xref: /netbsd-src/sys/nfs/nfs_socket.c (revision d9ac053afa92f3d7a493f791e10558f738234217)
1 /*	$NetBSD: nfs_socket.c,v 1.135 2006/06/07 22:34:17 kardel 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 1.135 2006/06/07 22:34:17 kardel 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 			case NFSERR_STALEWRITEVERF:
1393 				error = EINVAL;
1394 				break;
1395 
1396 			default:
1397 #ifdef DIAGNOSTIC
1398 				printf("Invalid rpc error code %d\n", error);
1399 #endif
1400 				error = EINVAL;
1401 				break;
1402 			}
1403 
1404 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1405 				*mrp = mrep;
1406 				*mdp = md;
1407 				*dposp = dpos;
1408 				error |= NFSERR_RETERR;
1409 			} else
1410 				m_freem(mrep);
1411 			goto nfsmout;
1412 		}
1413 
1414 		/*
1415 		 * note which credential worked to minimize number of retries.
1416 		 */
1417 		if (use_opencred)
1418 			np->n_flag |= NUSEOPENCRED;
1419 		else
1420 			np->n_flag &= ~NUSEOPENCRED;
1421 
1422 #ifndef NFS_V2_ONLY
1423 		/*
1424 		 * For nqnfs, get any lease in reply
1425 		 */
1426 		if (nmp->nm_flag & NFSMNT_NQNFS) {
1427 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1428 			if (*tl) {
1429 				nqlflag = fxdr_unsigned(int, *tl);
1430 				nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1431 				cachable = fxdr_unsigned(int, *tl++);
1432 				reqtime += fxdr_unsigned(int, *tl++);
1433 				if (reqtime > time_second) {
1434 				    frev = fxdr_hyper(tl);
1435 				    nqnfs_clientlease(nmp, np, nqlflag,
1436 					cachable, reqtime, frev);
1437 				}
1438 			}
1439 		}
1440 #endif
1441 		*mrp = mrep;
1442 		*mdp = md;
1443 		*dposp = dpos;
1444 
1445 		KASSERT(error == 0);
1446 		goto nfsmout;
1447 	}
1448 	m_freem(mrep);
1449 	error = EPROTONOSUPPORT;
1450 nfsmout:
1451 	KASSERT(kauth_cred_getrefcnt(acred) == 1);
1452 	kauth_cred_free(acred);
1453 	m_freem(rep->r_mreq);
1454 	free((caddr_t)rep, M_NFSREQ);
1455 	m_freem(mrest_backup);
1456 	return (error);
1457 }
1458 #endif /* NFS */
1459 
1460 /*
1461  * Generate the rpc reply header
1462  * siz arg. is used to decide if adding a cluster is worthwhile
1463  */
1464 int
1465 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1466 	int siz;
1467 	struct nfsrv_descript *nd;
1468 	struct nfssvc_sock *slp;
1469 	int err;
1470 	int cache;
1471 	u_quad_t *frev;
1472 	struct mbuf **mrq;
1473 	struct mbuf **mbp;
1474 	caddr_t *bposp;
1475 {
1476 	u_int32_t *tl;
1477 	struct mbuf *mreq;
1478 	caddr_t bpos;
1479 	struct mbuf *mb;
1480 
1481 	mreq = m_gethdr(M_WAIT, MT_DATA);
1482 	MCLAIM(mreq, &nfs_mowner);
1483 	mb = mreq;
1484 	/*
1485 	 * If this is a big reply, use a cluster else
1486 	 * try and leave leading space for the lower level headers.
1487 	 */
1488 	siz += RPC_REPLYSIZ;
1489 	if (siz >= max_datalen) {
1490 		m_clget(mreq, M_WAIT);
1491 	} else
1492 		mreq->m_data += max_hdr;
1493 	tl = mtod(mreq, u_int32_t *);
1494 	mreq->m_len = 6 * NFSX_UNSIGNED;
1495 	bpos = ((caddr_t)tl) + mreq->m_len;
1496 	*tl++ = txdr_unsigned(nd->nd_retxid);
1497 	*tl++ = rpc_reply;
1498 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1499 		*tl++ = rpc_msgdenied;
1500 		if (err & NFSERR_AUTHERR) {
1501 			*tl++ = rpc_autherr;
1502 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1503 			mreq->m_len -= NFSX_UNSIGNED;
1504 			bpos -= NFSX_UNSIGNED;
1505 		} else {
1506 			*tl++ = rpc_mismatch;
1507 			*tl++ = txdr_unsigned(RPC_VER2);
1508 			*tl = txdr_unsigned(RPC_VER2);
1509 		}
1510 	} else {
1511 		*tl++ = rpc_msgaccepted;
1512 
1513 		/*
1514 		 * For Kerberos authentication, we must send the nickname
1515 		 * verifier back, otherwise just RPCAUTH_NULL.
1516 		 */
1517 		if (nd->nd_flag & ND_KERBFULL) {
1518 			struct nfsuid *nuidp;
1519 			struct timeval ktvin, ktvout;
1520 
1521 			memset(&ktvout, 0, sizeof ktvout);	/* XXX gcc */
1522 
1523 			LIST_FOREACH(nuidp,
1524 			    NUIDHASH(slp, kauth_cred_geteuid(nd->nd_cr)),
1525 			    nu_hash) {
1526 				if (kauth_cred_geteuid(nuidp->nu_cr) ==
1527 				kauth_cred_geteuid(nd->nd_cr) &&
1528 				    (!nd->nd_nam2 || netaddr_match(
1529 				    NU_NETFAM(nuidp), &nuidp->nu_haddr,
1530 				    nd->nd_nam2)))
1531 					break;
1532 			}
1533 			if (nuidp) {
1534 				ktvin.tv_sec =
1535 				    txdr_unsigned(nuidp->nu_timestamp.tv_sec
1536 					- 1);
1537 				ktvin.tv_usec =
1538 				    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1539 
1540 				/*
1541 				 * Encrypt the timestamp in ecb mode using the
1542 				 * session key.
