xref: /freebsd/sbin/init/init.c (revision 47dd1d1b619cc035b82b49a91a25544309ff95ae)
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 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 		execv(shell, argv);
1090 		stall("can't exec %s for %s: %m", shell, script);
1091 		_exit(1);	/* force single user mode */
1092 	}
1093 
1094 	if (pid == -1) {
1095 		emergency("can't fork for %s on %s: %m", shell, script);
1096 		while (waitpid(-1, (int *) 0, WNOHANG) > 0)
1097 			continue;
1098 		sleep(STALL_TIMEOUT);
1099 		return (state_func_t) single_user;
1100 	}
1101 
1102 	/*
1103 	 * Copied from single_user().  This is a bit paranoid.
1104 	 */
1105 	requested_transition = 0;
1106 	do {
1107 		if ((wpid = waitpid(-1, &status, WUNTRACED)) != -1)
1108 			collect_child(wpid);
1109 		if (wpid == -1) {
1110 			if (requested_transition == death_single ||
1111 			    requested_transition == reroot)
1112 				return (state_func_t) requested_transition;
1113 			if (errno == EINTR)
1114 				continue;
1115 			warning("wait for %s on %s failed: %m; going to "
1116 			    "single user mode", shell, script);
1117 			return (state_func_t) single_user;
1118 		}
1119 		if (wpid == pid && WIFSTOPPED(status)) {
1120 			warning("init: %s on %s stopped, restarting\n",
1121 			    shell, script);
1122 			kill(pid, SIGCONT);
1123 			wpid = -1;
1124 		}
1125 	} while (wpid != pid);
1126 
1127 	if (WIFSIGNALED(status) && WTERMSIG(status) == SIGTERM &&
1128 	    requested_transition == catatonia) {
1129 		/* /etc/rc executed /sbin/reboot; wait for the end quietly */
1130 		sigset_t s;
1131 
1132 		sigfillset(&s);
1133 		for (;;)
1134 			sigsuspend(&s);
1135 	}
1136 
1137 	if (!WIFEXITED(status)) {
1138 		warning("%s on %s terminated abnormally, going to single "
1139 		    "user mode", shell, script);
1140 		return (state_func_t) single_user;
1141 	}
1142 
1143 	if (WEXITSTATUS(status))
1144 		return (state_func_t) single_user;
1145 
1146 	return (state_func_t) 0;
1147 }
1148 
1149 /*
1150  * Open the session database.
1151  *
1152  * NB: We could pass in the size here; is it necessary?
1153  */
1154 static int
1155 start_session_db(void)
1156 {
1157 	if (session_db && (*session_db->close)(session_db))
1158 		emergency("session database close: %s", strerror(errno));
1159 	if ((session_db = dbopen(NULL, O_RDWR, 0, DB_HASH, NULL)) == NULL) {
1160 		emergency("session database open: %s", strerror(errno));
1161 		return (1);
1162 	}
1163 	return (0);
1164 
1165 }
1166 
1167 /*
1168  * Add a new login session.
1169  */
1170 static void
1171 add_session(session_t *sp)
1172 {
1173 	DBT key;
1174 	DBT data;
1175 
1176 	key.data = &sp->se_process;
1177 	key.size = sizeof sp->se_process;
1178 	data.data = &sp;
1179 	data.size = sizeof sp;
1180 
1181 	if ((*session_db->put)(session_db, &key, &data, 0))
1182 		emergency("insert %d: %s", sp->se_process, strerror(errno));
1183 }
1184 
1185 /*
1186  * Delete an old login session.
1187  */
1188 static void
1189 del_session(session_t *sp)
1190 {
1191 	DBT key;
1192 
1193 	key.data = &sp->se_process;
1194 	key.size = sizeof sp->se_process;
1195 
1196 	if ((*session_db->del)(session_db, &key, 0))
1197 		emergency("delete %d: %s", sp->se_process, strerror(errno));
1198 }
1199 
1200 /*
1201  * Look up a login session by pid.
