xref: /freebsd/sys/kern/kern_shutdown.c (revision 9c999a259f00b35f0467acd351fea9157ed7e1e4)
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  * kern_reboot(9): Shut down the system cleanly to prepare for reboot, halt, or
425  * power off.
426  */
427 void
428 kern_reboot(int howto)
429 {
430 	static int once = 0;
431 
432 	/*
433 	 * Normal paths here don't hold Giant, but we can wind up here
434 	 * unexpectedly with it held.  Drop it now so we don't have to
435 	 * drop and pick it up elsewhere. The paths it is locking will
436 	 * never be returned to, and it is preferable to preclude
437 	 * deadlock than to lock against code that won't ever
438 	 * continue.
439 	 */
440 	while (mtx_owned(&Giant))
441 		mtx_unlock(&Giant);
442 
443 #if defined(SMP)
444 	/*
445 	 * Bind us to the first CPU so that all shutdown code runs there.  Some
446 	 * systems don't shutdown properly (i.e., ACPI power off) if we
447 	 * run on another processor.
448 	 */
449 	if (!SCHEDULER_STOPPED()) {
450 		thread_lock(curthread);
451 		sched_bind(curthread, CPU_FIRST());
452 		thread_unlock(curthread);
453 		KASSERT(PCPU_GET(cpuid) == CPU_FIRST(),
454 		    ("%s: not running on cpu 0", __func__));
455 	}
456 #endif
457 	/* We're in the process of rebooting. */
458 	rebooting = 1;
459 
460 	/* We are out of the debugger now. */
461 	kdb_active = 0;
462 
463 	/*
464 	 * Do any callouts that should be done BEFORE syncing the filesystems.
465 	 */
466 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
467 
468 	/*
469 	 * Now sync filesystems
470 	 */
471 	if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
472 		once = 1;
473 		bufshutdown(show_busybufs);
474 	}
475 
476 	print_uptime();
477 
478 	cngrab();
479 
480 	/*
481 	 * Ok, now do things that assume all filesystem activity has
482 	 * been completed.
483 	 */
484 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
485 
486 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
487 		doadump(TRUE);
488 
489 	/* Now that we're going to really halt the system... */
490 	EVENTHANDLER_INVOKE(shutdown_final, howto);
491 
492 	for(;;) ;	/* safety against shutdown_reset not working */
493 	/* NOTREACHED */
494 }
495 
496 /*
497  * The system call that results in changing the rootfs.
498  */
499 static int
500 kern_reroot(void)
501 {
502 	struct vnode *oldrootvnode, *vp;
503 	struct mount *mp, *devmp;
504 	int error;
505 
506 	if (curproc != initproc)
507 		return (EPERM);
508 
509 	/*
510 	 * Mark the filesystem containing currently-running executable
511 	 * (the temporary copy of init(8)) busy.
512 	 */
513 	vp = curproc->p_textvp;
514 	error = vn_lock(vp, LK_SHARED);
515 	if (error != 0)
516 		return (error);
517 	mp = vp->v_mount;
518 	error = vfs_busy(mp, MBF_NOWAIT);
519 	if (error != 0) {
520 		vfs_ref(mp);
521 		VOP_UNLOCK(vp);
522 		error = vfs_busy(mp, 0);
523 		vn_lock(vp, LK_SHARED | LK_RETRY);
524 		vfs_rel(mp);
525 		if (error != 0) {
526 			VOP_UNLOCK(vp);
527 			return (ENOENT);
528 		}
529 		if (VN_IS_DOOMED(vp)) {
530 			VOP_UNLOCK(vp);
531 			vfs_unbusy(mp);
532 			return (ENOENT);
533 		}
534 	}
535 	VOP_UNLOCK(vp);
536 
537 	/*
538 	 * Remove the filesystem containing currently-running executable
539 	 * from the mount list, to prevent it from being unmounted
540 	 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
541 	 *
542 	 * Also preserve /dev - forcibly unmounting it could cause driver
543 	 * reinitialization.
544 	 */
545 
546 	vfs_ref(rootdevmp);
547 	devmp = rootdevmp;
548 	rootdevmp = NULL;
549 
550 	mtx_lock(&mountlist_mtx);
551 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
552 	TAILQ_REMOVE(&mountlist, devmp, mnt_list);
553 	mtx_unlock(&mountlist_mtx);
554 
555 	oldrootvnode = rootvnode;
556 
557 	/*
558 	 * Unmount everything except for the two filesystems preserved above.
559 	 */
560 	vfs_unmountall();
561 
562 	/*
563 	 * Add /dev back; vfs_mountroot() will move it into its new place.
564 	 */
565 	mtx_lock(&mountlist_mtx);
566 	TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
567 	mtx_unlock(&mountlist_mtx);
568 	rootdevmp = devmp;
569 	vfs_rel(rootdevmp);
570 
571 	/*
572 	 * Mount the new rootfs.
573 	 */
574 	vfs_mountroot();
575 
576 	/*
577 	 * Update all references to the old rootvnode.
578 	 */
579 	mountcheckdirs(oldrootvnode, rootvnode);
580 
581 	/*
582 	 * Add the temporary filesystem back and unbusy it.
583 	 */
584 	mtx_lock(&mountlist_mtx);
585 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
586 	mtx_unlock(&mountlist_mtx);
587 	vfs_unbusy(mp);
588 
589 	return (0);
590 }
591 
592 /*
593  * If the shutdown was a clean halt, behave accordingly.
