xref: /freebsd/sys/kern/kern_shutdown.c (revision f5147e312f43a9050468de539aeafa072caa1a60)
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 #else
128 static int trace_on_panic = 0;
129 #endif
130 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
131     CTLFLAG_RWTUN | CTLFLAG_SECURE,
132     &trace_on_panic, 0, "Print stack trace on kernel panic");
133 #endif /* KDB */
134 
135 static int sync_on_panic = 0;
136 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN,
137 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
138 
139 static bool poweroff_on_panic = 0;
140 SYSCTL_BOOL(_kern, OID_AUTO, poweroff_on_panic, CTLFLAG_RWTUN,
141 	&poweroff_on_panic, 0, "Do a power off instead of a reboot on a panic");
142 
143 static bool powercycle_on_panic = 0;
144 SYSCTL_BOOL(_kern, OID_AUTO, powercycle_on_panic, CTLFLAG_RWTUN,
145 	&powercycle_on_panic, 0, "Do a power cycle instead of a reboot on a panic");
146 
147 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0,
148     "Shutdown environment");
149 
150 #ifndef DIAGNOSTIC
151 static int show_busybufs;
152 #else
153 static int show_busybufs = 1;
154 #endif
155 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
156 	&show_busybufs, 0, "");
157 
158 int suspend_blocked = 0;
159 SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW,
160 	&suspend_blocked, 0, "Block suspend due to a pending shutdown");
161 
162 #ifdef EKCD
163 FEATURE(ekcd, "Encrypted kernel crash dumps support");
164 
165 MALLOC_DEFINE(M_EKCD, "ekcd", "Encrypted kernel crash dumps data");
166 
167 struct kerneldumpcrypto {
168 	uint8_t			kdc_encryption;
169 	uint8_t			kdc_iv[KERNELDUMP_IV_MAX_SIZE];
170 	keyInstance		kdc_ki;
171 	cipherInstance		kdc_ci;
172 	uint32_t		kdc_dumpkeysize;
173 	struct kerneldumpkey	kdc_dumpkey[];
174 };
175 #endif
176 
177 struct kerneldumpcomp {
178 	uint8_t			kdc_format;
179 	struct compressor	*kdc_stream;
180 	uint8_t			*kdc_buf;
181 	size_t			kdc_resid;
182 };
183 
184 static struct kerneldumpcomp *kerneldumpcomp_create(struct dumperinfo *di,
185 		    uint8_t compression);
186 static void	kerneldumpcomp_destroy(struct dumperinfo *di);
187 static int	kerneldumpcomp_write_cb(void *base, size_t len, off_t off, void *arg);
188 
189 static int kerneldump_gzlevel = 6;
190 SYSCTL_INT(_kern, OID_AUTO, kerneldump_gzlevel, CTLFLAG_RWTUN,
191     &kerneldump_gzlevel, 0,
192     "Kernel crash dump compression level");
193 
194 /*
195  * Variable panicstr contains argument to first call to panic; used as flag
196  * to indicate that the kernel has already called panic.
197  */
198 const char *panicstr;
199 
200 int dumping;				/* system is dumping */
201 int rebooting;				/* system is rebooting */
202 static struct dumperinfo dumper;	/* our selected dumper */
203 
204 /* Context information for dump-debuggers. */
205 static struct pcb dumppcb;		/* Registers. */
206 lwpid_t dumptid;			/* Thread ID. */
207 
208 static struct cdevsw reroot_cdevsw = {
209      .d_version = D_VERSION,
210      .d_name    = "reroot",
211 };
212 
213 static void poweroff_wait(void *, int);
214 static void shutdown_halt(void *junk, int howto);
215 static void shutdown_panic(void *junk, int howto);
216 static void shutdown_reset(void *junk, int howto);
217 static int kern_reroot(void);
218 
219 /* register various local shutdown events */
220 static void
221 shutdown_conf(void *unused)
222 {
223 
224 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
225 	    SHUTDOWN_PRI_FIRST);
226 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
227 	    SHUTDOWN_PRI_LAST + 100);
228 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
229 	    SHUTDOWN_PRI_LAST + 100);
230 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
231 	    SHUTDOWN_PRI_LAST + 200);
232 }
233 
234 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
235 
236 /*
237  * The only reason this exists is to create the /dev/reroot/ directory,
238  * used by reroot code in init(8) as a mountpoint for tmpfs.
239  */
240 static void
241 reroot_conf(void *unused)
242 {
243 	int error;
244 	struct cdev *cdev;
245 
246 	error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev,
247 	    &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot");
248 	if (error != 0) {
249 		printf("%s: failed to create device node, error %d",
250 		    __func__, error);
251 	}
252 }
253 
254 SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL);
255 
256 /*
257  * The system call that results in a reboot.
