xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 1ca5c1b28139779176bd5c13ad7c5f25c0bcd5f8)
1 /*	$NetBSD: nfs_socket.c,v 1.70 2002/01/21 13:21:51 minoura Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1991, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)nfs_socket.c	8.5 (Berkeley) 3/30/95
39  */
40 
41 /*
42  * Socket operations for use by nfs
43  */
44 
45 #include <sys/cdefs.h>
46 __KERNEL_RCSID(0, "$NetBSD: nfs_socket.c,v 1.70 2002/01/21 13:21:51 minoura Exp $");
47 
48 #include "fs_nfs.h"
49 #include "opt_nfs.h"
50 #include "opt_nfsserver.h"
51 #include "opt_inet.h"
52 
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/callout.h>
56 #include <sys/proc.h>
57 #include <sys/mount.h>
58 #include <sys/kernel.h>
59 #include <sys/mbuf.h>
60 #include <sys/vnode.h>
61 #include <sys/domain.h>
62 #include <sys/protosw.h>
63 #include <sys/socket.h>
64 #include <sys/socketvar.h>
65 #include <sys/syslog.h>
66 #include <sys/tprintf.h>
67 #include <sys/namei.h>
68 #include <sys/signal.h>
69 #include <sys/signalvar.h>
70 
71 #include <netinet/in.h>
72 #include <netinet/tcp.h>
73 
74 #include <nfs/rpcv2.h>
75 #include <nfs/nfsproto.h>
76 #include <nfs/nfs.h>
77 #include <nfs/xdr_subs.h>
78 #include <nfs/nfsm_subs.h>
79 #include <nfs/nfsmount.h>
80 #include <nfs/nfsnode.h>
81 #include <nfs/nfsrtt.h>
82 #include <nfs/nqnfs.h>
83 #include <nfs/nfs_var.h>
84 
85 #define	TRUE	1
86 #define	FALSE	0
87 
88 /*
89  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
90  * Use the mean and mean deviation of rtt for the appropriate type of rpc
91  * for the frequent rpcs and a default for the others.
92  * The justification for doing "other" this way is that these rpcs
93  * happen so infrequently that timer est. would probably be stale.
94  * Also, since many of these rpcs are
95  * non-idempotent, a conservative timeout is desired.
96  * getattr, lookup - A+2D
97  * read, write     - A+4D
98  * other           - nm_timeo
99  */
100 #define	NFS_RTO(n, t) \
101 	((t) == 0 ? (n)->nm_timeo : \
102 	 ((t) < 3 ? \
103 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
104 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
105 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
106 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
107 /*
108  * External data, mostly RPC constants in XDR form
109  */
110 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
111 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
112 	rpc_auth_kerb;
113 extern u_int32_t nfs_prog, nqnfs_prog;
114 extern time_t nqnfsstarttime;
115 extern struct nfsstats nfsstats;
116 extern int nfsv3_procid[NFS_NPROCS];
117 extern int nfs_ticks;
118 
119 /*
120  * Defines which timer to use for the procnum.
121  * 0 - default
122  * 1 - getattr
123  * 2 - lookup
124  * 3 - read
125  * 4 - write
126  */
127 static const int proct[NFS_NPROCS] = {
128 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
129 	0, 0, 0,
130 };
131 
132 /*
133  * There is a congestion window for outstanding rpcs maintained per mount
134  * point. The cwnd size is adjusted in roughly the way that:
135  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
136  * SIGCOMM '88". ACM, August 1988.
137  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
138  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
139  * of rpcs is in progress.
140  * (The sent count and cwnd are scaled for integer arith.)
141  * Variants of "slow start" were tried and were found to be too much of a
142  * performance hit (ave. rtt 3 times larger),
143  * I suspect due to the large rtt that nfs rpcs have.
144  */
145 #define	NFS_CWNDSCALE	256
146 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
147 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
148 int nfsrtton = 0;
149 struct nfsrtt nfsrtt;
150 
151 struct callout nfs_timer_ch = CALLOUT_INITIALIZER;
152 
153 /*
154  * Initialize sockets and congestion for a new NFS connection.
155  * We do not free the sockaddr if error.
156  */
157 int
158 nfs_connect(nmp, rep)
159 	struct nfsmount *nmp;
160 	struct nfsreq *rep;
161 {
162 	struct socket *so;
163 	int s, error, rcvreserve, sndreserve;
164 	struct sockaddr *saddr;
165 	struct sockaddr_in *sin;
166 #ifdef INET6
167 	struct sockaddr_in6 *sin6;
168 #endif
169 	struct mbuf *m;
170 	u_int16_t tport;
171 
172 	nmp->nm_so = (struct socket *)0;
173 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
174 	error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
175 		nmp->nm_soproto);
176 	if (error)
177 		goto bad;
178 	so = nmp->nm_so;
179 	nmp->nm_soflags = so->so_proto->pr_flags;
180 
181 	/*
182 	 * Some servers require that the client port be a reserved port number.
183 	 */
184 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
185 		MGET(m, M_WAIT, MT_SONAME);
186 		sin = mtod(m, struct sockaddr_in *);
187 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
188 		sin->sin_family = AF_INET;
189 		sin->sin_addr.s_addr = INADDR_ANY;
190 		tport = IPPORT_RESERVED - 1;
191 		sin->sin_port = htons(tport);
192 		while ((error = sobind(so, m, &proc0)) == EADDRINUSE &&
193 		       --tport > IPPORT_RESERVED / 2)
194 			sin->sin_port = htons(tport);
195 		m_freem(m);
196 		if (error)
197 			goto bad;
198 	}
199 #ifdef INET6
200 	if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) {
201 		MGET(m, M_WAIT, MT_SONAME);
202 		sin6 = mtod(m, struct sockaddr_in6 *);
203 		sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6);
204 		sin6->sin6_family = AF_INET6;
205 		sin6->sin6_addr = in6addr_any;
206 		tport = IPV6PORT_RESERVED - 1;
207 		sin6->sin6_port = htons(tport);
208 		while ((error = sobind(so, m, &proc0)) == EADDRINUSE &&
209 		       --tport > IPV6PORT_RESERVED / 2)
210 			sin6->sin6_port = htons(tport);
211 		m_freem(m);
212 		if (error)
213 			goto bad;
214 	}
215 #endif
216 
217 	/*
218 	 * Protocols that do not require connections may be optionally left
219 	 * unconnected for servers that reply from a port other than NFS_PORT.
220 	 */
221 	if (nmp->nm_flag & NFSMNT_NOCONN) {
222 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
223 			error = ENOTCONN;
224 			goto bad;
225 		}
226 	} else {
227 		error = soconnect(so, nmp->nm_nam);
228 		if (error)
229 			goto bad;
230 
231 		/*
232 		 * Wait for the connection to complete. Cribbed from the
233 		 * connect system call but with the wait timing out so
234 		 * that interruptible mounts don't hang here for a long time.
235 		 */
236 		s = splsoftnet();
237 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
238 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
239 				"nfscn1", 2 * hz);
240 			if ((so->so_state & SS_ISCONNECTING) &&
241 			    so->so_error == 0 && rep &&
242 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
243 				so->so_state &= ~SS_ISCONNECTING;
244 				splx(s);
245 				goto bad;
246 			}
247 		}
248 		if (so->so_error) {
249 			error = so->so_error;
250 			so->so_error = 0;
251 			splx(s);
252 			goto bad;
253 		}
254 		splx(s);
255 	}
256 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
257 		so->so_rcv.sb_timeo = (5 * hz);
258 		so->so_snd.sb_timeo = (5 * hz);
259 	} else {
260 		so->so_rcv.sb_timeo = 0;
261 		so->so_snd.sb_timeo = 0;
262 	}
263 	if (nmp->nm_sotype == SOCK_DGRAM) {
264 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
265 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
266 		    NFS_MAXPKTHDR) * 2;
267 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
268 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
269 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
270 		    NFS_MAXPKTHDR) * 2;
271 	} else {
272 		if (nmp->nm_sotype != SOCK_STREAM)
273 			panic("nfscon sotype");
274 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
275 			MGET(m, M_WAIT, MT_SOOPTS);
276 			*mtod(m, int32_t *) = 1;
277 			m->m_len = sizeof(int32_t);
278 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
279 		}
280 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
281 			MGET(m, M_WAIT, MT_SOOPTS);
282 			*mtod(m, int32_t *) = 1;
283 			m->m_len = sizeof(int32_t);
284 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
285 		}
286 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
287 		    sizeof (u_int32_t)) * 2;
288 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
289 		    sizeof (u_int32_t)) * 2;
290 	}
291 	error = soreserve(so, sndreserve, rcvreserve);
292 	if (error)
293 		goto bad;
294 	so->so_rcv.sb_flags |= SB_NOINTR;
295 	so->so_snd.sb_flags |= SB_NOINTR;
296 
297 	/* Initialize other non-zero congestion variables */
298 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
299 		NFS_TIMEO << 3;
300 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
301 		nmp->nm_sdrtt[3] = 0;
302 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
303 	nmp->nm_sent = 0;
304 	nmp->nm_timeouts = 0;
305 	return (0);
306 
307 bad:
308 	nfs_disconnect(nmp);
309 	return (error);
310 }
311 
312 /*
313  * Reconnect routine:
314  * Called when a connection is broken on a reliable protocol.
