xref: /netbsd/sys/nfs/nfs_socket.c (revision 534b0ad8)
1 /*	$NetBSD: nfs_socket.c,v 1.201 2023/04/09 12:33:58 riastradh 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 1.201 2023/04/09 12:33:58 riastradh Exp $");
43 
44 #ifdef _KERNEL_OPT
45 #include "opt_nfs.h"
46 #include "opt_mbuftrace.h"
47 #endif
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/evcnt.h>
52 #include <sys/callout.h>
53 #include <sys/proc.h>
54 #include <sys/mount.h>
55 #include <sys/kernel.h>
56 #include <sys/kmem.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 #include <sys/time.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[nfs_proct[(r)->r_procnum] - 1]
106 #define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[nfs_proct[(r)->r_procnum] - 1]
107 
108 /*
109  * Defines which timer to use for the procnum.
110  * 0 - default
111  * 1 - getattr
112  * 2 - lookup
113  * 3 - read
114  * 4 - write
115  */
116 const int nfs_proct[NFS_NPROCS] = {
117 	[NFSPROC_NULL] = 0,
118 	[NFSPROC_GETATTR] = 1,
119 	[NFSPROC_SETATTR] = 0,
120 	[NFSPROC_LOOKUP] = 2,
121 	[NFSPROC_ACCESS] = 1,
122 	[NFSPROC_READLINK] = 3,
123 	[NFSPROC_READ] = 3,
124 	[NFSPROC_WRITE] = 4,
125 	[NFSPROC_CREATE] = 0,
126 	[NFSPROC_MKDIR] = 0,
127 	[NFSPROC_SYMLINK] = 0,
128 	[NFSPROC_MKNOD] = 0,
129 	[NFSPROC_REMOVE] = 0,
130 	[NFSPROC_RMDIR] = 0,
131 	[NFSPROC_RENAME] = 0,
132 	[NFSPROC_LINK] = 0,
133 	[NFSPROC_READDIR] = 3,
134 	[NFSPROC_READDIRPLUS] = 3,
135 	[NFSPROC_FSSTAT] = 0,
136 	[NFSPROC_FSINFO] = 0,
137 	[NFSPROC_PATHCONF] = 0,
138 	[NFSPROC_COMMIT] = 0,
139 	[NFSPROC_NOOP] = 0,
140 };
141 
142 #ifdef DEBUG
143 /*
144  * Avoid spamming the console with debugging messages.  We only print
145  * the nfs timer and reply error debugs every 10 seconds.
146  */
147 const struct timeval nfs_err_interval = { 10, 0 };
148 struct timeval nfs_reply_last_err_time;
149 struct timeval nfs_timer_last_err_time;
150 #endif
151 
152 /*
153  * There is a congestion window for outstanding rpcs maintained per mount
154  * point. The cwnd size is adjusted in roughly the way that:
155  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
156  * SIGCOMM '88". ACM, August 1988.
157  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
158  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
159  * of rpcs is in progress.
160  * (The sent count and cwnd are scaled for integer arith.)
161  * Variants of "slow start" were tried and were found to be too much of a
162  * performance hit (ave. rtt 3 times larger),
163  * I suspect due to the large rtt that nfs rpcs have.
164  */
165 int nfsrtton = 0;
166 struct nfsrtt nfsrtt;
167 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
168 struct nfsreqhead nfs_reqq;
169 kmutex_t nfs_reqq_lock;
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 static kmutex_t nfs_timer_lock;
175 static bool (*nfs_timer_srvvec)(void);
176 
177 /*
178  * Initialize sockets and congestion for a new NFS connection.
179  * We do not free the sockaddr if error.
180  */
181 int
nfs_connect(struct nfsmount * nmp,struct nfsreq * rep,struct lwp * l)182 nfs_connect(struct nfsmount *nmp, struct nfsreq *rep, struct lwp *l)
183 {
184 	struct socket *so;
185 	int error, rcvreserve, sndreserve;
186 	struct sockaddr *saddr;
187 	struct sockaddr_in sin;
188 	struct sockaddr_in6 sin6;
189 	int val;
190 
191 	nmp->nm_so = NULL;
192 	saddr = mtod(nmp->nm_nam, struct sockaddr *);
193 	error = socreate(saddr->sa_family, &nmp->nm_so,
194 		nmp->nm_sotype, nmp->nm_soproto, l, NULL);
195 	if (error)
196 		goto bad;
197 	so = nmp->nm_so;
198 #ifdef MBUFTRACE
199 	so->so_mowner = &nfs_mowner;
200 	so->so_rcv.sb_mowner = &nfs_mowner;
201 	so->so_snd.sb_mowner = &nfs_mowner;
202 #endif
203 	nmp->nm_soflags = so->so_proto->pr_flags;
204 
205 	/*
206 	 * Some servers require that the client port be a reserved port number.
207 	 */
208 	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
209 		val = IP_PORTRANGE_LOW;
210 
211 		if ((error = so_setsockopt(NULL, so, IPPROTO_IP, IP_PORTRANGE,
212 		    &val, sizeof(val))))
213 			goto bad;
214 		sin.sin_len = sizeof(struct sockaddr_in);
215 		sin.sin_family = AF_INET;
216 		sin.sin_addr.s_addr = INADDR_ANY;
217 		sin.sin_port = 0;
218 		error = sobind(so, (struct sockaddr *)&sin, &lwp0);
219 		if (error)
220 			goto bad;
221 	}
222 	if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) {
223 		val = IPV6_PORTRANGE_LOW;
224 
225 		if ((error = so_setsockopt(NULL, so, IPPROTO_IPV6,
226 		    IPV6_PORTRANGE, &val, sizeof(val))))
227 			goto bad;
228 		memset(&sin6, 0, sizeof(sin6));
229 		sin6.sin6_len = sizeof(struct sockaddr_in6);
230 		sin6.sin6_family = AF_INET6;
231 		error = sobind(so, (struct sockaddr *)&sin6, &lwp0);
232 		if (error)
233 			goto bad;
234 	}
235 
236 	/*
237 	 * Protocols that do not require connections may be optionally left
238 	 * unconnected for servers that reply from a port other than NFS_PORT.
