xref: /freebsd/sys/kern/kern_shutdown.c (revision 190cef3d52236565eb22e18b33e9e865ec634aa3)
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/mbuf.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/taskqueue.h>
77 #include <sys/vnode.h>
78 #include <sys/watchdog.h>
79 
80 #include <crypto/rijndael/rijndael-api-fst.h>
81 #include <crypto/sha2/sha256.h>
82 
83 #include <ddb/ddb.h>
84 
85 #include <machine/cpu.h>
86 #include <machine/dump.h>
87 #include <machine/pcb.h>
88 #include <machine/smp.h>
89 
90 #include <security/mac/mac_framework.h>
91 
92 #include <vm/vm.h>
93 #include <vm/vm_object.h>
94 #include <vm/vm_page.h>
95 #include <vm/vm_pager.h>
96 #include <vm/swap_pager.h>
97 
98 #include <sys/signalvar.h>
99 
100 static MALLOC_DEFINE(M_DUMPER, "dumper", "dumper block buffer");
101 
102 #ifndef PANIC_REBOOT_WAIT_TIME
103 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
104 #endif
105 static int panic_reboot_wait_time = PANIC_REBOOT_WAIT_TIME;
106 SYSCTL_INT(_kern, OID_AUTO, panic_reboot_wait_time, CTLFLAG_RWTUN,
107     &panic_reboot_wait_time, 0,
108     "Seconds to wait before rebooting after a panic");
109 
110 /*
111  * Note that stdarg.h and the ANSI style va_start macro is used for both
112  * ANSI and traditional C compilers.
113  */
114 #include <machine/stdarg.h>
115 
116 #ifdef KDB
117 #ifdef KDB_UNATTENDED
118 int debugger_on_panic = 0;
119 #else
120 int debugger_on_panic = 1;
121 #endif
122 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic,
123     CTLFLAG_RWTUN | CTLFLAG_SECURE,
124     &debugger_on_panic, 0, "Run debugger on kernel panic");
125 
126 #ifdef KDB_TRACE
127 static int trace_on_panic = 1;
128 static bool trace_all_panics = true;
129 #else
130 static int trace_on_panic = 0;
131 static bool trace_all_panics = false;
132 #endif
133 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
134     CTLFLAG_RWTUN | CTLFLAG_SECURE,
135     &trace_on_panic, 0, "Print stack trace on kernel panic");
136 SYSCTL_BOOL(_debug, OID_AUTO, trace_all_panics, CTLFLAG_RWTUN,
137     &trace_all_panics, 0, "Print stack traces on secondary kernel panics");
138 #endif /* KDB */
139 
140 static int sync_on_panic = 0;
141 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN,
142 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
143 
144 static bool poweroff_on_panic = 0;
145 SYSCTL_BOOL(_kern, OID_AUTO, poweroff_on_panic, CTLFLAG_RWTUN,
146 	&poweroff_on_panic, 0, "Do a power off instead of a reboot on a panic");
147 
148 static bool powercycle_on_panic = 0;
149 SYSCTL_BOOL(_kern, OID_AUTO, powercycle_on_panic, CTLFLAG_RWTUN,
150 	&powercycle_on_panic, 0, "Do a power cycle instead of a reboot on a panic");
151 
152 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0,
153     "Shutdown environment");
154 
155 #ifndef DIAGNOSTIC
156 static int show_busybufs;
157 #else
158 static int show_busybufs = 1;
159 #endif
160 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
161 	&show_busybufs, 0, "");
162 
163 int suspend_blocked = 0;
164 SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW,
165 	&suspend_blocked, 0, "Block suspend due to a pending shutdown");
166 
167 #ifdef EKCD
168 FEATURE(ekcd, "Encrypted kernel crash dumps support");
169 
170 MALLOC_DEFINE(M_EKCD, "ekcd", "Encrypted kernel crash dumps data");
171 
172 struct kerneldumpcrypto {
173 	uint8_t			kdc_encryption;
174 	uint8_t			kdc_iv[KERNELDUMP_IV_MAX_SIZE];
175 	keyInstance		kdc_ki;
176 	cipherInstance		kdc_ci;
177 	uint32_t		kdc_dumpkeysize;
178 	struct kerneldumpkey	kdc_dumpkey[];
179 };
180 #endif
181 
182 struct kerneldumpcomp {
183 	uint8_t			kdc_format;
184 	struct compressor	*kdc_stream;
185 	uint8_t			*kdc_buf;
186 	size_t			kdc_resid;
187 };
188 
189 static struct kerneldumpcomp *kerneldumpcomp_create(struct dumperinfo *di,
190 		    uint8_t compression);
191 static void	kerneldumpcomp_destroy(struct dumperinfo *di);
192 static int	kerneldumpcomp_write_cb(void *base, size_t len, off_t off, void *arg);
193 
194 static int kerneldump_gzlevel = 6;
195 SYSCTL_INT(_kern, OID_AUTO, kerneldump_gzlevel, CTLFLAG_RWTUN,
196     &kerneldump_gzlevel, 0,
197     "Kernel crash dump compression level");
198 
199 /*
200  * Variable panicstr contains argument to first call to panic; used as flag
201  * to indicate that the kernel has already called panic.
202  */
203 const char *panicstr;
204 
205 int dumping;				/* system is dumping */
206 int rebooting;				/* system is rebooting */
207 static struct dumperinfo dumper;	/* our selected dumper */
208 
209 /* Context information for dump-debuggers. */
210 static struct pcb dumppcb;		/* Registers. */
211 lwpid_t dumptid;			/* Thread ID. */
212 
213 static struct cdevsw reroot_cdevsw = {
214      .d_version = D_VERSION,
215      .d_name    = "reroot",
216 };
217 
218 static void poweroff_wait(void *, int);
219 static void shutdown_halt(void *junk, int howto);
220 static void shutdown_panic(void *junk, int howto);
221 static void shutdown_reset(void *junk, int howto);
222 static int kern_reroot(void);
223 
224 /* register various local shutdown events */
225 static void
226 shutdown_conf(void *unused)
227 {
228 
229 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
230 	    SHUTDOWN_PRI_FIRST);
231 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
232 	    SHUTDOWN_PRI_LAST + 100);
233 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
234 	    SHUTDOWN_PRI_LAST + 100);
235 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
236 	    SHUTDOWN_PRI_LAST + 200);
237 }
238 
239 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
240 
241 /*
242  * The only reason this exists is to create the /dev/reroot/ directory,
243  * used by reroot code in init(8) as a mountpoint for tmpfs.
