xref: /freebsd/usr.sbin/nfsd/nfsd.c (revision fe2494903422ba3b924eba82cb63a6a9188fad7a)
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 #ifndef lint
36 static const char copyright[] =
37 "@(#) Copyright (c) 1989, 1993, 1994\n\
38 	The Regents of the University of California.  All rights reserved.\n";
39 #endif /* not lint */
40 
41 #ifndef lint
42 #if 0
43 static char sccsid[] = "@(#)nfsd.c	8.9 (Berkeley) 3/29/95";
44 #endif
45 static const char rcsid[] =
46   "$FreeBSD$";
47 #endif /* not lint */
48 
49 #include <sys/param.h>
50 #include <sys/syslog.h>
51 #include <sys/wait.h>
52 #include <sys/mount.h>
53 #include <sys/fcntl.h>
54 #include <sys/linker.h>
55 #include <sys/module.h>
56 #include <sys/types.h>
57 #include <sys/stat.h>
58 #include <sys/sysctl.h>
59 #include <sys/ucred.h>
60 
61 #include <rpc/rpc.h>
62 #include <rpc/pmap_clnt.h>
63 #include <rpcsvc/nfs_prot.h>
64 
65 #include <netdb.h>
66 #include <arpa/inet.h>
67 #include <nfs/nfssvc.h>
68 
69 #include <fs/nfs/nfsproto.h>
70 #include <fs/nfs/nfskpiport.h>
71 #include <fs/nfs/nfs.h>
72 
73 #include <err.h>
74 #include <errno.h>
75 #include <signal.h>
76 #include <stdio.h>
77 #include <stdlib.h>
78 #include <string.h>
79 #include <unistd.h>
80 #include <sysexits.h>
81 
82 #include <getopt.h>
83 
84 static int	debug = 0;
85 
86 #define	NFSD_STABLERESTART	"/var/db/nfs-stablerestart"
87 #define	NFSD_STABLEBACKUP	"/var/db/nfs-stablerestart.bak"
88 #define	MAXNFSDCNT	256
89 #define	DEFNFSDCNT	 4
90 #define	NFS_VER2	 2
91 #define NFS_VER3	 3
92 #define NFS_VER4	 4
93 static pid_t children[MAXNFSDCNT]; /* PIDs of children */
94 static pid_t masterpid;		   /* PID of master/parent */
95 static int nfsdcnt;		/* number of children */
96 static int nfsdcnt_set;
97 static int minthreads;
98 static int maxthreads;
99 static int nfssvc_nfsd;		/* Set to correct NFSSVC_xxx flag */
100 static int stablefd = -1;	/* Fd for the stable restart file */
101 static int backupfd;		/* Fd for the backup stable restart file */
102 static const char *getopt_shortopts;
103 static const char *getopt_usage;
104 
105 static int minthreads_set;
106 static int maxthreads_set;
107 
108 static struct option longopts[] = {
109 	{ "debug", no_argument, &debug, 1 },
110 	{ "minthreads", required_argument, &minthreads_set, 1 },
111 	{ "maxthreads", required_argument, &maxthreads_set, 1 },
112 	{ "pnfs", required_argument, NULL, 'p' },
113 	{ "mirror", required_argument, NULL, 'm' },
114 	{ NULL, 0, NULL, 0}
115 };
116 
117 static void	cleanup(int);
118 static void	child_cleanup(int);
119 static void	killchildren(void);
120 static void	nfsd_exit(int);
121 static void	nonfs(int);
122 static void	reapchild(int);
123 static int	setbindhost(struct addrinfo **ia, const char *bindhost,
124 		    struct addrinfo hints);
125 static void	start_server(int, struct nfsd_nfsd_args *);
126 static void	unregistration(void);
127 static void	usage(void);
128 static void	open_stable(int *, int *);
129 static void	copy_stable(int, int);
130 static void	backup_stable(int);
131 static void	set_nfsdcnt(int);
132 static void	parse_dsserver(const char *, struct nfsd_nfsd_args *);
133 
134 /*
135  * Nfs server daemon mostly just a user context for nfssvc()
136  *
137  * 1 - do file descriptor and signal cleanup
138  * 2 - fork the nfsd(s)
139  * 3 - create server socket(s)
140  * 4 - register socket with rpcbind
141  *
142  * For connectionless protocols, just pass the socket into the kernel via.
143  * nfssvc().
144  * For connection based sockets, loop doing accepts. When you get a new
145  * socket from accept, pass the msgsock into the kernel via. nfssvc().
146  * The arguments are:
147  *	-r - reregister with rpcbind
148  *	-d - unregister with rpcbind
149  *	-t - support tcp nfs clients
150  *	-u - support udp nfs clients
151  *	-e - forces it to run a server that supports nfsv4
152  *	-p - enable a pNFS service
153  *	-m - set the mirroring level for a pNFS service
154  * followed by "n" which is the number of nfsds' to fork off
155  */
156 int
157 main(int argc, char **argv)
158 {
159 	struct nfsd_addsock_args addsockargs;
160 	struct addrinfo *ai_udp, *ai_tcp, *ai_udp6, *ai_tcp6, hints;
161 	struct netconfig *nconf_udp, *nconf_tcp, *nconf_udp6, *nconf_tcp6;
162 	struct netbuf nb_udp, nb_tcp, nb_udp6, nb_tcp6;
163 	struct sockaddr_storage peer;
164 	fd_set ready, sockbits;
165 	int ch, connect_type_cnt, i, maxsock, msgsock;
166 	socklen_t len;
167 	int on = 1, unregister, reregister, sock;
168 	int tcp6sock, ip6flag, tcpflag, tcpsock;
169 	int udpflag, ecode, error, s;
170 	int bindhostc, bindanyflag, rpcbreg, rpcbregcnt;
171 	int nfssvc_addsock;
172 	int longindex = 0;
173 	int nfs_minvers = NFS_VER2;
174 	size_t nfs_minvers_size;
175 	const char *lopt;
176 	char **bindhost = NULL;
177 	pid_t pid;
178 	struct nfsd_nfsd_args nfsdargs;
179 
180 	nfsdargs.mirrorcnt = 1;
