xref: /freebsd-14-stable/sys/kern/kern_shutdown.c (revision 261f70ffc8a85085c4bc3632818f77e60d4d076e)
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 dumping __read_mostly;		/* system is dumping */
233 int rebooting __read_mostly;		/* 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
shutdown_conf(void * unused)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_panic, NULL,
268 	    SHUTDOWN_PRI_LAST + 100);
269 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
270 	    SHUTDOWN_PRI_LAST + 200);
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
reroot_conf(void * unused)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
sys_reboot(struct thread * td,struct reboot_args * uap)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
shutdown_nice_task_fn(void * arg,int pending __unused)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
shutdown_nice(int howto)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
print_uptime(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
doadump(boolean_t textdump)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
reboottrace(int howto)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
kern_reboot(int howto)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 (!SCHEDULER_STOPPED() && 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 	/*
496 	 * Do any callouts that should be done BEFORE syncing the filesystems.
497 	 */
498 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
499 	BOOTTRACE("shutdown pre sync complete");
500 
501 	/*
502 	 * Now sync filesystems
503 	 */
504 	if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
505 		once = 1;
506 		BOOTTRACE("bufshutdown begin");
507 		bufshutdown(show_busybufs);
508 		BOOTTRACE("bufshutdown end");
509 	}
510 
511 	print_uptime();
512 
513 	cngrab();
514 
515 	/*
516 	 * Ok, now do things that assume all filesystem activity has
517 	 * been completed.
518 	 */
519 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
520 	BOOTTRACE("shutdown post sync complete");
521 
522 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
523 		doadump(TRUE);
524 
525 	/* Now that we're going to really halt the system... */
526 	BOOTTRACE("shutdown final begin");
527 
528 	if (shutdown_trace)
529 		boottrace_dump_console();
530 
531 	EVENTHANDLER_INVOKE(shutdown_final, howto);
532 
533 	/*
534 	 * Call this directly so that reset is attempted even if shutdown
535 	 * handlers are not yet registered.
536 	 */
537 	shutdown_reset(NULL, howto);
538 
539 	for(;;) ;	/* safety against shutdown_reset not working */
540 	/* NOTREACHED */
541 }
542 
543 /*
544  * The system call that results in changing the rootfs.
545  */
546 static int
kern_reroot(void)547 kern_reroot(void)
548 {
549 	struct vnode *oldrootvnode, *vp;
550 	struct mount *mp, *devmp;
551 	int error;
552 
553 	if (curproc != initproc)
554 		return (EPERM);
555 
556 	/*
557 	 * Mark the filesystem containing currently-running executable
558 	 * (the temporary copy of init(8)) busy.
559 	 */
560 	vp = curproc->p_textvp;
561 	error = vn_lock(vp, LK_SHARED);
562 	if (error != 0)
563 		return (error);
564 	mp = vp->v_mount;
565 	error = vfs_busy(mp, MBF_NOWAIT);
566 	if (error != 0) {
567 		vfs_ref(mp);
568 		VOP_UNLOCK(vp);
569 		error = vfs_busy(mp, 0);
570 		vn_lock(vp, LK_SHARED | LK_RETRY);
571 		vfs_rel(mp);
572 		if (error != 0) {
573 			VOP_UNLOCK(vp);
574 			return (ENOENT);
575 		}
576 		if (VN_IS_DOOMED(vp)) {
577 			VOP_UNLOCK(vp);
578 			vfs_unbusy(mp);
579 			return (ENOENT);
580 		}
581 	}
582 	VOP_UNLOCK(vp);
583 
584 	/*
585 	 * Remove the filesystem containing currently-running executable
586 	 * from the mount list, to prevent it from being unmounted
587 	 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
588 	 *
589 	 * Also preserve /dev - forcibly unmounting it could cause driver
590 	 * reinitialization.
591 	 */
592 
593 	vfs_ref(rootdevmp);
594 	devmp = rootdevmp;
595 	rootdevmp = NULL;
596 
597 	mtx_lock(&mountlist_mtx);
598 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
599 	TAILQ_REMOVE(&mountlist, devmp, mnt_list);
600 	mtx_unlock(&mountlist_mtx);
601 
602 	oldrootvnode = rootvnode;
603 
604 	/*
605 	 * Unmount everything except for the two filesystems preserved above.
606 	 */
607 	vfs_unmountall();
608 
609 	/*
610 	 * Add /dev back; vfs_mountroot() will move it into its new place.
611 	 */
612 	mtx_lock(&mountlist_mtx);
613 	TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
614 	mtx_unlock(&mountlist_mtx);
615 	rootdevmp = devmp;
616 	vfs_rel(rootdevmp);
617 
618 	/*
619 	 * Mount the new rootfs.
620 	 */
621 	vfs_mountroot();
622 
623 	/*
624 	 * Update all references to the old rootvnode.
625 	 */
626 	mountcheckdirs(oldrootvnode, rootvnode);
627 
628 	/*
629 	 * Add the temporary filesystem back and unbusy it.
630 	 */
631 	mtx_lock(&mountlist_mtx);
632 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
633 	mtx_unlock(&mountlist_mtx);
634 	vfs_unbusy(mp);
635 
636 	return (0);
637 }
638 
639 /*
640  * If the shutdown was a clean halt, behave accordingly.
641  */
642 static void
shutdown_halt(void * junk,int howto)643 shutdown_halt(void *junk, int howto)
644 {
645 
646 	if (howto & RB_HALT) {
647 		printf("\n");
648 		printf("The operating system has halted.\n");
649 		printf("Please press any key to reboot.\n\n");
650 
651 		wdog_kern_pat(WD_TO_NEVER);
652 
653 		switch (cngetc()) {
654 		case -1:		/* No console, just die */
655 			cpu_halt();
656 			/* NOTREACHED */
657 		default:
658 			break;
659 		}
660 	}
661 }
662 
663 /*
664  * Check to see if the system panicked, pause and then reboot
665  * according to the specified delay.
