1 /*
2 * Copyright (c) 1989, 1991, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95
37 */
38
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD: releng/5.0/sys/nfsclient/nfs_socket.c 104306 2002-10-01 17:15:53Z jmallett $");
41
42 /*
43 * Socket operations for use by nfs
44 */
45
46 #include "opt_inet6.h"
47
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/mount.h>
55 #include <sys/mutex.h>
56 #include <sys/proc.h>
57 #include <sys/protosw.h>
58 #include <sys/signalvar.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/sysctl.h>
62 #include <sys/syslog.h>
63 #include <sys/vnode.h>
64
65 #include <netinet/in.h>
66 #include <netinet/tcp.h>
67
68 #include <nfs/rpcv2.h>
69 #include <nfs/nfsproto.h>
70 #include <nfsclient/nfs.h>
71 #include <nfs/xdr_subs.h>
72 #include <nfsclient/nfsm_subs.h>
73 #include <nfsclient/nfsmount.h>
74 #include <nfsclient/nfsnode.h>
75
76 #define TRUE 1
77 #define FALSE 0
78
79 /*
80 * Estimate rto for an nfs rpc sent via. an unreliable datagram.
81 * Use the mean and mean deviation of rtt for the appropriate type of rpc
82 * for the frequent rpcs and a default for the others.
83 * The justification for doing "other" this way is that these rpcs
84 * happen so infrequently that timer est. would probably be stale.
85 * Also, since many of these rpcs are
86 * non-idempotent, a conservative timeout is desired.
87 * getattr, lookup - A+2D
88 * read, write - A+4D
89 * other - nm_timeo
90 */
91 #define NFS_RTO(n, t) \
92 ((t) == 0 ? (n)->nm_timeo : \
93 ((t) < 3 ? \
94 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
95 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
96 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
97 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
98
99 /*
100 * Defines which timer to use for the procnum.
101 * 0 - default
102 * 1 - getattr
103 * 2 - lookup
104 * 3 - read
105 * 4 - write
106 */
107 static int proct[NFS_NPROCS] = {
108 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
109 };
110
111 static int nfs_realign_test;
112 static int nfs_realign_count;
113 static int nfs_bufpackets = 4;
114
115 SYSCTL_DECL(_vfs_nfs);
116
117 SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_test, CTLFLAG_RW, &nfs_realign_test, 0, "");
118 SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_count, CTLFLAG_RW, &nfs_realign_count, 0, "");
119 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0, "");
120
121
122 /*
123 * There is a congestion window for outstanding rpcs maintained per mount
124 * point. The cwnd size is adjusted in roughly the way that:
125 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
126 * SIGCOMM '88". ACM, August 1988.
127 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
128 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
129 * of rpcs is in progress.
130 * (The sent count and cwnd are scaled for integer arith.)
131 * Variants of "slow start" were tried and were found to be too much of a
132 * performance hit (ave. rtt 3 times larger),
133 * I suspect due to the large rtt that nfs rpcs have.
134 */
135 #define NFS_CWNDSCALE 256
136 #define NFS_MAXCWND (NFS_CWNDSCALE * 32)
137 #define NFS_NBACKOFF 8
138 static int nfs_backoff[NFS_NBACKOFF] = { 2, 4, 8, 16, 32, 64, 128, 256, };
139 struct callout_handle nfs_timer_handle;
140
141 static int nfs_msg(struct thread *, char *, char *);
142 static int nfs_rcvlock(struct nfsreq *);
143 static void nfs_rcvunlock(struct nfsreq *);
144 static void nfs_realign(struct mbuf **pm, int hsiz);
145 static int nfs_receive(struct nfsreq *rep, struct sockaddr **aname,
146 struct mbuf **mp);
147 static int nfs_reply(struct nfsreq *);
148 static void nfs_softterm(struct nfsreq *rep);
149 static int nfs_reconnect(struct nfsreq *rep);
150
151 /*
152 * Initialize sockets and congestion for a new NFS connection.
153 * We do not free the sockaddr if error.
154 */
155 int
156 nfs_connect(struct nfsmount *nmp, struct nfsreq *rep)
157 {
158 struct socket *so;
159 int s, error, rcvreserve, sndreserve;
160 int pktscale;
161 struct sockaddr *saddr;
162 struct thread *td = &thread0; /* only used for socreate and sobind */
163
164 nmp->nm_so = NULL;
165 saddr = nmp->nm_nam;
166 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
167 nmp->nm_soproto, nmp->nm_mountp->mnt_cred, td);
168 if (error)
169 goto bad;
170 so = nmp->nm_so;
171 nmp->nm_soflags = so->so_proto->pr_flags;
172
173 /*
174 * Some servers require that the client port be a reserved port number.
175 */
176 if (nmp->nm_flag & NFSMNT_RESVPORT) {
177 struct sockopt sopt;
178 int ip, ip2, len;
179 struct sockaddr_in6 ssin;
180 struct sockaddr *sa;
181
182 bzero(&sopt, sizeof sopt);
183 switch(saddr->sa_family) {
184 case AF_INET:
185 sopt.sopt_level = IPPROTO_IP;
186 sopt.sopt_name = IP_PORTRANGE;
187 ip = IP_PORTRANGE_LOW;
188 ip2 = IP_PORTRANGE_DEFAULT;
189 len = sizeof (struct sockaddr_in);
190 break;
191 #ifdef INET6
192 case AF_INET6:
193 sopt.sopt_level = IPPROTO_IPV6;
194 sopt.sopt_name = IPV6_PORTRANGE;
195 ip = IPV6_PORTRANGE_LOW;
196 ip2 = IPV6_PORTRANGE_DEFAULT;
197 len = sizeof (struct sockaddr_in6);
198 break;
199 #endif
200 default:
201 goto noresvport;
202 }
203 sa = (struct sockaddr *)&ssin;
204 bzero(sa, len);
205 sa->sa_len = len;
206 sa->sa_family = saddr->sa_family;
207 sopt.sopt_dir = SOPT_SET;
208 sopt.sopt_val = (void *)&ip;
209 sopt.sopt_valsize = sizeof(ip);
210 error = sosetopt(so, &sopt);
211 if (error)
212 goto bad;
213 error = sobind(so, sa, td);
214 if (error)
215 goto bad;
216 ip = ip2;
217 error = sosetopt(so, &sopt);
218 if (error)
219 goto bad;
220 noresvport: ;
221 }
222
223 /*
224 * Protocols that do not require connections may be optionally left
225 * unconnected for servers that reply from a port other than NFS_PORT.