181 	nfsdargs.addr = NULL;
182 	nfsdargs.addrlen = 0;
183 	nfsdcnt = DEFNFSDCNT;
184 	unregister = reregister = tcpflag = maxsock = 0;
185 	bindanyflag = udpflag = connect_type_cnt = bindhostc = 0;
186 	getopt_shortopts = "ah:n:rdtuep:m:";
187 	getopt_usage =
188 	    "usage:\n"
189 	    "  nfsd [-ardtue] [-h bindip]\n"
190 	    "       [-n numservers] [--minthreads #] [--maxthreads #]\n"
191 	    "       [-p/--pnfs dsserver0:/dsserver0-mounted-on-dir,...,"
192 	    "dsserverN:/dsserverN-mounted-on-dir] [-m mirrorlevel]\n";
193 	while ((ch = getopt_long(argc, argv, getopt_shortopts, longopts,
194 		    &longindex)) != -1)
195 		switch (ch) {
196 		case 'a':
197 			bindanyflag = 1;
198 			break;
199 		case 'n':
200 			set_nfsdcnt(atoi(optarg));
201 			break;
202 		case 'h':
203 			bindhostc++;
204 			bindhost = realloc(bindhost,sizeof(char *)*bindhostc);
205 			if (bindhost == NULL)
206 				errx(1, "Out of memory");
207 			bindhost[bindhostc-1] = strdup(optarg);
208 			if (bindhost[bindhostc-1] == NULL)
209 				errx(1, "Out of memory");
210 			break;
211 		case 'r':
212 			reregister = 1;
213 			break;
214 		case 'd':
215 			unregister = 1;
216 			break;
217 		case 't':
218 			tcpflag = 1;
219 			break;
220 		case 'u':
221 			udpflag = 1;
222 			break;
223 		case 'e':
224 			/* now a no-op, since this is the default */
225 			break;
226 		case 'p':
227 			/* Parse out the DS server host names and mount pts. */
228 			parse_dsserver(optarg, &nfsdargs);
229 			break;
230 		case 'm':
231 			/* Set the mirror level for a pNFS service. */
232 			i = atoi(optarg);
233 			if (i < 2 || i > NFSDEV_MAXMIRRORS)
234 				errx(1, "Mirror level out of range 2<-->%d",
235 				    NFSDEV_MAXMIRRORS);
236 			nfsdargs.mirrorcnt = i;
237 			break;
238 		case 0:
239 			lopt = longopts[longindex].name;
240 			if (!strcmp(lopt, "minthreads")) {
241 				minthreads = atoi(optarg);
242 			} else if (!strcmp(lopt, "maxthreads")) {
243 				maxthreads = atoi(optarg);
244 			}
245 			break;
246 		default:
247 		case '?':
248 			usage();
249 		}
250 	if (!tcpflag && !udpflag)
251 		udpflag = 1;
252 	argv += optind;
253 	argc -= optind;
254 	if (minthreads_set && maxthreads_set && minthreads > maxthreads)
255 		errx(EX_USAGE,
256 		    "error: minthreads(%d) can't be greater than "
257 		    "maxthreads(%d)", minthreads, maxthreads);
258 
259 	/*
260 	 * XXX
261 	 * Backward compatibility, trailing number is the count of daemons.
262 	 */
263 	if (argc > 1)
264 		usage();
265 	if (argc == 1)
266 		set_nfsdcnt(atoi(argv[0]));
267 
268 	/*
269 	 * Unless the "-o" option was specified, try and run "nfsd".
270 	 * If "-o" was specified, try and run "nfsserver".
271 	 */
272 	if (modfind("nfsd") < 0) {
273 		/* Not present in kernel, try loading it */
274 		if (kldload("nfsd") < 0 || modfind("nfsd") < 0)
275 			errx(1, "NFS server is not available");
276 	}
277 
278 	ip6flag = 1;
279 	s = socket(AF_INET6, SOCK_DGRAM, IPPROTO_UDP);
280 	if (s == -1) {
281 		if (errno != EPROTONOSUPPORT && errno != EAFNOSUPPORT)
282 			err(1, "socket");
283 		ip6flag = 0;
284 	} else if (getnetconfigent("udp6") == NULL ||
285 		getnetconfigent("tcp6") == NULL) {
286 		ip6flag = 0;
287 	}
288 	if (s != -1)
289 		close(s);
290 
291 	if (bindhostc == 0 || bindanyflag) {
292 		bindhostc++;
293 		bindhost = realloc(bindhost,sizeof(char *)*bindhostc);
294 		if (bindhost == NULL)
295 			errx(1, "Out of memory");
296 		bindhost[bindhostc-1] = strdup("*");
297 		if (bindhost[bindhostc-1] == NULL)
298 			errx(1, "Out of memory");
299 	}
300 
301 	nfs_minvers_size = sizeof(nfs_minvers);
302 	error = sysctlbyname("vfs.nfsd.server_min_nfsvers", &nfs_minvers,
303 	    &nfs_minvers_size, NULL, 0);
304 	if (error != 0 || nfs_minvers < NFS_VER2 || nfs_minvers > NFS_VER4) {
305 		warnx("sysctlbyname(vfs.nfsd.server_min_nfsvers) failed,"
306 		    " defaulting to NFSv2");
307 		nfs_minvers = NFS_VER2;
308 	}
309 
310 	if (unregister) {
311 		unregistration();
312 		exit (0);
313 	}
314 	if (reregister) {
315 		if (udpflag) {
316 			memset(&hints, 0, sizeof hints);
317 			hints.ai_flags = AI_PASSIVE;
318 			hints.ai_family = AF_INET;
319 			hints.ai_socktype = SOCK_DGRAM;
320 			hints.ai_protocol = IPPROTO_UDP;
321 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp);
322 			if (ecode != 0)
323 				err(1, "getaddrinfo udp: %s", gai_strerror(ecode));
324 			nconf_udp = getnetconfigent("udp");
325 			if (nconf_udp == NULL)
326 				err(1, "getnetconfigent udp failed");
327 			nb_udp.buf = ai_udp->ai_addr;
328 			nb_udp.len = nb_udp.maxlen = ai_udp->ai_addrlen;
329 			if (nfs_minvers == NFS_VER2)
330 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp,
331 				    &nb_udp))
332 					err(1, "rpcb_set udp failed");
333 			if (nfs_minvers <= NFS_VER3)
334 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp,
335 				    &nb_udp))
336 					err(1, "rpcb_set udp failed");
