xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: nfs_socket.c,v 1.111 2004/09/17 14:11:26 skrll 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. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_socket.c	8.5 (Berkeley) 3/30/95
35  */
36 
37 /*
38  * Socket operations for use by nfs
39  */
40 
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: nfs_socket.c,v 1.111 2004/09/17 14:11:26 skrll Exp $");
43 
44 #include "fs_nfs.h"
45 #include "opt_nfs.h"
46 #include "opt_nfsserver.h"
47 #include "opt_mbuftrace.h"
48 #include "opt_inet.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/callout.h>
53 #include <sys/proc.h>
54 #include <sys/mount.h>
55 #include <sys/kernel.h>
56 #include <sys/mbuf.h>
57 #include <sys/vnode.h>
58 #include <sys/domain.h>
59 #include <sys/protosw.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
62 #include <sys/syslog.h>
63 #include <sys/tprintf.h>
64 #include <sys/namei.h>
65 #include <sys/signal.h>
66 #include <sys/signalvar.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/tcp.h>
70 
71 #include <nfs/rpcv2.h>
72 #include <nfs/nfsproto.h>
73 #include <nfs/nfs.h>
74 #include <nfs/xdr_subs.h>
75 #include <nfs/nfsm_subs.h>
76 #include <nfs/nfsmount.h>
77 #include <nfs/nfsnode.h>
78 #include <nfs/nfsrtt.h>
79 #include <nfs/nqnfs.h>
80 #include <nfs/nfs_var.h>
81 
82 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
83 #ifdef MBUFTRACE
84 struct mowner nfs_mowner = { "nfs" };
85 #endif
86 
87 /*
88  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
89  * Use the mean and mean deviation of rtt for the appropriate type of rpc
90  * for the frequent rpcs and a default for the others.
91  * The justification for doing "other" this way is that these rpcs
92  * happen so infrequently that timer est. would probably be stale.
93  * Also, since many of these rpcs are
94  * non-idempotent, a conservative timeout is desired.
95  * getattr, lookup - A+2D
96  * read, write     - A+4D
97  * other           - nm_timeo
98  */
99 #define	NFS_RTO(n, t) \
100 	((t) == 0 ? (n)->nm_timeo : \
101 	 ((t) < 3 ? \
102 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
103 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
104 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
105 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
106 /*
107  * External data, mostly RPC constants in XDR form
108  */
109 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
110 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
111 	rpc_auth_kerb;
112 extern u_int32_t nfs_prog, nqnfs_prog;
113 extern time_t nqnfsstarttime;
114 extern const int nfsv3_procid[NFS_NPROCS];
115 extern int nfs_ticks;
116 
117 /*
118  * Defines which timer to use for the procnum.
119  * 0 - default
120  * 1 - getattr
121  * 2 - lookup
122  * 3 - read
123  * 4 - write
124  */
125 static const int proct[NFS_NPROCS] = {
126 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
127 	0, 0, 0,
128 };
129 
130 /*
131  * There is a congestion window for outstanding rpcs maintained per mount
132  * point. The cwnd size is adjusted in roughly the way that:
133  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
134  * SIGCOMM '88". ACM, August 1988.
135  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
136  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
137  * of rpcs is in progress.
138  * (The sent count and cwnd are scaled for integer arith.)
139  * Variants of "slow start" were tried and were found to be too much of a
140  * performance hit (ave. rtt 3 times larger),
141  * I suspect due to the large rtt that nfs rpcs have.
142  */
143 #define	NFS_CWNDSCALE	256
144 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
145 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
146 int nfsrtton = 0;
147 struct nfsrtt nfsrtt;
148 struct nfsreqhead nfs_reqq;
149 
150 struct callout nfs_timer_ch = CALLOUT_INITIALIZER_SETFUNC(nfs_timer, NULL);
151 
152 /*
153  * Initialize sockets and congestion for a new NFS connection.
154  * We do not free the sockaddr if error.
155  */
156 int
157 nfs_connect(nmp, rep, p)
158 	struct nfsmount *nmp;
159 	struct nfsreq *rep;
160 	struct proc *p;
161 {
162 	struct socket *so;
163 	int s, error, rcvreserve, sndreserve;
164 	struct sockaddr *saddr;
165 	struct sockaddr_in *sin;
166 #ifdef INET6
167 	struct sockaddr_in6 *sin6;
168 #endif
169 	struct mbuf *m;
170 
171 	nmp->nm_so = (struct socket *)0;
172 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
173 	error = socreate(saddr->sa_family, &nmp->nm_so,
174 		nmp->nm_sotype, nmp->nm_soproto, p);
175 	if (error)
176 		goto bad;
177 	so = nmp->nm_so;
178 #ifdef MBUFTRACE
179 	so->so_mowner = &nfs_mowner;
180 	so->so_rcv.sb_mowner = &nfs_mowner;
181 	so->so_snd.sb_mowner = &nfs_mowner;
182 #endif
183 	nmp->nm_soflags = so->so_proto->pr_flags;
184 
185 	/*
186 	 * Some servers require that the client port be a reserved port number.
187 	 */
188 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
189 		m = m_get(M_WAIT, MT_SOOPTS);
190 		MCLAIM(m, so->so_mowner);
191 		*mtod(m, int32_t *) = IP_PORTRANGE_LOW;
192 		m->m_len = sizeof(int32_t);
193 		if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m)))
194 			goto bad;
195 		m = m_get(M_WAIT, MT_SONAME);
196 		MCLAIM(m, so->so_mowner);
197 		sin = mtod(m, struct sockaddr_in *);
198 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
199 		sin->sin_family = AF_INET;
200 		sin->sin_addr.s_addr = INADDR_ANY;
201 		sin->sin_port = 0;
202 		error = sobind(so, m, &proc0);
203 		m_freem(m);
204 		if (error)
205 			goto bad;
206 	}
207 #ifdef INET6
208 	if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) {
209 		m = m_get(M_WAIT, MT_SOOPTS);
210 		MCLAIM(m, so->so_mowner);
211 		*mtod(m, int32_t *) = IPV6_PORTRANGE_LOW;
212 		m->m_len = sizeof(int32_t);
213 		if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m)))
214 			goto bad;
215 		m = m_get(M_WAIT, MT_SONAME);
216 		MCLAIM(m, so->so_mowner);
217 		sin6 = mtod(m, struct sockaddr_in6 *);
218 		sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6);
219 		sin6->sin6_family = AF_INET6;
220 		sin6->sin6_addr = in6addr_any;
221 		sin6->sin6_port = 0;
222 		error = sobind(so, m, &proc0);
223 		m_freem(m);
224 		if (error)
225 			goto bad;
226 	}
227 #endif
228 
229 	/*
230 	 * Protocols that do not require connections may be optionally left
231 	 * unconnected for servers that reply from a port other than NFS_PORT.
232 	 */
233 	if (nmp->nm_flag & NFSMNT_NOCONN) {
234 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
235 			error = ENOTCONN;
236 			goto bad;
237 		}
238 	} else {
239 		error = soconnect(so, nmp->nm_nam, p);
240 		if (error)
241 			goto bad;
242 
243 		/*
244 		 * Wait for the connection to complete. Cribbed from the
245 		 * connect system call but with the wait timing out so
246 		 * that interruptible mounts don't hang here for a long time.
247 		 */
248 		s = splsoftnet();
249 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
250 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
251 				"nfscn1", 2 * hz);
252 			if ((so->so_state & SS_ISCONNECTING) &&
253 			    so->so_error == 0 && rep &&
254 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
255 				so->so_state &= ~SS_ISCONNECTING;
256 				splx(s);
257 				goto bad;
258 			}
259 		}
260 		if (so->so_error) {
261 			error = so->so_error;
262 			so->so_error = 0;
263 			splx(s);
264 			goto bad;
265 		}
266 		splx(s);
267 	}
268 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
269 		so->so_rcv.sb_timeo = (5 * hz);
270 		so->so_snd.sb_timeo = (5 * hz);
271 	} else {
272 		/*
273 		 * enable receive timeout to detect server crash and reconnect.
274 		 * otherwise, we can be stuck in soreceive forever.
275 		 */
276 		so->so_rcv.sb_timeo = (5 * hz);
277 		so->so_snd.sb_timeo = 0;
278 	}
279 	if (nmp->nm_sotype == SOCK_DGRAM) {
280 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
281 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
282 		    NFS_MAXPKTHDR) * 2;
283 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
284 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
285 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
286 		    NFS_MAXPKTHDR) * 2;
287 	} else {
288 		if (nmp->nm_sotype != SOCK_STREAM)
289 			panic("nfscon sotype");
290 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
291 			m = m_get(M_WAIT, MT_SOOPTS);
292 			MCLAIM(m, so->so_mowner);
293 			*mtod(m, int32_t *) = 1;
294 			m->m_len = sizeof(int32_t);
295 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
296 		}
297 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
298 			m = m_get(M_WAIT, MT_SOOPTS);
299 			MCLAIM(m, so->so_mowner);
300 			*mtod(m, int32_t *) = 1;
301 			m->m_len = sizeof(int32_t);
302 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
303 		}
304 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
305 		    sizeof (u_int32_t)) * 2;
306 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
307 		    sizeof (u_int32_t)) * 2;
308 	}
309 	error = soreserve(so, sndreserve, rcvreserve);
310 	if (error)
311 		goto bad;
312 	so->so_rcv.sb_flags |= SB_NOINTR;
313 	so->so_snd.sb_flags |= SB_NOINTR;
314 
315 	/* Initialize other non-zero congestion variables */
316 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
317 		NFS_TIMEO << 3;
318 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
319 		nmp->nm_sdrtt[3] = 0;
320 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
321 	nmp->nm_sent = 0;
322 	nmp->nm_timeouts = 0;
323 	return (0);
324 
325 bad:
326 	nfs_disconnect(nmp);
327 	return (error);
328 }
329 
330 /*
331  * Reconnect routine:
332  * Called when a connection is broken on a reliable protocol.
