xref: /netbsd-src/sys/nfs/nfs_socket.c (revision d20841bb642898112fe68f0ad3f7b26dddf56f07)
1 /*	$NetBSD: nfs_socket.c,v 1.100 2003/12/07 21:15:46 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. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_socket.c	8.5 (Berkeley) 3/30/95
35  */
36 
37 /*
38  * Socket operations for use by nfs
39  */
40 
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: nfs_socket.c,v 1.100 2003/12/07 21:15:46 fvdl Exp $");
43 
44 #include "fs_nfs.h"
45 #include "opt_nfs.h"
46 #include "opt_nfsserver.h"
47 #include "opt_mbuftrace.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 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
83 #ifdef MBUFTRACE
84 struct mowner nfs_mowner = { "nfs" };
85 #endif
86 
87 /*
88  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
89  * Use the mean and mean deviation of rtt for the appropriate type of rpc
90  * for the frequent rpcs and a default for the others.
91  * The justification for doing "other" this way is that these rpcs
92  * happen so infrequently that timer est. would probably be stale.
93  * Also, since many of these rpcs are
94  * non-idempotent, a conservative timeout is desired.
95  * getattr, lookup - A+2D
96  * read, write     - A+4D
97  * other           - nm_timeo
98  */
99 #define	NFS_RTO(n, t) \
100 	((t) == 0 ? (n)->nm_timeo : \
101 	 ((t) < 3 ? \
102 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
103 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
104 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
105 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
106 /*
107  * External data, mostly RPC constants in XDR form
108  */
109 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
110 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
111 	rpc_auth_kerb;
112 extern u_int32_t nfs_prog, nqnfs_prog;
113 extern time_t nqnfsstarttime;
114 extern const int nfsv3_procid[NFS_NPROCS];
115 extern int nfs_ticks;
116 
117 /*
118  * Defines which timer to use for the procnum.
119  * 0 - default
120  * 1 - getattr
121  * 2 - lookup
122  * 3 - read
123  * 4 - write
124  */
125 static const int proct[NFS_NPROCS] = {
126 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
127 	0, 0, 0,
128 };
129 
130 /*
131  * There is a congestion window for outstanding rpcs maintained per mount
132  * point. The cwnd size is adjusted in roughly the way that:
133  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
134  * SIGCOMM '88". ACM, August 1988.
135  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
136  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
137  * of rpcs is in progress.
138  * (The sent count and cwnd are scaled for integer arith.)
139  * Variants of "slow start" were tried and were found to be too much of a
140  * performance hit (ave. rtt 3 times larger),
141  * I suspect due to the large rtt that nfs rpcs have.
142  */
143 #define	NFS_CWNDSCALE	256
144 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
145 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
146 int nfsrtton = 0;
147 struct nfsrtt nfsrtt;
148 struct nfsreqhead nfs_reqq;
149 
150 struct callout nfs_timer_ch = CALLOUT_INITIALIZER_SETFUNC(nfs_timer, NULL);
151 
152 /*
153  * Initialize sockets and congestion for a new NFS connection.
154  * We do not free the sockaddr if error.
155  */
156 int
157 nfs_connect(nmp, rep)
158 	struct nfsmount *nmp;
159 	struct nfsreq *rep;
160 {
161 	struct socket *so;
162 	int s, error, rcvreserve, sndreserve;
163 	struct sockaddr *saddr;
164 	struct sockaddr_in *sin;
165 #ifdef INET6
166 	struct sockaddr_in6 *sin6;
167 #endif
168 	struct mbuf *m;
169 
170 	nmp->nm_so = (struct socket *)0;
171 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
172 	error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
173 		nmp->nm_soproto);
174 	if (error)
175 		goto bad;
176 	so = nmp->nm_so;
177 #ifdef MBUFTRACE
178 	so->so_mowner = &nfs_mowner;
179 	so->so_rcv.sb_mowner = &nfs_mowner;
180 	so->so_snd.sb_mowner = &nfs_mowner;
181 #endif
182 	nmp->nm_soflags = so->so_proto->pr_flags;
183 
184 	/*
185 	 * Some servers require that the client port be a reserved port number.
186 	 */
187 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
188 		m = m_get(M_WAIT, MT_SOOPTS);
189 		MCLAIM(m, so->so_mowner);
190 		*mtod(m, int32_t *) = IP_PORTRANGE_LOW;
191 		m->m_len = sizeof(int32_t);
192 		if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m)))
193 			goto bad;
194 		m = m_get(M_WAIT, MT_SONAME);
195 		MCLAIM(m, so->so_mowner);
196 		sin = mtod(m, struct sockaddr_in *);
197 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
198 		sin->sin_family = AF_INET;
199 		sin->sin_addr.s_addr = INADDR_ANY;
200 		sin->sin_port = 0;
201 		error = sobind(so, m, &proc0);
202 		m_freem(m);
203 		if (error)
204 			goto bad;
205 	}
206 #ifdef INET6
207 	if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) {
208 		m = m_get(M_WAIT, MT_SOOPTS);
209 		MCLAIM(m, so->so_mowner);
210 		*mtod(m, int32_t *) = IPV6_PORTRANGE_LOW;
211 		m->m_len = sizeof(int32_t);
212 		if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m)))
213 			goto bad;
214 		m = m_get(M_WAIT, MT_SONAME);
215 		MCLAIM(m, so->so_mowner);
216 		sin6 = mtod(m, struct sockaddr_in6 *);
217 		sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6);
218 		sin6->sin6_family = AF_INET6;
219 		sin6->sin6_addr = in6addr_any;
220 		sin6->sin6_port = 0;
221 		error = sobind(so, m, &proc0);
222 		m_freem(m);
223 		if (error)
224 			goto bad;
225 	}
226 #endif
227 
228 	/*
229 	 * Protocols that do not require connections may be optionally left
230 	 * unconnected for servers that reply from a port other than NFS_PORT.
231 	 */
232 	if (nmp->nm_flag & NFSMNT_NOCONN) {
233 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
234 			error = ENOTCONN;
235 			goto bad;
236 		}
237 	} else {
238 		error = soconnect(so, nmp->nm_nam);
239 		if (error)
240 			goto bad;
241 
242 		/*
243 		 * Wait for the connection to complete. Cribbed from the
244 		 * connect system call but with the wait timing out so
245 		 * that interruptible mounts don't hang here for a long time.
246 		 */
247 		s = splsoftnet();
248 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
249 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
250 				"nfscn1", 2 * hz);
251 			if ((so->so_state & SS_ISCONNECTING) &&
252 			    so->so_error == 0 && rep &&
253 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
254 				so->so_state &= ~SS_ISCONNECTING;
255 				splx(s);
256 				goto bad;
257 			}
258 		}
259 		if (so->so_error) {
260 			error = so->so_error;
261 			so->so_error = 0;
262 			splx(s);
263 			goto bad;
264 		}
265 		splx(s);
266 	}
267 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
268 		so->so_rcv.sb_timeo = (5 * hz);
269 		so->so_snd.sb_timeo = (5 * hz);
270 	} else {
271 		so->so_rcv.sb_timeo = 0;
272 		so->so_snd.sb_timeo = 0;
273 	}
274 	if (nmp->nm_sotype == SOCK_DGRAM) {
275 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
276 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
277 		    NFS_MAXPKTHDR) * 2;
278 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
279 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
280 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
281 		    NFS_MAXPKTHDR) * 2;
282 	} else {
283 		if (nmp->nm_sotype != SOCK_STREAM)
284 			panic("nfscon sotype");
285 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
286 			m = m_get(M_WAIT, MT_SOOPTS);
287 			MCLAIM(m, so->so_mowner);
288 			*mtod(m, int32_t *) = 1;
289 			m->m_len = sizeof(int32_t);
290 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
291 		}
292 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
293 			m = m_get(M_WAIT, MT_SOOPTS);
294 			MCLAIM(m, so->so_mowner);
295 			*mtod(m, int32_t *) = 1;
296 			m->m_len = sizeof(int32_t);
297 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
298 		}
299 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
300 		    sizeof (u_int32_t)) * 2;
301 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
302 		    sizeof (u_int32_t)) * 2;
303 	}
304 	error = soreserve(so, sndreserve, rcvreserve);
305 	if (error)
306 		goto bad;
307 	so->so_rcv.sb_flags |= SB_NOINTR;
308 	so->so_snd.sb_flags |= SB_NOINTR;
309 
310 	/* Initialize other non-zero congestion variables */
311 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
312 		NFS_TIMEO << 3;
313 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
314 		nmp->nm_sdrtt[3] = 0;
315 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
316 	nmp->nm_sent = 0;
317 	nmp->nm_timeouts = 0;
318 	return (0);
319 
320 bad:
321 	nfs_disconnect(nmp);
322 	return (error);
323 }
324 
325 /*
326  * Reconnect routine:
327  * Called when a connection is broken on a reliable protocol.
328  * - clean up the old socket
329  * - nfs_connect() again
330  * - set R_MUSTRESEND for all outstanding requests on mount point
331  * If this fails the mount point is DEAD!
332  * nb: Must be called with the nfs_sndlock() set on the mount point.
