xref: /openbsd-src/sys/nfs/nfs_socket.c (revision b2ea75c1b17e1a9a339660e7ed45cd24946b230e)
1 /*	$OpenBSD: nfs_socket.c,v 1.22 2001/06/25 03:28:09 csapuntz Exp $	*/
2 /*	$NetBSD: nfs_socket.c,v 1.27 1996/04/15 20:20:00 thorpej Exp $	*/
3 
4 /*
5  * Copyright (c) 1989, 1991, 1993, 1995
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * Rick Macklem at The University of Guelph.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the University of
22  *	California, Berkeley and its contributors.
23  * 4. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  *
39  *	@(#)nfs_socket.c	8.5 (Berkeley) 3/30/95
40  */
41 
42 /*
43  * Socket operations for use by nfs
44  */
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/kernel.h>
51 #include <sys/mbuf.h>
52 #include <sys/vnode.h>
53 #include <sys/domain.h>
54 #include <sys/protosw.h>
55 #include <sys/socket.h>
56 #include <sys/socketvar.h>
57 #include <sys/syslog.h>
58 #include <sys/tprintf.h>
59 #include <sys/namei.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/tcp.h>
63 
64 #include <nfs/rpcv2.h>
65 #include <nfs/nfsproto.h>
66 #include <nfs/nfs.h>
67 #include <nfs/xdr_subs.h>
68 #include <nfs/nfsm_subs.h>
69 #include <nfs/nfsmount.h>
70 #include <nfs/nfsnode.h>
71 #include <nfs/nfsrtt.h>
72 #include <nfs/nfs_var.h>
73 
74 #define	TRUE	1
75 #define	FALSE	0
76 
77 /*
78  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
79  * Use the mean and mean deviation of rtt for the appropriate type of rpc
80  * for the frequent rpcs and a default for the others.
81  * The justification for doing "other" this way is that these rpcs
82  * happen so infrequently that timer est. would probably be stale.
83  * Also, since many of these rpcs are
84  * non-idempotent, a conservative timeout is desired.
85  * getattr, lookup - A+2D
86  * read, write     - A+4D
87  * other           - nm_timeo
88  */
89 #define	NFS_RTO(n, t) \
90 	((t) == 0 ? (n)->nm_timeo : \
91 	 ((t) < 3 ? \
92 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
93 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
94 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
95 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
96 /*
97  * External data, mostly RPC constants in XDR form
98  */
99 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
100 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
101 	rpc_auth_kerb;
102 extern u_int32_t nfs_prog;
103 extern struct nfsstats nfsstats;
104 extern int nfsv3_procid[NFS_NPROCS];
105 extern int nfs_ticks;
106 
107 /*
108  * Defines which timer to use for the procnum.
109  * 0 - default
110  * 1 - getattr
111  * 2 - lookup
112  * 3 - read
113  * 4 - write
114  */
115 static int proct[NFS_NPROCS] = {
116 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
117 	0, 0, 0,
118 };
119 
120 /*
121  * There is a congestion window for outstanding rpcs maintained per mount
122  * point. The cwnd size is adjusted in roughly the way that:
123  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
124  * SIGCOMM '88". ACM, August 1988.
125  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
126  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
127  * of rpcs is in progress.
128  * (The sent count and cwnd are scaled for integer arith.)
129  * Variants of "slow start" were tried and were found to be too much of a
130  * performance hit (ave. rtt 3 times larger),
131  * I suspect due to the large rtt that nfs rpcs have.
132  */
133 #define	NFS_CWNDSCALE	256
134 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
135 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
136 int nfsrtton = 0;
137 struct nfsrtt nfsrtt;
138 
139 /*
140  * Initialize sockets and congestion for a new NFS connection.
141  * We do not free the sockaddr if error.
142  */
143 int
144 nfs_connect(nmp, rep)
145 	register struct nfsmount *nmp;
146 	struct nfsreq *rep;
147 {
148 	register struct socket *so;
149 	int s, error, rcvreserve, sndreserve;
150 	struct sockaddr *saddr;
151 	struct sockaddr_in *sin;
152 	struct mbuf *m;
153 
154 	nmp->nm_so = (struct socket *)0;
155 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
156 	error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
157 		nmp->nm_soproto);
158 	if (error)
159 		goto bad;
160 	so = nmp->nm_so;
161 	nmp->nm_soflags = so->so_proto->pr_flags;
162 
163 	/*
164 	 * Some servers require that the client port be a reserved port number.
165 	 * We always allocate a reserved port, as this prevents filehandle
166 	 * disclosure through UDP port capture.
167 	 */
168 	if (saddr->sa_family == AF_INET) {
169 		struct mbuf *mopt;
170 		int *ip;
171 
172 		MGET(mopt, M_WAIT, MT_SOOPTS);
173 		mopt->m_len = sizeof(int);
174 		ip = mtod(mopt, int *);
175 		*ip = IP_PORTRANGE_LOW;
176 		error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt);
177 		if (error)
178 			goto bad;
179 
180 		MGET(m, M_WAIT, MT_SONAME);
181 		sin = mtod(m, struct sockaddr_in *);
182 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
183 		sin->sin_family = AF_INET;
184 		sin->sin_addr.s_addr = INADDR_ANY;
185 		sin->sin_port = htons(0);
186 		error = sobind(so, m);
187 		m_freem(m);
188 		if (error)
189 			goto bad;
190 
191 		MGET(mopt, M_WAIT, MT_SOOPTS);
192 		mopt->m_len = sizeof(int);
193 		ip = mtod(mopt, int *);
194 		*ip = IP_PORTRANGE_DEFAULT;
195 		error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt);
196 		if (error)
197 			goto bad;
198 	}
199 
200 	/*
201 	 * Protocols that do not require connections may be optionally left
202 	 * unconnected for servers that reply from a port other than NFS_PORT.
203 	 */
204 	if (nmp->nm_flag & NFSMNT_NOCONN) {
205 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
206 			error = ENOTCONN;
207 			goto bad;
208 		}
209 	} else {
210 		error = soconnect(so, nmp->nm_nam);
211 		if (error)
212 			goto bad;
213 
214 		/*
215 		 * Wait for the connection to complete. Cribbed from the
216 		 * connect system call but with the wait timing out so
217 		 * that interruptible mounts don't hang here for a long time.
218 		 */
219 		s = splsoftnet();
220 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
221 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
222 				"nfscon", 2 * hz);
223 			if ((so->so_state & SS_ISCONNECTING) &&
224 			    so->so_error == 0 && rep &&
225 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
226 				so->so_state &= ~SS_ISCONNECTING;
227 				splx(s);
228 				goto bad;
229 			}
230 		}
231 		if (so->so_error) {
232 			error = so->so_error;
233 			so->so_error = 0;
234 			splx(s);
235 			goto bad;
236 		}
237 		splx(s);
238 	}
239 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
240 		so->so_rcv.sb_timeo = (5 * hz);
241 		so->so_snd.sb_timeo = (5 * hz);
242 	} else {
243 		so->so_rcv.sb_timeo = 0;
244 		so->so_snd.sb_timeo = 0;
245 	}
246 	if (nmp->nm_sotype == SOCK_DGRAM) {
247 		sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR;
248 		rcvreserve = max(nmp->nm_rsize, nmp->nm_readdirsize) +
249 		    NFS_MAXPKTHDR;
250 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
251 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
252 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
253 		    NFS_MAXPKTHDR) * 2;
254 	} else {
255 		if (nmp->nm_sotype != SOCK_STREAM)
256 			panic("nfscon sotype");
257 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
258 			MGET(m, M_WAIT, MT_SOOPTS);
259 			*mtod(m, int32_t *) = 1;
260 			m->m_len = sizeof(int32_t);
261 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
262 		}
263 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
264 			MGET(m, M_WAIT, MT_SOOPTS);
265 			*mtod(m, int32_t *) = 1;
266 			m->m_len = sizeof(int32_t);
267 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
268 		}
269 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
270 		    sizeof (u_int32_t)) * 2;
271 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
272 		    sizeof (u_int32_t)) * 2;
273 	}
274 	error = soreserve(so, sndreserve, rcvreserve);
275 	if (error)
276 		goto bad;
277 	so->so_rcv.sb_flags |= SB_NOINTR;
278 	so->so_snd.sb_flags |= SB_NOINTR;
279 
280 	/* Initialize other non-zero congestion variables */
281 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
282 		nmp->nm_srtt[4] = (NFS_TIMEO << 3);
283 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
284 		nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0;
285 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
286 	nmp->nm_sent = 0;
287 	nmp->nm_timeouts = 0;
288 	return (0);
289 
290 bad:
291 	nfs_disconnect(nmp);
292 	return (error);
293 }
294 
295 /*
296  * Reconnect routine:
297  * Called when a connection is broken on a reliable protocol.
