xref: /freebsd/sys/kern/kern_shutdown.c (revision a4bcd20486f8c20cc875b39bc75aa0d5a047373f)
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 int debugger_on_trap = 0;
131 SYSCTL_INT(_debug, OID_AUTO, debugger_on_trap,
132     CTLFLAG_RWTUN | CTLFLAG_SECURE,
133     &debugger_on_trap, 0, "Run debugger on kernel trap before panic");
134 
135 #ifdef KDB_TRACE
136 static int trace_on_panic = 1;
137 static bool trace_all_panics = true;
138 #else
139 static int trace_on_panic = 0;
140 static bool trace_all_panics = false;
141 #endif
142 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
143     CTLFLAG_RWTUN | CTLFLAG_SECURE,
144     &trace_on_panic, 0, "Print stack trace on kernel panic");
145 SYSCTL_BOOL(_debug, OID_AUTO, trace_all_panics, CTLFLAG_RWTUN,
146     &trace_all_panics, 0, "Print stack traces on secondary kernel panics");
147 #endif /* KDB */
148 
149 static int sync_on_panic = 0;
150 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN,
151 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
152 
153 static bool poweroff_on_panic = 0;
154 SYSCTL_BOOL(_kern, OID_AUTO, poweroff_on_panic, CTLFLAG_RWTUN,
155 	&poweroff_on_panic, 0, "Do a power off instead of a reboot on a panic");
156 
157 static bool powercycle_on_panic = 0;
158 SYSCTL_BOOL(_kern, OID_AUTO, powercycle_on_panic, CTLFLAG_RWTUN,
159 	&powercycle_on_panic, 0, "Do a power cycle instead of a reboot on a panic");
160 
161 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
162     "Shutdown environment");
163 
164 #ifndef DIAGNOSTIC
165 static int show_busybufs;
166 #else
167 static int show_busybufs = 1;
168 #endif
169 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
170     &show_busybufs, 0,
171     "Show busy buffers during shutdown");
172 
173 int suspend_blocked = 0;
174 SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW,
175 	&suspend_blocked, 0, "Block suspend due to a pending shutdown");
176 
177 #ifdef EKCD
178 FEATURE(ekcd, "Encrypted kernel crash dumps support");
179 
180 MALLOC_DEFINE(M_EKCD, "ekcd", "Encrypted kernel crash dumps data");
181 
182 struct kerneldumpcrypto {
183 	uint8_t			kdc_encryption;
184 	uint8_t			kdc_iv[KERNELDUMP_IV_MAX_SIZE];
185 	union {
186 		struct {
187 			keyInstance	aes_ki;
188 			cipherInstance	aes_ci;
189 		} u_aes;
190 		struct chacha_ctx	u_chacha;
191 	} u;
192 #define	kdc_ki	u.u_aes.aes_ki
193 #define	kdc_ci	u.u_aes.aes_ci
194 #define	kdc_chacha	u.u_chacha
195 	uint32_t		kdc_dumpkeysize;
196 	struct kerneldumpkey	kdc_dumpkey[];
197 };
198 #endif
199 
200 struct kerneldumpcomp {
201 	uint8_t			kdc_format;
202 	struct compressor	*kdc_stream;
203 	uint8_t			*kdc_buf;
204 	size_t			kdc_resid;
205 };
206 
207 static struct kerneldumpcomp *kerneldumpcomp_create(struct dumperinfo *di,
208 		    uint8_t compression);
209 static void	kerneldumpcomp_destroy(struct dumperinfo *di);
210 static int	kerneldumpcomp_write_cb(void *base, size_t len, off_t off, void *arg);
211 
212 static int kerneldump_gzlevel = 6;
213 SYSCTL_INT(_kern, OID_AUTO, kerneldump_gzlevel, CTLFLAG_RWTUN,
214     &kerneldump_gzlevel, 0,
215     "Kernel crash dump compression level");
216 
217 /*
218  * Variable panicstr contains argument to first call to panic; used as flag
219  * to indicate that the kernel has already called panic.
220  */
221 const char *panicstr;
222 bool __read_frequently panicked;
223 
224 int __read_mostly dumping;		/* system is dumping */
225 int rebooting;				/* system is rebooting */
226 /*
227  * Used to serialize between sysctl kern.shutdown.dumpdevname and list
228  * modifications via ioctl.
229  */
230 static struct mtx dumpconf_list_lk;
231 MTX_SYSINIT(dumper_configs, &dumpconf_list_lk, "dumper config list", MTX_DEF);
232 
233 /* Our selected dumper(s). */
234 static TAILQ_HEAD(dumpconflist, dumperinfo) dumper_configs =
235     TAILQ_HEAD_INITIALIZER(dumper_configs);
236 
237 /* Context information for dump-debuggers. */
238 static struct pcb dumppcb;		/* Registers. */
239 lwpid_t dumptid;			/* Thread ID. */
240 
241 static struct cdevsw reroot_cdevsw = {
242      .d_version = D_VERSION,
243      .d_name    = "reroot",
244 };
245 
246 static void poweroff_wait(void *, int);
247 static void shutdown_halt(void *junk, int howto);
248 static void shutdown_panic(void *junk, int howto);
249 static void shutdown_reset(void *junk, int howto);
250 static int kern_reroot(void);
251 
252 /* register various local shutdown events */
253 static void
254 shutdown_conf(void *unused)
255 {
256 
257 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
258 	    SHUTDOWN_PRI_FIRST);
259 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
260 	    SHUTDOWN_PRI_LAST + 100);
261 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
262 	    SHUTDOWN_PRI_LAST + 100);
263 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL,
264 	    SHUTDOWN_PRI_LAST + 200);
265 }
266 
267 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
268 
269 /*
270  * The only reason this exists is to create the /dev/reroot/ directory,
271  * used by reroot code in init(8) as a mountpoint for tmpfs.
272  */
273 static void
274 reroot_conf(void *unused)
275 {
276 	int error;
277 	struct cdev *cdev;
278 
279 	error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev,
280 	    &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot");
281 	if (error != 0) {
282 		printf("%s: failed to create device node, error %d",
283 		    __func__, error);
284 	}
285 }
286 
287 SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL);
288 
289 /*
290  * The system call that results in a reboot.
