xref: /freebsd/sys/fs/nfsserver/nfs_nfsdkrpc.c (revision f126890ac5386406dadf7c4cfa9566cbb56537c5)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
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 
36 #include <sys/cdefs.h>
37 #include "opt_inet6.h"
38 #include "opt_kgssapi.h"
39 #include "opt_kern_tls.h"
40 
41 #include <fs/nfs/nfsport.h>
42 
43 #include <rpc/rpc.h>
44 #include <rpc/rpcsec_gss.h>
45 #include <rpc/rpcsec_tls.h>
46 
47 #include <fs/nfsserver/nfs_fha_new.h>
48 
49 #include <security/mac/mac_framework.h>
50 
51 NFSDLOCKMUTEX;
52 NFSV4ROOTLOCKMUTEX;
53 char *nfsrv_zeropnfsdat = NULL;
54 
55 /*
56  * Mapping of old NFS Version 2 RPC numbers to generic numbers.
57  */
58 int newnfs_nfsv3_procid[NFS_V3NPROCS] = {
59 	NFSPROC_NULL,
60 	NFSPROC_GETATTR,
61 	NFSPROC_SETATTR,
62 	NFSPROC_NOOP,
63 	NFSPROC_LOOKUP,
64 	NFSPROC_READLINK,
65 	NFSPROC_READ,
66 	NFSPROC_NOOP,
67 	NFSPROC_WRITE,
68 	NFSPROC_CREATE,
69 	NFSPROC_REMOVE,
70 	NFSPROC_RENAME,
71 	NFSPROC_LINK,
72 	NFSPROC_SYMLINK,
73 	NFSPROC_MKDIR,
74 	NFSPROC_RMDIR,
75 	NFSPROC_READDIR,
76 	NFSPROC_FSSTAT,
77 	NFSPROC_NOOP,
78 	NFSPROC_NOOP,
79 	NFSPROC_NOOP,
80 	NFSPROC_NOOP,
81 };
82 
83 SYSCTL_DECL(_vfs_nfsd);
84 
85 NFSD_VNET_DEFINE_STATIC(int, nfs_privport) = 0;
86 SYSCTL_INT(_vfs_nfsd, OID_AUTO, nfs_privport, CTLFLAG_NFSD_VNET | CTLFLAG_RWTUN,
87     &NFSD_VNET_NAME(nfs_privport), 0,
88     "Only allow clients using a privileged port for NFSv2, 3 and 4");
89 
90 NFSD_VNET_DEFINE_STATIC(int, nfs_minvers) = NFS_VER2;
91 SYSCTL_INT(_vfs_nfsd, OID_AUTO, server_min_nfsvers,
92     CTLFLAG_NFSD_VNET | CTLFLAG_RWTUN, &NFSD_VNET_NAME(nfs_minvers), 0,
93     "The lowest version of NFS handled by the server");
94 
95 NFSD_VNET_DEFINE_STATIC(int, nfs_maxvers) = NFS_VER4;
96 SYSCTL_INT(_vfs_nfsd, OID_AUTO, server_max_nfsvers,
97     CTLFLAG_NFSD_VNET | CTLFLAG_RWTUN, &NFSD_VNET_NAME(nfs_maxvers), 0,
98     "The highest version of NFS handled by the server");
99 
100 static int nfs_proc(struct nfsrv_descript *, u_int32_t, SVCXPRT *xprt,
101     struct nfsrvcache **);
102 
103 extern u_long sb_max_adj;
104 extern int newnfs_numnfsd;
105 extern time_t nfsdev_time;
106 extern int nfsrv_writerpc[NFS_NPROCS];
107 extern volatile int nfsrv_devidcnt;
108 extern struct nfsv4_opflag nfsv4_opflag[NFSV42_NOPS];
109 extern int nfsd_debuglevel;
110 
111 NFSD_VNET_DECLARE(struct proc *, nfsd_master_proc);
112 
113 NFSD_VNET_DEFINE(SVCPOOL *, nfsrvd_pool);
114 NFSD_VNET_DEFINE(int, nfsrv_numnfsd) = 0;
115 NFSD_VNET_DEFINE(struct nfsv4lock, nfsd_suspend_lock);
116 
117 NFSD_VNET_DEFINE_STATIC(bool, nfsrvd_inited) = false;
118 
119 /*
120  * NFS server system calls
121  */
122 
123 static void
124 nfssvc_program(struct svc_req *rqst, SVCXPRT *xprt)
