xref: /openbsd-src/sys/nfs/nfs_socket.c (revision daf88648c0e349d5c02e1504293082072c981640)
1 /*	$OpenBSD: nfs_socket.c,v 1.46 2006/10/28 20:56:46 thib Exp $	*/
2 /*	$NetBSD: nfs_socket.c,v 1.27 1996/04/15 20:20:00 thorpej Exp $	*/
3 
4 /*
5  * Copyright (c) 1989, 1991, 1993, 1995
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * Rick Macklem at The University of Guelph.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)nfs_socket.c	8.5 (Berkeley) 3/30/95
36  */
37 
38 /*
39  * Socket operations for use by nfs
40  */
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/proc.h>
45 #include <sys/mount.h>
46 #include <sys/kernel.h>
47 #include <sys/mbuf.h>
48 #include <sys/vnode.h>
49 #include <sys/domain.h>
50 #include <sys/protosw.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 #include <sys/syslog.h>
54 #include <sys/tprintf.h>
55 #include <sys/namei.h>
56 
57 #include <netinet/in.h>
58 #include <netinet/tcp.h>
59 
60 #include <nfs/rpcv2.h>
61 #include <nfs/nfsproto.h>
62 #include <nfs/nfs.h>
63 #include <nfs/xdr_subs.h>
64 #include <nfs/nfsm_subs.h>
65 #include <nfs/nfsmount.h>
66 #include <nfs/nfsnode.h>
67 #include <nfs/nfsrtt.h>
68 #include <nfs/nfs_var.h>
69 
70 #define	TRUE	1
71 #define	FALSE	0
72 
73 /*
74  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
75  * Use the mean and mean deviation of rtt for the appropriate type of rpc
76  * for the frequent rpcs and a default for the others.
77  * The justification for doing "other" this way is that these rpcs
78  * happen so infrequently that timer est. would probably be stale.
79  * Also, since many of these rpcs are
80  * non-idempotent, a conservative timeout is desired.
81  * getattr, lookup - A+2D
82  * read, write     - A+4D
83  * other           - nm_timeo
84  */
85 #define	NFS_RTO(n, t) \
86 	((t) == 0 ? (n)->nm_timeo : \
87 	 ((t) < 3 ? \
88 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
89 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
90 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
91 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
92 /*
93  * External data, mostly RPC constants in XDR form
94  */
95 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
96 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
97 	rpc_auth_kerb;
98 extern u_int32_t nfs_prog;
99 extern struct nfsstats nfsstats;
100 extern int nfsv3_procid[NFS_NPROCS];
101 extern int nfs_ticks;
102 
103 /*
104  * Defines which timer to use for the procnum.
105  * 0 - default
106  * 1 - getattr
107  * 2 - lookup
108  * 3 - read
109  * 4 - write
110  */
111 static int proct[NFS_NPROCS] = {
112 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
113 	0, 0, 0,
114 };
115 
116 /*
117  * There is a congestion window for outstanding rpcs maintained per mount
118  * point. The cwnd size is adjusted in roughly the way that:
119  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
120  * SIGCOMM '88". ACM, August 1988.
121  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
122  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
123  * of rpcs is in progress.
124  * (The sent count and cwnd are scaled for integer arith.)
125  * Variants of "slow start" were tried and were found to be too much of a
126  * performance hit (ave. rtt 3 times larger),
127  * I suspect due to the large rtt that nfs rpcs have.
128  */
129 #define	NFS_CWNDSCALE	256
130 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
131 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
132 int nfsrtton = 0;
133 struct nfsrtt nfsrtt;
134 
135 void nfs_realign(struct mbuf **, int);
136 void nfs_realign_fixup(struct mbuf *, struct mbuf *, unsigned int *);
137 unsigned int nfs_realign_test = 0;
138 unsigned int nfs_realign_count = 0;
139 
140 struct nfsreqhead nfs_reqq;
141 
142 /*
143  * Initialize sockets and congestion for a new NFS connection.
144  * We do not free the sockaddr if error.
145  */
146 int
147 nfs_connect(nmp, rep)
148 	struct nfsmount *nmp;
149 	struct nfsreq *rep;
150 {
151 	struct socket *so;
152 	int s, error, rcvreserve, sndreserve;
153 	struct sockaddr *saddr;
154 	struct sockaddr_in *sin;
155 	struct mbuf *m;
156 
157 	nmp->nm_so = (struct socket *)0;
158 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
159 	error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
160 		nmp->nm_soproto);
161 	if (error)
162 		goto bad;
163 	so = nmp->nm_so;
164 	nmp->nm_soflags = so->so_proto->pr_flags;
165 
166 	/*
167 	 * Some servers require that the client port be a reserved port number.
168 	 * We always allocate a reserved port, as this prevents filehandle
169 	 * disclosure through UDP port capture.
170 	 */
171 	if (saddr->sa_family == AF_INET) {
172 		struct mbuf *mopt;
173 		int *ip;
174 
175 		MGET(mopt, M_WAIT, MT_SOOPTS);
176 		mopt->m_len = sizeof(int);
177 		ip = mtod(mopt, int *);
178 		*ip = IP_PORTRANGE_LOW;
179 		error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt);
180 		if (error)
181 			goto bad;
182 
183 		MGET(m, M_WAIT, MT_SONAME);
184 		sin = mtod(m, struct sockaddr_in *);
185 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
186 		sin->sin_family = AF_INET;
187 		sin->sin_addr.s_addr = INADDR_ANY;
188 		sin->sin_port = htons(0);
189 		error = sobind(so, m);
190 		m_freem(m);
191 		if (error)
192 			goto bad;
193 
194 		MGET(mopt, M_WAIT, MT_SOOPTS);
195 		mopt->m_len = sizeof(int);
196 		ip = mtod(mopt, int *);
197 		*ip = IP_PORTRANGE_DEFAULT;
198 		error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt);
199 		if (error)
200 			goto bad;
201 	}
202 
203 	/*
204 	 * Protocols that do not require connections may be optionally left
205 	 * unconnected for servers that reply from a port other than NFS_PORT.
206 	 */
207 	if (nmp->nm_flag & NFSMNT_NOCONN) {
208 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
209 			error = ENOTCONN;
210 			goto bad;
211 		}
212 	} else {
213 		error = soconnect(so, nmp->nm_nam);
214 		if (error)
215 			goto bad;
216 
217 		/*
218 		 * Wait for the connection to complete. Cribbed from the
219 		 * connect system call but with the wait timing out so
220 		 * that interruptible mounts don't hang here for a long time.
221 		 */
222 		s = splsoftnet();
223 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
224 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
225 				"nfscon", 2 * hz);
226 			if ((so->so_state & SS_ISCONNECTING) &&
227 			    so->so_error == 0 && rep &&
228 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
229 				so->so_state &= ~SS_ISCONNECTING;
230 				splx(s);
231 				goto bad;
232 			}
233 		}
234 		if (so->so_error) {
235 			error = so->so_error;
236 			so->so_error = 0;
237 			splx(s);
238 			goto bad;
239 		}
240 		splx(s);
241 	}
242 	/*
243 	 * Always set receive timeout to detect server crash and reconnect.
244 	 * Otherwise, we can get stuck in soreceive forever.
245 	 */
246 	so->so_rcv.sb_timeo = (5 * hz);
247 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT))
248 		so->so_snd.sb_timeo = (5 * hz);
249 	else
250 		so->so_snd.sb_timeo = 0;
251 	if (nmp->nm_sotype == SOCK_DGRAM) {
252 		sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR;
253 		rcvreserve = max(nmp->nm_rsize, nmp->nm_readdirsize) +
254 		    NFS_MAXPKTHDR;
255 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
256 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
257 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
258 		    NFS_MAXPKTHDR) * 2;
259 	} else {
260 		if (nmp->nm_sotype != SOCK_STREAM)
261 			panic("nfscon sotype");
262 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
263 			MGET(m, M_WAIT, MT_SOOPTS);
264 			*mtod(m, int32_t *) = 1;
265 			m->m_len = sizeof(int32_t);
266 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
267 		}
268 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
269 			MGET(m, M_WAIT, MT_SOOPTS);
270 			*mtod(m, int32_t *) = 1;
271 			m->m_len = sizeof(int32_t);
272 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
273 		}
274 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
275 		    sizeof (u_int32_t)) * 2;
276 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
277 		    sizeof (u_int32_t)) * 2;
278 	}
279 	error = soreserve(so, sndreserve, rcvreserve);
280 	if (error)
281 		goto bad;
282 	so->so_rcv.sb_flags |= SB_NOINTR;
283 	so->so_snd.sb_flags |= SB_NOINTR;
284 
285 	/* Initialize other non-zero congestion variables */
286 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] =
287 	    nmp->nm_srtt[3] = (NFS_TIMEO << 3);
288 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
289 	    nmp->nm_sdrtt[3] = 0;
290 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
291 	nmp->nm_sent = 0;
292 	nmp->nm_timeouts = 0;
293 	return (0);
294 
295 bad:
296 	nfs_disconnect(nmp);
297 	return (error);
298 }
299 
300 /*
301  * Reconnect routine:
302  * Called when a connection is broken on a reliable protocol.
