xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 89c5a767f8fc7a4633b2d409966e2becbb98ff92)
1 /*	$NetBSD: nfs_socket.c,v 1.53 1999/08/29 16:29:16 sommerfeld 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 "fs_nfs.h"
46 #include "opt_nfsserver.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/proc.h>
51 #include <sys/mount.h>
52 #include <sys/kernel.h>
53 #include <sys/mbuf.h>
54 #include <sys/vnode.h>
55 #include <sys/domain.h>
56 #include <sys/protosw.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/syslog.h>
60 #include <sys/tprintf.h>
61 #include <sys/namei.h>
62 #include <sys/signal.h>
63 #include <sys/signalvar.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/tcp.h>
67 
68 #include <nfs/rpcv2.h>
69 #include <nfs/nfsproto.h>
70 #include <nfs/nfs.h>
71 #include <nfs/xdr_subs.h>
72 #include <nfs/nfsm_subs.h>
73 #include <nfs/nfsmount.h>
74 #include <nfs/nfsnode.h>
75 #include <nfs/nfsrtt.h>
76 #include <nfs/nqnfs.h>
77 #include <nfs/nfs_var.h>
78 
79 #define	TRUE	1
80 #define	FALSE	0
81 
82 /*
83  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
84  * Use the mean and mean deviation of rtt for the appropriate type of rpc
85  * for the frequent rpcs and a default for the others.
86  * The justification for doing "other" this way is that these rpcs
87  * happen so infrequently that timer est. would probably be stale.
88  * Also, since many of these rpcs are
89  * non-idempotent, a conservative timeout is desired.
90  * getattr, lookup - A+2D
91  * read, write     - A+4D
92  * other           - nm_timeo
93  */
94 #define	NFS_RTO(n, t) \
95 	((t) == 0 ? (n)->nm_timeo : \
96 	 ((t) < 3 ? \
97 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
98 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
99 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
100 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
101 /*
102  * External data, mostly RPC constants in XDR form
103  */
104 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
105 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
106 	rpc_auth_kerb;
107 extern u_int32_t nfs_prog, nqnfs_prog;
108 extern time_t nqnfsstarttime;
109 extern struct nfsstats nfsstats;
110 extern int nfsv3_procid[NFS_NPROCS];
111 extern int nfs_ticks;
112 
113 /*
114  * Defines which timer to use for the procnum.
115  * 0 - default
116  * 1 - getattr
117  * 2 - lookup
118  * 3 - read
119  * 4 - write
120  */
121 static int proct[NFS_NPROCS] = {
122 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
123 	0, 0, 0,
124 };
125 
126 /*
127  * There is a congestion window for outstanding rpcs maintained per mount
128  * point. The cwnd size is adjusted in roughly the way that:
129  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
130  * SIGCOMM '88". ACM, August 1988.
131  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
132  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
133  * of rpcs is in progress.
134  * (The sent count and cwnd are scaled for integer arith.)
135  * Variants of "slow start" were tried and were found to be too much of a
136  * performance hit (ave. rtt 3 times larger),
137  * I suspect due to the large rtt that nfs rpcs have.
138  */
139 #define	NFS_CWNDSCALE	256
140 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
141 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
142 int nfsrtton = 0;
143 struct nfsrtt nfsrtt;
144 
145 /*
146  * Initialize sockets and congestion for a new NFS connection.
147  * We do not free the sockaddr if error.
148  */
149 int
150 nfs_connect(nmp, rep)
151 	register struct nfsmount *nmp;
152 	struct nfsreq *rep;
153 {
154 	register struct socket *so;
155 	int s, error, rcvreserve, sndreserve;
156 	struct sockaddr *saddr;
157 	struct sockaddr_in *sin;
158 	struct mbuf *m;
159 	u_int16_t tport;
160 
161 	nmp->nm_so = (struct socket *)0;
162 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
163 	error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
164 		nmp->nm_soproto);
165 	if (error)
166 		goto bad;
167 	so = nmp->nm_so;
168 	nmp->nm_soflags = so->so_proto->pr_flags;
169 
170 	/*
171 	 * Some servers require that the client port be a reserved port number.
172 	 */
173 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
174 		MGET(m, M_WAIT, MT_SONAME);
175 		sin = mtod(m, struct sockaddr_in *);
176 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
177 		sin->sin_family = AF_INET;
178 		sin->sin_addr.s_addr = INADDR_ANY;
179 		tport = IPPORT_RESERVED - 1;
180 		sin->sin_port = htons(tport);
181 		while ((error = sobind(so, m)) == EADDRINUSE &&
182 		       --tport > IPPORT_RESERVED / 2)
183 			sin->sin_port = htons(tport);
184 		m_freem(m);
185 		if (error)
186 			goto bad;
187 	}
188 
189 	/*
190 	 * Protocols that do not require connections may be optionally left
191 	 * unconnected for servers that reply from a port other than NFS_PORT.
192 	 */
193 	if (nmp->nm_flag & NFSMNT_NOCONN) {
194 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
195 			error = ENOTCONN;
196 			goto bad;
197 		}
198 	} else {
199 		error = soconnect(so, nmp->nm_nam);
200 		if (error)
201 			goto bad;
202 
203 		/*
204 		 * Wait for the connection to complete. Cribbed from the
205 		 * connect system call but with the wait timing out so
206 		 * that interruptible mounts don't hang here for a long time.
207 		 */
208 		s = splsoftnet();
209 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
210 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
211 				"nfscn1", 2 * hz);
212 			if ((so->so_state & SS_ISCONNECTING) &&
213 			    so->so_error == 0 && rep &&
214 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
215 				so->so_state &= ~SS_ISCONNECTING;
216 				splx(s);
217 				goto bad;
218 			}
219 		}
220 		if (so->so_error) {
221 			error = so->so_error;
222 			so->so_error = 0;
223 			splx(s);
224 			goto bad;
225 		}
226 		splx(s);
227 	}
228 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
229 		so->so_rcv.sb_timeo = (5 * hz);
230 		so->so_snd.sb_timeo = (5 * hz);
231 	} else {
232 		so->so_rcv.sb_timeo = 0;
233 		so->so_snd.sb_timeo = 0;
234 	}
235 	if (nmp->nm_sotype == SOCK_DGRAM) {
236 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
237 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
238 		    NFS_MAXPKTHDR) * 2;
239 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
240 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
241 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
242 		    NFS_MAXPKTHDR) * 2;
243 	} else {
244 		if (nmp->nm_sotype != SOCK_STREAM)
245 			panic("nfscon sotype");
246 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
247 			MGET(m, M_WAIT, MT_SOOPTS);
248 			*mtod(m, int32_t *) = 1;
249 			m->m_len = sizeof(int32_t);
250 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
251 		}
252 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
253 			MGET(m, M_WAIT, MT_SOOPTS);
254 			*mtod(m, int32_t *) = 1;
255 			m->m_len = sizeof(int32_t);
256 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
257 		}
258 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
259 		    sizeof (u_int32_t)) * 2;
260 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
261 		    sizeof (u_int32_t)) * 2;
262 	}
263 	error = soreserve(so, sndreserve, rcvreserve);
264 	if (error)
265 		goto bad;
266 	so->so_rcv.sb_flags |= SB_NOINTR;
267 	so->so_snd.sb_flags |= SB_NOINTR;
268 
269 	/* Initialize other non-zero congestion variables */
270 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
271 		nmp->nm_srtt[4] = (NFS_TIMEO << 3);
272 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
273 		nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0;
274 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
275 	nmp->nm_sent = 0;
276 	nmp->nm_timeouts = 0;
277 	return (0);
278 
279 bad:
280 	nfs_disconnect(nmp);
281 	return (error);
282 }
283 
284 /*
285  * Reconnect routine:
286  * Called when a connection is broken on a reliable protocol.
287  * - clean up the old socket
288  * - nfs_connect() again
289  * - set R_MUSTRESEND for all outstanding requests on mount point
290  * If this fails the mount point is DEAD!
291  * nb: Must be called with the nfs_sndlock() set on the mount point.
292  */
293 int
294 nfs_reconnect(rep)
295 	register struct nfsreq *rep;
296 {
297 	register struct nfsreq *rp;
298 	register struct nfsmount *nmp = rep->r_nmp;
299 	int error;
300 
301 	nfs_disconnect(nmp);
302 	while ((error = nfs_connect(nmp, rep)) != 0) {
303 		if (error == EINTR || error == ERESTART)
304 			return (EINTR);
305 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0);
306 	}
307 
308 	/*
309 	 * Loop through outstanding request list and fix up all requests
310 	 * on old socket.
311 	 */
312 	for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) {
313 		if (rp->r_nmp == nmp)
314 			rp->r_flags |= R_MUSTRESEND;
315 	}
316 	return (0);
317 }
318 
319 /*
320  * NFS disconnect. Clean up and unlink.