1202  */
1203 static session_t *
1204 find_session(pid_t pid)
1205 {
1206 	DBT key;
1207 	DBT data;
1208 	session_t *ret;
1209 
1210 	key.data = &pid;
1211 	key.size = sizeof pid;
1212 	if ((*session_db->get)(session_db, &key, &data, 0) != 0)
1213 		return 0;
1214 	bcopy(data.data, (char *)&ret, sizeof(ret));
1215 	return ret;
1216 }
1217 
1218 /*
1219  * Construct an argument vector from a command line.
1220  */
1221 static char **
1222 construct_argv(char *command)
1223 {
1224 	int argc = 0;
1225 	char **argv = (char **) malloc(((strlen(command) + 1) / 2 + 1)
1226 						* sizeof (char *));
1227 
1228 	if ((argv[argc++] = strk(command)) == NULL) {
1229 		free(argv);
1230 		return (NULL);
1231 	}
1232 	while ((argv[argc++] = strk((char *) 0)) != NULL)
1233 		continue;
1234 	return argv;
1235 }
1236 
1237 /*
1238  * Deallocate a session descriptor.
1239  */
1240 static void
1241 free_session(session_t *sp)
1242 {
1243 	free(sp->se_device);
1244 	if (sp->se_getty) {
1245 		free(sp->se_getty);
1246 		free(sp->se_getty_argv_space);
1247 		free(sp->se_getty_argv);
1248 	}
1249 	if (sp->se_window) {
1250 		free(sp->se_window);
1251 		free(sp->se_window_argv_space);
1252 		free(sp->se_window_argv);
1253 	}
1254 	if (sp->se_type)
1255 		free(sp->se_type);
1256 	free(sp);
1257 }
1258 
1259 /*
1260  * Allocate a new session descriptor.
1261  * Mark it SE_PRESENT.
1262  */
1263 static session_t *
1264 new_session(session_t *sprev, struct ttyent *typ)
1265 {
1266 	session_t *sp;
1267 
1268 	if ((typ->ty_status & TTY_ON) == 0 ||
1269 	    typ->ty_name == 0 ||
1270 	    typ->ty_getty == 0)
1271 		return 0;
1272 
1273 	sp = (session_t *) calloc(1, sizeof (session_t));
1274 
1275 	sp->se_flags |= SE_PRESENT;
1276 
1277 	if ((typ->ty_status & TTY_IFEXISTS) != 0)
1278 		sp->se_flags |= SE_IFEXISTS;
1279 
1280 	if ((typ->ty_status & TTY_IFCONSOLE) != 0)
1281 		sp->se_flags |= SE_IFCONSOLE;
1282 
1283 	if (asprintf(&sp->se_device, "%s%s", _PATH_DEV, typ->ty_name) < 0)
1284 		err(1, "asprintf");
1285 
1286 	if (setupargv(sp, typ) == 0) {
1287 		free_session(sp);
1288 		return (0);
1289 	}
1290 
1291 	sp->se_next = 0;
1292 	if (sprev == NULL) {
1293 		sessions = sp;
1294 		sp->se_prev = 0;
1295 	} else {
1296 		sprev->se_next = sp;
1297 		sp->se_prev = sprev;
1298 	}
1299 
1300 	return sp;
1301 }
1302 
1303 /*
1304  * Calculate getty and if useful window argv vectors.