125 {
126 	struct nfsrv_descript nd;
127 	struct nfsrvcache *rp = NULL;
128 	rpc_gss_rawcred_t *rcredp;
129 	int cacherep, credflavor, i, j;
130 	u_char *p;
131 #ifdef KERN_TLS
132 	u_int maxlen;
133 #endif
134 
135 	NFSD_CURVNET_SET_QUIET(NFSD_TD_TO_VNET(curthread));
136 	memset(&nd, 0, sizeof(nd));
137 	if (rqst->rq_vers == NFS_VER2) {
138 		if (rqst->rq_proc > NFSV2PROC_STATFS ||
139 		    newnfs_nfsv3_procid[rqst->rq_proc] == NFSPROC_NOOP) {
140 			svcerr_noproc(rqst);
141 			svc_freereq(rqst);
142 			goto out;
143 		}
144 		nd.nd_procnum = newnfs_nfsv3_procid[rqst->rq_proc];
145 		nd.nd_flag = ND_NFSV2;
146 	} else if (rqst->rq_vers == NFS_VER3) {
147 		if (rqst->rq_proc >= NFS_V3NPROCS) {
148 			svcerr_noproc(rqst);
149 			svc_freereq(rqst);
150 			goto out;
151 		}
152 		nd.nd_procnum = rqst->rq_proc;
153 		nd.nd_flag = ND_NFSV3;
154 	} else {
155 		if (rqst->rq_proc != NFSPROC_NULL &&
156 		    rqst->rq_proc != NFSV4PROC_COMPOUND) {
157 			svcerr_noproc(rqst);
158 			svc_freereq(rqst);
159 			goto out;
160 		}
161 		nd.nd_procnum = rqst->rq_proc;
162 		nd.nd_flag = ND_NFSV4;
163 	}
164 
165 	/*
166 	 * Note: we want rq_addr, not svc_getrpccaller for nd_nam2 -
167 	 * NFS_SRVMAXDATA uses a NULL value for nd_nam2 to detect TCP
168 	 * mounts.
169 	 */
170 	nd.nd_mrep = rqst->rq_args;
171 	rqst->rq_args = NULL;
172 	newnfs_realign(&nd.nd_mrep, M_WAITOK);
173 	nd.nd_md = nd.nd_mrep;
174 	nd.nd_dpos = mtod(nd.nd_md, caddr_t);
175 	nd.nd_nam = svc_getrpccaller(rqst);
176 	nd.nd_nam2 = rqst->rq_addr;
177 	nd.nd_mreq = NULL;
178 	nd.nd_cred = NULL;
179 
180 	if (NFSD_VNET(nfs_privport) != 0) {
181 		/* Check if source port is privileged */
182 		u_short port;
183 		struct sockaddr *nam = nd.nd_nam;
184 		struct sockaddr_in *sin;
185 
186 		sin = (struct sockaddr_in *)nam;
187 		/*
188 		 * INET/INET6 - same code:
189 		 *    sin_port and sin6_port are at same offset
190 		 */
191 		port = ntohs(sin->sin_port);
192 		if (port >= IPPORT_RESERVED &&
193 		    nd.nd_procnum != NFSPROC_NULL) {
194 #ifdef INET6
195 			char buf[INET6_ADDRSTRLEN];
196 #else
197 			char buf[INET_ADDRSTRLEN];
198 #endif
199 #ifdef INET6
200 #if defined(KLD_MODULE)
201 			/* Do not use ip6_sprintf: the nfs module should work without INET6. */
202 #define	ip6_sprintf(buf, a)						\
203 			(sprintf((buf), "%x:%x:%x:%x:%x:%x:%x:%x",	\
204 			    (a)->s6_addr16[0], (a)->s6_addr16[1],	\
205 			    (a)->s6_addr16[2], (a)->s6_addr16[3],	\
206 			    (a)->s6_addr16[4], (a)->s6_addr16[5],	\
207 			    (a)->s6_addr16[6], (a)->s6_addr16[7]),	\
208 			    (buf))
209 #endif
210 #endif
211 			printf("NFS request from unprivileged port (%s:%d)\n",
212 #ifdef INET6
213 			    sin->sin_family == AF_INET6 ?
