xref: /freebsd/usr.sbin/nfsd/nfsd.c (revision 2b833162)
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 static int nfs_minvers = NFS_VER2;
105 
106 static int minthreads_set;
107 static int maxthreads_set;
108 
109 static struct option longopts[] = {
110 	{ "debug", no_argument, &debug, 1 },
111 	{ "minthreads", required_argument, &minthreads_set, 1 },
112 	{ "maxthreads", required_argument, &maxthreads_set, 1 },
113 	{ "pnfs", required_argument, NULL, 'p' },
114 	{ "mirror", required_argument, NULL, 'm' },
115 	{ NULL, 0, NULL, 0}
116 };
117 
118 static void	cleanup(int);
119 static void	child_cleanup(int);
120 static void	killchildren(void);
121 static void	nfsd_exit(int);
122 static void	nonfs(int);
123 static void	reapchild(int);
124 static int	setbindhost(struct addrinfo **ia, const char *bindhost,
125 		    struct addrinfo hints);
126 static void	start_server(int, struct nfsd_nfsd_args *, const char *vhost);
127 static void	unregistration(void);
128 static void	usage(void);
129 static void	open_stable(int *, int *);
130 static void	copy_stable(int, int);
131 static void	backup_stable(int);
132 static void	set_nfsdcnt(int);
133 static void	parse_dsserver(const char *, struct nfsd_nfsd_args *);
134 
135 /*
136  * Nfs server daemon mostly just a user context for nfssvc()
137  *
138  * 1 - do file descriptor and signal cleanup
139  * 2 - fork the nfsd(s)
140  * 3 - create server socket(s)
141  * 4 - register socket with rpcbind
142  *
143  * For connectionless protocols, just pass the socket into the kernel via.
144  * nfssvc().
145  * For connection based sockets, loop doing accepts. When you get a new
146  * socket from accept, pass the msgsock into the kernel via. nfssvc().
147  * The arguments are:
148  *	-r - reregister with rpcbind
149  *	-d - unregister with rpcbind
150  *	-t - support tcp nfs clients
151  *	-u - support udp nfs clients
152  *	-e - forces it to run a server that supports nfsv4
153  *	-p - enable a pNFS service
154  *	-m - set the mirroring level for a pNFS service
155  * followed by "n" which is the number of nfsds' to fork off
156  */
157 int
158 main(int argc, char **argv)
159 {
160 	struct nfsd_addsock_args addsockargs;
161 	struct addrinfo *ai_udp, *ai_tcp, *ai_udp6, *ai_tcp6, hints;
162 	struct netconfig *nconf_udp, *nconf_tcp, *nconf_udp6, *nconf_tcp6;
163 	struct netbuf nb_udp, nb_tcp, nb_udp6, nb_tcp6;
164 	struct sockaddr_storage peer;
165 	fd_set ready, sockbits;
166 	int ch, connect_type_cnt, i, maxsock, msgsock;
167 	socklen_t len;
168 	int on = 1, unregister, reregister, sock;
169 	int tcp6sock, ip6flag, tcpflag, tcpsock;
170 	int udpflag, ecode, error, s;
171 	int bindhostc, bindanyflag, rpcbreg, rpcbregcnt;
172 	int nfssvc_addsock;
173 	int jailed, longindex = 0;
174 	size_t jailed_size, nfs_minvers_size;
175 	const char *lopt;
176 	char **bindhost = NULL;
177 	pid_t pid;
178 	struct nfsd_nfsd_args nfsdargs;
179 	const char *vhostname = NULL;
180 
181 	nfsdargs.mirrorcnt = 1;
182 	nfsdargs.addr = NULL;
183 	nfsdargs.addrlen = 0;
184 	nfsdcnt = DEFNFSDCNT;
185 	unregister = reregister = tcpflag = maxsock = 0;
186 	bindanyflag = udpflag = connect_type_cnt = bindhostc = 0;
187 	getopt_shortopts = "ah:n:rdtuep:m:V:";
188 	getopt_usage =
189 	    "usage:\n"
190 	    "  nfsd [-ardtue] [-h bindip]\n"
191 	    "       [-n numservers] [--minthreads #] [--maxthreads #]\n"
192 	    "       [-p/--pnfs dsserver0:/dsserver0-mounted-on-dir,...,"
193 	    "dsserverN:/dsserverN-mounted-on-dir] [-m mirrorlevel]\n"
194 	    "       [-V virtual_hostname]\n";
195 	while ((ch = getopt_long(argc, argv, getopt_shortopts, longopts,
196 		    &longindex)) != -1)
197 		switch (ch) {
198 		case 'V':
199 			if (strlen(optarg) <= MAXHOSTNAMELEN)
200 				vhostname = optarg;
201 			else
202 				warnx("Virtual host name (%s) is too long",
203 				    optarg);
204 			break;
205 		case 'a':
206 			bindanyflag = 1;
207 			break;
208 		case 'n':
209 			set_nfsdcnt(atoi(optarg));
210 			break;
211 		case 'h':
212 			bindhostc++;
213 			bindhost = realloc(bindhost,sizeof(char *)*bindhostc);
214 			if (bindhost == NULL)
215 				errx(1, "Out of memory");
216 			bindhost[bindhostc-1] = strdup(optarg);
217 			if (bindhost[bindhostc-1] == NULL)
218 				errx(1, "Out of memory");
219 			break;
220 		case 'r':
221 			reregister = 1;
222 			break;
223 		case 'd':
224 			unregister = 1;
225 			break;
226 		case 't':
227 			tcpflag = 1;
228 			break;
229 		case 'u':
230 			udpflag = 1;
231 			break;
232 		case 'e':
233 			/* now a no-op, since this is the default */
234 			break;
235 		case 'p':
236 			/* Parse out the DS server host names and mount pts. */
237 			parse_dsserver(optarg, &nfsdargs);
238 			break;
239 		case 'm':
240 			/* Set the mirror level for a pNFS service. */
241 			i = atoi(optarg);
242 			if (i < 2 || i > NFSDEV_MAXMIRRORS)
243 				errx(1, "Mirror level out of range 2<-->%d",
244 				    NFSDEV_MAXMIRRORS);
245 			nfsdargs.mirrorcnt = i;
246 			break;
247 		case 0:
248 			lopt = longopts[longindex].name;
249 			if (!strcmp(lopt, "minthreads")) {
250 				minthreads = atoi(optarg);
251 			} else if (!strcmp(lopt, "maxthreads")) {
252 				maxthreads = atoi(optarg);
253 			}
254 			break;
255 		default:
256 		case '?':
257 			usage();
258 		}
259 	if (!tcpflag && !udpflag)
260 		udpflag = 1;
261 	argv += optind;
262 	argc -= optind;
263 	if (minthreads_set && maxthreads_set && minthreads > maxthreads)
264 		errx(EX_USAGE,
265 		    "error: minthreads(%d) can't be greater than "
266 		    "maxthreads(%d)", minthreads, maxthreads);
267 
268 	/*
269 	 * XXX
270 	 * Backward compatibility, trailing number is the count of daemons.
