xref: /freebsd/sbin/init/init.c (revision 525fe93dc7487a1e63a90f6a2b956abc601963c1)
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 #include <sys/param.h>
36 #include <sys/boottrace.h>
37 #include <sys/ioctl.h>
38 #include <sys/mman.h>
39 #include <sys/mount.h>
40 #include <sys/reboot.h>
41 #include <sys/stat.h>
42 #include <sys/sysctl.h>
43 #include <sys/uio.h>
44 #include <sys/wait.h>
45 
46 #include <db.h>
47 #include <err.h>
48 #include <errno.h>
49 #include <fcntl.h>
50 #include <kenv.h>
51 #include <libutil.h>
52 #include <paths.h>
53 #include <signal.h>
54 #include <stdarg.h>
55 #include <stdbool.h>
56 #include <stdio.h>
57 #include <stdlib.h>
58 #include <string.h>
59 #include <syslog.h>
60 #include <time.h>
61 #include <ttyent.h>
62 #include <unistd.h>
63 
64 #ifdef SECURE
65 #include <pwd.h>
66 #endif
67 
68 #ifdef LOGIN_CAP
69 #include <login_cap.h>
70 #endif
71 
72 #include "mntopts.h"
73 #include "pathnames.h"
74 
75 /*
76  * Sleep times; used to prevent thrashing.
77  */
78 #define	GETTY_SPACING		 5	/* N secs minimum getty spacing */
79 #define	GETTY_SLEEP		30	/* sleep N secs after spacing problem */
80 #define	GETTY_NSPACE		 3	/* max. spacing count to bring reaction */
81 #define	WINDOW_WAIT		 3	/* wait N secs after starting window */
82 #define	STALL_TIMEOUT		30	/* wait N secs after warning */
83 #define	DEATH_WATCH		10	/* wait N secs for procs to die */
84 #define	DEATH_SCRIPT		120	/* wait for 2min for /etc/rc.shutdown */
85 #define	RESOURCE_RC		"daemon"
86 #define	RESOURCE_WINDOW		"default"
87 #define	RESOURCE_GETTY		"default"
88 #define SCRIPT_ARGV_SIZE 3 /* size of argv passed to execute_script, can be increased if needed */
89 
90 static void handle(sig_t, ...);
91 static void delset(sigset_t *, ...);
92 
93 static void stall(const char *, ...) __printflike(1, 2);
94 static void warning(const char *, ...) __printflike(1, 2);
95 static void emergency(const char *, ...) __printflike(1, 2);
96 static void disaster(int);
97 static void revoke_ttys(void);
98 static int  runshutdown(void);
99 static char *strk(char *);
100 static void runfinal(void);
101 
102 /*
103  * We really need a recursive typedef...
104  * The following at least guarantees that the return type of (*state_t)()
105  * is sufficiently wide to hold a function pointer.
106  */
107 typedef long (*state_func_t)(void);
108 typedef state_func_t (*state_t)(void);
109 
110 static state_func_t single_user(void);
111 static state_func_t runcom(void);
112 static state_func_t read_ttys(void);
113 static state_func_t multi_user(void);
114 static state_func_t clean_ttys(void);
115 static state_func_t catatonia(void);
116 static state_func_t death(void);
117 static state_func_t death_single(void);
118 static state_func_t reroot(void);
119 static state_func_t reroot_phase_two(void);
120 
121 static state_func_t run_script(const char *);
122 
123 static enum { AUTOBOOT, FASTBOOT } runcom_mode = AUTOBOOT;
124 
125 static bool Reboot = false;
126 static int howto = RB_AUTOBOOT;
127 
128 static bool devfs = false;
129 static char *init_path_argv0;
130 
131 static void transition(state_t);
132 static state_t requested_transition;
133 static state_t current_state = death_single;
134 
135 static void execute_script(char *argv[]);
136 static void open_console(void);
137 static const char *get_shell(void);
138 static void replace_init(char *path);
139 static void write_stderr(const char *message);
140 
141 typedef struct init_session {
142 	pid_t	se_process;		/* controlling process */
143 	time_t	se_started;		/* used to avoid thrashing */
144 	int	se_flags;		/* status of session */
145 #define	SE_SHUTDOWN	0x1		/* session won't be restarted */
146 #define	SE_PRESENT	0x2		/* session is in /etc/ttys */
147 #define	SE_IFEXISTS	0x4		/* session defined as "onifexists" */
148 #define	SE_IFCONSOLE	0x8		/* session defined as "onifconsole" */
149 	int	se_nspace;		/* spacing count */
150 	char	*se_device;		/* filename of port */
151 	char	*se_getty;		/* what to run on that port */
152 	char	*se_getty_argv_space;   /* pre-parsed argument array space */
153 	char	**se_getty_argv;	/* pre-parsed argument array */
154 	char	*se_window;		/* window system (started only once) */
155 	char	*se_window_argv_space;  /* pre-parsed argument array space */
156 	char	**se_window_argv;	/* pre-parsed argument array */
157 	char	*se_type;		/* default terminal type */
158 	struct	init_session *se_prev;
159 	struct	init_session *se_next;
160 } session_t;
161 
162 static void free_session(session_t *);
163 static session_t *new_session(session_t *, struct ttyent *);
164 static session_t *sessions;
165 
166 static char **construct_argv(char *);
167 static void start_window_system(session_t *);
168 static void collect_child(pid_t);
169 static pid_t start_getty(session_t *);
170 static void transition_handler(int);
171 static void alrm_handler(int);
172 static void setsecuritylevel(int);
173 static int getsecuritylevel(void);
174 static int setupargv(session_t *, struct ttyent *);
175 #ifdef LOGIN_CAP
176 static void setprocresources(const char *);
177 #endif
178 static bool clang;
179 
180 static int start_session_db(void);
181 static void add_session(session_t *);
182 static void del_session(session_t *);
183 static session_t *find_session(pid_t);
184 static DB *session_db;
185 
186 /*
187  * The mother of all processes.
