xref: /netbsd-src/sys/nfs/nfs_socket.c (revision 0920b4f20b78ab1ccd9f2312fbe10deaf000cbf3)
1 /*	$NetBSD: nfs_socket.c,v 1.163 2007/08/05 09:40:39 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1991, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_socket.c	8.5 (Berkeley) 3/30/95
35  */
36 
37 /*
38  * Socket operations for use by nfs
39  */
40 
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: nfs_socket.c,v 1.163 2007/08/05 09:40:39 yamt Exp $");
43 
44 #include "fs_nfs.h"
45 #include "opt_nfs.h"
46 #include "opt_nfsserver.h"
47 #include "opt_mbuftrace.h"
48 #include "opt_inet.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/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/mbuf.h>
58 #include <sys/vnode.h>
59 #include <sys/domain.h>
60 #include <sys/protosw.h>
61 #include <sys/socket.h>
62 #include <sys/socketvar.h>
63 #include <sys/syslog.h>
64 #include <sys/tprintf.h>
65 #include <sys/namei.h>
66 #include <sys/signal.h>
67 #include <sys/signalvar.h>
68 #include <sys/kauth.h>
69 
70 #include <netinet/in.h>
71 #include <netinet/tcp.h>
72 
73 #include <nfs/rpcv2.h>
74 #include <nfs/nfsproto.h>
75 #include <nfs/nfs.h>
76 #include <nfs/xdr_subs.h>
77 #include <nfs/nfsm_subs.h>
78 #include <nfs/nfsmount.h>
79 #include <nfs/nfsnode.h>
80 #include <nfs/nfsrtt.h>
81 #include <nfs/nfs_var.h>
82 
83 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
84 #ifdef MBUFTRACE
85 struct mowner nfs_mowner = 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 /*
118  * Defines which timer to use for the procnum.
119  * 0 - default
120  * 1 - getattr
121  * 2 - lookup
122  * 3 - read
123  * 4 - write
124  */
125 static const int proct[NFS_NPROCS] = {
126 	[NFSPROC_NULL] = 0,
127 	[NFSPROC_GETATTR] = 1,
128 	[NFSPROC_SETATTR] = 0,
129 	[NFSPROC_LOOKUP] = 2,
130 	[NFSPROC_ACCESS] = 1,
131 	[NFSPROC_READLINK] = 3,
132 	[NFSPROC_READ] = 3,
133 	[NFSPROC_WRITE] = 4,
134 	[NFSPROC_CREATE] = 0,
135 	[NFSPROC_MKDIR] = 0,
136 	[NFSPROC_SYMLINK] = 0,
137 	[NFSPROC_MKNOD] = 0,
138 	[NFSPROC_REMOVE] = 0,
139 	[NFSPROC_RMDIR] = 0,
140 	[NFSPROC_RENAME] = 0,
141 	[NFSPROC_LINK] = 0,
142 	[NFSPROC_READDIR] = 3,
143 	[NFSPROC_READDIRPLUS] = 3,
144 	[NFSPROC_FSSTAT] = 0,
145 	[NFSPROC_FSINFO] = 0,
146 	[NFSPROC_PATHCONF] = 0,
147 	[NFSPROC_COMMIT] = 0,
148 	[NFSPROC_NOOP] = 0,
149 };
150 
151 /*
152  * There is a congestion window for outstanding rpcs maintained per mount
153  * point. The cwnd size is adjusted in roughly the way that:
154  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
155  * SIGCOMM '88". ACM, August 1988.
156  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
157  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
158  * of rpcs is in progress.
159  * (The sent count and cwnd are scaled for integer arith.)
160  * Variants of "slow start" were tried and were found to be too much of a
161  * performance hit (ave. rtt 3 times larger),
162  * I suspect due to the large rtt that nfs rpcs have.
163  */
164 #define	NFS_CWNDSCALE	256
165 #define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
166 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
167 int nfsrtton = 0;
168 struct nfsrtt nfsrtt;
169 struct nfsreqhead nfs_reqq;
170 static callout_t nfs_timer_ch;
171 static struct evcnt nfs_timer_ev;
172 static struct evcnt nfs_timer_start_ev;
173 static struct evcnt nfs_timer_stop_ev;
174 
175 static int nfs_sndlock(struct nfsmount *, struct nfsreq *);
176 static void nfs_sndunlock(struct nfsmount *);
177 static int nfs_rcvlock(struct nfsmount *, struct nfsreq *);
178 static void nfs_rcvunlock(struct nfsmount *);
179 
180 /*
181  * Initialize sockets and congestion for a new NFS connection.
182  * We do not free the sockaddr if error.
183  */
184 int
185 nfs_connect(nmp, rep, l)
186 	struct nfsmount *nmp;
187 	struct nfsreq *rep;
188 	struct lwp *l;
189 {
190 	struct socket *so;
191 	int s, error, rcvreserve, sndreserve;
192 	struct sockaddr *saddr;
193 	struct sockaddr_in *sin;
194 #ifdef INET6
195 	struct sockaddr_in6 *sin6;
196 #endif
197 	struct mbuf *m;
198 
199 	nmp->nm_so = (struct socket *)0;
200 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
201 	error = socreate(saddr->sa_family, &nmp->nm_so,
202 		nmp->nm_sotype, nmp->nm_soproto, l);
203 	if (error)
204 		goto bad;
205 	so = nmp->nm_so;
206 #ifdef MBUFTRACE
207 	so->so_mowner = &nfs_mowner;
208 	so->so_rcv.sb_mowner = &nfs_mowner;
209 	so->so_snd.sb_mowner = &nfs_mowner;
210 #endif
211 	nmp->nm_soflags = so->so_proto->pr_flags;
212 
213 	/*
214 	 * Some servers require that the client port be a reserved port number.
215 	 */
216 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
217 		m = m_get(M_WAIT, MT_SOOPTS);
218 		MCLAIM(m, so->so_mowner);
219 		*mtod(m, int32_t *) = IP_PORTRANGE_LOW;
220 		m->m_len = sizeof(int32_t);
221 		if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m)))
222 			goto bad;
223 		m = m_get(M_WAIT, MT_SONAME);
224 		MCLAIM(m, so->so_mowner);
225 		sin = mtod(m, struct sockaddr_in *);
226 		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
227 		sin->sin_family = AF_INET;
228 		sin->sin_addr.s_addr = INADDR_ANY;
229 		sin->sin_port = 0;
230 		error = sobind(so, m, &lwp0);
231 		m_freem(m);
232 		if (error)
233 			goto bad;
234 	}
235 #ifdef INET6
236 	if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) {
237 		m = m_get(M_WAIT, MT_SOOPTS);
238 		MCLAIM(m, so->so_mowner);
239 		*mtod(m, int32_t *) = IPV6_PORTRANGE_LOW;
240 		m->m_len = sizeof(int32_t);
241 		if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m)))
242 			goto bad;
243 		m = m_get(M_WAIT, MT_SONAME);
244 		MCLAIM(m, so->so_mowner);
245 		sin6 = mtod(m, struct sockaddr_in6 *);
246 		sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6);
247 		sin6->sin6_family = AF_INET6;
248 		sin6->sin6_addr = in6addr_any;
249 		sin6->sin6_port = 0;
250 		error = sobind(so, m, &lwp0);
251 		m_freem(m);
252 		if (error)
253 			goto bad;
254 	}
255 #endif
256 
257 	/*
258 	 * Protocols that do not require connections may be optionally left
259 	 * unconnected for servers that reply from a port other than NFS_PORT.
260 	 */
261 	if (nmp->nm_flag & NFSMNT_NOCONN) {
262 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
263 			error = ENOTCONN;
264 			goto bad;
265 		}
266 	} else {
267 		error = soconnect(so, nmp->nm_nam, l);
268 		if (error)
269 			goto bad;
270 
271 		/*
272 		 * Wait for the connection to complete. Cribbed from the
273 		 * connect system call but with the wait timing out so
274 		 * that interruptible mounts don't hang here for a long time.
275 		 */
276 		s = splsoftnet();
277 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
278 			(void) tsleep((void *)&so->so_timeo, PSOCK,
279 				"nfscn1", 2 * hz);
280 			if ((so->so_state & SS_ISCONNECTING) &&
281 			    so->so_error == 0 && rep &&
282 			    (error = nfs_sigintr(nmp, rep, rep->r_lwp)) != 0){
283 				so->so_state &= ~SS_ISCONNECTING;
284 				splx(s);
285 				goto bad;
286 			}
287 		}
288 		if (so->so_error) {
289 			error = so->so_error;
290 			so->so_error = 0;
291 			splx(s);
292 			goto bad;
293 		}
294 		splx(s);
295 	}
296 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
297 		so->so_rcv.sb_timeo = (5 * hz);
298 		so->so_snd.sb_timeo = (5 * hz);
299 	} else {
300 		/*
301 		 * enable receive timeout to detect server crash and reconnect.
302 		 * otherwise, we can be stuck in soreceive forever.
303 		 */
304 		so->so_rcv.sb_timeo = (5 * hz);
305 		so->so_snd.sb_timeo = 0;
306 	}
307 	if (nmp->nm_sotype == SOCK_DGRAM) {
308 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
309 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
310 		    NFS_MAXPKTHDR) * 2;
311 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
312 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
313 		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
314 		    NFS_MAXPKTHDR) * 2;
315 	} else {
316 		if (nmp->nm_sotype != SOCK_STREAM)
317 			panic("nfscon sotype");
318 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
319 			m = m_get(M_WAIT, MT_SOOPTS);
320 			MCLAIM(m, so->so_mowner);
321 			*mtod(m, int32_t *) = 1;
322 			m->m_len = sizeof(int32_t);
323 			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
324 		}
325 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
326 			m = m_get(M_WAIT, MT_SOOPTS);
327 			MCLAIM(m, so->so_mowner);
328 			*mtod(m, int32_t *) = 1;
329 			m->m_len = sizeof(int32_t);
330 			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
331 		}
332 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
333 		    sizeof (u_int32_t)) * 2;
334 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
335 		    sizeof (u_int32_t)) * 2;
336 	}
337 	error = soreserve(so, sndreserve, rcvreserve);
338 	if (error)
339 		goto bad;
340 	so->so_rcv.sb_flags |= SB_NOINTR;
341 	so->so_snd.sb_flags |= SB_NOINTR;
342 
343 	/* Initialize other non-zero congestion variables */
344 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
345 		NFS_TIMEO << 3;
346 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
347 		nmp->nm_sdrtt[3] = 0;
348 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
349 	nmp->nm_sent = 0;
350 	nmp->nm_timeouts = 0;
351 	return (0);
352 
353 bad:
354 	nfs_disconnect(nmp);
355 	return (error);
356 }
357 
358 /*
359  * Reconnect routine:
360  * Called when a connection is broken on a reliable protocol.
361  * - clean up the old socket
362  * - nfs_connect() again
363  * - set R_MUSTRESEND for all outstanding requests on mount point
364  * If this fails the mount point is DEAD!
365  * nb: Must be called with the nfs_sndlock() set on the mount point.
