xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 76dfffe33547c37f8bdd446e3e4ab0f3c16cea4b)
1 /*	$NetBSD: nfs_socket.c,v 1.31 1996/10/13 01:39:07 christos Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1991, 1993, 1995
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. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)nfs_socket.c	8.5 (Berkeley) 3/30/95
39  */
40 
41 /*
42  * Socket operations for use by nfs
43  */
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/proc.h>
48 #include <sys/mount.h>
49 #include <sys/kernel.h>
50 #include <sys/mbuf.h>
51 #include <sys/vnode.h>
52 #include <sys/domain.h>
53 #include <sys/protosw.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/syslog.h>
57 #include <sys/tprintf.h>
58 #include <sys/namei.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/tcp.h>
62 
63 #include <nfs/rpcv2.h>
64 #include <nfs/nfsproto.h>
65 #include <nfs/nfs.h>
66 #include <nfs/xdr_subs.h>
67 #include <nfs/nfsm_subs.h>
68 #include <nfs/nfsmount.h>
69 #include <nfs/nfsnode.h>
70 #include <nfs/nfsrtt.h>
71 #include <nfs/nqnfs.h>
72 #include <nfs/nfs_var.h>
73 
74 #define	TRUE	1
75 #define	FALSE	0
76 
77 /*
78  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
79  * Use the mean and mean deviation of rtt for the appropriate type of rpc
80  * for the frequent rpcs and a default for the others.
81  * The justification for doing "other" this way is that these rpcs
82  * happen so infrequently that timer est. would probably be stale.
83  * Also, since many of these rpcs are
84  * non-idempotent, a conservative timeout is desired.
85  * getattr, lookup - A+2D
86  * read, write     - A+4D
87  * other           - nm_timeo
88  */
89 #define	NFS_RTO(n, t) \
90 	((t) == 0 ? (n)->nm_timeo : \
91 	 ((t) < 3 ? \
92 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
93 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
94 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
95 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
96 /*
97  * External data, mostly RPC constants in XDR form
98  */
99 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
100 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
101 	rpc_auth_kerb;
102 extern u_int32_t nfs_prog, nqnfs_prog;
103 extern time_t nqnfsstarttime;
104 extern struct nfsstats nfsstats;
105 extern int nfsv3_procid[NFS_NPROCS];
106 extern int nfs_ticks;
107 
108 /*
109  * Defines which timer to use for the procnum.
110  * 0 - default
111  * 1 - getattr
112  * 2 - lookup
113  * 3 - read
114  * 4 - write
115  */
116 static int proct[NFS_NPROCS] = {
117 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
118 	0, 0, 0,
119 };
120 
121 /*
122  * There is a congestion window for outstanding rpcs maintained per mount
123  * point. The cwnd size is adjusted in roughly the way that:
124  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
125  * SIGCOMM '88". ACM, August 1988.
126  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
127  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
128  * of rpcs is in progress.
129  * (The sent count and cwnd are scaled for integer arith.)
130  * Variants of "slow start" were tried and were found to be too much of a
131  * performance hit (ave. rtt 3 times larger),
132  * I suspect due to the large rtt that nfs rpcs have.
133  */
134 #define	NFS_CWNDSCALE	256
135 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
136 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
137 int nfsrtton = 0;
138 struct nfsrtt nfsrtt;
139 
140 /*
141  * Initialize sockets and congestion for a new NFS connection.
142  * We do not free the sockaddr if error.
143  */
144 int
145 nfs_connect(nmp, rep)
146 	register struct nfsmount *nmp;
147 	struct nfsreq *rep;
148 {
149 	register struct socket *so;
150 	int s, error, rcvreserve, sndreserve;
151 	struct sockaddr *saddr;
152 	struct sockaddr_in *sin;
153 	struct mbuf *m;
154 	u_int16_t tport;
155 
156 	nmp->nm_so = (struct socket *)0;
157 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
158 	error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
159 		nmp->nm_soproto);
160 	if (error)
161 		goto bad;
162 	so = nmp->nm_so;
163 	nmp->nm_soflags = so->so_proto->pr_flags;
164 
165 	/*
166 	 * Some servers require that the client port be a reserved port number.
167 	 */
168 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
169 		MGET(m, M_WAIT, MT_SONAME);
170 		sin = mtod(m, struct sockaddr_in *);
171 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
172 		sin->sin_family = AF_INET;
173 		sin->sin_addr.s_addr = INADDR_ANY;
174 		tport = IPPORT_RESERVED - 1;
175 		sin->sin_port = htons(tport);
176 		while ((error = sobind(so, m)) == EADDRINUSE &&
177 		       --tport > IPPORT_RESERVED / 2)
178 			sin->sin_port = htons(tport);
179 		m_freem(m);
180 		if (error)
181 			goto bad;
182 	}
183 
184 	/*
185 	 * Protocols that do not require connections may be optionally left
186 	 * unconnected for servers that reply from a port other than NFS_PORT.
187 	 */
188 	if (nmp->nm_flag & NFSMNT_NOCONN) {
189 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
190 			error = ENOTCONN;
191 			goto bad;
192 		}
193 	} else {
194 		error = soconnect(so, nmp->nm_nam);
195 		if (error)
196 			goto bad;
197 
198 		/*
199 		 * Wait for the connection to complete. Cribbed from the
200 		 * connect system call but with the wait timing out so
201 		 * that interruptible mounts don't hang here for a long time.
202 		 */
203 		s = splsoftnet();
204 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
205 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
206 				"nfscon", 2 * hz);
207 			if ((so->so_state & SS_ISCONNECTING) &&
208 			    so->so_error == 0 && rep &&
209 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
210 				so->so_state &= ~SS_ISCONNECTING;
211 				splx(s);
212 				goto bad;
213 			}
214 		}
215 		if (so->so_error) {
216 			error = so->so_error;
217 			so->so_error = 0;
218 			splx(s);
219 			goto bad;
220 		}
221 		splx(s);
222 	}
223 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
224 		so->so_rcv.sb_timeo = (5 * hz);
225 		so->so_snd.sb_timeo = (5 * hz);
226 	} else {
227 		so->so_rcv.sb_timeo = 0;
228 		so->so_snd.sb_timeo = 0;
229 	}
230 	if (nmp->nm_sotype == SOCK_DGRAM) {
231 		sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR;
232 		rcvreserve = max(nmp->nm_rsize, nmp->nm_readdirsize) +
233 		    NFS_MAXPKTHDR;
234 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
235 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
236 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
237 		    NFS_MAXPKTHDR) * 2;
238 	} else {
239 		if (nmp->nm_sotype != SOCK_STREAM)
240 			panic("nfscon sotype");
241 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
242 			MGET(m, M_WAIT, MT_SOOPTS);
243 			*mtod(m, int32_t *) = 1;
244 			m->m_len = sizeof(int32_t);
245 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
246 		}
247 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
248 			MGET(m, M_WAIT, MT_SOOPTS);
249 			*mtod(m, int32_t *) = 1;
250 			m->m_len = sizeof(int32_t);
251 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
252 		}
253 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
254 		    sizeof (u_int32_t)) * 2;
255 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
256 		    sizeof (u_int32_t)) * 2;
257 	}
258 	error = soreserve(so, sndreserve, rcvreserve);
259 	if (error)
260 		goto bad;
261 	so->so_rcv.sb_flags |= SB_NOINTR;
262 	so->so_snd.sb_flags |= SB_NOINTR;
263 
264 	/* Initialize other non-zero congestion variables */
265 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
266 		nmp->nm_srtt[4] = (NFS_TIMEO << 3);
267 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
268 		nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0;
269 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
270 	nmp->nm_sent = 0;
271 	nmp->nm_timeouts = 0;
272 	return (0);
273 
274 bad:
275 	nfs_disconnect(nmp);
276 	return (error);
277 }
278 
279 /*
280  * Reconnect routine:
281  * Called when a connection is broken on a reliable protocol.
282  * - clean up the old socket
283  * - nfs_connect() again
284  * - set R_MUSTRESEND for all outstanding requests on mount point
285  * If this fails the mount point is DEAD!
286  * nb: Must be called with the nfs_sndlock() set on the mount point.
287  */
288 int
289 nfs_reconnect(rep)
290 	register struct nfsreq *rep;
291 {
292 	register struct nfsreq *rp;
293 	register struct nfsmount *nmp = rep->r_nmp;
294 	int error;
295 
296 	nfs_disconnect(nmp);
297 	while ((error = nfs_connect(nmp, rep)) != 0) {
298 		if (error == EINTR || error == ERESTART)
299 			return (EINTR);
300 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0);
301 	}
302 
303 	/*
304 	 * Loop through outstanding request list and fix up all requests
305 	 * on old socket.
306 	 */
307 	for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) {
308 		if (rp->r_nmp == nmp)
309 			rp->r_flags |= R_MUSTRESEND;
310 	}
311 	return (0);
312 }
313 
314 /*
315  * NFS disconnect. Clean up and unlink.