333  */
334 int
335 nfs_reconnect(rep)
336 	struct nfsreq *rep;
337 {
338 	struct nfsreq *rp;
339 	struct nfsmount *nmp = rep->r_nmp;
340 	int error;
341 
342 	nfs_disconnect(nmp);
343 	while ((error = nfs_connect(nmp, rep)) != 0) {
344 		if (error == EINTR || error == ERESTART)
345 			return (EINTR);
346 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0);
347 	}
348 
349 	/*
350 	 * Loop through outstanding request list and fix up all requests
351 	 * on old socket.
352 	 */
353 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
354 		if (rp->r_nmp == nmp)
355 			rp->r_flags |= R_MUSTRESEND;
356 	}
357 	return (0);
358 }
359 
360 /*
361  * NFS disconnect. Clean up and unlink.
362  */
363 void
364 nfs_disconnect(nmp)
365 	struct nfsmount *nmp;
366 {
367 	struct socket *so;
368 	int drain = 0;
369 
370 	if (nmp->nm_so) {
371 		so = nmp->nm_so;
372 		nmp->nm_so = (struct socket *)0;
373 		soshutdown(so, 2);
374 		drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
375 		if (drain) {
376 			/*
377 			 * soshutdown() above should wake up the current
378 			 * listener.
379 			 * Now wake up those waiting for the receive lock, and
380 			 * wait for them to go away unhappy, to prevent *nmp
381 			 * from evaporating while they're sleeping.
382 			 */
383 			while (nmp->nm_waiters > 0) {
384 				wakeup (&nmp->nm_iflag);
385 				(void) tsleep(&nmp->nm_waiters, PVFS,
386 				    "nfsdis", 0);
387 			}
388 		}
389 		soclose(so);
390 	}
391 #ifdef DIAGNOSTIC
392 	if (drain && (nmp->nm_waiters > 0))
393 		panic("nfs_disconnect: waiters left after drain?");
394 #endif
395 }
396 
397 void
398 nfs_safedisconnect(nmp)
399 	struct nfsmount *nmp;
400 {
401 	struct nfsreq dummyreq;
402 
403 	memset(&dummyreq, 0, sizeof(dummyreq));
404 	dummyreq.r_nmp = nmp;
405 	nfs_rcvlock(&dummyreq); /* XXX ignored error return */
406 	nfs_disconnect(nmp);
407 	nfs_rcvunlock(nmp);
408 }
409 
410 /*
411  * This is the nfs send routine. For connection based socket types, it
412  * must be called with an nfs_sndlock() on the socket.
413  * "rep == NULL" indicates that it has been called from a server.
414  * For the client side:
415  * - return EINTR if the RPC is terminated, 0 otherwise
416  * - set R_MUSTRESEND if the send fails for any reason
417  * - do any cleanup required by recoverable socket errors (? ? ?)
418  * For the server side:
419  * - return EINTR or ERESTART if interrupted by a signal
420  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
421  * - do any cleanup required by recoverable socket errors (? ? ?)
422  */
423 int
424 nfs_send(so, nam, top, rep)
425 	struct socket *so;
426 	struct mbuf *nam;
427 	struct mbuf *top;
428 	struct nfsreq *rep;
429 {
430 	struct mbuf *sendnam;
431 	int error, soflags, flags;
432 
433 	if (rep) {
434 		if (rep->r_flags & R_SOFTTERM) {
435 			m_freem(top);
436 			return (EINTR);
437 		}
438 		if ((so = rep->r_nmp->nm_so) == NULL) {
439 			rep->r_flags |= R_MUSTRESEND;
440 			m_freem(top);
441 			return (0);
442 		}
443 		rep->r_flags &= ~R_MUSTRESEND;
444 		soflags = rep->r_nmp->nm_soflags;
445 	} else
446 		soflags = so->so_proto->pr_flags;
447 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
448 		sendnam = (struct mbuf *)0;
449 	else
450 		sendnam = nam;
451 	if (so->so_type == SOCK_SEQPACKET)
452 		flags = MSG_EOR;
453 	else
454 		flags = 0;
455 
456 	error = (*so->so_send)(so, sendnam, (struct uio *)0, top,
457 		(struct mbuf *)0, flags);
458 	if (error) {
459 		if (rep) {
460 			if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
461 				/*
462 				 * We're too fast for the network/driver,
463 				 * and UDP isn't flowcontrolled.
464 				 * We need to resend. This is not fatal,
465 				 * just try again.
466 				 *
467 				 * Could be smarter here by doing some sort
468 				 * of a backoff, but this is rare.
469 				 */
470 				rep->r_flags |= R_MUSTRESEND;
471 			} else {
472 				log(LOG_INFO, "nfs send error %d for %s\n",
473 				    error,
474 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
475 				/*
476 				 * Deal with errors for the client side.
477 				 */
478 				if (rep->r_flags & R_SOFTTERM)
479 					error = EINTR;
480 				else
481 					rep->r_flags |= R_MUSTRESEND;
482 			}
483 		} else {
484 			/*
485 			 * See above. This error can happen under normal
486 			 * circumstances and the log is too noisy.
487 			 * The error will still show up in nfsstat.
488 			 */
489 			if (error != ENOBUFS || so->so_type != SOCK_DGRAM)
490 				log(LOG_INFO, "nfsd send error %d\n", error);
491 		}
492 
493 		/*
494 		 * Handle any recoverable (soft) socket errors here. (? ? ?)
495 		 */
496 		if (error != EINTR && error != ERESTART &&
497 			error != EWOULDBLOCK && error != EPIPE)
498 			error = 0;
499 	}
500 	return (error);
501 }
502 
503 #ifdef NFS
504 /*
505  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
506  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
507  * Mark and consolidate the data into a new mbuf list.
508  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
509  *     small mbufs.
510  * For SOCK_STREAM we must be very careful to read an entire record once
511  * we have read any of it, even if the system call has been interrupted.
512  */
513 int
514 nfs_receive(rep, aname, mp)
515 	struct nfsreq *rep;
516 	struct mbuf **aname;
517 	struct mbuf **mp;
518 {
519 	struct socket *so;
520 	struct uio auio;
521 	struct iovec aio;
522 	struct mbuf *m;
523 	struct mbuf *control;
524 	u_int32_t len;
525 	struct mbuf **getnam;
526 	int error, sotype, rcvflg;
527 	struct proc *p = curproc;	/* XXX */
528 
529 	/*
530 	 * Set up arguments for soreceive()
531 	 */
532 	*mp = (struct mbuf *)0;
533 	*aname = (struct mbuf *)0;
534 	sotype = rep->r_nmp->nm_sotype;
535 
536 	/*
537 	 * For reliable protocols, lock against other senders/receivers
538 	 * in case a reconnect is necessary.
539 	 * For SOCK_STREAM, first get the Record Mark to find out how much
540 	 * more there is to get.
541 	 * We must lock the socket against other receivers
542 	 * until we have an entire rpc request/reply.
543 	 */
544 	if (sotype != SOCK_DGRAM) {
545 		error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
546 		if (error)
547 			return (error);
548 tryagain:
549 		/*
550 		 * Check for fatal errors and resending request.
551 		 */
552 		/*
553 		 * Ugh: If a reconnect attempt just happened, nm_so
554 		 * would have changed. NULL indicates a failed
555 		 * attempt that has essentially shut down this
556 		 * mount point.
557 		 */
558 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
559 			nfs_sndunlock(&rep->r_nmp->nm_iflag);
560 			return (EINTR);
561 		}
562 		so = rep->r_nmp->nm_so;
563 		if (!so) {
564 			error = nfs_reconnect(rep);
565 			if (error) {
566 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
567 				return (error);
568 			}
569 			goto tryagain;
570 		}
571 		while (rep->r_flags & R_MUSTRESEND) {
572 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
573 			nfsstats.rpcretries++;
574 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
575 			if (error) {
576 				if (error == EINTR || error == ERESTART ||
577 				    (error = nfs_reconnect(rep)) != 0) {
578 					nfs_sndunlock(&rep->r_nmp->nm_iflag);
579 					return (error);
580 				}
581 				goto tryagain;
582 			}
583 		}
584 		nfs_sndunlock(&rep->r_nmp->nm_iflag);
585 		if (sotype == SOCK_STREAM) {
586 			aio.iov_base = (caddr_t) &len;
587 			aio.iov_len = sizeof(u_int32_t);
588 			auio.uio_iov = &aio;
589 			auio.uio_iovcnt = 1;
590 			auio.uio_segflg = UIO_SYSSPACE;
591 			auio.uio_rw = UIO_READ;
592 			auio.uio_offset = 0;
593 			auio.uio_resid = sizeof(u_int32_t);
594 			auio.uio_procp = p;
595 			do {
596 			   rcvflg = MSG_WAITALL;
597 			   error = (*so->so_receive)(so, (struct mbuf **)0, &auio,
598 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
599 			   if (error == EWOULDBLOCK && rep) {
600 				if (rep->r_flags & R_SOFTTERM)
601 					return (EINTR);
602 			   }
603 			} while (error == EWOULDBLOCK);
604 			if (!error && auio.uio_resid > 0) {
605 			    /*
606 			     * Don't log a 0 byte receive; it means
607 			     * that the socket has been closed, and
608 			     * can happen during normal operation
609 			     * (forcible unmount or Solaris server).