271 	 */
272 	if (argc > 1)
273 		usage();
274 	if (argc == 1)
275 		set_nfsdcnt(atoi(argv[0]));
276 
277 	/*
278 	 * Unless the "-o" option was specified, try and run "nfsd".
279 	 * If "-o" was specified, try and run "nfsserver".
280 	 */
281 	if (modfind("nfsd") < 0) {
282 		/* Not present in kernel, try loading it */
283 		if (kldload("nfsd") < 0 || modfind("nfsd") < 0)
284 			errx(1, "NFS server is not available");
285 	}
286 
287 	ip6flag = 1;
288 	s = socket(AF_INET6, SOCK_DGRAM, IPPROTO_UDP);
289 	if (s == -1) {
290 		if (errno != EPROTONOSUPPORT && errno != EAFNOSUPPORT)
291 			err(1, "socket");
292 		ip6flag = 0;
293 	} else if (getnetconfigent("udp6") == NULL ||
294 		getnetconfigent("tcp6") == NULL) {
295 		ip6flag = 0;
296 	}
297 	if (s != -1)
298 		close(s);
299 
300 	if (bindhostc == 0 || bindanyflag) {
301 		bindhostc++;
302 		bindhost = realloc(bindhost,sizeof(char *)*bindhostc);
303 		if (bindhost == NULL)
304 			errx(1, "Out of memory");
305 		bindhost[bindhostc-1] = strdup("*");
306 		if (bindhost[bindhostc-1] == NULL)
307 			errx(1, "Out of memory");
308 	}
309 
310 	if (unregister) {
311 		/*
312 		 * Unregister before setting nfs_minvers, in case the
313 		 * value of vfs.nfsd.server_min_nfsvers has changed
314 		 * since registering with rpcbind.
315 		 */
316 		unregistration();
317 		exit (0);
318 	}
319 
320 	nfs_minvers_size = sizeof(nfs_minvers);
321 	error = sysctlbyname("vfs.nfsd.server_min_nfsvers", &nfs_minvers,
322 	    &nfs_minvers_size, NULL, 0);
323 	if (error != 0 || nfs_minvers < NFS_VER2 || nfs_minvers > NFS_VER4) {
324 		warnx("sysctlbyname(vfs.nfsd.server_min_nfsvers) failed,"
325 		    " defaulting to NFSv2");
326 		nfs_minvers = NFS_VER2;
327 	}
328 
329 	if (reregister) {
330 		if (udpflag) {
331 			memset(&hints, 0, sizeof hints);
332 			hints.ai_flags = AI_PASSIVE;
333 			hints.ai_family = AF_INET;
334 			hints.ai_socktype = SOCK_DGRAM;
335 			hints.ai_protocol = IPPROTO_UDP;
336 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp);
337 			if (ecode != 0)
338 				err(1, "getaddrinfo udp: %s", gai_strerror(ecode));
339 			nconf_udp = getnetconfigent("udp");
340 			if (nconf_udp == NULL)
341 				err(1, "getnetconfigent udp failed");
342 			nb_udp.buf = ai_udp->ai_addr;
343 			nb_udp.len = nb_udp.maxlen = ai_udp->ai_addrlen;
344 			if (nfs_minvers == NFS_VER2)
345 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp,
346 				    &nb_udp))
347 					err(1, "rpcb_set udp failed");
348 			if (nfs_minvers <= NFS_VER3)
349 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp,
350 				    &nb_udp))
351 					err(1, "rpcb_set udp failed");
352 			freeaddrinfo(ai_udp);
353 		}
354 		if (udpflag && ip6flag) {
355 			memset(&hints, 0, sizeof hints);
356 			hints.ai_flags = AI_PASSIVE;
357 			hints.ai_family = AF_INET6;
358 			hints.ai_socktype = SOCK_DGRAM;
359 			hints.ai_protocol = IPPROTO_UDP;
360 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp6);
361 			if (ecode != 0)
362 				err(1, "getaddrinfo udp6: %s", gai_strerror(ecode));
363 			nconf_udp6 = getnetconfigent("udp6");
364 			if (nconf_udp6 == NULL)
365 				err(1, "getnetconfigent udp6 failed");
366 			nb_udp6.buf = ai_udp6->ai_addr;
367 			nb_udp6.len = nb_udp6.maxlen = ai_udp6->ai_addrlen;
368 			if (nfs_minvers == NFS_VER2)
369 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp6,
370 				    &nb_udp6))
371 					err(1, "rpcb_set udp6 failed");
372 			if (nfs_minvers <= NFS_VER3)
373 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp6,
374 				    &nb_udp6))
375 					err(1, "rpcb_set udp6 failed");
376 			freeaddrinfo(ai_udp6);
377 		}
378 		if (tcpflag) {
379 			memset(&hints, 0, sizeof hints);
380 			hints.ai_flags = AI_PASSIVE;
381 			hints.ai_family = AF_INET;
382 			hints.ai_socktype = SOCK_STREAM;
383 			hints.ai_protocol = IPPROTO_TCP;
384 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_tcp);
385 			if (ecode != 0)
386 				err(1, "getaddrinfo tcp: %s", gai_strerror(ecode));
387 			nconf_tcp = getnetconfigent("tcp");
388 			if (nconf_tcp == NULL)
389 				err(1, "getnetconfigent tcp failed");
390 			nb_tcp.buf = ai_tcp->ai_addr;
391 			nb_tcp.len = nb_tcp.maxlen = ai_tcp->ai_addrlen;
392 			if (nfs_minvers == NFS_VER2)
393 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp,
394 				    &nb_tcp))
395 					err(1, "rpcb_set tcp failed");
396 			if (nfs_minvers <= NFS_VER3)
397 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp,
398 				    &nb_tcp))
399 					err(1, "rpcb_set tcp failed");
400 			freeaddrinfo(ai_tcp);
401 		}
402 		if (tcpflag && ip6flag) {
403 			memset(&hints, 0, sizeof hints);
404 			hints.ai_flags = AI_PASSIVE;
405 			hints.ai_family = AF_INET6;
406 			hints.ai_socktype = SOCK_STREAM;
407 			hints.ai_protocol = IPPROTO_TCP;
408 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_tcp6);
409 			if (ecode != 0)
410 				err(1, "getaddrinfo tcp6: %s", gai_strerror(ecode));
411 			nconf_tcp6 = getnetconfigent("tcp6");
412 			if (nconf_tcp6 == NULL)
413 				err(1, "getnetconfigent tcp6 failed");
414 			nb_tcp6.buf = ai_tcp6->ai_addr;
415 			nb_tcp6.len = nb_tcp6.maxlen = ai_tcp6->ai_addrlen;
416 			if (nfs_minvers == NFS_VER2)
417 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp6,
418 				    &nb_tcp6))
419 					err(1, "rpcb_set tcp6 failed");
420 			if (nfs_minvers <= NFS_VER3)
421 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp6,
422 				   &nb_tcp6))
423 					err(1, "rpcb_set tcp6 failed");
424 			freeaddrinfo(ai_tcp6);
425 		}
426 		exit (0);
427 	}
428 	if (debug == 0) {
429 		daemon(0, 0);
430 		(void)signal(SIGHUP, SIG_IGN);
431 		(void)signal(SIGINT, SIG_IGN);
432 		/*
433 		 * nfsd sits in the kernel most of the time.  It needs
434 		 * to ignore SIGTERM/SIGQUIT in order to stay alive as long
435 		 * as possible during a shutdown, otherwise loopback
436 		 * mounts will not be able to unmount.
