xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 6dffe8d42bd46273f674d7ab834e7be9b1af990e)
1 /*	$NetBSD: nfs_socket.c,v 1.180 2009/03/14 21:04:25 dsl 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 1.180 2009/03/14 21:04:25 dsl Exp $");
43 
44 #ifdef _KERNEL_OPT
45 #include "fs_nfs.h"
46 #include "opt_nfs.h"
47 #include "opt_mbuftrace.h"
48 #endif
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/evcnt.h>
53 #include <sys/callout.h>
54 #include <sys/proc.h>
55 #include <sys/mount.h>
56 #include <sys/kernel.h>
57 #include <sys/kmem.h>
58 #include <sys/mbuf.h>
59 #include <sys/vnode.h>
60 #include <sys/domain.h>
61 #include <sys/protosw.h>
62 #include <sys/socket.h>
63 #include <sys/socketvar.h>
64 #include <sys/syslog.h>
65 #include <sys/tprintf.h>
66 #include <sys/namei.h>
67 #include <sys/signal.h>
68 #include <sys/signalvar.h>
69 #include <sys/kauth.h>
70 
71 #include <netinet/in.h>
72 #include <netinet/tcp.h>
73 
74 #include <nfs/rpcv2.h>
75 #include <nfs/nfsproto.h>
76 #include <nfs/nfs.h>
77 #include <nfs/xdr_subs.h>
78 #include <nfs/nfsm_subs.h>
79 #include <nfs/nfsmount.h>
80 #include <nfs/nfsnode.h>
81 #include <nfs/nfsrtt.h>
82 #include <nfs/nfs_var.h>
83 
84 #ifdef MBUFTRACE
85 struct mowner nfs_mowner = MOWNER_INIT("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;
114 extern const int nfsv3_procid[NFS_NPROCS];
115 extern int nfs_ticks;
116 
117 #ifdef DEBUG
118 /*
119  * Avoid spamming the console with debugging messages.  We only print
120  * the nfs timer and reply error debugs every 10 seconds.
121  */
122 static const struct timeval nfs_err_interval = { 10, 0 };
123 static struct timeval nfs_reply_last_err_time;
124 static struct timeval nfs_timer_last_err_time;
125 #endif
126 
127 /*
128  * Defines which timer to use for the procnum.
129  * 0 - default
130  * 1 - getattr
131  * 2 - lookup
132  * 3 - read
133  * 4 - write
134  */
135 static const int proct[NFS_NPROCS] = {
136 	[NFSPROC_NULL] = 0,
137 	[NFSPROC_GETATTR] = 1,
138 	[NFSPROC_SETATTR] = 0,
139 	[NFSPROC_LOOKUP] = 2,
140 	[NFSPROC_ACCESS] = 1,
141 	[NFSPROC_READLINK] = 3,
142 	[NFSPROC_READ] = 3,
143 	[NFSPROC_WRITE] = 4,
144 	[NFSPROC_CREATE] = 0,
145 	[NFSPROC_MKDIR] = 0,
146 	[NFSPROC_SYMLINK] = 0,
147 	[NFSPROC_MKNOD] = 0,
148 	[NFSPROC_REMOVE] = 0,
149 	[NFSPROC_RMDIR] = 0,
150 	[NFSPROC_RENAME] = 0,
151 	[NFSPROC_LINK] = 0,
152 	[NFSPROC_READDIR] = 3,
153 	[NFSPROC_READDIRPLUS] = 3,
154 	[NFSPROC_FSSTAT] = 0,
155 	[NFSPROC_FSINFO] = 0,
156 	[NFSPROC_PATHCONF] = 0,
157 	[NFSPROC_COMMIT] = 0,
158 	[NFSPROC_NOOP] = 0,
159 };
160 
161 /*
162  * There is a congestion window for outstanding rpcs maintained per mount
163  * point. The cwnd size is adjusted in roughly the way that:
164  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
165  * SIGCOMM '88". ACM, August 1988.
166  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
167  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
168  * of rpcs is in progress.
169  * (The sent count and cwnd are scaled for integer arith.)
170  * Variants of "slow start" were tried and were found to be too much of a
171  * performance hit (ave. rtt 3 times larger),
172  * I suspect due to the large rtt that nfs rpcs have.
173  */
174 #define	NFS_CWNDSCALE	256
175 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
176 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
177 int nfsrtton = 0;
178 struct nfsrtt nfsrtt;
179 struct nfsreqhead nfs_reqq;
180 static callout_t nfs_timer_ch;
181 static struct evcnt nfs_timer_ev;
182 static struct evcnt nfs_timer_start_ev;
183 static struct evcnt nfs_timer_stop_ev;
184 static kmutex_t nfs_timer_lock;
185 static bool (*nfs_timer_srvvec)(void);
186 
187 #ifdef NFS
188 static int nfs_sndlock(struct nfsmount *, struct nfsreq *);
189 static void nfs_sndunlock(struct nfsmount *);
190 #endif
191 static int nfs_rcvlock(struct nfsmount *, struct nfsreq *);
192 static void nfs_rcvunlock(struct nfsmount *);
193 
194 /*
195  * Initialize sockets and congestion for a new NFS connection.
196  * We do not free the sockaddr if error.
197  */
198 int
199 nfs_connect(struct nfsmount *nmp, struct nfsreq *rep, struct lwp *l)
200 {
201 	struct socket *so;
202 	int error, rcvreserve, sndreserve;
203 	struct sockaddr *saddr;
204 	struct sockaddr_in *sin;
205 	struct sockaddr_in6 *sin6;
206 	struct mbuf *m;
207 	int val;
208 
209 	nmp->nm_so = (struct socket *)0;
210 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
211 	error = socreate(saddr->sa_family, &nmp->nm_so,
212 		nmp->nm_sotype, nmp->nm_soproto, l, NULL);
213 	if (error)
214 		goto bad;
215 	so = nmp->nm_so;
216 #ifdef MBUFTRACE
217 	so->so_mowner = &nfs_mowner;
218 	so->so_rcv.sb_mowner = &nfs_mowner;
219 	so->so_snd.sb_mowner = &nfs_mowner;
220 #endif
221 	nmp->nm_soflags = so->so_proto->pr_flags;
222 
223 	/*
224 	 * Some servers require that the client port be a reserved port number.
225 	 */
226 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
227 		val = IP_PORTRANGE_LOW;
228 
229 		if ((error = so_setsockopt(NULL, so, IPPROTO_IP, IP_PORTRANGE,
230 		    &val, sizeof(val))))
231 			goto bad;
232 		m = m_get(M_WAIT, MT_SONAME);
233 		MCLAIM(m, so->so_mowner);
234 		sin = mtod(m, struct sockaddr_in *);
235 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
236 		sin->sin_family = AF_INET;
237 		sin->sin_addr.s_addr = INADDR_ANY;
238 		sin->sin_port = 0;
239 		error = sobind(so, m, &lwp0);
240 		m_freem(m);
241 		if (error)
242 			goto bad;
243 	}
244 	if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) {
245 		val = IPV6_PORTRANGE_LOW;
246 
247 		if ((error = so_setsockopt(NULL, so, IPPROTO_IPV6,
248 		    IPV6_PORTRANGE, &val, sizeof(val))))
249 			goto bad;
250 		m = m_get(M_WAIT, MT_SONAME);
251 		MCLAIM(m, so->so_mowner);
252 		sin6 = mtod(m, struct sockaddr_in6 *);
253 		memset(sin6, 0, sizeof(*sin6));
254 		sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6);
255 		sin6->sin6_family = AF_INET6;
256 		error = sobind(so, m, &lwp0);
257 		m_freem(m);
258 		if (error)
259 			goto bad;
260 	}
261 
262 	/*
263 	 * Protocols that do not require connections may be optionally left
264 	 * unconnected for servers that reply from a port other than NFS_PORT.
265 	 */
266 	solock(so);
267 	if (nmp->nm_flag & NFSMNT_NOCONN) {
268 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
269 			sounlock(so);
270 			error = ENOTCONN;
271 			goto bad;
272 		}
273 	} else {
274 		error = soconnect(so, nmp->nm_nam, l);
275 		if (error) {
276 			sounlock(so);
277 			goto bad;
278 		}
279 
280 		/*
281 		 * Wait for the connection to complete. Cribbed from the
282 		 * connect system call but with the wait timing out so
283 		 * that interruptible mounts don't hang here for a long time.
284 		 */
285 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
286 			(void)sowait(so, false, 2 * hz);
287 			if ((so->so_state & SS_ISCONNECTING) &&
288 			    so->so_error == 0 && rep &&
289 			    (error = nfs_sigintr(nmp, rep, rep->r_lwp)) != 0){
290 				so->so_state &= ~SS_ISCONNECTING;
291 				sounlock(so);
292 				goto bad;
293 			}
294 		}
295 		if (so->so_error) {
296 			error = so->so_error;
297 			so->so_error = 0;
298 			sounlock(so);
299 			goto bad;
300 		}
301 	}
302 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
303 		so->so_rcv.sb_timeo = (5 * hz);
304 		so->so_snd.sb_timeo = (5 * hz);
305 	} else {
306 		/*
307 		 * enable receive timeout to detect server crash and reconnect.
308 		 * otherwise, we can be stuck in soreceive forever.
