xref: /netbsd-src/usr.sbin/rpc.statd/statd.c (revision fdecd6a253f999ae92b139670d9e15cc9df4497c)
1 /*	$NetBSD: statd.c,v 1.2 1997/05/17 15:52:52 christos Exp $	*/
2 
3 /*
4  * Copyright (c) 1995
5  *	A.R. Gordon (andrew.gordon@net-tel.co.uk).  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed for the FreeBSD project
18  * 4. Neither the name of the author nor the names of any co-contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY ANDREW GORDON AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 
37 /* main() function for status monitor daemon.  Some of the code in this	*/
38 /* file was generated by running rpcgen /usr/include/rpcsvc/sm_inter.x	*/
39 /* The actual program logic is in the file procs.c			*/
40 
41 #include <stdio.h>
42 #include <errno.h>
43 #include <fcntl.h>
44 #include <signal.h>
45 #include <string.h>
46 #include <syslog.h>
47 #include <unistd.h>
48 #include <sys/types.h>
49 #include <sys/mman.h>
50 #include <sys/wait.h>
51 
52 #include <rpc/rpc.h>
53 
54 #include "statd.h"
55 
56 #ifndef lint
57 static char rcsid[] = "$NetBSD: statd.c,v 1.2 1997/05/17 15:52:52 christos Exp $";
58 #endif				/* not lint */
59 
60 int     	debug = 0;		/* Controls syslog() for debug msgs */
61 int     	_rpcsvcdirty = 0;	/* XXX ??? */
62 FileLayout	*status_info;		/* Pointer to mmap()ed status file */
63 static int	status_fd;		/* File descriptor for the open file */
64 static off_t	status_file_len;	/* Current on-disc length of file */
65 
66 extern void sm_prog_1 __P((struct svc_req *, SVCXPRT *));
67 static void handle_sigchld __P((int));
68 
69 main(argc, argv)
70 	int argc;
71 	char **argv;
72 {
73 	SVCXPRT *transp;
74 	struct sigaction sa;
75 	int ch;
76 
77 	while ((ch = getopt(argc, argv, "d")) != (-1)) {
78 		switch (ch) {
79 		case 'd':
80 			debug = 1;
81 			break;
82 		default:
83 		case '?':
84 			errx(1, "usage: rpc.statd [-d]");
85 			/* NOTREACHED */
86 		}
87 	}
88 	(void)pmap_unset(SM_PROG, SM_VERS);
89 
90 	transp = svcudp_create(RPC_ANYSOCK);
91 	if (transp == NULL) {
92 		errx(1, "cannot create udp service.");
93 		/* NOTREACHED */
94 	}
95 	if (!svc_register(transp, SM_PROG, SM_VERS, sm_prog_1, IPPROTO_UDP)) {
96 		errx(1, "unable to register (SM_PROG, SM_VERS, udp).");
97 		/* NOTREACHED */
98 	}
99 	transp = svctcp_create(RPC_ANYSOCK, 0, 0);
100 	if (transp == NULL) {
101 		errx(1, "cannot create tcp service.");
102 		/* NOTREACHED */
103 	}
104 	if (!svc_register(transp, SM_PROG, SM_VERS, sm_prog_1, IPPROTO_TCP)) {
105 		errx(1, "unable to register (SM_PROG, SM_VERS, tcp).");
106 		/* NOTREACHED */
107 	}
108 	init_file("/var/db/statd.status");
109 
110 	/*
111 	 * Note that it is NOT sensible to run this program from inetd - the
112 	 * protocol assumes that it will run immediately at boot time.
113 	 */
114 	daemon(0, 0);
115 	openlog("rpc.statd", 0, LOG_DAEMON);
116 	if (debug)
117 		syslog(LOG_INFO, "Starting - debug enabled");
118 	else
119 		syslog(LOG_INFO, "Starting");
120 
121 	/* Install signal handler to collect exit status of child processes */
122 	sa.sa_handler = handle_sigchld;
123 	sigemptyset(&sa.sa_mask);
124 	sigaddset(&sa.sa_mask, SIGCHLD);
125 	sa.sa_flags = SA_RESTART;
126 	sigaction(SIGCHLD, &sa, NULL);
127 
128 	/* Initialisation now complete - start operating */
129 
130 	/*
131 	 * notify_hosts() forks a process (if necessary) to do the
132 	 * SM_NOTIFY calls, which may be slow.