244  */
245 static void
246 reroot_conf(void *unused)
247 {
248 	int error;
249 	struct cdev *cdev;
250 
251 	error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev,
252 	    &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot");
253 	if (error != 0) {
254 		printf("%s: failed to create device node, error %d",
255 		    __func__, error);
256 	}
257 }
258 
259 SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL);
260 
261 /*
262  * The system call that results in a reboot.
263  */
264 /* ARGSUSED */
265 int
266 sys_reboot(struct thread *td, struct reboot_args *uap)
267 {
268 	int error;
269 
270 	error = 0;
271 #ifdef MAC
272 	error = mac_system_check_reboot(td->td_ucred, uap->opt);
273 #endif
274 	if (error == 0)
275 		error = priv_check(td, PRIV_REBOOT);
276 	if (error == 0) {
277 		if (uap->opt & RB_REROOT)
278 			error = kern_reroot();
279 		else
280 			kern_reboot(uap->opt);
281 	}
282 	return (error);
283 }
284 
285 static void
286 shutdown_nice_task_fn(void *arg, int pending __unused)
287 {
288 	int howto;
289 
290 	howto = (uintptr_t)arg;
291 	/* Send a signal to init(8) and have it shutdown the world. */
292 	PROC_LOCK(initproc);
293 	if (howto & RB_POWEROFF)
294 		kern_psignal(initproc, SIGUSR2);
295 	else if (howto & RB_POWERCYCLE)
296 		kern_psignal(initproc, SIGWINCH);
297 	else if (howto & RB_HALT)
298 		kern_psignal(initproc, SIGUSR1);
299 	else
300 		kern_psignal(initproc, SIGINT);
301 	PROC_UNLOCK(initproc);
302 }
303 
304 static struct task shutdown_nice_task = TASK_INITIALIZER(0,
305     &shutdown_nice_task_fn, NULL);
306 
307 /*
308  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
309  */
310 void
311 shutdown_nice(int howto)
312 {
313 
314 	if (initproc != NULL && !SCHEDULER_STOPPED()) {
315 		shutdown_nice_task.ta_context = (void *)(uintptr_t)howto;
316 		taskqueue_enqueue(taskqueue_fast, &shutdown_nice_task);
317 	} else {
318 		/*
319 		 * No init(8) running, or scheduler would not allow it
320 		 * to run, so simply reboot.
321 		 */
322 		kern_reboot(howto | RB_NOSYNC);
323 	}
324 }
325 
326 static void
327 print_uptime(void)
328 {
329 	int f;
330 	struct timespec ts;
331 
332 	getnanouptime(&ts);
333 	printf("Uptime: ");
334 	f = 0;
335 	if (ts.tv_sec >= 86400) {
336 		printf("%ldd", (long)ts.tv_sec / 86400);
337 		ts.tv_sec %= 86400;
338 		f = 1;
339 	}
340 	if (f || ts.tv_sec >= 3600) {
341 		printf("%ldh", (long)ts.tv_sec / 3600);
342 		ts.tv_sec %= 3600;
343 		f = 1;
344 	}
345 	if (f || ts.tv_sec >= 60) {
346 		printf("%ldm", (long)ts.tv_sec / 60);
347 		ts.tv_sec %= 60;
348 		f = 1;
349 	}
350 	printf("%lds\n", (long)ts.tv_sec);
351 }
352 
353 int
354 doadump(boolean_t textdump)
355 {
356 	boolean_t coredump;
357 	int error;
358 
359 	error = 0;
360 	if (dumping)
361 		return (EBUSY);
362 	if (dumper.dumper == NULL)
363 		return (ENXIO);
364 
365 	savectx(&dumppcb);
366 	dumptid = curthread->td_tid;
367 	dumping++;
368 
369 	coredump = TRUE;
370 #ifdef DDB
371 	if (textdump && textdump_pending) {
372 		coredump = FALSE;
373 		textdump_dumpsys(&dumper);
374 	}
375 #endif
376 	if (coredump)
377 		error = dumpsys(&dumper);
378 
379 	dumping--;
380 	return (error);
381 }
382 
383 /*
384  * Shutdown the system cleanly to prepare for reboot, halt, or power off.
385  */
386 void
387 kern_reboot(int howto)
388 {
389 	static int once = 0;
390 
391 	/*
392 	 * Normal paths here don't hold Giant, but we can wind up here
393 	 * unexpectedly with it held.  Drop it now so we don't have to
394 	 * drop and pick it up elsewhere. The paths it is locking will
395 	 * never be returned to, and it is preferable to preclude
396 	 * deadlock than to lock against code that won't ever
397 	 * continue.
398 	 */
399 	while (mtx_owned(&Giant))
400 		mtx_unlock(&Giant);
401 
402 #if defined(SMP)
403 	/*
404 	 * Bind us to the first CPU so that all shutdown code runs there.  Some
405 	 * systems don't shutdown properly (i.e., ACPI power off) if we
406 	 * run on another processor.
407 	 */
408 	if (!SCHEDULER_STOPPED()) {
409 		thread_lock(curthread);
410 		sched_bind(curthread, CPU_FIRST());
411 		thread_unlock(curthread);
412 		KASSERT(PCPU_GET(cpuid) == CPU_FIRST(),
413 		    ("boot: not running on cpu 0"));
414 	}
415 #endif
416 	/* We're in the process of rebooting. */
417 	rebooting = 1;
418 
419 	/* We are out of the debugger now. */
420 	kdb_active = 0;
421 
422 	/*
423 	 * Do any callouts that should be done BEFORE syncing the filesystems.
424 	 */
425 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
426 
427 	/*
428 	 * Now sync filesystems
429 	 */
430 	if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
431 		once = 1;
432 		bufshutdown(show_busybufs);
433 	}
434 
435 	print_uptime();
436 
437 	cngrab();
438 
439 	/*
440 	 * Ok, now do things that assume all filesystem activity has
441 	 * been completed.