316  */
317 void
318 nfs_disconnect(nmp)
319 	register struct nfsmount *nmp;
320 {
321 	register struct socket *so;
322 
323 	if (nmp->nm_so) {
324 		so = nmp->nm_so;
325 		nmp->nm_so = (struct socket *)0;
326 		soshutdown(so, 2);
327 		soclose(so);
328 	}
329 }
330 
331 /*
332  * This is the nfs send routine. For connection based socket types, it
333  * must be called with an nfs_sndlock() on the socket.
334  * "rep == NULL" indicates that it has been called from a server.
335  * For the client side:
336  * - return EINTR if the RPC is terminated, 0 otherwise
337  * - set R_MUSTRESEND if the send fails for any reason
338  * - do any cleanup required by recoverable socket errors (???)
339  * For the server side:
340  * - return EINTR or ERESTART if interrupted by a signal
341  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
342  * - do any cleanup required by recoverable socket errors (???)
343  */
344 int
345 nfs_send(so, nam, top, rep)
346 	register struct socket *so;
347 	struct mbuf *nam;
348 	register struct mbuf *top;
349 	struct nfsreq *rep;
350 {
351 	struct mbuf *sendnam;
352 	int error, soflags, flags;
353 
354 	if (rep) {
355 		if (rep->r_flags & R_SOFTTERM) {
356 			m_freem(top);
357 			return (EINTR);
358 		}
359 		if ((so = rep->r_nmp->nm_so) == NULL) {
360 			rep->r_flags |= R_MUSTRESEND;
361 			m_freem(top);
362 			return (0);
363 		}
364 		rep->r_flags &= ~R_MUSTRESEND;
365 		soflags = rep->r_nmp->nm_soflags;
366 	} else
367 		soflags = so->so_proto->pr_flags;
368 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
369 		sendnam = (struct mbuf *)0;
370 	else
371 		sendnam = nam;
372 	if (so->so_type == SOCK_SEQPACKET)
373 		flags = MSG_EOR;
374 	else
375 		flags = 0;
376 
377 	error = sosend(so, sendnam, (struct uio *)0, top,
378 		(struct mbuf *)0, flags);
379 	if (error) {
380 		if (rep) {
381 			log(LOG_INFO, "nfs send error %d for server %s\n",error,
382 			    rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
383 			/*
384 			 * Deal with errors for the client side.
385 			 */
386 			if (rep->r_flags & R_SOFTTERM)
387 				error = EINTR;
388 			else
389 				rep->r_flags |= R_MUSTRESEND;
390 		} else
391 			log(LOG_INFO, "nfsd send error %d\n", error);
392 
393 		/*
394 		 * Handle any recoverable (soft) socket errors here. (???)
395 		 */
396 		if (error != EINTR && error != ERESTART &&
397 			error != EWOULDBLOCK && error != EPIPE)
398 			error = 0;
399 	}
400 	return (error);
401 }
402 
403 #ifdef NFSCLIENT
404 /*
405  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
406  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
407  * Mark and consolidate the data into a new mbuf list.
408  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
409  *     small mbufs.
410  * For SOCK_STREAM we must be very careful to read an entire record once
411  * we have read any of it, even if the system call has been interrupted.
412  */
413 int
414 nfs_receive(rep, aname, mp)
415 	register struct nfsreq *rep;
416 	struct mbuf **aname;
417 	struct mbuf **mp;
418 {
419 	register struct socket *so;
420 	struct uio auio;
421 	struct iovec aio;
422 	register struct mbuf *m;
423 	struct mbuf *control;
424 	u_int32_t len;
425 	struct mbuf **getnam;
426 	int error, sotype, rcvflg;
427 	struct proc *p = curproc;	/* XXX */
428 
429 	/*
430 	 * Set up arguments for soreceive()
431 	 */
432 	*mp = (struct mbuf *)0;
433 	*aname = (struct mbuf *)0;
434 	sotype = rep->r_nmp->nm_sotype;
435 
436 	/*
437 	 * For reliable protocols, lock against other senders/receivers
438 	 * in case a reconnect is necessary.
439 	 * For SOCK_STREAM, first get the Record Mark to find out how much
440 	 * more there is to get.
441 	 * We must lock the socket against other receivers
442 	 * until we have an entire rpc request/reply.
443 	 */
444 	if (sotype != SOCK_DGRAM) {
445 		error = nfs_sndlock(&rep->r_nmp->nm_flag, rep);
446 		if (error)
447 			return (error);
448 tryagain:
449 		/*
450 		 * Check for fatal errors and resending request.
451 		 */
452 		/*
453 		 * Ugh: If a reconnect attempt just happened, nm_so
454 		 * would have changed. NULL indicates a failed
455 		 * attempt that has essentially shut down this
456 		 * mount point.
457 		 */
458 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
459 			nfs_sndunlock(&rep->r_nmp->nm_flag);
460 			return (EINTR);
461 		}
462 		so = rep->r_nmp->nm_so;
463 		if (!so) {
464 			error = nfs_reconnect(rep);
465 			if (error) {
466 				nfs_sndunlock(&rep->r_nmp->nm_flag);
467 				return (error);
468 			}
469 			goto tryagain;
470 		}
471 		while (rep->r_flags & R_MUSTRESEND) {
472 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
473 			nfsstats.rpcretries++;
474 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
475 			if (error) {
476 				if (error == EINTR || error == ERESTART ||
477 				    (error = nfs_reconnect(rep)) != 0) {
478 					nfs_sndunlock(&rep->r_nmp->nm_flag);
479 					return (error);
480 				}
481 				goto tryagain;
482 			}
483 		}
484 		nfs_sndunlock(&rep->r_nmp->nm_flag);
485 		if (sotype == SOCK_STREAM) {
486 			aio.iov_base = (caddr_t) &len;
487 			aio.iov_len = sizeof(u_int32_t);
488 			auio.uio_iov = &aio;
489 			auio.uio_iovcnt = 1;
490 			auio.uio_segflg = UIO_SYSSPACE;
491 			auio.uio_rw = UIO_READ;
492 			auio.uio_offset = 0;
493 			auio.uio_resid = sizeof(u_int32_t);
494 			auio.uio_procp = p;
495 			do {
496 			   rcvflg = MSG_WAITALL;
497 			   error = soreceive(so, (struct mbuf **)0, &auio,
498 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
499 			   if (error == EWOULDBLOCK && rep) {
500 				if (rep->r_flags & R_SOFTTERM)
501 					return (EINTR);
502 			   }
503 			} while (error == EWOULDBLOCK);
504 			if (!error && auio.uio_resid > 0) {
505 			    log(LOG_INFO,
506 				 "short receive (%d/%d) from nfs server %s\n",
507 				 sizeof(u_int32_t) - auio.uio_resid,
508 				 sizeof(u_int32_t),
509 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
510 			    error = EPIPE;
511 			}
512 			if (error)
513 				goto errout;
514 			len = ntohl(len) & ~0x80000000;
515 			/*
516 			 * This is SERIOUS! We are out of sync with the sender
517 			 * and forcing a disconnect/reconnect is all I can do.
518 			 */
519 			if (len > NFS_MAXPACKET) {
520 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
521 				"impossible packet length",
522 				len,
523 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
524 			    error = EFBIG;
525 			    goto errout;
526 			}
527 			auio.uio_resid = len;
528 			do {
529 			    rcvflg = MSG_WAITALL;
530 			    error =  soreceive(so, (struct mbuf **)0,
531 				&auio, mp, (struct mbuf **)0, &rcvflg);
532 			} while (error == EWOULDBLOCK || error == EINTR ||
533 				 error == ERESTART);
534 			if (!error && auio.uio_resid > 0) {
535 			    log(LOG_INFO,
536 				"short receive (%d/%d) from nfs server %s\n",
537 				len - auio.uio_resid, len,
538 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
539 			    error = EPIPE;
540 			}
541 		} else {
542 			/*
543 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
544 			 * and soreceive() will return when it has either a
545 			 * control msg or a data msg.
546 			 * We have no use for control msg., but must grab them
547 			 * and then throw them away so we know what is going
548 			 * on.