594  */
595 static void
596 shutdown_halt(void *junk, int howto)
597 {
598 
599 	if (howto & RB_HALT) {
600 		printf("\n");
601 		printf("The operating system has halted.\n");
602 		printf("Please press any key to reboot.\n\n");
603 
604 		wdog_kern_pat(WD_TO_NEVER);
605 
606 		switch (cngetc()) {
607 		case -1:		/* No console, just die */
608 			cpu_halt();
609 			/* NOTREACHED */
610 		default:
611 			break;
612 		}
613 	}
614 }
615 
616 /*
617  * Check to see if the system paniced, pause and then reboot
618  * according to the specified delay.
619  */
620 static void
621 shutdown_panic(void *junk, int howto)
622 {
623 	int loop;
624 
625 	if (howto & RB_DUMP) {
626 		if (panic_reboot_wait_time != 0) {
627 			if (panic_reboot_wait_time != -1) {
628 				printf("Automatic reboot in %d seconds - "
629 				       "press a key on the console to abort\n",
630 					panic_reboot_wait_time);
631 				for (loop = panic_reboot_wait_time * 10;
632 				     loop > 0; --loop) {
633 					DELAY(1000 * 100); /* 1/10th second */
634 					/* Did user type a key? */
635 					if (cncheckc() != -1)
636 						break;
637 				}
638 				if (!loop)
639 					return;
640 			}
641 		} else { /* zero time specified - reboot NOW */
642 			return;
643 		}
644 		printf("--> Press a key on the console to reboot,\n");
645 		printf("--> or switch off the system now.\n");
646 		cngetc();
647 	}
648 }
649 
650 /*
651  * Everything done, now reset
652  */
653 static void
654 shutdown_reset(void *junk, int howto)
655 {
656 
657 	printf("Rebooting...\n");
658 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
659 
660 	/*
661 	 * Acquiring smp_ipi_mtx here has a double effect:
662 	 * - it disables interrupts avoiding CPU0 preemption
663 	 *   by fast handlers (thus deadlocking  against other CPUs)
664 	 * - it avoids deadlocks against smp_rendezvous() or, more
665 	 *   generally, threads busy-waiting, with this spinlock held,
666 	 *   and waiting for responses by threads on other CPUs
667 	 *   (ie. smp_tlb_shootdown()).
668 	 *
669 	 * For the !SMP case it just needs to handle the former problem.
670 	 */
671 #ifdef SMP
672 	mtx_lock_spin(&smp_ipi_mtx);
673 #else
674 	spinlock_enter();
675 #endif
676 
677 	cpu_reset();
678 	/* NOTREACHED */ /* assuming reset worked */
679 }
680 
681 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
682 static int kassert_warn_only = 0;
683 #ifdef KDB
684 static int kassert_do_kdb = 0;
685 #endif
686 #ifdef KTR
687 static int kassert_do_ktr = 0;
688 #endif
689 static int kassert_do_log = 1;
690 static int kassert_log_pps_limit = 4;
691 static int kassert_log_mute_at = 0;
692 static int kassert_log_panic_at = 0;
693 static int kassert_suppress_in_panic = 0;
694 static int kassert_warnings = 0;
695 
696 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
697     "kassert options");
698 
699 #ifdef KASSERT_PANIC_OPTIONAL
700 #define KASSERT_RWTUN	CTLFLAG_RWTUN
701 #else
702 #define KASSERT_RWTUN	CTLFLAG_RDTUN
703 #endif
704 
705 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, KASSERT_RWTUN,
706     &kassert_warn_only, 0,
707     "KASSERT triggers a panic (0) or just a warning (1)");
708 
709 #ifdef KDB
710 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, KASSERT_RWTUN,
711     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
712 #endif
713 
714 #ifdef KTR
715 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, KASSERT_RWTUN,
716     &kassert_do_ktr, 0,
717     "KASSERT does a KTR, set this to the KTRMASK you want");
718 #endif
719 
720 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, KASSERT_RWTUN,
721     &kassert_do_log, 0,
722     "If warn_only is enabled, log (1) or do not log (0) assertion violations");
723 
724 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RD | CTLFLAG_STATS,
725     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
726 
727 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, KASSERT_RWTUN,
728     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
729 
730 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, KASSERT_RWTUN,
731     &kassert_log_pps_limit, 0, "limit number of log messages per second");
732 
733 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, KASSERT_RWTUN,
734     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
735 
736 SYSCTL_INT(_debug_kassert, OID_AUTO, suppress_in_panic, KASSERT_RWTUN,
737     &kassert_suppress_in_panic, 0,
738     "KASSERTs will be suppressed while handling a panic");
739 #undef KASSERT_RWTUN
740 
741 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
742 
743 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
744     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_MPSAFE, NULL, 0,
745     kassert_sysctl_kassert, "I",
746     "set to trigger a test kassert");
747 
748 static int
749 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
750 {
751 	int error, i;
752 
753 	error = sysctl_wire_old_buffer(req, sizeof(int));
754 	if (error == 0) {
755 		i = 0;
756 		error = sysctl_handle_int(oidp, &i, 0, req);
757 	}
758 	if (error != 0 || req->newptr == NULL)
759 		return (error);
760 	KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
761 	return (0);
762 }
763 
764 #ifdef KASSERT_PANIC_OPTIONAL
765 /*
766  * Called by KASSERT, this decides if we will panic
767  * or if we will log via printf and/or ktr.
768  */
769 void
770 kassert_panic(const char *fmt, ...)
771 {
772 	static char buf[256];
773 	va_list ap;
774 
775 	va_start(ap, fmt);
776 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
777 	va_end(ap);
778 
779 	/*
780 	 * If we are suppressing secondary panics, log the warning but do not
781 	 * re-enter panic/kdb.