258  */
259 /* ARGSUSED */
260 int
261 sys_reboot(struct thread *td, struct reboot_args *uap)
262 {
263 	int error;
264 
265 	error = 0;
266 #ifdef MAC
267 	error = mac_system_check_reboot(td->td_ucred, uap->opt);
268 #endif
269 	if (error == 0)
270 		error = priv_check(td, PRIV_REBOOT);
271 	if (error == 0) {
272 		if (uap->opt & RB_REROOT)
273 			error = kern_reroot();
274 		else
275 			kern_reboot(uap->opt);
276 	}
277 	return (error);
278 }
279 
280 static void
281 shutdown_nice_task_fn(void *arg, int pending __unused)
282 {
283 	int howto;
284 
285 	howto = (uintptr_t)arg;
286 	/* Send a signal to init(8) and have it shutdown the world. */
287 	PROC_LOCK(initproc);
288 	if (howto & RB_POWEROFF)
289 		kern_psignal(initproc, SIGUSR2);
290 	else if (howto & RB_POWERCYCLE)
291 		kern_psignal(initproc, SIGWINCH);
292 	else if (howto & RB_HALT)
293 		kern_psignal(initproc, SIGUSR1);
294 	else
295 		kern_psignal(initproc, SIGINT);
296 	PROC_UNLOCK(initproc);
297 }
298 
299 static struct task shutdown_nice_task = TASK_INITIALIZER(0,
300     &shutdown_nice_task_fn, NULL);
301 
302 /*
303  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
304  */
305 void
306 shutdown_nice(int howto)
307 {
308 
309 	if (initproc != NULL && !SCHEDULER_STOPPED()) {
310 		shutdown_nice_task.ta_context = (void *)(uintptr_t)howto;
311 		taskqueue_enqueue(taskqueue_fast, &shutdown_nice_task);
312 	} else {
313 		/*
314 		 * No init(8) running, or scheduler would not allow it
315 		 * to run, so simply reboot.
316 		 */
317 		kern_reboot(howto | RB_NOSYNC);
318 	}
319 }
320 
321 static void
322 print_uptime(void)
323 {
324 	int f;
325 	struct timespec ts;
326 
327 	getnanouptime(&ts);
328 	printf("Uptime: ");
329 	f = 0;
330 	if (ts.tv_sec >= 86400) {
331 		printf("%ldd", (long)ts.tv_sec / 86400);
332 		ts.tv_sec %= 86400;
333 		f = 1;
334 	}
335 	if (f || ts.tv_sec >= 3600) {
336 		printf("%ldh", (long)ts.tv_sec / 3600);
337 		ts.tv_sec %= 3600;
338 		f = 1;
339 	}
340 	if (f || ts.tv_sec >= 60) {
341 		printf("%ldm", (long)ts.tv_sec / 60);
342 		ts.tv_sec %= 60;
343 		f = 1;
344 	}
345 	printf("%lds\n", (long)ts.tv_sec);
346 }
347 
348 int
349 doadump(boolean_t textdump)
350 {
351 	boolean_t coredump;
352 	int error;
353 
354 	error = 0;
355 	if (dumping)
356 		return (EBUSY);
357 	if (dumper.dumper == NULL)
358 		return (ENXIO);
359 
360 	savectx(&dumppcb);
361 	dumptid = curthread->td_tid;
362 	dumping++;
363 
364 	coredump = TRUE;
365 #ifdef DDB
366 	if (textdump && textdump_pending) {
367 		coredump = FALSE;
368 		textdump_dumpsys(&dumper);
369 	}
370 #endif
371 	if (coredump)
372 		error = dumpsys(&dumper);
373 
374 	dumping--;
375 	return (error);
376 }
377 
378 /*
379  * Shutdown the system cleanly to prepare for reboot, halt, or power off.
380  */
381 void
382 kern_reboot(int howto)
383 {
384 	static int once = 0;
385 
386 	/*
387 	 * Normal paths here don't hold Giant, but we can wind up here
388 	 * unexpectedly with it held.  Drop it now so we don't have to
389 	 * drop and pick it up elsewhere. The paths it is locking will
390 	 * never be returned to, and it is preferable to preclude
391 	 * deadlock than to lock against code that won't ever
392 	 * continue.
393 	 */
394 	while (mtx_owned(&Giant))
395 		mtx_unlock(&Giant);
396 
397 #if defined(SMP)
398 	/*
399 	 * Bind us to the first CPU so that all shutdown code runs there.  Some
400 	 * systems don't shutdown properly (i.e., ACPI power off) if we
401 	 * run on another processor.
402 	 */
403 	if (!SCHEDULER_STOPPED()) {
404 		thread_lock(curthread);
405 		sched_bind(curthread, CPU_FIRST());
406 		thread_unlock(curthread);
407 		KASSERT(PCPU_GET(cpuid) == CPU_FIRST(),
408 		    ("boot: not running on cpu 0"));
409 	}
410 #endif
411 	/* We're in the process of rebooting. */
412 	rebooting = 1;
413 
414 	/* We are out of the debugger now. */
415 	kdb_active = 0;
416 
417 	/*
418 	 * Do any callouts that should be done BEFORE syncing the filesystems.
419 	 */
420 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
421 
422 	/*
423 	 * Now sync filesystems
424 	 */
425 	if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
426 		once = 1;
427 		bufshutdown(show_busybufs);
428 	}
429 
430 	print_uptime();
431 
432 	cngrab();
433 
434 	/*
435 	 * Ok, now do things that assume all filesystem activity has
436 	 * been completed.
437 	 */
438 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
439 
440 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
441 		doadump(TRUE);
442 
443 	/* Now that we're going to really halt the system... */
444 	EVENTHANDLER_INVOKE(shutdown_final, howto);
445 
446 	for(;;) ;	/* safety against shutdown_reset not working */
447 	/* NOTREACHED */
448 }
449 
450 /*
451  * The system call that results in changing the rootfs.
452  */
453 static int
454 kern_reroot(void)
455 {
456 	struct vnode *oldrootvnode, *vp;
457 	struct mount *mp, *devmp;
458 	int error;
459 
460 	if (curproc != initproc)
461 		return (EPERM);
462 
463 	/*
464 	 * Mark the filesystem containing currently-running executable
465 	 * (the temporary copy of init(8)) busy.