315  * - clean up the old socket
316  * - nfs_connect() again
317  * - set R_MUSTRESEND for all outstanding requests on mount point
318  * If this fails the mount point is DEAD!
319  * nb: Must be called with the nfs_sndlock() set on the mount point.
320  */
321 int
322 nfs_reconnect(rep)
323 	struct nfsreq *rep;
324 {
325 	struct nfsreq *rp;
326 	struct nfsmount *nmp = rep->r_nmp;
327 	int error;
328 
329 	nfs_disconnect(nmp);
330 	while ((error = nfs_connect(nmp, rep)) != 0) {
331 		if (error == EINTR || error == ERESTART)
332 			return (EINTR);
333 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0);
334 	}
335 
336 	/*
337 	 * Loop through outstanding request list and fix up all requests
338 	 * on old socket.
339 	 */
340 	for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) {
341 		if (rp->r_nmp == nmp)
342 			rp->r_flags |= R_MUSTRESEND;
343 	}
344 	return (0);
345 }
346 
347 /*
348  * NFS disconnect. Clean up and unlink.
349  */
350 void
351 nfs_disconnect(nmp)
352 	struct nfsmount *nmp;
353 {
354 	struct socket *so;
355 	int drain = 0;
356 
357 	if (nmp->nm_so) {
358 		so = nmp->nm_so;
359 		nmp->nm_so = (struct socket *)0;
360 		soshutdown(so, 2);
361 		drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
362 		if (drain) {
363 			/*
364 			 * soshutdown() above should wake up the current
365 			 * listener.
366 			 * Now wake up those waiting for the recive lock, and
367 			 * wait for them to go away unhappy, to prevent *nmp
368 			 * from evaporating while they're sleeping.
369 			 */
370 			while (nmp->nm_waiters > 0) {
371 				wakeup (&nmp->nm_iflag);
372 				(void) tsleep(&nmp->nm_waiters, PVFS,
373 				    "nfsdis", 0);
374 			}
375 		}
376 		soclose(so);
377 	}
378 #ifdef DIAGNOSTIC
379 	if (drain && (nmp->nm_waiters > 0))
380 		panic("nfs_disconnect: waiters left after drain?\n");
381 #endif
382 }
383 
384 void
385 nfs_safedisconnect(nmp)
386 	struct nfsmount *nmp;
387 {
388 	struct nfsreq dummyreq;
389 
390 	memset(&dummyreq, 0, sizeof(dummyreq));
391 	dummyreq.r_nmp = nmp;
392 	nfs_rcvlock(&dummyreq); /* XXX ignored error return */
393 	nfs_disconnect(nmp);
394 	nfs_rcvunlock(&nmp->nm_iflag);
395 }
396 
397 /*
398  * This is the nfs send routine. For connection based socket types, it
399  * must be called with an nfs_sndlock() on the socket.
400  * "rep == NULL" indicates that it has been called from a server.
401  * For the client side:
402  * - return EINTR if the RPC is terminated, 0 otherwise
403  * - set R_MUSTRESEND if the send fails for any reason
404  * - do any cleanup required by recoverable socket errors (? ? ?)
405  * For the server side:
406  * - return EINTR or ERESTART if interrupted by a signal
407  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
408  * - do any cleanup required by recoverable socket errors (? ? ?)
409  */
410 int
411 nfs_send(so, nam, top, rep)
412 	struct socket *so;
413 	struct mbuf *nam;
414 	struct mbuf *top;
415 	struct nfsreq *rep;
416 {
417 	struct mbuf *sendnam;
418 	int error, soflags, flags;
419 
420 	if (rep) {
421 		if (rep->r_flags & R_SOFTTERM) {
422 			m_freem(top);
423 			return (EINTR);
424 		}
425 		if ((so = rep->r_nmp->nm_so) == NULL) {
426 			rep->r_flags |= R_MUSTRESEND;
427 			m_freem(top);
428 			return (0);
429 		}
430 		rep->r_flags &= ~R_MUSTRESEND;
431 		soflags = rep->r_nmp->nm_soflags;
432 	} else
433 		soflags = so->so_proto->pr_flags;
434 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
435 		sendnam = (struct mbuf *)0;
436 	else
437 		sendnam = nam;
438 	if (so->so_type == SOCK_SEQPACKET)
439 		flags = MSG_EOR;
440 	else
441 		flags = 0;
442 
443 	error = (*so->so_send)(so, sendnam, (struct uio *)0, top,
444 		(struct mbuf *)0, flags);
445 	if (error) {
446 		if (rep) {
447 			if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
448 				/*
449 				 * We're too fast for the network/driver,
450 				 * and UDP isn't flowcontrolled.
451 				 * We need to resend. This is not fatal,
452 				 * just try again.
453 				 *
454 				 * Could be smarter here by doing some sort
455 				 * of a backoff, but this is rare.
456 				 */
457 				rep->r_flags |= R_MUSTRESEND;
458 			} else {
459 				log(LOG_INFO, "nfs send error %d for %s\n",
460 				    error,
461 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
462 				/*
463 				 * Deal with errors for the client side.
464 				 */
465 				if (rep->r_flags & R_SOFTTERM)
466 					error = EINTR;
467 				else
468 					rep->r_flags |= R_MUSTRESEND;
469 			}
470 		} else {
471 			/*
472 			 * See above. This error can happen under normal
473 			 * circumstances and the log is too noisy.
474 			 * The error will still show up in nfsstat.
475 			 */
476 			if (error != ENOBUFS || so->so_type != SOCK_DGRAM)
477 				log(LOG_INFO, "nfsd send error %d\n", error);
478 		}
479 
480 		/*
481 		 * Handle any recoverable (soft) socket errors here. (? ? ?)
482 		 */
483 		if (error != EINTR && error != ERESTART &&
484 			error != EWOULDBLOCK && error != EPIPE)
485 			error = 0;
486 	}
487 	return (error);
488 }
489 
490 #ifdef NFS
491 /*
492  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
493  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
494  * Mark and consolidate the data into a new mbuf list.
495  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
496  *     small mbufs.
497  * For SOCK_STREAM we must be very careful to read an entire record once
498  * we have read any of it, even if the system call has been interrupted.
499  */
500 int
501 nfs_receive(rep, aname, mp)
502 	struct nfsreq *rep;
503 	struct mbuf **aname;
504 	struct mbuf **mp;
505 {
506 	struct socket *so;
507 	struct uio auio;
508 	struct iovec aio;
509 	struct mbuf *m;
510 	struct mbuf *control;
511 	u_int32_t len;
512 	struct mbuf **getnam;
513 	int error, sotype, rcvflg;
514 	struct proc *p = curproc;	/* XXX */
515 
516 	/*
517 	 * Set up arguments for soreceive()
518 	 */
519 	*mp = (struct mbuf *)0;
520 	*aname = (struct mbuf *)0;
521 	sotype = rep->r_nmp->nm_sotype;
522 
523 	/*
524 	 * For reliable protocols, lock against other senders/receivers
525 	 * in case a reconnect is necessary.
526 	 * For SOCK_STREAM, first get the Record Mark to find out how much
527 	 * more there is to get.
528 	 * We must lock the socket against other receivers
529 	 * until we have an entire rpc request/reply.
530 	 */
531 	if (sotype != SOCK_DGRAM) {
532 		error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
533 		if (error)
534 			return (error);
535 tryagain:
536 		/*
537 		 * Check for fatal errors and resending request.
538 		 */
539 		/*
540 		 * Ugh: If a reconnect attempt just happened, nm_so
541 		 * would have changed. NULL indicates a failed
542 		 * attempt that has essentially shut down this
543 		 * mount point.
544 		 */
545 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
546 			nfs_sndunlock(&rep->r_nmp->nm_iflag);
547 			return (EINTR);
548 		}
549 		so = rep->r_nmp->nm_so;
550 		if (!so) {
551 			error = nfs_reconnect(rep);
552 			if (error) {
553 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
554 				return (error);
555 			}
556 			goto tryagain;
557 		}
558 		while (rep->r_flags & R_MUSTRESEND) {
559 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
560 			nfsstats.rpcretries++;
561 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
562 			if (error) {
563 				if (error == EINTR || error == ERESTART ||
564 				    (error = nfs_reconnect(rep)) != 0) {
565 					nfs_sndunlock(&rep->r_nmp->nm_iflag);
566 					return (error);
567 				}
568 				goto tryagain;
569 			}
570 		}
571 		nfs_sndunlock(&rep->r_nmp->nm_iflag);
572 		if (sotype == SOCK_STREAM) {
573 			aio.iov_base = (caddr_t) &len;
574 			aio.iov_len = sizeof(u_int32_t);
575 			auio.uio_iov = &aio;
576 			auio.uio_iovcnt = 1;
577 			auio.uio_segflg = UIO_SYSSPACE;
578 			auio.uio_rw = UIO_READ;
579 			auio.uio_offset = 0;
580 			auio.uio_resid = sizeof(u_int32_t);
581 			auio.uio_procp = p;
582 			do {
583 			   rcvflg = MSG_WAITALL;
584 			   error = (*so->so_receive)(so, (struct mbuf **)0, &auio,
585 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
586 			   if (error == EWOULDBLOCK && rep) {
587 				if (rep->r_flags & R_SOFTTERM)
588 					return (EINTR);
589 			   }
590 			} while (error == EWOULDBLOCK);
591 			if (!error && auio.uio_resid > 0) {
592 			    /*
593 			     * Don't log a 0 byte receive; it means
594 			     * that the socket has been closed, and
595 			     * can happen during normal operation
596 			     * (forcible unmount or Solaris server).
