FreeBSD/Linux Kernel Cross Reference
sys/nfs/nfs_socket.c
1 /* $NetBSD: nfs_socket.c,v 1.112.2.2 2005/12/15 20:22:16 tron 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 1.112.2.2 2005/12/15 20:22:16 tron Exp $");
43
44 #include "fs_nfs.h"
45 #include "opt_nfs.h"
46 #include "opt_nfsserver.h"
47 #include "opt_mbuftrace.h"
48 #include "opt_inet.h"
49
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/callout.h>
53 #include <sys/proc.h>
54 #include <sys/mount.h>
55 #include <sys/kernel.h>
56 #include <sys/mbuf.h>
57 #include <sys/vnode.h>
58 #include <sys/domain.h>
59 #include <sys/protosw.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
62 #include <sys/syslog.h>
63 #include <sys/tprintf.h>
64 #include <sys/namei.h>
65 #include <sys/signal.h>
66 #include <sys/signalvar.h>
67
68 #include <netinet/in.h>
69 #include <netinet/tcp.h>
70
71 #include <nfs/rpcv2.h>
72 #include <nfs/nfsproto.h>
73 #include <nfs/nfs.h>
74 #include <nfs/xdr_subs.h>
75 #include <nfs/nfsm_subs.h>
76 #include <nfs/nfsmount.h>
77 #include <nfs/nfsnode.h>
78 #include <nfs/nfsrtt.h>
79 #include <nfs/nqnfs.h>
80 #include <nfs/nfs_var.h>
81
82 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
83 #ifdef MBUFTRACE
84 struct mowner nfs_mowner = { "nfs" };
85 #endif
86
87 /*
88 * Estimate rto for an nfs rpc sent via. an unreliable datagram.
89 * Use the mean and mean deviation of rtt for the appropriate type of rpc
90 * for the frequent rpcs and a default for the others.
91 * The justification for doing "other" this way is that these rpcs
92 * happen so infrequently that timer est. would probably be stale.
93 * Also, since many of these rpcs are
94 * non-idempotent, a conservative timeout is desired.
95 * getattr, lookup - A+2D
96 * read, write - A+4D
97 * other - nm_timeo
98 */
99 #define NFS_RTO(n, t) \
100 ((t) == 0 ? (n)->nm_timeo : \
101 ((t) < 3 ? \
102 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
103 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
104 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
105 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
106 /*
107 * External data, mostly RPC constants in XDR form
108 */
109 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
110 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
111 rpc_auth_kerb;
112 extern u_int32_t nfs_prog, nqnfs_prog;
113 extern time_t nqnfsstarttime;
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 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
127 0, 0, 0,
128 };
129
130 /*
131 * There is a congestion window for outstanding rpcs maintained per mount
132 * point. The cwnd size is adjusted in roughly the way that:
133 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
134 * SIGCOMM '88". ACM, August 1988.
135 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
136 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
137 * of rpcs is in progress.
138 * (The sent count and cwnd are scaled for integer arith.)
139 * Variants of "slow start" were tried and were found to be too much of a
140 * performance hit (ave. rtt 3 times larger),
141 * I suspect due to the large rtt that nfs rpcs have.
142 */
143 #define NFS_CWNDSCALE 256
144 #define NFS_MAXCWND (NFS_CWNDSCALE * 32)
145 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
146 int nfsrtton = 0;
147 struct nfsrtt nfsrtt;
148 struct nfsreqhead nfs_reqq;
149
150 struct callout nfs_timer_ch = CALLOUT_INITIALIZER_SETFUNC(nfs_timer, NULL);
151
152 /*
153 * Initialize sockets and congestion for a new NFS connection.
154 * We do not free the sockaddr if error.
155 */
156 int
157 nfs_connect(nmp, rep, p)
158 struct nfsmount *nmp;
159 struct nfsreq *rep;
160 struct proc *p;
161 {
162 struct socket *so;
163 int s, error, rcvreserve, sndreserve;
164 struct sockaddr *saddr;
165 struct sockaddr_in *sin;
166 #ifdef INET6
167 struct sockaddr_in6 *sin6;
168 #endif
169 struct mbuf *m;
170
171 nmp->nm_so = (struct socket *)0;
172 saddr = mtod(nmp->nm_nam, struct sockaddr *);
173 error = socreate(saddr->sa_family, &nmp->nm_so,
174 nmp->nm_sotype, nmp->nm_soproto, p);
175 if (error)
176 goto bad;
177 so = nmp->nm_so;
178 #ifdef MBUFTRACE
179 so->so_mowner = &nfs_mowner;
180 so->so_rcv.sb_mowner = &nfs_mowner;
181 so->so_snd.sb_mowner = &nfs_mowner;
182 #endif
183 nmp->nm_soflags = so->so_proto->pr_flags;
184
185 /*
186 * Some servers require that the client port be a reserved port number.
187 */
188 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
189 m = m_get(M_WAIT, MT_SOOPTS);
190 MCLAIM(m, so->so_mowner);
191 *mtod(m, int32_t *) = IP_PORTRANGE_LOW;
192 m->m_len = sizeof(int32_t);
193 if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m)))
194 goto bad;
195 m = m_get(M_WAIT, MT_SONAME);
196 MCLAIM(m, so->so_mowner);
197 sin = mtod(m, struct sockaddr_in *);
198 sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
199 sin->sin_family = AF_INET;
200 sin->sin_addr.s_addr = INADDR_ANY;
201 sin->sin_port = 0;
202 error = sobind(so, m, &proc0);
203 m_freem(m);
204 if (error)
205 goto bad;
206 }
207 #ifdef INET6
208 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) {
209 m = m_get(M_WAIT, MT_SOOPTS);
210 MCLAIM(m, so->so_mowner);
211 *mtod(m, int32_t *) = IPV6_PORTRANGE_LOW;
212 m->m_len = sizeof(int32_t);
213 if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m)))
214 goto bad;
215 m = m_get(M_WAIT, MT_SONAME);
216 MCLAIM(m, so->so_mowner);
217 sin6 = mtod(m, struct sockaddr_in6 *);
218 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6);
219 sin6->sin6_family = AF_INET6;
220 sin6->sin6_addr = in6addr_any;
221 sin6->sin6_port = 0;
222 error = sobind(so, m, &proc0);
223 m_freem(m);
224 if (error)
225 goto bad;
226 }
227 #endif
228
229 /*
230 * Protocols that do not require connections may be optionally left
231 * unconnected for servers that reply from a port other than NFS_PORT.
232 */
233 if (nmp->nm_flag & NFSMNT_NOCONN) {
234 if (nmp->nm_soflags & PR_CONNREQUIRED) {
235 error = ENOTCONN;
236 goto bad;
237 }
238 } else {
239 error = soconnect(so, nmp->nm_nam, p);
240 if (error)
241 goto bad;
242
243 /*
244 * Wait for the connection to complete. Cribbed from the
245 * connect system call but with the wait timing out so
246 * that interruptible mounts don't hang here for a long time.
247 */
248 s = splsoftnet();
249 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
250 (void) tsleep((caddr_t)&so->so_timeo, PSOCK,
251 "nfscn1", 2 * hz);
252 if ((so->so_state & SS_ISCONNECTING) &&
253 so->so_error == 0 && rep &&
254 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
255 so->so_state &= ~SS_ISCONNECTING;
256 splx(s);
257 goto bad;
258 }
259 }
260 if (so->so_error) {
261 error = so->so_error;
262 so->so_error = 0;
263 splx(s);
264 goto bad;
265 }
266 splx(s);
267 }
268 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
269 so->so_rcv.sb_timeo = (5 * hz);
270 so->so_snd.sb_timeo = (5 * hz);
271 } else {
272 /*
273 * enable receive timeout to detect server crash and reconnect.
274 * otherwise, we can be stuck in soreceive forever.
275 */
276 so->so_rcv.sb_timeo = (5 * hz);
277 so->so_snd.sb_timeo = 0;
278 }
279 if (nmp->nm_sotype == SOCK_DGRAM) {
280 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
281 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
282 NFS_MAXPKTHDR) * 2;
283 } else if (nmp->nm_sotype == SOCK_SEQPACKET) {
284 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
285 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
286 NFS_MAXPKTHDR) * 2;
287 } else {
288 if (nmp->nm_sotype != SOCK_STREAM)
289 panic("nfscon sotype");
290 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
291 m = m_get(M_WAIT, MT_SOOPTS);
292 MCLAIM(m, so->so_mowner);
293 *mtod(m, int32_t *) = 1;
294 m->m_len = sizeof(int32_t);
295 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
296 }
297 if (so->so_proto->pr_protocol == IPPROTO_TCP) {
298 m = m_get(M_WAIT, MT_SOOPTS);
299 MCLAIM(m, so->so_mowner);
300 *mtod(m, int32_t *) = 1;
301 m->m_len = sizeof(int32_t);
302 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
303 }
304 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
305 sizeof (u_int32_t)) * 2;
306 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
307 sizeof (u_int32_t)) * 2;
308 }
309 error = soreserve(so, sndreserve, rcvreserve);
310 if (error)
311 goto bad;
312 so->so_rcv.sb_flags |= SB_NOINTR;
313 so->so_snd.sb_flags |= SB_NOINTR;
314
315 /* Initialize other non-zero congestion variables */
316 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
317 NFS_TIMEO << 3;
318 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
319 nmp->nm_sdrtt[3] = 0;
320 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */
321 nmp->nm_sent = 0;
322 nmp->nm_timeouts = 0;
323 return (0);
324
325 bad:
326 nfs_disconnect(nmp);
327 return (error);
328 }
329
330 /*
331 * Reconnect routine:
332 * Called when a connection is broken on a reliable protocol.
333 * - clean up the old socket
334 * - nfs_connect() again
335 * - set R_MUSTRESEND for all outstanding requests on mount point
336 * If this fails the mount point is DEAD!
337 * nb: Must be called with the nfs_sndlock() set on the mount point.