214 			    ip6_sprintf(buf, &satosin6(sin)->sin6_addr) :
215 #if defined(KLD_MODULE)
216 #undef ip6_sprintf
217 #endif
218 #endif
219 			    inet_ntoa_r(sin->sin_addr, buf), port);
220 			svcerr_weakauth(rqst);
221 			svc_freereq(rqst);
222 			m_freem(nd.nd_mrep);
223 			goto out;
224 		}
225 	}
226 
227 	if (nd.nd_procnum != NFSPROC_NULL) {
228 		if (!svc_getcred(rqst, &nd.nd_cred, &credflavor)) {
229 			svcerr_weakauth(rqst);
230 			svc_freereq(rqst);
231 			m_freem(nd.nd_mrep);
232 			goto out;
233 		}
234 
235 		/* Set the flag based on credflavor */
236 		if (credflavor == RPCSEC_GSS_KRB5) {
237 			nd.nd_flag |= ND_GSS;
238 		} else if (credflavor == RPCSEC_GSS_KRB5I) {
239 			nd.nd_flag |= (ND_GSS | ND_GSSINTEGRITY);
240 		} else if (credflavor == RPCSEC_GSS_KRB5P) {
241 			nd.nd_flag |= (ND_GSS | ND_GSSPRIVACY);
242 		} else if (credflavor != AUTH_SYS) {
243 			svcerr_weakauth(rqst);
244 			svc_freereq(rqst);
245 			m_freem(nd.nd_mrep);
246 			goto out;
247 		}
248 
249 		/* Acquire the principal name for the RPCSEC_GSS cases. */
250 		if ((nd.nd_flag & (ND_NFSV4 | ND_GSS)) == (ND_NFSV4 | ND_GSS)) {
251 			rcredp = NULL;
252 			rpc_gss_getcred_call(rqst, &rcredp, NULL, NULL);
253 			/*
254 			 * The exported principal name consists of:
255 			 * - TOK_ID bytes with value of 4 and 1.
256 			 * - 2 byte mech length
257 			 * - mech
258 			 * - 4 byte principal name length
259 			 * - principal name
260 			 * A call to gss_import_name() would be an
261 			 * upcall to the gssd, so parse it here.
262 			 * See lib/libgssapi/gss_import_name.c for the
263 			 * above format.
264 			 */
265 			if (rcredp != NULL &&
266 			    rcredp->client_principal->len > 4 &&
267 			    rcredp->client_principal->name[0] == 4 &&
268 			    rcredp->client_principal->name[1] == 1) {
269 				/* Skip over the mech. */
270 				p = &rcredp->client_principal->name[2];
271 				i = (p[0] << 8) | p[1];
272 				p += i + 2;
273 				i += 4;
274 				/*
275 				 * Set "j" to a bogus length so that the
276 				 * "i + j" check will fail unless the below
277 				 * code sets "j" correctly.
278 				 */
279 				j = rcredp->client_principal->len;
280 				if (rcredp->client_principal->len > i + 4) {
281 					j = (p[0] << 24) | (p[1] << 16) |
282 					    (p[2] << 8) | p[3];
283 					i += 4;
284 					p += 4;
285 				}
286 				if (i + j == rcredp->client_principal->len) {
287 					nd.nd_principal = malloc(j + 1, M_TEMP,
288 					    M_WAITOK);
289 					nd.nd_princlen = j;
290 					memcpy(nd.nd_principal, p, j);
291 					nd.nd_principal[j] = '\0';
292 					NFSD_DEBUG(1, "nfssvc_program: "
293 					    "principal=%s\n", nd.nd_principal);
294 				}
295 			}
296 			if (nd.nd_princlen == 0)
297 				printf("nfssvc_program: cannot get RPCSEC_GSS "
298 				    "principal name\n");
299 		}
300 
301 		if ((xprt->xp_tls & RPCTLS_FLAGS_HANDSHAKE) != 0) {
302 			nd.nd_flag |= ND_TLS;
303 			if ((xprt->xp_tls & RPCTLS_FLAGS_VERIFIED) != 0)
304 				nd.nd_flag |= ND_TLSCERT;
305 			if ((xprt->xp_tls & RPCTLS_FLAGS_CERTUSER) != 0)
306 				nd.nd_flag |= ND_TLSCERTUSER;
307 		}
308 		nd.nd_maxextsiz = 16384;
309 #ifdef MAC
310 		mac_cred_associate_nfsd(nd.nd_cred);
311 #endif
312 		/*
313 		 * Get a refcnt (shared lock) on nfsd_suspend_lock.
