xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 990562bfef3015a6e18a209dde598d5588ea144c)
1 /*	$NetBSD: nfs_socket.c,v 1.78 2003/02/01 06:23:49 thorpej 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 1.78 2003/02/01 06:23:49 thorpej 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 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
86 
87 #define	TRUE	1
88 #define	FALSE	0
89 
90 /*
91  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
92  * Use the mean and mean deviation of rtt for the appropriate type of rpc
93  * for the frequent rpcs and a default for the others.
94  * The justification for doing "other" this way is that these rpcs
95  * happen so infrequently that timer est. would probably be stale.
96  * Also, since many of these rpcs are
97  * non-idempotent, a conservative timeout is desired.
98  * getattr, lookup - A+2D
99  * read, write     - A+4D
100  * other           - nm_timeo
101  */
102 #define	NFS_RTO(n, t) \
103 	((t) == 0 ? (n)->nm_timeo : \
104 	 ((t) < 3 ? \
105 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
106 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
107 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
108 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
109 /*
110  * External data, mostly RPC constants in XDR form
111  */
112 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
113 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
114 	rpc_auth_kerb;
115 extern u_int32_t nfs_prog, nqnfs_prog;
116 extern time_t nqnfsstarttime;
117 extern const int nfsv3_procid[NFS_NPROCS];
118 extern int nfs_ticks;
119 
120 /*
121  * Defines which timer to use for the procnum.
122  * 0 - default
123  * 1 - getattr
124  * 2 - lookup
125  * 3 - read
126  * 4 - write
127  */
128 static const int proct[NFS_NPROCS] = {
129 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
130 	0, 0, 0,
131 };
132 
133 /*
134  * There is a congestion window for outstanding rpcs maintained per mount
135  * point. The cwnd size is adjusted in roughly the way that:
136  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
137  * SIGCOMM '88". ACM, August 1988.
138  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
139  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
140  * of rpcs is in progress.
141  * (The sent count and cwnd are scaled for integer arith.)
142  * Variants of "slow start" were tried and were found to be too much of a
143  * performance hit (ave. rtt 3 times larger),
144  * I suspect due to the large rtt that nfs rpcs have.
145  */
146 #define	NFS_CWNDSCALE	256
147 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
148 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
149 int nfsrtton = 0;
150 struct nfsrtt nfsrtt;
151 struct nfsreqhead nfs_reqq;
152 
153 struct callout nfs_timer_ch = CALLOUT_INITIALIZER;
154 
155 /*
156  * Initialize sockets and congestion for a new NFS connection.
157  * We do not free the sockaddr if error.
158  */
159 int
160 nfs_connect(nmp, rep)
161 	struct nfsmount *nmp;
162 	struct nfsreq *rep;
163 {
164 	struct socket *so;
165 	int s, error, rcvreserve, sndreserve;
166 	struct sockaddr *saddr;
167 	struct sockaddr_in *sin;
168 #ifdef INET6
169 	struct sockaddr_in6 *sin6;
170 #endif
171 	struct mbuf *m;
172 
173 	nmp->nm_so = (struct socket *)0;
174 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
175 	error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
176 		nmp->nm_soproto);
177 	if (error)
178 		goto bad;
179 	so = nmp->nm_so;
180 	nmp->nm_soflags = so->so_proto->pr_flags;
181 
182 	/*
183 	 * Some servers require that the client port be a reserved port number.
184 	 */
185 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
186 		MGET(m, M_WAIT, MT_SOOPTS);
187 		*mtod(m, int32_t *) = IP_PORTRANGE_LOW;
188 		m->m_len = sizeof(int32_t);
189 		if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m)))
190 			goto bad;
191 		MGET(m, M_WAIT, MT_SONAME);
192 		sin = mtod(m, struct sockaddr_in *);
193 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
194 		sin->sin_family = AF_INET;
195 		sin->sin_addr.s_addr = INADDR_ANY;
196 		sin->sin_port = 0;
197 		error = sobind(so, m, &proc0);
198 		m_freem(m);
199 		if (error)
200 			goto bad;
201 	}
202 #ifdef INET6
203 	if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) {
204 		MGET(m, M_WAIT, MT_SOOPTS);
205 		*mtod(m, int32_t *) = IPV6_PORTRANGE_LOW;
206 		m->m_len = sizeof(int32_t);
207 		if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m)))
208 			goto bad;
209 		MGET(m, M_WAIT, MT_SONAME);
210 		sin6 = mtod(m, struct sockaddr_in6 *);
211 		sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6);
212 		sin6->sin6_family = AF_INET6;
213 		sin6->sin6_addr = in6addr_any;
214 		sin6->sin6_port = 0;
215 		error = sobind(so, m, &proc0);
216 		m_freem(m);
217 		if (error)
218 			goto bad;
219 	}
220 #endif
221 
222 	/*
223 	 * Protocols that do not require connections may be optionally left
224 	 * unconnected for servers that reply from a port other than NFS_PORT.
225 	 */
226 	if (nmp->nm_flag & NFSMNT_NOCONN) {
227 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
228 			error = ENOTCONN;
229 			goto bad;
230 		}
231 	} else {
232 		error = soconnect(so, nmp->nm_nam);
233 		if (error)
234 			goto bad;
235 
236 		/*
237 		 * Wait for the connection to complete. Cribbed from the
238 		 * connect system call but with the wait timing out so
239 		 * that interruptible mounts don't hang here for a long time.
240 		 */
241 		s = splsoftnet();
242 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
243 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
244 				"nfscn1", 2 * hz);
245 			if ((so->so_state & SS_ISCONNECTING) &&
246 			    so->so_error == 0 && rep &&
247 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
248 				so->so_state &= ~SS_ISCONNECTING;
249 				splx(s);
250 				goto bad;
251 			}
252 		}
253 		if (so->so_error) {
254 			error = so->so_error;
255 			so->so_error = 0;
256 			splx(s);
257 			goto bad;
258 		}
259 		splx(s);
260 	}
261 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
262 		so->so_rcv.sb_timeo = (5 * hz);
263 		so->so_snd.sb_timeo = (5 * hz);
264 	} else {
265 		so->so_rcv.sb_timeo = 0;
266 		so->so_snd.sb_timeo = 0;
267 	}
268 	if (nmp->nm_sotype == SOCK_DGRAM) {
269 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
270 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
271 		    NFS_MAXPKTHDR) * 2;
272 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
273 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
274 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
275 		    NFS_MAXPKTHDR) * 2;
276 	} else {
277 		if (nmp->nm_sotype != SOCK_STREAM)
278 			panic("nfscon sotype");
279 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
280 			MGET(m, M_WAIT, MT_SOOPTS);
281 			*mtod(m, int32_t *) = 1;
282 			m->m_len = sizeof(int32_t);
283 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
284 		}
285 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
286 			MGET(m, M_WAIT, MT_SOOPTS);
287 			*mtod(m, int32_t *) = 1;
288 			m->m_len = sizeof(int32_t);
289 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
290 		}
291 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
292 		    sizeof (u_int32_t)) * 2;
293 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
294 		    sizeof (u_int32_t)) * 2;
295 	}
296 	error = soreserve(so, sndreserve, rcvreserve);
297 	if (error)
298 		goto bad;
299 	so->so_rcv.sb_flags |= SB_NOINTR;
300 	so->so_snd.sb_flags |= SB_NOINTR;
301 
302 	/* Initialize other non-zero congestion variables */
303 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
304 		NFS_TIMEO << 3;
305 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
306 		nmp->nm_sdrtt[3] = 0;
307 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
308 	nmp->nm_sent = 0;
309 	nmp->nm_timeouts = 0;
310 	return (0);
311 
312 bad:
313 	nfs_disconnect(nmp);
314 	return (error);
315 }
316 
317 /*
318  * Reconnect routine:
319  * Called when a connection is broken on a reliable protocol.
320  * - clean up the old socket
321  * - nfs_connect() again
322  * - set R_MUSTRESEND for all outstanding requests on mount point
323  * If this fails the mount point is DEAD!
324  * nb: Must be called with the nfs_sndlock() set on the mount point.
325  */
326 int
327 nfs_reconnect(rep)
328 	struct nfsreq *rep;
329 {
330 	struct nfsreq *rp;
331 	struct nfsmount *nmp = rep->r_nmp;
332 	int error;
333 
334 	nfs_disconnect(nmp);
335 	while ((error = nfs_connect(nmp, rep)) != 0) {
336 		if (error == EINTR || error == ERESTART)
337 			return (EINTR);
338 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0);
339 	}
340 
341 	/*
342 	 * Loop through outstanding request list and fix up all requests
343 	 * on old socket.
344 	 */
345 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
346 		if (rp->r_nmp == nmp)
347 			rp->r_flags |= R_MUSTRESEND;
348 	}
349 	return (0);
350 }
351 
352 /*
353  * NFS disconnect. Clean up and unlink.
354  */
355 void
356 nfs_disconnect(nmp)
357 	struct nfsmount *nmp;
358 {
359 	struct socket *so;
360 	int drain = 0;
361 
362 	if (nmp->nm_so) {
363 		so = nmp->nm_so;
364 		nmp->nm_so = (struct socket *)0;
365 		soshutdown(so, 2);
366 		drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
367 		if (drain) {
368 			/*
369 			 * soshutdown() above should wake up the current
370 			 * listener.