597 			     */
598 			    if (auio.uio_resid != sizeof (u_int32_t))
599 			      log(LOG_INFO,
600 				 "short receive (%lu/%lu) from nfs server %s\n",
601 				 (u_long)sizeof(u_int32_t) - auio.uio_resid,
602 				 (u_long)sizeof(u_int32_t),
603 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
604 			    error = EPIPE;
605 			}
606 			if (error)
607 				goto errout;
608 			len = ntohl(len) & ~0x80000000;
609 			/*
610 			 * This is SERIOUS! We are out of sync with the sender
611 			 * and forcing a disconnect/reconnect is all I can do.
612 			 */
613 			if (len > NFS_MAXPACKET) {
614 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
615 				"impossible packet length",
616 				len,
617 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
618 			    error = EFBIG;
619 			    goto errout;
620 			}
621 			auio.uio_resid = len;
622 			do {
623 			    rcvflg = MSG_WAITALL;
624 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
625 				&auio, mp, (struct mbuf **)0, &rcvflg);
626 			} while (error == EWOULDBLOCK || error == EINTR ||
627 				 error == ERESTART);
628 			if (!error && auio.uio_resid > 0) {
629 			    if (len != auio.uio_resid)
630 			      log(LOG_INFO,
631 				"short receive (%lu/%d) from nfs server %s\n",
632 				(u_long)len - auio.uio_resid, len,
633 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
634 			    error = EPIPE;
635 			}
636 		} else {
637 			/*
638 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
639 			 * and soreceive() will return when it has either a
640 			 * control msg or a data msg.
641 			 * We have no use for control msg., but must grab them
642 			 * and then throw them away so we know what is going
643 			 * on.
644 			 */
645 			auio.uio_resid = len = 100000000; /* Anything Big */
646 			auio.uio_procp = p;
647 			do {
648 			    rcvflg = 0;
649 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
650 				&auio, mp, &control, &rcvflg);
651 			    if (control)
652 				m_freem(control);
653 			    if (error == EWOULDBLOCK && rep) {
654 				if (rep->r_flags & R_SOFTTERM)
655 					return (EINTR);
656 			    }
657 			} while (error == EWOULDBLOCK ||
658 				 (!error && *mp == NULL && control));
659 			if ((rcvflg & MSG_EOR) == 0)
660 				printf("Egad!!\n");
661 			if (!error && *mp == NULL)
662 				error = EPIPE;
663 			len -= auio.uio_resid;
664 		}
665 errout:
666 		if (error && error != EINTR && error != ERESTART) {
667 			m_freem(*mp);
668 			*mp = (struct mbuf *)0;
669 			if (error != EPIPE)
670 				log(LOG_INFO,
671 				    "receive error %d from nfs server %s\n",
672 				    error,
673 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
674 			error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
675 			if (!error)
676 				error = nfs_reconnect(rep);
677 			if (!error)
678 				goto tryagain;
679 			else
680 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
681 		}
682 	} else {
683 		if ((so = rep->r_nmp->nm_so) == NULL)
684 			return (EACCES);
685 		if (so->so_state & SS_ISCONNECTED)
686 			getnam = (struct mbuf **)0;
687 		else
688 			getnam = aname;
689 		auio.uio_resid = len = 1000000;
690 		auio.uio_procp = p;
691 		do {
692 			rcvflg = 0;
693 			error =  (*so->so_receive)(so, getnam, &auio, mp,
694 				(struct mbuf **)0, &rcvflg);
695 			if (error == EWOULDBLOCK &&
696 			    (rep->r_flags & R_SOFTTERM))
697 				return (EINTR);
698 		} while (error == EWOULDBLOCK);
699 		len -= auio.uio_resid;
700 		if (!error && *mp == NULL)
701 			error = EPIPE;
702 	}
703 	if (error) {
704 		m_freem(*mp);
705 		*mp = (struct mbuf *)0;
706 	}
707 	return (error);
708 }
709 
710 /*
711  * Implement receipt of reply on a socket.
712  * We must search through the list of received datagrams matching them
713  * with outstanding requests using the xid, until ours is found.
714  */
715 /* ARGSUSED */
716 int
717 nfs_reply(myrep)
718 	struct nfsreq *myrep;
719 {
720 	struct nfsreq *rep;
721 	struct nfsmount *nmp = myrep->r_nmp;
722 	int32_t t1;
723 	struct mbuf *mrep, *nam, *md;
724 	u_int32_t rxid, *tl;
725 	caddr_t dpos, cp2;
726 	int error;
727 
728 	/*
729 	 * Loop around until we get our own reply
730 	 */
731 	for (;;) {
732 		/*
733 		 * Lock against other receivers so that I don't get stuck in
734 		 * sbwait() after someone else has received my reply for me.
735 		 * Also necessary for connection based protocols to avoid
736 		 * race conditions during a reconnect.
737 		 */
738 		error = nfs_rcvlock(myrep);
739 		if (error == EALREADY)
740 			return (0);
741 		if (error)
742 			return (error);
743 		/*
744 		 * Get the next Rpc reply off the socket
745 		 */
746 		nmp->nm_waiters++;
747 		error = nfs_receive(myrep, &nam, &mrep);
748 		nfs_rcvunlock(&nmp->nm_iflag);
749 		if (error) {
750 
751 			if (nmp->nm_iflag & NFSMNT_DISMNT) {
752 				/*
753 				 * Oops, we're going away now..
754 				 */
755 				nmp->nm_waiters--;
756 				wakeup (&nmp->nm_waiters);
757 				return error;
758 			}
759 			nmp->nm_waiters--;
760 			/*
761 			 * Ignore routing errors on connectionless protocols? ?
762 			 */
763 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
764 				nmp->nm_so->so_error = 0;
765 #ifdef DEBUG
766 				printf("nfs_reply: ignoring error %d\n", error);
767 #endif
768 				if (myrep->r_flags & R_GETONEREP)
769 					return (0);
770 				continue;
771 			}
772 			return (error);
773 		}
774 		nmp->nm_waiters--;
775 		if (nam)
776 			m_freem(nam);
777 
778 		/*
779 		 * Get the xid and check that it is an rpc reply
780 		 */
781 		md = mrep;
782 		dpos = mtod(md, caddr_t);
783 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
784 		rxid = *tl++;
785 		if (*tl != rpc_reply) {
786 #ifndef NFS_V2_ONLY
787 			if (nmp->nm_flag & NFSMNT_NQNFS) {
788 				if (nqnfs_callback(nmp, mrep, md, dpos))
789 					nfsstats.rpcinvalid++;
790 			} else
791 #endif
792 			{
793 				nfsstats.rpcinvalid++;
794 				m_freem(mrep);
795 			}
796 nfsmout:
797 			if (myrep->r_flags & R_GETONEREP)
798 				return (0);
799 			continue;
800 		}
801 
802 		/*
803 		 * Loop through the request list to match up the reply
804 		 * Iff no match, just drop the datagram
805 		 */
806 		for (rep = nfs_reqq.tqh_first; rep != 0;
807 		    rep = rep->r_chain.tqe_next) {
808 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
809 				/* Found it.. */
810 				rep->r_mrep = mrep;
811 				rep->r_md = md;
812 				rep->r_dpos = dpos;
813 				if (nfsrtton) {
814 					struct rttl *rt;
815 
816 					rt = &nfsrtt.rttl[nfsrtt.pos];
817 					rt->proc = rep->r_procnum;
818 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
819 					rt->sent = nmp->nm_sent;
820 					rt->cwnd = nmp->nm_cwnd;
821 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
822 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
823 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
824 					rt->tstamp = time;
825 					if (rep->r_flags & R_TIMING)
826 						rt->rtt = rep->r_rtt;
827 					else
828 						rt->rtt = 1000000;
829 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
830 				}
831 				/*
832 				 * Update congestion window.
833 				 * Do the additive increase of
834 				 * one rpc/rtt.
835 				 */
836 				if (nmp->nm_cwnd <= nmp->nm_sent) {
837 					nmp->nm_cwnd +=
838 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
839 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
840 					if (nmp->nm_cwnd > NFS_MAXCWND)
841 						nmp->nm_cwnd = NFS_MAXCWND;
842 				}
843 				rep->r_flags &= ~R_SENT;
844 				nmp->nm_sent -= NFS_CWNDSCALE;
845 				/*
846 				 * Update rtt using a gain of 0.125 on the mean
847 				 * and a gain of 0.25 on the deviation.
848 				 */
849 				if (rep->r_flags & R_TIMING) {
850 					/*
851 					 * Since the timer resolution of
852 					 * NFS_HZ is so course, it can often
853 					 * result in r_rtt == 0. Since
854 					 * r_rtt == N means that the actual
855 					 * rtt is between N+dt and N+2-dt ticks,
856 					 * add 1.
