xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 481fca6e59249d8ffcf24fef7cfbe7b131bfb080)
1 /*	$NetBSD: nfs_socket.c,v 1.58 2000/06/27 17:52:33 mrg Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1991, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)nfs_socket.c	8.5 (Berkeley) 3/30/95
39  */
40 
41 /*
42  * Socket operations for use by nfs
43  */
44 
45 #include "fs_nfs.h"
46 #include "opt_nfsserver.h"
47 #include "opt_inet.h"
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/callout.h>
52 #include <sys/proc.h>
53 #include <sys/mount.h>
54 #include <sys/kernel.h>
55 #include <sys/mbuf.h>
56 #include <sys/vnode.h>
57 #include <sys/domain.h>
58 #include <sys/protosw.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/syslog.h>
62 #include <sys/tprintf.h>
63 #include <sys/namei.h>
64 #include <sys/signal.h>
65 #include <sys/signalvar.h>
66 
67 #include <netinet/in.h>
68 #include <netinet/tcp.h>
69 
70 #include <nfs/rpcv2.h>
71 #include <nfs/nfsproto.h>
72 #include <nfs/nfs.h>
73 #include <nfs/xdr_subs.h>
74 #include <nfs/nfsm_subs.h>
75 #include <nfs/nfsmount.h>
76 #include <nfs/nfsnode.h>
77 #include <nfs/nfsrtt.h>
78 #include <nfs/nqnfs.h>
79 #include <nfs/nfs_var.h>
80 
81 #define	TRUE	1
82 #define	FALSE	0
83 
84 /*
85  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
86  * Use the mean and mean deviation of rtt for the appropriate type of rpc
87  * for the frequent rpcs and a default for the others.
88  * The justification for doing "other" this way is that these rpcs
89  * happen so infrequently that timer est. would probably be stale.
90  * Also, since many of these rpcs are
91  * non-idempotent, a conservative timeout is desired.
92  * getattr, lookup - A+2D
93  * read, write     - A+4D
94  * other           - nm_timeo
95  */
96 #define	NFS_RTO(n, t) \
97 	((t) == 0 ? (n)->nm_timeo : \
98 	 ((t) < 3 ? \
99 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
100 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
101 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
102 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
103 /*
104  * External data, mostly RPC constants in XDR form
105  */
106 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
107 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
108 	rpc_auth_kerb;
109 extern u_int32_t nfs_prog, nqnfs_prog;
110 extern time_t nqnfsstarttime;
111 extern struct nfsstats nfsstats;
112 extern int nfsv3_procid[NFS_NPROCS];
113 extern int nfs_ticks;
114 
115 /*
116  * Defines which timer to use for the procnum.
117  * 0 - default
118  * 1 - getattr
119  * 2 - lookup
120  * 3 - read
121  * 4 - write
122  */
123 static int proct[NFS_NPROCS] = {
124 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
125 	0, 0, 0,
126 };
127 
128 /*
129  * There is a congestion window for outstanding rpcs maintained per mount
130  * point. The cwnd size is adjusted in roughly the way that:
131  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
132  * SIGCOMM '88". ACM, August 1988.
133  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
134  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
135  * of rpcs is in progress.
136  * (The sent count and cwnd are scaled for integer arith.)
137  * Variants of "slow start" were tried and were found to be too much of a
138  * performance hit (ave. rtt 3 times larger),
139  * I suspect due to the large rtt that nfs rpcs have.
140  */
141 #define	NFS_CWNDSCALE	256
142 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
143 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
144 int nfsrtton = 0;
145 struct nfsrtt nfsrtt;
146 
147 struct callout nfs_timer_ch = CALLOUT_INITIALIZER;
148 
149 /*
150  * Initialize sockets and congestion for a new NFS connection.
151  * We do not free the sockaddr if error.
152  */
153 int
154 nfs_connect(nmp, rep)
155 	struct nfsmount *nmp;
156 	struct nfsreq *rep;
157 {
158 	struct socket *so;
159 	int s, error, rcvreserve, sndreserve;
160 	struct sockaddr *saddr;
161 	struct sockaddr_in *sin;
162 #ifdef INET6
163 	struct sockaddr_in6 *sin6;
164 #endif
165 	struct mbuf *m;
166 	u_int16_t tport;
167 
168 	nmp->nm_so = (struct socket *)0;
169 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
170 	error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
171 		nmp->nm_soproto);
172 	if (error)
173 		goto bad;
174 	so = nmp->nm_so;
175 	nmp->nm_soflags = so->so_proto->pr_flags;
176 
177 	/*
178 	 * Some servers require that the client port be a reserved port number.
179 	 */
180 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
181 		MGET(m, M_WAIT, MT_SONAME);
182 		sin = mtod(m, struct sockaddr_in *);
183 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
184 		sin->sin_family = AF_INET;
185 		sin->sin_addr.s_addr = INADDR_ANY;
186 		tport = IPPORT_RESERVED - 1;
187 		sin->sin_port = htons(tport);
188 		while ((error = sobind(so, m)) == EADDRINUSE &&
189 		       --tport > IPPORT_RESERVED / 2)
190 			sin->sin_port = htons(tport);
191 		m_freem(m);
192 		if (error)
193 			goto bad;
194 	}
195 #ifdef INET6
196 	if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) {
197 		MGET(m, M_WAIT, MT_SONAME);
198 		sin6 = mtod(m, struct sockaddr_in6 *);
199 		sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6);
200 		sin6->sin6_family = AF_INET6;
201 		sin6->sin6_addr = in6addr_any;
202 		tport = IPV6PORT_RESERVED - 1;
203 		sin6->sin6_port = htons(tport);
204 		while ((error = sobind(so, m)) == EADDRINUSE &&
205 		       --tport > IPV6PORT_RESERVED / 2)
206 			sin6->sin6_port = htons(tport);
207 		m_freem(m);
208 		if (error)
209 			goto bad;
210 	}
211 #endif
212 
213 	/*
214 	 * Protocols that do not require connections may be optionally left
215 	 * unconnected for servers that reply from a port other than NFS_PORT.
216 	 */
217 	if (nmp->nm_flag & NFSMNT_NOCONN) {
218 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
219 			error = ENOTCONN;
220 			goto bad;
221 		}
222 	} else {
223 		error = soconnect(so, nmp->nm_nam);
224 		if (error)
225 			goto bad;
226 
227 		/*
228 		 * Wait for the connection to complete. Cribbed from the
229 		 * connect system call but with the wait timing out so
230 		 * that interruptible mounts don't hang here for a long time.
231 		 */
232 		s = splsoftnet();
233 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
234 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
235 				"nfscn1", 2 * hz);
236 			if ((so->so_state & SS_ISCONNECTING) &&
237 			    so->so_error == 0 && rep &&
238 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
239 				so->so_state &= ~SS_ISCONNECTING;
240 				splx(s);
241 				goto bad;
242 			}
243 		}
244 		if (so->so_error) {
245 			error = so->so_error;
246 			so->so_error = 0;
247 			splx(s);
248 			goto bad;
249 		}
250 		splx(s);
251 	}
252 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
253 		so->so_rcv.sb_timeo = (5 * hz);
254 		so->so_snd.sb_timeo = (5 * hz);
255 	} else {
256 		so->so_rcv.sb_timeo = 0;
257 		so->so_snd.sb_timeo = 0;
258 	}
259 	if (nmp->nm_sotype == SOCK_DGRAM) {
260 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
261 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
262 		    NFS_MAXPKTHDR) * 2;
263 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
264 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
265 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
266 		    NFS_MAXPKTHDR) * 2;
267 	} else {
268 		if (nmp->nm_sotype != SOCK_STREAM)
269 			panic("nfscon sotype");
270 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
271 			MGET(m, M_WAIT, MT_SOOPTS);
272 			*mtod(m, int32_t *) = 1;
273 			m->m_len = sizeof(int32_t);
274 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
275 		}
276 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
277 			MGET(m, M_WAIT, MT_SOOPTS);
278 			*mtod(m, int32_t *) = 1;
279 			m->m_len = sizeof(int32_t);
280 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
281 		}
282 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
283 		    sizeof (u_int32_t)) * 2;
284 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
285 		    sizeof (u_int32_t)) * 2;
286 	}
287 	error = soreserve(so, sndreserve, rcvreserve);
288 	if (error)
289 		goto bad;
290 	so->so_rcv.sb_flags |= SB_NOINTR;
291 	so->so_snd.sb_flags |= SB_NOINTR;
292 
293 	/* Initialize other non-zero congestion variables */
294 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
295 		nmp->nm_srtt[4] = (NFS_TIMEO << 3);
296 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
297 		nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0;
298 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
299 	nmp->nm_sent = 0;
300 	nmp->nm_timeouts = 0;
301 	return (0);
302 
303 bad:
304 	nfs_disconnect(nmp);
305 	return (error);
306 }
307 
308 /*
309  * Reconnect routine:
310  * Called when a connection is broken on a reliable protocol.
311  * - clean up the old socket
312  * - nfs_connect() again
313  * - set R_MUSTRESEND for all outstanding requests on mount point
314  * If this fails the mount point is DEAD!
