xref: /titanic_50/usr/src/cmd/utmpd/utmpd.c (revision eab227978ccdaa5a7cc9fd92ace768915dae3a2b)
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  * Copyright 2014, 2015 Shruti V Sampat <shrutisampat@gmail.com>
23  */
24 
25 /*
26  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
27  * Use is subject to license terms.
28  */
29 
30 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
31 /*	  All Rights Reserved  	*/
32 
33 /*
34  * Portions of such source code were derived from Berkeley 4.3 BSD
35  * under license from the Regents of the University of California.
36  */
37 
38 /*
39  * utmpd	- utmp daemon
40  *
41  *		This program receives requests from  pututxline(3)
42  *		via a named pipe to watch the process to make sure it cleans up
43  *		its utmpx entry on termination.
44  *		The program keeps a list of procs
45  *		and uses poll() on their /proc files to detect termination.
46  *		Also the  program periodically scans the /etc/utmpx file for
47  *		processes that aren't in the table so they can be watched.
48  *
49  *		If utmpd doesn't hear back over the pipe from pututline(3) that
50  *		the process has removed its entry it cleans the entry when the
51  *		the process terminates.
52  *		The AT&T Copyright above is there since we borrowed the pipe
53  *		mechanism from init(1m).
54  */
55 
56 
57 #include	<sys/types.h>
58 #include	<signal.h>
59 #include	<stdio.h>
60 #include	<stdio_ext.h>
61 #include	<unistd.h>
62 #include	<utmpx.h>
63 #include	<errno.h>
64 #include	<termio.h>
65 #include	<sys/termios.h>
66 #include	<sys/tty.h>
67 #include	<ctype.h>
68 #include	<sys/stat.h>
69 #include	<sys/statvfs.h>
70 #include	<fcntl.h>
71 #include	<time.h>
72 #include	<sys/stropts.h>
73 #include	<wait.h>
74 #include	<syslog.h>
75 #include	<stdlib.h>
76 #include	<string.h>
77 #include	<poll.h>
78 #include	<deflt.h>
79 #include	<procfs.h>
80 #include	<sys/resource.h>
81 #include	<limits.h>
82 
83 #define	dprintf(x)	if (Debug) (void) printf x
84 
85 /*
86  * Memory allocation keyed off MAX_FDS
87  */
88 #define	MAX_FDS		4064	/* Maximum # file descriptors */
89 #define	EXTRA_MARGIN	32	/* Allocate this many more FDS over Max_Fds */
90 /*
91  * MAX_POLLNV & RESETS - paranoia to cover an error case that might not exist
92  */
93 #define	MAX_POLL_ERRS	1024	/* Count of bad errors */
94 #define	MAX_RESETS	1024	/* Maximum times to reload tables */
95 #define	POLL_TIMEOUT	300	/* Default Timeout for poll() in seconds */
96 #define	CLEANIT		1	/* Used by rem_pid() */
97 #define	DONT_CLEAN	0	/* Used by rem_pid() */
98 #define	UTMP_DEFAULT	"/etc/default/utmpd"
99 #define	WARN_TIME	3600	/* seconds between utmp checks */
100 #define	WTMPX_UFREQ	60	/* seconds between updating WTMPX's atime */
101 
102 
103 /*
104  * The pidrec structure describes the data shipped down the pipe to
105  * us from the pututxline() library in
106  * lib/libc/port/gen/getutx.c
107  */
108 
109 /*
110  * pd_type's
111  */
112 #define	ADDPID	1
113 #define	REMPID	2
114 
115 struct	pidrec {
116 	int	pd_type;		/* Command type */
117 	pid_t	pd_pid;			/* pid to add or remove */
118 };
119 
120 
121 /*
122  * Since this program uses poll(2) and poll takes an array of file descriptors
123  * as an argument we maintain our data in tables.
124  * One table is the file descriptor array for poll, another parallel
125  * array is a table which contains the process ID of the corresponding
126  * open fd.  These tables are kept sorted by process ID for quick lookups.
127  */
128 
129 struct	pidentry {
130 	pid_t	pl_pid;			/* pid to watch for */
131 	int	pl_status;		/* Exit status of proc */
132 };
133 
134 static struct pidentry *pidtable = NULL;
135 
136 static pollfd_t *fdtable = NULL;
137 
138 static int	pidcnt = 0;		/* Number of procs being watched */
139 static char	*prog_name;		/* To save the invocation name away */
140 static char	*UTMPPIPE_DIR =	"/var/run";
141 static char	*UTMPPIPE = "/var/run/utmppipe";
142 static int	Pfd = -1;		/* File descriptor of named pipe */
143 static int	Poll_timeout = POLL_TIMEOUT;
144 static int	WTMPXfd = -1;		/* File descriptor of WTMPX_FILE */
145 static int	WTMPX_ufreq = WTMPX_UFREQ;
146 static int	Debug = 0;		/* Set by command line argument */
147 static int	Max_fds		= MAX_FDS;
148 
149 /*
150  * This program has three main components plus utilities and debug routines
151  *	Receiver - receives the process ID or process for us to watch.