337 			freeaddrinfo(ai_udp);
338 		}
339 		if (udpflag && ip6flag) {
340 			memset(&hints, 0, sizeof hints);
341 			hints.ai_flags = AI_PASSIVE;
342 			hints.ai_family = AF_INET6;
343 			hints.ai_socktype = SOCK_DGRAM;
344 			hints.ai_protocol = IPPROTO_UDP;
345 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp6);
346 			if (ecode != 0)
347 				err(1, "getaddrinfo udp6: %s", gai_strerror(ecode));
348 			nconf_udp6 = getnetconfigent("udp6");
349 			if (nconf_udp6 == NULL)
350 				err(1, "getnetconfigent udp6 failed");
351 			nb_udp6.buf = ai_udp6->ai_addr;
352 			nb_udp6.len = nb_udp6.maxlen = ai_udp6->ai_addrlen;
353 			if (nfs_minvers == NFS_VER2)
354 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp6,
355 				    &nb_udp6))
356 					err(1, "rpcb_set udp6 failed");
357 			if (nfs_minvers <= NFS_VER3)
358 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp6,
359 				    &nb_udp6))
360 					err(1, "rpcb_set udp6 failed");
361 			freeaddrinfo(ai_udp6);
362 		}
363 		if (tcpflag) {
364 			memset(&hints, 0, sizeof hints);
365 			hints.ai_flags = AI_PASSIVE;
366 			hints.ai_family = AF_INET;
367 			hints.ai_socktype = SOCK_STREAM;
368 			hints.ai_protocol = IPPROTO_TCP;
369 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_tcp);
370 			if (ecode != 0)
371 				err(1, "getaddrinfo tcp: %s", gai_strerror(ecode));
372 			nconf_tcp = getnetconfigent("tcp");
373 			if (nconf_tcp == NULL)
374 				err(1, "getnetconfigent tcp failed");
375 			nb_tcp.buf = ai_tcp->ai_addr;
376 			nb_tcp.len = nb_tcp.maxlen = ai_tcp->ai_addrlen;
377 			if (nfs_minvers == NFS_VER2)
378 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp,
379 				    &nb_tcp))
380 					err(1, "rpcb_set tcp failed");
381 			if (nfs_minvers <= NFS_VER3)
382 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp,
383 				    &nb_tcp))
384 					err(1, "rpcb_set tcp failed");
385 			freeaddrinfo(ai_tcp);
386 		}
387 		if (tcpflag && ip6flag) {
388 			memset(&hints, 0, sizeof hints);
389 			hints.ai_flags = AI_PASSIVE;
390 			hints.ai_family = AF_INET6;
391 			hints.ai_socktype = SOCK_STREAM;
392 			hints.ai_protocol = IPPROTO_TCP;
393 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_tcp6);
394 			if (ecode != 0)
395 				err(1, "getaddrinfo tcp6: %s", gai_strerror(ecode));
396 			nconf_tcp6 = getnetconfigent("tcp6");
397 			if (nconf_tcp6 == NULL)
398 				err(1, "getnetconfigent tcp6 failed");
399 			nb_tcp6.buf = ai_tcp6->ai_addr;
400 			nb_tcp6.len = nb_tcp6.maxlen = ai_tcp6->ai_addrlen;
401 			if (nfs_minvers == NFS_VER2)
402 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp6,
403 				    &nb_tcp6))
404 					err(1, "rpcb_set tcp6 failed");
405 			if (nfs_minvers <= NFS_VER3)
406 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp6,
407 				   &nb_tcp6))
408 					err(1, "rpcb_set tcp6 failed");
409 			freeaddrinfo(ai_tcp6);
410 		}
411 		exit (0);
412 	}
413 	if (debug == 0) {
414 		daemon(0, 0);
415 		(void)signal(SIGHUP, SIG_IGN);
416 		(void)signal(SIGINT, SIG_IGN);
417 		/*
418 		 * nfsd sits in the kernel most of the time.  It needs
419 		 * to ignore SIGTERM/SIGQUIT in order to stay alive as long
420 		 * as possible during a shutdown, otherwise loopback
421 		 * mounts will not be able to unmount.
422 		 */
423 		(void)signal(SIGTERM, SIG_IGN);
424 		(void)signal(SIGQUIT, SIG_IGN);
425 	}
426 	(void)signal(SIGSYS, nonfs);
427 	(void)signal(SIGCHLD, reapchild);
428 	(void)signal(SIGUSR2, backup_stable);
429 
430 	openlog("nfsd", LOG_PID | (debug ? LOG_PERROR : 0), LOG_DAEMON);
431 
432 	/*
433 	 * For V4, we open the stablerestart file and call nfssvc()
434 	 * to get it loaded. This is done before the daemons do the
435 	 * regular nfssvc() call to service NFS requests.
436 	 * (This way the file remains open until the last nfsd is killed
437 	 *  off.)
438 	 * It and the backup copy will be created as empty files
439 	 * the first time this nfsd is started and should never be
440 	 * deleted/replaced if at all possible. It should live on a
441 	 * local, non-volatile storage device that does not do hardware
442 	 * level write-back caching. (See SCSI doc for more information
443 	 * on how to prevent write-back caching on SCSI disks.)
444 	 */
445 	open_stable(&stablefd, &backupfd);
446 	if (stablefd < 0) {
447 		syslog(LOG_ERR, "Can't open %s: %m\n", NFSD_STABLERESTART);
448 		exit(1);
449 	}
450 	/* This system call will fail for old kernels, but that's ok. */
451 	nfssvc(NFSSVC_BACKUPSTABLE, NULL);
452 	if (nfssvc(NFSSVC_STABLERESTART, (caddr_t)&stablefd) < 0) {
453 		syslog(LOG_ERR, "Can't read stable storage file: %m\n");
454 		exit(1);
455 	}
456 	nfssvc_addsock = NFSSVC_NFSDADDSOCK;
457 	nfssvc_nfsd = NFSSVC_NFSDNFSD | NFSSVC_NEWSTRUCT;
458 
459 	if (tcpflag) {
460 		/*
461 		 * For TCP mode, we fork once to start the first
462 		 * kernel nfsd thread. The kernel will add more
463 		 * threads as needed.
