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 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95
33 */
34
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD: src/sys/nfsclient/nfs_socket.c,v 1.114.2.3 2006/04/02 04:15:34 cel Exp $");
37
38 /*
39 * Socket operations for use by nfs
40 */
41
42 #include "opt_inet6.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/mount.h>
51 #include <sys/mutex.h>
52 #include <sys/proc.h>
53 #include <sys/protosw.h>
54 #include <sys/signalvar.h>
55 #include <sys/socket.h>
56 #include <sys/socketvar.h>
57 #include <sys/sysctl.h>
58 #include <sys/syslog.h>
59 #include <sys/vnode.h>
60
61 #include <netinet/in.h>
62 #include <netinet/tcp.h>
63
64 #include <rpc/rpcclnt.h>
65
66 #include <nfs/rpcv2.h>
67 #include <nfs/nfsproto.h>
68 #include <nfsclient/nfs.h>
69 #include <nfs/xdr_subs.h>
70 #include <nfsclient/nfsm_subs.h>
71 #include <nfsclient/nfsmount.h>
72 #include <nfsclient/nfsnode.h>
73
74 #include <nfs4client/nfs4.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 static int nfs_reconnects;
115 static int nfs3_jukebox_delay = 10;
116
117 SYSCTL_DECL(_vfs_nfs);
118
119 SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_test, CTLFLAG_RW, &nfs_realign_test, 0, "");
120 SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_count, CTLFLAG_RW, &nfs_realign_count, 0, "");
121 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0, "");
122 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
123 "number of times the nfs client has had to reconnect");
124 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
125 "number of seconds to delay a retry after receiving EJUKEBOX");
126
127
128 /*
129 * There is a congestion window for outstanding rpcs maintained per mount
130 * point. The cwnd size is adjusted in roughly the way that:
131 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
132 * SIGCOMM '88". ACM, August 1988.
133 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
134 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
135 * of rpcs is in progress.
136 * (The sent count and cwnd are scaled for integer arith.)
137 * Variants of "slow start" were tried and were found to be too much of a
138 * performance hit (ave. rtt 3 times larger),
139 * I suspect due to the large rtt that nfs rpcs have.
140 */
141 #define NFS_CWNDSCALE 256
142 #define NFS_MAXCWND (NFS_CWNDSCALE * 32)
143 #define NFS_NBACKOFF 8
144 static int nfs_backoff[NFS_NBACKOFF] = { 2, 4, 8, 16, 32, 64, 128, 256, };
145 struct callout nfs_callout;
146
147 static int nfs_msg(struct thread *, const char *, const char *, int);
148 static int nfs_rcvlock(struct nfsreq *);
149 static void nfs_rcvunlock(struct nfsreq *);
150 static void nfs_realign(struct mbuf **pm, int hsiz);
151 static int nfs_receive(struct nfsreq *rep, struct sockaddr **aname,
152 struct mbuf **mp);
153 static int nfs_reply(struct nfsreq *);
154 static void nfs_softterm(struct nfsreq *rep);
155 static int nfs_reconnect(struct nfsreq *rep);
156
157 /*
158 * Initialize sockets and congestion for a new NFS connection.
159 * We do not free the sockaddr if error.
160 */
161 int
162 nfs_connect(struct nfsmount *nmp, struct nfsreq *rep)
163 {
164 struct socket *so;
165 int error, rcvreserve, sndreserve;
166 int pktscale;
167 struct sockaddr *saddr;
168 struct thread *td = &thread0; /* only used for socreate and sobind */
169
170 NET_ASSERT_GIANT();
171
172 nmp->nm_so = NULL;
173 saddr = nmp->nm_nam;
174 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
175 nmp->nm_soproto, nmp->nm_mountp->mnt_cred, td);
176 if (error)
177 goto bad;
178 so = nmp->nm_so;
179 nmp->nm_soflags = so->so_proto->pr_flags;
180
181 /*
182 * Some servers require that the client port be a reserved port number.
183 */
184 if (nmp->nm_flag & NFSMNT_RESVPORT) {
185 struct sockopt sopt;
186 int ip, ip2, len;
187 struct sockaddr_in6 ssin;
188 struct sockaddr *sa;
189
190 bzero(&sopt, sizeof sopt);
191 switch(saddr->sa_family) {
192 case AF_INET:
193 sopt.sopt_level = IPPROTO_IP;
194 sopt.sopt_name = IP_PORTRANGE;
195 ip = IP_PORTRANGE_LOW;
196 ip2 = IP_PORTRANGE_DEFAULT;
197 len = sizeof (struct sockaddr_in);
198 break;
199 #ifdef INET6
200 case AF_INET6:
201 sopt.sopt_level = IPPROTO_IPV6;
202 sopt.sopt_name = IPV6_PORTRANGE;
203 ip = IPV6_PORTRANGE_LOW;
204 ip2 = IPV6_PORTRANGE_DEFAULT;
205 len = sizeof (struct sockaddr_in6);
206 break;
207 #endif
208 default:
209 goto noresvport;
210 }
211 sa = (struct sockaddr *)&ssin;
212 bzero(sa, len);
213 sa->sa_len = len;
214 sa->sa_family = saddr->sa_family;
215 sopt.sopt_dir = SOPT_SET;
216 sopt.sopt_val = (void *)&ip;
217 sopt.sopt_valsize = sizeof(ip);
218 error = sosetopt(so, &sopt);
219 if (error)
220 goto bad;
221 error = sobind(so, sa, td);
222 if (error)
223 goto bad;
224 ip = ip2;
225 error = sosetopt(so, &sopt);
226 if (error)
227 goto bad;
228 noresvport: ;
229 }
230
231 /*
232 * Protocols that do not require connections may be optionally left
233 * unconnected for servers that reply from a port other than NFS_PORT.
234 */
235 if (nmp->nm_flag & NFSMNT_NOCONN) {
236 if (nmp->nm_soflags & PR_CONNREQUIRED) {
237 error = ENOTCONN;
238 goto bad;
239 }
240 } else {
241 error = soconnect(so, nmp->nm_nam, td);
242 if (error)
243 goto bad;
244
245 /*
246 * Wait for the connection to complete. Cribbed from the
247 * connect system call but with the wait timing out so
248 * that interruptible mounts don't hang here for a long time.
