xref: /netbsd-src/sys/nfs/nfs_socket.c (revision d48f14661dda8638fee055ba15d35bdfb29b9fa8)
1 /*	$NetBSD: nfs_socket.c,v 1.136 2006/06/30 09:55:06 yamt 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 1.136 2006/06/30 09:55:06 yamt 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 #include <sys/kauth.h>
68 
69 #include <netinet/in.h>
70 #include <netinet/tcp.h>
71 
72 #include <nfs/rpcv2.h>
73 #include <nfs/nfsproto.h>
74 #include <nfs/nfs.h>
75 #include <nfs/xdr_subs.h>
76 #include <nfs/nfsm_subs.h>
77 #include <nfs/nfsmount.h>
78 #include <nfs/nfsnode.h>
79 #include <nfs/nfsrtt.h>
80 #include <nfs/nqnfs.h>
81 #include <nfs/nfs_var.h>
82 
83 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
84 #ifdef MBUFTRACE
85 struct mowner nfs_mowner = { "nfs" };
86 #endif
87 
88 /*
89  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
90  * Use the mean and mean deviation of rtt for the appropriate type of rpc
91  * for the frequent rpcs and a default for the others.
92  * The justification for doing "other" this way is that these rpcs
93  * happen so infrequently that timer est. would probably be stale.
94  * Also, since many of these rpcs are
95  * non-idempotent, a conservative timeout is desired.
96  * getattr, lookup - A+2D
97  * read, write     - A+4D
98  * other           - nm_timeo
99  */
100 #define	NFS_RTO(n, t) \
101 	((t) == 0 ? (n)->nm_timeo : \
102 	 ((t) < 3 ? \
103 	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
104 	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
105 #define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
106 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
107 /*
108  * External data, mostly RPC constants in XDR form
109  */
110 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
111 	rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
112 	rpc_auth_kerb;
113 extern u_int32_t nfs_prog, nqnfs_prog;
114 extern time_t nqnfsstarttime;
115 extern const int nfsv3_procid[NFS_NPROCS];
116 extern int nfs_ticks;
117 
118 /*
119  * Defines which timer to use for the procnum.
120  * 0 - default
121  * 1 - getattr
122  * 2 - lookup
123  * 3 - read
124  * 4 - write
125  */
126 static const int proct[NFS_NPROCS] = {
127 	[NFSPROC_NULL] = 0,
128 	[NFSPROC_GETATTR] = 1,
129 	[NFSPROC_SETATTR] = 0,
130 	[NFSPROC_LOOKUP] = 2,
131 	[NFSPROC_ACCESS] = 1,
132 	[NFSPROC_READLINK] = 3,
133 	[NFSPROC_READ] = 3,
134 	[NFSPROC_WRITE] = 4,
135 	[NFSPROC_CREATE] = 0,
136 	[NFSPROC_MKDIR] = 0,
137 	[NFSPROC_SYMLINK] = 0,
138 	[NFSPROC_MKNOD] = 0,
139 	[NFSPROC_REMOVE] = 0,
140 	[NFSPROC_RMDIR] = 0,
141 	[NFSPROC_RENAME] = 0,
142 	[NFSPROC_LINK] = 0,
143 	[NFSPROC_READDIR] = 3,
144 	[NFSPROC_READDIRPLUS] = 3,
145 	[NFSPROC_FSSTAT] = 0,
146 	[NFSPROC_FSINFO] = 0,
147 	[NFSPROC_PATHCONF] = 0,
148 	[NFSPROC_COMMIT] = 0,
149 	[NQNFSPROC_GETLEASE] = 0,
150 	[NQNFSPROC_VACATED] = 0,
151 	[NQNFSPROC_EVICTED] = 0,
152 	[NFSPROC_NOOP] = 0,
153 };
154 
155 /*
156  * There is a congestion window for outstanding rpcs maintained per mount
157  * point. The cwnd size is adjusted in roughly the way that:
158  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
159  * SIGCOMM '88". ACM, August 1988.
160  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
161  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
162  * of rpcs is in progress.
163  * (The sent count and cwnd are scaled for integer arith.)
164  * Variants of "slow start" were tried and were found to be too much of a
165  * performance hit (ave. rtt 3 times larger),
166  * I suspect due to the large rtt that nfs rpcs have.
167  */
168 #define	NFS_CWNDSCALE	256
169 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
170 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
171 int nfsrtton = 0;
172 struct nfsrtt nfsrtt;
173 struct nfsreqhead nfs_reqq;
174 
175 struct callout nfs_timer_ch = CALLOUT_INITIALIZER_SETFUNC(nfs_timer, NULL);
176 
177 /*
178  * Initialize sockets and congestion for a new NFS connection.
179  * We do not free the sockaddr if error.
180  */
181 int
182 nfs_connect(nmp, rep, l)
183 	struct nfsmount *nmp;
184 	struct nfsreq *rep;
185 	struct lwp *l;
186 {
187 	struct socket *so;
188 	int s, error, rcvreserve, sndreserve;
189 	struct sockaddr *saddr;
190 	struct sockaddr_in *sin;
191 #ifdef INET6
192 	struct sockaddr_in6 *sin6;
193 #endif
194 	struct mbuf *m;
195 
196 	nmp->nm_so = (struct socket *)0;
197 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
198 	error = socreate(saddr->sa_family, &nmp->nm_so,
199 		nmp->nm_sotype, nmp->nm_soproto, l);
200 	if (error)
201 		goto bad;
202 	so = nmp->nm_so;
203 #ifdef MBUFTRACE
204 	so->so_mowner = &nfs_mowner;
205 	so->so_rcv.sb_mowner = &nfs_mowner;
206 	so->so_snd.sb_mowner = &nfs_mowner;
207 #endif
208 	nmp->nm_soflags = so->so_proto->pr_flags;
209 
210 	/*
211 	 * Some servers require that the client port be a reserved port number.
212 	 */
213 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
214 		m = m_get(M_WAIT, MT_SOOPTS);
215 		MCLAIM(m, so->so_mowner);
216 		*mtod(m, int32_t *) = IP_PORTRANGE_LOW;
217 		m->m_len = sizeof(int32_t);
218 		if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m)))
219 			goto bad;
220 		m = m_get(M_WAIT, MT_SONAME);
221 		MCLAIM(m, so->so_mowner);
222 		sin = mtod(m, struct sockaddr_in *);
223 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
224 		sin->sin_family = AF_INET;
225 		sin->sin_addr.s_addr = INADDR_ANY;
226 		sin->sin_port = 0;
227 		error = sobind(so, m, &lwp0);
228 		m_freem(m);
229 		if (error)
230 			goto bad;
231 	}
232 #ifdef INET6
233 	if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) {
234 		m = m_get(M_WAIT, MT_SOOPTS);
235 		MCLAIM(m, so->so_mowner);
236 		*mtod(m, int32_t *) = IPV6_PORTRANGE_LOW;
237 		m->m_len = sizeof(int32_t);
238 		if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m)))
239 			goto bad;
240 		m = m_get(M_WAIT, MT_SONAME);
241 		MCLAIM(m, so->so_mowner);
242 		sin6 = mtod(m, struct sockaddr_in6 *);
243 		sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6);
244 		sin6->sin6_family = AF_INET6;
245 		sin6->sin6_addr = in6addr_any;
246 		sin6->sin6_port = 0;
247 		error = sobind(so, m, &lwp0);
248 		m_freem(m);
249 		if (error)
250 			goto bad;
251 	}
252 #endif
253 
254 	/*
255 	 * Protocols that do not require connections may be optionally left
256 	 * unconnected for servers that reply from a port other than NFS_PORT.
257 	 */
258 	if (nmp->nm_flag & NFSMNT_NOCONN) {
259 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
260 			error = ENOTCONN;
261 			goto bad;
262 		}
263 	} else {
264 		error = soconnect(so, nmp->nm_nam, l);
265 		if (error)
266 			goto bad;
267 
268 		/*
269 		 * Wait for the connection to complete. Cribbed from the
270 		 * connect system call but with the wait timing out so
271 		 * that interruptible mounts don't hang here for a long time.
272 		 */
273 		s = splsoftnet();
274 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
275 			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
276 				"nfscn1", 2 * hz);
277 			if ((so->so_state & SS_ISCONNECTING) &&
278 			    so->so_error == 0 && rep &&
279 			    (error = nfs_sigintr(nmp, rep, rep->r_lwp)) != 0){
280 				so->so_state &= ~SS_ISCONNECTING;
281 				splx(s);
282 				goto bad;
283 			}
284 		}
285 		if (so->so_error) {
286 			error = so->so_error;
287 			so->so_error = 0;
288 			splx(s);
289 			goto bad;
290 		}
291 		splx(s);
292 	}
293 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
294 		so->so_rcv.sb_timeo = (5 * hz);
295 		so->so_snd.sb_timeo = (5 * hz);
296 	} else {
297 		/*
298 		 * enable receive timeout to detect server crash and reconnect.
299 		 * otherwise, we can be stuck in soreceive forever.
300 		 */
301 		so->so_rcv.sb_timeo = (5 * hz);
302 		so->so_snd.sb_timeo = 0;
303 	}
304 	if (nmp->nm_sotype == SOCK_DGRAM) {
305 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
306 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
307 		    NFS_MAXPKTHDR) * 2;
308 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
309 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
310 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
311 		    NFS_MAXPKTHDR) * 2;
312 	} else {
313 		if (nmp->nm_sotype != SOCK_STREAM)
314 			panic("nfscon sotype");
315 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
316 			m = m_get(M_WAIT, MT_SOOPTS);
317 			MCLAIM(m, so->so_mowner);
318 			*mtod(m, int32_t *) = 1;
319 			m->m_len = sizeof(int32_t);
320 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
321 		}
322 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
323 			m = m_get(M_WAIT, MT_SOOPTS);
324 			MCLAIM(m, so->so_mowner);
325 			*mtod(m, int32_t *) = 1;
326 			m->m_len = sizeof(int32_t);
327 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
328 		}
329 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
330 		    sizeof (u_int32_t)) * 2;
331 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
332 		    sizeof (u_int32_t)) * 2;
333 	}
334 	error = soreserve(so, sndreserve, rcvreserve);
335 	if (error)
336 		goto bad;
337 	so->so_rcv.sb_flags |= SB_NOINTR;
338 	so->so_snd.sb_flags |= SB_NOINTR;
339 
340 	/* Initialize other non-zero congestion variables */
341 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
342 		NFS_TIMEO << 3;
343 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
344 		nmp->nm_sdrtt[3] = 0;
345 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
346 	nmp->nm_sent = 0;
347 	nmp->nm_timeouts = 0;
348 	return (0);
349 
350 bad:
351 	nfs_disconnect(nmp);
352 	return (error);
353 }
354 
355 /*
356  * Reconnect routine:
357  * Called when a connection is broken on a reliable protocol.
358  * - clean up the old socket
359  * - nfs_connect() again
360  * - set R_MUSTRESEND for all outstanding requests on mount point
361  * If this fails the mount point is DEAD!
362  * nb: Must be called with the nfs_sndlock() set on the mount point.
363  */
364 int
365 nfs_reconnect(rep, l)
366 	struct nfsreq *rep;
367 	struct lwp *l;
368 {
369 	struct nfsreq *rp;
370 	struct nfsmount *nmp = rep->r_nmp;
371 	int error;
372 
373 	nfs_disconnect(nmp);
374 	while ((error = nfs_connect(nmp, rep, l)) != 0) {
375 		if (error == EINTR || error == ERESTART)
376 			return (EINTR);
377 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0);
378 	}
379 
380 	/*
381 	 * Loop through outstanding request list and fix up all requests
382 	 * on old socket.