366  */
367 int
368 nfs_reconnect(rep, l)
369 	struct nfsreq *rep;
370 	struct lwp *l;
371 {
372 	struct nfsreq *rp;
373 	struct nfsmount *nmp = rep->r_nmp;
374 	int error;
375 
376 	nfs_disconnect(nmp);
377 	while ((error = nfs_connect(nmp, rep, l)) != 0) {
378 		if (error == EINTR || error == ERESTART)
379 			return (EINTR);
380 		kpause("nfscn2", false, hz, NULL);
381 	}
382 
383 	/*
384 	 * Loop through outstanding request list and fix up all requests
385 	 * on old socket.
386 	 */
387 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
388 		if (rp->r_nmp == nmp) {
389 			if ((rp->r_flags & R_MUSTRESEND) == 0)
390 				rp->r_flags |= R_MUSTRESEND | R_REXMITTED;
391 			rp->r_rexmit = 0;
392 		}
393 	}
394 	return (0);
395 }
396 
397 /*
398  * NFS disconnect. Clean up and unlink.
399  */
400 void
401 nfs_disconnect(nmp)
402 	struct nfsmount *nmp;
403 {
404 	struct socket *so;
405 	int drain = 0;
406 
407 	if (nmp->nm_so) {
408 		so = nmp->nm_so;
409 		nmp->nm_so = (struct socket *)0;
410 		soshutdown(so, SHUT_RDWR);
411 		drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
412 		if (drain) {
413 			/*
414 			 * soshutdown() above should wake up the current
415 			 * listener.
416 			 * Now wake up those waiting for the receive lock, and
417 			 * wait for them to go away unhappy, to prevent *nmp
418 			 * from evaporating while they're sleeping.
419 			 */
420 			mutex_enter(&nmp->nm_lock);
421 			while (nmp->nm_waiters > 0) {
422 				cv_broadcast(&nmp->nm_rcvcv);
423 				cv_broadcast(&nmp->nm_sndcv);
424 				cv_wait(&nmp->nm_disconcv, &nmp->nm_lock);
425 			}
426 			mutex_exit(&nmp->nm_lock);
427 		}
428 		soclose(so);
429 	}
430 #ifdef DIAGNOSTIC
431 	if (drain && (nmp->nm_waiters > 0))
432 		panic("nfs_disconnect: waiters left after drain?");
433 #endif
434 }
435 
436 void
437 nfs_safedisconnect(nmp)
438 	struct nfsmount *nmp;
439 {
440 	struct nfsreq dummyreq;
441 
442 	memset(&dummyreq, 0, sizeof(dummyreq));
443 	dummyreq.r_nmp = nmp;
444 	nfs_rcvlock(nmp, &dummyreq); /* XXX ignored error return */
445 	nfs_disconnect(nmp);
446 	nfs_rcvunlock(nmp);
447 }
448 
449 /*
450  * This is the nfs send routine. For connection based socket types, it
451  * must be called with an nfs_sndlock() on the socket.
452  * "rep == NULL" indicates that it has been called from a server.
453  * For the client side:
454  * - return EINTR if the RPC is terminated, 0 otherwise
455  * - set R_MUSTRESEND if the send fails for any reason
456  * - do any cleanup required by recoverable socket errors (? ? ?)
457  * For the server side:
458  * - return EINTR or ERESTART if interrupted by a signal
459  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
460  * - do any cleanup required by recoverable socket errors (? ? ?)
461  */
462 int
463 nfs_send(so, nam, top, rep, l)
464 	struct socket *so;
465 	struct mbuf *nam;
466 	struct mbuf *top;
467 	struct nfsreq *rep;
468 	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, (struct uio *)0, top,
501 		    (struct mbuf *)0, flags,  l);
502 	if (error) {
503 		if (rep) {
504 			if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
505 				/*
506 				 * We're too fast for the network/driver,
507 				 * and UDP isn't flowcontrolled.
508 				 * We need to resend. This is not fatal,
509 				 * just try again.
510 				 *
511 				 * Could be smarter here by doing some sort
512 				 * of a backoff, but this is rare.
513 				 */
514 				rep->r_flags |= R_MUSTRESEND;
515 			} else {
516 				if (error != EPIPE)
517 					log(LOG_INFO,
518 					    "nfs send error %d for %s\n",
519 					    error,
520 					    rep->r_nmp->nm_mountp->
521 						    mnt_stat.f_mntfromname);
522 				/*
523 				 * Deal with errors for the client side.
524 				 */
525 				if (rep->r_flags & R_SOFTTERM)
526 					error = EINTR;
527 				else
528 					rep->r_flags |= R_MUSTRESEND;
529 			}
530 		} else {
531 			/*
532 			 * See above. This error can happen under normal
533 			 * circumstances and the log is too noisy.
534 			 * The error will still show up in nfsstat.
535 			 */
536 			if (error != ENOBUFS || so->so_type != SOCK_DGRAM)
537 				log(LOG_INFO, "nfsd send error %d\n", error);
538 		}
539 
540 		/*
541 		 * Handle any recoverable (soft) socket errors here. (? ? ?)
542 		 */
543 		if (error != EINTR && error != ERESTART &&
544 			error != EWOULDBLOCK && error != EPIPE)
545 			error = 0;
546 	}
547 	return (error);
548 }
549 
550 #ifdef NFS
551 /*
552  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
553  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
554  * Mark and consolidate the data into a new mbuf list.
555  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
556  *     small mbufs.
557  * For SOCK_STREAM we must be very careful to read an entire record once
558  * we have read any of it, even if the system call has been interrupted.
559  */
560 int
561 nfs_receive(rep, aname, mp, l)
562 	struct nfsreq *rep;
563 	struct mbuf **aname;
564 	struct mbuf **mp;
565 	struct lwp *l;
566 {
567 	struct socket *so;
568 	struct uio auio;
569 	struct iovec aio;
570 	struct mbuf *m;
571 	struct mbuf *control;
572 	u_int32_t len;
573 	struct mbuf **getnam;
574 	int error, sotype, rcvflg;
575 
576 	/*
577 	 * Set up arguments for soreceive()
578 	 */
579 	*mp = (struct mbuf *)0;
580 	*aname = (struct mbuf *)0;
581 	sotype = rep->r_nmp->nm_sotype;
582 
583 	/*
584 	 * For reliable protocols, lock against other senders/receivers
585 	 * in case a reconnect is necessary.
586 	 * For SOCK_STREAM, first get the Record Mark to find out how much
587 	 * more there is to get.
588 	 * We must lock the socket against other receivers
589 	 * until we have an entire rpc request/reply.
590 	 */
591 	if (sotype != SOCK_DGRAM) {
592 		error = nfs_sndlock(rep->r_nmp, rep);
593 		if (error)
594 			return (error);
595 tryagain:
596 		/*
597 		 * Check for fatal errors and resending request.
598 		 */
599 		/*
600 		 * Ugh: If a reconnect attempt just happened, nm_so
601 		 * would have changed. NULL indicates a failed
602 		 * attempt that has essentially shut down this
603 		 * mount point.
604 		 */
605 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
606 			nfs_sndunlock(rep->r_nmp);
607 			return (EINTR);
608 		}
609 		so = rep->r_nmp->nm_so;
610 		if (!so) {
611 			error = nfs_reconnect(rep, l);
612 			if (error) {
613 				nfs_sndunlock(rep->r_nmp);
614 				return (error);
615 			}
616 			goto tryagain;
617 		}
618 		while (rep->r_flags & R_MUSTRESEND) {
619 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
620 			nfsstats.rpcretries++;
621 			rep->r_rtt = 0;
622 			rep->r_flags &= ~R_TIMING;
623 			error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, l);
624 			if (error) {
625 				if (error == EINTR || error == ERESTART ||
626 				    (error = nfs_reconnect(rep, l)) != 0) {
627 					nfs_sndunlock(rep->r_nmp);
628 					return (error);
629 				}
630 				goto tryagain;
631 			}
632 		}
633 		nfs_sndunlock(rep->r_nmp);
634 		if (sotype == SOCK_STREAM) {
635 			aio.iov_base = (void *) &len;
636 			aio.iov_len = sizeof(u_int32_t);
637 			auio.uio_iov = &aio;
638 			auio.uio_iovcnt = 1;
639 			auio.uio_rw = UIO_READ;
640 			auio.uio_offset = 0;
641 			auio.uio_resid = sizeof(u_int32_t);
642 			UIO_SETUP_SYSSPACE(&auio);
643 			do {
644 			   rcvflg = MSG_WAITALL;
645 			   error = (*so->so_receive)(so, (struct mbuf **)0, &auio,
646 				(struct mbuf **)0, (struct mbuf **)0, &rcvflg);
647 			   if (error == EWOULDBLOCK && rep) {
648 				if (rep->r_flags & R_SOFTTERM)
649 					return (EINTR);
650 				/*
651 				 * if it seems that the server died after it
652 				 * received our request, set EPIPE so that
653 				 * we'll reconnect and retransmit requests.
654 				 */
655 				if (rep->r_rexmit >= rep->r_nmp->nm_retry) {
656 					nfsstats.rpctimeouts++;
657 					error = EPIPE;
658 				}
659 			   }
660 			} while (error == EWOULDBLOCK);
661 			if (!error && auio.uio_resid > 0) {
662 			    /*
663 			     * Don't log a 0 byte receive; it means
664 			     * that the socket has been closed, and
665 			     * can happen during normal operation
666 			     * (forcible unmount or Solaris server).
667 			     */
668 			    if (auio.uio_resid != sizeof (u_int32_t))
669 			      log(LOG_INFO,
670 				 "short receive (%lu/%lu) from nfs server %s\n",
671 				 (u_long)sizeof(u_int32_t) - auio.uio_resid,
672 				 (u_long)sizeof(u_int32_t),
673 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
674 			    error = EPIPE;
675 			}
676 			if (error)
677 				goto errout;
678 			len = ntohl(len) & ~0x80000000;
679 			/*
680 			 * This is SERIOUS! We are out of sync with the sender
681 			 * and forcing a disconnect/reconnect is all I can do.
682 			 */
683 			if (len > NFS_MAXPACKET) {
684 			    log(LOG_ERR, "%s (%d) from nfs server %s\n",
685 				"impossible packet length",
686 				len,
687 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
688 			    error = EFBIG;
689 			    goto errout;
690 			}
691 			auio.uio_resid = len;
692 			do {
693 			    rcvflg = MSG_WAITALL;
694 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
695 				&auio, mp, (struct mbuf **)0, &rcvflg);
696 			} while (error == EWOULDBLOCK || error == EINTR ||
697 				 error == ERESTART);
698 			if (!error && auio.uio_resid > 0) {
699 			    if (len != auio.uio_resid)
700 			      log(LOG_INFO,
701 				"short receive (%lu/%d) from nfs server %s\n",
702 				(u_long)len - auio.uio_resid, len,
703 				rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
704 			    error = EPIPE;
705 			}
706 		} else {
707 			/*
708 			 * NB: Since uio_resid is big, MSG_WAITALL is ignored
709 			 * and soreceive() will return when it has either a
710 			 * control msg or a data msg.
711 			 * We have no use for control msg., but must grab them
712 			 * and then throw them away so we know what is going
713 			 * on.
