xref: /freebsd/usr.sbin/nfsd/nfsd.c (revision 06c3fb27)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993, 1994
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its 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 THE REGENTS 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 REGENTS 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 #include <sys/param.h>
36 #include <sys/syslog.h>
37 #include <sys/wait.h>
38 #include <sys/mount.h>
39 #include <sys/fcntl.h>
40 #include <sys/linker.h>
41 #include <sys/module.h>
42 #include <sys/types.h>
43 #include <sys/stat.h>
44 #include <sys/sysctl.h>
45 #include <sys/ucred.h>
46 
47 #include <rpc/rpc.h>
48 #include <rpc/pmap_clnt.h>
49 #include <rpcsvc/nfs_prot.h>
50 
51 #include <netdb.h>
52 #include <arpa/inet.h>
53 #include <nfs/nfssvc.h>
54 
55 #include <fs/nfs/nfsproto.h>
56 #include <fs/nfs/nfskpiport.h>
57 #include <fs/nfs/nfs.h>
58 
59 #include <err.h>
60 #include <errno.h>
61 #include <signal.h>
62 #include <stdio.h>
63 #include <stdlib.h>
64 #include <string.h>
65 #include <unistd.h>
66 #include <sysexits.h>
67 
68 #include <getopt.h>
69 
70 static int	debug = 0;
71 
72 #define	NFSD_STABLERESTART	"/var/db/nfs-stablerestart"
73 #define	NFSD_STABLEBACKUP	"/var/db/nfs-stablerestart.bak"
74 #define	MAXNFSDCNT	256
75 #define	DEFNFSDCNT	 4
76 #define	NFS_VER2	 2
77 #define NFS_VER3	 3
78 #define NFS_VER4	 4
79 static pid_t children[MAXNFSDCNT]; /* PIDs of children */
80 static pid_t masterpid;		   /* PID of master/parent */
81 static int nfsdcnt;		/* number of children */
82 static int nfsdcnt_set;
83 static int minthreads;
84 static int maxthreads;
85 static int nfssvc_nfsd;		/* Set to correct NFSSVC_xxx flag */
86 static int stablefd = -1;	/* Fd for the stable restart file */
87 static int backupfd;		/* Fd for the backup stable restart file */
88 static const char *getopt_shortopts;
89 static const char *getopt_usage;
90 static int nfs_minvers = NFS_VER2;
91 
92 static int minthreads_set;
93 static int maxthreads_set;
94 
95 static struct option longopts[] = {
96 	{ "debug", no_argument, &debug, 1 },
97 	{ "minthreads", required_argument, &minthreads_set, 1 },
98 	{ "maxthreads", required_argument, &maxthreads_set, 1 },
99 	{ "pnfs", required_argument, NULL, 'p' },
100 	{ "mirror", required_argument, NULL, 'm' },
101 	{ NULL, 0, NULL, 0}
102 };
103 
104 static void	cleanup(int);
105 static void	child_cleanup(int);
106 static void	killchildren(void);
107 static void	nfsd_exit(int);
108 static void	nonfs(int);
109 static void	reapchild(int);
110 static int	setbindhost(struct addrinfo **ia, const char *bindhost,
111 		    struct addrinfo hints);
112 static void	start_server(int, struct nfsd_nfsd_args *, const char *vhost);
113 static void	unregistration(void);
114 static void	usage(void);
115 static void	open_stable(int *, int *);
116 static void	copy_stable(int, int);
117 static void	backup_stable(int);
118 static void	set_nfsdcnt(int);
119 static void	parse_dsserver(const char *, struct nfsd_nfsd_args *);
120 
121 /*
122  * Nfs server daemon mostly just a user context for nfssvc()
123  *
124  * 1 - do file descriptor and signal cleanup
125  * 2 - fork the nfsd(s)
126  * 3 - create server socket(s)
127  * 4 - register socket with rpcbind
128  *
129  * For connectionless protocols, just pass the socket into the kernel via.
130  * nfssvc().
131  * For connection based sockets, loop doing accepts. When you get a new
132  * socket from accept, pass the msgsock into the kernel via. nfssvc().
133  * The arguments are:
134  *	-r - reregister with rpcbind
135  *	-d - unregister with rpcbind
136  *	-t - support tcp nfs clients
137  *	-u - support udp nfs clients
138  *	-e - forces it to run a server that supports nfsv4
139  *	-p - enable a pNFS service
140  *	-m - set the mirroring level for a pNFS service
141  * followed by "n" which is the number of nfsds' to fork off
142  */
143 int
144 main(int argc, char **argv)
145 {
146 	struct nfsd_addsock_args addsockargs;
147 	struct addrinfo *ai_udp, *ai_tcp, *ai_udp6, *ai_tcp6, hints;
148 	struct netconfig *nconf_udp, *nconf_tcp, *nconf_udp6, *nconf_tcp6;
149 	struct netbuf nb_udp, nb_tcp, nb_udp6, nb_tcp6;
150 	struct sockaddr_storage peer;
151 	fd_set ready, sockbits;
152 	int ch, connect_type_cnt, i, maxsock, msgsock;
153 	socklen_t len;
154 	int on = 1, unregister, reregister, sock;
155 	int tcp6sock, ip6flag, tcpflag, tcpsock;
156 	int udpflag, ecode, error, s;
157 	int bindhostc, bindanyflag, rpcbreg, rpcbregcnt;
158 	int nfssvc_addsock;
159 	int jailed, longindex = 0;
160 	size_t jailed_size, nfs_minvers_size;
161 	const char *lopt;
162 	char **bindhost = NULL;
163 	pid_t pid;
164 	struct nfsd_nfsd_args nfsdargs;
165 	const char *vhostname = NULL;
166 
167 	nfsdargs.mirrorcnt = 1;
168 	nfsdargs.addr = NULL;
169 	nfsdargs.addrlen = 0;
170 	nfsdcnt = DEFNFSDCNT;
171 	unregister = reregister = tcpflag = maxsock = 0;
172 	bindanyflag = udpflag = connect_type_cnt = bindhostc = 0;
173 	getopt_shortopts = "ah:n:rdtuep:m:V:";
174 	getopt_usage =
175 	    "usage:\n"
176 	    "  nfsd [-ardtue] [-h bindip]\n"
177 	    "       [-n numservers] [--minthreads #] [--maxthreads #]\n"