291  */
292 /* ARGSUSED */
293 int
294 sys_reboot(struct thread *td, struct reboot_args *uap)
295 {
296 	int error;
297 
298 	error = 0;
299 #ifdef MAC
300 	error = mac_system_check_reboot(td->td_ucred, uap->opt);
301 #endif
302 	if (error == 0)
303 		error = priv_check(td, PRIV_REBOOT);
304 	if (error == 0) {
305 		if (uap->opt & RB_REROOT)
306 			error = kern_reroot();
307 		else
308 			kern_reboot(uap->opt);
309 	}
310 	return (error);
311 }
312 
313 static void
314 shutdown_nice_task_fn(void *arg, int pending __unused)
315 {
316 	int howto;
317 
318 	howto = (uintptr_t)arg;
319 	/* Send a signal to init(8) and have it shutdown the world. */
320 	PROC_LOCK(initproc);
321 	if (howto & RB_POWEROFF)
322 		kern_psignal(initproc, SIGUSR2);
323 	else if (howto & RB_POWERCYCLE)
324 		kern_psignal(initproc, SIGWINCH);
325 	else if (howto & RB_HALT)
326 		kern_psignal(initproc, SIGUSR1);
327 	else
328 		kern_psignal(initproc, SIGINT);
329 	PROC_UNLOCK(initproc);
330 }
331 
332 static struct task shutdown_nice_task = TASK_INITIALIZER(0,
333     &shutdown_nice_task_fn, NULL);
334 
335 /*
336  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
337  */
338 void
339 shutdown_nice(int howto)
340 {
341 
342 	if (initproc != NULL && !SCHEDULER_STOPPED()) {
343 		shutdown_nice_task.ta_context = (void *)(uintptr_t)howto;
344 		taskqueue_enqueue(taskqueue_fast, &shutdown_nice_task);
345 	} else {
346 		/*
347 		 * No init(8) running, or scheduler would not allow it
348 		 * to run, so simply reboot.
349 		 */
350 		kern_reboot(howto | RB_NOSYNC);
351 	}
352 }
353 
354 static void
355 print_uptime(void)
356 {
357 	int f;
358 	struct timespec ts;
359 
360 	getnanouptime(&ts);
361 	printf("Uptime: ");
362 	f = 0;
363 	if (ts.tv_sec >= 86400) {
364 		printf("%ldd", (long)ts.tv_sec / 86400);
365 		ts.tv_sec %= 86400;
366 		f = 1;
367 	}
368 	if (f || ts.tv_sec >= 3600) {
369 		printf("%ldh", (long)ts.tv_sec / 3600);
370 		ts.tv_sec %= 3600;
371 		f = 1;
372 	}
373 	if (f || ts.tv_sec >= 60) {
374 		printf("%ldm", (long)ts.tv_sec / 60);
375 		ts.tv_sec %= 60;
376 		f = 1;
377 	}
378 	printf("%lds\n", (long)ts.tv_sec);
379 }
380 
381 int
382 doadump(boolean_t textdump)
383 {
384 	boolean_t coredump;
385 	int error;
386 
387 	error = 0;
388 	if (dumping)
389 		return (EBUSY);
390 	if (TAILQ_EMPTY(&dumper_configs))
391 		return (ENXIO);
392 
393 	savectx(&dumppcb);
394 	dumptid = curthread->td_tid;
395 	dumping++;
396 
397 	coredump = TRUE;
398 #ifdef DDB
399 	if (textdump && textdump_pending) {
400 		coredump = FALSE;
401 		textdump_dumpsys(TAILQ_FIRST(&dumper_configs));
402 	}
403 #endif
404 	if (coredump) {
405 		struct dumperinfo *di;
406 
407 		TAILQ_FOREACH(di, &dumper_configs, di_next) {
408 			error = dumpsys(di);
409 			if (error == 0)
410 				break;
411 		}
412 	}
413 
414 	dumping--;
415 	return (error);
416 }
417 
418 /*
419  * Shutdown the system cleanly to prepare for reboot, halt, or power off.
420  */
421 void
422 kern_reboot(int howto)
423 {
424 	static int once = 0;
425 
426 	/*
427 	 * Normal paths here don't hold Giant, but we can wind up here
428 	 * unexpectedly with it held.  Drop it now so we don't have to
429 	 * drop and pick it up elsewhere. The paths it is locking will
430 	 * never be returned to, and it is preferable to preclude
431 	 * deadlock than to lock against code that won't ever
432 	 * continue.
433 	 */
434 	while (mtx_owned(&Giant))
435 		mtx_unlock(&Giant);
436 
437 #if defined(SMP)
438 	/*
439 	 * Bind us to the first CPU so that all shutdown code runs there.  Some
440 	 * systems don't shutdown properly (i.e., ACPI power off) if we
441 	 * run on another processor.
442 	 */
443 	if (!SCHEDULER_STOPPED()) {
444 		thread_lock(curthread);
445 		sched_bind(curthread, CPU_FIRST());
446 		thread_unlock(curthread);
447 		KASSERT(PCPU_GET(cpuid) == CPU_FIRST(),
448 		    ("boot: not running on cpu 0"));
449 	}
450 #endif
451 	/* We're in the process of rebooting. */
452 	rebooting = 1;
453 
454 	/* We are out of the debugger now. */
455 	kdb_active = 0;
456 
457 	/*
458 	 * Do any callouts that should be done BEFORE syncing the filesystems.
459 	 */
460 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
461 
462 	/*
463 	 * Now sync filesystems
464 	 */
465 	if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
466 		once = 1;
467 		bufshutdown(show_busybufs);
468 	}
469 
470 	print_uptime();
471 
472 	cngrab();
473 
474 	/*
475 	 * Ok, now do things that assume all filesystem activity has
476 	 * been completed.
477 	 */
478 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
479 
480 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
481 		doadump(TRUE);
482 
483 	/* Now that we're going to really halt the system... */
484 	EVENTHANDLER_INVOKE(shutdown_final, howto);
485 
486 	for(;;) ;	/* safety against shutdown_reset not working */
487 	/* NOTREACHED */
488 }
489 
490 /*
491  * The system call that results in changing the rootfs.
492  */
493 static int
494 kern_reroot(void)
495 {
496 	struct vnode *oldrootvnode, *vp;
497 	struct mount *mp, *devmp;
498 	int error;
499 
500 	if (curproc != initproc)
501 		return (EPERM);
502 
503 	/*
504 	 * Mark the filesystem containing currently-running executable
505 	 * (the temporary copy of init(8)) busy.
506 	 */
507 	vp = curproc->p_textvp;
508 	error = vn_lock(vp, LK_SHARED);
509 	if (error != 0)
510 		return (error);
511 	mp = vp->v_mount;
512 	error = vfs_busy(mp, MBF_NOWAIT);
513 	if (error != 0) {
514 		vfs_ref(mp);
515 		VOP_UNLOCK(vp);
516 		error = vfs_busy(mp, 0);
517 		vn_lock(vp, LK_SHARED | LK_RETRY);
518 		vfs_rel(mp);
519 		if (error != 0) {
520 			VOP_UNLOCK(vp);
521 			return (ENOENT);
522 		}
523 		if (VN_IS_DOOMED(vp)) {
524 			VOP_UNLOCK(vp);
525 			vfs_unbusy(mp);
526 			return (ENOENT);
527 		}
528 	}
529 	VOP_UNLOCK(vp);
530 
531 	/*
532 	 * Remove the filesystem containing currently-running executable
533 	 * from the mount list, to prevent it from being unmounted
534 	 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
535 	 *
536 	 * Also preserve /dev - forcibly unmounting it could cause driver
537 	 * reinitialization.