188  */
189 int
190 main(int argc, char *argv[])
191 {
192 	state_t initial_transition = runcom;
193 	char kenv_value[PATH_MAX];
194 	int c, error;
195 	struct sigaction sa;
196 	sigset_t mask;
197 
198 	/* Dispose of random users. */
199 	if (getuid() != 0)
200 		errx(1, "%s", strerror(EPERM));
201 
202 	BOOTTRACE("init(8) starting...");
203 
204 	/* System V users like to reexec init. */
205 	if (getpid() != 1) {
206 #ifdef COMPAT_SYSV_INIT
207 		/* So give them what they want */
208 		if (argc > 1) {
209 			if (strlen(argv[1]) == 1) {
210 				char runlevel = *argv[1];
211 				int sig;
212 
213 				switch (runlevel) {
214 				case '0': /* halt + poweroff */
215 					sig = SIGUSR2;
216 					break;
217 				case '1': /* single-user */
218 					sig = SIGTERM;
219 					break;
220 				case '6': /* reboot */
221 					sig = SIGINT;
222 					break;
223 				case 'c': /* block further logins */
224 					sig = SIGTSTP;
225 					break;
226 				case 'q': /* rescan /etc/ttys */
227 					sig = SIGHUP;
228 					break;
229 				case 'r': /* remount root */
230 					sig = SIGEMT;
231 					break;
232 				default:
233 					goto invalid;
234 				}
235 				kill(1, sig);
236 				_exit(0);
237 			} else
238 invalid:
239 				errx(1, "invalid run-level ``%s''", argv[1]);
240 		} else
241 #endif
242 			errx(1, "already running");
243 	}
244 
245 	init_path_argv0 = strdup(argv[0]);
246 	if (init_path_argv0 == NULL)
247 		err(1, "strdup");
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 = true;
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(disaster, SIGABRT, SIGFPE, SIGILL, SIGSEGV, SIGBUS, SIGSYS,
299 	    SIGXCPU, SIGXFSZ, 0);
300 	handle(transition_handler, SIGHUP, SIGINT, SIGEMT, SIGTERM, SIGTSTP,
301 	    SIGUSR1, SIGUSR2, SIGWINCH, 0);
302 	handle(alrm_handler, SIGALRM, 0);
303 	sigfillset(&mask);
304 	delset(&mask, SIGABRT, SIGFPE, SIGILL, SIGSEGV, SIGBUS, SIGSYS,
305 	    SIGXCPU, SIGXFSZ, SIGHUP, SIGINT, SIGEMT, SIGTERM, SIGTSTP,
306 	    SIGALRM, SIGUSR1, SIGUSR2, SIGWINCH, 0);
307 	sigprocmask(SIG_SETMASK, &mask, NULL);
308 	sigemptyset(&sa.sa_mask);
309 	sa.sa_flags = 0;
310 	sa.sa_handler = SIG_IGN;
311 	sigaction(SIGTTIN, &sa, NULL);
312 	sigaction(SIGTTOU, &sa, NULL);
313 
314 	/*
315 	 * Paranoia.
316 	 */
317 	close(0);
318 	close(1);
319 	close(2);
320 
321 	if (kenv(KENV_GET, "init_exec", kenv_value, sizeof(kenv_value)) > 0) {
322 		replace_init(kenv_value);
323 		_exit(0); /* reboot */
324 	}
325 
326 	if (kenv(KENV_GET, "init_script", kenv_value, sizeof(kenv_value)) > 0) {
327 		state_func_t next_transition;
328 
329 		if ((next_transition = run_script(kenv_value)) != NULL)
330 			initial_transition = (state_t) next_transition;
331 	}
332 
333 	if (kenv(KENV_GET, "init_chroot", kenv_value, sizeof(kenv_value)) > 0) {
334 		if (chdir(kenv_value) != 0 || chroot(".") != 0)
335 			warning("Can't chroot to %s: %m", kenv_value);
336 	}
337 
338 	/*
339 	 * Additional check if devfs needs to be mounted:
340 	 * If "/" and "/dev" have the same device number,
341 	 * then it hasn't been mounted yet.
342 	 */
343 	if (!devfs) {
344 		struct stat stst;
345 		dev_t root_devno;
346 
347 		stat("/", &stst);
348 		root_devno = stst.st_dev;
349 		if (stat("/dev", &stst) != 0)
350 			warning("Can't stat /dev: %m");
351 		else if (stst.st_dev == root_devno)
352 			devfs = true;
353 	}
354 
355 	if (devfs) {
356 		struct iovec iov[4];
357 		char *s;
358 		int i;
359 
360 		char _fstype[]	= "fstype";
361 		char _devfs[]	= "devfs";
362 		char _fspath[]	= "fspath";
363 		char _path_dev[]= _PATH_DEV;
364 
365 		iov[0].iov_base = _fstype;
366 		iov[0].iov_len = sizeof(_fstype);
367 		iov[1].iov_base = _devfs;
368 		iov[1].iov_len = sizeof(_devfs);
369 		iov[2].iov_base = _fspath;
370 		iov[2].iov_len = sizeof(_fspath);
371 		/*
372 		 * Try to avoid the trailing slash in _PATH_DEV.
373 		 * Be *very* defensive.
374 		 */
375 		s = strdup(_PATH_DEV);
376 		if (s != NULL) {
377 			i = strlen(s);
378 			if (i > 0 && s[i - 1] == '/')
379 				s[i - 1] = '\0';
380 			iov[3].iov_base = s;
381 			iov[3].iov_len = strlen(s) + 1;
382 		} else {
383 			iov[3].iov_base = _path_dev;
384 			iov[3].iov_len = sizeof(_path_dev);
385 		}
386 		nmount(iov, 4, 0);
387 		if (s != NULL)
388 			free(s);
389 	}
390 
391 	if (initial_transition != reroot_phase_two) {
392 		/*
393 		 * Unmount reroot leftovers.  This runs after init(8)
394 		 * gets reexecuted after reroot_phase_two() is done.
395 		 */
396 		error = unmount(_PATH_REROOT, MNT_FORCE);
397 		if (error != 0 && errno != EINVAL)
398 			warning("Cannot unmount %s: %m", _PATH_REROOT);
399 	}
400 
401 	/*
402 	 * Start the state machine.
403 	 */
404 	transition(initial_transition);
405 
406 	/*
407 	 * Should never reach here.
408 	 */
409 	return 1;
410 }
411 
412 /*
413  * Associate a function with a signal handler.
414  */
415 static void
416 handle(sig_t handler, ...)
417 {
418 	int sig;
419 	struct sigaction sa;
420 	sigset_t mask_everything;
421 	va_list ap;
422 	va_start(ap, handler);
423 
424 	sa.sa_handler = handler;
425 	sigfillset(&mask_everything);
426 
427 	while ((sig = va_arg(ap, int)) != 0) {
428 		sa.sa_mask = mask_everything;
429 		/* XXX SA_RESTART? */
430 		sa.sa_flags = sig == SIGCHLD ? SA_NOCLDSTOP : 0;
431 		sigaction(sig, &sa, NULL);
432 	}
433 	va_end(ap);
434 }
435 
436 /*
437  * Delete a set of signals from a mask.
438  */
439 static void
440 delset(sigset_t *maskp, ...)
441 {
442 	int sig;
443 	va_list ap;
444 	va_start(ap, maskp);
445 
446 	while ((sig = va_arg(ap, int)) != 0)
447 		sigdelset(maskp, sig);
448 	va_end(ap);
449 }
450 
451 /*
452  * Log a message and sleep for a while (to give someone an opportunity
453  * to read it and to save log or hardcopy output if the problem is chronic).
