xref: /freebsd/sys/fs/nfsserver/nfs_nfsdkrpc.c (revision f4b37ed0f8b307b1f3f0f630ca725d68f1dff30d)
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 		    newnfs_nfsv3_procid[rqst->rq_proc] == NFSPROC_NOOP) {
122 			svcerr_noproc(rqst);
123 			svc_freereq(rqst);
124 			goto out;
125 		}
126 		nd.nd_procnum = newnfs_nfsv3_procid[rqst->rq_proc];
127 		nd.nd_flag = ND_NFSV2;
128 	} else if (rqst->rq_vers == NFS_VER3) {
129 		if (rqst->rq_proc >= NFS_V3NPROCS) {
130 			svcerr_noproc(rqst);
131 			svc_freereq(rqst);
132 			goto out;
133 		}
134 		nd.nd_procnum = rqst->rq_proc;
135 		nd.nd_flag = ND_NFSV3;
136 	} else {
137 		if (rqst->rq_proc != NFSPROC_NULL &&
138 		    rqst->rq_proc != NFSV4PROC_COMPOUND) {
139 			svcerr_noproc(rqst);
140 			svc_freereq(rqst);
141 			goto out;
142 		}
143 		nd.nd_procnum = rqst->rq_proc;
144 		nd.nd_flag = ND_NFSV4;
145 	}
146 
147 	/*
148 	 * Note: we want rq_addr, not svc_getrpccaller for nd_nam2 -
149 	 * NFS_SRVMAXDATA uses a NULL value for nd_nam2 to detect TCP
150 	 * mounts.
151 	 */
152 	nd.nd_mrep = rqst->rq_args;
153 	rqst->rq_args = NULL;
154 	newnfs_realign(&nd.nd_mrep, M_WAITOK);
155 	nd.nd_md = nd.nd_mrep;
156 	nd.nd_dpos = mtod(nd.nd_md, caddr_t);
157 	nd.nd_nam = svc_getrpccaller(rqst);
158 	nd.nd_nam2 = rqst->rq_addr;
159 	nd.nd_mreq = NULL;
160 	nd.nd_cred = NULL;
161 
162 	if (nfs_privport && (nd.nd_flag & ND_NFSV4) == 0) {
163 		/* Check if source port is privileged */
164 		u_short port;
165 		struct sockaddr *nam = nd.nd_nam;
166 		struct sockaddr_in *sin;
167 
168 		sin = (struct sockaddr_in *)nam;
169 		/*
170 		 * INET/INET6 - same code:
171 		 *    sin_port and sin6_port are at same offset
172 		 */
173 		port = ntohs(sin->sin_port);
174 		if (port >= IPPORT_RESERVED &&
175 		    nd.nd_procnum != NFSPROC_NULL) {
176 #ifdef INET6
177 			char b6[INET6_ADDRSTRLEN];
178 #if defined(KLD_MODULE)
179 			/* Do not use ip6_sprintf: the nfs module should work without INET6. */
180 #define	ip6_sprintf(buf, a)						\
181 			(sprintf((buf), "%x:%x:%x:%x:%x:%x:%x:%x",	\
182 			    (a)->s6_addr16[0], (a)->s6_addr16[1],	\
183 			    (a)->s6_addr16[2], (a)->s6_addr16[3],	\
184 			    (a)->s6_addr16[4], (a)->s6_addr16[5],	\
185 			    (a)->s6_addr16[6], (a)->s6_addr16[7]),	\
186 			    (buf))
187 #endif
188 #endif
189 			printf("NFS request from unprivileged port (%s:%d)\n",
190 #ifdef INET6
191 			    sin->sin_family == AF_INET6 ?
192 			    ip6_sprintf(b6, &satosin6(sin)->sin6_addr) :
193 #if defined(KLD_MODULE)
194 #undef ip6_sprintf
195 #endif
196 #endif
197 			    inet_ntoa(sin->sin_addr), port);
198 			svcerr_weakauth(rqst);
199 			svc_freereq(rqst);
200 			m_freem(nd.nd_mrep);
201 			goto out;
202 		}
203 	}
204 
205 	if (nd.nd_procnum != NFSPROC_NULL) {
206 		if (!svc_getcred(rqst, &nd.nd_cred, &credflavor)) {
207 			svcerr_weakauth(rqst);
208 			svc_freereq(rqst);
209 			m_freem(nd.nd_mrep);
210 			goto out;
211 		}
212 
213 		/* Set the flag based on credflavor */
214 		if (credflavor == RPCSEC_GSS_KRB5) {
215 			nd.nd_flag |= ND_GSS;
216 		} else if (credflavor == RPCSEC_GSS_KRB5I) {
217 			nd.nd_flag |= (ND_GSS | ND_GSSINTEGRITY);
218 		} else if (credflavor == RPCSEC_GSS_KRB5P) {
219 			nd.nd_flag |= (ND_GSS | ND_GSSPRIVACY);
220 		} else if (credflavor != AUTH_SYS) {
221 			svcerr_weakauth(rqst);
222 			svc_freereq(rqst);
223 			m_freem(nd.nd_mrep);
224 			goto out;
225 		}
226 
227 #ifdef MAC
228 		mac_cred_associate_nfsd(nd.nd_cred);
229 #endif
230 		/*
231 		 * Get a refcnt (shared lock) on nfsd_suspend_lock.
