xref: /illumos-gate/usr/src/cmd/init/init.c (revision ddb365bfc9e868ad24ccdcb0dc91af18b10df082)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2021 OmniOS Community Edition (OmniOSce) Association.
24  * Copyright 2020 Oxide Computer Company
25  * Copyright (c) 2013 Gary Mills
26  *
27  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
28  */
29 
30 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
31 /*	  All Rights Reserved	*/
32 
33 /*
34  * University Copyright- Copyright (c) 1982, 1986, 1988
35  * The Regents of the University of California
36  * All Rights Reserved
37  *
38  * University Acknowledgment- Portions of this document are derived from
39  * software developed by the University of California, Berkeley, and its
40  * contributors.
41  */
42 
43 /*
44  * init(8) is the general process spawning program.  Its primary job is to
45  * start and restart svc.startd for smf(7).  For backwards-compatibility it also
46  * spawns and respawns processes according to /etc/inittab and the current
47  * run-level.  It reads /etc/default/inittab for general configuration.
48  *
49  * To change run-levels the system administrator runs init from the command
50  * line with a level name.  init signals svc.startd via libscf and directs the
51  * zone's init (pid 1 in the global zone) what to do by sending it a signal;
52  * these signal numbers are commonly refered to in the code as 'states'.  Valid
53  * run-levels are [sS0123456].  Additionally, init can be given directives
54  * [qQabc], which indicate actions to be taken pertaining to /etc/inittab.
55  *
56  * When init processes inittab entries, it finds processes that are to be
57  * spawned at various run-levels.  inittab contains the set of the levels for
58  * which each inittab entry is valid.
59  *
60  * State File and Restartability
61  *   Premature exit by init(8) is handled as a special case by the kernel:
62  *   init(8) will be immediately re-executed, retaining its original PID.  (PID
63  *   1 in the global zone.)  To track the processes it has previously spawned,
64  *   as well as other mutable state, init(8) regularly updates a state file
65  *   such that its subsequent invocations have knowledge of its various
66  *   dependent processes and duties.
67  *
68  * Process Contracts
69  *   We start svc.startd(8) in a contract and transfer inherited contracts when
70  *   restarting it.  Everything else is started using the legacy contract
71  *   template, and the created contracts are abandoned when they become empty.
72  *
73  * utmpx Entry Handling
74  *   Because init(8) no longer governs the startup process, its knowledge of
75  *   when utmpx becomes writable is indirect.  However, spawned processes
76  *   expect to be constructed with valid utmpx entries.  As a result, attempts
77  *   to write normal entries will be retried until successful.
78  *
79  * Maintenance Mode
80  *   In certain failure scenarios, init(8) will enter a maintenance mode, in
81  *   which it invokes sulogin(8) to allow the operator an opportunity to
82  *   repair the system.  Normally, this operation is performed as a
83  *   fork(2)-exec(2)-waitpid(3C) sequence with the parent waiting for repair or
84  *   diagnosis to be completed.  In the cases that fork(2) requests themselves
85  *   fail, init(8) will directly execute sulogin(8), and allow the kernel to
86  *   restart init(8) on exit from the operator session.
87  *
88  *   One scenario where init(8) enters its maintenance mode is when
89  *   svc.startd(8) begins to fail rapidly, defined as when the average time
90  *   between recent failures drops below a given threshold.
91  */
92 
93 #include <sys/contract/process.h>
94 #include <sys/ctfs.h>
95 #include <sys/stat.h>
96 #include <sys/statvfs.h>
97 #include <sys/stropts.h>
98 #include <sys/systeminfo.h>
99 #include <sys/time.h>
100 #include <sys/termios.h>
101 #include <sys/tty.h>
102 #include <sys/types.h>
103 #include <sys/utsname.h>
104 #include <sys/bootbanner.h>
105 
106 #include <bsm/adt_event.h>
107 #include <bsm/libbsm.h>
108 #include <security/pam_appl.h>
109 
110 #include <assert.h>
111 #include <ctype.h>
112 #include <definit.h>
113 #include <dirent.h>
114 #include <errno.h>
115 #include <fcntl.h>
116 #include <libcontract.h>
117 #include <libcontract_priv.h>
118 #include <libintl.h>
119 #include <libscf.h>
120 #include <libscf_priv.h>
121 #include <poll.h>
122 #include <procfs.h>
123 #include <signal.h>
124 #include <stdarg.h>
125 #include <stdio.h>
126 #include <stdio_ext.h>
127 #include <stdlib.h>
128 #include <string.h>
129 #include <strings.h>
130 #include <syslog.h>
131 #include <time.h>
132 #include <ulimit.h>
133 #include <unistd.h>
134 #include <utmpx.h>
135 #include <wait.h>
136 #include <zone.h>
137 #include <ucontext.h>
138 
139 #undef	sleep
140 
141 #define	fioctl(p, sptr, cmd)	ioctl(fileno(p), sptr, cmd)
142 #define	min(a, b)		(((a) < (b)) ? (a) : (b))
143 
144 #define	TRUE	1
145 #define	FALSE	0
146 #define	FAILURE	-1
147 
148 #define	UT_USER_SZ	32	/* Size of a utmpx ut_user field */
149 #define	UT_LINE_SZ	32	/* Size of a utmpx ut_line field */
150 
151 /*
152  * SLEEPTIME	The number of seconds "init" sleeps between wakeups if
153  *		nothing else requires this "init" wakeup.
154  */
155 #define	SLEEPTIME	(5 * 60)
156 
157 /*
158  * MAXCMDL	The maximum length of a command string in inittab.
159  */
160 #define	MAXCMDL	512
161 
162 /*
163  * EXEC		The length of the prefix string added to all comamnds
164  *		found in inittab.
165  */
166 #define	EXEC	(sizeof ("exec ") - 1)
167 
168 /*
169  * TWARN	The amount of time between warning signal, SIGTERM,
170  *		and the fatal kill signal, SIGKILL.
171  */
172 #define	TWARN	5
173 
174 #define	id_eq(x, y)	((x[0] == y[0] && x[1] == y[1] && x[2] == y[2] &&\
175 			x[3] == y[3]) ? TRUE : FALSE)
176 
177 /*
178  * The kernel's default umask is 022 these days; since some processes inherit
179  * their umask from init, init will set it from CMASK in /etc/default/init.
180  * init gets the default umask from the kernel, it sets it to 022 whenever
181  * it wants to create a file and reverts to CMASK afterwards.
182  */
183 
184 static int cmask;
185 
186 /*
187  * The following definitions, concluding with the 'lvls' array, provide a
188  * common mapping between level-name (like 'S'), signal number (state),
189  * run-level mask, and specific properties associated with a run-level.
190  * This array should be accessed using the routines lvlname_to_state(),
191  * lvlname_to_mask(), state_to_mask(), and state_to_flags().
192  */
193 
194 /*
195  * Correspondence of signals to init actions.
196  */
197 #define	LVLQ		SIGHUP
198 #define	LVL0		SIGINT
199 #define	LVL1		SIGQUIT
200 #define	LVL2		SIGILL
201 #define	LVL3		SIGTRAP
202 #define	LVL4		SIGIOT
203 #define	LVL5		SIGEMT
204 #define	LVL6		SIGFPE
205 #define	SINGLE_USER	SIGBUS
206 #define	LVLa		SIGSEGV
207 #define	LVLb		SIGSYS
208 #define	LVLc		SIGPIPE
209 
210 /*
211  * Bit Mask for each level.  Used to determine legal levels.
212  */
213 #define	MASK0	0x0001
214 #define	MASK1	0x0002
215 #define	MASK2	0x0004
216 #define	MASK3	0x0008
217 #define	MASK4	0x0010
218 #define	MASK5	0x0020
219 #define	MASK6	0x0040
220 #define	MASKSU	0x0080
221 #define	MASKa	0x0100
222 #define	MASKb	0x0200
223 #define	MASKc	0x0400
224 
225 #define	MASK_NUMERIC (MASK0 | MASK1 | MASK2 | MASK3 | MASK4 | MASK5 | MASK6)
226 #define	MASK_abc (MASKa | MASKb | MASKc)
227 
228 /*
229  * Flags to indicate properties of various states.
230  */
231 #define	LSEL_RUNLEVEL	0x0001	/* runlevels you can transition to */
232 
233 typedef struct lvl {
234 	int	lvl_state;
235 	int	lvl_mask;
236 	char	lvl_name;
237 	int	lvl_flags;
238 } lvl_t;
239 
240 static lvl_t lvls[] = {
241 	{ LVLQ,		0,	'Q', 0					},
242 	{ LVLQ,		0,	'q', 0					},
243 	{ LVL0,		MASK0,	'0', LSEL_RUNLEVEL			},
244 	{ LVL1,		MASK1,	'1', LSEL_RUNLEVEL			},
245 	{ LVL2,		MASK2,	'2', LSEL_RUNLEVEL			},
246 	{ LVL3,		MASK3,	'3', LSEL_RUNLEVEL			},
247 	{ LVL4,		MASK4,	'4', LSEL_RUNLEVEL			},
248 	{ LVL5,		MASK5,	'5', LSEL_RUNLEVEL			},
249 	{ LVL6,		MASK6,	'6', LSEL_RUNLEVEL			},
250 	{ SINGLE_USER,	MASKSU, 'S', LSEL_RUNLEVEL			},
251 	{ SINGLE_USER,	MASKSU, 's', LSEL_RUNLEVEL			},
252 	{ LVLa,		MASKa,	'a', 0					},
253 	{ LVLb,		MASKb,	'b', 0					},
254 	{ LVLc,		MASKc,	'c', 0					}
255 };
256 
257 #define	LVL_NELEMS (sizeof (lvls) / sizeof (lvl_t))
258 
259 /*
260  * Legal action field values.
261  */
262 #define	OFF		0	/* Kill process if on, else ignore */
263 #define	RESPAWN		1	/* Continuously restart process when it dies */
264 #define	ONDEMAND	RESPAWN	/* Respawn for a, b, c type processes */
265 #define	ONCE		2	/* Start process, do not respawn when dead */
266 #define	WAIT		3	/* Perform once and wait to complete */
267 #define	BOOT		4	/* Start at boot time only */
268 #define	BOOTWAIT	5	/* Start at boot time and wait to complete */
269 #define	POWERFAIL	6	/* Start on powerfail */
270 #define	POWERWAIT	7	/* Start and wait for complete on powerfail */
271 #define	INITDEFAULT	8	/* Default level "init" should start at */
272 #define	SYSINIT		9	/* Actions performed before init speaks */
273 
274 #define	M_OFF		0001
275 #define	M_RESPAWN	0002
276 #define	M_ONDEMAND	M_RESPAWN
277 #define	M_ONCE		0004
278 #define	M_WAIT		0010
279 #define	M_BOOT		0020
280 #define	M_BOOTWAIT	0040
281 #define	M_PF		0100
282 #define	M_PWAIT		0200
283 #define	M_INITDEFAULT	0400
284 #define	M_SYSINIT	01000
285 
286 /* States for the inittab parser in getcmd(). */
287 #define	ID	1
288 #define	LEVELS	2
289 #define	ACTION	3
290 #define	COMMAND	4
291 #define	COMMENT	5
292 
293 /*
294  * inittab entry id constants
295  */
296 #define	INITTAB_ENTRY_ID_SIZE 4
297 #define	INITTAB_ENTRY_ID_STR_FORMAT "%.4s"	/* if INITTAB_ENTRY_ID_SIZE */
298 						/* changes, this should */
299 						/* change accordingly */
300 
301 /*
302  * Init can be in any of three main states, "normal" mode where it is
303  * processing entries for the lines file in a normal fashion, "boot" mode,
304  * where it is only interested in the boot actions, and "powerfail" mode,
305  * where it is only interested in powerfail related actions. The following
306  * masks declare the legal actions for each mode.
307  */
308 #define	NORMAL_MODES	(M_OFF | M_RESPAWN | M_ONCE | M_WAIT)
309 #define	BOOT_MODES	(M_BOOT | M_BOOTWAIT)
310 #define	PF_MODES	(M_PF | M_PWAIT)
311 
312 struct PROC_TABLE {
313 	char	p_id[INITTAB_ENTRY_ID_SIZE];	/* Four letter unique id of */
314 						/* process */
315 	pid_t	p_pid;		/* Process id */
316 	short	p_count;	/* How many respawns of this command in */
317 				/*   the current series */
318 	long	p_time;		/* Start time for a series of respawns */
319 	short	p_flags;
320 	short	p_exit;		/* Exit status of a process which died */
321 };
322 
323 /*
324  * Flags for the "p_flags" word of a PROC_TABLE entry:
325  *
326  *	OCCUPIED	This slot in init's proc table is in use.
327  *
328  *	LIVING		Process is alive.
329  *
330  *	NOCLEANUP	efork() is not allowed to cleanup this entry even
331  *			if process is dead.
332  *
333  *	NAMED		This process has a name, i.e. came from inittab.
334  *
335  *	DEMANDREQUEST	Process started by a "telinit [abc]" command.  Processes
336  *			formed this way are respawnable and immune to level
337  *			changes as long as their entry exists in inittab.
338  *
339  *	TOUCHED		Flag used by remv() to determine whether it has looked
340  *			at an entry while checking for processes to be killed.
341  *
342  *	WARNED		Flag used by remv() to mark processes that have been
343  *			sent the SIGTERM signal.  If they don't die in 5
344  *			seconds, they are sent the SIGKILL signal.
345  *
346  *	KILLED		Flag used by remv() to mark procs that have been sent
347  *			the SIGTERM and SIGKILL signals.
348  *
349  *	PF_MASK		Bitwise or of legal flags, for sanity checking.
350  */
351 #define	OCCUPIED	01
352 #define	LIVING		02
353 #define	NOCLEANUP	04
354 #define	NAMED		010
355 #define	DEMANDREQUEST	020
356 #define	TOUCHED		040
357 #define	WARNED		0100
358 #define	KILLED		0200
359 #define	PF_MASK		0377
360 
361 /*
362  * Respawn limits for processes that are to be respawned:
363  *
364  *	SPAWN_INTERVAL	The number of seconds over which "init" will try to
365  *			respawn a process SPAWN_LIMIT times before it gets mad.
366  *
367  *	SPAWN_LIMIT	The number of respawns "init" will attempt in
368  *			SPAWN_INTERVAL seconds before it generates an
369  *			error message and inhibits further tries for
370  *			INHIBIT seconds.
371  *
372  *	INHIBIT		The number of seconds "init" ignores an entry it had
373  *			trouble spawning unless a "telinit Q" is received.
374  */
375 
376 #define	SPAWN_INTERVAL	(2*60)
377 #define	SPAWN_LIMIT	10
378 #define	INHIBIT		(5*60)
379 
380 /*
381  * The maximum number of decimal digits for an id_t.  (ceil(log10 (max_id)))
382  */
383 #define	ID_MAX_STR_LEN	10
384 
385 #define	NULLPROC	((struct PROC_TABLE *)(0))
386 #define	NO_ROOM		((struct PROC_TABLE *)(FAILURE))
387 
388 struct CMD_LINE {
389 	char c_id[INITTAB_ENTRY_ID_SIZE];	/* Four letter unique id of */
390 						/* process to be affected by */
391 						/* action */
392 	short c_levels;	/* Mask of legal levels for process */
393 	short c_action;	/* Mask for type of action required */
394 	char *c_command; /* Pointer to init command */
395 };
396 
397 struct	pidrec {
398 	int	pd_type;	/* Command type */
399 	pid_t	pd_pid;		/* pid to add or remove */
400 };
401 
402 /*
403  * pd_type's
404  */
405 #define	ADDPID	1
406 #define	REMPID	2
407 
408 static struct	pidlist {
409 	pid_t	pl_pid;		/* pid to watch for */
410 	int	pl_dflag;	/* Flag indicating SIGCLD from this pid */
411 	short	pl_exit;	/* Exit status of proc */
412 	struct	pidlist	*pl_next; /* Next in list */
413 } *Plhead, *Plfree;
414 
415 /*
416  * The following structure contains a set of modes for /dev/syscon
417  * and should match the default contents of /etc/ioctl.syscon.
418  */
419 static struct termios	dflt_termios = {
420 	.c_iflag = BRKINT|ICRNL|IXON|IMAXBEL,
421 	.c_oflag = OPOST|ONLCR|TAB3,
422 	.c_cflag = CS8|CREAD|B9600,
423 	.c_lflag = ISIG|ICANON|ECHO|ECHOE|ECHOK|ECHOCTL|ECHOKE|IEXTEN,
424 	.c_cc = { CINTR, CQUIT, CERASE, CKILL, CEOF, 0, 0, 0,
425 	    CSTART, CSTOP, CSWTCH, CDSUSP, CRPRNT, CFLUSH, CWERASE, CLNEXT,
426 	    CSTATUS, CERASE2, 0
427 	}
428 };
429 
430 static struct termios	stored_syscon_termios;
431 static int		write_ioctl = 0;	/* Rewrite /etc/ioctl.syscon */
432 
433 static union WAKEUP {
434 	struct WAKEFLAGS {
435 		unsigned w_usersignal : 1;	/* User sent signal to "init" */
436 		unsigned w_childdeath : 1;	/* An "init" child died */
437 		unsigned w_powerhit : 1;	/* OS experienced powerfail */
438 	}	w_flags;
439 	int w_mask;
440 } wakeup;
441 
442 
443 struct init_state {
444 	int			ist_runlevel;
445 	int			ist_num_proc;
446 	int			ist_utmpx_ok;
447 	struct PROC_TABLE	ist_proc_table[1];
448 };
449 
450 #define	cur_state	(g_state->ist_runlevel)
451 #define	num_proc	(g_state->ist_num_proc)
452 #define	proc_table	(g_state->ist_proc_table)
453 #define	utmpx_ok	(g_state->ist_utmpx_ok)
454 
455 /* Contract cookies. */
456 #define	ORDINARY_COOKIE		0
457 #define	STARTD_COOKIE		1
458 
459 
460 #ifndef NDEBUG
461 #define	bad_error(func, err)	{					\
462 	(void) fprintf(stderr, "%s:%d: %s() failed with unexpected "	\
463 	    "error %d.  Aborting.\n", __FILE__, __LINE__, (func), (err)); \
464 	abort();							\
465 }
466 #else
467 #define	bad_error(func, err)	abort()
468 #endif
469 
470 
471 /*
472  * Useful file and device names.
473  */
474 static char *CONSOLE	  = "/dev/console";	/* Real system console */
475 static char *INITPIPE_DIR = "/var/run";
476 static char *INITPIPE	  = "/var/run/initpipe";
477 
478 #define	INIT_STATE_DIR "/etc/svc/volatile"
479 static const char * const init_state_file = INIT_STATE_DIR "/init.state";
480 static const char * const init_next_state_file =
481 	INIT_STATE_DIR "/init-next.state";
482 
483 static const int init_num_proc = 20;	/* Initial size of process table. */
484 
485 static char *UTMPX	 = UTMPX_FILE;		/* Snapshot record file */
486 static char *WTMPX	 = WTMPX_FILE;		/* Long term record file */
487 static char *INITTAB	 = "/etc/inittab";	/* Script file for "init" */
488 static char *SYSTTY	 = "/dev/systty";	/* System Console */
489 static char *SYSCON	 = "/dev/syscon";	/* Virtual System console */
490 static char *IOCTLSYSCON = "/etc/ioctl.syscon";	/* Last syscon modes */
491 static char *ENVFILE	 = DEFINIT_DEFAULT_FILE; /* Default env. */
492 static char *SU	= "/etc/sulogin";	/* Super-user program for single user */
493 static char *SH	= "/sbin/sh";		/* Standard shell */
494 
495 /*
496  * Default Path.  /sbin is included in path only during sysinit phase
497  */
498 #define	DEF_PATH	"PATH=/usr/sbin:/usr/bin"
499 #define	INIT_PATH	"PATH=/sbin:/usr/sbin:/usr/bin"
500 
501 static int	prior_state;
502 static int	prev_state;	/* State "init" was in last time it woke */
503 static int	new_state;	/* State user wants "init" to go to. */
504 static int	lvlq_received;	/* Explicit request to examine state */
505 static int	op_modes = BOOT_MODES; /* Current state of "init" */
506 static int	Gchild = 0;	/* Flag to indicate "godchild" died, set in */
507 				/*   childeath() and cleared in cleanaux() */
508 static int	Pfd = -1;	/* fd to receive pids thru */
509 static unsigned int	spawncnt, pausecnt;
510 static int	rsflag;		/* Set if a respawn has taken place */
511 static volatile int time_up;	/* Flag set to TRUE by the alarm interrupt */
512 				/* routine each time an alarm interrupt */
513 				/* takes place. */
514 static int	sflg = 0;	/* Set if we were booted -s to single user */
515 static int	rflg = 0;	/* Set if booted -r, reconfigure devices */
516 static int	bflg = 0;	/* Set if booted -b, don't run rc scripts */
517 static pid_t	init_pid;	/* PID of "one true" init for current zone */
518 
519 static struct init_state *g_state = NULL;
520 static size_t	g_state_sz;
521 static int	booting = 1;	/* Set while we're booting. */
522 
523 /*
524  * Array for default global environment.
