xref: /freebsd/sys/kern/kern_shutdown.c (revision ca987d4641cdcd7f27e153db17c5bf064934faf5)
1 /*-
2  * Copyright (c) 1986, 1988, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_shutdown.c	8.3 (Berkeley) 1/21/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_ddb.h"
41 #include "opt_ekcd.h"
42 #include "opt_gzio.h"
43 #include "opt_kdb.h"
44 #include "opt_panic.h"
45 #include "opt_sched.h"
46 #include "opt_watchdog.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/conf.h>
53 #include <sys/cons.h>
54 #include <sys/eventhandler.h>
55 #include <sys/filedesc.h>
56 #include <sys/gzio.h>
57 #include <sys/jail.h>
58 #include <sys/kdb.h>
59 #include <sys/kernel.h>
60 #include <sys/kerneldump.h>
61 #include <sys/kthread.h>
62 #include <sys/ktr.h>
63 #include <sys/malloc.h>
64 #include <sys/mount.h>
65 #include <sys/priv.h>
66 #include <sys/proc.h>
67 #include <sys/reboot.h>
68 #include <sys/resourcevar.h>
69 #include <sys/rwlock.h>
70 #include <sys/sched.h>
71 #include <sys/smp.h>
72 #include <sys/sysctl.h>
73 #include <sys/sysproto.h>
74 #include <sys/vnode.h>
75 #include <sys/watchdog.h>
76 
77 #include <crypto/rijndael/rijndael-api-fst.h>
78 #include <crypto/sha2/sha256.h>
79 
80 #include <ddb/ddb.h>
81 
82 #include <machine/cpu.h>
83 #include <machine/dump.h>
84 #include <machine/pcb.h>
85 #include <machine/smp.h>
86 
87 #include <security/mac/mac_framework.h>
88 
89 #include <vm/vm.h>
90 #include <vm/vm_object.h>
91 #include <vm/vm_page.h>
92 #include <vm/vm_pager.h>
93 #include <vm/swap_pager.h>
94 
95 #include <sys/signalvar.h>
96 
97 static MALLOC_DEFINE(M_DUMPER, "dumper", "dumper block buffer");
98 
99 #ifndef PANIC_REBOOT_WAIT_TIME
100 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
101 #endif
102 static int panic_reboot_wait_time = PANIC_REBOOT_WAIT_TIME;
103 SYSCTL_INT(_kern, OID_AUTO, panic_reboot_wait_time, CTLFLAG_RWTUN,
104     &panic_reboot_wait_time, 0,
105     "Seconds to wait before rebooting after a panic");
106 
107 /*
108  * Note that stdarg.h and the ANSI style va_start macro is used for both
109  * ANSI and traditional C compilers.
110  */
111 #include <machine/stdarg.h>
112 
113 #ifdef KDB
114 #ifdef KDB_UNATTENDED
115 int debugger_on_panic = 0;
116 #else
117 int debugger_on_panic = 1;
118 #endif
119 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic,
120     CTLFLAG_RWTUN | CTLFLAG_SECURE,
121     &debugger_on_panic, 0, "Run debugger on kernel panic");
122 
123 #ifdef KDB_TRACE
124 static int trace_on_panic = 1;
125 #else
126 static int trace_on_panic = 0;
127 #endif
128 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
129     CTLFLAG_RWTUN | CTLFLAG_SECURE,
130     &trace_on_panic, 0, "Print stack trace on kernel panic");
131 #endif /* KDB */
132 
133 static int sync_on_panic = 0;
134 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN,
135 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
136 
137 static bool poweroff_on_panic = 0;
138 SYSCTL_BOOL(_kern, OID_AUTO, poweroff_on_panic, CTLFLAG_RWTUN,
139 	&poweroff_on_panic, 0, "Do a power off instead of a reboot on a panic");
140 
141 static bool powercycle_on_panic = 0;
142 SYSCTL_BOOL(_kern, OID_AUTO, powercycle_on_panic, CTLFLAG_RWTUN,
143 	&powercycle_on_panic, 0, "Do a power cycle instead of a reboot on a panic");
144 
145 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0,
146     "Shutdown environment");
147 
148 #ifndef DIAGNOSTIC
149 static int show_busybufs;
150 #else
151 static int show_busybufs = 1;
152 #endif
153 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
154 	&show_busybufs, 0, "");
155 
156 int suspend_blocked = 0;
