xref: /openbsd/sys/nfs/nfs_socket.c (revision 610f49f8)
1 /*	$OpenBSD: nfs_socket.c,v 1.23 2002/01/16 21:51:16 ericj Exp $	*/
2 /*	$NetBSD: nfs_socket.c,v 1.27 1996/04/15 20:20:00 thorpej Exp $	*/
3 
4 /*
5  * Copyright (c) 1989, 1991, 1993, 1995
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * Rick Macklem at The University of Guelph.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the University of
22  *	California, Berkeley and its contributors.
23  * 4. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  *
39  *	@(#)nfs_socket.c	8.5 (Berkeley) 3/30/95
40  */
41 
42 /*
43  * Socket operations for use by nfs
44  */
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/kernel.h>
51 #include <sys/mbuf.h>
52 #include <sys/vnode.h>
53 #include <sys/domain.h>
54 #include <sys/protosw.h>
55 #include <sys/socket.h>
56 #include <sys/socketvar.h>
57 #include <sys/syslog.h>
58 #include <sys/tprintf.h>
59 #include <sys/namei.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/tcp.h>
63 
64 #include <nfs/rpcv2.h>
65 #include <nfs/nfsproto.h>
66 #include <nfs/nfs.h>
67 #include <nfs/xdr_subs.h>
68 #include <nfs/nfsm_subs.h>
69 #include <nfs/nfsmount.h>
70 #include <nfs/nfsnode.h>
71 #include <nfs/nfsrtt.h>
72 #include <nfs/nfs_var.h>
73 
74 #define	TRUE	1
75 #define	FALSE	0
76 
77 /*
78  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
79  * Use the mean and mean deviation of rtt for the appropriate type of rpc
80  * for the frequent rpcs and a default for the others.
81  * The justification for doing "other" this way is that these rpcs
82  * happen so infrequently that timer est. would probably be stale.
83  * Also, since many of these rpcs are
84  * non-idempotent, a conservative timeout is desired.
85  * getattr, lookup - A+2D
86  * read, write     - A+4D
87  * other           - nm_timeo
88  */
89 #define	NFS_RTO(n, t) \
90 	((t) == 0 ? (n)->nm_timeo : \
91 	 ((t) < 3 ? \
92 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
93 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
94 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
95 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
96 /*
97  * External data, mostly RPC constants in XDR form
98  */
99 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
100 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
101 	rpc_auth_kerb;
102 extern u_int32_t nfs_prog;
103 extern struct nfsstats nfsstats;
104 extern int nfsv3_procid[NFS_NPROCS];
105 extern int nfs_ticks;
106 
107 /*
108  * Defines which timer to use for the procnum.
109  * 0 - default
110  * 1 - getattr
111  * 2 - lookup
112  * 3 - read
113  * 4 - write
114  */
115 static int proct[NFS_NPROCS] = {
116 	0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
117 	0, 0, 0,
118 };
119 
120 /*
121  * There is a congestion window for outstanding rpcs maintained per mount
122  * point. The cwnd size is adjusted in roughly the way that:
123  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
124  * SIGCOMM '88". ACM, August 1988.
125  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
126  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
127  * of rpcs is in progress.
128  * (The sent count and cwnd are scaled for integer arith.)
129  * Variants of "slow start" were tried and were found to be too much of a
130  * performance hit (ave. rtt 3 times larger),
131  * I suspect due to the large rtt that nfs rpcs have.
132  */
133 #define	NFS_CWNDSCALE	256
134 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
135 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
136 int nfsrtton = 0;
137 struct nfsrtt nfsrtt;
138 
139 /*
140  * Initialize sockets and congestion for a new NFS connection.
141  * We do not free the sockaddr if error.
142  */
143 int
144 nfs_connect(nmp, rep)
145 	struct nfsmount *nmp;
146 	struct nfsreq *rep;
147 {
148 	struct socket *so;
149 	int s, error, rcvreserve, sndreserve;
150 	struct sockaddr *saddr;
151 	struct sockaddr_in *sin;
152 	struct mbuf *m;
153 
154 	nmp->nm_so = (struct socket *)0;
155 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
156 	error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
157 		nmp->nm_soproto);
158 	if (error)
159 		goto bad;
160 	so = nmp->nm_so;
161 	nmp->nm_soflags = so->so_proto->pr_flags;
162 
163 	/*
164 	 * Some servers require that the client port be a reserved port number.
165 	 * We always allocate a reserved port, as this prevents filehandle
166 	 * disclosure through UDP port capture.
167 	 */
168 	if (saddr->sa_family == AF_INET) {
169 		struct mbuf *mopt;
170 		int *ip;
171 
172 		MGET(mopt, M_WAIT, MT_SOOPTS);
173 		mopt->m_len = sizeof(int);
174 		ip = mtod(mopt, int *);
175 		*ip = IP_PORTRANGE_LOW;
176 		error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt);
177 		if (error)
178 			goto bad;
179 
180 		MGET(m, M_WAIT, MT_SONAME);
181 		sin = mtod(m, struct sockaddr_in *);
182 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
183 		sin->sin_family = AF_INET;
184 		sin->sin_addr.s_addr = INADDR_ANY;
185 		sin->sin_port = htons(0);
186 		error = sobind(so, m);
187 		m_freem(m);
188 		if (error)
189 			goto bad;
190 
191 		MGET(mopt, M_WAIT, MT_SOOPTS);
192 		mopt->m_len = sizeof(int);
193 		ip = mtod(mopt, int *);
194 		*ip = IP_PORTRANGE_DEFAULT;
195 		error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt);
196 		if (error)
197 			goto bad;
198 	}
199 
200 	/*
201 	 * Protocols that do not require connections may be optionally left
202 	 * unconnected for servers that reply from a port other than NFS_PORT.
203 	 */
204 	if (nmp->nm_flag & NFSMNT_NOCONN) {
205 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
206 			error = ENOTCONN;
207 			goto bad;
208 		}
209 	} else {
210 		error = soconnect(so, nmp->nm_nam);
211 		if (error)
212 			goto bad;
213 
214 		/*
215 		 * Wait for the connection to complete. Cribbed from the
216 		 * connect system call but with the wait timing out so
217 		 * that interruptible mounts don't hang here for a long time.
218 		 */
219 		s = splsoftnet();
220 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
221 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
222 				"nfscon", 2 * hz);
223 			if ((so->so_state & SS_ISCONNECTING) &&
224 			    so->so_error == 0 && rep &&
225 			    (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
226 				so->so_state &= ~SS_ISCONNECTING;
227 				splx(s);
228 				goto bad;
229 			}
230 		}
231 		if (so->so_error) {
232 			error = so->so_error;
233 			so->so_error = 0;
234 			splx(s);
235 			goto bad;
236 		}
237 		splx(s);
238 	}
239 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
240 		so->so_rcv.sb_timeo = (5 * hz);
241 		so->so_snd.sb_timeo = (5 * hz);
242 	} else {
243 		so->so_rcv.sb_timeo = 0;
244 		so->so_snd.sb_timeo = 0;
245 	}
246 	if (nmp->nm_sotype == SOCK_DGRAM) {
247 		sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR;
248 		rcvreserve = max(nmp->nm_rsize, nmp->nm_readdirsize) +
249 		    NFS_MAXPKTHDR;
250 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
251 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
252 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
253 		    NFS_MAXPKTHDR) * 2;
254 	} else {
255 		if (nmp->nm_sotype != SOCK_STREAM)
256 			panic("nfscon sotype");
257 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
258 			MGET(m, M_WAIT, MT_SOOPTS);
259 			*mtod(m, int32_t *) = 1;
260 			m->m_len = sizeof(int32_t);
261 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
262 		}
263 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
264 			MGET(m, M_WAIT, MT_SOOPTS);
265 			*mtod(m, int32_t *) = 1;
266 			m->m_len = sizeof(int32_t);
267 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
268 		}
269 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
270 		    sizeof (u_int32_t)) * 2;
271 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
272 		    sizeof (u_int32_t)) * 2;
273 	}
274 	error = soreserve(so, sndreserve, rcvreserve);
275 	if (error)
276 		goto bad;
277 	so->so_rcv.sb_flags |= SB_NOINTR;
278 	so->so_snd.sb_flags |= SB_NOINTR;
279 
280 	/* Initialize other non-zero congestion variables */
281 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
282 		nmp->nm_srtt[4] = (NFS_TIMEO << 3);
283 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
284 		nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0;
285 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
286 	nmp->nm_sent = 0;
287 	nmp->nm_timeouts = 0;
288 	return (0);
289 
290 bad:
291 	nfs_disconnect(nmp);
292 	return (error);
293 }
294 
295 /*
296  * Reconnect routine:
297  * Called when a connection is broken on a reliable protocol.
298  * - clean up the old socket
299  * - nfs_connect() again
300  * - set R_MUSTRESEND for all outstanding requests on mount point
301  * If this fails the mount point is DEAD!