525  */
526 #define	MAXENVENT	24	/* Max number of default env variables + 1 */
527 				/* init can use three itself, so this leaves */
528 				/* 20 for the administrator in ENVFILE. */
529 static char	*glob_envp[MAXENVENT];	/* Array of environment strings */
530 static int	glob_envn;		/* Number of environment strings */
531 
532 
533 static struct pollfd	poll_fds[1];
534 static int		poll_nfds = 0;	/* poll_fds is uninitialized */
535 
536 /*
537  * Contracts constants
538  */
539 #define	SVC_INIT_PREFIX "init:/"
540 #define	SVC_AUX_SIZE (INITTAB_ENTRY_ID_SIZE + 1)
541 #define	SVC_FMRI_SIZE (sizeof (SVC_INIT_PREFIX) + INITTAB_ENTRY_ID_SIZE)
542 
543 static int	legacy_tmpl = -1;	/* fd for legacy contract template */
544 static int	startd_tmpl = -1;	/* fd for svc.startd's template */
545 static char	startd_svc_aux[SVC_AUX_SIZE];
546 
547 static char	startd_cline[256] = "";	/* svc.startd's command line */
548 static int	do_restart_startd = 1;	/* Whether to restart svc.startd. */
549 static char	*smf_options = NULL;	/* Options to give to startd. */
550 static int	smf_debug = 0;		/* Messages for debugging smf(7) */
551 static time_t	init_boot_time;		/* Substitute for kernel boot time. */
552 
553 #define	NSTARTD_FAILURE_TIMES	3		/* trigger after 3 failures */
554 #define	STARTD_FAILURE_RATE_NS	5000000000LL	/* 1 failure/5 seconds */
555 
556 static hrtime_t	startd_failure_time[NSTARTD_FAILURE_TIMES];
557 static uint_t	startd_failure_index;
558 
559 
560 static char	*prog_name(char *);
561 static int	state_to_mask(int);
562 static int	lvlname_to_mask(char, int *);
563 static void	lscf_set_runlevel(char);
564 static int	state_to_flags(int);
565 static char	state_to_name(int);
566 static int	lvlname_to_state(char);
567 static int	getcmd(struct CMD_LINE *, char *);
568 static int	realcon();
569 static int	spawn_processes();
570 static int	get_ioctl_syscon();
571 static int	account(short, struct PROC_TABLE *, char *);
572 static void	alarmclk();
573 static void	childeath(int);
574 static void	cleanaux();
575 static void	clearent(pid_t, short);
576 static void	console(boolean_t, char *, ...);
577 static void	init_signals(void);
578 static void	setup_pipe();
579 static void	killproc(pid_t);
580 static void	init_env();
581 static void	boot_init();
582 static void	powerfail();
583 static void	remv();
584 static void	write_ioctl_syscon();
585 static void	spawn(struct PROC_TABLE *, struct CMD_LINE *);
586 static void	setimer(int);
587 static void	siglvl(int, siginfo_t *, ucontext_t *);
588 static void	sigpoll(int);
589 static void	enter_maintenance(void);
590 static void	timer(int);
591 static void	userinit(int, char **);
592 static void	notify_pam_dead(struct utmpx *);
593 static long	waitproc(struct PROC_TABLE *);
594 static struct PROC_TABLE *efork(int, struct PROC_TABLE *, int);
595 static struct PROC_TABLE *findpslot(struct CMD_LINE *);
596 static void	increase_proc_table_size();
597 static void	st_init();
598 static void	st_write();
599 static void	contracts_init();
600 static void	contract_event(struct pollfd *);
601 static int	startd_run(const char *, int, ctid_t);
602 static void	startd_record_failure();
603 static int	startd_failure_rate_critical();
604 static char	*audit_boot_msg();
605 static int	audit_put_record(int, int, char *);
606 static void	update_boot_archive(int new_state);
607 static void	init_bootbanner_print(const char *, uint_t);
608 
609 int
610 main(int argc, char *argv[])
611 {
612 	int	chg_lvl_flag = FALSE, print_banner = FALSE;
613 	int	may_need_audit = 1;
614 	int	c;
615 	char	*msg;
616 
617 	/* Get a timestamp for use as boot time, if needed. */
618 	(void) time(&init_boot_time);
619 
620 	/* Get the default umask */
621 	cmask = umask(022);
622 	(void) umask(cmask);
623 
624 	/* Parse the arguments to init. Check for single user */
625 	opterr = 0;
626 	while ((c = getopt(argc, argv, "brsm:")) != EOF) {
627 		switch (c) {
628 		case 'b':
629 			rflg = 0;
630 			bflg = 1;
631 			if (!sflg)
632 				sflg++;
633 			break;
634 		case 'r':
635 			bflg = 0;
636 			rflg++;
637 			break;
638 		case 's':
639 			if (!bflg)
640 				sflg++;
641 			break;
642 		case 'm':
643 			smf_options = optarg;
644 			smf_debug = (strstr(smf_options, "debug") != NULL);
645 			break;
646 		}
647 	}
648 
649 	/*
650 	 * Determine if we are the main init, or a user invoked init, whose job
651 	 * it is to inform init to change levels or perform some other action.
652 	 */
653 	if (zone_getattr(getzoneid(), ZONE_ATTR_INITPID, &init_pid,
654 	    sizeof (init_pid)) != sizeof (init_pid)) {
655 		(void) fprintf(stderr, "could not get pid for init\n");
656 		return (1);
657 	}
658 
659 	/*
660 	 * If this PID is not the same as the "true" init for the zone, then we
661 	 * must be in 'user' mode.
662 	 */
663 	if (getpid() != init_pid) {
664 		userinit(argc, argv);
665 	}
666 
667 	if (getzoneid() != GLOBAL_ZONEID) {
668 		print_banner = TRUE;
669 	}
670 
671 	/*
672 	 * Initialize state (and set "booting").
673 	 */
674 	st_init();
675 
676 	if (booting && print_banner) {
677 		/*
678 		 * We want to print the boot banner as soon as
679 		 * possible.  In the global zone, the kernel does it,
680 		 * but we do not have that luxury in non-global zones,
681 		 * so we will print it here.
682 		 */
683 #ifdef	LEGACY_BANNER
684 		struct utsname un;
685 		char buf[BUFSIZ];
686 		const char *bits;
687 		int r;
688 
689 		(void) uname(&un);
690 		if ((r = sysinfo(SI_ADDRESS_WIDTH, buf, sizeof (buf))) > 0 &&
691 		    r < sizeof (buf)) {
692 			bits = buf;
693 		} else {
694 			bits = "64";
695 		}
696 
697 		console(B_FALSE,
698 		    "\n\n%s Release %s Version %s %s-bit\r\n",
699 		    un.sysname, un.release, un.version, bits);
700 		console(B_FALSE,
701 		    "Copyright (c) 1983, 2010, Oracle and/or its affiliates."
702 		    " All rights reserved.\r\n");
703 #else
704 		bootbanner_print(init_bootbanner_print, 0);
705 #endif
706 	}
707 
708 	/*
709 	 * Get the ioctl settings for /dev/syscon from /etc/ioctl.syscon
710 	 * so that it can be brought up in the state it was in when the
711 	 * system went down; or set to defaults if ioctl.syscon isn't
712 	 * valid.
713 	 *
714 	 * This needs to be done even if we're restarting so reset_modes()
715 	 * will work in case we need to go down to single user mode.
716 	 */
717 	write_ioctl = get_ioctl_syscon();
718 
719 	/*
720 	 * Set up all signals to be caught or ignored as appropriate.
721 	 */
722 	init_signals();
723 
724 	/* Load glob_envp from ENVFILE. */
725 	init_env();
726 
727 	contracts_init();
728 
729 	if (!booting) {
730 		/* cur_state should have been read in. */
731 
732 		op_modes = NORMAL_MODES;
733 
734 		/* Rewrite the ioctl file if it was bad. */
735 		if (write_ioctl)
736 			write_ioctl_syscon();
737 	} else {
738 		/*
739 		 * It's fine to boot up with state as zero, because
740 		 * startd will later tell us the real state.
741 		 */
742 		cur_state = 0;
743 		op_modes = BOOT_MODES;
744 
745 		boot_init();
746 	}
747 
748 	prev_state = prior_state = cur_state;
749 
750 	setup_pipe();
751 
752 	/*
753 	 * Here is the beginning of the main process loop.
754 	 */
755 	for (;;) {
756 		if (lvlq_received) {
757 			setup_pipe();
758 			lvlq_received = B_FALSE;
759 		}
760 
761 		/*
762 		 * Clean up any accounting records for dead "godchildren".
763 		 */
764 		if (Gchild)
765 			cleanaux();
766 
767 		/*
768 		 * If in "normal" mode, check all living processes and initiate
769 		 * kill sequence on those that should not be there anymore.
770 		 */
771 		if (op_modes == NORMAL_MODES && cur_state != LVLa &&
772 		    cur_state != LVLb && cur_state != LVLc)
773 			remv();
774 
775 		/*
776 		 * If a change in run levels is the reason we awoke, now do
777 		 * the accounting to report the change in the utmp file.
778 		 * Also report the change on the system console.
779 		 */
780 		if (chg_lvl_flag) {
781 			chg_lvl_flag = FALSE;
782 
783 			if (state_to_flags(cur_state) & LSEL_RUNLEVEL) {
784 				char rl = state_to_name(cur_state);
785 
786 				if (rl != -1)
787 					lscf_set_runlevel(rl);
788 			}
789 
790 			may_need_audit = 1;
791 		}
792 
793 		/*
794 		 * Scan the inittab file and spawn and respawn processes that
795 		 * should be alive in the current state. If inittab does not
796 		 * exist default to  single user mode.
797 		 */
798 		if (spawn_processes() == FAILURE) {
799 			prior_state = prev_state;
800 			cur_state = SINGLE_USER;
801 		}
802 
803 		/* If any respawns occurred, take note. */
804 		if (rsflag) {
805 			rsflag = 0;
806 			spawncnt++;
807 		}
808 
809 		/*
810 		 * If a powerfail signal was received during the last
811 		 * sequence, set mode to powerfail.  When spawn_processes() is
812 		 * entered the first thing it does is to check "powerhit".  If
813 		 * it is in PF_MODES then it clears "powerhit" and does
814 		 * a powerfail sequence.  If it is not in PF_MODES, then it
815 		 * puts itself in PF_MODES and then clears "powerhit".  Should
816 		 * "powerhit" get set again while spawn_processes() is working
817 		 * on a powerfail sequence, the following code  will see that
818 		 * spawn_processes() tries to execute the powerfail sequence
819 		 * again.  This guarantees that the powerfail sequence will be
820 		 * successfully completed before further processing takes
821 		 * place.
822 		 */
823 		if (wakeup.w_flags.w_powerhit) {
824 			op_modes = PF_MODES;
825 			/*
826 			 * Make sure that cur_state != prev_state so that
827 			 * ONCE and WAIT types work.
828 			 */
829 			prev_state = 0;
830 		} else if (op_modes != NORMAL_MODES) {
831 			/*
832 			 * If spawn_processes() was not just called while in
833 			 * normal mode, we set the mode to normal and it will
834 			 * be called again to check normal modes.  If we have
835 			 * just finished a powerfail sequence with prev_state
836 			 * equal to zero, we set prev_state equal to cur_state
837 			 * before the next pass through.
838 			 */
839 			if (op_modes == PF_MODES)
840 				prev_state = cur_state;
841 			op_modes = NORMAL_MODES;
842 		} else if (cur_state == LVLa || cur_state == LVLb ||
843 		    cur_state == LVLc) {
844 			/*
845 			 * If it was a change of levels that awakened us and the
846 			 * new level is one of the demand levels then reset
847 			 * cur_state to the previous state and do another scan
848 			 * to take care of the usual respawn actions.
849 			 */
850 			cur_state = prior_state;
851 			prior_state = prev_state;
852 			prev_state = cur_state;
853 		} else {
854 			prev_state = cur_state;
855 
856 			if (wakeup.w_mask == 0) {
857 				int ret;
858 
859 				if (may_need_audit && (cur_state == LVL3)) {
860 					msg = audit_boot_msg();
861 
862 					may_need_audit = 0;
863 					(void) audit_put_record(ADT_SUCCESS,
864 					    ADT_SUCCESS, msg);
865 					free(msg);
866 				}
867 
868 				/*
869 				 * "init" is finished with all actions for
870 				 * the current wakeup.
871 				 */
872 				ret = poll(poll_fds, poll_nfds,
873 				    SLEEPTIME * MILLISEC);
874 				pausecnt++;
875 				if (ret > 0)
876 					contract_event(&poll_fds[0]);
877 				else if (ret < 0 && errno != EINTR)
878 					console(B_TRUE, "poll() error: %s\n",
879 					    strerror(errno));
880 			}
881 
882 			if (wakeup.w_flags.w_usersignal) {
883 				/*
884 				 * Install the new level.  This could be a real
885 				 * change in levels  or a telinit [Q|a|b|c] or
886 				 * just a telinit to the same level at which
887 				 * we are running.
888 				 */
889 				if (new_state != cur_state) {
890 					if (new_state == LVLa ||
891 					    new_state == LVLb ||
892 					    new_state == LVLc) {
893 						prev_state = prior_state;
894 						prior_state = cur_state;
895 						cur_state = new_state;
896 					} else {
897 						prev_state = cur_state;
898 						if (cur_state >= 0)
899 							prior_state = cur_state;
900 						cur_state = new_state;
901 						chg_lvl_flag = TRUE;
902 					}
903 				}
904 
905 				new_state = 0;
906 			}
907 
908 			if (wakeup.w_flags.w_powerhit)
909 				op_modes = PF_MODES;
910 
911 			/*
912 			 * Clear all wakeup reasons.
913 			 */
914 			wakeup.w_mask = 0;
915 		}
916 	}
917 
918 	/*NOTREACHED*/
919 }
920 
921 static void
922 init_bootbanner_print(const char *line, uint_t num)
923 {
924 	const char *pfx = (num == 0) ? "\n\n" : "";
925 
926 	console(B_FALSE, "%s%s\r\n", pfx, line);
927 }
928 
929 static void
930 update_boot_archive(int new_state)
931 {
932 	if (new_state != LVL0 && new_state != LVL5 && new_state != LVL6)
933 		return;
934 
935 	if (getzoneid() != GLOBAL_ZONEID)
936 		return;
937 
938 	(void) system("/sbin/bootadm -ea update_all");
939 }
940 
941 /*
942  * void enter_maintenance()
943  *   A simple invocation of sulogin(8), with no baggage, in the case that we
944  *   are unable to activate svc.startd(8).  We fork; the child runs sulogin;
945  *   we wait for it to exit.
946  */
947 static void
948 enter_maintenance()
949 {
950 	struct PROC_TABLE	*su_process;
951 
952 	console(B_FALSE, "Requesting maintenance mode\n"
953 	    "(See /lib/svc/share/README for additional information.)\n");
954 	(void) sighold(SIGCLD);
955 	while ((su_process = efork(M_OFF, NULLPROC, NOCLEANUP)) == NO_ROOM)
956 		(void) pause();
957 	(void) sigrelse(SIGCLD);
958 	if (su_process == NULLPROC) {
959 		int fd;
960 
961 		(void) fclose(stdin);
962 		(void) fclose(stdout);
963 		(void) fclose(stderr);
964 		closefrom(0);
965 
966 		fd = open(SYSCON, O_RDWR | O_NOCTTY);
967 		if (fd >= 0) {
968 			(void) dup2(fd, 1);
969 			(void) dup2(fd, 2);
970 		} else {
971 			/*
972 			 * Need to issue an error message somewhere.
973 			 */
974 			syslog(LOG_CRIT, "init[%d]: cannot open %s; %s\n",
975 			    getpid(), SYSCON, strerror(errno));
976 		}
977 
978 		/*
979 		 * Execute the "su" program.
980 		 */
981 		(void) execle(SU, SU, "-", (char *)0, glob_envp);
982 		console(B_TRUE, "execle of %s failed: %s\n", SU,
983 		    strerror(errno));
984 		timer(5);
985 		exit(1);
986 	}
987 
988 	/*
989 	 * If we are the parent, wait around for the child to die
990 	 * or for "init" to be signaled to change levels.
991 	 */
992 	while (waitproc(su_process) == FAILURE) {
993 		/*
994 		 * All other reasons for waking are ignored when in
995 		 * single-user mode.  The only child we are interested
996 		 * in is being waited for explicitly by waitproc().
997 		 */
998 		wakeup.w_mask = 0;
999 	}
1000 }
1001 
1002 /*
1003  * remv() scans through "proc_table" and performs cleanup.  If
1004  * there is a process in the table, which shouldn't be here at
1005  * the current run level, then remv() kills the process.
1006  */
1007 static void
1008 remv()
1009 {
1010 	struct PROC_TABLE	*process;
1011 	struct CMD_LINE		cmd;
1012 	char			cmd_string[MAXCMDL];
1013 	int			change_level;
1014 
1015 	change_level = (cur_state != prev_state ? TRUE : FALSE);
1016 
1017 	/*
1018 	 * Clear the TOUCHED flag on all entries so that when we have
1019 	 * finished scanning inittab, we will be able to tell if we
1020 	 * have any processes for which there is no entry in inittab.
1021 	 */
1022 	for (process = proc_table;
1023 	    (process < proc_table + num_proc); process++) {
1024 		process->p_flags &= ~TOUCHED;
1025 	}
1026 
1027 	/*
1028 	 * Scan all inittab entries.
1029 	 */
1030 	while (getcmd(&cmd, &cmd_string[0]) == TRUE) {
1031 		/* Scan for process which goes with this entry in inittab. */
1032 		for (process = proc_table;
1033 		    (process < proc_table + num_proc); process++) {
1034 			if ((process->p_flags & OCCUPIED) == 0 ||
1035 			    !id_eq(process->p_id, cmd.c_id))
1036 				continue;
1037 
1038 			/*
1039 			 * This slot contains the process we are looking for.
1040 			 */
1041 
1042 			/*
1043 			 * Is the cur_state SINGLE_USER or is this process
1044 			 * marked as "off" or was this proc started by some
1045 			 * mechanism other than LVL{a|b|c} and the current level
1046 			 * does not support this process?
1047 			 */
1048 			if (cur_state == SINGLE_USER ||
1049 			    cmd.c_action == M_OFF ||
1050 			    ((cmd.c_levels & state_to_mask(cur_state)) == 0 &&
1051 			    (process->p_flags & DEMANDREQUEST) == 0)) {
1052 				if (process->p_flags & LIVING) {
1053 					/*
1054 					 * Touch this entry so we know we have
1055 					 * treated it.  Note that procs which
1056 					 * are already dead at this point and
1057 					 * should not be restarted are left
1058 					 * untouched.  This causes their slot to
1059 					 * be freed later after dead accounting
1060 					 * is done.
1061 					 */
1062 					process->p_flags |= TOUCHED;
1063 
1064 					if ((process->p_flags & KILLED) == 0) {
1065 						if (change_level) {
1066 							process->p_flags
1067 							    |= WARNED;
1068 							(void) kill(
1069 							    process->p_pid,
1070 							    SIGTERM);
1071 						} else {
1072 							/*
1073 							 * Fork a killing proc
1074 							 * so "init" can
1075 							 * continue without
1076 							 * having to pause for
1077 							 * TWARN seconds.
1078 							 */
1079 							killproc(
1080 							    process->p_pid);
1081 						}
1082 						process->p_flags |= KILLED;
1083 					}
1084 				}
1085 			} else {
1086 				/*
1087 				 * Process can exist at current level.  If it is
1088 				 * still alive or a DEMANDREQUEST we touch it so
1089 				 * it will be left alone.  Otherwise we leave it
1090 				 * untouched so it will be accounted for and
1091 				 * cleaned up later in remv().  Dead
1092 				 * DEMANDREQUESTs will be accounted but not
1093 				 * freed.
1094 				 */
1095 				if (process->p_flags &
1096 				    (LIVING|NOCLEANUP|DEMANDREQUEST))
1097 					process->p_flags |= TOUCHED;
1098 			}
1099 
1100 			break;
1101 		}
1102 	}
1103 
1104 	st_write();
1105 
1106 	/*
1107 	 * If this was a change of levels call, scan through the
1108 	 * process table for processes that were warned to die.  If any
1109 	 * are found that haven't left yet, sleep for TWARN seconds and
1110 	 * then send final terminations to any that haven't died yet.
1111 	 */
1112 	if (change_level) {
1113 
1114 		/*
1115 		 * Set the alarm for TWARN seconds on the assumption
1116 		 * that there will be some that need to be waited for.
1117 		 * This won't harm anything except we are guaranteed to
1118 		 * wakeup in TWARN seconds whether we need to or not.
