?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
1:1 Userland threading stage 2.20/4:
[dragonfly.git] / sys / vfs / nfs / nfs_socket.c
blob5daac501847a14642f80b91d62482feb3c0f3c0d
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 1.41 2007/02/21 15:46:48 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
65 #include <netinet/in.h>
66 #include <netinet/tcp.h>
67 #include <sys/thread2.h>
68
69 #include "rpcv2.h"
70 #include "nfsproto.h"
71 #include "nfs.h"
72 #include "xdr_subs.h"
73 #include "nfsm_subs.h"
74 #include "nfsmount.h"
75 #include "nfsnode.h"
76 #include "nfsrtt.h"
77
78 #define TRUE 1
79 #define FALSE 0
80
81 /*
82 * Estimate rto for an nfs rpc sent via. an unreliable datagram.
83 * Use the mean and mean deviation of rtt for the appropriate type of rpc
84 * for the frequent rpcs and a default for the others.
85 * The justification for doing "other" this way is that these rpcs
86 * happen so infrequently that timer est. would probably be stale.
87 * Also, since many of these rpcs are
88 * non-idempotent, a conservative timeout is desired.
89 * getattr, lookup - A+2D
90 * read, write - A+4D
91 * other - nm_timeo
92 */
93 #define NFS_RTO(n, t) \
94 ((t) == 0 ? (n)->nm_timeo : \
95 ((t) < 3 ? \
96 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
97 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
98 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
99 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
100 /*
101 * External data, mostly RPC constants in XDR form
102 */
103 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
104 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
105 rpc_auth_kerb;
106 extern u_int32_t nfs_prog;
107 extern struct nfsstats nfsstats;
108 extern int nfsv3_procid[NFS_NPROCS];
109 extern int nfs_ticks;
110
111 /*
112 * Defines which timer to use for the procnum.
113 * 0 - default
114 * 1 - getattr
115 * 2 - lookup
116 * 3 - read
117 * 4 - write
118 */
119 static int proct[NFS_NPROCS] = {
120 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
121 0, 0, 0,
122 };
123
124 static int nfs_realign_test;
125 static int nfs_realign_count;
126 static int nfs_bufpackets = 4;
127 static int nfs_timer_raced;
128
129 SYSCTL_DECL(_vfs_nfs);
130
131 SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_test, CTLFLAG_RW, &nfs_realign_test, 0, "");
132 SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_count, CTLFLAG_RW, &nfs_realign_count, 0, "");
133 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0, "");
134
135
136 /*
137 * There is a congestion window for outstanding rpcs maintained per mount
138 * point. The cwnd size is adjusted in roughly the way that:
139 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
140 * SIGCOMM '88". ACM, August 1988.
141 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
142 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
143 * of rpcs is in progress.
144 * (The sent count and cwnd are scaled for integer arith.)
145 * Variants of "slow start" were tried and were found to be too much of a
146 * performance hit (ave. rtt 3 times larger),
147 * I suspect due to the large rtt that nfs rpcs have.
148 */
149 #define NFS_CWNDSCALE 256
150 #define NFS_MAXCWND (NFS_CWNDSCALE * 32)
151 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
152 int nfsrtton = 0;
153 struct nfsrtt nfsrtt;
154 struct callout nfs_timer_handle;
155
156 static int nfs_msg (struct thread *,char *,char *);
157 static int nfs_rcvlock (struct nfsreq *);
158 static void nfs_rcvunlock (struct nfsreq *);
159 static void nfs_realign (struct mbuf **pm, int hsiz);
160 static int nfs_receive (struct nfsreq *rep, struct sockaddr **aname,
161 struct mbuf **mp);
162 static void nfs_softterm (struct nfsreq *rep);
163 static int nfs_reconnect (struct nfsreq *rep);
164 #ifndef NFS_NOSERVER
165 static int nfsrv_getstream (struct nfssvc_sock *, int, int *);
166
167 int (*nfsrv3_procs[NFS_NPROCS]) (struct nfsrv_descript *nd,
168 struct nfssvc_sock *slp,
169 struct thread *td,
170 struct mbuf **mreqp) = {
171 nfsrv_null,
172 nfsrv_getattr,
173 nfsrv_setattr,
174 nfsrv_lookup,
175 nfsrv3_access,
176 nfsrv_readlink,
177 nfsrv_read,
178 nfsrv_write,
179 nfsrv_create,
180 nfsrv_mkdir,
181 nfsrv_symlink,
182 nfsrv_mknod,
183 nfsrv_remove,
184 nfsrv_rmdir,
185 nfsrv_rename,
186 nfsrv_link,
187 nfsrv_readdir,
188 nfsrv_readdirplus,
189 nfsrv_statfs,
190 nfsrv_fsinfo,
191 nfsrv_pathconf,
192 nfsrv_commit,
193 nfsrv_noop,
194 nfsrv_noop,
195 nfsrv_noop,
196 nfsrv_noop
197 };
198 #endif /* NFS_NOSERVER */
199
200 /*
201 * Initialize sockets and congestion for a new NFS connection.
