xref: /freebsd/sys/kern/kern_shutdown.c (revision 22cf89c938886d14f5796fc49f9f020c23ea8eaf)
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 #include "opt_ddb.h"
41 #include "opt_ekcd.h"
42 #include "opt_kdb.h"
43 #include "opt_panic.h"
44 #include "opt_printf.h"
45 #include "opt_sched.h"
46 #include "opt_watchdog.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/boottrace.h>
52 #include <sys/buf.h>
53 #include <sys/conf.h>
54 #include <sys/compressor.h>
55 #include <sys/cons.h>
56 #include <sys/disk.h>
57 #include <sys/eventhandler.h>
58 #include <sys/filedesc.h>
59 #include <sys/jail.h>
60 #include <sys/kdb.h>
61 #include <sys/kernel.h>
62 #include <sys/kerneldump.h>
63 #include <sys/kthread.h>
64 #include <sys/ktr.h>
65 #include <sys/malloc.h>
66 #include <sys/mbuf.h>
67 #include <sys/mount.h>
68 #include <sys/priv.h>
69 #include <sys/proc.h>
70 #include <sys/reboot.h>
71 #include <sys/resourcevar.h>
72 #include <sys/rwlock.h>
73 #include <sys/sbuf.h>
74 #include <sys/sched.h>
75 #include <sys/smp.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/taskqueue.h>
79 #include <sys/vnode.h>
80 #include <sys/watchdog.h>
81 
82 #include <crypto/chacha20/chacha.h>
83 #include <crypto/rijndael/rijndael-api-fst.h>
84 #include <crypto/sha2/sha256.h>
85 
86 #include <ddb/ddb.h>
87 
88 #include <machine/cpu.h>
89 #include <machine/dump.h>
90 #include <machine/pcb.h>
91 #include <machine/smp.h>
92 
93 #include <security/mac/mac_framework.h>
94 
95 #include <vm/vm.h>
96 #include <vm/vm_object.h>
97 #include <vm/vm_page.h>
98 #include <vm/vm_pager.h>
99 #include <vm/swap_pager.h>
100 
101 #include <sys/signalvar.h>
102 
103 static MALLOC_DEFINE(M_DUMPER, "dumper", "dumper block buffer");
104 
105 #ifndef PANIC_REBOOT_WAIT_TIME
106 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
107 #endif
108 static int panic_reboot_wait_time = PANIC_REBOOT_WAIT_TIME;
109 SYSCTL_INT(_kern, OID_AUTO, panic_reboot_wait_time, CTLFLAG_RWTUN,
110     &panic_reboot_wait_time, 0,
111     "Seconds to wait before rebooting after a panic");
112 static int reboot_wait_time = 0;
113 SYSCTL_INT(_kern, OID_AUTO, reboot_wait_time, CTLFLAG_RWTUN,
114     &reboot_wait_time, 0,
115     "Seconds to wait before rebooting");
116 
117 /*
118  * Note that stdarg.h and the ANSI style va_start macro is used for both
119  * ANSI and traditional C compilers.
120  */
121 #include <machine/stdarg.h>
122 
123 #ifdef KDB
124 #ifdef KDB_UNATTENDED
125 int debugger_on_panic = 0;
126 #else
127 int debugger_on_panic = 1;
128 #endif
129 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic,
130     CTLFLAG_RWTUN, &debugger_on_panic, 0,
131     "Run debugger on kernel panic");
132 
133 static bool debugger_on_recursive_panic = false;
134 SYSCTL_BOOL(_debug, OID_AUTO, debugger_on_recursive_panic,
135     CTLFLAG_RWTUN, &debugger_on_recursive_panic, 0,
136     "Run debugger on recursive kernel panic");
137 
138 int debugger_on_trap = 0;
139 SYSCTL_INT(_debug, OID_AUTO, debugger_on_trap,
140     CTLFLAG_RWTUN, &debugger_on_trap, 0,
141     "Run debugger on kernel trap before panic");
142 
143 #ifdef KDB_TRACE
144 static int trace_on_panic = 1;
145 static bool trace_all_panics = true;
146 #else
147 static int trace_on_panic = 0;
148 static bool trace_all_panics = false;
149 #endif
150 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
151     CTLFLAG_RWTUN | CTLFLAG_SECURE,
152     &trace_on_panic, 0, "Print stack trace on kernel panic");
153 SYSCTL_BOOL(_debug, OID_AUTO, trace_all_panics, CTLFLAG_RWTUN,
154     &trace_all_panics, 0, "Print stack traces on secondary kernel panics");
155 #endif /* KDB */
156 
157 static int sync_on_panic = 0;
158 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN,
159 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
160 
161 static bool poweroff_on_panic = 0;
162 SYSCTL_BOOL(_kern, OID_AUTO, poweroff_on_panic, CTLFLAG_RWTUN,
163 	&poweroff_on_panic, 0, "Do a power off instead of a reboot on a panic");
164 
165 static bool powercycle_on_panic = 0;
166 SYSCTL_BOOL(_kern, OID_AUTO, powercycle_on_panic, CTLFLAG_RWTUN,
167 	&powercycle_on_panic, 0, "Do a power cycle instead of a reboot on a panic");
168 
169 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
170     "Shutdown environment");
171 
172 #ifndef DIAGNOSTIC
173 static int show_busybufs;
174 #else
175 static int show_busybufs = 1;
176 #endif
177 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
178     &show_busybufs, 0,
179     "Show busy buffers during shutdown");
180 
181 int suspend_blocked = 0;
182 SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW,
183 	&suspend_blocked, 0, "Block suspend due to a pending shutdown");
184 
185 #ifdef EKCD
186 FEATURE(ekcd, "Encrypted kernel crash dumps support");
187 
188 MALLOC_DEFINE(M_EKCD, "ekcd", "Encrypted kernel crash dumps data");
189 
190 struct kerneldumpcrypto {
191 	uint8_t			kdc_encryption;
192 	uint8_t			kdc_iv[KERNELDUMP_IV_MAX_SIZE];
193 	union {
194 		struct {
195 			keyInstance	aes_ki;
196 			cipherInstance	aes_ci;
197 		} u_aes;
198 		struct chacha_ctx	u_chacha;
199 	} u;
200 #define	kdc_ki	u.u_aes.aes_ki
201 #define	kdc_ci	u.u_aes.aes_ci
202 #define	kdc_chacha	u.u_chacha
203 	uint32_t		kdc_dumpkeysize;
204 	struct kerneldumpkey	kdc_dumpkey[];
205 };
206 #endif
207 
208 struct kerneldumpcomp {
209 	uint8_t			kdc_format;
210 	struct compressor	*kdc_stream;
211 	uint8_t			*kdc_buf;
212 	size_t			kdc_resid;
213 };
214 
215 static struct kerneldumpcomp *kerneldumpcomp_create(struct dumperinfo *di,
216 		    uint8_t compression);
217 static void	kerneldumpcomp_destroy(struct dumperinfo *di);
218 static int	kerneldumpcomp_write_cb(void *base, size_t len, off_t off, void *arg);
219 
220 static int kerneldump_gzlevel = 6;
221 SYSCTL_INT(_kern, OID_AUTO, kerneldump_gzlevel, CTLFLAG_RWTUN,
222     &kerneldump_gzlevel, 0,
223     "Kernel crash dump compression level");
224 
225 /*
226  * Variable panicstr contains argument to first call to panic; used as flag
227  * to indicate that the kernel has already called panic.