666  */
667 static void
shutdown_panic(void * junk,int howto)668 shutdown_panic(void *junk, int howto)
669 {
670 	int loop;
671 
672 	if (howto & RB_DUMP) {
673 		if (panic_reboot_wait_time != 0) {
674 			if (panic_reboot_wait_time != -1) {
675 				printf("Automatic reboot in %d seconds - "
676 				       "press a key on the console to abort\n",
677 					panic_reboot_wait_time);
678 				for (loop = panic_reboot_wait_time * 10;
679 				     loop > 0; --loop) {
680 					DELAY(1000 * 100); /* 1/10th second */
681 					/* Did user type a key? */
682 					if (cncheckc() != -1)
683 						break;
684 				}
685 				if (!loop)
686 					return;
687 			}
688 		} else { /* zero time specified - reboot NOW */
689 			return;
690 		}
691 		printf("--> Press a key on the console to reboot,\n");
692 		printf("--> or switch off the system now.\n");
693 		cngetc();
694 	}
695 }
696 
697 /*
698  * Everything done, now reset
699  */
700 static void
shutdown_reset(void * junk,int howto)701 shutdown_reset(void *junk, int howto)
702 {
703 
704 	printf("Rebooting...\n");
705 	DELAY(reboot_wait_time * 1000000);
706 
707 	/*
708 	 * Acquiring smp_ipi_mtx here has a double effect:
709 	 * - it disables interrupts avoiding CPU0 preemption
710 	 *   by fast handlers (thus deadlocking  against other CPUs)
711 	 * - it avoids deadlocks against smp_rendezvous() or, more
712 	 *   generally, threads busy-waiting, with this spinlock held,
713 	 *   and waiting for responses by threads on other CPUs
714 	 *   (ie. smp_tlb_shootdown()).
715 	 *
716 	 * For the !SMP case it just needs to handle the former problem.
717 	 */
718 #ifdef SMP
719 	mtx_lock_spin(&smp_ipi_mtx);
720 #else
721 	spinlock_enter();
722 #endif
723 
724 	cpu_reset();
725 	/* NOTREACHED */ /* assuming reset worked */
726 }
727 
728 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
729 static int kassert_warn_only = 0;
730 #ifdef KDB
731 static int kassert_do_kdb = 0;
732 #endif
733 #ifdef KTR
734 static int kassert_do_ktr = 0;
735 #endif
736 static int kassert_do_log = 1;
737 static int kassert_log_pps_limit = 4;
738 static int kassert_log_mute_at = 0;
739 static int kassert_log_panic_at = 0;
740 static int kassert_suppress_in_panic = 0;
741 static int kassert_warnings = 0;
742 
743 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
744     "kassert options");
745 
746 #ifdef KASSERT_PANIC_OPTIONAL
747 #define KASSERT_RWTUN	CTLFLAG_RWTUN
748 #else
749 #define KASSERT_RWTUN	CTLFLAG_RDTUN
750 #endif
751 
752 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, KASSERT_RWTUN,
753     &kassert_warn_only, 0,
754     "KASSERT triggers a panic (0) or just a warning (1)");
755 
756 #ifdef KDB
757 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, KASSERT_RWTUN,
758     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
759 #endif
760 
761 #ifdef KTR
762 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, KASSERT_RWTUN,
763     &kassert_do_ktr, 0,
764     "KASSERT does a KTR, set this to the KTRMASK you want");
765 #endif
766 
767 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, KASSERT_RWTUN,
768     &kassert_do_log, 0,
769     "If warn_only is enabled, log (1) or do not log (0) assertion violations");
770 
771 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RD | CTLFLAG_STATS,
772     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
773 
774 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, KASSERT_RWTUN,
775     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
776 
777 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, KASSERT_RWTUN,
778     &kassert_log_pps_limit, 0, "limit number of log messages per second");
779 
780 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, KASSERT_RWTUN,
781     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
782 
783 SYSCTL_INT(_debug_kassert, OID_AUTO, suppress_in_panic, KASSERT_RWTUN,
784     &kassert_suppress_in_panic, 0,
785     "KASSERTs will be suppressed while handling a panic");
786 #undef KASSERT_RWTUN
787 
788 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
789 
790 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
791     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_MPSAFE, NULL, 0,
792     kassert_sysctl_kassert, "I",
793     "set to trigger a test kassert");
794 
795 static int
kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)796 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
797 {
798 	int error, i;
799 
800 	error = sysctl_wire_old_buffer(req, sizeof(int));
801 	if (error == 0) {
802 		i = 0;
803 		error = sysctl_handle_int(oidp, &i, 0, req);
804 	}
805 	if (error != 0 || req->newptr == NULL)
806 		return (error);
807 	KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
808 	return (0);
809 }
810 
811 #ifdef KASSERT_PANIC_OPTIONAL
812 /*
813  * Called by KASSERT, this decides if we will panic
814  * or if we will log via printf and/or ktr.
815  */
816 void
kassert_panic(const char * fmt,...)817 kassert_panic(const char *fmt, ...)
818 {
819 	static char buf[256];
820 	va_list ap;
821 
822 	va_start(ap, fmt);
823 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
824 	va_end(ap);
825 
826 	/*
827 	 * If we are suppressing secondary panics, log the warning but do not
828 	 * re-enter panic/kdb.
829 	 */
830 	if (KERNEL_PANICKED() && kassert_suppress_in_panic) {
831 		if (kassert_do_log) {
832 			printf("KASSERT failed: %s\n", buf);
833 #ifdef KDB
834 			if (trace_all_panics && trace_on_panic)
835 				kdb_backtrace();
836 #endif
837 		}
838 		return;
839 	}
840 
841 	/*
842 	 * panic if we're not just warning, or if we've exceeded
843 	 * kassert_log_panic_at warnings.