610 			     */
611 			    if (auio.uio_resid != sizeof (u_int32_t))
612 			      log(LOG_INFO,
613 				 "short receive (%lu/%lu) from nfs server %s\n",
614 				 (u_long)sizeof(u_int32_t) - auio.uio_resid,
615 				 (u_long)sizeof(u_int32_t),
616 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
617 			    error = EPIPE;
618 			}
619 			if (error)
620 				goto errout;
621 			len = ntohl(len) & ~0x80000000;
622 			/*
623 			 * This is SERIOUS! We are out of sync with the sender
624 			 * and forcing a disconnect/reconnect is all I can do.
625 			 */
626 			if (len > NFS_MAXPACKET) {
627 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
628 				"impossible packet length",
629 				len,
630 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
631 			    error = EFBIG;
632 			    goto errout;
633 			}
634 			auio.uio_resid = len;
635 			do {
636 			    rcvflg = MSG_WAITALL;
637 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
638 				&auio, mp, (struct mbuf **)0, &rcvflg);
639 			} while (error == EWOULDBLOCK || error == EINTR ||
640 				 error == ERESTART);
641 			if (!error && auio.uio_resid > 0) {
642 			    if (len != auio.uio_resid)
643 			      log(LOG_INFO,
644 				"short receive (%lu/%d) from nfs server %s\n",
645 				(u_long)len - auio.uio_resid, len,
646 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
647 			    error = EPIPE;
648 			}
649 		} else {
650 			/*
651 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
652 			 * and soreceive() will return when it has either a
653 			 * control msg or a data msg.
654 			 * We have no use for control msg., but must grab them
655 			 * and then throw them away so we know what is going
656 			 * on.
657 			 */
658 			auio.uio_resid = len = 100000000; /* Anything Big */
659 			auio.uio_procp = p;
660 			do {
661 			    rcvflg = 0;
662 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
663 				&auio, mp, &control, &rcvflg);
664 			    if (control)
665 				m_freem(control);
666 			    if (error == EWOULDBLOCK && rep) {
667 				if (rep->r_flags & R_SOFTTERM)
668 					return (EINTR);
669 			    }
670 			} while (error == EWOULDBLOCK ||
671 				 (!error && *mp == NULL && control));
672 			if ((rcvflg & MSG_EOR) == 0)
673 				printf("Egad!!\n");
674 			if (!error && *mp == NULL)
675 				error = EPIPE;
676 			len -= auio.uio_resid;
677 		}
678 errout:
679 		if (error && error != EINTR && error != ERESTART) {
680 			m_freem(*mp);
681 			*mp = (struct mbuf *)0;
682 			if (error != EPIPE)
683 				log(LOG_INFO,
684 				    "receive error %d from nfs server %s\n",
685 				    error,
686 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
687 			error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
688 			if (!error)
689 				error = nfs_reconnect(rep);
690 			if (!error)
691 				goto tryagain;
692 			else
693 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
694 		}
695 	} else {
696 		if ((so = rep->r_nmp->nm_so) == NULL)
697 			return (EACCES);
698 		if (so->so_state & SS_ISCONNECTED)
699 			getnam = (struct mbuf **)0;
700 		else
701 			getnam = aname;
702 		auio.uio_resid = len = 1000000;
703 		auio.uio_procp = p;
704 		do {
705 			rcvflg = 0;
706 			error =  (*so->so_receive)(so, getnam, &auio, mp,
707 				(struct mbuf **)0, &rcvflg);
708 			if (error == EWOULDBLOCK &&
709 			    (rep->r_flags & R_SOFTTERM))
710 				return (EINTR);
711 		} while (error == EWOULDBLOCK);
712 		len -= auio.uio_resid;
713 		if (!error && *mp == NULL)
714 			error = EPIPE;
715 	}
716 	if (error) {
717 		m_freem(*mp);
718 		*mp = (struct mbuf *)0;
719 	}
720 	return (error);
721 }
722 
723 /*
724  * Implement receipt of reply on a socket.
725  * We must search through the list of received datagrams matching them
726  * with outstanding requests using the xid, until ours is found.
727  */
728 /* ARGSUSED */
729 int
730 nfs_reply(myrep)
731 	struct nfsreq *myrep;
732 {
733 	struct nfsreq *rep;
734 	struct nfsmount *nmp = myrep->r_nmp;
735 	int32_t t1;
736 	struct mbuf *mrep, *nam, *md;
737 	u_int32_t rxid, *tl;
738 	caddr_t dpos, cp2;
739 	int error;
740 
741 	/*
742 	 * Loop around until we get our own reply
743 	 */
744 	for (;;) {
745 		/*
746 		 * Lock against other receivers so that I don't get stuck in
747 		 * sbwait() after someone else has received my reply for me.
748 		 * Also necessary for connection based protocols to avoid
749 		 * race conditions during a reconnect.
750 		 */
751 		error = nfs_rcvlock(myrep);
752 		if (error == EALREADY)
753 			return (0);
754 		if (error)
755 			return (error);
756 		/*
757 		 * Get the next Rpc reply off the socket
758 		 */
759 		nmp->nm_waiters++;
760 		error = nfs_receive(myrep, &nam, &mrep);
761 		nfs_rcvunlock(nmp);
762 		if (error) {
763 
764 			if (nmp->nm_iflag & NFSMNT_DISMNT) {
765 				/*
766 				 * Oops, we're going away now..
767 				 */
768 				nmp->nm_waiters--;
769 				wakeup (&nmp->nm_waiters);
770 				return error;
771 			}
772 			nmp->nm_waiters--;
773 			/*
774 			 * Ignore routing errors on connectionless protocols? ?
775 			 */
776 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
777 				nmp->nm_so->so_error = 0;
778 #ifdef DEBUG
779 				printf("nfs_reply: ignoring error %d\n", error);
780 #endif
781 				if (myrep->r_flags & R_GETONEREP)
782 					return (0);
783 				continue;
784 			}
785 			return (error);
786 		}
787 		nmp->nm_waiters--;
788 		if (nam)
789 			m_freem(nam);
790 
791 		/*
792 		 * Get the xid and check that it is an rpc reply
793 		 */
794 		md = mrep;
795 		dpos = mtod(md, caddr_t);
796 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
797 		rxid = *tl++;
798 		if (*tl != rpc_reply) {
799 #ifndef NFS_V2_ONLY
800 			if (nmp->nm_flag & NFSMNT_NQNFS) {
801 				if (nqnfs_callback(nmp, mrep, md, dpos))
802 					nfsstats.rpcinvalid++;
803 			} else
804 #endif
805 			{
806 				nfsstats.rpcinvalid++;
807 				m_freem(mrep);
808 			}
809 nfsmout:
810 			if (myrep->r_flags & R_GETONEREP)
811 				return (0);
812 			continue;
813 		}
814 
815 		/*
816 		 * Loop through the request list to match up the reply
817 		 * Iff no match, just drop the datagram
818 		 */
819 		TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
820 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
821 				/* Found it.. */
822 				rep->r_mrep = mrep;
823 				rep->r_md = md;
824 				rep->r_dpos = dpos;
825 				if (nfsrtton) {
826 					struct rttl *rt;
827 
828 					rt = &nfsrtt.rttl[nfsrtt.pos];
829 					rt->proc = rep->r_procnum;
830 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
831 					rt->sent = nmp->nm_sent;
832 					rt->cwnd = nmp->nm_cwnd;
833 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
834 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
835 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
836 					rt->tstamp = time;
837 					if (rep->r_flags & R_TIMING)
838 						rt->rtt = rep->r_rtt;
839 					else
840 						rt->rtt = 1000000;
841 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
842 				}
843 				/*
844 				 * Update congestion window.
845 				 * Do the additive increase of
846 				 * one rpc/rtt.
847 				 */
848 				if (nmp->nm_cwnd <= nmp->nm_sent) {
849 					nmp->nm_cwnd +=
850 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
851 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
852 					if (nmp->nm_cwnd > NFS_MAXCWND)
853 						nmp->nm_cwnd = NFS_MAXCWND;
854 				}
855 				rep->r_flags &= ~R_SENT;
856 				nmp->nm_sent -= NFS_CWNDSCALE;
857 				/*
858 				 * Update rtt using a gain of 0.125 on the mean
859 				 * and a gain of 0.25 on the deviation.
860 				 */
861 				if (rep->r_flags & R_TIMING) {
862 					/*
863 					 * Since the timer resolution of
864 					 * NFS_HZ is so course, it can often
865 					 * result in r_rtt == 0. Since
866 					 * r_rtt == N means that the actual
867 					 * rtt is between N+dt and N+2-dt ticks,
868 					 * add 1.
869 					 */
870 					t1 = rep->r_rtt + 1;
871 					t1 -= (NFS_SRTT(rep) >> 3);
872 					NFS_SRTT(rep) += t1;
873 					if (t1 < 0)
874 						t1 = -t1;
875 					t1 -= (NFS_SDRTT(rep) >> 2);
876 					NFS_SDRTT(rep) += t1;
877 				}
878 				nmp->nm_timeouts = 0;
879 				break;
880 			}
881 		}
882 		/*
883 		 * If not matched to a request, drop it.
884 		 * If it's mine, get out.