228  */
229 const char *panicstr;
230 bool __read_frequently panicked;
231 
232 int __read_mostly dumping;		/* system is dumping */
233 int rebooting;				/* system is rebooting */
234 /*
235  * Used to serialize between sysctl kern.shutdown.dumpdevname and list
236  * modifications via ioctl.
237  */
238 static struct mtx dumpconf_list_lk;
239 MTX_SYSINIT(dumper_configs, &dumpconf_list_lk, "dumper config list", MTX_DEF);
240 
241 /* Our selected dumper(s). */
242 static TAILQ_HEAD(dumpconflist, dumperinfo) dumper_configs =
243     TAILQ_HEAD_INITIALIZER(dumper_configs);
244 
245 /* Context information for dump-debuggers, saved by the dump_savectx() macro. */
246 struct pcb dumppcb;			/* Registers. */
247 lwpid_t dumptid;			/* Thread ID. */
248 
249 static struct cdevsw reroot_cdevsw = {
250      .d_version = D_VERSION,
251      .d_name    = "reroot",
252 };
253 
254 static void poweroff_wait(void *, int);
255 static void shutdown_halt(void *junk, int howto);
256 static void shutdown_panic(void *junk, int howto);
257 static void shutdown_reset(void *junk, int howto);
258 static int kern_reroot(void);
259 
260 /* register various local shutdown events */
261 static void
262 shutdown_conf(void *unused)
263 {
264 
265 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
266 	    SHUTDOWN_PRI_FIRST);
267 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
268 	    SHUTDOWN_PRI_LAST + 100);
269 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
270 	    SHUTDOWN_PRI_LAST + 100);
271 }
272 
273 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
274 
275 /*
276  * The only reason this exists is to create the /dev/reroot/ directory,
277  * used by reroot code in init(8) as a mountpoint for tmpfs.
278  */
279 static void
280 reroot_conf(void *unused)
281 {
282 	int error;
283 	struct cdev *cdev;
284 
285 	error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev,
286 	    &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot");
287 	if (error != 0) {
288 		printf("%s: failed to create device node, error %d",
289 		    __func__, error);
290 	}
291 }
292 
293 SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL);
294 
295 /*
296  * The system call that results in a reboot.
297  */
298 /* ARGSUSED */
299 int
300 sys_reboot(struct thread *td, struct reboot_args *uap)
301 {
302 	int error;
303 
304 	error = 0;
305 #ifdef MAC
306 	error = mac_system_check_reboot(td->td_ucred, uap->opt);
307 #endif
308 	if (error == 0)
309 		error = priv_check(td, PRIV_REBOOT);
310 	if (error == 0) {
311 		if (uap->opt & RB_REROOT)
312 			error = kern_reroot();
313 		else
314 			kern_reboot(uap->opt);
315 	}
316 	return (error);
317 }
318 
319 static void
320 shutdown_nice_task_fn(void *arg, int pending __unused)
321 {
322 	int howto;
323 
324 	howto = (uintptr_t)arg;
325 	/* Send a signal to init(8) and have it shutdown the world. */
326 	PROC_LOCK(initproc);
327 	if ((howto & RB_POWEROFF) != 0) {
328 		BOOTTRACE("SIGUSR2 to init(8)");
329 		kern_psignal(initproc, SIGUSR2);
330 	} else if ((howto & RB_POWERCYCLE) != 0) {
331 		BOOTTRACE("SIGWINCH to init(8)");
332 		kern_psignal(initproc, SIGWINCH);
333 	} else if ((howto & RB_HALT) != 0) {
334 		BOOTTRACE("SIGUSR1 to init(8)");
335 		kern_psignal(initproc, SIGUSR1);
336 	} else {
337 		BOOTTRACE("SIGINT to init(8)");
338 		kern_psignal(initproc, SIGINT);
339 	}
340 	PROC_UNLOCK(initproc);
341 }
342 
343 static struct task shutdown_nice_task = TASK_INITIALIZER(0,
344     &shutdown_nice_task_fn, NULL);
345 
346 /*
347  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
348  */
349 void
350 shutdown_nice(int howto)
351 {
352 
353 	if (initproc != NULL && !SCHEDULER_STOPPED()) {
354 		BOOTTRACE("shutdown initiated");
355 		shutdown_nice_task.ta_context = (void *)(uintptr_t)howto;
356 		taskqueue_enqueue(taskqueue_fast, &shutdown_nice_task);
357 	} else {
358 		/*
359 		 * No init(8) running, or scheduler would not allow it
360 		 * to run, so simply reboot.
361 		 */
362 		kern_reboot(howto | RB_NOSYNC);
363 	}
364 }
365 
366 static void
367 print_uptime(void)
368 {
369 	int f;
370 	struct timespec ts;
371 
372 	getnanouptime(&ts);
373 	printf("Uptime: ");
374 	f = 0;
375 	if (ts.tv_sec >= 86400) {
376 		printf("%ldd", (long)ts.tv_sec / 86400);
377 		ts.tv_sec %= 86400;
378 		f = 1;
379 	}
380 	if (f || ts.tv_sec >= 3600) {
381 		printf("%ldh", (long)ts.tv_sec / 3600);
382 		ts.tv_sec %= 3600;
383 		f = 1;
384 	}
385 	if (f || ts.tv_sec >= 60) {
386 		printf("%ldm", (long)ts.tv_sec / 60);
387 		ts.tv_sec %= 60;
388 		f = 1;
389 	}
390 	printf("%lds\n", (long)ts.tv_sec);
391 }
392 
393 int
394 doadump(boolean_t textdump)
395 {
396 	boolean_t coredump;
397 	int error;
398 
399 	error = 0;
400 	if (dumping)
401 		return (EBUSY);
402 	if (TAILQ_EMPTY(&dumper_configs))
403 		return (ENXIO);
404 
405 	dump_savectx();
406 	dumping++;
407 
408 	coredump = TRUE;
409 #ifdef DDB
410 	if (textdump && textdump_pending) {
411 		coredump = FALSE;
412 		textdump_dumpsys(TAILQ_FIRST(&dumper_configs));
413 	}
414 #endif
415 	if (coredump) {
416 		struct dumperinfo *di;
417 
418 		TAILQ_FOREACH(di, &dumper_configs, di_next) {
419 			error = dumpsys(di);
420 			if (error == 0)
421 				break;
422 		}
423 	}
424 
425 	dumping--;
426 	return (error);
427 }
428 
429 /*
430  * Trace the shutdown reason.
431  */
432 static void
433 reboottrace(int howto)
434 {
435 	if ((howto & RB_DUMP) != 0) {
436 		if ((howto & RB_HALT) != 0)
437 			BOOTTRACE("system panic: halting...");
438 		if ((howto & RB_POWEROFF) != 0)
439 			BOOTTRACE("system panic: powering off...");
440 		if ((howto & (RB_HALT|RB_POWEROFF)) == 0)
441 			BOOTTRACE("system panic: rebooting...");
442 	} else {
443 		if ((howto & RB_HALT) != 0)
444 			BOOTTRACE("system halting...");
445 		if ((howto & RB_POWEROFF) != 0)
446 			BOOTTRACE("system powering off...");
447 		if ((howto & (RB_HALT|RB_POWEROFF)) == 0)
448 			BOOTTRACE("system rebooting...");
449 	}
450 }
451 
452 /*
453  * kern_reboot(9): Shut down the system cleanly to prepare for reboot, halt, or
454  * power off.
