?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Get rid of struct user/UAREA.
[dragonfly.git] / sys / vfs / nfs / nfs_socket.c
blob68ce06188dd9cac1428188701f8ca0707d24d2f4
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 * $FreeBSD: src/sys/nfs/nfs_socket.c,v 1.60.2.6 2003/03/26 01:44:46 alfred Exp $
38 * $DragonFly: src/sys/vfs/nfs/nfs_socket.c,v 1.42 2007/02/25 23:17:13 corecode Exp $
39 */
40
41 /*
42 * Socket operations for use by nfs
43 */
44
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/proc.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/kernel.h>
51 #include <sys/mbuf.h>
52 #include <sys/vnode.h>
53 #include <sys/fcntl.h>
54 #include <sys/protosw.h>
55 #include <sys/resourcevar.h>
56 #include <sys/socket.h>
57 #include <sys/socketvar.h>
58 #include <sys/socketops.h>
59 #include <sys/syslog.h>
60 #include <sys/thread.h>
61 #include <sys/tprintf.h>
62 #include <sys/sysctl.h>
63 #include <sys/signalvar.h>
64 #include <sys/signal2.h>
65
66 #include <netinet/in.h>
67 #include <netinet/tcp.h>
68 #include <sys/thread2.h>
69
70 #include "rpcv2.h"
71 #include "nfsproto.h"
72 #include "nfs.h"
73 #include "xdr_subs.h"
74 #include "nfsm_subs.h"
75 #include "nfsmount.h"
76 #include "nfsnode.h"
77 #include "nfsrtt.h"
78
79 #define TRUE 1
80 #define FALSE 0
81
82 /*
83 * Estimate rto for an nfs rpc sent via. an unreliable datagram.
84 * Use the mean and mean deviation of rtt for the appropriate type of rpc
85 * for the frequent rpcs and a default for the others.
86 * The justification for doing "other" this way is that these rpcs
87 * happen so infrequently that timer est. would probably be stale.
88 * Also, since many of these rpcs are
89 * non-idempotent, a conservative timeout is desired.
90 * getattr, lookup - A+2D
91 * read, write - A+4D
92 * other - nm_timeo
93 */
94 #define NFS_RTO(n, t) \
95 ((t) == 0 ? (n)->nm_timeo : \
96 ((t) < 3 ? \
97 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
98 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
99 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
100 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
101 /*
102 * External data, mostly RPC constants in XDR form
103 */
104 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
105 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
106 rpc_auth_kerb;
107 extern u_int32_t nfs_prog;
108 extern struct nfsstats nfsstats;
109 extern int nfsv3_procid[NFS_NPROCS];
110 extern int nfs_ticks;
111
112 /*
113 * Defines which timer to use for the procnum.
114 * 0 - default
115 * 1 - getattr
116 * 2 - lookup
117 * 3 - read
118 * 4 - write
119 */
120 static int proct[NFS_NPROCS] = {
121 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
122 0, 0, 0,
123 };
124
125 static int nfs_realign_test;
126 static int nfs_realign_count;
127 static int nfs_bufpackets = 4;
128 static int nfs_timer_raced;
129
130 SYSCTL_DECL(_vfs_nfs);
131
132 SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_test, CTLFLAG_RW, &nfs_realign_test, 0, "");
133 SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_count, CTLFLAG_RW, &nfs_realign_count, 0, "");
134 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0, "");
135
136
137 /*
138 * There is a congestion window for outstanding rpcs maintained per mount
139 * point. The cwnd size is adjusted in roughly the way that:
140 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
141 * SIGCOMM '88". ACM, August 1988.
142 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
143 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
144 * of rpcs is in progress.
145 * (The sent count and cwnd are scaled for integer arith.)
146 * Variants of "slow start" were tried and were found to be too much of a
147 * performance hit (ave. rtt 3 times larger),
148 * I suspect due to the large rtt that nfs rpcs have.
149 */
150 #define NFS_CWNDSCALE 256
151 #define NFS_MAXCWND (NFS_CWNDSCALE * 32)
152 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
153 int nfsrtton = 0;
154 struct nfsrtt nfsrtt;
155 struct callout nfs_timer_handle;
156
157 static int nfs_msg (struct thread *,char *,char *);
158 static int nfs_rcvlock (struct nfsreq *);
159 static void nfs_rcvunlock (struct nfsreq *);
160 static void nfs_realign (struct mbuf **pm, int hsiz);
161 static int nfs_receive (struct nfsreq *rep, struct sockaddr **aname,
162 struct mbuf **mp);
163 static void nfs_softterm (struct nfsreq *rep);
164 static int nfs_reconnect (struct nfsreq *rep);
165 #ifndef NFS_NOSERVER
166 static int nfsrv_getstream (struct nfssvc_sock *, int, int *);
167
168 int (*nfsrv3_procs[NFS_NPROCS]) (struct nfsrv_descript *nd,
169 struct nfssvc_sock *slp,
170 struct thread *td,
171 struct mbuf **mreqp) = {
172 nfsrv_null,
173 nfsrv_getattr,
174 nfsrv_setattr,
175 nfsrv_lookup,
176 nfsrv3_access,
177 nfsrv_readlink,
178 nfsrv_read,
179 nfsrv_write,
180 nfsrv_create,
181 nfsrv_mkdir,
182 nfsrv_symlink,
183 nfsrv_mknod,
184 nfsrv_remove,
185 nfsrv_rmdir,
186 nfsrv_rename,
187 nfsrv_link,
188 nfsrv_readdir,
189 nfsrv_readdirplus,
190 nfsrv_statfs,
191 nfsrv_fsinfo,
192 nfsrv_pathconf,
193 nfsrv_commit,
194 nfsrv_noop,
195 nfsrv_noop,
196 nfsrv_noop,
197 nfsrv_noop
198 };
199 #endif /* NFS_NOSERVER */
200
201 /*
202 * Initialize sockets and congestion for a new NFS connection.
203 * We do not free the sockaddr if error.
204 */
205 int
206 nfs_connect(struct nfsmount *nmp, struct nfsreq *rep)
207 {
208 struct socket *so;
209 int error, rcvreserve, sndreserve;
210 int pktscale;
211 struct sockaddr *saddr;
212 struct sockaddr_in *sin;
213 struct thread *td = &thread0; /* only used for socreate and sobind */
214
215 nmp->nm_so = (struct socket *)0;
216 saddr = nmp->nm_nam;
217 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
218 nmp->nm_soproto, td);
219 if (error)
220 goto bad;
221 so = nmp->nm_so;
222 nmp->nm_soflags = so->so_proto->pr_flags;
223
224 /*
225 * Some servers require that the client port be a reserved port number.
226 */
227 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
228 struct sockopt sopt;
229 int ip;
230 struct sockaddr_in ssin;
231
232 bzero(&sopt, sizeof sopt);
233 ip = IP_PORTRANGE_LOW;
234 sopt.sopt_level = IPPROTO_IP;
235 sopt.sopt_name = IP_PORTRANGE;
236 sopt.sopt_val = (void *)&ip;
237 sopt.sopt_valsize = sizeof(ip);
238 sopt.sopt_td = NULL;
239 error = sosetopt(so, &sopt);
240 if (error)
241 goto bad;
242 bzero(&ssin, sizeof ssin);
243 sin = &ssin;
244 sin->sin_len = sizeof (struct sockaddr_in);
245 sin->sin_family = AF_INET;
246 sin->sin_addr.s_addr = INADDR_ANY;
247 sin->sin_port = htons(0);
248 error = sobind(so, (struct sockaddr *)sin, td);
249 if (error)
250 goto bad;
251 bzero(&sopt, sizeof sopt);
252 ip = IP_PORTRANGE_DEFAULT;
253 sopt.sopt_level = IPPROTO_IP;
254 sopt.sopt_name = IP_PORTRANGE;
255 sopt.sopt_val = (void *)&ip;
256 sopt.sopt_valsize = sizeof(ip);
257 sopt.sopt_td = NULL;
258 error = sosetopt(so, &sopt);
259 if (error)
260 goto bad;
261 }
262
263 /*
264 * Protocols that do not require connections may be optionally left
265 * unconnected for servers that reply from a port other than NFS_PORT.
266 */
267 if (nmp->nm_flag & NFSMNT_NOCONN) {
268 if (nmp->nm_soflags & PR_CONNREQUIRED) {
269 error = ENOTCONN;
270 goto bad;
271 }
272 } else {
273 error = soconnect(so, nmp->nm_nam, td);
274 if (error)
275 goto bad;
276
277 /*
278 * Wait for the connection to complete. Cribbed from the
279 * connect system call but with the wait timing out so
280 * that interruptible mounts don't hang here for a long time.
281 */
282 crit_enter();
283 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
284 (void) tsleep((caddr_t)&so->so_timeo, 0,
285 "nfscon", 2 * hz);
286 if ((so->so_state & SS_ISCONNECTING) &&
287 so->so_error == 0 && rep &&
288 (error = nfs_sigintr(nmp, rep, rep->r_td)) != 0){
289 so->so_state &= ~SS_ISCONNECTING;
290 crit_exit();
291 goto bad;
292 }
293 }
294 if (so->so_error) {
295 error = so->so_error;
296 so->so_error = 0;
297 crit_exit();
298 goto bad;
299 }
300 crit_exit();
301 }
302 so->so_rcv.sb_timeo = (5 * hz);
303 so->so_snd.sb_timeo = (5 * hz);
304
305 /*
306 * Get buffer reservation size from sysctl, but impose reasonable
307 * limits.
