xref: /freebsd/sbin/init/init.c (revision eac7052fdebb90caf2f653e06187bdbca837b9c7)
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 revoke_ttys(void);
110 static int  runshutdown(void);
111 static char *strk(char *);
112 
113 /*
114  * We really need a recursive typedef...
115  * The following at least guarantees that the return type of (*state_t)()
116  * is sufficiently wide to hold a function pointer.
117  */
118 typedef long (*state_func_t)(void);
119 typedef state_func_t (*state_t)(void);
120 
121 static state_func_t single_user(void);
122 static state_func_t runcom(void);
123 static state_func_t read_ttys(void);
124 static state_func_t multi_user(void);
125 static state_func_t clean_ttys(void);
126 static state_func_t catatonia(void);
127 static state_func_t death(void);
128 static state_func_t death_single(void);
129 static state_func_t reroot(void);
130 static state_func_t reroot_phase_two(void);
131 
132 static state_func_t run_script(const char *);
133 
134 static enum { AUTOBOOT, FASTBOOT } runcom_mode = AUTOBOOT;
135 #define FALSE	0
136 #define TRUE	1
137 
138 static int Reboot = FALSE;
139 static int howto = RB_AUTOBOOT;
140 
141 static int devfs;
142 static char *init_path_argv0;
143 
144 static void transition(state_t);
145 static state_t requested_transition;
146 static state_t current_state = death_single;
147 
148 static void execute_script(char *argv[]);
149 static void open_console(void);
150 static const char *get_shell(void);
151 static void replace_init(char *path);
152 static void write_stderr(const char *message);
153 
154 typedef struct init_session {
155 	pid_t	se_process;		/* controlling process */
156 	time_t	se_started;		/* used to avoid thrashing */
157 	int	se_flags;		/* status of session */
158 #define	SE_SHUTDOWN	0x1		/* session won't be restarted */
159 #define	SE_PRESENT	0x2		/* session is in /etc/ttys */
160 #define	SE_IFEXISTS	0x4		/* session defined as "onifexists" */
161 #define	SE_IFCONSOLE	0x8		/* session defined as "onifconsole" */
162 	int	se_nspace;		/* spacing count */
163 	char	*se_device;		/* filename of port */
164 	char	*se_getty;		/* what to run on that port */
165 	char	*se_getty_argv_space;   /* pre-parsed argument array space */
166 	char	**se_getty_argv;	/* pre-parsed argument array */
167 	char	*se_window;		/* window system (started only once) */
168 	char	*se_window_argv_space;  /* pre-parsed argument array space */
169 	char	**se_window_argv;	/* pre-parsed argument array */
170 	char	*se_type;		/* default terminal type */
171 	struct	init_session *se_prev;
172 	struct	init_session *se_next;
173 } session_t;
174 
175 static void free_session(session_t *);
176 static session_t *new_session(session_t *, struct ttyent *);
177 static session_t *sessions;
178 
179 static char **construct_argv(char *);
180 static void start_window_system(session_t *);
181 static void collect_child(pid_t);
182 static pid_t start_getty(session_t *);
183 static void transition_handler(int);
184 static void alrm_handler(int);
185 static void setsecuritylevel(int);
186 static int getsecuritylevel(void);
187 static int setupargv(session_t *, struct ttyent *);
188 #ifdef LOGIN_CAP
189 static void setprocresources(const char *);
190 #endif
191 static int clang;
192 
193 static int start_session_db(void);
194 static void add_session(session_t *);
195 static void del_session(session_t *);
196 static session_t *find_session(pid_t);
197 static DB *session_db;
198 
199 /*
200  * The mother of all processes.
201  */
202 int
203 main(int argc, char *argv[])
204 {
205 	state_t initial_transition = runcom;
206 	char kenv_value[PATH_MAX];
207 	int c, error;
208 	struct sigaction sa;
209 	sigset_t mask;
210 
211 	/* Dispose of random users. */
212 	if (getuid() != 0)
213 		errx(1, "%s", strerror(EPERM));
214 
215 	/* System V users like to reexec init. */
216 	if (getpid() != 1) {
217 #ifdef COMPAT_SYSV_INIT
218 		/* So give them what they want */
219 		if (argc > 1) {
220 			if (strlen(argv[1]) == 1) {
221 				char runlevel = *argv[1];
222 				int sig;
223 
224 				switch (runlevel) {
225 				case '0': /* halt + poweroff */
226 					sig = SIGUSR2;
227 					break;
228 				case '1': /* single-user */
229 					sig = SIGTERM;
230 					break;
231 				case '6': /* reboot */
232 					sig = SIGINT;
233 					break;
234 				case 'c': /* block further logins */
235 					sig = SIGTSTP;
236 					break;
237 				case 'q': /* rescan /etc/ttys */
238 					sig = SIGHUP;
239 					break;
240 				case 'r': /* remount root */
241 					sig = SIGEMT;
242 					break;
243 				default:
244 					goto invalid;
245 				}
246 				kill(1, sig);
247 				_exit(0);
248 			} else
249 invalid:
250 				errx(1, "invalid run-level ``%s''", argv[1]);
251 		} else
252 #endif
253 			errx(1, "already running");
254 	}
255 
256 	init_path_argv0 = strdup(argv[0]);
257 	if (init_path_argv0 == NULL)
258 		err(1, "strdup");
259 
260 	/*
261 	 * Note that this does NOT open a file...
262 	 * Does 'init' deserve its own facility number?
263 	 */
264 	openlog("init", LOG_CONS, LOG_AUTH);
265 
266 	/*
267 	 * Create an initial session.
268 	 */
269 	if (setsid() < 0 && (errno != EPERM || getsid(0) != 1))
270 		warning("initial setsid() failed: %m");
271 
272 	/*
273 	 * Establish an initial user so that programs running
274 	 * single user do not freak out and die (like passwd).
275 	 */
276 	if (setlogin("root") < 0)
277 		warning("setlogin() failed: %m");
278 
279 	/*
280 	 * This code assumes that we always get arguments through flags,
281 	 * never through bits set in some random machine register.
282 	 */
283 	while ((c = getopt(argc, argv, "dsfr")) != -1)
284 		switch (c) {
285 		case 'd':
286 			devfs = 1;
287 			break;
288 		case 's':
289 			initial_transition = single_user;
290 			break;
291 		case 'f':
292 			runcom_mode = FASTBOOT;
293 			break;
294 		case 'r':
295 			initial_transition = reroot_phase_two;
296 			break;
297 		default:
298 			warning("unrecognized flag '-%c'", c);
299 			break;
300 		}
301 
302 	if (optind != argc)
303 		warning("ignoring excess arguments");
304 
305 	/*
306 	 * We catch or block signals rather than ignore them,
307 	 * so that they get reset on exec.
308 	 */
309 	handle(disaster, SIGABRT, SIGFPE, SIGILL, SIGSEGV, SIGBUS, SIGSYS,
310 	    SIGXCPU, 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, NULL);
319 	sigemptyset(&sa.sa_mask);
320 	sa.sa_flags = 0;
321 	sa.sa_handler = SIG_IGN;
322 	sigaction(SIGTTIN, &sa, NULL);
323 	sigaction(SIGTTOU, &sa, NULL);
324 
325 	/*
326 	 * Paranoia.
327 	 */
328 	close(0);
329 	close(1);
330 	close(2);
331 
332 	if (kenv(KENV_GET, "init_exec", kenv_value, sizeof(kenv_value)) > 0) {
333 		replace_init(kenv_value);
334 		_exit(0); /* reboot */
335 	}
336 
337 	if (kenv(KENV_GET, "init_script", kenv_value, sizeof(kenv_value)) > 0) {
338 		state_func_t next_transition;
339 
340 		if ((next_transition = run_script(kenv_value)) != NULL)
341 			initial_transition = (state_t) next_transition;
342 	}
343 
344 	if (kenv(KENV_GET, "init_chroot", kenv_value, sizeof(kenv_value)) > 0) {
345 		if (chdir(kenv_value) != 0 || chroot(".") != 0)
346 			warning("Can't chroot to %s: %m", kenv_value);
347 	}
348 
349 	/*
350 	 * Additional check if devfs needs to be mounted:
351 	 * If "/" and "/dev" have the same device number,
352 	 * then it hasn't been mounted yet.
