 	}
1189 
1190 	return (0);
1191 }
1192 
1193 static int
1194 dump_write_key(struct dumperinfo *di, off_t offset)
1195 {
1196 	struct kerneldumpcrypto *kdc;
1197 
1198 	kdc = di->kdc;
1199 	if (kdc == NULL)
1200 		return (0);
1201 	return (dump_write(di, kdc->kdc_dumpkey, 0, offset,
1202 	    kdc->kdc_dumpkeysize));
1203 }
1204 #endif /* EKCD */
1205 
1206 #ifdef GZIO
1207 static int
1208 kerneldumpgz_write_cb(void *base, size_t length, off_t offset, void *arg)
1209 {
1210 	struct dumperinfo *di;
1211 	size_t resid, rlength;
1212 	int error;
1213 
1214 	di = arg;
1215 
1216 	if (length % di->blocksize != 0) {
1217 		/*
1218 		 * This must be the final write after flushing the compression
1219 		 * stream. Write as many full blocks as possible and stash the
1220 		 * residual data in the dumper's block buffer. It will be
1221 		 * padded and written in dump_finish().
1222 		 */
1223 		rlength = rounddown(length, di->blocksize);
1224 		if (rlength != 0) {
1225 			error = _dump_append(di, base, 0, rlength);
1226 			if (error != 0)
1227 				return (error);
1228 		}
1229 		resid = length - rlength;
1230 		memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1231 		di->kdgz->kdgz_resid = resid;
1232 		return (EAGAIN);
1233 	}
1234 	return (_dump_append(di, base, 0, length));
1235 }
1236 #endif /* GZIO */
1237 
1238 /*
1239  * Write a kerneldumpheader at the specified offset. The header structure is 512
1240  * bytes in size, but we must pad to the device sector size.
1241  */
1242 static int
1243 dump_write_header(struct dumperinfo *di, struct kerneldumpheader *kdh,
1244     off_t offset)
1245 {
1246 	void *buf;
1247 	size_t hdrsz;
1248 
1249 	hdrsz = sizeof(*kdh);
1250 	if (hdrsz > di->blocksize)
1251 		return (ENOMEM);
1252 
1253 	if (hdrsz == di->blocksize)
1254 		buf = kdh;
1255 	else {
1256 		buf = di->blockbuf;
1257 		memset(buf, 0, di->blocksize);
1258 		memcpy(buf, kdh, hdrsz);
1259 	}
1260 
1261 	return (dump_write(di, buf, 0, offset, di->blocksize));
1262 }
1263 
1264 /*
1265  * Don't touch the first SIZEOF_METADATA bytes on the dump device.  This is to
1266  * protect us from metadata and metadata from us.
1267  */
1268 #define	SIZEOF_METADATA		(64 * 1024)
1269 
1270 /*
1271  * Do some preliminary setup for a kernel dump: initialize state for encryption,
1272  * if requested, and make sure that we have enough space on the dump device.
1273  *
1274  * We set things up so that the dump ends before the last sector of the dump
1275  * device, at which the trailing header is written.
1276  *
1277  *     +-----------+------+-----+----------------------------+------+
1278  *     |           | lhdr | key |    ... kernel dump ...     | thdr |
1279  *     +-----------+------+-----+----------------------------+------+
1280  *                   1 blk  opt <------- dump extent --------> 1 blk
1281  *
1282  * Dumps written using dump_append() start at the beginning of the extent.
1283  * Uncompressed dumps will use the entire extent, but compressed dumps typically
1284  * will not. The true length of the dump is recorded in the leading and trailing
1285  * headers once the dump has been completed.
1286  */
1287 int
1288 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1289 {
1290 	uint64_t dumpextent;
1291 	uint32_t keysize;
1292 
1293 #ifdef EKCD
1294 	int error = kerneldumpcrypto_init(di->kdc);
1295 	if (error != 0)
1296 		return (error);
1297 	keysize = kerneldumpcrypto_dumpkeysize(di->kdc);
1298 #else
1299 	keysize = 0;
1300 #endif
1301 
1302 	dumpextent = dtoh64(kdh->dumpextent);
1303 	if (di->mediasize < SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1304 	    keysize) {
1305 #ifdef GZIO
1306 		if (di->kdgz != NULL) {
1307 			/*
1308 			 * We don't yet know how much space the compressed dump
1309 			 * will occupy, so try to use the whole swap partition
1310 			 * (minus the first 64KB) in the hope that the
1311 			 * compressed dump will fit. If that doesn't turn out to
1312 			 * be enouch, the bounds checking in dump_write()
1313 			 * will catch us and cause the dump to fail.
1314 			 */
1315 			dumpextent = di->mediasize - SIZEOF_METADATA -
1316 			    2 * di->blocksize - keysize;
1317 			kdh->dumpextent = htod64(dumpextent);
1318 		} else
1319 #endif
1320 			return (E2BIG);
1321 	}
1322 
1323 	/* The offset at which to begin writing the dump. */
1324 	di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1325 	    dumpextent;
1326 
1327 	return (0);
1328 }
1329 
1330 static int
1331 _dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1332     size_t length)
1333 {
1334 	int error;
1335 
1336 #ifdef EKCD
1337 	if (di->kdc != NULL)
1338 		error = dump_encrypted_write(di, virtual, physical, di->dumpoff,
1339 		    length);
1340 	else
1341 #endif
1342 		error = dump_write(di, virtual, physical, di->dumpoff, length);
1343 	if (error == 0)
1344 		di->dumpoff += length;
1345 	return (error);
1346 }
1347 
1348 /*
1349  * Write to the dump device starting at dumpoff. When compression is enabled,
1350  * writes to the device will be performed using a callback that gets invoked
1351  * when the compression stream's output buffer is full.