437 		 */
438 		(void)signal(SIGTERM, SIG_IGN);
439 		(void)signal(SIGQUIT, SIG_IGN);
440 	}
441 	(void)signal(SIGSYS, nonfs);
442 	(void)signal(SIGCHLD, reapchild);
443 	(void)signal(SIGUSR2, backup_stable);
444 
445 	openlog("nfsd", LOG_PID | (debug ? LOG_PERROR : 0), LOG_DAEMON);
446 
447 	/*
448 	 * For V4, we open the stablerestart file and call nfssvc()
449 	 * to get it loaded. This is done before the daemons do the
450 	 * regular nfssvc() call to service NFS requests.
451 	 * (This way the file remains open until the last nfsd is killed
452 	 *  off.)
453 	 * It and the backup copy will be created as empty files
454 	 * the first time this nfsd is started and should never be
455 	 * deleted/replaced if at all possible. It should live on a
456 	 * local, non-volatile storage device that does not do hardware
457 	 * level write-back caching. (See SCSI doc for more information
458 	 * on how to prevent write-back caching on SCSI disks.)
459 	 */
460 	open_stable(&stablefd, &backupfd);
461 	if (stablefd < 0) {
462 		syslog(LOG_ERR, "Can't open %s: %m\n", NFSD_STABLERESTART);
463 		exit(1);
464 	}
465 	/* This system call will fail for old kernels, but that's ok. */
466 	nfssvc(NFSSVC_BACKUPSTABLE, NULL);
467 	if (nfssvc(NFSSVC_STABLERESTART, (caddr_t)&stablefd) < 0) {
468 		if (errno == EPERM) {
469 			jailed = 0;
470 			jailed_size = sizeof(jailed);
471 			sysctlbyname("security.jail.jailed", &jailed,
472 			    &jailed_size, NULL, 0);
473 			if (jailed != 0)
474 				syslog(LOG_ERR, "nfssvc stablerestart failed: "
475 				    "allow.nfsd might not be configured");
476 			else
477 				syslog(LOG_ERR, "nfssvc stablerestart failed");
478 		} else if (errno == ENXIO)
479 			syslog(LOG_ERR, "nfssvc stablerestart failed: is nfsd "
480 			    "already running?");
481 		else
482 			syslog(LOG_ERR, "Can't read stable storage file: %m\n");
483 		exit(1);
484 	}
485 	nfssvc_addsock = NFSSVC_NFSDADDSOCK;
486 	nfssvc_nfsd = NFSSVC_NFSDNFSD | NFSSVC_NEWSTRUCT;
487 
488 	if (tcpflag) {
489 		/*
490 		 * For TCP mode, we fork once to start the first
491 		 * kernel nfsd thread. The kernel will add more
492 		 * threads as needed.
493 		 */
494 		masterpid = getpid();
495 		pid = fork();
496 		if (pid == -1) {
497 			syslog(LOG_ERR, "fork: %m");
498 			nfsd_exit(1);
499 		}
500 		if (pid) {
501 			children[0] = pid;
502 		} else {
503 			(void)signal(SIGUSR1, child_cleanup);
504 			setproctitle("server");
505 			start_server(0, &nfsdargs, vhostname);
506 		}
507 	}
508 
509 	(void)signal(SIGUSR1, cleanup);
510 	FD_ZERO(&sockbits);
511 
512 	rpcbregcnt = 0;
513 	/* Set up the socket for udp and rpcb register it. */
514 	if (udpflag) {
515 		rpcbreg = 0;
516 		for (i = 0; i < bindhostc; i++) {
517 			memset(&hints, 0, sizeof hints);
518 			hints.ai_flags = AI_PASSIVE;
519 			hints.ai_family = AF_INET;
520 			hints.ai_socktype = SOCK_DGRAM;
521 			hints.ai_protocol = IPPROTO_UDP;
522 			if (setbindhost(&ai_udp, bindhost[i], hints) == 0) {
523 				rpcbreg = 1;
524 				rpcbregcnt++;
525 				if ((sock = socket(ai_udp->ai_family,
526 				    ai_udp->ai_socktype,
527 				    ai_udp->ai_protocol)) < 0) {
528 					syslog(LOG_ERR,
529 					    "can't create udp socket");
530 					nfsd_exit(1);
531 				}
532 				if (bind(sock, ai_udp->ai_addr,
533 				    ai_udp->ai_addrlen) < 0) {
534 					syslog(LOG_ERR,
535 					    "can't bind udp addr %s: %m",
536 					    bindhost[i]);
537 					nfsd_exit(1);
538 				}
539 				freeaddrinfo(ai_udp);
540 				addsockargs.sock = sock;
541 				addsockargs.name = NULL;
542 				addsockargs.namelen = 0;
543 				if (nfssvc(nfssvc_addsock, &addsockargs) < 0) {
544 					syslog(LOG_ERR, "can't Add UDP socket");
545 					nfsd_exit(1);
546 				}
547 				(void)close(sock);
548 			}
549 		}
550 		if (rpcbreg == 1) {
551 			memset(&hints, 0, sizeof hints);
552 			hints.ai_flags = AI_PASSIVE;
553 			hints.ai_family = AF_INET;
554 			hints.ai_socktype = SOCK_DGRAM;
555 			hints.ai_protocol = IPPROTO_UDP;
556 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp);
557 			if (ecode != 0) {
558 				syslog(LOG_ERR, "getaddrinfo udp: %s",
559 				   gai_strerror(ecode));
560 				nfsd_exit(1);
561 			}
562 			nconf_udp = getnetconfigent("udp");
563 			if (nconf_udp == NULL)
564 				err(1, "getnetconfigent udp failed");