178 	    "       [-p/--pnfs dsserver0:/dsserver0-mounted-on-dir,...,"
179 	    "dsserverN:/dsserverN-mounted-on-dir] [-m mirrorlevel]\n"
180 	    "       [-V virtual_hostname]\n";
181 	while ((ch = getopt_long(argc, argv, getopt_shortopts, longopts,
182 		    &longindex)) != -1)
183 		switch (ch) {
184 		case 'V':
185 			if (strlen(optarg) <= MAXHOSTNAMELEN)
186 				vhostname = optarg;
187 			else
188 				warnx("Virtual host name (%s) is too long",
189 				    optarg);
190 			break;
191 		case 'a':
192 			bindanyflag = 1;
193 			break;
194 		case 'n':
195 			set_nfsdcnt(atoi(optarg));
196 			break;
197 		case 'h':
198 			bindhostc++;
199 			bindhost = realloc(bindhost,sizeof(char *)*bindhostc);
200 			if (bindhost == NULL)
201 				errx(1, "Out of memory");
202 			bindhost[bindhostc-1] = strdup(optarg);
203 			if (bindhost[bindhostc-1] == NULL)
204 				errx(1, "Out of memory");
205 			break;
206 		case 'r':
207 			reregister = 1;
208 			break;
209 		case 'd':
210 			unregister = 1;
211 			break;
212 		case 't':
213 			tcpflag = 1;
214 			break;
215 		case 'u':
216 			udpflag = 1;
217 			break;
218 		case 'e':
219 			/* now a no-op, since this is the default */
220 			break;
221 		case 'p':
222 			/* Parse out the DS server host names and mount pts. */
223 			parse_dsserver(optarg, &nfsdargs);
224 			break;
225 		case 'm':
226 			/* Set the mirror level for a pNFS service. */
227 			i = atoi(optarg);
228 			if (i < 2 || i > NFSDEV_MAXMIRRORS)
229 				errx(1, "Mirror level out of range 2<-->%d",
230 				    NFSDEV_MAXMIRRORS);
231 			nfsdargs.mirrorcnt = i;
232 			break;
233 		case 0:
234 			lopt = longopts[longindex].name;
235 			if (!strcmp(lopt, "minthreads")) {
236 				minthreads = atoi(optarg);
237 			} else if (!strcmp(lopt, "maxthreads")) {
238 				maxthreads = atoi(optarg);
239 			}
240 			break;
241 		default:
242 		case '?':
243 			usage();
244 		}
245 	if (!tcpflag && !udpflag)
246 		udpflag = 1;
247 	argv += optind;
248 	argc -= optind;
249 	if (minthreads_set && maxthreads_set && minthreads > maxthreads)
250 		errx(EX_USAGE,
251 		    "error: minthreads(%d) can't be greater than "
252 		    "maxthreads(%d)", minthreads, maxthreads);
253 
254 	/*
255 	 * XXX
256 	 * Backward compatibility, trailing number is the count of daemons.
257 	 */
258 	if (argc > 1)
259 		usage();
260 	if (argc == 1)
261 		set_nfsdcnt(atoi(argv[0]));
262 
263 	/*
264 	 * Unless the "-o" option was specified, try and run "nfsd".
265 	 * If "-o" was specified, try and run "nfsserver".
266 	 */
267 	if (modfind("nfsd") < 0) {
268 		/* Not present in kernel, try loading it */
269 		if (kldload("nfsd") < 0 || modfind("nfsd") < 0)
270 			errx(1, "NFS server is not available");
271 	}
272 
273 	ip6flag = 1;
274 	s = socket(AF_INET6, SOCK_DGRAM, IPPROTO_UDP);
275 	if (s == -1) {
276 		if (errno != EPROTONOSUPPORT && errno != EAFNOSUPPORT)
277 			err(1, "socket");
278 		ip6flag = 0;
279 	} else if (getnetconfigent("udp6") == NULL ||
280 		getnetconfigent("tcp6") == NULL) {
281 		ip6flag = 0;
282 	}
283 	if (s != -1)
284 		close(s);
285 
286 	if (bindhostc == 0 || bindanyflag) {
287 		bindhostc++;
288 		bindhost = realloc(bindhost,sizeof(char *)*bindhostc);
289 		if (bindhost == NULL)
290 			errx(1, "Out of memory");
291 		bindhost[bindhostc-1] = strdup("*");
292 		if (bindhost[bindhostc-1] == NULL)
293 			errx(1, "Out of memory");
294 	}
295 
296 	if (unregister) {
297 		/*
298 		 * Unregister before setting nfs_minvers, in case the
299 		 * value of vfs.nfsd.server_min_nfsvers has changed
300 		 * since registering with rpcbind.
301 		 */
302 		unregistration();
303 		exit (0);
304 	}
305 
306 	nfs_minvers_size = sizeof(nfs_minvers);
307 	error = sysctlbyname("vfs.nfsd.server_min_nfsvers", &nfs_minvers,
308 	    &nfs_minvers_size, NULL, 0);
309 	if (error != 0 || nfs_minvers < NFS_VER2 || nfs_minvers > NFS_VER4) {
310 		warnx("sysctlbyname(vfs.nfsd.server_min_nfsvers) failed,"
311 		    " defaulting to NFSv2");
312 		nfs_minvers = NFS_VER2;
313 	}
314 
315 	if (reregister) {
316 		if (udpflag) {
317 			memset(&hints, 0, sizeof hints);
318 			hints.ai_flags = AI_PASSIVE;
319 			hints.ai_family = AF_INET;
320 			hints.ai_socktype = SOCK_DGRAM;
321 			hints.ai_protocol = IPPROTO_UDP;
322 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp);
323 			if (ecode != 0)
324 				err(1, "getaddrinfo udp: %s", gai_strerror(ecode));
325 			nconf_udp = getnetconfigent("udp");
326 			if (nconf_udp == NULL)
327 				err(1, "getnetconfigent udp failed");
328 			nb_udp.buf = ai_udp->ai_addr;
329 			nb_udp.len = nb_udp.maxlen = ai_udp->ai_addrlen;
330 			if (nfs_minvers == NFS_VER2)
331 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp,
332 				    &nb_udp))
333 					err(1, "rpcb_set udp failed");
334 			if (nfs_minvers <= NFS_VER3)
335 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp,
336 				    &nb_udp))
337 					err(1, "rpcb_set udp failed");
338 			freeaddrinfo(ai_udp);
339 		}
340 		if (udpflag && ip6flag) {
341 			memset(&hints, 0, sizeof hints);
342 			hints.ai_flags = AI_PASSIVE;
343 			hints.ai_family = AF_INET6;
344 			hints.ai_socktype = SOCK_DGRAM;
345 			hints.ai_protocol = IPPROTO_UDP;