844 	 */
845 	if (!kassert_warn_only ||
846 	    (kassert_log_panic_at > 0 &&
847 	     kassert_warnings >= kassert_log_panic_at)) {
848 		va_start(ap, fmt);
849 		vpanic(fmt, ap);
850 		/* NORETURN */
851 	}
852 #ifdef KTR
853 	if (kassert_do_ktr)
854 		CTR0(ktr_mask, buf);
855 #endif /* KTR */
856 	/*
857 	 * log if we've not yet met the mute limit.
858 	 */
859 	if (kassert_do_log &&
860 	    (kassert_log_mute_at == 0 ||
861 	     kassert_warnings < kassert_log_mute_at)) {
862 		static  struct timeval lasterr;
863 		static  int curerr;
864 
865 		if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
866 			printf("KASSERT failed: %s\n", buf);
867 			kdb_backtrace();
868 		}
869 	}
870 #ifdef KDB
871 	if (kassert_do_kdb) {
872 		kdb_enter(KDB_WHY_KASSERT, buf);
873 	}
874 #endif
875 	atomic_add_int(&kassert_warnings, 1);
876 }
877 #endif /* KASSERT_PANIC_OPTIONAL */
878 #endif
879 
880 /*
881  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
882  * and then reboots.  If we are called twice, then we avoid trying to sync
883  * the disks as this often leads to recursive panics.
884  */
885 void
panic(const char * fmt,...)886 panic(const char *fmt, ...)
887 {
888 	va_list ap;
889 
890 	va_start(ap, fmt);
891 	vpanic(fmt, ap);
892 }
893 
894 void
vpanic(const char * fmt,va_list ap)895 vpanic(const char *fmt, va_list ap)
896 {
897 #ifdef SMP
898 	cpuset_t other_cpus;
899 #endif
900 	struct thread *td = curthread;
901 	int bootopt, newpanic;
902 	static char buf[256];
903 
904 	spinlock_enter();
905 
906 #ifdef SMP
907 	/*
908 	 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
909 	 * concurrently entering panic.  Only the winner will proceed
910 	 * further.
911 	 */
912 	if (panicstr == NULL && !kdb_active) {
913 		other_cpus = all_cpus;
914 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
915 		stop_cpus_hard(other_cpus);
916 	}
917 #endif
918 
919 	/*
920 	 * Ensure that the scheduler is stopped while panicking, even if panic
921 	 * has been entered from kdb.
922 	 */
923 	td->td_stopsched = 1;
924 
925 	bootopt = RB_AUTOBOOT;
926 	newpanic = 0;
927 	if (KERNEL_PANICKED())
928 		bootopt |= RB_NOSYNC;
929 	else {
930 		bootopt |= RB_DUMP;
931 		panicstr = fmt;
932 		panicked = true;
933 		newpanic = 1;
934 	}
935 
936 	if (newpanic) {
937 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
938 		panicstr = buf;
939 		cngrab();
940 		printf("panic: %s\n", buf);
941 	} else {
942 		printf("panic: ");
943 		vprintf(fmt, ap);
944 		printf("\n");
945 	}
946 #ifdef SMP
947 	printf("cpuid = %d\n", PCPU_GET(cpuid));
948 #endif
949 	printf("time = %jd\n", (intmax_t )time_second);
950 #ifdef KDB
951 	if ((newpanic || trace_all_panics) && trace_on_panic)
952 		kdb_backtrace();
953 	if (debugger_on_panic)
954 		kdb_enter(KDB_WHY_PANIC, "panic");
955 	else if (!newpanic && debugger_on_recursive_panic)
956 		kdb_enter(KDB_WHY_PANIC, "re-panic");
957 #endif
958 	/*thread_lock(td); */
959 	td->td_flags |= TDF_INPANIC;
960 	/* thread_unlock(td); */
961 	if (!sync_on_panic)
962 		bootopt |= RB_NOSYNC;
963 	if (poweroff_on_panic)
964 		bootopt |= RB_POWEROFF;
965 	if (powercycle_on_panic)
966 		bootopt |= RB_POWERCYCLE;
967 	kern_reboot(bootopt);
968 }
969 
970 /*
971  * Support for poweroff delay.
972  *
973  * Please note that setting this delay too short might power off your machine
974  * before the write cache on your hard disk has been flushed, leading to
975  * soft-updates inconsistencies.
976  */
977 #ifndef POWEROFF_DELAY
978 # define POWEROFF_DELAY 5000
979 #endif
980 static int poweroff_delay = POWEROFF_DELAY;
981 
982 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
983     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
984 
985 static void
poweroff_wait(void * junk,int howto)986 poweroff_wait(void *junk, int howto)
987 {
988 
989 	if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
990 		return;
991 	DELAY(poweroff_delay * 1000);
992 }
993 
994 /*
995  * Some system processes (e.g. syncer) need to be stopped at appropriate
996  * points in their main loops prior to a system shutdown, so that they
997  * won't interfere with the shutdown process (e.g. by holding a disk buf
998  * to cause sync to fail).  For each of these system processes, register
999  * shutdown_kproc() as a handler for one of shutdown events.
