xref: /openbsd-src/sys/nfs/nfs_socket.c (revision 47911bd667ac77dc523b8a13ef40b012dbffa741)
1 /*	$OpenBSD: nfs_socket.c,v 1.32 2002/10/29 12:47:06 art Exp $	*/
2 /*	$NetBSD: nfs_socket.c,v 1.27 1996/04/15 20:20:00 thorpej Exp $	*/
3 
4 /*
5  * Copyright (c) 1989, 1991, 1993, 1995
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * Rick Macklem at The University of Guelph.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the University of
22  *	California, Berkeley and its contributors.
23  * 4. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  *
39  *	@(#)nfs_socket.c	8.5 (Berkeley) 3/30/95
40  */
41 
42 /*
43  * Socket operations for use by nfs
44  */
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/kernel.h>
51 #include <sys/mbuf.h>
52 #include <sys/vnode.h>
53 #include <sys/domain.h>
54 #include <sys/protosw.h>
55 #include <sys/socket.h>
56 #include <sys/socketvar.h>
57 #include <sys/syslog.h>
58 #include <sys/tprintf.h>
59 #include <sys/namei.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/tcp.h>
63 
64 #include <nfs/rpcv2.h>
65 #include <nfs/nfsproto.h>
66 #include <nfs/nfs.h>
67 #include <nfs/xdr_subs.h>
68 #include <nfs/nfsm_subs.h>
69 #include <nfs/nfsmount.h>
70 #include <nfs/nfsnode.h>
71 #include <nfs/nfsrtt.h>
72 #include <nfs/nfs_var.h>
73 
74 #define	TRUE	1
75 #define	FALSE	0
76 
77 /*
78  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
79  * Use the mean and mean deviation of rtt for the appropriate type of rpc
80  * for the frequent rpcs and a default for the others.
81  * The justification for doing "other" this way is that these rpcs
82  * happen so infrequently that timer est. would probably be stale.
83  * Also, since many of these rpcs are
84  * non-idempotent, a conservative timeout is desired.
85  * getattr, lookup - A+2D
86  * read, write     - A+4D
87  * other           - nm_timeo
88  */
89 #define	NFS_RTO(n, t) \
90 	((t) == 0 ? (n)->nm_timeo : \
91 	 ((t) < 3 ? \
92 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
93 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
94 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
95 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
96 /*
97  * External data, mostly RPC constants in XDR form
98  */
99 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
100 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
101 	rpc_auth_kerb;
102 extern u_int32_t nfs_prog;
103 extern struct nfsstats nfsstats;
104 extern int nfsv3_procid[NFS_NPROCS];
105 extern int nfs_ticks;
106 
107 /*
108  * Defines which timer to use for the procnum.
109  * 0 - default
110  * 1 - getattr
111  * 2 - lookup
112  * 3 - read
113  * 4 - write
114  */
115 static int proct[NFS_NPROCS] = {
116 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
117 	0, 0, 0,
118 };
119 
120 /*
121  * There is a congestion window for outstanding rpcs maintained per mount
122  * point. The cwnd size is adjusted in roughly the way that:
123  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
124  * SIGCOMM '88". ACM, August 1988.
125  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
126  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
127  * of rpcs is in progress.
128  * (The sent count and cwnd are scaled for integer arith.)
129  * Variants of "slow start" were tried and were found to be too much of a
130  * performance hit (ave. rtt 3 times larger),
131  * I suspect due to the large rtt that nfs rpcs have.
132  */
133 #define	NFS_CWNDSCALE	256
134 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
135 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
136 int nfsrtton = 0;
137 struct nfsrtt nfsrtt;
138 
139 void nfs_realign(struct mbuf **, int);
140 unsigned int nfs_realign_test = 0;
141 unsigned int nfs_realign_count = 0;
142 
143 struct nfsreqhead nfs_reqq;
144 
145 /*
146  * Initialize sockets and congestion for a new NFS connection.
147  * We do not free the sockaddr if error.
148  */
149 int
150 nfs_connect(nmp, rep)
151 	struct nfsmount *nmp;
152 	struct nfsreq *rep;
153 {
154 	struct socket *so;
155 	int s, error, rcvreserve, sndreserve;
156 	struct sockaddr *saddr;
157 	struct sockaddr_in *sin;
158 	struct mbuf *m;
159 
160 	nmp->nm_so = (struct socket *)0;
161 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
162 	error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
163 		nmp->nm_soproto);
164 	if (error)
165 		goto bad;
166 	so = nmp->nm_so;
167 	nmp->nm_soflags = so->so_proto->pr_flags;
168 
169 	/*
170 	 * Some servers require that the client port be a reserved port number.
171 	 * We always allocate a reserved port, as this prevents filehandle
172 	 * disclosure through UDP port capture.
173 	 */
174 	if (saddr->sa_family == AF_INET) {
175 		struct mbuf *mopt;
176 		int *ip;
177 
178 		MGET(mopt, M_WAIT, MT_SOOPTS);
179 		mopt->m_len = sizeof(int);
180 		ip = mtod(mopt, int *);
181 		*ip = IP_PORTRANGE_LOW;
182 		error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt);
183 		if (error)
184 			goto bad;
185 
186 		MGET(m, M_WAIT, MT_SONAME);
187 		sin = mtod(m, struct sockaddr_in *);
188 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
189 		sin->sin_family = AF_INET;
190 		sin->sin_addr.s_addr = INADDR_ANY;
191 		sin->sin_port = htons(0);
192 		error = sobind(so, m);
193 		m_freem(m);
194 		if (error)
195 			goto bad;
196 
197 		MGET(mopt, M_WAIT, MT_SOOPTS);
198 		mopt->m_len = sizeof(int);
199 		ip = mtod(mopt, int *);
200 		*ip = IP_PORTRANGE_DEFAULT;
201 		error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt);
202 		if (error)
203 			goto bad;
204 	}
205 
206 	/*
207 	 * Protocols that do not require connections may be optionally left
208 	 * unconnected for servers that reply from a port other than NFS_PORT.
209 	 */
210 	if (nmp->nm_flag & NFSMNT_NOCONN) {
211 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
212 			error = ENOTCONN;
213 			goto bad;
214 		}
215 	} else {
216 		error = soconnect(so, nmp->nm_nam);
217 		if (error)
218 			goto bad;
219 
220 		/*
221 		 * Wait for the connection to complete. Cribbed from the
222 		 * connect system call but with the wait timing out so
223 		 * that interruptible mounts don't hang here for a long time.
224 		 */
225 		s = splsoftnet();
226 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
227 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
228 				"nfscon", 2 * hz);
229 			if ((so->so_state & SS_ISCONNECTING) &&
230 			    so->so_error == 0 && rep &&
231 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
232 				so->so_state &= ~SS_ISCONNECTING;
233 				splx(s);
234 				goto bad;
235 			}
236 		}
237 		if (so->so_error) {
238 			error = so->so_error;
239 			so->so_error = 0;
240 			splx(s);
241 			goto bad;
242 		}
243 		splx(s);
244 	}
245 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
246 		so->so_rcv.sb_timeo = (5 * hz);
247 		so->so_snd.sb_timeo = (5 * hz);
248 	} else {
249 		so->so_rcv.sb_timeo = 0;
250 		so->so_snd.sb_timeo = 0;
251 	}
252 	if (nmp->nm_sotype == SOCK_DGRAM) {
253 		sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR;
254 		rcvreserve = max(nmp->nm_rsize, nmp->nm_readdirsize) +
255 		    NFS_MAXPKTHDR;
256 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
257 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
258 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
259 		    NFS_MAXPKTHDR) * 2;
260 	} else {
261 		if (nmp->nm_sotype != SOCK_STREAM)
262 			panic("nfscon sotype");
263 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
264 			MGET(m, M_WAIT, MT_SOOPTS);
265 			*mtod(m, int32_t *) = 1;
266 			m->m_len = sizeof(int32_t);
267 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
268 		}
269 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
270 			MGET(m, M_WAIT, MT_SOOPTS);
271 			*mtod(m, int32_t *) = 1;
272 			m->m_len = sizeof(int32_t);
273 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
274 		}
275 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
276 		    sizeof (u_int32_t)) * 2;
277 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
278 		    sizeof (u_int32_t)) * 2;
279 	}
280 	error = soreserve(so, sndreserve, rcvreserve);
281 	if (error)
282 		goto bad;
283 	so->so_rcv.sb_flags |= SB_NOINTR;
284 	so->so_snd.sb_flags |= SB_NOINTR;
285 
286 	/* Initialize other non-zero congestion variables */
287 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
288 		nmp->nm_srtt[4] = (NFS_TIMEO << 3);
289 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
290 		nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0;
291 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
292 	nmp->nm_sent = 0;
293 	nmp->nm_timeouts = 0;
294 	return (0);
295 
296 bad:
297 	nfs_disconnect(nmp);
298 	return (error);
299 }
300 
301 /*
302  * Reconnect routine:
303  * Called when a connection is broken on a reliable protocol.
