xref: /freebsd/sys/fs/nfsserver/nfs_nfsdkrpc.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet6.h"
38 #include "opt_kgssapi.h"
39 
40 #include <fs/nfs/nfsport.h>
41 
42 #include <rpc/rpc.h>
43 #include <rpc/rpcsec_gss.h>
44 
45 #include <nfs/nfs_fha.h>
46 #include <fs/nfsserver/nfs_fha_new.h>
47 
48 #include <security/mac/mac_framework.h>
49 
50 NFSDLOCKMUTEX;
51 NFSV4ROOTLOCKMUTEX;
52 struct nfsv4lock nfsd_suspend_lock;
53 
54 /*
55  * Mapping of old NFS Version 2 RPC numbers to generic numbers.
56  */
57 int newnfs_nfsv3_procid[NFS_V3NPROCS] = {
58 	NFSPROC_NULL,
59 	NFSPROC_GETATTR,
60 	NFSPROC_SETATTR,
61 	NFSPROC_NOOP,
62 	NFSPROC_LOOKUP,
63 	NFSPROC_READLINK,
64 	NFSPROC_READ,
65 	NFSPROC_NOOP,
66 	NFSPROC_WRITE,
67 	NFSPROC_CREATE,
68 	NFSPROC_REMOVE,
69 	NFSPROC_RENAME,
70 	NFSPROC_LINK,
71 	NFSPROC_SYMLINK,
72 	NFSPROC_MKDIR,
73 	NFSPROC_RMDIR,
74 	NFSPROC_READDIR,
75 	NFSPROC_FSSTAT,
76 	NFSPROC_NOOP,
77 	NFSPROC_NOOP,
78 	NFSPROC_NOOP,
79 	NFSPROC_NOOP,
80 };
81 
82 
83 SYSCTL_DECL(_vfs_nfsd);
84 
85 SVCPOOL		*nfsrvd_pool;
86 
87 static int	nfs_privport = 0;
88 SYSCTL_INT(_vfs_nfsd, OID_AUTO, nfs_privport, CTLFLAG_RWTUN,
89     &nfs_privport, 0,
90     "Only allow clients using a privileged port for NFSv2 and 3");
91 
92 static int	nfs_minvers = NFS_VER2;
93 SYSCTL_INT(_vfs_nfsd, OID_AUTO, server_min_nfsvers, CTLFLAG_RWTUN,
94     &nfs_minvers, 0, "The lowest version of NFS handled by the server");
95 
96 static int	nfs_maxvers = NFS_VER4;
97 SYSCTL_INT(_vfs_nfsd, OID_AUTO, server_max_nfsvers, CTLFLAG_RWTUN,
98     &nfs_maxvers, 0, "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 struct proc *nfsd_master_proc;
106 
107 /*
108  * NFS server system calls
109  */
110 
111 static void
112 nfssvc_program(struct svc_req *rqst, SVCXPRT *xprt)
113 {
114 	struct nfsrv_descript nd;
115 	struct nfsrvcache *rp = NULL;
116 	int cacherep, credflavor;
117 
118 	memset(&nd, 0, sizeof(nd));
119 	if (rqst->rq_vers == NFS_VER2) {
120 		if (rqst->rq_proc > NFSV2PROC_STATFS) {
121 			svcerr_noproc(rqst);
122 			svc_freereq(rqst);
123 			goto out;
124 		}
125 		nd.nd_procnum = newnfs_nfsv3_procid[rqst->rq_proc];
126 		nd.nd_flag = ND_NFSV2;
127 	} else if (rqst->rq_vers == NFS_VER3) {
128 		if (rqst->rq_proc >= NFS_V3NPROCS) {
129 			svcerr_noproc(rqst);
130 			svc_freereq(rqst);
131 			goto out;
132 		}
133 		nd.nd_procnum = rqst->rq_proc;
134 		nd.nd_flag = ND_NFSV3;
135 	} else {
136 		if (rqst->rq_proc != NFSPROC_NULL &&
137 		    rqst->rq_proc != NFSV4PROC_COMPOUND) {
138 			svcerr_noproc(rqst);
139 			svc_freereq(rqst);
140 			goto out;
141 		}
142 		nd.nd_procnum = rqst->rq_proc;
143 		nd.nd_flag = ND_NFSV4;
144 	}
145 
146 	/*
147 	 * Note: we want rq_addr, not svc_getrpccaller for nd_nam2 -
148 	 * NFS_SRVMAXDATA uses a NULL value for nd_nam2 to detect TCP
149 	 * mounts.