321  */
322 void
323 nfs_disconnect(nmp)
324 	register struct nfsmount *nmp;
325 {
326 	register struct socket *so;
327 	int drain = 0;
328 
329 	if (nmp->nm_so) {
330 		so = nmp->nm_so;
331 		nmp->nm_so = (struct socket *)0;
332 		soshutdown(so, 2);
333 		drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
334 		if (drain) {
335 			/*
336 			 * soshutdown() above should wake up the current
337 			 * listener.
338 			 * Now wake up those waiting for the recive lock, and
339 			 * wait for them to go away unhappy, to prevent *nmp
340 			 * from evaporating while they're sleeping.
341 			 */
342 			while (nmp->nm_waiters > 0) {
343 				wakeup (&nmp->nm_iflag);
344 				sleep(&nmp->nm_waiters, PVFS);
345 			}
346 		}
347 		soclose(so);
348 	}
349 #ifdef DIAGNOSTIC
350 	if (drain && (nmp->nm_waiters > 0))
351 		panic("nfs_disconnect: waiters left after drain?\n");
352 #endif
353 }
354 
355 void
356 nfs_safedisconnect(nmp)
357 	struct nfsmount *nmp;
358 {
359 	struct nfsreq dummyreq;
360 
361 	memset(&dummyreq, 0, sizeof(dummyreq));
362 	dummyreq.r_nmp = nmp;
363 	nfs_rcvlock(&dummyreq); /* XXX ignored error return */
364 	nfs_disconnect(nmp);
365 	nfs_rcvunlock(&nmp->nm_iflag);
366 }
367 
368 /*
369  * This is the nfs send routine. For connection based socket types, it
370  * must be called with an nfs_sndlock() on the socket.
371  * "rep == NULL" indicates that it has been called from a server.
372  * For the client side:
373  * - return EINTR if the RPC is terminated, 0 otherwise
374  * - set R_MUSTRESEND if the send fails for any reason
375  * - do any cleanup required by recoverable socket errors (???)
376  * For the server side:
377  * - return EINTR or ERESTART if interrupted by a signal
378  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
379  * - do any cleanup required by recoverable socket errors (???)
380  */
381 int
382 nfs_send(so, nam, top, rep)
383 	register struct socket *so;
384 	struct mbuf *nam;
385 	register struct mbuf *top;
386 	struct nfsreq *rep;
387 {
388 	struct mbuf *sendnam;
389 	int error, soflags, flags;
390 
391 	if (rep) {
392 		if (rep->r_flags & R_SOFTTERM) {
393 			m_freem(top);
394 			return (EINTR);
395 		}
396 		if ((so = rep->r_nmp->nm_so) == NULL) {
397 			rep->r_flags |= R_MUSTRESEND;
398 			m_freem(top);
399 			return (0);
400 		}
401 		rep->r_flags &= ~R_MUSTRESEND;
402 		soflags = rep->r_nmp->nm_soflags;
403 	} else
404 		soflags = so->so_proto->pr_flags;
405 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
406 		sendnam = (struct mbuf *)0;
407 	else
408 		sendnam = nam;
409 	if (so->so_type == SOCK_SEQPACKET)
410 		flags = MSG_EOR;
411 	else
412 		flags = 0;
413 
414 	error = (*so->so_send)(so, sendnam, (struct uio *)0, top,
415 		(struct mbuf *)0, flags);
416 	if (error) {
417 		if (rep) {
418 			log(LOG_INFO, "nfs send error %d for server %s\n",error,
419 			    rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
420 			/*
421 			 * Deal with errors for the client side.
422 			 */
423 			if (rep->r_flags & R_SOFTTERM)
424 				error = EINTR;
425 			else
426 				rep->r_flags |= R_MUSTRESEND;
427 		} else
428 			log(LOG_INFO, "nfsd send error %d\n", error);
429 
430 		/*
431 		 * Handle any recoverable (soft) socket errors here. (???)
432 		 */
433 		if (error != EINTR && error != ERESTART &&
434 			error != EWOULDBLOCK && error != EPIPE)
435 			error = 0;
436 	}
437 	return (error);
438 }
439 
440 #ifdef NFS
441 /*
442  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
443  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
444  * Mark and consolidate the data into a new mbuf list.
445  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
446  *     small mbufs.
447  * For SOCK_STREAM we must be very careful to read an entire record once
448  * we have read any of it, even if the system call has been interrupted.
449  */
450 int
451 nfs_receive(rep, aname, mp)
452 	register struct nfsreq *rep;
453 	struct mbuf **aname;
454 	struct mbuf **mp;
455 {
456 	register struct socket *so;
457 	struct uio auio;
458 	struct iovec aio;
459 	register struct mbuf *m;
460 	struct mbuf *control;
461 	u_int32_t len;
462 	struct mbuf **getnam;
463 	int error, sotype, rcvflg;
464 	struct proc *p = curproc;	/* XXX */
465 
466 	/*
467 	 * Set up arguments for soreceive()
468 	 */
469 	*mp = (struct mbuf *)0;
470 	*aname = (struct mbuf *)0;
471 	sotype = rep->r_nmp->nm_sotype;
472 
473 	/*
474 	 * For reliable protocols, lock against other senders/receivers
475 	 * in case a reconnect is necessary.
476 	 * For SOCK_STREAM, first get the Record Mark to find out how much
477 	 * more there is to get.
478 	 * We must lock the socket against other receivers
479 	 * until we have an entire rpc request/reply.
480 	 */
481 	if (sotype != SOCK_DGRAM) {
482 		error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
483 		if (error)
484 			return (error);
485 tryagain:
486 		/*
487 		 * Check for fatal errors and resending request.
488 		 */
489 		/*
490 		 * Ugh: If a reconnect attempt just happened, nm_so
491 		 * would have changed. NULL indicates a failed
492 		 * attempt that has essentially shut down this
493 		 * mount point.
494 		 */
495 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
496 			nfs_sndunlock(&rep->r_nmp->nm_iflag);
497 			return (EINTR);
498 		}
499 		so = rep->r_nmp->nm_so;
500 		if (!so) {
501 			error = nfs_reconnect(rep);
502 			if (error) {
503 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
504 				return (error);
505 			}
506 			goto tryagain;
507 		}
508 		while (rep->r_flags & R_MUSTRESEND) {
509 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
510 			nfsstats.rpcretries++;
511 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
512 			if (error) {
513 				if (error == EINTR || error == ERESTART ||
514 				    (error = nfs_reconnect(rep)) != 0) {
515 					nfs_sndunlock(&rep->r_nmp->nm_iflag);
516 					return (error);
517 				}
518 				goto tryagain;
519 			}
520 		}
521 		nfs_sndunlock(&rep->r_nmp->nm_iflag);
522 		if (sotype == SOCK_STREAM) {
523 			aio.iov_base = (caddr_t) &len;
524 			aio.iov_len = sizeof(u_int32_t);
525 			auio.uio_iov = &aio;
526 			auio.uio_iovcnt = 1;
527 			auio.uio_segflg = UIO_SYSSPACE;
528 			auio.uio_rw = UIO_READ;
529 			auio.uio_offset = 0;
530 			auio.uio_resid = sizeof(u_int32_t);
531 			auio.uio_procp = p;
532 			do {
533 			   rcvflg = MSG_WAITALL;
534 			   error = (*so->so_receive)(so, (struct mbuf **)0, &auio,
535 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
536 			   if (error == EWOULDBLOCK && rep) {
537 				if (rep->r_flags & R_SOFTTERM)
538 					return (EINTR);
539 			   }
540 			} while (error == EWOULDBLOCK);
541 			if (!error && auio.uio_resid > 0) {
542 			    /*
543 			     * Don't log a 0 byte receive; it means
544 			     * that the socket has been closed, and
545 			     * can happen during normal operation
546 			     * (forcible unmount or Solaris server).
547 			     */
548 			    if (auio.uio_resid != sizeof (u_int32_t))
549 			      log(LOG_INFO,
550 				 "short receive (%lu/%lu) from nfs server %s\n",
551 				 (u_long)sizeof(u_int32_t) - auio.uio_resid,
552 				 (u_long)sizeof(u_int32_t),
553 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
554 			    error = EPIPE;
555 			}
556 			if (error)
557 				goto errout;
558 			len = ntohl(len) & ~0x80000000;
559 			/*
560 			 * This is SERIOUS! We are out of sync with the sender
561 			 * and forcing a disconnect/reconnect is all I can do.
562 			 */
563 			if (len > NFS_MAXPACKET) {
564 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
565 				"impossible packet length",
566 				len,
567 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
568 			    error = EFBIG;
569 			    goto errout;
570 			}
571 			auio.uio_resid = len;
572 			do {
573 			    rcvflg = MSG_WAITALL;
574 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
575 				&auio, mp, (struct mbuf **)0, &rcvflg);
576 			} while (error == EWOULDBLOCK || error == EINTR ||
577 				 error == ERESTART);
578 			if (!error && auio.uio_resid > 0) {
579 			    if (len != auio.uio_resid)
580 			      log(LOG_INFO,
581 				"short receive (%lu/%d) from nfs server %s\n",
582 				(u_long)len - auio.uio_resid, len,
583 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
584 			    error = EPIPE;
585 			}
586 		} else {
587 			/*
588 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
589 			 * and soreceive() will return when it has either a
590 			 * control msg or a data msg.
