xref: /freebsd/sbin/init/init.c (revision 2c77ec54190294415c52a8da0c0d9c5a957c03a3)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Donn Seeley at Berkeley Software Design, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 #ifndef lint
36 static const char copyright[] =
37 "@(#) Copyright (c) 1991, 1993\n\
38 	The Regents of the University of California.  All rights reserved.\n";
39 #endif /* not lint */
40 
41 #ifndef lint
42 #if 0
43 static char sccsid[] = "@(#)init.c	8.1 (Berkeley) 7/15/93";
44 #endif
45 static const char rcsid[] =
46   "$FreeBSD$";
47 #endif /* not lint */
48 
49 #include <sys/param.h>
50 #include <sys/ioctl.h>
51 #include <sys/mman.h>
52 #include <sys/mount.h>
53 #include <sys/sysctl.h>
54 #include <sys/wait.h>
55 #include <sys/stat.h>
56 #include <sys/uio.h>
57 
58 #include <db.h>
59 #include <errno.h>
60 #include <fcntl.h>
61 #include <kenv.h>
62 #include <libutil.h>
63 #include <paths.h>
64 #include <signal.h>
65 #include <stdio.h>
66 #include <stdlib.h>
67 #include <string.h>
68 #include <syslog.h>
69 #include <time.h>
70 #include <ttyent.h>
71 #include <unistd.h>
72 #include <sys/reboot.h>
73 #include <err.h>
74 
75 #include <stdarg.h>
76 
77 #ifdef SECURE
78 #include <pwd.h>
79 #endif
80 
81 #ifdef LOGIN_CAP
82 #include <login_cap.h>
83 #endif
84 
85 #include "mntopts.h"
86 #include "pathnames.h"
87 
88 /*
89  * Sleep times; used to prevent thrashing.
90  */
91 #define	GETTY_SPACING		 5	/* N secs minimum getty spacing */
92 #define	GETTY_SLEEP		30	/* sleep N secs after spacing problem */
93 #define	GETTY_NSPACE		 3	/* max. spacing count to bring reaction */
94 #define	WINDOW_WAIT		 3	/* wait N secs after starting window */
95 #define	STALL_TIMEOUT		30	/* wait N secs after warning */
96 #define	DEATH_WATCH		10	/* wait N secs for procs to die */
97 #define	DEATH_SCRIPT		120	/* wait for 2min for /etc/rc.shutdown */
98 #define	RESOURCE_RC		"daemon"
99 #define	RESOURCE_WINDOW		"default"
100 #define	RESOURCE_GETTY		"default"
101 
102 static void handle(sig_t, ...);
103 static void delset(sigset_t *, ...);
104 
105 static void stall(const char *, ...) __printflike(1, 2);
106 static void warning(const char *, ...) __printflike(1, 2);
107 static void emergency(const char *, ...) __printflike(1, 2);
108 static void disaster(int);
109 static void badsys(int);
110 static void revoke_ttys(void);
111 static int  runshutdown(void);
112 static char *strk(char *);
113 
114 /*
115  * We really need a recursive typedef...
116  * The following at least guarantees that the return type of (*state_t)()
117  * is sufficiently wide to hold a function pointer.
118  */
119 typedef long (*state_func_t)(void);
120 typedef state_func_t (*state_t)(void);
121 
122 static state_func_t single_user(void);
123 static state_func_t runcom(void);
124 static state_func_t read_ttys(void);
125 static state_func_t multi_user(void);
126 static state_func_t clean_ttys(void);
127 static state_func_t catatonia(void);
128 static state_func_t death(void);
129 static state_func_t death_single(void);
130 static state_func_t reroot(void);
131 static state_func_t reroot_phase_two(void);
132 
133 static state_func_t run_script(const char *);
134 
135 static enum { AUTOBOOT, FASTBOOT } runcom_mode = AUTOBOOT;
136 #define FALSE	0
137 #define TRUE	1
138 
139 static int Reboot = FALSE;
140 static int howto = RB_AUTOBOOT;
141 
142 static int devfs;
143 static char *init_path_argv0;
144 
145 static void transition(state_t);
146 static state_t requested_transition;
147 static state_t current_state = death_single;
148 
149 static void open_console(void);
150 static const char *get_shell(void);
151 static void write_stderr(const char *message);
152 
153 typedef struct init_session {
154 	pid_t	se_process;		/* controlling process */
155 	time_t	se_started;		/* used to avoid thrashing */
156 	int	se_flags;		/* status of session */
157 #define	SE_SHUTDOWN	0x1		/* session won't be restarted */
158 #define	SE_PRESENT	0x2		/* session is in /etc/ttys */
159 #define	SE_IFEXISTS	0x4		/* session defined as "onifexists" */
160 #define	SE_IFCONSOLE	0x8		/* session defined as "onifconsole" */
161 	int	se_nspace;		/* spacing count */
162 	char	*se_device;		/* filename of port */
163 	char	*se_getty;		/* what to run on that port */
164 	char	*se_getty_argv_space;   /* pre-parsed argument array space */
165 	char	**se_getty_argv;	/* pre-parsed argument array */
166 	char	*se_window;		/* window system (started only once) */
167 	char	*se_window_argv_space;  /* pre-parsed argument array space */
168 	char	**se_window_argv;	/* pre-parsed argument array */
169 	char	*se_type;		/* default terminal type */
170 	struct	init_session *se_prev;
171 	struct	init_session *se_next;
172 } session_t;
173 
174 static void free_session(session_t *);
175 static session_t *new_session(session_t *, struct ttyent *);
176 static session_t *sessions;
177 
178 static char **construct_argv(char *);
179 static void start_window_system(session_t *);
180 static void collect_child(pid_t);
181 static pid_t start_getty(session_t *);
182 static void transition_handler(int);
183 static void alrm_handler(int);
184 static void setsecuritylevel(int);
185 static int getsecuritylevel(void);
186 static int setupargv(session_t *, struct ttyent *);
187 #ifdef LOGIN_CAP
188 static void setprocresources(const char *);
189 #endif
190 static int clang;
191 
192 static int start_session_db(void);
193 static void add_session(session_t *);
194 static void del_session(session_t *);
195 static session_t *find_session(pid_t);
196 static DB *session_db;
197 
198 /*
199  * The mother of all processes.
200  */
201 int
202 main(int argc, char *argv[])
203 {
204 	state_t initial_transition = runcom;
205 	char kenv_value[PATH_MAX];
206 	int c, error;
207 	struct sigaction sa;
208 	sigset_t mask;
209 
210 	/* Dispose of random users. */
211 	if (getuid() != 0)
212 		errx(1, "%s", strerror(EPERM));
213 
214 	/* System V users like to reexec init. */
215 	if (getpid() != 1) {
216 #ifdef COMPAT_SYSV_INIT
217 		/* So give them what they want */
218 		if (argc > 1) {
219 			if (strlen(argv[1]) == 1) {
220 				char runlevel = *argv[1];
221 				int sig;
222 
223 				switch (runlevel) {
224 				case '0': /* halt + poweroff */
225 					sig = SIGUSR2;
226 					break;
227 				case '1': /* single-user */
228 					sig = SIGTERM;
229 					break;
230 				case '6': /* reboot */
231 					sig = SIGINT;
232 					break;
233 				case 'c': /* block further logins */
234 					sig = SIGTSTP;
235 					break;
236 				case 'q': /* rescan /etc/ttys */
237 					sig = SIGHUP;
238 					break;
239 				case 'r': /* remount root */
240 					sig = SIGEMT;
241 					break;
242 				default:
243 					goto invalid;
244 				}
245 				kill(1, sig);
246 				_exit(0);
247 			} else
248 invalid:
249 				errx(1, "invalid run-level ``%s''", argv[1]);
250 		} else
251 #endif
252 			errx(1, "already running");
253 	}
254 
255 	init_path_argv0 = strdup(argv[0]);
256 	if (init_path_argv0 == NULL)
257 		err(1, "strdup");
258 
259 	/*
260 	 * Note that this does NOT open a file...
261 	 * Does 'init' deserve its own facility number?
262 	 */
263 	openlog("init", LOG_CONS, LOG_AUTH);
264 
265 	/*
266 	 * Create an initial session.
267 	 */
268 	if (setsid() < 0 && (errno != EPERM || getsid(0) != 1))
269 		warning("initial setsid() failed: %m");
270 
271 	/*
272 	 * Establish an initial user so that programs running
273 	 * single user do not freak out and die (like passwd).
