xref: /freebsd/sys/kern/kern_shutdown.c (revision 3ee5c55415a7b08c6c4c403cc6b96e30d768e1c9)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1986, 1988, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_shutdown.c	8.3 (Berkeley) 1/21/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_ddb.h"
43 #include "opt_ekcd.h"
44 #include "opt_kdb.h"
45 #include "opt_panic.h"
46 #include "opt_sched.h"
47 #include "opt_watchdog.h"
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/bio.h>
52 #include <sys/buf.h>
53 #include <sys/conf.h>
54 #include <sys/compressor.h>
55 #include <sys/cons.h>
56 #include <sys/eventhandler.h>
57 #include <sys/filedesc.h>
58 #include <sys/jail.h>
59 #include <sys/kdb.h>
60 #include <sys/kernel.h>
61 #include <sys/kerneldump.h>
62 #include <sys/kthread.h>
63 #include <sys/ktr.h>
64 #include <sys/malloc.h>
65 #include <sys/mount.h>
66 #include <sys/priv.h>
67 #include <sys/proc.h>
68 #include <sys/reboot.h>
69 #include <sys/resourcevar.h>
70 #include <sys/rwlock.h>
71 #include <sys/sched.h>
72 #include <sys/smp.h>
73 #include <sys/sysctl.h>
74 #include <sys/sysproto.h>
75 #include <sys/taskqueue.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 static bool trace_all_panics = true;
128 #else
129 static int trace_on_panic = 0;
130 static bool trace_all_panics = false;
131 #endif
132 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
133     CTLFLAG_RWTUN | CTLFLAG_SECURE,
134     &trace_on_panic, 0, "Print stack trace on kernel panic");
135 SYSCTL_BOOL(_debug, OID_AUTO, trace_all_panics, CTLFLAG_RWTUN,
136     &trace_all_panics, 0, "Print stack traces on secondary kernel panics");
137 #endif /* KDB */
138 
139 static int sync_on_panic = 0;
140 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN,
141 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
142 
143 static bool poweroff_on_panic = 0;
144 SYSCTL_BOOL(_kern, OID_AUTO, poweroff_on_panic, CTLFLAG_RWTUN,
145 	&poweroff_on_panic, 0, "Do a power off instead of a reboot on a panic");
146 
147 static bool powercycle_on_panic = 0;
148 SYSCTL_BOOL(_kern, OID_AUTO, powercycle_on_panic, CTLFLAG_RWTUN,
149 	&powercycle_on_panic, 0, "Do a power cycle instead of a reboot on a panic");
150 
151 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0,
152     "Shutdown environment");
153 
154 #ifndef DIAGNOSTIC
155 static int show_busybufs;
156 #else
157 static int show_busybufs = 1;
158 #endif
159 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
160 	&show_busybufs, 0, "");
161 
162 int suspend_blocked = 0;
163 SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW,
164 	&suspend_blocked, 0, "Block suspend due to a pending shutdown");
165 
166 #ifdef EKCD
167 FEATURE(ekcd, "Encrypted kernel crash dumps support");
168 
169 MALLOC_DEFINE(M_EKCD, "ekcd", "Encrypted kernel crash dumps data");
170 
171 struct kerneldumpcrypto {
172 	uint8_t			kdc_encryption;
173 	uint8_t			kdc_iv[KERNELDUMP_IV_MAX_SIZE];
174 	keyInstance		kdc_ki;
175 	cipherInstance		kdc_ci;
176 	uint32_t		kdc_dumpkeysize;
177 	struct kerneldumpkey	kdc_dumpkey[];
178 };
179 #endif
180 
181 struct kerneldumpcomp {
182 	uint8_t			kdc_format;
183 	struct compressor	*kdc_stream;
184 	uint8_t			*kdc_buf;
185 	size_t			kdc_resid;
186 };
187 
188 static struct kerneldumpcomp *kerneldumpcomp_create(struct dumperinfo *di,
189 		    uint8_t compression);
190 static void	kerneldumpcomp_destroy(struct dumperinfo *di);
191 static int	kerneldumpcomp_write_cb(void *base, size_t len, off_t off, void *arg);
192 
193 static int kerneldump_gzlevel = 6;
194 SYSCTL_INT(_kern, OID_AUTO, kerneldump_gzlevel, CTLFLAG_RWTUN,
195     &kerneldump_gzlevel, 0,
196     "Kernel crash dump compression level");
197 
198 /*
199  * Variable panicstr contains argument to first call to panic; used as flag
200  * to indicate that the kernel has already called panic.
201  */
202 const char *panicstr;
203 
204 int dumping;				/* system is dumping */
205 int rebooting;				/* system is rebooting */
206 static struct dumperinfo dumper;	/* our selected dumper */
207 
208 /* Context information for dump-debuggers. */
209 static struct pcb dumppcb;		/* Registers. */
210 lwpid_t dumptid;			/* Thread ID. */
211 
212 static struct cdevsw reroot_cdevsw = {
213      .d_version = D_VERSION,
214      .d_name    = "reroot",
215 };
216 
217 static void poweroff_wait(void *, int);
218 static void shutdown_halt(void *junk, int howto);
219 static void shutdown_panic(void *junk, int howto);
220 static void shutdown_reset(void *junk, int howto);
221 static int kern_reroot(void);
222 
223 /* register various local shutdown events */
224 static void
225 shutdown_conf(void *unused)
226 {
227 
228 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
229 	    SHUTDOWN_PRI_FIRST);
230 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
231 	    SHUTDOWN_PRI_LAST + 100);
232 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
233 	    SHUTDOWN_PRI_LAST + 100);
234 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
235 	    SHUTDOWN_PRI_LAST + 200);
236 }
237 
238 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
239 
240 /*
241  * The only reason this exists is to create the /dev/reroot/ directory,
242  * used by reroot code in init(8) as a mountpoint for tmpfs.