308 */
309 pktscale = nfs_bufpackets;
310 if (pktscale < 2)
311 pktscale = 2;
312 if (pktscale > 64)
313 pktscale = 64;
314
315 if (nmp->nm_sotype == SOCK_DGRAM) {
316 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
317 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
318 NFS_MAXPKTHDR) * pktscale;
319 } else if (nmp->nm_sotype == SOCK_SEQPACKET) {
320 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
321 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
322 NFS_MAXPKTHDR) * pktscale;
323 } else {
324 if (nmp->nm_sotype != SOCK_STREAM)
325 panic("nfscon sotype");
326 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
327 struct sockopt sopt;
328 int val;
329
330 bzero(&sopt, sizeof sopt);
331 sopt.sopt_level = SOL_SOCKET;
332 sopt.sopt_name = SO_KEEPALIVE;
333 sopt.sopt_val = &val;
334 sopt.sopt_valsize = sizeof val;
335 val = 1;
336 sosetopt(so, &sopt);
337 }
338 if (so->so_proto->pr_protocol == IPPROTO_TCP) {
339 struct sockopt sopt;
340 int val;
341
342 bzero(&sopt, sizeof sopt);
343 sopt.sopt_level = IPPROTO_TCP;
344 sopt.sopt_name = TCP_NODELAY;
345 sopt.sopt_val = &val;
346 sopt.sopt_valsize = sizeof val;
347 val = 1;
348 sosetopt(so, &sopt);
349 }
350 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
351 sizeof (u_int32_t)) * pktscale;
352 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
353 sizeof (u_int32_t)) * pktscale;
354 }
355 error = soreserve(so, sndreserve, rcvreserve,
356 &td->td_proc->p_rlimit[RLIMIT_SBSIZE]);
357 if (error)
358 goto bad;
359 so->so_rcv.sb_flags |= SB_NOINTR;
360 so->so_snd.sb_flags |= SB_NOINTR;
361
362 /* Initialize other non-zero congestion variables */
363 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] =
364 nmp->nm_srtt[3] = (NFS_TIMEO << 3);
365 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
366 nmp->nm_sdrtt[3] = 0;
367 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */
368 nmp->nm_sent = 0;
369 nmp->nm_timeouts = 0;
370 return (0);
371
372 bad:
373 nfs_disconnect(nmp);
374 return (error);
375 }
376
377 /*
378 * Reconnect routine:
379 * Called when a connection is broken on a reliable protocol.
380 * - clean up the old socket
381 * - nfs_connect() again
382 * - set R_MUSTRESEND for all outstanding requests on mount point
383 * If this fails the mount point is DEAD!
384 * nb: Must be called with the nfs_sndlock() set on the mount point.
385 */
386 static int
387 nfs_reconnect(struct nfsreq *rep)
388 {
389 struct nfsreq *rp;
390 struct nfsmount *nmp = rep->r_nmp;
391 int error;
392
393 nfs_disconnect(nmp);
394 while ((error = nfs_connect(nmp, rep)) != 0) {
395 if (error == EINTR || error == ERESTART)
396 return (EINTR);
397 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
398 }
399
400 /*
401 * Loop through outstanding request list and fix up all requests
402 * on old socket.
403 */
404 crit_enter();
405 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
406 if (rp->r_nmp == nmp)
407 rp->r_flags |= R_MUSTRESEND;
408 }
409 crit_exit();
410 return (0);
411 }
412
413 /*
414 * NFS disconnect. Clean up and unlink.
415 */
416 void
417 nfs_disconnect(struct nfsmount *nmp)
418 {
419 struct socket *so;
420
421 if (nmp->nm_so) {
422 so = nmp->nm_so;
423 nmp->nm_so = (struct socket *)0;
424 soshutdown(so, 2);
425 soclose(so, FNONBLOCK);
426 }
427 }
428
429 void
430 nfs_safedisconnect(struct nfsmount *nmp)
431 {
432 struct nfsreq dummyreq;
433
434 bzero(&dummyreq, sizeof(dummyreq));
435 dummyreq.r_nmp = nmp;
436 dummyreq.r_td = NULL;
437 nfs_rcvlock(&dummyreq);
438 nfs_disconnect(nmp);
439 nfs_rcvunlock(&dummyreq);
440 }
441
442 /*
443 * This is the nfs send routine. For connection based socket types, it
444 * must be called with an nfs_sndlock() on the socket.
445 * "rep == NULL" indicates that it has been called from a server.
446 * For the client side:
447 * - return EINTR if the RPC is terminated, 0 otherwise
448 * - set R_MUSTRESEND if the send fails for any reason
449 * - do any cleanup required by recoverable socket errors (?)
450 * For the server side:
451 * - return EINTR or ERESTART if interrupted by a signal
452 * - return EPIPE if a connection is lost for connection based sockets (TCP...)
453 * - do any cleanup required by recoverable socket errors (?)
454 */
455 int
456 nfs_send(struct socket *so, struct sockaddr *nam, struct mbuf *top,
457 struct nfsreq *rep)
458 {
459 struct sockaddr *sendnam;
460 int error, soflags, flags;
461
462 if (rep) {
463 if (rep->r_flags & R_SOFTTERM) {
464 m_freem(top);
465 return (EINTR);
466 }
467 if ((so = rep->r_nmp->nm_so) == NULL) {
468 rep->r_flags |= R_MUSTRESEND;
469 m_freem(top);
470 return (0);
471 }
472 rep->r_flags &= ~R_MUSTRESEND;
473 soflags = rep->r_nmp->nm_soflags;
474 } else
475 soflags = so->so_proto->pr_flags;
476 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
477 sendnam = (struct sockaddr *)0;
478 else
479 sendnam = nam;
480 if (so->so_type == SOCK_SEQPACKET)
481 flags = MSG_EOR;
482 else
483 flags = 0;
484
485 error = so_pru_sosend(so, sendnam, NULL, top, NULL, flags,
486 curthread /*XXX*/);
487 /*
488 * ENOBUFS for dgram sockets is transient and non fatal.
489 * No need to log, and no need to break a soft mount.
490 */
491 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
492 error = 0;
493 if (rep) /* do backoff retransmit on client */
494 rep->r_flags |= R_MUSTRESEND;
495 }
496
497 if (error) {
498 if (rep) {
499 log(LOG_INFO, "nfs send error %d for server %s\n",error,
500 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
501 /*
502 * Deal with errors for the client side.
503 */
504 if (rep->r_flags & R_SOFTTERM)
505 error = EINTR;
506 else
507 rep->r_flags |= R_MUSTRESEND;
508 } else
509 log(LOG_INFO, "nfsd send error %d\n", error);
510
511 /*
512 * Handle any recoverable (soft) socket errors here. (?)
513 */
514 if (error != EINTR && error != ERESTART &&
515 error != EWOULDBLOCK && error != EPIPE)
516 error = 0;
517 }
518 return (error);
519 }
520
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 static int
531 nfs_receive(struct nfsreq *rep, struct sockaddr **aname, struct mbuf **mp)
532 {
533 struct socket *so;
534 struct uio auio;
535 struct iovec aio;
536 struct mbuf *m;
537 struct mbuf *control;
538 u_int32_t len;
539 struct sockaddr **getnam;
540 int error, sotype, rcvflg;
541 struct thread *td = curthread; /* XXX */
542
543 /*
544 * Set up arguments for soreceive()
545 */
546 *mp = (struct mbuf *)0;
547 *aname = (struct sockaddr *)0;
548 sotype = rep->r_nmp->nm_sotype;
549
550 /*
551 * For reliable protocols, lock against other senders/receivers
552 * in case a reconnect is necessary.
553 * For SOCK_STREAM, first get the Record Mark to find out how much
554 * more there is to get.
555 * We must lock the socket against other receivers
556 * until we have an entire rpc request/reply.
557 */
558 if (sotype != SOCK_DGRAM) {
559 error = nfs_sndlock(rep);
560 if (error)
561 return (error);
562 tryagain:
563 /*
564 * Check for fatal errors and resending request.
565 */
566 /*
567 * Ugh: If a reconnect attempt just happened, nm_so
568 * would have changed. NULL indicates a failed
569 * attempt that has essentially shut down this
570 * mount point.
571 */
572 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
573 nfs_sndunlock(rep);
574 return (EINTR);
575 }
576 so = rep->r_nmp->nm_so;
577 if (!so) {
578 error = nfs_reconnect(rep);
579 if (error) {
580 nfs_sndunlock(rep);
581 return (error);
582 }
583 goto tryagain;
584 }
585 while (rep->r_flags & R_MUSTRESEND) {
586 m = m_copym(rep->r_mreq, 0, M_COPYALL, MB_WAIT);
587 nfsstats.rpcretries++;
588 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
589 if (error) {
590 if (error == EINTR || error == ERESTART ||
591 (error = nfs_reconnect(rep)) != 0) {
592 nfs_sndunlock(rep);
593 return (error);
594 }
595 goto tryagain;
596 }
597 }
598 nfs_sndunlock(rep);
599 if (sotype == SOCK_STREAM) {
600 aio.iov_base = (caddr_t) &len;
601 aio.iov_len = sizeof(u_int32_t);
602 auio.uio_iov = &aio;
603 auio.uio_iovcnt = 1;
604 auio.uio_segflg = UIO_SYSSPACE;
605 auio.uio_rw = UIO_READ;
606 auio.uio_offset = 0;
607 auio.uio_resid = sizeof(u_int32_t);
608 auio.uio_td = td;
609 do {
610 rcvflg = MSG_WAITALL;
611 error = so_pru_soreceive(so, NULL, &auio, NULL,
612 NULL, &rcvflg);
613 if (error == EWOULDBLOCK && rep) {
614 if (rep->r_flags & R_SOFTTERM)
615 return (EINTR);
616 }
617 } while (error == EWOULDBLOCK);
618 if (!error && auio.uio_resid > 0) {
619 /*
620 * Don't log a 0 byte receive; it means
621 * that the socket has been closed, and
622 * can happen during normal operation
623 * (forcible unmount or Solaris server).
624 */
625 if (auio.uio_resid != sizeof (u_int32_t))
626 log(LOG_INFO,
627 "short receive (%d/%d) from nfs server %s\n",
628 (int)(sizeof(u_int32_t) - auio.uio_resid),
629 (int)sizeof(u_int32_t),
630 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
631 error = EPIPE;
632 }
633 if (error)
634 goto errout;
635 len = ntohl(len) & ~0x80000000;
636 /*
637 * This is SERIOUS! We are out of sync with the sender
638 * and forcing a disconnect/reconnect is all I can do.