549 			 */
550 			auio.uio_resid = len = 100000000; /* Anything Big */
551 			auio.uio_procp = p;
552 			do {
553 			    rcvflg = 0;
554 			    error =  soreceive(so, (struct mbuf **)0,
555 				&auio, mp, &control, &rcvflg);
556 			    if (control)
557 				m_freem(control);
558 			    if (error == EWOULDBLOCK && rep) {
559 				if (rep->r_flags & R_SOFTTERM)
560 					return (EINTR);
561 			    }
562 			} while (error == EWOULDBLOCK ||
563 				 (!error && *mp == NULL && control));
564 			if ((rcvflg & MSG_EOR) == 0)
565 				printf("Egad!!\n");
566 			if (!error && *mp == NULL)
567 				error = EPIPE;
568 			len -= auio.uio_resid;
569 		}
570 errout:
571 		if (error && error != EINTR && error != ERESTART) {
572 			m_freem(*mp);
573 			*mp = (struct mbuf *)0;
574 			if (error != EPIPE)
575 				log(LOG_INFO,
576 				    "receive error %d from nfs server %s\n",
577 				    error,
578 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
579 			error = nfs_sndlock(&rep->r_nmp->nm_flag, rep);
580 			if (!error)
581 				error = nfs_reconnect(rep);
582 			if (!error)
583 				goto tryagain;
584 		}
585 	} else {
586 		if ((so = rep->r_nmp->nm_so) == NULL)
587 			return (EACCES);
588 		if (so->so_state & SS_ISCONNECTED)
589 			getnam = (struct mbuf **)0;
590 		else
591 			getnam = aname;
592 		auio.uio_resid = len = 1000000;
593 		auio.uio_procp = p;
594 		do {
595 			rcvflg = 0;
596 			error =  soreceive(so, getnam, &auio, mp,
597 				(struct mbuf **)0, &rcvflg);
598 			if (error == EWOULDBLOCK &&
599 			    (rep->r_flags & R_SOFTTERM))
600 				return (EINTR);
601 		} while (error == EWOULDBLOCK);
602 		len -= auio.uio_resid;
603 	}
604 	if (error) {
605 		m_freem(*mp);
606 		*mp = (struct mbuf *)0;
607 	}
608 	/*
609 	 * Search for any mbufs that are not a multiple of 4 bytes long
610 	 * or with m_data not longword aligned.
611 	 * These could cause pointer alignment problems, so copy them to
612 	 * well aligned mbufs.
613 	 */
614 	nfs_realign(*mp, 5 * NFSX_UNSIGNED);
615 	return (error);
616 }
617 
618 /*
619  * Implement receipt of reply on a socket.
620  * We must search through the list of received datagrams matching them
621  * with outstanding requests using the xid, until ours is found.
622  */
623 /* ARGSUSED */
624 int
625 nfs_reply(myrep)
626 	struct nfsreq *myrep;
627 {
628 	register struct nfsreq *rep;
629 	register struct nfsmount *nmp = myrep->r_nmp;
630 	register int32_t t1;
631 	struct mbuf *mrep, *nam, *md;
632 	u_int32_t rxid, *tl;
633 	caddr_t dpos, cp2;
634 	int error;
635 
636 	/*
637 	 * Loop around until we get our own reply
638 	 */
639 	for (;;) {
640 		/*
641 		 * Lock against other receivers so that I don't get stuck in
642 		 * sbwait() after someone else has received my reply for me.
643 		 * Also necessary for connection based protocols to avoid
644 		 * race conditions during a reconnect.
645 		 */
646 		error = nfs_rcvlock(myrep);
647 		if (error)
648 			return (error);
649 		/* Already received, bye bye */
650 		if (myrep->r_mrep != NULL) {
651 			nfs_rcvunlock(&nmp->nm_flag);
652 			return (0);
653 		}
654 		/*
655 		 * Get the next Rpc reply off the socket
656 		 */
657 		error = nfs_receive(myrep, &nam, &mrep);
658 		nfs_rcvunlock(&nmp->nm_flag);
659 		if (error) {
660 
661 			/*
662 			 * Ignore routing errors on connectionless protocols??
663 			 */
664 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
665 				nmp->nm_so->so_error = 0;
666 				if (myrep->r_flags & R_GETONEREP)
667 					return (0);
668 				continue;
669 			}
670 			return (error);
671 		}
672 		if (nam)
673 			m_freem(nam);
674 
675 		/*
676 		 * Get the xid and check that it is an rpc reply
677 		 */
678 		md = mrep;
679 		dpos = mtod(md, caddr_t);
680 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
681 		rxid = *tl++;
682 		if (*tl != rpc_reply) {
683 			if (nmp->nm_flag & NFSMNT_NQNFS) {
684 				if (nqnfs_callback(nmp, mrep, md, dpos))
685 					nfsstats.rpcinvalid++;
686 			} else {
687 				nfsstats.rpcinvalid++;
688 				m_freem(mrep);
689 			}
690 nfsmout:
691 			if (myrep->r_flags & R_GETONEREP)
692 				return (0);
693 			continue;
694 		}
695 
696 		/*
697 		 * Loop through the request list to match up the reply
698 		 * Iff no match, just drop the datagram
699 		 */
700 		for (rep = nfs_reqq.tqh_first; rep != 0;
701 		    rep = rep->r_chain.tqe_next) {
702 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
703 				/* Found it.. */
704 				rep->r_mrep = mrep;
705 				rep->r_md = md;
706 				rep->r_dpos = dpos;
707 				if (nfsrtton) {
708 					struct rttl *rt;
709 
710 					rt = &nfsrtt.rttl[nfsrtt.pos];
711 					rt->proc = rep->r_procnum;
712 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
713 					rt->sent = nmp->nm_sent;
714 					rt->cwnd = nmp->nm_cwnd;
715 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
716 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
717 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
718 					rt->tstamp = time;
719 					if (rep->r_flags & R_TIMING)
720 						rt->rtt = rep->r_rtt;
721 					else
722 						rt->rtt = 1000000;
723 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
724 				}
725 				/*
726 				 * Update congestion window.
727 				 * Do the additive increase of
728 				 * one rpc/rtt.
729 				 */
730 				if (nmp->nm_cwnd <= nmp->nm_sent) {
731 					nmp->nm_cwnd +=
732 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
733 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
734 					if (nmp->nm_cwnd > NFS_MAXCWND)
735 						nmp->nm_cwnd = NFS_MAXCWND;
736 				}
737 				rep->r_flags &= ~R_SENT;
738 				nmp->nm_sent -= NFS_CWNDSCALE;
739 				/*
740 				 * Update rtt using a gain of 0.125 on the mean
741 				 * and a gain of 0.25 on the deviation.
742 				 */
743 				if (rep->r_flags & R_TIMING) {
744 					/*
745 					 * Since the timer resolution of
746 					 * NFS_HZ is so course, it can often
747 					 * result in r_rtt == 0. Since
748 					 * r_rtt == N means that the actual
749 					 * rtt is between N+dt and N+2-dt ticks,
750 					 * add 1.
751 					 */
752 					t1 = rep->r_rtt + 1;
753 					t1 -= (NFS_SRTT(rep) >> 3);
754 					NFS_SRTT(rep) += t1;
755 					if (t1 < 0)
756 						t1 = -t1;
757 					t1 -= (NFS_SDRTT(rep) >> 2);
758 					NFS_SDRTT(rep) += t1;
759 				}
760 				nmp->nm_timeouts = 0;
761 				break;
762 			}
763 		}
764 		/*
765 		 * If not matched to a request, drop it.
766 		 * If it's mine, get out.
767 		 */
768 		if (rep == 0) {
769 			nfsstats.rpcunexpected++;
770 			m_freem(mrep);
771 		} else if (rep == myrep) {
772 			if (rep->r_mrep == NULL)
773 				panic("nfsreply nil");
774 			return (0);
775 		}
776 		if (myrep->r_flags & R_GETONEREP)
777 			return (0);
778 	}
779 }
780 
781 /*
782  * nfs_request - goes something like this
783  *	- fill in request struct
784  *	- links it into list
785  *	- calls nfs_send() for first transmit
786  *	- calls nfs_receive() to get reply
787  *	- break down rpc header and return with nfs reply pointed to
788  *	  by mrep or error
789  * nb: always frees up mreq mbuf list
790  */
791 int
792 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
793 	struct vnode *vp;
794 	struct mbuf *mrest;
795 	int procnum;
796 	struct proc *procp;
797 	struct ucred *cred;
798 	struct mbuf **mrp;
799 	struct mbuf **mdp;
800 	caddr_t *dposp;
801 {
802 	register struct mbuf *m, *mrep;
803 	register struct nfsreq *rep;
804 	register u_int32_t *tl;
805 	register int i;
806 	struct nfsmount *nmp;
807 	struct mbuf *md, *mheadend;
808 	struct nfsnode *np;
809 	char nickv[RPCX_NICKVERF];
810 	time_t reqtime, waituntil;
811 	caddr_t dpos, cp2;
812 	int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type;
813 	int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0;
814 	int verf_len, verf_type;
815 	u_int32_t xid;
816 	u_quad_t frev;
817 	char *auth_str, *verf_str;
818 	NFSKERBKEY_T key;		/* save session key */
819 
820 	nmp = VFSTONFS(vp->v_mount);
821 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
822 	rep->r_nmp = nmp;
823 	rep->r_vp = vp;
824 	rep->r_procp = procp;
825 	rep->r_procnum = procnum;
826 	i = 0;
827 	m = mrest;
828 	while (m) {
829 		i += m->m_len;
830 		m = m->m_next;
831 	}
832 	mrest_len = i;
833 
834 	/*
835 	 * Get the RPC header with authorization.
