xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 1f2744e6e4915c9da2a3f980279398c4cf7d5e6d)
1 /*	$NetBSD: nfs_socket.c,v 1.17 1994/08/17 14:43:51 mycroft Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1991, 1993
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	8.3 (Berkeley) 1/12/94
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 
59 #include <netinet/in.h>
60 #include <netinet/tcp.h>
61 #include <nfs/rpcv2.h>
62 #include <nfs/nfsv2.h>
63 #include <nfs/nfs.h>
64 #include <nfs/xdr_subs.h>
65 #include <nfs/nfsm_subs.h>
66 #include <nfs/nfsmount.h>
67 #include <nfs/nfsnode.h>
68 #include <nfs/nfsrtt.h>
69 #include <nfs/nqnfs.h>
70 
71 #define	TRUE	1
72 #define	FALSE	0
73 
74 /*
75  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
76  * Use the mean and mean deviation of rtt for the appropriate type of rpc
77  * for the frequent rpcs and a default for the others.
78  * The justification for doing "other" this way is that these rpcs
79  * happen so infrequently that timer est. would probably be stale.
80  * Also, since many of these rpcs are
81  * non-idempotent, a conservative timeout is desired.
82  * getattr, lookup - A+2D
83  * read, write     - A+4D
84  * other           - nm_timeo
85  */
86 #define	NFS_RTO(n, t) \
87 	((t) == 0 ? (n)->nm_timeo : \
88 	 ((t) < 3 ? \
89 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
90 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
91 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
92 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
93 /*
94  * External data, mostly RPC constants in XDR form
95  */
96 extern u_long rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, rpc_auth_unix,
97 	rpc_msgaccepted, rpc_call, rpc_autherr, rpc_rejectedcred,
98 	rpc_auth_kerb;
99 extern u_long nfs_prog, nfs_vers, nqnfs_prog, nqnfs_vers;
100 extern time_t nqnfsstarttime;
101 extern int nonidempotent[NFS_NPROCS];
102 
103 /*
104  * Maps errno values to nfs error numbers.
105  * Use NFSERR_IO as the catch all for ones not specifically defined in
106  * RFC 1094.
107  */
108 static int nfsrv_errmap[ELAST] = {
109   NFSERR_PERM,	NFSERR_NOENT,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
110   NFSERR_NXIO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
111   NFSERR_IO,	NFSERR_IO,	NFSERR_ACCES,	NFSERR_IO,	NFSERR_IO,
112   NFSERR_IO,	NFSERR_EXIST,	NFSERR_IO,	NFSERR_NODEV,	NFSERR_NOTDIR,
113   NFSERR_ISDIR,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
114   NFSERR_IO,	NFSERR_FBIG,	NFSERR_NOSPC,	NFSERR_IO,	NFSERR_ROFS,
115   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
116   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
117   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
118   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
119   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
120   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
121   NFSERR_IO,	NFSERR_IO,	NFSERR_NAMETOL,	NFSERR_IO,	NFSERR_IO,
122   NFSERR_NOTEMPTY, NFSERR_IO,	NFSERR_IO,	NFSERR_DQUOT,	NFSERR_STALE,
123   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
124   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
125   NFSERR_IO,
126 };
127 
128 /*
129  * Defines which timer to use for the procnum.
130  * 0 - default
131  * 1 - getattr
132  * 2 - lookup
133  * 3 - read
134  * 4 - write
135  */
136 static int proct[NFS_NPROCS] = {
137 	0, 1, 0, 0, 2, 3, 3, 0, 4, 0, 0, 0, 0, 0, 0, 0, 3, 0, 3, 0, 0, 0, 0,
138 };
139 
140 /*
141  * There is a congestion window for outstanding rpcs maintained per mount
142  * point. The cwnd size is adjusted in roughly the way that:
143  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
144  * SIGCOMM '88". ACM, August 1988.
145  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
146  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
147  * of rpcs is in progress.
148  * (The sent count and cwnd are scaled for integer arith.)
149  * Variants of "slow start" were tried and were found to be too much of a
150  * performance hit (ave. rtt 3 times larger),
151  * I suspect due to the large rtt that nfs rpcs have.
152  */
153 #define	NFS_CWNDSCALE	256
154 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
155 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
156 int	nfs_sbwait();
157 void	nfs_disconnect(), nfs_realign(), nfsrv_wakenfsd(), nfs_sndunlock();
158 void	nfs_rcvunlock(), nqnfs_serverd(), nqnfs_clientlease();
159 struct mbuf *nfsm_rpchead();
160 int nfsrtton = 0;
161 struct nfsrtt nfsrtt;
162 
163 /*
164  * Initialize sockets and congestion for a new NFS connection.
165  * We do not free the sockaddr if error.
166  */
167 nfs_connect(nmp, rep)
168 	register struct nfsmount *nmp;
169 	struct nfsreq *rep;
170 {
171 	register struct socket *so;
172 	int s, error, rcvreserve, sndreserve;
173 	struct sockaddr *saddr;
174 	struct sockaddr_in *sin;
175 	struct mbuf *m;
176 	u_short tport;
177 
178 	nmp->nm_so = (struct socket *)0;
179 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
180 	if (error = socreate(saddr->sa_family,
181 		&nmp->nm_so, nmp->nm_sotype, nmp->nm_soproto))
182 		goto bad;
183 	so = nmp->nm_so;
184 	nmp->nm_soflags = so->so_proto->pr_flags;
185 
186 	/*
187 	 * Some servers require that the client port be a reserved port number.
188 	 */
189 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
190 		MGET(m, M_WAIT, MT_SONAME);
191 		sin = mtod(m, struct sockaddr_in *);
192 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
193 		sin->sin_family = AF_INET;
194 		sin->sin_addr.s_addr = INADDR_ANY;
195 		tport = IPPORT_RESERVED - 1;
196 		sin->sin_port = htons(tport);
197 		while ((error = sobind(so, m)) == EADDRINUSE &&
198 		       --tport > IPPORT_RESERVED / 2)
199 			sin->sin_port = htons(tport);
200 		m_freem(m);
201 		if (error)
202 			goto bad;
203 	}
204 
205 	/*
206 	 * Protocols that do not require connections may be optionally left
207 	 * unconnected for servers that reply from a port other than NFS_PORT.
208 	 */
209 	if (nmp->nm_flag & NFSMNT_NOCONN) {
210 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
211 			error = ENOTCONN;
212 			goto bad;
213 		}
214 	} else {
215 		if (error = soconnect(so, nmp->nm_nam))
216 			goto bad;
217 
218 		/*
219 		 * Wait for the connection to complete. Cribbed from the
220 		 * connect system call but with the wait timing out so
221 		 * that interruptible mounts don't hang here for a long time.
222 		 */
223 		s = splnet();
224 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
225 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
226 				"nfscon", 2 * hz);
227 			if ((so->so_state & SS_ISCONNECTING) &&
228 			    so->so_error == 0 && rep &&
229 			    (error = nfs_sigintr(nmp, rep, rep->r_procp))) {
230 				so->so_state &= ~SS_ISCONNECTING;
231 				splx(s);
232 				goto bad;
233 			}
234 		}
235 		if (so->so_error) {
236 			error = so->so_error;
237 			so->so_error = 0;
238 			splx(s);
239 			goto bad;
240 		}
241 		splx(s);
242 	}
243 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
244 		so->so_rcv.sb_timeo = (5 * hz);
245 		so->so_snd.sb_timeo = (5 * hz);
246 	} else {
247 		so->so_rcv.sb_timeo = 0;
248 		so->so_snd.sb_timeo = 0;
249 	}
250 	if (nmp->nm_sotype == SOCK_DGRAM) {
251 		sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR;
252 		rcvreserve = nmp->nm_rsize + NFS_MAXPKTHDR;
253 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
254 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
255 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR) * 2;
256 	} else {
257 		if (nmp->nm_sotype != SOCK_STREAM)
258 			panic("nfscon sotype");
259 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
260 			MGET(m, M_WAIT, MT_SOOPTS);
261 			*mtod(m, int *) = 1;
262 			m->m_len = sizeof(int);
263 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
264 		}
265 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
266 			MGET(m, M_WAIT, MT_SOOPTS);
267 			*mtod(m, int *) = 1;
268 			m->m_len = sizeof(int);
269 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
270 		}
271 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + sizeof (u_long))
272 				* 2;
273 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + sizeof (u_long))
274 				* 2;
275 	}
276 	if (error = soreserve(so, sndreserve, rcvreserve))
277 		goto bad;
278 	so->so_rcv.sb_flags |= SB_NOINTR;
279 	so->so_snd.sb_flags |= SB_NOINTR;
280 
281 	/* Initialize other non-zero congestion variables */
282 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
283 		nmp->nm_srtt[4] = (NFS_TIMEO << 3);
284 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
285 		nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0;
286 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
287 	nmp->nm_sent = 0;
288 	nmp->nm_timeouts = 0;
289 	return (0);
290 
291 bad:
292 	nfs_disconnect(nmp);
293 	return (error);
294 }
295 
296 /*
297  * Reconnect routine:
298  * Called when a connection is broken on a reliable protocol.