314 		 * NFSSVC_SUSPENDNFSD will take an exclusive lock on
315 		 * nfsd_suspend_lock to suspend these threads.
316 		 * The call to nfsv4_lock() that precedes nfsv4_getref()
317 		 * ensures that the acquisition of the exclusive lock
318 		 * takes priority over acquisition of the shared lock by
319 		 * waiting for any exclusive lock request to complete.
320 		 * This must be done here, before the check of
321 		 * nfsv4root exports by nfsvno_v4rootexport().
322 		 */
323 		NFSLOCKV4ROOTMUTEX();
324 		nfsv4_lock(&NFSD_VNET(nfsd_suspend_lock), 0, NULL,
325 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
326 		nfsv4_getref(&NFSD_VNET(nfsd_suspend_lock), NULL,
327 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
328 		NFSUNLOCKV4ROOTMUTEX();
329 
330 		if ((nd.nd_flag & ND_NFSV4) != 0) {
331 			nd.nd_repstat = nfsvno_v4rootexport(&nd);
332 			if (nd.nd_repstat != 0) {
333 				NFSLOCKV4ROOTMUTEX();
334 				nfsv4_relref(&NFSD_VNET(nfsd_suspend_lock));
335 				NFSUNLOCKV4ROOTMUTEX();
336 				svcerr_weakauth(rqst);
337 				svc_freereq(rqst);
338 				m_freem(nd.nd_mrep);
339 				goto out;
340 			}
341 		}
342 
343 #ifdef KERN_TLS
344 		if ((xprt->xp_tls & RPCTLS_FLAGS_HANDSHAKE) != 0 &&
345 		    rpctls_getinfo(&maxlen, false, false))
346 			nd.nd_maxextsiz = maxlen;
347 #endif
348 		cacherep = nfs_proc(&nd, rqst->rq_xid, xprt, &rp);
349 		NFSLOCKV4ROOTMUTEX();
350 		nfsv4_relref(&NFSD_VNET(nfsd_suspend_lock));
351 		NFSUNLOCKV4ROOTMUTEX();
352 	} else {
353 		NFSMGET(nd.nd_mreq);
354 		nd.nd_mreq->m_len = 0;
355 		cacherep = RC_REPLY;
356 	}
357 	if (nd.nd_mrep != NULL)
358 		m_freem(nd.nd_mrep);
359 
360 	if (nd.nd_cred != NULL)
361 		crfree(nd.nd_cred);
362 
363 	if (cacherep == RC_DROPIT) {
364 		if (nd.nd_mreq != NULL)
365 			m_freem(nd.nd_mreq);
366 		svc_freereq(rqst);
367 		goto out;
368 	}
369 
370 	if (nd.nd_mreq == NULL) {
371 		svcerr_decode(rqst);
372 		svc_freereq(rqst);
373 		goto out;
374 	}
375 
376 	if (nd.nd_repstat & NFSERR_AUTHERR) {
377 		svcerr_auth(rqst, nd.nd_repstat & ~NFSERR_AUTHERR);
378 		if (nd.nd_mreq != NULL)
379 			m_freem(nd.nd_mreq);
380 	} else if (!svc_sendreply_mbuf(rqst, nd.nd_mreq)) {
381 		svcerr_systemerr(rqst);
382 	}
383 	if (rp != NULL) {
384 		nfsrvd_sentcache(rp, (rqst->rq_reply_seq != 0 ||
385 		    SVC_ACK(xprt, NULL)), rqst->rq_reply_seq);
386 	}
387 	svc_freereq(rqst);
388 
389 out:
390 	free(nd.nd_principal, M_TEMP);
391 	NFSD_CURVNET_RESTORE();
392 	ast_kclear(curthread);
393 	NFSEXITCODE(0);
394 }
395 
396 /*
397  * Check the cache and, optionally, do the RPC.