442 	 */
443 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
444 
445 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
446 		doadump(TRUE);
447 
448 	/* Now that we're going to really halt the system... */
449 	EVENTHANDLER_INVOKE(shutdown_final, howto);
450 
451 	for(;;) ;	/* safety against shutdown_reset not working */
452 	/* NOTREACHED */
453 }
454 
455 /*
456  * The system call that results in changing the rootfs.
457  */
458 static int
459 kern_reroot(void)
460 {
461 	struct vnode *oldrootvnode, *vp;
462 	struct mount *mp, *devmp;
463 	int error;
464 
465 	if (curproc != initproc)
466 		return (EPERM);
467 
468 	/*
469 	 * Mark the filesystem containing currently-running executable
470 	 * (the temporary copy of init(8)) busy.
471 	 */
472 	vp = curproc->p_textvp;
473 	error = vn_lock(vp, LK_SHARED);
474 	if (error != 0)
475 		return (error);
476 	mp = vp->v_mount;
477 	error = vfs_busy(mp, MBF_NOWAIT);
478 	if (error != 0) {
479 		vfs_ref(mp);
480 		VOP_UNLOCK(vp, 0);
481 		error = vfs_busy(mp, 0);
482 		vn_lock(vp, LK_SHARED | LK_RETRY);
483 		vfs_rel(mp);
484 		if (error != 0) {
485 			VOP_UNLOCK(vp, 0);
486 			return (ENOENT);
487 		}
488 		if (vp->v_iflag & VI_DOOMED) {
489 			VOP_UNLOCK(vp, 0);
490 			vfs_unbusy(mp);
491 			return (ENOENT);
492 		}
493 	}
494 	VOP_UNLOCK(vp, 0);
495 
496 	/*
497 	 * Remove the filesystem containing currently-running executable
498 	 * from the mount list, to prevent it from being unmounted
499 	 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
500 	 *
501 	 * Also preserve /dev - forcibly unmounting it could cause driver
502 	 * reinitialization.
503 	 */
504 
505 	vfs_ref(rootdevmp);
506 	devmp = rootdevmp;
507 	rootdevmp = NULL;
508 
509 	mtx_lock(&mountlist_mtx);
510 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
511 	TAILQ_REMOVE(&mountlist, devmp, mnt_list);
512 	mtx_unlock(&mountlist_mtx);
513 
514 	oldrootvnode = rootvnode;
515 
516 	/*
517 	 * Unmount everything except for the two filesystems preserved above.
518 	 */
519 	vfs_unmountall();
520 
521 	/*
522 	 * Add /dev back; vfs_mountroot() will move it into its new place.
523 	 */
524 	mtx_lock(&mountlist_mtx);
525 	TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
526 	mtx_unlock(&mountlist_mtx);
527 	rootdevmp = devmp;
528 	vfs_rel(rootdevmp);
529 
530 	/*
531 	 * Mount the new rootfs.
532 	 */
533 	vfs_mountroot();
534 
535 	/*
536 	 * Update all references to the old rootvnode.
537 	 */
538 	mountcheckdirs(oldrootvnode, rootvnode);
539 
540 	/*
541 	 * Add the temporary filesystem back and unbusy it.
542 	 */
543 	mtx_lock(&mountlist_mtx);
544 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
545 	mtx_unlock(&mountlist_mtx);
546 	vfs_unbusy(mp);
547 
548 	return (0);
549 }
550 
551 /*
552  * If the shutdown was a clean halt, behave accordingly.
553  */
554 static void
555 shutdown_halt(void *junk, int howto)
556 {
557 
558 	if (howto & RB_HALT) {
559 		printf("\n");
560 		printf("The operating system has halted.\n");
561 		printf("Please press any key to reboot.\n\n");
562 		switch (cngetc()) {
563 		case -1:		/* No console, just die */
564 			cpu_halt();
565 			/* NOTREACHED */
566 		default:
567 			break;
568 		}
569 	}
570 }
571 
572 /*
573  * Check to see if the system paniced, pause and then reboot
574  * according to the specified delay.
575  */
576 static void
577 shutdown_panic(void *junk, int howto)
578 {
579 	int loop;
580 
581 	if (howto & RB_DUMP) {
582 		if (panic_reboot_wait_time != 0) {
583 			if (panic_reboot_wait_time != -1) {
584 				printf("Automatic reboot in %d seconds - "
585 				       "press a key on the console to abort\n",
586 					panic_reboot_wait_time);
587 				for (loop = panic_reboot_wait_time * 10;
588 				     loop > 0; --loop) {
589 					DELAY(1000 * 100); /* 1/10th second */
590 					/* Did user type a key? */
591 					if (cncheckc() != -1)
592 						break;
593 				}
594 				if (!loop)
595 					return;
596 			}
597 		} else { /* zero time specified - reboot NOW */
598 			return;
599 		}
600 		printf("--> Press a key on the console to reboot,\n");
601 		printf("--> or switch off the system now.\n");
602 		cngetc();
603 	}
604 }
605 
606 /*
607  * Everything done, now reset
608  */
609 static void
610 shutdown_reset(void *junk, int howto)
611 {
612 
613 	printf("Rebooting...\n");
614 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
615 
616 	/*
617 	 * Acquiring smp_ipi_mtx here has a double effect:
618 	 * - it disables interrupts avoiding CPU0 preemption
619 	 *   by fast handlers (thus deadlocking  against other CPUs)
620 	 * - it avoids deadlocks against smp_rendezvous() or, more
621 	 *   generally, threads busy-waiting, with this spinlock held,
622 	 *   and waiting for responses by threads on other CPUs
623 	 *   (ie. smp_tlb_shootdown()).
624 	 *
625 	 * For the !SMP case it just needs to handle the former problem.