304  * - clean up the old socket
305  * - nfs_connect() again
306  * - set R_MUSTRESEND for all outstanding requests on mount point
307  * If this fails the mount point is DEAD!
308  * nb: Must be called with the nfs_sndlock() set on the mount point.
309  */
310 int
311 nfs_reconnect(rep)
312 	struct nfsreq *rep;
313 {
314 	struct nfsreq *rp;
315 	struct nfsmount *nmp = rep->r_nmp;
316 	int error;
317 
318 	nfs_disconnect(nmp);
319 	while ((error = nfs_connect(nmp, rep)) != 0) {
320 		if (error == EINTR || error == ERESTART)
321 			return (EINTR);
322 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0);
323 	}
324 
325 	/*
326 	 * Loop through outstanding request list and fix up all requests
327 	 * on old socket.
328 	 */
329 	for (rp = TAILQ_FIRST(&nfs_reqq); rp != NULL;
330 	    rp = TAILQ_NEXT(rp, r_chain)) {
331 		if (rp->r_nmp == nmp)
332 			rp->r_flags |= R_MUSTRESEND;
333 	}
334 	return (0);
335 }
336 
337 /*
338  * NFS disconnect. Clean up and unlink.
339  */
340 void
341 nfs_disconnect(nmp)
342 	struct nfsmount *nmp;
343 {
344 	struct socket *so;
345 
346 	if (nmp->nm_so) {
347 		so = nmp->nm_so;
348 		nmp->nm_so = (struct socket *)0;
349 		soshutdown(so, 2);
350 		soclose(so);
351 	}
352 }
353 
354 /*
355  * This is the nfs send routine. For connection based socket types, it
356  * must be called with an nfs_sndlock() on the socket.
357  * "rep == NULL" indicates that it has been called from a server.
358  * For the client side:
359  * - return EINTR if the RPC is terminated, 0 otherwise
360  * - set R_MUSTRESEND if the send fails for any reason
361  * - do any cleanup required by recoverable socket errors (???)
362  * For the server side:
363  * - return EINTR or ERESTART if interrupted by a signal
364  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
365  * - do any cleanup required by recoverable socket errors (???)
366  */
367 int
368 nfs_send(so, nam, top, rep)
369 	struct socket *so;
370 	struct mbuf *nam;
371 	struct mbuf *top;
372 	struct nfsreq *rep;
373 {
374 	struct mbuf *sendnam;
375 	int error, soflags, flags;
376 
377 	if (rep) {
378 		if (rep->r_flags & R_SOFTTERM) {
379 			m_freem(top);
380 			return (EINTR);
381 		}
382 		if ((so = rep->r_nmp->nm_so) == NULL) {
383 			rep->r_flags |= R_MUSTRESEND;
384 			m_freem(top);
385 			return (0);
386 		}
387 		rep->r_flags &= ~R_MUSTRESEND;
388 		soflags = rep->r_nmp->nm_soflags;
389 	} else
390 		soflags = so->so_proto->pr_flags;
391 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
392 		sendnam = (struct mbuf *)0;
393 	else
394 		sendnam = nam;
395 	if (so->so_type == SOCK_SEQPACKET)
396 		flags = MSG_EOR;
397 	else
398 		flags = 0;
399 
400 	error = sosend(so, sendnam, (struct uio *)0, top,
401 		(struct mbuf *)0, flags);
402 	if (error) {
403 		if (rep) {
404 			log(LOG_INFO, "nfs send error %d for server %s\n",error,
405 			    rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
406 			/*
407 			 * Deal with errors for the client side.
408 			 */
409 			if (rep->r_flags & R_SOFTTERM)
410 				error = EINTR;
411 			else
412 				rep->r_flags |= R_MUSTRESEND;
413 		} else
414 			log(LOG_INFO, "nfsd send error %d\n", error);
415 
416 		/*
417 		 * Handle any recoverable (soft) socket errors here. (???)
418 		 */
419 		if (error != EINTR && error != ERESTART &&
420 			error != EWOULDBLOCK && error != EPIPE)
421 			error = 0;
422 	}
423 	return (error);
424 }
425 
426 #ifdef NFSCLIENT
427 /*
428  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
429  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
430  * Mark and consolidate the data into a new mbuf list.
431  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
432  *     small mbufs.
433  * For SOCK_STREAM we must be very careful to read an entire record once
434  * we have read any of it, even if the system call has been interrupted.
435  */
436 int
437 nfs_receive(rep, aname, mp)
438 	struct nfsreq *rep;
439 	struct mbuf **aname;
440 	struct mbuf **mp;
441 {
442 	struct socket *so;
443 	struct uio auio;
444 	struct iovec aio;
445 	struct mbuf *m;
446 	struct mbuf *control;
447 	u_int32_t len;
448 	struct mbuf **getnam;
449 	int error, sotype, rcvflg;
450 	struct proc *p = curproc;	/* XXX */
451 
452 	/*
453 	 * Set up arguments for soreceive()
454 	 */
455 	*mp = (struct mbuf *)0;
456 	*aname = (struct mbuf *)0;
457 	sotype = rep->r_nmp->nm_sotype;
458 
459 	/*
460 	 * For reliable protocols, lock against other senders/receivers
461 	 * in case a reconnect is necessary.
462 	 * For SOCK_STREAM, first get the Record Mark to find out how much
463 	 * more there is to get.
464 	 * We must lock the socket against other receivers
465 	 * until we have an entire rpc request/reply.
466 	 */
467 	if (sotype != SOCK_DGRAM) {
468 		error = nfs_sndlock(&rep->r_nmp->nm_flag, rep);
469 		if (error)
470 			return (error);
471 tryagain:
472 		/*
473 		 * Check for fatal errors and resending request.
474 		 */
475 		/*
476 		 * Ugh: If a reconnect attempt just happened, nm_so
477 		 * would have changed. NULL indicates a failed
478 		 * attempt that has essentially shut down this
479 		 * mount point.
480 		 */
481 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
482 			nfs_sndunlock(&rep->r_nmp->nm_flag);
483 			return (EINTR);
484 		}
485 		so = rep->r_nmp->nm_so;
486 		if (!so) {
487 			error = nfs_reconnect(rep);
488 			if (error) {
489 				nfs_sndunlock(&rep->r_nmp->nm_flag);
490 				return (error);
491 			}
492 			goto tryagain;
493 		}
494 		while (rep->r_flags & R_MUSTRESEND) {
495 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
496 			nfsstats.rpcretries++;
497 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
498 			if (error) {
499 				if (error == EINTR || error == ERESTART ||
500 				    (error = nfs_reconnect(rep)) != 0) {
501 					nfs_sndunlock(&rep->r_nmp->nm_flag);
502 					return (error);
503 				}
504 				goto tryagain;
505 			}
506 		}
507 		nfs_sndunlock(&rep->r_nmp->nm_flag);
508 		if (sotype == SOCK_STREAM) {
509 			aio.iov_base = (caddr_t) &len;
510 			aio.iov_len = sizeof(u_int32_t);
511 			auio.uio_iov = &aio;
512 			auio.uio_iovcnt = 1;
513 			auio.uio_segflg = UIO_SYSSPACE;
514 			auio.uio_rw = UIO_READ;
515 			auio.uio_offset = 0;
516 			auio.uio_resid = sizeof(u_int32_t);
517 			auio.uio_procp = p;
518 			do {
519 			   rcvflg = MSG_WAITALL;
520 			   error = soreceive(so, (struct mbuf **)0, &auio,
521 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
522 			   if (error == EWOULDBLOCK && rep) {
523 				if (rep->r_flags & R_SOFTTERM)
524 					return (EINTR);
525 			   }
526 			} while (error == EWOULDBLOCK);
527 			if (!error && auio.uio_resid > 0) {
528 			    log(LOG_INFO,
529 				 "short receive (%d/%d) from nfs server %s\n",
530 				 sizeof(u_int32_t) - auio.uio_resid,
531 				 sizeof(u_int32_t),
532 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
533 			    error = EPIPE;
534 			}
535 			if (error)
536 				goto errout;
537 
538 			len = ntohl(len) & ~0x80000000;
539 			/*
540 			 * This is SERIOUS! We are out of sync with the sender
541 			 * and forcing a disconnect/reconnect is all I can do.
