xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 2a399c6883d870daece976daec6ffa7bb7f934ce)
1 /*	$NetBSD: nfs_socket.c,v 1.40 1997/11/16 23:23:20 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 		struct nfsreq dummyreq;
325 
326 		bzero(&dummyreq, sizeof(dummyreq));
327 		dummyreq.r_nmp = nmp;
328 		nfs_rcvlock(&dummyreq);
329 
330 		so = nmp->nm_so;
331 		nmp->nm_so = (struct socket *)0;
332 		soshutdown(so, 2);
333 		soclose(so);
334 
335 		nfs_rcvunlock(&nmp->nm_iflag);
336 	}
337 }
338 
339 /*
340  * This is the nfs send routine. For connection based socket types, it
341  * must be called with an nfs_sndlock() on the socket.
342  * "rep == NULL" indicates that it has been called from a server.
343  * For the client side:
344  * - return EINTR if the RPC is terminated, 0 otherwise
345  * - set R_MUSTRESEND if the send fails for any reason
346  * - do any cleanup required by recoverable socket errors (???)
347  * For the server side:
348  * - return EINTR or ERESTART if interrupted by a signal
349  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
350  * - do any cleanup required by recoverable socket errors (???)
351  */
352 int
353 nfs_send(so, nam, top, rep)
354 	register struct socket *so;
355 	struct mbuf *nam;
356 	register struct mbuf *top;
357 	struct nfsreq *rep;
358 {
359 	struct mbuf *sendnam;
360 	int error, soflags, flags;
361 
362 	if (rep) {
363 		if (rep->r_flags & R_SOFTTERM) {
364 			m_freem(top);
365 			return (EINTR);
366 		}
367 		if ((so = rep->r_nmp->nm_so) == NULL) {
368 			rep->r_flags |= R_MUSTRESEND;
369 			m_freem(top);
370 			return (0);
371 		}
372 		rep->r_flags &= ~R_MUSTRESEND;
373 		soflags = rep->r_nmp->nm_soflags;
374 	} else
375 		soflags = so->so_proto->pr_flags;
376 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
377 		sendnam = (struct mbuf *)0;
378 	else
379 		sendnam = nam;
380 	if (so->so_type == SOCK_SEQPACKET)
381 		flags = MSG_EOR;
382 	else
383 		flags = 0;
384 
385 	error = sosend(so, sendnam, (struct uio *)0, top,
386 		(struct mbuf *)0, flags);
387 	if (error) {
388 		if (rep) {
389 			log(LOG_INFO, "nfs send error %d for server %s\n",error,
390 			    rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
391 			/*
392 			 * Deal with errors for the client side.
393 			 */
394 			if (rep->r_flags & R_SOFTTERM)
395 				error = EINTR;
396 			else
397 				rep->r_flags |= R_MUSTRESEND;
398 		} else
399 			log(LOG_INFO, "nfsd send error %d\n", error);
400 
401 		/*
402 		 * Handle any recoverable (soft) socket errors here. (???)
403 		 */
404 		if (error != EINTR && error != ERESTART &&
405 			error != EWOULDBLOCK && error != EPIPE)
406 			error = 0;
407 	}
408 	return (error);
409 }
410 
411 #ifdef NFS
412 /*
413  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
414  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
415  * Mark and consolidate the data into a new mbuf list.
416  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
417  *     small mbufs.
418  * For SOCK_STREAM we must be very careful to read an entire record once
419  * we have read any of it, even if the system call has been interrupted.
420  */
421 int
422 nfs_receive(rep, aname, mp)
423 	register struct nfsreq *rep;
424 	struct mbuf **aname;
425 	struct mbuf **mp;
426 {
427 	register struct socket *so;
428 	struct uio auio;
429 	struct iovec aio;
430 	register struct mbuf *m;
431 	struct mbuf *control;
432 	u_int32_t len;
433 	struct mbuf **getnam;
434 	int error, sotype, rcvflg;
435 	struct proc *p = curproc;	/* XXX */
436 
437 	/*
438 	 * Set up arguments for soreceive()
439 	 */
440 	*mp = (struct mbuf *)0;
441 	*aname = (struct mbuf *)0;
442 	sotype = rep->r_nmp->nm_sotype;
443 
444 	/*
445 	 * For reliable protocols, lock against other senders/receivers
446 	 * in case a reconnect is necessary.
447 	 * For SOCK_STREAM, first get the Record Mark to find out how much
448 	 * more there is to get.
449 	 * We must lock the socket against other receivers
450 	 * until we have an entire rpc request/reply.
451 	 */
452 	if (sotype != SOCK_DGRAM) {
453 		error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
454 		if (error)
455 			return (error);
456 tryagain:
457 		/*
458 		 * Check for fatal errors and resending request.
459 		 */
460 		/*
461 		 * Ugh: If a reconnect attempt just happened, nm_so
462 		 * would have changed. NULL indicates a failed
463 		 * attempt that has essentially shut down this
464 		 * mount point.
465 		 */
466 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
467 			nfs_sndunlock(&rep->r_nmp->nm_iflag);
468 			return (EINTR);
469 		}
470 		so = rep->r_nmp->nm_so;
471 		if (!so) {
472 			error = nfs_reconnect(rep);
473 			if (error) {
474 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
475 				return (error);
476 			}
477 			goto tryagain;
478 		}
479 		while (rep->r_flags & R_MUSTRESEND) {
480 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
481 			nfsstats.rpcretries++;
482 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
483 			if (error) {
484 				if (error == EINTR || error == ERESTART ||
485 				    (error = nfs_reconnect(rep)) != 0) {
486 					nfs_sndunlock(&rep->r_nmp->nm_iflag);
487 					return (error);
488 				}
489 				goto tryagain;
490 			}
491 		}
492 		nfs_sndunlock(&rep->r_nmp->nm_iflag);
493 		if (sotype == SOCK_STREAM) {
494 			aio.iov_base = (caddr_t) &len;
495 			aio.iov_len = sizeof(u_int32_t);
496 			auio.uio_iov = &aio;
497 			auio.uio_iovcnt = 1;
498 			auio.uio_segflg = UIO_SYSSPACE;
499 			auio.uio_rw = UIO_READ;
500 			auio.uio_offset = 0;
501 			auio.uio_resid = sizeof(u_int32_t);
502 			auio.uio_procp = p;
503 			do {
504 			   rcvflg = MSG_WAITALL;
505 			   error = soreceive(so, (struct mbuf **)0, &auio,
506 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
507 			   if (error == EWOULDBLOCK && rep) {
508 				if (rep->r_flags & R_SOFTTERM)
509 					return (EINTR);
510 			   }
511 			} while (error == EWOULDBLOCK);
512 			if (!error && auio.uio_resid > 0) {
513 			    log(LOG_INFO,
514 				 "short receive (%d/%d) from nfs server %s\n",
515 				 sizeof(u_int32_t) - auio.uio_resid,
516 				 sizeof(u_int32_t),
517 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
518 			    error = EPIPE;
519 			}
520 			if (error)
521 				goto errout;
522 			len = ntohl(len) & ~0x80000000;
523 			/*
524 			 * This is SERIOUS! We are out of sync with the sender
525 			 * and forcing a disconnect/reconnect is all I can do.
