xref: /netbsd-src/usr.sbin/rpc.statd/statd.c (revision 4472dbe5e3bd91ef2540bada7a7ca7384627ff9b)
1 /*	$NetBSD: statd.c,v 1.14 1999/06/10 05:53:51 scottr Exp $	*/
2 
3 /*
4  * Copyright (c) 1997 Christos Zoulas. All rights reserved.
5  * Copyright (c) 1995
6  *	A.R. Gordon (andrew.gordon@net-tel.co.uk).  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed for the FreeBSD project
19  *	This product includes software developed by Christos Zoulas.
20  * 4. Neither the name of the author nor the names of any co-contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY ANDREW GORDON AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  */
37 
38 #include <sys/cdefs.h>
39 #ifndef lint
40 __RCSID("$NetBSD: statd.c,v 1.14 1999/06/10 05:53:51 scottr Exp $");
41 #endif
42 
43 /* main() function for status monitor daemon.  Some of the code in this	*/
44 /* file was generated by running rpcgen /usr/include/rpcsvc/sm_inter.x	*/
45 /* The actual program logic is in the file procs.c			*/
46 
47 #include <sys/param.h>
48 
49 #include <err.h>
50 #include <ctype.h>
51 #include <errno.h>
52 #include <fcntl.h>
53 #include <signal.h>
54 #include <stdio.h>
55 #include <stdlib.h>
56 #include <string.h>
57 #include <syslog.h>
58 #include <unistd.h>
59 #include <util.h>
60 #include <db.h>
61 
62 #include <rpc/rpc.h>
63 
64 #include "statd.h"
65 
66 struct sigaction sa;
67 int     	debug = 0;		/* Controls syslog() for debug msgs */
68 int     	_rpcsvcdirty = 0;	/* XXX ??? */
69 static DB	*db;			/* Database file */
70 
71 Header		 status_info;
72 
73 static char undefdata[] = "\0\1\2\3\4\5\6\7";
74 static DBT undefkey = {
75 	undefdata,
76 	sizeof(undefdata)
77 };
78 extern char *__progname;
79 
80 
81 /* statd.c */
82 static int walk_one __P((int (*fun )__P ((DBT *, DBT *, void *)), DBT *, DBT *, void *));
83 static int walk_db __P((int (*fun )__P ((DBT *, DBT *, void *)), void *));
84 static int reset_host __P((DBT *, DBT *, void *));
85 static int check_work __P((DBT *, DBT *, void *));
86 static int unmon_host __P((DBT *, DBT *, void *));
87 static int notify_one __P((DBT *, DBT *, void *));
88 static void init_file __P((char *));
89 static int notify_one_host __P((char *));
90 static void die __P((int)) __attribute__((__noreturn__));
91 
92 int main __P((int, char **));
93 
94 int
95 main(argc, argv)
96 	int argc;
97 	char **argv;
98 {
99 	SVCXPRT *transp;
100 	int ch;
101 
102 	while ((ch = getopt(argc, argv, "d")) != (-1)) {
103 		switch (ch) {
104 		case 'd':
105 			debug = 1;
106 			break;
107 		default:
108 		case '?':
109 			(void)fprintf(stderr, "usage: %s [-d]\n", __progname);
110 			exit(1);
111 			/* NOTREACHED */
112 		}
113 	}
114 	(void)pmap_unset(SM_PROG, SM_VERS);
115 
116 	transp = svcudp_create(RPC_ANYSOCK);
117 	if (transp == NULL) {
118 		errx(1, "cannot create udp service.");
119 		/* NOTREACHED */
120 	}
121 	if (!svc_register(transp, SM_PROG, SM_VERS, sm_prog_1, IPPROTO_UDP)) {
122 		errx(1, "unable to register (SM_PROG, SM_VERS, udp).");
123 		/* NOTREACHED */
124 	}
125 	transp = svctcp_create(RPC_ANYSOCK, 0, 0);
126 	if (transp == NULL) {
127 		errx(1, "cannot create tcp service.");
128 		/* NOTREACHED */
129 	}
130 	if (!svc_register(transp, SM_PROG, SM_VERS, sm_prog_1, IPPROTO_TCP)) {
131 		errx(1, "unable to register (SM_PROG, SM_VERS, tcp).");
132 		/* NOTREACHED */
133 	}
134 	init_file("/var/db/statd.status");
135 
136 	/*
137 	 * Note that it is NOT sensible to run this program from inetd - the
138 	 * protocol assumes that it will run immediately at boot time.
