xref: /freebsd/sys/kern/kern_shutdown.c (revision 6be3386466ab79a84b48429ae66244f21526d3df)
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_printf.h"
47 #include "opt_sched.h"
48 #include "opt_watchdog.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/bio.h>
53 #include <sys/buf.h>
54 #include <sys/conf.h>
55 #include <sys/compressor.h>
56 #include <sys/cons.h>
57 #include <sys/disk.h>
58 #include <sys/eventhandler.h>
59 #include <sys/filedesc.h>
60 #include <sys/jail.h>
61 #include <sys/kdb.h>
62 #include <sys/kernel.h>
63 #include <sys/kerneldump.h>
64 #include <sys/kthread.h>
65 #include <sys/ktr.h>
66 #include <sys/malloc.h>
67 #include <sys/mbuf.h>
68 #include <sys/mount.h>
69 #include <sys/priv.h>
70 #include <sys/proc.h>
71 #include <sys/reboot.h>
72 #include <sys/resourcevar.h>
73 #include <sys/rwlock.h>
74 #include <sys/sbuf.h>
75 #include <sys/sched.h>
76 #include <sys/smp.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysproto.h>
79 #include <sys/taskqueue.h>
80 #include <sys/vnode.h>
81 #include <sys/watchdog.h>
82 
83 #include <crypto/chacha20/chacha.h>
84 #include <crypto/rijndael/rijndael-api-fst.h>
85 #include <crypto/sha2/sha256.h>
86 
87 #include <ddb/ddb.h>
88 
89 #include <machine/cpu.h>
90 #include <machine/dump.h>
91 #include <machine/pcb.h>
92 #include <machine/smp.h>
93 
94 #include <security/mac/mac_framework.h>
95 
96 #include <vm/vm.h>
97 #include <vm/vm_object.h>
98 #include <vm/vm_page.h>
99 #include <vm/vm_pager.h>
100 #include <vm/swap_pager.h>
101 
102 #include <sys/signalvar.h>
103 
104 static MALLOC_DEFINE(M_DUMPER, "dumper", "dumper block buffer");
105 
106 #ifndef PANIC_REBOOT_WAIT_TIME
107 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
108 #endif
109 static int panic_reboot_wait_time = PANIC_REBOOT_WAIT_TIME;
110 SYSCTL_INT(_kern, OID_AUTO, panic_reboot_wait_time, CTLFLAG_RWTUN,
111     &panic_reboot_wait_time, 0,
112     "Seconds to wait before rebooting after a panic");
113 
114 /*
115  * Note that stdarg.h and the ANSI style va_start macro is used for both
116  * ANSI and traditional C compilers.
117  */
118 #include <machine/stdarg.h>
119 
120 #ifdef KDB
121 #ifdef KDB_UNATTENDED
122 int debugger_on_panic = 0;
123 #else
124 int debugger_on_panic = 1;
125 #endif
126 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic,
127     CTLFLAG_RWTUN | CTLFLAG_SECURE,
128     &debugger_on_panic, 0, "Run debugger on kernel panic");
129 
130 static bool debugger_on_recursive_panic = false;
131 SYSCTL_BOOL(_debug, OID_AUTO, debugger_on_recursive_panic,
132     CTLFLAG_RWTUN | CTLFLAG_SECURE,
133     &debugger_on_recursive_panic, 0, "Run debugger on recursive kernel panic");
134 
135 int debugger_on_trap = 0;
136 SYSCTL_INT(_debug, OID_AUTO, debugger_on_trap,
137     CTLFLAG_RWTUN | CTLFLAG_SECURE,
138     &debugger_on_trap, 0, "Run debugger on kernel trap before panic");
139 
140 #ifdef KDB_TRACE
141 static int trace_on_panic = 1;
142 static bool trace_all_panics = true;
143 #else
144 static int trace_on_panic = 0;
145 static bool trace_all_panics = false;
146 #endif
147 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
148     CTLFLAG_RWTUN | CTLFLAG_SECURE,
149     &trace_on_panic, 0, "Print stack trace on kernel panic");
150 SYSCTL_BOOL(_debug, OID_AUTO, trace_all_panics, CTLFLAG_RWTUN,
151     &trace_all_panics, 0, "Print stack traces on secondary kernel panics");
152 #endif /* KDB */
153 
154 static int sync_on_panic = 0;
155 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN,
156 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
157 
158 static bool poweroff_on_panic = 0;
159 SYSCTL_BOOL(_kern, OID_AUTO, poweroff_on_panic, CTLFLAG_RWTUN,
160 	&poweroff_on_panic, 0, "Do a power off instead of a reboot on a panic");
161 
162 static bool powercycle_on_panic = 0;
163 SYSCTL_BOOL(_kern, OID_AUTO, powercycle_on_panic, CTLFLAG_RWTUN,
164 	&powercycle_on_panic, 0, "Do a power cycle instead of a reboot on a panic");
165 
166 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
167     "Shutdown environment");
168 
169 #ifndef DIAGNOSTIC
170 static int show_busybufs;
171 #else
172 static int show_busybufs = 1;
173 #endif
174 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
175     &show_busybufs, 0,
176     "Show busy buffers during shutdown");
177 
178 int suspend_blocked = 0;
179 SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW,
180 	&suspend_blocked, 0, "Block suspend due to a pending shutdown");
181 
182 #ifdef EKCD
183 FEATURE(ekcd, "Encrypted kernel crash dumps support");
184 
185 MALLOC_DEFINE(M_EKCD, "ekcd", "Encrypted kernel crash dumps data");
186 
187 struct kerneldumpcrypto {
188 	uint8_t			kdc_encryption;
189 	uint8_t			kdc_iv[KERNELDUMP_IV_MAX_SIZE];
190 	union {
191 		struct {
192 			keyInstance	aes_ki;
193 			cipherInstance	aes_ci;
194 		} u_aes;
195 		struct chacha_ctx	u_chacha;
196 	} u;
197 #define	kdc_ki	u.u_aes.aes_ki
198 #define	kdc_ci	u.u_aes.aes_ci
199 #define	kdc_chacha	u.u_chacha
200 	uint32_t		kdc_dumpkeysize;
201 	struct kerneldumpkey	kdc_dumpkey[];
202 };
203 #endif
204 
205 struct kerneldumpcomp {
206 	uint8_t			kdc_format;
207 	struct compressor	*kdc_stream;
208 	uint8_t			*kdc_buf;
209 	size_t			kdc_resid;
210 };
211 
212 static struct kerneldumpcomp *kerneldumpcomp_create(struct dumperinfo *di,
213 		    uint8_t compression);
214 static void	kerneldumpcomp_destroy(struct dumperinfo *di);
215 static int	kerneldumpcomp_write_cb(void *base, size_t len, off_t off, void *arg);
216 
217 static int kerneldump_gzlevel = 6;
218 SYSCTL_INT(_kern, OID_AUTO, kerneldump_gzlevel, CTLFLAG_RWTUN,
219     &kerneldump_gzlevel, 0,
220     "Kernel crash dump compression level");
221 
222 /*
223  * Variable panicstr contains argument to first call to panic; used as flag
224  * to indicate that the kernel has already called panic.