249 */
250 SOCK_LOCK(so);
251 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
252 (void) msleep(&so->so_timeo, SOCK_MTX(so),
253 PSOCK, "nfscon", 2 * hz);
254 if ((so->so_state & SS_ISCONNECTING) &&
255 so->so_error == 0 && rep &&
256 (error = nfs_sigintr(nmp, rep, rep->r_td)) != 0) {
257 so->so_state &= ~SS_ISCONNECTING;
258 SOCK_UNLOCK(so);
259 goto bad;
260 }
261 }
262 if (so->so_error) {
263 error = so->so_error;
264 so->so_error = 0;
265 SOCK_UNLOCK(so);
266 goto bad;
267 }
268 SOCK_UNLOCK(so);
269 }
270 so->so_rcv.sb_timeo = 12 * hz;
271 so->so_snd.sb_timeo = 5 * hz;
272
273 /*
274 * Get buffer reservation size from sysctl, but impose reasonable
275 * limits.
276 */
277 pktscale = nfs_bufpackets;
278 if (pktscale < 2)
279 pktscale = 2;
280 if (pktscale > 64)
281 pktscale = 64;
282
283 if (nmp->nm_sotype == SOCK_DGRAM) {
284 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
285 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
286 NFS_MAXPKTHDR) * pktscale;
287 } else if (nmp->nm_sotype == SOCK_SEQPACKET) {
288 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
289 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
290 NFS_MAXPKTHDR) * pktscale;
291 } else {
292 if (nmp->nm_sotype != SOCK_STREAM)
293 panic("nfscon sotype");
294 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
295 struct sockopt sopt;
296 int val;
297
298 bzero(&sopt, sizeof sopt);
299 sopt.sopt_dir = SOPT_SET;
300 sopt.sopt_level = SOL_SOCKET;
301 sopt.sopt_name = SO_KEEPALIVE;
302 sopt.sopt_val = &val;
303 sopt.sopt_valsize = sizeof val;
304 val = 1;
305 sosetopt(so, &sopt);
306 }
307 if (so->so_proto->pr_protocol == IPPROTO_TCP) {
308 struct sockopt sopt;
309 int val;
310
311 bzero(&sopt, sizeof sopt);
312 sopt.sopt_dir = SOPT_SET;
313 sopt.sopt_level = IPPROTO_TCP;
314 sopt.sopt_name = TCP_NODELAY;
315 sopt.sopt_val = &val;
316 sopt.sopt_valsize = sizeof val;
317 val = 1;
318 sosetopt(so, &sopt);
319 }
320 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
321 sizeof (u_int32_t)) * pktscale;
322 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
323 sizeof (u_int32_t)) * pktscale;
324 }
325 error = soreserve(so, sndreserve, rcvreserve);
326 if (error)
327 goto bad;
328 SOCKBUF_LOCK(&so->so_rcv);
329 so->so_rcv.sb_flags |= SB_NOINTR;
330 SOCKBUF_UNLOCK(&so->so_rcv);
331 SOCKBUF_LOCK(&so->so_snd);
332 so->so_snd.sb_flags |= SB_NOINTR;
333 SOCKBUF_UNLOCK(&so->so_snd);
334
335 /* Initialize other non-zero congestion variables */
336 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] =
337 nmp->nm_srtt[3] = (NFS_TIMEO << 3);
338 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
339 nmp->nm_sdrtt[3] = 0;
340 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */
341 nmp->nm_sent = 0;
342 nmp->nm_timeouts = 0;
343 return (0);
344
345 bad:
346 nfs_disconnect(nmp);
347 return (error);
348 }
349
350 /*
351 * Reconnect routine:
352 * Called when a connection is broken on a reliable protocol.
353 * - clean up the old socket
354 * - nfs_connect() again
355 * - set R_MUSTRESEND for all outstanding requests on mount point
356 * If this fails the mount point is DEAD!
357 * nb: Must be called with the nfs_sndlock() set on the mount point.
358 */
359 static int
360 nfs_reconnect(struct nfsreq *rep)
361 {
362 struct nfsreq *rp;
363 struct nfsmount *nmp = rep->r_nmp;
364 int error;
365
366 nfs_reconnects++;
367 nfs_disconnect(nmp);
368 while ((error = nfs_connect(nmp, rep)) != 0) {
369 if (error == ERESTART)
370 error = EINTR;
371 if (error == EIO || error == EINTR)
372 return (error);
373 (void) tsleep(&lbolt, PSOCK, "nfscon", 0);
374 }
375
376 /*
377 * Loop through outstanding request list and fix up all requests
378 * on old socket.
379 */
380 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
381 if (rp->r_nmp == nmp)
382 rp->r_flags |= R_MUSTRESEND;
383 }
384 return (0);
385 }
386
387 /*
388 * NFS disconnect. Clean up and unlink.
389 */
390 void
391 nfs_disconnect(struct nfsmount *nmp)
392 {
393 struct socket *so;
394
395 NET_ASSERT_GIANT();
396
397 if (nmp->nm_so) {
398 so = nmp->nm_so;
399 nmp->nm_so = NULL;
400 soshutdown(so, SHUT_RDWR);
401 soclose(so);
402 }
403 }
404
405 void
406 nfs_safedisconnect(struct nfsmount *nmp)
407 {
408 struct nfsreq dummyreq;
409
410 bzero(&dummyreq, sizeof(dummyreq));
411 dummyreq.r_nmp = nmp;
412 nfs_rcvlock(&dummyreq);
413 nfs_disconnect(nmp);
414 nfs_rcvunlock(&dummyreq);
415 }
416
417 /*
418 * This is the nfs send routine. For connection based socket types, it
419 * must be called with an nfs_sndlock() on the socket.
420 * - return EINTR if the RPC is terminated, 0 otherwise
421 * - set R_MUSTRESEND if the send fails for any reason
422 * - do any cleanup required by recoverable socket errors (?)