157 SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW,
158 	&suspend_blocked, 0, "Block suspend due to a pending shutdown");
159 
160 #ifdef EKCD
161 FEATURE(ekcd, "Encrypted kernel crash dumps support");
162 
163 MALLOC_DEFINE(M_EKCD, "ekcd", "Encrypted kernel crash dumps data");
164 
165 struct kerneldumpcrypto {
166 	uint8_t			kdc_encryption;
167 	uint8_t			kdc_iv[KERNELDUMP_IV_MAX_SIZE];
168 	keyInstance		kdc_ki;
169 	cipherInstance		kdc_ci;
170 	uint32_t		kdc_dumpkeysize;
171 	struct kerneldumpkey	kdc_dumpkey[];
172 };
173 #endif
174 
175 #ifdef GZIO
176 struct kerneldumpgz {
177 	struct gzio_stream	*kdgz_stream;
178 	uint8_t			*kdgz_buf;
179 	size_t			kdgz_resid;
180 };
181 
182 static struct kerneldumpgz *kerneldumpgz_create(struct dumperinfo *di,
183 		    uint8_t compression);
184 static void	kerneldumpgz_destroy(struct dumperinfo *di);
185 static int	kerneldumpgz_write_cb(void *cb, size_t len, off_t off, void *arg);
186 
187 static int kerneldump_gzlevel = 6;
188 SYSCTL_INT(_kern, OID_AUTO, kerneldump_gzlevel, CTLFLAG_RWTUN,
189     &kerneldump_gzlevel, 0,
190     "Kernel crash dump gzip compression level");
191 #endif /* GZIO */
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 CPU 0 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, 0);
381 		thread_unlock(curthread);
382 		KASSERT(PCPU_GET(cpuid) == 0, ("boot: not running on cpu 0"));
383 	}
384 #endif
385 	/* We're in the process of rebooting. */
386 	rebooting = 1;
387 
388 	/* We are out of the debugger now. */
389 	kdb_active = 0;
390 
391 	/*
392 	 * Do any callouts that should be done BEFORE syncing the filesystems.
393 	 */
394 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
395 
396 	/*
397 	 * Now sync filesystems
398 	 */
399 	if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
400 		once = 1;
401 		bufshutdown(show_busybufs);
402 	}
403 
404 	print_uptime();
405 
406 	cngrab();
407 
408 	/*
409 	 * Ok, now do things that assume all filesystem activity has
410 	 * been completed.
411 	 */
412 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
413 
414 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
415 		doadump(TRUE);
416 
417 	/* Now that we're going to really halt the system... */
418 	EVENTHANDLER_INVOKE(shutdown_final, howto);
419 
420 	for(;;) ;	/* safety against shutdown_reset not working */
421 	/* NOTREACHED */
422 }
423 
424 /*
425  * The system call that results in changing the rootfs.
426  */
427 static int
428 kern_reroot(void)
429 {
430 	struct vnode *oldrootvnode, *vp;
431 	struct mount *mp, *devmp;
432 	int error;
433 
434 	if (curproc != initproc)
435 		return (EPERM);
436 
437 	/*
438 	 * Mark the filesystem containing currently-running executable
439 	 * (the temporary copy of init(8)) busy.
440 	 */
441 	vp = curproc->p_textvp;
442 	error = vn_lock(vp, LK_SHARED);
443 	if (error != 0)
444 		return (error);
445 	mp = vp->v_mount;
446 	error = vfs_busy(mp, MBF_NOWAIT);
447 	if (error != 0) {
448 		vfs_ref(mp);
449 		VOP_UNLOCK(vp, 0);
450 		error = vfs_busy(mp, 0);
451 		vn_lock(vp, LK_SHARED | LK_RETRY);
452 		vfs_rel(mp);
453 		if (error != 0) {
454 			VOP_UNLOCK(vp, 0);
455 			return (ENOENT);
456 		}
457 		if (vp->v_iflag & VI_DOOMED) {
458 			VOP_UNLOCK(vp, 0);
459 			vfs_unbusy(mp);
460 			return (ENOENT);
461 		}
462 	}
463 	VOP_UNLOCK(vp, 0);
464 
465 	/*
466 	 * Remove the filesystem containing currently-running executable
467 	 * from the mount list, to prevent it from being unmounted
468 	 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
469 	 *
470 	 * Also preserve /dev - forcibly unmounting it could cause driver
471 	 * reinitialization.