333  * - clean up the old socket
334  * - nfs_connect() again
335  * - set R_MUSTRESEND for all outstanding requests on mount point
336  * If this fails the mount point is DEAD!
337  * nb: Must be called with the nfs_sndlock() set on the mount point.
338  */
339 int
340 nfs_reconnect(rep, p)
341 	struct nfsreq *rep;
342 	struct proc *p;
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, p)) != 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 			if ((rp->r_flags & R_MUSTRESEND) == 0)
362 				rp->r_flags |= R_MUSTRESEND | R_REXMITTED;
363 			rp->r_rexmit = 0;
364 		}
365 	}
366 	return (0);
367 }
368 
369 /*
370  * NFS disconnect. Clean up and unlink.
371  */
372 void
373 nfs_disconnect(nmp)
374 	struct nfsmount *nmp;
375 {
376 	struct socket *so;
377 	int drain = 0;
378 
379 	if (nmp->nm_so) {
380 		so = nmp->nm_so;
381 		nmp->nm_so = (struct socket *)0;
382 		soshutdown(so, 2);
383 		drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
384 		if (drain) {
385 			/*
386 			 * soshutdown() above should wake up the current
387 			 * listener.
388 			 * Now wake up those waiting for the receive lock, and
389 			 * wait for them to go away unhappy, to prevent *nmp
390 			 * from evaporating while they're sleeping.
391 			 */
392 			while (nmp->nm_waiters > 0) {
393 				wakeup (&nmp->nm_iflag);
394 				(void) tsleep(&nmp->nm_waiters, PVFS,
395 				    "nfsdis", 0);
396 			}
397 		}
398 		soclose(so);
399 	}
400 #ifdef DIAGNOSTIC
401 	if (drain && (nmp->nm_waiters > 0))
402 		panic("nfs_disconnect: waiters left after drain?");
403 #endif
404 }
405 
406 void
407 nfs_safedisconnect(nmp)
408 	struct nfsmount *nmp;
409 {
410 	struct nfsreq dummyreq;
411 
412 	memset(&dummyreq, 0, sizeof(dummyreq));
413 	dummyreq.r_nmp = nmp;
414 	nfs_rcvlock(&dummyreq); /* XXX ignored error return */
415 	nfs_disconnect(nmp);
416 	nfs_rcvunlock(nmp);
417 }
418 
419 /*
420  * This is the nfs send routine. For connection based socket types, it
421  * must be called with an nfs_sndlock() on the socket.
422  * "rep == NULL" indicates that it has been called from a server.
423  * For the client side:
424  * - return EINTR if the RPC is terminated, 0 otherwise
425  * - set R_MUSTRESEND if the send fails for any reason
426  * - do any cleanup required by recoverable socket errors (? ? ?)
427  * For the server side:
428  * - return EINTR or ERESTART if interrupted by a signal
429  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
430  * - do any cleanup required by recoverable socket errors (? ? ?)
431  */
432 int
433 nfs_send(so, nam, top, rep, p)
434 	struct socket *so;
435 	struct mbuf *nam;
436 	struct mbuf *top;
437 	struct nfsreq *rep;
438 	struct proc *p;
439 {
440 	struct mbuf *sendnam;
441 	int error, soflags, flags;
442 
443 	/* XXX nfs_doio()/nfs_request() calls with  rep->r_procp == NULL */
444 	if (p == NULL && rep->r_procp == NULL)
445 		p = curproc;
446 
447 	if (rep) {
448 		if (rep->r_flags & R_SOFTTERM) {
449 			m_freem(top);
450 			return (EINTR);
451 		}
452 		if ((so = rep->r_nmp->nm_so) == NULL) {
453 			rep->r_flags |= R_MUSTRESEND;
454 			m_freem(top);
455 			return (0);
456 		}
457 		rep->r_flags &= ~R_MUSTRESEND;
458 		soflags = rep->r_nmp->nm_soflags;
459 	} else
460 		soflags = so->so_proto->pr_flags;
461 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
462 		sendnam = (struct mbuf *)0;
463 	else
464 		sendnam = nam;
465 	if (so->so_type == SOCK_SEQPACKET)
466 		flags = MSG_EOR;
467 	else
468 		flags = 0;
469 
470 	error = (*so->so_send)(so, sendnam, (struct uio *)0, top,
471 		    (struct mbuf *)0, flags,  p);
472 	if (error) {
473 		if (rep) {
474 			if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
475 				/*
476 				 * We're too fast for the network/driver,
477 				 * and UDP isn't flowcontrolled.
478 				 * We need to resend. This is not fatal,
479 				 * just try again.
480 				 *
481 				 * Could be smarter here by doing some sort
482 				 * of a backoff, but this is rare.
483 				 */
484 				rep->r_flags |= R_MUSTRESEND;
485 			} else {
486 				if (error != EPIPE)
487 					log(LOG_INFO,
488 					    "nfs send error %d for %s\n",
489 					    error,
490 					    rep->r_nmp->nm_mountp->
491 						    mnt_stat.f_mntfromname);
492 				/*
493 				 * Deal with errors for the client side.
494 				 */
495 				if (rep->r_flags & R_SOFTTERM)
496 					error = EINTR;
497 				else
498 					rep->r_flags |= R_MUSTRESEND;
499 			}
500 		} else {
501 			/*
502 			 * See above. This error can happen under normal
503 			 * circumstances and the log is too noisy.
504 			 * The error will still show up in nfsstat.
505 			 */
506 			if (error != ENOBUFS || so->so_type != SOCK_DGRAM)
507 				log(LOG_INFO, "nfsd send error %d\n", error);
508 		}
509 
510 		/*
511 		 * Handle any recoverable (soft) socket errors here. (? ? ?)
512 		 */
513 		if (error != EINTR && error != ERESTART &&
514 			error != EWOULDBLOCK && error != EPIPE)
515 			error = 0;
516 	}
517 	return (error);
518 }
519 
520 #ifdef NFS
521 /*
522  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
523  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
524  * Mark and consolidate the data into a new mbuf list.
525  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
526  *     small mbufs.
527  * For SOCK_STREAM we must be very careful to read an entire record once
528  * we have read any of it, even if the system call has been interrupted.
529  */
530 int
531 nfs_receive(rep, aname, mp, p)
532 	struct nfsreq *rep;
533 	struct mbuf **aname;
534 	struct mbuf **mp;
535 	struct proc *p;
536 {
537 	struct socket *so;
538 	struct uio auio;
539 	struct iovec aio;
540 	struct mbuf *m;
541 	struct mbuf *control;
542 	u_int32_t len;
543 	struct mbuf **getnam;
544 	int error, sotype, rcvflg;
545 
546 	/*
547 	 * Set up arguments for soreceive()
548 	 */
549 	*mp = (struct mbuf *)0;
550 	*aname = (struct mbuf *)0;
551 	sotype = rep->r_nmp->nm_sotype;
552 
553 	/*
554 	 * For reliable protocols, lock against other senders/receivers
555 	 * in case a reconnect is necessary.
556 	 * For SOCK_STREAM, first get the Record Mark to find out how much
557 	 * more there is to get.
558 	 * We must lock the socket against other receivers
559 	 * until we have an entire rpc request/reply.
560 	 */
561 	if (sotype != SOCK_DGRAM) {
562 		error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
563 		if (error)
564 			return (error);
565 tryagain:
566 		/*
567 		 * Check for fatal errors and resending request.
568 		 */
569 		/*
570 		 * Ugh: If a reconnect attempt just happened, nm_so
571 		 * would have changed. NULL indicates a failed
572 		 * attempt that has essentially shut down this
573 		 * mount point.
574 		 */
575 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
576 			nfs_sndunlock(&rep->r_nmp->nm_iflag);
577 			return (EINTR);
578 		}
579 		so = rep->r_nmp->nm_so;
580 		if (!so) {
581 			error = nfs_reconnect(rep, p);
582 			if (error) {
583 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
584 				return (error);
585 			}
586 			goto tryagain;
587 		}
588 		while (rep->r_flags & R_MUSTRESEND) {
589 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
590 			nfsstats.rpcretries++;
591 			rep->r_rtt = 0;
592 			rep->r_flags &= ~R_TIMING;
593 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, p);
594 			if (error) {
595 				if (error == EINTR || error == ERESTART ||
596 				    (error = nfs_reconnect(rep, p)) != 0) {
597 					nfs_sndunlock(&rep->r_nmp->nm_iflag);
598 					return (error);
599 				}
600 				goto tryagain;
601 			}
602 		}
603 		nfs_sndunlock(&rep->r_nmp->nm_iflag);
604 		if (sotype == SOCK_STREAM) {
605 			aio.iov_base = (caddr_t) &len;
606 			aio.iov_len = sizeof(u_int32_t);
607 			auio.uio_iov = &aio;
608 			auio.uio_iovcnt = 1;
609 			auio.uio_segflg = UIO_SYSSPACE;
610 			auio.uio_rw = UIO_READ;
611 			auio.uio_offset = 0;
612 			auio.uio_resid = sizeof(u_int32_t);
613 			auio.uio_procp = NULL;
614 			do {
615 			   rcvflg = MSG_WAITALL;
616 			   error = (*so->so_receive)(so, (struct mbuf **)0, &auio,
617 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
618 			   if (error == EWOULDBLOCK && rep) {
619 				if (rep->r_flags & R_SOFTTERM)
620 					return (EINTR);
621 				/*
622 				 * if it seems that the server died after it
623 				 * received our request, set EPIPE so that
624 				 * we'll reconnect and retransmit requests.
