xref: /freebsd/usr.sbin/nfsd/nfsd.c (revision 5e3190f700637fcfc1a52daeaa4a031fdd2557c7)
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 #endif /* not lint */
46 
47 #include <sys/param.h>
48 #include <sys/syslog.h>
49 #include <sys/wait.h>
50 #include <sys/mount.h>
51 #include <sys/fcntl.h>
52 #include <sys/linker.h>
53 #include <sys/module.h>
54 #include <sys/types.h>
55 #include <sys/stat.h>
56 #include <sys/sysctl.h>
57 #include <sys/ucred.h>
58 
59 #include <rpc/rpc.h>
60 #include <rpc/pmap_clnt.h>
61 #include <rpcsvc/nfs_prot.h>
62 
63 #include <netdb.h>
64 #include <arpa/inet.h>
65 #include <nfs/nfssvc.h>
66 
67 #include <fs/nfs/nfsproto.h>
68 #include <fs/nfs/nfskpiport.h>
69 #include <fs/nfs/nfs.h>
70 
71 #include <err.h>
72 #include <errno.h>
73 #include <signal.h>
74 #include <stdio.h>
75 #include <stdlib.h>
76 #include <string.h>
77 #include <unistd.h>
78 #include <sysexits.h>
79 
80 #include <getopt.h>
81 
82 static int	debug = 0;
83 
84 #define	NFSD_STABLERESTART	"/var/db/nfs-stablerestart"
85 #define	NFSD_STABLEBACKUP	"/var/db/nfs-stablerestart.bak"
86 #define	MAXNFSDCNT	256
87 #define	DEFNFSDCNT	 4
88 #define	NFS_VER2	 2
89 #define NFS_VER3	 3
90 #define NFS_VER4	 4
91 static pid_t children[MAXNFSDCNT]; /* PIDs of children */
92 static pid_t masterpid;		   /* PID of master/parent */
93 static int nfsdcnt;		/* number of children */
94 static int nfsdcnt_set;
95 static int minthreads;
96 static int maxthreads;
97 static int nfssvc_nfsd;		/* Set to correct NFSSVC_xxx flag */
98 static int stablefd = -1;	/* Fd for the stable restart file */
99 static int backupfd;		/* Fd for the backup stable restart file */
100 static const char *getopt_shortopts;
101 static const char *getopt_usage;
102 static int nfs_minvers = NFS_VER2;
103 
104 static int minthreads_set;
105 static int maxthreads_set;
106 
107 static struct option longopts[] = {
108 	{ "debug", no_argument, &debug, 1 },
109 	{ "minthreads", required_argument, &minthreads_set, 1 },
110 	{ "maxthreads", required_argument, &maxthreads_set, 1 },
111 	{ "pnfs", required_argument, NULL, 'p' },
112 	{ "mirror", required_argument, NULL, 'm' },
113 	{ NULL, 0, NULL, 0}
114 };
115 
116 static void	cleanup(int);
117 static void	child_cleanup(int);
118 static void	killchildren(void);
119 static void	nfsd_exit(int);
120 static void	nonfs(int);
121 static void	reapchild(int);
122 static int	setbindhost(struct addrinfo **ia, const char *bindhost,
123 		    struct addrinfo hints);
124 static void	start_server(int, struct nfsd_nfsd_args *, const char *vhost);
125 static void	unregistration(void);
126 static void	usage(void);
127 static void	open_stable(int *, int *);
128 static void	copy_stable(int, int);
129 static void	backup_stable(int);
130 static void	set_nfsdcnt(int);
131 static void	parse_dsserver(const char *, struct nfsd_nfsd_args *);
132 
133 /*
134  * Nfs server daemon mostly just a user context for nfssvc()
135  *
136  * 1 - do file descriptor and signal cleanup
137  * 2 - fork the nfsd(s)
138  * 3 - create server socket(s)
139  * 4 - register socket with rpcbind
140  *
141  * For connectionless protocols, just pass the socket into the kernel via.
142  * nfssvc().
143  * For connection based sockets, loop doing accepts. When you get a new
144  * socket from accept, pass the msgsock into the kernel via. nfssvc().
145  * The arguments are:
146  *	-r - reregister with rpcbind
147  *	-d - unregister with rpcbind
148  *	-t - support tcp nfs clients
149  *	-u - support udp nfs clients
150  *	-e - forces it to run a server that supports nfsv4
151  *	-p - enable a pNFS service
152  *	-m - set the mirroring level for a pNFS service
153  * followed by "n" which is the number of nfsds' to fork off
154  */
155 int
156 main(int argc, char **argv)
157 {
158 	struct nfsd_addsock_args addsockargs;
159 	struct addrinfo *ai_udp, *ai_tcp, *ai_udp6, *ai_tcp6, hints;
160 	struct netconfig *nconf_udp, *nconf_tcp, *nconf_udp6, *nconf_tcp6;
161 	struct netbuf nb_udp, nb_tcp, nb_udp6, nb_tcp6;
162 	struct sockaddr_storage peer;
163 	fd_set ready, sockbits;
164 	int ch, connect_type_cnt, i, maxsock, msgsock;
165 	socklen_t len;
166 	int on = 1, unregister, reregister, sock;
167 	int tcp6sock, ip6flag, tcpflag, tcpsock;
168 	int udpflag, ecode, error, s;
169 	int bindhostc, bindanyflag, rpcbreg, rpcbregcnt;
170 	int nfssvc_addsock;
171 	int jailed, longindex = 0;
172 	size_t jailed_size, nfs_minvers_size;
173 	const char *lopt;
174 	char **bindhost = NULL;
175 	pid_t pid;
176 	struct nfsd_nfsd_args nfsdargs;
177 	const char *vhostname = NULL;
178 
179 	nfsdargs.mirrorcnt = 1;
180 	nfsdargs.addr = NULL;
181 	nfsdargs.addrlen = 0;
182 	nfsdcnt = DEFNFSDCNT;
183 	unregister = reregister = tcpflag = maxsock = 0;
184 	bindanyflag = udpflag = connect_type_cnt = bindhostc = 0;
185 	getopt_shortopts = "ah:n:rdtuep:m:V:";
186 	getopt_usage =
187 	    "usage:\n"
188 	    "  nfsd [-ardtue] [-h bindip]\n"
189 	    "       [-n numservers] [--minthreads #] [--maxthreads #]\n"
190 	    "       [-p/--pnfs dsserver0:/dsserver0-mounted-on-dir,...,"
191 	    "dsserverN:/dsserverN-mounted-on-dir] [-m mirrorlevel]\n"