782 	 */
783 	if (panicstr != NULL && kassert_suppress_in_panic) {
784 		if (kassert_do_log) {
785 			printf("KASSERT failed: %s\n", buf);
786 #ifdef KDB
787 			if (trace_all_panics && trace_on_panic)
788 				kdb_backtrace();
789 #endif
790 		}
791 		return;
792 	}
793 
794 	/*
795 	 * panic if we're not just warning, or if we've exceeded
796 	 * kassert_log_panic_at warnings.
797 	 */
798 	if (!kassert_warn_only ||
799 	    (kassert_log_panic_at > 0 &&
800 	     kassert_warnings >= kassert_log_panic_at)) {
801 		va_start(ap, fmt);
802 		vpanic(fmt, ap);
803 		/* NORETURN */
804 	}
805 #ifdef KTR
806 	if (kassert_do_ktr)
807 		CTR0(ktr_mask, buf);
808 #endif /* KTR */
809 	/*
810 	 * log if we've not yet met the mute limit.
811 	 */
812 	if (kassert_do_log &&
813 	    (kassert_log_mute_at == 0 ||
814 	     kassert_warnings < kassert_log_mute_at)) {
815 		static  struct timeval lasterr;
816 		static  int curerr;
817 
818 		if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
819 			printf("KASSERT failed: %s\n", buf);
820 			kdb_backtrace();
821 		}
822 	}
823 #ifdef KDB
824 	if (kassert_do_kdb) {
825 		kdb_enter(KDB_WHY_KASSERT, buf);
826 	}
827 #endif
828 	atomic_add_int(&kassert_warnings, 1);
829 }
830 #endif /* KASSERT_PANIC_OPTIONAL */
831 #endif
832 
833 /*
834  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
835  * and then reboots.  If we are called twice, then we avoid trying to sync
836  * the disks as this often leads to recursive panics.
837  */
838 void
839 panic(const char *fmt, ...)
840 {
841 	va_list ap;
842 
843 	va_start(ap, fmt);
844 	vpanic(fmt, ap);
845 }
846 
847 void
848 vpanic(const char *fmt, va_list ap)
849 {
850 #ifdef SMP
851 	cpuset_t other_cpus;
852 #endif
853 	struct thread *td = curthread;
854 	int bootopt, newpanic;
855 	static char buf[256];
856 
857 	spinlock_enter();
858 
859 #ifdef SMP
860 	/*
861 	 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
862 	 * concurrently entering panic.  Only the winner will proceed
863 	 * further.
864 	 */
865 	if (panicstr == NULL && !kdb_active) {
866 		other_cpus = all_cpus;
867 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
868 		stop_cpus_hard(other_cpus);
869 	}
870 #endif
871 
872 	/*
873 	 * Ensure that the scheduler is stopped while panicking, even if panic
874 	 * has been entered from kdb.
875 	 */
876 	td->td_stopsched = 1;
877 
878 	bootopt = RB_AUTOBOOT;
879 	newpanic = 0;
880 	if (panicstr)
881 		bootopt |= RB_NOSYNC;
882 	else {
883 		bootopt |= RB_DUMP;
884 		panicstr = fmt;
885 		panicked = true;
886 		newpanic = 1;
887 	}
888 
889 	if (newpanic) {
890 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
891 		panicstr = buf;
892 		cngrab();
893 		printf("panic: %s\n", buf);
894 	} else {
895 		printf("panic: ");
896 		vprintf(fmt, ap);
897 		printf("\n");
898 	}
899 #ifdef SMP
900 	printf("cpuid = %d\n", PCPU_GET(cpuid));
901 #endif
902 	printf("time = %jd\n", (intmax_t )time_second);
903 #ifdef KDB
904 	if ((newpanic || trace_all_panics) && trace_on_panic)
905 		kdb_backtrace();
906 	if (debugger_on_panic)
907 		kdb_enter(KDB_WHY_PANIC, "panic");
908 	else if (!newpanic && debugger_on_recursive_panic)
909 		kdb_enter(KDB_WHY_PANIC, "re-panic");
910 #endif
911 	/*thread_lock(td); */
912 	td->td_flags |= TDF_INPANIC;
913 	/* thread_unlock(td); */
914 	if (!sync_on_panic)
915 		bootopt |= RB_NOSYNC;
916 	if (poweroff_on_panic)
917 		bootopt |= RB_POWEROFF;
918 	if (powercycle_on_panic)
919 		bootopt |= RB_POWERCYCLE;
920 	kern_reboot(bootopt);
921 }
922 
923 /*
924  * Support for poweroff delay.
925  *
926  * Please note that setting this delay too short might power off your machine
927  * before the write cache on your hard disk has been flushed, leading to
928  * soft-updates inconsistencies.
929  */
930 #ifndef POWEROFF_DELAY
931 # define POWEROFF_DELAY 5000
932 #endif
933 static int poweroff_delay = POWEROFF_DELAY;
934 
935 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
936     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
937 
938 static void
939 poweroff_wait(void *junk, int howto)
940 {
941 
942 	if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
943 		return;
944 	DELAY(poweroff_delay * 1000);
945 }
946 
947 /*
948  * Some system processes (e.g. syncer) need to be stopped at appropriate
949  * points in their main loops prior to a system shutdown, so that they
950  * won't interfere with the shutdown process (e.g. by holding a disk buf
951  * to cause sync to fail).  For each of these system processes, register
952  * shutdown_kproc() as a handler for one of shutdown events.