714 			 */
715 			auio.uio_resid = len = 100000000; /* Anything Big */
716 			/* not need to setup uio_vmspace */
717 			do {
718 			    rcvflg = 0;
719 			    error =  (*so->so_receive)(so, (struct mbuf **)0,
720 				&auio, mp, &control, &rcvflg);
721 			    if (control)
722 				m_freem(control);
723 			    if (error == EWOULDBLOCK && rep) {
724 				if (rep->r_flags & R_SOFTTERM)
725 					return (EINTR);
726 			    }
727 			} while (error == EWOULDBLOCK ||
728 				 (!error && *mp == NULL && control));
729 			if ((rcvflg & MSG_EOR) == 0)
730 				printf("Egad!!\n");
731 			if (!error && *mp == NULL)
732 				error = EPIPE;
733 			len -= auio.uio_resid;
734 		}
735 errout:
736 		if (error && error != EINTR && error != ERESTART) {
737 			m_freem(*mp);
738 			*mp = (struct mbuf *)0;
739 			if (error != EPIPE)
740 				log(LOG_INFO,
741 				    "receive error %d from nfs server %s\n",
742 				    error,
743 				 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
744 			error = nfs_sndlock(rep->r_nmp, rep);
745 			if (!error)
746 				error = nfs_reconnect(rep, l);
747 			if (!error)
748 				goto tryagain;
749 			else
750 				nfs_sndunlock(rep->r_nmp);
751 		}
752 	} else {
753 		if ((so = rep->r_nmp->nm_so) == NULL)
754 			return (EACCES);
755 		if (so->so_state & SS_ISCONNECTED)
756 			getnam = (struct mbuf **)0;
757 		else
758 			getnam = aname;
759 		auio.uio_resid = len = 1000000;
760 		/* not need to setup uio_vmspace */
761 		do {
762 			rcvflg = 0;
763 			error =  (*so->so_receive)(so, getnam, &auio, mp,
764 				(struct mbuf **)0, &rcvflg);
765 			if (error == EWOULDBLOCK &&
766 			    (rep->r_flags & R_SOFTTERM))
767 				return (EINTR);
768 		} while (error == EWOULDBLOCK);
769 		len -= auio.uio_resid;
770 		if (!error && *mp == NULL)
771 			error = EPIPE;
772 	}
773 	if (error) {
774 		m_freem(*mp);
775 		*mp = (struct mbuf *)0;
776 	}
777 	return (error);
778 }
779 
780 /*
781  * Implement receipt of reply on a socket.
782  * We must search through the list of received datagrams matching them
783  * with outstanding requests using the xid, until ours is found.
784  */
785 /* ARGSUSED */
786 int
787 nfs_reply(myrep, lwp)
788 	struct nfsreq *myrep;
789 	struct lwp *lwp;
790 {
791 	struct nfsreq *rep;
792 	struct nfsmount *nmp = myrep->r_nmp;
793 	int32_t t1;
794 	struct mbuf *mrep, *nam, *md;
795 	u_int32_t rxid, *tl;
796 	char *dpos, *cp2;
797 	int error;
798 
799 	/*
800 	 * Loop around until we get our own reply
801 	 */
802 	for (;;) {
803 		/*
804 		 * Lock against other receivers so that I don't get stuck in
805 		 * sbwait() after someone else has received my reply for me.
806 		 * Also necessary for connection based protocols to avoid
807 		 * race conditions during a reconnect.
808 		 */
809 		error = nfs_rcvlock(nmp, myrep);
810 		if (error == EALREADY)
811 			return (0);
812 		if (error)
813 			return (error);
814 		/*
815 		 * Get the next Rpc reply off the socket
816 		 */
817 
818 		mutex_enter(&nmp->nm_lock);
819 		nmp->nm_waiters++;
820 		mutex_exit(&nmp->nm_lock);
821 
822 		error = nfs_receive(myrep, &nam, &mrep, lwp);
823 
824 		mutex_enter(&nmp->nm_lock);
825 		nmp->nm_waiters--;
826 		cv_signal(&nmp->nm_disconcv);
827 		mutex_exit(&nmp->nm_lock);
828 
829 		if (error) {
830 			nfs_rcvunlock(nmp);
831 
832 			if (nmp->nm_iflag & NFSMNT_DISMNT) {
833 				/*
834 				 * Oops, we're going away now..
835 				 */
836 				return error;
837 			}
838 			/*
839 			 * Ignore routing errors on connectionless protocols? ?
840 			 */
841 			if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
842 				nmp->nm_so->so_error = 0;
843 #ifdef DEBUG
844 				printf("nfs_reply: ignoring error %d\n", error);
845 #endif
846 				continue;
847 			}
848 			return (error);
849 		}
850 		if (nam)
851 			m_freem(nam);
852 
853 		/*
854 		 * Get the xid and check that it is an rpc reply
855 		 */
856 		md = mrep;
857 		dpos = mtod(md, void *);
858 		nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
859 		rxid = *tl++;
860 		if (*tl != rpc_reply) {
861 			nfsstats.rpcinvalid++;
862 			m_freem(mrep);
863 nfsmout:
864 			nfs_rcvunlock(nmp);
865 			continue;
866 		}
867 
868 		/*
869 		 * Loop through the request list to match up the reply
870 		 * Iff no match, just drop the datagram
871 		 */
872 		TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
873 			if (rep->r_mrep == NULL && rxid == rep->r_xid) {
874 				/* Found it.. */
875 				rep->r_mrep = mrep;
876 				rep->r_md = md;
877 				rep->r_dpos = dpos;
878 				if (nfsrtton) {
879 					struct rttl *rt;
880 
881 					rt = &nfsrtt.rttl[nfsrtt.pos];
882 					rt->proc = rep->r_procnum;
883 					rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
884 					rt->sent = nmp->nm_sent;
885 					rt->cwnd = nmp->nm_cwnd;
886 					rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
887 					rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
888 					rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx;
889 					getmicrotime(&rt->tstamp);
890 					if (rep->r_flags & R_TIMING)
891 						rt->rtt = rep->r_rtt;
892 					else
893 						rt->rtt = 1000000;
894 					nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
895 				}
896 				/*
897 				 * Update congestion window.
898 				 * Do the additive increase of
899 				 * one rpc/rtt.
900 				 */
901 				if (nmp->nm_cwnd <= nmp->nm_sent) {
902 					nmp->nm_cwnd +=
903 					   (NFS_CWNDSCALE * NFS_CWNDSCALE +
904 					   (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
905 					if (nmp->nm_cwnd > NFS_MAXCWND)
906 						nmp->nm_cwnd = NFS_MAXCWND;
907 				}
908 				rep->r_flags &= ~R_SENT;
909 				nmp->nm_sent -= NFS_CWNDSCALE;
910 				/*
911 				 * Update rtt using a gain of 0.125 on the mean
912 				 * and a gain of 0.25 on the deviation.
913 				 */
914 				if (rep->r_flags & R_TIMING) {
915 					/*
916 					 * Since the timer resolution of
917 					 * NFS_HZ is so course, it can often
918 					 * result in r_rtt == 0. Since
919 					 * r_rtt == N means that the actual
920 					 * rtt is between N+dt and N+2-dt ticks,
921 					 * add 1.
922 					 */
923 					t1 = rep->r_rtt + 1;
924 					t1 -= (NFS_SRTT(rep) >> 3);
925 					NFS_SRTT(rep) += t1;
926 					if (t1 < 0)
927 						t1 = -t1;
928 					t1 -= (NFS_SDRTT(rep) >> 2);
929 					NFS_SDRTT(rep) += t1;
930 				}
931 				nmp->nm_timeouts = 0;
932 				break;
933 			}
934 		}
935 		nfs_rcvunlock(nmp);
936 		/*
937 		 * If not matched to a request, drop it.
938 		 * If it's mine, get out.
939 		 */
940 		if (rep == 0) {
941 			nfsstats.rpcunexpected++;
942 			m_freem(mrep);
943 		} else if (rep == myrep) {
944 			if (rep->r_mrep == NULL)
945 				panic("nfsreply nil");
946 			return (0);
947 		}
948 	}
949 }
950 
951 /*
952  * nfs_request - goes something like this
953  *	- fill in request struct
954  *	- links it into list
955  *	- calls nfs_send() for first transmit
956  *	- calls nfs_receive() to get reply
957  *	- break down rpc header and return with nfs reply pointed to
958  *	  by mrep or error
959  * nb: always frees up mreq mbuf list
960  */
961 int
962 nfs_request(np, mrest, procnum, lwp, cred, mrp, mdp, dposp, rexmitp)
963 	struct nfsnode *np;
964 	struct mbuf *mrest;
965 	int procnum;
966 	struct lwp *lwp;
967 	kauth_cred_t cred;
968 	struct mbuf **mrp;
969 	struct mbuf **mdp;
970 	char **dposp;
971 	int *rexmitp;
972 {
973 	struct mbuf *m, *mrep;
974 	struct nfsreq *rep;
975 	u_int32_t *tl;
976 	int i;
977 	struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount);
978 	struct mbuf *md, *mheadend;
979 	char nickv[RPCX_NICKVERF];
980 	time_t waituntil;
981 	char *dpos, *cp2;
982 	int t1, s, error = 0, mrest_len, auth_len, auth_type;
983 	int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0;
984 	int verf_len, verf_type;
985 	u_int32_t xid;
986 	char *auth_str, *verf_str;
987 	NFSKERBKEY_T key;		/* save session key */
988 	kauth_cred_t acred;
989 	struct mbuf *mrest_backup = NULL;
990 	kauth_cred_t origcred = NULL; /* XXX: gcc */
991 	bool retry_cred = true;
992 	bool use_opencred = (np->n_flag & NUSEOPENCRED) != 0;
993 
994 	if (rexmitp != NULL)
995 		*rexmitp = 0;
996 
997 	acred = kauth_cred_alloc();
998 
999 tryagain_cred:
1000 	KASSERT(cred != NULL);
1001 	MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
1002 	rep->r_nmp = nmp;
1003 	KASSERT(lwp == NULL || lwp == curlwp);
1004 	rep->r_lwp = lwp;
1005 	rep->r_procnum = procnum;
1006 	i = 0;
1007 	m = mrest;
1008 	while (m) {
1009 		i += m->m_len;
1010 		m = m->m_next;
1011 	}
1012 	mrest_len = i;
1013 
1014 	/*
1015 	 * Get the RPC header with authorization.
1016 	 */
1017 kerbauth:
1018 	verf_str = auth_str = (char *)0;
1019 	if (nmp->nm_flag & NFSMNT_KERB) {
1020 		verf_str = nickv;
1021 		verf_len = sizeof (nickv);
1022 		auth_type = RPCAUTH_KERB4;
1023 		memset((void *)key, 0, sizeof (key));
1024 		if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
1025 			&auth_len, verf_str, verf_len)) {
1026 			error = nfs_getauth(nmp, rep, cred, &auth_str,
1027 				&auth_len, verf_str, &verf_len, key);
1028 			if (error) {
1029 				free((void *)rep, M_NFSREQ);
1030 				m_freem(mrest);
1031 				KASSERT(kauth_cred_getrefcnt(acred) == 1);
1032 				kauth_cred_free(acred);
1033 				return (error);
1034 			}
1035 		}
1036 		retry_cred = false;
1037 	} else {
1038 		/* AUTH_UNIX */
1039 		uid_t uid;
1040 		gid_t gid;
1041 
1042 		/*
1043 		 * on the most unix filesystems, permission checks are
1044 		 * done when the file is open(2)'ed.