192 	    "       [-V virtual_hostname]\n";
193 	while ((ch = getopt_long(argc, argv, getopt_shortopts, longopts,
194 		    &longindex)) != -1)
195 		switch (ch) {
196 		case 'V':
197 			if (strlen(optarg) <= MAXHOSTNAMELEN)
198 				vhostname = optarg;
199 			else
200 				warnx("Virtual host name (%s) is too long",
201 				    optarg);
202 			break;
203 		case 'a':
204 			bindanyflag = 1;
205 			break;
206 		case 'n':
207 			set_nfsdcnt(atoi(optarg));
208 			break;
209 		case 'h':
210 			bindhostc++;
211 			bindhost = realloc(bindhost,sizeof(char *)*bindhostc);
212 			if (bindhost == NULL)
213 				errx(1, "Out of memory");
214 			bindhost[bindhostc-1] = strdup(optarg);
215 			if (bindhost[bindhostc-1] == NULL)
216 				errx(1, "Out of memory");
217 			break;
218 		case 'r':
219 			reregister = 1;
220 			break;
221 		case 'd':
222 			unregister = 1;
223 			break;
224 		case 't':
225 			tcpflag = 1;
226 			break;
227 		case 'u':
228 			udpflag = 1;
229 			break;
230 		case 'e':
231 			/* now a no-op, since this is the default */
232 			break;
233 		case 'p':
234 			/* Parse out the DS server host names and mount pts. */
235 			parse_dsserver(optarg, &nfsdargs);
236 			break;
237 		case 'm':
238 			/* Set the mirror level for a pNFS service. */
239 			i = atoi(optarg);
240 			if (i < 2 || i > NFSDEV_MAXMIRRORS)
241 				errx(1, "Mirror level out of range 2<-->%d",
242 				    NFSDEV_MAXMIRRORS);
243 			nfsdargs.mirrorcnt = i;
244 			break;
245 		case 0:
246 			lopt = longopts[longindex].name;
247 			if (!strcmp(lopt, "minthreads")) {
248 				minthreads = atoi(optarg);
249 			} else if (!strcmp(lopt, "maxthreads")) {
250 				maxthreads = atoi(optarg);
251 			}
252 			break;
253 		default:
254 		case '?':
255 			usage();
256 		}
257 	if (!tcpflag && !udpflag)
258 		udpflag = 1;
259 	argv += optind;
260 	argc -= optind;
261 	if (minthreads_set && maxthreads_set && minthreads > maxthreads)
262 		errx(EX_USAGE,
263 		    "error: minthreads(%d) can't be greater than "
264 		    "maxthreads(%d)", minthreads, maxthreads);
265 
266 	/*
267 	 * XXX
268 	 * Backward compatibility, trailing number is the count of daemons.
269 	 */
270 	if (argc > 1)
271 		usage();
272 	if (argc == 1)
273 		set_nfsdcnt(atoi(argv[0]));
274 
275 	/*
276 	 * Unless the "-o" option was specified, try and run "nfsd".
277 	 * If "-o" was specified, try and run "nfsserver".
278 	 */
279 	if (modfind("nfsd") < 0) {
280 		/* Not present in kernel, try loading it */
281 		if (kldload("nfsd") < 0 || modfind("nfsd") < 0)
282 			errx(1, "NFS server is not available");
283 	}
284 
285 	ip6flag = 1;
286 	s = socket(AF_INET6, SOCK_DGRAM, IPPROTO_UDP);
287 	if (s == -1) {
288 		if (errno != EPROTONOSUPPORT && errno != EAFNOSUPPORT)
289 			err(1, "socket");
290 		ip6flag = 0;
291 	} else if (getnetconfigent("udp6") == NULL ||
292 		getnetconfigent("tcp6") == NULL) {
293 		ip6flag = 0;
294 	}
295 	if (s != -1)
296 		close(s);
297 
298 	if (bindhostc == 0 || bindanyflag) {
299 		bindhostc++;
300 		bindhost = realloc(bindhost,sizeof(char *)*bindhostc);
301 		if (bindhost == NULL)
302 			errx(1, "Out of memory");
303 		bindhost[bindhostc-1] = strdup("*");
304 		if (bindhost[bindhostc-1] == NULL)
305 			errx(1, "Out of memory");
306 	}
307 
308 	if (unregister) {
309 		/*
310 		 * Unregister before setting nfs_minvers, in case the
311 		 * value of vfs.nfsd.server_min_nfsvers has changed
312 		 * since registering with rpcbind.
313 		 */
314 		unregistration();
315 		exit (0);
316 	}
317 
318 	nfs_minvers_size = sizeof(nfs_minvers);
319 	error = sysctlbyname("vfs.nfsd.server_min_nfsvers", &nfs_minvers,
320 	    &nfs_minvers_size, NULL, 0);
321 	if (error != 0 || nfs_minvers < NFS_VER2 || nfs_minvers > NFS_VER4) {
322 		warnx("sysctlbyname(vfs.nfsd.server_min_nfsvers) failed,"
323 		    " defaulting to NFSv2");
324 		nfs_minvers = NFS_VER2;
325 	}
326 
327 	if (reregister) {
328 		if (udpflag) {
329 			memset(&hints, 0, sizeof hints);
330 			hints.ai_flags = AI_PASSIVE;
331 			hints.ai_family = AF_INET;
332 			hints.ai_socktype = SOCK_DGRAM;
333 			hints.ai_protocol = IPPROTO_UDP;
334 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp);
335 			if (ecode != 0)
336 				err(1, "getaddrinfo udp: %s", gai_strerror(ecode));
337 			nconf_udp = getnetconfigent("udp");
338 			if (nconf_udp == NULL)
339 				err(1, "getnetconfigent udp failed");
340 			nb_udp.buf = ai_udp->ai_addr;
341 			nb_udp.len = nb_udp.maxlen = ai_udp->ai_addrlen;
342 			if (nfs_minvers == NFS_VER2)
343 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp,
344 				    &nb_udp))
345 					err(1, "rpcb_set udp failed");
346 			if (nfs_minvers <= NFS_VER3)
347 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp,
348 				    &nb_udp))
349 					err(1, "rpcb_set udp failed");
350 			freeaddrinfo(ai_udp);
351 		}
352 		if (udpflag && ip6flag) {
353 			memset(&hints, 0, sizeof hints);
354 			hints.ai_flags = AI_PASSIVE;
355 			hints.ai_family = AF_INET6;
356 			hints.ai_socktype = SOCK_DGRAM;
357 			hints.ai_protocol = IPPROTO_UDP;
358 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp6);