1119 		 */
1120 		setimer(TWARN);
1121 
1122 		/*
1123 		 * Scan for processes which should be dying.  We hope they
1124 		 * will die without having to be sent a SIGKILL signal.
1125 		 */
1126 		for (process = proc_table;
1127 		    (process < proc_table + num_proc); process++) {
1128 			/*
1129 			 * If this process should die, hasn't yet, and the
1130 			 * TWARN time hasn't expired yet, wait for process
1131 			 * to die or for timer to expire.
1132 			 */
1133 			while (time_up == FALSE &&
1134 			    (process->p_flags & (WARNED|LIVING|OCCUPIED)) ==
1135 			    (WARNED|LIVING|OCCUPIED))
1136 				(void) pause();
1137 
1138 			if (time_up == TRUE)
1139 				break;
1140 		}
1141 
1142 		/*
1143 		 * If we reached the end of the table without the timer
1144 		 * expiring, then there are no procs which will have to be
1145 		 * sent the SIGKILL signal.  If the timer has expired, then
1146 		 * it is necessary to scan the table again and send signals
1147 		 * to all processes which aren't going away nicely.
1148 		 */
1149 		if (time_up == TRUE) {
1150 			for (process = proc_table;
1151 			    (process < proc_table + num_proc); process++) {
1152 				if ((process->p_flags &
1153 				    (WARNED|LIVING|OCCUPIED)) ==
1154 				    (WARNED|LIVING|OCCUPIED))
1155 					(void) kill(process->p_pid, SIGKILL);
1156 			}
1157 		}
1158 		setimer(0);
1159 	}
1160 
1161 	/*
1162 	 * Rescan the proc_table for two kinds of entry, those marked LIVING,
1163 	 * NAMED, which don't have an entry in inittab (haven't been TOUCHED
1164 	 * by the above scanning), and haven't been sent kill signals, and
1165 	 * those entries marked not LIVING, NAMED.  The former procs are killed.
1166 	 * The latter have DEAD_PROCESS accounting done and the slot cleared.
1167 	 */
1168 	for (process = proc_table;
1169 	    (process < proc_table + num_proc); process++) {
1170 		if ((process->p_flags & (LIVING|NAMED|TOUCHED|KILLED|OCCUPIED))
1171 		    == (LIVING|NAMED|OCCUPIED)) {
1172 			killproc(process->p_pid);
1173 			process->p_flags |= KILLED;
1174 		} else if ((process->p_flags & (LIVING|NAMED|OCCUPIED)) ==
1175 		    (NAMED|OCCUPIED)) {
1176 			(void) account(DEAD_PROCESS, process, NULL);
1177 			/*
1178 			 * If this named proc hasn't been TOUCHED, then free the
1179 			 * space. It has either died of it's own accord, but
1180 			 * isn't respawnable or it was killed because it
1181 			 * shouldn't exist at this level.
1182 			 */
1183 			if ((process->p_flags & TOUCHED) == 0)
1184 				process->p_flags = 0;
1185 		}
1186 	}
1187 
1188 	st_write();
1189 }
1190 
1191 /*
1192  * Extract the svc.startd command line and whether to restart it from its
1193  * inittab entry.
1194  */
1195 /*ARGSUSED*/
1196 static void
1197 process_startd_line(struct CMD_LINE *cmd, char *cmd_string)
1198 {
1199 	size_t sz;
1200 
1201 	/* Save the command line. */
1202 	if (sflg || rflg) {
1203 		/* Also append -r or -s. */
1204 		(void) strlcpy(startd_cline, cmd_string, sizeof (startd_cline));
1205 		(void) strlcat(startd_cline, " -", sizeof (startd_cline));
1206 		if (sflg)
1207 			sz = strlcat(startd_cline, "s", sizeof (startd_cline));
1208 		if (rflg)
1209 			sz = strlcat(startd_cline, "r", sizeof (startd_cline));
1210 	} else {
1211 		sz = strlcpy(startd_cline, cmd_string, sizeof (startd_cline));
1212 	}
1213 
1214 	if (sz >= sizeof (startd_cline)) {
1215 		console(B_TRUE,
1216 		    "svc.startd command line too long.  Ignoring.\n");
1217 		startd_cline[0] = '\0';
1218 		return;
1219 	}
1220 }
1221 
1222 /*
1223  * spawn_processes() scans inittab for entries which should be run at this
1224  * mode.  Processes which should be running but are not, are started.
1225  */
1226 static int
1227 spawn_processes()
1228 {
1229 	struct PROC_TABLE		*pp;
1230 	struct CMD_LINE			cmd;
1231 	char				cmd_string[MAXCMDL];
1232 	short				lvl_mask;
1233 	int				status;
1234 
1235 	/*
1236 	 * First check the "powerhit" flag.  If it is set, make sure the modes
1237 	 * are PF_MODES and clear the "powerhit" flag.  Avoid the possible race
1238 	 * on the "powerhit" flag by disallowing a new powerfail interrupt
1239 	 * between the test of the powerhit flag and the clearing of it.
1240 	 */
1241 	if (wakeup.w_flags.w_powerhit) {
1242 		wakeup.w_flags.w_powerhit = 0;
1243 		op_modes = PF_MODES;
1244 	}
1245 	lvl_mask = state_to_mask(cur_state);
1246 
1247 	/*
1248 	 * Scan through all the entries in inittab.
1249 	 */
1250 	while ((status = getcmd(&cmd, &cmd_string[0])) == TRUE) {
1251 		if (id_eq(cmd.c_id, "smf")) {
1252 			process_startd_line(&cmd, cmd_string);
1253 			continue;
1254 		}
1255 
1256 retry_for_proc_slot:
1257 
1258 		/*
1259 		 * Find out if there is a process slot for this entry already.
1260 		 */
1261 		if ((pp = findpslot(&cmd)) == NULLPROC) {
1262 			/*
1263 			 * we've run out of proc table entries
1264 			 * increase proc_table.
1265 			 */
1266 			increase_proc_table_size();
1267 
1268 			/*
1269 			 * Retry now as we have an empty proc slot.
1270 			 * In case increase_proc_table_size() fails,
1271 			 * we will keep retrying.
1272 			 */
1273 			goto retry_for_proc_slot;
1274 		}
1275 
1276 		/*
1277 		 * If there is an entry, and it is marked as DEMANDREQUEST,
1278 		 * one of the levels a, b, or c is in its levels mask, and
1279 		 * the action field is ONDEMAND and ONDEMAND is a permissable
1280 		 * mode, and the process is dead, then respawn it.
1281 		 */
1282 		if (((pp->p_flags & (LIVING|DEMANDREQUEST)) == DEMANDREQUEST) &&
1283 		    (cmd.c_levels & MASK_abc) &&
1284 		    (cmd.c_action & op_modes) == M_ONDEMAND) {
1285 			spawn(pp, &cmd);
1286 			continue;
1287 		}
1288 
1289 		/*
1290 		 * If the action is not an action we are interested in,
1291 		 * skip the entry.
1292 		 */
1293 		if ((cmd.c_action & op_modes) == 0 || pp->p_flags & LIVING ||
1294 		    (cmd.c_levels & lvl_mask) == 0)
1295 			continue;
1296 
1297 		/*
1298 		 * If the modes are the normal modes (ONCE, WAIT, RESPAWN, OFF,
1299 		 * ONDEMAND) and the action field is either OFF or the action
1300 		 * field is ONCE or WAIT and the current level is the same as
1301 		 * the last level, then skip this entry.  ONCE and WAIT only
1302 		 * get run when the level changes.
1303 		 */
1304 		if (op_modes == NORMAL_MODES &&
1305 		    (cmd.c_action == M_OFF ||
1306 		    (cmd.c_action & (M_ONCE|M_WAIT)) &&
1307 		    cur_state == prev_state))
1308 			continue;
1309 
1310 		/*
1311 		 * At this point we are interested in performing the action for
1312 		 * this entry.  Actions fall into two categories, spinning off
1313 		 * a process and not waiting, and spinning off a process and
1314 		 * waiting for it to die.  If the action is ONCE, RESPAWN,
1315 		 * ONDEMAND, POWERFAIL, or BOOT we don't wait for the process
1316 		 * to die, for all other actions we do wait.
1317 		 */
1318 		if (cmd.c_action & (M_ONCE | M_RESPAWN | M_PF | M_BOOT)) {
1319 			spawn(pp, &cmd);
1320 
1321 		} else {
1322 			spawn(pp, &cmd);
1323 			while (waitproc(pp) == FAILURE)
1324 				;
1325 			(void) account(DEAD_PROCESS, pp, NULL);
1326 			pp->p_flags = 0;
1327 		}
1328 	}
1329 	return (status);
1330 }
1331 
1332 /*
1333  * spawn() spawns a shell, inserts the information about the process
1334  * process into the proc_table, and does the startup accounting.
1335  */
1336 static void
1337 spawn(struct PROC_TABLE *process, struct CMD_LINE *cmd)
1338 {
1339 	int		i;
1340 	int		modes, maxfiles;
1341 	time_t		now;
1342 	struct PROC_TABLE tmproc, *oprocess;
1343 
1344 	/*
1345 	 * The modes to be sent to efork() are 0 unless we are
1346 	 * spawning a LVLa, LVLb, or LVLc entry or we will be
1347 	 * waiting for the death of the child before continuing.
1348 	 */
1349 	modes = NAMED;
1350 	if (process->p_flags & DEMANDREQUEST || cur_state == LVLa ||
1351 	    cur_state == LVLb || cur_state == LVLc)
1352 		modes |= DEMANDREQUEST;
1353 	if ((cmd->c_action & (M_SYSINIT | M_WAIT | M_BOOTWAIT | M_PWAIT)) != 0)
1354 		modes |= NOCLEANUP;
1355 
1356 	/*
1357 	 * If this is a respawnable process, check the threshold
1358 	 * information to avoid excessive respawns.
1359 	 */
1360 	if (cmd->c_action & M_RESPAWN) {
1361 		/*
1362 		 * Add NOCLEANUP to all respawnable commands so that the
1363 		 * information about the frequency of respawns isn't lost.
1364 		 */
1365 		modes |= NOCLEANUP;
1366 		(void) time(&now);
1367 
1368 		/*
1369 		 * If no time is assigned, then this is the first time
1370 		 * this command is being processed in this series.  Assign
1371 		 * the current time.
1372 		 */
1373 		if (process->p_time == 0L)
1374 			process->p_time = now;
1375 
1376 		if (process->p_count++ == SPAWN_LIMIT) {
1377 
1378 			if ((now - process->p_time) < SPAWN_INTERVAL) {
1379 				/*
1380 				 * Process is respawning too rapidly.  Print
1381 				 * message and refuse to respawn it for now.
1382 				 */
1383 				console(B_TRUE, "Command is respawning too "
1384 				    "rapidly. Check for possible errors.\n"
1385 				    "id:%4s \"%s\"\n",
1386 				    &cmd->c_id[0], &cmd->c_command[EXEC]);
1387 				return;
1388 			}
1389 			process->p_time = now;
1390 			process->p_count = 0;
1391 
1392 		} else if (process->p_count > SPAWN_LIMIT) {
1393 			/*
1394 			 * If process has been respawning too rapidly and
1395 			 * the inhibit time limit hasn't expired yet, we
1396 			 * refuse to respawn.
1397 			 */
1398 			if (now - process->p_time < SPAWN_INTERVAL + INHIBIT)
1399 				return;
1400 			process->p_time = now;
1401 			process->p_count = 0;
1402 		}
1403 		rsflag = TRUE;
1404 	}
1405 
1406 	/*
1407 	 * Spawn a child process to execute this command.
1408 	 */
1409 	(void) sighold(SIGCLD);
1410 	oprocess = process;
1411 	while ((process = efork(cmd->c_action, oprocess, modes)) == NO_ROOM)
1412 		(void) pause();
1413 
1414 	if (process == NULLPROC) {
1415 
1416 		/*
1417 		 * We are the child.  We must make sure we get a different
1418 		 * file pointer for our references to utmpx.  Otherwise our
1419 		 * seeks and reads will compete with those of the parent.
1420 		 */
1421 		endutxent();
1422 
1423 		/*
1424 		 * Perform the accounting for the beginning of a process.
1425 		 * Note that all processes are initially "INIT_PROCESS"es.
1426 		 */
1427 		tmproc.p_id[0] = cmd->c_id[0];
1428 		tmproc.p_id[1] = cmd->c_id[1];
1429 		tmproc.p_id[2] = cmd->c_id[2];
1430 		tmproc.p_id[3] = cmd->c_id[3];
1431 		tmproc.p_pid = getpid();
1432 		tmproc.p_exit = 0;
1433 		(void) account(INIT_PROCESS, &tmproc,
1434 		    prog_name(&cmd->c_command[EXEC]));
1435 		maxfiles = ulimit(UL_GDESLIM, 0);
1436 		for (i = 0; i < maxfiles; i++)
1437 			(void) fcntl(i, F_SETFD, FD_CLOEXEC);
1438 
1439 		/*
1440 		 * Now exec a shell with the -c option and the command
1441 		 * from inittab.
1442 		 */
1443 		(void) execle(SH, "INITSH", "-c", cmd->c_command, (char *)0,
1444 		    glob_envp);
1445 		console(B_TRUE, "Command\n\"%s\"\n failed to execute.  errno "
1446 		    "= %d (exec of shell failed)\n", cmd->c_command, errno);
1447 
1448 		/*
1449 		 * Don't come back so quickly that "init" doesn't have a
1450 		 * chance to finish putting this child in "proc_table".
1451 		 */
1452 		timer(20);
1453 		exit(1);
1454 
1455 	}
1456 
1457 	/*
1458 	 * We are the parent.  Insert the necessary
1459 	 * information in the proc_table.
1460 	 */
1461 	process->p_id[0] = cmd->c_id[0];
1462 	process->p_id[1] = cmd->c_id[1];
1463 	process->p_id[2] = cmd->c_id[2];
1464 	process->p_id[3] = cmd->c_id[3];
1465 
1466 	st_write();
1467 
1468 	(void) sigrelse(SIGCLD);
1469 }
1470 
1471 /*
1472  * findpslot() finds the old slot in the process table for the
1473  * command with the same id, or it finds an empty slot.
1474  */
1475 static struct PROC_TABLE *
1476 findpslot(struct CMD_LINE *cmd)
1477 {
1478 	struct PROC_TABLE	*process;
1479 	struct PROC_TABLE	*empty = NULLPROC;
1480 
1481 	for (process = proc_table;
1482 	    (process < proc_table + num_proc); process++) {
1483 		if (process->p_flags & OCCUPIED &&
1484 		    id_eq(process->p_id, cmd->c_id))
1485 			break;
1486 
1487 		/*
1488 		 * If the entry is totally empty and "empty" is still 0,
1489 		 * remember where this hole is and make sure the slot is
1490 		 * zeroed out.
1491 		 */
1492 		if (empty == NULLPROC && (process->p_flags & OCCUPIED) == 0) {
1493 			empty = process;
1494 			process->p_id[0] = '\0';
1495 			process->p_id[1] = '\0';
1496 			process->p_id[2] = '\0';
1497 			process->p_id[3] = '\0';
1498 			process->p_pid = 0;
1499 			process->p_time = 0L;
1500 			process->p_count = 0;
1501 			process->p_flags = 0;
1502 			process->p_exit = 0;
1503 		}
1504 	}
1505 
1506 	/*
1507 	 * If there is no entry for this slot, then there should be an
1508 	 * empty slot.  If there is no empty slot, then we've run out
1509 	 * of proc_table space.  If the latter is true, empty will be
1510 	 * NULL and the caller will have to complain.
1511 	 */
1512 	if (process == (proc_table + num_proc))
1513 		process = empty;
1514 
1515 	return (process);
1516 }
1517 
1518 /*
1519  * getcmd() parses lines from inittab.  Each time it finds a command line
1520  * it will return TRUE as well as fill the passed CMD_LINE structure and
1521  * the shell command string.  When the end of inittab is reached, FALSE
1522  * is returned inittab is automatically opened if it is not currently open
1523  * and is closed when the end of the file is reached.
1524  */
1525 static FILE *fp_inittab = NULL;
1526 
1527 static int
1528 getcmd(struct CMD_LINE *cmd, char *shcmd)
1529 {
1530 	char	*ptr;
1531 	int	c, lastc, state;
1532 	char	*ptr1;
1533 	int	answer, i, proceed;
1534 	struct	stat	sbuf;
1535 	static char *actions[] = {
1536 		"off", "respawn", "ondemand", "once", "wait", "boot",
1537 		"bootwait", "powerfail", "powerwait", "initdefault",
1538 		"sysinit",
1539 	};
1540 	static short act_masks[] = {
1541 		M_OFF, M_RESPAWN, M_ONDEMAND, M_ONCE, M_WAIT, M_BOOT,
1542 		M_BOOTWAIT, M_PF, M_PWAIT, M_INITDEFAULT, M_SYSINIT,
1543 	};
1544 	/*
1545 	 * Only these actions will be allowed for entries which
1546 	 * are specified for single-user mode.
1547 	 */
1548 	short su_acts = M_INITDEFAULT | M_PF | M_PWAIT | M_WAIT;
1549 
1550 	if (fp_inittab == NULL) {
1551 		/*
1552 		 * Before attempting to open inittab we stat it to make
1553 		 * sure it currently exists and is not empty.  We try
1554 		 * several times because someone may have temporarily
1555 		 * unlinked or truncated the file.
1556 		 */
1557 		for (i = 0; i < 3; i++) {
1558 			if (stat(INITTAB, &sbuf) == -1) {
1559 				if (i == 2) {
1560 					console(B_TRUE,
1561 					    "Cannot stat %s, errno: %d\n",
1562 					    INITTAB, errno);
1563 					return (FAILURE);
1564 				} else {
1565 					timer(3);
1566 				}
1567 			} else if (sbuf.st_size < 10) {
1568 				if (i == 2) {
1569 					console(B_TRUE,
1570 					    "%s truncated or corrupted\n",
1571 					    INITTAB);
1572 					return (FAILURE);
1573 				} else {
1574 					timer(3);
1575 				}
1576 			} else {
1577 				break;
1578 			}
1579 		}
1580 
1581 		/*
1582 		 * If unable to open inittab, print error message and
1583 		 * return FAILURE to caller.
1584 		 */
1585 		if ((fp_inittab = fopen(INITTAB, "r")) == NULL) {
1586 			console(B_TRUE, "Cannot open %s errno: %d\n", INITTAB,
1587 			    errno);
1588 			return (FAILURE);
1589 		}
1590 	}
1591 
1592 	/*
1593 	 * Keep getting commands from inittab until you find a
1594 	 * good one or run out of file.
1595 	 */
1596 	for (answer = FALSE; answer == FALSE; ) {
1597 		/*
1598 		 * Zero out the cmd itself before trying next line.
1599 		 */
1600 		bzero(cmd, sizeof (struct CMD_LINE));
1601 
1602 		/*
1603 		 * Read in lines of inittab, parsing at colons, until a line is
1604 		 * read in which doesn't end with a backslash.  Do not start if
1605 		 * the first character read is an EOF.  Note that this means
1606 		 * that lines which don't end in a newline are still processed,
1607 		 * since the "for" will terminate normally once started,
1608 		 * regardless of whether line terminates with a newline or EOF.
1609 		 */
1610 		state = FAILURE;
1611 		if ((c = fgetc(fp_inittab)) == EOF) {
1612 			answer = FALSE;
1613 			(void) fclose(fp_inittab);
1614 			fp_inittab = NULL;
1615 			break;
1616 		}
1617 
1618 		for (proceed = TRUE, ptr = shcmd, state = ID, lastc = '\0';
1619 		    proceed && c != EOF;
1620 		    lastc = c, c = fgetc(fp_inittab)) {
1621 			/* If we're not in the FAILURE state and haven't */
1622 			/* yet reached the shell command field, process	 */
1623 			/* the line, otherwise just look for a real end	 */
1624 			/* of line.					 */
1625 			if (state != FAILURE && state != COMMAND) {
1626 			/*
1627 			 * Squeeze out spaces and tabs.
1628 			 */
1629 			if (c == ' ' || c == '\t')
1630 				continue;
1631 
1632 			/*
1633 			 * Ignore characters in a comment, except for the \n.
1634 			 */
1635 			if (state == COMMENT) {
1636 				if (c == '\n') {
1637 					lastc = ' ';
1638 					break;
1639 				} else {
1640 					continue;
1641 				}
1642 			}
1643 
1644 			/*
1645 			 * Detect comments (lines whose first non-whitespace
1646 			 * character is '#') by checking that we're at the
1647 			 * beginning of a line, have seen a '#', and haven't
1648 			 * yet accumulated any characters.
1649 			 */
1650 			if (state == ID && c == '#' && ptr == shcmd) {
1651 				state = COMMENT;
1652 				continue;
1653 			}
1654 
1655 			/*
1656 			 * If the character is a ':', then check the
1657 			 * previous field for correctness and advance
1658 			 * to the next field.