371 			 * Now wake up those waiting for the receive lock, and
372 			 * wait for them to go away unhappy, to prevent *nmp
373 			 * from evaporating while they're sleeping.
374 			 */
375 			while (nmp->nm_waiters > 0) {
376 				wakeup (&nmp->nm_iflag);
377 				(void) tsleep(&nmp->nm_waiters, PVFS,
378 				    "nfsdis", 0);
379 			}
380 		}
381 		soclose(so);
382 	}
383 #ifdef DIAGNOSTIC
384 	if (drain && (nmp->nm_waiters > 0))
385 		panic("nfs_disconnect: waiters left after drain?");
386 #endif
387 }
388 
389 void
390 nfs_safedisconnect(nmp)
391 	struct nfsmount *nmp;
392 {
393 	struct nfsreq dummyreq;
394 
395 	memset(&dummyreq, 0, sizeof(dummyreq));
396 	dummyreq.r_nmp = nmp;
397 	nfs_rcvlock(&dummyreq); /* XXX ignored error return */
398 	nfs_disconnect(nmp);
399 	nfs_rcvunlock(&nmp->nm_iflag);
400 }
401 
402 /*
403  * This is the nfs send routine. For connection based socket types, it
404  * must be called with an nfs_sndlock() on the socket.
405  * "rep == NULL" indicates that it has been called from a server.
406  * For the client side:
407  * - return EINTR if the RPC is terminated, 0 otherwise
408  * - set R_MUSTRESEND if the send fails for any reason
409  * - do any cleanup required by recoverable socket errors (? ? ?)
410  * For the server side:
411  * - return EINTR or ERESTART if interrupted by a signal
412  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
413  * - do any cleanup required by recoverable socket errors (? ? ?)
414  */
415 int
416 nfs_send(so, nam, top, rep)
417 	struct socket *so;
418 	struct mbuf *nam;
419 	struct mbuf *top;
420 	struct nfsreq *rep;
421 {
422 	struct mbuf *sendnam;
423 	int error, soflags, flags;
424 
425 	if (rep) {
426 		if (rep->r_flags & R_SOFTTERM) {
427 			m_freem(top);
428 			return (EINTR);
429 		}
430 		if ((so = rep->r_nmp->nm_so) == NULL) {
431 			rep->r_flags |= R_MUSTRESEND;
432 			m_freem(top);
433 			return (0);
434 		}
435 		rep->r_flags &= ~R_MUSTRESEND;
436 		soflags = rep->r_nmp->nm_soflags;
437 	} else
438 		soflags = so->so_proto->pr_flags;
439 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
440 		sendnam = (struct mbuf *)0;
441 	else
442 		sendnam = nam;
443 	if (so->so_type == SOCK_SEQPACKET)
444 		flags = MSG_EOR;
445 	else
446 		flags = 0;
447 
448 	error = (*so->so_send)(so, sendnam, (struct uio *)0, top,
449 		(struct mbuf *)0, flags);
450 	if (error) {
451 		if (rep) {
452 			if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
453 				/*
454 				 * We're too fast for the network/driver,
455 				 * and UDP isn't flowcontrolled.
456 				 * We need to resend. This is not fatal,
457 				 * just try again.
458 				 *
459 				 * Could be smarter here by doing some sort
460 				 * of a backoff, but this is rare.
461 				 */
462 				rep->r_flags |= R_MUSTRESEND;
463 			} else {
464 				log(LOG_INFO, "nfs send error %d for %s\n",
465 				    error,
466 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
467 				/*
468 				 * Deal with errors for the client side.
469 				 */
470 				if (rep->r_flags & R_SOFTTERM)
471 					error = EINTR;
472 				else
473 					rep->r_flags |= R_MUSTRESEND;
474 			}
475 		} else {
476 			/*
477 			 * See above. This error can happen under normal
478 			 * circumstances and the log is too noisy.
479 			 * The error will still show up in nfsstat.
480 			 */
481 			if (error != ENOBUFS || so->so_type != SOCK_DGRAM)
482 				log(LOG_INFO, "nfsd send error %d\n", error);
483 		}
484 
485 		/*
486 		 * Handle any recoverable (soft) socket errors here. (? ? ?)
487 		 */
488 		if (error != EINTR && error != ERESTART &&
489 			error != EWOULDBLOCK && error != EPIPE)
490 			error = 0;
491 	}
492 	return (error);
493 }
494 
495 #ifdef NFS
496 /*
497  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
498  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
499  * Mark and consolidate the data into a new mbuf list.
500  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
501  *     small mbufs.
502  * For SOCK_STREAM we must be very careful to read an entire record once
503  * we have read any of it, even if the system call has been interrupted.
504  */
505 int
506 nfs_receive(rep, aname, mp)
507 	struct nfsreq *rep;
508 	struct mbuf **aname;
509 	struct mbuf **mp;
510 {
511 	struct socket *so;
512 	struct uio auio;
513 	struct iovec aio;
514 	struct mbuf *m;
515 	struct mbuf *control;
516 	u_int32_t len;
517 	struct mbuf **getnam;
518 	int error, sotype, rcvflg;
519 	struct proc *p = curproc;	/* XXX */
520 
521 	/*
522 	 * Set up arguments for soreceive()
523 	 */
524 	*mp = (struct mbuf *)0;
525 	*aname = (struct mbuf *)0;
526 	sotype = rep->r_nmp->nm_sotype;
527 
528 	/*
529 	 * For reliable protocols, lock against other senders/receivers
530 	 * in case a reconnect is necessary.
531 	 * For SOCK_STREAM, first get the Record Mark to find out how much
532 	 * more there is to get.
533 	 * We must lock the socket against other receivers
534 	 * until we have an entire rpc request/reply.
535 	 */
536 	if (sotype != SOCK_DGRAM) {
537 		error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
538 		if (error)
539 			return (error);
540 tryagain:
541 		/*
542 		 * Check for fatal errors and resending request.
543 		 */
544 		/*
545 		 * Ugh: If a reconnect attempt just happened, nm_so
546 		 * would have changed. NULL indicates a failed
547 		 * attempt that has essentially shut down this
548 		 * mount point.
549 		 */
550 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
551 			nfs_sndunlock(&rep->r_nmp->nm_iflag);
552 			return (EINTR);
553 		}
554 		so = rep->r_nmp->nm_so;
555 		if (!so) {
556 			error = nfs_reconnect(rep);
557 			if (error) {
558 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
559 				return (error);
560 			}
561 			goto tryagain;
562 		}
563 		while (rep->r_flags & R_MUSTRESEND) {
564 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
565 			nfsstats.rpcretries++;
566 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
567 			if (error) {
568 				if (error == EINTR || error == ERESTART ||
569 				    (error = nfs_reconnect(rep)) != 0) {
570 					nfs_sndunlock(&rep->r_nmp->nm_iflag);
571 					return (error);
572 				}
573 				goto tryagain;
574 			}
575 		}
576 		nfs_sndunlock(&rep->r_nmp->nm_iflag);
577 		if (sotype == SOCK_STREAM) {
578 			aio.iov_base = (caddr_t) &len;
579 			aio.iov_len = sizeof(u_int32_t);
580 			auio.uio_iov = &aio;
581 			auio.uio_iovcnt = 1;
582 			auio.uio_segflg = UIO_SYSSPACE;
583 			auio.uio_rw = UIO_READ;
584 			auio.uio_offset = 0;
585 			auio.uio_resid = sizeof(u_int32_t);
586 			auio.uio_procp = p;
587 			do {
588 			   rcvflg = MSG_WAITALL;
589 			   error = (*so->so_receive)(so, (struct mbuf **)0, &auio,
590 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
591 			   if (error == EWOULDBLOCK && rep) {
592 				if (rep->r_flags & R_SOFTTERM)
593 					return (EINTR);
594 			   }
595 			} while (error == EWOULDBLOCK);
596 			if (!error && auio.uio_resid > 0) {
597 			    /*
598 			     * Don't log a 0 byte receive; it means
599 			     * that the socket has been closed, and
600 			     * can happen during normal operation
601 			     * (forcible unmount or Solaris server).
602 			     */
603 			    if (auio.uio_resid != sizeof (u_int32_t))
604 			      log(LOG_INFO,
605 				 "short receive (%lu/%lu) from nfs server %s\n",
606 				 (u_long)sizeof(u_int32_t) - auio.uio_resid,
607 				 (u_long)sizeof(u_int32_t),
608 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
609 			    error = EPIPE;
610 			}
611 			if (error)
612 				goto errout;
613 			len = ntohl(len) & ~0x80000000;
614 			/*
615 			 * This is SERIOUS! We are out of sync with the sender
616 			 * and forcing a disconnect/reconnect is all I can do.
