xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 4472dbe5e3bd91ef2540bada7a7ca7384627ff9b)
1 /*	$NetBSD: nfs_socket.c,v 1.56 2000/05/27 04:52:41 thorpej 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/callout.h>
51 #include <sys/proc.h>
52 #include <sys/mount.h>
53 #include <sys/kernel.h>
54 #include <sys/mbuf.h>
55 #include <sys/vnode.h>
56 #include <sys/domain.h>
57 #include <sys/protosw.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/syslog.h>
61 #include <sys/tprintf.h>
62 #include <sys/namei.h>
63 #include <sys/signal.h>
64 #include <sys/signalvar.h>
65 
66 #include <netinet/in.h>
67 #include <netinet/tcp.h>
68 
69 #include <nfs/rpcv2.h>
70 #include <nfs/nfsproto.h>
71 #include <nfs/nfs.h>
72 #include <nfs/xdr_subs.h>
73 #include <nfs/nfsm_subs.h>
74 #include <nfs/nfsmount.h>
75 #include <nfs/nfsnode.h>
76 #include <nfs/nfsrtt.h>
77 #include <nfs/nqnfs.h>
78 #include <nfs/nfs_var.h>
79 
80 #define	TRUE	1
81 #define	FALSE	0
82 
83 /*
84  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
85  * Use the mean and mean deviation of rtt for the appropriate type of rpc
86  * for the frequent rpcs and a default for the others.
87  * The justification for doing "other" this way is that these rpcs
88  * happen so infrequently that timer est. would probably be stale.
89  * Also, since many of these rpcs are
90  * non-idempotent, a conservative timeout is desired.
91  * getattr, lookup - A+2D
92  * read, write     - A+4D
93  * other           - nm_timeo
94  */
95 #define	NFS_RTO(n, t) \
96 	((t) == 0 ? (n)->nm_timeo : \
97 	 ((t) < 3 ? \
98 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
99 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
100 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
101 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
102 /*
103  * External data, mostly RPC constants in XDR form
104  */
105 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
106 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
107 	rpc_auth_kerb;
108 extern u_int32_t nfs_prog, nqnfs_prog;
109 extern time_t nqnfsstarttime;
110 extern struct nfsstats nfsstats;
111 extern int nfsv3_procid[NFS_NPROCS];
112 extern int nfs_ticks;
113 
114 /*
115  * Defines which timer to use for the procnum.
116  * 0 - default
117  * 1 - getattr
118  * 2 - lookup
119  * 3 - read
120  * 4 - write
121  */
122 static int proct[NFS_NPROCS] = {
123 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
124 	0, 0, 0,
125 };
126 
127 /*
128  * There is a congestion window for outstanding rpcs maintained per mount
129  * point. The cwnd size is adjusted in roughly the way that:
130  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
131  * SIGCOMM '88". ACM, August 1988.
132  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
133  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
134  * of rpcs is in progress.
135  * (The sent count and cwnd are scaled for integer arith.)
136  * Variants of "slow start" were tried and were found to be too much of a
137  * performance hit (ave. rtt 3 times larger),
138  * I suspect due to the large rtt that nfs rpcs have.
139  */
140 #define	NFS_CWNDSCALE	256
141 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
142 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
143 int nfsrtton = 0;
144 struct nfsrtt nfsrtt;
145 
146 struct callout nfs_timer_ch = CALLOUT_INITIALIZER;
147 
148 /*
149  * Initialize sockets and congestion for a new NFS connection.
150  * We do not free the sockaddr if error.
151  */
152 int
153 nfs_connect(nmp, rep)
154 	struct nfsmount *nmp;
155 	struct nfsreq *rep;
156 {
157 	struct socket *so;
158 	int s, error, rcvreserve, sndreserve;
159 	struct sockaddr *saddr;
160 	struct sockaddr_in *sin;
161 	struct mbuf *m;
162 	u_int16_t tport;
163 
164 	nmp->nm_so = (struct socket *)0;
165 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
166 	error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
167 		nmp->nm_soproto);
168 	if (error)
169 		goto bad;
170 	so = nmp->nm_so;
171 	nmp->nm_soflags = so->so_proto->pr_flags;
172 
173 	/*
174 	 * Some servers require that the client port be a reserved port number.
175 	 */
176 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
177 		MGET(m, M_WAIT, MT_SONAME);
178 		sin = mtod(m, struct sockaddr_in *);
179 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
180 		sin->sin_family = AF_INET;
181 		sin->sin_addr.s_addr = INADDR_ANY;
182 		tport = IPPORT_RESERVED - 1;
183 		sin->sin_port = htons(tport);
184 		while ((error = sobind(so, m)) == EADDRINUSE &&
185 		       --tport > IPPORT_RESERVED / 2)
186 			sin->sin_port = htons(tport);
187 		m_freem(m);
188 		if (error)
189 			goto bad;
190 	}
191 
192 	/*
193 	 * Protocols that do not require connections may be optionally left
194 	 * unconnected for servers that reply from a port other than NFS_PORT.
195 	 */
196 	if (nmp->nm_flag & NFSMNT_NOCONN) {
197 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
198 			error = ENOTCONN;
199 			goto bad;
200 		}
201 	} else {
202 		error = soconnect(so, nmp->nm_nam);
203 		if (error)
204 			goto bad;
205 
206 		/*
207 		 * Wait for the connection to complete. Cribbed from the
208 		 * connect system call but with the wait timing out so
209 		 * that interruptible mounts don't hang here for a long time.
210 		 */
211 		s = splsoftnet();
212 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
213 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
214 				"nfscn1", 2 * hz);
215 			if ((so->so_state & SS_ISCONNECTING) &&
216 			    so->so_error == 0 && rep &&
217 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
218 				so->so_state &= ~SS_ISCONNECTING;
219 				splx(s);
220 				goto bad;
221 			}
222 		}
223 		if (so->so_error) {
224 			error = so->so_error;
225 			so->so_error = 0;
226 			splx(s);
227 			goto bad;
228 		}
229 		splx(s);
230 	}
231 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
232 		so->so_rcv.sb_timeo = (5 * hz);
233 		so->so_snd.sb_timeo = (5 * hz);
234 	} else {
235 		so->so_rcv.sb_timeo = 0;
236 		so->so_snd.sb_timeo = 0;
237 	}
238 	if (nmp->nm_sotype == SOCK_DGRAM) {
239 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
240 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
241 		    NFS_MAXPKTHDR) * 2;
242 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
243 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
244 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
245 		    NFS_MAXPKTHDR) * 2;
246 	} else {
247 		if (nmp->nm_sotype != SOCK_STREAM)
248 			panic("nfscon sotype");
249 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
250 			MGET(m, M_WAIT, MT_SOOPTS);
251 			*mtod(m, int32_t *) = 1;
252 			m->m_len = sizeof(int32_t);
253 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
254 		}
255 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
256 			MGET(m, M_WAIT, MT_SOOPTS);
257 			*mtod(m, int32_t *) = 1;
258 			m->m_len = sizeof(int32_t);
259 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
260 		}
261 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
262 		    sizeof (u_int32_t)) * 2;
263 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
264 		    sizeof (u_int32_t)) * 2;
265 	}
266 	error = soreserve(so, sndreserve, rcvreserve);
267 	if (error)
268 		goto bad;
269 	so->so_rcv.sb_flags |= SB_NOINTR;
270 	so->so_snd.sb_flags |= SB_NOINTR;
271 
272 	/* Initialize other non-zero congestion variables */
273 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
274 		nmp->nm_srtt[4] = (NFS_TIMEO << 3);
275 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
276 		nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0;
277 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
278 	nmp->nm_sent = 0;
279 	nmp->nm_timeouts = 0;
280 	return (0);
281 
282 bad:
283 	nfs_disconnect(nmp);
284 	return (error);
285 }
286 
287 /*
288  * Reconnect routine:
289  * Called when a connection is broken on a reliable protocol.
290  * - clean up the old socket
291  * - nfs_connect() again
292  * - set R_MUSTRESEND for all outstanding requests on mount point
293  * If this fails the mount point is DEAD!
294  * nb: Must be called with the nfs_sndlock() set on the mount point.
295  */
296 int
297 nfs_reconnect(rep)
298 	struct nfsreq *rep;
299 {
300 	struct nfsreq *rp;
301 	struct nfsmount *nmp = rep->r_nmp;
302 	int error;
303 
304 	nfs_disconnect(nmp);
305 	while ((error = nfs_connect(nmp, rep)) != 0) {
306 		if (error == EINTR || error == ERESTART)
307 			return (EINTR);
308 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0);
309 	}
310 
311 	/*
312 	 * Loop through outstanding request list and fix up all requests
313 	 * on old socket.
314 	 */
315 	for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) {
316 		if (rp->r_nmp == nmp)
317 			rp->r_flags |= R_MUSTRESEND;
318 	}
319 	return (0);
320 }
321 
322 /*
323  * NFS disconnect. Clean up and unlink.