565 			nb_udp.buf = ai_udp->ai_addr;
566 			nb_udp.len = nb_udp.maxlen = ai_udp->ai_addrlen;
567 			if (nfs_minvers == NFS_VER2)
568 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp,
569 				    &nb_udp))
570 					err(1, "rpcb_set udp failed");
571 			if (nfs_minvers <= NFS_VER3)
572 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp,
573 				    &nb_udp))
574 					err(1, "rpcb_set udp failed");
575 			freeaddrinfo(ai_udp);
576 		}
577 	}
578 
579 	/* Set up the socket for udp6 and rpcb register it. */
580 	if (udpflag && ip6flag) {
581 		rpcbreg = 0;
582 		for (i = 0; i < bindhostc; i++) {
583 			memset(&hints, 0, sizeof hints);
584 			hints.ai_flags = AI_PASSIVE;
585 			hints.ai_family = AF_INET6;
586 			hints.ai_socktype = SOCK_DGRAM;
587 			hints.ai_protocol = IPPROTO_UDP;
588 			if (setbindhost(&ai_udp6, bindhost[i], hints) == 0) {
589 				rpcbreg = 1;
590 				rpcbregcnt++;
591 				if ((sock = socket(ai_udp6->ai_family,
592 				    ai_udp6->ai_socktype,
593 				    ai_udp6->ai_protocol)) < 0) {
594 					syslog(LOG_ERR,
595 						"can't create udp6 socket");
596 					nfsd_exit(1);
597 				}
598 				if (setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY,
599 				    &on, sizeof on) < 0) {
600 					syslog(LOG_ERR,
601 					    "can't set v6-only binding for "
602 					    "udp6 socket: %m");
603 					nfsd_exit(1);
604 				}
605 				if (bind(sock, ai_udp6->ai_addr,
606 				    ai_udp6->ai_addrlen) < 0) {
607 					syslog(LOG_ERR,
608 					    "can't bind udp6 addr %s: %m",
609 					    bindhost[i]);
610 					nfsd_exit(1);
611 				}
612 				freeaddrinfo(ai_udp6);
613 				addsockargs.sock = sock;
614 				addsockargs.name = NULL;
615 				addsockargs.namelen = 0;
616 				if (nfssvc(nfssvc_addsock, &addsockargs) < 0) {
617 					syslog(LOG_ERR,
618 					    "can't add UDP6 socket");
619 					nfsd_exit(1);
620 				}
621 				(void)close(sock);
622 			}
623 		}
624 		if (rpcbreg == 1) {
625 			memset(&hints, 0, sizeof hints);
626 			hints.ai_flags = AI_PASSIVE;
627 			hints.ai_family = AF_INET6;
628 			hints.ai_socktype = SOCK_DGRAM;
629 			hints.ai_protocol = IPPROTO_UDP;
630 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_udp6);
631 			if (ecode != 0) {
632 				syslog(LOG_ERR, "getaddrinfo udp6: %s",
633 				   gai_strerror(ecode));
634 				nfsd_exit(1);
635 			}
636 			nconf_udp6 = getnetconfigent("udp6");
637 			if (nconf_udp6 == NULL)
638 				err(1, "getnetconfigent udp6 failed");
639 			nb_udp6.buf = ai_udp6->ai_addr;
640 			nb_udp6.len = nb_udp6.maxlen = ai_udp6->ai_addrlen;
641 			if (nfs_minvers == NFS_VER2)
642 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_udp6,
643 				    &nb_udp6))
644 					err(1,
645 					    "rpcb_set udp6 failed");
646 			if (nfs_minvers <= NFS_VER3)
647 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_udp6,
648 				    &nb_udp6))
649 					err(1,
650 					    "rpcb_set udp6 failed");
651 			freeaddrinfo(ai_udp6);
652 		}
653 	}
654 
655 	/* Set up the socket for tcp and rpcb register it. */
656 	if (tcpflag) {
657 		rpcbreg = 0;
658 		for (i = 0; i < bindhostc; i++) {
659 			memset(&hints, 0, sizeof hints);
660 			hints.ai_flags = AI_PASSIVE;
661 			hints.ai_family = AF_INET;
662 			hints.ai_socktype = SOCK_STREAM;
663 			hints.ai_protocol = IPPROTO_TCP;
664 			if (setbindhost(&ai_tcp, bindhost[i], hints) == 0) {
665 				rpcbreg = 1;
666 				rpcbregcnt++;
667 				if ((tcpsock = socket(AF_INET, SOCK_STREAM,
668 				    0)) < 0) {
669 					syslog(LOG_ERR,
670 					    "can't create tcp socket");
671 					nfsd_exit(1);
672 				}
673 				if (setsockopt(tcpsock, SOL_SOCKET,
674 				    SO_REUSEADDR,
675 				    (char *)&on, sizeof(on)) < 0)
676 					syslog(LOG_ERR,
677 					     "setsockopt SO_REUSEADDR: %m");
678 				if (bind(tcpsock, ai_tcp->ai_addr,
679 				    ai_tcp->ai_addrlen) < 0) {
680 					syslog(LOG_ERR,
681 					    "can't bind tcp addr %s: %m",
682 					    bindhost[i]);
683 					nfsd_exit(1);
684 				}
685 				if (listen(tcpsock, -1) < 0) {
686 					syslog(LOG_ERR, "listen failed");
687 					nfsd_exit(1);
688 				}
689 				freeaddrinfo(ai_tcp);
690 				FD_SET(tcpsock, &sockbits);
691 				maxsock = tcpsock;
692 				connect_type_cnt++;
693 			}
694 		}
695 		if (rpcbreg == 1) {
696 			memset(&hints, 0, sizeof hints);
697 			hints.ai_flags = AI_PASSIVE;