303  * - clean up the old socket
304  * - nfs_connect() again
305  * - set R_MUSTRESEND for all outstanding requests on mount point
306  * If this fails the mount point is DEAD!
307  * nb: Must be called with the nfs_sndlock() set on the mount point.
308  */
309 int
310 nfs_reconnect(rep)
311 	struct nfsreq *rep;
312 {
313 	struct nfsreq *rp;
314 	struct nfsmount *nmp = rep->r_nmp;
315 	int error;
316 
317 	nfs_disconnect(nmp);
318 	while ((error = nfs_connect(nmp, rep)) != 0) {
319 		if (error == EINTR || error == ERESTART)
320 			return (EINTR);
321 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0);
322 	}
323 
324 	/*
325 	 * Loop through outstanding request list and fix up all requests
326 	 * on old socket.
327 	 */
328 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
329 		if (rp->r_nmp == nmp) {
330 			rp->r_flags |= R_MUSTRESEND;
331 			rp->r_rexmit = 0;
332 		}
333 	}
334 	return (0);
335 }
336 
337 /*
338  * NFS disconnect. Clean up and unlink.
339  */
340 void
341 nfs_disconnect(nmp)
342 	struct nfsmount *nmp;
343 {
344 	struct socket *so;
345 
346 	if (nmp->nm_so) {
347 		so = nmp->nm_so;
348 		nmp->nm_so = (struct socket *)0;
349 		soshutdown(so, 2);
350 		soclose(so);
351 	}
352 }
353 
354 /*
355  * This is the nfs send routine. For connection based socket types, it
356  * must be called with an nfs_sndlock() on the socket.
357  * "rep == NULL" indicates that it has been called from a server.
358  * For the client side:
359  * - return EINTR if the RPC is terminated, 0 otherwise
360  * - set R_MUSTRESEND if the send fails for any reason
361  * - do any cleanup required by recoverable socket errors (???)
362  * For the server side:
363  * - return EINTR or ERESTART if interrupted by a signal
364  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
365  * - do any cleanup required by recoverable socket errors (???)
366  */
367 int
368 nfs_send(so, nam, top, rep)
369 	struct socket *so;
370 	struct mbuf *nam;
371 	struct mbuf *top;
372 	struct nfsreq *rep;
373 {
374 	struct mbuf *sendnam;
375 	int error, soflags, flags;
376 
377 	if (rep) {
378 		if (rep->r_flags & R_SOFTTERM) {
379 			m_freem(top);
380 			return (EINTR);
381 		}
382 		if ((so = rep->r_nmp->nm_so) == NULL) {
383 			rep->r_flags |= R_MUSTRESEND;
384 			m_freem(top);
385 			return (0);
386 		}
387 		rep->r_flags &= ~R_MUSTRESEND;
388 		soflags = rep->r_nmp->nm_soflags;
389 	} else
390 		soflags = so->so_proto->pr_flags;
391 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
392 		sendnam = (struct mbuf *)0;
393 	else
394 		sendnam = nam;
395 	if (so->so_type == SOCK_SEQPACKET)
396 		flags = MSG_EOR;
397 	else
398 		flags = 0;
399 
400 	error = sosend(so, sendnam, (struct uio *)0, top,
401 		(struct mbuf *)0, flags);
402 	if (error) {
403 		if (rep) {
404 			log(LOG_INFO, "nfs send error %d for server %s\n",error,
405 			    rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
406 			/*
407 			 * Deal with errors for the client side.
408 			 */
409 			if (rep->r_flags & R_SOFTTERM)
410 				error = EINTR;
411 			else
412 				rep->r_flags |= R_MUSTRESEND;
413 		} else
414 			log(LOG_INFO, "nfsd send error %d\n", error);
415 
416 		/*
417 		 * Handle any recoverable (soft) socket errors here. (???)
418 		 */
419 		if (error != EINTR && error != ERESTART &&
420 			error != EWOULDBLOCK && error != EPIPE)
421 			error = 0;
422 	}
423 	return (error);
424 }
425 
426 #ifdef NFSCLIENT
427 /*
428  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
429  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
430  * Mark and consolidate the data into a new mbuf list.
431  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
432  *     small mbufs.
433  * For SOCK_STREAM we must be very careful to read an entire record once
434  * we have read any of it, even if the system call has been interrupted.
435  */
436 int
437 nfs_receive(rep, aname, mp)
438 	struct nfsreq *rep;
439 	struct mbuf **aname;
440 	struct mbuf **mp;
441 {
442 	struct socket *so;
443 	struct uio auio;
444 	struct iovec aio;
445 	struct mbuf *m;
446 	struct mbuf *control;
447 	u_int32_t len;
448 	struct mbuf **getnam;
449 	int error, sotype, rcvflg;
450 	struct proc *p = curproc;	/* XXX */
451 
452 	/*
453 	 * Set up arguments for soreceive()
454 	 */
455 	*mp = (struct mbuf *)0;
456 	*aname = (struct mbuf *)0;
457 	sotype = rep->r_nmp->nm_sotype;
458 
459 	/*
460 	 * For reliable protocols, lock against other senders/receivers
461 	 * in case a reconnect is necessary.
462 	 * For SOCK_STREAM, first get the Record Mark to find out how much
463 	 * more there is to get.
464 	 * We must lock the socket against other receivers
465 	 * until we have an entire rpc request/reply.
466 	 */
467 	if (sotype != SOCK_DGRAM) {
468 		error = nfs_sndlock(&rep->r_nmp->nm_flag, rep);
469 		if (error)
470 			return (error);
471 tryagain:
472 		/*
473 		 * Check for fatal errors and resending request.
474 		 */
475 		/*
476 		 * Ugh: If a reconnect attempt just happened, nm_so
477 		 * would have changed. NULL indicates a failed
478 		 * attempt that has essentially shut down this
479 		 * mount point.
480 		 */
481 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
482 			nfs_sndunlock(&rep->r_nmp->nm_flag);
483 			return (EINTR);
484 		}
485 		so = rep->r_nmp->nm_so;
486 		if (!so) {
487 			error = nfs_reconnect(rep);
488 			if (error) {
489 				nfs_sndunlock(&rep->r_nmp->nm_flag);
490 				return (error);
491 			}
492 			goto tryagain;
493 		}
494 		while (rep->r_flags & R_MUSTRESEND) {
495 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
496 			nfsstats.rpcretries++;
497 			rep->r_rtt = 0;
498 			rep->r_flags &= ~R_TIMING;
499 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
500 			if (error) {
501 				if (error == EINTR || error == ERESTART ||
502 				    (error = nfs_reconnect(rep)) != 0) {
503 					nfs_sndunlock(&rep->r_nmp->nm_flag);
504 					return (error);
505 				}
506 				goto tryagain;
507 			}
508 		}
509 		nfs_sndunlock(&rep->r_nmp->nm_flag);
510 		if (sotype == SOCK_STREAM) {
511 			aio.iov_base = (caddr_t) &len;
512 			aio.iov_len = sizeof(u_int32_t);
513 			auio.uio_iov = &aio;
514 			auio.uio_iovcnt = 1;
515 			auio.uio_segflg = UIO_SYSSPACE;
516 			auio.uio_rw = UIO_READ;
517 			auio.uio_offset = 0;
518 			auio.uio_resid = sizeof(u_int32_t);
519 			auio.uio_procp = p;
520 			do {
521 			   rcvflg = MSG_WAITALL;
522 			   error = soreceive(so, (struct mbuf **)0, &auio,
523 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
524 			   if (error == EWOULDBLOCK && rep) {
525 				if (rep->r_flags & R_SOFTTERM)
526 					return (EINTR);
527 				/*
528 				 * looks like the server died after it
529 				 * received the request, make sure
530 				 * that we will retransmit and we
531 				 * don't get stuck here forever.
532 				 */
533 				if (rep->r_rexmit >= rep->r_nmp->nm_retry) {
534 					nfsstats.rpctimeouts++;
535 					error = EPIPE;
536 				}
537 			   }
538 			} while (error == EWOULDBLOCK);
539 			if (!error && auio.uio_resid > 0) {
540 			    log(LOG_INFO,
541 				 "short receive (%d/%d) from nfs server %s\n",
542 				 sizeof(u_int32_t) - auio.uio_resid,
543 				 sizeof(u_int32_t),
544 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
545 			    error = EPIPE;
546 			}
547 			if (error)
548 				goto errout;
549 
550 			len = ntohl(len) & ~0x80000000;
551 			/*
552 			 * This is SERIOUS! We are out of sync with the sender
553 			 * and forcing a disconnect/reconnect is all I can do.
