xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 4d7e773266e3c3f48566c86c0ad52d51c6454fd1)
1 /*	$NetBSD: nfs_socket.c,v 1.39 1997/10/10 01:53:24 fvdl Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1991, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)nfs_socket.c	8.5 (Berkeley) 3/30/95
39  */
40 
41 /*
42  * Socket operations for use by nfs
43  */
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/proc.h>
48 #include <sys/mount.h>
49 #include <sys/kernel.h>
50 #include <sys/mbuf.h>
51 #include <sys/vnode.h>
52 #include <sys/domain.h>
53 #include <sys/protosw.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/syslog.h>
57 #include <sys/tprintf.h>
58 #include <sys/namei.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/tcp.h>
62 
63 #include <nfs/rpcv2.h>
64 #include <nfs/nfsproto.h>
65 #include <nfs/nfs.h>
66 #include <nfs/xdr_subs.h>
67 #include <nfs/nfsm_subs.h>
68 #include <nfs/nfsmount.h>
69 #include <nfs/nfsnode.h>
70 #include <nfs/nfsrtt.h>
71 #include <nfs/nqnfs.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, nqnfs_prog;
103 extern time_t nqnfsstarttime;
104 extern struct nfsstats nfsstats;
105 extern int nfsv3_procid[NFS_NPROCS];
106 extern int nfs_ticks;
107 
108 /*
109  * Defines which timer to use for the procnum.
110  * 0 - default
111  * 1 - getattr
112  * 2 - lookup
113  * 3 - read
114  * 4 - write
115  */
116 static int proct[NFS_NPROCS] = {
117 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
118 	0, 0, 0,
119 };
120 
121 /*
122  * There is a congestion window for outstanding rpcs maintained per mount
123  * point. The cwnd size is adjusted in roughly the way that:
124  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
125  * SIGCOMM '88". ACM, August 1988.
126  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
127  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
128  * of rpcs is in progress.
129  * (The sent count and cwnd are scaled for integer arith.)
130  * Variants of "slow start" were tried and were found to be too much of a
131  * performance hit (ave. rtt 3 times larger),
132  * I suspect due to the large rtt that nfs rpcs have.
133  */
134 #define	NFS_CWNDSCALE	256
135 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
136 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
137 int nfsrtton = 0;
138 struct nfsrtt nfsrtt;
139 
140 /*
141  * Initialize sockets and congestion for a new NFS connection.
142  * We do not free the sockaddr if error.
143  */
144 int
145 nfs_connect(nmp, rep)
146 	register struct nfsmount *nmp;
147 	struct nfsreq *rep;
148 {
149 	register struct socket *so;
150 	int s, error, rcvreserve, sndreserve;
151 	struct sockaddr *saddr;
152 	struct sockaddr_in *sin;
153 	struct mbuf *m;
154 	u_int16_t tport;
155 
156 	nmp->nm_so = (struct socket *)0;
157 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
158 	error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
159 		nmp->nm_soproto);
160 	if (error)
161 		goto bad;
162 	so = nmp->nm_so;
163 	nmp->nm_soflags = so->so_proto->pr_flags;
164 
165 	/*
166 	 * Some servers require that the client port be a reserved port number.
167 	 */
168 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
169 		MGET(m, M_WAIT, MT_SONAME);
170 		sin = mtod(m, struct sockaddr_in *);
171 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
172 		sin->sin_family = AF_INET;
173 		sin->sin_addr.s_addr = INADDR_ANY;
174 		tport = IPPORT_RESERVED - 1;
175 		sin->sin_port = htons(tport);
176 		while ((error = sobind(so, m)) == EADDRINUSE &&
177 		       --tport > IPPORT_RESERVED / 2)
178 			sin->sin_port = htons(tport);
179 		m_freem(m);
180 		if (error)
181 			goto bad;
182 	}
183 
184 	/*
185 	 * Protocols that do not require connections may be optionally left
186 	 * unconnected for servers that reply from a port other than NFS_PORT.
187 	 */
188 	if (nmp->nm_flag & NFSMNT_NOCONN) {
189 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
190 			error = ENOTCONN;
191 			goto bad;
192 		}
193 	} else {
194 		error = soconnect(so, nmp->nm_nam);
195 		if (error)
196 			goto bad;
197 
198 		/*
199 		 * Wait for the connection to complete. Cribbed from the
200 		 * connect system call but with the wait timing out so
201 		 * that interruptible mounts don't hang here for a long time.
202 		 */
203 		s = splsoftnet();
204 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
205 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
206 				"nfscon", 2 * hz);
207 			if ((so->so_state & SS_ISCONNECTING) &&
208 			    so->so_error == 0 && rep &&
209 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
210 				so->so_state &= ~SS_ISCONNECTING;
211 				splx(s);
212 				goto bad;
213 			}
214 		}
215 		if (so->so_error) {
216 			error = so->so_error;
217 			so->so_error = 0;
218 			splx(s);
219 			goto bad;
220 		}
221 		splx(s);
222 	}
223 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
224 		so->so_rcv.sb_timeo = (5 * hz);
225 		so->so_snd.sb_timeo = (5 * hz);
226 	} else {
227 		so->so_rcv.sb_timeo = 0;
228 		so->so_snd.sb_timeo = 0;
229 	}
230 	if (nmp->nm_sotype == SOCK_DGRAM) {
231 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
232 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
233 		    NFS_MAXPKTHDR) * 2;
234 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
235 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
236 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
237 		    NFS_MAXPKTHDR) * 2;
238 	} else {
239 		if (nmp->nm_sotype != SOCK_STREAM)
240 			panic("nfscon sotype");
241 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
242 			MGET(m, M_WAIT, MT_SOOPTS);
243 			*mtod(m, int32_t *) = 1;
244 			m->m_len = sizeof(int32_t);
245 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
246 		}
247 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
248 			MGET(m, M_WAIT, MT_SOOPTS);
249 			*mtod(m, int32_t *) = 1;
250 			m->m_len = sizeof(int32_t);
251 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
252 		}
253 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
254 		    sizeof (u_int32_t)) * 2;
255 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
256 		    sizeof (u_int32_t)) * 2;
257 	}
258 	error = soreserve(so, sndreserve, rcvreserve);
259 	if (error)
260 		goto bad;
261 	so->so_rcv.sb_flags |= SB_NOINTR;
262 	so->so_snd.sb_flags |= SB_NOINTR;
263 
264 	/* Initialize other non-zero congestion variables */
265 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
266 		nmp->nm_srtt[4] = (NFS_TIMEO << 3);
267 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
268 		nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0;
269 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
270 	nmp->nm_sent = 0;
271 	nmp->nm_timeouts = 0;
272 	return (0);
273 
274 bad:
275 	nfs_disconnect(nmp);
276 	return (error);
277 }
278 
279 /*
280  * Reconnect routine:
281  * Called when a connection is broken on a reliable protocol.
282  * - clean up the old socket
283  * - nfs_connect() again
284  * - set R_MUSTRESEND for all outstanding requests on mount point
285  * If this fails the mount point is DEAD!
286  * nb: Must be called with the nfs_sndlock() set on the mount point.
287  */
288 int
289 nfs_reconnect(rep)
290 	register struct nfsreq *rep;
291 {
292 	register struct nfsreq *rp;
293 	register struct nfsmount *nmp = rep->r_nmp;
294 	int error;
295 
296 	nfs_disconnect(nmp);
297 	while ((error = nfs_connect(nmp, rep)) != 0) {
298 		if (error == EINTR || error == ERESTART)
299 			return (EINTR);
300 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0);
301 	}
302 
303 	/*
304 	 * Loop through outstanding request list and fix up all requests
305 	 * on old socket.
306 	 */
307 	for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) {
308 		if (rp->r_nmp == nmp)
309 			rp->r_flags |= R_MUSTRESEND;
310 	}
311 	return (0);
312 }
313 
314 /*
315  * NFS disconnect. Clean up and unlink.
316  */
317 void
318 nfs_disconnect(nmp)
319 	register struct nfsmount *nmp;
320 {
321 	register struct socket *so;
322 
323 	if (nmp->nm_so) {
324 		so = nmp->nm_so;
325 		nmp->nm_so = (struct socket *)0;
326 		soshutdown(so, 2);
327 		soclose(so);
328 	}
329 }
330 
331 /*
332  * This is the nfs send routine. For connection based socket types, it
333  * must be called with an nfs_sndlock() on the socket.
334  * "rep == NULL" indicates that it has been called from a server.