591 			 * We have no use for control msg., but must grab them
592 			 * and then throw them away so we know what is going
593 			 * on.
594 			 */
595 			auio.uio_resid = len = 100000000; /* Anything Big */
596 			auio.uio_procp = p;
597 			do {
598 			    rcvflg = 0;
599 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
600 				&auio, mp, &control, &rcvflg);
601 			    if (control)
602 				m_freem(control);
603 			    if (error == EWOULDBLOCK && rep) {
604 				if (rep->r_flags & R_SOFTTERM)
605 					return (EINTR);
606 			    }
607 			} while (error == EWOULDBLOCK ||
608 				 (!error && *mp == NULL && control));
609 			if ((rcvflg & MSG_EOR) == 0)
610 				printf("Egad!!\n");
611 			if (!error && *mp == NULL)
612 				error = EPIPE;
613 			len -= auio.uio_resid;
614 		}
615 errout:
616 		if (error && error != EINTR && error != ERESTART) {
617 			m_freem(*mp);
618 			*mp = (struct mbuf *)0;
619 			if (error != EPIPE)
620 				log(LOG_INFO,
621 				    "receive error %d from nfs server %s\n",
622 				    error,
623 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
624 			error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
625 			if (!error)
626 				error = nfs_reconnect(rep);
627 			if (!error)
628 				goto tryagain;
629 			else
630 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
631 		}
632 	} else {
633 		if ((so = rep->r_nmp->nm_so) == NULL)
634 			return (EACCES);
635 		if (so->so_state & SS_ISCONNECTED)
636 			getnam = (struct mbuf **)0;
637 		else
638 			getnam = aname;
639 		auio.uio_resid = len = 1000000;
640 		auio.uio_procp = p;
641 		do {
642 			rcvflg = 0;
643 			error =  (*so->so_receive)(so, getnam, &auio, mp,
644 				(struct mbuf **)0, &rcvflg);
645 			if (error == EWOULDBLOCK &&
646 			    (rep->r_flags & R_SOFTTERM))
647 				return (EINTR);
648 		} while (error == EWOULDBLOCK);
649 		len -= auio.uio_resid;
650 		if (!error && *mp == NULL)
651 			error = EPIPE;
652 	}
653 	if (error) {
654 		m_freem(*mp);
655 		*mp = (struct mbuf *)0;
656 	}
657 	return (error);
658 }
659 
660 /*
661  * Implement receipt of reply on a socket.
662  * We must search through the list of received datagrams matching them
663  * with outstanding requests using the xid, until ours is found.
664  */
665 /* ARGSUSED */
666 int
667 nfs_reply(myrep)
668 	struct nfsreq *myrep;
669 {
670 	register struct nfsreq *rep;
671 	register struct nfsmount *nmp = myrep->r_nmp;
672 	register int32_t t1;
673 	struct mbuf *mrep, *nam, *md;
674 	u_int32_t rxid, *tl;
675 	caddr_t dpos, cp2;
676 	int error;
677 
678 	/*
679 	 * Loop around until we get our own reply
680 	 */
681 	for (;;) {
682 		/*
683 		 * Lock against other receivers so that I don't get stuck in
684 		 * sbwait() after someone else has received my reply for me.
685 		 * Also necessary for connection based protocols to avoid
686 		 * race conditions during a reconnect.
687 		 */
688 		error = nfs_rcvlock(myrep);
689 		if (error == EALREADY)
690 			return (0);
691 		if (error)
692 			return (error);
693 		/*
694 		 * Get the next Rpc reply off the socket
695 		 */
696 		nmp->nm_waiters++;
697 		error = nfs_receive(myrep, &nam, &mrep);
698 		nfs_rcvunlock(&nmp->nm_iflag);
699 		if (error) {
700 
701 			if (nmp->nm_iflag & NFSMNT_DISMNT) {
702 				/*
703 				 * Oops, we're going away now..
704 				 */
705 				nmp->nm_waiters--;
706 				wakeup (&nmp->nm_waiters);
707 				return error;
708 			}
709 			nmp->nm_waiters--;
710 			/*
711 			 * Ignore routing errors on connectionless protocols??
712 			 */
713 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
714 				nmp->nm_so->so_error = 0;
715 #ifdef DEBUG
716 				printf("nfs_reply: ignoring error %d\n", error);
717 #endif
718 				if (myrep->r_flags & R_GETONEREP)
719 					return (0);
720 				continue;
721 			}
722 			return (error);
723 		}
724 		nmp->nm_waiters--;
725 		if (nam)
726 			m_freem(nam);
727 
728 		/*
729 		 * Get the xid and check that it is an rpc reply
730 		 */
731 		md = mrep;
732 		dpos = mtod(md, caddr_t);
733 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
734 		rxid = *tl++;
735 		if (*tl != rpc_reply) {
736 			if (nmp->nm_flag & NFSMNT_NQNFS) {
737 				if (nqnfs_callback(nmp, mrep, md, dpos))
738 					nfsstats.rpcinvalid++;
739 			} else {
740 				nfsstats.rpcinvalid++;
741 				m_freem(mrep);
742 			}
743 nfsmout:
744 			if (myrep->r_flags & R_GETONEREP)
745 				return (0);
746 			continue;
747 		}
748 
749 		/*
750 		 * Loop through the request list to match up the reply
751 		 * Iff no match, just drop the datagram
752 		 */
753 		for (rep = nfs_reqq.tqh_first; rep != 0;
754 		    rep = rep->r_chain.tqe_next) {
755 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
756 				/* Found it.. */
757 				rep->r_mrep = mrep;
758 				rep->r_md = md;
759 				rep->r_dpos = dpos;
760 				if (nfsrtton) {
761 					struct rttl *rt;
762 
763 					rt = &nfsrtt.rttl[nfsrtt.pos];
764 					rt->proc = rep->r_procnum;
765 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
766 					rt->sent = nmp->nm_sent;
767 					rt->cwnd = nmp->nm_cwnd;
768 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
769 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
770 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
771 					rt->tstamp = time;
772 					if (rep->r_flags & R_TIMING)
773 						rt->rtt = rep->r_rtt;
774 					else
775 						rt->rtt = 1000000;
776 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
777 				}
778 				/*
779 				 * Update congestion window.
780 				 * Do the additive increase of
781 				 * one rpc/rtt.
782 				 */
783 				if (nmp->nm_cwnd <= nmp->nm_sent) {
784 					nmp->nm_cwnd +=
785 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
786 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
787 					if (nmp->nm_cwnd > NFS_MAXCWND)
788 						nmp->nm_cwnd = NFS_MAXCWND;
789 				}
790 				rep->r_flags &= ~R_SENT;
791 				nmp->nm_sent -= NFS_CWNDSCALE;
792 				/*
793 				 * Update rtt using a gain of 0.125 on the mean
794 				 * and a gain of 0.25 on the deviation.
795 				 */
796 				if (rep->r_flags & R_TIMING) {
797 					/*
798 					 * Since the timer resolution of
799 					 * NFS_HZ is so course, it can often
800 					 * result in r_rtt == 0. Since
801 					 * r_rtt == N means that the actual
802 					 * rtt is between N+dt and N+2-dt ticks,
803 					 * add 1.
804 					 */
805 					t1 = rep->r_rtt + 1;
806 					t1 -= (NFS_SRTT(rep) >> 3);
807 					NFS_SRTT(rep) += t1;
808 					if (t1 < 0)
809 						t1 = -t1;
810 					t1 -= (NFS_SDRTT(rep) >> 2);
811 					NFS_SDRTT(rep) += t1;
812 				}
813 				nmp->nm_timeouts = 0;
814 				break;
815 			}
816 		}
817 		/*
818 		 * If not matched to a request, drop it.
819 		 * If it's mine, get out.