455  */
456 void
457 kern_reboot(int howto)
458 {
459 	static int once = 0;
460 
461 	if (initproc != NULL && curproc != initproc)
462 		BOOTTRACE("kernel shutdown (dirty) started");
463 	else
464 		BOOTTRACE("kernel shutdown (clean) started");
465 
466 	/*
467 	 * Normal paths here don't hold Giant, but we can wind up here
468 	 * unexpectedly with it held.  Drop it now so we don't have to
469 	 * drop and pick it up elsewhere. The paths it is locking will
470 	 * never be returned to, and it is preferable to preclude
471 	 * deadlock than to lock against code that won't ever
472 	 * continue.
473 	 */
474 	while (mtx_owned(&Giant))
475 		mtx_unlock(&Giant);
476 
477 #if defined(SMP)
478 	/*
479 	 * Bind us to the first CPU so that all shutdown code runs there.  Some
480 	 * systems don't shutdown properly (i.e., ACPI power off) if we
481 	 * run on another processor.
482 	 */
483 	if (!SCHEDULER_STOPPED()) {
484 		thread_lock(curthread);
485 		sched_bind(curthread, CPU_FIRST());
486 		thread_unlock(curthread);
487 		KASSERT(PCPU_GET(cpuid) == CPU_FIRST(),
488 		    ("%s: not running on cpu 0", __func__));
489 	}
490 #endif
491 	/* We're in the process of rebooting. */
492 	rebooting = 1;
493 	reboottrace(howto);
494 
495 	/* We are out of the debugger now. */
496 	kdb_active = 0;
497 
498 	/*
499 	 * Do any callouts that should be done BEFORE syncing the filesystems.
500 	 */
501 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
502 	BOOTTRACE("shutdown pre sync complete");
503 
504 	/*
505 	 * Now sync filesystems
506 	 */
507 	if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
508 		once = 1;
509 		BOOTTRACE("bufshutdown begin");
510 		bufshutdown(show_busybufs);
511 		BOOTTRACE("bufshutdown end");
512 	}
513 
514 	print_uptime();
515 
516 	cngrab();
517 
518 	/*
519 	 * Ok, now do things that assume all filesystem activity has
520 	 * been completed.
521 	 */
522 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
523 	BOOTTRACE("shutdown post sync complete");
524 
525 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
526 		doadump(TRUE);
527 
528 	/* Now that we're going to really halt the system... */
529 	BOOTTRACE("shutdown final begin");
530 
531 	if (shutdown_trace)
532 		boottrace_dump_console();
533 
534 	EVENTHANDLER_INVOKE(shutdown_final, howto);
535 
536 	/*
537 	 * Call this directly so that reset is attempted even if shutdown
538 	 * handlers are not yet registered.
539 	 */
540 	shutdown_reset(NULL, howto);
541 
542 	for(;;) ;	/* safety against shutdown_reset not working */
543 	/* NOTREACHED */
544 }
545 
546 /*
547  * The system call that results in changing the rootfs.
548  */
549 static int
550 kern_reroot(void)
551 {
552 	struct vnode *oldrootvnode, *vp;
553 	struct mount *mp, *devmp;
554 	int error;
555 
556 	if (curproc != initproc)
557 		return (EPERM);
558 
559 	/*
560 	 * Mark the filesystem containing currently-running executable
561 	 * (the temporary copy of init(8)) busy.
562 	 */
563 	vp = curproc->p_textvp;
564 	error = vn_lock(vp, LK_SHARED);
565 	if (error != 0)
566 		return (error);
567 	mp = vp->v_mount;
568 	error = vfs_busy(mp, MBF_NOWAIT);
569 	if (error != 0) {
570 		vfs_ref(mp);
571 		VOP_UNLOCK(vp);
572 		error = vfs_busy(mp, 0);
573 		vn_lock(vp, LK_SHARED | LK_RETRY);
574 		vfs_rel(mp);
575 		if (error != 0) {
576 			VOP_UNLOCK(vp);
577 			return (ENOENT);
578 		}
579 		if (VN_IS_DOOMED(vp)) {
580 			VOP_UNLOCK(vp);
581 			vfs_unbusy(mp);
582 			return (ENOENT);
583 		}
584 	}
585 	VOP_UNLOCK(vp);
586 
587 	/*
588 	 * Remove the filesystem containing currently-running executable
589 	 * from the mount list, to prevent it from being unmounted
590 	 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
591 	 *
592 	 * Also preserve /dev - forcibly unmounting it could cause driver
593 	 * reinitialization.
594 	 */
595 
596 	vfs_ref(rootdevmp);
597 	devmp = rootdevmp;
598 	rootdevmp = NULL;
599 
600 	mtx_lock(&mountlist_mtx);
601 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
602 	TAILQ_REMOVE(&mountlist, devmp, mnt_list);
603 	mtx_unlock(&mountlist_mtx);
604 
605 	oldrootvnode = rootvnode;
606 
607 	/*
608 	 * Unmount everything except for the two filesystems preserved above.
609 	 */
610 	vfs_unmountall();
611 
612 	/*
613 	 * Add /dev back; vfs_mountroot() will move it into its new place.
614 	 */
615 	mtx_lock(&mountlist_mtx);
616 	TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
617 	mtx_unlock(&mountlist_mtx);
618 	rootdevmp = devmp;
619 	vfs_rel(rootdevmp);
620 
621 	/*
622 	 * Mount the new rootfs.
623 	 */
624 	vfs_mountroot();
625 
626 	/*
627 	 * Update all references to the old rootvnode.
628 	 */
629 	mountcheckdirs(oldrootvnode, rootvnode);
630 
631 	/*
632 	 * Add the temporary filesystem back and unbusy it.
633 	 */
634 	mtx_lock(&mountlist_mtx);
635 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
636 	mtx_unlock(&mountlist_mtx);
637 	vfs_unbusy(mp);
638 
639 	return (0);
640 }
641 
642 /*
643  * If the shutdown was a clean halt, behave accordingly.
644  */
645 static void
646 shutdown_halt(void *junk, int howto)
647 {
648 
649 	if (howto & RB_HALT) {
650 		printf("\n");
651 		printf("The operating system has halted.\n");
652 		printf("Please press any key to reboot.\n\n");
653 
654 		wdog_kern_pat(WD_TO_NEVER);
655 
656 		switch (cngetc()) {
657 		case -1:		/* No console, just die */
658 			cpu_halt();
659 			/* NOTREACHED */
660 		default:
661 			break;
662 		}
663 	}
664 }
665 
666 /*
667  * Check to see if the system panicked, pause and then reboot
668  * according to the specified delay.
669  */
670 static void
671 shutdown_panic(void *junk, int howto)
672 {
673 	int loop;
674 
675 	if (howto & RB_DUMP) {
676 		if (panic_reboot_wait_time != 0) {
677 			if (panic_reboot_wait_time != -1) {
678 				printf("Automatic reboot in %d seconds - "
679 				       "press a key on the console to abort\n",
680 					panic_reboot_wait_time);
681 				for (loop = panic_reboot_wait_time * 10;
682 				     loop > 0; --loop) {
683 					DELAY(1000 * 100); /* 1/10th second */
684 					/* Did user type a key? */
685 					if (cncheckc() != -1)
686 						break;
687 				}
688 				if (!loop)
689 					return;
690 			}
691 		} else { /* zero time specified - reboot NOW */
692 			return;
693 		}
694 		printf("--> Press a key on the console to reboot,\n");
695 		printf("--> or switch off the system now.\n");
696 		cngetc();
697 	}
698 }
699 
700 /*
701  * Everything done, now reset
702  */
703 static void
704 shutdown_reset(void *junk, int howto)
705 {
706 
707 	printf("Rebooting...\n");
708 	DELAY(reboot_wait_time * 1000000);
709 
710 	/*
711 	 * Acquiring smp_ipi_mtx here has a double effect:
712 	 * - it disables interrupts avoiding CPU0 preemption
713 	 *   by fast handlers (thus deadlocking  against other CPUs)
714 	 * - it avoids deadlocks against smp_rendezvous() or, more
715 	 *   generally, threads busy-waiting, with this spinlock held,
716 	 *   and waiting for responses by threads on other CPUs
717 	 *   (ie. smp_tlb_shootdown()).