454  * NB: should send a message to the session logger to avoid blocking.
455  */
456 static void
457 stall(const char *message, ...)
458 {
459 	va_list ap;
460 	va_start(ap, message);
461 
462 	vsyslog(LOG_ALERT, message, ap);
463 	va_end(ap);
464 	sleep(STALL_TIMEOUT);
465 }
466 
467 /*
468  * Like stall(), but doesn't sleep.
469  * If cpp had variadic macros, the two functions could be #defines for another.
470  * NB: should send a message to the session logger to avoid blocking.
471  */
472 static void
473 warning(const char *message, ...)
474 {
475 	va_list ap;
476 	va_start(ap, message);
477 
478 	vsyslog(LOG_ALERT, message, ap);
479 	va_end(ap);
480 }
481 
482 /*
483  * Log an emergency message.
484  * NB: should send a message to the session logger to avoid blocking.
485  */
486 static void
487 emergency(const char *message, ...)
488 {
489 	va_list ap;
490 	va_start(ap, message);
491 
492 	vsyslog(LOG_EMERG, message, ap);
493 	va_end(ap);
494 }
495 
496 /*
497  * Catch an unexpected signal.
498  */
499 static void
500 disaster(int sig)
501 {
502 
503 	emergency("fatal signal: %s",
504 	    (unsigned)sig < NSIG ? sys_siglist[sig] : "unknown signal");
505 
506 	sleep(STALL_TIMEOUT);
507 	_exit(sig);		/* reboot */
508 }
509 
510 /*
511  * Get the security level of the kernel.
512  */
513 static int
514 getsecuritylevel(void)
515 {
516 #ifdef KERN_SECURELVL
517 	int name[2], curlevel;
518 	size_t len;
519 
520 	name[0] = CTL_KERN;
521 	name[1] = KERN_SECURELVL;
522 	len = sizeof curlevel;
523 	if (sysctl(name, 2, &curlevel, &len, NULL, 0) == -1) {
524 		emergency("cannot get kernel security level: %m");
525 		return (-1);
526 	}
527 	return (curlevel);
528 #else
529 	return (-1);
530 #endif
531 }
532 
533 /*
534  * Set the security level of the kernel.
535  */
536 static void
537 setsecuritylevel(int newlevel)
538 {
539 #ifdef KERN_SECURELVL
540 	int name[2], curlevel;
541 
542 	curlevel = getsecuritylevel();
543 	if (newlevel == curlevel)
544 		return;
545 	name[0] = CTL_KERN;
546 	name[1] = KERN_SECURELVL;
547 	if (sysctl(name, 2, NULL, NULL, &newlevel, sizeof newlevel) == -1) {
548 		emergency(
549 		    "cannot change kernel security level from %d to %d: %m",
550 		    curlevel, newlevel);
551 		return;
552 	}
553 #ifdef SECURE
554 	warning("kernel security level changed from %d to %d",
555 	    curlevel, newlevel);
556 #endif
557 #endif
558 }
559 
560 /*
561  * Change states in the finite state machine.
562  * The initial state is passed as an argument.
563  */
564 static void
565 transition(state_t s)
566 {
567 
568 	current_state = s;
569 	for (;;)
570 		current_state = (state_t) (*current_state)();
571 }
572 
573 /*
574  * Start a session and allocate a controlling terminal.
575  * Only called by children of init after forking.
576  */
577 static void
578 open_console(void)
579 {
580 	int fd;
581 
582 	/*
583 	 * Try to open /dev/console.  Open the device with O_NONBLOCK to
584 	 * prevent potential blocking on a carrier.
585 	 */
586 	revoke(_PATH_CONSOLE);
587 	if ((fd = open(_PATH_CONSOLE, O_RDWR | O_NONBLOCK)) != -1) {
588 		(void)fcntl(fd, F_SETFL, fcntl(fd, F_GETFL) & ~O_NONBLOCK);
589 		if (login_tty(fd) == 0)
590 			return;
591 		close(fd);
592 	}
593 
594 	/* No luck.  Log output to file if possible. */
595 	if ((fd = open(_PATH_DEVNULL, O_RDWR)) == -1) {
596 		stall("cannot open null device.");
597 		_exit(1);
598 	}
599 	if (fd != STDIN_FILENO) {
600 		dup2(fd, STDIN_FILENO);
601 		close(fd);
602 	}
603 	fd = open(_PATH_INITLOG, O_WRONLY | O_APPEND | O_CREAT, 0644);
604 	if (fd == -1)
605 		dup2(STDIN_FILENO, STDOUT_FILENO);
606 	else if (fd != STDOUT_FILENO) {
607 		dup2(fd, STDOUT_FILENO);
608 		close(fd);
609 	}
610 	dup2(STDOUT_FILENO, STDERR_FILENO);
611 }
612 
613 static const char *
614 get_shell(void)
615 {
616 	static char kenv_value[PATH_MAX];
617 
618 	if (kenv(KENV_GET, "init_shell", kenv_value, sizeof(kenv_value)) > 0)
619 		return kenv_value;
620 	else
621 		return _PATH_BSHELL;
622 }
623 
624 static void
625 write_stderr(const char *message)
626 {
627 
628 	write(STDERR_FILENO, message, strlen(message));
629 }
630 
631 static int
632 read_file(const char *path, void **bufp, size_t *bufsizep)
633 {
634 	struct stat sb;
635 	size_t bufsize;
636 	void *buf;
637 	ssize_t nbytes;
638 	int error, fd;
639 
640 	fd = open(path, O_RDONLY);
641 	if (fd < 0) {
642 		emergency("%s: %m", path);
643 		return (-1);
644 	}
645 
646 	error = fstat(fd, &sb);
647 	if (error != 0) {
648 		emergency("fstat: %m");
649 		close(fd);
650 		return (error);
651 	}
652 
653 	bufsize = sb.st_size;
654 	buf = malloc(bufsize);
655 	if (buf == NULL) {
656 		emergency("malloc: %m");
657 		close(fd);
658 		return (error);
659 	}
660 
661 	nbytes = read(fd, buf, bufsize);
662 	if (nbytes != (ssize_t)bufsize) {
663 		emergency("read: %m");
664 		close(fd);
665 		free(buf);
666 		return (error);
667 	}
668 
669 	error = close(fd);
670 	if (error != 0) {
671 		emergency("close: %m");
672 		free(buf);
673 		return (error);
674 	}
675 
676 	*bufp = buf;
677 	*bufsizep = bufsize;
678 
679 	return (0);
680 }
681 
682 static int
683 create_file(const char *path, const void *buf, size_t bufsize)
684 {
685 	ssize_t nbytes;
686 	int error, fd;
687 
688 	fd = open(path, O_WRONLY | O_CREAT | O_EXCL, 0700);
689 	if (fd < 0) {
690 		emergency("%s: %m", path);
691 		return (-1);
692 	}
693 
694 	nbytes = write(fd, buf, bufsize);
695 	if (nbytes != (ssize_t)bufsize) {
696 		emergency("write: %m");
697 		close(fd);
698 		return (-1);
699 	}
700 
701 	error = close(fd);
702 	if (error != 0) {
703 		emergency("close: %m");
704 		return (-1);
705 	}
706 
707 	return (0);
708 }
709 
710 static int
711 mount_tmpfs(const char *fspath)
712 {
713 	struct iovec *iov;
714 	char errmsg[255];
715 	int error, iovlen;
716 
717 	iov = NULL;
718 	iovlen = 0;
719 	memset(errmsg, 0, sizeof(errmsg));
720 	build_iovec(&iov, &iovlen, "fstype",
721 	    __DECONST(void *, "tmpfs"), (size_t)-1);
722 	build_iovec(&iov, &iovlen, "fspath",
723 	    __DECONST(void *, fspath), (size_t)-1);
724 	build_iovec(&iov, &iovlen, "errmsg",
725 	    errmsg, sizeof(errmsg));
726 
727 	error = nmount(iov, iovlen, 0);
728 	if (error != 0) {
729 		if (*errmsg != '\0') {
730 			emergency("cannot mount tmpfs on %s: %s: %m",
731 			    fspath, errmsg);
732 		} else {
733 			emergency("cannot mount tmpfs on %s: %m",
734 			    fspath);
735 		}
736 		return (error);
737 	}
738 	return (0);
739 }
740 
741 static state_func_t
742 reroot(void)
743 {
744 	void *buf;
745 	size_t bufsize;
746 	int error;
747 
748 	buf = NULL;
749 	bufsize = 0;
750 
751 	revoke_ttys();
752 	runshutdown();
753 
754 	/*
755 	 * Make sure nobody can interfere with our scheme.