626 	 */
627 #ifdef SMP
628 	mtx_lock_spin(&smp_ipi_mtx);
629 #else
630 	spinlock_enter();
631 #endif
632 
633 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
634 	cpu_reset();
635 	/* NOTREACHED */ /* assuming reset worked */
636 }
637 
638 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
639 static int kassert_warn_only = 0;
640 #ifdef KDB
641 static int kassert_do_kdb = 0;
642 #endif
643 #ifdef KTR
644 static int kassert_do_ktr = 0;
645 #endif
646 static int kassert_do_log = 1;
647 static int kassert_log_pps_limit = 4;
648 static int kassert_log_mute_at = 0;
649 static int kassert_log_panic_at = 0;
650 static int kassert_suppress_in_panic = 0;
651 static int kassert_warnings = 0;
652 
653 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW, NULL, "kassert options");
654 
655 #ifdef KASSERT_PANIC_OPTIONAL
656 #define KASSERT_RWTUN	CTLFLAG_RWTUN
657 #else
658 #define KASSERT_RWTUN	CTLFLAG_RDTUN
659 #endif
660 
661 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, KASSERT_RWTUN,
662     &kassert_warn_only, 0,
663     "KASSERT triggers a panic (0) or just a warning (1)");
664 
665 #ifdef KDB
666 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, KASSERT_RWTUN,
667     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
668 #endif
669 
670 #ifdef KTR
671 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, KASSERT_RWTUN,
672     &kassert_do_ktr, 0,
673     "KASSERT does a KTR, set this to the KTRMASK you want");
674 #endif
675 
676 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, KASSERT_RWTUN,
677     &kassert_do_log, 0,
678     "If warn_only is enabled, log (1) or do not log (0) assertion violations");
679 
680 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, KASSERT_RWTUN,
681     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
682 
683 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, KASSERT_RWTUN,
684     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
685 
686 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, KASSERT_RWTUN,
687     &kassert_log_pps_limit, 0, "limit number of log messages per second");
688 
689 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, KASSERT_RWTUN,
690     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
691 
692 SYSCTL_INT(_debug_kassert, OID_AUTO, suppress_in_panic, KASSERT_RWTUN,
693     &kassert_suppress_in_panic, 0,
694     "KASSERTs will be suppressed while handling a panic");
695 #undef KASSERT_RWTUN
696 
697 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
698 
699 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
700     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE, NULL, 0,
701     kassert_sysctl_kassert, "I", "set to trigger a test kassert");
702 
703 static int
704 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
705 {
706 	int error, i;
707 
708 	error = sysctl_wire_old_buffer(req, sizeof(int));
709 	if (error == 0) {
710 		i = 0;
711 		error = sysctl_handle_int(oidp, &i, 0, req);
712 	}
713 	if (error != 0 || req->newptr == NULL)
714 		return (error);
715 	KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
716 	return (0);
717 }
718 
719 #ifdef KASSERT_PANIC_OPTIONAL
720 /*
721  * Called by KASSERT, this decides if we will panic
722  * or if we will log via printf and/or ktr.
723  */
724 void
725 kassert_panic(const char *fmt, ...)
726 {
727 	static char buf[256];
728 	va_list ap;
729 
730 	va_start(ap, fmt);
731 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
732 	va_end(ap);
733 
734 	/*
735 	 * If we are suppressing secondary panics, log the warning but do not
736 	 * re-enter panic/kdb.
737 	 */
738 	if (panicstr != NULL && kassert_suppress_in_panic) {
739 		if (kassert_do_log) {
740 			printf("KASSERT failed: %s\n", buf);
741 #ifdef KDB
742 			if (trace_all_panics && trace_on_panic)
743 				kdb_backtrace();
744 #endif
745 		}
746 		return;
747 	}
748 
749 	/*
750 	 * panic if we're not just warning, or if we've exceeded
751 	 * kassert_log_panic_at warnings.
752 	 */
753 	if (!kassert_warn_only ||
754 	    (kassert_log_panic_at > 0 &&
755 	     kassert_warnings >= kassert_log_panic_at)) {
756 		va_start(ap, fmt);
757 		vpanic(fmt, ap);
758 		/* NORETURN */
759 	}
760 #ifdef KTR
761 	if (kassert_do_ktr)
762 		CTR0(ktr_mask, buf);
763 #endif /* KTR */
764 	/*
765 	 * log if we've not yet met the mute limit.
766 	 */
767 	if (kassert_do_log &&
768 	    (kassert_log_mute_at == 0 ||
769 	     kassert_warnings < kassert_log_mute_at)) {
770 		static  struct timeval lasterr;
771 		static  int curerr;
772 
773 		if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
774 			printf("KASSERT failed: %s\n", buf);
775 			kdb_backtrace();
776 		}
777 	}
778 #ifdef KDB
779 	if (kassert_do_kdb) {
780 		kdb_enter(KDB_WHY_KASSERT, buf);
781 	}
782 #endif
783 	atomic_add_int(&kassert_warnings, 1);
784 }
785 #endif /* KASSERT_PANIC_OPTIONAL */
786 #endif
787 
788 /*
789  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
790  * and then reboots.  If we are called twice, then we avoid trying to sync
791  * the disks as this often leads to recursive panics.
792  */
793 void
794 panic(const char *fmt, ...)
795 {
796 	va_list ap;
797 
798 	va_start(ap, fmt);
799 	vpanic(fmt, ap);
800 }
801 
802 void
803 vpanic(const char *fmt, va_list ap)
804 {
805 #ifdef SMP
806 	cpuset_t other_cpus;
807 #endif
808 	struct thread *td = curthread;
809 	int bootopt, newpanic;
810 	static char buf[256];
811 
812 	spinlock_enter();
813 
814 #ifdef SMP
815 	/*
816 	 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
817 	 * concurrently entering panic.  Only the winner will proceed
818 	 * further.
819 	 */
820 	if (panicstr == NULL && !kdb_active) {
821 		other_cpus = all_cpus;
822 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
823 		stop_cpus_hard(other_cpus);
824 	}
825 #endif
826 
827 	/*
828 	 * Ensure that the scheduler is stopped while panicking, even if panic
829 	 * has been entered from kdb.