243  */
244 static void
245 reroot_conf(void *unused)
246 {
247 	int error;
248 	struct cdev *cdev;
249 
250 	error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev,
251 	    &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot");
252 	if (error != 0) {
253 		printf("%s: failed to create device node, error %d",
254 		    __func__, error);
255 	}
256 }
257 
258 SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL);
259 
260 /*
261  * The system call that results in a reboot.
262  */
263 /* ARGSUSED */
264 int
265 sys_reboot(struct thread *td, struct reboot_args *uap)
266 {
267 	int error;
268 
269 	error = 0;
270 #ifdef MAC
271 	error = mac_system_check_reboot(td->td_ucred, uap->opt);
272 #endif
273 	if (error == 0)
274 		error = priv_check(td, PRIV_REBOOT);
275 	if (error == 0) {
276 		if (uap->opt & RB_REROOT)
277 			error = kern_reroot();
278 		else
279 			kern_reboot(uap->opt);
280 	}
281 	return (error);
282 }
283 
284 static void
285 shutdown_nice_task_fn(void *arg, int pending __unused)
286 {
287 	int howto;
288 
289 	howto = (uintptr_t)arg;
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 }
302 
303 static struct task shutdown_nice_task = TASK_INITIALIZER(0,
304     &shutdown_nice_task_fn, NULL);
305 
306 /*
307  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
308  */
309 void
310 shutdown_nice(int howto)
311 {
312 
313 	if (initproc != NULL && !SCHEDULER_STOPPED()) {
314 		shutdown_nice_task.ta_context = (void *)(uintptr_t)howto;
315 		taskqueue_enqueue(taskqueue_fast, &shutdown_nice_task);
316 	} else {
317 		/*
318 		 * No init(8) running, or scheduler would not allow it
319 		 * to run, so simply reboot.
320 		 */
321 		kern_reboot(howto | RB_NOSYNC);
322 	}
323 }
324 
325 static void
326 print_uptime(void)
327 {
328 	int f;
329 	struct timespec ts;
330 
331 	getnanouptime(&ts);
332 	printf("Uptime: ");
333 	f = 0;
334 	if (ts.tv_sec >= 86400) {
335 		printf("%ldd", (long)ts.tv_sec / 86400);
336 		ts.tv_sec %= 86400;
337 		f = 1;
338 	}
339 	if (f || ts.tv_sec >= 3600) {
340 		printf("%ldh", (long)ts.tv_sec / 3600);
341 		ts.tv_sec %= 3600;
342 		f = 1;
343 	}
344 	if (f || ts.tv_sec >= 60) {
345 		printf("%ldm", (long)ts.tv_sec / 60);
346 		ts.tv_sec %= 60;
347 		f = 1;
348 	}
349 	printf("%lds\n", (long)ts.tv_sec);
350 }
351 
352 int
353 doadump(boolean_t textdump)
354 {
355 	boolean_t coredump;
356 	int error;
357 
358 	error = 0;
359 	if (dumping)
360 		return (EBUSY);
361 	if (dumper.dumper == NULL)
362 		return (ENXIO);
363 
364 	savectx(&dumppcb);
365 	dumptid = curthread->td_tid;
366 	dumping++;
367 
368 	coredump = TRUE;
369 #ifdef DDB
370 	if (textdump && textdump_pending) {
371 		coredump = FALSE;
372 		textdump_dumpsys(&dumper);
373 	}
374 #endif
375 	if (coredump)
376 		error = dumpsys(&dumper);
377 
378 	dumping--;
379 	return (error);
380 }
381 
382 /*
383  * Shutdown the system cleanly to prepare for reboot, halt, or power off.
384  */
385 void
386 kern_reboot(int howto)
387 {
388 	static int once = 0;
389 
390 	/*
391 	 * Normal paths here don't hold Giant, but we can wind up here
392 	 * unexpectedly with it held.  Drop it now so we don't have to
393 	 * drop and pick it up elsewhere. The paths it is locking will
394 	 * never be returned to, and it is preferable to preclude
395 	 * deadlock than to lock against code that won't ever
396 	 * continue.
397 	 */
398 	while (mtx_owned(&Giant))
399 		mtx_unlock(&Giant);
400 
401 #if defined(SMP)
402 	/*
403 	 * Bind us to the first CPU so that all shutdown code runs there.  Some
404 	 * systems don't shutdown properly (i.e., ACPI power off) if we
405 	 * run on another processor.
406 	 */
407 	if (!SCHEDULER_STOPPED()) {
408 		thread_lock(curthread);
409 		sched_bind(curthread, CPU_FIRST());
410 		thread_unlock(curthread);
411 		KASSERT(PCPU_GET(cpuid) == CPU_FIRST(),
412 		    ("boot: not running on cpu 0"));
413 	}
414 #endif
415 	/* We're in the process of rebooting. */
416 	rebooting = 1;
417 
418 	/* We are out of the debugger now. */
419 	kdb_active = 0;
420 
421 	/*
422 	 * Do any callouts that should be done BEFORE syncing the filesystems.
423 	 */
424 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
425 
426 	/*
427 	 * Now sync filesystems
428 	 */
429 	if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
430 		once = 1;
431 		bufshutdown(show_busybufs);
432 	}
433 
434 	print_uptime();
435 
436 	cngrab();
437 
438 	/*
439 	 * Ok, now do things that assume all filesystem activity has
440 	 * been completed.
441 	 */
442 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
443 
444 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
445 		doadump(TRUE);
446 
447 	/* Now that we're going to really halt the system... */
448 	EVENTHANDLER_INVOKE(shutdown_final, howto);
449 
450 	for(;;) ;	/* safety against shutdown_reset not working */
451 	/* NOTREACHED */
452 }
453 
454 /*
455  * The system call that results in changing the rootfs.