820 		 */
821 		if (rep == 0) {
822 			nfsstats.rpcunexpected++;
823 			m_freem(mrep);
824 		} else if (rep == myrep) {
825 			if (rep->r_mrep == NULL)
826 				panic("nfsreply nil");
827 			return (0);
828 		}
829 		if (myrep->r_flags & R_GETONEREP)
830 			return (0);
831 	}
832 }
833 
834 /*
835  * nfs_request - goes something like this
836  *	- fill in request struct
837  *	- links it into list
838  *	- calls nfs_send() for first transmit
839  *	- calls nfs_receive() to get reply
840  *	- break down rpc header and return with nfs reply pointed to
841  *	  by mrep or error
842  * nb: always frees up mreq mbuf list
843  */
844 int
845 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
846 	struct vnode *vp;
847 	struct mbuf *mrest;
848 	int procnum;
849 	struct proc *procp;
850 	struct ucred *cred;
851 	struct mbuf **mrp;
852 	struct mbuf **mdp;
853 	caddr_t *dposp;
854 {
855 	register struct mbuf *m, *mrep;
856 	register struct nfsreq *rep;
857 	register u_int32_t *tl;
858 	register int i;
859 	struct nfsmount *nmp;
860 	struct mbuf *md, *mheadend;
861 	struct nfsnode *np;
862 	char nickv[RPCX_NICKVERF];
863 	time_t reqtime, waituntil;
864 	caddr_t dpos, cp2;
865 	int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type;
866 	int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0;
867 	int verf_len, verf_type;
868 	u_int32_t xid;
869 	u_quad_t frev;
870 	char *auth_str, *verf_str;
871 	NFSKERBKEY_T key;		/* save session key */
872 
873 	nmp = VFSTONFS(vp->v_mount);
874 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
875 	rep->r_nmp = nmp;
876 	rep->r_vp = vp;
877 	rep->r_procp = procp;
878 	rep->r_procnum = procnum;
879 	i = 0;
880 	m = mrest;
881 	while (m) {
882 		i += m->m_len;
883 		m = m->m_next;
884 	}
885 	mrest_len = i;
886 
887 	/*
888 	 * Get the RPC header with authorization.
889 	 */
890 kerbauth:
891 	verf_str = auth_str = (char *)0;
892 	if (nmp->nm_flag & NFSMNT_KERB) {
893 		verf_str = nickv;
894 		verf_len = sizeof (nickv);
895 		auth_type = RPCAUTH_KERB4;
896 		memset((caddr_t)key, 0, sizeof (key));
897 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
898 			&auth_len, verf_str, verf_len)) {
899 			error = nfs_getauth(nmp, rep, cred, &auth_str,
900 				&auth_len, verf_str, &verf_len, key);
901 			if (error) {
902 				free((caddr_t)rep, M_NFSREQ);
903 				m_freem(mrest);
904 				return (error);
905 			}
906 		}
907 	} else {
908 		auth_type = RPCAUTH_UNIX;
909 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
910 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
911 			5 * NFSX_UNSIGNED;
912 	}
913 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
914 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
915 	if (auth_str)
916 		free(auth_str, M_TEMP);
917 
918 	/*
919 	 * For stream protocols, insert a Sun RPC Record Mark.
920 	 */
921 	if (nmp->nm_sotype == SOCK_STREAM) {
922 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
923 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
924 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
925 	}
926 	rep->r_mreq = m;
927 	rep->r_xid = xid;
928 tryagain:
929 	if (nmp->nm_flag & NFSMNT_SOFT)
930 		rep->r_retry = nmp->nm_retry;
931 	else
932 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
933 	rep->r_rtt = rep->r_rexmit = 0;
934 	if (proct[procnum] > 0)
935 		rep->r_flags = R_TIMING;
936 	else
937 		rep->r_flags = 0;
938 	rep->r_mrep = NULL;
939 
940 	/*
941 	 * Do the client side RPC.
942 	 */
943 	nfsstats.rpcrequests++;
944 	/*
945 	 * Chain request into list of outstanding requests. Be sure
946 	 * to put it LAST so timer finds oldest requests first.
947 	 */
948 	s = splsoftnet();
949 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
950 
951 	/* Get send time for nqnfs */
952 	reqtime = time.tv_sec;
953 
954 	/*
955 	 * If backing off another request or avoiding congestion, don't
956 	 * send this one now but let timer do it. If not timing a request,
957 	 * do it now.
958 	 */
959 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
960 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
961 		nmp->nm_sent < nmp->nm_cwnd)) {
962 		splx(s);
963 		if (nmp->nm_soflags & PR_CONNREQUIRED)
964 			error = nfs_sndlock(&nmp->nm_iflag, rep);
965 		if (!error) {
966 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
967 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep);
968 			if (nmp->nm_soflags & PR_CONNREQUIRED)
969 				nfs_sndunlock(&nmp->nm_iflag);
970 		}
971 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
972 			nmp->nm_sent += NFS_CWNDSCALE;
973 			rep->r_flags |= R_SENT;
974 		}
975 	} else {
976 		splx(s);
977 		rep->r_rtt = -1;
978 	}
979 
980 	/*
981 	 * Wait for the reply from our send or the timer's.
982 	 */
983 	if (!error || error == EPIPE)
984 		error = nfs_reply(rep);
985 
986 	/*
987 	 * RPC done, unlink the request.
988 	 */
989 	s = splsoftnet();
990 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
991 	splx(s);
992 
993 	/*
994 	 * Decrement the outstanding request count.
995 	 */
996 	if (rep->r_flags & R_SENT) {
997 		rep->r_flags &= ~R_SENT;	/* paranoia */
998 		nmp->nm_sent -= NFS_CWNDSCALE;
999 	}
1000 
1001 	/*
1002 	 * If there was a successful reply and a tprintf msg.
1003 	 * tprintf a response.
1004 	 */
1005 	if (!error && (rep->r_flags & R_TPRINTFMSG))
1006 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1007 		    "is alive again");
1008 	mrep = rep->r_mrep;
1009 	md = rep->r_md;
1010 	dpos = rep->r_dpos;
1011 	if (error) {
1012 		m_freem(rep->r_mreq);
1013 		free((caddr_t)rep, M_NFSREQ);
1014 		return (error);
1015 	}
1016 
1017 	/*
1018 	 * break down the rpc header and check if ok
1019 	 */
1020 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1021 	if (*tl++ == rpc_msgdenied) {
1022 		if (*tl == rpc_mismatch)
1023 			error = EOPNOTSUPP;
1024 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1025 			if (!failed_auth) {
1026 				failed_auth++;
1027 				mheadend->m_next = (struct mbuf *)0;
1028 				m_freem(mrep);
1029 				m_freem(rep->r_mreq);
1030 				goto kerbauth;
1031 			} else
1032 				error = EAUTH;
1033 		} else
1034 			error = EACCES;
1035 		m_freem(mrep);
1036 		m_freem(rep->r_mreq);
1037 		free((caddr_t)rep, M_NFSREQ);
1038 		return (error);
1039 	}
1040 
1041 	/*
1042 	 * Grab any Kerberos verifier, otherwise just throw it away.
1043 	 */
1044 	verf_type = fxdr_unsigned(int, *tl++);
1045 	i = fxdr_unsigned(int32_t, *tl);
1046 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1047 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1048 		if (error)
1049 			goto nfsmout;
1050 	} else if (i > 0)
1051 		nfsm_adv(nfsm_rndup(i));
1052 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1053 	/* 0 == ok */
1054 	if (*tl == 0) {
1055 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1056 		if (*tl != 0) {
1057 			error = fxdr_unsigned(int, *tl);
1058 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1059 				error == NFSERR_TRYLATER) {
1060 				m_freem(mrep);
1061 				error = 0;
1062 				waituntil = time.tv_sec + trylater_delay;
1063 				while (time.tv_sec < waituntil)
1064 					(void) tsleep((caddr_t)&lbolt,
1065 						PSOCK, "nqnfstry", 0);
1066 				trylater_delay *= nfs_backoff[trylater_cnt];
1067 				if (trylater_cnt < 7)
1068 					trylater_cnt++;
1069 				goto tryagain;
1070 			}
1071 
1072 			/*
1073 			 * If the File Handle was stale, invalidate the
1074 			 * lookup cache, just in case.
1075 			 */
1076 			if (error == ESTALE)
1077 				cache_purge(vp);
1078 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1079 				*mrp = mrep;
1080 				*mdp = md;
1081 				*dposp = dpos;
1082 				error |= NFSERR_RETERR;
1083 			} else
1084 				m_freem(mrep);
1085 			m_freem(rep->r_mreq);
1086 			free((caddr_t)rep, M_NFSREQ);
1087 			return (error);
1088 		}
1089 
1090 		/*
1091 		 * For nqnfs, get any lease in reply
1092 		 */
1093 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1094 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1095 			if (*tl) {
1096 				np = VTONFS(vp);
1097 				nqlflag = fxdr_unsigned(int, *tl);
1098 				nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1099 				cachable = fxdr_unsigned(int, *tl++);
1100 				reqtime += fxdr_unsigned(int, *tl++);
1101 				if (reqtime > time.tv_sec) {
1102 				    frev = fxdr_hyper(tl);
1103 				    nqnfs_clientlease(nmp, np, nqlflag,
1104 					cachable, reqtime, frev);
1105 				}
1106 			}
1107 		}
1108 		*mrp = mrep;
1109 		*mdp = md;
1110 		*dposp = dpos;
1111 		m_freem(rep->r_mreq);
1112 		FREE((caddr_t)rep, M_NFSREQ);
1113 		return (0);
1114 	}
1115 	m_freem(mrep);
1116 	error = EPROTONOSUPPORT;
1117 nfsmout:
1118 	m_freem(rep->r_mreq);
1119 	free((caddr_t)rep, M_NFSREQ);
1120 	return (error);
1121 }
1122 #endif /* NFS */
1123 
1124 /*
1125  * Generate the rpc reply header
1126  * siz arg. is used to decide if adding a cluster is worthwhile
1127  */
1128 int
1129 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1130 	int siz;
1131 	struct nfsrv_descript *nd;
1132 	struct nfssvc_sock *slp;
1133 	int err;
1134 	int cache;
1135 	u_quad_t *frev;
1136 	struct mbuf **mrq;
1137 	struct mbuf **mbp;
1138 	caddr_t *bposp;
1139 {
1140 	register u_int32_t *tl;
1141 	register struct mbuf *mreq;
1142 	caddr_t bpos;
1143 	struct mbuf *mb, *mb2;
1144 
1145 	MGETHDR(mreq, M_WAIT, MT_DATA);
1146 	mb = mreq;
1147 	/*
1148 	 * If this is a big reply, use a cluster else
1149 	 * try and leave leading space for the lower level headers.