302  * nb: Must be called with the nfs_sndlock() set on the mount point.
303  */
304 int
305 nfs_reconnect(rep)
306 	struct nfsreq *rep;
307 {
308 	struct nfsreq *rp;
309 	struct nfsmount *nmp = rep->r_nmp;
310 	int error;
311 
312 	nfs_disconnect(nmp);
313 	while ((error = nfs_connect(nmp, rep)) != 0) {
314 		if (error == EINTR || error == ERESTART)
315 			return (EINTR);
316 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0);
317 	}
318 
319 	/*
320 	 * Loop through outstanding request list and fix up all requests
321 	 * on old socket.
322 	 */
323 	for (rp = TAILQ_FIRST(&nfs_reqq); rp != NULL;
324 	    rp = TAILQ_NEXT(rp, r_chain)) {
325 		if (rp->r_nmp == nmp)
326 			rp->r_flags |= R_MUSTRESEND;
327 	}
328 	return (0);
329 }
330 
331 /*
332  * NFS disconnect. Clean up and unlink.
333  */
334 void
335 nfs_disconnect(nmp)
336 	struct nfsmount *nmp;
337 {
338 	struct socket *so;
339 
340 	if (nmp->nm_so) {
341 		so = nmp->nm_so;
342 		nmp->nm_so = (struct socket *)0;
343 		soshutdown(so, 2);
344 		soclose(so);
345 	}
346 }
347 
348 /*
349  * This is the nfs send routine. For connection based socket types, it
350  * must be called with an nfs_sndlock() on the socket.
351  * "rep == NULL" indicates that it has been called from a server.
352  * For the client side:
353  * - return EINTR if the RPC is terminated, 0 otherwise
354  * - set R_MUSTRESEND if the send fails for any reason
355  * - do any cleanup required by recoverable socket errors (???)
356  * For the server side:
357  * - return EINTR or ERESTART if interrupted by a signal
358  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
359  * - do any cleanup required by recoverable socket errors (???)
360  */
361 int
362 nfs_send(so, nam, top, rep)
363 	struct socket *so;
364 	struct mbuf *nam;
365 	struct mbuf *top;
366 	struct nfsreq *rep;
367 {
368 	struct mbuf *sendnam;
369 	int error, soflags, flags;
370 
371 	if (rep) {
372 		if (rep->r_flags & R_SOFTTERM) {
373 			m_freem(top);
374 			return (EINTR);
375 		}
376 		if ((so = rep->r_nmp->nm_so) == NULL) {
377 			rep->r_flags |= R_MUSTRESEND;
378 			m_freem(top);
379 			return (0);
380 		}
381 		rep->r_flags &= ~R_MUSTRESEND;
382 		soflags = rep->r_nmp->nm_soflags;
383 	} else
384 		soflags = so->so_proto->pr_flags;
385 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
386 		sendnam = (struct mbuf *)0;
387 	else
388 		sendnam = nam;
389 	if (so->so_type == SOCK_SEQPACKET)
390 		flags = MSG_EOR;
391 	else
392 		flags = 0;
393 
394 	error = sosend(so, sendnam, (struct uio *)0, top,
395 		(struct mbuf *)0, flags);
396 	if (error) {
397 		if (rep) {
398 			log(LOG_INFO, "nfs send error %d for server %s\n",error,
399 			    rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
400 			/*
401 			 * Deal with errors for the client side.
402 			 */
403 			if (rep->r_flags & R_SOFTTERM)
404 				error = EINTR;
405 			else
406 				rep->r_flags |= R_MUSTRESEND;
407 		} else
408 			log(LOG_INFO, "nfsd send error %d\n", error);
409 
410 		/*
411 		 * Handle any recoverable (soft) socket errors here. (???)
412 		 */
413 		if (error != EINTR && error != ERESTART &&
414 			error != EWOULDBLOCK && error != EPIPE)
415 			error = 0;
416 	}
417 	return (error);
418 }
419 
420 #ifdef NFSCLIENT
421 /*
422  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
423  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
424  * Mark and consolidate the data into a new mbuf list.
425  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
426  *     small mbufs.
427  * For SOCK_STREAM we must be very careful to read an entire record once
428  * we have read any of it, even if the system call has been interrupted.
429  */
430 int
431 nfs_receive(rep, aname, mp)
432 	struct nfsreq *rep;
433 	struct mbuf **aname;
434 	struct mbuf **mp;
435 {
436 	struct socket *so;
437 	struct uio auio;
438 	struct iovec aio;
439 	struct mbuf *m;
440 	struct mbuf *control;
441 	u_int32_t len;
442 	struct mbuf **getnam;
443 	int error, sotype, rcvflg;
444 	struct proc *p = curproc;	/* XXX */
445 
446 	/*
447 	 * Set up arguments for soreceive()
448 	 */
449 	*mp = (struct mbuf *)0;
450 	*aname = (struct mbuf *)0;
451 	sotype = rep->r_nmp->nm_sotype;
452 
453 	/*
454 	 * For reliable protocols, lock against other senders/receivers
455 	 * in case a reconnect is necessary.
456 	 * For SOCK_STREAM, first get the Record Mark to find out how much
457 	 * more there is to get.
458 	 * We must lock the socket against other receivers
459 	 * until we have an entire rpc request/reply.
460 	 */
461 	if (sotype != SOCK_DGRAM) {
462 		error = nfs_sndlock(&rep->r_nmp->nm_flag, rep);
463 		if (error)
464 			return (error);
465 tryagain:
466 		/*
467 		 * Check for fatal errors and resending request.
468 		 */
469 		/*
470 		 * Ugh: If a reconnect attempt just happened, nm_so
471 		 * would have changed. NULL indicates a failed
472 		 * attempt that has essentially shut down this
473 		 * mount point.
474 		 */
475 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
476 			nfs_sndunlock(&rep->r_nmp->nm_flag);
477 			return (EINTR);
478 		}
479 		so = rep->r_nmp->nm_so;
480 		if (!so) {
481 			error = nfs_reconnect(rep);
482 			if (error) {
483 				nfs_sndunlock(&rep->r_nmp->nm_flag);
484 				return (error);
485 			}
486 			goto tryagain;
487 		}
488 		while (rep->r_flags & R_MUSTRESEND) {
489 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
490 			nfsstats.rpcretries++;
491 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
492 			if (error) {
493 				if (error == EINTR || error == ERESTART ||
494 				    (error = nfs_reconnect(rep)) != 0) {
495 					nfs_sndunlock(&rep->r_nmp->nm_flag);
496 					return (error);
497 				}
498 				goto tryagain;
499 			}
500 		}
501 		nfs_sndunlock(&rep->r_nmp->nm_flag);
502 		if (sotype == SOCK_STREAM) {
503 			aio.iov_base = (caddr_t) &len;
504 			aio.iov_len = sizeof(u_int32_t);
505 			auio.uio_iov = &aio;
506 			auio.uio_iovcnt = 1;
507 			auio.uio_segflg = UIO_SYSSPACE;
508 			auio.uio_rw = UIO_READ;
509 			auio.uio_offset = 0;
510 			auio.uio_resid = sizeof(u_int32_t);
511 			auio.uio_procp = p;
512 			do {
513 			   rcvflg = MSG_WAITALL;
514 			   error = soreceive(so, (struct mbuf **)0, &auio,
515 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
516 			   if (error == EWOULDBLOCK && rep) {
517 				if (rep->r_flags & R_SOFTTERM)
518 					return (EINTR);
519 			   }
520 			} while (error == EWOULDBLOCK);
521 			if (!error && auio.uio_resid > 0) {
522 			    log(LOG_INFO,
523 				 "short receive (%d/%d) from nfs server %s\n",
524 				 sizeof(u_int32_t) - auio.uio_resid,
525 				 sizeof(u_int32_t),
526 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
527 			    error = EPIPE;
528 			}
529 			if (error)
530 				goto errout;
531 			len = ntohl(len) & ~0x80000000;
532 			/*
533 			 * This is SERIOUS! We are out of sync with the sender
534 			 * and forcing a disconnect/reconnect is all I can do.
535 			 */
536 			if (len > NFS_MAXPACKET) {
537 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
538 				"impossible packet length",
539 				len,
540 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
541 			    error = EFBIG;
542 			    goto errout;
543 			}
544 			auio.uio_resid = len;
545 			do {
546 			    rcvflg = MSG_WAITALL;
547 			    error =  soreceive(so, (struct mbuf **)0,
548 				&auio, mp, (struct mbuf **)0, &rcvflg);
549 			} while (error == EWOULDBLOCK || error == EINTR ||
550 				 error == ERESTART);
551 			if (!error && auio.uio_resid > 0) {
552 			    log(LOG_INFO,
553 				"short receive (%d/%d) from nfs server %s\n",
554 				len - auio.uio_resid, len,
555 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
556 			    error = EPIPE;
557 			}
558 		} else {
559 			/*
560 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
561 			 * and soreceive() will return when it has either a
562 			 * control msg or a data msg.
