xref: /netbsd-src/usr.sbin/rpc.statd/statd.c (revision d0fed6c87ddc40a8bffa6f99e7433ddfc864dd83)
1 /*	$NetBSD: statd.c,v 1.1 1997/03/10 06:28:31 scottr 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 1.1 1997/03/10 06:28:31 scottr 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(struct svc_req * rqstp, SVCXPRT * transp);
67 static void handle_sigchld();
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, code, scp)
152 	int sig, code;
153 	struct sigcontext *scp;
154 {
155 	int     pid, status;
156 	pid = wait4(-1, &status, WNOHANG, (struct rusage *) 0);
157 	if (!pid)
158 		syslog(LOG_ERR, "Phantom SIGCHLD??");
159 	else if (status)
160 		syslog(LOG_ERR, "Child %d failed with status %d", pid,
161 		    WEXITSTATUS(status));
162 	else if (debug)
163 		syslog(LOG_DEBUG, "Child %d exited OK", pid);
164 }
165 
166 /* sync_file --------------------------------------------------------------- */
167 /*
168  * Purpose:	Packaged call of msync() to flush changes to mmap()ed file
169  * Returns:	Nothing.  Errors to syslog.
170  */
171 void
172 sync_file()
173 {
174 	if (msync((void *)status_info, 0) < 0)
175 		syslog(LOG_ERR, "msync() failed: %s", strerror(errno));
176 }
177 
178 /* find_host -------------------------------------------------------------- */
179 /*
180  * Purpose:	Find the entry in the status file for a given host
181  * Returns:	Pointer to that entry in the mmap() region, or NULL.
182  * Notes:	Also creates entries if requested.
183  *		Failure to create also returns NULL.
184  */
185 HostInfo *
186 find_host(hostname, create)
187 	char *hostname;
188 	int create;
189 {
190 	HostInfo *hp;
191 	HostInfo *spare_slot = NULL;
192 	HostInfo *result = NULL;
193 	int i;
194 
195 	for (i = 0, hp = status_info->hosts; i < status_info->noOfHosts;
196 	    i++, hp++) {
197 		if (!strncasecmp(hostname, hp->hostname, SM_MAXSTRLEN)) {
198 			result = hp;
199 			break;
200 		}
201 		if (!spare_slot && !hp->monList && !hp->notifyReqd)
202 			spare_slot = hp;
203 	}
204 
205 	/* Return if entry found, or if not asked to create one. */
206 	if (result || !create)
207 		return (result);
208 
209 	/*
210 	 * Now create an entry, using the spare slot if one was found or
211 	 * adding to the end of the list otherwise, extending file if req'd
212 	 */
213 	if (!spare_slot) {
214 		off_t desired_size;
215 		spare_slot = &status_info->hosts[status_info->noOfHosts];
216 		desired_size = ((char *)spare_slot - (char *)status_info) +
217 		    sizeof(HostInfo);
218 
219 		if (desired_size > status_file_len) {
220 			/* Extend file by writing 1 byte of junk at the
221 			 * desired end pos	 */
222 			lseek(status_fd, desired_size - 1, SEEK_SET);
223 			i = write(status_fd, &i, 1);
224 			if (i < 1) {
225 				syslog(LOG_ERR, "Unable to extend status file");
226 				return (NULL);
227 			}
228 			status_file_len = desired_size;
229 		}
230 		status_info->noOfHosts++;
231 	}
232 	/*
233 	 * Initialise the spare slot that has been found/created
234 	 * Note that we do not msync(), since the caller is presumed to be
235 	 * about to modify the entry further
236 	 */
237 	memset(spare_slot, 0, sizeof(HostInfo));
238 	strncpy(spare_slot->hostname, hostname, SM_MAXSTRLEN);
239 	return (spare_slot);
240 }
241 
242 /* init_file -------------------------------------------------------------- */
243 /*
244  * Purpose:	Open file, create if necessary, initialise it.
245  * Returns:	Nothing - exits on error
246  * Notes:	Called before process becomes daemon, hence logs to
247  *		stderr rather than syslog.