1000  */
1001 static int kproc_shutdown_wait = 60;
1002 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
1003     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
1004 
1005 void
kproc_shutdown(void * arg,int howto)1006 kproc_shutdown(void *arg, int howto)
1007 {
1008 	struct proc *p;
1009 	int error;
1010 
1011 	if (SCHEDULER_STOPPED())
1012 		return;
1013 
1014 	p = (struct proc *)arg;
1015 	printf("Waiting (max %d seconds) for system process `%s' to stop... ",
1016 	    kproc_shutdown_wait, p->p_comm);
1017 	error = kproc_suspend(p, kproc_shutdown_wait * hz);
1018 
1019 	if (error == EWOULDBLOCK)
1020 		printf("timed out\n");
1021 	else
1022 		printf("done\n");
1023 }
1024 
1025 void
kthread_shutdown(void * arg,int howto)1026 kthread_shutdown(void *arg, int howto)
1027 {
1028 	struct thread *td;
1029 	int error;
1030 
1031 	if (SCHEDULER_STOPPED())
1032 		return;
1033 
1034 	td = (struct thread *)arg;
1035 	printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
1036 	    kproc_shutdown_wait, td->td_name);
1037 	error = kthread_suspend(td, kproc_shutdown_wait * hz);
1038 
1039 	if (error == EWOULDBLOCK)
1040 		printf("timed out\n");
1041 	else
1042 		printf("done\n");
1043 }
1044 
1045 static int
dumpdevname_sysctl_handler(SYSCTL_HANDLER_ARGS)1046 dumpdevname_sysctl_handler(SYSCTL_HANDLER_ARGS)
1047 {
1048 	char buf[256];
1049 	struct dumperinfo *di;
1050 	struct sbuf sb;
1051 	int error;
1052 
1053 	error = sysctl_wire_old_buffer(req, 0);
1054 	if (error != 0)
1055 		return (error);
1056 
1057 	sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1058 
1059 	mtx_lock(&dumpconf_list_lk);
1060 	TAILQ_FOREACH(di, &dumper_configs, di_next) {
1061 		if (di != TAILQ_FIRST(&dumper_configs))
1062 			sbuf_putc(&sb, ',');
1063 		sbuf_cat(&sb, di->di_devname);
1064 	}
1065 	mtx_unlock(&dumpconf_list_lk);
1066 
1067 	error = sbuf_finish(&sb);
1068 	sbuf_delete(&sb);
1069 	return (error);
1070 }
1071 SYSCTL_PROC(_kern_shutdown, OID_AUTO, dumpdevname,
1072     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, &dumper_configs, 0,
1073     dumpdevname_sysctl_handler, "A",
1074     "Device(s) for kernel dumps");
1075 
1076 static int _dump_append(struct dumperinfo *di, void *virtual, size_t length);
1077 
1078 #ifdef EKCD
1079 static struct kerneldumpcrypto *
kerneldumpcrypto_create(size_t blocksize,uint8_t encryption,const uint8_t * key,uint32_t encryptedkeysize,const uint8_t * encryptedkey)1080 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
1081     const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1082 {
1083 	struct kerneldumpcrypto *kdc;
1084 	struct kerneldumpkey *kdk;
1085 	uint32_t dumpkeysize;
1086 
1087 	dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
1088 	kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
1089 
1090 	arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
1091 
1092 	kdc->kdc_encryption = encryption;
1093 	switch (kdc->kdc_encryption) {
1094 	case KERNELDUMP_ENC_AES_256_CBC:
1095 		if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
1096 			goto failed;
1097 		break;
1098 	case KERNELDUMP_ENC_CHACHA20:
1099 		chacha_keysetup(&kdc->kdc_chacha, key, 256);
1100 		break;
1101 	default:
1102 		goto failed;
1103 	}
1104 
1105 	kdc->kdc_dumpkeysize = dumpkeysize;
1106 	kdk = kdc->kdc_dumpkey;
1107 	kdk->kdk_encryption = kdc->kdc_encryption;
1108 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1109 	kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
1110 	memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
1111 
1112 	return (kdc);
1113 failed:
1114 	zfree(kdc, M_EKCD);
1115 	return (NULL);
1116 }
1117 
1118 static int
kerneldumpcrypto_init(struct kerneldumpcrypto * kdc)1119 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
1120 {
1121 	uint8_t hash[SHA256_DIGEST_LENGTH];
1122 	SHA256_CTX ctx;
1123 	struct kerneldumpkey *kdk;
1124 	int error;
1125 
1126 	error = 0;
1127 
1128 	if (kdc == NULL)
1129 		return (0);
1130 
1131 	/*
1132 	 * When a user enters ddb it can write a crash dump multiple times.
1133 	 * Each time it should be encrypted using a different IV.
1134 	 */
1135 	SHA256_Init(&ctx);
1136 	SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1137 	SHA256_Final(hash, &ctx);
1138 	bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1139 
1140 	switch (kdc->kdc_encryption) {
1141 	case KERNELDUMP_ENC_AES_256_CBC:
1142 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1143 		    kdc->kdc_iv) <= 0) {
1144 			error = EINVAL;
1145 			goto out;
1146 		}
1147 		break;
1148 	case KERNELDUMP_ENC_CHACHA20:
1149 		chacha_ivsetup(&kdc->kdc_chacha, kdc->kdc_iv, NULL);
1150 		break;
1151 	default:
1152 		error = EINVAL;
1153 		goto out;
1154 	}
1155 
1156 	kdk = kdc->kdc_dumpkey;
1157 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1158 out:
1159 	explicit_bzero(hash, sizeof(hash));
1160 	return (error);
1161 }
1162 
1163 static uint32_t
kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto * kdc)1164 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
1165 {
1166 
1167 	if (kdc == NULL)
1168 		return (0);
1169 	return (kdc->kdc_dumpkeysize);
1170 }
1171 #endif /* EKCD */
1172 
1173 static struct kerneldumpcomp *
kerneldumpcomp_create(struct dumperinfo * di,uint8_t compression)1174 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression)
1175 {
1176 	struct kerneldumpcomp *kdcomp;
1177 	int format;
1178 
1179 	switch (compression) {
1180 	case KERNELDUMP_COMP_GZIP:
1181 		format = COMPRESS_GZIP;
1182 		break;
1183 	case KERNELDUMP_COMP_ZSTD:
1184 		format = COMPRESS_ZSTD;
1185 		break;
1186 	default:
1187 		return (NULL);
1188 	}
1189 
1190 	kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO);
1191 	kdcomp->kdc_format = compression;
1192 	kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb,
1193 	    format, di->maxiosize, kerneldump_gzlevel, di);
1194 	if (kdcomp->kdc_stream == NULL) {
1195 		free(kdcomp, M_DUMPER);
1196 		return (NULL);
1197 	}
1198 	kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1199 	return (kdcomp);
1200 }
1201 
1202 static void
kerneldumpcomp_destroy(struct dumperinfo * di)1203 kerneldumpcomp_destroy(struct dumperinfo *di)
1204 {
1205 	struct kerneldumpcomp *kdcomp;
1206 
1207 	kdcomp = di->kdcomp;
1208 	if (kdcomp == NULL)
1209 		return;
1210 	compressor_fini(kdcomp->kdc_stream);
1211 	zfree(kdcomp->kdc_buf, M_DUMPER);
1212 	free(kdcomp, M_DUMPER);
1213 }
1214 
1215 /*
1216  * Free a dumper. Must not be present on global list.