1045 		 * ie. once a file is successfully open'ed,
1046 		 * following i/o operations never fail with EACCES.
1047 		 * we try to follow the semantics as far as possible.
1048 		 *
1049 		 * note that we expect that the nfs server always grant
1050 		 * accesses by the file's owner.
1051 		 */
1052 		origcred = cred;
1053 		switch (procnum) {
1054 		case NFSPROC_READ:
1055 		case NFSPROC_WRITE:
1056 		case NFSPROC_COMMIT:
1057 			uid = np->n_vattr->va_uid;
1058 			gid = np->n_vattr->va_gid;
1059 			if (kauth_cred_geteuid(cred) == uid &&
1060 			    kauth_cred_getegid(cred) == gid) {
1061 				retry_cred = false;
1062 				break;
1063 			}
1064 			if (use_opencred)
1065 				break;
1066 			kauth_cred_setuid(acred, uid);
1067 			kauth_cred_seteuid(acred, uid);
1068 			kauth_cred_setsvuid(acred, uid);
1069 			kauth_cred_setgid(acred, gid);
1070 			kauth_cred_setegid(acred, gid);
1071 			kauth_cred_setsvgid(acred, gid);
1072 			cred = acred;
1073 			break;
1074 		default:
1075 			retry_cred = false;
1076 			break;
1077 		}
1078 		/*
1079 		 * backup mbuf chain if we can need it later to retry.
1080 		 *
1081 		 * XXX maybe we can keep a direct reference to
1082 		 * mrest without doing m_copym, but it's ...ugly.
1083 		 */
1084 		if (retry_cred)
1085 			mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT);
1086 		auth_type = RPCAUTH_UNIX;
1087 		/* XXX elad - ngroups */
1088 		auth_len = (((kauth_cred_ngroups(cred) > nmp->nm_numgrps) ?
1089 			nmp->nm_numgrps : kauth_cred_ngroups(cred)) << 2) +
1090 			5 * NFSX_UNSIGNED;
1091 	}
1092 	m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
1093 	     auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
1094 	if (auth_str)
1095 		free(auth_str, M_TEMP);
1096 
1097 	/*
1098 	 * For stream protocols, insert a Sun RPC Record Mark.
1099 	 */
1100 	if (nmp->nm_sotype == SOCK_STREAM) {
1101 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
1102 		*mtod(m, u_int32_t *) = htonl(0x80000000 |
1103 			 (m->m_pkthdr.len - NFSX_UNSIGNED));
1104 	}
1105 	rep->r_mreq = m;
1106 	rep->r_xid = xid;
1107 tryagain:
1108 	if (nmp->nm_flag & NFSMNT_SOFT)
1109 		rep->r_retry = nmp->nm_retry;
1110 	else
1111 		rep->r_retry = NFS_MAXREXMIT + 1;	/* past clip limit */
1112 	rep->r_rtt = rep->r_rexmit = 0;
1113 	if (proct[procnum] > 0)
1114 		rep->r_flags = R_TIMING;
1115 	else
1116 		rep->r_flags = 0;
1117 	rep->r_mrep = NULL;
1118 
1119 	/*
1120 	 * Do the client side RPC.
1121 	 */
1122 	nfsstats.rpcrequests++;
1123 	/*
1124 	 * Chain request into list of outstanding requests. Be sure
1125 	 * to put it LAST so timer finds oldest requests first.
1126 	 */
1127 	s = splsoftnet();
1128 	TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1129 	nfs_timer_start();
1130 
1131 	/*
1132 	 * If backing off another request or avoiding congestion, don't
1133 	 * send this one now but let timer do it. If not timing a request,
1134 	 * do it now.
1135 	 */
1136 	if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1137 		(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1138 		nmp->nm_sent < nmp->nm_cwnd)) {
1139 		splx(s);
1140 		if (nmp->nm_soflags & PR_CONNREQUIRED)
1141 			error = nfs_sndlock(nmp, rep);
1142 		if (!error) {
1143 			m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
1144 			error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp);
1145 			if (nmp->nm_soflags & PR_CONNREQUIRED)
1146 				nfs_sndunlock(nmp);
1147 		}
1148 		if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1149 			nmp->nm_sent += NFS_CWNDSCALE;
1150 			rep->r_flags |= R_SENT;
1151 		}
1152 	} else {
1153 		splx(s);
1154 		rep->r_rtt = -1;
1155 	}
1156 
1157 	/*
1158 	 * Wait for the reply from our send or the timer's.
1159 	 */
1160 	if (!error || error == EPIPE)
1161 		error = nfs_reply(rep, lwp);
1162 
1163 	/*
1164 	 * RPC done, unlink the request.
1165 	 */
1166 	s = splsoftnet();
1167 	TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1168 	splx(s);
1169 
1170 	/*
1171 	 * Decrement the outstanding request count.
1172 	 */
1173 	if (rep->r_flags & R_SENT) {
1174 		rep->r_flags &= ~R_SENT;	/* paranoia */
1175 		nmp->nm_sent -= NFS_CWNDSCALE;
1176 	}
1177 
1178 	if (rexmitp != NULL) {
1179 		int rexmit;
1180 
1181 		if (nmp->nm_sotype != SOCK_DGRAM)
1182 			rexmit = (rep->r_flags & R_REXMITTED) != 0;
1183 		else
1184 			rexmit = rep->r_rexmit;
1185 		*rexmitp = rexmit;
1186 	}
1187 
1188 	/*
1189 	 * If there was a successful reply and a tprintf msg.
1190 	 * tprintf a response.
1191 	 */
1192 	if (!error && (rep->r_flags & R_TPRINTFMSG))
1193 		nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1194 		    "is alive again");
1195 	mrep = rep->r_mrep;
1196 	md = rep->r_md;
1197 	dpos = rep->r_dpos;
1198 	if (error)
1199 		goto nfsmout;
1200 
1201 	/*
1202 	 * break down the rpc header and check if ok
1203 	 */
1204 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1205 	if (*tl++ == rpc_msgdenied) {
1206 		if (*tl == rpc_mismatch)
1207 			error = EOPNOTSUPP;
1208 		else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1209 			if (!failed_auth) {
1210 				failed_auth++;
1211 				mheadend->m_next = (struct mbuf *)0;
1212 				m_freem(mrep);
1213 				m_freem(rep->r_mreq);
1214 				goto kerbauth;
1215 			} else
1216 				error = EAUTH;
1217 		} else
1218 			error = EACCES;
1219 		m_freem(mrep);
1220 		goto nfsmout;
1221 	}
1222 
1223 	/*
1224 	 * Grab any Kerberos verifier, otherwise just throw it away.
1225 	 */
1226 	verf_type = fxdr_unsigned(int, *tl++);
1227 	i = fxdr_unsigned(int32_t, *tl);
1228 	if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1229 		error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1230 		if (error)
1231 			goto nfsmout;
1232 	} else if (i > 0)
1233 		nfsm_adv(nfsm_rndup(i));
1234 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1235 	/* 0 == ok */
1236 	if (*tl == 0) {
1237 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1238 		if (*tl != 0) {
1239 			error = fxdr_unsigned(int, *tl);
1240 			switch (error) {
1241 			case NFSERR_PERM:
1242 				error = EPERM;
1243 				break;
1244 
1245 			case NFSERR_NOENT:
1246 				error = ENOENT;
1247 				break;
1248 
1249 			case NFSERR_IO:
1250 				error = EIO;
1251 				break;
1252 
1253 			case NFSERR_NXIO:
1254 				error = ENXIO;
1255 				break;
1256 
1257 			case NFSERR_ACCES:
1258 				error = EACCES;
1259 				if (!retry_cred)
1260 					break;
1261 				m_freem(mrep);
1262 				m_freem(rep->r_mreq);
1263 				FREE(rep, M_NFSREQ);
1264 				use_opencred = !use_opencred;
1265 				if (mrest_backup == NULL) {
1266 					/* m_copym failure */
1267 					KASSERT(
1268 					    kauth_cred_getrefcnt(acred) == 1);
1269 					kauth_cred_free(acred);
1270 					return ENOMEM;
1271 				}
1272 				mrest = mrest_backup;
1273 				mrest_backup = NULL;
1274 				cred = origcred;
1275 				error = 0;
1276 				retry_cred = false;
1277 				goto tryagain_cred;
1278 
1279 			case NFSERR_EXIST:
1280 				error = EEXIST;
1281 				break;
1282 
1283 			case NFSERR_XDEV:
1284 				error = EXDEV;
1285 				break;
1286 
1287 			case NFSERR_NODEV:
1288 				error = ENODEV;
1289 				break;
1290 
1291 			case NFSERR_NOTDIR:
1292 				error = ENOTDIR;
1293 				break;
1294 
1295 			case NFSERR_ISDIR:
1296 				error = EISDIR;
1297 				break;
1298 
1299 			case NFSERR_INVAL:
1300 				error = EINVAL;
1301 				break;
1302 
1303 			case NFSERR_FBIG:
1304 				error = EFBIG;
1305 				break;
1306 
1307 			case NFSERR_NOSPC:
1308 				error = ENOSPC;
1309 				break;
1310 
1311 			case NFSERR_ROFS:
1312 				error = EROFS;
1313 				break;
1314 
1315 			case NFSERR_MLINK:
1316 				error = EMLINK;
1317 				break;
1318 
1319 			case NFSERR_TIMEDOUT:
1320 				error = ETIMEDOUT;
1321 				break;
1322 
1323 			case NFSERR_NAMETOL:
1324 				error = ENAMETOOLONG;
1325 				break;
1326 
1327 			case NFSERR_NOTEMPTY:
1328 				error = ENOTEMPTY;
1329 				break;
1330 
1331 			case NFSERR_DQUOT:
1332 				error = EDQUOT;
1333 				break;
1334 
1335 			case NFSERR_STALE:
1336 				/*
1337 				 * If the File Handle was stale, invalidate the
1338 				 * lookup cache, just in case.
1339 				 */
1340 				error = ESTALE;
1341 				cache_purge(NFSTOV(np));
1342 				break;
1343 
1344 			case NFSERR_REMOTE:
1345 				error = EREMOTE;
1346 				break;
1347 
1348 			case NFSERR_WFLUSH:
1349 			case NFSERR_BADHANDLE:
1350 			case NFSERR_NOT_SYNC:
1351 			case NFSERR_BAD_COOKIE:
1352 				error = EINVAL;
1353 				break;
1354 
1355 			case NFSERR_NOTSUPP:
1356 				error = ENOTSUP;
1357 				break;
1358 
1359 			case NFSERR_TOOSMALL:
1360 			case NFSERR_SERVERFAULT:
1361 			case NFSERR_BADTYPE:
1362 				error = EINVAL;
1363 				break;
1364 
1365 			case NFSERR_TRYLATER:
1366 				if ((nmp->nm_flag & NFSMNT_NFSV3) == 0)
1367 					break;
1368 				m_freem(mrep);
1369 				error = 0;
1370 				waituntil = time_second + trylater_delay;
1371 				while (time_second < waituntil) {
1372 					kpause("nfstrylater", false, hz, NULL);
1373 				}
1374 				trylater_delay *= NFS_TRYLATERDELMUL;
1375 				if (trylater_delay > NFS_TRYLATERDELMAX)
1376 					trylater_delay = NFS_TRYLATERDELMAX;
1377 				/*
1378 				 * RFC1813:
1379 				 * The client should wait and then try
1380 				 * the request with a new RPC transaction ID.