1543 				 */
1544 #ifdef NFSKERB
1545 				XXX
1546 #endif
1547 
1548 				*tl++ = rpc_auth_kerb;
1549 				*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1550 				*tl = ktvout.tv_sec;
1551 				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1552 				*tl++ = ktvout.tv_usec;
1553 				*tl++ = txdr_unsigned(
1554 				    kauth_cred_geteuid(nuidp->nu_cr));
1555 			} else {
1556 				*tl++ = 0;
1557 				*tl++ = 0;
1558 			}
1559 		} else {
1560 			*tl++ = 0;
1561 			*tl++ = 0;
1562 		}
1563 		switch (err) {
1564 		case EPROGUNAVAIL:
1565 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1566 			break;
1567 		case EPROGMISMATCH:
1568 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1569 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1570 			if (nd->nd_flag & ND_NQNFS) {
1571 				*tl++ = txdr_unsigned(3);
1572 				*tl = txdr_unsigned(3);
1573 			} else {
1574 				*tl++ = txdr_unsigned(2);
1575 				*tl = txdr_unsigned(3);
1576 			}
1577 			break;
1578 		case EPROCUNAVAIL:
1579 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1580 			break;
1581 		case EBADRPC:
1582 			*tl = txdr_unsigned(RPC_GARBAGE);
1583 			break;
1584 		default:
1585 			*tl = 0;
1586 			if (err != NFSERR_RETVOID) {
1587 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1588 				if (err)
1589 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1590 				else
1591 				    *tl = 0;
1592 			}
1593 			break;
1594 		};
1595 	}
1596 
1597 	/*
1598 	 * For nqnfs, piggyback lease as requested.
1599 	 */
1600 	if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1601 		if (nd->nd_flag & ND_LEASE) {
1602 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1603 			*tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1604 			*tl++ = txdr_unsigned(cache);
1605 			*tl++ = txdr_unsigned(nd->nd_duration);
1606 			txdr_hyper(*frev, tl);
1607 		} else {
1608 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1609 			*tl = 0;
1610 		}
1611 	}
1612 	if (mrq != NULL)
1613 		*mrq = mreq;
1614 	*mbp = mb;
1615 	*bposp = bpos;
1616 	if (err != 0 && err != NFSERR_RETVOID)
1617 		nfsstats.srvrpc_errs++;
1618 	return (0);
1619 }
1620 
1621 /*
1622  * Nfs timer routine
1623  * Scan the nfsreq list and retranmit any requests that have timed out
1624  * To avoid retransmission attempts on STREAM sockets (in the future) make
1625  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1626  */
1627 void
1628 nfs_timer(arg)
1629 	void *arg;	/* never used */
1630 {
1631 	struct nfsreq *rep;
1632 	struct mbuf *m;
1633 	struct socket *so;
1634 	struct nfsmount *nmp;
1635 	int timeo;
1636 	int s, error;
1637 #ifdef NFSSERVER
1638 	struct timeval tv;
1639 	struct nfssvc_sock *slp;
1640 	static long lasttime = 0;
1641 	u_quad_t cur_usec;
1642 #endif
1643 
1644 	s = splsoftnet();
1645 	TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
1646 		nmp = rep->r_nmp;
1647 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1648 			continue;
1649 		if (nfs_sigintr(nmp, rep, rep->r_lwp)) {
1650 			rep->r_flags |= R_SOFTTERM;
1651 			continue;
1652 		}
1653 		if (rep->r_rtt >= 0) {
1654 			rep->r_rtt++;
1655 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1656 				timeo = nmp->nm_timeo;
1657 			else
1658 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1659 			if (nmp->nm_timeouts > 0)
1660 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1661 			if (rep->r_rtt <= timeo)
1662 				continue;
1663 			if (nmp->nm_timeouts <
1664 			    (sizeof(nfs_backoff) / sizeof(nfs_backoff[0])))
1665 				nmp->nm_timeouts++;
1666 		}
1667 		/*
1668 		 * Check for server not responding
1669 		 */
1670 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1671 		     rep->r_rexmit > nmp->nm_deadthresh) {
1672 			nfs_msg(rep->r_lwp,
1673 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1674 			    "not responding");
1675 			rep->r_flags |= R_TPRINTFMSG;
1676 		}
1677 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1678 			nfsstats.rpctimeouts++;
1679 			rep->r_flags |= R_SOFTTERM;
1680 			continue;
1681 		}
1682 		if (nmp->nm_sotype != SOCK_DGRAM) {
1683 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1684 				rep->r_rexmit = NFS_MAXREXMIT;
1685 			continue;
1686 		}
1687 		if ((so = nmp->nm_so) == NULL)
1688 			continue;
1689 
1690 		/*
1691 		 * If there is enough space and the window allows..
1692 		 *	Resend it
1693 		 * Set r_rtt to -1 in case we fail to send it now.
1694 		 */
1695 		rep->r_rtt = -1;
1696 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1697 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1698 		    (rep->r_flags & R_SENT) ||
1699 		    nmp->nm_sent < nmp->nm_cwnd) &&
1700 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1701 		        if (so->so_state & SS_ISCONNECTED)
1702 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1703 			    (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0);
1704 			else
1705 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1706 			    nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0);
1707 			if (error) {
1708 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1709 #ifdef DEBUG
1710 					printf("nfs_timer: ignoring error %d\n",
1711 						error);
1712 #endif
1713 					so->so_error = 0;
1714 				}
1715 			} else {
1716 				/*
1717 				 * Iff first send, start timing
1718 				 * else turn timing off, backoff timer
1719 				 * and divide congestion window by 2.