563 			 * We have no use for control msg., but must grab them
564 			 * and then throw them away so we know what is going
565 			 * on.
566 			 */
567 			auio.uio_resid = len = 100000000; /* Anything Big */
568 			auio.uio_procp = p;
569 			do {
570 			    rcvflg = 0;
571 			    error =  soreceive(so, (struct mbuf **)0,
572 				&auio, mp, &control, &rcvflg);
573 			    if (control)
574 				m_freem(control);
575 			    if (error == EWOULDBLOCK && rep) {
576 				if (rep->r_flags & R_SOFTTERM)
577 					return (EINTR);
578 			    }
579 			} while (error == EWOULDBLOCK ||
580 				 (!error && *mp == NULL && control));
581 			if ((rcvflg & MSG_EOR) == 0)
582 				printf("Egad!!\n");
583 			if (!error && *mp == NULL)
584 				error = EPIPE;
585 			len -= auio.uio_resid;
586 		}
587 errout:
588 		if (error && error != EINTR && error != ERESTART) {
589 			m_freem(*mp);
590 			*mp = (struct mbuf *)0;
591 			if (error != EPIPE)
592 				log(LOG_INFO,
593 				    "receive error %d from nfs server %s\n",
594 				    error,
595 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
596 			error = nfs_sndlock(&rep->r_nmp->nm_flag, rep);
597 			if (!error)
598 				error = nfs_reconnect(rep);
599 			if (!error)
600 				goto tryagain;
601 		}
602 	} else {
603 		if ((so = rep->r_nmp->nm_so) == NULL)
604 			return (EACCES);
605 		if (so->so_state & SS_ISCONNECTED)
606 			getnam = (struct mbuf **)0;
607 		else
608 			getnam = aname;
609 		auio.uio_resid = len = 1000000;
610 		auio.uio_procp = p;
611 		do {
612 			rcvflg = 0;
613 			error =  soreceive(so, getnam, &auio, mp,
614 				(struct mbuf **)0, &rcvflg);
615 			if (error == EWOULDBLOCK &&
616 			    (rep->r_flags & R_SOFTTERM))
617 				return (EINTR);
618 		} while (error == EWOULDBLOCK);
619 		len -= auio.uio_resid;
620 	}
621 	if (error) {
622 		m_freem(*mp);
623 		*mp = (struct mbuf *)0;
624 	}
625 	/*
626 	 * Search for any mbufs that are not a multiple of 4 bytes long
627 	 * or with m_data not longword aligned.
628 	 * These could cause pointer alignment problems, so copy them to
629 	 * well aligned mbufs.
630 	 */
631 	nfs_realign(*mp, 5 * NFSX_UNSIGNED);
632 	return (error);
633 }
634 
635 /*
636  * Implement receipt of reply on a socket.
637  * We must search through the list of received datagrams matching them
638  * with outstanding requests using the xid, until ours is found.
639  */
640 /* ARGSUSED */
641 int
642 nfs_reply(myrep)
643 	struct nfsreq *myrep;
644 {
645 	struct nfsreq *rep;
646 	struct nfsmount *nmp = myrep->r_nmp;
647 	int32_t t1;
648 	struct mbuf *mrep, *nam, *md;
649 	u_int32_t rxid, *tl;
650 	caddr_t dpos, cp2;
651 	int error;
652 
653 	/*
654 	 * Loop around until we get our own reply
655 	 */
656 	for (;;) {
657 		/*
658 		 * Lock against other receivers so that I don't get stuck in
659 		 * sbwait() after someone else has received my reply for me.
660 		 * Also necessary for connection based protocols to avoid
661 		 * race conditions during a reconnect.
662 		 */
663 		error = nfs_rcvlock(myrep);
664 		if (error)
665 			return (error);
666 		/* Already received, bye bye */
667 		if (myrep->r_mrep != NULL) {
668 			nfs_rcvunlock(&nmp->nm_flag);
669 			return (0);
670 		}
671 		/*
672 		 * Get the next Rpc reply off the socket
673 		 */
674 		error = nfs_receive(myrep, &nam, &mrep);
675 		nfs_rcvunlock(&nmp->nm_flag);
676 		if (error) {
677 
678 			/*
679 			 * Ignore routing errors on connectionless protocols??
680 			 */
681 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
682 				nmp->nm_so->so_error = 0;
683 				if (myrep->r_flags & R_GETONEREP)
684 					return (0);
685 				continue;
686 			}
687 			return (error);
688 		}
689 		if (nam)
690 			m_freem(nam);
691 
692 		/*
693 		 * Get the xid and check that it is an rpc reply
694 		 */
695 		md = mrep;
696 		dpos = mtod(md, caddr_t);
697 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
698 		rxid = *tl++;
699 		if (*tl != rpc_reply) {
700 			nfsstats.rpcinvalid++;
701 			m_freem(mrep);
702 nfsmout:
703 			if (myrep->r_flags & R_GETONEREP)
704 				return (0);
705 			continue;
706 		}
707 
708 		/*
709 		 * Loop through the request list to match up the reply
710 		 * Iff no match, just drop the datagram
711 		 */
712 		for (rep = TAILQ_FIRST(&nfs_reqq); rep != NULL;
713 		    rep = TAILQ_NEXT(rep, r_chain)) {
714 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
715 				/* Found it.. */
716 				rep->r_mrep = mrep;
717 				rep->r_md = md;
718 				rep->r_dpos = dpos;
719 				if (nfsrtton) {
720 					struct rttl *rt;
721 
722 					rt = &nfsrtt.rttl[nfsrtt.pos];
723 					rt->proc = rep->r_procnum;
724 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
725 					rt->sent = nmp->nm_sent;
726 					rt->cwnd = nmp->nm_cwnd;
727 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
728 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
729 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
730 					rt->tstamp = time;
731 					if (rep->r_flags & R_TIMING)
732 						rt->rtt = rep->r_rtt;
733 					else
734 						rt->rtt = 1000000;
735 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
736 				}
737 				/*
738 				 * Update congestion window.
739 				 * Do the additive increase of
740 				 * one rpc/rtt.
741 				 */
742 				if (nmp->nm_cwnd <= nmp->nm_sent) {
743 					nmp->nm_cwnd +=
744 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
745 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
746 					if (nmp->nm_cwnd > NFS_MAXCWND)
747 						nmp->nm_cwnd = NFS_MAXCWND;
748 				}
749 				rep->r_flags &= ~R_SENT;
750 				nmp->nm_sent -= NFS_CWNDSCALE;
751 				/*
752 				 * Update rtt using a gain of 0.125 on the mean
753 				 * and a gain of 0.25 on the deviation.
754 				 */
755 				if (rep->r_flags & R_TIMING) {
756 					/*
757 					 * Since the timer resolution of
758 					 * NFS_HZ is so course, it can often
759 					 * result in r_rtt == 0. Since
760 					 * r_rtt == N means that the actual
761 					 * rtt is between N+dt and N+2-dt ticks,
762 					 * add 1.
763 					 */
764 					t1 = rep->r_rtt + 1;
765 					t1 -= (NFS_SRTT(rep) >> 3);
766 					NFS_SRTT(rep) += t1;
767 					if (t1 < 0)
768 						t1 = -t1;
769 					t1 -= (NFS_SDRTT(rep) >> 2);
770 					NFS_SDRTT(rep) += t1;
771 				}
772 				nmp->nm_timeouts = 0;
773 				break;
774 			}
775 		}
776 		/*
777 		 * If not matched to a request, drop it.
778 		 * If it's mine, get out.
779 		 */
780 		if (rep == 0) {
781 			nfsstats.rpcunexpected++;
782 			m_freem(mrep);
783 		} else if (rep == myrep) {
784 			if (rep->r_mrep == NULL)
785 				panic("nfsreply nil");
786 			return (0);
787 		}
788 		if (myrep->r_flags & R_GETONEREP)
789 			return (0);
790 	}
791 }
792 
793 /*
794  * nfs_request - goes something like this
795  *	- fill in request struct
796  *	- links it into list
797  *	- calls nfs_send() for first transmit
798  *	- calls nfs_receive() to get reply
799  *	- break down rpc header and return with nfs reply pointed to
800  *	  by mrep or error
801  * nb: always frees up mreq mbuf list
802  */
803 int
804 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
805 	struct vnode *vp;
806 	struct mbuf *mrest;
807 	int procnum;
808 	struct proc *procp;
809 	struct ucred *cred;
810 	struct mbuf **mrp;
811 	struct mbuf **mdp;
812 	caddr_t *dposp;
813 {
814 	struct mbuf *m, *mrep;
815 	struct nfsreq *rep;
816 	u_int32_t *tl;
817 	int i;
818 	struct nfsmount *nmp;
819 	struct mbuf *md, *mheadend;
820 	char nickv[RPCX_NICKVERF];
821 	time_t reqtime, waituntil;
822 	caddr_t dpos, cp2;
823 	int t1, s, error = 0, mrest_len, auth_len, auth_type;
824 	int trylater_delay = 15, trylater_cnt = 0, failed_auth = 0;
825 	int verf_len, verf_type;
826 	u_int32_t xid;
827 	char *auth_str, *verf_str;
828 	NFSKERBKEY_T key;		/* save session key */
829 
830 	nmp = VFSTONFS(vp->v_mount);
831 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
832 	rep->r_nmp = nmp;
833 	rep->r_vp = vp;
834 	rep->r_procp = procp;
835 	rep->r_procnum = procnum;
836 	i = 0;
837 	m = mrest;
838 	while (m) {
839 		i += m->m_len;
840 		m = m->m_next;
841 	}
842 	mrest_len = i;
843 
844 	/*
845 	 * Get the RPC header with authorization.