274 	 */
275 	if (setlogin("root") < 0)
276 		warning("setlogin() failed: %m");
277 
278 	/*
279 	 * This code assumes that we always get arguments through flags,
280 	 * never through bits set in some random machine register.
281 	 */
282 	while ((c = getopt(argc, argv, "dsfr")) != -1)
283 		switch (c) {
284 		case 'd':
285 			devfs = 1;
286 			break;
287 		case 's':
288 			initial_transition = single_user;
289 			break;
290 		case 'f':
291 			runcom_mode = FASTBOOT;
292 			break;
293 		case 'r':
294 			initial_transition = reroot_phase_two;
295 			break;
296 		default:
297 			warning("unrecognized flag '-%c'", c);
298 			break;
299 		}
300 
301 	if (optind != argc)
302 		warning("ignoring excess arguments");
303 
304 	/*
305 	 * We catch or block signals rather than ignore them,
306 	 * so that they get reset on exec.
307 	 */
308 	handle(badsys, SIGSYS, 0);
309 	handle(disaster, SIGABRT, SIGFPE, SIGILL, SIGSEGV, SIGBUS, SIGXCPU,
310 	    SIGXFSZ, 0);
311 	handle(transition_handler, SIGHUP, SIGINT, SIGEMT, SIGTERM, SIGTSTP,
312 	    SIGUSR1, SIGUSR2, SIGWINCH, 0);
313 	handle(alrm_handler, SIGALRM, 0);
314 	sigfillset(&mask);
315 	delset(&mask, SIGABRT, SIGFPE, SIGILL, SIGSEGV, SIGBUS, SIGSYS,
316 	    SIGXCPU, SIGXFSZ, SIGHUP, SIGINT, SIGEMT, SIGTERM, SIGTSTP,
317 	    SIGALRM, SIGUSR1, SIGUSR2, SIGWINCH, 0);
318 	sigprocmask(SIG_SETMASK, &mask, (sigset_t *) 0);
319 	sigemptyset(&sa.sa_mask);
320 	sa.sa_flags = 0;
321 	sa.sa_handler = SIG_IGN;
322 	sigaction(SIGTTIN, &sa, (struct sigaction *)0);
323 	sigaction(SIGTTOU, &sa, (struct sigaction *)0);
324 
325 	/*
326 	 * Paranoia.
327 	 */
328 	close(0);
329 	close(1);
330 	close(2);
331 
332 	if (kenv(KENV_GET, "init_script", kenv_value, sizeof(kenv_value)) > 0) {
333 		state_func_t next_transition;
334 
335 		if ((next_transition = run_script(kenv_value)) != NULL)
336 			initial_transition = (state_t) next_transition;
337 	}
338 
339 	if (kenv(KENV_GET, "init_chroot", kenv_value, sizeof(kenv_value)) > 0) {
340 		if (chdir(kenv_value) != 0 || chroot(".") != 0)
341 			warning("Can't chroot to %s: %m", kenv_value);
342 	}
343 
344 	/*
345 	 * Additional check if devfs needs to be mounted:
346 	 * If "/" and "/dev" have the same device number,
347 	 * then it hasn't been mounted yet.
348 	 */
349 	if (!devfs) {
350 		struct stat stst;
351 		dev_t root_devno;
352 
353 		stat("/", &stst);
354 		root_devno = stst.st_dev;
355 		if (stat("/dev", &stst) != 0)
356 			warning("Can't stat /dev: %m");
357 		else if (stst.st_dev == root_devno)
358 			devfs++;
359 	}
360 
361 	if (devfs) {
362 		struct iovec iov[4];
363 		char *s;
364 		int i;
365 
366 		char _fstype[]	= "fstype";
367 		char _devfs[]	= "devfs";
368 		char _fspath[]	= "fspath";
369 		char _path_dev[]= _PATH_DEV;
370 
371 		iov[0].iov_base = _fstype;
372 		iov[0].iov_len = sizeof(_fstype);
373 		iov[1].iov_base = _devfs;
374 		iov[1].iov_len = sizeof(_devfs);
375 		iov[2].iov_base = _fspath;
376 		iov[2].iov_len = sizeof(_fspath);
377 		/*
378 		 * Try to avoid the trailing slash in _PATH_DEV.
379 		 * Be *very* defensive.
380 		 */
381 		s = strdup(_PATH_DEV);
382 		if (s != NULL) {
383 			i = strlen(s);
384 			if (i > 0 && s[i - 1] == '/')
385 				s[i - 1] = '\0';
386 			iov[3].iov_base = s;
387 			iov[3].iov_len = strlen(s) + 1;
388 		} else {
389 			iov[3].iov_base = _path_dev;
390 			iov[3].iov_len = sizeof(_path_dev);
391 		}
392 		nmount(iov, 4, 0);
393 		if (s != NULL)
394 			free(s);
395 	}
396 
397 	if (initial_transition != reroot_phase_two) {
398 		/*
399 		 * Unmount reroot leftovers.  This runs after init(8)
400 		 * gets reexecuted after reroot_phase_two() is done.
401 		 */
402 		error = unmount(_PATH_REROOT, MNT_FORCE);
403 		if (error != 0 && errno != EINVAL)
404 			warning("Cannot unmount %s: %m", _PATH_REROOT);
405 	}
406 
407 	/*
408 	 * Start the state machine.
409 	 */
410 	transition(initial_transition);
411 
412 	/*
413 	 * Should never reach here.
414 	 */
415 	return 1;
416 }
417 
418 /*
419  * Associate a function with a signal handler.
420  */
421 static void
422 handle(sig_t handler, ...)
423 {
424 	int sig;
425 	struct sigaction sa;
426 	sigset_t mask_everything;
427 	va_list ap;
428 	va_start(ap, handler);
429 
430 	sa.sa_handler = handler;
431 	sigfillset(&mask_everything);
432 
433 	while ((sig = va_arg(ap, int)) != 0) {
434 		sa.sa_mask = mask_everything;
435 		/* XXX SA_RESTART? */
436 		sa.sa_flags = sig == SIGCHLD ? SA_NOCLDSTOP : 0;
437 		sigaction(sig, &sa, (struct sigaction *) 0);
438 	}
439 	va_end(ap);
440 }
441 
442 /*
443  * Delete a set of signals from a mask.
444  */
445 static void
446 delset(sigset_t *maskp, ...)
447 {
448 	int sig;
449 	va_list ap;
450 	va_start(ap, maskp);
451 
452 	while ((sig = va_arg(ap, int)) != 0)
453 		sigdelset(maskp, sig);
454 	va_end(ap);
455 }
456 
457 /*
458  * Log a message and sleep for a while (to give someone an opportunity
459  * to read it and to save log or hardcopy output if the problem is chronic).
460  * NB: should send a message to the session logger to avoid blocking.
461  */
462 static void
463 stall(const char *message, ...)
464 {
465 	va_list ap;
466 	va_start(ap, message);
467 
468 	vsyslog(LOG_ALERT, message, ap);
469 	va_end(ap);
470 	sleep(STALL_TIMEOUT);
471 }
472 
473 /*
474  * Like stall(), but doesn't sleep.
475  * If cpp had variadic macros, the two functions could be #defines for another.
476  * NB: should send a message to the session logger to avoid blocking.
477  */
478 static void
479 warning(const char *message, ...)
480 {
481 	va_list ap;
482 	va_start(ap, message);
483 
484 	vsyslog(LOG_ALERT, message, ap);
485 	va_end(ap);
486 }
487 
488 /*
489  * Log an emergency message.
490  * NB: should send a message to the session logger to avoid blocking.
491  */
492 static void
493 emergency(const char *message, ...)
494 {
495 	va_list ap;
496 	va_start(ap, message);
497 
498 	vsyslog(LOG_EMERG, message, ap);
499 	va_end(ap);
500 }
501 
502 /*
503  * Catch a SIGSYS signal.
504  *
505  * These may arise if a system does not support sysctl.
506  * We tolerate up to 25 of these, then throw in the towel.
507  */
508 static void
509 badsys(int sig)
510 {
511 	static int badcount = 0;
512 
513 	if (badcount++ < 25)
514 		return;
515 	disaster(sig);
516 }
517 
518 /*
519  * Catch an unexpected signal.
520  */
521 static void
522 disaster(int sig)
523 {
524 
525 	emergency("fatal signal: %s",
526 	    (unsigned)sig < NSIG ? sys_siglist[sig] : "unknown signal");
527 
528 	sleep(STALL_TIMEOUT);
529 	_exit(sig);		/* reboot */
530 }
531 
532 /*
533  * Get the security level of the kernel.