359 			if (ecode != 0)
360 				err(1, "getaddrinfo udp6: %s", gai_strerror(ecode));
361 			nconf_udp6 = getnetconfigent("udp6");
362 			if (nconf_udp6 == NULL)
363 				err(1, "getnetconfigent udp6 failed");
364 			nb_udp6.buf = ai_udp6->ai_addr;
365 			nb_udp6.len = nb_udp6.maxlen = ai_udp6->ai_addrlen;
366 			if (nfs_minvers == NFS_VER2)
367 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp6,
368 				    &nb_udp6))
369 					err(1, "rpcb_set udp6 failed");
370 			if (nfs_minvers <= NFS_VER3)
371 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp6,
372 				    &nb_udp6))
373 					err(1, "rpcb_set udp6 failed");
374 			freeaddrinfo(ai_udp6);
375 		}
376 		if (tcpflag) {
377 			memset(&hints, 0, sizeof hints);
378 			hints.ai_flags = AI_PASSIVE;
379 			hints.ai_family = AF_INET;
380 			hints.ai_socktype = SOCK_STREAM;
381 			hints.ai_protocol = IPPROTO_TCP;
382 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_tcp);
383 			if (ecode != 0)
384 				err(1, "getaddrinfo tcp: %s", gai_strerror(ecode));
385 			nconf_tcp = getnetconfigent("tcp");
386 			if (nconf_tcp == NULL)
387 				err(1, "getnetconfigent tcp failed");
388 			nb_tcp.buf = ai_tcp->ai_addr;
389 			nb_tcp.len = nb_tcp.maxlen = ai_tcp->ai_addrlen;
390 			if (nfs_minvers == NFS_VER2)
391 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp,
392 				    &nb_tcp))
393 					err(1, "rpcb_set tcp failed");
394 			if (nfs_minvers <= NFS_VER3)
395 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp,
396 				    &nb_tcp))
397 					err(1, "rpcb_set tcp failed");
398 			freeaddrinfo(ai_tcp);
399 		}
400 		if (tcpflag && ip6flag) {
401 			memset(&hints, 0, sizeof hints);
402 			hints.ai_flags = AI_PASSIVE;
403 			hints.ai_family = AF_INET6;
404 			hints.ai_socktype = SOCK_STREAM;
405 			hints.ai_protocol = IPPROTO_TCP;
406 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_tcp6);
407 			if (ecode != 0)
408 				err(1, "getaddrinfo tcp6: %s", gai_strerror(ecode));
409 			nconf_tcp6 = getnetconfigent("tcp6");
410 			if (nconf_tcp6 == NULL)
411 				err(1, "getnetconfigent tcp6 failed");
412 			nb_tcp6.buf = ai_tcp6->ai_addr;
413 			nb_tcp6.len = nb_tcp6.maxlen = ai_tcp6->ai_addrlen;
414 			if (nfs_minvers == NFS_VER2)
415 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp6,
416 				    &nb_tcp6))
417 					err(1, "rpcb_set tcp6 failed");
418 			if (nfs_minvers <= NFS_VER3)
419 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp6,
420 				   &nb_tcp6))
421 					err(1, "rpcb_set tcp6 failed");
422 			freeaddrinfo(ai_tcp6);
423 		}
424 		exit (0);
425 	}
426 	if (debug == 0) {
427 		daemon(0, 0);
428 		(void)signal(SIGHUP, SIG_IGN);
429 		(void)signal(SIGINT, SIG_IGN);
430 		/*
431 		 * nfsd sits in the kernel most of the time.  It needs
432 		 * to ignore SIGTERM/SIGQUIT in order to stay alive as long
433 		 * as possible during a shutdown, otherwise loopback
434 		 * mounts will not be able to unmount.
435 		 */
436 		(void)signal(SIGTERM, SIG_IGN);
437 		(void)signal(SIGQUIT, SIG_IGN);
438 	}
439 	(void)signal(SIGSYS, nonfs);
440 	(void)signal(SIGCHLD, reapchild);
441 	(void)signal(SIGUSR2, backup_stable);
442 
443 	openlog("nfsd", LOG_PID | (debug ? LOG_PERROR : 0), LOG_DAEMON);
444 
445 	/*
446 	 * For V4, we open the stablerestart file and call nfssvc()
447 	 * to get it loaded. This is done before the daemons do the
448 	 * regular nfssvc() call to service NFS requests.
449 	 * (This way the file remains open until the last nfsd is killed
450 	 *  off.)
451 	 * It and the backup copy will be created as empty files
452 	 * the first time this nfsd is started and should never be
453 	 * deleted/replaced if at all possible. It should live on a
454 	 * local, non-volatile storage device that does not do hardware
455 	 * level write-back caching. (See SCSI doc for more information
456 	 * on how to prevent write-back caching on SCSI disks.)
457 	 */
458 	open_stable(&stablefd, &backupfd);
459 	if (stablefd < 0) {
460 		syslog(LOG_ERR, "Can't open %s: %m\n", NFSD_STABLERESTART);
461 		exit(1);
462 	}
463 	/* This system call will fail for old kernels, but that's ok. */
464 	nfssvc(NFSSVC_BACKUPSTABLE, NULL);
465 	if (nfssvc(NFSSVC_STABLERESTART, (caddr_t)&stablefd) < 0) {
466 		if (errno == EPERM) {
467 			jailed = 0;
468 			jailed_size = sizeof(jailed);
469 			sysctlbyname("security.jail.jailed", &jailed,
470 			    &jailed_size, NULL, 0);
471 			if (jailed != 0)
472 				syslog(LOG_ERR, "nfssvc stablerestart failed: "
473 				    "allow.nfsd might not be configured");
474 			else
475 				syslog(LOG_ERR, "nfssvc stablerestart failed");
476 		} else if (errno == ENXIO)
477 			syslog(LOG_ERR, "nfssvc stablerestart failed: is nfsd "
478 			    "already running?");
479 		else
480 			syslog(LOG_ERR, "Can't read stable storage file: %m\n");
481 		exit(1);
482 	}
483 	nfssvc_addsock = NFSSVC_NFSDADDSOCK;
484 	nfssvc_nfsd = NFSSVC_NFSDNFSD | NFSSVC_NEWSTRUCT;
485 
486 	if (tcpflag) {
487 		/*
488 		 * For TCP mode, we fork once to start the first
489 		 * kernel nfsd thread. The kernel will add more
490 		 * threads as needed.
