xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 1ffa7b76c40339c17a0fb2a09fac93f287cfc046)
1 /*	$NetBSD: nfs_socket.c,v 1.83 2003/04/24 21:21:05 drochner 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 1.83 2003/04/24 21:21:05 drochner 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 #ifdef MBUFTRACE
87 struct mowner nfs_mowner = { "nfs" };
88 #endif
89 
90 #define	TRUE	1
91 #define	FALSE	0
92 
93 /*
94  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
95  * Use the mean and mean deviation of rtt for the appropriate type of rpc
96  * for the frequent rpcs and a default for the others.
97  * The justification for doing "other" this way is that these rpcs
98  * happen so infrequently that timer est. would probably be stale.
99  * Also, since many of these rpcs are
100  * non-idempotent, a conservative timeout is desired.
101  * getattr, lookup - A+2D
102  * read, write     - A+4D
103  * other           - nm_timeo
104  */
105 #define	NFS_RTO(n, t) \
106 	((t) == 0 ? (n)->nm_timeo : \
107 	 ((t) < 3 ? \
108 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
109 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
110 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
111 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
112 /*
113  * External data, mostly RPC constants in XDR form
114  */
115 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
116 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
117 	rpc_auth_kerb;
118 extern u_int32_t nfs_prog, nqnfs_prog;
119 extern time_t nqnfsstarttime;
120 extern const int nfsv3_procid[NFS_NPROCS];
121 extern int nfs_ticks;
122 
123 /*
124  * Defines which timer to use for the procnum.
125  * 0 - default
126  * 1 - getattr
127  * 2 - lookup
128  * 3 - read
129  * 4 - write
130  */
131 static const int proct[NFS_NPROCS] = {
132 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
133 	0, 0, 0,
134 };
135 
136 /*
137  * There is a congestion window for outstanding rpcs maintained per mount
138  * point. The cwnd size is adjusted in roughly the way that:
139  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
140  * SIGCOMM '88". ACM, August 1988.
141  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
142  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
143  * of rpcs is in progress.
144  * (The sent count and cwnd are scaled for integer arith.)
145  * Variants of "slow start" were tried and were found to be too much of a
146  * performance hit (ave. rtt 3 times larger),
147  * I suspect due to the large rtt that nfs rpcs have.
148  */
149 #define	NFS_CWNDSCALE	256
150 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
151 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
152 int nfsrtton = 0;
153 struct nfsrtt nfsrtt;
154 struct nfsreqhead nfs_reqq;
155 
156 struct callout nfs_timer_ch = CALLOUT_INITIALIZER;
157 
158 /*
159  * Initialize sockets and congestion for a new NFS connection.
160  * We do not free the sockaddr if error.
161  */
162 int
163 nfs_connect(nmp, rep)
164 	struct nfsmount *nmp;
165 	struct nfsreq *rep;
166 {
167 	struct socket *so;
168 	int s, error, rcvreserve, sndreserve;
169 	struct sockaddr *saddr;
170 	struct sockaddr_in *sin;
171 #ifdef INET6
172 	struct sockaddr_in6 *sin6;
173 #endif
174 	struct mbuf *m;
175 
176 	nmp->nm_so = (struct socket *)0;
177 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
178 	error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
179 		nmp->nm_soproto);
180 	if (error)
181 		goto bad;
182 	so = nmp->nm_so;
183 #ifdef MBUFTRACE
184 	so->so_mowner = &nfs_mowner;
185 	so->so_rcv.sb_mowner = &nfs_mowner;
186 	so->so_snd.sb_mowner = &nfs_mowner;
187 #endif
188 	nmp->nm_soflags = so->so_proto->pr_flags;
189 
190 	/*
191 	 * Some servers require that the client port be a reserved port number.
192 	 */
193 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
194 		m = m_get(M_WAIT, MT_SOOPTS);
195 		MCLAIM(m, so->so_mowner);
196 		*mtod(m, int32_t *) = IP_PORTRANGE_LOW;
197 		m->m_len = sizeof(int32_t);
198 		if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m)))
199 			goto bad;
200 		m = m_get(M_WAIT, MT_SONAME);
201 		MCLAIM(m, so->so_mowner);
202 		sin = mtod(m, struct sockaddr_in *);
203 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
204 		sin->sin_family = AF_INET;
205 		sin->sin_addr.s_addr = INADDR_ANY;
206 		sin->sin_port = 0;
207 		error = sobind(so, m, &proc0);
208 		m_freem(m);
209 		if (error)
210 			goto bad;
211 	}
212 #ifdef INET6
213 	if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) {
214 		m = m_get(M_WAIT, MT_SOOPTS);
215 		MCLAIM(m, so->so_mowner);
216 		*mtod(m, int32_t *) = IPV6_PORTRANGE_LOW;
217 		m->m_len = sizeof(int32_t);
218 		if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m)))
219 			goto bad;
220 		m = m_get(M_WAIT, MT_SONAME);
221 		MCLAIM(m, so->so_mowner);
222 		sin6 = mtod(m, struct sockaddr_in6 *);
223 		sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6);
224 		sin6->sin6_family = AF_INET6;
225 		sin6->sin6_addr = in6addr_any;
226 		sin6->sin6_port = 0;
227 		error = sobind(so, m, &proc0);
228 		m_freem(m);
229 		if (error)
230 			goto bad;
231 	}
232 #endif
233 
234 	/*
235 	 * Protocols that do not require connections may be optionally left
236 	 * unconnected for servers that reply from a port other than NFS_PORT.
237 	 */
238 	if (nmp->nm_flag & NFSMNT_NOCONN) {
239 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
240 			error = ENOTCONN;
241 			goto bad;
242 		}
243 	} else {
244 		error = soconnect(so, nmp->nm_nam);
245 		if (error)
246 			goto bad;
247 
248 		/*
249 		 * Wait for the connection to complete. Cribbed from the
250 		 * connect system call but with the wait timing out so
251 		 * that interruptible mounts don't hang here for a long time.
252 		 */
253 		s = splsoftnet();
254 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
255 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
256 				"nfscn1", 2 * hz);
257 			if ((so->so_state & SS_ISCONNECTING) &&
258 			    so->so_error == 0 && rep &&
259 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
260 				so->so_state &= ~SS_ISCONNECTING;
261 				splx(s);
262 				goto bad;
263 			}
264 		}
265 		if (so->so_error) {
266 			error = so->so_error;
267 			so->so_error = 0;
268 			splx(s);
269 			goto bad;
270 		}
271 		splx(s);
272 	}
273 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
274 		so->so_rcv.sb_timeo = (5 * hz);
275 		so->so_snd.sb_timeo = (5 * hz);
276 	} else {
277 		so->so_rcv.sb_timeo = 0;
278 		so->so_snd.sb_timeo = 0;
279 	}
280 	if (nmp->nm_sotype == SOCK_DGRAM) {
281 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
282 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
283 		    NFS_MAXPKTHDR) * 2;
284 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
285 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
286 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
287 		    NFS_MAXPKTHDR) * 2;
288 	} else {
289 		if (nmp->nm_sotype != SOCK_STREAM)
290 			panic("nfscon sotype");
291 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
292 			m = m_get(M_WAIT, MT_SOOPTS);
293 			MCLAIM(m, so->so_mowner);
294 			*mtod(m, int32_t *) = 1;
295 			m->m_len = sizeof(int32_t);
296 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
297 		}
298 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
299 			m = m_get(M_WAIT, MT_SOOPTS);
300 			MCLAIM(m, so->so_mowner);
301 			*mtod(m, int32_t *) = 1;
302 			m->m_len = sizeof(int32_t);
303 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
304 		}
305 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
306 		    sizeof (u_int32_t)) * 2;
307 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
308 		    sizeof (u_int32_t)) * 2;
309 	}
310 	error = soreserve(so, sndreserve, rcvreserve);
311 	if (error)
312 		goto bad;
313 	so->so_rcv.sb_flags |= SB_NOINTR;
314 	so->so_snd.sb_flags |= SB_NOINTR;
315 
316 	/* Initialize other non-zero congestion variables */
317 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
318 		NFS_TIMEO << 3;
319 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
320 		nmp->nm_sdrtt[3] = 0;
321 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
322 	nmp->nm_sent = 0;
323 	nmp->nm_timeouts = 0;
324 	return (0);
325 
326 bad:
327 	nfs_disconnect(nmp);
328 	return (error);
329 }
330 
331 /*
332  * Reconnect routine:
333  * Called when a connection is broken on a reliable protocol.
334  * - clean up the old socket
335  * - nfs_connect() again
336  * - set R_MUSTRESEND for all outstanding requests on mount point
337  * If this fails the mount point is DEAD!
338  * nb: Must be called with the nfs_sndlock() set on the mount point.
