xref: /freebsd/sys/kern/kern_shutdown.c (revision 2a2234c0f41da33b8cfc938e46b54a8234b64135)
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_suppress_in_panic = 1;
646 static int kassert_warnings = 0;
647 
648 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW, NULL, "kassert options");
649 
650 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, CTLFLAG_RWTUN,
651     &kassert_warn_only, 0,
652     "KASSERT triggers a panic (1) or just a warning (0)");
653 
654 #ifdef KDB
655 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, CTLFLAG_RWTUN,
656     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
657 #endif
658 
659 #ifdef KTR
660 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, CTLFLAG_RWTUN,
661     &kassert_do_ktr, 0,
662     "KASSERT does a KTR, set this to the KTRMASK you want");
663 #endif
664 
665 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, CTLFLAG_RWTUN,
666     &kassert_do_log, 0, "KASSERT triggers a panic (1) or just a warning (0)");
667 
668 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RWTUN,
669     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
670 
671 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, CTLFLAG_RWTUN,
672     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
673 
674 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, CTLFLAG_RWTUN,
675     &kassert_log_pps_limit, 0, "limit number of log messages per second");
676 
677 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, CTLFLAG_RWTUN,
678     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
679 
680 SYSCTL_INT(_debug_kassert, OID_AUTO, suppress_in_panic, CTLFLAG_RWTUN,
681     &kassert_suppress_in_panic, 0,
682     "KASSERTs will be suppressed while handling a panic");
683 
684 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
685 
686 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
687     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE, NULL, 0,
688     kassert_sysctl_kassert, "I", "set to trigger a test kassert");
689 
690 static int
691 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
692 {
693 	int error, i;
694 
695 	error = sysctl_wire_old_buffer(req, sizeof(int));
696 	if (error == 0) {
697 		i = 0;
698 		error = sysctl_handle_int(oidp, &i, 0, req);
699 	}
700 	if (error != 0 || req->newptr == NULL)
701 		return (error);
702 	KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
703 	return (0);
704 }
705 
706 /*
707  * Called by KASSERT, this decides if we will panic
708  * or if we will log via printf and/or ktr.
709  */
710 void
711 kassert_panic(const char *fmt, ...)
712 {
713 	static char buf[256];
714 	va_list ap;
715 
716 	/* If we already panic'd, don't create a double-fault. */
717 	if (panicstr != NULL && kassert_suppress_in_panic)
718 		return;
719 
720 	va_start(ap, fmt);
721 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
722 	va_end(ap);
723 
724 	/*
725 	 * panic if we're not just warning, or if we've exceeded
726 	 * kassert_log_panic_at warnings.
727 	 */
728 	if (!kassert_warn_only ||
729 	    (kassert_log_panic_at > 0 &&
730 	     kassert_warnings >= kassert_log_panic_at)) {
731 		va_start(ap, fmt);
732 		vpanic(fmt, ap);
733 		/* NORETURN */
734 	}
735 #ifdef KTR
736 	if (kassert_do_ktr)
737 		CTR0(ktr_mask, buf);
738 #endif /* KTR */
739 	/*
740 	 * log if we've not yet met the mute limit.
741 	 */
742 	if (kassert_do_log &&
743 	    (kassert_log_mute_at == 0 ||
744 	     kassert_warnings < kassert_log_mute_at)) {
745 		static  struct timeval lasterr;
746 		static  int curerr;
747 
748 		if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
749 			printf("KASSERT failed: %s\n", buf);
750 			kdb_backtrace();
751 		}
752 	}
753 #ifdef KDB
754 	if (kassert_do_kdb) {
755 		kdb_enter(KDB_WHY_KASSERT, buf);
756 	}
757 #endif
758 	atomic_add_int(&kassert_warnings, 1);
759 }
760 #endif
761 
762 /*
763  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
764  * and then reboots.  If we are called twice, then we avoid trying to sync
765  * the disks as this often leads to recursive panics.
766  */
767 void
768 panic(const char *fmt, ...)