226 */
227 if (nmp->nm_flag & NFSMNT_NOCONN) {
228 if (nmp->nm_soflags & PR_CONNREQUIRED) {
229 error = ENOTCONN;
230 goto bad;
231 }
232 } else {
233 error = soconnect(so, nmp->nm_nam, td);
234 if (error)
235 goto bad;
236
237 /*
238 * Wait for the connection to complete. Cribbed from the
239 * connect system call but with the wait timing out so
240 * that interruptible mounts don't hang here for a long time.
241 */
242 s = splnet();
243 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
244 (void) tsleep((caddr_t)&so->so_timeo,
245 PSOCK, "nfscon", 2 * hz);
246 if ((so->so_state & SS_ISCONNECTING) &&
247 so->so_error == 0 && rep &&
248 (error = nfs_sigintr(nmp, rep, rep->r_td)) != 0) {
249 so->so_state &= ~SS_ISCONNECTING;
250 splx(s);
251 goto bad;
252 }
253 }
254 if (so->so_error) {
255 error = so->so_error;
256 so->so_error = 0;
257 splx(s);
258 goto bad;
259 }
260 splx(s);
261 }
262 so->so_rcv.sb_timeo = 5 * hz;
263 so->so_snd.sb_timeo = 5 * hz;
264
265 /*
266 * Get buffer reservation size from sysctl, but impose reasonable
267 * limits.
268 */
269 pktscale = nfs_bufpackets;
270 if (pktscale < 2)
271 pktscale = 2;
272 if (pktscale > 64)
273 pktscale = 64;
274
275 if (nmp->nm_sotype == SOCK_DGRAM) {
276 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
277 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
278 NFS_MAXPKTHDR) * pktscale;
279 } else if (nmp->nm_sotype == SOCK_SEQPACKET) {
280 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
281 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
282 NFS_MAXPKTHDR) * pktscale;
283 } else {
284 if (nmp->nm_sotype != SOCK_STREAM)
285 panic("nfscon sotype");
286 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
287 struct sockopt sopt;
288 int val;
289
290 bzero(&sopt, sizeof sopt);
291 sopt.sopt_level = SOL_SOCKET;
292 sopt.sopt_name = SO_KEEPALIVE;
293 sopt.sopt_val = &val;
294 sopt.sopt_valsize = sizeof val;
295 val = 1;
296 sosetopt(so, &sopt);
297 }
298 if (so->so_proto->pr_protocol == IPPROTO_TCP) {
299 struct sockopt sopt;
300 int val;
301
302 bzero(&sopt, sizeof sopt);
303 sopt.sopt_level = IPPROTO_TCP;
304 sopt.sopt_name = TCP_NODELAY;
305 sopt.sopt_val = &val;
306 sopt.sopt_valsize = sizeof val;
307 val = 1;
308 sosetopt(so, &sopt);
309 }
310 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
311 sizeof (u_int32_t)) * pktscale;
312 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
313 sizeof (u_int32_t)) * pktscale;
314 }
315 error = soreserve(so, sndreserve, rcvreserve);
316 if (error)
317 goto bad;
318 so->so_rcv.sb_flags |= SB_NOINTR;
319 so->so_snd.sb_flags |= SB_NOINTR;
320
321 /* Initialize other non-zero congestion variables */
322 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] =
323 nmp->nm_srtt[3] = (NFS_TIMEO << 3);
324 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
325 nmp->nm_sdrtt[3] = 0;
326 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */
327 nmp->nm_sent = 0;
328 nmp->nm_timeouts = 0;
329 return (0);
330
331 bad:
332 nfs_disconnect(nmp);
333 return (error);
334 }
335
336 /*
337 * Reconnect routine:
338 * Called when a connection is broken on a reliable protocol.
339 * - clean up the old socket
340 * - nfs_connect() again
341 * - set R_MUSTRESEND for all outstanding requests on mount point
342 * If this fails the mount point is DEAD!
343 * nb: Must be called with the nfs_sndlock() set on the mount point.
344 */
345 static int
346 nfs_reconnect(struct nfsreq *rep)
347 {
348 struct nfsreq *rp;
349 struct nfsmount *nmp = rep->r_nmp;
350 int error;
351
352 nfs_disconnect(nmp);
353 while ((error = nfs_connect(nmp, rep)) != 0) {
354 if (error == EINTR || error == ERESTART)
355 return (EINTR);
356 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0);
357 }
358
359 /*
360 * Loop through outstanding request list and fix up all requests
361 * on old socket.
362 */
363 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
364 if (rp->r_nmp == nmp)
365 rp->r_flags |= R_MUSTRESEND;
366 }
367 return (0);
368 }
369
370 /*
371 * NFS disconnect. Clean up and unlink.