423 */
424 int
425 nfs_send(struct socket *so, struct sockaddr *nam, struct mbuf *top,
426 struct nfsreq *rep)
427 {
428 struct sockaddr *sendnam;
429 int error, error2, soflags, flags;
430
431 NET_ASSERT_GIANT();
432
433 KASSERT(rep, ("nfs_send: called with rep == NULL"));
434
435 error = nfs_sigintr(rep->r_nmp, rep, rep->r_td);
436 if (error) {
437 m_freem(top);
438 return (error);
439 }
440 if ((so = rep->r_nmp->nm_so) == NULL) {
441 rep->r_flags |= R_MUSTRESEND;
442 m_freem(top);
443 return (0);
444 }
445 rep->r_flags &= ~R_MUSTRESEND;
446 soflags = rep->r_nmp->nm_soflags;
447
448 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
449 sendnam = NULL;
450 else
451 sendnam = nam;
452 if (so->so_type == SOCK_SEQPACKET)
453 flags = MSG_EOR;
454 else
455 flags = 0;
456
457 error = so->so_proto->pr_usrreqs->pru_sosend(so, sendnam, 0, top, 0,
458 flags, curthread /*XXX*/);
459 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
460 error = 0;
461 rep->r_flags |= R_MUSTRESEND;
462 }
463
464 if (error) {
465 /*
466 * Don't report EPIPE errors on nfs sockets.
467 * These can be due to idle tcp mounts which will be closed by
468 * netapp, solaris, etc. if left idle too long.
469 */
470 if (error != EPIPE) {
471 log(LOG_INFO, "nfs send error %d for server %s\n",
472 error,
473 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
474 }
475 /*
476 * Deal with errors for the client side.
477 */
478 error2 = NFS_SIGREP(rep);
479 if (error2)
480 error = error2;
481 else
482 rep->r_flags |= R_MUSTRESEND;
483
484 /*
485 * Handle any recoverable (soft) socket errors here. (?)
486 */
487 if (error != EINTR && error != ERESTART && error != EIO &&
488 error != EWOULDBLOCK && error != EPIPE)
489 error = 0;
490 }
491 return (error);
492 }
493
494 /*
495 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
496 * done by soreceive(), but for SOCK_STREAM we must deal with the Record
497 * Mark and consolidate the data into a new mbuf list.
498 * nb: Sometimes TCP passes the data up to soreceive() in long lists of
499 * small mbufs.
500 * For SOCK_STREAM we must be very careful to read an entire record once
501 * we have read any of it, even if the system call has been interrupted.
502 */
503 static int
504 nfs_receive(struct nfsreq *rep, struct sockaddr **aname, struct mbuf **mp)
505 {
506 struct socket *so;
507 struct uio auio;
508 struct iovec aio;
509 struct mbuf *m;
510 struct mbuf *control;
511 u_int32_t len;
512 struct sockaddr **getnam;
513 int error, error2, sotype, rcvflg;
514 struct thread *td = curthread; /* XXX */
515
516 NET_ASSERT_GIANT();
517
518 /*
519 * Set up arguments for soreceive()
520 */
521 *mp = NULL;
522 *aname = NULL;
523 sotype = rep->r_nmp->nm_sotype;
524
525 /*
526 * For reliable protocols, lock against other senders/receivers
527 * in case a reconnect is necessary.
528 * For SOCK_STREAM, first get the Record Mark to find out how much
529 * more there is to get.
530 * We must lock the socket against other receivers
531 * until we have an entire rpc request/reply.
532 */
533 if (sotype != SOCK_DGRAM) {
534 error = nfs_sndlock(rep);
535 if (error)
536 return (error);
537 tryagain:
538 /*
539 * Check for fatal errors and resending request.
540 */
541 /*
542 * Ugh: If a reconnect attempt just happened, nm_so
543 * would have changed. NULL indicates a failed
544 * attempt that has essentially shut down this
545 * mount point.
546 */
547 if (rep->r_mrep || (error = NFS_SIGREP(rep)) != 0) {
548 nfs_sndunlock(rep);
549 return (error == 0 ? EINTR : error);
550 }
551 so = rep->r_nmp->nm_so;
552 if (!so) {
553 error = nfs_reconnect(rep);
554 if (error) {
555 nfs_sndunlock(rep);
556 return (error);
557 }
558 goto tryagain;
559 }
560 while (rep->r_flags & R_MUSTRESEND) {
561 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_TRYWAIT);
562 nfsstats.rpcretries++;
563 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
564 if (error) {
565 if (error == EINTR || error == ERESTART ||
566 error == EIO ||
567 (error = nfs_reconnect(rep)) != 0) {
568 nfs_sndunlock(rep);
569 return (error);
570 }
571 goto tryagain;
572 }
573 }
574 nfs_sndunlock(rep);
575 if (sotype == SOCK_STREAM) {
576 aio.iov_base = (caddr_t) &len;
577 aio.iov_len = sizeof(u_int32_t);
578 auio.uio_iov = &aio;
579 auio.uio_iovcnt = 1;
580 auio.uio_segflg = UIO_SYSSPACE;
581 auio.uio_rw = UIO_READ;
582 auio.uio_offset = 0;
583 auio.uio_resid = sizeof(u_int32_t);
584 auio.uio_td = td;
585 do {
586 rcvflg = MSG_WAITALL;
587 error = so->so_proto->pr_usrreqs->pru_soreceive
588 (so, NULL, &auio, NULL, NULL, &rcvflg);
589 if (error == EWOULDBLOCK) {
590 error2 = NFS_SIGREP(rep);
591 if (error2)
592 return (error2);
593 }
594 } while (0);
595 if (!error && auio.uio_resid > 0) {
596 /*
597 * Don't log a 0 byte receive; it means
598 * that the socket has been closed, and
599 * can happen during normal operation
600 * (forcible unmount or Solaris server).
601 */
602 if (auio.uio_resid != sizeof (u_int32_t))
603 log(LOG_INFO,
604 "short receive (%d/%d) from nfs server %s\n",
605 (int)(sizeof(u_int32_t) - auio.uio_resid),
606 (int)sizeof(u_int32_t),
607 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
608 error = EPIPE;
609 }
610 if (error)
611 goto errout;
612 len = ntohl(len) & ~0x80000000;
613 /*
614 * This is SERIOUS! We are out of sync with the sender
615 * and forcing a disconnect/reconnect is all I can do.
