xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 56a34939419542e88b386b2229be7565f4f45461)
1 /*	$NetBSD: nfs_socket.c,v 1.175 2008/11/23 08:22:07 mrg 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 1.175 2008/11/23 08:22:07 mrg 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(nmp, rep, l)
200 	struct nfsmount *nmp;
201 	struct nfsreq *rep;
202 	struct lwp *l;
203 {
204 	struct socket *so;
205 	int error, rcvreserve, sndreserve;
206 	struct sockaddr *saddr;
207 	struct sockaddr_in *sin;
208 	struct sockaddr_in6 *sin6;
209 	struct mbuf *m;
210 	int val;
211 
212 	nmp->nm_so = (struct socket *)0;
213 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
214 	error = socreate(saddr->sa_family, &nmp->nm_so,
215 		nmp->nm_sotype, nmp->nm_soproto, l, NULL);
216 	if (error)
217 		goto bad;
218 	so = nmp->nm_so;
219 #ifdef MBUFTRACE
220 	so->so_mowner = &nfs_mowner;
221 	so->so_rcv.sb_mowner = &nfs_mowner;
222 	so->so_snd.sb_mowner = &nfs_mowner;
223 #endif
224 	nmp->nm_soflags = so->so_proto->pr_flags;
225 
226 	/*
227 	 * Some servers require that the client port be a reserved port number.
228 	 */
229 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
230 		val = IP_PORTRANGE_LOW;
231 
232 		if ((error = so_setsockopt(NULL, so, IPPROTO_IP, IP_PORTRANGE,
233 		    &val, sizeof(val))))
234 			goto bad;
235 		m = m_get(M_WAIT, MT_SONAME);
236 		MCLAIM(m, so->so_mowner);
237 		sin = mtod(m, struct sockaddr_in *);
238 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
239 		sin->sin_family = AF_INET;
240 		sin->sin_addr.s_addr = INADDR_ANY;
241 		sin->sin_port = 0;
242 		error = sobind(so, m, &lwp0);
243 		m_freem(m);
244 		if (error)
245 			goto bad;
246 	}
247 	if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) {
248 		val = IPV6_PORTRANGE_LOW;
249 
250 		if ((error = so_setsockopt(NULL, so, IPPROTO_IPV6,
251 		    IPV6_PORTRANGE, &val, sizeof(val))))
252 			goto bad;
253 		m = m_get(M_WAIT, MT_SONAME);
254 		MCLAIM(m, so->so_mowner);
255 		sin6 = mtod(m, struct sockaddr_in6 *);
256 		memset(sin6, 0, sizeof(*sin6));
257 		sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6);
258 		sin6->sin6_family = AF_INET6;
259 		error = sobind(so, m, &lwp0);
260 		m_freem(m);
261 		if (error)
262 			goto bad;
263 	}
264 
265 	/*
266 	 * Protocols that do not require connections may be optionally left
267 	 * unconnected for servers that reply from a port other than NFS_PORT.
268 	 */
269 	solock(so);
270 	if (nmp->nm_flag & NFSMNT_NOCONN) {
271 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
272 			sounlock(so);
273 			error = ENOTCONN;
274 			goto bad;
275 		}
276 	} else {
277 		error = soconnect(so, nmp->nm_nam, l);
278 		if (error) {
279 			sounlock(so);
280 			goto bad;
281 		}
282 
283 		/*
284 		 * Wait for the connection to complete. Cribbed from the
285 		 * connect system call but with the wait timing out so
286 		 * that interruptible mounts don't hang here for a long time.
287 		 */
288 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
289 			(void)sowait(so, 2 * hz);
290 			if ((so->so_state & SS_ISCONNECTING) &&
291 			    so->so_error == 0 && rep &&
292 			    (error = nfs_sigintr(nmp, rep, rep->r_lwp)) != 0){
293 				so->so_state &= ~SS_ISCONNECTING;
294 				sounlock(so);
295 				goto bad;
296 			}
297 		}
298 		if (so->so_error) {
299 			error = so->so_error;
300 			so->so_error = 0;
301 			sounlock(so);
302 			goto bad;
303 		}
304 	}
305 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
306 		so->so_rcv.sb_timeo = (5 * hz);
307 		so->so_snd.sb_timeo = (5 * hz);
308 	} else {
309 		/*
310 		 * enable receive timeout to detect server crash and reconnect.
311 		 * otherwise, we can be stuck in soreceive forever.
312 		 */
313 		so->so_rcv.sb_timeo = (5 * hz);
314 		so->so_snd.sb_timeo = 0;
315 	}
316 	if (nmp->nm_sotype == SOCK_DGRAM) {
317 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
318 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
319 		    NFS_MAXPKTHDR) * 2;
320 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
321 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
322 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
323 		    NFS_MAXPKTHDR) * 2;
324 	} else {
325 		sounlock(so);
326 		if (nmp->nm_sotype != SOCK_STREAM)
327 			panic("nfscon sotype");
328 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
329 			val = 1;
330 			so_setsockopt(NULL, so, SOL_SOCKET, SO_KEEPALIVE, &val,
331 			    sizeof(val));
332 		}
333 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
334 			val = 1;
335 			so_setsockopt(NULL, so, IPPROTO_TCP, TCP_NODELAY, &val,
336 			    sizeof(val));
337 		}
338 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
339 		    sizeof (u_int32_t)) * 2;
340 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
341 		    sizeof (u_int32_t)) * 2;
342 		solock(so);
343 	}
344 	error = soreserve(so, sndreserve, rcvreserve);
345 	if (error) {
346 		sounlock(so);
347 		goto bad;
348 	}
349 	so->so_rcv.sb_flags |= SB_NOINTR;
350 	so->so_snd.sb_flags |= SB_NOINTR;
351 	sounlock(so);
352 
353 	/* Initialize other non-zero congestion variables */
354 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
355 		NFS_TIMEO << 3;
356 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
357 		nmp->nm_sdrtt[3] = 0;
358 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
359 	nmp->nm_sent = 0;
360 	nmp->nm_timeouts = 0;
361 	return (0);
362 
363 bad:
364 	nfs_disconnect(nmp);
365 	return (error);
366 }
367 
368 /*
369  * Reconnect routine:
370  * Called when a connection is broken on a reliable protocol.
371  * - clean up the old socket
372  * - nfs_connect() again
373  * - set R_MUSTRESEND for all outstanding requests on mount point
374  * If this fails the mount point is DEAD!
375  * nb: Must be called with the nfs_sndlock() set on the mount point.
376  */
377 int
378 nfs_reconnect(struct nfsreq *rep)
379 {
380 	struct nfsreq *rp;
381 	struct nfsmount *nmp = rep->r_nmp;
382 	int error;
383 
384 	nfs_disconnect(nmp);
385 	while ((error = nfs_connect(nmp, rep, &lwp0)) != 0) {
386 		if (error == EINTR || error == ERESTART)
387 			return (EINTR);
388 		kpause("nfscn2", false, hz, NULL);
389 	}
390 
391 	/*
392 	 * Loop through outstanding request list and fix up all requests
393 	 * on old socket.
394 	 */
395 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
396 		if (rp->r_nmp == nmp) {
397 			if ((rp->r_flags & R_MUSTRESEND) == 0)
398 				rp->r_flags |= R_MUSTRESEND | R_REXMITTED;
399 			rp->r_rexmit = 0;
400 		}
401 	}
402 	return (0);
403 }
404 
405 /*
406  * NFS disconnect. Clean up and unlink.
407  */
408 void
409 nfs_disconnect(nmp)
410 	struct nfsmount *nmp;
411 {
412 	struct socket *so;
413 	int drain = 0;
414 
415 	if (nmp->nm_so) {
416 		so = nmp->nm_so;
417 		nmp->nm_so = (struct socket *)0;
418 		solock(so);
419 		soshutdown(so, SHUT_RDWR);
420 		sounlock(so);
421 		drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
422 		if (drain) {
423 			/*
424 			 * soshutdown() above should wake up the current
425 			 * listener.
426 			 * Now wake up those waiting for the receive lock, and
427 			 * wait for them to go away unhappy, to prevent *nmp
428 			 * from evaporating while they're sleeping.
429 			 */
430 			mutex_enter(&nmp->nm_lock);
431 			while (nmp->nm_waiters > 0) {
432 				cv_broadcast(&nmp->nm_rcvcv);
433 				cv_broadcast(&nmp->nm_sndcv);
434 				cv_wait(&nmp->nm_disconcv, &nmp->nm_lock);
435 			}
436 			mutex_exit(&nmp->nm_lock);
437 		}
438 		soclose(so);
439 	}
440 #ifdef DIAGNOSTIC
441 	if (drain && (nmp->nm_waiters > 0))
442 		panic("nfs_disconnect: waiters left after drain?");
443 #endif
444 }
445 
446 void
447 nfs_safedisconnect(nmp)
448 	struct nfsmount *nmp;
449 {
450 	struct nfsreq dummyreq;
451 
452 	memset(&dummyreq, 0, sizeof(dummyreq));
453 	dummyreq.r_nmp = nmp;
454 	nfs_rcvlock(nmp, &dummyreq); /* XXX ignored error return */
455 	nfs_disconnect(nmp);
456 	nfs_rcvunlock(nmp);
457 }
458 
459 /*
460  * This is the nfs send routine. For connection based socket types, it
461  * must be called with an nfs_sndlock() on the socket.