1217  */
1218 void
dumper_destroy(struct dumperinfo * di)1219 dumper_destroy(struct dumperinfo *di)
1220 {
1221 
1222 	if (di == NULL)
1223 		return;
1224 
1225 	zfree(di->blockbuf, M_DUMPER);
1226 	kerneldumpcomp_destroy(di);
1227 #ifdef EKCD
1228 	zfree(di->kdcrypto, M_EKCD);
1229 #endif
1230 	zfree(di, M_DUMPER);
1231 }
1232 
1233 /*
1234  * Allocate and set up a new dumper from the provided template.
1235  */
1236 int
dumper_create(const struct dumperinfo * di_template,const char * devname,const struct diocskerneldump_arg * kda,struct dumperinfo ** dip)1237 dumper_create(const struct dumperinfo *di_template, const char *devname,
1238     const struct diocskerneldump_arg *kda, struct dumperinfo **dip)
1239 {
1240 	struct dumperinfo *newdi;
1241 	int error = 0;
1242 
1243 	if (dip == NULL)
1244 		return (EINVAL);
1245 
1246 	/* Allocate a new dumper */
1247 	newdi = malloc(sizeof(*newdi) + strlen(devname) + 1, M_DUMPER,
1248 	    M_WAITOK | M_ZERO);
1249 	memcpy(newdi, di_template, sizeof(*newdi));
1250 	newdi->blockbuf = NULL;
1251 	newdi->kdcrypto = NULL;
1252 	newdi->kdcomp = NULL;
1253 	strcpy(newdi->di_devname, devname);
1254 
1255 	if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
1256 #ifdef EKCD
1257 		newdi->kdcrypto = kerneldumpcrypto_create(newdi->blocksize,
1258 		    kda->kda_encryption, kda->kda_key,
1259 		    kda->kda_encryptedkeysize, kda->kda_encryptedkey);
1260 		if (newdi->kdcrypto == NULL) {
1261 			error = EINVAL;
1262 			goto cleanup;
1263 		}
1264 #else
1265 		error = EOPNOTSUPP;
1266 		goto cleanup;
1267 #endif
1268 	}
1269 	if (kda->kda_compression != KERNELDUMP_COMP_NONE) {
1270 #ifdef EKCD
1271 		/*
1272 		 * We can't support simultaneous unpadded block cipher
1273 		 * encryption and compression because there is no guarantee the
1274 		 * length of the compressed result is exactly a multiple of the
1275 		 * cipher block size.
1276 		 */
1277 		if (kda->kda_encryption == KERNELDUMP_ENC_AES_256_CBC) {
1278 			error = EOPNOTSUPP;
1279 			goto cleanup;
1280 		}
1281 #endif
1282 		newdi->kdcomp = kerneldumpcomp_create(newdi,
1283 		    kda->kda_compression);
1284 		if (newdi->kdcomp == NULL) {
1285 			error = EINVAL;
1286 			goto cleanup;
1287 		}
1288 	}
1289 	newdi->blockbuf = malloc(newdi->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1290 
1291 	*dip = newdi;
1292 	return (0);
1293 cleanup:
1294 	dumper_destroy(newdi);
1295 	return (error);
1296 }
1297 
1298 /*
1299  * Create a new dumper and register it in the global list.
1300  */
1301 int
dumper_insert(const struct dumperinfo * di_template,const char * devname,const struct diocskerneldump_arg * kda)1302 dumper_insert(const struct dumperinfo *di_template, const char *devname,
1303     const struct diocskerneldump_arg *kda)
1304 {
1305 	struct dumperinfo *newdi, *listdi;
1306 	bool inserted;
1307 	uint8_t index;
1308 	int error;
1309 
1310 	index = kda->kda_index;
1311 	MPASS(index != KDA_REMOVE && index != KDA_REMOVE_DEV &&
1312 	    index != KDA_REMOVE_ALL);
1313 
1314 	error = priv_check(curthread, PRIV_SETDUMPER);
1315 	if (error != 0)
1316 		return (error);
1317 
1318 	error = dumper_create(di_template, devname, kda, &newdi);
1319 	if (error != 0)
1320 		return (error);
1321 
1322 	/* Add the new configuration to the queue */
1323 	mtx_lock(&dumpconf_list_lk);
1324 	inserted = false;
1325 	TAILQ_FOREACH(listdi, &dumper_configs, di_next) {
1326 		if (index == 0) {
1327 			TAILQ_INSERT_BEFORE(listdi, newdi, di_next);
1328 			inserted = true;
1329 			break;
1330 		}
1331 		index--;
1332 	}
1333 	if (!inserted)
1334 		TAILQ_INSERT_TAIL(&dumper_configs, newdi, di_next);
1335 	mtx_unlock(&dumpconf_list_lk);
1336 
1337 	return (0);
1338 }
1339 
1340 #ifdef DDB
1341 void
dumper_ddb_insert(struct dumperinfo * newdi)1342 dumper_ddb_insert(struct dumperinfo *newdi)
1343 {
1344 	TAILQ_INSERT_HEAD(&dumper_configs, newdi, di_next);
1345 }
1346 
1347 void
dumper_ddb_remove(struct dumperinfo * di)1348 dumper_ddb_remove(struct dumperinfo *di)
1349 {
1350 	TAILQ_REMOVE(&dumper_configs, di, di_next);
1351 }
1352 #endif
1353 
1354 static bool
dumper_config_match(const struct dumperinfo * di,const char * devname,const struct diocskerneldump_arg * kda)1355 dumper_config_match(const struct dumperinfo *di, const char *devname,
1356     const struct diocskerneldump_arg *kda)
1357 {
1358 	if (kda->kda_index == KDA_REMOVE_ALL)
1359 		return (true);
1360 
1361 	if (strcmp(di->di_devname, devname) != 0)
1362 		return (false);
1363 
1364 	/*
1365 	 * Allow wildcard removal of configs matching a device on g_dev_orphan.