1381 				 */
1382 				nfs_renewxid(rep);
1383 				goto tryagain;
1384 
1385 			default:
1386 #ifdef DIAGNOSTIC
1387 				printf("Invalid rpc error code %d\n", error);
1388 #endif
1389 				error = EINVAL;
1390 				break;
1391 			}
1392 
1393 			if (nmp->nm_flag & NFSMNT_NFSV3) {
1394 				*mrp = mrep;
1395 				*mdp = md;
1396 				*dposp = dpos;
1397 				error |= NFSERR_RETERR;
1398 			} else
1399 				m_freem(mrep);
1400 			goto nfsmout;
1401 		}
1402 
1403 		/*
1404 		 * note which credential worked to minimize number of retries.
1405 		 */
1406 		if (use_opencred)
1407 			np->n_flag |= NUSEOPENCRED;
1408 		else
1409 			np->n_flag &= ~NUSEOPENCRED;
1410 
1411 		*mrp = mrep;
1412 		*mdp = md;
1413 		*dposp = dpos;
1414 
1415 		KASSERT(error == 0);
1416 		goto nfsmout;
1417 	}
1418 	m_freem(mrep);
1419 	error = EPROTONOSUPPORT;
1420 nfsmout:
1421 	KASSERT(kauth_cred_getrefcnt(acred) == 1);
1422 	kauth_cred_free(acred);
1423 	m_freem(rep->r_mreq);
1424 	free((void *)rep, M_NFSREQ);
1425 	m_freem(mrest_backup);
1426 	return (error);
1427 }
1428 #endif /* NFS */
1429 
1430 /*
1431  * Generate the rpc reply header
1432  * siz arg. is used to decide if adding a cluster is worthwhile
1433  */
1434 int
1435 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1436 	int siz;
1437 	struct nfsrv_descript *nd;
1438 	struct nfssvc_sock *slp;
1439 	int err;
1440 	int cache;
1441 	u_quad_t *frev;
1442 	struct mbuf **mrq;
1443 	struct mbuf **mbp;
1444 	char **bposp;
1445 {
1446 	u_int32_t *tl;
1447 	struct mbuf *mreq;
1448 	char *bpos;
1449 	struct mbuf *mb;
1450 
1451 	mreq = m_gethdr(M_WAIT, MT_DATA);
1452 	MCLAIM(mreq, &nfs_mowner);
1453 	mb = mreq;
1454 	/*
1455 	 * If this is a big reply, use a cluster else
1456 	 * try and leave leading space for the lower level headers.
1457 	 */
1458 	siz += RPC_REPLYSIZ;
1459 	if (siz >= max_datalen) {
1460 		m_clget(mreq, M_WAIT);
1461 	} else
1462 		mreq->m_data += max_hdr;
1463 	tl = mtod(mreq, u_int32_t *);
1464 	mreq->m_len = 6 * NFSX_UNSIGNED;
1465 	bpos = ((char *)tl) + mreq->m_len;
1466 	*tl++ = txdr_unsigned(nd->nd_retxid);
1467 	*tl++ = rpc_reply;
1468 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1469 		*tl++ = rpc_msgdenied;
1470 		if (err & NFSERR_AUTHERR) {
1471 			*tl++ = rpc_autherr;
1472 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1473 			mreq->m_len -= NFSX_UNSIGNED;
1474 			bpos -= NFSX_UNSIGNED;
1475 		} else {
1476 			*tl++ = rpc_mismatch;
1477 			*tl++ = txdr_unsigned(RPC_VER2);
1478 			*tl = txdr_unsigned(RPC_VER2);
1479 		}
1480 	} else {
1481 		*tl++ = rpc_msgaccepted;
1482 
1483 		/*
1484 		 * For Kerberos authentication, we must send the nickname
1485 		 * verifier back, otherwise just RPCAUTH_NULL.
1486 		 */
1487 		if (nd->nd_flag & ND_KERBFULL) {
1488 			struct nfsuid *nuidp;
1489 			struct timeval ktvin, ktvout;
1490 
1491 			memset(&ktvout, 0, sizeof ktvout);	/* XXX gcc */
1492 
1493 			LIST_FOREACH(nuidp,
1494 			    NUIDHASH(slp, kauth_cred_geteuid(nd->nd_cr)),
1495 			    nu_hash) {
1496 				if (kauth_cred_geteuid(nuidp->nu_cr) ==
1497 				kauth_cred_geteuid(nd->nd_cr) &&
1498 				    (!nd->nd_nam2 || netaddr_match(
1499 				    NU_NETFAM(nuidp), &nuidp->nu_haddr,
1500 				    nd->nd_nam2)))
1501 					break;
1502 			}
1503 			if (nuidp) {
1504 				ktvin.tv_sec =
1505 				    txdr_unsigned(nuidp->nu_timestamp.tv_sec
1506 					- 1);
1507 				ktvin.tv_usec =
1508 				    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1509 
1510 				/*
1511 				 * Encrypt the timestamp in ecb mode using the
1512 				 * session key.
1513 				 */
1514 #ifdef NFSKERB
1515 				XXX
1516 #endif
1517 
1518 				*tl++ = rpc_auth_kerb;
1519 				*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1520 				*tl = ktvout.tv_sec;
1521 				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1522 				*tl++ = ktvout.tv_usec;
1523 				*tl++ = txdr_unsigned(
1524 				    kauth_cred_geteuid(nuidp->nu_cr));
1525 			} else {
1526 				*tl++ = 0;
1527 				*tl++ = 0;
1528 			}
1529 		} else {
1530 			*tl++ = 0;
1531 			*tl++ = 0;
1532 		}
1533 		switch (err) {
1534 		case EPROGUNAVAIL:
1535 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
1536 			break;
1537 		case EPROGMISMATCH:
1538 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
1539 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1540 			*tl++ = txdr_unsigned(2);
1541 			*tl = txdr_unsigned(3);
1542 			break;
1543 		case EPROCUNAVAIL:
1544 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
1545 			break;
1546 		case EBADRPC:
1547 			*tl = txdr_unsigned(RPC_GARBAGE);
1548 			break;
1549 		default:
1550 			*tl = 0;
1551 			if (err != NFSERR_RETVOID) {
1552 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1553 				if (err)
1554 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1555 				else
1556 				    *tl = 0;
1557 			}
1558 			break;
1559 		};
1560 	}
1561 
1562 	if (mrq != NULL)
1563 		*mrq = mreq;
1564 	*mbp = mb;
1565 	*bposp = bpos;
1566 	if (err != 0 && err != NFSERR_RETVOID)
1567 		nfsstats.srvrpc_errs++;
1568 	return (0);
1569 }
1570 
1571 static void
1572 nfs_timer_schedule(void)
1573 {
1574 
1575 	callout_schedule(&nfs_timer_ch, nfs_ticks);
1576 }
1577 
1578 void
1579 nfs_timer_start(void)
1580 {
1581 
1582 	if (callout_pending(&nfs_timer_ch))
1583 		return;
1584 
1585 	nfs_timer_start_ev.ev_count++;
1586 	nfs_timer_schedule();
1587 }
1588 
1589 void
1590 nfs_timer_init(void)
1591 {
1592 
1593 	callout_init(&nfs_timer_ch, 0);
1594 	callout_setfunc(&nfs_timer_ch, nfs_timer, NULL);
1595 	evcnt_attach_dynamic(&nfs_timer_ev, EVCNT_TYPE_MISC, NULL,
1596 	    "nfs", "timer");
1597 	evcnt_attach_dynamic(&nfs_timer_start_ev, EVCNT_TYPE_MISC, NULL,
1598 	    "nfs", "timer start");
1599 	evcnt_attach_dynamic(&nfs_timer_stop_ev, EVCNT_TYPE_MISC, NULL,
1600 	    "nfs", "timer stop");
1601 }
1602 
1603 /*
1604  * Nfs timer routine
1605  * Scan the nfsreq list and retranmit any requests that have timed out
1606  * To avoid retransmission attempts on STREAM sockets (in the future) make
1607  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1608  * A non-NULL argument means 'initialize'.
1609  */
1610 void
1611 nfs_timer(void *arg)
1612 {
1613 	struct nfsreq *rep;
1614 	struct mbuf *m;
1615 	struct socket *so;
1616 	struct nfsmount *nmp;
1617 	int timeo;
1618 	int s, error;
1619 	bool more = false;
1620 #ifdef NFSSERVER
1621 	struct timeval tv;
1622 	struct nfssvc_sock *slp;
1623 	u_quad_t cur_usec;
1624 #endif
1625 
1626 	nfs_timer_ev.ev_count++;
1627 
1628 	s = splsoftnet();
1629 	TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
1630 		more = true;
1631 		nmp = rep->r_nmp;
1632 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1633 			continue;
1634 		if (nfs_sigintr(nmp, rep, rep->r_lwp)) {
1635 			rep->r_flags |= R_SOFTTERM;
1636 			continue;
1637 		}
1638 		if (rep->r_rtt >= 0) {
1639 			rep->r_rtt++;
1640 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1641 				timeo = nmp->nm_timeo;
1642 			else
1643 				timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1644 			if (nmp->nm_timeouts > 0)
1645 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1646 			if (rep->r_rtt <= timeo)
1647 				continue;
1648 			if (nmp->nm_timeouts <
1649 			    (sizeof(nfs_backoff) / sizeof(nfs_backoff[0])))
1650 				nmp->nm_timeouts++;
1651 		}
1652 		/*
1653 		 * Check for server not responding
1654 		 */
1655 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1656 		     rep->r_rexmit > nmp->nm_deadthresh) {
1657 			nfs_msg(rep->r_lwp,
1658 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
1659 			    "not responding");
1660 			rep->r_flags |= R_TPRINTFMSG;
1661 		}
1662 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
1663 			nfsstats.rpctimeouts++;
1664 			rep->r_flags |= R_SOFTTERM;
1665 			continue;
1666 		}
1667 		if (nmp->nm_sotype != SOCK_DGRAM) {
1668 			if (++rep->r_rexmit > NFS_MAXREXMIT)
1669 				rep->r_rexmit = NFS_MAXREXMIT;
1670 			continue;
1671 		}
1672 		if ((so = nmp->nm_so) == NULL)
1673 			continue;
1674 
1675 		/*
1676 		 * If there is enough space and the window allows..
1677 		 *	Resend it
1678 		 * Set r_rtt to -1 in case we fail to send it now.
1679 		 */
1680 		rep->r_rtt = -1;
1681 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1682 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1683 		    (rep->r_flags & R_SENT) ||
1684 		    nmp->nm_sent < nmp->nm_cwnd) &&
1685 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1686 		        if (so->so_state & SS_ISCONNECTED)
1687 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1688 			    (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0);
1689 			else
1690 			    error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1691 			    nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0);
1692 			if (error) {
1693 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1694 #ifdef DEBUG
1695 					printf("nfs_timer: ignoring error %d\n",
1696 						error);
1697 #endif
1698 					so->so_error = 0;
1699 				}
1700 			} else {
1701 				/*
1702 				 * Iff first send, start timing
1703 				 * else turn timing off, backoff timer
1704 				 * and divide congestion window by 2.