617 			 */
618 			if (len > NFS_MAXPACKET) {
619 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
620 				"impossible packet length",
621 				len,
622 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
623 			    error = EFBIG;
624 			    goto errout;
625 			}
626 			auio.uio_resid = len;
627 			do {
628 			    rcvflg = MSG_WAITALL;
629 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
630 				&auio, mp, (struct mbuf **)0, &rcvflg);
631 			} while (error == EWOULDBLOCK || error == EINTR ||
632 				 error == ERESTART);
633 			if (!error && auio.uio_resid > 0) {
634 			    if (len != auio.uio_resid)
635 			      log(LOG_INFO,
636 				"short receive (%lu/%d) from nfs server %s\n",
637 				(u_long)len - auio.uio_resid, len,
638 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
639 			    error = EPIPE;
640 			}
641 		} else {
642 			/*
643 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
644 			 * and soreceive() will return when it has either a
645 			 * control msg or a data msg.
646 			 * We have no use for control msg., but must grab them
647 			 * and then throw them away so we know what is going
648 			 * on.
649 			 */
650 			auio.uio_resid = len = 100000000; /* Anything Big */
651 			auio.uio_procp = p;
652 			do {
653 			    rcvflg = 0;
654 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
655 				&auio, mp, &control, &rcvflg);
656 			    if (control)
657 				m_freem(control);
658 			    if (error == EWOULDBLOCK && rep) {
659 				if (rep->r_flags & R_SOFTTERM)
660 					return (EINTR);
661 			    }
662 			} while (error == EWOULDBLOCK ||
663 				 (!error && *mp == NULL && control));
664 			if ((rcvflg & MSG_EOR) == 0)
665 				printf("Egad!!\n");
666 			if (!error && *mp == NULL)
667 				error = EPIPE;
668 			len -= auio.uio_resid;
669 		}
670 errout:
671 		if (error && error != EINTR && error != ERESTART) {
672 			m_freem(*mp);
673 			*mp = (struct mbuf *)0;
674 			if (error != EPIPE)
675 				log(LOG_INFO,
676 				    "receive error %d from nfs server %s\n",
677 				    error,
678 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
679 			error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
680 			if (!error)
681 				error = nfs_reconnect(rep);
682 			if (!error)
683 				goto tryagain;
684 			else
685 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
686 		}
687 	} else {
688 		if ((so = rep->r_nmp->nm_so) == NULL)
689 			return (EACCES);
690 		if (so->so_state & SS_ISCONNECTED)
691 			getnam = (struct mbuf **)0;
692 		else
693 			getnam = aname;
694 		auio.uio_resid = len = 1000000;
695 		auio.uio_procp = p;
696 		do {
697 			rcvflg = 0;
698 			error =  (*so->so_receive)(so, getnam, &auio, mp,
699 				(struct mbuf **)0, &rcvflg);
700 			if (error == EWOULDBLOCK &&
701 			    (rep->r_flags & R_SOFTTERM))
702 				return (EINTR);
703 		} while (error == EWOULDBLOCK);
704 		len -= auio.uio_resid;
705 		if (!error && *mp == NULL)
706 			error = EPIPE;
707 	}
708 	if (error) {
709 		m_freem(*mp);
710 		*mp = (struct mbuf *)0;
711 	}
712 	return (error);
713 }
714 
715 /*
716  * Implement receipt of reply on a socket.
717  * We must search through the list of received datagrams matching them
718  * with outstanding requests using the xid, until ours is found.
719  */
720 /* ARGSUSED */
721 int
722 nfs_reply(myrep)
723 	struct nfsreq *myrep;
724 {
725 	struct nfsreq *rep;
726 	struct nfsmount *nmp = myrep->r_nmp;
727 	int32_t t1;
728 	struct mbuf *mrep, *nam, *md;
729 	u_int32_t rxid, *tl;
730 	caddr_t dpos, cp2;
731 	int error;
732 
733 	/*
734 	 * Loop around until we get our own reply
735 	 */
736 	for (;;) {
737 		/*
738 		 * Lock against other receivers so that I don't get stuck in
739 		 * sbwait() after someone else has received my reply for me.
740 		 * Also necessary for connection based protocols to avoid
741 		 * race conditions during a reconnect.
742 		 */
743 		error = nfs_rcvlock(myrep);
744 		if (error == EALREADY)
745 			return (0);
746 		if (error)
747 			return (error);
748 		/*
749 		 * Get the next Rpc reply off the socket
750 		 */
751 		nmp->nm_waiters++;
752 		error = nfs_receive(myrep, &nam, &mrep);
753 		nfs_rcvunlock(&nmp->nm_iflag);
754 		if (error) {
755 
756 			if (nmp->nm_iflag & NFSMNT_DISMNT) {
757 				/*
758 				 * Oops, we're going away now..
759 				 */
760 				nmp->nm_waiters--;
761 				wakeup (&nmp->nm_waiters);
762 				return error;
763 			}
764 			nmp->nm_waiters--;
765 			/*
766 			 * Ignore routing errors on connectionless protocols? ?
767 			 */
768 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
769 				nmp->nm_so->so_error = 0;
770 #ifdef DEBUG
771 				printf("nfs_reply: ignoring error %d\n", error);
772 #endif
773 				if (myrep->r_flags & R_GETONEREP)
774 					return (0);
775 				continue;
776 			}
777 			return (error);
778 		}
779 		nmp->nm_waiters--;
780 		if (nam)
781 			m_freem(nam);
782 
783 		/*
784 		 * Get the xid and check that it is an rpc reply
785 		 */
786 		md = mrep;
787 		dpos = mtod(md, caddr_t);
788 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
789 		rxid = *tl++;
790 		if (*tl != rpc_reply) {
791 #ifndef NFS_V2_ONLY
792 			if (nmp->nm_flag & NFSMNT_NQNFS) {
793 				if (nqnfs_callback(nmp, mrep, md, dpos))
794 					nfsstats.rpcinvalid++;
795 			} else
796 #endif
797 			{
798 				nfsstats.rpcinvalid++;
799 				m_freem(mrep);
800 			}
801 nfsmout:
802 			if (myrep->r_flags & R_GETONEREP)
803 				return (0);
804 			continue;
805 		}
806 
807 		/*
808 		 * Loop through the request list to match up the reply
809 		 * Iff no match, just drop the datagram
810 		 */
811 		TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
812 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
813 				/* Found it.. */
814 				rep->r_mrep = mrep;
815 				rep->r_md = md;
816 				rep->r_dpos = dpos;
817 				if (nfsrtton) {
818 					struct rttl *rt;
819 
820 					rt = &nfsrtt.rttl[nfsrtt.pos];
821 					rt->proc = rep->r_procnum;
822 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
823 					rt->sent = nmp->nm_sent;
824 					rt->cwnd = nmp->nm_cwnd;
825 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
826 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
827 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
828 					rt->tstamp = time;
829 					if (rep->r_flags & R_TIMING)
830 						rt->rtt = rep->r_rtt;
831 					else
832 						rt->rtt = 1000000;
833 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
834 				}
835 				/*
836 				 * Update congestion window.
837 				 * Do the additive increase of
838 				 * one rpc/rtt.
839 				 */
840 				if (nmp->nm_cwnd <= nmp->nm_sent) {
841 					nmp->nm_cwnd +=
842 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
843 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
844 					if (nmp->nm_cwnd > NFS_MAXCWND)
845 						nmp->nm_cwnd = NFS_MAXCWND;
846 				}
847 				rep->r_flags &= ~R_SENT;
848 				nmp->nm_sent -= NFS_CWNDSCALE;
849 				/*
850 				 * Update rtt using a gain of 0.125 on the mean
851 				 * and a gain of 0.25 on the deviation.
852 				 */
853 				if (rep->r_flags & R_TIMING) {
854 					/*
855 					 * Since the timer resolution of
856 					 * NFS_HZ is so course, it can often
857 					 * result in r_rtt == 0. Since
858 					 * r_rtt == N means that the actual
859 					 * rtt is between N+dt and N+2-dt ticks,
860 					 * add 1.
861 					 */
862 					t1 = rep->r_rtt + 1;
863 					t1 -= (NFS_SRTT(rep) >> 3);
864 					NFS_SRTT(rep) += t1;
865 					if (t1 < 0)
866 						t1 = -t1;
867 					t1 -= (NFS_SDRTT(rep) >> 2);
868 					NFS_SDRTT(rep) += t1;
869 				}
870 				nmp->nm_timeouts = 0;
871 				break;
872 			}
873 		}
874 		/*
875 		 * If not matched to a request, drop it.
876 		 * If it's mine, get out.