232 		 * NFSSVC_SUSPENDNFSD will take an exclusive lock on
233 		 * nfsd_suspend_lock to suspend these threads.
234 		 * This must be done here, before the check of
235 		 * nfsv4root exports by nfsvno_v4rootexport().
236 		 */
237 		NFSLOCKV4ROOTMUTEX();
238 		nfsv4_getref(&nfsd_suspend_lock, NULL, NFSV4ROOTLOCKMUTEXPTR,
239 		    NULL);
240 		NFSUNLOCKV4ROOTMUTEX();
241 
242 		if ((nd.nd_flag & ND_NFSV4) != 0) {
243 			nd.nd_repstat = nfsvno_v4rootexport(&nd);
244 			if (nd.nd_repstat != 0) {
245 				NFSLOCKV4ROOTMUTEX();
246 				nfsv4_relref(&nfsd_suspend_lock);
247 				NFSUNLOCKV4ROOTMUTEX();
248 				svcerr_weakauth(rqst);
249 				svc_freereq(rqst);
250 				m_freem(nd.nd_mrep);
251 				goto out;
252 			}
253 		}
254 
255 		cacherep = nfs_proc(&nd, rqst->rq_xid, xprt, &rp);
256 		NFSLOCKV4ROOTMUTEX();
257 		nfsv4_relref(&nfsd_suspend_lock);
258 		NFSUNLOCKV4ROOTMUTEX();
259 	} else {
260 		NFSMGET(nd.nd_mreq);
261 		nd.nd_mreq->m_len = 0;
262 		cacherep = RC_REPLY;
263 	}
264 	if (nd.nd_mrep != NULL)
265 		m_freem(nd.nd_mrep);
266 
267 	if (nd.nd_cred != NULL)
268 		crfree(nd.nd_cred);
269 
270 	if (cacherep == RC_DROPIT) {
271 		if (nd.nd_mreq != NULL)
272 			m_freem(nd.nd_mreq);
273 		svc_freereq(rqst);
274 		goto out;
275 	}
276 
277 	if (nd.nd_mreq == NULL) {
278 		svcerr_decode(rqst);
279 		svc_freereq(rqst);
280 		goto out;
281 	}
282 
283 	if (nd.nd_repstat & NFSERR_AUTHERR) {
284 		svcerr_auth(rqst, nd.nd_repstat & ~NFSERR_AUTHERR);
285 		if (nd.nd_mreq != NULL)
286 			m_freem(nd.nd_mreq);
287 	} else if (!svc_sendreply_mbuf(rqst, nd.nd_mreq)) {
288 		svcerr_systemerr(rqst);
289 	}
290 	if (rp != NULL) {
291 		nfsrvd_sentcache(rp, (rqst->rq_reply_seq != 0 ||
292 		    SVC_ACK(xprt, NULL)), rqst->rq_reply_seq);
293 	}
294 	svc_freereq(rqst);
295 
296 out:
297 	if (softdep_ast_cleanup != NULL)
298 		softdep_ast_cleanup();
299 	NFSEXITCODE(0);
300 }
301 
302 /*
303  * Check the cache and, optionally, do the RPC.
304  * Return the appropriate cache response.
305  */
306 static int
307 nfs_proc(struct nfsrv_descript *nd, u_int32_t xid, SVCXPRT *xprt,
308     struct nfsrvcache **rpp)
309 {
310 	struct thread *td = curthread;
311 	int cacherep = RC_DOIT, isdgram, taglen = -1;
312 	struct mbuf *m;
313 	u_char tag[NFSV4_SMALLSTR + 1], *tagstr = NULL;
314 	u_int32_t minorvers = 0;
315 	uint32_t ack;
316 
317 	*rpp = NULL;
318 	if (nd->nd_nam2 == NULL) {
319 		nd->nd_flag |= ND_STREAMSOCK;
320 		isdgram = 0;
321 	} else {
322 		isdgram = 1;
323 	}
324 
325 	/*
326 	 * Two cases:
327 	 * 1 - For NFSv2 over UDP, if we are near our malloc/mget
328 	 *     limit, just drop the request. There is no
329 	 *     NFSERR_RESOURCE or NFSERR_DELAY for NFSv2 and the
330 	 *     client will timeout/retry over UDP in a little while.