830 	 */
831 	td->td_stopsched = 1;
832 
833 	bootopt = RB_AUTOBOOT;
834 	newpanic = 0;
835 	if (panicstr)
836 		bootopt |= RB_NOSYNC;
837 	else {
838 		bootopt |= RB_DUMP;
839 		panicstr = fmt;
840 		newpanic = 1;
841 	}
842 
843 	if (newpanic) {
844 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
845 		panicstr = buf;
846 		cngrab();
847 		printf("panic: %s\n", buf);
848 	} else {
849 		printf("panic: ");
850 		vprintf(fmt, ap);
851 		printf("\n");
852 	}
853 #ifdef SMP
854 	printf("cpuid = %d\n", PCPU_GET(cpuid));
855 #endif
856 	printf("time = %jd\n", (intmax_t )time_second);
857 #ifdef KDB
858 	if ((newpanic || trace_all_panics) && trace_on_panic)
859 		kdb_backtrace();
860 	if (debugger_on_panic)
861 		kdb_enter(KDB_WHY_PANIC, "panic");
862 #endif
863 	/*thread_lock(td); */
864 	td->td_flags |= TDF_INPANIC;
865 	/* thread_unlock(td); */
866 	if (!sync_on_panic)
867 		bootopt |= RB_NOSYNC;
868 	if (poweroff_on_panic)
869 		bootopt |= RB_POWEROFF;
870 	if (powercycle_on_panic)
871 		bootopt |= RB_POWERCYCLE;
872 	kern_reboot(bootopt);
873 }
874 
875 /*
876  * Support for poweroff delay.
877  *
878  * Please note that setting this delay too short might power off your machine
879  * before the write cache on your hard disk has been flushed, leading to
880  * soft-updates inconsistencies.
881  */
882 #ifndef POWEROFF_DELAY
883 # define POWEROFF_DELAY 5000
884 #endif
885 static int poweroff_delay = POWEROFF_DELAY;
886 
887 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
888     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
889 
890 static void
891 poweroff_wait(void *junk, int howto)
892 {
893 
894 	if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
895 		return;
896 	DELAY(poweroff_delay * 1000);
897 }
898 
899 /*
900  * Some system processes (e.g. syncer) need to be stopped at appropriate
901  * points in their main loops prior to a system shutdown, so that they
902  * won't interfere with the shutdown process (e.g. by holding a disk buf
903  * to cause sync to fail).  For each of these system processes, register
904  * shutdown_kproc() as a handler for one of shutdown events.
905  */
906 static int kproc_shutdown_wait = 60;
907 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
908     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
909 
910 void
911 kproc_shutdown(void *arg, int howto)
912 {
913 	struct proc *p;
914 	int error;
915 
916 	if (panicstr)
917 		return;
918 
919 	p = (struct proc *)arg;
920 	printf("Waiting (max %d seconds) for system process `%s' to stop... ",
921 	    kproc_shutdown_wait, p->p_comm);
922 	error = kproc_suspend(p, kproc_shutdown_wait * hz);
923 
924 	if (error == EWOULDBLOCK)
925 		printf("timed out\n");
926 	else
927 		printf("done\n");
928 }
929 
930 void
931 kthread_shutdown(void *arg, int howto)
932 {
933 	struct thread *td;
934 	int error;
935 
936 	if (panicstr)
937 		return;
938 
939 	td = (struct thread *)arg;
940 	printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
941 	    kproc_shutdown_wait, td->td_name);
942 	error = kthread_suspend(td, kproc_shutdown_wait * hz);
943 
944 	if (error == EWOULDBLOCK)
945 		printf("timed out\n");
946 	else
947 		printf("done\n");
948 }
949 
950 static char dumpdevname[sizeof(((struct cdev*)NULL)->si_name)];
951 SYSCTL_STRING(_kern_shutdown, OID_AUTO, dumpdevname, CTLFLAG_RD,
952     dumpdevname, 0, "Device for kernel dumps");
953 
954 static int	_dump_append(struct dumperinfo *di, void *virtual,
955 		    vm_offset_t physical, size_t length);
956 
957 #ifdef EKCD
958 static struct kerneldumpcrypto *
959 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
960     const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
961 {
962 	struct kerneldumpcrypto *kdc;
963 	struct kerneldumpkey *kdk;
964 	uint32_t dumpkeysize;
965 
966 	dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
967 	kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
968 
969 	arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
970 
971 	kdc->kdc_encryption = encryption;
972 	switch (kdc->kdc_encryption) {
973 	case KERNELDUMP_ENC_AES_256_CBC:
974 		if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
975 			goto failed;
976 		break;
977 	default:
978 		goto failed;
979 	}
980 
981 	kdc->kdc_dumpkeysize = dumpkeysize;
982 	kdk = kdc->kdc_dumpkey;
983 	kdk->kdk_encryption = kdc->kdc_encryption;
984 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
985 	kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
986 	memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
987 
988 	return (kdc);
989 failed:
990 	explicit_bzero(kdc, sizeof(*kdc) + dumpkeysize);
991 	free(kdc, M_EKCD);
992 	return (NULL);
993 }
994 
995 static int
996 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
997 {
998 	uint8_t hash[SHA256_DIGEST_LENGTH];
999 	SHA256_CTX ctx;
1000 	struct kerneldumpkey *kdk;
1001 	int error;
1002 
1003 	error = 0;
1004 
1005 	if (kdc == NULL)
1006 		return (0);
1007 
1008 	/*
1009 	 * When a user enters ddb it can write a crash dump multiple times.
1010 	 * Each time it should be encrypted using a different IV.