309 		 */
310 		so->so_rcv.sb_timeo = (5 * hz);
311 		so->so_snd.sb_timeo = 0;
312 	}
313 	if (nmp->nm_sotype == SOCK_DGRAM) {
314 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
315 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
316 		    NFS_MAXPKTHDR) * 2;
317 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
318 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
319 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
320 		    NFS_MAXPKTHDR) * 2;
321 	} else {
322 		sounlock(so);
323 		if (nmp->nm_sotype != SOCK_STREAM)
324 			panic("nfscon sotype");
325 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
326 			val = 1;
327 			so_setsockopt(NULL, so, SOL_SOCKET, SO_KEEPALIVE, &val,
328 			    sizeof(val));
329 		}
330 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
331 			val = 1;
332 			so_setsockopt(NULL, so, IPPROTO_TCP, TCP_NODELAY, &val,
333 			    sizeof(val));
334 		}
335 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
336 		    sizeof (u_int32_t)) * 2;
337 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
338 		    sizeof (u_int32_t)) * 2;
339 		solock(so);
340 	}
341 	error = soreserve(so, sndreserve, rcvreserve);
342 	if (error) {
343 		sounlock(so);
344 		goto bad;
345 	}
346 	so->so_rcv.sb_flags |= SB_NOINTR;
347 	so->so_snd.sb_flags |= SB_NOINTR;
348 	sounlock(so);
349 
350 	/* Initialize other non-zero congestion variables */
351 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
352 		NFS_TIMEO << 3;
353 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
354 		nmp->nm_sdrtt[3] = 0;
355 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
356 	nmp->nm_sent = 0;
357 	nmp->nm_timeouts = 0;
358 	return (0);
359 
360 bad:
361 	nfs_disconnect(nmp);
362 	return (error);
363 }
364 
365 /*
366  * Reconnect routine:
367  * Called when a connection is broken on a reliable protocol.
368  * - clean up the old socket
369  * - nfs_connect() again
370  * - set R_MUSTRESEND for all outstanding requests on mount point
371  * If this fails the mount point is DEAD!
372  * nb: Must be called with the nfs_sndlock() set on the mount point.
373  */
374 int
375 nfs_reconnect(struct nfsreq *rep)
376 {
377 	struct nfsreq *rp;
378 	struct nfsmount *nmp = rep->r_nmp;
379 	int error;
380 
381 	nfs_disconnect(nmp);
382 	while ((error = nfs_connect(nmp, rep, &lwp0)) != 0) {
383 		if (error == EINTR || error == ERESTART)
384 			return (EINTR);
385 		kpause("nfscn2", false, hz, NULL);
386 	}
387 
388 	/*
389 	 * Loop through outstanding request list and fix up all requests
390 	 * on old socket.
391 	 */
392 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
393 		if (rp->r_nmp == nmp) {
394 			if ((rp->r_flags & R_MUSTRESEND) == 0)
395 				rp->r_flags |= R_MUSTRESEND | R_REXMITTED;
396 			rp->r_rexmit = 0;
397 		}
398 	}
399 	return (0);
400 }
401 
402 /*
403  * NFS disconnect. Clean up and unlink.
404  */
405 void
406 nfs_disconnect(struct nfsmount *nmp)
407 {
408 	struct socket *so;
409 	int drain = 0;
410 
411 	if (nmp->nm_so) {
412 		so = nmp->nm_so;
413 		nmp->nm_so = (struct socket *)0;
414 		solock(so);
415 		soshutdown(so, SHUT_RDWR);
416 		sounlock(so);
417 		drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
418 		if (drain) {
419 			/*
420 			 * soshutdown() above should wake up the current
421 			 * listener.
422 			 * Now wake up those waiting for the receive lock, and
423 			 * wait for them to go away unhappy, to prevent *nmp
424 			 * from evaporating while they're sleeping.
425 			 */
426 			mutex_enter(&nmp->nm_lock);
427 			while (nmp->nm_waiters > 0) {
428 				cv_broadcast(&nmp->nm_rcvcv);
429 				cv_broadcast(&nmp->nm_sndcv);
430 				cv_wait(&nmp->nm_disconcv, &nmp->nm_lock);
431 			}
432 			mutex_exit(&nmp->nm_lock);
433 		}
434 		soclose(so);
435 	}
436 #ifdef DIAGNOSTIC
437 	if (drain && (nmp->nm_waiters > 0))
438 		panic("nfs_disconnect: waiters left after drain?");
439 #endif
440 }
441 
442 void
443 nfs_safedisconnect(struct nfsmount *nmp)
444 {
445 	struct nfsreq dummyreq;
446 
447 	memset(&dummyreq, 0, sizeof(dummyreq));
448 	dummyreq.r_nmp = nmp;
449 	nfs_rcvlock(nmp, &dummyreq); /* XXX ignored error return */
450 	nfs_disconnect(nmp);
451 	nfs_rcvunlock(nmp);
452 }
453 
454 /*
455  * This is the nfs send routine. For connection based socket types, it
456  * must be called with an nfs_sndlock() on the socket.
457  * "rep == NULL" indicates that it has been called from a server.
458  * For the client side:
459  * - return EINTR if the RPC is terminated, 0 otherwise
460  * - set R_MUSTRESEND if the send fails for any reason
461  * - do any cleanup required by recoverable socket errors (? ? ?)
462  * For the server side:
463  * - return EINTR or ERESTART if interrupted by a signal
464  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
465  * - do any cleanup required by recoverable socket errors (? ? ?)
466  */
467 int
468 nfs_send(struct socket *so, struct mbuf *nam, struct mbuf *top, struct nfsreq *rep, struct lwp *l)
469 {
470 	struct mbuf *sendnam;
471 	int error, soflags, flags;
472 
473 	/* XXX nfs_doio()/nfs_request() calls with  rep->r_lwp == NULL */
474 	if (l == NULL && rep->r_lwp == NULL)
475 		l = curlwp;
476 
477 	if (rep) {
478 		if (rep->r_flags & R_SOFTTERM) {
479 			m_freem(top);
480 			return (EINTR);
481 		}
482 		if ((so = rep->r_nmp->nm_so) == NULL) {
483 			rep->r_flags |= R_MUSTRESEND;
484 			m_freem(top);
485 			return (0);
486 		}
487 		rep->r_flags &= ~R_MUSTRESEND;
488 		soflags = rep->r_nmp->nm_soflags;
489 	} else
490 		soflags = so->so_proto->pr_flags;
491 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
492 		sendnam = (struct mbuf *)0;
493 	else
494 		sendnam = nam;
495 	if (so->so_type == SOCK_SEQPACKET)
496 		flags = MSG_EOR;
497 	else
498 		flags = 0;
499 
500 	error = (*so->so_send)(so, sendnam, NULL, top, NULL, flags,  l);
501 	if (error) {
502 		if (rep) {
503 			if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
504 				/*
505 				 * We're too fast for the network/driver,
506 				 * and UDP isn't flowcontrolled.
507 				 * We need to resend. This is not fatal,
508 				 * just try again.
509 				 *
510 				 * Could be smarter here by doing some sort
511 				 * of a backoff, but this is rare.
512 				 */
513 				rep->r_flags |= R_MUSTRESEND;
514 			} else {
515 				if (error != EPIPE)
516 					log(LOG_INFO,
517 					    "nfs send error %d for %s\n",
518 					    error,
519 					    rep->r_nmp->nm_mountp->
520 						    mnt_stat.f_mntfromname);
521 				/*
522 				 * Deal with errors for the client side.
523 				 */
524 				if (rep->r_flags & R_SOFTTERM)
525 					error = EINTR;
526 				else
527 					rep->r_flags |= R_MUSTRESEND;
528 			}
529 		} else {
530 			/*
531 			 * See above. This error can happen under normal
532 			 * circumstances and the log is too noisy.
533 			 * The error will still show up in nfsstat.
534 			 */
535 			if (error != ENOBUFS || so->so_type != SOCK_DGRAM)
536 				log(LOG_INFO, "nfsd send error %d\n", error);
537 		}
538 
539 		/*
540 		 * Handle any recoverable (soft) socket errors here. (? ? ?)
541 		 */
542 		if (error != EINTR && error != ERESTART &&
543 			error != EWOULDBLOCK && error != EPIPE)
544 			error = 0;
545 	}
546 	return (error);
547 }
548 
549 #ifdef NFS
550 /*
551  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
552  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
553  * Mark and consolidate the data into a new mbuf list.
554  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
555  *     small mbufs.
556  * For SOCK_STREAM we must be very careful to read an entire record once
557  * we have read any of it, even if the system call has been interrupted.
558  */
559 static int
560 nfs_receive(struct nfsreq *rep, struct mbuf **aname, struct mbuf **mp,
561     struct lwp *l)
562 {
563 	struct socket *so;
564 	struct uio auio;
565 	struct iovec aio;
566 	struct mbuf *m;
567 	struct mbuf *control;
568 	u_int32_t len;
569 	struct mbuf **getnam;
570 	int error, sotype, rcvflg;
571 
572 	/*
573 	 * Set up arguments for soreceive()
574 	 */
575 	*mp = (struct mbuf *)0;
576 	*aname = (struct mbuf *)0;
577 	sotype = rep->r_nmp->nm_sotype;
578 
579 	/*
580 	 * For reliable protocols, lock against other senders/receivers
581 	 * in case a reconnect is necessary.
582 	 * For SOCK_STREAM, first get the Record Mark to find out how much
583 	 * more there is to get.
584 	 * We must lock the socket against other receivers
585 	 * until we have an entire rpc request/reply.