491 		 */
492 		masterpid = getpid();
493 		pid = fork();
494 		if (pid == -1) {
495 			syslog(LOG_ERR, "fork: %m");
496 			nfsd_exit(1);
497 		}
498 		if (pid) {
499 			children[0] = pid;
500 		} else {
501 			(void)signal(SIGUSR1, child_cleanup);
502 			setproctitle("server");
503 			start_server(0, &nfsdargs, vhostname);
504 		}
505 	}
506 
507 	(void)signal(SIGUSR1, cleanup);
508 	FD_ZERO(&sockbits);
509 
510 	rpcbregcnt = 0;
511 	/* Set up the socket for udp and rpcb register it. */
512 	if (udpflag) {
513 		rpcbreg = 0;
514 		for (i = 0; i < bindhostc; i++) {
515 			memset(&hints, 0, sizeof hints);
516 			hints.ai_flags = AI_PASSIVE;
517 			hints.ai_family = AF_INET;
518 			hints.ai_socktype = SOCK_DGRAM;
519 			hints.ai_protocol = IPPROTO_UDP;
520 			if (setbindhost(&ai_udp, bindhost[i], hints) == 0) {
521 				rpcbreg = 1;
522 				rpcbregcnt++;
523 				if ((sock = socket(ai_udp->ai_family,
524 				    ai_udp->ai_socktype,
525 				    ai_udp->ai_protocol)) < 0) {
526 					syslog(LOG_ERR,
527 					    "can't create udp socket");
528 					nfsd_exit(1);
529 				}
530 				if (bind(sock, ai_udp->ai_addr,
531 				    ai_udp->ai_addrlen) < 0) {
532 					syslog(LOG_ERR,
533 					    "can't bind udp addr %s: %m",
534 					    bindhost[i]);
535 					nfsd_exit(1);
536 				}
537 				freeaddrinfo(ai_udp);
538 				addsockargs.sock = sock;
539 				addsockargs.name = NULL;
540 				addsockargs.namelen = 0;
541 				if (nfssvc(nfssvc_addsock, &addsockargs) < 0) {
542 					syslog(LOG_ERR, "can't Add UDP socket");
543 					nfsd_exit(1);
544 				}
545 				(void)close(sock);
546 			}
547 		}
548 		if (rpcbreg == 1) {
549 			memset(&hints, 0, sizeof hints);
550 			hints.ai_flags = AI_PASSIVE;
551 			hints.ai_family = AF_INET;
552 			hints.ai_socktype = SOCK_DGRAM;
553 			hints.ai_protocol = IPPROTO_UDP;
554 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp);
555 			if (ecode != 0) {
556 				syslog(LOG_ERR, "getaddrinfo udp: %s",
557 				   gai_strerror(ecode));
558 				nfsd_exit(1);
559 			}
560 			nconf_udp = getnetconfigent("udp");
561 			if (nconf_udp == NULL)
562 				err(1, "getnetconfigent udp failed");
563 			nb_udp.buf = ai_udp->ai_addr;
564 			nb_udp.len = nb_udp.maxlen = ai_udp->ai_addrlen;
565 			if (nfs_minvers == NFS_VER2)
566 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp,
567 				    &nb_udp))
568 					err(1, "rpcb_set udp failed");
569 			if (nfs_minvers <= NFS_VER3)
570 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp,
571 				    &nb_udp))
572 					err(1, "rpcb_set udp failed");
573 			freeaddrinfo(ai_udp);
574 		}
575 	}
576 
577 	/* Set up the socket for udp6 and rpcb register it. */
578 	if (udpflag && ip6flag) {
579 		rpcbreg = 0;
580 		for (i = 0; i < bindhostc; i++) {
581 			memset(&hints, 0, sizeof hints);
582 			hints.ai_flags = AI_PASSIVE;
583 			hints.ai_family = AF_INET6;
584 			hints.ai_socktype = SOCK_DGRAM;
585 			hints.ai_protocol = IPPROTO_UDP;
586 			if (setbindhost(&ai_udp6, bindhost[i], hints) == 0) {
587 				rpcbreg = 1;
588 				rpcbregcnt++;
589 				if ((sock = socket(ai_udp6->ai_family,
590 				    ai_udp6->ai_socktype,
591 				    ai_udp6->ai_protocol)) < 0) {
592 					syslog(LOG_ERR,
593 						"can't create udp6 socket");
594 					nfsd_exit(1);
595 				}
596 				if (setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY,
597 				    &on, sizeof on) < 0) {
598 					syslog(LOG_ERR,
599 					    "can't set v6-only binding for "
600 					    "udp6 socket: %m");
601 					nfsd_exit(1);
602 				}
603 				if (bind(sock, ai_udp6->ai_addr,
604 				    ai_udp6->ai_addrlen) < 0) {
605 					syslog(LOG_ERR,
606 					    "can't bind udp6 addr %s: %m",
607 					    bindhost[i]);
608 					nfsd_exit(1);
609 				}
610 				freeaddrinfo(ai_udp6);
611 				addsockargs.sock = sock;
612 				addsockargs.name = NULL;
613 				addsockargs.namelen = 0;
614 				if (nfssvc(nfssvc_addsock, &addsockargs) < 0) {
615 					syslog(LOG_ERR,
616 					    "can't add UDP6 socket");
617 					nfsd_exit(1);
618 				}
619 				(void)close(sock);
620 			}
621 		}
622 		if (rpcbreg == 1) {
623 			memset(&hints, 0, sizeof hints);
624 			hints.ai_flags = AI_PASSIVE;
625 			hints.ai_family = AF_INET6;
626 			hints.ai_socktype = SOCK_DGRAM;
627 			hints.ai_protocol = IPPROTO_UDP;
628 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp6);
629 			if (ecode != 0) {
630 				syslog(LOG_ERR, "getaddrinfo udp6: %s",
631 				   gai_strerror(ecode));
632 				nfsd_exit(1);
633 			}
634 			nconf_udp6 = getnetconfigent("udp6");
635 			if (nconf_udp6 == NULL)
636 				err(1, "getnetconfigent udp6 failed");
637 			nb_udp6.buf = ai_udp6->ai_addr;
638 			nb_udp6.len = nb_udp6.maxlen = ai_udp6->ai_addrlen;
639 			if (nfs_minvers == NFS_VER2)
640 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp6,
641 				    &nb_udp6))
642 					err(1,
643 					    "rpcb_set udp6 failed");
644 			if (nfs_minvers <= NFS_VER3)
645 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp6,
646 				    &nb_udp6))
647 					err(1,
648 					    "rpcb_set udp6 failed");
649 			freeaddrinfo(ai_udp6);
650 		}
651 	}
652 
653 	/* Set up the socket for tcp and rpcb register it. */
654 	if (tcpflag) {
655 		rpcbreg = 0;
656 		for (i = 0; i < bindhostc; i++) {
657 			memset(&hints, 0, sizeof hints);
658 			hints.ai_flags = AI_PASSIVE;
659 			hints.ai_family = AF_INET;
660 			hints.ai_socktype = SOCK_STREAM;