339  */
340 int
341 nfs_reconnect(rep)
342 	struct nfsreq *rep;
343 {
344 	struct nfsreq *rp;
345 	struct nfsmount *nmp = rep->r_nmp;
346 	int error;
347 
348 	nfs_disconnect(nmp);
349 	while ((error = nfs_connect(nmp, rep)) != 0) {
350 		if (error == EINTR || error == ERESTART)
351 			return (EINTR);
352 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0);
353 	}
354 
355 	/*
356 	 * Loop through outstanding request list and fix up all requests
357 	 * on old socket.
358 	 */
359 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
360 		if (rp->r_nmp == nmp)
361 			rp->r_flags |= R_MUSTRESEND;
362 	}
363 	return (0);
364 }
365 
366 /*
367  * NFS disconnect. Clean up and unlink.
368  */
369 void
370 nfs_disconnect(nmp)
371 	struct nfsmount *nmp;
372 {
373 	struct socket *so;
374 	int drain = 0;
375 
376 	if (nmp->nm_so) {
377 		so = nmp->nm_so;
378 		nmp->nm_so = (struct socket *)0;
379 		soshutdown(so, 2);
380 		drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
381 		if (drain) {
382 			/*
383 			 * soshutdown() above should wake up the current
384 			 * listener.
385 			 * Now wake up those waiting for the receive lock, and
386 			 * wait for them to go away unhappy, to prevent *nmp
387 			 * from evaporating while they're sleeping.
388 			 */
389 			while (nmp->nm_waiters > 0) {
390 				wakeup (&nmp->nm_iflag);
391 				(void) tsleep(&nmp->nm_waiters, PVFS,
392 				    "nfsdis", 0);
393 			}
394 		}
395 		soclose(so);
396 	}
397 #ifdef DIAGNOSTIC
398 	if (drain && (nmp->nm_waiters > 0))
399 		panic("nfs_disconnect: waiters left after drain?");
400 #endif
401 }
402 
403 void
404 nfs_safedisconnect(nmp)
405 	struct nfsmount *nmp;
406 {
407 	struct nfsreq dummyreq;
408 
409 	memset(&dummyreq, 0, sizeof(dummyreq));
410 	dummyreq.r_nmp = nmp;
411 	nfs_rcvlock(&dummyreq); /* XXX ignored error return */
412 	nfs_disconnect(nmp);
413 	nfs_rcvunlock(&nmp->nm_iflag);
414 }
415 
416 /*
417  * This is the nfs send routine. For connection based socket types, it
418  * must be called with an nfs_sndlock() on the socket.
419  * "rep == NULL" indicates that it has been called from a server.
420  * For the client side:
421  * - return EINTR if the RPC is terminated, 0 otherwise
422  * - set R_MUSTRESEND if the send fails for any reason
423  * - do any cleanup required by recoverable socket errors (? ? ?)
424  * For the server side:
425  * - return EINTR or ERESTART if interrupted by a signal
426  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
427  * - do any cleanup required by recoverable socket errors (? ? ?)
428  */
429 int
430 nfs_send(so, nam, top, rep)
431 	struct socket *so;
432 	struct mbuf *nam;
433 	struct mbuf *top;
434 	struct nfsreq *rep;
435 {
436 	struct mbuf *sendnam;
437 	int error, soflags, flags;
438 
439 	if (rep) {
440 		if (rep->r_flags & R_SOFTTERM) {
441 			m_freem(top);
442 			return (EINTR);
443 		}
444 		if ((so = rep->r_nmp->nm_so) == NULL) {
445 			rep->r_flags |= R_MUSTRESEND;
446 			m_freem(top);
447 			return (0);
448 		}
449 		rep->r_flags &= ~R_MUSTRESEND;
450 		soflags = rep->r_nmp->nm_soflags;
451 	} else
452 		soflags = so->so_proto->pr_flags;
453 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
454 		sendnam = (struct mbuf *)0;
455 	else
456 		sendnam = nam;
457 	if (so->so_type == SOCK_SEQPACKET)
458 		flags = MSG_EOR;
459 	else
460 		flags = 0;
461 
462 	error = (*so->so_send)(so, sendnam, (struct uio *)0, top,
463 		(struct mbuf *)0, flags);
464 	if (error) {
465 		if (rep) {
466 			if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
467 				/*
468 				 * We're too fast for the network/driver,
469 				 * and UDP isn't flowcontrolled.
470 				 * We need to resend. This is not fatal,
471 				 * just try again.
472 				 *
473 				 * Could be smarter here by doing some sort
474 				 * of a backoff, but this is rare.
475 				 */
476 				rep->r_flags |= R_MUSTRESEND;
477 			} else {
478 				log(LOG_INFO, "nfs send error %d for %s\n",
479 				    error,
480 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
481 				/*
482 				 * Deal with errors for the client side.
483 				 */
484 				if (rep->r_flags & R_SOFTTERM)
485 					error = EINTR;
486 				else
487 					rep->r_flags |= R_MUSTRESEND;
488 			}
489 		} else {
490 			/*
491 			 * See above. This error can happen under normal
492 			 * circumstances and the log is too noisy.
493 			 * The error will still show up in nfsstat.
494 			 */
495 			if (error != ENOBUFS || so->so_type != SOCK_DGRAM)
496 				log(LOG_INFO, "nfsd send error %d\n", error);
497 		}
498 
499 		/*
500 		 * Handle any recoverable (soft) socket errors here. (? ? ?)
501 		 */
502 		if (error != EINTR && error != ERESTART &&
503 			error != EWOULDBLOCK && error != EPIPE)
504 			error = 0;
505 	}
506 	return (error);
507 }
508 
509 #ifdef NFS
510 /*
511  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
512  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
513  * Mark and consolidate the data into a new mbuf list.
514  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
515  *     small mbufs.
516  * For SOCK_STREAM we must be very careful to read an entire record once
517  * we have read any of it, even if the system call has been interrupted.
518  */
519 int
520 nfs_receive(rep, aname, mp)
521 	struct nfsreq *rep;
522 	struct mbuf **aname;
523 	struct mbuf **mp;
524 {
525 	struct socket *so;
526 	struct uio auio;
527 	struct iovec aio;
528 	struct mbuf *m;
529 	struct mbuf *control;
530 	u_int32_t len;
531 	struct mbuf **getnam;
532 	int error, sotype, rcvflg;
533 	struct proc *p = curproc;	/* XXX */
534 
535 	/*
536 	 * Set up arguments for soreceive()
537 	 */
538 	*mp = (struct mbuf *)0;
539 	*aname = (struct mbuf *)0;
540 	sotype = rep->r_nmp->nm_sotype;
541 
542 	/*
543 	 * For reliable protocols, lock against other senders/receivers
544 	 * in case a reconnect is necessary.
545 	 * For SOCK_STREAM, first get the Record Mark to find out how much
546 	 * more there is to get.
547 	 * We must lock the socket against other receivers
548 	 * until we have an entire rpc request/reply.
549 	 */
550 	if (sotype != SOCK_DGRAM) {
551 		error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
552 		if (error)
553 			return (error);
554 tryagain:
555 		/*
556 		 * Check for fatal errors and resending request.
557 		 */
558 		/*
559 		 * Ugh: If a reconnect attempt just happened, nm_so
560 		 * would have changed. NULL indicates a failed
561 		 * attempt that has essentially shut down this
562 		 * mount point.
563 		 */
564 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
565 			nfs_sndunlock(&rep->r_nmp->nm_iflag);
566 			return (EINTR);
567 		}
568 		so = rep->r_nmp->nm_so;
569 		if (!so) {
570 			error = nfs_reconnect(rep);
571 			if (error) {
572 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
573 				return (error);
574 			}
575 			goto tryagain;
576 		}
577 		while (rep->r_flags & R_MUSTRESEND) {
578 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
579 			nfsstats.rpcretries++;
580 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
581 			if (error) {
582 				if (error == EINTR || error == ERESTART ||
583 				    (error = nfs_reconnect(rep)) != 0) {
584 					nfs_sndunlock(&rep->r_nmp->nm_iflag);
585 					return (error);
586 				}
587 				goto tryagain;
588 			}
589 		}
590 		nfs_sndunlock(&rep->r_nmp->nm_iflag);
591 		if (sotype == SOCK_STREAM) {
592 			aio.iov_base = (caddr_t) &len;
593 			aio.iov_len = sizeof(u_int32_t);
594 			auio.uio_iov = &aio;
595 			auio.uio_iovcnt = 1;
596 			auio.uio_segflg = UIO_SYSSPACE;
597 			auio.uio_rw = UIO_READ;
598 			auio.uio_offset = 0;
599 			auio.uio_resid = sizeof(u_int32_t);
600 			auio.uio_procp = p;
601 			do {
602 			   rcvflg = MSG_WAITALL;
603 			   error = (*so->so_receive)(so, (struct mbuf **)0, &auio,
604 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
605 			   if (error == EWOULDBLOCK && rep) {
606 				if (rep->r_flags & R_SOFTTERM)
607 					return (EINTR);
608 			   }
609 			} while (error == EWOULDBLOCK);
610 			if (!error && auio.uio_resid > 0) {
611 			    /*
612 			     * Don't log a 0 byte receive; it means
613 			     * that the socket has been closed, and
614 			     * can happen during normal operation
615 			     * (forcible unmount or Solaris server).