315  * nb: Must be called with the nfs_sndlock() set on the mount point.
316  */
317 int
318 nfs_reconnect(rep)
319 	struct nfsreq *rep;
320 {
321 	struct nfsreq *rp;
322 	struct nfsmount *nmp = rep->r_nmp;
323 	int error;
324 
325 	nfs_disconnect(nmp);
326 	while ((error = nfs_connect(nmp, rep)) != 0) {
327 		if (error == EINTR || error == ERESTART)
328 			return (EINTR);
329 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0);
330 	}
331 
332 	/*
333 	 * Loop through outstanding request list and fix up all requests
334 	 * on old socket.
335 	 */
336 	for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) {
337 		if (rp->r_nmp == nmp)
338 			rp->r_flags |= R_MUSTRESEND;
339 	}
340 	return (0);
341 }
342 
343 /*
344  * NFS disconnect. Clean up and unlink.
345  */
346 void
347 nfs_disconnect(nmp)
348 	struct nfsmount *nmp;
349 {
350 	struct socket *so;
351 	int drain = 0;
352 
353 	if (nmp->nm_so) {
354 		so = nmp->nm_so;
355 		nmp->nm_so = (struct socket *)0;
356 		soshutdown(so, 2);
357 		drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
358 		if (drain) {
359 			/*
360 			 * soshutdown() above should wake up the current
361 			 * listener.
362 			 * Now wake up those waiting for the recive lock, and
363 			 * wait for them to go away unhappy, to prevent *nmp
364 			 * from evaporating while they're sleeping.
365 			 */
366 			while (nmp->nm_waiters > 0) {
367 				wakeup (&nmp->nm_iflag);
368 				(void) tsleep(&nmp->nm_waiters, PVFS,
369 				    "nfsdis", 0);
370 			}
371 		}
372 		soclose(so);
373 	}
374 #ifdef DIAGNOSTIC
375 	if (drain && (nmp->nm_waiters > 0))
376 		panic("nfs_disconnect: waiters left after drain?\n");
377 #endif
378 }
379 
380 void
381 nfs_safedisconnect(nmp)
382 	struct nfsmount *nmp;
383 {
384 	struct nfsreq dummyreq;
385 
386 	memset(&dummyreq, 0, sizeof(dummyreq));
387 	dummyreq.r_nmp = nmp;
388 	nfs_rcvlock(&dummyreq); /* XXX ignored error return */
389 	nfs_disconnect(nmp);
390 	nfs_rcvunlock(&nmp->nm_iflag);
391 }
392 
393 /*
394  * This is the nfs send routine. For connection based socket types, it
395  * must be called with an nfs_sndlock() on the socket.
396  * "rep == NULL" indicates that it has been called from a server.
397  * For the client side:
398  * - return EINTR if the RPC is terminated, 0 otherwise
399  * - set R_MUSTRESEND if the send fails for any reason
400  * - do any cleanup required by recoverable socket errors (? ? ?)
401  * For the server side:
402  * - return EINTR or ERESTART if interrupted by a signal
403  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
404  * - do any cleanup required by recoverable socket errors (? ? ?)
405  */
406 int
407 nfs_send(so, nam, top, rep)
408 	struct socket *so;
409 	struct mbuf *nam;
410 	struct mbuf *top;
411 	struct nfsreq *rep;
412 {
413 	struct mbuf *sendnam;
414 	int error, soflags, flags;
415 
416 	if (rep) {
417 		if (rep->r_flags & R_SOFTTERM) {
418 			m_freem(top);
419 			return (EINTR);
420 		}
421 		if ((so = rep->r_nmp->nm_so) == NULL) {
422 			rep->r_flags |= R_MUSTRESEND;
423 			m_freem(top);
424 			return (0);
425 		}
426 		rep->r_flags &= ~R_MUSTRESEND;
427 		soflags = rep->r_nmp->nm_soflags;
428 	} else
429 		soflags = so->so_proto->pr_flags;
430 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
431 		sendnam = (struct mbuf *)0;
432 	else
433 		sendnam = nam;
434 	if (so->so_type == SOCK_SEQPACKET)
435 		flags = MSG_EOR;
436 	else
437 		flags = 0;
438 
439 	error = (*so->so_send)(so, sendnam, (struct uio *)0, top,
440 		(struct mbuf *)0, flags);
441 	if (error) {
442 		if (rep) {
443 			log(LOG_INFO, "nfs send error %d for server %s\n",error,
444 			    rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
445 			/*
446 			 * Deal with errors for the client side.
447 			 */
448 			if (rep->r_flags & R_SOFTTERM)
449 				error = EINTR;
450 			else
451 				rep->r_flags |= R_MUSTRESEND;
452 		} else
453 			log(LOG_INFO, "nfsd send error %d\n", error);
454 
455 		/*
456 		 * Handle any recoverable (soft) socket errors here. (? ? ?)
457 		 */
458 		if (error != EINTR && error != ERESTART &&
459 			error != EWOULDBLOCK && error != EPIPE)
460 			error = 0;
461 	}
462 	return (error);
463 }
464 
465 #ifdef NFS
466 /*
467  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
468  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
469  * Mark and consolidate the data into a new mbuf list.
470  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
471  *     small mbufs.
472  * For SOCK_STREAM we must be very careful to read an entire record once
473  * we have read any of it, even if the system call has been interrupted.
474  */
475 int
476 nfs_receive(rep, aname, mp)
477 	struct nfsreq *rep;
478 	struct mbuf **aname;
479 	struct mbuf **mp;
480 {
481 	struct socket *so;
482 	struct uio auio;
483 	struct iovec aio;
484 	struct mbuf *m;
485 	struct mbuf *control;
486 	u_int32_t len;
487 	struct mbuf **getnam;
488 	int error, sotype, rcvflg;
489 	struct proc *p = curproc;	/* XXX */
490 
491 	/*
492 	 * Set up arguments for soreceive()
493 	 */
494 	*mp = (struct mbuf *)0;
495 	*aname = (struct mbuf *)0;
496 	sotype = rep->r_nmp->nm_sotype;
497 
498 	/*
499 	 * For reliable protocols, lock against other senders/receivers
500 	 * in case a reconnect is necessary.
501 	 * For SOCK_STREAM, first get the Record Mark to find out how much
502 	 * more there is to get.
503 	 * We must lock the socket against other receivers
504 	 * until we have an entire rpc request/reply.
505 	 */
506 	if (sotype != SOCK_DGRAM) {
507 		error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
508 		if (error)
509 			return (error);
510 tryagain:
511 		/*
512 		 * Check for fatal errors and resending request.
513 		 */
514 		/*
515 		 * Ugh: If a reconnect attempt just happened, nm_so
516 		 * would have changed. NULL indicates a failed
517 		 * attempt that has essentially shut down this
518 		 * mount point.
519 		 */
520 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
521 			nfs_sndunlock(&rep->r_nmp->nm_iflag);
522 			return (EINTR);
523 		}
524 		so = rep->r_nmp->nm_so;
525 		if (!so) {
526 			error = nfs_reconnect(rep);
527 			if (error) {
528 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
529 				return (error);
530 			}
531 			goto tryagain;
532 		}
533 		while (rep->r_flags & R_MUSTRESEND) {
534 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
535 			nfsstats.rpcretries++;
536 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
537 			if (error) {
538 				if (error == EINTR || error == ERESTART ||
539 				    (error = nfs_reconnect(rep)) != 0) {
540 					nfs_sndunlock(&rep->r_nmp->nm_iflag);
541 					return (error);
542 				}
543 				goto tryagain;
544 			}
545 		}
546 		nfs_sndunlock(&rep->r_nmp->nm_iflag);
547 		if (sotype == SOCK_STREAM) {
548 			aio.iov_base = (caddr_t) &len;
549 			aio.iov_len = sizeof(u_int32_t);
550 			auio.uio_iov = &aio;
551 			auio.uio_iovcnt = 1;
552 			auio.uio_segflg = UIO_SYSSPACE;
553 			auio.uio_rw = UIO_READ;
554 			auio.uio_offset = 0;
555 			auio.uio_resid = sizeof(u_int32_t);
556 			auio.uio_procp = p;
557 			do {
558 			   rcvflg = MSG_WAITALL;
559 			   error = (*so->so_receive)(so, (struct mbuf **)0, &auio,
560 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
561 			   if (error == EWOULDBLOCK && rep) {
562 				if (rep->r_flags & R_SOFTTERM)
563 					return (EINTR);
564 			   }
565 			} while (error == EWOULDBLOCK);
566 			if (!error && auio.uio_resid > 0) {
567 			    /*
568 			     * Don't log a 0 byte receive; it means
569 			     * that the socket has been closed, and
570 			     * can happen during normal operation
571 			     * (forcible unmount or Solaris server).
