xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 3b01aba77a7a698587faaae455bbfe740923c1f5)
1 /*	$NetBSD: nfs_socket.c,v 1.67 2001/05/09 20:02:32 fvdl 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_nfs.h"
47 #include "opt_nfsserver.h"
48 #include "opt_inet.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/callout.h>
53 #include <sys/proc.h>
54 #include <sys/mount.h>
55 #include <sys/kernel.h>
56 #include <sys/mbuf.h>
57 #include <sys/vnode.h>
58 #include <sys/domain.h>
59 #include <sys/protosw.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
62 #include <sys/syslog.h>
63 #include <sys/tprintf.h>
64 #include <sys/namei.h>
65 #include <sys/signal.h>
66 #include <sys/signalvar.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/tcp.h>
70 
71 #include <nfs/rpcv2.h>
72 #include <nfs/nfsproto.h>
73 #include <nfs/nfs.h>
74 #include <nfs/xdr_subs.h>
75 #include <nfs/nfsm_subs.h>
76 #include <nfs/nfsmount.h>
77 #include <nfs/nfsnode.h>
78 #include <nfs/nfsrtt.h>
79 #include <nfs/nqnfs.h>
80 #include <nfs/nfs_var.h>
81 
82 #define	TRUE	1
83 #define	FALSE	0
84 
85 /*
86  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
87  * Use the mean and mean deviation of rtt for the appropriate type of rpc
88  * for the frequent rpcs and a default for the others.
89  * The justification for doing "other" this way is that these rpcs
90  * happen so infrequently that timer est. would probably be stale.
91  * Also, since many of these rpcs are
92  * non-idempotent, a conservative timeout is desired.
93  * getattr, lookup - A+2D
94  * read, write     - A+4D
95  * other           - nm_timeo
96  */
97 #define	NFS_RTO(n, t) \
98 	((t) == 0 ? (n)->nm_timeo : \
99 	 ((t) < 3 ? \
100 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
101 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
102 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
103 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
104 /*
105  * External data, mostly RPC constants in XDR form
106  */
107 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
108 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
109 	rpc_auth_kerb;
110 extern u_int32_t nfs_prog, nqnfs_prog;
111 extern time_t nqnfsstarttime;
112 extern struct nfsstats nfsstats;
113 extern int nfsv3_procid[NFS_NPROCS];
114 extern int nfs_ticks;
115 
116 /*
117  * Defines which timer to use for the procnum.
118  * 0 - default
119  * 1 - getattr
120  * 2 - lookup
121  * 3 - read
122  * 4 - write
123  */
124 static const int proct[NFS_NPROCS] = {
125 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
126 	0, 0, 0,
127 };
128 
129 /*
130  * There is a congestion window for outstanding rpcs maintained per mount
131  * point. The cwnd size is adjusted in roughly the way that:
132  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
133  * SIGCOMM '88". ACM, August 1988.
134  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
135  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
136  * of rpcs is in progress.
137  * (The sent count and cwnd are scaled for integer arith.)
138  * Variants of "slow start" were tried and were found to be too much of a
139  * performance hit (ave. rtt 3 times larger),
140  * I suspect due to the large rtt that nfs rpcs have.
141  */
142 #define	NFS_CWNDSCALE	256
143 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
144 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
145 int nfsrtton = 0;
146 struct nfsrtt nfsrtt;
147 
148 struct callout nfs_timer_ch = CALLOUT_INITIALIZER;
149 
150 /*
151  * Initialize sockets and congestion for a new NFS connection.
152  * We do not free the sockaddr if error.
153  */
154 int
155 nfs_connect(nmp, rep)
156 	struct nfsmount *nmp;
157 	struct nfsreq *rep;
158 {
159 	struct socket *so;
160 	int s, error, rcvreserve, sndreserve;
161 	struct sockaddr *saddr;
162 	struct sockaddr_in *sin;
163 #ifdef INET6
164 	struct sockaddr_in6 *sin6;
165 #endif
166 	struct mbuf *m;
167 	u_int16_t tport;
168 
169 	nmp->nm_so = (struct socket *)0;
170 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
171 	error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
172 		nmp->nm_soproto);
173 	if (error)
174 		goto bad;
175 	so = nmp->nm_so;
176 	nmp->nm_soflags = so->so_proto->pr_flags;
177 
178 	/*
179 	 * Some servers require that the client port be a reserved port number.
180 	 */
181 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
182 		MGET(m, M_WAIT, MT_SONAME);
183 		sin = mtod(m, struct sockaddr_in *);
184 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
185 		sin->sin_family = AF_INET;
186 		sin->sin_addr.s_addr = INADDR_ANY;
187 		tport = IPPORT_RESERVED - 1;
188 		sin->sin_port = htons(tport);
189 		while ((error = sobind(so, m, &proc0)) == EADDRINUSE &&
190 		       --tport > IPPORT_RESERVED / 2)
191 			sin->sin_port = htons(tport);
192 		m_freem(m);
193 		if (error)
194 			goto bad;
195 	}
196 #ifdef INET6
197 	if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) {
198 		MGET(m, M_WAIT, MT_SONAME);
199 		sin6 = mtod(m, struct sockaddr_in6 *);
200 		sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6);
201 		sin6->sin6_family = AF_INET6;
202 		sin6->sin6_addr = in6addr_any;
203 		tport = IPV6PORT_RESERVED - 1;
204 		sin6->sin6_port = htons(tport);
205 		while ((error = sobind(so, m, &proc0)) == EADDRINUSE &&
206 		       --tport > IPV6PORT_RESERVED / 2)
207 			sin6->sin6_port = htons(tport);
208 		m_freem(m);
209 		if (error)
210 			goto bad;
211 	}
212 #endif
213 
214 	/*
215 	 * Protocols that do not require connections may be optionally left
216 	 * unconnected for servers that reply from a port other than NFS_PORT.
217 	 */
218 	if (nmp->nm_flag & NFSMNT_NOCONN) {
219 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
220 			error = ENOTCONN;
221 			goto bad;
222 		}
223 	} else {
224 		error = soconnect(so, nmp->nm_nam);
225 		if (error)
226 			goto bad;
227 
228 		/*
229 		 * Wait for the connection to complete. Cribbed from the
230 		 * connect system call but with the wait timing out so
231 		 * that interruptible mounts don't hang here for a long time.
232 		 */
233 		s = splsoftnet();
234 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
235 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
236 				"nfscn1", 2 * hz);
237 			if ((so->so_state & SS_ISCONNECTING) &&
238 			    so->so_error == 0 && rep &&
239 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
240 				so->so_state &= ~SS_ISCONNECTING;
241 				splx(s);
242 				goto bad;
243 			}
244 		}
245 		if (so->so_error) {
246 			error = so->so_error;
247 			so->so_error = 0;
248 			splx(s);
249 			goto bad;
250 		}
251 		splx(s);
252 	}
253 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
254 		so->so_rcv.sb_timeo = (5 * hz);
255 		so->so_snd.sb_timeo = (5 * hz);
256 	} else {
257 		so->so_rcv.sb_timeo = 0;
258 		so->so_snd.sb_timeo = 0;
259 	}
260 	if (nmp->nm_sotype == SOCK_DGRAM) {
261 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
262 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
263 		    NFS_MAXPKTHDR) * 2;
264 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
265 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
266 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
267 		    NFS_MAXPKTHDR) * 2;
268 	} else {
269 		if (nmp->nm_sotype != SOCK_STREAM)
270 			panic("nfscon sotype");
271 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
272 			MGET(m, M_WAIT, MT_SOOPTS);
273 			*mtod(m, int32_t *) = 1;
274 			m->m_len = sizeof(int32_t);
275 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
276 		}
277 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
278 			MGET(m, M_WAIT, MT_SOOPTS);
279 			*mtod(m, int32_t *) = 1;
280 			m->m_len = sizeof(int32_t);
281 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
282 		}
283 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
284 		    sizeof (u_int32_t)) * 2;
285 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
286 		    sizeof (u_int32_t)) * 2;
287 	}
288 	error = soreserve(so, sndreserve, rcvreserve);
289 	if (error)
290 		goto bad;
291 	so->so_rcv.sb_flags |= SB_NOINTR;
292 	so->so_snd.sb_flags |= SB_NOINTR;
293 
294 	/* Initialize other non-zero congestion variables */
295 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
296 		nmp->nm_srtt[4] = (NFS_TIMEO << 3);
297 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
298 		nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0;
299 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
300 	nmp->nm_sent = 0;
301 	nmp->nm_timeouts = 0;
302 	return (0);
303 
304 bad:
305 	nfs_disconnect(nmp);
306 	return (error);
307 }
308 
309 /*
310  * Reconnect routine:
311  * Called when a connection is broken on a reliable protocol.
312  * - clean up the old socket
313  * - nfs_connect() again
314  * - set R_MUSTRESEND for all outstanding requests on mount point
315  * If this fails the mount point is DEAD!
316  * nb: Must be called with the nfs_sndlock() set on the mount point.