472 	 */
473 
474 	vfs_ref(rootdevmp);
475 	devmp = rootdevmp;
476 	rootdevmp = NULL;
477 
478 	mtx_lock(&mountlist_mtx);
479 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
480 	TAILQ_REMOVE(&mountlist, devmp, mnt_list);
481 	mtx_unlock(&mountlist_mtx);
482 
483 	oldrootvnode = rootvnode;
484 
485 	/*
486 	 * Unmount everything except for the two filesystems preserved above.
487 	 */
488 	vfs_unmountall();
489 
490 	/*
491 	 * Add /dev back; vfs_mountroot() will move it into its new place.
492 	 */
493 	mtx_lock(&mountlist_mtx);
494 	TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
495 	mtx_unlock(&mountlist_mtx);
496 	rootdevmp = devmp;
497 	vfs_rel(rootdevmp);
498 
499 	/*
500 	 * Mount the new rootfs.
501 	 */
502 	vfs_mountroot();
503 
504 	/*
505 	 * Update all references to the old rootvnode.
506 	 */
507 	mountcheckdirs(oldrootvnode, rootvnode);
508 
509 	/*
510 	 * Add the temporary filesystem back and unbusy it.
511 	 */
512 	mtx_lock(&mountlist_mtx);
513 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
514 	mtx_unlock(&mountlist_mtx);
515 	vfs_unbusy(mp);
516 
517 	return (0);
518 }
519 
520 /*
521  * If the shutdown was a clean halt, behave accordingly.
522  */
523 static void
524 shutdown_halt(void *junk, int howto)
525 {
526 
527 	if (howto & RB_HALT) {
528 		printf("\n");
529 		printf("The operating system has halted.\n");
530 		printf("Please press any key to reboot.\n\n");
531 		switch (cngetc()) {
532 		case -1:		/* No console, just die */
533 			cpu_halt();
534 			/* NOTREACHED */
535 		default:
536 			howto &= ~RB_HALT;
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 #ifdef GZIO
988 static struct kerneldumpgz *
989 kerneldumpgz_create(struct dumperinfo *di, uint8_t compression)
990 {
991 	struct kerneldumpgz *kdgz;
992 
993 	if (compression != KERNELDUMP_COMP_GZIP)
994 		return (NULL);
995 	kdgz = malloc(sizeof(*kdgz), M_DUMPER, M_WAITOK | M_ZERO);
996 	kdgz->kdgz_stream = gzio_init(kerneldumpgz_write_cb, GZIO_DEFLATE,
997 	    di->maxiosize, kerneldump_gzlevel, di);
998 	if (kdgz->kdgz_stream == NULL) {
999 		free(kdgz, M_DUMPER);
1000 		return (NULL);
1001 	}
1002 	kdgz->kdgz_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1003 	return (kdgz);
1004 }
1005 
1006 static void
1007 kerneldumpgz_destroy(struct dumperinfo *di)
1008 {
1009 	struct kerneldumpgz *kdgz;
1010 
1011 	kdgz = di->kdgz;
1012 	if (kdgz == NULL)
1013 		return;
1014 	gzio_fini(kdgz->kdgz_stream);
1015 	explicit_bzero(kdgz->kdgz_buf, di->maxiosize);
1016 	free(kdgz->kdgz_buf, M_DUMPER);
1017 	free(kdgz, M_DUMPER);
1018 }
1019 #endif /* GZIO */
1020 
1021 /* Registration of dumpers */
1022 int
1023 set_dumper(struct dumperinfo *di, const char *devname, struct thread *td,
1024     uint8_t compression, uint8_t encryption, const uint8_t *key,
1025     uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1026 {
1027 	size_t wantcopy;
1028 	int error;
1029 
1030 	error = priv_check(td, PRIV_SETDUMPER);
1031 	if (error != 0)
1032 		return (error);
1033 
1034 	if (di == NULL) {
1035 		error = 0;
1036 		goto cleanup;
1037 	}
1038 	if (dumper.dumper != NULL)
1039 		return (EBUSY);
1040 	dumper = *di;
1041 	dumper.blockbuf = NULL;
1042 	dumper.kdc = NULL;
1043 	dumper.kdgz = NULL;
1044 
1045 	if (encryption != KERNELDUMP_ENC_NONE) {
1046 #ifdef EKCD
1047 		dumper.kdc = kerneldumpcrypto_create(di->blocksize, encryption,
1048 		    key, encryptedkeysize, encryptedkey);
1049 		if (dumper.kdc == NULL) {
1050 			error = EINVAL;
1051 			goto cleanup;
1052 		}
1053 #else
1054 		error = EOPNOTSUPP;
1055 		goto cleanup;
1056 #endif
1057 	}
1058 
1059 	wantcopy = strlcpy(dumpdevname, devname, sizeof(dumpdevname));
1060 	if (wantcopy >= sizeof(dumpdevname)) {
1061 		printf("set_dumper: device name truncated from '%s' -> '%s'\n",
1062 		    devname, dumpdevname);
1063 	}
1064 
1065 	if (compression != KERNELDUMP_COMP_NONE) {
1066 #ifdef GZIO
1067 		/*
1068 		 * We currently can't support simultaneous encryption and
1069 		 * compression.