877 		 */
878 		if (rep == 0) {
879 			nfsstats.rpcunexpected++;
880 			m_freem(mrep);
881 		} else if (rep == myrep) {
882 			if (rep->r_mrep == NULL)
883 				panic("nfsreply nil");
884 			return (0);
885 		}
886 		if (myrep->r_flags & R_GETONEREP)
887 			return (0);
888 	}
889 }
890 
891 /*
892  * nfs_request - goes something like this
893  *	- fill in request struct
894  *	- links it into list
895  *	- calls nfs_send() for first transmit
896  *	- calls nfs_receive() to get reply
897  *	- break down rpc header and return with nfs reply pointed to
898  *	  by mrep or error
899  * nb: always frees up mreq mbuf list
900  */
901 int
902 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
903 	struct vnode *vp;
904 	struct mbuf *mrest;
905 	int procnum;
906 	struct proc *procp;
907 	struct ucred *cred;
908 	struct mbuf **mrp;
909 	struct mbuf **mdp;
910 	caddr_t *dposp;
911 {
912 	struct mbuf *m, *mrep;
913 	struct nfsreq *rep;
914 	u_int32_t *tl;
915 	int i;
916 	struct nfsmount *nmp;
917 	struct mbuf *md, *mheadend;
918 	char nickv[RPCX_NICKVERF];
919 	time_t reqtime, waituntil;
920 	caddr_t dpos, cp2;
921 	int t1, s, error = 0, mrest_len, auth_len, auth_type;
922 	int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0;
923 	int verf_len, verf_type;
924 	u_int32_t xid;
925 	char *auth_str, *verf_str;
926 	NFSKERBKEY_T key;		/* save session key */
927 #ifndef NFS_V2_ONLY
928 	int nqlflag, cachable;
929 	u_quad_t frev;
930 	struct nfsnode *np;
931 #endif
932 
933 	KASSERT(cred != NULL);
934 	nmp = VFSTONFS(vp->v_mount);
935 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
936 	rep->r_nmp = nmp;
937 	rep->r_vp = vp;
938 	rep->r_procp = procp;
939 	rep->r_procnum = procnum;
940 	i = 0;
941 	m = mrest;
942 	while (m) {
943 		i += m->m_len;
944 		m = m->m_next;
945 	}
946 	mrest_len = i;
947 
948 	/*
949 	 * Get the RPC header with authorization.
950 	 */
951 kerbauth:
952 	verf_str = auth_str = (char *)0;
953 	if (nmp->nm_flag & NFSMNT_KERB) {
954 		verf_str = nickv;
955 		verf_len = sizeof (nickv);
956 		auth_type = RPCAUTH_KERB4;
957 		memset((caddr_t)key, 0, sizeof (key));
958 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
959 			&auth_len, verf_str, verf_len)) {
960 			error = nfs_getauth(nmp, rep, cred, &auth_str,
961 				&auth_len, verf_str, &verf_len, key);
962 			if (error) {
963 				free((caddr_t)rep, M_NFSREQ);
964 				m_freem(mrest);
965 				return (error);
966 			}
967 		}
968 	} else {
969 		auth_type = RPCAUTH_UNIX;
970 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
971 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
972 			5 * NFSX_UNSIGNED;
973 	}
974 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
975 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
976 	if (auth_str)
977 		free(auth_str, M_TEMP);
978 
979 	/*
980 	 * For stream protocols, insert a Sun RPC Record Mark.
981 	 */
982 	if (nmp->nm_sotype == SOCK_STREAM) {
983 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
984 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
985 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
986 	}
987 	rep->r_mreq = m;
988 	rep->r_xid = xid;
989 tryagain:
990 	if (nmp->nm_flag & NFSMNT_SOFT)
991 		rep->r_retry = nmp->nm_retry;
992 	else
993 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
994 	rep->r_rtt = rep->r_rexmit = 0;
995 	if (proct[procnum] > 0)
996 		rep->r_flags = R_TIMING;
997 	else
998 		rep->r_flags = 0;
999 	rep->r_mrep = NULL;
1000 
1001 	/*
1002 	 * Do the client side RPC.
1003 	 */
1004 	nfsstats.rpcrequests++;
1005 	/*
1006 	 * Chain request into list of outstanding requests. Be sure
1007 	 * to put it LAST so timer finds oldest requests first.
1008 	 */
1009 	s = splsoftnet();
1010 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1011 
1012 	/* Get send time for nqnfs */
1013 	reqtime = time.tv_sec;
1014 
1015 	/*
1016 	 * If backing off another request or avoiding congestion, don't
1017 	 * send this one now but let timer do it. If not timing a request,
1018 	 * do it now.
1019 	 */
1020 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1021 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1022 		nmp->nm_sent < nmp->nm_cwnd)) {
1023 		splx(s);
1024 		if (nmp->nm_soflags & PR_CONNREQUIRED)
1025 			error = nfs_sndlock(&nmp->nm_iflag, rep);
1026 		if (!error) {
1027 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
1028 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep);
1029 			if (nmp->nm_soflags & PR_CONNREQUIRED)
1030 				nfs_sndunlock(&nmp->nm_iflag);
1031 		}
1032 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1033 			nmp->nm_sent += NFS_CWNDSCALE;
1034 			rep->r_flags |= R_SENT;
1035 		}
1036 	} else {
1037 		splx(s);
1038 		rep->r_rtt = -1;
1039 	}
1040 
1041 	/*
1042 	 * Wait for the reply from our send or the timer's.
1043 	 */
1044 	if (!error || error == EPIPE)
1045 		error = nfs_reply(rep);
1046 
1047 	/*
1048 	 * RPC done, unlink the request.
1049 	 */
1050 	s = splsoftnet();
1051 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1052 	splx(s);
1053 
1054 	/*
1055 	 * Decrement the outstanding request count.
1056 	 */
1057 	if (rep->r_flags & R_SENT) {
1058 		rep->r_flags &= ~R_SENT;	/* paranoia */
1059 		nmp->nm_sent -= NFS_CWNDSCALE;
1060 	}
1061 
1062 	/*
1063 	 * If there was a successful reply and a tprintf msg.
1064 	 * tprintf a response.
1065 	 */
1066 	if (!error && (rep->r_flags & R_TPRINTFMSG))
1067 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1068 		    "is alive again");
1069 	mrep = rep->r_mrep;
1070 	md = rep->r_md;
1071 	dpos = rep->r_dpos;
1072 	if (error) {
1073 		m_freem(rep->r_mreq);
1074 		free((caddr_t)rep, M_NFSREQ);
1075 		return (error);
1076 	}
1077 
1078 	/*
1079 	 * break down the rpc header and check if ok
1080 	 */
1081 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1082 	if (*tl++ == rpc_msgdenied) {
1083 		if (*tl == rpc_mismatch)
1084 			error = EOPNOTSUPP;
1085 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1086 			if (!failed_auth) {
1087 				failed_auth++;
1088 				mheadend->m_next = (struct mbuf *)0;
1089 				m_freem(mrep);
1090 				m_freem(rep->r_mreq);
1091 				goto kerbauth;
1092 			} else
1093 				error = EAUTH;
1094 		} else
1095 			error = EACCES;
1096 		m_freem(mrep);
1097 		m_freem(rep->r_mreq);
1098 		free((caddr_t)rep, M_NFSREQ);
1099 		return (error);
1100 	}
1101 
1102 	/*
1103 	 * Grab any Kerberos verifier, otherwise just throw it away.
1104 	 */
1105 	verf_type = fxdr_unsigned(int, *tl++);
1106 	i = fxdr_unsigned(int32_t, *tl);
1107 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1108 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1109 		if (error)
1110 			goto nfsmout;
1111 	} else if (i > 0)
1112 		nfsm_adv(nfsm_rndup(i));
1113 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1114 	/* 0 == ok */
1115 	if (*tl == 0) {
1116 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1117 		if (*tl != 0) {
1118 			error = fxdr_unsigned(int, *tl);
1119 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1120 				error == NFSERR_TRYLATER) {
1121 				m_freem(mrep);
1122 				error = 0;
1123 				waituntil = time.tv_sec + trylater_delay;
1124 				while (time.tv_sec < waituntil)
1125 					(void) tsleep((caddr_t)&lbolt,
1126 						PSOCK, "nqnfstry", 0);
1127 				trylater_delay *= nfs_backoff[trylater_cnt];
1128 				if (trylater_cnt < 7)
1129 					trylater_cnt++;
1130 				goto tryagain;
1131 			}
1132 
1133 			/*
1134 			 * If the File Handle was stale, invalidate the
1135 			 * lookup cache, just in case.