353 	 */
354 	if (!devfs) {
355 		struct stat stst;
356 		dev_t root_devno;
357 
358 		stat("/", &stst);
359 		root_devno = stst.st_dev;
360 		if (stat("/dev", &stst) != 0)
361 			warning("Can't stat /dev: %m");
362 		else if (stst.st_dev == root_devno)
363 			devfs++;
364 	}
365 
366 	if (devfs) {
367 		struct iovec iov[4];
368 		char *s;
369 		int i;
370 
371 		char _fstype[]	= "fstype";
372 		char _devfs[]	= "devfs";
373 		char _fspath[]	= "fspath";
374 		char _path_dev[]= _PATH_DEV;
375 
376 		iov[0].iov_base = _fstype;
377 		iov[0].iov_len = sizeof(_fstype);
378 		iov[1].iov_base = _devfs;
379 		iov[1].iov_len = sizeof(_devfs);
380 		iov[2].iov_base = _fspath;
381 		iov[2].iov_len = sizeof(_fspath);
382 		/*
383 		 * Try to avoid the trailing slash in _PATH_DEV.
384 		 * Be *very* defensive.
385 		 */
386 		s = strdup(_PATH_DEV);
387 		if (s != NULL) {
388 			i = strlen(s);
389 			if (i > 0 && s[i - 1] == '/')
390 				s[i - 1] = '\0';
391 			iov[3].iov_base = s;
392 			iov[3].iov_len = strlen(s) + 1;
393 		} else {
394 			iov[3].iov_base = _path_dev;
395 			iov[3].iov_len = sizeof(_path_dev);
396 		}
397 		nmount(iov, 4, 0);
398 		if (s != NULL)
399 			free(s);
400 	}
401 
402 	if (initial_transition != reroot_phase_two) {
403 		/*
404 		 * Unmount reroot leftovers.  This runs after init(8)
405 		 * gets reexecuted after reroot_phase_two() is done.
406 		 */
407 		error = unmount(_PATH_REROOT, MNT_FORCE);
408 		if (error != 0 && errno != EINVAL)
409 			warning("Cannot unmount %s: %m", _PATH_REROOT);
410 	}
411 
412 	/*
413 	 * Start the state machine.
414 	 */
415 	transition(initial_transition);
416 
417 	/*
418 	 * Should never reach here.
419 	 */
420 	return 1;
421 }
422 
423 /*
424  * Associate a function with a signal handler.
425  */
426 static void
427 handle(sig_t handler, ...)
428 {
429 	int sig;
430 	struct sigaction sa;
431 	sigset_t mask_everything;
432 	va_list ap;
433 	va_start(ap, handler);
434 
435 	sa.sa_handler = handler;
436 	sigfillset(&mask_everything);
437 
438 	while ((sig = va_arg(ap, int)) != 0) {
439 		sa.sa_mask = mask_everything;
440 		/* XXX SA_RESTART? */
441 		sa.sa_flags = sig == SIGCHLD ? SA_NOCLDSTOP : 0;
442 		sigaction(sig, &sa, NULL);
443 	}
444 	va_end(ap);
445 }
446 
447 /*
448  * Delete a set of signals from a mask.
449  */
450 static void
451 delset(sigset_t *maskp, ...)
452 {
453 	int sig;
454 	va_list ap;
455 	va_start(ap, maskp);
456 
457 	while ((sig = va_arg(ap, int)) != 0)
458 		sigdelset(maskp, sig);
459 	va_end(ap);
460 }
461 
462 /*
463  * Log a message and sleep for a while (to give someone an opportunity
464  * to read it and to save log or hardcopy output if the problem is chronic).
465  * NB: should send a message to the session logger to avoid blocking.
466  */
467 static void
468 stall(const char *message, ...)
469 {
470 	va_list ap;
471 	va_start(ap, message);
472 
473 	vsyslog(LOG_ALERT, message, ap);
474 	va_end(ap);
475 	sleep(STALL_TIMEOUT);
476 }
477 
478 /*
479  * Like stall(), but doesn't sleep.
480  * If cpp had variadic macros, the two functions could be #defines for another.
481  * NB: should send a message to the session logger to avoid blocking.
482  */
483 static void
484 warning(const char *message, ...)
485 {
486 	va_list ap;
487 	va_start(ap, message);
488 
489 	vsyslog(LOG_ALERT, message, ap);
490 	va_end(ap);
491 }
492 
493 /*
494  * Log an emergency message.
495  * NB: should send a message to the session logger to avoid blocking.
496  */
497 static void
498 emergency(const char *message, ...)
499 {
500 	va_list ap;
501 	va_start(ap, message);
502 
503 	vsyslog(LOG_EMERG, message, ap);
504 	va_end(ap);
505 }
506 
507 /*
508  * Catch an unexpected signal.
509  */
510 static void
511 disaster(int sig)
512 {
513 
514 	emergency("fatal signal: %s",
515 	    (unsigned)sig < NSIG ? sys_siglist[sig] : "unknown signal");
516 
517 	sleep(STALL_TIMEOUT);
518 	_exit(sig);		/* reboot */
519 }
520 
521 /*
522  * Get the security level of the kernel.
523  */
524 static int
525 getsecuritylevel(void)
526 {
527 #ifdef KERN_SECURELVL
528 	int name[2], curlevel;
529 	size_t len;
530 
531 	name[0] = CTL_KERN;
532 	name[1] = KERN_SECURELVL;
533 	len = sizeof curlevel;
534 	if (sysctl(name, 2, &curlevel, &len, NULL, 0) == -1) {
535 		emergency("cannot get kernel security level: %s",
536 		    strerror(errno));
537 		return (-1);
538 	}
539 	return (curlevel);
540 #else
541 	return (-1);
542 #endif
543 }
544 
545 /*
546  * Set the security level of the kernel.
547  */
548 static void
549 setsecuritylevel(int newlevel)
550 {
551 #ifdef KERN_SECURELVL
552 	int name[2], curlevel;
553 
554 	curlevel = getsecuritylevel();
555 	if (newlevel == curlevel)
556 		return;
557 	name[0] = CTL_KERN;
558 	name[1] = KERN_SECURELVL;
559 	if (sysctl(name, 2, NULL, NULL, &newlevel, sizeof newlevel) == -1) {
560 		emergency(
561 		    "cannot change kernel security level from %d to %d: %s",
562 		    curlevel, newlevel, strerror(errno));
563 		return;
564 	}
565 #ifdef SECURE
566 	warning("kernel security level changed from %d to %d",
567 	    curlevel, newlevel);
568 #endif
569 #endif
570 }
571 
572 /*
573  * Change states in the finite state machine.
574  * The initial state is passed as an argument.
575  */
576 static void
577 transition(state_t s)
578 {
579 
580 	current_state = s;
581 	for (;;)
582 		current_state = (state_t) (*current_state)();
583 }
584 
585 /*
586  * Start a session and allocate a controlling terminal.
587  * Only called by children of init after forking.
588  */
589 static void
590 open_console(void)
591 {
592 	int fd;
593 
594 	/*
595 	 * Try to open /dev/console.  Open the device with O_NONBLOCK to
596 	 * prevent potential blocking on a carrier.