1305  */
1306 static int
1307 setupargv(session_t *sp, struct ttyent *typ)
1308 {
1309 
1310 	if (sp->se_getty) {
1311 		free(sp->se_getty);
1312 		free(sp->se_getty_argv_space);
1313 		free(sp->se_getty_argv);
1314 	}
1315 	if (asprintf(&sp->se_getty, "%s %s", typ->ty_getty, typ->ty_name) < 0)
1316 		err(1, "asprintf");
1317 	sp->se_getty_argv_space = strdup(sp->se_getty);
1318 	sp->se_getty_argv = construct_argv(sp->se_getty_argv_space);
1319 	if (sp->se_getty_argv == NULL) {
1320 		warning("can't parse getty for port %s", sp->se_device);
1321 		free(sp->se_getty);
1322 		free(sp->se_getty_argv_space);
1323 		sp->se_getty = sp->se_getty_argv_space = 0;
1324 		return (0);
1325 	}
1326 	if (sp->se_window) {
1327 		free(sp->se_window);
1328 		free(sp->se_window_argv_space);
1329 		free(sp->se_window_argv);
1330 	}
1331 	sp->se_window = sp->se_window_argv_space = 0;
1332 	sp->se_window_argv = 0;
1333 	if (typ->ty_window) {
1334 		sp->se_window = strdup(typ->ty_window);
1335 		sp->se_window_argv_space = strdup(sp->se_window);
1336 		sp->se_window_argv = construct_argv(sp->se_window_argv_space);
1337 		if (sp->se_window_argv == NULL) {
1338 			warning("can't parse window for port %s",
1339 			    sp->se_device);
1340 			free(sp->se_window_argv_space);
1341 			free(sp->se_window);
1342 			sp->se_window = sp->se_window_argv_space = 0;
1343 			return (0);
1344 		}
1345 	}
1346 	if (sp->se_type)
1347 		free(sp->se_type);
1348 	sp->se_type = typ->ty_type ? strdup(typ->ty_type) : 0;
1349 	return (1);
1350 }
1351 
1352 /*
1353  * Walk the list of ttys and create sessions for each active line.
1354  */
1355 static state_func_t
1356 read_ttys(void)
1357 {
1358 	session_t *sp, *snext;
1359 	struct ttyent *typ;
1360 
1361 	/*
1362 	 * Destroy any previous session state.
1363 	 * There shouldn't be any, but just in case...
1364 	 */
1365 	for (sp = sessions; sp; sp = snext) {
1366 		snext = sp->se_next;
1367 		free_session(sp);
1368 	}
1369 	sessions = 0;
1370 	if (start_session_db())
1371 		return (state_func_t) single_user;
1372 
1373 	/*
1374 	 * Allocate a session entry for each active port.
1375 	 * Note that sp starts at 0.
1376 	 */
1377 	while ((typ = getttyent()) != NULL)
1378 		if ((snext = new_session(sp, typ)) != NULL)
1379 			sp = snext;
1380 
1381 	endttyent();
1382 
1383 	return (state_func_t) multi_user;
1384 }
1385 
1386 /*
1387  * Start a window system running.
1388  */
1389 static void
1390 start_window_system(session_t *sp)
1391 {
1392 	pid_t pid;
1393 	sigset_t mask;
1394 	char term[64], *env[2];
1395 	int status;
1396 
1397 	if ((pid = fork()) == -1) {
1398 		emergency("can't fork for window system on port %s: %m",
1399 		    sp->se_device);
1400 		/* hope that getty fails and we can try again */
1401 		return;
1402 	}
1403 	if (pid) {
1404 		waitpid(-1, &status, 0);
1405 		return;
1406 	}
1407 
1408 	/* reparent window process to the init to not make a zombie on exit */
1409 	if ((pid = fork()) == -1) {
1410 		emergency("can't fork for window system on port %s: %m",
1411 		    sp->se_device);
1412 		_exit(1);
1413 	}
1414 	if (pid)
1415 		_exit(0);
1416 
1417 	sigemptyset(&mask);
1418 	sigprocmask(SIG_SETMASK, &mask, (sigset_t *) 0);
1419 
1420 	if (setsid() < 0)
1421 		emergency("setsid failed (window) %m");
1422 
1423 #ifdef LOGIN_CAP
1424 	setprocresources(RESOURCE_WINDOW);
1425 #endif
1426 	if (sp->se_type) {
1427 		/* Don't use malloc after fork */
1428 		strcpy(term, "TERM=");
1429 		strlcat(term, sp->se_type, sizeof(term));
1430 		env[0] = term;
1431 		env[1] = 0;
1432 	}
1433 	else
1434 		env[0] = 0;
1435 	execve(sp->se_window_argv[0], sp->se_window_argv, env);
1436 	stall("can't exec window system '%s' for port %s: %m",
1437 		sp->se_window_argv[0], sp->se_device);
1438 	_exit(1);
1439 }
1440 
1441 /*
1442  * Start a login session running.