846 	 */
847 kerbauth:
848 	verf_str = auth_str = (char *)0;
849 	if (nmp->nm_flag & NFSMNT_KERB) {
850 		verf_str = nickv;
851 		verf_len = sizeof (nickv);
852 		auth_type = RPCAUTH_KERB4;
853 		bzero((caddr_t)key, sizeof (key));
854 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
855 			&auth_len, verf_str, verf_len)) {
856 			error = nfs_getauth(nmp, rep, cred, &auth_str,
857 				&auth_len, verf_str, &verf_len, key);
858 			if (error) {
859 				free((caddr_t)rep, M_NFSREQ);
860 				m_freem(mrest);
861 				return (error);
862 			}
863 		}
864 	} else {
865 		auth_type = RPCAUTH_UNIX;
866 		auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
867 			nmp->nm_numgrps : cred->cr_ngroups) << 2) +
868 			5 * NFSX_UNSIGNED;
869 	}
870 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
871 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
872 	if (auth_str)
873 		free(auth_str, M_TEMP);
874 
875 	/*
876 	 * For stream protocols, insert a Sun RPC Record Mark.
877 	 */
878 	if (nmp->nm_sotype == SOCK_STREAM) {
879 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
880 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
881 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
882 	}
883 	rep->r_mreq = m;
884 	rep->r_xid = xid;
885 tryagain:
886 	if (nmp->nm_flag & NFSMNT_SOFT)
887 		rep->r_retry = nmp->nm_retry;
888 	else
889 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
890 	rep->r_rtt = rep->r_rexmit = 0;
891 	if (proct[procnum] > 0)
892 		rep->r_flags = R_TIMING;
893 	else
894 		rep->r_flags = 0;
895 	rep->r_mrep = NULL;
896 
897 	/*
898 	 * Do the client side RPC.
899 	 */
900 	nfsstats.rpcrequests++;
901 	/*
902 	 * Chain request into list of outstanding requests. Be sure
903 	 * to put it LAST so timer finds oldest requests first.
904 	 */
905 	s = splsoftclock();
906 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
907 
908 	/* Get send time for nqnfs */
909 	reqtime = time.tv_sec;
910 
911 	/*
912 	 * If backing off another request or avoiding congestion, don't
913 	 * send this one now but let timer do it. If not timing a request,
914 	 * do it now.
915 	 */
916 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
917 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
918 		nmp->nm_sent < nmp->nm_cwnd)) {
919 		splx(s);
920 		if (nmp->nm_soflags & PR_CONNREQUIRED)
921 			error = nfs_sndlock(&nmp->nm_flag, rep);
922 		if (!error) {
923 			error = nfs_send(nmp->nm_so, nmp->nm_nam,
924 					m_copym(m, 0, M_COPYALL, M_WAIT),
925 					rep);
926 			if (nmp->nm_soflags & PR_CONNREQUIRED)
927 				nfs_sndunlock(&nmp->nm_flag);
928 		}
929 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
930 			nmp->nm_sent += NFS_CWNDSCALE;
931 			rep->r_flags |= R_SENT;
932 		}
933 	} else {
934 		splx(s);
935 		rep->r_rtt = -1;
936 	}
937 
938 	/*
939 	 * Wait for the reply from our send or the timer's.
940 	 */
941 	if (!error || error == EPIPE)
942 		error = nfs_reply(rep);
943 
944 	/*
945 	 * RPC done, unlink the request.
946 	 */
947 	s = splsoftclock();
948 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
949 	splx(s);
950 
951 	/*
952 	 * Decrement the outstanding request count.
953 	 */
954 	if (rep->r_flags & R_SENT) {
955 		rep->r_flags &= ~R_SENT;	/* paranoia */
956 		nmp->nm_sent -= NFS_CWNDSCALE;
957 	}
958 
959 	/*
960 	 * If there was a successful reply and a tprintf msg.
961 	 * tprintf a response.
962 	 */
963 	if (!error && (rep->r_flags & R_TPRINTFMSG))
964 		nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
965 		    "is alive again");
966 	mrep = rep->r_mrep;
967 	md = rep->r_md;
968 	dpos = rep->r_dpos;
969 	if (error) {
970 		m_freem(rep->r_mreq);
971 		free((caddr_t)rep, M_NFSREQ);
972 		return (error);
973 	}
974 
975 	/*
976 	 * break down the rpc header and check if ok
977 	 */
978 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
979 	if (*tl++ == rpc_msgdenied) {
980 		if (*tl == rpc_mismatch)
981 			error = EOPNOTSUPP;
982 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
983 			if (!failed_auth) {
984 				failed_auth++;
985 				mheadend->m_next = (struct mbuf *)0;
986 				m_freem(mrep);
987 				m_freem(rep->r_mreq);
988 				goto kerbauth;
989 			} else
990 				error = EAUTH;
991 		} else
992 			error = EACCES;
993 		m_freem(mrep);
994 		m_freem(rep->r_mreq);
995 		free((caddr_t)rep, M_NFSREQ);
996 		return (error);
997 	}
998 
999 	/*
1000 	 * Grab any Kerberos verifier, otherwise just throw it away.
1001 	 */
1002 	verf_type = fxdr_unsigned(int, *tl++);
1003 	i = fxdr_unsigned(int32_t, *tl);
1004 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1005 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1006 		if (error)
1007 			goto nfsmout;
1008 	} else if (i > 0)
1009 		nfsm_adv(nfsm_rndup(i));
1010 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1011 	/* 0 == ok */
1012 	if (*tl == 0) {
1013 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1014 		if (*tl != 0) {
1015 			error = fxdr_unsigned(int, *tl);
1016 			if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1017 				error == NFSERR_TRYLATER) {
1018 				m_freem(mrep);
1019 				error = 0;
1020 				waituntil = time.tv_sec + trylater_delay;
1021 				while (time.tv_sec < waituntil)
1022 					(void) tsleep((caddr_t)&lbolt,
1023 						PSOCK, "nqnfstry", 0);
1024 				trylater_delay *= nfs_backoff[trylater_cnt];
1025 				if (trylater_cnt < 7)
1026 					trylater_cnt++;
1027 				goto tryagain;
1028 			}
1029 
1030 			/*
1031 			 * If the File Handle was stale, invalidate the
1032 			 * lookup cache, just in case.
1033 			 */
1034 			if (error == ESTALE)
1035 				cache_purge(vp);
1036 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1037 				*mrp = mrep;
1038 				*mdp = md;
1039 				*dposp = dpos;
1040 				error |= NFSERR_RETERR;
1041 			} else
1042 				m_freem(mrep);
1043 			m_freem(rep->r_mreq);
1044 			free((caddr_t)rep, M_NFSREQ);
1045 			return (error);
1046 		}
1047 
1048 		*mrp = mrep;
1049 		*mdp = md;
1050 		*dposp = dpos;
1051 		m_freem(rep->r_mreq);
1052 		FREE((caddr_t)rep, M_NFSREQ);
1053 		return (0);
1054 	}
1055 	m_freem(mrep);
1056 	error = EPROTONOSUPPORT;
1057 nfsmout:
1058 	m_freem(rep->r_mreq);
1059 	free((caddr_t)rep, M_NFSREQ);
1060 	return (error);
1061 }
1062 #endif /* NFSCLIENT */
1063 
1064 /*
1065  * Generate the rpc reply header
1066  * siz arg. is used to decide if adding a cluster is worthwhile
1067  */
1068 int
1069 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1070 	int siz;
1071 	struct nfsrv_descript *nd;
1072 	struct nfssvc_sock *slp;
1073 	int err;
1074 	int cache;
1075 	u_quad_t *frev;
1076 	struct mbuf **mrq;
1077 	struct mbuf **mbp;
1078 	caddr_t *bposp;
1079 {
1080 	u_int32_t *tl;
1081 	struct mbuf *mreq;
1082 	caddr_t bpos;
1083 	struct mbuf *mb, *mb2;
1084 
1085 	MGETHDR(mreq, M_WAIT, MT_DATA);
1086 	mb = mreq;
1087 	/*
1088 	 * If this is a big reply, use a cluster else
1089 	 * try and leave leading space for the lower level headers.