542 			 */
543 			if (len > NFS_MAXPACKET) {
544 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
545 				"impossible packet length",
546 				len,
547 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
548 			    error = EFBIG;
549 			    goto errout;
550 			}
551 			auio.uio_resid = len;
552 			do {
553 			    rcvflg = MSG_WAITALL;
554 			    error =  soreceive(so, (struct mbuf **)0,
555 				&auio, mp, (struct mbuf **)0, &rcvflg);
556 			} while (error == EWOULDBLOCK || error == EINTR ||
557 				 error == ERESTART);
558 			if (!error && auio.uio_resid > 0) {
559 			    log(LOG_INFO,
560 				"short receive (%d/%d) from nfs server %s\n",
561 				len - auio.uio_resid, len,
562 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
563 			    error = EPIPE;
564 			}
565 		} else {
566 			/*
567 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
568 			 * and soreceive() will return when it has either a
569 			 * control msg or a data msg.
570 			 * We have no use for control msg., but must grab them
571 			 * and then throw them away so we know what is going
572 			 * on.
573 			 */
574 			auio.uio_resid = len = 100000000; /* Anything Big */
575 			auio.uio_procp = p;
576 			do {
577 			    rcvflg = 0;
578 			    error =  soreceive(so, (struct mbuf **)0,
579 				&auio, mp, &control, &rcvflg);
580 			    if (control)
581 				m_freem(control);
582 			    if (error == EWOULDBLOCK && rep) {
583 				if (rep->r_flags & R_SOFTTERM)
584 					return (EINTR);
585 			    }
586 			} while (error == EWOULDBLOCK ||
587 				 (!error && *mp == NULL && control));
588 			if ((rcvflg & MSG_EOR) == 0)
589 				printf("Egad!!\n");
590 			if (!error && *mp == NULL)
591 				error = EPIPE;
592 			len -= auio.uio_resid;
593 		}
594 errout:
595 		if (error && error != EINTR && error != ERESTART) {
596 			m_freem(*mp);
597 			*mp = (struct mbuf *)0;
598 			if (error != EPIPE)
599 				log(LOG_INFO,
600 				    "receive error %d from nfs server %s\n",
601 				    error,
602 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
603 			error = nfs_sndlock(&rep->r_nmp->nm_flag, rep);
604 			if (!error)
605 				error = nfs_reconnect(rep);
606 			if (!error)
607 				goto tryagain;
608 		}
609 	} else {
610 		if ((so = rep->r_nmp->nm_so) == NULL)
611 			return (EACCES);
612 		if (so->so_state & SS_ISCONNECTED)
613 			getnam = (struct mbuf **)0;
614 		else
615 			getnam = aname;
616 		auio.uio_resid = len = 1000000;
617 		auio.uio_procp = p;
618 		do {
619 			rcvflg = 0;
620 			error =  soreceive(so, getnam, &auio, mp,
621 				(struct mbuf **)0, &rcvflg);
622 			if (error == EWOULDBLOCK &&
623 			    (rep->r_flags & R_SOFTTERM))
624 				return (EINTR);
625 		} while (error == EWOULDBLOCK);
626 		len -= auio.uio_resid;
627 	}
628 	if (error) {
629 		m_freem(*mp);
630 		*mp = (struct mbuf *)0;
631 	}
632 	/*
633 	 * Search for any mbufs that are not a multiple of 4 bytes long
634 	 * or with m_data not longword aligned.
635 	 * These could cause pointer alignment problems, so copy them to
636 	 * well aligned mbufs.
637 	 */
638 	nfs_realign(mp, 5 * NFSX_UNSIGNED);
639 	return (error);
640 }
641 
642 /*
643  * Implement receipt of reply on a socket.
644  * We must search through the list of received datagrams matching them
645  * with outstanding requests using the xid, until ours is found.
646  */
647 /* ARGSUSED */
648 int
649 nfs_reply(myrep)
650 	struct nfsreq *myrep;
651 {
652 	struct nfsreq *rep;
653 	struct nfsmount *nmp = myrep->r_nmp;
654 	int32_t t1;
655 	struct mbuf *mrep, *nam, *md;
656 	u_int32_t rxid, *tl;
657 	caddr_t dpos, cp2;
658 	int error;
659 
660 	/*
661 	 * Loop around until we get our own reply
662 	 */
663 	for (;;) {
664 		/*
665 		 * Lock against other receivers so that I don't get stuck in
666 		 * sbwait() after someone else has received my reply for me.
667 		 * Also necessary for connection based protocols to avoid
668 		 * race conditions during a reconnect.
669 		 */
670 		error = nfs_rcvlock(myrep);
671 		if (error)
672 			return (error);
673 		/* Already received, bye bye */
674 		if (myrep->r_mrep != NULL) {
675 			nfs_rcvunlock(&nmp->nm_flag);
676 			return (0);
677 		}
678 		/*
679 		 * Get the next Rpc reply off the socket
680 		 */
681 		error = nfs_receive(myrep, &nam, &mrep);
682 		nfs_rcvunlock(&nmp->nm_flag);
683 		if (error) {
684 
685 			/*
686 			 * Ignore routing errors on connectionless protocols??
687 			 */
688 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
689 				nmp->nm_so->so_error = 0;
690 				if (myrep->r_flags & R_GETONEREP)
691 					return (0);
692 				continue;
693 			}
694 			return (error);
695 		}
696 		if (nam)
697 			m_freem(nam);
698 
699 		/*
700 		 * Get the xid and check that it is an rpc reply
701 		 */
702 		md = mrep;
703 		dpos = mtod(md, caddr_t);
704 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
705 		rxid = *tl++;
706 		if (*tl != rpc_reply) {
707 			nfsstats.rpcinvalid++;
708 			m_freem(mrep);
709 nfsmout:
710 			if (myrep->r_flags & R_GETONEREP)
711 				return (0);
712 			continue;
713 		}
714 
715 		/*
716 		 * Loop through the request list to match up the reply
717 		 * Iff no match, just drop the datagram
718 		 */
719 		for (rep = TAILQ_FIRST(&nfs_reqq); rep != NULL;
720 		    rep = TAILQ_NEXT(rep, r_chain)) {
721 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
722 				/* Found it.. */
723 				rep->r_mrep = mrep;
724 				rep->r_md = md;
725 				rep->r_dpos = dpos;
726 				if (nfsrtton) {
727 					struct rttl *rt;
728 
729 					rt = &nfsrtt.rttl[nfsrtt.pos];
730 					rt->proc = rep->r_procnum;
731 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
732 					rt->sent = nmp->nm_sent;
733 					rt->cwnd = nmp->nm_cwnd;
734 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
735 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
736 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
737 					rt->tstamp = time;
738 					if (rep->r_flags & R_TIMING)
739 						rt->rtt = rep->r_rtt;
740 					else
741 						rt->rtt = 1000000;
742 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
743 				}
744 				/*
745 				 * Update congestion window.
746 				 * Do the additive increase of
747 				 * one rpc/rtt.
748 				 */
749 				if (nmp->nm_cwnd <= nmp->nm_sent) {
750 					nmp->nm_cwnd +=
751 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
752 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
753 					if (nmp->nm_cwnd > NFS_MAXCWND)
754 						nmp->nm_cwnd = NFS_MAXCWND;
755 				}
756 				rep->r_flags &= ~R_SENT;
757 				nmp->nm_sent -= NFS_CWNDSCALE;
758 				/*
759 				 * Update rtt using a gain of 0.125 on the mean
760 				 * and a gain of 0.25 on the deviation.
761 				 */
762 				if (rep->r_flags & R_TIMING) {
763 					/*
764 					 * Since the timer resolution of
765 					 * NFS_HZ is so course, it can often
766 					 * result in r_rtt == 0. Since
767 					 * r_rtt == N means that the actual
768 					 * rtt is between N+dt and N+2-dt ticks,
769 					 * add 1.
770 					 */
771 					t1 = rep->r_rtt + 1;
772 					t1 -= (NFS_SRTT(rep) >> 3);
773 					NFS_SRTT(rep) += t1;
774 					if (t1 < 0)
775 						t1 = -t1;
776 					t1 -= (NFS_SDRTT(rep) >> 2);
777 					NFS_SDRTT(rep) += t1;
778 				}
779 				nmp->nm_timeouts = 0;
780 				break;
781 			}
782 		}
783 		/*
784 		 * If not matched to a request, drop it.
785 		 * If it's mine, get out.