299  * - clean up the old socket
300  * - nfs_connect() again
301  * - set R_MUSTRESEND for all outstanding requests on mount point
302  * If this fails the mount point is DEAD!
303  * nb: Must be called with the nfs_sndlock() set on the mount point.
304  */
305 nfs_reconnect(rep)
306 	register struct nfsreq *rep;
307 {
308 	register struct nfsreq *rp;
309 	register struct nfsmount *nmp = rep->r_nmp;
310 	int error;
311 
312 	nfs_disconnect(nmp);
313 	while (error = nfs_connect(nmp, rep)) {
314 		if (error == EINTR || error == ERESTART)
315 			return (EINTR);
316 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0);
317 	}
318 
319 	/*
320 	 * Loop through outstanding request list and fix up all requests
321 	 * on old socket.
322 	 */
323 	for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) {
324 		if (rp->r_nmp == nmp)
325 			rp->r_flags |= R_MUSTRESEND;
326 	}
327 	return (0);
328 }
329 
330 /*
331  * NFS disconnect. Clean up and unlink.
332  */
333 void
334 nfs_disconnect(nmp)
335 	register struct nfsmount *nmp;
336 {
337 	register struct socket *so;
338 
339 	if (nmp->nm_so) {
340 		so = nmp->nm_so;
341 		nmp->nm_so = (struct socket *)0;
342 		soshutdown(so, 2);
343 		soclose(so);
344 	}
345 }
346 
347 /*
348  * This is the nfs send routine. For connection based socket types, it
349  * must be called with an nfs_sndlock() on the socket.
350  * "rep == NULL" indicates that it has been called from a server.
351  * For the client side:
352  * - return EINTR if the RPC is terminated, 0 otherwise
353  * - set R_MUSTRESEND if the send fails for any reason
354  * - do any cleanup required by recoverable socket errors (???)
355  * For the server side:
356  * - return EINTR or ERESTART if interrupted by a signal
357  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
358  * - do any cleanup required by recoverable socket errors (???)
359  */
360 nfs_send(so, nam, top, rep)
361 	register struct socket *so;
362 	struct mbuf *nam;
363 	register struct mbuf *top;
364 	struct nfsreq *rep;
365 {
366 	struct mbuf *sendnam;
367 	int error, soflags, flags;
368 
369 	if (rep) {
370 		if (rep->r_flags & R_SOFTTERM) {
371 			m_freem(top);
372 			return (EINTR);
373 		}
374 		if ((so = rep->r_nmp->nm_so) == NULL) {
375 			rep->r_flags |= R_MUSTRESEND;
376 			m_freem(top);
377 			return (0);
378 		}
379 		rep->r_flags &= ~R_MUSTRESEND;
380 		soflags = rep->r_nmp->nm_soflags;
381 	} else
382 		soflags = so->so_proto->pr_flags;
383 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
384 		sendnam = (struct mbuf *)0;
385 	else
386 		sendnam = nam;
387 	if (so->so_type == SOCK_SEQPACKET)
388 		flags = MSG_EOR;
389 	else
390 		flags = 0;
391 
392 	error = sosend(so, sendnam, (struct uio *)0, top,
393 		(struct mbuf *)0, flags);
394 	if (error) {
395 		if (rep) {
396 			log(LOG_INFO, "nfs send error %d for server %s\n",error,
397 			    rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
398 			/*
399 			 * Deal with errors for the client side.
400 			 */
401 			if (rep->r_flags & R_SOFTTERM)
402 				error = EINTR;
403 			else
404 				rep->r_flags |= R_MUSTRESEND;
405 		} else
406 			log(LOG_INFO, "nfsd send error %d\n", error);
407 
408 		/*
409 		 * Handle any recoverable (soft) socket errors here. (???)
410 		 */
411 		if (error != EINTR && error != ERESTART &&
412 			error != EWOULDBLOCK && error != EPIPE)
413 			error = 0;
414 	}
415 	return (error);
416 }
417 
418 #ifdef NFSCLIENT
419 /*
420  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
421  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
422  * Mark and consolidate the data into a new mbuf list.
423  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
424  *     small mbufs.
425  * For SOCK_STREAM we must be very careful to read an entire record once
426  * we have read any of it, even if the system call has been interrupted.
427  */
428 nfs_receive(rep, aname, mp)
429 	register struct nfsreq *rep;
430 	struct mbuf **aname;
431 	struct mbuf **mp;
432 {
433 	register struct socket *so;
434 	struct uio auio;
435 	struct iovec aio;
436 	register struct mbuf *m;
437 	struct mbuf *control;
438 	u_long len;
439 	struct mbuf **getnam;
440 	int error, sotype, rcvflg;
441 	struct proc *p = curproc;	/* XXX */
442 
443 	/*
444 	 * Set up arguments for soreceive()
445 	 */
446 	*mp = (struct mbuf *)0;
447 	*aname = (struct mbuf *)0;
448 	sotype = rep->r_nmp->nm_sotype;
449 
450 	/*
451 	 * For reliable protocols, lock against other senders/receivers
452 	 * in case a reconnect is necessary.
453 	 * For SOCK_STREAM, first get the Record Mark to find out how much
454 	 * more there is to get.
455 	 * We must lock the socket against other receivers
456 	 * until we have an entire rpc request/reply.
457 	 */
458 	if (sotype != SOCK_DGRAM) {
459 		if (error = nfs_sndlock(&rep->r_nmp->nm_flag, rep))
460 			return (error);
461 tryagain:
462 		/*
463 		 * Check for fatal errors and resending request.
464 		 */
465 		/*
466 		 * Ugh: If a reconnect attempt just happened, nm_so
467 		 * would have changed. NULL indicates a failed
468 		 * attempt that has essentially shut down this
469 		 * mount point.
470 		 */
471 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
472 			nfs_sndunlock(&rep->r_nmp->nm_flag);
473 			return (EINTR);
474 		}
475 		if ((so = rep->r_nmp->nm_so) == NULL) {
476 			if (error = nfs_reconnect(rep)) {
477 				nfs_sndunlock(&rep->r_nmp->nm_flag);
478 				return (error);
479 			}
480 			goto tryagain;
481 		}
482 		while (rep->r_flags & R_MUSTRESEND) {
483 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
484 			nfsstats.rpcretries++;
485 			if (error = nfs_send(so, rep->r_nmp->nm_nam, m, rep)) {
486 				if (error == EINTR || error == ERESTART ||
487 				    (error = nfs_reconnect(rep))) {
488 					nfs_sndunlock(&rep->r_nmp->nm_flag);
489 					return (error);
490 				}
491 				goto tryagain;
492 			}
493 		}
494 		nfs_sndunlock(&rep->r_nmp->nm_flag);
495 		if (sotype == SOCK_STREAM) {
496 			aio.iov_base = (caddr_t) &len;
497 			aio.iov_len = sizeof(u_long);
498 			auio.uio_iov = &aio;
499 			auio.uio_iovcnt = 1;
500 			auio.uio_segflg = UIO_SYSSPACE;
501 			auio.uio_rw = UIO_READ;
502 			auio.uio_offset = 0;
503 			auio.uio_resid = sizeof(u_long);
504 			auio.uio_procp = p;
505 			do {
506 			   rcvflg = MSG_WAITALL;
507 			   error = soreceive(so, (struct mbuf **)0, &auio,
508 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
509 			   if (error == EWOULDBLOCK && rep) {
510 				if (rep->r_flags & R_SOFTTERM)
511 					return (EINTR);
512 			   }
513 			} while (error == EWOULDBLOCK);
514 			if (!error && auio.uio_resid > 0) {
515 			    log(LOG_INFO,
516 				 "short receive (%d/%d) from nfs server %s\n",
517 				 sizeof(u_long) - auio.uio_resid,
518 				 sizeof(u_long),
519 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
520 			    error = EPIPE;
521 			}
522 			if (error)
523 				goto errout;
524 			len = ntohl(len) & ~0x80000000;
525 			/*
526 			 * This is SERIOUS! We are out of sync with the sender
527 			 * and forcing a disconnect/reconnect is all I can do.