1070 		 */
1071 		if (encryption != KERNELDUMP_ENC_NONE) {
1072 			error = EOPNOTSUPP;
1073 			goto cleanup;
1074 		}
1075 		dumper.kdgz = kerneldumpgz_create(&dumper, compression);
1076 		if (dumper.kdgz == NULL) {
1077 			error = EINVAL;
1078 			goto cleanup;
1079 		}
1080 #else
1081 		error = EOPNOTSUPP;
1082 		goto cleanup;
1083 #endif
1084 	}
1085 
1086 	dumper.blockbuf = malloc(di->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1087 	return (0);
1088 cleanup:
1089 #ifdef EKCD
1090 	if (dumper.kdc != NULL) {
1091 		explicit_bzero(dumper.kdc, sizeof(*dumper.kdc) +
1092 		    dumper.kdc->kdc_dumpkeysize);
1093 		free(dumper.kdc, M_EKCD);
1094 	}
1095 #endif
1096 
1097 #ifdef GZIO
1098 	kerneldumpgz_destroy(&dumper);
1099 #endif
1100 
1101 	if (dumper.blockbuf != NULL) {
1102 		explicit_bzero(dumper.blockbuf, dumper.blocksize);
1103 		free(dumper.blockbuf, M_DUMPER);
1104 	}
1105 	explicit_bzero(&dumper, sizeof(dumper));
1106 	dumpdevname[0] = '\0';
1107 	return (error);
1108 }
1109 
1110 static int
1111 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1112 {
1113 
1114 	if (length != 0 && (offset < di->mediaoffset ||
1115 	    offset - di->mediaoffset + length > di->mediasize)) {
1116 		printf("Attempt to write outside dump device boundaries.\n"
1117 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1118 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
1119 		    (uintmax_t)length, (intmax_t)di->mediasize);
1120 		return (ENOSPC);
1121 	}
1122 	if (length % di->blocksize != 0) {
1123 		printf("Attempt to write partial block of length %ju.\n",
1124 		    (uintmax_t)length);
1125 		return (EINVAL);
1126 	}
1127 	if (offset % di->blocksize != 0) {
1128 		printf("Attempt to write at unaligned offset %jd.\n",
1129 		    (intmax_t)offset);
1130 		return (EINVAL);
1131 	}
1132 
1133 	return (0);
1134 }
1135 
1136 #ifdef EKCD
1137 static int
1138 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1139 {
1140 
1141 	switch (kdc->kdc_encryption) {
1142 	case KERNELDUMP_ENC_AES_256_CBC:
1143 		if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1144 		    8 * size, buf) <= 0) {
1145 			return (EIO);
1146 		}
1147 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1148 		    buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1149 			return (EIO);
1150 		}
1151 		break;
1152 	default:
1153 		return (EINVAL);
1154 	}
1155 
1156 	return (0);
1157 }
1158 
1159 /* Encrypt data and call dumper. */
1160 static int
1161 dump_encrypted_write(struct dumperinfo *di, void *virtual,
1162     vm_offset_t physical, off_t offset, size_t length)
1163 {
1164 	static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1165 	struct kerneldumpcrypto *kdc;
1166 	int error;
1167 	size_t nbytes;
1168 
1169 	kdc = di->kdc;
1170 
1171 	while (length > 0) {
1172 		nbytes = MIN(length, sizeof(buf));
1173 		bcopy(virtual, buf, nbytes);
1174 
1175 		if (dump_encrypt(kdc, buf, nbytes) != 0)
1176 			return (EIO);
1177 
1178 		error = dump_write(di, buf, physical, offset, nbytes);
1179 		if (error != 0)
1180 			return (error);
1181 
1182 		offset += nbytes;
1183 		virtual = (void *)((uint8_t *)virtual + nbytes);
1184 		length -= nbytes;
1185 	}
1186 
1187 	return (0);
1188 }
1189 
1190 static int
1191 dump_write_key(struct dumperinfo *di, off_t offset)
1192 {
1193 	struct kerneldumpcrypto *kdc;
1194 
1195 	kdc = di->kdc;
1196 	if (kdc == NULL)
1197 		return (0);