698 			hints.ai_family = AF_INET;
699 			hints.ai_socktype = SOCK_STREAM;
700 			hints.ai_protocol = IPPROTO_TCP;
701 			ecode = getaddrinfo(NULL, "nfs", &hints,
702 			     &ai_tcp);
703 			if (ecode != 0) {
704 				syslog(LOG_ERR, "getaddrinfo tcp: %s",
705 				   gai_strerror(ecode));
706 				nfsd_exit(1);
707 			}
708 			nconf_tcp = getnetconfigent("tcp");
709 			if (nconf_tcp == NULL)
710 				err(1, "getnetconfigent tcp failed");
711 			nb_tcp.buf = ai_tcp->ai_addr;
712 			nb_tcp.len = nb_tcp.maxlen = ai_tcp->ai_addrlen;
713 			if (nfs_minvers == NFS_VER2)
714 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp,
715 				    &nb_tcp))
716 					err(1, "rpcb_set tcp failed");
717 			if (nfs_minvers <= NFS_VER3)
718 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp,
719 				    &nb_tcp))
720 					err(1, "rpcb_set tcp failed");
721 			freeaddrinfo(ai_tcp);
722 		}
723 	}
724 
725 	/* Set up the socket for tcp6 and rpcb register it. */
726 	if (tcpflag && ip6flag) {
727 		rpcbreg = 0;
728 		for (i = 0; i < bindhostc; i++) {
729 			memset(&hints, 0, sizeof hints);
730 			hints.ai_flags = AI_PASSIVE;
731 			hints.ai_family = AF_INET6;
732 			hints.ai_socktype = SOCK_STREAM;
733 			hints.ai_protocol = IPPROTO_TCP;
734 			if (setbindhost(&ai_tcp6, bindhost[i], hints) == 0) {
735 				rpcbreg = 1;
736 				rpcbregcnt++;
737 				if ((tcp6sock = socket(ai_tcp6->ai_family,
738 				    ai_tcp6->ai_socktype,
739 				    ai_tcp6->ai_protocol)) < 0) {
740 					syslog(LOG_ERR,
741 					    "can't create tcp6 socket");
742 					nfsd_exit(1);
743 				}
744 				if (setsockopt(tcp6sock, SOL_SOCKET,
745 				    SO_REUSEADDR,
746 				    (char *)&on, sizeof(on)) < 0)
747 					syslog(LOG_ERR,
748 					    "setsockopt SO_REUSEADDR: %m");
749 				if (setsockopt(tcp6sock, IPPROTO_IPV6,
750 				    IPV6_V6ONLY, &on, sizeof on) < 0) {
751 					syslog(LOG_ERR,
752 					"can't set v6-only binding for tcp6 "
753 					    "socket: %m");
754 					nfsd_exit(1);
755 				}
756 				if (bind(tcp6sock, ai_tcp6->ai_addr,
757 				    ai_tcp6->ai_addrlen) < 0) {
758 					syslog(LOG_ERR,
759 					    "can't bind tcp6 addr %s: %m",
760 					    bindhost[i]);
761 					nfsd_exit(1);
762 				}
763 				if (listen(tcp6sock, -1) < 0) {
764 					syslog(LOG_ERR, "listen failed");
765 					nfsd_exit(1);
766 				}
767 				freeaddrinfo(ai_tcp6);
768 				FD_SET(tcp6sock, &sockbits);
769 				if (maxsock < tcp6sock)
770 					maxsock = tcp6sock;
771 				connect_type_cnt++;
772 			}
773 		}
774 		if (rpcbreg == 1) {
775 			memset(&hints, 0, sizeof hints);
776 			hints.ai_flags = AI_PASSIVE;
777 			hints.ai_family = AF_INET6;
778 			hints.ai_socktype = SOCK_STREAM;
779 			hints.ai_protocol = IPPROTO_TCP;
780 			ecode = getaddrinfo(NULL, "nfs", &hints, &ai_tcp6);
781 			if (ecode != 0) {
782 				syslog(LOG_ERR, "getaddrinfo tcp6: %s",
783 				   gai_strerror(ecode));
784 				nfsd_exit(1);
785 			}
786 			nconf_tcp6 = getnetconfigent("tcp6");
787 			if (nconf_tcp6 == NULL)
788 				err(1, "getnetconfigent tcp6 failed");
789 			nb_tcp6.buf = ai_tcp6->ai_addr;
790 			nb_tcp6.len = nb_tcp6.maxlen = ai_tcp6->ai_addrlen;
791 			if (nfs_minvers == NFS_VER2)
792 				if (!rpcb_set(NFS_PROGRAM, 2, nconf_tcp6,
793 				    &nb_tcp6))
794 					err(1, "rpcb_set tcp6 failed");
795 			if (nfs_minvers <= NFS_VER3)
796 				if (!rpcb_set(NFS_PROGRAM, 3, nconf_tcp6,
797 				    &nb_tcp6))
798 					err(1, "rpcb_set tcp6 failed");
799 			freeaddrinfo(ai_tcp6);
800 		}
801 	}
802 
803 	if (rpcbregcnt == 0) {
804 		syslog(LOG_ERR, "rpcb_set() failed, nothing to do: %m");
805 		nfsd_exit(1);
806 	}
807 
808 	if (tcpflag && connect_type_cnt == 0) {
809 		syslog(LOG_ERR, "tcp connects == 0, nothing to do: %m");
810 		nfsd_exit(1);
811 	}
812 
813 	setproctitle("master");
814 	/*
815 	 * We always want a master to have a clean way to shut nfsd down
816 	 * (with unregistration): if the master is killed, it unregisters and
817 	 * kills all children. If we run for UDP only (and so do not have to
818 	 * loop waiting for accept), we instead make the parent
819 	 * a "server" too. start_server will not return.
820 	 */
821 	if (!tcpflag)
822 		start_server(1, &nfsdargs, vhostname);
823 
824 	/*
825 	 * Loop forever accepting connections and passing the sockets
826 	 * into the kernel for the mounts.