150 	 */
151 	nd.nd_mrep = rqst->rq_args;
152 	rqst->rq_args = NULL;
153 	newnfs_realign(&nd.nd_mrep, M_WAITOK);
154 	nd.nd_md = nd.nd_mrep;
155 	nd.nd_dpos = mtod(nd.nd_md, caddr_t);
156 	nd.nd_nam = svc_getrpccaller(rqst);
157 	nd.nd_nam2 = rqst->rq_addr;
158 	nd.nd_mreq = NULL;
159 	nd.nd_cred = NULL;
160 
161 	if (nfs_privport && (nd.nd_flag & ND_NFSV4) == 0) {
162 		/* Check if source port is privileged */
163 		u_short port;
164 		struct sockaddr *nam = nd.nd_nam;
165 		struct sockaddr_in *sin;
166 
167 		sin = (struct sockaddr_in *)nam;
168 		/*
169 		 * INET/INET6 - same code:
170 		 *    sin_port and sin6_port are at same offset
171 		 */
172 		port = ntohs(sin->sin_port);
173 		if (port >= IPPORT_RESERVED &&
174 		    nd.nd_procnum != NFSPROC_NULL) {
175 #ifdef INET6
176 			char b6[INET6_ADDRSTRLEN];
177 #if defined(KLD_MODULE)
178 			/* Do not use ip6_sprintf: the nfs module should work without INET6. */
179 #define	ip6_sprintf(buf, a)						\
180 			(sprintf((buf), "%x:%x:%x:%x:%x:%x:%x:%x",	\
181 			    (a)->s6_addr16[0], (a)->s6_addr16[1],	\
182 			    (a)->s6_addr16[2], (a)->s6_addr16[3],	\
183 			    (a)->s6_addr16[4], (a)->s6_addr16[5],	\
184 			    (a)->s6_addr16[6], (a)->s6_addr16[7]),	\
185 			    (buf))
186 #endif
187 #endif
188 			printf("NFS request from unprivileged port (%s:%d)\n",
189 #ifdef INET6
190 			    sin->sin_family == AF_INET6 ?
191 			    ip6_sprintf(b6, &satosin6(sin)->sin6_addr) :
192 #if defined(KLD_MODULE)
193 #undef ip6_sprintf
194 #endif
195 #endif
196 			    inet_ntoa(sin->sin_addr), port);
197 			svcerr_weakauth(rqst);
198 			svc_freereq(rqst);
199 			m_freem(nd.nd_mrep);
200 			goto out;
201 		}
202 	}
203 
204 	if (nd.nd_procnum != NFSPROC_NULL) {
205 		if (!svc_getcred(rqst, &nd.nd_cred, &credflavor)) {
206 			svcerr_weakauth(rqst);
207 			svc_freereq(rqst);
208 			m_freem(nd.nd_mrep);
209 			goto out;
210 		}
211 
212 		/* Set the flag based on credflavor */
213 		if (credflavor == RPCSEC_GSS_KRB5) {
214 			nd.nd_flag |= ND_GSS;
215 		} else if (credflavor == RPCSEC_GSS_KRB5I) {
216 			nd.nd_flag |= (ND_GSS | ND_GSSINTEGRITY);
217 		} else if (credflavor == RPCSEC_GSS_KRB5P) {
218 			nd.nd_flag |= (ND_GSS | ND_GSSPRIVACY);
219 		} else if (credflavor != AUTH_SYS) {
220 			svcerr_weakauth(rqst);
221 			svc_freereq(rqst);
222 			m_freem(nd.nd_mrep);
223 			goto out;
224 		}
225 
226 #ifdef MAC
227 		mac_cred_associate_nfsd(nd.nd_cred);
228 #endif
229 		/*
230 		 * Get a refcnt (shared lock) on nfsd_suspend_lock.
231 		 * NFSSVC_SUSPENDNFSD will take an exclusive lock on
232 		 * nfsd_suspend_lock to suspend these threads.