769 {
770 	va_list ap;
771 
772 	va_start(ap, fmt);
773 	vpanic(fmt, ap);
774 }
775 
776 void
777 vpanic(const char *fmt, va_list ap)
778 {
779 #ifdef SMP
780 	cpuset_t other_cpus;
781 #endif
782 	struct thread *td = curthread;
783 	int bootopt, newpanic;
784 	static char buf[256];
785 
786 	spinlock_enter();
787 
788 #ifdef SMP
789 	/*
790 	 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
791 	 * concurrently entering panic.  Only the winner will proceed
792 	 * further.
793 	 */
794 	if (panicstr == NULL && !kdb_active) {
795 		other_cpus = all_cpus;
796 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
797 		stop_cpus_hard(other_cpus);
798 	}
799 #endif
800 
801 	/*
802 	 * Ensure that the scheduler is stopped while panicking, even if panic
803 	 * has been entered from kdb.
804 	 */
805 	td->td_stopsched = 1;
806 
807 	bootopt = RB_AUTOBOOT;
808 	newpanic = 0;
809 	if (panicstr)
810 		bootopt |= RB_NOSYNC;
811 	else {
812 		bootopt |= RB_DUMP;
813 		panicstr = fmt;
814 		newpanic = 1;
815 	}
816 
817 	if (newpanic) {
818 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
819 		panicstr = buf;
820 		cngrab();
821 		printf("panic: %s\n", buf);
822 	} else {
823 		printf("panic: ");
824 		vprintf(fmt, ap);
825 		printf("\n");
826 	}
827 #ifdef SMP
828 	printf("cpuid = %d\n", PCPU_GET(cpuid));
829 #endif
830 	printf("time = %jd\n", (intmax_t )time_second);
831 #ifdef KDB
832 	if (newpanic && trace_on_panic)
833 		kdb_backtrace();
834 	if (debugger_on_panic)
835 		kdb_enter(KDB_WHY_PANIC, "panic");
836 #endif
837 	/*thread_lock(td); */
838 	td->td_flags |= TDF_INPANIC;
839 	/* thread_unlock(td); */
840 	if (!sync_on_panic)
841 		bootopt |= RB_NOSYNC;
842 	if (poweroff_on_panic)
843 		bootopt |= RB_POWEROFF;
844 	if (powercycle_on_panic)
845 		bootopt |= RB_POWERCYCLE;
846 	kern_reboot(bootopt);
847 }
848 
849 /*
850  * Support for poweroff delay.
851  *
852  * Please note that setting this delay too short might power off your machine
853  * before the write cache on your hard disk has been flushed, leading to
854  * soft-updates inconsistencies.
855  */
856 #ifndef POWEROFF_DELAY
857 # define POWEROFF_DELAY 5000
858 #endif
859 static int poweroff_delay = POWEROFF_DELAY;
860 
861 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
862     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
863 
864 static void
865 poweroff_wait(void *junk, int howto)
866 {
867 
868 	if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
869 		return;
870 	DELAY(poweroff_delay * 1000);
871 }
872 
873 /*
874  * Some system processes (e.g. syncer) need to be stopped at appropriate
875  * points in their main loops prior to a system shutdown, so that they
876  * won't interfere with the shutdown process (e.g. by holding a disk buf
877  * to cause sync to fail).  For each of these system processes, register
878  * shutdown_kproc() as a handler for one of shutdown events.