464 		 */
465 		masterpid = getpid();
466 		pid = fork();
467 		if (pid == -1) {
468 			syslog(LOG_ERR, "fork: %m");
469 			nfsd_exit(1);
470 		}
471 		if (pid) {
472 			children[0] = pid;
473 		} else {
474 			(void)signal(SIGUSR1, child_cleanup);
475 			setproctitle("server");
476 			start_server(0, &nfsdargs);
477 		}
478 	}
479 
480 	(void)signal(SIGUSR1, cleanup);
481 	FD_ZERO(&sockbits);
482 
483 	rpcbregcnt = 0;
484 	/* Set up the socket for udp and rpcb register it. */
485 	if (udpflag) {
486 		rpcbreg = 0;
487 		for (i = 0; i < bindhostc; i++) {
488 			memset(&hints, 0, sizeof hints);
489 			hints.ai_flags = AI_PASSIVE;
490 			hints.ai_family = AF_INET;
491 			hints.ai_socktype = SOCK_DGRAM;
492 			hints.ai_protocol = IPPROTO_UDP;
493 			if (setbindhost(&ai_udp, bindhost[i], hints) == 0) {
494 				rpcbreg = 1;
495 				rpcbregcnt++;
496 				if ((sock = socket(ai_udp->ai_family,
497 				    ai_udp->ai_socktype,
498 				    ai_udp->ai_protocol)) < 0) {
499 					syslog(LOG_ERR,
500 					    "can't create udp socket");
501 					nfsd_exit(1);
502 				}
503 				if (bind(sock, ai_udp->ai_addr,
504 				    ai_udp->ai_addrlen) < 0) {
505 					syslog(LOG_ERR,
506 					    "can't bind udp addr %s: %m",
507 					    bindhost[i]);
508 					nfsd_exit(1);
509 				}
510 				freeaddrinfo(ai_udp);
511 				addsockargs.sock = sock;
512 				addsockargs.name = NULL;
513 				addsockargs.namelen = 0;
514 				if (nfssvc(nfssvc_addsock, &addsockargs) < 0) {
515 					syslog(LOG_ERR, "can't Add UDP socket");
516 					nfsd_exit(1);
517 				}
518 				(void)close(sock);
519 			}
520 		}
521 		if (rpcbreg == 1) {
522 			memset(&hints, 0, sizeof hints);
523 			hints.ai_flags = AI_PASSIVE;
524 			hints.ai_family = AF_INET;
525 			hints.ai_socktype = SOCK_DGRAM;
526 			hints.ai_protocol = IPPROTO_UDP;
527 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp);
528 			if (ecode != 0) {
529 				syslog(LOG_ERR, "getaddrinfo udp: %s",
530 				   gai_strerror(ecode));
531 				nfsd_exit(1);
532 			}
533 			nconf_udp = getnetconfigent("udp");
534 			if (nconf_udp == NULL)
535 				err(1, "getnetconfigent udp failed");
536 			nb_udp.buf = ai_udp->ai_addr;
537 			nb_udp.len = nb_udp.maxlen = ai_udp->ai_addrlen;
538 			if (nfs_minvers == NFS_VER2)
539 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp,
540 				    &nb_udp))
541 					err(1, "rpcb_set udp failed");
542 			if (nfs_minvers <= NFS_VER3)
543 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp,
544 				    &nb_udp))
545 					err(1, "rpcb_set udp failed");
546 			freeaddrinfo(ai_udp);
547 		}
548 	}
549 
550 	/* Set up the socket for udp6 and rpcb register it. */
551 	if (udpflag && ip6flag) {
552 		rpcbreg = 0;
553 		for (i = 0; i < bindhostc; i++) {
554 			memset(&hints, 0, sizeof hints);
555 			hints.ai_flags = AI_PASSIVE;
556 			hints.ai_family = AF_INET6;
557 			hints.ai_socktype = SOCK_DGRAM;
558 			hints.ai_protocol = IPPROTO_UDP;
559 			if (setbindhost(&ai_udp6, bindhost[i], hints) == 0) {
560 				rpcbreg = 1;
561 				rpcbregcnt++;
562 				if ((sock = socket(ai_udp6->ai_family,
563 				    ai_udp6->ai_socktype,
564 				    ai_udp6->ai_protocol)) < 0) {
565 					syslog(LOG_ERR,
566 						"can't create udp6 socket");
567 					nfsd_exit(1);
568 				}
569 				if (setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY,
570 				    &on, sizeof on) < 0) {
571 					syslog(LOG_ERR,
572 					    "can't set v6-only binding for "
573 					    "udp6 socket: %m");
574 					nfsd_exit(1);
575 				}
576 				if (bind(sock, ai_udp6->ai_addr,
577 				    ai_udp6->ai_addrlen) < 0) {
578 					syslog(LOG_ERR,
579 					    "can't bind udp6 addr %s: %m",
580 					    bindhost[i]);
581 					nfsd_exit(1);
582 				}
583 				freeaddrinfo(ai_udp6);
584 				addsockargs.sock = sock;
585 				addsockargs.name = NULL;
586 				addsockargs.namelen = 0;
587 				if (nfssvc(nfssvc_addsock, &addsockargs) < 0) {
588 					syslog(LOG_ERR,
589 					    "can't add UDP6 socket");
590 					nfsd_exit(1);
591 				}
592 				(void)close(sock);
593 			}
594 		}
595 		if (rpcbreg == 1) {
596 			memset(&hints, 0, sizeof hints);
597 			hints.ai_flags = AI_PASSIVE;
598 			hints.ai_family = AF_INET6;
599 			hints.ai_socktype = SOCK_DGRAM;
600 			hints.ai_protocol = IPPROTO_UDP;
601 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp6);
602 			if (ecode != 0) {
603 				syslog(LOG_ERR, "getaddrinfo udp6: %s",
604 				   gai_strerror(ecode));
605 				nfsd_exit(1);
606 			}
607 			nconf_udp6 = getnetconfigent("udp6");
608 			if (nconf_udp6 == NULL)
609 				err(1, "getnetconfigent udp6 failed");
610 			nb_udp6.buf = ai_udp6->ai_addr;
611 			nb_udp6.len = nb_udp6.maxlen = ai_udp6->ai_addrlen;
612 			if (nfs_minvers == NFS_VER2)
613 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp6,
614 				    &nb_udp6))
615 					err(1,
616 					    "rpcb_set udp6 failed");
617 			if (nfs_minvers <= NFS_VER3)
618 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp6,
619 				    &nb_udp6))
620 					err(1,
621 					    "rpcb_set udp6 failed");
622 			freeaddrinfo(ai_udp6);
623 		}
624 	}
625 
626 	/* Set up the socket for tcp and rpcb register it. */
627 	if (tcpflag) {
628 		rpcbreg = 0;
629 		for (i = 0; i < bindhostc; i++) {
630 			memset(&hints, 0, sizeof hints);
631 			hints.ai_flags = AI_PASSIVE;
632 			hints.ai_family = AF_INET;
633 			hints.ai_socktype = SOCK_STREAM;