133 	 */
134 	notify_hosts();
135 
136 	svc_run();		/* Should never return */
137 	exit(1);
138 }
139 
140 /* handle_sigchld ---------------------------------------------------------- */
141 /*
142  * Purpose:	Catch SIGCHLD and collect process status
143  * Returns:	Nothing.
144  * Notes:	No special action required, other than to collect the
145  *		process status and hence allow the child to die:
146  *		we only use child processes for asynchronous transmission
147  *		of SM_NOTIFY to other systems, so it is normal for the
148  *		children to exit when they have done their work.
149  */
150 static void
151 handle_sigchld(sig)
152 	int sig;
153 {
154 	int     pid, status;
155 	pid = wait4(-1, &status, WNOHANG, (struct rusage *) 0);
156 	if (!pid)
157 		syslog(LOG_ERR, "Phantom SIGCHLD??");
158 	else if (status)
159 		syslog(LOG_ERR, "Child %d failed with status %d", pid,
160 		    WEXITSTATUS(status));
161 	else if (debug)
162 		syslog(LOG_DEBUG, "Child %d exited OK", pid);
163 }
164 
165 /* sync_file --------------------------------------------------------------- */
166 /*
167  * Purpose:	Packaged call of msync() to flush changes to mmap()ed file
168  * Returns:	Nothing.  Errors to syslog.
169  */
170 void
171 sync_file()
172 {
173 	if (msync((void *)status_info, 0) < 0)
174 		syslog(LOG_ERR, "msync() failed: %s", strerror(errno));
175 }
176 
177 /* find_host -------------------------------------------------------------- */
178 /*
179  * Purpose:	Find the entry in the status file for a given host
180  * Returns:	Pointer to that entry in the mmap() region, or NULL.
181  * Notes:	Also creates entries if requested.
182  *		Failure to create also returns NULL.
183  */
184 HostInfo *
185 find_host(hostname, create)
186 	char *hostname;
187 	int create;
188 {
189 	HostInfo *hp;
190 	HostInfo *spare_slot = NULL;
191 	HostInfo *result = NULL;
192 	int i;
193 
194 	for (i = 0, hp = status_info->hosts; i < status_info->noOfHosts;
195 	    i++, hp++) {
196 		if (!strncasecmp(hostname, hp->hostname, SM_MAXSTRLEN)) {
197 			result = hp;
198 			break;
199 		}
200 		if (!spare_slot && !hp->monList && !hp->notifyReqd)
201 			spare_slot = hp;
202 	}
203 
204 	/* Return if entry found, or if not asked to create one. */
205 	if (result || !create)
206 		return (result);
207 
208 	/*
209 	 * Now create an entry, using the spare slot if one was found or
210 	 * adding to the end of the list otherwise, extending file if req'd
211 	 */
212 	if (!spare_slot) {
213 		off_t desired_size;
214 		spare_slot = &status_info->hosts[status_info->noOfHosts];
215 		desired_size = ((char *)spare_slot - (char *)status_info) +
216 		    sizeof(HostInfo);
217 
218 		if (desired_size > status_file_len) {
219 			/* Extend file by writing 1 byte of junk at the
220 			 * desired end pos	 */
221 			lseek(status_fd, desired_size - 1, SEEK_SET);
222 			i = write(status_fd, &i, 1);
223 			if (i < 1) {
224 				syslog(LOG_ERR, "Unable to extend status file");
225 				return (NULL);
226 			}
227 			status_file_len = desired_size;
228 		}
229 		status_info->noOfHosts++;
230 	}
231 	/*
232 	 * Initialise the spare slot that has been found/created
233 	 * Note that we do not msync(), since the caller is presumed to be
234 	 * about to modify the entry further
235 	 */
236 	memset(spare_slot, 0, sizeof(HostInfo));
237 	strncpy(spare_slot->hostname, hostname, SM_MAXSTRLEN);
238 	return (spare_slot);
239 }
240 
241 /* init_file -------------------------------------------------------------- */
242 /*
243  * Purpose:	Open file, create if necessary, initialise it.