526 			 */
527 			if (len > NFS_MAXPACKET) {
528 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
529 				"impossible packet length",
530 				len,
531 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
532 			    error = EFBIG;
533 			    goto errout;
534 			}
535 			auio.uio_resid = len;
536 			do {
537 			    rcvflg = MSG_WAITALL;
538 			    error =  soreceive(so, (struct mbuf **)0,
539 				&auio, mp, (struct mbuf **)0, &rcvflg);
540 			} while (error == EWOULDBLOCK || error == EINTR ||
541 				 error == ERESTART);
542 			if (!error && auio.uio_resid > 0) {
543 			    log(LOG_INFO,
544 				"short receive (%d/%d) from nfs server %s\n",
545 				len - auio.uio_resid, len,
546 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
547 			    error = EPIPE;
548 			}
549 		} else {
550 			/*
551 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
552 			 * and soreceive() will return when it has either a
553 			 * control msg or a data msg.
554 			 * We have no use for control msg., but must grab them
555 			 * and then throw them away so we know what is going
556 			 * on.
557 			 */
558 			auio.uio_resid = len = 100000000; /* Anything Big */
559 			auio.uio_procp = p;
560 			do {
561 			    rcvflg = 0;
562 			    error =  soreceive(so, (struct mbuf **)0,
563 				&auio, mp, &control, &rcvflg);
564 			    if (control)
565 				m_freem(control);
566 			    if (error == EWOULDBLOCK && rep) {
567 				if (rep->r_flags & R_SOFTTERM)
568 					return (EINTR);
569 			    }
570 			} while (error == EWOULDBLOCK ||
571 				 (!error && *mp == NULL && control));
572 			if ((rcvflg & MSG_EOR) == 0)
573 				printf("Egad!!\n");
574 			if (!error && *mp == NULL)
575 				error = EPIPE;
576 			len -= auio.uio_resid;
577 		}
578 errout:
579 		if (error && error != EINTR && error != ERESTART) {
580 			m_freem(*mp);
581 			*mp = (struct mbuf *)0;
582 			if (error != EPIPE)
583 				log(LOG_INFO,
584 				    "receive error %d from nfs server %s\n",
585 				    error,
586 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
587 			error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
588 			if (!error)
589 				error = nfs_reconnect(rep);
590 			if (!error)
591 				goto tryagain;
592 			else
593 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
594 		}
595 	} else {
596 		if ((so = rep->r_nmp->nm_so) == NULL)
597 			return (EACCES);
598 		if (so->so_state & SS_ISCONNECTED)
599 			getnam = (struct mbuf **)0;
600 		else
601 			getnam = aname;
602 		auio.uio_resid = len = 1000000;
603 		auio.uio_procp = p;
604 		do {
605 			rcvflg = 0;
606 			error =  soreceive(so, getnam, &auio, mp,
607 				(struct mbuf **)0, &rcvflg);
608 			if (error == EWOULDBLOCK &&
609 			    (rep->r_flags & R_SOFTTERM))
610 				return (EINTR);
611 		} while (error == EWOULDBLOCK);
612 		len -= auio.uio_resid;
613 	}
614 	if (error) {
615 		m_freem(*mp);
616 		*mp = (struct mbuf *)0;
617 	}
618 	return (error);
619 }
620 
621 /*
622  * Implement receipt of reply on a socket.
623  * We must search through the list of received datagrams matching them
624  * with outstanding requests using the xid, until ours is found.
625  */
626 /* ARGSUSED */
627 int
628 nfs_reply(myrep)
629 	struct nfsreq *myrep;
630 {
631 	register struct nfsreq *rep;
632 	register struct nfsmount *nmp = myrep->r_nmp;
633 	register int32_t t1;
634 	struct mbuf *mrep, *nam, *md;
635 	u_int32_t rxid, *tl;
636 	caddr_t dpos, cp2;
637 	int error;
638 
639 	/*
640 	 * Loop around until we get our own reply
641 	 */
642 	for (;;) {
643 		/*
644 		 * Lock against other receivers so that I don't get stuck in
645 		 * sbwait() after someone else has received my reply for me.
646 		 * Also necessary for connection based protocols to avoid
647 		 * race conditions during a reconnect.
648 		 */
649 		error = nfs_rcvlock(myrep);
650 		if (error == EALREADY)
651 			return (0);
652 		if (error)
653 			return (error);
654 		/*
655 		 * Get the next Rpc reply off the socket
656 		 */
657 		error = nfs_receive(myrep, &nam, &mrep);
658 		nfs_rcvunlock(&nmp->nm_iflag);
659 		if (error) {
660 
661 			/*
662 			 * Ignore routing errors on connectionless protocols??
663 			 */
664 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
665 				nmp->nm_so->so_error = 0;
666 #ifdef DEBUG
667 				printf("nfs_reply: ignoring error %d\n", error);
668 #endif
669 				if (myrep->r_flags & R_GETONEREP)
670 					return (0);
671 				continue;
672 			}
673 			return (error);
674 		}
675 		if (nam)
676 			m_freem(nam);
677 
678 		/*
679 		 * Get the xid and check that it is an rpc reply
680 		 */
681 		md = mrep;
682 		dpos = mtod(md, caddr_t);
683 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
684 		rxid = *tl++;
685 		if (*tl != rpc_reply) {
686 			if (nmp->nm_flag & NFSMNT_NQNFS) {
687 				if (nqnfs_callback(nmp, mrep, md, dpos))
688 					nfsstats.rpcinvalid++;
689 			} else {
690 				nfsstats.rpcinvalid++;
691 				m_freem(mrep);
692 			}
693 nfsmout:
694 			if (myrep->r_flags & R_GETONEREP)
695 				return (0);
696 			continue;
697 		}
698 
699 		/*
700 		 * Loop through the request list to match up the reply
701 		 * Iff no match, just drop the datagram
702 		 */
703 		for (rep = nfs_reqq.tqh_first; rep != 0;
704 		    rep = rep->r_chain.tqe_next) {
705 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
706 				/* Found it.. */
707 				rep->r_mrep = mrep;
708 				rep->r_md = md;
709 				rep->r_dpos = dpos;
710 				if (nfsrtton) {
711 					struct rttl *rt;
712 
713 					rt = &nfsrtt.rttl[nfsrtt.pos];
714 					rt->proc = rep->r_procnum;
715 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
716 					rt->sent = nmp->nm_sent;
717 					rt->cwnd = nmp->nm_cwnd;
718 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
719 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
720 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
721 					rt->tstamp = time;
722 					if (rep->r_flags & R_TIMING)
723 						rt->rtt = rep->r_rtt;
724 					else
725 						rt->rtt = 1000000;
726 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
727 				}
728 				/*
729 				 * Update congestion window.
730 				 * Do the additive increase of
731 				 * one rpc/rtt.
732 				 */
733 				if (nmp->nm_cwnd <= nmp->nm_sent) {
734 					nmp->nm_cwnd +=
735 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
736 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
737 					if (nmp->nm_cwnd > NFS_MAXCWND)
738 						nmp->nm_cwnd = NFS_MAXCWND;
739 				}
740 				rep->r_flags &= ~R_SENT;
741 				nmp->nm_sent -= NFS_CWNDSCALE;
742 				/*
743 				 * Update rtt using a gain of 0.125 on the mean
744 				 * and a gain of 0.25 on the deviation.
745 				 */
746 				if (rep->r_flags & R_TIMING) {
747 					/*
748 					 * Since the timer resolution of
749 					 * NFS_HZ is so course, it can often
750 					 * result in r_rtt == 0. Since
751 					 * r_rtt == N means that the actual
752 					 * rtt is between N+dt and N+2-dt ticks,
753 					 * add 1.