827 	 */
828 	for (;;) {
829 		ready = sockbits;
830 		if (connect_type_cnt > 1) {
831 			if (select(maxsock + 1,
832 			    &ready, NULL, NULL, NULL) < 1) {
833 				error = errno;
834 				if (error == EINTR)
835 					continue;
836 				syslog(LOG_ERR, "select failed: %m");
837 				nfsd_exit(1);
838 			}
839 		}
840 		for (tcpsock = 0; tcpsock <= maxsock; tcpsock++) {
841 			if (FD_ISSET(tcpsock, &ready)) {
842 				len = sizeof(peer);
843 				if ((msgsock = accept(tcpsock,
844 				    (struct sockaddr *)&peer, &len)) < 0) {
845 					error = errno;
846 					syslog(LOG_ERR, "accept failed: %m");
847 					if (error == ECONNABORTED ||
848 					    error == EINTR)
849 						continue;
850 					nfsd_exit(1);
851 				}
852 				if (setsockopt(msgsock, SOL_SOCKET,
853 				    SO_KEEPALIVE, (char *)&on, sizeof(on)) < 0)
854 					syslog(LOG_ERR,
855 					    "setsockopt SO_KEEPALIVE: %m");
856 				addsockargs.sock = msgsock;
857 				addsockargs.name = (caddr_t)&peer;
858 				addsockargs.namelen = len;
859 				nfssvc(nfssvc_addsock, &addsockargs);
860 				(void)close(msgsock);
861 			}
862 		}
863 	}
864 }
865 
866 static int
867 setbindhost(struct addrinfo **ai, const char *bindhost, struct addrinfo hints)
868 {
869 	int ecode;
870 	u_int32_t host_addr[4];  /* IPv4 or IPv6 */
871 	const char *hostptr;
872 
873 	if (bindhost == NULL || strcmp("*", bindhost) == 0)
874 		hostptr = NULL;
875 	else
876 		hostptr = bindhost;
877 
878 	if (hostptr != NULL) {
879 		switch (hints.ai_family) {
880 		case AF_INET:
881 			if (inet_pton(AF_INET, hostptr, host_addr) == 1) {
882 				hints.ai_flags = AI_NUMERICHOST;
883 			} else {
884 				if (inet_pton(AF_INET6, hostptr,
885 				    host_addr) == 1)
886 					return (1);
887 			}
888 			break;
889 		case AF_INET6:
890 			if (inet_pton(AF_INET6, hostptr, host_addr) == 1) {
891 				hints.ai_flags = AI_NUMERICHOST;
892 			} else {
893 				if (inet_pton(AF_INET, hostptr,
894 				    host_addr) == 1)
895 					return (1);
896 			}
897 			break;
898 		default:
899 			break;
900 		}
901 	}
902 
903 	ecode = getaddrinfo(hostptr, "nfs", &hints, ai);
904 	if (ecode != 0) {
905 		syslog(LOG_ERR, "getaddrinfo %s: %s", bindhost,
906 		    gai_strerror(ecode));
907 		return (1);
908 	}
909 	return (0);
910 }
911 
912 static void
913 set_nfsdcnt(int proposed)
914 {
915 
916 	if (proposed < 1) {
917 		warnx("nfsd count too low %d; reset to %d", proposed,
918 		    DEFNFSDCNT);
919 		nfsdcnt = DEFNFSDCNT;
920 	} else if (proposed > MAXNFSDCNT) {
921 		warnx("nfsd count too high %d; truncated to %d", proposed,
922 		    MAXNFSDCNT);
923 		nfsdcnt = MAXNFSDCNT;
924 	} else
925 		nfsdcnt = proposed;
926 	nfsdcnt_set = 1;
927 }
928 
929 static void
930 usage(void)
931 {
932 	(void)fprintf(stderr, "%s", getopt_usage);
933 	exit(1);
934 }
935 
936 static void
937 nonfs(__unused int signo)
938 {
939 	syslog(LOG_ERR, "missing system call: NFS not available");
940 }
941 
942 static void
943 reapchild(__unused int signo)
944 {
945 	pid_t pid;
946 	int i;
947 
948 	while ((pid = wait3(NULL, WNOHANG, NULL)) > 0) {
949 		for (i = 0; i < nfsdcnt; i++)
950 			if (pid == children[i])
951 				children[i] = -1;
952 	}
953 }
954 
955 static void
956 unregistration(void)
957 {
958 	if ((nfs_minvers == NFS_VER2 && !rpcb_unset(NFS_PROGRAM, 2, NULL)) ||
959 	    (nfs_minvers <= NFS_VER3 && !rpcb_unset(NFS_PROGRAM, 3, NULL)))
960 		syslog(LOG_ERR, "rpcb_unset failed");
961 }
962 
963 static void
964 killchildren(void)
965 {
966 	int i;
967 
968 	for (i = 0; i < nfsdcnt; i++) {
969 		if (children[i] > 0)
970 			kill(children[i], SIGKILL);
971 	}
972 }
973 
974 /*
975  * Cleanup master after SIGUSR1.
976  */
977 static void
978 cleanup(__unused int signo)
979 {
980 	nfsd_exit(0);
981 }
982 
983 /*
984  * Cleanup child after SIGUSR1.
985  */
986 static void
987 child_cleanup(__unused int signo)
988 {
989 	exit(0);
990 }
991 
992 static void
993 nfsd_exit(int status)
994 {
995 	killchildren();
996 	unregistration();
997 	exit(status);
998 }
999 
1000 static int
1001 get_tuned_nfsdcount(void)
1002 {
1003 	int ncpu, error, tuned_nfsdcnt;
1004 	size_t ncpu_size;
1005 
1006 	ncpu_size = sizeof(ncpu);
1007 	error = sysctlbyname("hw.ncpu", &ncpu, &ncpu_size, NULL, 0);
1008 	if (error) {
1009 		warnx("sysctlbyname(hw.ncpu) failed defaulting to %d nfs servers",
1010 		    DEFNFSDCNT);
1011 		tuned_nfsdcnt = DEFNFSDCNT;
1012 	} else {
1013 		tuned_nfsdcnt = ncpu * 8;
1014 	}
1015 	return tuned_nfsdcnt;
1016 }
1017 
1018 static void
1019 start_server(int master, struct nfsd_nfsd_args *nfsdargp, const char *vhost)
1020 {
1021 	char principal[MAXHOSTNAMELEN + 5];
1022 	int status, error;
1023 	char hostname[MAXHOSTNAMELEN + 1], *cp;
1024 	struct addrinfo *aip, hints;
1025 
1026 	status = 0;
1027 	if (vhost == NULL)
1028 		gethostname(hostname, sizeof (hostname));
1029 	else
1030 		strlcpy(hostname, vhost, sizeof (hostname));
1031 	snprintf(principal, sizeof (principal), "nfs@%s", hostname);
1032 	if ((cp = strchr(hostname, '.')) == NULL ||
1033 	    *(cp + 1) == '\0') {
1034 		/* If not fully qualified, try getaddrinfo() */
1035 		memset((void *)&hints, 0, sizeof (hints));
1036 		hints.ai_flags = AI_CANONNAME;
1037 		error = getaddrinfo(hostname, NULL, &hints, &aip);
1038 		if (error == 0) {
1039 			if (aip->ai_canonname != NULL &&
1040 			    (cp = strchr(aip->ai_canonname, '.')) !=
1041 			    NULL && *(cp + 1) != '\0')
1042 				snprintf(principal, sizeof (principal),
1043 				    "nfs@%s", aip->ai_canonname);
1044 			freeaddrinfo(aip);
1045 		}
1046 	}
1047 	nfsdargp->principal = principal;
1048 
1049 	if (nfsdcnt_set)
1050 		nfsdargp->minthreads = nfsdargp->maxthreads = nfsdcnt;
1051 	else {
1052 		nfsdargp->minthreads = minthreads_set ? minthreads : get_tuned_nfsdcount();
1053 		nfsdargp->maxthreads = maxthreads_set ? maxthreads : nfsdargp->minthreads;
1054 		if (nfsdargp->maxthreads < nfsdargp->minthreads)
1055 			nfsdargp->maxthreads = nfsdargp->minthreads;
1056 	}
1057 	error = nfssvc(nfssvc_nfsd, nfsdargp);
1058 	if (error < 0 && errno == EAUTH) {
1059 		/*
1060 		 * This indicates that it could not register the
1061 		 * rpcsec_gss credentials, usually because the
1062 		 * gssd daemon isn't running.