625 				 */
626 				if (rep->r_rexmit >= rep->r_nmp->nm_retry) {
627 					nfsstats.rpctimeouts++;
628 					error = EPIPE;
629 				}
630 			   }
631 			} while (error == EWOULDBLOCK);
632 			if (!error && auio.uio_resid > 0) {
633 			    /*
634 			     * Don't log a 0 byte receive; it means
635 			     * that the socket has been closed, and
636 			     * can happen during normal operation
637 			     * (forcible unmount or Solaris server).
638 			     */
639 			    if (auio.uio_resid != sizeof (u_int32_t))
640 			      log(LOG_INFO,
641 				 "short receive (%lu/%lu) from nfs server %s\n",
642 				 (u_long)sizeof(u_int32_t) - auio.uio_resid,
643 				 (u_long)sizeof(u_int32_t),
644 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
645 			    error = EPIPE;
646 			}
647 			if (error)
648 				goto errout;
649 			len = ntohl(len) & ~0x80000000;
650 			/*
651 			 * This is SERIOUS! We are out of sync with the sender
652 			 * and forcing a disconnect/reconnect is all I can do.
653 			 */
654 			if (len > NFS_MAXPACKET) {
655 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
656 				"impossible packet length",
657 				len,
658 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
659 			    error = EFBIG;
660 			    goto errout;
661 			}
662 			auio.uio_resid = len;
663 			do {
664 			    rcvflg = MSG_WAITALL;
665 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
666 				&auio, mp, (struct mbuf **)0, &rcvflg);
667 			} while (error == EWOULDBLOCK || error == EINTR ||
668 				 error == ERESTART);
669 			if (!error && auio.uio_resid > 0) {
670 			    if (len != auio.uio_resid)
671 			      log(LOG_INFO,
672 				"short receive (%lu/%d) from nfs server %s\n",
673 				(u_long)len - auio.uio_resid, len,
674 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
675 			    error = EPIPE;
676 			}
677 		} else {
678 			/*
679 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
680 			 * and soreceive() will return when it has either a
681 			 * control msg or a data msg.
682 			 * We have no use for control msg., but must grab them
683 			 * and then throw them away so we know what is going
684 			 * on.
685 			 */
686 			auio.uio_resid = len = 100000000; /* Anything Big */
687 			auio.uio_procp = p;
688 			do {
689 			    rcvflg = 0;
690 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
691 				&auio, mp, &control, &rcvflg);
692 			    if (control)
693 				m_freem(control);
694 			    if (error == EWOULDBLOCK && rep) {
695 				if (rep->r_flags & R_SOFTTERM)
696 					return (EINTR);
697 			    }
698 			} while (error == EWOULDBLOCK ||
699 				 (!error && *mp == NULL && control));
700 			if ((rcvflg & MSG_EOR) == 0)
701 				printf("Egad!!\n");
702 			if (!error && *mp == NULL)
703 				error = EPIPE;
704 			len -= auio.uio_resid;
705 		}
706 errout:
707 		if (error && error != EINTR && error != ERESTART) {
708 			m_freem(*mp);
709 			*mp = (struct mbuf *)0;
710 			if (error != EPIPE)
711 				log(LOG_INFO,
712 				    "receive error %d from nfs server %s\n",
713 				    error,
714 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
715 			error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
716 			if (!error)
717 				error = nfs_reconnect(rep, p);
718 			if (!error)
719 				goto tryagain;
720 			else
721 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
722 		}
723 	} else {
724 		if ((so = rep->r_nmp->nm_so) == NULL)
725 			return (EACCES);
726 		if (so->so_state & SS_ISCONNECTED)
727 			getnam = (struct mbuf **)0;
728 		else
729 			getnam = aname;
730 		auio.uio_resid = len = 1000000;
731 		auio.uio_procp = p;
732 		do {
733 			rcvflg = 0;
734 			error =  (*so->so_receive)(so, getnam, &auio, mp,
735 				(struct mbuf **)0, &rcvflg);
736 			if (error == EWOULDBLOCK &&
737 			    (rep->r_flags & R_SOFTTERM))
738 				return (EINTR);
739 		} while (error == EWOULDBLOCK);
740 		len -= auio.uio_resid;
741 		if (!error && *mp == NULL)
742 			error = EPIPE;
743 	}
744 	if (error) {
745 		m_freem(*mp);
746 		*mp = (struct mbuf *)0;
747 	}
748 	return (error);
749 }
750 
751 /*
752  * Implement receipt of reply on a socket.
753  * We must search through the list of received datagrams matching them
754  * with outstanding requests using the xid, until ours is found.
755  */
756 /* ARGSUSED */
757 int
758 nfs_reply(myrep, procp)
759 	struct nfsreq *myrep;
760 	struct proc *procp;
761 {
762 	struct nfsreq *rep;
763 	struct nfsmount *nmp = myrep->r_nmp;
764 	int32_t t1;
765 	struct mbuf *mrep, *nam, *md;
766 	u_int32_t rxid, *tl;
767 	caddr_t dpos, cp2;
768 	int error;
769 
770 	/*
771 	 * Loop around until we get our own reply
772 	 */
773 	for (;;) {
774 		/*
775 		 * Lock against other receivers so that I don't get stuck in
776 		 * sbwait() after someone else has received my reply for me.
777 		 * Also necessary for connection based protocols to avoid
778 		 * race conditions during a reconnect.
779 		 */
780 		error = nfs_rcvlock(myrep);
781 		if (error == EALREADY)
782 			return (0);
783 		if (error)
784 			return (error);
785 		/*
786 		 * Get the next Rpc reply off the socket
787 		 */
788 		nmp->nm_waiters++;
789 		error = nfs_receive(myrep, &nam, &mrep, procp);
790 		nfs_rcvunlock(nmp);
791 		if (error) {
792 
793 			if (nmp->nm_iflag & NFSMNT_DISMNT) {
794 				/*
795 				 * Oops, we're going away now..
796 				 */
797 				nmp->nm_waiters--;
798 				wakeup (&nmp->nm_waiters);
799 				return error;
800 			}
801 			nmp->nm_waiters--;
802 			/*
803 			 * Ignore routing errors on connectionless protocols? ?
804 			 */
805 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
806 				nmp->nm_so->so_error = 0;
807 #ifdef DEBUG
808 				printf("nfs_reply: ignoring error %d\n", error);
809 #endif
810 				if (myrep->r_flags & R_GETONEREP)
811 					return (0);
812 				continue;
813 			}
814 			return (error);
815 		}
816 		nmp->nm_waiters--;
817 		if (nam)
818 			m_freem(nam);
819 
820 		/*
821 		 * Get the xid and check that it is an rpc reply
822 		 */
823 		md = mrep;
824 		dpos = mtod(md, caddr_t);
825 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
826 		rxid = *tl++;
827 		if (*tl != rpc_reply) {
828 #ifndef NFS_V2_ONLY
829 			if (nmp->nm_flag & NFSMNT_NQNFS) {
830 				if (nqnfs_callback(nmp, mrep, md, dpos))
831 					nfsstats.rpcinvalid++;
832 			} else
833 #endif
834 			{
835 				nfsstats.rpcinvalid++;
836 				m_freem(mrep);
837 			}
838 nfsmout:
839 			if (myrep->r_flags & R_GETONEREP)
840 				return (0);
841 			continue;
842 		}
843 
844 		/*
845 		 * Loop through the request list to match up the reply
846 		 * Iff no match, just drop the datagram
847 		 */
848 		TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
849 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
850 				/* Found it.. */
851 				rep->r_mrep = mrep;
852 				rep->r_md = md;
853 				rep->r_dpos = dpos;
854 				if (nfsrtton) {
855 					struct rttl *rt;
856 
857 					rt = &nfsrtt.rttl[nfsrtt.pos];
858 					rt->proc = rep->r_procnum;
859 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
860 					rt->sent = nmp->nm_sent;
861 					rt->cwnd = nmp->nm_cwnd;
862 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
863 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
864 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx;
865 					rt->tstamp = time;
866 					if (rep->r_flags & R_TIMING)
867 						rt->rtt = rep->r_rtt;
868 					else
869 						rt->rtt = 1000000;
870 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
871 				}
872 				/*
873 				 * Update congestion window.
874 				 * Do the additive increase of
875 				 * one rpc/rtt.
876 				 */
877 				if (nmp->nm_cwnd <= nmp->nm_sent) {
878 					nmp->nm_cwnd +=
879 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
880 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
881 					if (nmp->nm_cwnd > NFS_MAXCWND)
882 						nmp->nm_cwnd = NFS_MAXCWND;
883 				}
884 				rep->r_flags &= ~R_SENT;
885 				nmp->nm_sent -= NFS_CWNDSCALE;
886 				/*
887 				 * Update rtt using a gain of 0.125 on the mean
888 				 * and a gain of 0.25 on the deviation.
889 				 */
890 				if (rep->r_flags & R_TIMING) {
891 					/*
892 					 * Since the timer resolution of
893 					 * NFS_HZ is so course, it can often
894 					 * result in r_rtt == 0. Since
895 					 * r_rtt == N means that the actual
896 					 * rtt is between N+dt and N+2-dt ticks,
897 					 * add 1.
898 					 */
899 					t1 = rep->r_rtt + 1;
900 					t1 -= (NFS_SRTT(rep) >> 3);
901 					NFS_SRTT(rep) += t1;
902 					if (t1 < 0)
903 						t1 = -t1;
904 					t1 -= (NFS_SDRTT(rep) >> 2);
905 					NFS_SDRTT(rep) += t1;
906 				}
907 				nmp->nm_timeouts = 0;
908 				break;
909 			}
910 		}
911 		/*
912 		 * If not matched to a request, drop it.
913 		 * If it's mine, get out.