152  *		   (Uses a named pipe to get messages)
153  *	Watcher	 - Use poll(2) to watch for processes to die so they
154  *		   can be cleaned up (get marked as DEAD_PROCESS)
155  *	Scanner	 - periodically scans the utmpx file for stale entries
156  *		   or live entries that we don't know about.
157  */
158 
159 static int wait_for_pids();	/* Watcher - uses poll */
160 static void scan_utmps();	/* Scanner, reads utmpx file */
161 static void drain_pipe();	/* Receiver - reads mesgs over UTMPPIPE */
162 static void setup_pipe();	/* For setting up receiver */
163 
164 static void add_pid();		/* Adds a process to the table */
165 static void rem_pid();		/* Removes a process from the table */
166 static int find_pid();		/* Finds a process in the table */
167 static int proc_to_fd();	/* Takes a pid and returns an fd for its proc */
168 static void load_tables();	/* Loads up the tables the first time around */
169 static int pidcmp();		/* For sorting pids */
170 
171 static void clean_entry();	/* Removes entry from our table and calls ... */
172 static void clean_utmpx_ent();	/* Cleans a utmpx entry */
173 
174 static void fatal() __NORETURN;	/* Prints error message and calls exit */
175 static void nonfatal();		/* Prints error message */
176 static void print_tables();	/* Prints out internal tables for Debug */
177 static int proc_is_alive(pid_t pid);	/* Check if a process is alive */
178 static void warn_utmp(void);
179 
180 /* Validate defaults from file and assign */
181 static int validate_default(char *defp, int *flag);
182 
183 /*
184  * main()  - Main does basic setup and calls wait_for_pids() to do the work
185  */
186 
187 int
188 main(int argc, char *argv[])
189 {
190 	char *defp;
191 	struct rlimit rlim;
192 	int i;
193 	time_t curtime, now;
194 	char msg[256];
195 
196 	prog_name = argv[0];			/* Save invocation name */
197 
198 	if (getuid() != 0)  {
199 		(void) fprintf(stderr,
200 		    "You must be root to run this program\n");
201 		fatal("You must be root to run this program");
202 	}
203 
204 	if (argc > 1) {
205 		if ((argc == 2 && (int)strlen(argv[1]) >= 2) &&
206 		    (argv[1][0] == '-' && argv[1][1] == 'd')) {
207 			Debug = 1;
208 		} else {
209 			(void) fprintf(stderr,
210 			    "%s: Wrong number of arguments\n", prog_name);
211 			(void) fprintf(stderr,
212 			    "Usage: %s [-debug]\n", prog_name);
213 			exit(2);
214 		}
215 	}
216 
217 	/*
218 	 * Read defaults file for poll timeout, WTMPX update frequency
219 	 * and maximum number of processes to monitor.
220 	 */
221 	if (defopen(UTMP_DEFAULT) == 0) {
222 		if ((defp = defread("SCAN_PERIOD=")) != NULL)
223 			if (validate_default(defp, &Poll_timeout) == -1) {
224 				(void) snprintf(msg, sizeof (msg), "SCAN_PERIOD"
225 				    " should be a positive integer, found %s",
226 				    defp);
227 				nonfatal(msg);
228 			}
229 		dprintf(("Poll timeout set to %d\n", Poll_timeout));
230 
231 		if ((defp = defread("WTMPX_UPDATE_FREQ=")) != NULL)
232 			if (validate_default(defp, &WTMPX_ufreq) == -1) {
233 				(void) snprintf(msg, sizeof (msg),
234 				    "WTMPX_UPDATE_FREQ should be a positive "
235 				    "integer, found %s", defp);
236 				nonfatal(msg);
237 			}
238 		dprintf(("WTMPX update frequency set to %d\n", WTMPX_ufreq));
239 
240 		/*
241 		 * Paranoia - if polling on large number of FDs is expensive /
242 		 * buggy the number can be set lower in the field.