383 	 */
384 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
385 		if (rp->r_nmp == nmp) {
386 			if ((rp->r_flags & R_MUSTRESEND) == 0)
387 				rp->r_flags |= R_MUSTRESEND | R_REXMITTED;
388 			rp->r_rexmit = 0;
389 		}
390 	}
391 	return (0);
392 }
393 
394 /*
395  * NFS disconnect. Clean up and unlink.
396  */
397 void
398 nfs_disconnect(nmp)
399 	struct nfsmount *nmp;
400 {
401 	struct socket *so;
402 	int drain = 0;
403 
404 	if (nmp->nm_so) {
405 		so = nmp->nm_so;
406 		nmp->nm_so = (struct socket *)0;
407 		soshutdown(so, 2);
408 		drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
409 		if (drain) {
410 			/*
411 			 * soshutdown() above should wake up the current
412 			 * listener.
413 			 * Now wake up those waiting for the receive lock, and
414 			 * wait for them to go away unhappy, to prevent *nmp
415 			 * from evaporating while they're sleeping.
416 			 */
417 			while (nmp->nm_waiters > 0) {
418 				wakeup (&nmp->nm_iflag);
419 				(void) tsleep(&nmp->nm_waiters, PVFS,
420 				    "nfsdis", 0);
421 			}
422 		}
423 		soclose(so);
424 	}
425 #ifdef DIAGNOSTIC
426 	if (drain && (nmp->nm_waiters > 0))
427 		panic("nfs_disconnect: waiters left after drain?");
428 #endif
429 }
430 
431 void
432 nfs_safedisconnect(nmp)
433 	struct nfsmount *nmp;
434 {
435 	struct nfsreq dummyreq;
436 
437 	memset(&dummyreq, 0, sizeof(dummyreq));
438 	dummyreq.r_nmp = nmp;
439 	nfs_rcvlock(&dummyreq); /* XXX ignored error return */
440 	nfs_disconnect(nmp);
441 	nfs_rcvunlock(nmp);
442 }
443 
444 /*
445  * This is the nfs send routine. For connection based socket types, it
446  * must be called with an nfs_sndlock() on the socket.
447  * "rep == NULL" indicates that it has been called from a server.
448  * For the client side:
449  * - return EINTR if the RPC is terminated, 0 otherwise
450  * - set R_MUSTRESEND if the send fails for any reason
451  * - do any cleanup required by recoverable socket errors (? ? ?)
452  * For the server side:
453  * - return EINTR or ERESTART if interrupted by a signal
454  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
455  * - do any cleanup required by recoverable socket errors (? ? ?)
456  */
457 int
458 nfs_send(so, nam, top, rep, l)
459 	struct socket *so;
460 	struct mbuf *nam;
461 	struct mbuf *top;
462 	struct nfsreq *rep;
463 	struct lwp *l;
464 {
465 	struct mbuf *sendnam;
466 	int error, soflags, flags;
467 
468 	/* XXX nfs_doio()/nfs_request() calls with  rep->r_lwp == NULL */
469 	if (l == NULL && rep->r_lwp == NULL)
470 		l = curlwp;
471 
472 	if (rep) {
473 		if (rep->r_flags & R_SOFTTERM) {
474 			m_freem(top);
475 			return (EINTR);
476 		}
477 		if ((so = rep->r_nmp->nm_so) == NULL) {
478 			rep->r_flags |= R_MUSTRESEND;
479 			m_freem(top);
480 			return (0);
481 		}
482 		rep->r_flags &= ~R_MUSTRESEND;
483 		soflags = rep->r_nmp->nm_soflags;
484 	} else
485 		soflags = so->so_proto->pr_flags;
486 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
487 		sendnam = (struct mbuf *)0;
488 	else
489 		sendnam = nam;
490 	if (so->so_type == SOCK_SEQPACKET)
491 		flags = MSG_EOR;
492 	else
493 		flags = 0;
494 
495 	error = (*so->so_send)(so, sendnam, (struct uio *)0, top,
496 		    (struct mbuf *)0, flags,  l);
497 	if (error) {
498 		if (rep) {
499 			if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
500 				/*
501 				 * We're too fast for the network/driver,
502 				 * and UDP isn't flowcontrolled.
503 				 * We need to resend. This is not fatal,
504 				 * just try again.
505 				 *
506 				 * Could be smarter here by doing some sort
507 				 * of a backoff, but this is rare.
508 				 */
509 				rep->r_flags |= R_MUSTRESEND;
510 			} else {
511 				if (error != EPIPE)
512 					log(LOG_INFO,
513 					    "nfs send error %d for %s\n",
514 					    error,
515 					    rep->r_nmp->nm_mountp->
516 						    mnt_stat.f_mntfromname);
517 				/*
518 				 * Deal with errors for the client side.
519 				 */
520 				if (rep->r_flags & R_SOFTTERM)
521 					error = EINTR;
522 				else
523 					rep->r_flags |= R_MUSTRESEND;
524 			}
525 		} else {
526 			/*
527 			 * See above. This error can happen under normal
528 			 * circumstances and the log is too noisy.
529 			 * The error will still show up in nfsstat.
530 			 */
531 			if (error != ENOBUFS || so->so_type != SOCK_DGRAM)
532 				log(LOG_INFO, "nfsd send error %d\n", error);
533 		}
534 
535 		/*
536 		 * Handle any recoverable (soft) socket errors here. (? ? ?)
537 		 */
538 		if (error != EINTR && error != ERESTART &&
539 			error != EWOULDBLOCK && error != EPIPE)
540 			error = 0;
541 	}
542 	return (error);
543 }
544 
545 #ifdef NFS
546 /*
547  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
548  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
549  * Mark and consolidate the data into a new mbuf list.
550  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
551  *     small mbufs.
552  * For SOCK_STREAM we must be very careful to read an entire record once
553  * we have read any of it, even if the system call has been interrupted.
554  */
555 int
556 nfs_receive(rep, aname, mp, l)
557 	struct nfsreq *rep;
558 	struct mbuf **aname;
559 	struct mbuf **mp;
560 	struct lwp *l;
561 {
562 	struct socket *so;
563 	struct uio auio;
564 	struct iovec aio;
565 	struct mbuf *m;
566 	struct mbuf *control;
567 	u_int32_t len;
568 	struct mbuf **getnam;
569 	int error, sotype, rcvflg;
570 
571 	/*
572 	 * Set up arguments for soreceive()
573 	 */
574 	*mp = (struct mbuf *)0;
575 	*aname = (struct mbuf *)0;
576 	sotype = rep->r_nmp->nm_sotype;
577 
578 	/*
579 	 * For reliable protocols, lock against other senders/receivers
580 	 * in case a reconnect is necessary.
581 	 * For SOCK_STREAM, first get the Record Mark to find out how much
582 	 * more there is to get.
583 	 * We must lock the socket against other receivers
584 	 * until we have an entire rpc request/reply.
585 	 */
586 	if (sotype != SOCK_DGRAM) {
587 		error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
588 		if (error)
589 			return (error);
590 tryagain:
591 		/*
592 		 * Check for fatal errors and resending request.
593 		 */
594 		/*
595 		 * Ugh: If a reconnect attempt just happened, nm_so
596 		 * would have changed. NULL indicates a failed
597 		 * attempt that has essentially shut down this
598 		 * mount point.
599 		 */
600 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
601 			nfs_sndunlock(&rep->r_nmp->nm_iflag);
602 			return (EINTR);
603 		}
604 		so = rep->r_nmp->nm_so;
605 		if (!so) {
606 			error = nfs_reconnect(rep, l);
607 			if (error) {
608 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
609 				return (error);
610 			}
611 			goto tryagain;
612 		}
613 		while (rep->r_flags & R_MUSTRESEND) {
614 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
615 			nfsstats.rpcretries++;
616 			rep->r_rtt = 0;
617 			rep->r_flags &= ~R_TIMING;
618 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, l);
619 			if (error) {
620 				if (error == EINTR || error == ERESTART ||
621 				    (error = nfs_reconnect(rep, l)) != 0) {
622 					nfs_sndunlock(&rep->r_nmp->nm_iflag);
623 					return (error);
624 				}
625 				goto tryagain;
626 			}
627 		}
628 		nfs_sndunlock(&rep->r_nmp->nm_iflag);
629 		if (sotype == SOCK_STREAM) {
630 			aio.iov_base = (caddr_t) &len;
631 			aio.iov_len = sizeof(u_int32_t);
632 			auio.uio_iov = &aio;
633 			auio.uio_iovcnt = 1;
634 			auio.uio_rw = UIO_READ;
635 			auio.uio_offset = 0;
636 			auio.uio_resid = sizeof(u_int32_t);
637 			UIO_SETUP_SYSSPACE(&auio);
638 			do {
639 			   rcvflg = MSG_WAITALL;
640 			   error = (*so->so_receive)(so, (struct mbuf **)0, &auio,
641 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
642 			   if (error == EWOULDBLOCK && rep) {
643 				if (rep->r_flags & R_SOFTTERM)
644 					return (EINTR);
645 				/*
646 				 * if it seems that the server died after it
647 				 * received our request, set EPIPE so that
648 				 * we'll reconnect and retransmit requests.
649 				 */
650 				if (rep->r_rexmit >= rep->r_nmp->nm_retry) {
651 					nfsstats.rpctimeouts++;
652 					error = EPIPE;
653 				}
654 			   }
655 			} while (error == EWOULDBLOCK);
656 			if (!error && auio.uio_resid > 0) {
657 			    /*
658 			     * Don't log a 0 byte receive; it means
659 			     * that the socket has been closed, and
660 			     * can happen during normal operation
661 			     * (forcible unmount or Solaris server).
662 			     */
663 			    if (auio.uio_resid != sizeof (u_int32_t))
664 			      log(LOG_INFO,
665 				 "short receive (%lu/%lu) from nfs server %s\n",
666 				 (u_long)sizeof(u_int32_t) - auio.uio_resid,
667 				 (u_long)sizeof(u_int32_t),
668 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
669 			    error = EPIPE;
670 			}
671 			if (error)
672 				goto errout;
673 			len = ntohl(len) & ~0x80000000;
674 			/*
675 			 * This is SERIOUS! We are out of sync with the sender
676 			 * and forcing a disconnect/reconnect is all I can do.
677 			 */
678 			if (len > NFS_MAXPACKET) {
679 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
680 				"impossible packet length",
681 				len,
682 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
683 			    error = EFBIG;
684 			    goto errout;
685 			}
686 			auio.uio_resid = len;
687 			do {
688 			    rcvflg = MSG_WAITALL;
689 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
690 				&auio, mp, (struct mbuf **)0, &rcvflg);
691 			} while (error == EWOULDBLOCK || error == EINTR ||
692 				 error == ERESTART);
693 			if (!error && auio.uio_resid > 0) {
694 			    if (len != auio.uio_resid)
695 			      log(LOG_INFO,
696 				"short receive (%lu/%d) from nfs server %s\n",
697 				(u_long)len - auio.uio_resid, len,
698 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
699 			    error = EPIPE;
700 			}
701 		} else {
702 			/*
703 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
704 			 * and soreceive() will return when it has either a
705 			 * control msg or a data msg.