534  */
535 static int
536 getsecuritylevel(void)
537 {
538 #ifdef KERN_SECURELVL
539 	int name[2], curlevel;
540 	size_t len;
541 
542 	name[0] = CTL_KERN;
543 	name[1] = KERN_SECURELVL;
544 	len = sizeof curlevel;
545 	if (sysctl(name, 2, &curlevel, &len, NULL, 0) == -1) {
546 		emergency("cannot get kernel security level: %s",
547 		    strerror(errno));
548 		return (-1);
549 	}
550 	return (curlevel);
551 #else
552 	return (-1);
553 #endif
554 }
555 
556 /*
557  * Set the security level of the kernel.
558  */
559 static void
560 setsecuritylevel(int newlevel)
561 {
562 #ifdef KERN_SECURELVL
563 	int name[2], curlevel;
564 
565 	curlevel = getsecuritylevel();
566 	if (newlevel == curlevel)
567 		return;
568 	name[0] = CTL_KERN;
569 	name[1] = KERN_SECURELVL;
570 	if (sysctl(name, 2, NULL, NULL, &newlevel, sizeof newlevel) == -1) {
571 		emergency(
572 		    "cannot change kernel security level from %d to %d: %s",
573 		    curlevel, newlevel, strerror(errno));
574 		return;
575 	}
576 #ifdef SECURE
577 	warning("kernel security level changed from %d to %d",
578 	    curlevel, newlevel);
579 #endif
580 #endif
581 }
582 
583 /*
584  * Change states in the finite state machine.
585  * The initial state is passed as an argument.
586  */
587 static void
588 transition(state_t s)
589 {
590 
591 	current_state = s;
592 	for (;;)
593 		current_state = (state_t) (*current_state)();
594 }
595 
596 /*
597  * Start a session and allocate a controlling terminal.
598  * Only called by children of init after forking.
599  */
600 static void
601 open_console(void)
602 {
603 	int fd;
604 
605 	/*
606 	 * Try to open /dev/console.  Open the device with O_NONBLOCK to
607 	 * prevent potential blocking on a carrier.
608 	 */
609 	revoke(_PATH_CONSOLE);
610 	if ((fd = open(_PATH_CONSOLE, O_RDWR | O_NONBLOCK)) != -1) {
611 		(void)fcntl(fd, F_SETFL, fcntl(fd, F_GETFL) & ~O_NONBLOCK);
612 		if (login_tty(fd) == 0)
613 			return;
614 		close(fd);
615 	}
616 
617 	/* No luck.  Log output to file if possible. */
618 	if ((fd = open(_PATH_DEVNULL, O_RDWR)) == -1) {
619 		stall("cannot open null device.");
620 		_exit(1);
621 	}
622 	if (fd != STDIN_FILENO) {
623 		dup2(fd, STDIN_FILENO);
624 		close(fd);
625 	}
626 	fd = open(_PATH_INITLOG, O_WRONLY | O_APPEND | O_CREAT, 0644);
627 	if (fd == -1)
628 		dup2(STDIN_FILENO, STDOUT_FILENO);
629 	else if (fd != STDOUT_FILENO) {
630 		dup2(fd, STDOUT_FILENO);
631 		close(fd);
632 	}
633 	dup2(STDOUT_FILENO, STDERR_FILENO);
634 }
635 
636 static const char *
637 get_shell(void)
638 {
639 	static char kenv_value[PATH_MAX];
640 
641 	if (kenv(KENV_GET, "init_shell", kenv_value, sizeof(kenv_value)) > 0)
642 		return kenv_value;
643 	else
644 		return _PATH_BSHELL;
645 }
646 
647 static void
648 write_stderr(const char *message)
649 {
650 
651 	write(STDERR_FILENO, message, strlen(message));
652 }
653 
654 static int
655 read_file(const char *path, void **bufp, size_t *bufsizep)
656 {
657 	struct stat sb;
658 	size_t bufsize;
659 	void *buf;
660 	ssize_t nbytes;
661 	int error, fd;
662 
663 	fd = open(path, O_RDONLY);
664 	if (fd < 0) {
665 		emergency("%s: %s", path, strerror(errno));
666 		return (-1);
667 	}
668 
669 	error = fstat(fd, &sb);
670 	if (error != 0) {
671 		emergency("fstat: %s", strerror(errno));
672 		close(fd);
673 		return (error);
674 	}
675 
676 	bufsize = sb.st_size;
677 	buf = malloc(bufsize);
678 	if (buf == NULL) {
679 		emergency("malloc: %s", strerror(errno));
680 		close(fd);
681 		return (error);
682 	}
683 
684 	nbytes = read(fd, buf, bufsize);
685 	if (nbytes != (ssize_t)bufsize) {
686 		emergency("read: %s", strerror(errno));
687 		close(fd);
688 		free(buf);
689 		return (error);
690 	}
691 
692 	error = close(fd);
693 	if (error != 0) {
694 		emergency("close: %s", strerror(errno));
695 		free(buf);
696 		return (error);
697 	}
698 
699 	*bufp = buf;
700 	*bufsizep = bufsize;
701 
702 	return (0);
703 }
704 
705 static int
706 create_file(const char *path, const void *buf, size_t bufsize)
707 {
708 	ssize_t nbytes;
709 	int error, fd;
710 
711 	fd = open(path, O_WRONLY | O_CREAT | O_EXCL, 0700);
712 	if (fd < 0) {
713 		emergency("%s: %s", path, strerror(errno));
714 		return (-1);
715 	}
716 
717 	nbytes = write(fd, buf, bufsize);
718 	if (nbytes != (ssize_t)bufsize) {
719 		emergency("write: %s", strerror(errno));
720 		close(fd);
721 		return (-1);
722 	}
723 
724 	error = close(fd);
725 	if (error != 0) {
726 		emergency("close: %s", strerror(errno));
727 		return (-1);
728 	}
729 
730 	return (0);
731 }
732 
733 static int
734 mount_tmpfs(const char *fspath)
735 {
736 	struct iovec *iov;
737 	char errmsg[255];
738 	int error, iovlen;
739 
740 	iov = NULL;
741 	iovlen = 0;
742 	memset(errmsg, 0, sizeof(errmsg));
743 	build_iovec(&iov, &iovlen, "fstype",
744 	    __DECONST(void *, "tmpfs"), (size_t)-1);
745 	build_iovec(&iov, &iovlen, "fspath",
746 	    __DECONST(void *, fspath), (size_t)-1);
747 	build_iovec(&iov, &iovlen, "errmsg",
748 	    errmsg, sizeof(errmsg));
749 
750 	error = nmount(iov, iovlen, 0);
751 	if (error != 0) {
752 		if (*errmsg != '\0') {
753 			emergency("cannot mount tmpfs on %s: %s: %s",
754 			    fspath, errmsg, strerror(errno));
755 		} else {
756 			emergency("cannot mount tmpfs on %s: %s",
757 			    fspath, strerror(errno));
758 		}
759 		return (error);
760 	}
761 	return (0);
762 }
763 
764 static state_func_t
765 reroot(void)
766 {
767 	void *buf;
768 	size_t bufsize;
769 	int error;
770 
771 	buf = NULL;
772 	bufsize = 0;
773 
774 	revoke_ttys();
775 	runshutdown();
776 
777 	/*
778 	 * Make sure nobody can interfere with our scheme.
779 	 * Ignore ESRCH, which can apparently happen when
780 	 * there are no processes to kill.
781 	 */
782 	error = kill(-1, SIGKILL);
783 	if (error != 0 && errno != ESRCH) {
784 		emergency("kill(2) failed: %s", strerror(errno));
785 		goto out;
786 	}
787 
788 	/*
789 	 * Copy the init binary into tmpfs, so that we can unmount
790 	 * the old rootfs without committing suicide.
791 	 */
792 	error = read_file(init_path_argv0, &buf, &bufsize);
793 	if (error != 0)
794 		goto out;
795 	error = mount_tmpfs(_PATH_REROOT);
796 	if (error != 0)
797 		goto out;
798 	error = create_file(_PATH_REROOT_INIT, buf, bufsize);
799 	if (error != 0)
800 		goto out;
801 
802 	/*
803 	 * Execute the temporary init.
804 	 */
805 	execl(_PATH_REROOT_INIT, _PATH_REROOT_INIT, "-r", NULL);
806 	emergency("cannot exec %s: %s", _PATH_REROOT_INIT, strerror(errno));
807 
808 out:
809 	emergency("reroot failed; going to single user mode");
810 	free(buf);
811 	return (state_func_t) single_user;
812 }
813 
814 static state_func_t
815 reroot_phase_two(void)
816 {
817 	char init_path[PATH_MAX], *path, *path_component;
818 	size_t init_path_len;
819 	int nbytes, error;
820 
821 	/*
822 	 * Ask the kernel to mount the new rootfs.