239 	 */
240 	solock(so);
241 	if (nmp->nm_flag & NFSMNT_NOCONN) {
242 		if (nmp->nm_soflags & PR_CONNREQUIRED) {
243 			sounlock(so);
244 			error = ENOTCONN;
245 			goto bad;
246 		}
247 	} else {
248 		error = soconnect(so, mtod(nmp->nm_nam, struct sockaddr *), l);
249 		if (error) {
250 			sounlock(so);
251 			goto bad;
252 		}
253 
254 		/*
255 		 * Wait for the connection to complete. Cribbed from the
256 		 * connect system call but with the wait timing out so
257 		 * that interruptible mounts don't hang here for a long time.
258 		 */
259 		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
260 			(void)sowait(so, false, 2 * hz);
261 			if ((so->so_state & SS_ISCONNECTING) &&
262 			    so->so_error == 0 && rep &&
263 			    (error = nfs_sigintr(nmp, rep, rep->r_lwp)) != 0){
264 				so->so_state &= ~SS_ISCONNECTING;
265 				sounlock(so);
266 				goto bad;
267 			}
268 		}
269 		if (so->so_error) {
270 			error = so->so_error;
271 			so->so_error = 0;
272 			sounlock(so);
273 			goto bad;
274 		}
275 	}
276 	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
277 		so->so_rcv.sb_timeo = (5 * hz);
278 		so->so_snd.sb_timeo = (5 * hz);
279 	} else {
280 		/*
281 		 * enable receive timeout to detect server crash and reconnect.
282 		 * otherwise, we can be stuck in soreceive forever.
283 		 */
284 		so->so_rcv.sb_timeo = (5 * hz);
285 		so->so_snd.sb_timeo = 0;
286 	}
287 	if (nmp->nm_sotype == SOCK_DGRAM) {
288 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 3;
289 		rcvreserve = (uimax(nmp->nm_rsize, nmp->nm_readdirsize) +
290 		    NFS_MAXPKTHDR) * 2;
291 	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
292 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 3;
293 		rcvreserve = (uimax(nmp->nm_rsize, nmp->nm_readdirsize) +
294 		    NFS_MAXPKTHDR) * 3;
295 	} else {
296 		sounlock(so);
297 		if (nmp->nm_sotype != SOCK_STREAM)
298 			panic("nfscon sotype");
299 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
300 			val = 1;
301 			so_setsockopt(NULL, so, SOL_SOCKET, SO_KEEPALIVE, &val,
302 			    sizeof(val));
303 		}
304 		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
305 			val = 1;
306 			so_setsockopt(NULL, so, IPPROTO_TCP, TCP_NODELAY, &val,
307 			    sizeof(val));
308 		}
309 		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
310 		    sizeof (u_int32_t)) * 3;
311 		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
312 		    sizeof (u_int32_t)) * 3;
313 		solock(so);
314 	}
315 	error = soreserve(so, sndreserve, rcvreserve);
316 	if (error) {
317 		sounlock(so);
318 		goto bad;
319 	}
320 	so->so_rcv.sb_flags |= SB_NOINTR;
321 	so->so_snd.sb_flags |= SB_NOINTR;
322 	sounlock(so);
323 
324 	/* Initialize other non-zero congestion variables */
325 	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
326 		NFS_TIMEO << 3;
327 	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
328 		nmp->nm_sdrtt[3] = 0;
329 	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
330 	nmp->nm_sent = 0;
331 	nmp->nm_timeouts = 0;
332 	return (0);
333 
334 bad:
335 	nfs_disconnect(nmp);
336 	return (error);
337 }
338 
339 /*
340  * Reconnect routine:
341  * Called when a connection is broken on a reliable protocol.
342  * - clean up the old socket
343  * - nfs_connect() again
344  * - set R_MUSTRESEND for all outstanding requests on mount point
345  * If this fails the mount point is DEAD!
346  * nb: Must be called with the nfs_sndlock() set on the mount point.
347  */
348 int
nfs_reconnect(struct nfsreq * rep)349 nfs_reconnect(struct nfsreq *rep)
350 {
351 	struct nfsreq *rp;
352 	struct nfsmount *nmp = rep->r_nmp;
353 	int error, s;
354 	time_t before_ts;
355 
356 	nfs_disconnect(nmp);
357 
358 	/*
359 	 * Force unmount: do not try to reconnect
360 	 */
361 	if (nmp->nm_iflag & NFSMNT_DISMNTFORCE)
362 		return EIO;
363 
364 	before_ts = time_uptime;
365 	while ((error = nfs_connect(nmp, rep, &lwp0)) != 0) {
366 		if (error == EINTR || error == ERESTART)
367 			return (EINTR);
368 
369 		if (rep->r_flags & R_SOFTTERM)
370 			return (EIO);
371 
372 		/*
373 		 * Soft mount can fail here, but not too fast:
374 		 * we want to make sure we at least honoured
375 		 * NFS timeout.
376 		 */
377 		if ((nmp->nm_flag & NFSMNT_SOFT) &&
378 		    (time_uptime - before_ts > nmp->nm_timeo / NFS_HZ))
379 			return (EIO);
380 
381 		kpause("nfscn2", false, hz, NULL);
382 	}
383 
384 	/*
385 	 * Loop through outstanding request list and fix up all requests
386 	 * on old socket.