786 		 */
787 		if (rep == 0) {
788 			nfsstats.rpcunexpected++;
789 			m_freem(mrep);
790 		} else if (rep == myrep) {
791 			if (rep->r_mrep == NULL)
792 				panic("nfsreply nil");
793 			return (0);
794 		}
795 		if (myrep->r_flags & R_GETONEREP)
796 			return (0);
797 	}
798 }
799 
800 /*
801  * nfs_request - goes something like this
802  *	- fill in request struct
803  *	- links it into list
804  *	- calls nfs_send() for first transmit
805  *	- calls nfs_receive() to get reply
806  *	- break down rpc header and return with nfs reply pointed to
807  *	  by mrep or error
808  * nb: always frees up mreq mbuf list
809  */
810 int
811 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
812 	struct vnode *vp;
813 	struct mbuf *mrest;
814 	int procnum;
815 	struct proc *procp;
816 	struct ucred *cred;
817 	struct mbuf **mrp;
818 	struct mbuf **mdp;
819 	caddr_t *dposp;
820 {
821 	struct mbuf *m, *mrep;
822 	struct nfsreq *rep;
823 	u_int32_t *tl;
824 	int i;
825 	struct nfsmount *nmp;
826 	struct mbuf *md, *mheadend;
827 	char nickv[RPCX_NICKVERF];
828 	time_t reqtime, waituntil;
829 	caddr_t dpos, cp2;
830 	int t1, s, error = 0, mrest_len, auth_len, auth_type;
831 	int trylater_delay = 15, trylater_cnt = 0, failed_auth = 0;
832 	int verf_len, verf_type;
833 	u_int32_t xid;
834 	char *auth_str, *verf_str;
835 	NFSKERBKEY_T key;		/* save session key */
836 
837 	nmp = VFSTONFS(vp->v_mount);
838 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
839 	rep->r_nmp = nmp;
840 	rep->r_vp = vp;
841 	rep->r_procp = procp;
842 	rep->r_procnum = procnum;
843 	i = 0;
844 	m = mrest;
845 	while (m) {
846 		i += m->m_len;
847 		m = m->m_next;
848 	}
849 	mrest_len = i;
850 
851 	/*
852 	 * Get the RPC header with authorization.
853 	 */
854 kerbauth:
855 	verf_str = auth_str = (char *)0;
856 	if (nmp->nm_flag & NFSMNT_KERB) {
857 		verf_str = nickv;
858 		verf_len = sizeof (nickv);
859 		auth_type = RPCAUTH_KERB4;
860 		bzero((caddr_t)key, sizeof (key));
861 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
862 			&auth_len, verf_str, verf_len)) {
863 			error = nfs_getauth(nmp, rep, cred, &auth_str,
864 				&auth_len, verf_str, &verf_len, key);
865 			if (error) {
866 				free((caddr_t)rep, M_NFSREQ);
867 				m_freem(mrest);
868 				return (error);
869 			}
870 		}
871 	} else {
872 		auth_type = RPCAUTH_UNIX;
873 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
874 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
875 			5 * NFSX_UNSIGNED;
876 	}
877 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
878 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
879 	if (auth_str)
880 		free(auth_str, M_TEMP);
881 
882 	/*
883 	 * For stream protocols, insert a Sun RPC Record Mark.
884 	 */
885 	if (nmp->nm_sotype == SOCK_STREAM) {
886 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
887 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
888 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
889 	}
890 	rep->r_mreq = m;
891 	rep->r_xid = xid;
892 tryagain:
893 	if (nmp->nm_flag & NFSMNT_SOFT)
894 		rep->r_retry = nmp->nm_retry;
895 	else
896 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
897 	rep->r_rtt = rep->r_rexmit = 0;
898 	if (proct[procnum] > 0)
899 		rep->r_flags = R_TIMING;
900 	else
901 		rep->r_flags = 0;
902 	rep->r_mrep = NULL;
903 
904 	/*
905 	 * Do the client side RPC.
906 	 */
907 	nfsstats.rpcrequests++;
908 	/*
909 	 * Chain request into list of outstanding requests. Be sure
910 	 * to put it LAST so timer finds oldest requests first.
911 	 */
912 	s = splsoftclock();
913 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
914 
915 	/* Get send time for nqnfs */
916 	reqtime = time.tv_sec;
917 
918 	/*
919 	 * If backing off another request or avoiding congestion, don't
920 	 * send this one now but let timer do it. If not timing a request,
921 	 * do it now.
922 	 */
923 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
924 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
925 		nmp->nm_sent < nmp->nm_cwnd)) {
926 		splx(s);
927 		if (nmp->nm_soflags & PR_CONNREQUIRED)
928 			error = nfs_sndlock(&nmp->nm_flag, rep);
929 		if (!error) {
930 			error = nfs_send(nmp->nm_so, nmp->nm_nam,
931 					m_copym(m, 0, M_COPYALL, M_WAIT),
932 					rep);
933 			if (nmp->nm_soflags & PR_CONNREQUIRED)
934 				nfs_sndunlock(&nmp->nm_flag);
935 		}
936 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
937 			nmp->nm_sent += NFS_CWNDSCALE;
938 			rep->r_flags |= R_SENT;
939 		}
940 	} else {
941 		splx(s);
942 		rep->r_rtt = -1;
943 	}
944 
945 	/*
946 	 * Wait for the reply from our send or the timer's.
947 	 */
948 	if (!error || error == EPIPE)
949 		error = nfs_reply(rep);
950 
951 	/*
952 	 * RPC done, unlink the request.
953 	 */
954 	s = splsoftclock();
955 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
956 	splx(s);
957 
958 	/*
959 	 * Decrement the outstanding request count.
960 	 */
961 	if (rep->r_flags & R_SENT) {
962 		rep->r_flags &= ~R_SENT;	/* paranoia */
963 		nmp->nm_sent -= NFS_CWNDSCALE;
964 	}
965 
966 	/*
967 	 * If there was a successful reply and a tprintf msg.
968 	 * tprintf a response.
969 	 */
970 	if (!error && (rep->r_flags & R_TPRINTFMSG))
971 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
972 		    "is alive again");
973 	mrep = rep->r_mrep;
974 	md = rep->r_md;
975 	dpos = rep->r_dpos;
976 	if (error) {
977 		m_freem(rep->r_mreq);
978 		free((caddr_t)rep, M_NFSREQ);
979 		return (error);
980 	}
981 
982 	/*
983 	 * break down the rpc header and check if ok
984 	 */
985 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
986 	if (*tl++ == rpc_msgdenied) {
987 		if (*tl == rpc_mismatch)
988 			error = EOPNOTSUPP;
989 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
990 			if (!failed_auth) {
991 				failed_auth++;
992 				mheadend->m_next = (struct mbuf *)0;
993 				m_freem(mrep);
994 				m_freem(rep->r_mreq);
995 				goto kerbauth;
996 			} else
997 				error = EAUTH;
998 		} else
999 			error = EACCES;
1000 		m_freem(mrep);
1001 		m_freem(rep->r_mreq);
1002 		free((caddr_t)rep, M_NFSREQ);
1003 		return (error);
1004 	}
1005 
1006 	/*
1007 	 * Grab any Kerberos verifier, otherwise just throw it away.
1008 	 */
1009 	verf_type = fxdr_unsigned(int, *tl++);
1010 	i = fxdr_unsigned(int32_t, *tl);
1011 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1012 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1013 		if (error)
1014 			goto nfsmout;
1015 	} else if (i > 0)
1016 		nfsm_adv(nfsm_rndup(i));
1017 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1018 	/* 0 == ok */
1019 	if (*tl == 0) {
1020 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1021 		if (*tl != 0) {
1022 			error = fxdr_unsigned(int, *tl);
1023 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1024 				error == NFSERR_TRYLATER) {
1025 				m_freem(mrep);
1026 				error = 0;
1027 				waituntil = time.tv_sec + trylater_delay;
1028 				while (time.tv_sec < waituntil)
1029 					(void) tsleep((caddr_t)&lbolt,
1030 						PSOCK, "nqnfstry", 0);
1031 				trylater_delay *= nfs_backoff[trylater_cnt];
1032 				if (trylater_cnt < 7)
1033 					trylater_cnt++;
1034 				goto tryagain;
1035 			}
1036 
1037 			/*
1038 			 * If the File Handle was stale, invalidate the
1039 			 * lookup cache, just in case.