756 	 * Ignore ESRCH, which can apparently happen when
757 	 * there are no processes to kill.
758 	 */
759 	error = kill(-1, SIGKILL);
760 	if (error != 0 && errno != ESRCH) {
761 		emergency("kill(2) failed: %m");
762 		goto out;
763 	}
764 
765 	/*
766 	 * Copy the init binary into tmpfs, so that we can unmount
767 	 * the old rootfs without committing suicide.
768 	 */
769 	error = read_file(init_path_argv0, &buf, &bufsize);
770 	if (error != 0)
771 		goto out;
772 	error = mount_tmpfs(_PATH_REROOT);
773 	if (error != 0)
774 		goto out;
775 	error = create_file(_PATH_REROOT_INIT, buf, bufsize);
776 	if (error != 0)
777 		goto out;
778 
779 	/*
780 	 * Execute the temporary init.
781 	 */
782 	execl(_PATH_REROOT_INIT, _PATH_REROOT_INIT, "-r", NULL);
783 	emergency("cannot exec %s: %m", _PATH_REROOT_INIT);
784 
785 out:
786 	emergency("reroot failed; going to single user mode");
787 	free(buf);
788 	return (state_func_t) single_user;
789 }
790 
791 static state_func_t
792 reroot_phase_two(void)
793 {
794 	char init_path[PATH_MAX], *path, *path_component;
795 	size_t init_path_len;
796 	int nbytes, error;
797 
798 	/*
799 	 * Ask the kernel to mount the new rootfs.
800 	 */
801 	error = reboot(RB_REROOT);
802 	if (error != 0) {
803 		emergency("RB_REBOOT failed: %m");
804 		goto out;
805 	}
806 
807 	/*
808 	 * Figure out where the destination init(8) binary is.  Note that
809 	 * the path could be different than what we've started with.  Use
810 	 * the value from kenv, if set, or the one from sysctl otherwise.
811 	 * The latter defaults to a hardcoded value, but can be overridden
812 	 * by a build time option.
813 	 */
814 	nbytes = kenv(KENV_GET, "init_path", init_path, sizeof(init_path));
815 	if (nbytes <= 0) {
816 		init_path_len = sizeof(init_path);
817 		error = sysctlbyname("kern.init_path",
818 		    init_path, &init_path_len, NULL, 0);
819 		if (error != 0) {
820 			emergency("failed to retrieve kern.init_path: %m");
821 			goto out;
822 		}
823 	}
824 
825 	/*
826 	 * Repeat the init search logic from sys/kern/init_path.c
827 	 */
828 	path_component = init_path;
829 	while ((path = strsep(&path_component, ":")) != NULL) {
830 		/*
831 		 * Execute init(8) from the new rootfs.
832 		 */
833 		execl(path, path, NULL);
834 	}
835 	emergency("cannot exec init from %s: %m", init_path);
836 
837 out:
838 	emergency("reroot failed; going to single user mode");
839 	return (state_func_t) single_user;
840 }
841 
842 /*
843  * Bring the system up single user.
844  */
845 static state_func_t
846 single_user(void)
847 {
848 	pid_t pid, wpid;
849 	int status;
850 	sigset_t mask;
851 	const char *shell;
852 	char *argv[2];
853 	struct timeval tv, tn;
854 #ifdef SECURE
855 	struct ttyent *typ;
856 	struct passwd *pp;
857 	static const char banner[] =
858 		"Enter root password, or ^D to go multi-user\n";
859 	char *clear, *password;
860 #endif
861 #ifdef DEBUGSHELL
862 	char altshell[128];
863 #endif
864 
865 	if (Reboot) {
866 		/* Instead of going single user, let's reboot the machine */
867 		BOOTTRACE("shutting down the system");
868 		sync();
869 		/* Run scripts after all processes have been terminated. */
870 		runfinal();
871 		if (reboot(howto) == -1) {
872 			emergency("reboot(%#x) failed, %m", howto);
873 			_exit(1); /* panic and reboot */
874 		}
875 		warning("reboot(%#x) returned", howto);
876 		_exit(0); /* panic as well */
877 	}
878 
879 	BOOTTRACE("going to single user mode");
880 	shell = get_shell();
881 
882 	if ((pid = fork()) == 0) {
883 		/*
884 		 * Start the single user session.
885 		 */
886 		open_console();
887 
888 #ifdef SECURE
889 		/*
890 		 * Check the root password.
891 		 * We don't care if the console is 'on' by default;
892 		 * it's the only tty that can be 'off' and 'secure'.