225  */
226 const char *panicstr;
227 bool __read_frequently panicked;
228 
229 int __read_mostly dumping;		/* system is dumping */
230 int rebooting;				/* system is rebooting */
231 /*
232  * Used to serialize between sysctl kern.shutdown.dumpdevname and list
233  * modifications via ioctl.
234  */
235 static struct mtx dumpconf_list_lk;
236 MTX_SYSINIT(dumper_configs, &dumpconf_list_lk, "dumper config list", MTX_DEF);
237 
238 /* Our selected dumper(s). */
239 static TAILQ_HEAD(dumpconflist, dumperinfo) dumper_configs =
240     TAILQ_HEAD_INITIALIZER(dumper_configs);
241 
242 /* Context information for dump-debuggers. */
243 static struct pcb dumppcb;		/* Registers. */
244 lwpid_t dumptid;			/* Thread ID. */
245 
246 static struct cdevsw reroot_cdevsw = {
247      .d_version = D_VERSION,
248      .d_name    = "reroot",
249 };
250 
251 static void poweroff_wait(void *, int);
252 static void shutdown_halt(void *junk, int howto);
253 static void shutdown_panic(void *junk, int howto);
254 static void shutdown_reset(void *junk, int howto);
255 static int kern_reroot(void);
256 
257 /* register various local shutdown events */
258 static void
259 shutdown_conf(void *unused)
260 {
261 
262 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
263 	    SHUTDOWN_PRI_FIRST);
264 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
265 	    SHUTDOWN_PRI_LAST + 100);
266 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
267 	    SHUTDOWN_PRI_LAST + 100);
268 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
269 	    SHUTDOWN_PRI_LAST + 200);
270 }
271 
272 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
273 
274 /*
275  * The only reason this exists is to create the /dev/reroot/ directory,
276  * used by reroot code in init(8) as a mountpoint for tmpfs.
277  */
278 static void
279 reroot_conf(void *unused)
280 {
281 	int error;
282 	struct cdev *cdev;
283 
284 	error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev,
285 	    &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot");
286 	if (error != 0) {
287 		printf("%s: failed to create device node, error %d",
288 		    __func__, error);
289 	}
290 }
291 
292 SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL);
293 
294 /*
295  * The system call that results in a reboot.
296  */
297 /* ARGSUSED */
298 int
299 sys_reboot(struct thread *td, struct reboot_args *uap)
300 {
301 	int error;
302 
303 	error = 0;
304 #ifdef MAC
305 	error = mac_system_check_reboot(td->td_ucred, uap->opt);
306 #endif
307 	if (error == 0)
308 		error = priv_check(td, PRIV_REBOOT);
309 	if (error == 0) {
310 		if (uap->opt & RB_REROOT)
311 			error = kern_reroot();
312 		else
313 			kern_reboot(uap->opt);
314 	}
315 	return (error);
316 }
317 
318 static void
319 shutdown_nice_task_fn(void *arg, int pending __unused)
320 {
321 	int howto;
322 
323 	howto = (uintptr_t)arg;
324 	/* Send a signal to init(8) and have it shutdown the world. */
325 	PROC_LOCK(initproc);
326 	if (howto & RB_POWEROFF)
327 		kern_psignal(initproc, SIGUSR2);
328 	else if (howto & RB_POWERCYCLE)
329 		kern_psignal(initproc, SIGWINCH);
330 	else if (howto & RB_HALT)
331 		kern_psignal(initproc, SIGUSR1);
332 	else
333 		kern_psignal(initproc, SIGINT);
334 	PROC_UNLOCK(initproc);
335 }
336 
337 static struct task shutdown_nice_task = TASK_INITIALIZER(0,
338     &shutdown_nice_task_fn, NULL);
339 
340 /*
341  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
342  */
343 void
344 shutdown_nice(int howto)
345 {
346 
347 	if (initproc != NULL && !SCHEDULER_STOPPED()) {
348 		shutdown_nice_task.ta_context = (void *)(uintptr_t)howto;
349 		taskqueue_enqueue(taskqueue_fast, &shutdown_nice_task);
350 	} else {
351 		/*
352 		 * No init(8) running, or scheduler would not allow it
353 		 * to run, so simply reboot.
354 		 */
355 		kern_reboot(howto | RB_NOSYNC);
356 	}
357 }
358 
359 static void
360 print_uptime(void)
361 {
362 	int f;
363 	struct timespec ts;
364 
365 	getnanouptime(&ts);
366 	printf("Uptime: ");
367 	f = 0;
368 	if (ts.tv_sec >= 86400) {
369 		printf("%ldd", (long)ts.tv_sec / 86400);
370 		ts.tv_sec %= 86400;
371 		f = 1;
372 	}
373 	if (f || ts.tv_sec >= 3600) {
374 		printf("%ldh", (long)ts.tv_sec / 3600);
375 		ts.tv_sec %= 3600;
376 		f = 1;
377 	}
378 	if (f || ts.tv_sec >= 60) {
379 		printf("%ldm", (long)ts.tv_sec / 60);
380 		ts.tv_sec %= 60;
381 		f = 1;
382 	}
383 	printf("%lds\n", (long)ts.tv_sec);
384 }
385 
386 int
387 doadump(boolean_t textdump)
388 {
389 	boolean_t coredump;
390 	int error;
391 
392 	error = 0;
393 	if (dumping)
394 		return (EBUSY);
395 	if (TAILQ_EMPTY(&dumper_configs))
396 		return (ENXIO);
397 
398 	savectx(&dumppcb);
399 	dumptid = curthread->td_tid;
400 	dumping++;
401 
402 	coredump = TRUE;
403 #ifdef DDB
404 	if (textdump && textdump_pending) {
405 		coredump = FALSE;
406 		textdump_dumpsys(TAILQ_FIRST(&dumper_configs));
407 	}
408 #endif
409 	if (coredump) {
410 		struct dumperinfo *di;
411 
412 		TAILQ_FOREACH(di, &dumper_configs, di_next) {
413 			error = dumpsys(di);
414 			if (error == 0)
415 				break;
416 		}
417 	}
418 
419 	dumping--;
420 	return (error);
421 }
422 
423 /*
424  * Shutdown the system cleanly to prepare for reboot, halt, or power off.
425  */
426 void
427 kern_reboot(int howto)
428 {
429 	static int once = 0;
430 
431 	/*
432 	 * Normal paths here don't hold Giant, but we can wind up here
433 	 * unexpectedly with it held.  Drop it now so we don't have to
434 	 * drop and pick it up elsewhere. The paths it is locking will
435 	 * never be returned to, and it is preferable to preclude
436 	 * deadlock than to lock against code that won't ever
437 	 * continue.
438 	 */
439 	while (mtx_owned(&Giant))
440 		mtx_unlock(&Giant);
441 
442 #if defined(SMP)
443 	/*
444 	 * Bind us to the first CPU so that all shutdown code runs there.  Some
445 	 * systems don't shutdown properly (i.e., ACPI power off) if we
446 	 * run on another processor.