597 	 */
598 	revoke(_PATH_CONSOLE);
599 	if ((fd = open(_PATH_CONSOLE, O_RDWR | O_NONBLOCK)) != -1) {
600 		(void)fcntl(fd, F_SETFL, fcntl(fd, F_GETFL) & ~O_NONBLOCK);
601 		if (login_tty(fd) == 0)
602 			return;
603 		close(fd);
604 	}
605 
606 	/* No luck.  Log output to file if possible. */
607 	if ((fd = open(_PATH_DEVNULL, O_RDWR)) == -1) {
608 		stall("cannot open null device.");
609 		_exit(1);
610 	}
611 	if (fd != STDIN_FILENO) {
612 		dup2(fd, STDIN_FILENO);
613 		close(fd);
614 	}
615 	fd = open(_PATH_INITLOG, O_WRONLY | O_APPEND | O_CREAT, 0644);
616 	if (fd == -1)
617 		dup2(STDIN_FILENO, STDOUT_FILENO);
618 	else if (fd != STDOUT_FILENO) {
619 		dup2(fd, STDOUT_FILENO);
620 		close(fd);
621 	}
622 	dup2(STDOUT_FILENO, STDERR_FILENO);
623 }
624 
625 static const char *
626 get_shell(void)
627 {
628 	static char kenv_value[PATH_MAX];
629 
630 	if (kenv(KENV_GET, "init_shell", kenv_value, sizeof(kenv_value)) > 0)
631 		return kenv_value;
632 	else
633 		return _PATH_BSHELL;
634 }
635 
636 static void
637 write_stderr(const char *message)
638 {
639 
640 	write(STDERR_FILENO, message, strlen(message));
641 }
642 
643 static int
644 read_file(const char *path, void **bufp, size_t *bufsizep)
645 {
646 	struct stat sb;
647 	size_t bufsize;
648 	void *buf;
649 	ssize_t nbytes;
650 	int error, fd;
651 
652 	fd = open(path, O_RDONLY);
653 	if (fd < 0) {
654 		emergency("%s: %s", path, strerror(errno));
655 		return (-1);
656 	}
657 
658 	error = fstat(fd, &sb);
659 	if (error != 0) {
660 		emergency("fstat: %s", strerror(errno));
661 		close(fd);
662 		return (error);
663 	}
664 
665 	bufsize = sb.st_size;
666 	buf = malloc(bufsize);
667 	if (buf == NULL) {
668 		emergency("malloc: %s", strerror(errno));
669 		close(fd);
670 		return (error);
671 	}
672 
673 	nbytes = read(fd, buf, bufsize);
674 	if (nbytes != (ssize_t)bufsize) {
675 		emergency("read: %s", strerror(errno));
676 		close(fd);
677 		free(buf);
678 		return (error);
679 	}
680 
681 	error = close(fd);
682 	if (error != 0) {
683 		emergency("close: %s", strerror(errno));
684 		free(buf);
685 		return (error);
686 	}
687 
688 	*bufp = buf;
689 	*bufsizep = bufsize;
690 
691 	return (0);
692 }
693 
694 static int
695 create_file(const char *path, const void *buf, size_t bufsize)
696 {
697 	ssize_t nbytes;
698 	int error, fd;
699 
700 	fd = open(path, O_WRONLY | O_CREAT | O_EXCL, 0700);
701 	if (fd < 0) {
702 		emergency("%s: %s", path, strerror(errno));
703 		return (-1);
704 	}
705 
706 	nbytes = write(fd, buf, bufsize);
707 	if (nbytes != (ssize_t)bufsize) {
708 		emergency("write: %s", strerror(errno));
709 		close(fd);
710 		return (-1);
711 	}
712 
713 	error = close(fd);
714 	if (error != 0) {
715 		emergency("close: %s", strerror(errno));
716 		return (-1);
717 	}
718 
719 	return (0);
720 }
721 
722 static int
723 mount_tmpfs(const char *fspath)
724 {
725 	struct iovec *iov;
726 	char errmsg[255];
727 	int error, iovlen;
728 
729 	iov = NULL;
730 	iovlen = 0;
731 	memset(errmsg, 0, sizeof(errmsg));
732 	build_iovec(&iov, &iovlen, "fstype",
733 	    __DECONST(void *, "tmpfs"), (size_t)-1);
734 	build_iovec(&iov, &iovlen, "fspath",
735 	    __DECONST(void *, fspath), (size_t)-1);
736 	build_iovec(&iov, &iovlen, "errmsg",
737 	    errmsg, sizeof(errmsg));
738 
739 	error = nmount(iov, iovlen, 0);
740 	if (error != 0) {
741 		if (*errmsg != '\0') {
742 			emergency("cannot mount tmpfs on %s: %s: %s",
743 			    fspath, errmsg, strerror(errno));
744 		} else {
745 			emergency("cannot mount tmpfs on %s: %s",
746 			    fspath, strerror(errno));
747 		}
748 		return (error);
749 	}
750 	return (0);
751 }
752 
753 static state_func_t
754 reroot(void)
755 {
756 	void *buf;
757 	size_t bufsize;
758 	int error;
759 
760 	buf = NULL;
761 	bufsize = 0;
762 
763 	revoke_ttys();
764 	runshutdown();
765 
766 	/*
767 	 * Make sure nobody can interfere with our scheme.
768 	 * Ignore ESRCH, which can apparently happen when
769 	 * there are no processes to kill.
770 	 */
771 	error = kill(-1, SIGKILL);
772 	if (error != 0 && errno != ESRCH) {
773 		emergency("kill(2) failed: %s", strerror(errno));
774 		goto out;
775 	}
776 
777 	/*
778 	 * Copy the init binary into tmpfs, so that we can unmount
779 	 * the old rootfs without committing suicide.
780 	 */
781 	error = read_file(init_path_argv0, &buf, &bufsize);
782 	if (error != 0)
783 		goto out;
784 	error = mount_tmpfs(_PATH_REROOT);
785 	if (error != 0)
786 		goto out;
787 	error = create_file(_PATH_REROOT_INIT, buf, bufsize);
788 	if (error != 0)
789 		goto out;
790 
791 	/*
792 	 * Execute the temporary init.
793 	 */
794 	execl(_PATH_REROOT_INIT, _PATH_REROOT_INIT, "-r", NULL);
795 	emergency("cannot exec %s: %s", _PATH_REROOT_INIT, strerror(errno));
796 
797 out:
798 	emergency("reroot failed; going to single user mode");
799 	free(buf);
800 	return (state_func_t) single_user;
801 }
802 
803 static state_func_t
804 reroot_phase_two(void)
805 {
806 	char init_path[PATH_MAX], *path, *path_component;
807 	size_t init_path_len;
808 	int nbytes, error;
809 
810 	/*
811 	 * Ask the kernel to mount the new rootfs.
812 	 */
813 	error = reboot(RB_REROOT);
814 	if (error != 0) {
815 		emergency("RB_REBOOT failed: %s", strerror(errno));
816 		goto out;
817 	}
818 
819 	/*
820 	 * Figure out where the destination init(8) binary is.  Note that
821 	 * the path could be different than what we've started with.  Use
822 	 * the value from kenv, if set, or the one from sysctl otherwise.
823 	 * The latter defaults to a hardcoded value, but can be overridden
824 	 * by a build time option.
825 	 */
826 	nbytes = kenv(KENV_GET, "init_path", init_path, sizeof(init_path));
827 	if (nbytes <= 0) {
828 		init_path_len = sizeof(init_path);
829 		error = sysctlbyname("kern.init_path",
830 		    init_path, &init_path_len, NULL, 0);
831 		if (error != 0) {
832 			emergency("failed to retrieve kern.init_path: %s",
833 			    strerror(errno));
834 			goto out;
835 		}
836 	}
837 
838 	/*
839 	 * Repeat the init search logic from sys/kern/init_path.c
840 	 */
841 	path_component = init_path;
842 	while ((path = strsep(&path_component, ":")) != NULL) {
843 		/*
844 		 * Execute init(8) from the new rootfs.
845 		 */
846 		execl(path, path, NULL);
847 	}
848 	emergency("cannot exec init from %s: %s", init_path, strerror(errno));
849 
850 out:
851 	emergency("reroot failed; going to single user mode");
852 	return (state_func_t) single_user;
853 }
854 
855 /*
856  * Bring the system up single user.