528 			 */
529 			if (len > NFS_MAXPACKET) {
530 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
531 				"impossible packet length",
532 				len,
533 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
534 			    error = EFBIG;
535 			    goto errout;
536 			}
537 			auio.uio_resid = len;
538 			do {
539 			    rcvflg = MSG_WAITALL;
540 			    error =  soreceive(so, (struct mbuf **)0,
541 				&auio, mp, (struct mbuf **)0, &rcvflg);
542 			} while (error == EWOULDBLOCK || error == EINTR ||
543 				 error == ERESTART);
544 			if (!error && auio.uio_resid > 0) {
545 			    log(LOG_INFO,
546 				"short receive (%d/%d) from nfs server %s\n",
547 				len - auio.uio_resid, len,
548 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
549 			    error = EPIPE;
550 			}
551 		} else {
552 			/*
553 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
554 			 * and soreceive() will return when it has either a
555 			 * control msg or a data msg.
556 			 * We have no use for control msg., but must grab them
557 			 * and then throw them away so we know what is going
558 			 * on.
559 			 */
560 			auio.uio_resid = len = 100000000; /* Anything Big */
561 			auio.uio_procp = p;
562 			do {
563 			    rcvflg = 0;
564 			    error =  soreceive(so, (struct mbuf **)0,
565 				&auio, mp, &control, &rcvflg);
566 			    if (control)
567 				m_freem(control);
568 			    if (error == EWOULDBLOCK && rep) {
569 				if (rep->r_flags & R_SOFTTERM)
570 					return (EINTR);
571 			    }
572 			} while (error == EWOULDBLOCK ||
573 				 (!error && *mp == NULL && control));
574 			if ((rcvflg & MSG_EOR) == 0)
575 				printf("Egad!!\n");
576 			if (!error && *mp == NULL)
577 				error = EPIPE;
578 			len -= auio.uio_resid;
579 		}
580 errout:
581 		if (error && error != EINTR && error != ERESTART) {
582 			m_freem(*mp);
583 			*mp = (struct mbuf *)0;
584 			if (error != EPIPE)
585 				log(LOG_INFO,
586 				    "receive error %d from nfs server %s\n",
587 				    error,
588 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
589 			error = nfs_sndlock(&rep->r_nmp->nm_flag, rep);
590 			if (!error)
591 				error = nfs_reconnect(rep);
592 			if (!error)
593 				goto tryagain;
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 	/*
619 	 * Search for any mbufs that are not a multiple of 4 bytes long
620 	 * or with m_data not longword aligned.
621 	 * These could cause pointer alignment problems, so copy them to
622 	 * well aligned mbufs.
623 	 */
624 	nfs_realign(*mp, 5 * NFSX_UNSIGNED);
625 	return (error);
626 }
627 
628 /*
629  * Implement receipt of reply on a socket.
630  * We must search through the list of received datagrams matching them
631  * with outstanding requests using the xid, until ours is found.
632  */
633 /* ARGSUSED */
634 nfs_reply(myrep)
635 	struct nfsreq *myrep;
636 {
637 	register struct nfsreq *rep;
638 	register struct nfsmount *nmp = myrep->r_nmp;
639 	register long t1;
640 	struct mbuf *mrep, *nam, *md;
641 	u_long rxid, *tl;
642 	caddr_t dpos, cp2;
643 	int error;
644 
645 	/*
646 	 * Loop around until we get our own reply
647 	 */
648 	for (;;) {
649 		/*
650 		 * Lock against other receivers so that I don't get stuck in
651 		 * sbwait() after someone else has received my reply for me.
652 		 * Also necessary for connection based protocols to avoid
653 		 * race conditions during a reconnect.
654 		 */
655 		if (error = nfs_rcvlock(myrep))
656 			return (error);
657 		/* Already received, bye bye */
658 		if (myrep->r_mrep != NULL) {
659 			nfs_rcvunlock(&nmp->nm_flag);
660 			return (0);
661 		}
662 		/*
663 		 * Get the next Rpc reply off the socket
664 		 */
665 		error = nfs_receive(myrep, &nam, &mrep);
666 		nfs_rcvunlock(&nmp->nm_flag);
667 		if (error) {
668 
669 			/*
670 			 * Ignore routing errors on connectionless protocols??
671 			 */
672 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
673 				nmp->nm_so->so_error = 0;
674 				if (myrep->r_flags & R_GETONEREP)
675 					return (0);
676 				continue;
677 			}
678 			return (error);
679 		}
680 		if (nam)
681 			m_freem(nam);
682 
683 		/*
684 		 * Get the xid and check that it is an rpc reply
685 		 */
686 		md = mrep;
687 		dpos = mtod(md, caddr_t);
688 		nfsm_dissect(tl, u_long *, 2*NFSX_UNSIGNED);
689 		rxid = *tl++;
690 		if (*tl != rpc_reply) {
691 			if (nmp->nm_flag & NFSMNT_NQNFS) {
692 				if (nqnfs_callback(nmp, mrep, md, dpos))
693 					nfsstats.rpcinvalid++;
694 			} else {
695 				nfsstats.rpcinvalid++;
696 				m_freem(mrep);
697 			}
698 nfsmout:
699 			if (myrep->r_flags & R_GETONEREP)
700 				return (0);
701 			continue;
702 		}
703 
704 		/*
705 		 * Loop through the request list to match up the reply
706 		 * Iff no match, just drop the datagram
707 		 */
708 		for (rep = nfs_reqq.tqh_first; rep != 0;
709 		    rep = rep->r_chain.tqe_next) {
710 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
711 				/* Found it.. */
712 				rep->r_mrep = mrep;
713 				rep->r_md = md;
714 				rep->r_dpos = dpos;
715 				if (nfsrtton) {
716 					struct rttl *rt;
717 
718 					rt = &nfsrtt.rttl[nfsrtt.pos];
719 					rt->proc = rep->r_procnum;
720 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
721 					rt->sent = nmp->nm_sent;
722 					rt->cwnd = nmp->nm_cwnd;
723 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
724 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
725 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
726 					rt->tstamp = time;
727 					if (rep->r_flags & R_TIMING)
728 						rt->rtt = rep->r_rtt;
729 					else
730 						rt->rtt = 1000000;
731 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
732 				}
733 				/*
734 				 * Update congestion window.
735 				 * Do the additive increase of
736 				 * one rpc/rtt.
737 				 */
738 				if (nmp->nm_cwnd <= nmp->nm_sent) {
739 					nmp->nm_cwnd +=
740 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
741 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
742 					if (nmp->nm_cwnd > NFS_MAXCWND)
743 						nmp->nm_cwnd = NFS_MAXCWND;
744 				}
745 				rep->r_flags &= ~R_SENT;
746 				nmp->nm_sent -= NFS_CWNDSCALE;
747 				/*
748 				 * Update rtt using a gain of 0.125 on the mean
749 				 * and a gain of 0.25 on the deviation.