324  */
325 void
326 nfs_disconnect(nmp)
327 	struct nfsmount *nmp;
328 {
329 	struct socket *so;
330 	int drain = 0;
331 
332 	if (nmp->nm_so) {
333 		so = nmp->nm_so;
334 		nmp->nm_so = (struct socket *)0;
335 		soshutdown(so, 2);
336 		drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
337 		if (drain) {
338 			/*
339 			 * soshutdown() above should wake up the current
340 			 * listener.
341 			 * Now wake up those waiting for the recive lock, and
342 			 * wait for them to go away unhappy, to prevent *nmp
343 			 * from evaporating while they're sleeping.
344 			 */
345 			while (nmp->nm_waiters > 0) {
346 				wakeup (&nmp->nm_iflag);
347 				(void) tsleep(&nmp->nm_waiters, PVFS,
348 				    "nfsdis", 0);
349 			}
350 		}
351 		soclose(so);
352 	}
353 #ifdef DIAGNOSTIC
354 	if (drain && (nmp->nm_waiters > 0))
355 		panic("nfs_disconnect: waiters left after drain?\n");
356 #endif
357 }
358 
359 void
360 nfs_safedisconnect(nmp)
361 	struct nfsmount *nmp;
362 {
363 	struct nfsreq dummyreq;
364 
365 	memset(&dummyreq, 0, sizeof(dummyreq));
366 	dummyreq.r_nmp = nmp;
367 	nfs_rcvlock(&dummyreq); /* XXX ignored error return */
368 	nfs_disconnect(nmp);
369 	nfs_rcvunlock(&nmp->nm_iflag);
370 }
371 
372 /*
373  * This is the nfs send routine. For connection based socket types, it
374  * must be called with an nfs_sndlock() on the socket.
375  * "rep == NULL" indicates that it has been called from a server.
376  * For the client side:
377  * - return EINTR if the RPC is terminated, 0 otherwise
378  * - set R_MUSTRESEND if the send fails for any reason
379  * - do any cleanup required by recoverable socket errors (???)
380  * For the server side:
381  * - return EINTR or ERESTART if interrupted by a signal
382  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
383  * - do any cleanup required by recoverable socket errors (???)
384  */
385 int
386 nfs_send(so, nam, top, rep)
387 	struct socket *so;
388 	struct mbuf *nam;
389 	struct mbuf *top;
390 	struct nfsreq *rep;
391 {
392 	struct mbuf *sendnam;
393 	int error, soflags, flags;
394 
395 	if (rep) {
396 		if (rep->r_flags & R_SOFTTERM) {
397 			m_freem(top);
398 			return (EINTR);
399 		}
400 		if ((so = rep->r_nmp->nm_so) == NULL) {
401 			rep->r_flags |= R_MUSTRESEND;
402 			m_freem(top);
403 			return (0);
404 		}
405 		rep->r_flags &= ~R_MUSTRESEND;
406 		soflags = rep->r_nmp->nm_soflags;
407 	} else
408 		soflags = so->so_proto->pr_flags;
409 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
410 		sendnam = (struct mbuf *)0;
411 	else
412 		sendnam = nam;
413 	if (so->so_type == SOCK_SEQPACKET)
414 		flags = MSG_EOR;
415 	else
416 		flags = 0;
417 
418 	error = (*so->so_send)(so, sendnam, (struct uio *)0, top,
419 		(struct mbuf *)0, flags);
420 	if (error) {
421 		if (rep) {
422 			log(LOG_INFO, "nfs send error %d for server %s\n",error,
423 			    rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
424 			/*
425 			 * Deal with errors for the client side.
426 			 */
427 			if (rep->r_flags & R_SOFTTERM)
428 				error = EINTR;
429 			else
430 				rep->r_flags |= R_MUSTRESEND;
431 		} else
432 			log(LOG_INFO, "nfsd send error %d\n", error);
433 
434 		/*
435 		 * Handle any recoverable (soft) socket errors here. (???)
436 		 */
437 		if (error != EINTR && error != ERESTART &&
438 			error != EWOULDBLOCK && error != EPIPE)
439 			error = 0;
440 	}
441 	return (error);
442 }
443 
444 #ifdef NFS
445 /*
446  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
447  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
448  * Mark and consolidate the data into a new mbuf list.
449  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
450  *     small mbufs.
451  * For SOCK_STREAM we must be very careful to read an entire record once
452  * we have read any of it, even if the system call has been interrupted.
453  */
454 int
455 nfs_receive(rep, aname, mp)
456 	struct nfsreq *rep;
457 	struct mbuf **aname;
458 	struct mbuf **mp;
459 {
460 	struct socket *so;
461 	struct uio auio;
462 	struct iovec aio;
463 	struct mbuf *m;
464 	struct mbuf *control;
465 	u_int32_t len;
466 	struct mbuf **getnam;
467 	int error, sotype, rcvflg;
468 	struct proc *p = curproc;	/* XXX */
469 
470 	/*
471 	 * Set up arguments for soreceive()
472 	 */
473 	*mp = (struct mbuf *)0;
474 	*aname = (struct mbuf *)0;
475 	sotype = rep->r_nmp->nm_sotype;
476 
477 	/*
478 	 * For reliable protocols, lock against other senders/receivers
479 	 * in case a reconnect is necessary.
480 	 * For SOCK_STREAM, first get the Record Mark to find out how much
481 	 * more there is to get.
482 	 * We must lock the socket against other receivers
483 	 * until we have an entire rpc request/reply.
484 	 */
485 	if (sotype != SOCK_DGRAM) {
486 		error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
487 		if (error)
488 			return (error);
489 tryagain:
490 		/*
491 		 * Check for fatal errors and resending request.
492 		 */
493 		/*
494 		 * Ugh: If a reconnect attempt just happened, nm_so
495 		 * would have changed. NULL indicates a failed
496 		 * attempt that has essentially shut down this
497 		 * mount point.
498 		 */
499 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
500 			nfs_sndunlock(&rep->r_nmp->nm_iflag);
501 			return (EINTR);
502 		}
503 		so = rep->r_nmp->nm_so;
504 		if (!so) {
505 			error = nfs_reconnect(rep);
506 			if (error) {
507 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
508 				return (error);
509 			}
510 			goto tryagain;
511 		}
512 		while (rep->r_flags & R_MUSTRESEND) {
513 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
514 			nfsstats.rpcretries++;
515 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
516 			if (error) {
517 				if (error == EINTR || error == ERESTART ||
518 				    (error = nfs_reconnect(rep)) != 0) {
519 					nfs_sndunlock(&rep->r_nmp->nm_iflag);
520 					return (error);
521 				}
522 				goto tryagain;
523 			}
524 		}
525 		nfs_sndunlock(&rep->r_nmp->nm_iflag);
526 		if (sotype == SOCK_STREAM) {
527 			aio.iov_base = (caddr_t) &len;
528 			aio.iov_len = sizeof(u_int32_t);
529 			auio.uio_iov = &aio;
530 			auio.uio_iovcnt = 1;
531 			auio.uio_segflg = UIO_SYSSPACE;
532 			auio.uio_rw = UIO_READ;
533 			auio.uio_offset = 0;
534 			auio.uio_resid = sizeof(u_int32_t);
535 			auio.uio_procp = p;
536 			do {
537 			   rcvflg = MSG_WAITALL;
538 			   error = (*so->so_receive)(so, (struct mbuf **)0, &auio,
539 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
540 			   if (error == EWOULDBLOCK && rep) {
541 				if (rep->r_flags & R_SOFTTERM)
542 					return (EINTR);
543 			   }
544 			} while (error == EWOULDBLOCK);
545 			if (!error && auio.uio_resid > 0) {
546 			    /*
547 			     * Don't log a 0 byte receive; it means
548 			     * that the socket has been closed, and
549 			     * can happen during normal operation
550 			     * (forcible unmount or Solaris server).
551 			     */
552 			    if (auio.uio_resid != sizeof (u_int32_t))
553 			      log(LOG_INFO,
554 				 "short receive (%lu/%lu) from nfs server %s\n",
555 				 (u_long)sizeof(u_int32_t) - auio.uio_resid,
556 				 (u_long)sizeof(u_int32_t),
557 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
558 			    error = EPIPE;
559 			}
560 			if (error)
561 				goto errout;
562 			len = ntohl(len) & ~0x80000000;
563 			/*
564 			 * This is SERIOUS! We are out of sync with the sender
565 			 * and forcing a disconnect/reconnect is all I can do.
566 			 */
567 			if (len > NFS_MAXPACKET) {
568 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
569 				"impossible packet length",
570 				len,
571 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
572 			    error = EFBIG;
573 			    goto errout;
574 			}
575 			auio.uio_resid = len;
576 			do {
577 			    rcvflg = MSG_WAITALL;
578 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
579 				&auio, mp, (struct mbuf **)0, &rcvflg);
580 			} while (error == EWOULDBLOCK || error == EINTR ||
581 				 error == ERESTART);
582 			if (!error && auio.uio_resid > 0) {
583 			    if (len != auio.uio_resid)
584 			      log(LOG_INFO,
585 				"short receive (%lu/%d) from nfs server %s\n",
586 				(u_long)len - auio.uio_resid, len,
587 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
588 			    error = EPIPE;
589 			}
590 		} else {
591 			/*
592 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
593 			 * and soreceive() will return when it has either a
594 			 * control msg or a data msg.