338 */
339 int
340 nfs_reconnect(rep, p)
341 struct nfsreq *rep;
342 struct proc *p;
343 {
344 struct nfsreq *rp;
345 struct nfsmount *nmp = rep->r_nmp;
346 int error;
347
348 nfs_disconnect(nmp);
349 while ((error = nfs_connect(nmp, rep, p)) != 0) {
350 if (error == EINTR || error == ERESTART)
351 return (EINTR);
352 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0);
353 }
354
355 /*
356 * Loop through outstanding request list and fix up all requests
357 * on old socket.
358 */
359 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
360 if (rp->r_nmp == nmp) {
361 if ((rp->r_flags & R_MUSTRESEND) == 0)
362 rp->r_flags |= R_MUSTRESEND | R_REXMITTED;
363 rp->r_rexmit = 0;
364 }
365 }
366 return (0);
367 }
368
369 /*
370 * NFS disconnect. Clean up and unlink.
371 */
372 void
373 nfs_disconnect(nmp)
374 struct nfsmount *nmp;
375 {
376 struct socket *so;
377 int drain = 0;
378
379 if (nmp->nm_so) {
380 so = nmp->nm_so;
381 nmp->nm_so = (struct socket *)0;
382 soshutdown(so, 2);
383 drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0;
384 if (drain) {
385 /*
386 * soshutdown() above should wake up the current
387 * listener.
388 * Now wake up those waiting for the receive lock, and
389 * wait for them to go away unhappy, to prevent *nmp
390 * from evaporating while they're sleeping.
391 */
392 while (nmp->nm_waiters > 0) {
393 wakeup (&nmp->nm_iflag);
394 (void) tsleep(&nmp->nm_waiters, PVFS,
395 "nfsdis", 0);
396 }
397 }
398 soclose(so);
399 }
400 #ifdef DIAGNOSTIC
401 if (drain && (nmp->nm_waiters > 0))
402 panic("nfs_disconnect: waiters left after drain?");
403 #endif
404 }
405
406 void
407 nfs_safedisconnect(nmp)
408 struct nfsmount *nmp;
409 {
410 struct nfsreq dummyreq;
411
412 memset(&dummyreq, 0, sizeof(dummyreq));
413 dummyreq.r_nmp = nmp;
414 nfs_rcvlock(&dummyreq); /* XXX ignored error return */
415 nfs_disconnect(nmp);
416 nfs_rcvunlock(nmp);
417 }
418
419 /*
420 * This is the nfs send routine. For connection based socket types, it
421 * must be called with an nfs_sndlock() on the socket.
422 * "rep == NULL" indicates that it has been called from a server.
423 * For the client side:
424 * - return EINTR if the RPC is terminated, 0 otherwise
425 * - set R_MUSTRESEND if the send fails for any reason
426 * - do any cleanup required by recoverable socket errors (? ? ?)
427 * For the server side:
428 * - return EINTR or ERESTART if interrupted by a signal
429 * - return EPIPE if a connection is lost for connection based sockets (TCP...)
430 * - do any cleanup required by recoverable socket errors (? ? ?)
431 */
432 int
433 nfs_send(so, nam, top, rep, p)
434 struct socket *so;
435 struct mbuf *nam;
436 struct mbuf *top;
437 struct nfsreq *rep;
438 struct proc *p;
439 {
440 struct mbuf *sendnam;
441 int error, soflags, flags;
442
443 /* XXX nfs_doio()/nfs_request() calls with rep->r_procp == NULL */
444 if (p == NULL && rep->r_procp == NULL)
445 p = curproc;
446
447 if (rep) {
448 if (rep->r_flags & R_SOFTTERM) {
449 m_freem(top);
450 return (EINTR);
451 }
452 if ((so = rep->r_nmp->nm_so) == NULL) {
453 rep->r_flags |= R_MUSTRESEND;
454 m_freem(top);
455 return (0);
456 }
457 rep->r_flags &= ~R_MUSTRESEND;
458 soflags = rep->r_nmp->nm_soflags;
459 } else
460 soflags = so->so_proto->pr_flags;
461 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
462 sendnam = (struct mbuf *)0;
463 else
464 sendnam = nam;
465 if (so->so_type == SOCK_SEQPACKET)
466 flags = MSG_EOR;
467 else
468 flags = 0;
469
470 error = (*so->so_send)(so, sendnam, (struct uio *)0, top,
471 (struct mbuf *)0, flags, p);
472 if (error) {
473 if (rep) {
474 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
475 /*
476 * We're too fast for the network/driver,
477 * and UDP isn't flowcontrolled.
478 * We need to resend. This is not fatal,
479 * just try again.
480 *
481 * Could be smarter here by doing some sort
482 * of a backoff, but this is rare.
483 */
484 rep->r_flags |= R_MUSTRESEND;
485 } else {
486 if (error != EPIPE)
487 log(LOG_INFO,
488 "nfs send error %d for %s\n",
489 error,
490 rep->r_nmp->nm_mountp->
491 mnt_stat.f_mntfromname);
492 /*
493 * Deal with errors for the client side.
494 */
495 if (rep->r_flags & R_SOFTTERM)
496 error = EINTR;
497 else
498 rep->r_flags |= R_MUSTRESEND;
499 }
500 } else {
501 /*
502 * See above. This error can happen under normal
503 * circumstances and the log is too noisy.
504 * The error will still show up in nfsstat.
505 */
506 if (error != ENOBUFS || so->so_type != SOCK_DGRAM)
507 log(LOG_INFO, "nfsd send error %d\n", error);
508 }
509
510 /*
511 * Handle any recoverable (soft) socket errors here. (? ? ?)
512 */
513 if (error != EINTR && error != ERESTART &&
514 error != EWOULDBLOCK && error != EPIPE)
515 error = 0;
516 }
517 return (error);
518 }
519
520 #ifdef NFS
521 /*
522 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
523 * done by soreceive(), but for SOCK_STREAM we must deal with the Record
524 * Mark and consolidate the data into a new mbuf list.
525 * nb: Sometimes TCP passes the data up to soreceive() in long lists of
526 * small mbufs.
527 * For SOCK_STREAM we must be very careful to read an entire record once
528 * we have read any of it, even if the system call has been interrupted.
529 */
530 int
531 nfs_receive(rep, aname, mp, p)
532 struct nfsreq *rep;
533 struct mbuf **aname;
534 struct mbuf **mp;
535 struct proc *p;
536 {
537 struct socket *so;
538 struct uio auio;
539 struct iovec aio;
540 struct mbuf *m;
541 struct mbuf *control;
542 u_int32_t len;
543 struct mbuf **getnam;
544 int error, sotype, rcvflg;
545
546 /*
547 * Set up arguments for soreceive()
548 */
549 *mp = (struct mbuf *)0;
550 *aname = (struct mbuf *)0;
551 sotype = rep->r_nmp->nm_sotype;
552
553 /*
554 * For reliable protocols, lock against other senders/receivers
555 * in case a reconnect is necessary.
556 * For SOCK_STREAM, first get the Record Mark to find out how much
557 * more there is to get.
558 * We must lock the socket against other receivers
559 * until we have an entire rpc request/reply.
560 */
561 if (sotype != SOCK_DGRAM) {
562 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
563 if (error)
564 return (error);
565 tryagain:
566 /*
567 * Check for fatal errors and resending request.
568 */
569 /*
570 * Ugh: If a reconnect attempt just happened, nm_so
571 * would have changed. NULL indicates a failed
572 * attempt that has essentially shut down this
573 * mount point.
574 */
575 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
576 nfs_sndunlock(&rep->r_nmp->nm_iflag);
577 return (EINTR);
578 }
579 so = rep->r_nmp->nm_so;
580 if (!so) {
581 error = nfs_reconnect(rep, p);
582 if (error) {
583 nfs_sndunlock(&rep->r_nmp->nm_iflag);
584 return (error);
585 }
586 goto tryagain;
587 }
588 while (rep->r_flags & R_MUSTRESEND) {
589 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
590 nfsstats.rpcretries++;
591 rep->r_rtt = 0;
592 rep->r_flags &= ~R_TIMING;
593 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, p);
594 if (error) {
595 if (error == EINTR || error == ERESTART ||
596 (error = nfs_reconnect(rep, p)) != 0) {
597 nfs_sndunlock(&rep->r_nmp->nm_iflag);
598 return (error);
599 }
600 goto tryagain;
601 }
602 }
603 nfs_sndunlock(&rep->r_nmp->nm_iflag);
604 if (sotype == SOCK_STREAM) {
605 aio.iov_base = (caddr_t) &len;
606 aio.iov_len = sizeof(u_int32_t);
607 auio.uio_iov = &aio;
608 auio.uio_iovcnt = 1;
609 auio.uio_segflg = UIO_SYSSPACE;
610 auio.uio_rw = UIO_READ;
611 auio.uio_offset = 0;
612 auio.uio_resid = sizeof(u_int32_t);
613 auio.uio_procp = NULL;
614 do {
615 rcvflg = MSG_WAITALL;
616 error = (*so->so_receive)(so, (struct mbuf **)0, &auio,
617 (struct mbuf **)0, (struct mbuf **)0, &rcvflg);
618 if (error == EWOULDBLOCK && rep) {
619 if (rep->r_flags & R_SOFTTERM)
620 return (EINTR);
621 /*
622 * if it seems that the server died after it
623 * received our request, set EPIPE so that
624 * we'll reconnect and retransmit requests.
625 */
626 if (rep->r_rexmit >= rep->r_nmp->nm_retry) {
627 nfsstats.rpctimeouts++;
628 error = EPIPE;
629 }
630 }
631 } while (error == EWOULDBLOCK);
632 if (!error && auio.uio_resid > 0) {
633 /*
634 * Don't log a 0 byte receive; it means
635 * that the socket has been closed, and
636 * can happen during normal operation
637 * (forcible unmount or Solaris server).
