Home | History | Annotate | Line # | Download | only in server
nfs_nfsdkrpc.c revision 1.1
      1 /*	$NetBSD: nfs_nfsdkrpc.c,v 1.1 2013/09/30 07:19:53 dholland Exp $	*/
      2 /*-
      3  * Copyright (c) 1989, 1993
      4  *	The Regents of the University of California.  All rights reserved.
      5  *
      6  * This code is derived from software contributed to Berkeley by
      7  * Rick Macklem at The University of Guelph.
      8  *
      9  * Redistribution and use in source and binary forms, with or without
     10  * modification, are permitted provided that the following conditions
     11  * are met:
     12  * 1. Redistributions of source code must retain the above copyright
     13  *    notice, this list of conditions and the following disclaimer.
     14  * 2. Redistributions in binary form must reproduce the above copyright
     15  *    notice, this list of conditions and the following disclaimer in the
     16  *    documentation and/or other materials provided with the distribution.
     17  * 4. Neither the name of the University nor the names of its contributors
     18  *    may be used to endorse or promote products derived from this software
     19  *    without specific prior written permission.
     20  *
     21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     31  * SUCH DAMAGE.
     32  *
     33  */
     34 
     35 #include <sys/cdefs.h>
     36 /* __FBSDID("FreeBSD: head/sys/fs/nfsserver/nfs_nfsdkrpc.c 249596 2013-04-17 22:42:43Z ken "); */
     37 __RCSID("$NetBSD: nfs_nfsdkrpc.c,v 1.1 2013/09/30 07:19:53 dholland Exp $");
     38 
     39 #include "opt_inet6.h"
     40 #include "opt_kgssapi.h"
     41 
     42 #include <fs/nfs/nfsport.h>
     43 
     44 #include <rpc/rpc.h>
     45 #include <rpc/rpcsec_gss.h>
     46 
     47 #include <nfs/nfs_fha.h>
     48 #include <fs/nfsserver/nfs_fha_new.h>
     49 
     50 #include <security/mac/mac_framework.h>
     51 
     52 NFSDLOCKMUTEX;
     53 NFSV4ROOTLOCKMUTEX;
     54 struct nfsv4lock nfsd_suspend_lock;
     55 
     56 /*
     57  * Mapping of old NFS Version 2 RPC numbers to generic numbers.
     58  */
     59 int newnfs_nfsv3_procid[NFS_V3NPROCS] = {
     60 	NFSPROC_NULL,
     61 	NFSPROC_GETATTR,
     62 	NFSPROC_SETATTR,
     63 	NFSPROC_NOOP,
     64 	NFSPROC_LOOKUP,
     65 	NFSPROC_READLINK,
     66 	NFSPROC_READ,
     67 	NFSPROC_NOOP,
     68 	NFSPROC_WRITE,
     69 	NFSPROC_CREATE,
     70 	NFSPROC_REMOVE,
     71 	NFSPROC_RENAME,
     72 	NFSPROC_LINK,
     73 	NFSPROC_SYMLINK,
     74 	NFSPROC_MKDIR,
     75 	NFSPROC_RMDIR,
     76 	NFSPROC_READDIR,
     77 	NFSPROC_FSSTAT,
     78 	NFSPROC_NOOP,
     79 	NFSPROC_NOOP,
     80 	NFSPROC_NOOP,
     81 	NFSPROC_NOOP,
     82 };
     83 
     84 
     85 SYSCTL_DECL(_vfs_nfsd);
     86 
     87 SVCPOOL		*nfsrvd_pool;
     88 
     89 static int	nfs_privport = 0;
     90 SYSCTL_INT(_vfs_nfsd, OID_AUTO, nfs_privport, CTLFLAG_RW,
     91     &nfs_privport, 0,
     92     "Only allow clients using a privileged port for NFSv2 and 3");
     93 
     94 static int	nfs_minvers = NFS_VER2;
     95 SYSCTL_INT(_vfs_nfsd, OID_AUTO, server_min_nfsvers, CTLFLAG_RW,
     96     &nfs_minvers, 0, "The lowest version of NFS handled by the server");
     97 
     98 static int	nfs_maxvers = NFS_VER4;