661 			hints.ai_protocol = IPPROTO_TCP;
662 			if (setbindhost(&ai_tcp, bindhost[i], hints) == 0) {
663 				rpcbreg = 1;
664 				rpcbregcnt++;
665 				if ((tcpsock = socket(AF_INET, SOCK_STREAM,
666 				    0)) < 0) {
667 					syslog(LOG_ERR,
668 					    "can't create tcp socket");
669 					nfsd_exit(1);
670 				}
671 				if (setsockopt(tcpsock, SOL_SOCKET,
672 				    SO_REUSEADDR,
673 				    (char *)&on, sizeof(on)) < 0)
674 					syslog(LOG_ERR,
675 					     "setsockopt SO_REUSEADDR: %m");
676 				if (bind(tcpsock, ai_tcp->ai_addr,
677 				    ai_tcp->ai_addrlen) < 0) {
678 					syslog(LOG_ERR,
679 					    "can't bind tcp addr %s: %m",
680 					    bindhost[i]);
681 					nfsd_exit(1);
682 				}
683 				if (listen(tcpsock, -1) < 0) {
684 					syslog(LOG_ERR, "listen failed");
685 					nfsd_exit(1);
686 				}
687 				freeaddrinfo(ai_tcp);
688 				FD_SET(tcpsock, &sockbits);
689 				maxsock = tcpsock;
690 				connect_type_cnt++;
691 			}
692 		}
693 		if (rpcbreg == 1) {
694 			memset(&hints, 0, sizeof hints);
695 			hints.ai_flags = AI_PASSIVE;
696 			hints.ai_family = AF_INET;
697 			hints.ai_socktype = SOCK_STREAM;
698 			hints.ai_protocol = IPPROTO_TCP;
699 			ecode = getaddrinfo(NULL, "nfs", &hints,
700 			     &ai_tcp);
701 			if (ecode != 0) {
702 				syslog(LOG_ERR, "getaddrinfo tcp: %s",
703 				   gai_strerror(ecode));
704 				nfsd_exit(1);
705 			}
706 			nconf_tcp = getnetconfigent("tcp");
707 			if (nconf_tcp == NULL)
708 				err(1, "getnetconfigent tcp failed");
709 			nb_tcp.buf = ai_tcp->ai_addr;
710 			nb_tcp.len = nb_tcp.maxlen = ai_tcp->ai_addrlen;
711 			if (nfs_minvers == NFS_VER2)
712 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp,
713 				    &nb_tcp))
714 					err(1, "rpcb_set tcp failed");
715 			if (nfs_minvers <= NFS_VER3)
716 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp,
717 				    &nb_tcp))
718 					err(1, "rpcb_set tcp failed");
719 			freeaddrinfo(ai_tcp);
720 		}
721 	}
722 
723 	/* Set up the socket for tcp6 and rpcb register it. */
724 	if (tcpflag && ip6flag) {
725 		rpcbreg = 0;
726 		for (i = 0; i < bindhostc; i++) {
727 			memset(&hints, 0, sizeof hints);
728 			hints.ai_flags = AI_PASSIVE;
729 			hints.ai_family = AF_INET6;
730 			hints.ai_socktype = SOCK_STREAM;
731 			hints.ai_protocol = IPPROTO_TCP;
732 			if (setbindhost(&ai_tcp6, bindhost[i], hints) == 0) {
733 				rpcbreg = 1;
734 				rpcbregcnt++;
735 				if ((tcp6sock = socket(ai_tcp6->ai_family,
736 				    ai_tcp6->ai_socktype,
737 				    ai_tcp6->ai_protocol)) < 0) {
738 					syslog(LOG_ERR,
739 					    "can't create tcp6 socket");
740 					nfsd_exit(1);
741 				}
742 				if (setsockopt(tcp6sock, SOL_SOCKET,
743 				    SO_REUSEADDR,
744 				    (char *)&on, sizeof(on)) < 0)
745 					syslog(LOG_ERR,
746 					    "setsockopt SO_REUSEADDR: %m");
747 				if (setsockopt(tcp6sock, IPPROTO_IPV6,
748 				    IPV6_V6ONLY, &on, sizeof on) < 0) {
749 					syslog(LOG_ERR,
750 					"can't set v6-only binding for tcp6 "
751 					    "socket: %m");
752 					nfsd_exit(1);
753 				}
754 				if (bind(tcp6sock, ai_tcp6->ai_addr,
755 				    ai_tcp6->ai_addrlen) < 0) {
756 					syslog(LOG_ERR,
757 					    "can't bind tcp6 addr %s: %m",
758 					    bindhost[i]);
759 					nfsd_exit(1);
760 				}
761 				if (listen(tcp6sock, -1) < 0) {
762 					syslog(LOG_ERR, "listen failed");
763 					nfsd_exit(1);
764 				}
765 				freeaddrinfo(ai_tcp6);
766 				FD_SET(tcp6sock, &sockbits);
767 				if (maxsock < tcp6sock)
768 					maxsock = tcp6sock;
769 				connect_type_cnt++;
770 			}
771 		}
772 		if (rpcbreg == 1) {
773 			memset(&hints, 0, sizeof hints);
774 			hints.ai_flags = AI_PASSIVE;
775 			hints.ai_family = AF_INET6;
776 			hints.ai_socktype = SOCK_STREAM;
777 			hints.ai_protocol = IPPROTO_TCP;
778 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_tcp6);
779 			if (ecode != 0) {
780 				syslog(LOG_ERR, "getaddrinfo tcp6: %s",
781 				   gai_strerror(ecode));
782 				nfsd_exit(1);
783 			}
784 			nconf_tcp6 = getnetconfigent("tcp6");
785 			if (nconf_tcp6 == NULL)
786 				err(1, "getnetconfigent tcp6 failed");
787 			nb_tcp6.buf = ai_tcp6->ai_addr;
788 			nb_tcp6.len = nb_tcp6.maxlen = ai_tcp6->ai_addrlen;
789 			if (nfs_minvers == NFS_VER2)
790 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp6,
791 				    &nb_tcp6))
792 					err(1, "rpcb_set tcp6 failed");
793 			if (nfs_minvers <= NFS_VER3)
794 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp6,
795 				    &nb_tcp6))
796 					err(1, "rpcb_set tcp6 failed");
797 			freeaddrinfo(ai_tcp6);
798 		}
799 	}
800 
801 	if (rpcbregcnt == 0) {
802 		syslog(LOG_ERR, "rpcb_set() failed, nothing to do: %m");
803 		nfsd_exit(1);
804 	}
805 
806 	if (tcpflag && connect_type_cnt == 0) {
807 		syslog(LOG_ERR, "tcp connects == 0, nothing to do: %m");
808 		nfsd_exit(1);
809 	}
810 
811 	setproctitle("master");
812 	/*
813 	 * We always want a master to have a clean way to shut nfsd down
814 	 * (with unregistration): if the master is killed, it unregisters and
815 	 * kills all children. If we run for UDP only (and so do not have to
816 	 * loop waiting for accept), we instead make the parent
817 	 * a "server" too. start_server will not return.
818 	 */
819 	if (!tcpflag)
820 		start_server(1, &nfsdargs, vhostname);
821 
822 	/*
823 	 * Loop forever accepting connections and passing the sockets
824 	 * into the kernel for the mounts.