538 	 */
539 
540 	vfs_ref(rootdevmp);
541 	devmp = rootdevmp;
542 	rootdevmp = NULL;
543 
544 	mtx_lock(&mountlist_mtx);
545 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
546 	TAILQ_REMOVE(&mountlist, devmp, mnt_list);
547 	mtx_unlock(&mountlist_mtx);
548 
549 	oldrootvnode = rootvnode;
550 
551 	/*
552 	 * Unmount everything except for the two filesystems preserved above.
553 	 */
554 	vfs_unmountall();
555 
556 	/*
557 	 * Add /dev back; vfs_mountroot() will move it into its new place.
558 	 */
559 	mtx_lock(&mountlist_mtx);
560 	TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
561 	mtx_unlock(&mountlist_mtx);
562 	rootdevmp = devmp;
563 	vfs_rel(rootdevmp);
564 
565 	/*
566 	 * Mount the new rootfs.
567 	 */
568 	vfs_mountroot();
569 
570 	/*
571 	 * Update all references to the old rootvnode.
572 	 */
573 	mountcheckdirs(oldrootvnode, rootvnode);
574 
575 	/*
576 	 * Add the temporary filesystem back and unbusy it.
577 	 */
578 	mtx_lock(&mountlist_mtx);
579 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
580 	mtx_unlock(&mountlist_mtx);
581 	vfs_unbusy(mp);
582 
583 	return (0);
584 }
585 
586 /*
587  * If the shutdown was a clean halt, behave accordingly.
588  */
589 static void
590 shutdown_halt(void *junk, int howto)
591 {
592 
593 	if (howto & RB_HALT) {
594 		printf("\n");
595 		printf("The operating system has halted.\n");
596 		printf("Please press any key to reboot.\n\n");
597 
598 		wdog_kern_pat(WD_TO_NEVER);
599 
600 		switch (cngetc()) {
601 		case -1:		/* No console, just die */
602 			cpu_halt();
603 			/* NOTREACHED */
604 		default:
605 			break;
606 		}
607 	}
608 }
609 
610 /*
611  * Check to see if the system paniced, pause and then reboot
612  * according to the specified delay.
613  */
614 static void
615 shutdown_panic(void *junk, int howto)
616 {
617 	int loop;
618 
619 	if (howto & RB_DUMP) {
620 		if (panic_reboot_wait_time != 0) {
621 			if (panic_reboot_wait_time != -1) {
622 				printf("Automatic reboot in %d seconds - "
623 				       "press a key on the console to abort\n",
624 					panic_reboot_wait_time);
625 				for (loop = panic_reboot_wait_time * 10;
626 				     loop > 0; --loop) {
627 					DELAY(1000 * 100); /* 1/10th second */
628 					/* Did user type a key? */
629 					if (cncheckc() != -1)
630 						break;
631 				}
632 				if (!loop)
633 					return;
634 			}
635 		} else { /* zero time specified - reboot NOW */
636 			return;
637 		}
638 		printf("--> Press a key on the console to reboot,\n");
639 		printf("--> or switch off the system now.\n");
640 		cngetc();
641 	}
642 }
643 
644 /*
645  * Everything done, now reset
646  */
647 static void
648 shutdown_reset(void *junk, int howto)
649 {
650 
651 	printf("Rebooting...\n");
652 	DELAY(1000000);	/* wait 1 sec for printf's to complete and be read */
653 
654 	/*
655 	 * Acquiring smp_ipi_mtx here has a double effect:
656 	 * - it disables interrupts avoiding CPU0 preemption
657 	 *   by fast handlers (thus deadlocking  against other CPUs)
658 	 * - it avoids deadlocks against smp_rendezvous() or, more
659 	 *   generally, threads busy-waiting, with this spinlock held,
660 	 *   and waiting for responses by threads on other CPUs
661 	 *   (ie. smp_tlb_shootdown()).
662 	 *
663 	 * For the !SMP case it just needs to handle the former problem.
664 	 */
665 #ifdef SMP
666 	mtx_lock_spin(&smp_ipi_mtx);
667 #else
668 	spinlock_enter();
669 #endif
670 
671 	cpu_reset();
672 	/* NOTREACHED */ /* assuming reset worked */
673 }
674 
675 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
676 static int kassert_warn_only = 0;
677 #ifdef KDB
678 static int kassert_do_kdb = 0;
679 #endif
680 #ifdef KTR
681 static int kassert_do_ktr = 0;
682 #endif
683 static int kassert_do_log = 1;
684 static int kassert_log_pps_limit = 4;
685 static int kassert_log_mute_at = 0;
686 static int kassert_log_panic_at = 0;
687 static int kassert_suppress_in_panic = 0;
688 static int kassert_warnings = 0;
689 
690 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
691     "kassert options");
692 
693 #ifdef KASSERT_PANIC_OPTIONAL
694 #define KASSERT_RWTUN	CTLFLAG_RWTUN
695 #else
696 #define KASSERT_RWTUN	CTLFLAG_RDTUN
697 #endif
698 
699 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, KASSERT_RWTUN,
700     &kassert_warn_only, 0,
701     "KASSERT triggers a panic (0) or just a warning (1)");
702 
703 #ifdef KDB
704 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, KASSERT_RWTUN,
705     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
706 #endif
707 
708 #ifdef KTR
709 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, KASSERT_RWTUN,
710     &kassert_do_ktr, 0,
711     "KASSERT does a KTR, set this to the KTRMASK you want");
712 #endif
713 
714 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, KASSERT_RWTUN,
715     &kassert_do_log, 0,
716     "If warn_only is enabled, log (1) or do not log (0) assertion violations");
717 
718 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RD | CTLFLAG_STATS,
719     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
720 
721 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, KASSERT_RWTUN,
722     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
723 
724 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, KASSERT_RWTUN,
725     &kassert_log_pps_limit, 0, "limit number of log messages per second");
726 
727 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, KASSERT_RWTUN,
728     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
729 
730 SYSCTL_INT(_debug_kassert, OID_AUTO, suppress_in_panic, KASSERT_RWTUN,
731     &kassert_suppress_in_panic, 0,
732     "KASSERTs will be suppressed while handling a panic");
733 #undef KASSERT_RWTUN
734 
735 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
736 
737 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
738     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_NEEDGIANT, NULL, 0,
739     kassert_sysctl_kassert, "I",
740     "set to trigger a test kassert");
741 
742 static int
743 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
744 {
745 	int error, i;
746 
747 	error = sysctl_wire_old_buffer(req, sizeof(int));
748 	if (error == 0) {
749 		i = 0;
750 		error = sysctl_handle_int(oidp, &i, 0, req);
751 	}
752 	if (error != 0 || req->newptr == NULL)
753 		return (error);
754 	KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
755 	return (0);
756 }
757 
758 #ifdef KASSERT_PANIC_OPTIONAL
759 /*
760  * Called by KASSERT, this decides if we will panic
761  * or if we will log via printf and/or ktr.