885 		 */
886 		if (rep == 0) {
887 			nfsstats.rpcunexpected++;
888 			m_freem(mrep);
889 		} else if (rep == myrep) {
890 			if (rep->r_mrep == NULL)
891 				panic("nfsreply nil");
892 			return (0);
893 		}
894 		if (myrep->r_flags & R_GETONEREP)
895 			return (0);
896 	}
897 }
898 
899 /*
900  * nfs_request - goes something like this
901  *	- fill in request struct
902  *	- links it into list
903  *	- calls nfs_send() for first transmit
904  *	- calls nfs_receive() to get reply
905  *	- break down rpc header and return with nfs reply pointed to
906  *	  by mrep or error
907  * nb: always frees up mreq mbuf list
908  */
909 int
910 nfs_request(np, mrest, procnum, procp, cred, mrp, mdp, dposp)
911 	struct nfsnode *np;
912 	struct mbuf *mrest;
913 	int procnum;
914 	struct proc *procp;
915 	struct ucred *cred;
916 	struct mbuf **mrp;
917 	struct mbuf **mdp;
918 	caddr_t *dposp;
919 {
920 	struct mbuf *m, *mrep;
921 	struct nfsreq *rep;
922 	u_int32_t *tl;
923 	int i;
924 	struct nfsmount *nmp;
925 	struct mbuf *md, *mheadend;
926 	char nickv[RPCX_NICKVERF];
927 	time_t reqtime, waituntil;
928 	caddr_t dpos, cp2;
929 	int t1, s, error = 0, mrest_len, auth_len, auth_type;
930 	int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0;
931 	int verf_len, verf_type;
932 	u_int32_t xid;
933 	char *auth_str, *verf_str;
934 	NFSKERBKEY_T key;		/* save session key */
935 	struct ucred acred;
936 #ifndef NFS_V2_ONLY
937 	int nqlflag, cachable;
938 	u_quad_t frev;
939 #endif
940 
941 	KASSERT(cred != NULL);
942 	nmp = VFSTONFS(np->n_vnode->v_mount);
943 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
944 	rep->r_nmp = nmp;
945 	rep->r_procp = procp;
946 	rep->r_procnum = procnum;
947 	i = 0;
948 	m = mrest;
949 	while (m) {
950 		i += m->m_len;
951 		m = m->m_next;
952 	}
953 	mrest_len = i;
954 
955 	/*
956 	 * Get the RPC header with authorization.
957 	 */
958 kerbauth:
959 	verf_str = auth_str = (char *)0;
960 	if (nmp->nm_flag & NFSMNT_KERB) {
961 		verf_str = nickv;
962 		verf_len = sizeof (nickv);
963 		auth_type = RPCAUTH_KERB4;
964 		memset((caddr_t)key, 0, sizeof (key));
965 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
966 			&auth_len, verf_str, verf_len)) {
967 			error = nfs_getauth(nmp, rep, cred, &auth_str,
968 				&auth_len, verf_str, &verf_len, key);
969 			if (error) {
970 				free((caddr_t)rep, M_NFSREQ);
971 				m_freem(mrest);
972 				return (error);
973 			}
974 		}
975 	} else {
976 		switch (procnum) {
977 		case NFSPROC_READ:
978 		case NFSPROC_WRITE:
979 		case NFSPROC_COMMIT:
980 			acred.cr_uid = np->n_vattr->va_uid;
981 			acred.cr_gid = np->n_vattr->va_gid;
982 			acred.cr_ngroups = 0;
983 			acred.cr_ref = 2;	/* Just to be safe.. */
984 			cred = &acred;
985 			break;
986 		}
987 		auth_type = RPCAUTH_UNIX;
988 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
989 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
990 			5 * NFSX_UNSIGNED;
991 	}
992 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
993 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
994 	if (auth_str)
995 		free(auth_str, M_TEMP);
996 
997 	/*
998 	 * For stream protocols, insert a Sun RPC Record Mark.
999 	 */
1000 	if (nmp->nm_sotype == SOCK_STREAM) {
1001 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
1002 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
1003 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
1004 	}
1005 	rep->r_mreq = m;
1006 	rep->r_xid = xid;
1007 tryagain:
1008 	if (nmp->nm_flag & NFSMNT_SOFT)
1009 		rep->r_retry = nmp->nm_retry;
1010 	else
1011 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
1012 	rep->r_rtt = rep->r_rexmit = 0;
1013 	if (proct[procnum] > 0)
1014 		rep->r_flags = R_TIMING;
1015 	else
1016 		rep->r_flags = 0;
1017 	rep->r_mrep = NULL;
1018 
1019 	/*
1020 	 * Do the client side RPC.
1021 	 */
1022 	nfsstats.rpcrequests++;
1023 	/*
1024 	 * Chain request into list of outstanding requests. Be sure
1025 	 * to put it LAST so timer finds oldest requests first.
1026 	 */
1027 	s = splsoftnet();
1028 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1029 
1030 	/* Get send time for nqnfs */
1031 	reqtime = time.tv_sec;
1032 
1033 	/*
1034 	 * If backing off another request or avoiding congestion, don't
1035 	 * send this one now but let timer do it. If not timing a request,
1036 	 * do it now.
1037 	 */
1038 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1039 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1040 		nmp->nm_sent < nmp->nm_cwnd)) {
1041 		splx(s);
1042 		if (nmp->nm_soflags & PR_CONNREQUIRED)
1043 			error = nfs_sndlock(&nmp->nm_iflag, rep);
1044 		if (!error) {
1045 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
1046 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep);
1047 			if (nmp->nm_soflags & PR_CONNREQUIRED)
1048 				nfs_sndunlock(&nmp->nm_iflag);
1049 		}
1050 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1051 			nmp->nm_sent += NFS_CWNDSCALE;
1052 			rep->r_flags |= R_SENT;
1053 		}
1054 	} else {
1055 		splx(s);
1056 		rep->r_rtt = -1;
1057 	}
1058 
1059 	/*
1060 	 * Wait for the reply from our send or the timer's.
1061 	 */
1062 	if (!error || error == EPIPE)
1063 		error = nfs_reply(rep);
1064 
1065 	/*
1066 	 * RPC done, unlink the request.
1067 	 */
1068 	s = splsoftnet();
1069 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1070 	splx(s);
1071 
1072 	/*
1073 	 * Decrement the outstanding request count.
1074 	 */
1075 	if (rep->r_flags & R_SENT) {
1076 		rep->r_flags &= ~R_SENT;	/* paranoia */
1077 		nmp->nm_sent -= NFS_CWNDSCALE;
1078 	}
1079 
1080 	/*
1081 	 * If there was a successful reply and a tprintf msg.
1082 	 * tprintf a response.
1083 	 */
1084 	if (!error && (rep->r_flags & R_TPRINTFMSG))
1085 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1086 		    "is alive again");
1087 	mrep = rep->r_mrep;
1088 	md = rep->r_md;
1089 	dpos = rep->r_dpos;
1090 	if (error) {
1091 		m_freem(rep->r_mreq);
1092 		free((caddr_t)rep, M_NFSREQ);
1093 		return (error);
1094 	}
1095 
1096 	/*
1097 	 * break down the rpc header and check if ok
1098 	 */
1099 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1100 	if (*tl++ == rpc_msgdenied) {
1101 		if (*tl == rpc_mismatch)
1102 			error = EOPNOTSUPP;
1103 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1104 			if (!failed_auth) {
1105 				failed_auth++;
1106 				mheadend->m_next = (struct mbuf *)0;
1107 				m_freem(mrep);
1108 				m_freem(rep->r_mreq);
1109 				goto kerbauth;
1110 			} else
1111 				error = EAUTH;
1112 		} else
1113 			error = EACCES;
1114 		m_freem(mrep);
1115 		m_freem(rep->r_mreq);
1116 		free((caddr_t)rep, M_NFSREQ);
1117 		return (error);
1118 	}
1119 
1120 	/*
1121 	 * Grab any Kerberos verifier, otherwise just throw it away.
1122 	 */
1123 	verf_type = fxdr_unsigned(int, *tl++);
1124 	i = fxdr_unsigned(int32_t, *tl);
1125 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1126 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1127 		if (error)
1128 			goto nfsmout;
1129 	} else if (i > 0)
1130 		nfsm_adv(nfsm_rndup(i));
1131 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1132 	/* 0 == ok */
1133 	if (*tl == 0) {
1134 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1135 		if (*tl != 0) {
1136 			error = fxdr_unsigned(int, *tl);
1137 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1138 				error == NFSERR_TRYLATER) {
1139 				m_freem(mrep);
1140 				error = 0;
1141 				waituntil = time.tv_sec + trylater_delay;
1142 				while (time.tv_sec < waituntil)
1143 					(void) tsleep((caddr_t)&lbolt,
1144 						PSOCK, "nqnfstry", 0);
1145 				trylater_delay *= NFS_TRYLATERDELMUL;
1146 				if (trylater_delay > NFS_TRYLATERDELMAX)
1147 					trylater_delay = NFS_TRYLATERDELMAX;
1148 				/*
1149 				 * RFC1813:
1150 				 * The client should wait and then try
1151 				 * the request with a new RPC transaction ID.
1152 				 */
1153 				nfs_renewxid(rep);
1154 				goto tryagain;
1155 			}
1156 
1157 			/*
1158 			 * If the File Handle was stale, invalidate the
1159 			 * lookup cache, just in case.