893 		 */
894 		typ = getttynam("console");
895 		pp = getpwnam("root");
896 		if (typ && (typ->ty_status & TTY_SECURE) == 0 &&
897 		    pp && *pp->pw_passwd) {
898 			write_stderr(banner);
899 			for (;;) {
900 				clear = getpass("Password:");
901 				if (clear == NULL || *clear == '\0')
902 					_exit(0);
903 				password = crypt(clear, pp->pw_passwd);
904 				explicit_bzero(clear, _PASSWORD_LEN);
905 				if (password != NULL &&
906 				    strcmp(password, pp->pw_passwd) == 0)
907 					break;
908 				warning("single-user login failed\n");
909 			}
910 		}
911 		endttyent();
912 		endpwent();
913 #endif /* SECURE */
914 
915 #ifdef DEBUGSHELL
916 		{
917 			char *cp = altshell;
918 			int num;
919 
920 #define	SHREQUEST "Enter full pathname of shell or RETURN for "
921 			write_stderr(SHREQUEST);
922 			write_stderr(shell);
923 			write_stderr(": ");
924 			while ((num = read(STDIN_FILENO, cp, 1)) != -1 &&
925 			    num != 0 && *cp != '\n' && cp < &altshell[127])
926 				cp++;
927 			*cp = '\0';
928 			if (altshell[0] != '\0')
929 				shell = altshell;
930 		}
931 #endif /* DEBUGSHELL */
932 
933 		/*
934 		 * Unblock signals.
935 		 * We catch all the interesting ones,
936 		 * and those are reset to SIG_DFL on exec.
937 		 */
938 		sigemptyset(&mask);
939 		sigprocmask(SIG_SETMASK, &mask, NULL);
940 
941 		/*
942 		 * Fire off a shell.
943 		 * If the default one doesn't work, try the Bourne shell.
944 		 */
945 
946 		char name[] = "-sh";
947 
948 		argv[0] = name;
949 		argv[1] = NULL;
950 		execv(shell, argv);
951 		emergency("can't exec %s for single user: %m", shell);
952 		execv(_PATH_BSHELL, argv);
953 		emergency("can't exec %s for single user: %m", _PATH_BSHELL);
954 		sleep(STALL_TIMEOUT);
955 		_exit(1);
956 	}
957 
958 	if (pid == -1) {
959 		/*
960 		 * We are seriously hosed.  Do our best.
961 		 */
962 		emergency("can't fork single-user shell, trying again");
963 		while (waitpid(-1, (int *) 0, WNOHANG) > 0)
964 			continue;
965 		return (state_func_t) single_user;
966 	}
967 
968 	requested_transition = 0;
969 	do {
970 		if ((wpid = waitpid(-1, &status, WUNTRACED)) != -1)
971 			collect_child(wpid);
972 		if (wpid == -1) {
973 			if (errno == EINTR)
974 				continue;
975 			warning("wait for single-user shell failed: %m; restarting");
976 			return (state_func_t) single_user;
977 		}
978 		if (wpid == pid && WIFSTOPPED(status)) {
979 			warning("init: shell stopped, restarting\n");
980 			kill(pid, SIGCONT);
981 			wpid = -1;
982 		}
983 	} while (wpid != pid && !requested_transition);
984 
985 	if (requested_transition)
986 		return (state_func_t) requested_transition;
987 
988 	if (!WIFEXITED(status)) {
989 		if (WTERMSIG(status) == SIGKILL) {
990 			/*
991 			 *  reboot(8) killed shell?
992 			 */
993 			warning("single user shell terminated.");
994 			gettimeofday(&tv, NULL);
995 			tn = tv;
996 			tv.tv_sec += STALL_TIMEOUT;
997 			while (tv.tv_sec > tn.tv_sec || (tv.tv_sec ==
998 			    tn.tv_sec && tv.tv_usec > tn.tv_usec)) {
999 				sleep(1);
1000 				gettimeofday(&tn, NULL);
1001 			}
1002 			_exit(0);
1003 		} else {
1004 			warning("single user shell terminated, restarting");
1005 			return (state_func_t) single_user;
1006 		}
1007 	}
1008 
1009 	runcom_mode = FASTBOOT;
1010 	return (state_func_t) runcom;
1011 }
1012 
1013 /*
1014  * Run the system startup script.
1015  */
1016 static state_func_t
1017 runcom(void)
1018 {
1019 	state_func_t next_transition;
1020 
1021 	BOOTTRACE("/etc/rc starting...");
1022 	if ((next_transition = run_script(_PATH_RUNCOM)) != NULL)
1023 		return next_transition;
1024 	BOOTTRACE("/etc/rc finished");
1025 
1026 	runcom_mode = AUTOBOOT;		/* the default */
1027 	return (state_func_t) read_ttys;
1028 }
1029 
1030 static void
1031 execute_script(char *argv[])
1032 {
1033 	struct sigaction sa;
1034 	char* sh_argv[3 + SCRIPT_ARGV_SIZE];
1035 	const char *shell, *script;
1036 	int error, sh_argv_len, i;
1037 
1038 	bzero(&sa, sizeof(sa));
1039 	sigemptyset(&sa.sa_mask);
1040 	sa.sa_handler = SIG_IGN;
1041 	sigaction(SIGTSTP, &sa, NULL);
1042 	sigaction(SIGHUP, &sa, NULL);
1043 
1044 	open_console();
1045 
1046 	sigprocmask(SIG_SETMASK, &sa.sa_mask, NULL);
1047 #ifdef LOGIN_CAP
1048 	setprocresources(RESOURCE_RC);
1049 #endif
1050 
1051 	/*
1052 	 * Try to directly execute the script first.  If it
1053 	 * fails, try the old method of passing the script path
1054 	 * to sh(1).  Don't complain if it fails because of
1055 	 * the missing execute bit.
1056 	 */
1057 	script = argv[0];
1058 	error = access(script, X_OK);
1059 	if (error == 0) {
1060 		execv(script, argv);
1061 		warning("can't directly exec %s: %m", script);
1062 	} else if (errno != EACCES) {
1063 		warning("can't access %s: %m", script);
1064 	}
1065 
1066 	shell = get_shell();
1067 	sh_argv[0] = __DECONST(char*, shell);
1068 	sh_argv_len = 1;
1069 #ifdef SECURE
1070 	if (strcmp(shell, _PATH_BSHELL) == 0) {
1071 		sh_argv[1] = __DECONST(char*, "-o");
1072 		sh_argv[2] = __DECONST(char*, "verify");
1073 		sh_argv_len = 3;
1074 	}
1075 #endif
1076 	for (i = 0; i != SCRIPT_ARGV_SIZE; ++i)
1077 		sh_argv[i + sh_argv_len] = argv[i];
1078 	execv(shell, sh_argv);
1079 	stall("can't exec %s for %s: %m", shell, script);
1080 }
1081 
1082 /*
1083  * Execute binary, replacing init(8) as PID 1.
1084  */
1085 static void
1086 replace_init(char *path)
1087 {
1088 	char *argv[SCRIPT_ARGV_SIZE];
1089 
1090 	argv[0] = path;
1091 	argv[1] = NULL;
1092 
1093 	execute_script(argv);
1094 }
1095 
1096 /*
1097  * Run a shell script.