554 			 */
555 			if (len > NFS_MAXPACKET) {
556 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
557 				"impossible packet length",
558 				len,
559 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
560 			    error = EFBIG;
561 			    goto errout;
562 			}
563 			auio.uio_resid = len;
564 			do {
565 			    rcvflg = MSG_WAITALL;
566 			    error =  soreceive(so, (struct mbuf **)0,
567 				&auio, mp, (struct mbuf **)0, &rcvflg);
568 			} while (error == EWOULDBLOCK || error == EINTR ||
569 				 error == ERESTART);
570 			if (!error && auio.uio_resid > 0) {
571 			    log(LOG_INFO,
572 				"short receive (%d/%d) from nfs server %s\n",
573 				len - auio.uio_resid, len,
574 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
575 			    error = EPIPE;
576 			}
577 		} else {
578 			/*
579 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
580 			 * and soreceive() will return when it has either a
581 			 * control msg or a data msg.
582 			 * We have no use for control msg., but must grab them
583 			 * and then throw them away so we know what is going
584 			 * on.
585 			 */
586 			auio.uio_resid = len = 100000000; /* Anything Big */
587 			auio.uio_procp = p;
588 			do {
589 			    rcvflg = 0;
590 			    error =  soreceive(so, (struct mbuf **)0,
591 				&auio, mp, &control, &rcvflg);
592 			    if (control)
593 				m_freem(control);
594 			    if (error == EWOULDBLOCK && rep) {
595 				if (rep->r_flags & R_SOFTTERM)
596 					return (EINTR);
597 			    }
598 			} while (error == EWOULDBLOCK ||
599 				 (!error && *mp == NULL && control));
600 			if ((rcvflg & MSG_EOR) == 0)
601 				printf("Egad!!\n");
602 			if (!error && *mp == NULL)
603 				error = EPIPE;
604 			len -= auio.uio_resid;
605 		}
606 errout:
607 		if (error && error != EINTR && error != ERESTART) {
608 			m_freem(*mp);
609 			*mp = (struct mbuf *)0;
610 			if (error != EPIPE)
611 				log(LOG_INFO,
612 				    "receive error %d from nfs server %s\n",
613 				    error,
614 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
615 			error = nfs_sndlock(&rep->r_nmp->nm_flag, rep);
616 			if (!error) {
617 				error = nfs_reconnect(rep);
618 				if (!error)
619 					goto tryagain;
620 				nfs_sndunlock(&rep->r_nmp->nm_flag);
621 			}
622 		}
623 	} else {
624 		if ((so = rep->r_nmp->nm_so) == NULL)
625 			return (EACCES);
626 		if (so->so_state & SS_ISCONNECTED)
627 			getnam = (struct mbuf **)0;
628 		else
629 			getnam = aname;
630 		auio.uio_resid = len = 1000000;
631 		auio.uio_procp = p;
632 		do {
633 			rcvflg = 0;
634 			error =  soreceive(so, getnam, &auio, mp,
635 				(struct mbuf **)0, &rcvflg);
636 			if (error == EWOULDBLOCK &&
637 			    (rep->r_flags & R_SOFTTERM))
638 				return (EINTR);
639 		} while (error == EWOULDBLOCK);
640 		len -= auio.uio_resid;
641 	}
642 	if (error) {
643 		m_freem(*mp);
644 		*mp = (struct mbuf *)0;
645 	}
646 	/*
647 	 * Search for any mbufs that are not a multiple of 4 bytes long
648 	 * or with m_data not longword aligned.
649 	 * These could cause pointer alignment problems, so copy them to
650 	 * well aligned mbufs.
651 	 */
652 	nfs_realign(mp, 5 * NFSX_UNSIGNED);
653 	return (error);
654 }
655 
656 /*
657  * Implement receipt of reply on a socket.
658  * We must search through the list of received datagrams matching them
659  * with outstanding requests using the xid, until ours is found.
660  */
661 /* ARGSUSED */
662 int
663 nfs_reply(myrep)
664 	struct nfsreq *myrep;
665 {
666 	struct nfsreq *rep;
667 	struct nfsmount *nmp = myrep->r_nmp;
668 	int32_t t1;
669 	struct mbuf *mrep, *nam, *md;
670 	u_int32_t rxid, *tl;
671 	caddr_t dpos, cp2;
672 	int error;
673 
674 	/*
675 	 * Loop around until we get our own reply
676 	 */
677 	for (;;) {
678 		/*
679 		 * Lock against other receivers so that I don't get stuck in
680 		 * sbwait() after someone else has received my reply for me.
681 		 * Also necessary for connection based protocols to avoid
682 		 * race conditions during a reconnect.
683 		 */
684 		error = nfs_rcvlock(myrep);
685 		if (error)
686 			return (error);
687 		/* Already received, bye bye */
688 		if (myrep->r_mrep != NULL) {
689 			nfs_rcvunlock(&nmp->nm_flag);
690 			return (0);
691 		}
692 		/*
693 		 * Get the next Rpc reply off the socket
694 		 */
695 		error = nfs_receive(myrep, &nam, &mrep);
696 		nfs_rcvunlock(&nmp->nm_flag);
697 		if (error) {
698 
699 			/*
700 			 * Ignore routing errors on connectionless protocols??
701 			 */
702 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
703 				if (nmp->nm_so)
704 					nmp->nm_so->so_error = 0;
705 				if (myrep->r_flags & R_GETONEREP)
706 					return (0);
707 				continue;
708 			}
709 			return (error);
710 		}
711 		if (nam)
712 			m_freem(nam);
713 
714 		/*
715 		 * Get the xid and check that it is an rpc reply
716 		 */
717 		md = mrep;
718 		dpos = mtod(md, caddr_t);
719 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
720 		rxid = *tl++;
721 		if (*tl != rpc_reply) {
722 			nfsstats.rpcinvalid++;
723 			m_freem(mrep);
724 nfsmout:
725 			if (myrep->r_flags & R_GETONEREP)
726 				return (0);
727 			continue;
728 		}
729 
730 		/*
731 		 * Loop through the request list to match up the reply
732 		 * Iff no match, just drop the datagram
733 		 */
734 		for (rep = TAILQ_FIRST(&nfs_reqq); rep != NULL;
735 		    rep = TAILQ_NEXT(rep, r_chain)) {
736 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
737 				/* Found it.. */
738 				rep->r_mrep = mrep;
739 				rep->r_md = md;
740 				rep->r_dpos = dpos;
741 				if (nfsrtton) {
742 					struct rttl *rt;
743 
744 					rt = &nfsrtt.rttl[nfsrtt.pos];
745 					rt->proc = rep->r_procnum;
746 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
747 					rt->sent = nmp->nm_sent;
748 					rt->cwnd = nmp->nm_cwnd;
749 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
750 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
751 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
752 					getmicrotime(&rt->tstamp);
753 					if (rep->r_flags & R_TIMING)
754 						rt->rtt = rep->r_rtt;
755 					else
756 						rt->rtt = 1000000;
757 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
758 				}
759 				/*
760 				 * Update congestion window.
761 				 * Do the additive increase of
762 				 * one rpc/rtt.
763 				 */
764 				if (nmp->nm_cwnd <= nmp->nm_sent) {
765 					nmp->nm_cwnd +=
766 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
767 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
768 					if (nmp->nm_cwnd > NFS_MAXCWND)
769 						nmp->nm_cwnd = NFS_MAXCWND;
770 				}
771 				rep->r_flags &= ~R_SENT;
772 				nmp->nm_sent -= NFS_CWNDSCALE;
773 				/*
774 				 * Update rtt using a gain of 0.125 on the mean
775 				 * and a gain of 0.25 on the deviation.
776 				 */
777 				if (rep->r_flags & R_TIMING) {
778 					/*
779 					 * Since the timer resolution of
780 					 * NFS_HZ is so course, it can often
781 					 * result in r_rtt == 0. Since
782 					 * r_rtt == N means that the actual
783 					 * rtt is between N+dt and N+2-dt ticks,
784 					 * add 1.
785 					 */
786 					t1 = rep->r_rtt + 1;
787 					t1 -= (NFS_SRTT(rep) >> 3);
788 					NFS_SRTT(rep) += t1;
789 					if (t1 < 0)
790 						t1 = -t1;
791 					t1 -= (NFS_SDRTT(rep) >> 2);
792 					NFS_SDRTT(rep) += t1;
793 				}
794 				nmp->nm_timeouts = 0;
795 				break;
796 			}
797 		}
798 		/*
799 		 * If not matched to a request, drop it.
800 		 * If it's mine, get out.