447 	 */
448 	if (!SCHEDULER_STOPPED()) {
449 		thread_lock(curthread);
450 		sched_bind(curthread, CPU_FIRST());
451 		thread_unlock(curthread);
452 		KASSERT(PCPU_GET(cpuid) == CPU_FIRST(),
453 		    ("boot: not running on cpu 0"));
454 	}
455 #endif
456 	/* We're in the process of rebooting. */
457 	rebooting = 1;
458 
459 	/* We are out of the debugger now. */
460 	kdb_active = 0;
461 
462 	/*
463 	 * Do any callouts that should be done BEFORE syncing the filesystems.
464 	 */
465 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
466 
467 	/*
468 	 * Now sync filesystems
469 	 */
470 	if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
471 		once = 1;
472 		bufshutdown(show_busybufs);
473 	}
474 
475 	print_uptime();
476 
477 	cngrab();
478 
479 	/*
480 	 * Ok, now do things that assume all filesystem activity has
481 	 * been completed.
482 	 */
483 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
484 
485 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
486 		doadump(TRUE);
487 
488 	/* Now that we're going to really halt the system... */
489 	EVENTHANDLER_INVOKE(shutdown_final, howto);
490 
491 	for(;;) ;	/* safety against shutdown_reset not working */
492 	/* NOTREACHED */
493 }
494 
495 /*
496  * The system call that results in changing the rootfs.
497  */
498 static int
499 kern_reroot(void)
500 {
501 	struct vnode *oldrootvnode, *vp;
502 	struct mount *mp, *devmp;
503 	int error;
504 
505 	if (curproc != initproc)
506 		return (EPERM);
507 
508 	/*
509 	 * Mark the filesystem containing currently-running executable
510 	 * (the temporary copy of init(8)) busy.
511 	 */
512 	vp = curproc->p_textvp;
513 	error = vn_lock(vp, LK_SHARED);
514 	if (error != 0)
515 		return (error);
516 	mp = vp->v_mount;
517 	error = vfs_busy(mp, MBF_NOWAIT);
518 	if (error != 0) {
519 		vfs_ref(mp);
520 		VOP_UNLOCK(vp);
521 		error = vfs_busy(mp, 0);
522 		vn_lock(vp, LK_SHARED | LK_RETRY);
523 		vfs_rel(mp);
524 		if (error != 0) {
525 			VOP_UNLOCK(vp);
526 			return (ENOENT);
527 		}
528 		if (VN_IS_DOOMED(vp)) {
529 			VOP_UNLOCK(vp);
530 			vfs_unbusy(mp);
531 			return (ENOENT);
532 		}
533 	}
534 	VOP_UNLOCK(vp);
535 
536 	/*
537 	 * Remove the filesystem containing currently-running executable
538 	 * from the mount list, to prevent it from being unmounted
539 	 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
540 	 *
541 	 * Also preserve /dev - forcibly unmounting it could cause driver
542 	 * reinitialization.
543 	 */
544 
545 	vfs_ref(rootdevmp);
546 	devmp = rootdevmp;
547 	rootdevmp = NULL;
548 
549 	mtx_lock(&mountlist_mtx);
550 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
551 	TAILQ_REMOVE(&mountlist, devmp, mnt_list);
552 	mtx_unlock(&mountlist_mtx);
553 
554 	oldrootvnode = rootvnode;
555 
556 	/*
557 	 * Unmount everything except for the two filesystems preserved above.
558 	 */
559 	vfs_unmountall();
560 
561 	/*
562 	 * Add /dev back; vfs_mountroot() will move it into its new place.
563 	 */
564 	mtx_lock(&mountlist_mtx);
565 	TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
566 	mtx_unlock(&mountlist_mtx);
567 	rootdevmp = devmp;
568 	vfs_rel(rootdevmp);
569 
570 	/*
571 	 * Mount the new rootfs.
572 	 */
573 	vfs_mountroot();
574 
575 	/*
576 	 * Update all references to the old rootvnode.
577 	 */
578 	mountcheckdirs(oldrootvnode, rootvnode);
579 
580 	/*
581 	 * Add the temporary filesystem back and unbusy it.
582 	 */
583 	mtx_lock(&mountlist_mtx);
584 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
585 	mtx_unlock(&mountlist_mtx);
586 	vfs_unbusy(mp);
587 
588 	return (0);
589 }
590 
591 /*
592  * If the shutdown was a clean halt, behave accordingly.
593  */
594 static void
595 shutdown_halt(void *junk, int howto)
596 {
597 
598 	if (howto & RB_HALT) {
599 		printf("\n");
600 		printf("The operating system has halted.\n");
601 		printf("Please press any key to reboot.\n\n");
602 
603 		wdog_kern_pat(WD_TO_NEVER);
604 
605 		switch (cngetc()) {
606 		case -1:		/* No console, just die */
607 			cpu_halt();
608 			/* NOTREACHED */
609 		default:
610 			break;
611 		}
612 	}
613 }
614 
615 /*
616  * Check to see if the system paniced, pause and then reboot
617  * according to the specified delay.
618  */
619 static void
620 shutdown_panic(void *junk, int howto)
621 {
622 	int loop;
623 
624 	if (howto & RB_DUMP) {
625 		if (panic_reboot_wait_time != 0) {
626 			if (panic_reboot_wait_time != -1) {
627 				printf("Automatic reboot in %d seconds - "
628 				       "press a key on the console to abort\n",
629 					panic_reboot_wait_time);
630 				for (loop = panic_reboot_wait_time * 10;
631 				     loop > 0; --loop) {
632 					DELAY(1000 * 100); /* 1/10th second */
633 					/* Did user type a key? */
634 					if (cncheckc() != -1)
635 						break;
636 				}
637 				if (!loop)
638 					return;
639 			}
640 		} else { /* zero time specified - reboot NOW */
641 			return;
642 		}
643 		printf("--> Press a key on the console to reboot,\n");
644 		printf("--> or switch off the system now.\n");
645 		cngetc();
646 	}
647 }
648 
649 /*
650  * Everything done, now reset
651  */
652 static void
653 shutdown_reset(void *junk, int howto)
654 {
655 
656 	printf("Rebooting...\n");
657 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
658 
659 	/*
660 	 * Acquiring smp_ipi_mtx here has a double effect:
661 	 * - it disables interrupts avoiding CPU0 preemption
662 	 *   by fast handlers (thus deadlocking  against other CPUs)
663 	 * - it avoids deadlocks against smp_rendezvous() or, more
664 	 *   generally, threads busy-waiting, with this spinlock held,
665 	 *   and waiting for responses by threads on other CPUs
666 	 *   (ie. smp_tlb_shootdown()).
667 	 *
668 	 * For the !SMP case it just needs to handle the former problem.