372 */
373 void
374 nfs_disconnect(struct nfsmount *nmp)
375 {
376 struct socket *so;
377
378 if (nmp->nm_so) {
379 so = nmp->nm_so;
380 nmp->nm_so = NULL;
381 soshutdown(so, 2);
382 soclose(so);
383 }
384 }
385
386 void
387 nfs_safedisconnect(struct nfsmount *nmp)
388 {
389 struct nfsreq dummyreq;
390
391 bzero(&dummyreq, sizeof(dummyreq));
392 dummyreq.r_nmp = nmp;
393 nfs_rcvlock(&dummyreq);
394 nfs_disconnect(nmp);
395 nfs_rcvunlock(&dummyreq);
396 }
397
398 /*
399 * This is the nfs send routine. For connection based socket types, it
400 * must be called with an nfs_sndlock() on the socket.
401 * - return EINTR if the RPC is terminated, 0 otherwise
402 * - set R_MUSTRESEND if the send fails for any reason
403 * - do any cleanup required by recoverable socket errors (?)
404 */
405 int
406 nfs_send(struct socket *so, struct sockaddr *nam, struct mbuf *top,
407 struct nfsreq *rep)
408 {
409 struct sockaddr *sendnam;
410 int error, soflags, flags;
411
412 KASSERT(rep, ("nfs_send: called with rep == NULL"));
413
414 if (rep->r_flags & R_SOFTTERM) {
415 m_freem(top);
416 return (EINTR);
417 }
418 if ((so = rep->r_nmp->nm_so) == NULL) {
419 rep->r_flags |= R_MUSTRESEND;
420 m_freem(top);
421 return (0);
422 }
423 rep->r_flags &= ~R_MUSTRESEND;
424 soflags = rep->r_nmp->nm_soflags;
425
426 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
427 sendnam = NULL;
428 else
429 sendnam = nam;
430 if (so->so_type == SOCK_SEQPACKET)
431 flags = MSG_EOR;
432 else
433 flags = 0;
434
435 error = so->so_proto->pr_usrreqs->pru_sosend(so, sendnam, 0, top, 0,
436 flags, curthread /*XXX*/);
437 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
438 error = 0;
439 rep->r_flags |= R_MUSTRESEND;
440 }
441
442 if (error) {
443 log(LOG_INFO, "nfs send error %d for server %s\n", error,
444 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
445 /*
446 * Deal with errors for the client side.
447 */
448 if (rep->r_flags & R_SOFTTERM)
449 error = EINTR;
450 else
451 rep->r_flags |= R_MUSTRESEND;
452
453 /*
454 * Handle any recoverable (soft) socket errors here. (?)
455 */
456 if (error != EINTR && error != ERESTART &&
457 error != EWOULDBLOCK && error != EPIPE)
458 error = 0;
459 }
460 return (error);
461 }
462
463 /*
464 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
465 * done by soreceive(), but for SOCK_STREAM we must deal with the Record
466 * Mark and consolidate the data into a new mbuf list.
467 * nb: Sometimes TCP passes the data up to soreceive() in long lists of
468 * small mbufs.
469 * For SOCK_STREAM we must be very careful to read an entire record once
470 * we have read any of it, even if the system call has been interrupted.
471 */
472 static int
473 nfs_receive(struct nfsreq *rep, struct sockaddr **aname, struct mbuf **mp)
474 {
475 struct socket *so;
476 struct uio auio;
477 struct iovec aio;
478 struct mbuf *m;
479 struct mbuf *control;
480 u_int32_t len;
481 struct sockaddr **getnam;
482 int error, sotype, rcvflg;
483 struct thread *td = curthread; /* XXX */
484
485 /*
486 * Set up arguments for soreceive()
487 */
488 *mp = NULL;
489 *aname = NULL;
490 sotype = rep->r_nmp->nm_sotype;
491
492 /*
493 * For reliable protocols, lock against other senders/receivers
494 * in case a reconnect is necessary.
495 * For SOCK_STREAM, first get the Record Mark to find out how much
496 * more there is to get.
497 * We must lock the socket against other receivers
498 * until we have an entire rpc request/reply.
499 */
500 if (sotype != SOCK_DGRAM) {
501 error = nfs_sndlock(rep);
502 if (error)
503 return (error);
504 tryagain:
505 /*
506 * Check for fatal errors and resending request.
507 */
508 /*
509 * Ugh: If a reconnect attempt just happened, nm_so
510 * would have changed. NULL indicates a failed
511 * attempt that has essentially shut down this
512 * mount point.
513 */
514 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
515 nfs_sndunlock(rep);
516 return (EINTR);
517 }
518 so = rep->r_nmp->nm_so;
519 if (!so) {
520 error = nfs_reconnect(rep);
521 if (error) {
522 nfs_sndunlock(rep);
523 return (error);
524 }
525 goto tryagain;
526 }
527 while (rep->r_flags & R_MUSTRESEND) {
528 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_TRYWAIT);
529 nfsstats.rpcretries++;
530 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
531 if (error) {
532 if (error == EINTR || error == ERESTART ||
533 (error = nfs_reconnect(rep)) != 0) {
534 nfs_sndunlock(rep);
535 return (error);
536 }
537 goto tryagain;
538 }
539 }
540 nfs_sndunlock(rep);
541 if (sotype == SOCK_STREAM) {
542 aio.iov_base = (caddr_t) &len;
543 aio.iov_len = sizeof(u_int32_t);
544 auio.uio_iov = &aio;
545 auio.uio_iovcnt = 1;
546 auio.uio_segflg = UIO_SYSSPACE;
547 auio.uio_rw = UIO_READ;
548 auio.uio_offset = 0;
549 auio.uio_resid = sizeof(u_int32_t);
550 auio.uio_td = td;
551 do {
552 rcvflg = MSG_WAITALL;
553 error = so->so_proto->pr_usrreqs->pru_soreceive
554 (so, NULL, &auio, NULL, NULL, &rcvflg);
555 if (error == EWOULDBLOCK && rep) {
556 if (rep->r_flags & R_SOFTTERM)
557 return (EINTR);
558 }
559 } while (error == EWOULDBLOCK);
560 if (!error && auio.uio_resid > 0) {
561 /*
562 * Don't log a 0 byte receive; it means
563 * that the socket has been closed, and
564 * can happen during normal operation
565 * (forcible unmount or Solaris server).