836 	 */
837 kerbauth:
838 	verf_str = auth_str = (char *)0;
839 	if (nmp->nm_flag & NFSMNT_KERB) {
840 		verf_str = nickv;
841 		verf_len = sizeof (nickv);
842 		auth_type = RPCAUTH_KERB4;
843 		bzero((caddr_t)key, sizeof (key));
844 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
845 			&auth_len, verf_str, verf_len)) {
846 			error = nfs_getauth(nmp, rep, cred, &auth_str,
847 				&auth_len, verf_str, &verf_len, key);
848 			if (error) {
849 				free((caddr_t)rep, M_NFSREQ);
850 				m_freem(mrest);
851 				return (error);
852 			}
853 		}
854 	} else {
855 		auth_type = RPCAUTH_UNIX;
856 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
857 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
858 			5 * NFSX_UNSIGNED;
859 	}
860 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
861 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
862 	if (auth_str)
863 		free(auth_str, M_TEMP);
864 
865 	/*
866 	 * For stream protocols, insert a Sun RPC Record Mark.
867 	 */
868 	if (nmp->nm_sotype == SOCK_STREAM) {
869 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
870 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
871 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
872 	}
873 	rep->r_mreq = m;
874 	rep->r_xid = xid;
875 tryagain:
876 	if (nmp->nm_flag & NFSMNT_SOFT)
877 		rep->r_retry = nmp->nm_retry;
878 	else
879 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
880 	rep->r_rtt = rep->r_rexmit = 0;
881 	if (proct[procnum] > 0)
882 		rep->r_flags = R_TIMING;
883 	else
884 		rep->r_flags = 0;
885 	rep->r_mrep = NULL;
886 
887 	/*
888 	 * Do the client side RPC.
889 	 */
890 	nfsstats.rpcrequests++;
891 	/*
892 	 * Chain request into list of outstanding requests. Be sure
893 	 * to put it LAST so timer finds oldest requests first.
894 	 */
895 	s = splsoftclock();
896 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
897 
898 	/* Get send time for nqnfs */
899 	reqtime = time.tv_sec;
900 
901 	/*
902 	 * If backing off another request or avoiding congestion, don't
903 	 * send this one now but let timer do it. If not timing a request,
904 	 * do it now.
905 	 */
906 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
907 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
908 		nmp->nm_sent < nmp->nm_cwnd)) {
909 		splx(s);
910 		if (nmp->nm_soflags & PR_CONNREQUIRED)
911 			error = nfs_sndlock(&nmp->nm_flag, rep);
912 		if (!error) {
913 			m = m_copym(m, 0, M_COPYALL, M_WAIT);
914 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep);
915 			if (nmp->nm_soflags & PR_CONNREQUIRED)
916 				nfs_sndunlock(&nmp->nm_flag);
917 		}
918 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
919 			nmp->nm_sent += NFS_CWNDSCALE;
920 			rep->r_flags |= R_SENT;
921 		}
922 	} else {
923 		splx(s);
924 		rep->r_rtt = -1;
925 	}
926 
927 	/*
928 	 * Wait for the reply from our send or the timer's.
929 	 */
930 	if (!error || error == EPIPE)
931 		error = nfs_reply(rep);
932 
933 	/*
934 	 * RPC done, unlink the request.
935 	 */
936 	s = splsoftclock();
937 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
938 	splx(s);
939 
940 	/*
941 	 * Decrement the outstanding request count.
942 	 */
943 	if (rep->r_flags & R_SENT) {
944 		rep->r_flags &= ~R_SENT;	/* paranoia */
945 		nmp->nm_sent -= NFS_CWNDSCALE;
946 	}
947 
948 	/*
949 	 * If there was a successful reply and a tprintf msg.
950 	 * tprintf a response.
951 	 */
952 	if (!error && (rep->r_flags & R_TPRINTFMSG))
953 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
954 		    "is alive again");
955 	mrep = rep->r_mrep;
956 	md = rep->r_md;
957 	dpos = rep->r_dpos;
958 	if (error) {
959 		m_freem(rep->r_mreq);
960 		free((caddr_t)rep, M_NFSREQ);
961 		return (error);
962 	}
963 
964 	/*
965 	 * break down the rpc header and check if ok
966 	 */
967 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
968 	if (*tl++ == rpc_msgdenied) {
969 		if (*tl == rpc_mismatch)
970 			error = EOPNOTSUPP;
971 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
972 			if (!failed_auth) {
973 				failed_auth++;
974 				mheadend->m_next = (struct mbuf *)0;
975 				m_freem(mrep);
976 				m_freem(rep->r_mreq);
977 				goto kerbauth;
978 			} else
979 				error = EAUTH;
980 		} else
981 			error = EACCES;
982 		m_freem(mrep);
983 		m_freem(rep->r_mreq);
984 		free((caddr_t)rep, M_NFSREQ);
985 		return (error);
986 	}
987 
988 	/*
989 	 * Grab any Kerberos verifier, otherwise just throw it away.
990 	 */
991 	verf_type = fxdr_unsigned(int, *tl++);
992 	i = fxdr_unsigned(int32_t, *tl);
993 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
994 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
995 		if (error)
996 			goto nfsmout;
997 	} else if (i > 0)
998 		nfsm_adv(nfsm_rndup(i));
999 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1000 	/* 0 == ok */
1001 	if (*tl == 0) {
1002 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1003 		if (*tl != 0) {
1004 			error = fxdr_unsigned(int, *tl);
1005 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1006 				error == NFSERR_TRYLATER) {
1007 				m_freem(mrep);
1008 				error = 0;
1009 				waituntil = time.tv_sec + trylater_delay;
1010 				while (time.tv_sec < waituntil)
1011 					(void) tsleep((caddr_t)&lbolt,
1012 						PSOCK, "nqnfstry", 0);
1013 				trylater_delay *= nfs_backoff[trylater_cnt];
1014 				if (trylater_cnt < 7)
1015 					trylater_cnt++;
1016 				goto tryagain;
1017 			}
1018 
1019 			/*
1020 			 * If the File Handle was stale, invalidate the
1021 			 * lookup cache, just in case.
1022 			 */
1023 			if (error == ESTALE)
1024 				cache_purge(vp);
1025 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1026 				*mrp = mrep;
1027 				*mdp = md;
1028 				*dposp = dpos;
1029 				error |= NFSERR_RETERR;
1030 			} else
1031 				m_freem(mrep);
1032 			m_freem(rep->r_mreq);
1033 			free((caddr_t)rep, M_NFSREQ);
1034 			return (error);
1035 		}
1036 
1037 		/*
1038 		 * For nqnfs, get any lease in reply
1039 		 */
1040 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1041 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1042 			if (*tl) {
1043 				np = VTONFS(vp);
1044 				nqlflag = fxdr_unsigned(int, *tl);
1045 				nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1046 				cachable = fxdr_unsigned(int, *tl++);
1047 				reqtime += fxdr_unsigned(int, *tl++);
1048 				if (reqtime > time.tv_sec) {
1049 				    fxdr_hyper(tl, &frev);
1050 				    nqnfs_clientlease(nmp, np, nqlflag,
1051 					cachable, reqtime, frev);
1052 				}
1053 			}
1054 		}
1055 		*mrp = mrep;
1056 		*mdp = md;
1057 		*dposp = dpos;
1058 		m_freem(rep->r_mreq);
1059 		FREE((caddr_t)rep, M_NFSREQ);
1060 		return (0);
1061 	}
1062 	m_freem(mrep);
1063 	error = EPROTONOSUPPORT;
1064 nfsmout:
1065 	m_freem(rep->r_mreq);
1066 	free((caddr_t)rep, M_NFSREQ);
1067 	return (error);
1068 }
1069 #endif /* NFSCLIENT */
1070 
1071 /*
1072  * Generate the rpc reply header
1073  * siz arg. is used to decide if adding a cluster is worthwhile
1074  */
1075 int
1076 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1077 	int siz;
1078 	struct nfsrv_descript *nd;
1079 	struct nfssvc_sock *slp;
1080 	int err;
1081 	int cache;
1082 	u_quad_t *frev;
1083 	struct mbuf **mrq;
1084 	struct mbuf **mbp;
1085 	caddr_t *bposp;
1086 {
1087 	register u_int32_t *tl;
1088 	register struct mbuf *mreq;
1089 	caddr_t bpos;
1090 	struct mbuf *mb, *mb2;
1091 
1092 	MGETHDR(mreq, M_WAIT, MT_DATA);
1093 	mb = mreq;
1094 	/*
1095 	 * If this is a big reply, use a cluster else
1096 	 * try and leave leading space for the lower level headers.