634 			hints.ai_protocol = IPPROTO_TCP;
635 			if (setbindhost(&ai_tcp, bindhost[i], hints) == 0) {
636 				rpcbreg = 1;
637 				rpcbregcnt++;
638 				if ((tcpsock = socket(AF_INET, SOCK_STREAM,
639 				    0)) < 0) {
640 					syslog(LOG_ERR,
641 					    "can't create tcp socket");
642 					nfsd_exit(1);
643 				}
644 				if (setsockopt(tcpsock, SOL_SOCKET,
645 				    SO_REUSEADDR,
646 				    (char *)&on, sizeof(on)) < 0)
647 					syslog(LOG_ERR,
648 					     "setsockopt SO_REUSEADDR: %m");
649 				if (bind(tcpsock, ai_tcp->ai_addr,
650 				    ai_tcp->ai_addrlen) < 0) {
651 					syslog(LOG_ERR,
652 					    "can't bind tcp addr %s: %m",
653 					    bindhost[i]);
654 					nfsd_exit(1);
655 				}
656 				if (listen(tcpsock, -1) < 0) {
657 					syslog(LOG_ERR, "listen failed");
658 					nfsd_exit(1);
659 				}
660 				freeaddrinfo(ai_tcp);
661 				FD_SET(tcpsock, &sockbits);
662 				maxsock = tcpsock;
663 				connect_type_cnt++;
664 			}
665 		}
666 		if (rpcbreg == 1) {
667 			memset(&hints, 0, sizeof hints);
668 			hints.ai_flags = AI_PASSIVE;
669 			hints.ai_family = AF_INET;
670 			hints.ai_socktype = SOCK_STREAM;
671 			hints.ai_protocol = IPPROTO_TCP;
672 			ecode = getaddrinfo(NULL, "nfs", &hints,
673 			     &ai_tcp);
674 			if (ecode != 0) {
675 				syslog(LOG_ERR, "getaddrinfo tcp: %s",
676 				   gai_strerror(ecode));
677 				nfsd_exit(1);
678 			}
679 			nconf_tcp = getnetconfigent("tcp");
680 			if (nconf_tcp == NULL)
681 				err(1, "getnetconfigent tcp failed");
682 			nb_tcp.buf = ai_tcp->ai_addr;
683 			nb_tcp.len = nb_tcp.maxlen = ai_tcp->ai_addrlen;
684 			if (nfs_minvers == NFS_VER2)
685 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp,
686 				    &nb_tcp))
687 					err(1, "rpcb_set tcp failed");
688 			if (nfs_minvers <= NFS_VER3)
689 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp,
690 				    &nb_tcp))
691 					err(1, "rpcb_set tcp failed");
692 			freeaddrinfo(ai_tcp);
693 		}
694 	}
695 
696 	/* Set up the socket for tcp6 and rpcb register it. */
697 	if (tcpflag && ip6flag) {
698 		rpcbreg = 0;
699 		for (i = 0; i < bindhostc; i++) {
700 			memset(&hints, 0, sizeof hints);
701 			hints.ai_flags = AI_PASSIVE;
702 			hints.ai_family = AF_INET6;
703 			hints.ai_socktype = SOCK_STREAM;
704 			hints.ai_protocol = IPPROTO_TCP;
705 			if (setbindhost(&ai_tcp6, bindhost[i], hints) == 0) {
706 				rpcbreg = 1;
707 				rpcbregcnt++;
708 				if ((tcp6sock = socket(ai_tcp6->ai_family,
709 				    ai_tcp6->ai_socktype,
710 				    ai_tcp6->ai_protocol)) < 0) {
711 					syslog(LOG_ERR,
712 					    "can't create tcp6 socket");
713 					nfsd_exit(1);
714 				}
715 				if (setsockopt(tcp6sock, SOL_SOCKET,
716 				    SO_REUSEADDR,
717 				    (char *)&on, sizeof(on)) < 0)
718 					syslog(LOG_ERR,
719 					    "setsockopt SO_REUSEADDR: %m");
720 				if (setsockopt(tcp6sock, IPPROTO_IPV6,
721 				    IPV6_V6ONLY, &on, sizeof on) < 0) {
722 					syslog(LOG_ERR,
723 					"can't set v6-only binding for tcp6 "
724 					    "socket: %m");
725 					nfsd_exit(1);
726 				}
727 				if (bind(tcp6sock, ai_tcp6->ai_addr,
728 				    ai_tcp6->ai_addrlen) < 0) {
729 					syslog(LOG_ERR,
730 					    "can't bind tcp6 addr %s: %m",
731 					    bindhost[i]);
732 					nfsd_exit(1);
733 				}
734 				if (listen(tcp6sock, -1) < 0) {
735 					syslog(LOG_ERR, "listen failed");
736 					nfsd_exit(1);
737 				}
738 				freeaddrinfo(ai_tcp6);
739 				FD_SET(tcp6sock, &sockbits);
740 				if (maxsock < tcp6sock)
741 					maxsock = tcp6sock;
742 				connect_type_cnt++;
743 			}
744 		}
745 		if (rpcbreg == 1) {
746 			memset(&hints, 0, sizeof hints);
747 			hints.ai_flags = AI_PASSIVE;
748 			hints.ai_family = AF_INET6;
749 			hints.ai_socktype = SOCK_STREAM;
750 			hints.ai_protocol = IPPROTO_TCP;
751 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_tcp6);
752 			if (ecode != 0) {
753 				syslog(LOG_ERR, "getaddrinfo tcp6: %s",
754 				   gai_strerror(ecode));
755 				nfsd_exit(1);
756 			}
757 			nconf_tcp6 = getnetconfigent("tcp6");
758 			if (nconf_tcp6 == NULL)
759 				err(1, "getnetconfigent tcp6 failed");
760 			nb_tcp6.buf = ai_tcp6->ai_addr;
761 			nb_tcp6.len = nb_tcp6.maxlen = ai_tcp6->ai_addrlen;
762 			if (nfs_minvers == NFS_VER2)
763 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp6,
764 				    &nb_tcp6))
765 					err(1, "rpcb_set tcp6 failed");
766 			if (nfs_minvers <= NFS_VER3)
767 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp6,
768 				    &nb_tcp6))
769 					err(1, "rpcb_set tcp6 failed");
770 			freeaddrinfo(ai_tcp6);
771 		}
772 	}
773 
774 	if (rpcbregcnt == 0) {
775 		syslog(LOG_ERR, "rpcb_set() failed, nothing to do: %m");
776 		nfsd_exit(1);
777 	}
778 
779 	if (tcpflag && connect_type_cnt == 0) {
780 		syslog(LOG_ERR, "tcp connects == 0, nothing to do: %m");
781 		nfsd_exit(1);
782 	}
783 
784 	setproctitle("master");
785 	/*
786 	 * We always want a master to have a clean way to shut nfsd down
787 	 * (with unregistration): if the master is killed, it unregisters and
788 	 * kills all children. If we run for UDP only (and so do not have to
789 	 * loop waiting for accept), we instead make the parent
790 	 * a "server" too. start_server will not return.
791 	 */
792 	if (!tcpflag)
793 		start_server(1, &nfsdargs);
794 
795 	/*
796 	 * Loop forever accepting connections and passing the sockets
797 	 * into the kernel for the mounts.