1011 	 */
1012 	SHA256_Init(&ctx);
1013 	SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1014 	SHA256_Final(hash, &ctx);
1015 	bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1016 
1017 	switch (kdc->kdc_encryption) {
1018 	case KERNELDUMP_ENC_AES_256_CBC:
1019 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1020 		    kdc->kdc_iv) <= 0) {
1021 			error = EINVAL;
1022 			goto out;
1023 		}
1024 		break;
1025 	default:
1026 		error = EINVAL;
1027 		goto out;
1028 	}
1029 
1030 	kdk = kdc->kdc_dumpkey;
1031 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1032 out:
1033 	explicit_bzero(hash, sizeof(hash));
1034 	return (error);
1035 }
1036 
1037 static uint32_t
1038 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
1039 {
1040 
1041 	if (kdc == NULL)
1042 		return (0);
1043 	return (kdc->kdc_dumpkeysize);
1044 }
1045 #endif /* EKCD */
1046 
1047 static struct kerneldumpcomp *
1048 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression)
1049 {
1050 	struct kerneldumpcomp *kdcomp;
1051 	int format;
1052 
1053 	switch (compression) {
1054 	case KERNELDUMP_COMP_GZIP:
1055 		format = COMPRESS_GZIP;
1056 		break;
1057 	case KERNELDUMP_COMP_ZSTD:
1058 		format = COMPRESS_ZSTD;
1059 		break;
1060 	default:
1061 		return (NULL);
1062 	}
1063 
1064 	kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO);
1065 	kdcomp->kdc_format = compression;
1066 	kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb,
1067 	    format, di->maxiosize, kerneldump_gzlevel, di);
1068 	if (kdcomp->kdc_stream == NULL) {
1069 		free(kdcomp, M_DUMPER);
1070 		return (NULL);
1071 	}
1072 	kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1073 	return (kdcomp);
1074 }
1075 
1076 static void
1077 kerneldumpcomp_destroy(struct dumperinfo *di)
1078 {
1079 	struct kerneldumpcomp *kdcomp;
1080 
1081 	kdcomp = di->kdcomp;
1082 	if (kdcomp == NULL)
1083 		return;
1084 	compressor_fini(kdcomp->kdc_stream);
1085 	explicit_bzero(kdcomp->kdc_buf, di->maxiosize);
1086 	free(kdcomp->kdc_buf, M_DUMPER);
1087 	free(kdcomp, M_DUMPER);
1088 }
1089 
1090 /* Registration of dumpers */
1091 int
1092 set_dumper(struct dumperinfo *di, const char *devname, struct thread *td,
1093     uint8_t compression, uint8_t encryption, const uint8_t *key,
1094     uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1095 {
1096 	size_t wantcopy;
1097 	int error;
1098 
1099 	error = priv_check(td, PRIV_SETDUMPER);
1100 	if (error != 0)
1101 		return (error);
1102 
1103 	if (dumper.dumper != NULL)
1104 		return (EBUSY);
1105 	dumper = *di;
1106 	dumper.blockbuf = NULL;
1107 	dumper.kdcrypto = NULL;
1108 	dumper.kdcomp = NULL;
1109 
1110 	if (encryption != KERNELDUMP_ENC_NONE) {
1111 #ifdef EKCD
1112 		dumper.kdcrypto = kerneldumpcrypto_create(di->blocksize,
1113 		    encryption, key, encryptedkeysize, encryptedkey);
1114 		if (dumper.kdcrypto == NULL) {
1115 			error = EINVAL;
1116 			goto cleanup;
1117 		}
1118 #else
1119 		error = EOPNOTSUPP;
1120 		goto cleanup;
1121 #endif
1122 	}
1123 
1124 	wantcopy = strlcpy(dumpdevname, devname, sizeof(dumpdevname));
1125 	if (wantcopy >= sizeof(dumpdevname)) {
1126 		printf("set_dumper: device name truncated from '%s' -> '%s'\n",
1127 		    devname, dumpdevname);
1128 	}
1129 
1130 	if (compression != KERNELDUMP_COMP_NONE) {
1131 		/*
1132 		 * We currently can't support simultaneous encryption and
1133 		 * compression.
1134 		 */
1135 		if (encryption != KERNELDUMP_ENC_NONE) {
1136 			error = EOPNOTSUPP;
1137 			goto cleanup;
1138 		}
1139 		dumper.kdcomp = kerneldumpcomp_create(&dumper, compression);
1140 		if (dumper.kdcomp == NULL) {
1141 			error = EINVAL;
1142 			goto cleanup;
1143 		}
1144 	}
1145 
1146 	dumper.blockbuf = malloc(di->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1147 	return (0);
1148 
1149 cleanup:
1150 	(void)clear_dumper(td);
1151 	return (error);
1152 }
1153 
1154 int
1155 clear_dumper(struct thread *td)
1156 {
1157 	int error;
1158 
1159 	error = priv_check(td, PRIV_SETDUMPER);
1160 	if (error != 0)
1161 		return (error);
1162 
1163 #ifdef NETDUMP
1164 	netdump_mbuf_drain();
1165 #endif
1166 
1167 #ifdef EKCD
1168 	if (dumper.kdcrypto != NULL) {
1169 		explicit_bzero(dumper.kdcrypto, sizeof(*dumper.kdcrypto) +
1170 		    dumper.kdcrypto->kdc_dumpkeysize);
1171 		free(dumper.kdcrypto, M_EKCD);
1172 	}
1173 #endif
1174 
1175 	kerneldumpcomp_destroy(&dumper);
1176 
1177 	if (dumper.blockbuf != NULL) {
1178 		explicit_bzero(dumper.blockbuf, dumper.blocksize);
1179 		free(dumper.blockbuf, M_DUMPER);
1180 	}
1181 	explicit_bzero(&dumper, sizeof(dumper));
1182 	dumpdevname[0] = '\0';
1183 	return (0);
1184 }
1185 
1186 static int
1187 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1188 {
1189 
1190 	if (di->mediasize > 0 && length != 0 && (offset < di->mediaoffset ||
1191 	    offset - di->mediaoffset + length > di->mediasize)) {
1192 		if (di->kdcomp != NULL && offset >= di->mediaoffset) {
1193 			printf(
1194 		    "Compressed dump failed to fit in device boundaries.\n");
1195 			return (E2BIG);
1196 		}
1197 
1198 		printf("Attempt to write outside dump device boundaries.\n"
1199 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1200 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
1201 		    (uintmax_t)length, (intmax_t)di->mediasize);
1202 		return (ENOSPC);
1203 	}
1204 	if (length % di->blocksize != 0) {
1205 		printf("Attempt to write partial block of length %ju.\n",
1206 		    (uintmax_t)length);
1207 		return (EINVAL);
1208 	}
1209 	if (offset % di->blocksize != 0) {