243 		 */
244 		if ((defp = defread("MAX_FDS=")) != NULL)
245 			if (validate_default(defp, &Max_fds) == -1) {
246 				(void) snprintf(msg, sizeof (msg), "MAX_FDS "
247 				    "should be a positive integer, found %s",
248 				    defp);
249 				nonfatal(msg);
250 			}
251 		dprintf(("Max fds set to %d\n", Max_fds));
252 		(void) defopen((char *)NULL);
253 	}
254 
255 	if (Debug == 0) {
256 		/*
257 		 * Daemonize ourselves
258 		 */
259 		if (fork()) {
260 			exit(0);
261 		}
262 		(void) close(0);
263 		(void) close(1);
264 		(void) close(2);
265 		/*
266 		 * We open these to avoid accidentally writing to a proc file
267 		 */
268 		(void) open("/dev/null", O_RDONLY);
269 		(void) open("/dev/null", O_WRONLY);
270 		(void) open("/dev/null", O_WRONLY);
271 		(void) setsid();		/* release process from tty */
272 	}
273 
274 	openlog(prog_name, LOG_PID, LOG_DAEMON);	/* For error messages */
275 	warn_utmp();	/* check to see if utmp came back by accident */
276 
277 	/*
278 	 * Allocate the pidtable and fdtable.  An earlier version did
279 	 * this as we go, but this is simpler.
280 	 */
281 	if ((pidtable = malloc(Max_fds * sizeof (struct pidentry))) == NULL)
282 		fatal("Malloc failed");
283 	if ((fdtable = malloc(Max_fds * sizeof (pollfd_t))) == NULL)
284 		fatal("Malloc failed");
285 
286 	/*
287 	 * Up the limit on FDs
288 	 */
289 	if (getrlimit(RLIMIT_NOFILE, &rlim) == 0) {
290 		rlim.rlim_cur = Max_fds + EXTRA_MARGIN + 1;
291 		rlim.rlim_max = Max_fds + EXTRA_MARGIN + 1;
292 		if (setrlimit(RLIMIT_NOFILE, &rlim) != 0) {
293 			fatal("Out of File Descriptors");
294 		}
295 	} else
296 		fatal("getrlimit returned failure");
297 
298 	(void) enable_extended_FILE_stdio(-1, -1);
299 
300 	if ((WTMPXfd = open(WTMPX_FILE, O_RDONLY)) < 0)
301 		nonfatal("WARNING: unable to open " WTMPX_FILE " for update.");
302 
303 	/*
304 	 * Loop here scanning the utmpx file and waiting for processes
305 	 * to terminate.  Most of the activity is directed out of wait_for_pids.
306 	 * If wait_for_pids fails we reload the table and try again.
307 	 */
308 
309 	curtime = time(NULL);
310 	dprintf(("utmp warning timer set to %d seconds\n", WARN_TIME));
311 
312 	for (i = 0; i < MAX_RESETS; i++) {
313 		load_tables();
314 		while (wait_for_pids() == 1) {
315 			now = time(NULL);
316 			if ((now - curtime) >= WARN_TIME) {
317 				dprintf(("utmp warning timer expired\n"));
318 				warn_utmp();
319 				curtime = now;
320 			}
321 		}
322 	}
323 
324 	(void) close(WTMPXfd);
325 
326 	/*
327 	 * We only get here if we had a bunch of resets - so give up
328 	 */
329 	fatal("Too many resets, giving up");
330 	return (1);
331 }
332 
333 /*
334  * load_tables()	- Designed to be called repeatedly if we need to
335  *			  restart things.  Zeros the pidcount, and loads
336  *			  the tables by scanning utmpx
337  */
338 
339 static void
340 load_tables()
341 {
342 	int i;
343 
344 	dprintf(("Load tables\n"));
345 
346 	/*
347 	 * Close any open files.
348 	 */
349 	for (i = 0; i < pidcnt; i++)
350 		(void) close(fdtable[i].fd);
351 
352 	pidcnt = 0;
353 	Pfd = -1;
354 	setup_pipe();		/* Setup the pipe to receive messages */
355 	scan_utmps();		/* Read in USER procs entries to watch */
356 }
357 
358 
359 /*
360  *			*** The Watcher ***
361  *
362  * Wait_for_pids	- wait for the termination of a process in the table.
363  *			  Returns 1 on normal exist, 0 on failure.
364  */
365 
366 static int
367 wait_for_pids()
368 {
369 	register struct pollfd *pfd;
370 	register int i;
371 	pid_t pid;
372 	int ret_val = 0;
373 	int timeout;
374 	static time_t last_timeout  = 0;
375 	static int bad_error  = 0;	/* Count of POLL errors */
376 
377 	/*
378 	 * First time through we initialize last_timeout to now.
379 	 */
380 	if (last_timeout == 0)
381 		last_timeout = time(NULL);
382 
383 	/*
384 	 * Recalculate timeout - checking to see if time expired.