1160 			 */
1161 			if (error == ESTALE)
1162 				cache_purge(NFSTOV(np));
1163 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1164 				*mrp = mrep;
1165 				*mdp = md;
1166 				*dposp = dpos;
1167 				error |= NFSERR_RETERR;
1168 			} else
1169 				m_freem(mrep);
1170 			m_freem(rep->r_mreq);
1171 			free((caddr_t)rep, M_NFSREQ);
1172 			return (error);
1173 		}
1174 
1175 #ifndef NFS_V2_ONLY
1176 		/*
1177 		 * For nqnfs, get any lease in reply
1178 		 */
1179 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1180 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1181 			if (*tl) {
1182 				nqlflag = fxdr_unsigned(int, *tl);
1183 				nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1184 				cachable = fxdr_unsigned(int, *tl++);
1185 				reqtime += fxdr_unsigned(int, *tl++);
1186 				if (reqtime > time.tv_sec) {
1187 				    frev = fxdr_hyper(tl);
1188 				    nqnfs_clientlease(nmp, np, nqlflag,
1189 					cachable, reqtime, frev);
1190 				}
1191 			}
1192 		}
1193 #endif
1194 		*mrp = mrep;
1195 		*mdp = md;
1196 		*dposp = dpos;
1197 		m_freem(rep->r_mreq);
1198 		FREE((caddr_t)rep, M_NFSREQ);
1199 		return (0);
1200 	}
1201 	m_freem(mrep);
1202 	error = EPROTONOSUPPORT;
1203 nfsmout:
1204 	m_freem(rep->r_mreq);
1205 	free((caddr_t)rep, M_NFSREQ);
1206 	return (error);
1207 }
1208 #endif /* NFS */
1209 
1210 /*
1211  * Generate the rpc reply header
1212  * siz arg. is used to decide if adding a cluster is worthwhile
1213  */
1214 int
1215 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1216 	int siz;
1217 	struct nfsrv_descript *nd;
1218 	struct nfssvc_sock *slp;
1219 	int err;
1220 	int cache;
1221 	u_quad_t *frev;
1222 	struct mbuf **mrq;
1223 	struct mbuf **mbp;
1224 	caddr_t *bposp;
1225 {
1226 	u_int32_t *tl;
1227 	struct mbuf *mreq;
1228 	caddr_t bpos;
1229 	struct mbuf *mb;
1230 
1231 	mreq = m_gethdr(M_WAIT, MT_DATA);
1232 	MCLAIM(mreq, &nfs_mowner);
1233 	mb = mreq;
1234 	/*
1235 	 * If this is a big reply, use a cluster else
1236 	 * try and leave leading space for the lower level headers.
1237 	 */
1238 	siz += RPC_REPLYSIZ;
1239 	if (siz >= max_datalen) {
1240 		m_clget(mreq, M_WAIT);
1241 	} else
1242 		mreq->m_data += max_hdr;
1243 	tl = mtod(mreq, u_int32_t *);
1244 	mreq->m_len = 6 * NFSX_UNSIGNED;
1245 	bpos = ((caddr_t)tl) + mreq->m_len;
1246 	*tl++ = txdr_unsigned(nd->nd_retxid);
1247 	*tl++ = rpc_reply;
1248 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1249 		*tl++ = rpc_msgdenied;
1250 		if (err & NFSERR_AUTHERR) {
1251 			*tl++ = rpc_autherr;
1252 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1253 			mreq->m_len -= NFSX_UNSIGNED;
1254 			bpos -= NFSX_UNSIGNED;
1255 		} else {
1256 			*tl++ = rpc_mismatch;
1257 			*tl++ = txdr_unsigned(RPC_VER2);
1258 			*tl = txdr_unsigned(RPC_VER2);
1259 		}
1260 	} else {
1261 		*tl++ = rpc_msgaccepted;
1262 
1263 		/*
1264 		 * For Kerberos authentication, we must send the nickname
1265 		 * verifier back, otherwise just RPCAUTH_NULL.
1266 		 */
1267 		if (nd->nd_flag & ND_KERBFULL) {
1268 			struct nfsuid *nuidp;
1269 			struct timeval ktvin, ktvout;
1270 
1271 			LIST_FOREACH(nuidp, NUIDHASH(slp, nd->nd_cr.cr_uid),
1272 			    nu_hash) {
1273 				if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1274 				    (!nd->nd_nam2 || netaddr_match(
1275 				    NU_NETFAM(nuidp), &nuidp->nu_haddr,
1276 				    nd->nd_nam2)))
1277 					break;
1278 			}
1279 			if (nuidp) {
1280 				ktvin.tv_sec =
1281 				    txdr_unsigned(nuidp->nu_timestamp.tv_sec
1282 					- 1);
1283 				ktvin.tv_usec =
1284 				    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1285 
1286 				/*
1287 				 * Encrypt the timestamp in ecb mode using the
1288 				 * session key.
1289 				 */
1290 #ifdef NFSKERB
1291 				XXX
1292 #endif
1293 
1294 				*tl++ = rpc_auth_kerb;
1295 				*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1296 				*tl = ktvout.tv_sec;
1297 				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1298 				*tl++ = ktvout.tv_usec;
1299 				*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1300 			} else {
1301 				*tl++ = 0;
1302 				*tl++ = 0;
1303 			}
1304 		} else {
1305 			*tl++ = 0;
1306 			*tl++ = 0;
1307 		}
1308 		switch (err) {
1309 		case EPROGUNAVAIL:
1310 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1311 			break;
1312 		case EPROGMISMATCH:
1313 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1314 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1315 			if (nd->nd_flag & ND_NQNFS) {
1316 				*tl++ = txdr_unsigned(3);
1317 				*tl = txdr_unsigned(3);
1318 			} else {
1319 				*tl++ = txdr_unsigned(2);
1320 				*tl = txdr_unsigned(3);
1321 			}
1322 			break;
1323 		case EPROCUNAVAIL:
1324 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1325 			break;
1326 		case EBADRPC:
1327 			*tl = txdr_unsigned(RPC_GARBAGE);
1328 			break;
1329 		default:
1330 			*tl = 0;
1331 			if (err != NFSERR_RETVOID) {
1332 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1333 				if (err)
1334 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1335 				else
1336 				    *tl = 0;
1337 			}
1338 			break;
1339 		};
1340 	}
1341 
1342 	/*
1343 	 * For nqnfs, piggyback lease as requested.
1344 	 */
1345 	if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1346 		if (nd->nd_flag & ND_LEASE) {
1347 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1348 			*tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1349 			*tl++ = txdr_unsigned(cache);
1350 			*tl++ = txdr_unsigned(nd->nd_duration);
1351 			txdr_hyper(*frev, tl);
1352 		} else {
1353 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1354 			*tl = 0;
1355 		}
1356 	}
1357 	if (mrq != NULL)
1358 		*mrq = mreq;
1359 	*mbp = mb;
1360 	*bposp = bpos;
1361 	if (err != 0 && err != NFSERR_RETVOID)
1362 		nfsstats.srvrpc_errs++;
1363 	return (0);
1364 }
1365 
1366 /*
1367  * Nfs timer routine
1368  * Scan the nfsreq list and retranmit any requests that have timed out
1369  * To avoid retransmission attempts on STREAM sockets (in the future) make
1370  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1371  */
1372 void
1373 nfs_timer(arg)
1374 	void *arg;	/* never used */
1375 {
1376 	struct nfsreq *rep;
1377 	struct mbuf *m;
1378 	struct socket *so;
1379 	struct nfsmount *nmp;
1380 	int timeo;
1381 	int s, error;
1382 #ifdef NFSSERVER
1383 	struct nfssvc_sock *slp;
1384 	static long lasttime = 0;
1385 	u_quad_t cur_usec;
1386 #endif
1387 
1388 	s = splsoftnet();
1389 	TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
1390 		nmp = rep->r_nmp;
1391 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1392 			continue;
1393 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1394 			rep->r_flags |= R_SOFTTERM;
1395 			continue;
1396 		}
1397 		if (rep->r_rtt >= 0) {
1398 			rep->r_rtt++;
1399 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1400 				timeo = nmp->nm_timeo;
1401 			else
1402 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1403 			if (nmp->nm_timeouts > 0)
1404 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1405 			if (rep->r_rtt <= timeo)
1406 				continue;
1407 			if (nmp->nm_timeouts <
1408 			    (sizeof(nfs_backoff) / sizeof(nfs_backoff[0])))
1409 				nmp->nm_timeouts++;
1410 		}
1411 		/*
1412 		 * Check for server not responding
1413 		 */
1414 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1415 		     rep->r_rexmit > nmp->nm_deadthresh) {
1416 			nfs_msg(rep->r_procp,
1417 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1418 			    "not responding");
1419 			rep->r_flags |= R_TPRINTFMSG;
1420 		}
1421 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1422 			nfsstats.rpctimeouts++;
1423 			rep->r_flags |= R_SOFTTERM;
1424 			continue;
1425 		}
1426 		if (nmp->nm_sotype != SOCK_DGRAM) {
1427 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1428 				rep->r_rexmit = NFS_MAXREXMIT;
1429 			continue;
1430 		}
1431 		if ((so = nmp->nm_so) == NULL)
1432 			continue;
1433 
1434 		/*
1435 		 * If there is enough space and the window allows..