616 			     */
617 			    if (auio.uio_resid != sizeof (u_int32_t))
618 			      log(LOG_INFO,
619 				 "short receive (%lu/%lu) from nfs server %s\n",
620 				 (u_long)sizeof(u_int32_t) - auio.uio_resid,
621 				 (u_long)sizeof(u_int32_t),
622 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
623 			    error = EPIPE;
624 			}
625 			if (error)
626 				goto errout;
627 			len = ntohl(len) & ~0x80000000;
628 			/*
629 			 * This is SERIOUS! We are out of sync with the sender
630 			 * and forcing a disconnect/reconnect is all I can do.
631 			 */
632 			if (len > NFS_MAXPACKET) {
633 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
634 				"impossible packet length",
635 				len,
636 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
637 			    error = EFBIG;
638 			    goto errout;
639 			}
640 			auio.uio_resid = len;
641 			do {
642 			    rcvflg = MSG_WAITALL;
643 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
644 				&auio, mp, (struct mbuf **)0, &rcvflg);
645 			} while (error == EWOULDBLOCK || error == EINTR ||
646 				 error == ERESTART);
647 			if (!error && auio.uio_resid > 0) {
648 			    if (len != auio.uio_resid)
649 			      log(LOG_INFO,
650 				"short receive (%lu/%d) from nfs server %s\n",
651 				(u_long)len - auio.uio_resid, len,
652 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
653 			    error = EPIPE;
654 			}
655 		} else {
656 			/*
657 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
658 			 * and soreceive() will return when it has either a
659 			 * control msg or a data msg.
660 			 * We have no use for control msg., but must grab them
661 			 * and then throw them away so we know what is going
662 			 * on.
663 			 */
664 			auio.uio_resid = len = 100000000; /* Anything Big */
665 			auio.uio_procp = p;
666 			do {
667 			    rcvflg = 0;
668 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
669 				&auio, mp, &control, &rcvflg);
670 			    if (control)
671 				m_freem(control);
672 			    if (error == EWOULDBLOCK && rep) {
673 				if (rep->r_flags & R_SOFTTERM)
674 					return (EINTR);
675 			    }
676 			} while (error == EWOULDBLOCK ||
677 				 (!error && *mp == NULL && control));
678 			if ((rcvflg & MSG_EOR) == 0)
679 				printf("Egad!!\n");
680 			if (!error && *mp == NULL)
681 				error = EPIPE;
682 			len -= auio.uio_resid;
683 		}
684 errout:
685 		if (error && error != EINTR && error != ERESTART) {
686 			m_freem(*mp);
687 			*mp = (struct mbuf *)0;
688 			if (error != EPIPE)
689 				log(LOG_INFO,
690 				    "receive error %d from nfs server %s\n",
691 				    error,
692 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
693 			error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
694 			if (!error)
695 				error = nfs_reconnect(rep);
696 			if (!error)
697 				goto tryagain;
698 			else
699 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
700 		}
701 	} else {
702 		if ((so = rep->r_nmp->nm_so) == NULL)
703 			return (EACCES);
704 		if (so->so_state & SS_ISCONNECTED)
705 			getnam = (struct mbuf **)0;
706 		else
707 			getnam = aname;
708 		auio.uio_resid = len = 1000000;
709 		auio.uio_procp = p;
710 		do {
711 			rcvflg = 0;
712 			error =  (*so->so_receive)(so, getnam, &auio, mp,
713 				(struct mbuf **)0, &rcvflg);
714 			if (error == EWOULDBLOCK &&
715 			    (rep->r_flags & R_SOFTTERM))
716 				return (EINTR);
717 		} while (error == EWOULDBLOCK);
718 		len -= auio.uio_resid;
719 		if (!error && *mp == NULL)
720 			error = EPIPE;
721 	}
722 	if (error) {
723 		m_freem(*mp);
724 		*mp = (struct mbuf *)0;
725 	}
726 	return (error);
727 }
728 
729 /*
730  * Implement receipt of reply on a socket.
731  * We must search through the list of received datagrams matching them
732  * with outstanding requests using the xid, until ours is found.
733  */
734 /* ARGSUSED */
735 int
736 nfs_reply(myrep)
737 	struct nfsreq *myrep;
738 {
739 	struct nfsreq *rep;
740 	struct nfsmount *nmp = myrep->r_nmp;
741 	int32_t t1;
742 	struct mbuf *mrep, *nam, *md;
743 	u_int32_t rxid, *tl;
744 	caddr_t dpos, cp2;
745 	int error;
746 
747 	/*
748 	 * Loop around until we get our own reply
749 	 */
750 	for (;;) {
751 		/*
752 		 * Lock against other receivers so that I don't get stuck in
753 		 * sbwait() after someone else has received my reply for me.
754 		 * Also necessary for connection based protocols to avoid
755 		 * race conditions during a reconnect.
756 		 */
757 		error = nfs_rcvlock(myrep);
758 		if (error == EALREADY)
759 			return (0);
760 		if (error)
761 			return (error);
762 		/*
763 		 * Get the next Rpc reply off the socket
764 		 */
765 		nmp->nm_waiters++;
766 		error = nfs_receive(myrep, &nam, &mrep);
767 		nfs_rcvunlock(&nmp->nm_iflag);
768 		if (error) {
769 
770 			if (nmp->nm_iflag & NFSMNT_DISMNT) {
771 				/*
772 				 * Oops, we're going away now..
773 				 */
774 				nmp->nm_waiters--;
775 				wakeup (&nmp->nm_waiters);
776 				return error;
777 			}
778 			nmp->nm_waiters--;
779 			/*
780 			 * Ignore routing errors on connectionless protocols? ?
781 			 */
782 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
783 				nmp->nm_so->so_error = 0;
784 #ifdef DEBUG
785 				printf("nfs_reply: ignoring error %d\n", error);
786 #endif
787 				if (myrep->r_flags & R_GETONEREP)
788 					return (0);
789 				continue;
790 			}
791 			return (error);
792 		}
793 		nmp->nm_waiters--;
794 		if (nam)
795 			m_freem(nam);
796 
797 		/*
798 		 * Get the xid and check that it is an rpc reply
799 		 */
800 		md = mrep;
801 		dpos = mtod(md, caddr_t);
802 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
803 		rxid = *tl++;
804 		if (*tl != rpc_reply) {
805 #ifndef NFS_V2_ONLY
806 			if (nmp->nm_flag & NFSMNT_NQNFS) {
807 				if (nqnfs_callback(nmp, mrep, md, dpos))
808 					nfsstats.rpcinvalid++;
809 			} else
810 #endif
811 			{
812 				nfsstats.rpcinvalid++;
813 				m_freem(mrep);
814 			}
815 nfsmout:
816 			if (myrep->r_flags & R_GETONEREP)
817 				return (0);
818 			continue;
819 		}
820 
821 		/*
822 		 * Loop through the request list to match up the reply
823 		 * Iff no match, just drop the datagram
824 		 */
825 		TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
826 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
827 				/* Found it.. */
828 				rep->r_mrep = mrep;
829 				rep->r_md = md;
830 				rep->r_dpos = dpos;
831 				if (nfsrtton) {
832 					struct rttl *rt;
833 
834 					rt = &nfsrtt.rttl[nfsrtt.pos];
835 					rt->proc = rep->r_procnum;
836 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
837 					rt->sent = nmp->nm_sent;
838 					rt->cwnd = nmp->nm_cwnd;
839 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
840 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
841 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
842 					rt->tstamp = time;
843 					if (rep->r_flags & R_TIMING)
844 						rt->rtt = rep->r_rtt;
845 					else
846 						rt->rtt = 1000000;
847 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
848 				}
849 				/*
850 				 * Update congestion window.
851 				 * Do the additive increase of
852 				 * one rpc/rtt.
853 				 */
854 				if (nmp->nm_cwnd <= nmp->nm_sent) {
855 					nmp->nm_cwnd +=
856 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
857 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
858 					if (nmp->nm_cwnd > NFS_MAXCWND)
859 						nmp->nm_cwnd = NFS_MAXCWND;
860 				}
861 				rep->r_flags &= ~R_SENT;
862 				nmp->nm_sent -= NFS_CWNDSCALE;
863 				/*
864 				 * Update rtt using a gain of 0.125 on the mean
865 				 * and a gain of 0.25 on the deviation.
866 				 */
867 				if (rep->r_flags & R_TIMING) {
868 					/*
869 					 * Since the timer resolution of
870 					 * NFS_HZ is so course, it can often
871 					 * result in r_rtt == 0. Since
872 					 * r_rtt == N means that the actual
873 					 * rtt is between N+dt and N+2-dt ticks,
874 					 * add 1.
875 					 */
876 					t1 = rep->r_rtt + 1;
877 					t1 -= (NFS_SRTT(rep) >> 3);
878 					NFS_SRTT(rep) += t1;
879 					if (t1 < 0)
880 						t1 = -t1;
881 					t1 -= (NFS_SDRTT(rep) >> 2);
882 					NFS_SDRTT(rep) += t1;
883 				}
884 				nmp->nm_timeouts = 0;
885 				break;
886 			}
887 		}
888 		/*
889 		 * If not matched to a request, drop it.