762  */
763 void
764 kassert_panic(const char *fmt, ...)
765 {
766 	static char buf[256];
767 	va_list ap;
768 
769 	va_start(ap, fmt);
770 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
771 	va_end(ap);
772 
773 	/*
774 	 * If we are suppressing secondary panics, log the warning but do not
775 	 * re-enter panic/kdb.
776 	 */
777 	if (panicstr != NULL && kassert_suppress_in_panic) {
778 		if (kassert_do_log) {
779 			printf("KASSERT failed: %s\n", buf);
780 #ifdef KDB
781 			if (trace_all_panics && trace_on_panic)
782 				kdb_backtrace();
783 #endif
784 		}
785 		return;
786 	}
787 
788 	/*
789 	 * panic if we're not just warning, or if we've exceeded
790 	 * kassert_log_panic_at warnings.
791 	 */
792 	if (!kassert_warn_only ||
793 	    (kassert_log_panic_at > 0 &&
794 	     kassert_warnings >= kassert_log_panic_at)) {
795 		va_start(ap, fmt);
796 		vpanic(fmt, ap);
797 		/* NORETURN */
798 	}
799 #ifdef KTR
800 	if (kassert_do_ktr)
801 		CTR0(ktr_mask, buf);
802 #endif /* KTR */
803 	/*
804 	 * log if we've not yet met the mute limit.
805 	 */
806 	if (kassert_do_log &&
807 	    (kassert_log_mute_at == 0 ||
808 	     kassert_warnings < kassert_log_mute_at)) {
809 		static  struct timeval lasterr;
810 		static  int curerr;
811 
812 		if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
813 			printf("KASSERT failed: %s\n", buf);
814 			kdb_backtrace();
815 		}
816 	}
817 #ifdef KDB
818 	if (kassert_do_kdb) {
819 		kdb_enter(KDB_WHY_KASSERT, buf);
820 	}
821 #endif
822 	atomic_add_int(&kassert_warnings, 1);
823 }
824 #endif /* KASSERT_PANIC_OPTIONAL */
825 #endif
826 
827 /*
828  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
829  * and then reboots.  If we are called twice, then we avoid trying to sync
830  * the disks as this often leads to recursive panics.
831  */
832 void
833 panic(const char *fmt, ...)
834 {
835 	va_list ap;
836 
837 	va_start(ap, fmt);
838 	vpanic(fmt, ap);
839 }
840 
841 void
842 vpanic(const char *fmt, va_list ap)
843 {
844 #ifdef SMP
845 	cpuset_t other_cpus;
846 #endif
847 	struct thread *td = curthread;
848 	int bootopt, newpanic;
849 	static char buf[256];
850 
851 	spinlock_enter();
852 
853 #ifdef SMP
854 	/*
855 	 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
856 	 * concurrently entering panic.  Only the winner will proceed
857 	 * further.
858 	 */
859 	if (panicstr == NULL && !kdb_active) {
860 		other_cpus = all_cpus;
861 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
862 		stop_cpus_hard(other_cpus);
863 	}
864 #endif
865 
866 	/*
867 	 * Ensure that the scheduler is stopped while panicking, even if panic
868 	 * has been entered from kdb.
869 	 */
870 	td->td_stopsched = 1;
871 
872 	bootopt = RB_AUTOBOOT;
873 	newpanic = 0;
874 	if (panicstr)
875 		bootopt |= RB_NOSYNC;
876 	else {
877 		bootopt |= RB_DUMP;
878 		panicstr = fmt;
879 		panicked = true;
880 		newpanic = 1;
881 	}
882 
883 	if (newpanic) {
884 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
885 		panicstr = buf;
886 		cngrab();
887 		printf("panic: %s\n", buf);
888 	} else {
889 		printf("panic: ");
890 		vprintf(fmt, ap);
891 		printf("\n");
892 	}
893 #ifdef SMP
894 	printf("cpuid = %d\n", PCPU_GET(cpuid));
895 #endif
896 	printf("time = %jd\n", (intmax_t )time_second);
897 #ifdef KDB
898 	if ((newpanic || trace_all_panics) && trace_on_panic)
899 		kdb_backtrace();
900 	if (debugger_on_panic)
901 		kdb_enter(KDB_WHY_PANIC, "panic");
902 #endif
903 	/*thread_lock(td); */
904 	td->td_flags |= TDF_INPANIC;
905 	/* thread_unlock(td); */
906 	if (!sync_on_panic)
907 		bootopt |= RB_NOSYNC;
908 	if (poweroff_on_panic)
909 		bootopt |= RB_POWEROFF;
910 	if (powercycle_on_panic)
911 		bootopt |= RB_POWERCYCLE;
912 	kern_reboot(bootopt);
913 }
914 
915 /*
916  * Support for poweroff delay.
917  *
918  * Please note that setting this delay too short might power off your machine
919  * before the write cache on your hard disk has been flushed, leading to
920  * soft-updates inconsistencies.
921  */
922 #ifndef POWEROFF_DELAY
923 # define POWEROFF_DELAY 5000
924 #endif
925 static int poweroff_delay = POWEROFF_DELAY;
926 
927 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
928     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
929 
930 static void
931 poweroff_wait(void *junk, int howto)
932 {
933 
934 	if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
935 		return;
936 	DELAY(poweroff_delay * 1000);
937 }
938 
939 /*
940  * Some system processes (e.g. syncer) need to be stopped at appropriate
941  * points in their main loops prior to a system shutdown, so that they
942  * won't interfere with the shutdown process (e.g. by holding a disk buf
943  * to cause sync to fail).  For each of these system processes, register
944  * shutdown_kproc() as a handler for one of shutdown events.