1090 	 */
1091 	siz += RPC_REPLYSIZ;
1092 	if (siz >= max_datalen) {
1093 		MCLGET(mreq, M_WAIT);
1094 	} else
1095 		mreq->m_data += max_hdr;
1096 	tl = mtod(mreq, u_int32_t *);
1097 	mreq->m_len = 6 * NFSX_UNSIGNED;
1098 	bpos = ((caddr_t)tl) + mreq->m_len;
1099 	*tl++ = txdr_unsigned(nd->nd_retxid);
1100 	*tl++ = rpc_reply;
1101 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1102 		*tl++ = rpc_msgdenied;
1103 		if (err & NFSERR_AUTHERR) {
1104 			*tl++ = rpc_autherr;
1105 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1106 			mreq->m_len -= NFSX_UNSIGNED;
1107 			bpos -= NFSX_UNSIGNED;
1108 		} else {
1109 			*tl++ = rpc_mismatch;
1110 			*tl++ = txdr_unsigned(RPC_VER2);
1111 			*tl = txdr_unsigned(RPC_VER2);
1112 		}
1113 	} else {
1114 		*tl++ = rpc_msgaccepted;
1115 
1116 		/*
1117 		 * For Kerberos authentication, we must send the nickname
1118 		 * verifier back, otherwise just RPCAUTH_NULL.
1119 		 */
1120 		if (nd->nd_flag & ND_KERBFULL) {
1121 		    struct nfsuid *nuidp;
1122 		    struct timeval ktvin, ktvout;
1123 
1124 		    for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1125 		 	nuidp != NULL; nuidp = LIST_NEXT(nuidp, nu_hash)) {
1126 			if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1127 			    (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1128 			     &nuidp->nu_haddr, nd->nd_nam2)))
1129 			    break;
1130 		    }
1131 		    if (nuidp) {
1132 			ktvin.tv_sec =
1133 			    txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1134 			ktvin.tv_usec =
1135 			    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1136 
1137 			/*
1138 			 * Encrypt the timestamp in ecb mode using the
1139 			 * session key.
1140 			 */
1141 #ifdef NFSKERB
1142 			XXX
1143 #endif
1144 
1145 			*tl++ = rpc_auth_kerb;
1146 			*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1147 			*tl = ktvout.tv_sec;
1148 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1149 			*tl++ = ktvout.tv_usec;
1150 			*tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1151 		    } else {
1152 			*tl++ = 0;
1153 			*tl++ = 0;
1154 		    }
1155 		} else {
1156 			*tl++ = 0;
1157 			*tl++ = 0;
1158 		}
1159 		switch (err) {
1160 		case EPROGUNAVAIL:
1161 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1162 			break;
1163 		case EPROGMISMATCH:
1164 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1165 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1166 			*tl++ = txdr_unsigned(2);
1167 			*tl = txdr_unsigned(3);
1168 			break;
1169 		case EPROCUNAVAIL:
1170 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1171 			break;
1172 		case EBADRPC:
1173 			*tl = txdr_unsigned(RPC_GARBAGE);
1174 			break;
1175 		default:
1176 			*tl = 0;
1177 			if (err != NFSERR_RETVOID) {
1178 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1179 				if (err)
1180 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1181 				else
1182 				    *tl = 0;
1183 			}
1184 			break;
1185 		};
1186 	}
1187 
1188 	*mrq = mreq;
1189 	if (mrq != NULL)
1190 		*mbp = mb;
1191 	*bposp = bpos;
1192 	if (err != 0 && err != NFSERR_RETVOID)
1193 		nfsstats.srvrpc_errs++;
1194 	return (0);
1195 }
1196 
1197 /*
1198  * Nfs timer routine
1199  * Scan the nfsreq list and retranmit any requests that have timed out
1200  * To avoid retransmission attempts on STREAM sockets (in the future) make
1201  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1202  */
1203 void
1204 nfs_timer(arg)
1205 	void *arg;
1206 {
1207 	struct timeout *to = (struct timeout *)arg;
1208 	struct nfsreq *rep;
1209 	struct mbuf *m;
1210 	struct socket *so;
1211 	struct nfsmount *nmp;
1212 	int timeo;
1213 	int s, error;
1214 #ifdef NFSSERVER
1215 	struct nfssvc_sock *slp;
1216 	u_quad_t cur_usec;
1217 #endif
1218 
1219 	s = splsoftnet();
1220 	for (rep = TAILQ_FIRST(&nfs_reqq); rep != NULL;
1221 	    rep = TAILQ_NEXT(rep, r_chain)) {
1222 		nmp = rep->r_nmp;
1223 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1224 			continue;
1225 		if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1226 			rep->r_flags |= R_SOFTTERM;
1227 			continue;
1228 		}
1229 		if (rep->r_rtt >= 0) {
1230 			rep->r_rtt++;
1231 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1232 				timeo = nmp->nm_timeo;
1233 			else
1234 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1235 			if (nmp->nm_timeouts > 0)
1236 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1237 			if (rep->r_rtt <= timeo)
1238 				continue;
1239 			if (nmp->nm_timeouts < 8)
1240 				nmp->nm_timeouts++;
1241 		}
1242 		/*
1243 		 * Check for server not responding
1244 		 */
1245 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1246 		     rep->r_rexmit > nmp->nm_deadthresh) {
1247 			nfs_msg(rep->r_procp,
1248 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1249 			    "not responding");
1250 			rep->r_flags |= R_TPRINTFMSG;
1251 		}
1252 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1253 			nfsstats.rpctimeouts++;
1254 			rep->r_flags |= R_SOFTTERM;
1255 			continue;
1256 		}
1257 		if (nmp->nm_sotype != SOCK_DGRAM) {
1258 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1259 				rep->r_rexmit = NFS_MAXREXMIT;
1260 			continue;
1261 		}
1262 		if ((so = nmp->nm_so) == NULL)
1263 			continue;
1264 
1265 		/*
1266 		 * If there is enough space and the window allows..
1267 		 *	Resend it
1268 		 * Set r_rtt to -1 in case we fail to send it now.
1269 		 */
1270 		rep->r_rtt = -1;
1271 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1272 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1273 		    (rep->r_flags & R_SENT) ||
1274 		    nmp->nm_sent < nmp->nm_cwnd) &&
1275 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1276 			if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
1277 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1278 			    (struct mbuf *)0, (struct mbuf *)0);
1279 			else
1280 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1281 			    nmp->nm_nam, (struct mbuf *)0);
1282 			if (error) {
1283 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
1284 					so->so_error = 0;
1285 			} else {
1286 				/*
1287 				 * Iff first send, start timing
1288 				 * else turn timing off, backoff timer
1289 				 * and divide congestion window by 2.
1290 				 */
1291 				if (rep->r_flags & R_SENT) {
1292 					rep->r_flags &= ~R_TIMING;
1293 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1294 						rep->r_rexmit = NFS_MAXREXMIT;
1295 					nmp->nm_cwnd >>= 1;
1296 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1297 						nmp->nm_cwnd = NFS_CWNDSCALE;
1298 					nfsstats.rpcretries++;
1299 				} else {
1300 					rep->r_flags |= R_SENT;
1301 					nmp->nm_sent += NFS_CWNDSCALE;
1302 				}
1303 				rep->r_rtt = 0;
1304 			}
1305 		}
1306 	}
1307 
1308 #ifdef NFSSERVER
1309 	/*
1310 	 * Scan the write gathering queues for writes that need to be
1311 	 * completed now.
1312 	 */
1313 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1314 	for (slp = TAILQ_FIRST(&nfssvc_sockhead); slp != NULL;
1315 	    slp = TAILQ_NEXT(slp, ns_chain)) {
1316 	    if (LIST_FIRST(&slp->ns_tq) &&
1317 	        LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec)
1318 		nfsrv_wakenfsd(slp);
1319 	}
1320 #endif /* NFSSERVER */
1321 	splx(s);
1322 	timeout_add(to, nfs_ticks);
1323 }
1324 
1325 /*
1326  * Test for a termination condition pending on the process.
1327  * This is used for NFSMNT_INT mounts.
1328  */
1329 int
1330 nfs_sigintr(nmp, rep, p)
1331 	struct nfsmount *nmp;
1332 	struct nfsreq *rep;
1333 	struct proc *p;
1334 {
1335 
1336 	if (rep && (rep->r_flags & R_SOFTTERM))
1337 		return (EINTR);
1338 	if (!(nmp->nm_flag & NFSMNT_INT))
1339 		return (0);
1340 	if (p && p->p_siglist &&
1341 	    (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) &
1342 	    NFSINT_SIGMASK))
1343 		return (EINTR);
1344 	return (0);
1345 }
1346 
1347 /*
1348  * Lock a socket against others.
1349  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1350  * and also to avoid race conditions between the processes with nfs requests
1351  * in progress when a reconnect is necessary.