825 	 */
826 	for (;;) {
827 		ready = sockbits;
828 		if (connect_type_cnt > 1) {
829 			if (select(maxsock + 1,
830 			    &ready, NULL, NULL, NULL) < 1) {
831 				error = errno;
832 				if (error == EINTR)
833 					continue;
834 				syslog(LOG_ERR, "select failed: %m");
835 				nfsd_exit(1);
836 			}
837 		}
838 		for (tcpsock = 0; tcpsock <= maxsock; tcpsock++) {
839 			if (FD_ISSET(tcpsock, &ready)) {
840 				len = sizeof(peer);
841 				if ((msgsock = accept(tcpsock,
842 				    (struct sockaddr *)&peer, &len)) < 0) {
843 					error = errno;
844 					syslog(LOG_ERR, "accept failed: %m");
845 					if (error == ECONNABORTED ||
846 					    error == EINTR)
847 						continue;
848 					nfsd_exit(1);
849 				}
850 				if (setsockopt(msgsock, SOL_SOCKET,
851 				    SO_KEEPALIVE, (char *)&on, sizeof(on)) < 0)
852 					syslog(LOG_ERR,
853 					    "setsockopt SO_KEEPALIVE: %m");
854 				addsockargs.sock = msgsock;
855 				addsockargs.name = (caddr_t)&peer;
856 				addsockargs.namelen = len;
857 				nfssvc(nfssvc_addsock, &addsockargs);
858 				(void)close(msgsock);
859 			}
860 		}
861 	}
862 }
863 
864 static int
865 setbindhost(struct addrinfo **ai, const char *bindhost, struct addrinfo hints)
866 {
867 	int ecode;
868 	u_int32_t host_addr[4];  /* IPv4 or IPv6 */
869 	const char *hostptr;
870 
871 	if (bindhost == NULL || strcmp("*", bindhost) == 0)
872 		hostptr = NULL;
873 	else
874 		hostptr = bindhost;
875 
876 	if (hostptr != NULL) {
877 		switch (hints.ai_family) {
878 		case AF_INET:
879 			if (inet_pton(AF_INET, hostptr, host_addr) == 1) {
880 				hints.ai_flags = AI_NUMERICHOST;
881 			} else {
882 				if (inet_pton(AF_INET6, hostptr,
883 				    host_addr) == 1)
884 					return (1);
885 			}
886 			break;
887 		case AF_INET6:
888 			if (inet_pton(AF_INET6, hostptr, host_addr) == 1) {
889 				hints.ai_flags = AI_NUMERICHOST;
890 			} else {
891 				if (inet_pton(AF_INET, hostptr,
892 				    host_addr) == 1)
893 					return (1);
894 			}
895 			break;
896 		default:
897 			break;
898 		}
899 	}
900 
901 	ecode = getaddrinfo(hostptr, "nfs", &hints, ai);
902 	if (ecode != 0) {
903 		syslog(LOG_ERR, "getaddrinfo %s: %s", bindhost,
904 		    gai_strerror(ecode));
905 		return (1);
906 	}
907 	return (0);
908 }
909 
910 static void
911 set_nfsdcnt(int proposed)
912 {
913 
914 	if (proposed < 1) {
915 		warnx("nfsd count too low %d; reset to %d", proposed,
916 		    DEFNFSDCNT);
917 		nfsdcnt = DEFNFSDCNT;
918 	} else if (proposed > MAXNFSDCNT) {
919 		warnx("nfsd count too high %d; truncated to %d", proposed,
920 		    MAXNFSDCNT);
921 		nfsdcnt = MAXNFSDCNT;
922 	} else
923 		nfsdcnt = proposed;
924 	nfsdcnt_set = 1;
925 }
926 
927 static void
928 usage(void)
929 {
930 	(void)fprintf(stderr, "%s", getopt_usage);
931 	exit(1);
932 }
933 
934 static void
935 nonfs(__unused int signo)
936 {
937 	syslog(LOG_ERR, "missing system call: NFS not available");
938 }
939 
940 static void
941 reapchild(__unused int signo)
942 {
943 	pid_t pid;
944 	int i;
945 
946 	while ((pid = wait3(NULL, WNOHANG, NULL)) > 0) {
947 		for (i = 0; i < nfsdcnt; i++)
948 			if (pid == children[i])
949 				children[i] = -1;
950 	}
951 }
952 
953 static void
954 unregistration(void)
955 {
956 	if ((nfs_minvers == NFS_VER2 && !rpcb_unset(NFS_PROGRAM, 2, NULL)) ||
957 	    (nfs_minvers <= NFS_VER3 && !rpcb_unset(NFS_PROGRAM, 3, NULL)))
958 		syslog(LOG_ERR, "rpcb_unset failed");
959 }
960 
961 static void
962 killchildren(void)
963 {
964 	int i;
965 
966 	for (i = 0; i < nfsdcnt; i++) {
967 		if (children[i] > 0)
968 			kill(children[i], SIGKILL);
969 	}
970 }
971 
972 /*
973  * Cleanup master after SIGUSR1.
974  */
975 static void
976 cleanup(__unused int signo)
977 {
978 	nfsd_exit(0);
979 }
980 
981 /*
982  * Cleanup child after SIGUSR1.
983  */
984 static void
985 child_cleanup(__unused int signo)
986 {
987 	exit(0);
988 }
989 
990 static void
991 nfsd_exit(int status)
992 {
993 	killchildren();
994 	unregistration();
995 	exit(status);
996 }
997 
998 static int
999 get_tuned_nfsdcount(void)
1000 {
1001 	int ncpu, error, tuned_nfsdcnt;
1002 	size_t ncpu_size;
1003 
1004 	ncpu_size = sizeof(ncpu);
1005 	error = sysctlbyname("hw.ncpu", &ncpu, &ncpu_size, NULL, 0);
1006 	if (error) {
1007 		warnx("sysctlbyname(hw.ncpu) failed defaulting to %d nfs servers",
1008 		    DEFNFSDCNT);
1009 		tuned_nfsdcnt = DEFNFSDCNT;
1010 	} else {
1011 		tuned_nfsdcnt = ncpu * 8;
1012 	}
1013 	return tuned_nfsdcnt;
1014 }
1015 
1016 static void
1017 start_server(int master, struct nfsd_nfsd_args *nfsdargp, const char *vhost)
1018 {
1019 	char principal[MAXHOSTNAMELEN + 5];
1020 	int status, error;
1021 	char hostname[MAXHOSTNAMELEN + 1], *cp;
1022 	struct addrinfo *aip, hints;
1023 
1024 	status = 0;
1025 	if (vhost == NULL)
1026 		gethostname(hostname, sizeof (hostname));
1027 	else
1028 		strlcpy(hostname, vhost, sizeof (hostname));
1029 	snprintf(principal, sizeof (principal), "nfs@%s", hostname);
1030 	if ((cp = strchr(hostname, '.')) == NULL ||
1031 	    *(cp + 1) == '\0') {
1032 		/* If not fully qualified, try getaddrinfo() */
1033 		memset((void *)&hints, 0, sizeof (hints));
1034 		hints.ai_flags = AI_CANONNAME;
1035 		error = getaddrinfo(hostname, NULL, &hints, &aip);
1036 		if (error == 0) {
1037 			if (aip->ai_canonname != NULL &&
1038 			    (cp = strchr(aip->ai_canonname, '.')) !=
1039 			    NULL && *(cp + 1) != '\0')
1040 				snprintf(principal, sizeof (principal),
1041 				    "nfs@%s", aip->ai_canonname);
1042 			freeaddrinfo(aip);
1043 		}
1044 	}
1045 	nfsdargp->principal = principal;
1046 
1047 	if (nfsdcnt_set)
1048 		nfsdargp->minthreads = nfsdargp->maxthreads = nfsdcnt;
1049 	else {
1050 		nfsdargp->minthreads = minthreads_set ? minthreads : get_tuned_nfsdcount();
1051 		nfsdargp->maxthreads = maxthreads_set ? maxthreads : nfsdargp->minthreads;
1052 		if (nfsdargp->maxthreads < nfsdargp->minthreads)
1053 			nfsdargp->maxthreads = nfsdargp->minthreads;
1054 	}
1055 	error = nfssvc(nfssvc_nfsd, nfsdargp);
1056 	if (error < 0 && errno == EAUTH) {
1057 		/*
1058 		 * This indicates that it could not register the
1059 		 * rpcsec_gss credentials, usually because the
1060 		 * gssd daemon isn't running.