298  * - clean up the old socket
299  * - nfs_connect() again
300  * - set R_MUSTRESEND for all outstanding requests on mount point
301  * If this fails the mount point is DEAD!
302  * nb: Must be called with the nfs_sndlock() set on the mount point.
303  */
304 int
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)) != 0) {
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 int
361 nfs_send(so, nam, top, rep)
362 	register struct socket *so;
363 	struct mbuf *nam;
364 	register struct mbuf *top;
365 	struct nfsreq *rep;
366 {
367 	struct mbuf *sendnam;
368 	int error, soflags, flags;
369 
370 	if (rep) {
371 		if (rep->r_flags & R_SOFTTERM) {
372 			m_freem(top);
373 			return (EINTR);
374 		}
375 		if ((so = rep->r_nmp->nm_so) == NULL) {
376 			rep->r_flags |= R_MUSTRESEND;
377 			m_freem(top);
378 			return (0);
379 		}
380 		rep->r_flags &= ~R_MUSTRESEND;
381 		soflags = rep->r_nmp->nm_soflags;
382 	} else
383 		soflags = so->so_proto->pr_flags;
384 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
385 		sendnam = (struct mbuf *)0;
386 	else
387 		sendnam = nam;
388 	if (so->so_type == SOCK_SEQPACKET)
389 		flags = MSG_EOR;
390 	else
391 		flags = 0;
392 
393 	error = sosend(so, sendnam, (struct uio *)0, top,
394 		(struct mbuf *)0, flags);
395 	if (error) {
396 		if (rep) {
397 			log(LOG_INFO, "nfs send error %d for server %s\n",error,
398 			    rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
399 			/*
400 			 * Deal with errors for the client side.
401 			 */
402 			if (rep->r_flags & R_SOFTTERM)
403 				error = EINTR;
404 			else
405 				rep->r_flags |= R_MUSTRESEND;
406 		} else
407 			log(LOG_INFO, "nfsd send error %d\n", error);
408 
409 		/*
410 		 * Handle any recoverable (soft) socket errors here. (???)
411 		 */
412 		if (error != EINTR && error != ERESTART &&
413 			error != EWOULDBLOCK && error != EPIPE)
414 			error = 0;
415 	}
416 	return (error);
417 }
418 
419 #ifdef NFSCLIENT
420 /*
421  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
422  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
423  * Mark and consolidate the data into a new mbuf list.
424  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
425  *     small mbufs.
426  * For SOCK_STREAM we must be very careful to read an entire record once
427  * we have read any of it, even if the system call has been interrupted.
428  */
429 int
430 nfs_receive(rep, aname, mp)
431 	register struct nfsreq *rep;
432 	struct mbuf **aname;
433 	struct mbuf **mp;
434 {
435 	register struct socket *so;
436 	struct uio auio;
437 	struct iovec aio;
438 	register struct mbuf *m;
439 	struct mbuf *control;
440 	u_int32_t len;
441 	struct mbuf **getnam;
442 	int error, sotype, rcvflg;
443 	struct proc *p = curproc;	/* XXX */
444 
445 	/*
446 	 * Set up arguments for soreceive()
447 	 */
448 	*mp = (struct mbuf *)0;
449 	*aname = (struct mbuf *)0;
450 	sotype = rep->r_nmp->nm_sotype;
451 
452 	/*
453 	 * For reliable protocols, lock against other senders/receivers
454 	 * in case a reconnect is necessary.
455 	 * For SOCK_STREAM, first get the Record Mark to find out how much
456 	 * more there is to get.
457 	 * We must lock the socket against other receivers
458 	 * until we have an entire rpc request/reply.
459 	 */
460 	if (sotype != SOCK_DGRAM) {
461 		error = nfs_sndlock(&rep->r_nmp->nm_flag, rep);
462 		if (error)
463 			return (error);
464 tryagain:
465 		/*
466 		 * Check for fatal errors and resending request.
467 		 */
468 		/*
469 		 * Ugh: If a reconnect attempt just happened, nm_so
470 		 * would have changed. NULL indicates a failed
471 		 * attempt that has essentially shut down this
472 		 * mount point.
473 		 */
474 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
475 			nfs_sndunlock(&rep->r_nmp->nm_flag);
476 			return (EINTR);
477 		}
478 		so = rep->r_nmp->nm_so;
479 		if (!so) {
480 			error = nfs_reconnect(rep);
481 			if (error) {
482 				nfs_sndunlock(&rep->r_nmp->nm_flag);
483 				return (error);
484 			}
485 			goto tryagain;
486 		}
487 		while (rep->r_flags & R_MUSTRESEND) {
488 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
489 			nfsstats.rpcretries++;
490 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
491 			if (error) {
492 				if (error == EINTR || error == ERESTART ||
493 				    (error = nfs_reconnect(rep)) != 0) {
494 					nfs_sndunlock(&rep->r_nmp->nm_flag);
495 					return (error);
496 				}
497 				goto tryagain;
498 			}
499 		}
500 		nfs_sndunlock(&rep->r_nmp->nm_flag);
501 		if (sotype == SOCK_STREAM) {
502 			aio.iov_base = (caddr_t) &len;
503 			aio.iov_len = sizeof(u_int32_t);
504 			auio.uio_iov = &aio;
505 			auio.uio_iovcnt = 1;
506 			auio.uio_segflg = UIO_SYSSPACE;
507 			auio.uio_rw = UIO_READ;
508 			auio.uio_offset = 0;
509 			auio.uio_resid = sizeof(u_int32_t);
510 			auio.uio_procp = p;
511 			do {
512 			   rcvflg = MSG_WAITALL;
513 			   error = soreceive(so, (struct mbuf **)0, &auio,
514 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
515 			   if (error == EWOULDBLOCK && rep) {
516 				if (rep->r_flags & R_SOFTTERM)
517 					return (EINTR);
518 			   }
519 			} while (error == EWOULDBLOCK);
520 			if (!error && auio.uio_resid > 0) {
521 			    log(LOG_INFO,
522 				 "short receive (%d/%d) from nfs server %s\n",
523 				 sizeof(u_int32_t) - auio.uio_resid,
524 				 sizeof(u_int32_t),
525 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
526 			    error = EPIPE;
527 			}
528 			if (error)
529 				goto errout;
530 			len = ntohl(len) & ~0x80000000;
531 			/*
532 			 * This is SERIOUS! We are out of sync with the sender
533 			 * and forcing a disconnect/reconnect is all I can do.
534 			 */
535 			if (len > NFS_MAXPACKET) {
536 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
537 				"impossible packet length",
538 				len,
539 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
540 			    error = EFBIG;
541 			    goto errout;
542 			}
543 			auio.uio_resid = len;
544 			do {
545 			    rcvflg = MSG_WAITALL;
546 			    error =  soreceive(so, (struct mbuf **)0,
547 				&auio, mp, (struct mbuf **)0, &rcvflg);
548 			} while (error == EWOULDBLOCK || error == EINTR ||
549 				 error == ERESTART);
550 			if (!error && auio.uio_resid > 0) {
551 			    log(LOG_INFO,
552 				"short receive (%d/%d) from nfs server %s\n",
553 				len - auio.uio_resid, len,
554 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
555 			    error = EPIPE;
556 			}
557 		} else {
558 			/*
559 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
560 			 * and soreceive() will return when it has either a
561 			 * control msg or a data msg.
562 			 * We have no use for control msg., but must grab them
563 			 * and then throw them away so we know what is going
564 			 * on.