669 	 */
670 #ifdef SMP
671 	mtx_lock_spin(&smp_ipi_mtx);
672 #else
673 	spinlock_enter();
674 #endif
675 
676 	cpu_reset();
677 	/* NOTREACHED */ /* assuming reset worked */
678 }
679 
680 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
681 static int kassert_warn_only = 0;
682 #ifdef KDB
683 static int kassert_do_kdb = 0;
684 #endif
685 #ifdef KTR
686 static int kassert_do_ktr = 0;
687 #endif
688 static int kassert_do_log = 1;
689 static int kassert_log_pps_limit = 4;
690 static int kassert_log_mute_at = 0;
691 static int kassert_log_panic_at = 0;
692 static int kassert_suppress_in_panic = 0;
693 static int kassert_warnings = 0;
694 
695 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
696     "kassert options");
697 
698 #ifdef KASSERT_PANIC_OPTIONAL
699 #define KASSERT_RWTUN	CTLFLAG_RWTUN
700 #else
701 #define KASSERT_RWTUN	CTLFLAG_RDTUN
702 #endif
703 
704 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, KASSERT_RWTUN,
705     &kassert_warn_only, 0,
706     "KASSERT triggers a panic (0) or just a warning (1)");
707 
708 #ifdef KDB
709 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, KASSERT_RWTUN,
710     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
711 #endif
712 
713 #ifdef KTR
714 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, KASSERT_RWTUN,
715     &kassert_do_ktr, 0,
716     "KASSERT does a KTR, set this to the KTRMASK you want");
717 #endif
718 
719 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, KASSERT_RWTUN,
720     &kassert_do_log, 0,
721     "If warn_only is enabled, log (1) or do not log (0) assertion violations");
722 
723 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RD | CTLFLAG_STATS,
724     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
725 
726 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, KASSERT_RWTUN,
727     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
728 
729 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, KASSERT_RWTUN,
730     &kassert_log_pps_limit, 0, "limit number of log messages per second");
731 
732 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, KASSERT_RWTUN,
733     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
734 
735 SYSCTL_INT(_debug_kassert, OID_AUTO, suppress_in_panic, KASSERT_RWTUN,
736     &kassert_suppress_in_panic, 0,
737     "KASSERTs will be suppressed while handling a panic");
738 #undef KASSERT_RWTUN
739 
740 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
741 
742 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
743     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_NEEDGIANT, NULL, 0,
744     kassert_sysctl_kassert, "I",
745     "set to trigger a test kassert");
746 
747 static int
748 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
749 {
750 	int error, i;
751 
752 	error = sysctl_wire_old_buffer(req, sizeof(int));
753 	if (error == 0) {
754 		i = 0;
755 		error = sysctl_handle_int(oidp, &i, 0, req);
756 	}
757 	if (error != 0 || req->newptr == NULL)
758 		return (error);
759 	KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
760 	return (0);
761 }
762 
763 #ifdef KASSERT_PANIC_OPTIONAL
764 /*
765  * Called by KASSERT, this decides if we will panic
766  * or if we will log via printf and/or ktr.
767  */
768 void
769 kassert_panic(const char *fmt, ...)
770 {
771 	static char buf[256];
772 	va_list ap;
773 
774 	va_start(ap, fmt);
775 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
776 	va_end(ap);
777 
778 	/*
779 	 * If we are suppressing secondary panics, log the warning but do not
780 	 * re-enter panic/kdb.
781 	 */
782 	if (panicstr != NULL && kassert_suppress_in_panic) {
783 		if (kassert_do_log) {
784 			printf("KASSERT failed: %s\n", buf);
785 #ifdef KDB
786 			if (trace_all_panics && trace_on_panic)
787 				kdb_backtrace();
788 #endif
789 		}
790 		return;
791 	}
792 
793 	/*
794 	 * panic if we're not just warning, or if we've exceeded
795 	 * kassert_log_panic_at warnings.
796 	 */
797 	if (!kassert_warn_only ||
798 	    (kassert_log_panic_at > 0 &&
799 	     kassert_warnings >= kassert_log_panic_at)) {
800 		va_start(ap, fmt);
801 		vpanic(fmt, ap);
802 		/* NORETURN */
803 	}
804 #ifdef KTR
805 	if (kassert_do_ktr)
806 		CTR0(ktr_mask, buf);
807 #endif /* KTR */
808 	/*
809 	 * log if we've not yet met the mute limit.
810 	 */
811 	if (kassert_do_log &&
812 	    (kassert_log_mute_at == 0 ||
813 	     kassert_warnings < kassert_log_mute_at)) {
814 		static  struct timeval lasterr;
815 		static  int curerr;
816 
817 		if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
818 			printf("KASSERT failed: %s\n", buf);
819 			kdb_backtrace();
820 		}
821 	}
822 #ifdef KDB
823 	if (kassert_do_kdb) {
824 		kdb_enter(KDB_WHY_KASSERT, buf);
825 	}
826 #endif
827 	atomic_add_int(&kassert_warnings, 1);
828 }
829 #endif /* KASSERT_PANIC_OPTIONAL */
830 #endif
831 
832 /*
833  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
834  * and then reboots.  If we are called twice, then we avoid trying to sync
835  * the disks as this often leads to recursive panics.
836  */
837 void
838 panic(const char *fmt, ...)
839 {
840 	va_list ap;
841 
842 	va_start(ap, fmt);
843 	vpanic(fmt, ap);
844 }
845 
846 void
847 vpanic(const char *fmt, va_list ap)
848 {
849 #ifdef SMP
850 	cpuset_t other_cpus;
851 #endif
852 	struct thread *td = curthread;
853 	int bootopt, newpanic;
854 	static char buf[256];
855 
856 	spinlock_enter();
857 
858 #ifdef SMP
859 	/*
860 	 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
861 	 * concurrently entering panic.  Only the winner will proceed
862 	 * further.
863 	 */
864 	if (panicstr == NULL && !kdb_active) {
865 		other_cpus = all_cpus;
866 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
867 		stop_cpus_hard(other_cpus);
868 	}
869 #endif
870 
871 	/*
872 	 * Ensure that the scheduler is stopped while panicking, even if panic
873 	 * has been entered from kdb.
874 	 */
875 	td->td_stopsched = 1;
876 
877 	bootopt = RB_AUTOBOOT;
878 	newpanic = 0;
879 	if (panicstr)
880 		bootopt |= RB_NOSYNC;
881 	else {
882 		bootopt |= RB_DUMP;
883 		panicstr = fmt;
884 		panicked = true;
885 		newpanic = 1;
886 	}
887 
888 	if (newpanic) {
889 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
890 		panicstr = buf;
891 		cngrab();
892 		printf("panic: %s\n", buf);
893 	} else {
894 		printf("panic: ");
895 		vprintf(fmt, ap);
896 		printf("\n");
897 	}
898 #ifdef SMP
899 	printf("cpuid = %d\n", PCPU_GET(cpuid));
900 #endif
901 	printf("time = %jd\n", (intmax_t )time_second);
902 #ifdef KDB
903 	if ((newpanic || trace_all_panics) && trace_on_panic)
904 		kdb_backtrace();
905 	if (debugger_on_panic)
906 		kdb_enter(KDB_WHY_PANIC, "panic");
907 	else if (!newpanic && debugger_on_recursive_panic)
908 		kdb_enter(KDB_WHY_PANIC, "re-panic");
909 #endif
910 	/*thread_lock(td); */
911 	td->td_flags |= TDF_INPANIC;
912 	/* thread_unlock(td); */
913 	if (!sync_on_panic)
914 		bootopt |= RB_NOSYNC;
915 	if (poweroff_on_panic)
916 		bootopt |= RB_POWEROFF;
917 	if (powercycle_on_panic)
918 		bootopt |= RB_POWERCYCLE;
919 	kern_reboot(bootopt);
920 }
921 
922 /*
923  * Support for poweroff delay.