914 		 */
915 		if (rep == 0) {
916 			nfsstats.rpcunexpected++;
917 			m_freem(mrep);
918 		} else if (rep == myrep) {
919 			if (rep->r_mrep == NULL)
920 				panic("nfsreply nil");
921 			return (0);
922 		}
923 		if (myrep->r_flags & R_GETONEREP)
924 			return (0);
925 	}
926 }
927 
928 /*
929  * nfs_request - goes something like this
930  *	- fill in request struct
931  *	- links it into list
932  *	- calls nfs_send() for first transmit
933  *	- calls nfs_receive() to get reply
934  *	- break down rpc header and return with nfs reply pointed to
935  *	  by mrep or error
936  * nb: always frees up mreq mbuf list
937  */
938 int
939 nfs_request(np, mrest, procnum, procp, cred, mrp, mdp, dposp, rexmitp)
940 	struct nfsnode *np;
941 	struct mbuf *mrest;
942 	int procnum;
943 	struct proc *procp;
944 	struct ucred *cred;
945 	struct mbuf **mrp;
946 	struct mbuf **mdp;
947 	caddr_t *dposp;
948 	int *rexmitp;
949 {
950 	struct mbuf *m, *mrep;
951 	struct nfsreq *rep;
952 	u_int32_t *tl;
953 	int i;
954 	struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount);
955 	struct mbuf *md, *mheadend;
956 	char nickv[RPCX_NICKVERF];
957 	time_t reqtime, waituntil;
958 	caddr_t dpos, cp2;
959 	int t1, s, error = 0, mrest_len, auth_len, auth_type;
960 	int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0;
961 	int verf_len, verf_type;
962 	u_int32_t xid;
963 	char *auth_str, *verf_str;
964 	NFSKERBKEY_T key;		/* save session key */
965 	struct ucred acred;
966 #ifndef NFS_V2_ONLY
967 	int nqlflag, cachable;
968 	u_quad_t frev;
969 #endif
970 	struct mbuf *mrest_backup = NULL;
971 	struct ucred *origcred = NULL; /* XXX: gcc */
972 	boolean_t retry_cred = TRUE;
973 	boolean_t use_opencred = (np->n_flag & NUSEOPENCRED) != 0;
974 
975 	if (rexmitp != NULL)
976 		*rexmitp = 0;
977 
978 tryagain_cred:
979 	KASSERT(cred != NULL);
980 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
981 	rep->r_nmp = nmp;
982 	rep->r_procp = procp;
983 	rep->r_procnum = procnum;
984 	i = 0;
985 	m = mrest;
986 	while (m) {
987 		i += m->m_len;
988 		m = m->m_next;
989 	}
990 	mrest_len = i;
991 
992 	/*
993 	 * Get the RPC header with authorization.
994 	 */
995 kerbauth:
996 	verf_str = auth_str = (char *)0;
997 	if (nmp->nm_flag & NFSMNT_KERB) {
998 		verf_str = nickv;
999 		verf_len = sizeof (nickv);
1000 		auth_type = RPCAUTH_KERB4;
1001 		memset((caddr_t)key, 0, sizeof (key));
1002 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
1003 			&auth_len, verf_str, verf_len)) {
1004 			error = nfs_getauth(nmp, rep, cred, &auth_str,
1005 				&auth_len, verf_str, &verf_len, key);
1006 			if (error) {
1007 				free((caddr_t)rep, M_NFSREQ);
1008 				m_freem(mrest);
1009 				return (error);
1010 			}
1011 		}
1012 		retry_cred = FALSE;
1013 	} else {
1014 		/* AUTH_UNIX */
1015 		uid_t uid;
1016 		gid_t gid;
1017 
1018 		/*
1019 		 * on the most unix filesystems, permission checks are
1020 		 * done when the file is open(2)'ed.
1021 		 * ie. once a file is successfully open'ed,
1022 		 * following i/o operations never fail with EACCES.
1023 		 * we try to follow the semantics as far as possible.
1024 		 *
1025 		 * note that we expect that the nfs server always grant
1026 		 * accesses by the file's owner.
1027 		 */
1028 		origcred = cred;
1029 		switch (procnum) {
1030 		case NFSPROC_READ:
1031 		case NFSPROC_WRITE:
1032 		case NFSPROC_COMMIT:
1033 			uid = np->n_vattr->va_uid;
1034 			gid = np->n_vattr->va_gid;
1035 			if (cred->cr_uid == uid && cred->cr_gid == gid) {
1036 				retry_cred = FALSE;
1037 				break;
1038 			}
1039 			if (use_opencred)
1040 				break;
1041 			acred.cr_uid = uid;
1042 			acred.cr_gid = gid;
1043 			acred.cr_ngroups = 0;
1044 			acred.cr_ref = 2;	/* Just to be safe.. */
1045 			cred = &acred;
1046 			break;
1047 		default:
1048 			retry_cred = FALSE;
1049 			break;
1050 		}
1051 		/*
1052 		 * backup mbuf chain if we can need it later to retry.
1053 		 *
1054 		 * XXX maybe we can keep a direct reference to
1055 		 * mrest without doing m_copym, but it's ...ugly.
1056 		 */
1057 		if (retry_cred)
1058 			mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT);
1059 		auth_type = RPCAUTH_UNIX;
1060 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
1061 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
1062 			5 * NFSX_UNSIGNED;
1063 	}
1064 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
1065 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
1066 	if (auth_str)
1067 		free(auth_str, M_TEMP);
1068 
1069 	/*
1070 	 * For stream protocols, insert a Sun RPC Record Mark.
1071 	 */
1072 	if (nmp->nm_sotype == SOCK_STREAM) {
1073 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
1074 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
1075 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
1076 	}
1077 	rep->r_mreq = m;
1078 	rep->r_xid = xid;
1079 tryagain:
1080 	if (nmp->nm_flag & NFSMNT_SOFT)
1081 		rep->r_retry = nmp->nm_retry;
1082 	else
1083 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
1084 	rep->r_rtt = rep->r_rexmit = 0;
1085 	if (proct[procnum] > 0)
1086 		rep->r_flags = R_TIMING;
1087 	else
1088 		rep->r_flags = 0;
1089 	rep->r_mrep = NULL;
1090 
1091 	/*
1092 	 * Do the client side RPC.
1093 	 */
1094 	nfsstats.rpcrequests++;
1095 	/*
1096 	 * Chain request into list of outstanding requests. Be sure
1097 	 * to put it LAST so timer finds oldest requests first.
1098 	 */
1099 	s = splsoftnet();
1100 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1101 
1102 	/* Get send time for nqnfs */
1103 	reqtime = time.tv_sec;
1104 
1105 	/*
1106 	 * If backing off another request or avoiding congestion, don't
1107 	 * send this one now but let timer do it. If not timing a request,
1108 	 * do it now.
1109 	 */
1110 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1111 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1112 		nmp->nm_sent < nmp->nm_cwnd)) {
1113 		splx(s);
1114 		if (nmp->nm_soflags & PR_CONNREQUIRED)
1115 			error = nfs_sndlock(&nmp->nm_iflag, rep);
1116 		if (!error) {
1117 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
1118 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, procp);
1119 			if (nmp->nm_soflags & PR_CONNREQUIRED)
1120 				nfs_sndunlock(&nmp->nm_iflag);
1121 		}
1122 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1123 			nmp->nm_sent += NFS_CWNDSCALE;
1124 			rep->r_flags |= R_SENT;
1125 		}
1126 	} else {
1127 		splx(s);
1128 		rep->r_rtt = -1;
1129 	}
1130 
1131 	/*
1132 	 * Wait for the reply from our send or the timer's.
1133 	 */
1134 	if (!error || error == EPIPE)
1135 		error = nfs_reply(rep, procp);
1136 
1137 	/*
1138 	 * RPC done, unlink the request.
1139 	 */
1140 	s = splsoftnet();
1141 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1142 	splx(s);
1143 
1144 	/*
1145 	 * Decrement the outstanding request count.
1146 	 */
1147 	if (rep->r_flags & R_SENT) {
1148 		rep->r_flags &= ~R_SENT;	/* paranoia */
1149 		nmp->nm_sent -= NFS_CWNDSCALE;
1150 	}
1151 
1152 	if (rexmitp != NULL) {
1153 		int rexmit;
1154 
1155 		if (nmp->nm_sotype != SOCK_DGRAM)
1156 			rexmit = (rep->r_flags & R_REXMITTED) != 0;
1157 		else
1158 			rexmit = rep->r_rexmit;
1159 		*rexmitp = rexmit;
1160 	}
1161 
1162 	/*
1163 	 * If there was a successful reply and a tprintf msg.
1164 	 * tprintf a response.
1165 	 */
1166 	if (!error && (rep->r_flags & R_TPRINTFMSG))
1167 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1168 		    "is alive again");
1169 	mrep = rep->r_mrep;
1170 	md = rep->r_md;
1171 	dpos = rep->r_dpos;
1172 	if (error)
1173 		goto nfsmout;
1174 
1175 	/*
1176 	 * break down the rpc header and check if ok
1177 	 */
1178 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1179 	if (*tl++ == rpc_msgdenied) {
1180 		if (*tl == rpc_mismatch)
1181 			error = EOPNOTSUPP;
1182 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1183 			if (!failed_auth) {
1184 				failed_auth++;
1185 				mheadend->m_next = (struct mbuf *)0;
1186 				m_freem(mrep);
1187 				m_freem(rep->r_mreq);
1188 				goto kerbauth;
1189 			} else
1190 				error = EAUTH;
1191 		} else
1192 			error = EACCES;
1193 		m_freem(mrep);
1194 		goto nfsmout;
1195 	}
1196 
1197 	/*
1198 	 * Grab any Kerberos verifier, otherwise just throw it away.