456  */
457 static int
458 kern_reroot(void)
459 {
460 	struct vnode *oldrootvnode, *vp;
461 	struct mount *mp, *devmp;
462 	int error;
463 
464 	if (curproc != initproc)
465 		return (EPERM);
466 
467 	/*
468 	 * Mark the filesystem containing currently-running executable
469 	 * (the temporary copy of init(8)) busy.
470 	 */
471 	vp = curproc->p_textvp;
472 	error = vn_lock(vp, LK_SHARED);
473 	if (error != 0)
474 		return (error);
475 	mp = vp->v_mount;
476 	error = vfs_busy(mp, MBF_NOWAIT);
477 	if (error != 0) {
478 		vfs_ref(mp);
479 		VOP_UNLOCK(vp, 0);
480 		error = vfs_busy(mp, 0);
481 		vn_lock(vp, LK_SHARED | LK_RETRY);
482 		vfs_rel(mp);
483 		if (error != 0) {
484 			VOP_UNLOCK(vp, 0);
485 			return (ENOENT);
486 		}
487 		if (vp->v_iflag & VI_DOOMED) {
488 			VOP_UNLOCK(vp, 0);
489 			vfs_unbusy(mp);
490 			return (ENOENT);
491 		}
492 	}
493 	VOP_UNLOCK(vp, 0);
494 
495 	/*
496 	 * Remove the filesystem containing currently-running executable
497 	 * from the mount list, to prevent it from being unmounted
498 	 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
499 	 *
500 	 * Also preserve /dev - forcibly unmounting it could cause driver
501 	 * reinitialization.
502 	 */
503 
504 	vfs_ref(rootdevmp);
505 	devmp = rootdevmp;
506 	rootdevmp = NULL;
507 
508 	mtx_lock(&mountlist_mtx);
509 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
510 	TAILQ_REMOVE(&mountlist, devmp, mnt_list);
511 	mtx_unlock(&mountlist_mtx);
512 
513 	oldrootvnode = rootvnode;
514 
515 	/*
516 	 * Unmount everything except for the two filesystems preserved above.
517 	 */
518 	vfs_unmountall();
519 
520 	/*
521 	 * Add /dev back; vfs_mountroot() will move it into its new place.
522 	 */
523 	mtx_lock(&mountlist_mtx);
524 	TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
525 	mtx_unlock(&mountlist_mtx);
526 	rootdevmp = devmp;
527 	vfs_rel(rootdevmp);
528 
529 	/*
530 	 * Mount the new rootfs.
531 	 */
532 	vfs_mountroot();
533 
534 	/*
535 	 * Update all references to the old rootvnode.
536 	 */
537 	mountcheckdirs(oldrootvnode, rootvnode);
538 
539 	/*
540 	 * Add the temporary filesystem back and unbusy it.
541 	 */
542 	mtx_lock(&mountlist_mtx);
543 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
544 	mtx_unlock(&mountlist_mtx);
545 	vfs_unbusy(mp);
546 
547 	return (0);
548 }
549 
550 /*
551  * If the shutdown was a clean halt, behave accordingly.
552  */
553 static void
554 shutdown_halt(void *junk, int howto)
555 {
556 
557 	if (howto & RB_HALT) {
558 		printf("\n");
559 		printf("The operating system has halted.\n");
560 		printf("Please press any key to reboot.\n\n");
561 		switch (cngetc()) {
562 		case -1:		/* No console, just die */
563 			cpu_halt();
564 			/* NOTREACHED */
565 		default:
566 			break;
567 		}
568 	}
569 }
570 
571 /*
572  * Check to see if the system paniced, pause and then reboot
573  * according to the specified delay.
574  */
575 static void
576 shutdown_panic(void *junk, int howto)
577 {
578 	int loop;
579 
580 	if (howto & RB_DUMP) {
581 		if (panic_reboot_wait_time != 0) {
582 			if (panic_reboot_wait_time != -1) {
583 				printf("Automatic reboot in %d seconds - "
584 				       "press a key on the console to abort\n",
585 					panic_reboot_wait_time);
586 				for (loop = panic_reboot_wait_time * 10;
587 				     loop > 0; --loop) {
588 					DELAY(1000 * 100); /* 1/10th second */
589 					/* Did user type a key? */
590 					if (cncheckc() != -1)
591 						break;
592 				}
593 				if (!loop)
594 					return;
595 			}
596 		} else { /* zero time specified - reboot NOW */
597 			return;
598 		}
599 		printf("--> Press a key on the console to reboot,\n");
600 		printf("--> or switch off the system now.\n");
601 		cngetc();
602 	}
603 }
604 
605 /*
606  * Everything done, now reset
607  */
608 static void
609 shutdown_reset(void *junk, int howto)
610 {
611 
612 	printf("Rebooting...\n");
613 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
614 
615 	/*
616 	 * Acquiring smp_ipi_mtx here has a double effect:
617 	 * - it disables interrupts avoiding CPU0 preemption
618 	 *   by fast handlers (thus deadlocking  against other CPUs)
619 	 * - it avoids deadlocks against smp_rendezvous() or, more
620 	 *   generally, threads busy-waiting, with this spinlock held,
621 	 *   and waiting for responses by threads on other CPUs
622 	 *   (ie. smp_tlb_shootdown()).
623 	 *
624 	 * For the !SMP case it just needs to handle the former problem.