335  * For the client side:
336  * - return EINTR if the RPC is terminated, 0 otherwise
337  * - set R_MUSTRESEND if the send fails for any reason
338  * - do any cleanup required by recoverable socket errors (???)
339  * For the server side:
340  * - return EINTR or ERESTART if interrupted by a signal
341  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
342  * - do any cleanup required by recoverable socket errors (???)
343  */
344 int
345 nfs_send(so, nam, top, rep)
346 	register struct socket *so;
347 	struct mbuf *nam;
348 	register struct mbuf *top;
349 	struct nfsreq *rep;
350 {
351 	struct mbuf *sendnam;
352 	int error, soflags, flags;
353 
354 	if (rep) {
355 		if (rep->r_flags & R_SOFTTERM) {
356 			m_freem(top);
357 			return (EINTR);
358 		}
359 		if ((so = rep->r_nmp->nm_so) == NULL) {
360 			rep->r_flags |= R_MUSTRESEND;
361 			m_freem(top);
362 			return (0);
363 		}
364 		rep->r_flags &= ~R_MUSTRESEND;
365 		soflags = rep->r_nmp->nm_soflags;
366 	} else
367 		soflags = so->so_proto->pr_flags;
368 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
369 		sendnam = (struct mbuf *)0;
370 	else
371 		sendnam = nam;
372 	if (so->so_type == SOCK_SEQPACKET)
373 		flags = MSG_EOR;
374 	else
375 		flags = 0;
376 
377 	error = sosend(so, sendnam, (struct uio *)0, top,
378 		(struct mbuf *)0, flags);
379 	if (error) {
380 		if (rep) {
381 			log(LOG_INFO, "nfs send error %d for server %s\n",error,
382 			    rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
383 			/*
384 			 * Deal with errors for the client side.
385 			 */
386 			if (rep->r_flags & R_SOFTTERM)
387 				error = EINTR;
388 			else
389 				rep->r_flags |= R_MUSTRESEND;
390 		} else
391 			log(LOG_INFO, "nfsd send error %d\n", error);
392 
393 		/*
394 		 * Handle any recoverable (soft) socket errors here. (???)
395 		 */
396 		if (error != EINTR && error != ERESTART &&
397 			error != EWOULDBLOCK && error != EPIPE)
398 			error = 0;
399 	}
400 	return (error);
401 }
402 
403 #ifdef NFS
404 /*
405  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
406  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
407  * Mark and consolidate the data into a new mbuf list.
408  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
409  *     small mbufs.
410  * For SOCK_STREAM we must be very careful to read an entire record once
411  * we have read any of it, even if the system call has been interrupted.
412  */
413 int
414 nfs_receive(rep, aname, mp)
415 	register struct nfsreq *rep;
416 	struct mbuf **aname;
417 	struct mbuf **mp;
418 {
419 	register struct socket *so;
420 	struct uio auio;
421 	struct iovec aio;
422 	register struct mbuf *m;
423 	struct mbuf *control;
424 	u_int32_t len;
425 	struct mbuf **getnam;
426 	int error, sotype, rcvflg;
427 	struct proc *p = curproc;	/* XXX */
428 
429 	/*
430 	 * Set up arguments for soreceive()
431 	 */
432 	*mp = (struct mbuf *)0;
433 	*aname = (struct mbuf *)0;
434 	sotype = rep->r_nmp->nm_sotype;
435 
436 	/*
437 	 * For reliable protocols, lock against other senders/receivers
438 	 * in case a reconnect is necessary.
439 	 * For SOCK_STREAM, first get the Record Mark to find out how much
440 	 * more there is to get.
441 	 * We must lock the socket against other receivers
442 	 * until we have an entire rpc request/reply.
443 	 */
444 	if (sotype != SOCK_DGRAM) {
445 		error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
446 		if (error)
447 			return (error);
448 tryagain:
449 		/*
450 		 * Check for fatal errors and resending request.
451 		 */
452 		/*
453 		 * Ugh: If a reconnect attempt just happened, nm_so
454 		 * would have changed. NULL indicates a failed
455 		 * attempt that has essentially shut down this
456 		 * mount point.
457 		 */
458 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
459 			nfs_sndunlock(&rep->r_nmp->nm_iflag);
460 			return (EINTR);
461 		}
462 		so = rep->r_nmp->nm_so;
463 		if (!so) {
464 			error = nfs_reconnect(rep);
465 			if (error) {
466 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
467 				return (error);
468 			}
469 			goto tryagain;
470 		}
471 		while (rep->r_flags & R_MUSTRESEND) {
472 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
473 			nfsstats.rpcretries++;
474 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
475 			if (error) {
476 				if (error == EINTR || error == ERESTART ||
477 				    (error = nfs_reconnect(rep)) != 0) {
478 					nfs_sndunlock(&rep->r_nmp->nm_iflag);
479 					return (error);
480 				}
481 				goto tryagain;
482 			}
483 		}
484 		nfs_sndunlock(&rep->r_nmp->nm_iflag);
485 		if (sotype == SOCK_STREAM) {
486 			aio.iov_base = (caddr_t) &len;
487 			aio.iov_len = sizeof(u_int32_t);
488 			auio.uio_iov = &aio;
489 			auio.uio_iovcnt = 1;
490 			auio.uio_segflg = UIO_SYSSPACE;
491 			auio.uio_rw = UIO_READ;
492 			auio.uio_offset = 0;
493 			auio.uio_resid = sizeof(u_int32_t);
494 			auio.uio_procp = p;
495 			do {
496 			   rcvflg = MSG_WAITALL;
497 			   error = soreceive(so, (struct mbuf **)0, &auio,
498 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
499 			   if (error == EWOULDBLOCK && rep) {
500 				if (rep->r_flags & R_SOFTTERM)
501 					return (EINTR);
502 			   }
503 			} while (error == EWOULDBLOCK);
504 			if (!error && auio.uio_resid > 0) {
505 			    log(LOG_INFO,
506 				 "short receive (%d/%d) from nfs server %s\n",
507 				 sizeof(u_int32_t) - auio.uio_resid,
508 				 sizeof(u_int32_t),
509 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
510 			    error = EPIPE;
511 			}
512 			if (error)
513 				goto errout;
514 			len = ntohl(len) & ~0x80000000;
515 			/*
516 			 * This is SERIOUS! We are out of sync with the sender
517 			 * and forcing a disconnect/reconnect is all I can do.
518 			 */
519 			if (len > NFS_MAXPACKET) {
520 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
521 				"impossible packet length",
522 				len,
523 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
524 			    error = EFBIG;
525 			    goto errout;
526 			}
527 			auio.uio_resid = len;
528 			do {
529 			    rcvflg = MSG_WAITALL;
530 			    error =  soreceive(so, (struct mbuf **)0,
531 				&auio, mp, (struct mbuf **)0, &rcvflg);
532 			} while (error == EWOULDBLOCK || error == EINTR ||
533 				 error == ERESTART);
534 			if (!error && auio.uio_resid > 0) {
535 			    log(LOG_INFO,
536 				"short receive (%d/%d) from nfs server %s\n",
537 				len - auio.uio_resid, len,
538 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
539 			    error = EPIPE;
540 			}
541 		} else {
542 			/*
543 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
544 			 * and soreceive() will return when it has either a
545 			 * control msg or a data msg.
546 			 * We have no use for control msg., but must grab them
547 			 * and then throw them away so we know what is going
548 			 * on.
549 			 */
550 			auio.uio_resid = len = 100000000; /* Anything Big */
551 			auio.uio_procp = p;
552 			do {
553 			    rcvflg = 0;
554 			    error =  soreceive(so, (struct mbuf **)0,
555 				&auio, mp, &control, &rcvflg);
556 			    if (control)
557 				m_freem(control);
558 			    if (error == EWOULDBLOCK && rep) {
559 				if (rep->r_flags & R_SOFTTERM)
560 					return (EINTR);
561 			    }
562 			} while (error == EWOULDBLOCK ||
563 				 (!error && *mp == NULL && control));
564 			if ((rcvflg & MSG_EOR) == 0)
565 				printf("Egad!!\n");
566 			if (!error && *mp == NULL)
567 				error = EPIPE;
568 			len -= auio.uio_resid;
569 		}
570 errout:
571 		if (error && error != EINTR && error != ERESTART) {
572 			m_freem(*mp);
573 			*mp = (struct mbuf *)0;
574 			if (error != EPIPE)
575 				log(LOG_INFO,
576 				    "receive error %d from nfs server %s\n",
577 				    error,
578 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
579 			error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
580 			if (!error)
581 				error = nfs_reconnect(rep);
582 			if (!error)
583 				goto tryagain;
584 			else
585 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
586 		}
587 	} else {
588 		if ((so = rep->r_nmp->nm_so) == NULL)
589 			return (EACCES);
590 		if (so->so_state & SS_ISCONNECTED)
591 			getnam = (struct mbuf **)0;
592 		else
593 			getnam = aname;
594 		auio.uio_resid = len = 1000000;
595 		auio.uio_procp = p;
596 		do {
597 			rcvflg = 0;
598 			error =  soreceive(so, getnam, &auio, mp,
599 				(struct mbuf **)0, &rcvflg);
600 			if (error == EWOULDBLOCK &&
601 			    (rep->r_flags & R_SOFTTERM))
602 				return (EINTR);
603 		} while (error == EWOULDBLOCK);
604 		len -= auio.uio_resid;
605 	}
606 	if (error) {
607 		m_freem(*mp);
608 		*mp = (struct mbuf *)0;
609 	}
610 	return (error);
611 }
612 
613 /*
614  * Implement receipt of reply on a socket.