202 * We do not free the sockaddr if error.
203 */
204 int
205 nfs_connect(struct nfsmount *nmp, struct nfsreq *rep)
206 {
207 struct socket *so;
208 int error, rcvreserve, sndreserve;
209 int pktscale;
210 struct sockaddr *saddr;
211 struct sockaddr_in *sin;
212 struct thread *td = &thread0; /* only used for socreate and sobind */
213
214 nmp->nm_so = (struct socket *)0;
215 saddr = nmp->nm_nam;
216 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
217 nmp->nm_soproto, td);
218 if (error)
219 goto bad;
220 so = nmp->nm_so;
221 nmp->nm_soflags = so->so_proto->pr_flags;
222
223 /*
224 * Some servers require that the client port be a reserved port number.
225 */
226 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
227 struct sockopt sopt;
228 int ip;
229 struct sockaddr_in ssin;
230
231 bzero(&sopt, sizeof sopt);
232 ip = IP_PORTRANGE_LOW;
233 sopt.sopt_level = IPPROTO_IP;
234 sopt.sopt_name = IP_PORTRANGE;
235 sopt.sopt_val = (void *)&ip;
236 sopt.sopt_valsize = sizeof(ip);
237 sopt.sopt_td = NULL;
238 error = sosetopt(so, &sopt);
239 if (error)
240 goto bad;
241 bzero(&ssin, sizeof ssin);
242 sin = &ssin;
243 sin->sin_len = sizeof (struct sockaddr_in);
244 sin->sin_family = AF_INET;
245 sin->sin_addr.s_addr = INADDR_ANY;
246 sin->sin_port = htons(0);
247 error = sobind(so, (struct sockaddr *)sin, td);
248 if (error)
249 goto bad;
250 bzero(&sopt, sizeof sopt);
251 ip = IP_PORTRANGE_DEFAULT;
252 sopt.sopt_level = IPPROTO_IP;
253 sopt.sopt_name = IP_PORTRANGE;
254 sopt.sopt_val = (void *)&ip;
255 sopt.sopt_valsize = sizeof(ip);
256 sopt.sopt_td = NULL;
257 error = sosetopt(so, &sopt);
258 if (error)
259 goto bad;
260 }
261
262 /*
263 * Protocols that do not require connections may be optionally left
264 * unconnected for servers that reply from a port other than NFS_PORT.
265 */
266 if (nmp->nm_flag & NFSMNT_NOCONN) {
267 if (nmp->nm_soflags & PR_CONNREQUIRED) {
268 error = ENOTCONN;
269 goto bad;
270 }
271 } else {
272 error = soconnect(so, nmp->nm_nam, td);
273 if (error)
274 goto bad;
275
276 /*
277 * Wait for the connection to complete. Cribbed from the
278 * connect system call but with the wait timing out so
279 * that interruptible mounts don't hang here for a long time.
280 */
281 crit_enter();
282 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
283 (void) tsleep((caddr_t)&so->so_timeo, 0,
284 "nfscon", 2 * hz);
285 if ((so->so_state & SS_ISCONNECTING) &&
286 so->so_error == 0 && rep &&
287 (error = nfs_sigintr(nmp, rep, rep->r_td)) != 0){
288 so->so_state &= ~SS_ISCONNECTING;
289 crit_exit();
290 goto bad;
291 }
292 }
293 if (so->so_error) {
294 error = so->so_error;
295 so->so_error = 0;
296 crit_exit();
297 goto bad;
298 }
299 crit_exit();
300 }
301 so->so_rcv.sb_timeo = (5 * hz);
302 so->so_snd.sb_timeo = (5 * hz);
303
304 /*
305 * Get buffer reservation size from sysctl, but impose reasonable
306 * limits.