750 				 */
751 				if (rep->r_flags & R_TIMING) {
752 					/*
753 					 * Since the timer resolution of
754 					 * NFS_HZ is so course, it can often
755 					 * result in r_rtt == 0. Since
756 					 * r_rtt == N means that the actual
757 					 * rtt is between N+dt and N+2-dt ticks,
758 					 * add 1.
759 					 */
760 					t1 = rep->r_rtt + 1;
761 					t1 -= (NFS_SRTT(rep) >> 3);
762 					NFS_SRTT(rep) += t1;
763 					if (t1 < 0)
764 						t1 = -t1;
765 					t1 -= (NFS_SDRTT(rep) >> 2);
766 					NFS_SDRTT(rep) += t1;
767 				}
768 				nmp->nm_timeouts = 0;
769 				break;
770 			}
771 		}
772 		/*
773 		 * If not matched to a request, drop it.
774 		 * If it's mine, get out.
775 		 */
776 		if (rep == 0) {
777 			nfsstats.rpcunexpected++;
778 			m_freem(mrep);
779 		} else if (rep == myrep) {
780 			if (rep->r_mrep == NULL)
781 				panic("nfsreply nil");
782 			return (0);
783 		}
784 		if (myrep->r_flags & R_GETONEREP)
785 			return (0);
786 	}
787 }
788 
789 /*
790  * nfs_request - goes something like this
791  *	- fill in request struct
792  *	- links it into list
793  *	- calls nfs_send() for first transmit
794  *	- calls nfs_receive() to get reply
795  *	- break down rpc header and return with nfs reply pointed to
796  *	  by mrep or error
797  * nb: always frees up mreq mbuf list
798  */
799 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
800 	struct vnode *vp;
801 	struct mbuf *mrest;
802 	int procnum;
803 	struct proc *procp;
804 	struct ucred *cred;
805 	struct mbuf **mrp;
806 	struct mbuf **mdp;
807 	caddr_t *dposp;
808 {
809 	register struct mbuf *m, *mrep;
810 	register struct nfsreq *rep;
811 	register u_long *tl;
812 	register int i;
813 	struct nfsmount *nmp;
814 	struct mbuf *md, *mheadend;
815 	struct nfsreq *reph;
816 	struct nfsnode *np;
817 	time_t reqtime, waituntil;
818 	caddr_t dpos, cp2;
819 	int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type;
820 	int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0;
821 	u_long xid;
822 	u_quad_t frev;
823 	char *auth_str;
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 	auth_str = (char *)0;
844 	if (nmp->nm_flag & NFSMNT_KERB) {
845 		if (failed_auth) {
846 			error = nfs_getauth(nmp, rep, cred, &auth_type,
847 				&auth_str, &auth_len);
848 			if (error) {
849 				free((caddr_t)rep, M_NFSREQ);
850 				m_freem(mrest);
851 				return (error);
852 			}
853 		} else {
854 			auth_type = RPCAUTH_UNIX;
855 			auth_len = 5 * NFSX_UNSIGNED;
856 		}
857 	} else {
858 		auth_type = RPCAUTH_UNIX;
859 		if (cred->cr_ngroups < 1)
860 			panic("nfsreq nogrps");
861 		auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ?
862 			nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) +
863 			5 * NFSX_UNSIGNED;
864 	}
865 	m = nfsm_rpchead(cred, (nmp->nm_flag & NFSMNT_NQNFS), procnum,
866 	     auth_type, auth_len, auth_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_long *) = 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 = splsoftclock();
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_flag, 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_flag);
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 = splsoftclock();
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_long *, 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 (*tl == rpc_rejectedcred && failed_auth == 0) {
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 	 * skip over the auth_verf, someday we may want to cache auth_short's
995 	 * for nfs_reqhead(), but for now just dump it
996 	 */
997 	if (*++tl != 0) {
998 		i = nfsm_rndup(fxdr_unsigned(long, *tl));
999 		nfsm_adv(i);
1000 	}
1001 	nfsm_dissect(tl, u_long *, NFSX_UNSIGNED);
1002 	/* 0 == ok */
1003 	if (*tl == 0) {
1004 		nfsm_dissect(tl, u_long *, NFSX_UNSIGNED);
1005 		if (*tl != 0) {
1006 			error = fxdr_unsigned(int, *tl);
1007 			m_freem(mrep);
1008 			if ((nmp->nm_flag & NFSMNT_NQNFS) &&
1009 			    error == NQNFS_TRYLATER) {
1010 				error = 0;
1011 				waituntil = time.tv_sec + trylater_delay;
1012 				while (time.tv_sec < waituntil)
1013 					(void) tsleep((caddr_t)&lbolt,
1014 						PSOCK, "nqnfstry", 0);
1015 				trylater_delay *= nfs_backoff[trylater_cnt];
1016 				if (trylater_cnt < 7)
1017 					trylater_cnt++;
1018 				goto tryagain;
1019 			}
1020 
1021 			/*
1022 			 * If the File Handle was stale, invalidate the
1023 			 * lookup cache, just in case.
1024 			 */
1025 			if (error == ESTALE)
1026 				cache_purge(vp);
1027 			m_freem(rep->r_mreq);
1028 			free((caddr_t)rep, M_NFSREQ);
1029 			return (error);
1030 		}
1031 
1032 		/*
1033 		 * For nqnfs, get any lease in reply
1034 		 */
1035 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1036 			nfsm_dissect(tl, u_long *, NFSX_UNSIGNED);
1037 			if (*tl) {
1038 				np = VTONFS(vp);
1039 				nqlflag = fxdr_unsigned(int, *tl);
1040 				nfsm_dissect(tl, u_long *, 4*NFSX_UNSIGNED);
1041 				cachable = fxdr_unsigned(int, *tl++);
1042 				reqtime += fxdr_unsigned(int, *tl++);
1043 				if (reqtime > time.tv_sec) {
1044 				    fxdr_hyper(tl, &frev);
1045 				    nqnfs_clientlease(nmp, np, nqlflag,
1046 					cachable, reqtime, frev);
1047 				}
1048 			}
1049 		}
1050 		*mrp = mrep;
1051 		*mdp = md;
1052 		*dposp = dpos;
1053 		m_freem(rep->r_mreq);
1054 		FREE((caddr_t)rep, M_NFSREQ);
1055 		return (0);
1056 	}
1057 	m_freem(mrep);
1058 	m_freem(rep->r_mreq);
1059 	free((caddr_t)rep, M_NFSREQ);
1060 	error = EPROTONOSUPPORT;
1061 nfsmout:
1062 	return (error);
1063 }
1064 #endif /* NFSCLIENT */
1065 
1066 /*
1067  * Generate the rpc reply header
1068  * siz arg. is used to decide if adding a cluster is worthwhile
1069  */
1070 nfs_rephead(siz, nd, err, cache, frev, mrq, mbp, bposp)
1071 	int siz;
1072 	struct nfsd *nd;
1073 	int err;
1074 	int cache;
1075 	u_quad_t *frev;
1076 	struct mbuf **mrq;
1077 	struct mbuf **mbp;
1078 	caddr_t *bposp;
1079 {
1080 	register u_long *tl;
1081 	register struct mbuf *mreq;
1082 	caddr_t bpos;
1083 	struct mbuf *mb, *mb2;
1084 
1085 	MGETHDR(mreq, M_WAIT, MT_DATA);
1086 	mb = mreq;
1087 	/*
1088 	 * If this is a big reply, use a cluster else
1089 	 * try and leave leading space for the lower level headers.