1720 				 */
1721 				if (rep->r_flags & R_SENT) {
1722 					rep->r_flags &= ~R_TIMING;
1723 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1724 						rep->r_rexmit = NFS_MAXREXMIT;
1725 					nmp->nm_cwnd >>= 1;
1726 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1727 						nmp->nm_cwnd = NFS_CWNDSCALE;
1728 					nfsstats.rpcretries++;
1729 				} else {
1730 					rep->r_flags |= R_SENT;
1731 					nmp->nm_sent += NFS_CWNDSCALE;
1732 				}
1733 				rep->r_rtt = 0;
1734 			}
1735 		}
1736 	}
1737 
1738 #ifdef NFSSERVER
1739 	/*
1740 	 * Call the nqnfs server timer once a second to handle leases.
1741 	 */
1742 	if (lasttime != time_second) {
1743 		lasttime = time_second;
1744 		nqnfs_serverd();
1745 	}
1746 
1747 	/*
1748 	 * Scan the write gathering queues for writes that need to be
1749 	 * completed now.
1750 	 */
1751 	getmicrotime(&tv);
1752 	cur_usec = (u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec;
1753 	TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
1754 	    if (LIST_FIRST(&slp->ns_tq) &&
1755 		LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec)
1756 		nfsrv_wakenfsd(slp);
1757 	}
1758 #endif /* NFSSERVER */
1759 	splx(s);
1760 	callout_schedule(&nfs_timer_ch, nfs_ticks);
1761 }
1762 
1763 /*ARGSUSED*/
1764 void
1765 nfs_exit(p, v)
1766 	struct proc *p;
1767 	void *v;
1768 {
1769 	struct nfsreq *rp;
1770 	int s = splsoftnet();
1771 
1772 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
1773 		if (rp->r_lwp && rp->r_lwp->l_proc == p)
1774 			TAILQ_REMOVE(&nfs_reqq, rp, r_chain);
1775 	}
1776 	splx(s);
1777 }
1778 
1779 /*
1780  * Test for a termination condition pending on the process.
1781  * This is used for NFSMNT_INT mounts.
1782  */
1783 int
1784 nfs_sigintr(nmp, rep, l)
1785 	struct nfsmount *nmp;
1786 	struct nfsreq *rep;
1787 	struct lwp *l;
1788 {
1789 	sigset_t ss;
1790 
1791 	if (rep && (rep->r_flags & R_SOFTTERM))
1792 		return (EINTR);
1793 	if (!(nmp->nm_flag & NFSMNT_INT))
1794 		return (0);
1795 	if (l) {
1796 		sigpending1(l->l_proc, &ss);
1797 #if 0
1798 		sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss);
1799 #endif
1800 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1801 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1802 		    sigismember(&ss, SIGQUIT))
1803 			return (EINTR);
1804 	}
1805 	return (0);
1806 }
1807 
1808 /*
1809  * Lock a socket against others.
1810  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1811  * and also to avoid race conditions between the processes with nfs requests
1812  * in progress when a reconnect is necessary.
1813  */
1814 int
1815 nfs_sndlock(flagp, rep)
1816 	int *flagp;
1817 	struct nfsreq *rep;
1818 {
1819 	struct lwp *l;
1820 	int slpflag = 0, slptimeo = 0;
1821 
1822 	if (rep) {
1823 		l = rep->r_lwp;
1824 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1825 			slpflag = PCATCH;
1826 	} else
1827 		l = (struct lwp *)0;
1828 	while (*flagp & NFSMNT_SNDLOCK) {
1829 		if (rep && nfs_sigintr(rep->r_nmp, rep, l))
1830 			return (EINTR);
1831 		*flagp |= NFSMNT_WANTSND;
1832 		(void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1833 			slptimeo);
1834 		if (slpflag == PCATCH) {
1835 			slpflag = 0;
1836 			slptimeo = 2 * hz;
1837 		}
1838 	}
1839 	*flagp |= NFSMNT_SNDLOCK;
1840 	return (0);
1841 }
1842 
1843 /*
1844  * Unlock the stream socket for others.
1845  */
1846 void
1847 nfs_sndunlock(flagp)
1848 	int *flagp;
1849 {
1850 
1851 	if ((*flagp & NFSMNT_SNDLOCK) == 0)
1852 		panic("nfs sndunlock");
1853 	*flagp &= ~NFSMNT_SNDLOCK;
1854 	if (*flagp & NFSMNT_WANTSND) {
1855 		*flagp &= ~NFSMNT_WANTSND;
1856 		wakeup((caddr_t)flagp);
1857 	}
1858 }
1859 
1860 int
1861 nfs_rcvlock(rep)
1862 	struct nfsreq *rep;
1863 {
1864 	struct nfsmount *nmp = rep->r_nmp;
1865 	int *flagp = &nmp->nm_iflag;
1866 	int slpflag, slptimeo = 0;
1867 	int error = 0;
1868 
1869 	if (*flagp & NFSMNT_DISMNT)
1870 		return EIO;
1871 
1872 	if (*flagp & NFSMNT_INT)
1873 		slpflag = PCATCH;
1874 	else
1875 		slpflag = 0;
1876 	simple_lock(&nmp->nm_slock);
1877 	while (*flagp & NFSMNT_RCVLOCK) {
1878 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) {
1879 			error = EINTR;
1880 			goto quit;
1881 		}
1882 		*flagp |= NFSMNT_WANTRCV;
1883 		nmp->nm_waiters++;
1884 		(void) ltsleep(flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1885 			slptimeo, &nmp->nm_slock);
1886 		nmp->nm_waiters--;
1887 		if (*flagp & NFSMNT_DISMNT) {
1888 			wakeup(&nmp->nm_waiters);
1889 			error = EIO;
1890 			goto quit;
1891 		}
1892 		/* If our reply was received while we were sleeping,
1893 		 * then just return without taking the lock to avoid a
1894 		 * situation where a single iod could 'capture' the
1895 		 * receive lock.