625 	 */
626 #ifdef SMP
627 	mtx_lock_spin(&smp_ipi_mtx);
628 #else
629 	spinlock_enter();
630 #endif
631 
632 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
633 	cpu_reset();
634 	/* NOTREACHED */ /* assuming reset worked */
635 }
636 
637 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
638 static int kassert_warn_only = 0;
639 #ifdef KDB
640 static int kassert_do_kdb = 0;
641 #endif
642 #ifdef KTR
643 static int kassert_do_ktr = 0;
644 #endif
645 static int kassert_do_log = 1;
646 static int kassert_log_pps_limit = 4;
647 static int kassert_log_mute_at = 0;
648 static int kassert_log_panic_at = 0;
649 static int kassert_suppress_in_panic = 0;
650 static int kassert_warnings = 0;
651 
652 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW, NULL, "kassert options");
653 
654 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, CTLFLAG_RWTUN,
655     &kassert_warn_only, 0,
656     "KASSERT triggers a panic (1) or just a warning (0)");
657 
658 #ifdef KDB
659 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, CTLFLAG_RWTUN,
660     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
661 #endif
662 
663 #ifdef KTR
664 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, CTLFLAG_RWTUN,
665     &kassert_do_ktr, 0,
666     "KASSERT does a KTR, set this to the KTRMASK you want");
667 #endif
668 
669 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, CTLFLAG_RWTUN,
670     &kassert_do_log, 0,
671     "If warn_only is enabled, log (1) or do not log (0) assertion violations");
672 
673 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RWTUN,
674     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
675 
676 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, CTLFLAG_RWTUN,
677     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
678 
679 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, CTLFLAG_RWTUN,
680     &kassert_log_pps_limit, 0, "limit number of log messages per second");
681 
682 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, CTLFLAG_RWTUN,
683     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
684 
685 SYSCTL_INT(_debug_kassert, OID_AUTO, suppress_in_panic, CTLFLAG_RWTUN,
686     &kassert_suppress_in_panic, 0,
687     "KASSERTs will be suppressed while handling a panic");
688 
689 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
690 
691 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
692     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE, NULL, 0,
693     kassert_sysctl_kassert, "I", "set to trigger a test kassert");
694 
695 static int
696 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
697 {
698 	int error, i;
699 
700 	error = sysctl_wire_old_buffer(req, sizeof(int));
701 	if (error == 0) {
702 		i = 0;
703 		error = sysctl_handle_int(oidp, &i, 0, req);
704 	}
705 	if (error != 0 || req->newptr == NULL)
706 		return (error);
707 	KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
708 	return (0);
709 }
710 
711 /*
712  * Called by KASSERT, this decides if we will panic
713  * or if we will log via printf and/or ktr.
714  */
715 void
716 kassert_panic(const char *fmt, ...)
717 {
718 	static char buf[256];
719 	va_list ap;
720 
721 	va_start(ap, fmt);
722 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
723 	va_end(ap);
724 
725 	/*
726 	 * If we are suppressing secondary panics, log the warning but do not
727 	 * re-enter panic/kdb.
728 	 */
729 	if (panicstr != NULL && kassert_suppress_in_panic) {
730 		if (kassert_do_log) {
731 			printf("KASSERT failed: %s\n", buf);
732 #ifdef KDB
733 			if (trace_all_panics && trace_on_panic)
734 				kdb_backtrace();
735 #endif
736 		}
737 		return;
738 	}
739 
740 	/*
741 	 * panic if we're not just warning, or if we've exceeded
742 	 * kassert_log_panic_at warnings.
743 	 */
744 	if (!kassert_warn_only ||
745 	    (kassert_log_panic_at > 0 &&
746 	     kassert_warnings >= kassert_log_panic_at)) {
747 		va_start(ap, fmt);
748 		vpanic(fmt, ap);
749 		/* NORETURN */
750 	}
751 #ifdef KTR
752 	if (kassert_do_ktr)
753 		CTR0(ktr_mask, buf);
754 #endif /* KTR */
755 	/*
756 	 * log if we've not yet met the mute limit.
757 	 */
758 	if (kassert_do_log &&
759 	    (kassert_log_mute_at == 0 ||
760 	     kassert_warnings < kassert_log_mute_at)) {
761 		static  struct timeval lasterr;
762 		static  int curerr;
763 
764 		if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
765 			printf("KASSERT failed: %s\n", buf);
766 			kdb_backtrace();
767 		}
768 	}
769 #ifdef KDB
770 	if (kassert_do_kdb) {
771 		kdb_enter(KDB_WHY_KASSERT, buf);
772 	}
773 #endif
774 	atomic_add_int(&kassert_warnings, 1);
775 }
776 #endif
777 
778 /*
779  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
780  * and then reboots.  If we are called twice, then we avoid trying to sync
781  * the disks as this often leads to recursive panics.
782  */
783 void
784 panic(const char *fmt, ...)
785 {
786 	va_list ap;
787 
788 	va_start(ap, fmt);
789 	vpanic(fmt, ap);
790 }
791 
792 void
793 vpanic(const char *fmt, va_list ap)
794 {
795 #ifdef SMP
796 	cpuset_t other_cpus;
797 #endif
798 	struct thread *td = curthread;
799 	int bootopt, newpanic;
800 	static char buf[256];
801 
802 	spinlock_enter();
803 
804 #ifdef SMP
805 	/*
806 	 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
807 	 * concurrently entering panic.  Only the winner will proceed
808 	 * further.
809 	 */
810 	if (panicstr == NULL && !kdb_active) {
811 		other_cpus = all_cpus;
812 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
813 		stop_cpus_hard(other_cpus);
814 	}
815 #endif
816 
817 	/*
818 	 * Ensure that the scheduler is stopped while panicking, even if panic
819 	 * has been entered from kdb.