798 	 */
799 	for (;;) {
800 		ready = sockbits;
801 		if (connect_type_cnt > 1) {
802 			if (select(maxsock + 1,
803 			    &ready, NULL, NULL, NULL) < 1) {
804 				error = errno;
805 				if (error == EINTR)
806 					continue;
807 				syslog(LOG_ERR, "select failed: %m");
808 				nfsd_exit(1);
809 			}
810 		}
811 		for (tcpsock = 0; tcpsock <= maxsock; tcpsock++) {
812 			if (FD_ISSET(tcpsock, &ready)) {
813 				len = sizeof(peer);
814 				if ((msgsock = accept(tcpsock,
815 				    (struct sockaddr *)&peer, &len)) < 0) {
816 					error = errno;
817 					syslog(LOG_ERR, "accept failed: %m");
818 					if (error == ECONNABORTED ||
819 					    error == EINTR)
820 						continue;
821 					nfsd_exit(1);
822 				}
823 				if (setsockopt(msgsock, SOL_SOCKET,
824 				    SO_KEEPALIVE, (char *)&on, sizeof(on)) < 0)
825 					syslog(LOG_ERR,
826 					    "setsockopt SO_KEEPALIVE: %m");
827 				addsockargs.sock = msgsock;
828 				addsockargs.name = (caddr_t)&peer;
829 				addsockargs.namelen = len;
830 				nfssvc(nfssvc_addsock, &addsockargs);
831 				(void)close(msgsock);
832 			}
833 		}
834 	}
835 }
836 
837 static int
838 setbindhost(struct addrinfo **ai, const char *bindhost, struct addrinfo hints)
839 {
840 	int ecode;
841 	u_int32_t host_addr[4];  /* IPv4 or IPv6 */
842 	const char *hostptr;
843 
844 	if (bindhost == NULL || strcmp("*", bindhost) == 0)
845 		hostptr = NULL;
846 	else
847 		hostptr = bindhost;
848 
849 	if (hostptr != NULL) {
850 		switch (hints.ai_family) {
851 		case AF_INET:
852 			if (inet_pton(AF_INET, hostptr, host_addr) == 1) {
853 				hints.ai_flags = AI_NUMERICHOST;
854 			} else {
855 				if (inet_pton(AF_INET6, hostptr,
856 				    host_addr) == 1)
857 					return (1);
858 			}
859 			break;
860 		case AF_INET6:
861 			if (inet_pton(AF_INET6, hostptr, host_addr) == 1) {
862 				hints.ai_flags = AI_NUMERICHOST;
863 			} else {
864 				if (inet_pton(AF_INET, hostptr,
865 				    host_addr) == 1)
866 					return (1);
867 			}
868 			break;
869 		default:
870 			break;
871 		}
872 	}
873 
874 	ecode = getaddrinfo(hostptr, "nfs", &hints, ai);
875 	if (ecode != 0) {
876 		syslog(LOG_ERR, "getaddrinfo %s: %s", bindhost,
877 		    gai_strerror(ecode));
878 		return (1);
879 	}
880 	return (0);
881 }
882 
883 static void
884 set_nfsdcnt(int proposed)
885 {
886 
887 	if (proposed < 1) {
888 		warnx("nfsd count too low %d; reset to %d", proposed,
889 		    DEFNFSDCNT);
890 		nfsdcnt = DEFNFSDCNT;
891 	} else if (proposed > MAXNFSDCNT) {
892 		warnx("nfsd count too high %d; truncated to %d", proposed,
893 		    MAXNFSDCNT);
894 		nfsdcnt = MAXNFSDCNT;
895 	} else
896 		nfsdcnt = proposed;
897 	nfsdcnt_set = 1;
898 }
899 
900 static void
901 usage(void)
902 {
903 	(void)fprintf(stderr, "%s", getopt_usage);
904 	exit(1);
905 }
906 
907 static void
908 nonfs(__unused int signo)
909 {
910 	syslog(LOG_ERR, "missing system call: NFS not available");
911 }
912 
913 static void
914 reapchild(__unused int signo)
915 {
916 	pid_t pid;
917 	int i;
918 
919 	while ((pid = wait3(NULL, WNOHANG, NULL)) > 0) {
920 		for (i = 0; i < nfsdcnt; i++)
921 			if (pid == children[i])
922 				children[i] = -1;
923 	}
924 }
925 
926 static void
927 unregistration(void)
928 {
929 	if ((!rpcb_unset(NFS_PROGRAM, 2, NULL)) ||
930 	    (!rpcb_unset(NFS_PROGRAM, 3, NULL)))
931 		syslog(LOG_ERR, "rpcb_unset failed");
932 }
933 
934 static void
935 killchildren(void)
936 {
937 	int i;
938 
939 	for (i = 0; i < nfsdcnt; i++) {
940 		if (children[i] > 0)
941 			kill(children[i], SIGKILL);
942 	}
943 }
944 
945 /*
946  * Cleanup master after SIGUSR1.
947  */
948 static void
949 cleanup(__unused int signo)
950 {
951 	nfsd_exit(0);
952 }
953 
954 /*
955  * Cleanup child after SIGUSR1.
956  */
957 static void
958 child_cleanup(__unused int signo)
959 {
960 	exit(0);
961 }
962 
963 static void
964 nfsd_exit(int status)
965 {
966 	killchildren();
967 	unregistration();
968 	exit(status);
969 }
970 
971 static int
972 get_tuned_nfsdcount(void)
973 {
974 	int ncpu, error, tuned_nfsdcnt;
975 	size_t ncpu_size;
976 
977 	ncpu_size = sizeof(ncpu);
978 	error = sysctlbyname("hw.ncpu", &ncpu, &ncpu_size, NULL, 0);
979 	if (error) {
980 		warnx("sysctlbyname(hw.ncpu) failed defaulting to %d nfs servers",
981 		    DEFNFSDCNT);
982 		tuned_nfsdcnt = DEFNFSDCNT;
983 	} else {
984 		tuned_nfsdcnt = ncpu * 8;
985 	}
986 	return tuned_nfsdcnt;
987 }
988 
989 static void
990 start_server(int master, struct nfsd_nfsd_args *nfsdargp)
991 {
992 	char principal[MAXHOSTNAMELEN + 5];
993 	int status, error;
994 	char hostname[MAXHOSTNAMELEN + 1], *cp;
995 	struct addrinfo *aip, hints;
996 
997 	status = 0;
998 	gethostname(hostname, sizeof (hostname));
999 	snprintf(principal, sizeof (principal), "nfs@%s", hostname);
1000 	if ((cp = strchr(hostname, '.')) == NULL ||
1001 	    *(cp + 1) == '\0') {
1002 		/* If not fully qualified, try getaddrinfo() */
1003 		memset((void *)&hints, 0, sizeof (hints));
1004 		hints.ai_flags = AI_CANONNAME;
1005 		error = getaddrinfo(hostname, NULL, &hints, &aip);
1006 		if (error == 0) {
1007 			if (aip->ai_canonname != NULL &&
1008 			    (cp = strchr(aip->ai_canonname, '.')) !=
1009 			    NULL && *(cp + 1) != '\0')
1010 				snprintf(principal, sizeof (principal),
1011 				    "nfs@%s", aip->ai_canonname);
1012 			freeaddrinfo(aip);
1013 		}
1014 	}
1015 	nfsdargp->principal = principal;
1016 
1017 	if (nfsdcnt_set)
1018 		nfsdargp->minthreads = nfsdargp->maxthreads = nfsdcnt;
1019 	else {
1020 		nfsdargp->minthreads = minthreads_set ? minthreads : get_tuned_nfsdcount();
1021 		nfsdargp->maxthreads = maxthreads_set ? maxthreads : nfsdargp->minthreads;
1022 		if (nfsdargp->maxthreads < nfsdargp->minthreads)
1023 			nfsdargp->maxthreads = nfsdargp->minthreads;
1024 	}
1025 	error = nfssvc(nfssvc_nfsd, nfsdargp);
1026 	if (error < 0 && errno == EAUTH) {
1027 		/*
1028 		 * This indicates that it could not register the
1029 		 * rpcsec_gss credentials, usually because the
1030 		 * gssd daemon isn't running.