462  * "rep == NULL" indicates that it has been called from a server.
463  * For the client side:
464  * - return EINTR if the RPC is terminated, 0 otherwise
465  * - set R_MUSTRESEND if the send fails for any reason
466  * - do any cleanup required by recoverable socket errors (? ? ?)
467  * For the server side:
468  * - return EINTR or ERESTART if interrupted by a signal
469  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
470  * - do any cleanup required by recoverable socket errors (? ? ?)
471  */
472 int
473 nfs_send(so, nam, top, rep, l)
474 	struct socket *so;
475 	struct mbuf *nam;
476 	struct mbuf *top;
477 	struct nfsreq *rep;
478 	struct lwp *l;
479 {
480 	struct mbuf *sendnam;
481 	int error, soflags, flags;
482 
483 	/* XXX nfs_doio()/nfs_request() calls with  rep->r_lwp == NULL */
484 	if (l == NULL && rep->r_lwp == NULL)
485 		l = curlwp;
486 
487 	if (rep) {
488 		if (rep->r_flags & R_SOFTTERM) {
489 			m_freem(top);
490 			return (EINTR);
491 		}
492 		if ((so = rep->r_nmp->nm_so) == NULL) {
493 			rep->r_flags |= R_MUSTRESEND;
494 			m_freem(top);
495 			return (0);
496 		}
497 		rep->r_flags &= ~R_MUSTRESEND;
498 		soflags = rep->r_nmp->nm_soflags;
499 	} else
500 		soflags = so->so_proto->pr_flags;
501 	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
502 		sendnam = (struct mbuf *)0;
503 	else
504 		sendnam = nam;
505 	if (so->so_type == SOCK_SEQPACKET)
506 		flags = MSG_EOR;
507 	else
508 		flags = 0;
509 
510 	error = (*so->so_send)(so, sendnam, NULL, top, NULL, flags,  l);
511 	if (error) {
512 		if (rep) {
513 			if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
514 				/*
515 				 * We're too fast for the network/driver,
516 				 * and UDP isn't flowcontrolled.
517 				 * We need to resend. This is not fatal,
518 				 * just try again.
519 				 *
520 				 * Could be smarter here by doing some sort
521 				 * of a backoff, but this is rare.
522 				 */
523 				rep->r_flags |= R_MUSTRESEND;
524 			} else {
525 				if (error != EPIPE)
526 					log(LOG_INFO,
527 					    "nfs send error %d for %s\n",
528 					    error,
529 					    rep->r_nmp->nm_mountp->
530 						    mnt_stat.f_mntfromname);
531 				/*
532 				 * Deal with errors for the client side.
533 				 */
534 				if (rep->r_flags & R_SOFTTERM)
535 					error = EINTR;
536 				else
537 					rep->r_flags |= R_MUSTRESEND;
538 			}
539 		} else {
540 			/*
541 			 * See above. This error can happen under normal
542 			 * circumstances and the log is too noisy.
543 			 * The error will still show up in nfsstat.
544 			 */
545 			if (error != ENOBUFS || so->so_type != SOCK_DGRAM)
546 				log(LOG_INFO, "nfsd send error %d\n", error);
547 		}
548 
549 		/*
550 		 * Handle any recoverable (soft) socket errors here. (? ? ?)
551 		 */
552 		if (error != EINTR && error != ERESTART &&
553 			error != EWOULDBLOCK && error != EPIPE)
554 			error = 0;
555 	}
556 	return (error);
557 }
558 
559 #ifdef NFS
560 /*
561  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
562  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
563  * Mark and consolidate the data into a new mbuf list.
564  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
565  *     small mbufs.
566  * For SOCK_STREAM we must be very careful to read an entire record once
567  * we have read any of it, even if the system call has been interrupted.
568  */
569 static int
570 nfs_receive(struct nfsreq *rep, struct mbuf **aname, struct mbuf **mp,
571     struct lwp *l)
572 {
573 	struct socket *so;
574 	struct uio auio;
575 	struct iovec aio;
576 	struct mbuf *m;
577 	struct mbuf *control;
578 	u_int32_t len;
579 	struct mbuf **getnam;
580 	int error, sotype, rcvflg;
581 
582 	/*
583 	 * Set up arguments for soreceive()
584 	 */
585 	*mp = (struct mbuf *)0;
586 	*aname = (struct mbuf *)0;
587 	sotype = rep->r_nmp->nm_sotype;
588 
589 	/*
590 	 * For reliable protocols, lock against other senders/receivers
591 	 * in case a reconnect is necessary.
592 	 * For SOCK_STREAM, first get the Record Mark to find out how much
593 	 * more there is to get.
594 	 * We must lock the socket against other receivers
595 	 * until we have an entire rpc request/reply.
596 	 */
597 	if (sotype != SOCK_DGRAM) {
598 		error = nfs_sndlock(rep->r_nmp, rep);
599 		if (error)
600 			return (error);
601 tryagain:
602 		/*
603 		 * Check for fatal errors and resending request.
604 		 */
605 		/*
606 		 * Ugh: If a reconnect attempt just happened, nm_so
607 		 * would have changed. NULL indicates a failed
608 		 * attempt that has essentially shut down this
609 		 * mount point.
610 		 */
611 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
612 			nfs_sndunlock(rep->r_nmp);
613 			return (EINTR);
614 		}
615 		so = rep->r_nmp->nm_so;
616 		if (!so) {
617 			error = nfs_reconnect(rep);
618 			if (error) {
619 				nfs_sndunlock(rep->r_nmp);
620 				return (error);
621 			}
622 			goto tryagain;
623 		}
624 		while (rep->r_flags & R_MUSTRESEND) {
625 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
626 			nfsstats.rpcretries++;
627 			rep->r_rtt = 0;
628 			rep->r_flags &= ~R_TIMING;
629 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, l);
630 			if (error) {
631 				if (error == EINTR || error == ERESTART ||
632 				    (error = nfs_reconnect(rep)) != 0) {
633 					nfs_sndunlock(rep->r_nmp);
634 					return (error);
635 				}
636 				goto tryagain;
637 			}
638 		}
639 		nfs_sndunlock(rep->r_nmp);
640 		if (sotype == SOCK_STREAM) {
641 			aio.iov_base = (void *) &len;
642 			aio.iov_len = sizeof(u_int32_t);
643 			auio.uio_iov = &aio;
644 			auio.uio_iovcnt = 1;
645 			auio.uio_rw = UIO_READ;
646 			auio.uio_offset = 0;
647 			auio.uio_resid = sizeof(u_int32_t);
648 			UIO_SETUP_SYSSPACE(&auio);
649 			do {
650 			   rcvflg = MSG_WAITALL;
651 			   error = (*so->so_receive)(so, (struct mbuf **)0, &auio,
652 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
653 			   if (error == EWOULDBLOCK && rep) {
654 				if (rep->r_flags & R_SOFTTERM)
655 					return (EINTR);
656 				/*
657 				 * if it seems that the server died after it
658 				 * received our request, set EPIPE so that
659 				 * we'll reconnect and retransmit requests.
660 				 */
661 				if (rep->r_rexmit >= rep->r_nmp->nm_retry) {
662 					nfsstats.rpctimeouts++;
663 					error = EPIPE;
664 				}
665 			   }
666 			} while (error == EWOULDBLOCK);
667 			if (!error && auio.uio_resid > 0) {
668 			    /*
669 			     * Don't log a 0 byte receive; it means
670 			     * that the socket has been closed, and
671 			     * can happen during normal operation
672 			     * (forcible unmount or Solaris server).
673 			     */
674 			    if (auio.uio_resid != sizeof (u_int32_t))
675 			      log(LOG_INFO,
676 				 "short receive (%lu/%lu) from nfs server %s\n",
677 				 (u_long)sizeof(u_int32_t) - auio.uio_resid,
678 				 (u_long)sizeof(u_int32_t),
679 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
680 			    error = EPIPE;
681 			}
682 			if (error)
683 				goto errout;
684 			len = ntohl(len) & ~0x80000000;
685 			/*
686 			 * This is SERIOUS! We are out of sync with the sender
687 			 * and forcing a disconnect/reconnect is all I can do.