823 	 */
824 	error = reboot(RB_REROOT);
825 	if (error != 0) {
826 		emergency("RB_REBOOT failed: %s", strerror(errno));
827 		goto out;
828 	}
829 
830 	/*
831 	 * Figure out where the destination init(8) binary is.  Note that
832 	 * the path could be different than what we've started with.  Use
833 	 * the value from kenv, if set, or the one from sysctl otherwise.
834 	 * The latter defaults to a hardcoded value, but can be overridden
835 	 * by a build time option.
836 	 */
837 	nbytes = kenv(KENV_GET, "init_path", init_path, sizeof(init_path));
838 	if (nbytes <= 0) {
839 		init_path_len = sizeof(init_path);
840 		error = sysctlbyname("kern.init_path",
841 		    init_path, &init_path_len, NULL, 0);
842 		if (error != 0) {
843 			emergency("failed to retrieve kern.init_path: %s",
844 			    strerror(errno));
845 			goto out;
846 		}
847 	}
848 
849 	/*
850 	 * Repeat the init search logic from sys/kern/init_path.c
851 	 */
852 	path_component = init_path;
853 	while ((path = strsep(&path_component, ":")) != NULL) {
854 		/*
855 		 * Execute init(8) from the new rootfs.
856 		 */
857 		execl(path, path, NULL);
858 	}
859 	emergency("cannot exec init from %s: %s", init_path, strerror(errno));
860 
861 out:
862 	emergency("reroot failed; going to single user mode");
863 	return (state_func_t) single_user;
864 }
865 
866 /*
867  * Bring the system up single user.
868  */
869 static state_func_t
870 single_user(void)
871 {
872 	pid_t pid, wpid;
873 	int status;
874 	sigset_t mask;
875 	const char *shell;
876 	char *argv[2];
877 	struct timeval tv, tn;
878 #ifdef SECURE
879 	struct ttyent *typ;
880 	struct passwd *pp;
881 	static const char banner[] =
882 		"Enter root password, or ^D to go multi-user\n";
883 	char *clear, *password;
884 #endif
885 #ifdef DEBUGSHELL
886 	char altshell[128];
887 #endif
888 
889 	if (Reboot) {
890 		/* Instead of going single user, let's reboot the machine */
891 		sync();
892 		if (reboot(howto) == -1) {
893 			emergency("reboot(%#x) failed, %s", howto,
894 			    strerror(errno));
895 			_exit(1); /* panic and reboot */
896 		}
897 		warning("reboot(%#x) returned", howto);
898 		_exit(0); /* panic as well */
899 	}
900 
901 	shell = get_shell();
902 
903 	if ((pid = fork()) == 0) {
904 		/*
905 		 * Start the single user session.
906 		 */
907 		open_console();
908 
909 #ifdef SECURE
910 		/*
911 		 * Check the root password.
912 		 * We don't care if the console is 'on' by default;
913 		 * it's the only tty that can be 'off' and 'secure'.
914 		 */
915 		typ = getttynam("console");
916 		pp = getpwnam("root");
917 		if (typ && (typ->ty_status & TTY_SECURE) == 0 &&
918 		    pp && *pp->pw_passwd) {
919 			write_stderr(banner);
920 			for (;;) {
921 				clear = getpass("Password:");
922 				if (clear == NULL || *clear == '\0')
923 					_exit(0);
924 				password = crypt(clear, pp->pw_passwd);
925 				bzero(clear, _PASSWORD_LEN);
926 				if (password != NULL &&
927 				    strcmp(password, pp->pw_passwd) == 0)
928 					break;
929 				warning("single-user login failed\n");
930 			}
931 		}
932 		endttyent();
933 		endpwent();
934 #endif /* SECURE */
935 
936 #ifdef DEBUGSHELL
937 		{
938 			char *cp = altshell;
939 			int num;
940 
941 #define	SHREQUEST "Enter full pathname of shell or RETURN for "
942 			write_stderr(SHREQUEST);
943 			write_stderr(shell);
944 			write_stderr(": ");
945 			while ((num = read(STDIN_FILENO, cp, 1)) != -1 &&
946 			    num != 0 && *cp != '\n' && cp < &altshell[127])
947 				cp++;
948 			*cp = '\0';
949 			if (altshell[0] != '\0')
950 				shell = altshell;
951 		}
952 #endif /* DEBUGSHELL */
953 
954 		/*
955 		 * Unblock signals.
956 		 * We catch all the interesting ones,
957 		 * and those are reset to SIG_DFL on exec.
958 		 */
959 		sigemptyset(&mask);
960 		sigprocmask(SIG_SETMASK, &mask, (sigset_t *) 0);
961 
962 		/*
963 		 * Fire off a shell.
964 		 * If the default one doesn't work, try the Bourne shell.
965 		 */
966 
967 		char name[] = "-sh";
968 
969 		argv[0] = name;
970 		argv[1] = 0;
971 		execv(shell, argv);
972 		emergency("can't exec %s for single user: %m", shell);
973 		execv(_PATH_BSHELL, argv);
974 		emergency("can't exec %s for single user: %m", _PATH_BSHELL);
975 		sleep(STALL_TIMEOUT);
976 		_exit(1);
977 	}
978 
979 	if (pid == -1) {
980 		/*
981 		 * We are seriously hosed.  Do our best.
982 		 */
983 		emergency("can't fork single-user shell, trying again");
984 		while (waitpid(-1, (int *) 0, WNOHANG) > 0)
985 			continue;
986 		return (state_func_t) single_user;
987 	}
988 
989 	requested_transition = 0;
990 	do {
991 		if ((wpid = waitpid(-1, &status, WUNTRACED)) != -1)
992 			collect_child(wpid);
993 		if (wpid == -1) {
994 			if (errno == EINTR)
995 				continue;
996 			warning("wait for single-user shell failed: %m; restarting");
997 			return (state_func_t) single_user;
998 		}
999 		if (wpid == pid && WIFSTOPPED(status)) {
1000 			warning("init: shell stopped, restarting\n");
1001 			kill(pid, SIGCONT);
1002 			wpid = -1;
1003 		}
1004 	} while (wpid != pid && !requested_transition);
1005 
1006 	if (requested_transition)
1007 		return (state_func_t) requested_transition;
1008 
1009 	if (!WIFEXITED(status)) {
1010 		if (WTERMSIG(status) == SIGKILL) {
1011 			/*
1012 			 *  reboot(8) killed shell?
1013 			 */
1014 			warning("single user shell terminated.");
1015 			gettimeofday(&tv, NULL);
1016 			tn = tv;
1017 			tv.tv_sec += STALL_TIMEOUT;
1018 			while (tv.tv_sec > tn.tv_sec || (tv.tv_sec ==
1019 			    tn.tv_sec && tv.tv_usec > tn.tv_usec)) {
1020 				sleep(1);
1021 				gettimeofday(&tn, NULL);
1022 			}
1023 			_exit(0);
1024 		} else {
1025 			warning("single user shell terminated, restarting");
1026 			return (state_func_t) single_user;
1027 		}
1028 	}
1029 
1030 	runcom_mode = FASTBOOT;
1031 	return (state_func_t) runcom;
1032 }
1033 
1034 /*
1035  * Run the system startup script.
1036  */
1037 static state_func_t
1038 runcom(void)
1039 {
1040 	state_func_t next_transition;
1041 
1042 	if ((next_transition = run_script(_PATH_RUNCOM)) != NULL)
1043 		return next_transition;
1044 
1045 	runcom_mode = AUTOBOOT;		/* the default */
1046 	return (state_func_t) read_ttys;
1047 }
1048 
1049 /*
1050  * Run a shell script.
1051  * Returns 0 on success, otherwise the next transition to enter:
1052  *  - single_user if fork/execv/waitpid failed, or if the script
1053  *    terminated with a signal or exit code != 0.
1054  *  - death_single if a SIGTERM was delivered to init(8).