595 			 * We have no use for control msg., but must grab them
596 			 * and then throw them away so we know what is going
597 			 * on.
598 			 */
599 			auio.uio_resid = len = 100000000; /* Anything Big */
600 			auio.uio_procp = p;
601 			do {
602 			    rcvflg = 0;
603 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
604 				&auio, mp, &control, &rcvflg);
605 			    if (control)
606 				m_freem(control);
607 			    if (error == EWOULDBLOCK && rep) {
608 				if (rep->r_flags & R_SOFTTERM)
609 					return (EINTR);
610 			    }
611 			} while (error == EWOULDBLOCK ||
612 				 (!error && *mp == NULL && control));
613 			if ((rcvflg & MSG_EOR) == 0)
614 				printf("Egad!!\n");
615 			if (!error && *mp == NULL)
616 				error = EPIPE;
617 			len -= auio.uio_resid;
618 		}
619 errout:
620 		if (error && error != EINTR && error != ERESTART) {
621 			m_freem(*mp);
622 			*mp = (struct mbuf *)0;
623 			if (error != EPIPE)
624 				log(LOG_INFO,
625 				    "receive error %d from nfs server %s\n",
626 				    error,
627 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
628 			error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
629 			if (!error)
630 				error = nfs_reconnect(rep);
631 			if (!error)
632 				goto tryagain;
633 			else
634 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
635 		}
636 	} else {
637 		if ((so = rep->r_nmp->nm_so) == NULL)
638 			return (EACCES);
639 		if (so->so_state & SS_ISCONNECTED)
640 			getnam = (struct mbuf **)0;
641 		else
642 			getnam = aname;
643 		auio.uio_resid = len = 1000000;
644 		auio.uio_procp = p;
645 		do {
646 			rcvflg = 0;
647 			error =  (*so->so_receive)(so, getnam, &auio, mp,
648 				(struct mbuf **)0, &rcvflg);
649 			if (error == EWOULDBLOCK &&
650 			    (rep->r_flags & R_SOFTTERM))
651 				return (EINTR);
652 		} while (error == EWOULDBLOCK);
653 		len -= auio.uio_resid;
654 		if (!error && *mp == NULL)
655 			error = EPIPE;
656 	}
657 	if (error) {
658 		m_freem(*mp);
659 		*mp = (struct mbuf *)0;
660 	}
661 	return (error);
662 }
663 
664 /*
665  * Implement receipt of reply on a socket.
666  * We must search through the list of received datagrams matching them
667  * with outstanding requests using the xid, until ours is found.
668  */
669 /* ARGSUSED */
670 int
671 nfs_reply(myrep)
672 	struct nfsreq *myrep;
673 {
674 	struct nfsreq *rep;
675 	struct nfsmount *nmp = myrep->r_nmp;
676 	int32_t t1;
677 	struct mbuf *mrep, *nam, *md;
678 	u_int32_t rxid, *tl;
679 	caddr_t dpos, cp2;
680 	int error;
681 
682 	/*
683 	 * Loop around until we get our own reply
684 	 */
685 	for (;;) {
686 		/*
687 		 * Lock against other receivers so that I don't get stuck in
688 		 * sbwait() after someone else has received my reply for me.
689 		 * Also necessary for connection based protocols to avoid
690 		 * race conditions during a reconnect.
691 		 */
692 		error = nfs_rcvlock(myrep);
693 		if (error == EALREADY)
694 			return (0);
695 		if (error)
696 			return (error);
697 		/*
698 		 * Get the next Rpc reply off the socket
699 		 */
700 		nmp->nm_waiters++;
701 		error = nfs_receive(myrep, &nam, &mrep);
702 		nfs_rcvunlock(&nmp->nm_iflag);
703 		if (error) {
704 
705 			if (nmp->nm_iflag & NFSMNT_DISMNT) {
706 				/*
707 				 * Oops, we're going away now..
708 				 */
709 				nmp->nm_waiters--;
710 				wakeup (&nmp->nm_waiters);
711 				return error;
712 			}
713 			nmp->nm_waiters--;
714 			/*
715 			 * Ignore routing errors on connectionless protocols??
716 			 */
717 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
718 				nmp->nm_so->so_error = 0;
719 #ifdef DEBUG
720 				printf("nfs_reply: ignoring error %d\n", error);
721 #endif
722 				if (myrep->r_flags & R_GETONEREP)
723 					return (0);
724 				continue;
725 			}
726 			return (error);
727 		}
728 		nmp->nm_waiters--;
729 		if (nam)
730 			m_freem(nam);
731 
732 		/*
733 		 * Get the xid and check that it is an rpc reply
734 		 */
735 		md = mrep;
736 		dpos = mtod(md, caddr_t);
737 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
738 		rxid = *tl++;
739 		if (*tl != rpc_reply) {
740 			if (nmp->nm_flag & NFSMNT_NQNFS) {
741 				if (nqnfs_callback(nmp, mrep, md, dpos))
742 					nfsstats.rpcinvalid++;
743 			} else {
744 				nfsstats.rpcinvalid++;
745 				m_freem(mrep);
746 			}
747 nfsmout:
748 			if (myrep->r_flags & R_GETONEREP)
749 				return (0);
750 			continue;
751 		}
752 
753 		/*
754 		 * Loop through the request list to match up the reply
755 		 * Iff no match, just drop the datagram
756 		 */
757 		for (rep = nfs_reqq.tqh_first; rep != 0;
758 		    rep = rep->r_chain.tqe_next) {
759 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
760 				/* Found it.. */
761 				rep->r_mrep = mrep;
762 				rep->r_md = md;
763 				rep->r_dpos = dpos;
764 				if (nfsrtton) {
765 					struct rttl *rt;
766 
767 					rt = &nfsrtt.rttl[nfsrtt.pos];
768 					rt->proc = rep->r_procnum;
769 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
770 					rt->sent = nmp->nm_sent;
771 					rt->cwnd = nmp->nm_cwnd;
772 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
773 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
774 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
775 					rt->tstamp = time;
776 					if (rep->r_flags & R_TIMING)
777 						rt->rtt = rep->r_rtt;
778 					else
779 						rt->rtt = 1000000;
780 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
781 				}
782 				/*
783 				 * Update congestion window.
784 				 * Do the additive increase of
785 				 * one rpc/rtt.
786 				 */
787 				if (nmp->nm_cwnd <= nmp->nm_sent) {
788 					nmp->nm_cwnd +=
789 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
790 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
791 					if (nmp->nm_cwnd > NFS_MAXCWND)
792 						nmp->nm_cwnd = NFS_MAXCWND;
793 				}
794 				rep->r_flags &= ~R_SENT;
795 				nmp->nm_sent -= NFS_CWNDSCALE;
796 				/*
797 				 * Update rtt using a gain of 0.125 on the mean
798 				 * and a gain of 0.25 on the deviation.
799 				 */
800 				if (rep->r_flags & R_TIMING) {
801 					/*
802 					 * Since the timer resolution of
803 					 * NFS_HZ is so course, it can often
804 					 * result in r_rtt == 0. Since
805 					 * r_rtt == N means that the actual
806 					 * rtt is between N+dt and N+2-dt ticks,
807 					 * add 1.
808 					 */
809 					t1 = rep->r_rtt + 1;
810 					t1 -= (NFS_SRTT(rep) >> 3);
811 					NFS_SRTT(rep) += t1;
812 					if (t1 < 0)
813 						t1 = -t1;
814 					t1 -= (NFS_SDRTT(rep) >> 2);
815 					NFS_SDRTT(rep) += t1;
816 				}
817 				nmp->nm_timeouts = 0;
818 				break;
819 			}
820 		}
821 		/*
822 		 * If not matched to a request, drop it.
823 		 * If it's mine, get out.