1443  */
1444 static pid_t
1445 start_getty(session_t *sp)
1446 {
1447 	pid_t pid;
1448 	sigset_t mask;
1449 	time_t current_time = time((time_t *) 0);
1450 	int too_quick = 0;
1451 	char term[64], *env[2];
1452 
1453 	if (current_time >= sp->se_started &&
1454 	    current_time - sp->se_started < GETTY_SPACING) {
1455 		if (++sp->se_nspace > GETTY_NSPACE) {
1456 			sp->se_nspace = 0;
1457 			too_quick = 1;
1458 		}
1459 	} else
1460 		sp->se_nspace = 0;
1461 
1462 	/*
1463 	 * fork(), not vfork() -- we can't afford to block.
1464 	 */
1465 	if ((pid = fork()) == -1) {
1466 		emergency("can't fork for getty on port %s: %m", sp->se_device);
1467 		return -1;
1468 	}
1469 
1470 	if (pid)
1471 		return pid;
1472 
1473 	if (too_quick) {
1474 		warning("getty repeating too quickly on port %s, sleeping %d secs",
1475 		    sp->se_device, GETTY_SLEEP);
1476 		sleep((unsigned) GETTY_SLEEP);
1477 	}
1478 
1479 	if (sp->se_window) {
1480 		start_window_system(sp);
1481 		sleep(WINDOW_WAIT);
1482 	}
1483 
1484 	sigemptyset(&mask);
1485 	sigprocmask(SIG_SETMASK, &mask, (sigset_t *) 0);
1486 
1487 #ifdef LOGIN_CAP
1488 	setprocresources(RESOURCE_GETTY);
1489 #endif
1490 	if (sp->se_type) {
1491 		/* Don't use malloc after fork */
1492 		strcpy(term, "TERM=");
1493 		strlcat(term, sp->se_type, sizeof(term));
1494 		env[0] = term;
1495 		env[1] = 0;
1496 	} else
1497 		env[0] = 0;
1498 	execve(sp->se_getty_argv[0], sp->se_getty_argv, env);
1499 	stall("can't exec getty '%s' for port %s: %m",
1500 		sp->se_getty_argv[0], sp->se_device);
1501 	_exit(1);
1502 }
1503 
1504 /*
1505  * Return 1 if the session is defined as "onifexists"
1506  * or "onifconsole" and the device node does not exist.
1507  */
1508 static int
1509 session_has_no_tty(session_t *sp)
1510 {
1511 	int fd;
1512 
1513 	if ((sp->se_flags & SE_IFEXISTS) == 0 &&
1514 	    (sp->se_flags & SE_IFCONSOLE) == 0)
1515 		return (0);
1516 
1517 	fd = open(sp->se_device, O_RDONLY | O_NONBLOCK, 0);
1518 	if (fd < 0) {
1519 		if (errno == ENOENT)
1520 			return (1);
1521 		return (0);
1522 	}
1523 
1524 	close(fd);
1525 	return (0);
1526 }
1527 
1528 /*
1529  * Collect exit status for a child.
1530  * If an exiting login, start a new login running.
1531  */
1532 static void
1533 collect_child(pid_t pid)
1534 {
1535 	session_t *sp, *sprev, *snext;
1536 
1537 	if (! sessions)
1538 		return;
1539 
1540 	if (! (sp = find_session(pid)))
1541 		return;
1542 
1543 	del_session(sp);
1544 	sp->se_process = 0;
1545 
1546 	if (sp->se_flags & SE_SHUTDOWN ||
1547 	    session_has_no_tty(sp)) {
1548 		if ((sprev = sp->se_prev) != NULL)
1549 			sprev->se_next = sp->se_next;
1550 		else
1551 			sessions = sp->se_next;
1552 		if ((snext = sp->se_next) != NULL)
1553 			snext->se_prev = sp->se_prev;
1554 		free_session(sp);
1555 		return;
1556 	}
1557 
1558 	if ((pid = start_getty(sp)) == -1) {
1559 		/* serious trouble */
1560 		requested_transition = clean_ttys;
1561 		return;
1562 	}
1563 
1564 	sp->se_process = pid;
1565 	sp->se_started = time((time_t *) 0);
1566 	add_session(sp);
1567 }
1568 
1569 /*
1570  * Catch a signal and request a state transition.