385 	 */
386 
387 	if ((timeout = Poll_timeout - (time(NULL) - last_timeout)) <= 0) {
388 		timeout = Poll_timeout;
389 		last_timeout = time(NULL);
390 		scan_utmps();
391 	}
392 
393 	fdtable[0].events = POLLRDNORM;
394 
395 	for (i = 0; i < (timeout / WTMPX_ufreq); i++) {
396 
397 		/*
398 		 * Loop here while getting EAGAIN
399 		 */
400 
401 		while ((ret_val = poll(fdtable, pidcnt, WTMPX_ufreq*1000)) < 0)
402 			if (errno == EAGAIN)
403 				(void) sleep(2);
404 			else
405 				fatal("poll");
406 		/*
407 		 * The results of pread(2) are discarded; we only want
408 		 * to update the access time of WTMPX_FILE.
409 		 * Periodically touching WTMPX helps determine when the
410 		 * OS became unavailable when the OS boots again .
411 		 * See PSARC 2004/462 for more information.
412 		 */
413 
414 		(void) pread(WTMPXfd, (void *)&pid, sizeof (pid), 0);
415 
416 		if (ret_val)		/* file descriptor(s) need attention */
417 			break;
418 	}
419 
420 	/*
421 	 * If ret_val == 0 the poll timed out - reset last_time and
422 	 * call scan_utmps
423 	 */
424 	if (ret_val == 0) {
425 		last_timeout = time(NULL);
426 		scan_utmps();
427 		return (1);
428 	}
429 
430 	/*
431 	 * Check the pipe file descriptor
432 	 */
433 	if (fdtable[0].revents & POLLRDNORM) {
434 		drain_pipe();
435 		fdtable[0].revents = 0;
436 		ret_val--;
437 	}
438 
439 	(void) sleep(5);	/* Give parents time to cleanup children */
440 
441 	/*
442 	 * We got here because the status of one of the pids that
443 	 * we are polling on has changed, so search the table looking
444 	 * for the entry.
445 	 *
446 	 * The table is scanned backwards so that entries can be removed
447 	 * while we go since the table is compacted from high down to low
448 	 */
449 	for (i = pidcnt - 1; i > 0; i--) {
450 		/*
451 		 * Break out of the loop if we've processed all the entries.
452 		 */
453 		if (ret_val == 0)
454 			break;
455 
456 		pfd = &fdtable[i];
457 
458 		if (pfd->fd < 0) {
459 			rem_pid((pid_t)0, i, DONT_CLEAN);
460 			continue;
461 		}
462 		/*
463 		 * POLLHUP	- Process terminated
464 		 */
465 		if (pfd->revents & POLLHUP) {
466 			psinfo_t psinfo;
467 
468 			if (pread(pfd->fd, &psinfo, sizeof (psinfo), (off_t)0)
469 			    != sizeof (psinfo)) {
470 				dprintf(("! %d: terminated, status 0x%.4x\n", \
471 				    (int)pidtable[i].pl_pid, psinfo.pr_wstat));
472 				pidtable[i].pl_status = psinfo.pr_wstat;
473 
474 			} else {
475 				dprintf(("! %d: terminated\n", \
476 				    (int)pidtable[i].pl_pid));
477 				pidtable[i].pl_status = 0;
478 			}
479 			/*
480 			 * PID gets removed when terminated only
481 			 */
482 			rem_pid((pid_t)0, i, CLEANIT);
483 			ret_val--;
484 			continue;
485 		}
486 		/*
487 		 * POLLNVAL and POLLERR
488 		 *	These error's shouldn't occurr but until their fixed
489 		 *	we perform some simple error recovery.
490 		 */
491 		if (pfd->revents & (POLLNVAL|POLLERR)) {
492 			dprintf(("Poll Err = %d pid = %d i = %d\n", \
493 			    pfd->revents, (int)pidtable[i].pl_pid, i));
494 
495 			pid = pidtable[i].pl_pid; /* Save pid for below */
496 			/*
497 			 * If its POLLNVAL we just remove the process for
498 			 * now, it will get picked up in the next scan.
499 			 * POLLERR pids get re-added after being deleted.
500 			 */
501 			if (pfd->revents & POLLNVAL) {
502 				rem_pid((pid_t)0, i, DONT_CLEAN);
503 			} else {			/* Else... POLLERR */
504 				rem_pid((pid_t)0, i, DONT_CLEAN);
505 				add_pid(pid);
506 			}
507 
508 			if (bad_error++ > MAX_POLL_ERRS) {
509 				bad_error = 0;
510 				return (0);	/* 0 Indicates severe error */
511 			}
512 			ret_val--;
513 			continue;
514 		}
515 
516 		/*
517 		 * No more bits should be set in revents but check anyway
518 		 */
519 		if (pfd->revents != 0) {
520 			dprintf(("%d: unknown err %d\n", \
521 			    (int)pidtable[i].pl_pid, pfd->revents));
522 
523 			rem_pid((pid_t)0, i, DONT_CLEAN);
524 			ret_val--;
525 
526 			if (bad_error++ > MAX_POLL_ERRS) {
527 				bad_error = 0;
528 				return (0);	/* 0 Indicates severe error */
529 			}
530 			return (1);
531 		}
532 	}
533 	return (1);			/* 1 Indicates Everything okay */
534 }
535 
536 /*
537  *		*** The Scanner ***
538  *
539  * scan_utmps()		- Scan the utmpx file.