824 		 */
825 		if (rep == 0) {
826 			nfsstats.rpcunexpected++;
827 			m_freem(mrep);
828 		} else if (rep == myrep) {
829 			if (rep->r_mrep == NULL)
830 				panic("nfsreply nil");
831 			return (0);
832 		}
833 		if (myrep->r_flags & R_GETONEREP)
834 			return (0);
835 	}
836 }
837 
838 /*
839  * nfs_request - goes something like this
840  *	- fill in request struct
841  *	- links it into list
842  *	- calls nfs_send() for first transmit
843  *	- calls nfs_receive() to get reply
844  *	- break down rpc header and return with nfs reply pointed to
845  *	  by mrep or error
846  * nb: always frees up mreq mbuf list
847  */
848 int
849 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
850 	struct vnode *vp;
851 	struct mbuf *mrest;
852 	int procnum;
853 	struct proc *procp;
854 	struct ucred *cred;
855 	struct mbuf **mrp;
856 	struct mbuf **mdp;
857 	caddr_t *dposp;
858 {
859 	struct mbuf *m, *mrep;
860 	struct nfsreq *rep;
861 	u_int32_t *tl;
862 	int i;
863 	struct nfsmount *nmp;
864 	struct mbuf *md, *mheadend;
865 	struct nfsnode *np;
866 	char nickv[RPCX_NICKVERF];
867 	time_t reqtime, waituntil;
868 	caddr_t dpos, cp2;
869 	int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type;
870 	int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0;
871 	int verf_len, verf_type;
872 	u_int32_t xid;
873 	u_quad_t frev;
874 	char *auth_str, *verf_str;
875 	NFSKERBKEY_T key;		/* save session key */
876 
877 	nmp = VFSTONFS(vp->v_mount);
878 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
879 	rep->r_nmp = nmp;
880 	rep->r_vp = vp;
881 	rep->r_procp = procp;
882 	rep->r_procnum = procnum;
883 	i = 0;
884 	m = mrest;
885 	while (m) {
886 		i += m->m_len;
887 		m = m->m_next;
888 	}
889 	mrest_len = i;
890 
891 	/*
892 	 * Get the RPC header with authorization.
893 	 */
894 kerbauth:
895 	verf_str = auth_str = (char *)0;
896 	if (nmp->nm_flag & NFSMNT_KERB) {
897 		verf_str = nickv;
898 		verf_len = sizeof (nickv);
899 		auth_type = RPCAUTH_KERB4;
900 		memset((caddr_t)key, 0, sizeof (key));
901 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
902 			&auth_len, verf_str, verf_len)) {
903 			error = nfs_getauth(nmp, rep, cred, &auth_str,
904 				&auth_len, verf_str, &verf_len, key);
905 			if (error) {
906 				free((caddr_t)rep, M_NFSREQ);
907 				m_freem(mrest);
908 				return (error);
909 			}
910 		}
911 	} else {
912 		auth_type = RPCAUTH_UNIX;
913 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
914 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
915 			5 * NFSX_UNSIGNED;
916 	}
917 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
918 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
919 	if (auth_str)
920 		free(auth_str, M_TEMP);
921 
922 	/*
923 	 * For stream protocols, insert a Sun RPC Record Mark.
924 	 */
925 	if (nmp->nm_sotype == SOCK_STREAM) {
926 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
927 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
928 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
929 	}
930 	rep->r_mreq = m;
931 	rep->r_xid = xid;
932 tryagain:
933 	if (nmp->nm_flag & NFSMNT_SOFT)
934 		rep->r_retry = nmp->nm_retry;
935 	else
936 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
937 	rep->r_rtt = rep->r_rexmit = 0;
938 	if (proct[procnum] > 0)
939 		rep->r_flags = R_TIMING;
940 	else
941 		rep->r_flags = 0;
942 	rep->r_mrep = NULL;
943 
944 	/*
945 	 * Do the client side RPC.
946 	 */
947 	nfsstats.rpcrequests++;
948 	/*
949 	 * Chain request into list of outstanding requests. Be sure
950 	 * to put it LAST so timer finds oldest requests first.
951 	 */
952 	s = splsoftnet();
953 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
954 
955 	/* Get send time for nqnfs */
956 	reqtime = time.tv_sec;
957 
958 	/*
959 	 * If backing off another request or avoiding congestion, don't
960 	 * send this one now but let timer do it. If not timing a request,
961 	 * do it now.
962 	 */
963 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
964 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
965 		nmp->nm_sent < nmp->nm_cwnd)) {
966 		splx(s);
967 		if (nmp->nm_soflags & PR_CONNREQUIRED)
968 			error = nfs_sndlock(&nmp->nm_iflag, rep);
969 		if (!error) {
970 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
971 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep);
972 			if (nmp->nm_soflags & PR_CONNREQUIRED)
973 				nfs_sndunlock(&nmp->nm_iflag);
974 		}
975 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
976 			nmp->nm_sent += NFS_CWNDSCALE;
977 			rep->r_flags |= R_SENT;
978 		}
979 	} else {
980 		splx(s);
981 		rep->r_rtt = -1;
982 	}
983 
984 	/*
985 	 * Wait for the reply from our send or the timer's.
986 	 */
987 	if (!error || error == EPIPE)
988 		error = nfs_reply(rep);
989 
990 	/*
991 	 * RPC done, unlink the request.
992 	 */
993 	s = splsoftnet();
994 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
995 	splx(s);
996 
997 	/*
998 	 * Decrement the outstanding request count.
999 	 */
1000 	if (rep->r_flags & R_SENT) {
1001 		rep->r_flags &= ~R_SENT;	/* paranoia */
1002 		nmp->nm_sent -= NFS_CWNDSCALE;
1003 	}
1004 
1005 	/*
1006 	 * If there was a successful reply and a tprintf msg.
1007 	 * tprintf a response.
1008 	 */
1009 	if (!error && (rep->r_flags & R_TPRINTFMSG))
1010 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1011 		    "is alive again");
1012 	mrep = rep->r_mrep;
1013 	md = rep->r_md;
1014 	dpos = rep->r_dpos;
1015 	if (error) {
1016 		m_freem(rep->r_mreq);
1017 		free((caddr_t)rep, M_NFSREQ);
1018 		return (error);
1019 	}
1020 
1021 	/*
1022 	 * break down the rpc header and check if ok
1023 	 */
1024 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1025 	if (*tl++ == rpc_msgdenied) {
1026 		if (*tl == rpc_mismatch)
1027 			error = EOPNOTSUPP;
1028 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1029 			if (!failed_auth) {
1030 				failed_auth++;
1031 				mheadend->m_next = (struct mbuf *)0;
1032 				m_freem(mrep);
1033 				m_freem(rep->r_mreq);
1034 				goto kerbauth;
1035 			} else
1036 				error = EAUTH;
1037 		} else
1038 			error = EACCES;
1039 		m_freem(mrep);
1040 		m_freem(rep->r_mreq);
1041 		free((caddr_t)rep, M_NFSREQ);
1042 		return (error);
1043 	}
1044 
1045 	/*
1046 	 * Grab any Kerberos verifier, otherwise just throw it away.
1047 	 */
1048 	verf_type = fxdr_unsigned(int, *tl++);
1049 	i = fxdr_unsigned(int32_t, *tl);
1050 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1051 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1052 		if (error)
1053 			goto nfsmout;
1054 	} else if (i > 0)
1055 		nfsm_adv(nfsm_rndup(i));
1056 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1057 	/* 0 == ok */
1058 	if (*tl == 0) {
1059 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1060 		if (*tl != 0) {
1061 			error = fxdr_unsigned(int, *tl);
1062 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1063 				error == NFSERR_TRYLATER) {
1064 				m_freem(mrep);
1065 				error = 0;
1066 				waituntil = time.tv_sec + trylater_delay;
1067 				while (time.tv_sec < waituntil)
1068 					(void) tsleep((caddr_t)&lbolt,
1069 						PSOCK, "nqnfstry", 0);
1070 				trylater_delay *= nfs_backoff[trylater_cnt];
1071 				if (trylater_cnt < 7)
1072 					trylater_cnt++;
1073 				goto tryagain;
1074 			}
1075 
1076 			/*
1077 			 * If the File Handle was stale, invalidate the
1078 			 * lookup cache, just in case.
1079 			 */
1080 			if (error == ESTALE)
1081 				cache_purge(vp);
1082 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1083 				*mrp = mrep;
1084 				*mdp = md;
1085 				*dposp = dpos;
1086 				error |= NFSERR_RETERR;
1087 			} else
1088 				m_freem(mrep);
1089 			m_freem(rep->r_mreq);
1090 			free((caddr_t)rep, M_NFSREQ);
1091 			return (error);
1092 		}
1093 
1094 		/*
1095 		 * For nqnfs, get any lease in reply
1096 		 */
1097 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1098 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1099 			if (*tl) {
1100 				np = VTONFS(vp);
1101 				nqlflag = fxdr_unsigned(int, *tl);
1102 				nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1103 				cachable = fxdr_unsigned(int, *tl++);
1104 				reqtime += fxdr_unsigned(int, *tl++);
1105 				if (reqtime > time.tv_sec) {
1106 				    frev = fxdr_hyper(tl);
1107 				    nqnfs_clientlease(nmp, np, nqlflag,
1108 					cachable, reqtime, frev);
1109 				}
1110 			}
1111 		}
1112 		*mrp = mrep;
1113 		*mdp = md;
1114 		*dposp = dpos;
1115 		m_freem(rep->r_mreq);
1116 		FREE((caddr_t)rep, M_NFSREQ);
1117 		return (0);
1118 	}
1119 	m_freem(mrep);
1120 	error = EPROTONOSUPPORT;
1121 nfsmout:
1122 	m_freem(rep->r_mreq);
1123 	free((caddr_t)rep, M_NFSREQ);
1124 	return (error);
1125 }
1126 #endif /* NFS */
1127 
1128 /*
1129  * Generate the rpc reply header
1130  * siz arg. is used to decide if adding a cluster is worthwhile
1131  */
1132 int
1133 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1134 	int siz;
1135 	struct nfsrv_descript *nd;
1136 	struct nfssvc_sock *slp;
1137 	int err;
1138 	int cache;
1139 	u_quad_t *frev;
1140 	struct mbuf **mrq;
1141 	struct mbuf **mbp;
1142 	caddr_t *bposp;
1143 {
1144 	u_int32_t *tl;
1145 	struct mbuf *mreq;
1146 	caddr_t bpos;
1147 	struct mbuf *mb, *mb2;
1148 
1149 	MGETHDR(mreq, M_WAIT, MT_DATA);
1150 	mb = mreq;
1151 	/*
1152 	 * If this is a big reply, use a cluster else
1153 	 * try and leave leading space for the lower level headers.