565 			 */
566 			auio.uio_resid = len = 100000000; /* Anything Big */
567 			auio.uio_procp = p;
568 			do {
569 			    rcvflg = 0;
570 			    error =  soreceive(so, (struct mbuf **)0,
571 				&auio, mp, &control, &rcvflg);
572 			    if (control)
573 				m_freem(control);
574 			    if (error == EWOULDBLOCK && rep) {
575 				if (rep->r_flags & R_SOFTTERM)
576 					return (EINTR);
577 			    }
578 			} while (error == EWOULDBLOCK ||
579 				 (!error && *mp == NULL && control));
580 			if ((rcvflg & MSG_EOR) == 0)
581 				printf("Egad!!\n");
582 			if (!error && *mp == NULL)
583 				error = EPIPE;
584 			len -= auio.uio_resid;
585 		}
586 errout:
587 		if (error && error != EINTR && error != ERESTART) {
588 			m_freem(*mp);
589 			*mp = (struct mbuf *)0;
590 			if (error != EPIPE)
591 				log(LOG_INFO,
592 				    "receive error %d from nfs server %s\n",
593 				    error,
594 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
595 			error = nfs_sndlock(&rep->r_nmp->nm_flag, rep);
596 			if (!error)
597 				error = nfs_reconnect(rep);
598 			if (!error)
599 				goto tryagain;
600 		}
601 	} else {
602 		if ((so = rep->r_nmp->nm_so) == NULL)
603 			return (EACCES);
604 		if (so->so_state & SS_ISCONNECTED)
605 			getnam = (struct mbuf **)0;
606 		else
607 			getnam = aname;
608 		auio.uio_resid = len = 1000000;
609 		auio.uio_procp = p;
610 		do {
611 			rcvflg = 0;
612 			error =  soreceive(so, getnam, &auio, mp,
613 				(struct mbuf **)0, &rcvflg);
614 			if (error == EWOULDBLOCK &&
615 			    (rep->r_flags & R_SOFTTERM))
616 				return (EINTR);
617 		} while (error == EWOULDBLOCK);
618 		len -= auio.uio_resid;
619 	}
620 	if (error) {
621 		m_freem(*mp);
622 		*mp = (struct mbuf *)0;
623 	}
624 	/*
625 	 * Search for any mbufs that are not a multiple of 4 bytes long
626 	 * or with m_data not longword aligned.
627 	 * These could cause pointer alignment problems, so copy them to
628 	 * well aligned mbufs.
629 	 */
630 	nfs_realign(*mp, 5 * NFSX_UNSIGNED);
631 	return (error);
632 }
633 
634 /*
635  * Implement receipt of reply on a socket.
636  * We must search through the list of received datagrams matching them
637  * with outstanding requests using the xid, until ours is found.
638  */
639 /* ARGSUSED */
640 int
641 nfs_reply(myrep)
642 	struct nfsreq *myrep;
643 {
644 	register struct nfsreq *rep;
645 	register struct nfsmount *nmp = myrep->r_nmp;
646 	register int32_t t1;
647 	struct mbuf *mrep, *nam, *md;
648 	u_int32_t rxid, *tl;
649 	caddr_t dpos, cp2;
650 	int error;
651 
652 	/*
653 	 * Loop around until we get our own reply
654 	 */
655 	for (;;) {
656 		/*
657 		 * Lock against other receivers so that I don't get stuck in
658 		 * sbwait() after someone else has received my reply for me.
659 		 * Also necessary for connection based protocols to avoid
660 		 * race conditions during a reconnect.
661 		 */
662 		error = nfs_rcvlock(myrep);
663 		if (error)
664 			return (error);
665 		/* Already received, bye bye */
666 		if (myrep->r_mrep != NULL) {
667 			nfs_rcvunlock(&nmp->nm_flag);
668 			return (0);
669 		}
670 		/*
671 		 * Get the next Rpc reply off the socket
672 		 */
673 		error = nfs_receive(myrep, &nam, &mrep);
674 		nfs_rcvunlock(&nmp->nm_flag);
675 		if (error) {
676 
677 			/*
678 			 * Ignore routing errors on connectionless protocols??
679 			 */
680 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
681 				nmp->nm_so->so_error = 0;
682 				if (myrep->r_flags & R_GETONEREP)
683 					return (0);
684 				continue;
685 			}
686 			return (error);
687 		}
688 		if (nam)
689 			m_freem(nam);
690 
691 		/*
692 		 * Get the xid and check that it is an rpc reply
693 		 */
694 		md = mrep;
695 		dpos = mtod(md, caddr_t);
696 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
697 		rxid = *tl++;
698 		if (*tl != rpc_reply) {
699 			nfsstats.rpcinvalid++;
700 			m_freem(mrep);
701 nfsmout:
702 			if (myrep->r_flags & R_GETONEREP)
703 				return (0);
704 			continue;
705 		}
706 
707 		/*
708 		 * Loop through the request list to match up the reply
709 		 * Iff no match, just drop the datagram
710 		 */
711 		for (rep = nfs_reqq.tqh_first; rep != 0;
712 		    rep = rep->r_chain.tqe_next) {
713 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
714 				/* Found it.. */
715 				rep->r_mrep = mrep;
716 				rep->r_md = md;
717 				rep->r_dpos = dpos;
718 				if (nfsrtton) {
719 					struct rttl *rt;
720 
721 					rt = &nfsrtt.rttl[nfsrtt.pos];
722 					rt->proc = rep->r_procnum;
723 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
724 					rt->sent = nmp->nm_sent;
725 					rt->cwnd = nmp->nm_cwnd;
726 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
727 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
728 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
729 					rt->tstamp = time;
730 					if (rep->r_flags & R_TIMING)
731 						rt->rtt = rep->r_rtt;
732 					else
733 						rt->rtt = 1000000;
734 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
735 				}
736 				/*
737 				 * Update congestion window.
738 				 * Do the additive increase of
739 				 * one rpc/rtt.
740 				 */
741 				if (nmp->nm_cwnd <= nmp->nm_sent) {
742 					nmp->nm_cwnd +=
743 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
744 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
745 					if (nmp->nm_cwnd > NFS_MAXCWND)
746 						nmp->nm_cwnd = NFS_MAXCWND;
747 				}
748 				rep->r_flags &= ~R_SENT;
749 				nmp->nm_sent -= NFS_CWNDSCALE;
750 				/*
751 				 * Update rtt using a gain of 0.125 on the mean
752 				 * and a gain of 0.25 on the deviation.
753 				 */
754 				if (rep->r_flags & R_TIMING) {
755 					/*
756 					 * Since the timer resolution of
757 					 * NFS_HZ is so course, it can often
758 					 * result in r_rtt == 0. Since
759 					 * r_rtt == N means that the actual
760 					 * rtt is between N+dt and N+2-dt ticks,
761 					 * add 1.
762 					 */
763 					t1 = rep->r_rtt + 1;
764 					t1 -= (NFS_SRTT(rep) >> 3);
765 					NFS_SRTT(rep) += t1;
766 					if (t1 < 0)
767 						t1 = -t1;
768 					t1 -= (NFS_SDRTT(rep) >> 2);
769 					NFS_SDRTT(rep) += t1;
770 				}
771 				nmp->nm_timeouts = 0;
772 				break;
773 			}
774 		}
775 		/*
776 		 * If not matched to a request, drop it.
777 		 * If it's mine, get out.
778 		 */
779 		if (rep == 0) {
780 			nfsstats.rpcunexpected++;
781 			m_freem(mrep);
782 		} else if (rep == myrep) {
783 			if (rep->r_mrep == NULL)
784 				panic("nfsreply nil");
785 			return (0);
786 		}
787 		if (myrep->r_flags & R_GETONEREP)
788 			return (0);
789 	}
790 }
791 
792 /*
793  * nfs_request - goes something like this
794  *	- fill in request struct
795  *	- links it into list
796  *	- calls nfs_send() for first transmit
797  *	- calls nfs_receive() to get reply
798  *	- break down rpc header and return with nfs reply pointed to
799  *	  by mrep or error
800  * nb: always frees up mreq mbuf list
801  */
802 int
803 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
804 	struct vnode *vp;
805 	struct mbuf *mrest;
806 	int procnum;
807 	struct proc *procp;
808 	struct ucred *cred;
809 	struct mbuf **mrp;
810 	struct mbuf **mdp;
811 	caddr_t *dposp;
812 {
813 	register struct mbuf *m, *mrep;
814 	register struct nfsreq *rep;
815 	register u_int32_t *tl;
816 	register int i;
817 	struct nfsmount *nmp;
818 	struct mbuf *md, *mheadend;
819 	char nickv[RPCX_NICKVERF];
820 	time_t reqtime, waituntil;
821 	caddr_t dpos, cp2;
822 	int t1, s, error = 0, mrest_len, auth_len, auth_type;
823 	int trylater_delay = 15, trylater_cnt = 0, failed_auth = 0;
824 	int verf_len, verf_type;
825 	u_int32_t xid;
826 	char *auth_str, *verf_str;
827 	NFSKERBKEY_T key;		/* save session key */
828 
829 	nmp = VFSTONFS(vp->v_mount);
830 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
831 	rep->r_nmp = nmp;
832 	rep->r_vp = vp;
833 	rep->r_procp = procp;
834 	rep->r_procnum = procnum;
835 	i = 0;
836 	m = mrest;
837 	while (m) {
838 		i += m->m_len;
839 		m = m->m_next;
840 	}
841 	mrest_len = i;
842 
843 	/*
844 	 * Get the RPC header with authorization.