566 */
567 if (auio.uio_resid != sizeof (u_int32_t))
568 log(LOG_INFO,
569 "short receive (%d/%d) from nfs server %s\n",
570 (int)(sizeof(u_int32_t) - auio.uio_resid),
571 (int)sizeof(u_int32_t),
572 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
573 error = EPIPE;
574 }
575 if (error)
576 goto errout;
577 len = ntohl(len) & ~0x80000000;
578 /*
579 * This is SERIOUS! We are out of sync with the sender
580 * and forcing a disconnect/reconnect is all I can do.
581 */
582 if (len > NFS_MAXPACKET) {
583 log(LOG_ERR, "%s (%d) from nfs server %s\n",
584 "impossible packet length",
585 len,
586 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
587 error = EFBIG;
588 goto errout;
589 }
590 auio.uio_resid = len;
591 do {
592 rcvflg = MSG_WAITALL;
593 error = so->so_proto->pr_usrreqs->pru_soreceive
594 (so, NULL,
595 &auio, mp, NULL, &rcvflg);
596 } while (error == EWOULDBLOCK || error == EINTR ||
597 error == ERESTART);
598 if (!error && auio.uio_resid > 0) {
599 if (len != auio.uio_resid)
600 log(LOG_INFO,
601 "short receive (%d/%d) from nfs server %s\n",
602 len - auio.uio_resid, len,
603 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
604 error = EPIPE;
605 }
606 } else {
607 /*
608 * NB: Since uio_resid is big, MSG_WAITALL is ignored
609 * and soreceive() will return when it has either a
610 * control msg or a data msg.
611 * We have no use for control msg., but must grab them
612 * and then throw them away so we know what is going
613 * on.
614 */
615 auio.uio_resid = len = 100000000; /* Anything Big */
616 auio.uio_td = td;
617 do {
618 rcvflg = 0;
619 error = so->so_proto->pr_usrreqs->pru_soreceive
620 (so, NULL,
621 &auio, mp, &control, &rcvflg);
622 if (control)
623 m_freem(control);
624 if (error == EWOULDBLOCK && rep) {
625 if (rep->r_flags & R_SOFTTERM)
626 return (EINTR);
627 }
628 } while (error == EWOULDBLOCK ||
629 (!error && *mp == NULL && control));
630 if ((rcvflg & MSG_EOR) == 0)
631 printf("Egad!!\n");
632 if (!error && *mp == NULL)
633 error = EPIPE;
634 len -= auio.uio_resid;
635 }
636 errout:
637 if (error && error != EINTR && error != ERESTART) {
638 m_freem(*mp);
639 *mp = NULL;
640 if (error != EPIPE)
641 log(LOG_INFO,
642 "receive error %d from nfs server %s\n",
643 error,
644 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
645 error = nfs_sndlock(rep);
646 if (!error) {
647 error = nfs_reconnect(rep);
648 if (!error)
649 goto tryagain;
650 else
651 nfs_sndunlock(rep);
652 }
653 }
654 } else {
655 if ((so = rep->r_nmp->nm_so) == NULL)
656 return (EACCES);
657 if (so->so_state & SS_ISCONNECTED)
658 getnam = NULL;
659 else
660 getnam = aname;
661 auio.uio_resid = len = 1000000;
662 auio.uio_td = td;
663 do {
664 rcvflg = 0;
665 error = so->so_proto->pr_usrreqs->pru_soreceive
666 (so, getnam, &auio, mp,
667 NULL, &rcvflg);
668 if (error == EWOULDBLOCK &&
669 (rep->r_flags & R_SOFTTERM))
670 return (EINTR);
671 } while (error == EWOULDBLOCK);
672 len -= auio.uio_resid;
673 }
674 if (error) {
675 m_freem(*mp);
676 *mp = NULL;
677 }
678 /*
679 * Search for any mbufs that are not a multiple of 4 bytes long
680 * or with m_data not longword aligned.
681 * These could cause pointer alignment problems, so copy them to
682 * well aligned mbufs.
683 */
684 nfs_realign(mp, 5 * NFSX_UNSIGNED);
685 return (error);
686 }
687
688 /*
689 * Implement receipt of reply on a socket.
690 * We must search through the list of received datagrams matching them
691 * with outstanding requests using the xid, until ours is found.
692 */
693 /* ARGSUSED */
694 static int
695 nfs_reply(struct nfsreq *myrep)
696 {
697 struct nfsreq *rep;
698 struct nfsmount *nmp = myrep->r_nmp;
699 int32_t t1;
700 struct mbuf *mrep, *md;
701 struct sockaddr *nam;
702 u_int32_t rxid, *tl;
703 caddr_t dpos;
704 int error;
705
706 /*
707 * Loop around until we get our own reply
708 */
709 for (;;) {
710 /*
711 * Lock against other receivers so that I don't get stuck in
712 * sbwait() after someone else has received my reply for me.
713 * Also necessary for connection based protocols to avoid
714 * race conditions during a reconnect.
715 * If nfs_rcvlock() returns EALREADY, that means that
716 * the reply has already been recieved by another
717 * process and we can return immediately. In this
718 * case, the lock is not taken to avoid races with
719 * other processes.
720 */
721 error = nfs_rcvlock(myrep);
722 if (error == EALREADY)
723 return (0);
724 if (error)
725 return (error);
726 /*
727 * Get the next Rpc reply off the socket
728 */
729 error = nfs_receive(myrep, &nam, &mrep);
730 nfs_rcvunlock(myrep);
731 if (error) {
732
733 /*
734 * Ignore routing errors on connectionless protocols??