945  */
946 static int kproc_shutdown_wait = 60;
947 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
948     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
949 
950 void
951 kproc_shutdown(void *arg, int howto)
952 {
953 	struct proc *p;
954 	int error;
955 
956 	if (panicstr)
957 		return;
958 
959 	p = (struct proc *)arg;
960 	printf("Waiting (max %d seconds) for system process `%s' to stop... ",
961 	    kproc_shutdown_wait, p->p_comm);
962 	error = kproc_suspend(p, kproc_shutdown_wait * hz);
963 
964 	if (error == EWOULDBLOCK)
965 		printf("timed out\n");
966 	else
967 		printf("done\n");
968 }
969 
970 void
971 kthread_shutdown(void *arg, int howto)
972 {
973 	struct thread *td;
974 	int error;
975 
976 	if (panicstr)
977 		return;
978 
979 	td = (struct thread *)arg;
980 	printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
981 	    kproc_shutdown_wait, td->td_name);
982 	error = kthread_suspend(td, kproc_shutdown_wait * hz);
983 
984 	if (error == EWOULDBLOCK)
985 		printf("timed out\n");
986 	else
987 		printf("done\n");
988 }
989 
990 static int
991 dumpdevname_sysctl_handler(SYSCTL_HANDLER_ARGS)
992 {
993 	char buf[256];
994 	struct dumperinfo *di;
995 	struct sbuf sb;
996 	int error;
997 
998 	error = sysctl_wire_old_buffer(req, 0);
999 	if (error != 0)
1000 		return (error);
1001 
1002 	sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1003 
1004 	mtx_lock(&dumpconf_list_lk);
1005 	TAILQ_FOREACH(di, &dumper_configs, di_next) {
1006 		if (di != TAILQ_FIRST(&dumper_configs))
1007 			sbuf_putc(&sb, ',');
1008 		sbuf_cat(&sb, di->di_devname);
1009 	}
1010 	mtx_unlock(&dumpconf_list_lk);
1011 
1012 	error = sbuf_finish(&sb);
1013 	sbuf_delete(&sb);
1014 	return (error);
1015 }
1016 SYSCTL_PROC(_kern_shutdown, OID_AUTO, dumpdevname,
1017     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, &dumper_configs, 0,
1018     dumpdevname_sysctl_handler, "A",
1019     "Device(s) for kernel dumps");
1020 
1021 static int	_dump_append(struct dumperinfo *di, void *virtual,
1022 		    vm_offset_t physical, size_t length);
1023 
1024 #ifdef EKCD
1025 static struct kerneldumpcrypto *
1026 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
1027     const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1028 {
1029 	struct kerneldumpcrypto *kdc;
1030 	struct kerneldumpkey *kdk;
1031 	uint32_t dumpkeysize;
1032 
1033 	dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
1034 	kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
1035 
1036 	arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
1037 
1038 	kdc->kdc_encryption = encryption;
1039 	switch (kdc->kdc_encryption) {
1040 	case KERNELDUMP_ENC_AES_256_CBC:
1041 		if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
1042 			goto failed;
1043 		break;
1044 	case KERNELDUMP_ENC_CHACHA20:
1045 		chacha_keysetup(&kdc->kdc_chacha, key, 256);
1046 		break;
1047 	default:
1048 		goto failed;
1049 	}
1050 
1051 	kdc->kdc_dumpkeysize = dumpkeysize;
1052 	kdk = kdc->kdc_dumpkey;
1053 	kdk->kdk_encryption = kdc->kdc_encryption;
1054 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1055 	kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
1056 	memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
1057 
1058 	return (kdc);
1059 failed:
1060 	zfree(kdc, M_EKCD);
1061 	return (NULL);
1062 }
1063 
1064 static int
1065 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
1066 {
1067 	uint8_t hash[SHA256_DIGEST_LENGTH];
1068 	SHA256_CTX ctx;
1069 	struct kerneldumpkey *kdk;
1070 	int error;
1071 
1072 	error = 0;
1073 
1074 	if (kdc == NULL)
1075 		return (0);
1076 
1077 	/*
1078 	 * When a user enters ddb it can write a crash dump multiple times.
1079 	 * Each time it should be encrypted using a different IV.
1080 	 */
1081 	SHA256_Init(&ctx);
1082 	SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1083 	SHA256_Final(hash, &ctx);
1084 	bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1085 
1086 	switch (kdc->kdc_encryption) {
1087 	case KERNELDUMP_ENC_AES_256_CBC:
1088 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1089 		    kdc->kdc_iv) <= 0) {
1090 			error = EINVAL;
1091 			goto out;
1092 		}
1093 		break;
1094 	case KERNELDUMP_ENC_CHACHA20:
1095 		chacha_ivsetup(&kdc->kdc_chacha, kdc->kdc_iv, NULL);
1096 		break;
1097 	default:
1098 		error = EINVAL;
1099 		goto out;
1100 	}
1101 
1102 	kdk = kdc->kdc_dumpkey;
1103 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1104 out:
1105 	explicit_bzero(hash, sizeof(hash));
1106 	return (error);
1107 }
1108 
1109 static uint32_t
1110 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
1111 {
1112 
1113 	if (kdc == NULL)
1114 		return (0);
1115 	return (kdc->kdc_dumpkeysize);
1116 }
1117 #endif /* EKCD */
1118 
1119 static struct kerneldumpcomp *
1120 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression)
1121 {
1122 	struct kerneldumpcomp *kdcomp;
1123 	int format;
1124 
1125 	switch (compression) {
1126 	case KERNELDUMP_COMP_GZIP:
1127 		format = COMPRESS_GZIP;
1128 		break;
1129 	case KERNELDUMP_COMP_ZSTD:
1130 		format = COMPRESS_ZSTD;
1131 		break;
1132 	default:
1133 		return (NULL);
1134 	}
1135 
1136 	kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO);
1137 	kdcomp->kdc_format = compression;
1138 	kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb,
1139 	    format, di->maxiosize, kerneldump_gzlevel, di);
1140 	if (kdcomp->kdc_stream == NULL) {
1141 		free(kdcomp, M_DUMPER);
1142 		return (NULL);
1143 	}
1144 	kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1145 	return (kdcomp);
1146 }
1147 
1148 static void
1149 kerneldumpcomp_destroy(struct dumperinfo *di)
1150 {
1151 	struct kerneldumpcomp *kdcomp;
1152 
1153 	kdcomp = di->kdcomp;
1154 	if (kdcomp == NULL)
1155 		return;
1156 	compressor_fini(kdcomp->kdc_stream);
1157 	zfree(kdcomp->kdc_buf, M_DUMPER);
1158 	free(kdcomp, M_DUMPER);
1159 }
1160 
1161 /*
1162  * Must not be present on global list.