586 	 */
587 	if (sotype != SOCK_DGRAM) {
588 		error = nfs_sndlock(rep->r_nmp, rep);
589 		if (error)
590 			return (error);
591 tryagain:
592 		/*
593 		 * Check for fatal errors and resending request.
594 		 */
595 		/*
596 		 * Ugh: If a reconnect attempt just happened, nm_so
597 		 * would have changed. NULL indicates a failed
598 		 * attempt that has essentially shut down this
599 		 * mount point.
600 		 */
601 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
602 			nfs_sndunlock(rep->r_nmp);
603 			return (EINTR);
604 		}
605 		so = rep->r_nmp->nm_so;
606 		if (!so) {
607 			error = nfs_reconnect(rep);
608 			if (error) {
609 				nfs_sndunlock(rep->r_nmp);
610 				return (error);
611 			}
612 			goto tryagain;
613 		}
614 		while (rep->r_flags & R_MUSTRESEND) {
615 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
616 			nfsstats.rpcretries++;
617 			rep->r_rtt = 0;
618 			rep->r_flags &= ~R_TIMING;
619 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, l);
620 			if (error) {
621 				if (error == EINTR || error == ERESTART ||
622 				    (error = nfs_reconnect(rep)) != 0) {
623 					nfs_sndunlock(rep->r_nmp);
624 					return (error);
625 				}
626 				goto tryagain;
627 			}
628 		}
629 		nfs_sndunlock(rep->r_nmp);
630 		if (sotype == SOCK_STREAM) {
631 			aio.iov_base = (void *) &len;
632 			aio.iov_len = sizeof(u_int32_t);
633 			auio.uio_iov = &aio;
634 			auio.uio_iovcnt = 1;
635 			auio.uio_rw = UIO_READ;
636 			auio.uio_offset = 0;
637 			auio.uio_resid = sizeof(u_int32_t);
638 			UIO_SETUP_SYSSPACE(&auio);
639 			do {
640 			   rcvflg = MSG_WAITALL;
641 			   error = (*so->so_receive)(so, (struct mbuf **)0, &auio,
642 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
643 			   if (error == EWOULDBLOCK && rep) {
644 				if (rep->r_flags & R_SOFTTERM)
645 					return (EINTR);
646 				/*
647 				 * if it seems that the server died after it
648 				 * received our request, set EPIPE so that
649 				 * we'll reconnect and retransmit requests.
650 				 */
651 				if (rep->r_rexmit >= rep->r_nmp->nm_retry) {
652 					nfsstats.rpctimeouts++;
653 					error = EPIPE;
654 				}
655 			   }
656 			} while (error == EWOULDBLOCK);
657 			if (!error && auio.uio_resid > 0) {
658 			    /*
659 			     * Don't log a 0 byte receive; it means
660 			     * that the socket has been closed, and
661 			     * can happen during normal operation
662 			     * (forcible unmount or Solaris server).
663 			     */
664 			    if (auio.uio_resid != sizeof (u_int32_t))
665 			      log(LOG_INFO,
666 				 "short receive (%lu/%lu) from nfs server %s\n",
667 				 (u_long)sizeof(u_int32_t) - auio.uio_resid,
668 				 (u_long)sizeof(u_int32_t),
669 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
670 			    error = EPIPE;
671 			}
672 			if (error)
673 				goto errout;
674 			len = ntohl(len) & ~0x80000000;
675 			/*
676 			 * This is SERIOUS! We are out of sync with the sender
677 			 * and forcing a disconnect/reconnect is all I can do.
678 			 */
679 			if (len > NFS_MAXPACKET) {
680 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
681 				"impossible packet length",
682 				len,
683 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
684 			    error = EFBIG;
685 			    goto errout;
686 			}
687 			auio.uio_resid = len;
688 			do {
689 			    rcvflg = MSG_WAITALL;
690 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
691 				&auio, mp, (struct mbuf **)0, &rcvflg);
692 			} while (error == EWOULDBLOCK || error == EINTR ||
693 				 error == ERESTART);
694 			if (!error && auio.uio_resid > 0) {
695 			    if (len != auio.uio_resid)
696 			      log(LOG_INFO,
697 				"short receive (%lu/%d) from nfs server %s\n",
698 				(u_long)len - auio.uio_resid, len,
699 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
700 			    error = EPIPE;
701 			}
702 		} else {
703 			/*
704 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
705 			 * and soreceive() will return when it has either a
706 			 * control msg or a data msg.
707 			 * We have no use for control msg., but must grab them
708 			 * and then throw them away so we know what is going
709 			 * on.
710 			 */
711 			auio.uio_resid = len = 100000000; /* Anything Big */
712 			/* not need to setup uio_vmspace */
713 			do {
714 			    rcvflg = 0;
715 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
716 				&auio, mp, &control, &rcvflg);
717 			    if (control)
718 				m_freem(control);
719 			    if (error == EWOULDBLOCK && rep) {
720 				if (rep->r_flags & R_SOFTTERM)
721 					return (EINTR);
722 			    }
723 			} while (error == EWOULDBLOCK ||
724 				 (!error && *mp == NULL && control));
725 			if ((rcvflg & MSG_EOR) == 0)
726 				printf("Egad!!\n");
727 			if (!error && *mp == NULL)
728 				error = EPIPE;
729 			len -= auio.uio_resid;
730 		}
731 errout:
732 		if (error && error != EINTR && error != ERESTART) {
733 			m_freem(*mp);
734 			*mp = (struct mbuf *)0;
735 			if (error != EPIPE)
736 				log(LOG_INFO,
737 				    "receive error %d from nfs server %s\n",
738 				    error,
739 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
740 			error = nfs_sndlock(rep->r_nmp, rep);
741 			if (!error)
742 				error = nfs_reconnect(rep);
743 			if (!error)
744 				goto tryagain;
745 			else
746 				nfs_sndunlock(rep->r_nmp);
747 		}
748 	} else {
749 		if ((so = rep->r_nmp->nm_so) == NULL)
750 			return (EACCES);
751 		if (so->so_state & SS_ISCONNECTED)
752 			getnam = (struct mbuf **)0;
753 		else
754 			getnam = aname;
755 		auio.uio_resid = len = 1000000;
756 		/* not need to setup uio_vmspace */
757 		do {
758 			rcvflg = 0;
759 			error =  (*so->so_receive)(so, getnam, &auio, mp,
760 				(struct mbuf **)0, &rcvflg);
761 			if (error == EWOULDBLOCK &&
762 			    (rep->r_flags & R_SOFTTERM))
763 				return (EINTR);
764 		} while (error == EWOULDBLOCK);
765 		len -= auio.uio_resid;
766 		if (!error && *mp == NULL)
767 			error = EPIPE;
768 	}
769 	if (error) {
770 		m_freem(*mp);
771 		*mp = (struct mbuf *)0;
772 	}
773 	return (error);
774 }
775 
776 /*
777  * Implement receipt of reply on a socket.
778  * We must search through the list of received datagrams matching them
779  * with outstanding requests using the xid, until ours is found.
780  */
781 /* ARGSUSED */
782 static int
783 nfs_reply(struct nfsreq *myrep, struct lwp *lwp)
784 {
785 	struct nfsreq *rep;
786 	struct nfsmount *nmp = myrep->r_nmp;
787 	int32_t t1;
788 	struct mbuf *mrep, *nam, *md;
789 	u_int32_t rxid, *tl;
790 	char *dpos, *cp2;
791 	int error;
792 
793 	/*
794 	 * Loop around until we get our own reply
795 	 */
796 	for (;;) {
797 		/*
798 		 * Lock against other receivers so that I don't get stuck in
799 		 * sbwait() after someone else has received my reply for me.
800 		 * Also necessary for connection based protocols to avoid
801 		 * race conditions during a reconnect.
802 		 */
803 		error = nfs_rcvlock(nmp, myrep);
804 		if (error == EALREADY)
805 			return (0);
806 		if (error)
807 			return (error);
808 		/*
809 		 * Get the next Rpc reply off the socket
810 		 */
811 
812 		mutex_enter(&nmp->nm_lock);
813 		nmp->nm_waiters++;
814 		mutex_exit(&nmp->nm_lock);
815 
816 		error = nfs_receive(myrep, &nam, &mrep, lwp);
817 
818 		mutex_enter(&nmp->nm_lock);
819 		nmp->nm_waiters--;
820 		cv_signal(&nmp->nm_disconcv);
821 		mutex_exit(&nmp->nm_lock);
822 
823 		if (error) {
824 			nfs_rcvunlock(nmp);
825 
826 			if (nmp->nm_iflag & NFSMNT_DISMNT) {
827 				/*
828 				 * Oops, we're going away now..
829 				 */
830 				return error;
831 			}
832 			/*
833 			 * Ignore routing errors on connectionless protocols? ?