1063 		 * (only the experimental server with nfsv4)
1064 		 */
1065 		syslog(LOG_ERR, "No gssd, using AUTH_SYS only");
1066 		principal[0] = '\0';
1067 		error = nfssvc(nfssvc_nfsd, nfsdargp);
1068 	}
1069 	if (error < 0) {
1070 		if (errno == ENXIO) {
1071 			syslog(LOG_ERR, "Bad -p option, cannot run");
1072 			if (masterpid != 0 && master == 0)
1073 				kill(masterpid, SIGUSR1);
1074 		} else
1075 			syslog(LOG_ERR, "nfssvc: %m");
1076 		status = 1;
1077 	}
1078 	if (master)
1079 		nfsd_exit(status);
1080 	else
1081 		exit(status);
1082 }
1083 
1084 /*
1085  * Open the stable restart file and return the file descriptor for it.
1086  */
1087 static void
1088 open_stable(int *stable_fdp, int *backup_fdp)
1089 {
1090 	int stable_fd, backup_fd = -1, ret;
1091 	struct stat st, backup_st;
1092 
1093 	/* Open and stat the stable restart file. */
1094 	stable_fd = open(NFSD_STABLERESTART, O_RDWR, 0);
1095 	if (stable_fd < 0)
1096 		stable_fd = open(NFSD_STABLERESTART, O_RDWR | O_CREAT, 0600);
1097 	if (stable_fd >= 0) {
1098 		ret = fstat(stable_fd, &st);
1099 		if (ret < 0) {
1100 			close(stable_fd);
1101 			stable_fd = -1;
1102 		}
1103 	}
1104 
1105 	/* Open and stat the backup stable restart file. */
1106 	if (stable_fd >= 0) {
1107 		backup_fd = open(NFSD_STABLEBACKUP, O_RDWR, 0);
1108 		if (backup_fd < 0)
1109 			backup_fd = open(NFSD_STABLEBACKUP, O_RDWR | O_CREAT,
1110 			    0600);
1111 		if (backup_fd >= 0) {
1112 			ret = fstat(backup_fd, &backup_st);
1113 			if (ret < 0) {
1114 				close(backup_fd);
1115 				backup_fd = -1;
1116 			}
1117 		}
1118 		if (backup_fd < 0) {
1119 			close(stable_fd);
1120 			stable_fd = -1;
1121 		}
1122 	}
1123 
1124 	*stable_fdp = stable_fd;
1125 	*backup_fdp = backup_fd;
1126 	if (stable_fd < 0)
1127 		return;
1128 
1129 	/* Sync up the 2 files, as required. */
1130 	if (st.st_size > 0)
1131 		copy_stable(stable_fd, backup_fd);
1132 	else if (backup_st.st_size > 0)
1133 		copy_stable(backup_fd, stable_fd);
1134 }
1135 
1136 /*
1137  * Copy the stable restart file to the backup or vice versa.
1138  */
1139 static void
1140 copy_stable(int from_fd, int to_fd)
1141 {
1142 	int cnt, ret;
1143 	static char buf[1024];
1144 
1145 	ret = lseek(from_fd, (off_t)0, SEEK_SET);
1146 	if (ret >= 0)
1147 		ret = lseek(to_fd, (off_t)0, SEEK_SET);
1148 	if (ret >= 0)
1149 		ret = ftruncate(to_fd, (off_t)0);
1150 	if (ret >= 0)
1151 		do {
1152 			cnt = read(from_fd, buf, 1024);
1153 			if (cnt > 0)
1154 				ret = write(to_fd, buf, cnt);
1155 			else if (cnt < 0)
1156 				ret = cnt;
1157 		} while (cnt > 0 && ret >= 0);
1158 	if (ret >= 0)
1159 		ret = fsync(to_fd);
1160 	if (ret < 0)
1161 		syslog(LOG_ERR, "stable restart copy failure: %m");
1162 }
1163 
1164 /*
1165  * Back up the stable restart file when indicated by the kernel.
1166  */
1167 static void
1168 backup_stable(__unused int signo)
1169 {
1170 
1171 	if (stablefd >= 0)
1172 		copy_stable(stablefd, backupfd);
1173 }
1174 
1175 /*
1176  * Parse the pNFS string and extract the DS servers and ports numbers.