1352  */
1353 int
1354 nfs_sndlock(flagp, rep)
1355 	int *flagp;
1356 	struct nfsreq *rep;
1357 {
1358 	struct proc *p;
1359 	int slpflag = 0, slptimeo = 0;
1360 
1361 	if (rep) {
1362 		p = rep->r_procp;
1363 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1364 			slpflag = PCATCH;
1365 	} else
1366 		p = (struct proc *)0;
1367 	while (*flagp & NFSMNT_SNDLOCK) {
1368 		if (nfs_sigintr(rep->r_nmp, rep, p))
1369 			return (EINTR);
1370 		*flagp |= NFSMNT_WANTSND;
1371 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1372 			slptimeo);
1373 		if (slpflag == PCATCH) {
1374 			slpflag = 0;
1375 			slptimeo = 2 * hz;
1376 		}
1377 	}
1378 	*flagp |= NFSMNT_SNDLOCK;
1379 	return (0);
1380 }
1381 
1382 /*
1383  * Unlock the stream socket for others.
1384  */
1385 void
1386 nfs_sndunlock(flagp)
1387 	int *flagp;
1388 {
1389 
1390 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1391 		panic("nfs sndunlock");
1392 	*flagp &= ~NFSMNT_SNDLOCK;
1393 	if (*flagp & NFSMNT_WANTSND) {
1394 		*flagp &= ~NFSMNT_WANTSND;
1395 		wakeup((caddr_t)flagp);
1396 	}
1397 }
1398 
1399 int
1400 nfs_rcvlock(rep)
1401 	struct nfsreq *rep;
1402 {
1403 	int *flagp = &rep->r_nmp->nm_flag;
1404 	int slpflag, slptimeo = 0;
1405 
1406 	if (*flagp & NFSMNT_INT)
1407 		slpflag = PCATCH;
1408 	else
1409 		slpflag = 0;
1410 	while (*flagp & NFSMNT_RCVLOCK) {
1411 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1412 			return (EINTR);
1413 		*flagp |= NFSMNT_WANTRCV;
1414 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1415 			slptimeo);
1416 		if (slpflag == PCATCH) {
1417 			slpflag = 0;
1418 			slptimeo = 2 * hz;
1419 		}
1420 	}
1421 	*flagp |= NFSMNT_RCVLOCK;
1422 	return (0);
1423 }
1424 
1425 /*
1426  * Unlock the stream socket for others.
1427  */
1428 void
1429 nfs_rcvunlock(flagp)
1430 	int *flagp;
1431 {
1432 
1433 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1434 		panic("nfs rcvunlock");
1435 	*flagp &= ~NFSMNT_RCVLOCK;
1436 	if (*flagp & NFSMNT_WANTRCV) {
1437 		*flagp &= ~NFSMNT_WANTRCV;
1438 		wakeup((caddr_t)flagp);
1439 	}
1440 }
1441 
1442 /*
1443  * Check for badly aligned mbuf data areas and
1444  * realign data in an mbuf list by copying the data areas up, as required.
1445  */
1446 void
1447 nfs_realign(m, hsiz)
1448 	struct mbuf *m;
1449 	int hsiz;
1450 {
1451 	struct mbuf *m2;
1452 	int siz, mlen, olen;
1453 	caddr_t tcp, fcp;
1454 	struct mbuf *mnew;
1455 
1456 	while (m) {
1457 	    /*
1458 	     * This never happens for UDP, rarely happens for TCP
1459 	     * but frequently happens for iso transport.
1460 	     */
1461 	    if ((m->m_len & 0x3) || (mtod(m, long) & 0x3)) {
1462 		olen = m->m_len;
1463 		fcp = mtod(m, caddr_t);
1464 		if ((long)fcp & 0x3) {
1465 		        if (m->m_flags & M_PKTHDR)
1466 				m_tag_delete_chain(m, NULL);
1467 			m->m_flags &= ~M_PKTHDR;
1468 			if (m->m_flags & M_EXT)
1469 				m->m_data = m->m_ext.ext_buf +
1470 					((m->m_ext.ext_size - olen) & ~0x3);
1471 			else
1472 				m->m_data = m->m_dat;
1473 		}
1474 		m->m_len = 0;
1475 		tcp = mtod(m, caddr_t);
1476 		mnew = m;
1477 		m2 = m->m_next;
1478 
1479 		/*
1480 		 * If possible, only put the first invariant part
1481 		 * of the RPC header in the first mbuf.
1482 		 */
1483 		mlen = M_TRAILINGSPACE(m);
1484 		if (olen <= hsiz && mlen > hsiz)
1485 			mlen = hsiz;
1486 
1487 		/* Loop through the mbuf list consolidating data. */
1488 		while (m) {
1489 			while (olen > 0) {
1490 				if (mlen == 0) {
1491 				        if (m2->m_flags & M_PKTHDR)
1492 						m_tag_delete_chain(m2, NULL);
1493 					m2->m_flags &= ~M_PKTHDR;
1494 					if (m2->m_flags & M_EXT)
1495 						m2->m_data = m2->m_ext.ext_buf;
1496 					else
1497 						m2->m_data = m2->m_dat;
1498 					m2->m_len = 0;
1499 					mlen = M_TRAILINGSPACE(m2);
1500 					tcp = mtod(m2, caddr_t);
1501 					mnew = m2;
1502 					m2 = m2->m_next;
1503 				}
1504 				siz = min(mlen, olen);
1505 				if (tcp != fcp)
1506 					bcopy(fcp, tcp, siz);
1507 				mnew->m_len += siz;
1508 				mlen -= siz;
1509 				olen -= siz;
1510 				tcp += siz;
1511 				fcp += siz;
1512 			}
1513 			m = m->m_next;
1514 			if (m) {
1515 				olen = m->m_len;
1516 				fcp = mtod(m, caddr_t);
1517 			}
1518 		}
1519 
1520 		/*
1521 		 * Finally, set m_len == 0 for any trailing mbufs that have
1522 		 * been copied out of.
1523 		 */
1524 		while (m2) {
1525 			m2->m_len = 0;
1526 			m2 = m2->m_next;
1527 		}
1528 		return;
1529 	    }
1530 	    m = m->m_next;
1531 	}
1532 }
1533 
1534 /*
1535  * Parse an RPC request
1536  * - verify it
1537  * - fill in the cred struct.
1538  */
1539 int
1540 nfs_getreq(nd, nfsd, has_header)
1541 	struct nfsrv_descript *nd;
1542 	struct nfsd *nfsd;
1543 	int has_header;
1544 {
1545 	int len, i;
1546 	u_int32_t *tl;
1547 	int32_t t1;
1548 	struct uio uio;
1549 	struct iovec iov;
1550 	caddr_t dpos, cp2, cp;
1551 	u_int32_t nfsvers, auth_type;
1552 	uid_t nickuid;
1553 	int error = 0, ticklen;
1554 	struct mbuf *mrep, *md;
1555 	struct nfsuid *nuidp;
1556 	struct timeval tvin, tvout;
1557 
1558 	mrep = nd->nd_mrep;
1559 	md = nd->nd_md;
1560 	dpos = nd->nd_dpos;
1561 	if (has_header) {
1562 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1563 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1564 		if (*tl++ != rpc_call) {
1565 			m_freem(mrep);
1566 			return (EBADRPC);
1567 		}
1568 	} else
1569 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1570 	nd->nd_repstat = 0;
1571 	nd->nd_flag = 0;
1572 	if (*tl++ != rpc_vers) {
1573 		nd->nd_repstat = ERPCMISMATCH;
1574 		nd->nd_procnum = NFSPROC_NOOP;
1575 		return (0);
1576 	}
1577 	if (*tl != nfs_prog) {
1578 		nd->nd_repstat = EPROGUNAVAIL;
1579 		nd->nd_procnum = NFSPROC_NOOP;
1580 		return (0);
1581 	}
1582 	tl++;
1583 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1584 	if (nfsvers != NFS_VER2 && nfsvers != NFS_VER3) {
1585 		nd->nd_repstat = EPROGMISMATCH;
1586 		nd->nd_procnum = NFSPROC_NOOP;
1587 		return (0);
1588 	}
1589 	if (nfsvers == NFS_VER3)
1590 		nd->nd_flag = ND_NFSV3;
1591 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1592 	if (nd->nd_procnum == NFSPROC_NULL)
1593 		return (0);
1594 	if (nd->nd_procnum >= NFS_NPROCS ||
1595 		(nd->nd_procnum > NFSPROC_COMMIT) ||
1596 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1597 		nd->nd_repstat = EPROCUNAVAIL;
1598 		nd->nd_procnum = NFSPROC_NOOP;
1599 		return (0);
1600 	}
1601 	if ((nd->nd_flag & ND_NFSV3) == 0)
1602 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1603 	auth_type = *tl++;
1604 	len = fxdr_unsigned(int, *tl++);
1605 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1606 		m_freem(mrep);
1607 		return (EBADRPC);
1608 	}
1609 
1610 	nd->nd_flag &= ~ND_KERBAUTH;
1611 	/*
1612 	 * Handle auth_unix or auth_kerb.