317  */
318 int
319 nfs_reconnect(rep)
320 	struct nfsreq *rep;
321 {
322 	struct nfsreq *rp;
323 	struct nfsmount *nmp = rep->r_nmp;
324 	int error;
325 
326 	nfs_disconnect(nmp);
327 	while ((error = nfs_connect(nmp, rep)) != 0) {
328 		if (error == EINTR || error == ERESTART)
329 			return (EINTR);
330 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0);
331 	}
332 
333 	/*
334 	 * Loop through outstanding request list and fix up all requests
335 	 * on old socket.
336 	 */
337 	for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) {
338 		if (rp->r_nmp == nmp)
339 			rp->r_flags |= R_MUSTRESEND;
340 	}
341 	return (0);
342 }
343 
344 /*
345  * NFS disconnect. Clean up and unlink.
346  */
347 void
348 nfs_disconnect(nmp)
349 	struct nfsmount *nmp;
350 {
351 	struct socket *so;
352 	int drain = 0;
353 
354 	if (nmp->nm_so) {
355 		so = nmp->nm_so;
356 		nmp->nm_so = (struct socket *)0;
357 		soshutdown(so, 2);
358 		drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
359 		if (drain) {
360 			/*
361 			 * soshutdown() above should wake up the current
362 			 * listener.
363 			 * Now wake up those waiting for the recive lock, and
364 			 * wait for them to go away unhappy, to prevent *nmp
365 			 * from evaporating while they're sleeping.
366 			 */
367 			while (nmp->nm_waiters > 0) {
368 				wakeup (&nmp->nm_iflag);
369 				(void) tsleep(&nmp->nm_waiters, PVFS,
370 				    "nfsdis", 0);
371 			}
372 		}
373 		soclose(so);
374 	}
375 #ifdef DIAGNOSTIC
376 	if (drain && (nmp->nm_waiters > 0))
377 		panic("nfs_disconnect: waiters left after drain?\n");
378 #endif
379 }
380 
381 void
382 nfs_safedisconnect(nmp)
383 	struct nfsmount *nmp;
384 {
385 	struct nfsreq dummyreq;
386 
387 	memset(&dummyreq, 0, sizeof(dummyreq));
388 	dummyreq.r_nmp = nmp;
389 	nfs_rcvlock(&dummyreq); /* XXX ignored error return */
390 	nfs_disconnect(nmp);
391 	nfs_rcvunlock(&nmp->nm_iflag);
392 }
393 
394 /*
395  * This is the nfs send routine. For connection based socket types, it
396  * must be called with an nfs_sndlock() on the socket.
397  * "rep == NULL" indicates that it has been called from a server.
398  * For the client side:
399  * - return EINTR if the RPC is terminated, 0 otherwise
400  * - set R_MUSTRESEND if the send fails for any reason
401  * - do any cleanup required by recoverable socket errors (? ? ?)
402  * For the server side:
403  * - return EINTR or ERESTART if interrupted by a signal
404  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
405  * - do any cleanup required by recoverable socket errors (? ? ?)
406  */
407 int
408 nfs_send(so, nam, top, rep)
409 	struct socket *so;
410 	struct mbuf *nam;
411 	struct mbuf *top;
412 	struct nfsreq *rep;
413 {
414 	struct mbuf *sendnam;
415 	int error, soflags, flags;
416 
417 	if (rep) {
418 		if (rep->r_flags & R_SOFTTERM) {
419 			m_freem(top);
420 			return (EINTR);
421 		}
422 		if ((so = rep->r_nmp->nm_so) == NULL) {
423 			rep->r_flags |= R_MUSTRESEND;
424 			m_freem(top);
425 			return (0);
426 		}
427 		rep->r_flags &= ~R_MUSTRESEND;
428 		soflags = rep->r_nmp->nm_soflags;
429 	} else
430 		soflags = so->so_proto->pr_flags;
431 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
432 		sendnam = (struct mbuf *)0;
433 	else
434 		sendnam = nam;
435 	if (so->so_type == SOCK_SEQPACKET)
436 		flags = MSG_EOR;
437 	else
438 		flags = 0;
439 
440 	error = (*so->so_send)(so, sendnam, (struct uio *)0, top,
441 		(struct mbuf *)0, flags);
442 	if (error) {
443 		if (rep) {
444 			if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
445 				/*
446 				 * We're too fast for the network/driver,
447 				 * and UDP isn't flowcontrolled.
448 				 * We need to resend. This is not fatal,
449 				 * just try again.
450 				 *
451 				 * Could be smarter here by doing some sort
452 				 * of a backoff, but this is rare.
453 				 */
454 				rep->r_flags |= R_MUSTRESEND;
455 			} else {
456 				log(LOG_INFO, "nfs send error %d for %s\n",
457 				    error,
458 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
459 				/*
460 				 * Deal with errors for the client side.
461 				 */
462 				if (rep->r_flags & R_SOFTTERM)
463 					error = EINTR;
464 				else
465 					rep->r_flags |= R_MUSTRESEND;
466 			}
467 		} else {
468 			/*
469 			 * See above. This error can happen under normal
470 			 * circumstances and the log is too noisy.
471 			 * The error will still show up in nfsstat.
472 			 */
473 			if (error != ENOBUFS || so->so_type != SOCK_DGRAM)
474 				log(LOG_INFO, "nfsd send error %d\n", error);
475 		}
476 
477 		/*
478 		 * Handle any recoverable (soft) socket errors here. (? ? ?)
479 		 */
480 		if (error != EINTR && error != ERESTART &&
481 			error != EWOULDBLOCK && error != EPIPE)
482 			error = 0;
483 	}
484 	return (error);
485 }
486 
487 #ifdef NFS
488 /*
489  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
490  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
491  * Mark and consolidate the data into a new mbuf list.
492  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
493  *     small mbufs.
494  * For SOCK_STREAM we must be very careful to read an entire record once
495  * we have read any of it, even if the system call has been interrupted.
496  */
497 int
498 nfs_receive(rep, aname, mp)
499 	struct nfsreq *rep;
500 	struct mbuf **aname;
501 	struct mbuf **mp;
502 {
503 	struct socket *so;
504 	struct uio auio;
505 	struct iovec aio;
506 	struct mbuf *m;
507 	struct mbuf *control;
508 	u_int32_t len;
509 	struct mbuf **getnam;
510 	int error, sotype, rcvflg;
511 	struct proc *p = curproc;	/* XXX */
512 
513 	/*
514 	 * Set up arguments for soreceive()
515 	 */
516 	*mp = (struct mbuf *)0;
517 	*aname = (struct mbuf *)0;
518 	sotype = rep->r_nmp->nm_sotype;
519 
520 	/*
521 	 * For reliable protocols, lock against other senders/receivers
522 	 * in case a reconnect is necessary.
523 	 * For SOCK_STREAM, first get the Record Mark to find out how much
524 	 * more there is to get.
525 	 * We must lock the socket against other receivers
526 	 * until we have an entire rpc request/reply.
527 	 */
528 	if (sotype != SOCK_DGRAM) {
529 		error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
530 		if (error)
531 			return (error);
532 tryagain:
533 		/*
534 		 * Check for fatal errors and resending request.
535 		 */
536 		/*
537 		 * Ugh: If a reconnect attempt just happened, nm_so
538 		 * would have changed. NULL indicates a failed
539 		 * attempt that has essentially shut down this
540 		 * mount point.
541 		 */
542 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
543 			nfs_sndunlock(&rep->r_nmp->nm_iflag);
544 			return (EINTR);
545 		}
546 		so = rep->r_nmp->nm_so;
547 		if (!so) {
548 			error = nfs_reconnect(rep);
549 			if (error) {
550 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
551 				return (error);
552 			}
553 			goto tryagain;
554 		}
555 		while (rep->r_flags & R_MUSTRESEND) {
556 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
557 			nfsstats.rpcretries++;
558 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
559 			if (error) {
560 				if (error == EINTR || error == ERESTART ||
561 				    (error = nfs_reconnect(rep)) != 0) {
562 					nfs_sndunlock(&rep->r_nmp->nm_iflag);
563 					return (error);
564 				}
565 				goto tryagain;
566 			}
567 		}
568 		nfs_sndunlock(&rep->r_nmp->nm_iflag);
569 		if (sotype == SOCK_STREAM) {
570 			aio.iov_base = (caddr_t) &len;
571 			aio.iov_len = sizeof(u_int32_t);
572 			auio.uio_iov = &aio;
573 			auio.uio_iovcnt = 1;
574 			auio.uio_segflg = UIO_SYSSPACE;
575 			auio.uio_rw = UIO_READ;
576 			auio.uio_offset = 0;
577 			auio.uio_resid = sizeof(u_int32_t);
578 			auio.uio_procp = p;
579 			do {
580 			   rcvflg = MSG_WAITALL;
581 			   error = (*so->so_receive)(so, (struct mbuf **)0, &auio,
582 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
583 			   if (error == EWOULDBLOCK && rep) {
584 				if (rep->r_flags & R_SOFTTERM)
585 					return (EINTR);
586 			   }
587 			} while (error == EWOULDBLOCK);
588 			if (!error && auio.uio_resid > 0) {
589 			    /*
590 			     * Don't log a 0 byte receive; it means
591 			     * that the socket has been closed, and
592 			     * can happen during normal operation
593 			     * (forcible unmount or Solaris server).