688 			 */
689 			if (len > NFS_MAXPACKET) {
690 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
691 				"impossible packet length",
692 				len,
693 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
694 			    error = EFBIG;
695 			    goto errout;
696 			}
697 			auio.uio_resid = len;
698 			do {
699 			    rcvflg = MSG_WAITALL;
700 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
701 				&auio, mp, (struct mbuf **)0, &rcvflg);
702 			} while (error == EWOULDBLOCK || error == EINTR ||
703 				 error == ERESTART);
704 			if (!error && auio.uio_resid > 0) {
705 			    if (len != auio.uio_resid)
706 			      log(LOG_INFO,
707 				"short receive (%lu/%d) from nfs server %s\n",
708 				(u_long)len - auio.uio_resid, len,
709 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
710 			    error = EPIPE;
711 			}
712 		} else {
713 			/*
714 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
715 			 * and soreceive() will return when it has either a
716 			 * control msg or a data msg.
717 			 * We have no use for control msg., but must grab them
718 			 * and then throw them away so we know what is going
719 			 * on.
720 			 */
721 			auio.uio_resid = len = 100000000; /* Anything Big */
722 			/* not need to setup uio_vmspace */
723 			do {
724 			    rcvflg = 0;
725 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
726 				&auio, mp, &control, &rcvflg);
727 			    if (control)
728 				m_freem(control);
729 			    if (error == EWOULDBLOCK && rep) {
730 				if (rep->r_flags & R_SOFTTERM)
731 					return (EINTR);
732 			    }
733 			} while (error == EWOULDBLOCK ||
734 				 (!error && *mp == NULL && control));
735 			if ((rcvflg & MSG_EOR) == 0)
736 				printf("Egad!!\n");
737 			if (!error && *mp == NULL)
738 				error = EPIPE;
739 			len -= auio.uio_resid;
740 		}
741 errout:
742 		if (error && error != EINTR && error != ERESTART) {
743 			m_freem(*mp);
744 			*mp = (struct mbuf *)0;
745 			if (error != EPIPE)
746 				log(LOG_INFO,
747 				    "receive error %d from nfs server %s\n",
748 				    error,
749 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
750 			error = nfs_sndlock(rep->r_nmp, rep);
751 			if (!error)
752 				error = nfs_reconnect(rep);
753 			if (!error)
754 				goto tryagain;
755 			else
756 				nfs_sndunlock(rep->r_nmp);
757 		}
758 	} else {
759 		if ((so = rep->r_nmp->nm_so) == NULL)
760 			return (EACCES);
761 		if (so->so_state & SS_ISCONNECTED)
762 			getnam = (struct mbuf **)0;
763 		else
764 			getnam = aname;
765 		auio.uio_resid = len = 1000000;
766 		/* not need to setup uio_vmspace */
767 		do {
768 			rcvflg = 0;
769 			error =  (*so->so_receive)(so, getnam, &auio, mp,
770 				(struct mbuf **)0, &rcvflg);
771 			if (error == EWOULDBLOCK &&
772 			    (rep->r_flags & R_SOFTTERM))
773 				return (EINTR);
774 		} while (error == EWOULDBLOCK);
775 		len -= auio.uio_resid;
776 		if (!error && *mp == NULL)
777 			error = EPIPE;
778 	}
779 	if (error) {
780 		m_freem(*mp);
781 		*mp = (struct mbuf *)0;
782 	}
783 	return (error);
784 }
785 
786 /*
787  * Implement receipt of reply on a socket.
788  * We must search through the list of received datagrams matching them
789  * with outstanding requests using the xid, until ours is found.
790  */
791 /* ARGSUSED */
792 static int
793 nfs_reply(struct nfsreq *myrep, struct lwp *lwp)
794 {
795 	struct nfsreq *rep;
796 	struct nfsmount *nmp = myrep->r_nmp;
797 	int32_t t1;
798 	struct mbuf *mrep, *nam, *md;
799 	u_int32_t rxid, *tl;
800 	char *dpos, *cp2;
801 	int error;
802 
803 	/*
804 	 * Loop around until we get our own reply
805 	 */
806 	for (;;) {
807 		/*
808 		 * Lock against other receivers so that I don't get stuck in
809 		 * sbwait() after someone else has received my reply for me.
810 		 * Also necessary for connection based protocols to avoid
811 		 * race conditions during a reconnect.
812 		 */
813 		error = nfs_rcvlock(nmp, myrep);
814 		if (error == EALREADY)
815 			return (0);
816 		if (error)
817 			return (error);
818 		/*
819 		 * Get the next Rpc reply off the socket
820 		 */
821 
822 		mutex_enter(&nmp->nm_lock);
823 		nmp->nm_waiters++;
824 		mutex_exit(&nmp->nm_lock);
825 
826 		error = nfs_receive(myrep, &nam, &mrep, lwp);
827 
828 		mutex_enter(&nmp->nm_lock);
829 		nmp->nm_waiters--;
830 		cv_signal(&nmp->nm_disconcv);
831 		mutex_exit(&nmp->nm_lock);
832 
833 		if (error) {
834 			nfs_rcvunlock(nmp);
835 
836 			if (nmp->nm_iflag & NFSMNT_DISMNT) {
837 				/*
838 				 * Oops, we're going away now..
839 				 */
840 				return error;
841 			}
842 			/*
843 			 * Ignore routing errors on connectionless protocols? ?
844 			 */
845 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
846 				nmp->nm_so->so_error = 0;
847 #ifdef DEBUG
848 				if (ratecheck(&nfs_reply_last_err_time,
849 				    &nfs_err_interval))
850 					printf("%s: ignoring error %d\n",
851 					       __func__, error);
852 #endif
853 				continue;
854 			}
855 			return (error);
856 		}
857 		if (nam)
858 			m_freem(nam);
859 
860 		/*
861 		 * Get the xid and check that it is an rpc reply
862 		 */
863 		md = mrep;
864 		dpos = mtod(md, void *);
865 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
866 		rxid = *tl++;
867 		if (*tl != rpc_reply) {
868 			nfsstats.rpcinvalid++;
869 			m_freem(mrep);
870 nfsmout:
871 			nfs_rcvunlock(nmp);
872 			continue;
873 		}
874 
875 		/*
876 		 * Loop through the request list to match up the reply
877 		 * Iff no match, just drop the datagram
878 		 */
879 		TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
880 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
881 				/* Found it.. */
882 				rep->r_mrep = mrep;
883 				rep->r_md = md;
884 				rep->r_dpos = dpos;
885 				if (nfsrtton) {
886 					struct rttl *rt;
887 
888 					rt = &nfsrtt.rttl[nfsrtt.pos];
889 					rt->proc = rep->r_procnum;
890 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
891 					rt->sent = nmp->nm_sent;
892 					rt->cwnd = nmp->nm_cwnd;
893 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
894 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
895 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx;
896 					getmicrotime(&rt->tstamp);
897 					if (rep->r_flags & R_TIMING)
898 						rt->rtt = rep->r_rtt;
899 					else
900 						rt->rtt = 1000000;
901 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
902 				}
903 				/*
904 				 * Update congestion window.
905 				 * Do the additive increase of
906 				 * one rpc/rtt.
907 				 */
908 				if (nmp->nm_cwnd <= nmp->nm_sent) {
909 					nmp->nm_cwnd +=
910 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
911 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
912 					if (nmp->nm_cwnd > NFS_MAXCWND)
913 						nmp->nm_cwnd = NFS_MAXCWND;
914 				}
915 				rep->r_flags &= ~R_SENT;
916 				nmp->nm_sent -= NFS_CWNDSCALE;
917 				/*
918 				 * Update rtt using a gain of 0.125 on the mean
919 				 * and a gain of 0.25 on the deviation.
920 				 */
921 				if (rep->r_flags & R_TIMING) {
922 					/*
923 					 * Since the timer resolution of
924 					 * NFS_HZ is so course, it can often
925 					 * result in r_rtt == 0. Since
926 					 * r_rtt == N means that the actual
927 					 * rtt is between N+dt and N+2-dt ticks,
928 					 * add 1.
929 					 */
930 					t1 = rep->r_rtt + 1;
931 					t1 -= (NFS_SRTT(rep) >> 3);
932 					NFS_SRTT(rep) += t1;
933 					if (t1 < 0)
934 						t1 = -t1;
935 					t1 -= (NFS_SDRTT(rep) >> 2);
936 					NFS_SDRTT(rep) += t1;
937 				}
938 				nmp->nm_timeouts = 0;
939 				break;
940 			}
941 		}
942 		nfs_rcvunlock(nmp);
943 		/*
944 		 * If not matched to a request, drop it.
945 		 * If it's mine, get out.