1055  */
1056 static state_func_t
1057 run_script(const char *script)
1058 {
1059 	pid_t pid, wpid;
1060 	int error, status;
1061 	char *argv[4];
1062 	const char *shell;
1063 	struct sigaction sa;
1064 
1065 	shell = get_shell();
1066 
1067 	if ((pid = fork()) == 0) {
1068 		sigemptyset(&sa.sa_mask);
1069 		sa.sa_flags = 0;
1070 		sa.sa_handler = SIG_IGN;
1071 		sigaction(SIGTSTP, &sa, (struct sigaction *)0);
1072 		sigaction(SIGHUP, &sa, (struct sigaction *)0);
1073 
1074 		open_console();
1075 
1076 		char _sh[]		= "sh";
1077 		char _autoboot[]	= "autoboot";
1078 
1079 		argv[0] = _sh;
1080 		argv[1] = __DECONST(char *, script);
1081 		argv[2] = runcom_mode == AUTOBOOT ? _autoboot : 0;
1082 		argv[3] = 0;
1083 
1084 		sigprocmask(SIG_SETMASK, &sa.sa_mask, (sigset_t *) 0);
1085 
1086 #ifdef LOGIN_CAP
1087 		setprocresources(RESOURCE_RC);
1088 #endif
1089 
1090 		/*
1091 		 * Try to directly execute the script first.  If it
1092 		 * fails, try the old method of passing the script path
1093 		 * to sh(1).  Don't complain if it fails because of
1094 		 * the missing execute bit.
1095 		 */
1096 		error = access(script, X_OK);
1097 		if (error == 0) {
1098 			execv(script, argv + 1);
1099 			warning("can't exec %s: %m", script);
1100 		} else if (errno != EACCES) {
1101 			warning("can't access %s: %m", script);
1102 		}
1103 
1104 		execv(shell, argv);
1105 		stall("can't exec %s for %s: %m", shell, script);
1106 		_exit(1);	/* force single user mode */
1107 	}
1108 
1109 	if (pid == -1) {
1110 		emergency("can't fork for %s on %s: %m", shell, script);
1111 		while (waitpid(-1, (int *) 0, WNOHANG) > 0)
1112 			continue;
1113 		sleep(STALL_TIMEOUT);
1114 		return (state_func_t) single_user;
1115 	}
1116 
1117 	/*
1118 	 * Copied from single_user().  This is a bit paranoid.
1119 	 */
1120 	requested_transition = 0;
1121 	do {
1122 		if ((wpid = waitpid(-1, &status, WUNTRACED)) != -1)
1123 			collect_child(wpid);
1124 		if (wpid == -1) {
1125 			if (requested_transition == death_single ||
1126 			    requested_transition == reroot)
1127 				return (state_func_t) requested_transition;
1128 			if (errno == EINTR)
1129 				continue;
1130 			warning("wait for %s on %s failed: %m; going to "
1131 			    "single user mode", shell, script);
1132 			return (state_func_t) single_user;
1133 		}
1134 		if (wpid == pid && WIFSTOPPED(status)) {
1135 			warning("init: %s on %s stopped, restarting\n",
1136 			    shell, script);
1137 			kill(pid, SIGCONT);
1138 			wpid = -1;
1139 		}
1140 	} while (wpid != pid);
1141 
1142 	if (WIFSIGNALED(status) && WTERMSIG(status) == SIGTERM &&
1143 	    requested_transition == catatonia) {
1144 		/* /etc/rc executed /sbin/reboot; wait for the end quietly */
1145 		sigset_t s;
1146 
1147 		sigfillset(&s);
1148 		for (;;)
1149 			sigsuspend(&s);
1150 	}
1151 
1152 	if (!WIFEXITED(status)) {
1153 		warning("%s on %s terminated abnormally, going to single "
1154 		    "user mode", shell, script);
1155 		return (state_func_t) single_user;
1156 	}
1157 
1158 	if (WEXITSTATUS(status))
1159 		return (state_func_t) single_user;
1160 
1161 	return (state_func_t) 0;
1162 }
1163 
1164 /*
1165  * Open the session database.
1166  *
1167  * NB: We could pass in the size here; is it necessary?
1168  */
1169 static int
1170 start_session_db(void)
1171 {
1172 	if (session_db && (*session_db->close)(session_db))
1173 		emergency("session database close: %s", strerror(errno));
1174 	if ((session_db = dbopen(NULL, O_RDWR, 0, DB_HASH, NULL)) == NULL) {
1175 		emergency("session database open: %s", strerror(errno));
1176 		return (1);
1177 	}
1178 	return (0);
1179 
1180 }
1181 
1182 /*
1183  * Add a new login session.
1184  */
1185 static void
1186 add_session(session_t *sp)
1187 {
1188 	DBT key;
1189 	DBT data;
1190 
1191 	key.data = &sp->se_process;
1192 	key.size = sizeof sp->se_process;
1193 	data.data = &sp;
1194 	data.size = sizeof sp;
1195 
1196 	if ((*session_db->put)(session_db, &key, &data, 0))
1197 		emergency("insert %d: %s", sp->se_process, strerror(errno));
1198 }
1199 
1200 /*
1201  * Delete an old login session.
1202  */
1203 static void
1204 del_session(session_t *sp)
1205 {
1206 	DBT key;
1207 
1208 	key.data = &sp->se_process;
1209 	key.size = sizeof sp->se_process;
1210 
1211 	if ((*session_db->del)(session_db, &key, 0))
1212 		emergency("delete %d: %s", sp->se_process, strerror(errno));
1213 }
1214 
1215 /*
1216  * Look up a login session by pid.
1217  */
1218 static session_t *
1219 find_session(pid_t pid)
1220 {
1221 	DBT key;
1222 	DBT data;
1223 	session_t *ret;
1224 
1225 	key.data = &pid;
1226 	key.size = sizeof pid;
1227 	if ((*session_db->get)(session_db, &key, &data, 0) != 0)
1228 		return 0;
1229 	bcopy(data.data, (char *)&ret, sizeof(ret));
1230 	return ret;
1231 }
1232 
1233 /*
1234  * Construct an argument vector from a command line.
1235  */
1236 static char **
1237 construct_argv(char *command)
1238 {
1239 	int argc = 0;
1240 	char **argv = (char **) malloc(((strlen(command) + 1) / 2 + 1)
1241 						* sizeof (char *));
1242 
1243 	if ((argv[argc++] = strk(command)) == NULL) {
1244 		free(argv);
1245 		return (NULL);
1246 	}
1247 	while ((argv[argc++] = strk((char *) 0)) != NULL)
1248 		continue;
1249 	return argv;
1250 }
1251 
1252 /*
1253  * Deallocate a session descriptor.
1254  */
1255 static void
1256 free_session(session_t *sp)
1257 {
1258 	free(sp->se_device);
1259 	if (sp->se_getty) {
1260 		free(sp->se_getty);
1261 		free(sp->se_getty_argv_space);
1262 		free(sp->se_getty_argv);
1263 	}
1264 	if (sp->se_window) {
1265 		free(sp->se_window);
1266 		free(sp->se_window_argv_space);
1267 		free(sp->se_window_argv);
1268 	}
1269 	if (sp->se_type)
1270 		free(sp->se_type);
1271 	free(sp);
1272 }
1273 
1274 /*
1275  * Allocate a new session descriptor.
1276  * Mark it SE_PRESENT.
1277  */
1278 static session_t *
1279 new_session(session_t *sprev, struct ttyent *typ)
1280 {
1281 	session_t *sp;
1282 
1283 	if ((typ->ty_status & TTY_ON) == 0 ||
1284 	    typ->ty_name == 0 ||
1285 	    typ->ty_getty == 0)
1286 		return 0;
1287 
1288 	sp = (session_t *) calloc(1, sizeof (session_t));
1289 
1290 	sp->se_flags |= SE_PRESENT;
1291 
1292 	if ((typ->ty_status & TTY_IFEXISTS) != 0)
1293 		sp->se_flags |= SE_IFEXISTS;
1294 
1295 	if ((typ->ty_status & TTY_IFCONSOLE) != 0)
1296 		sp->se_flags |= SE_IFCONSOLE;
1297 
1298 	if (asprintf(&sp->se_device, "%s%s", _PATH_DEV, typ->ty_name) < 0)
1299 		err(1, "asprintf");
1300 
1301 	if (setupargv(sp, typ) == 0) {
1302 		free_session(sp);
1303 		return (0);
1304 	}
1305 
1306 	sp->se_next = 0;
1307 	if (sprev == NULL) {
1308 		sessions = sp;
1309 		sp->se_prev = 0;
1310 	} else {
1311 		sprev->se_next = sp;
1312 		sp->se_prev = sprev;
1313 	}
1314 
1315 	return sp;
1316 }
1317 
1318 /*
1319  * Calculate getty and if useful window argv vectors.