387 	 */
388 	s = splsoftnet();
389 	mutex_enter(&nfs_reqq_lock);
390 	TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
391 		if (rp->r_nmp == nmp) {
392 			if ((rp->r_flags & R_MUSTRESEND) == 0)
393 				rp->r_flags |= R_MUSTRESEND | R_REXMITTED;
394 			rp->r_rexmit = 0;
395 		}
396 	}
397 	mutex_exit(&nfs_reqq_lock);
398 	splx(s);
399 	return (0);
400 }
401 
402 /*
403  * NFS disconnect. Clean up and unlink.
404  */
405 void
nfs_disconnect(struct nfsmount * nmp)406 nfs_disconnect(struct nfsmount *nmp)
407 {
408 	struct socket *so;
409 	int drain = 0;
410 
411 	if (nmp->nm_so) {
412 		so = nmp->nm_so;
413 		nmp->nm_so = NULL;
414 		solock(so);
415 		soshutdown(so, SHUT_RDWR);
416 		sounlock(so);
417 		drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
418 		if (drain) {
419 			/*
420 			 * soshutdown() above should wake up the current
421 			 * listener.
422 			 * Now wake up those waiting for the receive lock, and
423 			 * wait for them to go away unhappy, to prevent *nmp
424 			 * from evaporating while they're sleeping.
425 			 */
426 			mutex_enter(&nmp->nm_lock);
427 			while (nmp->nm_waiters > 0) {
428 				cv_broadcast(&nmp->nm_rcvcv);
429 				cv_broadcast(&nmp->nm_sndcv);
430 				cv_wait(&nmp->nm_disconcv, &nmp->nm_lock);
431 			}
432 			mutex_exit(&nmp->nm_lock);
433 		}
434 		soclose(so);
435 	}
436 #ifdef DIAGNOSTIC
437 	if (drain && (nmp->nm_waiters > 0))
438 		panic("nfs_disconnect: waiters left after drain?");
439 #endif
440 }
441 
442 void
nfs_safedisconnect(struct nfsmount * nmp)443 nfs_safedisconnect(struct nfsmount *nmp)
444 {
445 	struct nfsreq dummyreq;
446 
447 	memset(&dummyreq, 0, sizeof(dummyreq));
448 	dummyreq.r_nmp = nmp;
449 	nfs_rcvlock(nmp, &dummyreq); /* XXX ignored error return */
450 	nfs_disconnect(nmp);
451 	nfs_rcvunlock(nmp);
452 }
453 
454 /*
455  * This is the nfs send routine. For connection based socket types, it
456  * must be called with an nfs_sndlock() on the socket.
457  * "rep == NULL" indicates that it has been called from a server.
458  * For the client side:
459  * - return EINTR if the RPC is terminated, 0 otherwise
460  * - set R_MUSTRESEND if the send fails for any reason
461  * - do any cleanup required by recoverable socket errors (? ? ?)
462  * For the server side:
463  * - return EINTR or ERESTART if interrupted by a signal
464  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
465  * - do any cleanup required by recoverable socket errors (? ? ?)
466  */
467 int
nfs_send(struct socket * so,struct mbuf * nam,struct mbuf * top,struct nfsreq * rep,struct lwp * l)468 nfs_send(struct socket *so, struct mbuf *nam, struct mbuf *top, struct nfsreq *rep, struct lwp *l)
469 {
470 	struct sockaddr *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 = NULL;
493 	else
494 		sendnam = mtod(nam, struct sockaddr *);
495 	if (so->so_type == SOCK_SEQPACKET)
496 		flags = MSG_EOR;
497 	else
498 		flags = 0;
499 
500 	error = (*so->so_send)(so, sendnam, NULL, top, NULL, flags,  l);
501 	if (error) {
502 		if (rep) {
503 			if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
504 				/*
505 				 * We're too fast for the network/driver,
506 				 * and UDP isn't flowcontrolled.
507 				 * We need to resend. This is not fatal,
508 				 * just try again.
509 				 *
510 				 * Could be smarter here by doing some sort
511 				 * of a backoff, but this is rare.
512 				 */
513 				rep->r_flags |= R_MUSTRESEND;
514 			} else {
515 				if (error != EPIPE)
516 					log(LOG_INFO,
517 					    "nfs send error %d for %s\n",
518 					    error,
519 					    rep->r_nmp->nm_mountp->
520 						    mnt_stat.f_mntfromname);
521 				/*
522 				 * Deal with errors for the client side.
523 				 */
524 				if (rep->r_flags & R_SOFTTERM)
525 					error = EINTR;
526 				else if (error != EMSGSIZE)
527 					rep->r_flags |= R_MUSTRESEND;
528 			}
529 		} else {
530 			/*
531 			 * See above. This error can happen under normal
532 			 * circumstances and the log is too noisy.
533 			 * The error will still show up in nfsstat.
534 			 */
535 			if (error != ENOBUFS || so->so_type != SOCK_DGRAM)
536 				log(LOG_INFO, "nfsd send error %d\n", error);
537 		}
538 
539 		/*
540 		 * Handle any recoverable (soft) socket errors here. (? ? ?)