834 			 */
835 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
836 				nmp->nm_so->so_error = 0;
837 #ifdef DEBUG
838 				if (ratecheck(&nfs_reply_last_err_time,
839 				    &nfs_err_interval))
840 					printf("%s: ignoring error %d\n",
841 					       __func__, error);
842 #endif
843 				continue;
844 			}
845 			return (error);
846 		}
847 		if (nam)
848 			m_freem(nam);
849 
850 		/*
851 		 * Get the xid and check that it is an rpc reply
852 		 */
853 		md = mrep;
854 		dpos = mtod(md, void *);
855 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
856 		rxid = *tl++;
857 		if (*tl != rpc_reply) {
858 			nfsstats.rpcinvalid++;
859 			m_freem(mrep);
860 nfsmout:
861 			nfs_rcvunlock(nmp);
862 			continue;
863 		}
864 
865 		/*
866 		 * Loop through the request list to match up the reply
867 		 * Iff no match, just drop the datagram
868 		 */
869 		TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
870 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
871 				/* Found it.. */
872 				rep->r_mrep = mrep;
873 				rep->r_md = md;
874 				rep->r_dpos = dpos;
875 				if (nfsrtton) {
876 					struct rttl *rt;
877 
878 					rt = &nfsrtt.rttl[nfsrtt.pos];
879 					rt->proc = rep->r_procnum;
880 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
881 					rt->sent = nmp->nm_sent;
882 					rt->cwnd = nmp->nm_cwnd;
883 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
884 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
885 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx;
886 					getmicrotime(&rt->tstamp);
887 					if (rep->r_flags & R_TIMING)
888 						rt->rtt = rep->r_rtt;
889 					else
890 						rt->rtt = 1000000;
891 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
892 				}
893 				/*
894 				 * Update congestion window.
895 				 * Do the additive increase of
896 				 * one rpc/rtt.
897 				 */
898 				if (nmp->nm_cwnd <= nmp->nm_sent) {
899 					nmp->nm_cwnd +=
900 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
901 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
902 					if (nmp->nm_cwnd > NFS_MAXCWND)
903 						nmp->nm_cwnd = NFS_MAXCWND;
904 				}
905 				rep->r_flags &= ~R_SENT;
906 				nmp->nm_sent -= NFS_CWNDSCALE;
907 				/*
908 				 * Update rtt using a gain of 0.125 on the mean
909 				 * and a gain of 0.25 on the deviation.
910 				 */
911 				if (rep->r_flags & R_TIMING) {
912 					/*
913 					 * Since the timer resolution of
914 					 * NFS_HZ is so course, it can often
915 					 * result in r_rtt == 0. Since
916 					 * r_rtt == N means that the actual
917 					 * rtt is between N+dt and N+2-dt ticks,
918 					 * add 1.
919 					 */
920 					t1 = rep->r_rtt + 1;
921 					t1 -= (NFS_SRTT(rep) >> 3);
922 					NFS_SRTT(rep) += t1;
923 					if (t1 < 0)
924 						t1 = -t1;
925 					t1 -= (NFS_SDRTT(rep) >> 2);
926 					NFS_SDRTT(rep) += t1;
927 				}
928 				nmp->nm_timeouts = 0;
929 				break;
930 			}
931 		}
932 		nfs_rcvunlock(nmp);
933 		/*
934 		 * If not matched to a request, drop it.
935 		 * If it's mine, get out.
936 		 */
937 		if (rep == 0) {
938 			nfsstats.rpcunexpected++;
939 			m_freem(mrep);
940 		} else if (rep == myrep) {
941 			if (rep->r_mrep == NULL)
942 				panic("nfsreply nil");
943 			return (0);
944 		}
945 	}
946 }
947 
948 /*
949  * nfs_request - goes something like this
950  *	- fill in request struct
951  *	- links it into list
952  *	- calls nfs_send() for first transmit
953  *	- calls nfs_receive() to get reply
954  *	- break down rpc header and return with nfs reply pointed to
955  *	  by mrep or error
956  * nb: always frees up mreq mbuf list
957  */
958 int
959 nfs_request(struct nfsnode *np, struct mbuf *mrest, int procnum, struct lwp *lwp, kauth_cred_t cred, struct mbuf **mrp, struct mbuf **mdp, char **dposp, int *rexmitp)
960 {
961 	struct mbuf *m, *mrep;
962 	struct nfsreq *rep;
963 	u_int32_t *tl;
964 	int i;
965 	struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount);
966 	struct mbuf *md, *mheadend;
967 	char nickv[RPCX_NICKVERF];
968 	time_t waituntil;
969 	char *dpos, *cp2;
970 	int t1, s, error = 0, mrest_len, auth_len, auth_type;
971 	int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0;
972 	int verf_len, verf_type;
973 	u_int32_t xid;
974 	char *auth_str, *verf_str;
975 	NFSKERBKEY_T key;		/* save session key */
976 	kauth_cred_t acred;
977 	struct mbuf *mrest_backup = NULL;
978 	kauth_cred_t origcred = NULL; /* XXX: gcc */
979 	bool retry_cred = true;
980 	bool use_opencred = (np->n_flag & NUSEOPENCRED) != 0;
981 
982 	if (rexmitp != NULL)
983 		*rexmitp = 0;
984 
985 	acred = kauth_cred_alloc();
986 
987 tryagain_cred:
988 	KASSERT(cred != NULL);
989 	rep = kmem_alloc(sizeof(*rep), KM_SLEEP);
990 	rep->r_nmp = nmp;
991 	KASSERT(lwp == NULL || lwp == curlwp);
992 	rep->r_lwp = lwp;
993 	rep->r_procnum = procnum;
994 	i = 0;
995 	m = mrest;
996 	while (m) {
997 		i += m->m_len;
998 		m = m->m_next;
999 	}
1000 	mrest_len = i;
1001 
1002 	/*
1003 	 * Get the RPC header with authorization.
1004 	 */
1005 kerbauth:
1006 	verf_str = auth_str = (char *)0;
1007 	if (nmp->nm_flag & NFSMNT_KERB) {
1008 		verf_str = nickv;
1009 		verf_len = sizeof (nickv);
1010 		auth_type = RPCAUTH_KERB4;
1011 		memset((void *)key, 0, sizeof (key));
1012 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
1013 			&auth_len, verf_str, verf_len)) {
1014 			error = nfs_getauth(nmp, rep, cred, &auth_str,
1015 				&auth_len, verf_str, &verf_len, key);
1016 			if (error) {
1017 				kmem_free(rep, sizeof(*rep));
1018 				m_freem(mrest);
1019 				KASSERT(kauth_cred_getrefcnt(acred) == 1);
1020 				kauth_cred_free(acred);
1021 				return (error);
1022 			}
1023 		}
1024 		retry_cred = false;
1025 	} else {
1026 		/* AUTH_UNIX */
1027 		uid_t uid;
1028 		gid_t gid;
1029 
1030 		/*
1031 		 * on the most unix filesystems, permission checks are
1032 		 * done when the file is open(2)'ed.
1033 		 * ie. once a file is successfully open'ed,
1034 		 * following i/o operations never fail with EACCES.
1035 		 * we try to follow the semantics as far as possible.
1036 		 *
1037 		 * note that we expect that the nfs server always grant
1038 		 * accesses by the file's owner.
1039 		 */
1040 		origcred = cred;
1041 		switch (procnum) {
1042 		case NFSPROC_READ:
1043 		case NFSPROC_WRITE:
1044 		case NFSPROC_COMMIT:
1045 			uid = np->n_vattr->va_uid;
1046 			gid = np->n_vattr->va_gid;
1047 			if (kauth_cred_geteuid(cred) == uid &&
1048 			    kauth_cred_getegid(cred) == gid) {
1049 				retry_cred = false;
1050 				break;
1051 			}
1052 			if (use_opencred)
1053 				break;
1054 			kauth_cred_setuid(acred, uid);
1055 			kauth_cred_seteuid(acred, uid);
1056 			kauth_cred_setsvuid(acred, uid);
1057 			kauth_cred_setgid(acred, gid);
1058 			kauth_cred_setegid(acred, gid);
1059 			kauth_cred_setsvgid(acred, gid);
1060 			cred = acred;
1061 			break;
1062 		default:
1063 			retry_cred = false;
1064 			break;
1065 		}
1066 		/*
1067 		 * backup mbuf chain if we can need it later to retry.
1068 		 *
1069 		 * XXX maybe we can keep a direct reference to
1070 		 * mrest without doing m_copym, but it's ...ugly.
1071 		 */
1072 		if (retry_cred)
1073 			mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT);
1074 		auth_type = RPCAUTH_UNIX;
1075 		/* XXX elad - ngroups */
1076 		auth_len = (((kauth_cred_ngroups(cred) > nmp->nm_numgrps) ?
1077 			nmp->nm_numgrps : kauth_cred_ngroups(cred)) << 2) +
1078 			5 * NFSX_UNSIGNED;
1079 	}
1080 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
1081 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
1082 	if (auth_str)
1083 		free(auth_str, M_TEMP);
1084 
1085 	/*
1086 	 * For stream protocols, insert a Sun RPC Record Mark.
1087 	 */
1088 	if (nmp->nm_sotype == SOCK_STREAM) {
1089 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
1090 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
1091 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
1092 	}
1093 	rep->r_mreq = m;
1094 	rep->r_xid = xid;
1095 tryagain:
1096 	if (nmp->nm_flag & NFSMNT_SOFT)
1097 		rep->r_retry = nmp->nm_retry;
1098 	else
1099 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
1100 	rep->r_rtt = rep->r_rexmit = 0;
1101 	if (proct[procnum] > 0)
1102 		rep->r_flags = R_TIMING;
1103 	else
1104 		rep->r_flags = 0;
1105 	rep->r_mrep = NULL;
1106 
1107 	/*
1108 	 * Do the client side RPC.
1109 	 */
1110 	nfsstats.rpcrequests++;
1111 	/*
1112 	 * Chain request into list of outstanding requests. Be sure
1113 	 * to put it LAST so timer finds oldest requests first.