801 		 */
802 		if (rep == 0) {
803 			nfsstats.rpcunexpected++;
804 			m_freem(mrep);
805 		} else if (rep == myrep) {
806 			if (rep->r_mrep == NULL)
807 				panic("nfsreply nil");
808 			return (0);
809 		}
810 		if (myrep->r_flags & R_GETONEREP)
811 			return (0);
812 	}
813 }
814 
815 /*
816  * nfs_request - goes something like this
817  *	- fill in request struct
818  *	- links it into list
819  *	- calls nfs_send() for first transmit
820  *	- calls nfs_receive() to get reply
821  *	- break down rpc header and return with nfs reply pointed to
822  *	  by mrep or error
823  * nb: always frees up mreq mbuf list
824  */
825 int
826 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
827 	struct vnode *vp;
828 	struct mbuf *mrest;
829 	int procnum;
830 	struct proc *procp;
831 	struct ucred *cred;
832 	struct mbuf **mrp;
833 	struct mbuf **mdp;
834 	caddr_t *dposp;
835 {
836 	struct mbuf *m, *mrep;
837 	struct nfsreq *rep;
838 	u_int32_t *tl;
839 	int i;
840 	struct nfsmount *nmp;
841 	struct mbuf *md, *mheadend;
842 	char nickv[RPCX_NICKVERF];
843 	time_t reqtime, waituntil;
844 	caddr_t dpos, cp2;
845 	int t1, s, error = 0, mrest_len, auth_len, auth_type;
846 	int trylater_delay, failed_auth = 0;
847 	int verf_len, verf_type;
848 	u_int32_t xid;
849 	char *auth_str, *verf_str;
850 	NFSKERBKEY_T key;		/* save session key */
851 
852 	trylater_delay = NFS_MINTIMEO;
853 
854 	nmp = VFSTONFS(vp->v_mount);
855 	rep = pool_get(&nfsreqpl, PR_WAITOK);
856 	rep->r_nmp = nmp;
857 	rep->r_vp = vp;
858 	rep->r_procp = procp;
859 	rep->r_procnum = procnum;
860 	i = 0;
861 	m = mrest;
862 	while (m) {
863 		i += m->m_len;
864 		m = m->m_next;
865 	}
866 	mrest_len = i;
867 
868 	/*
869 	 * Get the RPC header with authorization.
870 	 */
871 kerbauth:
872 	verf_str = auth_str = (char *)0;
873 	if (nmp->nm_flag & NFSMNT_KERB) {
874 		verf_str = nickv;
875 		verf_len = sizeof (nickv);
876 		auth_type = RPCAUTH_KERB4;
877 		bzero((caddr_t)key, sizeof (key));
878 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
879 			&auth_len, verf_str, verf_len)) {
880 			error = nfs_getauth(nmp, rep, cred, &auth_str,
881 				&auth_len, verf_str, &verf_len, key);
882 			if (error) {
883 				pool_put(&nfsreqpl, rep);
884 				m_freem(mrest);
885 				return (error);
886 			}
887 		}
888 	} else {
889 		auth_type = RPCAUTH_UNIX;
890 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
891 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
892 			5 * NFSX_UNSIGNED;
893 	}
894 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
895 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
896 	if (auth_str)
897 		free(auth_str, M_TEMP);
898 
899 	/*
900 	 * For stream protocols, insert a Sun RPC Record Mark.
901 	 */
902 	if (nmp->nm_sotype == SOCK_STREAM) {
903 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
904 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
905 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
906 	}
907 	rep->r_mreq = m;
908 	rep->r_xid = xid;
909 tryagain:
910 	if (nmp->nm_flag & NFSMNT_SOFT)
911 		rep->r_retry = nmp->nm_retry;
912 	else
913 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
914 	rep->r_rtt = rep->r_rexmit = 0;
915 	if (proct[procnum] > 0)
916 		rep->r_flags = R_TIMING;
917 	else
918 		rep->r_flags = 0;
919 	rep->r_mrep = NULL;
920 
921 	/*
922 	 * Do the client side RPC.
923 	 */
924 	nfsstats.rpcrequests++;
925 	/*
926 	 * Chain request into list of outstanding requests. Be sure
927 	 * to put it LAST so timer finds oldest requests first.
928 	 */
929 	s = splsoftnet();
930 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
931 
932 	/* Get send time for nqnfs */
933 	reqtime = time_second;
934 
935 	/*
936 	 * If backing off another request or avoiding congestion, don't
937 	 * send this one now but let timer do it. If not timing a request,
938 	 * do it now.
939 	 */
940 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
941 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
942 		nmp->nm_sent < nmp->nm_cwnd)) {
943 		splx(s);
944 		if (nmp->nm_soflags & PR_CONNREQUIRED)
945 			error = nfs_sndlock(&nmp->nm_flag, rep);
946 		if (!error) {
947 			error = nfs_send(nmp->nm_so, nmp->nm_nam,
948 					m_copym(m, 0, M_COPYALL, M_WAIT),
949 					rep);
950 			if (nmp->nm_soflags & PR_CONNREQUIRED)
951 				nfs_sndunlock(&nmp->nm_flag);
952 		}
953 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
954 			nmp->nm_sent += NFS_CWNDSCALE;
955 			rep->r_flags |= R_SENT;
956 		}
957 	} else {
958 		splx(s);
959 		rep->r_rtt = -1;
960 	}
961 
962 	/*
963 	 * Wait for the reply from our send or the timer's.
964 	 */
965 	if (!error || error == EPIPE)
966 		error = nfs_reply(rep);
967 
968 	/*
969 	 * RPC done, unlink the request.
970 	 */
971 	s = splsoftnet();
972 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
973 	splx(s);
974 
975 	/*
976 	 * Decrement the outstanding request count.
977 	 */
978 	if (rep->r_flags & R_SENT) {
979 		rep->r_flags &= ~R_SENT;	/* paranoia */
980 		nmp->nm_sent -= NFS_CWNDSCALE;
981 	}
982 
983 	/*
984 	 * If there was a successful reply and a tprintf msg.
985 	 * tprintf a response.
986 	 */
987 	if (!error && (rep->r_flags & R_TPRINTFMSG))
988 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
989 		    "is alive again");
990 	mrep = rep->r_mrep;
991 	md = rep->r_md;
992 	dpos = rep->r_dpos;
993 	if (error) {
994 		m_freem(rep->r_mreq);
995 		pool_put(&nfsreqpl, rep);
996 		return (error);
997 	}
998 
999 	/*
1000 	 * break down the rpc header and check if ok
1001 	 */
1002 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1003 	if (*tl++ == rpc_msgdenied) {
1004 		if (*tl == rpc_mismatch)
1005 			error = EOPNOTSUPP;
1006 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1007 			if (!failed_auth) {
1008 				failed_auth++;
1009 				mheadend->m_next = (struct mbuf *)0;
1010 				m_freem(mrep);
1011 				m_freem(rep->r_mreq);
1012 				goto kerbauth;
1013 			} else
1014 				error = EAUTH;
1015 		} else
1016 			error = EACCES;
1017 		m_freem(mrep);
1018 		m_freem(rep->r_mreq);
1019 		pool_put(&nfsreqpl, rep);
1020 		return (error);
1021 	}
1022 
1023 	/*
1024 	 * Grab any Kerberos verifier, otherwise just throw it away.
1025 	 */
1026 	verf_type = fxdr_unsigned(int, *tl++);
1027 	i = fxdr_unsigned(int32_t, *tl);
1028 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1029 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1030 		if (error)
1031 			goto nfsmout;
1032 	} else if (i > 0)
1033 		nfsm_adv(nfsm_rndup(i));
1034 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1035 	/* 0 == ok */
1036 	if (*tl == 0) {
1037 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1038 		if (*tl != 0) {
1039 			error = fxdr_unsigned(int, *tl);
1040 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1041 				error == NFSERR_TRYLATER) {
1042 				m_freem(mrep);
1043 				error = 0;
1044 				waituntil = time_second + trylater_delay;
1045 				while (time_second < waituntil)
1046 					(void) tsleep((caddr_t)&lbolt,
1047 						PSOCK, "nqnfstry", 0);
1048 				trylater_delay *= NFS_TIMEOUTMUL;
1049 				if (trylater_delay > NFS_MAXTIMEO)
1050 					trylater_delay = NFS_MAXTIMEO;
1051 
1052 				goto tryagain;
1053 			}
1054 
1055 			/*
1056 			 * If the File Handle was stale, invalidate the
1057 			 * lookup cache, just in case.
1058 			 */
1059 			if (error == ESTALE)
1060 				cache_purge(vp);
1061 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1062 				*mrp = mrep;
1063 				*mdp = md;
1064 				*dposp = dpos;
1065 				error |= NFSERR_RETERR;
1066 			} else
1067 				m_freem(mrep);
1068 			m_freem(rep->r_mreq);
1069 			pool_put(&nfsreqpl, rep);
1070 			return (error);
1071 		}
1072 
1073 		*mrp = mrep;
1074 		*mdp = md;
1075 		*dposp = dpos;
1076 		m_freem(rep->r_mreq);
1077 		pool_put(&nfsreqpl, rep);
1078 		return (0);
1079 	}
1080 	m_freem(mrep);
1081 	error = EPROTONOSUPPORT;
1082 nfsmout:
1083 	m_freem(rep->r_mreq);
1084 	pool_put(&nfsreqpl, rep);
1085 	return (error);
1086 }
1087 #endif /* NFSCLIENT */
1088 
1089 /*
1090  * Generate the rpc reply header
1091  * siz arg. is used to decide if adding a cluster is worthwhile
1092  */
1093 int
1094 nfs_rephead(siz, nd, slp, err, frev, mrq, mbp, bposp)
1095 	int siz;
1096 	struct nfsrv_descript *nd;
1097 	struct nfssvc_sock *slp;
1098 	int err;
1099 	u_quad_t *frev;
1100 	struct mbuf **mrq;
1101 	struct mbuf **mbp;
1102 	caddr_t *bposp;
1103 {
1104 	u_int32_t *tl;
1105 	struct mbuf *mreq;
1106 	caddr_t bpos;
1107 	struct mbuf *mb, *mb2;
1108 
1109 	MGETHDR(mreq, M_WAIT, MT_DATA);
1110 	mb = mreq;
1111 	/*
1112 	 * If this is a big reply, use a cluster else
1113 	 * try and leave leading space for the lower level headers.