845 	 */
846 kerbauth:
847 	verf_str = auth_str = (char *)0;
848 	if (nmp->nm_flag & NFSMNT_KERB) {
849 		verf_str = nickv;
850 		verf_len = sizeof (nickv);
851 		auth_type = RPCAUTH_KERB4;
852 		bzero((caddr_t)key, sizeof (key));
853 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
854 			&auth_len, verf_str, verf_len)) {
855 			error = nfs_getauth(nmp, rep, cred, &auth_str,
856 				&auth_len, verf_str, &verf_len, key);
857 			if (error) {
858 				free((caddr_t)rep, M_NFSREQ);
859 				m_freem(mrest);
860 				return (error);
861 			}
862 		}
863 	} else {
864 		auth_type = RPCAUTH_UNIX;
865 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
866 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
867 			5 * NFSX_UNSIGNED;
868 	}
869 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
870 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
871 	if (auth_str)
872 		free(auth_str, M_TEMP);
873 
874 	/*
875 	 * For stream protocols, insert a Sun RPC Record Mark.
876 	 */
877 	if (nmp->nm_sotype == SOCK_STREAM) {
878 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
879 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
880 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
881 	}
882 	rep->r_mreq = m;
883 	rep->r_xid = xid;
884 tryagain:
885 	if (nmp->nm_flag & NFSMNT_SOFT)
886 		rep->r_retry = nmp->nm_retry;
887 	else
888 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
889 	rep->r_rtt = rep->r_rexmit = 0;
890 	if (proct[procnum] > 0)
891 		rep->r_flags = R_TIMING;
892 	else
893 		rep->r_flags = 0;
894 	rep->r_mrep = NULL;
895 
896 	/*
897 	 * Do the client side RPC.
898 	 */
899 	nfsstats.rpcrequests++;
900 	/*
901 	 * Chain request into list of outstanding requests. Be sure
902 	 * to put it LAST so timer finds oldest requests first.
903 	 */
904 	s = splsoftclock();
905 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
906 
907 	/* Get send time for nqnfs */
908 	reqtime = time.tv_sec;
909 
910 	/*
911 	 * If backing off another request or avoiding congestion, don't
912 	 * send this one now but let timer do it. If not timing a request,
913 	 * do it now.
914 	 */
915 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
916 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
917 		nmp->nm_sent < nmp->nm_cwnd)) {
918 		splx(s);
919 		if (nmp->nm_soflags & PR_CONNREQUIRED)
920 			error = nfs_sndlock(&nmp->nm_flag, rep);
921 		if (!error) {
922 			error = nfs_send(nmp->nm_so, nmp->nm_nam,
923 					m_copym(m, 0, M_COPYALL, M_WAIT),
924 					rep);
925 			if (nmp->nm_soflags & PR_CONNREQUIRED)
926 				nfs_sndunlock(&nmp->nm_flag);
927 		}
928 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
929 			nmp->nm_sent += NFS_CWNDSCALE;
930 			rep->r_flags |= R_SENT;
931 		}
932 	} else {
933 		splx(s);
934 		rep->r_rtt = -1;
935 	}
936 
937 	/*
938 	 * Wait for the reply from our send or the timer's.
939 	 */
940 	if (!error || error == EPIPE)
941 		error = nfs_reply(rep);
942 
943 	/*
944 	 * RPC done, unlink the request.
945 	 */
946 	s = splsoftclock();
947 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
948 	splx(s);
949 
950 	/*
951 	 * Decrement the outstanding request count.
952 	 */
953 	if (rep->r_flags & R_SENT) {
954 		rep->r_flags &= ~R_SENT;	/* paranoia */
955 		nmp->nm_sent -= NFS_CWNDSCALE;
956 	}
957 
958 	/*
959 	 * If there was a successful reply and a tprintf msg.
960 	 * tprintf a response.
961 	 */
962 	if (!error && (rep->r_flags & R_TPRINTFMSG))
963 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
964 		    "is alive again");
965 	mrep = rep->r_mrep;
966 	md = rep->r_md;
967 	dpos = rep->r_dpos;
968 	if (error) {
969 		m_freem(rep->r_mreq);
970 		free((caddr_t)rep, M_NFSREQ);
971 		return (error);
972 	}
973 
974 	/*
975 	 * break down the rpc header and check if ok
976 	 */
977 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
978 	if (*tl++ == rpc_msgdenied) {
979 		if (*tl == rpc_mismatch)
980 			error = EOPNOTSUPP;
981 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
982 			if (!failed_auth) {
983 				failed_auth++;
984 				mheadend->m_next = (struct mbuf *)0;
985 				m_freem(mrep);
986 				m_freem(rep->r_mreq);
987 				goto kerbauth;
988 			} else
989 				error = EAUTH;
990 		} else
991 			error = EACCES;
992 		m_freem(mrep);
993 		m_freem(rep->r_mreq);
994 		free((caddr_t)rep, M_NFSREQ);
995 		return (error);
996 	}
997 
998 	/*
999 	 * Grab any Kerberos verifier, otherwise just throw it away.
1000 	 */
1001 	verf_type = fxdr_unsigned(int, *tl++);
1002 	i = fxdr_unsigned(int32_t, *tl);
1003 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1004 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1005 		if (error)
1006 			goto nfsmout;
1007 	} else if (i > 0)
1008 		nfsm_adv(nfsm_rndup(i));
1009 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1010 	/* 0 == ok */
1011 	if (*tl == 0) {
1012 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1013 		if (*tl != 0) {
1014 			error = fxdr_unsigned(int, *tl);
1015 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1016 				error == NFSERR_TRYLATER) {
1017 				m_freem(mrep);
1018 				error = 0;
1019 				waituntil = time.tv_sec + trylater_delay;
1020 				while (time.tv_sec < waituntil)
1021 					(void) tsleep((caddr_t)&lbolt,
1022 						PSOCK, "nqnfstry", 0);
1023 				trylater_delay *= nfs_backoff[trylater_cnt];
1024 				if (trylater_cnt < 7)
1025 					trylater_cnt++;
1026 				goto tryagain;
1027 			}
1028 
1029 			/*
1030 			 * If the File Handle was stale, invalidate the
1031 			 * lookup cache, just in case.
1032 			 */
1033 			if (error == ESTALE)
1034 				cache_purge(vp);
1035 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1036 				*mrp = mrep;
1037 				*mdp = md;
1038 				*dposp = dpos;
1039 				error |= NFSERR_RETERR;
1040 			} else
1041 				m_freem(mrep);
1042 			m_freem(rep->r_mreq);
1043 			free((caddr_t)rep, M_NFSREQ);
1044 			return (error);
1045 		}
1046 
1047 		*mrp = mrep;
1048 		*mdp = md;
1049 		*dposp = dpos;
1050 		m_freem(rep->r_mreq);
1051 		FREE((caddr_t)rep, M_NFSREQ);
1052 		return (0);
1053 	}
1054 	m_freem(mrep);
1055 	error = EPROTONOSUPPORT;
1056 nfsmout:
1057 	m_freem(rep->r_mreq);
1058 	free((caddr_t)rep, M_NFSREQ);
1059 	return (error);
1060 }
1061 #endif /* NFSCLIENT */
1062 
1063 /*
1064  * Generate the rpc reply header
1065  * siz arg. is used to decide if adding a cluster is worthwhile
1066  */
1067 int
1068 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1069 	int siz;
1070 	struct nfsrv_descript *nd;
1071 	struct nfssvc_sock *slp;
1072 	int err;
1073 	int cache;
1074 	u_quad_t *frev;
1075 	struct mbuf **mrq;
1076 	struct mbuf **mbp;
1077 	caddr_t *bposp;
1078 {
1079 	register u_int32_t *tl;
1080 	register struct mbuf *mreq;
1081 	caddr_t bpos;
1082 	struct mbuf *mb, *mb2;
1083 
1084 	MGETHDR(mreq, M_WAIT, MT_DATA);
1085 	mb = mreq;
1086 	/*
1087 	 * If this is a big reply, use a cluster else
1088 	 * try and leave leading space for the lower level headers.