1114 	 */
1115 	s = splsoftnet();
1116 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1117 	nfs_timer_start();
1118 
1119 	/*
1120 	 * If backing off another request or avoiding congestion, don't
1121 	 * send this one now but let timer do it. If not timing a request,
1122 	 * do it now.
1123 	 */
1124 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1125 	    (nmp->nm_flag & NFSMNT_DUMBTIMR) || nmp->nm_sent < nmp->nm_cwnd)) {
1126 		splx(s);
1127 		if (nmp->nm_soflags & PR_CONNREQUIRED)
1128 			error = nfs_sndlock(nmp, rep);
1129 		if (!error) {
1130 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
1131 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp);
1132 			if (nmp->nm_soflags & PR_CONNREQUIRED)
1133 				nfs_sndunlock(nmp);
1134 		}
1135 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1136 			nmp->nm_sent += NFS_CWNDSCALE;
1137 			rep->r_flags |= R_SENT;
1138 		}
1139 	} else {
1140 		splx(s);
1141 		rep->r_rtt = -1;
1142 	}
1143 
1144 	/*
1145 	 * Wait for the reply from our send or the timer's.
1146 	 */
1147 	if (!error || error == EPIPE)
1148 		error = nfs_reply(rep, lwp);
1149 
1150 	/*
1151 	 * RPC done, unlink the request.
1152 	 */
1153 	s = splsoftnet();
1154 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1155 	splx(s);
1156 
1157 	/*
1158 	 * Decrement the outstanding request count.
1159 	 */
1160 	if (rep->r_flags & R_SENT) {
1161 		rep->r_flags &= ~R_SENT;	/* paranoia */
1162 		nmp->nm_sent -= NFS_CWNDSCALE;
1163 	}
1164 
1165 	if (rexmitp != NULL) {
1166 		int rexmit;
1167 
1168 		if (nmp->nm_sotype != SOCK_DGRAM)
1169 			rexmit = (rep->r_flags & R_REXMITTED) != 0;
1170 		else
1171 			rexmit = rep->r_rexmit;
1172 		*rexmitp = rexmit;
1173 	}
1174 
1175 	/*
1176 	 * If there was a successful reply and a tprintf msg.
1177 	 * tprintf a response.
1178 	 */
1179 	if (!error && (rep->r_flags & R_TPRINTFMSG))
1180 		nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1181 		    "is alive again");
1182 	mrep = rep->r_mrep;
1183 	md = rep->r_md;
1184 	dpos = rep->r_dpos;
1185 	if (error)
1186 		goto nfsmout;
1187 
1188 	/*
1189 	 * break down the rpc header and check if ok
1190 	 */
1191 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1192 	if (*tl++ == rpc_msgdenied) {
1193 		if (*tl == rpc_mismatch)
1194 			error = EOPNOTSUPP;
1195 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1196 			if (!failed_auth) {
1197 				failed_auth++;
1198 				mheadend->m_next = (struct mbuf *)0;
1199 				m_freem(mrep);
1200 				m_freem(rep->r_mreq);
1201 				goto kerbauth;
1202 			} else
1203 				error = EAUTH;
1204 		} else
1205 			error = EACCES;
1206 		m_freem(mrep);
1207 		goto nfsmout;
1208 	}
1209 
1210 	/*
1211 	 * Grab any Kerberos verifier, otherwise just throw it away.
1212 	 */
1213 	verf_type = fxdr_unsigned(int, *tl++);
1214 	i = fxdr_unsigned(int32_t, *tl);
1215 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1216 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1217 		if (error)
1218 			goto nfsmout;
1219 	} else if (i > 0)
1220 		nfsm_adv(nfsm_rndup(i));
1221 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1222 	/* 0 == ok */
1223 	if (*tl == 0) {
1224 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1225 		if (*tl != 0) {
1226 			error = fxdr_unsigned(int, *tl);
1227 			switch (error) {
1228 			case NFSERR_PERM:
1229 				error = EPERM;
1230 				break;
1231 
1232 			case NFSERR_NOENT:
1233 				error = ENOENT;
1234 				break;
1235 
1236 			case NFSERR_IO:
1237 				error = EIO;
1238 				break;
1239 
1240 			case NFSERR_NXIO:
1241 				error = ENXIO;
1242 				break;
1243 
1244 			case NFSERR_ACCES:
1245 				error = EACCES;
1246 				if (!retry_cred)
1247 					break;
1248 				m_freem(mrep);
1249 				m_freem(rep->r_mreq);
1250 				kmem_free(rep, sizeof(*rep));
1251 				use_opencred = !use_opencred;
1252 				if (mrest_backup == NULL) {
1253 					/* m_copym failure */
1254 					KASSERT(
1255 					    kauth_cred_getrefcnt(acred) == 1);
1256 					kauth_cred_free(acred);
1257 					return ENOMEM;
1258 				}
1259 				mrest = mrest_backup;
1260 				mrest_backup = NULL;
1261 				cred = origcred;
1262 				error = 0;
1263 				retry_cred = false;
1264 				goto tryagain_cred;
1265 
1266 			case NFSERR_EXIST:
1267 				error = EEXIST;
1268 				break;
1269 
1270 			case NFSERR_XDEV:
1271 				error = EXDEV;
1272 				break;
1273 
1274 			case NFSERR_NODEV:
1275 				error = ENODEV;
1276 				break;
1277 
1278 			case NFSERR_NOTDIR:
1279 				error = ENOTDIR;
1280 				break;
1281 
1282 			case NFSERR_ISDIR:
1283 				error = EISDIR;
1284 				break;
1285 
1286 			case NFSERR_INVAL:
1287 				error = EINVAL;
1288 				break;
1289 
1290 			case NFSERR_FBIG:
1291 				error = EFBIG;
1292 				break;
1293 
1294 			case NFSERR_NOSPC:
1295 				error = ENOSPC;
1296 				break;
1297 
1298 			case NFSERR_ROFS:
1299 				error = EROFS;
1300 				break;
1301 
1302 			case NFSERR_MLINK:
1303 				error = EMLINK;
1304 				break;
1305 
1306 			case NFSERR_TIMEDOUT:
1307 				error = ETIMEDOUT;
1308 				break;
1309 
1310 			case NFSERR_NAMETOL:
1311 				error = ENAMETOOLONG;
1312 				break;
1313 
1314 			case NFSERR_NOTEMPTY:
1315 				error = ENOTEMPTY;
1316 				break;
1317 
1318 			case NFSERR_DQUOT:
1319 				error = EDQUOT;
1320 				break;
1321 
1322 			case NFSERR_STALE:
1323 				/*
1324 				 * If the File Handle was stale, invalidate the
1325 				 * lookup cache, just in case.
1326 				 */
1327 				error = ESTALE;
1328 				cache_purge(NFSTOV(np));
1329 				break;
1330 
1331 			case NFSERR_REMOTE:
1332 				error = EREMOTE;
1333 				break;
1334 
1335 			case NFSERR_WFLUSH:
1336 			case NFSERR_BADHANDLE:
1337 			case NFSERR_NOT_SYNC:
1338 			case NFSERR_BAD_COOKIE:
1339 				error = EINVAL;
1340 				break;
1341 
1342 			case NFSERR_NOTSUPP:
1343 				error = ENOTSUP;
1344 				break;
1345 
1346 			case NFSERR_TOOSMALL:
1347 			case NFSERR_SERVERFAULT:
1348 			case NFSERR_BADTYPE:
1349 				error = EINVAL;
1350 				break;
1351 
1352 			case NFSERR_TRYLATER:
1353 				if ((nmp->nm_flag & NFSMNT_NFSV3) == 0)
1354 					break;
1355 				m_freem(mrep);
1356 				error = 0;
1357 				waituntil = time_second + trylater_delay;
1358 				while (time_second < waituntil) {
1359 					kpause("nfstrylater", false, hz, NULL);
1360 				}
1361 				trylater_delay *= NFS_TRYLATERDELMUL;
1362 				if (trylater_delay > NFS_TRYLATERDELMAX)
1363 					trylater_delay = NFS_TRYLATERDELMAX;
1364 				/*
1365 				 * RFC1813:
1366 				 * The client should wait and then try
1367 				 * the request with a new RPC transaction ID.
1368 				 */
1369 				nfs_renewxid(rep);
1370 				goto tryagain;
1371 
1372 			default:
1373 #ifdef DIAGNOSTIC
1374 				printf("Invalid rpc error code %d\n", error);
1375 #endif
1376 				error = EINVAL;
1377 				break;
1378 			}
1379 
1380 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1381 				*mrp = mrep;
1382 				*mdp = md;
1383 				*dposp = dpos;
1384 				error |= NFSERR_RETERR;
1385 			} else
1386 				m_freem(mrep);
1387 			goto nfsmout;
1388 		}
1389 
1390 		/*
1391 		 * note which credential worked to minimize number of retries.