572 			     */
573 			    if (auio.uio_resid != sizeof (u_int32_t))
574 			      log(LOG_INFO,
575 				 "short receive (%lu/%lu) from nfs server %s\n",
576 				 (u_long)sizeof(u_int32_t) - auio.uio_resid,
577 				 (u_long)sizeof(u_int32_t),
578 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
579 			    error = EPIPE;
580 			}
581 			if (error)
582 				goto errout;
583 			len = ntohl(len) & ~0x80000000;
584 			/*
585 			 * This is SERIOUS! We are out of sync with the sender
586 			 * and forcing a disconnect/reconnect is all I can do.
587 			 */
588 			if (len > NFS_MAXPACKET) {
589 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
590 				"impossible packet length",
591 				len,
592 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
593 			    error = EFBIG;
594 			    goto errout;
595 			}
596 			auio.uio_resid = len;
597 			do {
598 			    rcvflg = MSG_WAITALL;
599 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
600 				&auio, mp, (struct mbuf **)0, &rcvflg);
601 			} while (error == EWOULDBLOCK || error == EINTR ||
602 				 error == ERESTART);
603 			if (!error && auio.uio_resid > 0) {
604 			    if (len != auio.uio_resid)
605 			      log(LOG_INFO,
606 				"short receive (%lu/%d) from nfs server %s\n",
607 				(u_long)len - auio.uio_resid, len,
608 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
609 			    error = EPIPE;
610 			}
611 		} else {
612 			/*
613 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
614 			 * and soreceive() will return when it has either a
615 			 * control msg or a data msg.
616 			 * We have no use for control msg., but must grab them
617 			 * and then throw them away so we know what is going
618 			 * on.
619 			 */
620 			auio.uio_resid = len = 100000000; /* Anything Big */
621 			auio.uio_procp = p;
622 			do {
623 			    rcvflg = 0;
624 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
625 				&auio, mp, &control, &rcvflg);
626 			    if (control)
627 				m_freem(control);
628 			    if (error == EWOULDBLOCK && rep) {
629 				if (rep->r_flags & R_SOFTTERM)
630 					return (EINTR);
631 			    }
632 			} while (error == EWOULDBLOCK ||
633 				 (!error && *mp == NULL && control));
634 			if ((rcvflg & MSG_EOR) == 0)
635 				printf("Egad!!\n");
636 			if (!error && *mp == NULL)
637 				error = EPIPE;
638 			len -= auio.uio_resid;
639 		}
640 errout:
641 		if (error && error != EINTR && error != ERESTART) {
642 			m_freem(*mp);
643 			*mp = (struct mbuf *)0;
644 			if (error != EPIPE)
645 				log(LOG_INFO,
646 				    "receive error %d from nfs server %s\n",
647 				    error,
648 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
649 			error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
650 			if (!error)
651 				error = nfs_reconnect(rep);
652 			if (!error)
653 				goto tryagain;
654 			else
655 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
656 		}
657 	} else {
658 		if ((so = rep->r_nmp->nm_so) == NULL)
659 			return (EACCES);
660 		if (so->so_state & SS_ISCONNECTED)
661 			getnam = (struct mbuf **)0;
662 		else
663 			getnam = aname;
664 		auio.uio_resid = len = 1000000;
665 		auio.uio_procp = p;
666 		do {
667 			rcvflg = 0;
668 			error =  (*so->so_receive)(so, getnam, &auio, mp,
669 				(struct mbuf **)0, &rcvflg);
670 			if (error == EWOULDBLOCK &&
671 			    (rep->r_flags & R_SOFTTERM))
672 				return (EINTR);
673 		} while (error == EWOULDBLOCK);
674 		len -= auio.uio_resid;
675 		if (!error && *mp == NULL)
676 			error = EPIPE;
677 	}
678 	if (error) {
679 		m_freem(*mp);
680 		*mp = (struct mbuf *)0;
681 	}
682 	return (error);
683 }
684 
685 /*
686  * Implement receipt of reply on a socket.
687  * We must search through the list of received datagrams matching them
688  * with outstanding requests using the xid, until ours is found.
689  */
690 /* ARGSUSED */
691 int
692 nfs_reply(myrep)
693 	struct nfsreq *myrep;
694 {
695 	struct nfsreq *rep;
696 	struct nfsmount *nmp = myrep->r_nmp;
697 	int32_t t1;
698 	struct mbuf *mrep, *nam, *md;
699 	u_int32_t rxid, *tl;
700 	caddr_t dpos, cp2;
701 	int error;
702 
703 	/*
704 	 * Loop around until we get our own reply
705 	 */
706 	for (;;) {
707 		/*
708 		 * Lock against other receivers so that I don't get stuck in
709 		 * sbwait() after someone else has received my reply for me.
710 		 * Also necessary for connection based protocols to avoid
711 		 * race conditions during a reconnect.
712 		 */
713 		error = nfs_rcvlock(myrep);
714 		if (error == EALREADY)
715 			return (0);
716 		if (error)
717 			return (error);
718 		/*
719 		 * Get the next Rpc reply off the socket
720 		 */
721 		nmp->nm_waiters++;
722 		error = nfs_receive(myrep, &nam, &mrep);
723 		nfs_rcvunlock(&nmp->nm_iflag);
724 		if (error) {
725 
726 			if (nmp->nm_iflag & NFSMNT_DISMNT) {
727 				/*
728 				 * Oops, we're going away now..
729 				 */
730 				nmp->nm_waiters--;
731 				wakeup (&nmp->nm_waiters);
732 				return error;
733 			}
734 			nmp->nm_waiters--;
735 			/*
736 			 * Ignore routing errors on connectionless protocols? ?
737 			 */
738 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
739 				nmp->nm_so->so_error = 0;
740 #ifdef DEBUG
741 				printf("nfs_reply: ignoring error %d\n", error);
742 #endif
743 				if (myrep->r_flags & R_GETONEREP)
744 					return (0);
745 				continue;
746 			}
747 			return (error);
748 		}
749 		nmp->nm_waiters--;
750 		if (nam)
751 			m_freem(nam);
752 
753 		/*
754 		 * Get the xid and check that it is an rpc reply
755 		 */
756 		md = mrep;
757 		dpos = mtod(md, caddr_t);
758 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
759 		rxid = *tl++;
760 		if (*tl != rpc_reply) {
761 			if (nmp->nm_flag & NFSMNT_NQNFS) {
762 				if (nqnfs_callback(nmp, mrep, md, dpos))
763 					nfsstats.rpcinvalid++;
764 			} else {
765 				nfsstats.rpcinvalid++;
766 				m_freem(mrep);
767 			}
768 nfsmout:
769 			if (myrep->r_flags & R_GETONEREP)
770 				return (0);
771 			continue;
772 		}
773 
774 		/*
775 		 * Loop through the request list to match up the reply
776 		 * Iff no match, just drop the datagram
777 		 */
778 		for (rep = nfs_reqq.tqh_first; rep != 0;
779 		    rep = rep->r_chain.tqe_next) {
780 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
781 				/* Found it.. */
782 				rep->r_mrep = mrep;
783 				rep->r_md = md;
784 				rep->r_dpos = dpos;
785 				if (nfsrtton) {
786 					struct rttl *rt;
787 
788 					rt = &nfsrtt.rttl[nfsrtt.pos];
789 					rt->proc = rep->r_procnum;
790 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
791 					rt->sent = nmp->nm_sent;
792 					rt->cwnd = nmp->nm_cwnd;
793 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
794 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
795 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
796 					rt->tstamp = time;
797 					if (rep->r_flags & R_TIMING)
798 						rt->rtt = rep->r_rtt;
799 					else
800 						rt->rtt = 1000000;
801 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
802 				}
803 				/*
804 				 * Update congestion window.
805 				 * Do the additive increase of
806 				 * one rpc/rtt.
807 				 */
808 				if (nmp->nm_cwnd <= nmp->nm_sent) {
809 					nmp->nm_cwnd +=
810 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
811 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
812 					if (nmp->nm_cwnd > NFS_MAXCWND)
813 						nmp->nm_cwnd = NFS_MAXCWND;
814 				}
815 				rep->r_flags &= ~R_SENT;
816 				nmp->nm_sent -= NFS_CWNDSCALE;
817 				/*
818 				 * Update rtt using a gain of 0.125 on the mean
819 				 * and a gain of 0.25 on the deviation.
820 				 */
821 				if (rep->r_flags & R_TIMING) {
822 					/*
823 					 * Since the timer resolution of
824 					 * NFS_HZ is so course, it can often
825 					 * result in r_rtt == 0. Since
826 					 * r_rtt == N means that the actual
827 					 * rtt is between N+dt and N+2-dt ticks,
828 					 * add 1.
829 					 */
830 					t1 = rep->r_rtt + 1;
831 					t1 -= (NFS_SRTT(rep) >> 3);
832 					NFS_SRTT(rep) += t1;
833 					if (t1 < 0)
834 						t1 = -t1;
835 					t1 -= (NFS_SDRTT(rep) >> 2);
836 					NFS_SDRTT(rep) += t1;
837 				}
838 				nmp->nm_timeouts = 0;
839 				break;
840 			}
841 		}
842 		/*
843 		 * If not matched to a request, drop it.
844 		 * If it's mine, get out.