1061 		 * (only the experimental server with nfsv4)
1062 		 */
1063 		syslog(LOG_ERR, "No gssd, using AUTH_SYS only");
1064 		principal[0] = '\0';
1065 		error = nfssvc(nfssvc_nfsd, nfsdargp);
1066 	}
1067 	if (error < 0) {
1068 		if (errno == ENXIO) {
1069 			syslog(LOG_ERR, "Bad -p option, cannot run");
1070 			if (masterpid != 0 && master == 0)
1071 				kill(masterpid, SIGUSR1);
1072 		} else
1073 			syslog(LOG_ERR, "nfssvc: %m");
1074 		status = 1;
1075 	}
1076 	if (master)
1077 		nfsd_exit(status);
1078 	else
1079 		exit(status);
1080 }
1081 
1082 /*
1083  * Open the stable restart file and return the file descriptor for it.
1084  */
1085 static void
1086 open_stable(int *stable_fdp, int *backup_fdp)
1087 {
1088 	int stable_fd, backup_fd = -1, ret;
1089 	struct stat st, backup_st;
1090 
1091 	/* Open and stat the stable restart file. */
1092 	stable_fd = open(NFSD_STABLERESTART, O_RDWR, 0);
1093 	if (stable_fd < 0)
1094 		stable_fd = open(NFSD_STABLERESTART, O_RDWR | O_CREAT, 0600);
1095 	if (stable_fd >= 0) {
1096 		ret = fstat(stable_fd, &st);
1097 		if (ret < 0) {
1098 			close(stable_fd);
1099 			stable_fd = -1;
1100 		}
1101 	}
1102 
1103 	/* Open and stat the backup stable restart file. */
1104 	if (stable_fd >= 0) {
1105 		backup_fd = open(NFSD_STABLEBACKUP, O_RDWR, 0);
1106 		if (backup_fd < 0)
1107 			backup_fd = open(NFSD_STABLEBACKUP, O_RDWR | O_CREAT,
1108 			    0600);
1109 		if (backup_fd >= 0) {
1110 			ret = fstat(backup_fd, &backup_st);
1111 			if (ret < 0) {
1112 				close(backup_fd);
1113 				backup_fd = -1;
1114 			}
1115 		}
1116 		if (backup_fd < 0) {
1117 			close(stable_fd);
1118 			stable_fd = -1;
1119 		}
1120 	}
1121 
1122 	*stable_fdp = stable_fd;
1123 	*backup_fdp = backup_fd;
1124 	if (stable_fd < 0)
1125 		return;
1126 
1127 	/* Sync up the 2 files, as required. */
1128 	if (st.st_size > 0)
1129 		copy_stable(stable_fd, backup_fd);
1130 	else if (backup_st.st_size > 0)
1131 		copy_stable(backup_fd, stable_fd);
1132 }
1133 
1134 /*
1135  * Copy the stable restart file to the backup or vice versa.
1136  */
1137 static void
1138 copy_stable(int from_fd, int to_fd)
1139 {
1140 	int cnt, ret;
1141 	static char buf[1024];
1142 
1143 	ret = lseek(from_fd, (off_t)0, SEEK_SET);
1144 	if (ret >= 0)
1145 		ret = lseek(to_fd, (off_t)0, SEEK_SET);
1146 	if (ret >= 0)
1147 		ret = ftruncate(to_fd, (off_t)0);
1148 	if (ret >= 0)
1149 		do {
1150 			cnt = read(from_fd, buf, 1024);
1151 			if (cnt > 0)
1152 				ret = write(to_fd, buf, cnt);
1153 			else if (cnt < 0)
1154 				ret = cnt;
1155 		} while (cnt > 0 && ret >= 0);
1156 	if (ret >= 0)
1157 		ret = fsync(to_fd);
1158 	if (ret < 0)
1159 		syslog(LOG_ERR, "stable restart copy failure: %m");
1160 }
1161 
1162 /*
1163  * Back up the stable restart file when indicated by the kernel.
1164  */
1165 static void
1166 backup_stable(__unused int signo)
1167 {
1168 
1169 	if (stablefd >= 0)
1170 		copy_stable(stablefd, backupfd);
1171 }
1172 
1173 /*
1174  * Parse the pNFS string and extract the DS servers and ports numbers.