233 		 * This must be done here, before the check of
234 		 * nfsv4root exports by nfsvno_v4rootexport().
235 		 */
236 		NFSLOCKV4ROOTMUTEX();
237 		nfsv4_getref(&nfsd_suspend_lock, NULL, NFSV4ROOTLOCKMUTEXPTR,
238 		    NULL);
239 		NFSUNLOCKV4ROOTMUTEX();
240 
241 		if ((nd.nd_flag & ND_NFSV4) != 0) {
242 			nd.nd_repstat = nfsvno_v4rootexport(&nd);
243 			if (nd.nd_repstat != 0) {
244 				NFSLOCKV4ROOTMUTEX();
245 				nfsv4_relref(&nfsd_suspend_lock);
246 				NFSUNLOCKV4ROOTMUTEX();
247 				svcerr_weakauth(rqst);
248 				svc_freereq(rqst);
249 				m_freem(nd.nd_mrep);
250 				goto out;
251 			}
252 		}
253 
254 		cacherep = nfs_proc(&nd, rqst->rq_xid, xprt, &rp);
255 		NFSLOCKV4ROOTMUTEX();
256 		nfsv4_relref(&nfsd_suspend_lock);
257 		NFSUNLOCKV4ROOTMUTEX();
258 	} else {
259 		NFSMGET(nd.nd_mreq);
260 		nd.nd_mreq->m_len = 0;
261 		cacherep = RC_REPLY;
262 	}
263 	if (nd.nd_mrep != NULL)
264 		m_freem(nd.nd_mrep);
265 
266 	if (nd.nd_cred != NULL)
267 		crfree(nd.nd_cred);
268 
269 	if (cacherep == RC_DROPIT) {
270 		if (nd.nd_mreq != NULL)
271 			m_freem(nd.nd_mreq);
272 		svc_freereq(rqst);
273 		goto out;
274 	}
275 
276 	if (nd.nd_mreq == NULL) {
277 		svcerr_decode(rqst);
278 		svc_freereq(rqst);
279 		goto out;
280 	}
281 
282 	if (nd.nd_repstat & NFSERR_AUTHERR) {
283 		svcerr_auth(rqst, nd.nd_repstat & ~NFSERR_AUTHERR);
284 		if (nd.nd_mreq != NULL)
285 			m_freem(nd.nd_mreq);
286 	} else if (!svc_sendreply_mbuf(rqst, nd.nd_mreq)) {
287 		svcerr_systemerr(rqst);
288 	}
289 	if (rp != NULL) {
290 		nfsrvd_sentcache(rp, (rqst->rq_reply_seq != 0 ||
291 		    SVC_ACK(xprt, NULL)), rqst->rq_reply_seq);
292 	}
293 	svc_freereq(rqst);
294 
295 out:
296 	NFSEXITCODE(0);
297 }
298 
299 /*
300  * Check the cache and, optionally, do the RPC.
301  * Return the appropriate cache response.
302  */
303 static int
304 nfs_proc(struct nfsrv_descript *nd, u_int32_t xid, SVCXPRT *xprt,
305     struct nfsrvcache **rpp)
306 {
307 	struct thread *td = curthread;
308 	int cacherep = RC_DOIT, isdgram, taglen = -1;
309 	struct mbuf *m;
310 	u_char tag[NFSV4_SMALLSTR + 1], *tagstr = NULL;
311 	u_int32_t minorvers = 0;
312 	uint32_t ack;
313 
314 	*rpp = NULL;
315 	if (nd->nd_nam2 == NULL) {
316 		nd->nd_flag |= ND_STREAMSOCK;
317 		isdgram = 0;
318 	} else {
319 		isdgram = 1;
320 	}
321 
322 	/*
323 	 * Two cases:
324 	 * 1 - For NFSv2 over UDP, if we are near our malloc/mget
325 	 *     limit, just drop the request. There is no
326 	 *     NFSERR_RESOURCE or NFSERR_DELAY for NFSv2 and the
327 	 *     client will timeout/retry over UDP in a little while.