1097 	 */
1098 	siz += RPC_REPLYSIZ;
1099 	if (siz >= MINCLSIZE) {
1100 		MCLGET(mreq, M_WAIT);
1101 	} else
1102 		mreq->m_data += max_hdr;
1103 	tl = mtod(mreq, u_int32_t *);
1104 	mreq->m_len = 6 * NFSX_UNSIGNED;
1105 	bpos = ((caddr_t)tl) + mreq->m_len;
1106 	*tl++ = txdr_unsigned(nd->nd_retxid);
1107 	*tl++ = rpc_reply;
1108 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1109 		*tl++ = rpc_msgdenied;
1110 		if (err & NFSERR_AUTHERR) {
1111 			*tl++ = rpc_autherr;
1112 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1113 			mreq->m_len -= NFSX_UNSIGNED;
1114 			bpos -= NFSX_UNSIGNED;
1115 		} else {
1116 			*tl++ = rpc_mismatch;
1117 			*tl++ = txdr_unsigned(RPC_VER2);
1118 			*tl = txdr_unsigned(RPC_VER2);
1119 		}
1120 	} else {
1121 		*tl++ = rpc_msgaccepted;
1122 
1123 		/*
1124 		 * For Kerberos authentication, we must send the nickname
1125 		 * verifier back, otherwise just RPCAUTH_NULL.
1126 		 */
1127 		if (nd->nd_flag & ND_KERBFULL) {
1128 		    register struct nfsuid *nuidp;
1129 		    struct timeval ktvin, ktvout;
1130 
1131 		    for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1132 			nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1133 			if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1134 			    (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1135 			     &nuidp->nu_haddr, nd->nd_nam2)))
1136 			    break;
1137 		    }
1138 		    if (nuidp) {
1139 			ktvin.tv_sec =
1140 			    txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1141 			ktvin.tv_usec =
1142 			    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1143 
1144 			/*
1145 			 * Encrypt the timestamp in ecb mode using the
1146 			 * session key.
1147 			 */
1148 #ifdef NFSKERB
1149 			XXX
1150 #endif
1151 
1152 			*tl++ = rpc_auth_kerb;
1153 			*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1154 			*tl = ktvout.tv_sec;
1155 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1156 			*tl++ = ktvout.tv_usec;
1157 			*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1158 		    } else {
1159 			*tl++ = 0;
1160 			*tl++ = 0;
1161 		    }
1162 		} else {
1163 			*tl++ = 0;
1164 			*tl++ = 0;
1165 		}
1166 		switch (err) {
1167 		case EPROGUNAVAIL:
1168 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1169 			break;
1170 		case EPROGMISMATCH:
1171 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1172 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1173 			if (nd->nd_flag & ND_NQNFS) {
1174 				*tl++ = txdr_unsigned(3);
1175 				*tl = txdr_unsigned(3);
1176 			} else {
1177 				*tl++ = txdr_unsigned(2);
1178 				*tl = txdr_unsigned(3);
1179 			}
1180 			break;
1181 		case EPROCUNAVAIL:
1182 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1183 			break;
1184 		case EBADRPC:
1185 			*tl = txdr_unsigned(RPC_GARBAGE);
1186 			break;
1187 		default:
1188 			*tl = 0;
1189 			if (err != NFSERR_RETVOID) {
1190 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1191 				if (err)
1192 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1193 				else
1194 				    *tl = 0;
1195 			}
1196 			break;
1197 		};
1198 	}
1199 
1200 	/*
1201 	 * For nqnfs, piggyback lease as requested.
1202 	 */
1203 	if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1204 		if (nd->nd_flag & ND_LEASE) {
1205 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1206 			*tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1207 			*tl++ = txdr_unsigned(cache);
1208 			*tl++ = txdr_unsigned(nd->nd_duration);
1209 			txdr_hyper(frev, tl);
1210 		} else {
1211 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1212 			*tl = 0;
1213 		}
1214 	}
1215 	*mrq = mreq;
1216 	*mbp = mb;
1217 	*bposp = bpos;
1218 	if (err != 0 && err != NFSERR_RETVOID)
1219 		nfsstats.srvrpc_errs++;
1220 	return (0);
1221 }
1222 
1223 /*
1224  * Nfs timer routine
1225  * Scan the nfsreq list and retranmit any requests that have timed out
1226  * To avoid retransmission attempts on STREAM sockets (in the future) make
1227  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1228  */
1229 void
1230 nfs_timer(arg)
1231 	void *arg;	/* never used */
1232 {
1233 	register struct nfsreq *rep;
1234 	register struct mbuf *m;
1235 	register struct socket *so;
1236 	register struct nfsmount *nmp;
1237 	register int timeo;
1238 	int s, error;
1239 #ifdef NFSSERVER
1240 	register struct nfssvc_sock *slp;
1241 	static long lasttime = 0;
1242 	u_quad_t cur_usec;
1243 #endif
1244 
1245 	s = splsoftnet();
1246 	for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) {
1247 		nmp = rep->r_nmp;
1248 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1249 			continue;
1250 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1251 			rep->r_flags |= R_SOFTTERM;
1252 			continue;
1253 		}
1254 		if (rep->r_rtt >= 0) {
1255 			rep->r_rtt++;
1256 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1257 				timeo = nmp->nm_timeo;
1258 			else
1259 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1260 			if (nmp->nm_timeouts > 0)
1261 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1262 			if (rep->r_rtt <= timeo)
1263 				continue;
1264 			if (nmp->nm_timeouts < 8)
1265 				nmp->nm_timeouts++;
1266 		}
1267 		/*
1268 		 * Check for server not responding
1269 		 */
1270 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1271 		     rep->r_rexmit > nmp->nm_deadthresh) {
1272 			nfs_msg(rep->r_procp,
1273 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1274 			    "not responding");
1275 			rep->r_flags |= R_TPRINTFMSG;
1276 		}
1277 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1278 			nfsstats.rpctimeouts++;
1279 			rep->r_flags |= R_SOFTTERM;
1280 			continue;
1281 		}
1282 		if (nmp->nm_sotype != SOCK_DGRAM) {
1283 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1284 				rep->r_rexmit = NFS_MAXREXMIT;
1285 			continue;
1286 		}
1287 		if ((so = nmp->nm_so) == NULL)
1288 			continue;
1289 
1290 		/*
1291 		 * If there is enough space and the window allows..
1292 		 *	Resend it
1293 		 * Set r_rtt to -1 in case we fail to send it now.
1294 		 */
1295 		rep->r_rtt = -1;
1296 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1297 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1298 		    (rep->r_flags & R_SENT) ||
1299 		    nmp->nm_sent < nmp->nm_cwnd) &&
1300 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1301 			if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
1302 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1303 			    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1304 			else
1305 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1306 			    nmp->nm_nam, (struct mbuf *)0, (struct proc *)0);
1307 			if (error) {
1308 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
1309 					so->so_error = 0;
1310 			} else {
1311 				/*
1312 				 * Iff first send, start timing
1313 				 * else turn timing off, backoff timer
1314 				 * and divide congestion window by 2.
1315 				 */
1316 				if (rep->r_flags & R_SENT) {
1317 					rep->r_flags &= ~R_TIMING;
1318 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1319 						rep->r_rexmit = NFS_MAXREXMIT;
1320 					nmp->nm_cwnd >>= 1;
1321 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1322 						nmp->nm_cwnd = NFS_CWNDSCALE;
1323 					nfsstats.rpcretries++;
1324 				} else {
1325 					rep->r_flags |= R_SENT;
1326 					nmp->nm_sent += NFS_CWNDSCALE;
1327 				}
1328 				rep->r_rtt = 0;
1329 			}
1330 		}
1331 	}
1332 
1333 #ifdef NFSSERVER
1334 	/*
1335 	 * Call the nqnfs server timer once a second to handle leases.
1336 	 */
1337 	if (lasttime != time.tv_sec) {
1338 		lasttime = time.tv_sec;
1339 		nqnfs_serverd();
1340 	}
1341 
1342 	/*
1343 	 * Scan the write gathering queues for writes that need to be
1344 	 * completed now.
1345 	 */
1346 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1347 	for (slp = nfssvc_sockhead.tqh_first; slp != 0;
1348 	    slp = slp->ns_chain.tqe_next) {
1349 	    if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1350 		nfsrv_wakenfsd(slp);
1351 	}
1352 #endif /* NFSSERVER */
1353 	splx(s);
1354 	timeout(nfs_timer, (void *)0, nfs_ticks);
1355 }
1356 
1357 /*
1358  * Test for a termination condition pending on the process.
1359  * This is used for NFSMNT_INT mounts.
1360  */
1361 int
1362 nfs_sigintr(nmp, rep, p)
1363 	struct nfsmount *nmp;
1364 	struct nfsreq *rep;
1365 	register struct proc *p;
1366 {
1367 
1368 	if (rep && (rep->r_flags & R_SOFTTERM))
1369 		return (EINTR);
1370 	if (!(nmp->nm_flag & NFSMNT_INT))
1371 		return (0);
1372 	if (p && p->p_siglist &&
1373 	    (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) &
1374 	    NFSINT_SIGMASK))
1375 		return (EINTR);
1376 	return (0);
1377 }
1378 
1379 /*
1380  * Lock a socket against others.
1381  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1382  * and also to avoid race conditions between the processes with nfs requests
1383  * in progress when a reconnect is necessary.