615  * We must search through the list of received datagrams matching them
616  * with outstanding requests using the xid, until ours is found.
617  */
618 /* ARGSUSED */
619 int
620 nfs_reply(myrep)
621 	struct nfsreq *myrep;
622 {
623 	register struct nfsreq *rep;
624 	register struct nfsmount *nmp = myrep->r_nmp;
625 	register int32_t t1;
626 	struct mbuf *mrep, *nam, *md;
627 	u_int32_t rxid, *tl;
628 	caddr_t dpos, cp2;
629 	int error;
630 
631 	/*
632 	 * Loop around until we get our own reply
633 	 */
634 	for (;;) {
635 		/*
636 		 * Lock against other receivers so that I don't get stuck in
637 		 * sbwait() after someone else has received my reply for me.
638 		 * Also necessary for connection based protocols to avoid
639 		 * race conditions during a reconnect.
640 		 */
641 		error = nfs_rcvlock(myrep);
642 		if (error == EALREADY)
643 			return (0);
644 		if (error)
645 			return (error);
646 		/*
647 		 * Get the next Rpc reply off the socket
648 		 */
649 		error = nfs_receive(myrep, &nam, &mrep);
650 		nfs_rcvunlock(&nmp->nm_iflag);
651 		if (error) {
652 
653 			/*
654 			 * Ignore routing errors on connectionless protocols??
655 			 */
656 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
657 				nmp->nm_so->so_error = 0;
658 #ifdef DEBUG
659 				printf("nfs_reply: ignoring error %d\n", error);
660 #endif
661 				if (myrep->r_flags & R_GETONEREP)
662 					return (0);
663 				continue;
664 			}
665 			return (error);
666 		}
667 		if (nam)
668 			m_freem(nam);
669 
670 		/*
671 		 * XXX: Temporary work-around for unexplained lossage with a
672 		 * XXX: netmask that's not byte-aligned, i.e. 255.255.255.192.
673 		 * XXX: See PR kern/3579 for details.
674 		 */
675 		if (mrep == 0) {
676 			printf("nfs_reply: null mbuf from nfs_receive()\n");
677 			continue;
678 		}
679 
680 		/*
681 		 * Get the xid and check that it is an rpc reply
682 		 */
683 		md = mrep;
684 		dpos = mtod(md, caddr_t);
685 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
686 		rxid = *tl++;
687 		if (*tl != rpc_reply) {
688 			if (nmp->nm_flag & NFSMNT_NQNFS) {
689 				if (nqnfs_callback(nmp, mrep, md, dpos))
690 					nfsstats.rpcinvalid++;
691 			} else {
692 				nfsstats.rpcinvalid++;
693 				m_freem(mrep);
694 			}
695 nfsmout:
696 			if (myrep->r_flags & R_GETONEREP)
697 				return (0);
698 			continue;
699 		}
700 
701 		/*
702 		 * Loop through the request list to match up the reply
703 		 * Iff no match, just drop the datagram
704 		 */
705 		for (rep = nfs_reqq.tqh_first; rep != 0;
706 		    rep = rep->r_chain.tqe_next) {
707 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
708 				/* Found it.. */
709 				rep->r_mrep = mrep;
710 				rep->r_md = md;
711 				rep->r_dpos = dpos;
712 				if (nfsrtton) {
713 					struct rttl *rt;
714 
715 					rt = &nfsrtt.rttl[nfsrtt.pos];
716 					rt->proc = rep->r_procnum;
717 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
718 					rt->sent = nmp->nm_sent;
719 					rt->cwnd = nmp->nm_cwnd;
720 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
721 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
722 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
723 					rt->tstamp = time;
724 					if (rep->r_flags & R_TIMING)
725 						rt->rtt = rep->r_rtt;
726 					else
727 						rt->rtt = 1000000;
728 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
729 				}
730 				/*
731 				 * Update congestion window.
732 				 * Do the additive increase of
733 				 * one rpc/rtt.
734 				 */
735 				if (nmp->nm_cwnd <= nmp->nm_sent) {
736 					nmp->nm_cwnd +=
737 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
738 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
739 					if (nmp->nm_cwnd > NFS_MAXCWND)
740 						nmp->nm_cwnd = NFS_MAXCWND;
741 				}
742 				rep->r_flags &= ~R_SENT;
743 				nmp->nm_sent -= NFS_CWNDSCALE;
744 				/*
745 				 * Update rtt using a gain of 0.125 on the mean
746 				 * and a gain of 0.25 on the deviation.
747 				 */
748 				if (rep->r_flags & R_TIMING) {
749 					/*
750 					 * Since the timer resolution of
751 					 * NFS_HZ is so course, it can often
752 					 * result in r_rtt == 0. Since
753 					 * r_rtt == N means that the actual
754 					 * rtt is between N+dt and N+2-dt ticks,
755 					 * add 1.
756 					 */
757 					t1 = rep->r_rtt + 1;
758 					t1 -= (NFS_SRTT(rep) >> 3);
759 					NFS_SRTT(rep) += t1;
760 					if (t1 < 0)
761 						t1 = -t1;
762 					t1 -= (NFS_SDRTT(rep) >> 2);
763 					NFS_SDRTT(rep) += t1;
764 				}
765 				nmp->nm_timeouts = 0;
766 				break;
767 			}
768 		}
769 		/*
770 		 * If not matched to a request, drop it.
771 		 * If it's mine, get out.
772 		 */
773 		if (rep == 0) {
774 			nfsstats.rpcunexpected++;
775 			m_freem(mrep);
776 		} else if (rep == myrep) {
777 			if (rep->r_mrep == NULL)
778 				panic("nfsreply nil");
779 			return (0);
780 		}
781 		if (myrep->r_flags & R_GETONEREP)
782 			return (0);
783 	}
784 }
785 
786 /*
787  * nfs_request - goes something like this
788  *	- fill in request struct
789  *	- links it into list
790  *	- calls nfs_send() for first transmit
791  *	- calls nfs_receive() to get reply
792  *	- break down rpc header and return with nfs reply pointed to
793  *	  by mrep or error
794  * nb: always frees up mreq mbuf list
795  */
796 int
797 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
798 	struct vnode *vp;
799 	struct mbuf *mrest;
800 	int procnum;
801 	struct proc *procp;
802 	struct ucred *cred;
803 	struct mbuf **mrp;
804 	struct mbuf **mdp;
805 	caddr_t *dposp;
806 {
807 	register struct mbuf *m, *mrep;
808 	register struct nfsreq *rep;
809 	register u_int32_t *tl;
810 	register int i;
811 	struct nfsmount *nmp;
812 	struct mbuf *md, *mheadend;
813 	struct nfsnode *np;
814 	char nickv[RPCX_NICKVERF];
815 	time_t reqtime, waituntil;
816 	caddr_t dpos, cp2;
817 	int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type;
818 	int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0;
819 	int verf_len, verf_type;
820 	u_int32_t xid;
821 	u_quad_t frev;
822 	char *auth_str, *verf_str;
823 	NFSKERBKEY_T key;		/* save session key */
824 
825 	nmp = VFSTONFS(vp->v_mount);
826 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
827 	rep->r_nmp = nmp;
828 	rep->r_vp = vp;
829 	rep->r_procp = procp;
830 	rep->r_procnum = procnum;
831 	i = 0;
832 	m = mrest;
833 	while (m) {
834 		i += m->m_len;
835 		m = m->m_next;
836 	}
837 	mrest_len = i;
838 
839 	/*
840 	 * Get the RPC header with authorization.