541 		 */
542 		if (error != EINTR && error != ERESTART &&
543 		    error != EWOULDBLOCK && error != EPIPE &&
544 		    error != EMSGSIZE)
545 			error = 0;
546 	}
547 	return (error);
548 }
549 
550 /*
551  * Generate the rpc reply header
552  * siz arg. is used to decide if adding a cluster is worthwhile
553  */
554 int
nfs_rephead(int siz,struct nfsrv_descript * nd,struct nfssvc_sock * slp,int err,int cache,u_quad_t * frev,struct mbuf ** mrq,struct mbuf ** mbp,char ** bposp)555 nfs_rephead(int siz, struct nfsrv_descript *nd, struct nfssvc_sock *slp, int err, int cache, u_quad_t *frev, struct mbuf **mrq, struct mbuf **mbp, char **bposp)
556 {
557 	u_int32_t *tl;
558 	struct mbuf *mreq;
559 	char *bpos;
560 	struct mbuf *mb;
561 
562 	mreq = m_gethdr(M_WAIT, MT_DATA);
563 	MCLAIM(mreq, &nfs_mowner);
564 	mb = mreq;
565 	/*
566 	 * If this is a big reply, use a cluster else
567 	 * try and leave leading space for the lower level headers.
568 	 */
569 	siz += RPC_REPLYSIZ;
570 	if (siz >= max_datalen) {
571 		m_clget(mreq, M_WAIT);
572 	} else
573 		mreq->m_data += max_hdr;
574 	tl = mtod(mreq, u_int32_t *);
575 	mreq->m_len = 6 * NFSX_UNSIGNED;
576 	bpos = ((char *)tl) + mreq->m_len;
577 	*tl++ = txdr_unsigned(nd->nd_retxid);
578 	*tl++ = rpc_reply;
579 	if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
580 		*tl++ = rpc_msgdenied;
581 		if (err & NFSERR_AUTHERR) {
582 			*tl++ = rpc_autherr;
583 			*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
584 			mreq->m_len -= NFSX_UNSIGNED;
585 			bpos -= NFSX_UNSIGNED;
586 		} else {
587 			*tl++ = rpc_mismatch;
588 			*tl++ = txdr_unsigned(RPC_VER2);
589 			*tl = txdr_unsigned(RPC_VER2);
590 		}
591 	} else {
592 		*tl++ = rpc_msgaccepted;
593 
594 		/*
595 		 * For Kerberos authentication, we must send the nickname
596 		 * verifier back, otherwise just RPCAUTH_NULL.
597 		 */
598 		if (nd->nd_flag & ND_KERBFULL) {
599 			struct nfsuid *nuidp;
600 			struct timeval ktvin, ktvout;
601 
602 			memset(&ktvout, 0, sizeof ktvout);	/* XXX gcc */
603 
604 			LIST_FOREACH(nuidp,
605 			    NUIDHASH(slp, kauth_cred_geteuid(nd->nd_cr)),
606 			    nu_hash) {
607 				if (kauth_cred_geteuid(nuidp->nu_cr) ==
608 				kauth_cred_geteuid(nd->nd_cr) &&
609 				    (!nd->nd_nam2 || netaddr_match(
610 				    NU_NETFAM(nuidp), &nuidp->nu_haddr,
611 				    nd->nd_nam2)))
612 					break;
613 			}
614 			if (nuidp) {
615 				ktvin.tv_sec =
616 				    txdr_unsigned(nuidp->nu_timestamp.tv_sec
617 					- 1);
618 				ktvin.tv_usec =
619 				    txdr_unsigned(nuidp->nu_timestamp.tv_usec);
620 
621 				/*
622 				 * Encrypt the timestamp in ecb mode using the
623 				 * session key.
624 				 */
625 #ifdef NFSKERB
626 				XXX
627 #else
628 				(void)ktvin.tv_sec;
629 #endif
630 
631 				*tl++ = rpc_auth_kerb;
632 				*tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
633 				*tl = ktvout.tv_sec;
634 				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
635 				*tl++ = ktvout.tv_usec;
636 				*tl++ = txdr_unsigned(
637 				    kauth_cred_geteuid(nuidp->nu_cr));
638 			} else {
639 				*tl++ = 0;
640 				*tl++ = 0;
641 			}
642 		} else {
643 			*tl++ = 0;
644 			*tl++ = 0;
645 		}
646 		switch (err) {
647 		case EPROGUNAVAIL:
648 			*tl = txdr_unsigned(RPC_PROGUNAVAIL);
649 			break;
650 		case EPROGMISMATCH:
651 			*tl = txdr_unsigned(RPC_PROGMISMATCH);
652 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
653 			*tl++ = txdr_unsigned(2);
654 			*tl = txdr_unsigned(3);
655 			break;
656 		case EPROCUNAVAIL:
657 			*tl = txdr_unsigned(RPC_PROCUNAVAIL);
658 			break;
659 		case EBADRPC:
660 			*tl = txdr_unsigned(RPC_GARBAGE);
661 			break;
662 		default:
663 			*tl = 0;
664 			if (err != NFSERR_RETVOID) {
665 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
666 				if (err)
667 				    *tl = txdr_unsigned(nfsrv_errmap(nd, err));
668 				else
669 				    *tl = 0;
670 			}
671 			break;
672 		};
673 	}
674 
675 	if (mrq != NULL)
676 		*mrq = mreq;
677 	*mbp = mb;
678 	*bposp = bpos;
679 	if (err != 0 && err != NFSERR_RETVOID)
680 		nfsstats.srvrpc_errs++;
681 	return (0);
682 }
683 
684 static void
nfs_timer_schedule(void)685 nfs_timer_schedule(void)
686 {
687 
688 	callout_schedule(&nfs_timer_ch, nfs_ticks);
689 }
690 
691 void
nfs_timer_start(void)692 nfs_timer_start(void)
693 {
694 
695 	if (callout_pending(&nfs_timer_ch))
696 		return;
697 
698 	nfs_timer_start_ev.ev_count++;
699 	nfs_timer_schedule();