307 */
308 pktscale = nfs_bufpackets;
309 if (pktscale < 2)
310 pktscale = 2;
311 if (pktscale > 64)
312 pktscale = 64;
313
314 if (nmp->nm_sotype == SOCK_DGRAM) {
315 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
316 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
317 NFS_MAXPKTHDR) * pktscale;
318 } else if (nmp->nm_sotype == SOCK_SEQPACKET) {
319 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
320 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
321 NFS_MAXPKTHDR) * pktscale;
322 } else {
323 if (nmp->nm_sotype != SOCK_STREAM)
324 panic("nfscon sotype");
325 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
326 struct sockopt sopt;
327 int val;
328
329 bzero(&sopt, sizeof sopt);
330 sopt.sopt_level = SOL_SOCKET;
331 sopt.sopt_name = SO_KEEPALIVE;
332 sopt.sopt_val = &val;
333 sopt.sopt_valsize = sizeof val;
334 val = 1;
335 sosetopt(so, &sopt);
336 }
337 if (so->so_proto->pr_protocol == IPPROTO_TCP) {
338 struct sockopt sopt;
339 int val;
340
341 bzero(&sopt, sizeof sopt);
342 sopt.sopt_level = IPPROTO_TCP;
343 sopt.sopt_name = TCP_NODELAY;
344 sopt.sopt_val = &val;
345 sopt.sopt_valsize = sizeof val;
346 val = 1;
347 sosetopt(so, &sopt);
348 }
349 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
350 sizeof (u_int32_t)) * pktscale;
351 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
352 sizeof (u_int32_t)) * pktscale;
353 }
354 error = soreserve(so, sndreserve, rcvreserve,
355 &td->td_proc->p_rlimit[RLIMIT_SBSIZE]);
356 if (error)
357 goto bad;
358 so->so_rcv.sb_flags |= SB_NOINTR;
359 so->so_snd.sb_flags |= SB_NOINTR;
360
361 /* Initialize other non-zero congestion variables */
362 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] =
363 nmp->nm_srtt[3] = (NFS_TIMEO << 3);
364 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
365 nmp->nm_sdrtt[3] = 0;
366 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */
367 nmp->nm_sent = 0;
368 nmp->nm_timeouts = 0;
369 return (0);
370
371 bad:
372 nfs_disconnect(nmp);
373 return (error);
374 }
375
376 /*
377 * Reconnect routine:
378 * Called when a connection is broken on a reliable protocol.
379 * - clean up the old socket
380 * - nfs_connect() again
381 * - set R_MUSTRESEND for all outstanding requests on mount point
382 * If this fails the mount point is DEAD!
383 * nb: Must be called with the nfs_sndlock() set on the mount point.
384 */
385 static int
386 nfs_reconnect(struct nfsreq *rep)
387 {
388 struct nfsreq *rp;
389 struct nfsmount *nmp = rep->r_nmp;
390 int error;
391
392 nfs_disconnect(nmp);
393 while ((error = nfs_connect(nmp, rep)) != 0) {
394 if (error == EINTR || error == ERESTART)
395 return (EINTR);
396 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
397 }
398
399 /*
400 * Loop through outstanding request list and fix up all requests
401 * on old socket.
402 */
403 crit_enter();
404 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
405 if (rp->r_nmp == nmp)
406 rp->r_flags |= R_MUSTRESEND;
407 }
408 crit_exit();
409 return (0);
410 }
411
412 /*
413 * NFS disconnect. Clean up and unlink.