1150 	 */
1151 	siz += RPC_REPLYSIZ;
1152 	if (siz >= max_datalen) {
1153 		MCLGET(mreq, M_WAIT);
1154 	} else
1155 		mreq->m_data += max_hdr;
1156 	tl = mtod(mreq, u_int32_t *);
1157 	mreq->m_len = 6 * NFSX_UNSIGNED;
1158 	bpos = ((caddr_t)tl) + mreq->m_len;
1159 	*tl++ = txdr_unsigned(nd->nd_retxid);
1160 	*tl++ = rpc_reply;
1161 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1162 		*tl++ = rpc_msgdenied;
1163 		if (err & NFSERR_AUTHERR) {
1164 			*tl++ = rpc_autherr;
1165 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1166 			mreq->m_len -= NFSX_UNSIGNED;
1167 			bpos -= NFSX_UNSIGNED;
1168 		} else {
1169 			*tl++ = rpc_mismatch;
1170 			*tl++ = txdr_unsigned(RPC_VER2);
1171 			*tl = txdr_unsigned(RPC_VER2);
1172 		}
1173 	} else {
1174 		*tl++ = rpc_msgaccepted;
1175 
1176 		/*
1177 		 * For Kerberos authentication, we must send the nickname
1178 		 * verifier back, otherwise just RPCAUTH_NULL.
1179 		 */
1180 		if (nd->nd_flag & ND_KERBFULL) {
1181 		    register struct nfsuid *nuidp;
1182 		    struct timeval ktvin, ktvout;
1183 
1184 		    for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1185 			nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1186 			if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1187 			    (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1188 			     &nuidp->nu_haddr, nd->nd_nam2)))
1189 			    break;
1190 		    }
1191 		    if (nuidp) {
1192 			ktvin.tv_sec =
1193 			    txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1194 			ktvin.tv_usec =
1195 			    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1196 
1197 			/*
1198 			 * Encrypt the timestamp in ecb mode using the
1199 			 * session key.
1200 			 */
1201 #ifdef NFSKERB
1202 			XXX
1203 #endif
1204 
1205 			*tl++ = rpc_auth_kerb;
1206 			*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1207 			*tl = ktvout.tv_sec;
1208 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1209 			*tl++ = ktvout.tv_usec;
1210 			*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1211 		    } else {
1212 			*tl++ = 0;
1213 			*tl++ = 0;
1214 		    }
1215 		} else {
1216 			*tl++ = 0;
1217 			*tl++ = 0;
1218 		}
1219 		switch (err) {
1220 		case EPROGUNAVAIL:
1221 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1222 			break;
1223 		case EPROGMISMATCH:
1224 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1225 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1226 			if (nd->nd_flag & ND_NQNFS) {
1227 				*tl++ = txdr_unsigned(3);
1228 				*tl = txdr_unsigned(3);
1229 			} else {
1230 				*tl++ = txdr_unsigned(2);
1231 				*tl = txdr_unsigned(3);
1232 			}
1233 			break;
1234 		case EPROCUNAVAIL:
1235 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1236 			break;
1237 		case EBADRPC:
1238 			*tl = txdr_unsigned(RPC_GARBAGE);
1239 			break;
1240 		default:
1241 			*tl = 0;
1242 			if (err != NFSERR_RETVOID) {
1243 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1244 				if (err)
1245 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1246 				else
1247 				    *tl = 0;
1248 			}
1249 			break;
1250 		};
1251 	}
1252 
1253 	/*
1254 	 * For nqnfs, piggyback lease as requested.
1255 	 */
1256 	if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1257 		if (nd->nd_flag & ND_LEASE) {
1258 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1259 			*tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1260 			*tl++ = txdr_unsigned(cache);
1261 			*tl++ = txdr_unsigned(nd->nd_duration);
1262 			txdr_hyper(*frev, tl);
1263 		} else {
1264 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1265 			*tl = 0;
1266 		}
1267 	}
1268 	if (mrq != NULL)
1269 		*mrq = mreq;
1270 	*mbp = mb;
1271 	*bposp = bpos;
1272 	if (err != 0 && err != NFSERR_RETVOID)
1273 		nfsstats.srvrpc_errs++;
1274 	return (0);
1275 }
1276 
1277 /*
1278  * Nfs timer routine
1279  * Scan the nfsreq list and retranmit any requests that have timed out
1280  * To avoid retransmission attempts on STREAM sockets (in the future) make
1281  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1282  */
1283 void
1284 nfs_timer(arg)
1285 	void *arg;	/* never used */
1286 {
1287 	register struct nfsreq *rep;
1288 	register struct mbuf *m;
1289 	register struct socket *so;
1290 	register struct nfsmount *nmp;
1291 	register int timeo;
1292 	int s, error;
1293 #ifdef NFSSERVER
1294 	register struct nfssvc_sock *slp;
1295 	static long lasttime = 0;
1296 	u_quad_t cur_usec;
1297 #endif
1298 
1299 	s = splsoftnet();
1300 	for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) {
1301 		nmp = rep->r_nmp;
1302 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1303 			continue;
1304 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1305 			rep->r_flags |= R_SOFTTERM;
1306 			continue;
1307 		}
1308 		if (rep->r_rtt >= 0) {
1309 			rep->r_rtt++;
1310 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1311 				timeo = nmp->nm_timeo;
1312 			else
1313 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1314 			if (nmp->nm_timeouts > 0)
1315 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1316 			if (rep->r_rtt <= timeo)
1317 				continue;
1318 			if (nmp->nm_timeouts < 8)
1319 				nmp->nm_timeouts++;
1320 		}
1321 		/*
1322 		 * Check for server not responding
1323 		 */
1324 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1325 		     rep->r_rexmit > nmp->nm_deadthresh) {
1326 			nfs_msg(rep->r_procp,
1327 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1328 			    "not responding");
1329 			rep->r_flags |= R_TPRINTFMSG;
1330 		}
1331 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1332 			nfsstats.rpctimeouts++;
1333 			rep->r_flags |= R_SOFTTERM;
1334 			continue;
1335 		}
1336 		if (nmp->nm_sotype != SOCK_DGRAM) {
1337 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1338 				rep->r_rexmit = NFS_MAXREXMIT;
1339 			continue;
1340 		}
1341 		if ((so = nmp->nm_so) == NULL)
1342 			continue;
1343 
1344 		/*
1345 		 * If there is enough space and the window allows..
1346 		 *	Resend it
1347 		 * Set r_rtt to -1 in case we fail to send it now.
1348 		 */
1349 		rep->r_rtt = -1;
1350 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1351 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1352 		    (rep->r_flags & R_SENT) ||
1353 		    nmp->nm_sent < nmp->nm_cwnd) &&
1354 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1355 		        if (so->so_state & SS_ISCONNECTED)
1356 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1357 			    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1358 			else
1359 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1360 			    nmp->nm_nam, (struct mbuf *)0, (struct proc *)0);
1361 			if (error) {
1362 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1363 #ifdef DEBUG
1364 					printf("nfs_timer: ignoring error %d\n",
1365 						error);
1366 #endif
1367 					so->so_error = 0;
1368 				}
1369 			} else {
1370 				/*
1371 				 * Iff first send, start timing
1372 				 * else turn timing off, backoff timer
1373 				 * and divide congestion window by 2.
1374 				 */
1375 				if (rep->r_flags & R_SENT) {
1376 					rep->r_flags &= ~R_TIMING;
1377 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1378 						rep->r_rexmit = NFS_MAXREXMIT;
1379 					nmp->nm_cwnd >>= 1;
1380 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1381 						nmp->nm_cwnd = NFS_CWNDSCALE;
1382 					nfsstats.rpcretries++;
1383 				} else {
1384 					rep->r_flags |= R_SENT;
1385 					nmp->nm_sent += NFS_CWNDSCALE;
1386 				}
1387 				rep->r_rtt = 0;
1388 			}
1389 		}
1390 	}
1391 
1392 #ifdef NFSSERVER
1393 	/*
1394 	 * Call the nqnfs server timer once a second to handle leases.