1154 	 */
1155 	siz += RPC_REPLYSIZ;
1156 	if (siz >= max_datalen) {
1157 		MCLGET(mreq, M_WAIT);
1158 	} else
1159 		mreq->m_data += max_hdr;
1160 	tl = mtod(mreq, u_int32_t *);
1161 	mreq->m_len = 6 * NFSX_UNSIGNED;
1162 	bpos = ((caddr_t)tl) + mreq->m_len;
1163 	*tl++ = txdr_unsigned(nd->nd_retxid);
1164 	*tl++ = rpc_reply;
1165 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1166 		*tl++ = rpc_msgdenied;
1167 		if (err & NFSERR_AUTHERR) {
1168 			*tl++ = rpc_autherr;
1169 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1170 			mreq->m_len -= NFSX_UNSIGNED;
1171 			bpos -= NFSX_UNSIGNED;
1172 		} else {
1173 			*tl++ = rpc_mismatch;
1174 			*tl++ = txdr_unsigned(RPC_VER2);
1175 			*tl = txdr_unsigned(RPC_VER2);
1176 		}
1177 	} else {
1178 		*tl++ = rpc_msgaccepted;
1179 
1180 		/*
1181 		 * For Kerberos authentication, we must send the nickname
1182 		 * verifier back, otherwise just RPCAUTH_NULL.
1183 		 */
1184 		if (nd->nd_flag & ND_KERBFULL) {
1185 		    struct nfsuid *nuidp;
1186 		    struct timeval ktvin, ktvout;
1187 
1188 		    for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1189 			nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1190 			if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1191 			    (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1192 			     &nuidp->nu_haddr, nd->nd_nam2)))
1193 			    break;
1194 		    }
1195 		    if (nuidp) {
1196 			ktvin.tv_sec =
1197 			    txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1198 			ktvin.tv_usec =
1199 			    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1200 
1201 			/*
1202 			 * Encrypt the timestamp in ecb mode using the
1203 			 * session key.
1204 			 */
1205 #ifdef NFSKERB
1206 			XXX
1207 #endif
1208 
1209 			*tl++ = rpc_auth_kerb;
1210 			*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1211 			*tl = ktvout.tv_sec;
1212 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1213 			*tl++ = ktvout.tv_usec;
1214 			*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1215 		    } else {
1216 			*tl++ = 0;
1217 			*tl++ = 0;
1218 		    }
1219 		} else {
1220 			*tl++ = 0;
1221 			*tl++ = 0;
1222 		}
1223 		switch (err) {
1224 		case EPROGUNAVAIL:
1225 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1226 			break;
1227 		case EPROGMISMATCH:
1228 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1229 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1230 			if (nd->nd_flag & ND_NQNFS) {
1231 				*tl++ = txdr_unsigned(3);
1232 				*tl = txdr_unsigned(3);
1233 			} else {
1234 				*tl++ = txdr_unsigned(2);
1235 				*tl = txdr_unsigned(3);
1236 			}
1237 			break;
1238 		case EPROCUNAVAIL:
1239 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1240 			break;
1241 		case EBADRPC:
1242 			*tl = txdr_unsigned(RPC_GARBAGE);
1243 			break;
1244 		default:
1245 			*tl = 0;
1246 			if (err != NFSERR_RETVOID) {
1247 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1248 				if (err)
1249 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1250 				else
1251 				    *tl = 0;
1252 			}
1253 			break;
1254 		};
1255 	}
1256 
1257 	/*
1258 	 * For nqnfs, piggyback lease as requested.
1259 	 */
1260 	if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1261 		if (nd->nd_flag & ND_LEASE) {
1262 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1263 			*tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1264 			*tl++ = txdr_unsigned(cache);
1265 			*tl++ = txdr_unsigned(nd->nd_duration);
1266 			txdr_hyper(*frev, tl);
1267 		} else {
1268 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1269 			*tl = 0;
1270 		}
1271 	}
1272 	if (mrq != NULL)
1273 		*mrq = mreq;
1274 	*mbp = mb;
1275 	*bposp = bpos;
1276 	if (err != 0 && err != NFSERR_RETVOID)
1277 		nfsstats.srvrpc_errs++;
1278 	return (0);
1279 }
1280 
1281 /*
1282  * Nfs timer routine
1283  * Scan the nfsreq list and retranmit any requests that have timed out
1284  * To avoid retransmission attempts on STREAM sockets (in the future) make
1285  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1286  */
1287 void
1288 nfs_timer(arg)
1289 	void *arg;	/* never used */
1290 {
1291 	struct nfsreq *rep;
1292 	struct mbuf *m;
1293 	struct socket *so;
1294 	struct nfsmount *nmp;
1295 	int timeo;
1296 	int s, error;
1297 #ifdef NFSSERVER
1298 	struct nfssvc_sock *slp;
1299 	static long lasttime = 0;
1300 	u_quad_t cur_usec;
1301 #endif
1302 
1303 	s = splsoftnet();
1304 	for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) {
1305 		nmp = rep->r_nmp;
1306 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1307 			continue;
1308 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1309 			rep->r_flags |= R_SOFTTERM;
1310 			continue;
1311 		}
1312 		if (rep->r_rtt >= 0) {
1313 			rep->r_rtt++;
1314 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1315 				timeo = nmp->nm_timeo;
1316 			else
1317 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1318 			if (nmp->nm_timeouts > 0)
1319 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1320 			if (rep->r_rtt <= timeo)
1321 				continue;
1322 			if (nmp->nm_timeouts < 8)
1323 				nmp->nm_timeouts++;
1324 		}
1325 		/*
1326 		 * Check for server not responding
1327 		 */
1328 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1329 		     rep->r_rexmit > nmp->nm_deadthresh) {
1330 			nfs_msg(rep->r_procp,
1331 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1332 			    "not responding");
1333 			rep->r_flags |= R_TPRINTFMSG;
1334 		}
1335 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1336 			nfsstats.rpctimeouts++;
1337 			rep->r_flags |= R_SOFTTERM;
1338 			continue;
1339 		}
1340 		if (nmp->nm_sotype != SOCK_DGRAM) {
1341 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1342 				rep->r_rexmit = NFS_MAXREXMIT;
1343 			continue;
1344 		}
1345 		if ((so = nmp->nm_so) == NULL)
1346 			continue;
1347 
1348 		/*
1349 		 * If there is enough space and the window allows..
1350 		 *	Resend it
1351 		 * Set r_rtt to -1 in case we fail to send it now.
1352 		 */
1353 		rep->r_rtt = -1;
1354 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1355 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1356 		    (rep->r_flags & R_SENT) ||
1357 		    nmp->nm_sent < nmp->nm_cwnd) &&
1358 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1359 		        if (so->so_state & SS_ISCONNECTED)
1360 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1361 			    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1362 			else
1363 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1364 			    nmp->nm_nam, (struct mbuf *)0, (struct proc *)0);
1365 			if (error) {
1366 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1367 #ifdef DEBUG
1368 					printf("nfs_timer: ignoring error %d\n",
1369 						error);
1370 #endif
1371 					so->so_error = 0;
1372 				}
1373 			} else {
1374 				/*
1375 				 * Iff first send, start timing
1376 				 * else turn timing off, backoff timer
1377 				 * and divide congestion window by 2.
1378 				 */
1379 				if (rep->r_flags & R_SENT) {
1380 					rep->r_flags &= ~R_TIMING;
1381 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1382 						rep->r_rexmit = NFS_MAXREXMIT;
1383 					nmp->nm_cwnd >>= 1;
1384 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1385 						nmp->nm_cwnd = NFS_CWNDSCALE;
1386 					nfsstats.rpcretries++;
1387 				} else {
1388 					rep->r_flags |= R_SENT;
1389 					nmp->nm_sent += NFS_CWNDSCALE;
1390 				}
1391 				rep->r_rtt = 0;
1392 			}
1393 		}
1394 	}
1395 
1396 #ifdef NFSSERVER
1397 	/*
1398 	 * Call the nqnfs server timer once a second to handle leases.
