xref: /freebsd/sys/fs/nfsserver/nfs_nfsdkrpc.c (revision b985c9ca)
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) = 1;
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 			static struct timeval privport_ratecheck = {
195 				.tv_sec = 0, .tv_usec = 0
196 			};
197 			static const struct timeval privport_ratecheck_int = {
198 				.tv_sec = 1, .tv_usec = 0
199 			};
200 #ifdef INET6
201 			char buf[INET6_ADDRSTRLEN];
202 #else
203 			char buf[INET_ADDRSTRLEN];
204 #endif
205 #ifdef INET6
206 #if defined(KLD_MODULE)
207 			/* Do not use ip6_sprintf: the nfs module should work without INET6. */
208 #define	ip6_sprintf(buf, a)						\
209 			(sprintf((buf), "%x:%x:%x:%x:%x:%x:%x:%x",	\
210 			    (a)->s6_addr16[0], (a)->s6_addr16[1],	\
211 			    (a)->s6_addr16[2], (a)->s6_addr16[3],	\
212 			    (a)->s6_addr16[4], (a)->s6_addr16[5],	\
213 			    (a)->s6_addr16[6], (a)->s6_addr16[7]),	\
214 			    (buf))
215 #endif
216 #endif
217 			if (ratecheck(&privport_ratecheck,
218 			    &privport_ratecheck_int)) {
219 				printf(
220 			    "NFS request from unprivileged port (%s:%d)\n",
221 #ifdef INET6
222 				    sin->sin_family == AF_INET6 ?
223 				    ip6_sprintf(buf, &satosin6(sin)->sin6_addr) :
224 #if defined(KLD_MODULE)
225 #undef ip6_sprintf
226 #endif
227 #endif
228 				    inet_ntoa_r(sin->sin_addr, buf), port);
229 			}
230 			svcerr_weakauth(rqst);
231 			svc_freereq(rqst);
232 			m_freem(nd.nd_mrep);
233 			goto out;
234 		}
235 	}
236 
237 	if (nd.nd_procnum != NFSPROC_NULL) {
238 		if (!svc_getcred(rqst, &nd.nd_cred, &credflavor)) {
239 			svcerr_weakauth(rqst);
240 			svc_freereq(rqst);
241 			m_freem(nd.nd_mrep);
242 			goto out;
243 		}
244 
245 		/* Set the flag based on credflavor */
246 		if (credflavor == RPCSEC_GSS_KRB5) {
247 			nd.nd_flag |= ND_GSS;
248 		} else if (credflavor == RPCSEC_GSS_KRB5I) {
249 			nd.nd_flag |= (ND_GSS | ND_GSSINTEGRITY);
250 		} else if (credflavor == RPCSEC_GSS_KRB5P) {
251 			nd.nd_flag |= (ND_GSS | ND_GSSPRIVACY);
252 		} else if (credflavor != AUTH_SYS) {
253 			svcerr_weakauth(rqst);
254 			svc_freereq(rqst);
255 			m_freem(nd.nd_mrep);
256 			goto out;
257 		}
258 
259 		/* Acquire the principal name for the RPCSEC_GSS cases. */
260 		if ((nd.nd_flag & (ND_NFSV4 | ND_GSS)) == (ND_NFSV4 | ND_GSS)) {
261 			rcredp = NULL;
262 			rpc_gss_getcred_call(rqst, &rcredp, NULL, NULL);
263 			/*
264 			 * The exported principal name consists of:
265 			 * - TOK_ID bytes with value of 4 and 1.
266 			 * - 2 byte mech length
267 			 * - mech
268 			 * - 4 byte principal name length
269 			 * - principal name
270 			 * A call to gss_import_name() would be an
271 			 * upcall to the gssd, so parse it here.
272 			 * See lib/libgssapi/gss_import_name.c for the
273 			 * above format.
274 			 */
275 			if (rcredp != NULL &&
276 			    rcredp->client_principal->len > 4 &&
277 			    rcredp->client_principal->name[0] == 4 &&
278 			    rcredp->client_principal->name[1] == 1) {
279 				/* Skip over the mech. */
280 				p = &rcredp->client_principal->name[2];
281 				i = (p[0] << 8) | p[1];
282 				p += i + 2;
283 				i += 4;
284 				/*
285 				 * Set "j" to a bogus length so that the
286 				 * "i + j" check will fail unless the below
287 				 * code sets "j" correctly.