414 */
415 void
416 nfs_disconnect(struct nfsmount *nmp)
417 {
418 struct socket *so;
419
420 if (nmp->nm_so) {
421 so = nmp->nm_so;
422 nmp->nm_so = (struct socket *)0;
423 soshutdown(so, 2);
424 soclose(so, FNONBLOCK);
425 }
426 }
427
428 void
429 nfs_safedisconnect(struct nfsmount *nmp)
430 {
431 struct nfsreq dummyreq;
432
433 bzero(&dummyreq, sizeof(dummyreq));
434 dummyreq.r_nmp = nmp;
435 dummyreq.r_td = NULL;
436 nfs_rcvlock(&dummyreq);
437 nfs_disconnect(nmp);
438 nfs_rcvunlock(&dummyreq);
439 }
440
441 /*
442 * This is the nfs send routine. For connection based socket types, it
443 * must be called with an nfs_sndlock() on the socket.
444 * "rep == NULL" indicates that it has been called from a server.
445 * For the client side:
446 * - return EINTR if the RPC is terminated, 0 otherwise
447 * - set R_MUSTRESEND if the send fails for any reason
448 * - do any cleanup required by recoverable socket errors (?)
449 * For the server side:
450 * - return EINTR or ERESTART if interrupted by a signal
451 * - return EPIPE if a connection is lost for connection based sockets (TCP...)
452 * - do any cleanup required by recoverable socket errors (?)
453 */
454 int
455 nfs_send(struct socket *so, struct sockaddr *nam, struct mbuf *top,
456 struct nfsreq *rep)
457 {
458 struct sockaddr *sendnam;
459 int error, soflags, flags;
460
461 if (rep) {
462 if (rep->r_flags & R_SOFTTERM) {
463 m_freem(top);
464 return (EINTR);
465 }
466 if ((so = rep->r_nmp->nm_so) == NULL) {
467 rep->r_flags |= R_MUSTRESEND;
468 m_freem(top);
469 return (0);
470 }
471 rep->r_flags &= ~R_MUSTRESEND;
472 soflags = rep->r_nmp->nm_soflags;
473 } else
474 soflags = so->so_proto->pr_flags;
475 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
476 sendnam = (struct sockaddr *)0;
477 else
478 sendnam = nam;
479 if (so->so_type == SOCK_SEQPACKET)
480 flags = MSG_EOR;
481 else
482 flags = 0;
483
484 error = so_pru_sosend(so, sendnam, NULL, top, NULL, flags,
485 curthread /*XXX*/);
486 /*
487 * ENOBUFS for dgram sockets is transient and non fatal.
488 * No need to log, and no need to break a soft mount.
489 */
490 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
491 error = 0;
492 if (rep) /* do backoff retransmit on client */
493 rep->r_flags |= R_MUSTRESEND;
494 }
495
496 if (error) {
497 if (rep) {
498 log(LOG_INFO, "nfs send error %d for server %s\n",error,
499 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
500 /*
501 * Deal with errors for the client side.
502 */
503 if (rep->r_flags & R_SOFTTERM)
504 error = EINTR;
505 else
506 rep->r_flags |= R_MUSTRESEND;
507 } else
508 log(LOG_INFO, "nfsd send error %d\n", error);
509
510 /*
511 * Handle any recoverable (soft) socket errors here. (?)
512 */
513 if (error != EINTR && error != ERESTART &&
514 error != EWOULDBLOCK && error != EPIPE)
515 error = 0;
516 }
517 return (error);
518 }
519
520 /*
521 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
522 * done by soreceive(), but for SOCK_STREAM we must deal with the Record
523 * Mark and consolidate the data into a new mbuf list.
524 * nb: Sometimes TCP passes the data up to soreceive() in long lists of
525 * small mbufs.
526 * For SOCK_STREAM we must be very careful to read an entire record once
527 * we have read any of it, even if the system call has been interrupted.
528 */
529 static int
530 nfs_receive(struct nfsreq *rep, struct sockaddr **aname, struct mbuf **mp)
531 {
532 struct socket *so;
533 struct uio auio;
534 struct iovec aio;
535 struct mbuf *m;
536 struct mbuf *control;
537 u_int32_t len;
538 struct sockaddr **getnam;
539 int error, sotype, rcvflg;
540 struct thread *td = curthread; /* XXX */
541
542 /*
543 * Set up arguments for soreceive()
544 */
545 *mp = (struct mbuf *)0;
546 *aname = (struct sockaddr *)0;
547 sotype = rep->r_nmp->nm_sotype;
548
549 /*
550 * For reliable protocols, lock against other senders/receivers
551 * in case a reconnect is necessary.