1040 			 */
1041 			if (error == ESTALE)
1042 				cache_purge(vp);
1043 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1044 				*mrp = mrep;
1045 				*mdp = md;
1046 				*dposp = dpos;
1047 				error |= NFSERR_RETERR;
1048 			} else
1049 				m_freem(mrep);
1050 			m_freem(rep->r_mreq);
1051 			free((caddr_t)rep, M_NFSREQ);
1052 			return (error);
1053 		}
1054 
1055 		*mrp = mrep;
1056 		*mdp = md;
1057 		*dposp = dpos;
1058 		m_freem(rep->r_mreq);
1059 		FREE((caddr_t)rep, M_NFSREQ);
1060 		return (0);
1061 	}
1062 	m_freem(mrep);
1063 	error = EPROTONOSUPPORT;
1064 nfsmout:
1065 	m_freem(rep->r_mreq);
1066 	free((caddr_t)rep, M_NFSREQ);
1067 	return (error);
1068 }
1069 #endif /* NFSCLIENT */
1070 
1071 /*
1072  * Generate the rpc reply header
1073  * siz arg. is used to decide if adding a cluster is worthwhile
1074  */
1075 int
1076 nfs_rephead(siz, nd, slp, err, frev, mrq, mbp, bposp)
1077 	int siz;
1078 	struct nfsrv_descript *nd;
1079 	struct nfssvc_sock *slp;
1080 	int err;
1081 	u_quad_t *frev;
1082 	struct mbuf **mrq;
1083 	struct mbuf **mbp;
1084 	caddr_t *bposp;
1085 {
1086 	u_int32_t *tl;
1087 	struct mbuf *mreq;
1088 	caddr_t bpos;
1089 	struct mbuf *mb, *mb2;
1090 
1091 	MGETHDR(mreq, M_WAIT, MT_DATA);
1092 	mb = mreq;
1093 	/*
1094 	 * If this is a big reply, use a cluster else
1095 	 * try and leave leading space for the lower level headers.
1096 	 */
1097 	siz += RPC_REPLYSIZ;
1098 	if (siz >= max_datalen) {
1099 		MCLGET(mreq, M_WAIT);
1100 	} else
1101 		mreq->m_data += max_hdr;
1102 	tl = mtod(mreq, u_int32_t *);
1103 	mreq->m_len = 6 * NFSX_UNSIGNED;
1104 	bpos = ((caddr_t)tl) + mreq->m_len;
1105 	*tl++ = txdr_unsigned(nd->nd_retxid);
1106 	*tl++ = rpc_reply;
1107 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1108 		*tl++ = rpc_msgdenied;
1109 		if (err & NFSERR_AUTHERR) {
1110 			*tl++ = rpc_autherr;
1111 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1112 			mreq->m_len -= NFSX_UNSIGNED;
1113 			bpos -= NFSX_UNSIGNED;
1114 		} else {
1115 			*tl++ = rpc_mismatch;
1116 			*tl++ = txdr_unsigned(RPC_VER2);
1117 			*tl = txdr_unsigned(RPC_VER2);
1118 		}
1119 	} else {
1120 		*tl++ = rpc_msgaccepted;
1121 
1122 		/*
1123 		 * For Kerberos authentication, we must send the nickname
1124 		 * verifier back, otherwise just RPCAUTH_NULL.
1125 		 */
1126 		if (nd->nd_flag & ND_KERBFULL) {
1127 		    struct nfsuid *nuidp;
1128 		    struct timeval ktvin, ktvout;
1129 
1130 		    for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1131 		 	nuidp != NULL; nuidp = LIST_NEXT(nuidp, nu_hash)) {
1132 			if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1133 			    (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1134 			     &nuidp->nu_haddr, nd->nd_nam2)))
1135 			    break;
1136 		    }
1137 		    if (nuidp) {
1138 			ktvin.tv_sec =
1139 			    txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1140 			ktvin.tv_usec =
1141 			    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1142 
1143 			*tl++ = rpc_auth_kerb;
1144 			*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1145 			*tl = ktvout.tv_sec;
1146 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1147 			*tl++ = ktvout.tv_usec;
1148 			*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1149 		    } else {
1150 			*tl++ = 0;
1151 			*tl++ = 0;
1152 		    }
1153 		} else {
1154 			*tl++ = 0;
1155 			*tl++ = 0;
1156 		}
1157 		switch (err) {
1158 		case EPROGUNAVAIL:
1159 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1160 			break;
1161 		case EPROGMISMATCH:
1162 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1163 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1164 			*tl++ = txdr_unsigned(2);
1165 			*tl = txdr_unsigned(3);
1166 			break;
1167 		case EPROCUNAVAIL:
1168 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1169 			break;
1170 		case EBADRPC:
1171 			*tl = txdr_unsigned(RPC_GARBAGE);
1172 			break;
1173 		default:
1174 			*tl = 0;
1175 			if (err != NFSERR_RETVOID) {
1176 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1177 				if (err)
1178 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1179 				else
1180 				    *tl = 0;
1181 			}
1182 			break;
1183 		};
1184 	}
1185 
1186 	*mrq = mreq;
1187 	if (mbp != NULL)
1188 		*mbp = mb;
1189 	*bposp = bpos;
1190 	if (err != 0 && err != NFSERR_RETVOID)
1191 		nfsstats.srvrpc_errs++;
1192 	return (0);
1193 }
1194 
1195 /*
1196  * Nfs timer routine
1197  * Scan the nfsreq list and retranmit any requests that have timed out
1198  * To avoid retransmission attempts on STREAM sockets (in the future) make
1199  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1200  */
1201 void
1202 nfs_timer(arg)
1203 	void *arg;
1204 {
1205 	struct timeout *to = (struct timeout *)arg;
1206 	struct nfsreq *rep;
1207 	struct mbuf *m;
1208 	struct socket *so;
1209 	struct nfsmount *nmp;
1210 	int timeo;
1211 	int s, error;
1212 #ifdef NFSSERVER
1213 	struct nfssvc_sock *slp;
1214 	u_quad_t cur_usec;
1215 #endif
1216 
1217 	s = splsoftnet();
1218 	for (rep = TAILQ_FIRST(&nfs_reqq); rep != NULL;
1219 	    rep = TAILQ_NEXT(rep, r_chain)) {
1220 		nmp = rep->r_nmp;
1221 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1222 			continue;
1223 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1224 			rep->r_flags |= R_SOFTTERM;
1225 			continue;
1226 		}
1227 		if (rep->r_rtt >= 0) {
1228 			rep->r_rtt++;
1229 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1230 				timeo = nmp->nm_timeo;
1231 			else
1232 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1233 			if (nmp->nm_timeouts > 0)
1234 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1235 			if (rep->r_rtt <= timeo)
1236 				continue;
1237 			if (nmp->nm_timeouts < 8)
1238 				nmp->nm_timeouts++;
1239 		}
1240 		/*
1241 		 * Check for server not responding
1242 		 */
1243 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1244 		     rep->r_rexmit > nmp->nm_deadthresh) {
1245 			nfs_msg(rep->r_procp,
1246 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1247 			    "not responding");
1248 			rep->r_flags |= R_TPRINTFMSG;
1249 		}
1250 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1251 			nfsstats.rpctimeouts++;
1252 			rep->r_flags |= R_SOFTTERM;
1253 			continue;
1254 		}
1255 		if (nmp->nm_sotype != SOCK_DGRAM) {
1256 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1257 				rep->r_rexmit = NFS_MAXREXMIT;
1258 			continue;
1259 		}
1260 		if ((so = nmp->nm_so) == NULL)
1261 			continue;
1262 
1263 		/*
1264 		 * If there is enough space and the window allows..
1265 		 *	Resend it
1266 		 * Set r_rtt to -1 in case we fail to send it now.
1267 		 */
1268 		rep->r_rtt = -1;
1269 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1270 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1271 		    (rep->r_flags & R_SENT) ||
1272 		    nmp->nm_sent < nmp->nm_cwnd) &&
1273 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1274 			if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
1275 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1276 			    (struct mbuf *)0, (struct mbuf *)0);
1277 			else
1278 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1279 			    nmp->nm_nam, (struct mbuf *)0);
1280 			if (error) {
1281 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
1282 					so->so_error = 0;
1283 			} else {
1284 				/*
1285 				 * Iff first send, start timing
1286 				 * else turn timing off, backoff timer
1287 				 * and divide congestion window by 2.
1288 				 */
1289 				if (rep->r_flags & R_SENT) {
1290 					rep->r_flags &= ~R_TIMING;
1291 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1292 						rep->r_rexmit = NFS_MAXREXMIT;
1293 					nmp->nm_cwnd >>= 1;
1294 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1295 						nmp->nm_cwnd = NFS_CWNDSCALE;
1296 					nfsstats.rpcretries++;
1297 				} else {
1298 					rep->r_flags |= R_SENT;
1299 					nmp->nm_sent += NFS_CWNDSCALE;
1300 				}
1301 				rep->r_rtt = 0;
1302 			}
1303 		}
1304 	}
1305 
1306 #ifdef NFSSERVER
1307 	/*
1308 	 * Scan the write gathering queues for writes that need to be
1309 	 * completed now.