639 */
640 if (len > NFS_MAXPACKET) {
641 log(LOG_ERR, "%s (%d) from nfs server %s\n",
642 "impossible packet length",
643 len,
644 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
645 error = EFBIG;
646 goto errout;
647 }
648 auio.uio_resid = len;
649 do {
650 rcvflg = MSG_WAITALL;
651 error = so_pru_soreceive(so, NULL, &auio, mp,
652 NULL, &rcvflg);
653 } while (error == EWOULDBLOCK || error == EINTR ||
654 error == ERESTART);
655 if (!error && auio.uio_resid > 0) {
656 if (len != auio.uio_resid)
657 log(LOG_INFO,
658 "short receive (%d/%d) from nfs server %s\n",
659 len - auio.uio_resid, len,
660 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
661 error = EPIPE;
662 }
663 } else {
664 /*
665 * NB: Since uio_resid is big, MSG_WAITALL is ignored
666 * and soreceive() will return when it has either a
667 * control msg or a data msg.
668 * We have no use for control msg., but must grab them
669 * and then throw them away so we know what is going
670 * on.
671 */
672 auio.uio_resid = len = 100000000; /* Anything Big */
673 auio.uio_td = td;
674 do {
675 rcvflg = 0;
676 error = so_pru_soreceive(so, NULL, &auio, mp,
677 &control, &rcvflg);
678 if (control)
679 m_freem(control);
680 if (error == EWOULDBLOCK && rep) {
681 if (rep->r_flags & R_SOFTTERM)
682 return (EINTR);
683 }
684 } while (error == EWOULDBLOCK ||
685 (!error && *mp == NULL && control));
686 if ((rcvflg & MSG_EOR) == 0)
687 kprintf("Egad!!\n");
688 if (!error && *mp == NULL)
689 error = EPIPE;
690 len -= auio.uio_resid;
691 }
692 errout:
693 if (error && error != EINTR && error != ERESTART) {
694 m_freem(*mp);
695 *mp = (struct mbuf *)0;
696 if (error != EPIPE)
697 log(LOG_INFO,
698 "receive error %d from nfs server %s\n",
699 error,
700 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
701 error = nfs_sndlock(rep);
702 if (!error) {
703 error = nfs_reconnect(rep);
704 if (!error)
705 goto tryagain;
706 else
707 nfs_sndunlock(rep);
708 }
709 }
710 } else {
711 if ((so = rep->r_nmp->nm_so) == NULL)
712 return (EACCES);
713 if (so->so_state & SS_ISCONNECTED)
714 getnam = (struct sockaddr **)0;
715 else
716 getnam = aname;
717 auio.uio_resid = len = 1000000;
718 auio.uio_td = td;
719 do {
720 rcvflg = 0;
721 error = so_pru_soreceive(so, getnam, &auio, mp, NULL,
722 &rcvflg);
723 if (error == EWOULDBLOCK &&
724 (rep->r_flags & R_SOFTTERM))
725 return (EINTR);
726 } while (error == EWOULDBLOCK);
727 len -= auio.uio_resid;
728 }
729 if (error) {
730 m_freem(*mp);
731 *mp = (struct mbuf *)0;
732 }
733 /*
734 * Search for any mbufs that are not a multiple of 4 bytes long
735 * or with m_data not longword aligned.
736 * These could cause pointer alignment problems, so copy them to
737 * well aligned mbufs.
738 */
739 nfs_realign(mp, 5 * NFSX_UNSIGNED);
740 return (error);
741 }
742
743 /*
744 * Implement receipt of reply on a socket.
745 * We must search through the list of received datagrams matching them
746 * with outstanding requests using the xid, until ours is found.
747 */
748 /* ARGSUSED */
749 int
750 nfs_reply(struct nfsreq *myrep)
751 {
752 struct nfsreq *rep;
753 struct nfsmount *nmp = myrep->r_nmp;
754 int32_t t1;
755 struct mbuf *mrep, *md;
756 struct sockaddr *nam;
757 u_int32_t rxid, *tl;
758 caddr_t dpos, cp2;
759 int error;
760
761 /*
762 * Loop around until we get our own reply
763 */
764 for (;;) {
765 /*
766 * Lock against other receivers so that I don't get stuck in
767 * sbwait() after someone else has received my reply for me.
768 * Also necessary for connection based protocols to avoid
769 * race conditions during a reconnect.
770 * If nfs_rcvlock() returns EALREADY, that means that
771 * the reply has already been recieved by another
772 * process and we can return immediately. In this
773 * case, the lock is not taken to avoid races with
774 * other processes.
775 */
776 error = nfs_rcvlock(myrep);
777 if (error == EALREADY)
778 return (0);
779 if (error)
780 return (error);
781 /*
782 * Get the next Rpc reply off the socket
783 */
784 error = nfs_receive(myrep, &nam, &mrep);
785 nfs_rcvunlock(myrep);
786 if (error) {
787 /*
788 * Ignore routing errors on connectionless protocols??
789 */
790 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
791 nmp->nm_so->so_error = 0;
792 if (myrep->r_flags & R_GETONEREP)
793 return (0);
794 continue;
795 }
796 return (error);
797 }
798 if (nam)
799 FREE(nam, M_SONAME);
800
801 /*
802 * Get the xid and check that it is an rpc reply
803 */
804 md = mrep;
805 dpos = mtod(md, caddr_t);
806 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
807 rxid = *tl++;
808 if (*tl != rpc_reply) {
809 nfsstats.rpcinvalid++;
810 m_freem(mrep);
811 nfsmout:
812 if (myrep->r_flags & R_GETONEREP)
813 return (0);
814 continue;
815 }
816
817 /*
818 * Loop through the request list to match up the reply
819 * Iff no match, just drop the datagram. On match, set
820 * r_mrep atomically to prevent the timer from messing
821 * around with the request after we have exited the critical
822 * section.
823 */
824 crit_enter();
825 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
826 if (rep->r_mrep == NULL && rxid == rep->r_xid) {
827 rep->r_mrep = mrep;
828 break;
829 }
830 }
831 crit_exit();
832
833 /*
834 * Fill in the rest of the reply if we found a match.
835 */
836 if (rep) {
837 rep->r_md = md;
838 rep->r_dpos = dpos;
839 if (nfsrtton) {
840 struct rttl *rt;
841
842 rt = &nfsrtt.rttl[nfsrtt.pos];
843 rt->proc = rep->r_procnum;
844 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
845 rt->sent = nmp->nm_sent;
846 rt->cwnd = nmp->nm_cwnd;
847 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
848 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
849 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
850 getmicrotime(&rt->tstamp);
851 if (rep->r_flags & R_TIMING)
852 rt->rtt = rep->r_rtt;
853 else
854 rt->rtt = 1000000;
855 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
856 }
857 /*
858 * Update congestion window.
859 * Do the additive increase of
860 * one rpc/rtt.
861 */
862 if (nmp->nm_cwnd <= nmp->nm_sent) {
863 nmp->nm_cwnd +=
864 (NFS_CWNDSCALE * NFS_CWNDSCALE +
865 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
866 if (nmp->nm_cwnd > NFS_MAXCWND)
867 nmp->nm_cwnd = NFS_MAXCWND;
868 }
869 crit_enter(); /* nfs_timer interlock for nm_sent */
870 if (rep->r_flags & R_SENT) {
871 rep->r_flags &= ~R_SENT;
872 nmp->nm_sent -= NFS_CWNDSCALE;
873 }
874 crit_exit();
875 /*
876 * Update rtt using a gain of 0.125 on the mean
877 * and a gain of 0.25 on the deviation.
878 */
879 if (rep->r_flags & R_TIMING) {
880 /*
881 * Since the timer resolution of
882 * NFS_HZ is so course, it can often
883 * result in r_rtt == 0. Since
884 * r_rtt == N means that the actual
885 * rtt is between N+dt and N+2-dt ticks,
886 * add 1.
887 */
888 t1 = rep->r_rtt + 1;
889 t1 -= (NFS_SRTT(rep) >> 3);
890 NFS_SRTT(rep) += t1;
891 if (t1 < 0)
892 t1 = -t1;
893 t1 -= (NFS_SDRTT(rep) >> 2);
894 NFS_SDRTT(rep) += t1;
895 }
896 nmp->nm_timeouts = 0;
897 }
898 /*
899 * If not matched to a request, drop it.
900 * If it's mine, get out.
901 */
902 if (rep == NULL) {
903 nfsstats.rpcunexpected++;
904 m_freem(mrep);
905 } else if (rep == myrep) {
906 if (rep->r_mrep == NULL)
907 panic("nfsreply nil");
908 return (0);
909 }
910 if (myrep->r_flags & R_GETONEREP)
911 return (0);
912 }
913 }
914
915 /*
916 * nfs_request - goes something like this
917 * - fill in request struct
918 * - links it into list
919 * - calls nfs_send() for first transmit
920 * - calls nfs_receive() to get reply
921 * - break down rpc header and return with nfs reply pointed to
922 * by mrep or error
923 * nb: always frees up mreq mbuf list
924 */
925 int
926 nfs_request(struct vnode *vp, struct mbuf *mrest, int procnum,
927 struct thread *td, struct ucred *cred, struct mbuf **mrp,
928 struct mbuf **mdp, caddr_t *dposp)
929 {
930 struct mbuf *mrep, *m2;
931 struct nfsreq *rep;
932 u_int32_t *tl;
933 int i;
934 struct nfsmount *nmp;
935 struct mbuf *m, *md, *mheadend;
936 char nickv[RPCX_NICKVERF];
937 time_t waituntil;
938 caddr_t dpos, cp2;
939 int t1, error = 0, mrest_len, auth_len, auth_type;
940 int trylater_delay = 15, trylater_cnt = 0, failed_auth = 0;
941 int verf_len, verf_type;
942 u_int32_t xid;
943 char *auth_str, *verf_str;
944 NFSKERBKEY_T key; /* save session key */
945
946 /* Reject requests while attempting a forced unmount. */
947 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) {
948 m_freem(mrest);
949 return (ESTALE);
950 }
951 nmp = VFSTONFS(vp->v_mount);
952 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
953 rep->r_nmp = nmp;
954 rep->r_vp = vp;
955 rep->r_td = td;
956 rep->r_procnum = procnum;
957 rep->r_mreq = NULL;
958 i = 0;
959 m = mrest;
960 while (m) {
961 i += m->m_len;
962 m = m->m_next;
963 }
964 mrest_len = i;
965
966 /*
967 * Get the RPC header with authorization.