398  * Return the appropriate cache response.
399  */
400 static int
401 nfs_proc(struct nfsrv_descript *nd, u_int32_t xid, SVCXPRT *xprt,
402     struct nfsrvcache **rpp)
403 {
404 	int cacherep = RC_DOIT, isdgram, taglen = -1;
405 	struct mbuf *m;
406 	u_char tag[NFSV4_SMALLSTR + 1], *tagstr = NULL;
407 	u_int32_t minorvers = 0;
408 	uint32_t ack;
409 
410 	*rpp = NULL;
411 	if (nd->nd_nam2 == NULL) {
412 		nd->nd_flag |= ND_STREAMSOCK;
413 		isdgram = 0;
414 	} else {
415 		isdgram = 1;
416 	}
417 
418 	/*
419 	 * Two cases:
420 	 * 1 - For NFSv2 over UDP, if we are near our malloc/mget
421 	 *     limit, just drop the request. There is no
422 	 *     NFSERR_RESOURCE or NFSERR_DELAY for NFSv2 and the
423 	 *     client will timeout/retry over UDP in a little while.
424 	 * 2 - nd_repstat == 0 && nd_mreq == NULL, which
425 	 *     means a normal nfs rpc, so check the cache
426 	 */
427 	if ((nd->nd_flag & ND_NFSV2) && nd->nd_nam2 != NULL &&
428 	    nfsrv_mallocmget_limit()) {
429 		cacherep = RC_DROPIT;
430 	} else {
431 		/*
432 		 * For NFSv3, play it safe and assume that the client is
433 		 * doing retries on the same TCP connection.
434 		 */
435 		if ((nd->nd_flag & (ND_NFSV4 | ND_STREAMSOCK)) ==
436 		    ND_STREAMSOCK)
437 			nd->nd_flag |= ND_SAMETCPCONN;
438 		nd->nd_retxid = xid;
439 		nd->nd_tcpconntime = NFSD_MONOSEC;
440 		nd->nd_sockref = xprt->xp_sockref;
441 		if ((nd->nd_flag & ND_NFSV4) != 0)
442 			nfsd_getminorvers(nd, tag, &tagstr, &taglen,
443 			    &minorvers);
444 		if ((nd->nd_flag & ND_NFSV41) != 0)
445 			/* NFSv4.1 caches replies in the session slots. */
446 			cacherep = RC_DOIT;
447 		else {
448 			cacherep = nfsrvd_getcache(nd);
449 			ack = 0;
450 			SVC_ACK(xprt, &ack);
451 			nfsrc_trimcache(xprt->xp_sockref, ack, 0);
452 		}
453 	}
454 
455 	/*
456 	 * Handle the request. There are three cases.
457 	 * RC_DOIT - do the RPC
458 	 * RC_REPLY - return the reply already created
459 	 * RC_DROPIT - just throw the request away
460 	 */
461 	if (cacherep == RC_DOIT) {
462 		if ((nd->nd_flag & ND_NFSV41) != 0)
463 			nd->nd_xprt = xprt;
464 		nfsrvd_dorpc(nd, isdgram, tagstr, taglen, minorvers);
465 		if ((nd->nd_flag & ND_NFSV41) != 0) {
466 			if (nd->nd_repstat != NFSERR_REPLYFROMCACHE &&
467 			    (nd->nd_flag & ND_SAVEREPLY) != 0) {
468 				/* Cache a copy of the reply. */
469 				m = m_copym(nd->nd_mreq, 0, M_COPYALL,
470 				    M_WAITOK);
471 			} else
472 				m = NULL;
473 			if ((nd->nd_flag & ND_HASSEQUENCE) != 0)
474 				nfsrv_cache_session(nd, &m);
475 			if (nd->nd_repstat == NFSERR_REPLYFROMCACHE) {
476 				nd->nd_repstat = 0;
477 				if (m != NULL) {
478 					m_freem(nd->nd_mreq);
479 					nd->nd_mreq = m;
480 				}
481 			}
482 			cacherep = RC_REPLY;
483 		} else {
484 			if (nd->nd_repstat == NFSERR_DONTREPLY)
485 				cacherep = RC_DROPIT;
486 			else
487 				cacherep = RC_REPLY;
488 			*rpp = nfsrvd_updatecache(nd);
489 		}
490 	}
491 	if (tagstr != NULL && taglen > NFSV4_SMALLSTR)
492 		free(tagstr, M_TEMP);
493 
494 	NFSEXITCODE2(0, nd);
495 	return (cacherep);
496 }
497 
498 static void
499 nfssvc_loss(SVCXPRT *xprt)
500 {
501 	uint32_t ack;
502 
503 	ack = 0;
504 	SVC_ACK(xprt, &ack);
505 	NFSD_CURVNET_SET(NFSD_TD_TO_VNET(curthread));
506 	nfsrc_trimcache(xprt->xp_sockref, ack, 1);
507 	NFSD_CURVNET_RESTORE();
508 }
509 
510 /*
511  * Adds a socket to the list for servicing by nfsds.