1896 		 */
1897 		if (rep->r_mrep != NULL) {
1898 			error = EALREADY;
1899 			goto quit;
1900 		}
1901 		if (slpflag == PCATCH) {
1902 			slpflag = 0;
1903 			slptimeo = 2 * hz;
1904 		}
1905 	}
1906 	*flagp |= NFSMNT_RCVLOCK;
1907 quit:
1908 	simple_unlock(&nmp->nm_slock);
1909 	return error;
1910 }
1911 
1912 /*
1913  * Unlock the stream socket for others.
1914  */
1915 void
1916 nfs_rcvunlock(nmp)
1917 	struct nfsmount *nmp;
1918 {
1919 	int *flagp = &nmp->nm_iflag;
1920 
1921 	simple_lock(&nmp->nm_slock);
1922 	if ((*flagp & NFSMNT_RCVLOCK) == 0)
1923 		panic("nfs rcvunlock");
1924 	*flagp &= ~NFSMNT_RCVLOCK;
1925 	if (*flagp & NFSMNT_WANTRCV) {
1926 		*flagp &= ~NFSMNT_WANTRCV;
1927 		wakeup((caddr_t)flagp);
1928 	}
1929 	simple_unlock(&nmp->nm_slock);
1930 }
1931 
1932 /*
1933  * Parse an RPC request
1934  * - verify it
1935  * - allocate and fill in the cred.
1936  */
1937 int
1938 nfs_getreq(nd, nfsd, has_header)
1939 	struct nfsrv_descript *nd;
1940 	struct nfsd *nfsd;
1941 	int has_header;
1942 {
1943 	int len, i;
1944 	u_int32_t *tl;
1945 	int32_t t1;
1946 	struct uio uio;
1947 	struct iovec iov;
1948 	caddr_t dpos, cp2, cp;
1949 	u_int32_t nfsvers, auth_type;
1950 	uid_t nickuid;
1951 	int error = 0, nqnfs = 0, ticklen;
1952 	struct mbuf *mrep, *md;
1953 	struct nfsuid *nuidp;
1954 	struct timeval tvin, tvout;
1955 
1956 	memset(&tvout, 0, sizeof tvout);	/* XXX gcc */
1957 
1958 	KASSERT(nd->nd_cr == NULL);
1959 	mrep = nd->nd_mrep;
1960 	md = nd->nd_md;
1961 	dpos = nd->nd_dpos;
1962 	if (has_header) {
1963 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1964 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1965 		if (*tl++ != rpc_call) {
1966 			m_freem(mrep);
1967 			return (EBADRPC);
1968 		}
1969 	} else
1970 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1971 	nd->nd_repstat = 0;
1972 	nd->nd_flag = 0;
1973 	if (*tl++ != rpc_vers) {
1974 		nd->nd_repstat = ERPCMISMATCH;
1975 		nd->nd_procnum = NFSPROC_NOOP;
1976 		return (0);
1977 	}
1978 	if (*tl != nfs_prog) {
1979 		if (*tl == nqnfs_prog)
1980 			nqnfs++;
1981 		else {
1982 			nd->nd_repstat = EPROGUNAVAIL;
1983 			nd->nd_procnum = NFSPROC_NOOP;
1984 			return (0);
1985 		}
1986 	}
1987 	tl++;
1988 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1989 	if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1990 		(nfsvers != NQNFS_VER3 && nqnfs)) {
1991 		nd->nd_repstat = EPROGMISMATCH;
1992 		nd->nd_procnum = NFSPROC_NOOP;
1993 		return (0);
1994 	}
1995 	if (nqnfs)
1996 		nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1997 	else if (nfsvers == NFS_VER3)
1998 		nd->nd_flag = ND_NFSV3;
1999 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
2000 	if (nd->nd_procnum == NFSPROC_NULL)
2001 		return (0);
2002 	if (nd->nd_procnum >= NFS_NPROCS ||
2003 		(!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
2004 		(!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
2005 		nd->nd_repstat = EPROCUNAVAIL;
2006 		nd->nd_procnum = NFSPROC_NOOP;
2007 		return (0);
2008 	}
2009 	if ((nd->nd_flag & ND_NFSV3) == 0)
2010 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
2011 	auth_type = *tl++;
2012 	len = fxdr_unsigned(int, *tl++);
2013 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
2014 		m_freem(mrep);
2015 		return (EBADRPC);
2016 	}
2017 
2018 	nd->nd_flag &= ~ND_KERBAUTH;
2019 	/*
2020 	 * Handle auth_unix or auth_kerb.