1177  */
1178 static void
1179 parse_dsserver(const char *optionarg, struct nfsd_nfsd_args *nfsdargp)
1180 {
1181 	char *cp, *cp2, *dsaddr, *dshost, *dspath, *dsvol, nfsprt[9];
1182 	char *mdspath, *mdsp, ip6[INET6_ADDRSTRLEN];
1183 	const char *ad;
1184 	int ecode;
1185 	u_int adsiz, dsaddrcnt, dshostcnt, dspathcnt, hostsiz, pathsiz;
1186 	u_int mdspathcnt;
1187 	size_t dsaddrsiz, dshostsiz, dspathsiz, nfsprtsiz, mdspathsiz;
1188 	struct addrinfo hints, *ai_tcp, *res;
1189 	struct sockaddr_in sin;
1190 	struct sockaddr_in6 sin6;
1191 
1192 	cp = strdup(optionarg);
1193 	if (cp == NULL)
1194 		errx(1, "Out of memory");
1195 
1196 	/* Now, do the host names. */
1197 	dspathsiz = 1024;
1198 	dspathcnt = 0;
1199 	dspath = malloc(dspathsiz);
1200 	if (dspath == NULL)
1201 		errx(1, "Out of memory");
1202 	dshostsiz = 1024;
1203 	dshostcnt = 0;
1204 	dshost = malloc(dshostsiz);
1205 	if (dshost == NULL)
1206 		errx(1, "Out of memory");
1207 	dsaddrsiz = 1024;
1208 	dsaddrcnt = 0;
1209 	dsaddr = malloc(dsaddrsiz);
1210 	if (dsaddr == NULL)
1211 		errx(1, "Out of memory");
1212 	mdspathsiz = 1024;
1213 	mdspathcnt = 0;
1214 	mdspath = malloc(mdspathsiz);
1215 	if (mdspath == NULL)
1216 		errx(1, "Out of memory");
1217 
1218 	/* Put the NFS port# in "." form. */
1219 	snprintf(nfsprt, 9, ".%d.%d", 2049 >> 8, 2049 & 0xff);
1220 	nfsprtsiz = strlen(nfsprt);
1221 
1222 	ai_tcp = NULL;
1223 	/* Loop around for each DS server name. */
1224 	do {
1225 		cp2 = strchr(cp, ',');
1226 		if (cp2 != NULL) {
1227 			/* Not the last DS in the list. */
1228 			*cp2++ = '\0';
1229 			if (*cp2 == '\0')
1230 				usage();
1231 		}
1232 
1233 		dsvol = strchr(cp, ':');
1234 		if (dsvol == NULL || *(dsvol + 1) == '\0')
1235 			usage();
1236 		*dsvol++ = '\0';
1237 
1238 		/* Optional path for MDS file system to be stored on DS. */
1239 		mdsp = strchr(dsvol, '#');
1240 		if (mdsp != NULL) {
1241 			if (*(mdsp + 1) == '\0' || mdsp <= dsvol)
1242 				usage();
1243 			*mdsp++ = '\0';
1244 		}
1245 
1246 		/* Append this pathname to dspath. */
1247 		pathsiz = strlen(dsvol);
1248 		if (dspathcnt + pathsiz + 1 > dspathsiz) {
1249 			dspathsiz *= 2;
1250 			dspath = realloc(dspath, dspathsiz);
1251 			if (dspath == NULL)
1252 				errx(1, "Out of memory");
1253 		}
1254 		strcpy(&dspath[dspathcnt], dsvol);
1255 		dspathcnt += pathsiz + 1;
1256 
1257 		/* Append this pathname to mdspath. */
1258 		if (mdsp != NULL)
1259 			pathsiz = strlen(mdsp);
1260 		else
1261 			pathsiz = 0;
1262 		if (mdspathcnt + pathsiz + 1 > mdspathsiz) {
1263 			mdspathsiz *= 2;
1264 			mdspath = realloc(mdspath, mdspathsiz);
1265 			if (mdspath == NULL)
1266 				errx(1, "Out of memory");
1267 		}
1268 		if (mdsp != NULL)
1269 			strcpy(&mdspath[mdspathcnt], mdsp);
1270 		else
1271 			mdspath[mdspathcnt] = '\0';
1272 		mdspathcnt += pathsiz + 1;
1273 
1274 		if (ai_tcp != NULL)
1275 			freeaddrinfo(ai_tcp);
1276 
1277 		/* Get the fully qualified domain name and IP address. */
1278 		memset(&hints, 0, sizeof(hints));
1279 		hints.ai_flags = AI_CANONNAME | AI_ADDRCONFIG;
1280 		hints.ai_family = PF_UNSPEC;
1281 		hints.ai_socktype = SOCK_STREAM;
1282 		hints.ai_protocol = IPPROTO_TCP;
1283 		ecode = getaddrinfo(cp, NULL, &hints, &ai_tcp);
1284 		if (ecode != 0)
1285 			err(1, "getaddrinfo pnfs: %s %s", cp,
1286 			    gai_strerror(ecode));
1287 		ad = NULL;
1288 		for (res = ai_tcp; res != NULL; res = res->ai_next) {
1289 			if (res->ai_addr->sa_family == AF_INET) {
1290 				if (res->ai_addrlen < sizeof(sin))
1291 					err(1, "getaddrinfo() returned "
1292 					    "undersized IPv4 address");
1293 				/*
1294 				 * Mips cares about sockaddr_in alignment,
1295 				 * so copy the address.
1296 				 */
1297 				memcpy(&sin, res->ai_addr, sizeof(sin));
1298 				ad = inet_ntoa(sin.sin_addr);
1299 				break;
1300 			} else if (res->ai_family == AF_INET6) {
1301 				if (res->ai_addrlen < sizeof(sin6))
1302 					err(1, "getaddrinfo() returned "
1303 					    "undersized IPv6 address");
1304 				/*
1305 				 * Mips cares about sockaddr_in6 alignment,
1306 				 * so copy the address.
1307 				 */
1308 				memcpy(&sin6, res->ai_addr, sizeof(sin6));
1309 				ad = inet_ntop(AF_INET6, &sin6.sin6_addr, ip6,
1310 				    sizeof(ip6));
1311 
1312 				/*
1313 				 * XXX
1314 				 * Since a link local address will only
1315 				 * work if the client and DS are in the
1316 				 * same scope zone, only use it if it is
1317 				 * the only address.
1318 				 */
1319 				if (ad != NULL &&
1320 				    !IN6_IS_ADDR_LINKLOCAL(&sin6.sin6_addr))
1321 					break;
1322 			}
1323 		}
1324 		if (ad == NULL)
1325 			err(1, "No IP address for %s", cp);
1326 
1327 		/* Append this address to dsaddr. */
1328 		adsiz = strlen(ad);
1329 		if (dsaddrcnt + adsiz + nfsprtsiz + 1 > dsaddrsiz) {
1330 			dsaddrsiz *= 2;
1331 			dsaddr = realloc(dsaddr, dsaddrsiz);
1332 			if (dsaddr == NULL)
1333 				errx(1, "Out of memory");
1334 		}
1335 		strcpy(&dsaddr[dsaddrcnt], ad);
1336 		strcat(&dsaddr[dsaddrcnt], nfsprt);
1337 		dsaddrcnt += adsiz + nfsprtsiz + 1;
1338 
1339 		/* Append this hostname to dshost. */
1340 		hostsiz = strlen(ai_tcp->ai_canonname);
1341 		if (dshostcnt + hostsiz + 1 > dshostsiz) {
1342 			dshostsiz *= 2;
1343 			dshost = realloc(dshost, dshostsiz);
1344 			if (dshost == NULL)
1345 				errx(1, "Out of memory");
1346 		}
1347 		strcpy(&dshost[dshostcnt], ai_tcp->ai_canonname);
1348 		dshostcnt += hostsiz + 1;
1349 
1350 		cp = cp2;
1351 	} while (cp != NULL);
1352 
1353 	nfsdargp->addr = dsaddr;
1354 	nfsdargp->addrlen = dsaddrcnt;
1355 	nfsdargp->dnshost = dshost;
1356 	nfsdargp->dnshostlen = dshostcnt;
1357 	nfsdargp->dspath = dspath;
1358 	nfsdargp->dspathlen = dspathcnt;
1359 	nfsdargp->mdspath = mdspath;
1360 	nfsdargp->mdspathlen = mdspathcnt;
1361 	freeaddrinfo(ai_tcp);
1362 }
1363 
1364