594 			     */
595 			    if (auio.uio_resid != sizeof (u_int32_t))
596 			      log(LOG_INFO,
597 				 "short receive (%lu/%lu) from nfs server %s\n",
598 				 (u_long)sizeof(u_int32_t) - auio.uio_resid,
599 				 (u_long)sizeof(u_int32_t),
600 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
601 			    error = EPIPE;
602 			}
603 			if (error)
604 				goto errout;
605 			len = ntohl(len) & ~0x80000000;
606 			/*
607 			 * This is SERIOUS! We are out of sync with the sender
608 			 * and forcing a disconnect/reconnect is all I can do.
609 			 */
610 			if (len > NFS_MAXPACKET) {
611 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
612 				"impossible packet length",
613 				len,
614 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
615 			    error = EFBIG;
616 			    goto errout;
617 			}
618 			auio.uio_resid = len;
619 			do {
620 			    rcvflg = MSG_WAITALL;
621 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
622 				&auio, mp, (struct mbuf **)0, &rcvflg);
623 			} while (error == EWOULDBLOCK || error == EINTR ||
624 				 error == ERESTART);
625 			if (!error && auio.uio_resid > 0) {
626 			    if (len != auio.uio_resid)
627 			      log(LOG_INFO,
628 				"short receive (%lu/%d) from nfs server %s\n",
629 				(u_long)len - auio.uio_resid, len,
630 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
631 			    error = EPIPE;
632 			}
633 		} else {
634 			/*
635 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
636 			 * and soreceive() will return when it has either a
637 			 * control msg or a data msg.
638 			 * We have no use for control msg., but must grab them
639 			 * and then throw them away so we know what is going
640 			 * on.
641 			 */
642 			auio.uio_resid = len = 100000000; /* Anything Big */
643 			auio.uio_procp = p;
644 			do {
645 			    rcvflg = 0;
646 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
647 				&auio, mp, &control, &rcvflg);
648 			    if (control)
649 				m_freem(control);
650 			    if (error == EWOULDBLOCK && rep) {
651 				if (rep->r_flags & R_SOFTTERM)
652 					return (EINTR);
653 			    }
654 			} while (error == EWOULDBLOCK ||
655 				 (!error && *mp == NULL && control));
656 			if ((rcvflg & MSG_EOR) == 0)
657 				printf("Egad!!\n");
658 			if (!error && *mp == NULL)
659 				error = EPIPE;
660 			len -= auio.uio_resid;
661 		}
662 errout:
663 		if (error && error != EINTR && error != ERESTART) {
664 			m_freem(*mp);
665 			*mp = (struct mbuf *)0;
666 			if (error != EPIPE)
667 				log(LOG_INFO,
668 				    "receive error %d from nfs server %s\n",
669 				    error,
670 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
671 			error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
672 			if (!error)
673 				error = nfs_reconnect(rep);
674 			if (!error)
675 				goto tryagain;
676 			else
677 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
678 		}
679 	} else {
680 		if ((so = rep->r_nmp->nm_so) == NULL)
681 			return (EACCES);
682 		if (so->so_state & SS_ISCONNECTED)
683 			getnam = (struct mbuf **)0;
684 		else
685 			getnam = aname;
686 		auio.uio_resid = len = 1000000;
687 		auio.uio_procp = p;
688 		do {
689 			rcvflg = 0;
690 			error =  (*so->so_receive)(so, getnam, &auio, mp,
691 				(struct mbuf **)0, &rcvflg);
692 			if (error == EWOULDBLOCK &&
693 			    (rep->r_flags & R_SOFTTERM))
694 				return (EINTR);
695 		} while (error == EWOULDBLOCK);
696 		len -= auio.uio_resid;
697 		if (!error && *mp == NULL)
698 			error = EPIPE;
699 	}
700 	if (error) {
701 		m_freem(*mp);
702 		*mp = (struct mbuf *)0;
703 	}
704 	return (error);
705 }
706 
707 /*
708  * Implement receipt of reply on a socket.
709  * We must search through the list of received datagrams matching them
710  * with outstanding requests using the xid, until ours is found.
711  */
712 /* ARGSUSED */
713 int
714 nfs_reply(myrep)
715 	struct nfsreq *myrep;
716 {
717 	struct nfsreq *rep;
718 	struct nfsmount *nmp = myrep->r_nmp;
719 	int32_t t1;
720 	struct mbuf *mrep, *nam, *md;
721 	u_int32_t rxid, *tl;
722 	caddr_t dpos, cp2;
723 	int error;
724 
725 	/*
726 	 * Loop around until we get our own reply
727 	 */
728 	for (;;) {
729 		/*
730 		 * Lock against other receivers so that I don't get stuck in
731 		 * sbwait() after someone else has received my reply for me.
732 		 * Also necessary for connection based protocols to avoid
733 		 * race conditions during a reconnect.
734 		 */
735 		error = nfs_rcvlock(myrep);
736 		if (error == EALREADY)
737 			return (0);
738 		if (error)
739 			return (error);
740 		/*
741 		 * Get the next Rpc reply off the socket
742 		 */
743 		nmp->nm_waiters++;
744 		error = nfs_receive(myrep, &nam, &mrep);
745 		nfs_rcvunlock(&nmp->nm_iflag);
746 		if (error) {
747 
748 			if (nmp->nm_iflag & NFSMNT_DISMNT) {
749 				/*
750 				 * Oops, we're going away now..
751 				 */
752 				nmp->nm_waiters--;
753 				wakeup (&nmp->nm_waiters);
754 				return error;
755 			}
756 			nmp->nm_waiters--;
757 			/*
758 			 * Ignore routing errors on connectionless protocols? ?
759 			 */
760 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
761 				nmp->nm_so->so_error = 0;
762 #ifdef DEBUG
763 				printf("nfs_reply: ignoring error %d\n", error);
764 #endif
765 				if (myrep->r_flags & R_GETONEREP)
766 					return (0);
767 				continue;
768 			}
769 			return (error);
770 		}
771 		nmp->nm_waiters--;
772 		if (nam)
773 			m_freem(nam);
774 
775 		/*
776 		 * Get the xid and check that it is an rpc reply
777 		 */
778 		md = mrep;
779 		dpos = mtod(md, caddr_t);
780 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
781 		rxid = *tl++;
782 		if (*tl != rpc_reply) {
783 #ifndef NFS_V2_ONLY
784 			if (nmp->nm_flag & NFSMNT_NQNFS) {
785 				if (nqnfs_callback(nmp, mrep, md, dpos))
786 					nfsstats.rpcinvalid++;
787 			} else
788 #endif
789 			{
790 				nfsstats.rpcinvalid++;
791 				m_freem(mrep);
792 			}
793 nfsmout:
794 			if (myrep->r_flags & R_GETONEREP)
795 				return (0);
796 			continue;
797 		}
798 
799 		/*
800 		 * Loop through the request list to match up the reply
801 		 * Iff no match, just drop the datagram
802 		 */
803 		for (rep = nfs_reqq.tqh_first; rep != 0;
804 		    rep = rep->r_chain.tqe_next) {
805 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
806 				/* Found it.. */
807 				rep->r_mrep = mrep;
808 				rep->r_md = md;
809 				rep->r_dpos = dpos;
810 				if (nfsrtton) {
811 					struct rttl *rt;
812 
813 					rt = &nfsrtt.rttl[nfsrtt.pos];
814 					rt->proc = rep->r_procnum;
815 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
816 					rt->sent = nmp->nm_sent;
817 					rt->cwnd = nmp->nm_cwnd;
818 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
819 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
820 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
821 					rt->tstamp = time;
822 					if (rep->r_flags & R_TIMING)
823 						rt->rtt = rep->r_rtt;
824 					else
825 						rt->rtt = 1000000;
826 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
827 				}
828 				/*
829 				 * Update congestion window.
830 				 * Do the additive increase of
831 				 * one rpc/rtt.
832 				 */
833 				if (nmp->nm_cwnd <= nmp->nm_sent) {
834 					nmp->nm_cwnd +=
835 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
836 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
837 					if (nmp->nm_cwnd > NFS_MAXCWND)
838 						nmp->nm_cwnd = NFS_MAXCWND;
839 				}
840 				rep->r_flags &= ~R_SENT;
841 				nmp->nm_sent -= NFS_CWNDSCALE;
842 				/*
843 				 * Update rtt using a gain of 0.125 on the mean
844 				 * and a gain of 0.25 on the deviation.
845 				 */
846 				if (rep->r_flags & R_TIMING) {
847 					/*
848 					 * Since the timer resolution of
849 					 * NFS_HZ is so course, it can often
850 					 * result in r_rtt == 0. Since
851 					 * r_rtt == N means that the actual
852 					 * rtt is between N+dt and N+2-dt ticks,
853 					 * add 1.
854 					 */
855 					t1 = rep->r_rtt + 1;
856 					t1 -= (NFS_SRTT(rep) >> 3);
857 					NFS_SRTT(rep) += t1;
858 					if (t1 < 0)
859 						t1 = -t1;
860 					t1 -= (NFS_SDRTT(rep) >> 2);
861 					NFS_SDRTT(rep) += t1;
862 				}
863 				nmp->nm_timeouts = 0;
864 				break;
865 			}
866 		}
867 		/*
868 		 * If not matched to a request, drop it.
