xref: /freebsd/sbin/init/init.c (revision 49b49cda41feabe3439f7318e8bf40e3896c7bf4)
1 /*-
2  * Copyright (c) 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Donn Seeley at Berkeley Software Design, Inc.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #ifndef lint
34 static const char copyright[] =
35 "@(#) Copyright (c) 1991, 1993\n\
36 	The Regents of the University of California.  All rights reserved.\n";
37 #endif /* not lint */
38 
39 #ifndef lint
40 #if 0
41 static char sccsid[] = "@(#)init.c	8.1 (Berkeley) 7/15/93";
42 #endif
43 static const char rcsid[] =
44   "$FreeBSD$";
45 #endif /* not lint */
46 
47 #include <sys/param.h>
48 #include <sys/ioctl.h>
49 #include <sys/mman.h>
50 #include <sys/mount.h>
51 #include <sys/sysctl.h>
52 #include <sys/wait.h>
53 #include <sys/stat.h>
54 #include <sys/uio.h>
55 
56 #include <db.h>
57 #include <errno.h>
58 #include <fcntl.h>
59 #include <kenv.h>
60 #include <libutil.h>
61 #include <paths.h>
62 #include <signal.h>
63 #include <stdio.h>
64 #include <stdlib.h>
65 #include <string.h>
66 #include <syslog.h>
67 #include <time.h>
68 #include <ttyent.h>
69 #include <unistd.h>
70 #include <sys/reboot.h>
71 #include <err.h>
72 
73 #include <stdarg.h>
74 
75 #ifdef SECURE
76 #include <pwd.h>
77 #endif
78 
79 #ifdef LOGIN_CAP
80 #include <login_cap.h>
81 #endif
82 
83 #include "mntopts.h"
84 #include "pathnames.h"
85 
86 /*
87  * Sleep times; used to prevent thrashing.
88  */
89 #define	GETTY_SPACING		 5	/* N secs minimum getty spacing */
90 #define	GETTY_SLEEP		30	/* sleep N secs after spacing problem */
91 #define	GETTY_NSPACE		 3	/* max. spacing count to bring reaction */
92 #define	WINDOW_WAIT		 3	/* wait N secs after starting window */
93 #define	STALL_TIMEOUT		30	/* wait N secs after warning */
94 #define	DEATH_WATCH		10	/* wait N secs for procs to die */
95 #define	DEATH_SCRIPT		120	/* wait for 2min for /etc/rc.shutdown */
96 #define	RESOURCE_RC		"daemon"
97 #define	RESOURCE_WINDOW		"default"
98 #define	RESOURCE_GETTY		"default"
99 
100 static void handle(sig_t, ...);
101 static void delset(sigset_t *, ...);
102 
103 static void stall(const char *, ...) __printflike(1, 2);
104 static void warning(const char *, ...) __printflike(1, 2);
105 static void emergency(const char *, ...) __printflike(1, 2);
106 static void disaster(int);
107 static void badsys(int);
108 static void revoke_ttys(void);
109 static int  runshutdown(void);
110 static char *strk(char *);
111 
112 /*
113  * We really need a recursive typedef...
114  * The following at least guarantees that the return type of (*state_t)()
115  * is sufficiently wide to hold a function pointer.
116  */
117 typedef long (*state_func_t)(void);
118 typedef state_func_t (*state_t)(void);
119 
120 static state_func_t single_user(void);
121 static state_func_t runcom(void);
122 static state_func_t read_ttys(void);
123 static state_func_t multi_user(void);
124 static state_func_t clean_ttys(void);
125 static state_func_t catatonia(void);
126 static state_func_t death(void);
127 static state_func_t death_single(void);
128 static state_func_t reroot(void);
129 static state_func_t reroot_phase_two(void);
130 
131 static state_func_t run_script(const char *);
132 
133 static enum { AUTOBOOT, FASTBOOT } runcom_mode = AUTOBOOT;
134 #define FALSE	0
135 #define TRUE	1
136 
137 static int Reboot = FALSE;
138 static int howto = RB_AUTOBOOT;
139 
140 static int devfs;
141 
142 static void transition(state_t);
143 static state_t requested_transition;
144 static state_t current_state = death_single;
145 
146 static void open_console(void);
147 static const char *get_shell(void);
148 static void write_stderr(const char *message);
149 
150 typedef struct init_session {
151 	pid_t	se_process;		/* controlling process */
152 	time_t	se_started;		/* used to avoid thrashing */
153 	int	se_flags;		/* status of session */
154 #define	SE_SHUTDOWN	0x1		/* session won't be restarted */
155 #define	SE_PRESENT	0x2		/* session is in /etc/ttys */
156 	int	se_nspace;		/* spacing count */
157 	char	*se_device;		/* filename of port */
158 	char	*se_getty;		/* what to run on that port */
159 	char	*se_getty_argv_space;   /* pre-parsed argument array space */
160 	char	**se_getty_argv;	/* pre-parsed argument array */
161 	char	*se_window;		/* window system (started only once) */
162 	char	*se_window_argv_space;  /* pre-parsed argument array space */
163 	char	**se_window_argv;	/* pre-parsed argument array */
164 	char	*se_type;		/* default terminal type */
165 	struct	init_session *se_prev;
166 	struct	init_session *se_next;
167 } session_t;
168 
169 static void free_session(session_t *);
170 static session_t *new_session(session_t *, struct ttyent *);
171 static session_t *sessions;
172 
173 static char **construct_argv(char *);
174 static void start_window_system(session_t *);
175 static void collect_child(pid_t);
176 static pid_t start_getty(session_t *);
177 static void transition_handler(int);
178 static void alrm_handler(int);
179 static void setsecuritylevel(int);
180 static int getsecuritylevel(void);
181 static int setupargv(session_t *, struct ttyent *);
182 #ifdef LOGIN_CAP
183 static void setprocresources(const char *);
184 #endif
185 static int clang;
186 
187 static int start_session_db(void);
188 static void add_session(session_t *);
189 static void del_session(session_t *);
190 static session_t *find_session(pid_t);
191 static DB *session_db;
192 
193 /*
194  * The mother of all processes.
195  */
196 int
197 main(int argc, char *argv[])
198 {
199 	state_t initial_transition = runcom;
200 	char kenv_value[PATH_MAX];
201 	int c, error;
202 	struct sigaction sa;
203 	sigset_t mask;
204 
205 	/* Dispose of random users. */
206 	if (getuid() != 0)
207 		errx(1, "%s", strerror(EPERM));
208 
209 	/* System V users like to reexec init. */
210 	if (getpid() != 1) {
211 #ifdef COMPAT_SYSV_INIT
212 		/* So give them what they want */
213 		if (argc > 1) {
214 			if (strlen(argv[1]) == 1) {
215 				char runlevel = *argv[1];
216 				int sig;
217 
218 				switch (runlevel) {
219 				case '0': /* halt + poweroff */
220 					sig = SIGUSR2;
221 					break;
222 				case '1': /* single-user */
223 					sig = SIGTERM;
224 					break;
225 				case '6': /* reboot */
226 					sig = SIGINT;
227 					break;
228 				case 'c': /* block further logins */
229 					sig = SIGTSTP;
230 					break;
231 				case 'q': /* rescan /etc/ttys */
232 					sig = SIGHUP;
233 					break;
234 				case 'r': /* remount root */
235 					sig = SIGEMT;
236 					break;
237 				default:
238 					goto invalid;
239 				}
240 				kill(1, sig);
241 				_exit(0);
242 			} else
243 invalid:
244 				errx(1, "invalid run-level ``%s''", argv[1]);
245 		} else
246 #endif
247 			errx(1, "already running");
248 	}
249 	/*
250 	 * Note that this does NOT open a file...
251 	 * Does 'init' deserve its own facility number?
252 	 */
253 	openlog("init", LOG_CONS, LOG_AUTH);
254 
255 	/*
256 	 * Create an initial session.
257 	 */
258 	if (setsid() < 0 && (errno != EPERM || getsid(0) != 1))
259 		warning("initial setsid() failed: %m");
260 
261 	/*
262 	 * Establish an initial user so that programs running
263 	 * single user do not freak out and die (like passwd).