1384  */
1385 int
1386 nfs_sndlock(flagp, rep)
1387 	register int *flagp;
1388 	struct nfsreq *rep;
1389 {
1390 	struct proc *p;
1391 	int slpflag = 0, slptimeo = 0;
1392 
1393 	if (rep) {
1394 		p = rep->r_procp;
1395 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1396 			slpflag = PCATCH;
1397 	} else
1398 		p = (struct proc *)0;
1399 	while (*flagp & NFSMNT_SNDLOCK) {
1400 		if (nfs_sigintr(rep->r_nmp, rep, p))
1401 			return (EINTR);
1402 		*flagp |= NFSMNT_WANTSND;
1403 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1404 			slptimeo);
1405 		if (slpflag == PCATCH) {
1406 			slpflag = 0;
1407 			slptimeo = 2 * hz;
1408 		}
1409 	}
1410 	*flagp |= NFSMNT_SNDLOCK;
1411 	return (0);
1412 }
1413 
1414 /*
1415  * Unlock the stream socket for others.
1416  */
1417 void
1418 nfs_sndunlock(flagp)
1419 	register int *flagp;
1420 {
1421 
1422 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1423 		panic("nfs sndunlock");
1424 	*flagp &= ~NFSMNT_SNDLOCK;
1425 	if (*flagp & NFSMNT_WANTSND) {
1426 		*flagp &= ~NFSMNT_WANTSND;
1427 		wakeup((caddr_t)flagp);
1428 	}
1429 }
1430 
1431 int
1432 nfs_rcvlock(rep)
1433 	register struct nfsreq *rep;
1434 {
1435 	register int *flagp = &rep->r_nmp->nm_flag;
1436 	int slpflag, slptimeo = 0;
1437 
1438 	if (*flagp & NFSMNT_INT)
1439 		slpflag = PCATCH;
1440 	else
1441 		slpflag = 0;
1442 	while (*flagp & NFSMNT_RCVLOCK) {
1443 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1444 			return (EINTR);
1445 		*flagp |= NFSMNT_WANTRCV;
1446 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1447 			slptimeo);
1448 		if (slpflag == PCATCH) {
1449 			slpflag = 0;
1450 			slptimeo = 2 * hz;
1451 		}
1452 	}
1453 	*flagp |= NFSMNT_RCVLOCK;
1454 	return (0);
1455 }
1456 
1457 /*
1458  * Unlock the stream socket for others.
1459  */
1460 void
1461 nfs_rcvunlock(flagp)
1462 	register int *flagp;
1463 {
1464 
1465 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1466 		panic("nfs rcvunlock");
1467 	*flagp &= ~NFSMNT_RCVLOCK;
1468 	if (*flagp & NFSMNT_WANTRCV) {
1469 		*flagp &= ~NFSMNT_WANTRCV;
1470 		wakeup((caddr_t)flagp);
1471 	}
1472 }
1473 
1474 /*
1475  * Check for badly aligned mbuf data areas and
1476  * realign data in an mbuf list by copying the data areas up, as required.
1477  */
1478 void
1479 nfs_realign(m, hsiz)
1480 	register struct mbuf *m;
1481 	int hsiz;
1482 {
1483 	register struct mbuf *m2;
1484 	register int siz, mlen, olen;
1485 	register caddr_t tcp, fcp;
1486 	struct mbuf *mnew;
1487 
1488 	while (m) {
1489 	    /*
1490 	     * This never happens for UDP, rarely happens for TCP
1491 	     * but frequently happens for iso transport.
1492 	     */
1493 	    if ((m->m_len & 0x3) || (mtod(m, long) & 0x3)) {
1494 		olen = m->m_len;
1495 		fcp = mtod(m, caddr_t);
1496 		if ((long)fcp & 0x3) {
1497 			m->m_flags &= ~M_PKTHDR;
1498 			if (m->m_flags & M_EXT)
1499 				m->m_data = m->m_ext.ext_buf +
1500 					((m->m_ext.ext_size - olen) & ~0x3);
1501 			else
1502 				m->m_data = m->m_dat;
1503 		}
1504 		m->m_len = 0;
1505 		tcp = mtod(m, caddr_t);
1506 		mnew = m;
1507 		m2 = m->m_next;
1508 
1509 		/*
1510 		 * If possible, only put the first invariant part
1511 		 * of the RPC header in the first mbuf.
1512 		 */
1513 		mlen = M_TRAILINGSPACE(m);
1514 		if (olen <= hsiz && mlen > hsiz)
1515 			mlen = hsiz;
1516 
1517 		/*
1518 		 * Loop through the mbuf list consolidating data.
1519 		 */
1520 		while (m) {
1521 			while (olen > 0) {
1522 				if (mlen == 0) {
1523 					m2->m_flags &= ~M_PKTHDR;
1524 					if (m2->m_flags & M_EXT)
1525 						m2->m_data = m2->m_ext.ext_buf;
1526 					else
1527 						m2->m_data = m2->m_dat;
1528 					m2->m_len = 0;
1529 					mlen = M_TRAILINGSPACE(m2);
1530 					tcp = mtod(m2, caddr_t);
1531 					mnew = m2;
1532 					m2 = m2->m_next;
1533 				}
1534 				siz = min(mlen, olen);
1535 				if (tcp != fcp)
1536 					bcopy(fcp, tcp, siz);
1537 				mnew->m_len += siz;
1538 				mlen -= siz;
1539 				olen -= siz;
1540 				tcp += siz;
1541 				fcp += siz;
1542 			}
1543 			m = m->m_next;
1544 			if (m) {
1545 				olen = m->m_len;
1546 				fcp = mtod(m, caddr_t);
1547 			}
1548 		}
1549 
1550 		/*
1551 		 * Finally, set m_len == 0 for any trailing mbufs that have
1552 		 * been copied out of.
1553 		 */
1554 		while (m2) {
1555 			m2->m_len = 0;
1556 			m2 = m2->m_next;
1557 		}
1558 		return;
1559 	    }
1560 	    m = m->m_next;
1561 	}
1562 }
1563 
1564 /*
1565  * Parse an RPC request
1566  * - verify it
1567  * - fill in the cred struct.
1568  */
1569 int
1570 nfs_getreq(nd, nfsd, has_header)
1571 	register struct nfsrv_descript *nd;
1572 	struct nfsd *nfsd;
1573 	int has_header;
1574 {
1575 	register int len, i;
1576 	register u_int32_t *tl;
1577 	register int32_t t1;
1578 	struct uio uio;
1579 	struct iovec iov;
1580 	caddr_t dpos, cp2, cp;
1581 	u_int32_t nfsvers, auth_type;
1582 	uid_t nickuid;
1583 	int error = 0, nqnfs = 0, ticklen;
1584 	struct mbuf *mrep, *md;
1585 	register struct nfsuid *nuidp;
1586 	struct timeval tvin, tvout;
1587 
1588 	mrep = nd->nd_mrep;
1589 	md = nd->nd_md;
1590 	dpos = nd->nd_dpos;
1591 	if (has_header) {
1592 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1593 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1594 		if (*tl++ != rpc_call) {
1595 			m_freem(mrep);
1596 			return (EBADRPC);
1597 		}
1598 	} else
1599 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1600 	nd->nd_repstat = 0;
1601 	nd->nd_flag = 0;
1602 	if (*tl++ != rpc_vers) {
1603 		nd->nd_repstat = ERPCMISMATCH;
1604 		nd->nd_procnum = NFSPROC_NOOP;
1605 		return (0);
1606 	}
1607 	if (*tl != nfs_prog) {
1608 		if (*tl == nqnfs_prog)
1609 			nqnfs++;
1610 		else {
1611 			nd->nd_repstat = EPROGUNAVAIL;
1612 			nd->nd_procnum = NFSPROC_NOOP;
1613 			return (0);
1614 		}
1615 	}
1616 	tl++;
1617 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1618 	if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1619 		(nfsvers != NQNFS_VER3 && nqnfs)) {
1620 		nd->nd_repstat = EPROGMISMATCH;
1621 		nd->nd_procnum = NFSPROC_NOOP;
1622 		return (0);
1623 	}
1624 	if (nqnfs)
1625 		nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1626 	else if (nfsvers == NFS_VER3)
1627 		nd->nd_flag = ND_NFSV3;
1628 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1629 	if (nd->nd_procnum == NFSPROC_NULL)
1630 		return (0);
1631 	if (nd->nd_procnum >= NFS_NPROCS ||
1632 		(!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1633 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1634 		nd->nd_repstat = EPROCUNAVAIL;
1635 		nd->nd_procnum = NFSPROC_NOOP;
1636 		return (0);
1637 	}
1638 	if ((nd->nd_flag & ND_NFSV3) == 0)
1639 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1640 	auth_type = *tl++;
1641 	len = fxdr_unsigned(int, *tl++);
1642 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1643 		m_freem(mrep);
1644 		return (EBADRPC);
1645 	}
1646 
1647 	nd->nd_flag &= ~ND_KERBAUTH;
1648 	/*
1649 	 * Handle auth_unix or auth_kerb.