953  */
954 static int kproc_shutdown_wait = 60;
955 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
956     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
957 
958 void
959 kproc_shutdown(void *arg, int howto)
960 {
961 	struct proc *p;
962 	int error;
963 
964 	if (panicstr)
965 		return;
966 
967 	p = (struct proc *)arg;
968 	printf("Waiting (max %d seconds) for system process `%s' to stop... ",
969 	    kproc_shutdown_wait, p->p_comm);
970 	error = kproc_suspend(p, kproc_shutdown_wait * hz);
971 
972 	if (error == EWOULDBLOCK)
973 		printf("timed out\n");
974 	else
975 		printf("done\n");
976 }
977 
978 void
979 kthread_shutdown(void *arg, int howto)
980 {
981 	struct thread *td;
982 	int error;
983 
984 	if (panicstr)
985 		return;
986 
987 	td = (struct thread *)arg;
988 	printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
989 	    kproc_shutdown_wait, td->td_name);
990 	error = kthread_suspend(td, kproc_shutdown_wait * hz);
991 
992 	if (error == EWOULDBLOCK)
993 		printf("timed out\n");
994 	else
995 		printf("done\n");
996 }
997 
998 static int
999 dumpdevname_sysctl_handler(SYSCTL_HANDLER_ARGS)
1000 {
1001 	char buf[256];
1002 	struct dumperinfo *di;
1003 	struct sbuf sb;
1004 	int error;
1005 
1006 	error = sysctl_wire_old_buffer(req, 0);
1007 	if (error != 0)
1008 		return (error);
1009 
1010 	sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1011 
1012 	mtx_lock(&dumpconf_list_lk);
1013 	TAILQ_FOREACH(di, &dumper_configs, di_next) {
1014 		if (di != TAILQ_FIRST(&dumper_configs))
1015 			sbuf_putc(&sb, ',');
1016 		sbuf_cat(&sb, di->di_devname);
1017 	}
1018 	mtx_unlock(&dumpconf_list_lk);
1019 
1020 	error = sbuf_finish(&sb);
1021 	sbuf_delete(&sb);
1022 	return (error);
1023 }
1024 SYSCTL_PROC(_kern_shutdown, OID_AUTO, dumpdevname,
1025     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, &dumper_configs, 0,
1026     dumpdevname_sysctl_handler, "A",
1027     "Device(s) for kernel dumps");
1028 
1029 static int	_dump_append(struct dumperinfo *di, void *virtual,
1030 		    vm_offset_t physical, size_t length);
1031 
1032 #ifdef EKCD
1033 static struct kerneldumpcrypto *
1034 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
1035     const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1036 {
1037 	struct kerneldumpcrypto *kdc;
1038 	struct kerneldumpkey *kdk;
1039 	uint32_t dumpkeysize;
1040 
1041 	dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
1042 	kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
1043 
1044 	arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
1045 
1046 	kdc->kdc_encryption = encryption;
1047 	switch (kdc->kdc_encryption) {
1048 	case KERNELDUMP_ENC_AES_256_CBC:
1049 		if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
1050 			goto failed;
1051 		break;
1052 	case KERNELDUMP_ENC_CHACHA20:
1053 		chacha_keysetup(&kdc->kdc_chacha, key, 256);
1054 		break;
1055 	default:
1056 		goto failed;
1057 	}
1058 
1059 	kdc->kdc_dumpkeysize = dumpkeysize;
1060 	kdk = kdc->kdc_dumpkey;
1061 	kdk->kdk_encryption = kdc->kdc_encryption;
1062 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1063 	kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
1064 	memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
1065 
1066 	return (kdc);
1067 failed:
1068 	zfree(kdc, M_EKCD);
1069 	return (NULL);
1070 }
1071 
1072 static int
1073 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
1074 {
1075 	uint8_t hash[SHA256_DIGEST_LENGTH];
1076 	SHA256_CTX ctx;
1077 	struct kerneldumpkey *kdk;
1078 	int error;
1079 
1080 	error = 0;
1081 
1082 	if (kdc == NULL)
1083 		return (0);
1084 
1085 	/*
1086 	 * When a user enters ddb it can write a crash dump multiple times.
1087 	 * Each time it should be encrypted using a different IV.
1088 	 */
1089 	SHA256_Init(&ctx);
1090 	SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1091 	SHA256_Final(hash, &ctx);
1092 	bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1093 
1094 	switch (kdc->kdc_encryption) {
1095 	case KERNELDUMP_ENC_AES_256_CBC:
1096 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1097 		    kdc->kdc_iv) <= 0) {
1098 			error = EINVAL;
1099 			goto out;
1100 		}
1101 		break;
1102 	case KERNELDUMP_ENC_CHACHA20:
1103 		chacha_ivsetup(&kdc->kdc_chacha, kdc->kdc_iv, NULL);
1104 		break;
1105 	default:
1106 		error = EINVAL;
1107 		goto out;
1108 	}
1109 
1110 	kdk = kdc->kdc_dumpkey;
1111 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1112 out:
1113 	explicit_bzero(hash, sizeof(hash));
1114 	return (error);
1115 }
1116 
1117 static uint32_t
1118 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
1119 {
1120 
1121 	if (kdc == NULL)
1122 		return (0);
1123 	return (kdc->kdc_dumpkeysize);
1124 }
1125 #endif /* EKCD */
1126 
1127 static struct kerneldumpcomp *
1128 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression)
1129 {
1130 	struct kerneldumpcomp *kdcomp;
1131 	int format;
1132 
1133 	switch (compression) {
1134 	case KERNELDUMP_COMP_GZIP:
1135 		format = COMPRESS_GZIP;
1136 		break;
1137 	case KERNELDUMP_COMP_ZSTD:
1138 		format = COMPRESS_ZSTD;
1139 		break;
1140 	default:
1141 		return (NULL);
1142 	}
1143 
1144 	kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO);
1145 	kdcomp->kdc_format = compression;
1146 	kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb,
1147 	    format, di->maxiosize, kerneldump_gzlevel, di);
1148 	if (kdcomp->kdc_stream == NULL) {
1149 		free(kdcomp, M_DUMPER);
1150 		return (NULL);
1151 	}
1152 	kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1153 	return (kdcomp);
1154 }
1155 
1156 static void
1157 kerneldumpcomp_destroy(struct dumperinfo *di)
1158 {
1159 	struct kerneldumpcomp *kdcomp;
1160 
1161 	kdcomp = di->kdcomp;
1162 	if (kdcomp == NULL)
1163 		return;
1164 	compressor_fini(kdcomp->kdc_stream);
1165 	zfree(kdcomp->kdc_buf, M_DUMPER);
1166 	free(kdcomp, M_DUMPER);
1167 }
1168 
1169 /*
1170  * Must not be present on global list.