264 	 */
265 	if (setlogin("root") < 0)
266 		warning("setlogin() failed: %m");
267 
268 	/*
269 	 * This code assumes that we always get arguments through flags,
270 	 * never through bits set in some random machine register.
271 	 */
272 	while ((c = getopt(argc, argv, "dsfr")) != -1)
273 		switch (c) {
274 		case 'd':
275 			devfs = 1;
276 			break;
277 		case 's':
278 			initial_transition = single_user;
279 			break;
280 		case 'f':
281 			runcom_mode = FASTBOOT;
282 			break;
283 		case 'r':
284 			initial_transition = reroot_phase_two;
285 			break;
286 		default:
287 			warning("unrecognized flag '-%c'", c);
288 			break;
289 		}
290 
291 	if (optind != argc)
292 		warning("ignoring excess arguments");
293 
294 	/*
295 	 * We catch or block signals rather than ignore them,
296 	 * so that they get reset on exec.
297 	 */
298 	handle(badsys, SIGSYS, 0);
299 	handle(disaster, SIGABRT, SIGFPE, SIGILL, SIGSEGV, SIGBUS, SIGXCPU,
300 	    SIGXFSZ, 0);
301 	handle(transition_handler, SIGHUP, SIGINT, SIGEMT, SIGTERM, SIGTSTP,
302 	    SIGUSR1, SIGUSR2, 0);
303 	handle(alrm_handler, SIGALRM, 0);
304 	sigfillset(&mask);
305 	delset(&mask, SIGABRT, SIGFPE, SIGILL, SIGSEGV, SIGBUS, SIGSYS,
306 	    SIGXCPU, SIGXFSZ, SIGHUP, SIGINT, SIGEMT, SIGTERM, SIGTSTP,
307 	    SIGALRM, SIGUSR1, SIGUSR2, 0);
308 	sigprocmask(SIG_SETMASK, &mask, (sigset_t *) 0);
309 	sigemptyset(&sa.sa_mask);
310 	sa.sa_flags = 0;
311 	sa.sa_handler = SIG_IGN;
312 	sigaction(SIGTTIN, &sa, (struct sigaction *)0);
313 	sigaction(SIGTTOU, &sa, (struct sigaction *)0);
314 
315 	/*
316 	 * Paranoia.
317 	 */
318 	close(0);
319 	close(1);
320 	close(2);
321 
322 	if (kenv(KENV_GET, "init_script", kenv_value, sizeof(kenv_value)) > 0) {
323 		state_func_t next_transition;
324 
325 		if ((next_transition = run_script(kenv_value)) != 0)
326 			initial_transition = (state_t) next_transition;
327 	}
328 
329 	if (kenv(KENV_GET, "init_chroot", kenv_value, sizeof(kenv_value)) > 0) {
330 		if (chdir(kenv_value) != 0 || chroot(".") != 0)
331 			warning("Can't chroot to %s: %m", kenv_value);
332 	}
333 
334 	/*
335 	 * Additional check if devfs needs to be mounted:
336 	 * If "/" and "/dev" have the same device number,
337 	 * then it hasn't been mounted yet.
338 	 */
339 	if (!devfs) {
340 		struct stat stst;
341 		dev_t root_devno;
342 
343 		stat("/", &stst);
344 		root_devno = stst.st_dev;
345 		if (stat("/dev", &stst) != 0)
346 			warning("Can't stat /dev: %m");
347 		else if (stst.st_dev == root_devno)
348 			devfs++;
349 	}
350 
351 	if (devfs) {
352 		struct iovec iov[4];
353 		char *s;
354 		int i;
355 
356 		char _fstype[]	= "fstype";
357 		char _devfs[]	= "devfs";
358 		char _fspath[]	= "fspath";
359 		char _path_dev[]= _PATH_DEV;
360 
361 		iov[0].iov_base = _fstype;
362 		iov[0].iov_len = sizeof(_fstype);
363 		iov[1].iov_base = _devfs;
364 		iov[1].iov_len = sizeof(_devfs);
365 		iov[2].iov_base = _fspath;
366 		iov[2].iov_len = sizeof(_fspath);
367 		/*
368 		 * Try to avoid the trailing slash in _PATH_DEV.
369 		 * Be *very* defensive.
370 		 */
371 		s = strdup(_PATH_DEV);
372 		if (s != NULL) {
373 			i = strlen(s);
374 			if (i > 0 && s[i - 1] == '/')
375 				s[i - 1] = '\0';
376 			iov[3].iov_base = s;
377 			iov[3].iov_len = strlen(s) + 1;
378 		} else {
379 			iov[3].iov_base = _path_dev;
380 			iov[3].iov_len = sizeof(_path_dev);
381 		}
382 		nmount(iov, 4, 0);
383 		if (s != NULL)
384 			free(s);
385 	}
386 
387 	if (initial_transition != reroot_phase_two) {
388 		/*
389 		 * Unmount reroot leftovers.  This runs after init(8)
390 		 * gets reexecuted after reroot_phase_two() is done.
391 		 */
392 		error = unmount(_PATH_REROOT, MNT_FORCE);
393 		if (error != 0 && errno != EINVAL)
394 			warning("Cannot unmount %s: %m", _PATH_REROOT);
395 	}
396 
397 	/*
398 	 * Start the state machine.
399 	 */
400 	transition(initial_transition);
401 
402 	/*
403 	 * Should never reach here.
404 	 */
405 	return 1;
406 }
407 
408 /*
409  * Associate a function with a signal handler.
410  */
411 static void
412 handle(sig_t handler, ...)
413 {
414 	int sig;
415 	struct sigaction sa;
416 	sigset_t mask_everything;
417 	va_list ap;
418 	va_start(ap, handler);
419 
420 	sa.sa_handler = handler;
421 	sigfillset(&mask_everything);
422 
423 	while ((sig = va_arg(ap, int)) != 0) {
424 		sa.sa_mask = mask_everything;
425 		/* XXX SA_RESTART? */
426 		sa.sa_flags = sig == SIGCHLD ? SA_NOCLDSTOP : 0;
427 		sigaction(sig, &sa, (struct sigaction *) 0);
428 	}
429 	va_end(ap);
430 }
431 
432 /*
433  * Delete a set of signals from a mask.
434  */
435 static void
436 delset(sigset_t *maskp, ...)
437 {
438 	int sig;
439 	va_list ap;
440 	va_start(ap, maskp);
441 
442 	while ((sig = va_arg(ap, int)) != 0)
443 		sigdelset(maskp, sig);
444 	va_end(ap);
445 }
446 
447 /*
448  * Log a message and sleep for a while (to give someone an opportunity
449  * to read it and to save log or hardcopy output if the problem is chronic).
450  * NB: should send a message to the session logger to avoid blocking.
451  */
452 static void
453 stall(const char *message, ...)
454 {
455 	va_list ap;
456 	va_start(ap, message);
457 
458 	vsyslog(LOG_ALERT, message, ap);
459 	va_end(ap);
460 	sleep(STALL_TIMEOUT);
461 }
462 
463 /*
464  * Like stall(), but doesn't sleep.
465  * If cpp had variadic macros, the two functions could be #defines for another.
466  * NB: should send a message to the session logger to avoid blocking.
467  */
468 static void
469 warning(const char *message, ...)
470 {
471 	va_list ap;
472 	va_start(ap, message);
473 
474 	vsyslog(LOG_ALERT, message, ap);
475 	va_end(ap);
476 }
477 
478 /*
479  * Log an emergency message.
480  * NB: should send a message to the session logger to avoid blocking.
481  */
482 static void
483 emergency(const char *message, ...)
484 {
485 	va_list ap;
486 	va_start(ap, message);
487 
488 	vsyslog(LOG_EMERG, message, ap);
489 	va_end(ap);
490 }
491 
492 /*
493  * Catch a SIGSYS signal.
494  *
495  * These may arise if a system does not support sysctl.
496  * We tolerate up to 25 of these, then throw in the towel.
497  */
498 static void
499 badsys(int sig)
500 {
501 	static int badcount = 0;
502 
503 	if (badcount++ < 25)
504 		return;
505 	disaster(sig);
506 }
507 
508 /*
509  * Catch an unexpected signal.