328 	 * 2 - nd_repstat == 0 && nd_mreq == NULL, which
329 	 *     means a normal nfs rpc, so check the cache
330 	 */
331 	if ((nd->nd_flag & ND_NFSV2) && nd->nd_nam2 != NULL &&
332 	    nfsrv_mallocmget_limit()) {
333 		cacherep = RC_DROPIT;
334 	} else {
335 		/*
336 		 * For NFSv3, play it safe and assume that the client is
337 		 * doing retries on the same TCP connection.
338 		 */
339 		if ((nd->nd_flag & (ND_NFSV4 | ND_STREAMSOCK)) ==
340 		    ND_STREAMSOCK)
341 			nd->nd_flag |= ND_SAMETCPCONN;
342 		nd->nd_retxid = xid;
343 		nd->nd_tcpconntime = NFSD_MONOSEC;
344 		nd->nd_sockref = xprt->xp_sockref;
345 		if ((nd->nd_flag & ND_NFSV4) != 0)
346 			nfsd_getminorvers(nd, tag, &tagstr, &taglen,
347 			    &minorvers);
348 		if ((nd->nd_flag & ND_NFSV41) != 0)
349 			/* NFSv4.1 caches replies in the session slots. */
350 			cacherep = RC_DOIT;
351 		else {
352 			cacherep = nfsrvd_getcache(nd);
353 			ack = 0;
354 			SVC_ACK(xprt, &ack);
355 			nfsrc_trimcache(xprt->xp_sockref, ack, 0);
356 		}
357 	}
358 
359 	/*
360 	 * Handle the request. There are three cases.
361 	 * RC_DOIT - do the RPC
362 	 * RC_REPLY - return the reply already created
363 	 * RC_DROPIT - just throw the request away
364 	 */
365 	if (cacherep == RC_DOIT) {
366 		if ((nd->nd_flag & ND_NFSV41) != 0)
367 			nd->nd_xprt = xprt;
368 		nfsrvd_dorpc(nd, isdgram, tagstr, taglen, minorvers, td);
369 		if ((nd->nd_flag & ND_NFSV41) != 0) {
370 			if (nd->nd_repstat != NFSERR_REPLYFROMCACHE &&
371 			    (nd->nd_flag & ND_SAVEREPLY) != 0) {
372 				/* Cache a copy of the reply. */
373 				m = m_copym(nd->nd_mreq, 0, M_COPYALL,
374 				    M_WAITOK);
375 			} else
376 				m = NULL;
377 			if ((nd->nd_flag & ND_HASSEQUENCE) != 0)
378 				nfsrv_cache_session(nd->nd_sessionid,
379 				    nd->nd_slotid, nd->nd_repstat, &m);
380 			if (nd->nd_repstat == NFSERR_REPLYFROMCACHE)
381 				nd->nd_repstat = 0;
382 			cacherep = RC_REPLY;
383 		} else {
384 			if (nd->nd_repstat == NFSERR_DONTREPLY)
385 				cacherep = RC_DROPIT;
386 			else
387 				cacherep = RC_REPLY;
388 			*rpp = nfsrvd_updatecache(nd);
389 		}
390 	}
391 	if (tagstr != NULL && taglen > NFSV4_SMALLSTR)
392 		free(tagstr, M_TEMP);
393 
394 	NFSEXITCODE2(0, nd);
395 	return (cacherep);
396 }
397 
398 static void
399 nfssvc_loss(SVCXPRT *xprt)
400 {
401 	uint32_t ack;
402 
403 	ack = 0;
404 	SVC_ACK(xprt, &ack);
405 	nfsrc_trimcache(xprt->xp_sockref, ack, 1);
406 }
407 
408 /*
409  * Adds a socket to the list for servicing by nfsds.
410  */
411 int
412 nfsrvd_addsock(struct file *fp)
413 {
414 	int siz;
415 	struct socket *so;
416 	int error = 0;
417 	SVCXPRT *xprt;
418 	static u_int64_t sockref = 0;
419 
420 	so = fp->f_data;
421 
422 	siz = sb_max_adj;
423 	error = soreserve(so, siz, siz);
424 	if (error)
425 		goto out;
426 
427 	/*
428 	 * Steal the socket from userland so that it doesn't close
429 	 * unexpectedly.