616 */
617 if (len > NFS_MAXPACKET) {
618 log(LOG_ERR, "%s (%d) from nfs server %s\n",
619 "impossible packet length",
620 len,
621 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
622 error = EFBIG;
623 goto errout;
624 }
625 auio.uio_resid = len;
626 do {
627 rcvflg = MSG_WAITALL;
628 error = so->so_proto->pr_usrreqs->pru_soreceive
629 (so, NULL,
630 &auio, mp, NULL, &rcvflg);
631 } while (0);
632 if (!error && auio.uio_resid > 0) {
633 if (len != auio.uio_resid)
634 log(LOG_INFO,
635 "short receive (%d/%d) from nfs server %s\n",
636 len - auio.uio_resid, len,
637 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
638 error = EPIPE;
639 }
640 } else {
641 /*
642 * NB: Since uio_resid is big, MSG_WAITALL is ignored
643 * and soreceive() will return when it has either a
644 * control msg or a data msg.
645 * We have no use for control msg., but must grab them
646 * and then throw them away so we know what is going
647 * on.
648 */
649 auio.uio_resid = len = 100000000; /* Anything Big */
650 auio.uio_td = td;
651 do {
652 rcvflg = 0;
653 error = so->so_proto->pr_usrreqs->pru_soreceive
654 (so, NULL,
655 &auio, mp, &control, &rcvflg);
656 if (control)
657 m_freem(control);
658 if (error == EWOULDBLOCK && rep) {
659 error2 = NFS_SIGREP(rep);
660 if (error2)
661 return (error2);
662 }
663 } while (!error && *mp == NULL && control);
664 if ((rcvflg & MSG_EOR) == 0)
665 printf("Egad!!\n");
666 if (!error && *mp == NULL)
667 error = EPIPE;
668 len -= auio.uio_resid;
669 }
670 errout:
671 if (error && error != EINTR && error != EIO &&
672 error != ERESTART) {
673 m_freem(*mp);
674 *mp = NULL;
675 if (error != EPIPE && error != EWOULDBLOCK)
676 log(LOG_INFO,
677 "receive error %d from nfs server %s\n",
678 error,
679 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
680 error = nfs_sndlock(rep);
681 if (!error) {
682 error = nfs_reconnect(rep);
683 if (!error)
684 goto tryagain;
685 else
686 nfs_sndunlock(rep);
687 }
688 }
689 } else {
690 /*
691 * We may have failed while rebinding the datagram socket
692 * so attempt a rebind here.
693 */
694 if ((so = rep->r_nmp->nm_so) == NULL) {
695 error = nfs_sndlock(rep);
696 if (!error) {
697 error = nfs_reconnect(rep);
698 nfs_sndunlock(rep);
699 }
700 if (error)
701 return (error);
702 so = rep->r_nmp->nm_so;
703 }
704 if (so->so_state & SS_ISCONNECTED)
705 getnam = NULL;
706 else
707 getnam = aname;
708 auio.uio_resid = len = 1000000;
709 auio.uio_td = td;
710 do {
711 rcvflg = 0;
712 error = so->so_proto->pr_usrreqs->pru_soreceive
713 (so, getnam, &auio, mp,
714 NULL, &rcvflg);
715 if (error) {
716 error2 = NFS_SIGREP(rep);
717 if (error2) {
718 error = error2;
719 goto dgramout;
720 }
721 }
722 if (error) {
723 error2 = nfs_sndlock(rep);
724 if (!error2) {
725 error2 = nfs_reconnect(rep);
726 if (error2)
727 error = error2;
728 else
729 so = rep->r_nmp->nm_so;
730 nfs_sndunlock(rep);
731 } else {
732 error = error2;
733 }
734 }
735 } while (error == EWOULDBLOCK);
736 dgramout:
737 len -= auio.uio_resid;
738 }
739 if (error) {
740 m_freem(*mp);
741 *mp = NULL;
742 }
743 /*
744 * Search for any mbufs that are not a multiple of 4 bytes long
745 * or with m_data not longword aligned.
746 * These could cause pointer alignment problems, so copy them to
747 * well aligned mbufs.
748 */
749 nfs_realign(mp, 5 * NFSX_UNSIGNED);
750 return (error);
751 }
752
753 /*
754 * Implement receipt of reply on a socket.
755 * We must search through the list of received datagrams matching them
756 * with outstanding requests using the xid, until ours is found.
757 */
758 /* ARGSUSED */
759 static int
760 nfs_reply(struct nfsreq *myrep)
761 {
762 struct nfsreq *rep;
763 struct nfsmount *nmp = myrep->r_nmp;
764 int32_t t1;
765 struct mbuf *mrep, *md;
766 struct sockaddr *nam;
767 u_int32_t rxid, *tl;
768 caddr_t dpos;
769 int error;
770
771 /*
772 * Loop around until we get our own reply
773 */
774 for (;;) {
775 /*
776 * Lock against other receivers so that I don't get stuck in
777 * sbwait() after someone else has received my reply for me.
778 * Also necessary for connection based protocols to avoid
779 * race conditions during a reconnect.
780 * If nfs_rcvlock() returns EALREADY, that means that
781 * the reply has already been recieved by another
782 * process and we can return immediately. In this
783 * case, the lock is not taken to avoid races with
784 * other processes.
785 */
786 error = nfs_rcvlock(myrep);
787 if (error == EALREADY)
788 return (0);
789 if (error)
790 return (error);
791 /*
792 * Get the next Rpc reply off the socket
793 */
794 error = nfs_receive(myrep, &nam, &mrep);
795 nfs_rcvunlock(myrep);
796 if (error) {
797
798 /*
799 * Ignore routing errors on connectionless protocols??
800 */
801 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
802 nmp->nm_so->so_error = 0;
803 if (myrep->r_flags & R_GETONEREP)
804 return (0);
805 continue;
806 }
807 return (error);
808 }
809 if (nam)
810 FREE(nam, M_SONAME);
811
812 /*
813 * Get the xid and check that it is an rpc reply
814 */
815 md = mrep;
816 dpos = mtod(md, caddr_t);
817 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
818 rxid = *tl++;
819 if (*tl != rpc_reply) {
820 nfsstats.rpcinvalid++;
821 m_freem(mrep);
822 nfsmout:
823 if (myrep->r_flags & R_GETONEREP)
824 return (0);
825 continue;
826 }
827
828 /*
829 * Loop through the request list to match up the reply
830 * Iff no match, just drop the datagram
831 */
832 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
833 if (rep->r_mrep == NULL && rxid == rep->r_xid) {
834 /* Found it.. */
835 rep->r_mrep = mrep;
836 rep->r_md = md;
837 rep->r_dpos = dpos;
838 /*
839 * Update congestion window.