2021 	 */
2022 	if (auth_type == rpc_auth_unix) {
2023 		uid_t uid;
2024 		gid_t gid, *grbuf;
2025 
2026 		nd->nd_cr = kauth_cred_alloc();
2027 		len = fxdr_unsigned(int, *++tl);
2028 		if (len < 0 || len > NFS_MAXNAMLEN) {
2029 			m_freem(mrep);
2030 			error = EBADRPC;
2031 			goto errout;
2032 		}
2033 		nfsm_adv(nfsm_rndup(len));
2034 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2035 
2036 		uid = fxdr_unsigned(uid_t, *tl++);
2037 		gid = fxdr_unsigned(gid_t, *tl++);
2038 		kauth_cred_setuid(nd->nd_cr, uid);
2039 		kauth_cred_setgid(nd->nd_cr, gid);
2040 		kauth_cred_seteuid(nd->nd_cr, uid);
2041 		kauth_cred_setsvuid(nd->nd_cr, gid);
2042 		kauth_cred_setegid(nd->nd_cr, uid);
2043 		kauth_cred_setsvgid(nd->nd_cr, gid);
2044 
2045 		len = fxdr_unsigned(int, *tl);
2046 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
2047 			m_freem(mrep);
2048 			error = EBADRPC;
2049 			goto errout;
2050 		}
2051 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
2052 
2053 		grbuf = malloc(len * sizeof(gid_t), M_TEMP, M_WAITOK);
2054 		for (i = 0; i < len; i++) {
2055 			if (i < NGROUPS) /* XXX elad */
2056 				grbuf[i] = fxdr_unsigned(gid_t, *tl++);
2057 			else
2058 				tl++;
2059 		}
2060 		kauth_cred_setgroups(nd->nd_cr, grbuf, min(len, NGROUPS), -1);
2061 		free(grbuf, M_TEMP);
2062 
2063 		len = fxdr_unsigned(int, *++tl);
2064 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
2065 			m_freem(mrep);
2066 			error = EBADRPC;
2067 			goto errout;
2068 		}
2069 		if (len > 0)
2070 			nfsm_adv(nfsm_rndup(len));
2071 	} else if (auth_type == rpc_auth_kerb) {
2072 		switch (fxdr_unsigned(int, *tl++)) {
2073 		case RPCAKN_FULLNAME:
2074 			ticklen = fxdr_unsigned(int, *tl);
2075 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
2076 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
2077 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
2078 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
2079 				m_freem(mrep);
2080 				error = EBADRPC;
2081 				goto errout;
2082 			}
2083 			uio.uio_offset = 0;
2084 			uio.uio_iov = &iov;
2085 			uio.uio_iovcnt = 1;
2086 			UIO_SETUP_SYSSPACE(&uio);
2087 			iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
2088 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
2089 			nfsm_mtouio(&uio, uio.uio_resid);
2090 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2091 			if (*tl++ != rpc_auth_kerb ||
2092 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
2093 				printf("Bad kerb verifier\n");
2094 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2095 				nd->nd_procnum = NFSPROC_NOOP;
2096 				return (0);
2097 			}
2098 			nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
2099 			tl = (u_int32_t *)cp;
2100 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
2101 				printf("Not fullname kerb verifier\n");
2102 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2103 				nd->nd_procnum = NFSPROC_NOOP;
2104 				return (0);
2105 			}
2106 			cp += NFSX_UNSIGNED;
2107 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
2108 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
2109 			nd->nd_flag |= ND_KERBFULL;
2110 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
2111 			break;
2112 		case RPCAKN_NICKNAME:
2113 			if (len != 2 * NFSX_UNSIGNED) {
2114 				printf("Kerb nickname short\n");
2115 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
2116 				nd->nd_procnum = NFSPROC_NOOP;
2117 				return (0);
2118 			}
2119 			nickuid = fxdr_unsigned(uid_t, *tl);
2120 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2121 			if (*tl++ != rpc_auth_kerb ||
2122 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
2123 				printf("Kerb nick verifier bad\n");
2124 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2125 				nd->nd_procnum = NFSPROC_NOOP;
2126 				return (0);
2127 			}
2128 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2129 			tvin.tv_sec = *tl++;
2130 			tvin.tv_usec = *tl;
2131 
2132 			LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid),
2133 			    nu_hash) {
2134 				if (kauth_cred_geteuid(nuidp->nu_cr) == nickuid &&
2135 				    (!nd->nd_nam2 ||
2136 				     netaddr_match(NU_NETFAM(nuidp),
2137 				      &nuidp->nu_haddr, nd->nd_nam2)))
2138 					break;
2139 			}
2140 			if (!nuidp) {
2141 				nd->nd_repstat =
2142 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
2143 				nd->nd_procnum = NFSPROC_NOOP;
2144 				return (0);
2145 			}
2146 
2147 			/*
2148 			 * Now, decrypt the timestamp using the session key
2149 			 * and validate it.
2150 			 */
2151 #ifdef NFSKERB
2152 			XXX
2153 #endif
2154 
2155 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
2156 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
2157 			if (nuidp->nu_expire < time_second ||
2158 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
2159 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
2160 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
2161 				nuidp->nu_expire = 0;
2162 				nd->nd_repstat =
2163 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
2164 				nd->nd_procnum = NFSPROC_NOOP;
2165 				return (0);
2166 			}
2167 			kauth_cred_hold(nuidp->nu_cr);
2168 			nd->nd_cr = nuidp->nu_cr;
2169 			nd->nd_flag |= ND_KERBNICK;
2170 		}
2171 	} else {
2172 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
2173 		nd->nd_procnum = NFSPROC_NOOP;
2174 		return (0);
2175 	}
2176 
2177 	/*
2178 	 * For nqnfs, get piggybacked lease request.
2179 	 */
2180 	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
2181 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2182 		nd->nd_flag |= fxdr_unsigned(int, *tl);
2183 		if (nd->nd_flag & ND_LEASE) {
2184 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2185 			nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
2186 		} else
2187 			nd->nd_duration = NQ_MINLEASE;
2188 	} else
2189 		nd->nd_duration = NQ_MINLEASE;
2190 	nd->nd_md = md;
2191 	nd->nd_dpos = dpos;
2192 	KASSERT((nd->nd_cr == NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) != 0)
2193 	     || (nd->nd_cr != NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) == 0));
2194 	return (0);
2195 nfsmout:
2196 errout:
2197 	KASSERT(error != 0);
2198 	if (nd->nd_cr != NULL) {
2199 		kauth_cred_free(nd->nd_cr);
2200 		nd->nd_cr = NULL;
2201 	}
2202 	return (error);
2203 }
2204 
2205 int
2206 nfs_msg(l, server, msg)
2207 	struct lwp *l;
2208 	const char *server, *msg;
2209 {
2210 	tpr_t tpr;
2211 
2212 	if (l)
2213 		tpr = tprintf_open(l->l_proc);
2214 	else
2215 		tpr = NULL;
2216 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
2217 	tprintf_close(tpr);
2218 	return (0);
2219 }
2220 
2221 #ifdef NFSSERVER
2222 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
2223 				    struct nfssvc_sock *, struct lwp *,
2224 				    struct mbuf **)) = {
2225 	nfsrv_null,
2226 	nfsrv_getattr,
2227 	nfsrv_setattr,
2228 	nfsrv_lookup,
2229 	nfsrv3_access,
2230 	nfsrv_readlink,
2231 	nfsrv_read,
2232 	nfsrv_write,
2233 	nfsrv_create,
2234 	nfsrv_mkdir,
2235 	nfsrv_symlink,
2236 	nfsrv_mknod,
2237 	nfsrv_remove,
2238 	nfsrv_rmdir,
2239 	nfsrv_rename,
2240 	nfsrv_link,
2241 	nfsrv_readdir,
2242 	nfsrv_readdirplus,
2243 	nfsrv_statfs,
2244 	nfsrv_fsinfo,
2245 	nfsrv_pathconf,
2246 	nfsrv_commit,
2247 	nqnfsrv_getlease,
2248 	nqnfsrv_vacated,
2249 	nfsrv_noop,
2250 	nfsrv_noop
2251 };
2252 
2253 /*
2254  * Socket upcall routine for the nfsd sockets.