552 * For SOCK_STREAM, first get the Record Mark to find out how much
553 * more there is to get.
554 * We must lock the socket against other receivers
555 * until we have an entire rpc request/reply.
556 */
557 if (sotype != SOCK_DGRAM) {
558 error = nfs_sndlock(rep);
559 if (error)
560 return (error);
561 tryagain:
562 /*
563 * Check for fatal errors and resending request.
564 */
565 /*
566 * Ugh: If a reconnect attempt just happened, nm_so
567 * would have changed. NULL indicates a failed
568 * attempt that has essentially shut down this
569 * mount point.
570 */
571 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
572 nfs_sndunlock(rep);
573 return (EINTR);
574 }
575 so = rep->r_nmp->nm_so;
576 if (!so) {
577 error = nfs_reconnect(rep);
578 if (error) {
579 nfs_sndunlock(rep);
580 return (error);
581 }
582 goto tryagain;
583 }
584 while (rep->r_flags & R_MUSTRESEND) {
585 m = m_copym(rep->r_mreq, 0, M_COPYALL, MB_WAIT);
586 nfsstats.rpcretries++;
587 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
588 if (error) {
589 if (error == EINTR || error == ERESTART ||
590 (error = nfs_reconnect(rep)) != 0) {
591 nfs_sndunlock(rep);
592 return (error);
593 }
594 goto tryagain;
595 }
596 }
597 nfs_sndunlock(rep);
598 if (sotype == SOCK_STREAM) {
599 aio.iov_base = (caddr_t) &len;
600 aio.iov_len = sizeof(u_int32_t);
601 auio.uio_iov = &aio;
602 auio.uio_iovcnt = 1;
603 auio.uio_segflg = UIO_SYSSPACE;
604 auio.uio_rw = UIO_READ;
605 auio.uio_offset = 0;
606 auio.uio_resid = sizeof(u_int32_t);
607 auio.uio_td = td;
608 do {
609 rcvflg = MSG_WAITALL;
610 error = so_pru_soreceive(so, NULL, &auio, NULL,
611 NULL, &rcvflg);
612 if (error == EWOULDBLOCK && rep) {
613 if (rep->r_flags & R_SOFTTERM)
614 return (EINTR);
615 }
616 } while (error == EWOULDBLOCK);
617 if (!error && auio.uio_resid > 0) {
618 /*
619 * Don't log a 0 byte receive; it means
620 * that the socket has been closed, and
621 * can happen during normal operation
622 * (forcible unmount or Solaris server).
623 */
624 if (auio.uio_resid != sizeof (u_int32_t))
625 log(LOG_INFO,
626 "short receive (%d/%d) from nfs server %s\n",
627 (int)(sizeof(u_int32_t) - auio.uio_resid),
628 (int)sizeof(u_int32_t),
629 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
630 error = EPIPE;
631 }
632 if (error)
633 goto errout;
634 len = ntohl(len) & ~0x80000000;
635 /*
636 * This is SERIOUS! We are out of sync with the sender
637 * and forcing a disconnect/reconnect is all I can do.
638 */
639 if (len > NFS_MAXPACKET) {
640 log(LOG_ERR, "%s (%d) from nfs server %s\n",
641 "impossible packet length",
642 len,
643 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
644 error = EFBIG;
645 goto errout;
646 }
647 auio.uio_resid = len;
648 do {
649 rcvflg = MSG_WAITALL;
650 error = so_pru_soreceive(so, NULL, &auio, mp,
651 NULL, &rcvflg);
652 } while (error == EWOULDBLOCK || error == EINTR ||
653 error == ERESTART);
654 if (!error && auio.uio_resid > 0) {
655 if (len != auio.uio_resid)
656 log(LOG_INFO,
657 "short receive (%d/%d) from nfs server %s\n",
658 len - auio.uio_resid, len,
659 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
660 error = EPIPE;
661 }
662 } else {
663 /*
664 * NB: Since uio_resid is big, MSG_WAITALL is ignored
665 * and soreceive() will return when it has either a
666 * control msg or a data msg.