     99 SYSCTL_INT(_vfs_nfsd, OID_AUTO, server_max_nfsvers, CTLFLAG_RW,
    100     &nfs_maxvers, 0, "The highest version of NFS handled by the server");
    101 
    102 static int nfs_proc(struct nfsrv_descript *, u_int32_t, struct socket *,
    103     u_int64_t, struct nfsrvcache **);
    104 
    105 extern u_long sb_max_adj;
    106 extern int newnfs_numnfsd;
    107 extern struct proc *nfsd_master_proc;
    108 
    109 /*
    110  * NFS server system calls
    111  */
    112 
    113 static void
    114 nfssvc_program(struct svc_req *rqst, SVCXPRT *xprt)
    115 {
    116 	struct nfsrv_descript nd;
    117 	struct nfsrvcache *rp = NULL;
    118 	int cacherep, credflavor;
    119 
    120 	memset(&nd, 0, sizeof(nd));
    121 	if (rqst->rq_vers == NFS_VER2) {
    122 		if (rqst->rq_proc > NFSV2PROC_STATFS) {
    123 			svcerr_noproc(rqst);
    124 			svc_freereq(rqst);
    125 			goto out;
    126 		}
    127 		nd.nd_procnum = newnfs_nfsv3_procid[rqst->rq_proc];
    128 		nd.nd_flag = ND_NFSV2;
    129 	} else if (rqst->rq_vers == NFS_VER3) {
    130 		if (rqst->rq_proc >= NFS_V3NPROCS) {
    131 			svcerr_noproc(rqst);
    132 			svc_freereq(rqst);
    133 			goto out;
    134 		}
    135 		nd.nd_procnum = rqst->rq_proc;
    136 		nd.nd_flag = ND_NFSV3;
    137 	} else {
    138 		if (rqst->rq_proc != NFSPROC_NULL &&
    139 		    rqst->rq_proc != NFSV4PROC_COMPOUND) {
    140 			svcerr_noproc(rqst);
    141 			svc_freereq(rqst);
    142 			goto out;
    143 		}
    144 		nd.nd_procnum = rqst->rq_proc;
    145 		nd.nd_flag = ND_NFSV4;
    146 	}
    147 
    148 	/*
    149 	 * Note: we want rq_addr, not svc_getrpccaller for nd_nam2 -
    150 	 * NFS_SRVMAXDATA uses a NULL value for nd_nam2 to detect TCP
    151 	 * mounts.
    152 	 */
    153 	nd.nd_mrep = rqst->rq_args;
    154 	rqst->rq_args = NULL;
    155 	newnfs_realign(&nd.nd_mrep, M_WAITOK);
    156 	nd.nd_md = nd.nd_mrep;
    157 	nd.nd_dpos = mtod(nd.nd_md, caddr_t);
    158 	nd.nd_nam = svc_getrpccaller(rqst);
    159 	nd.nd_nam2 = rqst->rq_addr;
    160 	nd.nd_mreq = NULL;
    161 	nd.nd_cred = NULL;
    162 
    163 	if (nfs_privport && (nd.nd_flag & ND_NFSV4) == 0) {
    164 		/* Check if source port is privileged */
    165 		u_short port;
    166 		struct sockaddr *nam = nd.nd_nam;
    167 		struct sockaddr_in *sin;
    168 
    169 		sin = (struct sockaddr_in *)nam;
    170 		/*
    171 		 * INET/INET6 - same code:
    172 		 *    sin_port and sin6_port are at same offset
    173 		 */
    174 		port = ntohs(sin->sin_port);
    175 		if (port >= IPPORT_RESERVED &&
    176 		    nd.nd_procnum != NFSPROC_NULL) {
    177 #ifdef INET6
    178 			char b6[INET6_ADDRSTRLEN];
    179 #if defined(KLD_MODULE)
    180 			/* Do not use ip6_sprintf: the nfs module should work without INET6. */
    181 #define	ip6_sprintf(buf, a)						\
    182 			(sprintf((buf), "%x:%x:%x:%x:%x:%x:%x:%x",	\
    183 			    (a)->s6_addr16[0], (a)->s6_addr16[1],	\
    184 			    (a)->s6_addr16[2], (a)->s6_addr16[3],	\
    185 			    (a)->s6_addr16[4], (a)->s6_addr16[5],	\
    186 			    (a)->s6_addr16[6], (a)->s6_addr16[7]),	\
    187 			    (buf))
    188 #endif
    189 #endif
    190 			printf("NFS request from unprivileged port (%s:%d)\n",
    191 #ifdef INET6
    192 			    sin->sin_family == AF_INET6 ?