139 	 */
140 	if (!debug)
141 		daemon(0, 0);
142 	pidfile(NULL);
143 	openlog("rpc.statd", 0, LOG_DAEMON);
144 	if (debug)
145 		syslog(LOG_INFO, "Starting - debug enabled");
146 	else
147 		syslog(LOG_INFO, "Starting");
148 
149 	sa.sa_handler = die;
150 	sa.sa_flags = 0;
151 	sigemptyset(&sa.sa_mask);
152 	(void)sigaction(SIGTERM, &sa, NULL);
153 	(void)sigaction(SIGQUIT, &sa, NULL);
154 	(void)sigaction(SIGHUP, &sa, NULL);
155 	(void)sigaction(SIGINT, &sa, NULL);
156 
157 	sa.sa_handler = SIG_IGN;
158 	sa.sa_flags = SA_RESTART;
159 	sigemptyset(&sa.sa_mask);
160 	sigaddset(&sa.sa_mask, SIGALRM);
161 
162 	/* Initialisation now complete - start operating */
163 
164 	/* Notify hosts that need it */
165 	notify_handler(0);
166 
167 	while (1)
168 		svc_run();		/* Should never return */
169 	die(0);
170 }
171 
172 /* notify_handler ---------------------------------------------------------- */
173 /*
174  * Purpose:	Catch SIGALRM and collect process status
175  * Returns:	Nothing.
176  * Notes:	No special action required, other than to collect the
177  *		process status and hence allow the child to die:
178  *		we only use child processes for asynchronous transmission
179  *		of SM_NOTIFY to other systems, so it is normal for the
180  *		children to exit when they have done their work.
181  */
182 void
183 notify_handler(sig)
184 	int sig;
185 {
186 	time_t now;
187 
188 	NO_ALARM;
189 	sa.sa_handler = SIG_IGN;
190 	(void)sigaction(SIGALRM, &sa, NULL);
191 
192 	now = time(NULL);
193 
194 	(void) walk_db(notify_one, &now);
195 
196 	if (walk_db(check_work, &now) == 0) {
197 		/*
198 		 * No more work to be done.
199 		 */
200 		CLR_ALARM;
201 		return;
202 	}
203 	sync_file();
204 	ALARM;
205 	alarm(5);
206 }
207 
208 /* sync_file --------------------------------------------------------------- */
209 /*
210  * Purpose:	Packaged call of msync() to flush changes to mmap()ed file
211  * Returns:	Nothing.  Errors to syslog.