869 		 * If it's mine, get out.
870 		 */
871 		if (rep == 0) {
872 			nfsstats.rpcunexpected++;
873 			m_freem(mrep);
874 		} else if (rep == myrep) {
875 			if (rep->r_mrep == NULL)
876 				panic("nfsreply nil");
877 			return (0);
878 		}
879 		if (myrep->r_flags & R_GETONEREP)
880 			return (0);
881 	}
882 }
883 
884 /*
885  * nfs_request - goes something like this
886  *	- fill in request struct
887  *	- links it into list
888  *	- calls nfs_send() for first transmit
889  *	- calls nfs_receive() to get reply
890  *	- break down rpc header and return with nfs reply pointed to
891  *	  by mrep or error
892  * nb: always frees up mreq mbuf list
893  */
894 int
895 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
896 	struct vnode *vp;
897 	struct mbuf *mrest;
898 	int procnum;
899 	struct proc *procp;
900 	struct ucred *cred;
901 	struct mbuf **mrp;
902 	struct mbuf **mdp;
903 	caddr_t *dposp;
904 {
905 	struct mbuf *m, *mrep;
906 	struct nfsreq *rep;
907 	u_int32_t *tl;
908 	int i;
909 	struct nfsmount *nmp;
910 	struct mbuf *md, *mheadend;
911 	char nickv[RPCX_NICKVERF];
912 	time_t reqtime, waituntil;
913 	caddr_t dpos, cp2;
914 	int t1, s, error = 0, mrest_len, auth_len, auth_type;
915 	int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0;
916 	int verf_len, verf_type;
917 	u_int32_t xid;
918 	char *auth_str, *verf_str;
919 	NFSKERBKEY_T key;		/* save session key */
920 #ifndef NFS_V2_ONLY
921 	int nqlflag, cachable;
922 	u_quad_t frev;
923 	struct nfsnode *np;
924 #endif
925 
926 	KASSERT(cred != NULL);
927 	nmp = VFSTONFS(vp->v_mount);
928 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
929 	rep->r_nmp = nmp;
930 	rep->r_vp = vp;
931 	rep->r_procp = procp;
932 	rep->r_procnum = procnum;
933 	i = 0;
934 	m = mrest;
935 	while (m) {
936 		i += m->m_len;
937 		m = m->m_next;
938 	}
939 	mrest_len = i;
940 
941 	/*
942 	 * Get the RPC header with authorization.
943 	 */
944 kerbauth:
945 	verf_str = auth_str = (char *)0;
946 	if (nmp->nm_flag & NFSMNT_KERB) {
947 		verf_str = nickv;
948 		verf_len = sizeof (nickv);
949 		auth_type = RPCAUTH_KERB4;
950 		memset((caddr_t)key, 0, sizeof (key));
951 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
952 			&auth_len, verf_str, verf_len)) {
953 			error = nfs_getauth(nmp, rep, cred, &auth_str,
954 				&auth_len, verf_str, &verf_len, key);
955 			if (error) {
956 				free((caddr_t)rep, M_NFSREQ);
957 				m_freem(mrest);
958 				return (error);
959 			}
960 		}
961 	} else {
962 		auth_type = RPCAUTH_UNIX;
963 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
964 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
965 			5 * NFSX_UNSIGNED;
966 	}
967 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
968 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
969 	if (auth_str)
970 		free(auth_str, M_TEMP);
971 
972 	/*
973 	 * For stream protocols, insert a Sun RPC Record Mark.
974 	 */
975 	if (nmp->nm_sotype == SOCK_STREAM) {
976 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
977 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
978 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
979 	}
980 	rep->r_mreq = m;
981 	rep->r_xid = xid;
982 tryagain:
983 	if (nmp->nm_flag & NFSMNT_SOFT)
984 		rep->r_retry = nmp->nm_retry;
985 	else
986 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
987 	rep->r_rtt = rep->r_rexmit = 0;
988 	if (proct[procnum] > 0)
989 		rep->r_flags = R_TIMING;
990 	else
991 		rep->r_flags = 0;
992 	rep->r_mrep = NULL;
993 
994 	/*
995 	 * Do the client side RPC.
996 	 */
997 	nfsstats.rpcrequests++;
998 	/*
999 	 * Chain request into list of outstanding requests. Be sure
1000 	 * to put it LAST so timer finds oldest requests first.
1001 	 */
1002 	s = splsoftnet();
1003 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1004 
1005 	/* Get send time for nqnfs */
1006 	reqtime = time.tv_sec;
1007 
1008 	/*
1009 	 * If backing off another request or avoiding congestion, don't
1010 	 * send this one now but let timer do it. If not timing a request,
1011 	 * do it now.
1012 	 */
1013 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1014 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1015 		nmp->nm_sent < nmp->nm_cwnd)) {
1016 		splx(s);
1017 		if (nmp->nm_soflags & PR_CONNREQUIRED)
1018 			error = nfs_sndlock(&nmp->nm_iflag, rep);
1019 		if (!error) {
1020 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
1021 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep);
1022 			if (nmp->nm_soflags & PR_CONNREQUIRED)
1023 				nfs_sndunlock(&nmp->nm_iflag);
1024 		}
1025 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1026 			nmp->nm_sent += NFS_CWNDSCALE;
1027 			rep->r_flags |= R_SENT;
1028 		}
1029 	} else {
1030 		splx(s);
1031 		rep->r_rtt = -1;
1032 	}
1033 
1034 	/*
1035 	 * Wait for the reply from our send or the timer's.
1036 	 */
1037 	if (!error || error == EPIPE)
1038 		error = nfs_reply(rep);
1039 
1040 	/*
1041 	 * RPC done, unlink the request.
1042 	 */
1043 	s = splsoftnet();
1044 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1045 	splx(s);
1046 
1047 	/*
1048 	 * Decrement the outstanding request count.
1049 	 */
1050 	if (rep->r_flags & R_SENT) {
1051 		rep->r_flags &= ~R_SENT;	/* paranoia */
1052 		nmp->nm_sent -= NFS_CWNDSCALE;
1053 	}
1054 
1055 	/*
1056 	 * If there was a successful reply and a tprintf msg.
1057 	 * tprintf a response.
1058 	 */
1059 	if (!error && (rep->r_flags & R_TPRINTFMSG))
1060 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1061 		    "is alive again");
1062 	mrep = rep->r_mrep;
1063 	md = rep->r_md;
1064 	dpos = rep->r_dpos;
1065 	if (error) {
1066 		m_freem(rep->r_mreq);
1067 		free((caddr_t)rep, M_NFSREQ);
1068 		return (error);
1069 	}
1070 
1071 	/*
1072 	 * break down the rpc header and check if ok
1073 	 */
1074 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1075 	if (*tl++ == rpc_msgdenied) {
1076 		if (*tl == rpc_mismatch)
1077 			error = EOPNOTSUPP;
1078 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1079 			if (!failed_auth) {
1080 				failed_auth++;
1081 				mheadend->m_next = (struct mbuf *)0;
1082 				m_freem(mrep);
1083 				m_freem(rep->r_mreq);
1084 				goto kerbauth;
1085 			} else
1086 				error = EAUTH;
1087 		} else
1088 			error = EACCES;
1089 		m_freem(mrep);
1090 		m_freem(rep->r_mreq);
1091 		free((caddr_t)rep, M_NFSREQ);
1092 		return (error);
1093 	}
1094 
1095 	/*
1096 	 * Grab any Kerberos verifier, otherwise just throw it away.
1097 	 */
1098 	verf_type = fxdr_unsigned(int, *tl++);
1099 	i = fxdr_unsigned(int32_t, *tl);
1100 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1101 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1102 		if (error)
1103 			goto nfsmout;
1104 	} else if (i > 0)
1105 		nfsm_adv(nfsm_rndup(i));
1106 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1107 	/* 0 == ok */
1108 	if (*tl == 0) {
1109 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1110 		if (*tl != 0) {
1111 			error = fxdr_unsigned(int, *tl);
1112 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1113 				error == NFSERR_TRYLATER) {
1114 				m_freem(mrep);
1115 				error = 0;
1116 				waituntil = time.tv_sec + trylater_delay;
1117 				while (time.tv_sec < waituntil)
1118 					(void) tsleep((caddr_t)&lbolt,
1119 						PSOCK, "nqnfstry", 0);
1120 				trylater_delay *= nfs_backoff[trylater_cnt];
1121 				if (trylater_cnt < 7)
1122 					trylater_cnt++;
1123 				goto tryagain;
1124 			}
1125 
1126 			/*
1127 			 * If the File Handle was stale, invalidate the
1128 			 * lookup cache, just in case.