1613 	 */
1614 	if (auth_type == rpc_auth_unix) {
1615 		len = fxdr_unsigned(int, *++tl);
1616 		if (len < 0 || len > NFS_MAXNAMLEN) {
1617 			m_freem(mrep);
1618 			return (EBADRPC);
1619 		}
1620 		nfsm_adv(nfsm_rndup(len));
1621 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1622 		bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred));
1623 		nd->nd_cr.cr_ref = 1;
1624 		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1625 		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1626 		len = fxdr_unsigned(int, *tl);
1627 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1628 			m_freem(mrep);
1629 			return (EBADRPC);
1630 		}
1631 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1632 		for (i = 0; i < len; i++)
1633 		    if (i < NGROUPS)
1634 			nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1635 		    else
1636 			tl++;
1637 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1638 		if (nd->nd_cr.cr_ngroups > 1)
1639 		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1640 		len = fxdr_unsigned(int, *++tl);
1641 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1642 			m_freem(mrep);
1643 			return (EBADRPC);
1644 		}
1645 		if (len > 0)
1646 			nfsm_adv(nfsm_rndup(len));
1647 	} else if (auth_type == rpc_auth_kerb) {
1648 		switch (fxdr_unsigned(int, *tl++)) {
1649 		case RPCAKN_FULLNAME:
1650 			ticklen = fxdr_unsigned(int, *tl);
1651 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1652 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1653 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1654 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1655 				m_freem(mrep);
1656 				return (EBADRPC);
1657 			}
1658 			uio.uio_offset = 0;
1659 			uio.uio_iov = &iov;
1660 			uio.uio_iovcnt = 1;
1661 			uio.uio_segflg = UIO_SYSSPACE;
1662 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1663 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1664 			nfsm_mtouio(&uio, uio.uio_resid);
1665 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1666 			if (*tl++ != rpc_auth_kerb ||
1667 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1668 				printf("Bad kerb verifier\n");
1669 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1670 				nd->nd_procnum = NFSPROC_NOOP;
1671 				return (0);
1672 			}
1673 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1674 			tl = (u_int32_t *)cp;
1675 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1676 				printf("Not fullname kerb verifier\n");
1677 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1678 				nd->nd_procnum = NFSPROC_NOOP;
1679 				return (0);
1680 			}
1681 			cp += NFSX_UNSIGNED;
1682 			bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED);
1683 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1684 			nd->nd_flag |= ND_KERBFULL;
1685 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1686 			break;
1687 		case RPCAKN_NICKNAME:
1688 			if (len != 2 * NFSX_UNSIGNED) {
1689 				printf("Kerb nickname short\n");
1690 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1691 				nd->nd_procnum = NFSPROC_NOOP;
1692 				return (0);
1693 			}
1694 			nickuid = fxdr_unsigned(uid_t, *tl);
1695 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1696 			if (*tl++ != rpc_auth_kerb ||
1697 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1698 				printf("Kerb nick verifier bad\n");
1699 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1700 				nd->nd_procnum = NFSPROC_NOOP;
1701 				return (0);
1702 			}
1703 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1704 			tvin.tv_sec = *tl++;
1705 			tvin.tv_usec = *tl;
1706 
1707 			for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1708 			    nuidp != NULL; nuidp = LIST_NEXT(nuidp, nu_hash)) {
1709 				if (nuidp->nu_cr.cr_uid == nickuid &&
1710 				    (!nd->nd_nam2 ||
1711 				     netaddr_match(NU_NETFAM(nuidp),
1712 				      &nuidp->nu_haddr, nd->nd_nam2)))
1713 					break;
1714 			}
1715 			if (!nuidp) {
1716 				nd->nd_repstat =
1717 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1718 				nd->nd_procnum = NFSPROC_NOOP;
1719 				return (0);
1720 			}
1721 
1722 			/*
1723 			 * Now, decrypt the timestamp using the session key
1724 			 * and validate it.
1725 			 */
1726 #ifdef NFSKERB
1727 			XXX
1728 #endif
1729 
1730 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1731 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1732 			if (nuidp->nu_expire < time.tv_sec ||
1733 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1734 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1735 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1736 				nuidp->nu_expire = 0;
1737 				nd->nd_repstat =
1738 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1739 				nd->nd_procnum = NFSPROC_NOOP;
1740 				return (0);
1741 			}
1742 			nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1743 			nd->nd_flag |= ND_KERBNICK;
1744 		};
1745 	} else {
1746 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1747 		nd->nd_procnum = NFSPROC_NOOP;
1748 		return (0);
1749 	}
1750 
1751 	nd->nd_md = md;
1752 	nd->nd_dpos = dpos;
1753 	return (0);
1754 nfsmout:
1755 	return (error);
1756 }
1757 
1758 int
1759 nfs_msg(p, server, msg)
1760 	struct proc *p;
1761 	char *server, *msg;
1762 {
1763 	tpr_t tpr;
1764 
1765 	if (p)
1766 		tpr = tprintf_open(p);
1767 	else
1768 		tpr = NULL;
1769 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1770 	tprintf_close(tpr);
1771 	return (0);
1772 }
1773 
1774 #ifdef NFSSERVER
1775 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
1776 				    struct nfssvc_sock *, struct proc *,
1777 				    struct mbuf **)) = {
1778 	nfsrv_null,
1779 	nfsrv_getattr,
1780 	nfsrv_setattr,
1781 	nfsrv_lookup,
1782 	nfsrv3_access,
1783 	nfsrv_readlink,
1784 	nfsrv_read,
1785 	nfsrv_write,
1786 	nfsrv_create,
1787 	nfsrv_mkdir,
1788 	nfsrv_symlink,
1789 	nfsrv_mknod,
1790 	nfsrv_remove,
1791 	nfsrv_rmdir,
1792 	nfsrv_rename,
1793 	nfsrv_link,
1794 	nfsrv_readdir,
1795 	nfsrv_readdirplus,
1796 	nfsrv_statfs,
1797 	nfsrv_fsinfo,
1798 	nfsrv_pathconf,
1799 	nfsrv_commit,
1800 	nfsrv_noop,
1801 	nfsrv_noop,
1802 	nfsrv_noop,
1803 	nfsrv_noop
1804 };
1805 
1806 /*
1807  * Socket upcall routine for the nfsd sockets.
1808  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
1809  * Essentially do as much as possible non-blocking, else punt and it will
1810  * be called with M_WAIT from an nfsd.
1811  */
1812 void
1813 nfsrv_rcv(so, arg, waitflag)
1814 	struct socket *so;
1815 	caddr_t arg;
1816 	int waitflag;
1817 {
1818 	struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
1819 	struct mbuf *m;
1820 	struct mbuf *mp, *nam;
1821 	struct uio auio;
1822 	int flags, error;
1823 
1824 	if ((slp->ns_flag & SLP_VALID) == 0)
1825 		return;
1826 #ifdef notdef
1827 	/*
1828 	 * Define this to test for nfsds handling this under heavy load.
1829 	 */
1830 	if (waitflag == M_DONTWAIT) {
1831 		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
1832 	}
1833 #endif
1834 	auio.uio_procp = NULL;
1835 	if (so->so_type == SOCK_STREAM) {
1836 		/*
1837 		 * If there are already records on the queue, defer soreceive()
1838 		 * to an nfsd so that there is feedback to the TCP layer that
1839 		 * the nfs servers are heavily loaded.
1840 		 */
1841 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
1842 			slp->ns_flag |= SLP_NEEDQ;
1843 			goto dorecs;
1844 		}
1845 
1846 		/*
1847 		 * Do soreceive().
1848 		 */
1849 		auio.uio_resid = 1000000000;
1850 		flags = MSG_DONTWAIT;
1851 		error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
1852 		if (error || mp == (struct mbuf *)0) {
1853 			if (error == EWOULDBLOCK)
1854 				slp->ns_flag |= SLP_NEEDQ;
1855 			else
1856 				slp->ns_flag |= SLP_DISCONN;
1857 			goto dorecs;
1858 		}
1859 		m = mp;
1860 		if (slp->ns_rawend) {
1861 			slp->ns_rawend->m_next = m;
1862 			slp->ns_cc += 1000000000 - auio.uio_resid;
1863 		} else {
1864 			slp->ns_raw = m;
1865 			slp->ns_cc = 1000000000 - auio.uio_resid;
1866 		}
1867 		while (m->m_next)
1868 			m = m->m_next;
1869 		slp->ns_rawend = m;
1870 
1871 		/*
1872 		 * Now try and parse record(s) out of the raw stream data.