540  *			  For each USER_PROCESS check
541  *			  if its alive or dead.  If alive and its not in
542  *			  our table to be watched, put it there.  If its
543  *			  dead, remove it from our table and clean it up.
544  */
545 
546 static void
547 scan_utmps()
548 {
549 	struct	utmpx	*utmpx;
550 	int	i;
551 
552 	dprintf(("Scan utmps\n"));
553 	/*
554 	 * Scan utmpx.
555 	 */
556 	setutxent();
557 	while ((utmpx = getutxent()) != NULL) {
558 		if (utmpx->ut_type == USER_PROCESS) {
559 			/*
560 			 * Is the process alive?
561 			 */
562 			if (proc_is_alive(utmpx->ut_pid)) {
563 				/*
564 				 * Yes, the process is alive, so add it if we
565 				 * don't have it in our table.
566 				 */
567 				if (find_pid(utmpx->ut_pid, &i) == 0)
568 					add_pid(utmpx->ut_pid);	/* No, add it */
569 			} else {
570 				/*
571 				 * No, the process is dead, so remove it if its
572 				 * in our table, otherwise just clean it.
573 				 */
574 				if (find_pid(utmpx->ut_pid, &i) == 1)
575 					rem_pid(utmpx->ut_pid, i, CLEANIT);
576 				else
577 					clean_utmpx_ent(utmpx);
578 			}
579 		}
580 	}
581 	/*
582 	 * Close it to flush the buffer.
583 	 */
584 	endutxent();
585 }
586 
587 
588 /*
589  *			*** Receiver Routines ***
590  */
591 
592 /*
593  * setup_pipe	- Set up the pipe to read pids over
594  */
595 
596 static void
597 setup_pipe()
598 {
599 
600 	struct statvfs statvfs_buf;
601 	/*
602 	 * This code & comments swiped from init and left stock since it works
603 	 */
604 
605 	if (Pfd < 0) {
606 		if ((statvfs(UTMPPIPE_DIR, &statvfs_buf) == 0) &&
607 		    ((statvfs_buf.f_flag & ST_RDONLY) == 0)) {
608 			(void) unlink(UTMPPIPE);
609 			(void) mknod(UTMPPIPE, S_IFIFO | 0600, 0);
610 		}
611 		Pfd = open(UTMPPIPE, O_RDWR | O_NDELAY);
612 	}
613 	if (Pfd < 0)
614 		nonfatal(UTMPPIPE);
615 	/*
616 	 * This code from init modified to be poll based instead of SIGPOLL,
617 	 * signal based.
618 	 */
619 
620 	if (Pfd >= 0) {
621 		/*
622 		 * Read pipe in message discard mode.  When read reads a
623 		 * pidrec size record, the remainder of the message will
624 		 * be discarded.  Though there shouldn't be any it will
625 		 * help resynch if someone else wrote some garbage.
626 		 */
627 		(void) ioctl(Pfd, I_SRDOPT, RMSGD);
628 	}
629 
630 	/*
631 	 * My code.  We use slot 0 in the table to hold the fd of the pipe
632 	 */
633 	add_pid(0);			/* Proc 0 guaranteed to get slot 0 */
634 	fdtable[0].fd = Pfd;		/* Pfd could be -1, should be okay */
635 	fdtable[0].events = POLLRDNORM;
636 }
637 
638 /*
639  * drain_pipe()		- The receiver routine that reads the pipe
640  */
641 
642 static void
643 drain_pipe()
644 {
645 	struct pidrec prec;
646 	register struct pidrec *p = &prec;
647 	int bytes_read;
648 	int i;
649 
650 	for (;;) {
651 		/*
652 		 * Important Note: Either read will really fail (in which case
653 		 * return is all we can do) or will get EAGAIN (Pfd was opened
654 		 * O_NDELAY), in which case we also want to return.
655 		 */
656 
657 		if ((bytes_read = read(Pfd, p, sizeof (struct pidrec))) !=
658 		    sizeof (struct pidrec))  {
659 			/*
660 			 * Something went wrong reading, so read until pipe
661 			 * is empty
662 			 */
663 			if (bytes_read > 0)
664 				while (read(Pfd, p, sizeof (struct pidrec)) > 0)
665 					;
666 			return;
667 		}
668 
669 		dprintf(("drain_pipe: Recd command %d, pid %d\n",
670 		    p->pd_type, (int)p->pd_pid));
671 		switch (p->pd_type) {
672 		case ADDPID:
673 			/*
674 			 * Check if we already have the process, adding it
675 			 * if we don't.