638 */
639 if (auio.uio_resid != sizeof (u_int32_t))
640 log(LOG_INFO,
641 "short receive (%lu/%lu) from nfs server %s\n",
642 (u_long)sizeof(u_int32_t) - auio.uio_resid,
643 (u_long)sizeof(u_int32_t),
644 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
645 error = EPIPE;
646 }
647 if (error)
648 goto errout;
649 len = ntohl(len) & ~0x80000000;
650 /*
651 * This is SERIOUS! We are out of sync with the sender
652 * and forcing a disconnect/reconnect is all I can do.
653 */
654 if (len > NFS_MAXPACKET) {
655 log(LOG_ERR, "%s (%d) from nfs server %s\n",
656 "impossible packet length",
657 len,
658 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
659 error = EFBIG;
660 goto errout;
661 }
662 auio.uio_resid = len;
663 do {
664 rcvflg = MSG_WAITALL;
665 error = (*so->so_receive)(so, (struct mbuf **)0,
666 &auio, mp, (struct mbuf **)0, &rcvflg);
667 } while (error == EWOULDBLOCK || error == EINTR ||
668 error == ERESTART);
669 if (!error && auio.uio_resid > 0) {
670 if (len != auio.uio_resid)
671 log(LOG_INFO,
672 "short receive (%lu/%d) from nfs server %s\n",
673 (u_long)len - auio.uio_resid, len,
674 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
675 error = EPIPE;
676 }
677 } else {
678 /*
679 * NB: Since uio_resid is big, MSG_WAITALL is ignored
680 * and soreceive() will return when it has either a
681 * control msg or a data msg.
682 * We have no use for control msg., but must grab them
683 * and then throw them away so we know what is going
684 * on.
685 */
686 auio.uio_resid = len = 100000000; /* Anything Big */
687 auio.uio_procp = p;
688 do {
689 rcvflg = 0;
690 error = (*so->so_receive)(so, (struct mbuf **)0,
691 &auio, mp, &control, &rcvflg);
692 if (control)
693 m_freem(control);
694 if (error == EWOULDBLOCK && rep) {
695 if (rep->r_flags & R_SOFTTERM)
696 return (EINTR);
697 }
698 } while (error == EWOULDBLOCK ||
699 (!error && *mp == NULL && control));
700 if ((rcvflg & MSG_EOR) == 0)
701 printf("Egad!!\n");
702 if (!error && *mp == NULL)
703 error = EPIPE;
704 len -= auio.uio_resid;
705 }
706 errout:
707 if (error && error != EINTR && error != ERESTART) {
708 m_freem(*mp);
709 *mp = (struct mbuf *)0;
710 if (error != EPIPE)
711 log(LOG_INFO,
712 "receive error %d from nfs server %s\n",
713 error,
714 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
715 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep);
716 if (!error)
717 error = nfs_reconnect(rep, p);
718 if (!error)
719 goto tryagain;
720 else
721 nfs_sndunlock(&rep->r_nmp->nm_iflag);
722 }
723 } else {
724 if ((so = rep->r_nmp->nm_so) == NULL)
725 return (EACCES);
726 if (so->so_state & SS_ISCONNECTED)
727 getnam = (struct mbuf **)0;
728 else
729 getnam = aname;
730 auio.uio_resid = len = 1000000;
731 auio.uio_procp = p;
732 do {
733 rcvflg = 0;
734 error = (*so->so_receive)(so, getnam, &auio, mp,
735 (struct mbuf **)0, &rcvflg);
736 if (error == EWOULDBLOCK &&
737 (rep->r_flags & R_SOFTTERM))
738 return (EINTR);
739 } while (error == EWOULDBLOCK);
740 len -= auio.uio_resid;
741 if (!error && *mp == NULL)
742 error = EPIPE;
743 }
744 if (error) {
745 m_freem(*mp);
746 *mp = (struct mbuf *)0;
747 }
748 return (error);
749 }
750
751 /*
752 * Implement receipt of reply on a socket.
753 * We must search through the list of received datagrams matching them
754 * with outstanding requests using the xid, until ours is found.
755 */
756 /* ARGSUSED */
757 int
758 nfs_reply(myrep, procp)
759 struct nfsreq *myrep;
760 struct proc *procp;
761 {
762 struct nfsreq *rep;
763 struct nfsmount *nmp = myrep->r_nmp;
764 int32_t t1;
765 struct mbuf *mrep, *nam, *md;
766 u_int32_t rxid, *tl;
767 caddr_t dpos, cp2;
768 int error;
769
770 /*
771 * Loop around until we get our own reply
772 */
773 for (;;) {
774 /*
775 * Lock against other receivers so that I don't get stuck in
776 * sbwait() after someone else has received my reply for me.
777 * Also necessary for connection based protocols to avoid
778 * race conditions during a reconnect.
779 */
780 error = nfs_rcvlock(myrep);
781 if (error == EALREADY)
782 return (0);
783 if (error)
784 return (error);
785 /*
786 * Get the next Rpc reply off the socket
787 */
788 nmp->nm_waiters++;
789 error = nfs_receive(myrep, &nam, &mrep, procp);
790 nfs_rcvunlock(nmp);
791 if (error) {
792
793 if (nmp->nm_iflag & NFSMNT_DISMNT) {
794 /*
795 * Oops, we're going away now..
796 */
797 nmp->nm_waiters--;
798 wakeup (&nmp->nm_waiters);
799 return error;
800 }
801 nmp->nm_waiters--;
802 /*
803 * Ignore routing errors on connectionless protocols? ?
804 */
805 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
806 nmp->nm_so->so_error = 0;
807 #ifdef DEBUG
808 printf("nfs_reply: ignoring error %d\n", error);
809 #endif
810 if (myrep->r_flags & R_GETONEREP)
811 return (0);
812 continue;
813 }
814 return (error);
815 }
816 nmp->nm_waiters--;
817 if (nam)
818 m_freem(nam);
819
820 /*
821 * Get the xid and check that it is an rpc reply
822 */
823 md = mrep;
824 dpos = mtod(md, caddr_t);
825 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
826 rxid = *tl++;
827 if (*tl != rpc_reply) {
828 #ifndef NFS_V2_ONLY
829 if (nmp->nm_flag & NFSMNT_NQNFS) {
830 if (nqnfs_callback(nmp, mrep, md, dpos))
831 nfsstats.rpcinvalid++;
832 } else
833 #endif
834 {
835 nfsstats.rpcinvalid++;
836 m_freem(mrep);
837 }
838 nfsmout:
839 if (myrep->r_flags & R_GETONEREP)
840 return (0);
841 continue;
842 }
843
844 /*
845 * Loop through the request list to match up the reply
846 * Iff no match, just drop the datagram
847 */
848 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
849 if (rep->r_mrep == NULL && rxid == rep->r_xid) {
850 /* Found it.. */
851 rep->r_mrep = mrep;
852 rep->r_md = md;
853 rep->r_dpos = dpos;
854 if (nfsrtton) {
855 struct rttl *rt;
856
857 rt = &nfsrtt.rttl[nfsrtt.pos];
858 rt->proc = rep->r_procnum;
859 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
860 rt->sent = nmp->nm_sent;
861 rt->cwnd = nmp->nm_cwnd;
862 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
863 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
864 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx;
865 rt->tstamp = time;
866 if (rep->r_flags & R_TIMING)
867 rt->rtt = rep->r_rtt;
868 else
869 rt->rtt = 1000000;
870 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
871 }
872 /*
873 * Update congestion window.
874 * Do the additive increase of
875 * one rpc/rtt.
876 */
877 if (nmp->nm_cwnd <= nmp->nm_sent) {
878 nmp->nm_cwnd +=
879 (NFS_CWNDSCALE * NFS_CWNDSCALE +
880 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
881 if (nmp->nm_cwnd > NFS_MAXCWND)
882 nmp->nm_cwnd = NFS_MAXCWND;
883 }
884 rep->r_flags &= ~R_SENT;
885 nmp->nm_sent -= NFS_CWNDSCALE;
886 /*
887 * Update rtt using a gain of 0.125 on the mean
888 * and a gain of 0.25 on the deviation.
889 */
890 if (rep->r_flags & R_TIMING) {
891 /*
892 * Since the timer resolution of
893 * NFS_HZ is so course, it can often
894 * result in r_rtt == 0. Since
895 * r_rtt == N means that the actual
896 * rtt is between N+dt and N+2-dt ticks,
897 * add 1.
898 */
899 t1 = rep->r_rtt + 1;
900 t1 -= (NFS_SRTT(rep) >> 3);
901 NFS_SRTT(rep) += t1;
902 if (t1 < 0)
903 t1 = -t1;
904 t1 -= (NFS_SDRTT(rep) >> 2);
905 NFS_SDRTT(rep) += t1;
906 }
907 nmp->nm_timeouts = 0;
908 break;
909 }
910 }
911 /*
912 * If not matched to a request, drop it.
913 * If it's mine, get out.