1392 		 */
1393 		if (use_opencred)
1394 			np->n_flag |= NUSEOPENCRED;
1395 		else
1396 			np->n_flag &= ~NUSEOPENCRED;
1397 
1398 		*mrp = mrep;
1399 		*mdp = md;
1400 		*dposp = dpos;
1401 
1402 		KASSERT(error == 0);
1403 		goto nfsmout;
1404 	}
1405 	m_freem(mrep);
1406 	error = EPROTONOSUPPORT;
1407 nfsmout:
1408 	KASSERT(kauth_cred_getrefcnt(acred) == 1);
1409 	kauth_cred_free(acred);
1410 	m_freem(rep->r_mreq);
1411 	kmem_free(rep, sizeof(*rep));
1412 	m_freem(mrest_backup);
1413 	return (error);
1414 }
1415 #endif /* NFS */
1416 
1417 /*
1418  * Generate the rpc reply header
1419  * siz arg. is used to decide if adding a cluster is worthwhile
1420  */
1421 int
1422 nfs_rephead(int siz, struct nfsrv_descript *nd, struct nfssvc_sock *slp, int err, int cache, u_quad_t *frev, struct mbuf **mrq, struct mbuf **mbp, char **bposp)
1423 {
1424 	u_int32_t *tl;
1425 	struct mbuf *mreq;
1426 	char *bpos;
1427 	struct mbuf *mb;
1428 
1429 	mreq = m_gethdr(M_WAIT, MT_DATA);
1430 	MCLAIM(mreq, &nfs_mowner);
1431 	mb = mreq;
1432 	/*
1433 	 * If this is a big reply, use a cluster else
1434 	 * try and leave leading space for the lower level headers.
1435 	 */
1436 	siz += RPC_REPLYSIZ;
1437 	if (siz >= max_datalen) {
1438 		m_clget(mreq, M_WAIT);
1439 	} else
1440 		mreq->m_data += max_hdr;
1441 	tl = mtod(mreq, u_int32_t *);
1442 	mreq->m_len = 6 * NFSX_UNSIGNED;
1443 	bpos = ((char *)tl) + mreq->m_len;
1444 	*tl++ = txdr_unsigned(nd->nd_retxid);
1445 	*tl++ = rpc_reply;
1446 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1447 		*tl++ = rpc_msgdenied;
1448 		if (err & NFSERR_AUTHERR) {
1449 			*tl++ = rpc_autherr;
1450 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1451 			mreq->m_len -= NFSX_UNSIGNED;
1452 			bpos -= NFSX_UNSIGNED;
1453 		} else {
1454 			*tl++ = rpc_mismatch;
1455 			*tl++ = txdr_unsigned(RPC_VER2);
1456 			*tl = txdr_unsigned(RPC_VER2);
1457 		}
1458 	} else {
1459 		*tl++ = rpc_msgaccepted;
1460 
1461 		/*
1462 		 * For Kerberos authentication, we must send the nickname
1463 		 * verifier back, otherwise just RPCAUTH_NULL.
1464 		 */
1465 		if (nd->nd_flag & ND_KERBFULL) {
1466 			struct nfsuid *nuidp;
1467 			struct timeval ktvin, ktvout;
1468 
1469 			memset(&ktvout, 0, sizeof ktvout);	/* XXX gcc */
1470 
1471 			LIST_FOREACH(nuidp,
1472 			    NUIDHASH(slp, kauth_cred_geteuid(nd->nd_cr)),
1473 			    nu_hash) {
1474 				if (kauth_cred_geteuid(nuidp->nu_cr) ==
1475 				kauth_cred_geteuid(nd->nd_cr) &&
1476 				    (!nd->nd_nam2 || netaddr_match(
1477 				    NU_NETFAM(nuidp), &nuidp->nu_haddr,
1478 				    nd->nd_nam2)))
1479 					break;
1480 			}
1481 			if (nuidp) {
1482 				ktvin.tv_sec =
1483 				    txdr_unsigned(nuidp->nu_timestamp.tv_sec
1484 					- 1);
1485 				ktvin.tv_usec =
1486 				    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1487 
1488 				/*
1489 				 * Encrypt the timestamp in ecb mode using the
1490 				 * session key.
1491 				 */
1492 #ifdef NFSKERB
1493 				XXX
1494 #endif
1495 
1496 				*tl++ = rpc_auth_kerb;
1497 				*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1498 				*tl = ktvout.tv_sec;
1499 				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1500 				*tl++ = ktvout.tv_usec;
1501 				*tl++ = txdr_unsigned(
1502 				    kauth_cred_geteuid(nuidp->nu_cr));
1503 			} else {
1504 				*tl++ = 0;
1505 				*tl++ = 0;
1506 			}
1507 		} else {
1508 			*tl++ = 0;
1509 			*tl++ = 0;
1510 		}
1511 		switch (err) {
1512 		case EPROGUNAVAIL:
1513 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1514 			break;
1515 		case EPROGMISMATCH:
1516 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1517 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1518 			*tl++ = txdr_unsigned(2);
1519 			*tl = txdr_unsigned(3);
1520 			break;
1521 		case EPROCUNAVAIL:
1522 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1523 			break;
1524 		case EBADRPC:
1525 			*tl = txdr_unsigned(RPC_GARBAGE);
1526 			break;
1527 		default:
1528 			*tl = 0;
1529 			if (err != NFSERR_RETVOID) {
1530 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1531 				if (err)
1532 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1533 				else
1534 				    *tl = 0;
1535 			}
1536 			break;
1537 		};
1538 	}
1539 
1540 	if (mrq != NULL)
1541 		*mrq = mreq;
1542 	*mbp = mb;
1543 	*bposp = bpos;
1544 	if (err != 0 && err != NFSERR_RETVOID)
1545 		nfsstats.srvrpc_errs++;
1546 	return (0);
1547 }
1548 
1549 static void
1550 nfs_timer_schedule(void)
1551 {
1552 
1553 	callout_schedule(&nfs_timer_ch, nfs_ticks);
1554 }
1555 
1556 void
1557 nfs_timer_start(void)
1558 {
1559 
1560 	if (callout_pending(&nfs_timer_ch))
1561 		return;
1562 
1563 	nfs_timer_start_ev.ev_count++;
1564 	nfs_timer_schedule();
1565 }
1566 
1567 void
1568 nfs_timer_init(void)
1569 {
1570 
1571 	mutex_init(&nfs_timer_lock, MUTEX_DEFAULT, IPL_NONE);
1572 	callout_init(&nfs_timer_ch, 0);
1573 	callout_setfunc(&nfs_timer_ch, nfs_timer, NULL);
1574 	evcnt_attach_dynamic(&nfs_timer_ev, EVCNT_TYPE_MISC, NULL,
1575 	    "nfs", "timer");
1576 	evcnt_attach_dynamic(&nfs_timer_start_ev, EVCNT_TYPE_MISC, NULL,
1577 	    "nfs", "timer start");
1578 	evcnt_attach_dynamic(&nfs_timer_stop_ev, EVCNT_TYPE_MISC, NULL,
1579 	    "nfs", "timer stop");
1580 }
1581 
1582 void
1583 nfs_timer_fini(void)
1584 {
1585 
1586 	callout_halt(&nfs_timer_ch, NULL);
1587 	callout_destroy(&nfs_timer_ch);
1588 	mutex_destroy(&nfs_timer_lock);
1589 	evcnt_detach(&nfs_timer_ev);
1590 	evcnt_detach(&nfs_timer_start_ev);
1591 	evcnt_detach(&nfs_timer_stop_ev);
1592 }
1593 
1594 void
1595 nfs_timer_srvinit(bool (*func)(void))
1596 {
1597 
1598 	nfs_timer_srvvec = func;
1599 }
1600 
1601 void
1602 nfs_timer_srvfini(void)
1603 {
1604 
1605 	mutex_enter(&nfs_timer_lock);
1606 	nfs_timer_srvvec = NULL;
1607 	mutex_exit(&nfs_timer_lock);
1608 }
1609 
1610 
1611 /*
1612  * Nfs timer routine
1613  * Scan the nfsreq list and retranmit any requests that have timed out
1614  * To avoid retransmission attempts on STREAM sockets (in the future) make
1615  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1616  */
1617 void
1618 nfs_timer(void *arg)
1619 {
1620 	struct nfsreq *rep;
1621 	struct mbuf *m;
1622 	struct socket *so;
1623 	struct nfsmount *nmp;
1624 	int timeo;
1625 	int error;
1626 	bool more = false;
1627 
1628 	nfs_timer_ev.ev_count++;
1629 
1630 	mutex_enter(softnet_lock);	/* XXX PR 40491 */
1631 	TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
1632 		more = true;
1633 		nmp = rep->r_nmp;
1634 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1635 			continue;
1636 		if (nfs_sigintr(nmp, rep, rep->r_lwp)) {
1637 			rep->r_flags |= R_SOFTTERM;
1638 			continue;
1639 		}
1640 		if (rep->r_rtt >= 0) {
1641 			rep->r_rtt++;
1642 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1643 				timeo = nmp->nm_timeo;
1644 			else
1645 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1646 			if (nmp->nm_timeouts > 0)
1647 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1648 			if (timeo > NFS_MAXTIMEO)
1649 				timeo = NFS_MAXTIMEO;
1650 			if (rep->r_rtt <= timeo)
1651 				continue;
1652 			if (nmp->nm_timeouts <
1653 			    (sizeof(nfs_backoff) / sizeof(nfs_backoff[0])))
1654 				nmp->nm_timeouts++;
1655 		}
1656 		/*
1657 		 * Check for server not responding
1658 		 */
1659 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1660 		     rep->r_rexmit > nmp->nm_deadthresh) {
1661 			nfs_msg(rep->r_lwp,
1662 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1663 			    "not responding");
1664 			rep->r_flags |= R_TPRINTFMSG;
1665 		}
1666 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1667 			nfsstats.rpctimeouts++;
1668 			rep->r_flags |= R_SOFTTERM;
1669 			continue;
1670 		}
1671 		if (nmp->nm_sotype != SOCK_DGRAM) {
1672 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1673 				rep->r_rexmit = NFS_MAXREXMIT;
1674 			continue;
1675 		}
1676 		if ((so = nmp->nm_so) == NULL)
1677 			continue;
1678 
1679 		/*
1680 		 * If there is enough space and the window allows..