    193 			    ip6_sprintf(b6, &satosin6(sin)->sin6_addr) :
    194 #if defined(KLD_MODULE)
    195 #undef ip6_sprintf
    196 #endif
    197 #endif
    198 			    inet_ntoa(sin->sin_addr), port);
    199 			svcerr_weakauth(rqst);
    200 			svc_freereq(rqst);
    201 			m_freem(nd.nd_mrep);
    202 			goto out;
    203 		}
    204 	}
    205 
    206 	if (nd.nd_procnum != NFSPROC_NULL) {
    207 		if (!svc_getcred(rqst, &nd.nd_cred, &credflavor)) {
    208 			svcerr_weakauth(rqst);
    209 			svc_freereq(rqst);
    210 			m_freem(nd.nd_mrep);
    211 			goto out;
    212 		}
    213 
    214 		/* Set the flag based on credflavor */
    215 		if (credflavor == RPCSEC_GSS_KRB5) {
    216 			nd.nd_flag |= ND_GSS;
    217 		} else if (credflavor == RPCSEC_GSS_KRB5I) {
    218 			nd.nd_flag |= (ND_GSS | ND_GSSINTEGRITY);
    219 		} else if (credflavor == RPCSEC_GSS_KRB5P) {
    220 			nd.nd_flag |= (ND_GSS | ND_GSSPRIVACY);
    221 		} else if (credflavor != AUTH_SYS) {
    222 			svcerr_weakauth(rqst);
    223 			svc_freereq(rqst);
    224 			m_freem(nd.nd_mrep);
    225 			goto out;
    226 		}
    227 
    228 #ifdef MAC
    229 		mac_cred_associate_nfsd(nd.nd_cred);
    230 #endif
    231 		/*
    232 		 * Get a refcnt (shared lock) on nfsd_suspend_lock.
    233 		 * NFSSVC_SUSPENDNFSD will take an exclusive lock on
    234 		 * nfsd_suspend_lock to suspend these threads.
    235 		 * This must be done here, before the check of
    236 		 * nfsv4root exports by nfsvno_v4rootexport().
    237 		 */
    238 		NFSLOCKV4ROOTMUTEX();
    239 		nfsv4_getref(&nfsd_suspend_lock, NULL, NFSV4ROOTLOCKMUTEXPTR,
    240 		    NULL);
    241 		NFSUNLOCKV4ROOTMUTEX();
    242 
    243 		if ((nd.nd_flag & ND_NFSV4) != 0) {
    244 			nd.nd_repstat = nfsvno_v4rootexport(&nd);
    245 			if (nd.nd_repstat != 0) {
    246 				NFSLOCKV4ROOTMUTEX();
    247 				nfsv4_relref(&nfsd_suspend_lock);
    248 				NFSUNLOCKV4ROOTMUTEX();
    249 				svcerr_weakauth(rqst);
    250 				svc_freereq(rqst);
    251 				m_freem(nd.nd_mrep);
    252 				goto out;
    253 			}
    254 		}
    255 
    256 		cacherep = nfs_proc(&nd, rqst->rq_xid, xprt->xp_socket,
    257 		    xprt->xp_sockref, &rp);
    258 		NFSLOCKV4ROOTMUTEX();
    259 		nfsv4_relref(&nfsd_suspend_lock);
    260 		NFSUNLOCKV4ROOTMUTEX();
    261 	} else {
    262 		NFSMGET(nd.nd_mreq);
    263 		nd.nd_mreq->m_len = 0;
    264 		cacherep = RC_REPLY;
    265 	}
    266 	if (nd.nd_mrep != NULL)
    267 		m_freem(nd.nd_mrep);
    268 
    269 	if (nd.nd_cred != NULL)
    270 		crfree(nd.nd_cred);
    271 
    272 	if (cacherep == RC_DROPIT) {
    273 		if (nd.nd_mreq != NULL)
    274 			m_freem(nd.nd_mreq);
    275 		svc_freereq(rqst);
    276 		goto out;
    277 	}
    278 
    279 	if (nd.nd_mreq == NULL) {
    280 		svcerr_decode(rqst);
    281 		svc_freereq(rqst);
    282 		goto out;
    283 	}
    284 
    285 	if (nd.nd_repstat & NFSERR_AUTHERR) {
    286 		svcerr_auth(rqst, nd.nd_repstat & ~NFSERR_AUTHERR);
    287 		if (nd.nd_mreq != NULL)
    288 			m_freem(nd.nd_mreq);
    289 	} else if (!svc_sendreply_mbuf(rqst, nd.nd_mreq)) {
    290 		svcerr_systemerr(rqst);
    291 	}
    292 	if (rp != NULL)
    293 		nfsrvd_sentcache(rp, xprt->xp_socket, 0);
    294 	svc_freereq(rqst);
    295 
    296 out:
    297 	NFSEXITCODE(0);
    298 }
    299 
    300 /*
    301  * Check the cache and, optionally, do the RPC.