212  */
213 void
214 sync_file()
215 {
216 	DBT data;
217 
218 	data.data = &status_info;
219 	data.size = sizeof(status_info);
220 	switch ((*db->put)(db, &undefkey, &data, 0)) {
221 	case 0:
222 		return;
223 	case -1:
224 		goto bad;
225 	default:
226 		abort();
227 	}
228 	if ((*db->sync)(db, 0) == -1) {
229 bad:
230 		syslog(LOG_ERR, "database corrupted %m");
231 		die(1);
232 	}
233 }
234 
235 /* change_host -------------------------------------------------------------- */
236 /*
237  * Purpose:	Update/Create an entry for host
238  * Returns:	Nothing
239  * Notes:
240  *
241  */
242 void
243 change_host(hostname, hp)
244 	char *hostname;
245 	HostInfo *hp;
246 {
247 	DBT key, data;
248 	char *ptr;
249 
250 	for (ptr = hostname; *ptr; ptr++)
251 		if (isupper((unsigned char) *ptr))
252 			*ptr = tolower((unsigned char) *ptr);
253 
254 	key.data = hostname;
255 	key.size = ptr - hostname + 1;
256 	data.data = hp;
257 	data.size = sizeof(*hp);
258 
259 	switch ((*db->put)(db, &key, &data, 0)) {
260 	case -1:
261 		syslog(LOG_ERR, "database corrupted %m");
262 		die(1);
263 	case 0:
264 		return;
265 	default:
266 		abort();
267 	}
268 }
269 
270 
271 /* find_host -------------------------------------------------------------- */
272 /*
273  * Purpose:	Find the entry in the status file for a given host
274  * Returns:	Copy of entry in hd, or NULL
275  * Notes:
276  *
277  */
278 HostInfo *
279 find_host(hostname, hp)
280 	char *hostname;
281 	HostInfo *hp;
282 {
283 	DBT key, data;
284 	char *ptr;
285 
286 	for (ptr = hostname; *ptr; ptr++)
287 		if (isupper((unsigned char) *ptr))
288 			*ptr = tolower((unsigned char) *ptr);
289 
290 	key.data = hostname;
291 	key.size = ptr - hostname + 1;
292 	switch ((*db->get)(db, &key, &data, 0)) {
293 	case 0:
294 		if (data.size != sizeof(*hp))
295 			goto bad;
296 		return memcpy(hp, data.data, sizeof(*hp));
297 	case 1:
298 		return NULL;
299 	case -1:
300 		goto bad;
301 	default:
302 		abort();
303 	}
304 
305 bad:
306 	syslog(LOG_ERR, "Database corrupted %m");
307 	return NULL;
308 }
309 
310 /* walk_one ------------------------------------------------------------- */
311 /*
312  * Purpose:	Call the given function if the element is valid
313  * Returns:	Nothing - exits on error
314  * Notes:
315  */
316 static int
317 walk_one(fun, key, data, ptr)
318 	int (*fun) __P((DBT *, DBT *, void *));
319 	DBT *key, *data;
320 	void *ptr;
321 {
322 	if (key->size == undefkey.size &&
323 	    memcmp(key->data, undefkey.data, key->size) == 0)
324 		return 0;
325 	if (data->size != sizeof(HostInfo)) {
326 		syslog(LOG_ERR, "Bad data in database");
327 		die(1);
328 	}
329 
330 	return (*fun)(key, data, ptr);
331 }
332 
333 /* walk_db -------------------------------------------------------------- */
334 /*
335  * Purpose:	Iterate over all elements calling the given function
336  * Returns:	-1 if function failed, 0 on success
337  * Notes:
338  */
339 static int
340 walk_db(fun, ptr)
341 	int (*fun) __P((DBT *, DBT *, void *));
342 	void *ptr;
343 {
344 	DBT key, data;
345 
346 	switch ((*db->seq)(db, &key, &data, R_FIRST)) {
347 	case -1:
348 		goto bad;
349 	case 1:
350 		/* We should have at least the magic entry at this point */
351 		abort();
352 	case 0:
353 		if (walk_one(fun, &key, &data, ptr) == -1)
354 			return -1;
355 		break;
356 	default:
357 		abort();
358 	}
359 
360 
361 	for (;;)
362 		switch ((*db->seq)(db, &key, &data, R_NEXT)) {
363 		case -1:
364 			goto bad;
365 		case 1:
366 			if (walk_one(fun, &key, &data, ptr) == -1)
367 				return -1;
368 			break;
369 		case 0:
370 			return 0;
371 		default:
372 			abort();
373 		}
374 bad:
375 	syslog(LOG_ERR, "Corrupted database %m");
376 	die(1);
377 }
378 
379 /* reset_host ------------------------------------------------------------ */
380 /*
381  * Purpose:	Clean up existing hosts in file.
382  * Returns:	Always success 0.
383  * Notes:	Clean-up of existing file - monitored hosts will have a
384  *		pointer to a list of clients, which refers to memory in
385  *		the previous incarnation of the program and so are
386  *		meaningless now.  These pointers are zeroed and the fact
387  *		that the host was previously monitored is recorded by
388  *		setting the notifyReqd flag, which will in due course
389  *		cause a SM_NOTIFY to be sent.