1031 		 * (only the experimental server with nfsv4)
1032 		 */
1033 		syslog(LOG_ERR, "No gssd, using AUTH_SYS only");
1034 		principal[0] = '\0';
1035 		error = nfssvc(nfssvc_nfsd, nfsdargp);
1036 	}
1037 	if (error < 0) {
1038 		if (errno == ENXIO) {
1039 			syslog(LOG_ERR, "Bad -p option, cannot run");
1040 			if (masterpid != 0 && master == 0)
1041 				kill(masterpid, SIGUSR1);
1042 		} else
1043 			syslog(LOG_ERR, "nfssvc: %m");
1044 		status = 1;
1045 	}
1046 	if (master)
1047 		nfsd_exit(status);
1048 	else
1049 		exit(status);
1050 }
1051 
1052 /*
1053  * Open the stable restart file and return the file descriptor for it.
1054  */
1055 static void
1056 open_stable(int *stable_fdp, int *backup_fdp)
1057 {
1058 	int stable_fd, backup_fd = -1, ret;
1059 	struct stat st, backup_st;
1060 
1061 	/* Open and stat the stable restart file. */
1062 	stable_fd = open(NFSD_STABLERESTART, O_RDWR, 0);
1063 	if (stable_fd < 0)
1064 		stable_fd = open(NFSD_STABLERESTART, O_RDWR | O_CREAT, 0600);
1065 	if (stable_fd >= 0) {
1066 		ret = fstat(stable_fd, &st);
1067 		if (ret < 0) {
1068 			close(stable_fd);
1069 			stable_fd = -1;
1070 		}
1071 	}
1072 
1073 	/* Open and stat the backup stable restart file. */
1074 	if (stable_fd >= 0) {
1075 		backup_fd = open(NFSD_STABLEBACKUP, O_RDWR, 0);
1076 		if (backup_fd < 0)
1077 			backup_fd = open(NFSD_STABLEBACKUP, O_RDWR | O_CREAT,
1078 			    0600);
1079 		if (backup_fd >= 0) {
1080 			ret = fstat(backup_fd, &backup_st);
1081 			if (ret < 0) {
1082 				close(backup_fd);
1083 				backup_fd = -1;
1084 			}
1085 		}
1086 		if (backup_fd < 0) {
1087 			close(stable_fd);
1088 			stable_fd = -1;
1089 		}
1090 	}
1091 
1092 	*stable_fdp = stable_fd;
1093 	*backup_fdp = backup_fd;
1094 	if (stable_fd < 0)
1095 		return;
1096 
1097 	/* Sync up the 2 files, as required. */
1098 	if (st.st_size > 0)
1099 		copy_stable(stable_fd, backup_fd);
1100 	else if (backup_st.st_size > 0)
1101 		copy_stable(backup_fd, stable_fd);
1102 }
1103 
1104 /*
1105  * Copy the stable restart file to the backup or vice versa.
1106  */
1107 static void
1108 copy_stable(int from_fd, int to_fd)
1109 {
1110 	int cnt, ret;
1111 	static char buf[1024];
1112 
1113 	ret = lseek(from_fd, (off_t)0, SEEK_SET);
1114 	if (ret >= 0)
1115 		ret = lseek(to_fd, (off_t)0, SEEK_SET);
1116 	if (ret >= 0)
1117 		ret = ftruncate(to_fd, (off_t)0);
1118 	if (ret >= 0)
1119 		do {
1120 			cnt = read(from_fd, buf, 1024);
1121 			if (cnt > 0)
1122 				ret = write(to_fd, buf, cnt);
1123 			else if (cnt < 0)
1124 				ret = cnt;
1125 		} while (cnt > 0 && ret >= 0);
1126 	if (ret >= 0)
1127 		ret = fsync(to_fd);
1128 	if (ret < 0)
1129 		syslog(LOG_ERR, "stable restart copy failure: %m");
1130 }
1131 
1132 /*
1133  * Back up the stable restart file when indicated by the kernel.
1134  */
1135 static void
1136 backup_stable(__unused int signo)
1137 {
1138 
1139 	if (stablefd >= 0)
1140 		copy_stable(stablefd, backupfd);
1141 }
1142 
1143 /*
1144  * Parse the pNFS string and extract the DS servers and ports numbers.