1436 		 *	Resend it
1437 		 * Set r_rtt to -1 in case we fail to send it now.
1438 		 */
1439 		rep->r_rtt = -1;
1440 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1441 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1442 		    (rep->r_flags & R_SENT) ||
1443 		    nmp->nm_sent < nmp->nm_cwnd) &&
1444 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1445 		        if (so->so_state & SS_ISCONNECTED)
1446 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1447 			    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1448 			else
1449 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1450 			    nmp->nm_nam, (struct mbuf *)0, (struct proc *)0);
1451 			if (error) {
1452 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1453 #ifdef DEBUG
1454 					printf("nfs_timer: ignoring error %d\n",
1455 						error);
1456 #endif
1457 					so->so_error = 0;
1458 				}
1459 			} else {
1460 				/*
1461 				 * Iff first send, start timing
1462 				 * else turn timing off, backoff timer
1463 				 * and divide congestion window by 2.
1464 				 */
1465 				if (rep->r_flags & R_SENT) {
1466 					rep->r_flags &= ~R_TIMING;
1467 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1468 						rep->r_rexmit = NFS_MAXREXMIT;
1469 					nmp->nm_cwnd >>= 1;
1470 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1471 						nmp->nm_cwnd = NFS_CWNDSCALE;
1472 					nfsstats.rpcretries++;
1473 				} else {
1474 					rep->r_flags |= R_SENT;
1475 					nmp->nm_sent += NFS_CWNDSCALE;
1476 				}
1477 				rep->r_rtt = 0;
1478 			}
1479 		}
1480 	}
1481 
1482 #ifdef NFSSERVER
1483 	/*
1484 	 * Call the nqnfs server timer once a second to handle leases.
1485 	 */
1486 	if (lasttime != time.tv_sec) {
1487 		lasttime = time.tv_sec;
1488 		nqnfs_serverd();
1489 	}
1490 
1491 	/*
1492 	 * Scan the write gathering queues for writes that need to be
1493 	 * completed now.
1494 	 */
1495 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1496 	TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
1497 	    if (LIST_FIRST(&slp->ns_tq) &&
1498 		LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec)
1499 		nfsrv_wakenfsd(slp);
1500 	}
1501 #endif /* NFSSERVER */
1502 	splx(s);
1503 	callout_schedule(&nfs_timer_ch, nfs_ticks);
1504 }
1505 
1506 /*ARGSUSED*/
1507 void
1508 nfs_exit(p, v)
1509 	struct proc *p;
1510 	void *v;
1511 {
1512 	struct nfsreq *rp;
1513 	int s = splsoftnet();
1514 
1515 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
1516 		if (rp->r_procp == p)
1517 			TAILQ_REMOVE(&nfs_reqq, rp, r_chain);
1518 	}
1519 	splx(s);
1520 }
1521 
1522 /*
1523  * Test for a termination condition pending on the process.
1524  * This is used for NFSMNT_INT mounts.
1525  */
1526 int
1527 nfs_sigintr(nmp, rep, p)
1528 	struct nfsmount *nmp;
1529 	struct nfsreq *rep;
1530 	struct proc *p;
1531 {
1532 	sigset_t ss;
1533 
1534 	if (rep && (rep->r_flags & R_SOFTTERM))
1535 		return (EINTR);
1536 	if (!(nmp->nm_flag & NFSMNT_INT))
1537 		return (0);
1538 	if (p) {
1539 		sigpending1(p, &ss);
1540 #if 0
1541 		sigminusset(&p->p_sigctx.ps_sigignore, &ss);
1542 #endif
1543 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1544 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1545 		    sigismember(&ss, SIGQUIT))
1546 			return (EINTR);
1547 	}
1548 	return (0);
1549 }
1550 
1551 /*
1552  * Lock a socket against others.
1553  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1554  * and also to avoid race conditions between the processes with nfs requests
1555  * in progress when a reconnect is necessary.
1556  */
1557 int
1558 nfs_sndlock(flagp, rep)
1559 	int *flagp;
1560 	struct nfsreq *rep;
1561 {
1562 	struct proc *p;
1563 	int slpflag = 0, slptimeo = 0;
1564 
1565 	if (rep) {
1566 		p = rep->r_procp;
1567 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1568 			slpflag = PCATCH;
1569 	} else
1570 		p = (struct proc *)0;
1571 	while (*flagp & NFSMNT_SNDLOCK) {
1572 		if (nfs_sigintr(rep->r_nmp, rep, p))
1573 			return (EINTR);
1574 		*flagp |= NFSMNT_WANTSND;
1575 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1576 			slptimeo);
1577 		if (slpflag == PCATCH) {
1578 			slpflag = 0;
1579 			slptimeo = 2 * hz;
1580 		}
1581 	}
1582 	*flagp |= NFSMNT_SNDLOCK;
1583 	return (0);
1584 }
1585 
1586 /*
1587  * Unlock the stream socket for others.
1588  */
1589 void
1590 nfs_sndunlock(flagp)
1591 	int *flagp;
1592 {
1593 
1594 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1595 		panic("nfs sndunlock");
1596 	*flagp &= ~NFSMNT_SNDLOCK;
1597 	if (*flagp & NFSMNT_WANTSND) {
1598 		*flagp &= ~NFSMNT_WANTSND;
1599 		wakeup((caddr_t)flagp);
1600 	}
1601 }
1602 
1603 int
1604 nfs_rcvlock(rep)
1605 	struct nfsreq *rep;
1606 {
1607 	struct nfsmount *nmp = rep->r_nmp;
1608 	int *flagp = &nmp->nm_iflag;
1609 	int slpflag, slptimeo = 0;
1610 	int error = 0;
1611 
1612 	if (*flagp & NFSMNT_DISMNT)
1613 		return EIO;
1614 
1615 	if (*flagp & NFSMNT_INT)
1616 		slpflag = PCATCH;
1617 	else
1618 		slpflag = 0;
1619 	simple_lock(&nmp->nm_slock);
1620 	while (*flagp & NFSMNT_RCVLOCK) {
1621 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) {
1622 			error = EINTR;
1623 			goto quit;
1624 		}
1625 		*flagp |= NFSMNT_WANTRCV;
1626 		nmp->nm_waiters++;
1627 		(void) ltsleep(flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1628 			slptimeo, &nmp->nm_slock);
1629 		nmp->nm_waiters--;
1630 		if (*flagp & NFSMNT_DISMNT) {
1631 			wakeup(&nmp->nm_waiters);
1632 			error = EIO;
1633 			goto quit;
1634 		}
1635 		/* If our reply was received while we were sleeping,
1636 		 * then just return without taking the lock to avoid a
1637 		 * situation where a single iod could 'capture' the
1638 		 * receive lock.
1639 		 */
1640 		if (rep->r_mrep != NULL) {
1641 			error = EALREADY;
1642 			goto quit;
1643 		}
1644 		if (slpflag == PCATCH) {
1645 			slpflag = 0;
1646 			slptimeo = 2 * hz;
1647 		}
1648 	}
1649 	*flagp |= NFSMNT_RCVLOCK;
1650 quit:
1651 	simple_unlock(&nmp->nm_slock);
1652 	return error;
1653 }
1654 
1655 /*
1656  * Unlock the stream socket for others.
1657  */
1658 void
1659 nfs_rcvunlock(nmp)
1660 	struct nfsmount *nmp;
1661 {
1662 	int *flagp = &nmp->nm_iflag;
1663 
1664 	simple_lock(&nmp->nm_slock);
1665 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1666 		panic("nfs rcvunlock");
1667 	*flagp &= ~NFSMNT_RCVLOCK;
1668 	if (*flagp & NFSMNT_WANTRCV) {
1669 		*flagp &= ~NFSMNT_WANTRCV;
1670 		wakeup((caddr_t)flagp);
1671 	}
1672 	simple_unlock(&nmp->nm_slock);
1673 }
1674 
1675 /*
1676  * Parse an RPC request
1677  * - verify it
1678  * - fill in the cred struct.