845 		 */
846 		if (rep == 0) {
847 			nfsstats.rpcunexpected++;
848 			m_freem(mrep);
849 		} else if (rep == myrep) {
850 			if (rep->r_mrep == NULL)
851 				panic("nfsreply nil");
852 			return (0);
853 		}
854 		if (myrep->r_flags & R_GETONEREP)
855 			return (0);
856 	}
857 }
858 
859 /*
860  * nfs_request - goes something like this
861  *	- fill in request struct
862  *	- links it into list
863  *	- calls nfs_send() for first transmit
864  *	- calls nfs_receive() to get reply
865  *	- break down rpc header and return with nfs reply pointed to
866  *	  by mrep or error
867  * nb: always frees up mreq mbuf list
868  */
869 int
870 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
871 	struct vnode *vp;
872 	struct mbuf *mrest;
873 	int procnum;
874 	struct proc *procp;
875 	struct ucred *cred;
876 	struct mbuf **mrp;
877 	struct mbuf **mdp;
878 	caddr_t *dposp;
879 {
880 	struct mbuf *m, *mrep;
881 	struct nfsreq *rep;
882 	u_int32_t *tl;
883 	int i;
884 	struct nfsmount *nmp;
885 	struct mbuf *md, *mheadend;
886 	struct nfsnode *np;
887 	char nickv[RPCX_NICKVERF];
888 	time_t reqtime, waituntil;
889 	caddr_t dpos, cp2;
890 	int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type;
891 	int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0;
892 	int verf_len, verf_type;
893 	u_int32_t xid;
894 	u_quad_t frev;
895 	char *auth_str, *verf_str;
896 	NFSKERBKEY_T key;		/* save session key */
897 
898 	nmp = VFSTONFS(vp->v_mount);
899 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
900 	rep->r_nmp = nmp;
901 	rep->r_vp = vp;
902 	rep->r_procp = procp;
903 	rep->r_procnum = procnum;
904 	i = 0;
905 	m = mrest;
906 	while (m) {
907 		i += m->m_len;
908 		m = m->m_next;
909 	}
910 	mrest_len = i;
911 
912 	/*
913 	 * Get the RPC header with authorization.
914 	 */
915 kerbauth:
916 	verf_str = auth_str = (char *)0;
917 	if (nmp->nm_flag & NFSMNT_KERB) {
918 		verf_str = nickv;
919 		verf_len = sizeof (nickv);
920 		auth_type = RPCAUTH_KERB4;
921 		memset((caddr_t)key, 0, sizeof (key));
922 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
923 			&auth_len, verf_str, verf_len)) {
924 			error = nfs_getauth(nmp, rep, cred, &auth_str,
925 				&auth_len, verf_str, &verf_len, key);
926 			if (error) {
927 				free((caddr_t)rep, M_NFSREQ);
928 				m_freem(mrest);
929 				return (error);
930 			}
931 		}
932 	} else {
933 		auth_type = RPCAUTH_UNIX;
934 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
935 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
936 			5 * NFSX_UNSIGNED;
937 	}
938 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
939 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
940 	if (auth_str)
941 		free(auth_str, M_TEMP);
942 
943 	/*
944 	 * For stream protocols, insert a Sun RPC Record Mark.
945 	 */
946 	if (nmp->nm_sotype == SOCK_STREAM) {
947 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
948 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
949 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
950 	}
951 	rep->r_mreq = m;
952 	rep->r_xid = xid;
953 tryagain:
954 	if (nmp->nm_flag & NFSMNT_SOFT)
955 		rep->r_retry = nmp->nm_retry;
956 	else
957 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
958 	rep->r_rtt = rep->r_rexmit = 0;
959 	if (proct[procnum] > 0)
960 		rep->r_flags = R_TIMING;
961 	else
962 		rep->r_flags = 0;
963 	rep->r_mrep = NULL;
964 
965 	/*
966 	 * Do the client side RPC.
967 	 */
968 	nfsstats.rpcrequests++;
969 	/*
970 	 * Chain request into list of outstanding requests. Be sure
971 	 * to put it LAST so timer finds oldest requests first.
972 	 */
973 	s = splsoftnet();
974 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
975 
976 	/* Get send time for nqnfs */
977 	reqtime = time.tv_sec;
978 
979 	/*
980 	 * If backing off another request or avoiding congestion, don't
981 	 * send this one now but let timer do it. If not timing a request,
982 	 * do it now.
983 	 */
984 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
985 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
986 		nmp->nm_sent < nmp->nm_cwnd)) {
987 		splx(s);
988 		if (nmp->nm_soflags & PR_CONNREQUIRED)
989 			error = nfs_sndlock(&nmp->nm_iflag, rep);
990 		if (!error) {
991 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
992 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep);
993 			if (nmp->nm_soflags & PR_CONNREQUIRED)
994 				nfs_sndunlock(&nmp->nm_iflag);
995 		}
996 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
997 			nmp->nm_sent += NFS_CWNDSCALE;
998 			rep->r_flags |= R_SENT;
999 		}
1000 	} else {
1001 		splx(s);
1002 		rep->r_rtt = -1;
1003 	}
1004 
1005 	/*
1006 	 * Wait for the reply from our send or the timer's.
1007 	 */
1008 	if (!error || error == EPIPE)
1009 		error = nfs_reply(rep);
1010 
1011 	/*
1012 	 * RPC done, unlink the request.
1013 	 */
1014 	s = splsoftnet();
1015 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1016 	splx(s);
1017 
1018 	/*
1019 	 * Decrement the outstanding request count.
1020 	 */
1021 	if (rep->r_flags & R_SENT) {
1022 		rep->r_flags &= ~R_SENT;	/* paranoia */
1023 		nmp->nm_sent -= NFS_CWNDSCALE;
1024 	}
1025 
1026 	/*
1027 	 * If there was a successful reply and a tprintf msg.
1028 	 * tprintf a response.
1029 	 */
1030 	if (!error && (rep->r_flags & R_TPRINTFMSG))
1031 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1032 		    "is alive again");
1033 	mrep = rep->r_mrep;
1034 	md = rep->r_md;
1035 	dpos = rep->r_dpos;
1036 	if (error) {
1037 		m_freem(rep->r_mreq);
1038 		free((caddr_t)rep, M_NFSREQ);
1039 		return (error);
1040 	}
1041 
1042 	/*
1043 	 * break down the rpc header and check if ok
1044 	 */
1045 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1046 	if (*tl++ == rpc_msgdenied) {
1047 		if (*tl == rpc_mismatch)
1048 			error = EOPNOTSUPP;
1049 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1050 			if (!failed_auth) {
1051 				failed_auth++;
1052 				mheadend->m_next = (struct mbuf *)0;
1053 				m_freem(mrep);
1054 				m_freem(rep->r_mreq);
1055 				goto kerbauth;
1056 			} else
1057 				error = EAUTH;
1058 		} else
1059 			error = EACCES;
1060 		m_freem(mrep);
1061 		m_freem(rep->r_mreq);
1062 		free((caddr_t)rep, M_NFSREQ);
1063 		return (error);
1064 	}
1065 
1066 	/*
1067 	 * Grab any Kerberos verifier, otherwise just throw it away.
1068 	 */
1069 	verf_type = fxdr_unsigned(int, *tl++);
1070 	i = fxdr_unsigned(int32_t, *tl);
1071 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1072 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1073 		if (error)
1074 			goto nfsmout;
1075 	} else if (i > 0)
1076 		nfsm_adv(nfsm_rndup(i));
1077 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1078 	/* 0 == ok */
1079 	if (*tl == 0) {
1080 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1081 		if (*tl != 0) {
1082 			error = fxdr_unsigned(int, *tl);
1083 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1084 				error == NFSERR_TRYLATER) {
1085 				m_freem(mrep);
1086 				error = 0;
1087 				waituntil = time.tv_sec + trylater_delay;
1088 				while (time.tv_sec < waituntil)
1089 					(void) tsleep((caddr_t)&lbolt,
1090 						PSOCK, "nqnfstry", 0);
1091 				trylater_delay *= nfs_backoff[trylater_cnt];
1092 				if (trylater_cnt < 7)
1093 					trylater_cnt++;
1094 				goto tryagain;
1095 			}
1096 
1097 			/*
1098 			 * If the File Handle was stale, invalidate the
1099 			 * lookup cache, just in case.