331 	 * 2 - nd_repstat == 0 && nd_mreq == NULL, which
332 	 *     means a normal nfs rpc, so check the cache
333 	 */
334 	if ((nd->nd_flag & ND_NFSV2) && nd->nd_nam2 != NULL &&
335 	    nfsrv_mallocmget_limit()) {
336 		cacherep = RC_DROPIT;
337 	} else {
338 		/*
339 		 * For NFSv3, play it safe and assume that the client is
340 		 * doing retries on the same TCP connection.
341 		 */
342 		if ((nd->nd_flag & (ND_NFSV4 | ND_STREAMSOCK)) ==
343 		    ND_STREAMSOCK)
344 			nd->nd_flag |= ND_SAMETCPCONN;
345 		nd->nd_retxid = xid;
346 		nd->nd_tcpconntime = NFSD_MONOSEC;
347 		nd->nd_sockref = xprt->xp_sockref;
348 		if ((nd->nd_flag & ND_NFSV4) != 0)
349 			nfsd_getminorvers(nd, tag, &tagstr, &taglen,
350 			    &minorvers);
351 		if ((nd->nd_flag & ND_NFSV41) != 0)
352 			/* NFSv4.1 caches replies in the session slots. */
353 			cacherep = RC_DOIT;
354 		else {
355 			cacherep = nfsrvd_getcache(nd);
356 			ack = 0;
357 			SVC_ACK(xprt, &ack);
358 			nfsrc_trimcache(xprt->xp_sockref, ack, 0);
359 		}
360 	}
361 
362 	/*
363 	 * Handle the request. There are three cases.
364 	 * RC_DOIT - do the RPC
365 	 * RC_REPLY - return the reply already created
366 	 * RC_DROPIT - just throw the request away
367 	 */
368 	if (cacherep == RC_DOIT) {
369 		if ((nd->nd_flag & ND_NFSV41) != 0)
370 			nd->nd_xprt = xprt;
371 		nfsrvd_dorpc(nd, isdgram, tagstr, taglen, minorvers, td);
372 		if ((nd->nd_flag & ND_NFSV41) != 0) {
373 			if (nd->nd_repstat != NFSERR_REPLYFROMCACHE &&
374 			    (nd->nd_flag & ND_SAVEREPLY) != 0) {
375 				/* Cache a copy of the reply. */
376 				m = m_copym(nd->nd_mreq, 0, M_COPYALL,
377 				    M_WAITOK);
378 			} else
379 				m = NULL;
380 			if ((nd->nd_flag & ND_HASSEQUENCE) != 0)
381 				nfsrv_cache_session(nd->nd_sessionid,
382 				    nd->nd_slotid, nd->nd_repstat, &m);
383 			if (nd->nd_repstat == NFSERR_REPLYFROMCACHE)
384 				nd->nd_repstat = 0;
385 			cacherep = RC_REPLY;
386 		} else {
387 			if (nd->nd_repstat == NFSERR_DONTREPLY)
388 				cacherep = RC_DROPIT;
389 			else
390 				cacherep = RC_REPLY;
391 			*rpp = nfsrvd_updatecache(nd);
392 		}
393 	}
394 	if (tagstr != NULL && taglen > NFSV4_SMALLSTR)
395 		free(tagstr, M_TEMP);
396 
397 	NFSEXITCODE2(0, nd);
398 	return (cacherep);
399 }
400 
401 static void
402 nfssvc_loss(SVCXPRT *xprt)
403 {
404 	uint32_t ack;
405 
406 	ack = 0;
407 	SVC_ACK(xprt, &ack);
408 	nfsrc_trimcache(xprt->xp_sockref, ack, 1);
409 }
410 
411 /*
412  * Adds a socket to the list for servicing by nfsds.
413  */
414 int
415 nfsrvd_addsock(struct file *fp)
416 {
417 	int siz;
418 	struct socket *so;
419 	int error = 0;
420 	SVCXPRT *xprt;
421 	static u_int64_t sockref = 0;
422 
423 	so = fp->f_data;
424 
425 	siz = sb_max_adj;
426 	error = soreserve(so, siz, siz);
427 	if (error)
428 		goto out;
429 
430 	/*
431 	 * Steal the socket from userland so that it doesn't close
432 	 * unexpectedly.