1163  */
1164 static void
1165 free_single_dumper(struct dumperinfo *di)
1166 {
1167 
1168 	if (di == NULL)
1169 		return;
1170 
1171 	zfree(di->blockbuf, M_DUMPER);
1172 
1173 	kerneldumpcomp_destroy(di);
1174 
1175 #ifdef EKCD
1176 	zfree(di->kdcrypto, M_EKCD);
1177 #endif
1178 	zfree(di, M_DUMPER);
1179 }
1180 
1181 /* Registration of dumpers */
1182 int
1183 dumper_insert(const struct dumperinfo *di_template, const char *devname,
1184     const struct diocskerneldump_arg *kda)
1185 {
1186 	struct dumperinfo *newdi, *listdi;
1187 	bool inserted;
1188 	uint8_t index;
1189 	int error;
1190 
1191 	index = kda->kda_index;
1192 	MPASS(index != KDA_REMOVE && index != KDA_REMOVE_DEV &&
1193 	    index != KDA_REMOVE_ALL);
1194 
1195 	error = priv_check(curthread, PRIV_SETDUMPER);
1196 	if (error != 0)
1197 		return (error);
1198 
1199 	newdi = malloc(sizeof(*newdi) + strlen(devname) + 1, M_DUMPER, M_WAITOK
1200 	    | M_ZERO);
1201 	memcpy(newdi, di_template, sizeof(*newdi));
1202 	newdi->blockbuf = NULL;
1203 	newdi->kdcrypto = NULL;
1204 	newdi->kdcomp = NULL;
1205 	strcpy(newdi->di_devname, devname);
1206 
1207 	if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
1208 #ifdef EKCD
1209 		newdi->kdcrypto = kerneldumpcrypto_create(di_template->blocksize,
1210 		    kda->kda_encryption, kda->kda_key,
1211 		    kda->kda_encryptedkeysize, kda->kda_encryptedkey);
1212 		if (newdi->kdcrypto == NULL) {
1213 			error = EINVAL;
1214 			goto cleanup;
1215 		}
1216 #else
1217 		error = EOPNOTSUPP;
1218 		goto cleanup;
1219 #endif
1220 	}
1221 	if (kda->kda_compression != KERNELDUMP_COMP_NONE) {
1222 #ifdef EKCD
1223 		/*
1224 		 * We can't support simultaneous unpadded block cipher
1225 		 * encryption and compression because there is no guarantee the
1226 		 * length of the compressed result is exactly a multiple of the
1227 		 * cipher block size.
1228 		 */
1229 		if (kda->kda_encryption == KERNELDUMP_ENC_AES_256_CBC) {
1230 			error = EOPNOTSUPP;
1231 			goto cleanup;
1232 		}
1233 #endif
1234 		newdi->kdcomp = kerneldumpcomp_create(newdi,
1235 		    kda->kda_compression);
1236 		if (newdi->kdcomp == NULL) {
1237 			error = EINVAL;
1238 			goto cleanup;
1239 		}
1240 	}
1241 
1242 	newdi->blockbuf = malloc(newdi->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1243 
1244 	/* Add the new configuration to the queue */
1245 	mtx_lock(&dumpconf_list_lk);
1246 	inserted = false;
1247 	TAILQ_FOREACH(listdi, &dumper_configs, di_next) {
1248 		if (index == 0) {
1249 			TAILQ_INSERT_BEFORE(listdi, newdi, di_next);
1250 			inserted = true;
1251 			break;
1252 		}
1253 		index--;
1254 	}
1255 	if (!inserted)
1256 		TAILQ_INSERT_TAIL(&dumper_configs, newdi, di_next);
1257 	mtx_unlock(&dumpconf_list_lk);
1258 
1259 	return (0);
1260 
1261 cleanup:
1262 	free_single_dumper(newdi);
1263 	return (error);
1264 }
1265 
1266 #ifdef DDB
1267 void
1268 dumper_ddb_insert(struct dumperinfo *newdi)
1269 {
1270 	TAILQ_INSERT_HEAD(&dumper_configs, newdi, di_next);
1271 }
1272 
1273 void
1274 dumper_ddb_remove(struct dumperinfo *di)
1275 {
1276 	TAILQ_REMOVE(&dumper_configs, di, di_next);
1277 }
1278 #endif
1279 
1280 static bool
1281 dumper_config_match(const struct dumperinfo *di, const char *devname,
1282     const struct diocskerneldump_arg *kda)
1283 {
1284 	if (kda->kda_index == KDA_REMOVE_ALL)
1285 		return (true);
1286 
1287 	if (strcmp(di->di_devname, devname) != 0)
1288 		return (false);
1289 
1290 	/*
1291 	 * Allow wildcard removal of configs matching a device on g_dev_orphan.
1292 	 */
1293 	if (kda->kda_index == KDA_REMOVE_DEV)
1294 		return (true);
1295 
1296 	if (di->kdcomp != NULL) {
1297 		if (di->kdcomp->kdc_format != kda->kda_compression)
1298 			return (false);
1299 	} else if (kda->kda_compression != KERNELDUMP_COMP_NONE)
1300 		return (false);
1301 #ifdef EKCD
1302 	if (di->kdcrypto != NULL) {
1303 		if (di->kdcrypto->kdc_encryption != kda->kda_encryption)
1304 			return (false);
1305 		/*
1306 		 * Do we care to verify keys match to delete?  It seems weird
1307 		 * to expect multiple fallback dump configurations on the same
1308 		 * device that only differ in crypto key.
1309 		 */
1310 	} else
1311 #endif
1312 		if (kda->kda_encryption != KERNELDUMP_ENC_NONE)
1313 			return (false);
1314 
1315 	return (true);
1316 }
1317 
1318 int
1319 dumper_remove(const char *devname, const struct diocskerneldump_arg *kda)
1320 {
1321 	struct dumperinfo *di, *sdi;
1322 	bool found;
1323 	int error;
1324 
1325 	error = priv_check(curthread, PRIV_SETDUMPER);
1326 	if (error != 0)
1327 		return (error);
1328 
1329 	/*
1330 	 * Try to find a matching configuration, and kill it.
1331 	 *
1332 	 * NULL 'kda' indicates remove any configuration matching 'devname',
1333 	 * which may remove multiple configurations in atypical configurations.