    302  * Return the appropriate cache response.
    303  */
    304 static int
    305 nfs_proc(struct nfsrv_descript *nd, u_int32_t xid, struct socket *so,
    306     u_int64_t sockref, struct nfsrvcache **rpp)
    307 {
    308 	struct thread *td = curthread;
    309 	int cacherep = RC_DOIT, isdgram;
    310 
    311 	*rpp = NULL;
    312 	if (nd->nd_nam2 == NULL) {
    313 		nd->nd_flag |= ND_STREAMSOCK;
    314 		isdgram = 0;
    315 	} else {
    316 		isdgram = 1;
    317 	}
    318 
    319 	/*
    320 	 * Two cases:
    321 	 * 1 - For NFSv2 over UDP, if we are near our malloc/mget
    322 	 *     limit, just drop the request. There is no
    323 	 *     NFSERR_RESOURCE or NFSERR_DELAY for NFSv2 and the
    324 	 *     client will timeout/retry over UDP in a little while.
    325 	 * 2 - nd_repstat == 0 && nd_mreq == NULL, which
    326 	 *     means a normal nfs rpc, so check the cache
    327 	 */
    328 	if ((nd->nd_flag & ND_NFSV2) && nd->nd_nam2 != NULL &&
    329 	    nfsrv_mallocmget_limit()) {
    330 		cacherep = RC_DROPIT;
    331 	} else {
    332 		/*
    333 		 * For NFSv3, play it safe and assume that the client is
    334 		 * doing retries on the same TCP connection.
    335 		 */
    336 		if ((nd->nd_flag & (ND_NFSV4 | ND_STREAMSOCK)) ==
    337 		    ND_STREAMSOCK)
    338 			nd->nd_flag |= ND_SAMETCPCONN;
    339 		nd->nd_retxid = xid;
    340 		nd->nd_tcpconntime = NFSD_MONOSEC;
    341 		nd->nd_sockref = sockref;
    342 		cacherep = nfsrvd_getcache(nd, so);
    343 	}
    344 
    345 	/*
    346 	 * Handle the request. There are three cases.
    347 	 * RC_DOIT - do the RPC
    348 	 * RC_REPLY - return the reply already created
    349 	 * RC_DROPIT - just throw the request away
    350 	 */
    351 	if (cacherep == RC_DOIT) {
    352 		nfsrvd_dorpc(nd, isdgram, td);
    353 		if (nd->nd_repstat == NFSERR_DONTREPLY)
    354 			cacherep = RC_DROPIT;
    355 		else
    356 			cacherep = RC_REPLY;
    357 		*rpp = nfsrvd_updatecache(nd, so);
    358 	}
    359 
    360 	NFSEXITCODE2(0, nd);
    361 	return (cacherep);
    362 }
    363 
    364 /*
    365  * Adds a socket to the list for servicing by nfsds.
    366  */
    367 int
    368 nfsrvd_addsock(struct file *fp)
    369 {
    370 	int siz;
    371 	struct socket *so;
    372 	int error = 0;
    373 	SVCXPRT *xprt;
    374 	static u_int64_t sockref = 0;
    375 
    376 	so = fp->f_data;
    377 
    378 	siz = sb_max_adj;
    379 	error = soreserve(so, siz, siz);
    380 	if (error)
    381 		goto out;
    382 
    383 	/*
    384 	 * Steal the socket from userland so that it doesn't close
    385 	 * unexpectedly.