433 	 */
434 	if (so->so_type == SOCK_DGRAM)
435 		xprt = svc_dg_create(nfsrvd_pool, so, 0, 0);
436 	else
437 		xprt = svc_vc_create(nfsrvd_pool, so, 0, 0);
438 	if (xprt) {
439 		fp->f_ops = &badfileops;
440 		fp->f_data = NULL;
441 		xprt->xp_sockref = ++sockref;
442 		if (nfs_minvers == NFS_VER2)
443 			svc_reg(xprt, NFS_PROG, NFS_VER2, nfssvc_program,
444 			    NULL);
445 		if (nfs_minvers <= NFS_VER3 && nfs_maxvers >= NFS_VER3)
446 			svc_reg(xprt, NFS_PROG, NFS_VER3, nfssvc_program,
447 			    NULL);
448 		if (nfs_maxvers >= NFS_VER4)
449 			svc_reg(xprt, NFS_PROG, NFS_VER4, nfssvc_program,
450 			    NULL);
451 		if (so->so_type == SOCK_STREAM)
452 			svc_loss_reg(xprt, nfssvc_loss);
453 		SVC_RELEASE(xprt);
454 	}
455 
456 out:
457 	NFSEXITCODE(error);
458 	return (error);
459 }
460 
461 /*
462  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
463  * until it is killed by a signal.
464  */
465 int
466 nfsrvd_nfsd(struct thread *td, struct nfsd_nfsd_args *args)
467 {
468 	char principal[MAXHOSTNAMELEN + 5];
469 	struct proc *p;
470 	int error = 0;
471 	bool_t ret2, ret3, ret4;
472 
473 	error = copyinstr(args->principal, principal, sizeof (principal),
474 	    NULL);
475 	if (error)
476 		goto out;
477 
478 	/*
479 	 * Only the first nfsd actually does any work. The RPC code
480 	 * adds threads to it as needed. Any extra processes offered
481 	 * by nfsd just exit. If nfsd is new enough, it will call us
482 	 * once with a structure that specifies how many threads to
483 	 * use.
484 	 */
485 	NFSD_LOCK();
486 	if (newnfs_numnfsd == 0) {
487 		p = td->td_proc;
488 		PROC_LOCK(p);
489 		p->p_flag2 |= P2_AST_SU;
490 		PROC_UNLOCK(p);
491 		newnfs_numnfsd++;
492 
493 		NFSD_UNLOCK();
494 
495 		/* An empty string implies AUTH_SYS only. */
496 		if (principal[0] != '\0') {
497 			ret2 = rpc_gss_set_svc_name_call(principal,
498 			    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG, NFS_VER2);
499 			ret3 = rpc_gss_set_svc_name_call(principal,
500 			    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG, NFS_VER3);
501 			ret4 = rpc_gss_set_svc_name_call(principal,
502 			    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG, NFS_VER4);
503 
504 			if (!ret2 || !ret3 || !ret4)
505 				printf("nfsd: can't register svc name\n");
506 		}
507 
508 		nfsrvd_pool->sp_minthreads = args->minthreads;
509 		nfsrvd_pool->sp_maxthreads = args->maxthreads;
510 
511 		svc_run(nfsrvd_pool);
512 
513 		if (principal[0] != '\0') {
514 			rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER2);
515 			rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER3);
516 			rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER4);
517 		}
518 
519 		NFSD_LOCK();
520 		newnfs_numnfsd--;
521 		nfsrvd_init(1);
522 		PROC_LOCK(p);
523 		p->p_flag2 &= ~P2_AST_SU;
524 		PROC_UNLOCK(p);
525 	}
526 	NFSD_UNLOCK();
527 
528 out:
529 	NFSEXITCODE(error);
530 	return (error);
531 }
532 
533 /*
534  * Initialize the data structures for the server.
535  * Handshake with any new nfsds starting up to avoid any chance of
536  * corruption.
537  */
538 void
539 nfsrvd_init(int terminating)
540 {
541 
542 	NFSD_LOCK_ASSERT();
543 
544 	if (terminating) {
545 		nfsd_master_proc = NULL;
546 		NFSD_UNLOCK();
547 		svcpool_destroy(nfsrvd_pool);
548 		nfsrvd_pool = NULL;
549 		NFSD_LOCK();
550 	}
551 
552 	NFSD_UNLOCK();
553 
554 	nfsrvd_pool = svcpool_create("nfsd", SYSCTL_STATIC_CHILDREN(_vfs_nfsd));
555 	nfsrvd_pool->sp_rcache = NULL;
556 	nfsrvd_pool->sp_assign = fhanew_assign;
557 	nfsrvd_pool->sp_done = fha_nd_complete;
558 
559 	NFSD_LOCK();
560 }
561 
562