890 		 * If it's mine, get out.
891 		 */
892 		if (rep == 0) {
893 			nfsstats.rpcunexpected++;
894 			m_freem(mrep);
895 		} else if (rep == myrep) {
896 			if (rep->r_mrep == NULL)
897 				panic("nfsreply nil");
898 			return (0);
899 		}
900 		if (myrep->r_flags & R_GETONEREP)
901 			return (0);
902 	}
903 }
904 
905 /*
906  * nfs_request - goes something like this
907  *	- fill in request struct
908  *	- links it into list
909  *	- calls nfs_send() for first transmit
910  *	- calls nfs_receive() to get reply
911  *	- break down rpc header and return with nfs reply pointed to
912  *	  by mrep or error
913  * nb: always frees up mreq mbuf list
914  */
915 int
916 nfs_request(np, mrest, procnum, procp, cred, mrp, mdp, dposp)
917 	struct nfsnode *np;
918 	struct mbuf *mrest;
919 	int procnum;
920 	struct proc *procp;
921 	struct ucred *cred;
922 	struct mbuf **mrp;
923 	struct mbuf **mdp;
924 	caddr_t *dposp;
925 {
926 	struct mbuf *m, *mrep;
927 	struct nfsreq *rep;
928 	u_int32_t *tl;
929 	int i;
930 	struct nfsmount *nmp;
931 	struct mbuf *md, *mheadend;
932 	char nickv[RPCX_NICKVERF];
933 	time_t reqtime, waituntil;
934 	caddr_t dpos, cp2;
935 	int t1, s, error = 0, mrest_len, auth_len, auth_type;
936 	int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0;
937 	int verf_len, verf_type;
938 	u_int32_t xid;
939 	char *auth_str, *verf_str;
940 	NFSKERBKEY_T key;		/* save session key */
941 #ifndef NFS_V2_ONLY
942 	int nqlflag, cachable;
943 	u_quad_t frev;
944 #endif
945 
946 	KASSERT(cred != NULL);
947 	nmp = VFSTONFS(np->n_vnode->v_mount);
948 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
949 	rep->r_nmp = nmp;
950 	rep->r_procp = procp;
951 	rep->r_procnum = procnum;
952 	i = 0;
953 	m = mrest;
954 	while (m) {
955 		i += m->m_len;
956 		m = m->m_next;
957 	}
958 	mrest_len = i;
959 
960 	/*
961 	 * Get the RPC header with authorization.
962 	 */
963 kerbauth:
964 	verf_str = auth_str = (char *)0;
965 	if (nmp->nm_flag & NFSMNT_KERB) {
966 		verf_str = nickv;
967 		verf_len = sizeof (nickv);
968 		auth_type = RPCAUTH_KERB4;
969 		memset((caddr_t)key, 0, sizeof (key));
970 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
971 			&auth_len, verf_str, verf_len)) {
972 			error = nfs_getauth(nmp, rep, cred, &auth_str,
973 				&auth_len, verf_str, &verf_len, key);
974 			if (error) {
975 				free((caddr_t)rep, M_NFSREQ);
976 				m_freem(mrest);
977 				return (error);
978 			}
979 		}
980 	} else {
981 		auth_type = RPCAUTH_UNIX;
982 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
983 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
984 			5 * NFSX_UNSIGNED;
985 	}
986 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
987 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
988 	if (auth_str)
989 		free(auth_str, M_TEMP);
990 
991 	/*
992 	 * For stream protocols, insert a Sun RPC Record Mark.
993 	 */
994 	if (nmp->nm_sotype == SOCK_STREAM) {
995 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
996 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
997 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
998 	}
999 	rep->r_mreq = m;
1000 	rep->r_xid = xid;
1001 tryagain:
1002 	if (nmp->nm_flag & NFSMNT_SOFT)
1003 		rep->r_retry = nmp->nm_retry;
1004 	else
1005 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
1006 	rep->r_rtt = rep->r_rexmit = 0;
1007 	if (proct[procnum] > 0)
1008 		rep->r_flags = R_TIMING;
1009 	else
1010 		rep->r_flags = 0;
1011 	rep->r_mrep = NULL;
1012 
1013 	/*
1014 	 * Do the client side RPC.
1015 	 */
1016 	nfsstats.rpcrequests++;
1017 	/*
1018 	 * Chain request into list of outstanding requests. Be sure
1019 	 * to put it LAST so timer finds oldest requests first.
1020 	 */
1021 	s = splsoftnet();
1022 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1023 
1024 	/* Get send time for nqnfs */
1025 	reqtime = time.tv_sec;
1026 
1027 	/*
1028 	 * If backing off another request or avoiding congestion, don't
1029 	 * send this one now but let timer do it. If not timing a request,
1030 	 * do it now.
1031 	 */
1032 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1033 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1034 		nmp->nm_sent < nmp->nm_cwnd)) {
1035 		splx(s);
1036 		if (nmp->nm_soflags & PR_CONNREQUIRED)
1037 			error = nfs_sndlock(&nmp->nm_iflag, rep);
1038 		if (!error) {
1039 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
1040 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep);
1041 			if (nmp->nm_soflags & PR_CONNREQUIRED)
1042 				nfs_sndunlock(&nmp->nm_iflag);
1043 		}
1044 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1045 			nmp->nm_sent += NFS_CWNDSCALE;
1046 			rep->r_flags |= R_SENT;
1047 		}
1048 	} else {
1049 		splx(s);
1050 		rep->r_rtt = -1;
1051 	}
1052 
1053 	/*
1054 	 * Wait for the reply from our send or the timer's.
1055 	 */
1056 	if (!error || error == EPIPE)
1057 		error = nfs_reply(rep);
1058 
1059 	/*
1060 	 * RPC done, unlink the request.
1061 	 */
1062 	s = splsoftnet();
1063 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1064 	splx(s);
1065 
1066 	/*
1067 	 * Decrement the outstanding request count.
1068 	 */
1069 	if (rep->r_flags & R_SENT) {
1070 		rep->r_flags &= ~R_SENT;	/* paranoia */
1071 		nmp->nm_sent -= NFS_CWNDSCALE;
1072 	}
1073 
1074 	/*
1075 	 * If there was a successful reply and a tprintf msg.
1076 	 * tprintf a response.
1077 	 */
1078 	if (!error && (rep->r_flags & R_TPRINTFMSG))
1079 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1080 		    "is alive again");
1081 	mrep = rep->r_mrep;
1082 	md = rep->r_md;
1083 	dpos = rep->r_dpos;
1084 	if (error) {
1085 		m_freem(rep->r_mreq);
1086 		free((caddr_t)rep, M_NFSREQ);
1087 		return (error);
1088 	}
1089 
1090 	/*
1091 	 * break down the rpc header and check if ok
1092 	 */
1093 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1094 	if (*tl++ == rpc_msgdenied) {
1095 		if (*tl == rpc_mismatch)
1096 			error = EOPNOTSUPP;
1097 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1098 			if (!failed_auth) {
1099 				failed_auth++;
1100 				mheadend->m_next = (struct mbuf *)0;
1101 				m_freem(mrep);
1102 				m_freem(rep->r_mreq);
1103 				goto kerbauth;
1104 			} else
1105 				error = EAUTH;
1106 		} else
1107 			error = EACCES;
1108 		m_freem(mrep);
1109 		m_freem(rep->r_mreq);
1110 		free((caddr_t)rep, M_NFSREQ);
1111 		return (error);
1112 	}
1113 
1114 	/*
1115 	 * Grab any Kerberos verifier, otherwise just throw it away.
1116 	 */
1117 	verf_type = fxdr_unsigned(int, *tl++);
1118 	i = fxdr_unsigned(int32_t, *tl);
1119 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1120 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1121 		if (error)
1122 			goto nfsmout;
1123 	} else if (i > 0)
1124 		nfsm_adv(nfsm_rndup(i));
1125 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1126 	/* 0 == ok */
1127 	if (*tl == 0) {
1128 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1129 		if (*tl != 0) {
1130 			error = fxdr_unsigned(int, *tl);
1131 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1132 				error == NFSERR_TRYLATER) {
1133 				m_freem(mrep);
1134 				error = 0;
1135 				waituntil = time.tv_sec + trylater_delay;
1136 				while (time.tv_sec < waituntil)
1137 					(void) tsleep((caddr_t)&lbolt,
1138 						PSOCK, "nqnfstry", 0);
1139 				trylater_delay *= nfs_backoff[trylater_cnt];
1140 				if (trylater_cnt < 7)
1141 					trylater_cnt++;
1142 				goto tryagain;
1143 			}
1144 
1145 			/*
1146 			 * If the File Handle was stale, invalidate the
1147 			 * lookup cache, just in case.