706 			 * We have no use for control msg., but must grab them
707 			 * and then throw them away so we know what is going
708 			 * on.
709 			 */
710 			auio.uio_resid = len = 100000000; /* Anything Big */
711 			/* not need to setup uio_vmspace */
712 			do {
713 			    rcvflg = 0;
714 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
715 				&auio, mp, &control, &rcvflg);
716 			    if (control)
717 				m_freem(control);
718 			    if (error == EWOULDBLOCK && rep) {
719 				if (rep->r_flags & R_SOFTTERM)
720 					return (EINTR);
721 			    }
722 			} while (error == EWOULDBLOCK ||
723 				 (!error && *mp == NULL && control));
724 			if ((rcvflg & MSG_EOR) == 0)
725 				printf("Egad!!\n");
726 			if (!error && *mp == NULL)
727 				error = EPIPE;
728 			len -= auio.uio_resid;
729 		}
730 errout:
731 		if (error && error != EINTR && error != ERESTART) {
732 			m_freem(*mp);
733 			*mp = (struct mbuf *)0;
734 			if (error != EPIPE)
735 				log(LOG_INFO,
736 				    "receive error %d from nfs server %s\n",
737 				    error,
738 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
739 			error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
740 			if (!error)
741 				error = nfs_reconnect(rep, l);
742 			if (!error)
743 				goto tryagain;
744 			else
745 				nfs_sndunlock(&rep->r_nmp->nm_iflag);
746 		}
747 	} else {
748 		if ((so = rep->r_nmp->nm_so) == NULL)
749 			return (EACCES);
750 		if (so->so_state & SS_ISCONNECTED)
751 			getnam = (struct mbuf **)0;
752 		else
753 			getnam = aname;
754 		auio.uio_resid = len = 1000000;
755 		/* not need to setup uio_vmspace */
756 		do {
757 			rcvflg = 0;
758 			error =  (*so->so_receive)(so, getnam, &auio, mp,
759 				(struct mbuf **)0, &rcvflg);
760 			if (error == EWOULDBLOCK &&
761 			    (rep->r_flags & R_SOFTTERM))
762 				return (EINTR);
763 		} while (error == EWOULDBLOCK);
764 		len -= auio.uio_resid;
765 		if (!error && *mp == NULL)
766 			error = EPIPE;
767 	}
768 	if (error) {
769 		m_freem(*mp);
770 		*mp = (struct mbuf *)0;
771 	}
772 	return (error);
773 }
774 
775 /*
776  * Implement receipt of reply on a socket.
777  * We must search through the list of received datagrams matching them
778  * with outstanding requests using the xid, until ours is found.
779  */
780 /* ARGSUSED */
781 int
782 nfs_reply(myrep, lwp)
783 	struct nfsreq *myrep;
784 	struct lwp *lwp;
785 {
786 	struct nfsreq *rep;
787 	struct nfsmount *nmp = myrep->r_nmp;
788 	int32_t t1;
789 	struct mbuf *mrep, *nam, *md;
790 	u_int32_t rxid, *tl;
791 	caddr_t dpos, cp2;
792 	int error;
793 
794 	/*
795 	 * Loop around until we get our own reply
796 	 */
797 	for (;;) {
798 		/*
799 		 * Lock against other receivers so that I don't get stuck in
800 		 * sbwait() after someone else has received my reply for me.
801 		 * Also necessary for connection based protocols to avoid
802 		 * race conditions during a reconnect.
803 		 */
804 		error = nfs_rcvlock(myrep);
805 		if (error == EALREADY)
806 			return (0);
807 		if (error)
808 			return (error);
809 		/*
810 		 * Get the next Rpc reply off the socket
811 		 */
812 		nmp->nm_waiters++;
813 		error = nfs_receive(myrep, &nam, &mrep, lwp);
814 		nfs_rcvunlock(nmp);
815 		if (error) {
816 
817 			if (nmp->nm_iflag & NFSMNT_DISMNT) {
818 				/*
819 				 * Oops, we're going away now..
820 				 */
821 				nmp->nm_waiters--;
822 				wakeup (&nmp->nm_waiters);
823 				return error;
824 			}
825 			nmp->nm_waiters--;
826 			/*
827 			 * Ignore routing errors on connectionless protocols? ?
828 			 */
829 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
830 				nmp->nm_so->so_error = 0;
831 #ifdef DEBUG
832 				printf("nfs_reply: ignoring error %d\n", error);
833 #endif
834 				if (myrep->r_flags & R_GETONEREP)
835 					return (0);
836 				continue;
837 			}
838 			return (error);
839 		}
840 		nmp->nm_waiters--;
841 		if (nam)
842 			m_freem(nam);
843 
844 		/*
845 		 * Get the xid and check that it is an rpc reply
846 		 */
847 		md = mrep;
848 		dpos = mtod(md, caddr_t);
849 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
850 		rxid = *tl++;
851 		if (*tl != rpc_reply) {
852 #ifndef NFS_V2_ONLY
853 			if (nmp->nm_flag & NFSMNT_NQNFS) {
854 				if (nqnfs_callback(nmp, mrep, md, dpos,
855 				    myrep->r_lwp))
856 					nfsstats.rpcinvalid++;
857 			} else
858 #endif
859 			{
860 				nfsstats.rpcinvalid++;
861 				m_freem(mrep);
862 			}
863 nfsmout:
864 			if (myrep->r_flags & R_GETONEREP)
865 				return (0);
866 			continue;
867 		}
868 
869 		/*
870 		 * Loop through the request list to match up the reply
871 		 * Iff no match, just drop the datagram
872 		 */
873 		TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
874 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
875 				/* Found it.. */
876 				rep->r_mrep = mrep;
877 				rep->r_md = md;
878 				rep->r_dpos = dpos;
879 				if (nfsrtton) {
880 					struct rttl *rt;
881 
882 					rt = &nfsrtt.rttl[nfsrtt.pos];
883 					rt->proc = rep->r_procnum;
884 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
885 					rt->sent = nmp->nm_sent;
886 					rt->cwnd = nmp->nm_cwnd;
887 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
888 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
889 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx;
890 					getmicrotime(&rt->tstamp);
891 					if (rep->r_flags & R_TIMING)
892 						rt->rtt = rep->r_rtt;
893 					else
894 						rt->rtt = 1000000;
895 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
896 				}
897 				/*
898 				 * Update congestion window.
899 				 * Do the additive increase of
900 				 * one rpc/rtt.
901 				 */
902 				if (nmp->nm_cwnd <= nmp->nm_sent) {
903 					nmp->nm_cwnd +=
904 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
905 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
906 					if (nmp->nm_cwnd > NFS_MAXCWND)
907 						nmp->nm_cwnd = NFS_MAXCWND;
908 				}
909 				rep->r_flags &= ~R_SENT;
910 				nmp->nm_sent -= NFS_CWNDSCALE;
911 				/*
912 				 * Update rtt using a gain of 0.125 on the mean
913 				 * and a gain of 0.25 on the deviation.
914 				 */
915 				if (rep->r_flags & R_TIMING) {
916 					/*
917 					 * Since the timer resolution of
918 					 * NFS_HZ is so course, it can often
919 					 * result in r_rtt == 0. Since
920 					 * r_rtt == N means that the actual
921 					 * rtt is between N+dt and N+2-dt ticks,
922 					 * add 1.
923 					 */
924 					t1 = rep->r_rtt + 1;
925 					t1 -= (NFS_SRTT(rep) >> 3);
926 					NFS_SRTT(rep) += t1;
927 					if (t1 < 0)
928 						t1 = -t1;
929 					t1 -= (NFS_SDRTT(rep) >> 2);
930 					NFS_SDRTT(rep) += t1;
931 				}
932 				nmp->nm_timeouts = 0;
933 				break;
934 			}
935 		}
936 		/*
937 		 * If not matched to a request, drop it.
938 		 * If it's mine, get out.
939 		 */
940 		if (rep == 0) {
941 			nfsstats.rpcunexpected++;
942 			m_freem(mrep);
943 		} else if (rep == myrep) {
944 			if (rep->r_mrep == NULL)
945 				panic("nfsreply nil");
946 			return (0);
947 		}
948 		if (myrep->r_flags & R_GETONEREP)
949 			return (0);
950 	}
951 }
952 
953 /*
954  * nfs_request - goes something like this
955  *	- fill in request struct
956  *	- links it into list
957  *	- calls nfs_send() for first transmit
958  *	- calls nfs_receive() to get reply
959  *	- break down rpc header and return with nfs reply pointed to
960  *	  by mrep or error
961  * nb: always frees up mreq mbuf list
962  */
963 int
964 nfs_request(np, mrest, procnum, lwp, cred, mrp, mdp, dposp, rexmitp)
965 	struct nfsnode *np;
966 	struct mbuf *mrest;
967 	int procnum;
968 	struct lwp *lwp;
969 	kauth_cred_t cred;
970 	struct mbuf **mrp;
971 	struct mbuf **mdp;
972 	caddr_t *dposp;
973 	int *rexmitp;
974 {
975 	struct mbuf *m, *mrep;
976 	struct nfsreq *rep;
977 	u_int32_t *tl;
978 	int i;
979 	struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount);
980 	struct mbuf *md, *mheadend;
981 	char nickv[RPCX_NICKVERF];
982 	time_t reqtime, waituntil;
983 	caddr_t dpos, cp2;
984 	int t1, s, error = 0, mrest_len, auth_len, auth_type;
985 	int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0;
986 	int verf_len, verf_type;
987 	u_int32_t xid;
988 	char *auth_str, *verf_str;
989 	NFSKERBKEY_T key;		/* save session key */
990 	kauth_cred_t acred;
991 #ifndef NFS_V2_ONLY
992 	int nqlflag, cachable;
993 	u_quad_t frev;
994 #endif
995 	struct mbuf *mrest_backup = NULL;
996 	kauth_cred_t origcred = NULL; /* XXX: gcc */
997 	boolean_t retry_cred = TRUE;
998 	boolean_t use_opencred = (np->n_flag & NUSEOPENCRED) != 0;
999 
1000 	if (rexmitp != NULL)
1001 		*rexmitp = 0;
1002 
1003 	acred = kauth_cred_alloc();
1004 
1005 tryagain_cred:
1006 	KASSERT(cred != NULL);
1007 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
1008 	rep->r_nmp = nmp;
1009 	rep->r_lwp = lwp;
1010 	rep->r_procnum = procnum;
1011 	i = 0;
1012 	m = mrest;
1013 	while (m) {
1014 		i += m->m_len;
1015 		m = m->m_next;
1016 	}
1017 	mrest_len = i;
1018 
1019 	/*
1020 	 * Get the RPC header with authorization.
1021 	 */
1022 kerbauth:
1023 	verf_str = auth_str = (char *)0;
1024 	if (nmp->nm_flag & NFSMNT_KERB) {
1025 		verf_str = nickv;
1026 		verf_len = sizeof (nickv);
1027 		auth_type = RPCAUTH_KERB4;
1028 		memset((caddr_t)key, 0, sizeof (key));
1029 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
1030 			&auth_len, verf_str, verf_len)) {
1031 			error = nfs_getauth(nmp, rep, cred, &auth_str,
1032 				&auth_len, verf_str, &verf_len, key);
1033 			if (error) {
1034 				free((caddr_t)rep, M_NFSREQ);
1035 				m_freem(mrest);
1036 				KASSERT(kauth_cred_getrefcnt(acred) == 1);
1037 				kauth_cred_free(acred);
1038 				return (error);
1039 			}
1040 		}
1041 		retry_cred = FALSE;
1042 	} else {
1043 		/* AUTH_UNIX */
1044 		uid_t uid;
1045 		gid_t gid;
1046 
1047 		/*
1048 		 * on the most unix filesystems, permission checks are
1049 		 * done when the file is open(2)'ed.