857  */
858 static state_func_t
859 single_user(void)
860 {
861 	pid_t pid, wpid;
862 	int status;
863 	sigset_t mask;
864 	const char *shell;
865 	char *argv[2];
866 	struct timeval tv, tn;
867 #ifdef SECURE
868 	struct ttyent *typ;
869 	struct passwd *pp;
870 	static const char banner[] =
871 		"Enter root password, or ^D to go multi-user\n";
872 	char *clear, *password;
873 #endif
874 #ifdef DEBUGSHELL
875 	char altshell[128];
876 #endif
877 
878 	if (Reboot) {
879 		/* Instead of going single user, let's reboot the machine */
880 		sync();
881 		if (reboot(howto) == -1) {
882 			emergency("reboot(%#x) failed, %s", howto,
883 			    strerror(errno));
884 			_exit(1); /* panic and reboot */
885 		}
886 		warning("reboot(%#x) returned", howto);
887 		_exit(0); /* panic as well */
888 	}
889 
890 	shell = get_shell();
891 
892 	if ((pid = fork()) == 0) {
893 		/*
894 		 * Start the single user session.
895 		 */
896 		open_console();
897 
898 #ifdef SECURE
899 		/*
900 		 * Check the root password.
901 		 * We don't care if the console is 'on' by default;
902 		 * it's the only tty that can be 'off' and 'secure'.
903 		 */
904 		typ = getttynam("console");
905 		pp = getpwnam("root");
906 		if (typ && (typ->ty_status & TTY_SECURE) == 0 &&
907 		    pp && *pp->pw_passwd) {
908 			write_stderr(banner);
909 			for (;;) {
910 				clear = getpass("Password:");
911 				if (clear == NULL || *clear == '\0')
912 					_exit(0);
913 				password = crypt(clear, pp->pw_passwd);
914 				bzero(clear, _PASSWORD_LEN);
915 				if (password != NULL &&
916 				    strcmp(password, pp->pw_passwd) == 0)
917 					break;
918 				warning("single-user login failed\n");
919 			}
920 		}
921 		endttyent();
922 		endpwent();
923 #endif /* SECURE */
924 
925 #ifdef DEBUGSHELL
926 		{
927 			char *cp = altshell;
928 			int num;
929 
930 #define	SHREQUEST "Enter full pathname of shell or RETURN for "
931 			write_stderr(SHREQUEST);
932 			write_stderr(shell);
933 			write_stderr(": ");
934 			while ((num = read(STDIN_FILENO, cp, 1)) != -1 &&
935 			    num != 0 && *cp != '\n' && cp < &altshell[127])
936 				cp++;
937 			*cp = '\0';
938 			if (altshell[0] != '\0')
939 				shell = altshell;
940 		}
941 #endif /* DEBUGSHELL */
942 
943 		/*
944 		 * Unblock signals.
945 		 * We catch all the interesting ones,
946 		 * and those are reset to SIG_DFL on exec.
947 		 */
948 		sigemptyset(&mask);
949 		sigprocmask(SIG_SETMASK, &mask, NULL);
950 
951 		/*
952 		 * Fire off a shell.
953 		 * If the default one doesn't work, try the Bourne shell.
954 		 */
955 
956 		char name[] = "-sh";
957 
958 		argv[0] = name;
959 		argv[1] = NULL;
960 		execv(shell, argv);
961 		emergency("can't exec %s for single user: %m", shell);
962 		execv(_PATH_BSHELL, argv);
963 		emergency("can't exec %s for single user: %m", _PATH_BSHELL);
964 		sleep(STALL_TIMEOUT);
965 		_exit(1);
966 	}
967 
968 	if (pid == -1) {
969 		/*
970 		 * We are seriously hosed.  Do our best.
971 		 */
972 		emergency("can't fork single-user shell, trying again");
973 		while (waitpid(-1, (int *) 0, WNOHANG) > 0)
974 			continue;
975 		return (state_func_t) single_user;
976 	}
977 
978 	requested_transition = 0;
979 	do {
980 		if ((wpid = waitpid(-1, &status, WUNTRACED)) != -1)
981 			collect_child(wpid);
982 		if (wpid == -1) {
983 			if (errno == EINTR)
984 				continue;
985 			warning("wait for single-user shell failed: %m; restarting");
986 			return (state_func_t) single_user;
987 		}
988 		if (wpid == pid && WIFSTOPPED(status)) {
989 			warning("init: shell stopped, restarting\n");
990 			kill(pid, SIGCONT);
991 			wpid = -1;
992 		}
993 	} while (wpid != pid && !requested_transition);
994 
995 	if (requested_transition)
996 		return (state_func_t) requested_transition;
997 
998 	if (!WIFEXITED(status)) {
999 		if (WTERMSIG(status) == SIGKILL) {
1000 			/*
1001 			 *  reboot(8) killed shell?
1002 			 */
1003 			warning("single user shell terminated.");
1004 			gettimeofday(&tv, NULL);
1005 			tn = tv;
1006 			tv.tv_sec += STALL_TIMEOUT;
1007 			while (tv.tv_sec > tn.tv_sec || (tv.tv_sec ==
1008 			    tn.tv_sec && tv.tv_usec > tn.tv_usec)) {
1009 				sleep(1);
1010 				gettimeofday(&tn, NULL);
1011 			}
1012 			_exit(0);
1013 		} else {
1014 			warning("single user shell terminated, restarting");
1015 			return (state_func_t) single_user;
1016 		}
1017 	}
1018 
1019 	runcom_mode = FASTBOOT;
1020 	return (state_func_t) runcom;
1021 }
1022 
1023 /*
1024  * Run the system startup script.
1025  */
1026 static state_func_t
1027 runcom(void)
1028 {
1029 	state_func_t next_transition;
1030 
1031 	if ((next_transition = run_script(_PATH_RUNCOM)) != NULL)
1032 		return next_transition;
1033 
1034 	runcom_mode = AUTOBOOT;		/* the default */
1035 	return (state_func_t) read_ttys;
1036 }
1037 
1038 static void
1039 execute_script(char *argv[])
1040 {
1041 	struct sigaction sa;
1042 	const char *shell, *script;
1043 	int error;
1044 
1045 	bzero(&sa, sizeof(sa));
1046 	sigemptyset(&sa.sa_mask);
1047 	sa.sa_handler = SIG_IGN;
1048 	sigaction(SIGTSTP, &sa, NULL);
1049 	sigaction(SIGHUP, &sa, NULL);
1050 
1051 	open_console();
1052 
1053 	sigprocmask(SIG_SETMASK, &sa.sa_mask, NULL);
1054 #ifdef LOGIN_CAP
1055 	setprocresources(RESOURCE_RC);
1056 #endif
1057 
1058 	/*
1059 	 * Try to directly execute the script first.  If it
1060 	 * fails, try the old method of passing the script path
1061 	 * to sh(1).  Don't complain if it fails because of
1062 	 * the missing execute bit.
1063 	 */
1064 	script = argv[1];
1065 	error = access(script, X_OK);
1066 	if (error == 0) {
1067 		execv(script, argv + 1);
1068 		warning("can't directly exec %s: %m", script);
1069 	} else if (errno != EACCES) {
1070 		warning("can't access %s: %m", script);
1071 	}
1072 
1073 	shell = get_shell();
1074 	execv(shell, argv);
1075 	stall("can't exec %s for %s: %m", shell, script);
1076 }
1077 
1078 /*
1079  * Execute binary, replacing init(8) as PID 1.
1080  */
1081 static void
1082 replace_init(char *path)
1083 {
1084 	char *argv[3];
1085 	char sh[] = "sh";
1086 
1087 	argv[0] = sh;
1088 	argv[1] = path;
1089 	argv[2] = NULL;
1090 
1091 	execute_script(argv);
1092 }
1093 
1094 /*
1095  * Run a shell script.
1096  * Returns 0 on success, otherwise the next transition to enter:
1097  *  - single_user if fork/execv/waitpid failed, or if the script
1098  *    terminated with a signal or exit code != 0.
1099  *  - death_single if a SIGTERM was delivered to init(8).