1089 	 */
1090 	siz += RPC_REPLYSIZ;
1091 	if (siz >= max_datalen) {
1092 		MCLGET(mreq, M_WAIT);
1093 	} else
1094 		mreq->m_data += max_hdr;
1095 	tl = mtod(mreq, u_int32_t *);
1096 	mreq->m_len = 6 * NFSX_UNSIGNED;
1097 	bpos = ((caddr_t)tl) + mreq->m_len;
1098 	*tl++ = txdr_unsigned(nd->nd_retxid);
1099 	*tl++ = rpc_reply;
1100 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1101 		*tl++ = rpc_msgdenied;
1102 		if (err & NFSERR_AUTHERR) {
1103 			*tl++ = rpc_autherr;
1104 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1105 			mreq->m_len -= NFSX_UNSIGNED;
1106 			bpos -= NFSX_UNSIGNED;
1107 		} else {
1108 			*tl++ = rpc_mismatch;
1109 			*tl++ = txdr_unsigned(RPC_VER2);
1110 			*tl = txdr_unsigned(RPC_VER2);
1111 		}
1112 	} else {
1113 		*tl++ = rpc_msgaccepted;
1114 
1115 		/*
1116 		 * For Kerberos authentication, we must send the nickname
1117 		 * verifier back, otherwise just RPCAUTH_NULL.
1118 		 */
1119 		if (nd->nd_flag & ND_KERBFULL) {
1120 		    register struct nfsuid *nuidp;
1121 		    struct timeval ktvin, ktvout;
1122 
1123 		    for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1124 			nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1125 			if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1126 			    (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1127 			     &nuidp->nu_haddr, nd->nd_nam2)))
1128 			    break;
1129 		    }
1130 		    if (nuidp) {
1131 			ktvin.tv_sec =
1132 			    txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1133 			ktvin.tv_usec =
1134 			    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1135 
1136 			/*
1137 			 * Encrypt the timestamp in ecb mode using the
1138 			 * session key.
1139 			 */
1140 #ifdef NFSKERB
1141 			XXX
1142 #endif
1143 
1144 			*tl++ = rpc_auth_kerb;
1145 			*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1146 			*tl = ktvout.tv_sec;
1147 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1148 			*tl++ = ktvout.tv_usec;
1149 			*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1150 		    } else {
1151 			*tl++ = 0;
1152 			*tl++ = 0;
1153 		    }
1154 		} else {
1155 			*tl++ = 0;
1156 			*tl++ = 0;
1157 		}
1158 		switch (err) {
1159 		case EPROGUNAVAIL:
1160 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1161 			break;
1162 		case EPROGMISMATCH:
1163 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1164 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1165 			*tl++ = txdr_unsigned(2);
1166 			*tl = txdr_unsigned(3);
1167 			break;
1168 		case EPROCUNAVAIL:
1169 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1170 			break;
1171 		case EBADRPC:
1172 			*tl = txdr_unsigned(RPC_GARBAGE);
1173 			break;
1174 		default:
1175 			*tl = 0;
1176 			if (err != NFSERR_RETVOID) {
1177 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1178 				if (err)
1179 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1180 				else
1181 				    *tl = 0;
1182 			}
1183 			break;
1184 		};
1185 	}
1186 
1187 	*mrq = mreq;
1188 	if (mrq != NULL)
1189 		*mbp = mb;
1190 	*bposp = bpos;
1191 	if (err != 0 && err != NFSERR_RETVOID)
1192 		nfsstats.srvrpc_errs++;
1193 	return (0);
1194 }
1195 
1196 /*
1197  * Nfs timer routine
1198  * Scan the nfsreq list and retranmit any requests that have timed out
1199  * To avoid retransmission attempts on STREAM sockets (in the future) make
1200  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1201  */
1202 void
1203 nfs_timer(arg)
1204 	void *arg;
1205 {
1206 	struct timeout *to = (struct timeout *)arg;
1207 	struct nfsreq *rep;
1208 	struct mbuf *m;
1209 	struct socket *so;
1210 	struct nfsmount *nmp;
1211 	int timeo;
1212 	int s, error;
1213 #ifdef NFSSERVER
1214 	struct nfssvc_sock *slp;
1215 	u_quad_t cur_usec;
1216 #endif
1217 
1218 	s = splsoftnet();
1219 	for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) {
1220 		nmp = rep->r_nmp;
1221 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1222 			continue;
1223 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1224 			rep->r_flags |= R_SOFTTERM;
1225 			continue;
1226 		}
1227 		if (rep->r_rtt >= 0) {
1228 			rep->r_rtt++;
1229 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1230 				timeo = nmp->nm_timeo;
1231 			else
1232 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1233 			if (nmp->nm_timeouts > 0)
1234 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1235 			if (rep->r_rtt <= timeo)
1236 				continue;
1237 			if (nmp->nm_timeouts < 8)
1238 				nmp->nm_timeouts++;
1239 		}
1240 		/*
1241 		 * Check for server not responding
1242 		 */
1243 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1244 		     rep->r_rexmit > nmp->nm_deadthresh) {
1245 			nfs_msg(rep->r_procp,
1246 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1247 			    "not responding");
1248 			rep->r_flags |= R_TPRINTFMSG;
1249 		}
1250 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1251 			nfsstats.rpctimeouts++;
1252 			rep->r_flags |= R_SOFTTERM;
1253 			continue;
1254 		}
1255 		if (nmp->nm_sotype != SOCK_DGRAM) {
1256 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1257 				rep->r_rexmit = NFS_MAXREXMIT;
1258 			continue;
1259 		}
1260 		if ((so = nmp->nm_so) == NULL)
1261 			continue;
1262 
1263 		/*
1264 		 * If there is enough space and the window allows..
1265 		 *	Resend it
1266 		 * Set r_rtt to -1 in case we fail to send it now.
1267 		 */
1268 		rep->r_rtt = -1;
1269 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1270 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1271 		    (rep->r_flags & R_SENT) ||
1272 		    nmp->nm_sent < nmp->nm_cwnd) &&
1273 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1274 			if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
1275 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1276 			    (struct mbuf *)0, (struct mbuf *)0);
1277 			else
1278 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1279 			    nmp->nm_nam, (struct mbuf *)0);
1280 			if (error) {
1281 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
1282 					so->so_error = 0;
1283 			} else {
1284 				/*
1285 				 * Iff first send, start timing
1286 				 * else turn timing off, backoff timer
1287 				 * and divide congestion window by 2.
1288 				 */
1289 				if (rep->r_flags & R_SENT) {
1290 					rep->r_flags &= ~R_TIMING;
1291 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1292 						rep->r_rexmit = NFS_MAXREXMIT;
1293 					nmp->nm_cwnd >>= 1;
1294 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1295 						nmp->nm_cwnd = NFS_CWNDSCALE;
1296 					nfsstats.rpcretries++;
1297 				} else {
1298 					rep->r_flags |= R_SENT;
1299 					nmp->nm_sent += NFS_CWNDSCALE;
1300 				}
1301 				rep->r_rtt = 0;
1302 			}
1303 		}
1304 	}
1305 
1306 #ifdef NFSSERVER
1307 	/*
1308 	 * Scan the write gathering queues for writes that need to be
1309 	 * completed now.
1310 	 */
1311 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1312 	for (slp = nfssvc_sockhead.tqh_first; slp != 0;
1313 	    slp = slp->ns_chain.tqe_next) {
1314 	    if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1315 		nfsrv_wakenfsd(slp);
1316 	}
1317 #endif /* NFSSERVER */
1318 	splx(s);
1319 	timeout_add(to, nfs_ticks);
1320 }
1321 
1322 /*
1323  * Test for a termination condition pending on the process.
1324  * This is used for NFSMNT_INT mounts.
1325  */
1326 int
1327 nfs_sigintr(nmp, rep, p)
1328 	struct nfsmount *nmp;
1329 	struct nfsreq *rep;
1330 	register struct proc *p;
1331 {
1332 
1333 	if (rep && (rep->r_flags & R_SOFTTERM))
1334 		return (EINTR);
1335 	if (!(nmp->nm_flag & NFSMNT_INT))
1336 		return (0);
1337 	if (p && p->p_siglist &&
1338 	    (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) &
1339 	    NFSINT_SIGMASK))
1340 		return (EINTR);
1341 	return (0);
1342 }
1343 
1344 /*
1345  * Lock a socket against others.
1346  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1347  * and also to avoid race conditions between the processes with nfs requests
1348  * in progress when a reconnect is necessary.
1349  */
1350 int
1351 nfs_sndlock(flagp, rep)
1352 	register int *flagp;
1353 	struct nfsreq *rep;
1354 {
1355 	struct proc *p;
1356 	int slpflag = 0, slptimeo = 0;
1357 
1358 	if (rep) {
1359 		p = rep->r_procp;
1360 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1361 			slpflag = PCATCH;
1362 	} else
1363 		p = (struct proc *)0;
1364 	while (*flagp & NFSMNT_SNDLOCK) {
1365 		if (nfs_sigintr(rep->r_nmp, rep, p))
1366 			return (EINTR);
1367 		*flagp |= NFSMNT_WANTSND;
1368 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1369 			slptimeo);
1370 		if (slpflag == PCATCH) {
1371 			slpflag = 0;
1372 			slptimeo = 2 * hz;
1373 		}
1374 	}
1375 	*flagp |= NFSMNT_SNDLOCK;
1376 	return (0);
1377 }
1378 
1379 /*
1380  * Unlock the stream socket for others.