346 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp6);
347 			if (ecode != 0)
348 				err(1, "getaddrinfo udp6: %s", gai_strerror(ecode));
349 			nconf_udp6 = getnetconfigent("udp6");
350 			if (nconf_udp6 == NULL)
351 				err(1, "getnetconfigent udp6 failed");
352 			nb_udp6.buf = ai_udp6->ai_addr;
353 			nb_udp6.len = nb_udp6.maxlen = ai_udp6->ai_addrlen;
354 			if (nfs_minvers == NFS_VER2)
355 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp6,
356 				    &nb_udp6))
357 					err(1, "rpcb_set udp6 failed");
358 			if (nfs_minvers <= NFS_VER3)
359 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp6,
360 				    &nb_udp6))
361 					err(1, "rpcb_set udp6 failed");
362 			freeaddrinfo(ai_udp6);
363 		}
364 		if (tcpflag) {
365 			memset(&hints, 0, sizeof hints);
366 			hints.ai_flags = AI_PASSIVE;
367 			hints.ai_family = AF_INET;
368 			hints.ai_socktype = SOCK_STREAM;
369 			hints.ai_protocol = IPPROTO_TCP;
370 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_tcp);
371 			if (ecode != 0)
372 				err(1, "getaddrinfo tcp: %s", gai_strerror(ecode));
373 			nconf_tcp = getnetconfigent("tcp");
374 			if (nconf_tcp == NULL)
375 				err(1, "getnetconfigent tcp failed");
376 			nb_tcp.buf = ai_tcp->ai_addr;
377 			nb_tcp.len = nb_tcp.maxlen = ai_tcp->ai_addrlen;
378 			if (nfs_minvers == NFS_VER2)
379 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp,
380 				    &nb_tcp))
381 					err(1, "rpcb_set tcp failed");
382 			if (nfs_minvers <= NFS_VER3)
383 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp,
384 				    &nb_tcp))
385 					err(1, "rpcb_set tcp failed");
386 			freeaddrinfo(ai_tcp);
387 		}
388 		if (tcpflag && ip6flag) {
389 			memset(&hints, 0, sizeof hints);
390 			hints.ai_flags = AI_PASSIVE;
391 			hints.ai_family = AF_INET6;
392 			hints.ai_socktype = SOCK_STREAM;
393 			hints.ai_protocol = IPPROTO_TCP;
394 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_tcp6);
395 			if (ecode != 0)
396 				err(1, "getaddrinfo tcp6: %s", gai_strerror(ecode));
397 			nconf_tcp6 = getnetconfigent("tcp6");
398 			if (nconf_tcp6 == NULL)
399 				err(1, "getnetconfigent tcp6 failed");
400 			nb_tcp6.buf = ai_tcp6->ai_addr;
401 			nb_tcp6.len = nb_tcp6.maxlen = ai_tcp6->ai_addrlen;
402 			if (nfs_minvers == NFS_VER2)
403 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp6,
404 				    &nb_tcp6))
405 					err(1, "rpcb_set tcp6 failed");
406 			if (nfs_minvers <= NFS_VER3)
407 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp6,
408 				   &nb_tcp6))
409 					err(1, "rpcb_set tcp6 failed");
410 			freeaddrinfo(ai_tcp6);
411 		}
412 		exit (0);
413 	}
414 	if (debug == 0) {
415 		daemon(0, 0);
416 		(void)signal(SIGHUP, SIG_IGN);
417 		(void)signal(SIGINT, SIG_IGN);
418 		/*
419 		 * nfsd sits in the kernel most of the time.  It needs
420 		 * to ignore SIGTERM/SIGQUIT in order to stay alive as long
421 		 * as possible during a shutdown, otherwise loopback
422 		 * mounts will not be able to unmount.
423 		 */
424 		(void)signal(SIGTERM, SIG_IGN);
425 		(void)signal(SIGQUIT, SIG_IGN);
426 	}
427 	(void)signal(SIGSYS, nonfs);
428 	(void)signal(SIGCHLD, reapchild);
429 	(void)signal(SIGUSR2, backup_stable);
430 
431 	openlog("nfsd", LOG_PID | (debug ? LOG_PERROR : 0), LOG_DAEMON);
432 
433 	/*
434 	 * For V4, we open the stablerestart file and call nfssvc()
435 	 * to get it loaded. This is done before the daemons do the
436 	 * regular nfssvc() call to service NFS requests.
437 	 * (This way the file remains open until the last nfsd is killed
438 	 *  off.)
439 	 * It and the backup copy will be created as empty files
440 	 * the first time this nfsd is started and should never be
441 	 * deleted/replaced if at all possible. It should live on a
442 	 * local, non-volatile storage device that does not do hardware
443 	 * level write-back caching. (See SCSI doc for more information
444 	 * on how to prevent write-back caching on SCSI disks.)
445 	 */
446 	open_stable(&stablefd, &backupfd);
447 	if (stablefd < 0) {
448 		syslog(LOG_ERR, "Can't open %s: %m\n", NFSD_STABLERESTART);
449 		exit(1);
450 	}
451 	/* This system call will fail for old kernels, but that's ok. */
452 	nfssvc(NFSSVC_BACKUPSTABLE, NULL);
453 	if (nfssvc(NFSSVC_STABLERESTART, (caddr_t)&stablefd) < 0) {
454 		if (errno == EPERM) {
455 			jailed = 0;
456 			jailed_size = sizeof(jailed);
457 			sysctlbyname("security.jail.jailed", &jailed,
458 			    &jailed_size, NULL, 0);
459 			if (jailed != 0)
460 				syslog(LOG_ERR, "nfssvc stablerestart failed: "
461 				    "allow.nfsd might not be configured");
462 			else
463 				syslog(LOG_ERR, "nfssvc stablerestart failed");
464 		} else if (errno == ENXIO)
465 			syslog(LOG_ERR, "nfssvc stablerestart failed: is nfsd "
466 			    "already running?");
467 		else
468 			syslog(LOG_ERR, "Can't read stable storage file: %m\n");
469 		exit(1);
470 	}
471 	nfssvc_addsock = NFSSVC_NFSDADDSOCK;
472 	nfssvc_nfsd = NFSSVC_NFSDNFSD | NFSSVC_NEWSTRUCT;
473 
474 	if (tcpflag) {
475 		/*
476 		 * For TCP mode, we fork once to start the first
477 		 * kernel nfsd thread. The kernel will add more
478 		 * threads as needed.