968 */
969 kerbauth:
970 verf_str = auth_str = (char *)0;
971 if (nmp->nm_flag & NFSMNT_KERB) {
972 verf_str = nickv;
973 verf_len = sizeof (nickv);
974 auth_type = RPCAUTH_KERB4;
975 bzero((caddr_t)key, sizeof (key));
976 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
977 &auth_len, verf_str, verf_len)) {
978 error = nfs_getauth(nmp, rep, cred, &auth_str,
979 &auth_len, verf_str, &verf_len, key);
980 if (error) {
981 kfree((caddr_t)rep, M_NFSREQ);
982 m_freem(mrest);
983 return (error);
984 }
985 }
986 } else {
987 auth_type = RPCAUTH_UNIX;
988 if (cred->cr_ngroups < 1)
989 panic("nfsreq nogrps");
990 auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ?
991 nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) +
992 5 * NFSX_UNSIGNED;
993 }
994 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
995 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
996 if (auth_str)
997 kfree(auth_str, M_TEMP);
998
999 /*
1000 * For stream protocols, insert a Sun RPC Record Mark.
1001 */
1002 if (nmp->nm_sotype == SOCK_STREAM) {
1003 M_PREPEND(m, NFSX_UNSIGNED, MB_WAIT);
1004 if (m == NULL) {
1005 kfree(rep, M_NFSREQ);
1006 return (ENOBUFS);
1007 }
1008 *mtod(m, u_int32_t *) = htonl(0x80000000 |
1009 (m->m_pkthdr.len - NFSX_UNSIGNED));
1010 }
1011 rep->r_mreq = m;
1012 rep->r_xid = xid;
1013 tryagain:
1014 if (nmp->nm_flag & NFSMNT_SOFT)
1015 rep->r_retry = nmp->nm_retry;
1016 else
1017 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */
1018 rep->r_rtt = rep->r_rexmit = 0;
1019 if (proct[procnum] > 0)
1020 rep->r_flags = R_TIMING | R_MASKTIMER;
1021 else
1022 rep->r_flags = R_MASKTIMER;
1023 rep->r_mrep = NULL;
1024
1025 /*
1026 * Do the client side RPC.
1027 */
1028 nfsstats.rpcrequests++;
1029
1030 /*
1031 * Chain request into list of outstanding requests. Be sure
1032 * to put it LAST so timer finds oldest requests first. Note
1033 * that R_MASKTIMER is set at the moment to prevent any timer
1034 * action on this request while we are still doing processing on
1035 * it below. splsoftclock() primarily protects nm_sent. Note
1036 * that we may block in this code so there is no atomicy guarentee.
1037 */
1038 crit_enter();
1039 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1040
1041 /*
1042 * If backing off another request or avoiding congestion, don't
1043 * send this one now but let timer do it. If not timing a request,
1044 * do it now.
1045 *
1046 * Even though the timer will not mess with our request there is
1047 * still the possibility that we will race a reply (which clears
1048 * R_SENT), especially on localhost connections, so be very careful
1049 * when setting R_SENT. We could set R_SENT prior to calling
1050 * nfs_send() but why bother if the response occurs that quickly?
1051 */
1052 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1053 (nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1054 nmp->nm_sent < nmp->nm_cwnd)) {
1055 if (nmp->nm_soflags & PR_CONNREQUIRED)
1056 error = nfs_sndlock(rep);
1057 if (!error) {
1058 m2 = m_copym(m, 0, M_COPYALL, MB_WAIT);
1059 error = nfs_send(nmp->nm_so, nmp->nm_nam, m2, rep);
1060 if (nmp->nm_soflags & PR_CONNREQUIRED)
1061 nfs_sndunlock(rep);
1062 }
1063 if (!error && (rep->r_flags & R_MUSTRESEND) == 0 &&
1064 rep->r_mrep == NULL) {
1065 KASSERT((rep->r_flags & R_SENT) == 0,
1066 ("R_SENT ASSERT %p", rep));
1067 nmp->nm_sent += NFS_CWNDSCALE;
1068 rep->r_flags |= R_SENT;
1069 }
1070 } else {
1071 rep->r_rtt = -1;
1072 }
1073
1074 /*
1075 * Let the timer do what it will with the request, then
1076 * wait for the reply from our send or the timer's.
1077 */
1078 if (!error || error == EPIPE) {
1079 rep->r_flags &= ~R_MASKTIMER;
1080 crit_exit();
1081 error = nfs_reply(rep);
1082 crit_enter();
1083 }
1084
1085 /*
1086 * RPC done, unlink the request, but don't rip it out from under
1087 * the callout timer.
1088 */
1089 while (rep->r_flags & R_LOCKED) {
1090 nfs_timer_raced = 1;
1091 tsleep(&nfs_timer_raced, 0, "nfstrac", 0);
1092 }
1093 TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1094
1095 /*
1096 * Decrement the outstanding request count.
1097 */
1098 if (rep->r_flags & R_SENT) {
1099 rep->r_flags &= ~R_SENT;
1100 nmp->nm_sent -= NFS_CWNDSCALE;
1101 }
1102 crit_exit();
1103
1104 /*
1105 * If there was a successful reply and a tprintf msg.
1106 * tprintf a response.
1107 */
1108 if (!error && (rep->r_flags & R_TPRINTFMSG))
1109 nfs_msg(rep->r_td, nmp->nm_mountp->mnt_stat.f_mntfromname,
1110 "is alive again");
1111 mrep = rep->r_mrep;
1112 md = rep->r_md;
1113 dpos = rep->r_dpos;
1114 if (error) {
1115 m_freem(rep->r_mreq);
1116 kfree((caddr_t)rep, M_NFSREQ);
1117 return (error);
1118 }
1119
1120 /*
1121 * break down the rpc header and check if ok
1122 */
1123 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1124 if (*tl++ == rpc_msgdenied) {
1125 if (*tl == rpc_mismatch)
1126 error = EOPNOTSUPP;
1127 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1128 if (!failed_auth) {
1129 failed_auth++;
1130 mheadend->m_next = (struct mbuf *)0;
1131 m_freem(mrep);
1132 m_freem(rep->r_mreq);
1133 goto kerbauth;
1134 } else
1135 error = EAUTH;
1136 } else
1137 error = EACCES;
1138 m_freem(mrep);
1139 m_freem(rep->r_mreq);
1140 kfree((caddr_t)rep, M_NFSREQ);
1141 return (error);
1142 }
1143
1144 /*
1145 * Grab any Kerberos verifier, otherwise just throw it away.
1146 */
1147 verf_type = fxdr_unsigned(int, *tl++);
1148 i = fxdr_unsigned(int32_t, *tl);
1149 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1150 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1151 if (error)
1152 goto nfsmout;
1153 } else if (i > 0)
1154 nfsm_adv(nfsm_rndup(i));
1155 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1156 /* 0 == ok */
1157 if (*tl == 0) {
1158 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1159 if (*tl != 0) {
1160 error = fxdr_unsigned(int, *tl);
1161 if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1162 error == NFSERR_TRYLATER) {
1163 m_freem(mrep);
1164 error = 0;
1165 waituntil = time_second + trylater_delay;
1166 while (time_second < waituntil)
1167 (void) tsleep((caddr_t)&lbolt,
1168 0, "nqnfstry", 0);
1169 trylater_delay *= nfs_backoff[trylater_cnt];
1170 if (trylater_cnt < 7)
1171 trylater_cnt++;
1172 goto tryagain;
1173 }
1174
1175 /*
1176 * If the File Handle was stale, invalidate the
1177 * lookup cache, just in case.
1178 *
1179 * To avoid namecache<->vnode deadlocks we must
1180 * release the vnode lock if we hold it.
1181 */
1182 if (error == ESTALE) {
1183 int ltype;
1184
1185 ltype = lockstatus(&vp->v_lock, curthread);
1186 if (ltype == LK_EXCLUSIVE || ltype == LK_SHARED)
1187 lockmgr(&vp->v_lock, LK_RELEASE);
1188 cache_inval_vp(vp, CINV_CHILDREN);
1189 if (ltype == LK_EXCLUSIVE || ltype == LK_SHARED)
1190 lockmgr(&vp->v_lock, ltype);
1191 }
1192 if (nmp->nm_flag & NFSMNT_NFSV3) {
1193 *mrp = mrep;
1194 *mdp = md;
1195 *dposp = dpos;
1196 error |= NFSERR_RETERR;
1197 } else
1198 m_freem(mrep);
1199 m_freem(rep->r_mreq);
1200 kfree((caddr_t)rep, M_NFSREQ);
1201 return (error);
1202 }
1203
1204 *mrp = mrep;
1205 *mdp = md;
1206 *dposp = dpos;
1207 m_freem(rep->r_mreq);
1208 FREE((caddr_t)rep, M_NFSREQ);
1209 return (0);
1210 }
1211 m_freem(mrep);
1212 error = EPROTONOSUPPORT;
1213 nfsmout:
1214 m_freem(rep->r_mreq);
1215 kfree((caddr_t)rep, M_NFSREQ);
1216 return (error);
1217 }
1218
1219 #ifndef NFS_NOSERVER
1220 /*
1221 * Generate the rpc reply header
1222 * siz arg. is used to decide if adding a cluster is worthwhile
1223 */
1224 int
1225 nfs_rephead(int siz, struct nfsrv_descript *nd, struct nfssvc_sock *slp,
1226 int err, struct mbuf **mrq, struct mbuf **mbp, caddr_t *bposp)
1227 {
1228 u_int32_t *tl;
1229 struct mbuf *mreq;
1230 caddr_t bpos;
1231 struct mbuf *mb, *mb2;
1232
1233 siz += RPC_REPLYSIZ;
1234 mb = mreq = m_getl(max_hdr + siz, MB_WAIT, MT_DATA, M_PKTHDR, NULL);
1235 mreq->m_pkthdr.len = 0;
1236 /*
1237 * If this is not a cluster, try and leave leading space
1238 * for the lower level headers.