924  *
925  * Please note that setting this delay too short might power off your machine
926  * before the write cache on your hard disk has been flushed, leading to
927  * soft-updates inconsistencies.
928  */
929 #ifndef POWEROFF_DELAY
930 # define POWEROFF_DELAY 5000
931 #endif
932 static int poweroff_delay = POWEROFF_DELAY;
933 
934 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
935     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
936 
937 static void
938 poweroff_wait(void *junk, int howto)
939 {
940 
941 	if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
942 		return;
943 	DELAY(poweroff_delay * 1000);
944 }
945 
946 /*
947  * Some system processes (e.g. syncer) need to be stopped at appropriate
948  * points in their main loops prior to a system shutdown, so that they
949  * won't interfere with the shutdown process (e.g. by holding a disk buf
950  * to cause sync to fail).  For each of these system processes, register
951  * shutdown_kproc() as a handler for one of shutdown events.
952  */
953 static int kproc_shutdown_wait = 60;
954 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
955     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
956 
957 void
958 kproc_shutdown(void *arg, int howto)
959 {
960 	struct proc *p;
961 	int error;
962 
963 	if (panicstr)
964 		return;
965 
966 	p = (struct proc *)arg;
967 	printf("Waiting (max %d seconds) for system process `%s' to stop... ",
968 	    kproc_shutdown_wait, p->p_comm);
969 	error = kproc_suspend(p, kproc_shutdown_wait * hz);
970 
971 	if (error == EWOULDBLOCK)
972 		printf("timed out\n");
973 	else
974 		printf("done\n");
975 }
976 
977 void
978 kthread_shutdown(void *arg, int howto)
979 {
980 	struct thread *td;
981 	int error;
982 
983 	if (panicstr)
984 		return;
985 
986 	td = (struct thread *)arg;
987 	printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
988 	    kproc_shutdown_wait, td->td_name);
989 	error = kthread_suspend(td, kproc_shutdown_wait * hz);
990 
991 	if (error == EWOULDBLOCK)
992 		printf("timed out\n");
993 	else
994 		printf("done\n");
995 }
996 
997 static int
998 dumpdevname_sysctl_handler(SYSCTL_HANDLER_ARGS)
999 {
1000 	char buf[256];
1001 	struct dumperinfo *di;
1002 	struct sbuf sb;
1003 	int error;
1004 
1005 	error = sysctl_wire_old_buffer(req, 0);
1006 	if (error != 0)
1007 		return (error);
1008 
1009 	sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1010 
1011 	mtx_lock(&dumpconf_list_lk);
1012 	TAILQ_FOREACH(di, &dumper_configs, di_next) {
1013 		if (di != TAILQ_FIRST(&dumper_configs))
1014 			sbuf_putc(&sb, ',');
1015 		sbuf_cat(&sb, di->di_devname);
1016 	}
1017 	mtx_unlock(&dumpconf_list_lk);
1018 
1019 	error = sbuf_finish(&sb);
1020 	sbuf_delete(&sb);
1021 	return (error);
1022 }
1023 SYSCTL_PROC(_kern_shutdown, OID_AUTO, dumpdevname,
1024     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, &dumper_configs, 0,
1025     dumpdevname_sysctl_handler, "A",
1026     "Device(s) for kernel dumps");
1027 
1028 static int	_dump_append(struct dumperinfo *di, void *virtual,
1029 		    vm_offset_t physical, size_t length);
1030 
1031 #ifdef EKCD
1032 static struct kerneldumpcrypto *
1033 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
1034     const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1035 {
1036 	struct kerneldumpcrypto *kdc;
1037 	struct kerneldumpkey *kdk;
1038 	uint32_t dumpkeysize;
1039 
1040 	dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
1041 	kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
1042 
1043 	arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
1044 
1045 	kdc->kdc_encryption = encryption;
1046 	switch (kdc->kdc_encryption) {
1047 	case KERNELDUMP_ENC_AES_256_CBC:
1048 		if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
1049 			goto failed;
1050 		break;
1051 	case KERNELDUMP_ENC_CHACHA20:
1052 		chacha_keysetup(&kdc->kdc_chacha, key, 256);
1053 		break;
1054 	default:
1055 		goto failed;
1056 	}
1057 
1058 	kdc->kdc_dumpkeysize = dumpkeysize;
1059 	kdk = kdc->kdc_dumpkey;
1060 	kdk->kdk_encryption = kdc->kdc_encryption;
1061 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1062 	kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
1063 	memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
1064 
1065 	return (kdc);
1066 failed:
1067 	zfree(kdc, M_EKCD);
1068 	return (NULL);
1069 }
1070 
1071 static int
1072 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
1073 {
1074 	uint8_t hash[SHA256_DIGEST_LENGTH];
1075 	SHA256_CTX ctx;
1076 	struct kerneldumpkey *kdk;
1077 	int error;
1078 
1079 	error = 0;
1080 
1081 	if (kdc == NULL)
1082 		return (0);
1083 
1084 	/*
1085 	 * When a user enters ddb it can write a crash dump multiple times.
1086 	 * Each time it should be encrypted using a different IV.
1087 	 */
1088 	SHA256_Init(&ctx);
1089 	SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1090 	SHA256_Final(hash, &ctx);
1091 	bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1092 
1093 	switch (kdc->kdc_encryption) {
1094 	case KERNELDUMP_ENC_AES_256_CBC:
1095 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1096 		    kdc->kdc_iv) <= 0) {
1097 			error = EINVAL;
1098 			goto out;
1099 		}
1100 		break;
1101 	case KERNELDUMP_ENC_CHACHA20:
1102 		chacha_ivsetup(&kdc->kdc_chacha, kdc->kdc_iv, NULL);
1103 		break;
1104 	default:
1105 		error = EINVAL;
1106 		goto out;
1107 	}
1108 
1109 	kdk = kdc->kdc_dumpkey;
1110 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1111 out:
1112 	explicit_bzero(hash, sizeof(hash));
1113 	return (error);
1114 }
1115 
1116 static uint32_t
1117 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
1118 {
1119 
1120 	if (kdc == NULL)
1121 		return (0);
1122 	return (kdc->kdc_dumpkeysize);
1123 }
1124 #endif /* EKCD */
1125 
1126 static struct kerneldumpcomp *
1127 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression)
1128 {
1129 	struct kerneldumpcomp *kdcomp;
1130 	int format;
1131 
1132 	switch (compression) {
1133 	case KERNELDUMP_COMP_GZIP:
1134 		format = COMPRESS_GZIP;
1135 		break;
1136 	case KERNELDUMP_COMP_ZSTD:
1137 		format = COMPRESS_ZSTD;
1138 		break;
1139 	default:
1140 		return (NULL);
1141 	}
1142 
1143 	kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO);
1144 	kdcomp->kdc_format = compression;
1145 	kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb,
1146 	    format, di->maxiosize, kerneldump_gzlevel, di);
1147 	if (kdcomp->kdc_stream == NULL) {
1148 		free(kdcomp, M_DUMPER);
1149 		return (NULL);
1150 	}
1151 	kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1152 	return (kdcomp);
1153 }
1154 
1155 static void
1156 kerneldumpcomp_destroy(struct dumperinfo *di)
1157 {
1158 	struct kerneldumpcomp *kdcomp;
1159 
1160 	kdcomp = di->kdcomp;
1161 	if (kdcomp == NULL)
1162 		return;
1163 	compressor_fini(kdcomp->kdc_stream);
1164 	zfree(kdcomp->kdc_buf, M_DUMPER);
1165 	free(kdcomp, M_DUMPER);
1166 }
1167 
1168 /*
1169  * Must not be present on global list.