1100 			 */
1101 			if (error == ESTALE)
1102 				cache_purge(vp);
1103 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1104 				*mrp = mrep;
1105 				*mdp = md;
1106 				*dposp = dpos;
1107 				error |= NFSERR_RETERR;
1108 			} else
1109 				m_freem(mrep);
1110 			m_freem(rep->r_mreq);
1111 			free((caddr_t)rep, M_NFSREQ);
1112 			return (error);
1113 		}
1114 
1115 		/*
1116 		 * For nqnfs, get any lease in reply
1117 		 */
1118 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1119 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1120 			if (*tl) {
1121 				np = VTONFS(vp);
1122 				nqlflag = fxdr_unsigned(int, *tl);
1123 				nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1124 				cachable = fxdr_unsigned(int, *tl++);
1125 				reqtime += fxdr_unsigned(int, *tl++);
1126 				if (reqtime > time.tv_sec) {
1127 				    frev = fxdr_hyper(tl);
1128 				    nqnfs_clientlease(nmp, np, nqlflag,
1129 					cachable, reqtime, frev);
1130 				}
1131 			}
1132 		}
1133 		*mrp = mrep;
1134 		*mdp = md;
1135 		*dposp = dpos;
1136 		m_freem(rep->r_mreq);
1137 		FREE((caddr_t)rep, M_NFSREQ);
1138 		return (0);
1139 	}
1140 	m_freem(mrep);
1141 	error = EPROTONOSUPPORT;
1142 nfsmout:
1143 	m_freem(rep->r_mreq);
1144 	free((caddr_t)rep, M_NFSREQ);
1145 	return (error);
1146 }
1147 #endif /* NFS */
1148 
1149 /*
1150  * Generate the rpc reply header
1151  * siz arg. is used to decide if adding a cluster is worthwhile
1152  */
1153 int
1154 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1155 	int siz;
1156 	struct nfsrv_descript *nd;
1157 	struct nfssvc_sock *slp;
1158 	int err;
1159 	int cache;
1160 	u_quad_t *frev;
1161 	struct mbuf **mrq;
1162 	struct mbuf **mbp;
1163 	caddr_t *bposp;
1164 {
1165 	u_int32_t *tl;
1166 	struct mbuf *mreq;
1167 	caddr_t bpos;
1168 	struct mbuf *mb, *mb2;
1169 
1170 	MGETHDR(mreq, M_WAIT, MT_DATA);
1171 	mb = mreq;
1172 	/*
1173 	 * If this is a big reply, use a cluster else
1174 	 * try and leave leading space for the lower level headers.
1175 	 */
1176 	siz += RPC_REPLYSIZ;
1177 	if (siz >= max_datalen) {
1178 		MCLGET(mreq, M_WAIT);
1179 	} else
1180 		mreq->m_data += max_hdr;
1181 	tl = mtod(mreq, u_int32_t *);
1182 	mreq->m_len = 6 * NFSX_UNSIGNED;
1183 	bpos = ((caddr_t)tl) + mreq->m_len;
1184 	*tl++ = txdr_unsigned(nd->nd_retxid);
1185 	*tl++ = rpc_reply;
1186 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1187 		*tl++ = rpc_msgdenied;
1188 		if (err & NFSERR_AUTHERR) {
1189 			*tl++ = rpc_autherr;
1190 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1191 			mreq->m_len -= NFSX_UNSIGNED;
1192 			bpos -= NFSX_UNSIGNED;
1193 		} else {
1194 			*tl++ = rpc_mismatch;
1195 			*tl++ = txdr_unsigned(RPC_VER2);
1196 			*tl = txdr_unsigned(RPC_VER2);
1197 		}
1198 	} else {
1199 		*tl++ = rpc_msgaccepted;
1200 
1201 		/*
1202 		 * For Kerberos authentication, we must send the nickname
1203 		 * verifier back, otherwise just RPCAUTH_NULL.
1204 		 */
1205 		if (nd->nd_flag & ND_KERBFULL) {
1206 		    struct nfsuid *nuidp;
1207 		    struct timeval ktvin, ktvout;
1208 
1209 		    for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1210 			nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1211 			if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1212 			    (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1213 			     &nuidp->nu_haddr, nd->nd_nam2)))
1214 			    break;
1215 		    }
1216 		    if (nuidp) {
1217 			ktvin.tv_sec =
1218 			    txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1219 			ktvin.tv_usec =
1220 			    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1221 
1222 			/*
1223 			 * Encrypt the timestamp in ecb mode using the
1224 			 * session key.
1225 			 */
1226 #ifdef NFSKERB
1227 			XXX
1228 #endif
1229 
1230 			*tl++ = rpc_auth_kerb;
1231 			*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1232 			*tl = ktvout.tv_sec;
1233 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1234 			*tl++ = ktvout.tv_usec;
1235 			*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1236 		    } else {
1237 			*tl++ = 0;
1238 			*tl++ = 0;
1239 		    }
1240 		} else {
1241 			*tl++ = 0;
1242 			*tl++ = 0;
1243 		}
1244 		switch (err) {
1245 		case EPROGUNAVAIL:
1246 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1247 			break;
1248 		case EPROGMISMATCH:
1249 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1250 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1251 			if (nd->nd_flag & ND_NQNFS) {
1252 				*tl++ = txdr_unsigned(3);
1253 				*tl = txdr_unsigned(3);
1254 			} else {
1255 				*tl++ = txdr_unsigned(2);
1256 				*tl = txdr_unsigned(3);
1257 			}
1258 			break;
1259 		case EPROCUNAVAIL:
1260 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1261 			break;
1262 		case EBADRPC:
1263 			*tl = txdr_unsigned(RPC_GARBAGE);
1264 			break;
1265 		default:
1266 			*tl = 0;
1267 			if (err != NFSERR_RETVOID) {
1268 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1269 				if (err)
1270 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1271 				else
1272 				    *tl = 0;
1273 			}
1274 			break;
1275 		};
1276 	}
1277 
1278 	/*
1279 	 * For nqnfs, piggyback lease as requested.
1280 	 */
1281 	if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1282 		if (nd->nd_flag & ND_LEASE) {
1283 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1284 			*tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1285 			*tl++ = txdr_unsigned(cache);
1286 			*tl++ = txdr_unsigned(nd->nd_duration);
1287 			txdr_hyper(*frev, tl);
1288 		} else {
1289 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1290 			*tl = 0;
1291 		}
1292 	}
1293 	if (mrq != NULL)
1294 		*mrq = mreq;
1295 	*mbp = mb;
1296 	*bposp = bpos;
1297 	if (err != 0 && err != NFSERR_RETVOID)
1298 		nfsstats.srvrpc_errs++;
1299 	return (0);
1300 }
1301 
1302 /*
1303  * Nfs timer routine
1304  * Scan the nfsreq list and retranmit any requests that have timed out
1305  * To avoid retransmission attempts on STREAM sockets (in the future) make
1306  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1307  */
1308 void
1309 nfs_timer(arg)
1310 	void *arg;	/* never used */
1311 {
1312 	struct nfsreq *rep;
1313 	struct mbuf *m;
1314 	struct socket *so;
1315 	struct nfsmount *nmp;
1316 	int timeo;
1317 	int s, error;
1318 #ifdef NFSSERVER
1319 	struct nfssvc_sock *slp;
1320 	static long lasttime = 0;
1321 	u_quad_t cur_usec;
1322 #endif
1323 
1324 	s = splsoftnet();
1325 	for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) {
1326 		nmp = rep->r_nmp;
1327 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1328 			continue;
1329 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1330 			rep->r_flags |= R_SOFTTERM;
1331 			continue;
1332 		}
1333 		if (rep->r_rtt >= 0) {
1334 			rep->r_rtt++;
1335 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1336 				timeo = nmp->nm_timeo;
1337 			else
1338 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1339 			if (nmp->nm_timeouts > 0)
1340 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1341 			if (rep->r_rtt <= timeo)
1342 				continue;
1343 			if (nmp->nm_timeouts < 8)
1344 				nmp->nm_timeouts++;
1345 		}
1346 		/*
1347 		 * Check for server not responding
1348 		 */
1349 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1350 		     rep->r_rexmit > nmp->nm_deadthresh) {
1351 			nfs_msg(rep->r_procp,
1352 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1353 			    "not responding");
1354 			rep->r_flags |= R_TPRINTFMSG;
1355 		}
1356 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1357 			nfsstats.rpctimeouts++;
1358 			rep->r_flags |= R_SOFTTERM;
1359 			continue;
1360 		}
1361 		if (nmp->nm_sotype != SOCK_DGRAM) {
1362 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1363 				rep->r_rexmit = NFS_MAXREXMIT;
1364 			continue;
1365 		}
1366 		if ((so = nmp->nm_so) == NULL)
1367 			continue;
1368 
1369 		/*
1370 		 * If there is enough space and the window allows..
1371 		 *	Resend it
1372 		 * Set r_rtt to -1 in case we fail to send it now.
1373 		 */
1374 		rep->r_rtt = -1;
1375 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1376 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1377 		    (rep->r_flags & R_SENT) ||
1378 		    nmp->nm_sent < nmp->nm_cwnd) &&
1379 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1380 		        if (so->so_state & SS_ISCONNECTED)
1381 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1382 			    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1383 			else
1384 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1385 			    nmp->nm_nam, (struct mbuf *)0, (struct proc *)0);
1386 			if (error) {
1387 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1388 #ifdef DEBUG
1389 					printf("nfs_timer: ignoring error %d\n",
1390 						error);
1391 #endif
1392 					so->so_error = 0;
1393 				}
1394 			} else {
1395 				/*
1396 				 * Iff first send, start timing
1397 				 * else turn timing off, backoff timer
1398 				 * and divide congestion window by 2.