479 		 */
480 		masterpid = getpid();
481 		pid = fork();
482 		if (pid == -1) {
483 			syslog(LOG_ERR, "fork: %m");
484 			nfsd_exit(1);
485 		}
486 		if (pid) {
487 			children[0] = pid;
488 		} else {
489 			(void)signal(SIGUSR1, child_cleanup);
490 			setproctitle("server");
491 			start_server(0, &nfsdargs, vhostname);
492 		}
493 	}
494 
495 	(void)signal(SIGUSR1, cleanup);
496 	FD_ZERO(&sockbits);
497 
498 	rpcbregcnt = 0;
499 	/* Set up the socket for udp and rpcb register it. */
500 	if (udpflag) {
501 		rpcbreg = 0;
502 		for (i = 0; i < bindhostc; i++) {
503 			memset(&hints, 0, sizeof hints);
504 			hints.ai_flags = AI_PASSIVE;
505 			hints.ai_family = AF_INET;
506 			hints.ai_socktype = SOCK_DGRAM;
507 			hints.ai_protocol = IPPROTO_UDP;
508 			if (setbindhost(&ai_udp, bindhost[i], hints) == 0) {
509 				rpcbreg = 1;
510 				rpcbregcnt++;
511 				if ((sock = socket(ai_udp->ai_family,
512 				    ai_udp->ai_socktype,
513 				    ai_udp->ai_protocol)) < 0) {
514 					syslog(LOG_ERR,
515 					    "can't create udp socket");
516 					nfsd_exit(1);
517 				}
518 				if (bind(sock, ai_udp->ai_addr,
519 				    ai_udp->ai_addrlen) < 0) {
520 					syslog(LOG_ERR,
521 					    "can't bind udp addr %s: %m",
522 					    bindhost[i]);
523 					nfsd_exit(1);
524 				}
525 				freeaddrinfo(ai_udp);
526 				addsockargs.sock = sock;
527 				addsockargs.name = NULL;
528 				addsockargs.namelen = 0;
529 				if (nfssvc(nfssvc_addsock, &addsockargs) < 0) {
530 					syslog(LOG_ERR, "can't Add UDP socket");
531 					nfsd_exit(1);
532 				}
533 				(void)close(sock);
534 			}
535 		}
536 		if (rpcbreg == 1) {
537 			memset(&hints, 0, sizeof hints);
538 			hints.ai_flags = AI_PASSIVE;
539 			hints.ai_family = AF_INET;
540 			hints.ai_socktype = SOCK_DGRAM;
541 			hints.ai_protocol = IPPROTO_UDP;
542 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp);
543 			if (ecode != 0) {
544 				syslog(LOG_ERR, "getaddrinfo udp: %s",
545 				   gai_strerror(ecode));
546 				nfsd_exit(1);
547 			}
548 			nconf_udp = getnetconfigent("udp");
549 			if (nconf_udp == NULL)
550 				err(1, "getnetconfigent udp failed");
551 			nb_udp.buf = ai_udp->ai_addr;
552 			nb_udp.len = nb_udp.maxlen = ai_udp->ai_addrlen;
553 			if (nfs_minvers == NFS_VER2)
554 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp,
555 				    &nb_udp))
556 					err(1, "rpcb_set udp failed");
557 			if (nfs_minvers <= NFS_VER3)
558 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp,
559 				    &nb_udp))
560 					err(1, "rpcb_set udp failed");
561 			freeaddrinfo(ai_udp);
562 		}
563 	}
564 
565 	/* Set up the socket for udp6 and rpcb register it. */
566 	if (udpflag && ip6flag) {
567 		rpcbreg = 0;
568 		for (i = 0; i < bindhostc; i++) {
569 			memset(&hints, 0, sizeof hints);
570 			hints.ai_flags = AI_PASSIVE;
571 			hints.ai_family = AF_INET6;
572 			hints.ai_socktype = SOCK_DGRAM;
573 			hints.ai_protocol = IPPROTO_UDP;
574 			if (setbindhost(&ai_udp6, bindhost[i], hints) == 0) {
575 				rpcbreg = 1;
576 				rpcbregcnt++;
577 				if ((sock = socket(ai_udp6->ai_family,
578 				    ai_udp6->ai_socktype,
579 				    ai_udp6->ai_protocol)) < 0) {
580 					syslog(LOG_ERR,
581 						"can't create udp6 socket");
582 					nfsd_exit(1);
583 				}
584 				if (setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY,
585 				    &on, sizeof on) < 0) {
586 					syslog(LOG_ERR,
587 					    "can't set v6-only binding for "
588 					    "udp6 socket: %m");
589 					nfsd_exit(1);
590 				}
591 				if (bind(sock, ai_udp6->ai_addr,
592 				    ai_udp6->ai_addrlen) < 0) {
593 					syslog(LOG_ERR,
594 					    "can't bind udp6 addr %s: %m",
595 					    bindhost[i]);
596 					nfsd_exit(1);
597 				}
598 				freeaddrinfo(ai_udp6);
599 				addsockargs.sock = sock;
600 				addsockargs.name = NULL;
601 				addsockargs.namelen = 0;
602 				if (nfssvc(nfssvc_addsock, &addsockargs) < 0) {
603 					syslog(LOG_ERR,
604 					    "can't add UDP6 socket");
605 					nfsd_exit(1);
606 				}
607 				(void)close(sock);
608 			}
609 		}
610 		if (rpcbreg == 1) {
611 			memset(&hints, 0, sizeof hints);
612 			hints.ai_flags = AI_PASSIVE;
613 			hints.ai_family = AF_INET6;
614 			hints.ai_socktype = SOCK_DGRAM;
615 			hints.ai_protocol = IPPROTO_UDP;
616 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp6);
617 			if (ecode != 0) {
618 				syslog(LOG_ERR, "getaddrinfo udp6: %s",
619 				   gai_strerror(ecode));
620 				nfsd_exit(1);
621 			}
622 			nconf_udp6 = getnetconfigent("udp6");
623 			if (nconf_udp6 == NULL)
624 				err(1, "getnetconfigent udp6 failed");
625 			nb_udp6.buf = ai_udp6->ai_addr;
626 			nb_udp6.len = nb_udp6.maxlen = ai_udp6->ai_addrlen;
627 			if (nfs_minvers == NFS_VER2)
628 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp6,
629 				    &nb_udp6))
630 					err(1,
631 					    "rpcb_set udp6 failed");
632 			if (nfs_minvers <= NFS_VER3)
633 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp6,
634 				    &nb_udp6))
635 					err(1,
636 					    "rpcb_set udp6 failed");
637 			freeaddrinfo(ai_udp6);
638 		}
639 	}
640 
641 	/* Set up the socket for tcp and rpcb register it. */
642 	if (tcpflag) {
643 		rpcbreg = 0;
644 		for (i = 0; i < bindhostc; i++) {
645 			memset(&hints, 0, sizeof hints);
646 			hints.ai_flags = AI_PASSIVE;
647 			hints.ai_family = AF_INET;
648 			hints.ai_socktype = SOCK_STREAM;