1381  */
1382 void
1383 nfs_sndunlock(flagp)
1384 	register int *flagp;
1385 {
1386 
1387 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1388 		panic("nfs sndunlock");
1389 	*flagp &= ~NFSMNT_SNDLOCK;
1390 	if (*flagp & NFSMNT_WANTSND) {
1391 		*flagp &= ~NFSMNT_WANTSND;
1392 		wakeup((caddr_t)flagp);
1393 	}
1394 }
1395 
1396 int
1397 nfs_rcvlock(rep)
1398 	register struct nfsreq *rep;
1399 {
1400 	register int *flagp = &rep->r_nmp->nm_flag;
1401 	int slpflag, slptimeo = 0;
1402 
1403 	if (*flagp & NFSMNT_INT)
1404 		slpflag = PCATCH;
1405 	else
1406 		slpflag = 0;
1407 	while (*flagp & NFSMNT_RCVLOCK) {
1408 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1409 			return (EINTR);
1410 		*flagp |= NFSMNT_WANTRCV;
1411 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1412 			slptimeo);
1413 		if (slpflag == PCATCH) {
1414 			slpflag = 0;
1415 			slptimeo = 2 * hz;
1416 		}
1417 	}
1418 	*flagp |= NFSMNT_RCVLOCK;
1419 	return (0);
1420 }
1421 
1422 /*
1423  * Unlock the stream socket for others.
1424  */
1425 void
1426 nfs_rcvunlock(flagp)
1427 	register int *flagp;
1428 {
1429 
1430 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1431 		panic("nfs rcvunlock");
1432 	*flagp &= ~NFSMNT_RCVLOCK;
1433 	if (*flagp & NFSMNT_WANTRCV) {
1434 		*flagp &= ~NFSMNT_WANTRCV;
1435 		wakeup((caddr_t)flagp);
1436 	}
1437 }
1438 
1439 /*
1440  * Check for badly aligned mbuf data areas and
1441  * realign data in an mbuf list by copying the data areas up, as required.
1442  */
1443 void
1444 nfs_realign(m, hsiz)
1445 	register struct mbuf *m;
1446 	int hsiz;
1447 {
1448 	register struct mbuf *m2;
1449 	register int siz, mlen, olen;
1450 	register caddr_t tcp, fcp;
1451 	struct mbuf *mnew;
1452 
1453 	while (m) {
1454 	    /*
1455 	     * This never happens for UDP, rarely happens for TCP
1456 	     * but frequently happens for iso transport.
1457 	     */
1458 	    if ((m->m_len & 0x3) || (mtod(m, long) & 0x3)) {
1459 		olen = m->m_len;
1460 		fcp = mtod(m, caddr_t);
1461 		if ((long)fcp & 0x3) {
1462 		        if (m->m_flags & M_PKTHDR)
1463 				m_tag_delete_chain(m, NULL);
1464 			m->m_flags &= ~M_PKTHDR;
1465 			if (m->m_flags & M_EXT)
1466 				m->m_data = m->m_ext.ext_buf +
1467 					((m->m_ext.ext_size - olen) & ~0x3);
1468 			else
1469 				m->m_data = m->m_dat;
1470 		}
1471 		m->m_len = 0;
1472 		tcp = mtod(m, caddr_t);
1473 		mnew = m;
1474 		m2 = m->m_next;
1475 
1476 		/*
1477 		 * If possible, only put the first invariant part
1478 		 * of the RPC header in the first mbuf.
1479 		 */
1480 		mlen = M_TRAILINGSPACE(m);
1481 		if (olen <= hsiz && mlen > hsiz)
1482 			mlen = hsiz;
1483 
1484 		/* Loop through the mbuf list consolidating data. */
1485 		while (m) {
1486 			while (olen > 0) {
1487 				if (mlen == 0) {
1488 				        if (m2->m_flags & M_PKTHDR)
1489 						m_tag_delete_chain(m2, NULL);
1490 					m2->m_flags &= ~M_PKTHDR;
1491 					if (m2->m_flags & M_EXT)
1492 						m2->m_data = m2->m_ext.ext_buf;
1493 					else
1494 						m2->m_data = m2->m_dat;
1495 					m2->m_len = 0;
1496 					mlen = M_TRAILINGSPACE(m2);
1497 					tcp = mtod(m2, caddr_t);
1498 					mnew = m2;
1499 					m2 = m2->m_next;
1500 				}
1501 				siz = min(mlen, olen);
1502 				if (tcp != fcp)
1503 					bcopy(fcp, tcp, siz);
1504 				mnew->m_len += siz;
1505 				mlen -= siz;
1506 				olen -= siz;
1507 				tcp += siz;
1508 				fcp += siz;
1509 			}
1510 			m = m->m_next;
1511 			if (m) {
1512 				olen = m->m_len;
1513 				fcp = mtod(m, caddr_t);
1514 			}
1515 		}
1516 
1517 		/*
1518 		 * Finally, set m_len == 0 for any trailing mbufs that have
1519 		 * been copied out of.
1520 		 */
1521 		while (m2) {
1522 			m2->m_len = 0;
1523 			m2 = m2->m_next;
1524 		}
1525 		return;
1526 	    }
1527 	    m = m->m_next;
1528 	}
1529 }
1530 
1531 /*
1532  * Parse an RPC request
1533  * - verify it
1534  * - fill in the cred struct.
1535  */
1536 int
1537 nfs_getreq(nd, nfsd, has_header)
1538 	register struct nfsrv_descript *nd;
1539 	struct nfsd *nfsd;
1540 	int has_header;
1541 {
1542 	register int len, i;
1543 	register u_int32_t *tl;
1544 	register int32_t t1;
1545 	struct uio uio;
1546 	struct iovec iov;
1547 	caddr_t dpos, cp2, cp;
1548 	u_int32_t nfsvers, auth_type;
1549 	uid_t nickuid;
1550 	int error = 0, ticklen;
1551 	struct mbuf *mrep, *md;
1552 	register struct nfsuid *nuidp;
1553 	struct timeval tvin, tvout;
1554 
1555 	mrep = nd->nd_mrep;
1556 	md = nd->nd_md;
1557 	dpos = nd->nd_dpos;
1558 	if (has_header) {
1559 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1560 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1561 		if (*tl++ != rpc_call) {
1562 			m_freem(mrep);
1563 			return (EBADRPC);
1564 		}
1565 	} else
1566 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1567 	nd->nd_repstat = 0;
1568 	nd->nd_flag = 0;
1569 	if (*tl++ != rpc_vers) {
1570 		nd->nd_repstat = ERPCMISMATCH;
1571 		nd->nd_procnum = NFSPROC_NOOP;
1572 		return (0);
1573 	}
1574 	if (*tl != nfs_prog) {
1575 		nd->nd_repstat = EPROGUNAVAIL;
1576 		nd->nd_procnum = NFSPROC_NOOP;
1577 		return (0);
1578 	}
1579 	tl++;
1580 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1581 	if (nfsvers != NFS_VER2 && nfsvers != NFS_VER3) {
1582 		nd->nd_repstat = EPROGMISMATCH;
1583 		nd->nd_procnum = NFSPROC_NOOP;
1584 		return (0);
1585 	}
1586 	if (nfsvers == NFS_VER3)
1587 		nd->nd_flag = ND_NFSV3;
1588 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1589 	if (nd->nd_procnum == NFSPROC_NULL)
1590 		return (0);
1591 	if (nd->nd_procnum >= NFS_NPROCS ||
1592 		(nd->nd_procnum > NFSPROC_COMMIT) ||
1593 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1594 		nd->nd_repstat = EPROCUNAVAIL;
1595 		nd->nd_procnum = NFSPROC_NOOP;
1596 		return (0);
1597 	}
1598 	if ((nd->nd_flag & ND_NFSV3) == 0)
1599 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1600 	auth_type = *tl++;
1601 	len = fxdr_unsigned(int, *tl++);
1602 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1603 		m_freem(mrep);
1604 		return (EBADRPC);
1605 	}
1606 
1607 	nd->nd_flag &= ~ND_KERBAUTH;
1608 	/*
1609 	 * Handle auth_unix or auth_kerb.