1239 */
1240 if ((max_hdr + siz) < MINCLSIZE)
1241 mreq->m_data += max_hdr;
1242 tl = mtod(mreq, u_int32_t *);
1243 mreq->m_len = 6 * NFSX_UNSIGNED;
1244 bpos = ((caddr_t)tl) + mreq->m_len;
1245 *tl++ = txdr_unsigned(nd->nd_retxid);
1246 *tl++ = rpc_reply;
1247 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1248 *tl++ = rpc_msgdenied;
1249 if (err & NFSERR_AUTHERR) {
1250 *tl++ = rpc_autherr;
1251 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1252 mreq->m_len -= NFSX_UNSIGNED;
1253 bpos -= NFSX_UNSIGNED;
1254 } else {
1255 *tl++ = rpc_mismatch;
1256 *tl++ = txdr_unsigned(RPC_VER2);
1257 *tl = txdr_unsigned(RPC_VER2);
1258 }
1259 } else {
1260 *tl++ = rpc_msgaccepted;
1261
1262 /*
1263 * For Kerberos authentication, we must send the nickname
1264 * verifier back, otherwise just RPCAUTH_NULL.
1265 */
1266 if (nd->nd_flag & ND_KERBFULL) {
1267 struct nfsuid *nuidp;
1268 struct timeval ktvin, ktvout;
1269
1270 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1271 nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1272 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1273 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1274 &nuidp->nu_haddr, nd->nd_nam2)))
1275 break;
1276 }
1277 if (nuidp) {
1278 ktvin.tv_sec =
1279 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1280 ktvin.tv_usec =
1281 txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1282
1283 /*
1284 * Encrypt the timestamp in ecb mode using the
1285 * session key.
1286 */
1287 #ifdef NFSKERB
1288 XXX
1289 #endif
1290
1291 *tl++ = rpc_auth_kerb;
1292 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1293 *tl = ktvout.tv_sec;
1294 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1295 *tl++ = ktvout.tv_usec;
1296 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1297 } else {
1298 *tl++ = 0;
1299 *tl++ = 0;
1300 }
1301 } else {
1302 *tl++ = 0;
1303 *tl++ = 0;
1304 }
1305 switch (err) {
1306 case EPROGUNAVAIL:
1307 *tl = txdr_unsigned(RPC_PROGUNAVAIL);
1308 break;
1309 case EPROGMISMATCH:
1310 *tl = txdr_unsigned(RPC_PROGMISMATCH);
1311 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1312 *tl++ = txdr_unsigned(2);
1313 *tl = txdr_unsigned(3);
1314 break;
1315 case EPROCUNAVAIL:
1316 *tl = txdr_unsigned(RPC_PROCUNAVAIL);
1317 break;
1318 case EBADRPC:
1319 *tl = txdr_unsigned(RPC_GARBAGE);
1320 break;
1321 default:
1322 *tl = 0;
1323 if (err != NFSERR_RETVOID) {
1324 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1325 if (err)
1326 *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1327 else
1328 *tl = 0;
1329 }
1330 break;
1331 };
1332 }
1333
1334 if (mrq != NULL)
1335 *mrq = mreq;
1336 *mbp = mb;
1337 *bposp = bpos;
1338 if (err != 0 && err != NFSERR_RETVOID)
1339 nfsstats.srvrpc_errs++;
1340 return (0);
1341 }
1342
1343
1344 #endif /* NFS_NOSERVER */
1345 /*
1346 * Nfs timer routine
1347 * Scan the nfsreq list and retranmit any requests that have timed out
1348 * To avoid retransmission attempts on STREAM sockets (in the future) make
1349 * sure to set the r_retry field to 0 (implies nm_retry == 0).
1350 */
1351 void
1352 nfs_timer(void *arg /* never used */)
1353 {
1354 struct nfsreq *rep;
1355 struct mbuf *m;
1356 struct socket *so;
1357 struct nfsmount *nmp;
1358 int timeo;
1359 int error;
1360 #ifndef NFS_NOSERVER
1361 struct nfssvc_sock *slp;
1362 u_quad_t cur_usec;
1363 #endif /* NFS_NOSERVER */
1364 struct thread *td = &thread0; /* XXX for credentials, will break if sleep */
1365
1366 crit_enter();
1367 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
1368 nmp = rep->r_nmp;
1369 if (rep->r_mrep || (rep->r_flags & (R_SOFTTERM|R_MASKTIMER)))
1370 continue;
1371 rep->r_flags |= R_LOCKED;
1372 if (nfs_sigintr(nmp, rep, rep->r_td)) {
1373 nfs_softterm(rep);
1374 goto skip;
1375 }
1376 if (rep->r_rtt >= 0) {
1377 rep->r_rtt++;
1378 if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1379 timeo = nmp->nm_timeo;
1380 else
1381 timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1382 if (nmp->nm_timeouts > 0)
1383 timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1384 if (rep->r_rtt <= timeo)
1385 goto skip;
1386 if (nmp->nm_timeouts < 8)
1387 nmp->nm_timeouts++;
1388 }
1389 /*
1390 * Check for server not responding
1391 */
1392 if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1393 rep->r_rexmit > nmp->nm_deadthresh) {
1394 nfs_msg(rep->r_td,
1395 nmp->nm_mountp->mnt_stat.f_mntfromname,
1396 "not responding");
1397 rep->r_flags |= R_TPRINTFMSG;
1398 }
1399 if (rep->r_rexmit >= rep->r_retry) { /* too many */
1400 nfsstats.rpctimeouts++;
1401 nfs_softterm(rep);
1402 goto skip;
1403 }
1404 if (nmp->nm_sotype != SOCK_DGRAM) {
1405 if (++rep->r_rexmit > NFS_MAXREXMIT)
1406 rep->r_rexmit = NFS_MAXREXMIT;
1407 goto skip;
1408 }
1409 if ((so = nmp->nm_so) == NULL)
1410 goto skip;
1411
1412 /*
1413 * If there is enough space and the window allows..
1414 * Resend it
1415 * Set r_rtt to -1 in case we fail to send it now.
1416 */
1417 rep->r_rtt = -1;
1418 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1419 ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1420 (rep->r_flags & R_SENT) ||
1421 nmp->nm_sent < nmp->nm_cwnd) &&
1422 (m = m_copym(rep->r_mreq, 0, M_COPYALL, MB_DONTWAIT))){
1423 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
1424 error = so_pru_send(so, 0, m, (struct sockaddr *)0,
1425 (struct mbuf *)0, td);
1426 else
1427 error = so_pru_send(so, 0, m, nmp->nm_nam,
1428 (struct mbuf *)0, td);
1429 if (error) {
1430 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
1431 so->so_error = 0;
1432 } else if (rep->r_mrep == NULL) {
1433 /*
1434 * Iff first send, start timing
1435 * else turn timing off, backoff timer
1436 * and divide congestion window by 2.
1437 *
1438 * It is possible for the so_pru_send() to
1439 * block and for us to race a reply so we
1440 * only do this if the reply field has not
1441 * been filled in. R_LOCKED will prevent
1442 * the request from being ripped out from under
1443 * us entirely.
1444 */
1445 if (rep->r_flags & R_SENT) {
1446 rep->r_flags &= ~R_TIMING;
1447 if (++rep->r_rexmit > NFS_MAXREXMIT)
1448 rep->r_rexmit = NFS_MAXREXMIT;
1449 nmp->nm_cwnd >>= 1;
1450 if (nmp->nm_cwnd < NFS_CWNDSCALE)
1451 nmp->nm_cwnd = NFS_CWNDSCALE;
1452 nfsstats.rpcretries++;
1453 } else {
1454 rep->r_flags |= R_SENT;
1455 nmp->nm_sent += NFS_CWNDSCALE;
1456 }
1457 rep->r_rtt = 0;
1458 }
1459 }
1460 skip:
1461 rep->r_flags &= ~R_LOCKED;
1462 }
1463 #ifndef NFS_NOSERVER
1464
1465 /*
1466 * Scan the write gathering queues for writes that need to be
1467 * completed now.
1468 */
1469 cur_usec = nfs_curusec();
1470 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
1471 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1472 nfsrv_wakenfsd(slp, 1);
1473 }
1474 #endif /* NFS_NOSERVER */
1475
1476 /*
1477 * Due to possible blocking, a client operation may be waiting for
1478 * us to finish processing this request so it can remove it.
1479 */
1480 if (nfs_timer_raced) {
1481 nfs_timer_raced = 0;
1482 wakeup(&nfs_timer_raced);
1483 }
1484 crit_exit();
1485 callout_reset(&nfs_timer_handle, nfs_ticks, nfs_timer, NULL);
1486 }
1487
1488 /*
1489 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1490 * wait for all requests to complete. This is used by forced unmounts
1491 * to terminate any outstanding RPCs.
1492 */
1493 int
1494 nfs_nmcancelreqs(struct nfsmount *nmp)
1495 {
1496 struct nfsreq *req;
1497 int i;
1498
1499 crit_enter();
1500 TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
1501 if (nmp != req->r_nmp || req->r_mrep != NULL ||
1502 (req->r_flags & R_SOFTTERM)) {
1503 continue;
1504 }
1505 nfs_softterm(req);
1506 }
1507 crit_exit();
1508
1509 for (i = 0; i < 30; i++) {
1510 crit_enter();
1511 TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
1512 if (nmp == req->r_nmp)
1513 break;
1514 }
1515 crit_exit();
1516 if (req == NULL)
1517 return (0);
1518 tsleep(&lbolt, 0, "nfscancel", 0);
1519 }
1520 return (EBUSY);
1521 }
1522
1523 /*
1524 * Flag a request as being about to terminate (due to NFSMNT_INT/NFSMNT_SOFT).