857 					 */
858 					t1 = rep->r_rtt + 1;
859 					t1 -= (NFS_SRTT(rep) >> 3);
860 					NFS_SRTT(rep) += t1;
861 					if (t1 < 0)
862 						t1 = -t1;
863 					t1 -= (NFS_SDRTT(rep) >> 2);
864 					NFS_SDRTT(rep) += t1;
865 				}
866 				nmp->nm_timeouts = 0;
867 				break;
868 			}
869 		}
870 		/*
871 		 * If not matched to a request, drop it.
872 		 * If it's mine, get out.
873 		 */
874 		if (rep == 0) {
875 			nfsstats.rpcunexpected++;
876 			m_freem(mrep);
877 		} else if (rep == myrep) {
878 			if (rep->r_mrep == NULL)
879 				panic("nfsreply nil");
880 			return (0);
881 		}
882 		if (myrep->r_flags & R_GETONEREP)
883 			return (0);
884 	}
885 }
886 
887 /*
888  * nfs_request - goes something like this
889  *	- fill in request struct
890  *	- links it into list
891  *	- calls nfs_send() for first transmit
892  *	- calls nfs_receive() to get reply
893  *	- break down rpc header and return with nfs reply pointed to
894  *	  by mrep or error
895  * nb: always frees up mreq mbuf list
896  */
897 int
898 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
899 	struct vnode *vp;
900 	struct mbuf *mrest;
901 	int procnum;
902 	struct proc *procp;
903 	struct ucred *cred;
904 	struct mbuf **mrp;
905 	struct mbuf **mdp;
906 	caddr_t *dposp;
907 {
908 	struct mbuf *m, *mrep;
909 	struct nfsreq *rep;
910 	u_int32_t *tl;
911 	int i;
912 	struct nfsmount *nmp;
913 	struct mbuf *md, *mheadend;
914 	char nickv[RPCX_NICKVERF];
915 	time_t reqtime, waituntil;
916 	caddr_t dpos, cp2;
917 	int t1, s, error = 0, mrest_len, auth_len, auth_type;
918 	int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0;
919 	int verf_len, verf_type;
920 	u_int32_t xid;
921 	char *auth_str, *verf_str;
922 	NFSKERBKEY_T key;		/* save session key */
923 #ifndef NFS_V2_ONLY
924 	int nqlflag, cachable;
925 	u_quad_t frev;
926 	struct nfsnode *np;
927 #endif
928 
929 	KASSERT(cred != NULL);
930 	nmp = VFSTONFS(vp->v_mount);
931 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
932 	rep->r_nmp = nmp;
933 	rep->r_vp = vp;
934 	rep->r_procp = procp;
935 	rep->r_procnum = procnum;
936 	i = 0;
937 	m = mrest;
938 	while (m) {
939 		i += m->m_len;
940 		m = m->m_next;
941 	}
942 	mrest_len = i;
943 
944 	/*
945 	 * Get the RPC header with authorization.
946 	 */
947 kerbauth:
948 	verf_str = auth_str = (char *)0;
949 	if (nmp->nm_flag & NFSMNT_KERB) {
950 		verf_str = nickv;
951 		verf_len = sizeof (nickv);
952 		auth_type = RPCAUTH_KERB4;
953 		memset((caddr_t)key, 0, sizeof (key));
954 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
955 			&auth_len, verf_str, verf_len)) {
956 			error = nfs_getauth(nmp, rep, cred, &auth_str,
957 				&auth_len, verf_str, &verf_len, key);
958 			if (error) {
959 				free((caddr_t)rep, M_NFSREQ);
960 				m_freem(mrest);
961 				return (error);
962 			}
963 		}
964 	} else {
965 		auth_type = RPCAUTH_UNIX;
966 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
967 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
968 			5 * NFSX_UNSIGNED;
969 	}
970 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
971 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
972 	if (auth_str)
973 		free(auth_str, M_TEMP);
974 
975 	/*
976 	 * For stream protocols, insert a Sun RPC Record Mark.
977 	 */
978 	if (nmp->nm_sotype == SOCK_STREAM) {
979 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
980 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
981 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
982 	}
983 	rep->r_mreq = m;
984 	rep->r_xid = xid;
985 tryagain:
986 	if (nmp->nm_flag & NFSMNT_SOFT)
987 		rep->r_retry = nmp->nm_retry;
988 	else
989 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
990 	rep->r_rtt = rep->r_rexmit = 0;
991 	if (proct[procnum] > 0)
992 		rep->r_flags = R_TIMING;
993 	else
994 		rep->r_flags = 0;
995 	rep->r_mrep = NULL;
996 
997 	/*
998 	 * Do the client side RPC.
999 	 */
1000 	nfsstats.rpcrequests++;
1001 	/*
1002 	 * Chain request into list of outstanding requests. Be sure
1003 	 * to put it LAST so timer finds oldest requests first.
1004 	 */
1005 	s = splsoftnet();
1006 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1007 
1008 	/* Get send time for nqnfs */
1009 	reqtime = time.tv_sec;
1010 
1011 	/*
1012 	 * If backing off another request or avoiding congestion, don't
1013 	 * send this one now but let timer do it. If not timing a request,
1014 	 * do it now.
1015 	 */
1016 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1017 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1018 		nmp->nm_sent < nmp->nm_cwnd)) {
1019 		splx(s);
1020 		if (nmp->nm_soflags & PR_CONNREQUIRED)
1021 			error = nfs_sndlock(&nmp->nm_iflag, rep);
1022 		if (!error) {
1023 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
1024 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep);
1025 			if (nmp->nm_soflags & PR_CONNREQUIRED)
1026 				nfs_sndunlock(&nmp->nm_iflag);
1027 		}
1028 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1029 			nmp->nm_sent += NFS_CWNDSCALE;
1030 			rep->r_flags |= R_SENT;
1031 		}
1032 	} else {
1033 		splx(s);
1034 		rep->r_rtt = -1;
1035 	}
1036 
1037 	/*
1038 	 * Wait for the reply from our send or the timer's.
1039 	 */
1040 	if (!error || error == EPIPE)
1041 		error = nfs_reply(rep);
1042 
1043 	/*
1044 	 * RPC done, unlink the request.
1045 	 */
1046 	s = splsoftnet();
1047 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1048 	splx(s);
1049 
1050 	/*
1051 	 * Decrement the outstanding request count.
1052 	 */
1053 	if (rep->r_flags & R_SENT) {
1054 		rep->r_flags &= ~R_SENT;	/* paranoia */
1055 		nmp->nm_sent -= NFS_CWNDSCALE;
1056 	}
1057 
1058 	/*
1059 	 * If there was a successful reply and a tprintf msg.
1060 	 * tprintf a response.
1061 	 */
1062 	if (!error && (rep->r_flags & R_TPRINTFMSG))
1063 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1064 		    "is alive again");
1065 	mrep = rep->r_mrep;
1066 	md = rep->r_md;
1067 	dpos = rep->r_dpos;
1068 	if (error) {
1069 		m_freem(rep->r_mreq);
1070 		free((caddr_t)rep, M_NFSREQ);
1071 		return (error);
1072 	}
1073 
1074 	/*
1075 	 * break down the rpc header and check if ok
1076 	 */
1077 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1078 	if (*tl++ == rpc_msgdenied) {
1079 		if (*tl == rpc_mismatch)
1080 			error = EOPNOTSUPP;
1081 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1082 			if (!failed_auth) {
1083 				failed_auth++;
1084 				mheadend->m_next = (struct mbuf *)0;
1085 				m_freem(mrep);
1086 				m_freem(rep->r_mreq);
1087 				goto kerbauth;
1088 			} else
1089 				error = EAUTH;
1090 		} else
1091 			error = EACCES;
1092 		m_freem(mrep);
1093 		m_freem(rep->r_mreq);
1094 		free((caddr_t)rep, M_NFSREQ);
1095 		return (error);
1096 	}
1097 
1098 	/*
1099 	 * Grab any Kerberos verifier, otherwise just throw it away.
1100 	 */
1101 	verf_type = fxdr_unsigned(int, *tl++);
1102 	i = fxdr_unsigned(int32_t, *tl);
1103 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1104 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1105 		if (error)
1106 			goto nfsmout;
1107 	} else if (i > 0)
1108 		nfsm_adv(nfsm_rndup(i));
1109 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1110 	/* 0 == ok */
1111 	if (*tl == 0) {
1112 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1113 		if (*tl != 0) {
1114 			error = fxdr_unsigned(int, *tl);
1115 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1116 				error == NFSERR_TRYLATER) {
1117 				m_freem(mrep);
1118 				error = 0;
1119 				waituntil = time.tv_sec + trylater_delay;
1120 				while (time.tv_sec < waituntil)
1121 					(void) tsleep((caddr_t)&lbolt,
1122 						PSOCK, "nqnfstry", 0);
1123 				trylater_delay *= nfs_backoff[trylater_cnt];
1124 				if (trylater_cnt < 7)
1125 					trylater_cnt++;
1126 				goto tryagain;
1127 			}
1128 
1129 			/*
1130 			 * If the File Handle was stale, invalidate the
1131 			 * lookup cache, just in case.