1320  */
1321 static int
1322 setupargv(session_t *sp, struct ttyent *typ)
1323 {
1324 
1325 	if (sp->se_getty) {
1326 		free(sp->se_getty);
1327 		free(sp->se_getty_argv_space);
1328 		free(sp->se_getty_argv);
1329 	}
1330 	if (asprintf(&sp->se_getty, "%s %s", typ->ty_getty, typ->ty_name) < 0)
1331 		err(1, "asprintf");
1332 	sp->se_getty_argv_space = strdup(sp->se_getty);
1333 	sp->se_getty_argv = construct_argv(sp->se_getty_argv_space);
1334 	if (sp->se_getty_argv == NULL) {
1335 		warning("can't parse getty for port %s", sp->se_device);
1336 		free(sp->se_getty);
1337 		free(sp->se_getty_argv_space);
1338 		sp->se_getty = sp->se_getty_argv_space = 0;
1339 		return (0);
1340 	}
1341 	if (sp->se_window) {
1342 		free(sp->se_window);
1343 		free(sp->se_window_argv_space);
1344 		free(sp->se_window_argv);
1345 	}
1346 	sp->se_window = sp->se_window_argv_space = 0;
1347 	sp->se_window_argv = 0;
1348 	if (typ->ty_window) {
1349 		sp->se_window = strdup(typ->ty_window);
1350 		sp->se_window_argv_space = strdup(sp->se_window);
1351 		sp->se_window_argv = construct_argv(sp->se_window_argv_space);
1352 		if (sp->se_window_argv == NULL) {
1353 			warning("can't parse window for port %s",
1354 			    sp->se_device);
1355 			free(sp->se_window_argv_space);
1356 			free(sp->se_window);
1357 			sp->se_window = sp->se_window_argv_space = 0;
1358 			return (0);
1359 		}
1360 	}
1361 	if (sp->se_type)
1362 		free(sp->se_type);
1363 	sp->se_type = typ->ty_type ? strdup(typ->ty_type) : 0;
1364 	return (1);
1365 }
1366 
1367 /*
1368  * Walk the list of ttys and create sessions for each active line.
1369  */
1370 static state_func_t
1371 read_ttys(void)
1372 {
1373 	session_t *sp, *snext;
1374 	struct ttyent *typ;
1375 
1376 	/*
1377 	 * Destroy any previous session state.
1378 	 * There shouldn't be any, but just in case...
1379 	 */
1380 	for (sp = sessions; sp; sp = snext) {
1381 		snext = sp->se_next;
1382 		free_session(sp);
1383 	}
1384 	sessions = 0;
1385 	if (start_session_db())
1386 		return (state_func_t) single_user;
1387 
1388 	/*
1389 	 * Allocate a session entry for each active port.
1390 	 * Note that sp starts at 0.
1391 	 */
1392 	while ((typ = getttyent()) != NULL)
1393 		if ((snext = new_session(sp, typ)) != NULL)
1394 			sp = snext;
1395 
1396 	endttyent();
1397 
1398 	return (state_func_t) multi_user;
1399 }
1400 
1401 /*
1402  * Start a window system running.
1403  */
1404 static void
1405 start_window_system(session_t *sp)
1406 {
1407 	pid_t pid;
1408 	sigset_t mask;
1409 	char term[64], *env[2];
1410 	int status;
1411 
1412 	if ((pid = fork()) == -1) {
1413 		emergency("can't fork for window system on port %s: %m",
1414 		    sp->se_device);
1415 		/* hope that getty fails and we can try again */
1416 		return;
1417 	}
1418 	if (pid) {
1419 		waitpid(-1, &status, 0);
1420 		return;
1421 	}
1422 
1423 	/* reparent window process to the init to not make a zombie on exit */
1424 	if ((pid = fork()) == -1) {
1425 		emergency("can't fork for window system on port %s: %m",
1426 		    sp->se_device);
1427 		_exit(1);
1428 	}
1429 	if (pid)
1430 		_exit(0);
1431 
1432 	sigemptyset(&mask);
1433 	sigprocmask(SIG_SETMASK, &mask, (sigset_t *) 0);
1434 
1435 	if (setsid() < 0)
1436 		emergency("setsid failed (window) %m");
1437 
1438 #ifdef LOGIN_CAP
1439 	setprocresources(RESOURCE_WINDOW);
1440 #endif
1441 	if (sp->se_type) {
1442 		/* Don't use malloc after fork */
1443 		strcpy(term, "TERM=");
1444 		strlcat(term, sp->se_type, sizeof(term));
1445 		env[0] = term;
1446 		env[1] = 0;
1447 	}
1448 	else
1449 		env[0] = 0;
1450 	execve(sp->se_window_argv[0], sp->se_window_argv, env);
1451 	stall("can't exec window system '%s' for port %s: %m",
1452 		sp->se_window_argv[0], sp->se_device);
1453 	_exit(1);
1454 }
1455 
1456 /*
1457  * Start a login session running.
1458  */
1459 static pid_t
1460 start_getty(session_t *sp)
1461 {
1462 	pid_t pid;
1463 	sigset_t mask;
1464 	time_t current_time = time((time_t *) 0);
1465 	int too_quick = 0;
1466 	char term[64], *env[2];
1467 
1468 	if (current_time >= sp->se_started &&
1469 	    current_time - sp->se_started < GETTY_SPACING) {
1470 		if (++sp->se_nspace > GETTY_NSPACE) {
1471 			sp->se_nspace = 0;
1472 			too_quick = 1;
1473 		}
1474 	} else
1475 		sp->se_nspace = 0;
1476 
1477 	/*
1478 	 * fork(), not vfork() -- we can't afford to block.
1479 	 */
1480 	if ((pid = fork()) == -1) {
1481 		emergency("can't fork for getty on port %s: %m", sp->se_device);
1482 		return -1;
1483 	}
1484 
1485 	if (pid)
1486 		return pid;
1487 
1488 	if (too_quick) {
1489 		warning("getty repeating too quickly on port %s, sleeping %d secs",
1490 		    sp->se_device, GETTY_SLEEP);
1491 		sleep((unsigned) GETTY_SLEEP);
1492 	}
1493 
1494 	if (sp->se_window) {
1495 		start_window_system(sp);
1496 		sleep(WINDOW_WAIT);
1497 	}
1498 
1499 	sigemptyset(&mask);
1500 	sigprocmask(SIG_SETMASK, &mask, (sigset_t *) 0);
1501 
1502 #ifdef LOGIN_CAP
1503 	setprocresources(RESOURCE_GETTY);
1504 #endif
1505 	if (sp->se_type) {
1506 		/* Don't use malloc after fork */
1507 		strcpy(term, "TERM=");
1508 		strlcat(term, sp->se_type, sizeof(term));
1509 		env[0] = term;
1510 		env[1] = 0;
1511 	} else
1512 		env[0] = 0;
1513 	execve(sp->se_getty_argv[0], sp->se_getty_argv, env);
1514 	stall("can't exec getty '%s' for port %s: %m",
1515 		sp->se_getty_argv[0], sp->se_device);
1516 	_exit(1);
1517 }
1518 
1519 /*
1520  * Return 1 if the session is defined as "onifexists"
1521  * or "onifconsole" and the device node does not exist.
1522  */
1523 static int
1524 session_has_no_tty(session_t *sp)
1525 {
1526 	int fd;
1527 
1528 	if ((sp->se_flags & SE_IFEXISTS) == 0 &&
1529 	    (sp->se_flags & SE_IFCONSOLE) == 0)
1530 		return (0);
1531 
1532 	fd = open(sp->se_device, O_RDONLY | O_NONBLOCK, 0);
1533 	if (fd < 0) {
1534 		if (errno == ENOENT)
1535 			return (1);
1536 		return (0);
1537 	}
1538 
1539 	close(fd);
1540 	return (0);
1541 }
1542 
1543 /*
1544  * Collect exit status for a child.
1545  * If an exiting login, start a new login running.
1546  */
1547 static void
1548 collect_child(pid_t pid)
1549 {
1550 	session_t *sp, *sprev, *snext;
1551 
1552 	if (! sessions)
1553 		return;
1554 
1555 	if (! (sp = find_session(pid)))
1556 		return;
1557 
1558 	del_session(sp);
1559 	sp->se_process = 0;
1560 
1561 	if (sp->se_flags & SE_SHUTDOWN ||
1562 	    session_has_no_tty(sp)) {
1563 		if ((sprev = sp->se_prev) != NULL)
1564 			sprev->se_next = sp->se_next;
1565 		else
1566 			sessions = sp->se_next;
1567 		if ((snext = sp->se_next) != NULL)
1568 			snext->se_prev = sp->se_prev;
1569 		free_session(sp);
1570 		return;
1571 	}
1572 
1573 	if ((pid = start_getty(sp)) == -1) {
1574 		/* serious trouble */
1575 		requested_transition = clean_ttys;
1576 		return;
1577 	}
1578 
1579 	sp->se_process = pid;
1580 	sp->se_started = time((time_t *) 0);
1581 	add_session(sp);
1582 }
1583 
1584 /*
1585  * Catch a signal and request a state transition.