1366 	 */
1367 	if (kda->kda_index == KDA_REMOVE_DEV)
1368 		return (true);
1369 
1370 	if (di->kdcomp != NULL) {
1371 		if (di->kdcomp->kdc_format != kda->kda_compression)
1372 			return (false);
1373 	} else if (kda->kda_compression != KERNELDUMP_COMP_NONE)
1374 		return (false);
1375 #ifdef EKCD
1376 	if (di->kdcrypto != NULL) {
1377 		if (di->kdcrypto->kdc_encryption != kda->kda_encryption)
1378 			return (false);
1379 		/*
1380 		 * Do we care to verify keys match to delete?  It seems weird
1381 		 * to expect multiple fallback dump configurations on the same
1382 		 * device that only differ in crypto key.
1383 		 */
1384 	} else
1385 #endif
1386 		if (kda->kda_encryption != KERNELDUMP_ENC_NONE)
1387 			return (false);
1388 
1389 	return (true);
1390 }
1391 
1392 /*
1393  * Remove and free the requested dumper(s) from the global list.
1394  */
1395 int
dumper_remove(const char * devname,const struct diocskerneldump_arg * kda)1396 dumper_remove(const char *devname, const struct diocskerneldump_arg *kda)
1397 {
1398 	struct dumperinfo *di, *sdi;
1399 	bool found;
1400 	int error;
1401 
1402 	error = priv_check(curthread, PRIV_SETDUMPER);
1403 	if (error != 0)
1404 		return (error);
1405 
1406 	/*
1407 	 * Try to find a matching configuration, and kill it.
1408 	 *
1409 	 * NULL 'kda' indicates remove any configuration matching 'devname',
1410 	 * which may remove multiple configurations in atypical configurations.
1411 	 */
1412 	found = false;
1413 	mtx_lock(&dumpconf_list_lk);
1414 	TAILQ_FOREACH_SAFE(di, &dumper_configs, di_next, sdi) {
1415 		if (dumper_config_match(di, devname, kda)) {
1416 			found = true;
1417 			TAILQ_REMOVE(&dumper_configs, di, di_next);
1418 			dumper_destroy(di);
1419 		}
1420 	}
1421 	mtx_unlock(&dumpconf_list_lk);
1422 
1423 	/* Only produce ENOENT if a more targeted match didn't match. */
1424 	if (!found && kda->kda_index == KDA_REMOVE)
1425 		return (ENOENT);
1426 	return (0);
1427 }
1428 
1429 static int
dump_check_bounds(struct dumperinfo * di,off_t offset,size_t length)1430 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1431 {
1432 
1433 	if (di->mediasize > 0 && length != 0 && (offset < di->mediaoffset ||
1434 	    offset - di->mediaoffset + length > di->mediasize)) {
1435 		if (di->kdcomp != NULL && offset >= di->mediaoffset) {
1436 			printf(
1437 		    "Compressed dump failed to fit in device boundaries.\n");
1438 			return (E2BIG);
1439 		}
1440 
1441 		printf("Attempt to write outside dump device boundaries.\n"
1442 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1443 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
1444 		    (uintmax_t)length, (intmax_t)di->mediasize);
1445 		return (ENOSPC);
1446 	}
1447 	if (length % di->blocksize != 0) {
1448 		printf("Attempt to write partial block of length %ju.\n",
1449 		    (uintmax_t)length);
1450 		return (EINVAL);
1451 	}
1452 	if (offset % di->blocksize != 0) {
1453 		printf("Attempt to write at unaligned offset %jd.\n",
1454 		    (intmax_t)offset);
1455 		return (EINVAL);
1456 	}
1457 
1458 	return (0);
1459 }
1460 
1461 #ifdef EKCD
1462 static int
dump_encrypt(struct kerneldumpcrypto * kdc,uint8_t * buf,size_t size)1463 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1464 {
1465 
1466 	switch (kdc->kdc_encryption) {
1467 	case KERNELDUMP_ENC_AES_256_CBC:
1468 		if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1469 		    8 * size, buf) <= 0) {
1470 			return (EIO);
1471 		}
1472 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1473 		    buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1474 			return (EIO);
1475 		}
1476 		break;
1477 	case KERNELDUMP_ENC_CHACHA20:
1478 		chacha_encrypt_bytes(&kdc->kdc_chacha, buf, buf, size);
1479 		break;
1480 	default:
1481 		return (EINVAL);
1482 	}
1483 
1484 	return (0);
1485 }
1486 
1487 /* Encrypt data and call dumper. */
1488 static int
dump_encrypted_write(struct dumperinfo * di,void * virtual,off_t offset,size_t length)1489 dump_encrypted_write(struct dumperinfo *di, void *virtual, off_t offset,
1490     size_t length)
1491 {
1492 	static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1493 	struct kerneldumpcrypto *kdc;
1494 	int error;
1495 	size_t nbytes;
1496 
1497 	kdc = di->kdcrypto;
1498 
1499 	while (length > 0) {
1500 		nbytes = MIN(length, sizeof(buf));
1501 		bcopy(virtual, buf, nbytes);
1502 
1503 		if (dump_encrypt(kdc, buf, nbytes) != 0)
1504 			return (EIO);
1505 
1506 		error = dump_write(di, buf, offset, nbytes);
1507 		if (error != 0)
1508 			return (error);
1509 
1510 		offset += nbytes;
1511 		virtual = (void *)((uint8_t *)virtual + nbytes);
1512 		length -= nbytes;
1513 	}
1514 
1515 	return (0);
1516 }
1517 #endif /* EKCD */
1518 
1519 static int
kerneldumpcomp_write_cb(void * base,size_t length,off_t offset,void * arg)1520 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1521 {
1522 	struct dumperinfo *di;
1523 	size_t resid, rlength;
1524 	int error;
1525 
1526 	di = arg;
1527 
1528 	if (length % di->blocksize != 0) {
1529 		/*
1530 		 * This must be the final write after flushing the compression
1531 		 * stream. Write as many full blocks as possible and stash the
1532 		 * residual data in the dumper's block buffer. It will be
1533 		 * padded and written in dump_finish().