1659 			 */
1660 			if (c == ':') {
1661 				switch (state) {
1662 
1663 				case ID :
1664 				/*
1665 				 * Check to see that there are only
1666 				 * 1 to 4 characters for the id.
1667 				 */
1668 				if ((i = ptr - shcmd) < 1 || i > 4) {
1669 					state = FAILURE;
1670 				} else {
1671 					bcopy(shcmd, &cmd->c_id[0], i);
1672 					ptr = shcmd;
1673 					state = LEVELS;
1674 				}
1675 				break;
1676 
1677 				case LEVELS :
1678 				/*
1679 				 * Build a mask for all the levels for
1680 				 * which this command will be legal.
1681 				 */
1682 				for (cmd->c_levels = 0, ptr1 = shcmd;
1683 				    ptr1 < ptr; ptr1++) {
1684 					int mask;
1685 					if (lvlname_to_mask(*ptr1,
1686 					    &mask) == -1) {
1687 						state = FAILURE;
1688 						break;
1689 					}
1690 					cmd->c_levels |= mask;
1691 				}
1692 				if (state != FAILURE) {
1693 					state = ACTION;
1694 					ptr = shcmd;	/* Reset the buffer */
1695 				}
1696 				break;
1697 
1698 				case ACTION :
1699 				/*
1700 				 * Null terminate the string in shcmd buffer and
1701 				 * then try to match against legal actions.  If
1702 				 * the field is of length 0, then the default of
1703 				 * "RESPAWN" is used if the id is numeric,
1704 				 * otherwise the default is "OFF".
1705 				 */
1706 				if (ptr == shcmd) {
1707 					if (isdigit(cmd->c_id[0]) &&
1708 					    (cmd->c_id[1] == '\0' ||
1709 					    isdigit(cmd->c_id[1])) &&
1710 					    (cmd->c_id[2] == '\0' ||
1711 					    isdigit(cmd->c_id[2])) &&
1712 					    (cmd->c_id[3] == '\0' ||
1713 					    isdigit(cmd->c_id[3])))
1714 						cmd->c_action = M_RESPAWN;
1715 					else
1716 						cmd->c_action = M_OFF;
1717 				} else {
1718 					for (cmd->c_action = 0, i = 0,
1719 					    *ptr = '\0';
1720 					    i <
1721 					    sizeof (actions)/sizeof (char *);
1722 					    i++) {
1723 					if (strcmp(shcmd, actions[i]) == 0) {
1724 						if ((cmd->c_levels & MASKSU) &&
1725 						    !(act_masks[i] & su_acts))
1726 							cmd->c_action = 0;
1727 						else
1728 							cmd->c_action =
1729 							    act_masks[i];
1730 						break;
1731 					}
1732 					}
1733 				}
1734 
1735 				/*
1736 				 * If the action didn't match any legal action,
1737 				 * set state to FAILURE.
1738 				 */
1739 				if (cmd->c_action == 0) {
1740 					state = FAILURE;
1741 				} else {
1742 					state = COMMAND;
1743 					(void) strcpy(shcmd, "exec ");
1744 				}
1745 				ptr = shcmd + EXEC;
1746 				break;
1747 				}
1748 				continue;
1749 			}
1750 		}
1751 
1752 		/* If the character is a '\n', then this is the end of a */
1753 		/* line.  If the '\n' wasn't preceded by a backslash, */
1754 		/* it is also the end of an inittab command.  If it was */
1755 		/* preceded by a backslash then the next line is a */
1756 		/* continuation.  Note that the continuation '\n' falls */
1757 		/* through and is treated like other characters and is */
1758 		/* stored in the shell command line. */
1759 		if (c == '\n' && lastc != '\\') {
1760 			proceed = FALSE;
1761 			*ptr = '\0';
1762 			break;
1763 		}
1764 
1765 		/* For all other characters just stuff them into the */
1766 		/* command as long as there aren't too many of them. */
1767 		/* Make sure there is room for a terminating '\0' also. */
1768 		if (ptr >= shcmd + MAXCMDL - 1)
1769 			state = FAILURE;
1770 		else
1771 			*ptr++ = (char)c;
1772 
1773 		/* If the character we just stored was a quoted	*/
1774 		/* backslash, then change "c" to '\0', so that this	*/
1775 		/* backslash will not cause a subsequent '\n' to appear */
1776 		/* quoted.  In otherwords '\' '\' '\n' is the real end */
1777 		/* of a command, while '\' '\n' is a continuation. */
1778 		if (c == '\\' && lastc == '\\')
1779 			c = '\0';
1780 		}
1781 
1782 		/*
1783 		 * Make sure all the fields are properly specified
1784 		 * for a good command line.
1785 		 */
1786 		if (state == COMMAND) {
1787 			answer = TRUE;
1788 			cmd->c_command = shcmd;
1789 
1790 			/*
1791 			 * If no default level was supplied, insert
1792 			 * all numerical levels.
1793 			 */
1794 			if (cmd->c_levels == 0)
1795 				cmd->c_levels = MASK_NUMERIC;
1796 
1797 			/*
1798 			 * If no action has been supplied, declare this
1799 			 * entry to be OFF.
1800 			 */
1801 			if (cmd->c_action == 0)
1802 				cmd->c_action = M_OFF;
1803 
1804 			/*
1805 			 * If no shell command has been supplied, make sure
1806 			 * there is a null string in the command field.
1807 			 */
1808 			if (ptr == shcmd + EXEC)
1809 				*shcmd = '\0';
1810 		} else
1811 			answer = FALSE;
1812 
1813 		/*
1814 		 * If we have reached the end of inittab, then close it
1815 		 * and quit trying to find a good command line.
1816 		 */
1817 		if (c == EOF) {
1818 			(void) fclose(fp_inittab);
1819 			fp_inittab = NULL;
1820 			break;
1821 		}
1822 	}
1823 	return (answer);
1824 }
1825 
1826 /*
1827  * lvlname_to_state(): convert the character name of a state to its level
1828  * (its corresponding signal number).
1829  */
1830 static int
1831 lvlname_to_state(char name)
1832 {
1833 	int i;
1834 	for (i = 0; i < LVL_NELEMS; i++) {
1835 		if (lvls[i].lvl_name == name)
1836 			return (lvls[i].lvl_state);
1837 	}
1838 	return (-1);
1839 }
1840 
1841 /*
1842  * state_to_name(): convert the level to the character name.
1843  */
1844 static char
1845 state_to_name(int state)
1846 {
1847 	int i;
1848 	for (i = 0; i < LVL_NELEMS; i++) {
1849 		if (lvls[i].lvl_state == state)
1850 			return (lvls[i].lvl_name);
1851 	}
1852 	return (-1);
1853 }
1854 
1855 /*
1856  * state_to_mask(): return the mask corresponding to a signal number
1857  */
1858 static int
1859 state_to_mask(int state)
1860 {
1861 	int i;
1862 	for (i = 0; i < LVL_NELEMS; i++) {
1863 		if (lvls[i].lvl_state == state)
1864 			return (lvls[i].lvl_mask);
1865 	}
1866 	return (0);	/* return 0, since that represents an empty mask */
1867 }
1868 
1869 /*
1870  * lvlname_to_mask(): return the mask corresponding to a levels character name
1871  */
1872 static int
1873 lvlname_to_mask(char name, int *mask)
1874 {
1875 	int i;
1876 	for (i = 0; i < LVL_NELEMS; i++) {
1877 		if (lvls[i].lvl_name == name) {
1878 			*mask = lvls[i].lvl_mask;
1879 			return (0);
1880 		}
1881 	}
1882 	return (-1);
1883 }
1884 
1885 /*
1886  * state_to_flags(): return the flags corresponding to a runlevel.  These
1887  * indicate properties of that runlevel.
1888  */
1889 static int
1890 state_to_flags(int state)
1891 {
1892 	int i;
1893 	for (i = 0; i < LVL_NELEMS; i++) {
1894 		if (lvls[i].lvl_state == state)
1895 			return (lvls[i].lvl_flags);
1896 	}
1897 	return (0);
1898 }
1899 
1900 /*
1901  * killproc() creates a child which kills the process specified by pid.
1902  */
1903 void
1904 killproc(pid_t pid)
1905 {
1906 	struct PROC_TABLE	*process;
1907 
1908 	(void) sighold(SIGCLD);
1909 	while ((process = efork(M_OFF, NULLPROC, 0)) == NO_ROOM)
1910 		(void) pause();
1911 	(void) sigrelse(SIGCLD);
1912 
1913 	if (process == NULLPROC) {
1914 		/*
1915 		 * efork() sets all signal handlers to the default, so reset
1916 		 * the ALRM handler to make timer() work as expected.
1917 		 */
1918 		(void) sigset(SIGALRM, alarmclk);
1919 
1920 		/*
1921 		 * We are the child.  Try to terminate the process nicely
1922 		 * first using SIGTERM and if it refuses to die in TWARN
1923 		 * seconds kill it with SIGKILL.
1924 		 */
1925 		(void) kill(pid, SIGTERM);
1926 		(void) timer(TWARN);
1927 		(void) kill(pid, SIGKILL);
1928 		(void) exit(0);
1929 	}
1930 }
1931 
1932 /*
1933  * Set up the default environment for all procs to be forked from init.
1934  * Read the values from the /etc/default/init file, except for PATH.  If
1935  * there is not enough room in the environment array, the environment
1936  * lines that don't fit are discarded and a message is written to the console.
1937  */
1938 void
1939 init_env()
1940 {
1941 	void		*dstate;
1942 	const char	*tokp;
1943 
1944 	glob_envp[0] = malloc((unsigned)(strlen(DEF_PATH)+2));
1945 	(void) strcpy(glob_envp[0], DEF_PATH);
1946 	glob_envn = 1;
1947 
1948 	if (rflg) {
1949 		glob_envp[1] =
1950 		    malloc((unsigned)(strlen("_DVFS_RECONFIG=YES")+2));
1951 		(void) strcpy(glob_envp[1], "_DVFS_RECONFIG=YES");
1952 		++glob_envn;
1953 	} else if (bflg == 1) {
1954 		glob_envp[1] =
1955 		    malloc((unsigned)(strlen("RB_NOBOOTRC=YES")+2));
1956 		(void) strcpy(glob_envp[1], "RB_NOBOOTRC=YES");
1957 		++glob_envn;
1958 	}
1959 
1960 	if (definit_open(ENVFILE, &dstate) != 0) {
1961 		console(B_TRUE,
1962 		    "Cannot open %s. Environment not initialized.\n",
1963 		    ENVFILE);
1964 		return;
1965 	}
1966 
1967 	while ((tokp = definit_token(dstate)) != NULL &&
1968 	    glob_envn < MAXENVENT - 2) {
1969 
1970 		if (strncmp(tokp, "CMASK=", sizeof ("CMASK=") - 1) == 0) {
1971 			long t;
1972 
1973 			/* We know there's an = */
1974 			t = strtol(strchr(tokp, '=') + 1, NULL, 8);
1975 
1976 			/* Sanity */
1977 			if (t >= DEFINIT_MIN_UMASK && t <= DEFINIT_MAX_UMASK)
1978 				cmask = (int)t;
1979 			(void) umask(cmask);
1980 			continue;
1981 		}
1982 		glob_envp[glob_envn] = strdup(tokp);
1983 		if (glob_envp[glob_envn] == NULL) {
1984 			console(B_TRUE, "Out of memory building environment, "
1985 			    "truncated.\n");
1986 			break;
1987 		}
1988 		if (++glob_envn >= MAXENVENT - 1) {
1989 			console(B_TRUE, "Too many variables in %s; "
1990 			    "environment not fully initialized.\n", ENVFILE);
1991 			break;
1992 		}
1993 	}
1994 
1995 	/*
1996 	 * Append a null pointer to the environment array to mark its end.
1997 	 */
1998 	glob_envp[glob_envn] = NULL;
1999 
2000 	definit_close(dstate);
2001 }
2002 
2003 /*
2004  * boot_init(): Do initialization things that should be done at boot.
2005  */
2006 void
2007 boot_init()
2008 {
2009 	int i;
2010 	struct PROC_TABLE *process, *oprocess;
2011 	struct CMD_LINE	cmd;
2012 	char	line[MAXCMDL];
2013 	char	svc_aux[SVC_AUX_SIZE];
2014 	char	init_svc_fmri[SVC_FMRI_SIZE];
2015 	char *old_path;
2016 	int maxfiles;
2017 
2018 	/* Use INIT_PATH for sysinit cmds */
2019 	old_path = glob_envp[0];
2020 	glob_envp[0] = malloc((unsigned)(strlen(INIT_PATH)+2));
2021 	(void) strcpy(glob_envp[0], INIT_PATH);
2022 
2023 	/*
2024 	 * Scan inittab(5) and process the special svc.startd entry, initdefault
2025 	 * and sysinit entries.
2026 	 */
2027 	while (getcmd(&cmd, &line[0]) == TRUE) {
2028 		if (startd_tmpl >= 0 && id_eq(cmd.c_id, "smf")) {
2029 			process_startd_line(&cmd, line);
2030 			(void) snprintf(startd_svc_aux, SVC_AUX_SIZE,
2031 			    INITTAB_ENTRY_ID_STR_FORMAT, cmd.c_id);
2032 		} else if (cmd.c_action == M_INITDEFAULT) {
2033 			/*
2034 			 * initdefault is no longer meaningful, as the SMF
2035 			 * milestone controls what (legacy) run level we
2036 			 * boot to.
2037 			 */
2038 			console(B_TRUE,
2039 			    "Ignoring legacy \"initdefault\" entry.\n");
2040 		} else if (cmd.c_action == M_SYSINIT) {
2041 			/*
2042 			 * Execute the "sysinit" entry and wait for it to
2043 			 * complete.  No bookkeeping is performed on these
2044 			 * entries because we avoid writing to the file system
2045 			 * until after there has been an chance to check it.
2046 			 */
2047 			if (process = findpslot(&cmd)) {
2048 				(void) sighold(SIGCLD);
2049 				(void) snprintf(svc_aux, SVC_AUX_SIZE,
2050 				    INITTAB_ENTRY_ID_STR_FORMAT, cmd.c_id);
2051 				(void) snprintf(init_svc_fmri, SVC_FMRI_SIZE,
2052 				    SVC_INIT_PREFIX INITTAB_ENTRY_ID_STR_FORMAT,
2053 				    cmd.c_id);
2054 				if (legacy_tmpl >= 0) {
2055 					(void) ct_pr_tmpl_set_svc_fmri(
2056 					    legacy_tmpl, init_svc_fmri);
2057 					(void) ct_pr_tmpl_set_svc_aux(
2058 					    legacy_tmpl, svc_aux);
2059 				}
2060 
2061 				for (oprocess = process;
2062 				    (process = efork(M_OFF, oprocess,
2063 				    (NAMED|NOCLEANUP))) == NO_ROOM;
2064 				    /* CSTYLED */)
2065 					;
2066 				(void) sigrelse(SIGCLD);
2067 
2068 				if (process == NULLPROC) {
2069 					maxfiles = ulimit(UL_GDESLIM, 0);
2070 
2071 					for (i = 0; i < maxfiles; i++)
2072 						(void) fcntl(i, F_SETFD,
2073 						    FD_CLOEXEC);
2074 					(void) execle(SH, "INITSH", "-c",
2075 					    cmd.c_command,
2076 					    (char *)0, glob_envp);
2077 					console(B_TRUE,
2078 "Command\n\"%s\"\n failed to execute.  errno = %d (exec of shell failed)\n",
2079 					    cmd.c_command, errno);
2080 					exit(1);
2081 				} else
2082 					while (waitproc(process) == FAILURE)
2083 						;
2084 				process->p_flags = 0;
2085 				st_write();
2086 			}
2087 		}
2088 	}
2089 
2090 	/* Restore the path. */
2091 	free(glob_envp[0]);
2092 	glob_envp[0] = old_path;
2093 
2094 	/*
2095 	 * This will enable st_write() to complain about init_state_file.
2096 	 */
2097 	booting = 0;
2098 
2099 	/*
2100 	 * If the /etc/ioctl.syscon didn't exist or had invalid contents write
2101 	 * out a correct version.
2102 	 */
2103 	if (write_ioctl)
2104 		write_ioctl_syscon();
2105 
2106 	/*
2107 	 * Start svc.startd(8), which does most of the work.
2108 	 */
2109 	if (startd_cline[0] != '\0' && startd_tmpl >= 0) {
2110 		/* Start svc.startd. */
2111 		if (startd_run(startd_cline, startd_tmpl, 0) == -1)
2112 			cur_state = SINGLE_USER;
2113 	} else {
2114 		console(B_TRUE, "Absent svc.startd entry or bad "
2115 		    "contract template.  Not starting svc.startd.\n");
2116 		enter_maintenance();
2117 	}
2118 }
2119 
2120 /*
2121  * init_signals(): Initialize all signals to either be caught or ignored.
2122  */
2123 void
2124 init_signals(void)
2125 {
2126 	struct sigaction act;
2127 	int i;
2128 
2129 	/*
2130 	 * Start by ignoring all signals, then selectively re-enable some.
2131 	 * The SIG_IGN disposition will only affect asynchronous signals:
2132 	 * any signal that we trigger synchronously that doesn't end up
2133 	 * being handled by siglvl() will be forcibly delivered by the kernel.
2134 	 */
2135 	for (i = SIGHUP; i <= SIGRTMAX; i++)
2136 		(void) sigset(i, SIG_IGN);
2137 
2138 	/*
2139 	 * Handle all level-changing signals using siglvl() and set sa_mask so
2140 	 * that all level-changing signals are blocked while in siglvl().
2141 	 */
2142 	act.sa_handler = siglvl;
2143 	act.sa_flags = SA_SIGINFO;
2144 	(void) sigemptyset(&act.sa_mask);
2145 
2146 	(void) sigaddset(&act.sa_mask, LVLQ);
2147 	(void) sigaddset(&act.sa_mask, LVL0);
2148 	(void) sigaddset(&act.sa_mask, LVL1);
2149 	(void) sigaddset(&act.sa_mask, LVL2);
2150 	(void) sigaddset(&act.sa_mask, LVL3);
2151 	(void) sigaddset(&act.sa_mask, LVL4);
2152 	(void) sigaddset(&act.sa_mask, LVL5);
2153 	(void) sigaddset(&act.sa_mask, LVL6);
2154 	(void) sigaddset(&act.sa_mask, SINGLE_USER);
2155 	(void) sigaddset(&act.sa_mask, LVLa);
2156 	(void) sigaddset(&act.sa_mask, LVLb);
2157 	(void) sigaddset(&act.sa_mask, LVLc);
2158 
2159 	(void) sigaction(LVLQ, &act, NULL);
2160 	(void) sigaction(LVL0, &act, NULL);
2161 	(void) sigaction(LVL1, &act, NULL);
2162 	(void) sigaction(LVL2, &act, NULL);
2163 	(void) sigaction(LVL3, &act, NULL);
2164 	(void) sigaction(LVL4, &act, NULL);
2165 	(void) sigaction(LVL5, &act, NULL);
2166 	(void) sigaction(LVL6, &act, NULL);
2167 	(void) sigaction(SINGLE_USER, &act, NULL);
2168 	(void) sigaction(LVLa, &act, NULL);
2169 	(void) sigaction(LVLb, &act, NULL);
2170 	(void) sigaction(LVLc, &act, NULL);
2171 
2172 	(void) sigset(SIGALRM, alarmclk);
2173 	alarmclk();
2174 
2175 	(void) sigset(SIGCLD, childeath);
2176 	(void) sigset(SIGPWR, powerfail);
2177 }
2178 
2179 /*
2180  * Set up pipe for "godchildren". If the file exists and is a pipe just open
2181  * it. Else, if the file system is r/w create it.  Otherwise, defer its
2182  * creation and open until after /var/run has been mounted.  This function is
2183  * only called on startup and when explicitly requested via LVLQ.
2184  */
2185 void
2186 setup_pipe()
2187 {
2188 	struct stat stat_buf;
2189 	struct statvfs statvfs_buf;
2190 	struct sigaction act;
2191 
2192 	/*
2193 	 * Always close the previous pipe descriptor as the mounted filesystems
2194 	 * may have changed.
2195 	 */
2196 	if (Pfd >= 0)
2197 		(void) close(Pfd);
2198 
2199 	if ((stat(INITPIPE, &stat_buf) == 0) &&
2200 	    ((stat_buf.st_mode & (S_IFMT|S_IRUSR)) == (S_IFIFO|S_IRUSR)))
2201 		Pfd = open(INITPIPE, O_RDWR | O_NDELAY);
2202 	else
2203 		if ((statvfs(INITPIPE_DIR, &statvfs_buf) == 0) &&
2204 		    ((statvfs_buf.f_flag & ST_RDONLY) == 0)) {
2205 			(void) unlink(INITPIPE);
2206 			(void) mknod(INITPIPE, S_IFIFO | 0600, 0);
2207 			Pfd = open(INITPIPE, O_RDWR | O_NDELAY);
2208 		}
2209 
2210 	if (Pfd >= 0) {
2211 		(void) ioctl(Pfd, I_SETSIG, S_INPUT);
2212 		/*
2213 		 * Read pipe in message discard mode.