466 	 */
467 	vp = curproc->p_textvp;
468 	error = vn_lock(vp, LK_SHARED);
469 	if (error != 0)
470 		return (error);
471 	mp = vp->v_mount;
472 	error = vfs_busy(mp, MBF_NOWAIT);
473 	if (error != 0) {
474 		vfs_ref(mp);
475 		VOP_UNLOCK(vp, 0);
476 		error = vfs_busy(mp, 0);
477 		vn_lock(vp, LK_SHARED | LK_RETRY);
478 		vfs_rel(mp);
479 		if (error != 0) {
480 			VOP_UNLOCK(vp, 0);
481 			return (ENOENT);
482 		}
483 		if (vp->v_iflag & VI_DOOMED) {
484 			VOP_UNLOCK(vp, 0);
485 			vfs_unbusy(mp);
486 			return (ENOENT);
487 		}
488 	}
489 	VOP_UNLOCK(vp, 0);
490 
491 	/*
492 	 * Remove the filesystem containing currently-running executable
493 	 * from the mount list, to prevent it from being unmounted
494 	 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
495 	 *
496 	 * Also preserve /dev - forcibly unmounting it could cause driver
497 	 * reinitialization.
498 	 */
499 
500 	vfs_ref(rootdevmp);
501 	devmp = rootdevmp;
502 	rootdevmp = NULL;
503 
504 	mtx_lock(&mountlist_mtx);
505 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
506 	TAILQ_REMOVE(&mountlist, devmp, mnt_list);
507 	mtx_unlock(&mountlist_mtx);
508 
509 	oldrootvnode = rootvnode;
510 
511 	/*
512 	 * Unmount everything except for the two filesystems preserved above.
513 	 */
514 	vfs_unmountall();
515 
516 	/*
517 	 * Add /dev back; vfs_mountroot() will move it into its new place.
518 	 */
519 	mtx_lock(&mountlist_mtx);
520 	TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
521 	mtx_unlock(&mountlist_mtx);
522 	rootdevmp = devmp;
523 	vfs_rel(rootdevmp);
524 
525 	/*
526 	 * Mount the new rootfs.
527 	 */
528 	vfs_mountroot();
529 
530 	/*
531 	 * Update all references to the old rootvnode.
532 	 */
533 	mountcheckdirs(oldrootvnode, rootvnode);
534 
535 	/*
536 	 * Add the temporary filesystem back and unbusy it.
537 	 */
538 	mtx_lock(&mountlist_mtx);
539 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
540 	mtx_unlock(&mountlist_mtx);
541 	vfs_unbusy(mp);
542 
543 	return (0);
544 }
545 
546 /*
547  * If the shutdown was a clean halt, behave accordingly.
548  */
549 static void
550 shutdown_halt(void *junk, int howto)
551 {
552 
553 	if (howto & RB_HALT) {
554 		printf("\n");
555 		printf("The operating system has halted.\n");
556 		printf("Please press any key to reboot.\n\n");
557 		switch (cngetc()) {
558 		case -1:		/* No console, just die */
559 			cpu_halt();
560 			/* NOTREACHED */
561 		default:
562 			break;
563 		}
564 	}
565 }
566 
567 /*
568  * Check to see if the system paniced, pause and then reboot
569  * according to the specified delay.
570  */
571 static void
572 shutdown_panic(void *junk, int howto)
573 {
574 	int loop;
575 
576 	if (howto & RB_DUMP) {
577 		if (panic_reboot_wait_time != 0) {
578 			if (panic_reboot_wait_time != -1) {
579 				printf("Automatic reboot in %d seconds - "
580 				       "press a key on the console to abort\n",
581 					panic_reboot_wait_time);
582 				for (loop = panic_reboot_wait_time * 10;
583 				     loop > 0; --loop) {
584 					DELAY(1000 * 100); /* 1/10th second */
585 					/* Did user type a key? */
586 					if (cncheckc() != -1)
587 						break;
588 				}
589 				if (!loop)
590 					return;
591 			}
592 		} else { /* zero time specified - reboot NOW */
593 			return;
594 		}
595 		printf("--> Press a key on the console to reboot,\n");
596 		printf("--> or switch off the system now.\n");
597 		cngetc();
598 	}
599 }
600 
601 /*
602  * Everything done, now reset
603  */
604 static void
605 shutdown_reset(void *junk, int howto)
606 {
607 
608 	printf("Rebooting...\n");
609 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
610 
611 	/*
612 	 * Acquiring smp_ipi_mtx here has a double effect:
613 	 * - it disables interrupts avoiding CPU0 preemption
614 	 *   by fast handlers (thus deadlocking  against other CPUs)
615 	 * - it avoids deadlocks against smp_rendezvous() or, more
616 	 *   generally, threads busy-waiting, with this spinlock held,
617 	 *   and waiting for responses by threads on other CPUs
618 	 *   (ie. smp_tlb_shootdown()).
619 	 *
620 	 * For the !SMP case it just needs to handle the former problem.