1129 			 */
1130 			if (error == ESTALE)
1131 				cache_purge(vp);
1132 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1133 				*mrp = mrep;
1134 				*mdp = md;
1135 				*dposp = dpos;
1136 				error |= NFSERR_RETERR;
1137 			} else
1138 				m_freem(mrep);
1139 			m_freem(rep->r_mreq);
1140 			free((caddr_t)rep, M_NFSREQ);
1141 			return (error);
1142 		}
1143 
1144 #ifndef NFS_V2_ONLY
1145 		/*
1146 		 * For nqnfs, get any lease in reply
1147 		 */
1148 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1149 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1150 			if (*tl) {
1151 				np = VTONFS(vp);
1152 				nqlflag = fxdr_unsigned(int, *tl);
1153 				nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1154 				cachable = fxdr_unsigned(int, *tl++);
1155 				reqtime += fxdr_unsigned(int, *tl++);
1156 				if (reqtime > time.tv_sec) {
1157 				    frev = fxdr_hyper(tl);
1158 				    nqnfs_clientlease(nmp, np, nqlflag,
1159 					cachable, reqtime, frev);
1160 				}
1161 			}
1162 		}
1163 #endif
1164 		*mrp = mrep;
1165 		*mdp = md;
1166 		*dposp = dpos;
1167 		m_freem(rep->r_mreq);
1168 		FREE((caddr_t)rep, M_NFSREQ);
1169 		return (0);
1170 	}
1171 	m_freem(mrep);
1172 	error = EPROTONOSUPPORT;
1173 nfsmout:
1174 	m_freem(rep->r_mreq);
1175 	free((caddr_t)rep, M_NFSREQ);
1176 	return (error);
1177 }
1178 #endif /* NFS */
1179 
1180 /*
1181  * Generate the rpc reply header
1182  * siz arg. is used to decide if adding a cluster is worthwhile
1183  */
1184 int
1185 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1186 	int siz;
1187 	struct nfsrv_descript *nd;
1188 	struct nfssvc_sock *slp;
1189 	int err;
1190 	int cache;
1191 	u_quad_t *frev;
1192 	struct mbuf **mrq;
1193 	struct mbuf **mbp;
1194 	caddr_t *bposp;
1195 {
1196 	u_int32_t *tl;
1197 	struct mbuf *mreq;
1198 	caddr_t bpos;
1199 	struct mbuf *mb, *mb2;
1200 
1201 	MGETHDR(mreq, M_WAIT, MT_DATA);
1202 	mb = mreq;
1203 	/*
1204 	 * If this is a big reply, use a cluster else
1205 	 * try and leave leading space for the lower level headers.
1206 	 */
1207 	siz += RPC_REPLYSIZ;
1208 	if (siz >= max_datalen) {
1209 		MCLGET(mreq, M_WAIT);
1210 	} else
1211 		mreq->m_data += max_hdr;
1212 	tl = mtod(mreq, u_int32_t *);
1213 	mreq->m_len = 6 * NFSX_UNSIGNED;
1214 	bpos = ((caddr_t)tl) + mreq->m_len;
1215 	*tl++ = txdr_unsigned(nd->nd_retxid);
1216 	*tl++ = rpc_reply;
1217 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1218 		*tl++ = rpc_msgdenied;
1219 		if (err & NFSERR_AUTHERR) {
1220 			*tl++ = rpc_autherr;
1221 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1222 			mreq->m_len -= NFSX_UNSIGNED;
1223 			bpos -= NFSX_UNSIGNED;
1224 		} else {
1225 			*tl++ = rpc_mismatch;
1226 			*tl++ = txdr_unsigned(RPC_VER2);
1227 			*tl = txdr_unsigned(RPC_VER2);
1228 		}
1229 	} else {
1230 		*tl++ = rpc_msgaccepted;
1231 
1232 		/*
1233 		 * For Kerberos authentication, we must send the nickname
1234 		 * verifier back, otherwise just RPCAUTH_NULL.
1235 		 */
1236 		if (nd->nd_flag & ND_KERBFULL) {
1237 		    struct nfsuid *nuidp;
1238 		    struct timeval ktvin, ktvout;
1239 
1240 		    for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1241 			nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1242 			if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1243 			    (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1244 			     &nuidp->nu_haddr, nd->nd_nam2)))
1245 			    break;
1246 		    }
1247 		    if (nuidp) {
1248 			ktvin.tv_sec =
1249 			    txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1250 			ktvin.tv_usec =
1251 			    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1252 
1253 			/*
1254 			 * Encrypt the timestamp in ecb mode using the
1255 			 * session key.
1256 			 */
1257 #ifdef NFSKERB
1258 			XXX
1259 #endif
1260 
1261 			*tl++ = rpc_auth_kerb;
1262 			*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1263 			*tl = ktvout.tv_sec;
1264 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1265 			*tl++ = ktvout.tv_usec;
1266 			*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1267 		    } else {
1268 			*tl++ = 0;
1269 			*tl++ = 0;
1270 		    }
1271 		} else {
1272 			*tl++ = 0;
1273 			*tl++ = 0;
1274 		}
1275 		switch (err) {
1276 		case EPROGUNAVAIL:
1277 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1278 			break;
1279 		case EPROGMISMATCH:
1280 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1281 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1282 			if (nd->nd_flag & ND_NQNFS) {
1283 				*tl++ = txdr_unsigned(3);
1284 				*tl = txdr_unsigned(3);
1285 			} else {
1286 				*tl++ = txdr_unsigned(2);
1287 				*tl = txdr_unsigned(3);
1288 			}
1289 			break;
1290 		case EPROCUNAVAIL:
1291 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1292 			break;
1293 		case EBADRPC:
1294 			*tl = txdr_unsigned(RPC_GARBAGE);
1295 			break;
1296 		default:
1297 			*tl = 0;
1298 			if (err != NFSERR_RETVOID) {
1299 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1300 				if (err)
1301 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1302 				else
1303 				    *tl = 0;
1304 			}
1305 			break;
1306 		};
1307 	}
1308 
1309 	/*
1310 	 * For nqnfs, piggyback lease as requested.
1311 	 */
1312 	if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1313 		if (nd->nd_flag & ND_LEASE) {
1314 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1315 			*tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1316 			*tl++ = txdr_unsigned(cache);
1317 			*tl++ = txdr_unsigned(nd->nd_duration);
1318 			txdr_hyper(*frev, tl);
1319 		} else {
1320 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1321 			*tl = 0;
1322 		}
1323 	}
1324 	if (mrq != NULL)
1325 		*mrq = mreq;
1326 	*mbp = mb;
1327 	*bposp = bpos;
1328 	if (err != 0 && err != NFSERR_RETVOID)
1329 		nfsstats.srvrpc_errs++;
1330 	return (0);
1331 }
1332 
1333 /*
1334  * Nfs timer routine
1335  * Scan the nfsreq list and retranmit any requests that have timed out
1336  * To avoid retransmission attempts on STREAM sockets (in the future) make
1337  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1338  */
1339 void
1340 nfs_timer(arg)
1341 	void *arg;	/* never used */
1342 {
1343 	struct nfsreq *rep;
1344 	struct mbuf *m;
1345 	struct socket *so;
1346 	struct nfsmount *nmp;
1347 	int timeo;
1348 	int s, error;
1349 #ifdef NFSSERVER
1350 	struct nfssvc_sock *slp;
1351 	static long lasttime = 0;
1352 	u_quad_t cur_usec;
1353 #endif
1354 
1355 	s = splsoftnet();
1356 	for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) {
1357 		nmp = rep->r_nmp;
1358 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1359 			continue;
1360 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1361 			rep->r_flags |= R_SOFTTERM;
1362 			continue;
1363 		}
1364 		if (rep->r_rtt >= 0) {
1365 			rep->r_rtt++;
1366 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1367 				timeo = nmp->nm_timeo;
1368 			else
1369 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1370 			if (nmp->nm_timeouts > 0)
1371 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1372 			if (rep->r_rtt <= timeo)
1373 				continue;
1374 			if (nmp->nm_timeouts < 8)
1375 				nmp->nm_timeouts++;
1376 		}
1377 		/*
1378 		 * Check for server not responding
1379 		 */
1380 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1381 		     rep->r_rexmit > nmp->nm_deadthresh) {
1382 			nfs_msg(rep->r_procp,
1383 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1384 			    "not responding");
1385 			rep->r_flags |= R_TPRINTFMSG;
1386 		}
1387 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1388 			nfsstats.rpctimeouts++;
1389 			rep->r_flags |= R_SOFTTERM;
1390 			continue;
1391 		}
1392 		if (nmp->nm_sotype != SOCK_DGRAM) {
1393 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1394 				rep->r_rexmit = NFS_MAXREXMIT;
1395 			continue;
1396 		}
1397 		if ((so = nmp->nm_so) == NULL)
1398 			continue;
1399 
1400 		/*
1401 		 * If there is enough space and the window allows..
1402 		 *	Resend it
1403 		 * Set r_rtt to -1 in case we fail to send it now.
1404 		 */
1405 		rep->r_rtt = -1;
1406 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1407 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1408 		    (rep->r_flags & R_SENT) ||
1409 		    nmp->nm_sent < nmp->nm_cwnd) &&
1410 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1411 		        if (so->so_state & SS_ISCONNECTED)
1412 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1413 			    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1414 			else
1415 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1416 			    nmp->nm_nam, (struct mbuf *)0, (struct proc *)0);
1417 			if (error) {
1418 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1419 #ifdef DEBUG
1420 					printf("nfs_timer: ignoring error %d\n",
1421 						error);
1422 #endif
1423 					so->so_error = 0;
1424 				}
1425 			} else {
1426 				/*
1427 				 * Iff first send, start timing
1428 				 * else turn timing off, backoff timer
1429 				 * and divide congestion window by 2.