1171  */
1172 static void
1173 free_single_dumper(struct dumperinfo *di)
1174 {
1175 
1176 	if (di == NULL)
1177 		return;
1178 
1179 	zfree(di->blockbuf, M_DUMPER);
1180 
1181 	kerneldumpcomp_destroy(di);
1182 
1183 #ifdef EKCD
1184 	zfree(di->kdcrypto, M_EKCD);
1185 #endif
1186 	zfree(di, M_DUMPER);
1187 }
1188 
1189 /* Registration of dumpers */
1190 int
1191 dumper_insert(const struct dumperinfo *di_template, const char *devname,
1192     const struct diocskerneldump_arg *kda)
1193 {
1194 	struct dumperinfo *newdi, *listdi;
1195 	bool inserted;
1196 	uint8_t index;
1197 	int error;
1198 
1199 	index = kda->kda_index;
1200 	MPASS(index != KDA_REMOVE && index != KDA_REMOVE_DEV &&
1201 	    index != KDA_REMOVE_ALL);
1202 
1203 	error = priv_check(curthread, PRIV_SETDUMPER);
1204 	if (error != 0)
1205 		return (error);
1206 
1207 	newdi = malloc(sizeof(*newdi) + strlen(devname) + 1, M_DUMPER, M_WAITOK
1208 	    | M_ZERO);
1209 	memcpy(newdi, di_template, sizeof(*newdi));
1210 	newdi->blockbuf = NULL;
1211 	newdi->kdcrypto = NULL;
1212 	newdi->kdcomp = NULL;
1213 	strcpy(newdi->di_devname, devname);
1214 
1215 	if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
1216 #ifdef EKCD
1217 		newdi->kdcrypto = kerneldumpcrypto_create(di_template->blocksize,
1218 		    kda->kda_encryption, kda->kda_key,
1219 		    kda->kda_encryptedkeysize, kda->kda_encryptedkey);
1220 		if (newdi->kdcrypto == NULL) {
1221 			error = EINVAL;
1222 			goto cleanup;
1223 		}
1224 #else
1225 		error = EOPNOTSUPP;
1226 		goto cleanup;
1227 #endif
1228 	}
1229 	if (kda->kda_compression != KERNELDUMP_COMP_NONE) {
1230 #ifdef EKCD
1231 		/*
1232 		 * We can't support simultaneous unpadded block cipher
1233 		 * encryption and compression because there is no guarantee the
1234 		 * length of the compressed result is exactly a multiple of the
1235 		 * cipher block size.
1236 		 */
1237 		if (kda->kda_encryption == KERNELDUMP_ENC_AES_256_CBC) {
1238 			error = EOPNOTSUPP;
1239 			goto cleanup;
1240 		}
1241 #endif
1242 		newdi->kdcomp = kerneldumpcomp_create(newdi,
1243 		    kda->kda_compression);
1244 		if (newdi->kdcomp == NULL) {
1245 			error = EINVAL;
1246 			goto cleanup;
1247 		}
1248 	}
1249 
1250 	newdi->blockbuf = malloc(newdi->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1251 
1252 	/* Add the new configuration to the queue */
1253 	mtx_lock(&dumpconf_list_lk);
1254 	inserted = false;
1255 	TAILQ_FOREACH(listdi, &dumper_configs, di_next) {
1256 		if (index == 0) {
1257 			TAILQ_INSERT_BEFORE(listdi, newdi, di_next);
1258 			inserted = true;
1259 			break;
1260 		}
1261 		index--;
1262 	}
1263 	if (!inserted)
1264 		TAILQ_INSERT_TAIL(&dumper_configs, newdi, di_next);
1265 	mtx_unlock(&dumpconf_list_lk);
1266 
1267 	return (0);
1268 
1269 cleanup:
1270 	free_single_dumper(newdi);
1271 	return (error);
1272 }
1273 
1274 #ifdef DDB
1275 void
1276 dumper_ddb_insert(struct dumperinfo *newdi)
1277 {
1278 	TAILQ_INSERT_HEAD(&dumper_configs, newdi, di_next);
1279 }
1280 
1281 void
1282 dumper_ddb_remove(struct dumperinfo *di)
1283 {
1284 	TAILQ_REMOVE(&dumper_configs, di, di_next);
1285 }
1286 #endif
1287 
1288 static bool
1289 dumper_config_match(const struct dumperinfo *di, const char *devname,
1290     const struct diocskerneldump_arg *kda)
1291 {
1292 	if (kda->kda_index == KDA_REMOVE_ALL)
1293 		return (true);
1294 
1295 	if (strcmp(di->di_devname, devname) != 0)
1296 		return (false);
1297 
1298 	/*
1299 	 * Allow wildcard removal of configs matching a device on g_dev_orphan.