621 	 */
622 #ifdef SMP
623 	mtx_lock_spin(&smp_ipi_mtx);
624 #else
625 	spinlock_enter();
626 #endif
627 
628 	/* cpu_boot(howto); */ /* doesn't do anything at the moment */
629 	cpu_reset();
630 	/* NOTREACHED */ /* assuming reset worked */
631 }
632 
633 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
634 static int kassert_warn_only = 0;
635 #ifdef KDB
636 static int kassert_do_kdb = 0;
637 #endif
638 #ifdef KTR
639 static int kassert_do_ktr = 0;
640 #endif
641 static int kassert_do_log = 1;
642 static int kassert_log_pps_limit = 4;
643 static int kassert_log_mute_at = 0;
644 static int kassert_log_panic_at = 0;
645 static int kassert_warnings = 0;
646 
647 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW, NULL, "kassert options");
648 
649 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, CTLFLAG_RWTUN,
650     &kassert_warn_only, 0,
651     "KASSERT triggers a panic (1) or just a warning (0)");
652 
653 #ifdef KDB
654 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, CTLFLAG_RWTUN,
655     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
656 #endif
657 
658 #ifdef KTR
659 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, CTLFLAG_RWTUN,
660     &kassert_do_ktr, 0,
661     "KASSERT does a KTR, set this to the KTRMASK you want");
662 #endif
663 
664 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, CTLFLAG_RWTUN,
665     &kassert_do_log, 0, "KASSERT triggers a panic (1) or just a warning (0)");
666 
667 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RWTUN,
668     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
669 
670 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, CTLFLAG_RWTUN,
671     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
672 
673 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, CTLFLAG_RWTUN,
674     &kassert_log_pps_limit, 0, "limit number of log messages per second");
675 
676 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, CTLFLAG_RWTUN,
677     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
678 
679 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
680 
681 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
682     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE, NULL, 0,
683     kassert_sysctl_kassert, "I", "set to trigger a test kassert");
684 
685 static int
686 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
687 {
688 	int error, i;
689 
690 	error = sysctl_wire_old_buffer(req, sizeof(int));
691 	if (error == 0) {
692 		i = 0;
693 		error = sysctl_handle_int(oidp, &i, 0, req);
694 	}
695 	if (error != 0 || req->newptr == NULL)
696 		return (error);
697 	KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
698 	return (0);
699 }
700 
701 /*
702  * Called by KASSERT, this decides if we will panic
703  * or if we will log via printf and/or ktr.
704  */
705 void
706 kassert_panic(const char *fmt, ...)
707 {
708 	static char buf[256];
709 	va_list ap;
710 
711 	va_start(ap, fmt);
712 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
713 	va_end(ap);
714 
715 	/*
716 	 * panic if we're not just warning, or if we've exceeded
717 	 * kassert_log_panic_at warnings.
718 	 */
719 	if (!kassert_warn_only ||
720 	    (kassert_log_panic_at > 0 &&
721 	     kassert_warnings >= kassert_log_panic_at)) {
722 		va_start(ap, fmt);
723 		vpanic(fmt, ap);
724 		/* NORETURN */
725 	}
726 #ifdef KTR
727 	if (kassert_do_ktr)
728 		CTR0(ktr_mask, buf);
729 #endif /* KTR */
730 	/*
731 	 * log if we've not yet met the mute limit.
732 	 */
733 	if (kassert_do_log &&
734 	    (kassert_log_mute_at == 0 ||
735 	     kassert_warnings < kassert_log_mute_at)) {
736 		static  struct timeval lasterr;
737 		static  int curerr;
738 
739 		if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
740 			printf("KASSERT failed: %s\n", buf);
741 			kdb_backtrace();
742 		}
743 	}
744 #ifdef KDB
745 	if (kassert_do_kdb) {
746 		kdb_enter(KDB_WHY_KASSERT, buf);
747 	}
748 #endif
749 	atomic_add_int(&kassert_warnings, 1);
750 }
751 #endif
752 
753 /*
754  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
755  * and then reboots.  If we are called twice, then we avoid trying to sync
756  * the disks as this often leads to recursive panics.
757  */
758 void
759 panic(const char *fmt, ...)
760 {
761 	va_list ap;
762 
763 	va_start(ap, fmt);
764 	vpanic(fmt, ap);
765 }
766 
767 void
768 vpanic(const char *fmt, va_list ap)
769 {
770 #ifdef SMP
771 	cpuset_t other_cpus;
772 #endif
773 	struct thread *td = curthread;
774 	int bootopt, newpanic;
775 	static char buf[256];
776 
777 	spinlock_enter();
778 
779 #ifdef SMP
780 	/*
781 	 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
782 	 * concurrently entering panic.  Only the winner will proceed
783 	 * further.
784 	 */
785 	if (panicstr == NULL && !kdb_active) {
786 		other_cpus = all_cpus;
787 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
788 		stop_cpus_hard(other_cpus);
789 	}
790 #endif
791 
792 	/*
793 	 * Ensure that the scheduler is stopped while panicking, even if panic
794 	 * has been entered from kdb.
795 	 */
796 	td->td_stopsched = 1;
797 
798 	bootopt = RB_AUTOBOOT;
799 	newpanic = 0;
800 	if (panicstr)
801 		bootopt |= RB_NOSYNC;
802 	else {
803 		bootopt |= RB_DUMP;
804 		panicstr = fmt;
805 		newpanic = 1;
806 	}
807 
808 	if (newpanic) {
809 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
810 		panicstr = buf;
811 		cngrab();
812 		printf("panic: %s\n", buf);
813 	} else {
814 		printf("panic: ");
815 		vprintf(fmt, ap);
816 		printf("\n");
817 	}
818 #ifdef SMP
819 	printf("cpuid = %d\n", PCPU_GET(cpuid));
820 #endif
821 	printf("time = %jd\n", (intmax_t )time_second);
822 #ifdef KDB
823 	if (newpanic && trace_on_panic)
824 		kdb_backtrace();
825 	if (debugger_on_panic)
826 		kdb_enter(KDB_WHY_PANIC, "panic");
827 #endif
828 	/*thread_lock(td); */
829 	td->td_flags |= TDF_INPANIC;
830 	/* thread_unlock(td); */
831 	if (!sync_on_panic)
832 		bootopt |= RB_NOSYNC;
833 	if (poweroff_on_panic)
834 		bootopt |= RB_POWEROFF;
835 	if (powercycle_on_panic)
836 		bootopt |= RB_POWERCYCLE;
837 	kern_reboot(bootopt);
838 }
839 
840 /*
841  * Support for poweroff delay.