676 			 */
677 			if (find_pid(p->pd_pid, &i) == 0)
678 				add_pid(p->pd_pid);
679 			break;
680 
681 		case REMPID:
682 			rem_pid(p->pd_pid, -1, DONT_CLEAN);
683 			break;
684 		default:
685 			nonfatal("Bad message on utmppipe\n");
686 				break;
687 		}
688 	}
689 }
690 
691 
692 /*
693  *		*** Utilities for add and removing entries in the tables ***
694  */
695 
696 /*
697  * add_pid	- add a pid to the fd table and the pidtable.
698  *		  these tables are sorted tables for quick lookups.
699  *
700  */
701 static void
702 add_pid(pid_t pid)
703 {
704 	int fd = 0;
705 	int i = 0, move_amt;
706 	int j;
707 	static int first_time = 1;
708 
709 	/*
710 	 * Check to see if the pid is already in our table, or being passed
711 	 * pid zero.
712 	 */
713 	if (pidcnt != 0 && (find_pid(pid, &j) == 1 || pid == 0))
714 		return;
715 
716 	if (pidcnt >= Max_fds) {
717 		if (first_time == 1) {
718 			/*
719 			 * Print this error only once
720 			 */
721 			nonfatal("File Descriptor limit exceeded");
722 			first_time = 0;
723 		}
724 		return;
725 	}
726 	/*
727 	 * Open the /proc file checking if there's still a valid proc file.
728 	 */
729 	if (pid != 0 && (fd = proc_to_fd(pid)) == -1) {
730 		/*
731 		 * No so the process died before we got to watch for him
732 		 */
733 		return;
734 	}
735 
736 	/*
737 	 * We only do this code if we're not putting in the first element
738 	 * Which we know will be for proc zero which is used by setup_pipe
739 	 * for its pipe fd.
740 	 */
741 	if (pidcnt != 0) {
742 		for (i = 0; i < pidcnt; i++) {
743 			if (pid <= pidtable[i].pl_pid)
744 				break;
745 		}
746 
747 		/*
748 		 * Handle the case where we're not sticking our entry on the
749 		 * the end, or overwriting an existing entry.
750 		 */
751 		if (i != pidcnt && pid != pidtable[i].pl_pid) {
752 
753 			move_amt = pidcnt - i;
754 			/*
755 			 * Move table down
756 			 */
757 			if (move_amt != 0) {
758 				(void) memmove(&pidtable[i+1], &pidtable[i],
759 				    move_amt * sizeof (struct pidentry));
760 				(void) memmove(&fdtable[i+1], &fdtable[i],
761 				    move_amt * sizeof (pollfd_t));
762 			}
763 		}
764 	}
765 
766 	/*
767 	 * Fill in the events field for poll and copy the entry into the array
768 	 */
769 	fdtable[i].events = 0;
770 	fdtable[i].revents = 0;
771 	fdtable[i].fd = fd;
772 
773 	/*
774 	 * Likewise, setup pid field and pointer (index) to the fdtable entry
775 	 */
776 	pidtable[i].pl_pid = pid;
777 
778 	pidcnt++;			/* Bump the pid count */
779 	dprintf(("  add_pid: pid = %d fd = %d index = %d pidcnt = %d\n",
780 	    (int)pid, fd, i, pidcnt));
781 }
782 
783 
784 /*
785  * rem_pid	- Remove an entry from the table and check to see if its
786  *		  not in the utmpx file.
787  *		  If i != -1 don't look up the pid, use i as index
788  *
789  * pid          - Pid of process to clean or 0 if we don't know it
790  *
791  * i            - Index into table or -1 if we need to look it up
792  *
793  * clean_it     - Clean the entry, or just remove from table?
794  */
795 
796 static void
797 rem_pid(pid_t pid, int i, int clean_it)
798 {
799 	int move_amt;
800 
801 	dprintf(("  rem_pid: pid = %d i = %d", (int)pid, i));
802 
803 	/*
804 	 * Don't allow slot 0 in the table to be removed - utmppipe fd
805 	 */
806 	if ((i == -1 && pid == 0) || (i == 0))	{
807 		dprintf((" - attempted to remove proc 0\n"));
808 		return;
809 	}
810 
811 	if (i != -1 || find_pid(pid, &i) == 1) {	/* Found the entry */
812 		(void) close(fdtable[i].fd);	/* We're done with the fd */
813 
814 		dprintf((" fd = %d\n", fdtable[i].fd));
815 
816 		if (clean_it == CLEANIT)
817 			clean_entry(i);
818 
819 		move_amt = (pidcnt - i) - 1;
820 		/*
821 		 * Remove entries from the tables.