820 	 */
821 	td->td_stopsched = 1;
822 
823 	bootopt = RB_AUTOBOOT;
824 	newpanic = 0;
825 	if (panicstr)
826 		bootopt |= RB_NOSYNC;
827 	else {
828 		bootopt |= RB_DUMP;
829 		panicstr = fmt;
830 		newpanic = 1;
831 	}
832 
833 	if (newpanic) {
834 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
835 		panicstr = buf;
836 		cngrab();
837 		printf("panic: %s\n", buf);
838 	} else {
839 		printf("panic: ");
840 		vprintf(fmt, ap);
841 		printf("\n");
842 	}
843 #ifdef SMP
844 	printf("cpuid = %d\n", PCPU_GET(cpuid));
845 #endif
846 	printf("time = %jd\n", (intmax_t )time_second);
847 #ifdef KDB
848 	if ((newpanic || trace_all_panics) && trace_on_panic)
849 		kdb_backtrace();
850 	if (debugger_on_panic)
851 		kdb_enter(KDB_WHY_PANIC, "panic");
852 #endif
853 	/*thread_lock(td); */
854 	td->td_flags |= TDF_INPANIC;
855 	/* thread_unlock(td); */
856 	if (!sync_on_panic)
857 		bootopt |= RB_NOSYNC;
858 	if (poweroff_on_panic)
859 		bootopt |= RB_POWEROFF;
860 	if (powercycle_on_panic)
861 		bootopt |= RB_POWERCYCLE;
862 	kern_reboot(bootopt);
863 }
864 
865 /*
866  * Support for poweroff delay.
867  *
868  * Please note that setting this delay too short might power off your machine
869  * before the write cache on your hard disk has been flushed, leading to
870  * soft-updates inconsistencies.
871  */
872 #ifndef POWEROFF_DELAY
873 # define POWEROFF_DELAY 5000
874 #endif
875 static int poweroff_delay = POWEROFF_DELAY;
876 
877 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
878     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
879 
880 static void
881 poweroff_wait(void *junk, int howto)
882 {
883 
884 	if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
885 		return;
886 	DELAY(poweroff_delay * 1000);
887 }
888 
889 /*
890  * Some system processes (e.g. syncer) need to be stopped at appropriate
891  * points in their main loops prior to a system shutdown, so that they
892  * won't interfere with the shutdown process (e.g. by holding a disk buf
893  * to cause sync to fail).  For each of these system processes, register
894  * shutdown_kproc() as a handler for one of shutdown events.
895  */
896 static int kproc_shutdown_wait = 60;
897 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
898     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
899 
900 void
901 kproc_shutdown(void *arg, int howto)
902 {
903 	struct proc *p;
904 	int error;
905 
906 	if (panicstr)
907 		return;
908 
909 	p = (struct proc *)arg;
910 	printf("Waiting (max %d seconds) for system process `%s' to stop... ",
911 	    kproc_shutdown_wait, p->p_comm);
912 	error = kproc_suspend(p, kproc_shutdown_wait * hz);
913 
914 	if (error == EWOULDBLOCK)
915 		printf("timed out\n");
916 	else
917 		printf("done\n");
918 }
919 
920 void
921 kthread_shutdown(void *arg, int howto)
922 {
923 	struct thread *td;
924 	int error;
925 
926 	if (panicstr)
927 		return;
928 
929 	td = (struct thread *)arg;
930 	printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
931 	    kproc_shutdown_wait, td->td_name);
932 	error = kthread_suspend(td, kproc_shutdown_wait * hz);
933 
934 	if (error == EWOULDBLOCK)
935 		printf("timed out\n");
936 	else
937 		printf("done\n");
938 }
939 
940 static char dumpdevname[sizeof(((struct cdev*)NULL)->si_name)];
941 SYSCTL_STRING(_kern_shutdown, OID_AUTO, dumpdevname, CTLFLAG_RD,
942     dumpdevname, 0, "Device for kernel dumps");
943 
944 static int	_dump_append(struct dumperinfo *di, void *virtual,
945 		    vm_offset_t physical, size_t length);
946 
947 #ifdef EKCD
948 static struct kerneldumpcrypto *
949 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
950     const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
951 {
952 	struct kerneldumpcrypto *kdc;
953 	struct kerneldumpkey *kdk;
954 	uint32_t dumpkeysize;
955 
956 	dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
957 	kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
958 
959 	arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
960 
961 	kdc->kdc_encryption = encryption;
962 	switch (kdc->kdc_encryption) {
963 	case KERNELDUMP_ENC_AES_256_CBC:
964 		if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
965 			goto failed;
966 		break;
967 	default:
968 		goto failed;
969 	}
970 
971 	kdc->kdc_dumpkeysize = dumpkeysize;
972 	kdk = kdc->kdc_dumpkey;
973 	kdk->kdk_encryption = kdc->kdc_encryption;
974 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
975 	kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
976 	memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
977 
978 	return (kdc);
979 failed:
980 	explicit_bzero(kdc, sizeof(*kdc) + dumpkeysize);
981 	free(kdc, M_EKCD);
982 	return (NULL);
983 }
984 
985 static int
986 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
987 {
988 	uint8_t hash[SHA256_DIGEST_LENGTH];
989 	SHA256_CTX ctx;
990 	struct kerneldumpkey *kdk;
991 	int error;
992 
993 	error = 0;
994 
995 	if (kdc == NULL)
996 		return (0);
997 
998 	/*
999 	 * When a user enters ddb it can write a crash dump multiple times.
1000 	 * Each time it should be encrypted using a different IV.