842  *
843  * Please note that setting this delay too short might power off your machine
844  * before the write cache on your hard disk has been flushed, leading to
845  * soft-updates inconsistencies.
846  */
847 #ifndef POWEROFF_DELAY
848 # define POWEROFF_DELAY 5000
849 #endif
850 static int poweroff_delay = POWEROFF_DELAY;
851 
852 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
853     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
854 
855 static void
856 poweroff_wait(void *junk, int howto)
857 {
858 
859 	if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
860 		return;
861 	DELAY(poweroff_delay * 1000);
862 }
863 
864 /*
865  * Some system processes (e.g. syncer) need to be stopped at appropriate
866  * points in their main loops prior to a system shutdown, so that they
867  * won't interfere with the shutdown process (e.g. by holding a disk buf
868  * to cause sync to fail).  For each of these system processes, register
869  * shutdown_kproc() as a handler for one of shutdown events.
870  */
871 static int kproc_shutdown_wait = 60;
872 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
873     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
874 
875 void
876 kproc_shutdown(void *arg, int howto)
877 {
878 	struct proc *p;
879 	int error;
880 
881 	if (panicstr)
882 		return;
883 
884 	p = (struct proc *)arg;
885 	printf("Waiting (max %d seconds) for system process `%s' to stop... ",
886 	    kproc_shutdown_wait, p->p_comm);
887 	error = kproc_suspend(p, kproc_shutdown_wait * hz);
888 
889 	if (error == EWOULDBLOCK)
890 		printf("timed out\n");
891 	else
892 		printf("done\n");
893 }
894 
895 void
896 kthread_shutdown(void *arg, int howto)
897 {
898 	struct thread *td;
899 	int error;
900 
901 	if (panicstr)
902 		return;
903 
904 	td = (struct thread *)arg;
905 	printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
906 	    kproc_shutdown_wait, td->td_name);
907 	error = kthread_suspend(td, kproc_shutdown_wait * hz);
908 
909 	if (error == EWOULDBLOCK)
910 		printf("timed out\n");
911 	else
912 		printf("done\n");
913 }
914 
915 static char dumpdevname[sizeof(((struct cdev*)NULL)->si_name)];
916 SYSCTL_STRING(_kern_shutdown, OID_AUTO, dumpdevname, CTLFLAG_RD,
917     dumpdevname, 0, "Device for kernel dumps");
918 
919 static int	_dump_append(struct dumperinfo *di, void *virtual,
920 		    vm_offset_t physical, size_t length);
921 
922 #ifdef EKCD
923 static struct kerneldumpcrypto *
924 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
925     const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
926 {
927 	struct kerneldumpcrypto *kdc;
928 	struct kerneldumpkey *kdk;
929 	uint32_t dumpkeysize;
930 
931 	dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
932 	kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
933 
934 	arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
935 
936 	kdc->kdc_encryption = encryption;
937 	switch (kdc->kdc_encryption) {
938 	case KERNELDUMP_ENC_AES_256_CBC:
939 		if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
940 			goto failed;
941 		break;
942 	default:
943 		goto failed;
944 	}
945 
946 	kdc->kdc_dumpkeysize = dumpkeysize;
947 	kdk = kdc->kdc_dumpkey;
948 	kdk->kdk_encryption = kdc->kdc_encryption;
949 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
950 	kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
951 	memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
952 
953 	return (kdc);
954 failed:
955 	explicit_bzero(kdc, sizeof(*kdc) + dumpkeysize);
956 	free(kdc, M_EKCD);
957 	return (NULL);
958 }
959 
960 static int
961 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
962 {
963 	uint8_t hash[SHA256_DIGEST_LENGTH];
964 	SHA256_CTX ctx;
965 	struct kerneldumpkey *kdk;
966 	int error;
967 
968 	error = 0;
969 
970 	if (kdc == NULL)
971 		return (0);
972 
973 	/*
974 	 * When a user enters ddb it can write a crash dump multiple times.
975 	 * Each time it should be encrypted using a different IV.