1586  */
1587 static void
1588 transition_handler(int sig)
1589 {
1590 
1591 	switch (sig) {
1592 	case SIGHUP:
1593 		if (current_state == read_ttys || current_state == multi_user ||
1594 		    current_state == clean_ttys || current_state == catatonia)
1595 			requested_transition = clean_ttys;
1596 		break;
1597 	case SIGWINCH:
1598 	case SIGUSR2:
1599 		howto = sig == SIGUSR2 ? RB_POWEROFF : RB_POWERCYCLE;
1600 	case SIGUSR1:
1601 		howto |= RB_HALT;
1602 	case SIGINT:
1603 		Reboot = TRUE;
1604 	case SIGTERM:
1605 		if (current_state == read_ttys || current_state == multi_user ||
1606 		    current_state == clean_ttys || current_state == catatonia)
1607 			requested_transition = death;
1608 		else
1609 			requested_transition = death_single;
1610 		break;
1611 	case SIGTSTP:
1612 		if (current_state == runcom || current_state == read_ttys ||
1613 		    current_state == clean_ttys ||
1614 		    current_state == multi_user || current_state == catatonia)
1615 			requested_transition = catatonia;
1616 		break;
1617 	case SIGEMT:
1618 		requested_transition = reroot;
1619 		break;
1620 	default:
1621 		requested_transition = 0;
1622 		break;
1623 	}
1624 }
1625 
1626 /*
1627  * Take the system multiuser.
1628  */
1629 static state_func_t
1630 multi_user(void)
1631 {
1632 	pid_t pid;
1633 	session_t *sp;
1634 
1635 	requested_transition = 0;
1636 
1637 	/*
1638 	 * If the administrator has not set the security level to -1
1639 	 * to indicate that the kernel should not run multiuser in secure
1640 	 * mode, and the run script has not set a higher level of security
1641 	 * than level 1, then put the kernel into secure mode.
1642 	 */
1643 	if (getsecuritylevel() == 0)
1644 		setsecuritylevel(1);
1645 
1646 	for (sp = sessions; sp; sp = sp->se_next) {
1647 		if (sp->se_process)
1648 			continue;
1649 		if (session_has_no_tty(sp))
1650 			continue;
1651 		if ((pid = start_getty(sp)) == -1) {
1652 			/* serious trouble */
1653 			requested_transition = clean_ttys;
1654 			break;
1655 		}
1656 		sp->se_process = pid;
1657 		sp->se_started = time((time_t *) 0);
1658 		add_session(sp);
1659 	}
1660 
1661 	while (!requested_transition)
1662 		if ((pid = waitpid(-1, (int *) 0, 0)) != -1)
1663 			collect_child(pid);
1664 
1665 	return (state_func_t) requested_transition;
1666 }
1667 
1668 /*
1669  * This is an (n*2)+(n^2) algorithm.  We hope it isn't run often...
1670  */
1671 static state_func_t
1672 clean_ttys(void)
1673 {
1674 	session_t *sp, *sprev;
1675 	struct ttyent *typ;
1676 	int devlen;
1677 	char *old_getty, *old_window, *old_type;
1678 
1679 	/*
1680 	 * mark all sessions for death, (!SE_PRESENT)
1681 	 * as we find or create new ones they'll be marked as keepers,
1682 	 * we'll later nuke all the ones not found in /etc/ttys
1683 	 */
1684 	for (sp = sessions; sp != NULL; sp = sp->se_next)
1685 		sp->se_flags &= ~SE_PRESENT;
1686 
1687 	devlen = sizeof(_PATH_DEV) - 1;
1688 	while ((typ = getttyent()) != NULL) {
1689 		for (sprev = 0, sp = sessions; sp; sprev = sp, sp = sp->se_next)
1690 			if (strcmp(typ->ty_name, sp->se_device + devlen) == 0)
1691 				break;
1692 
1693 		if (sp) {
1694 			/* we want this one to live */
1695 			sp->se_flags |= SE_PRESENT;
1696 			if ((typ->ty_status & TTY_ON) == 0 ||
1697 			    typ->ty_getty == 0) {
1698 				sp->se_flags |= SE_SHUTDOWN;
1699 				kill(sp->se_process, SIGHUP);
1700 				continue;
1701 			}
1702 			sp->se_flags &= ~SE_SHUTDOWN;
1703 			old_getty = sp->se_getty ? strdup(sp->se_getty) : 0;
1704 			old_window = sp->se_window ? strdup(sp->se_window) : 0;
1705 			old_type = sp->se_type ? strdup(sp->se_type) : 0;
1706 			if (setupargv(sp, typ) == 0) {
1707 				warning("can't parse getty for port %s",
1708 					sp->se_device);
1709 				sp->se_flags |= SE_SHUTDOWN;
1710 				kill(sp->se_process, SIGHUP);
1711 			}
1712 			else if (   !old_getty
1713 				 || (!old_type && sp->se_type)
1714 				 || (old_type && !sp->se_type)
1715 				 || (!old_window && sp->se_window)
1716 				 || (old_window && !sp->se_window)
1717 				 || (strcmp(old_getty, sp->se_getty) != 0)
1718 				 || (old_window && strcmp(old_window, sp->se_window) != 0)
1719 				 || (old_type && strcmp(old_type, sp->se_type) != 0)
1720 				) {
1721 				/* Don't set SE_SHUTDOWN here */
1722 				sp->se_nspace = 0;
1723 				sp->se_started = 0;
1724 				kill(sp->se_process, SIGHUP);
1725 			}
1726 			if (old_getty)
1727 				free(old_getty);
1728 			if (old_window)
1729 				free(old_window);
1730 			if (old_type)
1731 				free(old_type);
1732 			continue;
1733 		}
1734 
1735 		new_session(sprev, typ);
1736 	}
1737 
1738 	endttyent();
1739 
1740 	/*
1741 	 * sweep through and kill all deleted sessions
1742 	 * ones who's /etc/ttys line was deleted (SE_PRESENT unset)
1743 	 */
1744 	for (sp = sessions; sp != NULL; sp = sp->se_next) {
1745 		if ((sp->se_flags & SE_PRESENT) == 0) {
1746 			sp->se_flags |= SE_SHUTDOWN;
1747 			kill(sp->se_process, SIGHUP);
1748 		}
1749 	}
1750 
1751 	return (state_func_t) multi_user;
1752 }
1753 
1754 /*
1755  * Block further logins.
1756  */
1757 static state_func_t
1758 catatonia(void)
1759 {
1760 	session_t *sp;
1761 
1762 	for (sp = sessions; sp; sp = sp->se_next)
1763 		sp->se_flags |= SE_SHUTDOWN;
1764 
1765 	return (state_func_t) multi_user;
1766 }
1767 
1768 /*
1769  * Note SIGALRM.
1770  */
1771 static void
1772 alrm_handler(int sig)
1773 {
1774 
1775 	(void)sig;
1776 	clang = 1;
1777 }
1778 
1779 /*
1780  * Bring the system down to single user.
1781  */
1782 static state_func_t
1783 death(void)
1784 {
1785 	int block, blocked;
1786 	size_t len;
1787 
1788 	/* Temporarily block suspend. */
1789 	len = sizeof(blocked);
1790 	block = 1;
1791 	if (sysctlbyname("kern.suspend_blocked", &blocked, &len,
1792 	    &block, sizeof(block)) == -1)
1793 		blocked = 0;
1794 
1795 	/*
1796 	 * Also revoke the TTY here.  Because runshutdown() may reopen
1797 	 * the TTY whose getty we're killing here, there is no guarantee
1798 	 * runshutdown() will perform the initial open() call, causing
1799 	 * the terminal attributes to be misconfigured.
1800 	 */
1801 	revoke_ttys();
1802 
1803 	/* Try to run the rc.shutdown script within a period of time */
1804 	runshutdown();
1805 
1806 	/* Unblock suspend if we blocked it. */
1807 	if (!blocked)
1808 		sysctlbyname("kern.suspend_blocked", NULL, NULL,
1809 		    &blocked, sizeof(blocked));
1810 
1811 	return (state_func_t) death_single;
1812 }
1813 
1814 /*
1815  * Do what is necessary to reinitialize single user mode or reboot
1816  * from an incomplete state.