879  */
880 static int kproc_shutdown_wait = 60;
881 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
882     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
883 
884 void
885 kproc_shutdown(void *arg, int howto)
886 {
887 	struct proc *p;
888 	int error;
889 
890 	if (panicstr)
891 		return;
892 
893 	p = (struct proc *)arg;
894 	printf("Waiting (max %d seconds) for system process `%s' to stop... ",
895 	    kproc_shutdown_wait, p->p_comm);
896 	error = kproc_suspend(p, kproc_shutdown_wait * hz);
897 
898 	if (error == EWOULDBLOCK)
899 		printf("timed out\n");
900 	else
901 		printf("done\n");
902 }
903 
904 void
905 kthread_shutdown(void *arg, int howto)
906 {
907 	struct thread *td;
908 	int error;
909 
910 	if (panicstr)
911 		return;
912 
913 	td = (struct thread *)arg;
914 	printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
915 	    kproc_shutdown_wait, td->td_name);
916 	error = kthread_suspend(td, kproc_shutdown_wait * hz);
917 
918 	if (error == EWOULDBLOCK)
919 		printf("timed out\n");
920 	else
921 		printf("done\n");
922 }
923 
924 static char dumpdevname[sizeof(((struct cdev*)NULL)->si_name)];
925 SYSCTL_STRING(_kern_shutdown, OID_AUTO, dumpdevname, CTLFLAG_RD,
926     dumpdevname, 0, "Device for kernel dumps");
927 
928 static int	_dump_append(struct dumperinfo *di, void *virtual,
929 		    vm_offset_t physical, size_t length);
930 
931 #ifdef EKCD
932 static struct kerneldumpcrypto *
933 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
934     const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
935 {
936 	struct kerneldumpcrypto *kdc;
937 	struct kerneldumpkey *kdk;
938 	uint32_t dumpkeysize;
939 
940 	dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
941 	kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
942 
943 	arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
944 
945 	kdc->kdc_encryption = encryption;
946 	switch (kdc->kdc_encryption) {
947 	case KERNELDUMP_ENC_AES_256_CBC:
948 		if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
949 			goto failed;
950 		break;
951 	default:
952 		goto failed;
953 	}
954 
955 	kdc->kdc_dumpkeysize = dumpkeysize;
956 	kdk = kdc->kdc_dumpkey;
957 	kdk->kdk_encryption = kdc->kdc_encryption;
958 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
959 	kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
960 	memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
961 
962 	return (kdc);
963 failed:
964 	explicit_bzero(kdc, sizeof(*kdc) + dumpkeysize);
965 	free(kdc, M_EKCD);
966 	return (NULL);
967 }
968 
969 static int
970 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
971 {
972 	uint8_t hash[SHA256_DIGEST_LENGTH];
973 	SHA256_CTX ctx;
974 	struct kerneldumpkey *kdk;
975 	int error;
976 
977 	error = 0;
978 
979 	if (kdc == NULL)
980 		return (0);
981 
982 	/*
983 	 * When a user enters ddb it can write a crash dump multiple times.
984 	 * Each time it should be encrypted using a different IV.
985 	 */
986 	SHA256_Init(&ctx);
987 	SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
988 	SHA256_Final(hash, &ctx);
989 	bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
990 
991 	switch (kdc->kdc_encryption) {
992 	case KERNELDUMP_ENC_AES_256_CBC:
993 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
994 		    kdc->kdc_iv) <= 0) {
995 			error = EINVAL;
996 			goto out;
997 		}
998 		break;
999 	default:
1000 		error = EINVAL;
1001 		goto out;
1002 	}
1003 
1004 	kdk = kdc->kdc_dumpkey;
1005 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1006 out:
1007 	explicit_bzero(hash, sizeof(hash));
1008 	return (error);
1009 }
1010 
1011 static uint32_t
1012 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
1013 {
1014 
1015 	if (kdc == NULL)
1016 		return (0);
1017 	return (kdc->kdc_dumpkeysize);
1018 }
1019 #endif /* EKCD */
1020 
1021 static struct kerneldumpcomp *
1022 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression)
1023 {