1610 	 */
1611 	if (auth_type == rpc_auth_unix) {
1612 		len = fxdr_unsigned(int, *++tl);
1613 		if (len < 0 || len > NFS_MAXNAMLEN) {
1614 			m_freem(mrep);
1615 			return (EBADRPC);
1616 		}
1617 		nfsm_adv(nfsm_rndup(len));
1618 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1619 		bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred));
1620 		nd->nd_cr.cr_ref = 1;
1621 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1622 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1623 		len = fxdr_unsigned(int, *tl);
1624 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1625 			m_freem(mrep);
1626 			return (EBADRPC);
1627 		}
1628 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1629 		for (i = 0; i < len; i++)
1630 		    if (i < NGROUPS)
1631 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1632 		    else
1633 			tl++;
1634 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1635 		if (nd->nd_cr.cr_ngroups > 1)
1636 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1637 		len = fxdr_unsigned(int, *++tl);
1638 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1639 			m_freem(mrep);
1640 			return (EBADRPC);
1641 		}
1642 		if (len > 0)
1643 			nfsm_adv(nfsm_rndup(len));
1644 	} else if (auth_type == rpc_auth_kerb) {
1645 		switch (fxdr_unsigned(int, *tl++)) {
1646 		case RPCAKN_FULLNAME:
1647 			ticklen = fxdr_unsigned(int, *tl);
1648 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1649 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1650 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1651 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1652 				m_freem(mrep);
1653 				return (EBADRPC);
1654 			}
1655 			uio.uio_offset = 0;
1656 			uio.uio_iov = &iov;
1657 			uio.uio_iovcnt = 1;
1658 			uio.uio_segflg = UIO_SYSSPACE;
1659 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1660 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1661 			nfsm_mtouio(&uio, uio.uio_resid);
1662 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1663 			if (*tl++ != rpc_auth_kerb ||
1664 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1665 				printf("Bad kerb verifier\n");
1666 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1667 				nd->nd_procnum = NFSPROC_NOOP;
1668 				return (0);
1669 			}
1670 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1671 			tl = (u_int32_t *)cp;
1672 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1673 				printf("Not fullname kerb verifier\n");
1674 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1675 				nd->nd_procnum = NFSPROC_NOOP;
1676 				return (0);
1677 			}
1678 			cp += NFSX_UNSIGNED;
1679 			bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED);
1680 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1681 			nd->nd_flag |= ND_KERBFULL;
1682 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1683 			break;
1684 		case RPCAKN_NICKNAME:
1685 			if (len != 2 * NFSX_UNSIGNED) {
1686 				printf("Kerb nickname short\n");
1687 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1688 				nd->nd_procnum = NFSPROC_NOOP;
1689 				return (0);
1690 			}
1691 			nickuid = fxdr_unsigned(uid_t, *tl);
1692 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1693 			if (*tl++ != rpc_auth_kerb ||
1694 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1695 				printf("Kerb nick verifier bad\n");
1696 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1697 				nd->nd_procnum = NFSPROC_NOOP;
1698 				return (0);
1699 			}
1700 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1701 			tvin.tv_sec = *tl++;
1702 			tvin.tv_usec = *tl;
1703 
1704 			for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1705 			    nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1706 				if (nuidp->nu_cr.cr_uid == nickuid &&
1707 				    (!nd->nd_nam2 ||
1708 				     netaddr_match(NU_NETFAM(nuidp),
1709 				      &nuidp->nu_haddr, nd->nd_nam2)))
1710 					break;
1711 			}
1712 			if (!nuidp) {
1713 				nd->nd_repstat =
1714 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1715 				nd->nd_procnum = NFSPROC_NOOP;
1716 				return (0);
1717 			}
1718 
1719 			/*
1720 			 * Now, decrypt the timestamp using the session key
1721 			 * and validate it.
1722 			 */
1723 #ifdef NFSKERB
1724 			XXX
1725 #endif
1726 
1727 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1728 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1729 			if (nuidp->nu_expire < time.tv_sec ||
1730 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1731 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1732 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1733 				nuidp->nu_expire = 0;
1734 				nd->nd_repstat =
1735 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1736 				nd->nd_procnum = NFSPROC_NOOP;
1737 				return (0);
1738 			}
1739 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1740 			nd->nd_flag |= ND_KERBNICK;
1741 		};
1742 	} else {
1743 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1744 		nd->nd_procnum = NFSPROC_NOOP;
1745 		return (0);
1746 	}
1747 
1748 	nd->nd_md = md;
1749 	nd->nd_dpos = dpos;
1750 	return (0);
1751 nfsmout:
1752 	return (error);
1753 }
1754 
1755 int
1756 nfs_msg(p, server, msg)
1757 	struct proc *p;
1758 	char *server, *msg;
1759 {
1760 	tpr_t tpr;
1761 
1762 	if (p)
1763 		tpr = tprintf_open(p);
1764 	else
1765 		tpr = NULL;
1766 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1767 	tprintf_close(tpr);
1768 	return (0);
1769 }
1770 
1771 #ifdef NFSSERVER
1772 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
1773 				    struct nfssvc_sock *, struct proc *,
1774 				    struct mbuf **)) = {
1775 	nfsrv_null,
1776 	nfsrv_getattr,
1777 	nfsrv_setattr,
1778 	nfsrv_lookup,
1779 	nfsrv3_access,
1780 	nfsrv_readlink,
1781 	nfsrv_read,
1782 	nfsrv_write,
1783 	nfsrv_create,
1784 	nfsrv_mkdir,
1785 	nfsrv_symlink,
1786 	nfsrv_mknod,
1787 	nfsrv_remove,
1788 	nfsrv_rmdir,
1789 	nfsrv_rename,
1790 	nfsrv_link,
1791 	nfsrv_readdir,
1792 	nfsrv_readdirplus,
1793 	nfsrv_statfs,
1794 	nfsrv_fsinfo,
1795 	nfsrv_pathconf,
1796 	nfsrv_commit,
1797 	nfsrv_noop,
1798 	nfsrv_noop,
1799 	nfsrv_noop,
1800 	nfsrv_noop
1801 };
1802 
1803 /*
1804  * Socket upcall routine for the nfsd sockets.
1805  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1806  * Essentially do as much as possible non-blocking, else punt and it will
1807  * be called with M_WAIT from an nfsd.
1808  */
1809 void
1810 nfsrv_rcv(so, arg, waitflag)
1811 	struct socket *so;
1812 	caddr_t arg;
1813 	int waitflag;
1814 {
1815 	register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1816 	register struct mbuf *m;
1817 	struct mbuf *mp, *nam;
1818 	struct uio auio;
1819 	int flags, error;
1820 
1821 	if ((slp->ns_flag & SLP_VALID) == 0)
1822 		return;
1823 #ifdef notdef
1824 	/*
1825 	 * Define this to test for nfsds handling this under heavy load.
1826 	 */
1827 	if (waitflag == M_DONTWAIT) {
1828 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1829 	}
1830 #endif
1831 	auio.uio_procp = NULL;
1832 	if (so->so_type == SOCK_STREAM) {
1833 		/*
1834 		 * If there are already records on the queue, defer soreceive()
1835 		 * to an nfsd so that there is feedback to the TCP layer that
1836 		 * the nfs servers are heavily loaded.
1837 		 */
1838 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1839 			slp->ns_flag |= SLP_NEEDQ;
1840 			goto dorecs;
1841 		}
1842 
1843 		/*
1844 		 * Do soreceive().
1845 		 */
1846 		auio.uio_resid = 1000000000;
1847 		flags = MSG_DONTWAIT;
1848 		error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1849 		if (error || mp == (struct mbuf *)0) {
1850 			if (error == EWOULDBLOCK)
1851 				slp->ns_flag |= SLP_NEEDQ;
1852 			else
1853 				slp->ns_flag |= SLP_DISCONN;
1854 			goto dorecs;
1855 		}
1856 		m = mp;
1857 		if (slp->ns_rawend) {
1858 			slp->ns_rawend->m_next = m;
1859 			slp->ns_cc += 1000000000 - auio.uio_resid;
1860 		} else {
1861 			slp->ns_raw = m;
1862 			slp->ns_cc = 1000000000 - auio.uio_resid;
1863 		}
1864 		while (m->m_next)
1865 			m = m->m_next;
1866 		slp->ns_rawend = m;
1867 
1868 		/*
1869 		 * Now try and parse record(s) out of the raw stream data.