430 	 */
431 	if (so->so_type == SOCK_DGRAM)
432 		xprt = svc_dg_create(nfsrvd_pool, so, 0, 0);
433 	else
434 		xprt = svc_vc_create(nfsrvd_pool, so, 0, 0);
435 	if (xprt) {
436 		fp->f_ops = &badfileops;
437 		fp->f_data = NULL;
438 		xprt->xp_sockref = ++sockref;
439 		if (nfs_minvers == NFS_VER2)
440 			svc_reg(xprt, NFS_PROG, NFS_VER2, nfssvc_program,
441 			    NULL);
442 		if (nfs_minvers <= NFS_VER3 && nfs_maxvers >= NFS_VER3)
443 			svc_reg(xprt, NFS_PROG, NFS_VER3, nfssvc_program,
444 			    NULL);
445 		if (nfs_maxvers >= NFS_VER4)
446 			svc_reg(xprt, NFS_PROG, NFS_VER4, nfssvc_program,
447 			    NULL);
448 		if (so->so_type == SOCK_STREAM)
449 			svc_loss_reg(xprt, nfssvc_loss);
450 		SVC_RELEASE(xprt);
451 	}
452 
453 out:
454 	NFSEXITCODE(error);
455 	return (error);
456 }
457 
458 /*
459  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
460  * until it is killed by a signal.
461  */
462 int
463 nfsrvd_nfsd(struct thread *td, struct nfsd_nfsd_args *args)
464 {
465 	char principal[MAXHOSTNAMELEN + 5];
466 	int error = 0;
467 	bool_t ret2, ret3, ret4;
468 
469 	error = copyinstr(args->principal, principal, sizeof (principal),
470 	    NULL);
471 	if (error)
472 		goto out;
473 
474 	/*
475 	 * Only the first nfsd actually does any work. The RPC code
476 	 * adds threads to it as needed. Any extra processes offered
477 	 * by nfsd just exit. If nfsd is new enough, it will call us
478 	 * once with a structure that specifies how many threads to
479 	 * use.
480 	 */
481 	NFSD_LOCK();
482 	if (newnfs_numnfsd == 0) {
483 		newnfs_numnfsd++;
484 
485 		NFSD_UNLOCK();
486 
487 		/* An empty string implies AUTH_SYS only. */
488 		if (principal[0] != '\0') {
489 			ret2 = rpc_gss_set_svc_name_call(principal,
490 			    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG, NFS_VER2);
491 			ret3 = rpc_gss_set_svc_name_call(principal,
492 			    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG, NFS_VER3);
493 			ret4 = rpc_gss_set_svc_name_call(principal,
494 			    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG, NFS_VER4);
495 
496 			if (!ret2 || !ret3 || !ret4)
497 				printf("nfsd: can't register svc name\n");
498 		}
499 
500 		nfsrvd_pool->sp_minthreads = args->minthreads;
501 		nfsrvd_pool->sp_maxthreads = args->maxthreads;
502 
503 		svc_run(nfsrvd_pool);
504 
505 		if (principal[0] != '\0') {
506 			rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER2);
507 			rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER3);
508 			rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER4);
509 		}
510 
511 		NFSD_LOCK();
512 		newnfs_numnfsd--;
513 		nfsrvd_init(1);
514 	}
515 	NFSD_UNLOCK();
516 
517 out:
518 	NFSEXITCODE(error);
519 	return (error);
520 }
521 
522 /*
523  * Initialize the data structures for the server.
524  * Handshake with any new nfsds starting up to avoid any chance of
525  * corruption.
526  */
527 void
528 nfsrvd_init(int terminating)
529 {
530 
531 	NFSD_LOCK_ASSERT();
532 
533 	if (terminating) {
534 		nfsd_master_proc = NULL;
535 		NFSD_UNLOCK();
536 		svcpool_destroy(nfsrvd_pool);
537 		nfsrvd_pool = NULL;
538 		NFSD_LOCK();
539 	}
540 
541 	NFSD_UNLOCK();
542 
543 	nfsrvd_pool = svcpool_create("nfsd", SYSCTL_STATIC_CHILDREN(_vfs_nfsd));
544 	nfsrvd_pool->sp_rcache = NULL;
545 	nfsrvd_pool->sp_assign = fhanew_assign;
546 	nfsrvd_pool->sp_done = fha_nd_complete;
547 
548 	NFSD_LOCK();
549 }
550 
551