1399 	 */
1400 	if (lasttime != time.tv_sec) {
1401 		lasttime = time.tv_sec;
1402 		nqnfs_serverd();
1403 	}
1404 
1405 	/*
1406 	 * Scan the write gathering queues for writes that need to be
1407 	 * completed now.
1408 	 */
1409 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1410 	for (slp = nfssvc_sockhead.tqh_first; slp != 0;
1411 	    slp = slp->ns_chain.tqe_next) {
1412 	    if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1413 		nfsrv_wakenfsd(slp);
1414 	}
1415 #endif /* NFSSERVER */
1416 	splx(s);
1417 	callout_reset(&nfs_timer_ch, nfs_ticks, nfs_timer, NULL);
1418 }
1419 
1420 /*
1421  * Test for a termination condition pending on the process.
1422  * This is used for NFSMNT_INT mounts.
1423  */
1424 int
1425 nfs_sigintr(nmp, rep, p)
1426 	struct nfsmount *nmp;
1427 	struct nfsreq *rep;
1428 	struct proc *p;
1429 {
1430 	sigset_t ss;
1431 
1432 	if (rep && (rep->r_flags & R_SOFTTERM))
1433 		return (EINTR);
1434 	if (!(nmp->nm_flag & NFSMNT_INT))
1435 		return (0);
1436 	if (p) {
1437 		sigpending1(p, &ss);
1438 #if 0
1439 		sigminusset(&p->p_sigignore, &ss);
1440 #endif
1441 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1442 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1443 		    sigismember(&ss, SIGQUIT))
1444 			return (EINTR);
1445 	}
1446 	return (0);
1447 }
1448 
1449 /*
1450  * Lock a socket against others.
1451  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1452  * and also to avoid race conditions between the processes with nfs requests
1453  * in progress when a reconnect is necessary.
1454  */
1455 int
1456 nfs_sndlock(flagp, rep)
1457 	int *flagp;
1458 	struct nfsreq *rep;
1459 {
1460 	struct proc *p;
1461 	int slpflag = 0, slptimeo = 0;
1462 
1463 	if (rep) {
1464 		p = rep->r_procp;
1465 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1466 			slpflag = PCATCH;
1467 	} else
1468 		p = (struct proc *)0;
1469 	while (*flagp & NFSMNT_SNDLOCK) {
1470 		if (nfs_sigintr(rep->r_nmp, rep, p))
1471 			return (EINTR);
1472 		*flagp |= NFSMNT_WANTSND;
1473 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1474 			slptimeo);
1475 		if (slpflag == PCATCH) {
1476 			slpflag = 0;
1477 			slptimeo = 2 * hz;
1478 		}
1479 	}
1480 	*flagp |= NFSMNT_SNDLOCK;
1481 	return (0);
1482 }
1483 
1484 /*
1485  * Unlock the stream socket for others.
1486  */
1487 void
1488 nfs_sndunlock(flagp)
1489 	int *flagp;
1490 {
1491 
1492 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1493 		panic("nfs sndunlock");
1494 	*flagp &= ~NFSMNT_SNDLOCK;
1495 	if (*flagp & NFSMNT_WANTSND) {
1496 		*flagp &= ~NFSMNT_WANTSND;
1497 		wakeup((caddr_t)flagp);
1498 	}
1499 }
1500 
1501 int
1502 nfs_rcvlock(rep)
1503 	struct nfsreq *rep;
1504 {
1505 	struct nfsmount *nmp = rep->r_nmp;
1506 	int *flagp = &nmp->nm_iflag;
1507 	int slpflag, slptimeo = 0;
1508 
1509 	if (*flagp & NFSMNT_DISMNT)
1510 		return EIO;
1511 
1512 	if (*flagp & NFSMNT_INT)
1513 		slpflag = PCATCH;
1514 	else
1515 		slpflag = 0;
1516 	while (*flagp & NFSMNT_RCVLOCK) {
1517 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1518 			return (EINTR);
1519 		*flagp |= NFSMNT_WANTRCV;
1520 		nmp->nm_waiters++;
1521 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1522 			slptimeo);
1523 		nmp->nm_waiters--;
1524 		if (*flagp & NFSMNT_DISMNT) {
1525 			wakeup(&nmp->nm_waiters);
1526 			return EIO;
1527 		}
1528 		/* If our reply was received while we were sleeping,
1529 		 * then just return without taking the lock to avoid a
1530 		 * situation where a single iod could 'capture' the
1531 		 * receive lock.
1532 		 */
1533 		if (rep->r_mrep != NULL)
1534 			return (EALREADY);
1535 		if (slpflag == PCATCH) {
1536 			slpflag = 0;
1537 			slptimeo = 2 * hz;
1538 		}
1539 	}
1540 	*flagp |= NFSMNT_RCVLOCK;
1541 	return (0);
1542 }
1543 
1544 /*
1545  * Unlock the stream socket for others.
1546  */
1547 void
1548 nfs_rcvunlock(flagp)
1549 	int *flagp;
1550 {
1551 
1552 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1553 		panic("nfs rcvunlock");
1554 	*flagp &= ~NFSMNT_RCVLOCK;
1555 	if (*flagp & NFSMNT_WANTRCV) {
1556 		*flagp &= ~NFSMNT_WANTRCV;
1557 		wakeup((caddr_t)flagp);
1558 	}
1559 }
1560 
1561 /*
1562  * Parse an RPC request
1563  * - verify it
1564  * - fill in the cred struct.
1565  */
1566 int
1567 nfs_getreq(nd, nfsd, has_header)
1568 	struct nfsrv_descript *nd;
1569 	struct nfsd *nfsd;
1570 	int has_header;
1571 {
1572 	int len, i;
1573 	u_int32_t *tl;
1574 	int32_t t1;
1575 	struct uio uio;
1576 	struct iovec iov;
1577 	caddr_t dpos, cp2, cp;
1578 	u_int32_t nfsvers, auth_type;
1579 	uid_t nickuid;
1580 	int error = 0, nqnfs = 0, ticklen;
1581 	struct mbuf *mrep, *md;
1582 	struct nfsuid *nuidp;
1583 	struct timeval tvin, tvout;
1584 
1585 	mrep = nd->nd_mrep;
1586 	md = nd->nd_md;
1587 	dpos = nd->nd_dpos;
1588 	if (has_header) {
1589 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1590 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1591 		if (*tl++ != rpc_call) {
1592 			m_freem(mrep);
1593 			return (EBADRPC);
1594 		}
1595 	} else
1596 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1597 	nd->nd_repstat = 0;
1598 	nd->nd_flag = 0;
1599 	if (*tl++ != rpc_vers) {
1600 		nd->nd_repstat = ERPCMISMATCH;
1601 		nd->nd_procnum = NFSPROC_NOOP;
1602 		return (0);
1603 	}
1604 	if (*tl != nfs_prog) {
1605 		if (*tl == nqnfs_prog)
1606 			nqnfs++;
1607 		else {
1608 			nd->nd_repstat = EPROGUNAVAIL;
1609 			nd->nd_procnum = NFSPROC_NOOP;
1610 			return (0);
1611 		}
1612 	}
1613 	tl++;
1614 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1615 	if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1616 		(nfsvers != NQNFS_VER3 && nqnfs)) {
1617 		nd->nd_repstat = EPROGMISMATCH;
1618 		nd->nd_procnum = NFSPROC_NOOP;
1619 		return (0);
1620 	}
1621 	if (nqnfs)
1622 		nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1623 	else if (nfsvers == NFS_VER3)
1624 		nd->nd_flag = ND_NFSV3;
1625 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1626 	if (nd->nd_procnum == NFSPROC_NULL)
1627 		return (0);
1628 	if (nd->nd_procnum >= NFS_NPROCS ||
1629 		(!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1630 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1631 		nd->nd_repstat = EPROCUNAVAIL;
1632 		nd->nd_procnum = NFSPROC_NOOP;
1633 		return (0);
1634 	}
1635 	if ((nd->nd_flag & ND_NFSV3) == 0)
1636 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1637 	auth_type = *tl++;
1638 	len = fxdr_unsigned(int, *tl++);
1639 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1640 		m_freem(mrep);
1641 		return (EBADRPC);
1642 	}
1643 
1644 	nd->nd_flag &= ~ND_KERBAUTH;
1645 	/*
1646 	 * Handle auth_unix or auth_kerb.