1399 				 */
1400 				if (rep->r_flags & R_SENT) {
1401 					rep->r_flags &= ~R_TIMING;
1402 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1403 						rep->r_rexmit = NFS_MAXREXMIT;
1404 					nmp->nm_cwnd >>= 1;
1405 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1406 						nmp->nm_cwnd = NFS_CWNDSCALE;
1407 					nfsstats.rpcretries++;
1408 				} else {
1409 					rep->r_flags |= R_SENT;
1410 					nmp->nm_sent += NFS_CWNDSCALE;
1411 				}
1412 				rep->r_rtt = 0;
1413 			}
1414 		}
1415 	}
1416 
1417 #ifdef NFSSERVER
1418 	/*
1419 	 * Call the nqnfs server timer once a second to handle leases.
1420 	 */
1421 	if (lasttime != time.tv_sec) {
1422 		lasttime = time.tv_sec;
1423 		nqnfs_serverd();
1424 	}
1425 
1426 	/*
1427 	 * Scan the write gathering queues for writes that need to be
1428 	 * completed now.
1429 	 */
1430 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1431 	for (slp = nfssvc_sockhead.tqh_first; slp != 0;
1432 	    slp = slp->ns_chain.tqe_next) {
1433 	    if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1434 		nfsrv_wakenfsd(slp);
1435 	}
1436 #endif /* NFSSERVER */
1437 	splx(s);
1438 	callout_reset(&nfs_timer_ch, nfs_ticks, nfs_timer, NULL);
1439 }
1440 
1441 /*
1442  * Test for a termination condition pending on the process.
1443  * This is used for NFSMNT_INT mounts.
1444  */
1445 int
1446 nfs_sigintr(nmp, rep, p)
1447 	struct nfsmount *nmp;
1448 	struct nfsreq *rep;
1449 	struct proc *p;
1450 {
1451 	sigset_t ss;
1452 
1453 	if (rep && (rep->r_flags & R_SOFTTERM))
1454 		return (EINTR);
1455 	if (!(nmp->nm_flag & NFSMNT_INT))
1456 		return (0);
1457 	if (p) {
1458 		sigpending1(p, &ss);
1459 #if 0
1460 		sigminusset(&p->p_sigignore, &ss);
1461 #endif
1462 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1463 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1464 		    sigismember(&ss, SIGQUIT))
1465 			return (EINTR);
1466 	}
1467 	return (0);
1468 }
1469 
1470 /*
1471  * Lock a socket against others.
1472  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1473  * and also to avoid race conditions between the processes with nfs requests
1474  * in progress when a reconnect is necessary.
1475  */
1476 int
1477 nfs_sndlock(flagp, rep)
1478 	int *flagp;
1479 	struct nfsreq *rep;
1480 {
1481 	struct proc *p;
1482 	int slpflag = 0, slptimeo = 0;
1483 
1484 	if (rep) {
1485 		p = rep->r_procp;
1486 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1487 			slpflag = PCATCH;
1488 	} else
1489 		p = (struct proc *)0;
1490 	while (*flagp & NFSMNT_SNDLOCK) {
1491 		if (nfs_sigintr(rep->r_nmp, rep, p))
1492 			return (EINTR);
1493 		*flagp |= NFSMNT_WANTSND;
1494 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1495 			slptimeo);
1496 		if (slpflag == PCATCH) {
1497 			slpflag = 0;
1498 			slptimeo = 2 * hz;
1499 		}
1500 	}
1501 	*flagp |= NFSMNT_SNDLOCK;
1502 	return (0);
1503 }
1504 
1505 /*
1506  * Unlock the stream socket for others.
1507  */
1508 void
1509 nfs_sndunlock(flagp)
1510 	int *flagp;
1511 {
1512 
1513 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1514 		panic("nfs sndunlock");
1515 	*flagp &= ~NFSMNT_SNDLOCK;
1516 	if (*flagp & NFSMNT_WANTSND) {
1517 		*flagp &= ~NFSMNT_WANTSND;
1518 		wakeup((caddr_t)flagp);
1519 	}
1520 }
1521 
1522 int
1523 nfs_rcvlock(rep)
1524 	struct nfsreq *rep;
1525 {
1526 	struct nfsmount *nmp = rep->r_nmp;
1527 	int *flagp = &nmp->nm_iflag;
1528 	int slpflag, slptimeo = 0;
1529 
1530 	if (*flagp & NFSMNT_DISMNT)
1531 		return EIO;
1532 
1533 	if (*flagp & NFSMNT_INT)
1534 		slpflag = PCATCH;
1535 	else
1536 		slpflag = 0;
1537 	while (*flagp & NFSMNT_RCVLOCK) {
1538 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1539 			return (EINTR);
1540 		*flagp |= NFSMNT_WANTRCV;
1541 		nmp->nm_waiters++;
1542 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1543 			slptimeo);
1544 		nmp->nm_waiters--;
1545 		if (*flagp & NFSMNT_DISMNT) {
1546 			wakeup(&nmp->nm_waiters);
1547 			return EIO;
1548 		}
1549 		/* If our reply was received while we were sleeping,
1550 		 * then just return without taking the lock to avoid a
1551 		 * situation where a single iod could 'capture' the
1552 		 * receive lock.
1553 		 */
1554 		if (rep->r_mrep != NULL)
1555 			return (EALREADY);
1556 		if (slpflag == PCATCH) {
1557 			slpflag = 0;
1558 			slptimeo = 2 * hz;
1559 		}
1560 	}
1561 	*flagp |= NFSMNT_RCVLOCK;
1562 	return (0);
1563 }
1564 
1565 /*
1566  * Unlock the stream socket for others.
1567  */
1568 void
1569 nfs_rcvunlock(flagp)
1570 	int *flagp;
1571 {
1572 
1573 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1574 		panic("nfs rcvunlock");
1575 	*flagp &= ~NFSMNT_RCVLOCK;
1576 	if (*flagp & NFSMNT_WANTRCV) {
1577 		*flagp &= ~NFSMNT_WANTRCV;
1578 		wakeup((caddr_t)flagp);
1579 	}
1580 }
1581 
1582 /*
1583  * Parse an RPC request
1584  * - verify it
1585  * - fill in the cred struct.
1586  */
1587 int
1588 nfs_getreq(nd, nfsd, has_header)
1589 	struct nfsrv_descript *nd;
1590 	struct nfsd *nfsd;
1591 	int has_header;
1592 {
1593 	int len, i;
1594 	u_int32_t *tl;
1595 	int32_t t1;
1596 	struct uio uio;
1597 	struct iovec iov;
1598 	caddr_t dpos, cp2, cp;
1599 	u_int32_t nfsvers, auth_type;
1600 	uid_t nickuid;
1601 	int error = 0, nqnfs = 0, ticklen;
1602 	struct mbuf *mrep, *md;
1603 	struct nfsuid *nuidp;
1604 	struct timeval tvin, tvout;
1605 
1606 	mrep = nd->nd_mrep;
1607 	md = nd->nd_md;
1608 	dpos = nd->nd_dpos;
1609 	if (has_header) {
1610 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1611 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1612 		if (*tl++ != rpc_call) {
1613 			m_freem(mrep);
1614 			return (EBADRPC);
1615 		}
1616 	} else
1617 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1618 	nd->nd_repstat = 0;
1619 	nd->nd_flag = 0;
1620 	if (*tl++ != rpc_vers) {
1621 		nd->nd_repstat = ERPCMISMATCH;
1622 		nd->nd_procnum = NFSPROC_NOOP;
1623 		return (0);
1624 	}
1625 	if (*tl != nfs_prog) {
1626 		if (*tl == nqnfs_prog)
1627 			nqnfs++;
1628 		else {
1629 			nd->nd_repstat = EPROGUNAVAIL;
1630 			nd->nd_procnum = NFSPROC_NOOP;
1631 			return (0);
1632 		}
1633 	}
1634 	tl++;
1635 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1636 	if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1637 		(nfsvers != NQNFS_VER3 && nqnfs)) {
1638 		nd->nd_repstat = EPROGMISMATCH;
1639 		nd->nd_procnum = NFSPROC_NOOP;
1640 		return (0);
1641 	}
1642 	if (nqnfs)
1643 		nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1644 	else if (nfsvers == NFS_VER3)
1645 		nd->nd_flag = ND_NFSV3;
1646 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1647 	if (nd->nd_procnum == NFSPROC_NULL)
1648 		return (0);
1649 	if (nd->nd_procnum >= NFS_NPROCS ||
1650 		(!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1651 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1652 		nd->nd_repstat = EPROCUNAVAIL;
1653 		nd->nd_procnum = NFSPROC_NOOP;
1654 		return (0);
1655 	}
1656 	if ((nd->nd_flag & ND_NFSV3) == 0)
1657 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1658 	auth_type = *tl++;
1659 	len = fxdr_unsigned(int, *tl++);
1660 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1661 		m_freem(mrep);
1662 		return (EBADRPC);
1663 	}
1664 
1665 	nd->nd_flag &= ~ND_KERBAUTH;
1666 	/*
1667 	 * Handle auth_unix or auth_kerb.