822 		 */
823 		(void) memmove(&pidtable[i], &pidtable[i+1],
824 		    move_amt * sizeof (struct pidentry));
825 
826 		(void) memmove(&fdtable[i], &fdtable[i+1],
827 		    move_amt * sizeof (pollfd_t));
828 
829 		/*
830 		 * decrement the pid count - one less pid to worry about
831 		 */
832 		pidcnt--;
833 	}
834 	if (i == -1)
835 		dprintf((" - entry not found \n"));
836 }
837 
838 
839 /*
840  * find_pid	- Returns an index into the pidtable of the specifed pid,
841  *		  else -1 if not found
842  */
843 
844 static int
845 find_pid(pid_t pid, int *i)
846 {
847 	struct pidentry pe;
848 	struct pidentry *p;
849 
850 	pe.pl_pid = pid;
851 	p = bsearch(&pe, pidtable, pidcnt, sizeof (struct pidentry), pidcmp);
852 
853 	if (p == NULL)
854 		return (0);
855 	else {
856 		*i = p - (struct pidentry *)pidtable;
857 		return (1);
858 	}
859 }
860 
861 
862 /*
863  * Pidcmp - Used by besearch for sorting and finding  process IDs.
864  */
865 
866 static int
867 pidcmp(struct pidentry *a, struct pidentry *b)
868 {
869 	if (b == NULL || a == NULL)
870 		return (0);
871 	return (a->pl_pid - b->pl_pid);
872 }
873 
874 
875 /*
876  * proc_to_fd	- Take a process ID and return an open file descriptor to the
877  *		  /proc file for the specified process.
878  */
879 static int
880 proc_to_fd(pid_t pid)
881 {
882 	char procname[64];
883 	int fd, dfd;
884 
885 	(void) sprintf(procname, "/proc/%d/psinfo", (int)pid);
886 
887 	if ((fd = open(procname, O_RDONLY)) >= 0) {
888 		/*
889 		 * dup the fd above the low order values to assure
890 		 * stdio works for other fds - paranoia.
891 		 */
892 		if (fd < EXTRA_MARGIN) {
893 			dfd = fcntl(fd, F_DUPFD, EXTRA_MARGIN);
894 			if (dfd > 0) {
895 				(void) close(fd);
896 				fd = dfd;
897 			}
898 		}
899 		/*
900 		 * More paranoia - set the close on exec flag
901 		 */
902 		(void) fcntl(fd, F_SETFD, 1);
903 		return (fd);
904 	}
905 	if (errno == ENOENT)
906 		return (-1);
907 
908 	if (errno == EMFILE) {
909 		/*
910 		 * This is fatal, since libc won't be able to allocate
911 		 * any fds for the pututxline() routines
912 		 */
913 		fatal("Out of file descriptors");
914 	}
915 	fatal(procname);		/* Only get here on error */
916 	return (-1);
917 }
918 
919 
920 /*
921  *		*** Utmpx Cleaning Utilities ***
922  */
923 
924 /*
925  * Clean_entry	- Cleans the specified entry - where i is an index
926  *		  into the pid_table.
927  */
928 static void
929 clean_entry(int i)
930 {
931 	struct utmpx *u;
932 
933 	if (pidcnt == 0)
934 		return;
935 
936 	dprintf(("    Cleaning %d\n", (int)pidtable[i].pl_pid));
937 
938 	/*
939 	 * Double check if the process is dead.
940 	 */
941 	if (proc_is_alive(pidtable[i].pl_pid)) {
942 		dprintf(("      Bad attempt to clean %d\n",
943 		    (int)pidtable[i].pl_pid));
944 		return;
945 	}
946 
947 	/*
948 	 * Find the entry that corresponds to this pid.
949 	 * Do nothing if entry not found in utmpx file.
950 	 */
951 	setutxent();
952 	while ((u = getutxent()) != NULL) {
953 		if (u->ut_pid == pidtable[i].pl_pid) {
954 			if (u->ut_type == USER_PROCESS) {
955 				clean_utmpx_ent(u);
956 			}
957 		}
958 	}
959 	endutxent();
960 }
961 
962 
963 /*
964  * clean_utmpx_ent	- Clean a utmpx entry
965  */
966 
967 static void
968 clean_utmpx_ent(struct utmpx *u)
969 {
970 	dprintf(("      clean_utmpx_ent: %d\n", (int)u->ut_pid));
971 	u->ut_type = DEAD_PROCESS;
972 	(void) time(&u->ut_xtime);
973 	(void) pututxline(u);
974 	updwtmpx(WTMPX_FILE, u);
975 	/*
976 	 * XXX update wtmp for ! nonuserx entries?