649 			hints.ai_protocol = IPPROTO_TCP;
650 			if (setbindhost(&ai_tcp, bindhost[i], hints) == 0) {
651 				rpcbreg = 1;
652 				rpcbregcnt++;
653 				if ((tcpsock = socket(AF_INET, SOCK_STREAM,
654 				    0)) < 0) {
655 					syslog(LOG_ERR,
656 					    "can't create tcp socket");
657 					nfsd_exit(1);
658 				}
659 				if (setsockopt(tcpsock, SOL_SOCKET,
660 				    SO_REUSEADDR,
661 				    (char *)&on, sizeof(on)) < 0)
662 					syslog(LOG_ERR,
663 					     "setsockopt SO_REUSEADDR: %m");
664 				if (bind(tcpsock, ai_tcp->ai_addr,
665 				    ai_tcp->ai_addrlen) < 0) {
666 					syslog(LOG_ERR,
667 					    "can't bind tcp addr %s: %m",
668 					    bindhost[i]);
669 					nfsd_exit(1);
670 				}
671 				if (listen(tcpsock, -1) < 0) {
672 					syslog(LOG_ERR, "listen failed");
673 					nfsd_exit(1);
674 				}
675 				freeaddrinfo(ai_tcp);
676 				FD_SET(tcpsock, &sockbits);
677 				maxsock = tcpsock;
678 				connect_type_cnt++;
679 			}
680 		}
681 		if (rpcbreg == 1) {
682 			memset(&hints, 0, sizeof hints);
683 			hints.ai_flags = AI_PASSIVE;
684 			hints.ai_family = AF_INET;
685 			hints.ai_socktype = SOCK_STREAM;
686 			hints.ai_protocol = IPPROTO_TCP;
687 			ecode = getaddrinfo(NULL, "nfs", &hints,
688 			     &ai_tcp);
689 			if (ecode != 0) {
690 				syslog(LOG_ERR, "getaddrinfo tcp: %s",
691 				   gai_strerror(ecode));
692 				nfsd_exit(1);
693 			}
694 			nconf_tcp = getnetconfigent("tcp");
695 			if (nconf_tcp == NULL)
696 				err(1, "getnetconfigent tcp failed");
697 			nb_tcp.buf = ai_tcp->ai_addr;
698 			nb_tcp.len = nb_tcp.maxlen = ai_tcp->ai_addrlen;
699 			if (nfs_minvers == NFS_VER2)
700 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp,
701 				    &nb_tcp))
702 					err(1, "rpcb_set tcp failed");
703 			if (nfs_minvers <= NFS_VER3)
704 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp,
705 				    &nb_tcp))
706 					err(1, "rpcb_set tcp failed");
707 			freeaddrinfo(ai_tcp);
708 		}
709 	}
710 
711 	/* Set up the socket for tcp6 and rpcb register it. */
712 	if (tcpflag && ip6flag) {
713 		rpcbreg = 0;
714 		for (i = 0; i < bindhostc; i++) {
715 			memset(&hints, 0, sizeof hints);
716 			hints.ai_flags = AI_PASSIVE;
717 			hints.ai_family = AF_INET6;
718 			hints.ai_socktype = SOCK_STREAM;
719 			hints.ai_protocol = IPPROTO_TCP;
720 			if (setbindhost(&ai_tcp6, bindhost[i], hints) == 0) {
721 				rpcbreg = 1;
722 				rpcbregcnt++;
723 				if ((tcp6sock = socket(ai_tcp6->ai_family,
724 				    ai_tcp6->ai_socktype,
725 				    ai_tcp6->ai_protocol)) < 0) {
726 					syslog(LOG_ERR,
727 					    "can't create tcp6 socket");
728 					nfsd_exit(1);
729 				}
730 				if (setsockopt(tcp6sock, SOL_SOCKET,
731 				    SO_REUSEADDR,
732 				    (char *)&on, sizeof(on)) < 0)
733 					syslog(LOG_ERR,
734 					    "setsockopt SO_REUSEADDR: %m");
735 				if (setsockopt(tcp6sock, IPPROTO_IPV6,
736 				    IPV6_V6ONLY, &on, sizeof on) < 0) {
737 					syslog(LOG_ERR,
738 					"can't set v6-only binding for tcp6 "
739 					    "socket: %m");
740 					nfsd_exit(1);
741 				}
742 				if (bind(tcp6sock, ai_tcp6->ai_addr,
743 				    ai_tcp6->ai_addrlen) < 0) {
744 					syslog(LOG_ERR,
745 					    "can't bind tcp6 addr %s: %m",
746 					    bindhost[i]);
747 					nfsd_exit(1);
748 				}
749 				if (listen(tcp6sock, -1) < 0) {
750 					syslog(LOG_ERR, "listen failed");
751 					nfsd_exit(1);
752 				}
753 				freeaddrinfo(ai_tcp6);
754 				FD_SET(tcp6sock, &sockbits);
755 				if (maxsock < tcp6sock)
756 					maxsock = tcp6sock;
757 				connect_type_cnt++;
758 			}
759 		}
760 		if (rpcbreg == 1) {
761 			memset(&hints, 0, sizeof hints);
762 			hints.ai_flags = AI_PASSIVE;
763 			hints.ai_family = AF_INET6;
764 			hints.ai_socktype = SOCK_STREAM;
765 			hints.ai_protocol = IPPROTO_TCP;
766 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_tcp6);
767 			if (ecode != 0) {
768 				syslog(LOG_ERR, "getaddrinfo tcp6: %s",
769 				   gai_strerror(ecode));
770 				nfsd_exit(1);
771 			}
772 			nconf_tcp6 = getnetconfigent("tcp6");
773 			if (nconf_tcp6 == NULL)
774 				err(1, "getnetconfigent tcp6 failed");
775 			nb_tcp6.buf = ai_tcp6->ai_addr;
776 			nb_tcp6.len = nb_tcp6.maxlen = ai_tcp6->ai_addrlen;
777 			if (nfs_minvers == NFS_VER2)
778 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp6,
779 				    &nb_tcp6))
780 					err(1, "rpcb_set tcp6 failed");
781 			if (nfs_minvers <= NFS_VER3)
782 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp6,
783 				    &nb_tcp6))
784 					err(1, "rpcb_set tcp6 failed");
785 			freeaddrinfo(ai_tcp6);
786 		}
787 	}
788 
789 	if (rpcbregcnt == 0) {
790 		syslog(LOG_ERR, "rpcb_set() failed, nothing to do: %m");
791 		nfsd_exit(1);
792 	}
793 
794 	if (tcpflag && connect_type_cnt == 0) {
795 		syslog(LOG_ERR, "tcp connects == 0, nothing to do: %m");
796 		nfsd_exit(1);
797 	}
798 
799 	setproctitle("master");
800 	/*
801 	 * We always want a master to have a clean way to shut nfsd down
802 	 * (with unregistration): if the master is killed, it unregisters and
803 	 * kills all children. If we run for UDP only (and so do not have to
804 	 * loop waiting for accept), we instead make the parent
805 	 * a "server" too. start_server will not return.
806 	 */
807 	if (!tcpflag)
808 		start_server(1, &nfsdargs, vhostname);
809 
810 	/*
811 	 * Loop forever accepting connections and passing the sockets
812 	 * into the kernel for the mounts.