2255  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
2256  * Essentially do as much as possible non-blocking, else punt and it will
2257  * be called with M_WAIT from an nfsd.
2258  */
2259 void
2260 nfsrv_rcv(so, arg, waitflag)
2261 	struct socket *so;
2262 	caddr_t arg;
2263 	int waitflag;
2264 {
2265 	struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
2266 	struct mbuf *m;
2267 	struct mbuf *mp, *nam;
2268 	struct uio auio;
2269 	int flags, error;
2270 	int setflags = 0;
2271 
2272 	error = nfsdsock_lock(slp, (waitflag != M_DONTWAIT));
2273 	if (error) {
2274 		setflags |= SLP_NEEDQ;
2275 		goto dorecs_unlocked;
2276 	}
2277 
2278 	KASSERT(so == slp->ns_so);
2279 #define NFS_TEST_HEAVY
2280 #ifdef NFS_TEST_HEAVY
2281 	/*
2282 	 * Define this to test for nfsds handling this under heavy load.
2283 	 *
2284 	 * XXX it isn't safe to call so_receive from so_upcall context.
2285 	 */
2286 	if (waitflag == M_DONTWAIT) {
2287 		setflags |= SLP_NEEDQ;
2288 		goto dorecs;
2289 	}
2290 #endif
2291 	simple_lock(&slp->ns_lock);
2292 	slp->ns_flag &= ~SLP_NEEDQ;
2293 	simple_unlock(&slp->ns_lock);
2294 	if (so->so_type == SOCK_STREAM) {
2295 #ifndef NFS_TEST_HEAVY
2296 		/*
2297 		 * If there are already records on the queue, defer soreceive()
2298 		 * to an nfsd so that there is feedback to the TCP layer that
2299 		 * the nfs servers are heavily loaded.
2300 		 */
2301 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
2302 			setflags |= SLP_NEEDQ;
2303 			goto dorecs;
2304 		}
2305 #endif
2306 
2307 		/*
2308 		 * Do soreceive().
2309 		 */
2310 		auio.uio_resid = 1000000000;
2311 		/* not need to setup uio_vmspace */
2312 		flags = MSG_DONTWAIT;
2313 		error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, &flags);
2314 		if (error || mp == NULL) {
2315 			if (error == EWOULDBLOCK)
2316 				setflags |= SLP_NEEDQ;
2317 			else
2318 				setflags |= SLP_DISCONN;
2319 			goto dorecs;
2320 		}
2321 		m = mp;
2322 		if (slp->ns_rawend) {
2323 			slp->ns_rawend->m_next = m;
2324 			slp->ns_cc += 1000000000 - auio.uio_resid;
2325 		} else {
2326 			slp->ns_raw = m;
2327 			slp->ns_cc = 1000000000 - auio.uio_resid;
2328 		}
2329 		while (m->m_next)
2330 			m = m->m_next;
2331 		slp->ns_rawend = m;
2332 
2333 		/*
2334 		 * Now try and parse record(s) out of the raw stream data.
2335 		 */
2336 		error = nfsrv_getstream(slp, waitflag);
2337 		if (error) {
2338 			if (error == EPERM)
2339 				setflags |= SLP_DISCONN;
2340 			else
2341 				setflags |= SLP_NEEDQ;
2342 		}
2343 	} else {
2344 		do {
2345 			auio.uio_resid = 1000000000;
2346 			/* not need to setup uio_vmspace */
2347 			flags = MSG_DONTWAIT;
2348 			error = (*so->so_receive)(so, &nam, &auio, &mp, NULL,
2349 			    &flags);
2350 			if (mp) {
2351 				if (nam) {
2352 					m = nam;
2353 					m->m_next = mp;
2354 				} else
2355 					m = mp;
2356 				if (slp->ns_recend)
2357 					slp->ns_recend->m_nextpkt = m;
2358 				else
2359 					slp->ns_rec = m;
2360 				slp->ns_recend = m;
2361 				m->m_nextpkt = (struct mbuf *)0;
2362 			}
2363 			if (error) {
2364 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2365 				    && error != EWOULDBLOCK) {
2366 					setflags |= SLP_DISCONN;
2367 					goto dorecs;
2368 				}
2369 			}
2370 		} while (mp);
2371 	}
2372 dorecs:
2373 	nfsdsock_unlock(slp);
2374 
2375 dorecs_unlocked:
2376 	/*
2377 	 * Now try and process the request records, non-blocking.