1050 		 * ie. once a file is successfully open'ed,
1051 		 * following i/o operations never fail with EACCES.
1052 		 * we try to follow the semantics as far as possible.
1053 		 *
1054 		 * note that we expect that the nfs server always grant
1055 		 * accesses by the file's owner.
1056 		 */
1057 		origcred = cred;
1058 		switch (procnum) {
1059 		case NFSPROC_READ:
1060 		case NFSPROC_WRITE:
1061 		case NFSPROC_COMMIT:
1062 			uid = np->n_vattr->va_uid;
1063 			gid = np->n_vattr->va_gid;
1064 			if (kauth_cred_geteuid(cred) == uid &&
1065 			    kauth_cred_getegid(cred) == gid) {
1066 				retry_cred = FALSE;
1067 				break;
1068 			}
1069 			if (use_opencred)
1070 				break;
1071 			kauth_cred_setuid(acred, uid);
1072 			kauth_cred_seteuid(acred, uid);
1073 			kauth_cred_setsvuid(acred, uid);
1074 			kauth_cred_setgid(acred, gid);
1075 			kauth_cred_setegid(acred, gid);
1076 			kauth_cred_setsvgid(acred, gid);
1077 			cred = acred;
1078 			break;
1079 		default:
1080 			retry_cred = FALSE;
1081 			break;
1082 		}
1083 		/*
1084 		 * backup mbuf chain if we can need it later to retry.
1085 		 *
1086 		 * XXX maybe we can keep a direct reference to
1087 		 * mrest without doing m_copym, but it's ...ugly.
1088 		 */
1089 		if (retry_cred)
1090 			mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT);
1091 		auth_type = RPCAUTH_UNIX;
1092 		/* XXX elad - ngroups */
1093 		auth_len = (((kauth_cred_ngroups(cred) > nmp->nm_numgrps) ?
1094 			nmp->nm_numgrps : kauth_cred_ngroups(cred)) << 2) +
1095 			5 * NFSX_UNSIGNED;
1096 	}
1097 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
1098 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
1099 	if (auth_str)
1100 		free(auth_str, M_TEMP);
1101 
1102 	/*
1103 	 * For stream protocols, insert a Sun RPC Record Mark.
1104 	 */
1105 	if (nmp->nm_sotype == SOCK_STREAM) {
1106 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
1107 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
1108 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
1109 	}
1110 	rep->r_mreq = m;
1111 	rep->r_xid = xid;
1112 tryagain:
1113 	if (nmp->nm_flag & NFSMNT_SOFT)
1114 		rep->r_retry = nmp->nm_retry;
1115 	else
1116 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
1117 	rep->r_rtt = rep->r_rexmit = 0;
1118 	if (proct[procnum] > 0)
1119 		rep->r_flags = R_TIMING;
1120 	else
1121 		rep->r_flags = 0;
1122 	rep->r_mrep = NULL;
1123 
1124 	/*
1125 	 * Do the client side RPC.
1126 	 */
1127 	nfsstats.rpcrequests++;
1128 	/*
1129 	 * Chain request into list of outstanding requests. Be sure
1130 	 * to put it LAST so timer finds oldest requests first.
1131 	 */
1132 	s = splsoftnet();
1133 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1134 
1135 	/* Get send time for nqnfs */
1136 	reqtime = time_second;
1137 
1138 	/*
1139 	 * If backing off another request or avoiding congestion, don't
1140 	 * send this one now but let timer do it. If not timing a request,
1141 	 * do it now.
1142 	 */
1143 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1144 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1145 		nmp->nm_sent < nmp->nm_cwnd)) {
1146 		splx(s);
1147 		if (nmp->nm_soflags & PR_CONNREQUIRED)
1148 			error = nfs_sndlock(&nmp->nm_iflag, rep);
1149 		if (!error) {
1150 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
1151 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp);
1152 			if (nmp->nm_soflags & PR_CONNREQUIRED)
1153 				nfs_sndunlock(&nmp->nm_iflag);
1154 		}
1155 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1156 			nmp->nm_sent += NFS_CWNDSCALE;
1157 			rep->r_flags |= R_SENT;
1158 		}
1159 	} else {
1160 		splx(s);
1161 		rep->r_rtt = -1;
1162 	}
1163 
1164 	/*
1165 	 * Wait for the reply from our send or the timer's.
1166 	 */
1167 	if (!error || error == EPIPE)
1168 		error = nfs_reply(rep, lwp);
1169 
1170 	/*
1171 	 * RPC done, unlink the request.
1172 	 */
1173 	s = splsoftnet();
1174 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1175 	splx(s);
1176 
1177 	/*
1178 	 * Decrement the outstanding request count.
1179 	 */
1180 	if (rep->r_flags & R_SENT) {
1181 		rep->r_flags &= ~R_SENT;	/* paranoia */
1182 		nmp->nm_sent -= NFS_CWNDSCALE;
1183 	}
1184 
1185 	if (rexmitp != NULL) {
1186 		int rexmit;
1187 
1188 		if (nmp->nm_sotype != SOCK_DGRAM)
1189 			rexmit = (rep->r_flags & R_REXMITTED) != 0;
1190 		else
1191 			rexmit = rep->r_rexmit;
1192 		*rexmitp = rexmit;
1193 	}
1194 
1195 	/*
1196 	 * If there was a successful reply and a tprintf msg.
1197 	 * tprintf a response.
1198 	 */
1199 	if (!error && (rep->r_flags & R_TPRINTFMSG))
1200 		nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1201 		    "is alive again");
1202 	mrep = rep->r_mrep;
1203 	md = rep->r_md;
1204 	dpos = rep->r_dpos;
1205 	if (error)
1206 		goto nfsmout;
1207 
1208 	/*
1209 	 * break down the rpc header and check if ok
1210 	 */
1211 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1212 	if (*tl++ == rpc_msgdenied) {
1213 		if (*tl == rpc_mismatch)
1214 			error = EOPNOTSUPP;
1215 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1216 			if (!failed_auth) {
1217 				failed_auth++;
1218 				mheadend->m_next = (struct mbuf *)0;
1219 				m_freem(mrep);
1220 				m_freem(rep->r_mreq);
1221 				goto kerbauth;
1222 			} else
1223 				error = EAUTH;
1224 		} else
1225 			error = EACCES;
1226 		m_freem(mrep);
1227 		goto nfsmout;
1228 	}
1229 
1230 	/*
1231 	 * Grab any Kerberos verifier, otherwise just throw it away.
1232 	 */
1233 	verf_type = fxdr_unsigned(int, *tl++);
1234 	i = fxdr_unsigned(int32_t, *tl);
1235 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1236 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1237 		if (error)
1238 			goto nfsmout;
1239 	} else if (i > 0)
1240 		nfsm_adv(nfsm_rndup(i));
1241 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1242 	/* 0 == ok */
1243 	if (*tl == 0) {
1244 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1245 		if (*tl != 0) {
1246 			error = fxdr_unsigned(int, *tl);
1247 			switch (error) {
1248 			case NFSERR_PERM:
1249 				error = EPERM;
1250 				break;
1251 
1252 			case NFSERR_NOENT:
1253 				error = ENOENT;
1254 				break;
1255 
1256 			case NFSERR_IO:
1257 				error = EIO;
1258 				break;
1259 
1260 			case NFSERR_NXIO:
1261 				error = ENXIO;
1262 				break;
1263 
1264 			case NFSERR_ACCES:
1265 				error = EACCES;
1266 				if (!retry_cred)
1267 					break;
1268 				m_freem(mrep);
1269 				m_freem(rep->r_mreq);
1270 				FREE(rep, M_NFSREQ);
1271 				use_opencred = !use_opencred;
1272 				if (mrest_backup == NULL) {
1273 					/* m_copym failure */
1274 					KASSERT(
1275 					    kauth_cred_getrefcnt(acred) == 1);
1276 					kauth_cred_free(acred);
1277 					return ENOMEM;
1278 				}
1279 				mrest = mrest_backup;
1280 				mrest_backup = NULL;
1281 				cred = origcred;
1282 				error = 0;
1283 				retry_cred = FALSE;
1284 				goto tryagain_cred;
1285 
1286 			case NFSERR_EXIST:
1287 				error = EEXIST;
1288 				break;
1289 
1290 			case NFSERR_XDEV:
1291 				error = EXDEV;
1292 				break;
1293 
1294 			case NFSERR_NODEV:
1295 				error = ENODEV;
1296 				break;
1297 
1298 			case NFSERR_NOTDIR:
1299 				error = ENOTDIR;
1300 				break;
1301 
1302 			case NFSERR_ISDIR:
1303 				error = EISDIR;
1304 				break;
1305 
1306 			case NFSERR_INVAL:
1307 				error = EINVAL;
1308 				break;
1309 
1310 			case NFSERR_FBIG:
1311 				error = EFBIG;
1312 				break;
1313 
1314 			case NFSERR_NOSPC:
1315 				error = ENOSPC;
1316 				break;
1317 
1318 			case NFSERR_ROFS:
1319 				error = EROFS;
1320 				break;
1321 
1322 			case NFSERR_MLINK:
1323 				error = EMLINK;
1324 				break;
1325 
1326 			case NFSERR_TIMEDOUT:
1327 				error = ETIMEDOUT;
1328 				break;
1329 
1330 			case NFSERR_NAMETOL:
1331 				error = ENAMETOOLONG;
1332 				break;
1333 
1334 			case NFSERR_NOTEMPTY:
1335 				error = ENOTEMPTY;
1336 				break;
1337 
1338 			case NFSERR_DQUOT:
1339 				error = EDQUOT;
1340 				break;
1341 
1342 			case NFSERR_STALE:
1343 				/*
1344 				 * If the File Handle was stale, invalidate the
1345 				 * lookup cache, just in case.
1346 				 */
1347 				error = ESTALE;
1348 				cache_purge(NFSTOV(np));
1349 				break;
1350 
1351 			case NFSERR_REMOTE:
1352 				error = EREMOTE;
1353 				break;
1354 
1355 			case NFSERR_WFLUSH:
1356 			case NFSERR_BADHANDLE:
1357 			case NFSERR_NOT_SYNC:
1358 			case NFSERR_BAD_COOKIE:
1359 				error = EINVAL;
1360 				break;
1361 
1362 			case NFSERR_NOTSUPP:
1363 				error = ENOTSUP;
1364 				break;
1365 
1366 			case NFSERR_TOOSMALL:
1367 			case NFSERR_SERVERFAULT:
1368 			case NFSERR_BADTYPE:
1369 				error = EINVAL;
1370 				break;
1371 
1372 			case NFSERR_TRYLATER:
1373 				if ((nmp->nm_flag & NFSMNT_NFSV3) == 0)
1374 					break;
1375 				m_freem(mrep);
1376 				error = 0;
1377 				waituntil = time_second + trylater_delay;
1378 				while (time_second < waituntil)
1379 					(void) tsleep((caddr_t)&lbolt,
1380 						PSOCK, "nqnfstry", 0);
1381 				trylater_delay *= NFS_TRYLATERDELMUL;
1382 				if (trylater_delay > NFS_TRYLATERDELMAX)
1383 					trylater_delay = NFS_TRYLATERDELMAX;
1384 				/*
1385 				 * RFC1813:
1386 				 * The client should wait and then try
1387 				 * the request with a new RPC transaction ID.