1145  */
1146 static void
1147 parse_dsserver(const char *optionarg, struct nfsd_nfsd_args *nfsdargp)
1148 {
1149 	char *cp, *cp2, *dsaddr, *dshost, *dspath, *dsvol, nfsprt[9];
1150 	char *mdspath, *mdsp, ip6[INET6_ADDRSTRLEN];
1151 	const char *ad;
1152 	int ecode;
1153 	u_int adsiz, dsaddrcnt, dshostcnt, dspathcnt, hostsiz, pathsiz;
1154 	u_int mdspathcnt;
1155 	size_t dsaddrsiz, dshostsiz, dspathsiz, nfsprtsiz, mdspathsiz;
1156 	struct addrinfo hints, *ai_tcp, *res;
1157 	struct sockaddr_in sin;
1158 	struct sockaddr_in6 sin6;
1159 
1160 	cp = strdup(optionarg);
1161 	if (cp == NULL)
1162 		errx(1, "Out of memory");
1163 
1164 	/* Now, do the host names. */
1165 	dspathsiz = 1024;
1166 	dspathcnt = 0;
1167 	dspath = malloc(dspathsiz);
1168 	if (dspath == NULL)
1169 		errx(1, "Out of memory");
1170 	dshostsiz = 1024;
1171 	dshostcnt = 0;
1172 	dshost = malloc(dshostsiz);
1173 	if (dshost == NULL)
1174 		errx(1, "Out of memory");
1175 	dsaddrsiz = 1024;
1176 	dsaddrcnt = 0;
1177 	dsaddr = malloc(dsaddrsiz);
1178 	if (dsaddr == NULL)
1179 		errx(1, "Out of memory");
1180 	mdspathsiz = 1024;
1181 	mdspathcnt = 0;
1182 	mdspath = malloc(mdspathsiz);
1183 	if (mdspath == NULL)
1184 		errx(1, "Out of memory");
1185 
1186 	/* Put the NFS port# in "." form. */
1187 	snprintf(nfsprt, 9, ".%d.%d", 2049 >> 8, 2049 & 0xff);
1188 	nfsprtsiz = strlen(nfsprt);
1189 
1190 	ai_tcp = NULL;
1191 	/* Loop around for each DS server name. */
1192 	do {
1193 		cp2 = strchr(cp, ',');
1194 		if (cp2 != NULL) {
1195 			/* Not the last DS in the list. */
1196 			*cp2++ = '\0';
1197 			if (*cp2 == '\0')
1198 				usage();
1199 		}
1200 
1201 		dsvol = strchr(cp, ':');
1202 		if (dsvol == NULL || *(dsvol + 1) == '\0')
1203 			usage();
1204 		*dsvol++ = '\0';
1205 
1206 		/* Optional path for MDS file system to be stored on DS. */
1207 		mdsp = strchr(dsvol, '#');
1208 		if (mdsp != NULL) {
1209 			if (*(mdsp + 1) == '\0' || mdsp <= dsvol)
1210 				usage();
1211 			*mdsp++ = '\0';
1212 		}
1213 
1214 		/* Append this pathname to dspath. */
1215 		pathsiz = strlen(dsvol);
1216 		if (dspathcnt + pathsiz + 1 > dspathsiz) {
1217 			dspathsiz *= 2;
1218 			dspath = realloc(dspath, dspathsiz);
1219 			if (dspath == NULL)
1220 				errx(1, "Out of memory");
1221 		}
1222 		strcpy(&dspath[dspathcnt], dsvol);
1223 		dspathcnt += pathsiz + 1;
1224 
1225 		/* Append this pathname to mdspath. */
1226 		if (mdsp != NULL)
1227 			pathsiz = strlen(mdsp);
1228 		else
1229 			pathsiz = 0;
1230 		if (mdspathcnt + pathsiz + 1 > mdspathsiz) {
1231 			mdspathsiz *= 2;
1232 			mdspath = realloc(mdspath, mdspathsiz);
1233 			if (mdspath == NULL)
1234 				errx(1, "Out of memory");
1235 		}
1236 		if (mdsp != NULL)
1237 			strcpy(&mdspath[mdspathcnt], mdsp);
1238 		else
1239 			mdspath[mdspathcnt] = '\0';
1240 		mdspathcnt += pathsiz + 1;
1241 
1242 		if (ai_tcp != NULL)
1243 			freeaddrinfo(ai_tcp);
1244 
1245 		/* Get the fully qualified domain name and IP address. */
1246 		memset(&hints, 0, sizeof(hints));
1247 		hints.ai_flags = AI_CANONNAME | AI_ADDRCONFIG;
1248 		hints.ai_family = PF_UNSPEC;
1249 		hints.ai_socktype = SOCK_STREAM;
1250 		hints.ai_protocol = IPPROTO_TCP;
1251 		ecode = getaddrinfo(cp, NULL, &hints, &ai_tcp);
1252 		if (ecode != 0)
1253 			err(1, "getaddrinfo pnfs: %s %s", cp,
1254 			    gai_strerror(ecode));
1255 		ad = NULL;
1256 		for (res = ai_tcp; res != NULL; res = res->ai_next) {
1257 			if (res->ai_addr->sa_family == AF_INET) {
1258 				if (res->ai_addrlen < sizeof(sin))
1259 					err(1, "getaddrinfo() returned "
1260 					    "undersized IPv4 address");
1261 				/*
1262 				 * Mips cares about sockaddr_in alignment,
1263 				 * so copy the address.
1264 				 */
1265 				memcpy(&sin, res->ai_addr, sizeof(sin));
1266 				ad = inet_ntoa(sin.sin_addr);
1267 				break;
1268 			} else if (res->ai_family == AF_INET6) {
1269 				if (res->ai_addrlen < sizeof(sin6))
1270 					err(1, "getaddrinfo() returned "
1271 					    "undersized IPv6 address");
1272 				/*
1273 				 * Mips cares about sockaddr_in6 alignment,
1274 				 * so copy the address.
1275 				 */
1276 				memcpy(&sin6, res->ai_addr, sizeof(sin6));
1277 				ad = inet_ntop(AF_INET6, &sin6.sin6_addr, ip6,
1278 				    sizeof(ip6));
1279 
1280 				/*
1281 				 * XXX
1282 				 * Since a link local address will only
1283 				 * work if the client and DS are in the
1284 				 * same scope zone, only use it if it is
1285 				 * the only address.
1286 				 */
1287 				if (ad != NULL &&
1288 				    !IN6_IS_ADDR_LINKLOCAL(&sin6.sin6_addr))
1289 					break;
1290 			}
1291 		}
1292 		if (ad == NULL)
1293 			err(1, "No IP address for %s", cp);
1294 
1295 		/* Append this address to dsaddr. */
1296 		adsiz = strlen(ad);
1297 		if (dsaddrcnt + adsiz + nfsprtsiz + 1 > dsaddrsiz) {
1298 			dsaddrsiz *= 2;
1299 			dsaddr = realloc(dsaddr, dsaddrsiz);
1300 			if (dsaddr == NULL)
1301 				errx(1, "Out of memory");
1302 		}
1303 		strcpy(&dsaddr[dsaddrcnt], ad);
1304 		strcat(&dsaddr[dsaddrcnt], nfsprt);
1305 		dsaddrcnt += adsiz + nfsprtsiz + 1;
1306 
1307 		/* Append this hostname to dshost. */
1308 		hostsiz = strlen(ai_tcp->ai_canonname);
1309 		if (dshostcnt + hostsiz + 1 > dshostsiz) {
1310 			dshostsiz *= 2;
1311 			dshost = realloc(dshost, dshostsiz);
1312 			if (dshost == NULL)
1313 				errx(1, "Out of memory");
1314 		}
1315 		strcpy(&dshost[dshostcnt], ai_tcp->ai_canonname);
1316 		dshostcnt += hostsiz + 1;
1317 
1318 		cp = cp2;
1319 	} while (cp != NULL);
1320 
1321 	nfsdargp->addr = dsaddr;
1322 	nfsdargp->addrlen = dsaddrcnt;
1323 	nfsdargp->dnshost = dshost;
1324 	nfsdargp->dnshostlen = dshostcnt;
1325 	nfsdargp->dspath = dspath;
1326 	nfsdargp->dspathlen = dspathcnt;
1327 	nfsdargp->mdspath = mdspath;
1328 	nfsdargp->mdspathlen = mdspathcnt;
1329 	freeaddrinfo(ai_tcp);
1330 }
1331 
1332