390  *
391  *		Note that if we crash twice in quick succession, some hosts
392  *		may already have notifyReqd set, where we didn't manage to
393  *		notify them before the second crash occurred.
394  */
395 static int
396 reset_host(key, data, ptr)
397 	DBT *key, *data;
398 	void *ptr;
399 {
400 	HostInfo *hi = data->data;
401 
402 	if (hi->monList) {
403 		hi->notifyReqd = *(time_t *) data;
404 		hi->attempts = 0;
405 		hi->monList = NULL;
406 	}
407 	return 0;
408 }
409 
410 /* check_work ------------------------------------------------------------ */
411 /*
412  * Purpose:	Check if there is work to be done.
413  * Returns:	0 if there is no work to be done -1 if there is.
414  * Notes:
415  */
416 static int
417 check_work(key, data, ptr)
418 	DBT *key, *data;
419 	void *ptr;
420 {
421 	HostInfo *hi = data->data;
422 
423 	return hi->notifyReqd ? -1 : 0;
424 }
425 
426 /* unmon_host ------------------------------------------------------------ */
427 /*
428  * Purpose:	Unmonitor a host
429  * Returns:	0
430  * Notes:
431  */
432 static int
433 unmon_host(key, data, ptr)
434 	DBT *key, *data;
435 	void *ptr;
436 {
437 	char *name = key->data;
438 	HostInfo *hi = data->data;
439 
440 	if (do_unmon(name, hi, ptr))
441 		change_host(name, hi);
442 	return 0;
443 }
444 
445 /* notify_one ------------------------------------------------------------ */
446 /*
447  * Purpose:	Notify one host.
448  * Returns:	0 if success -1 on failure
449  * Notes:
450  */
451 static int
452 notify_one(key, data, ptr)
453 	DBT *key, *data;
454 	void *ptr;
455 {
456 	time_t now = *(time_t *) ptr;
457 	char *name = key->data;
458 	HostInfo *hi = data->data;
459 
460 	if (hi->notifyReqd == 0 || hi->notifyReqd > now)
461 		return 0;
462 
463 	if (notify_one_host(name)) {
464 give_up:
465 		hi->notifyReqd = 0;
466 		hi->attempts = 0;
467 		switch ((*db->put)(db, key, data, 0)) {
468 		case -1:
469 			syslog(LOG_ERR, "Error storing %s (%m)", name);
470 		case 0:
471 			return 0;
472 
473 		default:
474 			abort();
475 		}
476 	}
477 	else {
478 		/*
479 		 * If one of the initial attempts fails, we wait
480 		 * for a while and have another go.  This is necessary
481 		 * because when we have crashed, (eg. a power outage)
482 		 * it is quite possible that we won't be able to
483 		 * contact all monitored hosts immediately on restart,
484 		 * either because they crashed too and take longer
485 		 * to come up (in which case the notification isn't
486 		 * really required), or more importantly if some
487 		 * router etc. needed to reach the monitored host
488 		 * has not come back up yet.  In this case, we will
489 		 * be a bit late in re-establishing locks (after the
490 		 * grace period) but that is the best we can do.  We
491 		 * try 10 times at 5 sec intervals, 10 more times at
492 		 * 1 minute intervals, then 24 more times at hourly
493 		 * intervals, finally giving up altogether if the
494 		 * host hasn't come back to life after 24 hours.
495 		 */
496 		if (hi->attempts++ >= 44)
497 			goto give_up;
498 		else if (hi->attempts < 10)
499 			hi->notifyReqd += 5;
500 		else if (hi->attempts < 20)
501 			hi->notifyReqd += 60;
502 		else
503 			hi->notifyReqd += 60 * 60;
504 		return -1;
505 	}
506 }
507 
508 /* init_file -------------------------------------------------------------- */
509 /*
510  * Purpose:	Open file, create if necessary, initialise it.