1525 * The nm_send count is decremented now to avoid deadlocks when the process in
1526 * soreceive() hasn't yet managed to send its own request.
1527 *
1528 * This routine must be called at splsoftclock() to protect r_flags and
1529 * nm_sent.
1530 */
1531
1532 static void
1533 nfs_softterm(struct nfsreq *rep)
1534 {
1535 rep->r_flags |= R_SOFTTERM;
1536
1537 if (rep->r_flags & R_SENT) {
1538 rep->r_nmp->nm_sent -= NFS_CWNDSCALE;
1539 rep->r_flags &= ~R_SENT;
1540 }
1541 }
1542
1543 /*
1544 * Test for a termination condition pending on the process.
1545 * This is used for NFSMNT_INT mounts.
1546 */
1547 int
1548 nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct thread *td)
1549 {
1550 sigset_t tmpset;
1551 struct proc *p;
1552 struct lwp *lp;
1553
1554 if (rep && (rep->r_flags & R_SOFTTERM))
1555 return (EINTR);
1556 /* Terminate all requests while attempting a forced unmount. */
1557 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1558 return (EINTR);
1559 if (!(nmp->nm_flag & NFSMNT_INT))
1560 return (0);
1561 /* td might be NULL YYY */
1562 if (td == NULL || (p = td->td_proc) == NULL)
1563 return (0);
1564
1565 lp = td->td_lwp;
1566 tmpset = lwp_sigpend(lp);
1567 SIGSETNAND(tmpset, lp->lwp_sigmask);
1568 SIGSETNAND(tmpset, p->p_sigignore);
1569 if (SIGNOTEMPTY(tmpset) && NFSINT_SIGMASK(tmpset))
1570 return (EINTR);
1571
1572 return (0);
1573 }
1574
1575 /*
1576 * Lock a socket against others.
1577 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1578 * and also to avoid race conditions between the processes with nfs requests
1579 * in progress when a reconnect is necessary.
1580 */
1581 int
1582 nfs_sndlock(struct nfsreq *rep)
1583 {
1584 int *statep = &rep->r_nmp->nm_state;
1585 struct thread *td;
1586 int slptimeo;
1587 int slpflag;
1588 int error;
1589
1590 slpflag = 0;
1591 slptimeo = 0;
1592 td = rep->r_td;
1593 if (rep->r_nmp->nm_flag & NFSMNT_INT)
1594 slpflag = PCATCH;
1595
1596 error = 0;
1597 crit_enter();
1598 while (*statep & NFSSTA_SNDLOCK) {
1599 *statep |= NFSSTA_WANTSND;
1600 if (nfs_sigintr(rep->r_nmp, rep, td)) {
1601 error = EINTR;
1602 break;
1603 }
1604 tsleep((caddr_t)statep, slpflag, "nfsndlck", slptimeo);
1605 if (slpflag == PCATCH) {
1606 slpflag = 0;
1607 slptimeo = 2 * hz;
1608 }
1609 }
1610 /* Always fail if our request has been cancelled. */
1611 if ((rep->r_flags & R_SOFTTERM))
1612 error = EINTR;
1613 if (error == 0)
1614 *statep |= NFSSTA_SNDLOCK;
1615 crit_exit();
1616 return (error);
1617 }
1618
1619 /*
1620 * Unlock the stream socket for others.
1621 */
1622 void
1623 nfs_sndunlock(struct nfsreq *rep)
1624 {
1625 int *statep = &rep->r_nmp->nm_state;
1626
1627 if ((*statep & NFSSTA_SNDLOCK) == 0)
1628 panic("nfs sndunlock");
1629 crit_enter();
1630 *statep &= ~NFSSTA_SNDLOCK;
1631 if (*statep & NFSSTA_WANTSND) {
1632 *statep &= ~NFSSTA_WANTSND;
1633 wakeup((caddr_t)statep);
1634 }
1635 crit_exit();
1636 }
1637
1638 static int
1639 nfs_rcvlock(struct nfsreq *rep)
1640 {
1641 int *statep = &rep->r_nmp->nm_state;
1642 int slpflag;
1643 int slptimeo;
1644 int error;
1645
1646 /*
1647 * Unconditionally check for completion in case another nfsiod
1648 * get the packet while the caller was blocked, before the caller
1649 * called us. Packet reception is handled by mainline code which
1650 * is protected by the BGL at the moment.
1651 *
1652 * We do not strictly need the second check just before the
1653 * tsleep(), but it's good defensive programming.
1654 */
1655 if (rep->r_mrep != NULL)
1656 return (EALREADY);
1657
1658 if (rep->r_nmp->nm_flag & NFSMNT_INT)
1659 slpflag = PCATCH;
1660 else
1661 slpflag = 0;
1662 slptimeo = 0;
1663 error = 0;
1664 crit_enter();
1665 while (*statep & NFSSTA_RCVLOCK) {
1666 if (nfs_sigintr(rep->r_nmp, rep, rep->r_td)) {
1667 error = EINTR;
1668 break;
1669 }
1670 if (rep->r_mrep != NULL) {
1671 error = EALREADY;
1672 break;
1673 }
1674 *statep |= NFSSTA_WANTRCV;
1675 tsleep((caddr_t)statep, slpflag, "nfsrcvlk", slptimeo);
1676 /*
1677 * If our reply was recieved while we were sleeping,
1678 * then just return without taking the lock to avoid a
1679 * situation where a single iod could 'capture' the
1680 * recieve lock.
1681 */
1682 if (rep->r_mrep != NULL) {
1683 error = EALREADY;
1684 break;
1685 }
1686 if (slpflag == PCATCH) {
1687 slpflag = 0;
1688 slptimeo = 2 * hz;
1689 }
1690 }
1691 if (error == 0) {
1692 *statep |= NFSSTA_RCVLOCK;
1693 rep->r_nmp->nm_rcvlock_td = curthread; /* DEBUGGING */
1694 }
1695 crit_exit();
1696 return (error);
1697 }
1698
1699 /*
1700 * Unlock the stream socket for others.
1701 */
1702 static void
1703 nfs_rcvunlock(struct nfsreq *rep)
1704 {
1705 int *statep = &rep->r_nmp->nm_state;
1706
1707 if ((*statep & NFSSTA_RCVLOCK) == 0)
1708 panic("nfs rcvunlock");
1709 crit_enter();
1710 rep->r_nmp->nm_rcvlock_td = (void *)-1; /* DEBUGGING */
1711 *statep &= ~NFSSTA_RCVLOCK;
1712 if (*statep & NFSSTA_WANTRCV) {
1713 *statep &= ~NFSSTA_WANTRCV;
1714 wakeup((caddr_t)statep);
1715 }
1716 crit_exit();
1717 }
1718
1719 /*
1720 * nfs_realign:
1721 *
1722 * Check for badly aligned mbuf data and realign by copying the unaligned
1723 * portion of the data into a new mbuf chain and freeing the portions
1724 * of the old chain that were replaced.
1725 *
1726 * We cannot simply realign the data within the existing mbuf chain
1727 * because the underlying buffers may contain other rpc commands and
1728 * we cannot afford to overwrite them.
1729 *
1730 * We would prefer to avoid this situation entirely. The situation does
1731 * not occur with NFS/UDP and is supposed to only occassionally occur
1732 * with TCP. Use vfs.nfs.realign_count and realign_test to check this.
1733 */
1734 static void
1735 nfs_realign(struct mbuf **pm, int hsiz)
1736 {
1737 struct mbuf *m;
1738 struct mbuf *n = NULL;
1739 int off = 0;
1740
1741 ++nfs_realign_test;
1742
1743 while ((m = *pm) != NULL) {
1744 if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) {
1745 n = m_getl(m->m_len, MB_WAIT, MT_DATA, 0, NULL);
1746 n->m_len = 0;
1747 break;
1748 }
1749 pm = &m->m_next;
1750 }
1751
1752 /*
1753 * If n is non-NULL, loop on m copying data, then replace the
1754 * portion of the chain that had to be realigned.
1755 */
1756 if (n != NULL) {
1757 ++nfs_realign_count;
1758 while (m) {
1759 m_copyback(n, off, m->m_len, mtod(m, caddr_t));
1760 off += m->m_len;
1761 m = m->m_next;
1762 }
1763 m_freem(*pm);
1764 *pm = n;
1765 }
1766 }
1767
1768 #ifndef NFS_NOSERVER
1769
1770 /*
1771 * Parse an RPC request
1772 * - verify it
1773 * - fill in the cred struct.
1774 */
1775 int
1776 nfs_getreq(struct nfsrv_descript *nd, struct nfsd *nfsd, int has_header)
1777 {
1778 int len, i;
1779 u_int32_t *tl;
1780 int32_t t1;
1781 struct uio uio;
1782 struct iovec iov;
1783 caddr_t dpos, cp2, cp;
1784 u_int32_t nfsvers, auth_type;
1785 uid_t nickuid;
1786 int error = 0, ticklen;
1787 struct mbuf *mrep, *md;
1788 struct nfsuid *nuidp;
1789 struct timeval tvin, tvout;
1790 #if 0 /* until encrypted keys are implemented */
1791 NFSKERBKEYSCHED_T keys; /* stores key schedule */
1792 #endif
1793
1794 mrep = nd->nd_mrep;
1795 md = nd->nd_md;
1796 dpos = nd->nd_dpos;
1797 if (has_header) {
1798 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1799 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1800 if (*tl++ != rpc_call) {
1801 m_freem(mrep);
1802 return (EBADRPC);
1803 }
1804 } else
1805 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1806 nd->nd_repstat = 0;
1807 nd->nd_flag = 0;
1808 if (*tl++ != rpc_vers) {
1809 nd->nd_repstat = ERPCMISMATCH;
1810 nd->nd_procnum = NFSPROC_NOOP;
1811 return (0);
1812 }
1813 if (*tl != nfs_prog) {
1814 nd->nd_repstat = EPROGUNAVAIL;
1815 nd->nd_procnum = NFSPROC_NOOP;
1816 return (0);
1817 }
1818 tl++;
1819 nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1820 if (nfsvers < NFS_VER2 || nfsvers > NFS_VER3) {
1821 nd->nd_repstat = EPROGMISMATCH;
1822 nd->nd_procnum = NFSPROC_NOOP;
1823 return (0);
1824 }
1825 if (nfsvers == NFS_VER3)
1826 nd->nd_flag = ND_NFSV3;
1827 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1828 if (nd->nd_procnum == NFSPROC_NULL)
1829 return (0);
1830 if (nd->nd_procnum >= NFS_NPROCS ||
1831 (nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1832 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1833 nd->nd_repstat = EPROCUNAVAIL;
1834 nd->nd_procnum = NFSPROC_NOOP;
1835 return (0);
1836 }
1837 if ((nd->nd_flag & ND_NFSV3) == 0)
1838 nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1839 auth_type = *tl++;
1840 len = fxdr_unsigned(int, *tl++);
1841 if (len < 0 || len > RPCAUTH_MAXSIZ) {
1842 m_freem(mrep);
1843 return (EBADRPC);
1844 }
1845
1846 nd->nd_flag &= ~ND_KERBAUTH;
1847 /*
1848 * Handle auth_unix or auth_kerb.