2214 		 */
2215 		(void) ioctl(Pfd, I_SRDOPT, RMSGD);
2216 
2217 		act.sa_handler = sigpoll;
2218 		act.sa_flags = 0;
2219 		(void) sigemptyset(&act.sa_mask);
2220 		(void) sigaddset(&act.sa_mask, SIGCLD);
2221 		(void) sigaction(SIGPOLL, &act, NULL);
2222 	}
2223 }
2224 
2225 /*
2226  * siglvl - handle an asynchronous signal from init(8) telling us that we
2227  * should change the current run level.  We set new_state accordingly.
2228  */
2229 void
2230 siglvl(int sig, siginfo_t *sip, ucontext_t *ucp)
2231 {
2232 	struct PROC_TABLE *process;
2233 	struct sigaction act;
2234 
2235 	/*
2236 	 * If the signal was from the kernel (rather than init(8)) then init
2237 	 * itself tripped the signal.  That is, we might have a bug and tripped
2238 	 * a real SIGSEGV instead of receiving it as an alias for SIGLVLa.  In
2239 	 * such a case we reset the disposition to SIG_DFL, block all signals
2240 	 * in uc_mask but the current one, and return to the interrupted ucp
2241 	 * to effect an appropriate death.  The kernel will then restart us.
2242 	 *
2243 	 * The one exception to SI_FROMKERNEL() is SIGFPE (a.k.a. LVL6), which
2244 	 * the kernel can send us when it wants to effect an orderly reboot.
2245 	 * For this case we must also verify si_code is zero, rather than a
2246 	 * code such as FPE_INTDIV which a bug might have triggered.
2247 	 */
2248 	if (sip != NULL && SI_FROMKERNEL(sip) &&
2249 	    (sig != SIGFPE || sip->si_code == 0)) {
2250 
2251 		(void) sigemptyset(&act.sa_mask);
2252 		act.sa_handler = SIG_DFL;
2253 		act.sa_flags = 0;
2254 		(void) sigaction(sig, &act, NULL);
2255 
2256 		(void) sigfillset(&ucp->uc_sigmask);
2257 		(void) sigdelset(&ucp->uc_sigmask, sig);
2258 		ucp->uc_flags |= UC_SIGMASK;
2259 
2260 		(void) setcontext(ucp);
2261 	}
2262 
2263 	/*
2264 	 * If the signal received is a LVLQ signal, do not really
2265 	 * change levels, just restate the current level.  If the
2266 	 * signal is not a LVLQ, set the new level to the signal
2267 	 * received.
2268 	 */
2269 	if (sig == LVLQ) {
2270 		new_state = cur_state;
2271 		lvlq_received = B_TRUE;
2272 	} else {
2273 		new_state = sig;
2274 	}
2275 
2276 	/*
2277 	 * Clear all times and repeat counts in the process table
2278 	 * since either the level is changing or the user has editted
2279 	 * the inittab file and wants us to look at it again.
2280 	 * If the user has fixed a typo, we don't want residual timing
2281 	 * data preventing the fixed command line from executing.
2282 	 */
2283 	for (process = proc_table;
2284 	    (process < proc_table + num_proc); process++) {
2285 		process->p_time = 0L;
2286 		process->p_count = 0;
2287 	}
2288 
2289 	/*
2290 	 * Set the flag to indicate that a "user signal" was received.
2291 	 */
2292 	wakeup.w_flags.w_usersignal = 1;
2293 }
2294 
2295 
2296 /*
2297  * alarmclk
2298  */
2299 static void
2300 alarmclk()
2301 {
2302 	time_up = TRUE;
2303 }
2304 
2305 /*
2306  * childeath_single():
2307  *
2308  * This used to be the SIGCLD handler and it was set with signal()
2309  * (as opposed to sigset()).  When a child exited we'd come to the
2310  * handler, wait for the child, and reenable the handler with
2311  * signal() just before returning.  The implementation of signal()
2312  * checks with waitid() for waitable children and sends a SIGCLD
2313  * if there are some.  If children are exiting faster than the
2314  * handler can run we keep sending signals and the handler never
2315  * gets to return and eventually the stack runs out and init dies.
2316  * To prevent that we set the handler with sigset() so the handler
2317  * doesn't need to be reset, and in childeath() (see below) we
2318  * call childeath_single() as long as there are children to be
2319  * waited for.  If a child exits while init is in the handler a
2320  * SIGCLD will be pending and delivered on return from the handler.
2321  * If the child was already waited for the handler will have nothing
2322  * to do and return, otherwise the child will be waited for.
2323  */
2324 static void
2325 childeath_single(pid_t pid, int status)
2326 {
2327 	struct PROC_TABLE	*process;
2328 	struct pidlist		*pp;
2329 
2330 	/*
2331 	 * Scan the process table to see if we are interested in this process.
2332 	 */
2333 	for (process = proc_table;
2334 	    (process < proc_table + num_proc); process++) {
2335 		if ((process->p_flags & (LIVING|OCCUPIED)) ==
2336 		    (LIVING|OCCUPIED) && process->p_pid == pid) {
2337 
2338 			/*
2339 			 * Mark this process as having died and store the exit
2340 			 * status.  Also set the wakeup flag for a dead child
2341 			 * and break out of the loop.
2342 			 */
2343 			process->p_flags &= ~LIVING;
2344 			process->p_exit = (short)status;
2345 			wakeup.w_flags.w_childdeath = 1;
2346 
2347 			return;
2348 		}
2349 	}
2350 
2351 	/*
2352 	 * No process was found above, look through auxiliary list.
2353 	 */
2354 	(void) sighold(SIGPOLL);
2355 	pp = Plhead;
2356 	while (pp) {
2357 		if (pid > pp->pl_pid) {
2358 			/*
2359 			 * Keep on looking.
2360 			 */
2361 			pp = pp->pl_next;
2362 			continue;
2363 		} else if (pid < pp->pl_pid) {
2364 			/*
2365 			 * Not in the list.
2366 			 */
2367 			break;
2368 		} else {
2369 			/*
2370 			 * This is a dead "godchild".
2371 			 */
2372 			pp->pl_dflag = 1;
2373 			pp->pl_exit = (short)status;
2374 			wakeup.w_flags.w_childdeath = 1;
2375 			Gchild = 1;	/* Notice to call cleanaux(). */
2376 			break;
2377 		}
2378 	}
2379 
2380 	(void) sigrelse(SIGPOLL);
2381 }
2382 
2383 /* ARGSUSED */
2384 static void
2385 childeath(int signo)
2386 {
2387 	pid_t pid;
2388 	int status;
2389 
2390 	while ((pid = waitpid(-1, &status, WNOHANG)) > 0)
2391 		childeath_single(pid, status);
2392 }
2393 
2394 static void
2395 powerfail()
2396 {
2397 	(void) nice(-19);
2398 	wakeup.w_flags.w_powerhit = 1;
2399 }
2400 
2401 /*
2402  * efork() forks a child and the parent inserts the process in its table
2403  * of processes that are directly a result of forks that it has performed.
2404  * The child just changes the "global" with the process id for this process
2405  * to it's new value.
2406  * If efork() is called with a pointer into the proc_table it uses that slot,
2407  * otherwise it searches for a free slot.  Regardless of how it was called,
2408  * it returns the pointer to the proc_table entry
2409  *
2410  * The SIGCLD signal is blocked (held) before calling efork()
2411  * and is unblocked (released) after efork() returns.
2412  *
2413  * Ideally, this should be rewritten to use modern signal semantics.
2414  */
2415 static struct PROC_TABLE *
2416 efork(int action, struct PROC_TABLE *process, int modes)
2417 {
2418 	pid_t	childpid;
2419 	struct PROC_TABLE *proc;
2420 	int		i;
2421 	/*
2422 	 * Freshen up the proc_table, removing any entries for dead processes
2423 	 * that don't have NOCLEANUP set.  Perform the necessary accounting.
2424 	 */
2425 	for (proc = proc_table; (proc < proc_table + num_proc); proc++) {
2426 		if ((proc->p_flags & (OCCUPIED|LIVING|NOCLEANUP)) ==
2427 		    (OCCUPIED)) {
2428 			/*
2429 			 * Is this a named process?
2430 			 * If so, do the necessary bookkeeping.
2431 			 */
2432 			if (proc->p_flags & NAMED)
2433 				(void) account(DEAD_PROCESS, proc, NULL);
2434 
2435 			/*
2436 			 * Free this entry for new usage.
2437 			 */
2438 			proc->p_flags = 0;
2439 		}
2440 	}
2441 
2442 	while ((childpid = fork()) == FAILURE) {
2443 		/*
2444 		 * Shorten the alarm timer in case someone else's child dies
2445 		 * and free up a slot in the process table.
2446 		 */
2447 		setimer(5);
2448 
2449 		/*
2450 		 * Wait for some children to die.  Since efork()
2451 		 * is always called with SIGCLD blocked, unblock
2452 		 * it here so that child death signals can come in.
2453 		 */
2454 		(void) sigrelse(SIGCLD);
2455 		(void) pause();
2456 		(void) sighold(SIGCLD);
2457 		setimer(0);
2458 	}
2459 
2460 	if (childpid != 0) {
2461 
2462 		if (process == NULLPROC) {
2463 			/*
2464 			 * No proc table pointer specified so search
2465 			 * for a free slot.
2466 			 */
2467 			for (process = proc_table;  process->p_flags != 0 &&
2468 			    (process < proc_table + num_proc); process++)
2469 					;
2470 
2471 			if (process == (proc_table + num_proc)) {
2472 				int old_proc_table_size = num_proc;
2473 
2474 				/* Increase the process table size */
2475 				increase_proc_table_size();
2476 				if (old_proc_table_size == num_proc) {
2477 					/* didn't grow: memory failure */
2478 					return (NO_ROOM);
2479 				} else {
2480 					process =
2481 					    proc_table + old_proc_table_size;
2482 				}
2483 			}
2484 
2485 			process->p_time = 0L;
2486 			process->p_count = 0;
2487 		}
2488 		process->p_id[0] = '\0';
2489 		process->p_id[1] = '\0';
2490 		process->p_id[2] = '\0';
2491 		process->p_id[3] = '\0';
2492 		process->p_pid = childpid;
2493 		process->p_flags = (LIVING | OCCUPIED | modes);
2494 		process->p_exit = 0;
2495 
2496 		st_write();
2497 	} else {
2498 		if ((action & (M_WAIT | M_BOOTWAIT)) == 0)
2499 			(void) setpgrp();
2500 
2501 		process = NULLPROC;
2502 
2503 		/*
2504 		 * Reset all signals to the system defaults.
2505 		 */
2506 		for (i = SIGHUP; i <= SIGRTMAX; i++)
2507 			(void) sigset(i, SIG_DFL);
2508 
2509 		/*
2510 		 * POSIX B.2.2.2 advises that init should set SIGTTOU,
2511 		 * SIGTTIN, and SIGTSTP to SIG_IGN.
2512 		 *
2513 		 * Make sure that SIGXCPU and SIGXFSZ also remain ignored,
2514 		 * for backward compatibility.
2515 		 */
2516 		(void) sigset(SIGTTIN, SIG_IGN);
2517 		(void) sigset(SIGTTOU, SIG_IGN);
2518 		(void) sigset(SIGTSTP, SIG_IGN);
2519 		(void) sigset(SIGXCPU, SIG_IGN);
2520 		(void) sigset(SIGXFSZ, SIG_IGN);
2521 	}
2522 	return (process);
2523 }
2524 
2525 
2526 /*
2527  * waitproc() waits for a specified process to die.  For this function to
2528  * work, the specified process must already in the proc_table.  waitproc()
2529  * returns the exit status of the specified process when it dies.
2530  */
2531 static long
2532 waitproc(struct PROC_TABLE *process)
2533 {
2534 	int		answer;
2535 	sigset_t	oldmask, newmask, zeromask;
2536 
2537 	(void) sigemptyset(&zeromask);
2538 	(void) sigemptyset(&newmask);
2539 
2540 	(void) sigaddset(&newmask, SIGCLD);
2541 
2542 	/* Block SIGCLD and save the current signal mask */
2543 	if (sigprocmask(SIG_BLOCK, &newmask, &oldmask) < 0)
2544 		perror("SIG_BLOCK error");
2545 
2546 	/*
2547 	 * Wait around until the process dies.
2548 	 */
2549 	if (process->p_flags & LIVING)
2550 		(void) sigsuspend(&zeromask);
2551 
2552 	/* Reset signal mask to unblock SIGCLD */
2553 	if (sigprocmask(SIG_SETMASK, &oldmask, NULL) < 0)
2554 		perror("SIG_SETMASK error");
2555 
2556 	if (process->p_flags & LIVING)
2557 		return (FAILURE);
2558 
2559 	/*
2560 	 * Make sure to only return 16 bits so that answer will always
2561 	 * be positive whenever the process of interest really died.
2562 	 */
2563 	answer = (process->p_exit & 0xffff);
2564 
2565 	/*
2566 	 * Free the slot in the proc_table.
2567 	 */
2568 	process->p_flags = 0;
2569 	return (answer);
2570 }
2571 
2572 /*
2573  * notify_pam_dead(): calls into the PAM framework to close the given session.
2574  */
2575 static void
2576 notify_pam_dead(struct utmpx *up)
2577 {
2578 	pam_handle_t *pamh;
2579 	char user[sizeof (up->ut_user) + 1];
2580 	char ttyn[sizeof (up->ut_line) + 1];
2581 	char host[sizeof (up->ut_host) + 1];
2582 
2583 	/*
2584 	 * PAM does not take care of updating utmpx/wtmpx.
2585 	 */
2586 	(void) snprintf(user, sizeof (user), "%s", up->ut_user);
2587 	(void) snprintf(ttyn, sizeof (ttyn), "%s", up->ut_line);
2588 	(void) snprintf(host, sizeof (host), "%s", up->ut_host);
2589 
2590 	if (pam_start("init", user, NULL, &pamh) == PAM_SUCCESS)  {
2591 		(void) pam_set_item(pamh, PAM_TTY, ttyn);
2592 		(void) pam_set_item(pamh, PAM_RHOST, host);
2593 		(void) pam_close_session(pamh, 0);
2594 		(void) pam_end(pamh, PAM_SUCCESS);
2595 	}
2596 }
2597 
2598 /*
2599  * Check you can access utmpx (As / may be read-only and
2600  * /var may not be mounted yet).
2601  */
2602 static int
2603 access_utmpx(void)
2604 {
2605 	do {
2606 		utmpx_ok = (access(UTMPX, R_OK|W_OK) == 0);
2607 	} while (!utmpx_ok && errno == EINTR);
2608 
2609 	return (utmpx_ok);
2610 }
2611 
2612 /*
2613  * account() updates entries in utmpx and appends new entries to the end of
2614  * wtmpx (assuming they exist).  The program argument indicates the name of
2615  * program if INIT_PROCESS, otherwise should be NULL.
2616  *
2617  * account() only blocks for INIT_PROCESS requests.
2618  *
2619  * Returns non-zero if write failed.
2620  */
2621 static int
2622 account(short state, struct PROC_TABLE *process, char *program)
2623 {
2624 	struct utmpx utmpbuf, *u, *oldu;
2625 	int tmplen;
2626 	char fail_buf[UT_LINE_SZ];
2627 	sigset_t block, unblock;
2628 
2629 	if (!utmpx_ok && !access_utmpx()) {
2630 		return (-1);
2631 	}
2632 
2633 	/*
2634 	 * Set up the prototype for the utmp structure we want to write.
2635 	 */
2636 	u = &utmpbuf;
2637 	(void) memset(u, 0, sizeof (struct utmpx));
2638 
2639 	/*
2640 	 * Fill in the various fields of the utmp structure.
2641 	 */
2642 	u->ut_id[0] = process->p_id[0];
2643 	u->ut_id[1] = process->p_id[1];
2644 	u->ut_id[2] = process->p_id[2];
2645 	u->ut_id[3] = process->p_id[3];
2646 	u->ut_pid = process->p_pid;
2647 
2648 	/*
2649 	 * Fill the "ut_exit" structure.
2650 	 */
2651 	u->ut_exit.e_termination = WTERMSIG(process->p_exit);
2652 	u->ut_exit.e_exit = WEXITSTATUS(process->p_exit);
2653 	u->ut_type = state;
2654 
2655 	(void) time(&u->ut_tv.tv_sec);
2656 
2657 	/*
2658 	 * Block signals for utmp update.
2659 	 */
2660 	(void) sigfillset(&block);
2661 	(void) sigprocmask(SIG_BLOCK, &block, &unblock);
2662 
2663 	/*
2664 	 * See if there already is such an entry in the "utmpx" file.
2665 	 */
2666 	setutxent();	/* Start at beginning of utmpx file. */
2667 
2668 	if ((oldu = getutxid(u)) != NULL) {
2669 		/*
2670 		 * Copy in the old "user", "line" and "host" fields
2671 		 * to our new structure.
2672 		 */
2673 		bcopy(oldu->ut_user, u->ut_user, sizeof (u->ut_user));
2674 		bcopy(oldu->ut_line, u->ut_line, sizeof (u->ut_line));
2675 		bcopy(oldu->ut_host, u->ut_host, sizeof (u->ut_host));
2676 		u->ut_syslen = (tmplen = strlen(u->ut_host)) ?
2677 		    min(tmplen + 1, sizeof (u->ut_host)) : 0;
2678 
2679 		if (oldu->ut_type == USER_PROCESS && state == DEAD_PROCESS) {
2680 			notify_pam_dead(oldu);
2681 		}
2682 	}
2683 
2684 	/*
2685 	 * Perform special accounting. Insert the special string into the
2686 	 * ut_line array. For INIT_PROCESSes put in the name of the
2687 	 * program in the "ut_user" field.
2688 	 */
2689 	switch (state) {
2690 	case INIT_PROCESS:
2691 		(void) strncpy(u->ut_user, program, sizeof (u->ut_user));
2692 		(void) strcpy(fail_buf, "INIT_PROCESS");
2693 		break;
2694 
2695 	default:
2696 		(void) strlcpy(fail_buf, u->ut_id, sizeof (u->ut_id) + 1);
2697 		break;
2698 	}
2699 
2700 	/*
2701 	 * Write out the updated entry to utmpx file.
2702 	 */
2703 	if (pututxline(u) == NULL) {
2704 		console(B_TRUE, "Failed write of utmpx entry: \"%s\": %s\n",
2705 		    fail_buf, strerror(errno));
2706 		endutxent();
2707 		(void) sigprocmask(SIG_SETMASK, &unblock, NULL);
2708 		return (-1);
2709 	}
2710 
2711 	/*
2712 	 * If we're able to write to utmpx, then attempt to add to the
2713 	 * end of the wtmpx file.
2714 	 */
2715 	updwtmpx(WTMPX, u);
2716 
2717 	endutxent();
2718 
2719 	(void) sigprocmask(SIG_SETMASK, &unblock, NULL);
2720 
2721 	return (0);
2722 }
2723 
2724 static void
2725 clearent(pid_t pid, short status)
2726 {
2727 	struct utmpx *up;
2728 	sigset_t block, unblock;
2729 
2730 	/*
2731 	 * Block signals for utmp update.
2732 	 */
2733 	(void) sigfillset(&block);
2734 	(void) sigprocmask(SIG_BLOCK, &block, &unblock);
2735 
2736 	/*
2737 	 * No error checking for now.
2738 	 */
2739 
2740 	setutxent();
2741 	while (up = getutxent()) {
2742 		if (up->ut_pid == pid) {
2743 			if (up->ut_type == DEAD_PROCESS) {
2744 				/*
2745 				 * Cleaned up elsewhere.
2746 				 */
2747 				continue;
2748 			}
2749 
2750 			notify_pam_dead(up);
2751 
2752 			up->ut_type = DEAD_PROCESS;
2753 			up->ut_exit.e_termination = WTERMSIG(status);
2754 			up->ut_exit.e_exit = WEXITSTATUS(status);
2755 			(void) time(&up->ut_tv.tv_sec);
2756 
2757 			(void) pututxline(up);
2758 			/*
2759 			 * Now attempt to add to the end of the
2760 			 * wtmp and wtmpx files.  Do not create
2761 			 * if they don't already exist.
2762 			 */
2763 			updwtmpx(WTMPX, up);
2764 
2765 			break;
2766 		}
2767 	}
2768 
2769 	endutxent();
2770 	(void) sigprocmask(SIG_SETMASK, &unblock, NULL);
2771 }
2772 
2773 /*
2774  * prog_name() searches for the word or unix path name and
2775  * returns a pointer to the last element of the pathname.
2776  */
2777 static char *
2778 prog_name(char *string)
2779 {
2780 	char	*ptr, *ptr2;
2781 	static char word[UT_USER_SZ + 1];
2782 
2783 	/*
2784 	 * Search for the first word skipping leading spaces and tabs.