700 }
701 
702 void
nfs_timer_init(void)703 nfs_timer_init(void)
704 {
705 
706 	mutex_init(&nfs_timer_lock, MUTEX_DEFAULT, IPL_NONE);
707 	callout_init(&nfs_timer_ch, 0);
708 	callout_setfunc(&nfs_timer_ch, nfs_timer, NULL);
709 	evcnt_attach_dynamic(&nfs_timer_ev, EVCNT_TYPE_MISC, NULL,
710 	    "nfs", "timer");
711 	evcnt_attach_dynamic(&nfs_timer_start_ev, EVCNT_TYPE_MISC, NULL,
712 	    "nfs", "timer start");
713 	evcnt_attach_dynamic(&nfs_timer_stop_ev, EVCNT_TYPE_MISC, NULL,
714 	    "nfs", "timer stop");
715 }
716 
717 void
nfs_timer_fini(void)718 nfs_timer_fini(void)
719 {
720 
721 	callout_halt(&nfs_timer_ch, NULL);
722 	callout_destroy(&nfs_timer_ch);
723 	mutex_destroy(&nfs_timer_lock);
724 	evcnt_detach(&nfs_timer_ev);
725 	evcnt_detach(&nfs_timer_start_ev);
726 	evcnt_detach(&nfs_timer_stop_ev);
727 }
728 
729 void
nfs_timer_srvinit(bool (* func)(void))730 nfs_timer_srvinit(bool (*func)(void))
731 {
732 
733 	nfs_timer_srvvec = func;
734 }
735 
736 void
nfs_timer_srvfini(void)737 nfs_timer_srvfini(void)
738 {
739 
740 	mutex_enter(&nfs_timer_lock);
741 	nfs_timer_srvvec = NULL;
742 	mutex_exit(&nfs_timer_lock);
743 }
744 
745 
746 /*
747  * Nfs timer routine
748  * Scan the nfsreq list and retranmit any requests that have timed out
749  * To avoid retransmission attempts on STREAM sockets (in the future) make
750  * sure to set the r_retry field to 0 (implies nm_retry == 0).
751  */
752 void
nfs_timer(void * arg)753 nfs_timer(void *arg)
754 {
755 	struct nfsreq *rep;
756 	struct mbuf *m;
757 	struct socket *so;
758 	struct nfsmount *nmp;
759 	int timeo;
760 	int error;
761 	bool more = false;
762 
763 	nfs_timer_ev.ev_count++;
764 
765 	mutex_enter(&nfs_reqq_lock);
766 	TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
767 		more = true;
768 		nmp = rep->r_nmp;
769 		if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
770 			continue;
771 		if (nfs_sigintr(nmp, rep, rep->r_lwp)) {
772 			rep->r_flags |= R_SOFTTERM;
773 			continue;
774 		}
775 		if (rep->r_rtt >= 0) {
776 			rep->r_rtt++;
777 			if (nmp->nm_flag & NFSMNT_DUMBTIMR)
778 				timeo = nmp->nm_timeo;
779 			else
780 				timeo = NFS_RTO(nmp, nfs_proct[rep->r_procnum]);
781 			if (nmp->nm_timeouts > 0)
782 				timeo *= nfs_backoff[nmp->nm_timeouts - 1];
783 			if (timeo > NFS_MAXTIMEO)
784 				timeo = NFS_MAXTIMEO;
785 			if (rep->r_rtt <= timeo)
786 				continue;
787 			if (nmp->nm_timeouts <
788 			    (sizeof(nfs_backoff) / sizeof(nfs_backoff[0])))
789 				nmp->nm_timeouts++;
790 		}
791 		/*
792 		 * Check for server not responding
793 		 */
794 		if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
795 		     rep->r_rexmit > nmp->nm_deadthresh) {
796 			nfs_msg(rep->r_lwp,
797 			    nmp->nm_mountp->mnt_stat.f_mntfromname,
798 			    "not responding");
799 			rep->r_flags |= R_TPRINTFMSG;
800 		}
801 		if (rep->r_rexmit >= rep->r_retry) {	/* too many */
802 			nfsstats.rpctimeouts++;
803 			rep->r_flags |= R_SOFTTERM;
804 			continue;
805 		}
806 		if (nmp->nm_sotype != SOCK_DGRAM) {
807 			if (++rep->r_rexmit > NFS_MAXREXMIT)
808 				rep->r_rexmit = NFS_MAXREXMIT;
809 			continue;
810 		}
811 		if ((so = nmp->nm_so) == NULL)
812 			continue;
813 
814 		/*
815 		 * If there is enough space and the window allows..
816 		 *	Resend it
817 		 * Set r_rtt to -1 in case we fail to send it now.
818 		 */
819 		solock(so);
820 		rep->r_rtt = -1;
821 		if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
822 		   ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
823 		    (rep->r_flags & R_SENT) ||
824 		    nmp->nm_sent < nmp->nm_cwnd) &&
825 		   (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
826 		        if (so->so_state & SS_ISCONNECTED)
827 			    error = (*so->so_proto->pr_usrreqs->pr_send)(so,
828 			    m, NULL, NULL, NULL);
829 			else
830 			    error = (*so->so_proto->pr_usrreqs->pr_send)(so,
831 				m, mtod(nmp->nm_nam, struct sockaddr *),
832 				NULL, NULL);
833 			if (error) {
834 				if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
835 #ifdef DEBUG
836 					if (ratecheck(&nfs_timer_last_err_time,
837 					    &nfs_err_interval))
838 						printf("%s: ignoring error "
839 						       "%d\n", __func__, error);
840 #endif
841 					so->so_error = 0;
842 				}
843 			} else {
844 				/*
845 				 * Iff first send, start timing
846 				 * else turn timing off, backoff timer
847 				 * and divide congestion window by 2.