1650 	 */
1651 	if (auth_type == rpc_auth_unix) {
1652 		len = fxdr_unsigned(int, *++tl);
1653 		if (len < 0 || len > NFS_MAXNAMLEN) {
1654 			m_freem(mrep);
1655 			return (EBADRPC);
1656 		}
1657 		nfsm_adv(nfsm_rndup(len));
1658 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1659 		bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred));
1660 		nd->nd_cr.cr_ref = 1;
1661 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1662 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1663 		len = fxdr_unsigned(int, *tl);
1664 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1665 			m_freem(mrep);
1666 			return (EBADRPC);
1667 		}
1668 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1669 		for (i = 0; i < len; i++)
1670 		    if (i < NGROUPS)
1671 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1672 		    else
1673 			tl++;
1674 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1675 		if (nd->nd_cr.cr_ngroups > 1)
1676 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1677 		len = fxdr_unsigned(int, *++tl);
1678 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1679 			m_freem(mrep);
1680 			return (EBADRPC);
1681 		}
1682 		if (len > 0)
1683 			nfsm_adv(nfsm_rndup(len));
1684 	} else if (auth_type == rpc_auth_kerb) {
1685 		switch (fxdr_unsigned(int, *tl++)) {
1686 		case RPCAKN_FULLNAME:
1687 			ticklen = fxdr_unsigned(int, *tl);
1688 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1689 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1690 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1691 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1692 				m_freem(mrep);
1693 				return (EBADRPC);
1694 			}
1695 			uio.uio_offset = 0;
1696 			uio.uio_iov = &iov;
1697 			uio.uio_iovcnt = 1;
1698 			uio.uio_segflg = UIO_SYSSPACE;
1699 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1700 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1701 			nfsm_mtouio(&uio, uio.uio_resid);
1702 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1703 			if (*tl++ != rpc_auth_kerb ||
1704 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1705 				printf("Bad kerb verifier\n");
1706 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1707 				nd->nd_procnum = NFSPROC_NOOP;
1708 				return (0);
1709 			}
1710 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1711 			tl = (u_int32_t *)cp;
1712 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1713 				printf("Not fullname kerb verifier\n");
1714 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1715 				nd->nd_procnum = NFSPROC_NOOP;
1716 				return (0);
1717 			}
1718 			cp += NFSX_UNSIGNED;
1719 			bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED);
1720 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1721 			nd->nd_flag |= ND_KERBFULL;
1722 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1723 			break;
1724 		case RPCAKN_NICKNAME:
1725 			if (len != 2 * NFSX_UNSIGNED) {
1726 				printf("Kerb nickname short\n");
1727 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1728 				nd->nd_procnum = NFSPROC_NOOP;
1729 				return (0);
1730 			}
1731 			nickuid = fxdr_unsigned(uid_t, *tl);
1732 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1733 			if (*tl++ != rpc_auth_kerb ||
1734 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1735 				printf("Kerb nick verifier bad\n");
1736 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1737 				nd->nd_procnum = NFSPROC_NOOP;
1738 				return (0);
1739 			}
1740 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1741 			tvin.tv_sec = *tl++;
1742 			tvin.tv_usec = *tl;
1743 
1744 			for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1745 			    nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1746 				if (nuidp->nu_cr.cr_uid == nickuid &&
1747 				    (!nd->nd_nam2 ||
1748 				     netaddr_match(NU_NETFAM(nuidp),
1749 				      &nuidp->nu_haddr, nd->nd_nam2)))
1750 					break;
1751 			}
1752 			if (!nuidp) {
1753 				nd->nd_repstat =
1754 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1755 				nd->nd_procnum = NFSPROC_NOOP;
1756 				return (0);
1757 			}
1758 
1759 			/*
1760 			 * Now, decrypt the timestamp using the session key
1761 			 * and validate it.
1762 			 */
1763 #ifdef NFSKERB
1764 			XXX
1765 #endif
1766 
1767 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1768 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1769 			if (nuidp->nu_expire < time.tv_sec ||
1770 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1771 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1772 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1773 				nuidp->nu_expire = 0;
1774 				nd->nd_repstat =
1775 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1776 				nd->nd_procnum = NFSPROC_NOOP;
1777 				return (0);
1778 			}
1779 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1780 			nd->nd_flag |= ND_KERBNICK;
1781 		};
1782 	} else {
1783 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1784 		nd->nd_procnum = NFSPROC_NOOP;
1785 		return (0);
1786 	}
1787 
1788 	/*
1789 	 * For nqnfs, get piggybacked lease request.
1790 	 */
1791 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
1792 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1793 		nd->nd_flag |= fxdr_unsigned(int, *tl);
1794 		if (nd->nd_flag & ND_LEASE) {
1795 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1796 			nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
1797 		} else
1798 			nd->nd_duration = NQ_MINLEASE;
1799 	} else
1800 		nd->nd_duration = NQ_MINLEASE;
1801 	nd->nd_md = md;
1802 	nd->nd_dpos = dpos;
1803 	return (0);
1804 nfsmout:
1805 	return (error);
1806 }
1807 
1808 int
1809 nfs_msg(p, server, msg)
1810 	struct proc *p;
1811 	char *server, *msg;
1812 {
1813 	tpr_t tpr;
1814 
1815 	if (p)
1816 		tpr = tprintf_open(p);
1817 	else
1818 		tpr = NULL;
1819 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1820 	tprintf_close(tpr);
1821 	return (0);
1822 }
1823 
1824 #ifdef NFSSERVER
1825 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
1826 				    struct nfssvc_sock *, struct proc *,
1827 				    struct mbuf **)) = {
1828 	nfsrv_null,
1829 	nfsrv_getattr,
1830 	nfsrv_setattr,
1831 	nfsrv_lookup,
1832 	nfsrv3_access,
1833 	nfsrv_readlink,
1834 	nfsrv_read,
1835 	nfsrv_write,
1836 	nfsrv_create,
1837 	nfsrv_mkdir,
1838 	nfsrv_symlink,
1839 	nfsrv_mknod,
1840 	nfsrv_remove,
1841 	nfsrv_rmdir,
1842 	nfsrv_rename,
1843 	nfsrv_link,
1844 	nfsrv_readdir,
1845 	nfsrv_readdirplus,
1846 	nfsrv_statfs,
1847 	nfsrv_fsinfo,
1848 	nfsrv_pathconf,
1849 	nfsrv_commit,
1850 	nqnfsrv_getlease,
1851 	nqnfsrv_vacated,
1852 	nfsrv_noop,
1853 	nfsrv_noop
1854 };
1855 
1856 /*
1857  * Socket upcall routine for the nfsd sockets.
1858  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1859  * Essentially do as much as possible non-blocking, else punt and it will
1860  * be called with M_WAIT from an nfsd.
1861  */
1862 void
1863 nfsrv_rcv(so, arg, waitflag)
1864 	struct socket *so;
1865 	caddr_t arg;
1866 	int waitflag;
1867 {
1868 	register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1869 	register struct mbuf *m;
1870 	struct mbuf *mp, *nam;
1871 	struct uio auio;
1872 	int flags, error;
1873 
1874 	if ((slp->ns_flag & SLP_VALID) == 0)
1875 		return;
1876 #ifdef notdef
1877 	/*
1878 	 * Define this to test for nfsds handling this under heavy load.
1879 	 */
1880 	if (waitflag == M_DONTWAIT) {
1881 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1882 	}
1883 #endif
1884 	auio.uio_procp = NULL;
1885 	if (so->so_type == SOCK_STREAM) {
1886 		/*
1887 		 * If there are already records on the queue, defer soreceive()
1888 		 * to an nfsd so that there is feedback to the TCP layer that
1889 		 * the nfs servers are heavily loaded.
1890 		 */
1891 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1892 			slp->ns_flag |= SLP_NEEDQ;
1893 			goto dorecs;
1894 		}
1895 
1896 		/*
1897 		 * Do soreceive().
1898 		 */
1899 		auio.uio_resid = 1000000000;
1900 		flags = MSG_DONTWAIT;
1901 		error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1902 		if (error || mp == (struct mbuf *)0) {
1903 			if (error == EWOULDBLOCK)
1904 				slp->ns_flag |= SLP_NEEDQ;
1905 			else
1906 				slp->ns_flag |= SLP_DISCONN;
1907 			goto dorecs;
1908 		}
1909 		m = mp;
1910 		if (slp->ns_rawend) {
1911 			slp->ns_rawend->m_next = m;
1912 			slp->ns_cc += 1000000000 - auio.uio_resid;
1913 		} else {
1914 			slp->ns_raw = m;
1915 			slp->ns_cc = 1000000000 - auio.uio_resid;
1916 		}
1917 		while (m->m_next)
1918 			m = m->m_next;
1919 		slp->ns_rawend = m;
1920 
1921 		/*
1922 		 * Now try and parse record(s) out of the raw stream data.