1090 	 */
1091 	siz += RPC_REPLYSIZ;
1092 	if (siz >= MINCLSIZE) {
1093 		MCLGET(mreq, M_WAIT);
1094 	} else
1095 		mreq->m_data += max_hdr;
1096 	tl = mtod(mreq, u_long *);
1097 	mreq->m_len = 6*NFSX_UNSIGNED;
1098 	bpos = ((caddr_t)tl)+mreq->m_len;
1099 	*tl++ = txdr_unsigned(nd->nd_retxid);
1100 	*tl++ = rpc_reply;
1101 	if (err == ERPCMISMATCH || err == NQNFS_AUTHERR) {
1102 		*tl++ = rpc_msgdenied;
1103 		if (err == NQNFS_AUTHERR) {
1104 			*tl++ = rpc_autherr;
1105 			*tl = rpc_rejectedcred;
1106 			mreq->m_len -= NFSX_UNSIGNED;
1107 			bpos -= NFSX_UNSIGNED;
1108 		} else {
1109 			*tl++ = rpc_mismatch;
1110 			*tl++ = txdr_unsigned(2);
1111 			*tl = txdr_unsigned(2);
1112 		}
1113 	} else {
1114 		*tl++ = rpc_msgaccepted;
1115 		*tl++ = 0;
1116 		*tl++ = 0;
1117 		switch (err) {
1118 		case EPROGUNAVAIL:
1119 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1120 			break;
1121 		case EPROGMISMATCH:
1122 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1123 			nfsm_build(tl, u_long *, 2*NFSX_UNSIGNED);
1124 			*tl++ = txdr_unsigned(2);
1125 			*tl = txdr_unsigned(2);	/* someday 3 */
1126 			break;
1127 		case EPROCUNAVAIL:
1128 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1129 			break;
1130 		default:
1131 			*tl = 0;
1132 			if (err != VNOVAL) {
1133 				nfsm_build(tl, u_long *, NFSX_UNSIGNED);
1134 				if (err)
1135 					*tl = txdr_unsigned(nfsrv_errmap[err - 1]);
1136 				else
1137 					*tl = 0;
1138 			}
1139 			break;
1140 		};
1141 	}
1142 
1143 	/*
1144 	 * For nqnfs, piggyback lease as requested.
1145 	 */
1146 	if (nd->nd_nqlflag != NQL_NOVAL && err == 0) {
1147 		if (nd->nd_nqlflag) {
1148 			nfsm_build(tl, u_long *, 5*NFSX_UNSIGNED);
1149 			*tl++ = txdr_unsigned(nd->nd_nqlflag);
1150 			*tl++ = txdr_unsigned(cache);
1151 			*tl++ = txdr_unsigned(nd->nd_duration);
1152 			txdr_hyper(frev, tl);
1153 		} else {
1154 			if (nd->nd_nqlflag != 0)
1155 				panic("nqreph");
1156 			nfsm_build(tl, u_long *, NFSX_UNSIGNED);
1157 			*tl = 0;
1158 		}
1159 	}
1160 	*mrq = mreq;
1161 	*mbp = mb;
1162 	*bposp = bpos;
1163 	if (err != 0 && err != VNOVAL)
1164 		nfsstats.srvrpc_errs++;
1165 	return (0);
1166 }
1167 
1168 /*
1169  * Nfs timer routine
1170  * Scan the nfsreq list and retranmit any requests that have timed out
1171  * To avoid retransmission attempts on STREAM sockets (in the future) make
1172  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1173  */
1174 void
1175 nfs_timer(arg)
1176 	void *arg;
1177 {
1178 	register struct nfsreq *rep;
1179 	register struct mbuf *m;
1180 	register struct socket *so;
1181 	register struct nfsmount *nmp;
1182 	register int timeo;
1183 	static long lasttime = 0;
1184 	int s, error;
1185 
1186 	s = splnet();
1187 	for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) {
1188 		nmp = rep->r_nmp;
1189 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1190 			continue;
1191 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1192 			rep->r_flags |= R_SOFTTERM;
1193 			continue;
1194 		}
1195 		if (rep->r_rtt >= 0) {
1196 			rep->r_rtt++;
1197 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1198 				timeo = nmp->nm_timeo;
1199 			else
1200 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1201 			if (nmp->nm_timeouts > 0)
1202 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1203 			if (rep->r_rtt <= timeo)
1204 				continue;
1205 			if (nmp->nm_timeouts < 8)
1206 				nmp->nm_timeouts++;
1207 		}
1208 		/*
1209 		 * Check for server not responding
1210 		 */
1211 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1212 		     rep->r_rexmit > nmp->nm_deadthresh) {
1213 			nfs_msg(rep->r_procp,
1214 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1215 			    "not responding");
1216 			rep->r_flags |= R_TPRINTFMSG;
1217 		}
1218 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1219 			nfsstats.rpctimeouts++;
1220 			rep->r_flags |= R_SOFTTERM;
1221 			continue;
1222 		}
1223 		if (nmp->nm_sotype != SOCK_DGRAM) {
1224 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1225 				rep->r_rexmit = NFS_MAXREXMIT;
1226 			continue;
1227 		}
1228 		if ((so = nmp->nm_so) == NULL)
1229 			continue;
1230 
1231 		/*
1232 		 * If there is enough space and the window allows..
1233 		 *	Resend it
1234 		 * Set r_rtt to -1 in case we fail to send it now.
1235 		 */
1236 		rep->r_rtt = -1;
1237 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1238 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1239 		    (rep->r_flags & R_SENT) ||
1240 		    nmp->nm_sent < nmp->nm_cwnd) &&
1241 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1242 			if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
1243 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1244 			    (struct mbuf *)0, (struct mbuf *)0);
1245 			else
1246 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1247 			    nmp->nm_nam, (struct mbuf *)0);
1248 			if (error) {
1249 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
1250 					so->so_error = 0;
1251 			} else {
1252 				/*
1253 				 * Iff first send, start timing
1254 				 * else turn timing off, backoff timer
1255 				 * and divide congestion window by 2.
1256 				 */
1257 				if (rep->r_flags & R_SENT) {
1258 					rep->r_flags &= ~R_TIMING;
1259 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1260 						rep->r_rexmit = NFS_MAXREXMIT;
1261 					nmp->nm_cwnd >>= 1;
1262 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1263 						nmp->nm_cwnd = NFS_CWNDSCALE;
1264 					nfsstats.rpcretries++;
1265 				} else {
1266 					rep->r_flags |= R_SENT;
1267 					nmp->nm_sent += NFS_CWNDSCALE;
1268 				}
1269 				rep->r_rtt = 0;
1270 			}
1271 		}
1272 	}
1273 
1274 #ifdef NFSSERVER
1275 	/*
1276 	 * Call the nqnfs server timer once a second to handle leases.
1277 	 */
1278 	if (lasttime != time.tv_sec) {
1279 		lasttime = time.tv_sec;
1280 		nqnfs_serverd();
1281 	}
1282 #endif /* NFSSERVER */
1283 	splx(s);
1284 	timeout(nfs_timer, (void *)0, hz / NFS_HZ);
1285 }
1286 
1287 /*
1288  * Test for a termination condition pending on the process.
1289  * This is used for NFSMNT_INT mounts.
1290  */
1291 nfs_sigintr(nmp, rep, p)
1292 	struct nfsmount *nmp;
1293 	struct nfsreq *rep;
1294 	register struct proc *p;
1295 {
1296 
1297 	if (rep && (rep->r_flags & R_SOFTTERM))
1298 		return (EINTR);
1299 	if (!(nmp->nm_flag & NFSMNT_INT))
1300 		return (0);
1301 	if (p && p->p_siglist &&
1302 	    (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) &
1303 	    NFSINT_SIGMASK))
1304 		return (EINTR);
1305 	return (0);
1306 }
1307 
1308 /*
1309  * Lock a socket against others.
1310  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1311  * and also to avoid race conditions between the processes with nfs requests
1312  * in progress when a reconnect is necessary.
1313  */
1314 nfs_sndlock(flagp, rep)
1315 	register int *flagp;
1316 	struct nfsreq *rep;
1317 {
1318 	struct proc *p;
1319 	int slpflag = 0, slptimeo = 0;
1320 
1321 	if (rep) {
1322 		p = rep->r_procp;
1323 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1324 			slpflag = PCATCH;
1325 	} else
1326 		p = (struct proc *)0;
1327 	while (*flagp & NFSMNT_SNDLOCK) {
1328 		if (nfs_sigintr(rep->r_nmp, rep, p))
1329 			return (EINTR);
1330 		*flagp |= NFSMNT_WANTSND;
1331 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1332 			slptimeo);
1333 		if (slpflag == PCATCH) {
1334 			slpflag = 0;
1335 			slptimeo = 2 * hz;
1336 		}
1337 	}
1338 	*flagp |= NFSMNT_SNDLOCK;
1339 	return (0);
1340 }
1341 
1342 /*
1343  * Unlock the stream socket for others.