1571  */
1572 static void
1573 transition_handler(int sig)
1574 {
1575 
1576 	switch (sig) {
1577 	case SIGHUP:
1578 		if (current_state == read_ttys || current_state == multi_user ||
1579 		    current_state == clean_ttys || current_state == catatonia)
1580 			requested_transition = clean_ttys;
1581 		break;
1582 	case SIGWINCH:
1583 	case SIGUSR2:
1584 		howto = sig == SIGUSR2 ? RB_POWEROFF : RB_POWERCYCLE;
1585 	case SIGUSR1:
1586 		howto |= RB_HALT;
1587 	case SIGINT:
1588 		Reboot = TRUE;
1589 	case SIGTERM:
1590 		if (current_state == read_ttys || current_state == multi_user ||
1591 		    current_state == clean_ttys || current_state == catatonia)
1592 			requested_transition = death;
1593 		else
1594 			requested_transition = death_single;
1595 		break;
1596 	case SIGTSTP:
1597 		if (current_state == runcom || current_state == read_ttys ||
1598 		    current_state == clean_ttys ||
1599 		    current_state == multi_user || current_state == catatonia)
1600 			requested_transition = catatonia;
1601 		break;
1602 	case SIGEMT:
1603 		requested_transition = reroot;
1604 		break;
1605 	default:
1606 		requested_transition = 0;
1607 		break;
1608 	}
1609 }
1610 
1611 /*
1612  * Take the system multiuser.
1613  */
1614 static state_func_t
1615 multi_user(void)
1616 {
1617 	pid_t pid;
1618 	session_t *sp;
1619 
1620 	requested_transition = 0;
1621 
1622 	/*
1623 	 * If the administrator has not set the security level to -1
1624 	 * to indicate that the kernel should not run multiuser in secure
1625 	 * mode, and the run script has not set a higher level of security
1626 	 * than level 1, then put the kernel into secure mode.
1627 	 */
1628 	if (getsecuritylevel() == 0)
1629 		setsecuritylevel(1);
1630 
1631 	for (sp = sessions; sp; sp = sp->se_next) {
1632 		if (sp->se_process)
1633 			continue;
1634 		if (session_has_no_tty(sp))
1635 			continue;
1636 		if ((pid = start_getty(sp)) == -1) {
1637 			/* serious trouble */
1638 			requested_transition = clean_ttys;
1639 			break;
1640 		}
1641 		sp->se_process = pid;
1642 		sp->se_started = time((time_t *) 0);
1643 		add_session(sp);
1644 	}
1645 
1646 	while (!requested_transition)
1647 		if ((pid = waitpid(-1, (int *) 0, 0)) != -1)
1648 			collect_child(pid);
1649 
1650 	return (state_func_t) requested_transition;
1651 }
1652 
1653 /*
1654  * This is an (n*2)+(n^2) algorithm.  We hope it isn't run often...