510  */
511 static void
512 disaster(int sig)
513 {
514 
515 	emergency("fatal signal: %s",
516 	    (unsigned)sig < NSIG ? sys_siglist[sig] : "unknown signal");
517 
518 	sleep(STALL_TIMEOUT);
519 	_exit(sig);		/* reboot */
520 }
521 
522 /*
523  * Get the security level of the kernel.
524  */
525 static int
526 getsecuritylevel(void)
527 {
528 #ifdef KERN_SECURELVL
529 	int name[2], curlevel;
530 	size_t len;
531 
532 	name[0] = CTL_KERN;
533 	name[1] = KERN_SECURELVL;
534 	len = sizeof curlevel;
535 	if (sysctl(name, 2, &curlevel, &len, NULL, 0) == -1) {
536 		emergency("cannot get kernel security level: %s",
537 		    strerror(errno));
538 		return (-1);
539 	}
540 	return (curlevel);
541 #else
542 	return (-1);
543 #endif
544 }
545 
546 /*
547  * Set the security level of the kernel.
548  */
549 static void
550 setsecuritylevel(int newlevel)
551 {
552 #ifdef KERN_SECURELVL
553 	int name[2], curlevel;
554 
555 	curlevel = getsecuritylevel();
556 	if (newlevel == curlevel)
557 		return;
558 	name[0] = CTL_KERN;
559 	name[1] = KERN_SECURELVL;
560 	if (sysctl(name, 2, NULL, NULL, &newlevel, sizeof newlevel) == -1) {
561 		emergency(
562 		    "cannot change kernel security level from %d to %d: %s",
563 		    curlevel, newlevel, strerror(errno));
564 		return;
565 	}
566 #ifdef SECURE
567 	warning("kernel security level changed from %d to %d",
568 	    curlevel, newlevel);
569 #endif
570 #endif
571 }
572 
573 /*
574  * Change states in the finite state machine.
575  * The initial state is passed as an argument.
576  */
577 static void
578 transition(state_t s)
579 {
580 
581 	current_state = s;
582 	for (;;)
583 		current_state = (state_t) (*current_state)();
584 }
585 
586 /*
587  * Start a session and allocate a controlling terminal.
588  * Only called by children of init after forking.
589  */
590 static void
591 open_console(void)
592 {
593 	int fd;
594 
595 	/*
596 	 * Try to open /dev/console.  Open the device with O_NONBLOCK to
597 	 * prevent potential blocking on a carrier.
598 	 */
599 	revoke(_PATH_CONSOLE);
600 	if ((fd = open(_PATH_CONSOLE, O_RDWR | O_NONBLOCK)) != -1) {
601 		(void)fcntl(fd, F_SETFL, fcntl(fd, F_GETFL) & ~O_NONBLOCK);
602 		if (login_tty(fd) == 0)
603 			return;
604 		close(fd);
605 	}
606 
607 	/* No luck.  Log output to file if possible. */
608 	if ((fd = open(_PATH_DEVNULL, O_RDWR)) == -1) {
609 		stall("cannot open null device.");
610 		_exit(1);
611 	}
612 	if (fd != STDIN_FILENO) {
613 		dup2(fd, STDIN_FILENO);
614 		close(fd);
615 	}
616 	fd = open(_PATH_INITLOG, O_WRONLY | O_APPEND | O_CREAT, 0644);
617 	if (fd == -1)
618 		dup2(STDIN_FILENO, STDOUT_FILENO);
619 	else if (fd != STDOUT_FILENO) {
620 		dup2(fd, STDOUT_FILENO);
621 		close(fd);
622 	}
623 	dup2(STDOUT_FILENO, STDERR_FILENO);
624 }
625 
626 static const char *
627 get_shell(void)
628 {
629 	static char kenv_value[PATH_MAX];
630 
631 	if (kenv(KENV_GET, "init_shell", kenv_value, sizeof(kenv_value)) > 0)
632 		return kenv_value;
633 	else
634 		return _PATH_BSHELL;
635 }
636 
637 static void
638 write_stderr(const char *message)
639 {
640 
641 	write(STDERR_FILENO, message, strlen(message));
642 }
643 
644 static int
645 read_file(const char *path, void **bufp, size_t *bufsizep)
646 {
647 	struct stat sb;
648 	size_t bufsize;
649 	void *buf;
650 	ssize_t nbytes;
651 	int error, fd;
652 
653 	fd = open(path, O_RDONLY);
654 	if (fd < 0) {
655 		emergency("%s: %s", path, strerror(errno));
656 		return (-1);
657 	}
658 
659 	error = fstat(fd, &sb);
660 	if (error != 0) {
661 		emergency("fstat: %s", strerror(errno));
662 		close(fd);
663 		return (error);
664 	}
665 
666 	bufsize = sb.st_size;
667 	buf = malloc(bufsize);
668 	if (buf == NULL) {
669 		emergency("malloc: %s", strerror(errno));
670 		close(fd);
671 		return (error);
672 	}
673 
674 	nbytes = read(fd, buf, bufsize);
675 	if (nbytes != (ssize_t)bufsize) {
676 		emergency("read: %s", strerror(errno));
677 		close(fd);
678 		free(buf);
679 		return (error);
680 	}
681 
682 	error = close(fd);
683 	if (error != 0) {
684 		emergency("close: %s", strerror(errno));
685 		free(buf);
686 		return (error);
687 	}
688 
689 	*bufp = buf;
690 	*bufsizep = bufsize;
691 
692 	return (0);
693 }
694 
695 static int
696 create_file(const char *path, const void *buf, size_t bufsize)
697 {
698 	ssize_t nbytes;
699 	int error, fd;
700 
701 	fd = open(path, O_WRONLY | O_CREAT | O_EXCL, 0700);
702 	if (fd < 0) {
703 		emergency("%s: %s", path, strerror(errno));
704 		return (-1);
705 	}
706 
707 	nbytes = write(fd, buf, bufsize);
708 	if (nbytes != (ssize_t)bufsize) {
709 		emergency("write: %s", strerror(errno));
710 		close(fd);
711 		return (-1);
712 	}
713 
714 	error = close(fd);
715 	if (error != 0) {
716 		emergency("close: %s", strerror(errno));
717 		return (-1);
718 	}
719 
720 	return (0);
721 }
722 
723 static int
724 mount_tmpfs(const char *fspath)
725 {
726 	struct iovec *iov;
727 	char errmsg[255];
728 	int error, iovlen;
729 
730 	iov = NULL;
731 	iovlen = 0;
732 	memset(errmsg, 0, sizeof(errmsg));
733 	build_iovec(&iov, &iovlen, "fstype",
734 	    __DECONST(void *, "tmpfs"), (size_t)-1);
735 	build_iovec(&iov, &iovlen, "fspath",
736 	    __DECONST(void *, fspath), (size_t)-1);
737 	build_iovec(&iov, &iovlen, "errmsg",
738 	    errmsg, sizeof(errmsg));
739 
740 	error = nmount(iov, iovlen, 0);
741 	if (error != 0) {
742 		if (*errmsg != '\0') {
743 			emergency("cannot mount tmpfs on %s: %s: %s",
744 			    fspath, errmsg, strerror(errno));
745 		} else {
746 			emergency("cannot mount tmpfs on %s: %s",
747 			    fspath, strerror(errno));
748 		}
749 		return (error);
750 	}
751 	return (0);
752 }
753 
754 static state_func_t
755 reroot(void)
756 {
757 	void *buf;
758 	char init_path[PATH_MAX];
759 	size_t bufsize, init_path_len;
760 	int error, name[4];
761 
762 	buf = NULL;
763 	bufsize = 0;
764 
765 	name[0] = CTL_KERN;
766 	name[1] = KERN_PROC;
767 	name[2] = KERN_PROC_PATHNAME;
768 	name[3] = -1;
769 	init_path_len = sizeof(init_path);
770 	error = sysctl(name, 4, init_path, &init_path_len, NULL, 0);
771 	if (error != 0) {
772 		emergency("failed to get kern.proc.pathname: %s",
773 		    strerror(errno));
774 		goto out;
775 	}
776 
777 	revoke_ttys();
778 	runshutdown();
779 
780 	/*
781 	 * Make sure nobody can interfere with our scheme.