1024 	struct kerneldumpcomp *kdcomp;
1025 	int format;
1026 
1027 	switch (compression) {
1028 	case KERNELDUMP_COMP_GZIP:
1029 		format = COMPRESS_GZIP;
1030 		break;
1031 	case KERNELDUMP_COMP_ZSTD:
1032 		format = COMPRESS_ZSTD;
1033 		break;
1034 	default:
1035 		return (NULL);
1036 	}
1037 
1038 	kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO);
1039 	kdcomp->kdc_format = compression;
1040 	kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb,
1041 	    format, di->maxiosize, kerneldump_gzlevel, di);
1042 	if (kdcomp->kdc_stream == NULL) {
1043 		free(kdcomp, M_DUMPER);
1044 		return (NULL);
1045 	}
1046 	kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1047 	return (kdcomp);
1048 }
1049 
1050 static void
1051 kerneldumpcomp_destroy(struct dumperinfo *di)
1052 {
1053 	struct kerneldumpcomp *kdcomp;
1054 
1055 	kdcomp = di->kdcomp;
1056 	if (kdcomp == NULL)
1057 		return;
1058 	compressor_fini(kdcomp->kdc_stream);
1059 	explicit_bzero(kdcomp->kdc_buf, di->maxiosize);
1060 	free(kdcomp->kdc_buf, M_DUMPER);
1061 	free(kdcomp, M_DUMPER);
1062 }
1063 
1064 /* Registration of dumpers */
1065 int
1066 set_dumper(struct dumperinfo *di, const char *devname, struct thread *td,
1067     uint8_t compression, uint8_t encryption, const uint8_t *key,
1068     uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1069 {
1070 	size_t wantcopy;
1071 	int error;
1072 
1073 	error = priv_check(td, PRIV_SETDUMPER);
1074 	if (error != 0)
1075 		return (error);
1076 
1077 	if (di == NULL) {
1078 		error = 0;
1079 		goto cleanup;
1080 	}
1081 	if (dumper.dumper != NULL)
1082 		return (EBUSY);
1083 	dumper = *di;
1084 	dumper.blockbuf = NULL;
1085 	dumper.kdcrypto = NULL;
1086 	dumper.kdcomp = NULL;
1087 
1088 	if (encryption != KERNELDUMP_ENC_NONE) {
1089 #ifdef EKCD
1090 		dumper.kdcrypto = kerneldumpcrypto_create(di->blocksize,
1091 		    encryption, key, encryptedkeysize, encryptedkey);
1092 		if (dumper.kdcrypto == NULL) {
1093 			error = EINVAL;
1094 			goto cleanup;
1095 		}
1096 #else
1097 		error = EOPNOTSUPP;
1098 		goto cleanup;
1099 #endif
1100 	}
1101 
1102 	wantcopy = strlcpy(dumpdevname, devname, sizeof(dumpdevname));
1103 	if (wantcopy >= sizeof(dumpdevname)) {
1104 		printf("set_dumper: device name truncated from '%s' -> '%s'\n",
1105 		    devname, dumpdevname);
1106 	}
1107 
1108 	if (compression != KERNELDUMP_COMP_NONE) {
1109 		/*
1110 		 * We currently can't support simultaneous encryption and
1111 		 * compression.
1112 		 */
1113 		if (encryption != KERNELDUMP_ENC_NONE) {
1114 			error = EOPNOTSUPP;
1115 			goto cleanup;
1116 		}
1117 		dumper.kdcomp = kerneldumpcomp_create(&dumper, compression);
1118 		if (dumper.kdcomp == NULL) {
1119 			error = EINVAL;
1120 			goto cleanup;
1121 		}
1122 	}
1123 
1124 	dumper.blockbuf = malloc(di->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1125 	return (0);
1126 cleanup:
1127 #ifdef EKCD
1128 	if (dumper.kdcrypto != NULL) {
1129 		explicit_bzero(dumper.kdcrypto, sizeof(*dumper.kdcrypto) +
1130 		    dumper.kdcrypto->kdc_dumpkeysize);
1131 		free(dumper.kdcrypto, M_EKCD);
1132 	}
1133 #endif
1134 
1135 	kerneldumpcomp_destroy(&dumper);
1136 
1137 	if (dumper.blockbuf != NULL) {
1138 		explicit_bzero(dumper.blockbuf, dumper.blocksize);
1139 		free(dumper.blockbuf, M_DUMPER);
1140 	}
1141 	explicit_bzero(&dumper, sizeof(dumper));
1142 	dumpdevname[0] = '\0';
1143 	return (error);
1144 }
1145 
1146 static int
1147 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1148 {
1149 
1150 	if (length != 0 && (offset < di->mediaoffset ||
1151 	    offset - di->mediaoffset + length > di->mediasize)) {
1152 		if (di->kdcomp != NULL && offset >= di->mediaoffset) {
1153 			printf(
1154 		    "Compressed dump failed to fit in device boundaries.\n");
1155 			return (E2BIG);
1156 		}
1157 
1158 		printf("Attempt to write outside dump device boundaries.\n"
1159 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1160 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