840 * Do the additive increase of
841 * one rpc/rtt.
842 */
843 if (nmp->nm_cwnd <= nmp->nm_sent) {
844 nmp->nm_cwnd +=
845 (NFS_CWNDSCALE * NFS_CWNDSCALE +
846 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
847 if (nmp->nm_cwnd > NFS_MAXCWND)
848 nmp->nm_cwnd = NFS_MAXCWND;
849 }
850 if (rep->r_flags & R_SENT) {
851 rep->r_flags &= ~R_SENT;
852 nmp->nm_sent -= NFS_CWNDSCALE;
853 }
854 /*
855 * Update rtt using a gain of 0.125 on the mean
856 * and a gain of 0.25 on the deviation.
857 */
858 if (rep->r_flags & R_TIMING) {
859 /*
860 * Since the timer resolution of
861 * NFS_HZ is so course, it can often
862 * result in r_rtt == 0. Since
863 * r_rtt == N means that the actual
864 * rtt is between N+dt and N+2-dt ticks,
865 * add 1.
866 */
867 t1 = rep->r_rtt + 1;
868 t1 -= (NFS_SRTT(rep) >> 3);
869 NFS_SRTT(rep) += t1;
870 if (t1 < 0)
871 t1 = -t1;
872 t1 -= (NFS_SDRTT(rep) >> 2);
873 NFS_SDRTT(rep) += t1;
874 }
875 nmp->nm_timeouts = 0;
876 break;
877 }
878 }
879 /*
880 * If not matched to a request, drop it.
881 * If it's mine, get out.
882 */
883 if (rep == 0) {
884 nfsstats.rpcunexpected++;
885 m_freem(mrep);
886 } else if (rep == myrep) {
887 if (rep->r_mrep == NULL)
888 panic("nfsreply nil");
889 return (0);
890 }
891 if (myrep->r_flags & R_GETONEREP)
892 return (0);
893 }
894 }
895
896 /*
897 * nfs_request - goes something like this
898 * - fill in request struct
899 * - links it into list
900 * - calls nfs_send() for first transmit
901 * - calls nfs_receive() to get reply
902 * - break down rpc header and return with nfs reply pointed to
903 * by mrep or error
904 * nb: always frees up mreq mbuf list
905 */
906 int
907 nfs_request(struct vnode *vp, struct mbuf *mrest, int procnum,
908 struct thread *td, struct ucred *cred, struct mbuf **mrp,
909 struct mbuf **mdp, caddr_t *dposp)
910 {
911 struct mbuf *mrep, *m2;
912 struct nfsreq *rep;
913 u_int32_t *tl;
914 int i;
915 struct nfsmount *nmp;
916 struct mbuf *m, *md, *mheadend;
917 time_t waituntil;
918 caddr_t dpos;
919 int s, error = 0, mrest_len, auth_len, auth_type;
920 struct timeval now;
921 u_int32_t xid;
922
923 /* Reject requests while attempting a forced unmount. */
924 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) {
925 m_freem(mrest);
926 return (ESTALE);
927 }
928 nmp = VFSTONFS(vp->v_mount);
929 if ((nmp->nm_flag & NFSMNT_NFSV4) != 0)
930 return nfs4_request(vp, mrest, procnum, td, cred, mrp, mdp, dposp);
931 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
932 rep->r_nmp = nmp;
933 rep->r_vp = vp;
934 rep->r_td = td;
935 rep->r_procnum = procnum;
936
937 getmicrouptime(&now);
938 rep->r_lastmsg = now.tv_sec -
939 ((nmp->nm_tprintf_delay) - (nmp->nm_tprintf_initial_delay));
940 mrest_len = m_length(mrest, NULL);
941
942 /*
943 * Get the RPC header with authorization.
944 */
945 auth_type = RPCAUTH_UNIX;
946 if (cred->cr_ngroups < 1)
947 panic("nfsreq nogrps");
948 auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ?
949 nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) +
950 5 * NFSX_UNSIGNED;
951 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
952 mrest, mrest_len, &mheadend, &xid);
953
954 /*
955 * For stream protocols, insert a Sun RPC Record Mark.
956 */
957 if (nmp->nm_sotype == SOCK_STREAM) {
958 M_PREPEND(m, NFSX_UNSIGNED, M_TRYWAIT);
959 *mtod(m, u_int32_t *) = htonl(0x80000000 |
960 (m->m_pkthdr.len - NFSX_UNSIGNED));
961 }
962 rep->r_mreq = m;
963 rep->r_xid = xid;
964 tryagain:
965 if (nmp->nm_flag & NFSMNT_SOFT)
966 rep->r_retry = nmp->nm_retry;
967 else
968 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */
969 rep->r_rtt = rep->r_rexmit = 0;
970 if (proct[procnum] > 0)
971 rep->r_flags = R_TIMING;
972 else
973 rep->r_flags = 0;
974 rep->r_mrep = NULL;
975
976 /*
977 * Do the client side RPC.
978 */
979 nfsstats.rpcrequests++;
980 /*
981 * Chain request into list of outstanding requests. Be sure
982 * to put it LAST so timer finds oldest requests first.
983 */
984 s = splsoftclock();
985 if (TAILQ_EMPTY(&nfs_reqq))
986 callout_reset(&nfs_callout, nfs_ticks, nfs_timer, NULL);
987 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
988
989 /*
990 * If backing off another request or avoiding congestion, don't
991 * send this one now but let timer do it. If not timing a request,
992 * do it now.