1300 	 */
1301 	if (kda->kda_index == KDA_REMOVE_DEV)
1302 		return (true);
1303 
1304 	if (di->kdcomp != NULL) {
1305 		if (di->kdcomp->kdc_format != kda->kda_compression)
1306 			return (false);
1307 	} else if (kda->kda_compression != KERNELDUMP_COMP_NONE)
1308 		return (false);
1309 #ifdef EKCD
1310 	if (di->kdcrypto != NULL) {
1311 		if (di->kdcrypto->kdc_encryption != kda->kda_encryption)
1312 			return (false);
1313 		/*
1314 		 * Do we care to verify keys match to delete?  It seems weird
1315 		 * to expect multiple fallback dump configurations on the same
1316 		 * device that only differ in crypto key.
1317 		 */
1318 	} else
1319 #endif
1320 		if (kda->kda_encryption != KERNELDUMP_ENC_NONE)
1321 			return (false);
1322 
1323 	return (true);
1324 }
1325 
1326 int
1327 dumper_remove(const char *devname, const struct diocskerneldump_arg *kda)
1328 {
1329 	struct dumperinfo *di, *sdi;
1330 	bool found;
1331 	int error;
1332 
1333 	error = priv_check(curthread, PRIV_SETDUMPER);
1334 	if (error != 0)
1335 		return (error);
1336 
1337 	/*
1338 	 * Try to find a matching configuration, and kill it.
1339 	 *
1340 	 * NULL 'kda' indicates remove any configuration matching 'devname',
1341 	 * which may remove multiple configurations in atypical configurations.
1342 	 */
1343 	found = false;
1344 	mtx_lock(&dumpconf_list_lk);
1345 	TAILQ_FOREACH_SAFE(di, &dumper_configs, di_next, sdi) {
1346 		if (dumper_config_match(di, devname, kda)) {
1347 			found = true;
1348 			TAILQ_REMOVE(&dumper_configs, di, di_next);
1349 			free_single_dumper(di);
1350 		}
1351 	}
1352 	mtx_unlock(&dumpconf_list_lk);
1353 
1354 	/* Only produce ENOENT if a more targeted match didn't match. */
1355 	if (!found && kda->kda_index == KDA_REMOVE)
1356 		return (ENOENT);
1357 	return (0);
1358 }
1359 
1360 static int
1361 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1362 {
1363 
1364 	if (di->mediasize > 0 && length != 0 && (offset < di->mediaoffset ||
1365 	    offset - di->mediaoffset + length > di->mediasize)) {
1366 		if (di->kdcomp != NULL && offset >= di->mediaoffset) {
1367 			printf(
1368 		    "Compressed dump failed to fit in device boundaries.\n");
1369 			return (E2BIG);
1370 		}
1371 
1372 		printf("Attempt to write outside dump device boundaries.\n"
1373 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1374 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
1375 		    (uintmax_t)length, (intmax_t)di->mediasize);
1376 		return (ENOSPC);
1377 	}
1378 	if (length % di->blocksize != 0) {
1379 		printf("Attempt to write partial block of length %ju.\n",
1380 		    (uintmax_t)length);
1381 		return (EINVAL);
1382 	}
1383 	if (offset % di->blocksize != 0) {
1384 		printf("Attempt to write at unaligned offset %jd.\n",
1385 		    (intmax_t)offset);
1386 		return (EINVAL);
1387 	}
1388 
1389 	return (0);
1390 }
1391 
1392 #ifdef EKCD
1393 static int
1394 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1395 {
1396 
1397 	switch (kdc->kdc_encryption) {
1398 	case KERNELDUMP_ENC_AES_256_CBC:
1399 		if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1400 		    8 * size, buf) <= 0) {
1401 			return (EIO);
1402 		}
1403 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1404 		    buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1405 			return (EIO);
1406 		}
1407 		break;
1408 	case KERNELDUMP_ENC_CHACHA20:
1409 		chacha_encrypt_bytes(&kdc->kdc_chacha, buf, buf, size);
1410 		break;
1411 	default:
1412 		return (EINVAL);
1413 	}
1414 
1415 	return (0);
1416 }
1417 
1418 /* Encrypt data and call dumper. */
1419 static int
1420 dump_encrypted_write(struct dumperinfo *di, void *virtual,
1421     vm_offset_t physical, off_t offset, size_t length)
1422 {
1423 	static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1424 	struct kerneldumpcrypto *kdc;
1425 	int error;
1426 	size_t nbytes;
1427 
1428 	kdc = di->kdcrypto;
1429 
1430 	while (length > 0) {
1431 		nbytes = MIN(length, sizeof(buf));
1432 		bcopy(virtual, buf, nbytes);
1433 
1434 		if (dump_encrypt(kdc, buf, nbytes) != 0)
1435 			return (EIO);
1436 
1437 		error = dump_write(di, buf, physical, offset, nbytes);
1438 		if (error != 0)
1439 			return (error);
1440 
1441 		offset += nbytes;
1442 		virtual = (void *)((uint8_t *)virtual + nbytes);
1443 		length -= nbytes;
1444 	}
1445 
1446 	return (0);
1447 }
1448 #endif /* EKCD */
1449 
1450 static int
1451 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1452 {
1453 	struct dumperinfo *di;
1454 	size_t resid, rlength;
1455 	int error;
1456 
1457 	di = arg;
1458 
1459 	if (length % di->blocksize != 0) {
1460 		/*
1461 		 * This must be the final write after flushing the compression
1462 		 * stream. Write as many full blocks as possible and stash the
1463 		 * residual data in the dumper's block buffer. It will be
1464 		 * padded and written in dump_finish().