1161 		    (uintmax_t)length, (intmax_t)di->mediasize);
1162 		return (ENOSPC);
1163 	}
1164 	if (length % di->blocksize != 0) {
1165 		printf("Attempt to write partial block of length %ju.\n",
1166 		    (uintmax_t)length);
1167 		return (EINVAL);
1168 	}
1169 	if (offset % di->blocksize != 0) {
1170 		printf("Attempt to write at unaligned offset %jd.\n",
1171 		    (intmax_t)offset);
1172 		return (EINVAL);
1173 	}
1174 
1175 	return (0);
1176 }
1177 
1178 #ifdef EKCD
1179 static int
1180 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1181 {
1182 
1183 	switch (kdc->kdc_encryption) {
1184 	case KERNELDUMP_ENC_AES_256_CBC:
1185 		if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1186 		    8 * size, buf) <= 0) {
1187 			return (EIO);
1188 		}
1189 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1190 		    buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1191 			return (EIO);
1192 		}
1193 		break;
1194 	default:
1195 		return (EINVAL);
1196 	}
1197 
1198 	return (0);
1199 }
1200 
1201 /* Encrypt data and call dumper. */
1202 static int
1203 dump_encrypted_write(struct dumperinfo *di, void *virtual,
1204     vm_offset_t physical, off_t offset, size_t length)
1205 {
1206 	static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1207 	struct kerneldumpcrypto *kdc;
1208 	int error;
1209 	size_t nbytes;
1210 
1211 	kdc = di->kdcrypto;
1212 
1213 	while (length > 0) {
1214 		nbytes = MIN(length, sizeof(buf));
1215 		bcopy(virtual, buf, nbytes);
1216 
1217 		if (dump_encrypt(kdc, buf, nbytes) != 0)
1218 			return (EIO);
1219 
1220 		error = dump_write(di, buf, physical, offset, nbytes);
1221 		if (error != 0)
1222 			return (error);
1223 
1224 		offset += nbytes;
1225 		virtual = (void *)((uint8_t *)virtual + nbytes);
1226 		length -= nbytes;
1227 	}
1228 
1229 	return (0);
1230 }
1231 
1232 static int
1233 dump_write_key(struct dumperinfo *di, off_t offset)
1234 {
1235 	struct kerneldumpcrypto *kdc;
1236 
1237 	kdc = di->kdcrypto;
1238 	if (kdc == NULL)
1239 		return (0);
1240 	return (dump_write(di, kdc->kdc_dumpkey, 0, offset,
1241 	    kdc->kdc_dumpkeysize));
1242 }
1243 #endif /* EKCD */
1244 
1245 static int
1246 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1247 {
1248 	struct dumperinfo *di;
1249 	size_t resid, rlength;
1250 	int error;
1251 
1252 	di = arg;
1253 
1254 	if (length % di->blocksize != 0) {
1255 		/*
1256 		 * This must be the final write after flushing the compression
1257 		 * stream. Write as many full blocks as possible and stash the
1258 		 * residual data in the dumper's block buffer. It will be
1259 		 * padded and written in dump_finish().
1260 		 */
1261 		rlength = rounddown(length, di->blocksize);
1262 		if (rlength != 0) {
1263 			error = _dump_append(di, base, 0, rlength);
1264 			if (error != 0)
1265 				return (error);
1266 		}
1267 		resid = length - rlength;
1268 		memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1269 		di->kdcomp->kdc_resid = resid;
1270 		return (EAGAIN);
1271 	}
1272 	return (_dump_append(di, base, 0, length));
1273 }
1274 
1275 /*
1276  * Write a kerneldumpheader at the specified offset. The header structure is 512
1277  * bytes in size, but we must pad to the device sector size.
1278  */
1279 static int
1280 dump_write_header(struct dumperinfo *di, struct kerneldumpheader *kdh,
1281     off_t offset)
1282 {
1283 	void *buf;
1284 	size_t hdrsz;
1285 
1286 	hdrsz = sizeof(*kdh);
1287 	if (hdrsz > di->blocksize)
1288 		return (ENOMEM);
1289 
1290 	if (hdrsz == di->blocksize)
1291 		buf = kdh;
1292 	else {
1293 		buf = di->blockbuf;
1294 		memset(buf, 0, di->blocksize);
1295 		memcpy(buf, kdh, hdrsz);
1296 	}
1297 
1298 	return (dump_write(di, buf, 0, offset, di->blocksize));
1299 }
1300 
1301 /*
1302  * Don't touch the first SIZEOF_METADATA bytes on the dump device.  This is to
1303  * protect us from metadata and metadata from us.