1705 				 */
1706 				if (rep->r_flags & R_SENT) {
1707 					rep->r_flags &= ~R_TIMING;
1708 					if (++rep->r_rexmit > NFS_MAXREXMIT)
1709 						rep->r_rexmit = NFS_MAXREXMIT;
1710 					nmp->nm_cwnd >>= 1;
1711 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
1712 						nmp->nm_cwnd = NFS_CWNDSCALE;
1713 					nfsstats.rpcretries++;
1714 				} else {
1715 					rep->r_flags |= R_SENT;
1716 					nmp->nm_sent += NFS_CWNDSCALE;
1717 				}
1718 				rep->r_rtt = 0;
1719 			}
1720 		}
1721 	}
1722 
1723 #ifdef NFSSERVER
1724 	/*
1725 	 * Scan the write gathering queues for writes that need to be
1726 	 * completed now.
1727 	 */
1728 	getmicrotime(&tv);
1729 	cur_usec = (u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec;
1730 	TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
1731 		if (LIST_FIRST(&slp->ns_tq)) {
1732 			if (LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
1733 				nfsrv_wakenfsd(slp);
1734 			}
1735 			more = true;
1736 		}
1737 	}
1738 #endif /* NFSSERVER */
1739 	splx(s);
1740 	if (more) {
1741 		nfs_timer_schedule();
1742 	} else {
1743 		nfs_timer_stop_ev.ev_count++;
1744 	}
1745 }
1746 
1747 /*
1748  * Test for a termination condition pending on the process.
1749  * This is used for NFSMNT_INT mounts.
1750  */
1751 int
1752 nfs_sigintr(nmp, rep, l)
1753 	struct nfsmount *nmp;
1754 	struct nfsreq *rep;
1755 	struct lwp *l;
1756 {
1757 	sigset_t ss;
1758 
1759 	if (rep && (rep->r_flags & R_SOFTTERM))
1760 		return (EINTR);
1761 	if (!(nmp->nm_flag & NFSMNT_INT))
1762 		return (0);
1763 	if (l) {
1764 		sigpending1(l, &ss);
1765 #if 0
1766 		sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss);
1767 #endif
1768 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1769 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1770 		    sigismember(&ss, SIGQUIT))
1771 			return (EINTR);
1772 	}
1773 	return (0);
1774 }
1775 
1776 /*
1777  * Lock a socket against others.
1778  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1779  * and also to avoid race conditions between the processes with nfs requests
1780  * in progress when a reconnect is necessary.
1781  */
1782 static int
1783 nfs_sndlock(struct nfsmount *nmp, struct nfsreq *rep)
1784 {
1785 	struct lwp *l;
1786 	int timeo = 0;
1787 	bool catch = false;
1788 	int error = 0;
1789 
1790 	if (rep) {
1791 		l = rep->r_lwp;
1792 		if (rep->r_nmp->nm_flag & NFSMNT_INT)
1793 			catch = true;
1794 	} else
1795 		l = NULL;
1796 	mutex_enter(&nmp->nm_lock);
1797 	while ((nmp->nm_iflag & NFSMNT_SNDLOCK) != 0) {
1798 		if (rep && nfs_sigintr(rep->r_nmp, rep, l)) {
1799 			error = EINTR;
1800 			goto quit;
1801 		}
1802 		if (catch) {
1803 			cv_timedwait_sig(&nmp->nm_sndcv, &nmp->nm_lock, timeo);
1804 		} else {
1805 			cv_timedwait(&nmp->nm_sndcv, &nmp->nm_lock, timeo);
1806 		}
1807 		if (catch) {
1808 			catch = false;
1809 			timeo = 2 * hz;
1810 		}
1811 	}
1812 	nmp->nm_iflag |= NFSMNT_SNDLOCK;
1813 quit:
1814 	mutex_exit(&nmp->nm_lock);
1815 	return error;
1816 }
1817 
1818 /*
1819  * Unlock the stream socket for others.
1820  */
1821 static void
1822 nfs_sndunlock(struct nfsmount *nmp)
1823 {
1824 
1825 	mutex_enter(&nmp->nm_lock);
1826 	if ((nmp->nm_iflag & NFSMNT_SNDLOCK) == 0)
1827 		panic("nfs sndunlock");
1828 	nmp->nm_iflag &= ~NFSMNT_SNDLOCK;
1829 	cv_signal(&nmp->nm_sndcv);
1830 	mutex_exit(&nmp->nm_lock);
1831 }
1832 
1833 static int
1834 nfs_rcvlock(struct nfsmount *nmp, struct nfsreq *rep)
1835 {
1836 	int *flagp = &nmp->nm_iflag;
1837 	int slptimeo = 0;
1838 	bool catch;
1839 	int error = 0;
1840 
1841 	KASSERT(nmp == rep->r_nmp);
1842 
1843 	catch = (nmp->nm_flag & NFSMNT_INT) != 0;
1844 	mutex_enter(&nmp->nm_lock);
1845 	while (/* CONSTCOND */ true) {
1846 		if (*flagp & NFSMNT_DISMNT) {
1847 			cv_signal(&nmp->nm_disconcv);
1848 			error = EIO;
1849 			break;
1850 		}
1851 		/* If our reply was received while we were sleeping,
1852 		 * then just return without taking the lock to avoid a
1853 		 * situation where a single iod could 'capture' the
1854 		 * receive lock.
1855 		 */
1856 		if (rep->r_mrep != NULL) {
1857 			error = EALREADY;
1858 			break;
1859 		}
1860 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) {
1861 			error = EINTR;
1862 			break;
1863 		}
1864 		if ((*flagp & NFSMNT_RCVLOCK) == 0) {
1865 			*flagp |= NFSMNT_RCVLOCK;
1866 			break;
1867 		}
1868 		if (catch) {
1869 			cv_timedwait_sig(&nmp->nm_rcvcv, &nmp->nm_lock,
1870 			    slptimeo);
1871 		} else {
1872 			cv_timedwait(&nmp->nm_rcvcv, &nmp->nm_lock,
1873 			    slptimeo);
1874 		}
1875 		if (catch) {
1876 			catch = false;
1877 			slptimeo = 2 * hz;
1878 		}
1879 	}
1880 	mutex_exit(&nmp->nm_lock);
1881 	return error;
1882 }
1883 
1884 /*
1885  * Unlock the stream socket for others.
1886  */
1887 static void
1888 nfs_rcvunlock(struct nfsmount *nmp)
1889 {
1890 
1891 	mutex_enter(&nmp->nm_lock);
1892 	if ((nmp->nm_iflag & NFSMNT_RCVLOCK) == 0)
1893 		panic("nfs rcvunlock");
1894 	nmp->nm_iflag &= ~NFSMNT_RCVLOCK;
1895 	cv_broadcast(&nmp->nm_rcvcv);
1896 	mutex_exit(&nmp->nm_lock);
1897 }
1898 
1899 /*
1900  * Parse an RPC request
1901  * - verify it
1902  * - allocate and fill in the cred.
1903  */
1904 int
1905 nfs_getreq(nd, nfsd, has_header)
1906 	struct nfsrv_descript *nd;
1907 	struct nfsd *nfsd;
1908 	int has_header;
1909 {
1910 	int len, i;
1911 	u_int32_t *tl;
1912 	int32_t t1;
1913 	struct uio uio;
1914 	struct iovec iov;
1915 	char *dpos, *cp2, *cp;
1916 	u_int32_t nfsvers, auth_type;
1917 	uid_t nickuid;
1918 	int error = 0, ticklen;
1919 	struct mbuf *mrep, *md;
1920 	struct nfsuid *nuidp;
1921 	struct timeval tvin, tvout;
1922 
1923 	memset(&tvout, 0, sizeof tvout);	/* XXX gcc */
1924 
1925 	KASSERT(nd->nd_cr == NULL);
1926 	mrep = nd->nd_mrep;
1927 	md = nd->nd_md;
1928 	dpos = nd->nd_dpos;
1929 	if (has_header) {
1930 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1931 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1932 		if (*tl++ != rpc_call) {
1933 			m_freem(mrep);
1934 			return (EBADRPC);
1935 		}
1936 	} else
1937 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1938 	nd->nd_repstat = 0;
1939 	nd->nd_flag = 0;
1940 	if (*tl++ != rpc_vers) {
1941 		nd->nd_repstat = ERPCMISMATCH;
1942 		nd->nd_procnum = NFSPROC_NOOP;
1943 		return (0);
1944 	}
1945 	if (*tl != nfs_prog) {
1946 		nd->nd_repstat = EPROGUNAVAIL;
1947 		nd->nd_procnum = NFSPROC_NOOP;
1948 		return (0);
1949 	}
1950 	tl++;
1951 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1952 	if (nfsvers < NFS_VER2 || nfsvers > NFS_VER3) {
1953 		nd->nd_repstat = EPROGMISMATCH;
1954 		nd->nd_procnum = NFSPROC_NOOP;
1955 		return (0);
1956 	}
1957 	if (nfsvers == NFS_VER3)
1958 		nd->nd_flag = ND_NFSV3;
1959 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1960 	if (nd->nd_procnum == NFSPROC_NULL)
1961 		return (0);
1962 	if (nd->nd_procnum > NFSPROC_COMMIT ||
1963 	    (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1964 		nd->nd_repstat = EPROCUNAVAIL;
1965 		nd->nd_procnum = NFSPROC_NOOP;
1966 		return (0);
1967 	}
1968 	if ((nd->nd_flag & ND_NFSV3) == 0)
1969 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1970 	auth_type = *tl++;
1971 	len = fxdr_unsigned(int, *tl++);
1972 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1973 		m_freem(mrep);
1974 		return (EBADRPC);
1975 	}
1976 
1977 	nd->nd_flag &= ~ND_KERBAUTH;
1978 	/*
1979 	 * Handle auth_unix or auth_kerb.