1679  */
1680 int
1681 nfs_getreq(nd, nfsd, has_header)
1682 	struct nfsrv_descript *nd;
1683 	struct nfsd *nfsd;
1684 	int has_header;
1685 {
1686 	int len, i;
1687 	u_int32_t *tl;
1688 	int32_t t1;
1689 	struct uio uio;
1690 	struct iovec iov;
1691 	caddr_t dpos, cp2, cp;
1692 	u_int32_t nfsvers, auth_type;
1693 	uid_t nickuid;
1694 	int error = 0, nqnfs = 0, ticklen;
1695 	struct mbuf *mrep, *md;
1696 	struct nfsuid *nuidp;
1697 	struct timeval tvin, tvout;
1698 
1699 	mrep = nd->nd_mrep;
1700 	md = nd->nd_md;
1701 	dpos = nd->nd_dpos;
1702 	if (has_header) {
1703 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1704 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1705 		if (*tl++ != rpc_call) {
1706 			m_freem(mrep);
1707 			return (EBADRPC);
1708 		}
1709 	} else
1710 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1711 	nd->nd_repstat = 0;
1712 	nd->nd_flag = 0;
1713 	if (*tl++ != rpc_vers) {
1714 		nd->nd_repstat = ERPCMISMATCH;
1715 		nd->nd_procnum = NFSPROC_NOOP;
1716 		return (0);
1717 	}
1718 	if (*tl != nfs_prog) {
1719 		if (*tl == nqnfs_prog)
1720 			nqnfs++;
1721 		else {
1722 			nd->nd_repstat = EPROGUNAVAIL;
1723 			nd->nd_procnum = NFSPROC_NOOP;
1724 			return (0);
1725 		}
1726 	}
1727 	tl++;
1728 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1729 	if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1730 		(nfsvers != NQNFS_VER3 && nqnfs)) {
1731 		nd->nd_repstat = EPROGMISMATCH;
1732 		nd->nd_procnum = NFSPROC_NOOP;
1733 		return (0);
1734 	}
1735 	if (nqnfs)
1736 		nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1737 	else if (nfsvers == NFS_VER3)
1738 		nd->nd_flag = ND_NFSV3;
1739 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1740 	if (nd->nd_procnum == NFSPROC_NULL)
1741 		return (0);
1742 	if (nd->nd_procnum >= NFS_NPROCS ||
1743 		(!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1744 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1745 		nd->nd_repstat = EPROCUNAVAIL;
1746 		nd->nd_procnum = NFSPROC_NOOP;
1747 		return (0);
1748 	}
1749 	if ((nd->nd_flag & ND_NFSV3) == 0)
1750 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1751 	auth_type = *tl++;
1752 	len = fxdr_unsigned(int, *tl++);
1753 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1754 		m_freem(mrep);
1755 		return (EBADRPC);
1756 	}
1757 
1758 	nd->nd_flag &= ~ND_KERBAUTH;
1759 	/*
1760 	 * Handle auth_unix or auth_kerb.
1761 	 */
1762 	if (auth_type == rpc_auth_unix) {
1763 		len = fxdr_unsigned(int, *++tl);
1764 		if (len < 0 || len > NFS_MAXNAMLEN) {
1765 			m_freem(mrep);
1766 			return (EBADRPC);
1767 		}
1768 		nfsm_adv(nfsm_rndup(len));
1769 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1770 		memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred));
1771 		nd->nd_cr.cr_ref = 1;
1772 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1773 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1774 		len = fxdr_unsigned(int, *tl);
1775 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1776 			m_freem(mrep);
1777 			return (EBADRPC);
1778 		}
1779 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1780 		for (i = 0; i < len; i++)
1781 		    if (i < NGROUPS)
1782 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1783 		    else
1784 			tl++;
1785 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1786 		if (nd->nd_cr.cr_ngroups > 1)
1787 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1788 		len = fxdr_unsigned(int, *++tl);
1789 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1790 			m_freem(mrep);
1791 			return (EBADRPC);
1792 		}
1793 		if (len > 0)
1794 			nfsm_adv(nfsm_rndup(len));
1795 	} else if (auth_type == rpc_auth_kerb) {
1796 		switch (fxdr_unsigned(int, *tl++)) {
1797 		case RPCAKN_FULLNAME:
1798 			ticklen = fxdr_unsigned(int, *tl);
1799 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1800 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1801 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1802 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1803 				m_freem(mrep);
1804 				return (EBADRPC);
1805 			}
1806 			uio.uio_offset = 0;
1807 			uio.uio_iov = &iov;
1808 			uio.uio_iovcnt = 1;
1809 			uio.uio_segflg = UIO_SYSSPACE;
1810 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1811 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1812 			nfsm_mtouio(&uio, uio.uio_resid);
1813 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1814 			if (*tl++ != rpc_auth_kerb ||
1815 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1816 				printf("Bad kerb verifier\n");
1817 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1818 				nd->nd_procnum = NFSPROC_NOOP;
1819 				return (0);
1820 			}
1821 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1822 			tl = (u_int32_t *)cp;
1823 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1824 				printf("Not fullname kerb verifier\n");
1825 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1826 				nd->nd_procnum = NFSPROC_NOOP;
1827 				return (0);
1828 			}
1829 			cp += NFSX_UNSIGNED;
1830 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
1831 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1832 			nd->nd_flag |= ND_KERBFULL;
1833 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1834 			break;
1835 		case RPCAKN_NICKNAME:
1836 			if (len != 2 * NFSX_UNSIGNED) {
1837 				printf("Kerb nickname short\n");
1838 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1839 				nd->nd_procnum = NFSPROC_NOOP;
1840 				return (0);
1841 			}
1842 			nickuid = fxdr_unsigned(uid_t, *tl);
1843 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1844 			if (*tl++ != rpc_auth_kerb ||
1845 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1846 				printf("Kerb nick verifier bad\n");
1847 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1848 				nd->nd_procnum = NFSPROC_NOOP;
1849 				return (0);
1850 			}
1851 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1852 			tvin.tv_sec = *tl++;
1853 			tvin.tv_usec = *tl;
1854 
1855 			LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid),
1856 			    nu_hash) {
1857 				if (nuidp->nu_cr.cr_uid == nickuid &&
1858 				    (!nd->nd_nam2 ||
1859 				     netaddr_match(NU_NETFAM(nuidp),
1860 				      &nuidp->nu_haddr, nd->nd_nam2)))
1861 					break;
1862 			}
1863 			if (!nuidp) {
1864 				nd->nd_repstat =
1865 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1866 				nd->nd_procnum = NFSPROC_NOOP;
1867 				return (0);
1868 			}
1869 
1870 			/*
1871 			 * Now, decrypt the timestamp using the session key
1872 			 * and validate it.
1873 			 */
1874 #ifdef NFSKERB
1875 			XXX
1876 #endif
1877 
1878 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1879 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1880 			if (nuidp->nu_expire < time.tv_sec ||
1881 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1882 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1883 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1884 				nuidp->nu_expire = 0;
1885 				nd->nd_repstat =
1886 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1887 				nd->nd_procnum = NFSPROC_NOOP;
1888 				return (0);
1889 			}
1890 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1891 			nd->nd_flag |= ND_KERBNICK;
1892 		};
1893 	} else {
1894 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1895 		nd->nd_procnum = NFSPROC_NOOP;
1896 		return (0);
1897 	}
1898 
1899 	/*
1900 	 * For nqnfs, get piggybacked lease request.
1901 	 */
1902 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
1903 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1904 		nd->nd_flag |= fxdr_unsigned(int, *tl);
1905 		if (nd->nd_flag & ND_LEASE) {
1906 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1907 			nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
1908 		} else
1909 			nd->nd_duration = NQ_MINLEASE;
1910 	} else
1911 		nd->nd_duration = NQ_MINLEASE;
1912 	nd->nd_md = md;
1913 	nd->nd_dpos = dpos;
1914 	return (0);
1915 nfsmout:
1916 	return (error);
1917 }
1918 
1919 int
1920 nfs_msg(p, server, msg)
1921 	struct proc *p;
1922 	char *server, *msg;
1923 {
1924 	tpr_t tpr;
1925 
1926 	if (p)
1927 		tpr = tprintf_open(p);
1928 	else
1929 		tpr = NULL;
1930 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1931 	tprintf_close(tpr);
1932 	return (0);
1933 }
1934 
1935 #ifdef NFSSERVER
1936 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
1937 				    struct nfssvc_sock *, struct proc *,
1938 				    struct mbuf **)) = {
1939 	nfsrv_null,
1940 	nfsrv_getattr,
1941 	nfsrv_setattr,
1942 	nfsrv_lookup,
1943 	nfsrv3_access,
1944 	nfsrv_readlink,
1945 	nfsrv_read,
1946 	nfsrv_write,
1947 	nfsrv_create,
1948 	nfsrv_mkdir,
1949 	nfsrv_symlink,
1950 	nfsrv_mknod,
1951 	nfsrv_remove,
1952 	nfsrv_rmdir,
1953 	nfsrv_rename,
1954 	nfsrv_link,
1955 	nfsrv_readdir,
1956 	nfsrv_readdirplus,
1957 	nfsrv_statfs,
1958 	nfsrv_fsinfo,
1959 	nfsrv_pathconf,
1960 	nfsrv_commit,
1961 	nqnfsrv_getlease,
1962 	nqnfsrv_vacated,
1963 	nfsrv_noop,
1964 	nfsrv_noop
1965 };
1966 
1967 /*
1968  * Socket upcall routine for the nfsd sockets.
1969  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1970  * Essentially do as much as possible non-blocking, else punt and it will
1971  * be called with M_WAIT from an nfsd.
1972  */
1973 void
1974 nfsrv_rcv(so, arg, waitflag)
1975 	struct socket *so;
1976 	caddr_t arg;
1977 	int waitflag;
1978 {
1979 	struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1980 	struct mbuf *m;
1981 	struct mbuf *mp, *nam;
1982 	struct uio auio;
1983 	int flags, error;
1984 
1985 	if ((slp->ns_flag & SLP_VALID) == 0)
1986 		return;
1987 #ifdef notdef
1988 	/*
1989 	 * Define this to test for nfsds handling this under heavy load.
1990 	 */
1991 	if (waitflag == M_DONTWAIT) {
1992 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1993 	}
1994 #endif
1995 	auio.uio_procp = NULL;
1996 	if (so->so_type == SOCK_STREAM) {
1997 		/*
1998 		 * If there are already records on the queue, defer soreceive()
1999 		 * to an nfsd so that there is feedback to the TCP layer that
2000 		 * the nfs servers are heavily loaded.