754 					 */
755 					t1 = rep->r_rtt + 1;
756 					t1 -= (NFS_SRTT(rep) >> 3);
757 					NFS_SRTT(rep) += t1;
758 					if (t1 < 0)
759 						t1 = -t1;
760 					t1 -= (NFS_SDRTT(rep) >> 2);
761 					NFS_SDRTT(rep) += t1;
762 				}
763 				nmp->nm_timeouts = 0;
764 				break;
765 			}
766 		}
767 		/*
768 		 * If not matched to a request, drop it.
769 		 * If it's mine, get out.
770 		 */
771 		if (rep == 0) {
772 			nfsstats.rpcunexpected++;
773 			m_freem(mrep);
774 		} else if (rep == myrep) {
775 			if (rep->r_mrep == NULL)
776 				panic("nfsreply nil");
777 			return (0);
778 		}
779 		if (myrep->r_flags & R_GETONEREP)
780 			return (0);
781 	}
782 }
783 
784 /*
785  * nfs_request - goes something like this
786  *	- fill in request struct
787  *	- links it into list
788  *	- calls nfs_send() for first transmit
789  *	- calls nfs_receive() to get reply
790  *	- break down rpc header and return with nfs reply pointed to
791  *	  by mrep or error
792  * nb: always frees up mreq mbuf list
793  */
794 int
795 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
796 	struct vnode *vp;
797 	struct mbuf *mrest;
798 	int procnum;
799 	struct proc *procp;
800 	struct ucred *cred;
801 	struct mbuf **mrp;
802 	struct mbuf **mdp;
803 	caddr_t *dposp;
804 {
805 	register struct mbuf *m, *mrep;
806 	register struct nfsreq *rep;
807 	register u_int32_t *tl;
808 	register int i;
809 	struct nfsmount *nmp;
810 	struct mbuf *md, *mheadend;
811 	struct nfsnode *np;
812 	char nickv[RPCX_NICKVERF];
813 	time_t reqtime, waituntil;
814 	caddr_t dpos, cp2;
815 	int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type;
816 	int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0;
817 	int verf_len, verf_type;
818 	u_int32_t xid;
819 	u_quad_t frev;
820 	char *auth_str, *verf_str;
821 	NFSKERBKEY_T key;		/* save session key */
822 
823 	nmp = VFSTONFS(vp->v_mount);
824 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
825 	rep->r_nmp = nmp;
826 	rep->r_vp = vp;
827 	rep->r_procp = procp;
828 	rep->r_procnum = procnum;
829 	i = 0;
830 	m = mrest;
831 	while (m) {
832 		i += m->m_len;
833 		m = m->m_next;
834 	}
835 	mrest_len = i;
836 
837 	/*
838 	 * Get the RPC header with authorization.
839 	 */
840 kerbauth:
841 	verf_str = auth_str = (char *)0;
842 	if (nmp->nm_flag & NFSMNT_KERB) {
843 		verf_str = nickv;
844 		verf_len = sizeof (nickv);
845 		auth_type = RPCAUTH_KERB4;
846 		bzero((caddr_t)key, sizeof (key));
847 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
848 			&auth_len, verf_str, verf_len)) {
849 			error = nfs_getauth(nmp, rep, cred, &auth_str,
850 				&auth_len, verf_str, &verf_len, key);
851 			if (error) {
852 				free((caddr_t)rep, M_NFSREQ);
853 				m_freem(mrest);
854 				return (error);
855 			}
856 		}
857 	} else {
858 		auth_type = RPCAUTH_UNIX;
859 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
860 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
861 			5 * NFSX_UNSIGNED;
862 	}
863 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
864 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
865 	if (auth_str)
866 		free(auth_str, M_TEMP);
867 
868 	/*
869 	 * For stream protocols, insert a Sun RPC Record Mark.
870 	 */
871 	if (nmp->nm_sotype == SOCK_STREAM) {
872 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
873 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
874 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
875 	}
876 	rep->r_mreq = m;
877 	rep->r_xid = xid;
878 tryagain:
879 	if (nmp->nm_flag & NFSMNT_SOFT)
880 		rep->r_retry = nmp->nm_retry;
881 	else
882 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
883 	rep->r_rtt = rep->r_rexmit = 0;
884 	if (proct[procnum] > 0)
885 		rep->r_flags = R_TIMING;
886 	else
887 		rep->r_flags = 0;
888 	rep->r_mrep = NULL;
889 
890 	/*
891 	 * Do the client side RPC.
892 	 */
893 	nfsstats.rpcrequests++;
894 	/*
895 	 * Chain request into list of outstanding requests. Be sure
896 	 * to put it LAST so timer finds oldest requests first.
897 	 */
898 	s = splsoftnet();
899 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
900 
901 	/* Get send time for nqnfs */
902 	reqtime = time.tv_sec;
903 
904 	/*
905 	 * If backing off another request or avoiding congestion, don't
906 	 * send this one now but let timer do it. If not timing a request,
907 	 * do it now.
908 	 */
909 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
910 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
911 		nmp->nm_sent < nmp->nm_cwnd)) {
912 		splx(s);
913 		if (nmp->nm_soflags & PR_CONNREQUIRED)
914 			error = nfs_sndlock(&nmp->nm_iflag, rep);
915 		if (!error) {
916 			m = m_copym(m, 0, M_COPYALL, M_WAIT);
917 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep);
918 			if (nmp->nm_soflags & PR_CONNREQUIRED)
919 				nfs_sndunlock(&nmp->nm_iflag);
920 		}
921 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
922 			nmp->nm_sent += NFS_CWNDSCALE;
923 			rep->r_flags |= R_SENT;
924 		}
925 	} else {
926 		splx(s);
927 		rep->r_rtt = -1;
928 	}
929 
930 	/*
931 	 * Wait for the reply from our send or the timer's.
932 	 */
933 	if (!error || error == EPIPE)
934 		error = nfs_reply(rep);
935 
936 	/*
937 	 * RPC done, unlink the request.
938 	 */
939 	s = splsoftnet();
940 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
941 	splx(s);
942 
943 	/*
944 	 * Decrement the outstanding request count.
945 	 */
946 	if (rep->r_flags & R_SENT) {
947 		rep->r_flags &= ~R_SENT;	/* paranoia */
948 		nmp->nm_sent -= NFS_CWNDSCALE;
949 	}
950 
951 	/*
952 	 * If there was a successful reply and a tprintf msg.
953 	 * tprintf a response.
954 	 */
955 	if (!error && (rep->r_flags & R_TPRINTFMSG))
956 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
957 		    "is alive again");
958 	mrep = rep->r_mrep;
959 	md = rep->r_md;
960 	dpos = rep->r_dpos;
961 	if (error) {
962 		m_freem(rep->r_mreq);
963 		free((caddr_t)rep, M_NFSREQ);
964 		return (error);
965 	}
966 
967 	/*
968 	 * break down the rpc header and check if ok
969 	 */
970 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
971 	if (*tl++ == rpc_msgdenied) {
972 		if (*tl == rpc_mismatch)
973 			error = EOPNOTSUPP;
974 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
975 			if (!failed_auth) {
976 				failed_auth++;
977 				mheadend->m_next = (struct mbuf *)0;
978 				m_freem(mrep);
979 				m_freem(rep->r_mreq);
980 				goto kerbauth;
981 			} else
982 				error = EAUTH;
983 		} else
984 			error = EACCES;
985 		m_freem(mrep);
986 		m_freem(rep->r_mreq);
987 		free((caddr_t)rep, M_NFSREQ);
988 		return (error);
989 	}
990 
991 	/*
992 	 * Grab any Kerberos verifier, otherwise just throw it away.