1647 	 */
1648 	if (auth_type == rpc_auth_unix) {
1649 		len = fxdr_unsigned(int, *++tl);
1650 		if (len < 0 || len > NFS_MAXNAMLEN) {
1651 			m_freem(mrep);
1652 			return (EBADRPC);
1653 		}
1654 		nfsm_adv(nfsm_rndup(len));
1655 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1656 		memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred));
1657 		nd->nd_cr.cr_ref = 1;
1658 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1659 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1660 		len = fxdr_unsigned(int, *tl);
1661 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1662 			m_freem(mrep);
1663 			return (EBADRPC);
1664 		}
1665 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1666 		for (i = 0; i < len; i++)
1667 		    if (i < NGROUPS)
1668 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1669 		    else
1670 			tl++;
1671 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1672 		if (nd->nd_cr.cr_ngroups > 1)
1673 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1674 		len = fxdr_unsigned(int, *++tl);
1675 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1676 			m_freem(mrep);
1677 			return (EBADRPC);
1678 		}
1679 		if (len > 0)
1680 			nfsm_adv(nfsm_rndup(len));
1681 	} else if (auth_type == rpc_auth_kerb) {
1682 		switch (fxdr_unsigned(int, *tl++)) {
1683 		case RPCAKN_FULLNAME:
1684 			ticklen = fxdr_unsigned(int, *tl);
1685 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1686 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1687 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1688 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1689 				m_freem(mrep);
1690 				return (EBADRPC);
1691 			}
1692 			uio.uio_offset = 0;
1693 			uio.uio_iov = &iov;
1694 			uio.uio_iovcnt = 1;
1695 			uio.uio_segflg = UIO_SYSSPACE;
1696 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1697 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1698 			nfsm_mtouio(&uio, uio.uio_resid);
1699 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1700 			if (*tl++ != rpc_auth_kerb ||
1701 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1702 				printf("Bad kerb verifier\n");
1703 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1704 				nd->nd_procnum = NFSPROC_NOOP;
1705 				return (0);
1706 			}
1707 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1708 			tl = (u_int32_t *)cp;
1709 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1710 				printf("Not fullname kerb verifier\n");
1711 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1712 				nd->nd_procnum = NFSPROC_NOOP;
1713 				return (0);
1714 			}
1715 			cp += NFSX_UNSIGNED;
1716 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
1717 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1718 			nd->nd_flag |= ND_KERBFULL;
1719 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1720 			break;
1721 		case RPCAKN_NICKNAME:
1722 			if (len != 2 * NFSX_UNSIGNED) {
1723 				printf("Kerb nickname short\n");
1724 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1725 				nd->nd_procnum = NFSPROC_NOOP;
1726 				return (0);
1727 			}
1728 			nickuid = fxdr_unsigned(uid_t, *tl);
1729 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1730 			if (*tl++ != rpc_auth_kerb ||
1731 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1732 				printf("Kerb nick verifier bad\n");
1733 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1734 				nd->nd_procnum = NFSPROC_NOOP;
1735 				return (0);
1736 			}
1737 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1738 			tvin.tv_sec = *tl++;
1739 			tvin.tv_usec = *tl;
1740 
1741 			for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1742 			    nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1743 				if (nuidp->nu_cr.cr_uid == nickuid &&
1744 				    (!nd->nd_nam2 ||
1745 				     netaddr_match(NU_NETFAM(nuidp),
1746 				      &nuidp->nu_haddr, nd->nd_nam2)))
1747 					break;
1748 			}
1749 			if (!nuidp) {
1750 				nd->nd_repstat =
1751 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1752 				nd->nd_procnum = NFSPROC_NOOP;
1753 				return (0);
1754 			}
1755 
1756 			/*
1757 			 * Now, decrypt the timestamp using the session key
1758 			 * and validate it.
1759 			 */
1760 #ifdef NFSKERB
1761 			XXX
1762 #endif
1763 
1764 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1765 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1766 			if (nuidp->nu_expire < time.tv_sec ||
1767 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1768 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1769 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1770 				nuidp->nu_expire = 0;
1771 				nd->nd_repstat =
1772 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1773 				nd->nd_procnum = NFSPROC_NOOP;
1774 				return (0);
1775 			}
1776 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1777 			nd->nd_flag |= ND_KERBNICK;
1778 		};
1779 	} else {
1780 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1781 		nd->nd_procnum = NFSPROC_NOOP;
1782 		return (0);
1783 	}
1784 
1785 	/*
1786 	 * For nqnfs, get piggybacked lease request.
1787 	 */
1788 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
1789 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1790 		nd->nd_flag |= fxdr_unsigned(int, *tl);
1791 		if (nd->nd_flag & ND_LEASE) {
1792 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1793 			nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
1794 		} else
1795 			nd->nd_duration = NQ_MINLEASE;
1796 	} else
1797 		nd->nd_duration = NQ_MINLEASE;
1798 	nd->nd_md = md;
1799 	nd->nd_dpos = dpos;
1800 	return (0);
1801 nfsmout:
1802 	return (error);
1803 }
1804 
1805 int
1806 nfs_msg(p, server, msg)
1807 	struct proc *p;
1808 	char *server, *msg;
1809 {
1810 	tpr_t tpr;
1811 
1812 	if (p)
1813 		tpr = tprintf_open(p);
1814 	else
1815 		tpr = NULL;
1816 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1817 	tprintf_close(tpr);
1818 	return (0);
1819 }
1820 
1821 #ifdef NFSSERVER
1822 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
1823 				    struct nfssvc_sock *, struct proc *,
1824 				    struct mbuf **)) = {
1825 	nfsrv_null,
1826 	nfsrv_getattr,
1827 	nfsrv_setattr,
1828 	nfsrv_lookup,
1829 	nfsrv3_access,
1830 	nfsrv_readlink,
1831 	nfsrv_read,
1832 	nfsrv_write,
1833 	nfsrv_create,
1834 	nfsrv_mkdir,
1835 	nfsrv_symlink,
1836 	nfsrv_mknod,
1837 	nfsrv_remove,
1838 	nfsrv_rmdir,
1839 	nfsrv_rename,
1840 	nfsrv_link,
1841 	nfsrv_readdir,
1842 	nfsrv_readdirplus,
1843 	nfsrv_statfs,
1844 	nfsrv_fsinfo,
1845 	nfsrv_pathconf,
1846 	nfsrv_commit,
1847 	nqnfsrv_getlease,
1848 	nqnfsrv_vacated,
1849 	nfsrv_noop,
1850 	nfsrv_noop
1851 };
1852 
1853 /*
1854  * Socket upcall routine for the nfsd sockets.
1855  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1856  * Essentially do as much as possible non-blocking, else punt and it will
1857  * be called with M_WAIT from an nfsd.
1858  */
1859 void
1860 nfsrv_rcv(so, arg, waitflag)
1861 	struct socket *so;
1862 	caddr_t arg;
1863 	int waitflag;
1864 {
1865 	struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1866 	struct mbuf *m;
1867 	struct mbuf *mp, *nam;
1868 	struct uio auio;
1869 	int flags, error;
1870 
1871 	if ((slp->ns_flag & SLP_VALID) == 0)
1872 		return;
1873 #ifdef notdef
1874 	/*
1875 	 * Define this to test for nfsds handling this under heavy load.
1876 	 */
1877 	if (waitflag == M_DONTWAIT) {
1878 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1879 	}
1880 #endif
1881 	auio.uio_procp = NULL;
1882 	if (so->so_type == SOCK_STREAM) {
1883 		/*
1884 		 * If there are already records on the queue, defer soreceive()
1885 		 * to an nfsd so that there is feedback to the TCP layer that
1886 		 * the nfs servers are heavily loaded.
1887 		 */
1888 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1889 			slp->ns_flag |= SLP_NEEDQ;
1890 			goto dorecs;
1891 		}
1892 
1893 		/*
1894 		 * Do soreceive().
1895 		 */
1896 		auio.uio_resid = 1000000000;
1897 		flags = MSG_DONTWAIT;
1898 		error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1899 		if (error || mp == (struct mbuf *)0) {
1900 			if (error == EWOULDBLOCK)
1901 				slp->ns_flag |= SLP_NEEDQ;
1902 			else
1903 				slp->ns_flag |= SLP_DISCONN;
1904 			goto dorecs;
1905 		}
1906 		m = mp;
1907 		if (slp->ns_rawend) {
1908 			slp->ns_rawend->m_next = m;
1909 			slp->ns_cc += 1000000000 - auio.uio_resid;
1910 		} else {
1911 			slp->ns_raw = m;
1912 			slp->ns_cc = 1000000000 - auio.uio_resid;
1913 		}
1914 		while (m->m_next)
1915 			m = m->m_next;
1916 		slp->ns_rawend = m;
1917 
1918 		/*
1919 		 * Now try and parse record(s) out of the raw stream data.