1148 			 */
1149 			if (error == ESTALE)
1150 				cache_purge(NFSTOV(np));
1151 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1152 				*mrp = mrep;
1153 				*mdp = md;
1154 				*dposp = dpos;
1155 				error |= NFSERR_RETERR;
1156 			} else
1157 				m_freem(mrep);
1158 			m_freem(rep->r_mreq);
1159 			free((caddr_t)rep, M_NFSREQ);
1160 			return (error);
1161 		}
1162 
1163 #ifndef NFS_V2_ONLY
1164 		/*
1165 		 * For nqnfs, get any lease in reply
1166 		 */
1167 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1168 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1169 			if (*tl) {
1170 				nqlflag = fxdr_unsigned(int, *tl);
1171 				nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1172 				cachable = fxdr_unsigned(int, *tl++);
1173 				reqtime += fxdr_unsigned(int, *tl++);
1174 				if (reqtime > time.tv_sec) {
1175 				    frev = fxdr_hyper(tl);
1176 				    nqnfs_clientlease(nmp, np, nqlflag,
1177 					cachable, reqtime, frev);
1178 				}
1179 			}
1180 		}
1181 #endif
1182 		*mrp = mrep;
1183 		*mdp = md;
1184 		*dposp = dpos;
1185 		m_freem(rep->r_mreq);
1186 		FREE((caddr_t)rep, M_NFSREQ);
1187 		return (0);
1188 	}
1189 	m_freem(mrep);
1190 	error = EPROTONOSUPPORT;
1191 nfsmout:
1192 	m_freem(rep->r_mreq);
1193 	free((caddr_t)rep, M_NFSREQ);
1194 	return (error);
1195 }
1196 #endif /* NFS */
1197 
1198 /*
1199  * Generate the rpc reply header
1200  * siz arg. is used to decide if adding a cluster is worthwhile
1201  */
1202 int
1203 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1204 	int siz;
1205 	struct nfsrv_descript *nd;
1206 	struct nfssvc_sock *slp;
1207 	int err;
1208 	int cache;
1209 	u_quad_t *frev;
1210 	struct mbuf **mrq;
1211 	struct mbuf **mbp;
1212 	caddr_t *bposp;
1213 {
1214 	u_int32_t *tl;
1215 	struct mbuf *mreq;
1216 	caddr_t bpos;
1217 	struct mbuf *mb;
1218 
1219 	mreq = m_gethdr(M_WAIT, MT_DATA);
1220 	MCLAIM(mreq, &nfs_mowner);
1221 	mb = mreq;
1222 	/*
1223 	 * If this is a big reply, use a cluster else
1224 	 * try and leave leading space for the lower level headers.
1225 	 */
1226 	siz += RPC_REPLYSIZ;
1227 	if (siz >= max_datalen) {
1228 		m_clget(mreq, M_WAIT);
1229 	} else
1230 		mreq->m_data += max_hdr;
1231 	tl = mtod(mreq, u_int32_t *);
1232 	mreq->m_len = 6 * NFSX_UNSIGNED;
1233 	bpos = ((caddr_t)tl) + mreq->m_len;
1234 	*tl++ = txdr_unsigned(nd->nd_retxid);
1235 	*tl++ = rpc_reply;
1236 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1237 		*tl++ = rpc_msgdenied;
1238 		if (err & NFSERR_AUTHERR) {
1239 			*tl++ = rpc_autherr;
1240 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1241 			mreq->m_len -= NFSX_UNSIGNED;
1242 			bpos -= NFSX_UNSIGNED;
1243 		} else {
1244 			*tl++ = rpc_mismatch;
1245 			*tl++ = txdr_unsigned(RPC_VER2);
1246 			*tl = txdr_unsigned(RPC_VER2);
1247 		}
1248 	} else {
1249 		*tl++ = rpc_msgaccepted;
1250 
1251 		/*
1252 		 * For Kerberos authentication, we must send the nickname
1253 		 * verifier back, otherwise just RPCAUTH_NULL.
1254 		 */
1255 		if (nd->nd_flag & ND_KERBFULL) {
1256 		    struct nfsuid *nuidp;
1257 		    struct timeval ktvin, ktvout;
1258 
1259 		    LIST_FOREACH(nuidp, NUIDHASH(slp, nd->nd_cr.cr_uid),
1260 			nu_hash) {
1261 			if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1262 			    (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1263 			     &nuidp->nu_haddr, nd->nd_nam2)))
1264 			    break;
1265 		    }
1266 		    if (nuidp) {
1267 			ktvin.tv_sec =
1268 			    txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1269 			ktvin.tv_usec =
1270 			    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1271 
1272 			/*
1273 			 * Encrypt the timestamp in ecb mode using the
1274 			 * session key.
1275 			 */
1276 #ifdef NFSKERB
1277 			XXX
1278 #endif
1279 
1280 			*tl++ = rpc_auth_kerb;
1281 			*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1282 			*tl = ktvout.tv_sec;
1283 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1284 			*tl++ = ktvout.tv_usec;
1285 			*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1286 		    } else {
1287 			*tl++ = 0;
1288 			*tl++ = 0;
1289 		    }
1290 		} else {
1291 			*tl++ = 0;
1292 			*tl++ = 0;
1293 		}
1294 		switch (err) {
1295 		case EPROGUNAVAIL:
1296 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1297 			break;
1298 		case EPROGMISMATCH:
1299 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1300 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1301 			if (nd->nd_flag & ND_NQNFS) {
1302 				*tl++ = txdr_unsigned(3);
1303 				*tl = txdr_unsigned(3);
1304 			} else {
1305 				*tl++ = txdr_unsigned(2);
1306 				*tl = txdr_unsigned(3);
1307 			}
1308 			break;
1309 		case EPROCUNAVAIL:
1310 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1311 			break;
1312 		case EBADRPC:
1313 			*tl = txdr_unsigned(RPC_GARBAGE);
1314 			break;
1315 		default:
1316 			*tl = 0;
1317 			if (err != NFSERR_RETVOID) {
1318 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1319 				if (err)
1320 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1321 				else
1322 				    *tl = 0;
1323 			}
1324 			break;
1325 		};
1326 	}
1327 
1328 	/*
1329 	 * For nqnfs, piggyback lease as requested.
1330 	 */
1331 	if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1332 		if (nd->nd_flag & ND_LEASE) {
1333 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1334 			*tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1335 			*tl++ = txdr_unsigned(cache);
1336 			*tl++ = txdr_unsigned(nd->nd_duration);
1337 			txdr_hyper(*frev, tl);
1338 		} else {
1339 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1340 			*tl = 0;
1341 		}
1342 	}
1343 	if (mrq != NULL)
1344 		*mrq = mreq;
1345 	*mbp = mb;
1346 	*bposp = bpos;
1347 	if (err != 0 && err != NFSERR_RETVOID)
1348 		nfsstats.srvrpc_errs++;
1349 	return (0);
1350 }
1351 
1352 /*
1353  * Nfs timer routine
1354  * Scan the nfsreq list and retranmit any requests that have timed out
1355  * To avoid retransmission attempts on STREAM sockets (in the future) make
1356  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1357  */
1358 void
1359 nfs_timer(arg)
1360 	void *arg;	/* never used */
1361 {
1362 	struct nfsreq *rep;
1363 	struct mbuf *m;
1364 	struct socket *so;
1365 	struct nfsmount *nmp;
1366 	int timeo;
1367 	int s, error;
1368 #ifdef NFSSERVER
1369 	struct nfssvc_sock *slp;
1370 	static long lasttime = 0;
1371 	u_quad_t cur_usec;
1372 #endif
1373 
1374 	s = splsoftnet();
1375 	TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
1376 		nmp = rep->r_nmp;
1377 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1378 			continue;
1379 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1380 			rep->r_flags |= R_SOFTTERM;
1381 			continue;
1382 		}
1383 		if (rep->r_rtt >= 0) {
1384 			rep->r_rtt++;
1385 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1386 				timeo = nmp->nm_timeo;
1387 			else
1388 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1389 			if (nmp->nm_timeouts > 0)
1390 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1391 			if (rep->r_rtt <= timeo)
1392 				continue;
1393 			if (nmp->nm_timeouts < 8)
1394 				nmp->nm_timeouts++;
1395 		}
1396 		/*
1397 		 * Check for server not responding
1398 		 */
1399 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1400 		     rep->r_rexmit > nmp->nm_deadthresh) {
1401 			nfs_msg(rep->r_procp,
1402 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1403 			    "not responding");
1404 			rep->r_flags |= R_TPRINTFMSG;
1405 		}
1406 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1407 			nfsstats.rpctimeouts++;
1408 			rep->r_flags |= R_SOFTTERM;
1409 			continue;
1410 		}
1411 		if (nmp->nm_sotype != SOCK_DGRAM) {
1412 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1413 				rep->r_rexmit = NFS_MAXREXMIT;
1414 			continue;
1415 		}
1416 		if ((so = nmp->nm_so) == NULL)
1417 			continue;
1418 
1419 		/*
1420 		 * If there is enough space and the window allows..