    386 	 */
    387 	if (so->so_type == SOCK_DGRAM)
    388 		xprt = svc_dg_create(nfsrvd_pool, so, 0, 0);
    389 	else
    390 		xprt = svc_vc_create(nfsrvd_pool, so, 0, 0);
    391 	if (xprt) {
    392 		fp->f_ops = &badfileops;
    393 		fp->f_data = NULL;
    394 		xprt->xp_sockref = ++sockref;
    395 		if (nfs_minvers == NFS_VER2)
    396 			svc_reg(xprt, NFS_PROG, NFS_VER2, nfssvc_program,
    397 			    NULL);
    398 		if (nfs_minvers <= NFS_VER3 && nfs_maxvers >= NFS_VER3)
    399 			svc_reg(xprt, NFS_PROG, NFS_VER3, nfssvc_program,
    400 			    NULL);
    401 		if (nfs_maxvers >= NFS_VER4)
    402 			svc_reg(xprt, NFS_PROG, NFS_VER4, nfssvc_program,
    403 			    NULL);
    404 		SVC_RELEASE(xprt);
    405 	}
    406 
    407 out:
    408 	NFSEXITCODE(error);
    409 	return (error);
    410 }
    411 
    412 /*
    413  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
    414  * until it is killed by a signal.
    415  */
    416 int
    417 nfsrvd_nfsd(struct thread *td, struct nfsd_nfsd_args *args)
    418 {
    419 	char principal[MAXHOSTNAMELEN + 5];
    420 	int error = 0;
    421 	bool_t ret2, ret3, ret4;
    422 
    423 	error = copyinstr(args->principal, principal, sizeof (principal),
    424 	    NULL);
    425 	if (error)
    426 		goto out;
    427 
    428 	/*
    429 	 * Only the first nfsd actually does any work. The RPC code
    430 	 * adds threads to it as needed. Any extra processes offered
    431 	 * by nfsd just exit. If nfsd is new enough, it will call us
    432 	 * once with a structure that specifies how many threads to
    433 	 * use.
    434 	 */
    435 	NFSD_LOCK();
    436 	if (newnfs_numnfsd == 0) {
    437 		newnfs_numnfsd++;
    438 
    439 		NFSD_UNLOCK();
    440 
    441 		/* An empty string implies AUTH_SYS only. */
    442 		if (principal[0] != '\0') {
    443 			ret2 = rpc_gss_set_svc_name_call(principal,
    444 			    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG, NFS_VER2);
    445 			ret3 = rpc_gss_set_svc_name_call(principal,
    446 			    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG, NFS_VER3);
    447 			ret4 = rpc_gss_set_svc_name_call(principal,
    448 			    "kerberosv5", GSS_C_INDEFINITE, NFS_PROG, NFS_VER4);
    449 
    450 			if (!ret2 || !ret3 || !ret4)
    451 				printf("nfsd: can't register svc name\n");
    452 		}
    453 
    454 		nfsrvd_pool->sp_minthreads = args->minthreads;
    455 		nfsrvd_pool->sp_maxthreads = args->maxthreads;
    456 
    457 		svc_run(nfsrvd_pool);
    458 
    459 		if (principal[0] != '\0') {
    460 			rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER2);
    461 			rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER3);
    462 			rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER4);
    463 		}
    464 
    465 		NFSD_LOCK();
    466 		newnfs_numnfsd--;
    467 		nfsrvd_init(1);
    468 	}
    469 	NFSD_UNLOCK();
    470 
    471 out:
    472 	NFSEXITCODE(error);
    473 	return (error);
    474 }
    475 
    476 /*
    477  * Initialize the data structures for the server.
    478  * Handshake with any new nfsds starting up to avoid any chance of
    479  * corruption.
    480  */
    481 void
    482 nfsrvd_init(int terminating)
    483 {
    484 
    485 	NFSD_LOCK_ASSERT();
    486 
    487 	if (terminating) {
    488 		nfsd_master_proc = NULL;
    489 		NFSD_UNLOCK();
    490 		svcpool_destroy(nfsrvd_pool);
    491 		nfsrvd_pool = NULL;
    492 		NFSD_LOCK();
    493 	}
    494 
    495 	NFSD_UNLOCK();
    496 
    497 	nfsrvd_pool = svcpool_create("nfsd", SYSCTL_STATIC_CHILDREN(_vfs_nfsd));
    498 	nfsrvd_pool->sp_rcache = NULL;
    499 	nfsrvd_pool->sp_assign = fhanew_assign;
    500 	nfsrvd_pool->sp_done = fha_nd_complete;
    501 
    502 	NFSD_LOCK();
    503 }
    504 
    505