1114 	 */
1115 	siz += RPC_REPLYSIZ;
1116 	if (siz >= max_datalen) {
1117 		MCLGET(mreq, M_WAIT);
1118 	} else
1119 		mreq->m_data += max_hdr;
1120 	tl = mtod(mreq, u_int32_t *);
1121 	mreq->m_len = 6 * NFSX_UNSIGNED;
1122 	bpos = ((caddr_t)tl) + mreq->m_len;
1123 	*tl++ = txdr_unsigned(nd->nd_retxid);
1124 	*tl++ = rpc_reply;
1125 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1126 		*tl++ = rpc_msgdenied;
1127 		if (err & NFSERR_AUTHERR) {
1128 			*tl++ = rpc_autherr;
1129 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1130 			mreq->m_len -= NFSX_UNSIGNED;
1131 			bpos -= NFSX_UNSIGNED;
1132 		} else {
1133 			*tl++ = rpc_mismatch;
1134 			*tl++ = txdr_unsigned(RPC_VER2);
1135 			*tl = txdr_unsigned(RPC_VER2);
1136 		}
1137 	} else {
1138 		*tl++ = rpc_msgaccepted;
1139 
1140 		/*
1141 		 * For Kerberos authentication, we must send the nickname
1142 		 * verifier back, otherwise just RPCAUTH_NULL.
1143 		 */
1144 		if (nd->nd_flag & ND_KERBFULL) {
1145 		    struct nfsuid *nuidp;
1146 		    struct timeval ktvin, ktvout;
1147 
1148 		    LIST_FOREACH(nuidp, NUIDHASH(slp, nd->nd_cr.cr_uid),
1149 		     nu_hash) {
1150 			if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1151 			    (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1152 			     &nuidp->nu_haddr, nd->nd_nam2)))
1153 			    break;
1154 		    }
1155 		    if (nuidp) {
1156 			ktvin.tv_sec =
1157 			    txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1158 			ktvin.tv_usec =
1159 			    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1160 
1161 			*tl++ = rpc_auth_kerb;
1162 			*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1163 			*tl = ktvout.tv_sec;
1164 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1165 			*tl++ = ktvout.tv_usec;
1166 			*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1167 		    } else {
1168 			*tl++ = 0;
1169 			*tl++ = 0;
1170 		    }
1171 		} else {
1172 			*tl++ = 0;
1173 			*tl++ = 0;
1174 		}
1175 		switch (err) {
1176 		case EPROGUNAVAIL:
1177 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1178 			break;
1179 		case EPROGMISMATCH:
1180 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1181 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1182 			*tl++ = txdr_unsigned(2);
1183 			*tl = txdr_unsigned(3);
1184 			break;
1185 		case EPROCUNAVAIL:
1186 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1187 			break;
1188 		case EBADRPC:
1189 			*tl = txdr_unsigned(RPC_GARBAGE);
1190 			break;
1191 		default:
1192 			*tl = 0;
1193 			if (err != NFSERR_RETVOID) {
1194 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1195 				if (err)
1196 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1197 				else
1198 				    *tl = 0;
1199 			}
1200 			break;
1201 		};
1202 	}
1203 
1204 	*mrq = mreq;
1205 	if (mbp != NULL)
1206 		*mbp = mb;
1207 	*bposp = bpos;
1208 	if (err != 0 && err != NFSERR_RETVOID)
1209 		nfsstats.srvrpc_errs++;
1210 	return (0);
1211 }
1212 
1213 /*
1214  * Nfs timer routine
1215  * Scan the nfsreq list and retranmit any requests that have timed out
1216  * To avoid retransmission attempts on STREAM sockets (in the future) make
1217  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1218  */
1219 void
1220 nfs_timer(arg)
1221 	void *arg;
1222 {
1223 	struct timeout *to = (struct timeout *)arg;
1224 	struct nfsreq *rep;
1225 	struct mbuf *m;
1226 	struct socket *so;
1227 	struct nfsmount *nmp;
1228 	int timeo;
1229 	int s, error;
1230 #ifdef NFSSERVER
1231 	struct nfssvc_sock *slp;
1232 	struct timeval tv;
1233 	u_quad_t cur_usec;
1234 #endif
1235 
1236 	s = splsoftnet();
1237 	for (rep = TAILQ_FIRST(&nfs_reqq); rep != NULL;
1238 	    rep = TAILQ_NEXT(rep, r_chain)) {
1239 		nmp = rep->r_nmp;
1240 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1241 			continue;
1242 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1243 			rep->r_flags |= R_SOFTTERM;
1244 			continue;
1245 		}
1246 		if (rep->r_rtt >= 0) {
1247 			rep->r_rtt++;
1248 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1249 				timeo = nmp->nm_timeo;
1250 			else
1251 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1252 			if (nmp->nm_timeouts > 0)
1253 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1254 			if (rep->r_rtt <= timeo)
1255 				continue;
1256 			if (nmp->nm_timeouts < 8)
1257 				nmp->nm_timeouts++;
1258 		}
1259 		/*
1260 		 * Check for server not responding
1261 		 */
1262 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1263 		     rep->r_rexmit > nmp->nm_deadthresh) {
1264 			nfs_msg(rep->r_procp,
1265 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1266 			    "not responding");
1267 			rep->r_flags |= R_TPRINTFMSG;
1268 		}
1269 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1270 			nfsstats.rpctimeouts++;
1271 			rep->r_flags |= R_SOFTTERM;
1272 			continue;
1273 		}
1274 		if (nmp->nm_sotype != SOCK_DGRAM) {
1275 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1276 				rep->r_rexmit = NFS_MAXREXMIT;
1277 			continue;
1278 		}
1279 		if ((so = nmp->nm_so) == NULL)
1280 			continue;
1281 
1282 		/*
1283 		 * If there is enough space and the window allows..
1284 		 *	Resend it
1285 		 * Set r_rtt to -1 in case we fail to send it now.
1286 		 */
1287 		rep->r_rtt = -1;
1288 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1289 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1290 		    (rep->r_flags & R_SENT) ||
1291 		    nmp->nm_sent < nmp->nm_cwnd) &&
1292 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1293 			if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
1294 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1295 			    (struct mbuf *)0, (struct mbuf *)0);
1296 			else
1297 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1298 			    nmp->nm_nam, (struct mbuf *)0);
1299 			if (error) {
1300 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
1301 					so->so_error = 0;
1302 			} else {
1303 				/*
1304 				 * Iff first send, start timing
1305 				 * else turn timing off, backoff timer
1306 				 * and divide congestion window by 2.
1307 				 */
1308 				if (rep->r_flags & R_SENT) {
1309 					rep->r_flags &= ~R_TIMING;
1310 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1311 						rep->r_rexmit = NFS_MAXREXMIT;
1312 					nmp->nm_cwnd >>= 1;
1313 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1314 						nmp->nm_cwnd = NFS_CWNDSCALE;
1315 					nfsstats.rpcretries++;
1316 				} else {
1317 					rep->r_flags |= R_SENT;
1318 					nmp->nm_sent += NFS_CWNDSCALE;
1319 				}
1320 				rep->r_rtt = 0;
1321 			}
1322 		}
1323 	}
1324 
1325 #ifdef NFSSERVER
1326 	/*
1327 	 * Scan the write gathering queues for writes that need to be
1328 	 * completed now.
1329 	 */
1330 	getmicrotime(&tv);
1331 	cur_usec = (u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec;
1332 	for (slp = TAILQ_FIRST(&nfssvc_sockhead); slp != NULL;
1333 	    slp = TAILQ_NEXT(slp, ns_chain)) {
1334 	    if (LIST_FIRST(&slp->ns_tq) &&
1335 	        LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec)
1336 		nfsrv_wakenfsd(slp);
1337 	}
1338 #endif /* NFSSERVER */
1339 	splx(s);
1340 	timeout_add(to, nfs_ticks);
1341 }
1342 
1343 /*
1344  * Test for a termination condition pending on the process.
1345  * This is used for NFSMNT_INT mounts.