667 * We have no use for control msg., but must grab them
668 * and then throw them away so we know what is going
669 * on.
670 */
671 auio.uio_resid = len = 100000000; /* Anything Big */
672 auio.uio_td = td;
673 do {
674 rcvflg = 0;
675 error = so_pru_soreceive(so, NULL, &auio, mp,
676 &control, &rcvflg);
677 if (control)
678 m_freem(control);
679 if (error == EWOULDBLOCK && rep) {
680 if (rep->r_flags & R_SOFTTERM)
681 return (EINTR);
682 }
683 } while (error == EWOULDBLOCK ||
684 (!error && *mp == NULL && control));
685 if ((rcvflg & MSG_EOR) == 0)
686 kprintf("Egad!!\n");
687 if (!error && *mp == NULL)
688 error = EPIPE;
689 len -= auio.uio_resid;
690 }
691 errout:
692 if (error && error != EINTR && error != ERESTART) {
693 m_freem(*mp);
694 *mp = (struct mbuf *)0;
695 if (error != EPIPE)
696 log(LOG_INFO,
697 "receive error %d from nfs server %s\n",
698 error,
699 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
700 error = nfs_sndlock(rep);
701 if (!error) {
702 error = nfs_reconnect(rep);
703 if (!error)
704 goto tryagain;
705 else
706 nfs_sndunlock(rep);
707 }
708 }
709 } else {
710 if ((so = rep->r_nmp->nm_so) == NULL)
711 return (EACCES);
712 if (so->so_state & SS_ISCONNECTED)
713 getnam = (struct sockaddr **)0;
714 else
715 getnam = aname;
716 auio.uio_resid = len = 1000000;
717 auio.uio_td = td;
718 do {
719 rcvflg = 0;
720 error = so_pru_soreceive(so, getnam, &auio, mp, NULL,
721 &rcvflg);
722 if (error == EWOULDBLOCK &&
723 (rep->r_flags & R_SOFTTERM))
724 return (EINTR);
725 } while (error == EWOULDBLOCK);
726 len -= auio.uio_resid;
727 }
728 if (error) {
729 m_freem(*mp);
730 *mp = (struct mbuf *)0;
731 }
732 /*
733 * Search for any mbufs that are not a multiple of 4 bytes long
734 * or with m_data not longword aligned.
735 * These could cause pointer alignment problems, so copy them to
736 * well aligned mbufs.
737 */
738 nfs_realign(mp, 5 * NFSX_UNSIGNED);
739 return (error);
740 }
741
742 /*
743 * Implement receipt of reply on a socket.
744 * We must search through the list of received datagrams matching them
745 * with outstanding requests using the xid, until ours is found.
746 */
747 /* ARGSUSED */
748 int
749 nfs_reply(struct nfsreq *myrep)
750 {
751 struct nfsreq *rep;
752 struct nfsmount *nmp = myrep->r_nmp;
753 int32_t t1;
754 struct mbuf *mrep, *md;
755 struct sockaddr *nam;
756 u_int32_t rxid, *tl;
757 caddr_t dpos, cp2;
758 int error;
759
760 /*
761 * Loop around until we get our own reply
762 */
763 for (;;) {
764 /*
765 * Lock against other receivers so that I don't get stuck in
766 * sbwait() after someone else has received my reply for me.
767 * Also necessary for connection based protocols to avoid
768 * race conditions during a reconnect.
769 * If nfs_rcvlock() returns EALREADY, that means that
770 * the reply has already been recieved by another
771 * process and we can return immediately. In this
772 * case, the lock is not taken to avoid races with
773 * other processes.
774 */
775 error = nfs_rcvlock(myrep);
776 if (error == EALREADY)
777 return (0);
778 if (error)
779 return (error);
780 /*
781 * Get the next Rpc reply off the socket
782 */
783 error = nfs_receive(myrep, &nam, &mrep);
784 nfs_rcvunlock(myrep);
785 if (error) {
786 /*
787 * Ignore routing errors on connectionless protocols??