1344  */
1345 void
1346 nfs_sndunlock(flagp)
1347 	register int *flagp;
1348 {
1349 
1350 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1351 		panic("nfs sndunlock");
1352 	*flagp &= ~NFSMNT_SNDLOCK;
1353 	if (*flagp & NFSMNT_WANTSND) {
1354 		*flagp &= ~NFSMNT_WANTSND;
1355 		wakeup((caddr_t)flagp);
1356 	}
1357 }
1358 
1359 nfs_rcvlock(rep)
1360 	register struct nfsreq *rep;
1361 {
1362 	register int *flagp = &rep->r_nmp->nm_flag;
1363 	int slpflag, slptimeo = 0;
1364 
1365 	if (*flagp & NFSMNT_INT)
1366 		slpflag = PCATCH;
1367 	else
1368 		slpflag = 0;
1369 	while (*flagp & NFSMNT_RCVLOCK) {
1370 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1371 			return (EINTR);
1372 		*flagp |= NFSMNT_WANTRCV;
1373 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1374 			slptimeo);
1375 		if (slpflag == PCATCH) {
1376 			slpflag = 0;
1377 			slptimeo = 2 * hz;
1378 		}
1379 	}
1380 	*flagp |= NFSMNT_RCVLOCK;
1381 	return (0);
1382 }
1383 
1384 /*
1385  * Unlock the stream socket for others.
1386  */
1387 void
1388 nfs_rcvunlock(flagp)
1389 	register int *flagp;
1390 {
1391 
1392 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1393 		panic("nfs rcvunlock");
1394 	*flagp &= ~NFSMNT_RCVLOCK;
1395 	if (*flagp & NFSMNT_WANTRCV) {
1396 		*flagp &= ~NFSMNT_WANTRCV;
1397 		wakeup((caddr_t)flagp);
1398 	}
1399 }
1400 
1401 /*
1402  * Check for badly aligned mbuf data areas and
1403  * realign data in an mbuf list by copying the data areas up, as required.
1404  */
1405 void
1406 nfs_realign(m, hsiz)
1407 	register struct mbuf *m;
1408 	int hsiz;
1409 {
1410 	register struct mbuf *m2;
1411 	register int siz, mlen, olen;
1412 	register caddr_t tcp, fcp;
1413 	struct mbuf *mnew;
1414 
1415 	while (m) {
1416 	    /*
1417 	     * This never happens for UDP, rarely happens for TCP
1418 	     * but frequently happens for iso transport.
1419 	     */
1420 	    if ((m->m_len & 0x3) || (mtod(m, int) & 0x3)) {
1421 		olen = m->m_len;
1422 		fcp = mtod(m, caddr_t);
1423 		if ((int)fcp & 0x3) {
1424 			m->m_flags &= ~M_PKTHDR;
1425 			if (m->m_flags & M_EXT)
1426 				m->m_data = m->m_ext.ext_buf +
1427 					((m->m_ext.ext_size - olen) & ~0x3);
1428 			else
1429 				m->m_data = m->m_dat;
1430 		}
1431 		m->m_len = 0;
1432 		tcp = mtod(m, caddr_t);
1433 		mnew = m;
1434 		m2 = m->m_next;
1435 
1436 		/*
1437 		 * If possible, only put the first invariant part
1438 		 * of the RPC header in the first mbuf.
1439 		 */
1440 		mlen = M_TRAILINGSPACE(m);
1441 		if (olen <= hsiz && mlen > hsiz)
1442 			mlen = hsiz;
1443 
1444 		/*
1445 		 * Loop through the mbuf list consolidating data.
1446 		 */
1447 		while (m) {
1448 			while (olen > 0) {
1449 				if (mlen == 0) {
1450 					m2->m_flags &= ~M_PKTHDR;
1451 					if (m2->m_flags & M_EXT)
1452 						m2->m_data = m2->m_ext.ext_buf;
1453 					else
1454 						m2->m_data = m2->m_dat;
1455 					m2->m_len = 0;
1456 					mlen = M_TRAILINGSPACE(m2);
1457 					tcp = mtod(m2, caddr_t);
1458 					mnew = m2;
1459 					m2 = m2->m_next;
1460 				}
1461 				siz = min(mlen, olen);
1462 				if (tcp != fcp)
1463 					bcopy(fcp, tcp, siz);
1464 				mnew->m_len += siz;
1465 				mlen -= siz;
1466 				olen -= siz;
1467 				tcp += siz;
1468 				fcp += siz;
1469 			}
1470 			m = m->m_next;
1471 			if (m) {
1472 				olen = m->m_len;
1473 				fcp = mtod(m, caddr_t);
1474 			}
1475 		}
1476 
1477 		/*
1478 		 * Finally, set m_len == 0 for any trailing mbufs that have
1479 		 * been copied out of.
1480 		 */
1481 		while (m2) {
1482 			m2->m_len = 0;
1483 			m2 = m2->m_next;
1484 		}
1485 		return;
1486 	    }
1487 	    m = m->m_next;
1488 	}
1489 }
1490 
1491 /*
1492  * Parse an RPC request
1493  * - verify it
1494  * - fill in the cred struct.
1495  */
1496 nfs_getreq(nd, has_header)
1497 	register struct nfsd *nd;
1498 	int has_header;
1499 {
1500 	register int len, i;
1501 	register u_long *tl;
1502 	register long t1;
1503 	struct uio uio;
1504 	struct iovec iov;
1505 	caddr_t dpos, cp2;
1506 	u_long nfsvers, auth_type;
1507 	int error = 0, nqnfs = 0;
1508 	struct mbuf *mrep, *md;
1509 
1510 	mrep = nd->nd_mrep;
1511 	md = nd->nd_md;
1512 	dpos = nd->nd_dpos;
1513 	if (has_header) {
1514 		nfsm_dissect(tl, u_long *, 10*NFSX_UNSIGNED);
1515 		nd->nd_retxid = fxdr_unsigned(u_long, *tl++);
1516 		if (*tl++ != rpc_call) {
1517 			m_freem(mrep);
1518 			return (EBADRPC);
1519 		}
1520 	} else {
1521 		nfsm_dissect(tl, u_long *, 8*NFSX_UNSIGNED);
1522 	}
1523 	nd->nd_repstat = 0;
1524 	if (*tl++ != rpc_vers) {
1525 		nd->nd_repstat = ERPCMISMATCH;
1526 		nd->nd_procnum = NFSPROC_NOOP;
1527 		return (0);
1528 	}
1529 	nfsvers = nfs_vers;
1530 	if (*tl != nfs_prog) {
1531 		if (*tl == nqnfs_prog) {
1532 			nqnfs++;
1533 			nfsvers = nqnfs_vers;
1534 		} else {
1535 			nd->nd_repstat = EPROGUNAVAIL;
1536 			nd->nd_procnum = NFSPROC_NOOP;
1537 			return (0);
1538 		}
1539 	}
1540 	tl++;
1541 	if (*tl++ != nfsvers) {
1542 		nd->nd_repstat = EPROGMISMATCH;
1543 		nd->nd_procnum = NFSPROC_NOOP;
1544 		return (0);
1545 	}
1546 	nd->nd_procnum = fxdr_unsigned(u_long, *tl++);
1547 	if (nd->nd_procnum == NFSPROC_NULL)
1548 		return (0);
1549 	if (nd->nd_procnum >= NFS_NPROCS ||
1550 		(!nqnfs && nd->nd_procnum > NFSPROC_STATFS) ||
1551 		(*tl != rpc_auth_unix && *tl != rpc_auth_kerb)) {
1552 		nd->nd_repstat = EPROCUNAVAIL;
1553 		nd->nd_procnum = NFSPROC_NOOP;
1554 		return (0);
1555 	}
1556 	auth_type = *tl++;
1557 	len = fxdr_unsigned(int, *tl++);
1558 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1559 		m_freem(mrep);
1560 		return (EBADRPC);
1561 	}
1562 
1563 	/*
1564 	 * Handle auth_unix or auth_kerb.