1199 	 */
1200 	verf_type = fxdr_unsigned(int, *tl++);
1201 	i = fxdr_unsigned(int32_t, *tl);
1202 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1203 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1204 		if (error)
1205 			goto nfsmout;
1206 	} else if (i > 0)
1207 		nfsm_adv(nfsm_rndup(i));
1208 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1209 	/* 0 == ok */
1210 	if (*tl == 0) {
1211 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1212 		if (*tl != 0) {
1213 			error = fxdr_unsigned(int, *tl);
1214 			if (error == NFSERR_ACCES && retry_cred) {
1215 				m_freem(mrep);
1216 				m_freem(rep->r_mreq);
1217 				FREE(rep, M_NFSREQ);
1218 				use_opencred = !use_opencred;
1219 				if (mrest_backup == NULL)
1220 					return ENOMEM; /* m_copym failure */
1221 				mrest = mrest_backup;
1222 				mrest_backup = NULL;
1223 				cred = origcred;
1224 				error = 0;
1225 				retry_cred = FALSE;
1226 				goto tryagain_cred;
1227 			}
1228 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1229 				error == NFSERR_TRYLATER) {
1230 				m_freem(mrep);
1231 				error = 0;
1232 				waituntil = time.tv_sec + trylater_delay;
1233 				while (time.tv_sec < waituntil)
1234 					(void) tsleep((caddr_t)&lbolt,
1235 						PSOCK, "nqnfstry", 0);
1236 				trylater_delay *= NFS_TRYLATERDELMUL;
1237 				if (trylater_delay > NFS_TRYLATERDELMAX)
1238 					trylater_delay = NFS_TRYLATERDELMAX;
1239 				/*
1240 				 * RFC1813:
1241 				 * The client should wait and then try
1242 				 * the request with a new RPC transaction ID.
1243 				 */
1244 				nfs_renewxid(rep);
1245 				goto tryagain;
1246 			}
1247 
1248 			/*
1249 			 * If the File Handle was stale, invalidate the
1250 			 * lookup cache, just in case.
1251 			 */
1252 			if (error == ESTALE)
1253 				cache_purge(NFSTOV(np));
1254 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1255 				*mrp = mrep;
1256 				*mdp = md;
1257 				*dposp = dpos;
1258 				error |= NFSERR_RETERR;
1259 			} else
1260 				m_freem(mrep);
1261 			goto nfsmout;
1262 		}
1263 
1264 		/*
1265 		 * note which credential worked to minimize number of retries.
1266 		 */
1267 		if (use_opencred)
1268 			np->n_flag |= NUSEOPENCRED;
1269 		else
1270 			np->n_flag &= ~NUSEOPENCRED;
1271 
1272 #ifndef NFS_V2_ONLY
1273 		/*
1274 		 * For nqnfs, get any lease in reply
1275 		 */
1276 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1277 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1278 			if (*tl) {
1279 				nqlflag = fxdr_unsigned(int, *tl);
1280 				nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1281 				cachable = fxdr_unsigned(int, *tl++);
1282 				reqtime += fxdr_unsigned(int, *tl++);
1283 				if (reqtime > time.tv_sec) {
1284 				    frev = fxdr_hyper(tl);
1285 				    nqnfs_clientlease(nmp, np, nqlflag,
1286 					cachable, reqtime, frev);
1287 				}
1288 			}
1289 		}
1290 #endif
1291 		*mrp = mrep;
1292 		*mdp = md;
1293 		*dposp = dpos;
1294 
1295 		KASSERT(error == 0);
1296 		goto nfsmout;
1297 	}
1298 	m_freem(mrep);
1299 	error = EPROTONOSUPPORT;
1300 nfsmout:
1301 	m_freem(rep->r_mreq);
1302 	free((caddr_t)rep, M_NFSREQ);
1303 	m_freem(mrest_backup);
1304 	return (error);
1305 }
1306 #endif /* NFS */
1307 
1308 /*
1309  * Generate the rpc reply header
1310  * siz arg. is used to decide if adding a cluster is worthwhile
1311  */
1312 int
1313 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1314 	int siz;
1315 	struct nfsrv_descript *nd;
1316 	struct nfssvc_sock *slp;
1317 	int err;
1318 	int cache;
1319 	u_quad_t *frev;
1320 	struct mbuf **mrq;
1321 	struct mbuf **mbp;
1322 	caddr_t *bposp;
1323 {
1324 	u_int32_t *tl;
1325 	struct mbuf *mreq;
1326 	caddr_t bpos;
1327 	struct mbuf *mb;
1328 
1329 	mreq = m_gethdr(M_WAIT, MT_DATA);
1330 	MCLAIM(mreq, &nfs_mowner);
1331 	mb = mreq;
1332 	/*
1333 	 * If this is a big reply, use a cluster else
1334 	 * try and leave leading space for the lower level headers.
1335 	 */
1336 	siz += RPC_REPLYSIZ;
1337 	if (siz >= max_datalen) {
1338 		m_clget(mreq, M_WAIT);
1339 	} else
1340 		mreq->m_data += max_hdr;
1341 	tl = mtod(mreq, u_int32_t *);
1342 	mreq->m_len = 6 * NFSX_UNSIGNED;
1343 	bpos = ((caddr_t)tl) + mreq->m_len;
1344 	*tl++ = txdr_unsigned(nd->nd_retxid);
1345 	*tl++ = rpc_reply;
1346 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1347 		*tl++ = rpc_msgdenied;
1348 		if (err & NFSERR_AUTHERR) {
1349 			*tl++ = rpc_autherr;
1350 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1351 			mreq->m_len -= NFSX_UNSIGNED;
1352 			bpos -= NFSX_UNSIGNED;
1353 		} else {
1354 			*tl++ = rpc_mismatch;
1355 			*tl++ = txdr_unsigned(RPC_VER2);
1356 			*tl = txdr_unsigned(RPC_VER2);
1357 		}
1358 	} else {
1359 		*tl++ = rpc_msgaccepted;
1360 
1361 		/*
1362 		 * For Kerberos authentication, we must send the nickname
1363 		 * verifier back, otherwise just RPCAUTH_NULL.
1364 		 */
1365 		if (nd->nd_flag & ND_KERBFULL) {
1366 			struct nfsuid *nuidp;
1367 			struct timeval ktvin, ktvout;
1368 
1369 			LIST_FOREACH(nuidp, NUIDHASH(slp, nd->nd_cr.cr_uid),
1370 			    nu_hash) {
1371 				if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1372 				    (!nd->nd_nam2 || netaddr_match(
1373 				    NU_NETFAM(nuidp), &nuidp->nu_haddr,
1374 				    nd->nd_nam2)))
1375 					break;
1376 			}
1377 			if (nuidp) {
1378 				ktvin.tv_sec =
1379 				    txdr_unsigned(nuidp->nu_timestamp.tv_sec
1380 					- 1);
1381 				ktvin.tv_usec =
1382 				    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1383 
1384 				/*
1385 				 * Encrypt the timestamp in ecb mode using the
1386 				 * session key.
1387 				 */
1388 #ifdef NFSKERB
1389 				XXX
1390 #endif
1391 
1392 				*tl++ = rpc_auth_kerb;
1393 				*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1394 				*tl = ktvout.tv_sec;
1395 				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1396 				*tl++ = ktvout.tv_usec;
1397 				*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1398 			} else {
1399 				*tl++ = 0;
1400 				*tl++ = 0;
1401 			}
1402 		} else {
1403 			*tl++ = 0;
1404 			*tl++ = 0;
1405 		}
1406 		switch (err) {
1407 		case EPROGUNAVAIL:
1408 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1409 			break;
1410 		case EPROGMISMATCH:
1411 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1412 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1413 			if (nd->nd_flag & ND_NQNFS) {
1414 				*tl++ = txdr_unsigned(3);
1415 				*tl = txdr_unsigned(3);
1416 			} else {
1417 				*tl++ = txdr_unsigned(2);
1418 				*tl = txdr_unsigned(3);
1419 			}
1420 			break;
1421 		case EPROCUNAVAIL:
1422 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1423 			break;
1424 		case EBADRPC:
1425 			*tl = txdr_unsigned(RPC_GARBAGE);
1426 			break;
1427 		default:
1428 			*tl = 0;
1429 			if (err != NFSERR_RETVOID) {
1430 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1431 				if (err)
1432 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1433 				else
1434 				    *tl = 0;
1435 			}
1436 			break;
1437 		};
1438 	}
1439 
1440 	/*
1441 	 * For nqnfs, piggyback lease as requested.