1920 		 */
1921 		error = nfsrv_getstream(slp, waitflag);
1922 		if (error) {
1923 			if (error == EPERM)
1924 				slp->ns_flag |= SLP_DISCONN;
1925 			else
1926 				slp->ns_flag |= SLP_NEEDQ;
1927 		}
1928 	} else {
1929 		do {
1930 			auio.uio_resid = 1000000000;
1931 			flags = MSG_DONTWAIT;
1932 			error = (*so->so_receive)(so, &nam, &auio, &mp,
1933 						(struct mbuf **)0, &flags);
1934 			if (mp) {
1935 				if (nam) {
1936 					m = nam;
1937 					m->m_next = mp;
1938 				} else
1939 					m = mp;
1940 				if (slp->ns_recend)
1941 					slp->ns_recend->m_nextpkt = m;
1942 				else
1943 					slp->ns_rec = m;
1944 				slp->ns_recend = m;
1945 				m->m_nextpkt = (struct mbuf *)0;
1946 			}
1947 			if (error) {
1948 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
1949 					&& error != EWOULDBLOCK) {
1950 					slp->ns_flag |= SLP_DISCONN;
1951 					goto dorecs;
1952 				}
1953 			}
1954 		} while (mp);
1955 	}
1956 
1957 	/*
1958 	 * Now try and process the request records, non-blocking.
1959 	 */
1960 dorecs:
1961 	if (waitflag == M_DONTWAIT &&
1962 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
1963 		nfsrv_wakenfsd(slp);
1964 }
1965 
1966 /*
1967  * Try and extract an RPC request from the mbuf data list received on a
1968  * stream socket. The "waitflag" argument indicates whether or not it
1969  * can sleep.
1970  */
1971 int
1972 nfsrv_getstream(slp, waitflag)
1973 	struct nfssvc_sock *slp;
1974 	int waitflag;
1975 {
1976 	struct mbuf *m, **mpp;
1977 	char *cp1, *cp2;
1978 	int len;
1979 	struct mbuf *om, *m2, *recm = NULL;
1980 	u_int32_t recmark;
1981 
1982 	if (slp->ns_flag & SLP_GETSTREAM)
1983 		panic("nfs getstream");
1984 	slp->ns_flag |= SLP_GETSTREAM;
1985 	for (;;) {
1986 	    if (slp->ns_reclen == 0) {
1987 		if (slp->ns_cc < NFSX_UNSIGNED) {
1988 			slp->ns_flag &= ~SLP_GETSTREAM;
1989 			return (0);
1990 		}
1991 		m = slp->ns_raw;
1992 		if (m->m_len >= NFSX_UNSIGNED) {
1993 			memcpy((caddr_t)&recmark, mtod(m, caddr_t), NFSX_UNSIGNED);
1994 			m->m_data += NFSX_UNSIGNED;
1995 			m->m_len -= NFSX_UNSIGNED;
1996 		} else {
1997 			cp1 = (caddr_t)&recmark;
1998 			cp2 = mtod(m, caddr_t);
1999 			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
2000 				while (m->m_len == 0) {
2001 					m = m->m_next;
2002 					cp2 = mtod(m, caddr_t);
2003 				}
2004 				*cp1++ = *cp2++;
2005 				m->m_data++;
2006 				m->m_len--;
2007 			}
2008 		}
2009 		slp->ns_cc -= NFSX_UNSIGNED;
2010 		recmark = ntohl(recmark);
2011 		slp->ns_reclen = recmark & ~0x80000000;
2012 		if (recmark & 0x80000000)
2013 			slp->ns_flag |= SLP_LASTFRAG;
2014 		else
2015 			slp->ns_flag &= ~SLP_LASTFRAG;
2016 		if (slp->ns_reclen > NFS_MAXPACKET) {
2017 			slp->ns_flag &= ~SLP_GETSTREAM;
2018 			return (EPERM);
2019 		}
2020 	    }
2021 
2022 	    /*
2023 	     * Now get the record part.
2024 	     */
2025 	    if (slp->ns_cc == slp->ns_reclen) {
2026 		recm = slp->ns_raw;
2027 		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2028 		slp->ns_cc = slp->ns_reclen = 0;
2029 	    } else if (slp->ns_cc > slp->ns_reclen) {
2030 		len = 0;
2031 		m = slp->ns_raw;
2032 		om = (struct mbuf *)0;
2033 		while (len < slp->ns_reclen) {
2034 			if ((len + m->m_len) > slp->ns_reclen) {
2035 				size_t left = slp->ns_reclen - len;
2036 
2037 				MGETHDR(m2, waitflag, m->m_type);
2038 				if (m2 == NULL) {
2039 					slp->ns_flag &= ~SLP_GETSTREAM;
2040 					return (EWOULDBLOCK);
2041 				}
2042 				if (left > MHLEN) {
2043 					MCLGET(m2, waitflag);
2044 					if (!(m2->m_flags & M_EXT)) {
2045 						m_freem(m2);
2046 						slp->ns_flag &= ~SLP_GETSTREAM;
2047 						return (EWOULDBLOCK);
2048 					}
2049 				}
2050 				memcpy(mtod(m2, caddr_t), mtod(m, caddr_t),
2051 				    left);
2052 				m2->m_len = left;
2053 				m->m_data += left;
2054 				m->m_len -= left;
2055 				if (om) {
2056 					om->m_next = m2;
2057 					recm = slp->ns_raw;
2058 				} else
2059 					recm = m2;
2060 				len = slp->ns_reclen;
2061 			} else if ((len + m->m_len) == slp->ns_reclen) {
2062 				om = m;
2063 				len += m->m_len;
2064 				m = m->m_next;
2065 				recm = slp->ns_raw;
2066 				om->m_next = (struct mbuf *)0;
2067 			} else {
2068 				om = m;
2069 				len += m->m_len;
2070 				m = m->m_next;
2071 			}
2072 		}
2073 		slp->ns_raw = m;
2074 		slp->ns_cc -= len;
2075 		slp->ns_reclen = 0;
2076 	    } else {
2077 		slp->ns_flag &= ~SLP_GETSTREAM;
2078 		return (0);
2079 	    }
2080 
2081 	    /*
2082 	     * Accumulate the fragments into a record.
2083 	     */
2084 	    mpp = &slp->ns_frag;
2085 	    while (*mpp)
2086 		mpp = &((*mpp)->m_next);
2087 	    *mpp = recm;
2088 	    if (slp->ns_flag & SLP_LASTFRAG) {
2089 		if (slp->ns_recend)
2090 		    slp->ns_recend->m_nextpkt = slp->ns_frag;
2091 		else
2092 		    slp->ns_rec = slp->ns_frag;
2093 		slp->ns_recend = slp->ns_frag;
2094 		slp->ns_frag = (struct mbuf *)0;
2095 	    }
2096 	}
2097 }
2098 
2099 /*
2100  * Parse an RPC header.
2101  */
2102 int
2103 nfsrv_dorec(slp, nfsd, ndp)
2104 	struct nfssvc_sock *slp;
2105 	struct nfsd *nfsd;
2106 	struct nfsrv_descript **ndp;
2107 {
2108 	struct mbuf *m, *nam;
2109 	struct nfsrv_descript *nd;
2110 	int error;
2111 
2112 	*ndp = NULL;
2113 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2114 	    (m = slp->ns_rec) == (struct mbuf *)0)
2115 		return (ENOBUFS);
2116 	slp->ns_rec = m->m_nextpkt;
2117 	if (slp->ns_rec)
2118 		m->m_nextpkt = (struct mbuf *)0;
2119 	else
2120 		slp->ns_recend = (struct mbuf *)0;
2121 	if (m->m_type == MT_SONAME) {
2122 		nam = m;
2123 		m = m->m_next;
2124 		nam->m_next = NULL;
2125 	} else
2126 		nam = NULL;
2127 	MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2128 		M_NFSRVDESC, M_WAITOK);
2129 	nd->nd_md = nd->nd_mrep = m;
2130 	nd->nd_nam2 = nam;
2131 	nd->nd_dpos = mtod(m, caddr_t);
2132 	error = nfs_getreq(nd, nfsd, TRUE);
2133 	if (error) {
2134 		m_freem(nam);
2135 		free((caddr_t)nd, M_NFSRVDESC);
2136 		return (error);
2137 	}
2138 	*ndp = nd;
2139 	nfsd->nfsd_nd = nd;
2140 	return (0);
2141 }
2142 
2143 
2144 /*
2145  * Search for a sleeping nfsd and wake it up.
2146  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2147  * running nfsds will go look for the work in the nfssvc_sock list.
2148  */
2149 void
2150 nfsrv_wakenfsd(slp)
2151 	struct nfssvc_sock *slp;
2152 {
2153 	struct nfsd *nd;
2154 
2155 	if ((slp->ns_flag & SLP_VALID) == 0)
2156 		return;
2157 	for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) {
2158 		if (nd->nfsd_flag & NFSD_WAITING) {
2159 			nd->nfsd_flag &= ~NFSD_WAITING;
2160 			if (nd->nfsd_slp)
2161 				panic("nfsd wakeup");
2162 			slp->ns_sref++;
2163 			nd->nfsd_slp = slp;
2164 			wakeup((caddr_t)nd);
2165 			return;
2166 		}
2167 	}
2168 	slp->ns_flag |= SLP_DOREC;
2169 	nfsd_head_flag |= NFSD_CHECKSLP;
2170 }
2171 #endif /* NFSSERVER */
2172