512  */
513 int
514 nfsrvd_addsock(struct file *fp)
515 {
516 	int siz;
517 	struct socket *so;
518 	int error = 0;
519 	SVCXPRT *xprt;
520 	static u_int64_t sockref = 0;
521 
522 	so = fp->f_data;
523 
524 	siz = sb_max_adj;
525 	error = soreserve(so, siz, siz);
526 	if (error)
527 		goto out;
528 
529 	/*
530 	 * Steal the socket from userland so that it doesn't close
531 	 * unexpectedly.
532 	 */
533 	if (so->so_type == SOCK_DGRAM)
534 		xprt = svc_dg_create(NFSD_VNET(nfsrvd_pool), so, 0, 0);
535 	else
536 		xprt = svc_vc_create(NFSD_VNET(nfsrvd_pool), so, 0, 0);
537 	if (xprt) {
538 		fp->f_ops = &badfileops;
539 		fp->f_data = NULL;
540 		xprt->xp_sockref = ++sockref;
541 		if (NFSD_VNET(nfs_minvers) == NFS_VER2)
542 			svc_reg(xprt, NFS_PROG, NFS_VER2, nfssvc_program,
543 			    NULL);
544 		if (NFSD_VNET(nfs_minvers) <= NFS_VER3 &&
545 		    NFSD_VNET(nfs_maxvers) >= NFS_VER3)
546 			svc_reg(xprt, NFS_PROG, NFS_VER3, nfssvc_program,
547 			    NULL);
548 		if (NFSD_VNET(nfs_maxvers) >= NFS_VER4)
549 			svc_reg(xprt, NFS_PROG, NFS_VER4, nfssvc_program,
550 			    NULL);
551 		if (so->so_type == SOCK_STREAM)
552 			svc_loss_reg(xprt, nfssvc_loss);
553 		SVC_RELEASE(xprt);
554 	} else
555 		error = EPERM;
556 
557 out:
558 	NFSEXITCODE(error);
559 	return (error);
560 }
561 
562 /*
563  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
564  * until it is killed by a signal.
565  */
566 int
567 nfsrvd_nfsd(struct thread *td, struct nfsd_nfsd_args *args)
568 {
569 	char principal[MAXHOSTNAMELEN + 5];
570 	struct proc *p;
571 	int error = 0;
572 	bool_t ret2, ret3, ret4;
573 
574 	error = copyinstr(args->principal, principal, sizeof (principal),
575 	    NULL);
576 	if (error)
577 		goto out;
578 
579 	/*
580 	 * Only the first nfsd actually does any work. The RPC code
581 	 * adds threads to it as needed. Any extra processes offered
582 	 * by nfsd just exit. If nfsd is new enough, it will call us
583 	 * once with a structure that specifies how many threads to
584 	 * use.