288 				 */
289 				j = rcredp->client_principal->len;
290 				if (rcredp->client_principal->len > i + 4) {
291 					j = (p[0] << 24) | (p[1] << 16) |
292 					    (p[2] << 8) | p[3];
293 					i += 4;
294 					p += 4;
295 				}
296 				if (i + j == rcredp->client_principal->len) {
297 					nd.nd_principal = malloc(j + 1, M_TEMP,
298 					    M_WAITOK);
299 					nd.nd_princlen = j;
300 					memcpy(nd.nd_principal, p, j);
301 					nd.nd_principal[j] = '\0';
302 					NFSD_DEBUG(1, "nfssvc_program: "
303 					    "principal=%s\n", nd.nd_principal);
304 				}
305 			}
306 			if (nd.nd_princlen == 0)
307 				printf("nfssvc_program: cannot get RPCSEC_GSS "
308 				    "principal name\n");
309 		}
310 
311 		if ((xprt->xp_tls & RPCTLS_FLAGS_HANDSHAKE) != 0) {
312 			nd.nd_flag |= ND_TLS;
313 			if ((xprt->xp_tls & RPCTLS_FLAGS_VERIFIED) != 0)
314 				nd.nd_flag |= ND_TLSCERT;
315 			if ((xprt->xp_tls & RPCTLS_FLAGS_CERTUSER) != 0)
316 				nd.nd_flag |= ND_TLSCERTUSER;
317 		}
318 		nd.nd_maxextsiz = 16384;
319 #ifdef MAC
320 		mac_cred_associate_nfsd(nd.nd_cred);
321 #endif
322 		/*
323 		 * Get a refcnt (shared lock) on nfsd_suspend_lock.
324 		 * NFSSVC_SUSPENDNFSD will take an exclusive lock on
325 		 * nfsd_suspend_lock to suspend these threads.
326 		 * The call to nfsv4_lock() that precedes nfsv4_getref()
327 		 * ensures that the acquisition of the exclusive lock
328 		 * takes priority over acquisition of the shared lock by
329 		 * waiting for any exclusive lock request to complete.
330 		 * This must be done here, before the check of
331 		 * nfsv4root exports by nfsvno_v4rootexport().
332 		 */
333 		NFSLOCKV4ROOTMUTEX();
334 		nfsv4_lock(&NFSD_VNET(nfsd_suspend_lock), 0, NULL,
335 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
336 		nfsv4_getref(&NFSD_VNET(nfsd_suspend_lock), NULL,
337 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
338 		NFSUNLOCKV4ROOTMUTEX();
339 
340 		if ((nd.nd_flag & ND_NFSV4) != 0) {
341 			nd.nd_repstat = nfsvno_v4rootexport(&nd);
342 			if (nd.nd_repstat != 0) {
343 				NFSLOCKV4ROOTMUTEX();
344 				nfsv4_relref(&NFSD_VNET(nfsd_suspend_lock));
345 				NFSUNLOCKV4ROOTMUTEX();
346 				svcerr_weakauth(rqst);
347 				svc_freereq(rqst);
348 				m_freem(nd.nd_mrep);
349 				goto out;
350 			}
351 		}
352 
353 #ifdef KERN_TLS
354 		if ((xprt->xp_tls & RPCTLS_FLAGS_HANDSHAKE) != 0 &&
355 		    rpctls_getinfo(&maxlen, false, false))
356 			nd.nd_maxextsiz = maxlen;
357 #endif
358 		cacherep = nfs_proc(&nd, rqst->rq_xid, xprt, &rp);
359 		NFSLOCKV4ROOTMUTEX();
360 		nfsv4_relref(&NFSD_VNET(nfsd_suspend_lock));
361 		NFSUNLOCKV4ROOTMUTEX();
362 	} else {
363 		NFSMGET(nd.nd_mreq);
364 		nd.nd_mreq->m_len = 0;
365 		cacherep = RC_REPLY;
366 	}
367 	if (nd.nd_mrep != NULL)
368 		m_freem(nd.nd_mrep);
369 
370 	if (nd.nd_cred != NULL)
371 		crfree(nd.nd_cred);
372 
373 	if (cacherep == RC_DROPIT) {
374 		if (nd.nd_mreq != NULL)
375 			m_freem(nd.nd_mreq);
376 		svc_freereq(rqst);
377 		goto out;
378 	}
379 
380 	if (nd.nd_mreq == NULL) {
381 		svcerr_decode(rqst);
382 		svc_freereq(rqst);
383 		goto out;
384 	}
385 
386 	if (nd.nd_repstat & NFSERR_AUTHERR) {
387 		svcerr_auth(rqst, nd.nd_repstat & ~NFSERR_AUTHERR);
388 		if (nd.nd_mreq != NULL)
389 			m_freem(nd.nd_mreq);
390 	} else if (!svc_sendreply_mbuf(rqst, nd.nd_mreq)) {
391 		svcerr_systemerr(rqst);
392 	}
393 	if (rp != NULL) {
394 		nfsrvd_sentcache(rp, (rqst->rq_reply_seq != 0 ||
395 		    SVC_ACK(xprt, NULL)), rqst->rq_reply_seq);
396 	}
397 	svc_freereq(rqst);
398 
399 out:
400 	free(nd.nd_principal, M_TEMP);
401 	NFSD_CURVNET_RESTORE();
402 	ast_kclear(curthread);
403 	NFSEXITCODE(0);
404 }
405 
406 /*
407  * Check the cache and, optionally, do the RPC.