1655  */
1656 static state_func_t
1657 clean_ttys(void)
1658 {
1659 	session_t *sp, *sprev;
1660 	struct ttyent *typ;
1661 	int devlen;
1662 	char *old_getty, *old_window, *old_type;
1663 
1664 	/*
1665 	 * mark all sessions for death, (!SE_PRESENT)
1666 	 * as we find or create new ones they'll be marked as keepers,
1667 	 * we'll later nuke all the ones not found in /etc/ttys
1668 	 */
1669 	for (sp = sessions; sp != NULL; sp = sp->se_next)
1670 		sp->se_flags &= ~SE_PRESENT;
1671 
1672 	devlen = sizeof(_PATH_DEV) - 1;
1673 	while ((typ = getttyent()) != NULL) {
1674 		for (sprev = 0, sp = sessions; sp; sprev = sp, sp = sp->se_next)
1675 			if (strcmp(typ->ty_name, sp->se_device + devlen) == 0)
1676 				break;
1677 
1678 		if (sp) {
1679 			/* we want this one to live */
1680 			sp->se_flags |= SE_PRESENT;
1681 			if ((typ->ty_status & TTY_ON) == 0 ||
1682 			    typ->ty_getty == 0) {
1683 				sp->se_flags |= SE_SHUTDOWN;
1684 				kill(sp->se_process, SIGHUP);
1685 				continue;
1686 			}
1687 			sp->se_flags &= ~SE_SHUTDOWN;
1688 			old_getty = sp->se_getty ? strdup(sp->se_getty) : 0;
1689 			old_window = sp->se_window ? strdup(sp->se_window) : 0;
1690 			old_type = sp->se_type ? strdup(sp->se_type) : 0;
1691 			if (setupargv(sp, typ) == 0) {
1692 				warning("can't parse getty for port %s",
1693 					sp->se_device);
1694 				sp->se_flags |= SE_SHUTDOWN;
1695 				kill(sp->se_process, SIGHUP);
1696 			}
1697 			else if (   !old_getty
1698 				 || (!old_type && sp->se_type)
1699 				 || (old_type && !sp->se_type)
1700 				 || (!old_window && sp->se_window)
1701 				 || (old_window && !sp->se_window)
1702 				 || (strcmp(old_getty, sp->se_getty) != 0)
1703 				 || (old_window && strcmp(old_window, sp->se_window) != 0)
1704 				 || (old_type && strcmp(old_type, sp->se_type) != 0)
1705 				) {
1706 				/* Don't set SE_SHUTDOWN here */
1707 				sp->se_nspace = 0;
1708 				sp->se_started = 0;
1709 				kill(sp->se_process, SIGHUP);
1710 			}
1711 			if (old_getty)
1712 				free(old_getty);
1713 			if (old_window)
1714 				free(old_window);
1715 			if (old_type)
1716 				free(old_type);
1717 			continue;
1718 		}
1719 
1720 		new_session(sprev, typ);
1721 	}
1722 
1723 	endttyent();
1724 
1725 	/*
1726 	 * sweep through and kill all deleted sessions
1727 	 * ones who's /etc/ttys line was deleted (SE_PRESENT unset)
1728 	 */
1729 	for (sp = sessions; sp != NULL; sp = sp->se_next) {
1730 		if ((sp->se_flags & SE_PRESENT) == 0) {
1731 			sp->se_flags |= SE_SHUTDOWN;
1732 			kill(sp->se_process, SIGHUP);
1733 		}
1734 	}
1735 
1736 	return (state_func_t) multi_user;
1737 }
1738 
1739 /*
1740  * Block further logins.
1741  */
1742 static state_func_t
1743 catatonia(void)
1744 {
1745 	session_t *sp;
1746 
1747 	for (sp = sessions; sp; sp = sp->se_next)
1748 		sp->se_flags |= SE_SHUTDOWN;
1749 
1750 	return (state_func_t) multi_user;
1751 }
1752 
1753 /*
1754  * Note SIGALRM.
1755  */
1756 static void
1757 alrm_handler(int sig)
1758 {
1759 
1760 	(void)sig;
1761 	clang = 1;
1762 }
1763 
1764 /*
1765  * Bring the system down to single user.
1766  */
1767 static state_func_t
1768 death(void)
1769 {
1770 	int block, blocked;
1771 	size_t len;
1772 
1773 	/* Temporarily block suspend. */
1774 	len = sizeof(blocked);
1775 	block = 1;
1776 	if (sysctlbyname("kern.suspend_blocked", &blocked, &len,
1777 	    &block, sizeof(block)) == -1)
1778 		blocked = 0;
1779 
1780 	/*
1781 	 * Also revoke the TTY here.  Because runshutdown() may reopen
1782 	 * the TTY whose getty we're killing here, there is no guarantee
1783 	 * runshutdown() will perform the initial open() call, causing
1784 	 * the terminal attributes to be misconfigured.
1785 	 */
1786 	revoke_ttys();
1787 
1788 	/* Try to run the rc.shutdown script within a period of time */
1789 	runshutdown();
1790 
1791 	/* Unblock suspend if we blocked it. */
1792 	if (!blocked)
1793 		sysctlbyname("kern.suspend_blocked", NULL, NULL,
1794 		    &blocked, sizeof(blocked));
1795 
1796 	return (state_func_t) death_single;
1797 }
1798 
1799 /*
1800  * Do what is necessary to reinitialize single user mode or reboot
1801  * from an incomplete state.