977 	 */
978 }
979 
980 /*
981  *		*** Error Handling and Debugging Routines ***
982  */
983 
984 /*
985  * fatal - Catastrophic failure
986  */
987 
988 static void
989 fatal(char *str)
990 {
991 	int oerrno = errno;
992 
993 	syslog(LOG_ALERT, "%s", str);
994 	if (Debug == 1) {
995 		if ((errno = oerrno) != 0)
996 			perror(prog_name);
997 		dprintf(("%s\n", str));
998 	}
999 	exit(1);
1000 }
1001 
1002 /*
1003  * nonfatal - Non-Catastrophic failure - print message and errno
1004  */
1005 
1006 static void
1007 nonfatal(char *str)
1008 {
1009 	syslog(LOG_WARNING, "%s", str);
1010 
1011 	if (Debug == 1) {
1012 		if (errno != 0)
1013 			perror(prog_name);
1014 		dprintf(("%c%s\n", 7, str));
1015 		print_tables();
1016 		(void) sleep(5);	/* Time to read debug messages */
1017 	}
1018 }
1019 
1020 /*
1021  * print_tables	- Print internal tables - for debugging
1022  */
1023 
1024 static void
1025 print_tables()
1026 {
1027 	int i;
1028 
1029 	if (Debug == 0)
1030 		return;
1031 
1032 	dprintf(("pidtable: "));
1033 	for (i = 0; i < pidcnt; i++)
1034 		dprintf(("%d: %d  ", i, (int)pidtable[i].pl_pid));
1035 	dprintf(("\n"));
1036 	dprintf(("fdtable:  "));
1037 	for (i = 0; i < pidcnt; i++)
1038 		dprintf(("%d: %d  ", i, fdtable[i].fd));
1039 	dprintf(("\n"));
1040 }
1041 
1042 /*
1043  * proc_is_alive	- Check to see if a process is alive AND its
1044  *			  not a zombie.  Returns 1 if process is alive
1045  *			  and zero if it is dead or a zombie.
1046  */
1047 
1048 static int
1049 proc_is_alive(pid_t pid)
1050 {
1051 	char psinfoname[64];
1052 	int fd;
1053 	psinfo_t psinfo;
1054 
1055 	if (kill(pid, 0) != 0)
1056 		return (0);		/* Kill failed - no process */
1057 
1058 	/*
1059 	 * The process exists, so check if it's a zombie.
1060 	 */
1061 	(void) sprintf(psinfoname, "/proc/%d/psinfo", (int)pid);
1062 
1063 	if ((fd = open(psinfoname, O_RDONLY)) < 0 ||
1064 	    read(fd, &psinfo, sizeof (psinfo)) != sizeof (psinfo)) {
1065 		/*
1066 		 * We either couldn't open the proc, or we did but the
1067 		 * read of the psinfo file failed, so pid is nonexistent.
1068 		 */
1069 		psinfo.pr_nlwp = 0;
1070 	}
1071 	if (fd >= 0)
1072 		(void) close(fd);
1073 
1074 	/* if pr_nlwp == 0, process is a zombie */
1075 	return (psinfo.pr_nlwp != 0);
1076 }
1077 
1078 /*
1079  * warn_utmp -	/var/adm/utmp has been deprecated. It should no longer
1080  *		be used.  Applications that try to directly manipulate
1081  *		it may cause problems. Since the file is no longer
1082  *		shipped, if it appears on a system it's because an
1083  *		old application created it.  We'll have utmpd
1084  *		complain about it periodically.
1085  */
1086 
1087 static void
1088 warn_utmp()
1089 {
1090 	struct stat s;
1091 
1092 	if (lstat(UTMP_FILE, &s) == 0 &&
1093 	    s.st_size % sizeof (struct utmp) == 0) {
1094 		nonfatal("WARNING: /var/adm/utmp exists!\nSee "
1095 		    "utmp(4) for more information");
1096 	}
1097 }
1098 
1099 /*
1100  * validate_default - validate and assign defaults.
1101  */
1102 
1103 static int
1104 validate_default(char *defp, int *flag)
1105 {
1106 	long lval;
1107 	char *endptr;
1108 
1109 	errno = 0;
1110 	lval = strtol(defp, &endptr, 10);
1111 
1112 	if (errno != 0 || lval > INT_MAX || lval <= 0)
1113 		return (-1);
1114 
1115 	while (isspace(*endptr) != 0)
1116 		endptr++;
1117 
1118 	if (*endptr != '\0')
1119 		return (-1);
1120 
1121 	*flag = lval;
1122 	return (0);
1123 }
1124