1430 				 */
1431 				if (rep->r_flags & R_SENT) {
1432 					rep->r_flags &= ~R_TIMING;
1433 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1434 						rep->r_rexmit = NFS_MAXREXMIT;
1435 					nmp->nm_cwnd >>= 1;
1436 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1437 						nmp->nm_cwnd = NFS_CWNDSCALE;
1438 					nfsstats.rpcretries++;
1439 				} else {
1440 					rep->r_flags |= R_SENT;
1441 					nmp->nm_sent += NFS_CWNDSCALE;
1442 				}
1443 				rep->r_rtt = 0;
1444 			}
1445 		}
1446 	}
1447 
1448 #ifdef NFSSERVER
1449 	/*
1450 	 * Call the nqnfs server timer once a second to handle leases.
1451 	 */
1452 	if (lasttime != time.tv_sec) {
1453 		lasttime = time.tv_sec;
1454 		nqnfs_serverd();
1455 	}
1456 
1457 	/*
1458 	 * Scan the write gathering queues for writes that need to be
1459 	 * completed now.
1460 	 */
1461 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1462 	for (slp = nfssvc_sockhead.tqh_first; slp != 0;
1463 	    slp = slp->ns_chain.tqe_next) {
1464 	    if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1465 		nfsrv_wakenfsd(slp);
1466 	}
1467 #endif /* NFSSERVER */
1468 	splx(s);
1469 	callout_reset(&nfs_timer_ch, nfs_ticks, nfs_timer, NULL);
1470 }
1471 
1472 /*
1473  * Test for a termination condition pending on the process.
1474  * This is used for NFSMNT_INT mounts.
1475  */
1476 int
1477 nfs_sigintr(nmp, rep, p)
1478 	struct nfsmount *nmp;
1479 	struct nfsreq *rep;
1480 	struct proc *p;
1481 {
1482 	sigset_t ss;
1483 
1484 	if (rep && (rep->r_flags & R_SOFTTERM))
1485 		return (EINTR);
1486 	if (!(nmp->nm_flag & NFSMNT_INT))
1487 		return (0);
1488 	if (p) {
1489 		sigpending1(p, &ss);
1490 #if 0
1491 		sigminusset(&p->p_sigctx.ps_sigignore, &ss);
1492 #endif
1493 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1494 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1495 		    sigismember(&ss, SIGQUIT))
1496 			return (EINTR);
1497 	}
1498 	return (0);
1499 }
1500 
1501 /*
1502  * Lock a socket against others.
1503  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1504  * and also to avoid race conditions between the processes with nfs requests
1505  * in progress when a reconnect is necessary.
1506  */
1507 int
1508 nfs_sndlock(flagp, rep)
1509 	int *flagp;
1510 	struct nfsreq *rep;
1511 {
1512 	struct proc *p;
1513 	int slpflag = 0, slptimeo = 0;
1514 
1515 	if (rep) {
1516 		p = rep->r_procp;
1517 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1518 			slpflag = PCATCH;
1519 	} else
1520 		p = (struct proc *)0;
1521 	while (*flagp & NFSMNT_SNDLOCK) {
1522 		if (nfs_sigintr(rep->r_nmp, rep, p))
1523 			return (EINTR);
1524 		*flagp |= NFSMNT_WANTSND;
1525 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1526 			slptimeo);
1527 		if (slpflag == PCATCH) {
1528 			slpflag = 0;
1529 			slptimeo = 2 * hz;
1530 		}
1531 	}
1532 	*flagp |= NFSMNT_SNDLOCK;
1533 	return (0);
1534 }
1535 
1536 /*
1537  * Unlock the stream socket for others.
1538  */
1539 void
1540 nfs_sndunlock(flagp)
1541 	int *flagp;
1542 {
1543 
1544 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1545 		panic("nfs sndunlock");
1546 	*flagp &= ~NFSMNT_SNDLOCK;
1547 	if (*flagp & NFSMNT_WANTSND) {
1548 		*flagp &= ~NFSMNT_WANTSND;
1549 		wakeup((caddr_t)flagp);
1550 	}
1551 }
1552 
1553 int
1554 nfs_rcvlock(rep)
1555 	struct nfsreq *rep;
1556 {
1557 	struct nfsmount *nmp = rep->r_nmp;
1558 	int *flagp = &nmp->nm_iflag;
1559 	int slpflag, slptimeo = 0;
1560 
1561 	if (*flagp & NFSMNT_DISMNT)
1562 		return EIO;
1563 
1564 	if (*flagp & NFSMNT_INT)
1565 		slpflag = PCATCH;
1566 	else
1567 		slpflag = 0;
1568 	while (*flagp & NFSMNT_RCVLOCK) {
1569 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1570 			return (EINTR);
1571 		*flagp |= NFSMNT_WANTRCV;
1572 		nmp->nm_waiters++;
1573 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1574 			slptimeo);
1575 		nmp->nm_waiters--;
1576 		if (*flagp & NFSMNT_DISMNT) {
1577 			wakeup(&nmp->nm_waiters);
1578 			return EIO;
1579 		}
1580 		/* If our reply was received while we were sleeping,
1581 		 * then just return without taking the lock to avoid a
1582 		 * situation where a single iod could 'capture' the
1583 		 * receive lock.
1584 		 */
1585 		if (rep->r_mrep != NULL)
1586 			return (EALREADY);
1587 		if (slpflag == PCATCH) {
1588 			slpflag = 0;
1589 			slptimeo = 2 * hz;
1590 		}
1591 	}
1592 	*flagp |= NFSMNT_RCVLOCK;
1593 	return (0);
1594 }
1595 
1596 /*
1597  * Unlock the stream socket for others.
1598  */
1599 void
1600 nfs_rcvunlock(flagp)
1601 	int *flagp;
1602 {
1603 
1604 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1605 		panic("nfs rcvunlock");
1606 	*flagp &= ~NFSMNT_RCVLOCK;
1607 	if (*flagp & NFSMNT_WANTRCV) {
1608 		*flagp &= ~NFSMNT_WANTRCV;
1609 		wakeup((caddr_t)flagp);
1610 	}
1611 }
1612 
1613 /*
1614  * Parse an RPC request
1615  * - verify it
1616  * - fill in the cred struct.
1617  */
1618 int
1619 nfs_getreq(nd, nfsd, has_header)
1620 	struct nfsrv_descript *nd;
1621 	struct nfsd *nfsd;
1622 	int has_header;
1623 {
1624 	int len, i;
1625 	u_int32_t *tl;
1626 	int32_t t1;
1627 	struct uio uio;
1628 	struct iovec iov;
1629 	caddr_t dpos, cp2, cp;
1630 	u_int32_t nfsvers, auth_type;
1631 	uid_t nickuid;
1632 	int error = 0, nqnfs = 0, ticklen;
1633 	struct mbuf *mrep, *md;
1634 	struct nfsuid *nuidp;
1635 	struct timeval tvin, tvout;
1636 
1637 	mrep = nd->nd_mrep;
1638 	md = nd->nd_md;
1639 	dpos = nd->nd_dpos;
1640 	if (has_header) {
1641 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1642 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1643 		if (*tl++ != rpc_call) {
1644 			m_freem(mrep);
1645 			return (EBADRPC);
1646 		}
1647 	} else
1648 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1649 	nd->nd_repstat = 0;
1650 	nd->nd_flag = 0;
1651 	if (*tl++ != rpc_vers) {
1652 		nd->nd_repstat = ERPCMISMATCH;
1653 		nd->nd_procnum = NFSPROC_NOOP;
1654 		return (0);
1655 	}
1656 	if (*tl != nfs_prog) {
1657 		if (*tl == nqnfs_prog)
1658 			nqnfs++;
1659 		else {
1660 			nd->nd_repstat = EPROGUNAVAIL;
1661 			nd->nd_procnum = NFSPROC_NOOP;
1662 			return (0);
1663 		}
1664 	}
1665 	tl++;
1666 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1667 	if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1668 		(nfsvers != NQNFS_VER3 && nqnfs)) {
1669 		nd->nd_repstat = EPROGMISMATCH;
1670 		nd->nd_procnum = NFSPROC_NOOP;
1671 		return (0);
1672 	}
1673 	if (nqnfs)
1674 		nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1675 	else if (nfsvers == NFS_VER3)
1676 		nd->nd_flag = ND_NFSV3;
1677 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1678 	if (nd->nd_procnum == NFSPROC_NULL)
1679 		return (0);
1680 	if (nd->nd_procnum >= NFS_NPROCS ||
1681 		(!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1682 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1683 		nd->nd_repstat = EPROCUNAVAIL;
1684 		nd->nd_procnum = NFSPROC_NOOP;
1685 		return (0);
1686 	}
1687 	if ((nd->nd_flag & ND_NFSV3) == 0)
1688 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1689 	auth_type = *tl++;
1690 	len = fxdr_unsigned(int, *tl++);
1691 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1692 		m_freem(mrep);
1693 		return (EBADRPC);
1694 	}
1695 
1696 	nd->nd_flag &= ~ND_KERBAUTH;
1697 	/*
1698 	 * Handle auth_unix or auth_kerb.