1802  */
1803 static state_func_t
1804 death_single(void)
1805 {
1806 	int i;
1807 	pid_t pid;
1808 	static const int death_sigs[2] = { SIGTERM, SIGKILL };
1809 
1810 	revoke(_PATH_CONSOLE);
1811 
1812 	for (i = 0; i < 2; ++i) {
1813 		if (kill(-1, death_sigs[i]) == -1 && errno == ESRCH)
1814 			return (state_func_t) single_user;
1815 
1816 		clang = 0;
1817 		alarm(DEATH_WATCH);
1818 		do
1819 			if ((pid = waitpid(-1, (int *)0, 0)) != -1)
1820 				collect_child(pid);
1821 		while (clang == 0 && errno != ECHILD);
1822 
1823 		if (errno == ECHILD)
1824 			return (state_func_t) single_user;
1825 	}
1826 
1827 	warning("some processes would not die; ps axl advised");
1828 
1829 	return (state_func_t) single_user;
1830 }
1831 
1832 static void
1833 revoke_ttys(void)
1834 {
1835 	session_t *sp;
1836 
1837 	for (sp = sessions; sp; sp = sp->se_next) {
1838 		sp->se_flags |= SE_SHUTDOWN;
1839 		kill(sp->se_process, SIGHUP);
1840 		revoke(sp->se_device);
1841 	}
1842 }
1843 
1844 /*
1845  * Run the system shutdown script.
1846  *
1847  * Exit codes:      XXX I should document more
1848  * -2       shutdown script terminated abnormally
1849  * -1       fatal error - can't run script
1850  * 0        good.
1851  * >0       some error (exit code)
1852  */
1853 static int
1854 runshutdown(void)
1855 {
1856 	pid_t pid, wpid;
1857 	int status;
1858 	int shutdowntimeout;
1859 	size_t len;
1860 	char *argv[4];
1861 	const char *shell;
1862 	struct sigaction sa;
1863 	struct stat sb;
1864 
1865 	/*
1866 	 * rc.shutdown is optional, so to prevent any unnecessary
1867 	 * complaints from the shell we simply don't run it if the
1868 	 * file does not exist. If the stat() here fails for other
1869 	 * reasons, we'll let the shell complain.
1870 	 */
1871 	if (stat(_PATH_RUNDOWN, &sb) == -1 && errno == ENOENT)
1872 		return 0;
1873 
1874 	shell = get_shell();
1875 
1876 	if ((pid = fork()) == 0) {
1877 		sigemptyset(&sa.sa_mask);
1878 		sa.sa_flags = 0;
1879 		sa.sa_handler = SIG_IGN;
1880 		sigaction(SIGTSTP, &sa, (struct sigaction *)0);
1881 		sigaction(SIGHUP, &sa, (struct sigaction *)0);
1882 
1883 		open_console();
1884 
1885 		char _sh[]	= "sh";
1886 		char _reboot[]	= "reboot";
1887 		char _single[]	= "single";
1888 		char _path_rundown[] = _PATH_RUNDOWN;
1889 
1890 		argv[0] = _sh;
1891 		argv[1] = _path_rundown;
1892 		argv[2] = Reboot ? _reboot : _single;
1893 		argv[3] = 0;
1894 
1895 		sigprocmask(SIG_SETMASK, &sa.sa_mask, (sigset_t *) 0);
1896 
1897 #ifdef LOGIN_CAP
1898 		setprocresources(RESOURCE_RC);
1899 #endif
1900 		execv(shell, argv);
1901 		warning("can't exec %s for %s: %m", shell, _PATH_RUNDOWN);
1902 		_exit(1);	/* force single user mode */
1903 	}
1904 
1905 	if (pid == -1) {
1906 		emergency("can't fork for %s on %s: %m", shell, _PATH_RUNDOWN);
1907 		while (waitpid(-1, (int *) 0, WNOHANG) > 0)
1908 			continue;
1909 		sleep(STALL_TIMEOUT);
1910 		return -1;
1911 	}
1912 
1913 	len = sizeof(shutdowntimeout);
1914 	if (sysctlbyname("kern.init_shutdown_timeout", &shutdowntimeout, &len,
1915 	    NULL, 0) == -1 || shutdowntimeout < 2)
1916 		shutdowntimeout = DEATH_SCRIPT;
1917 	alarm(shutdowntimeout);
1918 	clang = 0;
1919 	/*
1920 	 * Copied from single_user().  This is a bit paranoid.