1817  */
1818 static state_func_t
1819 death_single(void)
1820 {
1821 	int i;
1822 	pid_t pid;
1823 	static const int death_sigs[2] = { SIGTERM, SIGKILL };
1824 
1825 	revoke(_PATH_CONSOLE);
1826 
1827 	for (i = 0; i < 2; ++i) {
1828 		if (kill(-1, death_sigs[i]) == -1 && errno == ESRCH)
1829 			return (state_func_t) single_user;
1830 
1831 		clang = 0;
1832 		alarm(DEATH_WATCH);
1833 		do
1834 			if ((pid = waitpid(-1, (int *)0, 0)) != -1)
1835 				collect_child(pid);
1836 		while (clang == 0 && errno != ECHILD);
1837 
1838 		if (errno == ECHILD)
1839 			return (state_func_t) single_user;
1840 	}
1841 
1842 	warning("some processes would not die; ps axl advised");
1843 
1844 	return (state_func_t) single_user;
1845 }
1846 
1847 static void
1848 revoke_ttys(void)
1849 {
1850 	session_t *sp;
1851 
1852 	for (sp = sessions; sp; sp = sp->se_next) {
1853 		sp->se_flags |= SE_SHUTDOWN;
1854 		kill(sp->se_process, SIGHUP);
1855 		revoke(sp->se_device);
1856 	}
1857 }
1858 
1859 /*
1860  * Run the system shutdown script.
1861  *
1862  * Exit codes:      XXX I should document more
1863  * -2       shutdown script terminated abnormally
1864  * -1       fatal error - can't run script
1865  * 0        good.
1866  * >0       some error (exit code)
1867  */
1868 static int
1869 runshutdown(void)
1870 {
1871 	pid_t pid, wpid;
1872 	int error, status;
1873 	int shutdowntimeout;
1874 	size_t len;
1875 	char *argv[4];
1876 	const char *shell;
1877 	struct sigaction sa;
1878 	struct stat sb;
1879 
1880 	/*
1881 	 * rc.shutdown is optional, so to prevent any unnecessary
1882 	 * complaints from the shell we simply don't run it if the
1883 	 * file does not exist. If the stat() here fails for other
1884 	 * reasons, we'll let the shell complain.
1885 	 */
1886 	if (stat(_PATH_RUNDOWN, &sb) == -1 && errno == ENOENT)
1887 		return 0;
1888 
1889 	shell = get_shell();
1890 
1891 	if ((pid = fork()) == 0) {
1892 		sigemptyset(&sa.sa_mask);
1893 		sa.sa_flags = 0;
1894 		sa.sa_handler = SIG_IGN;
1895 		sigaction(SIGTSTP, &sa, (struct sigaction *)0);
1896 		sigaction(SIGHUP, &sa, (struct sigaction *)0);
1897 
1898 		open_console();
1899 
1900 		char _sh[]	= "sh";
1901 		char _reboot[]	= "reboot";
1902 		char _single[]	= "single";
1903 		char _path_rundown[] = _PATH_RUNDOWN;
1904 
1905 		argv[0] = _sh;
1906 		argv[1] = _path_rundown;
1907 		argv[2] = Reboot ? _reboot : _single;
1908 		argv[3] = 0;
1909 
1910 		sigprocmask(SIG_SETMASK, &sa.sa_mask, (sigset_t *) 0);
1911 
1912 #ifdef LOGIN_CAP
1913 		setprocresources(RESOURCE_RC);
1914 #endif
1915 
1916 		/*
1917 		 * Try to directly execute the script first.  If it
1918 		 * fails, try the old method of passing the script path
1919 		 * to sh(1).  Don't complain if it fails because of
1920 		 * the missing execute bit.
1921 		 */
1922 		error = access(_path_rundown, X_OK);
1923 		if (error == 0) {
1924 			execv(_path_rundown, argv + 1);
1925 			warning("can't exec %s: %m", _path_rundown);
1926 		} else if (errno != EACCES) {
1927 			warning("can't access %s: %m", _path_rundown);
1928 		}
1929 
1930 		execv(shell, argv);
1931 		warning("can't exec %s for %s: %m", shell, _PATH_RUNDOWN);
1932 		_exit(1);	/* force single user mode */
1933 	}
1934 
1935 	if (pid == -1) {
1936 		emergency("can't fork for %s on %s: %m", shell, _PATH_RUNDOWN);
1937 		while (waitpid(-1, (int *) 0, WNOHANG) > 0)
1938 			continue;
1939 		sleep(STALL_TIMEOUT);
1940 		return -1;
1941 	}
1942 
1943 	len = sizeof(shutdowntimeout);
1944 	if (sysctlbyname("kern.init_shutdown_timeout", &shutdowntimeout, &len,
1945 	    NULL, 0) == -1 || shutdowntimeout < 2)
1946 		shutdowntimeout = DEATH_SCRIPT;
1947 	alarm(shutdowntimeout);
1948 	clang = 0;
1949 	/*
1950 	 * Copied from single_user().  This is a bit paranoid.
1951 	 * Use the same ALRM handler.
1952 	 */
1953 	do {
1954 		if ((wpid = waitpid(-1, &status, WUNTRACED)) != -1)
1955 			collect_child(wpid);
1956 		if (clang == 1) {
1957 			/* we were waiting for the sub-shell */
1958 			kill(wpid, SIGTERM);
1959 			warning("timeout expired for %s on %s: %m; going to "
1960 			    "single user mode", shell, _PATH_RUNDOWN);
1961 			return -1;
1962 		}
1963 		if (wpid == -1) {
1964 			if (errno == EINTR)
1965 				continue;
1966 			warning("wait for %s on %s failed: %m; going to "
1967 			    "single user mode", shell, _PATH_RUNDOWN);
1968 			return -1;
1969 		}
1970 		if (wpid == pid && WIFSTOPPED(status)) {
1971 			warning("init: %s on %s stopped, restarting\n",
1972 				shell, _PATH_RUNDOWN);
1973 			kill(pid, SIGCONT);
1974 			wpid = -1;
1975 		}
1976 	} while (wpid != pid && !clang);
1977 
1978 	/* Turn off the alarm */
1979 	alarm(0);
1980 
1981 	if (WIFSIGNALED(status) && WTERMSIG(status) == SIGTERM &&
1982 	    requested_transition == catatonia) {
1983 		/*
1984 		 * /etc/rc.shutdown executed /sbin/reboot;
1985 		 * wait for the end quietly
1986 		 */
1987 		sigset_t s;
1988 
1989 		sigfillset(&s);
1990 		for (;;)
1991 			sigsuspend(&s);
1992 	}
1993 
1994 	if (!WIFEXITED(status)) {
1995 		warning("%s on %s terminated abnormally, going to "
1996 		    "single user mode", shell, _PATH_RUNDOWN);
1997 		return -2;
1998 	}
1999 
2000 	if ((status = WEXITSTATUS(status)) != 0)
2001 		warning("%s returned status %d", _PATH_RUNDOWN, status);
2002 
2003 	return status;
2004 }
2005 
2006 static char *
2007 strk(char *p)
2008 {
2009 	static char *t;
2010 	char *q;
2011 	int c;
2012 
2013 	if (p)
2014 		t = p;
2015 	if (!t)
2016 		return 0;
2017 
2018 	c = *t;
2019 	while (c == ' ' || c == '\t' )
2020 		c = *++t;
2021 	if (!c) {
2022 		t = 0;
2023 		return 0;
2024 	}
2025 	q = t;
2026 	if (c == '\'') {
2027 		c = *++t;
2028 		q = t;
2029 		while (c && c != '\'')
2030 			c = *++t;
2031 		if (!c)  /* unterminated string */
2032 			q = t = 0;
2033 		else
2034 			*t++ = 0;
2035 	} else {
2036 		while (c && c != ' ' && c != '\t' )
2037 			c = *++t;
2038 		*t++ = 0;
2039 		if (!c)
2040 			t = 0;
2041 	}
2042 	return q;
2043 }
2044 
2045 #ifdef LOGIN_CAP
2046 static void
2047 setprocresources(const char *cname)
2048 {
2049 	login_cap_t *lc;
2050 	if ((lc = login_getclassbyname(cname, NULL)) != NULL) {
2051 		setusercontext(lc, (struct passwd*)NULL, 0,
2052 		    LOGIN_SETPRIORITY | LOGIN_SETRESOURCES |
2053 		    LOGIN_SETLOGINCLASS | LOGIN_SETCPUMASK);
2054 		login_close(lc);
2055 	}
2056 }
2057 #endif
2058