244  * Returns:	Nothing - exits on error
245  * Notes:	Called before process becomes daemon, hence logs to
246  *		stderr rather than syslog.
247  *		Opens the file, then mmap()s it for ease of access.
248  *		Also performs initial clean-up of the file, zeroing
249  *		monitor list pointers, setting the notifyReqd flag in
250  *		all hosts that had a monitor list, and incrementing
251  *		the state number to the next even value.
252  */
253 void
254 init_file(filename)
255 	char *filename;
256 {
257 	char buf[HEADER_LEN];
258 	int new_file = FALSE;
259 	int i;
260 
261 	/* try to open existing file - if not present, create one */
262 	status_fd = open(filename, O_RDWR);
263 	if ((status_fd < 0) && (errno == ENOENT)) {
264 		status_fd = open(filename, O_RDWR | O_CREAT, 0644);
265 		new_file = TRUE;
266 	}
267 	if (status_fd < 0) {
268 		err(1, "unable to open status file %s", filename);
269 		/* NOTREACHED */
270 	}
271 
272 	/*
273 	 * File now open.  mmap() it, with a generous size to allow for
274 	 * later growth, where we will extend the file but not re-map it.
275 	 */
276 	status_info = (FileLayout *)mmap(NULL, 0x10000000,
277 	    PROT_READ | PROT_WRITE, MAP_SHARED, status_fd, 0);
278 
279 	if (status_info == (FileLayout *)(-1)) {
280 		perror("rpc.statd");
281 		fprintf(stderr, "Unable to mmap() status file\n");
282 	}
283 	status_file_len = lseek(status_fd, 0L, SEEK_END);
284 
285 	/*
286 	 * If the file was not newly created, validate the contents, and if
287 	 * defective, re-create from scratch.
288 	 */
289 	if (!new_file) {
290 		if ((status_file_len < HEADER_LEN) || (status_file_len <
291 		    (HEADER_LEN + sizeof(HostInfo) * status_info->noOfHosts))) {
292 			fprintf(stderr, "rpc.statd: status file is corrupt\n");
293 			new_file = TRUE;
294 		}
295 	}
296 	/* Initialisation of a new, empty file. */
297 	if (new_file) {
298 		memset(buf, 0, sizeof(buf));
299 		lseek(status_fd, 0L, SEEK_SET);
300 		write(status_fd, buf, HEADER_LEN);
301 		status_file_len = HEADER_LEN;
302 	} else {
303 		/*
304 		 * Clean-up of existing file - monitored hosts will have a
305 		 * pointer to a list of clients, which refers to memory in
306 		 * the previous incarnation of the program and so are
307 		 * meaningless now.  These pointers are zeroed and the fact
308 		 * that the host was previously monitored is recorded by
309 		 * setting the notifyReqd flag, which will in due course
310 		 * cause a SM_NOTIFY to be sent.
311 		 *
312 		 * Note that if we crash twice in quick succession, some hosts
313 		 * may already have notifyReqd set, where we didn't manage to
314 		 * notify them before the second crash occurred.
315 		 */
316 		for (i = 0; i < status_info->noOfHosts; i++) {
317 			HostInfo *this_host = &status_info->hosts[i];
318 
319 			if (this_host->monList) {
320 				this_host->notifyReqd = TRUE;
321 				this_host->monList = NULL;
322 			}
323 		}
324 		/* Select the next higher even number for the state counter */
325 		status_info->ourState =
326 		    (status_info->ourState + 2) & 0xfffffffe;
327 		status_info->ourState++;	/* XXX - ??? */
328 	}
329 }
330 
331 /* notify_one_host --------------------------------------------------------- */
332 /*
333  * Purpose:	Perform SM_NOTIFY procedure at specified host
334  * Returns:	TRUE if success, FALSE if failed.