1210 		printf("Attempt to write at unaligned offset %jd.\n",
1211 		    (intmax_t)offset);
1212 		return (EINVAL);
1213 	}
1214 
1215 	return (0);
1216 }
1217 
1218 #ifdef EKCD
1219 static int
1220 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1221 {
1222 
1223 	switch (kdc->kdc_encryption) {
1224 	case KERNELDUMP_ENC_AES_256_CBC:
1225 		if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1226 		    8 * size, buf) <= 0) {
1227 			return (EIO);
1228 		}
1229 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1230 		    buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1231 			return (EIO);
1232 		}
1233 		break;
1234 	default:
1235 		return (EINVAL);
1236 	}
1237 
1238 	return (0);
1239 }
1240 
1241 /* Encrypt data and call dumper. */
1242 static int
1243 dump_encrypted_write(struct dumperinfo *di, void *virtual,
1244     vm_offset_t physical, off_t offset, size_t length)
1245 {
1246 	static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1247 	struct kerneldumpcrypto *kdc;
1248 	int error;
1249 	size_t nbytes;
1250 
1251 	kdc = di->kdcrypto;
1252 
1253 	while (length > 0) {
1254 		nbytes = MIN(length, sizeof(buf));
1255 		bcopy(virtual, buf, nbytes);
1256 
1257 		if (dump_encrypt(kdc, buf, nbytes) != 0)
1258 			return (EIO);
1259 
1260 		error = dump_write(di, buf, physical, offset, nbytes);
1261 		if (error != 0)
1262 			return (error);
1263 
1264 		offset += nbytes;
1265 		virtual = (void *)((uint8_t *)virtual + nbytes);
1266 		length -= nbytes;
1267 	}
1268 
1269 	return (0);
1270 }
1271 #endif /* EKCD */
1272 
1273 static int
1274 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1275 {
1276 	struct dumperinfo *di;
1277 	size_t resid, rlength;
1278 	int error;
1279 
1280 	di = arg;
1281 
1282 	if (length % di->blocksize != 0) {
1283 		/*
1284 		 * This must be the final write after flushing the compression
1285 		 * stream. Write as many full blocks as possible and stash the
1286 		 * residual data in the dumper's block buffer. It will be
1287 		 * padded and written in dump_finish().
1288 		 */
1289 		rlength = rounddown(length, di->blocksize);
1290 		if (rlength != 0) {
1291 			error = _dump_append(di, base, 0, rlength);
1292 			if (error != 0)
1293 				return (error);
1294 		}
1295 		resid = length - rlength;
1296 		memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1297 		di->kdcomp->kdc_resid = resid;
1298 		return (EAGAIN);
1299 	}
1300 	return (_dump_append(di, base, 0, length));
1301 }
1302 
1303 /*
1304  * Write kernel dump headers at the beginning and end of the dump extent.
1305  * Write the kernel dump encryption key after the leading header if we were
1306  * configured to do so.
1307  */
1308 static int
1309 dump_write_headers(struct dumperinfo *di, struct kerneldumpheader *kdh)
1310 {
1311 #ifdef EKCD
1312 	struct kerneldumpcrypto *kdc;
1313 #endif
1314 	void *buf, *key;
1315 	size_t hdrsz;
1316 	uint64_t extent;
1317 	uint32_t keysize;
1318 	int error;
1319 
1320 	hdrsz = sizeof(*kdh);
1321 	if (hdrsz > di->blocksize)
1322 		return (ENOMEM);
1323 
1324 #ifdef EKCD
1325 	kdc = di->kdcrypto;
1326 	key = kdc->kdc_dumpkey;
1327 	keysize = kerneldumpcrypto_dumpkeysize(kdc);
1328 #else
1329 	key = NULL;
1330 	keysize = 0;
1331 #endif
1332 
1333 	/*
1334 	 * If the dump device has special handling for headers, let it take care
1335 	 * of writing them out.
1336 	 */
1337 	if (di->dumper_hdr != NULL)
1338 		return (di->dumper_hdr(di, kdh, key, keysize));
1339 
1340 	if (hdrsz == di->blocksize)
1341 		buf = kdh;
1342 	else {
1343 		buf = di->blockbuf;
1344 		memset(buf, 0, di->blocksize);
1345 		memcpy(buf, kdh, hdrsz);
1346 	}
1347 
1348 	extent = dtoh64(kdh->dumpextent);
1349 #ifdef EKCD
1350 	if (kdc != NULL) {
1351 		error = dump_write(di, kdc->kdc_dumpkey, 0,
1352 		    di->mediaoffset + di->mediasize - di->blocksize - extent -
1353 		    keysize, keysize);
1354 		if (error != 0)
1355 			return (error);
1356 	}
1357 #endif
1358 
1359 	error = dump_write(di, buf, 0,
1360 	    di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1361 	    keysize, di->blocksize);
1362 	if (error == 0)
1363 		error = dump_write(di, buf, 0, di->mediaoffset + di->mediasize -
1364 		    di->blocksize, di->blocksize);
1365 	return (error);
1366 }
1367 
1368 /*
1369  * Don't touch the first SIZEOF_METADATA bytes on the dump device.  This is to
1370  * protect us from metadata and metadata from us.
1371  */
1372 #define	SIZEOF_METADATA		(64 * 1024)
1373 
1374 /*
1375  * Do some preliminary setup for a kernel dump: initialize state for encryption,
1376  * if requested, and make sure that we have enough space on the dump device.
1377  *
1378  * We set things up so that the dump ends before the last sector of the dump
1379  * device, at which the trailing header is written.
1380  *
1381  *     +-----------+------+-----+----------------------------+------+
1382  *     |           | lhdr | key |    ... kernel dump ...     | thdr |
1383  *     +-----------+------+-----+----------------------------+------+
1384  *                   1 blk  opt <------- dump extent --------> 1 blk
1385  *
1386  * Dumps written using dump_append() start at the beginning of the extent.
1387  * Uncompressed dumps will use the entire extent, but compressed dumps typically
1388  * will not. The true length of the dump is recorded in the leading and trailing
1389  * headers once the dump has been completed.
1390  *
1391  * The dump device may provide a callback, in which case it will initialize
1392  * dumpoff and take care of laying out the headers.