976 	 */
977 	SHA256_Init(&ctx);
978 	SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
979 	SHA256_Final(hash, &ctx);
980 	bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
981 
982 	switch (kdc->kdc_encryption) {
983 	case KERNELDUMP_ENC_AES_256_CBC:
984 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
985 		    kdc->kdc_iv) <= 0) {
986 			error = EINVAL;
987 			goto out;
988 		}
989 		break;
990 	default:
991 		error = EINVAL;
992 		goto out;
993 	}
994 
995 	kdk = kdc->kdc_dumpkey;
996 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
997 out:
998 	explicit_bzero(hash, sizeof(hash));
999 	return (error);
1000 }
1001 
1002 static uint32_t
1003 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
1004 {
1005 
1006 	if (kdc == NULL)
1007 		return (0);
1008 	return (kdc->kdc_dumpkeysize);
1009 }
1010 #endif /* EKCD */
1011 
1012 static struct kerneldumpcomp *
1013 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression)
1014 {
1015 	struct kerneldumpcomp *kdcomp;
1016 	int format;
1017 
1018 	switch (compression) {
1019 	case KERNELDUMP_COMP_GZIP:
1020 		format = COMPRESS_GZIP;
1021 		break;
1022 	case KERNELDUMP_COMP_ZSTD:
1023 		format = COMPRESS_ZSTD;
1024 		break;
1025 	default:
1026 		return (NULL);
1027 	}
1028 
1029 	kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO);
1030 	kdcomp->kdc_format = compression;
1031 	kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb,
1032 	    format, di->maxiosize, kerneldump_gzlevel, di);
1033 	if (kdcomp->kdc_stream == NULL) {
1034 		free(kdcomp, M_DUMPER);
1035 		return (NULL);
1036 	}
1037 	kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1038 	return (kdcomp);
1039 }
1040 
1041 static void
1042 kerneldumpcomp_destroy(struct dumperinfo *di)
1043 {
1044 	struct kerneldumpcomp *kdcomp;
1045 
1046 	kdcomp = di->kdcomp;
1047 	if (kdcomp == NULL)
1048 		return;
1049 	compressor_fini(kdcomp->kdc_stream);
1050 	explicit_bzero(kdcomp->kdc_buf, di->maxiosize);
1051 	free(kdcomp->kdc_buf, M_DUMPER);
1052 	free(kdcomp, M_DUMPER);
1053 }
1054 
1055 /* Registration of dumpers */
1056 int
1057 set_dumper(struct dumperinfo *di, const char *devname, struct thread *td,
1058     uint8_t compression, uint8_t encryption, const uint8_t *key,
1059     uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1060 {
1061 	size_t wantcopy;
1062 	int error;
1063 
1064 	error = priv_check(td, PRIV_SETDUMPER);
1065 	if (error != 0)
1066 		return (error);
1067 
1068 	if (di == NULL) {
1069 		error = 0;
1070 		goto cleanup;
1071 	}
1072 	if (dumper.dumper != NULL)
1073 		return (EBUSY);
1074 	dumper = *di;
1075 	dumper.blockbuf = NULL;
1076 	dumper.kdcrypto = NULL;
1077 	dumper.kdcomp = NULL;
1078 
1079 	if (encryption != KERNELDUMP_ENC_NONE) {
1080 #ifdef EKCD
1081 		dumper.kdcrypto = kerneldumpcrypto_create(di->blocksize,
1082 		    encryption, key, encryptedkeysize, encryptedkey);
1083 		if (dumper.kdcrypto == NULL) {
1084 			error = EINVAL;
1085 			goto cleanup;
1086 		}
1087 #else
1088 		error = EOPNOTSUPP;
1089 		goto cleanup;
1090 #endif
1091 	}
1092 
1093 	wantcopy = strlcpy(dumpdevname, devname, sizeof(dumpdevname));
1094 	if (wantcopy >= sizeof(dumpdevname)) {
1095 		printf("set_dumper: device name truncated from '%s' -> '%s'\n",
1096 		    devname, dumpdevname);
1097 	}
1098 
1099 	if (compression != KERNELDUMP_COMP_NONE) {
1100 		/*
1101 		 * We currently can't support simultaneous encryption and
1102 		 * compression.
1103 		 */
1104 		if (encryption != KERNELDUMP_ENC_NONE) {
1105 			error = EOPNOTSUPP;
1106 			goto cleanup;
1107 		}
1108 		dumper.kdcomp = kerneldumpcomp_create(&dumper, compression);
1109 		if (dumper.kdcomp == NULL) {
1110 			error = EINVAL;
1111 			goto cleanup;
1112 		}
1113 	}
1114 
1115 	dumper.blockbuf = malloc(di->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1116 	return (0);
1117 cleanup:
1118 #ifdef EKCD
1119 	if (dumper.kdcrypto != NULL) {
1120 		explicit_bzero(dumper.kdcrypto, sizeof(*dumper.kdcrypto) +
1121 		    dumper.kdcrypto->kdc_dumpkeysize);
1122 		free(dumper.kdcrypto, M_EKCD);
1123 	}
1124 #endif
1125 
1126 	kerneldumpcomp_destroy(&dumper);
1127 
1128 	if (dumper.blockbuf != NULL) {
1129 		explicit_bzero(dumper.blockbuf, dumper.blocksize);
1130 		free(dumper.blockbuf, M_DUMPER);
1131 	}
1132 	explicit_bzero(&dumper, sizeof(dumper));
1133 	dumpdevname[0] = '\0';
1134 	return (error);
1135 }
1136 
1137 static int
1138 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1139 {
1140 
1141 	if (length != 0 && (offset < di->mediaoffset ||
1142 	    offset - di->mediaoffset + length > di->mediasize)) {
1143 		if (di->kdcomp != NULL && offset >= di->mediaoffset) {
1144 			printf(
1145 		    "Compressed dump failed to fit in device boundaries.\n");
1146 			return (E2BIG);
1147 		}
1148 
1149 		printf("Attempt to write outside dump device boundaries.\n"
1150 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1151 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
1152 		    (uintmax_t)length, (intmax_t)di->mediasize);
1153 		return (ENOSPC);
1154 	}
1155 	if (length % di->blocksize != 0) {
1156 		printf("Attempt to write partial block of length %ju.\n",
1157 		    (uintmax_t)length);
1158 		return (EINVAL);
1159 	}
1160 	if (offset % di->blocksize != 0) {
1161 		printf("Attempt to write at unaligned offset %jd.\n",
1162 		    (intmax_t)offset);
1163 		return (EINVAL);
1164 	}