1334 	 */
1335 	found = false;
1336 	mtx_lock(&dumpconf_list_lk);
1337 	TAILQ_FOREACH_SAFE(di, &dumper_configs, di_next, sdi) {
1338 		if (dumper_config_match(di, devname, kda)) {
1339 			found = true;
1340 			TAILQ_REMOVE(&dumper_configs, di, di_next);
1341 			free_single_dumper(di);
1342 		}
1343 	}
1344 	mtx_unlock(&dumpconf_list_lk);
1345 
1346 	/* Only produce ENOENT if a more targeted match didn't match. */
1347 	if (!found && kda->kda_index == KDA_REMOVE)
1348 		return (ENOENT);
1349 	return (0);
1350 }
1351 
1352 static int
1353 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1354 {
1355 
1356 	if (di->mediasize > 0 && length != 0 && (offset < di->mediaoffset ||
1357 	    offset - di->mediaoffset + length > di->mediasize)) {
1358 		if (di->kdcomp != NULL && offset >= di->mediaoffset) {
1359 			printf(
1360 		    "Compressed dump failed to fit in device boundaries.\n");
1361 			return (E2BIG);
1362 		}
1363 
1364 		printf("Attempt to write outside dump device boundaries.\n"
1365 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1366 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
1367 		    (uintmax_t)length, (intmax_t)di->mediasize);
1368 		return (ENOSPC);
1369 	}
1370 	if (length % di->blocksize != 0) {
1371 		printf("Attempt to write partial block of length %ju.\n",
1372 		    (uintmax_t)length);
1373 		return (EINVAL);
1374 	}
1375 	if (offset % di->blocksize != 0) {
1376 		printf("Attempt to write at unaligned offset %jd.\n",
1377 		    (intmax_t)offset);
1378 		return (EINVAL);
1379 	}
1380 
1381 	return (0);
1382 }
1383 
1384 #ifdef EKCD
1385 static int
1386 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1387 {
1388 
1389 	switch (kdc->kdc_encryption) {
1390 	case KERNELDUMP_ENC_AES_256_CBC:
1391 		if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1392 		    8 * size, buf) <= 0) {
1393 			return (EIO);
1394 		}
1395 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1396 		    buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1397 			return (EIO);
1398 		}
1399 		break;
1400 	case KERNELDUMP_ENC_CHACHA20:
1401 		chacha_encrypt_bytes(&kdc->kdc_chacha, buf, buf, size);
1402 		break;
1403 	default:
1404 		return (EINVAL);
1405 	}
1406 
1407 	return (0);
1408 }
1409 
1410 /* Encrypt data and call dumper. */
1411 static int
1412 dump_encrypted_write(struct dumperinfo *di, void *virtual,
1413     vm_offset_t physical, off_t offset, size_t length)
1414 {
1415 	static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1416 	struct kerneldumpcrypto *kdc;
1417 	int error;
1418 	size_t nbytes;
1419 
1420 	kdc = di->kdcrypto;
1421 
1422 	while (length > 0) {
1423 		nbytes = MIN(length, sizeof(buf));
1424 		bcopy(virtual, buf, nbytes);
1425 
1426 		if (dump_encrypt(kdc, buf, nbytes) != 0)
1427 			return (EIO);
1428 
1429 		error = dump_write(di, buf, physical, offset, nbytes);
1430 		if (error != 0)
1431 			return (error);
1432 
1433 		offset += nbytes;
1434 		virtual = (void *)((uint8_t *)virtual + nbytes);
1435 		length -= nbytes;
1436 	}
1437 
1438 	return (0);
1439 }
1440 #endif /* EKCD */
1441 
1442 static int
1443 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1444 {
1445 	struct dumperinfo *di;
1446 	size_t resid, rlength;
1447 	int error;
1448 
1449 	di = arg;
1450 
1451 	if (length % di->blocksize != 0) {
1452 		/*
1453 		 * This must be the final write after flushing the compression
1454 		 * stream. Write as many full blocks as possible and stash the
1455 		 * residual data in the dumper's block buffer. It will be
1456 		 * padded and written in dump_finish().
1457 		 */
1458 		rlength = rounddown(length, di->blocksize);
1459 		if (rlength != 0) {
1460 			error = _dump_append(di, base, 0, rlength);
1461 			if (error != 0)
1462 				return (error);
1463 		}
1464 		resid = length - rlength;
1465 		memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1466 		bzero((uint8_t *)di->blockbuf + resid, di->blocksize - resid);
1467 		di->kdcomp->kdc_resid = resid;
1468 		return (EAGAIN);
1469 	}
1470 	return (_dump_append(di, base, 0, length));
1471 }
1472 
1473 /*
1474  * Write kernel dump headers at the beginning and end of the dump extent.
1475  * Write the kernel dump encryption key after the leading header if we were
1476  * configured to do so.
1477  */
1478 static int
1479 dump_write_headers(struct dumperinfo *di, struct kerneldumpheader *kdh)
1480 {
1481 #ifdef EKCD
1482 	struct kerneldumpcrypto *kdc;
1483 #endif
1484 	void *buf, *key;
1485 	size_t hdrsz;
1486 	uint64_t extent;
1487 	uint32_t keysize;
1488 	int error;
1489 
1490 	hdrsz = sizeof(*kdh);
1491 	if (hdrsz > di->blocksize)
1492 		return (ENOMEM);
1493 
1494 #ifdef EKCD
1495 	kdc = di->kdcrypto;
1496 	key = kdc->kdc_dumpkey;
1497 	keysize = kerneldumpcrypto_dumpkeysize(kdc);
1498 #else
1499 	key = NULL;
1500 	keysize = 0;
1501 #endif
1502 
1503 	/*
1504 	 * If the dump device has special handling for headers, let it take care
1505 	 * of writing them out.
1506 	 */
1507 	if (di->dumper_hdr != NULL)
1508 		return (di->dumper_hdr(di, kdh, key, keysize));
1509 
1510 	if (hdrsz == di->blocksize)
1511 		buf = kdh;
1512 	else {
1513 		buf = di->blockbuf;
1514 		memset(buf, 0, di->blocksize);
1515 		memcpy(buf, kdh, hdrsz);
1516 	}
1517 
1518 	extent = dtoh64(kdh->dumpextent);
1519 #ifdef EKCD
1520 	if (kdc != NULL) {
1521 		error = dump_write(di, kdc->kdc_dumpkey, 0,
1522 		    di->mediaoffset + di->mediasize - di->blocksize - extent -
1523 		    keysize, keysize);
1524 		if (error != 0)
1525 			return (error);
1526 	}
1527 #endif
1528 
1529 	error = dump_write(di, buf, 0,
1530 	    di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1531 	    keysize, di->blocksize);
1532 	if (error == 0)
1533 		error = dump_write(di, buf, 0, di->mediaoffset + di->mediasize -
1534 		    di->blocksize, di->blocksize);
1535 	return (error);
1536 }
1537 
1538 /*
1539  * Don't touch the first SIZEOF_METADATA bytes on the dump device.  This is to
1540  * protect us from metadata and metadata from us.
1541  */
1542 #define	SIZEOF_METADATA		(64 * 1024)
1543 
1544 /*
1545  * Do some preliminary setup for a kernel dump: initialize state for encryption,
1546  * if requested, and make sure that we have enough space on the dump device.
1547  *
1548  * We set things up so that the dump ends before the last sector of the dump
1549  * device, at which the trailing header is written.
1550  *
1551  *     +-----------+------+-----+----------------------------+------+
1552  *     |           | lhdr | key |    ... kernel dump ...     | thdr |
1553  *     +-----------+------+-----+----------------------------+------+
1554  *                   1 blk  opt <------- dump extent --------> 1 blk
1555  *
1556  * Dumps written using dump_append() start at the beginning of the extent.