1534 		 */
1535 		rlength = rounddown(length, di->blocksize);
1536 		if (rlength != 0) {
1537 			error = _dump_append(di, base, rlength);
1538 			if (error != 0)
1539 				return (error);
1540 		}
1541 		resid = length - rlength;
1542 		memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1543 		bzero((uint8_t *)di->blockbuf + resid, di->blocksize - resid);
1544 		di->kdcomp->kdc_resid = resid;
1545 		return (EAGAIN);
1546 	}
1547 	return (_dump_append(di, base, length));
1548 }
1549 
1550 /*
1551  * Write kernel dump headers at the beginning and end of the dump extent.
1552  * Write the kernel dump encryption key after the leading header if we were
1553  * configured to do so.
1554  */
1555 static int
dump_write_headers(struct dumperinfo * di,struct kerneldumpheader * kdh)1556 dump_write_headers(struct dumperinfo *di, struct kerneldumpheader *kdh)
1557 {
1558 #ifdef EKCD
1559 	struct kerneldumpcrypto *kdc;
1560 #endif
1561 	void *buf;
1562 	size_t hdrsz;
1563 	uint64_t extent;
1564 	uint32_t keysize;
1565 	int error;
1566 
1567 	hdrsz = sizeof(*kdh);
1568 	if (hdrsz > di->blocksize)
1569 		return (ENOMEM);
1570 
1571 #ifdef EKCD
1572 	kdc = di->kdcrypto;
1573 	keysize = kerneldumpcrypto_dumpkeysize(kdc);
1574 #else
1575 	keysize = 0;
1576 #endif
1577 
1578 	/*
1579 	 * If the dump device has special handling for headers, let it take care
1580 	 * of writing them out.
1581 	 */
1582 	if (di->dumper_hdr != NULL)
1583 		return (di->dumper_hdr(di, kdh));
1584 
1585 	if (hdrsz == di->blocksize)
1586 		buf = kdh;
1587 	else {
1588 		buf = di->blockbuf;
1589 		memset(buf, 0, di->blocksize);
1590 		memcpy(buf, kdh, hdrsz);
1591 	}
1592 
1593 	extent = dtoh64(kdh->dumpextent);
1594 #ifdef EKCD
1595 	if (kdc != NULL) {
1596 		error = dump_write(di, kdc->kdc_dumpkey,
1597 		    di->mediaoffset + di->mediasize - di->blocksize - extent -
1598 		    keysize, keysize);
1599 		if (error != 0)
1600 			return (error);
1601 	}
1602 #endif
1603 
1604 	error = dump_write(di, buf,
1605 	    di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1606 	    keysize, di->blocksize);
1607 	if (error == 0)
1608 		error = dump_write(di, buf, di->mediaoffset + di->mediasize -
1609 		    di->blocksize, di->blocksize);
1610 	return (error);
1611 }
1612 
1613 /*
1614  * Don't touch the first SIZEOF_METADATA bytes on the dump device.  This is to
1615  * protect us from metadata and metadata from us.
1616  */
1617 #define	SIZEOF_METADATA		(64 * 1024)
1618 
1619 /*
1620  * Do some preliminary setup for a kernel dump: initialize state for encryption,
1621  * if requested, and make sure that we have enough space on the dump device.
1622  *
1623  * We set things up so that the dump ends before the last sector of the dump
1624  * device, at which the trailing header is written.
1625  *
1626  *     +-----------+------+-----+----------------------------+------+
1627  *     |           | lhdr | key |    ... kernel dump ...     | thdr |
1628  *     +-----------+------+-----+----------------------------+------+
1629  *                   1 blk  opt <------- dump extent --------> 1 blk
1630  *
1631  * Dumps written using dump_append() start at the beginning of the extent.
1632  * Uncompressed dumps will use the entire extent, but compressed dumps typically
1633  * will not. The true length of the dump is recorded in the leading and trailing
1634  * headers once the dump has been completed.
1635  *
1636  * The dump device may provide a callback, in which case it will initialize
1637  * dumpoff and take care of laying out the headers.
1638  */
1639 int
dump_start(struct dumperinfo * di,struct kerneldumpheader * kdh)1640 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1641 {
1642 #ifdef EKCD
1643 	struct kerneldumpcrypto *kdc;
1644 #endif
1645 	void *key;
1646 	uint64_t dumpextent, span;
1647 	uint32_t keysize;
1648 	int error;
1649 
1650 #ifdef EKCD
1651 	/* Send the key before the dump so a partial dump is still usable. */
1652 	kdc = di->kdcrypto;
1653 	error = kerneldumpcrypto_init(kdc);
1654 	if (error != 0)
1655 		return (error);
1656 	keysize = kerneldumpcrypto_dumpkeysize(kdc);
1657 	key = keysize > 0 ? kdc->kdc_dumpkey : NULL;
1658 #else
1659 	error = 0;
1660 	keysize = 0;
1661 	key = NULL;
1662 #endif
1663 
1664 	if (di->dumper_start != NULL) {
1665 		error = di->dumper_start(di, key, keysize);
1666 	} else {
1667 		dumpextent = dtoh64(kdh->dumpextent);
1668 		span = SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1669 		    keysize;
1670 		if (di->mediasize < span) {
1671 			if (di->kdcomp == NULL)
1672 				return (E2BIG);
1673 
1674 			/*
1675 			 * We don't yet know how much space the compressed dump
1676 			 * will occupy, so try to use the whole swap partition
1677 			 * (minus the first 64KB) in the hope that the
1678 			 * compressed dump will fit. If that doesn't turn out to
1679 			 * be enough, the bounds checking in dump_write()
1680 			 * will catch us and cause the dump to fail.