782 	 */
783 	error = kill(-1, SIGKILL);
784 	if (error != 0) {
785 		emergency("kill(2) failed: %s", strerror(errno));
786 		goto out;
787 	}
788 
789 	/*
790 	 * Copy the init binary into tmpfs, so that we can unmount
791 	 * the old rootfs without committing suicide.
792 	 */
793 	error = read_file(init_path, &buf, &bufsize);
794 	if (error != 0)
795 		goto out;
796 	error = mount_tmpfs(_PATH_REROOT);
797 	if (error != 0)
798 		goto out;
799 	error = create_file(_PATH_REROOT_INIT, buf, bufsize);
800 	if (error != 0)
801 		goto out;
802 
803 	/*
804 	 * Execute the temporary init.
805 	 */
806 	execl(_PATH_REROOT_INIT, _PATH_REROOT_INIT, "-r", NULL);
807 	emergency("cannot exec %s: %s", _PATH_REROOT_INIT, strerror(errno));
808 
809 out:
810 	emergency("reroot failed; going to single user mode");
811 	free(buf);
812 	return (state_func_t) single_user;
813 }
814 
815 static state_func_t
816 reroot_phase_two(void)
817 {
818 	char init_path[PATH_MAX], *path, *path_component;
819 	size_t init_path_len;
820 	int nbytes, error;
821 
822 	/*
823 	 * Ask the kernel to mount the new rootfs.
824 	 */
825 	error = reboot(RB_REROOT);
826 	if (error != 0) {
827 		emergency("RB_REBOOT failed: %s", strerror(errno));
828 		goto out;
829 	}
830 
831 	/*
832 	 * Figure out where the destination init(8) binary is.  Note that
833 	 * the path could be different than what we've started with.  Use
834 	 * the value from kenv, if set, or the one from sysctl otherwise.
835 	 * The latter defaults to a hardcoded value, but can be overridden
836 	 * by a build time option.
837 	 */
838 	nbytes = kenv(KENV_GET, "init_path", init_path, sizeof(init_path));
839 	if (nbytes <= 0) {
840 		init_path_len = sizeof(init_path);
841 		error = sysctlbyname("kern.init_path",
842 		    init_path, &init_path_len, NULL, 0);
843 		if (error != 0) {
844 			emergency("failed to retrieve kern.init_path: %s",
845 			    strerror(errno));
846 			goto out;
847 		}
848 	}
849 
850 	/*
851 	 * Repeat the init search logic from sys/kern/init_path.c
852 	 */
853 	path_component = init_path;
854 	while ((path = strsep(&path_component, ":")) != NULL) {
855 		/*
856 		 * Execute init(8) from the new rootfs.
857 		 */
858 		execl(path, path, NULL);
859 	}
860 	emergency("cannot exec init from %s: %s", init_path, strerror(errno));
861 
862 out:
863 	emergency("reroot failed; going to single user mode");
864 	return (state_func_t) single_user;
865 }
866 
867 /*
868  * Bring the system up single user.
869  */
870 static state_func_t
871 single_user(void)
872 {
873 	pid_t pid, wpid;
874 	int status;
875 	sigset_t mask;
876 	const char *shell;
877 	char *argv[2];
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 		reboot(howto);
893 		_exit(0);
894 	}
895 
896 	shell = get_shell();
897 
898 	if ((pid = fork()) == 0) {
899 		/*
900 		 * Start the single user session.
901 		 */
902 		open_console();
903 
904 #ifdef SECURE
905 		/*
906 		 * Check the root password.
907 		 * We don't care if the console is 'on' by default;
908 		 * it's the only tty that can be 'off' and 'secure'.
909 		 */
910 		typ = getttynam("console");
911 		pp = getpwnam("root");
912 		if (typ && (typ->ty_status & TTY_SECURE) == 0 &&
913 		    pp && *pp->pw_passwd) {
914 			write_stderr(banner);
915 			for (;;) {
916 				clear = getpass("Password:");
917 				if (clear == 0 || *clear == '\0')
918 					_exit(0);
919 				password = crypt(clear, pp->pw_passwd);
920 				bzero(clear, _PASSWORD_LEN);
921 				if (password == NULL ||
922 				    strcmp(password, pp->pw_passwd) == 0)
923 					break;
924 				warning("single-user login failed\n");
925 			}
926 		}
927 		endttyent();
928 		endpwent();
929 #endif /* SECURE */
930 
931 #ifdef DEBUGSHELL
932 		{
933 			char *cp = altshell;
934 			int num;
935 
936 #define	SHREQUEST "Enter full pathname of shell or RETURN for "
937 			write_stderr(SHREQUEST);
938 			write_stderr(shell);
939 			write_stderr(": ");
940 			while ((num = read(STDIN_FILENO, cp, 1)) != -1 &&
941 			    num != 0 && *cp != '\n' && cp < &altshell[127])
942 				cp++;
943 			*cp = '\0';
944 			if (altshell[0] != '\0')
945 				shell = altshell;
946 		}
947 #endif /* DEBUGSHELL */
948 
949 		/*
950 		 * Unblock signals.
951 		 * We catch all the interesting ones,
952 		 * and those are reset to SIG_DFL on exec.
953 		 */
954 		sigemptyset(&mask);
955 		sigprocmask(SIG_SETMASK, &mask, (sigset_t *) 0);
956 
957 		/*
958 		 * Fire off a shell.
959 		 * If the default one doesn't work, try the Bourne shell.
960 		 */
961 
962 		char name[] = "-sh";
963 
964 		argv[0] = name;
965 		argv[1] = 0;
966 		execv(shell, argv);
967 		emergency("can't exec %s for single user: %m", shell);
968 		execv(_PATH_BSHELL, argv);
969 		emergency("can't exec %s for single user: %m", _PATH_BSHELL);
970 		sleep(STALL_TIMEOUT);
971 		_exit(1);
972 	}
973 
974 	if (pid == -1) {
975 		/*
976 		 * We are seriously hosed.  Do our best.
977 		 */
978 		emergency("can't fork single-user shell, trying again");
979 		while (waitpid(-1, (int *) 0, WNOHANG) > 0)
980 			continue;
981 		return (state_func_t) single_user;
982 	}
983 
984 	requested_transition = 0;
985 	do {
986 		if ((wpid = waitpid(-1, &status, WUNTRACED)) != -1)
987 			collect_child(wpid);
988 		if (wpid == -1) {
989 			if (errno == EINTR)
990 				continue;
991 			warning("wait for single-user shell failed: %m; restarting");
992 			return (state_func_t) single_user;
993 		}
994 		if (wpid == pid && WIFSTOPPED(status)) {
995 			warning("init: shell stopped, restarting\n");
996 			kill(pid, SIGCONT);
997 			wpid = -1;
998 		}
999 	} while (wpid != pid && !requested_transition);
1000 
1001 	if (requested_transition)
1002 		return (state_func_t) requested_transition;
1003 
1004 	if (!WIFEXITED(status)) {
1005 		if (WTERMSIG(status) == SIGKILL) {
1006 			/*
1007 			 *  reboot(8) killed shell?
1008 			 */
1009 			warning("single user shell terminated.");
1010 			sleep(STALL_TIMEOUT);
1011 			_exit(0);
1012 		} else {
1013 			warning("single user shell terminated, restarting");
1014 			return (state_func_t) single_user;
1015 		}
1016 	}
1017 
1018 	runcom_mode = FASTBOOT;
1019 	return (state_func_t) runcom;
1020 }
1021 
1022 /*
1023  * Run the system startup script.
1024  */
1025 static state_func_t
1026 runcom(void)
1027 {
1028 	state_func_t next_transition;
1029 
1030 	if ((next_transition = run_script(_PATH_RUNCOM)) != 0)
1031 		return next_transition;
1032 
1033 	runcom_mode = AUTOBOOT;		/* the default */
1034 	return (state_func_t) read_ttys;
1035 }
1036 
1037 /*
1038  * Run a shell script.
1039  * Returns 0 on success, otherwise the next transition to enter:
1040  *  - single_user if fork/execv/waitpid failed, or if the script
1041  *    terminated with a signal or exit code != 0.
1042  *  - death_single if a SIGTERM was delivered to init(8).