841 	 */
842 kerbauth:
843 	verf_str = auth_str = (char *)0;
844 	if (nmp->nm_flag & NFSMNT_KERB) {
845 		verf_str = nickv;
846 		verf_len = sizeof (nickv);
847 		auth_type = RPCAUTH_KERB4;
848 		bzero((caddr_t)key, sizeof (key));
849 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
850 			&auth_len, verf_str, verf_len)) {
851 			error = nfs_getauth(nmp, rep, cred, &auth_str,
852 				&auth_len, verf_str, &verf_len, key);
853 			if (error) {
854 				free((caddr_t)rep, M_NFSREQ);
855 				m_freem(mrest);
856 				return (error);
857 			}
858 		}
859 	} else {
860 		auth_type = RPCAUTH_UNIX;
861 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
862 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
863 			5 * NFSX_UNSIGNED;
864 	}
865 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
866 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
867 	if (auth_str)
868 		free(auth_str, M_TEMP);
869 
870 	/*
871 	 * For stream protocols, insert a Sun RPC Record Mark.
872 	 */
873 	if (nmp->nm_sotype == SOCK_STREAM) {
874 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
875 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
876 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
877 	}
878 	rep->r_mreq = m;
879 	rep->r_xid = xid;
880 tryagain:
881 	if (nmp->nm_flag & NFSMNT_SOFT)
882 		rep->r_retry = nmp->nm_retry;
883 	else
884 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
885 	rep->r_rtt = rep->r_rexmit = 0;
886 	if (proct[procnum] > 0)
887 		rep->r_flags = R_TIMING;
888 	else
889 		rep->r_flags = 0;
890 	rep->r_mrep = NULL;
891 
892 	/*
893 	 * Do the client side RPC.
894 	 */
895 	nfsstats.rpcrequests++;
896 	/*
897 	 * Chain request into list of outstanding requests. Be sure
898 	 * to put it LAST so timer finds oldest requests first.
899 	 */
900 	s = splsoftnet();
901 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
902 
903 	/* Get send time for nqnfs */
904 	reqtime = time.tv_sec;
905 
906 	/*
907 	 * If backing off another request or avoiding congestion, don't
908 	 * send this one now but let timer do it. If not timing a request,
909 	 * do it now.
910 	 */
911 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
912 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
913 		nmp->nm_sent < nmp->nm_cwnd)) {
914 		splx(s);
915 		if (nmp->nm_soflags & PR_CONNREQUIRED)
916 			error = nfs_sndlock(&nmp->nm_iflag, rep);
917 		if (!error) {
918 			m = m_copym(m, 0, M_COPYALL, M_WAIT);
919 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep);
920 			if (nmp->nm_soflags & PR_CONNREQUIRED)
921 				nfs_sndunlock(&nmp->nm_iflag);
922 		}
923 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
924 			nmp->nm_sent += NFS_CWNDSCALE;
925 			rep->r_flags |= R_SENT;
926 		}
927 	} else {
928 		splx(s);
929 		rep->r_rtt = -1;
930 	}
931 
932 	/*
933 	 * Wait for the reply from our send or the timer's.
934 	 */
935 	if (!error || error == EPIPE)
936 		error = nfs_reply(rep);
937 
938 	/*
939 	 * RPC done, unlink the request.
940 	 */
941 	s = splsoftnet();
942 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
943 	splx(s);
944 
945 	/*
946 	 * Decrement the outstanding request count.
947 	 */
948 	if (rep->r_flags & R_SENT) {
949 		rep->r_flags &= ~R_SENT;	/* paranoia */
950 		nmp->nm_sent -= NFS_CWNDSCALE;
951 	}
952 
953 	/*
954 	 * If there was a successful reply and a tprintf msg.
955 	 * tprintf a response.
956 	 */
957 	if (!error && (rep->r_flags & R_TPRINTFMSG))
958 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
959 		    "is alive again");
960 	mrep = rep->r_mrep;
961 	md = rep->r_md;
962 	dpos = rep->r_dpos;
963 	if (error) {
964 		m_freem(rep->r_mreq);
965 		free((caddr_t)rep, M_NFSREQ);
966 		return (error);
967 	}
968 
969 	/*
970 	 * break down the rpc header and check if ok
971 	 */
972 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
973 	if (*tl++ == rpc_msgdenied) {
974 		if (*tl == rpc_mismatch)
975 			error = EOPNOTSUPP;
976 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
977 			if (!failed_auth) {
978 				failed_auth++;
979 				mheadend->m_next = (struct mbuf *)0;
980 				m_freem(mrep);
981 				m_freem(rep->r_mreq);
982 				goto kerbauth;
983 			} else
984 				error = EAUTH;
985 		} else
986 			error = EACCES;
987 		m_freem(mrep);
988 		m_freem(rep->r_mreq);
989 		free((caddr_t)rep, M_NFSREQ);
990 		return (error);
991 	}
992 
993 	/*
994 	 * Grab any Kerberos verifier, otherwise just throw it away.
995 	 */
996 	verf_type = fxdr_unsigned(int, *tl++);
997 	i = fxdr_unsigned(int32_t, *tl);
998 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
999 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1000 		if (error)
1001 			goto nfsmout;
1002 	} else if (i > 0)
1003 		nfsm_adv(nfsm_rndup(i));
1004 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1005 	/* 0 == ok */
1006 	if (*tl == 0) {
1007 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1008 		if (*tl != 0) {
1009 			error = fxdr_unsigned(int, *tl);
1010 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1011 				error == NFSERR_TRYLATER) {
1012 				m_freem(mrep);
1013 				error = 0;
1014 				waituntil = time.tv_sec + trylater_delay;
1015 				while (time.tv_sec < waituntil)
1016 					(void) tsleep((caddr_t)&lbolt,
1017 						PSOCK, "nqnfstry", 0);
1018 				trylater_delay *= nfs_backoff[trylater_cnt];
1019 				if (trylater_cnt < 7)
1020 					trylater_cnt++;
1021 				goto tryagain;
1022 			}
1023 
1024 			/*
1025 			 * If the File Handle was stale, invalidate the
1026 			 * lookup cache, just in case.
1027 			 */
1028 			if (error == ESTALE)
1029 				cache_purge(vp);
1030 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1031 				*mrp = mrep;
1032 				*mdp = md;
1033 				*dposp = dpos;
1034 				error |= NFSERR_RETERR;
1035 			} else
1036 				m_freem(mrep);
1037 			m_freem(rep->r_mreq);
1038 			free((caddr_t)rep, M_NFSREQ);
1039 			return (error);
1040 		}
1041 
1042 		/*
1043 		 * For nqnfs, get any lease in reply
1044 		 */
1045 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1046 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1047 			if (*tl) {
1048 				np = VTONFS(vp);
1049 				nqlflag = fxdr_unsigned(int, *tl);
1050 				nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1051 				cachable = fxdr_unsigned(int, *tl++);
1052 				reqtime += fxdr_unsigned(int, *tl++);
1053 				if (reqtime > time.tv_sec) {
1054 				    fxdr_hyper(tl, &frev);
1055 				    nqnfs_clientlease(nmp, np, nqlflag,
1056 					cachable, reqtime, frev);
1057 				}
1058 			}
1059 		}
1060 		*mrp = mrep;
1061 		*mdp = md;
1062 		*dposp = dpos;
1063 		m_freem(rep->r_mreq);
1064 		FREE((caddr_t)rep, M_NFSREQ);
1065 		return (0);
1066 	}
1067 	m_freem(mrep);
1068 	error = EPROTONOSUPPORT;
1069 nfsmout:
1070 	m_freem(rep->r_mreq);
1071 	free((caddr_t)rep, M_NFSREQ);
1072 	return (error);
1073 }
1074 #endif /* NFS */
1075 
1076 /*
1077  * Generate the rpc reply header
1078  * siz arg. is used to decide if adding a cluster is worthwhile
1079  */
1080 int
1081 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1082 	int siz;
1083 	struct nfsrv_descript *nd;
1084 	struct nfssvc_sock *slp;
1085 	int err;
1086 	int cache;
1087 	u_quad_t *frev;
1088 	struct mbuf **mrq;
1089 	struct mbuf **mbp;
1090 	caddr_t *bposp;
1091 {
1092 	register u_int32_t *tl;
1093 	register struct mbuf *mreq;
1094 	caddr_t bpos;
1095 	struct mbuf *mb, *mb2;
1096 
1097 	MGETHDR(mreq, M_WAIT, MT_DATA);
1098 	mb = mreq;
1099 	/*
1100 	 * If this is a big reply, use a cluster else
1101 	 * try and leave leading space for the lower level headers.