1980 	 */
1981 	if (auth_type == rpc_auth_unix) {
1982 		uid_t uid;
1983 		gid_t gid, *grbuf;
1984 
1985 		nd->nd_cr = kauth_cred_alloc();
1986 		len = fxdr_unsigned(int, *++tl);
1987 		if (len < 0 || len > NFS_MAXNAMLEN) {
1988 			m_freem(mrep);
1989 			error = EBADRPC;
1990 			goto errout;
1991 		}
1992 		nfsm_adv(nfsm_rndup(len));
1993 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1994 
1995 		uid = fxdr_unsigned(uid_t, *tl++);
1996 		gid = fxdr_unsigned(gid_t, *tl++);
1997 		kauth_cred_setuid(nd->nd_cr, uid);
1998 		kauth_cred_seteuid(nd->nd_cr, uid);
1999 		kauth_cred_setsvuid(nd->nd_cr, uid);
2000 		kauth_cred_setgid(nd->nd_cr, gid);
2001 		kauth_cred_setegid(nd->nd_cr, gid);
2002 		kauth_cred_setsvgid(nd->nd_cr, gid);
2003 
2004 		len = fxdr_unsigned(int, *tl);
2005 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
2006 			m_freem(mrep);
2007 			error = EBADRPC;
2008 			goto errout;
2009 		}
2010 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
2011 
2012 		grbuf = malloc(len * sizeof(gid_t), M_TEMP, M_WAITOK);
2013 		for (i = 0; i < len; i++) {
2014 			if (i < NGROUPS) /* XXX elad */
2015 				grbuf[i] = fxdr_unsigned(gid_t, *tl++);
2016 			else
2017 				tl++;
2018 		}
2019 		kauth_cred_setgroups(nd->nd_cr, grbuf, min(len, NGROUPS), -1,
2020 		    UIO_SYSSPACE);
2021 		free(grbuf, M_TEMP);
2022 
2023 		len = fxdr_unsigned(int, *++tl);
2024 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
2025 			m_freem(mrep);
2026 			error = EBADRPC;
2027 			goto errout;
2028 		}
2029 		if (len > 0)
2030 			nfsm_adv(nfsm_rndup(len));
2031 	} else if (auth_type == rpc_auth_kerb) {
2032 		switch (fxdr_unsigned(int, *tl++)) {
2033 		case RPCAKN_FULLNAME:
2034 			ticklen = fxdr_unsigned(int, *tl);
2035 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
2036 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
2037 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
2038 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
2039 				m_freem(mrep);
2040 				error = EBADRPC;
2041 				goto errout;
2042 			}
2043 			uio.uio_offset = 0;
2044 			uio.uio_iov = &iov;
2045 			uio.uio_iovcnt = 1;
2046 			UIO_SETUP_SYSSPACE(&uio);
2047 			iov.iov_base = (void *)&nfsd->nfsd_authstr[4];
2048 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
2049 			nfsm_mtouio(&uio, uio.uio_resid);
2050 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2051 			if (*tl++ != rpc_auth_kerb ||
2052 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
2053 				printf("Bad kerb verifier\n");
2054 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2055 				nd->nd_procnum = NFSPROC_NOOP;
2056 				return (0);
2057 			}
2058 			nfsm_dissect(cp, void *, 4 * NFSX_UNSIGNED);
2059 			tl = (u_int32_t *)cp;
2060 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
2061 				printf("Not fullname kerb verifier\n");
2062 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2063 				nd->nd_procnum = NFSPROC_NOOP;
2064 				return (0);
2065 			}
2066 			cp += NFSX_UNSIGNED;
2067 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
2068 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
2069 			nd->nd_flag |= ND_KERBFULL;
2070 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
2071 			break;
2072 		case RPCAKN_NICKNAME:
2073 			if (len != 2 * NFSX_UNSIGNED) {
2074 				printf("Kerb nickname short\n");
2075 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
2076 				nd->nd_procnum = NFSPROC_NOOP;
2077 				return (0);
2078 			}
2079 			nickuid = fxdr_unsigned(uid_t, *tl);
2080 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2081 			if (*tl++ != rpc_auth_kerb ||
2082 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
2083 				printf("Kerb nick verifier bad\n");
2084 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2085 				nd->nd_procnum = NFSPROC_NOOP;
2086 				return (0);
2087 			}
2088 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2089 			tvin.tv_sec = *tl++;
2090 			tvin.tv_usec = *tl;
2091 
2092 			LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid),
2093 			    nu_hash) {
2094 				if (kauth_cred_geteuid(nuidp->nu_cr) == nickuid &&
2095 				    (!nd->nd_nam2 ||
2096 				     netaddr_match(NU_NETFAM(nuidp),
2097 				      &nuidp->nu_haddr, nd->nd_nam2)))
2098 					break;
2099 			}
2100 			if (!nuidp) {
2101 				nd->nd_repstat =
2102 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
2103 				nd->nd_procnum = NFSPROC_NOOP;
2104 				return (0);
2105 			}
2106 
2107 			/*
2108 			 * Now, decrypt the timestamp using the session key
2109 			 * and validate it.
2110 			 */
2111 #ifdef NFSKERB
2112 			XXX
2113 #endif
2114 
2115 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
2116 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
2117 			if (nuidp->nu_expire < time_second ||
2118 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
2119 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
2120 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
2121 				nuidp->nu_expire = 0;
2122 				nd->nd_repstat =
2123 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
2124 				nd->nd_procnum = NFSPROC_NOOP;
2125 				return (0);
2126 			}
2127 			kauth_cred_hold(nuidp->nu_cr);
2128 			nd->nd_cr = nuidp->nu_cr;
2129 			nd->nd_flag |= ND_KERBNICK;
2130 		}
2131 	} else {
2132 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
2133 		nd->nd_procnum = NFSPROC_NOOP;
2134 		return (0);
2135 	}
2136 
2137 	nd->nd_md = md;
2138 	nd->nd_dpos = dpos;
2139 	KASSERT((nd->nd_cr == NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) != 0)
2140 	     || (nd->nd_cr != NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) == 0));
2141 	return (0);
2142 nfsmout:
2143 errout:
2144 	KASSERT(error != 0);
2145 	if (nd->nd_cr != NULL) {
2146 		kauth_cred_free(nd->nd_cr);
2147 		nd->nd_cr = NULL;
2148 	}
2149 	return (error);
2150 }
2151 
2152 int
2153 nfs_msg(l, server, msg)
2154 	struct lwp *l;
2155 	const char *server, *msg;
2156 {
2157 	tpr_t tpr;
2158 
2159 	if (l)
2160 		tpr = tprintf_open(l->l_proc);
2161 	else
2162 		tpr = NULL;
2163 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
2164 	tprintf_close(tpr);
2165 	return (0);
2166 }
2167 
2168 #ifdef NFSSERVER
2169 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
2170 				    struct nfssvc_sock *, struct lwp *,
2171 				    struct mbuf **)) = {
2172 	nfsrv_null,
2173 	nfsrv_getattr,
2174 	nfsrv_setattr,
2175 	nfsrv_lookup,
2176 	nfsrv3_access,
2177 	nfsrv_readlink,
2178 	nfsrv_read,
2179 	nfsrv_write,
2180 	nfsrv_create,
2181 	nfsrv_mkdir,
2182 	nfsrv_symlink,
2183 	nfsrv_mknod,
2184 	nfsrv_remove,
2185 	nfsrv_rmdir,
2186 	nfsrv_rename,
2187 	nfsrv_link,
2188 	nfsrv_readdir,
2189 	nfsrv_readdirplus,
2190 	nfsrv_statfs,
2191 	nfsrv_fsinfo,
2192 	nfsrv_pathconf,
2193 	nfsrv_commit,
2194 	nfsrv_noop
2195 };
2196 
2197 /*
2198  * Socket upcall routine for the nfsd sockets.
2199  * The void *arg is a pointer to the "struct nfssvc_sock".
2200  * Essentially do as much as possible non-blocking, else punt and it will
2201  * be called with M_WAIT from an nfsd.
2202  */
2203 void
2204 nfsrv_rcv(so, arg, waitflag)
2205 	struct socket *so;
2206 	void *arg;
2207 	int waitflag;
2208 {
2209 	struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
2210 	struct mbuf *m;
2211 	struct mbuf *mp, *nam;
2212 	struct uio auio;
2213 	int flags, error;
2214 	int setflags = 0;
2215 
2216 	error = nfsdsock_lock(slp, (waitflag != M_DONTWAIT));
2217 	if (error) {
2218 		setflags |= SLP_NEEDQ;
2219 		goto dorecs_unlocked;
2220 	}
2221 
2222 	KASSERT(so == slp->ns_so);
2223 #define NFS_TEST_HEAVY
2224 #ifdef NFS_TEST_HEAVY
2225 	/*
2226 	 * Define this to test for nfsds handling this under heavy load.
2227 	 *
2228 	 * XXX it isn't safe to call so_receive from so_upcall context.
2229 	 */
2230 	if (waitflag == M_DONTWAIT) {
2231 		setflags |= SLP_NEEDQ;
2232 		goto dorecs;
2233 	}
2234 #endif
2235 	mutex_enter(&slp->ns_lock);
2236 	slp->ns_flag &= ~SLP_NEEDQ;
2237 	mutex_exit(&slp->ns_lock);
2238 	if (so->so_type == SOCK_STREAM) {
2239 #ifndef NFS_TEST_HEAVY
2240 		/*
2241 		 * If there are already records on the queue, defer soreceive()
2242 		 * to an nfsd so that there is feedback to the TCP layer that
2243 		 * the nfs servers are heavily loaded.
2244 		 */
2245 		if (slp->ns_rec && waitflag == M_DONTWAIT) {
2246 			setflags |= SLP_NEEDQ;
2247 			goto dorecs;
2248 		}
2249 #endif
2250 
2251 		/*
2252 		 * Do soreceive().
2253 		 */
2254 		auio.uio_resid = 1000000000;
2255 		/* not need to setup uio_vmspace */
2256 		flags = MSG_DONTWAIT;
2257 		error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, &flags);
2258 		if (error || mp == NULL) {
2259 			if (error == EWOULDBLOCK)
2260 				setflags |= SLP_NEEDQ;
2261 			else
2262 				setflags |= SLP_DISCONN;
2263 			goto dorecs;
2264 		}
2265 		m = mp;
2266 		m_claimm(m, &nfs_mowner);
2267 		if (slp->ns_rawend) {
2268 			slp->ns_rawend->m_next = m;
2269 			slp->ns_cc += 1000000000 - auio.uio_resid;
2270 		} else {
2271 			slp->ns_raw = m;
2272 			slp->ns_cc = 1000000000 - auio.uio_resid;
2273 		}
2274 		while (m->m_next)
2275 			m = m->m_next;
2276 		slp->ns_rawend = m;
2277 
2278 		/*
2279 		 * Now try and parse record(s) out of the raw stream data.
2280 		 */
2281 		error = nfsrv_getstream(slp, waitflag);
2282 		if (error) {
2283 			if (error == EPERM)
2284 				setflags |= SLP_DISCONN;
2285 			else
2286 				setflags |= SLP_NEEDQ;
2287 		}
2288 	} else {
2289 		do {
2290 			auio.uio_resid = 1000000000;
2291 			/* not need to setup uio_vmspace */
2292 			flags = MSG_DONTWAIT;
2293 			error = (*so->so_receive)(so, &nam, &auio, &mp, NULL,
2294 			    &flags);
2295 			if (mp) {
2296 				if (nam) {
2297 					m = nam;
2298 					m->m_next = mp;
2299 				} else
2300 					m = mp;
2301 				m_claimm(m, &nfs_mowner);
2302 				if (slp->ns_recend)
2303 					slp->ns_recend->m_nextpkt = m;
2304 				else
2305 					slp->ns_rec = m;
2306 				slp->ns_recend = m;
2307 				m->m_nextpkt = (struct mbuf *)0;
2308 			}
2309 			if (error) {
2310 				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2311 				    && error != EWOULDBLOCK) {
2312 					setflags |= SLP_DISCONN;
2313 					goto dorecs;
2314 				}
2315 			}
2316 		} while (mp);
2317 	}
2318 dorecs:
2319 	nfsdsock_unlock(slp);
2320 
2321 dorecs_unlocked:
2322 	/*
2323 	 * Now try and process the request records, non-blocking.