1304  */
1305 #define	SIZEOF_METADATA		(64 * 1024)
1306 
1307 /*
1308  * Do some preliminary setup for a kernel dump: initialize state for encryption,
1309  * if requested, and make sure that we have enough space on the dump device.
1310  *
1311  * We set things up so that the dump ends before the last sector of the dump
1312  * device, at which the trailing header is written.
1313  *
1314  *     +-----------+------+-----+----------------------------+------+
1315  *     |           | lhdr | key |    ... kernel dump ...     | thdr |
1316  *     +-----------+------+-----+----------------------------+------+
1317  *                   1 blk  opt <------- dump extent --------> 1 blk
1318  *
1319  * Dumps written using dump_append() start at the beginning of the extent.
1320  * Uncompressed dumps will use the entire extent, but compressed dumps typically
1321  * will not. The true length of the dump is recorded in the leading and trailing
1322  * headers once the dump has been completed.
1323  */
1324 int
1325 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1326 {
1327 	uint64_t dumpextent;
1328 	uint32_t keysize;
1329 
1330 #ifdef EKCD
1331 	int error = kerneldumpcrypto_init(di->kdcrypto);
1332 	if (error != 0)
1333 		return (error);
1334 	keysize = kerneldumpcrypto_dumpkeysize(di->kdcrypto);
1335 #else
1336 	keysize = 0;
1337 #endif
1338 
1339 	dumpextent = dtoh64(kdh->dumpextent);
1340 	if (di->mediasize < SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1341 	    keysize) {
1342 		if (di->kdcomp != NULL) {
1343 			/*
1344 			 * We don't yet know how much space the compressed dump
1345 			 * will occupy, so try to use the whole swap partition
1346 			 * (minus the first 64KB) in the hope that the
1347 			 * compressed dump will fit. If that doesn't turn out to
1348 			 * be enough, the bounds checking in dump_write()
1349 			 * will catch us and cause the dump to fail.
1350 			 */
1351 			dumpextent = di->mediasize - SIZEOF_METADATA -
1352 			    2 * di->blocksize - keysize;
1353 			kdh->dumpextent = htod64(dumpextent);
1354 		} else
1355 			return (E2BIG);
1356 	}
1357 
1358 	/* The offset at which to begin writing the dump. */
1359 	di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1360 	    dumpextent;
1361 
1362 	return (0);
1363 }
1364 
1365 static int
1366 _dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1367     size_t length)
1368 {
1369 	int error;
1370 
1371 #ifdef EKCD
1372 	if (di->kdcrypto != NULL)
1373 		error = dump_encrypted_write(di, virtual, physical, di->dumpoff,
1374 		    length);
1375 	else
1376 #endif
1377 		error = dump_write(di, virtual, physical, di->dumpoff, length);
1378 	if (error == 0)
1379 		di->dumpoff += length;
1380 	return (error);
1381 }
1382 
1383 /*
1384  * Write to the dump device starting at dumpoff. When compression is enabled,
1385  * writes to the device will be performed using a callback that gets invoked
1386  * when the compression stream's output buffer is full.
1387  */
1388 int
1389 dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1390     size_t length)
1391 {
1392 	void *buf;
1393 
1394 	if (di->kdcomp != NULL) {
1395 		/* Bounce through a buffer to avoid CRC errors. */
1396 		if (length > di->maxiosize)
1397 			return (EINVAL);
1398 		buf = di->kdcomp->kdc_buf;
1399 		memmove(buf, virtual, length);
1400 		return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1401 	}
1402 	return (_dump_append(di, virtual, physical, length));
1403 }
1404 
1405 /*
1406  * Write to the dump device at the specified offset.