1565 	 */
1566 	if (auth_type == rpc_auth_unix) {
1567 		len = fxdr_unsigned(int, *++tl);
1568 		if (len < 0 || len > NFS_MAXNAMLEN) {
1569 			m_freem(mrep);
1570 			return (EBADRPC);
1571 		}
1572 		nfsm_adv(nfsm_rndup(len));
1573 		nfsm_dissect(tl, u_long *, 3*NFSX_UNSIGNED);
1574 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1575 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1576 		len = fxdr_unsigned(int, *tl);
1577 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1578 			m_freem(mrep);
1579 			return (EBADRPC);
1580 		}
1581 		nfsm_dissect(tl, u_long *, (len + 2)*NFSX_UNSIGNED);
1582 		for (i = 1; i <= len; i++)
1583 			if (i < NGROUPS)
1584 				nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1585 			else
1586 				tl++;
1587 		nd->nd_cr.cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1);
1588 	} else if (auth_type == rpc_auth_kerb) {
1589 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1590 		nd->nd_authlen = fxdr_unsigned(int, *tl);
1591 		uio.uio_resid = nfsm_rndup(nd->nd_authlen);
1592 		if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1593 			m_freem(mrep);
1594 			return (EBADRPC);
1595 		}
1596 		uio.uio_offset = 0;
1597 		uio.uio_iov = &iov;
1598 		uio.uio_iovcnt = 1;
1599 		uio.uio_segflg = UIO_SYSSPACE;
1600 		iov.iov_base = (caddr_t)nd->nd_authstr;
1601 		iov.iov_len = RPCAUTH_MAXSIZ;
1602 		nfsm_mtouio(&uio, uio.uio_resid);
1603 		nfsm_dissect(tl, u_long *, 2 * NFSX_UNSIGNED);
1604 		nd->nd_flag |= NFSD_NEEDAUTH;
1605 	}
1606 
1607 	/*
1608 	 * Do we have any use for the verifier.
1609 	 * According to the "Remote Procedure Call Protocol Spec." it
1610 	 * should be AUTH_NULL, but some clients make it AUTH_UNIX?
1611 	 * For now, just skip over it
1612 	 */
1613 	len = fxdr_unsigned(int, *++tl);
1614 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1615 		m_freem(mrep);
1616 		return (EBADRPC);
1617 	}
1618 	if (len > 0) {
1619 		nfsm_adv(nfsm_rndup(len));
1620 	}
1621 
1622 	/*
1623 	 * For nqnfs, get piggybacked lease request.
1624 	 */
1625 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
1626 		nfsm_dissect(tl, u_long *, NFSX_UNSIGNED);
1627 		nd->nd_nqlflag = fxdr_unsigned(int, *tl);
1628 		if (nd->nd_nqlflag) {
1629 			nfsm_dissect(tl, u_long *, NFSX_UNSIGNED);
1630 			nd->nd_duration = fxdr_unsigned(int, *tl);
1631 		} else
1632 			nd->nd_duration = NQ_MINLEASE;
1633 	} else {
1634 		nd->nd_nqlflag = NQL_NOVAL;
1635 		nd->nd_duration = NQ_MINLEASE;
1636 	}
1637 	nd->nd_md = md;
1638 	nd->nd_dpos = dpos;
1639 	return (0);
1640 nfsmout:
1641 	return (error);
1642 }
1643 
1644 nfs_msg(p, server, msg)
1645 	struct proc *p;
1646 	char *server, *msg;
1647 {
1648 	tpr_t tpr;
1649 
1650 	if (p)
1651 		tpr = tprintf_open(p);
1652 	else
1653 		tpr = NULL;
1654 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1655 	tprintf_close(tpr);
1656 }
1657 
1658 #ifdef NFSSERVER
1659 int	nfsrv_null(),
1660 	nfsrv_getattr(),
1661 	nfsrv_setattr(),
1662 	nfsrv_lookup(),
1663 	nfsrv_readlink(),
1664 	nfsrv_read(),
1665 	nfsrv_write(),
1666 	nfsrv_create(),
1667 	nfsrv_remove(),
1668 	nfsrv_rename(),
1669 	nfsrv_link(),
1670 	nfsrv_symlink(),
1671 	nfsrv_mkdir(),
1672 	nfsrv_rmdir(),
1673 	nfsrv_readdir(),
1674 	nfsrv_statfs(),
1675 	nfsrv_noop(),
1676 	nqnfsrv_readdirlook(),
1677 	nqnfsrv_getlease(),
1678 	nqnfsrv_vacated(),
1679 	nqnfsrv_access();
1680 
1681 int (*nfsrv_procs[NFS_NPROCS])() = {
1682 	nfsrv_null,
1683 	nfsrv_getattr,
1684 	nfsrv_setattr,
1685 	nfsrv_noop,
1686 	nfsrv_lookup,
1687 	nfsrv_readlink,
1688 	nfsrv_read,
1689 	nfsrv_noop,
1690 	nfsrv_write,
1691 	nfsrv_create,
1692 	nfsrv_remove,
1693 	nfsrv_rename,
1694 	nfsrv_link,
1695 	nfsrv_symlink,
1696 	nfsrv_mkdir,
1697 	nfsrv_rmdir,
1698 	nfsrv_readdir,
1699 	nfsrv_statfs,
1700 	nqnfsrv_readdirlook,
1701 	nqnfsrv_getlease,
1702 	nqnfsrv_vacated,
1703 	nfsrv_noop,
1704 	nqnfsrv_access,
1705 };
1706 
1707 /*
1708  * Socket upcall routine for the nfsd sockets.
1709  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1710  * Essentially do as much as possible non-blocking, else punt and it will
1711  * be called with M_WAIT from an nfsd.
1712  */
1713 void
1714 nfsrv_rcv(so, arg, waitflag)
1715 	struct socket *so;
1716 	caddr_t arg;
1717 	int waitflag;
1718 {
1719 	register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1720 	register struct mbuf *m;
1721 	struct mbuf *mp, *nam;
1722 	struct uio auio;
1723 	int flags, error;
1724 
1725 	if ((slp->ns_flag & SLP_VALID) == 0)
1726 		return;
1727 #ifdef notdef
1728 	/*
1729 	 * Define this to test for nfsds handling this under heavy load.
1730 	 */
1731 	if (waitflag == M_DONTWAIT) {
1732 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1733 	}
1734 #endif
1735 	auio.uio_procp = NULL;
1736 	if (so->so_type == SOCK_STREAM) {
1737 		/*
1738 		 * If there are already records on the queue, defer soreceive()
1739 		 * to an nfsd so that there is feedback to the TCP layer that
1740 		 * the nfs servers are heavily loaded.
1741 		 */
1742 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1743 			slp->ns_flag |= SLP_NEEDQ;
1744 			goto dorecs;
1745 		}
1746 
1747 		/*
1748 		 * Do soreceive().
1749 		 */
1750 		auio.uio_resid = 1000000000;
1751 		flags = MSG_DONTWAIT;
1752 		error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1753 		if (error || mp == (struct mbuf *)0) {
1754 			if (error == EWOULDBLOCK)
1755 				slp->ns_flag |= SLP_NEEDQ;
1756 			else
1757 				slp->ns_flag |= SLP_DISCONN;
1758 			goto dorecs;
1759 		}
1760 		m = mp;
1761 		if (slp->ns_rawend) {
1762 			slp->ns_rawend->m_next = m;
1763 			slp->ns_cc += 1000000000 - auio.uio_resid;
1764 		} else {
1765 			slp->ns_raw = m;
1766 			slp->ns_cc = 1000000000 - auio.uio_resid;
1767 		}
1768 		while (m->m_next)
1769 			m = m->m_next;
1770 		slp->ns_rawend = m;
1771 
1772 		/*
1773 		 * Now try and parse record(s) out of the raw stream data.