1165 
1166 	return (0);
1167 }
1168 
1169 #ifdef EKCD
1170 static int
1171 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1172 {
1173 
1174 	switch (kdc->kdc_encryption) {
1175 	case KERNELDUMP_ENC_AES_256_CBC:
1176 		if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1177 		    8 * size, buf) <= 0) {
1178 			return (EIO);
1179 		}
1180 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1181 		    buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1182 			return (EIO);
1183 		}
1184 		break;
1185 	default:
1186 		return (EINVAL);
1187 	}
1188 
1189 	return (0);
1190 }
1191 
1192 /* Encrypt data and call dumper. */
1193 static int
1194 dump_encrypted_write(struct dumperinfo *di, void *virtual,
1195     vm_offset_t physical, off_t offset, size_t length)
1196 {
1197 	static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1198 	struct kerneldumpcrypto *kdc;
1199 	int error;
1200 	size_t nbytes;
1201 
1202 	kdc = di->kdcrypto;
1203 
1204 	while (length > 0) {
1205 		nbytes = MIN(length, sizeof(buf));
1206 		bcopy(virtual, buf, nbytes);
1207 
1208 		if (dump_encrypt(kdc, buf, nbytes) != 0)
1209 			return (EIO);
1210 
1211 		error = dump_write(di, buf, physical, offset, nbytes);
1212 		if (error != 0)
1213 			return (error);
1214 
1215 		offset += nbytes;
1216 		virtual = (void *)((uint8_t *)virtual + nbytes);
1217 		length -= nbytes;
1218 	}
1219 
1220 	return (0);
1221 }
1222 
1223 static int
1224 dump_write_key(struct dumperinfo *di, off_t offset)
1225 {
1226 	struct kerneldumpcrypto *kdc;
1227 
1228 	kdc = di->kdcrypto;
1229 	if (kdc == NULL)
1230 		return (0);
1231 	return (dump_write(di, kdc->kdc_dumpkey, 0, offset,
1232 	    kdc->kdc_dumpkeysize));
1233 }
1234 #endif /* EKCD */
1235 
1236 static int
1237 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1238 {
1239 	struct dumperinfo *di;
1240 	size_t resid, rlength;
1241 	int error;
1242 
1243 	di = arg;
1244 
1245 	if (length % di->blocksize != 0) {
1246 		/*
1247 		 * This must be the final write after flushing the compression
1248 		 * stream. Write as many full blocks as possible and stash the
1249 		 * residual data in the dumper's block buffer. It will be
1250 		 * padded and written in dump_finish().
1251 		 */
1252 		rlength = rounddown(length, di->blocksize);
1253 		if (rlength != 0) {
1254 			error = _dump_append(di, base, 0, rlength);
1255 			if (error != 0)
1256 				return (error);
1257 		}
1258 		resid = length - rlength;
1259 		memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1260 		di->kdcomp->kdc_resid = resid;
1261 		return (EAGAIN);
1262 	}
1263 	return (_dump_append(di, base, 0, length));
1264 }
1265 
1266 /*
1267  * Write a kerneldumpheader at the specified offset. The header structure is 512
1268  * bytes in size, but we must pad to the device sector size.
1269  */
1270 static int
1271 dump_write_header(struct dumperinfo *di, struct kerneldumpheader *kdh,
1272     off_t offset)
1273 {
1274 	void *buf;
1275 	size_t hdrsz;
1276 
1277 	hdrsz = sizeof(*kdh);
1278 	if (hdrsz > di->blocksize)
1279 		return (ENOMEM);
1280 
1281 	if (hdrsz == di->blocksize)
1282 		buf = kdh;
1283 	else {
1284 		buf = di->blockbuf;
1285 		memset(buf, 0, di->blocksize);
1286 		memcpy(buf, kdh, hdrsz);
1287 	}
1288 
1289 	return (dump_write(di, buf, 0, offset, di->blocksize));
1290 }
1291 
1292 /*
1293  * Don't touch the first SIZEOF_METADATA bytes on the dump device.  This is to
1294  * protect us from metadata and metadata from us.
1295  */
1296 #define	SIZEOF_METADATA		(64 * 1024)
1297 
1298 /*
1299  * Do some preliminary setup for a kernel dump: initialize state for encryption,
1300  * if requested, and make sure that we have enough space on the dump device.
1301  *
1302  * We set things up so that the dump ends before the last sector of the dump
1303  * device, at which the trailing header is written.
1304  *
1305  *     +-----------+------+-----+----------------------------+------+
1306  *     |           | lhdr | key |    ... kernel dump ...     | thdr |
1307  *     +-----------+------+-----+----------------------------+------+
1308  *                   1 blk  opt <------- dump extent --------> 1 blk
1309  *
1310  * Dumps written using dump_append() start at the beginning of the extent.
1311  * Uncompressed dumps will use the entire extent, but compressed dumps typically
1312  * will not. The true length of the dump is recorded in the leading and trailing
1313  * headers once the dump has been completed.
1314  */
1315 int
1316 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1317 {
1318 	uint64_t dumpextent;
1319 	uint32_t keysize;
1320 
1321 #ifdef EKCD
1322 	int error = kerneldumpcrypto_init(di->kdcrypto);
1323 	if (error != 0)
1324 		return (error);
1325 	keysize = kerneldumpcrypto_dumpkeysize(di->kdcrypto);
1326 #else
1327 	keysize = 0;
1328 #endif
1329 
1330 	dumpextent = dtoh64(kdh->dumpextent);
1331 	if (di->mediasize < SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1332 	    keysize) {
1333 		if (di->kdcomp != NULL) {
1334 			/*
1335 			 * We don't yet know how much space the compressed dump
1336 			 * will occupy, so try to use the whole swap partition
1337 			 * (minus the first 64KB) in the hope that the
1338 			 * compressed dump will fit. If that doesn't turn out to
1339 			 * be enough, the bounds checking in dump_write()
1340 			 * will catch us and cause the dump to fail.