2378 	 */
2379 	if (setflags) {
2380 		simple_lock(&slp->ns_lock);
2381 		slp->ns_flag |= setflags;
2382 		simple_unlock(&slp->ns_lock);
2383 	}
2384 	if (waitflag == M_DONTWAIT &&
2385 	    (slp->ns_rec || (slp->ns_flag & (SLP_DISCONN | SLP_NEEDQ)) != 0)) {
2386 		nfsrv_wakenfsd(slp);
2387 	}
2388 }
2389 
2390 int
2391 nfsdsock_lock(struct nfssvc_sock *slp, boolean_t waitok)
2392 {
2393 
2394 	simple_lock(&slp->ns_lock);
2395 	while ((slp->ns_flag & (SLP_BUSY|SLP_VALID)) == SLP_BUSY) {
2396 		if (!waitok) {
2397 			simple_unlock(&slp->ns_lock);
2398 			return EWOULDBLOCK;
2399 		}
2400 		slp->ns_flag |= SLP_WANT;
2401 		ltsleep(&slp->ns_flag, PSOCK, "nslock", 0, &slp->ns_lock);
2402 	}
2403 	if ((slp->ns_flag & SLP_VALID) == 0) {
2404 		simple_unlock(&slp->ns_lock);
2405 		return EINVAL;
2406 	}
2407 	slp->ns_flag |= SLP_BUSY;
2408 	simple_unlock(&slp->ns_lock);
2409 
2410 	return 0;
2411 }
2412 
2413 void
2414 nfsdsock_unlock(struct nfssvc_sock *slp)
2415 {
2416 
2417 	KASSERT((slp->ns_flag & SLP_BUSY) != 0);
2418 
2419 	simple_lock(&slp->ns_lock);
2420 	if ((slp->ns_flag & SLP_WANT) != 0) {
2421 		wakeup(&slp->ns_flag);
2422 	}
2423 	slp->ns_flag &= ~(SLP_BUSY|SLP_WANT);
2424 	simple_unlock(&slp->ns_lock);
2425 }
2426 
2427 int
2428 nfsdsock_drain(struct nfssvc_sock *slp)
2429 {
2430 	int error = 0;
2431 
2432 	simple_lock(&slp->ns_lock);
2433 	if ((slp->ns_flag & SLP_VALID) == 0) {
2434 		error = EINVAL;
2435 		goto done;
2436 	}
2437 	slp->ns_flag &= ~SLP_VALID;
2438 	while ((slp->ns_flag & SLP_BUSY) != 0) {
2439 		slp->ns_flag |= SLP_WANT;
2440 		ltsleep(&slp->ns_flag, PSOCK, "nsdrain", 0, &slp->ns_lock);
2441 	}
2442 done:
2443 	simple_unlock(&slp->ns_lock);
2444 
2445 	return error;
2446 }
2447 
2448 /*
2449  * Try and extract an RPC request from the mbuf data list received on a
2450  * stream socket. The "waitflag" argument indicates whether or not it
2451  * can sleep.
2452  */
2453 int
2454 nfsrv_getstream(slp, waitflag)
2455 	struct nfssvc_sock *slp;
2456 	int waitflag;
2457 {
2458 	struct mbuf *m, **mpp;
2459 	struct mbuf *recm;
2460 	u_int32_t recmark;
2461 	int error = 0;
2462 
2463 	for (;;) {
2464 		if (slp->ns_reclen == 0) {
2465 			if (slp->ns_cc < NFSX_UNSIGNED) {
2466 				break;
2467 			}
2468 			m = slp->ns_raw;
2469 			m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark);
2470 			m_adj(m, NFSX_UNSIGNED);
2471 			slp->ns_cc -= NFSX_UNSIGNED;
2472 			recmark = ntohl(recmark);
2473 			slp->ns_reclen = recmark & ~0x80000000;
2474 			if (recmark & 0x80000000)
2475 				slp->ns_flag |= SLP_LASTFRAG;
2476 			else
2477 				slp->ns_flag &= ~SLP_LASTFRAG;
2478 			if (slp->ns_reclen > NFS_MAXPACKET) {
2479 				error = EPERM;
2480 				break;
2481 			}
2482 		}
2483 
2484 		/*
2485 		 * Now get the record part.
2486 		 *
2487 		 * Note that slp->ns_reclen may be 0.  Linux sometimes
2488 		 * generates 0-length records.
2489 		 */
2490 		if (slp->ns_cc == slp->ns_reclen) {
2491 			recm = slp->ns_raw;
2492 			slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2493 			slp->ns_cc = slp->ns_reclen = 0;
2494 		} else if (slp->ns_cc > slp->ns_reclen) {
2495 			recm = slp->ns_raw;
2496 			m = m_split(recm, slp->ns_reclen, waitflag);
2497 			if (m == NULL) {
2498 				error = EWOULDBLOCK;
2499 				break;
2500 			}
2501 			m_claimm(recm, &nfs_mowner);
2502 			slp->ns_raw = m;
2503 			if (m->m_next == NULL)
2504 				slp->ns_rawend = m;
2505 			slp->ns_cc -= slp->ns_reclen;
2506 			slp->ns_reclen = 0;
2507 		} else {
2508 			break;
2509 		}
2510 
2511 		/*
2512 		 * Accumulate the fragments into a record.
2513 		 */
2514 		mpp = &slp->ns_frag;
2515 		while (*mpp)
2516 			mpp = &((*mpp)->m_next);
2517 		*mpp = recm;
2518 		if (slp->ns_flag & SLP_LASTFRAG) {
2519 			if (slp->ns_recend)
2520 				slp->ns_recend->m_nextpkt = slp->ns_frag;
2521 			else
2522 				slp->ns_rec = slp->ns_frag;
2523 			slp->ns_recend = slp->ns_frag;
2524 			slp->ns_frag = (struct mbuf *)0;
2525 		}
2526 	}
2527 
2528 	return error;
2529 }
2530 
2531 /*
2532  * Parse an RPC header.