1465 		 */
1466 		rlength = rounddown(length, di->blocksize);
1467 		if (rlength != 0) {
1468 			error = _dump_append(di, base, 0, rlength);
1469 			if (error != 0)
1470 				return (error);
1471 		}
1472 		resid = length - rlength;
1473 		memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1474 		bzero((uint8_t *)di->blockbuf + resid, di->blocksize - resid);
1475 		di->kdcomp->kdc_resid = resid;
1476 		return (EAGAIN);
1477 	}
1478 	return (_dump_append(di, base, 0, length));
1479 }
1480 
1481 /*
1482  * Write kernel dump headers at the beginning and end of the dump extent.
1483  * Write the kernel dump encryption key after the leading header if we were
1484  * configured to do so.
1485  */
1486 static int
1487 dump_write_headers(struct dumperinfo *di, struct kerneldumpheader *kdh)
1488 {
1489 #ifdef EKCD
1490 	struct kerneldumpcrypto *kdc;
1491 #endif
1492 	void *buf;
1493 	size_t hdrsz;
1494 	uint64_t extent;
1495 	uint32_t keysize;
1496 	int error;
1497 
1498 	hdrsz = sizeof(*kdh);
1499 	if (hdrsz > di->blocksize)
1500 		return (ENOMEM);
1501 
1502 #ifdef EKCD
1503 	kdc = di->kdcrypto;
1504 	keysize = kerneldumpcrypto_dumpkeysize(kdc);
1505 #else
1506 	keysize = 0;
1507 #endif
1508 
1509 	/*
1510 	 * If the dump device has special handling for headers, let it take care
1511 	 * of writing them out.
1512 	 */
1513 	if (di->dumper_hdr != NULL)
1514 		return (di->dumper_hdr(di, kdh));
1515 
1516 	if (hdrsz == di->blocksize)
1517 		buf = kdh;
1518 	else {
1519 		buf = di->blockbuf;
1520 		memset(buf, 0, di->blocksize);
1521 		memcpy(buf, kdh, hdrsz);
1522 	}
1523 
1524 	extent = dtoh64(kdh->dumpextent);
1525 #ifdef EKCD
1526 	if (kdc != NULL) {
1527 		error = dump_write(di, kdc->kdc_dumpkey, 0,
1528 		    di->mediaoffset + di->mediasize - di->blocksize - extent -
1529 		    keysize, keysize);
1530 		if (error != 0)
1531 			return (error);
1532 	}
1533 #endif
1534 
1535 	error = dump_write(di, buf, 0,
1536 	    di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1537 	    keysize, di->blocksize);
1538 	if (error == 0)
1539 		error = dump_write(di, buf, 0, di->mediaoffset + di->mediasize -
1540 		    di->blocksize, di->blocksize);
1541 	return (error);
1542 }
1543 
1544 /*
1545  * Don't touch the first SIZEOF_METADATA bytes on the dump device.  This is to
1546  * protect us from metadata and metadata from us.
1547  */
1548 #define	SIZEOF_METADATA		(64 * 1024)
1549 
1550 /*
1551  * Do some preliminary setup for a kernel dump: initialize state for encryption,
1552  * if requested, and make sure that we have enough space on the dump device.
1553  *
1554  * We set things up so that the dump ends before the last sector of the dump
1555  * device, at which the trailing header is written.
1556  *
1557  *     +-----------+------+-----+----------------------------+------+
1558  *     |           | lhdr | key |    ... kernel dump ...     | thdr |
1559  *     +-----------+------+-----+----------------------------+------+
1560  *                   1 blk  opt <------- dump extent --------> 1 blk
1561  *
1562  * Dumps written using dump_append() start at the beginning of the extent.
1563  * Uncompressed dumps will use the entire extent, but compressed dumps typically
1564  * will not. The true length of the dump is recorded in the leading and trailing
1565  * headers once the dump has been completed.
1566  *
1567  * The dump device may provide a callback, in which case it will initialize
1568  * dumpoff and take care of laying out the headers.
1569  */
1570 int
1571 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1572 {
1573 #ifdef EKCD
1574 	struct kerneldumpcrypto *kdc;
1575 #endif
1576 	void *key;
1577 	uint64_t dumpextent, span;
1578 	uint32_t keysize;
1579 	int error;
1580 
1581 #ifdef EKCD
1582 	/* Send the key before the dump so a partial dump is still usable. */
1583 	kdc = di->kdcrypto;
1584 	error = kerneldumpcrypto_init(kdc);
1585 	if (error != 0)
1586 		return (error);
1587 	keysize = kerneldumpcrypto_dumpkeysize(kdc);
1588 	key = keysize > 0 ? kdc->kdc_dumpkey : NULL;
1589 #else
1590 	error = 0;
1591 	keysize = 0;
1592 	key = NULL;
1593 #endif
1594 
1595 	if (di->dumper_start != NULL) {
1596 		error = di->dumper_start(di, key, keysize);
1597 	} else {
1598 		dumpextent = dtoh64(kdh->dumpextent);
1599 		span = SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1600 		    keysize;
1601 		if (di->mediasize < span) {
1602 			if (di->kdcomp == NULL)
1603 				return (E2BIG);
1604 
1605 			/*
1606 			 * We don't yet know how much space the compressed dump
1607 			 * will occupy, so try to use the whole swap partition
1608 			 * (minus the first 64KB) in the hope that the
1609 			 * compressed dump will fit. If that doesn't turn out to
1610 			 * be enough, the bounds checking in dump_write()
1611 			 * will catch us and cause the dump to fail.