1346  */
1347 int
1348 nfs_sigintr(nmp, rep, p)
1349 	struct nfsmount *nmp;
1350 	struct nfsreq *rep;
1351 	struct proc *p;
1352 {
1353 
1354 	if (rep && (rep->r_flags & R_SOFTTERM))
1355 		return (EINTR);
1356 	if (!(nmp->nm_flag & NFSMNT_INT))
1357 		return (0);
1358 	if (p && p->p_siglist &&
1359 	    (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) &
1360 	    NFSINT_SIGMASK))
1361 		return (EINTR);
1362 	return (0);
1363 }
1364 
1365 /*
1366  * Lock a socket against others.
1367  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1368  * and also to avoid race conditions between the processes with nfs requests
1369  * in progress when a reconnect is necessary.
1370  */
1371 int
1372 nfs_sndlock(flagp, rep)
1373 	int *flagp;
1374 	struct nfsreq *rep;
1375 {
1376 	struct proc *p;
1377 	int slpflag = 0, slptimeo = 0;
1378 
1379 	if (rep) {
1380 		p = rep->r_procp;
1381 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1382 			slpflag = PCATCH;
1383 	} else
1384 		p = (struct proc *)0;
1385 	while (*flagp & NFSMNT_SNDLOCK) {
1386 		if (nfs_sigintr(rep->r_nmp, rep, p))
1387 			return (EINTR);
1388 		*flagp |= NFSMNT_WANTSND;
1389 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1390 			slptimeo);
1391 		if (slpflag == PCATCH) {
1392 			slpflag = 0;
1393 			slptimeo = 2 * hz;
1394 		}
1395 	}
1396 	*flagp |= NFSMNT_SNDLOCK;
1397 	return (0);
1398 }
1399 
1400 /*
1401  * Unlock the stream socket for others.
1402  */
1403 void
1404 nfs_sndunlock(flagp)
1405 	int *flagp;
1406 {
1407 
1408 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1409 		panic("nfs sndunlock");
1410 	*flagp &= ~NFSMNT_SNDLOCK;
1411 	if (*flagp & NFSMNT_WANTSND) {
1412 		*flagp &= ~NFSMNT_WANTSND;
1413 		wakeup((caddr_t)flagp);
1414 	}
1415 }
1416 
1417 int
1418 nfs_rcvlock(rep)
1419 	struct nfsreq *rep;
1420 {
1421 	int *flagp = &rep->r_nmp->nm_flag;
1422 	int slpflag, slptimeo = 0;
1423 
1424 	if (*flagp & NFSMNT_INT)
1425 		slpflag = PCATCH;
1426 	else
1427 		slpflag = 0;
1428 	while (*flagp & NFSMNT_RCVLOCK) {
1429 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1430 			return (EINTR);
1431 		*flagp |= NFSMNT_WANTRCV;
1432 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1433 			slptimeo);
1434 		if (slpflag == PCATCH) {
1435 			slpflag = 0;
1436 			slptimeo = 2 * hz;
1437 		}
1438 	}
1439 	*flagp |= NFSMNT_RCVLOCK;
1440 	return (0);
1441 }
1442 
1443 /*
1444  * Unlock the stream socket for others.
1445  */
1446 void
1447 nfs_rcvunlock(flagp)
1448 	int *flagp;
1449 {
1450 
1451 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1452 		panic("nfs rcvunlock");
1453 	*flagp &= ~NFSMNT_RCVLOCK;
1454 	if (*flagp & NFSMNT_WANTRCV) {
1455 		*flagp &= ~NFSMNT_WANTRCV;
1456 		wakeup((caddr_t)flagp);
1457 	}
1458 }
1459 
1460 /*
1461  * Auxiliary routine to align the length of mbuf copies made with m_copyback().
1462  */
1463 void
1464 nfs_realign_fixup(struct mbuf *m, struct mbuf *n, unsigned int *off)
1465 {
1466 	size_t padding;
1467 
1468 	/*
1469 	 * The maximum number of bytes that m_copyback() places in a mbuf is
1470 	 * always an aligned quantity, so realign happens at the chain's tail.
1471 	 */
1472 	while (n->m_next != NULL)
1473 		n = n->m_next;
1474 
1475 	/*
1476 	 * Pad from the next elements in the source chain. Loop until the
1477 	 * destination chain is aligned, or the end of the source is reached.
1478 	 */
1479 	do {
1480 		m = m->m_next;
1481 		if (m == NULL)
1482 			return;
1483 
1484 		padding = min(ALIGN(n->m_len) - n->m_len, m->m_len);
1485 		if (padding > M_TRAILINGSPACE(n))
1486 			panic("nfs_realign_fixup: no memory to pad to");
1487 
1488 		bcopy(mtod(m, void *), mtod(n, char *) + n->m_len, padding);
1489 
1490 		n->m_len += padding;
1491 		m_adj(m, padding);
1492 		*off += padding;
1493 
1494 	} while (!ALIGNED_POINTER(n->m_len, void *));
1495 }
1496 
1497 /*
1498  * The NFS RPC parsing code uses the data address and the length of mbuf
1499  * structures to calculate on-memory addresses. This function makes sure these
1500  * parameters are correctly aligned.
1501  */
1502 void
1503 nfs_realign(struct mbuf **pm, int hsiz)
1504 {
1505 	struct mbuf *m;
1506 	struct mbuf *n = NULL;
1507 	unsigned int off = 0;
1508 
1509 	++nfs_realign_test;
1510 	while ((m = *pm) != NULL) {
1511 		if (!ALIGNED_POINTER(m->m_data, void *) ||
1512 		    !ALIGNED_POINTER(m->m_len,  void *)) {
1513 			MGET(n, M_WAIT, MT_DATA);
1514 			if (ALIGN(m->m_len) >= MINCLSIZE) {
1515 				MCLGET(n, M_WAIT);
1516 			}
1517 			n->m_len = 0;
1518 			break;
1519 		}
1520 		pm = &m->m_next;
1521 	}
1522 	/*
1523 	 * If n is non-NULL, loop on m copying data, then replace the
1524 	 * portion of the chain that had to be realigned.
1525 	 */
1526 	if (n != NULL) {
1527 		++nfs_realign_count;
1528 		while (m) {
1529 			m_copyback(n, off, m->m_len, mtod(m, caddr_t));
1530 
1531 			/*
1532 			 * If an unaligned amount of memory was copied, fix up
1533 			 * the last mbuf created by m_copyback().
1534 			 */
1535 			if (!ALIGNED_POINTER(m->m_len, void *))
1536 				nfs_realign_fixup(m, n, &off);
1537 
1538 			off += m->m_len;
1539 			m = m->m_next;
1540 		}
1541 		m_freem(*pm);
1542 		*pm = n;
1543 	}
1544 }
1545 
1546 
1547 /*
1548  * Parse an RPC request
1549  * - verify it
1550  * - fill in the cred struct.
1551  */
1552 int
1553 nfs_getreq(nd, nfsd, has_header)
1554 	struct nfsrv_descript *nd;
1555 	struct nfsd *nfsd;
1556 	int has_header;
1557 {
1558 	int len, i;
1559 	u_int32_t *tl;
1560 	int32_t t1;
1561 	struct uio uio;
1562 	struct iovec iov;
1563 	caddr_t dpos, cp2, cp;
1564 	u_int32_t nfsvers, auth_type;
1565 	uid_t nickuid;
1566 	int error = 0, ticklen;
1567 	struct mbuf *mrep, *md;
1568 	struct nfsuid *nuidp;
1569 	struct timeval tvin, tvout;
1570 
1571 	mrep = nd->nd_mrep;
1572 	md = nd->nd_md;
1573 	dpos = nd->nd_dpos;
1574 	if (has_header) {
1575 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1576 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1577 		if (*tl++ != rpc_call) {
1578 			m_freem(mrep);
1579 			return (EBADRPC);
1580 		}
1581 	} else
1582 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1583 	nd->nd_repstat = 0;
1584 	nd->nd_flag = 0;
1585 	if (*tl++ != rpc_vers) {
1586 		nd->nd_repstat = ERPCMISMATCH;
1587 		nd->nd_procnum = NFSPROC_NOOP;
1588 		return (0);
1589 	}
1590 	if (*tl != nfs_prog) {
1591 		nd->nd_repstat = EPROGUNAVAIL;
1592 		nd->nd_procnum = NFSPROC_NOOP;
1593 		return (0);
1594 	}
1595 	tl++;
1596 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1597 	if (nfsvers != NFS_VER2 && nfsvers != NFS_VER3) {
1598 		nd->nd_repstat = EPROGMISMATCH;
1599 		nd->nd_procnum = NFSPROC_NOOP;
1600 		return (0);
1601 	}
1602 	if (nfsvers == NFS_VER3)
1603 		nd->nd_flag = ND_NFSV3;
1604 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1605 	if (nd->nd_procnum == NFSPROC_NULL)
1606 		return (0);
1607 	if (nd->nd_procnum >= NFS_NPROCS ||
1608 		(nd->nd_procnum > NFSPROC_COMMIT) ||
1609 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1610 		nd->nd_repstat = EPROCUNAVAIL;
1611 		nd->nd_procnum = NFSPROC_NOOP;
1612 		return (0);
1613 	}
1614 	if ((nd->nd_flag & ND_NFSV3) == 0)
1615 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1616 	auth_type = *tl++;
1617 	len = fxdr_unsigned(int, *tl++);
1618 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1619 		m_freem(mrep);
1620 		return (EBADRPC);
1621 	}
1622 
1623 	nd->nd_flag &= ~ND_KERBAUTH;
1624 	/*
1625 	 * Handle auth_unix or auth_kerb.