1136 			 */
1137 			if (error == ESTALE)
1138 				cache_purge(vp);
1139 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1140 				*mrp = mrep;
1141 				*mdp = md;
1142 				*dposp = dpos;
1143 				error |= NFSERR_RETERR;
1144 			} else
1145 				m_freem(mrep);
1146 			m_freem(rep->r_mreq);
1147 			free((caddr_t)rep, M_NFSREQ);
1148 			return (error);
1149 		}
1150 
1151 #ifndef NFS_V2_ONLY
1152 		/*
1153 		 * For nqnfs, get any lease in reply
1154 		 */
1155 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1156 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1157 			if (*tl) {
1158 				np = VTONFS(vp);
1159 				nqlflag = fxdr_unsigned(int, *tl);
1160 				nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1161 				cachable = fxdr_unsigned(int, *tl++);
1162 				reqtime += fxdr_unsigned(int, *tl++);
1163 				if (reqtime > time.tv_sec) {
1164 				    frev = fxdr_hyper(tl);
1165 				    nqnfs_clientlease(nmp, np, nqlflag,
1166 					cachable, reqtime, frev);
1167 				}
1168 			}
1169 		}
1170 #endif
1171 		*mrp = mrep;
1172 		*mdp = md;
1173 		*dposp = dpos;
1174 		m_freem(rep->r_mreq);
1175 		FREE((caddr_t)rep, M_NFSREQ);
1176 		return (0);
1177 	}
1178 	m_freem(mrep);
1179 	error = EPROTONOSUPPORT;
1180 nfsmout:
1181 	m_freem(rep->r_mreq);
1182 	free((caddr_t)rep, M_NFSREQ);
1183 	return (error);
1184 }
1185 #endif /* NFS */
1186 
1187 /*
1188  * Generate the rpc reply header
1189  * siz arg. is used to decide if adding a cluster is worthwhile
1190  */
1191 int
1192 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1193 	int siz;
1194 	struct nfsrv_descript *nd;
1195 	struct nfssvc_sock *slp;
1196 	int err;
1197 	int cache;
1198 	u_quad_t *frev;
1199 	struct mbuf **mrq;
1200 	struct mbuf **mbp;
1201 	caddr_t *bposp;
1202 {
1203 	u_int32_t *tl;
1204 	struct mbuf *mreq;
1205 	caddr_t bpos;
1206 	struct mbuf *mb, *mb2;
1207 
1208 	MGETHDR(mreq, M_WAIT, MT_DATA);
1209 	mb = mreq;
1210 	/*
1211 	 * If this is a big reply, use a cluster else
1212 	 * try and leave leading space for the lower level headers.
1213 	 */
1214 	siz += RPC_REPLYSIZ;
1215 	if (siz >= max_datalen) {
1216 		MCLGET(mreq, M_WAIT);
1217 	} else
1218 		mreq->m_data += max_hdr;
1219 	tl = mtod(mreq, u_int32_t *);
1220 	mreq->m_len = 6 * NFSX_UNSIGNED;
1221 	bpos = ((caddr_t)tl) + mreq->m_len;
1222 	*tl++ = txdr_unsigned(nd->nd_retxid);
1223 	*tl++ = rpc_reply;
1224 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1225 		*tl++ = rpc_msgdenied;
1226 		if (err & NFSERR_AUTHERR) {
1227 			*tl++ = rpc_autherr;
1228 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1229 			mreq->m_len -= NFSX_UNSIGNED;
1230 			bpos -= NFSX_UNSIGNED;
1231 		} else {
1232 			*tl++ = rpc_mismatch;
1233 			*tl++ = txdr_unsigned(RPC_VER2);
1234 			*tl = txdr_unsigned(RPC_VER2);
1235 		}
1236 	} else {
1237 		*tl++ = rpc_msgaccepted;
1238 
1239 		/*
1240 		 * For Kerberos authentication, we must send the nickname
1241 		 * verifier back, otherwise just RPCAUTH_NULL.
1242 		 */
1243 		if (nd->nd_flag & ND_KERBFULL) {
1244 		    struct nfsuid *nuidp;
1245 		    struct timeval ktvin, ktvout;
1246 
1247 		    for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1248 			nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1249 			if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1250 			    (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1251 			     &nuidp->nu_haddr, nd->nd_nam2)))
1252 			    break;
1253 		    }
1254 		    if (nuidp) {
1255 			ktvin.tv_sec =
1256 			    txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1257 			ktvin.tv_usec =
1258 			    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1259 
1260 			/*
1261 			 * Encrypt the timestamp in ecb mode using the
1262 			 * session key.
1263 			 */
1264 #ifdef NFSKERB
1265 			XXX
1266 #endif
1267 
1268 			*tl++ = rpc_auth_kerb;
1269 			*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1270 			*tl = ktvout.tv_sec;
1271 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1272 			*tl++ = ktvout.tv_usec;
1273 			*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1274 		    } else {
1275 			*tl++ = 0;
1276 			*tl++ = 0;
1277 		    }
1278 		} else {
1279 			*tl++ = 0;
1280 			*tl++ = 0;
1281 		}
1282 		switch (err) {
1283 		case EPROGUNAVAIL:
1284 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1285 			break;
1286 		case EPROGMISMATCH:
1287 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1288 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1289 			if (nd->nd_flag & ND_NQNFS) {
1290 				*tl++ = txdr_unsigned(3);
1291 				*tl = txdr_unsigned(3);
1292 			} else {
1293 				*tl++ = txdr_unsigned(2);
1294 				*tl = txdr_unsigned(3);
1295 			}
1296 			break;
1297 		case EPROCUNAVAIL:
1298 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1299 			break;
1300 		case EBADRPC:
1301 			*tl = txdr_unsigned(RPC_GARBAGE);
1302 			break;
1303 		default:
1304 			*tl = 0;
1305 			if (err != NFSERR_RETVOID) {
1306 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1307 				if (err)
1308 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1309 				else
1310 				    *tl = 0;
1311 			}
1312 			break;
1313 		};
1314 	}
1315 
1316 	/*
1317 	 * For nqnfs, piggyback lease as requested.
1318 	 */
1319 	if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1320 		if (nd->nd_flag & ND_LEASE) {
1321 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1322 			*tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1323 			*tl++ = txdr_unsigned(cache);
1324 			*tl++ = txdr_unsigned(nd->nd_duration);
1325 			txdr_hyper(*frev, tl);
1326 		} else {
1327 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1328 			*tl = 0;
1329 		}
1330 	}
1331 	if (mrq != NULL)
1332 		*mrq = mreq;
1333 	*mbp = mb;
1334 	*bposp = bpos;
1335 	if (err != 0 && err != NFSERR_RETVOID)
1336 		nfsstats.srvrpc_errs++;
1337 	return (0);
1338 }
1339 
1340 /*
1341  * Nfs timer routine
1342  * Scan the nfsreq list and retranmit any requests that have timed out
1343  * To avoid retransmission attempts on STREAM sockets (in the future) make
1344  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1345  */
1346 void
1347 nfs_timer(arg)
1348 	void *arg;	/* never used */
1349 {
1350 	struct nfsreq *rep;
1351 	struct mbuf *m;
1352 	struct socket *so;
1353 	struct nfsmount *nmp;
1354 	int timeo;
1355 	int s, error;
1356 #ifdef NFSSERVER
1357 	struct nfssvc_sock *slp;
1358 	static long lasttime = 0;
1359 	u_quad_t cur_usec;
1360 #endif
1361 
1362 	s = splsoftnet();
1363 	TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
1364 		nmp = rep->r_nmp;
1365 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1366 			continue;
1367 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1368 			rep->r_flags |= R_SOFTTERM;
1369 			continue;
1370 		}
1371 		if (rep->r_rtt >= 0) {
1372 			rep->r_rtt++;
1373 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1374 				timeo = nmp->nm_timeo;
1375 			else
1376 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1377 			if (nmp->nm_timeouts > 0)
1378 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1379 			if (rep->r_rtt <= timeo)
1380 				continue;
1381 			if (nmp->nm_timeouts < 8)
1382 				nmp->nm_timeouts++;
1383 		}
1384 		/*
1385 		 * Check for server not responding
1386 		 */
1387 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1388 		     rep->r_rexmit > nmp->nm_deadthresh) {
1389 			nfs_msg(rep->r_procp,
1390 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1391 			    "not responding");
1392 			rep->r_flags |= R_TPRINTFMSG;
1393 		}
1394 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1395 			nfsstats.rpctimeouts++;
1396 			rep->r_flags |= R_SOFTTERM;
1397 			continue;
1398 		}
1399 		if (nmp->nm_sotype != SOCK_DGRAM) {
1400 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1401 				rep->r_rexmit = NFS_MAXREXMIT;
1402 			continue;
1403 		}
1404 		if ((so = nmp->nm_so) == NULL)
1405 			continue;
1406 
1407 		/*
1408 		 * If there is enough space and the window allows..
1409 		 *	Resend it
1410 		 * Set r_rtt to -1 in case we fail to send it now.
1411 		 */
1412 		rep->r_rtt = -1;
1413 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1414 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1415 		    (rep->r_flags & R_SENT) ||
1416 		    nmp->nm_sent < nmp->nm_cwnd) &&
1417 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1418 		        if (so->so_state & SS_ISCONNECTED)
1419 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1420 			    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1421 			else
1422 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1423 			    nmp->nm_nam, (struct mbuf *)0, (struct proc *)0);
1424 			if (error) {
1425 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1426 #ifdef DEBUG
1427 					printf("nfs_timer: ignoring error %d\n",
1428 						error);
1429 #endif
1430 					so->so_error = 0;
1431 				}
1432 			} else {
1433 				/*
1434 				 * Iff first send, start timing
1435 				 * else turn timing off, backoff timer
1436 				 * and divide congestion window by 2.
1437 				 */
1438 				if (rep->r_flags & R_SENT) {
1439 					rep->r_flags &= ~R_TIMING;
1440 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1441 						rep->r_rexmit = NFS_MAXREXMIT;
1442 					nmp->nm_cwnd >>= 1;
1443 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1444 						nmp->nm_cwnd = NFS_CWNDSCALE;
1445 					nfsstats.rpcretries++;
1446 				} else {
1447 					rep->r_flags |= R_SENT;
1448 					nmp->nm_sent += NFS_CWNDSCALE;
1449 				}
1450 				rep->r_rtt = 0;
1451 			}
1452 		}
1453 	}
1454 
1455 #ifdef NFSSERVER
1456 	/*
1457 	 * Call the nqnfs server timer once a second to handle leases.