408  * Return the appropriate cache response.
409  */
410 static int
411 nfs_proc(struct nfsrv_descript *nd, u_int32_t xid, SVCXPRT *xprt,
412     struct nfsrvcache **rpp)
413 {
414 	int cacherep = RC_DOIT, isdgram, taglen = -1;
415 	struct mbuf *m;
416 	u_char tag[NFSV4_SMALLSTR + 1], *tagstr = NULL;
417 	u_int32_t minorvers = 0;
418 	uint32_t ack;
419 
420 	*rpp = NULL;
421 	if (nd->nd_nam2 == NULL) {
422 		nd->nd_flag |= ND_STREAMSOCK;
423 		isdgram = 0;
424 	} else {
425 		isdgram = 1;
426 	}
427 
428 	/*
429 	 * Two cases:
430 	 * 1 - For NFSv2 over UDP, if we are near our malloc/mget
431 	 *     limit, just drop the request. There is no
432 	 *     NFSERR_RESOURCE or NFSERR_DELAY for NFSv2 and the
433 	 *     client will timeout/retry over UDP in a little while.
434 	 * 2 - nd_repstat == 0 && nd_mreq == NULL, which
435 	 *     means a normal nfs rpc, so check the cache
436 	 */
437 	if ((nd->nd_flag & ND_NFSV2) && nd->nd_nam2 != NULL &&
438 	    nfsrv_mallocmget_limit()) {
439 		cacherep = RC_DROPIT;
440 	} else {
441 		/*
442 		 * For NFSv3, play it safe and assume that the client is
443 		 * doing retries on the same TCP connection.
444 		 */
445 		if ((nd->nd_flag & (ND_NFSV4 | ND_STREAMSOCK)) ==
446 		    ND_STREAMSOCK)
447 			nd->nd_flag |= ND_SAMETCPCONN;
448 		nd->nd_retxid = xid;
449 		nd->nd_tcpconntime = NFSD_MONOSEC;
450 		nd->nd_sockref = xprt->xp_sockref;
451 		if ((nd->nd_flag & ND_NFSV4) != 0)
452 			nfsd_getminorvers(nd, tag, &tagstr, &taglen,
453 			    &minorvers);
454 		if ((nd->nd_flag & ND_NFSV41) != 0)
455 			/* NFSv4.1 caches replies in the session slots. */
456 			cacherep = RC_DOIT;
457 		else {
458 			cacherep = nfsrvd_getcache(nd);
459 			ack = 0;
460 			SVC_ACK(xprt, &ack);
461 			nfsrc_trimcache(xprt->xp_sockref, ack, 0);
462 		}
463 	}
464 
465 	/*
466 	 * Handle the request. There are three cases.