1170  */
1171 static void
1172 free_single_dumper(struct dumperinfo *di)
1173 {
1174 
1175 	if (di == NULL)
1176 		return;
1177 
1178 	zfree(di->blockbuf, M_DUMPER);
1179 
1180 	kerneldumpcomp_destroy(di);
1181 
1182 #ifdef EKCD
1183 	zfree(di->kdcrypto, M_EKCD);
1184 #endif
1185 	zfree(di, M_DUMPER);
1186 }
1187 
1188 /* Registration of dumpers */
1189 int
1190 dumper_insert(const struct dumperinfo *di_template, const char *devname,
1191     const struct diocskerneldump_arg *kda)
1192 {
1193 	struct dumperinfo *newdi, *listdi;
1194 	bool inserted;
1195 	uint8_t index;
1196 	int error;
1197 
1198 	index = kda->kda_index;
1199 	MPASS(index != KDA_REMOVE && index != KDA_REMOVE_DEV &&
1200 	    index != KDA_REMOVE_ALL);
1201 
1202 	error = priv_check(curthread, PRIV_SETDUMPER);
1203 	if (error != 0)
1204 		return (error);
1205 
1206 	newdi = malloc(sizeof(*newdi) + strlen(devname) + 1, M_DUMPER, M_WAITOK
1207 	    | M_ZERO);
1208 	memcpy(newdi, di_template, sizeof(*newdi));
1209 	newdi->blockbuf = NULL;
1210 	newdi->kdcrypto = NULL;
1211 	newdi->kdcomp = NULL;
1212 	strcpy(newdi->di_devname, devname);
1213 
1214 	if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
1215 #ifdef EKCD
1216 		newdi->kdcrypto = kerneldumpcrypto_create(di_template->blocksize,
1217 		    kda->kda_encryption, kda->kda_key,
1218 		    kda->kda_encryptedkeysize, kda->kda_encryptedkey);
1219 		if (newdi->kdcrypto == NULL) {
1220 			error = EINVAL;
1221 			goto cleanup;
1222 		}
1223 #else
1224 		error = EOPNOTSUPP;
1225 		goto cleanup;
1226 #endif
1227 	}
1228 	if (kda->kda_compression != KERNELDUMP_COMP_NONE) {
1229 #ifdef EKCD
1230 		/*
1231 		 * We can't support simultaneous unpadded block cipher
1232 		 * encryption and compression because there is no guarantee the
1233 		 * length of the compressed result is exactly a multiple of the
1234 		 * cipher block size.
1235 		 */
1236 		if (kda->kda_encryption == KERNELDUMP_ENC_AES_256_CBC) {
1237 			error = EOPNOTSUPP;
1238 			goto cleanup;
1239 		}
1240 #endif
1241 		newdi->kdcomp = kerneldumpcomp_create(newdi,
1242 		    kda->kda_compression);
1243 		if (newdi->kdcomp == NULL) {
1244 			error = EINVAL;
1245 			goto cleanup;
1246 		}
1247 	}
1248 
1249 	newdi->blockbuf = malloc(newdi->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1250 
1251 	/* Add the new configuration to the queue */
1252 	mtx_lock(&dumpconf_list_lk);
1253 	inserted = false;
1254 	TAILQ_FOREACH(listdi, &dumper_configs, di_next) {
1255 		if (index == 0) {
1256 			TAILQ_INSERT_BEFORE(listdi, newdi, di_next);
1257 			inserted = true;
1258 			break;
1259 		}
1260 		index--;
1261 	}
1262 	if (!inserted)
1263 		TAILQ_INSERT_TAIL(&dumper_configs, newdi, di_next);
1264 	mtx_unlock(&dumpconf_list_lk);
1265 
1266 	return (0);
1267 
1268 cleanup:
1269 	free_single_dumper(newdi);
1270 	return (error);
1271 }
1272 
1273 #ifdef DDB
1274 void
1275 dumper_ddb_insert(struct dumperinfo *newdi)
1276 {
1277 	TAILQ_INSERT_HEAD(&dumper_configs, newdi, di_next);
1278 }
1279 
1280 void
1281 dumper_ddb_remove(struct dumperinfo *di)
1282 {
1283 	TAILQ_REMOVE(&dumper_configs, di, di_next);
1284 }
1285 #endif
1286 
1287 static bool
1288 dumper_config_match(const struct dumperinfo *di, const char *devname,
1289     const struct diocskerneldump_arg *kda)
1290 {
1291 	if (kda->kda_index == KDA_REMOVE_ALL)
1292 		return (true);
1293 
1294 	if (strcmp(di->di_devname, devname) != 0)
1295 		return (false);
1296 
1297 	/*
1298 	 * Allow wildcard removal of configs matching a device on g_dev_orphan.
1299 	 */
1300 	if (kda->kda_index == KDA_REMOVE_DEV)
1301 		return (true);
1302 
1303 	if (di->kdcomp != NULL) {
1304 		if (di->kdcomp->kdc_format != kda->kda_compression)
1305 			return (false);
1306 	} else if (kda->kda_compression != KERNELDUMP_COMP_NONE)
1307 		return (false);
1308 #ifdef EKCD
1309 	if (di->kdcrypto != NULL) {
1310 		if (di->kdcrypto->kdc_encryption != kda->kda_encryption)
1311 			return (false);
1312 		/*
1313 		 * Do we care to verify keys match to delete?  It seems weird
1314 		 * to expect multiple fallback dump configurations on the same
1315 		 * device that only differ in crypto key.
1316 		 */
1317 	} else
1318 #endif
1319 		if (kda->kda_encryption != KERNELDUMP_ENC_NONE)
1320 			return (false);
1321 
1322 	return (true);
1323 }
1324 
1325 int
1326 dumper_remove(const char *devname, const struct diocskerneldump_arg *kda)
1327 {
1328 	struct dumperinfo *di, *sdi;
1329 	bool found;
1330 	int error;
1331 
1332 	error = priv_check(curthread, PRIV_SETDUMPER);
1333 	if (error != 0)
1334 		return (error);
1335 
1336 	/*
1337 	 * Try to find a matching configuration, and kill it.
1338 	 *
1339 	 * NULL 'kda' indicates remove any configuration matching 'devname',
1340 	 * which may remove multiple configurations in atypical configurations.