946 		 */
947 		if (rep == 0) {
948 			nfsstats.rpcunexpected++;
949 			m_freem(mrep);
950 		} else if (rep == myrep) {
951 			if (rep->r_mrep == NULL)
952 				panic("nfsreply nil");
953 			return (0);
954 		}
955 	}
956 }
957 
958 /*
959  * nfs_request - goes something like this
960  *	- fill in request struct
961  *	- links it into list
962  *	- calls nfs_send() for first transmit
963  *	- calls nfs_receive() to get reply
964  *	- break down rpc header and return with nfs reply pointed to
965  *	  by mrep or error
966  * nb: always frees up mreq mbuf list
967  */
968 int
969 nfs_request(np, mrest, procnum, lwp, cred, mrp, mdp, dposp, rexmitp)
970 	struct nfsnode *np;
971 	struct mbuf *mrest;
972 	int procnum;
973 	struct lwp *lwp;
974 	kauth_cred_t cred;
975 	struct mbuf **mrp;
976 	struct mbuf **mdp;
977 	char **dposp;
978 	int *rexmitp;
979 {
980 	struct mbuf *m, *mrep;
981 	struct nfsreq *rep;
982 	u_int32_t *tl;
983 	int i;
984 	struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount);
985 	struct mbuf *md, *mheadend;
986 	char nickv[RPCX_NICKVERF];
987 	time_t waituntil;
988 	char *dpos, *cp2;
989 	int t1, s, error = 0, mrest_len, auth_len, auth_type;
990 	int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0;
991 	int verf_len, verf_type;
992 	u_int32_t xid;
993 	char *auth_str, *verf_str;
994 	NFSKERBKEY_T key;		/* save session key */
995 	kauth_cred_t acred;
996 	struct mbuf *mrest_backup = NULL;
997 	kauth_cred_t origcred = NULL; /* XXX: gcc */
998 	bool retry_cred = true;
999 	bool use_opencred = (np->n_flag & NUSEOPENCRED) != 0;
1000 
1001 	if (rexmitp != NULL)
1002 		*rexmitp = 0;
1003 
1004 	acred = kauth_cred_alloc();
1005 
1006 tryagain_cred:
1007 	KASSERT(cred != NULL);
1008 	rep = kmem_alloc(sizeof(*rep), KM_SLEEP);
1009 	rep->r_nmp = nmp;
1010 	KASSERT(lwp == NULL || lwp == curlwp);
1011 	rep->r_lwp = lwp;
1012 	rep->r_procnum = procnum;
1013 	i = 0;
1014 	m = mrest;
1015 	while (m) {
1016 		i += m->m_len;
1017 		m = m->m_next;
1018 	}
1019 	mrest_len = i;
1020 
1021 	/*
1022 	 * Get the RPC header with authorization.
1023 	 */
1024 kerbauth:
1025 	verf_str = auth_str = (char *)0;
1026 	if (nmp->nm_flag & NFSMNT_KERB) {
1027 		verf_str = nickv;
1028 		verf_len = sizeof (nickv);
1029 		auth_type = RPCAUTH_KERB4;
1030 		memset((void *)key, 0, sizeof (key));
1031 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
1032 			&auth_len, verf_str, verf_len)) {
1033 			error = nfs_getauth(nmp, rep, cred, &auth_str,
1034 				&auth_len, verf_str, &verf_len, key);
1035 			if (error) {
1036 				kmem_free(rep, sizeof(*rep));
1037 				m_freem(mrest);
1038 				KASSERT(kauth_cred_getrefcnt(acred) == 1);
1039 				kauth_cred_free(acred);
1040 				return (error);
1041 			}
1042 		}
1043 		retry_cred = false;
1044 	} else {
1045 		/* AUTH_UNIX */
1046 		uid_t uid;
1047 		gid_t gid;
1048 
1049 		/*
1050 		 * on the most unix filesystems, permission checks are
1051 		 * done when the file is open(2)'ed.
1052 		 * ie. once a file is successfully open'ed,
1053 		 * following i/o operations never fail with EACCES.
1054 		 * we try to follow the semantics as far as possible.
1055 		 *
1056 		 * note that we expect that the nfs server always grant
1057 		 * accesses by the file's owner.
1058 		 */
1059 		origcred = cred;
1060 		switch (procnum) {
1061 		case NFSPROC_READ:
1062 		case NFSPROC_WRITE:
1063 		case NFSPROC_COMMIT:
1064 			uid = np->n_vattr->va_uid;
1065 			gid = np->n_vattr->va_gid;
1066 			if (kauth_cred_geteuid(cred) == uid &&
1067 			    kauth_cred_getegid(cred) == gid) {
1068 				retry_cred = false;
1069 				break;
1070 			}
1071 			if (use_opencred)
1072 				break;
1073 			kauth_cred_setuid(acred, uid);
1074 			kauth_cred_seteuid(acred, uid);
1075 			kauth_cred_setsvuid(acred, uid);
1076 			kauth_cred_setgid(acred, gid);
1077 			kauth_cred_setegid(acred, gid);
1078 			kauth_cred_setsvgid(acred, gid);
1079 			cred = acred;
1080 			break;
1081 		default:
1082 			retry_cred = false;
1083 			break;
1084 		}
1085 		/*
1086 		 * backup mbuf chain if we can need it later to retry.
1087 		 *
1088 		 * XXX maybe we can keep a direct reference to
1089 		 * mrest without doing m_copym, but it's ...ugly.
1090 		 */
1091 		if (retry_cred)
1092 			mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT);
1093 		auth_type = RPCAUTH_UNIX;
1094 		/* XXX elad - ngroups */
1095 		auth_len = (((kauth_cred_ngroups(cred) > nmp->nm_numgrps) ?
1096 			nmp->nm_numgrps : kauth_cred_ngroups(cred)) << 2) +
1097 			5 * NFSX_UNSIGNED;
1098 	}
1099 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
1100 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
1101 	if (auth_str)
1102 		free(auth_str, M_TEMP);
1103 
1104 	/*
1105 	 * For stream protocols, insert a Sun RPC Record Mark.
1106 	 */
1107 	if (nmp->nm_sotype == SOCK_STREAM) {
1108 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
1109 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
1110 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
1111 	}
1112 	rep->r_mreq = m;
1113 	rep->r_xid = xid;
1114 tryagain:
1115 	if (nmp->nm_flag & NFSMNT_SOFT)
1116 		rep->r_retry = nmp->nm_retry;
1117 	else
1118 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
1119 	rep->r_rtt = rep->r_rexmit = 0;
1120 	if (proct[procnum] > 0)
1121 		rep->r_flags = R_TIMING;
1122 	else
1123 		rep->r_flags = 0;
1124 	rep->r_mrep = NULL;
1125 
1126 	/*
1127 	 * Do the client side RPC.
1128 	 */
1129 	nfsstats.rpcrequests++;
1130 	/*
1131 	 * Chain request into list of outstanding requests. Be sure
1132 	 * to put it LAST so timer finds oldest requests first.
1133 	 */
1134 	s = splsoftnet();
1135 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1136 	nfs_timer_start();
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) || nmp->nm_sent < nmp->nm_cwnd)) {
1145 		splx(s);
1146 		if (nmp->nm_soflags & PR_CONNREQUIRED)
1147 			error = nfs_sndlock(nmp, rep);
1148 		if (!error) {
1149 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
1150 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp);
1151 			if (nmp->nm_soflags & PR_CONNREQUIRED)
1152 				nfs_sndunlock(nmp);
1153 		}
1154 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1155 			nmp->nm_sent += NFS_CWNDSCALE;
1156 			rep->r_flags |= R_SENT;
1157 		}
1158 	} else {
1159 		splx(s);
1160 		rep->r_rtt = -1;
1161 	}
1162 
1163 	/*
1164 	 * Wait for the reply from our send or the timer's.
1165 	 */
1166 	if (!error || error == EPIPE)
1167 		error = nfs_reply(rep, lwp);
1168 
1169 	/*
1170 	 * RPC done, unlink the request.
1171 	 */
1172 	s = splsoftnet();
1173 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1174 	splx(s);
1175 
1176 	/*
1177 	 * Decrement the outstanding request count.
1178 	 */
1179 	if (rep->r_flags & R_SENT) {
1180 		rep->r_flags &= ~R_SENT;	/* paranoia */
1181 		nmp->nm_sent -= NFS_CWNDSCALE;
1182 	}
1183 
1184 	if (rexmitp != NULL) {
1185 		int rexmit;
1186 
1187 		if (nmp->nm_sotype != SOCK_DGRAM)
1188 			rexmit = (rep->r_flags & R_REXMITTED) != 0;
1189 		else
1190 			rexmit = rep->r_rexmit;
1191 		*rexmitp = rexmit;
1192 	}
1193 
1194 	/*
1195 	 * If there was a successful reply and a tprintf msg.
1196 	 * tprintf a response.