1132 			 */
1133 			if (error == ESTALE)
1134 				cache_purge(vp);
1135 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1136 				*mrp = mrep;
1137 				*mdp = md;
1138 				*dposp = dpos;
1139 				error |= NFSERR_RETERR;
1140 			} else
1141 				m_freem(mrep);
1142 			m_freem(rep->r_mreq);
1143 			free((caddr_t)rep, M_NFSREQ);
1144 			return (error);
1145 		}
1146 
1147 #ifndef NFS_V2_ONLY
1148 		/*
1149 		 * For nqnfs, get any lease in reply
1150 		 */
1151 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1152 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1153 			if (*tl) {
1154 				np = VTONFS(vp);
1155 				nqlflag = fxdr_unsigned(int, *tl);
1156 				nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1157 				cachable = fxdr_unsigned(int, *tl++);
1158 				reqtime += fxdr_unsigned(int, *tl++);
1159 				if (reqtime > time.tv_sec) {
1160 				    frev = fxdr_hyper(tl);
1161 				    nqnfs_clientlease(nmp, np, nqlflag,
1162 					cachable, reqtime, frev);
1163 				}
1164 			}
1165 		}
1166 #endif
1167 		*mrp = mrep;
1168 		*mdp = md;
1169 		*dposp = dpos;
1170 		m_freem(rep->r_mreq);
1171 		FREE((caddr_t)rep, M_NFSREQ);
1172 		return (0);
1173 	}
1174 	m_freem(mrep);
1175 	error = EPROTONOSUPPORT;
1176 nfsmout:
1177 	m_freem(rep->r_mreq);
1178 	free((caddr_t)rep, M_NFSREQ);
1179 	return (error);
1180 }
1181 #endif /* NFS */
1182 
1183 /*
1184  * Generate the rpc reply header
1185  * siz arg. is used to decide if adding a cluster is worthwhile
1186  */
1187 int
1188 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1189 	int siz;
1190 	struct nfsrv_descript *nd;
1191 	struct nfssvc_sock *slp;
1192 	int err;
1193 	int cache;
1194 	u_quad_t *frev;
1195 	struct mbuf **mrq;
1196 	struct mbuf **mbp;
1197 	caddr_t *bposp;
1198 {
1199 	u_int32_t *tl;
1200 	struct mbuf *mreq;
1201 	caddr_t bpos;
1202 	struct mbuf *mb, *mb2;
1203 
1204 	MGETHDR(mreq, M_WAIT, MT_DATA);
1205 	mb = mreq;
1206 	/*
1207 	 * If this is a big reply, use a cluster else
1208 	 * try and leave leading space for the lower level headers.
1209 	 */
1210 	siz += RPC_REPLYSIZ;
1211 	if (siz >= max_datalen) {
1212 		MCLGET(mreq, M_WAIT);
1213 	} else
1214 		mreq->m_data += max_hdr;
1215 	tl = mtod(mreq, u_int32_t *);
1216 	mreq->m_len = 6 * NFSX_UNSIGNED;
1217 	bpos = ((caddr_t)tl) + mreq->m_len;
1218 	*tl++ = txdr_unsigned(nd->nd_retxid);
1219 	*tl++ = rpc_reply;
1220 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1221 		*tl++ = rpc_msgdenied;
1222 		if (err & NFSERR_AUTHERR) {
1223 			*tl++ = rpc_autherr;
1224 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1225 			mreq->m_len -= NFSX_UNSIGNED;
1226 			bpos -= NFSX_UNSIGNED;
1227 		} else {
1228 			*tl++ = rpc_mismatch;
1229 			*tl++ = txdr_unsigned(RPC_VER2);
1230 			*tl = txdr_unsigned(RPC_VER2);
1231 		}
1232 	} else {
1233 		*tl++ = rpc_msgaccepted;
1234 
1235 		/*
1236 		 * For Kerberos authentication, we must send the nickname
1237 		 * verifier back, otherwise just RPCAUTH_NULL.
1238 		 */
1239 		if (nd->nd_flag & ND_KERBFULL) {
1240 		    struct nfsuid *nuidp;
1241 		    struct timeval ktvin, ktvout;
1242 
1243 		    for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1244 			nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1245 			if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1246 			    (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1247 			     &nuidp->nu_haddr, nd->nd_nam2)))
1248 			    break;
1249 		    }
1250 		    if (nuidp) {
1251 			ktvin.tv_sec =
1252 			    txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1253 			ktvin.tv_usec =
1254 			    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1255 
1256 			/*
1257 			 * Encrypt the timestamp in ecb mode using the
1258 			 * session key.
1259 			 */
1260 #ifdef NFSKERB
1261 			XXX
1262 #endif
1263 
1264 			*tl++ = rpc_auth_kerb;
1265 			*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1266 			*tl = ktvout.tv_sec;
1267 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1268 			*tl++ = ktvout.tv_usec;
1269 			*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1270 		    } else {
1271 			*tl++ = 0;
1272 			*tl++ = 0;
1273 		    }
1274 		} else {
1275 			*tl++ = 0;
1276 			*tl++ = 0;
1277 		}
1278 		switch (err) {
1279 		case EPROGUNAVAIL:
1280 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1281 			break;
1282 		case EPROGMISMATCH:
1283 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1284 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1285 			if (nd->nd_flag & ND_NQNFS) {
1286 				*tl++ = txdr_unsigned(3);
1287 				*tl = txdr_unsigned(3);
1288 			} else {
1289 				*tl++ = txdr_unsigned(2);
1290 				*tl = txdr_unsigned(3);
1291 			}
1292 			break;
1293 		case EPROCUNAVAIL:
1294 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1295 			break;
1296 		case EBADRPC:
1297 			*tl = txdr_unsigned(RPC_GARBAGE);
1298 			break;
1299 		default:
1300 			*tl = 0;
1301 			if (err != NFSERR_RETVOID) {
1302 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1303 				if (err)
1304 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1305 				else
1306 				    *tl = 0;
1307 			}
1308 			break;
1309 		};
1310 	}
1311 
1312 	/*
1313 	 * For nqnfs, piggyback lease as requested.
1314 	 */
1315 	if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1316 		if (nd->nd_flag & ND_LEASE) {
1317 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1318 			*tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1319 			*tl++ = txdr_unsigned(cache);
1320 			*tl++ = txdr_unsigned(nd->nd_duration);
1321 			txdr_hyper(*frev, tl);
1322 		} else {
1323 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1324 			*tl = 0;
1325 		}
1326 	}
1327 	if (mrq != NULL)
1328 		*mrq = mreq;
1329 	*mbp = mb;
1330 	*bposp = bpos;
1331 	if (err != 0 && err != NFSERR_RETVOID)
1332 		nfsstats.srvrpc_errs++;
1333 	return (0);
1334 }
1335 
1336 /*
1337  * Nfs timer routine
1338  * Scan the nfsreq list and retranmit any requests that have timed out
1339  * To avoid retransmission attempts on STREAM sockets (in the future) make
1340  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1341  */
1342 void
1343 nfs_timer(arg)
1344 	void *arg;	/* never used */
1345 {
1346 	struct nfsreq *rep;
1347 	struct mbuf *m;
1348 	struct socket *so;
1349 	struct nfsmount *nmp;
1350 	int timeo;
1351 	int s, error;
1352 #ifdef NFSSERVER
1353 	struct nfssvc_sock *slp;
1354 	static long lasttime = 0;
1355 	u_quad_t cur_usec;
1356 #endif
1357 
1358 	s = splsoftnet();
1359 	for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) {
1360 		nmp = rep->r_nmp;
1361 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1362 			continue;
1363 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1364 			rep->r_flags |= R_SOFTTERM;
1365 			continue;
1366 		}
1367 		if (rep->r_rtt >= 0) {
1368 			rep->r_rtt++;
1369 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1370 				timeo = nmp->nm_timeo;
1371 			else
1372 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1373 			if (nmp->nm_timeouts > 0)
1374 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1375 			if (rep->r_rtt <= timeo)
1376 				continue;
1377 			if (nmp->nm_timeouts < 8)
1378 				nmp->nm_timeouts++;
1379 		}
1380 		/*
1381 		 * Check for server not responding
1382 		 */
1383 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1384 		     rep->r_rexmit > nmp->nm_deadthresh) {
1385 			nfs_msg(rep->r_procp,
1386 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1387 			    "not responding");
1388 			rep->r_flags |= R_TPRINTFMSG;
1389 		}
1390 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1391 			nfsstats.rpctimeouts++;
1392 			rep->r_flags |= R_SOFTTERM;
1393 			continue;
1394 		}
1395 		if (nmp->nm_sotype != SOCK_DGRAM) {
1396 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1397 				rep->r_rexmit = NFS_MAXREXMIT;
1398 			continue;
1399 		}
1400 		if ((so = nmp->nm_so) == NULL)
1401 			continue;
1402 
1403 		/*
1404 		 * If there is enough space and the window allows..
1405 		 *	Resend it
1406 		 * Set r_rtt to -1 in case we fail to send it now.
1407 		 */
1408 		rep->r_rtt = -1;
1409 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1410 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1411 		    (rep->r_flags & R_SENT) ||
1412 		    nmp->nm_sent < nmp->nm_cwnd) &&
1413 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1414 		        if (so->so_state & SS_ISCONNECTED)
1415 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1416 			    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1417 			else
1418 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1419 			    nmp->nm_nam, (struct mbuf *)0, (struct proc *)0);
1420 			if (error) {
1421 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1422 #ifdef DEBUG
1423 					printf("nfs_timer: ignoring error %d\n",
1424 						error);
1425 #endif
1426 					so->so_error = 0;
1427 				}
1428 			} else {
1429 				/*
1430 				 * Iff first send, start timing
1431 				 * else turn timing off, backoff timer
1432 				 * and divide congestion window by 2.