1043  */
1044 static state_func_t
1045 run_script(const char *script)
1046 {
1047 	pid_t pid, wpid;
1048 	int status;
1049 	char *argv[4];
1050 	const char *shell;
1051 	struct sigaction sa;
1052 
1053 	shell = get_shell();
1054 
1055 	if ((pid = fork()) == 0) {
1056 		sigemptyset(&sa.sa_mask);
1057 		sa.sa_flags = 0;
1058 		sa.sa_handler = SIG_IGN;
1059 		sigaction(SIGTSTP, &sa, (struct sigaction *)0);
1060 		sigaction(SIGHUP, &sa, (struct sigaction *)0);
1061 
1062 		open_console();
1063 
1064 		char _sh[]		= "sh";
1065 		char _autoboot[]	= "autoboot";
1066 
1067 		argv[0] = _sh;
1068 		argv[1] = __DECONST(char *, script);
1069 		argv[2] = runcom_mode == AUTOBOOT ? _autoboot : 0;
1070 		argv[3] = 0;
1071 
1072 		sigprocmask(SIG_SETMASK, &sa.sa_mask, (sigset_t *) 0);
1073 
1074 #ifdef LOGIN_CAP
1075 		setprocresources(RESOURCE_RC);
1076 #endif
1077 		execv(shell, argv);
1078 		stall("can't exec %s for %s: %m", shell, script);
1079 		_exit(1);	/* force single user mode */
1080 	}
1081 
1082 	if (pid == -1) {
1083 		emergency("can't fork for %s on %s: %m", shell, script);
1084 		while (waitpid(-1, (int *) 0, WNOHANG) > 0)
1085 			continue;
1086 		sleep(STALL_TIMEOUT);
1087 		return (state_func_t) single_user;
1088 	}
1089 
1090 	/*
1091 	 * Copied from single_user().  This is a bit paranoid.
1092 	 */
1093 	requested_transition = 0;
1094 	do {
1095 		if ((wpid = waitpid(-1, &status, WUNTRACED)) != -1)
1096 			collect_child(wpid);
1097 		if (wpid == -1) {
1098 			if (requested_transition == death_single ||
1099 			    requested_transition == reroot)
1100 				return (state_func_t) requested_transition;
1101 			if (errno == EINTR)
1102 				continue;
1103 			warning("wait for %s on %s failed: %m; going to "
1104 			    "single user mode", shell, script);
1105 			return (state_func_t) single_user;
1106 		}
1107 		if (wpid == pid && WIFSTOPPED(status)) {
1108 			warning("init: %s on %s stopped, restarting\n",
1109 			    shell, script);
1110 			kill(pid, SIGCONT);
1111 			wpid = -1;
1112 		}
1113 	} while (wpid != pid);
1114 
1115 	if (WIFSIGNALED(status) && WTERMSIG(status) == SIGTERM &&
1116 	    requested_transition == catatonia) {
1117 		/* /etc/rc executed /sbin/reboot; wait for the end quietly */
1118 		sigset_t s;
1119 
1120 		sigfillset(&s);
1121 		for (;;)
1122 			sigsuspend(&s);
1123 	}
1124 
1125 	if (!WIFEXITED(status)) {
1126 		warning("%s on %s terminated abnormally, going to single "
1127 		    "user mode", shell, script);
1128 		return (state_func_t) single_user;
1129 	}
1130 
1131 	if (WEXITSTATUS(status))
1132 		return (state_func_t) single_user;
1133 
1134 	return (state_func_t) 0;
1135 }
1136 
1137 /*
1138  * Open the session database.
1139  *
1140  * NB: We could pass in the size here; is it necessary?
1141  */
1142 static int
1143 start_session_db(void)
1144 {
1145 	if (session_db && (*session_db->close)(session_db))
1146 		emergency("session database close: %s", strerror(errno));
1147 	if ((session_db = dbopen(NULL, O_RDWR, 0, DB_HASH, NULL)) == 0) {
1148 		emergency("session database open: %s", strerror(errno));
1149 		return (1);
1150 	}
1151 	return (0);
1152 
1153 }
1154 
1155 /*
1156  * Add a new login session.
1157  */
1158 static void
1159 add_session(session_t *sp)
1160 {
1161 	DBT key;
1162 	DBT data;
1163 
1164 	key.data = &sp->se_process;
1165 	key.size = sizeof sp->se_process;
1166 	data.data = &sp;
1167 	data.size = sizeof sp;
1168 
1169 	if ((*session_db->put)(session_db, &key, &data, 0))
1170 		emergency("insert %d: %s", sp->se_process, strerror(errno));
1171 }
1172 
1173 /*
1174  * Delete an old login session.
1175  */
1176 static void
1177 del_session(session_t *sp)
1178 {
1179 	DBT key;
1180 
1181 	key.data = &sp->se_process;
1182 	key.size = sizeof sp->se_process;
1183 
1184 	if ((*session_db->del)(session_db, &key, 0))
1185 		emergency("delete %d: %s", sp->se_process, strerror(errno));
1186 }
1187 
1188 /*
1189  * Look up a login session by pid.
1190  */
1191 static session_t *
1192 find_session(pid_t pid)
1193 {
1194 	DBT key;
1195 	DBT data;
1196 	session_t *ret;
1197 
1198 	key.data = &pid;
1199 	key.size = sizeof pid;
1200 	if ((*session_db->get)(session_db, &key, &data, 0) != 0)
1201 		return 0;
1202 	bcopy(data.data, (char *)&ret, sizeof(ret));
1203 	return ret;
1204 }
1205 
1206 /*
1207  * Construct an argument vector from a command line.
1208  */
1209 static char **
1210 construct_argv(char *command)
1211 {
1212 	int argc = 0;
1213 	char **argv = (char **) malloc(((strlen(command) + 1) / 2 + 1)
1214 						* sizeof (char *));
1215 
1216 	if ((argv[argc++] = strk(command)) == 0) {
1217 		free(argv);
1218 		return (NULL);
1219 	}
1220 	while ((argv[argc++] = strk((char *) 0)) != NULL)
1221 		continue;
1222 	return argv;
1223 }
1224 
1225 /*
1226  * Deallocate a session descriptor.
1227  */
1228 static void
1229 free_session(session_t *sp)
1230 {
1231 	free(sp->se_device);
1232 	if (sp->se_getty) {
1233 		free(sp->se_getty);
1234 		free(sp->se_getty_argv_space);
1235 		free(sp->se_getty_argv);
1236 	}
1237 	if (sp->se_window) {
1238 		free(sp->se_window);
1239 		free(sp->se_window_argv_space);
1240 		free(sp->se_window_argv);
1241 	}
1242 	if (sp->se_type)
1243 		free(sp->se_type);
1244 	free(sp);
1245 }
1246 
1247 /*
1248  * Allocate a new session descriptor.
1249  * Mark it SE_PRESENT.
1250  */
1251 static session_t *
1252 new_session(session_t *sprev, struct ttyent *typ)
1253 {
1254 	session_t *sp;
1255 	int fd;
1256 
1257 	if ((typ->ty_status & TTY_ON) == 0 ||
1258 	    typ->ty_name == 0 ||
1259 	    typ->ty_getty == 0)
1260 		return 0;
1261 
1262 	sp = (session_t *) calloc(1, sizeof (session_t));
1263 
1264 	sp->se_flags |= SE_PRESENT;
1265 
1266 	sp->se_device = malloc(sizeof(_PATH_DEV) + strlen(typ->ty_name));
1267 	sprintf(sp->se_device, "%s%s", _PATH_DEV, typ->ty_name);
1268 
1269 	/*
1270 	 * Attempt to open the device, if we get "device not configured"
1271 	 * then don't add the device to the session list.
1272 	 */
1273 	if ((fd = open(sp->se_device, O_RDONLY | O_NONBLOCK, 0)) < 0) {
1274 		if (errno == ENXIO) {
1275 			free_session(sp);
1276 			return (0);
1277 		}
1278 	} else
1279 		close(fd);
1280 
1281 	if (setupargv(sp, typ) == 0) {
1282 		free_session(sp);
1283 		return (0);
1284 	}
1285 
1286 	sp->se_next = 0;
1287 	if (sprev == 0) {
1288 		sessions = sp;
1289 		sp->se_prev = 0;
1290 	} else {
1291 		sprev->se_next = sp;
1292 		sp->se_prev = sprev;
1293 	}
1294 
1295 	return sp;
1296 }
1297 
1298 /*
1299  * Calculate getty and if useful window argv vectors.