1001 	 */
1002 	SHA256_Init(&ctx);
1003 	SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1004 	SHA256_Final(hash, &ctx);
1005 	bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1006 
1007 	switch (kdc->kdc_encryption) {
1008 	case KERNELDUMP_ENC_AES_256_CBC:
1009 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1010 		    kdc->kdc_iv) <= 0) {
1011 			error = EINVAL;
1012 			goto out;
1013 		}
1014 		break;
1015 	default:
1016 		error = EINVAL;
1017 		goto out;
1018 	}
1019 
1020 	kdk = kdc->kdc_dumpkey;
1021 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1022 out:
1023 	explicit_bzero(hash, sizeof(hash));
1024 	return (error);
1025 }
1026 
1027 static uint32_t
1028 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
1029 {
1030 
1031 	if (kdc == NULL)
1032 		return (0);
1033 	return (kdc->kdc_dumpkeysize);
1034 }
1035 #endif /* EKCD */
1036 
1037 static struct kerneldumpcomp *
1038 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression)
1039 {
1040 	struct kerneldumpcomp *kdcomp;
1041 	int format;
1042 
1043 	switch (compression) {
1044 	case KERNELDUMP_COMP_GZIP:
1045 		format = COMPRESS_GZIP;
1046 		break;
1047 	case KERNELDUMP_COMP_ZSTD:
1048 		format = COMPRESS_ZSTD;
1049 		break;
1050 	default:
1051 		return (NULL);
1052 	}
1053 
1054 	kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO);
1055 	kdcomp->kdc_format = compression;
1056 	kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb,
1057 	    format, di->maxiosize, kerneldump_gzlevel, di);
1058 	if (kdcomp->kdc_stream == NULL) {
1059 		free(kdcomp, M_DUMPER);
1060 		return (NULL);
1061 	}
1062 	kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1063 	return (kdcomp);
1064 }
1065 
1066 static void
1067 kerneldumpcomp_destroy(struct dumperinfo *di)
1068 {
1069 	struct kerneldumpcomp *kdcomp;
1070 
1071 	kdcomp = di->kdcomp;
1072 	if (kdcomp == NULL)
1073 		return;
1074 	compressor_fini(kdcomp->kdc_stream);
1075 	explicit_bzero(kdcomp->kdc_buf, di->maxiosize);
1076 	free(kdcomp->kdc_buf, M_DUMPER);
1077 	free(kdcomp, M_DUMPER);
1078 }
1079 
1080 /* Registration of dumpers */
1081 int
1082 set_dumper(struct dumperinfo *di, const char *devname, struct thread *td,
1083     uint8_t compression, uint8_t encryption, const uint8_t *key,
1084     uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1085 {
1086 	size_t wantcopy;
1087 	int error;
1088 
1089 	error = priv_check(td, PRIV_SETDUMPER);
1090 	if (error != 0)
1091 		return (error);
1092 
1093 	if (di == NULL) {
1094 		error = 0;
1095 		goto cleanup;
1096 	}
1097 	if (dumper.dumper != NULL)
1098 		return (EBUSY);
1099 	dumper = *di;
1100 	dumper.blockbuf = NULL;
1101 	dumper.kdcrypto = NULL;
1102 	dumper.kdcomp = NULL;
1103 
1104 	if (encryption != KERNELDUMP_ENC_NONE) {
1105 #ifdef EKCD
1106 		dumper.kdcrypto = kerneldumpcrypto_create(di->blocksize,
1107 		    encryption, key, encryptedkeysize, encryptedkey);
1108 		if (dumper.kdcrypto == NULL) {
1109 			error = EINVAL;
1110 			goto cleanup;
1111 		}
1112 #else
1113 		error = EOPNOTSUPP;
1114 		goto cleanup;
1115 #endif
1116 	}
1117 
1118 	wantcopy = strlcpy(dumpdevname, devname, sizeof(dumpdevname));
1119 	if (wantcopy >= sizeof(dumpdevname)) {
1120 		printf("set_dumper: device name truncated from '%s' -> '%s'\n",
1121 		    devname, dumpdevname);
1122 	}
1123 
1124 	if (compression != KERNELDUMP_COMP_NONE) {
1125 		/*
1126 		 * We currently can't support simultaneous encryption and
1127 		 * compression.
1128 		 */
1129 		if (encryption != KERNELDUMP_ENC_NONE) {
1130 			error = EOPNOTSUPP;
1131 			goto cleanup;
1132 		}
1133 		dumper.kdcomp = kerneldumpcomp_create(&dumper, compression);
1134 		if (dumper.kdcomp == NULL) {
1135 			error = EINVAL;
1136 			goto cleanup;
1137 		}
1138 	}
1139 
1140 	dumper.blockbuf = malloc(di->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1141 	return (0);
1142 cleanup:
1143 #ifdef EKCD
1144 	if (dumper.kdcrypto != NULL) {
1145 		explicit_bzero(dumper.kdcrypto, sizeof(*dumper.kdcrypto) +
1146 		    dumper.kdcrypto->kdc_dumpkeysize);
1147 		free(dumper.kdcrypto, M_EKCD);
1148 	}
1149 #endif
1150 
1151 	kerneldumpcomp_destroy(&dumper);
1152 
1153 	if (dumper.blockbuf != NULL) {
1154 		explicit_bzero(dumper.blockbuf, dumper.blocksize);
1155 		free(dumper.blockbuf, M_DUMPER);
1156 	}
1157 	explicit_bzero(&dumper, sizeof(dumper));
1158 	dumpdevname[0] = '\0';
1159 	return (error);
1160 }
1161 
1162 static int
1163 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1164 {
1165 
1166 	if (length != 0 && (offset < di->mediaoffset ||
1167 	    offset - di->mediaoffset + length > di->mediasize)) {
1168 		if (di->kdcomp != NULL && offset >= di->mediaoffset) {
1169 			printf(
1170 		    "Compressed dump failed to fit in device boundaries.\n");
1171 			return (E2BIG);
1172 		}
1173 
1174 		printf("Attempt to write outside dump device boundaries.\n"
1175 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1176 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
1177 		    (uintmax_t)length, (intmax_t)di->mediasize);
1178 		return (ENOSPC);
1179 	}
1180 	if (length % di->blocksize != 0) {
1181 		printf("Attempt to write partial block of length %ju.\n",
1182 		    (uintmax_t)length);
1183 		return (EINVAL);
1184 	}
1185 	if (offset % di->blocksize != 0) {
1186 		printf("Attempt to write at unaligned offset %jd.\n",
1187 		    (intmax_t)offset);
1188 		return (EINVAL);