718 	 *
719 	 * For the !SMP case it just needs to handle the former problem.
720 	 */
721 #ifdef SMP
722 	mtx_lock_spin(&smp_ipi_mtx);
723 #else
724 	spinlock_enter();
725 #endif
726 
727 	cpu_reset();
728 	/* NOTREACHED */ /* assuming reset worked */
729 }
730 
731 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
732 static int kassert_warn_only = 0;
733 #ifdef KDB
734 static int kassert_do_kdb = 0;
735 #endif
736 #ifdef KTR
737 static int kassert_do_ktr = 0;
738 #endif
739 static int kassert_do_log = 1;
740 static int kassert_log_pps_limit = 4;
741 static int kassert_log_mute_at = 0;
742 static int kassert_log_panic_at = 0;
743 static int kassert_suppress_in_panic = 0;
744 static int kassert_warnings = 0;
745 
746 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
747     "kassert options");
748 
749 #ifdef KASSERT_PANIC_OPTIONAL
750 #define KASSERT_RWTUN	CTLFLAG_RWTUN
751 #else
752 #define KASSERT_RWTUN	CTLFLAG_RDTUN
753 #endif
754 
755 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, KASSERT_RWTUN,
756     &kassert_warn_only, 0,
757     "KASSERT triggers a panic (0) or just a warning (1)");
758 
759 #ifdef KDB
760 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, KASSERT_RWTUN,
761     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
762 #endif
763 
764 #ifdef KTR
765 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, KASSERT_RWTUN,
766     &kassert_do_ktr, 0,
767     "KASSERT does a KTR, set this to the KTRMASK you want");
768 #endif
769 
770 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, KASSERT_RWTUN,
771     &kassert_do_log, 0,
772     "If warn_only is enabled, log (1) or do not log (0) assertion violations");
773 
774 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RD | CTLFLAG_STATS,
775     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
776 
777 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, KASSERT_RWTUN,
778     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
779 
780 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, KASSERT_RWTUN,
781     &kassert_log_pps_limit, 0, "limit number of log messages per second");
782 
783 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, KASSERT_RWTUN,
784     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
785 
786 SYSCTL_INT(_debug_kassert, OID_AUTO, suppress_in_panic, KASSERT_RWTUN,
787     &kassert_suppress_in_panic, 0,
788     "KASSERTs will be suppressed while handling a panic");
789 #undef KASSERT_RWTUN
790 
791 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
792 
793 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
794     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_MPSAFE, NULL, 0,
795     kassert_sysctl_kassert, "I",
796     "set to trigger a test kassert");
797 
798 static int
799 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
800 {
801 	int error, i;
802 
803 	error = sysctl_wire_old_buffer(req, sizeof(int));
804 	if (error == 0) {
805 		i = 0;
806 		error = sysctl_handle_int(oidp, &i, 0, req);
807 	}
808 	if (error != 0 || req->newptr == NULL)
809 		return (error);
810 	KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
811 	return (0);
812 }
813 
814 #ifdef KASSERT_PANIC_OPTIONAL
815 /*
816  * Called by KASSERT, this decides if we will panic
817  * or if we will log via printf and/or ktr.
818  */
819 void
820 kassert_panic(const char *fmt, ...)
821 {
822 	static char buf[256];
823 	va_list ap;
824 
825 	va_start(ap, fmt);
826 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
827 	va_end(ap);
828 
829 	/*
830 	 * If we are suppressing secondary panics, log the warning but do not
831 	 * re-enter panic/kdb.
832 	 */
833 	if (KERNEL_PANICKED() && kassert_suppress_in_panic) {
834 		if (kassert_do_log) {
835 			printf("KASSERT failed: %s\n", buf);
836 #ifdef KDB
837 			if (trace_all_panics && trace_on_panic)
838 				kdb_backtrace();
839 #endif
840 		}
841 		return;
842 	}
843 
844 	/*
845 	 * panic if we're not just warning, or if we've exceeded
846 	 * kassert_log_panic_at warnings.
847 	 */
848 	if (!kassert_warn_only ||
849 	    (kassert_log_panic_at > 0 &&
850 	     kassert_warnings >= kassert_log_panic_at)) {
851 		va_start(ap, fmt);
852 		vpanic(fmt, ap);
853 		/* NORETURN */
854 	}
855 #ifdef KTR
856 	if (kassert_do_ktr)
857 		CTR0(ktr_mask, buf);
858 #endif /* KTR */
859 	/*
860 	 * log if we've not yet met the mute limit.
861 	 */
862 	if (kassert_do_log &&
863 	    (kassert_log_mute_at == 0 ||
864 	     kassert_warnings < kassert_log_mute_at)) {
865 		static  struct timeval lasterr;
866 		static  int curerr;
867 
868 		if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
869 			printf("KASSERT failed: %s\n", buf);
870 			kdb_backtrace();
871 		}
872 	}
873 #ifdef KDB
874 	if (kassert_do_kdb) {
875 		kdb_enter(KDB_WHY_KASSERT, buf);
876 	}
877 #endif
878 	atomic_add_int(&kassert_warnings, 1);
879 }
880 #endif /* KASSERT_PANIC_OPTIONAL */
881 #endif
882 
883 /*
884  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
885  * and then reboots.  If we are called twice, then we avoid trying to sync
886  * the disks as this often leads to recursive panics.
887  */
888 void
889 panic(const char *fmt, ...)
890 {
891 	va_list ap;
892 
893 	va_start(ap, fmt);
894 	vpanic(fmt, ap);
895 }
896 
897 void
898 vpanic(const char *fmt, va_list ap)
899 {
900 #ifdef SMP
901 	cpuset_t other_cpus;
902 #endif
903 	struct thread *td = curthread;
904 	int bootopt, newpanic;
905 	static char buf[256];
906 
907 	spinlock_enter();
908 
909 #ifdef SMP
910 	/*
911 	 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
912 	 * concurrently entering panic.  Only the winner will proceed
913 	 * further.
914 	 */
915 	if (panicstr == NULL && !kdb_active) {
916 		other_cpus = all_cpus;
917 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
918 		stop_cpus_hard(other_cpus);
919 	}
920 #endif
921 
922 	/*
923 	 * Ensure that the scheduler is stopped while panicking, even if panic
924 	 * has been entered from kdb.
925 	 */
926 	td->td_stopsched = 1;
927 
928 	bootopt = RB_AUTOBOOT;
929 	newpanic = 0;
930 	if (KERNEL_PANICKED())
931 		bootopt |= RB_NOSYNC;
932 	else {
933 		bootopt |= RB_DUMP;
934 		panicstr = fmt;
935 		panicked = true;
936 		newpanic = 1;
937 	}
938 
939 	if (newpanic) {
940 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
941 		panicstr = buf;
942 		cngrab();
943 		printf("panic: %s\n", buf);
944 	} else {
945 		printf("panic: ");
946 		vprintf(fmt, ap);
947 		printf("\n");
948 	}
949 #ifdef SMP
950 	printf("cpuid = %d\n", PCPU_GET(cpuid));
951 #endif
952 	printf("time = %jd\n", (intmax_t )time_second);
953 #ifdef KDB
954 	if ((newpanic || trace_all_panics) && trace_on_panic)
955 		kdb_backtrace();
956 	if (debugger_on_panic)
957 		kdb_enter(KDB_WHY_PANIC, "panic");
958 	else if (!newpanic && debugger_on_recursive_panic)
959 		kdb_enter(KDB_WHY_PANIC, "re-panic");
960 #endif
961 	/*thread_lock(td); */
962 	td->td_flags |= TDF_INPANIC;
963 	/* thread_unlock(td); */
964 	if (!sync_on_panic)
965 		bootopt |= RB_NOSYNC;
966 	if (poweroff_on_panic)
967 		bootopt |= RB_POWEROFF;
968 	if (powercycle_on_panic)
969 		bootopt |= RB_POWERCYCLE;
970 	kern_reboot(bootopt);
971 }
972 
973 /*
974  * Support for poweroff delay.