1310 	 */
1311 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1312 	for (slp = TAILQ_FIRST(&nfssvc_sockhead); slp != NULL;
1313 	    slp = TAILQ_NEXT(slp, ns_chain)) {
1314 	    if (LIST_FIRST(&slp->ns_tq) &&
1315 	        LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec)
1316 		nfsrv_wakenfsd(slp);
1317 	}
1318 #endif /* NFSSERVER */
1319 	splx(s);
1320 	timeout_add(to, nfs_ticks);
1321 }
1322 
1323 /*
1324  * Test for a termination condition pending on the process.
1325  * This is used for NFSMNT_INT mounts.
1326  */
1327 int
1328 nfs_sigintr(nmp, rep, p)
1329 	struct nfsmount *nmp;
1330 	struct nfsreq *rep;
1331 	struct proc *p;
1332 {
1333 
1334 	if (rep && (rep->r_flags & R_SOFTTERM))
1335 		return (EINTR);
1336 	if (!(nmp->nm_flag & NFSMNT_INT))
1337 		return (0);
1338 	if (p && p->p_siglist &&
1339 	    (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) &
1340 	    NFSINT_SIGMASK))
1341 		return (EINTR);
1342 	return (0);
1343 }
1344 
1345 /*
1346  * Lock a socket against others.
1347  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1348  * and also to avoid race conditions between the processes with nfs requests
1349  * in progress when a reconnect is necessary.
1350  */
1351 int
1352 nfs_sndlock(flagp, rep)
1353 	int *flagp;
1354 	struct nfsreq *rep;
1355 {
1356 	struct proc *p;
1357 	int slpflag = 0, slptimeo = 0;
1358 
1359 	if (rep) {
1360 		p = rep->r_procp;
1361 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1362 			slpflag = PCATCH;
1363 	} else
1364 		p = (struct proc *)0;
1365 	while (*flagp & NFSMNT_SNDLOCK) {
1366 		if (nfs_sigintr(rep->r_nmp, rep, p))
1367 			return (EINTR);
1368 		*flagp |= NFSMNT_WANTSND;
1369 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1370 			slptimeo);
1371 		if (slpflag == PCATCH) {
1372 			slpflag = 0;
1373 			slptimeo = 2 * hz;
1374 		}
1375 	}
1376 	*flagp |= NFSMNT_SNDLOCK;
1377 	return (0);
1378 }
1379 
1380 /*
1381  * Unlock the stream socket for others.
1382  */
1383 void
1384 nfs_sndunlock(flagp)
1385 	int *flagp;
1386 {
1387 
1388 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1389 		panic("nfs sndunlock");
1390 	*flagp &= ~NFSMNT_SNDLOCK;
1391 	if (*flagp & NFSMNT_WANTSND) {
1392 		*flagp &= ~NFSMNT_WANTSND;
1393 		wakeup((caddr_t)flagp);
1394 	}
1395 }
1396 
1397 int
1398 nfs_rcvlock(rep)
1399 	struct nfsreq *rep;
1400 {
1401 	int *flagp = &rep->r_nmp->nm_flag;
1402 	int slpflag, slptimeo = 0;
1403 
1404 	if (*flagp & NFSMNT_INT)
1405 		slpflag = PCATCH;
1406 	else
1407 		slpflag = 0;
1408 	while (*flagp & NFSMNT_RCVLOCK) {
1409 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1410 			return (EINTR);
1411 		*flagp |= NFSMNT_WANTRCV;
1412 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1413 			slptimeo);
1414 		if (slpflag == PCATCH) {
1415 			slpflag = 0;
1416 			slptimeo = 2 * hz;
1417 		}
1418 	}
1419 	*flagp |= NFSMNT_RCVLOCK;
1420 	return (0);
1421 }
1422 
1423 /*
1424  * Unlock the stream socket for others.
1425  */
1426 void
1427 nfs_rcvunlock(flagp)
1428 	int *flagp;
1429 {
1430 
1431 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1432 		panic("nfs rcvunlock");
1433 	*flagp &= ~NFSMNT_RCVLOCK;
1434 	if (*flagp & NFSMNT_WANTRCV) {
1435 		*flagp &= ~NFSMNT_WANTRCV;
1436 		wakeup((caddr_t)flagp);
1437 	}
1438 }
1439 
1440 /*
1441  *  NFS parsing code requires 32-bit alignment
1442  */
1443 void
1444 nfs_realign(struct mbuf **pm, int hsiz)
1445 {
1446 	struct mbuf *m;
1447 	struct mbuf *n = NULL;
1448 	int off = 0;
1449 
1450 	++nfs_realign_test;
1451 	while ((m = *pm) != NULL) {
1452 		if ((m->m_len & 0x3) || (mtod(m, long) & 0x3)) {
1453 			MGET(n, M_WAIT, MT_DATA);
1454 			if (m->m_len >= MINCLSIZE) {
1455 				MCLGET(n, M_WAIT);
1456 			}
1457 			n->m_len = 0;
1458 			break;
1459 		}
1460 		pm = &m->m_next;
1461 	}
1462 	/*
1463 	 * If n is non-NULL, loop on m copying data, then replace the
1464 	 * portion of the chain that had to be realigned.
1465 	 */
1466 	if (n != NULL) {
1467 		++nfs_realign_count;
1468 		while (m) {
1469 			m_copyback(n, off, m->m_len, mtod(m, caddr_t));
1470 			off += m->m_len;
1471 			m = m->m_next;
1472 		}
1473 		m_freem(*pm);
1474 		*pm = n;
1475 	}
1476 }
1477 
1478 
1479 /*
1480  * Parse an RPC request
1481  * - verify it
1482  * - fill in the cred struct.
1483  */
1484 int
1485 nfs_getreq(nd, nfsd, has_header)
1486 	struct nfsrv_descript *nd;
1487 	struct nfsd *nfsd;
1488 	int has_header;
1489 {
1490 	int len, i;
1491 	u_int32_t *tl;
1492 	int32_t t1;
1493 	struct uio uio;
1494 	struct iovec iov;
1495 	caddr_t dpos, cp2, cp;
1496 	u_int32_t nfsvers, auth_type;
1497 	uid_t nickuid;
1498 	int error = 0, ticklen;
1499 	struct mbuf *mrep, *md;
1500 	struct nfsuid *nuidp;
1501 	struct timeval tvin, tvout;
1502 
1503 	mrep = nd->nd_mrep;
1504 	md = nd->nd_md;
1505 	dpos = nd->nd_dpos;
1506 	if (has_header) {
1507 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1508 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1509 		if (*tl++ != rpc_call) {
1510 			m_freem(mrep);
1511 			return (EBADRPC);
1512 		}
1513 	} else
1514 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1515 	nd->nd_repstat = 0;
1516 	nd->nd_flag = 0;
1517 	if (*tl++ != rpc_vers) {
1518 		nd->nd_repstat = ERPCMISMATCH;
1519 		nd->nd_procnum = NFSPROC_NOOP;
1520 		return (0);
1521 	}
1522 	if (*tl != nfs_prog) {
1523 		nd->nd_repstat = EPROGUNAVAIL;
1524 		nd->nd_procnum = NFSPROC_NOOP;
1525 		return (0);
1526 	}
1527 	tl++;
1528 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1529 	if (nfsvers != NFS_VER2 && nfsvers != NFS_VER3) {
1530 		nd->nd_repstat = EPROGMISMATCH;
1531 		nd->nd_procnum = NFSPROC_NOOP;
1532 		return (0);
1533 	}
1534 	if (nfsvers == NFS_VER3)
1535 		nd->nd_flag = ND_NFSV3;
1536 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1537 	if (nd->nd_procnum == NFSPROC_NULL)
1538 		return (0);
1539 	if (nd->nd_procnum >= NFS_NPROCS ||
1540 		(nd->nd_procnum > NFSPROC_COMMIT) ||
1541 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1542 		nd->nd_repstat = EPROCUNAVAIL;
1543 		nd->nd_procnum = NFSPROC_NOOP;
1544 		return (0);
1545 	}
1546 	if ((nd->nd_flag & ND_NFSV3) == 0)
1547 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1548 	auth_type = *tl++;
1549 	len = fxdr_unsigned(int, *tl++);
1550 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1551 		m_freem(mrep);
1552 		return (EBADRPC);
1553 	}
1554 
1555 	nd->nd_flag &= ~ND_KERBAUTH;
1556 	/*
1557 	 * Handle auth_unix or auth_kerb.