848 				 */
849 				if (rep->r_flags & R_SENT) {
850 					rep->r_flags &= ~R_TIMING;
851 					if (++rep->r_rexmit > NFS_MAXREXMIT)
852 						rep->r_rexmit = NFS_MAXREXMIT;
853 					nmp->nm_cwnd >>= 1;
854 					if (nmp->nm_cwnd < NFS_CWNDSCALE)
855 						nmp->nm_cwnd = NFS_CWNDSCALE;
856 					nfsstats.rpcretries++;
857 				} else {
858 					rep->r_flags |= R_SENT;
859 					nmp->nm_sent += NFS_CWNDSCALE;
860 				}
861 				rep->r_rtt = 0;
862 			}
863 		}
864 		sounlock(so);
865 	}
866 	mutex_exit(&nfs_reqq_lock);
867 
868 	mutex_enter(&nfs_timer_lock);
869 	if (nfs_timer_srvvec != NULL) {
870 		more |= (*nfs_timer_srvvec)();
871 	}
872 	mutex_exit(&nfs_timer_lock);
873 
874 	if (more) {
875 		nfs_timer_schedule();
876 	} else {
877 		nfs_timer_stop_ev.ev_count++;
878 	}
879 }
880 
881 /*
882  * Test for a termination condition pending on the process.
883  * This is used for NFSMNT_INT mounts.
884  */
885 int
nfs_sigintr(struct nfsmount * nmp,struct nfsreq * rep,struct lwp * l)886 nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct lwp *l)
887 {
888 	sigset_t ss;
889 
890 	if (rep && (rep->r_flags & R_SOFTTERM))
891 		return (EINTR);
892 	if (!(nmp->nm_flag & NFSMNT_INT))
893 		return (0);
894 	if (l) {
895 		sigpending1(l, &ss);
896 #if 0
897 		sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss);
898 #endif
899 		if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
900 		    sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
901 		    sigismember(&ss, SIGQUIT))
902 			return (EINTR);
903 	}
904 	return (0);
905 }
906 
907 int
nfs_rcvlock(struct nfsmount * nmp,struct nfsreq * rep)908 nfs_rcvlock(struct nfsmount *nmp, struct nfsreq *rep)
909 {
910 	int *flagp = &nmp->nm_iflag;
911 	int slptimeo = 0;
912 	bool catch_p;
913 	int error = 0;
914 
915 	KASSERT(nmp == rep->r_nmp);
916 
917 	if (nmp->nm_flag & NFSMNT_SOFT)
918 		slptimeo = nmp->nm_retry * nmp->nm_timeo;
919 
920 	if (nmp->nm_iflag & NFSMNT_DISMNTFORCE)
921 		slptimeo = hz;
922 
923 	catch_p = (nmp->nm_flag & NFSMNT_INT) != 0;
924 	mutex_enter(&nmp->nm_lock);
925 	while (/* CONSTCOND */ true) {
926 		if (*flagp & NFSMNT_DISMNT) {
927 			cv_signal(&nmp->nm_disconcv);
928 			error = EIO;
929 			break;
930 		}
931 		/* If our reply was received while we were sleeping,
932 		 * then just return without taking the lock to avoid a
933 		 * situation where a single iod could 'capture' the
934 		 * receive lock.
935 		 */
936 		if (rep->r_mrep != NULL) {
937 			cv_signal(&nmp->nm_rcvcv);
938 			error = EALREADY;
939 			break;
940 		}
941 		if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) {
942 			cv_signal(&nmp->nm_rcvcv);
943 			error = EINTR;
944 			break;
945 		}
946 		if ((*flagp & NFSMNT_RCVLOCK) == 0) {
947 			*flagp |= NFSMNT_RCVLOCK;
948 			break;
949 		}
950 		if (catch_p) {
951 			error = cv_timedwait_sig(&nmp->nm_rcvcv, &nmp->nm_lock,
952 			    slptimeo);
953 		} else {
954 			error = cv_timedwait(&nmp->nm_rcvcv, &nmp->nm_lock,
955 			    slptimeo);
956 		}
957 		if (error) {
958 			if ((error == EWOULDBLOCK) &&
959 			    (nmp->nm_flag & NFSMNT_SOFT)) {
960 				error = EIO;
961 				break;
962 			}
963 			error = 0;
964 		}
965 		if (catch_p) {
966 			catch_p = false;
967 			slptimeo = 2 * hz;
968 		}
969 	}
970 	mutex_exit(&nmp->nm_lock);
971 	return error;
972 }
973 
974 /*
975  * Unlock the stream socket for others.
976  */
977 void
nfs_rcvunlock(struct nfsmount * nmp)978 nfs_rcvunlock(struct nfsmount *nmp)
979 {
980 
981 	mutex_enter(&nmp->nm_lock);
982 	if ((nmp->nm_iflag & NFSMNT_RCVLOCK) == 0)
983 		panic("nfs rcvunlock");
984 	nmp->nm_iflag &= ~NFSMNT_RCVLOCK;
985 	cv_signal(&nmp->nm_rcvcv);
986 	mutex_exit(&nmp->nm_lock);
987 }
988 
989 /*
990  * Parse an RPC request
991  * - verify it
992  * - allocate and fill in the cred.