1388 				 */
1389 				nfs_renewxid(rep);
1390 				goto tryagain;
1391 
1392 			default:
1393 #ifdef DIAGNOSTIC
1394 				printf("Invalid rpc error code %d\n", error);
1395 #endif
1396 				error = EINVAL;
1397 				break;
1398 			}
1399 
1400 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1401 				*mrp = mrep;
1402 				*mdp = md;
1403 				*dposp = dpos;
1404 				error |= NFSERR_RETERR;
1405 			} else
1406 				m_freem(mrep);
1407 			goto nfsmout;
1408 		}
1409 
1410 		/*
1411 		 * note which credential worked to minimize number of retries.
1412 		 */
1413 		if (use_opencred)
1414 			np->n_flag |= NUSEOPENCRED;
1415 		else
1416 			np->n_flag &= ~NUSEOPENCRED;
1417 
1418 #ifndef NFS_V2_ONLY
1419 		/*
1420 		 * For nqnfs, get any lease in reply
1421 		 */
1422 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1423 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1424 			if (*tl) {
1425 				nqlflag = fxdr_unsigned(int, *tl);
1426 				nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1427 				cachable = fxdr_unsigned(int, *tl++);
1428 				reqtime += fxdr_unsigned(int, *tl++);
1429 				if (reqtime > time_second) {
1430 				    frev = fxdr_hyper(tl);
1431 				    nqnfs_clientlease(nmp, np, nqlflag,
1432 					cachable, reqtime, frev);
1433 				}
1434 			}
1435 		}
1436 #endif
1437 		*mrp = mrep;
1438 		*mdp = md;
1439 		*dposp = dpos;
1440 
1441 		KASSERT(error == 0);
1442 		goto nfsmout;
1443 	}
1444 	m_freem(mrep);
1445 	error = EPROTONOSUPPORT;
1446 nfsmout:
1447 	KASSERT(kauth_cred_getrefcnt(acred) == 1);
1448 	kauth_cred_free(acred);
1449 	m_freem(rep->r_mreq);
1450 	free((caddr_t)rep, M_NFSREQ);
1451 	m_freem(mrest_backup);
1452 	return (error);
1453 }
1454 #endif /* NFS */
1455 
1456 /*
1457  * Generate the rpc reply header
1458  * siz arg. is used to decide if adding a cluster is worthwhile
1459  */
1460 int
1461 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1462 	int siz;
1463 	struct nfsrv_descript *nd;
1464 	struct nfssvc_sock *slp;
1465 	int err;
1466 	int cache;
1467 	u_quad_t *frev;
1468 	struct mbuf **mrq;
1469 	struct mbuf **mbp;
1470 	caddr_t *bposp;
1471 {
1472 	u_int32_t *tl;
1473 	struct mbuf *mreq;
1474 	caddr_t bpos;
1475 	struct mbuf *mb;
1476 
1477 	mreq = m_gethdr(M_WAIT, MT_DATA);
1478 	MCLAIM(mreq, &nfs_mowner);
1479 	mb = mreq;
1480 	/*
1481 	 * If this is a big reply, use a cluster else
1482 	 * try and leave leading space for the lower level headers.
1483 	 */
1484 	siz += RPC_REPLYSIZ;
1485 	if (siz >= max_datalen) {
1486 		m_clget(mreq, M_WAIT);
1487 	} else
1488 		mreq->m_data += max_hdr;
1489 	tl = mtod(mreq, u_int32_t *);
1490 	mreq->m_len = 6 * NFSX_UNSIGNED;
1491 	bpos = ((caddr_t)tl) + mreq->m_len;
1492 	*tl++ = txdr_unsigned(nd->nd_retxid);
1493 	*tl++ = rpc_reply;
1494 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1495 		*tl++ = rpc_msgdenied;
1496 		if (err & NFSERR_AUTHERR) {
1497 			*tl++ = rpc_autherr;
1498 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1499 			mreq->m_len -= NFSX_UNSIGNED;
1500 			bpos -= NFSX_UNSIGNED;
1501 		} else {
1502 			*tl++ = rpc_mismatch;
1503 			*tl++ = txdr_unsigned(RPC_VER2);
1504 			*tl = txdr_unsigned(RPC_VER2);
1505 		}
1506 	} else {
1507 		*tl++ = rpc_msgaccepted;
1508 
1509 		/*
1510 		 * For Kerberos authentication, we must send the nickname
1511 		 * verifier back, otherwise just RPCAUTH_NULL.
1512 		 */
1513 		if (nd->nd_flag & ND_KERBFULL) {
1514 			struct nfsuid *nuidp;
1515 			struct timeval ktvin, ktvout;
1516 
1517 			memset(&ktvout, 0, sizeof ktvout);	/* XXX gcc */
1518 
1519 			LIST_FOREACH(nuidp,
1520 			    NUIDHASH(slp, kauth_cred_geteuid(nd->nd_cr)),
1521 			    nu_hash) {
1522 				if (kauth_cred_geteuid(nuidp->nu_cr) ==
1523 				kauth_cred_geteuid(nd->nd_cr) &&
1524 				    (!nd->nd_nam2 || netaddr_match(
1525 				    NU_NETFAM(nuidp), &nuidp->nu_haddr,
1526 				    nd->nd_nam2)))
1527 					break;
1528 			}
1529 			if (nuidp) {
1530 				ktvin.tv_sec =
1531 				    txdr_unsigned(nuidp->nu_timestamp.tv_sec
1532 					- 1);
1533 				ktvin.tv_usec =
1534 				    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1535 
1536 				/*
1537 				 * Encrypt the timestamp in ecb mode using the
1538 				 * session key.
1539 				 */
1540 #ifdef NFSKERB
1541 				XXX
1542 #endif
1543 
1544 				*tl++ = rpc_auth_kerb;
1545 				*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1546 				*tl = ktvout.tv_sec;
1547 				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1548 				*tl++ = ktvout.tv_usec;
1549 				*tl++ = txdr_unsigned(
1550 				    kauth_cred_geteuid(nuidp->nu_cr));
1551 			} else {
1552 				*tl++ = 0;
1553 				*tl++ = 0;
1554 			}
1555 		} else {
1556 			*tl++ = 0;
1557 			*tl++ = 0;
1558 		}
1559 		switch (err) {
1560 		case EPROGUNAVAIL:
1561 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1562 			break;
1563 		case EPROGMISMATCH:
1564 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1565 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1566 			if (nd->nd_flag & ND_NQNFS) {
1567 				*tl++ = txdr_unsigned(3);
1568 				*tl = txdr_unsigned(3);
1569 			} else {
1570 				*tl++ = txdr_unsigned(2);
1571 				*tl = txdr_unsigned(3);
1572 			}
1573 			break;
1574 		case EPROCUNAVAIL:
1575 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1576 			break;
1577 		case EBADRPC:
1578 			*tl = txdr_unsigned(RPC_GARBAGE);
1579 			break;
1580 		default:
1581 			*tl = 0;
1582 			if (err != NFSERR_RETVOID) {
1583 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1584 				if (err)
1585 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1586 				else
1587 				    *tl = 0;
1588 			}
1589 			break;
1590 		};
1591 	}
1592 
1593 	/*
1594 	 * For nqnfs, piggyback lease as requested.
1595 	 */
1596 	if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1597 		if (nd->nd_flag & ND_LEASE) {
1598 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1599 			*tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1600 			*tl++ = txdr_unsigned(cache);
1601 			*tl++ = txdr_unsigned(nd->nd_duration);
1602 			txdr_hyper(*frev, tl);
1603 		} else {
1604 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1605 			*tl = 0;
1606 		}
1607 	}
1608 	if (mrq != NULL)
1609 		*mrq = mreq;
1610 	*mbp = mb;
1611 	*bposp = bpos;
1612 	if (err != 0 && err != NFSERR_RETVOID)
1613 		nfsstats.srvrpc_errs++;
1614 	return (0);
1615 }
1616 
1617 /*
1618  * Nfs timer routine
1619  * Scan the nfsreq list and retranmit any requests that have timed out
1620  * To avoid retransmission attempts on STREAM sockets (in the future) make
1621  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1622  */
1623 void
1624 nfs_timer(arg)
1625 	void *arg;	/* never used */
1626 {
1627 	struct nfsreq *rep;
1628 	struct mbuf *m;
1629 	struct socket *so;
1630 	struct nfsmount *nmp;
1631 	int timeo;
1632 	int s, error;
1633 #ifdef NFSSERVER
1634 	struct timeval tv;
1635 	struct nfssvc_sock *slp;
1636 	static long lasttime = 0;
1637 	u_quad_t cur_usec;
1638 #endif
1639 
1640 	s = splsoftnet();
1641 	TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
1642 		nmp = rep->r_nmp;
1643 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1644 			continue;
1645 		if (nfs_sigintr(nmp, rep, rep->r_lwp)) {
1646 			rep->r_flags |= R_SOFTTERM;
1647 			continue;
1648 		}
1649 		if (rep->r_rtt >= 0) {
1650 			rep->r_rtt++;
1651 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1652 				timeo = nmp->nm_timeo;
1653 			else
1654 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1655 			if (nmp->nm_timeouts > 0)
1656 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1657 			if (rep->r_rtt <= timeo)
1658 				continue;
1659 			if (nmp->nm_timeouts <
1660 			    (sizeof(nfs_backoff) / sizeof(nfs_backoff[0])))
1661 				nmp->nm_timeouts++;
1662 		}
1663 		/*
1664 		 * Check for server not responding
1665 		 */
1666 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1667 		     rep->r_rexmit > nmp->nm_deadthresh) {
1668 			nfs_msg(rep->r_lwp,
1669 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1670 			    "not responding");
1671 			rep->r_flags |= R_TPRINTFMSG;
1672 		}
1673 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1674 			nfsstats.rpctimeouts++;
1675 			rep->r_flags |= R_SOFTTERM;
1676 			continue;
1677 		}
1678 		if (nmp->nm_sotype != SOCK_DGRAM) {
1679 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1680 				rep->r_rexmit = NFS_MAXREXMIT;
1681 			continue;
1682 		}
1683 		if ((so = nmp->nm_so) == NULL)
1684 			continue;
1685 
1686 		/*
1687 		 * If there is enough space and the window allows..
1688 		 *	Resend it
1689 		 * Set r_rtt to -1 in case we fail to send it now.
1690 		 */
1691 		rep->r_rtt = -1;
1692 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1693 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1694 		    (rep->r_flags & R_SENT) ||
1695 		    nmp->nm_sent < nmp->nm_cwnd) &&
1696 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1697 		        if (so->so_state & SS_ISCONNECTED)
1698 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1699 			    (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0);
1700 			else
1701 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1702 			    nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0);
1703 			if (error) {
1704 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1705 #ifdef DEBUG
1706 					printf("nfs_timer: ignoring error %d\n",
1707 						error);
1708 #endif
1709 					so->so_error = 0;
1710 				}
1711 			} else {
1712 				/*
1713 				 * Iff first send, start timing
1714 				 * else turn timing off, backoff timer
1715 				 * and divide congestion window by 2.