1395 	 */
1396 	if (lasttime != time.tv_sec) {
1397 		lasttime = time.tv_sec;
1398 		nqnfs_serverd();
1399 	}
1400 
1401 	/*
1402 	 * Scan the write gathering queues for writes that need to be
1403 	 * completed now.
1404 	 */
1405 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1406 	for (slp = nfssvc_sockhead.tqh_first; slp != 0;
1407 	    slp = slp->ns_chain.tqe_next) {
1408 	    if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1409 		nfsrv_wakenfsd(slp);
1410 	}
1411 #endif /* NFSSERVER */
1412 	splx(s);
1413 	timeout(nfs_timer, (void *)0, nfs_ticks);
1414 }
1415 
1416 /*
1417  * Test for a termination condition pending on the process.
1418  * This is used for NFSMNT_INT mounts.
1419  */
1420 int
1421 nfs_sigintr(nmp, rep, p)
1422 	struct nfsmount *nmp;
1423 	struct nfsreq *rep;
1424 	register struct proc *p;
1425 {
1426 	sigset_t ss;
1427 
1428 	if (rep && (rep->r_flags & R_SOFTTERM))
1429 		return (EINTR);
1430 	if (!(nmp->nm_flag & NFSMNT_INT))
1431 		return (0);
1432 	if (p) {
1433 		sigpending1(p, &ss);
1434 #if 0
1435 		sigminusset(&p->p_sigignore, &ss);
1436 #endif
1437 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1438 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1439 		    sigismember(&ss, SIGQUIT))
1440 			return (EINTR);
1441 	}
1442 	return (0);
1443 }
1444 
1445 /*
1446  * Lock a socket against others.
1447  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1448  * and also to avoid race conditions between the processes with nfs requests
1449  * in progress when a reconnect is necessary.
1450  */
1451 int
1452 nfs_sndlock(flagp, rep)
1453 	register int *flagp;
1454 	struct nfsreq *rep;
1455 {
1456 	struct proc *p;
1457 	int slpflag = 0, slptimeo = 0;
1458 
1459 	if (rep) {
1460 		p = rep->r_procp;
1461 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1462 			slpflag = PCATCH;
1463 	} else
1464 		p = (struct proc *)0;
1465 	while (*flagp & NFSMNT_SNDLOCK) {
1466 		if (nfs_sigintr(rep->r_nmp, rep, p))
1467 			return (EINTR);
1468 		*flagp |= NFSMNT_WANTSND;
1469 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1470 			slptimeo);
1471 		if (slpflag == PCATCH) {
1472 			slpflag = 0;
1473 			slptimeo = 2 * hz;
1474 		}
1475 	}
1476 	*flagp |= NFSMNT_SNDLOCK;
1477 	return (0);
1478 }
1479 
1480 /*
1481  * Unlock the stream socket for others.
1482  */
1483 void
1484 nfs_sndunlock(flagp)
1485 	register int *flagp;
1486 {
1487 
1488 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1489 		panic("nfs sndunlock");
1490 	*flagp &= ~NFSMNT_SNDLOCK;
1491 	if (*flagp & NFSMNT_WANTSND) {
1492 		*flagp &= ~NFSMNT_WANTSND;
1493 		wakeup((caddr_t)flagp);
1494 	}
1495 }
1496 
1497 int
1498 nfs_rcvlock(rep)
1499 	register struct nfsreq *rep;
1500 {
1501 	struct nfsmount *nmp = rep->r_nmp;
1502 	register int *flagp = &nmp->nm_iflag;
1503 	int slpflag, slptimeo = 0;
1504 
1505 	if (*flagp & NFSMNT_DISMNT)
1506 		return EIO;
1507 
1508 	if (*flagp & NFSMNT_INT)
1509 		slpflag = PCATCH;
1510 	else
1511 		slpflag = 0;
1512 	while (*flagp & NFSMNT_RCVLOCK) {
1513 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1514 			return (EINTR);
1515 		*flagp |= NFSMNT_WANTRCV;
1516 		nmp->nm_waiters++;
1517 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1518 			slptimeo);
1519 		nmp->nm_waiters--;
1520 		if (*flagp & NFSMNT_DISMNT) {
1521 			wakeup(&nmp->nm_waiters);
1522 			return EIO;
1523 		}
1524 		/* If our reply was received while we were sleeping,
1525 		 * then just return without taking the lock to avoid a
1526 		 * situation where a single iod could 'capture' the
1527 		 * receive lock.
1528 		 */
1529 		if (rep->r_mrep != NULL)
1530 			return (EALREADY);
1531 		if (slpflag == PCATCH) {
1532 			slpflag = 0;
1533 			slptimeo = 2 * hz;
1534 		}
1535 	}
1536 	*flagp |= NFSMNT_RCVLOCK;
1537 	return (0);
1538 }
1539 
1540 /*
1541  * Unlock the stream socket for others.
1542  */
1543 void
1544 nfs_rcvunlock(flagp)
1545 	register int *flagp;
1546 {
1547 
1548 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1549 		panic("nfs rcvunlock");
1550 	*flagp &= ~NFSMNT_RCVLOCK;
1551 	if (*flagp & NFSMNT_WANTRCV) {
1552 		*flagp &= ~NFSMNT_WANTRCV;
1553 		wakeup((caddr_t)flagp);
1554 	}
1555 }
1556 
1557 /*
1558  * Parse an RPC request
1559  * - verify it
1560  * - fill in the cred struct.
1561  */
1562 int
1563 nfs_getreq(nd, nfsd, has_header)
1564 	register struct nfsrv_descript *nd;
1565 	struct nfsd *nfsd;
1566 	int has_header;
1567 {
1568 	register int len, i;
1569 	register u_int32_t *tl;
1570 	register int32_t t1;
1571 	struct uio uio;
1572 	struct iovec iov;
1573 	caddr_t dpos, cp2, cp;
1574 	u_int32_t nfsvers, auth_type;
1575 	uid_t nickuid;
1576 	int error = 0, nqnfs = 0, ticklen;
1577 	struct mbuf *mrep, *md;
1578 	register struct nfsuid *nuidp;
1579 	struct timeval tvin, tvout;
1580 
1581 	mrep = nd->nd_mrep;
1582 	md = nd->nd_md;
1583 	dpos = nd->nd_dpos;
1584 	if (has_header) {
1585 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1586 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1587 		if (*tl++ != rpc_call) {
1588 			m_freem(mrep);
1589 			return (EBADRPC);
1590 		}
1591 	} else
1592 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1593 	nd->nd_repstat = 0;
1594 	nd->nd_flag = 0;
1595 	if (*tl++ != rpc_vers) {
1596 		nd->nd_repstat = ERPCMISMATCH;
1597 		nd->nd_procnum = NFSPROC_NOOP;
1598 		return (0);
1599 	}
1600 	if (*tl != nfs_prog) {
1601 		if (*tl == nqnfs_prog)
1602 			nqnfs++;
1603 		else {
1604 			nd->nd_repstat = EPROGUNAVAIL;
1605 			nd->nd_procnum = NFSPROC_NOOP;
1606 			return (0);
1607 		}
1608 	}
1609 	tl++;
1610 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1611 	if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1612 		(nfsvers != NQNFS_VER3 && nqnfs)) {
1613 		nd->nd_repstat = EPROGMISMATCH;
1614 		nd->nd_procnum = NFSPROC_NOOP;
1615 		return (0);
1616 	}
1617 	if (nqnfs)
1618 		nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1619 	else if (nfsvers == NFS_VER3)
1620 		nd->nd_flag = ND_NFSV3;
1621 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1622 	if (nd->nd_procnum == NFSPROC_NULL)
1623 		return (0);
1624 	if (nd->nd_procnum >= NFS_NPROCS ||
1625 		(!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1626 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1627 		nd->nd_repstat = EPROCUNAVAIL;
1628 		nd->nd_procnum = NFSPROC_NOOP;
1629 		return (0);
1630 	}
1631 	if ((nd->nd_flag & ND_NFSV3) == 0)
1632 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1633 	auth_type = *tl++;
1634 	len = fxdr_unsigned(int, *tl++);
1635 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1636 		m_freem(mrep);
1637 		return (EBADRPC);
1638 	}
1639 
1640 	nd->nd_flag &= ~ND_KERBAUTH;
1641 	/*
1642 	 * Handle auth_unix or auth_kerb.