1407  */
1408 int
1409 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1410     off_t offset, size_t length)
1411 {
1412 	int error;
1413 
1414 	error = dump_check_bounds(di, offset, length);
1415 	if (error != 0)
1416 		return (error);
1417 	return (di->dumper(di->priv, virtual, physical, offset, length));
1418 }
1419 
1420 /*
1421  * Perform kernel dump finalization: flush the compression stream, if necessary,
1422  * write the leading and trailing kernel dump headers now that we know the true
1423  * length of the dump, and optionally write the encryption key following the
1424  * leading header.
1425  */
1426 int
1427 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1428 {
1429 	uint64_t extent;
1430 	uint32_t keysize;
1431 	int error;
1432 
1433 	extent = dtoh64(kdh->dumpextent);
1434 
1435 #ifdef EKCD
1436 	keysize = kerneldumpcrypto_dumpkeysize(di->kdcrypto);
1437 #else
1438 	keysize = 0;
1439 #endif
1440 
1441 	if (di->kdcomp != NULL) {
1442 		error = compressor_flush(di->kdcomp->kdc_stream);
1443 		if (error == EAGAIN) {
1444 			/* We have residual data in di->blockbuf. */
1445 			error = dump_write(di, di->blockbuf, 0, di->dumpoff,
1446 			    di->blocksize);
1447 			di->dumpoff += di->kdcomp->kdc_resid;
1448 			di->kdcomp->kdc_resid = 0;
1449 		}
1450 		if (error != 0)
1451 			return (error);
1452 
1453 		/*
1454 		 * We now know the size of the compressed dump, so update the
1455 		 * header accordingly and recompute parity.
1456 		 */
1457 		kdh->dumplength = htod64(di->dumpoff -
1458 		    (di->mediaoffset + di->mediasize - di->blocksize - extent));
1459 		kdh->parity = 0;
1460 		kdh->parity = kerneldump_parity(kdh);
1461 
1462 		compressor_reset(di->kdcomp->kdc_stream);
1463 	}
1464 
1465 	/*
1466 	 * Write kerneldump headers at the beginning and end of the dump extent.
1467 	 * Write the key after the leading header.
1468 	 */
1469 	error = dump_write_header(di, kdh,
1470 	    di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1471 	    keysize);
1472 	if (error != 0)
1473 		return (error);
1474 
1475 #ifdef EKCD
1476 	error = dump_write_key(di,
1477 	    di->mediaoffset + di->mediasize - di->blocksize - extent - keysize);
1478 	if (error != 0)
1479 		return (error);
1480 #endif
1481 
1482 	error = dump_write_header(di, kdh,
1483 	    di->mediaoffset + di->mediasize - di->blocksize);
1484 	if (error != 0)
1485 		return (error);
1486 
1487 	(void)dump_write(di, NULL, 0, 0, 0);
1488 	return (0);
1489 }
1490 
1491 void
1492 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1493     char *magic, uint32_t archver, uint64_t dumplen)
1494 {
1495 	size_t dstsize;
1496 
1497 	bzero(kdh, sizeof(*kdh));
1498 	strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1499 	strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1500 	kdh->version = htod32(KERNELDUMPVERSION);
1501 	kdh->architectureversion = htod32(archver);
1502 	kdh->dumplength = htod64(dumplen);
1503 	kdh->dumpextent = kdh->dumplength;
1504 	kdh->dumptime = htod64(time_second);
1505 #ifdef EKCD
1506 	kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1507 #else
1508 	kdh->dumpkeysize = 0;
1509 #endif
1510 	kdh->blocksize = htod32(di->blocksize);
1511 	strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1512 	dstsize = sizeof(kdh->versionstring);
1513 	if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1514 		kdh->versionstring[dstsize - 2] = '\n';
1515 	if (panicstr != NULL)
1516 		strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1517 	if (di->kdcomp != NULL)
1518 		kdh->compression = di->kdcomp->kdc_format;
1519 	kdh->parity = kerneldump_parity(kdh);
1520 }
1521 
1522 #ifdef DDB
1523 DB_SHOW_COMMAND(panic, db_show_panic)
1524 {
1525 
1526 	if (panicstr == NULL)
1527 		db_printf("panicstr not set\n");
1528 	else
1529 		db_printf("panic: %s\n", panicstr);
1530 }
1531 #endif
1532