2533  */
2534 int
2535 nfsrv_dorec(slp, nfsd, ndp)
2536 	struct nfssvc_sock *slp;
2537 	struct nfsd *nfsd;
2538 	struct nfsrv_descript **ndp;
2539 {
2540 	struct mbuf *m, *nam;
2541 	struct nfsrv_descript *nd;
2542 	int error;
2543 
2544 	*ndp = NULL;
2545 
2546 	if (nfsdsock_lock(slp, TRUE)) {
2547 		return ENOBUFS;
2548 	}
2549 	m = slp->ns_rec;
2550 	if (m == NULL) {
2551 		nfsdsock_unlock(slp);
2552 		return ENOBUFS;
2553 	}
2554 	slp->ns_rec = m->m_nextpkt;
2555 	if (slp->ns_rec)
2556 		m->m_nextpkt = NULL;
2557 	else
2558 		slp->ns_recend = NULL;
2559 	nfsdsock_unlock(slp);
2560 
2561 	if (m->m_type == MT_SONAME) {
2562 		nam = m;
2563 		m = m->m_next;
2564 		nam->m_next = NULL;
2565 	} else
2566 		nam = NULL;
2567 	nd = nfsdreq_alloc();
2568 	nd->nd_md = nd->nd_mrep = m;
2569 	nd->nd_nam2 = nam;
2570 	nd->nd_dpos = mtod(m, caddr_t);
2571 	error = nfs_getreq(nd, nfsd, TRUE);
2572 	if (error) {
2573 		m_freem(nam);
2574 		nfsdreq_free(nd);
2575 		return (error);
2576 	}
2577 	*ndp = nd;
2578 	nfsd->nfsd_nd = nd;
2579 	return (0);
2580 }
2581 
2582 /*
2583  * Search for a sleeping nfsd and wake it up.
2584  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2585  * running nfsds will go look for the work in the nfssvc_sock list.
2586  */
2587 void
2588 nfsrv_wakenfsd(slp)
2589 	struct nfssvc_sock *slp;
2590 {
2591 	struct nfsd *nd;
2592 
2593 	if ((slp->ns_flag & SLP_VALID) == 0)
2594 		return;
2595 	simple_lock(&nfsd_slock);
2596 	if (slp->ns_flag & SLP_DOREC) {
2597 		simple_unlock(&nfsd_slock);
2598 		return;
2599 	}
2600 	nd = SLIST_FIRST(&nfsd_idle_head);
2601 	if (nd) {
2602 		SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle);
2603 		simple_unlock(&nfsd_slock);
2604 
2605 		if (nd->nfsd_slp)
2606 			panic("nfsd wakeup");
2607 		slp->ns_sref++;
2608 		nd->nfsd_slp = slp;
2609 		wakeup(nd);
2610 		return;
2611 	}
2612 	slp->ns_flag |= SLP_DOREC;
2613 	nfsd_head_flag |= NFSD_CHECKSLP;
2614 	TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending);
2615 	simple_unlock(&nfsd_slock);
2616 }
2617 
2618 int
2619 nfsdsock_sendreply(struct nfssvc_sock *slp, struct nfsrv_descript *nd)
2620 {
2621 	int error;
2622 
2623 	if (nd->nd_mrep != NULL) {
2624 		m_freem(nd->nd_mrep);
2625 		nd->nd_mrep = NULL;
2626 	}
2627 
2628 	simple_lock(&slp->ns_lock);
2629 	if ((slp->ns_flag & SLP_SENDING) != 0) {
2630 		SIMPLEQ_INSERT_TAIL(&slp->ns_sendq, nd, nd_sendq);
2631 		simple_unlock(&slp->ns_lock);
2632 		return 0;
2633 	}
2634 	KASSERT(SIMPLEQ_EMPTY(&slp->ns_sendq));
2635 	slp->ns_flag |= SLP_SENDING;
2636 	simple_unlock(&slp->ns_lock);
2637 
2638 again:
2639 	error = nfs_send(slp->ns_so, nd->nd_nam2, nd->nd_mreq, NULL, curlwp);
2640 	if (nd->nd_nam2) {
2641 		m_free(nd->nd_nam2);
2642 	}
2643 	nfsdreq_free(nd);
2644 
2645 	simple_lock(&slp->ns_lock);
2646 	KASSERT((slp->ns_flag & SLP_SENDING) != 0);
2647 	nd = SIMPLEQ_FIRST(&slp->ns_sendq);
2648 	if (nd != NULL) {
2649 		SIMPLEQ_REMOVE_HEAD(&slp->ns_sendq, nd_sendq);
2650 		simple_unlock(&slp->ns_lock);
2651 		goto again;
2652 	}
2653 	slp->ns_flag &= ~SLP_SENDING;
2654 	simple_unlock(&slp->ns_lock);
2655 
2656 	return error;
2657 }
2658 #endif /* NFSSERVER */
2659 
2660 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY))
2661 static struct pool nfs_srvdesc_pool;
2662 
2663 void
2664 nfsdreq_init(void)
2665 {
2666 
2667 	pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript),
2668 	    0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr);
2669 }
2670 
2671 struct nfsrv_descript *
2672 nfsdreq_alloc(void)
2673 {
2674 	struct nfsrv_descript *nd;
2675 
2676 	nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK);
2677 	nd->nd_cr = NULL;
2678 	return nd;
2679 }
2680 
2681 void
2682 nfsdreq_free(struct nfsrv_descript *nd)
2683 {
2684 	kauth_cred_t cr;
2685 
2686 	cr = nd->nd_cr;
2687 	if (cr != NULL) {
2688 		KASSERT(kauth_cred_getrefcnt(cr) == 1);
2689 		kauth_cred_free(cr);
2690 	}
2691 	pool_put(&nfs_srvdesc_pool, nd);
2692 }
2693 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */
2694