1923 		 */
1924 		error = nfsrv_getstream(slp, waitflag);
1925 		if (error) {
1926 			if (error == EPERM)
1927 				slp->ns_flag |= SLP_DISCONN;
1928 			else
1929 				slp->ns_flag |= SLP_NEEDQ;
1930 		}
1931 	} else {
1932 		do {
1933 			auio.uio_resid = 1000000000;
1934 			flags = MSG_DONTWAIT;
1935 			error = soreceive(so, &nam, &auio, &mp,
1936 						(struct mbuf **)0, &flags);
1937 			if (mp) {
1938 				nfs_realign(mp, 10 * NFSX_UNSIGNED);
1939 				if (nam) {
1940 					m = nam;
1941 					m->m_next = mp;
1942 				} else
1943 					m = mp;
1944 				if (slp->ns_recend)
1945 					slp->ns_recend->m_nextpkt = m;
1946 				else
1947 					slp->ns_rec = m;
1948 				slp->ns_recend = m;
1949 				m->m_nextpkt = (struct mbuf *)0;
1950 			}
1951 			if (error) {
1952 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
1953 					&& error != EWOULDBLOCK) {
1954 					slp->ns_flag |= SLP_DISCONN;
1955 					goto dorecs;
1956 				}
1957 			}
1958 		} while (mp);
1959 	}
1960 
1961 	/*
1962 	 * Now try and process the request records, non-blocking.
1963 	 */
1964 dorecs:
1965 	if (waitflag == M_DONTWAIT &&
1966 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
1967 		nfsrv_wakenfsd(slp);
1968 }
1969 
1970 /*
1971  * Try and extract an RPC request from the mbuf data list received on a
1972  * stream socket. The "waitflag" argument indicates whether or not it
1973  * can sleep.
1974  */
1975 int
1976 nfsrv_getstream(slp, waitflag)
1977 	register struct nfssvc_sock *slp;
1978 	int waitflag;
1979 {
1980 	register struct mbuf *m, **mpp;
1981 	register char *cp1, *cp2;
1982 	register int len;
1983 	struct mbuf *om, *m2, *recm = NULL;
1984 	u_int32_t recmark;
1985 
1986 	if (slp->ns_flag & SLP_GETSTREAM)
1987 		panic("nfs getstream");
1988 	slp->ns_flag |= SLP_GETSTREAM;
1989 	for (;;) {
1990 	    if (slp->ns_reclen == 0) {
1991 		if (slp->ns_cc < NFSX_UNSIGNED) {
1992 			slp->ns_flag &= ~SLP_GETSTREAM;
1993 			return (0);
1994 		}
1995 		m = slp->ns_raw;
1996 		if (m->m_len >= NFSX_UNSIGNED) {
1997 			bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
1998 			m->m_data += NFSX_UNSIGNED;
1999 			m->m_len -= NFSX_UNSIGNED;
2000 		} else {
2001 			cp1 = (caddr_t)&recmark;
2002 			cp2 = mtod(m, caddr_t);
2003 			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
2004 				while (m->m_len == 0) {
2005 					m = m->m_next;
2006 					cp2 = mtod(m, caddr_t);
2007 				}
2008 				*cp1++ = *cp2++;
2009 				m->m_data++;
2010 				m->m_len--;
2011 			}
2012 		}
2013 		slp->ns_cc -= NFSX_UNSIGNED;
2014 		recmark = ntohl(recmark);
2015 		slp->ns_reclen = recmark & ~0x80000000;
2016 		if (recmark & 0x80000000)
2017 			slp->ns_flag |= SLP_LASTFRAG;
2018 		else
2019 			slp->ns_flag &= ~SLP_LASTFRAG;
2020 		if (slp->ns_reclen > NFS_MAXPACKET) {
2021 			slp->ns_flag &= ~SLP_GETSTREAM;
2022 			return (EPERM);
2023 		}
2024 	    }
2025 
2026 	    /*
2027 	     * Now get the record part.
2028 	     */
2029 	    if (slp->ns_cc == slp->ns_reclen) {
2030 		recm = slp->ns_raw;
2031 		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2032 		slp->ns_cc = slp->ns_reclen = 0;
2033 	    } else if (slp->ns_cc > slp->ns_reclen) {
2034 		len = 0;
2035 		m = slp->ns_raw;
2036 		om = (struct mbuf *)0;
2037 		while (len < slp->ns_reclen) {
2038 			if ((len + m->m_len) > slp->ns_reclen) {
2039 				m2 = m_copym(m, 0, slp->ns_reclen - len,
2040 					waitflag);
2041 				if (m2) {
2042 					if (om) {
2043 						om->m_next = m2;
2044 						recm = slp->ns_raw;
2045 					} else
2046 						recm = m2;
2047 					m->m_data += slp->ns_reclen - len;
2048 					m->m_len -= slp->ns_reclen - len;
2049 					len = slp->ns_reclen;
2050 				} else {
2051 					slp->ns_flag &= ~SLP_GETSTREAM;
2052 					return (EWOULDBLOCK);
2053 				}
2054 			} else if ((len + m->m_len) == slp->ns_reclen) {
2055 				om = m;
2056 				len += m->m_len;
2057 				m = m->m_next;
2058 				recm = slp->ns_raw;
2059 				om->m_next = (struct mbuf *)0;
2060 			} else {
2061 				om = m;
2062 				len += m->m_len;
2063 				m = m->m_next;
2064 			}
2065 		}
2066 		slp->ns_raw = m;
2067 		slp->ns_cc -= len;
2068 		slp->ns_reclen = 0;
2069 	    } else {
2070 		slp->ns_flag &= ~SLP_GETSTREAM;
2071 		return (0);
2072 	    }
2073 
2074 	    /*
2075 	     * Accumulate the fragments into a record.
2076 	     */
2077 	    mpp = &slp->ns_frag;
2078 	    while (*mpp)
2079 		mpp = &((*mpp)->m_next);
2080 	    *mpp = recm;
2081 	    if (slp->ns_flag & SLP_LASTFRAG) {
2082 		nfs_realign(slp->ns_frag, 10 * NFSX_UNSIGNED);
2083 		if (slp->ns_recend)
2084 		    slp->ns_recend->m_nextpkt = slp->ns_frag;
2085 		else
2086 		    slp->ns_rec = slp->ns_frag;
2087 		slp->ns_recend = slp->ns_frag;
2088 		slp->ns_frag = (struct mbuf *)0;
2089 	    }
2090 	}
2091 }
2092 
2093 /*
2094  * Parse an RPC header.
2095  */
2096 int
2097 nfsrv_dorec(slp, nfsd, ndp)
2098 	register struct nfssvc_sock *slp;
2099 	struct nfsd *nfsd;
2100 	struct nfsrv_descript **ndp;
2101 {
2102 	register struct mbuf *m, *nam;
2103 	register struct nfsrv_descript *nd;
2104 	int error;
2105 
2106 	*ndp = NULL;
2107 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2108 	    (m = slp->ns_rec) == (struct mbuf *)0)
2109 		return (ENOBUFS);
2110 	slp->ns_rec = m->m_nextpkt;
2111 	if (slp->ns_rec)
2112 		m->m_nextpkt = (struct mbuf *)0;
2113 	else
2114 		slp->ns_recend = (struct mbuf *)0;
2115 	if (m->m_type == MT_SONAME) {
2116 		nam = m;
2117 		m = m->m_next;
2118 		nam->m_next = NULL;
2119 	} else
2120 		nam = NULL;
2121 	MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2122 		M_NFSRVDESC, M_WAITOK);
2123 	nd->nd_md = nd->nd_mrep = m;
2124 	nd->nd_nam2 = nam;
2125 	nd->nd_dpos = mtod(m, caddr_t);
2126 	error = nfs_getreq(nd, nfsd, TRUE);
2127 	if (error) {
2128 		m_freem(nam);
2129 		free((caddr_t)nd, M_NFSRVDESC);
2130 		return (error);
2131 	}
2132 	*ndp = nd;
2133 	nfsd->nfsd_nd = nd;
2134 	return (0);
2135 }
2136 
2137 
2138 /*
2139  * Search for a sleeping nfsd and wake it up.
2140  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2141  * running nfsds will go look for the work in the nfssvc_sock list.
2142  */
2143 void
2144 nfsrv_wakenfsd(slp)
2145 	struct nfssvc_sock *slp;
2146 {
2147 	register struct nfsd *nd;
2148 
2149 	if ((slp->ns_flag & SLP_VALID) == 0)
2150 		return;
2151 	for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) {
2152 		if (nd->nfsd_flag & NFSD_WAITING) {
2153 			nd->nfsd_flag &= ~NFSD_WAITING;
2154 			if (nd->nfsd_slp)
2155 				panic("nfsd wakeup");
2156 			slp->ns_sref++;
2157 			nd->nfsd_slp = slp;
2158 			wakeup((caddr_t)nd);
2159 			return;
2160 		}
2161 	}
2162 	slp->ns_flag |= SLP_DOREC;
2163 	nfsd_head_flag |= NFSD_CHECKSLP;
2164 }
2165 #endif /* NFSSERVER */
2166