1681 			 */
1682 			dumpextent = di->mediasize - span + dumpextent;
1683 			kdh->dumpextent = htod64(dumpextent);
1684 		}
1685 
1686 		/*
1687 		 * The offset at which to begin writing the dump.
1688 		 */
1689 		di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1690 		    dumpextent;
1691 	}
1692 	di->origdumpoff = di->dumpoff;
1693 	return (error);
1694 }
1695 
1696 static int
_dump_append(struct dumperinfo * di,void * virtual,size_t length)1697 _dump_append(struct dumperinfo *di, void *virtual, size_t length)
1698 {
1699 	int error;
1700 
1701 #ifdef EKCD
1702 	if (di->kdcrypto != NULL)
1703 		error = dump_encrypted_write(di, virtual, di->dumpoff, length);
1704 	else
1705 #endif
1706 		error = dump_write(di, virtual, di->dumpoff, length);
1707 	if (error == 0)
1708 		di->dumpoff += length;
1709 	return (error);
1710 }
1711 
1712 /*
1713  * Write to the dump device starting at dumpoff. When compression is enabled,
1714  * writes to the device will be performed using a callback that gets invoked
1715  * when the compression stream's output buffer is full.
1716  */
1717 int
dump_append(struct dumperinfo * di,void * virtual,size_t length)1718 dump_append(struct dumperinfo *di, void *virtual, size_t length)
1719 {
1720 	void *buf;
1721 
1722 	if (di->kdcomp != NULL) {
1723 		/* Bounce through a buffer to avoid CRC errors. */
1724 		if (length > di->maxiosize)
1725 			return (EINVAL);
1726 		buf = di->kdcomp->kdc_buf;
1727 		memmove(buf, virtual, length);
1728 		return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1729 	}
1730 	return (_dump_append(di, virtual, length));
1731 }
1732 
1733 /*
1734  * Write to the dump device at the specified offset.
1735  */
1736 int
dump_write(struct dumperinfo * di,void * virtual,off_t offset,size_t length)1737 dump_write(struct dumperinfo *di, void *virtual, off_t offset, size_t length)
1738 {
1739 	int error;
1740 
1741 	error = dump_check_bounds(di, offset, length);
1742 	if (error != 0)
1743 		return (error);
1744 	return (di->dumper(di->priv, virtual, offset, length));
1745 }
1746 
1747 /*
1748  * Perform kernel dump finalization: flush the compression stream, if necessary,
1749  * write the leading and trailing kernel dump headers now that we know the true
1750  * length of the dump, and optionally write the encryption key following the
1751  * leading header.
1752  */
1753 int
dump_finish(struct dumperinfo * di,struct kerneldumpheader * kdh)1754 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1755 {
1756 	int error;
1757 
1758 	if (di->kdcomp != NULL) {
1759 		error = compressor_flush(di->kdcomp->kdc_stream);
1760 		if (error == EAGAIN) {
1761 			/* We have residual data in di->blockbuf. */
1762 			error = _dump_append(di, di->blockbuf, di->blocksize);
1763 			if (error == 0)
1764 				/* Compensate for _dump_append()'s adjustment. */
1765 				di->dumpoff -= di->blocksize - di->kdcomp->kdc_resid;
1766 			di->kdcomp->kdc_resid = 0;
1767 		}
1768 		if (error != 0)
1769 			return (error);
1770 
1771 		/*
1772 		 * We now know the size of the compressed dump, so update the
1773 		 * header accordingly and recompute parity.
1774 		 */
1775 		kdh->dumplength = htod64(di->dumpoff - di->origdumpoff);
1776 		kdh->parity = 0;
1777 		kdh->parity = kerneldump_parity(kdh);
1778 
1779 		compressor_reset(di->kdcomp->kdc_stream);
1780 	}
1781 
1782 	error = dump_write_headers(di, kdh);
1783 	if (error != 0)
1784 		return (error);
1785 
1786 	(void)dump_write(di, NULL, 0, 0);
1787 	return (0);
1788 }
1789 
1790 void
dump_init_header(const struct dumperinfo * di,struct kerneldumpheader * kdh,const char * magic,uint32_t archver,uint64_t dumplen)1791 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1792     const char *magic, uint32_t archver, uint64_t dumplen)
1793 {
1794 	size_t dstsize;
1795 
1796 	bzero(kdh, sizeof(*kdh));
1797 	strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1798 	strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1799 	kdh->version = htod32(KERNELDUMPVERSION);
1800 	kdh->architectureversion = htod32(archver);
1801 	kdh->dumplength = htod64(dumplen);
1802 	kdh->dumpextent = kdh->dumplength;
1803 	kdh->dumptime = htod64(time_second);
1804 #ifdef EKCD
1805 	kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1806 #else
1807 	kdh->dumpkeysize = 0;
1808 #endif
1809 	kdh->blocksize = htod32(di->blocksize);
1810 	strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1811 	dstsize = sizeof(kdh->versionstring);
1812 	if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1813 		kdh->versionstring[dstsize - 2] = '\n';
1814 	if (panicstr != NULL)
1815 		strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1816 	if (di->kdcomp != NULL)
1817 		kdh->compression = di->kdcomp->kdc_format;
1818 	kdh->parity = kerneldump_parity(kdh);
1819 }
1820 
1821 #ifdef DDB
DB_SHOW_COMMAND_FLAGS(panic,db_show_panic,DB_CMD_MEMSAFE)1822 DB_SHOW_COMMAND_FLAGS(panic, db_show_panic, DB_CMD_MEMSAFE)
1823 {
1824 
1825 	if (panicstr == NULL)
1826 		db_printf("panicstr not set\n");
1827 	else
1828 		db_printf("panic: %s\n", panicstr);
1829 }
1830 #endif
1831