1098  * Returns 0 on success, otherwise the next transition to enter:
1099  *  - single_user if fork/execv/waitpid failed, or if the script
1100  *    terminated with a signal or exit code != 0.
1101  *  - death_single if a SIGTERM was delivered to init(8).
1102  */
1103 static state_func_t
1104 run_script(const char *script)
1105 {
1106 	pid_t pid, wpid;
1107 	int status;
1108 	char *argv[SCRIPT_ARGV_SIZE];
1109 	const char *shell;
1110 
1111 	shell = get_shell();
1112 
1113 	if ((pid = fork()) == 0) {
1114 
1115 		char _autoboot[] = "autoboot";
1116 
1117 		argv[0] = __DECONST(char *, script);
1118 		argv[1] = runcom_mode == AUTOBOOT ? _autoboot : NULL;
1119 		argv[2] = 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 (requested_transition == death_single ||
1142 		    requested_transition == reroot)
1143 			return (state_func_t) requested_transition;
1144 		if (wpid == -1) {
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: %m");
1191 	if ((session_db = dbopen(NULL, O_RDWR, 0, DB_HASH, NULL)) == NULL) {
1192 		emergency("session database open: %m");
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: %m", sp->se_process);
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: %m", sp->se_process);
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 static const char *
1602 get_current_state(void)
1603 {
1604 
1605 	if (current_state == single_user)
1606 		return ("single-user");
1607 	if (current_state == runcom)
1608 		return ("runcom");
1609 	if (current_state == read_ttys)
1610 		return ("read-ttys");
1611 	if (current_state == multi_user)
1612 		return ("multi-user");
1613 	if (current_state == clean_ttys)
1614 		return ("clean-ttys");
1615 	if (current_state == catatonia)
1616 		return ("catatonia");
1617 	if (current_state == death)
1618 		return ("death");
1619 	if (current_state == death_single)
1620 		return ("death-single");
1621 	return ("unknown");
1622 }
1623 
1624 static void
1625 boottrace_transition(int sig)
1626 {
1627 	const char *action;
1628 
1629 	switch (sig) {
1630 	case SIGUSR2:
1631 		action = "halt & poweroff";
1632 		break;
1633 	case SIGUSR1:
1634 		action = "halt";
1635 		break;
1636 	case SIGINT:
1637 		action = "reboot";
1638 		break;
1639 	case SIGWINCH:
1640 		action = "powercycle";
1641 		break;
1642 	case SIGTERM:
1643 		action = Reboot ? "reboot" : "single-user";
1644 		break;
1645 	default:
1646 		BOOTTRACE("signal %d from %s", sig, get_current_state());
1647 		return;
1648 	}
1649 
1650 	/* Trace the shutdown reason. */
1651 	SHUTTRACE("%s from %s", action, get_current_state());
1652 }
1653 
1654 /*
1655  * Catch a signal and request a state transition.
1656  */
1657 static void
1658 transition_handler(int sig)
1659 {
1660 
1661 	boottrace_transition(sig);
1662 	switch (sig) {
1663 	case SIGHUP:
1664 		if (current_state == read_ttys || current_state == multi_user ||
1665 		    current_state == clean_ttys || current_state == catatonia)
1666 			requested_transition = clean_ttys;
1667 		break;
1668 	case SIGUSR2:
1669 		howto = RB_POWEROFF;
1670 	case SIGUSR1:
1671 		howto |= RB_HALT;
1672 	case SIGWINCH:
1673 	case SIGINT:
1674 		if (sig == SIGWINCH)
1675 			howto |= RB_POWERCYCLE;
1676 		Reboot = true;
1677 	case SIGTERM:
1678 		if (current_state == read_ttys || current_state == multi_user ||
1679 		    current_state == clean_ttys || current_state == catatonia)
1680 			requested_transition = death;
1681 		else
1682 			requested_transition = death_single;
1683 		break;
1684 	case SIGTSTP:
1685 		if (current_state == runcom || current_state == read_ttys ||
1686 		    current_state == clean_ttys ||
1687 		    current_state == multi_user || current_state == catatonia)
1688 			requested_transition = catatonia;
1689 		break;
1690 	case SIGEMT:
1691 		requested_transition = reroot;
1692 		break;
1693 	default:
1694 		requested_transition = 0;
1695 		break;
1696 	}
1697 }
1698 
1699 /*
1700  * Take the system multiuser.
1701  */
1702 static state_func_t
1703 multi_user(void)
1704 {
1705 	static bool inmultiuser = false;
1706 	pid_t pid;
1707 	session_t *sp;
1708 
1709 	requested_transition = 0;
1710 
1711 	/*
1712 	 * If the administrator has not set the security level to -1
1713 	 * to indicate that the kernel should not run multiuser in secure
1714 	 * mode, and the run script has not set a higher level of security
1715 	 * than level 1, then put the kernel into secure mode.
1716 	 */
1717 	if (getsecuritylevel() == 0)
1718 		setsecuritylevel(1);
1719 
1720 	for (sp = sessions; sp; sp = sp->se_next) {
1721 		if (sp->se_process)
1722 			continue;
1723 		if (session_has_no_tty(sp))
1724 			continue;
1725 		if ((pid = start_getty(sp)) == -1) {
1726 			/* serious trouble */
1727 			requested_transition = clean_ttys;
1728 			break;
1729 		}
1730 		sp->se_process = pid;
1731 		sp->se_started = time((time_t *) 0);
1732 		add_session(sp);
1733 	}
1734 
1735 	if (requested_transition == 0 && !inmultiuser) {
1736 		inmultiuser = true;
1737 		/* This marks the change from boot-time tracing to run-time. */
1738 		RUNTRACE("multi-user start");
1739 	}
1740 	while (!requested_transition)
1741 		if ((pid = waitpid(-1, (int *) 0, 0)) != -1)
1742 			collect_child(pid);
1743 
1744 	return (state_func_t) requested_transition;
1745 }
1746 
1747 /*
1748  * This is an (n*2)+(n^2) algorithm.  We hope it isn't run often...