1421 		 *	Resend it
1422 		 * Set r_rtt to -1 in case we fail to send it now.
1423 		 */
1424 		rep->r_rtt = -1;
1425 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1426 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1427 		    (rep->r_flags & R_SENT) ||
1428 		    nmp->nm_sent < nmp->nm_cwnd) &&
1429 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1430 		        if (so->so_state & SS_ISCONNECTED)
1431 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1432 			    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1433 			else
1434 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1435 			    nmp->nm_nam, (struct mbuf *)0, (struct proc *)0);
1436 			if (error) {
1437 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1438 #ifdef DEBUG
1439 					printf("nfs_timer: ignoring error %d\n",
1440 						error);
1441 #endif
1442 					so->so_error = 0;
1443 				}
1444 			} else {
1445 				/*
1446 				 * Iff first send, start timing
1447 				 * else turn timing off, backoff timer
1448 				 * and divide congestion window by 2.
1449 				 */
1450 				if (rep->r_flags & R_SENT) {
1451 					rep->r_flags &= ~R_TIMING;
1452 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1453 						rep->r_rexmit = NFS_MAXREXMIT;
1454 					nmp->nm_cwnd >>= 1;
1455 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1456 						nmp->nm_cwnd = NFS_CWNDSCALE;
1457 					nfsstats.rpcretries++;
1458 				} else {
1459 					rep->r_flags |= R_SENT;
1460 					nmp->nm_sent += NFS_CWNDSCALE;
1461 				}
1462 				rep->r_rtt = 0;
1463 			}
1464 		}
1465 	}
1466 
1467 #ifdef NFSSERVER
1468 	/*
1469 	 * Call the nqnfs server timer once a second to handle leases.
1470 	 */
1471 	if (lasttime != time.tv_sec) {
1472 		lasttime = time.tv_sec;
1473 		nqnfs_serverd();
1474 	}
1475 
1476 	/*
1477 	 * Scan the write gathering queues for writes that need to be
1478 	 * completed now.
1479 	 */
1480 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1481 	TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
1482 	    if (LIST_FIRST(&slp->ns_tq) &&
1483 		LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec)
1484 		nfsrv_wakenfsd(slp);
1485 	}
1486 #endif /* NFSSERVER */
1487 	splx(s);
1488 	callout_reset(&nfs_timer_ch, nfs_ticks, nfs_timer, NULL);
1489 }
1490 
1491 /*ARGSUSED*/
1492 void
1493 nfs_exit(p, v)
1494 	struct proc *p;
1495 	void *v;
1496 {
1497 	struct nfsreq *rp;
1498 	int s = splsoftnet();
1499 
1500 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
1501 		if (rp->r_procp == p)
1502 			TAILQ_REMOVE(&nfs_reqq, rp, r_chain);
1503 	}
1504 	splx(s);
1505 }
1506 
1507 /*
1508  * Test for a termination condition pending on the process.
1509  * This is used for NFSMNT_INT mounts.
1510  */
1511 int
1512 nfs_sigintr(nmp, rep, p)
1513 	struct nfsmount *nmp;
1514 	struct nfsreq *rep;
1515 	struct proc *p;
1516 {
1517 	sigset_t ss;
1518 
1519 	if (rep && (rep->r_flags & R_SOFTTERM))
1520 		return (EINTR);
1521 	if (!(nmp->nm_flag & NFSMNT_INT))
1522 		return (0);
1523 	if (p) {
1524 		sigpending1(p, &ss);
1525 #if 0
1526 		sigminusset(&p->p_sigctx.ps_sigignore, &ss);
1527 #endif
1528 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1529 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1530 		    sigismember(&ss, SIGQUIT))
1531 			return (EINTR);
1532 	}
1533 	return (0);
1534 }
1535 
1536 /*
1537  * Lock a socket against others.
1538  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1539  * and also to avoid race conditions between the processes with nfs requests
1540  * in progress when a reconnect is necessary.
1541  */
1542 int
1543 nfs_sndlock(flagp, rep)
1544 	int *flagp;
1545 	struct nfsreq *rep;
1546 {
1547 	struct proc *p;
1548 	int slpflag = 0, slptimeo = 0;
1549 
1550 	if (rep) {
1551 		p = rep->r_procp;
1552 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1553 			slpflag = PCATCH;
1554 	} else
1555 		p = (struct proc *)0;
1556 	while (*flagp & NFSMNT_SNDLOCK) {
1557 		if (nfs_sigintr(rep->r_nmp, rep, p))
1558 			return (EINTR);
1559 		*flagp |= NFSMNT_WANTSND;
1560 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1561 			slptimeo);
1562 		if (slpflag == PCATCH) {
1563 			slpflag = 0;
1564 			slptimeo = 2 * hz;
1565 		}
1566 	}
1567 	*flagp |= NFSMNT_SNDLOCK;
1568 	return (0);
1569 }
1570 
1571 /*
1572  * Unlock the stream socket for others.
1573  */
1574 void
1575 nfs_sndunlock(flagp)
1576 	int *flagp;
1577 {
1578 
1579 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1580 		panic("nfs sndunlock");
1581 	*flagp &= ~NFSMNT_SNDLOCK;
1582 	if (*flagp & NFSMNT_WANTSND) {
1583 		*flagp &= ~NFSMNT_WANTSND;
1584 		wakeup((caddr_t)flagp);
1585 	}
1586 }
1587 
1588 int
1589 nfs_rcvlock(rep)
1590 	struct nfsreq *rep;
1591 {
1592 	struct nfsmount *nmp = rep->r_nmp;
1593 	int *flagp = &nmp->nm_iflag;
1594 	int slpflag, slptimeo = 0;
1595 
1596 	if (*flagp & NFSMNT_DISMNT)
1597 		return EIO;
1598 
1599 	if (*flagp & NFSMNT_INT)
1600 		slpflag = PCATCH;
1601 	else
1602 		slpflag = 0;
1603 	while (*flagp & NFSMNT_RCVLOCK) {
1604 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1605 			return (EINTR);
1606 		*flagp |= NFSMNT_WANTRCV;
1607 		nmp->nm_waiters++;
1608 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1609 			slptimeo);
1610 		nmp->nm_waiters--;
1611 		if (*flagp & NFSMNT_DISMNT) {
1612 			wakeup(&nmp->nm_waiters);
1613 			return EIO;
1614 		}
1615 		/* If our reply was received while we were sleeping,
1616 		 * then just return without taking the lock to avoid a
1617 		 * situation where a single iod could 'capture' the
1618 		 * receive lock.
1619 		 */
1620 		if (rep->r_mrep != NULL)
1621 			return (EALREADY);
1622 		if (slpflag == PCATCH) {
1623 			slpflag = 0;
1624 			slptimeo = 2 * hz;
1625 		}
1626 	}
1627 	*flagp |= NFSMNT_RCVLOCK;
1628 	return (0);
1629 }
1630 
1631 /*
1632  * Unlock the stream socket for others.
1633  */
1634 void
1635 nfs_rcvunlock(flagp)
1636 	int *flagp;
1637 {
1638 
1639 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1640 		panic("nfs rcvunlock");
1641 	*flagp &= ~NFSMNT_RCVLOCK;
1642 	if (*flagp & NFSMNT_WANTRCV) {
1643 		*flagp &= ~NFSMNT_WANTRCV;
1644 		wakeup((caddr_t)flagp);
1645 	}
1646 }
1647 
1648 /*
1649  * Parse an RPC request
1650  * - verify it
1651  * - fill in the cred struct.
1652  */
1653 int
1654 nfs_getreq(nd, nfsd, has_header)
1655 	struct nfsrv_descript *nd;
1656 	struct nfsd *nfsd;
1657 	int has_header;
1658 {
1659 	int len, i;
1660 	u_int32_t *tl;
1661 	int32_t t1;
1662 	struct uio uio;
1663 	struct iovec iov;
1664 	caddr_t dpos, cp2, cp;
1665 	u_int32_t nfsvers, auth_type;
1666 	uid_t nickuid;
1667 	int error = 0, nqnfs = 0, ticklen;
1668 	struct mbuf *mrep, *md;
1669 	struct nfsuid *nuidp;
1670 	struct timeval tvin, tvout;
1671 
1672 	mrep = nd->nd_mrep;
1673 	md = nd->nd_md;
1674 	dpos = nd->nd_dpos;
1675 	if (has_header) {
1676 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1677 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1678 		if (*tl++ != rpc_call) {
1679 			m_freem(mrep);
1680 			return (EBADRPC);
1681 		}
1682 	} else
1683 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1684 	nd->nd_repstat = 0;
1685 	nd->nd_flag = 0;
1686 	if (*tl++ != rpc_vers) {
1687 		nd->nd_repstat = ERPCMISMATCH;
1688 		nd->nd_procnum = NFSPROC_NOOP;
1689 		return (0);
1690 	}
1691 	if (*tl != nfs_prog) {
1692 		if (*tl == nqnfs_prog)
1693 			nqnfs++;
1694 		else {
1695 			nd->nd_repstat = EPROGUNAVAIL;
1696 			nd->nd_procnum = NFSPROC_NOOP;
1697 			return (0);
1698 		}
1699 	}
1700 	tl++;
1701 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1702 	if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1703 		(nfsvers != NQNFS_VER3 && nqnfs)) {
1704 		nd->nd_repstat = EPROGMISMATCH;
1705 		nd->nd_procnum = NFSPROC_NOOP;
1706 		return (0);
1707 	}
1708 	if (nqnfs)
1709 		nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1710 	else if (nfsvers == NFS_VER3)
1711 		nd->nd_flag = ND_NFSV3;
1712 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1713 	if (nd->nd_procnum == NFSPROC_NULL)
1714 		return (0);
1715 	if (nd->nd_procnum >= NFS_NPROCS ||
1716 		(!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1717 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1718 		nd->nd_repstat = EPROCUNAVAIL;
1719 		nd->nd_procnum = NFSPROC_NOOP;
1720 		return (0);
1721 	}
1722 	if ((nd->nd_flag & ND_NFSV3) == 0)
1723 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1724 	auth_type = *tl++;
1725 	len = fxdr_unsigned(int, *tl++);
1726 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1727 		m_freem(mrep);
1728 		return (EBADRPC);
1729 	}
1730 
1731 	nd->nd_flag &= ~ND_KERBAUTH;
1732 	/*
1733 	 * Handle auth_unix or auth_kerb.