1433 				 */
1434 				if (rep->r_flags & R_SENT) {
1435 					rep->r_flags &= ~R_TIMING;
1436 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1437 						rep->r_rexmit = NFS_MAXREXMIT;
1438 					nmp->nm_cwnd >>= 1;
1439 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1440 						nmp->nm_cwnd = NFS_CWNDSCALE;
1441 					nfsstats.rpcretries++;
1442 				} else {
1443 					rep->r_flags |= R_SENT;
1444 					nmp->nm_sent += NFS_CWNDSCALE;
1445 				}
1446 				rep->r_rtt = 0;
1447 			}
1448 		}
1449 	}
1450 
1451 #ifdef NFSSERVER
1452 	/*
1453 	 * Call the nqnfs server timer once a second to handle leases.
1454 	 */
1455 	if (lasttime != time.tv_sec) {
1456 		lasttime = time.tv_sec;
1457 		nqnfs_serverd();
1458 	}
1459 
1460 	/*
1461 	 * Scan the write gathering queues for writes that need to be
1462 	 * completed now.
1463 	 */
1464 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1465 	for (slp = nfssvc_sockhead.tqh_first; slp != 0;
1466 	    slp = slp->ns_chain.tqe_next) {
1467 	    if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1468 		nfsrv_wakenfsd(slp);
1469 	}
1470 #endif /* NFSSERVER */
1471 	splx(s);
1472 	callout_reset(&nfs_timer_ch, nfs_ticks, nfs_timer, NULL);
1473 }
1474 
1475 /*
1476  * Test for a termination condition pending on the process.
1477  * This is used for NFSMNT_INT mounts.
1478  */
1479 int
1480 nfs_sigintr(nmp, rep, p)
1481 	struct nfsmount *nmp;
1482 	struct nfsreq *rep;
1483 	struct proc *p;
1484 {
1485 	sigset_t ss;
1486 
1487 	if (rep && (rep->r_flags & R_SOFTTERM))
1488 		return (EINTR);
1489 	if (!(nmp->nm_flag & NFSMNT_INT))
1490 		return (0);
1491 	if (p) {
1492 		sigpending1(p, &ss);
1493 #if 0
1494 		sigminusset(&p->p_sigctx.ps_sigignore, &ss);
1495 #endif
1496 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1497 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1498 		    sigismember(&ss, SIGQUIT))
1499 			return (EINTR);
1500 	}
1501 	return (0);
1502 }
1503 
1504 /*
1505  * Lock a socket against others.
1506  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1507  * and also to avoid race conditions between the processes with nfs requests
1508  * in progress when a reconnect is necessary.
1509  */
1510 int
1511 nfs_sndlock(flagp, rep)
1512 	int *flagp;
1513 	struct nfsreq *rep;
1514 {
1515 	struct proc *p;
1516 	int slpflag = 0, slptimeo = 0;
1517 
1518 	if (rep) {
1519 		p = rep->r_procp;
1520 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1521 			slpflag = PCATCH;
1522 	} else
1523 		p = (struct proc *)0;
1524 	while (*flagp & NFSMNT_SNDLOCK) {
1525 		if (nfs_sigintr(rep->r_nmp, rep, p))
1526 			return (EINTR);
1527 		*flagp |= NFSMNT_WANTSND;
1528 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1529 			slptimeo);
1530 		if (slpflag == PCATCH) {
1531 			slpflag = 0;
1532 			slptimeo = 2 * hz;
1533 		}
1534 	}
1535 	*flagp |= NFSMNT_SNDLOCK;
1536 	return (0);
1537 }
1538 
1539 /*
1540  * Unlock the stream socket for others.
1541  */
1542 void
1543 nfs_sndunlock(flagp)
1544 	int *flagp;
1545 {
1546 
1547 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1548 		panic("nfs sndunlock");
1549 	*flagp &= ~NFSMNT_SNDLOCK;
1550 	if (*flagp & NFSMNT_WANTSND) {
1551 		*flagp &= ~NFSMNT_WANTSND;
1552 		wakeup((caddr_t)flagp);
1553 	}
1554 }
1555 
1556 int
1557 nfs_rcvlock(rep)
1558 	struct nfsreq *rep;
1559 {
1560 	struct nfsmount *nmp = rep->r_nmp;
1561 	int *flagp = &nmp->nm_iflag;
1562 	int slpflag, slptimeo = 0;
1563 
1564 	if (*flagp & NFSMNT_DISMNT)
1565 		return EIO;
1566 
1567 	if (*flagp & NFSMNT_INT)
1568 		slpflag = PCATCH;
1569 	else
1570 		slpflag = 0;
1571 	while (*flagp & NFSMNT_RCVLOCK) {
1572 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1573 			return (EINTR);
1574 		*flagp |= NFSMNT_WANTRCV;
1575 		nmp->nm_waiters++;
1576 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1577 			slptimeo);
1578 		nmp->nm_waiters--;
1579 		if (*flagp & NFSMNT_DISMNT) {
1580 			wakeup(&nmp->nm_waiters);
1581 			return EIO;
1582 		}
1583 		/* If our reply was received while we were sleeping,
1584 		 * then just return without taking the lock to avoid a
1585 		 * situation where a single iod could 'capture' the
1586 		 * receive lock.
1587 		 */
1588 		if (rep->r_mrep != NULL)
1589 			return (EALREADY);
1590 		if (slpflag == PCATCH) {
1591 			slpflag = 0;
1592 			slptimeo = 2 * hz;
1593 		}
1594 	}
1595 	*flagp |= NFSMNT_RCVLOCK;
1596 	return (0);
1597 }
1598 
1599 /*
1600  * Unlock the stream socket for others.
1601  */
1602 void
1603 nfs_rcvunlock(flagp)
1604 	int *flagp;
1605 {
1606 
1607 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1608 		panic("nfs rcvunlock");
1609 	*flagp &= ~NFSMNT_RCVLOCK;
1610 	if (*flagp & NFSMNT_WANTRCV) {
1611 		*flagp &= ~NFSMNT_WANTRCV;
1612 		wakeup((caddr_t)flagp);
1613 	}
1614 }
1615 
1616 /*
1617  * Parse an RPC request
1618  * - verify it
1619  * - fill in the cred struct.
1620  */
1621 int
1622 nfs_getreq(nd, nfsd, has_header)
1623 	struct nfsrv_descript *nd;
1624 	struct nfsd *nfsd;
1625 	int has_header;
1626 {
1627 	int len, i;
1628 	u_int32_t *tl;
1629 	int32_t t1;
1630 	struct uio uio;
1631 	struct iovec iov;
1632 	caddr_t dpos, cp2, cp;
1633 	u_int32_t nfsvers, auth_type;
1634 	uid_t nickuid;
1635 	int error = 0, nqnfs = 0, ticklen;
1636 	struct mbuf *mrep, *md;
1637 	struct nfsuid *nuidp;
1638 	struct timeval tvin, tvout;
1639 
1640 	mrep = nd->nd_mrep;
1641 	md = nd->nd_md;
1642 	dpos = nd->nd_dpos;
1643 	if (has_header) {
1644 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1645 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1646 		if (*tl++ != rpc_call) {
1647 			m_freem(mrep);
1648 			return (EBADRPC);
1649 		}
1650 	} else
1651 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1652 	nd->nd_repstat = 0;
1653 	nd->nd_flag = 0;
1654 	if (*tl++ != rpc_vers) {
1655 		nd->nd_repstat = ERPCMISMATCH;
1656 		nd->nd_procnum = NFSPROC_NOOP;
1657 		return (0);
1658 	}
1659 	if (*tl != nfs_prog) {
1660 		if (*tl == nqnfs_prog)
1661 			nqnfs++;
1662 		else {
1663 			nd->nd_repstat = EPROGUNAVAIL;
1664 			nd->nd_procnum = NFSPROC_NOOP;
1665 			return (0);
1666 		}
1667 	}
1668 	tl++;
1669 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1670 	if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1671 		(nfsvers != NQNFS_VER3 && nqnfs)) {
1672 		nd->nd_repstat = EPROGMISMATCH;
1673 		nd->nd_procnum = NFSPROC_NOOP;
1674 		return (0);
1675 	}
1676 	if (nqnfs)
1677 		nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1678 	else if (nfsvers == NFS_VER3)
1679 		nd->nd_flag = ND_NFSV3;
1680 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1681 	if (nd->nd_procnum == NFSPROC_NULL)
1682 		return (0);
1683 	if (nd->nd_procnum >= NFS_NPROCS ||
1684 		(!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1685 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1686 		nd->nd_repstat = EPROCUNAVAIL;
1687 		nd->nd_procnum = NFSPROC_NOOP;
1688 		return (0);
1689 	}
1690 	if ((nd->nd_flag & ND_NFSV3) == 0)
1691 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1692 	auth_type = *tl++;
1693 	len = fxdr_unsigned(int, *tl++);
1694 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1695 		m_freem(mrep);
1696 		return (EBADRPC);
1697 	}
1698 
1699 	nd->nd_flag &= ~ND_KERBAUTH;
1700 	/*
1701 	 * Handle auth_unix or auth_kerb.