1643 	 */
1644 	if (auth_type == rpc_auth_unix) {
1645 		len = fxdr_unsigned(int, *++tl);
1646 		if (len < 0 || len > NFS_MAXNAMLEN) {
1647 			m_freem(mrep);
1648 			return (EBADRPC);
1649 		}
1650 		nfsm_adv(nfsm_rndup(len));
1651 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1652 		memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred));
1653 		nd->nd_cr.cr_ref = 1;
1654 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1655 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1656 		len = fxdr_unsigned(int, *tl);
1657 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1658 			m_freem(mrep);
1659 			return (EBADRPC);
1660 		}
1661 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1662 		for (i = 0; i < len; i++)
1663 		    if (i < NGROUPS)
1664 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1665 		    else
1666 			tl++;
1667 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1668 		if (nd->nd_cr.cr_ngroups > 1)
1669 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1670 		len = fxdr_unsigned(int, *++tl);
1671 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1672 			m_freem(mrep);
1673 			return (EBADRPC);
1674 		}
1675 		if (len > 0)
1676 			nfsm_adv(nfsm_rndup(len));
1677 	} else if (auth_type == rpc_auth_kerb) {
1678 		switch (fxdr_unsigned(int, *tl++)) {
1679 		case RPCAKN_FULLNAME:
1680 			ticklen = fxdr_unsigned(int, *tl);
1681 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1682 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1683 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1684 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1685 				m_freem(mrep);
1686 				return (EBADRPC);
1687 			}
1688 			uio.uio_offset = 0;
1689 			uio.uio_iov = &iov;
1690 			uio.uio_iovcnt = 1;
1691 			uio.uio_segflg = UIO_SYSSPACE;
1692 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1693 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1694 			nfsm_mtouio(&uio, uio.uio_resid);
1695 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1696 			if (*tl++ != rpc_auth_kerb ||
1697 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1698 				printf("Bad kerb verifier\n");
1699 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1700 				nd->nd_procnum = NFSPROC_NOOP;
1701 				return (0);
1702 			}
1703 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1704 			tl = (u_int32_t *)cp;
1705 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1706 				printf("Not fullname kerb verifier\n");
1707 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1708 				nd->nd_procnum = NFSPROC_NOOP;
1709 				return (0);
1710 			}
1711 			cp += NFSX_UNSIGNED;
1712 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
1713 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1714 			nd->nd_flag |= ND_KERBFULL;
1715 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1716 			break;
1717 		case RPCAKN_NICKNAME:
1718 			if (len != 2 * NFSX_UNSIGNED) {
1719 				printf("Kerb nickname short\n");
1720 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1721 				nd->nd_procnum = NFSPROC_NOOP;
1722 				return (0);
1723 			}
1724 			nickuid = fxdr_unsigned(uid_t, *tl);
1725 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1726 			if (*tl++ != rpc_auth_kerb ||
1727 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1728 				printf("Kerb nick verifier bad\n");
1729 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1730 				nd->nd_procnum = NFSPROC_NOOP;
1731 				return (0);
1732 			}
1733 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1734 			tvin.tv_sec = *tl++;
1735 			tvin.tv_usec = *tl;
1736 
1737 			for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1738 			    nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1739 				if (nuidp->nu_cr.cr_uid == nickuid &&
1740 				    (!nd->nd_nam2 ||
1741 				     netaddr_match(NU_NETFAM(nuidp),
1742 				      &nuidp->nu_haddr, nd->nd_nam2)))
1743 					break;
1744 			}
1745 			if (!nuidp) {
1746 				nd->nd_repstat =
1747 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1748 				nd->nd_procnum = NFSPROC_NOOP;
1749 				return (0);
1750 			}
1751 
1752 			/*
1753 			 * Now, decrypt the timestamp using the session key
1754 			 * and validate it.
1755 			 */
1756 #ifdef NFSKERB
1757 			XXX
1758 #endif
1759 
1760 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1761 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1762 			if (nuidp->nu_expire < time.tv_sec ||
1763 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1764 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1765 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1766 				nuidp->nu_expire = 0;
1767 				nd->nd_repstat =
1768 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1769 				nd->nd_procnum = NFSPROC_NOOP;
1770 				return (0);
1771 			}
1772 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1773 			nd->nd_flag |= ND_KERBNICK;
1774 		};
1775 	} else {
1776 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1777 		nd->nd_procnum = NFSPROC_NOOP;
1778 		return (0);
1779 	}
1780 
1781 	/*
1782 	 * For nqnfs, get piggybacked lease request.
1783 	 */
1784 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
1785 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1786 		nd->nd_flag |= fxdr_unsigned(int, *tl);
1787 		if (nd->nd_flag & ND_LEASE) {
1788 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1789 			nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
1790 		} else
1791 			nd->nd_duration = NQ_MINLEASE;
1792 	} else
1793 		nd->nd_duration = NQ_MINLEASE;
1794 	nd->nd_md = md;
1795 	nd->nd_dpos = dpos;
1796 	return (0);
1797 nfsmout:
1798 	return (error);
1799 }
1800 
1801 int
1802 nfs_msg(p, server, msg)
1803 	struct proc *p;
1804 	char *server, *msg;
1805 {
1806 	tpr_t tpr;
1807 
1808 	if (p)
1809 		tpr = tprintf_open(p);
1810 	else
1811 		tpr = NULL;
1812 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1813 	tprintf_close(tpr);
1814 	return (0);
1815 }
1816 
1817 #ifdef NFSSERVER
1818 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
1819 				    struct nfssvc_sock *, struct proc *,
1820 				    struct mbuf **)) = {
1821 	nfsrv_null,
1822 	nfsrv_getattr,
1823 	nfsrv_setattr,
1824 	nfsrv_lookup,
1825 	nfsrv3_access,
1826 	nfsrv_readlink,
1827 	nfsrv_read,
1828 	nfsrv_write,
1829 	nfsrv_create,
1830 	nfsrv_mkdir,
1831 	nfsrv_symlink,
1832 	nfsrv_mknod,
1833 	nfsrv_remove,
1834 	nfsrv_rmdir,
1835 	nfsrv_rename,
1836 	nfsrv_link,
1837 	nfsrv_readdir,
1838 	nfsrv_readdirplus,
1839 	nfsrv_statfs,
1840 	nfsrv_fsinfo,
1841 	nfsrv_pathconf,
1842 	nfsrv_commit,
1843 	nqnfsrv_getlease,
1844 	nqnfsrv_vacated,
1845 	nfsrv_noop,
1846 	nfsrv_noop
1847 };
1848 
1849 /*
1850  * Socket upcall routine for the nfsd sockets.
1851  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1852  * Essentially do as much as possible non-blocking, else punt and it will
1853  * be called with M_WAIT from an nfsd.
1854  */
1855 void
1856 nfsrv_rcv(so, arg, waitflag)
1857 	struct socket *so;
1858 	caddr_t arg;
1859 	int waitflag;
1860 {
1861 	register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1862 	register struct mbuf *m;
1863 	struct mbuf *mp, *nam;
1864 	struct uio auio;
1865 	int flags, error;
1866 
1867 	if ((slp->ns_flag & SLP_VALID) == 0)
1868 		return;
1869 #ifdef notdef
1870 	/*
1871 	 * Define this to test for nfsds handling this under heavy load.
1872 	 */
1873 	if (waitflag == M_DONTWAIT) {
1874 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1875 	}
1876 #endif
1877 	auio.uio_procp = NULL;
1878 	if (so->so_type == SOCK_STREAM) {
1879 		/*
1880 		 * If there are already records on the queue, defer soreceive()
1881 		 * to an nfsd so that there is feedback to the TCP layer that
1882 		 * the nfs servers are heavily loaded.
1883 		 */
1884 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1885 			slp->ns_flag |= SLP_NEEDQ;
1886 			goto dorecs;
1887 		}
1888 
1889 		/*
1890 		 * Do soreceive().
1891 		 */
1892 		auio.uio_resid = 1000000000;
1893 		flags = MSG_DONTWAIT;
1894 		error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1895 		if (error || mp == (struct mbuf *)0) {
1896 			if (error == EWOULDBLOCK)
1897 				slp->ns_flag |= SLP_NEEDQ;
1898 			else
1899 				slp->ns_flag |= SLP_DISCONN;
1900 			goto dorecs;
1901 		}
1902 		m = mp;
1903 		if (slp->ns_rawend) {
1904 			slp->ns_rawend->m_next = m;
1905 			slp->ns_cc += 1000000000 - auio.uio_resid;
1906 		} else {
1907 			slp->ns_raw = m;
1908 			slp->ns_cc = 1000000000 - auio.uio_resid;
1909 		}
1910 		while (m->m_next)
1911 			m = m->m_next;
1912 		slp->ns_rawend = m;
1913 
1914 		/*
1915 		 * Now try and parse record(s) out of the raw stream data.