735 */
736 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
737 nmp->nm_so->so_error = 0;
738 if (myrep->r_flags & R_GETONEREP)
739 return (0);
740 continue;
741 }
742 return (error);
743 }
744 if (nam)
745 FREE(nam, M_SONAME);
746
747 /*
748 * Get the xid and check that it is an rpc reply
749 */
750 md = mrep;
751 dpos = mtod(md, caddr_t);
752 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
753 rxid = *tl++;
754 if (*tl != rpc_reply) {
755 nfsstats.rpcinvalid++;
756 m_freem(mrep);
757 nfsmout:
758 if (myrep->r_flags & R_GETONEREP)
759 return (0);
760 continue;
761 }
762
763 /*
764 * Loop through the request list to match up the reply
765 * Iff no match, just drop the datagram
766 */
767 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
768 if (rep->r_mrep == NULL && rxid == rep->r_xid) {
769 /* Found it.. */
770 rep->r_mrep = mrep;
771 rep->r_md = md;
772 rep->r_dpos = dpos;
773 /*
774 * Update congestion window.
775 * Do the additive increase of
776 * one rpc/rtt.
777 */
778 if (nmp->nm_cwnd <= nmp->nm_sent) {
779 nmp->nm_cwnd +=
780 (NFS_CWNDSCALE * NFS_CWNDSCALE +
781 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
782 if (nmp->nm_cwnd > NFS_MAXCWND)
783 nmp->nm_cwnd = NFS_MAXCWND;
784 }
785 if (rep->r_flags & R_SENT) {
786 rep->r_flags &= ~R_SENT;
787 nmp->nm_sent -= NFS_CWNDSCALE;
788 }
789 /*
790 * Update rtt using a gain of 0.125 on the mean
791 * and a gain of 0.25 on the deviation.
792 */
793 if (rep->r_flags & R_TIMING) {
794 /*
795 * Since the timer resolution of
796 * NFS_HZ is so course, it can often
797 * result in r_rtt == 0. Since
798 * r_rtt == N means that the actual
799 * rtt is between N+dt and N+2-dt ticks,
800 * add 1.
801 */
802 t1 = rep->r_rtt + 1;
803 t1 -= (NFS_SRTT(rep) >> 3);
804 NFS_SRTT(rep) += t1;
805 if (t1 < 0)
806 t1 = -t1;
807 t1 -= (NFS_SDRTT(rep) >> 2);
808 NFS_SDRTT(rep) += t1;
809 }
810 nmp->nm_timeouts = 0;
811 break;
812 }
813 }
814 /*
815 * If not matched to a request, drop it.
816 * If it's mine, get out.
817 */
818 if (rep == 0) {
819 nfsstats.rpcunexpected++;
820 m_freem(mrep);
821 } else if (rep == myrep) {
822 if (rep->r_mrep == NULL)
823 panic("nfsreply nil");
824 return (0);
825 }
826 if (myrep->r_flags & R_GETONEREP)
827 return (0);
828 }
829 }
830
831 /*
832 * nfs_request - goes something like this
833 * - fill in request struct
834 * - links it into list
835 * - calls nfs_send() for first transmit
836 * - calls nfs_receive() to get reply
837 * - break down rpc header and return with nfs reply pointed to
838 * by mrep or error
839 * nb: always frees up mreq mbuf list
840 */
841 /* XXX overloaded before */
842 #define NQ_TRYLATERDEL 15 /* Initial try later delay (sec) */
843
844 int
845 nfs_request(struct vnode *vp, struct mbuf *mrest, int procnum,
846 struct thread *td, struct ucred *cred, struct mbuf **mrp,
847 struct mbuf **mdp, caddr_t *dposp)
848 {
849 struct mbuf *mrep, *m2;
850 struct nfsreq *rep;
851 u_int32_t *tl;
852 int i;
853 struct nfsmount *nmp;
854 struct mbuf *m, *md, *mheadend;
855 time_t waituntil;
856 caddr_t dpos;
857 int s, error = 0, mrest_len, auth_len, auth_type;
858 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0;
859 u_int32_t xid;
860
861 /* Reject requests while attempting a forced unmount. */
862 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) {
863 m_freem(mrest);
864 return (ESTALE);
865 }
866 nmp = VFSTONFS(vp->v_mount);
867 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
868 rep->r_nmp = nmp;
869 rep->r_vp = vp;
870 rep->r_td = td;
871 rep->r_procnum = procnum;
872 mrest_len = m_length(mrest, NULL);
873
874 /*
875 * Get the RPC header with authorization.
876 */
877 auth_type = RPCAUTH_UNIX;
878 if (cred->cr_ngroups < 1)
879 panic("nfsreq nogrps");
880 auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ?
881 nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) +
882 5 * NFSX_UNSIGNED;
883 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
884 mrest, mrest_len, &mheadend, &xid);
885
886 /*
887 * For stream protocols, insert a Sun RPC Record Mark.
888 */
889 if (nmp->nm_sotype == SOCK_STREAM) {
890 M_PREPEND(m, NFSX_UNSIGNED, M_TRYWAIT);
891 *mtod(m, u_int32_t *) = htonl(0x80000000 |
892 (m->m_pkthdr.len - NFSX_UNSIGNED));
893 }
894 rep->r_mreq = m;
895 rep->r_xid = xid;
896 tryagain:
897 if (nmp->nm_flag & NFSMNT_SOFT)
898 rep->r_retry = nmp->nm_retry;
899 else
900 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */
901 rep->r_rtt = rep->r_rexmit = 0;
902 if (proct[procnum] > 0)
903 rep->r_flags = R_TIMING;
904 else
905 rep->r_flags = 0;
906 rep->r_mrep = NULL;
907
908 /*
909 * Do the client side RPC.