1734 	 */
1735 	if (auth_type == rpc_auth_unix) {
1736 		len = fxdr_unsigned(int, *++tl);
1737 		if (len < 0 || len > NFS_MAXNAMLEN) {
1738 			m_freem(mrep);
1739 			return (EBADRPC);
1740 		}
1741 		nfsm_adv(nfsm_rndup(len));
1742 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1743 		memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred));
1744 		nd->nd_cr.cr_ref = 1;
1745 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1746 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1747 		len = fxdr_unsigned(int, *tl);
1748 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1749 			m_freem(mrep);
1750 			return (EBADRPC);
1751 		}
1752 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1753 		for (i = 0; i < len; i++)
1754 		    if (i < NGROUPS)
1755 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1756 		    else
1757 			tl++;
1758 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1759 		if (nd->nd_cr.cr_ngroups > 1)
1760 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1761 		len = fxdr_unsigned(int, *++tl);
1762 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1763 			m_freem(mrep);
1764 			return (EBADRPC);
1765 		}
1766 		if (len > 0)
1767 			nfsm_adv(nfsm_rndup(len));
1768 	} else if (auth_type == rpc_auth_kerb) {
1769 		switch (fxdr_unsigned(int, *tl++)) {
1770 		case RPCAKN_FULLNAME:
1771 			ticklen = fxdr_unsigned(int, *tl);
1772 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1773 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1774 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1775 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1776 				m_freem(mrep);
1777 				return (EBADRPC);
1778 			}
1779 			uio.uio_offset = 0;
1780 			uio.uio_iov = &iov;
1781 			uio.uio_iovcnt = 1;
1782 			uio.uio_segflg = UIO_SYSSPACE;
1783 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1784 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1785 			nfsm_mtouio(&uio, uio.uio_resid);
1786 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1787 			if (*tl++ != rpc_auth_kerb ||
1788 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1789 				printf("Bad kerb verifier\n");
1790 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1791 				nd->nd_procnum = NFSPROC_NOOP;
1792 				return (0);
1793 			}
1794 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1795 			tl = (u_int32_t *)cp;
1796 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1797 				printf("Not fullname kerb verifier\n");
1798 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1799 				nd->nd_procnum = NFSPROC_NOOP;
1800 				return (0);
1801 			}
1802 			cp += NFSX_UNSIGNED;
1803 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
1804 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1805 			nd->nd_flag |= ND_KERBFULL;
1806 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1807 			break;
1808 		case RPCAKN_NICKNAME:
1809 			if (len != 2 * NFSX_UNSIGNED) {
1810 				printf("Kerb nickname short\n");
1811 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1812 				nd->nd_procnum = NFSPROC_NOOP;
1813 				return (0);
1814 			}
1815 			nickuid = fxdr_unsigned(uid_t, *tl);
1816 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1817 			if (*tl++ != rpc_auth_kerb ||
1818 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1819 				printf("Kerb nick verifier bad\n");
1820 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1821 				nd->nd_procnum = NFSPROC_NOOP;
1822 				return (0);
1823 			}
1824 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1825 			tvin.tv_sec = *tl++;
1826 			tvin.tv_usec = *tl;
1827 
1828 			LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid),
1829 			    nu_hash) {
1830 				if (nuidp->nu_cr.cr_uid == nickuid &&
1831 				    (!nd->nd_nam2 ||
1832 				     netaddr_match(NU_NETFAM(nuidp),
1833 				      &nuidp->nu_haddr, nd->nd_nam2)))
1834 					break;
1835 			}
1836 			if (!nuidp) {
1837 				nd->nd_repstat =
1838 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1839 				nd->nd_procnum = NFSPROC_NOOP;
1840 				return (0);
1841 			}
1842 
1843 			/*
1844 			 * Now, decrypt the timestamp using the session key
1845 			 * and validate it.
1846 			 */
1847 #ifdef NFSKERB
1848 			XXX
1849 #endif
1850 
1851 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1852 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1853 			if (nuidp->nu_expire < time.tv_sec ||
1854 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1855 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1856 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1857 				nuidp->nu_expire = 0;
1858 				nd->nd_repstat =
1859 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1860 				nd->nd_procnum = NFSPROC_NOOP;
1861 				return (0);
1862 			}
1863 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1864 			nd->nd_flag |= ND_KERBNICK;
1865 		};
1866 	} else {
1867 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1868 		nd->nd_procnum = NFSPROC_NOOP;
1869 		return (0);
1870 	}
1871 
1872 	/*
1873 	 * For nqnfs, get piggybacked lease request.
1874 	 */
1875 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
1876 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1877 		nd->nd_flag |= fxdr_unsigned(int, *tl);
1878 		if (nd->nd_flag & ND_LEASE) {
1879 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1880 			nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
1881 		} else
1882 			nd->nd_duration = NQ_MINLEASE;
1883 	} else
1884 		nd->nd_duration = NQ_MINLEASE;
1885 	nd->nd_md = md;
1886 	nd->nd_dpos = dpos;
1887 	return (0);
1888 nfsmout:
1889 	return (error);
1890 }
1891 
1892 int
1893 nfs_msg(p, server, msg)
1894 	struct proc *p;
1895 	char *server, *msg;
1896 {
1897 	tpr_t tpr;
1898 
1899 	if (p)
1900 		tpr = tprintf_open(p);
1901 	else
1902 		tpr = NULL;
1903 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1904 	tprintf_close(tpr);
1905 	return (0);
1906 }
1907 
1908 #ifdef NFSSERVER
1909 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
1910 				    struct nfssvc_sock *, struct proc *,
1911 				    struct mbuf **)) = {
1912 	nfsrv_null,
1913 	nfsrv_getattr,
1914 	nfsrv_setattr,
1915 	nfsrv_lookup,
1916 	nfsrv3_access,
1917 	nfsrv_readlink,
1918 	nfsrv_read,
1919 	nfsrv_write,
1920 	nfsrv_create,
1921 	nfsrv_mkdir,
1922 	nfsrv_symlink,
1923 	nfsrv_mknod,
1924 	nfsrv_remove,
1925 	nfsrv_rmdir,
1926 	nfsrv_rename,
1927 	nfsrv_link,
1928 	nfsrv_readdir,
1929 	nfsrv_readdirplus,
1930 	nfsrv_statfs,
1931 	nfsrv_fsinfo,
1932 	nfsrv_pathconf,
1933 	nfsrv_commit,
1934 	nqnfsrv_getlease,
1935 	nqnfsrv_vacated,
1936 	nfsrv_noop,
1937 	nfsrv_noop
1938 };
1939 
1940 /*
1941  * Socket upcall routine for the nfsd sockets.
1942  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1943  * Essentially do as much as possible non-blocking, else punt and it will
1944  * be called with M_WAIT from an nfsd.
1945  */
1946 void
1947 nfsrv_rcv(so, arg, waitflag)
1948 	struct socket *so;
1949 	caddr_t arg;
1950 	int waitflag;
1951 {
1952 	struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1953 	struct mbuf *m;
1954 	struct mbuf *mp, *nam;
1955 	struct uio auio;
1956 	int flags, error;
1957 
1958 	if ((slp->ns_flag & SLP_VALID) == 0)
1959 		return;
1960 #ifdef notdef
1961 	/*
1962 	 * Define this to test for nfsds handling this under heavy load.
1963 	 */
1964 	if (waitflag == M_DONTWAIT) {
1965 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1966 	}
1967 #endif
1968 	auio.uio_procp = NULL;
1969 	if (so->so_type == SOCK_STREAM) {
1970 		/*
1971 		 * If there are already records on the queue, defer soreceive()
1972 		 * to an nfsd so that there is feedback to the TCP layer that
1973 		 * the nfs servers are heavily loaded.