1189 	}
1190 
1191 	return (0);
1192 }
1193 
1194 #ifdef EKCD
1195 static int
1196 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1197 {
1198 
1199 	switch (kdc->kdc_encryption) {
1200 	case KERNELDUMP_ENC_AES_256_CBC:
1201 		if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1202 		    8 * size, buf) <= 0) {
1203 			return (EIO);
1204 		}
1205 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1206 		    buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1207 			return (EIO);
1208 		}
1209 		break;
1210 	default:
1211 		return (EINVAL);
1212 	}
1213 
1214 	return (0);
1215 }
1216 
1217 /* Encrypt data and call dumper. */
1218 static int
1219 dump_encrypted_write(struct dumperinfo *di, void *virtual,
1220     vm_offset_t physical, off_t offset, size_t length)
1221 {
1222 	static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1223 	struct kerneldumpcrypto *kdc;
1224 	int error;
1225 	size_t nbytes;
1226 
1227 	kdc = di->kdcrypto;
1228 
1229 	while (length > 0) {
1230 		nbytes = MIN(length, sizeof(buf));
1231 		bcopy(virtual, buf, nbytes);
1232 
1233 		if (dump_encrypt(kdc, buf, nbytes) != 0)
1234 			return (EIO);
1235 
1236 		error = dump_write(di, buf, physical, offset, nbytes);
1237 		if (error != 0)
1238 			return (error);
1239 
1240 		offset += nbytes;
1241 		virtual = (void *)((uint8_t *)virtual + nbytes);
1242 		length -= nbytes;
1243 	}
1244 
1245 	return (0);
1246 }
1247 
1248 static int
1249 dump_write_key(struct dumperinfo *di, off_t offset)
1250 {
1251 	struct kerneldumpcrypto *kdc;
1252 
1253 	kdc = di->kdcrypto;
1254 	if (kdc == NULL)
1255 		return (0);
1256 	return (dump_write(di, kdc->kdc_dumpkey, 0, offset,
1257 	    kdc->kdc_dumpkeysize));
1258 }
1259 #endif /* EKCD */
1260 
1261 static int
1262 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1263 {
1264 	struct dumperinfo *di;
1265 	size_t resid, rlength;
1266 	int error;
1267 
1268 	di = arg;
1269 
1270 	if (length % di->blocksize != 0) {
1271 		/*
1272 		 * This must be the final write after flushing the compression
1273 		 * stream. Write as many full blocks as possible and stash the
1274 		 * residual data in the dumper's block buffer. It will be
1275 		 * padded and written in dump_finish().
1276 		 */
1277 		rlength = rounddown(length, di->blocksize);
1278 		if (rlength != 0) {
1279 			error = _dump_append(di, base, 0, rlength);
1280 			if (error != 0)
1281 				return (error);
1282 		}
1283 		resid = length - rlength;
1284 		memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1285 		di->kdcomp->kdc_resid = resid;
1286 		return (EAGAIN);
1287 	}
1288 	return (_dump_append(di, base, 0, length));
1289 }
1290 
1291 /*
1292  * Write a kerneldumpheader at the specified offset. The header structure is 512
1293  * bytes in size, but we must pad to the device sector size.
1294  */
1295 static int
1296 dump_write_header(struct dumperinfo *di, struct kerneldumpheader *kdh,
1297     off_t offset)
1298 {
1299 	void *buf;
1300 	size_t hdrsz;
1301 
1302 	hdrsz = sizeof(*kdh);
1303 	if (hdrsz > di->blocksize)
1304 		return (ENOMEM);
1305 
1306 	if (hdrsz == di->blocksize)
1307 		buf = kdh;
1308 	else {
1309 		buf = di->blockbuf;
1310 		memset(buf, 0, di->blocksize);
1311 		memcpy(buf, kdh, hdrsz);
1312 	}
1313 
1314 	return (dump_write(di, buf, 0, offset, di->blocksize));
1315 }
1316 
1317 /*
1318  * Don't touch the first SIZEOF_METADATA bytes on the dump device.  This is to
1319  * protect us from metadata and metadata from us.
1320  */
1321 #define	SIZEOF_METADATA		(64 * 1024)
1322 
1323 /*
1324  * Do some preliminary setup for a kernel dump: initialize state for encryption,
1325  * if requested, and make sure that we have enough space on the dump device.
1326  *
1327  * We set things up so that the dump ends before the last sector of the dump
1328  * device, at which the trailing header is written.
1329  *
1330  *     +-----------+------+-----+----------------------------+------+
1331  *     |           | lhdr | key |    ... kernel dump ...     | thdr |
1332  *     +-----------+------+-----+----------------------------+------+
1333  *                   1 blk  opt <------- dump extent --------> 1 blk
1334  *
1335  * Dumps written using dump_append() start at the beginning of the extent.
1336  * Uncompressed dumps will use the entire extent, but compressed dumps typically
1337  * will not. The true length of the dump is recorded in the leading and trailing
1338  * headers once the dump has been completed.
1339  */
1340 int
1341 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1342 {
1343 	uint64_t dumpextent;
1344 	uint32_t keysize;
1345 
1346 #ifdef EKCD
1347 	int error = kerneldumpcrypto_init(di->kdcrypto);
1348 	if (error != 0)
1349 		return (error);
1350 	keysize = kerneldumpcrypto_dumpkeysize(di->kdcrypto);
1351 #else
1352 	keysize = 0;
1353 #endif
1354 
1355 	dumpextent = dtoh64(kdh->dumpextent);
1356 	if (di->mediasize < SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1357 	    keysize) {
1358 		if (di->kdcomp != NULL) {
1359 			/*
1360 			 * We don't yet know how much space the compressed dump
1361 			 * will occupy, so try to use the whole swap partition
1362 			 * (minus the first 64KB) in the hope that the
1363 			 * compressed dump will fit. If that doesn't turn out to
1364 			 * be enough, the bounds checking in dump_write()
1365 			 * will catch us and cause the dump to fail.