335  */
336 static int
337 notify_one_host(hostname)
338 	char *hostname;
339 {
340 	struct timeval timeout = {20, 0};	/* 20 secs timeout */
341 	CLIENT *cli;
342 	char dummy;
343 	stat_chge arg;
344 	char our_hostname[SM_MAXSTRLEN + 1];
345 
346 	gethostname(our_hostname, sizeof(our_hostname));
347 	our_hostname[SM_MAXSTRLEN] = '\0';
348 	arg.mon_name = our_hostname;
349 	arg.state = status_info->ourState;
350 
351 	if (debug)
352 		syslog(LOG_DEBUG, "Sending SM_NOTIFY to host %s from %s",
353 		    hostname, our_hostname);
354 
355 	cli = clnt_create(hostname, SM_PROG, SM_VERS, "udp");
356 	if (!cli) {
357 		syslog(LOG_ERR, "Failed to contact host %s%s", hostname,
358 		    clnt_spcreateerror(""));
359 		return (FALSE);
360 	}
361 	if (clnt_call(cli, SM_NOTIFY, xdr_stat_chge, &arg, xdr_void,
362 	    &dummy, timeout) != RPC_SUCCESS) {
363 		syslog(LOG_ERR, "Failed to contact rpc.statd at host %s",
364 		    hostname);
365 		clnt_destroy(cli);
366 		return (FALSE);
367 	}
368 	clnt_destroy(cli);
369 	return (TRUE);
370 }
371 
372 /* notify_hosts ------------------------------------------------------------ */
373 /*
374  * Purpose:	Send SM_NOTIFY to all hosts marked as requiring it
375  * Returns:	Nothing, immediately - forks a process to do the work.
376  * Notes:	Does nothing if there are no monitored hosts.
377  *		Called after all the initialisation has been done -
378  *		logs to syslog.
379  */
380 void
381 notify_hosts(void)
382 {
383 	HostInfo *hp;
384 	int i, attempts;
385 	int work_to_do = FALSE;
386 	pid_t pid;
387 
388 	/* First check if there is in fact any work to do. */
389 	for (i = status_info->noOfHosts, hp = status_info->hosts; i;
390 	    i--, hp++) {
391 		if (hp->notifyReqd) {
392 			work_to_do = TRUE;
393 			break;
394 		}
395 	}
396 
397 	if (!work_to_do)
398 		return;		/* No work found */
399 
400 	pid = fork();
401 	if (pid == -1) {
402 		syslog(LOG_ERR, "Unable to fork notify process - %s",
403 		    strerror(errno));
404 		return;
405 	}
406 	if (pid)
407 		return;
408 
409 	/*
410 	 * Here in the child process.  We continue until all the hosts marked
411 	 * as requiring notification have been duly notified.
412 	 * If one of the initial attempts fails, we sleep for a while and
413 	 * have another go.  This is necessary because when we have crashed,
414 	 * (eg. a power outage) it is quite possible that we won't be able to
415 	 * contact all monitored hosts immediately on restart, either because
416 	 * they crashed too and take longer to come up (in which case the
417 	 * notification isn't really required), or more importantly if some
418 	 * router etc. needed to reach the monitored host has not come back
419 	 * up yet.  In this case, we will be a bit late in re-establishing
420 	 * locks (after the grace period) but that is the best we can do.
421 	 * We try 10 times at 5 sec intervals, 10 more times at 1 minute
422 	 * intervals, then 24 more times at hourly intervals, finally
423 	 * giving up altogether if the host hasn't come back to life after
424 	 * 24 hours.
425 	 */
426 	for (attempts = 0; attempts < 44; attempts++) {
427 		work_to_do = FALSE;	/* Unless anything fails */
428 		for (i = status_info->noOfHosts, hp = status_info->hosts; i > 0;
429 		    i--, hp++) {
430 			if (hp->notifyReqd) {
431 				if (notify_one_host(hp->hostname)) {
432 					hp->notifyReqd = FALSE;
433 					sync_file();
434 				} else
435 					work_to_do = TRUE;
436 			}
437 		}
438 		if (!work_to_do)
439 			break;
440 		if (attempts < 10)
441 			sleep(5);
442 		else
443 			if (attempts < 20)
444 				sleep(60);
445 			else
446 				sleep(60 * 60);
447 	}
448 	exit(0);
449 }
450