1749  */
1750 static state_func_t
1751 clean_ttys(void)
1752 {
1753 	session_t *sp, *sprev;
1754 	struct ttyent *typ;
1755 	int devlen;
1756 	char *old_getty, *old_window, *old_type;
1757 
1758 	/*
1759 	 * mark all sessions for death, (!SE_PRESENT)
1760 	 * as we find or create new ones they'll be marked as keepers,
1761 	 * we'll later nuke all the ones not found in /etc/ttys
1762 	 */
1763 	for (sp = sessions; sp != NULL; sp = sp->se_next)
1764 		sp->se_flags &= ~SE_PRESENT;
1765 
1766 	devlen = sizeof(_PATH_DEV) - 1;
1767 	while ((typ = getttyent()) != NULL) {
1768 		for (sprev = 0, sp = sessions; sp; sprev = sp, sp = sp->se_next)
1769 			if (strcmp(typ->ty_name, sp->se_device + devlen) == 0)
1770 				break;
1771 
1772 		if (sp) {
1773 			/* we want this one to live */
1774 			sp->se_flags |= SE_PRESENT;
1775 			if ((typ->ty_status & TTY_ON) == 0 ||
1776 			    typ->ty_getty == 0) {
1777 				sp->se_flags |= SE_SHUTDOWN;
1778 				kill(sp->se_process, SIGHUP);
1779 				continue;
1780 			}
1781 			sp->se_flags &= ~SE_SHUTDOWN;
1782 			old_getty = sp->se_getty ? strdup(sp->se_getty) : 0;
1783 			old_window = sp->se_window ? strdup(sp->se_window) : 0;
1784 			old_type = sp->se_type ? strdup(sp->se_type) : 0;
1785 			if (setupargv(sp, typ) == 0) {
1786 				warning("can't parse getty for port %s",
1787 					sp->se_device);
1788 				sp->se_flags |= SE_SHUTDOWN;
1789 				kill(sp->se_process, SIGHUP);
1790 			}
1791 			else if (   !old_getty
1792 				 || (!old_type && sp->se_type)
1793 				 || (old_type && !sp->se_type)
1794 				 || (!old_window && sp->se_window)
1795 				 || (old_window && !sp->se_window)
1796 				 || (strcmp(old_getty, sp->se_getty) != 0)
1797 				 || (old_window && strcmp(old_window, sp->se_window) != 0)
1798 				 || (old_type && strcmp(old_type, sp->se_type) != 0)
1799 				) {
1800 				/* Don't set SE_SHUTDOWN here */
1801 				sp->se_nspace = 0;
1802 				sp->se_started = 0;
1803 				kill(sp->se_process, SIGHUP);
1804 			}
1805 			if (old_getty)
1806 				free(old_getty);
1807 			if (old_window)
1808 				free(old_window);
1809 			if (old_type)
1810 				free(old_type);
1811 			continue;
1812 		}
1813 
1814 		new_session(sprev, typ);
1815 	}
1816 
1817 	endttyent();
1818 
1819 	/*
1820 	 * sweep through and kill all deleted sessions
1821 	 * ones who's /etc/ttys line was deleted (SE_PRESENT unset)
1822 	 */
1823 	for (sp = sessions; sp != NULL; sp = sp->se_next) {
1824 		if ((sp->se_flags & SE_PRESENT) == 0) {
1825 			sp->se_flags |= SE_SHUTDOWN;
1826 			kill(sp->se_process, SIGHUP);
1827 		}
1828 	}
1829 
1830 	return (state_func_t) multi_user;
1831 }
1832 
1833 /*
1834  * Block further logins.
1835  */
1836 static state_func_t
1837 catatonia(void)
1838 {
1839 	session_t *sp;
1840 
1841 	for (sp = sessions; sp; sp = sp->se_next)
1842 		sp->se_flags |= SE_SHUTDOWN;
1843 
1844 	return (state_func_t) multi_user;
1845 }
1846 
1847 /*
1848  * Note SIGALRM.
1849  */
1850 static void
1851 alrm_handler(int sig)
1852 {
1853 
1854 	(void)sig;
1855 	clang = true;
1856 }
1857 
1858 /*
1859  * Bring the system down to single user.
1860  */
1861 static state_func_t
1862 death(void)
1863 {
1864 	int block, blocked;
1865 	size_t len;
1866 
1867 	/* Temporarily block suspend. */
1868 	len = sizeof(blocked);
1869 	block = 1;
1870 	if (sysctlbyname("kern.suspend_blocked", &blocked, &len,
1871 	    &block, sizeof(block)) == -1)
1872 		blocked = 0;
1873 
1874 	/*
1875 	 * Also revoke the TTY here.  Because runshutdown() may reopen
1876 	 * the TTY whose getty we're killing here, there is no guarantee
1877 	 * runshutdown() will perform the initial open() call, causing
1878 	 * the terminal attributes to be misconfigured.
1879 	 */
1880 	revoke_ttys();
1881 
1882 	/* Try to run the rc.shutdown script within a period of time */
1883 	runshutdown();
1884 
1885 	/* Unblock suspend if we blocked it. */
1886 	if (!blocked)
1887 		sysctlbyname("kern.suspend_blocked", NULL, NULL,
1888 		    &blocked, sizeof(blocked));
1889 
1890 	return (state_func_t) death_single;
1891 }
1892 
1893 /*
1894  * Do what is necessary to reinitialize single user mode or reboot
1895  * from an incomplete state.
1896  */
1897 static state_func_t
1898 death_single(void)
1899 {
1900 	int i;
1901 	pid_t pid;
1902 	static const int death_sigs[2] = { SIGTERM, SIGKILL };
1903 
1904 	revoke(_PATH_CONSOLE);
1905 
1906 	BOOTTRACE("start killing user processes");
1907 	for (i = 0; i < 2; ++i) {
1908 		if (kill(-1, death_sigs[i]) == -1 && errno == ESRCH)
1909 			return (state_func_t) single_user;
1910 
1911 		clang = false;
1912 		alarm(DEATH_WATCH);
1913 		do
1914 			if ((pid = waitpid(-1, (int *)0, 0)) != -1)
1915 				collect_child(pid);
1916 		while (!clang && errno != ECHILD);
1917 
1918 		if (errno == ECHILD)
1919 			return (state_func_t) single_user;
1920 	}
1921 
1922 	warning("some processes would not die; ps axl advised");
1923 
1924 	return (state_func_t) single_user;
1925 }
1926 
1927 static void
1928 revoke_ttys(void)
1929 {
1930 	session_t *sp;
1931 
1932 	for (sp = sessions; sp; sp = sp->se_next) {
1933 		sp->se_flags |= SE_SHUTDOWN;
1934 		kill(sp->se_process, SIGHUP);
1935 		revoke(sp->se_device);
1936 	}
1937 }
1938 
1939 /*
1940  * Run the system shutdown script.
1941  *
1942  * Exit codes:      XXX I should document more
1943  * -2       shutdown script terminated abnormally
1944  * -1       fatal error - can't run script
1945  * 0        good.
1946  * >0       some error (exit code)
1947  */
1948 static int
1949 runshutdown(void)
1950 {
1951 	pid_t pid, wpid;
1952 	int status;
1953 	int shutdowntimeout;
1954 	size_t len;
1955 	char *argv[SCRIPT_ARGV_SIZE];
1956 	struct stat sb;
1957 
1958 	BOOTTRACE("init(8): start rc.shutdown");
1959 
1960 	/*
1961 	 * rc.shutdown is optional, so to prevent any unnecessary
1962 	 * complaints from the shell we simply don't run it if the
1963 	 * file does not exist. If the stat() here fails for other
1964 	 * reasons, we'll let the shell complain.