1442 	 */
1443 	if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1444 		if (nd->nd_flag & ND_LEASE) {
1445 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1446 			*tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1447 			*tl++ = txdr_unsigned(cache);
1448 			*tl++ = txdr_unsigned(nd->nd_duration);
1449 			txdr_hyper(*frev, tl);
1450 		} else {
1451 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1452 			*tl = 0;
1453 		}
1454 	}
1455 	if (mrq != NULL)
1456 		*mrq = mreq;
1457 	*mbp = mb;
1458 	*bposp = bpos;
1459 	if (err != 0 && err != NFSERR_RETVOID)
1460 		nfsstats.srvrpc_errs++;
1461 	return (0);
1462 }
1463 
1464 /*
1465  * Nfs timer routine
1466  * Scan the nfsreq list and retranmit any requests that have timed out
1467  * To avoid retransmission attempts on STREAM sockets (in the future) make
1468  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1469  */
1470 void
1471 nfs_timer(arg)
1472 	void *arg;	/* never used */
1473 {
1474 	struct nfsreq *rep;
1475 	struct mbuf *m;
1476 	struct socket *so;
1477 	struct nfsmount *nmp;
1478 	int timeo;
1479 	int s, error;
1480 #ifdef NFSSERVER
1481 	struct nfssvc_sock *slp;
1482 	static long lasttime = 0;
1483 	u_quad_t cur_usec;
1484 #endif
1485 
1486 	s = splsoftnet();
1487 	TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
1488 		nmp = rep->r_nmp;
1489 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1490 			continue;
1491 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1492 			rep->r_flags |= R_SOFTTERM;
1493 			continue;
1494 		}
1495 		if (rep->r_rtt >= 0) {
1496 			rep->r_rtt++;
1497 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1498 				timeo = nmp->nm_timeo;
1499 			else
1500 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1501 			if (nmp->nm_timeouts > 0)
1502 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1503 			if (rep->r_rtt <= timeo)
1504 				continue;
1505 			if (nmp->nm_timeouts <
1506 			    (sizeof(nfs_backoff) / sizeof(nfs_backoff[0])))
1507 				nmp->nm_timeouts++;
1508 		}
1509 		/*
1510 		 * Check for server not responding
1511 		 */
1512 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1513 		     rep->r_rexmit > nmp->nm_deadthresh) {
1514 			nfs_msg(rep->r_procp,
1515 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1516 			    "not responding");
1517 			rep->r_flags |= R_TPRINTFMSG;
1518 		}
1519 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1520 			nfsstats.rpctimeouts++;
1521 			rep->r_flags |= R_SOFTTERM;
1522 			continue;
1523 		}
1524 		if (nmp->nm_sotype != SOCK_DGRAM) {
1525 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1526 				rep->r_rexmit = NFS_MAXREXMIT;
1527 			continue;
1528 		}
1529 		if ((so = nmp->nm_so) == NULL)
1530 			continue;
1531 
1532 		/*
1533 		 * If there is enough space and the window allows..
1534 		 *	Resend it
1535 		 * Set r_rtt to -1 in case we fail to send it now.
1536 		 */
1537 		rep->r_rtt = -1;
1538 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1539 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1540 		    (rep->r_flags & R_SENT) ||
1541 		    nmp->nm_sent < nmp->nm_cwnd) &&
1542 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1543 		        if (so->so_state & SS_ISCONNECTED)
1544 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1545 			    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1546 			else
1547 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1548 			    nmp->nm_nam, (struct mbuf *)0, (struct proc *)0);
1549 			if (error) {
1550 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1551 #ifdef DEBUG
1552 					printf("nfs_timer: ignoring error %d\n",
1553 						error);
1554 #endif
1555 					so->so_error = 0;
1556 				}
1557 			} else {
1558 				/*
1559 				 * Iff first send, start timing
1560 				 * else turn timing off, backoff timer
1561 				 * and divide congestion window by 2.
1562 				 */
1563 				if (rep->r_flags & R_SENT) {
1564 					rep->r_flags &= ~R_TIMING;
1565 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1566 						rep->r_rexmit = NFS_MAXREXMIT;
1567 					nmp->nm_cwnd >>= 1;
1568 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1569 						nmp->nm_cwnd = NFS_CWNDSCALE;
1570 					nfsstats.rpcretries++;
1571 				} else {
1572 					rep->r_flags |= R_SENT;
1573 					nmp->nm_sent += NFS_CWNDSCALE;
1574 				}
1575 				rep->r_rtt = 0;
1576 			}
1577 		}
1578 	}
1579 
1580 #ifdef NFSSERVER
1581 	/*
1582 	 * Call the nqnfs server timer once a second to handle leases.
1583 	 */
1584 	if (lasttime != time.tv_sec) {
1585 		lasttime = time.tv_sec;
1586 		nqnfs_serverd();
1587 	}
1588 
1589 	/*
1590 	 * Scan the write gathering queues for writes that need to be
1591 	 * completed now.
1592 	 */
1593 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1594 	TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
1595 	    if (LIST_FIRST(&slp->ns_tq) &&
1596 		LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec)
1597 		nfsrv_wakenfsd(slp);
1598 	}
1599 #endif /* NFSSERVER */
1600 	splx(s);
1601 	callout_schedule(&nfs_timer_ch, nfs_ticks);
1602 }
1603 
1604 /*ARGSUSED*/
1605 void
1606 nfs_exit(p, v)
1607 	struct proc *p;
1608 	void *v;
1609 {
1610 	struct nfsreq *rp;
1611 	int s = splsoftnet();
1612 
1613 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
1614 		if (rp->r_procp == p)
1615 			TAILQ_REMOVE(&nfs_reqq, rp, r_chain);
1616 	}
1617 	splx(s);
1618 }
1619 
1620 /*
1621  * Test for a termination condition pending on the process.
1622  * This is used for NFSMNT_INT mounts.
1623  */
1624 int
1625 nfs_sigintr(nmp, rep, p)
1626 	struct nfsmount *nmp;
1627 	struct nfsreq *rep;
1628 	struct proc *p;
1629 {
1630 	sigset_t ss;
1631 
1632 	if (rep && (rep->r_flags & R_SOFTTERM))
1633 		return (EINTR);
1634 	if (!(nmp->nm_flag & NFSMNT_INT))
1635 		return (0);
1636 	if (p) {
1637 		sigpending1(p, &ss);
1638 #if 0
1639 		sigminusset(&p->p_sigctx.ps_sigignore, &ss);
1640 #endif
1641 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1642 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1643 		    sigismember(&ss, SIGQUIT))
1644 			return (EINTR);
1645 	}
1646 	return (0);
1647 }
1648 
1649 /*
1650  * Lock a socket against others.
1651  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1652  * and also to avoid race conditions between the processes with nfs requests
1653  * in progress when a reconnect is necessary.
1654  */
1655 int
1656 nfs_sndlock(flagp, rep)
1657 	int *flagp;
1658 	struct nfsreq *rep;
1659 {
1660 	struct proc *p;
1661 	int slpflag = 0, slptimeo = 0;
1662 
1663 	if (rep) {
1664 		p = rep->r_procp;
1665 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1666 			slpflag = PCATCH;
1667 	} else
1668 		p = (struct proc *)0;
1669 	while (*flagp & NFSMNT_SNDLOCK) {
1670 		if (rep && nfs_sigintr(rep->r_nmp, rep, p))
1671 			return (EINTR);
1672 		*flagp |= NFSMNT_WANTSND;
1673 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1674 			slptimeo);
1675 		if (slpflag == PCATCH) {
1676 			slpflag = 0;
1677 			slptimeo = 2 * hz;
1678 		}
1679 	}
1680 	*flagp |= NFSMNT_SNDLOCK;
1681 	return (0);
1682 }
1683 
1684 /*
1685  * Unlock the stream socket for others.
1686  */
1687 void
1688 nfs_sndunlock(flagp)
1689 	int *flagp;
1690 {
1691 
1692 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1693 		panic("nfs sndunlock");
1694 	*flagp &= ~NFSMNT_SNDLOCK;
1695 	if (*flagp & NFSMNT_WANTSND) {
1696 		*flagp &= ~NFSMNT_WANTSND;
1697 		wakeup((caddr_t)flagp);
1698 	}
1699 }
1700 
1701 int
1702 nfs_rcvlock(rep)
1703 	struct nfsreq *rep;
1704 {
1705 	struct nfsmount *nmp = rep->r_nmp;
1706 	int *flagp = &nmp->nm_iflag;
1707 	int slpflag, slptimeo = 0;
1708 	int error = 0;
1709 
1710 	if (*flagp & NFSMNT_DISMNT)
1711 		return EIO;
1712 
1713 	if (*flagp & NFSMNT_INT)
1714 		slpflag = PCATCH;
1715 	else
1716 		slpflag = 0;
1717 	simple_lock(&nmp->nm_slock);
1718 	while (*flagp & NFSMNT_RCVLOCK) {
1719 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) {
1720 			error = EINTR;
1721 			goto quit;
1722 		}
1723 		*flagp |= NFSMNT_WANTRCV;
1724 		nmp->nm_waiters++;
1725 		(void) ltsleep(flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1726 			slptimeo, &nmp->nm_slock);
1727 		nmp->nm_waiters--;
1728 		if (*flagp & NFSMNT_DISMNT) {
1729 			wakeup(&nmp->nm_waiters);
1730 			error = EIO;
1731 			goto quit;
1732 		}
1733 		/* If our reply was received while we were sleeping,
1734 		 * then just return without taking the lock to avoid a
1735 		 * situation where a single iod could 'capture' the
1736 		 * receive lock.
1737 		 */
1738 		if (rep->r_mrep != NULL) {
1739 			error = EALREADY;
1740 			goto quit;
1741 		}
1742 		if (slpflag == PCATCH) {
1743 			slpflag = 0;
1744 			slptimeo = 2 * hz;
1745 		}
1746 	}
1747 	*flagp |= NFSMNT_RCVLOCK;
1748 quit:
1749 	simple_unlock(&nmp->nm_slock);
1750 	return error;
1751 }
1752 
1753 /*
1754  * Unlock the stream socket for others.
1755  */
1756 void
1757 nfs_rcvunlock(nmp)
1758 	struct nfsmount *nmp;
1759 {
1760 	int *flagp = &nmp->nm_iflag;
1761 
1762 	simple_lock(&nmp->nm_slock);
1763 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1764 		panic("nfs rcvunlock");
1765 	*flagp &= ~NFSMNT_RCVLOCK;
1766 	if (*flagp & NFSMNT_WANTRCV) {
1767 		*flagp &= ~NFSMNT_WANTRCV;
1768 		wakeup((caddr_t)flagp);
1769 	}
1770 	simple_unlock(&nmp->nm_slock);
1771 }
1772 
1773 /*
1774  * Parse an RPC request
1775  * - verify it
1776  * - fill in the cred struct.