1197 	 */
1198 	if (!error && (rep->r_flags & R_TPRINTFMSG))
1199 		nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1200 		    "is alive again");
1201 	mrep = rep->r_mrep;
1202 	md = rep->r_md;
1203 	dpos = rep->r_dpos;
1204 	if (error)
1205 		goto nfsmout;
1206 
1207 	/*
1208 	 * break down the rpc header and check if ok
1209 	 */
1210 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1211 	if (*tl++ == rpc_msgdenied) {
1212 		if (*tl == rpc_mismatch)
1213 			error = EOPNOTSUPP;
1214 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1215 			if (!failed_auth) {
1216 				failed_auth++;
1217 				mheadend->m_next = (struct mbuf *)0;
1218 				m_freem(mrep);
1219 				m_freem(rep->r_mreq);
1220 				goto kerbauth;
1221 			} else
1222 				error = EAUTH;
1223 		} else
1224 			error = EACCES;
1225 		m_freem(mrep);
1226 		goto nfsmout;
1227 	}
1228 
1229 	/*
1230 	 * Grab any Kerberos verifier, otherwise just throw it away.
1231 	 */
1232 	verf_type = fxdr_unsigned(int, *tl++);
1233 	i = fxdr_unsigned(int32_t, *tl);
1234 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1235 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1236 		if (error)
1237 			goto nfsmout;
1238 	} else if (i > 0)
1239 		nfsm_adv(nfsm_rndup(i));
1240 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1241 	/* 0 == ok */
1242 	if (*tl == 0) {
1243 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1244 		if (*tl != 0) {
1245 			error = fxdr_unsigned(int, *tl);
1246 			switch (error) {
1247 			case NFSERR_PERM:
1248 				error = EPERM;
1249 				break;
1250 
1251 			case NFSERR_NOENT:
1252 				error = ENOENT;
1253 				break;
1254 
1255 			case NFSERR_IO:
1256 				error = EIO;
1257 				break;
1258 
1259 			case NFSERR_NXIO:
1260 				error = ENXIO;
1261 				break;
1262 
1263 			case NFSERR_ACCES:
1264 				error = EACCES;
1265 				if (!retry_cred)
1266 					break;
1267 				m_freem(mrep);
1268 				m_freem(rep->r_mreq);
1269 				kmem_free(rep, sizeof(*rep));
1270 				use_opencred = !use_opencred;
1271 				if (mrest_backup == NULL) {
1272 					/* m_copym failure */
1273 					KASSERT(
1274 					    kauth_cred_getrefcnt(acred) == 1);
1275 					kauth_cred_free(acred);
1276 					return ENOMEM;
1277 				}
1278 				mrest = mrest_backup;
1279 				mrest_backup = NULL;
1280 				cred = origcred;
1281 				error = 0;
1282 				retry_cred = false;
1283 				goto tryagain_cred;
1284 
1285 			case NFSERR_EXIST:
1286 				error = EEXIST;
1287 				break;
1288 
1289 			case NFSERR_XDEV:
1290 				error = EXDEV;
1291 				break;
1292 
1293 			case NFSERR_NODEV:
1294 				error = ENODEV;
1295 				break;
1296 
1297 			case NFSERR_NOTDIR:
1298 				error = ENOTDIR;
1299 				break;
1300 
1301 			case NFSERR_ISDIR:
1302 				error = EISDIR;
1303 				break;
1304 
1305 			case NFSERR_INVAL:
1306 				error = EINVAL;
1307 				break;
1308 
1309 			case NFSERR_FBIG:
1310 				error = EFBIG;
1311 				break;
1312 
1313 			case NFSERR_NOSPC:
1314 				error = ENOSPC;
1315 				break;
1316 
1317 			case NFSERR_ROFS:
1318 				error = EROFS;
1319 				break;
1320 
1321 			case NFSERR_MLINK:
1322 				error = EMLINK;
1323 				break;
1324 
1325 			case NFSERR_TIMEDOUT:
1326 				error = ETIMEDOUT;
1327 				break;
1328 
1329 			case NFSERR_NAMETOL:
1330 				error = ENAMETOOLONG;
1331 				break;
1332 
1333 			case NFSERR_NOTEMPTY:
1334 				error = ENOTEMPTY;
1335 				break;
1336 
1337 			case NFSERR_DQUOT:
1338 				error = EDQUOT;
1339 				break;
1340 
1341 			case NFSERR_STALE:
1342 				/*
1343 				 * If the File Handle was stale, invalidate the
1344 				 * lookup cache, just in case.
1345 				 */
1346 				error = ESTALE;
1347 				cache_purge(NFSTOV(np));
1348 				break;
1349 
1350 			case NFSERR_REMOTE:
1351 				error = EREMOTE;
1352 				break;
1353 
1354 			case NFSERR_WFLUSH:
1355 			case NFSERR_BADHANDLE:
1356 			case NFSERR_NOT_SYNC:
1357 			case NFSERR_BAD_COOKIE:
1358 				error = EINVAL;
1359 				break;
1360 
1361 			case NFSERR_NOTSUPP:
1362 				error = ENOTSUP;
1363 				break;
1364 
1365 			case NFSERR_TOOSMALL:
1366 			case NFSERR_SERVERFAULT:
1367 			case NFSERR_BADTYPE:
1368 				error = EINVAL;
1369 				break;
1370 
1371 			case NFSERR_TRYLATER:
1372 				if ((nmp->nm_flag & NFSMNT_NFSV3) == 0)
1373 					break;
1374 				m_freem(mrep);
1375 				error = 0;
1376 				waituntil = time_second + trylater_delay;
1377 				while (time_second < waituntil) {
1378 					kpause("nfstrylater", false, hz, NULL);
1379 				}
1380 				trylater_delay *= NFS_TRYLATERDELMUL;
1381 				if (trylater_delay > NFS_TRYLATERDELMAX)
1382 					trylater_delay = NFS_TRYLATERDELMAX;
1383 				/*
1384 				 * RFC1813:
1385 				 * The client should wait and then try
1386 				 * the request with a new RPC transaction ID.
1387 				 */
1388 				nfs_renewxid(rep);
1389 				goto tryagain;
1390 
1391 			default:
1392 #ifdef DIAGNOSTIC
1393 				printf("Invalid rpc error code %d\n", error);
1394 #endif
1395 				error = EINVAL;
1396 				break;
1397 			}
1398 
1399 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1400 				*mrp = mrep;
1401 				*mdp = md;
1402 				*dposp = dpos;
1403 				error |= NFSERR_RETERR;
1404 			} else
1405 				m_freem(mrep);
1406 			goto nfsmout;
1407 		}
1408 
1409 		/*
1410 		 * note which credential worked to minimize number of retries.
1411 		 */
1412 		if (use_opencred)
1413 			np->n_flag |= NUSEOPENCRED;
1414 		else
1415 			np->n_flag &= ~NUSEOPENCRED;
1416 
1417 		*mrp = mrep;
1418 		*mdp = md;
1419 		*dposp = dpos;
1420 
1421 		KASSERT(error == 0);
1422 		goto nfsmout;
1423 	}
1424 	m_freem(mrep);
1425 	error = EPROTONOSUPPORT;
1426 nfsmout:
1427 	KASSERT(kauth_cred_getrefcnt(acred) == 1);
1428 	kauth_cred_free(acred);
1429 	m_freem(rep->r_mreq);
1430 	kmem_free(rep, sizeof(*rep));
1431 	m_freem(mrest_backup);
1432 	return (error);
1433 }
1434 #endif /* NFS */
1435 
1436 /*
1437  * Generate the rpc reply header
1438  * siz arg. is used to decide if adding a cluster is worthwhile
1439  */
1440 int
1441 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1442 	int siz;
1443 	struct nfsrv_descript *nd;
1444 	struct nfssvc_sock *slp;
1445 	int err;
1446 	int cache;
1447 	u_quad_t *frev;
1448 	struct mbuf **mrq;
1449 	struct mbuf **mbp;
1450 	char **bposp;
1451 {
1452 	u_int32_t *tl;
1453 	struct mbuf *mreq;
1454 	char *bpos;
1455 	struct mbuf *mb;
1456 
1457 	mreq = m_gethdr(M_WAIT, MT_DATA);
1458 	MCLAIM(mreq, &nfs_mowner);
1459 	mb = mreq;
1460 	/*
1461 	 * If this is a big reply, use a cluster else
1462 	 * try and leave leading space for the lower level headers.