1366 			 */
1367 			dumpextent = di->mediasize - SIZEOF_METADATA -
1368 			    2 * di->blocksize - keysize;
1369 			kdh->dumpextent = htod64(dumpextent);
1370 		} else
1371 			return (E2BIG);
1372 	}
1373 
1374 	/* The offset at which to begin writing the dump. */
1375 	di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1376 	    dumpextent;
1377 
1378 	return (0);
1379 }
1380 
1381 static int
1382 _dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1383     size_t length)
1384 {
1385 	int error;
1386 
1387 #ifdef EKCD
1388 	if (di->kdcrypto != NULL)
1389 		error = dump_encrypted_write(di, virtual, physical, di->dumpoff,
1390 		    length);
1391 	else
1392 #endif
1393 		error = dump_write(di, virtual, physical, di->dumpoff, length);
1394 	if (error == 0)
1395 		di->dumpoff += length;
1396 	return (error);
1397 }
1398 
1399 /*
1400  * Write to the dump device starting at dumpoff. When compression is enabled,
1401  * writes to the device will be performed using a callback that gets invoked
1402  * when the compression stream's output buffer is full.
1403  */
1404 int
1405 dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1406     size_t length)
1407 {
1408 	void *buf;
1409 
1410 	if (di->kdcomp != NULL) {
1411 		/* Bounce through a buffer to avoid CRC errors. */
1412 		if (length > di->maxiosize)
1413 			return (EINVAL);
1414 		buf = di->kdcomp->kdc_buf;
1415 		memmove(buf, virtual, length);
1416 		return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1417 	}
1418 	return (_dump_append(di, virtual, physical, length));
1419 }
1420 
1421 /*
1422  * Write to the dump device at the specified offset.
1423  */
1424 int
1425 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1426     off_t offset, size_t length)
1427 {
1428 	int error;
1429 
1430 	error = dump_check_bounds(di, offset, length);
1431 	if (error != 0)
1432 		return (error);
1433 	return (di->dumper(di->priv, virtual, physical, offset, length));
1434 }
1435 
1436 /*
1437  * Perform kernel dump finalization: flush the compression stream, if necessary,
1438  * write the leading and trailing kernel dump headers now that we know the true
1439  * length of the dump, and optionally write the encryption key following the
1440  * leading header.
1441  */
1442 int
1443 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1444 {
1445 	uint64_t extent;
1446 	uint32_t keysize;
1447 	int error;
1448 
1449 	extent = dtoh64(kdh->dumpextent);
1450 
1451 #ifdef EKCD
1452 	keysize = kerneldumpcrypto_dumpkeysize(di->kdcrypto);
1453 #else
1454 	keysize = 0;
1455 #endif
1456 
1457 	if (di->kdcomp != NULL) {
1458 		error = compressor_flush(di->kdcomp->kdc_stream);
1459 		if (error == EAGAIN) {
1460 			/* We have residual data in di->blockbuf. */
1461 			error = dump_write(di, di->blockbuf, 0, di->dumpoff,
1462 			    di->blocksize);
1463 			di->dumpoff += di->kdcomp->kdc_resid;
1464 			di->kdcomp->kdc_resid = 0;
1465 		}
1466 		if (error != 0)
1467 			return (error);
1468 
1469 		/*
1470 		 * We now know the size of the compressed dump, so update the
1471 		 * header accordingly and recompute parity.
1472 		 */
1473 		kdh->dumplength = htod64(di->dumpoff -
1474 		    (di->mediaoffset + di->mediasize - di->blocksize - extent));
1475 		kdh->parity = 0;
1476 		kdh->parity = kerneldump_parity(kdh);
1477 
1478 		compressor_reset(di->kdcomp->kdc_stream);
1479 	}
1480 
1481 	/*
1482 	 * Write kerneldump headers at the beginning and end of the dump extent.
1483 	 * Write the key after the leading header.
1484 	 */
1485 	error = dump_write_header(di, kdh,
1486 	    di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1487 	    keysize);
1488 	if (error != 0)
1489 		return (error);
1490 
1491 #ifdef EKCD
1492 	error = dump_write_key(di,
1493 	    di->mediaoffset + di->mediasize - di->blocksize - extent - keysize);
1494 	if (error != 0)
1495 		return (error);
1496 #endif
1497 
1498 	error = dump_write_header(di, kdh,
1499 	    di->mediaoffset + di->mediasize - di->blocksize);
1500 	if (error != 0)
1501 		return (error);
1502 
1503 	(void)dump_write(di, NULL, 0, 0, 0);
1504 	return (0);
1505 }
1506 
1507 void
1508 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1509     char *magic, uint32_t archver, uint64_t dumplen)
1510 {
1511 	size_t dstsize;
1512 
1513 	bzero(kdh, sizeof(*kdh));
1514 	strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1515 	strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1516 	kdh->version = htod32(KERNELDUMPVERSION);
1517 	kdh->architectureversion = htod32(archver);
1518 	kdh->dumplength = htod64(dumplen);
1519 	kdh->dumpextent = kdh->dumplength;
1520 	kdh->dumptime = htod64(time_second);
1521 #ifdef EKCD
1522 	kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1523 #else
1524 	kdh->dumpkeysize = 0;
1525 #endif
1526 	kdh->blocksize = htod32(di->blocksize);
1527 	strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1528 	dstsize = sizeof(kdh->versionstring);
1529 	if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1530 		kdh->versionstring[dstsize - 2] = '\n';
1531 	if (panicstr != NULL)
1532 		strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1533 	if (di->kdcomp != NULL)
1534 		kdh->compression = di->kdcomp->kdc_format;
1535 	kdh->parity = kerneldump_parity(kdh);
1536 }
1537 
1538 #ifdef DDB
1539 DB_SHOW_COMMAND(panic, db_show_panic)
1540 {
1541 
1542 	if (panicstr == NULL)
1543 		db_printf("panicstr not set\n");
1544 	else
1545 		db_printf("panic: %s\n", panicstr);
1546 }
1547 #endif
1548