248  *		Opens the file, then mmap()s it for ease of access.
249  *		Also performs initial clean-up of the file, zeroing
250  *		monitor list pointers, setting the notifyReqd flag in
251  *		all hosts that had a monitor list, and incrementing
252  *		the state number to the next even value.
253  */
254 void
255 init_file(filename)
256 	char *filename;
257 {
258 	char buf[HEADER_LEN];
259 	int new_file = FALSE;
260 	int i;
261 
262 	/* try to open existing file - if not present, create one */
263 	status_fd = open(filename, O_RDWR);
264 	if ((status_fd < 0) && (errno == ENOENT)) {
265 		status_fd = open(filename, O_RDWR | O_CREAT, 0644);
266 		new_file = TRUE;
267 	}
268 	if (status_fd < 0) {
269 		err(1, "unable to open status file %s", filename);
270 		/* NOTREACHED */
271 	}
272 
273 	/*
274 	 * File now open.  mmap() it, with a generous size to allow for
275 	 * later growth, where we will extend the file but not re-map it.
276 	 */
277 	status_info = (FileLayout *)mmap(NULL, 0x10000000,
278 	    PROT_READ | PROT_WRITE, MAP_SHARED, status_fd, 0);
279 
280 	if (status_info == (FileLayout *)(-1)) {
281 		perror("rpc.statd");
282 		fprintf(stderr, "Unable to mmap() status file\n");
283 	}
284 	status_file_len = lseek(status_fd, 0L, SEEK_END);
285 
286 	/*
287 	 * If the file was not newly created, validate the contents, and if
288 	 * defective, re-create from scratch.
289 	 */
290 	if (!new_file) {
291 		if ((status_file_len < HEADER_LEN) || (status_file_len <
292 		    (HEADER_LEN + sizeof(HostInfo) * status_info->noOfHosts))) {
293 			fprintf(stderr, "rpc.statd: status file is corrupt\n");
294 			new_file = TRUE;
295 		}
296 	}
297 	/* Initialisation of a new, empty file. */
298 	if (new_file) {
299 		memset(buf, 0, sizeof(buf));
300 		lseek(status_fd, 0L, SEEK_SET);
301 		write(status_fd, buf, HEADER_LEN);
302 		status_file_len = HEADER_LEN;
303 	} else {
304 		/*
305 		 * Clean-up of existing file - monitored hosts will have a
306 		 * pointer to a list of clients, which refers to memory in
307 		 * the previous incarnation of the program and so are
308 		 * meaningless now.  These pointers are zeroed and the fact
309 		 * that the host was previously monitored is recorded by
310 		 * setting the notifyReqd flag, which will in due course
311 		 * cause a SM_NOTIFY to be sent.
312 		 *
313 		 * Note that if we crash twice in quick succession, some hosts
314 		 * may already have notifyReqd set, where we didn't manage to
315 		 * notify them before the second crash occurred.
316 		 */
317 		for (i = 0; i < status_info->noOfHosts; i++) {
318 			HostInfo *this_host = &status_info->hosts[i];
319 
320 			if (this_host->monList) {
321 				this_host->notifyReqd = TRUE;
322 				this_host->monList = NULL;
323 			}
324 		}
325 		/* Select the next higher even number for the state counter */
326 		status_info->ourState =
327 		    (status_info->ourState + 2) & 0xfffffffe;
328 		status_info->ourState++;	/* XXX - ??? */
329 	}
330 }
331 
332 /* notify_one_host --------------------------------------------------------- */
333 /*
334  * Purpose:	Perform SM_NOTIFY procedure at specified host
335  * Returns:	TRUE if success, FALSE if failed.