1463 	 */
1464 	siz += RPC_REPLYSIZ;
1465 	if (siz >= max_datalen) {
1466 		m_clget(mreq, M_WAIT);
1467 	} else
1468 		mreq->m_data += max_hdr;
1469 	tl = mtod(mreq, u_int32_t *);
1470 	mreq->m_len = 6 * NFSX_UNSIGNED;
1471 	bpos = ((char *)tl) + mreq->m_len;
1472 	*tl++ = txdr_unsigned(nd->nd_retxid);
1473 	*tl++ = rpc_reply;
1474 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1475 		*tl++ = rpc_msgdenied;
1476 		if (err & NFSERR_AUTHERR) {
1477 			*tl++ = rpc_autherr;
1478 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1479 			mreq->m_len -= NFSX_UNSIGNED;
1480 			bpos -= NFSX_UNSIGNED;
1481 		} else {
1482 			*tl++ = rpc_mismatch;
1483 			*tl++ = txdr_unsigned(RPC_VER2);
1484 			*tl = txdr_unsigned(RPC_VER2);
1485 		}
1486 	} else {
1487 		*tl++ = rpc_msgaccepted;
1488 
1489 		/*
1490 		 * For Kerberos authentication, we must send the nickname
1491 		 * verifier back, otherwise just RPCAUTH_NULL.
1492 		 */
1493 		if (nd->nd_flag & ND_KERBFULL) {
1494 			struct nfsuid *nuidp;
1495 			struct timeval ktvin, ktvout;
1496 
1497 			memset(&ktvout, 0, sizeof ktvout);	/* XXX gcc */
1498 
1499 			LIST_FOREACH(nuidp,
1500 			    NUIDHASH(slp, kauth_cred_geteuid(nd->nd_cr)),
1501 			    nu_hash) {
1502 				if (kauth_cred_geteuid(nuidp->nu_cr) ==
1503 				kauth_cred_geteuid(nd->nd_cr) &&
1504 				    (!nd->nd_nam2 || netaddr_match(
1505 				    NU_NETFAM(nuidp), &nuidp->nu_haddr,
1506 				    nd->nd_nam2)))
1507 					break;
1508 			}
1509 			if (nuidp) {
1510 				ktvin.tv_sec =
1511 				    txdr_unsigned(nuidp->nu_timestamp.tv_sec
1512 					- 1);
1513 				ktvin.tv_usec =
1514 				    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1515 
1516 				/*
1517 				 * Encrypt the timestamp in ecb mode using the
1518 				 * session key.
1519 				 */
1520 #ifdef NFSKERB
1521 				XXX
1522 #endif
1523 
1524 				*tl++ = rpc_auth_kerb;
1525 				*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1526 				*tl = ktvout.tv_sec;
1527 				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1528 				*tl++ = ktvout.tv_usec;
1529 				*tl++ = txdr_unsigned(
1530 				    kauth_cred_geteuid(nuidp->nu_cr));
1531 			} else {
1532 				*tl++ = 0;
1533 				*tl++ = 0;
1534 			}
1535 		} else {
1536 			*tl++ = 0;
1537 			*tl++ = 0;
1538 		}
1539 		switch (err) {
1540 		case EPROGUNAVAIL:
1541 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1542 			break;
1543 		case EPROGMISMATCH:
1544 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1545 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1546 			*tl++ = txdr_unsigned(2);
1547 			*tl = txdr_unsigned(3);
1548 			break;
1549 		case EPROCUNAVAIL:
1550 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1551 			break;
1552 		case EBADRPC:
1553 			*tl = txdr_unsigned(RPC_GARBAGE);
1554 			break;
1555 		default:
1556 			*tl = 0;
1557 			if (err != NFSERR_RETVOID) {
1558 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1559 				if (err)
1560 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1561 				else
1562 				    *tl = 0;
1563 			}
1564 			break;
1565 		};
1566 	}
1567 
1568 	if (mrq != NULL)
1569 		*mrq = mreq;
1570 	*mbp = mb;
1571 	*bposp = bpos;
1572 	if (err != 0 && err != NFSERR_RETVOID)
1573 		nfsstats.srvrpc_errs++;
1574 	return (0);
1575 }
1576 
1577 static void
1578 nfs_timer_schedule(void)
1579 {
1580 
1581 	callout_schedule(&nfs_timer_ch, nfs_ticks);
1582 }
1583 
1584 void
1585 nfs_timer_start(void)
1586 {
1587 
1588 	if (callout_pending(&nfs_timer_ch))
1589 		return;
1590 
1591 	nfs_timer_start_ev.ev_count++;
1592 	nfs_timer_schedule();
1593 }
1594 
1595 void
1596 nfs_timer_init(void)
1597 {
1598 
1599 	mutex_init(&nfs_timer_lock, MUTEX_DEFAULT, IPL_NONE);
1600 	callout_init(&nfs_timer_ch, 0);
1601 	callout_setfunc(&nfs_timer_ch, nfs_timer, NULL);
1602 	evcnt_attach_dynamic(&nfs_timer_ev, EVCNT_TYPE_MISC, NULL,
1603 	    "nfs", "timer");
1604 	evcnt_attach_dynamic(&nfs_timer_start_ev, EVCNT_TYPE_MISC, NULL,
1605 	    "nfs", "timer start");
1606 	evcnt_attach_dynamic(&nfs_timer_stop_ev, EVCNT_TYPE_MISC, NULL,
1607 	    "nfs", "timer stop");
1608 }
1609 
1610 void
1611 nfs_timer_fini(void)
1612 {
1613 
1614 	callout_halt(&nfs_timer_ch, NULL);
1615 	callout_destroy(&nfs_timer_ch);
1616 	mutex_destroy(&nfs_timer_lock);
1617 	evcnt_detach(&nfs_timer_ev);
1618 	evcnt_detach(&nfs_timer_start_ev);
1619 	evcnt_detach(&nfs_timer_stop_ev);
1620 }
1621 
1622 void
1623 nfs_timer_srvinit(bool (*func)(void))
1624 {
1625 
1626 	nfs_timer_srvvec = func;
1627 }
1628 
1629 void
1630 nfs_timer_srvfini(void)
1631 {
1632 
1633 	mutex_enter(&nfs_timer_lock);
1634 	nfs_timer_srvvec = NULL;
1635 	mutex_exit(&nfs_timer_lock);
1636 }
1637 
1638 
1639 /*
1640  * Nfs timer routine
1641  * Scan the nfsreq list and retranmit any requests that have timed out
1642  * To avoid retransmission attempts on STREAM sockets (in the future) make
1643  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1644  */
1645 void
1646 nfs_timer(void *arg)
1647 {
1648 	struct nfsreq *rep;
1649 	struct mbuf *m;
1650 	struct socket *so;
1651 	struct nfsmount *nmp;
1652 	int timeo;
1653 	int s, error;
1654 	bool more = false;
1655 
1656 	nfs_timer_ev.ev_count++;
1657 
1658 	s = splsoftnet();
1659 	TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
1660 		more = true;
1661 		nmp = rep->r_nmp;
1662 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1663 			continue;
1664 		if (nfs_sigintr(nmp, rep, rep->r_lwp)) {
1665 			rep->r_flags |= R_SOFTTERM;
1666 			continue;
1667 		}
1668 		if (rep->r_rtt >= 0) {
1669 			rep->r_rtt++;
1670 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1671 				timeo = nmp->nm_timeo;
1672 			else
1673 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1674 			if (nmp->nm_timeouts > 0)
1675 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1676 			if (rep->r_rtt <= timeo)
1677 				continue;
1678 			if (nmp->nm_timeouts <
1679 			    (sizeof(nfs_backoff) / sizeof(nfs_backoff[0])))
1680 				nmp->nm_timeouts++;
1681 		}
1682 		/*
1683 		 * Check for server not responding
1684 		 */
1685 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1686 		     rep->r_rexmit > nmp->nm_deadthresh) {
1687 			nfs_msg(rep->r_lwp,
1688 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1689 			    "not responding");
1690 			rep->r_flags |= R_TPRINTFMSG;
1691 		}
1692 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1693 			nfsstats.rpctimeouts++;
1694 			rep->r_flags |= R_SOFTTERM;
1695 			continue;
1696 		}
1697 		if (nmp->nm_sotype != SOCK_DGRAM) {
1698 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1699 				rep->r_rexmit = NFS_MAXREXMIT;
1700 			continue;
1701 		}
1702 		if ((so = nmp->nm_so) == NULL)
1703 			continue;
1704 
1705 		/*
1706 		 * If there is enough space and the window allows..
1707 		 *	Resend it
1708 		 * Set r_rtt to -1 in case we fail to send it now.
1709 		 */
1710 		solock(so);
1711 		rep->r_rtt = -1;
1712 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1713 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1714 		    (rep->r_flags & R_SENT) ||
1715 		    nmp->nm_sent < nmp->nm_cwnd) &&
1716 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1717 		        if (so->so_state & SS_ISCONNECTED)
1718 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1719 			    (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0);
1720 			else
1721 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1722 			    nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0);
1723 			if (error) {
1724 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1725 #ifdef DEBUG
1726 					if (ratecheck(&nfs_timer_last_err_time,
1727 					    &nfs_err_interval))
1728 						printf("%s: ignoring error "
1729 						       "%d\n", __func__, error);
1730 #endif
1731 					so->so_error = 0;
1732 				}
1733 			} else {
1734 				/*
1735 				 * Iff first send, start timing
1736 				 * else turn timing off, backoff timer
1737 				 * and divide congestion window by 2.