1716 				 */
1717 				if (rep->r_flags & R_SENT) {
1718 					rep->r_flags &= ~R_TIMING;
1719 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1720 						rep->r_rexmit = NFS_MAXREXMIT;
1721 					nmp->nm_cwnd >>= 1;
1722 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1723 						nmp->nm_cwnd = NFS_CWNDSCALE;
1724 					nfsstats.rpcretries++;
1725 				} else {
1726 					rep->r_flags |= R_SENT;
1727 					nmp->nm_sent += NFS_CWNDSCALE;
1728 				}
1729 				rep->r_rtt = 0;
1730 			}
1731 		}
1732 	}
1733 
1734 #ifdef NFSSERVER
1735 	/*
1736 	 * Call the nqnfs server timer once a second to handle leases.
1737 	 */
1738 	if (lasttime != time_second) {
1739 		lasttime = time_second;
1740 		nqnfs_serverd();
1741 	}
1742 
1743 	/*
1744 	 * Scan the write gathering queues for writes that need to be
1745 	 * completed now.
1746 	 */
1747 	getmicrotime(&tv);
1748 	cur_usec = (u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec;
1749 	TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
1750 	    if (LIST_FIRST(&slp->ns_tq) &&
1751 		LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec)
1752 		nfsrv_wakenfsd(slp);
1753 	}
1754 #endif /* NFSSERVER */
1755 	splx(s);
1756 	callout_schedule(&nfs_timer_ch, nfs_ticks);
1757 }
1758 
1759 /*ARGSUSED*/
1760 void
1761 nfs_exit(p, v)
1762 	struct proc *p;
1763 	void *v;
1764 {
1765 	struct nfsreq *rp;
1766 	int s = splsoftnet();
1767 
1768 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
1769 		if (rp->r_lwp && rp->r_lwp->l_proc == p)
1770 			TAILQ_REMOVE(&nfs_reqq, rp, r_chain);
1771 	}
1772 	splx(s);
1773 }
1774 
1775 /*
1776  * Test for a termination condition pending on the process.
1777  * This is used for NFSMNT_INT mounts.
1778  */
1779 int
1780 nfs_sigintr(nmp, rep, l)
1781 	struct nfsmount *nmp;
1782 	struct nfsreq *rep;
1783 	struct lwp *l;
1784 {
1785 	sigset_t ss;
1786 
1787 	if (rep && (rep->r_flags & R_SOFTTERM))
1788 		return (EINTR);
1789 	if (!(nmp->nm_flag & NFSMNT_INT))
1790 		return (0);
1791 	if (l) {
1792 		sigpending1(l->l_proc, &ss);
1793 #if 0
1794 		sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss);
1795 #endif
1796 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1797 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1798 		    sigismember(&ss, SIGQUIT))
1799 			return (EINTR);
1800 	}
1801 	return (0);
1802 }
1803 
1804 /*
1805  * Lock a socket against others.
1806  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1807  * and also to avoid race conditions between the processes with nfs requests
1808  * in progress when a reconnect is necessary.
1809  */
1810 int
1811 nfs_sndlock(flagp, rep)
1812 	int *flagp;
1813 	struct nfsreq *rep;
1814 {
1815 	struct lwp *l;
1816 	int slpflag = 0, slptimeo = 0;
1817 
1818 	if (rep) {
1819 		l = rep->r_lwp;
1820 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1821 			slpflag = PCATCH;
1822 	} else
1823 		l = (struct lwp *)0;
1824 	while (*flagp & NFSMNT_SNDLOCK) {
1825 		if (rep && nfs_sigintr(rep->r_nmp, rep, l))
1826 			return (EINTR);
1827 		*flagp |= NFSMNT_WANTSND;
1828 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1829 			slptimeo);
1830 		if (slpflag == PCATCH) {
1831 			slpflag = 0;
1832 			slptimeo = 2 * hz;
1833 		}
1834 	}
1835 	*flagp |= NFSMNT_SNDLOCK;
1836 	return (0);
1837 }
1838 
1839 /*
1840  * Unlock the stream socket for others.
1841  */
1842 void
1843 nfs_sndunlock(flagp)
1844 	int *flagp;
1845 {
1846 
1847 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1848 		panic("nfs sndunlock");
1849 	*flagp &= ~NFSMNT_SNDLOCK;
1850 	if (*flagp & NFSMNT_WANTSND) {
1851 		*flagp &= ~NFSMNT_WANTSND;
1852 		wakeup((caddr_t)flagp);
1853 	}
1854 }
1855 
1856 int
1857 nfs_rcvlock(rep)
1858 	struct nfsreq *rep;
1859 {
1860 	struct nfsmount *nmp = rep->r_nmp;
1861 	int *flagp = &nmp->nm_iflag;
1862 	int slpflag, slptimeo = 0;
1863 	int error = 0;
1864 
1865 	if (*flagp & NFSMNT_DISMNT)
1866 		return EIO;
1867 
1868 	if (*flagp & NFSMNT_INT)
1869 		slpflag = PCATCH;
1870 	else
1871 		slpflag = 0;
1872 	simple_lock(&nmp->nm_slock);
1873 	while (*flagp & NFSMNT_RCVLOCK) {
1874 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) {
1875 			error = EINTR;
1876 			goto quit;
1877 		}
1878 		*flagp |= NFSMNT_WANTRCV;
1879 		nmp->nm_waiters++;
1880 		(void) ltsleep(flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1881 			slptimeo, &nmp->nm_slock);
1882 		nmp->nm_waiters--;
1883 		if (*flagp & NFSMNT_DISMNT) {
1884 			wakeup(&nmp->nm_waiters);
1885 			error = EIO;
1886 			goto quit;
1887 		}
1888 		/* If our reply was received while we were sleeping,
1889 		 * then just return without taking the lock to avoid a
1890 		 * situation where a single iod could 'capture' the
1891 		 * receive lock.
1892 		 */
1893 		if (rep->r_mrep != NULL) {
1894 			error = EALREADY;
1895 			goto quit;
1896 		}
1897 		if (slpflag == PCATCH) {
1898 			slpflag = 0;
1899 			slptimeo = 2 * hz;
1900 		}
1901 	}
1902 	*flagp |= NFSMNT_RCVLOCK;
1903 quit:
1904 	simple_unlock(&nmp->nm_slock);
1905 	return error;
1906 }
1907 
1908 /*
1909  * Unlock the stream socket for others.
1910  */
1911 void
1912 nfs_rcvunlock(nmp)
1913 	struct nfsmount *nmp;
1914 {
1915 	int *flagp = &nmp->nm_iflag;
1916 
1917 	simple_lock(&nmp->nm_slock);
1918 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1919 		panic("nfs rcvunlock");
1920 	*flagp &= ~NFSMNT_RCVLOCK;
1921 	if (*flagp & NFSMNT_WANTRCV) {
1922 		*flagp &= ~NFSMNT_WANTRCV;
1923 		wakeup((caddr_t)flagp);
1924 	}
1925 	simple_unlock(&nmp->nm_slock);
1926 }
1927 
1928 /*
1929  * Parse an RPC request
1930  * - verify it
1931  * - allocate and fill in the cred.
1932  */
1933 int
1934 nfs_getreq(nd, nfsd, has_header)
1935 	struct nfsrv_descript *nd;
1936 	struct nfsd *nfsd;
1937 	int has_header;
1938 {
1939 	int len, i;
1940 	u_int32_t *tl;
1941 	int32_t t1;
1942 	struct uio uio;
1943 	struct iovec iov;
1944 	caddr_t dpos, cp2, cp;
1945 	u_int32_t nfsvers, auth_type;
1946 	uid_t nickuid;
1947 	int error = 0, nqnfs = 0, ticklen;
1948 	struct mbuf *mrep, *md;
1949 	struct nfsuid *nuidp;
1950 	struct timeval tvin, tvout;
1951 
1952 	memset(&tvout, 0, sizeof tvout);	/* XXX gcc */
1953 
1954 	KASSERT(nd->nd_cr == NULL);
1955 	mrep = nd->nd_mrep;
1956 	md = nd->nd_md;
1957 	dpos = nd->nd_dpos;
1958 	if (has_header) {
1959 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1960 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1961 		if (*tl++ != rpc_call) {
1962 			m_freem(mrep);
1963 			return (EBADRPC);
1964 		}
1965 	} else
1966 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1967 	nd->nd_repstat = 0;
1968 	nd->nd_flag = 0;
1969 	if (*tl++ != rpc_vers) {
1970 		nd->nd_repstat = ERPCMISMATCH;
1971 		nd->nd_procnum = NFSPROC_NOOP;
1972 		return (0);
1973 	}
1974 	if (*tl != nfs_prog) {
1975 		if (*tl == nqnfs_prog)
1976 			nqnfs++;
1977 		else {
1978 			nd->nd_repstat = EPROGUNAVAIL;
1979 			nd->nd_procnum = NFSPROC_NOOP;
1980 			return (0);
1981 		}
1982 	}
1983 	tl++;
1984 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1985 	if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1986 		(nfsvers != NQNFS_VER3 && nqnfs)) {
1987 		nd->nd_repstat = EPROGMISMATCH;
1988 		nd->nd_procnum = NFSPROC_NOOP;
1989 		return (0);
1990 	}
1991 	if (nqnfs)
1992 		nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1993 	else if (nfsvers == NFS_VER3)
1994 		nd->nd_flag = ND_NFSV3;
1995 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1996 	if (nd->nd_procnum == NFSPROC_NULL)
1997 		return (0);
1998 	if (nd->nd_procnum >= NFS_NPROCS ||
1999 		(!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
2000 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
2001 		nd->nd_repstat = EPROCUNAVAIL;
2002 		nd->nd_procnum = NFSPROC_NOOP;
2003 		return (0);
2004 	}
2005 	if ((nd->nd_flag & ND_NFSV3) == 0)
2006 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
2007 	auth_type = *tl++;
2008 	len = fxdr_unsigned(int, *tl++);
2009 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
2010 		m_freem(mrep);
2011 		return (EBADRPC);
2012 	}
2013 
2014 	nd->nd_flag &= ~ND_KERBAUTH;
2015 	/*
2016 	 * Handle auth_unix or auth_kerb.