1102 	 */
1103 	siz += RPC_REPLYSIZ;
1104 	if (siz >= MINCLSIZE) {
1105 		MCLGET(mreq, M_WAIT);
1106 	} else
1107 		mreq->m_data += max_hdr;
1108 	tl = mtod(mreq, u_int32_t *);
1109 	mreq->m_len = 6 * NFSX_UNSIGNED;
1110 	bpos = ((caddr_t)tl) + mreq->m_len;
1111 	*tl++ = txdr_unsigned(nd->nd_retxid);
1112 	*tl++ = rpc_reply;
1113 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1114 		*tl++ = rpc_msgdenied;
1115 		if (err & NFSERR_AUTHERR) {
1116 			*tl++ = rpc_autherr;
1117 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1118 			mreq->m_len -= NFSX_UNSIGNED;
1119 			bpos -= NFSX_UNSIGNED;
1120 		} else {
1121 			*tl++ = rpc_mismatch;
1122 			*tl++ = txdr_unsigned(RPC_VER2);
1123 			*tl = txdr_unsigned(RPC_VER2);
1124 		}
1125 	} else {
1126 		*tl++ = rpc_msgaccepted;
1127 
1128 		/*
1129 		 * For Kerberos authentication, we must send the nickname
1130 		 * verifier back, otherwise just RPCAUTH_NULL.
1131 		 */
1132 		if (nd->nd_flag & ND_KERBFULL) {
1133 		    register struct nfsuid *nuidp;
1134 		    struct timeval ktvin, ktvout;
1135 
1136 		    for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1137 			nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1138 			if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1139 			    (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1140 			     &nuidp->nu_haddr, nd->nd_nam2)))
1141 			    break;
1142 		    }
1143 		    if (nuidp) {
1144 			ktvin.tv_sec =
1145 			    txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1146 			ktvin.tv_usec =
1147 			    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1148 
1149 			/*
1150 			 * Encrypt the timestamp in ecb mode using the
1151 			 * session key.
1152 			 */
1153 #ifdef NFSKERB
1154 			XXX
1155 #endif
1156 
1157 			*tl++ = rpc_auth_kerb;
1158 			*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1159 			*tl = ktvout.tv_sec;
1160 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1161 			*tl++ = ktvout.tv_usec;
1162 			*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1163 		    } else {
1164 			*tl++ = 0;
1165 			*tl++ = 0;
1166 		    }
1167 		} else {
1168 			*tl++ = 0;
1169 			*tl++ = 0;
1170 		}
1171 		switch (err) {
1172 		case EPROGUNAVAIL:
1173 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1174 			break;
1175 		case EPROGMISMATCH:
1176 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1177 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1178 			if (nd->nd_flag & ND_NQNFS) {
1179 				*tl++ = txdr_unsigned(3);
1180 				*tl = txdr_unsigned(3);
1181 			} else {
1182 				*tl++ = txdr_unsigned(2);
1183 				*tl = txdr_unsigned(3);
1184 			}
1185 			break;
1186 		case EPROCUNAVAIL:
1187 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1188 			break;
1189 		case EBADRPC:
1190 			*tl = txdr_unsigned(RPC_GARBAGE);
1191 			break;
1192 		default:
1193 			*tl = 0;
1194 			if (err != NFSERR_RETVOID) {
1195 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1196 				if (err)
1197 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1198 				else
1199 				    *tl = 0;
1200 			}
1201 			break;
1202 		};
1203 	}
1204 
1205 	/*
1206 	 * For nqnfs, piggyback lease as requested.
1207 	 */
1208 	if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1209 		if (nd->nd_flag & ND_LEASE) {
1210 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1211 			*tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1212 			*tl++ = txdr_unsigned(cache);
1213 			*tl++ = txdr_unsigned(nd->nd_duration);
1214 			txdr_hyper(frev, tl);
1215 		} else {
1216 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1217 			*tl = 0;
1218 		}
1219 	}
1220 	if (mrq != NULL)
1221 		*mrq = mreq;
1222 	*mbp = mb;
1223 	*bposp = bpos;
1224 	if (err != 0 && err != NFSERR_RETVOID)
1225 		nfsstats.srvrpc_errs++;
1226 	return (0);
1227 }
1228 
1229 /*
1230  * Nfs timer routine
1231  * Scan the nfsreq list and retranmit any requests that have timed out
1232  * To avoid retransmission attempts on STREAM sockets (in the future) make
1233  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1234  */
1235 void
1236 nfs_timer(arg)
1237 	void *arg;	/* never used */
1238 {
1239 	register struct nfsreq *rep;
1240 	register struct mbuf *m;
1241 	register struct socket *so;
1242 	register struct nfsmount *nmp;
1243 	register int timeo;
1244 	int s, error;
1245 #ifdef NFSSERVER
1246 	register struct nfssvc_sock *slp;
1247 	static long lasttime = 0;
1248 	u_quad_t cur_usec;
1249 #endif
1250 
1251 	s = splsoftnet();
1252 	for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) {
1253 		nmp = rep->r_nmp;
1254 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1255 			continue;
1256 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1257 			rep->r_flags |= R_SOFTTERM;
1258 			continue;
1259 		}
1260 		if (rep->r_rtt >= 0) {
1261 			rep->r_rtt++;
1262 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1263 				timeo = nmp->nm_timeo;
1264 			else
1265 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1266 			if (nmp->nm_timeouts > 0)
1267 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1268 			if (rep->r_rtt <= timeo)
1269 				continue;
1270 			if (nmp->nm_timeouts < 8)
1271 				nmp->nm_timeouts++;
1272 		}
1273 		/*
1274 		 * Check for server not responding
1275 		 */
1276 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1277 		     rep->r_rexmit > nmp->nm_deadthresh) {
1278 			nfs_msg(rep->r_procp,
1279 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1280 			    "not responding");
1281 			rep->r_flags |= R_TPRINTFMSG;
1282 		}
1283 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1284 			nfsstats.rpctimeouts++;
1285 			rep->r_flags |= R_SOFTTERM;
1286 			continue;
1287 		}
1288 		if (nmp->nm_sotype != SOCK_DGRAM) {
1289 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1290 				rep->r_rexmit = NFS_MAXREXMIT;
1291 			continue;
1292 		}
1293 		if ((so = nmp->nm_so) == NULL)
1294 			continue;
1295 
1296 		/*
1297 		 * If there is enough space and the window allows..
1298 		 *	Resend it
1299 		 * Set r_rtt to -1 in case we fail to send it now.
1300 		 */
1301 		rep->r_rtt = -1;
1302 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1303 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1304 		    (rep->r_flags & R_SENT) ||
1305 		    nmp->nm_sent < nmp->nm_cwnd) &&
1306 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1307 			if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
1308 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1309 			    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1310 			else
1311 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1312 			    nmp->nm_nam, (struct mbuf *)0, (struct proc *)0);
1313 			if (error) {
1314 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1315 #ifdef DEBUG
1316 					printf("nfs_timer: ignoring error %d\n",
1317 						error);
1318 #endif
1319 					so->so_error = 0;
1320 				}
1321 			} else {
1322 				/*
1323 				 * Iff first send, start timing
1324 				 * else turn timing off, backoff timer
1325 				 * and divide congestion window by 2.
1326 				 */
1327 				if (rep->r_flags & R_SENT) {
1328 					rep->r_flags &= ~R_TIMING;
1329 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1330 						rep->r_rexmit = NFS_MAXREXMIT;
1331 					nmp->nm_cwnd >>= 1;
1332 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1333 						nmp->nm_cwnd = NFS_CWNDSCALE;
1334 					nfsstats.rpcretries++;
1335 				} else {
1336 					rep->r_flags |= R_SENT;
1337 					nmp->nm_sent += NFS_CWNDSCALE;
1338 				}
1339 				rep->r_rtt = 0;
1340 			}
1341 		}
1342 	}
1343 
1344 #ifdef NFSSERVER
1345 	/*
1346 	 * Call the nqnfs server timer once a second to handle leases.
1347 	 */
1348 	if (lasttime != time.tv_sec) {
1349 		lasttime = time.tv_sec;
1350 		nqnfs_serverd();
1351 	}
1352 
1353 	/*
1354 	 * Scan the write gathering queues for writes that need to be
1355 	 * completed now.