1300  */
1301 static int
1302 setupargv(session_t *sp, struct ttyent *typ)
1303 {
1304 
1305 	if (sp->se_getty) {
1306 		free(sp->se_getty);
1307 		free(sp->se_getty_argv_space);
1308 		free(sp->se_getty_argv);
1309 	}
1310 	sp->se_getty = malloc(strlen(typ->ty_getty) + strlen(typ->ty_name) + 2);
1311 	sprintf(sp->se_getty, "%s %s", typ->ty_getty, typ->ty_name);
1312 	sp->se_getty_argv_space = strdup(sp->se_getty);
1313 	sp->se_getty_argv = construct_argv(sp->se_getty_argv_space);
1314 	if (sp->se_getty_argv == 0) {
1315 		warning("can't parse getty for port %s", sp->se_device);
1316 		free(sp->se_getty);
1317 		free(sp->se_getty_argv_space);
1318 		sp->se_getty = sp->se_getty_argv_space = 0;
1319 		return (0);
1320 	}
1321 	if (sp->se_window) {
1322 		free(sp->se_window);
1323 		free(sp->se_window_argv_space);
1324 		free(sp->se_window_argv);
1325 	}
1326 	sp->se_window = sp->se_window_argv_space = 0;
1327 	sp->se_window_argv = 0;
1328 	if (typ->ty_window) {
1329 		sp->se_window = strdup(typ->ty_window);
1330 		sp->se_window_argv_space = strdup(sp->se_window);
1331 		sp->se_window_argv = construct_argv(sp->se_window_argv_space);
1332 		if (sp->se_window_argv == 0) {
1333 			warning("can't parse window for port %s",
1334 			    sp->se_device);
1335 			free(sp->se_window_argv_space);
1336 			free(sp->se_window);
1337 			sp->se_window = sp->se_window_argv_space = 0;
1338 			return (0);
1339 		}
1340 	}
1341 	if (sp->se_type)
1342 		free(sp->se_type);
1343 	sp->se_type = typ->ty_type ? strdup(typ->ty_type) : 0;
1344 	return (1);
1345 }
1346 
1347 /*
1348  * Walk the list of ttys and create sessions for each active line.
1349  */
1350 static state_func_t
1351 read_ttys(void)
1352 {
1353 	session_t *sp, *snext;
1354 	struct ttyent *typ;
1355 
1356 	/*
1357 	 * Destroy any previous session state.
1358 	 * There shouldn't be any, but just in case...
1359 	 */
1360 	for (sp = sessions; sp; sp = snext) {
1361 		snext = sp->se_next;
1362 		free_session(sp);
1363 	}
1364 	sessions = 0;
1365 	if (start_session_db())
1366 		return (state_func_t) single_user;
1367 
1368 	/*
1369 	 * Allocate a session entry for each active port.
1370 	 * Note that sp starts at 0.
1371 	 */
1372 	while ((typ = getttyent()) != NULL)
1373 		if ((snext = new_session(sp, typ)) != NULL)
1374 			sp = snext;
1375 
1376 	endttyent();
1377 
1378 	return (state_func_t) multi_user;
1379 }
1380 
1381 /*
1382  * Start a window system running.
1383  */
1384 static void
1385 start_window_system(session_t *sp)
1386 {
1387 	pid_t pid;
1388 	sigset_t mask;
1389 	char term[64], *env[2];
1390 	int status;
1391 
1392 	if ((pid = fork()) == -1) {
1393 		emergency("can't fork for window system on port %s: %m",
1394 		    sp->se_device);
1395 		/* hope that getty fails and we can try again */
1396 		return;
1397 	}
1398 	if (pid) {
1399 		waitpid(-1, &status, 0);
1400 		return;
1401 	}
1402 
1403 	/* reparent window process to the init to not make a zombie on exit */
1404 	if ((pid = fork()) == -1) {
1405 		emergency("can't fork for window system on port %s: %m",
1406 		    sp->se_device);
1407 		_exit(1);
1408 	}
1409 	if (pid)
1410 		_exit(0);
1411 
1412 	sigemptyset(&mask);
1413 	sigprocmask(SIG_SETMASK, &mask, (sigset_t *) 0);
1414 
1415 	if (setsid() < 0)
1416 		emergency("setsid failed (window) %m");
1417 
1418 #ifdef LOGIN_CAP
1419 	setprocresources(RESOURCE_WINDOW);
1420 #endif
1421 	if (sp->se_type) {
1422 		/* Don't use malloc after fork */
1423 		strcpy(term, "TERM=");
1424 		strncat(term, sp->se_type, sizeof(term) - 6);
1425 		env[0] = term;
1426 		env[1] = 0;
1427 	}
1428 	else
1429 		env[0] = 0;
1430 	execve(sp->se_window_argv[0], sp->se_window_argv, env);
1431 	stall("can't exec window system '%s' for port %s: %m",
1432 		sp->se_window_argv[0], sp->se_device);
1433 	_exit(1);
1434 }
1435 
1436 /*
1437  * Start a login session running.
1438  */
1439 static pid_t
1440 start_getty(session_t *sp)
1441 {
1442 	pid_t pid;
1443 	sigset_t mask;
1444 	time_t current_time = time((time_t *) 0);
1445 	int too_quick = 0;
1446 	char term[64], *env[2];
1447 
1448 	if (current_time >= sp->se_started &&
1449 	    current_time - sp->se_started < GETTY_SPACING) {
1450 		if (++sp->se_nspace > GETTY_NSPACE) {
1451 			sp->se_nspace = 0;
1452 			too_quick = 1;
1453 		}
1454 	} else
1455 		sp->se_nspace = 0;
1456 
1457 	/*
1458 	 * fork(), not vfork() -- we can't afford to block.
1459 	 */
1460 	if ((pid = fork()) == -1) {
1461 		emergency("can't fork for getty on port %s: %m", sp->se_device);
1462 		return -1;
1463 	}
1464 
1465 	if (pid)
1466 		return pid;
1467 
1468 	if (too_quick) {
1469 		warning("getty repeating too quickly on port %s, sleeping %d secs",
1470 		    sp->se_device, GETTY_SLEEP);
1471 		sleep((unsigned) GETTY_SLEEP);
1472 	}
1473 
1474 	if (sp->se_window) {
1475 		start_window_system(sp);
1476 		sleep(WINDOW_WAIT);
1477 	}
1478 
1479 	sigemptyset(&mask);
1480 	sigprocmask(SIG_SETMASK, &mask, (sigset_t *) 0);
1481 
1482 #ifdef LOGIN_CAP
1483 	setprocresources(RESOURCE_GETTY);
1484 #endif
1485 	if (sp->se_type) {
1486 		/* Don't use malloc after fork */
1487 		strcpy(term, "TERM=");
1488 		strncat(term, sp->se_type, sizeof(term) - 6);
1489 		env[0] = term;
1490 		env[1] = 0;
1491 	} else
1492 		env[0] = 0;
1493 	execve(sp->se_getty_argv[0], sp->se_getty_argv, env);
1494 	stall("can't exec getty '%s' for port %s: %m",
1495 		sp->se_getty_argv[0], sp->se_device);
1496 	_exit(1);
1497 }
1498 
1499 /*
1500  * Collect exit status for a child.
1501  * If an exiting login, start a new login running.
1502  */
1503 static void
1504 collect_child(pid_t pid)
1505 {
1506 	session_t *sp, *sprev, *snext;
1507 
1508 	if (! sessions)
1509 		return;
1510 
1511 	if (! (sp = find_session(pid)))
1512 		return;
1513 
1514 	del_session(sp);
1515 	sp->se_process = 0;
1516 
1517 	if (sp->se_flags & SE_SHUTDOWN) {
1518 		if ((sprev = sp->se_prev) != NULL)
1519 			sprev->se_next = sp->se_next;
1520 		else
1521 			sessions = sp->se_next;
1522 		if ((snext = sp->se_next) != NULL)
1523 			snext->se_prev = sp->se_prev;
1524 		free_session(sp);
1525 		return;
1526 	}
1527 
1528 	if ((pid = start_getty(sp)) == -1) {
1529 		/* serious trouble */
1530 		requested_transition = clean_ttys;
1531 		return;
1532 	}
1533 
1534 	sp->se_process = pid;
1535 	sp->se_started = time((time_t *) 0);
1536 	add_session(sp);
1537 }
1538 
1539 /*
1540  * Catch a signal and request a state transition.