1393  */
1394 int
1395 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1396 {
1397 	uint64_t dumpextent, span;
1398 	uint32_t keysize;
1399 	int error;
1400 
1401 #ifdef EKCD
1402 	error = kerneldumpcrypto_init(di->kdcrypto);
1403 	if (error != 0)
1404 		return (error);
1405 	keysize = kerneldumpcrypto_dumpkeysize(di->kdcrypto);
1406 #else
1407 	error = 0;
1408 	keysize = 0;
1409 #endif
1410 
1411 	if (di->dumper_start != NULL) {
1412 		error = di->dumper_start(di);
1413 	} else {
1414 		dumpextent = dtoh64(kdh->dumpextent);
1415 		span = SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1416 		    keysize;
1417 		if (di->mediasize < span) {
1418 			if (di->kdcomp == NULL)
1419 				return (E2BIG);
1420 
1421 			/*
1422 			 * We don't yet know how much space the compressed dump
1423 			 * will occupy, so try to use the whole swap partition
1424 			 * (minus the first 64KB) in the hope that the
1425 			 * compressed dump will fit. If that doesn't turn out to
1426 			 * be enough, the bounds checking in dump_write()
1427 			 * will catch us and cause the dump to fail.
1428 			 */
1429 			dumpextent = di->mediasize - span + dumpextent;
1430 			kdh->dumpextent = htod64(dumpextent);
1431 		}
1432 
1433 		/*
1434 		 * The offset at which to begin writing the dump.
1435 		 */
1436 		di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1437 		    dumpextent;
1438 	}
1439 	di->origdumpoff = di->dumpoff;
1440 	return (error);
1441 }
1442 
1443 static int
1444 _dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1445     size_t length)
1446 {
1447 	int error;
1448 
1449 #ifdef EKCD
1450 	if (di->kdcrypto != NULL)
1451 		error = dump_encrypted_write(di, virtual, physical, di->dumpoff,
1452 		    length);
1453 	else
1454 #endif
1455 		error = dump_write(di, virtual, physical, di->dumpoff, length);
1456 	if (error == 0)
1457 		di->dumpoff += length;
1458 	return (error);
1459 }
1460 
1461 /*
1462  * Write to the dump device starting at dumpoff. When compression is enabled,
1463  * writes to the device will be performed using a callback that gets invoked
1464  * when the compression stream's output buffer is full.
1465  */
1466 int
1467 dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1468     size_t length)
1469 {
1470 	void *buf;
1471 
1472 	if (di->kdcomp != NULL) {
1473 		/* Bounce through a buffer to avoid CRC errors. */
1474 		if (length > di->maxiosize)
1475 			return (EINVAL);
1476 		buf = di->kdcomp->kdc_buf;
1477 		memmove(buf, virtual, length);
1478 		return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1479 	}
1480 	return (_dump_append(di, virtual, physical, length));
1481 }
1482 
1483 /*
1484  * Write to the dump device at the specified offset.
1485  */
1486 int
1487 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1488     off_t offset, size_t length)
1489 {
1490 	int error;
1491 
1492 	error = dump_check_bounds(di, offset, length);
1493 	if (error != 0)
1494 		return (error);
1495 	return (di->dumper(di->priv, virtual, physical, offset, length));
1496 }
1497 
1498 /*
1499  * Perform kernel dump finalization: flush the compression stream, if necessary,
1500  * write the leading and trailing kernel dump headers now that we know the true
1501  * length of the dump, and optionally write the encryption key following the
1502  * leading header.
1503  */
1504 int
1505 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1506 {
1507 	int error;
1508 
1509 	if (di->kdcomp != NULL) {
1510 		error = compressor_flush(di->kdcomp->kdc_stream);
1511 		if (error == EAGAIN) {
1512 			/* We have residual data in di->blockbuf. */
1513 			error = dump_write(di, di->blockbuf, 0, di->dumpoff,
1514 			    di->blocksize);
1515 			di->dumpoff += di->kdcomp->kdc_resid;
1516 			di->kdcomp->kdc_resid = 0;
1517 		}
1518 		if (error != 0)
1519 			return (error);
1520 
1521 		/*
1522 		 * We now know the size of the compressed dump, so update the
1523 		 * header accordingly and recompute parity.
1524 		 */
1525 		kdh->dumplength = htod64(di->dumpoff - di->origdumpoff);
1526 		kdh->parity = 0;
1527 		kdh->parity = kerneldump_parity(kdh);
1528 
1529 		compressor_reset(di->kdcomp->kdc_stream);
1530 	}
1531 
1532 	error = dump_write_headers(di, kdh);
1533 	if (error != 0)
1534 		return (error);
1535 
1536 	(void)dump_write(di, NULL, 0, 0, 0);
1537 	return (0);
1538 }
1539 
1540 void
1541 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1542     char *magic, uint32_t archver, uint64_t dumplen)
1543 {
1544 	size_t dstsize;
1545 
1546 	bzero(kdh, sizeof(*kdh));
1547 	strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1548 	strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1549 	kdh->version = htod32(KERNELDUMPVERSION);
1550 	kdh->architectureversion = htod32(archver);
1551 	kdh->dumplength = htod64(dumplen);
1552 	kdh->dumpextent = kdh->dumplength;
1553 	kdh->dumptime = htod64(time_second);
1554 #ifdef EKCD
1555 	kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1556 #else
1557 	kdh->dumpkeysize = 0;
1558 #endif
1559 	kdh->blocksize = htod32(di->blocksize);
1560 	strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1561 	dstsize = sizeof(kdh->versionstring);
1562 	if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1563 		kdh->versionstring[dstsize - 2] = '\n';
1564 	if (panicstr != NULL)
1565 		strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1566 	if (di->kdcomp != NULL)
1567 		kdh->compression = di->kdcomp->kdc_format;
1568 	kdh->parity = kerneldump_parity(kdh);
1569 }
1570 
1571 #ifdef DDB
1572 DB_SHOW_COMMAND(panic, db_show_panic)
1573 {
1574 
1575 	if (panicstr == NULL)
1576 		db_printf("panicstr not set\n");
1577 	else
1578 		db_printf("panic: %s\n", panicstr);
1579 }
1580 #endif
1581