993 	 */
994 	verf_type = fxdr_unsigned(int, *tl++);
995 	i = fxdr_unsigned(int32_t, *tl);
996 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
997 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
998 		if (error)
999 			goto nfsmout;
1000 	} else if (i > 0)
1001 		nfsm_adv(nfsm_rndup(i));
1002 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1003 	/* 0 == ok */
1004 	if (*tl == 0) {
1005 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1006 		if (*tl != 0) {
1007 			error = fxdr_unsigned(int, *tl);
1008 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1009 				error == NFSERR_TRYLATER) {
1010 				m_freem(mrep);
1011 				error = 0;
1012 				waituntil = time.tv_sec + trylater_delay;
1013 				while (time.tv_sec < waituntil)
1014 					(void) tsleep((caddr_t)&lbolt,
1015 						PSOCK, "nqnfstry", 0);
1016 				trylater_delay *= nfs_backoff[trylater_cnt];
1017 				if (trylater_cnt < 7)
1018 					trylater_cnt++;
1019 				goto tryagain;
1020 			}
1021 
1022 			/*
1023 			 * If the File Handle was stale, invalidate the
1024 			 * lookup cache, just in case.
1025 			 */
1026 			if (error == ESTALE)
1027 				cache_purge(vp);
1028 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1029 				*mrp = mrep;
1030 				*mdp = md;
1031 				*dposp = dpos;
1032 				error |= NFSERR_RETERR;
1033 			} else
1034 				m_freem(mrep);
1035 			m_freem(rep->r_mreq);
1036 			free((caddr_t)rep, M_NFSREQ);
1037 			return (error);
1038 		}
1039 
1040 		/*
1041 		 * For nqnfs, get any lease in reply
1042 		 */
1043 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1044 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1045 			if (*tl) {
1046 				np = VTONFS(vp);
1047 				nqlflag = fxdr_unsigned(int, *tl);
1048 				nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1049 				cachable = fxdr_unsigned(int, *tl++);
1050 				reqtime += fxdr_unsigned(int, *tl++);
1051 				if (reqtime > time.tv_sec) {
1052 				    fxdr_hyper(tl, &frev);
1053 				    nqnfs_clientlease(nmp, np, nqlflag,
1054 					cachable, reqtime, frev);
1055 				}
1056 			}
1057 		}
1058 		*mrp = mrep;
1059 		*mdp = md;
1060 		*dposp = dpos;
1061 		m_freem(rep->r_mreq);
1062 		FREE((caddr_t)rep, M_NFSREQ);
1063 		return (0);
1064 	}
1065 	m_freem(mrep);
1066 	error = EPROTONOSUPPORT;
1067 nfsmout:
1068 	m_freem(rep->r_mreq);
1069 	free((caddr_t)rep, M_NFSREQ);
1070 	return (error);
1071 }
1072 #endif /* NFS */
1073 
1074 /*
1075  * Generate the rpc reply header
1076  * siz arg. is used to decide if adding a cluster is worthwhile
1077  */
1078 int
1079 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1080 	int siz;
1081 	struct nfsrv_descript *nd;
1082 	struct nfssvc_sock *slp;
1083 	int err;
1084 	int cache;
1085 	u_quad_t *frev;
1086 	struct mbuf **mrq;
1087 	struct mbuf **mbp;
1088 	caddr_t *bposp;
1089 {
1090 	register u_int32_t *tl;
1091 	register struct mbuf *mreq;
1092 	caddr_t bpos;
1093 	struct mbuf *mb, *mb2;
1094 
1095 	MGETHDR(mreq, M_WAIT, MT_DATA);
1096 	mb = mreq;
1097 	/*
1098 	 * If this is a big reply, use a cluster else
1099 	 * try and leave leading space for the lower level headers.
1100 	 */
1101 	siz += RPC_REPLYSIZ;
1102 	if (siz >= MINCLSIZE) {
1103 		MCLGET(mreq, M_WAIT);
1104 	} else
1105 		mreq->m_data += max_hdr;
1106 	tl = mtod(mreq, u_int32_t *);
1107 	mreq->m_len = 6 * NFSX_UNSIGNED;
1108 	bpos = ((caddr_t)tl) + mreq->m_len;
1109 	*tl++ = txdr_unsigned(nd->nd_retxid);
1110 	*tl++ = rpc_reply;
1111 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1112 		*tl++ = rpc_msgdenied;
1113 		if (err & NFSERR_AUTHERR) {
1114 			*tl++ = rpc_autherr;
1115 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1116 			mreq->m_len -= NFSX_UNSIGNED;
1117 			bpos -= NFSX_UNSIGNED;
1118 		} else {
1119 			*tl++ = rpc_mismatch;
1120 			*tl++ = txdr_unsigned(RPC_VER2);
1121 			*tl = txdr_unsigned(RPC_VER2);
1122 		}
1123 	} else {
1124 		*tl++ = rpc_msgaccepted;
1125 
1126 		/*
1127 		 * For Kerberos authentication, we must send the nickname
1128 		 * verifier back, otherwise just RPCAUTH_NULL.
1129 		 */
1130 		if (nd->nd_flag & ND_KERBFULL) {
1131 		    register struct nfsuid *nuidp;
1132 		    struct timeval ktvin, ktvout;
1133 
1134 		    for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1135 			nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1136 			if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1137 			    (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1138 			     &nuidp->nu_haddr, nd->nd_nam2)))
1139 			    break;
1140 		    }
1141 		    if (nuidp) {
1142 			ktvin.tv_sec =
1143 			    txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1144 			ktvin.tv_usec =
1145 			    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1146 
1147 			/*
1148 			 * Encrypt the timestamp in ecb mode using the
1149 			 * session key.
1150 			 */
1151 #ifdef NFSKERB
1152 			XXX
1153 #endif
1154 
1155 			*tl++ = rpc_auth_kerb;
1156 			*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1157 			*tl = ktvout.tv_sec;
1158 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1159 			*tl++ = ktvout.tv_usec;
1160 			*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1161 		    } else {
1162 			*tl++ = 0;
1163 			*tl++ = 0;
1164 		    }
1165 		} else {
1166 			*tl++ = 0;
1167 			*tl++ = 0;
1168 		}
1169 		switch (err) {
1170 		case EPROGUNAVAIL:
1171 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1172 			break;
1173 		case EPROGMISMATCH:
1174 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1175 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1176 			if (nd->nd_flag & ND_NQNFS) {
1177 				*tl++ = txdr_unsigned(3);
1178 				*tl = txdr_unsigned(3);
1179 			} else {
1180 				*tl++ = txdr_unsigned(2);
1181 				*tl = txdr_unsigned(3);
1182 			}
1183 			break;
1184 		case EPROCUNAVAIL:
1185 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1186 			break;
1187 		case EBADRPC:
1188 			*tl = txdr_unsigned(RPC_GARBAGE);
1189 			break;
1190 		default:
1191 			*tl = 0;
1192 			if (err != NFSERR_RETVOID) {
1193 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1194 				if (err)
1195 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1196 				else
1197 				    *tl = 0;
1198 			}
1199 			break;
1200 		};
1201 	}
1202 
1203 	/*
1204 	 * For nqnfs, piggyback lease as requested.