1626 	 */
1627 	if (auth_type == rpc_auth_unix) {
1628 		len = fxdr_unsigned(int, *++tl);
1629 		if (len < 0 || len > NFS_MAXNAMLEN) {
1630 			m_freem(mrep);
1631 			return (EBADRPC);
1632 		}
1633 		nfsm_adv(nfsm_rndup(len));
1634 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1635 		bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred));
1636 		nd->nd_cr.cr_ref = 1;
1637 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1638 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1639 		len = fxdr_unsigned(int, *tl);
1640 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1641 			m_freem(mrep);
1642 			return (EBADRPC);
1643 		}
1644 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1645 		for (i = 0; i < len; i++)
1646 		    if (i < NGROUPS)
1647 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1648 		    else
1649 			tl++;
1650 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1651 		if (nd->nd_cr.cr_ngroups > 1)
1652 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1653 		len = fxdr_unsigned(int, *++tl);
1654 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1655 			m_freem(mrep);
1656 			return (EBADRPC);
1657 		}
1658 		if (len > 0)
1659 			nfsm_adv(nfsm_rndup(len));
1660 	} else if (auth_type == rpc_auth_kerb) {
1661 		switch (fxdr_unsigned(int, *tl++)) {
1662 		case RPCAKN_FULLNAME:
1663 			ticklen = fxdr_unsigned(int, *tl);
1664 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1665 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1666 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1667 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1668 				m_freem(mrep);
1669 				return (EBADRPC);
1670 			}
1671 			uio.uio_offset = 0;
1672 			uio.uio_iov = &iov;
1673 			uio.uio_iovcnt = 1;
1674 			uio.uio_segflg = UIO_SYSSPACE;
1675 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1676 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1677 			nfsm_mtouio(&uio, uio.uio_resid);
1678 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1679 			if (*tl++ != rpc_auth_kerb ||
1680 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1681 				printf("Bad kerb verifier\n");
1682 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1683 				nd->nd_procnum = NFSPROC_NOOP;
1684 				return (0);
1685 			}
1686 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1687 			tl = (u_int32_t *)cp;
1688 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1689 				printf("Not fullname kerb verifier\n");
1690 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1691 				nd->nd_procnum = NFSPROC_NOOP;
1692 				return (0);
1693 			}
1694 			cp += NFSX_UNSIGNED;
1695 			bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED);
1696 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1697 			nd->nd_flag |= ND_KERBFULL;
1698 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1699 			break;
1700 		case RPCAKN_NICKNAME:
1701 			if (len != 2 * NFSX_UNSIGNED) {
1702 				printf("Kerb nickname short\n");
1703 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1704 				nd->nd_procnum = NFSPROC_NOOP;
1705 				return (0);
1706 			}
1707 			nickuid = fxdr_unsigned(uid_t, *tl);
1708 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1709 			if (*tl++ != rpc_auth_kerb ||
1710 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1711 				printf("Kerb nick verifier bad\n");
1712 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1713 				nd->nd_procnum = NFSPROC_NOOP;
1714 				return (0);
1715 			}
1716 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1717 			tvin.tv_sec = *tl++;
1718 			tvin.tv_usec = *tl;
1719 
1720 			LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid),
1721 			    nu_hash) {
1722 				if (nuidp->nu_cr.cr_uid == nickuid &&
1723 				    (!nd->nd_nam2 ||
1724 				     netaddr_match(NU_NETFAM(nuidp),
1725 				      &nuidp->nu_haddr, nd->nd_nam2)))
1726 					break;
1727 			}
1728 			if (!nuidp) {
1729 				nd->nd_repstat =
1730 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1731 				nd->nd_procnum = NFSPROC_NOOP;
1732 				return (0);
1733 			}
1734 
1735 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1736 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1737 			if (nuidp->nu_expire < time_second ||
1738 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1739 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1740 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1741 				nuidp->nu_expire = 0;
1742 				nd->nd_repstat =
1743 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1744 				nd->nd_procnum = NFSPROC_NOOP;
1745 				return (0);
1746 			}
1747 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1748 			nd->nd_flag |= ND_KERBNICK;
1749 		};
1750 	} else {
1751 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1752 		nd->nd_procnum = NFSPROC_NOOP;
1753 		return (0);
1754 	}
1755 
1756 	nd->nd_md = md;
1757 	nd->nd_dpos = dpos;
1758 	return (0);
1759 nfsmout:
1760 	return (error);
1761 }
1762 
1763 int
1764 nfs_msg(p, server, msg)
1765 	struct proc *p;
1766 	char *server, *msg;
1767 {
1768 	tpr_t tpr;
1769 
1770 	if (p)
1771 		tpr = tprintf_open(p);
1772 	else
1773 		tpr = NULL;
1774 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1775 	tprintf_close(tpr);
1776 	return (0);
1777 }
1778 
1779 #ifdef NFSSERVER
1780 int (*nfsrv3_procs[NFS_NPROCS])(struct nfsrv_descript *,
1781 				    struct nfssvc_sock *, struct proc *,
1782 				    struct mbuf **) = {
1783 	nfsrv_null,
1784 	nfsrv_getattr,
1785 	nfsrv_setattr,
1786 	nfsrv_lookup,
1787 	nfsrv3_access,
1788 	nfsrv_readlink,
1789 	nfsrv_read,
1790 	nfsrv_write,
1791 	nfsrv_create,
1792 	nfsrv_mkdir,
1793 	nfsrv_symlink,
1794 	nfsrv_mknod,
1795 	nfsrv_remove,
1796 	nfsrv_rmdir,
1797 	nfsrv_rename,
1798 	nfsrv_link,
1799 	nfsrv_readdir,
1800 	nfsrv_readdirplus,
1801 	nfsrv_statfs,
1802 	nfsrv_fsinfo,
1803 	nfsrv_pathconf,
1804 	nfsrv_commit,
1805 	nfsrv_noop,
1806 	nfsrv_noop,
1807 	nfsrv_noop,
1808 	nfsrv_noop
1809 };
1810 
1811 /*
1812  * Socket upcall routine for the nfsd sockets.
1813  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1814  * Essentially do as much as possible non-blocking, else punt and it will
1815  * be called with M_WAIT from an nfsd.
1816  */
1817 void
1818 nfsrv_rcv(so, arg, waitflag)
1819 	struct socket *so;
1820 	caddr_t arg;
1821 	int waitflag;
1822 {
1823 	struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1824 	struct mbuf *m;
1825 	struct mbuf *mp, *nam;
1826 	struct uio auio;
1827 	int flags, error;
1828 
1829 	if ((slp->ns_flag & SLP_VALID) == 0)
1830 		return;
1831 #ifdef notdef
1832 	/*
1833 	 * Define this to test for nfsds handling this under heavy load.
1834 	 */
1835 	if (waitflag == M_DONTWAIT) {
1836 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1837 	}
1838 #endif
1839 	auio.uio_procp = NULL;
1840 	if (so->so_type == SOCK_STREAM) {
1841 		/*
1842 		 * If there are already records on the queue, defer soreceive()
1843 		 * to an nfsd so that there is feedback to the TCP layer that
1844 		 * the nfs servers are heavily loaded.
1845 		 */
1846 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1847 			slp->ns_flag |= SLP_NEEDQ;
1848 			goto dorecs;
1849 		}
1850 
1851 		/*
1852 		 * Do soreceive().
1853 		 */
1854 		auio.uio_resid = 1000000000;
1855 		flags = MSG_DONTWAIT;
1856 		error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1857 		if (error || mp == (struct mbuf *)0) {
1858 			if (error == EWOULDBLOCK)
1859 				slp->ns_flag |= SLP_NEEDQ;
1860 			else
1861 				slp->ns_flag |= SLP_DISCONN;
1862 			goto dorecs;
1863 		}
1864 		m = mp;
1865 		if (slp->ns_rawend) {
1866 			slp->ns_rawend->m_next = m;
1867 			slp->ns_cc += 1000000000 - auio.uio_resid;
1868 		} else {
1869 			slp->ns_raw = m;
1870 			slp->ns_cc = 1000000000 - auio.uio_resid;
1871 		}
1872 		while (m->m_next)
1873 			m = m->m_next;
1874 		slp->ns_rawend = m;
1875 
1876 		/*
1877 		 * Now try and parse record(s) out of the raw stream data.