1557  * Uncompressed dumps will use the entire extent, but compressed dumps typically
1558  * will not. The true length of the dump is recorded in the leading and trailing
1559  * headers once the dump has been completed.
1560  *
1561  * The dump device may provide a callback, in which case it will initialize
1562  * dumpoff and take care of laying out the headers.
1563  */
1564 int
1565 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1566 {
1567 	uint64_t dumpextent, span;
1568 	uint32_t keysize;
1569 	int error;
1570 
1571 #ifdef EKCD
1572 	error = kerneldumpcrypto_init(di->kdcrypto);
1573 	if (error != 0)
1574 		return (error);
1575 	keysize = kerneldumpcrypto_dumpkeysize(di->kdcrypto);
1576 #else
1577 	error = 0;
1578 	keysize = 0;
1579 #endif
1580 
1581 	if (di->dumper_start != NULL) {
1582 		error = di->dumper_start(di);
1583 	} else {
1584 		dumpextent = dtoh64(kdh->dumpextent);
1585 		span = SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1586 		    keysize;
1587 		if (di->mediasize < span) {
1588 			if (di->kdcomp == NULL)
1589 				return (E2BIG);
1590 
1591 			/*
1592 			 * We don't yet know how much space the compressed dump
1593 			 * will occupy, so try to use the whole swap partition
1594 			 * (minus the first 64KB) in the hope that the
1595 			 * compressed dump will fit. If that doesn't turn out to
1596 			 * be enough, the bounds checking in dump_write()
1597 			 * will catch us and cause the dump to fail.
1598 			 */
1599 			dumpextent = di->mediasize - span + dumpextent;
1600 			kdh->dumpextent = htod64(dumpextent);
1601 		}
1602 
1603 		/*
1604 		 * The offset at which to begin writing the dump.
1605 		 */
1606 		di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1607 		    dumpextent;
1608 	}
1609 	di->origdumpoff = di->dumpoff;
1610 	return (error);
1611 }
1612 
1613 static int
1614 _dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1615     size_t length)
1616 {
1617 	int error;
1618 
1619 #ifdef EKCD
1620 	if (di->kdcrypto != NULL)
1621 		error = dump_encrypted_write(di, virtual, physical, di->dumpoff,
1622 		    length);
1623 	else
1624 #endif
1625 		error = dump_write(di, virtual, physical, di->dumpoff, length);
1626 	if (error == 0)
1627 		di->dumpoff += length;
1628 	return (error);
1629 }
1630 
1631 /*
1632  * Write to the dump device starting at dumpoff. When compression is enabled,
1633  * writes to the device will be performed using a callback that gets invoked
1634  * when the compression stream's output buffer is full.
1635  */
1636 int
1637 dump_append(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1638     size_t length)
1639 {
1640 	void *buf;
1641 
1642 	if (di->kdcomp != NULL) {
1643 		/* Bounce through a buffer to avoid CRC errors. */
1644 		if (length > di->maxiosize)
1645 			return (EINVAL);
1646 		buf = di->kdcomp->kdc_buf;
1647 		memmove(buf, virtual, length);
1648 		return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1649 	}
1650 	return (_dump_append(di, virtual, physical, length));
1651 }
1652 
1653 /*
1654  * Write to the dump device at the specified offset.
1655  */
1656 int
1657 dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical,
1658     off_t offset, size_t length)
1659 {
1660 	int error;
1661 
1662 	error = dump_check_bounds(di, offset, length);
1663 	if (error != 0)
1664 		return (error);
1665 	return (di->dumper(di->priv, virtual, physical, offset, length));
1666 }
1667 
1668 /*
1669  * Perform kernel dump finalization: flush the compression stream, if necessary,
1670  * write the leading and trailing kernel dump headers now that we know the true
1671  * length of the dump, and optionally write the encryption key following the
1672  * leading header.
1673  */
1674 int
1675 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1676 {
1677 	int error;
1678 
1679 	if (di->kdcomp != NULL) {
1680 		error = compressor_flush(di->kdcomp->kdc_stream);
1681 		if (error == EAGAIN) {
1682 			/* We have residual data in di->blockbuf. */
1683 			error = _dump_append(di, di->blockbuf, 0, di->blocksize);
1684 			if (error == 0)
1685 				/* Compensate for _dump_append()'s adjustment. */
1686 				di->dumpoff -= di->blocksize - di->kdcomp->kdc_resid;
1687 			di->kdcomp->kdc_resid = 0;
1688 		}
1689 		if (error != 0)
1690 			return (error);
1691 
1692 		/*
1693 		 * We now know the size of the compressed dump, so update the
1694 		 * header accordingly and recompute parity.
1695 		 */
1696 		kdh->dumplength = htod64(di->dumpoff - di->origdumpoff);
1697 		kdh->parity = 0;
1698 		kdh->parity = kerneldump_parity(kdh);
1699 
1700 		compressor_reset(di->kdcomp->kdc_stream);
1701 	}
1702 
1703 	error = dump_write_headers(di, kdh);
1704 	if (error != 0)
1705 		return (error);
1706 
1707 	(void)dump_write(di, NULL, 0, 0, 0);
1708 	return (0);
1709 }
1710 
1711 void
1712 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1713     const char *magic, uint32_t archver, uint64_t dumplen)
1714 {
1715 	size_t dstsize;
1716 
1717 	bzero(kdh, sizeof(*kdh));
1718 	strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1719 	strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1720 	kdh->version = htod32(KERNELDUMPVERSION);
1721 	kdh->architectureversion = htod32(archver);
1722 	kdh->dumplength = htod64(dumplen);
1723 	kdh->dumpextent = kdh->dumplength;
1724 	kdh->dumptime = htod64(time_second);
1725 #ifdef EKCD
1726 	kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1727 #else
1728 	kdh->dumpkeysize = 0;
1729 #endif
1730 	kdh->blocksize = htod32(di->blocksize);
1731 	strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1732 	dstsize = sizeof(kdh->versionstring);
1733 	if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1734 		kdh->versionstring[dstsize - 2] = '\n';
1735 	if (panicstr != NULL)
1736 		strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1737 	if (di->kdcomp != NULL)
1738 		kdh->compression = di->kdcomp->kdc_format;
1739 	kdh->parity = kerneldump_parity(kdh);
1740 }
1741 
1742 #ifdef DDB
1743 DB_SHOW_COMMAND(panic, db_show_panic)
1744 {
1745 
1746 	if (panicstr == NULL)
1747 		db_printf("panicstr not set\n");
1748 	else
1749 		db_printf("panic: %s\n", panicstr);
1750 }
1751 #endif
1752