1668 	 */
1669 	if (auth_type == rpc_auth_unix) {
1670 		len = fxdr_unsigned(int, *++tl);
1671 		if (len < 0 || len > NFS_MAXNAMLEN) {
1672 			m_freem(mrep);
1673 			return (EBADRPC);
1674 		}
1675 		nfsm_adv(nfsm_rndup(len));
1676 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1677 		memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred));
1678 		nd->nd_cr.cr_ref = 1;
1679 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1680 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1681 		len = fxdr_unsigned(int, *tl);
1682 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1683 			m_freem(mrep);
1684 			return (EBADRPC);
1685 		}
1686 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1687 		for (i = 0; i < len; i++)
1688 		    if (i < NGROUPS)
1689 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1690 		    else
1691 			tl++;
1692 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1693 		if (nd->nd_cr.cr_ngroups > 1)
1694 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1695 		len = fxdr_unsigned(int, *++tl);
1696 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1697 			m_freem(mrep);
1698 			return (EBADRPC);
1699 		}
1700 		if (len > 0)
1701 			nfsm_adv(nfsm_rndup(len));
1702 	} else if (auth_type == rpc_auth_kerb) {
1703 		switch (fxdr_unsigned(int, *tl++)) {
1704 		case RPCAKN_FULLNAME:
1705 			ticklen = fxdr_unsigned(int, *tl);
1706 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1707 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1708 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1709 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1710 				m_freem(mrep);
1711 				return (EBADRPC);
1712 			}
1713 			uio.uio_offset = 0;
1714 			uio.uio_iov = &iov;
1715 			uio.uio_iovcnt = 1;
1716 			uio.uio_segflg = UIO_SYSSPACE;
1717 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1718 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1719 			nfsm_mtouio(&uio, uio.uio_resid);
1720 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1721 			if (*tl++ != rpc_auth_kerb ||
1722 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1723 				printf("Bad kerb verifier\n");
1724 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1725 				nd->nd_procnum = NFSPROC_NOOP;
1726 				return (0);
1727 			}
1728 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1729 			tl = (u_int32_t *)cp;
1730 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1731 				printf("Not fullname kerb verifier\n");
1732 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1733 				nd->nd_procnum = NFSPROC_NOOP;
1734 				return (0);
1735 			}
1736 			cp += NFSX_UNSIGNED;
1737 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
1738 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1739 			nd->nd_flag |= ND_KERBFULL;
1740 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1741 			break;
1742 		case RPCAKN_NICKNAME:
1743 			if (len != 2 * NFSX_UNSIGNED) {
1744 				printf("Kerb nickname short\n");
1745 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1746 				nd->nd_procnum = NFSPROC_NOOP;
1747 				return (0);
1748 			}
1749 			nickuid = fxdr_unsigned(uid_t, *tl);
1750 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1751 			if (*tl++ != rpc_auth_kerb ||
1752 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1753 				printf("Kerb nick verifier bad\n");
1754 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1755 				nd->nd_procnum = NFSPROC_NOOP;
1756 				return (0);
1757 			}
1758 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1759 			tvin.tv_sec = *tl++;
1760 			tvin.tv_usec = *tl;
1761 
1762 			for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1763 			    nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1764 				if (nuidp->nu_cr.cr_uid == nickuid &&
1765 				    (!nd->nd_nam2 ||
1766 				     netaddr_match(NU_NETFAM(nuidp),
1767 				      &nuidp->nu_haddr, nd->nd_nam2)))
1768 					break;
1769 			}
1770 			if (!nuidp) {
1771 				nd->nd_repstat =
1772 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1773 				nd->nd_procnum = NFSPROC_NOOP;
1774 				return (0);
1775 			}
1776 
1777 			/*
1778 			 * Now, decrypt the timestamp using the session key
1779 			 * and validate it.
1780 			 */
1781 #ifdef NFSKERB
1782 			XXX
1783 #endif
1784 
1785 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1786 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1787 			if (nuidp->nu_expire < time.tv_sec ||
1788 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1789 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1790 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1791 				nuidp->nu_expire = 0;
1792 				nd->nd_repstat =
1793 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1794 				nd->nd_procnum = NFSPROC_NOOP;
1795 				return (0);
1796 			}
1797 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1798 			nd->nd_flag |= ND_KERBNICK;
1799 		};
1800 	} else {
1801 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1802 		nd->nd_procnum = NFSPROC_NOOP;
1803 		return (0);
1804 	}
1805 
1806 	/*
1807 	 * For nqnfs, get piggybacked lease request.
1808 	 */
1809 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
1810 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1811 		nd->nd_flag |= fxdr_unsigned(int, *tl);
1812 		if (nd->nd_flag & ND_LEASE) {
1813 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1814 			nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
1815 		} else
1816 			nd->nd_duration = NQ_MINLEASE;
1817 	} else
1818 		nd->nd_duration = NQ_MINLEASE;
1819 	nd->nd_md = md;
1820 	nd->nd_dpos = dpos;
1821 	return (0);
1822 nfsmout:
1823 	return (error);
1824 }
1825 
1826 int
1827 nfs_msg(p, server, msg)
1828 	struct proc *p;
1829 	char *server, *msg;
1830 {
1831 	tpr_t tpr;
1832 
1833 	if (p)
1834 		tpr = tprintf_open(p);
1835 	else
1836 		tpr = NULL;
1837 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1838 	tprintf_close(tpr);
1839 	return (0);
1840 }
1841 
1842 #ifdef NFSSERVER
1843 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
1844 				    struct nfssvc_sock *, struct proc *,
1845 				    struct mbuf **)) = {
1846 	nfsrv_null,
1847 	nfsrv_getattr,
1848 	nfsrv_setattr,
1849 	nfsrv_lookup,
1850 	nfsrv3_access,
1851 	nfsrv_readlink,
1852 	nfsrv_read,
1853 	nfsrv_write,
1854 	nfsrv_create,
1855 	nfsrv_mkdir,
1856 	nfsrv_symlink,
1857 	nfsrv_mknod,
1858 	nfsrv_remove,
1859 	nfsrv_rmdir,
1860 	nfsrv_rename,
1861 	nfsrv_link,
1862 	nfsrv_readdir,
1863 	nfsrv_readdirplus,
1864 	nfsrv_statfs,
1865 	nfsrv_fsinfo,
1866 	nfsrv_pathconf,
1867 	nfsrv_commit,
1868 	nqnfsrv_getlease,
1869 	nqnfsrv_vacated,
1870 	nfsrv_noop,
1871 	nfsrv_noop
1872 };
1873 
1874 /*
1875  * Socket upcall routine for the nfsd sockets.
1876  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1877  * Essentially do as much as possible non-blocking, else punt and it will
1878  * be called with M_WAIT from an nfsd.
1879  */
1880 void
1881 nfsrv_rcv(so, arg, waitflag)
1882 	struct socket *so;
1883 	caddr_t arg;
1884 	int waitflag;
1885 {
1886 	struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1887 	struct mbuf *m;
1888 	struct mbuf *mp, *nam;
1889 	struct uio auio;
1890 	int flags, error;
1891 
1892 	if ((slp->ns_flag & SLP_VALID) == 0)
1893 		return;
1894 #ifdef notdef
1895 	/*
1896 	 * Define this to test for nfsds handling this under heavy load.
1897 	 */
1898 	if (waitflag == M_DONTWAIT) {
1899 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1900 	}
1901 #endif
1902 	auio.uio_procp = NULL;
1903 	if (so->so_type == SOCK_STREAM) {
1904 		/*
1905 		 * If there are already records on the queue, defer soreceive()
1906 		 * to an nfsd so that there is feedback to the TCP layer that
1907 		 * the nfs servers are heavily loaded.
1908 		 */
1909 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1910 			slp->ns_flag |= SLP_NEEDQ;
1911 			goto dorecs;
1912 		}
1913 
1914 		/*
1915 		 * Do soreceive().
1916 		 */
1917 		auio.uio_resid = 1000000000;
1918 		flags = MSG_DONTWAIT;
1919 		error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1920 		if (error || mp == (struct mbuf *)0) {
1921 			if (error == EWOULDBLOCK)
1922 				slp->ns_flag |= SLP_NEEDQ;
1923 			else
1924 				slp->ns_flag |= SLP_DISCONN;
1925 			goto dorecs;
1926 		}
1927 		m = mp;
1928 		if (slp->ns_rawend) {
1929 			slp->ns_rawend->m_next = m;
1930 			slp->ns_cc += 1000000000 - auio.uio_resid;
1931 		} else {
1932 			slp->ns_raw = m;
1933 			slp->ns_cc = 1000000000 - auio.uio_resid;
1934 		}
1935 		while (m->m_next)
1936 			m = m->m_next;
1937 		slp->ns_rawend = m;
1938 
1939 		/*
1940 		 * Now try and parse record(s) out of the raw stream data.