993 */
994 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
995 (nmp->nm_flag & NFSMNT_DUMBTIMR) ||
996 nmp->nm_sent < nmp->nm_cwnd)) {
997 splx(s);
998 error = nfs_sndlock(rep);
999 if (!error) {
1000 m2 = m_copym(m, 0, M_COPYALL, M_TRYWAIT);
1001 error = nfs_send(nmp->nm_so, nmp->nm_nam, m2, rep);
1002 nfs_sndunlock(rep);
1003 }
1004 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1005 nmp->nm_sent += NFS_CWNDSCALE;
1006 rep->r_flags |= R_SENT;
1007 }
1008 } else {
1009 splx(s);
1010 rep->r_rtt = -1;
1011 }
1012
1013 /*
1014 * Wait for the reply from our send or the timer's.
1015 */
1016 if (!error || error == EPIPE)
1017 error = nfs_reply(rep);
1018
1019 /*
1020 * RPC done, unlink the request.
1021 */
1022 s = splsoftclock();
1023 TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1024 if (TAILQ_EMPTY(&nfs_reqq))
1025 callout_stop(&nfs_callout);
1026 splx(s);
1027
1028 /*
1029 * Decrement the outstanding request count.
1030 */
1031 if (rep->r_flags & R_SENT) {
1032 rep->r_flags &= ~R_SENT; /* paranoia */
1033 nmp->nm_sent -= NFS_CWNDSCALE;
1034 }
1035
1036 /*
1037 * If there was a successful reply and a tprintf msg.
1038 * tprintf a response.
1039 */
1040 if (!error)
1041 nfs_up(rep, nmp, rep->r_td, "is alive again", NFSSTA_TIMEO);
1042 mrep = rep->r_mrep;
1043 md = rep->r_md;
1044 dpos = rep->r_dpos;
1045 if (error) {
1046 m_freem(rep->r_mreq);
1047 free((caddr_t)rep, M_NFSREQ);
1048 return (error);
1049 }
1050
1051 /*
1052 * break down the rpc header and check if ok
1053 */
1054 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
1055 if (*tl++ == rpc_msgdenied) {
1056 if (*tl == rpc_mismatch)
1057 error = EOPNOTSUPP;
1058 else
1059 error = EACCES;
1060 m_freem(mrep);
1061 m_freem(rep->r_mreq);
1062 free((caddr_t)rep, M_NFSREQ);
1063 return (error);
1064 }
1065
1066 /*
1067 * Just throw away any verifyer (ie: kerberos etc).
1068 */
1069 i = fxdr_unsigned(int, *tl++); /* verf type */
1070 i = fxdr_unsigned(int32_t, *tl); /* len */
1071 if (i > 0)
1072 nfsm_adv(nfsm_rndup(i));
1073 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1074 /* 0 == ok */
1075 if (*tl == 0) {
1076 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1077 if (*tl != 0) {
1078 error = fxdr_unsigned(int, *tl);
1079 if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1080 error == NFSERR_TRYLATER) {
1081 m_freem(mrep);
1082 error = 0;
1083 waituntil = time_second + nfs3_jukebox_delay;
1084 while (time_second < waituntil)
1085 (void) tsleep(&lbolt,
1086 PSOCK, "nqnfstry", 0);
1087 goto tryagain;
1088 }
1089
1090 /*
1091 * If the File Handle was stale, invalidate the
1092 * lookup cache, just in case.
1093 */
1094 if (error == ESTALE)
1095 cache_purge(vp);
1096 if (nmp->nm_flag & NFSMNT_NFSV3) {
1097 *mrp = mrep;
1098 *mdp = md;
1099 *dposp = dpos;
1100 error |= NFSERR_RETERR;
1101 } else
1102 m_freem(mrep);
1103 m_freem(rep->r_mreq);
1104 free((caddr_t)rep, M_NFSREQ);
1105 return (error);
1106 }
1107
1108 *mrp = mrep;
1109 *mdp = md;
1110 *dposp = dpos;
1111 m_freem(rep->r_mreq);
1112 FREE((caddr_t)rep, M_NFSREQ);
1113 return (0);
1114 }
1115 m_freem(mrep);
1116 error = EPROTONOSUPPORT;
1117 nfsmout:
1118 m_freem(rep->r_mreq);
1119 free((caddr_t)rep, M_NFSREQ);
1120 return (error);
1121 }
1122
1123 /*
1124 * Nfs timer routine
1125 * Scan the nfsreq list and retranmit any requests that have timed out
1126 * To avoid retransmission attempts on STREAM sockets (in the future) make
1127 * sure to set the r_retry field to 0 (implies nm_retry == 0).
1128 */
1129 void
1130 nfs_timer(void *arg)
1131 {
1132 struct nfsreq *rep;
1133 struct mbuf *m;
1134 struct socket *so;
1135 struct nfsmount *nmp;
1136 int timeo;
1137 int s, error;
1138 struct timeval now;
1139
1140 getmicrouptime(&now);
1141 s = splnet();
1142 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
1143 nmp = rep->r_nmp;
1144 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1145 continue;
1146 if (nfs_sigintr(nmp, rep, rep->r_td))
1147 continue;
1148 if (nmp->nm_tprintf_initial_delay != 0 &&
1149 (rep->r_rexmit > 2 || (rep->r_flags & R_RESENDERR)) &&
1150 rep->r_lastmsg + nmp->nm_tprintf_delay < now.tv_sec) {
1151 rep->r_lastmsg = now.tv_sec;
1152 nfs_down(rep, nmp, rep->r_td, "not responding",
1153 0, NFSSTA_TIMEO);
1154 #if 0
1155 if (!(nmp->nm_state & NFSSTA_MOUNTED)) {
1156 /* we're not yet completely mounted and */
1157 /* we can't complete an RPC, so we fail */
1158 nfsstats.rpctimeouts++;
1159 nfs_softterm(rep);
1160 continue;
1161 }
1162 #endif
1163 }
1164 if (rep->r_rtt >= 0) {
1165 rep->r_rtt++;
1166 if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1167 timeo = nmp->nm_timeo;
1168 else
1169 timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1170 if (nmp->nm_timeouts > 0)
1171 timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1172 if (rep->r_rtt <= timeo)
1173 continue;
1174 if (nmp->nm_timeouts < NFS_NBACKOFF)
1175 nmp->nm_timeouts++;
1176 }
1177 if (rep->r_rexmit >= rep->r_retry) { /* too many */
1178 nfsstats.rpctimeouts++;
1179 nfs_softterm(rep);
1180 continue;
1181 }
1182 if (nmp->nm_sotype != SOCK_DGRAM) {
1183 if (++rep->r_rexmit > NFS_MAXREXMIT)
1184 rep->r_rexmit = NFS_MAXREXMIT;
1185 continue;
1186 }
1187 if ((so = nmp->nm_so) == NULL)
1188 continue;
1189
1190 /*
1191 * If there is enough space and the window allows..