1558 	 */
1559 	if (auth_type == rpc_auth_unix) {
1560 		len = fxdr_unsigned(int, *++tl);
1561 		if (len < 0 || len > NFS_MAXNAMLEN) {
1562 			m_freem(mrep);
1563 			return (EBADRPC);
1564 		}
1565 		nfsm_adv(nfsm_rndup(len));
1566 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1567 		bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred));
1568 		nd->nd_cr.cr_ref = 1;
1569 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1570 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1571 		len = fxdr_unsigned(int, *tl);
1572 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1573 			m_freem(mrep);
1574 			return (EBADRPC);
1575 		}
1576 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1577 		for (i = 0; i < len; i++)
1578 		    if (i < NGROUPS)
1579 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1580 		    else
1581 			tl++;
1582 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1583 		if (nd->nd_cr.cr_ngroups > 1)
1584 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1585 		len = fxdr_unsigned(int, *++tl);
1586 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1587 			m_freem(mrep);
1588 			return (EBADRPC);
1589 		}
1590 		if (len > 0)
1591 			nfsm_adv(nfsm_rndup(len));
1592 	} else if (auth_type == rpc_auth_kerb) {
1593 		switch (fxdr_unsigned(int, *tl++)) {
1594 		case RPCAKN_FULLNAME:
1595 			ticklen = fxdr_unsigned(int, *tl);
1596 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1597 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1598 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1599 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1600 				m_freem(mrep);
1601 				return (EBADRPC);
1602 			}
1603 			uio.uio_offset = 0;
1604 			uio.uio_iov = &iov;
1605 			uio.uio_iovcnt = 1;
1606 			uio.uio_segflg = UIO_SYSSPACE;
1607 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1608 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1609 			nfsm_mtouio(&uio, uio.uio_resid);
1610 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1611 			if (*tl++ != rpc_auth_kerb ||
1612 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1613 				printf("Bad kerb verifier\n");
1614 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1615 				nd->nd_procnum = NFSPROC_NOOP;
1616 				return (0);
1617 			}
1618 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1619 			tl = (u_int32_t *)cp;
1620 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1621 				printf("Not fullname kerb verifier\n");
1622 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1623 				nd->nd_procnum = NFSPROC_NOOP;
1624 				return (0);
1625 			}
1626 			cp += NFSX_UNSIGNED;
1627 			bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED);
1628 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1629 			nd->nd_flag |= ND_KERBFULL;
1630 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1631 			break;
1632 		case RPCAKN_NICKNAME:
1633 			if (len != 2 * NFSX_UNSIGNED) {
1634 				printf("Kerb nickname short\n");
1635 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1636 				nd->nd_procnum = NFSPROC_NOOP;
1637 				return (0);
1638 			}
1639 			nickuid = fxdr_unsigned(uid_t, *tl);
1640 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1641 			if (*tl++ != rpc_auth_kerb ||
1642 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1643 				printf("Kerb nick verifier bad\n");
1644 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1645 				nd->nd_procnum = NFSPROC_NOOP;
1646 				return (0);
1647 			}
1648 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1649 			tvin.tv_sec = *tl++;
1650 			tvin.tv_usec = *tl;
1651 
1652 			for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1653 			    nuidp != NULL; nuidp = LIST_NEXT(nuidp, nu_hash)) {
1654 				if (nuidp->nu_cr.cr_uid == nickuid &&
1655 				    (!nd->nd_nam2 ||
1656 				     netaddr_match(NU_NETFAM(nuidp),
1657 				      &nuidp->nu_haddr, nd->nd_nam2)))
1658 					break;
1659 			}
1660 			if (!nuidp) {
1661 				nd->nd_repstat =
1662 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1663 				nd->nd_procnum = NFSPROC_NOOP;
1664 				return (0);
1665 			}
1666 
1667 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1668 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1669 			if (nuidp->nu_expire < time.tv_sec ||
1670 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1671 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1672 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1673 				nuidp->nu_expire = 0;
1674 				nd->nd_repstat =
1675 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1676 				nd->nd_procnum = NFSPROC_NOOP;
1677 				return (0);
1678 			}
1679 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1680 			nd->nd_flag |= ND_KERBNICK;
1681 		};
1682 	} else {
1683 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1684 		nd->nd_procnum = NFSPROC_NOOP;
1685 		return (0);
1686 	}
1687 
1688 	nd->nd_md = md;
1689 	nd->nd_dpos = dpos;
1690 	return (0);
1691 nfsmout:
1692 	return (error);
1693 }
1694 
1695 int
1696 nfs_msg(p, server, msg)
1697 	struct proc *p;
1698 	char *server, *msg;
1699 {
1700 	tpr_t tpr;
1701 
1702 	if (p)
1703 		tpr = tprintf_open(p);
1704 	else
1705 		tpr = NULL;
1706 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1707 	tprintf_close(tpr);
1708 	return (0);
1709 }
1710 
1711 #ifdef NFSSERVER
1712 int (*nfsrv3_procs[NFS_NPROCS])(struct nfsrv_descript *,
1713 				    struct nfssvc_sock *, struct proc *,
1714 				    struct mbuf **) = {
1715 	nfsrv_null,
1716 	nfsrv_getattr,
1717 	nfsrv_setattr,
1718 	nfsrv_lookup,
1719 	nfsrv3_access,
1720 	nfsrv_readlink,
1721 	nfsrv_read,
1722 	nfsrv_write,
1723 	nfsrv_create,
1724 	nfsrv_mkdir,
1725 	nfsrv_symlink,
1726 	nfsrv_mknod,
1727 	nfsrv_remove,
1728 	nfsrv_rmdir,
1729 	nfsrv_rename,
1730 	nfsrv_link,
1731 	nfsrv_readdir,
1732 	nfsrv_readdirplus,
1733 	nfsrv_statfs,
1734 	nfsrv_fsinfo,
1735 	nfsrv_pathconf,
1736 	nfsrv_commit,
1737 	nfsrv_noop,
1738 	nfsrv_noop,
1739 	nfsrv_noop,
1740 	nfsrv_noop
1741 };
1742 
1743 /*
1744  * Socket upcall routine for the nfsd sockets.
1745  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1746  * Essentially do as much as possible non-blocking, else punt and it will
1747  * be called with M_WAIT from an nfsd.
1748  */
1749 void
1750 nfsrv_rcv(so, arg, waitflag)
1751 	struct socket *so;
1752 	caddr_t arg;
1753 	int waitflag;
1754 {
1755 	struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1756 	struct mbuf *m;
1757 	struct mbuf *mp, *nam;
1758 	struct uio auio;
1759 	int flags, error;
1760 
1761 	if ((slp->ns_flag & SLP_VALID) == 0)
1762 		return;
1763 #ifdef notdef
1764 	/*
1765 	 * Define this to test for nfsds handling this under heavy load.
1766 	 */
1767 	if (waitflag == M_DONTWAIT) {
1768 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1769 	}
1770 #endif
1771 	auio.uio_procp = NULL;
1772 	if (so->so_type == SOCK_STREAM) {
1773 		/*
1774 		 * If there are already records on the queue, defer soreceive()
1775 		 * to an nfsd so that there is feedback to the TCP layer that
1776 		 * the nfs servers are heavily loaded.
1777 		 */
1778 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1779 			slp->ns_flag |= SLP_NEEDQ;
1780 			goto dorecs;
1781 		}
1782 
1783 		/*
1784 		 * Do soreceive().
1785 		 */
1786 		auio.uio_resid = 1000000000;
1787 		flags = MSG_DONTWAIT;
1788 		error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1789 		if (error || mp == (struct mbuf *)0) {
1790 			if (error == EWOULDBLOCK)
1791 				slp->ns_flag |= SLP_NEEDQ;
1792 			else
1793 				slp->ns_flag |= SLP_DISCONN;
1794 			goto dorecs;
1795 		}
1796 		m = mp;
1797 		if (slp->ns_rawend) {
1798 			slp->ns_rawend->m_next = m;
1799 			slp->ns_cc += 1000000000 - auio.uio_resid;
1800 		} else {
1801 			slp->ns_raw = m;
1802 			slp->ns_cc = 1000000000 - auio.uio_resid;
1803 		}
1804 		while (m->m_next)
1805 			m = m->m_next;
1806 		slp->ns_rawend = m;
1807 
1808 		/*
1809 		 * Now try and parse record(s) out of the raw stream data.