1699 	 */
1700 	if (auth_type == rpc_auth_unix) {
1701 		len = fxdr_unsigned(int, *++tl);
1702 		if (len < 0 || len > NFS_MAXNAMLEN) {
1703 			m_freem(mrep);
1704 			return (EBADRPC);
1705 		}
1706 		nfsm_adv(nfsm_rndup(len));
1707 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1708 		memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred));
1709 		nd->nd_cr.cr_ref = 1;
1710 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1711 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1712 		len = fxdr_unsigned(int, *tl);
1713 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1714 			m_freem(mrep);
1715 			return (EBADRPC);
1716 		}
1717 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1718 		for (i = 0; i < len; i++)
1719 		    if (i < NGROUPS)
1720 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1721 		    else
1722 			tl++;
1723 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1724 		if (nd->nd_cr.cr_ngroups > 1)
1725 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1726 		len = fxdr_unsigned(int, *++tl);
1727 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1728 			m_freem(mrep);
1729 			return (EBADRPC);
1730 		}
1731 		if (len > 0)
1732 			nfsm_adv(nfsm_rndup(len));
1733 	} else if (auth_type == rpc_auth_kerb) {
1734 		switch (fxdr_unsigned(int, *tl++)) {
1735 		case RPCAKN_FULLNAME:
1736 			ticklen = fxdr_unsigned(int, *tl);
1737 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1738 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1739 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1740 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1741 				m_freem(mrep);
1742 				return (EBADRPC);
1743 			}
1744 			uio.uio_offset = 0;
1745 			uio.uio_iov = &iov;
1746 			uio.uio_iovcnt = 1;
1747 			uio.uio_segflg = UIO_SYSSPACE;
1748 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1749 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1750 			nfsm_mtouio(&uio, uio.uio_resid);
1751 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1752 			if (*tl++ != rpc_auth_kerb ||
1753 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1754 				printf("Bad kerb verifier\n");
1755 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1756 				nd->nd_procnum = NFSPROC_NOOP;
1757 				return (0);
1758 			}
1759 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1760 			tl = (u_int32_t *)cp;
1761 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1762 				printf("Not fullname kerb verifier\n");
1763 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1764 				nd->nd_procnum = NFSPROC_NOOP;
1765 				return (0);
1766 			}
1767 			cp += NFSX_UNSIGNED;
1768 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
1769 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1770 			nd->nd_flag |= ND_KERBFULL;
1771 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1772 			break;
1773 		case RPCAKN_NICKNAME:
1774 			if (len != 2 * NFSX_UNSIGNED) {
1775 				printf("Kerb nickname short\n");
1776 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1777 				nd->nd_procnum = NFSPROC_NOOP;
1778 				return (0);
1779 			}
1780 			nickuid = fxdr_unsigned(uid_t, *tl);
1781 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1782 			if (*tl++ != rpc_auth_kerb ||
1783 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1784 				printf("Kerb nick verifier bad\n");
1785 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1786 				nd->nd_procnum = NFSPROC_NOOP;
1787 				return (0);
1788 			}
1789 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1790 			tvin.tv_sec = *tl++;
1791 			tvin.tv_usec = *tl;
1792 
1793 			for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1794 			    nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1795 				if (nuidp->nu_cr.cr_uid == nickuid &&
1796 				    (!nd->nd_nam2 ||
1797 				     netaddr_match(NU_NETFAM(nuidp),
1798 				      &nuidp->nu_haddr, nd->nd_nam2)))
1799 					break;
1800 			}
1801 			if (!nuidp) {
1802 				nd->nd_repstat =
1803 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1804 				nd->nd_procnum = NFSPROC_NOOP;
1805 				return (0);
1806 			}
1807 
1808 			/*
1809 			 * Now, decrypt the timestamp using the session key
1810 			 * and validate it.
1811 			 */
1812 #ifdef NFSKERB
1813 			XXX
1814 #endif
1815 
1816 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1817 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1818 			if (nuidp->nu_expire < time.tv_sec ||
1819 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1820 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1821 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1822 				nuidp->nu_expire = 0;
1823 				nd->nd_repstat =
1824 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1825 				nd->nd_procnum = NFSPROC_NOOP;
1826 				return (0);
1827 			}
1828 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1829 			nd->nd_flag |= ND_KERBNICK;
1830 		};
1831 	} else {
1832 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1833 		nd->nd_procnum = NFSPROC_NOOP;
1834 		return (0);
1835 	}
1836 
1837 	/*
1838 	 * For nqnfs, get piggybacked lease request.
1839 	 */
1840 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
1841 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1842 		nd->nd_flag |= fxdr_unsigned(int, *tl);
1843 		if (nd->nd_flag & ND_LEASE) {
1844 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1845 			nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
1846 		} else
1847 			nd->nd_duration = NQ_MINLEASE;
1848 	} else
1849 		nd->nd_duration = NQ_MINLEASE;
1850 	nd->nd_md = md;
1851 	nd->nd_dpos = dpos;
1852 	return (0);
1853 nfsmout:
1854 	return (error);
1855 }
1856 
1857 int
1858 nfs_msg(p, server, msg)
1859 	struct proc *p;
1860 	char *server, *msg;
1861 {
1862 	tpr_t tpr;
1863 
1864 	if (p)
1865 		tpr = tprintf_open(p);
1866 	else
1867 		tpr = NULL;
1868 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1869 	tprintf_close(tpr);
1870 	return (0);
1871 }
1872 
1873 #ifdef NFSSERVER
1874 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
1875 				    struct nfssvc_sock *, struct proc *,
1876 				    struct mbuf **)) = {
1877 	nfsrv_null,
1878 	nfsrv_getattr,
1879 	nfsrv_setattr,
1880 	nfsrv_lookup,
1881 	nfsrv3_access,
1882 	nfsrv_readlink,
1883 	nfsrv_read,
1884 	nfsrv_write,
1885 	nfsrv_create,
1886 	nfsrv_mkdir,
1887 	nfsrv_symlink,
1888 	nfsrv_mknod,
1889 	nfsrv_remove,
1890 	nfsrv_rmdir,
1891 	nfsrv_rename,
1892 	nfsrv_link,
1893 	nfsrv_readdir,
1894 	nfsrv_readdirplus,
1895 	nfsrv_statfs,
1896 	nfsrv_fsinfo,
1897 	nfsrv_pathconf,
1898 	nfsrv_commit,
1899 	nqnfsrv_getlease,
1900 	nqnfsrv_vacated,
1901 	nfsrv_noop,
1902 	nfsrv_noop
1903 };
1904 
1905 /*
1906  * Socket upcall routine for the nfsd sockets.
1907  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1908  * Essentially do as much as possible non-blocking, else punt and it will
1909  * be called with M_WAIT from an nfsd.
1910  */
1911 void
1912 nfsrv_rcv(so, arg, waitflag)
1913 	struct socket *so;
1914 	caddr_t arg;
1915 	int waitflag;
1916 {
1917 	struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1918 	struct mbuf *m;
1919 	struct mbuf *mp, *nam;
1920 	struct uio auio;
1921 	int flags, error;
1922 
1923 	if ((slp->ns_flag & SLP_VALID) == 0)
1924 		return;
1925 #ifdef notdef
1926 	/*
1927 	 * Define this to test for nfsds handling this under heavy load.
1928 	 */
1929 	if (waitflag == M_DONTWAIT) {
1930 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1931 	}
1932 #endif
1933 	auio.uio_procp = NULL;
1934 	if (so->so_type == SOCK_STREAM) {
1935 		/*
1936 		 * If there are already records on the queue, defer soreceive()
1937 		 * to an nfsd so that there is feedback to the TCP layer that
1938 		 * the nfs servers are heavily loaded.
1939 		 */
1940 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1941 			slp->ns_flag |= SLP_NEEDQ;
1942 			goto dorecs;
1943 		}
1944 
1945 		/*
1946 		 * Do soreceive().
1947 		 */
1948 		auio.uio_resid = 1000000000;
1949 		flags = MSG_DONTWAIT;
1950 		error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1951 		if (error || mp == (struct mbuf *)0) {
1952 			if (error == EWOULDBLOCK)
1953 				slp->ns_flag |= SLP_NEEDQ;
1954 			else
1955 				slp->ns_flag |= SLP_DISCONN;
1956 			goto dorecs;
1957 		}
1958 		m = mp;
1959 		if (slp->ns_rawend) {
1960 			slp->ns_rawend->m_next = m;
1961 			slp->ns_cc += 1000000000 - auio.uio_resid;
1962 		} else {
1963 			slp->ns_raw = m;
1964 			slp->ns_cc = 1000000000 - auio.uio_resid;
1965 		}
1966 		while (m->m_next)
1967 			m = m->m_next;
1968 		slp->ns_rawend = m;
1969 
1970 		/*
1971 		 * Now try and parse record(s) out of the raw stream data.