2785 	 */
2786 	while (*string == ' ' || *string == '\t')
2787 		string++;
2788 
2789 	/*
2790 	 * If the first non-space non-tab character is not one allowed in
2791 	 * a word, return a pointer to a null string, otherwise parse the
2792 	 * pathname.
2793 	 */
2794 	if (*string != '.' && *string != '/' && *string != '_' &&
2795 	    (*string < 'a' || *string > 'z') &&
2796 	    (*string < 'A' || * string > 'Z') &&
2797 	    (*string < '0' || *string > '9'))
2798 		return ("");
2799 
2800 	/*
2801 	 * Parse the pathname looking forward for '/', ' ', '\t', '\n' or
2802 	 * '\0'.  Each time a '/' is found, move "ptr" to one past the
2803 	 * '/', thus when a ' ', '\t', '\n', or '\0' is found, "ptr" will
2804 	 * point to the last element of the pathname.
2805 	 */
2806 	for (ptr = string; *string != ' ' && *string != '\t' &&
2807 	    *string != '\n' && *string != '\0'; string++) {
2808 		if (*string == '/')
2809 			ptr = string+1;
2810 	}
2811 
2812 	/*
2813 	 * Copy out up to the size of the "ut_user" array into "word",
2814 	 * null terminate it and return a pointer to it.
2815 	 */
2816 	for (ptr2 = &word[0]; ptr2 < &word[UT_USER_SZ] &&
2817 	    ptr < string; /* CSTYLED */)
2818 		*ptr2++ = *ptr++;
2819 
2820 	*ptr2 = '\0';
2821 	return (&word[0]);
2822 }
2823 
2824 
2825 /*
2826  * realcon() returns a nonzero value if there is a character device
2827  * associated with SYSCON that has the same device number as CONSOLE.
2828  */
2829 static int
2830 realcon()
2831 {
2832 	struct stat sconbuf, conbuf;
2833 
2834 	if (stat(SYSCON, &sconbuf) != -1 &&
2835 	    stat(CONSOLE, &conbuf) != -1 &&
2836 	    S_ISCHR(sconbuf.st_mode) &&
2837 	    S_ISCHR(conbuf.st_mode) &&
2838 	    sconbuf.st_rdev == conbuf.st_rdev) {
2839 		return (1);
2840 	} else {
2841 		return (0);
2842 	}
2843 }
2844 
2845 
2846 /*
2847  * get_ioctl_syscon() retrieves the SYSCON settings from the IOCTLSYSCON file.
2848  * Returns true if the IOCTLSYSCON file needs to be written (with
2849  * write_ioctl_syscon() below)
2850  */
2851 static int
2852 get_ioctl_syscon()
2853 {
2854 	FILE	*fp;
2855 	unsigned int	iflags, oflags, cflags, lflags, ldisc, cc[18];
2856 	int		i, valid_format = 0;
2857 
2858 	/*
2859 	 * Read in the previous modes for SYSCON from IOCTLSYSCON.
2860 	 */
2861 	if ((fp = fopen(IOCTLSYSCON, "r")) == NULL) {
2862 		stored_syscon_termios = dflt_termios;
2863 		console(B_TRUE,
2864 		    "warning:%s does not exist, default settings assumed\n",
2865 		    IOCTLSYSCON);
2866 	} else {
2867 
2868 		i = fscanf(fp,
2869 	    "%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x",
2870 		    &iflags, &oflags, &cflags, &lflags,
2871 		    &cc[0], &cc[1], &cc[2], &cc[3], &cc[4], &cc[5], &cc[6],
2872 		    &cc[7], &cc[8], &cc[9], &cc[10], &cc[11], &cc[12], &cc[13],
2873 		    &cc[14], &cc[15], &cc[16], &cc[17]);
2874 
2875 		if (i == 22) {
2876 			stored_syscon_termios.c_iflag = iflags;
2877 			stored_syscon_termios.c_oflag = oflags;
2878 			stored_syscon_termios.c_cflag = cflags;
2879 			stored_syscon_termios.c_lflag = lflags;
2880 			for (i = 0; i < 18; i++)
2881 				stored_syscon_termios.c_cc[i] = (char)cc[i];
2882 			valid_format = 1;
2883 		} else if (i == 13) {
2884 		rewind(fp);
2885 		i = fscanf(fp, "%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x:%x",
2886 		    &iflags, &oflags, &cflags, &lflags, &ldisc, &cc[0], &cc[1],
2887 		    &cc[2], &cc[3], &cc[4], &cc[5], &cc[6], &cc[7]);
2888 
2889 		/*
2890 		 * If the file is formatted properly, use the values to
2891 		 * initialize the console terminal condition.
2892 		 */
2893 		stored_syscon_termios.c_iflag = (ushort_t)iflags;
2894 		stored_syscon_termios.c_oflag = (ushort_t)oflags;
2895 		stored_syscon_termios.c_cflag = (ushort_t)cflags;
2896 		stored_syscon_termios.c_lflag = (ushort_t)lflags;
2897 		for (i = 0; i < 8; i++)
2898 			stored_syscon_termios.c_cc[i] = (char)cc[i];
2899 		valid_format = 1;
2900 		}
2901 		(void) fclose(fp);
2902 
2903 		/* If the file is badly formatted, use the default settings. */
2904 		if (!valid_format)
2905 			stored_syscon_termios = dflt_termios;
2906 	}
2907 
2908 	/* If the file had a bad format, rewrite it later. */
2909 	return (!valid_format);
2910 }
2911 
2912 
2913 static void
2914 write_ioctl_syscon()
2915 {
2916 	FILE *fp;
2917 	int i;
2918 
2919 	(void) unlink(SYSCON);
2920 	(void) link(SYSTTY, SYSCON);
2921 	(void) umask(022);
2922 	fp = fopen(IOCTLSYSCON, "w");
2923 
2924 	(void) fprintf(fp, "%x:%x:%x:%x:0", stored_syscon_termios.c_iflag,
2925 	    stored_syscon_termios.c_oflag, stored_syscon_termios.c_cflag,
2926 	    stored_syscon_termios.c_lflag);
2927 	for (i = 0; i < 8; ++i)
2928 		(void) fprintf(fp, ":%x", stored_syscon_termios.c_cc[i]);
2929 	(void) putc('\n', fp);
2930 
2931 	(void) fflush(fp);
2932 	(void) fsync(fileno(fp));
2933 	(void) fclose(fp);
2934 	(void) umask(cmask);
2935 }
2936 
2937 
2938 /*
2939  * void console(boolean_t, char *, ...)
2940  *   Outputs the requested message to the system console.  Note that the number
2941  *   of arguments passed to console() should be determined by the print format.
2942  *
2943  *   The "prefix" parameter indicates whether or not "INIT: " should precede the
2944  *   message.
2945  *
2946  *   To make sure we write to the console in a sane fashion, we use the modes
2947  *   we keep in stored_syscon_termios (which we read out of /etc/ioctl.syscon).
2948  *   Afterwards we restore whatever modes were already there.
2949  */
2950 /* PRINTFLIKE2 */
2951 static void
2952 console(boolean_t prefix, char *format, ...)
2953 {
2954 	char	outbuf[BUFSIZ];
2955 	va_list	args;
2956 	int fd, getret;
2957 	struct termios old_syscon_termios;
2958 	FILE *f;
2959 
2960 	/*
2961 	 * We open SYSCON anew each time in case it has changed (see
2962 	 * userinit()).
2963 	 */
2964 	if ((fd = open(SYSCON, O_RDWR | O_NOCTTY)) < 0 ||
2965 	    (f = fdopen(fd, "r+")) == NULL) {
2966 		if (prefix)
2967 			syslog(LOG_WARNING, "INIT: ");
2968 		va_start(args, format);
2969 		vsyslog(LOG_WARNING, format, args);
2970 		va_end(args);
2971 		if (fd >= 0)
2972 			(void) close(fd);
2973 		return;
2974 	}
2975 	setbuf(f, &outbuf[0]);
2976 
2977 	getret = tcgetattr(fd, &old_syscon_termios);
2978 	old_syscon_termios.c_cflag &= ~HUPCL;
2979 	if (realcon())
2980 		/* Don't overwrite cflag of real console. */
2981 		stored_syscon_termios.c_cflag = old_syscon_termios.c_cflag;
2982 
2983 	stored_syscon_termios.c_cflag &= ~HUPCL;
2984 
2985 	(void) tcsetattr(fd, TCSANOW, &stored_syscon_termios);
2986 
2987 	if (prefix)
2988 		(void) fprintf(f, "\nINIT: ");
2989 	va_start(args, format);
2990 	(void) vfprintf(f, format, args);
2991 	va_end(args);
2992 
2993 	if (getret == 0)
2994 		(void) tcsetattr(fd, TCSADRAIN, &old_syscon_termios);
2995 
2996 	(void) fclose(f);
2997 }
2998 
2999 /*
3000  * timer() is a substitute for sleep() which uses alarm() and pause().
3001  */
3002 static void
3003 timer(int waitime)
3004 {
3005 	setimer(waitime);
3006 	while (time_up == FALSE)
3007 		(void) pause();
3008 }
3009 
3010 static void
3011 setimer(int timelimit)
3012 {
3013 	alarmclk();
3014 	(void) alarm(timelimit);
3015 	time_up = (timelimit ? FALSE : TRUE);
3016 }
3017 
3018 /*
3019  * Fails with
3020  *   ENOMEM - out of memory
3021  *   ECONNABORTED - repository connection broken
3022  *   EPERM - permission denied
3023  *   EACCES - backend access denied
3024  *   EROFS - backend readonly
3025  */
3026 static int
3027 get_or_add_startd(scf_instance_t *inst)
3028 {
3029 	scf_handle_t *h;
3030 	scf_scope_t *scope = NULL;
3031 	scf_service_t *svc = NULL;
3032 	int ret = 0;
3033 
3034 	h = scf_instance_handle(inst);
3035 
3036 	if (scf_handle_decode_fmri(h, SCF_SERVICE_STARTD, NULL, NULL, inst,
3037 	    NULL, NULL, SCF_DECODE_FMRI_EXACT) == 0)
3038 		return (0);
3039 
3040 	switch (scf_error()) {
3041 	case SCF_ERROR_CONNECTION_BROKEN:
3042 		return (ECONNABORTED);
3043 
3044 	case SCF_ERROR_NOT_FOUND:
3045 		break;
3046 
3047 	case SCF_ERROR_HANDLE_MISMATCH:
3048 	case SCF_ERROR_INVALID_ARGUMENT:
3049 	case SCF_ERROR_CONSTRAINT_VIOLATED:
3050 	default:
3051 		bad_error("scf_handle_decode_fmri", scf_error());
3052 	}
3053 
3054 	/* Make sure we're right, since we're adding piece-by-piece. */
3055 	assert(strcmp(SCF_SERVICE_STARTD,
3056 	    "svc:/system/svc/restarter:default") == 0);
3057 
3058 	if ((scope = scf_scope_create(h)) == NULL ||
3059 	    (svc = scf_service_create(h)) == NULL) {
3060 		ret = ENOMEM;
3061 		goto out;
3062 	}
3063 
3064 get_scope:
3065 	if (scf_handle_get_scope(h, SCF_SCOPE_LOCAL, scope) != 0) {
3066 		switch (scf_error()) {
3067 		case SCF_ERROR_CONNECTION_BROKEN:
3068 			ret = ECONNABORTED;
3069 			goto out;
3070 
3071 		case SCF_ERROR_NOT_FOUND:
3072 			(void) fputs(gettext(
3073 			    "smf(7) repository missing local scope.\n"),
3074 			    stderr);
3075 			exit(1);
3076 			/* NOTREACHED */
3077 
3078 		case SCF_ERROR_HANDLE_MISMATCH:
3079 		case SCF_ERROR_INVALID_ARGUMENT:
3080 		default:
3081 			bad_error("scf_handle_get_scope", scf_error());
3082 		}
3083 	}
3084 
3085 get_svc:
3086 	if (scf_scope_get_service(scope, "system/svc/restarter", svc) != 0) {
3087 		switch (scf_error()) {
3088 		case SCF_ERROR_CONNECTION_BROKEN:
3089 			ret = ECONNABORTED;
3090 			goto out;
3091 
3092 		case SCF_ERROR_DELETED:
3093 			goto get_scope;
3094 
3095 		case SCF_ERROR_NOT_FOUND:
3096 			break;
3097 
3098 		case SCF_ERROR_HANDLE_MISMATCH:
3099 		case SCF_ERROR_INVALID_ARGUMENT:
3100 		case SCF_ERROR_NOT_SET:
3101 		default:
3102 			bad_error("scf_scope_get_service", scf_error());
3103 		}
3104 
3105 add_svc:
3106 		if (scf_scope_add_service(scope, "system/svc/restarter", svc) !=
3107 		    0) {
3108 			switch (scf_error()) {
3109 			case SCF_ERROR_CONNECTION_BROKEN:
3110 				ret = ECONNABORTED;
3111 				goto out;
3112 
3113 			case SCF_ERROR_EXISTS:
3114 				goto get_svc;
3115 
3116 			case SCF_ERROR_PERMISSION_DENIED:
3117 				ret = EPERM;
3118 				goto out;
3119 
3120 			case SCF_ERROR_BACKEND_ACCESS:
3121 				ret = EACCES;
3122 				goto out;
3123 
3124 			case SCF_ERROR_BACKEND_READONLY:
3125 				ret = EROFS;
3126 				goto out;
3127 
3128 			case SCF_ERROR_HANDLE_MISMATCH:
3129 			case SCF_ERROR_INVALID_ARGUMENT:
3130 			case SCF_ERROR_NOT_SET:
3131 			default:
3132 				bad_error("scf_scope_add_service", scf_error());
3133 			}
3134 		}
3135 	}
3136 
3137 get_inst:
3138 	if (scf_service_get_instance(svc, "default", inst) != 0) {
3139 		switch (scf_error()) {
3140 		case SCF_ERROR_CONNECTION_BROKEN:
3141 			ret = ECONNABORTED;
3142 			goto out;
3143 
3144 		case SCF_ERROR_DELETED:
3145 			goto add_svc;
3146 
3147 		case SCF_ERROR_NOT_FOUND:
3148 			break;
3149 
3150 		case SCF_ERROR_HANDLE_MISMATCH:
3151 		case SCF_ERROR_INVALID_ARGUMENT:
3152 		case SCF_ERROR_NOT_SET:
3153 		default:
3154 			bad_error("scf_service_get_instance", scf_error());
3155 		}
3156 
3157 		if (scf_service_add_instance(svc, "default", inst) !=
3158 		    0) {
3159 			switch (scf_error()) {
3160 			case SCF_ERROR_CONNECTION_BROKEN:
3161 				ret = ECONNABORTED;
3162 				goto out;
3163 
3164 			case SCF_ERROR_DELETED:
3165 				goto add_svc;
3166 
3167 			case SCF_ERROR_EXISTS:
3168 				goto get_inst;
3169 
3170 			case SCF_ERROR_PERMISSION_DENIED:
3171 				ret = EPERM;
3172 				goto out;
3173 
3174 			case SCF_ERROR_BACKEND_ACCESS:
3175 				ret = EACCES;
3176 				goto out;
3177 
3178 			case SCF_ERROR_BACKEND_READONLY:
3179 				ret = EROFS;
3180 				goto out;
3181 
3182 			case SCF_ERROR_HANDLE_MISMATCH:
3183 			case SCF_ERROR_INVALID_ARGUMENT:
3184 			case SCF_ERROR_NOT_SET:
3185 			default:
3186 				bad_error("scf_service_add_instance",
3187 				    scf_error());
3188 			}
3189 		}
3190 	}
3191 
3192 	ret = 0;
3193 
3194 out:
3195 	scf_service_destroy(svc);
3196 	scf_scope_destroy(scope);
3197 	return (ret);
3198 }
3199 
3200 /*
3201  * Fails with
3202  *   ECONNABORTED - repository connection broken
3203  *   ECANCELED - the transaction's property group was deleted
3204  */
3205 static int
3206 transaction_add_set(scf_transaction_t *tx, scf_transaction_entry_t *ent,
3207     const char *pname, scf_type_t type)
3208 {
3209 change_type:
3210 	if (scf_transaction_property_change_type(tx, ent, pname, type) == 0)
3211 		return (0);
3212 
3213 	switch (scf_error()) {
3214 	case SCF_ERROR_CONNECTION_BROKEN:
3215 		return (ECONNABORTED);
3216 
3217 	case SCF_ERROR_DELETED:
3218 		return (ECANCELED);
3219 
3220 	case SCF_ERROR_NOT_FOUND:
3221 		goto new;
3222 
3223 	case SCF_ERROR_HANDLE_MISMATCH:
3224 	case SCF_ERROR_INVALID_ARGUMENT:
3225 	case SCF_ERROR_NOT_BOUND:
3226 	case SCF_ERROR_NOT_SET:
3227 	default:
3228 		bad_error("scf_transaction_property_change_type", scf_error());
3229 	}
3230 
3231 new:
3232 	if (scf_transaction_property_new(tx, ent, pname, type) == 0)
3233 		return (0);
3234 
3235 	switch (scf_error()) {
3236 	case SCF_ERROR_CONNECTION_BROKEN:
3237 		return (ECONNABORTED);
3238 
3239 	case SCF_ERROR_DELETED:
3240 		return (ECANCELED);
3241 
3242 	case SCF_ERROR_EXISTS:
3243 		goto change_type;
3244 
3245 	case SCF_ERROR_HANDLE_MISMATCH:
3246 	case SCF_ERROR_INVALID_ARGUMENT:
3247 	case SCF_ERROR_NOT_BOUND:
3248 	case SCF_ERROR_NOT_SET:
3249 	default:
3250 		bad_error("scf_transaction_property_new", scf_error());
3251 		/* NOTREACHED */
3252 	}
3253 }
3254 
3255 static void
3256 scferr(void)
3257 {
3258 	switch (scf_error()) {
3259 	case SCF_ERROR_NO_MEMORY:
3260 		console(B_TRUE, gettext("Out of memory.\n"));
3261 		break;
3262 
3263 	case SCF_ERROR_CONNECTION_BROKEN:
3264 		console(B_TRUE, gettext(
3265 		    "Connection to smf(7) repository server broken.\n"));
3266 		break;
3267 
3268 	case SCF_ERROR_NO_RESOURCES:
3269 		console(B_TRUE, gettext(
3270 		    "smf(7) repository server is out of memory.\n"));
3271 		break;
3272 
3273 	case SCF_ERROR_PERMISSION_DENIED:
3274 		console(B_TRUE, gettext("Insufficient privileges.\n"));
3275 		break;
3276 
3277 	default:
3278 		console(B_TRUE, gettext("libscf error: %s\n"),
3279 		    scf_strerror(scf_error()));
3280 	}
3281 }
3282 
3283 static void
3284 lscf_set_runlevel(char rl)
3285 {
3286 	scf_handle_t *h;
3287 	scf_instance_t *inst = NULL;
3288 	scf_propertygroup_t *pg = NULL;
3289 	scf_transaction_t *tx = NULL;
3290 	scf_transaction_entry_t *ent = NULL;
3291 	scf_value_t *val = NULL;
3292 	char buf[2];
3293 	int r;
3294 
3295 	h = scf_handle_create(SCF_VERSION);
3296 	if (h == NULL) {
3297 		scferr();
3298 		return;
3299 	}
3300 
3301 	if (scf_handle_bind(h) != 0) {
3302 		switch (scf_error()) {
3303 		case SCF_ERROR_NO_SERVER:
3304 			console(B_TRUE,
3305 			    gettext("smf(7) repository server not running.\n"));
3306 			goto bail;
3307 
3308 		default:
3309 			scferr();
3310 			goto bail;
3311 		}
3312 	}
3313 
3314 	if ((inst = scf_instance_create(h)) == NULL ||
3315 	    (pg = scf_pg_create(h)) == NULL ||
3316 	    (val = scf_value_create(h)) == NULL ||
3317 	    (tx = scf_transaction_create(h)) == NULL ||
3318 	    (ent = scf_entry_create(h)) == NULL) {
3319 		scferr();
3320 		goto bail;
3321 	}
3322 
3323 get_inst:
3324 	r = get_or_add_startd(inst);
3325 	switch (r) {
3326 	case 0:
3327 		break;
3328 
3329 	case ENOMEM:
3330 	case ECONNABORTED:
3331 	case EPERM:
3332 	case EACCES:
3333 	case EROFS:
3334 		scferr();
3335 		goto bail;
3336 	default:
3337 		bad_error("get_or_add_startd", r);
3338 	}
3339 
3340 get_pg:
3341 	if (scf_instance_get_pg(inst, SCF_PG_OPTIONS_OVR, pg) != 0) {
3342 		switch (scf_error()) {
3343 		case SCF_ERROR_CONNECTION_BROKEN:
3344 			scferr();
3345 			goto bail;
3346 
3347 		case SCF_ERROR_DELETED:
3348 			goto get_inst;
3349 
3350 		case SCF_ERROR_NOT_FOUND:
3351 			break;
3352 
3353 		case SCF_ERROR_HANDLE_MISMATCH:
3354 		case SCF_ERROR_INVALID_ARGUMENT:
3355 		case SCF_ERROR_NOT_SET:
3356 		default:
3357 			bad_error("scf_instance_get_pg", scf_error());
3358 		}
3359 
3360 add_pg:
3361 		if (scf_instance_add_pg(inst, SCF_PG_OPTIONS_OVR,
3362 		    SCF_PG_OPTIONS_OVR_TYPE, SCF_PG_OPTIONS_OVR_FLAGS, pg) !=
3363 		    0) {
3364 			switch (scf_error()) {
3365 			case SCF_ERROR_CONNECTION_BROKEN:
3366 			case SCF_ERROR_PERMISSION_DENIED:
3367 			case SCF_ERROR_BACKEND_ACCESS:
3368 				scferr();
3369 				goto bail;
3370 
3371 			case SCF_ERROR_DELETED:
3372 				goto get_inst;
3373 
3374 			case SCF_ERROR_EXISTS:
3375 				goto get_pg;
3376 
3377 			case SCF_ERROR_HANDLE_MISMATCH:
3378 			case SCF_ERROR_INVALID_ARGUMENT:
3379 			case SCF_ERROR_NOT_SET:
3380 			default:
3381 				bad_error("scf_instance_add_pg", scf_error());
3382 			}
3383 		}
3384 	}
3385 
3386 	buf[0] = rl;
3387 	buf[1] = '\0';
3388 	r = scf_value_set_astring(val, buf);
3389 	assert(r == 0);
3390 
3391 	for (;;) {
3392 		if (scf_transaction_start(tx, pg) != 0) {
3393 			switch (scf_error()) {
3394 			case SCF_ERROR_CONNECTION_BROKEN:
3395 			case SCF_ERROR_PERMISSION_DENIED:
3396 			case SCF_ERROR_BACKEND_ACCESS:
3397 				scferr();
3398 				goto bail;
3399 
3400 			case SCF_ERROR_DELETED:
3401 				goto add_pg;
3402 
3403 			case SCF_ERROR_HANDLE_MISMATCH:
3404 			case SCF_ERROR_NOT_BOUND:
3405 			case SCF_ERROR_IN_USE:
3406 			case SCF_ERROR_NOT_SET:
3407 			default:
3408 				bad_error("scf_transaction_start", scf_error());
3409 			}
3410 		}
3411 
3412 		r = transaction_add_set(tx, ent, "runlevel", SCF_TYPE_ASTRING);
3413 		switch (r) {
3414 		case 0:
3415 			break;
3416 
3417 		case ECONNABORTED:
3418 			scferr();
3419 			goto bail;
3420 
3421 		case ECANCELED:
3422 			scf_transaction_reset(tx);
3423 			goto add_pg;
3424 
3425 		default:
3426 			bad_error("transaction_add_set", r);
3427 		}
3428 
3429 		r = scf_entry_add_value(ent, val);
3430 		assert(r == 0);
3431 
3432 		r = scf_transaction_commit(tx);
3433 		if (r == 1)
3434 			break;
3435 
3436 		if (r != 0) {
3437 			switch (scf_error()) {
3438 			case SCF_ERROR_CONNECTION_BROKEN:
3439 			case SCF_ERROR_PERMISSION_DENIED:
3440 			case SCF_ERROR_BACKEND_ACCESS:
3441 			case SCF_ERROR_BACKEND_READONLY:
3442 				scferr();
3443 				goto bail;
3444 
3445 			case SCF_ERROR_DELETED:
3446 				scf_transaction_reset(tx);
3447 				goto add_pg;
3448 
3449 			case SCF_ERROR_INVALID_ARGUMENT:
3450 			case SCF_ERROR_NOT_BOUND:
3451 			case SCF_ERROR_NOT_SET:
3452 			default:
3453 				bad_error("scf_transaction_commit",
3454 				    scf_error());
3455 			}
3456 		}
3457 
3458 		scf_transaction_reset(tx);
3459 		(void) scf_pg_update(pg);
3460 	}
3461 
3462 bail:
3463 	scf_transaction_destroy(tx);
3464 	scf_entry_destroy(ent);
3465 	scf_value_destroy(val);
3466 	scf_pg_destroy(pg);
3467 	scf_instance_destroy(inst);
3468 
3469 	(void) scf_handle_unbind(h);
3470 	scf_handle_destroy(h);
3471 }
3472 
3473 /*
3474  * Function to handle requests from users to main init running as process 1.