1777  */
1778 int
1779 nfs_getreq(nd, nfsd, has_header)
1780 	struct nfsrv_descript *nd;
1781 	struct nfsd *nfsd;
1782 	int has_header;
1783 {
1784 	int len, i;
1785 	u_int32_t *tl;
1786 	int32_t t1;
1787 	struct uio uio;
1788 	struct iovec iov;
1789 	caddr_t dpos, cp2, cp;
1790 	u_int32_t nfsvers, auth_type;
1791 	uid_t nickuid;
1792 	int error = 0, nqnfs = 0, ticklen;
1793 	struct mbuf *mrep, *md;
1794 	struct nfsuid *nuidp;
1795 	struct timeval tvin, tvout;
1796 
1797 	mrep = nd->nd_mrep;
1798 	md = nd->nd_md;
1799 	dpos = nd->nd_dpos;
1800 	if (has_header) {
1801 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1802 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1803 		if (*tl++ != rpc_call) {
1804 			m_freem(mrep);
1805 			return (EBADRPC);
1806 		}
1807 	} else
1808 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1809 	nd->nd_repstat = 0;
1810 	nd->nd_flag = 0;
1811 	if (*tl++ != rpc_vers) {
1812 		nd->nd_repstat = ERPCMISMATCH;
1813 		nd->nd_procnum = NFSPROC_NOOP;
1814 		return (0);
1815 	}
1816 	if (*tl != nfs_prog) {
1817 		if (*tl == nqnfs_prog)
1818 			nqnfs++;
1819 		else {
1820 			nd->nd_repstat = EPROGUNAVAIL;
1821 			nd->nd_procnum = NFSPROC_NOOP;
1822 			return (0);
1823 		}
1824 	}
1825 	tl++;
1826 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1827 	if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1828 		(nfsvers != NQNFS_VER3 && nqnfs)) {
1829 		nd->nd_repstat = EPROGMISMATCH;
1830 		nd->nd_procnum = NFSPROC_NOOP;
1831 		return (0);
1832 	}
1833 	if (nqnfs)
1834 		nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1835 	else if (nfsvers == NFS_VER3)
1836 		nd->nd_flag = ND_NFSV3;
1837 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1838 	if (nd->nd_procnum == NFSPROC_NULL)
1839 		return (0);
1840 	if (nd->nd_procnum >= NFS_NPROCS ||
1841 		(!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1842 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1843 		nd->nd_repstat = EPROCUNAVAIL;
1844 		nd->nd_procnum = NFSPROC_NOOP;
1845 		return (0);
1846 	}
1847 	if ((nd->nd_flag & ND_NFSV3) == 0)
1848 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1849 	auth_type = *tl++;
1850 	len = fxdr_unsigned(int, *tl++);
1851 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1852 		m_freem(mrep);
1853 		return (EBADRPC);
1854 	}
1855 
1856 	nd->nd_flag &= ~ND_KERBAUTH;
1857 	/*
1858 	 * Handle auth_unix or auth_kerb.
1859 	 */
1860 	if (auth_type == rpc_auth_unix) {
1861 		len = fxdr_unsigned(int, *++tl);
1862 		if (len < 0 || len > NFS_MAXNAMLEN) {
1863 			m_freem(mrep);
1864 			return (EBADRPC);
1865 		}
1866 		nfsm_adv(nfsm_rndup(len));
1867 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1868 		memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred));
1869 		nd->nd_cr.cr_ref = 1;
1870 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1871 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1872 		len = fxdr_unsigned(int, *tl);
1873 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1874 			m_freem(mrep);
1875 			return (EBADRPC);
1876 		}
1877 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1878 		for (i = 0; i < len; i++)
1879 		    if (i < NGROUPS)
1880 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1881 		    else
1882 			tl++;
1883 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1884 		if (nd->nd_cr.cr_ngroups > 1)
1885 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1886 		len = fxdr_unsigned(int, *++tl);
1887 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1888 			m_freem(mrep);
1889 			return (EBADRPC);
1890 		}
1891 		if (len > 0)
1892 			nfsm_adv(nfsm_rndup(len));
1893 	} else if (auth_type == rpc_auth_kerb) {
1894 		switch (fxdr_unsigned(int, *tl++)) {
1895 		case RPCAKN_FULLNAME:
1896 			ticklen = fxdr_unsigned(int, *tl);
1897 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1898 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1899 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1900 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1901 				m_freem(mrep);
1902 				return (EBADRPC);
1903 			}
1904 			uio.uio_offset = 0;
1905 			uio.uio_iov = &iov;
1906 			uio.uio_iovcnt = 1;
1907 			uio.uio_segflg = UIO_SYSSPACE;
1908 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1909 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1910 			nfsm_mtouio(&uio, uio.uio_resid);
1911 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1912 			if (*tl++ != rpc_auth_kerb ||
1913 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1914 				printf("Bad kerb verifier\n");
1915 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1916 				nd->nd_procnum = NFSPROC_NOOP;
1917 				return (0);
1918 			}
1919 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1920 			tl = (u_int32_t *)cp;
1921 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1922 				printf("Not fullname kerb verifier\n");
1923 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1924 				nd->nd_procnum = NFSPROC_NOOP;
1925 				return (0);
1926 			}
1927 			cp += NFSX_UNSIGNED;
1928 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
1929 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1930 			nd->nd_flag |= ND_KERBFULL;
1931 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1932 			break;
1933 		case RPCAKN_NICKNAME:
1934 			if (len != 2 * NFSX_UNSIGNED) {
1935 				printf("Kerb nickname short\n");
1936 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1937 				nd->nd_procnum = NFSPROC_NOOP;
1938 				return (0);
1939 			}
1940 			nickuid = fxdr_unsigned(uid_t, *tl);
1941 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1942 			if (*tl++ != rpc_auth_kerb ||
1943 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1944 				printf("Kerb nick verifier bad\n");
1945 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1946 				nd->nd_procnum = NFSPROC_NOOP;
1947 				return (0);
1948 			}
1949 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1950 			tvin.tv_sec = *tl++;
1951 			tvin.tv_usec = *tl;
1952 
1953 			LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid),
1954 			    nu_hash) {
1955 				if (nuidp->nu_cr.cr_uid == nickuid &&
1956 				    (!nd->nd_nam2 ||
1957 				     netaddr_match(NU_NETFAM(nuidp),
1958 				      &nuidp->nu_haddr, nd->nd_nam2)))
1959 					break;
1960 			}
1961 			if (!nuidp) {
1962 				nd->nd_repstat =
1963 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1964 				nd->nd_procnum = NFSPROC_NOOP;
1965 				return (0);
1966 			}
1967 
1968 			/*
1969 			 * Now, decrypt the timestamp using the session key
1970 			 * and validate it.
1971 			 */
1972 #ifdef NFSKERB
1973 			XXX
1974 #endif
1975 
1976 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1977 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1978 			if (nuidp->nu_expire < time.tv_sec ||
1979 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1980 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1981 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1982 				nuidp->nu_expire = 0;
1983 				nd->nd_repstat =
1984 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1985 				nd->nd_procnum = NFSPROC_NOOP;
1986 				return (0);
1987 			}
1988 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1989 			nd->nd_flag |= ND_KERBNICK;
1990 		};
1991 	} else {
1992 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1993 		nd->nd_procnum = NFSPROC_NOOP;
1994 		return (0);
1995 	}
1996 
1997 	/*
1998 	 * For nqnfs, get piggybacked lease request.
1999 	 */
2000 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
2001 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2002 		nd->nd_flag |= fxdr_unsigned(int, *tl);
2003 		if (nd->nd_flag & ND_LEASE) {
2004 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2005 			nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
2006 		} else
2007 			nd->nd_duration = NQ_MINLEASE;
2008 	} else
2009 		nd->nd_duration = NQ_MINLEASE;
2010 	nd->nd_md = md;
2011 	nd->nd_dpos = dpos;
2012 	return (0);
2013 nfsmout:
2014 	return (error);
2015 }
2016 
2017 int
2018 nfs_msg(p, server, msg)
2019 	struct proc *p;
2020 	char *server, *msg;
2021 {
2022 	tpr_t tpr;
2023 
2024 	if (p)
2025 		tpr = tprintf_open(p);
2026 	else
2027 		tpr = NULL;
2028 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
2029 	tprintf_close(tpr);
2030 	return (0);
2031 }
2032 
2033 #ifdef NFSSERVER
2034 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
2035 				    struct nfssvc_sock *, struct proc *,
2036 				    struct mbuf **)) = {
2037 	nfsrv_null,
2038 	nfsrv_getattr,
2039 	nfsrv_setattr,
2040 	nfsrv_lookup,
2041 	nfsrv3_access,
2042 	nfsrv_readlink,
2043 	nfsrv_read,
2044 	nfsrv_write,
2045 	nfsrv_create,
2046 	nfsrv_mkdir,
2047 	nfsrv_symlink,
2048 	nfsrv_mknod,
2049 	nfsrv_remove,
2050 	nfsrv_rmdir,
2051 	nfsrv_rename,
2052 	nfsrv_link,
2053 	nfsrv_readdir,
2054 	nfsrv_readdirplus,
2055 	nfsrv_statfs,
2056 	nfsrv_fsinfo,
2057 	nfsrv_pathconf,
2058 	nfsrv_commit,
2059 	nqnfsrv_getlease,
2060 	nqnfsrv_vacated,
2061 	nfsrv_noop,
2062 	nfsrv_noop
2063 };
2064 
2065 /*
2066  * Socket upcall routine for the nfsd sockets.
2067  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
2068  * Essentially do as much as possible non-blocking, else punt and it will
2069  * be called with M_WAIT from an nfsd.
2070  */
2071 void
2072 nfsrv_rcv(so, arg, waitflag)
2073 	struct socket *so;
2074 	caddr_t arg;
2075 	int waitflag;
2076 {
2077 	struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
2078 	struct mbuf *m;
2079 	struct mbuf *mp, *nam;
2080 	struct uio auio;
2081 	int flags, error;
2082 
2083 	if ((slp->ns_flag & SLP_VALID) == 0)
2084 		return;
2085 #ifdef notdef
2086 	/*
2087 	 * Define this to test for nfsds handling this under heavy load.