1972 		 */
1973 		error = nfsrv_getstream(slp, waitflag);
1974 		if (error) {
1975 			if (error == EPERM)
1976 				slp->ns_flag |= SLP_DISCONN;
1977 			else
1978 				slp->ns_flag |= SLP_NEEDQ;
1979 		}
1980 	} else {
1981 		do {
1982 			auio.uio_resid = 1000000000;
1983 			flags = MSG_DONTWAIT;
1984 			error = (*so->so_receive)(so, &nam, &auio, &mp,
1985 						(struct mbuf **)0, &flags);
1986 			if (mp) {
1987 				if (nam) {
1988 					m = nam;
1989 					m->m_next = mp;
1990 				} else
1991 					m = mp;
1992 				if (slp->ns_recend)
1993 					slp->ns_recend->m_nextpkt = m;
1994 				else
1995 					slp->ns_rec = m;
1996 				slp->ns_recend = m;
1997 				m->m_nextpkt = (struct mbuf *)0;
1998 			}
1999 			if (error) {
2000 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2001 					&& error != EWOULDBLOCK) {
2002 					slp->ns_flag |= SLP_DISCONN;
2003 					goto dorecs;
2004 				}
2005 			}
2006 		} while (mp);
2007 	}
2008 
2009 	/*
2010 	 * Now try and process the request records, non-blocking.
2011 	 */
2012 dorecs:
2013 	if (waitflag == M_DONTWAIT &&
2014 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
2015 		nfsrv_wakenfsd(slp);
2016 }
2017 
2018 /*
2019  * Try and extract an RPC request from the mbuf data list received on a
2020  * stream socket. The "waitflag" argument indicates whether or not it
2021  * can sleep.
2022  */
2023 int
2024 nfsrv_getstream(slp, waitflag)
2025 	struct nfssvc_sock *slp;
2026 	int waitflag;
2027 {
2028 	struct mbuf *m, **mpp;
2029 	char *cp1, *cp2;
2030 	int len;
2031 	struct mbuf *om, *m2, *recm = NULL;
2032 	u_int32_t recmark;
2033 
2034 	if (slp->ns_flag & SLP_GETSTREAM)
2035 		panic("nfs getstream");
2036 	slp->ns_flag |= SLP_GETSTREAM;
2037 	for (;;) {
2038 	    if (slp->ns_reclen == 0) {
2039 		if (slp->ns_cc < NFSX_UNSIGNED) {
2040 			slp->ns_flag &= ~SLP_GETSTREAM;
2041 			return (0);
2042 		}
2043 		m = slp->ns_raw;
2044 		if (m->m_len >= NFSX_UNSIGNED) {
2045 			memcpy((caddr_t)&recmark, mtod(m, caddr_t), NFSX_UNSIGNED);
2046 			m->m_data += NFSX_UNSIGNED;
2047 			m->m_len -= NFSX_UNSIGNED;
2048 		} else {
2049 			cp1 = (caddr_t)&recmark;
2050 			cp2 = mtod(m, caddr_t);
2051 			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
2052 				while (m->m_len == 0) {
2053 					m = m->m_next;
2054 					cp2 = mtod(m, caddr_t);
2055 				}
2056 				*cp1++ = *cp2++;
2057 				m->m_data++;
2058 				m->m_len--;
2059 			}
2060 		}
2061 		slp->ns_cc -= NFSX_UNSIGNED;
2062 		recmark = ntohl(recmark);
2063 		slp->ns_reclen = recmark & ~0x80000000;
2064 		if (recmark & 0x80000000)
2065 			slp->ns_flag |= SLP_LASTFRAG;
2066 		else
2067 			slp->ns_flag &= ~SLP_LASTFRAG;
2068 		if (slp->ns_reclen > NFS_MAXPACKET) {
2069 			slp->ns_flag &= ~SLP_GETSTREAM;
2070 			return (EPERM);
2071 		}
2072 	    }
2073 
2074 	    /*
2075 	     * Now get the record part.
2076 	     */
2077 	    if (slp->ns_cc == slp->ns_reclen) {
2078 		recm = slp->ns_raw;
2079 		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2080 		slp->ns_cc = slp->ns_reclen = 0;
2081 	    } else if (slp->ns_cc > slp->ns_reclen) {
2082 		len = 0;
2083 		m = slp->ns_raw;
2084 		om = (struct mbuf *)0;
2085 		while (len < slp->ns_reclen) {
2086 			if ((len + m->m_len) > slp->ns_reclen) {
2087 				size_t left = slp->ns_reclen - len;
2088 
2089 				MGETHDR(m2, waitflag, m->m_type);
2090 				if (m2 == NULL) {
2091 					slp->ns_flag &= ~SLP_GETSTREAM;
2092 					return (EWOULDBLOCK);
2093 				}
2094 				if (left > MHLEN) {
2095 					MCLGET(m2, waitflag);
2096 					if (!(m2->m_flags & M_EXT)) {
2097 						m_freem(m2);
2098 						slp->ns_flag &= ~SLP_GETSTREAM;
2099 						return (EWOULDBLOCK);
2100 					}
2101 				}
2102 				memcpy(mtod(m2, caddr_t), mtod(m, caddr_t),
2103 				    left);
2104 				m2->m_len = left;
2105 				m->m_data += left;
2106 				m->m_len -= left;
2107 				if (om) {
2108 					om->m_next = m2;
2109 					recm = slp->ns_raw;
2110 				} else
2111 					recm = m2;
2112 				len = slp->ns_reclen;
2113 			} else if ((len + m->m_len) == slp->ns_reclen) {
2114 				om = m;
2115 				len += m->m_len;
2116 				m = m->m_next;
2117 				recm = slp->ns_raw;
2118 				om->m_next = (struct mbuf *)0;
2119 			} else {
2120 				om = m;
2121 				len += m->m_len;
2122 				m = m->m_next;
2123 			}
2124 		}
2125 		slp->ns_raw = m;
2126 		slp->ns_cc -= len;
2127 		slp->ns_reclen = 0;
2128 	    } else {
2129 		slp->ns_flag &= ~SLP_GETSTREAM;
2130 		return (0);
2131 	    }
2132 
2133 	    /*
2134 	     * Accumulate the fragments into a record.
2135 	     */
2136 	    mpp = &slp->ns_frag;
2137 	    while (*mpp)
2138 		mpp = &((*mpp)->m_next);
2139 	    *mpp = recm;
2140 	    if (slp->ns_flag & SLP_LASTFRAG) {
2141 		if (slp->ns_recend)
2142 		    slp->ns_recend->m_nextpkt = slp->ns_frag;
2143 		else
2144 		    slp->ns_rec = slp->ns_frag;
2145 		slp->ns_recend = slp->ns_frag;
2146 		slp->ns_frag = (struct mbuf *)0;
2147 	    }
2148 	}
2149 }
2150 
2151 /*
2152  * Parse an RPC header.
2153  */
2154 int
2155 nfsrv_dorec(slp, nfsd, ndp)
2156 	struct nfssvc_sock *slp;
2157 	struct nfsd *nfsd;
2158 	struct nfsrv_descript **ndp;
2159 {
2160 	struct mbuf *m, *nam;
2161 	struct nfsrv_descript *nd;
2162 	int error;
2163 
2164 	*ndp = NULL;
2165 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2166 	    (m = slp->ns_rec) == (struct mbuf *)0)
2167 		return (ENOBUFS);
2168 	slp->ns_rec = m->m_nextpkt;
2169 	if (slp->ns_rec)
2170 		m->m_nextpkt = (struct mbuf *)0;
2171 	else
2172 		slp->ns_recend = (struct mbuf *)0;
2173 	if (m->m_type == MT_SONAME) {
2174 		nam = m;
2175 		m = m->m_next;
2176 		nam->m_next = NULL;
2177 	} else
2178 		nam = NULL;
2179 	MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2180 		M_NFSRVDESC, M_WAITOK);
2181 	nd->nd_md = nd->nd_mrep = m;
2182 	nd->nd_nam2 = nam;
2183 	nd->nd_dpos = mtod(m, caddr_t);
2184 	error = nfs_getreq(nd, nfsd, TRUE);
2185 	if (error) {
2186 		m_freem(nam);
2187 		free((caddr_t)nd, M_NFSRVDESC);
2188 		return (error);
2189 	}
2190 	*ndp = nd;
2191 	nfsd->nfsd_nd = nd;
2192 	return (0);
2193 }
2194 
2195 
2196 /*
2197  * Search for a sleeping nfsd and wake it up.
2198  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2199  * running nfsds will go look for the work in the nfssvc_sock list.
2200  */
2201 void
2202 nfsrv_wakenfsd(slp)
2203 	struct nfssvc_sock *slp;
2204 {
2205 	struct nfsd *nd;
2206 
2207 	if ((slp->ns_flag & SLP_VALID) == 0)
2208 		return;
2209 	for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) {
2210 		if (nd->nfsd_flag & NFSD_WAITING) {
2211 			nd->nfsd_flag &= ~NFSD_WAITING;
2212 			if (nd->nfsd_slp)
2213 				panic("nfsd wakeup");
2214 			slp->ns_sref++;
2215 			nd->nfsd_slp = slp;
2216 			wakeup((caddr_t)nd);
2217 			return;
2218 		}
2219 	}
2220 	slp->ns_flag |= SLP_DOREC;
2221 	nfsd_head_flag |= NFSD_CHECKSLP;
2222 }
2223 #endif /* NFSSERVER */
2224