2017 	 */
2018 	if (auth_type == rpc_auth_unix) {
2019 		uid_t uid;
2020 		gid_t gid, *grbuf;
2021 
2022 		nd->nd_cr = kauth_cred_alloc();
2023 		len = fxdr_unsigned(int, *++tl);
2024 		if (len < 0 || len > NFS_MAXNAMLEN) {
2025 			m_freem(mrep);
2026 			error = EBADRPC;
2027 			goto errout;
2028 		}
2029 		nfsm_adv(nfsm_rndup(len));
2030 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2031 
2032 		uid = fxdr_unsigned(uid_t, *tl++);
2033 		gid = fxdr_unsigned(gid_t, *tl++);
2034 		kauth_cred_setuid(nd->nd_cr, uid);
2035 		kauth_cred_setgid(nd->nd_cr, gid);
2036 		kauth_cred_seteuid(nd->nd_cr, uid);
2037 		kauth_cred_setsvuid(nd->nd_cr, gid);
2038 		kauth_cred_setegid(nd->nd_cr, uid);
2039 		kauth_cred_setsvgid(nd->nd_cr, gid);
2040 
2041 		len = fxdr_unsigned(int, *tl);
2042 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
2043 			m_freem(mrep);
2044 			error = EBADRPC;
2045 			goto errout;
2046 		}
2047 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
2048 
2049 		grbuf = malloc(len * sizeof(gid_t), M_TEMP, M_WAITOK);
2050 		for (i = 0; i < len; i++) {
2051 			if (i < NGROUPS) /* XXX elad */
2052 				grbuf[i] = fxdr_unsigned(gid_t, *tl++);
2053 			else
2054 				tl++;
2055 		}
2056 		kauth_cred_setgroups(nd->nd_cr, grbuf, min(len, NGROUPS), -1);
2057 		free(grbuf, M_TEMP);
2058 
2059 		len = fxdr_unsigned(int, *++tl);
2060 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
2061 			m_freem(mrep);
2062 			error = EBADRPC;
2063 			goto errout;
2064 		}
2065 		if (len > 0)
2066 			nfsm_adv(nfsm_rndup(len));
2067 	} else if (auth_type == rpc_auth_kerb) {
2068 		switch (fxdr_unsigned(int, *tl++)) {
2069 		case RPCAKN_FULLNAME:
2070 			ticklen = fxdr_unsigned(int, *tl);
2071 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
2072 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
2073 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
2074 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
2075 				m_freem(mrep);
2076 				error = EBADRPC;
2077 				goto errout;
2078 			}
2079 			uio.uio_offset = 0;
2080 			uio.uio_iov = &iov;
2081 			uio.uio_iovcnt = 1;
2082 			UIO_SETUP_SYSSPACE(&uio);
2083 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
2084 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
2085 			nfsm_mtouio(&uio, uio.uio_resid);
2086 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2087 			if (*tl++ != rpc_auth_kerb ||
2088 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
2089 				printf("Bad kerb verifier\n");
2090 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2091 				nd->nd_procnum = NFSPROC_NOOP;
2092 				return (0);
2093 			}
2094 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
2095 			tl = (u_int32_t *)cp;
2096 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
2097 				printf("Not fullname kerb verifier\n");
2098 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2099 				nd->nd_procnum = NFSPROC_NOOP;
2100 				return (0);
2101 			}
2102 			cp += NFSX_UNSIGNED;
2103 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
2104 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
2105 			nd->nd_flag |= ND_KERBFULL;
2106 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
2107 			break;
2108 		case RPCAKN_NICKNAME:
2109 			if (len != 2 * NFSX_UNSIGNED) {
2110 				printf("Kerb nickname short\n");
2111 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
2112 				nd->nd_procnum = NFSPROC_NOOP;
2113 				return (0);
2114 			}
2115 			nickuid = fxdr_unsigned(uid_t, *tl);
2116 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2117 			if (*tl++ != rpc_auth_kerb ||
2118 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
2119 				printf("Kerb nick verifier bad\n");
2120 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2121 				nd->nd_procnum = NFSPROC_NOOP;
2122 				return (0);
2123 			}
2124 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2125 			tvin.tv_sec = *tl++;
2126 			tvin.tv_usec = *tl;
2127 
2128 			LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid),
2129 			    nu_hash) {
2130 				if (kauth_cred_geteuid(nuidp->nu_cr) == nickuid &&
2131 				    (!nd->nd_nam2 ||
2132 				     netaddr_match(NU_NETFAM(nuidp),
2133 				      &nuidp->nu_haddr, nd->nd_nam2)))
2134 					break;
2135 			}
2136 			if (!nuidp) {
2137 				nd->nd_repstat =
2138 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
2139 				nd->nd_procnum = NFSPROC_NOOP;
2140 				return (0);
2141 			}
2142 
2143 			/*
2144 			 * Now, decrypt the timestamp using the session key
2145 			 * and validate it.
2146 			 */
2147 #ifdef NFSKERB
2148 			XXX
2149 #endif
2150 
2151 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
2152 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
2153 			if (nuidp->nu_expire < time_second ||
2154 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
2155 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
2156 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
2157 				nuidp->nu_expire = 0;
2158 				nd->nd_repstat =
2159 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
2160 				nd->nd_procnum = NFSPROC_NOOP;
2161 				return (0);
2162 			}
2163 			kauth_cred_hold(nuidp->nu_cr);
2164 			nd->nd_cr = nuidp->nu_cr;
2165 			nd->nd_flag |= ND_KERBNICK;
2166 		}
2167 	} else {
2168 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
2169 		nd->nd_procnum = NFSPROC_NOOP;
2170 		return (0);
2171 	}
2172 
2173 	/*
2174 	 * For nqnfs, get piggybacked lease request.
2175 	 */
2176 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
2177 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2178 		nd->nd_flag |= fxdr_unsigned(int, *tl);
2179 		if (nd->nd_flag & ND_LEASE) {
2180 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2181 			nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
2182 		} else
2183 			nd->nd_duration = NQ_MINLEASE;
2184 	} else
2185 		nd->nd_duration = NQ_MINLEASE;
2186 	nd->nd_md = md;
2187 	nd->nd_dpos = dpos;
2188 	KASSERT((nd->nd_cr == NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) != 0)
2189 	     || (nd->nd_cr != NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) == 0));
2190 	return (0);
2191 nfsmout:
2192 errout:
2193 	KASSERT(error != 0);
2194 	if (nd->nd_cr != NULL) {
2195 		kauth_cred_free(nd->nd_cr);
2196 		nd->nd_cr = NULL;
2197 	}
2198 	return (error);
2199 }
2200 
2201 int
2202 nfs_msg(l, server, msg)
2203 	struct lwp *l;
2204 	const char *server, *msg;
2205 {
2206 	tpr_t tpr;
2207 
2208 	if (l)
2209 		tpr = tprintf_open(l->l_proc);
2210 	else
2211 		tpr = NULL;
2212 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
2213 	tprintf_close(tpr);
2214 	return (0);
2215 }
2216 
2217 #ifdef NFSSERVER
2218 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
2219 				    struct nfssvc_sock *, struct lwp *,
2220 				    struct mbuf **)) = {
2221 	nfsrv_null,
2222 	nfsrv_getattr,
2223 	nfsrv_setattr,
2224 	nfsrv_lookup,
2225 	nfsrv3_access,
2226 	nfsrv_readlink,
2227 	nfsrv_read,
2228 	nfsrv_write,
2229 	nfsrv_create,
2230 	nfsrv_mkdir,
2231 	nfsrv_symlink,
2232 	nfsrv_mknod,
2233 	nfsrv_remove,
2234 	nfsrv_rmdir,
2235 	nfsrv_rename,
2236 	nfsrv_link,
2237 	nfsrv_readdir,
2238 	nfsrv_readdirplus,
2239 	nfsrv_statfs,
2240 	nfsrv_fsinfo,
2241 	nfsrv_pathconf,
2242 	nfsrv_commit,
2243 	nqnfsrv_getlease,
2244 	nqnfsrv_vacated,
2245 	nfsrv_noop,
2246 	nfsrv_noop
2247 };
2248 
2249 /*
2250  * Socket upcall routine for the nfsd sockets.
2251  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
2252  * Essentially do as much as possible non-blocking, else punt and it will
2253  * be called with M_WAIT from an nfsd.
2254  */
2255 void
2256 nfsrv_rcv(so, arg, waitflag)
2257 	struct socket *so;
2258 	caddr_t arg;
2259 	int waitflag;
2260 {
2261 	struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
2262 	struct mbuf *m;
2263 	struct mbuf *mp, *nam;
2264 	struct uio auio;
2265 	int flags, error;
2266 	int setflags = 0;
2267 
2268 	error = nfsdsock_lock(slp, (waitflag != M_DONTWAIT));
2269 	if (error) {
2270 		setflags |= SLP_NEEDQ;
2271 		goto dorecs_unlocked;
2272 	}
2273 
2274 	KASSERT(so == slp->ns_so);
2275 #define NFS_TEST_HEAVY
2276 #ifdef NFS_TEST_HEAVY
2277 	/*
2278 	 * Define this to test for nfsds handling this under heavy load.
2279 	 *
2280 	 * XXX it isn't safe to call so_receive from so_upcall context.
2281 	 */
2282 	if (waitflag == M_DONTWAIT) {
2283 		setflags |= SLP_NEEDQ;
2284 		goto dorecs;
2285 	}
2286 #endif
2287 	simple_lock(&slp->ns_lock);
2288 	slp->ns_flag &= ~SLP_NEEDQ;
2289 	simple_unlock(&slp->ns_lock);
2290 	if (so->so_type == SOCK_STREAM) {
2291 #ifndef NFS_TEST_HEAVY
2292 		/*
2293 		 * If there are already records on the queue, defer soreceive()
2294 		 * to an nfsd so that there is feedback to the TCP layer that
2295 		 * the nfs servers are heavily loaded.
2296 		 */
2297 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
2298 			setflags |= SLP_NEEDQ;
2299 			goto dorecs;
2300 		}
2301 #endif
2302 
2303 		/*
2304 		 * Do soreceive().
2305 		 */
2306 		auio.uio_resid = 1000000000;
2307 		/* not need to setup uio_vmspace */
2308 		flags = MSG_DONTWAIT;
2309 		error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, &flags);
2310 		if (error || mp == NULL) {
2311 			if (error == EWOULDBLOCK)
2312 				setflags |= SLP_NEEDQ;
2313 			else
2314 				setflags |= SLP_DISCONN;
2315 			goto dorecs;
2316 		}
2317 		m = mp;
2318 		if (slp->ns_rawend) {
2319 			slp->ns_rawend->m_next = m;
2320 			slp->ns_cc += 1000000000 - auio.uio_resid;
2321 		} else {
2322 			slp->ns_raw = m;
2323 			slp->ns_cc = 1000000000 - auio.uio_resid;
2324 		}
2325 		while (m->m_next)
2326 			m = m->m_next;
2327 		slp->ns_rawend = m;
2328 
2329 		/*
2330 		 * Now try and parse record(s) out of the raw stream data.
2331 		 */
2332 		error = nfsrv_getstream(slp, waitflag);
2333 		if (error) {
2334 			if (error == EPERM)
2335 				setflags |= SLP_DISCONN;
2336 			else
2337 				setflags |= SLP_NEEDQ;
2338 		}
2339 	} else {
2340 		do {
2341 			auio.uio_resid = 1000000000;
2342 			/* not need to setup uio_vmspace */
2343 			flags = MSG_DONTWAIT;
2344 			error = (*so->so_receive)(so, &nam, &auio, &mp, NULL,
2345 			    &flags);
2346 			if (mp) {
2347 				if (nam) {
2348 					m = nam;
2349 					m->m_next = mp;
2350 				} else
2351 					m = mp;
2352 				if (slp->ns_recend)
2353 					slp->ns_recend->m_nextpkt = m;
2354 				else
2355 					slp->ns_rec = m;
2356 				slp->ns_recend = m;
2357 				m->m_nextpkt = (struct mbuf *)0;
2358 			}
2359 			if (error) {
2360 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2361 				    && error != EWOULDBLOCK) {
2362 					setflags |= SLP_DISCONN;
2363 					goto dorecs;
2364 				}
2365 			}
2366 		} while (mp);
2367 	}
2368 dorecs:
2369 	nfsdsock_unlock(slp);
2370 
2371 dorecs_unlocked:
2372 	/*
2373 	 * Now try and process the request records, non-blocking.