975  *
976  * Please note that setting this delay too short might power off your machine
977  * before the write cache on your hard disk has been flushed, leading to
978  * soft-updates inconsistencies.
979  */
980 #ifndef POWEROFF_DELAY
981 # define POWEROFF_DELAY 5000
982 #endif
983 static int poweroff_delay = POWEROFF_DELAY;
984 
985 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
986     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
987 
988 static void
989 poweroff_wait(void *junk, int howto)
990 {
991 
992 	if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
993 		return;
994 	DELAY(poweroff_delay * 1000);
995 }
996 
997 /*
998  * Some system processes (e.g. syncer) need to be stopped at appropriate
999  * points in their main loops prior to a system shutdown, so that they
1000  * won't interfere with the shutdown process (e.g. by holding a disk buf
1001  * to cause sync to fail).  For each of these system processes, register
1002  * shutdown_kproc() as a handler for one of shutdown events.
1003  */
1004 static int kproc_shutdown_wait = 60;
1005 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
1006     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
1007 
1008 void
1009 kproc_shutdown(void *arg, int howto)
1010 {
1011 	struct proc *p;
1012 	int error;
1013 
1014 	if (KERNEL_PANICKED())
1015 		return;
1016 
1017 	p = (struct proc *)arg;
1018 	printf("Waiting (max %d seconds) for system process `%s' to stop... ",
1019 	    kproc_shutdown_wait, p->p_comm);
1020 	error = kproc_suspend(p, kproc_shutdown_wait * hz);
1021 
1022 	if (error == EWOULDBLOCK)
1023 		printf("timed out\n");
1024 	else
1025 		printf("done\n");
1026 }
1027 
1028 void
1029 kthread_shutdown(void *arg, int howto)
1030 {
1031 	struct thread *td;
1032 	int error;
1033 
1034 	if (KERNEL_PANICKED())
1035 		return;
1036 
1037 	td = (struct thread *)arg;
1038 	printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
1039 	    kproc_shutdown_wait, td->td_name);
1040 	error = kthread_suspend(td, kproc_shutdown_wait * hz);
1041 
1042 	if (error == EWOULDBLOCK)
1043 		printf("timed out\n");
1044 	else
1045 		printf("done\n");
1046 }
1047 
1048 static int
1049 dumpdevname_sysctl_handler(SYSCTL_HANDLER_ARGS)
1050 {
1051 	char buf[256];
1052 	struct dumperinfo *di;
1053 	struct sbuf sb;
1054 	int error;
1055 
1056 	error = sysctl_wire_old_buffer(req, 0);
1057 	if (error != 0)
1058 		return (error);
1059 
1060 	sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1061 
1062 	mtx_lock(&dumpconf_list_lk);
1063 	TAILQ_FOREACH(di, &dumper_configs, di_next) {
1064 		if (di != TAILQ_FIRST(&dumper_configs))
1065 			sbuf_putc(&sb, ',');
1066 		sbuf_cat(&sb, di->di_devname);
1067 	}
1068 	mtx_unlock(&dumpconf_list_lk);
1069 
1070 	error = sbuf_finish(&sb);
1071 	sbuf_delete(&sb);
1072 	return (error);
1073 }
1074 SYSCTL_PROC(_kern_shutdown, OID_AUTO, dumpdevname,
1075     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, &dumper_configs, 0,
1076     dumpdevname_sysctl_handler, "A",
1077     "Device(s) for kernel dumps");
1078 
1079 static int _dump_append(struct dumperinfo *di, void *virtual, size_t length);
1080 
1081 #ifdef EKCD
1082 static struct kerneldumpcrypto *
1083 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
1084     const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1085 {
1086 	struct kerneldumpcrypto *kdc;
1087 	struct kerneldumpkey *kdk;
1088 	uint32_t dumpkeysize;
1089 
1090 	dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
1091 	kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
1092 
1093 	arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
1094 
1095 	kdc->kdc_encryption = encryption;
1096 	switch (kdc->kdc_encryption) {
1097 	case KERNELDUMP_ENC_AES_256_CBC:
1098 		if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
1099 			goto failed;
1100 		break;
1101 	case KERNELDUMP_ENC_CHACHA20:
1102 		chacha_keysetup(&kdc->kdc_chacha, key, 256);
1103 		break;
1104 	default:
1105 		goto failed;
1106 	}
1107 
1108 	kdc->kdc_dumpkeysize = dumpkeysize;
1109 	kdk = kdc->kdc_dumpkey;
1110 	kdk->kdk_encryption = kdc->kdc_encryption;
1111 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1112 	kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
1113 	memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
1114 
1115 	return (kdc);
1116 failed:
1117 	zfree(kdc, M_EKCD);
1118 	return (NULL);
1119 }
1120 
1121 static int
1122 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
1123 {
1124 	uint8_t hash[SHA256_DIGEST_LENGTH];
1125 	SHA256_CTX ctx;
1126 	struct kerneldumpkey *kdk;
1127 	int error;
1128 
1129 	error = 0;
1130 
1131 	if (kdc == NULL)
1132 		return (0);
1133 
1134 	/*
1135 	 * When a user enters ddb it can write a crash dump multiple times.
1136 	 * Each time it should be encrypted using a different IV.
1137 	 */
1138 	SHA256_Init(&ctx);
1139 	SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1140 	SHA256_Final(hash, &ctx);
1141 	bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1142 
1143 	switch (kdc->kdc_encryption) {
1144 	case KERNELDUMP_ENC_AES_256_CBC:
1145 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1146 		    kdc->kdc_iv) <= 0) {
1147 			error = EINVAL;
1148 			goto out;
1149 		}
1150 		break;
1151 	case KERNELDUMP_ENC_CHACHA20:
1152 		chacha_ivsetup(&kdc->kdc_chacha, kdc->kdc_iv, NULL);
1153 		break;
1154 	default:
1155 		error = EINVAL;
1156 		goto out;
1157 	}
1158 
1159 	kdk = kdc->kdc_dumpkey;
1160 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1161 out:
1162 	explicit_bzero(hash, sizeof(hash));
1163 	return (error);
1164 }
1165 
1166 static uint32_t
1167 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
1168 {
1169 
1170 	if (kdc == NULL)
1171 		return (0);
1172 	return (kdc->kdc_dumpkeysize);
1173 }
1174 #endif /* EKCD */
1175 
1176 static struct kerneldumpcomp *
1177 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression)
1178 {
1179 	struct kerneldumpcomp *kdcomp;
1180 	int format;
1181 
1182 	switch (compression) {
1183 	case KERNELDUMP_COMP_GZIP:
1184 		format = COMPRESS_GZIP;
1185 		break;
1186 	case KERNELDUMP_COMP_ZSTD:
1187 		format = COMPRESS_ZSTD;
1188 		break;
1189 	default:
1190 		return (NULL);
1191 	}
1192 
1193 	kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO);
1194 	kdcomp->kdc_format = compression;
1195 	kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb,
1196 	    format, di->maxiosize, kerneldump_gzlevel, di);
1197 	if (kdcomp->kdc_stream == NULL) {
1198 		free(kdcomp, M_DUMPER);
1199 		return (NULL);
1200 	}
1201 	kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1202 	return (kdcomp);
1203 }
1204 
1205 static void
1206 kerneldumpcomp_destroy(struct dumperinfo *di)
1207 {
1208 	struct kerneldumpcomp *kdcomp;
1209 
1210 	kdcomp = di->kdcomp;
1211 	if (kdcomp == NULL)
1212 		return;
1213 	compressor_fini(kdcomp->kdc_stream);
1214 	zfree(kdcomp->kdc_buf, M_DUMPER);
1215 	free(kdcomp, M_DUMPER);
1216 }
1217 
1218 /*
1219  * Free a dumper. Must not be present on global list.