1870 		 */
1871 		error = nfsrv_getstream(slp, waitflag);
1872 		if (error) {
1873 			if (error == EPERM)
1874 				slp->ns_flag |= SLP_DISCONN;
1875 			else
1876 				slp->ns_flag |= SLP_NEEDQ;
1877 		}
1878 	} else {
1879 		do {
1880 			auio.uio_resid = 1000000000;
1881 			flags = MSG_DONTWAIT;
1882 			error = soreceive(so, &nam, &auio, &mp,
1883 						(struct mbuf **)0, &flags);
1884 			if (mp) {
1885 				nfs_realign(mp, 10 * NFSX_UNSIGNED);
1886 				if (nam) {
1887 					m = nam;
1888 					m->m_next = mp;
1889 				} else
1890 					m = mp;
1891 				if (slp->ns_recend)
1892 					slp->ns_recend->m_nextpkt = m;
1893 				else
1894 					slp->ns_rec = m;
1895 				slp->ns_recend = m;
1896 				m->m_nextpkt = (struct mbuf *)0;
1897 			}
1898 			if (error) {
1899 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
1900 					&& error != EWOULDBLOCK) {
1901 					slp->ns_flag |= SLP_DISCONN;
1902 					goto dorecs;
1903 				}
1904 			}
1905 		} while (mp);
1906 	}
1907 
1908 	/*
1909 	 * Now try and process the request records, non-blocking.
1910 	 */
1911 dorecs:
1912 	if (waitflag == M_DONTWAIT &&
1913 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
1914 		nfsrv_wakenfsd(slp);
1915 }
1916 
1917 /*
1918  * Try and extract an RPC request from the mbuf data list received on a
1919  * stream socket. The "waitflag" argument indicates whether or not it
1920  * can sleep.
1921  */
1922 int
1923 nfsrv_getstream(slp, waitflag)
1924 	register struct nfssvc_sock *slp;
1925 	int waitflag;
1926 {
1927 	register struct mbuf *m, **mpp;
1928 	register char *cp1, *cp2;
1929 	register int len;
1930 	struct mbuf *om, *m2, *recm = NULL;
1931 	u_int32_t recmark;
1932 
1933 	if (slp->ns_flag & SLP_GETSTREAM)
1934 		panic("nfs getstream");
1935 	slp->ns_flag |= SLP_GETSTREAM;
1936 	for (;;) {
1937 	    if (slp->ns_reclen == 0) {
1938 		if (slp->ns_cc < NFSX_UNSIGNED) {
1939 			slp->ns_flag &= ~SLP_GETSTREAM;
1940 			return (0);
1941 		}
1942 		m = slp->ns_raw;
1943 		if (m->m_len >= NFSX_UNSIGNED) {
1944 			bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
1945 			m->m_data += NFSX_UNSIGNED;
1946 			m->m_len -= NFSX_UNSIGNED;
1947 		} else {
1948 			cp1 = (caddr_t)&recmark;
1949 			cp2 = mtod(m, caddr_t);
1950 			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
1951 				while (m->m_len == 0) {
1952 					m = m->m_next;
1953 					cp2 = mtod(m, caddr_t);
1954 				}
1955 				*cp1++ = *cp2++;
1956 				m->m_data++;
1957 				m->m_len--;
1958 			}
1959 		}
1960 		slp->ns_cc -= NFSX_UNSIGNED;
1961 		recmark = ntohl(recmark);
1962 		slp->ns_reclen = recmark & ~0x80000000;
1963 		if (recmark & 0x80000000)
1964 			slp->ns_flag |= SLP_LASTFRAG;
1965 		else
1966 			slp->ns_flag &= ~SLP_LASTFRAG;
1967 		if (slp->ns_reclen > NFS_MAXPACKET) {
1968 			slp->ns_flag &= ~SLP_GETSTREAM;
1969 			return (EPERM);
1970 		}
1971 	    }
1972 
1973 	    /*
1974 	     * Now get the record part.
1975 	     */
1976 	    if (slp->ns_cc == slp->ns_reclen) {
1977 		recm = slp->ns_raw;
1978 		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
1979 		slp->ns_cc = slp->ns_reclen = 0;
1980 	    } else if (slp->ns_cc > slp->ns_reclen) {
1981 		len = 0;
1982 		m = slp->ns_raw;
1983 		om = (struct mbuf *)0;
1984 		while (len < slp->ns_reclen) {
1985 			if ((len + m->m_len) > slp->ns_reclen) {
1986 				m2 = m_copym(m, 0, slp->ns_reclen - len,
1987 					waitflag);
1988 				if (m2) {
1989 					if (om) {
1990 						om->m_next = m2;
1991 						recm = slp->ns_raw;
1992 					} else
1993 						recm = m2;
1994 					m->m_data += slp->ns_reclen - len;
1995 					m->m_len -= slp->ns_reclen - len;
1996 					len = slp->ns_reclen;
1997 				} else {
1998 					slp->ns_flag &= ~SLP_GETSTREAM;
1999 					return (EWOULDBLOCK);
2000 				}
2001 			} else if ((len + m->m_len) == slp->ns_reclen) {
2002 				om = m;
2003 				len += m->m_len;
2004 				m = m->m_next;
2005 				recm = slp->ns_raw;
2006 				om->m_next = (struct mbuf *)0;
2007 			} else {
2008 				om = m;
2009 				len += m->m_len;
2010 				m = m->m_next;
2011 			}
2012 		}
2013 		slp->ns_raw = m;
2014 		slp->ns_cc -= len;
2015 		slp->ns_reclen = 0;
2016 	    } else {
2017 		slp->ns_flag &= ~SLP_GETSTREAM;
2018 		return (0);
2019 	    }
2020 
2021 	    /*
2022 	     * Accumulate the fragments into a record.
2023 	     */
2024 	    mpp = &slp->ns_frag;
2025 	    while (*mpp)
2026 		mpp = &((*mpp)->m_next);
2027 	    *mpp = recm;
2028 	    if (slp->ns_flag & SLP_LASTFRAG) {
2029 		nfs_realign(slp->ns_frag, 10 * NFSX_UNSIGNED);
2030 		if (slp->ns_recend)
2031 		    slp->ns_recend->m_nextpkt = slp->ns_frag;
2032 		else
2033 		    slp->ns_rec = slp->ns_frag;
2034 		slp->ns_recend = slp->ns_frag;
2035 		slp->ns_frag = (struct mbuf *)0;
2036 	    }
2037 	}
2038 }
2039 
2040 /*
2041  * Parse an RPC header.
2042  */
2043 int
2044 nfsrv_dorec(slp, nfsd, ndp)
2045 	register struct nfssvc_sock *slp;
2046 	struct nfsd *nfsd;
2047 	struct nfsrv_descript **ndp;
2048 {
2049 	register struct mbuf *m, *nam;
2050 	register struct nfsrv_descript *nd;
2051 	int error;
2052 
2053 	*ndp = NULL;
2054 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2055 	    (m = slp->ns_rec) == (struct mbuf *)0)
2056 		return (ENOBUFS);
2057 	slp->ns_rec = m->m_nextpkt;
2058 	if (slp->ns_rec)
2059 		m->m_nextpkt = (struct mbuf *)0;
2060 	else
2061 		slp->ns_recend = (struct mbuf *)0;
2062 	if (m->m_type == MT_SONAME) {
2063 		nam = m;
2064 		m = m->m_next;
2065 		nam->m_next = NULL;
2066 	} else
2067 		nam = NULL;
2068 	MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2069 		M_NFSRVDESC, M_WAITOK);
2070 	nd->nd_md = nd->nd_mrep = m;
2071 	nd->nd_nam2 = nam;
2072 	nd->nd_dpos = mtod(m, caddr_t);
2073 	error = nfs_getreq(nd, nfsd, TRUE);
2074 	if (error) {
2075 		m_freem(nam);
2076 		free((caddr_t)nd, M_NFSRVDESC);
2077 		return (error);
2078 	}
2079 	*ndp = nd;
2080 	nfsd->nfsd_nd = nd;
2081 	return (0);
2082 }
2083 
2084 
2085 /*
2086  * Search for a sleeping nfsd and wake it up.
2087  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2088  * running nfsds will go look for the work in the nfssvc_sock list.
2089  */
2090 void
2091 nfsrv_wakenfsd(slp)
2092 	struct nfssvc_sock *slp;
2093 {
2094 	register struct nfsd *nd;
2095 
2096 	if ((slp->ns_flag & SLP_VALID) == 0)
2097 		return;
2098 	for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) {
2099 		if (nd->nfsd_flag & NFSD_WAITING) {
2100 			nd->nfsd_flag &= ~NFSD_WAITING;
2101 			if (nd->nfsd_slp)
2102 				panic("nfsd wakeup");
2103 			slp->ns_sref++;
2104 			nd->nfsd_slp = slp;
2105 			wakeup((caddr_t)nd);
2106 			return;
2107 		}
2108 	}
2109 	slp->ns_flag |= SLP_DOREC;
2110 	nfsd_head_flag |= NFSD_CHECKSLP;
2111 }
2112 #endif /* NFSSERVER */
2113