1702 	 */
1703 	if (auth_type == rpc_auth_unix) {
1704 		len = fxdr_unsigned(int, *++tl);
1705 		if (len < 0 || len > NFS_MAXNAMLEN) {
1706 			m_freem(mrep);
1707 			return (EBADRPC);
1708 		}
1709 		nfsm_adv(nfsm_rndup(len));
1710 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1711 		memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred));
1712 		nd->nd_cr.cr_ref = 1;
1713 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1714 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1715 		len = fxdr_unsigned(int, *tl);
1716 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1717 			m_freem(mrep);
1718 			return (EBADRPC);
1719 		}
1720 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1721 		for (i = 0; i < len; i++)
1722 		    if (i < NGROUPS)
1723 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1724 		    else
1725 			tl++;
1726 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1727 		if (nd->nd_cr.cr_ngroups > 1)
1728 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1729 		len = fxdr_unsigned(int, *++tl);
1730 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1731 			m_freem(mrep);
1732 			return (EBADRPC);
1733 		}
1734 		if (len > 0)
1735 			nfsm_adv(nfsm_rndup(len));
1736 	} else if (auth_type == rpc_auth_kerb) {
1737 		switch (fxdr_unsigned(int, *tl++)) {
1738 		case RPCAKN_FULLNAME:
1739 			ticklen = fxdr_unsigned(int, *tl);
1740 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1741 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1742 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1743 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1744 				m_freem(mrep);
1745 				return (EBADRPC);
1746 			}
1747 			uio.uio_offset = 0;
1748 			uio.uio_iov = &iov;
1749 			uio.uio_iovcnt = 1;
1750 			uio.uio_segflg = UIO_SYSSPACE;
1751 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1752 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1753 			nfsm_mtouio(&uio, uio.uio_resid);
1754 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1755 			if (*tl++ != rpc_auth_kerb ||
1756 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1757 				printf("Bad kerb verifier\n");
1758 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1759 				nd->nd_procnum = NFSPROC_NOOP;
1760 				return (0);
1761 			}
1762 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1763 			tl = (u_int32_t *)cp;
1764 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1765 				printf("Not fullname kerb verifier\n");
1766 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1767 				nd->nd_procnum = NFSPROC_NOOP;
1768 				return (0);
1769 			}
1770 			cp += NFSX_UNSIGNED;
1771 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
1772 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1773 			nd->nd_flag |= ND_KERBFULL;
1774 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1775 			break;
1776 		case RPCAKN_NICKNAME:
1777 			if (len != 2 * NFSX_UNSIGNED) {
1778 				printf("Kerb nickname short\n");
1779 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1780 				nd->nd_procnum = NFSPROC_NOOP;
1781 				return (0);
1782 			}
1783 			nickuid = fxdr_unsigned(uid_t, *tl);
1784 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1785 			if (*tl++ != rpc_auth_kerb ||
1786 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1787 				printf("Kerb nick verifier bad\n");
1788 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1789 				nd->nd_procnum = NFSPROC_NOOP;
1790 				return (0);
1791 			}
1792 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1793 			tvin.tv_sec = *tl++;
1794 			tvin.tv_usec = *tl;
1795 
1796 			for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1797 			    nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1798 				if (nuidp->nu_cr.cr_uid == nickuid &&
1799 				    (!nd->nd_nam2 ||
1800 				     netaddr_match(NU_NETFAM(nuidp),
1801 				      &nuidp->nu_haddr, nd->nd_nam2)))
1802 					break;
1803 			}
1804 			if (!nuidp) {
1805 				nd->nd_repstat =
1806 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1807 				nd->nd_procnum = NFSPROC_NOOP;
1808 				return (0);
1809 			}
1810 
1811 			/*
1812 			 * Now, decrypt the timestamp using the session key
1813 			 * and validate it.
1814 			 */
1815 #ifdef NFSKERB
1816 			XXX
1817 #endif
1818 
1819 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1820 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1821 			if (nuidp->nu_expire < time.tv_sec ||
1822 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1823 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1824 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1825 				nuidp->nu_expire = 0;
1826 				nd->nd_repstat =
1827 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1828 				nd->nd_procnum = NFSPROC_NOOP;
1829 				return (0);
1830 			}
1831 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1832 			nd->nd_flag |= ND_KERBNICK;
1833 		};
1834 	} else {
1835 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1836 		nd->nd_procnum = NFSPROC_NOOP;
1837 		return (0);
1838 	}
1839 
1840 	/*
1841 	 * For nqnfs, get piggybacked lease request.
1842 	 */
1843 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
1844 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1845 		nd->nd_flag |= fxdr_unsigned(int, *tl);
1846 		if (nd->nd_flag & ND_LEASE) {
1847 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1848 			nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
1849 		} else
1850 			nd->nd_duration = NQ_MINLEASE;
1851 	} else
1852 		nd->nd_duration = NQ_MINLEASE;
1853 	nd->nd_md = md;
1854 	nd->nd_dpos = dpos;
1855 	return (0);
1856 nfsmout:
1857 	return (error);
1858 }
1859 
1860 int
1861 nfs_msg(p, server, msg)
1862 	struct proc *p;
1863 	char *server, *msg;
1864 {
1865 	tpr_t tpr;
1866 
1867 	if (p)
1868 		tpr = tprintf_open(p);
1869 	else
1870 		tpr = NULL;
1871 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1872 	tprintf_close(tpr);
1873 	return (0);
1874 }
1875 
1876 #ifdef NFSSERVER
1877 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
1878 				    struct nfssvc_sock *, struct proc *,
1879 				    struct mbuf **)) = {
1880 	nfsrv_null,
1881 	nfsrv_getattr,
1882 	nfsrv_setattr,
1883 	nfsrv_lookup,
1884 	nfsrv3_access,
1885 	nfsrv_readlink,
1886 	nfsrv_read,
1887 	nfsrv_write,
1888 	nfsrv_create,
1889 	nfsrv_mkdir,
1890 	nfsrv_symlink,
1891 	nfsrv_mknod,
1892 	nfsrv_remove,
1893 	nfsrv_rmdir,
1894 	nfsrv_rename,
1895 	nfsrv_link,
1896 	nfsrv_readdir,
1897 	nfsrv_readdirplus,
1898 	nfsrv_statfs,
1899 	nfsrv_fsinfo,
1900 	nfsrv_pathconf,
1901 	nfsrv_commit,
1902 	nqnfsrv_getlease,
1903 	nqnfsrv_vacated,
1904 	nfsrv_noop,
1905 	nfsrv_noop
1906 };
1907 
1908 /*
1909  * Socket upcall routine for the nfsd sockets.
1910  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1911  * Essentially do as much as possible non-blocking, else punt and it will
1912  * be called with M_WAIT from an nfsd.
1913  */
1914 void
1915 nfsrv_rcv(so, arg, waitflag)
1916 	struct socket *so;
1917 	caddr_t arg;
1918 	int waitflag;
1919 {
1920 	struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1921 	struct mbuf *m;
1922 	struct mbuf *mp, *nam;
1923 	struct uio auio;
1924 	int flags, error;
1925 
1926 	if ((slp->ns_flag & SLP_VALID) == 0)
1927 		return;
1928 #ifdef notdef
1929 	/*
1930 	 * Define this to test for nfsds handling this under heavy load.
1931 	 */
1932 	if (waitflag == M_DONTWAIT) {
1933 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1934 	}
1935 #endif
1936 	auio.uio_procp = NULL;
1937 	if (so->so_type == SOCK_STREAM) {
1938 		/*
1939 		 * If there are already records on the queue, defer soreceive()
1940 		 * to an nfsd so that there is feedback to the TCP layer that
1941 		 * the nfs servers are heavily loaded.
1942 		 */
1943 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1944 			slp->ns_flag |= SLP_NEEDQ;
1945 			goto dorecs;
1946 		}
1947 
1948 		/*
1949 		 * Do soreceive().
1950 		 */
1951 		auio.uio_resid = 1000000000;
1952 		flags = MSG_DONTWAIT;
1953 		error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1954 		if (error || mp == (struct mbuf *)0) {
1955 			if (error == EWOULDBLOCK)
1956 				slp->ns_flag |= SLP_NEEDQ;
1957 			else
1958 				slp->ns_flag |= SLP_DISCONN;
1959 			goto dorecs;
1960 		}
1961 		m = mp;
1962 		if (slp->ns_rawend) {
1963 			slp->ns_rawend->m_next = m;
1964 			slp->ns_cc += 1000000000 - auio.uio_resid;
1965 		} else {
1966 			slp->ns_raw = m;
1967 			slp->ns_cc = 1000000000 - auio.uio_resid;
1968 		}
1969 		while (m->m_next)
1970 			m = m->m_next;
1971 		slp->ns_rawend = m;
1972 
1973 		/*
1974 		 * Now try and parse record(s) out of the raw stream data.