1356 	 */
1357 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1358 	for (slp = nfssvc_sockhead.tqh_first; slp != 0;
1359 	    slp = slp->ns_chain.tqe_next) {
1360 	    if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1361 		nfsrv_wakenfsd(slp);
1362 	}
1363 #endif /* NFSSERVER */
1364 	splx(s);
1365 	timeout(nfs_timer, (void *)0, nfs_ticks);
1366 }
1367 
1368 /*
1369  * Test for a termination condition pending on the process.
1370  * This is used for NFSMNT_INT mounts.
1371  */
1372 int
1373 nfs_sigintr(nmp, rep, p)
1374 	struct nfsmount *nmp;
1375 	struct nfsreq *rep;
1376 	register struct proc *p;
1377 {
1378 
1379 	if (rep && (rep->r_flags & R_SOFTTERM))
1380 		return (EINTR);
1381 	if (!(nmp->nm_flag & NFSMNT_INT))
1382 		return (0);
1383 	if (p && p->p_siglist &&
1384 	    (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) &
1385 	    NFSINT_SIGMASK))
1386 		return (EINTR);
1387 	return (0);
1388 }
1389 
1390 /*
1391  * Lock a socket against others.
1392  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1393  * and also to avoid race conditions between the processes with nfs requests
1394  * in progress when a reconnect is necessary.
1395  */
1396 int
1397 nfs_sndlock(flagp, rep)
1398 	register int *flagp;
1399 	struct nfsreq *rep;
1400 {
1401 	struct proc *p;
1402 	int slpflag = 0, slptimeo = 0;
1403 
1404 	if (rep) {
1405 		p = rep->r_procp;
1406 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1407 			slpflag = PCATCH;
1408 	} else
1409 		p = (struct proc *)0;
1410 	while (*flagp & NFSMNT_SNDLOCK) {
1411 		if (nfs_sigintr(rep->r_nmp, rep, p))
1412 			return (EINTR);
1413 		*flagp |= NFSMNT_WANTSND;
1414 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1415 			slptimeo);
1416 		if (slpflag == PCATCH) {
1417 			slpflag = 0;
1418 			slptimeo = 2 * hz;
1419 		}
1420 	}
1421 	*flagp |= NFSMNT_SNDLOCK;
1422 	return (0);
1423 }
1424 
1425 /*
1426  * Unlock the stream socket for others.
1427  */
1428 void
1429 nfs_sndunlock(flagp)
1430 	register int *flagp;
1431 {
1432 
1433 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1434 		panic("nfs sndunlock");
1435 	*flagp &= ~NFSMNT_SNDLOCK;
1436 	if (*flagp & NFSMNT_WANTSND) {
1437 		*flagp &= ~NFSMNT_WANTSND;
1438 		wakeup((caddr_t)flagp);
1439 	}
1440 }
1441 
1442 int
1443 nfs_rcvlock(rep)
1444 	register struct nfsreq *rep;
1445 {
1446 	register int *flagp = &rep->r_nmp->nm_iflag;
1447 	int slpflag, slptimeo = 0;
1448 
1449 	if (*flagp & NFSMNT_INT)
1450 		slpflag = PCATCH;
1451 	else
1452 		slpflag = 0;
1453 	while (*flagp & NFSMNT_RCVLOCK) {
1454 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1455 			return (EINTR);
1456 		*flagp |= NFSMNT_WANTRCV;
1457 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1458 			slptimeo);
1459 		/* If our reply was received while we were sleeping,
1460 		 * then just return without taking the lock to avoid a
1461 		 * situation where a single iod could 'capture' the
1462 		 * receive lock.
1463 		 */
1464 		if (rep->r_mrep != NULL)
1465 			return (EALREADY);
1466 		if (slpflag == PCATCH) {
1467 			slpflag = 0;
1468 			slptimeo = 2 * hz;
1469 		}
1470 	}
1471 	*flagp |= NFSMNT_RCVLOCK;
1472 	return (0);
1473 }
1474 
1475 /*
1476  * Unlock the stream socket for others.
1477  */
1478 void
1479 nfs_rcvunlock(flagp)
1480 	register int *flagp;
1481 {
1482 
1483 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1484 		panic("nfs rcvunlock");
1485 	*flagp &= ~NFSMNT_RCVLOCK;
1486 	if (*flagp & NFSMNT_WANTRCV) {
1487 		*flagp &= ~NFSMNT_WANTRCV;
1488 		wakeup((caddr_t)flagp);
1489 	}
1490 }
1491 
1492 /*
1493  * Parse an RPC request
1494  * - verify it
1495  * - fill in the cred struct.
1496  */
1497 int
1498 nfs_getreq(nd, nfsd, has_header)
1499 	register struct nfsrv_descript *nd;
1500 	struct nfsd *nfsd;
1501 	int has_header;
1502 {
1503 	register int len, i;
1504 	register u_int32_t *tl;
1505 	register int32_t t1;
1506 	struct uio uio;
1507 	struct iovec iov;
1508 	caddr_t dpos, cp2, cp;
1509 	u_int32_t nfsvers, auth_type;
1510 	uid_t nickuid;
1511 	int error = 0, nqnfs = 0, ticklen;
1512 	struct mbuf *mrep, *md;
1513 	register struct nfsuid *nuidp;
1514 	struct timeval tvin, tvout;
1515 
1516 	mrep = nd->nd_mrep;
1517 	md = nd->nd_md;
1518 	dpos = nd->nd_dpos;
1519 	if (has_header) {
1520 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1521 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1522 		if (*tl++ != rpc_call) {
1523 			m_freem(mrep);
1524 			return (EBADRPC);
1525 		}
1526 	} else
1527 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1528 	nd->nd_repstat = 0;
1529 	nd->nd_flag = 0;
1530 	if (*tl++ != rpc_vers) {
1531 		nd->nd_repstat = ERPCMISMATCH;
1532 		nd->nd_procnum = NFSPROC_NOOP;
1533 		return (0);
1534 	}
1535 	if (*tl != nfs_prog) {
1536 		if (*tl == nqnfs_prog)
1537 			nqnfs++;
1538 		else {
1539 			nd->nd_repstat = EPROGUNAVAIL;
1540 			nd->nd_procnum = NFSPROC_NOOP;
1541 			return (0);
1542 		}
1543 	}
1544 	tl++;
1545 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1546 	if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1547 		(nfsvers != NQNFS_VER3 && nqnfs)) {
1548 		nd->nd_repstat = EPROGMISMATCH;
1549 		nd->nd_procnum = NFSPROC_NOOP;
1550 		return (0);
1551 	}
1552 	if (nqnfs)
1553 		nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1554 	else if (nfsvers == NFS_VER3)
1555 		nd->nd_flag = ND_NFSV3;
1556 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1557 	if (nd->nd_procnum == NFSPROC_NULL)
1558 		return (0);
1559 	if (nd->nd_procnum >= NFS_NPROCS ||
1560 		(!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1561 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1562 		nd->nd_repstat = EPROCUNAVAIL;
1563 		nd->nd_procnum = NFSPROC_NOOP;
1564 		return (0);
1565 	}
1566 	if ((nd->nd_flag & ND_NFSV3) == 0)
1567 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1568 	auth_type = *tl++;
1569 	len = fxdr_unsigned(int, *tl++);
1570 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1571 		m_freem(mrep);
1572 		return (EBADRPC);
1573 	}
1574 
1575 	nd->nd_flag &= ~ND_KERBAUTH;
1576 	/*
1577 	 * Handle auth_unix or auth_kerb.