1878 		 */
1879 		error = nfsrv_getstream(slp, waitflag);
1880 		if (error) {
1881 			if (error == EPERM)
1882 				slp->ns_flag |= SLP_DISCONN;
1883 			else
1884 				slp->ns_flag |= SLP_NEEDQ;
1885 		}
1886 	} else {
1887 		do {
1888 			auio.uio_resid = 1000000000;
1889 			flags = MSG_DONTWAIT;
1890 			error = soreceive(so, &nam, &auio, &mp,
1891 						(struct mbuf **)0, &flags);
1892 			if (mp) {
1893 				if (nam) {
1894 					m = nam;
1895 					m->m_next = mp;
1896 				} else
1897 					m = mp;
1898 				if (slp->ns_recend)
1899 					slp->ns_recend->m_nextpkt = m;
1900 				else
1901 					slp->ns_rec = m;
1902 				slp->ns_recend = m;
1903 				m->m_nextpkt = (struct mbuf *)0;
1904 			}
1905 			if (error) {
1906 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
1907 					&& error != EWOULDBLOCK) {
1908 					slp->ns_flag |= SLP_DISCONN;
1909 					goto dorecs;
1910 				}
1911 			}
1912 		} while (mp);
1913 	}
1914 
1915 	/*
1916 	 * Now try and process the request records, non-blocking.
1917 	 */
1918 dorecs:
1919 	if (waitflag == M_DONTWAIT &&
1920 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
1921 		nfsrv_wakenfsd(slp);
1922 }
1923 
1924 /*
1925  * Try and extract an RPC request from the mbuf data list received on a
1926  * stream socket. The "waitflag" argument indicates whether or not it
1927  * can sleep.
1928  */
1929 int
1930 nfsrv_getstream(slp, waitflag)
1931 	struct nfssvc_sock *slp;
1932 	int waitflag;
1933 {
1934 	struct mbuf *m, **mpp;
1935 	char *cp1, *cp2;
1936 	int len;
1937 	struct mbuf *om, *m2, *recm;
1938 	u_int32_t recmark;
1939 
1940 	if (slp->ns_flag & SLP_GETSTREAM)
1941 		panic("nfs getstream");
1942 	slp->ns_flag |= SLP_GETSTREAM;
1943 	for (;;) {
1944 	    if (slp->ns_reclen == 0) {
1945 		if (slp->ns_cc < NFSX_UNSIGNED) {
1946 			slp->ns_flag &= ~SLP_GETSTREAM;
1947 			return (0);
1948 		}
1949 		m = slp->ns_raw;
1950 		if (m->m_len >= NFSX_UNSIGNED) {
1951 			bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
1952 			m->m_data += NFSX_UNSIGNED;
1953 			m->m_len -= NFSX_UNSIGNED;
1954 		} else {
1955 			cp1 = (caddr_t)&recmark;
1956 			cp2 = mtod(m, caddr_t);
1957 			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
1958 				while (m->m_len == 0) {
1959 					m = m->m_next;
1960 					cp2 = mtod(m, caddr_t);
1961 				}
1962 				*cp1++ = *cp2++;
1963 				m->m_data++;
1964 				m->m_len--;
1965 			}
1966 		}
1967 		slp->ns_cc -= NFSX_UNSIGNED;
1968 		recmark = ntohl(recmark);
1969 		slp->ns_reclen = recmark & ~0x80000000;
1970 		if (recmark & 0x80000000)
1971 			slp->ns_flag |= SLP_LASTFRAG;
1972 		else
1973 			slp->ns_flag &= ~SLP_LASTFRAG;
1974 		if (slp->ns_reclen > NFS_MAXPACKET) {
1975 			slp->ns_flag &= ~SLP_GETSTREAM;
1976 			return (EPERM);
1977 		}
1978 	    }
1979 
1980 	    /*
1981 	     * Now get the record part.
1982 	     */
1983 	    recm = NULL;
1984 	    if (slp->ns_cc == slp->ns_reclen) {
1985 		recm = slp->ns_raw;
1986 		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
1987 		slp->ns_cc = slp->ns_reclen = 0;
1988 	    } else if (slp->ns_cc > slp->ns_reclen) {
1989 		len = 0;
1990 		m = slp->ns_raw;
1991 		om = (struct mbuf *)0;
1992 		while (len < slp->ns_reclen) {
1993 			if ((len + m->m_len) > slp->ns_reclen) {
1994 				m2 = m_copym(m, 0, slp->ns_reclen - len,
1995 					waitflag);
1996 				if (m2) {
1997 					if (om) {
1998 						om->m_next = m2;
1999 						recm = slp->ns_raw;
2000 					} else
2001 						recm = m2;
2002 					m->m_data += slp->ns_reclen - len;
2003 					m->m_len -= slp->ns_reclen - len;
2004 					len = slp->ns_reclen;
2005 				} else {
2006 					slp->ns_flag &= ~SLP_GETSTREAM;
2007 					return (EWOULDBLOCK);
2008 				}
2009 			} else if ((len + m->m_len) == slp->ns_reclen) {
2010 				om = m;
2011 				len += m->m_len;
2012 				m = m->m_next;
2013 				recm = slp->ns_raw;
2014 				om->m_next = (struct mbuf *)0;
2015 			} else {
2016 				om = m;
2017 				len += m->m_len;
2018 				m = m->m_next;
2019 			}
2020 		}
2021 		slp->ns_raw = m;
2022 		slp->ns_cc -= len;
2023 		slp->ns_reclen = 0;
2024 	    } else {
2025 		slp->ns_flag &= ~SLP_GETSTREAM;
2026 		return (0);
2027 	    }
2028 
2029 	    /*
2030 	     * Accumulate the fragments into a record.
2031 	     */
2032 	    mpp = &slp->ns_frag;
2033 	    while (*mpp)
2034 		mpp = &((*mpp)->m_next);
2035 	    *mpp = recm;
2036 	    if (slp->ns_flag & SLP_LASTFRAG) {
2037 		if (slp->ns_recend)
2038 		    slp->ns_recend->m_nextpkt = slp->ns_frag;
2039 		else
2040 		    slp->ns_rec = slp->ns_frag;
2041 		slp->ns_recend = slp->ns_frag;
2042 		slp->ns_frag = (struct mbuf *)0;
2043 	    }
2044 	}
2045 }
2046 
2047 /*
2048  * Parse an RPC header.
2049  */
2050 int
2051 nfsrv_dorec(slp, nfsd, ndp)
2052 	struct nfssvc_sock *slp;
2053 	struct nfsd *nfsd;
2054 	struct nfsrv_descript **ndp;
2055 {
2056 	struct mbuf *m, *nam;
2057 	struct nfsrv_descript *nd;
2058 	int error;
2059 
2060 	*ndp = NULL;
2061 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2062 	    (m = slp->ns_rec) == (struct mbuf *)0)
2063 		return (ENOBUFS);
2064 	slp->ns_rec = m->m_nextpkt;
2065 	if (slp->ns_rec)
2066 		m->m_nextpkt = (struct mbuf *)0;
2067 	else
2068 		slp->ns_recend = (struct mbuf *)0;
2069 	if (m->m_type == MT_SONAME) {
2070 		nam = m;
2071 		m = m->m_next;
2072 		nam->m_next = NULL;
2073 	} else
2074 		nam = NULL;
2075 	MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2076 		M_NFSRVDESC, M_WAITOK);
2077 	nfs_realign(&m, 10 * NFSX_UNSIGNED);
2078 	nd->nd_md = nd->nd_mrep = m;
2079 	nd->nd_nam2 = nam;
2080 	nd->nd_dpos = mtod(m, caddr_t);
2081 	error = nfs_getreq(nd, nfsd, TRUE);
2082 	if (error) {
2083 		m_freem(nam);
2084 		free((caddr_t)nd, M_NFSRVDESC);
2085 		return (error);
2086 	}
2087 	*ndp = nd;
2088 	nfsd->nfsd_nd = nd;
2089 	return (0);
2090 }
2091 
2092 
2093 /*
2094  * Search for a sleeping nfsd and wake it up.
2095  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2096  * running nfsds will go look for the work in the nfssvc_sock list.
2097  */
2098 void
2099 nfsrv_wakenfsd(slp)
2100 	struct nfssvc_sock *slp;
2101 {
2102 	struct nfsd *nd;
2103 
2104 	if ((slp->ns_flag & SLP_VALID) == 0)
2105 		return;
2106 	for (nd = TAILQ_FIRST(&nfsd_head); nd != NULL;
2107 	    nd = TAILQ_NEXT(nd, nfsd_chain)) {
2108 		if (nd->nfsd_flag & NFSD_WAITING) {
2109 			nd->nfsd_flag &= ~NFSD_WAITING;
2110 			if (nd->nfsd_slp)
2111 				panic("nfsd wakeup");
2112 			slp->ns_sref++;
2113 			nd->nfsd_slp = slp;
2114 			wakeup((caddr_t)nd);
2115 			return;
2116 		}
2117 	}
2118 	slp->ns_flag |= SLP_DOREC;
2119 	nfsd_head_flag |= NFSD_CHECKSLP;
2120 }
2121 #endif /* NFSSERVER */
2122