336  */
337 static int
338 notify_one_host(hostname)
339 	char *hostname;
340 {
341 	struct timeval timeout = {20, 0};	/* 20 secs timeout */
342 	CLIENT *cli;
343 	char dummy;
344 	stat_chge arg;
345 	char our_hostname[SM_MAXSTRLEN + 1];
346 
347 	gethostname(our_hostname, sizeof(our_hostname));
348 	our_hostname[SM_MAXSTRLEN] = '\0';
349 	arg.mon_name = our_hostname;
350 	arg.state = status_info->ourState;
351 
352 	if (debug)
353 		syslog(LOG_DEBUG, "Sending SM_NOTIFY to host %s from %s",
354 		    hostname, our_hostname);
355 
356 	cli = clnt_create(hostname, SM_PROG, SM_VERS, "udp");
357 	if (!cli) {
358 		syslog(LOG_ERR, "Failed to contact host %s%s", hostname,
359 		    clnt_spcreateerror(""));
360 		return (FALSE);
361 	}
362 	if (clnt_call(cli, SM_NOTIFY, xdr_stat_chge, &arg, xdr_void,
363 	    &dummy, timeout) != RPC_SUCCESS) {
364 		syslog(LOG_ERR, "Failed to contact rpc.statd at host %s",
365 		    hostname);
366 		clnt_destroy(cli);
367 		return (FALSE);
368 	}
369 	clnt_destroy(cli);
370 	return (TRUE);
371 }
372 
373 /* notify_hosts ------------------------------------------------------------ */
374 /*
375  * Purpose:	Send SM_NOTIFY to all hosts marked as requiring it
376  * Returns:	Nothing, immediately - forks a process to do the work.
377  * Notes:	Does nothing if there are no monitored hosts.
378  *		Called after all the initialisation has been done -
379  *		logs to syslog.
380  */
381 void
382 notify_hosts(void)
383 {
384 	HostInfo *hp;
385 	int i, attempts;
386 	int work_to_do = FALSE;
387 	pid_t pid;
388 
389 	/* First check if there is in fact any work to do. */
390 	for (i = status_info->noOfHosts, hp = status_info->hosts; i;
391 	    i--, hp++) {
392 		if (hp->notifyReqd) {
393 			work_to_do = TRUE;
394 			break;
395 		}
396 	}
397 
398 	if (!work_to_do)
399 		return;		/* No work found */
400 
401 	pid = fork();
402 	if (pid == -1) {
403 		syslog(LOG_ERR, "Unable to fork notify process - %s",
404 		    strerror(errno));
405 		return;
406 	}
407 	if (pid)
408 		return;
409 
410 	/*
411 	 * Here in the child process.  We continue until all the hosts marked
412 	 * as requiring notification have been duly notified.
413 	 * If one of the initial attempts fails, we sleep for a while and
414 	 * have another go.  This is necessary because when we have crashed,
415 	 * (eg. a power outage) it is quite possible that we won't be able to
416 	 * contact all monitored hosts immediately on restart, either because
417 	 * they crashed too and take longer to come up (in which case the
418 	 * notification isn't really required), or more importantly if some
419 	 * router etc. needed to reach the monitored host has not come back
420 	 * up yet.  In this case, we will be a bit late in re-establishing
421 	 * locks (after the grace period) but that is the best we can do.
422 	 * We try 10 times at 5 sec intervals, 10 more times at 1 minute
423 	 * intervals, then 24 more times at hourly intervals, finally
424 	 * giving up altogether if the host hasn't come back to life after
425 	 * 24 hours.
426 	 */
427 	for (attempts = 0; attempts < 44; attempts++) {
428 		work_to_do = FALSE;	/* Unless anything fails */
429 		for (i = status_info->noOfHosts, hp = status_info->hosts; i > 0;
430 		    i--, hp++) {
431 			if (hp->notifyReqd) {
432 				if (notify_one_host(hp->hostname)) {
433 					hp->notifyReqd = FALSE;
434 					sync_file();
435 				} else
436 					work_to_do = TRUE;
437 			}
438 		}
439 		if (!work_to_do)
440 			break;
441 		if (attempts < 10)
442 			sleep(5);
443 		else
444 			if (attempts < 20)
445 				sleep(60);
446 			else
447 				sleep(60 * 60);
448 	}
449 	exit(0);
450 }
451