1205 	 */
1206 	if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1207 		if (nd->nd_flag & ND_LEASE) {
1208 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1209 			*tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1210 			*tl++ = txdr_unsigned(cache);
1211 			*tl++ = txdr_unsigned(nd->nd_duration);
1212 			txdr_hyper(frev, tl);
1213 		} else {
1214 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1215 			*tl = 0;
1216 		}
1217 	}
1218 	if (mrq != NULL)
1219 		*mrq = mreq;
1220 	*mbp = mb;
1221 	*bposp = bpos;
1222 	if (err != 0 && err != NFSERR_RETVOID)
1223 		nfsstats.srvrpc_errs++;
1224 	return (0);
1225 }
1226 
1227 /*
1228  * Nfs timer routine
1229  * Scan the nfsreq list and retranmit any requests that have timed out
1230  * To avoid retransmission attempts on STREAM sockets (in the future) make
1231  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1232  */
1233 void
1234 nfs_timer(arg)
1235 	void *arg;	/* never used */
1236 {
1237 	register struct nfsreq *rep;
1238 	register struct mbuf *m;
1239 	register struct socket *so;
1240 	register struct nfsmount *nmp;
1241 	register int timeo;
1242 	int s, error;
1243 #ifdef NFSSERVER
1244 	register struct nfssvc_sock *slp;
1245 	static long lasttime = 0;
1246 	u_quad_t cur_usec;
1247 #endif
1248 
1249 	s = splsoftnet();
1250 	for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) {
1251 		nmp = rep->r_nmp;
1252 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1253 			continue;
1254 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1255 			rep->r_flags |= R_SOFTTERM;
1256 			continue;
1257 		}
1258 		if (rep->r_rtt >= 0) {
1259 			rep->r_rtt++;
1260 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1261 				timeo = nmp->nm_timeo;
1262 			else
1263 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1264 			if (nmp->nm_timeouts > 0)
1265 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1266 			if (rep->r_rtt <= timeo)
1267 				continue;
1268 			if (nmp->nm_timeouts < 8)
1269 				nmp->nm_timeouts++;
1270 		}
1271 		/*
1272 		 * Check for server not responding
1273 		 */
1274 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1275 		     rep->r_rexmit > nmp->nm_deadthresh) {
1276 			nfs_msg(rep->r_procp,
1277 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1278 			    "not responding");
1279 			rep->r_flags |= R_TPRINTFMSG;
1280 		}
1281 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1282 			nfsstats.rpctimeouts++;
1283 			rep->r_flags |= R_SOFTTERM;
1284 			continue;
1285 		}
1286 		if (nmp->nm_sotype != SOCK_DGRAM) {
1287 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1288 				rep->r_rexmit = NFS_MAXREXMIT;
1289 			continue;
1290 		}
1291 		if ((so = nmp->nm_so) == NULL)
1292 			continue;
1293 
1294 		/*
1295 		 * If there is enough space and the window allows..
1296 		 *	Resend it
1297 		 * Set r_rtt to -1 in case we fail to send it now.
1298 		 */
1299 		rep->r_rtt = -1;
1300 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1301 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1302 		    (rep->r_flags & R_SENT) ||
1303 		    nmp->nm_sent < nmp->nm_cwnd) &&
1304 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1305 		        if (so->so_state & SS_ISCONNECTED)
1306 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1307 			    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1308 			else
1309 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1310 			    nmp->nm_nam, (struct mbuf *)0, (struct proc *)0);
1311 			if (error) {
1312 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1313 #ifdef DEBUG
1314 					printf("nfs_timer: ignoring error %d\n",
1315 						error);
1316 #endif
1317 					so->so_error = 0;
1318 				}
1319 			} else {
1320 				/*
1321 				 * Iff first send, start timing
1322 				 * else turn timing off, backoff timer
1323 				 * and divide congestion window by 2.
1324 				 */
1325 				if (rep->r_flags & R_SENT) {
1326 					rep->r_flags &= ~R_TIMING;
1327 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1328 						rep->r_rexmit = NFS_MAXREXMIT;
1329 					nmp->nm_cwnd >>= 1;
1330 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1331 						nmp->nm_cwnd = NFS_CWNDSCALE;
1332 					nfsstats.rpcretries++;
1333 				} else {
1334 					rep->r_flags |= R_SENT;
1335 					nmp->nm_sent += NFS_CWNDSCALE;
1336 				}
1337 				rep->r_rtt = 0;
1338 			}
1339 		}
1340 	}
1341 
1342 #ifdef NFSSERVER
1343 	/*
1344 	 * Call the nqnfs server timer once a second to handle leases.
1345 	 */
1346 	if (lasttime != time.tv_sec) {
1347 		lasttime = time.tv_sec;
1348 		nqnfs_serverd();
1349 	}
1350 
1351 	/*
1352 	 * Scan the write gathering queues for writes that need to be
1353 	 * completed now.
1354 	 */
1355 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1356 	for (slp = nfssvc_sockhead.tqh_first; slp != 0;
1357 	    slp = slp->ns_chain.tqe_next) {
1358 	    if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1359 		nfsrv_wakenfsd(slp);
1360 	}
1361 #endif /* NFSSERVER */
1362 	splx(s);
1363 	timeout(nfs_timer, (void *)0, nfs_ticks);
1364 }
1365 
1366 /*
1367  * Test for a termination condition pending on the process.
1368  * This is used for NFSMNT_INT mounts.
1369  */
1370 int
1371 nfs_sigintr(nmp, rep, p)
1372 	struct nfsmount *nmp;
1373 	struct nfsreq *rep;
1374 	register struct proc *p;
1375 {
1376 
1377 	if (rep && (rep->r_flags & R_SOFTTERM))
1378 		return (EINTR);
1379 	if (!(nmp->nm_flag & NFSMNT_INT))
1380 		return (0);
1381 	if (p && p->p_siglist &&
1382 	    (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) &
1383 	    NFSINT_SIGMASK))
1384 		return (EINTR);
1385 	return (0);
1386 }
1387 
1388 /*
1389  * Lock a socket against others.
1390  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1391  * and also to avoid race conditions between the processes with nfs requests
1392  * in progress when a reconnect is necessary.
1393  */
1394 int
1395 nfs_sndlock(flagp, rep)
1396 	register int *flagp;
1397 	struct nfsreq *rep;
1398 {
1399 	struct proc *p;
1400 	int slpflag = 0, slptimeo = 0;
1401 
1402 	if (rep) {
1403 		p = rep->r_procp;
1404 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1405 			slpflag = PCATCH;
1406 	} else
1407 		p = (struct proc *)0;
1408 	while (*flagp & NFSMNT_SNDLOCK) {
1409 		if (nfs_sigintr(rep->r_nmp, rep, p))
1410 			return (EINTR);
1411 		*flagp |= NFSMNT_WANTSND;
1412 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1413 			slptimeo);
1414 		if (slpflag == PCATCH) {
1415 			slpflag = 0;
1416 			slptimeo = 2 * hz;
1417 		}
1418 	}
1419 	*flagp |= NFSMNT_SNDLOCK;
1420 	return (0);
1421 }
1422 
1423 /*
1424  * Unlock the stream socket for others.
1425  */
1426 void
1427 nfs_sndunlock(flagp)
1428 	register int *flagp;
1429 {
1430 
1431 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1432 		panic("nfs sndunlock");
1433 	*flagp &= ~NFSMNT_SNDLOCK;
1434 	if (*flagp & NFSMNT_WANTSND) {
1435 		*flagp &= ~NFSMNT_WANTSND;
1436 		wakeup((caddr_t)flagp);
1437 	}
1438 }
1439 
1440 int
1441 nfs_rcvlock(rep)
1442 	register struct nfsreq *rep;
1443 {
1444 	register int *flagp = &rep->r_nmp->nm_iflag;
1445 	int slpflag, slptimeo = 0;
1446 
1447 	if (*flagp & NFSMNT_INT)
1448 		slpflag = PCATCH;
1449 	else
1450 		slpflag = 0;
1451 	while (*flagp & NFSMNT_RCVLOCK) {
1452 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1453 			return (EINTR);
1454 		*flagp |= NFSMNT_WANTRCV;
1455 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1456 			slptimeo);
1457 		/* If our reply was received while we were sleeping,
1458 		 * then just return without taking the lock to avoid a
1459 		 * situation where a single iod could 'capture' the
1460 		 * receive lock.