1541  */
1542 static void
1543 transition_handler(int sig)
1544 {
1545 
1546 	switch (sig) {
1547 	case SIGHUP:
1548 		if (current_state == read_ttys || current_state == multi_user ||
1549 		    current_state == clean_ttys || current_state == catatonia)
1550 			requested_transition = clean_ttys;
1551 		break;
1552 	case SIGUSR2:
1553 		howto = RB_POWEROFF;
1554 	case SIGUSR1:
1555 		howto |= RB_HALT;
1556 	case SIGINT:
1557 		Reboot = TRUE;
1558 	case SIGTERM:
1559 		if (current_state == read_ttys || current_state == multi_user ||
1560 		    current_state == clean_ttys || current_state == catatonia)
1561 			requested_transition = death;
1562 		else
1563 			requested_transition = death_single;
1564 		break;
1565 	case SIGTSTP:
1566 		if (current_state == runcom || current_state == read_ttys ||
1567 		    current_state == clean_ttys ||
1568 		    current_state == multi_user || current_state == catatonia)
1569 			requested_transition = catatonia;
1570 		break;
1571 	case SIGEMT:
1572 		requested_transition = reroot;
1573 		break;
1574 	default:
1575 		requested_transition = 0;
1576 		break;
1577 	}
1578 }
1579 
1580 /*
1581  * Take the system multiuser.
1582  */
1583 static state_func_t
1584 multi_user(void)
1585 {
1586 	pid_t pid;
1587 	session_t *sp;
1588 
1589 	requested_transition = 0;
1590 
1591 	/*
1592 	 * If the administrator has not set the security level to -1
1593 	 * to indicate that the kernel should not run multiuser in secure
1594 	 * mode, and the run script has not set a higher level of security
1595 	 * than level 1, then put the kernel into secure mode.
1596 	 */
1597 	if (getsecuritylevel() == 0)
1598 		setsecuritylevel(1);
1599 
1600 	for (sp = sessions; sp; sp = sp->se_next) {
1601 		if (sp->se_process)
1602 			continue;
1603 		if ((pid = start_getty(sp)) == -1) {
1604 			/* serious trouble */
1605 			requested_transition = clean_ttys;
1606 			break;
1607 		}
1608 		sp->se_process = pid;
1609 		sp->se_started = time((time_t *) 0);
1610 		add_session(sp);
1611 	}
1612 
1613 	while (!requested_transition)
1614 		if ((pid = waitpid(-1, (int *) 0, 0)) != -1)
1615 			collect_child(pid);
1616 
1617 	return (state_func_t) requested_transition;
1618 }
1619 
1620 /*
1621  * This is an (n*2)+(n^2) algorithm.  We hope it isn't run often...
1622  */
1623 static state_func_t
1624 clean_ttys(void)
1625 {
1626 	session_t *sp, *sprev;
1627 	struct ttyent *typ;
1628 	int devlen;
1629 	char *old_getty, *old_window, *old_type;
1630 
1631 	/*
1632 	 * mark all sessions for death, (!SE_PRESENT)
1633 	 * as we find or create new ones they'll be marked as keepers,
1634 	 * we'll later nuke all the ones not found in /etc/ttys
1635 	 */
1636 	for (sp = sessions; sp != NULL; sp = sp->se_next)
1637 		sp->se_flags &= ~SE_PRESENT;
1638 
1639 	devlen = sizeof(_PATH_DEV) - 1;
1640 	while ((typ = getttyent()) != NULL) {
1641 		for (sprev = 0, sp = sessions; sp; sprev = sp, sp = sp->se_next)
1642 			if (strcmp(typ->ty_name, sp->se_device + devlen) == 0)
1643 				break;
1644 
1645 		if (sp) {
1646 			/* we want this one to live */
1647 			sp->se_flags |= SE_PRESENT;
1648 			if ((typ->ty_status & TTY_ON) == 0 ||
1649 			    typ->ty_getty == 0) {
1650 				sp->se_flags |= SE_SHUTDOWN;
1651 				kill(sp->se_process, SIGHUP);
1652 				continue;
1653 			}
1654 			sp->se_flags &= ~SE_SHUTDOWN;
1655 			old_getty = sp->se_getty ? strdup(sp->se_getty) : 0;
1656 			old_window = sp->se_window ? strdup(sp->se_window) : 0;
1657 			old_type = sp->se_type ? strdup(sp->se_type) : 0;
1658 			if (setupargv(sp, typ) == 0) {
1659 				warning("can't parse getty for port %s",
1660 					sp->se_device);
1661 				sp->se_flags |= SE_SHUTDOWN;
1662 				kill(sp->se_process, SIGHUP);
1663 			}
1664 			else if (   !old_getty
1665 				 || (!old_type && sp->se_type)
1666 				 || (old_type && !sp->se_type)
1667 				 || (!old_window && sp->se_window)
1668 				 || (old_window && !sp->se_window)
1669 				 || (strcmp(old_getty, sp->se_getty) != 0)
1670 				 || (old_window && strcmp(old_window, sp->se_window) != 0)
1671 				 || (old_type && strcmp(old_type, sp->se_type) != 0)
1672 				) {
1673 				/* Don't set SE_SHUTDOWN here */
1674 				sp->se_nspace = 0;
1675 				sp->se_started = 0;
1676 				kill(sp->se_process, SIGHUP);
1677 			}
1678 			if (old_getty)
1679 				free(old_getty);
1680 			if (old_window)
1681 				free(old_window);
1682 			if (old_type)
1683 				free(old_type);
1684 			continue;
1685 		}
1686 
1687 		new_session(sprev, typ);
1688 	}
1689 
1690 	endttyent();
1691 
1692 	/*
1693 	 * sweep through and kill all deleted sessions
1694 	 * ones who's /etc/ttys line was deleted (SE_PRESENT unset)
1695 	 */
1696 	for (sp = sessions; sp != NULL; sp = sp->se_next) {
1697 		if ((sp->se_flags & SE_PRESENT) == 0) {
1698 			sp->se_flags |= SE_SHUTDOWN;
1699 			kill(sp->se_process, SIGHUP);
1700 		}
1701 	}
1702 
1703 	return (state_func_t) multi_user;
1704 }
1705 
1706 /*
1707  * Block further logins.
1708  */
1709 static state_func_t
1710 catatonia(void)
1711 {
1712 	session_t *sp;
1713 
1714 	for (sp = sessions; sp; sp = sp->se_next)
1715 		sp->se_flags |= SE_SHUTDOWN;
1716 
1717 	return (state_func_t) multi_user;
1718 }
1719 
1720 /*
1721  * Note SIGALRM.
1722  */
1723 static void
1724 alrm_handler(int sig)
1725 {
1726 
1727 	(void)sig;
1728 	clang = 1;
1729 }
1730 
1731 /*
1732  * Bring the system down to single user.
1733  */
1734 static state_func_t
1735 death(void)
1736 {
1737 	int block, blocked;
1738 	size_t len;
1739 
1740 	/* Temporarily block suspend. */
1741 	len = sizeof(blocked);
1742 	block = 1;
1743 	if (sysctlbyname("kern.suspend_blocked", &blocked, &len,
1744 	    &block, sizeof(block)) == -1)
1745 		blocked = 0;
1746 
1747 	/*
1748 	 * Also revoke the TTY here.  Because runshutdown() may reopen
1749 	 * the TTY whose getty we're killing here, there is no guarantee
1750 	 * runshutdown() will perform the initial open() call, causing
1751 	 * the terminal attributes to be misconfigured.
1752 	 */
1753 	revoke_ttys();
1754 
1755 	/* Try to run the rc.shutdown script within a period of time */
1756 	runshutdown();
1757 
1758 	/* Unblock suspend if we blocked it. */
1759 	if (!blocked)
1760 		sysctlbyname("kern.suspend_blocked", NULL, NULL,
1761 		    &blocked, sizeof(blocked));
1762 
1763 	return (state_func_t) death_single;
1764 }
1765 
1766 /*
1767  * Do what is necessary to reinitialize single user mode or reboot
1768  * from an incomplete state.