1192 * Resend it
1193 * Set r_rtt to -1 in case we fail to send it now.
1194 */
1195 rep->r_rtt = -1;
1196 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1197 ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1198 (rep->r_flags & R_SENT) ||
1199 nmp->nm_sent < nmp->nm_cwnd) &&
1200 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1201 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
1202 error = (*so->so_proto->pr_usrreqs->pru_send)
1203 (so, 0, m, NULL, NULL, curthread);
1204 else
1205 error = (*so->so_proto->pr_usrreqs->pru_send)
1206 (so, 0, m, nmp->nm_nam, NULL, curthread);
1207 if (error) {
1208 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
1209 so->so_error = 0;
1210 rep->r_flags |= R_RESENDERR;
1211 } else {
1212 /*
1213 * Iff first send, start timing
1214 * else turn timing off, backoff timer
1215 * and divide congestion window by 2.
1216 */
1217 rep->r_flags &= ~R_RESENDERR;
1218 if (rep->r_flags & R_SENT) {
1219 rep->r_flags &= ~R_TIMING;
1220 if (++rep->r_rexmit > NFS_MAXREXMIT)
1221 rep->r_rexmit = NFS_MAXREXMIT;
1222 nmp->nm_cwnd >>= 1;
1223 if (nmp->nm_cwnd < NFS_CWNDSCALE)
1224 nmp->nm_cwnd = NFS_CWNDSCALE;
1225 nfsstats.rpcretries++;
1226 } else {
1227 rep->r_flags |= R_SENT;
1228 nmp->nm_sent += NFS_CWNDSCALE;
1229 }
1230 rep->r_rtt = 0;
1231 }
1232 }
1233 }
1234 splx(s);
1235 callout_reset(&nfs_callout, nfs_ticks, nfs_timer, NULL);
1236 }
1237
1238 /*
1239 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1240 * wait for all requests to complete. This is used by forced unmounts
1241 * to terminate any outstanding RPCs.
1242 */
1243 int
1244 nfs_nmcancelreqs(nmp)
1245 struct nfsmount *nmp;
1246 {
1247 struct nfsreq *req;
1248 int i, s;
1249
1250 s = splnet();
1251 TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
1252 if (nmp != req->r_nmp || req->r_mrep != NULL ||
1253 (req->r_flags & R_SOFTTERM))
1254 continue;
1255 nfs_softterm(req);
1256 }
1257 splx(s);
1258
1259 for (i = 0; i < 30; i++) {
1260 s = splnet();
1261 TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
1262 if (nmp == req->r_nmp)
1263 break;
1264 }
1265 splx(s);
1266 if (req == NULL)
1267 return (0);
1268 tsleep(&lbolt, PSOCK, "nfscancel", 0);
1269 }
1270 return (EBUSY);
1271 }
1272
1273 /*
1274 * Flag a request as being about to terminate (due to NFSMNT_INT/NFSMNT_SOFT).
1275 * The nm_send count is decremented now to avoid deadlocks when the process in
1276 * soreceive() hasn't yet managed to send its own request.
1277 */
1278
1279 static void
1280 nfs_softterm(struct nfsreq *rep)
1281 {
1282
1283 rep->r_flags |= R_SOFTTERM;
1284 if (rep->r_flags & R_SENT) {
1285 rep->r_nmp->nm_sent -= NFS_CWNDSCALE;
1286 rep->r_flags &= ~R_SENT;
1287 }
1288 }
1289
1290 /*
1291 * Test for a termination condition pending on the process.
1292 * This is used for NFSMNT_INT mounts.
1293 */
1294 int
1295 nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct thread *td)
1296 {
1297 struct proc *p;
1298 sigset_t tmpset;
1299
1300 if ((nmp->nm_flag & NFSMNT_NFSV4) != 0)
1301 return nfs4_sigintr(nmp, rep, td);
1302 if (rep && (rep->r_flags & R_SOFTTERM))
1303 return (EIO);
1304 /* Terminate all requests while attempting a forced unmount. */
1305 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1306 return (EIO);
1307 if (!(nmp->nm_flag & NFSMNT_INT))
1308 return (0);
1309 if (td == NULL)
1310 return (0);
1311
1312 p = td->td_proc;
1313 PROC_LOCK(p);
1314 tmpset = p->p_siglist;
1315 SIGSETNAND(tmpset, td->td_sigmask);
1316 mtx_lock(&p->p_sigacts->ps_mtx);
1317 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1318 mtx_unlock(&p->p_sigacts->ps_mtx);
1319 if (SIGNOTEMPTY(p->p_siglist) && NFSINT_SIGMASK(tmpset)) {
1320 PROC_UNLOCK(p);
1321 return (EINTR);
1322 }
1323 PROC_UNLOCK(p);
1324
1325 return (0);
1326 }
1327
1328 /*
1329 * Lock a socket against others.
1330 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1331 * and also to avoid race conditions between the processes with nfs requests
1332 * in progress when a reconnect is necessary.
1333 */
1334 int
1335 nfs_sndlock(struct nfsreq *rep)
1336 {
1337 int *statep = &rep->r_nmp->nm_state;
1338 struct thread *td;
1339 int error, slpflag = 0, slptimeo = 0;
1340
1341 td = rep->r_td;
1342 if (rep->r_nmp->nm_flag & NFSMNT_INT)
1343 slpflag = PCATCH;
1344 while (*statep & NFSSTA_SNDLOCK) {
1345 error = nfs_sigintr(rep->r_nmp, rep, td);
1346 if (error)
1347 return (error);
1348 *statep |= NFSSTA_WANTSND;
1349 (void) tsleep(statep, slpflag | (PZERO - 1),
1350 "nfsndlck", slptimeo);
1351 if (slpflag == PCATCH) {
1352 slpflag = 0;
1353 slptimeo = 2 * hz;
1354 }
1355 }
1356 *statep |= NFSSTA_SNDLOCK;
1357 return (0);
1358 }
1359
1360 /*
1361 * Unlock the stream socket for others.