914 */
915 if (rep == 0) {
916 nfsstats.rpcunexpected++;
917 m_freem(mrep);
918 } else if (rep == myrep) {
919 if (rep->r_mrep == NULL)
920 panic("nfsreply nil");
921 return (0);
922 }
923 if (myrep->r_flags & R_GETONEREP)
924 return (0);
925 }
926 }
927
928 /*
929 * nfs_request - goes something like this
930 * - fill in request struct
931 * - links it into list
932 * - calls nfs_send() for first transmit
933 * - calls nfs_receive() to get reply
934 * - break down rpc header and return with nfs reply pointed to
935 * by mrep or error
936 * nb: always frees up mreq mbuf list
937 */
938 int
939 nfs_request(np, mrest, procnum, procp, cred, mrp, mdp, dposp, rexmitp)
940 struct nfsnode *np;
941 struct mbuf *mrest;
942 int procnum;
943 struct proc *procp;
944 struct ucred *cred;
945 struct mbuf **mrp;
946 struct mbuf **mdp;
947 caddr_t *dposp;
948 int *rexmitp;
949 {
950 struct mbuf *m, *mrep;
951 struct nfsreq *rep;
952 u_int32_t *tl;
953 int i;
954 struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount);
955 struct mbuf *md, *mheadend;
956 char nickv[RPCX_NICKVERF];
957 time_t reqtime, waituntil;
958 caddr_t dpos, cp2;
959 int t1, s, error = 0, mrest_len, auth_len, auth_type;
960 int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0;
961 int verf_len, verf_type;
962 u_int32_t xid;
963 char *auth_str, *verf_str;
964 NFSKERBKEY_T key; /* save session key */
965 struct ucred acred;
966 #ifndef NFS_V2_ONLY
967 int nqlflag, cachable;
968 u_quad_t frev;
969 #endif
970 struct mbuf *mrest_backup = NULL;
971 struct ucred *origcred = NULL; /* XXX: gcc */
972 boolean_t retry_cred = TRUE;
973 boolean_t use_opencred = (np->n_flag & NUSEOPENCRED) != 0;
974
975 if (rexmitp != NULL)
976 *rexmitp = 0;
977
978 tryagain_cred:
979 KASSERT(cred != NULL);
980 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
981 rep->r_nmp = nmp;
982 rep->r_procp = procp;
983 rep->r_procnum = procnum;
984 i = 0;
985 m = mrest;
986 while (m) {
987 i += m->m_len;
988 m = m->m_next;
989 }
990 mrest_len = i;
991
992 /*
993 * Get the RPC header with authorization.
994 */
995 kerbauth:
996 verf_str = auth_str = (char *)0;
997 if (nmp->nm_flag & NFSMNT_KERB) {
998 verf_str = nickv;
999 verf_len = sizeof (nickv);
1000 auth_type = RPCAUTH_KERB4;
1001 memset((caddr_t)key, 0, sizeof (key));
1002 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
1003 &auth_len, verf_str, verf_len)) {
1004 error = nfs_getauth(nmp, rep, cred, &auth_str,
1005 &auth_len, verf_str, &verf_len, key);
1006 if (error) {
1007 free((caddr_t)rep, M_NFSREQ);
1008 m_freem(mrest);
1009 return (error);
1010 }
1011 }
1012 retry_cred = FALSE;
1013 } else {
1014 /* AUTH_UNIX */
1015 uid_t uid;
1016 gid_t gid;
1017
1018 /*
1019 * on the most unix filesystems, permission checks are
1020 * done when the file is open(2)'ed.
1021 * ie. once a file is successfully open'ed,
1022 * following i/o operations never fail with EACCES.
1023 * we try to follow the semantics as far as possible.
1024 *
1025 * note that we expect that the nfs server always grant
1026 * accesses by the file's owner.
1027 */
1028 origcred = cred;
1029 switch (procnum) {
1030 case NFSPROC_READ:
1031 case NFSPROC_WRITE:
1032 case NFSPROC_COMMIT:
1033 uid = np->n_vattr->va_uid;
1034 gid = np->n_vattr->va_gid;
1035 if (cred->cr_uid == uid && cred->cr_gid == gid) {
1036 retry_cred = FALSE;
1037 break;
1038 }
1039 if (use_opencred)
1040 break;
1041 acred.cr_uid = uid;
1042 acred.cr_gid = gid;
1043 acred.cr_ngroups = 0;
1044 acred.cr_ref = 2; /* Just to be safe.. */
1045 cred = &acred;
1046 break;
1047 default:
1048 retry_cred = FALSE;
1049 break;
1050 }
1051 /*
1052 * backup mbuf chain if we can need it later to retry.
1053 *
1054 * XXX maybe we can keep a direct reference to
1055 * mrest without doing m_copym, but it's ...ugly.
1056 */
1057 if (retry_cred)
1058 mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT);
1059 auth_type = RPCAUTH_UNIX;
1060 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ?
1061 nmp->nm_numgrps : cred->cr_ngroups) << 2) +
1062 5 * NFSX_UNSIGNED;
1063 }
1064 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
1065 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
1066 if (auth_str)
1067 free(auth_str, M_TEMP);
1068
1069 /*
1070 * For stream protocols, insert a Sun RPC Record Mark.
1071 */
1072 if (nmp->nm_sotype == SOCK_STREAM) {
1073 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
1074 *mtod(m, u_int32_t *) = htonl(0x80000000 |
1075 (m->m_pkthdr.len - NFSX_UNSIGNED));
1076 }
1077 rep->r_mreq = m;
1078 rep->r_xid = xid;
1079 tryagain:
1080 if (nmp->nm_flag & NFSMNT_SOFT)
1081 rep->r_retry = nmp->nm_retry;
1082 else
1083 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */
1084 rep->r_rtt = rep->r_rexmit = 0;
1085 if (proct[procnum] > 0)
1086 rep->r_flags = R_TIMING;
1087 else
1088 rep->r_flags = 0;
1089 rep->r_mrep = NULL;
1090
1091 /*
1092 * Do the client side RPC.
1093 */
1094 nfsstats.rpcrequests++;
1095 /*
1096 * Chain request into list of outstanding requests. Be sure
1097 * to put it LAST so timer finds oldest requests first.
1098 */
1099 s = splsoftnet();
1100 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1101
1102 /* Get send time for nqnfs */
1103 reqtime = time.tv_sec;
1104
1105 /*
1106 * If backing off another request or avoiding congestion, don't
1107 * send this one now but let timer do it. If not timing a request,
1108 * do it now.
1109 */
1110 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1111 (nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1112 nmp->nm_sent < nmp->nm_cwnd)) {
1113 splx(s);
1114 if (nmp->nm_soflags & PR_CONNREQUIRED)
1115 error = nfs_sndlock(&nmp->nm_iflag, rep);
1116 if (!error) {
1117 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
1118 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, procp);
1119 if (nmp->nm_soflags & PR_CONNREQUIRED)
1120 nfs_sndunlock(&nmp->nm_iflag);
1121 }
1122 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1123 nmp->nm_sent += NFS_CWNDSCALE;
1124 rep->r_flags |= R_SENT;
1125 }
1126 } else {
1127 splx(s);
1128 rep->r_rtt = -1;
1129 }
1130
1131 /*
1132 * Wait for the reply from our send or the timer's.
1133 */
1134 if (!error || error == EPIPE)
1135 error = nfs_reply(rep, procp);
1136
1137 /*
1138 * RPC done, unlink the request.
1139 */
1140 s = splsoftnet();
1141 TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1142 splx(s);
1143
1144 /*
1145 * Decrement the outstanding request count.
1146 */
1147 if (rep->r_flags & R_SENT) {
1148 rep->r_flags &= ~R_SENT; /* paranoia */
1149 nmp->nm_sent -= NFS_CWNDSCALE;
1150 }
1151
1152 if (rexmitp != NULL) {
1153 int rexmit;
1154
1155 if (nmp->nm_sotype != SOCK_DGRAM)
1156 rexmit = (rep->r_flags & R_REXMITTED) != 0;
1157 else
1158 rexmit = rep->r_rexmit;
1159 *rexmitp = rexmit;
1160 }
1161
1162 /*
1163 * If there was a successful reply and a tprintf msg.
1164 * tprintf a response.
1165 */
1166 if (!error && (rep->r_flags & R_TPRINTFMSG))
1167 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1168 "is alive again");
1169 mrep = rep->r_mrep;
1170 md = rep->r_md;
1171 dpos = rep->r_dpos;
1172 if (error)
1173 goto nfsmout;
1174
1175 /*
1176 * break down the rpc header and check if ok
1177 */
1178 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1179 if (*tl++ == rpc_msgdenied) {
1180 if (*tl == rpc_mismatch)
1181 error = EOPNOTSUPP;
1182 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1183 if (!failed_auth) {
1184 failed_auth++;
1185 mheadend->m_next = (struct mbuf *)0;
1186 m_freem(mrep);
1187 m_freem(rep->r_mreq);
1188 goto kerbauth;
1189 } else
1190 error = EAUTH;
1191 } else
1192 error = EACCES;
1193 m_freem(mrep);
1194 goto nfsmout;
1195 }
1196
1197 /*
1198 * Grab any Kerberos verifier, otherwise just throw it away.
1199 */
1200 verf_type = fxdr_unsigned(int, *tl++);
1201 i = fxdr_unsigned(int32_t, *tl);
1202 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1203 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1204 if (error)
1205 goto nfsmout;
1206 } else if (i > 0)
1207 nfsm_adv(nfsm_rndup(i));
1208 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1209 /* 0 == ok */
1210 if (*tl == 0) {
1211 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1212 if (*tl != 0) {
1213 error = fxdr_unsigned(int, *tl);
1214 switch (error) {
1215 case NFSERR_PERM:
1216 error = EPERM;
1217 break;
1218
1219 case NFSERR_NOENT:
1220 error = ENOENT;
1221 break;
1222
1223 case NFSERR_IO:
1224 error = EIO;
1225 break;
1226
1227 case NFSERR_NXIO:
1228 error = ENXIO;
1229 break;
1230
1231 case NFSERR_ACCES:
1232 error = EACCES;
1233 if (!retry_cred)
1234 break;
1235 m_freem(mrep);
1236 m_freem(rep->r_mreq);
1237 FREE(rep, M_NFSREQ);
1238 use_opencred = !use_opencred;
1239 if (mrest_backup == NULL)
1240 /* m_copym failure */
1241 return ENOMEM;
1242 mrest = mrest_backup;
1243 mrest_backup = NULL;
1244 cred = origcred;
1245 error = 0;
1246 retry_cred = FALSE;
1247 goto tryagain_cred;
1248
1249 case NFSERR_EXIST:
1250 error = EEXIST;
1251 break;
1252
1253 case NFSERR_XDEV:
1254 error = EXDEV;
1255 break;
1256
1257 case NFSERR_NODEV:
1258 error = ENODEV;
1259 break;
1260
1261 case NFSERR_NOTDIR:
1262 error = ENOTDIR;
1263 break;
1264
1265 case NFSERR_ISDIR:
1266 error = EISDIR;
1267 break;
1268
1269 case NFSERR_INVAL:
1270 error = EINVAL;
1271 break;
1272
1273 case NFSERR_FBIG:
1274 error = EFBIG;
1275 break;
1276
1277 case NFSERR_NOSPC:
1278 error = ENOSPC;
1279 break;
1280
1281 case NFSERR_ROFS:
1282 error = EROFS;
1283 break;
1284
1285 case NFSERR_MLINK:
1286 error = EMLINK;
1287 break;
1288
1289 case NFSERR_TIMEDOUT:
1290 error = ETIMEDOUT;
1291 break;
1292
1293 case NFSERR_NAMETOL:
1294 error = ENAMETOOLONG;
1295 break;
1296
1297 case NFSERR_NOTEMPTY:
1298 error = ENOTEMPTY;
1299 break;
1300
1301 case NFSERR_DQUOT:
1302 error = EDQUOT;
1303 break;
1304
1305 case NFSERR_STALE:
1306 /*
1307 * If the File Handle was stale, invalidate the
1308 * lookup cache, just in case.