1578 	 */
1579 	if (auth_type == rpc_auth_unix) {
1580 		len = fxdr_unsigned(int, *++tl);
1581 		if (len < 0 || len > NFS_MAXNAMLEN) {
1582 			m_freem(mrep);
1583 			return (EBADRPC);
1584 		}
1585 		nfsm_adv(nfsm_rndup(len));
1586 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1587 		bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred));
1588 		nd->nd_cr.cr_ref = 1;
1589 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1590 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1591 		len = fxdr_unsigned(int, *tl);
1592 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1593 			m_freem(mrep);
1594 			return (EBADRPC);
1595 		}
1596 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1597 		for (i = 0; i < len; i++)
1598 		    if (i < NGROUPS)
1599 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1600 		    else
1601 			tl++;
1602 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1603 		if (nd->nd_cr.cr_ngroups > 1)
1604 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1605 		len = fxdr_unsigned(int, *++tl);
1606 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1607 			m_freem(mrep);
1608 			return (EBADRPC);
1609 		}
1610 		if (len > 0)
1611 			nfsm_adv(nfsm_rndup(len));
1612 	} else if (auth_type == rpc_auth_kerb) {
1613 		switch (fxdr_unsigned(int, *tl++)) {
1614 		case RPCAKN_FULLNAME:
1615 			ticklen = fxdr_unsigned(int, *tl);
1616 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1617 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1618 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1619 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1620 				m_freem(mrep);
1621 				return (EBADRPC);
1622 			}
1623 			uio.uio_offset = 0;
1624 			uio.uio_iov = &iov;
1625 			uio.uio_iovcnt = 1;
1626 			uio.uio_segflg = UIO_SYSSPACE;
1627 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1628 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1629 			nfsm_mtouio(&uio, uio.uio_resid);
1630 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1631 			if (*tl++ != rpc_auth_kerb ||
1632 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1633 				printf("Bad kerb verifier\n");
1634 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1635 				nd->nd_procnum = NFSPROC_NOOP;
1636 				return (0);
1637 			}
1638 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1639 			tl = (u_int32_t *)cp;
1640 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1641 				printf("Not fullname kerb verifier\n");
1642 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1643 				nd->nd_procnum = NFSPROC_NOOP;
1644 				return (0);
1645 			}
1646 			cp += NFSX_UNSIGNED;
1647 			bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED);
1648 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1649 			nd->nd_flag |= ND_KERBFULL;
1650 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1651 			break;
1652 		case RPCAKN_NICKNAME:
1653 			if (len != 2 * NFSX_UNSIGNED) {
1654 				printf("Kerb nickname short\n");
1655 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1656 				nd->nd_procnum = NFSPROC_NOOP;
1657 				return (0);
1658 			}
1659 			nickuid = fxdr_unsigned(uid_t, *tl);
1660 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1661 			if (*tl++ != rpc_auth_kerb ||
1662 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1663 				printf("Kerb nick verifier bad\n");
1664 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1665 				nd->nd_procnum = NFSPROC_NOOP;
1666 				return (0);
1667 			}
1668 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1669 			tvin.tv_sec = *tl++;
1670 			tvin.tv_usec = *tl;
1671 
1672 			for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1673 			    nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1674 				if (nuidp->nu_cr.cr_uid == nickuid &&
1675 				    (!nd->nd_nam2 ||
1676 				     netaddr_match(NU_NETFAM(nuidp),
1677 				      &nuidp->nu_haddr, nd->nd_nam2)))
1678 					break;
1679 			}
1680 			if (!nuidp) {
1681 				nd->nd_repstat =
1682 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1683 				nd->nd_procnum = NFSPROC_NOOP;
1684 				return (0);
1685 			}
1686 
1687 			/*
1688 			 * Now, decrypt the timestamp using the session key
1689 			 * and validate it.
1690 			 */
1691 #ifdef NFSKERB
1692 			XXX
1693 #endif
1694 
1695 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1696 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1697 			if (nuidp->nu_expire < time.tv_sec ||
1698 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1699 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1700 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1701 				nuidp->nu_expire = 0;
1702 				nd->nd_repstat =
1703 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1704 				nd->nd_procnum = NFSPROC_NOOP;
1705 				return (0);
1706 			}
1707 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1708 			nd->nd_flag |= ND_KERBNICK;
1709 		};
1710 	} else {
1711 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1712 		nd->nd_procnum = NFSPROC_NOOP;
1713 		return (0);
1714 	}
1715 
1716 	/*
1717 	 * For nqnfs, get piggybacked lease request.
1718 	 */
1719 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
1720 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1721 		nd->nd_flag |= fxdr_unsigned(int, *tl);
1722 		if (nd->nd_flag & ND_LEASE) {
1723 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1724 			nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
1725 		} else
1726 			nd->nd_duration = NQ_MINLEASE;
1727 	} else
1728 		nd->nd_duration = NQ_MINLEASE;
1729 	nd->nd_md = md;
1730 	nd->nd_dpos = dpos;
1731 	return (0);
1732 nfsmout:
1733 	return (error);
1734 }
1735 
1736 int
1737 nfs_msg(p, server, msg)
1738 	struct proc *p;
1739 	char *server, *msg;
1740 {
1741 	tpr_t tpr;
1742 
1743 	if (p)
1744 		tpr = tprintf_open(p);
1745 	else
1746 		tpr = NULL;
1747 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1748 	tprintf_close(tpr);
1749 	return (0);
1750 }
1751 
1752 #ifdef NFSSERVER
1753 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
1754 				    struct nfssvc_sock *, struct proc *,
1755 				    struct mbuf **)) = {
1756 	nfsrv_null,
1757 	nfsrv_getattr,
1758 	nfsrv_setattr,
1759 	nfsrv_lookup,
1760 	nfsrv3_access,
1761 	nfsrv_readlink,
1762 	nfsrv_read,
1763 	nfsrv_write,
1764 	nfsrv_create,
1765 	nfsrv_mkdir,
1766 	nfsrv_symlink,
1767 	nfsrv_mknod,
1768 	nfsrv_remove,
1769 	nfsrv_rmdir,
1770 	nfsrv_rename,
1771 	nfsrv_link,
1772 	nfsrv_readdir,
1773 	nfsrv_readdirplus,
1774 	nfsrv_statfs,
1775 	nfsrv_fsinfo,
1776 	nfsrv_pathconf,
1777 	nfsrv_commit,
1778 	nqnfsrv_getlease,
1779 	nqnfsrv_vacated,
1780 	nfsrv_noop,
1781 	nfsrv_noop
1782 };
1783 
1784 /*
1785  * Socket upcall routine for the nfsd sockets.
1786  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1787  * Essentially do as much as possible non-blocking, else punt and it will
1788  * be called with M_WAIT from an nfsd.
1789  */
1790 void
1791 nfsrv_rcv(so, arg, waitflag)
1792 	struct socket *so;
1793 	caddr_t arg;
1794 	int waitflag;
1795 {
1796 	register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1797 	register struct mbuf *m;
1798 	struct mbuf *mp, *nam;
1799 	struct uio auio;
1800 	int flags, error;
1801 
1802 	if ((slp->ns_flag & SLP_VALID) == 0)
1803 		return;
1804 #ifdef notdef
1805 	/*
1806 	 * Define this to test for nfsds handling this under heavy load.
1807 	 */
1808 	if (waitflag == M_DONTWAIT) {
1809 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1810 	}
1811 #endif
1812 	auio.uio_procp = NULL;
1813 	if (so->so_type == SOCK_STREAM) {
1814 		/*
1815 		 * If there are already records on the queue, defer soreceive()
1816 		 * to an nfsd so that there is feedback to the TCP layer that
1817 		 * the nfs servers are heavily loaded.
1818 		 */
1819 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1820 			slp->ns_flag |= SLP_NEEDQ;
1821 			goto dorecs;
1822 		}
1823 
1824 		/*
1825 		 * Do soreceive().
1826 		 */
1827 		auio.uio_resid = 1000000000;
1828 		flags = MSG_DONTWAIT;
1829 		error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1830 		if (error || mp == (struct mbuf *)0) {
1831 			if (error == EWOULDBLOCK)
1832 				slp->ns_flag |= SLP_NEEDQ;
1833 			else
1834 				slp->ns_flag |= SLP_DISCONN;
1835 			goto dorecs;
1836 		}
1837 		m = mp;
1838 		if (slp->ns_rawend) {
1839 			slp->ns_rawend->m_next = m;
1840 			slp->ns_cc += 1000000000 - auio.uio_resid;
1841 		} else {
1842 			slp->ns_raw = m;
1843 			slp->ns_cc = 1000000000 - auio.uio_resid;
1844 		}
1845 		while (m->m_next)
1846 			m = m->m_next;
1847 		slp->ns_rawend = m;
1848 
1849 		/*
1850 		 * Now try and parse record(s) out of the raw stream data.
1851 		 */
1852 		error = nfsrv_getstream(slp, waitflag);
1853 		if (error) {
1854 			if (error == EPERM)
1855 				slp->ns_flag |= SLP_DISCONN;
1856 			else
1857 				slp->ns_flag |= SLP_NEEDQ;
1858 		}
1859 	} else {
1860 		do {
1861 			auio.uio_resid = 1000000000;
1862 			flags = MSG_DONTWAIT;
1863 			error = soreceive(so, &nam, &auio, &mp,
1864 						(struct mbuf **)0, &flags);
1865 			if (mp) {
1866 				if (nam) {
1867 					m = nam;
1868 					m->m_next = mp;
1869 				} else
1870 					m = mp;
1871 				if (slp->ns_recend)
1872 					slp->ns_recend->m_nextpkt = m;
1873 				else
1874 					slp->ns_rec = m;
1875 				slp->ns_recend = m;
1876 				m->m_nextpkt = (struct mbuf *)0;
1877 			}
1878 			if (error) {
1879 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
1880 					&& error != EWOULDBLOCK) {
1881 					slp->ns_flag |= SLP_DISCONN;
1882 					goto dorecs;
1883 				}
1884 			}
1885 		} while (mp);
1886 	}
1887 
1888 	/*
1889 	 * Now try and process the request records, non-blocking.