1220  */
1221 void
1222 dumper_destroy(struct dumperinfo *di)
1223 {
1224 
1225 	if (di == NULL)
1226 		return;
1227 
1228 	zfree(di->blockbuf, M_DUMPER);
1229 	kerneldumpcomp_destroy(di);
1230 #ifdef EKCD
1231 	zfree(di->kdcrypto, M_EKCD);
1232 #endif
1233 	zfree(di, M_DUMPER);
1234 }
1235 
1236 /*
1237  * Allocate and set up a new dumper from the provided template.
1238  */
1239 int
1240 dumper_create(const struct dumperinfo *di_template, const char *devname,
1241     const struct diocskerneldump_arg *kda, struct dumperinfo **dip)
1242 {
1243 	struct dumperinfo *newdi;
1244 	int error = 0;
1245 
1246 	if (dip == NULL)
1247 		return (EINVAL);
1248 
1249 	/* Allocate a new dumper */
1250 	newdi = malloc(sizeof(*newdi) + strlen(devname) + 1, M_DUMPER,
1251 	    M_WAITOK | M_ZERO);
1252 	memcpy(newdi, di_template, sizeof(*newdi));
1253 	newdi->blockbuf = NULL;
1254 	newdi->kdcrypto = NULL;
1255 	newdi->kdcomp = NULL;
1256 	strcpy(newdi->di_devname, devname);
1257 
1258 	if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
1259 #ifdef EKCD
1260 		newdi->kdcrypto = kerneldumpcrypto_create(newdi->blocksize,
1261 		    kda->kda_encryption, kda->kda_key,
1262 		    kda->kda_encryptedkeysize, kda->kda_encryptedkey);
1263 		if (newdi->kdcrypto == NULL) {
1264 			error = EINVAL;
1265 			goto cleanup;
1266 		}
1267 #else
1268 		error = EOPNOTSUPP;
1269 		goto cleanup;
1270 #endif
1271 	}
1272 	if (kda->kda_compression != KERNELDUMP_COMP_NONE) {
1273 #ifdef EKCD
1274 		/*
1275 		 * We can't support simultaneous unpadded block cipher
1276 		 * encryption and compression because there is no guarantee the
1277 		 * length of the compressed result is exactly a multiple of the
1278 		 * cipher block size.
1279 		 */
1280 		if (kda->kda_encryption == KERNELDUMP_ENC_AES_256_CBC) {
1281 			error = EOPNOTSUPP;
1282 			goto cleanup;
1283 		}
1284 #endif
1285 		newdi->kdcomp = kerneldumpcomp_create(newdi,
1286 		    kda->kda_compression);
1287 		if (newdi->kdcomp == NULL) {
1288 			error = EINVAL;
1289 			goto cleanup;
1290 		}
1291 	}
1292 	newdi->blockbuf = malloc(newdi->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1293 
1294 	*dip = newdi;
1295 	return (0);
1296 cleanup:
1297 	dumper_destroy(newdi);
1298 	return (error);
1299 }
1300 
1301 /*
1302  * Create a new dumper and register it in the global list.
1303  */
1304 int
1305 dumper_insert(const struct dumperinfo *di_template, const char *devname,
1306     const struct diocskerneldump_arg *kda)
1307 {
1308 	struct dumperinfo *newdi, *listdi;
1309 	bool inserted;
1310 	uint8_t index;
1311 	int error;
1312 
1313 	index = kda->kda_index;
1314 	MPASS(index != KDA_REMOVE && index != KDA_REMOVE_DEV &&
1315 	    index != KDA_REMOVE_ALL);
1316 
1317 	error = priv_check(curthread, PRIV_SETDUMPER);
1318 	if (error != 0)
1319 		return (error);
1320 
1321 	error = dumper_create(di_template, devname, kda, &newdi);
1322 	if (error != 0)
1323 		return (error);
1324 
1325 	/* Add the new configuration to the queue */
1326 	mtx_lock(&dumpconf_list_lk);
1327 	inserted = false;
1328 	TAILQ_FOREACH(listdi, &dumper_configs, di_next) {
1329 		if (index == 0) {
1330 			TAILQ_INSERT_BEFORE(listdi, newdi, di_next);
1331 			inserted = true;
1332 			break;
1333 		}
1334 		index--;
1335 	}
1336 	if (!inserted)
1337 		TAILQ_INSERT_TAIL(&dumper_configs, newdi, di_next);
1338 	mtx_unlock(&dumpconf_list_lk);
1339 
1340 	return (0);
1341 }
1342 
1343 #ifdef DDB
1344 void
1345 dumper_ddb_insert(struct dumperinfo *newdi)
1346 {
1347 	TAILQ_INSERT_HEAD(&dumper_configs, newdi, di_next);
1348 }
1349 
1350 void
1351 dumper_ddb_remove(struct dumperinfo *di)
1352 {
1353 	TAILQ_REMOVE(&dumper_configs, di, di_next);
1354 }
1355 #endif
1356 
1357 static bool
1358 dumper_config_match(const struct dumperinfo *di, const char *devname,
1359     const struct diocskerneldump_arg *kda)
1360 {
1361 	if (kda->kda_index == KDA_REMOVE_ALL)
1362 		return (true);
1363 
1364 	if (strcmp(di->di_devname, devname) != 0)
1365 		return (false);
1366 
1367 	/*
1368 	 * Allow wildcard removal of configs matching a device on g_dev_orphan.
1369 	 */
1370 	if (kda->kda_index == KDA_REMOVE_DEV)
1371 		return (true);
1372 
1373 	if (di->kdcomp != NULL) {
1374 		if (di->kdcomp->kdc_format != kda->kda_compression)
1375 			return (false);
1376 	} else if (kda->kda_compression != KERNELDUMP_COMP_NONE)
1377 		return (false);
1378 #ifdef EKCD
1379 	if (di->kdcrypto != NULL) {
1380 		if (di->kdcrypto->kdc_encryption != kda->kda_encryption)
1381 			return (false);
1382 		/*
1383 		 * Do we care to verify keys match to delete?  It seems weird
1384 		 * to expect multiple fallback dump configurations on the same
1385 		 * device that only differ in crypto key.
1386 		 */
1387 	} else
1388 #endif
1389 		if (kda->kda_encryption != KERNELDUMP_ENC_NONE)
1390 			return (false);
1391 
1392 	return (true);
1393 }
1394 
1395 /*
1396  * Remove and free the requested dumper(s) from the global list.