1309 */
1310 error = ESTALE;
1311 cache_purge(NFSTOV(np));
1312 break;
1313
1314 case NFSERR_REMOTE:
1315 error = EREMOTE;
1316 break;
1317
1318 case NFSERR_WFLUSH:
1319 case NFSERR_BADHANDLE:
1320 case NFSERR_NOT_SYNC:
1321 case NFSERR_BAD_COOKIE:
1322 error = EINVAL;
1323 break;
1324
1325 case NFSERR_NOTSUPP:
1326 error = ENOTSUP;
1327 break;
1328
1329 case NFSERR_TOOSMALL:
1330 case NFSERR_SERVERFAULT:
1331 case NFSERR_BADTYPE:
1332 error = EINVAL;
1333 break;
1334
1335 case NFSERR_TRYLATER:
1336 if ((nmp->nm_flag & NFSMNT_NFSV3) == 0)
1337 break;
1338 m_freem(mrep);
1339 error = 0;
1340 waituntil = time.tv_sec + trylater_delay;
1341 while (time.tv_sec < waituntil)
1342 (void) tsleep((caddr_t)&lbolt,
1343 PSOCK, "nqnfstry", 0);
1344 trylater_delay *= NFS_TRYLATERDELMUL;
1345 if (trylater_delay > NFS_TRYLATERDELMAX)
1346 trylater_delay = NFS_TRYLATERDELMAX;
1347 /*
1348 * RFC1813:
1349 * The client should wait and then try
1350 * the request with a new RPC transaction ID.
1351 */
1352 nfs_renewxid(rep);
1353 goto tryagain;
1354
1355 case NFSERR_STALEWRITEVERF:
1356 error = EINVAL;
1357 break;
1358
1359 default:
1360 #ifdef DIAGNOSTIC
1361 printf("Invalid rpc error code %d\n", error);
1362 #endif
1363 error = EINVAL;
1364 break;
1365 }
1366
1367 if (nmp->nm_flag & NFSMNT_NFSV3) {
1368 *mrp = mrep;
1369 *mdp = md;
1370 *dposp = dpos;
1371 error |= NFSERR_RETERR;
1372 } else
1373 m_freem(mrep);
1374 goto nfsmout;
1375 }
1376
1377 /*
1378 * note which credential worked to minimize number of retries.
1379 */
1380 if (use_opencred)
1381 np->n_flag |= NUSEOPENCRED;
1382 else
1383 np->n_flag &= ~NUSEOPENCRED;
1384
1385 #ifndef NFS_V2_ONLY
1386 /*
1387 * For nqnfs, get any lease in reply
1388 */
1389 if (nmp->nm_flag & NFSMNT_NQNFS) {
1390 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1391 if (*tl) {
1392 nqlflag = fxdr_unsigned(int, *tl);
1393 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1394 cachable = fxdr_unsigned(int, *tl++);
1395 reqtime += fxdr_unsigned(int, *tl++);
1396 if (reqtime > time.tv_sec) {
1397 frev = fxdr_hyper(tl);
1398 nqnfs_clientlease(nmp, np, nqlflag,
1399 cachable, reqtime, frev);
1400 }
1401 }
1402 }
1403 #endif
1404 *mrp = mrep;
1405 *mdp = md;
1406 *dposp = dpos;
1407
1408 KASSERT(error == 0);
1409 goto nfsmout;
1410 }
1411 m_freem(mrep);
1412 error = EPROTONOSUPPORT;
1413 nfsmout:
1414 m_freem(rep->r_mreq);
1415 free((caddr_t)rep, M_NFSREQ);
1416 m_freem(mrest_backup);
1417 return (error);
1418 }
1419 #endif /* NFS */
1420
1421 /*
1422 * Generate the rpc reply header
1423 * siz arg. is used to decide if adding a cluster is worthwhile
1424 */
1425 int
1426 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1427 int siz;
1428 struct nfsrv_descript *nd;
1429 struct nfssvc_sock *slp;
1430 int err;
1431 int cache;
1432 u_quad_t *frev;
1433 struct mbuf **mrq;
1434 struct mbuf **mbp;
1435 caddr_t *bposp;
1436 {
1437 u_int32_t *tl;
1438 struct mbuf *mreq;
1439 caddr_t bpos;
1440 struct mbuf *mb;
1441
1442 mreq = m_gethdr(M_WAIT, MT_DATA);
1443 MCLAIM(mreq, &nfs_mowner);
1444 mb = mreq;
1445 /*
1446 * If this is a big reply, use a cluster else
1447 * try and leave leading space for the lower level headers.
1448 */
1449 siz += RPC_REPLYSIZ;
1450 if (siz >= max_datalen) {
1451 m_clget(mreq, M_WAIT);
1452 } else
1453 mreq->m_data += max_hdr;
1454 tl = mtod(mreq, u_int32_t *);
1455 mreq->m_len = 6 * NFSX_UNSIGNED;
1456 bpos = ((caddr_t)tl) + mreq->m_len;
1457 *tl++ = txdr_unsigned(nd->nd_retxid);
1458 *tl++ = rpc_reply;
1459 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1460 *tl++ = rpc_msgdenied;
1461 if (err & NFSERR_AUTHERR) {
1462 *tl++ = rpc_autherr;
1463 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1464 mreq->m_len -= NFSX_UNSIGNED;
1465 bpos -= NFSX_UNSIGNED;
1466 } else {
1467 *tl++ = rpc_mismatch;
1468 *tl++ = txdr_unsigned(RPC_VER2);
1469 *tl = txdr_unsigned(RPC_VER2);
1470 }
1471 } else {
1472 *tl++ = rpc_msgaccepted;
1473
1474 /*
1475 * For Kerberos authentication, we must send the nickname
1476 * verifier back, otherwise just RPCAUTH_NULL.
1477 */
1478 if (nd->nd_flag & ND_KERBFULL) {
1479 struct nfsuid *nuidp;
1480 struct timeval ktvin, ktvout;
1481
1482 LIST_FOREACH(nuidp, NUIDHASH(slp, nd->nd_cr.cr_uid),
1483 nu_hash) {
1484 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1485 (!nd->nd_nam2 || netaddr_match(
1486 NU_NETFAM(nuidp), &nuidp->nu_haddr,
1487 nd->nd_nam2)))
1488 break;
1489 }
1490 if (nuidp) {
1491 ktvin.tv_sec =
1492 txdr_unsigned(nuidp->nu_timestamp.tv_sec
1493 - 1);
1494 ktvin.tv_usec =
1495 txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1496
1497 /*
1498 * Encrypt the timestamp in ecb mode using the
1499 * session key.
1500 */
1501 #ifdef NFSKERB
1502 XXX
1503 #endif
1504
1505 *tl++ = rpc_auth_kerb;
1506 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1507 *tl = ktvout.tv_sec;
1508 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1509 *tl++ = ktvout.tv_usec;
1510 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1511 } else {
1512 *tl++ = 0;
1513 *tl++ = 0;
1514 }
1515 } else {
1516 *tl++ = 0;
1517 *tl++ = 0;
1518 }
1519 switch (err) {
1520 case EPROGUNAVAIL:
1521 *tl = txdr_unsigned(RPC_PROGUNAVAIL);
1522 break;
1523 case EPROGMISMATCH:
1524 *tl = txdr_unsigned(RPC_PROGMISMATCH);
1525 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1526 if (nd->nd_flag & ND_NQNFS) {
1527 *tl++ = txdr_unsigned(3);
1528 *tl = txdr_unsigned(3);
1529 } else {
1530 *tl++ = txdr_unsigned(2);
1531 *tl = txdr_unsigned(3);
1532 }
1533 break;
1534 case EPROCUNAVAIL:
1535 *tl = txdr_unsigned(RPC_PROCUNAVAIL);
1536 break;
1537 case EBADRPC:
1538 *tl = txdr_unsigned(RPC_GARBAGE);
1539 break;
1540 default:
1541 *tl = 0;
1542 if (err != NFSERR_RETVOID) {
1543 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1544 if (err)
1545 *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1546 else
1547 *tl = 0;
1548 }
1549 break;
1550 };
1551 }
1552
1553 /*
1554 * For nqnfs, piggyback lease as requested.
1555 */
1556 if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1557 if (nd->nd_flag & ND_LEASE) {
1558 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1559 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1560 *tl++ = txdr_unsigned(cache);
1561 *tl++ = txdr_unsigned(nd->nd_duration);
1562 txdr_hyper(*frev, tl);
1563 } else {
1564 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1565 *tl = 0;
1566 }
1567 }
1568 if (mrq != NULL)
1569 *mrq = mreq;
1570 *mbp = mb;
1571 *bposp = bpos;
1572 if (err != 0 && err != NFSERR_RETVOID)
1573 nfsstats.srvrpc_errs++;
1574 return (0);
1575 }
1576
1577 /*
1578 * Nfs timer routine
1579 * Scan the nfsreq list and retranmit any requests that have timed out
1580 * To avoid retransmission attempts on STREAM sockets (in the future) make
1581 * sure to set the r_retry field to 0 (implies nm_retry == 0).