467 	 * RC_DOIT - do the RPC
468 	 * RC_REPLY - return the reply already created
469 	 * RC_DROPIT - just throw the request away
470 	 */
471 	if (cacherep == RC_DOIT) {
472 		if ((nd->nd_flag & ND_NFSV41) != 0)
473 			nd->nd_xprt = xprt;
474 		nfsrvd_dorpc(nd, isdgram, tagstr, taglen, minorvers);
475 		if ((nd->nd_flag & ND_NFSV41) != 0) {
476 			if (nd->nd_repstat != NFSERR_REPLYFROMCACHE &&
477 			    (nd->nd_flag & ND_SAVEREPLY) != 0) {
478 				/* Cache a copy of the reply. */
479 				m = m_copym(nd->nd_mreq, 0, M_COPYALL,
480 				    M_WAITOK);
481 			} else
482 				m = NULL;
483 			if ((nd->nd_flag & ND_HASSEQUENCE) != 0)
484 				nfsrv_cache_session(nd, &m);
485 			if (nd->nd_repstat == NFSERR_REPLYFROMCACHE) {
486 				nd->nd_repstat = 0;
487 				if (m != NULL) {
488 					m_freem(nd->nd_mreq);
489 					nd->nd_mreq = m;
490 				}
491 			}
492 			cacherep = RC_REPLY;
493 		} else {
494 			if (nd->nd_repstat == NFSERR_DONTREPLY)
495 				cacherep = RC_DROPIT;
496 			else
497 				cacherep = RC_REPLY;
498 			*rpp = nfsrvd_updatecache(nd);
499 		}
500 	}
501 	if (tagstr != NULL && taglen > NFSV4_SMALLSTR)
502 		free(tagstr, M_TEMP);
503 
504 	NFSEXITCODE2(0, nd);
505 	return (cacherep);
506 }
507 
508 static void
509 nfssvc_loss(SVCXPRT *xprt)
510 {
511 	uint32_t ack;
512 
513 	ack = 0;
514 	SVC_ACK(xprt, &ack);
515 	NFSD_CURVNET_SET(NFSD_TD_TO_VNET(curthread));
516 	nfsrc_trimcache(xprt->xp_sockref, ack, 1);
517 	NFSD_CURVNET_RESTORE();
518 }
519 
520 /*
521  * Adds a socket to the list for servicing by nfsds.
522  */
523 int
524 nfsrvd_addsock(struct file *fp)
525 {
526 	int siz;
527 	struct socket *so;
528 	int error = 0;
529 	SVCXPRT *xprt;
530 	static u_int64_t sockref = 0;
531 
532 	so = fp->f_data;
533 
534 	siz = sb_max_adj;
535 	error = soreserve(so, siz, siz);
536 	if (error)
537 		goto out;
538 
539 	/*
540 	 * Steal the socket from userland so that it doesn't close
541 	 * unexpectedly.
542 	 */
543 	if (so->so_type == SOCK_DGRAM)
544 		xprt = svc_dg_create(NFSD_VNET(nfsrvd_pool), so, 0, 0);
545 	else
546 		xprt = svc_vc_create(NFSD_VNET(nfsrvd_pool), so, 0, 0);
547 	if (xprt) {
548 		fp->f_ops = &badfileops;
549 		fp->f_data = NULL;
550 		xprt->xp_sockref = ++sockref;
551 		if (NFSD_VNET(nfs_minvers) == NFS_VER2)
552 			svc_reg(xprt, NFS_PROG, NFS_VER2, nfssvc_program,
553 			    NULL);
554 		if (NFSD_VNET(nfs_minvers) <= NFS_VER3 &&
555 		    NFSD_VNET(nfs_maxvers) >= NFS_VER3)
556 			svc_reg(xprt, NFS_PROG, NFS_VER3, nfssvc_program,
557 			    NULL);
558 		if (NFSD_VNET(nfs_maxvers) >= NFS_VER4)
559 			svc_reg(xprt, NFS_PROG, NFS_VER4, nfssvc_program,
560 			    NULL);
561 		if (so->so_type == SOCK_STREAM)
562 			svc_loss_reg(xprt, nfssvc_loss);
563 		SVC_RELEASE(xprt);
564 	} else
565 		error = EPERM;
566 
567 out:
568 	NFSEXITCODE(error);
569 	return (error);
570 }
571 
572 /*
573  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
574  * until it is killed by a signal.
575  */
576 int
577 nfsrvd_nfsd(struct thread *td, struct nfsd_nfsd_args *args)
578 {
579 	char principal[MAXHOSTNAMELEN + 5];
580 	struct proc *p;
581 	int error = 0;
582 	bool_t ret2, ret3, ret4;
583 
584 	error = copyinstr(args->principal, principal, sizeof (principal),
585 	    NULL);
586 	if (error)
587 		goto out;
588 
589 	/*
590 	 * Only the first nfsd actually does any work. The RPC code
591 	 * adds threads to it as needed. Any extra processes offered
592 	 * by nfsd just exit. If nfsd is new enough, it will call us
593 	 * once with a structure that specifies how many threads to
594 	 * use.