1890 	 */
1891 dorecs:
1892 	if (waitflag == M_DONTWAIT &&
1893 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
1894 		nfsrv_wakenfsd(slp);
1895 }
1896 
1897 /*
1898  * Try and extract an RPC request from the mbuf data list received on a
1899  * stream socket. The "waitflag" argument indicates whether or not it
1900  * can sleep.
1901  */
1902 int
1903 nfsrv_getstream(slp, waitflag)
1904 	register struct nfssvc_sock *slp;
1905 	int waitflag;
1906 {
1907 	register struct mbuf *m, **mpp;
1908 	register char *cp1, *cp2;
1909 	register int len;
1910 	struct mbuf *om, *m2, *recm = NULL;
1911 	u_int32_t recmark;
1912 
1913 	if (slp->ns_flag & SLP_GETSTREAM)
1914 		panic("nfs getstream");
1915 	slp->ns_flag |= SLP_GETSTREAM;
1916 	for (;;) {
1917 	    if (slp->ns_reclen == 0) {
1918 		if (slp->ns_cc < NFSX_UNSIGNED) {
1919 			slp->ns_flag &= ~SLP_GETSTREAM;
1920 			return (0);
1921 		}
1922 		m = slp->ns_raw;
1923 		if (m->m_len >= NFSX_UNSIGNED) {
1924 			bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
1925 			m->m_data += NFSX_UNSIGNED;
1926 			m->m_len -= NFSX_UNSIGNED;
1927 		} else {
1928 			cp1 = (caddr_t)&recmark;
1929 			cp2 = mtod(m, caddr_t);
1930 			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
1931 				while (m->m_len == 0) {
1932 					m = m->m_next;
1933 					cp2 = mtod(m, caddr_t);
1934 				}
1935 				*cp1++ = *cp2++;
1936 				m->m_data++;
1937 				m->m_len--;
1938 			}
1939 		}
1940 		slp->ns_cc -= NFSX_UNSIGNED;
1941 		recmark = ntohl(recmark);
1942 		slp->ns_reclen = recmark & ~0x80000000;
1943 		if (recmark & 0x80000000)
1944 			slp->ns_flag |= SLP_LASTFRAG;
1945 		else
1946 			slp->ns_flag &= ~SLP_LASTFRAG;
1947 		if (slp->ns_reclen > NFS_MAXPACKET) {
1948 			slp->ns_flag &= ~SLP_GETSTREAM;
1949 			return (EPERM);
1950 		}
1951 	    }
1952 
1953 	    /*
1954 	     * Now get the record part.
1955 	     */
1956 	    if (slp->ns_cc == slp->ns_reclen) {
1957 		recm = slp->ns_raw;
1958 		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
1959 		slp->ns_cc = slp->ns_reclen = 0;
1960 	    } else if (slp->ns_cc > slp->ns_reclen) {
1961 		len = 0;
1962 		m = slp->ns_raw;
1963 		om = (struct mbuf *)0;
1964 		while (len < slp->ns_reclen) {
1965 			if ((len + m->m_len) > slp->ns_reclen) {
1966 				size_t left = slp->ns_reclen - len;
1967 
1968 				MGETHDR(m2, waitflag, m->m_type);
1969 				if (m2 == NULL) {
1970 					slp->ns_flag &= ~SLP_GETSTREAM;
1971 					return (EWOULDBLOCK);
1972 				}
1973 				if (left > MHLEN) {
1974 					MCLGET(m2, waitflag);
1975 					if (!(m2->m_flags & M_EXT)) {
1976 						m_freem(m2);
1977 						slp->ns_flag &= ~SLP_GETSTREAM;
1978 						return (EWOULDBLOCK);
1979 					}
1980 				}
1981 				bcopy(mtod(m, caddr_t), mtod(m2, caddr_t),
1982 				    left);
1983 				m2->m_len = left;
1984 				m->m_data += left;
1985 				m->m_len -= left;
1986 				if (om) {
1987 					om->m_next = m2;
1988 					recm = slp->ns_raw;
1989 				} else
1990 					recm = m2;
1991 				len = slp->ns_reclen;
1992 			} else if ((len + m->m_len) == slp->ns_reclen) {
1993 				om = m;
1994 				len += m->m_len;
1995 				m = m->m_next;
1996 				recm = slp->ns_raw;
1997 				om->m_next = (struct mbuf *)0;
1998 			} else {
1999 				om = m;
2000 				len += m->m_len;
2001 				m = m->m_next;
2002 			}
2003 		}
2004 		slp->ns_raw = m;
2005 		slp->ns_cc -= len;
2006 		slp->ns_reclen = 0;
2007 	    } else {
2008 		slp->ns_flag &= ~SLP_GETSTREAM;
2009 		return (0);
2010 	    }
2011 
2012 	    /*
2013 	     * Accumulate the fragments into a record.
2014 	     */
2015 	    mpp = &slp->ns_frag;
2016 	    while (*mpp)
2017 		mpp = &((*mpp)->m_next);
2018 	    *mpp = recm;
2019 	    if (slp->ns_flag & SLP_LASTFRAG) {
2020 		if (slp->ns_recend)
2021 		    slp->ns_recend->m_nextpkt = slp->ns_frag;
2022 		else
2023 		    slp->ns_rec = slp->ns_frag;
2024 		slp->ns_recend = slp->ns_frag;
2025 		slp->ns_frag = (struct mbuf *)0;
2026 	    }
2027 	}
2028 }
2029 
2030 /*
2031  * Parse an RPC header.
2032  */
2033 int
2034 nfsrv_dorec(slp, nfsd, ndp)
2035 	register struct nfssvc_sock *slp;
2036 	struct nfsd *nfsd;
2037 	struct nfsrv_descript **ndp;
2038 {
2039 	register struct mbuf *m, *nam;
2040 	register struct nfsrv_descript *nd;
2041 	int error;
2042 
2043 	*ndp = NULL;
2044 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2045 	    (m = slp->ns_rec) == (struct mbuf *)0)
2046 		return (ENOBUFS);
2047 	slp->ns_rec = m->m_nextpkt;
2048 	if (slp->ns_rec)
2049 		m->m_nextpkt = (struct mbuf *)0;
2050 	else
2051 		slp->ns_recend = (struct mbuf *)0;
2052 	if (m->m_type == MT_SONAME) {
2053 		nam = m;
2054 		m = m->m_next;
2055 		nam->m_next = NULL;
2056 	} else
2057 		nam = NULL;
2058 	MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2059 		M_NFSRVDESC, M_WAITOK);
2060 	nd->nd_md = nd->nd_mrep = m;
2061 	nd->nd_nam2 = nam;
2062 	nd->nd_dpos = mtod(m, caddr_t);
2063 	error = nfs_getreq(nd, nfsd, TRUE);
2064 	if (error) {
2065 		m_freem(nam);
2066 		free((caddr_t)nd, M_NFSRVDESC);
2067 		return (error);
2068 	}
2069 	*ndp = nd;
2070 	nfsd->nfsd_nd = nd;
2071 	return (0);
2072 }
2073 
2074 
2075 /*
2076  * Search for a sleeping nfsd and wake it up.
2077  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2078  * running nfsds will go look for the work in the nfssvc_sock list.
2079  */
2080 void
2081 nfsrv_wakenfsd(slp)
2082 	struct nfssvc_sock *slp;
2083 {
2084 	register struct nfsd *nd;
2085 
2086 	if ((slp->ns_flag & SLP_VALID) == 0)
2087 		return;
2088 	for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) {
2089 		if (nd->nfsd_flag & NFSD_WAITING) {
2090 			nd->nfsd_flag &= ~NFSD_WAITING;
2091 			if (nd->nfsd_slp)
2092 				panic("nfsd wakeup");
2093 			slp->ns_sref++;
2094 			nd->nfsd_slp = slp;
2095 			wakeup((caddr_t)nd);
2096 			return;
2097 		}
2098 	}
2099 	slp->ns_flag |= SLP_DOREC;
2100 	nfsd_head_flag |= NFSD_CHECKSLP;
2101 }
2102 #endif /* NFSSERVER */
2103