1100  */
1101 static state_func_t
1102 run_script(const char *script)
1103 {
1104 	pid_t pid, wpid;
1105 	int status;
1106 	char *argv[4];
1107 	const char *shell;
1108 
1109 	shell = get_shell();
1110 
1111 	if ((pid = fork()) == 0) {
1112 
1113 		char _sh[]		= "sh";
1114 		char _autoboot[]	= "autoboot";
1115 
1116 		argv[0] = _sh;
1117 		argv[1] = __DECONST(char *, script);
1118 		argv[2] = runcom_mode == AUTOBOOT ? _autoboot : 0;
1119 		argv[3] = NULL;
1120 
1121 		execute_script(argv);
1122 		sleep(STALL_TIMEOUT);
1123 		_exit(1);	/* force single user mode */
1124 	}
1125 
1126 	if (pid == -1) {
1127 		emergency("can't fork for %s on %s: %m", shell, script);
1128 		while (waitpid(-1, (int *) 0, WNOHANG) > 0)
1129 			continue;
1130 		sleep(STALL_TIMEOUT);
1131 		return (state_func_t) single_user;
1132 	}
1133 
1134 	/*
1135 	 * Copied from single_user().  This is a bit paranoid.
1136 	 */
1137 	requested_transition = 0;
1138 	do {
1139 		if ((wpid = waitpid(-1, &status, WUNTRACED)) != -1)
1140 			collect_child(wpid);
1141 		if (wpid == -1) {
1142 			if (requested_transition == death_single ||
1143 			    requested_transition == reroot)
1144 				return (state_func_t) requested_transition;
1145 			if (errno == EINTR)
1146 				continue;
1147 			warning("wait for %s on %s failed: %m; going to "
1148 			    "single user mode", shell, script);
1149 			return (state_func_t) single_user;
1150 		}
1151 		if (wpid == pid && WIFSTOPPED(status)) {
1152 			warning("init: %s on %s stopped, restarting\n",
1153 			    shell, script);
1154 			kill(pid, SIGCONT);
1155 			wpid = -1;
1156 		}
1157 	} while (wpid != pid);
1158 
1159 	if (WIFSIGNALED(status) && WTERMSIG(status) == SIGTERM &&
1160 	    requested_transition == catatonia) {
1161 		/* /etc/rc executed /sbin/reboot; wait for the end quietly */
1162 		sigset_t s;
1163 
1164 		sigfillset(&s);
1165 		for (;;)
1166 			sigsuspend(&s);
1167 	}
1168 
1169 	if (!WIFEXITED(status)) {
1170 		warning("%s on %s terminated abnormally, going to single "
1171 		    "user mode", shell, script);
1172 		return (state_func_t) single_user;
1173 	}
1174 
1175 	if (WEXITSTATUS(status))
1176 		return (state_func_t) single_user;
1177 
1178 	return (state_func_t) 0;
1179 }
1180 
1181 /*
1182  * Open the session database.
1183  *
1184  * NB: We could pass in the size here; is it necessary?
1185  */
1186 static int
1187 start_session_db(void)
1188 {
1189 	if (session_db && (*session_db->close)(session_db))
1190 		emergency("session database close: %s", strerror(errno));
1191 	if ((session_db = dbopen(NULL, O_RDWR, 0, DB_HASH, NULL)) == NULL) {
1192 		emergency("session database open: %s", strerror(errno));
1193 		return (1);
1194 	}
1195 	return (0);
1196 
1197 }
1198 
1199 /*
1200  * Add a new login session.
1201  */
1202 static void
1203 add_session(session_t *sp)
1204 {
1205 	DBT key;
1206 	DBT data;
1207 
1208 	key.data = &sp->se_process;
1209 	key.size = sizeof sp->se_process;
1210 	data.data = &sp;
1211 	data.size = sizeof sp;
1212 
1213 	if ((*session_db->put)(session_db, &key, &data, 0))
1214 		emergency("insert %d: %s", sp->se_process, strerror(errno));
1215 }
1216 
1217 /*
1218  * Delete an old login session.
1219  */
1220 static void
1221 del_session(session_t *sp)
1222 {
1223 	DBT key;
1224 
1225 	key.data = &sp->se_process;
1226 	key.size = sizeof sp->se_process;
1227 
1228 	if ((*session_db->del)(session_db, &key, 0))
1229 		emergency("delete %d: %s", sp->se_process, strerror(errno));
1230 }
1231 
1232 /*
1233  * Look up a login session by pid.
1234  */
1235 static session_t *
1236 find_session(pid_t pid)
1237 {
1238 	DBT key;
1239 	DBT data;
1240 	session_t *ret;
1241 
1242 	key.data = &pid;
1243 	key.size = sizeof pid;
1244 	if ((*session_db->get)(session_db, &key, &data, 0) != 0)
1245 		return 0;
1246 	bcopy(data.data, (char *)&ret, sizeof(ret));
1247 	return ret;
1248 }
1249 
1250 /*
1251  * Construct an argument vector from a command line.
1252  */
1253 static char **
1254 construct_argv(char *command)
1255 {
1256 	int argc = 0;
1257 	char **argv = (char **) malloc(((strlen(command) + 1) / 2 + 1)
1258 						* sizeof (char *));
1259 
1260 	if ((argv[argc++] = strk(command)) == NULL) {
1261 		free(argv);
1262 		return (NULL);
1263 	}
1264 	while ((argv[argc++] = strk((char *) 0)) != NULL)
1265 		continue;
1266 	return argv;
1267 }
1268 
1269 /*
1270  * Deallocate a session descriptor.
1271  */
1272 static void
1273 free_session(session_t *sp)
1274 {
1275 	free(sp->se_device);
1276 	if (sp->se_getty) {
1277 		free(sp->se_getty);
1278 		free(sp->se_getty_argv_space);
1279 		free(sp->se_getty_argv);
1280 	}
1281 	if (sp->se_window) {
1282 		free(sp->se_window);
1283 		free(sp->se_window_argv_space);
1284 		free(sp->se_window_argv);
1285 	}
1286 	if (sp->se_type)
1287 		free(sp->se_type);
1288 	free(sp);
1289 }
1290 
1291 /*
1292  * Allocate a new session descriptor.
1293  * Mark it SE_PRESENT.
1294  */
1295 static session_t *
1296 new_session(session_t *sprev, struct ttyent *typ)
1297 {
1298 	session_t *sp;
1299 
1300 	if ((typ->ty_status & TTY_ON) == 0 ||
1301 	    typ->ty_name == 0 ||
1302 	    typ->ty_getty == 0)
1303 		return 0;
1304 
1305 	sp = (session_t *) calloc(1, sizeof (session_t));
1306 
1307 	sp->se_flags |= SE_PRESENT;
1308 
1309 	if ((typ->ty_status & TTY_IFEXISTS) != 0)
1310 		sp->se_flags |= SE_IFEXISTS;
1311 
1312 	if ((typ->ty_status & TTY_IFCONSOLE) != 0)
1313 		sp->se_flags |= SE_IFCONSOLE;
1314 
1315 	if (asprintf(&sp->se_device, "%s%s", _PATH_DEV, typ->ty_name) < 0)
1316 		err(1, "asprintf");
1317 
1318 	if (setupargv(sp, typ) == 0) {
1319 		free_session(sp);
1320 		return (0);
1321 	}
1322 
1323 	sp->se_next = 0;
1324 	if (sprev == NULL) {
1325 		sessions = sp;
1326 		sp->se_prev = 0;
1327 	} else {
1328 		sprev->se_next = sp;
1329 		sp->se_prev = sprev;
1330 	}
1331 
1332 	return sp;
1333 }
1334 
1335 /*
1336  * Calculate getty and if useful window argv vectors.