813 	 */
814 	for (;;) {
815 		ready = sockbits;
816 		if (connect_type_cnt > 1) {
817 			if (select(maxsock + 1,
818 			    &ready, NULL, NULL, NULL) < 1) {
819 				error = errno;
820 				if (error == EINTR)
821 					continue;
822 				syslog(LOG_ERR, "select failed: %m");
823 				nfsd_exit(1);
824 			}
825 		}
826 		for (tcpsock = 0; tcpsock <= maxsock; tcpsock++) {
827 			if (FD_ISSET(tcpsock, &ready)) {
828 				len = sizeof(peer);
829 				if ((msgsock = accept(tcpsock,
830 				    (struct sockaddr *)&peer, &len)) < 0) {
831 					error = errno;
832 					syslog(LOG_ERR, "accept failed: %m");
833 					if (error == ECONNABORTED ||
834 					    error == EINTR)
835 						continue;
836 					nfsd_exit(1);
837 				}
838 				if (setsockopt(msgsock, SOL_SOCKET,
839 				    SO_KEEPALIVE, (char *)&on, sizeof(on)) < 0)
840 					syslog(LOG_ERR,
841 					    "setsockopt SO_KEEPALIVE: %m");
842 				addsockargs.sock = msgsock;
843 				addsockargs.name = (caddr_t)&peer;
844 				addsockargs.namelen = len;
845 				nfssvc(nfssvc_addsock, &addsockargs);
846 				(void)close(msgsock);
847 			}
848 		}
849 	}
850 }
851 
852 static int
853 setbindhost(struct addrinfo **ai, const char *bindhost, struct addrinfo hints)
854 {
855 	int ecode;
856 	u_int32_t host_addr[4];  /* IPv4 or IPv6 */
857 	const char *hostptr;
858 
859 	if (bindhost == NULL || strcmp("*", bindhost) == 0)
860 		hostptr = NULL;
861 	else
862 		hostptr = bindhost;
863 
864 	if (hostptr != NULL) {
865 		switch (hints.ai_family) {
866 		case AF_INET:
867 			if (inet_pton(AF_INET, hostptr, host_addr) == 1) {
868 				hints.ai_flags = AI_NUMERICHOST;
869 			} else {
870 				if (inet_pton(AF_INET6, hostptr,
871 				    host_addr) == 1)
872 					return (1);
873 			}
874 			break;
875 		case AF_INET6:
876 			if (inet_pton(AF_INET6, hostptr, host_addr) == 1) {
877 				hints.ai_flags = AI_NUMERICHOST;
878 			} else {
879 				if (inet_pton(AF_INET, hostptr,
880 				    host_addr) == 1)
881 					return (1);
882 			}
883 			break;
884 		default:
885 			break;
886 		}
887 	}
888 
889 	ecode = getaddrinfo(hostptr, "nfs", &hints, ai);
890 	if (ecode != 0) {
891 		syslog(LOG_ERR, "getaddrinfo %s: %s", bindhost,
892 		    gai_strerror(ecode));
893 		return (1);
894 	}
895 	return (0);
896 }
897 
898 static void
899 set_nfsdcnt(int proposed)
900 {
901 
902 	if (proposed < 1) {
903 		warnx("nfsd count too low %d; reset to %d", proposed,
904 		    DEFNFSDCNT);
905 		nfsdcnt = DEFNFSDCNT;
906 	} else if (proposed > MAXNFSDCNT) {
907 		warnx("nfsd count too high %d; truncated to %d", proposed,
908 		    MAXNFSDCNT);
909 		nfsdcnt = MAXNFSDCNT;
910 	} else
911 		nfsdcnt = proposed;
912 	nfsdcnt_set = 1;
913 }
914 
915 static void
916 usage(void)
917 {
918 	(void)fprintf(stderr, "%s", getopt_usage);
919 	exit(1);
920 }
921 
922 static void
923 nonfs(__unused int signo)
924 {
925 	syslog(LOG_ERR, "missing system call: NFS not available");
926 }
927 
928 static void
929 reapchild(__unused int signo)
930 {
931 	pid_t pid;
932 	int i;
933 
934 	while ((pid = wait3(NULL, WNOHANG, NULL)) > 0) {
935 		for (i = 0; i < nfsdcnt; i++)
936 			if (pid == children[i])
937 				children[i] = -1;
938 	}
939 }
940 
941 static void
942 unregistration(void)
943 {
944 	if ((nfs_minvers == NFS_VER2 && !rpcb_unset(NFS_PROGRAM, 2, NULL)) ||
945 	    (nfs_minvers <= NFS_VER3 && !rpcb_unset(NFS_PROGRAM, 3, NULL)))
946 		syslog(LOG_ERR, "rpcb_unset failed");
947 }
948 
949 static void
950 killchildren(void)
951 {
952 	int i;
953 
954 	for (i = 0; i < nfsdcnt; i++) {
955 		if (children[i] > 0)
956 			kill(children[i], SIGKILL);
957 	}
958 }
959 
960 /*
961  * Cleanup master after SIGUSR1.
962  */
963 static void
964 cleanup(__unused int signo)
965 {
966 	nfsd_exit(0);
967 }
968 
969 /*
970  * Cleanup child after SIGUSR1.
971  */
972 static void
973 child_cleanup(__unused int signo)
974 {
975 	exit(0);
976 }
977 
978 static void
979 nfsd_exit(int status)
980 {
981 	killchildren();
982 	unregistration();
983 	exit(status);
984 }
985 
986 static int
987 get_tuned_nfsdcount(void)
988 {
989 	int ncpu, error, tuned_nfsdcnt;
990 	size_t ncpu_size;
991 
992 	ncpu_size = sizeof(ncpu);
993 	error = sysctlbyname("hw.ncpu", &ncpu, &ncpu_size, NULL, 0);
994 	if (error) {
995 		warnx("sysctlbyname(hw.ncpu) failed defaulting to %d nfs servers",
996 		    DEFNFSDCNT);
997 		tuned_nfsdcnt = DEFNFSDCNT;
998 	} else {
999 		tuned_nfsdcnt = ncpu * 8;
1000 	}
1001 	return tuned_nfsdcnt;
1002 }
1003 
1004 static void
1005 start_server(int master, struct nfsd_nfsd_args *nfsdargp, const char *vhost)
1006 {
1007 	char principal[MAXHOSTNAMELEN + 5];
1008 	int status, error;
1009 	char hostname[MAXHOSTNAMELEN + 1], *cp;
1010 	struct addrinfo *aip, hints;
1011 
1012 	status = 0;
1013 	if (vhost == NULL)
1014 		gethostname(hostname, sizeof (hostname));
1015 	else
1016 		strlcpy(hostname, vhost, sizeof (hostname));
1017 	snprintf(principal, sizeof (principal), "nfs@%s", hostname);
1018 	if ((cp = strchr(hostname, '.')) == NULL ||
1019 	    *(cp + 1) == '\0') {
1020 		/* If not fully qualified, try getaddrinfo() */
1021 		memset((void *)&hints, 0, sizeof (hints));
1022 		hints.ai_flags = AI_CANONNAME;
1023 		error = getaddrinfo(hostname, NULL, &hints, &aip);
1024 		if (error == 0) {
1025 			if (aip->ai_canonname != NULL &&
1026 			    (cp = strchr(aip->ai_canonname, '.')) !=
1027 			    NULL && *(cp + 1) != '\0')
1028 				snprintf(principal, sizeof (principal),
1029 				    "nfs@%s", aip->ai_canonname);
1030 			freeaddrinfo(aip);
1031 		}
1032 	}
1033 	nfsdargp->principal = principal;
1034 
1035 	if (nfsdcnt_set)
1036 		nfsdargp->minthreads = nfsdargp->maxthreads = nfsdcnt;
1037 	else {
1038 		nfsdargp->minthreads = minthreads_set ? minthreads : get_tuned_nfsdcount();
1039 		nfsdargp->maxthreads = maxthreads_set ? maxthreads : nfsdargp->minthreads;
1040 		if (nfsdargp->maxthreads < nfsdargp->minthreads)
1041 			nfsdargp->maxthreads = nfsdargp->minthreads;
1042 	}
1043 	error = nfssvc(nfssvc_nfsd, nfsdargp);
1044 	if (error < 0 && errno == EAUTH) {
1045 		/*
1046 		 * This indicates that it could not register the
1047 		 * rpcsec_gss credentials, usually because the
1048 		 * gssd daemon isn't running.