595 	 */
596 	NFSD_LOCK();
597 	if (NFSD_VNET(nfsrv_numnfsd) == 0) {
598 		nfsdev_time = time_second;
599 		p = td->td_proc;
600 		PROC_LOCK(p);
601 		p->p_flag2 |= P2_AST_SU;
602 		PROC_UNLOCK(p);
603 		newnfs_numnfsd++;	/* Total num for all vnets. */
604 		NFSD_VNET(nfsrv_numnfsd)++;	/* Num for this vnet. */
605 
606 		NFSD_UNLOCK();
607 		error = nfsrv_createdevids(args, td);
608 		if (error == 0) {
609 			/* An empty string implies AUTH_SYS only. */
610 			if (principal[0] != '\0') {
611 				ret2 = rpc_gss_set_svc_name_call(principal,
612 				    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG,
613 				    NFS_VER2);
614 				ret3 = rpc_gss_set_svc_name_call(principal,
615 				    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG,
616 				    NFS_VER3);
617 				ret4 = rpc_gss_set_svc_name_call(principal,
618 				    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG,
619 				    NFS_VER4);
620 
621 				if (!ret2 || !ret3 || !ret4)
622 					printf("nfsd: can't register svc "
623 					    "name %s jid:%d\n", principal,
624 					    td->td_ucred->cr_prison->pr_id);
625 			}
626 
627 			NFSD_VNET(nfsrvd_pool)->sp_minthreads =
628 			    args->minthreads;
629 			NFSD_VNET(nfsrvd_pool)->sp_maxthreads =
630 			    args->maxthreads;
631 
632 			/*
633 			 * If this is a pNFS service, make Getattr do a
634 			 * vn_start_write(), so it can do a vn_set_extattr().
635 			 */
636 			if (nfsrv_devidcnt > 0) {
637 				nfsrv_writerpc[NFSPROC_GETATTR] = 1;
638 				nfsv4_opflag[NFSV4OP_GETATTR].modifyfs = 1;
639 			}
640 
641 			svc_run(NFSD_VNET(nfsrvd_pool));
642 
643 			/* Reset Getattr to not do a vn_start_write(). */
644 			nfsrv_writerpc[NFSPROC_GETATTR] = 0;
645 			nfsv4_opflag[NFSV4OP_GETATTR].modifyfs = 0;
646 
647 			if (principal[0] != '\0') {
648 				rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER2);
649 				rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER3);
650 				rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER4);
651 			}
652 		}
653 		NFSD_LOCK();
654 		newnfs_numnfsd--;
655 		NFSD_VNET(nfsrv_numnfsd)--;
656 		nfsrvd_init(1);
657 		PROC_LOCK(p);
658 		p->p_flag2 &= ~P2_AST_SU;
659 		PROC_UNLOCK(p);
660 	}
661 	NFSD_UNLOCK();
662 
663 out:
664 	NFSEXITCODE(error);
665 	return (error);
666 }
667 
668 /*
669  * Initialize the data structures for the server.
670  * Handshake with any new nfsds starting up to avoid any chance of
671  * corruption.
672  */
673 void
674 nfsrvd_init(int terminating)
675 {
676 
677 	NFSD_LOCK_ASSERT();
678 
679 	if (terminating) {
680 		NFSD_VNET(nfsd_master_proc) = NULL;
681 		NFSD_UNLOCK();
682 		nfsrv_freealllayoutsanddevids();
683 		nfsrv_freeallbackchannel_xprts();
684 		svcpool_close(NFSD_VNET(nfsrvd_pool));
685 		free(nfsrv_zeropnfsdat, M_TEMP);
686 		nfsrv_zeropnfsdat = NULL;
687 		NFSD_LOCK();
688 	} else {
689 		/* Initialize per-vnet globals once per vnet. */
690 		if (NFSD_VNET(nfsrvd_inited))
691 			return;
692 		NFSD_VNET(nfsrvd_inited) = true;
693 		NFSD_UNLOCK();
694 		NFSD_VNET(nfsrvd_pool) = svcpool_create("nfsd",
695 		    SYSCTL_STATIC_CHILDREN(_vfs_nfsd));
696 		NFSD_VNET(nfsrvd_pool)->sp_rcache = NULL;
697 		NFSD_VNET(nfsrvd_pool)->sp_assign = fhanew_assign;
698 		NFSD_VNET(nfsrvd_pool)->sp_done = fhanew_nd_complete;
699 		NFSD_LOCK();
700 	}
701 }
702