1352  */
1353 int
1354 dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1355     size_t length)
1356 {
1357 #ifdef GZIO
1358 	void *buf;
1359 
1360 	if (di->kdgz != NULL) {
1361 		/* Bounce through a buffer to avoid gzip CRC errors. */
1362 		if (length > di->maxiosize)
1363 			return (EINVAL);
1364 		buf = di->kdgz->kdgz_buf;
1365 		memmove(buf, virtual, length);
1366 		return (gzio_write(di->kdgz->kdgz_stream, buf, length));
1367 	}
1368 #endif
1369 	return (_dump_append(di, virtual, physical, length));
1370 }
1371 
1372 /*
1373  * Write to the dump device at the specified offset.
1374  */
1375 int
1376 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1377     off_t offset, size_t length)
1378 {
1379 	int error;
1380 
1381 	error = dump_check_bounds(di, offset, length);
1382 	if (error != 0)
1383 		return (error);
1384 	return (di->dumper(di->priv, virtual, physical, offset, length));
1385 }
1386 
1387 /*
1388  * Perform kernel dump finalization: flush the compression stream, if necessary,
1389  * write the leading and trailing kernel dump headers now that we know the true
1390  * length of the dump, and optionally write the encryption key following the
1391  * leading header.
1392  */
1393 int
1394 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1395 {
1396 	uint64_t extent;
1397 	uint32_t keysize;
1398 	int error;
1399 
1400 	extent = dtoh64(kdh->dumpextent);
1401 
1402 #ifdef EKCD
1403 	keysize = kerneldumpcrypto_dumpkeysize(di->kdc);
1404 #else
1405 	keysize = 0;
1406 #endif
1407 
1408 #ifdef GZIO
1409 	if (di->kdgz != NULL) {
1410 		error = gzio_flush(di->kdgz->kdgz_stream);
1411 		if (error == EAGAIN) {
1412 			/* We have residual data in di->blockbuf. */
1413 			error = dump_write(di, di->blockbuf, 0, di->dumpoff,
1414 			    di->blocksize);
1415 			di->dumpoff += di->kdgz->kdgz_resid;
1416 			di->kdgz->kdgz_resid = 0;
1417 		}
1418 		if (error != 0)
1419 			return (error);
1420 
1421 		/*
1422 		 * We now know the size of the compressed dump, so update the
1423 		 * header accordingly and recompute parity.
1424 		 */
1425 		kdh->dumplength = htod64(di->dumpoff -
1426 		    (di->mediaoffset + di->mediasize - di->blocksize - extent));
1427 		kdh->parity = 0;
1428 		kdh->parity = kerneldump_parity(kdh);
1429 
1430 		gzio_reset(di->kdgz->kdgz_stream);
1431 	}
1432 #endif
1433 
1434 	/*
1435 	 * Write kerneldump headers at the beginning and end of the dump extent.
1436 	 * Write the key after the leading header.
1437 	 */
1438 	error = dump_write_header(di, kdh,
1439 	    di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1440 	    keysize);
1441 	if (error != 0)
1442 		return (error);
1443 
1444 #ifdef EKCD
1445 	error = dump_write_key(di,
1446 	    di->mediaoffset + di->mediasize - di->blocksize - extent - keysize);
1447 	if (error != 0)
1448 		return (error);
1449 #endif
1450 
1451 	error = dump_write_header(di, kdh,
1452 	    di->mediaoffset + di->mediasize - di->blocksize);
1453 	if (error != 0)
1454 		return (error);
1455 
1456 	(void)dump_write(di, NULL, 0, 0, 0);
1457 	return (0);
1458 }
1459 
1460 void
1461 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1462     char *magic, uint32_t archver, uint64_t dumplen)
1463 {
1464 	size_t dstsize;
1465 
1466 	bzero(kdh, sizeof(*kdh));
1467 	strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1468 	strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1469 	kdh->version = htod32(KERNELDUMPVERSION);
1470 	kdh->architectureversion = htod32(archver);
1471 	kdh->dumplength = htod64(dumplen);
1472 	kdh->dumpextent = kdh->dumplength;
1473 	kdh->dumptime = htod64(time_second);
1474 #ifdef EKCD
1475 	kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdc));
1476 #else
1477 	kdh->dumpkeysize = 0;
1478 #endif
1479 	kdh->blocksize = htod32(di->blocksize);
1480 	strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1481 	dstsize = sizeof(kdh->versionstring);
1482 	if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1483 		kdh->versionstring[dstsize - 2] = '\n';
1484 	if (panicstr != NULL)
1485 		strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1486 #ifdef GZIO
1487 	if (di->kdgz != NULL)
1488 		kdh->compression = KERNELDUMP_COMP_GZIP;
1489 #endif
1490 	kdh->parity = kerneldump_parity(kdh);
1491 }
1492 
1493 #ifdef DDB
1494 DB_SHOW_COMMAND(panic, db_show_panic)
1495 {
1496 
1497 	if (panicstr == NULL)
1498 		db_printf("panicstr not set\n");
1499 	else
1500 		db_printf("panic: %s\n", panicstr);
1501 }
1502 #endif
1503