1337  */
1338 static int
1339 setupargv(session_t *sp, struct ttyent *typ)
1340 {
1341 
1342 	if (sp->se_getty) {
1343 		free(sp->se_getty);
1344 		free(sp->se_getty_argv_space);
1345 		free(sp->se_getty_argv);
1346 	}
1347 	if (asprintf(&sp->se_getty, "%s %s", typ->ty_getty, typ->ty_name) < 0)
1348 		err(1, "asprintf");
1349 	sp->se_getty_argv_space = strdup(sp->se_getty);
1350 	sp->se_getty_argv = construct_argv(sp->se_getty_argv_space);
1351 	if (sp->se_getty_argv == NULL) {
1352 		warning("can't parse getty for port %s", sp->se_device);
1353 		free(sp->se_getty);
1354 		free(sp->se_getty_argv_space);
1355 		sp->se_getty = sp->se_getty_argv_space = 0;
1356 		return (0);
1357 	}
1358 	if (sp->se_window) {
1359 		free(sp->se_window);
1360 		free(sp->se_window_argv_space);
1361 		free(sp->se_window_argv);
1362 	}
1363 	sp->se_window = sp->se_window_argv_space = 0;
1364 	sp->se_window_argv = 0;
1365 	if (typ->ty_window) {
1366 		sp->se_window = strdup(typ->ty_window);
1367 		sp->se_window_argv_space = strdup(sp->se_window);
1368 		sp->se_window_argv = construct_argv(sp->se_window_argv_space);
1369 		if (sp->se_window_argv == NULL) {
1370 			warning("can't parse window for port %s",
1371 			    sp->se_device);
1372 			free(sp->se_window_argv_space);
1373 			free(sp->se_window);
1374 			sp->se_window = sp->se_window_argv_space = 0;
1375 			return (0);
1376 		}
1377 	}
1378 	if (sp->se_type)
1379 		free(sp->se_type);
1380 	sp->se_type = typ->ty_type ? strdup(typ->ty_type) : 0;
1381 	return (1);
1382 }
1383 
1384 /*
1385  * Walk the list of ttys and create sessions for each active line.
1386  */
1387 static state_func_t
1388 read_ttys(void)
1389 {
1390 	session_t *sp, *snext;
1391 	struct ttyent *typ;
1392 
1393 	/*
1394 	 * Destroy any previous session state.
1395 	 * There shouldn't be any, but just in case...
1396 	 */
1397 	for (sp = sessions; sp; sp = snext) {
1398 		snext = sp->se_next;
1399 		free_session(sp);
1400 	}
1401 	sessions = 0;
1402 	if (start_session_db())
1403 		return (state_func_t) single_user;
1404 
1405 	/*
1406 	 * Allocate a session entry for each active port.
1407 	 * Note that sp starts at 0.
1408 	 */
1409 	while ((typ = getttyent()) != NULL)
1410 		if ((snext = new_session(sp, typ)) != NULL)
1411 			sp = snext;
1412 
1413 	endttyent();
1414 
1415 	return (state_func_t) multi_user;
1416 }
1417 
1418 /*
1419  * Start a window system running.
1420  */
1421 static void
1422 start_window_system(session_t *sp)
1423 {
1424 	pid_t pid;
1425 	sigset_t mask;
1426 	char term[64], *env[2];
1427 	int status;
1428 
1429 	if ((pid = fork()) == -1) {
1430 		emergency("can't fork for window system on port %s: %m",
1431 		    sp->se_device);
1432 		/* hope that getty fails and we can try again */
1433 		return;
1434 	}
1435 	if (pid) {
1436 		waitpid(-1, &status, 0);
1437 		return;
1438 	}
1439 
1440 	/* reparent window process to the init to not make a zombie on exit */
1441 	if ((pid = fork()) == -1) {
1442 		emergency("can't fork for window system on port %s: %m",
1443 		    sp->se_device);
1444 		_exit(1);
1445 	}
1446 	if (pid)
1447 		_exit(0);
1448 
1449 	sigemptyset(&mask);
1450 	sigprocmask(SIG_SETMASK, &mask, NULL);
1451 
1452 	if (setsid() < 0)
1453 		emergency("setsid failed (window) %m");
1454 
1455 #ifdef LOGIN_CAP
1456 	setprocresources(RESOURCE_WINDOW);
1457 #endif
1458 	if (sp->se_type) {
1459 		/* Don't use malloc after fork */
1460 		strcpy(term, "TERM=");
1461 		strlcat(term, sp->se_type, sizeof(term));
1462 		env[0] = term;
1463 		env[1] = NULL;
1464 	}
1465 	else
1466 		env[0] = NULL;
1467 	execve(sp->se_window_argv[0], sp->se_window_argv, env);
1468 	stall("can't exec window system '%s' for port %s: %m",
1469 		sp->se_window_argv[0], sp->se_device);
1470 	_exit(1);
1471 }
1472 
1473 /*
1474  * Start a login session running.
1475  */
1476 static pid_t
1477 start_getty(session_t *sp)
1478 {
1479 	pid_t pid;
1480 	sigset_t mask;
1481 	time_t current_time = time((time_t *) 0);
1482 	int too_quick = 0;
1483 	char term[64], *env[2];
1484 
1485 	if (current_time >= sp->se_started &&
1486 	    current_time - sp->se_started < GETTY_SPACING) {
1487 		if (++sp->se_nspace > GETTY_NSPACE) {
1488 			sp->se_nspace = 0;
1489 			too_quick = 1;
1490 		}
1491 	} else
1492 		sp->se_nspace = 0;
1493 
1494 	/*
1495 	 * fork(), not vfork() -- we can't afford to block.
1496 	 */
1497 	if ((pid = fork()) == -1) {
1498 		emergency("can't fork for getty on port %s: %m", sp->se_device);
1499 		return -1;
1500 	}
1501 
1502 	if (pid)
1503 		return pid;
1504 
1505 	if (too_quick) {
1506 		warning("getty repeating too quickly on port %s, sleeping %d secs",
1507 		    sp->se_device, GETTY_SLEEP);
1508 		sleep((unsigned) GETTY_SLEEP);
1509 	}
1510 
1511 	if (sp->se_window) {
1512 		start_window_system(sp);
1513 		sleep(WINDOW_WAIT);
1514 	}
1515 
1516 	sigemptyset(&mask);
1517 	sigprocmask(SIG_SETMASK, &mask, NULL);
1518 
1519 #ifdef LOGIN_CAP
1520 	setprocresources(RESOURCE_GETTY);
1521 #endif
1522 	if (sp->se_type) {
1523 		/* Don't use malloc after fork */
1524 		strcpy(term, "TERM=");
1525 		strlcat(term, sp->se_type, sizeof(term));
1526 		env[0] = term;
1527 		env[1] = NULL;
1528 	} else
1529 		env[0] = NULL;
1530 	execve(sp->se_getty_argv[0], sp->se_getty_argv, env);
1531 	stall("can't exec getty '%s' for port %s: %m",
1532 		sp->se_getty_argv[0], sp->se_device);
1533 	_exit(1);
1534 }
1535 
1536 /*
1537  * Return 1 if the session is defined as "onifexists"
1538  * or "onifconsole" and the device node does not exist.
1539  */
1540 static int
1541 session_has_no_tty(session_t *sp)
1542 {
1543 	int fd;
1544 
1545 	if ((sp->se_flags & SE_IFEXISTS) == 0 &&
1546 	    (sp->se_flags & SE_IFCONSOLE) == 0)
1547 		return (0);
1548 
1549 	fd = open(sp->se_device, O_RDONLY | O_NONBLOCK, 0);
1550 	if (fd < 0) {
1551 		if (errno == ENOENT)
1552 			return (1);
1553 		return (0);
1554 	}
1555 
1556 	close(fd);
1557 	return (0);
1558 }
1559 
1560 /*
1561  * Collect exit status for a child.
1562  * If an exiting login, start a new login running.