511  * Returns:	Nothing - exits on error
512  * Notes:	Called before process becomes daemon, hence logs to
513  *		stderr rather than syslog.
514  *		Opens the file, then mmap()s it for ease of access.
515  *		Also performs initial clean-up of the file, zeroing
516  *		monitor list pointers, setting the notifyReqd flag in
517  *		all hosts that had a monitor list, and incrementing
518  *		the state number to the next even value.
519  */
520 static void
521 init_file(filename)
522 	char *filename;
523 {
524 	DBT data;
525 
526 	db = dbopen(filename, O_RDWR|O_CREAT|O_NDELAY|O_EXLOCK, 0644, DB_HASH,
527 	    NULL);
528 	if (db == NULL)
529 		err(1, "Cannot open `%s'", filename);
530 
531 	switch ((*db->get)(db, &undefkey, &data, 0)) {
532 	case 1:
533 		/* New database */
534 		(void)memset(&status_info, 0, sizeof(status_info));
535 		sync_file();
536 		return;
537 
538 	case -1:
539 		err(1, "error accessing database (%m)");
540 	case 0:
541 		/* Existing database */
542 		if (data.size != sizeof(status_info))
543 			errx(1, "database corrupted %lu != %lu",
544 			    (u_long)data.size, (u_long)sizeof(status_info));
545 		break;
546 	default:
547 		abort();
548 	}
549 
550 	reset_database();
551 	return;
552 }
553 
554 /* reset_database --------------------------------------------------------- */
555 /*
556  * Purpose:	Clears the statd database
557  * Returns:	Nothing
558  * Notes:	If this is not called on reset, it will leak memory.
559  */
560 void
561 reset_database()
562 {
563 	time_t now = time(NULL);
564 	walk_db(reset_host, &now);
565 
566 	/* Select the next higher even number for the state counter */
567 	status_info.ourState =
568 	    (status_info.ourState + 2) & 0xfffffffe;
569 	status_info.ourState++;	/* XXX - ??? */
570 	sync_file();
571 }
572 
573 /* unmon_hosts --------------------------------------------------------- */
574 /*
575  * Purpose:	Unmonitor all the hosts
576  * Returns:	Nothing
577  * Notes:
578  */
579 void
580 unmon_hosts()
581 {
582 	time_t now = time(NULL);
583 	walk_db(unmon_host, &now);
584 	sync_file();
585 }
586 
587 static int
588 notify_one_host(hostname)
589 	char *hostname;
590 {
591 	struct timeval timeout = {20, 0};	/* 20 secs timeout */
592 	CLIENT *cli;
593 	char dummy;
594 	stat_chge arg;
595 	char our_hostname[MAXHOSTNAMELEN + 1];
596 
597 	gethostname(our_hostname, sizeof(our_hostname));
598 	our_hostname[sizeof(our_hostname) - 1] = '\0';
599 	our_hostname[SM_MAXSTRLEN] = '\0';
600 	arg.mon_name = our_hostname;
601 	arg.state = status_info.ourState;
602 
603 	if (debug)
604 		syslog(LOG_DEBUG, "Sending SM_NOTIFY to host %s from %s",
605 		    hostname, our_hostname);
606 
607 	cli = clnt_create(hostname, SM_PROG, SM_VERS, "udp");
608 	if (!cli) {
609 		syslog(LOG_ERR, "Failed to contact host %s%s", hostname,
610 		    clnt_spcreateerror(""));
611 		return (FALSE);
612 	}
613 	if (clnt_call(cli, SM_NOTIFY, xdr_stat_chge, &arg, xdr_void,
614 	    &dummy, timeout) != RPC_SUCCESS) {
615 		syslog(LOG_ERR, "Failed to contact rpc.statd at host %s",
616 		    hostname);
617 		clnt_destroy(cli);
618 		return (FALSE);
619 	}
620 	clnt_destroy(cli);
621 	return (TRUE);
622 }
623 
624 
625 static void
626 die(n)
627 	int n;
628 {
629 	(*db->close)(db);
630 	exit(n);
631 }
632