1458 	 */
1459 	if (lasttime != time.tv_sec) {
1460 		lasttime = time.tv_sec;
1461 		nqnfs_serverd();
1462 	}
1463 
1464 	/*
1465 	 * Scan the write gathering queues for writes that need to be
1466 	 * completed now.
1467 	 */
1468 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1469 	TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
1470 	    if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1471 		nfsrv_wakenfsd(slp);
1472 	}
1473 #endif /* NFSSERVER */
1474 	splx(s);
1475 	callout_reset(&nfs_timer_ch, nfs_ticks, nfs_timer, NULL);
1476 }
1477 
1478 /*ARGSUSED*/
1479 void
1480 nfs_exit(p, v)
1481 	struct proc *p;
1482 	void *v;
1483 {
1484 	struct nfsreq *rp;
1485 	int s = splsoftnet();
1486 
1487 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
1488 		if (rp->r_procp == p)
1489 			TAILQ_REMOVE(&nfs_reqq, rp, r_chain);
1490 	}
1491 	splx(s);
1492 }
1493 
1494 /*
1495  * Test for a termination condition pending on the process.
1496  * This is used for NFSMNT_INT mounts.
1497  */
1498 int
1499 nfs_sigintr(nmp, rep, p)
1500 	struct nfsmount *nmp;
1501 	struct nfsreq *rep;
1502 	struct proc *p;
1503 {
1504 	sigset_t ss;
1505 
1506 	if (rep && (rep->r_flags & R_SOFTTERM))
1507 		return (EINTR);
1508 	if (!(nmp->nm_flag & NFSMNT_INT))
1509 		return (0);
1510 	if (p) {
1511 		sigpending1(p, &ss);
1512 #if 0
1513 		sigminusset(&p->p_sigctx.ps_sigignore, &ss);
1514 #endif
1515 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1516 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1517 		    sigismember(&ss, SIGQUIT))
1518 			return (EINTR);
1519 	}
1520 	return (0);
1521 }
1522 
1523 /*
1524  * Lock a socket against others.
1525  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1526  * and also to avoid race conditions between the processes with nfs requests
1527  * in progress when a reconnect is necessary.
1528  */
1529 int
1530 nfs_sndlock(flagp, rep)
1531 	int *flagp;
1532 	struct nfsreq *rep;
1533 {
1534 	struct proc *p;
1535 	int slpflag = 0, slptimeo = 0;
1536 
1537 	if (rep) {
1538 		p = rep->r_procp;
1539 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1540 			slpflag = PCATCH;
1541 	} else
1542 		p = (struct proc *)0;
1543 	while (*flagp & NFSMNT_SNDLOCK) {
1544 		if (nfs_sigintr(rep->r_nmp, rep, p))
1545 			return (EINTR);
1546 		*flagp |= NFSMNT_WANTSND;
1547 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1548 			slptimeo);
1549 		if (slpflag == PCATCH) {
1550 			slpflag = 0;
1551 			slptimeo = 2 * hz;
1552 		}
1553 	}
1554 	*flagp |= NFSMNT_SNDLOCK;
1555 	return (0);
1556 }
1557 
1558 /*
1559  * Unlock the stream socket for others.
1560  */
1561 void
1562 nfs_sndunlock(flagp)
1563 	int *flagp;
1564 {
1565 
1566 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1567 		panic("nfs sndunlock");
1568 	*flagp &= ~NFSMNT_SNDLOCK;
1569 	if (*flagp & NFSMNT_WANTSND) {
1570 		*flagp &= ~NFSMNT_WANTSND;
1571 		wakeup((caddr_t)flagp);
1572 	}
1573 }
1574 
1575 int
1576 nfs_rcvlock(rep)
1577 	struct nfsreq *rep;
1578 {
1579 	struct nfsmount *nmp = rep->r_nmp;
1580 	int *flagp = &nmp->nm_iflag;
1581 	int slpflag, slptimeo = 0;
1582 
1583 	if (*flagp & NFSMNT_DISMNT)
1584 		return EIO;
1585 
1586 	if (*flagp & NFSMNT_INT)
1587 		slpflag = PCATCH;
1588 	else
1589 		slpflag = 0;
1590 	while (*flagp & NFSMNT_RCVLOCK) {
1591 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1592 			return (EINTR);
1593 		*flagp |= NFSMNT_WANTRCV;
1594 		nmp->nm_waiters++;
1595 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1596 			slptimeo);
1597 		nmp->nm_waiters--;
1598 		if (*flagp & NFSMNT_DISMNT) {
1599 			wakeup(&nmp->nm_waiters);
1600 			return EIO;
1601 		}
1602 		/* If our reply was received while we were sleeping,
1603 		 * then just return without taking the lock to avoid a
1604 		 * situation where a single iod could 'capture' the
1605 		 * receive lock.
1606 		 */
1607 		if (rep->r_mrep != NULL)
1608 			return (EALREADY);
1609 		if (slpflag == PCATCH) {
1610 			slpflag = 0;
1611 			slptimeo = 2 * hz;
1612 		}
1613 	}
1614 	*flagp |= NFSMNT_RCVLOCK;
1615 	return (0);
1616 }
1617 
1618 /*
1619  * Unlock the stream socket for others.
1620  */
1621 void
1622 nfs_rcvunlock(flagp)
1623 	int *flagp;
1624 {
1625 
1626 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1627 		panic("nfs rcvunlock");
1628 	*flagp &= ~NFSMNT_RCVLOCK;
1629 	if (*flagp & NFSMNT_WANTRCV) {
1630 		*flagp &= ~NFSMNT_WANTRCV;
1631 		wakeup((caddr_t)flagp);
1632 	}
1633 }
1634 
1635 /*
1636  * Parse an RPC request
1637  * - verify it
1638  * - fill in the cred struct.
1639  */
1640 int
1641 nfs_getreq(nd, nfsd, has_header)
1642 	struct nfsrv_descript *nd;
1643 	struct nfsd *nfsd;
1644 	int has_header;
1645 {
1646 	int len, i;
1647 	u_int32_t *tl;
1648 	int32_t t1;
1649 	struct uio uio;
1650 	struct iovec iov;
1651 	caddr_t dpos, cp2, cp;
1652 	u_int32_t nfsvers, auth_type;
1653 	uid_t nickuid;
1654 	int error = 0, nqnfs = 0, ticklen;
1655 	struct mbuf *mrep, *md;
1656 	struct nfsuid *nuidp;
1657 	struct timeval tvin, tvout;
1658 
1659 	mrep = nd->nd_mrep;
1660 	md = nd->nd_md;
1661 	dpos = nd->nd_dpos;
1662 	if (has_header) {
1663 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1664 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1665 		if (*tl++ != rpc_call) {
1666 			m_freem(mrep);
1667 			return (EBADRPC);
1668 		}
1669 	} else
1670 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1671 	nd->nd_repstat = 0;
1672 	nd->nd_flag = 0;
1673 	if (*tl++ != rpc_vers) {
1674 		nd->nd_repstat = ERPCMISMATCH;
1675 		nd->nd_procnum = NFSPROC_NOOP;
1676 		return (0);
1677 	}
1678 	if (*tl != nfs_prog) {
1679 		if (*tl == nqnfs_prog)
1680 			nqnfs++;
1681 		else {
1682 			nd->nd_repstat = EPROGUNAVAIL;
1683 			nd->nd_procnum = NFSPROC_NOOP;
1684 			return (0);
1685 		}
1686 	}
1687 	tl++;
1688 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1689 	if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1690 		(nfsvers != NQNFS_VER3 && nqnfs)) {
1691 		nd->nd_repstat = EPROGMISMATCH;
1692 		nd->nd_procnum = NFSPROC_NOOP;
1693 		return (0);
1694 	}
1695 	if (nqnfs)
1696 		nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1697 	else if (nfsvers == NFS_VER3)
1698 		nd->nd_flag = ND_NFSV3;
1699 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1700 	if (nd->nd_procnum == NFSPROC_NULL)
1701 		return (0);
1702 	if (nd->nd_procnum >= NFS_NPROCS ||
1703 		(!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1704 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1705 		nd->nd_repstat = EPROCUNAVAIL;
1706 		nd->nd_procnum = NFSPROC_NOOP;
1707 		return (0);
1708 	}
1709 	if ((nd->nd_flag & ND_NFSV3) == 0)
1710 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1711 	auth_type = *tl++;
1712 	len = fxdr_unsigned(int, *tl++);
1713 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1714 		m_freem(mrep);
1715 		return (EBADRPC);
1716 	}
1717 
1718 	nd->nd_flag &= ~ND_KERBAUTH;
1719 	/*
1720 	 * Handle auth_unix or auth_kerb.