910 */
911 nfsstats.rpcrequests++;
912 /*
913 * Chain request into list of outstanding requests. Be sure
914 * to put it LAST so timer finds oldest requests first.
915 */
916 s = splsoftclock();
917 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
918
919 /*
920 * If backing off another request or avoiding congestion, don't
921 * send this one now but let timer do it. If not timing a request,
922 * do it now.
923 */
924 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
925 (nmp->nm_flag & NFSMNT_DUMBTIMR) ||
926 nmp->nm_sent < nmp->nm_cwnd)) {
927 splx(s);
928 if (nmp->nm_soflags & PR_CONNREQUIRED)
929 error = nfs_sndlock(rep);
930 if (!error) {
931 m2 = m_copym(m, 0, M_COPYALL, M_TRYWAIT);
932 error = nfs_send(nmp->nm_so, nmp->nm_nam, m2, rep);
933 if (nmp->nm_soflags & PR_CONNREQUIRED)
934 nfs_sndunlock(rep);
935 }
936 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
937 nmp->nm_sent += NFS_CWNDSCALE;
938 rep->r_flags |= R_SENT;
939 }
940 } else {
941 splx(s);
942 rep->r_rtt = -1;
943 }
944
945 /*
946 * Wait for the reply from our send or the timer's.
947 */
948 if (!error || error == EPIPE)
949 error = nfs_reply(rep);
950
951 /*
952 * RPC done, unlink the request.
953 */
954 s = splsoftclock();
955 TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
956 splx(s);
957
958 /*
959 * Decrement the outstanding request count.
960 */
961 if (rep->r_flags & R_SENT) {
962 rep->r_flags &= ~R_SENT; /* paranoia */
963 nmp->nm_sent -= NFS_CWNDSCALE;
964 }
965
966 /*
967 * If there was a successful reply and a tprintf msg.
968 * tprintf a response.
969 */
970 if (!error && (rep->r_flags & R_TPRINTFMSG))
971 nfs_msg(rep->r_td, nmp->nm_mountp->mnt_stat.f_mntfromname,
972 "is alive again");
973 mrep = rep->r_mrep;
974 md = rep->r_md;
975 dpos = rep->r_dpos;
976 if (error) {
977 m_freem(rep->r_mreq);
978 free((caddr_t)rep, M_NFSREQ);
979 return (error);
980 }
981
982 /*
983 * break down the rpc header and check if ok
984 */
985 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
986 if (*tl++ == rpc_msgdenied) {
987 if (*tl == rpc_mismatch)
988 error = EOPNOTSUPP;
989 else
990 error = EACCES;
991 m_freem(mrep);
992 m_freem(rep->r_mreq);
993 free((caddr_t)rep, M_NFSREQ);
994 return (error);
995 }
996
997 /*
998 * Just throw away any verifyer (ie: kerberos etc).
999 */
1000 i = fxdr_unsigned(int, *tl++); /* verf type */
1001 i = fxdr_unsigned(int32_t, *tl); /* len */
1002 if (i > 0)
1003 nfsm_adv(nfsm_rndup(i));
1004 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1005 /* 0 == ok */
1006 if (*tl == 0) {
1007 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1008 if (*tl != 0) {
1009 error = fxdr_unsigned(int, *tl);
1010 if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1011 error == NFSERR_TRYLATER) {
1012 m_freem(mrep);
1013 error = 0;
1014 waituntil = time_second + trylater_delay;
1015 while (time_second < waituntil)
1016 (void) tsleep((caddr_t)&lbolt,
1017 PSOCK, "nqnfstry", 0);
1018 trylater_delay *= nfs_backoff[trylater_cnt];
1019 if (trylater_cnt < NFS_NBACKOFF - 1)
1020 trylater_cnt++;
1021 goto tryagain;
1022 }
1023
1024 /*
1025 * If the File Handle was stale, invalidate the
1026 * lookup cache, just in case.
1027 */
1028 if (error == ESTALE)
1029 cache_purge(vp);
1030 if (nmp->nm_flag & NFSMNT_NFSV3) {
1031 *mrp = mrep;
1032 *mdp = md;
1033 *dposp = dpos;
1034 error |= NFSERR_RETERR;
1035 } else
1036 m_freem(mrep);
1037 m_freem(rep->r_mreq);
1038 free((caddr_t)rep, M_NFSREQ);
1039 return (error);
1040 }
1041
1042 *mrp = mrep;
1043 *mdp = md;
1044 *dposp = dpos;
1045 m_freem(rep->r_mreq);
1046 FREE((caddr_t)rep, M_NFSREQ);
1047 return (0);
1048 }
1049 m_freem(mrep);
1050 error = EPROTONOSUPPORT;
1051 nfsmout:
1052 m_freem(rep->r_mreq);
1053 free((caddr_t)rep, M_NFSREQ);
1054 return (error);
1055 }
1056
1057 /*
1058 * Nfs timer routine
1059 * Scan the nfsreq list and retranmit any requests that have timed out
1060 * To avoid retransmission attempts on STREAM sockets (in the future) make
1061 * sure to set the r_retry field to 0 (implies nm_retry == 0).