1362 */
1363 void
1364 nfs_sndunlock(struct nfsreq *rep)
1365 {
1366 int *statep = &rep->r_nmp->nm_state;
1367
1368 if ((*statep & NFSSTA_SNDLOCK) == 0)
1369 panic("nfs sndunlock");
1370 *statep &= ~NFSSTA_SNDLOCK;
1371 if (*statep & NFSSTA_WANTSND) {
1372 *statep &= ~NFSSTA_WANTSND;
1373 wakeup(statep);
1374 }
1375 }
1376
1377 static int
1378 nfs_rcvlock(struct nfsreq *rep)
1379 {
1380 int *statep = &rep->r_nmp->nm_state;
1381 int error, slpflag, slptimeo = 0;
1382
1383 if (rep->r_nmp->nm_flag & NFSMNT_INT)
1384 slpflag = PCATCH;
1385 else
1386 slpflag = 0;
1387 while (*statep & NFSSTA_RCVLOCK) {
1388 error = nfs_sigintr(rep->r_nmp, rep, rep->r_td);
1389 if (error)
1390 return (error);
1391 *statep |= NFSSTA_WANTRCV;
1392 (void) tsleep(statep, slpflag | (PZERO - 1), "nfsrcvlk",
1393 slptimeo);
1394 /*
1395 * If our reply was recieved while we were sleeping,
1396 * then just return without taking the lock to avoid a
1397 * situation where a single iod could 'capture' the
1398 * recieve lock.
1399 */
1400 if (rep->r_mrep != NULL)
1401 return (EALREADY);
1402 if (slpflag == PCATCH) {
1403 slpflag = 0;
1404 slptimeo = 2 * hz;
1405 }
1406 }
1407 /* Always fail if our request has been cancelled. */
1408 if (rep != NULL && (error = NFS_SIGREP(rep)) != 0)
1409 return (error);
1410 *statep |= NFSSTA_RCVLOCK;
1411 return (0);
1412 }
1413
1414 /*
1415 * Unlock the stream socket for others.
1416 */
1417 static void
1418 nfs_rcvunlock(struct nfsreq *rep)
1419 {
1420 int *statep = &rep->r_nmp->nm_state;
1421
1422 if ((*statep & NFSSTA_RCVLOCK) == 0)
1423 panic("nfs rcvunlock");
1424 *statep &= ~NFSSTA_RCVLOCK;
1425 if (*statep & NFSSTA_WANTRCV) {
1426 *statep &= ~NFSSTA_WANTRCV;
1427 wakeup(statep);
1428 }
1429 }
1430
1431 /*
1432 * nfs_realign:
1433 *
1434 * Check for badly aligned mbuf data and realign by copying the unaligned
1435 * portion of the data into a new mbuf chain and freeing the portions
1436 * of the old chain that were replaced.
1437 *
1438 * We cannot simply realign the data within the existing mbuf chain
1439 * because the underlying buffers may contain other rpc commands and
1440 * we cannot afford to overwrite them.
1441 *
1442 * We would prefer to avoid this situation entirely. The situation does
1443 * not occur with NFS/UDP and is supposed to only occassionally occur
1444 * with TCP. Use vfs.nfs.realign_count and realign_test to check this.
1445 */
1446 static void
1447 nfs_realign(struct mbuf **pm, int hsiz)
1448 {
1449 struct mbuf *m;
1450 struct mbuf *n = NULL;
1451 int off = 0;
1452
1453 ++nfs_realign_test;
1454 while ((m = *pm) != NULL) {
1455 if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) {
1456 MGET(n, M_TRYWAIT, MT_DATA);
1457 if (m->m_len >= MINCLSIZE) {
1458 MCLGET(n, M_TRYWAIT);
1459 }
1460 n->m_len = 0;
1461 break;
1462 }
1463 pm = &m->m_next;
1464 }
1465 /*
1466 * If n is non-NULL, loop on m copying data, then replace the
1467 * portion of the chain that had to be realigned.
1468 */
1469 if (n != NULL) {
1470 ++nfs_realign_count;
1471 while (m) {
1472 m_copyback(n, off, m->m_len, mtod(m, caddr_t));
1473 off += m->m_len;
1474 m = m->m_next;
1475 }
1476 m_freem(*pm);
1477 *pm = n;
1478 }
1479 }
1480
1481
1482 static int
1483 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1484 {
1485 struct proc *p;
1486
1487 p = td ? td->td_proc : NULL;
1488 if (error) {
1489 tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n", server,
1490 msg, error);
1491 } else {
1492 tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1493 }
1494 return (0);
1495 }
1496
1497 void
1498 nfs_down(rep, nmp, td, msg, error, flags)
1499 struct nfsreq *rep;
1500 struct nfsmount *nmp;
1501 struct thread *td;
1502 const char *msg;
1503 int error, flags;
1504 {
1505
1506 if (nmp == NULL)
1507 return;
1508 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1509 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1510 VQ_NOTRESP, 0);
1511 nmp->nm_state |= NFSSTA_TIMEO;
1512 }
1513 #ifdef NFSSTA_LOCKTIMEO
1514 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1515 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1516 VQ_NOTRESPLOCK, 0);
1517 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1518 }
1519 #endif
1520 if (rep)
1521 rep->r_flags |= R_TPRINTFMSG;
1522 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1523 }
1524
1525 void
1526 nfs_up(rep, nmp, td, msg, flags)
1527 struct nfsreq *rep;
1528 struct nfsmount *nmp;
1529 struct thread *td;
1530 const char *msg;
1531 int flags;
1532 {
1533 if (nmp == NULL)
1534 return;
1535 if ((rep == NULL) || (rep->r_flags & R_TPRINTFMSG) != 0)
1536 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1537 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1538 nmp->nm_state &= ~NFSSTA_TIMEO;
1539 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1540 VQ_NOTRESP, 1);
1541 }
1542 #ifdef NFSSTA_LOCKTIMEO
1543 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1544 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1545 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1546 VQ_NOTRESPLOCK, 1);
1547 }
1548 #endif
1549 }
1550
Cache object: 9035f994d6a6a0267dd0ef7203124dd4
|