1681 		 *	Resend it
1682 		 * Set r_rtt to -1 in case we fail to send it now.
1683 		 */
1684 		/* solock(so);		XXX PR 40491 */
1685 		rep->r_rtt = -1;
1686 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1687 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1688 		    (rep->r_flags & R_SENT) ||
1689 		    nmp->nm_sent < nmp->nm_cwnd) &&
1690 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1691 		        if (so->so_state & SS_ISCONNECTED)
1692 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1693 			    (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0);
1694 			else
1695 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1696 			    nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0);
1697 			if (error) {
1698 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1699 #ifdef DEBUG
1700 					if (ratecheck(&nfs_timer_last_err_time,
1701 					    &nfs_err_interval))
1702 						printf("%s: ignoring error "
1703 						       "%d\n", __func__, error);
1704 #endif
1705 					so->so_error = 0;
1706 				}
1707 			} else {
1708 				/*
1709 				 * Iff first send, start timing
1710 				 * else turn timing off, backoff timer
1711 				 * and divide congestion window by 2.
1712 				 */
1713 				if (rep->r_flags & R_SENT) {
1714 					rep->r_flags &= ~R_TIMING;
1715 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1716 						rep->r_rexmit = NFS_MAXREXMIT;
1717 					nmp->nm_cwnd >>= 1;
1718 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1719 						nmp->nm_cwnd = NFS_CWNDSCALE;
1720 					nfsstats.rpcretries++;
1721 				} else {
1722 					rep->r_flags |= R_SENT;
1723 					nmp->nm_sent += NFS_CWNDSCALE;
1724 				}
1725 				rep->r_rtt = 0;
1726 			}
1727 		}
1728 		/* sounlock(so);	XXX PR 40491 */
1729 	}
1730 	mutex_exit(softnet_lock);	/* XXX PR 40491 */
1731 
1732 	mutex_enter(&nfs_timer_lock);
1733 	if (nfs_timer_srvvec != NULL) {
1734 		more |= (*nfs_timer_srvvec)();
1735 	}
1736 	mutex_exit(&nfs_timer_lock);
1737 
1738 	if (more) {
1739 		nfs_timer_schedule();
1740 	} else {
1741 		nfs_timer_stop_ev.ev_count++;
1742 	}
1743 }
1744 
1745 /*
1746  * Test for a termination condition pending on the process.
1747  * This is used for NFSMNT_INT mounts.
1748  */
1749 int
1750 nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct lwp *l)
1751 {
1752 	sigset_t ss;
1753 
1754 	if (rep && (rep->r_flags & R_SOFTTERM))
1755 		return (EINTR);
1756 	if (!(nmp->nm_flag & NFSMNT_INT))
1757 		return (0);
1758 	if (l) {
1759 		sigpending1(l, &ss);
1760 #if 0
1761 		sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss);
1762 #endif
1763 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1764 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1765 		    sigismember(&ss, SIGQUIT))
1766 			return (EINTR);
1767 	}
1768 	return (0);
1769 }
1770 
1771 #ifdef NFS
1772 /*
1773  * Lock a socket against others.
1774  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1775  * and also to avoid race conditions between the processes with nfs requests
1776  * in progress when a reconnect is necessary.
1777  */
1778 static int
1779 nfs_sndlock(struct nfsmount *nmp, struct nfsreq *rep)
1780 {
1781 	struct lwp *l;
1782 	int timeo = 0;
1783 	bool catch = false;
1784 	int error = 0;
1785 
1786 	if (rep) {
1787 		l = rep->r_lwp;
1788 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1789 			catch = true;
1790 	} else
1791 		l = NULL;
1792 	mutex_enter(&nmp->nm_lock);
1793 	while ((nmp->nm_iflag & NFSMNT_SNDLOCK) != 0) {
1794 		if (rep && nfs_sigintr(rep->r_nmp, rep, l)) {
1795 			error = EINTR;
1796 			goto quit;
1797 		}
1798 		if (catch) {
1799 			cv_timedwait_sig(&nmp->nm_sndcv, &nmp->nm_lock, timeo);
1800 		} else {
1801 			cv_timedwait(&nmp->nm_sndcv, &nmp->nm_lock, timeo);
1802 		}
1803 		if (catch) {
1804 			catch = false;
1805 			timeo = 2 * hz;
1806 		}
1807 	}
1808 	nmp->nm_iflag |= NFSMNT_SNDLOCK;
1809 quit:
1810 	mutex_exit(&nmp->nm_lock);
1811 	return error;
1812 }
1813 
1814 /*
1815  * Unlock the stream socket for others.
1816  */
1817 static void
1818 nfs_sndunlock(struct nfsmount *nmp)
1819 {
1820 
1821 	mutex_enter(&nmp->nm_lock);
1822 	if ((nmp->nm_iflag & NFSMNT_SNDLOCK) == 0)
1823 		panic("nfs sndunlock");
1824 	nmp->nm_iflag &= ~NFSMNT_SNDLOCK;
1825 	cv_signal(&nmp->nm_sndcv);
1826 	mutex_exit(&nmp->nm_lock);
1827 }
1828 #endif /* NFS */
1829 
1830 static int
1831 nfs_rcvlock(struct nfsmount *nmp, struct nfsreq *rep)
1832 {
1833 	int *flagp = &nmp->nm_iflag;
1834 	int slptimeo = 0;
1835 	bool catch;
1836 	int error = 0;
1837 
1838 	KASSERT(nmp == rep->r_nmp);
1839 
1840 	catch = (nmp->nm_flag & NFSMNT_INT) != 0;
1841 	mutex_enter(&nmp->nm_lock);
1842 	while (/* CONSTCOND */ true) {
1843 		if (*flagp & NFSMNT_DISMNT) {
1844 			cv_signal(&nmp->nm_disconcv);
1845 			error = EIO;
1846 			break;
1847 		}
1848 		/* If our reply was received while we were sleeping,
1849 		 * then just return without taking the lock to avoid a
1850 		 * situation where a single iod could 'capture' the
1851 		 * receive lock.
1852 		 */
1853 		if (rep->r_mrep != NULL) {
1854 			error = EALREADY;
1855 			break;
1856 		}
1857 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) {
1858 			error = EINTR;
1859 			break;
1860 		}
1861 		if ((*flagp & NFSMNT_RCVLOCK) == 0) {
1862 			*flagp |= NFSMNT_RCVLOCK;
1863 			break;
1864 		}
1865 		if (catch) {
1866 			cv_timedwait_sig(&nmp->nm_rcvcv, &nmp->nm_lock,
1867 			    slptimeo);
1868 		} else {
1869 			cv_timedwait(&nmp->nm_rcvcv, &nmp->nm_lock,
1870 			    slptimeo);
1871 		}
1872 		if (catch) {
1873 			catch = false;
1874 			slptimeo = 2 * hz;
1875 		}
1876 	}
1877 	mutex_exit(&nmp->nm_lock);
1878 	return error;
1879 }
1880 
1881 /*
1882  * Unlock the stream socket for others.
1883  */
1884 static void
1885 nfs_rcvunlock(struct nfsmount *nmp)
1886 {
1887 
1888 	mutex_enter(&nmp->nm_lock);
1889 	if ((nmp->nm_iflag & NFSMNT_RCVLOCK) == 0)
1890 		panic("nfs rcvunlock");
1891 	nmp->nm_iflag &= ~NFSMNT_RCVLOCK;
1892 	cv_broadcast(&nmp->nm_rcvcv);
1893 	mutex_exit(&nmp->nm_lock);
1894 }
1895 
1896 /*
1897  * Parse an RPC request
1898  * - verify it
1899  * - allocate and fill in the cred.