1461 		 */
1462 		if (rep->r_mrep != NULL)
1463 			return (EALREADY);
1464 		if (slpflag == PCATCH) {
1465 			slpflag = 0;
1466 			slptimeo = 2 * hz;
1467 		}
1468 	}
1469 	*flagp |= NFSMNT_RCVLOCK;
1470 	return (0);
1471 }
1472 
1473 /*
1474  * Unlock the stream socket for others.
1475  */
1476 void
1477 nfs_rcvunlock(flagp)
1478 	register int *flagp;
1479 {
1480 
1481 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1482 		panic("nfs rcvunlock");
1483 	*flagp &= ~NFSMNT_RCVLOCK;
1484 	if (*flagp & NFSMNT_WANTRCV) {
1485 		*flagp &= ~NFSMNT_WANTRCV;
1486 		wakeup((caddr_t)flagp);
1487 	}
1488 }
1489 
1490 /*
1491  * Parse an RPC request
1492  * - verify it
1493  * - fill in the cred struct.
1494  */
1495 int
1496 nfs_getreq(nd, nfsd, has_header)
1497 	register struct nfsrv_descript *nd;
1498 	struct nfsd *nfsd;
1499 	int has_header;
1500 {
1501 	register int len, i;
1502 	register u_int32_t *tl;
1503 	register int32_t t1;
1504 	struct uio uio;
1505 	struct iovec iov;
1506 	caddr_t dpos, cp2, cp;
1507 	u_int32_t nfsvers, auth_type;
1508 	uid_t nickuid;
1509 	int error = 0, nqnfs = 0, ticklen;
1510 	struct mbuf *mrep, *md;
1511 	register struct nfsuid *nuidp;
1512 	struct timeval tvin, tvout;
1513 
1514 	mrep = nd->nd_mrep;
1515 	md = nd->nd_md;
1516 	dpos = nd->nd_dpos;
1517 	if (has_header) {
1518 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1519 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1520 		if (*tl++ != rpc_call) {
1521 			m_freem(mrep);
1522 			return (EBADRPC);
1523 		}
1524 	} else
1525 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1526 	nd->nd_repstat = 0;
1527 	nd->nd_flag = 0;
1528 	if (*tl++ != rpc_vers) {
1529 		nd->nd_repstat = ERPCMISMATCH;
1530 		nd->nd_procnum = NFSPROC_NOOP;
1531 		return (0);
1532 	}
1533 	if (*tl != nfs_prog) {
1534 		if (*tl == nqnfs_prog)
1535 			nqnfs++;
1536 		else {
1537 			nd->nd_repstat = EPROGUNAVAIL;
1538 			nd->nd_procnum = NFSPROC_NOOP;
1539 			return (0);
1540 		}
1541 	}
1542 	tl++;
1543 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1544 	if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1545 		(nfsvers != NQNFS_VER3 && nqnfs)) {
1546 		nd->nd_repstat = EPROGMISMATCH;
1547 		nd->nd_procnum = NFSPROC_NOOP;
1548 		return (0);
1549 	}
1550 	if (nqnfs)
1551 		nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1552 	else if (nfsvers == NFS_VER3)
1553 		nd->nd_flag = ND_NFSV3;
1554 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1555 	if (nd->nd_procnum == NFSPROC_NULL)
1556 		return (0);
1557 	if (nd->nd_procnum >= NFS_NPROCS ||
1558 		(!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1559 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1560 		nd->nd_repstat = EPROCUNAVAIL;
1561 		nd->nd_procnum = NFSPROC_NOOP;
1562 		return (0);
1563 	}
1564 	if ((nd->nd_flag & ND_NFSV3) == 0)
1565 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1566 	auth_type = *tl++;
1567 	len = fxdr_unsigned(int, *tl++);
1568 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1569 		m_freem(mrep);
1570 		return (EBADRPC);
1571 	}
1572 
1573 	nd->nd_flag &= ~ND_KERBAUTH;
1574 	/*
1575 	 * Handle auth_unix or auth_kerb.
1576 	 */
1577 	if (auth_type == rpc_auth_unix) {
1578 		len = fxdr_unsigned(int, *++tl);
1579 		if (len < 0 || len > NFS_MAXNAMLEN) {
1580 			m_freem(mrep);
1581 			return (EBADRPC);
1582 		}
1583 		nfsm_adv(nfsm_rndup(len));
1584 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1585 		bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred));
1586 		nd->nd_cr.cr_ref = 1;
1587 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1588 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1589 		len = fxdr_unsigned(int, *tl);
1590 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1591 			m_freem(mrep);
1592 			return (EBADRPC);
1593 		}
1594 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1595 		for (i = 0; i < len; i++)
1596 		    if (i < NGROUPS)
1597 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1598 		    else
1599 			tl++;
1600 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1601 		if (nd->nd_cr.cr_ngroups > 1)
1602 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1603 		len = fxdr_unsigned(int, *++tl);
1604 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1605 			m_freem(mrep);
1606 			return (EBADRPC);
1607 		}
1608 		if (len > 0)
1609 			nfsm_adv(nfsm_rndup(len));
1610 	} else if (auth_type == rpc_auth_kerb) {
1611 		switch (fxdr_unsigned(int, *tl++)) {
1612 		case RPCAKN_FULLNAME:
1613 			ticklen = fxdr_unsigned(int, *tl);
1614 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1615 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1616 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1617 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1618 				m_freem(mrep);
1619 				return (EBADRPC);
1620 			}
1621 			uio.uio_offset = 0;
1622 			uio.uio_iov = &iov;
1623 			uio.uio_iovcnt = 1;
1624 			uio.uio_segflg = UIO_SYSSPACE;
1625 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1626 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1627 			nfsm_mtouio(&uio, uio.uio_resid);
1628 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1629 			if (*tl++ != rpc_auth_kerb ||
1630 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1631 				printf("Bad kerb verifier\n");
1632 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1633 				nd->nd_procnum = NFSPROC_NOOP;
1634 				return (0);
1635 			}
1636 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1637 			tl = (u_int32_t *)cp;
1638 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1639 				printf("Not fullname kerb verifier\n");
1640 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1641 				nd->nd_procnum = NFSPROC_NOOP;
1642 				return (0);
1643 			}
1644 			cp += NFSX_UNSIGNED;
1645 			bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED);
1646 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1647 			nd->nd_flag |= ND_KERBFULL;
1648 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1649 			break;
1650 		case RPCAKN_NICKNAME:
1651 			if (len != 2 * NFSX_UNSIGNED) {
1652 				printf("Kerb nickname short\n");
1653 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1654 				nd->nd_procnum = NFSPROC_NOOP;
1655 				return (0);
1656 			}
1657 			nickuid = fxdr_unsigned(uid_t, *tl);
1658 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1659 			if (*tl++ != rpc_auth_kerb ||
1660 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1661 				printf("Kerb nick verifier bad\n");
1662 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1663 				nd->nd_procnum = NFSPROC_NOOP;
1664 				return (0);
1665 			}
1666 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1667 			tvin.tv_sec = *tl++;
1668 			tvin.tv_usec = *tl;
1669 
1670 			for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1671 			    nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1672 				if (nuidp->nu_cr.cr_uid == nickuid &&
1673 				    (!nd->nd_nam2 ||
1674 				     netaddr_match(NU_NETFAM(nuidp),
1675 				      &nuidp->nu_haddr, nd->nd_nam2)))
1676 					break;
1677 			}
1678 			if (!nuidp) {
1679 				nd->nd_repstat =
1680 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1681 				nd->nd_procnum = NFSPROC_NOOP;
1682 				return (0);
1683 			}
1684 
1685 			/*
1686 			 * Now, decrypt the timestamp using the session key
1687 			 * and validate it.