1849 */
1850 if (auth_type == rpc_auth_unix) {
1851 len = fxdr_unsigned(int, *++tl);
1852 if (len < 0 || len > NFS_MAXNAMLEN) {
1853 m_freem(mrep);
1854 return (EBADRPC);
1855 }
1856 nfsm_adv(nfsm_rndup(len));
1857 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1858 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred));
1859 nd->nd_cr.cr_ref = 1;
1860 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1861 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1862 len = fxdr_unsigned(int, *tl);
1863 if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1864 m_freem(mrep);
1865 return (EBADRPC);
1866 }
1867 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1868 for (i = 1; i <= len; i++)
1869 if (i < NGROUPS)
1870 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1871 else
1872 tl++;
1873 nd->nd_cr.cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1);
1874 if (nd->nd_cr.cr_ngroups > 1)
1875 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1876 len = fxdr_unsigned(int, *++tl);
1877 if (len < 0 || len > RPCAUTH_MAXSIZ) {
1878 m_freem(mrep);
1879 return (EBADRPC);
1880 }
1881 if (len > 0)
1882 nfsm_adv(nfsm_rndup(len));
1883 } else if (auth_type == rpc_auth_kerb) {
1884 switch (fxdr_unsigned(int, *tl++)) {
1885 case RPCAKN_FULLNAME:
1886 ticklen = fxdr_unsigned(int, *tl);
1887 *((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1888 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1889 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1890 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1891 m_freem(mrep);
1892 return (EBADRPC);
1893 }
1894 uio.uio_offset = 0;
1895 uio.uio_iov = &iov;
1896 uio.uio_iovcnt = 1;
1897 uio.uio_segflg = UIO_SYSSPACE;
1898 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1899 iov.iov_len = RPCAUTH_MAXSIZ - 4;
1900 nfsm_mtouio(&uio, uio.uio_resid);
1901 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1902 if (*tl++ != rpc_auth_kerb ||
1903 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1904 kprintf("Bad kerb verifier\n");
1905 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1906 nd->nd_procnum = NFSPROC_NOOP;
1907 return (0);
1908 }
1909 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1910 tl = (u_int32_t *)cp;
1911 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1912 kprintf("Not fullname kerb verifier\n");
1913 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1914 nd->nd_procnum = NFSPROC_NOOP;
1915 return (0);
1916 }
1917 cp += NFSX_UNSIGNED;
1918 bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED);
1919 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1920 nd->nd_flag |= ND_KERBFULL;
1921 nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1922 break;
1923 case RPCAKN_NICKNAME:
1924 if (len != 2 * NFSX_UNSIGNED) {
1925 kprintf("Kerb nickname short\n");
1926 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1927 nd->nd_procnum = NFSPROC_NOOP;
1928 return (0);
1929 }
1930 nickuid = fxdr_unsigned(uid_t, *tl);
1931 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1932 if (*tl++ != rpc_auth_kerb ||
1933 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1934 kprintf("Kerb nick verifier bad\n");
1935 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1936 nd->nd_procnum = NFSPROC_NOOP;
1937 return (0);
1938 }
1939 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1940 tvin.tv_sec = *tl++;
1941 tvin.tv_usec = *tl;
1942
1943 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1944 nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1945 if (nuidp->nu_cr.cr_uid == nickuid &&
1946 (!nd->nd_nam2 ||
1947 netaddr_match(NU_NETFAM(nuidp),
1948 &nuidp->nu_haddr, nd->nd_nam2)))
1949 break;
1950 }
1951 if (!nuidp) {
1952 nd->nd_repstat =
1953 (NFSERR_AUTHERR|AUTH_REJECTCRED);
1954 nd->nd_procnum = NFSPROC_NOOP;
1955 return (0);
1956 }
1957
1958 /*
1959 * Now, decrypt the timestamp using the session key
1960 * and validate it.
1961 */
1962 #ifdef NFSKERB
1963 XXX
1964 #endif
1965
1966 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1967 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1968 if (nuidp->nu_expire < time_second ||
1969 nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1970 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1971 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1972 nuidp->nu_expire = 0;
1973 nd->nd_repstat =
1974 (NFSERR_AUTHERR|AUTH_REJECTVERF);
1975 nd->nd_procnum = NFSPROC_NOOP;
1976 return (0);
1977 }
1978 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1979 nd->nd_flag |= ND_KERBNICK;
1980 };
1981 } else {
1982 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1983 nd->nd_procnum = NFSPROC_NOOP;
1984 return (0);
1985 }
1986
1987 nd->nd_md = md;
1988 nd->nd_dpos = dpos;
1989 return (0);
1990 nfsmout:
1991 return (error);
1992 }
1993
1994 #endif
1995
1996 /*
1997 * Send a message to the originating process's terminal. The thread and/or
1998 * process may be NULL. YYY the thread should not be NULL but there may
1999 * still be some uio_td's that are still being passed as NULL through to
2000 * nfsm_request().
2001 */
2002 static int
2003 nfs_msg(struct thread *td, char *server, char *msg)
2004 {
2005 tpr_t tpr;
2006
2007 if (td && td->td_proc)
2008 tpr = tprintf_open(td->td_proc);
2009 else
2010 tpr = NULL;
2011 tprintf(tpr, "nfs server %s: %s\n", server, msg);
2012 tprintf_close(tpr);
2013 return (0);
2014 }
2015
2016 #ifndef NFS_NOSERVER
2017 /*
2018 * Socket upcall routine for the nfsd sockets.
2019 * The caddr_t arg is a pointer to the "struct nfssvc_sock".
2020 * Essentially do as much as possible non-blocking, else punt and it will
2021 * be called with MB_WAIT from an nfsd.
2022 */
2023 void
2024 nfsrv_rcv(struct socket *so, void *arg, int waitflag)
2025 {
2026 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
2027 struct mbuf *m;
2028 struct mbuf *mp;
2029 struct sockaddr *nam;
2030 struct uio auio;
2031 int flags, error;
2032 int nparallel_wakeup = 0;
2033
2034 if ((slp->ns_flag & SLP_VALID) == 0)
2035 return;
2036
2037 /*
2038 * Do not allow an infinite number of completed RPC records to build
2039 * up before we stop reading data from the socket. Otherwise we could
2040 * end up holding onto an unreasonable number of mbufs for requests
2041 * waiting for service.
2042 *
2043 * This should give pretty good feedback to the TCP
2044 * layer and prevents a memory crunch for other protocols.
2045 *
2046 * Note that the same service socket can be dispatched to several
2047 * nfs servers simultaniously.
2048 *
2049 * the tcp protocol callback calls us with MB_DONTWAIT.
2050 * nfsd calls us with MB_WAIT (typically).
2051 */
2052 if (waitflag == MB_DONTWAIT && slp->ns_numrec >= nfsd_waiting / 2 + 1) {
2053 slp->ns_flag |= SLP_NEEDQ;
2054 goto dorecs;
2055 }
2056
2057 /*
2058 * Handle protocol specifics to parse an RPC request. We always
2059 * pull from the socket using non-blocking I/O.
2060 */
2061 auio.uio_td = NULL;
2062 if (so->so_type == SOCK_STREAM) {
2063 /*
2064 * The data has to be read in an orderly fashion from a TCP
2065 * stream, unlike a UDP socket. It is possible for soreceive
2066 * and/or nfsrv_getstream() to block, so make sure only one
2067 * entity is messing around with the TCP stream at any given
2068 * moment. The receive sockbuf's lock in soreceive is not
2069 * sufficient.
2070 *
2071 * Note that this procedure can be called from any number of
2072 * NFS severs *OR* can be upcalled directly from a TCP
2073 * protocol thread.
2074 */
2075 if (slp->ns_flag & SLP_GETSTREAM) {
2076 slp->ns_flag |= SLP_NEEDQ;
2077 goto dorecs;
2078 }
2079 slp->ns_flag |= SLP_GETSTREAM;
2080
2081 /*
2082 * Do soreceive().
2083 */
2084 auio.uio_resid = 1000000000;
2085 flags = MSG_DONTWAIT;
2086 error = so_pru_soreceive(so, &nam, &auio, &mp, NULL, &flags);
2087 if (error || mp == (struct mbuf *)0) {
2088 if (error == EWOULDBLOCK)
2089 slp->ns_flag |= SLP_NEEDQ;
2090 else
2091 slp->ns_flag |= SLP_DISCONN;
2092 slp->ns_flag &= ~SLP_GETSTREAM;
2093 goto dorecs;
2094 }
2095 m = mp;
2096 if (slp->ns_rawend) {
2097 slp->ns_rawend->m_next = m;
2098 slp->ns_cc += 1000000000 - auio.uio_resid;
2099 } else {
2100 slp->ns_raw = m;
2101 slp->ns_cc = 1000000000 - auio.uio_resid;
2102 }
2103 while (m->m_next)
2104 m = m->m_next;
2105 slp->ns_rawend = m;
2106
2107 /*
2108 * Now try and parse as many record(s) as we can out of the
2109 * raw stream data.