1900  */
1901 int
1902 nfs_getreq(struct nfsrv_descript *nd, struct nfsd *nfsd, int has_header)
1903 {
1904 	int len, i;
1905 	u_int32_t *tl;
1906 	int32_t t1;
1907 	struct uio uio;
1908 	struct iovec iov;
1909 	char *dpos, *cp2, *cp;
1910 	u_int32_t nfsvers, auth_type;
1911 	uid_t nickuid;
1912 	int error = 0, ticklen;
1913 	struct mbuf *mrep, *md;
1914 	struct nfsuid *nuidp;
1915 	struct timeval tvin, tvout;
1916 
1917 	memset(&tvout, 0, sizeof tvout);	/* XXX gcc */
1918 
1919 	KASSERT(nd->nd_cr == NULL);
1920 	mrep = nd->nd_mrep;
1921 	md = nd->nd_md;
1922 	dpos = nd->nd_dpos;
1923 	if (has_header) {
1924 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1925 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1926 		if (*tl++ != rpc_call) {
1927 			m_freem(mrep);
1928 			return (EBADRPC);
1929 		}
1930 	} else
1931 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1932 	nd->nd_repstat = 0;
1933 	nd->nd_flag = 0;
1934 	if (*tl++ != rpc_vers) {
1935 		nd->nd_repstat = ERPCMISMATCH;
1936 		nd->nd_procnum = NFSPROC_NOOP;
1937 		return (0);
1938 	}
1939 	if (*tl != nfs_prog) {
1940 		nd->nd_repstat = EPROGUNAVAIL;
1941 		nd->nd_procnum = NFSPROC_NOOP;
1942 		return (0);
1943 	}
1944 	tl++;
1945 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1946 	if (nfsvers < NFS_VER2 || nfsvers > NFS_VER3) {
1947 		nd->nd_repstat = EPROGMISMATCH;
1948 		nd->nd_procnum = NFSPROC_NOOP;
1949 		return (0);
1950 	}
1951 	if (nfsvers == NFS_VER3)
1952 		nd->nd_flag = ND_NFSV3;
1953 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1954 	if (nd->nd_procnum == NFSPROC_NULL)
1955 		return (0);
1956 	if (nd->nd_procnum > NFSPROC_COMMIT ||
1957 	    (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1958 		nd->nd_repstat = EPROCUNAVAIL;
1959 		nd->nd_procnum = NFSPROC_NOOP;
1960 		return (0);
1961 	}
1962 	if ((nd->nd_flag & ND_NFSV3) == 0)
1963 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1964 	auth_type = *tl++;
1965 	len = fxdr_unsigned(int, *tl++);
1966 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1967 		m_freem(mrep);
1968 		return (EBADRPC);
1969 	}
1970 
1971 	nd->nd_flag &= ~ND_KERBAUTH;
1972 	/*
1973 	 * Handle auth_unix or auth_kerb.
1974 	 */
1975 	if (auth_type == rpc_auth_unix) {
1976 		uid_t uid;
1977 		gid_t gid;
1978 
1979 		nd->nd_cr = kauth_cred_alloc();
1980 		len = fxdr_unsigned(int, *++tl);
1981 		if (len < 0 || len > NFS_MAXNAMLEN) {
1982 			m_freem(mrep);
1983 			error = EBADRPC;
1984 			goto errout;
1985 		}
1986 		nfsm_adv(nfsm_rndup(len));
1987 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1988 
1989 		uid = fxdr_unsigned(uid_t, *tl++);
1990 		gid = fxdr_unsigned(gid_t, *tl++);
1991 		kauth_cred_setuid(nd->nd_cr, uid);
1992 		kauth_cred_seteuid(nd->nd_cr, uid);
1993 		kauth_cred_setsvuid(nd->nd_cr, uid);
1994 		kauth_cred_setgid(nd->nd_cr, gid);
1995 		kauth_cred_setegid(nd->nd_cr, gid);
1996 		kauth_cred_setsvgid(nd->nd_cr, gid);
1997 
1998 		len = fxdr_unsigned(int, *tl);
1999 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
2000 			m_freem(mrep);
2001 			error = EBADRPC;
2002 			goto errout;
2003 		}
2004 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
2005 
2006 		if (len > 0) {
2007 			size_t grbuf_size = min(len, NGROUPS) * sizeof(gid_t);
2008 			gid_t *grbuf = kmem_alloc(grbuf_size, KM_SLEEP);
2009 
2010 			for (i = 0; i < len; i++) {
2011 				if (i < NGROUPS) /* XXX elad */
2012 					grbuf[i] = fxdr_unsigned(gid_t, *tl++);
2013 				else
2014 					tl++;
2015 			}
2016 			kauth_cred_setgroups(nd->nd_cr, grbuf,
2017 			    min(len, NGROUPS), -1, UIO_SYSSPACE);
2018 			kmem_free(grbuf, grbuf_size);
2019 		}
2020 
2021 		len = fxdr_unsigned(int, *++tl);
2022 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
2023 			m_freem(mrep);
2024 			error = EBADRPC;
2025 			goto errout;
2026 		}
2027 		if (len > 0)
2028 			nfsm_adv(nfsm_rndup(len));
2029 	} else if (auth_type == rpc_auth_kerb) {
2030 		switch (fxdr_unsigned(int, *tl++)) {
2031 		case RPCAKN_FULLNAME:
2032 			ticklen = fxdr_unsigned(int, *tl);
2033 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
2034 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
2035 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
2036 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
2037 				m_freem(mrep);
2038 				error = EBADRPC;
2039 				goto errout;
2040 			}
2041 			uio.uio_offset = 0;
2042 			uio.uio_iov = &iov;
2043 			uio.uio_iovcnt = 1;
2044 			UIO_SETUP_SYSSPACE(&uio);
2045 			iov.iov_base = (void *)&nfsd->nfsd_authstr[4];
2046 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
2047 			nfsm_mtouio(&uio, uio.uio_resid);
2048 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2049 			if (*tl++ != rpc_auth_kerb ||
2050 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
2051 				printf("Bad kerb verifier\n");
2052 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2053 				nd->nd_procnum = NFSPROC_NOOP;
2054 				return (0);
2055 			}
2056 			nfsm_dissect(cp, void *, 4 * NFSX_UNSIGNED);
2057 			tl = (u_int32_t *)cp;
2058 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
2059 				printf("Not fullname kerb verifier\n");
2060 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2061 				nd->nd_procnum = NFSPROC_NOOP;
2062 				return (0);
2063 			}
2064 			cp += NFSX_UNSIGNED;
2065 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
2066 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
2067 			nd->nd_flag |= ND_KERBFULL;
2068 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
2069 			break;
2070 		case RPCAKN_NICKNAME:
2071 			if (len != 2 * NFSX_UNSIGNED) {
2072 				printf("Kerb nickname short\n");
2073 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
2074 				nd->nd_procnum = NFSPROC_NOOP;
2075 				return (0);
2076 			}
2077 			nickuid = fxdr_unsigned(uid_t, *tl);
2078 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2079 			if (*tl++ != rpc_auth_kerb ||
2080 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
2081 				printf("Kerb nick verifier bad\n");
2082 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2083 				nd->nd_procnum = NFSPROC_NOOP;
2084 				return (0);
2085 			}
2086 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2087 			tvin.tv_sec = *tl++;
2088 			tvin.tv_usec = *tl;
2089 
2090 			LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid),
2091 			    nu_hash) {
2092 				if (kauth_cred_geteuid(nuidp->nu_cr) == nickuid &&
2093 				    (!nd->nd_nam2 ||
2094 				     netaddr_match(NU_NETFAM(nuidp),
2095 				      &nuidp->nu_haddr, nd->nd_nam2)))
2096 					break;
2097 			}
2098 			if (!nuidp) {
2099 				nd->nd_repstat =
2100 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
2101 				nd->nd_procnum = NFSPROC_NOOP;
2102 				return (0);
2103 			}
2104 
2105 			/*
2106 			 * Now, decrypt the timestamp using the session key
2107 			 * and validate it.
2108 			 */
2109 #ifdef NFSKERB
2110 			XXX
2111 #endif
2112 
2113 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
2114 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
2115 			if (nuidp->nu_expire < time_second ||
2116 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
2117 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
2118 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
2119 				nuidp->nu_expire = 0;
2120 				nd->nd_repstat =
2121 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
2122 				nd->nd_procnum = NFSPROC_NOOP;
2123 				return (0);
2124 			}
2125 			kauth_cred_hold(nuidp->nu_cr);
2126 			nd->nd_cr = nuidp->nu_cr;
2127 			nd->nd_flag |= ND_KERBNICK;
2128 		}
2129 	} else {
2130 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
2131 		nd->nd_procnum = NFSPROC_NOOP;
2132 		return (0);
2133 	}
2134 
2135 	nd->nd_md = md;
2136 	nd->nd_dpos = dpos;
2137 	KASSERT((nd->nd_cr == NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) != 0)
2138 	     || (nd->nd_cr != NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) == 0));
2139 	return (0);
2140 nfsmout:
2141 errout:
2142 	KASSERT(error != 0);
2143 	if (nd->nd_cr != NULL) {
2144 		kauth_cred_free(nd->nd_cr);
2145 		nd->nd_cr = NULL;
2146 	}
2147 	return (error);
2148 }
2149 
2150 int
2151 nfs_msg(struct lwp *l, const char *server, const char *msg)
2152 {
2153 	tpr_t tpr;
2154 
2155 	if (l)
2156 		tpr = tprintf_open(l->l_proc);
2157 	else
2158 		tpr = NULL;
2159 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
2160 	tprintf_close(tpr);
2161 	return (0);
2162 }
2163 
2164 static struct pool nfs_srvdesc_pool;
2165 
2166 void
2167 nfsdreq_init(void)
2168 {
2169 
2170 	pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript),
2171 	    0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr, IPL_NONE);
2172 }
2173 
2174 void
2175 nfsdreq_fini(void)
2176 {
2177 
2178 	pool_destroy(&nfs_srvdesc_pool);
2179 }
2180 
2181 struct nfsrv_descript *
2182 nfsdreq_alloc(void)
2183 {
2184 	struct nfsrv_descript *nd;
2185 
2186 	nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK);
2187 	nd->nd_cr = NULL;
2188 	return nd;
2189 }
2190 
2191 void
2192 nfsdreq_free(struct nfsrv_descript *nd)
2193 {
2194 	kauth_cred_t cr;
2195 
2196 	cr = nd->nd_cr;
2197 	if (cr != NULL) {
2198 		kauth_cred_free(cr);
2199 	}
2200 	pool_put(&nfs_srvdesc_pool, nd);
2201 }
2202