1975 		 */
1976 		error = nfsrv_getstream(slp, waitflag);
1977 		if (error) {
1978 			if (error == EPERM)
1979 				slp->ns_flag |= SLP_DISCONN;
1980 			else
1981 				slp->ns_flag |= SLP_NEEDQ;
1982 		}
1983 	} else {
1984 		do {
1985 			auio.uio_resid = 1000000000;
1986 			flags = MSG_DONTWAIT;
1987 			error = (*so->so_receive)(so, &nam, &auio, &mp,
1988 						(struct mbuf **)0, &flags);
1989 			if (mp) {
1990 				if (nam) {
1991 					m = nam;
1992 					m->m_next = mp;
1993 				} else
1994 					m = mp;
1995 				if (slp->ns_recend)
1996 					slp->ns_recend->m_nextpkt = m;
1997 				else
1998 					slp->ns_rec = m;
1999 				slp->ns_recend = m;
2000 				m->m_nextpkt = (struct mbuf *)0;
2001 			}
2002 			if (error) {
2003 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2004 					&& error != EWOULDBLOCK) {
2005 					slp->ns_flag |= SLP_DISCONN;
2006 					goto dorecs;
2007 				}
2008 			}
2009 		} while (mp);
2010 	}
2011 
2012 	/*
2013 	 * Now try and process the request records, non-blocking.
2014 	 */
2015 dorecs:
2016 	if (waitflag == M_DONTWAIT &&
2017 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
2018 		nfsrv_wakenfsd(slp);
2019 }
2020 
2021 /*
2022  * Try and extract an RPC request from the mbuf data list received on a
2023  * stream socket. The "waitflag" argument indicates whether or not it
2024  * can sleep.
2025  */
2026 int
2027 nfsrv_getstream(slp, waitflag)
2028 	struct nfssvc_sock *slp;
2029 	int waitflag;
2030 {
2031 	struct mbuf *m, **mpp;
2032 	char *cp1, *cp2;
2033 	int len;
2034 	struct mbuf *om, *m2, *recm = NULL;
2035 	u_int32_t recmark;
2036 
2037 	if (slp->ns_flag & SLP_GETSTREAM)
2038 		panic("nfs getstream");
2039 	slp->ns_flag |= SLP_GETSTREAM;
2040 	for (;;) {
2041 	    if (slp->ns_reclen == 0) {
2042 		if (slp->ns_cc < NFSX_UNSIGNED) {
2043 			slp->ns_flag &= ~SLP_GETSTREAM;
2044 			return (0);
2045 		}
2046 		m = slp->ns_raw;
2047 		if (m->m_len >= NFSX_UNSIGNED) {
2048 			memcpy((caddr_t)&recmark, mtod(m, caddr_t), NFSX_UNSIGNED);
2049 			m->m_data += NFSX_UNSIGNED;
2050 			m->m_len -= NFSX_UNSIGNED;
2051 		} else {
2052 			cp1 = (caddr_t)&recmark;
2053 			cp2 = mtod(m, caddr_t);
2054 			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
2055 				while (m->m_len == 0) {
2056 					m = m->m_next;
2057 					cp2 = mtod(m, caddr_t);
2058 				}
2059 				*cp1++ = *cp2++;
2060 				m->m_data++;
2061 				m->m_len--;
2062 			}
2063 			slp->ns_raw = m;
2064 		}
2065 		slp->ns_cc -= NFSX_UNSIGNED;
2066 		recmark = ntohl(recmark);
2067 		slp->ns_reclen = recmark & ~0x80000000;
2068 		if (recmark & 0x80000000)
2069 			slp->ns_flag |= SLP_LASTFRAG;
2070 		else
2071 			slp->ns_flag &= ~SLP_LASTFRAG;
2072 		if (slp->ns_reclen > NFS_MAXPACKET) {
2073 			slp->ns_flag &= ~SLP_GETSTREAM;
2074 			return (EPERM);
2075 		}
2076 	    }
2077 
2078 	    /*
2079 	     * Now get the record part.
2080 	     */
2081 	    if (slp->ns_cc == slp->ns_reclen) {
2082 		recm = slp->ns_raw;
2083 		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2084 		slp->ns_cc = slp->ns_reclen = 0;
2085 	    } else if (slp->ns_cc > slp->ns_reclen) {
2086 		len = 0;
2087 		m = slp->ns_raw;
2088 		om = (struct mbuf *)0;
2089 		while (len < slp->ns_reclen) {
2090 			if ((len + m->m_len) > slp->ns_reclen) {
2091 				size_t left = slp->ns_reclen - len;
2092 
2093 				MGETHDR(m2, waitflag, m->m_type);
2094 				if (m2 == NULL) {
2095 					slp->ns_flag &= ~SLP_GETSTREAM;
2096 					return (EWOULDBLOCK);
2097 				}
2098 				if (left > MHLEN) {
2099 					MCLGET(m2, waitflag);
2100 					if (!(m2->m_flags & M_EXT)) {
2101 						m_freem(m2);
2102 						slp->ns_flag &= ~SLP_GETSTREAM;
2103 						return (EWOULDBLOCK);
2104 					}
2105 				}
2106 				memcpy(mtod(m2, caddr_t), mtod(m, caddr_t),
2107 				    left);
2108 				m2->m_len = left;
2109 				m->m_data += left;
2110 				m->m_len -= left;
2111 				if (om) {
2112 					om->m_next = m2;
2113 					recm = slp->ns_raw;
2114 				} else
2115 					recm = m2;
2116 				len = slp->ns_reclen;
2117 			} else if ((len + m->m_len) == slp->ns_reclen) {
2118 				om = m;
2119 				len += m->m_len;
2120 				m = m->m_next;
2121 				recm = slp->ns_raw;
2122 				om->m_next = (struct mbuf *)0;
2123 			} else {
2124 				om = m;
2125 				len += m->m_len;
2126 				m = m->m_next;
2127 			}
2128 		}
2129 		slp->ns_raw = m;
2130 		slp->ns_cc -= len;
2131 		slp->ns_reclen = 0;
2132 	    } else {
2133 		slp->ns_flag &= ~SLP_GETSTREAM;
2134 		return (0);
2135 	    }
2136 
2137 	    /*
2138 	     * Accumulate the fragments into a record.
2139 	     */
2140 	    mpp = &slp->ns_frag;
2141 	    while (*mpp)
2142 		mpp = &((*mpp)->m_next);
2143 	    *mpp = recm;
2144 	    if (slp->ns_flag & SLP_LASTFRAG) {
2145 		if (slp->ns_recend)
2146 		    slp->ns_recend->m_nextpkt = slp->ns_frag;
2147 		else
2148 		    slp->ns_rec = slp->ns_frag;
2149 		slp->ns_recend = slp->ns_frag;
2150 		slp->ns_frag = (struct mbuf *)0;
2151 	    }
2152 	}
2153 }
2154 
2155 /*
2156  * Parse an RPC header.
2157  */
2158 int
2159 nfsrv_dorec(slp, nfsd, ndp)
2160 	struct nfssvc_sock *slp;
2161 	struct nfsd *nfsd;
2162 	struct nfsrv_descript **ndp;
2163 {
2164 	struct mbuf *m, *nam;
2165 	struct nfsrv_descript *nd;
2166 	int error;
2167 
2168 	*ndp = NULL;
2169 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2170 	    (m = slp->ns_rec) == (struct mbuf *)0)
2171 		return (ENOBUFS);
2172 	slp->ns_rec = m->m_nextpkt;
2173 	if (slp->ns_rec)
2174 		m->m_nextpkt = (struct mbuf *)0;
2175 	else
2176 		slp->ns_recend = (struct mbuf *)0;
2177 	if (m->m_type == MT_SONAME) {
2178 		nam = m;
2179 		m = m->m_next;
2180 		nam->m_next = NULL;
2181 	} else
2182 		nam = NULL;
2183 	MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2184 		M_NFSRVDESC, M_WAITOK);
2185 	nd->nd_md = nd->nd_mrep = m;
2186 	nd->nd_nam2 = nam;
2187 	nd->nd_dpos = mtod(m, caddr_t);
2188 	error = nfs_getreq(nd, nfsd, TRUE);
2189 	if (error) {
2190 		m_freem(nam);
2191 		free((caddr_t)nd, M_NFSRVDESC);
2192 		return (error);
2193 	}
2194 	*ndp = nd;
2195 	nfsd->nfsd_nd = nd;
2196 	return (0);
2197 }
2198 
2199 
2200 /*
2201  * Search for a sleeping nfsd and wake it up.
2202  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2203  * running nfsds will go look for the work in the nfssvc_sock list.
2204  */
2205 void
2206 nfsrv_wakenfsd(slp)
2207 	struct nfssvc_sock *slp;
2208 {
2209 	struct nfsd *nd;
2210 
2211 	if ((slp->ns_flag & SLP_VALID) == 0)
2212 		return;
2213 	for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) {
2214 		if (nd->nfsd_flag & NFSD_WAITING) {
2215 			nd->nfsd_flag &= ~NFSD_WAITING;
2216 			if (nd->nfsd_slp)
2217 				panic("nfsd wakeup");
2218 			slp->ns_sref++;
2219 			nd->nfsd_slp = slp;
2220 			wakeup((caddr_t)nd);
2221 			return;
2222 		}
2223 	}
2224 	slp->ns_flag |= SLP_DOREC;
2225 	nfsd_head_flag |= NFSD_CHECKSLP;
2226 }
2227 #endif /* NFSSERVER */
2228