1062 */
1063 void
1064 nfs_timer(void *arg)
1065 {
1066 struct nfsreq *rep;
1067 struct mbuf *m;
1068 struct socket *so;
1069 struct nfsmount *nmp;
1070 int timeo;
1071 int s, error;
1072 struct thread *td;
1073
1074 td = &thread0; /* XXX for credentials, may break if sleep */
1075 s = splnet();
1076 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
1077 nmp = rep->r_nmp;
1078 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1079 continue;
1080 if (nfs_sigintr(nmp, rep, rep->r_td)) {
1081 nfs_softterm(rep);
1082 continue;
1083 }
1084 if (rep->r_rtt >= 0) {
1085 rep->r_rtt++;
1086 if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1087 timeo = nmp->nm_timeo;
1088 else
1089 timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1090 if (nmp->nm_timeouts > 0)
1091 timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1092 if (rep->r_rtt <= timeo)
1093 continue;
1094 if (nmp->nm_timeouts < NFS_NBACKOFF)
1095 nmp->nm_timeouts++;
1096 }
1097 /*
1098 * Check for server not responding
1099 */
1100 if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1101 rep->r_rexmit > nmp->nm_deadthresh) {
1102 char buf[40];
1103 sprintf(buf, "not responding %d > %d",
1104 rep->r_rexmit, nmp->nm_deadthresh);
1105 nfs_msg(rep->r_td,
1106 nmp->nm_mountp->mnt_stat.f_mntfromname,
1107 buf /* "not responding" */);
1108 rep->r_flags |= R_TPRINTFMSG;
1109 }
1110 if (rep->r_rexmit >= rep->r_retry) { /* too many */
1111 nfsstats.rpctimeouts++;
1112 nfs_softterm(rep);
1113 continue;
1114 }
1115 if (nmp->nm_sotype != SOCK_DGRAM) {
1116 if (++rep->r_rexmit > NFS_MAXREXMIT)
1117 rep->r_rexmit = NFS_MAXREXMIT;
1118 continue;
1119 }
1120 if ((so = nmp->nm_so) == NULL)
1121 continue;
1122
1123 /*
1124 * If there is enough space and the window allows..
1125 * Resend it
1126 * Set r_rtt to -1 in case we fail to send it now.
1127 */
1128 rep->r_rtt = -1;
1129 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1130 ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1131 (rep->r_flags & R_SENT) ||
1132 nmp->nm_sent < nmp->nm_cwnd) &&
1133 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1134 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
1135 error = (*so->so_proto->pr_usrreqs->pru_send)
1136 (so, 0, m, NULL, NULL, td);
1137 else
1138 error = (*so->so_proto->pr_usrreqs->pru_send)
1139 (so, 0, m, nmp->nm_nam, NULL, td);
1140 if (error) {
1141 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
1142 so->so_error = 0;
1143 } else {
1144 /*
1145 * Iff first send, start timing
1146 * else turn timing off, backoff timer
1147 * and divide congestion window by 2.
1148 */
1149 if (rep->r_flags & R_SENT) {
1150 rep->r_flags &= ~R_TIMING;
1151 if (++rep->r_rexmit > NFS_MAXREXMIT)
1152 rep->r_rexmit = NFS_MAXREXMIT;
1153 nmp->nm_cwnd >>= 1;
1154 if (nmp->nm_cwnd < NFS_CWNDSCALE)
1155 nmp->nm_cwnd = NFS_CWNDSCALE;
1156 nfsstats.rpcretries++;
1157 } else {
1158 rep->r_flags |= R_SENT;
1159 nmp->nm_sent += NFS_CWNDSCALE;
1160 }
1161 rep->r_rtt = 0;
1162 }
1163 }
1164 }
1165 splx(s);
1166 nfs_timer_handle = timeout(nfs_timer, NULL, nfs_ticks);
1167 }
1168
1169 /*
1170 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1171 * wait for all requests to complete. This is used by forced unmounts
1172 * to terminate any outstanding RPCs.
1173 */
1174 int
1175 nfs_nmcancelreqs(nmp)
1176 struct nfsmount *nmp;
1177 {
1178 struct nfsreq *req;
1179 int i, s;
1180
1181 s = splnet();
1182 TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
1183 if (nmp != req->r_nmp || req->r_mrep != NULL ||
1184 (req->r_flags & R_SOFTTERM))
1185 continue;
1186 nfs_softterm(req);
1187 }
1188 splx(s);
1189
1190 for (i = 0; i < 30; i++) {
1191 s = splnet();
1192 TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
1193 if (nmp == req->r_nmp)
1194 break;
1195 }
1196 splx(s);
1197 if (req == NULL)
1198 return (0);
1199 tsleep(&lbolt, PSOCK, "nfscancel", 0);
1200 }
1201 return (EBUSY);
1202 }
1203
1204 /*
1205 * Flag a request as being about to terminate (due to NFSMNT_INT/NFSMNT_SOFT).
1206 * The nm_send count is decremented now to avoid deadlocks when the process in
1207 * soreceive() hasn't yet managed to send its own request.
1208 */
1209
1210 static void
1211 nfs_softterm(struct nfsreq *rep)
1212 {
1213
1214 rep->r_flags |= R_SOFTTERM;
1215 if (rep->r_flags & R_SENT) {
1216 rep->r_nmp->nm_sent -= NFS_CWNDSCALE;
1217 rep->r_flags &= ~R_SENT;
1218 }
1219 }
1220
1221 /*
1222 * Test for a termination condition pending on the process.
1223 * This is used for NFSMNT_INT mounts.
1224 */
1225 int
1226 nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct thread *td)
1227 {
1228 struct proc *p;
1229 sigset_t tmpset;
1230
1231 if (rep && (rep->r_flags & R_SOFTTERM))
1232 return (EINTR);
1233 /* Terminate all requests while attempting a forced unmount. */
1234 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1235 return (EINTR);
1236 if (!(nmp->nm_flag & NFSMNT_INT))
1237 return (0);
1238 if (td == NULL)
1239 return (0);
1240
1241 p = td->td_proc;
1242 tmpset = p->p_siglist;
1243 SIGSETNAND(tmpset, p->p_sigmask);
1244 SIGSETNAND(tmpset, p->p_sigignore);
1245 if (SIGNOTEMPTY(p->p_siglist) && NFSINT_SIGMASK(tmpset))
1246 return (EINTR);
1247
1248 return (0);
1249 }
1250
1251 /*
1252 * Lock a socket against others.
1253 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1254 * and also to avoid race conditions between the processes with nfs requests
1255 * in progress when a reconnect is necessary.