1049 		 * (only the experimental server with nfsv4)
1050 		 */
1051 		syslog(LOG_ERR, "No gssd, using AUTH_SYS only");
1052 		principal[0] = '\0';
1053 		error = nfssvc(nfssvc_nfsd, nfsdargp);
1054 	}
1055 	if (error < 0) {
1056 		if (errno == ENXIO) {
1057 			syslog(LOG_ERR, "Bad -p option, cannot run");
1058 			if (masterpid != 0 && master == 0)
1059 				kill(masterpid, SIGUSR1);
1060 		} else
1061 			syslog(LOG_ERR, "nfssvc: %m");
1062 		status = 1;
1063 	}
1064 	if (master)
1065 		nfsd_exit(status);
1066 	else
1067 		exit(status);
1068 }
1069 
1070 /*
1071  * Open the stable restart file and return the file descriptor for it.
1072  */
1073 static void
1074 open_stable(int *stable_fdp, int *backup_fdp)
1075 {
1076 	int stable_fd, backup_fd = -1, ret;
1077 	struct stat st, backup_st;
1078 
1079 	/* Open and stat the stable restart file. */
1080 	stable_fd = open(NFSD_STABLERESTART, O_RDWR, 0);
1081 	if (stable_fd < 0)
1082 		stable_fd = open(NFSD_STABLERESTART, O_RDWR | O_CREAT, 0600);
1083 	if (stable_fd >= 0) {
1084 		ret = fstat(stable_fd, &st);
1085 		if (ret < 0) {
1086 			close(stable_fd);
1087 			stable_fd = -1;
1088 		}
1089 	}
1090 
1091 	/* Open and stat the backup stable restart file. */
1092 	if (stable_fd >= 0) {
1093 		backup_fd = open(NFSD_STABLEBACKUP, O_RDWR, 0);
1094 		if (backup_fd < 0)
1095 			backup_fd = open(NFSD_STABLEBACKUP, O_RDWR | O_CREAT,
1096 			    0600);
1097 		if (backup_fd >= 0) {
1098 			ret = fstat(backup_fd, &backup_st);
1099 			if (ret < 0) {
1100 				close(backup_fd);
1101 				backup_fd = -1;
1102 			}
1103 		}
1104 		if (backup_fd < 0) {
1105 			close(stable_fd);
1106 			stable_fd = -1;
1107 		}
1108 	}
1109 
1110 	*stable_fdp = stable_fd;
1111 	*backup_fdp = backup_fd;
1112 	if (stable_fd < 0)
1113 		return;
1114 
1115 	/* Sync up the 2 files, as required. */
1116 	if (st.st_size > 0)
1117 		copy_stable(stable_fd, backup_fd);
1118 	else if (backup_st.st_size > 0)
1119 		copy_stable(backup_fd, stable_fd);
1120 }
1121 
1122 /*
1123  * Copy the stable restart file to the backup or vice versa.
1124  */
1125 static void
1126 copy_stable(int from_fd, int to_fd)
1127 {
1128 	int cnt, ret;
1129 	static char buf[1024];
1130 
1131 	ret = lseek(from_fd, (off_t)0, SEEK_SET);
1132 	if (ret >= 0)
1133 		ret = lseek(to_fd, (off_t)0, SEEK_SET);
1134 	if (ret >= 0)
1135 		ret = ftruncate(to_fd, (off_t)0);
1136 	if (ret >= 0)
1137 		do {
1138 			cnt = read(from_fd, buf, 1024);
1139 			if (cnt > 0)
1140 				ret = write(to_fd, buf, cnt);
1141 			else if (cnt < 0)
1142 				ret = cnt;
1143 		} while (cnt > 0 && ret >= 0);
1144 	if (ret >= 0)
1145 		ret = fsync(to_fd);
1146 	if (ret < 0)
1147 		syslog(LOG_ERR, "stable restart copy failure: %m");
1148 }
1149 
1150 /*
1151  * Back up the stable restart file when indicated by the kernel.
1152  */
1153 static void
1154 backup_stable(__unused int signo)
1155 {
1156 
1157 	if (stablefd >= 0)
1158 		copy_stable(stablefd, backupfd);
1159 }
1160 
1161 /*
1162  * Parse the pNFS string and extract the DS servers and ports numbers.