1563  */
1564 static void
1565 collect_child(pid_t pid)
1566 {
1567 	session_t *sp, *sprev, *snext;
1568 
1569 	if (! sessions)
1570 		return;
1571 
1572 	if (! (sp = find_session(pid)))
1573 		return;
1574 
1575 	del_session(sp);
1576 	sp->se_process = 0;
1577 
1578 	if (sp->se_flags & SE_SHUTDOWN ||
1579 	    session_has_no_tty(sp)) {
1580 		if ((sprev = sp->se_prev) != NULL)
1581 			sprev->se_next = sp->se_next;
1582 		else
1583 			sessions = sp->se_next;
1584 		if ((snext = sp->se_next) != NULL)
1585 			snext->se_prev = sp->se_prev;
1586 		free_session(sp);
1587 		return;
1588 	}
1589 
1590 	if ((pid = start_getty(sp)) == -1) {
1591 		/* serious trouble */
1592 		requested_transition = clean_ttys;
1593 		return;
1594 	}
1595 
1596 	sp->se_process = pid;
1597 	sp->se_started = time((time_t *) 0);
1598 	add_session(sp);
1599 }
1600 
1601 /*
1602  * Catch a signal and request a state transition.
1603  */
1604 static void
1605 transition_handler(int sig)
1606 {
1607 
1608 	switch (sig) {
1609 	case SIGHUP:
1610 		if (current_state == read_ttys || current_state == multi_user ||
1611 		    current_state == clean_ttys || current_state == catatonia)
1612 			requested_transition = clean_ttys;
1613 		break;
1614 	case SIGUSR2:
1615 		howto = RB_POWEROFF;
1616 	case SIGUSR1:
1617 		howto |= RB_HALT;
1618 	case SIGWINCH:
1619 	case SIGINT:
1620 		if (sig == SIGWINCH)
1621 			howto |= RB_POWERCYCLE;
1622 		Reboot = TRUE;
1623 	case SIGTERM:
1624 		if (current_state == read_ttys || current_state == multi_user ||
1625 		    current_state == clean_ttys || current_state == catatonia)
1626 			requested_transition = death;
1627 		else
1628 			requested_transition = death_single;
1629 		break;
1630 	case SIGTSTP:
1631 		if (current_state == runcom || current_state == read_ttys ||
1632 		    current_state == clean_ttys ||
1633 		    current_state == multi_user || current_state == catatonia)
1634 			requested_transition = catatonia;
1635 		break;
1636 	case SIGEMT:
1637 		requested_transition = reroot;
1638 		break;
1639 	default:
1640 		requested_transition = 0;
1641 		break;
1642 	}
1643 }
1644 
1645 /*
1646  * Take the system multiuser.
1647  */
1648 static state_func_t
1649 multi_user(void)
1650 {
1651 	pid_t pid;
1652 	session_t *sp;
1653 
1654 	requested_transition = 0;
1655 
1656 	/*
1657 	 * If the administrator has not set the security level to -1
1658 	 * to indicate that the kernel should not run multiuser in secure
1659 	 * mode, and the run script has not set a higher level of security
1660 	 * than level 1, then put the kernel into secure mode.
1661 	 */
1662 	if (getsecuritylevel() == 0)
1663 		setsecuritylevel(1);
1664 
1665 	for (sp = sessions; sp; sp = sp->se_next) {
1666 		if (sp->se_process)
1667 			continue;
1668 		if (session_has_no_tty(sp))
1669 			continue;
1670 		if ((pid = start_getty(sp)) == -1) {
1671 			/* serious trouble */
1672 			requested_transition = clean_ttys;
1673 			break;
1674 		}
1675 		sp->se_process = pid;
1676 		sp->se_started = time((time_t *) 0);
1677 		add_session(sp);
1678 	}
1679 
1680 	while (!requested_transition)
1681 		if ((pid = waitpid(-1, (int *) 0, 0)) != -1)
1682 			collect_child(pid);
1683 
1684 	return (state_func_t) requested_transition;
1685 }
1686 
1687 /*
1688  * This is an (n*2)+(n^2) algorithm.  We hope it isn't run often...
1689  */
1690 static state_func_t
1691 clean_ttys(void)
1692 {
1693 	session_t *sp, *sprev;
1694 	struct ttyent *typ;
1695 	int devlen;
1696 	char *old_getty, *old_window, *old_type;
1697 
1698 	/*
1699 	 * mark all sessions for death, (!SE_PRESENT)
1700 	 * as we find or create new ones they'll be marked as keepers,
1701 	 * we'll later nuke all the ones not found in /etc/ttys
1702 	 */
1703 	for (sp = sessions; sp != NULL; sp = sp->se_next)
1704 		sp->se_flags &= ~SE_PRESENT;
1705 
1706 	devlen = sizeof(_PATH_DEV) - 1;
1707 	while ((typ = getttyent()) != NULL) {
1708 		for (sprev = 0, sp = sessions; sp; sprev = sp, sp = sp->se_next)
1709 			if (strcmp(typ->ty_name, sp->se_device + devlen) == 0)
1710 				break;
1711 
1712 		if (sp) {
1713 			/* we want this one to live */
1714 			sp->se_flags |= SE_PRESENT;
1715 			if ((typ->ty_status & TTY_ON) == 0 ||
1716 			    typ->ty_getty == 0) {
1717 				sp->se_flags |= SE_SHUTDOWN;
1718 				kill(sp->se_process, SIGHUP);
1719 				continue;
1720 			}
1721 			sp->se_flags &= ~SE_SHUTDOWN;
1722 			old_getty = sp->se_getty ? strdup(sp->se_getty) : 0;
1723 			old_window = sp->se_window ? strdup(sp->se_window) : 0;
1724 			old_type = sp->se_type ? strdup(sp->se_type) : 0;
1725 			if (setupargv(sp, typ) == 0) {
1726 				warning("can't parse getty for port %s",
1727 					sp->se_device);
1728 				sp->se_flags |= SE_SHUTDOWN;
1729 				kill(sp->se_process, SIGHUP);
1730 			}
1731 			else if (   !old_getty
1732 				 || (!old_type && sp->se_type)
1733 				 || (old_type && !sp->se_type)
1734 				 || (!old_window && sp->se_window)
1735 				 || (old_window && !sp->se_window)
1736 				 || (strcmp(old_getty, sp->se_getty) != 0)
1737 				 || (old_window && strcmp(old_window, sp->se_window) != 0)
1738 				 || (old_type && strcmp(old_type, sp->se_type) != 0)
1739 				) {
1740 				/* Don't set SE_SHUTDOWN here */
1741 				sp->se_nspace = 0;
1742 				sp->se_started = 0;
1743 				kill(sp->se_process, SIGHUP);
1744 			}
1745 			if (old_getty)
1746 				free(old_getty);
1747 			if (old_window)
1748 				free(old_window);
1749 			if (old_type)
1750 				free(old_type);
1751 			continue;
1752 		}
1753 
1754 		new_session(sprev, typ);
1755 	}
1756 
1757 	endttyent();
1758 
1759 	/*
1760 	 * sweep through and kill all deleted sessions
1761 	 * ones who's /etc/ttys line was deleted (SE_PRESENT unset)
1762 	 */
1763 	for (sp = sessions; sp != NULL; sp = sp->se_next) {
1764 		if ((sp->se_flags & SE_PRESENT) == 0) {
1765 			sp->se_flags |= SE_SHUTDOWN;
1766 			kill(sp->se_process, SIGHUP);
1767 		}
1768 	}
1769 
1770 	return (state_func_t) multi_user;
1771 }
1772 
1773 /*
1774  * Block further logins.
1775  */
1776 static state_func_t
1777 catatonia(void)
1778 {
1779 	session_t *sp;
1780 
1781 	for (sp = sessions; sp; sp = sp->se_next)
1782 		sp->se_flags |= SE_SHUTDOWN;
1783 
1784 	return (state_func_t) multi_user;
1785 }
1786 
1787 /*
1788  * Note SIGALRM.
1789  */
1790 static void
1791 alrm_handler(int sig)
1792 {
1793 
1794 	(void)sig;
1795 	clang = 1;
1796 }
1797 
1798 /*
1799  * Bring the system down to single user.
1800  */
1801 static state_func_t
1802 death(void)
1803 {
1804 	int block, blocked;
1805 	size_t len;
1806 
1807 	/* Temporarily block suspend. */
1808 	len = sizeof(blocked);
1809 	block = 1;
1810 	if (sysctlbyname("kern.suspend_blocked", &blocked, &len,
1811 	    &block, sizeof(block)) == -1)
1812 		blocked = 0;
1813 
1814 	/*
1815 	 * Also revoke the TTY here.  Because runshutdown() may reopen
1816 	 * the TTY whose getty we're killing here, there is no guarantee
1817 	 * runshutdown() will perform the initial open() call, causing
1818 	 * the terminal attributes to be misconfigured.