1974 		 */
1975 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1976 			slp->ns_flag |= SLP_NEEDQ;
1977 			goto dorecs;
1978 		}
1979 
1980 		/*
1981 		 * Do soreceive().
1982 		 */
1983 		auio.uio_resid = 1000000000;
1984 		flags = MSG_DONTWAIT;
1985 		error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1986 		if (error || mp == (struct mbuf *)0) {
1987 			if (error == EWOULDBLOCK)
1988 				slp->ns_flag |= SLP_NEEDQ;
1989 			else
1990 				slp->ns_flag |= SLP_DISCONN;
1991 			goto dorecs;
1992 		}
1993 		m = mp;
1994 		if (slp->ns_rawend) {
1995 			slp->ns_rawend->m_next = m;
1996 			slp->ns_cc += 1000000000 - auio.uio_resid;
1997 		} else {
1998 			slp->ns_raw = m;
1999 			slp->ns_cc = 1000000000 - auio.uio_resid;
2000 		}
2001 		while (m->m_next)
2002 			m = m->m_next;
2003 		slp->ns_rawend = m;
2004 
2005 		/*
2006 		 * Now try and parse record(s) out of the raw stream data.
2007 		 */
2008 		error = nfsrv_getstream(slp, waitflag);
2009 		if (error) {
2010 			if (error == EPERM)
2011 				slp->ns_flag |= SLP_DISCONN;
2012 			else
2013 				slp->ns_flag |= SLP_NEEDQ;
2014 		}
2015 	} else {
2016 		do {
2017 			auio.uio_resid = 1000000000;
2018 			flags = MSG_DONTWAIT;
2019 			error = (*so->so_receive)(so, &nam, &auio, &mp,
2020 						(struct mbuf **)0, &flags);
2021 			if (mp) {
2022 				if (nam) {
2023 					m = nam;
2024 					m->m_next = mp;
2025 				} else
2026 					m = mp;
2027 				if (slp->ns_recend)
2028 					slp->ns_recend->m_nextpkt = m;
2029 				else
2030 					slp->ns_rec = m;
2031 				slp->ns_recend = m;
2032 				m->m_nextpkt = (struct mbuf *)0;
2033 			}
2034 			if (error) {
2035 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2036 					&& error != EWOULDBLOCK) {
2037 					slp->ns_flag |= SLP_DISCONN;
2038 					goto dorecs;
2039 				}
2040 			}
2041 		} while (mp);
2042 	}
2043 
2044 	/*
2045 	 * Now try and process the request records, non-blocking.
2046 	 */
2047 dorecs:
2048 	if (waitflag == M_DONTWAIT &&
2049 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
2050 		nfsrv_wakenfsd(slp);
2051 }
2052 
2053 /*
2054  * Try and extract an RPC request from the mbuf data list received on a
2055  * stream socket. The "waitflag" argument indicates whether or not it
2056  * can sleep.
2057  */
2058 int
2059 nfsrv_getstream(slp, waitflag)
2060 	struct nfssvc_sock *slp;
2061 	int waitflag;
2062 {
2063 	struct mbuf *m, **mpp;
2064 	struct mbuf *recm;
2065 	u_int32_t recmark;
2066 
2067 	if (slp->ns_flag & SLP_GETSTREAM)
2068 		panic("nfs getstream");
2069 	slp->ns_flag |= SLP_GETSTREAM;
2070 	for (;;) {
2071 		if (slp->ns_reclen == 0) {
2072 			if (slp->ns_cc < NFSX_UNSIGNED) {
2073 				slp->ns_flag &= ~SLP_GETSTREAM;
2074 				return (0);
2075 			}
2076 			m = slp->ns_raw;
2077 			m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark);
2078 			m_adj(m, NFSX_UNSIGNED);
2079 			slp->ns_cc -= NFSX_UNSIGNED;
2080 			recmark = ntohl(recmark);
2081 			slp->ns_reclen = recmark & ~0x80000000;
2082 			if (recmark & 0x80000000)
2083 				slp->ns_flag |= SLP_LASTFRAG;
2084 			else
2085 				slp->ns_flag &= ~SLP_LASTFRAG;
2086 			if (slp->ns_reclen > NFS_MAXPACKET) {
2087 				slp->ns_flag &= ~SLP_GETSTREAM;
2088 				return (EPERM);
2089 			}
2090 		}
2091 
2092 		/*
2093 		 * Now get the record part.
2094 		 *
2095 		 * Note that slp->ns_reclen may be 0.  Linux sometimes
2096 		 * generates 0-length records.
2097 		 */
2098 		if (slp->ns_cc == slp->ns_reclen) {
2099 			recm = slp->ns_raw;
2100 			slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2101 			slp->ns_cc = slp->ns_reclen = 0;
2102 		} else if (slp->ns_cc > slp->ns_reclen) {
2103 			recm = slp->ns_raw;
2104 			m = m_split(recm, slp->ns_reclen, waitflag);
2105 			if (m == NULL) {
2106 				slp->ns_flag &= ~SLP_GETSTREAM;
2107 				return (EWOULDBLOCK);
2108 			}
2109 			m_claim(recm, &nfs_mowner);
2110 			slp->ns_raw = m;
2111 			if (m->m_next == NULL)
2112 				slp->ns_rawend = m;
2113 			slp->ns_cc -= slp->ns_reclen;
2114 			slp->ns_reclen = 0;
2115 		} else {
2116 			slp->ns_flag &= ~SLP_GETSTREAM;
2117 			return (0);
2118 		}
2119 
2120 		/*
2121 		 * Accumulate the fragments into a record.
2122 		 */
2123 		mpp = &slp->ns_frag;
2124 		while (*mpp)
2125 			mpp = &((*mpp)->m_next);
2126 		*mpp = recm;
2127 		if (slp->ns_flag & SLP_LASTFRAG) {
2128 			if (slp->ns_recend)
2129 				slp->ns_recend->m_nextpkt = slp->ns_frag;
2130 			else
2131 				slp->ns_rec = slp->ns_frag;
2132 			slp->ns_recend = slp->ns_frag;
2133 			slp->ns_frag = (struct mbuf *)0;
2134 		}
2135 	}
2136 }
2137 
2138 /*
2139  * Parse an RPC header.
2140  */
2141 int
2142 nfsrv_dorec(slp, nfsd, ndp)
2143 	struct nfssvc_sock *slp;
2144 	struct nfsd *nfsd;
2145 	struct nfsrv_descript **ndp;
2146 {
2147 	struct mbuf *m, *nam;
2148 	struct nfsrv_descript *nd;
2149 	int error;
2150 
2151 	*ndp = NULL;
2152 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2153 	    (m = slp->ns_rec) == (struct mbuf *)0)
2154 		return (ENOBUFS);
2155 	slp->ns_rec = m->m_nextpkt;
2156 	if (slp->ns_rec)
2157 		m->m_nextpkt = (struct mbuf *)0;
2158 	else
2159 		slp->ns_recend = (struct mbuf *)0;
2160 	if (m->m_type == MT_SONAME) {
2161 		nam = m;
2162 		m = m->m_next;
2163 		nam->m_next = NULL;
2164 	} else
2165 		nam = NULL;
2166 	MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2167 		M_NFSRVDESC, M_WAITOK);
2168 	nd->nd_md = nd->nd_mrep = m;
2169 	nd->nd_nam2 = nam;
2170 	nd->nd_dpos = mtod(m, caddr_t);
2171 	error = nfs_getreq(nd, nfsd, TRUE);
2172 	if (error) {
2173 		m_freem(nam);
2174 		free((caddr_t)nd, M_NFSRVDESC);
2175 		return (error);
2176 	}
2177 	*ndp = nd;
2178 	nfsd->nfsd_nd = nd;
2179 	return (0);
2180 }
2181 
2182 
2183 /*
2184  * Search for a sleeping nfsd and wake it up.
2185  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2186  * running nfsds will go look for the work in the nfssvc_sock list.
2187  */
2188 void
2189 nfsrv_wakenfsd(slp)
2190 	struct nfssvc_sock *slp;
2191 {
2192 	struct nfsd *nd;
2193 
2194 	if ((slp->ns_flag & SLP_VALID) == 0)
2195 		return;
2196 	TAILQ_FOREACH(nd, &nfsd_head, nfsd_chain) {
2197 		if (nd->nfsd_flag & NFSD_WAITING) {
2198 			nd->nfsd_flag &= ~NFSD_WAITING;
2199 			if (nd->nfsd_slp)
2200 				panic("nfsd wakeup");
2201 			slp->ns_sref++;
2202 			nd->nfsd_slp = slp;
2203 			wakeup((caddr_t)nd);
2204 			return;
2205 		}
2206 	}
2207 	slp->ns_flag |= SLP_DOREC;
2208 	nfsd_head_flag |= NFSD_CHECKSLP;
2209 }
2210 #endif /* NFSSERVER */
2211