1397  */
1398 int
1399 dumper_remove(const char *devname, const struct diocskerneldump_arg *kda)
1400 {
1401 	struct dumperinfo *di, *sdi;
1402 	bool found;
1403 	int error;
1404 
1405 	error = priv_check(curthread, PRIV_SETDUMPER);
1406 	if (error != 0)
1407 		return (error);
1408 
1409 	/*
1410 	 * Try to find a matching configuration, and kill it.
1411 	 *
1412 	 * NULL 'kda' indicates remove any configuration matching 'devname',
1413 	 * which may remove multiple configurations in atypical configurations.
1414 	 */
1415 	found = false;
1416 	mtx_lock(&dumpconf_list_lk);
1417 	TAILQ_FOREACH_SAFE(di, &dumper_configs, di_next, sdi) {
1418 		if (dumper_config_match(di, devname, kda)) {
1419 			found = true;
1420 			TAILQ_REMOVE(&dumper_configs, di, di_next);
1421 			dumper_destroy(di);
1422 		}
1423 	}
1424 	mtx_unlock(&dumpconf_list_lk);
1425 
1426 	/* Only produce ENOENT if a more targeted match didn't match. */
1427 	if (!found && kda->kda_index == KDA_REMOVE)
1428 		return (ENOENT);
1429 	return (0);
1430 }
1431 
1432 static int
1433 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1434 {
1435 
1436 	if (di->mediasize > 0 && length != 0 && (offset < di->mediaoffset ||
1437 	    offset - di->mediaoffset + length > di->mediasize)) {
1438 		if (di->kdcomp != NULL && offset >= di->mediaoffset) {
1439 			printf(
1440 		    "Compressed dump failed to fit in device boundaries.\n");
1441 			return (E2BIG);
1442 		}
1443 
1444 		printf("Attempt to write outside dump device boundaries.\n"
1445 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1446 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
1447 		    (uintmax_t)length, (intmax_t)di->mediasize);
1448 		return (ENOSPC);
1449 	}
1450 	if (length % di->blocksize != 0) {
1451 		printf("Attempt to write partial block of length %ju.\n",
1452 		    (uintmax_t)length);
1453 		return (EINVAL);
1454 	}
1455 	if (offset % di->blocksize != 0) {
1456 		printf("Attempt to write at unaligned offset %jd.\n",
1457 		    (intmax_t)offset);
1458 		return (EINVAL);
1459 	}
1460 
1461 	return (0);
1462 }
1463 
1464 #ifdef EKCD
1465 static int
1466 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1467 {
1468 
1469 	switch (kdc->kdc_encryption) {
1470 	case KERNELDUMP_ENC_AES_256_CBC:
1471 		if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1472 		    8 * size, buf) <= 0) {
1473 			return (EIO);
1474 		}
1475 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1476 		    buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1477 			return (EIO);
1478 		}
1479 		break;
1480 	case KERNELDUMP_ENC_CHACHA20:
1481 		chacha_encrypt_bytes(&kdc->kdc_chacha, buf, buf, size);
1482 		break;
1483 	default:
1484 		return (EINVAL);
1485 	}
1486 
1487 	return (0);
1488 }
1489 
1490 /* Encrypt data and call dumper. */
1491 static int
1492 dump_encrypted_write(struct dumperinfo *di, void *virtual, off_t offset,
1493     size_t length)
1494 {
1495 	static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1496 	struct kerneldumpcrypto *kdc;
1497 	int error;
1498 	size_t nbytes;
1499 
1500 	kdc = di->kdcrypto;
1501 
1502 	while (length > 0) {
1503 		nbytes = MIN(length, sizeof(buf));
1504 		bcopy(virtual, buf, nbytes);
1505 
1506 		if (dump_encrypt(kdc, buf, nbytes) != 0)
1507 			return (EIO);
1508 
1509 		error = dump_write(di, buf, offset, nbytes);
1510 		if (error != 0)
1511 			return (error);
1512 
1513 		offset += nbytes;
1514 		virtual = (void *)((uint8_t *)virtual + nbytes);
1515 		length -= nbytes;
1516 	}
1517 
1518 	return (0);
1519 }
1520 #endif /* EKCD */
1521 
1522 static int
1523 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1524 {
1525 	struct dumperinfo *di;
1526 	size_t resid, rlength;
1527 	int error;
1528 
1529 	di = arg;
1530 
1531 	if (length % di->blocksize != 0) {
1532 		/*
1533 		 * This must be the final write after flushing the compression
1534 		 * stream. Write as many full blocks as possible and stash the
1535 		 * residual data in the dumper's block buffer. It will be
1536 		 * padded and written in dump_finish().
1537 		 */
1538 		rlength = rounddown(length, di->blocksize);
1539 		if (rlength != 0) {
1540 			error = _dump_append(di, base, rlength);
1541 			if (error != 0)
1542 				return (error);
1543 		}
1544 		resid = length - rlength;
1545 		memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1546 		bzero((uint8_t *)di->blockbuf + resid, di->blocksize - resid);
1547 		di->kdcomp->kdc_resid = resid;
1548 		return (EAGAIN);
1549 	}
1550 	return (_dump_append(di, base, length));
1551 }
1552 
1553 /*
1554  * Write kernel dump headers at the beginning and end of the dump extent.
1555  * Write the kernel dump encryption key after the leading header if we were
1556  * configured to do so.
1557  */
1558 static int
1559 dump_write_headers(struct dumperinfo *di, struct kerneldumpheader *kdh)
1560 {
1561 #ifdef EKCD
1562 	struct kerneldumpcrypto *kdc;
1563 #endif
1564 	void *buf;
1565 	size_t hdrsz;
1566 	uint64_t extent;
1567 	uint32_t keysize;
1568 	int error;
1569 
1570 	hdrsz = sizeof(*kdh);
1571 	if (hdrsz > di->blocksize)
1572 		return (ENOMEM);
1573 
1574 #ifdef EKCD
1575 	kdc = di->kdcrypto;
1576 	keysize = kerneldumpcrypto_dumpkeysize(kdc);
1577 #else
1578 	keysize = 0;
1579 #endif
1580 
1581 	/*
1582 	 * If the dump device has special handling for headers, let it take care
1583 	 * of writing them out.
1584 	 */
1585 	if (di->dumper_hdr != NULL)
1586 		return (di->dumper_hdr(di, kdh));
1587 
1588 	if (hdrsz == di->blocksize)
1589 		buf = kdh;
1590 	else {
1591 		buf = di->blockbuf;
1592 		memset(buf, 0, di->blocksize);
1593 		memcpy(buf, kdh, hdrsz);
1594 	}
1595 
1596 	extent = dtoh64(kdh->dumpextent);
1597 #ifdef EKCD
1598 	if (kdc != NULL) {
1599 		error = dump_write(di, kdc->kdc_dumpkey,
1600 		    di->mediaoffset + di->mediasize - di->blocksize - extent -
1601 		    keysize, keysize);
1602 		if (error != 0)
1603 			return (error);
1604 	}
1605 #endif
1606 
1607 	error = dump_write(di, buf,
1608 	    di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1609 	    keysize, di->blocksize);
1610 	if (error == 0)
1611 		error = dump_write(di, buf, di->mediaoffset + di->mediasize -
1612 		    di->blocksize, di->blocksize);
1613 	return (error);
1614 }
1615 
1616 /*
1617  * Don't touch the first SIZEOF_METADATA bytes on the dump device.  This is to
1618  * protect us from metadata and metadata from us.
1619  */
1620 #define	SIZEOF_METADATA		(64 * 1024)
1621 
1622 /*
1623  * Do some preliminary setup for a kernel dump: initialize state for encryption,
1624  * if requested, and make sure that we have enough space on the dump device.