1721 	 */
1722 	if (auth_type == rpc_auth_unix) {
1723 		len = fxdr_unsigned(int, *++tl);
1724 		if (len < 0 || len > NFS_MAXNAMLEN) {
1725 			m_freem(mrep);
1726 			return (EBADRPC);
1727 		}
1728 		nfsm_adv(nfsm_rndup(len));
1729 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1730 		memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred));
1731 		nd->nd_cr.cr_ref = 1;
1732 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1733 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1734 		len = fxdr_unsigned(int, *tl);
1735 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1736 			m_freem(mrep);
1737 			return (EBADRPC);
1738 		}
1739 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1740 		for (i = 0; i < len; i++)
1741 		    if (i < NGROUPS)
1742 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1743 		    else
1744 			tl++;
1745 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1746 		if (nd->nd_cr.cr_ngroups > 1)
1747 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1748 		len = fxdr_unsigned(int, *++tl);
1749 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1750 			m_freem(mrep);
1751 			return (EBADRPC);
1752 		}
1753 		if (len > 0)
1754 			nfsm_adv(nfsm_rndup(len));
1755 	} else if (auth_type == rpc_auth_kerb) {
1756 		switch (fxdr_unsigned(int, *tl++)) {
1757 		case RPCAKN_FULLNAME:
1758 			ticklen = fxdr_unsigned(int, *tl);
1759 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1760 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1761 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1762 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1763 				m_freem(mrep);
1764 				return (EBADRPC);
1765 			}
1766 			uio.uio_offset = 0;
1767 			uio.uio_iov = &iov;
1768 			uio.uio_iovcnt = 1;
1769 			uio.uio_segflg = UIO_SYSSPACE;
1770 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1771 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1772 			nfsm_mtouio(&uio, uio.uio_resid);
1773 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1774 			if (*tl++ != rpc_auth_kerb ||
1775 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1776 				printf("Bad kerb verifier\n");
1777 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1778 				nd->nd_procnum = NFSPROC_NOOP;
1779 				return (0);
1780 			}
1781 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1782 			tl = (u_int32_t *)cp;
1783 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1784 				printf("Not fullname kerb verifier\n");
1785 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1786 				nd->nd_procnum = NFSPROC_NOOP;
1787 				return (0);
1788 			}
1789 			cp += NFSX_UNSIGNED;
1790 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
1791 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1792 			nd->nd_flag |= ND_KERBFULL;
1793 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1794 			break;
1795 		case RPCAKN_NICKNAME:
1796 			if (len != 2 * NFSX_UNSIGNED) {
1797 				printf("Kerb nickname short\n");
1798 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1799 				nd->nd_procnum = NFSPROC_NOOP;
1800 				return (0);
1801 			}
1802 			nickuid = fxdr_unsigned(uid_t, *tl);
1803 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1804 			if (*tl++ != rpc_auth_kerb ||
1805 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1806 				printf("Kerb nick verifier bad\n");
1807 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1808 				nd->nd_procnum = NFSPROC_NOOP;
1809 				return (0);
1810 			}
1811 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1812 			tvin.tv_sec = *tl++;
1813 			tvin.tv_usec = *tl;
1814 
1815 			for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1816 			    nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1817 				if (nuidp->nu_cr.cr_uid == nickuid &&
1818 				    (!nd->nd_nam2 ||
1819 				     netaddr_match(NU_NETFAM(nuidp),
1820 				      &nuidp->nu_haddr, nd->nd_nam2)))
1821 					break;
1822 			}
1823 			if (!nuidp) {
1824 				nd->nd_repstat =
1825 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1826 				nd->nd_procnum = NFSPROC_NOOP;
1827 				return (0);
1828 			}
1829 
1830 			/*
1831 			 * Now, decrypt the timestamp using the session key
1832 			 * and validate it.
1833 			 */
1834 #ifdef NFSKERB
1835 			XXX
1836 #endif
1837 
1838 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1839 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1840 			if (nuidp->nu_expire < time.tv_sec ||
1841 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1842 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1843 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1844 				nuidp->nu_expire = 0;
1845 				nd->nd_repstat =
1846 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1847 				nd->nd_procnum = NFSPROC_NOOP;
1848 				return (0);
1849 			}
1850 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1851 			nd->nd_flag |= ND_KERBNICK;
1852 		};
1853 	} else {
1854 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1855 		nd->nd_procnum = NFSPROC_NOOP;
1856 		return (0);
1857 	}
1858 
1859 	/*
1860 	 * For nqnfs, get piggybacked lease request.
1861 	 */
1862 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
1863 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1864 		nd->nd_flag |= fxdr_unsigned(int, *tl);
1865 		if (nd->nd_flag & ND_LEASE) {
1866 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1867 			nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
1868 		} else
1869 			nd->nd_duration = NQ_MINLEASE;
1870 	} else
1871 		nd->nd_duration = NQ_MINLEASE;
1872 	nd->nd_md = md;
1873 	nd->nd_dpos = dpos;
1874 	return (0);
1875 nfsmout:
1876 	return (error);
1877 }
1878 
1879 int
1880 nfs_msg(p, server, msg)
1881 	struct proc *p;
1882 	char *server, *msg;
1883 {
1884 	tpr_t tpr;
1885 
1886 	if (p)
1887 		tpr = tprintf_open(p);
1888 	else
1889 		tpr = NULL;
1890 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1891 	tprintf_close(tpr);
1892 	return (0);
1893 }
1894 
1895 #ifdef NFSSERVER
1896 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
1897 				    struct nfssvc_sock *, struct proc *,
1898 				    struct mbuf **)) = {
1899 	nfsrv_null,
1900 	nfsrv_getattr,
1901 	nfsrv_setattr,
1902 	nfsrv_lookup,
1903 	nfsrv3_access,
1904 	nfsrv_readlink,
1905 	nfsrv_read,
1906 	nfsrv_write,
1907 	nfsrv_create,
1908 	nfsrv_mkdir,
1909 	nfsrv_symlink,
1910 	nfsrv_mknod,
1911 	nfsrv_remove,
1912 	nfsrv_rmdir,
1913 	nfsrv_rename,
1914 	nfsrv_link,
1915 	nfsrv_readdir,
1916 	nfsrv_readdirplus,
1917 	nfsrv_statfs,
1918 	nfsrv_fsinfo,
1919 	nfsrv_pathconf,
1920 	nfsrv_commit,
1921 	nqnfsrv_getlease,
1922 	nqnfsrv_vacated,
1923 	nfsrv_noop,
1924 	nfsrv_noop
1925 };
1926 
1927 /*
1928  * Socket upcall routine for the nfsd sockets.
1929  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1930  * Essentially do as much as possible non-blocking, else punt and it will
1931  * be called with M_WAIT from an nfsd.
1932  */
1933 void
1934 nfsrv_rcv(so, arg, waitflag)
1935 	struct socket *so;
1936 	caddr_t arg;
1937 	int waitflag;
1938 {
1939 	struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1940 	struct mbuf *m;
1941 	struct mbuf *mp, *nam;
1942 	struct uio auio;
1943 	int flags, error;
1944 
1945 	if ((slp->ns_flag & SLP_VALID) == 0)
1946 		return;
1947 #ifdef notdef
1948 	/*
1949 	 * Define this to test for nfsds handling this under heavy load.
1950 	 */
1951 	if (waitflag == M_DONTWAIT) {
1952 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1953 	}
1954 #endif
1955 	auio.uio_procp = NULL;
1956 	if (so->so_type == SOCK_STREAM) {
1957 		/*
1958 		 * If there are already records on the queue, defer soreceive()
1959 		 * to an nfsd so that there is feedback to the TCP layer that
1960 		 * the nfs servers are heavily loaded.
1961 		 */
1962 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1963 			slp->ns_flag |= SLP_NEEDQ;
1964 			goto dorecs;
1965 		}
1966 
1967 		/*
1968 		 * Do soreceive().
1969 		 */
1970 		auio.uio_resid = 1000000000;
1971 		flags = MSG_DONTWAIT;
1972 		error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1973 		if (error || mp == (struct mbuf *)0) {
1974 			if (error == EWOULDBLOCK)
1975 				slp->ns_flag |= SLP_NEEDQ;
1976 			else
1977 				slp->ns_flag |= SLP_DISCONN;
1978 			goto dorecs;
1979 		}
1980 		m = mp;
1981 		if (slp->ns_rawend) {
1982 			slp->ns_rawend->m_next = m;
1983 			slp->ns_cc += 1000000000 - auio.uio_resid;
1984 		} else {
1985 			slp->ns_raw = m;
1986 			slp->ns_cc = 1000000000 - auio.uio_resid;
1987 		}
1988 		while (m->m_next)
1989 			m = m->m_next;
1990 		slp->ns_rawend = m;
1991 
1992 		/*
1993 		 * Now try and parse record(s) out of the raw stream data.