1582 */
1583 void
1584 nfs_timer(arg)
1585 void *arg; /* never used */
1586 {
1587 struct nfsreq *rep;
1588 struct mbuf *m;
1589 struct socket *so;
1590 struct nfsmount *nmp;
1591 int timeo;
1592 int s, error;
1593 #ifdef NFSSERVER
1594 struct nfssvc_sock *slp;
1595 static long lasttime = 0;
1596 u_quad_t cur_usec;
1597 #endif
1598
1599 s = splsoftnet();
1600 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
1601 nmp = rep->r_nmp;
1602 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1603 continue;
1604 if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1605 rep->r_flags |= R_SOFTTERM;
1606 continue;
1607 }
1608 if (rep->r_rtt >= 0) {
1609 rep->r_rtt++;
1610 if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1611 timeo = nmp->nm_timeo;
1612 else
1613 timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1614 if (nmp->nm_timeouts > 0)
1615 timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1616 if (rep->r_rtt <= timeo)
1617 continue;
1618 if (nmp->nm_timeouts <
1619 (sizeof(nfs_backoff) / sizeof(nfs_backoff[0])))
1620 nmp->nm_timeouts++;
1621 }
1622 /*
1623 * Check for server not responding
1624 */
1625 if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1626 rep->r_rexmit > nmp->nm_deadthresh) {
1627 nfs_msg(rep->r_procp,
1628 nmp->nm_mountp->mnt_stat.f_mntfromname,
1629 "not responding");
1630 rep->r_flags |= R_TPRINTFMSG;
1631 }
1632 if (rep->r_rexmit >= rep->r_retry) { /* too many */
1633 nfsstats.rpctimeouts++;
1634 rep->r_flags |= R_SOFTTERM;
1635 continue;
1636 }
1637 if (nmp->nm_sotype != SOCK_DGRAM) {
1638 if (++rep->r_rexmit > NFS_MAXREXMIT)
1639 rep->r_rexmit = NFS_MAXREXMIT;
1640 continue;
1641 }
1642 if ((so = nmp->nm_so) == NULL)
1643 continue;
1644
1645 /*
1646 * If there is enough space and the window allows..
1647 * Resend it
1648 * Set r_rtt to -1 in case we fail to send it now.
1649 */
1650 rep->r_rtt = -1;
1651 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1652 ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1653 (rep->r_flags & R_SENT) ||
1654 nmp->nm_sent < nmp->nm_cwnd) &&
1655 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1656 if (so->so_state & SS_ISCONNECTED)
1657 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1658 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1659 else
1660 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m,
1661 nmp->nm_nam, (struct mbuf *)0, (struct proc *)0);
1662 if (error) {
1663 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
1664 #ifdef DEBUG
1665 printf("nfs_timer: ignoring error %d\n",
1666 error);
1667 #endif
1668 so->so_error = 0;
1669 }
1670 } else {
1671 /*
1672 * Iff first send, start timing
1673 * else turn timing off, backoff timer
1674 * and divide congestion window by 2.
1675 */
1676 if (rep->r_flags & R_SENT) {
1677 rep->r_flags &= ~R_TIMING;
1678 if (++rep->r_rexmit > NFS_MAXREXMIT)
1679 rep->r_rexmit = NFS_MAXREXMIT;
1680 nmp->nm_cwnd >>= 1;
1681 if (nmp->nm_cwnd < NFS_CWNDSCALE)
1682 nmp->nm_cwnd = NFS_CWNDSCALE;
1683 nfsstats.rpcretries++;
1684 } else {
1685 rep->r_flags |= R_SENT;
1686 nmp->nm_sent += NFS_CWNDSCALE;
1687 }
1688 rep->r_rtt = 0;
1689 }
1690 }
1691 }
1692
1693 #ifdef NFSSERVER
1694 /*
1695 * Call the nqnfs server timer once a second to handle leases.
1696 */
1697 if (lasttime != time.tv_sec) {
1698 lasttime = time.tv_sec;
1699 nqnfs_serverd();
1700 }
1701
1702 /*
1703 * Scan the write gathering queues for writes that need to be
1704 * completed now.
1705 */
1706 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1707 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
1708 if (LIST_FIRST(&slp->ns_tq) &&
1709 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec)
1710 nfsrv_wakenfsd(slp);
1711 }
1712 #endif /* NFSSERVER */
1713 splx(s);
1714 callout_schedule(&nfs_timer_ch, nfs_ticks);
1715 }
1716
1717 /*ARGSUSED*/
1718 void
1719 nfs_exit(p, v)
1720 struct proc *p;
1721 void *v;
1722 {
1723 struct nfsreq *rp;
1724 int s = splsoftnet();
1725
1726 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
1727 if (rp->r_procp == p)
1728 TAILQ_REMOVE(&nfs_reqq, rp, r_chain);
1729 }
1730 splx(s);
1731 }
1732
1733 /*
1734 * Test for a termination condition pending on the process.
1735 * This is used for NFSMNT_INT mounts.
1736 */
1737 int
1738 nfs_sigintr(nmp, rep, p)
1739 struct nfsmount *nmp;
1740 struct nfsreq *rep;
1741 struct proc *p;
1742 {
1743 sigset_t ss;
1744
1745 if (rep && (rep->r_flags & R_SOFTTERM))
1746 return (EINTR);
1747 if (!(nmp->nm_flag & NFSMNT_INT))
1748 return (0);
1749 if (p) {
1750 sigpending1(p, &ss);
1751 #if 0
1752 sigminusset(&p->p_sigctx.ps_sigignore, &ss);
1753 #endif
1754 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) ||
1755 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) ||
1756 sigismember(&ss, SIGQUIT))
1757 return (EINTR);
1758 }
1759 return (0);
1760 }
1761
1762 /*
1763 * Lock a socket against others.
1764 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1765 * and also to avoid race conditions between the processes with nfs requests
1766 * in progress when a reconnect is necessary.
1767 */
1768 int
1769 nfs_sndlock(flagp, rep)
1770 int *flagp;
1771 struct nfsreq *rep;
1772 {
1773 struct proc *p;
1774 int slpflag = 0, slptimeo = 0;
1775
1776 if (rep) {
1777 p = rep->r_procp;
1778 if (rep->r_nmp->nm_flag & NFSMNT_INT)
1779 slpflag = PCATCH;
1780 } else
1781 p = (struct proc *)0;
1782 while (*flagp & NFSMNT_SNDLOCK) {
1783 if (rep && nfs_sigintr(rep->r_nmp, rep, p))
1784 return (EINTR);
1785 *flagp |= NFSMNT_WANTSND;
1786 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck",
1787 slptimeo);
1788 if (slpflag == PCATCH) {
1789 slpflag = 0;
1790 slptimeo = 2 * hz;
1791 }
1792 }
1793 *flagp |= NFSMNT_SNDLOCK;
1794 return (0);
1795 }
1796
1797 /*
1798 * Unlock the stream socket for others.
1799 */
1800 void
1801 nfs_sndunlock(flagp)
1802 int *flagp;
1803 {
1804
1805 if ((*flagp & NFSMNT_SNDLOCK) == 0)
1806 panic("nfs sndunlock");
1807 *flagp &= ~NFSMNT_SNDLOCK;
1808 if (*flagp & NFSMNT_WANTSND) {
1809 *flagp &= ~NFSMNT_WANTSND;
1810 wakeup((caddr_t)flagp);
1811 }
1812 }
1813
1814 int
1815 nfs_rcvlock(rep)
1816 struct nfsreq *rep;
1817 {
1818 struct nfsmount *nmp = rep->r_nmp;
1819 int *flagp = &nmp->nm_iflag;
1820 int slpflag, slptimeo = 0;
1821 int error = 0;
1822
1823 if (*flagp & NFSMNT_DISMNT)
1824 return EIO;
1825
1826 if (*flagp & NFSMNT_INT)
1827 slpflag = PCATCH;
1828 else
1829 slpflag = 0;
1830 simple_lock(&nmp->nm_slock);
1831 while (*flagp & NFSMNT_RCVLOCK) {
1832 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) {
1833 error = EINTR;
1834 goto quit;
1835 }
1836 *flagp |= NFSMNT_WANTRCV;
1837 nmp->nm_waiters++;
1838 (void) ltsleep(flagp, slpflag | (PZERO - 1), "nfsrcvlk",
1839 slptimeo, &nmp->nm_slock);
1840 nmp->nm_waiters--;
1841 if (*flagp & NFSMNT_DISMNT) {
1842 wakeup(&nmp->nm_waiters);
1843 error = EIO;
1844 goto quit;
1845 }
1846 /* If our reply was received while we were sleeping,
1847 * then just return without taking the lock to avoid a
1848 * situation where a single iod could 'capture' the
1849 * receive lock.
1850 */
1851 if (rep->r_mrep != NULL) {
1852 error = EALREADY;
1853 goto quit;
1854 }
1855 if (slpflag == PCATCH) {
1856 slpflag = 0;
1857 slptimeo = 2 * hz;
1858 }
1859 }
1860 *flagp |= NFSMNT_RCVLOCK;
1861 quit:
1862 simple_unlock(&nmp->nm_slock);
1863 return error;
1864 }
1865
1866 /*
1867 * Unlock the stream socket for others.
1868 */
1869 void
1870 nfs_rcvunlock(nmp)
1871 struct nfsmount *nmp;
1872 {
1873 int *flagp = &nmp->nm_iflag;
1874
1875 simple_lock(&nmp->nm_slock);
1876 if ((*flagp & NFSMNT_RCVLOCK) == 0)
1877 panic("nfs rcvunlock");
1878 *flagp &= ~NFSMNT_RCVLOCK;
1879 if (*flagp & NFSMNT_WANTRCV) {
1880 *flagp &= ~NFSMNT_WANTRCV;
1881 wakeup((caddr_t)flagp);
1882 }
1883 simple_unlock(&nmp->nm_slock);
1884 }
1885
1886 /*
1887 * Parse an RPC request
1888 * - verify it
1889 * - fill in the cred struct.