1916 		 */
1917 		error = nfsrv_getstream(slp, waitflag);
1918 		if (error) {
1919 			if (error == EPERM)
1920 				slp->ns_flag |= SLP_DISCONN;
1921 			else
1922 				slp->ns_flag |= SLP_NEEDQ;
1923 		}
1924 	} else {
1925 		do {
1926 			auio.uio_resid = 1000000000;
1927 			flags = MSG_DONTWAIT;
1928 			error = (*so->so_receive)(so, &nam, &auio, &mp,
1929 						(struct mbuf **)0, &flags);
1930 			if (mp) {
1931 				if (nam) {
1932 					m = nam;
1933 					m->m_next = mp;
1934 				} else
1935 					m = mp;
1936 				if (slp->ns_recend)
1937 					slp->ns_recend->m_nextpkt = m;
1938 				else
1939 					slp->ns_rec = m;
1940 				slp->ns_recend = m;
1941 				m->m_nextpkt = (struct mbuf *)0;
1942 			}
1943 			if (error) {
1944 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
1945 					&& error != EWOULDBLOCK) {
1946 					slp->ns_flag |= SLP_DISCONN;
1947 					goto dorecs;
1948 				}
1949 			}
1950 		} while (mp);
1951 	}
1952 
1953 	/*
1954 	 * Now try and process the request records, non-blocking.
1955 	 */
1956 dorecs:
1957 	if (waitflag == M_DONTWAIT &&
1958 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
1959 		nfsrv_wakenfsd(slp);
1960 }
1961 
1962 /*
1963  * Try and extract an RPC request from the mbuf data list received on a
1964  * stream socket. The "waitflag" argument indicates whether or not it
1965  * can sleep.
1966  */
1967 int
1968 nfsrv_getstream(slp, waitflag)
1969 	register struct nfssvc_sock *slp;
1970 	int waitflag;
1971 {
1972 	register struct mbuf *m, **mpp;
1973 	register char *cp1, *cp2;
1974 	register int len;
1975 	struct mbuf *om, *m2, *recm = NULL;
1976 	u_int32_t recmark;
1977 
1978 	if (slp->ns_flag & SLP_GETSTREAM)
1979 		panic("nfs getstream");
1980 	slp->ns_flag |= SLP_GETSTREAM;
1981 	for (;;) {
1982 	    if (slp->ns_reclen == 0) {
1983 		if (slp->ns_cc < NFSX_UNSIGNED) {
1984 			slp->ns_flag &= ~SLP_GETSTREAM;
1985 			return (0);
1986 		}
1987 		m = slp->ns_raw;
1988 		if (m->m_len >= NFSX_UNSIGNED) {
1989 			memcpy((caddr_t)&recmark, mtod(m, caddr_t), NFSX_UNSIGNED);
1990 			m->m_data += NFSX_UNSIGNED;
1991 			m->m_len -= NFSX_UNSIGNED;
1992 		} else {
1993 			cp1 = (caddr_t)&recmark;
1994 			cp2 = mtod(m, caddr_t);
1995 			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
1996 				while (m->m_len == 0) {
1997 					m = m->m_next;
1998 					cp2 = mtod(m, caddr_t);
1999 				}
2000 				*cp1++ = *cp2++;
2001 				m->m_data++;
2002 				m->m_len--;
2003 			}
2004 		}
2005 		slp->ns_cc -= NFSX_UNSIGNED;
2006 		recmark = ntohl(recmark);
2007 		slp->ns_reclen = recmark & ~0x80000000;
2008 		if (recmark & 0x80000000)
2009 			slp->ns_flag |= SLP_LASTFRAG;
2010 		else
2011 			slp->ns_flag &= ~SLP_LASTFRAG;
2012 		if (slp->ns_reclen > NFS_MAXPACKET) {
2013 			slp->ns_flag &= ~SLP_GETSTREAM;
2014 			return (EPERM);
2015 		}
2016 	    }
2017 
2018 	    /*
2019 	     * Now get the record part.
2020 	     */
2021 	    if (slp->ns_cc == slp->ns_reclen) {
2022 		recm = slp->ns_raw;
2023 		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2024 		slp->ns_cc = slp->ns_reclen = 0;
2025 	    } else if (slp->ns_cc > slp->ns_reclen) {
2026 		len = 0;
2027 		m = slp->ns_raw;
2028 		om = (struct mbuf *)0;
2029 		while (len < slp->ns_reclen) {
2030 			if ((len + m->m_len) > slp->ns_reclen) {
2031 				size_t left = slp->ns_reclen - len;
2032 
2033 				MGETHDR(m2, waitflag, m->m_type);
2034 				if (m2 == NULL) {
2035 					slp->ns_flag &= ~SLP_GETSTREAM;
2036 					return (EWOULDBLOCK);
2037 				}
2038 				if (left > MHLEN) {
2039 					MCLGET(m2, waitflag);
2040 					if (!(m2->m_flags & M_EXT)) {
2041 						m_freem(m2);
2042 						slp->ns_flag &= ~SLP_GETSTREAM;
2043 						return (EWOULDBLOCK);
2044 					}
2045 				}
2046 				memcpy(mtod(m2, caddr_t), mtod(m, caddr_t),
2047 				    left);
2048 				m2->m_len = left;
2049 				m->m_data += left;
2050 				m->m_len -= left;
2051 				if (om) {
2052 					om->m_next = m2;
2053 					recm = slp->ns_raw;
2054 				} else
2055 					recm = m2;
2056 				len = slp->ns_reclen;
2057 			} else if ((len + m->m_len) == slp->ns_reclen) {
2058 				om = m;
2059 				len += m->m_len;
2060 				m = m->m_next;
2061 				recm = slp->ns_raw;
2062 				om->m_next = (struct mbuf *)0;
2063 			} else {
2064 				om = m;
2065 				len += m->m_len;
2066 				m = m->m_next;
2067 			}
2068 		}
2069 		slp->ns_raw = m;
2070 		slp->ns_cc -= len;
2071 		slp->ns_reclen = 0;
2072 	    } else {
2073 		slp->ns_flag &= ~SLP_GETSTREAM;
2074 		return (0);
2075 	    }
2076 
2077 	    /*
2078 	     * Accumulate the fragments into a record.
2079 	     */
2080 	    mpp = &slp->ns_frag;
2081 	    while (*mpp)
2082 		mpp = &((*mpp)->m_next);
2083 	    *mpp = recm;
2084 	    if (slp->ns_flag & SLP_LASTFRAG) {
2085 		if (slp->ns_recend)
2086 		    slp->ns_recend->m_nextpkt = slp->ns_frag;
2087 		else
2088 		    slp->ns_rec = slp->ns_frag;
2089 		slp->ns_recend = slp->ns_frag;
2090 		slp->ns_frag = (struct mbuf *)0;
2091 	    }
2092 	}
2093 }
2094 
2095 /*
2096  * Parse an RPC header.
2097  */
2098 int
2099 nfsrv_dorec(slp, nfsd, ndp)
2100 	register struct nfssvc_sock *slp;
2101 	struct nfsd *nfsd;
2102 	struct nfsrv_descript **ndp;
2103 {
2104 	register struct mbuf *m, *nam;
2105 	register struct nfsrv_descript *nd;
2106 	int error;
2107 
2108 	*ndp = NULL;
2109 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2110 	    (m = slp->ns_rec) == (struct mbuf *)0)
2111 		return (ENOBUFS);
2112 	slp->ns_rec = m->m_nextpkt;
2113 	if (slp->ns_rec)
2114 		m->m_nextpkt = (struct mbuf *)0;
2115 	else
2116 		slp->ns_recend = (struct mbuf *)0;
2117 	if (m->m_type == MT_SONAME) {
2118 		nam = m;
2119 		m = m->m_next;
2120 		nam->m_next = NULL;
2121 	} else
2122 		nam = NULL;
2123 	MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2124 		M_NFSRVDESC, M_WAITOK);
2125 	nd->nd_md = nd->nd_mrep = m;
2126 	nd->nd_nam2 = nam;
2127 	nd->nd_dpos = mtod(m, caddr_t);
2128 	error = nfs_getreq(nd, nfsd, TRUE);
2129 	if (error) {
2130 		m_freem(nam);
2131 		free((caddr_t)nd, M_NFSRVDESC);
2132 		return (error);
2133 	}
2134 	*ndp = nd;
2135 	nfsd->nfsd_nd = nd;
2136 	return (0);
2137 }
2138 
2139 
2140 /*
2141  * Search for a sleeping nfsd and wake it up.
2142  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2143  * running nfsds will go look for the work in the nfssvc_sock list.
2144  */
2145 void
2146 nfsrv_wakenfsd(slp)
2147 	struct nfssvc_sock *slp;
2148 {
2149 	register struct nfsd *nd;
2150 
2151 	if ((slp->ns_flag & SLP_VALID) == 0)
2152 		return;
2153 	for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) {
2154 		if (nd->nfsd_flag & NFSD_WAITING) {
2155 			nd->nfsd_flag &= ~NFSD_WAITING;
2156 			if (nd->nfsd_slp)
2157 				panic("nfsd wakeup");
2158 			slp->ns_sref++;
2159 			nd->nfsd_slp = slp;
2160 			wakeup((caddr_t)nd);
2161 			return;
2162 		}
2163 	}
2164 	slp->ns_flag |= SLP_DOREC;
2165 	nfsd_head_flag |= NFSD_CHECKSLP;
2166 }
2167 #endif /* NFSSERVER */
2168