1198 	return (dump_write(di, kdc->kdc_dumpkey, 0, offset,
1199 	    kdc->kdc_dumpkeysize));
1200 }
1201 #endif /* EKCD */
1202 
1203 #ifdef GZIO
1204 static int
1205 kerneldumpgz_write_cb(void *base, size_t length, off_t offset, void *arg)
1206 {
1207 	struct dumperinfo *di;
1208 	size_t resid, rlength;
1209 	int error;
1210 
1211 	di = arg;
1212 
1213 	if (length % di->blocksize != 0) {
1214 		/*
1215 		 * This must be the final write after flushing the compression
1216 		 * stream. Write as many full blocks as possible and stash the
1217 		 * residual data in the dumper's block buffer. It will be
1218 		 * padded and written in dump_finish().
1219 		 */
1220 		rlength = rounddown(length, di->blocksize);
1221 		if (rlength != 0) {
1222 			error = _dump_append(di, base, 0, rlength);
1223 			if (error != 0)
1224 				return (error);
1225 		}
1226 		resid = length - rlength;
1227 		memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1228 		di->kdgz->kdgz_resid = resid;
1229 		return (EAGAIN);
1230 	}
1231 	return (_dump_append(di, base, 0, length));
1232 }
1233 #endif /* GZIO */
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->kdc);
1292 	if (error != 0)
1293 		return (error);
1294 	keysize = kerneldumpcrypto_dumpkeysize(di->kdc);
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 #ifdef GZIO
1303 		if (di->kdgz != NULL) {
1304 			/*
1305 			 * We don't yet know how much space the compressed dump
1306 			 * will occupy, so try to use the whole swap partition
1307 			 * (minus the first 64KB) in the hope that the
1308 			 * compressed dump will fit. If that doesn't turn out to
1309 			 * be enouch, the bounds checking in dump_write()
1310 			 * will catch us and cause the dump to fail.
1311 			 */
1312 			dumpextent = di->mediasize - SIZEOF_METADATA -
1313 			    2 * di->blocksize - keysize;
1314 			kdh->dumpextent = htod64(dumpextent);
1315 		} else
1316 #endif
1317 			return (E2BIG);
1318 	}
1319 
1320 	/* The offset at which to begin writing the dump. */
1321 	di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1322 	    dumpextent;
1323 
1324 	return (0);
1325 }
1326 
1327 static int
1328 _dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1329     size_t length)
1330 {
1331 	int error;
1332 
1333 #ifdef EKCD
1334 	if (di->kdc != NULL)
1335 		error = dump_encrypted_write(di, virtual, physical, di->dumpoff,
1336 		    length);
1337 	else
1338 #endif
1339 		error = dump_write(di, virtual, physical, di->dumpoff, length);
1340 	if (error == 0)
1341 		di->dumpoff += length;
1342 	return (error);
1343 }
1344 
1345 /*
1346  * Write to the dump device starting at dumpoff. When compression is enabled,
1347  * writes to the device will be performed using a callback that gets invoked
1348  * when the compression stream's output buffer is full.
1349  */
1350 int
1351 dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1352     size_t length)
1353 {
1354 #ifdef GZIO
1355 	void *buf;
1356 
1357 	if (di->kdgz != NULL) {
1358 		/* Bounce through a buffer to avoid gzip CRC errors. */
1359 		if (length > di->maxiosize)
1360 			return (EINVAL);
1361 		buf = di->kdgz->kdgz_buf;
1362 		memmove(buf, virtual, length);
1363 		return (gzio_write(di->kdgz->kdgz_stream, buf, length));
1364 	}
1365 #endif
1366 	return (_dump_append(di, virtual, physical, length));
1367 }
1368 
1369 /*
1370  * Write to the dump device at the specified offset.