1810 		 */
1811 		error = nfsrv_getstream(slp, waitflag);
1812 		if (error) {
1813 			if (error == EPERM)
1814 				slp->ns_flag |= SLP_DISCONN;
1815 			else
1816 				slp->ns_flag |= SLP_NEEDQ;
1817 		}
1818 	} else {
1819 		do {
1820 			auio.uio_resid = 1000000000;
1821 			flags = MSG_DONTWAIT;
1822 			error = soreceive(so, &nam, &auio, &mp,
1823 						(struct mbuf **)0, &flags);
1824 			if (mp) {
1825 				if (nam) {
1826 					m = nam;
1827 					m->m_next = mp;
1828 				} else
1829 					m = mp;
1830 				if (slp->ns_recend)
1831 					slp->ns_recend->m_nextpkt = m;
1832 				else
1833 					slp->ns_rec = m;
1834 				slp->ns_recend = m;
1835 				m->m_nextpkt = (struct mbuf *)0;
1836 			}
1837 			if (error) {
1838 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
1839 					&& error != EWOULDBLOCK) {
1840 					slp->ns_flag |= SLP_DISCONN;
1841 					goto dorecs;
1842 				}
1843 			}
1844 		} while (mp);
1845 	}
1846 
1847 	/*
1848 	 * Now try and process the request records, non-blocking.
1849 	 */
1850 dorecs:
1851 	if (waitflag == M_DONTWAIT &&
1852 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
1853 		nfsrv_wakenfsd(slp);
1854 }
1855 
1856 /*
1857  * Try and extract an RPC request from the mbuf data list received on a
1858  * stream socket. The "waitflag" argument indicates whether or not it
1859  * can sleep.
1860  */
1861 int
1862 nfsrv_getstream(slp, waitflag)
1863 	struct nfssvc_sock *slp;
1864 	int waitflag;
1865 {
1866 	struct mbuf *m, **mpp;
1867 	char *cp1, *cp2;
1868 	int len;
1869 	struct mbuf *om, *m2, *recm;
1870 	u_int32_t recmark;
1871 
1872 	if (slp->ns_flag & SLP_GETSTREAM)
1873 		panic("nfs getstream");
1874 	slp->ns_flag |= SLP_GETSTREAM;
1875 	for (;;) {
1876 	    if (slp->ns_reclen == 0) {
1877 		if (slp->ns_cc < NFSX_UNSIGNED) {
1878 			slp->ns_flag &= ~SLP_GETSTREAM;
1879 			return (0);
1880 		}
1881 		m = slp->ns_raw;
1882 		if (m->m_len >= NFSX_UNSIGNED) {
1883 			bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
1884 			m->m_data += NFSX_UNSIGNED;
1885 			m->m_len -= NFSX_UNSIGNED;
1886 		} else {
1887 			cp1 = (caddr_t)&recmark;
1888 			cp2 = mtod(m, caddr_t);
1889 			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
1890 				while (m->m_len == 0) {
1891 					m = m->m_next;
1892 					cp2 = mtod(m, caddr_t);
1893 				}
1894 				*cp1++ = *cp2++;
1895 				m->m_data++;
1896 				m->m_len--;
1897 			}
1898 		}
1899 		slp->ns_cc -= NFSX_UNSIGNED;
1900 		recmark = ntohl(recmark);
1901 		slp->ns_reclen = recmark & ~0x80000000;
1902 		if (recmark & 0x80000000)
1903 			slp->ns_flag |= SLP_LASTFRAG;
1904 		else
1905 			slp->ns_flag &= ~SLP_LASTFRAG;
1906 		if (slp->ns_reclen > NFS_MAXPACKET) {
1907 			slp->ns_flag &= ~SLP_GETSTREAM;
1908 			return (EPERM);
1909 		}
1910 	    }
1911 
1912 	    /*
1913 	     * Now get the record part.
1914 	     */
1915 	    recm = NULL;
1916 	    if (slp->ns_cc == slp->ns_reclen) {
1917 		recm = slp->ns_raw;
1918 		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
1919 		slp->ns_cc = slp->ns_reclen = 0;
1920 	    } else if (slp->ns_cc > slp->ns_reclen) {
1921 		len = 0;
1922 		m = slp->ns_raw;
1923 		om = (struct mbuf *)0;
1924 		while (len < slp->ns_reclen) {
1925 			if ((len + m->m_len) > slp->ns_reclen) {
1926 				m2 = m_copym(m, 0, slp->ns_reclen - len,
1927 					waitflag);
1928 				if (m2) {
1929 					if (om) {
1930 						om->m_next = m2;
1931 						recm = slp->ns_raw;
1932 					} else
1933 						recm = m2;
1934 					m->m_data += slp->ns_reclen - len;
1935 					m->m_len -= slp->ns_reclen - len;
1936 					len = slp->ns_reclen;
1937 				} else {
1938 					slp->ns_flag &= ~SLP_GETSTREAM;
1939 					return (EWOULDBLOCK);
1940 				}
1941 			} else if ((len + m->m_len) == slp->ns_reclen) {
1942 				om = m;
1943 				len += m->m_len;
1944 				m = m->m_next;
1945 				recm = slp->ns_raw;
1946 				om->m_next = (struct mbuf *)0;
1947 			} else {
1948 				om = m;
1949 				len += m->m_len;
1950 				m = m->m_next;
1951 			}
1952 		}
1953 		slp->ns_raw = m;
1954 		slp->ns_cc -= len;
1955 		slp->ns_reclen = 0;
1956 	    } else {
1957 		slp->ns_flag &= ~SLP_GETSTREAM;
1958 		return (0);
1959 	    }
1960 
1961 	    /*
1962 	     * Accumulate the fragments into a record.
1963 	     */
1964 	    mpp = &slp->ns_frag;
1965 	    while (*mpp)
1966 		mpp = &((*mpp)->m_next);
1967 	    *mpp = recm;
1968 	    if (slp->ns_flag & SLP_LASTFRAG) {
1969 		if (slp->ns_recend)
1970 		    slp->ns_recend->m_nextpkt = slp->ns_frag;
1971 		else
1972 		    slp->ns_rec = slp->ns_frag;
1973 		slp->ns_recend = slp->ns_frag;
1974 		slp->ns_frag = (struct mbuf *)0;
1975 	    }
1976 	}
1977 }
1978 
1979 /*
1980  * Parse an RPC header.
1981  */
1982 int
1983 nfsrv_dorec(slp, nfsd, ndp)
1984 	struct nfssvc_sock *slp;
1985 	struct nfsd *nfsd;
1986 	struct nfsrv_descript **ndp;
1987 {
1988 	struct mbuf *m, *nam;
1989 	struct nfsrv_descript *nd;
1990 	int error;
1991 
1992 	*ndp = NULL;
1993 	if ((slp->ns_flag & SLP_VALID) == 0 ||
1994 	    (m = slp->ns_rec) == (struct mbuf *)0)
1995 		return (ENOBUFS);
1996 	slp->ns_rec = m->m_nextpkt;
1997 	if (slp->ns_rec)
1998 		m->m_nextpkt = (struct mbuf *)0;
1999 	else
2000 		slp->ns_recend = (struct mbuf *)0;
2001 	if (m->m_type == MT_SONAME) {
2002 		nam = m;
2003 		m = m->m_next;
2004 		nam->m_next = NULL;
2005 	} else
2006 		nam = NULL;
2007 	MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2008 		M_NFSRVDESC, M_WAITOK);
2009 	nfs_realign(&m, 10 * NFSX_UNSIGNED);
2010 	nd->nd_md = nd->nd_mrep = m;
2011 	nd->nd_nam2 = nam;
2012 	nd->nd_dpos = mtod(m, caddr_t);
2013 	error = nfs_getreq(nd, nfsd, TRUE);
2014 	if (error) {
2015 		m_freem(nam);
2016 		free((caddr_t)nd, M_NFSRVDESC);
2017 		return (error);
2018 	}
2019 	*ndp = nd;
2020 	nfsd->nfsd_nd = nd;
2021 	return (0);
2022 }
2023 
2024 
2025 /*
2026  * Search for a sleeping nfsd and wake it up.
2027  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2028  * running nfsds will go look for the work in the nfssvc_sock list.
2029  */
2030 void
2031 nfsrv_wakenfsd(slp)
2032 	struct nfssvc_sock *slp;
2033 {
2034 	struct nfsd *nd;
2035 
2036 	if ((slp->ns_flag & SLP_VALID) == 0)
2037 		return;
2038 	for (nd = TAILQ_FIRST(&nfsd_head); nd != NULL;
2039 	    nd = TAILQ_NEXT(nd, nfsd_chain)) {
2040 		if (nd->nfsd_flag & NFSD_WAITING) {
2041 			nd->nfsd_flag &= ~NFSD_WAITING;
2042 			if (nd->nfsd_slp)
2043 				panic("nfsd wakeup");
2044 			slp->ns_sref++;
2045 			nd->nfsd_slp = slp;
2046 			wakeup((caddr_t)nd);
2047 			return;
2048 		}
2049 	}
2050 	slp->ns_flag |= SLP_DOREC;
2051 	nfsd_head_flag |= NFSD_CHECKSLP;
2052 }
2053 #endif /* NFSSERVER */
2054