1890 */
1891 int
1892 nfs_getreq(nd, nfsd, has_header)
1893 struct nfsrv_descript *nd;
1894 struct nfsd *nfsd;
1895 int has_header;
1896 {
1897 int len, i;
1898 u_int32_t *tl;
1899 int32_t t1;
1900 struct uio uio;
1901 struct iovec iov;
1902 caddr_t dpos, cp2, cp;
1903 u_int32_t nfsvers, auth_type;
1904 uid_t nickuid;
1905 int error = 0, nqnfs = 0, ticklen;
1906 struct mbuf *mrep, *md;
1907 struct nfsuid *nuidp;
1908 struct timeval tvin, tvout;
1909
1910 mrep = nd->nd_mrep;
1911 md = nd->nd_md;
1912 dpos = nd->nd_dpos;
1913 if (has_header) {
1914 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1915 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1916 if (*tl++ != rpc_call) {
1917 m_freem(mrep);
1918 return (EBADRPC);
1919 }
1920 } else
1921 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1922 nd->nd_repstat = 0;
1923 nd->nd_flag = 0;
1924 if (*tl++ != rpc_vers) {
1925 nd->nd_repstat = ERPCMISMATCH;
1926 nd->nd_procnum = NFSPROC_NOOP;
1927 return (0);
1928 }
1929 if (*tl != nfs_prog) {
1930 if (*tl == nqnfs_prog)
1931 nqnfs++;
1932 else {
1933 nd->nd_repstat = EPROGUNAVAIL;
1934 nd->nd_procnum = NFSPROC_NOOP;
1935 return (0);
1936 }
1937 }
1938 tl++;
1939 nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1940 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1941 (nfsvers != NQNFS_VER3 && nqnfs)) {
1942 nd->nd_repstat = EPROGMISMATCH;
1943 nd->nd_procnum = NFSPROC_NOOP;
1944 return (0);
1945 }
1946 if (nqnfs)
1947 nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1948 else if (nfsvers == NFS_VER3)
1949 nd->nd_flag = ND_NFSV3;
1950 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1951 if (nd->nd_procnum == NFSPROC_NULL)
1952 return (0);
1953 if (nd->nd_procnum >= NFS_NPROCS ||
1954 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1955 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1956 nd->nd_repstat = EPROCUNAVAIL;
1957 nd->nd_procnum = NFSPROC_NOOP;
1958 return (0);
1959 }
1960 if ((nd->nd_flag & ND_NFSV3) == 0)
1961 nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1962 auth_type = *tl++;
1963 len = fxdr_unsigned(int, *tl++);
1964 if (len < 0 || len > RPCAUTH_MAXSIZ) {
1965 m_freem(mrep);
1966 return (EBADRPC);
1967 }
1968
1969 nd->nd_flag &= ~ND_KERBAUTH;
1970 /*
1971 * Handle auth_unix or auth_kerb.
1972 */
1973 if (auth_type == rpc_auth_unix) {
1974 len = fxdr_unsigned(int, *++tl);
1975 if (len < 0 || len > NFS_MAXNAMLEN) {
1976 m_freem(mrep);
1977 return (EBADRPC);
1978 }
1979 nfsm_adv(nfsm_rndup(len));
1980 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1981 memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred));
1982 nd->nd_cr.cr_ref = 1;
1983 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1984 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1985 len = fxdr_unsigned(int, *tl);
1986 if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1987 m_freem(mrep);
1988 return (EBADRPC);
1989 }
1990 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1991 for (i = 0; i < len; i++)
1992 if (i < NGROUPS)
1993 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1994 else
1995 tl++;
1996 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
1997 if (nd->nd_cr.cr_ngroups > 1)
1998 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1999 len = fxdr_unsigned(int, *++tl);
2000 if (len < 0 || len > RPCAUTH_MAXSIZ) {
2001 m_freem(mrep);
2002 return (EBADRPC);
2003 }
2004 if (len > 0)
2005 nfsm_adv(nfsm_rndup(len));
2006 } else if (auth_type == rpc_auth_kerb) {
2007 switch (fxdr_unsigned(int, *tl++)) {
2008 case RPCAKN_FULLNAME:
2009 ticklen = fxdr_unsigned(int, *tl);
2010 *((u_int32_t *)nfsd->nfsd_authstr) = *tl;
2011 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
2012 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
2013 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
2014 m_freem(mrep);
2015 return (EBADRPC);
2016 }
2017 uio.uio_offset = 0;
2018 uio.uio_iov = &iov;
2019 uio.uio_iovcnt = 1;
2020 uio.uio_segflg = UIO_SYSSPACE;
2021 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
2022 iov.iov_len = RPCAUTH_MAXSIZ - 4;
2023 nfsm_mtouio(&uio, uio.uio_resid);
2024 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2025 if (*tl++ != rpc_auth_kerb ||
2026 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
2027 printf("Bad kerb verifier\n");
2028 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2029 nd->nd_procnum = NFSPROC_NOOP;
2030 return (0);
2031 }
2032 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
2033 tl = (u_int32_t *)cp;
2034 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
2035 printf("Not fullname kerb verifier\n");
2036 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2037 nd->nd_procnum = NFSPROC_NOOP;
2038 return (0);
2039 }
2040 cp += NFSX_UNSIGNED;
2041 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED);
2042 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
2043 nd->nd_flag |= ND_KERBFULL;
2044 nfsd->nfsd_flag |= NFSD_NEEDAUTH;
2045 break;
2046 case RPCAKN_NICKNAME:
2047 if (len != 2 * NFSX_UNSIGNED) {
2048 printf("Kerb nickname short\n");
2049 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
2050 nd->nd_procnum = NFSPROC_NOOP;
2051 return (0);
2052 }
2053 nickuid = fxdr_unsigned(uid_t, *tl);
2054 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2055 if (*tl++ != rpc_auth_kerb ||
2056 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
2057 printf("Kerb nick verifier bad\n");
2058 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
2059 nd->nd_procnum = NFSPROC_NOOP;
2060 return (0);
2061 }
2062 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2063 tvin.tv_sec = *tl++;
2064 tvin.tv_usec = *tl;
2065
2066 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid),
2067 nu_hash) {
2068 if (nuidp->nu_cr.cr_uid == nickuid &&
2069 (!nd->nd_nam2 ||
2070 netaddr_match(NU_NETFAM(nuidp),
2071 &nuidp->nu_haddr, nd->nd_nam2)))
2072 break;
2073 }
2074 if (!nuidp) {
2075 nd->nd_repstat =
2076 (NFSERR_AUTHERR|AUTH_REJECTCRED);
2077 nd->nd_procnum = NFSPROC_NOOP;
2078 return (0);
2079 }
2080
2081 /*
2082 * Now, decrypt the timestamp using the session key
2083 * and validate it.
2084 */
2085 #ifdef NFSKERB
2086 XXX
2087 #endif
2088
2089 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
2090 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
2091 if (nuidp->nu_expire < time.tv_sec ||
2092 nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
2093 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
2094 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
2095 nuidp->nu_expire = 0;
2096 nd->nd_repstat =
2097 (NFSERR_AUTHERR|AUTH_REJECTVERF);
2098 nd->nd_procnum = NFSPROC_NOOP;
2099 return (0);
2100 }
2101 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
2102 nd->nd_flag |= ND_KERBNICK;
2103 };
2104 } else {
2105 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
2106 nd->nd_procnum = NFSPROC_NOOP;
2107 return (0);
2108 }
2109
2110 /*
2111 * For nqnfs, get piggybacked lease request.
2112 */
2113 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
2114 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2115 nd->nd_flag |= fxdr_unsigned(int, *tl);
2116 if (nd->nd_flag & ND_LEASE) {
2117 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2118 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl);
2119 } else
2120 nd->nd_duration = NQ_MINLEASE;
2121 } else
2122 nd->nd_duration = NQ_MINLEASE;
2123 nd->nd_md = md;
2124 nd->nd_dpos = dpos;
2125 return (0);
2126 nfsmout:
2127 return (error);
2128 }
2129
2130 int
2131 nfs_msg(p, server, msg)
2132 struct proc *p;
2133 char *server, *msg;
2134 {
2135 tpr_t tpr;
2136
2137 if (p)
2138 tpr = tprintf_open(p);
2139 else
2140 tpr = NULL;
2141 tprintf(tpr, "nfs server %s: %s\n", server, msg);
2142 tprintf_close(tpr);
2143 return (0);
2144 }
2145
2146 #ifdef NFSSERVER
2147 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
2148 struct nfssvc_sock *, struct proc *,
2149 struct mbuf **)) = {
2150 nfsrv_null,
2151 nfsrv_getattr,
2152 nfsrv_setattr,
2153 nfsrv_lookup,
2154 nfsrv3_access,
2155 nfsrv_readlink,
2156 nfsrv_read,
2157 nfsrv_write,
2158 nfsrv_create,
2159 nfsrv_mkdir,
2160 nfsrv_symlink,
2161 nfsrv_mknod,
2162 nfsrv_remove,
2163 nfsrv_rmdir,
2164 nfsrv_rename,
2165 nfsrv_link,
2166 nfsrv_readdir,
2167 nfsrv_readdirplus,
2168 nfsrv_statfs,
2169 nfsrv_fsinfo,
2170 nfsrv_pathconf,
2171 nfsrv_commit,
2172 nqnfsrv_getlease,
2173 nqnfsrv_vacated,
2174 nfsrv_noop,
2175 nfsrv_noop
2176 };
2177
2178 /*
2179 * Socket upcall routine for the nfsd sockets.
2180 * The caddr_t arg is a pointer to the "struct nfssvc_sock".