1688 			 */
1689 #ifdef NFSKERB
1690 			XXX
1691 #endif
1692 
1693 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1694 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1695 			if (nuidp->nu_expire < time.tv_sec ||
1696 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1697 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1698 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1699 				nuidp->nu_expire = 0;
1700 				nd->nd_repstat =
1701 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1702 				nd->nd_procnum = NFSPROC_NOOP;
1703 				return (0);
1704 			}
1705 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1706 			nd->nd_flag |= ND_KERBNICK;
1707 		};
1708 	} else {
1709 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1710 		nd->nd_procnum = NFSPROC_NOOP;
1711 		return (0);
1712 	}
1713 
1714 	/*
1715 	 * For nqnfs, get piggybacked lease request.
1716 	 */
1717 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
1718 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1719 		nd->nd_flag |= fxdr_unsigned(int, *tl);
1720 		if (nd->nd_flag & ND_LEASE) {
1721 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1722 			nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
1723 		} else
1724 			nd->nd_duration = NQ_MINLEASE;
1725 	} else
1726 		nd->nd_duration = NQ_MINLEASE;
1727 	nd->nd_md = md;
1728 	nd->nd_dpos = dpos;
1729 	return (0);
1730 nfsmout:
1731 	return (error);
1732 }
1733 
1734 int
1735 nfs_msg(p, server, msg)
1736 	struct proc *p;
1737 	char *server, *msg;
1738 {
1739 	tpr_t tpr;
1740 
1741 	if (p)
1742 		tpr = tprintf_open(p);
1743 	else
1744 		tpr = NULL;
1745 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1746 	tprintf_close(tpr);
1747 	return (0);
1748 }
1749 
1750 #ifdef NFSSERVER
1751 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
1752 				    struct nfssvc_sock *, struct proc *,
1753 				    struct mbuf **)) = {
1754 	nfsrv_null,
1755 	nfsrv_getattr,
1756 	nfsrv_setattr,
1757 	nfsrv_lookup,
1758 	nfsrv3_access,
1759 	nfsrv_readlink,
1760 	nfsrv_read,
1761 	nfsrv_write,
1762 	nfsrv_create,
1763 	nfsrv_mkdir,
1764 	nfsrv_symlink,
1765 	nfsrv_mknod,
1766 	nfsrv_remove,
1767 	nfsrv_rmdir,
1768 	nfsrv_rename,
1769 	nfsrv_link,
1770 	nfsrv_readdir,
1771 	nfsrv_readdirplus,
1772 	nfsrv_statfs,
1773 	nfsrv_fsinfo,
1774 	nfsrv_pathconf,
1775 	nfsrv_commit,
1776 	nqnfsrv_getlease,
1777 	nqnfsrv_vacated,
1778 	nfsrv_noop,
1779 	nfsrv_noop
1780 };
1781 
1782 /*
1783  * Socket upcall routine for the nfsd sockets.
1784  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1785  * Essentially do as much as possible non-blocking, else punt and it will
1786  * be called with M_WAIT from an nfsd.
1787  */
1788 void
1789 nfsrv_rcv(so, arg, waitflag)
1790 	struct socket *so;
1791 	caddr_t arg;
1792 	int waitflag;
1793 {
1794 	register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1795 	register struct mbuf *m;
1796 	struct mbuf *mp, *nam;
1797 	struct uio auio;
1798 	int flags, error;
1799 
1800 	if ((slp->ns_flag & SLP_VALID) == 0)
1801 		return;
1802 #ifdef notdef
1803 	/*
1804 	 * Define this to test for nfsds handling this under heavy load.
1805 	 */
1806 	if (waitflag == M_DONTWAIT) {
1807 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1808 	}
1809 #endif
1810 	auio.uio_procp = NULL;
1811 	if (so->so_type == SOCK_STREAM) {
1812 		/*
1813 		 * If there are already records on the queue, defer soreceive()
1814 		 * to an nfsd so that there is feedback to the TCP layer that
1815 		 * the nfs servers are heavily loaded.
1816 		 */
1817 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1818 			slp->ns_flag |= SLP_NEEDQ;
1819 			goto dorecs;
1820 		}
1821 
1822 		/*
1823 		 * Do soreceive().
1824 		 */
1825 		auio.uio_resid = 1000000000;
1826 		flags = MSG_DONTWAIT;
1827 		error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1828 		if (error || mp == (struct mbuf *)0) {
1829 			if (error == EWOULDBLOCK)
1830 				slp->ns_flag |= SLP_NEEDQ;
1831 			else
1832 				slp->ns_flag |= SLP_DISCONN;
1833 			goto dorecs;
1834 		}
1835 		m = mp;
1836 		if (slp->ns_rawend) {
1837 			slp->ns_rawend->m_next = m;
1838 			slp->ns_cc += 1000000000 - auio.uio_resid;
1839 		} else {
1840 			slp->ns_raw = m;
1841 			slp->ns_cc = 1000000000 - auio.uio_resid;
1842 		}
1843 		while (m->m_next)
1844 			m = m->m_next;
1845 		slp->ns_rawend = m;
1846 
1847 		/*
1848 		 * Now try and parse record(s) out of the raw stream data.
1849 		 */
1850 		error = nfsrv_getstream(slp, waitflag);
1851 		if (error) {
1852 			if (error == EPERM)
1853 				slp->ns_flag |= SLP_DISCONN;
1854 			else
1855 				slp->ns_flag |= SLP_NEEDQ;
1856 		}
1857 	} else {
1858 		do {
1859 			auio.uio_resid = 1000000000;
1860 			flags = MSG_DONTWAIT;
1861 			error = soreceive(so, &nam, &auio, &mp,
1862 						(struct mbuf **)0, &flags);
1863 			if (mp) {
1864 				if (nam) {
1865 					m = nam;
1866 					m->m_next = mp;
1867 				} else
1868 					m = mp;
1869 				if (slp->ns_recend)
1870 					slp->ns_recend->m_nextpkt = m;
1871 				else
1872 					slp->ns_rec = m;
1873 				slp->ns_recend = m;
1874 				m->m_nextpkt = (struct mbuf *)0;
1875 			}
1876 			if (error) {
1877 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
1878 					&& error != EWOULDBLOCK) {
1879 					slp->ns_flag |= SLP_DISCONN;
1880 					goto dorecs;
1881 				}
1882 			}
1883 		} while (mp);
1884 	}
1885 
1886 	/*
1887 	 * Now try and process the request records, non-blocking.
1888 	 */
1889 dorecs:
1890 	if (waitflag == M_DONTWAIT &&
1891 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
1892 		nfsrv_wakenfsd(slp);
1893 }
1894 
1895 /*
1896  * Try and extract an RPC request from the mbuf data list received on a
1897  * stream socket. The "waitflag" argument indicates whether or not it
1898  * can sleep.