1371  */
1372 int
1373 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1374     off_t offset, size_t length)
1375 {
1376 	int error;
1377 
1378 	error = dump_check_bounds(di, offset, length);
1379 	if (error != 0)
1380 		return (error);
1381 	return (di->dumper(di->priv, virtual, physical, offset, length));
1382 }
1383 
1384 /*
1385  * Perform kernel dump finalization: flush the compression stream, if necessary,
1386  * write the leading and trailing kernel dump headers now that we know the true
1387  * length of the dump, and optionally write the encryption key following the
1388  * leading header.
1389  */
1390 int
1391 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1392 {
1393 	uint64_t extent;
1394 	uint32_t keysize;
1395 	int error;
1396 
1397 	extent = dtoh64(kdh->dumpextent);
1398 
1399 #ifdef EKCD
1400 	keysize = kerneldumpcrypto_dumpkeysize(di->kdc);
1401 #else
1402 	keysize = 0;
1403 #endif
1404 
1405 #ifdef GZIO
1406 	if (di->kdgz != NULL) {
1407 		error = gzio_flush(di->kdgz->kdgz_stream);
1408 		if (error == EAGAIN) {
1409 			/* We have residual data in di->blockbuf. */
1410 			error = dump_write(di, di->blockbuf, 0, di->dumpoff,
1411 			    di->blocksize);
1412 			di->dumpoff += di->kdgz->kdgz_resid;
1413 			di->kdgz->kdgz_resid = 0;
1414 		}
1415 		if (error != 0)
1416 			return (error);
1417 
1418 		/*
1419 		 * We now know the size of the compressed dump, so update the
1420 		 * header accordingly and recompute parity.
1421 		 */
1422 		kdh->dumplength = htod64(di->dumpoff -
1423 		    (di->mediaoffset + di->mediasize - di->blocksize - extent));
1424 		kdh->parity = 0;
1425 		kdh->parity = kerneldump_parity(kdh);
1426 
1427 		gzio_reset(di->kdgz->kdgz_stream);
1428 	}
1429 #endif
1430 
1431 	/*
1432 	 * Write kerneldump headers at the beginning and end of the dump extent.
1433 	 * Write the key after the leading header.
1434 	 */
1435 	error = dump_write_header(di, kdh,
1436 	    di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1437 	    keysize);
1438 	if (error != 0)
1439 		return (error);
1440 
1441 #ifdef EKCD
1442 	error = dump_write_key(di,
1443 	    di->mediaoffset + di->mediasize - di->blocksize - extent - keysize);
1444 	if (error != 0)
1445 		return (error);
1446 #endif
1447 
1448 	error = dump_write_header(di, kdh,
1449 	    di->mediaoffset + di->mediasize - di->blocksize);
1450 	if (error != 0)
1451 		return (error);
1452 
1453 	(void)dump_write(di, NULL, 0, 0, 0);
1454 	return (0);
1455 }
1456 
1457 void
1458 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1459     char *magic, uint32_t archver, uint64_t dumplen)
1460 {
1461 	size_t dstsize;
1462 
1463 	bzero(kdh, sizeof(*kdh));
1464 	strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1465 	strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1466 	kdh->version = htod32(KERNELDUMPVERSION);
1467 	kdh->architectureversion = htod32(archver);
1468 	kdh->dumplength = htod64(dumplen);
1469 	kdh->dumpextent = kdh->dumplength;
1470 	kdh->dumptime = htod64(time_second);
1471 #ifdef EKCD
1472 	kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdc));
1473 #else
1474 	kdh->dumpkeysize = 0;
1475 #endif
1476 	kdh->blocksize = htod32(di->blocksize);
1477 	strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1478 	dstsize = sizeof(kdh->versionstring);
1479 	if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1480 		kdh->versionstring[dstsize - 2] = '\n';
1481 	if (panicstr != NULL)
1482 		strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1483 #ifdef GZIO
1484 	if (di->kdgz != NULL)
1485 		kdh->compression = KERNELDUMP_COMP_GZIP;
1486 #endif
1487 	kdh->parity = kerneldump_parity(kdh);
1488 }
1489 
1490 #ifdef DDB
1491 DB_SHOW_COMMAND(panic, db_show_panic)
1492 {
1493 
1494 	if (panicstr == NULL)
1495 		db_printf("panicstr not set\n");
1496 	else
1497 		db_printf("panic: %s\n", panicstr);
1498 }
1499 #endif
1500