1341 	 */
1342 	found = false;
1343 	mtx_lock(&dumpconf_list_lk);
1344 	TAILQ_FOREACH_SAFE(di, &dumper_configs, di_next, sdi) {
1345 		if (dumper_config_match(di, devname, kda)) {
1346 			found = true;
1347 			TAILQ_REMOVE(&dumper_configs, di, di_next);
1348 			free_single_dumper(di);
1349 		}
1350 	}
1351 	mtx_unlock(&dumpconf_list_lk);
1352 
1353 	/* Only produce ENOENT if a more targeted match didn't match. */
1354 	if (!found && kda->kda_index == KDA_REMOVE)
1355 		return (ENOENT);
1356 	return (0);
1357 }
1358 
1359 static int
1360 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1361 {
1362 
1363 	if (di->mediasize > 0 && length != 0 && (offset < di->mediaoffset ||
1364 	    offset - di->mediaoffset + length > di->mediasize)) {
1365 		if (di->kdcomp != NULL && offset >= di->mediaoffset) {
1366 			printf(
1367 		    "Compressed dump failed to fit in device boundaries.\n");
1368 			return (E2BIG);
1369 		}
1370 
1371 		printf("Attempt to write outside dump device boundaries.\n"
1372 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1373 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
1374 		    (uintmax_t)length, (intmax_t)di->mediasize);
1375 		return (ENOSPC);
1376 	}
1377 	if (length % di->blocksize != 0) {
1378 		printf("Attempt to write partial block of length %ju.\n",
1379 		    (uintmax_t)length);
1380 		return (EINVAL);
1381 	}
1382 	if (offset % di->blocksize != 0) {
1383 		printf("Attempt to write at unaligned offset %jd.\n",
1384 		    (intmax_t)offset);
1385 		return (EINVAL);
1386 	}
1387 
1388 	return (0);
1389 }
1390 
1391 #ifdef EKCD
1392 static int
1393 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1394 {
1395 
1396 	switch (kdc->kdc_encryption) {
1397 	case KERNELDUMP_ENC_AES_256_CBC:
1398 		if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1399 		    8 * size, buf) <= 0) {
1400 			return (EIO);
1401 		}
1402 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1403 		    buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1404 			return (EIO);
1405 		}
1406 		break;
1407 	case KERNELDUMP_ENC_CHACHA20:
1408 		chacha_encrypt_bytes(&kdc->kdc_chacha, buf, buf, size);
1409 		break;
1410 	default:
1411 		return (EINVAL);
1412 	}
1413 
1414 	return (0);
1415 }
1416 
1417 /* Encrypt data and call dumper. */
1418 static int
1419 dump_encrypted_write(struct dumperinfo *di, void *virtual,
1420     vm_offset_t physical, off_t offset, size_t length)
1421 {
1422 	static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1423 	struct kerneldumpcrypto *kdc;
1424 	int error;
1425 	size_t nbytes;
1426 
1427 	kdc = di->kdcrypto;
1428 
1429 	while (length > 0) {
1430 		nbytes = MIN(length, sizeof(buf));
1431 		bcopy(virtual, buf, nbytes);
1432 
1433 		if (dump_encrypt(kdc, buf, nbytes) != 0)
1434 			return (EIO);
1435 
1436 		error = dump_write(di, buf, physical, offset, nbytes);
1437 		if (error != 0)
1438 			return (error);
1439 
1440 		offset += nbytes;
1441 		virtual = (void *)((uint8_t *)virtual + nbytes);
1442 		length -= nbytes;
1443 	}
1444 
1445 	return (0);
1446 }
1447 #endif /* EKCD */
1448 
1449 static int
1450 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1451 {
1452 	struct dumperinfo *di;
1453 	size_t resid, rlength;
1454 	int error;
1455 
1456 	di = arg;
1457 
1458 	if (length % di->blocksize != 0) {
1459 		/*
1460 		 * This must be the final write after flushing the compression
1461 		 * stream. Write as many full blocks as possible and stash the
1462 		 * residual data in the dumper's block buffer. It will be
1463 		 * padded and written in dump_finish().
1464 		 */
1465 		rlength = rounddown(length, di->blocksize);
1466 		if (rlength != 0) {
1467 			error = _dump_append(di, base, 0, rlength);
1468 			if (error != 0)
1469 				return (error);
1470 		}
1471 		resid = length - rlength;
1472 		memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1473 		bzero((uint8_t *)di->blockbuf + resid, di->blocksize - resid);
1474 		di->kdcomp->kdc_resid = resid;
1475 		return (EAGAIN);
1476 	}
1477 	return (_dump_append(di, base, 0, length));
1478 }
1479 
1480 /*
1481  * Write kernel dump headers at the beginning and end of the dump extent.
1482  * Write the kernel dump encryption key after the leading header if we were
1483  * configured to do so.
1484  */
1485 static int
1486 dump_write_headers(struct dumperinfo *di, struct kerneldumpheader *kdh)
1487 {
1488 #ifdef EKCD
1489 	struct kerneldumpcrypto *kdc;
1490 #endif
1491 	void *buf, *key;
1492 	size_t hdrsz;
1493 	uint64_t extent;
1494 	uint32_t keysize;
1495 	int error;
1496 
1497 	hdrsz = sizeof(*kdh);
1498 	if (hdrsz > di->blocksize)
1499 		return (ENOMEM);
1500 
1501 #ifdef EKCD
1502 	kdc = di->kdcrypto;
1503 	key = kdc->kdc_dumpkey;
1504 	keysize = kerneldumpcrypto_dumpkeysize(kdc);
1505 #else
1506 	key = NULL;
1507 	keysize = 0;
1508 #endif
1509 
1510 	/*
1511 	 * If the dump device has special handling for headers, let it take care
1512 	 * of writing them out.
1513 	 */
1514 	if (di->dumper_hdr != NULL)
1515 		return (di->dumper_hdr(di, kdh, key, keysize));
1516 
1517 	if (hdrsz == di->blocksize)
1518 		buf = kdh;
1519 	else {
1520 		buf = di->blockbuf;
1521 		memset(buf, 0, di->blocksize);
1522 		memcpy(buf, kdh, hdrsz);
1523 	}
1524 
1525 	extent = dtoh64(kdh->dumpextent);
1526 #ifdef EKCD
1527 	if (kdc != NULL) {
1528 		error = dump_write(di, kdc->kdc_dumpkey, 0,
1529 		    di->mediaoffset + di->mediasize - di->blocksize - extent -
1530 		    keysize, keysize);
1531 		if (error != 0)
1532 			return (error);
1533 	}
1534 #endif
1535 
1536 	error = dump_write(di, buf, 0,
1537 	    di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1538 	    keysize, di->blocksize);
1539 	if (error == 0)
1540 		error = dump_write(di, buf, 0, di->mediaoffset + di->mediasize -
1541 		    di->blocksize, di->blocksize);
1542 	return (error);
1543 }
1544 
1545 /*
1546  * Don't touch the first SIZEOF_METADATA bytes on the dump device.  This is to
1547  * protect us from metadata and metadata from us.
1548  */
1549 #define	SIZEOF_METADATA		(64 * 1024)
1550 
1551 /*
1552  * Do some preliminary setup for a kernel dump: initialize state for encryption,
1553  * if requested, and make sure that we have enough space on the dump device.