993  */
994 int
nfs_getreq(struct nfsrv_descript * nd,struct nfsd * nfsd,int has_header)995 nfs_getreq(struct nfsrv_descript *nd, struct nfsd *nfsd, int has_header)
996 {
997 	int len, i;
998 	u_int32_t *tl;
999 	int32_t t1;
1000 	struct uio uio;
1001 	struct iovec iov;
1002 	char *dpos, *cp2, *cp;
1003 	u_int32_t nfsvers, auth_type;
1004 	uid_t nickuid;
1005 	int error = 0, ticklen;
1006 	struct mbuf *mrep, *md;
1007 	struct nfsuid *nuidp;
1008 	struct timeval tvin, tvout;
1009 
1010 	memset(&tvout, 0, sizeof tvout);	/* XXX gcc */
1011 
1012 	KASSERT(nd->nd_cr == NULL);
1013 	mrep = nd->nd_mrep;
1014 	md = nd->nd_md;
1015 	dpos = nd->nd_dpos;
1016 	if (has_header) {
1017 		nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1018 		nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1019 		if (*tl++ != rpc_call) {
1020 			m_freem(mrep);
1021 			return (EBADRPC);
1022 		}
1023 	} else
1024 		nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1025 	nd->nd_repstat = 0;
1026 	nd->nd_flag = 0;
1027 	if (*tl++ != rpc_vers) {
1028 		nd->nd_repstat = ERPCMISMATCH;
1029 		nd->nd_procnum = NFSPROC_NOOP;
1030 		return (0);
1031 	}
1032 	if (*tl != nfs_prog) {
1033 		nd->nd_repstat = EPROGUNAVAIL;
1034 		nd->nd_procnum = NFSPROC_NOOP;
1035 		return (0);
1036 	}
1037 	tl++;
1038 	nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1039 	if (nfsvers < NFS_VER2 || nfsvers > NFS_VER3) {
1040 		nd->nd_repstat = EPROGMISMATCH;
1041 		nd->nd_procnum = NFSPROC_NOOP;
1042 		return (0);
1043 	}
1044 	if (nfsvers == NFS_VER3)
1045 		nd->nd_flag = ND_NFSV3;
1046 	nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1047 	if (nd->nd_procnum == NFSPROC_NULL)
1048 		return (0);
1049 	if (nd->nd_procnum > NFSPROC_COMMIT ||
1050 	    (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1051 		nd->nd_repstat = EPROCUNAVAIL;
1052 		nd->nd_procnum = NFSPROC_NOOP;
1053 		return (0);
1054 	}
1055 	if ((nd->nd_flag & ND_NFSV3) == 0)
1056 		nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1057 	auth_type = *tl++;
1058 	len = fxdr_unsigned(int, *tl++);
1059 	if (len < 0 || len > RPCAUTH_MAXSIZ) {
1060 		m_freem(mrep);
1061 		return (EBADRPC);
1062 	}
1063 
1064 	nd->nd_flag &= ~ND_KERBAUTH;
1065 	/*
1066 	 * Handle auth_unix or auth_kerb.
1067 	 */
1068 	if (auth_type == rpc_auth_unix) {
1069 		uid_t uid;
1070 		gid_t gid;
1071 
1072 		nd->nd_cr = kauth_cred_alloc();
1073 		len = fxdr_unsigned(int, *++tl);
1074 		if (len < 0 || len > NFS_MAXNAMLEN) {
1075 			m_freem(mrep);
1076 			error = EBADRPC;
1077 			goto errout;
1078 		}
1079 		nfsm_adv(nfsm_rndup(len));
1080 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1081 
1082 		uid = fxdr_unsigned(uid_t, *tl++);
1083 		gid = fxdr_unsigned(gid_t, *tl++);
1084 		kauth_cred_setuid(nd->nd_cr, uid);
1085 		kauth_cred_seteuid(nd->nd_cr, uid);
1086 		kauth_cred_setsvuid(nd->nd_cr, uid);
1087 		kauth_cred_setgid(nd->nd_cr, gid);
1088 		kauth_cred_setegid(nd->nd_cr, gid);
1089 		kauth_cred_setsvgid(nd->nd_cr, gid);
1090 
1091 		len = fxdr_unsigned(int, *tl);
1092 		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1093 			m_freem(mrep);
1094 			error = EBADRPC;
1095 			goto errout;
1096 		}
1097 		nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1098 
1099 		if (len > 0) {
1100 			size_t grbuf_size = uimin(len, NGROUPS) * sizeof(gid_t);
1101 			gid_t *grbuf = kmem_alloc(grbuf_size, KM_SLEEP);
1102 
1103 			for (i = 0; i < len; i++) {
1104 				if (i < NGROUPS) /* XXX elad */
1105 					grbuf[i] = fxdr_unsigned(gid_t, *tl++);
1106 				else
1107 					tl++;
1108 			}
1109 			kauth_cred_setgroups(nd->nd_cr, grbuf,
1110 			    uimin(len, NGROUPS), -1, UIO_SYSSPACE);
1111 			kmem_free(grbuf, grbuf_size);
1112 		}
1113 
1114 		len = fxdr_unsigned(int, *++tl);
1115 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
1116 			m_freem(mrep);
1117 			error = EBADRPC;
1118 			goto errout;
1119 		}
1120 		if (len > 0)
1121 			nfsm_adv(nfsm_rndup(len));
1122 	} else if (auth_type == rpc_auth_kerb) {
1123 		switch (fxdr_unsigned(int, *tl++)) {
1124 		case RPCAKN_FULLNAME:
1125 			ticklen = fxdr_unsigned(int, *tl);
1126 			*((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1127 			uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1128 			nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1129 			if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1130 				m_freem(mrep);
1131 				error = EBADRPC;
1132 				goto errout;
1133 			}
1134 			uio.uio_offset = 0;
1135 			uio.uio_iov = &iov;
1136 			uio.uio_iovcnt = 1;
1137 			UIO_SETUP_SYSSPACE(&uio);
1138 			iov.iov_base = (void *)&nfsd->nfsd_authstr[4];
1139 			iov.iov_len = RPCAUTH_MAXSIZ - 4;
1140 			nfsm_mtouio(&uio, uio.uio_resid);
1141 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1142 			if (*tl++ != rpc_auth_kerb ||