1941 		 */
1942 		error = nfsrv_getstream(slp, waitflag);
1943 		if (error) {
1944 			if (error == EPERM)
1945 				slp->ns_flag |= SLP_DISCONN;
1946 			else
1947 				slp->ns_flag |= SLP_NEEDQ;
1948 		}
1949 	} else {
1950 		do {
1951 			auio.uio_resid = 1000000000;
1952 			flags = MSG_DONTWAIT;
1953 			error = (*so->so_receive)(so, &nam, &auio, &mp,
1954 						(struct mbuf **)0, &flags);
1955 			if (mp) {
1956 				if (nam) {
1957 					m = nam;
1958 					m->m_next = mp;
1959 				} else
1960 					m = mp;
1961 				if (slp->ns_recend)
1962 					slp->ns_recend->m_nextpkt = m;
1963 				else
1964 					slp->ns_rec = m;
1965 				slp->ns_recend = m;
1966 				m->m_nextpkt = (struct mbuf *)0;
1967 			}
1968 			if (error) {
1969 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
1970 					&& error != EWOULDBLOCK) {
1971 					slp->ns_flag |= SLP_DISCONN;
1972 					goto dorecs;
1973 				}
1974 			}
1975 		} while (mp);
1976 	}
1977 
1978 	/*
1979 	 * Now try and process the request records, non-blocking.
1980 	 */
1981 dorecs:
1982 	if (waitflag == M_DONTWAIT &&
1983 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
1984 		nfsrv_wakenfsd(slp);
1985 }
1986 
1987 /*
1988  * Try and extract an RPC request from the mbuf data list received on a
1989  * stream socket. The "waitflag" argument indicates whether or not it
1990  * can sleep.
1991  */
1992 int
1993 nfsrv_getstream(slp, waitflag)
1994 	struct nfssvc_sock *slp;
1995 	int waitflag;
1996 {
1997 	struct mbuf *m, **mpp;
1998 	char *cp1, *cp2;
1999 	int len;
2000 	struct mbuf *om, *m2, *recm = NULL;
2001 	u_int32_t recmark;
2002 
2003 	if (slp->ns_flag & SLP_GETSTREAM)
2004 		panic("nfs getstream");
2005 	slp->ns_flag |= SLP_GETSTREAM;
2006 	for (;;) {
2007 	    if (slp->ns_reclen == 0) {
2008 		if (slp->ns_cc < NFSX_UNSIGNED) {
2009 			slp->ns_flag &= ~SLP_GETSTREAM;
2010 			return (0);
2011 		}
2012 		m = slp->ns_raw;
2013 		if (m->m_len >= NFSX_UNSIGNED) {
2014 			memcpy((caddr_t)&recmark, mtod(m, caddr_t), NFSX_UNSIGNED);
2015 			m->m_data += NFSX_UNSIGNED;
2016 			m->m_len -= NFSX_UNSIGNED;
2017 		} else {
2018 			cp1 = (caddr_t)&recmark;
2019 			cp2 = mtod(m, caddr_t);
2020 			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
2021 				while (m->m_len == 0) {
2022 					m = m->m_next;
2023 					cp2 = mtod(m, caddr_t);
2024 				}
2025 				*cp1++ = *cp2++;
2026 				m->m_data++;
2027 				m->m_len--;
2028 			}
2029 		}
2030 		slp->ns_cc -= NFSX_UNSIGNED;
2031 		recmark = ntohl(recmark);
2032 		slp->ns_reclen = recmark & ~0x80000000;
2033 		if (recmark & 0x80000000)
2034 			slp->ns_flag |= SLP_LASTFRAG;
2035 		else
2036 			slp->ns_flag &= ~SLP_LASTFRAG;
2037 		if (slp->ns_reclen > NFS_MAXPACKET) {
2038 			slp->ns_flag &= ~SLP_GETSTREAM;
2039 			return (EPERM);
2040 		}
2041 	    }
2042 
2043 	    /*
2044 	     * Now get the record part.
2045 	     */
2046 	    if (slp->ns_cc == slp->ns_reclen) {
2047 		recm = slp->ns_raw;
2048 		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2049 		slp->ns_cc = slp->ns_reclen = 0;
2050 	    } else if (slp->ns_cc > slp->ns_reclen) {
2051 		len = 0;
2052 		m = slp->ns_raw;
2053 		om = (struct mbuf *)0;
2054 		while (len < slp->ns_reclen) {
2055 			if ((len + m->m_len) > slp->ns_reclen) {
2056 				size_t left = slp->ns_reclen - len;
2057 
2058 				MGETHDR(m2, waitflag, m->m_type);
2059 				if (m2 == NULL) {
2060 					slp->ns_flag &= ~SLP_GETSTREAM;
2061 					return (EWOULDBLOCK);
2062 				}
2063 				if (left > MHLEN) {
2064 					MCLGET(m2, waitflag);
2065 					if (!(m2->m_flags & M_EXT)) {
2066 						m_freem(m2);
2067 						slp->ns_flag &= ~SLP_GETSTREAM;
2068 						return (EWOULDBLOCK);
2069 					}
2070 				}
2071 				memcpy(mtod(m2, caddr_t), mtod(m, caddr_t),
2072 				    left);
2073 				m2->m_len = left;
2074 				m->m_data += left;
2075 				m->m_len -= left;
2076 				if (om) {
2077 					om->m_next = m2;
2078 					recm = slp->ns_raw;
2079 				} else
2080 					recm = m2;
2081 				len = slp->ns_reclen;
2082 			} else if ((len + m->m_len) == slp->ns_reclen) {
2083 				om = m;
2084 				len += m->m_len;
2085 				m = m->m_next;
2086 				recm = slp->ns_raw;
2087 				om->m_next = (struct mbuf *)0;
2088 			} else {
2089 				om = m;
2090 				len += m->m_len;
2091 				m = m->m_next;
2092 			}
2093 		}
2094 		slp->ns_raw = m;
2095 		slp->ns_cc -= len;
2096 		slp->ns_reclen = 0;
2097 	    } else {
2098 		slp->ns_flag &= ~SLP_GETSTREAM;
2099 		return (0);
2100 	    }
2101 
2102 	    /*
2103 	     * Accumulate the fragments into a record.
2104 	     */
2105 	    mpp = &slp->ns_frag;
2106 	    while (*mpp)
2107 		mpp = &((*mpp)->m_next);
2108 	    *mpp = recm;
2109 	    if (slp->ns_flag & SLP_LASTFRAG) {
2110 		if (slp->ns_recend)
2111 		    slp->ns_recend->m_nextpkt = slp->ns_frag;
2112 		else
2113 		    slp->ns_rec = slp->ns_frag;
2114 		slp->ns_recend = slp->ns_frag;
2115 		slp->ns_frag = (struct mbuf *)0;
2116 	    }
2117 	}
2118 }
2119 
2120 /*
2121  * Parse an RPC header.
2122  */
2123 int
2124 nfsrv_dorec(slp, nfsd, ndp)
2125 	struct nfssvc_sock *slp;
2126 	struct nfsd *nfsd;
2127 	struct nfsrv_descript **ndp;
2128 {
2129 	struct mbuf *m, *nam;
2130 	struct nfsrv_descript *nd;
2131 	int error;
2132 
2133 	*ndp = NULL;
2134 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2135 	    (m = slp->ns_rec) == (struct mbuf *)0)
2136 		return (ENOBUFS);
2137 	slp->ns_rec = m->m_nextpkt;
2138 	if (slp->ns_rec)
2139 		m->m_nextpkt = (struct mbuf *)0;
2140 	else
2141 		slp->ns_recend = (struct mbuf *)0;
2142 	if (m->m_type == MT_SONAME) {
2143 		nam = m;
2144 		m = m->m_next;
2145 		nam->m_next = NULL;
2146 	} else
2147 		nam = NULL;
2148 	MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2149 		M_NFSRVDESC, M_WAITOK);
2150 	nd->nd_md = nd->nd_mrep = m;
2151 	nd->nd_nam2 = nam;
2152 	nd->nd_dpos = mtod(m, caddr_t);
2153 	error = nfs_getreq(nd, nfsd, TRUE);
2154 	if (error) {
2155 		m_freem(nam);
2156 		free((caddr_t)nd, M_NFSRVDESC);
2157 		return (error);
2158 	}
2159 	*ndp = nd;
2160 	nfsd->nfsd_nd = nd;
2161 	return (0);
2162 }
2163 
2164 
2165 /*
2166  * Search for a sleeping nfsd and wake it up.
2167  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2168  * running nfsds will go look for the work in the nfssvc_sock list.
2169  */
2170 void
2171 nfsrv_wakenfsd(slp)
2172 	struct nfssvc_sock *slp;
2173 {
2174 	struct nfsd *nd;
2175 
2176 	if ((slp->ns_flag & SLP_VALID) == 0)
2177 		return;
2178 	for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) {
2179 		if (nd->nfsd_flag & NFSD_WAITING) {
2180 			nd->nfsd_flag &= ~NFSD_WAITING;
2181 			if (nd->nfsd_slp)
2182 				panic("nfsd wakeup");
2183 			slp->ns_sref++;
2184 			nd->nfsd_slp = slp;
2185 			wakeup((caddr_t)nd);
2186 			return;
2187 		}
2188 	}
2189 	slp->ns_flag |= SLP_DOREC;
2190 	nfsd_head_flag |= NFSD_CHECKSLP;
2191 }
2192 #endif /* NFSSERVER */
2193