1341 			 */
1342 			dumpextent = di->mediasize - SIZEOF_METADATA -
1343 			    2 * di->blocksize - keysize;
1344 			kdh->dumpextent = htod64(dumpextent);
1345 		} else
1346 			return (E2BIG);
1347 	}
1348 
1349 	/* The offset at which to begin writing the dump. */
1350 	di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1351 	    dumpextent;
1352 
1353 	return (0);
1354 }
1355 
1356 static int
1357 _dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1358     size_t length)
1359 {
1360 	int error;
1361 
1362 #ifdef EKCD
1363 	if (di->kdcrypto != NULL)
1364 		error = dump_encrypted_write(di, virtual, physical, di->dumpoff,
1365 		    length);
1366 	else
1367 #endif
1368 		error = dump_write(di, virtual, physical, di->dumpoff, length);
1369 	if (error == 0)
1370 		di->dumpoff += length;
1371 	return (error);
1372 }
1373 
1374 /*
1375  * Write to the dump device starting at dumpoff. When compression is enabled,
1376  * writes to the device will be performed using a callback that gets invoked
1377  * when the compression stream's output buffer is full.
1378  */
1379 int
1380 dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1381     size_t length)
1382 {
1383 	void *buf;
1384 
1385 	if (di->kdcomp != NULL) {
1386 		/* Bounce through a buffer to avoid CRC errors. */
1387 		if (length > di->maxiosize)
1388 			return (EINVAL);
1389 		buf = di->kdcomp->kdc_buf;
1390 		memmove(buf, virtual, length);
1391 		return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1392 	}
1393 	return (_dump_append(di, virtual, physical, length));
1394 }
1395 
1396 /*
1397  * Write to the dump device at the specified offset.
1398  */
1399 int
1400 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1401     off_t offset, size_t length)
1402 {
1403 	int error;
1404 
1405 	error = dump_check_bounds(di, offset, length);
1406 	if (error != 0)
1407 		return (error);
1408 	return (di->dumper(di->priv, virtual, physical, offset, length));
1409 }
1410 
1411 /*
1412  * Perform kernel dump finalization: flush the compression stream, if necessary,
1413  * write the leading and trailing kernel dump headers now that we know the true
1414  * length of the dump, and optionally write the encryption key following the
1415  * leading header.
1416  */
1417 int
1418 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1419 {
1420 	uint64_t extent;
1421 	uint32_t keysize;
1422 	int error;
1423 
1424 	extent = dtoh64(kdh->dumpextent);
1425 
1426 #ifdef EKCD
1427 	keysize = kerneldumpcrypto_dumpkeysize(di->kdcrypto);
1428 #else
1429 	keysize = 0;
1430 #endif
1431 
1432 	if (di->kdcomp != NULL) {
1433 		error = compressor_flush(di->kdcomp->kdc_stream);
1434 		if (error == EAGAIN) {
1435 			/* We have residual data in di->blockbuf. */
1436 			error = dump_write(di, di->blockbuf, 0, di->dumpoff,
1437 			    di->blocksize);
1438 			di->dumpoff += di->kdcomp->kdc_resid;
1439 			di->kdcomp->kdc_resid = 0;
1440 		}
1441 		if (error != 0)
1442 			return (error);
1443 
1444 		/*
1445 		 * We now know the size of the compressed dump, so update the
1446 		 * header accordingly and recompute parity.
1447 		 */
1448 		kdh->dumplength = htod64(di->dumpoff -
1449 		    (di->mediaoffset + di->mediasize - di->blocksize - extent));
1450 		kdh->parity = 0;
1451 		kdh->parity = kerneldump_parity(kdh);
1452 
1453 		compressor_reset(di->kdcomp->kdc_stream);
1454 	}
1455 
1456 	/*
1457 	 * Write kerneldump headers at the beginning and end of the dump extent.
1458 	 * Write the key after the leading header.
1459 	 */
1460 	error = dump_write_header(di, kdh,
1461 	    di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1462 	    keysize);
1463 	if (error != 0)
1464 		return (error);
1465 
1466 #ifdef EKCD
1467 	error = dump_write_key(di,
1468 	    di->mediaoffset + di->mediasize - di->blocksize - extent - keysize);
1469 	if (error != 0)
1470 		return (error);
1471 #endif
1472 
1473 	error = dump_write_header(di, kdh,
1474 	    di->mediaoffset + di->mediasize - di->blocksize);
1475 	if (error != 0)
1476 		return (error);
1477 
1478 	(void)dump_write(di, NULL, 0, 0, 0);
1479 	return (0);
1480 }
1481 
1482 void
1483 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1484     char *magic, uint32_t archver, uint64_t dumplen)
1485 {
1486 	size_t dstsize;
1487 
1488 	bzero(kdh, sizeof(*kdh));
1489 	strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1490 	strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1491 	kdh->version = htod32(KERNELDUMPVERSION);
1492 	kdh->architectureversion = htod32(archver);
1493 	kdh->dumplength = htod64(dumplen);
1494 	kdh->dumpextent = kdh->dumplength;
1495 	kdh->dumptime = htod64(time_second);
1496 #ifdef EKCD
1497 	kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1498 #else
1499 	kdh->dumpkeysize = 0;
1500 #endif
1501 	kdh->blocksize = htod32(di->blocksize);
1502 	strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1503 	dstsize = sizeof(kdh->versionstring);
1504 	if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1505 		kdh->versionstring[dstsize - 2] = '\n';
1506 	if (panicstr != NULL)
1507 		strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1508 	if (di->kdcomp != NULL)
1509 		kdh->compression = di->kdcomp->kdc_format;
1510 	kdh->parity = kerneldump_parity(kdh);
1511 }
1512 
1513 #ifdef DDB
1514 DB_SHOW_COMMAND(panic, db_show_panic)
1515 {
1516 
1517 	if (panicstr == NULL)
1518 		db_printf("panicstr not set\n");
1519 	else
1520 		db_printf("panic: %s\n", panicstr);
1521 }
1522 #endif
1523