1738 				 */
1739 				if (rep->r_flags & R_SENT) {
1740 					rep->r_flags &= ~R_TIMING;
1741 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1742 						rep->r_rexmit = NFS_MAXREXMIT;
1743 					nmp->nm_cwnd >>= 1;
1744 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1745 						nmp->nm_cwnd = NFS_CWNDSCALE;
1746 					nfsstats.rpcretries++;
1747 				} else {
1748 					rep->r_flags |= R_SENT;
1749 					nmp->nm_sent += NFS_CWNDSCALE;
1750 				}
1751 				rep->r_rtt = 0;
1752 			}
1753 		}
1754 		sounlock(so);
1755 	}
1756 	splx(s);
1757 
1758 	mutex_enter(&nfs_timer_lock);
1759 	if (nfs_timer_srvvec != NULL) {
1760 		more |= (*nfs_timer_srvvec)();
1761 	}
1762 	mutex_exit(&nfs_timer_lock);
1763 
1764 	if (more) {
1765 		nfs_timer_schedule();
1766 	} else {
1767 		nfs_timer_stop_ev.ev_count++;
1768 	}
1769 }
1770 
1771 /*
1772  * Test for a termination condition pending on the process.
1773  * This is used for NFSMNT_INT mounts.
1774  */
1775 int
1776 nfs_sigintr(nmp, rep, l)
1777 	struct nfsmount *nmp;
1778 	struct nfsreq *rep;
1779 	struct lwp *l;
1780 {
1781 	sigset_t ss;
1782 
1783 	if (rep && (rep->r_flags & R_SOFTTERM))
1784 		return (EINTR);
1785 	if (!(nmp->nm_flag & NFSMNT_INT))
1786 		return (0);
1787 	if (l) {
1788 		sigpending1(l, &ss);
1789 #if 0
1790 		sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss);
1791 #endif
1792 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1793 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1794 		    sigismember(&ss, SIGQUIT))
1795 			return (EINTR);
1796 	}
1797 	return (0);
1798 }
1799 
1800 #ifdef NFS
1801 /*
1802  * Lock a socket against others.
1803  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1804  * and also to avoid race conditions between the processes with nfs requests
1805  * in progress when a reconnect is necessary.
1806  */
1807 static int
1808 nfs_sndlock(struct nfsmount *nmp, struct nfsreq *rep)
1809 {
1810 	struct lwp *l;
1811 	int timeo = 0;
1812 	bool catch = false;
1813 	int error = 0;
1814 
1815 	if (rep) {
1816 		l = rep->r_lwp;
1817 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1818 			catch = true;
1819 	} else
1820 		l = NULL;
1821 	mutex_enter(&nmp->nm_lock);
1822 	while ((nmp->nm_iflag & NFSMNT_SNDLOCK) != 0) {
1823 		if (rep && nfs_sigintr(rep->r_nmp, rep, l)) {
1824 			error = EINTR;
1825 			goto quit;
1826 		}
1827 		if (catch) {
1828 			cv_timedwait_sig(&nmp->nm_sndcv, &nmp->nm_lock, timeo);
1829 		} else {
1830 			cv_timedwait(&nmp->nm_sndcv, &nmp->nm_lock, timeo);
1831 		}
1832 		if (catch) {
1833 			catch = false;
1834 			timeo = 2 * hz;
1835 		}
1836 	}
1837 	nmp->nm_iflag |= NFSMNT_SNDLOCK;
1838 quit:
1839 	mutex_exit(&nmp->nm_lock);
1840 	return error;
1841 }
1842 
1843 /*
1844  * Unlock the stream socket for others.
1845  */
1846 static void
1847 nfs_sndunlock(struct nfsmount *nmp)
1848 {
1849 
1850 	mutex_enter(&nmp->nm_lock);
1851 	if ((nmp->nm_iflag & NFSMNT_SNDLOCK) == 0)
1852 		panic("nfs sndunlock");
1853 	nmp->nm_iflag &= ~NFSMNT_SNDLOCK;
1854 	cv_signal(&nmp->nm_sndcv);
1855 	mutex_exit(&nmp->nm_lock);
1856 }
1857 #endif /* NFS */
1858 
1859 static int
1860 nfs_rcvlock(struct nfsmount *nmp, struct nfsreq *rep)
1861 {
1862 	int *flagp = &nmp->nm_iflag;
1863 	int slptimeo = 0;
1864 	bool catch;
1865 	int error = 0;
1866 
1867 	KASSERT(nmp == rep->r_nmp);
1868 
1869 	catch = (nmp->nm_flag & NFSMNT_INT) != 0;
1870 	mutex_enter(&nmp->nm_lock);
1871 	while (/* CONSTCOND */ true) {
1872 		if (*flagp & NFSMNT_DISMNT) {
1873 			cv_signal(&nmp->nm_disconcv);
1874 			error = EIO;
1875 			break;
1876 		}
1877 		/* If our reply was received while we were sleeping,
1878 		 * then just return without taking the lock to avoid a
1879 		 * situation where a single iod could 'capture' the
1880 		 * receive lock.
1881 		 */
1882 		if (rep->r_mrep != NULL) {
1883 			error = EALREADY;
1884 			break;
1885 		}
1886 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) {
1887 			error = EINTR;
1888 			break;
1889 		}
1890 		if ((*flagp & NFSMNT_RCVLOCK) == 0) {
1891 			*flagp |= NFSMNT_RCVLOCK;
1892 			break;
1893 		}
1894 		if (catch) {
1895 			cv_timedwait_sig(&nmp->nm_rcvcv, &nmp->nm_lock,
1896 			    slptimeo);
1897 		} else {
1898 			cv_timedwait(&nmp->nm_rcvcv, &nmp->nm_lock,
1899 			    slptimeo);
1900 		}
1901 		if (catch) {
1902 			catch = false;
1903 			slptimeo = 2 * hz;
1904 		}
1905 	}
1906 	mutex_exit(&nmp->nm_lock);
1907 	return error;
1908 }
1909 
1910 /*
1911  * Unlock the stream socket for others.
1912  */
1913 static void
1914 nfs_rcvunlock(struct nfsmount *nmp)
1915 {
1916 
1917 	mutex_enter(&nmp->nm_lock);
1918 	if ((nmp->nm_iflag & NFSMNT_RCVLOCK) == 0)
1919 		panic("nfs rcvunlock");
1920 	nmp->nm_iflag &= ~NFSMNT_RCVLOCK;
1921 	cv_broadcast(&nmp->nm_rcvcv);
1922 	mutex_exit(&nmp->nm_lock);
1923 }
1924 
1925 /*
1926  * Parse an RPC request
1927  * - verify it
1928  * - allocate and fill in the cred.
1929  */
1930 int
1931 nfs_getreq(nd, nfsd, has_header)
1932 	struct nfsrv_descript *nd;
1933 	struct nfsd *nfsd;
1934 	int has_header;
1935 {
1936 	int len, i;
1937 	u_int32_t *tl;
1938 	int32_t t1;
1939 	struct uio uio;
1940 	struct iovec iov;
1941 	char *dpos, *cp2, *cp;
1942 	u_int32_t nfsvers, auth_type;
1943 	uid_t nickuid;
1944 	int error = 0, ticklen;
1945 	struct mbuf *mrep, *md;
1946 	struct nfsuid *nuidp;
1947 	struct timeval tvin, tvout;
1948 
1949 	memset(&tvout, 0, sizeof tvout);	/* XXX gcc */
1950 
1951 	KASSERT(nd->nd_cr == NULL);
1952 	mrep = nd->nd_mrep;
1953 	md = nd->nd_md;
1954 	dpos = nd->nd_dpos;
1955 	if (has_header) {
1956 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1957 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1958 		if (*tl++ != rpc_call) {
1959 			m_freem(mrep);
1960 			return (EBADRPC);
1961 		}
1962 	} else
1963 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1964 	nd->nd_repstat = 0;
1965 	nd->nd_flag = 0;
1966 	if (*tl++ != rpc_vers) {
1967 		nd->nd_repstat = ERPCMISMATCH;
1968 		nd->nd_procnum = NFSPROC_NOOP;
1969 		return (0);
1970 	}
1971 	if (*tl != nfs_prog) {
1972 		nd->nd_repstat = EPROGUNAVAIL;
1973 		nd->nd_procnum = NFSPROC_NOOP;
1974 		return (0);
1975 	}
1976 	tl++;
1977 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1978 	if (nfsvers < NFS_VER2 || nfsvers > NFS_VER3) {
1979 		nd->nd_repstat = EPROGMISMATCH;
1980 		nd->nd_procnum = NFSPROC_NOOP;
1981 		return (0);
1982 	}
1983 	if (nfsvers == NFS_VER3)
1984 		nd->nd_flag = ND_NFSV3;
1985 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1986 	if (nd->nd_procnum == NFSPROC_NULL)
1987 		return (0);
1988 	if (nd->nd_procnum > NFSPROC_COMMIT ||
1989 	    (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1990 		nd->nd_repstat = EPROCUNAVAIL;
1991 		nd->nd_procnum = NFSPROC_NOOP;
1992 		return (0);
1993 	}
1994 	if ((nd->nd_flag & ND_NFSV3) == 0)
1995 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1996 	auth_type = *tl++;
1997 	len = fxdr_unsigned(int, *tl++);
1998 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1999 		m_freem(mrep);
2000 		return (EBADRPC);
2001 	}
2002 
2003 	nd->nd_flag &= ~ND_KERBAUTH;
2004 	/*
2005 	 * Handle auth_unix or auth_kerb.