1819 	 */
1820 	revoke_ttys();
1821 
1822 	/* Try to run the rc.shutdown script within a period of time */
1823 	runshutdown();
1824 
1825 	/* Unblock suspend if we blocked it. */
1826 	if (!blocked)
1827 		sysctlbyname("kern.suspend_blocked", NULL, NULL,
1828 		    &blocked, sizeof(blocked));
1829 
1830 	return (state_func_t) death_single;
1831 }
1832 
1833 /*
1834  * Do what is necessary to reinitialize single user mode or reboot
1835  * from an incomplete state.
1836  */
1837 static state_func_t
1838 death_single(void)
1839 {
1840 	int i;
1841 	pid_t pid;
1842 	static const int death_sigs[2] = { SIGTERM, SIGKILL };
1843 
1844 	revoke(_PATH_CONSOLE);
1845 
1846 	for (i = 0; i < 2; ++i) {
1847 		if (kill(-1, death_sigs[i]) == -1 && errno == ESRCH)
1848 			return (state_func_t) single_user;
1849 
1850 		clang = 0;
1851 		alarm(DEATH_WATCH);
1852 		do
1853 			if ((pid = waitpid(-1, (int *)0, 0)) != -1)
1854 				collect_child(pid);
1855 		while (clang == 0 && errno != ECHILD);
1856 
1857 		if (errno == ECHILD)
1858 			return (state_func_t) single_user;
1859 	}
1860 
1861 	warning("some processes would not die; ps axl advised");
1862 
1863 	return (state_func_t) single_user;
1864 }
1865 
1866 static void
1867 revoke_ttys(void)
1868 {
1869 	session_t *sp;
1870 
1871 	for (sp = sessions; sp; sp = sp->se_next) {
1872 		sp->se_flags |= SE_SHUTDOWN;
1873 		kill(sp->se_process, SIGHUP);
1874 		revoke(sp->se_device);
1875 	}
1876 }
1877 
1878 /*
1879  * Run the system shutdown script.
1880  *
1881  * Exit codes:      XXX I should document more
1882  * -2       shutdown script terminated abnormally
1883  * -1       fatal error - can't run script
1884  * 0        good.
1885  * >0       some error (exit code)
1886  */
1887 static int
1888 runshutdown(void)
1889 {
1890 	pid_t pid, wpid;
1891 	int status;
1892 	int shutdowntimeout;
1893 	size_t len;
1894 	char *argv[4];
1895 	struct stat sb;
1896 
1897 	/*
1898 	 * rc.shutdown is optional, so to prevent any unnecessary
1899 	 * complaints from the shell we simply don't run it if the
1900 	 * file does not exist. If the stat() here fails for other
1901 	 * reasons, we'll let the shell complain.
1902 	 */
1903 	if (stat(_PATH_RUNDOWN, &sb) == -1 && errno == ENOENT)
1904 		return 0;
1905 
1906 	if ((pid = fork()) == 0) {
1907 		char _sh[]	= "sh";
1908 		char _reboot[]	= "reboot";
1909 		char _single[]	= "single";
1910 		char _path_rundown[] = _PATH_RUNDOWN;
1911 
1912 		argv[0] = _sh;
1913 		argv[1] = _path_rundown;
1914 		argv[2] = Reboot ? _reboot : _single;
1915 		argv[3] = NULL;
1916 
1917 		execute_script(argv);
1918 		_exit(1);	/* force single user mode */
1919 	}
1920 
1921 	if (pid == -1) {
1922 		emergency("can't fork for %s: %m", _PATH_RUNDOWN);
1923 		while (waitpid(-1, (int *) 0, WNOHANG) > 0)
1924 			continue;
1925 		sleep(STALL_TIMEOUT);
1926 		return -1;
1927 	}
1928 
1929 	len = sizeof(shutdowntimeout);
1930 	if (sysctlbyname("kern.init_shutdown_timeout", &shutdowntimeout, &len,
1931 	    NULL, 0) == -1 || shutdowntimeout < 2)
1932 		shutdowntimeout = DEATH_SCRIPT;
1933 	alarm(shutdowntimeout);
1934 	clang = 0;
1935 	/*
1936 	 * Copied from single_user().  This is a bit paranoid.
1937 	 * Use the same ALRM handler.
1938 	 */
1939 	do {
1940 		if ((wpid = waitpid(-1, &status, WUNTRACED)) != -1)
1941 			collect_child(wpid);
1942 		if (clang == 1) {
1943 			/* we were waiting for the sub-shell */
1944 			kill(wpid, SIGTERM);
1945 			warning("timeout expired for %s: %m; going to "
1946 			    "single user mode", _PATH_RUNDOWN);
1947 			return -1;
1948 		}
1949 		if (wpid == -1) {
1950 			if (errno == EINTR)
1951 				continue;
1952 			warning("wait for %s failed: %m; going to "
1953 			    "single user mode", _PATH_RUNDOWN);
1954 			return -1;
1955 		}
1956 		if (wpid == pid && WIFSTOPPED(status)) {
1957 			warning("init: %s stopped, restarting\n",
1958 			    _PATH_RUNDOWN);
1959 			kill(pid, SIGCONT);
1960 			wpid = -1;
1961 		}
1962 	} while (wpid != pid && !clang);
1963 
1964 	/* Turn off the alarm */
1965 	alarm(0);
1966 
1967 	if (WIFSIGNALED(status) && WTERMSIG(status) == SIGTERM &&
1968 	    requested_transition == catatonia) {
1969 		/*
1970 		 * /etc/rc.shutdown executed /sbin/reboot;
1971 		 * wait for the end quietly
1972 		 */
1973 		sigset_t s;
1974 
1975 		sigfillset(&s);
1976 		for (;;)
1977 			sigsuspend(&s);
1978 	}
1979 
1980 	if (!WIFEXITED(status)) {
1981 		warning("%s terminated abnormally, going to "
1982 		    "single user mode", _PATH_RUNDOWN);
1983 		return -2;
1984 	}
1985 
1986 	if ((status = WEXITSTATUS(status)) != 0)
1987 		warning("%s returned status %d", _PATH_RUNDOWN, status);
1988 
1989 	return status;
1990 }
1991 
1992 static char *
1993 strk(char *p)
1994 {
1995 	static char *t;
1996 	char *q;
1997 	int c;
1998 
1999 	if (p)
2000 		t = p;
2001 	if (!t)
2002 		return 0;
2003 
2004 	c = *t;
2005 	while (c == ' ' || c == '\t' )
2006 		c = *++t;
2007 	if (!c) {
2008 		t = 0;
2009 		return 0;
2010 	}
2011 	q = t;
2012 	if (c == '\'') {
2013 		c = *++t;
2014 		q = t;
2015 		while (c && c != '\'')
2016 			c = *++t;
2017 		if (!c)  /* unterminated string */
2018 			q = t = 0;
2019 		else
2020 			*t++ = 0;
2021 	} else {
2022 		while (c && c != ' ' && c != '\t' )
2023 			c = *++t;
2024 		*t++ = 0;
2025 		if (!c)
2026 			t = 0;
2027 	}
2028 	return q;
2029 }
2030 
2031 #ifdef LOGIN_CAP
2032 static void
2033 setprocresources(const char *cname)
2034 {
2035 	login_cap_t *lc;
2036 	if ((lc = login_getclassbyname(cname, NULL)) != NULL) {
2037 		setusercontext(lc, (struct passwd*)NULL, 0,
2038 		    LOGIN_SETENV |
2039 		    LOGIN_SETPRIORITY | LOGIN_SETRESOURCES |
2040 		    LOGIN_SETLOGINCLASS | LOGIN_SETCPUMASK);
2041 		login_close(lc);
2042 	}
2043 }
2044 #endif
2045