1769  */
1770 static state_func_t
1771 death_single(void)
1772 {
1773 	int i;
1774 	pid_t pid;
1775 	static const int death_sigs[2] = { SIGTERM, SIGKILL };
1776 
1777 	revoke(_PATH_CONSOLE);
1778 
1779 	for (i = 0; i < 2; ++i) {
1780 		if (kill(-1, death_sigs[i]) == -1 && errno == ESRCH)
1781 			return (state_func_t) single_user;
1782 
1783 		clang = 0;
1784 		alarm(DEATH_WATCH);
1785 		do
1786 			if ((pid = waitpid(-1, (int *)0, 0)) != -1)
1787 				collect_child(pid);
1788 		while (clang == 0 && errno != ECHILD);
1789 
1790 		if (errno == ECHILD)
1791 			return (state_func_t) single_user;
1792 	}
1793 
1794 	warning("some processes would not die; ps axl advised");
1795 
1796 	return (state_func_t) single_user;
1797 }
1798 
1799 static void
1800 revoke_ttys(void)
1801 {
1802 	session_t *sp;
1803 
1804 	for (sp = sessions; sp; sp = sp->se_next) {
1805 		sp->se_flags |= SE_SHUTDOWN;
1806 		kill(sp->se_process, SIGHUP);
1807 		revoke(sp->se_device);
1808 	}
1809 }
1810 
1811 /*
1812  * Run the system shutdown script.
1813  *
1814  * Exit codes:      XXX I should document more
1815  * -2       shutdown script terminated abnormally
1816  * -1       fatal error - can't run script
1817  * 0        good.
1818  * >0       some error (exit code)
1819  */
1820 static int
1821 runshutdown(void)
1822 {
1823 	pid_t pid, wpid;
1824 	int status;
1825 	int shutdowntimeout;
1826 	size_t len;
1827 	char *argv[4];
1828 	const char *shell;
1829 	struct sigaction sa;
1830 	struct stat sb;
1831 
1832 	/*
1833 	 * rc.shutdown is optional, so to prevent any unnecessary
1834 	 * complaints from the shell we simply don't run it if the
1835 	 * file does not exist. If the stat() here fails for other
1836 	 * reasons, we'll let the shell complain.
1837 	 */
1838 	if (stat(_PATH_RUNDOWN, &sb) == -1 && errno == ENOENT)
1839 		return 0;
1840 
1841 	shell = get_shell();
1842 
1843 	if ((pid = fork()) == 0) {
1844 		sigemptyset(&sa.sa_mask);
1845 		sa.sa_flags = 0;
1846 		sa.sa_handler = SIG_IGN;
1847 		sigaction(SIGTSTP, &sa, (struct sigaction *)0);
1848 		sigaction(SIGHUP, &sa, (struct sigaction *)0);
1849 
1850 		open_console();
1851 
1852 		char _sh[]	= "sh";
1853 		char _reboot[]	= "reboot";
1854 		char _single[]	= "single";
1855 		char _path_rundown[] = _PATH_RUNDOWN;
1856 
1857 		argv[0] = _sh;
1858 		argv[1] = _path_rundown;
1859 		argv[2] = Reboot ? _reboot : _single;
1860 		argv[3] = 0;
1861 
1862 		sigprocmask(SIG_SETMASK, &sa.sa_mask, (sigset_t *) 0);
1863 
1864 #ifdef LOGIN_CAP
1865 		setprocresources(RESOURCE_RC);
1866 #endif
1867 		execv(shell, argv);
1868 		warning("can't exec %s for %s: %m", shell, _PATH_RUNDOWN);
1869 		_exit(1);	/* force single user mode */
1870 	}
1871 
1872 	if (pid == -1) {
1873 		emergency("can't fork for %s on %s: %m", shell, _PATH_RUNDOWN);
1874 		while (waitpid(-1, (int *) 0, WNOHANG) > 0)
1875 			continue;
1876 		sleep(STALL_TIMEOUT);
1877 		return -1;
1878 	}
1879 
1880 	len = sizeof(shutdowntimeout);
1881 	if (sysctlbyname("kern.init_shutdown_timeout", &shutdowntimeout, &len,
1882 	    NULL, 0) == -1 || shutdowntimeout < 2)
1883 		shutdowntimeout = DEATH_SCRIPT;
1884 	alarm(shutdowntimeout);
1885 	clang = 0;
1886 	/*
1887 	 * Copied from single_user().  This is a bit paranoid.
1888 	 * Use the same ALRM handler.
1889 	 */
1890 	do {
1891 		if ((wpid = waitpid(-1, &status, WUNTRACED)) != -1)
1892 			collect_child(wpid);
1893 		if (clang == 1) {
1894 			/* we were waiting for the sub-shell */
1895 			kill(wpid, SIGTERM);
1896 			warning("timeout expired for %s on %s: %m; going to "
1897 			    "single user mode", shell, _PATH_RUNDOWN);
1898 			return -1;
1899 		}
1900 		if (wpid == -1) {
1901 			if (errno == EINTR)
1902 				continue;
1903 			warning("wait for %s on %s failed: %m; going to "
1904 			    "single user mode", shell, _PATH_RUNDOWN);
1905 			return -1;
1906 		}
1907 		if (wpid == pid && WIFSTOPPED(status)) {
1908 			warning("init: %s on %s stopped, restarting\n",
1909 				shell, _PATH_RUNDOWN);
1910 			kill(pid, SIGCONT);
1911 			wpid = -1;
1912 		}
1913 	} while (wpid != pid && !clang);
1914 
1915 	/* Turn off the alarm */
1916 	alarm(0);
1917 
1918 	if (WIFSIGNALED(status) && WTERMSIG(status) == SIGTERM &&
1919 	    requested_transition == catatonia) {
1920 		/*
1921 		 * /etc/rc.shutdown executed /sbin/reboot;
1922 		 * wait for the end quietly
1923 		 */
1924 		sigset_t s;
1925 
1926 		sigfillset(&s);
1927 		for (;;)
1928 			sigsuspend(&s);
1929 	}
1930 
1931 	if (!WIFEXITED(status)) {
1932 		warning("%s on %s terminated abnormally, going to "
1933 		    "single user mode", shell, _PATH_RUNDOWN);
1934 		return -2;
1935 	}
1936 
1937 	if ((status = WEXITSTATUS(status)) != 0)
1938 		warning("%s returned status %d", _PATH_RUNDOWN, status);
1939 
1940 	return status;
1941 }
1942 
1943 static char *
1944 strk(char *p)
1945 {
1946 	static char *t;
1947 	char *q;
1948 	int c;
1949 
1950 	if (p)
1951 		t = p;
1952 	if (!t)
1953 		return 0;
1954 
1955 	c = *t;
1956 	while (c == ' ' || c == '\t' )
1957 		c = *++t;
1958 	if (!c) {
1959 		t = 0;
1960 		return 0;
1961 	}
1962 	q = t;
1963 	if (c == '\'') {
1964 		c = *++t;
1965 		q = t;
1966 		while (c && c != '\'')
1967 			c = *++t;
1968 		if (!c)  /* unterminated string */
1969 			q = t = 0;
1970 		else
1971 			*t++ = 0;
1972 	} else {
1973 		while (c && c != ' ' && c != '\t' )
1974 			c = *++t;
1975 		*t++ = 0;
1976 		if (!c)
1977 			t = 0;
1978 	}
1979 	return q;
1980 }
1981 
1982 #ifdef LOGIN_CAP
1983 static void
1984 setprocresources(const char *cname)
1985 {
1986 	login_cap_t *lc;
1987 	if ((lc = login_getclassbyname(cname, NULL)) != NULL) {
1988 		setusercontext(lc, (struct passwd*)NULL, 0,
1989 		    LOGIN_SETPRIORITY | LOGIN_SETRESOURCES |
1990 		    LOGIN_SETLOGINCLASS | LOGIN_SETCPUMASK);
1991 		login_close(lc);
1992 	}
1993 }
1994 #endif
1995