1774 		 */
1775 		if (error = nfsrv_getstream(slp, waitflag)) {
1776 			if (error == EPERM)
1777 				slp->ns_flag |= SLP_DISCONN;
1778 			else
1779 				slp->ns_flag |= SLP_NEEDQ;
1780 		}
1781 	} else {
1782 		do {
1783 			auio.uio_resid = 1000000000;
1784 			flags = MSG_DONTWAIT;
1785 			error = soreceive(so, &nam, &auio, &mp,
1786 						(struct mbuf **)0, &flags);
1787 			if (mp) {
1788 				nfs_realign(mp, 10 * NFSX_UNSIGNED);
1789 				if (nam) {
1790 					m = nam;
1791 					m->m_next = mp;
1792 				} else
1793 					m = mp;
1794 				if (slp->ns_recend)
1795 					slp->ns_recend->m_nextpkt = m;
1796 				else
1797 					slp->ns_rec = m;
1798 				slp->ns_recend = m;
1799 				m->m_nextpkt = (struct mbuf *)0;
1800 			}
1801 			if (error) {
1802 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
1803 					&& error != EWOULDBLOCK) {
1804 					slp->ns_flag |= SLP_DISCONN;
1805 					goto dorecs;
1806 				}
1807 			}
1808 		} while (mp);
1809 	}
1810 
1811 	/*
1812 	 * Now try and process the request records, non-blocking.
1813 	 */
1814 dorecs:
1815 	if (waitflag == M_DONTWAIT &&
1816 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
1817 		nfsrv_wakenfsd(slp);
1818 }
1819 
1820 /*
1821  * Try and extract an RPC request from the mbuf data list received on a
1822  * stream socket. The "waitflag" argument indicates whether or not it
1823  * can sleep.
1824  */
1825 nfsrv_getstream(slp, waitflag)
1826 	register struct nfssvc_sock *slp;
1827 	int waitflag;
1828 {
1829 	register struct mbuf *m;
1830 	register char *cp1, *cp2;
1831 	register int len;
1832 	struct mbuf *om, *m2, *recm;
1833 	u_long recmark;
1834 
1835 	if (slp->ns_flag & SLP_GETSTREAM)
1836 		panic("nfs getstream");
1837 	slp->ns_flag |= SLP_GETSTREAM;
1838 	for (;;) {
1839 	    if (slp->ns_reclen == 0) {
1840 		if (slp->ns_cc < NFSX_UNSIGNED) {
1841 			slp->ns_flag &= ~SLP_GETSTREAM;
1842 			return (0);
1843 		}
1844 		m = slp->ns_raw;
1845 		if (m->m_len >= NFSX_UNSIGNED) {
1846 			bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
1847 			m->m_data += NFSX_UNSIGNED;
1848 			m->m_len -= NFSX_UNSIGNED;
1849 		} else {
1850 			cp1 = (caddr_t)&recmark;
1851 			cp2 = mtod(m, caddr_t);
1852 			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
1853 				while (m->m_len == 0) {
1854 					m = m->m_next;
1855 					cp2 = mtod(m, caddr_t);
1856 				}
1857 				*cp1++ = *cp2++;
1858 				m->m_data++;
1859 				m->m_len--;
1860 			}
1861 		}
1862 		slp->ns_cc -= NFSX_UNSIGNED;
1863 		slp->ns_reclen = ntohl(recmark) & ~0x80000000;
1864 		if (slp->ns_reclen < NFS_MINPACKET || slp->ns_reclen > NFS_MAXPACKET) {
1865 			slp->ns_flag &= ~SLP_GETSTREAM;
1866 			return (EPERM);
1867 		}
1868 	    }
1869 
1870 	    /*
1871 	     * Now get the record part.
1872 	     */
1873 	    if (slp->ns_cc == slp->ns_reclen) {
1874 		recm = slp->ns_raw;
1875 		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
1876 		slp->ns_cc = slp->ns_reclen = 0;
1877 	    } else if (slp->ns_cc > slp->ns_reclen) {
1878 		len = 0;
1879 		m = slp->ns_raw;
1880 		om = (struct mbuf *)0;
1881 		while (len < slp->ns_reclen) {
1882 			if ((len + m->m_len) > slp->ns_reclen) {
1883 				m2 = m_copym(m, 0, slp->ns_reclen - len,
1884 					waitflag);
1885 				if (m2) {
1886 					if (om) {
1887 						om->m_next = m2;
1888 						recm = slp->ns_raw;
1889 					} else
1890 						recm = m2;
1891 					m->m_data += slp->ns_reclen - len;
1892 					m->m_len -= slp->ns_reclen - len;
1893 					len = slp->ns_reclen;
1894 				} else {
1895 					slp->ns_flag &= ~SLP_GETSTREAM;
1896 					return (EWOULDBLOCK);
1897 				}
1898 			} else if ((len + m->m_len) == slp->ns_reclen) {
1899 				om = m;
1900 				len += m->m_len;
1901 				m = m->m_next;
1902 				recm = slp->ns_raw;
1903 				om->m_next = (struct mbuf *)0;
1904 			} else {
1905 				om = m;
1906 				len += m->m_len;
1907 				m = m->m_next;
1908 			}
1909 		}
1910 		slp->ns_raw = m;
1911 		slp->ns_cc -= len;
1912 		slp->ns_reclen = 0;
1913 	    } else {
1914 		slp->ns_flag &= ~SLP_GETSTREAM;
1915 		return (0);
1916 	    }
1917 	    nfs_realign(recm, 10 * NFSX_UNSIGNED);
1918 	    if (slp->ns_recend)
1919 		slp->ns_recend->m_nextpkt = recm;
1920 	    else
1921 		slp->ns_rec = recm;
1922 	    slp->ns_recend = recm;
1923 	}
1924 }
1925 
1926 /*
1927  * Parse an RPC header.
1928  */
1929 nfsrv_dorec(slp, nd)
1930 	register struct nfssvc_sock *slp;
1931 	register struct nfsd *nd;
1932 {
1933 	register struct mbuf *m;
1934 	int error;
1935 
1936 	if ((slp->ns_flag & SLP_VALID) == 0 ||
1937 	    (m = slp->ns_rec) == (struct mbuf *)0)
1938 		return (ENOBUFS);
1939 	if (slp->ns_rec = m->m_nextpkt)
1940 		m->m_nextpkt = (struct mbuf *)0;
1941 	else
1942 		slp->ns_recend = (struct mbuf *)0;
1943 	if (m->m_type == MT_SONAME) {
1944 		nd->nd_nam = m;
1945 		nd->nd_md = nd->nd_mrep = m->m_next;
1946 		m->m_next = (struct mbuf *)0;
1947 	} else {
1948 		nd->nd_nam = (struct mbuf *)0;
1949 		nd->nd_md = nd->nd_mrep = m;
1950 	}
1951 	nd->nd_dpos = mtod(nd->nd_md, caddr_t);
1952 	if (error = nfs_getreq(nd, TRUE)) {
1953 		m_freem(nd->nd_nam);
1954 		return (error);
1955 	}
1956 	return (0);
1957 }
1958 
1959 /*
1960  * Search for a sleeping nfsd and wake it up.
1961  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
1962  * running nfsds will go look for the work in the nfssvc_sock list.
1963  */
1964 void
1965 nfsrv_wakenfsd(slp)
1966 	struct nfssvc_sock *slp;
1967 {
1968 	register struct nfsd *nd;
1969 
1970 	if ((slp->ns_flag & SLP_VALID) == 0)
1971 		return;
1972 	for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nd_chain.tqe_next) {
1973 		if (nd->nd_flag & NFSD_WAITING) {
1974 			nd->nd_flag &= ~NFSD_WAITING;
1975 			if (nd->nd_slp)
1976 				panic("nfsd wakeup");
1977 			slp->ns_sref++;
1978 			nd->nd_slp = slp;
1979 			wakeup((caddr_t)nd);
1980 			return;
1981 		}
1982 	}
1983 	slp->ns_flag |= SLP_DOREC;
1984 	nfsd_head_flag |= NFSD_CHECKSLP;
1985 }
1986 #endif /* NFSSERVER */
1987