585 	 */
586 	NFSD_LOCK();
587 	if (NFSD_VNET(nfsrv_numnfsd) == 0) {
588 		nfsdev_time = time_second;
589 		p = td->td_proc;
590 		PROC_LOCK(p);
591 		p->p_flag2 |= P2_AST_SU;
592 		PROC_UNLOCK(p);
593 		newnfs_numnfsd++;	/* Total num for all vnets. */
594 		NFSD_VNET(nfsrv_numnfsd)++;	/* Num for this vnet. */
595 
596 		NFSD_UNLOCK();
597 		error = nfsrv_createdevids(args, td);
598 		if (error == 0) {
599 			/* An empty string implies AUTH_SYS only. */
600 			if (principal[0] != '\0') {
601 				ret2 = rpc_gss_set_svc_name_call(principal,
602 				    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG,
603 				    NFS_VER2);
604 				ret3 = rpc_gss_set_svc_name_call(principal,
605 				    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG,
606 				    NFS_VER3);
607 				ret4 = rpc_gss_set_svc_name_call(principal,
608 				    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG,
609 				    NFS_VER4);
610 
611 				if (!ret2 || !ret3 || !ret4)
612 					printf("nfsd: can't register svc "
613 					    "name %s jid:%d\n", principal,
614 					    td->td_ucred->cr_prison->pr_id);
615 			}
616 
617 			NFSD_VNET(nfsrvd_pool)->sp_minthreads =
618 			    args->minthreads;
619 			NFSD_VNET(nfsrvd_pool)->sp_maxthreads =
620 			    args->maxthreads;
621 
622 			/*
623 			 * If this is a pNFS service, make Getattr do a
624 			 * vn_start_write(), so it can do a vn_set_extattr().
625 			 */
626 			if (nfsrv_devidcnt > 0) {
627 				nfsrv_writerpc[NFSPROC_GETATTR] = 1;
628 				nfsv4_opflag[NFSV4OP_GETATTR].modifyfs = 1;
629 			}
630 
631 			svc_run(NFSD_VNET(nfsrvd_pool));
632 
633 			/* Reset Getattr to not do a vn_start_write(). */
634 			nfsrv_writerpc[NFSPROC_GETATTR] = 0;
635 			nfsv4_opflag[NFSV4OP_GETATTR].modifyfs = 0;
636 
637 			if (principal[0] != '\0') {
638 				rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER2);
639 				rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER3);
640 				rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER4);
641 			}
642 		}
643 		NFSD_LOCK();
644 		newnfs_numnfsd--;
645 		NFSD_VNET(nfsrv_numnfsd)--;
646 		nfsrvd_init(1);
647 		PROC_LOCK(p);
648 		p->p_flag2 &= ~P2_AST_SU;
649 		PROC_UNLOCK(p);
650 	}
651 	NFSD_UNLOCK();
652 
653 out:
654 	NFSEXITCODE(error);
655 	return (error);
656 }
657 
658 /*
659  * Initialize the data structures for the server.
660  * Handshake with any new nfsds starting up to avoid any chance of
661  * corruption.
662  */
663 void
664 nfsrvd_init(int terminating)
665 {
666 
667 	NFSD_LOCK_ASSERT();
668 
669 	if (terminating) {
670 		NFSD_VNET(nfsd_master_proc) = NULL;
671 		NFSD_UNLOCK();
672 		nfsrv_freealllayoutsanddevids();
673 		nfsrv_freeallbackchannel_xprts();
674 		svcpool_close(NFSD_VNET(nfsrvd_pool));
675 		free(nfsrv_zeropnfsdat, M_TEMP);
676 		nfsrv_zeropnfsdat = NULL;
677 		NFSD_LOCK();
678 	} else {
679 		/* Initialize per-vnet globals once per vnet. */
680 		if (NFSD_VNET(nfsrvd_inited))
681 			return;
682 		NFSD_VNET(nfsrvd_inited) = true;
683 		NFSD_UNLOCK();
684 		NFSD_VNET(nfsrvd_pool) = svcpool_create("nfsd",
685 		    SYSCTL_STATIC_CHILDREN(_vfs_nfsd));
686 		NFSD_VNET(nfsrvd_pool)->sp_rcache = NULL;
687 		NFSD_VNET(nfsrvd_pool)->sp_assign = fhanew_assign;
688 		NFSD_VNET(nfsrvd_pool)->sp_done = fhanew_nd_complete;
689 		NFSD_LOCK();
690 	}
691 }
692