2088 	 */
2089 	if (waitflag == M_DONTWAIT) {
2090 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
2091 	}
2092 #endif
2093 	/* XXX: was NULL, soreceive() requires non-NULL uio->uio_procp */
2094 	auio.uio_procp = curproc;	/* XXX curproc */
2095 	if (so->so_type == SOCK_STREAM) {
2096 		/*
2097 		 * If there are already records on the queue, defer soreceive()
2098 		 * to an nfsd so that there is feedback to the TCP layer that
2099 		 * the nfs servers are heavily loaded.
2100 		 */
2101 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
2102 			slp->ns_flag |= SLP_NEEDQ;
2103 			goto dorecs;
2104 		}
2105 
2106 		/*
2107 		 * Do soreceive().
2108 		 */
2109 		auio.uio_resid = 1000000000;
2110 		flags = MSG_DONTWAIT;
2111 		error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
2112 		if (error || mp == (struct mbuf *)0) {
2113 			if (error == EWOULDBLOCK)
2114 				slp->ns_flag |= SLP_NEEDQ;
2115 			else
2116 				slp->ns_flag |= SLP_DISCONN;
2117 			goto dorecs;
2118 		}
2119 		m = mp;
2120 		if (slp->ns_rawend) {
2121 			slp->ns_rawend->m_next = m;
2122 			slp->ns_cc += 1000000000 - auio.uio_resid;
2123 		} else {
2124 			slp->ns_raw = m;
2125 			slp->ns_cc = 1000000000 - auio.uio_resid;
2126 		}
2127 		while (m->m_next)
2128 			m = m->m_next;
2129 		slp->ns_rawend = m;
2130 
2131 		/*
2132 		 * Now try and parse record(s) out of the raw stream data.
2133 		 */
2134 		error = nfsrv_getstream(slp, waitflag);
2135 		if (error) {
2136 			if (error == EPERM)
2137 				slp->ns_flag |= SLP_DISCONN;
2138 			else
2139 				slp->ns_flag |= SLP_NEEDQ;
2140 		}
2141 	} else {
2142 		do {
2143 			auio.uio_resid = 1000000000;
2144 			flags = MSG_DONTWAIT;
2145 			error = (*so->so_receive)(so, &nam, &auio, &mp,
2146 						(struct mbuf **)0, &flags);
2147 			if (mp) {
2148 				if (nam) {
2149 					m = nam;
2150 					m->m_next = mp;
2151 				} else
2152 					m = mp;
2153 				if (slp->ns_recend)
2154 					slp->ns_recend->m_nextpkt = m;
2155 				else
2156 					slp->ns_rec = m;
2157 				slp->ns_recend = m;
2158 				m->m_nextpkt = (struct mbuf *)0;
2159 			}
2160 			if (error) {
2161 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2162 					&& error != EWOULDBLOCK) {
2163 					slp->ns_flag |= SLP_DISCONN;
2164 					goto dorecs;
2165 				}
2166 			}
2167 		} while (mp);
2168 	}
2169 
2170 	/*
2171 	 * Now try and process the request records, non-blocking.
2172 	 */
2173 dorecs:
2174 	if (waitflag == M_DONTWAIT &&
2175 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
2176 		nfsrv_wakenfsd(slp);
2177 }
2178 
2179 /*
2180  * Try and extract an RPC request from the mbuf data list received on a
2181  * stream socket. The "waitflag" argument indicates whether or not it
2182  * can sleep.
2183  */
2184 int
2185 nfsrv_getstream(slp, waitflag)
2186 	struct nfssvc_sock *slp;
2187 	int waitflag;
2188 {
2189 	struct mbuf *m, **mpp;
2190 	struct mbuf *recm;
2191 	u_int32_t recmark;
2192 
2193 	if (slp->ns_flag & SLP_GETSTREAM)
2194 		panic("nfs getstream");
2195 	slp->ns_flag |= SLP_GETSTREAM;
2196 	for (;;) {
2197 		if (slp->ns_reclen == 0) {
2198 			if (slp->ns_cc < NFSX_UNSIGNED) {
2199 				slp->ns_flag &= ~SLP_GETSTREAM;
2200 				return (0);
2201 			}
2202 			m = slp->ns_raw;
2203 			m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark);
2204 			m_adj(m, NFSX_UNSIGNED);
2205 			slp->ns_cc -= NFSX_UNSIGNED;
2206 			recmark = ntohl(recmark);
2207 			slp->ns_reclen = recmark & ~0x80000000;
2208 			if (recmark & 0x80000000)
2209 				slp->ns_flag |= SLP_LASTFRAG;
2210 			else
2211 				slp->ns_flag &= ~SLP_LASTFRAG;
2212 			if (slp->ns_reclen > NFS_MAXPACKET) {
2213 				slp->ns_flag &= ~SLP_GETSTREAM;
2214 				return (EPERM);
2215 			}
2216 		}
2217 
2218 		/*
2219 		 * Now get the record part.
2220 		 *
2221 		 * Note that slp->ns_reclen may be 0.  Linux sometimes
2222 		 * generates 0-length records.
2223 		 */
2224 		if (slp->ns_cc == slp->ns_reclen) {
2225 			recm = slp->ns_raw;
2226 			slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2227 			slp->ns_cc = slp->ns_reclen = 0;
2228 		} else if (slp->ns_cc > slp->ns_reclen) {
2229 			recm = slp->ns_raw;
2230 			m = m_split(recm, slp->ns_reclen, waitflag);
2231 			if (m == NULL) {
2232 				slp->ns_flag &= ~SLP_GETSTREAM;
2233 				return (EWOULDBLOCK);
2234 			}
2235 			m_claimm(recm, &nfs_mowner);
2236 			slp->ns_raw = m;
2237 			if (m->m_next == NULL)
2238 				slp->ns_rawend = m;
2239 			slp->ns_cc -= slp->ns_reclen;
2240 			slp->ns_reclen = 0;
2241 		} else {
2242 			slp->ns_flag &= ~SLP_GETSTREAM;
2243 			return (0);
2244 		}
2245 
2246 		/*
2247 		 * Accumulate the fragments into a record.
2248 		 */
2249 		mpp = &slp->ns_frag;
2250 		while (*mpp)
2251 			mpp = &((*mpp)->m_next);
2252 		*mpp = recm;
2253 		if (slp->ns_flag & SLP_LASTFRAG) {
2254 			if (slp->ns_recend)
2255 				slp->ns_recend->m_nextpkt = slp->ns_frag;
2256 			else
2257 				slp->ns_rec = slp->ns_frag;
2258 			slp->ns_recend = slp->ns_frag;
2259 			slp->ns_frag = (struct mbuf *)0;
2260 		}
2261 	}
2262 }
2263 
2264 /*
2265  * Parse an RPC header.
2266  */
2267 int
2268 nfsrv_dorec(slp, nfsd, ndp)
2269 	struct nfssvc_sock *slp;
2270 	struct nfsd *nfsd;
2271 	struct nfsrv_descript **ndp;
2272 {
2273 	struct mbuf *m, *nam;
2274 	struct nfsrv_descript *nd;
2275 	int error;
2276 
2277 	*ndp = NULL;
2278 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2279 	    (m = slp->ns_rec) == (struct mbuf *)0)
2280 		return (ENOBUFS);
2281 	slp->ns_rec = m->m_nextpkt;
2282 	if (slp->ns_rec)
2283 		m->m_nextpkt = (struct mbuf *)0;
2284 	else
2285 		slp->ns_recend = (struct mbuf *)0;
2286 	if (m->m_type == MT_SONAME) {
2287 		nam = m;
2288 		m = m->m_next;
2289 		nam->m_next = NULL;
2290 	} else
2291 		nam = NULL;
2292 	nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK);
2293 	nd->nd_md = nd->nd_mrep = m;
2294 	nd->nd_nam2 = nam;
2295 	nd->nd_dpos = mtod(m, caddr_t);
2296 	error = nfs_getreq(nd, nfsd, TRUE);
2297 	if (error) {
2298 		m_freem(nam);
2299 		pool_put(&nfs_srvdesc_pool, nd);
2300 		return (error);
2301 	}
2302 	*ndp = nd;
2303 	nfsd->nfsd_nd = nd;
2304 	return (0);
2305 }
2306 
2307 
2308 /*
2309  * Search for a sleeping nfsd and wake it up.
2310  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2311  * running nfsds will go look for the work in the nfssvc_sock list.
2312  */
2313 void
2314 nfsrv_wakenfsd(slp)
2315 	struct nfssvc_sock *slp;
2316 {
2317 	struct nfsd *nd;
2318 
2319 	if ((slp->ns_flag & SLP_VALID) == 0)
2320 		return;
2321 	simple_lock(&nfsd_slock);
2322 	if (slp->ns_flag & SLP_DOREC) {
2323 		simple_unlock(&nfsd_slock);
2324 		return;
2325 	}
2326 	nd = SLIST_FIRST(&nfsd_idle_head);
2327 	if (nd) {
2328 		SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle);
2329 		simple_unlock(&nfsd_slock);
2330 
2331 		KASSERT(nd->nfsd_flag & NFSD_WAITING);
2332 		nd->nfsd_flag &= ~NFSD_WAITING;
2333 		if (nd->nfsd_slp)
2334 			panic("nfsd wakeup");
2335 		slp->ns_sref++;
2336 		nd->nfsd_slp = slp;
2337 		wakeup(nd);
2338 		return;
2339 	}
2340 	slp->ns_flag |= SLP_DOREC;
2341 	nfsd_head_flag |= NFSD_CHECKSLP;
2342 	TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending);
2343 	simple_unlock(&nfsd_slock);
2344 }
2345 #endif /* NFSSERVER */
2346