1873 		 */
1874 		error = nfsrv_getstream(slp, waitflag);
1875 		if (error) {
1876 			if (error == EPERM)
1877 				slp->ns_flag |= SLP_DISCONN;
1878 			else
1879 				slp->ns_flag |= SLP_NEEDQ;
1880 		}
1881 	} else {
1882 		do {
1883 			auio.uio_resid = 1000000000;
1884 			flags = MSG_DONTWAIT;
1885 			error = soreceive(so, &nam, &auio, &mp,
1886 						(struct mbuf **)0, &flags);
1887 			if (mp) {
1888 				nfs_realign(mp, 10 * NFSX_UNSIGNED);
1889 				if (nam) {
1890 					m = nam;
1891 					m->m_next = mp;
1892 				} else
1893 					m = mp;
1894 				if (slp->ns_recend)
1895 					slp->ns_recend->m_nextpkt = m;
1896 				else
1897 					slp->ns_rec = m;
1898 				slp->ns_recend = m;
1899 				m->m_nextpkt = (struct mbuf *)0;
1900 			}
1901 			if (error) {
1902 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
1903 					&& error != EWOULDBLOCK) {
1904 					slp->ns_flag |= SLP_DISCONN;
1905 					goto dorecs;
1906 				}
1907 			}
1908 		} while (mp);
1909 	}
1910 
1911 	/*
1912 	 * Now try and process the request records, non-blocking.
1913 	 */
1914 dorecs:
1915 	if (waitflag == M_DONTWAIT &&
1916 		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
1917 		nfsrv_wakenfsd(slp);
1918 }
1919 
1920 /*
1921  * Try and extract an RPC request from the mbuf data list received on a
1922  * stream socket. The "waitflag" argument indicates whether or not it
1923  * can sleep.
1924  */
1925 int
1926 nfsrv_getstream(slp, waitflag)
1927 	struct nfssvc_sock *slp;
1928 	int waitflag;
1929 {
1930 	struct mbuf *m, **mpp;
1931 	char *cp1, *cp2;
1932 	int len;
1933 	struct mbuf *om, *m2, *recm = NULL;
1934 	u_int32_t recmark;
1935 
1936 	if (slp->ns_flag & SLP_GETSTREAM)
1937 		panic("nfs getstream");
1938 	slp->ns_flag |= SLP_GETSTREAM;
1939 	for (;;) {
1940 	    if (slp->ns_reclen == 0) {
1941 		if (slp->ns_cc < NFSX_UNSIGNED) {
1942 			slp->ns_flag &= ~SLP_GETSTREAM;
1943 			return (0);
1944 		}
1945 		m = slp->ns_raw;
1946 		if (m->m_len >= NFSX_UNSIGNED) {
1947 			bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
1948 			m->m_data += NFSX_UNSIGNED;
1949 			m->m_len -= NFSX_UNSIGNED;
1950 		} else {
1951 			cp1 = (caddr_t)&recmark;
1952 			cp2 = mtod(m, caddr_t);
1953 			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
1954 				while (m->m_len == 0) {
1955 					m = m->m_next;
1956 					cp2 = mtod(m, caddr_t);
1957 				}
1958 				*cp1++ = *cp2++;
1959 				m->m_data++;
1960 				m->m_len--;
1961 			}
1962 		}
1963 		slp->ns_cc -= NFSX_UNSIGNED;
1964 		recmark = ntohl(recmark);
1965 		slp->ns_reclen = recmark & ~0x80000000;
1966 		if (recmark & 0x80000000)
1967 			slp->ns_flag |= SLP_LASTFRAG;
1968 		else
1969 			slp->ns_flag &= ~SLP_LASTFRAG;
1970 		if (slp->ns_reclen > NFS_MAXPACKET) {
1971 			slp->ns_flag &= ~SLP_GETSTREAM;
1972 			return (EPERM);
1973 		}
1974 	    }
1975 
1976 	    /*
1977 	     * Now get the record part.
1978 	     */
1979 	    if (slp->ns_cc == slp->ns_reclen) {
1980 		recm = slp->ns_raw;
1981 		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
1982 		slp->ns_cc = slp->ns_reclen = 0;
1983 	    } else if (slp->ns_cc > slp->ns_reclen) {
1984 		len = 0;
1985 		m = slp->ns_raw;
1986 		om = (struct mbuf *)0;
1987 		while (len < slp->ns_reclen) {
1988 			if ((len + m->m_len) > slp->ns_reclen) {
1989 				m2 = m_copym(m, 0, slp->ns_reclen - len,
1990 					waitflag);
1991 				if (m2) {
1992 					if (om) {
1993 						om->m_next = m2;
1994 						recm = slp->ns_raw;
1995 					} else
1996 						recm = m2;
1997 					m->m_data += slp->ns_reclen - len;
1998 					m->m_len -= slp->ns_reclen - len;
1999 					len = slp->ns_reclen;
2000 				} else {
2001 					slp->ns_flag &= ~SLP_GETSTREAM;
2002 					return (EWOULDBLOCK);
2003 				}
2004 			} else if ((len + m->m_len) == slp->ns_reclen) {
2005 				om = m;
2006 				len += m->m_len;
2007 				m = m->m_next;
2008 				recm = slp->ns_raw;
2009 				om->m_next = (struct mbuf *)0;
2010 			} else {
2011 				om = m;
2012 				len += m->m_len;
2013 				m = m->m_next;
2014 			}
2015 		}
2016 		slp->ns_raw = m;
2017 		slp->ns_cc -= len;
2018 		slp->ns_reclen = 0;
2019 	    } else {
2020 		slp->ns_flag &= ~SLP_GETSTREAM;
2021 		return (0);
2022 	    }
2023 
2024 	    /*
2025 	     * Accumulate the fragments into a record.
2026 	     */
2027 	    mpp = &slp->ns_frag;
2028 	    while (*mpp)
2029 		mpp = &((*mpp)->m_next);
2030 	    *mpp = recm;
2031 	    if (slp->ns_flag & SLP_LASTFRAG) {
2032 		nfs_realign(slp->ns_frag, 10 * NFSX_UNSIGNED);
2033 		if (slp->ns_recend)
2034 		    slp->ns_recend->m_nextpkt = slp->ns_frag;
2035 		else
2036 		    slp->ns_rec = slp->ns_frag;
2037 		slp->ns_recend = slp->ns_frag;
2038 		slp->ns_frag = (struct mbuf *)0;
2039 	    }
2040 	}
2041 }
2042 
2043 /*
2044  * Parse an RPC header.
2045  */
2046 int
2047 nfsrv_dorec(slp, nfsd, ndp)
2048 	struct nfssvc_sock *slp;
2049 	struct nfsd *nfsd;
2050 	struct nfsrv_descript **ndp;
2051 {
2052 	struct mbuf *m, *nam;
2053 	struct nfsrv_descript *nd;
2054 	int error;
2055 
2056 	*ndp = NULL;
2057 	if ((slp->ns_flag & SLP_VALID) == 0 ||
2058 	    (m = slp->ns_rec) == (struct mbuf *)0)
2059 		return (ENOBUFS);
2060 	slp->ns_rec = m->m_nextpkt;
2061 	if (slp->ns_rec)
2062 		m->m_nextpkt = (struct mbuf *)0;
2063 	else
2064 		slp->ns_recend = (struct mbuf *)0;
2065 	if (m->m_type == MT_SONAME) {
2066 		nam = m;
2067 		m = m->m_next;
2068 		nam->m_next = NULL;
2069 	} else
2070 		nam = NULL;
2071 	MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2072 		M_NFSRVDESC, M_WAITOK);
2073 	nd->nd_md = nd->nd_mrep = m;
2074 	nd->nd_nam2 = nam;
2075 	nd->nd_dpos = mtod(m, caddr_t);
2076 	error = nfs_getreq(nd, nfsd, TRUE);
2077 	if (error) {
2078 		m_freem(nam);
2079 		free((caddr_t)nd, M_NFSRVDESC);
2080 		return (error);
2081 	}
2082 	*ndp = nd;
2083 	nfsd->nfsd_nd = nd;
2084 	return (0);
2085 }
2086 
2087 
2088 /*
2089  * Search for a sleeping nfsd and wake it up.
2090  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2091  * running nfsds will go look for the work in the nfssvc_sock list.
2092  */
2093 void
2094 nfsrv_wakenfsd(slp)
2095 	struct nfssvc_sock *slp;
2096 {
2097 	struct nfsd *nd;
2098 
2099 	if ((slp->ns_flag & SLP_VALID) == 0)
2100 		return;
2101 	for (nd = TAILQ_FIRST(&nfsd_head); nd != NULL;
2102 	    nd = TAILQ_NEXT(nd, nfsd_chain)) {
2103 		if (nd->nfsd_flag & NFSD_WAITING) {
2104 			nd->nfsd_flag &= ~NFSD_WAITING;
2105 			if (nd->nfsd_slp)
2106 				panic("nfsd wakeup");
2107 			slp->ns_sref++;
2108 			nd->nfsd_slp = slp;
2109 			wakeup((caddr_t)nd);
2110 			return;
2111 		}
2112 	}
2113 	slp->ns_flag |= SLP_DOREC;
2114 	nfsd_head_flag |= NFSD_CHECKSLP;
2115 }
2116 #endif /* NFSSERVER */
2117