1554  *
1555  * We set things up so that the dump ends before the last sector of the dump
1556  * device, at which the trailing header is written.
1557  *
1558  *     +-----------+------+-----+----------------------------+------+
1559  *     |           | lhdr | key |    ... kernel dump ...     | thdr |
1560  *     +-----------+------+-----+----------------------------+------+
1561  *                   1 blk  opt <------- dump extent --------> 1 blk
1562  *
1563  * Dumps written using dump_append() start at the beginning of the extent.
1564  * Uncompressed dumps will use the entire extent, but compressed dumps typically
1565  * will not. The true length of the dump is recorded in the leading and trailing
1566  * headers once the dump has been completed.
1567  *
1568  * The dump device may provide a callback, in which case it will initialize
1569  * dumpoff and take care of laying out the headers.
1570  */
1571 int
1572 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1573 {
1574 	uint64_t dumpextent, span;
1575 	uint32_t keysize;
1576 	int error;
1577 
1578 #ifdef EKCD
1579 	error = kerneldumpcrypto_init(di->kdcrypto);
1580 	if (error != 0)
1581 		return (error);
1582 	keysize = kerneldumpcrypto_dumpkeysize(di->kdcrypto);
1583 #else
1584 	error = 0;
1585 	keysize = 0;
1586 #endif
1587 
1588 	if (di->dumper_start != NULL) {
1589 		error = di->dumper_start(di);
1590 	} else {
1591 		dumpextent = dtoh64(kdh->dumpextent);
1592 		span = SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1593 		    keysize;
1594 		if (di->mediasize < span) {
1595 			if (di->kdcomp == NULL)
1596 				return (E2BIG);
1597 
1598 			/*
1599 			 * We don't yet know how much space the compressed dump
1600 			 * will occupy, so try to use the whole swap partition
1601 			 * (minus the first 64KB) in the hope that the
1602 			 * compressed dump will fit. If that doesn't turn out to
1603 			 * be enough, the bounds checking in dump_write()
1604 			 * will catch us and cause the dump to fail.
1605 			 */
1606 			dumpextent = di->mediasize - span + dumpextent;
1607 			kdh->dumpextent = htod64(dumpextent);
1608 		}
1609 
1610 		/*
1611 		 * The offset at which to begin writing the dump.
1612 		 */
1613 		di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1614 		    dumpextent;
1615 	}
1616 	di->origdumpoff = di->dumpoff;
1617 	return (error);
1618 }
1619 
1620 static int
1621 _dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1622     size_t length)
1623 {
1624 	int error;
1625 
1626 #ifdef EKCD
1627 	if (di->kdcrypto != NULL)
1628 		error = dump_encrypted_write(di, virtual, physical, di->dumpoff,
1629 		    length);
1630 	else
1631 #endif
1632 		error = dump_write(di, virtual, physical, di->dumpoff, length);
1633 	if (error == 0)
1634 		di->dumpoff += length;
1635 	return (error);
1636 }
1637 
1638 /*
1639  * Write to the dump device starting at dumpoff. When compression is enabled,
1640  * writes to the device will be performed using a callback that gets invoked
1641  * when the compression stream's output buffer is full.
1642  */
1643 int
1644 dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1645     size_t length)
1646 {
1647 	void *buf;
1648 
1649 	if (di->kdcomp != NULL) {
1650 		/* Bounce through a buffer to avoid CRC errors. */
1651 		if (length > di->maxiosize)
1652 			return (EINVAL);
1653 		buf = di->kdcomp->kdc_buf;
1654 		memmove(buf, virtual, length);
1655 		return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1656 	}
1657 	return (_dump_append(di, virtual, physical, length));
1658 }
1659 
1660 /*
1661  * Write to the dump device at the specified offset.
1662  */
1663 int
1664 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1665     off_t offset, size_t length)
1666 {
1667 	int error;
1668 
1669 	error = dump_check_bounds(di, offset, length);
1670 	if (error != 0)
1671 		return (error);
1672 	return (di->dumper(di->priv, virtual, physical, offset, length));
1673 }
1674 
1675 /*
1676  * Perform kernel dump finalization: flush the compression stream, if necessary,
1677  * write the leading and trailing kernel dump headers now that we know the true
1678  * length of the dump, and optionally write the encryption key following the
1679  * leading header.
1680  */
1681 int
1682 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1683 {
1684 	int error;
1685 
1686 	if (di->kdcomp != NULL) {
1687 		error = compressor_flush(di->kdcomp->kdc_stream);
1688 		if (error == EAGAIN) {
1689 			/* We have residual data in di->blockbuf. */
1690 			error = _dump_append(di, di->blockbuf, 0, di->blocksize);
1691 			if (error == 0)
1692 				/* Compensate for _dump_append()'s adjustment. */
1693 				di->dumpoff -= di->blocksize - di->kdcomp->kdc_resid;
1694 			di->kdcomp->kdc_resid = 0;
1695 		}
1696 		if (error != 0)
1697 			return (error);
1698 
1699 		/*
1700 		 * We now know the size of the compressed dump, so update the
1701 		 * header accordingly and recompute parity.
1702 		 */
1703 		kdh->dumplength = htod64(di->dumpoff - di->origdumpoff);
1704 		kdh->parity = 0;
1705 		kdh->parity = kerneldump_parity(kdh);
1706 
1707 		compressor_reset(di->kdcomp->kdc_stream);
1708 	}
1709 
1710 	error = dump_write_headers(di, kdh);
1711 	if (error != 0)
1712 		return (error);
1713 
1714 	(void)dump_write(di, NULL, 0, 0, 0);
1715 	return (0);
1716 }
1717 
1718 void
1719 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1720     const char *magic, uint32_t archver, uint64_t dumplen)
1721 {
1722 	size_t dstsize;
1723 
1724 	bzero(kdh, sizeof(*kdh));
1725 	strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1726 	strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1727 	kdh->version = htod32(KERNELDUMPVERSION);
1728 	kdh->architectureversion = htod32(archver);
1729 	kdh->dumplength = htod64(dumplen);
1730 	kdh->dumpextent = kdh->dumplength;
1731 	kdh->dumptime = htod64(time_second);
1732 #ifdef EKCD
1733 	kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1734 #else
1735 	kdh->dumpkeysize = 0;
1736 #endif
1737 	kdh->blocksize = htod32(di->blocksize);
1738 	strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1739 	dstsize = sizeof(kdh->versionstring);
1740 	if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1741 		kdh->versionstring[dstsize - 2] = '\n';
1742 	if (panicstr != NULL)
1743 		strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1744 	if (di->kdcomp != NULL)
1745 		kdh->compression = di->kdcomp->kdc_format;
1746 	kdh->parity = kerneldump_parity(kdh);
1747 }
1748 
1749 #ifdef DDB
1750 DB_SHOW_COMMAND(panic, db_show_panic)
1751 {
1752 
1753 	if (panicstr == NULL)
1754 		db_printf("panicstr not set\n");
1755 	else
1756 		db_printf("panic: %s\n", panicstr);
1757 }
1758 #endif
1759