1256 */
1257 int
1258 nfs_sndlock(struct nfsreq *rep)
1259 {
1260 int *statep = &rep->r_nmp->nm_state;
1261 struct thread *td;
1262 int slpflag = 0, slptimeo = 0;
1263
1264 if (rep) {
1265 td = rep->r_td;
1266 if (rep->r_nmp->nm_flag & NFSMNT_INT)
1267 slpflag = PCATCH;
1268 } else
1269 td = NULL;
1270 while (*statep & NFSSTA_SNDLOCK) {
1271 if (nfs_sigintr(rep->r_nmp, rep, td))
1272 return (EINTR);
1273 *statep |= NFSSTA_WANTSND;
1274 (void) tsleep((caddr_t)statep, slpflag | (PZERO - 1),
1275 "nfsndlck", slptimeo);
1276 if (slpflag == PCATCH) {
1277 slpflag = 0;
1278 slptimeo = 2 * hz;
1279 }
1280 }
1281 *statep |= NFSSTA_SNDLOCK;
1282 return (0);
1283 }
1284
1285 /*
1286 * Unlock the stream socket for others.
1287 */
1288 void
1289 nfs_sndunlock(struct nfsreq *rep)
1290 {
1291 int *statep = &rep->r_nmp->nm_state;
1292
1293 if ((*statep & NFSSTA_SNDLOCK) == 0)
1294 panic("nfs sndunlock");
1295 *statep &= ~NFSSTA_SNDLOCK;
1296 if (*statep & NFSSTA_WANTSND) {
1297 *statep &= ~NFSSTA_WANTSND;
1298 wakeup((caddr_t)statep);
1299 }
1300 }
1301
1302 static int
1303 nfs_rcvlock(struct nfsreq *rep)
1304 {
1305 int *statep = &rep->r_nmp->nm_state;
1306 int slpflag, slptimeo = 0;
1307
1308 if (rep->r_nmp->nm_flag & NFSMNT_INT)
1309 slpflag = PCATCH;
1310 else
1311 slpflag = 0;
1312 while (*statep & NFSSTA_RCVLOCK) {
1313 if (nfs_sigintr(rep->r_nmp, rep, rep->r_td))
1314 return (EINTR);
1315 *statep |= NFSSTA_WANTRCV;
1316 (void) tsleep((caddr_t)statep, slpflag | (PZERO - 1), "nfsrcvlk",
1317 slptimeo);
1318 /*
1319 * If our reply was recieved while we were sleeping,
1320 * then just return without taking the lock to avoid a
1321 * situation where a single iod could 'capture' the
1322 * recieve lock.
1323 */
1324 if (rep->r_mrep != NULL)
1325 return (EALREADY);
1326 if (slpflag == PCATCH) {
1327 slpflag = 0;
1328 slptimeo = 2 * hz;
1329 }
1330 }
1331 /* Always fail if our request has been cancelled. */
1332 if (rep != NULL && (rep->r_flags & R_SOFTTERM))
1333 return (EINTR);
1334 *statep |= NFSSTA_RCVLOCK;
1335 return (0);
1336 }
1337
1338 /*
1339 * Unlock the stream socket for others.
1340 */
1341 static void
1342 nfs_rcvunlock(struct nfsreq *rep)
1343 {
1344 int *statep = &rep->r_nmp->nm_state;
1345
1346 if ((*statep & NFSSTA_RCVLOCK) == 0)
1347 panic("nfs rcvunlock");
1348 *statep &= ~NFSSTA_RCVLOCK;
1349 if (*statep & NFSSTA_WANTRCV) {
1350 *statep &= ~NFSSTA_WANTRCV;
1351 wakeup((caddr_t)statep);
1352 }
1353 }
1354
1355 /*
1356 * nfs_realign:
1357 *
1358 * Check for badly aligned mbuf data and realign by copying the unaligned
1359 * portion of the data into a new mbuf chain and freeing the portions
1360 * of the old chain that were replaced.
1361 *
1362 * We cannot simply realign the data within the existing mbuf chain
1363 * because the underlying buffers may contain other rpc commands and
1364 * we cannot afford to overwrite them.
1365 *
1366 * We would prefer to avoid this situation entirely. The situation does
1367 * not occur with NFS/UDP and is supposed to only occassionally occur
1368 * with TCP. Use vfs.nfs.realign_count and realign_test to check this.
1369 */
1370 static void
1371 nfs_realign(struct mbuf **pm, int hsiz)
1372 {
1373 struct mbuf *m;
1374 struct mbuf *n = NULL;
1375 int off = 0;
1376
1377 ++nfs_realign_test;
1378 while ((m = *pm) != NULL) {
1379 if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) {
1380 MGET(n, M_TRYWAIT, MT_DATA);
1381 if (m->m_len >= MINCLSIZE) {
1382 MCLGET(n, M_TRYWAIT);
1383 }
1384 n->m_len = 0;
1385 break;
1386 }
1387 pm = &m->m_next;
1388 }
1389 /*
1390 * If n is non-NULL, loop on m copying data, then replace the
1391 * portion of the chain that had to be realigned.
1392 */
1393 if (n != NULL) {
1394 ++nfs_realign_count;
1395 while (m) {
1396 m_copyback(n, off, m->m_len, mtod(m, caddr_t));
1397 off += m->m_len;
1398 m = m->m_next;
1399 }
1400 m_freem(*pm);
1401 *pm = n;
1402 }
1403 }
1404
1405
1406 static int
1407 nfs_msg(struct thread *td, char *server, char *msg)
1408 {
1409
1410 tprintf(td ? td->td_proc : NULL, LOG_INFO,
1411 "nfs server %s: %s\n", server, msg);
1412 return (0);
1413 }
Cache object: 8b3c755782aea60c5844c41f064ece95
|