1163  */
1164 static void
1165 parse_dsserver(const char *optionarg, struct nfsd_nfsd_args *nfsdargp)
1166 {
1167 	char *cp, *cp2, *dsaddr, *dshost, *dspath, *dsvol, nfsprt[9];
1168 	char *mdspath, *mdsp, ip6[INET6_ADDRSTRLEN];
1169 	const char *ad;
1170 	int ecode;
1171 	u_int adsiz, dsaddrcnt, dshostcnt, dspathcnt, hostsiz, pathsiz;
1172 	u_int mdspathcnt;
1173 	size_t dsaddrsiz, dshostsiz, dspathsiz, nfsprtsiz, mdspathsiz;
1174 	struct addrinfo hints, *ai_tcp, *res;
1175 	struct sockaddr_in sin;
1176 	struct sockaddr_in6 sin6;
1177 
1178 	cp = strdup(optionarg);
1179 	if (cp == NULL)
1180 		errx(1, "Out of memory");
1181 
1182 	/* Now, do the host names. */
1183 	dspathsiz = 1024;
1184 	dspathcnt = 0;
1185 	dspath = malloc(dspathsiz);
1186 	if (dspath == NULL)
1187 		errx(1, "Out of memory");
1188 	dshostsiz = 1024;
1189 	dshostcnt = 0;
1190 	dshost = malloc(dshostsiz);
1191 	if (dshost == NULL)
1192 		errx(1, "Out of memory");
1193 	dsaddrsiz = 1024;
1194 	dsaddrcnt = 0;
1195 	dsaddr = malloc(dsaddrsiz);
1196 	if (dsaddr == NULL)
1197 		errx(1, "Out of memory");
1198 	mdspathsiz = 1024;
1199 	mdspathcnt = 0;
1200 	mdspath = malloc(mdspathsiz);
1201 	if (mdspath == NULL)
1202 		errx(1, "Out of memory");
1203 
1204 	/* Put the NFS port# in "." form. */
1205 	snprintf(nfsprt, 9, ".%d.%d", 2049 >> 8, 2049 & 0xff);
1206 	nfsprtsiz = strlen(nfsprt);
1207 
1208 	ai_tcp = NULL;
1209 	/* Loop around for each DS server name. */
1210 	do {
1211 		cp2 = strchr(cp, ',');
1212 		if (cp2 != NULL) {
1213 			/* Not the last DS in the list. */
1214 			*cp2++ = '\0';
1215 			if (*cp2 == '\0')
1216 				usage();
1217 		}
1218 
1219 		dsvol = strchr(cp, ':');
1220 		if (dsvol == NULL || *(dsvol + 1) == '\0')
1221 			usage();
1222 		*dsvol++ = '\0';
1223 
1224 		/* Optional path for MDS file system to be stored on DS. */
1225 		mdsp = strchr(dsvol, '#');
1226 		if (mdsp != NULL) {
1227 			if (*(mdsp + 1) == '\0' || mdsp <= dsvol)
1228 				usage();
1229 			*mdsp++ = '\0';
1230 		}
1231 
1232 		/* Append this pathname to dspath. */
1233 		pathsiz = strlen(dsvol);
1234 		if (dspathcnt + pathsiz + 1 > dspathsiz) {
1235 			dspathsiz *= 2;
1236 			dspath = realloc(dspath, dspathsiz);
1237 			if (dspath == NULL)
1238 				errx(1, "Out of memory");
1239 		}
1240 		strcpy(&dspath[dspathcnt], dsvol);
1241 		dspathcnt += pathsiz + 1;
1242 
1243 		/* Append this pathname to mdspath. */
1244 		if (mdsp != NULL)
1245 			pathsiz = strlen(mdsp);
1246 		else
1247 			pathsiz = 0;
1248 		if (mdspathcnt + pathsiz + 1 > mdspathsiz) {
1249 			mdspathsiz *= 2;
1250 			mdspath = realloc(mdspath, mdspathsiz);
1251 			if (mdspath == NULL)
1252 				errx(1, "Out of memory");
1253 		}
1254 		if (mdsp != NULL)
1255 			strcpy(&mdspath[mdspathcnt], mdsp);
1256 		else
1257 			mdspath[mdspathcnt] = '\0';
1258 		mdspathcnt += pathsiz + 1;
1259 
1260 		if (ai_tcp != NULL)
1261 			freeaddrinfo(ai_tcp);
1262 
1263 		/* Get the fully qualified domain name and IP address. */
1264 		memset(&hints, 0, sizeof(hints));
1265 		hints.ai_flags = AI_CANONNAME | AI_ADDRCONFIG;
1266 		hints.ai_family = PF_UNSPEC;
1267 		hints.ai_socktype = SOCK_STREAM;
1268 		hints.ai_protocol = IPPROTO_TCP;
1269 		ecode = getaddrinfo(cp, NULL, &hints, &ai_tcp);
1270 		if (ecode != 0)
1271 			err(1, "getaddrinfo pnfs: %s %s", cp,
1272 			    gai_strerror(ecode));
1273 		ad = NULL;
1274 		for (res = ai_tcp; res != NULL; res = res->ai_next) {
1275 			if (res->ai_addr->sa_family == AF_INET) {
1276 				if (res->ai_addrlen < sizeof(sin))
1277 					err(1, "getaddrinfo() returned "
1278 					    "undersized IPv4 address");
1279 				/*
1280 				 * Mips cares about sockaddr_in alignment,
1281 				 * so copy the address.
1282 				 */
1283 				memcpy(&sin, res->ai_addr, sizeof(sin));
1284 				ad = inet_ntoa(sin.sin_addr);
1285 				break;
1286 			} else if (res->ai_family == AF_INET6) {
1287 				if (res->ai_addrlen < sizeof(sin6))
1288 					err(1, "getaddrinfo() returned "
1289 					    "undersized IPv6 address");
1290 				/*
1291 				 * Mips cares about sockaddr_in6 alignment,
1292 				 * so copy the address.
1293 				 */
1294 				memcpy(&sin6, res->ai_addr, sizeof(sin6));
1295 				ad = inet_ntop(AF_INET6, &sin6.sin6_addr, ip6,
1296 				    sizeof(ip6));
1297 
1298 				/*
1299 				 * XXX
1300 				 * Since a link local address will only
1301 				 * work if the client and DS are in the
1302 				 * same scope zone, only use it if it is
1303 				 * the only address.
1304 				 */
1305 				if (ad != NULL &&
1306 				    !IN6_IS_ADDR_LINKLOCAL(&sin6.sin6_addr))
1307 					break;
1308 			}
1309 		}
1310 		if (ad == NULL)
1311 			err(1, "No IP address for %s", cp);
1312 
1313 		/* Append this address to dsaddr. */
1314 		adsiz = strlen(ad);
1315 		if (dsaddrcnt + adsiz + nfsprtsiz + 1 > dsaddrsiz) {
1316 			dsaddrsiz *= 2;
1317 			dsaddr = realloc(dsaddr, dsaddrsiz);
1318 			if (dsaddr == NULL)
1319 				errx(1, "Out of memory");
1320 		}
1321 		strcpy(&dsaddr[dsaddrcnt], ad);
1322 		strcat(&dsaddr[dsaddrcnt], nfsprt);
1323 		dsaddrcnt += adsiz + nfsprtsiz + 1;
1324 
1325 		/* Append this hostname to dshost. */
1326 		hostsiz = strlen(ai_tcp->ai_canonname);
1327 		if (dshostcnt + hostsiz + 1 > dshostsiz) {
1328 			dshostsiz *= 2;
1329 			dshost = realloc(dshost, dshostsiz);
1330 			if (dshost == NULL)
1331 				errx(1, "Out of memory");
1332 		}
1333 		strcpy(&dshost[dshostcnt], ai_tcp->ai_canonname);
1334 		dshostcnt += hostsiz + 1;
1335 
1336 		cp = cp2;
1337 	} while (cp != NULL);
1338 
1339 	nfsdargp->addr = dsaddr;
1340 	nfsdargp->addrlen = dsaddrcnt;
1341 	nfsdargp->dnshost = dshost;
1342 	nfsdargp->dnshostlen = dshostcnt;
1343 	nfsdargp->dspath = dspath;
1344 	nfsdargp->dspathlen = dspathcnt;
1345 	nfsdargp->mdspath = mdspath;
1346 	nfsdargp->mdspathlen = mdspathcnt;
1347 	freeaddrinfo(ai_tcp);
1348 }
1349 
1350