2324 	 */
2325 	if (setflags) {
2326 		mutex_enter(&slp->ns_lock);
2327 		slp->ns_flag |= setflags;
2328 		mutex_exit(&slp->ns_lock);
2329 	}
2330 	if (waitflag == M_DONTWAIT &&
2331 	    (slp->ns_rec || (slp->ns_flag & (SLP_DISCONN | SLP_NEEDQ)) != 0)) {
2332 		nfsrv_wakenfsd(slp);
2333 	}
2334 }
2335 
2336 int
2337 nfsdsock_lock(struct nfssvc_sock *slp, bool waitok)
2338 {
2339 
2340 	mutex_enter(&slp->ns_lock);
2341 	while ((~slp->ns_flag & (SLP_BUSY|SLP_VALID)) == 0) {
2342 		if (!waitok) {
2343 			mutex_exit(&slp->ns_lock);
2344 			return EWOULDBLOCK;
2345 		}
2346 		cv_wait(&slp->ns_cv, &slp->ns_lock);
2347 	}
2348 	if ((slp->ns_flag & SLP_VALID) == 0) {
2349 		mutex_exit(&slp->ns_lock);
2350 		return EINVAL;
2351 	}
2352 	KASSERT((slp->ns_flag & SLP_BUSY) == 0);
2353 	slp->ns_flag |= SLP_BUSY;
2354 	mutex_exit(&slp->ns_lock);
2355 
2356 	return 0;
2357 }
2358 
2359 void
2360 nfsdsock_unlock(struct nfssvc_sock *slp)
2361 {
2362 
2363 	mutex_enter(&slp->ns_lock);
2364 	KASSERT((slp->ns_flag & SLP_BUSY) != 0);
2365 	KASSERT((slp->ns_flag & SLP_VALID) != 0);
2366 	cv_broadcast(&slp->ns_cv);
2367 	slp->ns_flag &= ~SLP_BUSY;
2368 	mutex_exit(&slp->ns_lock);
2369 }
2370 
2371 int
2372 nfsdsock_drain(struct nfssvc_sock *slp)
2373 {
2374 	int error = 0;
2375 
2376 	mutex_enter(&slp->ns_lock);
2377 	if ((slp->ns_flag & SLP_VALID) == 0) {
2378 		error = EINVAL;
2379 		goto done;
2380 	}
2381 	slp->ns_flag &= ~SLP_VALID;
2382 	while ((slp->ns_flag & SLP_BUSY) != 0) {
2383 		cv_wait(&slp->ns_cv, &slp->ns_lock);
2384 	}
2385 done:
2386 	mutex_exit(&slp->ns_lock);
2387 
2388 	return error;
2389 }
2390 
2391 /*
2392  * Try and extract an RPC request from the mbuf data list received on a
2393  * stream socket. The "waitflag" argument indicates whether or not it
2394  * can sleep.
2395  */
2396 int
2397 nfsrv_getstream(slp, waitflag)
2398 	struct nfssvc_sock *slp;
2399 	int waitflag;
2400 {
2401 	struct mbuf *m, **mpp;
2402 	struct mbuf *recm;
2403 	u_int32_t recmark;
2404 	int error = 0;
2405 
2406 	KASSERT((slp->ns_flag & SLP_BUSY) != 0);
2407 	for (;;) {
2408 		if (slp->ns_reclen == 0) {
2409 			if (slp->ns_cc < NFSX_UNSIGNED) {
2410 				break;
2411 			}
2412 			m = slp->ns_raw;
2413 			m_copydata(m, 0, NFSX_UNSIGNED, (void *)&recmark);
2414 			m_adj(m, NFSX_UNSIGNED);
2415 			slp->ns_cc -= NFSX_UNSIGNED;
2416 			recmark = ntohl(recmark);
2417 			slp->ns_reclen = recmark & ~0x80000000;
2418 			if (recmark & 0x80000000)
2419 				slp->ns_flag |= SLP_LASTFRAG;
2420 			else
2421 				slp->ns_flag &= ~SLP_LASTFRAG;
2422 			if (slp->ns_reclen > NFS_MAXPACKET) {
2423 				error = EPERM;
2424 				break;
2425 			}
2426 		}
2427 
2428 		/*
2429 		 * Now get the record part.
2430 		 *
2431 		 * Note that slp->ns_reclen may be 0.  Linux sometimes
2432 		 * generates 0-length records.
2433 		 */
2434 		if (slp->ns_cc == slp->ns_reclen) {
2435 			recm = slp->ns_raw;
2436 			slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2437 			slp->ns_cc = slp->ns_reclen = 0;
2438 		} else if (slp->ns_cc > slp->ns_reclen) {
2439 			recm = slp->ns_raw;
2440 			m = m_split(recm, slp->ns_reclen, waitflag);
2441 			if (m == NULL) {
2442 				error = EWOULDBLOCK;
2443 				break;
2444 			}
2445 			m_claimm(recm, &nfs_mowner);
2446 			slp->ns_raw = m;
2447 			if (m->m_next == NULL)
2448 				slp->ns_rawend = m;
2449 			slp->ns_cc -= slp->ns_reclen;
2450 			slp->ns_reclen = 0;
2451 		} else {
2452 			break;
2453 		}
2454 
2455 		/*
2456 		 * Accumulate the fragments into a record.
2457 		 */
2458 		mpp = &slp->ns_frag;
2459 		while (*mpp)
2460 			mpp = &((*mpp)->m_next);
2461 		*mpp = recm;
2462 		if (slp->ns_flag & SLP_LASTFRAG) {
2463 			if (slp->ns_recend)
2464 				slp->ns_recend->m_nextpkt = slp->ns_frag;
2465 			else
2466 				slp->ns_rec = slp->ns_frag;
2467 			slp->ns_recend = slp->ns_frag;
2468 			slp->ns_frag = (struct mbuf *)0;
2469 		}
2470 	}
2471 
2472 	return error;
2473 }
2474 
2475 /*
2476  * Parse an RPC header.
2477  */
2478 int
2479 nfsrv_dorec(slp, nfsd, ndp)
2480 	struct nfssvc_sock *slp;
2481 	struct nfsd *nfsd;
2482 	struct nfsrv_descript **ndp;
2483 {
2484 	struct mbuf *m, *nam;
2485 	struct nfsrv_descript *nd;
2486 	int error;
2487 
2488 	*ndp = NULL;
2489 
2490 	if (nfsdsock_lock(slp, true)) {
2491 		return ENOBUFS;
2492 	}
2493 	m = slp->ns_rec;
2494 	if (m == NULL) {
2495 		nfsdsock_unlock(slp);
2496 		return ENOBUFS;
2497 	}
2498 	slp->ns_rec = m->m_nextpkt;
2499 	if (slp->ns_rec)
2500 		m->m_nextpkt = NULL;
2501 	else
2502 		slp->ns_recend = NULL;
2503 	nfsdsock_unlock(slp);
2504 
2505 	if (m->m_type == MT_SONAME) {
2506 		nam = m;
2507 		m = m->m_next;
2508 		nam->m_next = NULL;
2509 	} else
2510 		nam = NULL;
2511 	nd = nfsdreq_alloc();
2512 	nd->nd_md = nd->nd_mrep = m;
2513 	nd->nd_nam2 = nam;
2514 	nd->nd_dpos = mtod(m, void *);
2515 	error = nfs_getreq(nd, nfsd, true);
2516 	if (error) {
2517 		m_freem(nam);
2518 		nfsdreq_free(nd);
2519 		return (error);
2520 	}
2521 	*ndp = nd;
2522 	nfsd->nfsd_nd = nd;
2523 	return (0);
2524 }
2525 
2526 /*
2527  * Search for a sleeping nfsd and wake it up.
2528  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2529  * running nfsds will go look for the work in the nfssvc_sock list.
2530  */
2531 void
2532 nfsrv_wakenfsd(slp)
2533 	struct nfssvc_sock *slp;
2534 {
2535 	struct nfsd *nd;
2536 
2537 	if ((slp->ns_flag & SLP_VALID) == 0)
2538 		return;
2539 	mutex_enter(&nfsd_lock);
2540 	if (slp->ns_flag & SLP_DOREC) {
2541 		mutex_exit(&nfsd_lock);
2542 		return;
2543 	}
2544 	nd = SLIST_FIRST(&nfsd_idle_head);
2545 	if (nd) {
2546 		SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle);
2547 		if (nd->nfsd_slp)
2548 			panic("nfsd wakeup");
2549 		slp->ns_sref++;
2550 		KASSERT(slp->ns_sref > 0);
2551 		nd->nfsd_slp = slp;
2552 		cv_signal(&nd->nfsd_cv);
2553 	} else {
2554 		slp->ns_flag |= SLP_DOREC;
2555 		nfsd_head_flag |= NFSD_CHECKSLP;
2556 		TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending);
2557 	}
2558 	mutex_exit(&nfsd_lock);
2559 }
2560 
2561 int
2562 nfsdsock_sendreply(struct nfssvc_sock *slp, struct nfsrv_descript *nd)
2563 {
2564 	int error;
2565 
2566 	if (nd->nd_mrep != NULL) {
2567 		m_freem(nd->nd_mrep);
2568 		nd->nd_mrep = NULL;
2569 	}
2570 
2571 	mutex_enter(&slp->ns_lock);
2572 	if ((slp->ns_flag & SLP_SENDING) != 0) {
2573 		SIMPLEQ_INSERT_TAIL(&slp->ns_sendq, nd, nd_sendq);
2574 		mutex_exit(&slp->ns_lock);
2575 		return 0;
2576 	}
2577 	KASSERT(SIMPLEQ_EMPTY(&slp->ns_sendq));
2578 	slp->ns_flag |= SLP_SENDING;
2579 	mutex_exit(&slp->ns_lock);
2580 
2581 again:
2582 	error = nfs_send(slp->ns_so, nd->nd_nam2, nd->nd_mreq, NULL, curlwp);
2583 	if (nd->nd_nam2) {
2584 		m_free(nd->nd_nam2);
2585 	}
2586 	nfsdreq_free(nd);
2587 
2588 	mutex_enter(&slp->ns_lock);
2589 	KASSERT((slp->ns_flag & SLP_SENDING) != 0);
2590 	nd = SIMPLEQ_FIRST(&slp->ns_sendq);
2591 	if (nd != NULL) {
2592 		SIMPLEQ_REMOVE_HEAD(&slp->ns_sendq, nd_sendq);
2593 		mutex_exit(&slp->ns_lock);
2594 		goto again;
2595 	}
2596 	slp->ns_flag &= ~SLP_SENDING;
2597 	mutex_exit(&slp->ns_lock);
2598 
2599 	return error;
2600 }
2601 #endif /* NFSSERVER */
2602 
2603 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY))
2604 static struct pool nfs_srvdesc_pool;
2605 
2606 void
2607 nfsdreq_init(void)
2608 {
2609 
2610 	pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript),
2611 	    0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr, IPL_NONE);
2612 }
2613 
2614 struct nfsrv_descript *
2615 nfsdreq_alloc(void)
2616 {
2617 	struct nfsrv_descript *nd;
2618 
2619 	nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK);
2620 	nd->nd_cr = NULL;
2621 	return nd;
2622 }
2623 
2624 void
2625 nfsdreq_free(struct nfsrv_descript *nd)
2626 {
2627 	kauth_cred_t cr;
2628 
2629 	cr = nd->nd_cr;
2630 	if (cr != NULL) {
2631 		kauth_cred_free(cr);
2632 	}
2633 	pool_put(&nfs_srvdesc_pool, nd);
2634 }
2635 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */
2636