2001 		 */
2002 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
2003 			slp->ns_flag |= SLP_NEEDQ;
2004 			goto dorecs;
2005 		}
2006 
2007 		/*
2008 		 * Do soreceive().
2009 		 */
2010 		auio.uio_resid = 1000000000;
2011 		flags = MSG_DONTWAIT;
2012 		error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
2013 		if (error || mp == (struct mbuf *)0) {
2014 			if (error == EWOULDBLOCK)
2015 				slp->ns_flag |= SLP_NEEDQ;
2016 			else
2017 				slp->ns_flag |= SLP_DISCONN;
2018 			goto dorecs;
2019 		}
2020 		m = mp;
2021 		if (slp->ns_rawend) {
2022 			slp->ns_rawend->m_next = m;
2023 			slp->ns_cc += 1000000000 - auio.uio_resid;
2024 		} else {
2025 			slp->ns_raw = m;
2026 			slp->ns_cc = 1000000000 - auio.uio_resid;
2027 		}
2028 		while (m->m_next)
2029 			m = m->m_next;
2030 		slp->ns_rawend = m;
2031 
2032 		/*
2033 		 * Now try and parse record(s) out of the raw stream data.
2034 		 */
2035 		error = nfsrv_getstream(slp, waitflag);
2036 		if (error) {
2037 			if (error == EPERM)
2038 				slp->ns_flag |= SLP_DISCONN;
2039 			else
2040 				slp->ns_flag |= SLP_NEEDQ;
2041 		}
2042 	} else {
2043 		do {
2044 			auio.uio_resid = 1000000000;
2045 			flags = MSG_DONTWAIT;
2046 			error = (*so->so_receive)(so, &nam, &auio, &mp,
2047 						(struct mbuf **)0, &flags);
2048 			if (mp) {
2049 				if (nam) {
2050 					m = nam;
2051 					m->m_next = mp;
2052 				} else
2053 					m = mp;
2054 				if (slp->ns_recend)
2055 					slp->ns_recend->m_nextpkt = m;
2056 				else
2057 					slp->ns_rec = m;
2058 				slp->ns_recend = m;
2059 				m->m_nextpkt = (struct mbuf *)0;
2060 			}
2061 			if (error) {
2062 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2063 					&& error != EWOULDBLOCK) {
2064 					slp->ns_flag |= SLP_DISCONN;
2065 					goto dorecs;
2066 				}
2067 			}
2068 		} while (mp);
2069 	}
2070 
2071 	/*
2072 	 * Now try and process the request records, non-blocking.
2073 	 */
2074 dorecs:
2075 	if (waitflag == M_DONTWAIT &&
2076 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
2077 		nfsrv_wakenfsd(slp);
2078 }
2079 
2080 /*
2081  * Try and extract an RPC request from the mbuf data list received on a
2082  * stream socket. The "waitflag" argument indicates whether or not it
2083  * can sleep.
2084  */
2085 int
2086 nfsrv_getstream(slp, waitflag)
2087 	struct nfssvc_sock *slp;
2088 	int waitflag;
2089 {
2090 	struct mbuf *m, **mpp;
2091 	struct mbuf *recm;
2092 	u_int32_t recmark;
2093 
2094 	if (slp->ns_flag & SLP_GETSTREAM)
2095 		panic("nfs getstream");
2096 	slp->ns_flag |= SLP_GETSTREAM;
2097 	for (;;) {
2098 		if (slp->ns_reclen == 0) {
2099 			if (slp->ns_cc < NFSX_UNSIGNED) {
2100 				slp->ns_flag &= ~SLP_GETSTREAM;
2101 				return (0);
2102 			}
2103 			m = slp->ns_raw;
2104 			m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark);
2105 			m_adj(m, NFSX_UNSIGNED);
2106 			slp->ns_cc -= NFSX_UNSIGNED;
2107 			recmark = ntohl(recmark);
2108 			slp->ns_reclen = recmark & ~0x80000000;
2109 			if (recmark & 0x80000000)
2110 				slp->ns_flag |= SLP_LASTFRAG;
2111 			else
2112 				slp->ns_flag &= ~SLP_LASTFRAG;
2113 			if (slp->ns_reclen > NFS_MAXPACKET) {
2114 				slp->ns_flag &= ~SLP_GETSTREAM;
2115 				return (EPERM);
2116 			}
2117 		}
2118 
2119 		/*
2120 		 * Now get the record part.
2121 		 *
2122 		 * Note that slp->ns_reclen may be 0.  Linux sometimes
2123 		 * generates 0-length records.
2124 		 */
2125 		if (slp->ns_cc == slp->ns_reclen) {
2126 			recm = slp->ns_raw;
2127 			slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2128 			slp->ns_cc = slp->ns_reclen = 0;
2129 		} else if (slp->ns_cc > slp->ns_reclen) {
2130 			recm = slp->ns_raw;
2131 			m = m_split(recm, slp->ns_reclen, waitflag);
2132 			if (m == NULL) {
2133 				slp->ns_flag &= ~SLP_GETSTREAM;
2134 				return (EWOULDBLOCK);
2135 			}
2136 			m_claim(recm, &nfs_mowner);
2137 			slp->ns_raw = m;
2138 			if (m->m_next == NULL)
2139 				slp->ns_rawend = m;
2140 			slp->ns_cc -= slp->ns_reclen;
2141 			slp->ns_reclen = 0;
2142 		} else {
2143 			slp->ns_flag &= ~SLP_GETSTREAM;
2144 			return (0);
2145 		}
2146 
2147 		/*
2148 		 * Accumulate the fragments into a record.
2149 		 */
2150 		mpp = &slp->ns_frag;
2151 		while (*mpp)
2152 			mpp = &((*mpp)->m_next);
2153 		*mpp = recm;
2154 		if (slp->ns_flag & SLP_LASTFRAG) {
2155 			if (slp->ns_recend)
2156 				slp->ns_recend->m_nextpkt = slp->ns_frag;
2157 			else
2158 				slp->ns_rec = slp->ns_frag;
2159 			slp->ns_recend = slp->ns_frag;
2160 			slp->ns_frag = (struct mbuf *)0;
2161 		}
2162 	}
2163 }
2164 
2165 /*
2166  * Parse an RPC header.
2167  */
2168 int
2169 nfsrv_dorec(slp, nfsd, ndp)
2170 	struct nfssvc_sock *slp;
2171 	struct nfsd *nfsd;
2172 	struct nfsrv_descript **ndp;
2173 {
2174 	struct mbuf *m, *nam;
2175 	struct nfsrv_descript *nd;
2176 	int error;
2177 
2178 	*ndp = NULL;
2179 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2180 	    (m = slp->ns_rec) == (struct mbuf *)0)
2181 		return (ENOBUFS);
2182 	slp->ns_rec = m->m_nextpkt;
2183 	if (slp->ns_rec)
2184 		m->m_nextpkt = (struct mbuf *)0;
2185 	else
2186 		slp->ns_recend = (struct mbuf *)0;
2187 	if (m->m_type == MT_SONAME) {
2188 		nam = m;
2189 		m = m->m_next;
2190 		nam->m_next = NULL;
2191 	} else
2192 		nam = NULL;
2193 	nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK);
2194 	nd->nd_md = nd->nd_mrep = m;
2195 	nd->nd_nam2 = nam;
2196 	nd->nd_dpos = mtod(m, caddr_t);
2197 	error = nfs_getreq(nd, nfsd, TRUE);
2198 	if (error) {
2199 		m_freem(nam);
2200 		pool_put(&nfs_srvdesc_pool, nd);
2201 		return (error);
2202 	}
2203 	*ndp = nd;
2204 	nfsd->nfsd_nd = nd;
2205 	return (0);
2206 }
2207 
2208 
2209 /*
2210  * Search for a sleeping nfsd and wake it up.
2211  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2212  * running nfsds will go look for the work in the nfssvc_sock list.
2213  */
2214 void
2215 nfsrv_wakenfsd(slp)
2216 	struct nfssvc_sock *slp;
2217 {
2218 	struct nfsd *nd;
2219 
2220 	if ((slp->ns_flag & SLP_VALID) == 0)
2221 		return;
2222 	simple_lock(&nfsd_slock);
2223 	if (slp->ns_flag & SLP_DOREC) {
2224 		simple_unlock(&nfsd_slock);
2225 		return;
2226 	}
2227 	nd = SLIST_FIRST(&nfsd_idle_head);
2228 	if (nd) {
2229 		SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle);
2230 		simple_unlock(&nfsd_slock);
2231 
2232 		KASSERT(nd->nfsd_flag & NFSD_WAITING);
2233 		nd->nfsd_flag &= ~NFSD_WAITING;
2234 		if (nd->nfsd_slp)
2235 			panic("nfsd wakeup");
2236 		slp->ns_sref++;
2237 		nd->nfsd_slp = slp;
2238 		wakeup(nd);
2239 		return;
2240 	}
2241 	slp->ns_flag |= SLP_DOREC;
2242 	nfsd_head_flag |= NFSD_CHECKSLP;
2243 	TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending);
2244 	simple_unlock(&nfsd_slock);
2245 }
2246 #endif /* NFSSERVER */
2247