3475  */
3476 static void
3477 userinit(int argc, char **argv)
3478 {
3479 	FILE	*fp;
3480 	char	*ln;
3481 	int	init_signal;
3482 	struct stat	sconbuf, conbuf;
3483 	const char *usage_msg = "Usage: init [0123456SsQqabc]\n";
3484 
3485 	/*
3486 	 * We are a user invoked init.  Is there an argument and is it
3487 	 * a single character?  If not, print usage message and quit.
3488 	 */
3489 	if (argc != 2 || argv[1][1] != '\0') {
3490 		(void) fprintf(stderr, usage_msg);
3491 		exit(0);
3492 	}
3493 
3494 	if ((init_signal = lvlname_to_state((char)argv[1][0])) == -1) {
3495 		(void) fprintf(stderr, usage_msg);
3496 		(void) audit_put_record(ADT_FAILURE, ADT_FAIL_VALUE_BAD_CMD,
3497 		    argv[1]);
3498 		exit(1);
3499 	}
3500 
3501 	if (init_signal == SINGLE_USER) {
3502 		/*
3503 		 * Make sure this process is talking to a legal tty line
3504 		 * and that /dev/syscon is linked to this line.
3505 		 */
3506 		ln = ttyname(0);	/* Get the name of tty */
3507 		if (ln == NULL) {
3508 			(void) fprintf(stderr,
3509 			    "Standard input not a tty line\n");
3510 			(void) audit_put_record(ADT_FAILURE,
3511 			    ADT_FAIL_VALUE_BAD_TTY, argv[1]);
3512 			exit(1);
3513 		}
3514 
3515 		if ((stat(ln, &sconbuf) != -1) &&
3516 		    (stat(SYSCON, &conbuf) == -1 ||
3517 		    sconbuf.st_rdev != conbuf.st_rdev)) {
3518 			/*
3519 			 * /dev/syscon needs to change.
3520 			 * Unlink /dev/syscon and relink it to the current line.
3521 			 */
3522 			if (lstat(SYSCON, &conbuf) != -1 &&
3523 			    unlink(SYSCON) == FAILURE) {
3524 				perror("Can't unlink /dev/syscon");
3525 				(void) fprintf(stderr,
3526 				    "Run command on the system console.\n");
3527 				(void) audit_put_record(ADT_FAILURE,
3528 				    ADT_FAIL_VALUE_PROGRAM, argv[1]);
3529 				exit(1);
3530 			}
3531 			if (symlink(ln, SYSCON) == FAILURE) {
3532 				(void) fprintf(stderr,
3533 				    "Can't symlink /dev/syscon to %s: %s", ln,
3534 				    strerror(errno));
3535 
3536 				/* Try to leave a syscon */
3537 				(void) link(SYSTTY, SYSCON);
3538 				(void) audit_put_record(ADT_FAILURE,
3539 				    ADT_FAIL_VALUE_PROGRAM, argv[1]);
3540 				exit(1);
3541 			}
3542 
3543 			/*
3544 			 * Try to leave a message on system console saying where
3545 			 * /dev/syscon is currently connected.
3546 			 */
3547 			if ((fp = fopen(SYSTTY, "r+")) != NULL) {
3548 				(void) fprintf(fp,
3549 				    "\n****	SYSCON CHANGED TO %s	****\n",
3550 				    ln);
3551 				(void) fclose(fp);
3552 			}
3553 		}
3554 	}
3555 
3556 	update_boot_archive(init_signal);
3557 
3558 	(void) audit_put_record(ADT_SUCCESS, ADT_SUCCESS, argv[1]);
3559 
3560 	/*
3561 	 * Signal init; init will take care of telling svc.startd.
3562 	 */
3563 	if (kill(init_pid, init_signal) == FAILURE) {
3564 		(void) fprintf(stderr, "Must be super-user\n");
3565 		(void) audit_put_record(ADT_FAILURE,
3566 		    ADT_FAIL_VALUE_AUTH, argv[1]);
3567 		exit(1);
3568 	}
3569 
3570 	exit(0);
3571 }
3572 
3573 
3574 #define	DELTA	25	/* Number of pidlist elements to allocate at a time */
3575 
3576 /* ARGSUSED */
3577 void
3578 sigpoll(int n)
3579 {
3580 	struct pidrec prec;
3581 	struct pidrec *p = &prec;
3582 	struct pidlist *plp;
3583 	struct pidlist *tp, *savetp;
3584 	int i;
3585 
3586 	if (Pfd < 0) {
3587 		return;
3588 	}
3589 
3590 	for (;;) {
3591 		/*
3592 		 * Important Note: Either read will really fail (in which case
3593 		 * return is all we can do) or will get EAGAIN (Pfd was opened
3594 		 * O_NDELAY), in which case we also want to return.
3595 		 * Always return from here!
3596 		 */
3597 		if (read(Pfd, p, sizeof (struct pidrec)) !=
3598 						sizeof (struct pidrec)) {
3599 			return;
3600 		}
3601 		switch (p->pd_type) {
3602 
3603 		case ADDPID:
3604 			/*
3605 			 * New "godchild", add to list.
3606 			 */
3607 			if (Plfree == NULL) {
3608 				plp = (struct pidlist *)calloc(DELTA,
3609 				    sizeof (struct pidlist));
3610 				if (plp == NULL) {
3611 					/* Can't save pid */
3612 					break;
3613 				}
3614 				/*
3615 				 * Point at 2nd record allocated, we'll use plp.
3616 				 */
3617 				tp = plp + 1;
3618 				/*
3619 				 * Link them into a chain.
3620 				 */
3621 				Plfree = tp;
3622 				for (i = 0; i < DELTA - 2; i++) {
3623 					tp->pl_next = tp + 1;
3624 					tp++;
3625 				}
3626 			} else {
3627 				plp = Plfree;
3628 				Plfree = plp->pl_next;
3629 			}
3630 			plp->pl_pid = p->pd_pid;
3631 			plp->pl_dflag = 0;
3632 			plp->pl_next = NULL;
3633 			/*
3634 			 * Note - pid list is kept in increasing order of pids.
3635 			 */
3636 			if (Plhead == NULL) {
3637 				Plhead = plp;
3638 				/* Back up to read next record */
3639 				break;
3640 			} else {
3641 				savetp = tp = Plhead;
3642 				while (tp) {
3643 					if (plp->pl_pid > tp->pl_pid) {
3644 						savetp = tp;
3645 						tp = tp->pl_next;
3646 						continue;
3647 					} else if (plp->pl_pid < tp->pl_pid) {
3648 						if (tp == Plhead) {
3649 							plp->pl_next = Plhead;
3650 							Plhead = plp;
3651 						} else {
3652 							plp->pl_next =
3653 							    savetp->pl_next;
3654 							savetp->pl_next = plp;
3655 						}
3656 						break;
3657 					} else {
3658 						/* Already in list! */
3659 						plp->pl_next = Plfree;
3660 						Plfree = plp;
3661 						break;
3662 					}
3663 				}
3664 				if (tp == NULL) {
3665 					/* Add to end of list */
3666 					savetp->pl_next = plp;
3667 				}
3668 			}
3669 			/* Back up to read next record. */
3670 			break;
3671 
3672 		case REMPID:
3673 			/*
3674 			 * This one was handled by someone else,
3675 			 * purge it from the list.
3676 			 */
3677 			if (Plhead == NULL) {
3678 				/* Back up to read next record. */
3679 				break;
3680 			}
3681 			savetp = tp = Plhead;
3682 			while (tp) {
3683 				if (p->pd_pid > tp->pl_pid) {
3684 					/* Keep on looking. */
3685 					savetp = tp;
3686 					tp = tp->pl_next;
3687 					continue;
3688 				} else if (p->pd_pid < tp->pl_pid) {
3689 					/* Not in list. */
3690 					break;
3691 				} else {
3692 					/* Found it. */
3693 					if (tp == Plhead)
3694 						Plhead = tp->pl_next;
3695 					else
3696 						savetp->pl_next = tp->pl_next;
3697 					tp->pl_next = Plfree;
3698 					Plfree = tp;
3699 					break;
3700 				}
3701 			}
3702 			/* Back up to read next record. */
3703 			break;
3704 		default:
3705 			console(B_TRUE, "Bad message on initpipe\n");
3706 			break;
3707 		}
3708 	}
3709 }
3710 
3711 
3712 static void
3713 cleanaux()
3714 {
3715 	struct pidlist *savep, *p;
3716 	pid_t	pid;
3717 	short	status;
3718 
3719 	(void) sighold(SIGCLD);
3720 	Gchild = 0;	/* Note - Safe to do this here since no SIGCLDs */
3721 	(void) sighold(SIGPOLL);
3722 	savep = p = Plhead;
3723 	while (p) {
3724 		if (p->pl_dflag) {
3725 			/*
3726 			 * Found an entry to delete,
3727 			 * remove it from list first.
3728 			 */
3729 			pid = p->pl_pid;
3730 			status = p->pl_exit;
3731 			if (p == Plhead) {
3732 				Plhead = p->pl_next;
3733 				p->pl_next = Plfree;
3734 				Plfree = p;
3735 				savep = p = Plhead;
3736 			} else {
3737 				savep->pl_next = p->pl_next;
3738 				p->pl_next = Plfree;
3739 				Plfree = p;
3740 				p = savep->pl_next;
3741 			}
3742 			clearent(pid, status);
3743 			continue;
3744 		}
3745 		savep = p;
3746 		p = p->pl_next;
3747 	}
3748 	(void) sigrelse(SIGPOLL);
3749 	(void) sigrelse(SIGCLD);
3750 }
3751 
3752 
3753 /*
3754  * /etc/inittab has more entries and we have run out of room in the proc_table
3755  * array. Double the size of proc_table to accomodate the extra entries.
3756  */
3757 static void
3758 increase_proc_table_size()
3759 {
3760 	sigset_t block, unblock;
3761 	void *ptr;
3762 	size_t delta = num_proc * sizeof (struct PROC_TABLE);
3763 
3764 
3765 	/*
3766 	 * Block signals for realloc.
3767 	 */
3768 	(void) sigfillset(&block);
3769 	(void) sigprocmask(SIG_BLOCK, &block, &unblock);
3770 
3771 
3772 	/*
3773 	 * On failure we just return because callers of this function check
3774 	 * for failure.
3775 	 */
3776 	do
3777 		ptr = realloc(g_state, g_state_sz + delta);
3778 	while (ptr == NULL && errno == EAGAIN)
3779 		;
3780 
3781 	if (ptr != NULL) {
3782 		/* ensure that the new part is initialized to zero */
3783 		bzero((caddr_t)ptr + g_state_sz, delta);
3784 
3785 		g_state = ptr;
3786 		g_state_sz += delta;
3787 		num_proc <<= 1;
3788 	}
3789 
3790 
3791 	/* unblock our signals before returning */
3792 	(void) sigprocmask(SIG_SETMASK, &unblock, NULL);
3793 }
3794 
3795 
3796 
3797 /*
3798  * Sanity check g_state.
3799  */
3800 static int
3801 st_sane()
3802 {
3803 	int i;
3804 	struct PROC_TABLE *ptp;
3805 
3806 
3807 	/* Note: cur_state is encoded as a signal number */
3808 	if (cur_state < 1 || cur_state == 9 || cur_state > 13)
3809 		return (0);
3810 
3811 	/* Check num_proc */
3812 	if (g_state_sz != sizeof (struct init_state) + (num_proc - 1) *
3813 	    sizeof (struct PROC_TABLE))
3814 		return (0);
3815 
3816 	/* Check proc_table */
3817 	for (i = 0, ptp = proc_table; i < num_proc; ++i, ++ptp) {
3818 		/* skip unoccupied entries */
3819 		if (!(ptp->p_flags & OCCUPIED))
3820 			continue;
3821 
3822 		/* p_flags has no bits outside of PF_MASK */
3823 		if (ptp->p_flags & ~(PF_MASK))
3824 			return (0);
3825 
3826 		/* 5 <= pid <= MAXPID */
3827 		if (ptp->p_pid < 5 || ptp->p_pid > MAXPID)
3828 			return (0);
3829 
3830 		/* p_count >= 0 */
3831 		if (ptp->p_count < 0)
3832 			return (0);
3833 
3834 		/* p_time >= 0 */
3835 		if (ptp->p_time < 0)
3836 			return (0);
3837 	}
3838 
3839 	return (1);
3840 }
3841 
3842 /*
3843  * Initialize our state.
3844  *
3845  * If the system just booted, then init_state_file, which is located on an
3846  * everpresent tmpfs filesystem, should not exist.
3847  *
3848  * If we were restarted, then init_state_file should exist, in
3849  * which case we'll read it in, sanity check it, and use it.
3850  *
3851  * Note: You can't call console() until proc_table is ready.
3852  */
3853 void
3854 st_init()
3855 {
3856 	struct stat stb;
3857 	int ret, st_fd, insane = 0;
3858 	size_t to_be_read;
3859 	char *ptr;
3860 
3861 
3862 	booting = 1;
3863 
3864 	do {
3865 		/*
3866 		 * If we can exclusively create the file, then we're the
3867 		 * initial invocation of init(8).
3868 		 */
3869 		st_fd = open(init_state_file, O_RDWR | O_CREAT | O_EXCL,
3870 		    S_IRUSR | S_IWUSR);
3871 	} while (st_fd == -1 && errno == EINTR);
3872 	if (st_fd != -1)
3873 		goto new_state;
3874 
3875 	booting = 0;
3876 
3877 	do {
3878 		st_fd = open(init_state_file, O_RDWR, S_IRUSR | S_IWUSR);
3879 	} while (st_fd == -1 && errno == EINTR);
3880 	if (st_fd == -1)
3881 		goto new_state;
3882 
3883 	/* Get the size of the file. */
3884 	do
3885 		ret = fstat(st_fd, &stb);
3886 	while (ret == -1 && errno == EINTR)
3887 		;
3888 	if (ret == -1)
3889 		goto new_state;
3890 
3891 	do
3892 		g_state = malloc(stb.st_size);
3893 	while (g_state == NULL && errno == EAGAIN)
3894 		;
3895 	if (g_state == NULL)
3896 		goto new_state;
3897 
3898 	to_be_read = stb.st_size;
3899 	ptr = (char *)g_state;
3900 	while (to_be_read > 0) {
3901 		ssize_t read_ret;
3902 
3903 		read_ret = read(st_fd, ptr, to_be_read);
3904 		if (read_ret < 0) {
3905 			if (errno == EINTR)
3906 				continue;
3907 
3908 			goto new_state;
3909 		}
3910 
3911 		to_be_read -= read_ret;
3912 		ptr += read_ret;
3913 	}
3914 
3915 	(void) close(st_fd);
3916 
3917 	g_state_sz = stb.st_size;
3918 
3919 	if (st_sane()) {
3920 		console(B_TRUE, "Restarting.\n");
3921 		return;
3922 	}
3923 
3924 	insane = 1;
3925 
3926 new_state:
3927 	if (st_fd >= 0)
3928 		(void) close(st_fd);
3929 	else
3930 		(void) unlink(init_state_file);
3931 
3932 	if (g_state != NULL)
3933 		free(g_state);
3934 
3935 	/* Something went wrong, so allocate new state. */
3936 	g_state_sz = sizeof (struct init_state) +
3937 	    ((init_num_proc - 1) * sizeof (struct PROC_TABLE));
3938 	do
3939 		g_state = calloc(1, g_state_sz);
3940 	while (g_state == NULL && errno == EAGAIN)
3941 		;
3942 	if (g_state == NULL) {
3943 		/* Fatal error! */
3944 		exit(errno);
3945 	}
3946 
3947 	g_state->ist_runlevel = -1;
3948 	num_proc = init_num_proc;
3949 
3950 	if (!booting) {
3951 		console(B_TRUE, "Restarting.\n");
3952 
3953 		/* Overwrite the bad state file. */
3954 		st_write();
3955 
3956 		if (!insane) {
3957 			console(B_TRUE,
3958 			    "Error accessing persistent state file `%s'.  "
3959 			    "Ignored.\n", init_state_file);
3960 		} else {
3961 			console(B_TRUE,
3962 			    "Persistent state file `%s' is invalid and was "
3963 			    "ignored.\n", init_state_file);
3964 		}
3965 	}
3966 }
3967 
3968 /*
3969  * Write g_state out to the state file.