1143 				fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1144 				printf("Bad kerb verifier\n");
1145 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1146 				nd->nd_procnum = NFSPROC_NOOP;
1147 				return (0);
1148 			}
1149 			nfsm_dissect(cp, void *, 4 * NFSX_UNSIGNED);
1150 			tl = (u_int32_t *)cp;
1151 			if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1152 				printf("Not fullname kerb verifier\n");
1153 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1154 				nd->nd_procnum = NFSPROC_NOOP;
1155 				return (0);
1156 			}
1157 			cp += NFSX_UNSIGNED;
1158 			memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
1159 			nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1160 			nd->nd_flag |= ND_KERBFULL;
1161 			nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1162 			break;
1163 		case RPCAKN_NICKNAME:
1164 			if (len != 2 * NFSX_UNSIGNED) {
1165 				printf("Kerb nickname short\n");
1166 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1167 				nd->nd_procnum = NFSPROC_NOOP;
1168 				return (0);
1169 			}
1170 			nickuid = fxdr_unsigned(uid_t, *tl);
1171 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1172 			if (*tl++ != rpc_auth_kerb ||
1173 				fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1174 				printf("Kerb nick verifier bad\n");
1175 				nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1176 				nd->nd_procnum = NFSPROC_NOOP;
1177 				return (0);
1178 			}
1179 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1180 			tvin.tv_sec = *tl++;
1181 			tvin.tv_usec = *tl;
1182 
1183 			LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid),
1184 			    nu_hash) {
1185 				if (kauth_cred_geteuid(nuidp->nu_cr) == nickuid &&
1186 				    (!nd->nd_nam2 ||
1187 				     netaddr_match(NU_NETFAM(nuidp),
1188 				      &nuidp->nu_haddr, nd->nd_nam2)))
1189 					break;
1190 			}
1191 			if (!nuidp) {
1192 				nd->nd_repstat =
1193 					(NFSERR_AUTHERR|AUTH_REJECTCRED);
1194 				nd->nd_procnum = NFSPROC_NOOP;
1195 				return (0);
1196 			}
1197 
1198 			/*
1199 			 * Now, decrypt the timestamp using the session key
1200 			 * and validate it.
1201 			 */
1202 #ifdef NFSKERB
1203 			XXX
1204 #else
1205 			(void)tvin.tv_sec;
1206 #endif
1207 
1208 			tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1209 			tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1210 			if (nuidp->nu_expire < time_second ||
1211 			    nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1212 			    (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1213 			     nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1214 				nuidp->nu_expire = 0;
1215 				nd->nd_repstat =
1216 				    (NFSERR_AUTHERR|AUTH_REJECTVERF);
1217 				nd->nd_procnum = NFSPROC_NOOP;
1218 				return (0);
1219 			}
1220 			kauth_cred_hold(nuidp->nu_cr);
1221 			nd->nd_cr = nuidp->nu_cr;
1222 			nd->nd_flag |= ND_KERBNICK;
1223 		}
1224 	} else {
1225 		nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1226 		nd->nd_procnum = NFSPROC_NOOP;
1227 		return (0);
1228 	}
1229 
1230 	nd->nd_md = md;
1231 	nd->nd_dpos = dpos;
1232 	KASSERT((nd->nd_cr == NULL) ==
1233 	    ((nfsd->nfsd_flag & NFSD_NEEDAUTH) != 0));
1234 	return (0);
1235 nfsmout:
1236 errout:
1237 	KASSERT(error != 0);
1238 	if (nd->nd_cr != NULL) {
1239 		kauth_cred_free(nd->nd_cr);
1240 		nd->nd_cr = NULL;
1241 	}
1242 	return (error);
1243 }
1244 
1245 int
nfs_msg(struct lwp * l,const char * server,const char * msg)1246 nfs_msg(struct lwp *l, const char *server, const char *msg)
1247 {
1248 	tpr_t tpr;
1249 
1250 #if 0 /* XXX nfs_timer can't block on proc_lock */
1251 	if (l)
1252 		tpr = tprintf_open(l->l_proc);
1253 	else
1254 #endif
1255 		tpr = NULL;
1256 	tprintf(tpr, "nfs server %s: %s\n", server, msg);
1257 	tprintf_close(tpr);
1258 	return (0);
1259 }
1260 
1261 static struct pool nfs_srvdesc_pool;
1262 
1263 void
nfsdreq_init(void)1264 nfsdreq_init(void)
1265 {
1266 
1267 	pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript),
1268 	    0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr, IPL_NONE);
1269 }
1270 
1271 void
nfsdreq_fini(void)1272 nfsdreq_fini(void)
1273 {
1274 
1275 	pool_destroy(&nfs_srvdesc_pool);
1276 }
1277 
1278 struct nfsrv_descript *
nfsdreq_alloc(void)1279 nfsdreq_alloc(void)
1280 {
1281 	struct nfsrv_descript *nd;
1282 
1283 	nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK);
1284 	nd->nd_cr = NULL;
1285 	return nd;
1286 }
1287 
1288 void
nfsdreq_free(struct nfsrv_descript * nd)1289 nfsdreq_free(struct nfsrv_descript *nd)
1290 {
1291 	kauth_cred_t cr;
1292 
1293 	cr = nd->nd_cr;
1294 	if (cr != NULL) {
1295 		kauth_cred_free(cr);
1296 	}
1297 	pool_put(&nfs_srvdesc_pool, nd);
1298 }
1299