2006 	 */
2007 	if (auth_type == rpc_auth_unix) {
2008 		uid_t uid;
2009 		gid_t gid;
2010 
2011 		nd->nd_cr = kauth_cred_alloc();
2012 		len = fxdr_unsigned(int, *++tl);
2013 		if (len < 0 || len > NFS_MAXNAMLEN) {
2014 			m_freem(mrep);
2015 			error = EBADRPC;
2016 			goto errout;
2017 		}
2018 		nfsm_adv(nfsm_rndup(len));
2019 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2020 
2021 		uid = fxdr_unsigned(uid_t, *tl++);
2022 		gid = fxdr_unsigned(gid_t, *tl++);
2023 		kauth_cred_setuid(nd->nd_cr, uid);
2024 		kauth_cred_seteuid(nd->nd_cr, uid);
2025 		kauth_cred_setsvuid(nd->nd_cr, uid);
2026 		kauth_cred_setgid(nd->nd_cr, gid);
2027 		kauth_cred_setegid(nd->nd_cr, gid);
2028 		kauth_cred_setsvgid(nd->nd_cr, gid);
2029 
2030 		len = fxdr_unsigned(int, *tl);
2031 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
2032 			m_freem(mrep);
2033 			error = EBADRPC;
2034 			goto errout;
2035 		}
2036 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
2037 
2038 		if (len > 0) {
2039 			size_t grbuf_size = min(len, NGROUPS) * sizeof(gid_t);
2040 			gid_t *grbuf = kmem_alloc(grbuf_size, KM_SLEEP);
2041 
2042 			for (i = 0; i < len; i++) {
2043 				if (i < NGROUPS) /* XXX elad */
2044 					grbuf[i] = fxdr_unsigned(gid_t, *tl++);
2045 				else
2046 					tl++;
2047 			}
2048 			kauth_cred_setgroups(nd->nd_cr, grbuf,
2049 			    min(len, NGROUPS), -1, UIO_SYSSPACE);
2050 			kmem_free(grbuf, grbuf_size);
2051 		}
2052 
2053 		len = fxdr_unsigned(int, *++tl);
2054 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
2055 			m_freem(mrep);
2056 			error = EBADRPC;
2057 			goto errout;
2058 		}
2059 		if (len > 0)
2060 			nfsm_adv(nfsm_rndup(len));
2061 	} else if (auth_type == rpc_auth_kerb) {
2062 		switch (fxdr_unsigned(int, *tl++)) {
2063 		case RPCAKN_FULLNAME:
2064 			ticklen = fxdr_unsigned(int, *tl);
2065 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
2066 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
2067 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
2068 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
2069 				m_freem(mrep);
2070 				error = EBADRPC;
2071 				goto errout;
2072 			}
2073 			uio.uio_offset = 0;
2074 			uio.uio_iov = &iov;
2075 			uio.uio_iovcnt = 1;
2076 			UIO_SETUP_SYSSPACE(&uio);
2077 			iov.iov_base = (void *)&nfsd->nfsd_authstr[4];
2078 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
2079 			nfsm_mtouio(&uio, uio.uio_resid);
2080 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2081 			if (*tl++ != rpc_auth_kerb ||
2082 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
2083 				printf("Bad kerb verifier\n");
2084 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2085 				nd->nd_procnum = NFSPROC_NOOP;
2086 				return (0);
2087 			}
2088 			nfsm_dissect(cp, void *, 4 * NFSX_UNSIGNED);
2089 			tl = (u_int32_t *)cp;
2090 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
2091 				printf("Not fullname kerb verifier\n");
2092 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2093 				nd->nd_procnum = NFSPROC_NOOP;
2094 				return (0);
2095 			}
2096 			cp += NFSX_UNSIGNED;
2097 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
2098 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
2099 			nd->nd_flag |= ND_KERBFULL;
2100 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
2101 			break;
2102 		case RPCAKN_NICKNAME:
2103 			if (len != 2 * NFSX_UNSIGNED) {
2104 				printf("Kerb nickname short\n");
2105 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
2106 				nd->nd_procnum = NFSPROC_NOOP;
2107 				return (0);
2108 			}
2109 			nickuid = fxdr_unsigned(uid_t, *tl);
2110 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2111 			if (*tl++ != rpc_auth_kerb ||
2112 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
2113 				printf("Kerb nick verifier bad\n");
2114 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2115 				nd->nd_procnum = NFSPROC_NOOP;
2116 				return (0);
2117 			}
2118 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2119 			tvin.tv_sec = *tl++;
2120 			tvin.tv_usec = *tl;
2121 
2122 			LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid),
2123 			    nu_hash) {
2124 				if (kauth_cred_geteuid(nuidp->nu_cr) == nickuid &&
2125 				    (!nd->nd_nam2 ||
2126 				     netaddr_match(NU_NETFAM(nuidp),
2127 				      &nuidp->nu_haddr, nd->nd_nam2)))
2128 					break;
2129 			}
2130 			if (!nuidp) {
2131 				nd->nd_repstat =
2132 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
2133 				nd->nd_procnum = NFSPROC_NOOP;
2134 				return (0);
2135 			}
2136 
2137 			/*
2138 			 * Now, decrypt the timestamp using the session key
2139 			 * and validate it.
2140 			 */
2141 #ifdef NFSKERB
2142 			XXX
2143 #endif
2144 
2145 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
2146 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
2147 			if (nuidp->nu_expire < time_second ||
2148 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
2149 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
2150 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
2151 				nuidp->nu_expire = 0;
2152 				nd->nd_repstat =
2153 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
2154 				nd->nd_procnum = NFSPROC_NOOP;
2155 				return (0);
2156 			}
2157 			kauth_cred_hold(nuidp->nu_cr);
2158 			nd->nd_cr = nuidp->nu_cr;
2159 			nd->nd_flag |= ND_KERBNICK;
2160 		}
2161 	} else {
2162 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
2163 		nd->nd_procnum = NFSPROC_NOOP;
2164 		return (0);
2165 	}
2166 
2167 	nd->nd_md = md;
2168 	nd->nd_dpos = dpos;
2169 	KASSERT((nd->nd_cr == NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) != 0)
2170 	     || (nd->nd_cr != NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) == 0));
2171 	return (0);
2172 nfsmout:
2173 errout:
2174 	KASSERT(error != 0);
2175 	if (nd->nd_cr != NULL) {
2176 		kauth_cred_free(nd->nd_cr);
2177 		nd->nd_cr = NULL;
2178 	}
2179 	return (error);
2180 }
2181 
2182 int
2183 nfs_msg(l, server, msg)
2184 	struct lwp *l;
2185 	const char *server, *msg;
2186 {
2187 	tpr_t tpr;
2188 
2189 	if (l)
2190 		tpr = tprintf_open(l->l_proc);
2191 	else
2192 		tpr = NULL;
2193 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
2194 	tprintf_close(tpr);
2195 	return (0);
2196 }
2197 
2198 static struct pool nfs_srvdesc_pool;
2199 
2200 void
2201 nfsdreq_init(void)
2202 {
2203 
2204 	pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript),
2205 	    0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr, IPL_NONE);
2206 }
2207 
2208 void
2209 nfsdreq_fini(void)
2210 {
2211 
2212 	pool_destroy(&nfs_srvdesc_pool);
2213 }
2214 
2215 struct nfsrv_descript *
2216 nfsdreq_alloc(void)
2217 {
2218 	struct nfsrv_descript *nd;
2219 
2220 	nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK);
2221 	nd->nd_cr = NULL;
2222 	return nd;
2223 }
2224 
2225 void
2226 nfsdreq_free(struct nfsrv_descript *nd)
2227 {
2228 	kauth_cred_t cr;
2229 
2230 	cr = nd->nd_cr;
2231 	if (cr != NULL) {
2232 		kauth_cred_free(cr);
2233 	}
2234 	pool_put(&nfs_srvdesc_pool, nd);
2235 }
2236