1625  *
1626  * We set things up so that the dump ends before the last sector of the dump
1627  * device, at which the trailing header is written.
1628  *
1629  *     +-----------+------+-----+----------------------------+------+
1630  *     |           | lhdr | key |    ... kernel dump ...     | thdr |
1631  *     +-----------+------+-----+----------------------------+------+
1632  *                   1 blk  opt <------- dump extent --------> 1 blk
1633  *
1634  * Dumps written using dump_append() start at the beginning of the extent.
1635  * Uncompressed dumps will use the entire extent, but compressed dumps typically
1636  * will not. The true length of the dump is recorded in the leading and trailing
1637  * headers once the dump has been completed.
1638  *
1639  * The dump device may provide a callback, in which case it will initialize
1640  * dumpoff and take care of laying out the headers.
1641  */
1642 int
1643 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1644 {
1645 #ifdef EKCD
1646 	struct kerneldumpcrypto *kdc;
1647 #endif
1648 	void *key;
1649 	uint64_t dumpextent, span;
1650 	uint32_t keysize;
1651 	int error;
1652 
1653 #ifdef EKCD
1654 	/* Send the key before the dump so a partial dump is still usable. */
1655 	kdc = di->kdcrypto;
1656 	error = kerneldumpcrypto_init(kdc);
1657 	if (error != 0)
1658 		return (error);
1659 	keysize = kerneldumpcrypto_dumpkeysize(kdc);
1660 	key = keysize > 0 ? kdc->kdc_dumpkey : NULL;
1661 #else
1662 	error = 0;
1663 	keysize = 0;
1664 	key = NULL;
1665 #endif
1666 
1667 	if (di->dumper_start != NULL) {
1668 		error = di->dumper_start(di, key, keysize);
1669 	} else {
1670 		dumpextent = dtoh64(kdh->dumpextent);
1671 		span = SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1672 		    keysize;
1673 		if (di->mediasize < span) {
1674 			if (di->kdcomp == NULL)
1675 				return (E2BIG);
1676 
1677 			/*
1678 			 * We don't yet know how much space the compressed dump
1679 			 * will occupy, so try to use the whole swap partition
1680 			 * (minus the first 64KB) in the hope that the
1681 			 * compressed dump will fit. If that doesn't turn out to
1682 			 * be enough, the bounds checking in dump_write()
1683 			 * will catch us and cause the dump to fail.
1684 			 */
1685 			dumpextent = di->mediasize - span + dumpextent;
1686 			kdh->dumpextent = htod64(dumpextent);
1687 		}
1688 
1689 		/*
1690 		 * The offset at which to begin writing the dump.
1691 		 */
1692 		di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1693 		    dumpextent;
1694 	}
1695 	di->origdumpoff = di->dumpoff;
1696 	return (error);
1697 }
1698 
1699 static int
1700 _dump_append(struct dumperinfo *di, void *virtual, size_t length)
1701 {
1702 	int error;
1703 
1704 #ifdef EKCD
1705 	if (di->kdcrypto != NULL)
1706 		error = dump_encrypted_write(di, virtual, di->dumpoff, length);
1707 	else
1708 #endif
1709 		error = dump_write(di, virtual, di->dumpoff, length);
1710 	if (error == 0)
1711 		di->dumpoff += length;
1712 	return (error);
1713 }
1714 
1715 /*
1716  * Write to the dump device starting at dumpoff. When compression is enabled,
1717  * writes to the device will be performed using a callback that gets invoked
1718  * when the compression stream's output buffer is full.
1719  */
1720 int
1721 dump_append(struct dumperinfo *di, void *virtual, size_t length)
1722 {
1723 	void *buf;
1724 
1725 	if (di->kdcomp != NULL) {
1726 		/* Bounce through a buffer to avoid CRC errors. */
1727 		if (length > di->maxiosize)
1728 			return (EINVAL);
1729 		buf = di->kdcomp->kdc_buf;
1730 		memmove(buf, virtual, length);
1731 		return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1732 	}
1733 	return (_dump_append(di, virtual, length));
1734 }
1735 
1736 /*
1737  * Write to the dump device at the specified offset.
1738  */
1739 int
1740 dump_write(struct dumperinfo *di, void *virtual, off_t offset, size_t length)
1741 {
1742 	int error;
1743 
1744 	error = dump_check_bounds(di, offset, length);
1745 	if (error != 0)
1746 		return (error);
1747 	return (di->dumper(di->priv, virtual, offset, length));
1748 }
1749 
1750 /*
1751  * Perform kernel dump finalization: flush the compression stream, if necessary,
1752  * write the leading and trailing kernel dump headers now that we know the true
1753  * length of the dump, and optionally write the encryption key following the
1754  * leading header.
1755  */
1756 int
1757 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1758 {
1759 	int error;
1760 
1761 	if (di->kdcomp != NULL) {
1762 		error = compressor_flush(di->kdcomp->kdc_stream);
1763 		if (error == EAGAIN) {
1764 			/* We have residual data in di->blockbuf. */
1765 			error = _dump_append(di, di->blockbuf, di->blocksize);
1766 			if (error == 0)
1767 				/* Compensate for _dump_append()'s adjustment. */
1768 				di->dumpoff -= di->blocksize - di->kdcomp->kdc_resid;
1769 			di->kdcomp->kdc_resid = 0;
1770 		}
1771 		if (error != 0)
1772 			return (error);
1773 
1774 		/*
1775 		 * We now know the size of the compressed dump, so update the
1776 		 * header accordingly and recompute parity.
1777 		 */
1778 		kdh->dumplength = htod64(di->dumpoff - di->origdumpoff);
1779 		kdh->parity = 0;
1780 		kdh->parity = kerneldump_parity(kdh);
1781 
1782 		compressor_reset(di->kdcomp->kdc_stream);
1783 	}
1784 
1785 	error = dump_write_headers(di, kdh);
1786 	if (error != 0)
1787 		return (error);
1788 
1789 	(void)dump_write(di, NULL, 0, 0);
1790 	return (0);
1791 }
1792 
1793 void
1794 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1795     const char *magic, uint32_t archver, uint64_t dumplen)
1796 {
1797 	size_t dstsize;
1798 
1799 	bzero(kdh, sizeof(*kdh));
1800 	strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1801 	strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1802 	kdh->version = htod32(KERNELDUMPVERSION);
1803 	kdh->architectureversion = htod32(archver);
1804 	kdh->dumplength = htod64(dumplen);
1805 	kdh->dumpextent = kdh->dumplength;
1806 	kdh->dumptime = htod64(time_second);
1807 #ifdef EKCD
1808 	kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1809 #else
1810 	kdh->dumpkeysize = 0;
1811 #endif
1812 	kdh->blocksize = htod32(di->blocksize);
1813 	strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1814 	dstsize = sizeof(kdh->versionstring);
1815 	if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1816 		kdh->versionstring[dstsize - 2] = '\n';
1817 	if (panicstr != NULL)
1818 		strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1819 	if (di->kdcomp != NULL)
1820 		kdh->compression = di->kdcomp->kdc_format;
1821 	kdh->parity = kerneldump_parity(kdh);
1822 }
1823 
1824 #ifdef DDB
1825 DB_SHOW_COMMAND_FLAGS(panic, db_show_panic, DB_CMD_MEMSAFE)
1826 {
1827 
1828 	if (panicstr == NULL)
1829 		db_printf("panicstr not set\n");
1830 	else
1831 		db_printf("panic: %s\n", panicstr);
1832 }
1833 #endif
1834