1994 		 */
1995 		error = nfsrv_getstream(slp, waitflag);
1996 		if (error) {
1997 			if (error == EPERM)
1998 				slp->ns_flag |= SLP_DISCONN;
1999 			else
2000 				slp->ns_flag |= SLP_NEEDQ;
2001 		}
2002 	} else {
2003 		do {
2004 			auio.uio_resid = 1000000000;
2005 			flags = MSG_DONTWAIT;
2006 			error = (*so->so_receive)(so, &nam, &auio, &mp,
2007 						(struct mbuf **)0, &flags);
2008 			if (mp) {
2009 				if (nam) {
2010 					m = nam;
2011 					m->m_next = mp;
2012 				} else
2013 					m = mp;
2014 				if (slp->ns_recend)
2015 					slp->ns_recend->m_nextpkt = m;
2016 				else
2017 					slp->ns_rec = m;
2018 				slp->ns_recend = m;
2019 				m->m_nextpkt = (struct mbuf *)0;
2020 			}
2021 			if (error) {
2022 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2023 					&& error != EWOULDBLOCK) {
2024 					slp->ns_flag |= SLP_DISCONN;
2025 					goto dorecs;
2026 				}
2027 			}
2028 		} while (mp);
2029 	}
2030 
2031 	/*
2032 	 * Now try and process the request records, non-blocking.
2033 	 */
2034 dorecs:
2035 	if (waitflag == M_DONTWAIT &&
2036 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
2037 		nfsrv_wakenfsd(slp);
2038 }
2039 
2040 /*
2041  * Try and extract an RPC request from the mbuf data list received on a
2042  * stream socket. The "waitflag" argument indicates whether or not it
2043  * can sleep.
2044  */
2045 int
2046 nfsrv_getstream(slp, waitflag)
2047 	struct nfssvc_sock *slp;
2048 	int waitflag;
2049 {
2050 	struct mbuf *m, **mpp;
2051 	char *cp1, *cp2;
2052 	int len;
2053 	struct mbuf *om, *m2, *recm;
2054 	u_int32_t recmark;
2055 
2056 	if (slp->ns_flag & SLP_GETSTREAM)
2057 		panic("nfs getstream");
2058 	slp->ns_flag |= SLP_GETSTREAM;
2059 	for (;;) {
2060 	    if (slp->ns_reclen == 0) {
2061 		if (slp->ns_cc < NFSX_UNSIGNED) {
2062 			slp->ns_flag &= ~SLP_GETSTREAM;
2063 			return (0);
2064 		}
2065 		m = slp->ns_raw;
2066 		if (m->m_len >= NFSX_UNSIGNED) {
2067 			memcpy((caddr_t)&recmark, mtod(m, caddr_t), NFSX_UNSIGNED);
2068 			m->m_data += NFSX_UNSIGNED;
2069 			m->m_len -= NFSX_UNSIGNED;
2070 		} else {
2071 			cp1 = (caddr_t)&recmark;
2072 			cp2 = mtod(m, caddr_t);
2073 			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
2074 				while (m->m_len == 0) {
2075 					m = m->m_next;
2076 					cp2 = mtod(m, caddr_t);
2077 				}
2078 				*cp1++ = *cp2++;
2079 				m->m_data++;
2080 				m->m_len--;
2081 			}
2082 		}
2083 		slp->ns_cc -= NFSX_UNSIGNED;
2084 		recmark = ntohl(recmark);
2085 		slp->ns_reclen = recmark & ~0x80000000;
2086 		if (recmark & 0x80000000)
2087 			slp->ns_flag |= SLP_LASTFRAG;
2088 		else
2089 			slp->ns_flag &= ~SLP_LASTFRAG;
2090 		if (slp->ns_reclen > NFS_MAXPACKET) {
2091 			slp->ns_flag &= ~SLP_GETSTREAM;
2092 			return (EPERM);
2093 		}
2094 	    }
2095 
2096 	    /*
2097 	     * Now get the record part.
2098 	     *
2099 	     * Note that slp->ns_reclen may be 0.  Linux sometimes
2100 	     * generates 0-length records.
2101 	     */
2102 	    if (slp->ns_cc == slp->ns_reclen) {
2103 		recm = slp->ns_raw;
2104 		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2105 		slp->ns_cc = slp->ns_reclen = 0;
2106 	    } else if (slp->ns_cc > slp->ns_reclen) {
2107 		len = 0;
2108 		m = slp->ns_raw;
2109 		recm = om = NULL;
2110 
2111 		while (len < slp->ns_reclen) {
2112 			if ((len + m->m_len) > slp->ns_reclen) {
2113 				size_t left = slp->ns_reclen - len;
2114 
2115 				MGETHDR(m2, waitflag, m->m_type);
2116 				if (m2 == NULL) {
2117 					slp->ns_flag &= ~SLP_GETSTREAM;
2118 					return (EWOULDBLOCK);
2119 				}
2120 				if (left > MHLEN) {
2121 					MCLGET(m2, waitflag);
2122 					if (!(m2->m_flags & M_EXT)) {
2123 						m_freem(m2);
2124 						slp->ns_flag &= ~SLP_GETSTREAM;
2125 						return (EWOULDBLOCK);
2126 					}
2127 				}
2128 				memcpy(mtod(m2, caddr_t), mtod(m, caddr_t),
2129 				    left);
2130 				m2->m_len = left;
2131 				m->m_data += left;
2132 				m->m_len -= left;
2133 				if (om) {
2134 					om->m_next = m2;
2135 					recm = slp->ns_raw;
2136 				} else
2137 					recm = m2;
2138 				len = slp->ns_reclen;
2139 			} else if ((len + m->m_len) == slp->ns_reclen) {
2140 				om = m;
2141 				len += m->m_len;
2142 				m = m->m_next;
2143 				recm = slp->ns_raw;
2144 				om->m_next = (struct mbuf *)0;
2145 			} else {
2146 				om = m;
2147 				len += m->m_len;
2148 				m = m->m_next;
2149 			}
2150 		}
2151 		slp->ns_raw = m;
2152 		slp->ns_cc -= len;
2153 		slp->ns_reclen = 0;
2154 	    } else {
2155 		slp->ns_flag &= ~SLP_GETSTREAM;
2156 		return (0);
2157 	    }
2158 
2159 	    /*
2160 	     * Accumulate the fragments into a record.
2161 	     */
2162 	    mpp = &slp->ns_frag;
2163 	    while (*mpp)
2164 		mpp = &((*mpp)->m_next);
2165 	    *mpp = recm;
2166 	    if (slp->ns_flag & SLP_LASTFRAG) {
2167 		if (slp->ns_recend)
2168 		    slp->ns_recend->m_nextpkt = slp->ns_frag;
2169 		else
2170 		    slp->ns_rec = slp->ns_frag;
2171 		slp->ns_recend = slp->ns_frag;
2172 		slp->ns_frag = (struct mbuf *)0;
2173 	    }
2174 	}
2175 }
2176 
2177 /*
2178  * Parse an RPC header.
2179  */
2180 int
2181 nfsrv_dorec(slp, nfsd, ndp)
2182 	struct nfssvc_sock *slp;
2183 	struct nfsd *nfsd;
2184 	struct nfsrv_descript **ndp;
2185 {
2186 	struct mbuf *m, *nam;
2187 	struct nfsrv_descript *nd;
2188 	int error;
2189 
2190 	*ndp = NULL;
2191 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2192 	    (m = slp->ns_rec) == (struct mbuf *)0)
2193 		return (ENOBUFS);
2194 	slp->ns_rec = m->m_nextpkt;
2195 	if (slp->ns_rec)
2196 		m->m_nextpkt = (struct mbuf *)0;
2197 	else
2198 		slp->ns_recend = (struct mbuf *)0;
2199 	if (m->m_type == MT_SONAME) {
2200 		nam = m;
2201 		m = m->m_next;
2202 		nam->m_next = NULL;
2203 	} else
2204 		nam = NULL;
2205 	MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2206 		M_NFSRVDESC, M_WAITOK);
2207 	nd->nd_md = nd->nd_mrep = m;
2208 	nd->nd_nam2 = nam;
2209 	nd->nd_dpos = mtod(m, caddr_t);
2210 	error = nfs_getreq(nd, nfsd, TRUE);
2211 	if (error) {
2212 		m_freem(nam);
2213 		free((caddr_t)nd, M_NFSRVDESC);
2214 		return (error);
2215 	}
2216 	*ndp = nd;
2217 	nfsd->nfsd_nd = nd;
2218 	return (0);
2219 }
2220 
2221 
2222 /*
2223  * Search for a sleeping nfsd and wake it up.
2224  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2225  * running nfsds will go look for the work in the nfssvc_sock list.
2226  */
2227 void
2228 nfsrv_wakenfsd(slp)
2229 	struct nfssvc_sock *slp;
2230 {
2231 	struct nfsd *nd;
2232 
2233 	if ((slp->ns_flag & SLP_VALID) == 0)
2234 		return;
2235 	TAILQ_FOREACH(nd, &nfsd_head, nfsd_chain) {
2236 		if (nd->nfsd_flag & NFSD_WAITING) {
2237 			nd->nfsd_flag &= ~NFSD_WAITING;
2238 			if (nd->nfsd_slp)
2239 				panic("nfsd wakeup");
2240 			slp->ns_sref++;
2241 			nd->nfsd_slp = slp;
2242 			wakeup((caddr_t)nd);
2243 			return;
2244 		}
2245 	}
2246 	slp->ns_flag |= SLP_DOREC;
2247 	nfsd_head_flag |= NFSD_CHECKSLP;
2248 }
2249 #endif /* NFSSERVER */
2250