2181 * Essentially do as much as possible non-blocking, else punt and it will
2182 * be called with M_WAIT from an nfsd.
2183 */
2184 void
2185 nfsrv_rcv(so, arg, waitflag)
2186 struct socket *so;
2187 caddr_t arg;
2188 int waitflag;
2189 {
2190 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
2191 struct mbuf *m;
2192 struct mbuf *mp, *nam;
2193 struct uio auio;
2194 int flags, error;
2195
2196 if ((slp->ns_flag & SLP_VALID) == 0)
2197 return;
2198 #if 1
2199 /*
2200 * Define this to test for nfsds handling this under heavy load.
2201 */
2202 if (waitflag == M_DONTWAIT) {
2203 slp->ns_flag |= SLP_NEEDQ; goto dorecs;
2204 }
2205 #endif
2206 /* XXX: was NULL, soreceive() requires non-NULL uio->uio_procp */
2207 auio.uio_procp = curproc; /* XXX curproc */
2208 if (so->so_type == SOCK_STREAM) {
2209 /*
2210 * If there are already records on the queue, defer soreceive()
2211 * to an nfsd so that there is feedback to the TCP layer that
2212 * the nfs servers are heavily loaded.
2213 */
2214 if (slp->ns_rec && waitflag == M_DONTWAIT) {
2215 slp->ns_flag |= SLP_NEEDQ;
2216 goto dorecs;
2217 }
2218
2219 /*
2220 * Do soreceive().
2221 */
2222 auio.uio_resid = 1000000000;
2223 flags = MSG_DONTWAIT;
2224 error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
2225 if (error || mp == (struct mbuf *)0) {
2226 if (error == EWOULDBLOCK)
2227 slp->ns_flag |= SLP_NEEDQ;
2228 else
2229 slp->ns_flag |= SLP_DISCONN;
2230 goto dorecs;
2231 }
2232 m = mp;
2233 if (slp->ns_rawend) {
2234 slp->ns_rawend->m_next = m;
2235 slp->ns_cc += 1000000000 - auio.uio_resid;
2236 } else {
2237 slp->ns_raw = m;
2238 slp->ns_cc = 1000000000 - auio.uio_resid;
2239 }
2240 while (m->m_next)
2241 m = m->m_next;
2242 slp->ns_rawend = m;
2243
2244 /*
2245 * Now try and parse record(s) out of the raw stream data.
2246 */
2247 error = nfsrv_getstream(slp, waitflag);
2248 if (error) {
2249 if (error == EPERM)
2250 slp->ns_flag |= SLP_DISCONN;
2251 else
2252 slp->ns_flag |= SLP_NEEDQ;
2253 }
2254 } else {
2255 do {
2256 auio.uio_resid = 1000000000;
2257 flags = MSG_DONTWAIT;
2258 error = (*so->so_receive)(so, &nam, &auio, &mp,
2259 (struct mbuf **)0, &flags);
2260 if (mp) {
2261 if (nam) {
2262 m = nam;
2263 m->m_next = mp;
2264 } else
2265 m = mp;
2266 if (slp->ns_recend)
2267 slp->ns_recend->m_nextpkt = m;
2268 else
2269 slp->ns_rec = m;
2270 slp->ns_recend = m;
2271 m->m_nextpkt = (struct mbuf *)0;
2272 }
2273 if (error) {
2274 if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2275 && error != EWOULDBLOCK) {
2276 slp->ns_flag |= SLP_DISCONN;
2277 goto dorecs;
2278 }
2279 }
2280 } while (mp);
2281 }
2282
2283 /*
2284 * Now try and process the request records, non-blocking.
2285 */
2286 dorecs:
2287 if (waitflag == M_DONTWAIT &&
2288 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
2289 nfsrv_wakenfsd(slp);
2290 }
2291
2292 /*
2293 * Try and extract an RPC request from the mbuf data list received on a
2294 * stream socket. The "waitflag" argument indicates whether or not it
2295 * can sleep.
2296 */
2297 int
2298 nfsrv_getstream(slp, waitflag)
2299 struct nfssvc_sock *slp;
2300 int waitflag;
2301 {
2302 struct mbuf *m, **mpp;
2303 struct mbuf *recm;
2304 u_int32_t recmark;
2305
2306 if (slp->ns_flag & SLP_GETSTREAM)
2307 panic("nfs getstream");
2308 slp->ns_flag |= SLP_GETSTREAM;
2309 for (;;) {
2310 if (slp->ns_reclen == 0) {
2311 if (slp->ns_cc < NFSX_UNSIGNED) {
2312 slp->ns_flag &= ~SLP_GETSTREAM;
2313 return (0);
2314 }
2315 m = slp->ns_raw;
2316 m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark);
2317 m_adj(m, NFSX_UNSIGNED);
2318 slp->ns_cc -= NFSX_UNSIGNED;
2319 recmark = ntohl(recmark);
2320 slp->ns_reclen = recmark & ~0x80000000;
2321 if (recmark & 0x80000000)
2322 slp->ns_flag |= SLP_LASTFRAG;
2323 else
2324 slp->ns_flag &= ~SLP_LASTFRAG;
2325 if (slp->ns_reclen > NFS_MAXPACKET) {
2326 slp->ns_flag &= ~SLP_GETSTREAM;
2327 return (EPERM);
2328 }
2329 }
2330
2331 /*
2332 * Now get the record part.
2333 *
2334 * Note that slp->ns_reclen may be 0. Linux sometimes
2335 * generates 0-length records.
2336 */
2337 if (slp->ns_cc == slp->ns_reclen) {
2338 recm = slp->ns_raw;
2339 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2340 slp->ns_cc = slp->ns_reclen = 0;
2341 } else if (slp->ns_cc > slp->ns_reclen) {
2342 recm = slp->ns_raw;
2343 m = m_split(recm, slp->ns_reclen, waitflag);
2344 if (m == NULL) {
2345 slp->ns_flag &= ~SLP_GETSTREAM;
2346 return (EWOULDBLOCK);
2347 }
2348 m_claimm(recm, &nfs_mowner);
2349 slp->ns_raw = m;
2350 if (m->m_next == NULL)
2351 slp->ns_rawend = m;
2352 slp->ns_cc -= slp->ns_reclen;
2353 slp->ns_reclen = 0;
2354 } else {
2355 slp->ns_flag &= ~SLP_GETSTREAM;
2356 return (0);
2357 }
2358
2359 /*
2360 * Accumulate the fragments into a record.
2361 */
2362 mpp = &slp->ns_frag;
2363 while (*mpp)
2364 mpp = &((*mpp)->m_next);
2365 *mpp = recm;
2366 if (slp->ns_flag & SLP_LASTFRAG) {
2367 if (slp->ns_recend)
2368 slp->ns_recend->m_nextpkt = slp->ns_frag;
2369 else
2370 slp->ns_rec = slp->ns_frag;
2371 slp->ns_recend = slp->ns_frag;
2372 slp->ns_frag = (struct mbuf *)0;
2373 }
2374 }
2375 }
2376
2377 /*
2378 * Parse an RPC header.
2379 */
2380 int
2381 nfsrv_dorec(slp, nfsd, ndp)
2382 struct nfssvc_sock *slp;
2383 struct nfsd *nfsd;
2384 struct nfsrv_descript **ndp;
2385 {
2386 struct mbuf *m, *nam;
2387 struct nfsrv_descript *nd;
2388 int error;
2389
2390 *ndp = NULL;
2391 if ((slp->ns_flag & SLP_VALID) == 0 ||
2392 (m = slp->ns_rec) == (struct mbuf *)0)
2393 return (ENOBUFS);
2394 slp->ns_rec = m->m_nextpkt;
2395 if (slp->ns_rec)
2396 m->m_nextpkt = (struct mbuf *)0;
2397 else
2398 slp->ns_recend = (struct mbuf *)0;
2399 if (m->m_type == MT_SONAME) {
2400 nam = m;
2401 m = m->m_next;
2402 nam->m_next = NULL;
2403 } else
2404 nam = NULL;
2405 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK);
2406 nd->nd_md = nd->nd_mrep = m;
2407 nd->nd_nam2 = nam;
2408 nd->nd_dpos = mtod(m, caddr_t);
2409 error = nfs_getreq(nd, nfsd, TRUE);
2410 if (error) {
2411 m_freem(nam);
2412 pool_put(&nfs_srvdesc_pool, nd);
2413 return (error);
2414 }
2415 *ndp = nd;
2416 nfsd->nfsd_nd = nd;
2417 return (0);
2418 }
2419
2420
2421 /*
2422 * Search for a sleeping nfsd and wake it up.
2423 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2424 * running nfsds will go look for the work in the nfssvc_sock list.
2425 */
2426 void
2427 nfsrv_wakenfsd(slp)
2428 struct nfssvc_sock *slp;
2429 {
2430 struct nfsd *nd;
2431
2432 if ((slp->ns_flag & SLP_VALID) == 0)
2433 return;
2434 simple_lock(&nfsd_slock);
2435 if (slp->ns_flag & SLP_DOREC) {
2436 simple_unlock(&nfsd_slock);
2437 return;
2438 }
2439 nd = SLIST_FIRST(&nfsd_idle_head);
2440 if (nd) {
2441 SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle);
2442 simple_unlock(&nfsd_slock);
2443
2444 KASSERT(nd->nfsd_flag & NFSD_WAITING);
2445 nd->nfsd_flag &= ~NFSD_WAITING;
2446 if (nd->nfsd_slp)
2447 panic("nfsd wakeup");
2448 slp->ns_sref++;
2449 nd->nfsd_slp = slp;
2450 wakeup(nd);
2451 return;
2452 }
2453 slp->ns_flag |= SLP_DOREC;
2454 nfsd_head_flag |= NFSD_CHECKSLP;
2455 TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending);
2456 simple_unlock(&nfsd_slock);
2457 }
2458 #endif /* NFSSERVER */
Cache object: 49e6d8f6da884f839e5264d5031759f7
|