2110 */
2111 error = nfsrv_getstream(slp, waitflag, &nparallel_wakeup);
2112 if (error) {
2113 if (error == EPERM)
2114 slp->ns_flag |= SLP_DISCONN;
2115 else
2116 slp->ns_flag |= SLP_NEEDQ;
2117 }
2118 slp->ns_flag &= ~SLP_GETSTREAM;
2119 } else {
2120 /*
2121 * For UDP soreceive typically pulls just one packet, loop
2122 * to get the whole batch.
2123 */
2124 do {
2125 auio.uio_resid = 1000000000;
2126 flags = MSG_DONTWAIT;
2127 error = so_pru_soreceive(so, &nam, &auio, &mp, NULL,
2128 &flags);
2129 if (mp) {
2130 struct nfsrv_rec *rec;
2131 int mf = (waitflag & MB_DONTWAIT) ?
2132 M_NOWAIT : M_WAITOK;
2133 rec = kmalloc(sizeof(struct nfsrv_rec),
2134 M_NFSRVDESC, mf);
2135 if (!rec) {
2136 if (nam)
2137 FREE(nam, M_SONAME);
2138 m_freem(mp);
2139 continue;
2140 }
2141 nfs_realign(&mp, 10 * NFSX_UNSIGNED);
2142 rec->nr_address = nam;
2143 rec->nr_packet = mp;
2144 STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
2145 ++slp->ns_numrec;
2146 ++nparallel_wakeup;
2147 }
2148 if (error) {
2149 if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2150 && error != EWOULDBLOCK) {
2151 slp->ns_flag |= SLP_DISCONN;
2152 goto dorecs;
2153 }
2154 }
2155 } while (mp);
2156 }
2157
2158 /*
2159 * If we were upcalled from the tcp protocol layer and we have
2160 * fully parsed records ready to go, or there is new data pending,
2161 * or something went wrong, try to wake up an nfsd thread to deal
2162 * with it.
2163 */
2164 dorecs:
2165 if (waitflag == MB_DONTWAIT && (slp->ns_numrec > 0
2166 || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) {
2167 nfsrv_wakenfsd(slp, nparallel_wakeup);
2168 }
2169 }
2170
2171 /*
2172 * Try and extract an RPC request from the mbuf data list received on a
2173 * stream socket. The "waitflag" argument indicates whether or not it
2174 * can sleep.
2175 */
2176 static int
2177 nfsrv_getstream(struct nfssvc_sock *slp, int waitflag, int *countp)
2178 {
2179 struct mbuf *m, **mpp;
2180 char *cp1, *cp2;
2181 int len;
2182 struct mbuf *om, *m2, *recm;
2183 u_int32_t recmark;
2184
2185 for (;;) {
2186 if (slp->ns_reclen == 0) {
2187 if (slp->ns_cc < NFSX_UNSIGNED)
2188 return (0);
2189 m = slp->ns_raw;
2190 if (m->m_len >= NFSX_UNSIGNED) {
2191 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
2192 m->m_data += NFSX_UNSIGNED;
2193 m->m_len -= NFSX_UNSIGNED;
2194 } else {
2195 cp1 = (caddr_t)&recmark;
2196 cp2 = mtod(m, caddr_t);
2197 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
2198 while (m->m_len == 0) {
2199 m = m->m_next;
2200 cp2 = mtod(m, caddr_t);
2201 }
2202 *cp1++ = *cp2++;
2203 m->m_data++;
2204 m->m_len--;
2205 }
2206 }
2207 slp->ns_cc -= NFSX_UNSIGNED;
2208 recmark = ntohl(recmark);
2209 slp->ns_reclen = recmark & ~0x80000000;
2210 if (recmark & 0x80000000)
2211 slp->ns_flag |= SLP_LASTFRAG;
2212 else
2213 slp->ns_flag &= ~SLP_LASTFRAG;
2214 if (slp->ns_reclen > NFS_MAXPACKET || slp->ns_reclen <= 0) {
2215 log(LOG_ERR, "%s (%d) from nfs client\n",
2216 "impossible packet length",
2217 slp->ns_reclen);
2218 return (EPERM);
2219 }
2220 }
2221
2222 /*
2223 * Now get the record part.
2224 *
2225 * Note that slp->ns_reclen may be 0. Linux sometimes
2226 * generates 0-length RPCs
2227 */
2228 recm = NULL;
2229 if (slp->ns_cc == slp->ns_reclen) {
2230 recm = slp->ns_raw;
2231 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2232 slp->ns_cc = slp->ns_reclen = 0;
2233 } else if (slp->ns_cc > slp->ns_reclen) {
2234 len = 0;
2235 m = slp->ns_raw;
2236 om = (struct mbuf *)0;
2237
2238 while (len < slp->ns_reclen) {
2239 if ((len + m->m_len) > slp->ns_reclen) {
2240 m2 = m_copym(m, 0, slp->ns_reclen - len,
2241 waitflag);
2242 if (m2) {
2243 if (om) {
2244 om->m_next = m2;
2245 recm = slp->ns_raw;
2246 } else
2247 recm = m2;
2248 m->m_data += slp->ns_reclen - len;
2249 m->m_len -= slp->ns_reclen - len;
2250 len = slp->ns_reclen;
2251 } else {
2252 return (EWOULDBLOCK);
2253 }
2254 } else if ((len + m->m_len) == slp->ns_reclen) {
2255 om = m;
2256 len += m->m_len;
2257 m = m->m_next;
2258 recm = slp->ns_raw;
2259 om->m_next = (struct mbuf *)0;
2260 } else {
2261 om = m;
2262 len += m->m_len;
2263 m = m->m_next;
2264 }
2265 }
2266 slp->ns_raw = m;
2267 slp->ns_cc -= len;
2268 slp->ns_reclen = 0;
2269 } else {
2270 return (0);
2271 }
2272
2273 /*
2274 * Accumulate the fragments into a record.
2275 */
2276 mpp = &slp->ns_frag;
2277 while (*mpp)
2278 mpp = &((*mpp)->m_next);
2279 *mpp = recm;
2280 if (slp->ns_flag & SLP_LASTFRAG) {
2281 struct nfsrv_rec *rec;
2282 int mf = (waitflag & MB_DONTWAIT) ? M_NOWAIT : M_WAITOK;
2283 rec = kmalloc(sizeof(struct nfsrv_rec), M_NFSRVDESC, mf);
2284 if (!rec) {
2285 m_freem(slp->ns_frag);
2286 } else {
2287 nfs_realign(&slp->ns_frag, 10 * NFSX_UNSIGNED);
2288 rec->nr_address = (struct sockaddr *)0;
2289 rec->nr_packet = slp->ns_frag;
2290 STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
2291 ++slp->ns_numrec;
2292 ++*countp;
2293 }
2294 slp->ns_frag = (struct mbuf *)0;
2295 }
2296 }
2297 }
2298
2299 /*
2300 * Parse an RPC header.
2301 */
2302 int
2303 nfsrv_dorec(struct nfssvc_sock *slp, struct nfsd *nfsd,
2304 struct nfsrv_descript **ndp)
2305 {
2306 struct nfsrv_rec *rec;
2307 struct mbuf *m;
2308 struct sockaddr *nam;
2309 struct nfsrv_descript *nd;
2310 int error;
2311
2312 *ndp = NULL;
2313 if ((slp->ns_flag & SLP_VALID) == 0 || !STAILQ_FIRST(&slp->ns_rec))
2314 return (ENOBUFS);
2315 rec = STAILQ_FIRST(&slp->ns_rec);
2316 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
2317 KKASSERT(slp->ns_numrec > 0);
2318 --slp->ns_numrec;
2319 nam = rec->nr_address;
2320 m = rec->nr_packet;
2321 kfree(rec, M_NFSRVDESC);
2322 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2323 M_NFSRVDESC, M_WAITOK);
2324 nd->nd_md = nd->nd_mrep = m;
2325 nd->nd_nam2 = nam;
2326 nd->nd_dpos = mtod(m, caddr_t);
2327 error = nfs_getreq(nd, nfsd, TRUE);
2328 if (error) {
2329 if (nam) {
2330 FREE(nam, M_SONAME);
2331 }
2332 kfree((caddr_t)nd, M_NFSRVDESC);
2333 return (error);
2334 }
2335 *ndp = nd;
2336 nfsd->nfsd_nd = nd;
2337 return (0);
2338 }
2339
2340 /*
2341 * Try to assign service sockets to nfsd threads based on the number
2342 * of new rpc requests that have been queued on the service socket.
2343 *
2344 * If no nfsd's are available or additonal requests are pending, set the
2345 * NFSD_CHECKSLP flag so that one of the running nfsds will go look for
2346 * the work in the nfssvc_sock list when it is finished processing its
2347 * current work. This flag is only cleared when an nfsd can not find
2348 * any new work to perform.
2349 */
2350 void
2351 nfsrv_wakenfsd(struct nfssvc_sock *slp, int nparallel)
2352 {
2353 struct nfsd *nd;
2354
2355 if ((slp->ns_flag & SLP_VALID) == 0)
2356 return;
2357 if (nparallel <= 1)
2358 nparallel = 1;
2359 TAILQ_FOREACH(nd, &nfsd_head, nfsd_chain) {
2360 if (nd->nfsd_flag & NFSD_WAITING) {
2361 nd->nfsd_flag &= ~NFSD_WAITING;
2362 if (nd->nfsd_slp)
2363 panic("nfsd wakeup");
2364 slp->ns_sref++;
2365 nd->nfsd_slp = slp;
2366 wakeup((caddr_t)nd);
2367 if (--nparallel == 0)
2368 break;
2369 }
2370 }
2371 if (nparallel) {
2372 slp->ns_flag |= SLP_DOREC;
2373 nfsd_head_flag |= NFSD_CHECKSLP;
2374 }
2375 }
2376 #endif /* NFS_NOSERVER */
DragonFly BSD