2374 	 */
2375 	if (setflags) {
2376 		simple_lock(&slp->ns_lock);
2377 		slp->ns_flag |= setflags;
2378 		simple_unlock(&slp->ns_lock);
2379 	}
2380 	if (waitflag == M_DONTWAIT &&
2381 	    (slp->ns_rec || (slp->ns_flag & (SLP_DISCONN | SLP_NEEDQ)) != 0)) {
2382 		nfsrv_wakenfsd(slp);
2383 	}
2384 }
2385 
2386 int
2387 nfsdsock_lock(struct nfssvc_sock *slp, boolean_t waitok)
2388 {
2389 
2390 	simple_lock(&slp->ns_lock);
2391 	while ((slp->ns_flag & (SLP_BUSY|SLP_VALID)) == SLP_BUSY) {
2392 		if (!waitok) {
2393 			simple_unlock(&slp->ns_lock);
2394 			return EWOULDBLOCK;
2395 		}
2396 		slp->ns_flag |= SLP_WANT;
2397 		ltsleep(&slp->ns_flag, PSOCK, "nslock", 0, &slp->ns_lock);
2398 	}
2399 	if ((slp->ns_flag & SLP_VALID) == 0) {
2400 		simple_unlock(&slp->ns_lock);
2401 		return EINVAL;
2402 	}
2403 	slp->ns_flag |= SLP_BUSY;
2404 	simple_unlock(&slp->ns_lock);
2405 
2406 	return 0;
2407 }
2408 
2409 void
2410 nfsdsock_unlock(struct nfssvc_sock *slp)
2411 {
2412 
2413 	KASSERT((slp->ns_flag & SLP_BUSY) != 0);
2414 
2415 	simple_lock(&slp->ns_lock);
2416 	if ((slp->ns_flag & SLP_WANT) != 0) {
2417 		wakeup(&slp->ns_flag);
2418 	}
2419 	slp->ns_flag &= ~(SLP_BUSY|SLP_WANT);
2420 	simple_unlock(&slp->ns_lock);
2421 }
2422 
2423 int
2424 nfsdsock_drain(struct nfssvc_sock *slp)
2425 {
2426 	int error = 0;
2427 
2428 	simple_lock(&slp->ns_lock);
2429 	if ((slp->ns_flag & SLP_VALID) == 0) {
2430 		error = EINVAL;
2431 		goto done;
2432 	}
2433 	slp->ns_flag &= ~SLP_VALID;
2434 	while ((slp->ns_flag & SLP_BUSY) != 0) {
2435 		slp->ns_flag |= SLP_WANT;
2436 		ltsleep(&slp->ns_flag, PSOCK, "nsdrain", 0, &slp->ns_lock);
2437 	}
2438 done:
2439 	simple_unlock(&slp->ns_lock);
2440 
2441 	return error;
2442 }
2443 
2444 /*
2445  * Try and extract an RPC request from the mbuf data list received on a
2446  * stream socket. The "waitflag" argument indicates whether or not it
2447  * can sleep.
2448  */
2449 int
2450 nfsrv_getstream(slp, waitflag)
2451 	struct nfssvc_sock *slp;
2452 	int waitflag;
2453 {
2454 	struct mbuf *m, **mpp;
2455 	struct mbuf *recm;
2456 	u_int32_t recmark;
2457 	int error = 0;
2458 
2459 	for (;;) {
2460 		if (slp->ns_reclen == 0) {
2461 			if (slp->ns_cc < NFSX_UNSIGNED) {
2462 				break;
2463 			}
2464 			m = slp->ns_raw;
2465 			m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark);
2466 			m_adj(m, NFSX_UNSIGNED);
2467 			slp->ns_cc -= NFSX_UNSIGNED;
2468 			recmark = ntohl(recmark);
2469 			slp->ns_reclen = recmark & ~0x80000000;
2470 			if (recmark & 0x80000000)
2471 				slp->ns_flag |= SLP_LASTFRAG;
2472 			else
2473 				slp->ns_flag &= ~SLP_LASTFRAG;
2474 			if (slp->ns_reclen > NFS_MAXPACKET) {
2475 				error = EPERM;
2476 				break;
2477 			}
2478 		}
2479 
2480 		/*
2481 		 * Now get the record part.
2482 		 *
2483 		 * Note that slp->ns_reclen may be 0.  Linux sometimes
2484 		 * generates 0-length records.
2485 		 */
2486 		if (slp->ns_cc == slp->ns_reclen) {
2487 			recm = slp->ns_raw;
2488 			slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2489 			slp->ns_cc = slp->ns_reclen = 0;
2490 		} else if (slp->ns_cc > slp->ns_reclen) {
2491 			recm = slp->ns_raw;
2492 			m = m_split(recm, slp->ns_reclen, waitflag);
2493 			if (m == NULL) {
2494 				error = EWOULDBLOCK;
2495 				break;
2496 			}
2497 			m_claimm(recm, &nfs_mowner);
2498 			slp->ns_raw = m;
2499 			if (m->m_next == NULL)
2500 				slp->ns_rawend = m;
2501 			slp->ns_cc -= slp->ns_reclen;
2502 			slp->ns_reclen = 0;
2503 		} else {
2504 			break;
2505 		}
2506 
2507 		/*
2508 		 * Accumulate the fragments into a record.
2509 		 */
2510 		mpp = &slp->ns_frag;
2511 		while (*mpp)
2512 			mpp = &((*mpp)->m_next);
2513 		*mpp = recm;
2514 		if (slp->ns_flag & SLP_LASTFRAG) {
2515 			if (slp->ns_recend)
2516 				slp->ns_recend->m_nextpkt = slp->ns_frag;
2517 			else
2518 				slp->ns_rec = slp->ns_frag;
2519 			slp->ns_recend = slp->ns_frag;
2520 			slp->ns_frag = (struct mbuf *)0;
2521 		}
2522 	}
2523 
2524 	return error;
2525 }
2526 
2527 /*
2528  * Parse an RPC header.
2529  */
2530 int
2531 nfsrv_dorec(slp, nfsd, ndp)
2532 	struct nfssvc_sock *slp;
2533 	struct nfsd *nfsd;
2534 	struct nfsrv_descript **ndp;
2535 {
2536 	struct mbuf *m, *nam;
2537 	struct nfsrv_descript *nd;
2538 	int error;
2539 
2540 	*ndp = NULL;
2541 
2542 	if (nfsdsock_lock(slp, TRUE)) {
2543 		return ENOBUFS;
2544 	}
2545 	m = slp->ns_rec;
2546 	if (m == NULL) {
2547 		nfsdsock_unlock(slp);
2548 		return ENOBUFS;
2549 	}
2550 	slp->ns_rec = m->m_nextpkt;
2551 	if (slp->ns_rec)
2552 		m->m_nextpkt = NULL;
2553 	else
2554 		slp->ns_recend = NULL;
2555 	nfsdsock_unlock(slp);
2556 
2557 	if (m->m_type == MT_SONAME) {
2558 		nam = m;
2559 		m = m->m_next;
2560 		nam->m_next = NULL;
2561 	} else
2562 		nam = NULL;
2563 	nd = nfsdreq_alloc();
2564 	nd->nd_md = nd->nd_mrep = m;
2565 	nd->nd_nam2 = nam;
2566 	nd->nd_dpos = mtod(m, caddr_t);
2567 	error = nfs_getreq(nd, nfsd, TRUE);
2568 	if (error) {
2569 		m_freem(nam);
2570 		nfsdreq_free(nd);
2571 		return (error);
2572 	}
2573 	*ndp = nd;
2574 	nfsd->nfsd_nd = nd;
2575 	return (0);
2576 }
2577 
2578 /*
2579  * Search for a sleeping nfsd and wake it up.
2580  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2581  * running nfsds will go look for the work in the nfssvc_sock list.
2582  */
2583 void
2584 nfsrv_wakenfsd(slp)
2585 	struct nfssvc_sock *slp;
2586 {
2587 	struct nfsd *nd;
2588 
2589 	if ((slp->ns_flag & SLP_VALID) == 0)
2590 		return;
2591 	simple_lock(&nfsd_slock);
2592 	if (slp->ns_flag & SLP_DOREC) {
2593 		simple_unlock(&nfsd_slock);
2594 		return;
2595 	}
2596 	nd = SLIST_FIRST(&nfsd_idle_head);
2597 	if (nd) {
2598 		SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle);
2599 		simple_unlock(&nfsd_slock);
2600 
2601 		if (nd->nfsd_slp)
2602 			panic("nfsd wakeup");
2603 		slp->ns_sref++;
2604 		nd->nfsd_slp = slp;
2605 		wakeup(nd);
2606 		return;
2607 	}
2608 	slp->ns_flag |= SLP_DOREC;
2609 	nfsd_head_flag |= NFSD_CHECKSLP;
2610 	TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending);
2611 	simple_unlock(&nfsd_slock);
2612 }
2613 
2614 int
2615 nfsdsock_sendreply(struct nfssvc_sock *slp, struct nfsrv_descript *nd)
2616 {
2617 	int error;
2618 
2619 	if (nd->nd_mrep != NULL) {
2620 		m_freem(nd->nd_mrep);
2621 		nd->nd_mrep = NULL;
2622 	}
2623 
2624 	simple_lock(&slp->ns_lock);
2625 	if ((slp->ns_flag & SLP_SENDING) != 0) {
2626 		SIMPLEQ_INSERT_TAIL(&slp->ns_sendq, nd, nd_sendq);
2627 		simple_unlock(&slp->ns_lock);
2628 		return 0;
2629 	}
2630 	KASSERT(SIMPLEQ_EMPTY(&slp->ns_sendq));
2631 	slp->ns_flag |= SLP_SENDING;
2632 	simple_unlock(&slp->ns_lock);
2633 
2634 again:
2635 	error = nfs_send(slp->ns_so, nd->nd_nam2, nd->nd_mreq, NULL, curlwp);
2636 	if (nd->nd_nam2) {
2637 		m_free(nd->nd_nam2);
2638 	}
2639 	nfsdreq_free(nd);
2640 
2641 	simple_lock(&slp->ns_lock);
2642 	KASSERT((slp->ns_flag & SLP_SENDING) != 0);
2643 	nd = SIMPLEQ_FIRST(&slp->ns_sendq);
2644 	if (nd != NULL) {
2645 		SIMPLEQ_REMOVE_HEAD(&slp->ns_sendq, nd_sendq);
2646 		simple_unlock(&slp->ns_lock);
2647 		goto again;
2648 	}
2649 	slp->ns_flag &= ~SLP_SENDING;
2650 	simple_unlock(&slp->ns_lock);
2651 
2652 	return error;
2653 }
2654 #endif /* NFSSERVER */
2655 
2656 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY))
2657 static struct pool nfs_srvdesc_pool;
2658 
2659 void
2660 nfsdreq_init(void)
2661 {
2662 
2663 	pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript),
2664 	    0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr);
2665 }
2666 
2667 struct nfsrv_descript *
2668 nfsdreq_alloc(void)
2669 {
2670 	struct nfsrv_descript *nd;
2671 
2672 	nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK);
2673 	nd->nd_cr = NULL;
2674 	return nd;
2675 }
2676 
2677 void
2678 nfsdreq_free(struct nfsrv_descript *nd)
2679 {
2680 	kauth_cred_t cr;
2681 
2682 	cr = nd->nd_cr;
2683 	if (cr != NULL) {
2684 		KASSERT(kauth_cred_getrefcnt(cr) == 1);
2685 		kauth_cred_free(cr);
2686 	}
2687 	pool_put(&nfs_srvdesc_pool, nd);
2688 }
2689 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */
2690