1921 	 * Use the same ALRM handler.
1922 	 */
1923 	do {
1924 		if ((wpid = waitpid(-1, &status, WUNTRACED)) != -1)
1925 			collect_child(wpid);
1926 		if (clang == 1) {
1927 			/* we were waiting for the sub-shell */
1928 			kill(wpid, SIGTERM);
1929 			warning("timeout expired for %s on %s: %m; going to "
1930 			    "single user mode", shell, _PATH_RUNDOWN);
1931 			return -1;
1932 		}
1933 		if (wpid == -1) {
1934 			if (errno == EINTR)
1935 				continue;
1936 			warning("wait for %s on %s failed: %m; going to "
1937 			    "single user mode", shell, _PATH_RUNDOWN);
1938 			return -1;
1939 		}
1940 		if (wpid == pid && WIFSTOPPED(status)) {
1941 			warning("init: %s on %s stopped, restarting\n",
1942 				shell, _PATH_RUNDOWN);
1943 			kill(pid, SIGCONT);
1944 			wpid = -1;
1945 		}
1946 	} while (wpid != pid && !clang);
1947 
1948 	/* Turn off the alarm */
1949 	alarm(0);
1950 
1951 	if (WIFSIGNALED(status) && WTERMSIG(status) == SIGTERM &&
1952 	    requested_transition == catatonia) {
1953 		/*
1954 		 * /etc/rc.shutdown executed /sbin/reboot;
1955 		 * wait for the end quietly
1956 		 */
1957 		sigset_t s;
1958 
1959 		sigfillset(&s);
1960 		for (;;)
1961 			sigsuspend(&s);
1962 	}
1963 
1964 	if (!WIFEXITED(status)) {
1965 		warning("%s on %s terminated abnormally, going to "
1966 		    "single user mode", shell, _PATH_RUNDOWN);
1967 		return -2;
1968 	}
1969 
1970 	if ((status = WEXITSTATUS(status)) != 0)
1971 		warning("%s returned status %d", _PATH_RUNDOWN, status);
1972 
1973 	return status;
1974 }
1975 
1976 static char *
1977 strk(char *p)
1978 {
1979 	static char *t;
1980 	char *q;
1981 	int c;
1982 
1983 	if (p)
1984 		t = p;
1985 	if (!t)
1986 		return 0;
1987 
1988 	c = *t;
1989 	while (c == ' ' || c == '\t' )
1990 		c = *++t;
1991 	if (!c) {
1992 		t = 0;
1993 		return 0;
1994 	}
1995 	q = t;
1996 	if (c == '\'') {
1997 		c = *++t;
1998 		q = t;
1999 		while (c && c != '\'')
2000 			c = *++t;
2001 		if (!c)  /* unterminated string */
2002 			q = t = 0;
2003 		else
2004 			*t++ = 0;
2005 	} else {
2006 		while (c && c != ' ' && c != '\t' )
2007 			c = *++t;
2008 		*t++ = 0;
2009 		if (!c)
2010 			t = 0;
2011 	}
2012 	return q;
2013 }
2014 
2015 #ifdef LOGIN_CAP
2016 static void
2017 setprocresources(const char *cname)
2018 {
2019 	login_cap_t *lc;
2020 	if ((lc = login_getclassbyname(cname, NULL)) != NULL) {
2021 		setusercontext(lc, (struct passwd*)NULL, 0,
2022 		    LOGIN_SETPRIORITY | LOGIN_SETRESOURCES |
2023 		    LOGIN_SETLOGINCLASS | LOGIN_SETCPUMASK);
2024 		login_close(lc);
2025 	}
2026 }
2027 #endif
2028