3970  */
3971 void
3972 st_write()
3973 {
3974 	static int complained = 0;
3975 
3976 	int st_fd;
3977 	char *cp;
3978 	size_t sz;
3979 	ssize_t ret;
3980 
3981 
3982 	do {
3983 		st_fd = open(init_next_state_file,
3984 		    O_WRONLY | O_CREAT | O_TRUNC, S_IRUSR | S_IWUSR);
3985 	} while (st_fd < 0 && errno == EINTR);
3986 	if (st_fd < 0)
3987 		goto err;
3988 
3989 	cp = (char *)g_state;
3990 	sz = g_state_sz;
3991 	while (sz > 0) {
3992 		ret = write(st_fd, cp, sz);
3993 		if (ret < 0) {
3994 			if (errno == EINTR)
3995 				continue;
3996 
3997 			goto err;
3998 		}
3999 
4000 		sz -= ret;
4001 		cp += ret;
4002 	}
4003 
4004 	(void) close(st_fd);
4005 	st_fd = -1;
4006 	if (rename(init_next_state_file, init_state_file)) {
4007 		(void) unlink(init_next_state_file);
4008 		goto err;
4009 	}
4010 	complained = 0;
4011 
4012 	return;
4013 
4014 err:
4015 	if (st_fd >= 0)
4016 		(void) close(st_fd);
4017 
4018 	if (!booting && !complained) {
4019 		/*
4020 		 * Only complain after the filesystem should have come up.
4021 		 * And only do it once so we don't loop between console()
4022 		 * & efork().
4023 		 */
4024 		complained = 1;
4025 		if (st_fd)
4026 			console(B_TRUE, "Couldn't write persistent state "
4027 			    "file `%s'.\n", init_state_file);
4028 		else
4029 			console(B_TRUE, "Couldn't move persistent state "
4030 			    "file `%s' to `%s'.\n", init_next_state_file,
4031 			    init_state_file);
4032 	}
4033 }
4034 
4035 /*
4036  * Create a contract with these parameters.
4037  */
4038 static int
4039 contract_make_template(uint_t info, uint_t critical, uint_t fatal,
4040     uint64_t cookie)
4041 {
4042 	int fd, err;
4043 
4044 	char *ioctl_tset_emsg =
4045 	    "Couldn't set \"%s\" contract template parameter: %s.\n";
4046 
4047 	do
4048 		fd = open64(CTFS_ROOT "/process/template", O_RDWR);
4049 	while (fd < 0 && errno == EINTR)
4050 		;
4051 	if (fd < 0) {
4052 		console(B_TRUE, "Couldn't create process template: %s.\n",
4053 		    strerror(errno));
4054 		return (-1);
4055 	}
4056 
4057 	if (err = ct_pr_tmpl_set_param(fd, CT_PR_INHERIT | CT_PR_REGENT))
4058 		console(B_TRUE, "Contract set template inherit, regent "
4059 		    "failed: %s.\n", strerror(err));
4060 
4061 	/*
4062 	 * These errors result in a misconfigured template, which is better
4063 	 * than no template at all, so warn but don't abort.
4064 	 */
4065 	if (err = ct_tmpl_set_informative(fd, info))
4066 		console(B_TRUE, ioctl_tset_emsg, "informative", strerror(err));
4067 
4068 	if (err = ct_tmpl_set_critical(fd, critical))
4069 		console(B_TRUE, ioctl_tset_emsg, "critical", strerror(err));
4070 
4071 	if (err = ct_pr_tmpl_set_fatal(fd, fatal))
4072 		console(B_TRUE, ioctl_tset_emsg, "fatal", strerror(err));
4073 
4074 	if (err = ct_tmpl_set_cookie(fd, cookie))
4075 		console(B_TRUE, ioctl_tset_emsg, "cookie", strerror(err));
4076 
4077 	(void) fcntl(fd, F_SETFD, FD_CLOEXEC);
4078 
4079 	return (fd);
4080 }
4081 
4082 /*
4083  * Create the templates and open an event file descriptor.  We use dup2(2) to
4084  * get these descriptors away from the stdin/stdout/stderr group.
4085  */
4086 static void
4087 contracts_init()
4088 {
4089 	int err, fd;
4090 
4091 	/*
4092 	 * Create & configure a legacy template.  We only want empty events so
4093 	 * we know when to abandon them.
4094 	 */
4095 	legacy_tmpl = contract_make_template(0, CT_PR_EV_EMPTY, CT_PR_EV_HWERR,
4096 	    ORDINARY_COOKIE);
4097 	if (legacy_tmpl >= 0) {
4098 		err = ct_tmpl_activate(legacy_tmpl);
4099 		if (err != 0) {
4100 			(void) close(legacy_tmpl);
4101 			legacy_tmpl = -1;
4102 			console(B_TRUE,
4103 			    "Couldn't activate legacy template (%s); "
4104 			    "legacy services will be in init's contract.\n",
4105 			    strerror(err));
4106 		}
4107 	} else
4108 		console(B_TRUE,
4109 		    "Legacy services will be in init's contract.\n");
4110 
4111 	if (dup2(legacy_tmpl, 255) == -1) {
4112 		console(B_TRUE, "Could not duplicate legacy template: %s.\n",
4113 		    strerror(errno));
4114 	} else {
4115 		(void) close(legacy_tmpl);
4116 		legacy_tmpl = 255;
4117 	}
4118 
4119 	(void) fcntl(legacy_tmpl, F_SETFD, FD_CLOEXEC);
4120 
4121 	startd_tmpl = contract_make_template(0, CT_PR_EV_EMPTY,
4122 	    CT_PR_EV_HWERR | CT_PR_EV_SIGNAL | CT_PR_EV_CORE, STARTD_COOKIE);
4123 
4124 	if (dup2(startd_tmpl, 254) == -1) {
4125 		console(B_TRUE, "Could not duplicate startd template: %s.\n",
4126 		    strerror(errno));
4127 	} else {
4128 		(void) close(startd_tmpl);
4129 		startd_tmpl = 254;
4130 	}
4131 
4132 	(void) fcntl(startd_tmpl, F_SETFD, FD_CLOEXEC);
4133 
4134 	if (legacy_tmpl < 0 && startd_tmpl < 0) {
4135 		/* The creation errors have already been reported. */
4136 		console(B_TRUE,
4137 		    "Ignoring contract events.  Core smf(7) services will not "
4138 		    "be restarted.\n");
4139 		return;
4140 	}
4141 
4142 	/*
4143 	 * Open an event endpoint.
4144 	 */
4145 	do
4146 		fd = open64(CTFS_ROOT "/process/pbundle", O_RDONLY);
4147 	while (fd < 0 && errno == EINTR)
4148 		;
4149 	if (fd < 0) {
4150 		console(B_TRUE,
4151 		    "Couldn't open process pbundle: %s.  Core smf(7) services "
4152 		    "will not be restarted.\n", strerror(errno));
4153 		return;
4154 	}
4155 
4156 	if (dup2(fd, 253) == -1) {
4157 		console(B_TRUE, "Could not duplicate process bundle: %s.\n",
4158 		    strerror(errno));
4159 	} else {
4160 		(void) close(fd);
4161 		fd = 253;
4162 	}
4163 
4164 	(void) fcntl(fd, F_SETFD, FD_CLOEXEC);
4165 
4166 	/* Reset in case we've been restarted. */
4167 	(void) ct_event_reset(fd);
4168 
4169 	poll_fds[0].fd = fd;
4170 	poll_fds[0].events = POLLIN;
4171 	poll_nfds = 1;
4172 }
4173 
4174 static int
4175 contract_getfile(ctid_t id, const char *name, int oflag)
4176 {
4177 	int fd;
4178 
4179 	do
4180 		fd = contract_open(id, "process", name, oflag);
4181 	while (fd < 0 && errno == EINTR)
4182 		;
4183 
4184 	if (fd < 0)
4185 		console(B_TRUE, "Couldn't open %s for contract %ld: %s.\n",
4186 		    name, id, strerror(errno));
4187 
4188 	return (fd);
4189 }
4190 
4191 static int
4192 contract_cookie(ctid_t id, uint64_t *cp)
4193 {
4194 	int fd, err;
4195 	ct_stathdl_t sh;
4196 
4197 	fd = contract_getfile(id, "status", O_RDONLY);
4198 	if (fd < 0)
4199 		return (-1);
4200 
4201 	err = ct_status_read(fd, CTD_COMMON, &sh);
4202 	if (err != 0) {
4203 		console(B_TRUE, "Couldn't read status of contract %ld: %s.\n",
4204 		    id, strerror(err));
4205 		(void) close(fd);
4206 		return (-1);
4207 	}
4208 
4209 	(void) close(fd);
4210 
4211 	*cp = ct_status_get_cookie(sh);
4212 
4213 	ct_status_free(sh);
4214 	return (0);
4215 }
4216 
4217 static void
4218 contract_ack(ct_evthdl_t e)
4219 {
4220 	int fd;
4221 
4222 	if (ct_event_get_flags(e) & CTE_INFO)
4223 		return;
4224 
4225 	fd = contract_getfile(ct_event_get_ctid(e), "ctl", O_WRONLY);
4226 	if (fd < 0)
4227 		return;
4228 
4229 	(void) ct_ctl_ack(fd, ct_event_get_evid(e));
4230 	(void) close(fd);
4231 }
4232 
4233 /*
4234  * Process a contract event.
4235  */
4236 static void
4237 contract_event(struct pollfd *poll)
4238 {
4239 	ct_evthdl_t e;
4240 	int err;
4241 	ctid_t ctid;
4242 
4243 	if (!(poll->revents & POLLIN)) {
4244 		if (poll->revents & POLLERR)
4245 			console(B_TRUE,
4246 			    "Unknown poll error on my process contract "
4247 			    "pbundle.\n");
4248 		return;
4249 	}
4250 
4251 	err = ct_event_read(poll->fd, &e);
4252 	if (err != 0) {
4253 		console(B_TRUE, "Error retrieving contract event: %s.\n",
4254 		    strerror(err));
4255 		return;
4256 	}
4257 
4258 	ctid = ct_event_get_ctid(e);
4259 
4260 	if (ct_event_get_type(e) == CT_PR_EV_EMPTY) {
4261 		uint64_t cookie;
4262 		int ret, abandon = 1;
4263 
4264 		/* If it's svc.startd, restart it.  Else, abandon. */
4265 		ret = contract_cookie(ctid, &cookie);
4266 
4267 		if (ret == 0) {
4268 			if (cookie == STARTD_COOKIE &&
4269 			    do_restart_startd) {
4270 				if (smf_debug)
4271 					console(B_TRUE, "Restarting "
4272 					    "svc.startd.\n");
4273 
4274 				/*
4275 				 * Account for the failure.  If the failure rate
4276 				 * exceeds a threshold, then drop to maintenance
4277 				 * mode.
4278 				 */
4279 				startd_record_failure();
4280 				if (startd_failure_rate_critical())
4281 					enter_maintenance();
4282 
4283 				if (startd_tmpl < 0)
4284 					console(B_TRUE,
4285 					    "Restarting svc.startd in "
4286 					    "improper contract (bad "
4287 					    "template).\n");
4288 
4289 				(void) startd_run(startd_cline, startd_tmpl,
4290 				    ctid);
4291 
4292 				abandon = 0;
4293 			}
4294 		}
4295 
4296 		if (abandon && (err = contract_abandon_id(ctid))) {
4297 			console(B_TRUE, "Couldn't abandon contract %ld: %s.\n",
4298 			    ctid, strerror(err));
4299 		}
4300 
4301 		/*
4302 		 * No need to acknowledge the event since either way the
4303 		 * originating contract should be abandoned.
4304 		 */
4305 	} else {
4306 		console(B_TRUE,
4307 		    "Received contract event of unexpected type %d from "
4308 		    "contract %ld.\n", ct_event_get_type(e), ctid);
4309 
4310 		if ((ct_event_get_flags(e) & (CTE_INFO | CTE_ACK)) == 0)
4311 			/* Allow unexpected critical events to be released. */
4312 			contract_ack(e);
4313 	}
4314 
4315 	ct_event_free(e);
4316 }
4317 
4318 /*
4319  * svc.startd(8) Management
4320  */
4321 
4322 /*
4323  * (Re)start svc.startd(8).  old_ctid should be the contract ID of the old
4324  * contract, or 0 if we're starting it for the first time.  If wait is true
4325  * we'll wait for and return the exit value of the child.
4326  */
4327 static int
4328 startd_run(const char *cline, int tmpl, ctid_t old_ctid)
4329 {
4330 	int err, i, ret, did_activate;
4331 	pid_t pid;
4332 	struct stat sb;
4333 
4334 	if (cline[0] == '\0')
4335 		return (-1);
4336 
4337 	/*
4338 	 * Don't restart startd if the system is rebooting or shutting down.
4339 	 */
4340 	do {
4341 		ret = stat("/etc/svc/volatile/resetting", &sb);
4342 	} while (ret == -1 && errno == EINTR);
4343 
4344 	if (ret == 0) {
4345 		if (smf_debug)
4346 			console(B_TRUE, "Quiescing for reboot.\n");
4347 		(void) pause();
4348 		return (-1);
4349 	}
4350 
4351 	err = ct_pr_tmpl_set_transfer(tmpl, old_ctid);
4352 	if (err == EINVAL) {
4353 		console(B_TRUE, "Remake startd_tmpl; reattempt transfer.\n");
4354 		tmpl = startd_tmpl = contract_make_template(0, CT_PR_EV_EMPTY,
4355 		    CT_PR_EV_HWERR, STARTD_COOKIE);
4356 
4357 		err = ct_pr_tmpl_set_transfer(tmpl, old_ctid);
4358 	}
4359 	if (err != 0) {
4360 		console(B_TRUE,
4361 		    "Couldn't set transfer parameter of contract template: "
4362 		    "%s.\n", strerror(err));
4363 	}
4364 
4365 	if ((err = ct_pr_tmpl_set_svc_fmri(startd_tmpl,
4366 	    SCF_SERVICE_STARTD)) != 0)
4367 		console(B_TRUE,
4368 		    "Can not set svc_fmri in contract template: %s\n",
4369 		    strerror(err));
4370 	if ((err = ct_pr_tmpl_set_svc_aux(startd_tmpl,
4371 	    startd_svc_aux)) != 0)
4372 		console(B_TRUE,
4373 		    "Can not set svc_aux in contract template: %s\n",
4374 		    strerror(err));
4375 	did_activate = !(ct_tmpl_activate(tmpl));
4376 	if (!did_activate)
4377 		console(B_TRUE,
4378 		    "Template activation failed; not starting \"%s\" in "
4379 		    "proper contract.\n", cline);
4380 
4381 	/* Hold SIGCLD so we can wait if necessary. */
4382 	(void) sighold(SIGCLD);
4383 
4384 	while ((pid = fork()) < 0) {
4385 		if (errno == EPERM) {
4386 			console(B_TRUE, "Insufficient permission to fork.\n");
4387 
4388 			/* Now that's a doozy. */
4389 			exit(1);
4390 		}
4391 
4392 		console(B_TRUE,
4393 		    "fork() for svc.startd failed: %s.  Will retry in 1 "
4394 		    "second...\n", strerror(errno));
4395 
4396 		(void) sleep(1);
4397 
4398 		/* Eventually give up? */
4399 	}
4400 
4401 	if (pid == 0) {
4402 		/* child */
4403 
4404 		/* See the comment in efork() */
4405 		for (i = SIGHUP; i <= SIGRTMAX; ++i) {
4406 			if (i == SIGTTOU || i == SIGTTIN || i == SIGTSTP)
4407 				(void) sigset(i, SIG_IGN);
4408 			else
4409 				(void) sigset(i, SIG_DFL);
4410 		}
4411 
4412 		if (smf_options != NULL) {
4413 			/* Put smf_options in the environment. */
4414 			glob_envp[glob_envn] =
4415 			    malloc(sizeof ("SMF_OPTIONS=") - 1 +
4416 			    strlen(smf_options) + 1);
4417 
4418 			if (glob_envp[glob_envn] != NULL) {
4419 				/* LINTED */
4420 				(void) sprintf(glob_envp[glob_envn],
4421 				    "SMF_OPTIONS=%s", smf_options);
4422 				glob_envp[glob_envn+1] = NULL;
4423 			} else {
4424 				console(B_TRUE,
4425 				    "Could not set SMF_OPTIONS (%s).\n",
4426 				    strerror(errno));
4427 			}
4428 		}
4429 
4430 		if (smf_debug)
4431 			console(B_TRUE, "Executing svc.startd\n");
4432 
4433 		(void) execle(SH, "INITSH", "-c", cline, NULL, glob_envp);
4434 
4435 		console(B_TRUE, "Could not exec \"%s\" (%s).\n", SH,
4436 		    strerror(errno));
4437 
4438 		exit(1);
4439 	}
4440 
4441 	/* parent */
4442 
4443 	if (did_activate) {
4444 		if (legacy_tmpl < 0 || ct_tmpl_activate(legacy_tmpl) != 0)
4445 			(void) ct_tmpl_clear(tmpl);
4446 	}
4447 
4448 	/* Clear the old_ctid reference so the kernel can reclaim it. */
4449 	if (old_ctid != 0)
4450 		(void) ct_pr_tmpl_set_transfer(tmpl, 0);
4451 
4452 	(void) sigrelse(SIGCLD);
4453 
4454 	return (0);
4455 }
4456 
4457 /*
4458  * void startd_record_failure(void)
4459  *   Place the current time in our circular array of svc.startd failures.
4460  */
4461 void
4462 startd_record_failure()
4463 {
4464 	int index = startd_failure_index++ % NSTARTD_FAILURE_TIMES;
4465 
4466 	startd_failure_time[index] = gethrtime();
4467 }
4468 
4469 /*
4470  * int startd_failure_rate_critical(void)
4471  *   Return true if the average failure interval is less than the permitted
4472  *   interval.  Implicit success if insufficient measurements for an average
4473  *   exist.
4474  */
4475 int
4476 startd_failure_rate_critical()
4477 {
4478 	int n = startd_failure_index;
4479 	hrtime_t avg_ns = 0;
4480 
4481 	if (startd_failure_index < NSTARTD_FAILURE_TIMES)
4482 		return (0);
4483 
4484 	avg_ns =
4485 	    (startd_failure_time[(n - 1) % NSTARTD_FAILURE_TIMES] -
4486 	    startd_failure_time[n % NSTARTD_FAILURE_TIMES]) /
4487 	    NSTARTD_FAILURE_TIMES;
4488 
4489 	return (avg_ns < STARTD_FAILURE_RATE_NS);
4490 }
4491 
4492 /*
4493  * returns string that must be free'd
4494  */
4495 
4496 static char
4497 *audit_boot_msg()
4498 {
4499 	char		*b, *p;
4500 	char		desc[] = "booted";
4501 	zoneid_t	zid = getzoneid();
4502 
4503 	b = malloc(sizeof (desc) + MAXNAMELEN + 3);
4504 	if (b == NULL)
4505 		return (b);
4506 
4507 	p = b;
4508 	p += strlcpy(p, desc, sizeof (desc));
4509 	if (zid != GLOBAL_ZONEID) {
4510 		p += strlcpy(p, ": ", 3);
4511 		(void) getzonenamebyid(zid, p, MAXNAMELEN);
4512 	}
4513 	return (b);
4514 }
4515 
4516 /*
4517  * Generate AUE_init_solaris audit record.  Return 1 if
4518  * auditing is enabled in case the caller cares.
4519  *
4520  * In the case of userint() or a local zone invocation of
4521  * one_true_init, the process initially contains the audit
4522  * characteristics of the process that invoked init.  The first pass
4523  * through here uses those characteristics then for the case of
4524  * one_true_init in a local zone, clears them so subsequent system
4525  * state changes won't be attributed to the person who booted the
4526  * zone.
4527  */
4528 static int
4529 audit_put_record(int pass_fail, int status, char *msg)
4530 {
4531 	adt_session_data_t	*ah;
4532 	adt_event_data_t	*event;
4533 
4534 	if (!adt_audit_enabled())
4535 		return (0);
4536 
4537 	/*
4538 	 * the PROC_DATA picks up the context to tell whether this is
4539 	 * an attributed record (auid = -2 is unattributed)
4540 	 */
4541 	if (adt_start_session(&ah, NULL, ADT_USE_PROC_DATA)) {
4542 		console(B_TRUE, "audit failure:  %s\n", strerror(errno));
4543 		return (1);
4544 	}
4545 	event = adt_alloc_event(ah, ADT_init_solaris);
4546 	if (event == NULL) {
4547 		console(B_TRUE, "audit failure:  %s\n", strerror(errno));
4548 		(void) adt_end_session(ah);
4549 		return (1);
4550 	}
4551 	event->adt_init_solaris.info = msg;	/* NULL is ok here */
4552 
4553 	if (adt_put_event(event, pass_fail, status)) {
4554 		console(B_TRUE, "audit failure:  %s\n", strerror(errno));
4555 		(void) adt_end_session(ah);
4556 		return (1);
4557 	}
4558 	adt_free_event(event);
4559 
4560 	(void) adt_end_session(ah);
4561 
4562 	return (1);
4563 }
4564