1175  */
1176 static void
1177 parse_dsserver(const char *optionarg, struct nfsd_nfsd_args *nfsdargp)
1178 {
1179 	char *cp, *cp2, *dsaddr, *dshost, *dspath, *dsvol, nfsprt[9];
1180 	char *mdspath, *mdsp, ip6[INET6_ADDRSTRLEN];
1181 	const char *ad;
1182 	int ecode;
1183 	u_int adsiz, dsaddrcnt, dshostcnt, dspathcnt, hostsiz, pathsiz;
1184 	u_int mdspathcnt;
1185 	size_t dsaddrsiz, dshostsiz, dspathsiz, nfsprtsiz, mdspathsiz;
1186 	struct addrinfo hints, *ai_tcp, *res;
1187 	struct sockaddr_in sin;
1188 	struct sockaddr_in6 sin6;
1189 
1190 	cp = strdup(optionarg);
1191 	if (cp == NULL)
1192 		errx(1, "Out of memory");
1193 
1194 	/* Now, do the host names. */
1195 	dspathsiz = 1024;
1196 	dspathcnt = 0;
1197 	dspath = malloc(dspathsiz);
1198 	if (dspath == NULL)
1199 		errx(1, "Out of memory");
1200 	dshostsiz = 1024;
1201 	dshostcnt = 0;
1202 	dshost = malloc(dshostsiz);
1203 	if (dshost == NULL)
1204 		errx(1, "Out of memory");
1205 	dsaddrsiz = 1024;
1206 	dsaddrcnt = 0;
1207 	dsaddr = malloc(dsaddrsiz);
1208 	if (dsaddr == NULL)
1209 		errx(1, "Out of memory");
1210 	mdspathsiz = 1024;
1211 	mdspathcnt = 0;
1212 	mdspath = malloc(mdspathsiz);
1213 	if (mdspath == NULL)
1214 		errx(1, "Out of memory");
1215 
1216 	/* Put the NFS port# in "." form. */
1217 	snprintf(nfsprt, 9, ".%d.%d", 2049 >> 8, 2049 & 0xff);
1218 	nfsprtsiz = strlen(nfsprt);
1219 
1220 	ai_tcp = NULL;
1221 	/* Loop around for each DS server name. */
1222 	do {
1223 		cp2 = strchr(cp, ',');
1224 		if (cp2 != NULL) {
1225 			/* Not the last DS in the list. */
1226 			*cp2++ = '\0';
1227 			if (*cp2 == '\0')
1228 				usage();
1229 		}
1230 
1231 		dsvol = strchr(cp, ':');
1232 		if (dsvol == NULL || *(dsvol + 1) == '\0')
1233 			usage();
1234 		*dsvol++ = '\0';
1235 
1236 		/* Optional path for MDS file system to be stored on DS. */
1237 		mdsp = strchr(dsvol, '#');
1238 		if (mdsp != NULL) {
1239 			if (*(mdsp + 1) == '\0' || mdsp <= dsvol)
1240 				usage();
1241 			*mdsp++ = '\0';
1242 		}
1243 
1244 		/* Append this pathname to dspath. */
1245 		pathsiz = strlen(dsvol);
1246 		if (dspathcnt + pathsiz + 1 > dspathsiz) {
1247 			dspathsiz *= 2;
1248 			dspath = realloc(dspath, dspathsiz);
1249 			if (dspath == NULL)
1250 				errx(1, "Out of memory");
1251 		}
1252 		strcpy(&dspath[dspathcnt], dsvol);
1253 		dspathcnt += pathsiz + 1;
1254 
1255 		/* Append this pathname to mdspath. */
1256 		if (mdsp != NULL)
1257 			pathsiz = strlen(mdsp);
1258 		else
1259 			pathsiz = 0;
1260 		if (mdspathcnt + pathsiz + 1 > mdspathsiz) {
1261 			mdspathsiz *= 2;
1262 			mdspath = realloc(mdspath, mdspathsiz);
1263 			if (mdspath == NULL)
1264 				errx(1, "Out of memory");
1265 		}
1266 		if (mdsp != NULL)
1267 			strcpy(&mdspath[mdspathcnt], mdsp);
1268 		else
1269 			mdspath[mdspathcnt] = '\0';
1270 		mdspathcnt += pathsiz + 1;
1271 
1272 		if (ai_tcp != NULL)
1273 			freeaddrinfo(ai_tcp);
1274 
1275 		/* Get the fully qualified domain name and IP address. */
1276 		memset(&hints, 0, sizeof(hints));
1277 		hints.ai_flags = AI_CANONNAME | AI_ADDRCONFIG;
1278 		hints.ai_family = PF_UNSPEC;
1279 		hints.ai_socktype = SOCK_STREAM;
1280 		hints.ai_protocol = IPPROTO_TCP;
1281 		ecode = getaddrinfo(cp, NULL, &hints, &ai_tcp);
1282 		if (ecode != 0)
1283 			err(1, "getaddrinfo pnfs: %s %s", cp,
1284 			    gai_strerror(ecode));
1285 		ad = NULL;
1286 		for (res = ai_tcp; res != NULL; res = res->ai_next) {
1287 			if (res->ai_addr->sa_family == AF_INET) {
1288 				if (res->ai_addrlen < sizeof(sin))
1289 					err(1, "getaddrinfo() returned "
1290 					    "undersized IPv4 address");
1291 				/*
1292 				 * Mips cares about sockaddr_in alignment,
1293 				 * so copy the address.
1294 				 */
1295 				memcpy(&sin, res->ai_addr, sizeof(sin));
1296 				ad = inet_ntoa(sin.sin_addr);
1297 				break;
1298 			} else if (res->ai_family == AF_INET6) {
1299 				if (res->ai_addrlen < sizeof(sin6))
1300 					err(1, "getaddrinfo() returned "
1301 					    "undersized IPv6 address");
1302 				/*
1303 				 * Mips cares about sockaddr_in6 alignment,
1304 				 * so copy the address.
1305 				 */
1306 				memcpy(&sin6, res->ai_addr, sizeof(sin6));
1307 				ad = inet_ntop(AF_INET6, &sin6.sin6_addr, ip6,
1308 				    sizeof(ip6));
1309 
1310 				/*
1311 				 * XXX
1312 				 * Since a link local address will only
1313 				 * work if the client and DS are in the
1314 				 * same scope zone, only use it if it is
1315 				 * the only address.
1316 				 */
1317 				if (ad != NULL &&
1318 				    !IN6_IS_ADDR_LINKLOCAL(&sin6.sin6_addr))
1319 					break;
1320 			}
1321 		}
1322 		if (ad == NULL)
1323 			err(1, "No IP address for %s", cp);
1324 
1325 		/* Append this address to dsaddr. */
1326 		adsiz = strlen(ad);
1327 		if (dsaddrcnt + adsiz + nfsprtsiz + 1 > dsaddrsiz) {
1328 			dsaddrsiz *= 2;
1329 			dsaddr = realloc(dsaddr, dsaddrsiz);
1330 			if (dsaddr == NULL)
1331 				errx(1, "Out of memory");
1332 		}
1333 		strcpy(&dsaddr[dsaddrcnt], ad);
1334 		strcat(&dsaddr[dsaddrcnt], nfsprt);
1335 		dsaddrcnt += adsiz + nfsprtsiz + 1;
1336 
1337 		/* Append this hostname to dshost. */
1338 		hostsiz = strlen(ai_tcp->ai_canonname);
1339 		if (dshostcnt + hostsiz + 1 > dshostsiz) {
1340 			dshostsiz *= 2;
1341 			dshost = realloc(dshost, dshostsiz);
1342 			if (dshost == NULL)
1343 				errx(1, "Out of memory");
1344 		}
1345 		strcpy(&dshost[dshostcnt], ai_tcp->ai_canonname);
1346 		dshostcnt += hostsiz + 1;
1347 
1348 		cp = cp2;
1349 	} while (cp != NULL);
1350 
1351 	nfsdargp->addr = dsaddr;
1352 	nfsdargp->addrlen = dsaddrcnt;
1353 	nfsdargp->dnshost = dshost;
1354 	nfsdargp->dnshostlen = dshostcnt;
1355 	nfsdargp->dspath = dspath;
1356 	nfsdargp->dspathlen = dspathcnt;
1357 	nfsdargp->mdspath = mdspath;
1358 	nfsdargp->mdspathlen = mdspathcnt;
1359 	freeaddrinfo(ai_tcp);
1360 }
1361 
1362