1612 			 */
1613 			dumpextent = di->mediasize - span + dumpextent;
1614 			kdh->dumpextent = htod64(dumpextent);
1615 		}
1616 
1617 		/*
1618 		 * The offset at which to begin writing the dump.
1619 		 */
1620 		di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1621 		    dumpextent;
1622 	}
1623 	di->origdumpoff = di->dumpoff;
1624 	return (error);
1625 }
1626 
1627 static int
1628 _dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1629     size_t length)
1630 {
1631 	int error;
1632 
1633 #ifdef EKCD
1634 	if (di->kdcrypto != NULL)
1635 		error = dump_encrypted_write(di, virtual, physical, di->dumpoff,
1636 		    length);
1637 	else
1638 #endif
1639 		error = dump_write(di, virtual, physical, di->dumpoff, length);
1640 	if (error == 0)
1641 		di->dumpoff += length;
1642 	return (error);
1643 }
1644 
1645 /*
1646  * Write to the dump device starting at dumpoff. When compression is enabled,
1647  * writes to the device will be performed using a callback that gets invoked
1648  * when the compression stream's output buffer is full.
1649  */
1650 int
1651 dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1652     size_t length)
1653 {
1654 	void *buf;
1655 
1656 	if (di->kdcomp != NULL) {
1657 		/* Bounce through a buffer to avoid CRC errors. */
1658 		if (length > di->maxiosize)
1659 			return (EINVAL);
1660 		buf = di->kdcomp->kdc_buf;
1661 		memmove(buf, virtual, length);
1662 		return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1663 	}
1664 	return (_dump_append(di, virtual, physical, length));
1665 }
1666 
1667 /*
1668  * Write to the dump device at the specified offset.
1669  */
1670 int
1671 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1672     off_t offset, size_t length)
1673 {
1674 	int error;
1675 
1676 	error = dump_check_bounds(di, offset, length);
1677 	if (error != 0)
1678 		return (error);
1679 	return (di->dumper(di->priv, virtual, physical, offset, length));
1680 }
1681 
1682 /*
1683  * Perform kernel dump finalization: flush the compression stream, if necessary,
1684  * write the leading and trailing kernel dump headers now that we know the true
1685  * length of the dump, and optionally write the encryption key following the
1686  * leading header.
1687  */
1688 int
1689 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1690 {
1691 	int error;
1692 
1693 	if (di->kdcomp != NULL) {
1694 		error = compressor_flush(di->kdcomp->kdc_stream);
1695 		if (error == EAGAIN) {
1696 			/* We have residual data in di->blockbuf. */
1697 			error = _dump_append(di, di->blockbuf, 0, di->blocksize);
1698 			if (error == 0)
1699 				/* Compensate for _dump_append()'s adjustment. */
1700 				di->dumpoff -= di->blocksize - di->kdcomp->kdc_resid;
1701 			di->kdcomp->kdc_resid = 0;
1702 		}
1703 		if (error != 0)
1704 			return (error);
1705 
1706 		/*
1707 		 * We now know the size of the compressed dump, so update the
1708 		 * header accordingly and recompute parity.
1709 		 */
1710 		kdh->dumplength = htod64(di->dumpoff - di->origdumpoff);
1711 		kdh->parity = 0;
1712 		kdh->parity = kerneldump_parity(kdh);
1713 
1714 		compressor_reset(di->kdcomp->kdc_stream);
1715 	}
1716 
1717 	error = dump_write_headers(di, kdh);
1718 	if (error != 0)
1719 		return (error);
1720 
1721 	(void)dump_write(di, NULL, 0, 0, 0);
1722 	return (0);
1723 }
1724 
1725 void
1726 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1727     const char *magic, uint32_t archver, uint64_t dumplen)
1728 {
1729 	size_t dstsize;
1730 
1731 	bzero(kdh, sizeof(*kdh));
1732 	strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1733 	strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1734 	kdh->version = htod32(KERNELDUMPVERSION);
1735 	kdh->architectureversion = htod32(archver);
1736 	kdh->dumplength = htod64(dumplen);
1737 	kdh->dumpextent = kdh->dumplength;
1738 	kdh->dumptime = htod64(time_second);
1739 #ifdef EKCD
1740 	kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1741 #else
1742 	kdh->dumpkeysize = 0;
1743 #endif
1744 	kdh->blocksize = htod32(di->blocksize);
1745 	strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1746 	dstsize = sizeof(kdh->versionstring);
1747 	if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1748 		kdh->versionstring[dstsize - 2] = '\n';
1749 	if (panicstr != NULL)
1750 		strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1751 	if (di->kdcomp != NULL)
1752 		kdh->compression = di->kdcomp->kdc_format;
1753 	kdh->parity = kerneldump_parity(kdh);
1754 }
1755 
1756 #ifdef DDB
1757 DB_SHOW_COMMAND(panic, db_show_panic)
1758 {
1759 
1760 	if (panicstr == NULL)
1761 		db_printf("panicstr not set\n");
1762 	else
1763 		db_printf("panic: %s\n", panicstr);
1764 }
1765 #endif
1766