1965 	 */
1966 	if (stat(_PATH_RUNDOWN, &sb) == -1 && errno == ENOENT)
1967 		return 0;
1968 
1969 	if ((pid = fork()) == 0) {
1970 		char _reboot[]	= "reboot";
1971 		char _single[]	= "single";
1972 		char _path_rundown[] = _PATH_RUNDOWN;
1973 
1974 		argv[0] = _path_rundown;
1975 		argv[1] = Reboot ? _reboot : _single;
1976 		argv[2] = NULL;
1977 
1978 		execute_script(argv);
1979 		_exit(1);	/* force single user mode */
1980 	}
1981 
1982 	if (pid == -1) {
1983 		emergency("can't fork for %s: %m", _PATH_RUNDOWN);
1984 		while (waitpid(-1, (int *) 0, WNOHANG) > 0)
1985 			continue;
1986 		sleep(STALL_TIMEOUT);
1987 		return -1;
1988 	}
1989 
1990 	len = sizeof(shutdowntimeout);
1991 	if (sysctlbyname("kern.init_shutdown_timeout", &shutdowntimeout, &len,
1992 	    NULL, 0) == -1 || shutdowntimeout < 2)
1993 		shutdowntimeout = DEATH_SCRIPT;
1994 	alarm(shutdowntimeout);
1995 	clang = false;
1996 	/*
1997 	 * Copied from single_user().  This is a bit paranoid.
1998 	 * Use the same ALRM handler.
1999 	 */
2000 	do {
2001 		if ((wpid = waitpid(-1, &status, WUNTRACED)) != -1)
2002 			collect_child(wpid);
2003 		if (clang) {
2004 			/* we were waiting for the sub-shell */
2005 			kill(wpid, SIGTERM);
2006 			warning("timeout expired for %s: %m; going to "
2007 			    "single user mode", _PATH_RUNDOWN);
2008 			BOOTTRACE("rc.shutdown's %d sec timeout expired",
2009 				  shutdowntimeout);
2010 			return -1;
2011 		}
2012 		if (wpid == -1) {
2013 			if (errno == EINTR)
2014 				continue;
2015 			warning("wait for %s failed: %m; going to "
2016 			    "single user mode", _PATH_RUNDOWN);
2017 			return -1;
2018 		}
2019 		if (wpid == pid && WIFSTOPPED(status)) {
2020 			warning("init: %s stopped, restarting\n",
2021 			    _PATH_RUNDOWN);
2022 			kill(pid, SIGCONT);
2023 			wpid = -1;
2024 		}
2025 	} while (wpid != pid && !clang);
2026 
2027 	/* Turn off the alarm */
2028 	alarm(0);
2029 
2030 	if (WIFSIGNALED(status) && WTERMSIG(status) == SIGTERM &&
2031 	    requested_transition == catatonia) {
2032 		/*
2033 		 * /etc/rc.shutdown executed /sbin/reboot;
2034 		 * wait for the end quietly
2035 		 */
2036 		sigset_t s;
2037 
2038 		sigfillset(&s);
2039 		for (;;)
2040 			sigsuspend(&s);
2041 	}
2042 
2043 	if (!WIFEXITED(status)) {
2044 		warning("%s terminated abnormally, going to "
2045 		    "single user mode", _PATH_RUNDOWN);
2046 		return -2;
2047 	}
2048 
2049 	if ((status = WEXITSTATUS(status)) != 0)
2050 		warning("%s returned status %d", _PATH_RUNDOWN, status);
2051 
2052 	return status;
2053 }
2054 
2055 static char *
2056 strk(char *p)
2057 {
2058 	static char *t;
2059 	char *q;
2060 	int c;
2061 
2062 	if (p)
2063 		t = p;
2064 	if (!t)
2065 		return 0;
2066 
2067 	c = *t;
2068 	while (c == ' ' || c == '\t' )
2069 		c = *++t;
2070 	if (!c) {
2071 		t = 0;
2072 		return 0;
2073 	}
2074 	q = t;
2075 	if (c == '\'') {
2076 		c = *++t;
2077 		q = t;
2078 		while (c && c != '\'')
2079 			c = *++t;
2080 		if (!c)  /* unterminated string */
2081 			q = t = 0;
2082 		else
2083 			*t++ = 0;
2084 	} else {
2085 		while (c && c != ' ' && c != '\t' )
2086 			c = *++t;
2087 		*t++ = 0;
2088 		if (!c)
2089 			t = 0;
2090 	}
2091 	return q;
2092 }
2093 
2094 #ifdef LOGIN_CAP
2095 static void
2096 setprocresources(const char *cname)
2097 {
2098 	login_cap_t *lc;
2099 	if ((lc = login_getclassbyname(cname, NULL)) != NULL) {
2100 		setusercontext(lc, (struct passwd*)NULL, 0,
2101 		    LOGIN_SETENV |
2102 		    LOGIN_SETPRIORITY | LOGIN_SETRESOURCES |
2103 		    LOGIN_SETLOGINCLASS | LOGIN_SETCPUMASK);
2104 		login_close(lc);
2105 	}
2106 }
2107 #endif
2108 
2109 /*
2110  * Run /etc/rc.final to execute scripts after all user processes have been
2111  * terminated.
2112  */
2113 static void
2114 runfinal(void)
2115 {
2116 	struct stat sb;
2117 	pid_t other_pid, pid;
2118 	sigset_t mask;
2119 
2120 	/* Avoid any surprises. */
2121 	alarm(0);
2122 
2123 	/* rc.final is optional. */
2124 	if (stat(_PATH_RUNFINAL, &sb) == -1 && errno == ENOENT)
2125 		return;
2126 	if (access(_PATH_RUNFINAL, X_OK) != 0) {
2127 		warning("%s exists, but not executable", _PATH_RUNFINAL);
2128 		return;
2129 	}
2130 
2131 	pid = fork();
2132 	if (pid == 0) {
2133 		/*
2134 		 * Reopen stdin/stdout/stderr so that scripts can write to
2135 		 * console.
2136 		 */
2137 		close(0);
2138 		open(_PATH_DEVNULL, O_RDONLY);
2139 		close(1);
2140 		close(2);
2141 		open_console();
2142 		dup2(1, 2);
2143 		sigemptyset(&mask);
2144 		sigprocmask(SIG_SETMASK, &mask, NULL);
2145 		signal(SIGCHLD, SIG_DFL);
2146 		execl(_PATH_RUNFINAL, _PATH_RUNFINAL, NULL);
2147 		perror("execl(" _PATH_RUNFINAL ") failed");
2148 		exit(1);
2149 	}
2150 
2151 	/* Wait for rc.final script to exit */
2152 	while ((other_pid = waitpid(-1, NULL, 0)) != pid && other_pid > 0) {
2153 		continue;
2154 	}
2155 }
2156