1899  */
1900 int
1901 nfsrv_getstream(slp, waitflag)
1902 	register struct nfssvc_sock *slp;
1903 	int waitflag;
1904 {
1905 	register struct mbuf *m, **mpp;
1906 	register char *cp1, *cp2;
1907 	register int len;
1908 	struct mbuf *om, *m2, *recm = NULL;
1909 	u_int32_t recmark;
1910 
1911 	if (slp->ns_flag & SLP_GETSTREAM)
1912 		panic("nfs getstream");
1913 	slp->ns_flag |= SLP_GETSTREAM;
1914 	for (;;) {
1915 	    if (slp->ns_reclen == 0) {
1916 		if (slp->ns_cc < NFSX_UNSIGNED) {
1917 			slp->ns_flag &= ~SLP_GETSTREAM;
1918 			return (0);
1919 		}
1920 		m = slp->ns_raw;
1921 		if (m->m_len >= NFSX_UNSIGNED) {
1922 			bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
1923 			m->m_data += NFSX_UNSIGNED;
1924 			m->m_len -= NFSX_UNSIGNED;
1925 		} else {
1926 			cp1 = (caddr_t)&recmark;
1927 			cp2 = mtod(m, caddr_t);
1928 			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
1929 				while (m->m_len == 0) {
1930 					m = m->m_next;
1931 					cp2 = mtod(m, caddr_t);
1932 				}
1933 				*cp1++ = *cp2++;
1934 				m->m_data++;
1935 				m->m_len--;
1936 			}
1937 		}
1938 		slp->ns_cc -= NFSX_UNSIGNED;
1939 		recmark = ntohl(recmark);
1940 		slp->ns_reclen = recmark & ~0x80000000;
1941 		if (recmark & 0x80000000)
1942 			slp->ns_flag |= SLP_LASTFRAG;
1943 		else
1944 			slp->ns_flag &= ~SLP_LASTFRAG;
1945 		if (slp->ns_reclen > NFS_MAXPACKET) {
1946 			slp->ns_flag &= ~SLP_GETSTREAM;
1947 			return (EPERM);
1948 		}
1949 	    }
1950 
1951 	    /*
1952 	     * Now get the record part.
1953 	     */
1954 	    if (slp->ns_cc == slp->ns_reclen) {
1955 		recm = slp->ns_raw;
1956 		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
1957 		slp->ns_cc = slp->ns_reclen = 0;
1958 	    } else if (slp->ns_cc > slp->ns_reclen) {
1959 		len = 0;
1960 		m = slp->ns_raw;
1961 		om = (struct mbuf *)0;
1962 		while (len < slp->ns_reclen) {
1963 			if ((len + m->m_len) > slp->ns_reclen) {
1964 				size_t left = slp->ns_reclen - len;
1965 
1966 				MGETHDR(m2, waitflag, m->m_type);
1967 				if (m2 == NULL) {
1968 					slp->ns_flag &= ~SLP_GETSTREAM;
1969 					return (EWOULDBLOCK);
1970 				}
1971 				if (left > MHLEN) {
1972 					MCLGET(m2, waitflag);
1973 					if (!(m2->m_flags & M_EXT)) {
1974 						m_freem(m2);
1975 						slp->ns_flag &= ~SLP_GETSTREAM;
1976 						return (EWOULDBLOCK);
1977 					}
1978 				}
1979 				bcopy(mtod(m, caddr_t), mtod(m2, caddr_t),
1980 				    left);
1981 				m2->m_len = left;
1982 				m->m_data += left;
1983 				m->m_len -= left;
1984 				if (om) {
1985 					om->m_next = m2;
1986 					recm = slp->ns_raw;
1987 				} else
1988 					recm = m2;
1989 				len = slp->ns_reclen;
1990 			} else if ((len + m->m_len) == slp->ns_reclen) {
1991 				om = m;
1992 				len += m->m_len;
1993 				m = m->m_next;
1994 				recm = slp->ns_raw;
1995 				om->m_next = (struct mbuf *)0;
1996 			} else {
1997 				om = m;
1998 				len += m->m_len;
1999 				m = m->m_next;
2000 			}
2001 		}
2002 		slp->ns_raw = m;
2003 		slp->ns_cc -= len;
2004 		slp->ns_reclen = 0;
2005 	    } else {
2006 		slp->ns_flag &= ~SLP_GETSTREAM;
2007 		return (0);
2008 	    }
2009 
2010 	    /*
2011 	     * Accumulate the fragments into a record.
2012 	     */
2013 	    mpp = &slp->ns_frag;
2014 	    while (*mpp)
2015 		mpp = &((*mpp)->m_next);
2016 	    *mpp = recm;
2017 	    if (slp->ns_flag & SLP_LASTFRAG) {
2018 		if (slp->ns_recend)
2019 		    slp->ns_recend->m_nextpkt = slp->ns_frag;
2020 		else
2021 		    slp->ns_rec = slp->ns_frag;
2022 		slp->ns_recend = slp->ns_frag;
2023 		slp->ns_frag = (struct mbuf *)0;
2024 	    }
2025 	}
2026 }
2027 
2028 /*
2029  * Parse an RPC header.
2030  */
2031 int
2032 nfsrv_dorec(slp, nfsd, ndp)
2033 	register struct nfssvc_sock *slp;
2034 	struct nfsd *nfsd;
2035 	struct nfsrv_descript **ndp;
2036 {
2037 	register struct mbuf *m, *nam;
2038 	register struct nfsrv_descript *nd;
2039 	int error;
2040 
2041 	*ndp = NULL;
2042 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2043 	    (m = slp->ns_rec) == (struct mbuf *)0)
2044 		return (ENOBUFS);
2045 	slp->ns_rec = m->m_nextpkt;
2046 	if (slp->ns_rec)
2047 		m->m_nextpkt = (struct mbuf *)0;
2048 	else
2049 		slp->ns_recend = (struct mbuf *)0;
2050 	if (m->m_type == MT_SONAME) {
2051 		nam = m;
2052 		m = m->m_next;
2053 		nam->m_next = NULL;
2054 	} else
2055 		nam = NULL;
2056 	MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2057 		M_NFSRVDESC, M_WAITOK);
2058 	nd->nd_md = nd->nd_mrep = m;
2059 	nd->nd_nam2 = nam;
2060 	nd->nd_dpos = mtod(m, caddr_t);
2061 	error = nfs_getreq(nd, nfsd, TRUE);
2062 	if (error) {
2063 		m_freem(nam);
2064 		free((caddr_t)nd, M_NFSRVDESC);
2065 		return (error);
2066 	}
2067 	*ndp = nd;
2068 	nfsd->nfsd_nd = nd;
2069 	return (0);
2070 }
2071 
2072 
2073 /*
2074  * Search for a sleeping nfsd and wake it up.
2075  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2076  * running nfsds will go look for the work in the nfssvc_sock list.
2077  */
2078 void
2079 nfsrv_wakenfsd(slp)
2080 	struct nfssvc_sock *slp;
2081 {
2082 	register struct nfsd *nd;
2083 
2084 	if ((slp->ns_flag & SLP_VALID) == 0)
2085 		return;
2086 	for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) {
2087 		if (nd->nfsd_flag & NFSD_WAITING) {
2088 			nd->nfsd_flag &= ~NFSD_WAITING;
2089 			if (nd->nfsd_slp)
2090 				panic("nfsd wakeup");
2091 			slp->ns_sref++;
2092 			nd->nfsd_slp = slp;
2093 			wakeup((caddr_t)nd);
2094 			return;
2095 		}
2096 	}
2097 	slp->ns_flag |= SLP_DOREC;
2098 	nfsd_head_flag |= NFSD_CHECKSLP;
2099 }
2100 #endif /* NFSSERVER */
2101