788 */
789 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
790 nmp->nm_so->so_error = 0;
791 if (myrep->r_flags & R_GETONEREP)
792 return (0);
793 continue;
794 }
795 return (error);
796 }
797 if (nam)
798 FREE(nam, M_SONAME);
799
800 /*
801 * Get the xid and check that it is an rpc reply
802 */
803 md = mrep;
804 dpos = mtod(md, caddr_t);
805 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
806 rxid = *tl++;
807 if (*tl != rpc_reply) {
808 nfsstats.rpcinvalid++;
809 m_freem(mrep);
810 nfsmout:
811 if (myrep->r_flags & R_GETONEREP)
812 return (0);
813 continue;
814 }
815
816 /*
817 * Loop through the request list to match up the reply
818 * Iff no match, just drop the datagram. On match, set
819 * r_mrep atomically to prevent the timer from messing
820 * around with the request after we have exited the critical
821 * section.
822 */
823 crit_enter();
824 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
825 if (rep->r_mrep == NULL && rxid == rep->r_xid) {
826 rep->r_mrep = mrep;
827 break;
828 }
829 }
830 crit_exit();
831
832 /*
833 * Fill in the rest of the reply if we found a match.
834 */
835 if (rep) {
836 rep->r_md = md;
837 rep->r_dpos = dpos;
838 if (nfsrtton) {
839 struct rttl *rt;
840
841 rt = &nfsrtt.rttl[nfsrtt.pos];
842 rt->proc = rep->r_procnum;
843 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
844 rt->sent = nmp->nm_sent;
845 rt->cwnd = nmp->nm_cwnd;
846 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
847 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
848 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
849 getmicrotime(&rt->tstamp);
850 if (rep->r_flags & R_TIMING)
851 rt->rtt = rep->r_rtt;
852 else
853 rt->rtt = 1000000;
854 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
855 }
856 /*
857 * Update congestion window.
858 * Do the additive increase of
859 * one rpc/rtt.
860 */
861 if (nmp->nm_cwnd <= nmp->nm_sent) {
862 nmp->nm_cwnd +=
863 (NFS_CWNDSCALE * NFS_CWNDSCALE +
864 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
865 if (nmp->nm_cwnd > NFS_MAXCWND)
866 nmp->nm_cwnd = NFS_MAXCWND;
867 }
868 crit_enter(); /* nfs_timer interlock for nm_sent */
869 if (rep->r_flags & R_SENT) {
870 rep->r_flags &= ~R_SENT;
871 nmp->nm_sent -= NFS_CWNDSCALE;
872 }
873 crit_exit();
874 /*
875 * Update rtt using a gain of 0.125 on the mean
876 * and a gain of 0.25 on the deviation.
877 */
878 if (rep->r_flags & R_TIMING) {
879 /*
880 * Since the timer resolution of
881 * NFS_HZ is so course, it can often
882 * result in r_rtt == 0. Since
883 * r_rtt == N means that the actual
884 * rtt is between N+dt and N+2-dt ticks,
885 * add 1.
886 */
887 t1 = rep->r_rtt + 1;
888 t1 -= (NFS_SRTT(rep) >> 3);
889 NFS_SRTT(rep) += t1;
890 if (t1 < 0)
891 t1 = -t1;
892 t1 -= (NFS_SDRTT(rep) >> 2);
893 NFS_SDRTT(rep) += t1;
894 }
895 nmp->nm_timeouts = 0;
896 }
897 /*
898 * If not matched to a request, drop it.
899 * If it's mine, get out.
900 */
901 if (rep == NULL) {
902 nfsstats.rpcunexpected++;
903 m_freem(mrep);
904 } else if (rep == myrep) {
905 if (rep->r_mrep == NULL)
906 panic("nfsreply nil");
907 return (0);
908 }
909 if (myrep->r_flags & R_GETONEREP)
910 return (0);
911 }
912 }
913
914 /*
915 * nfs_request - goes something like this
916 * - fill in request struct
917 * - links it into list
918 * - calls nfs_send() for first transmit
919 * - calls nfs_receive() to get reply
920 * - break down rpc header and return with nfs reply pointed to
921 * by mrep or error
922 * nb: always frees up mreq mbuf list
923 */
924 int
925 nfs_request(struct vnode *vp, struct mbuf *mrest, int procnum,
926 struct thread *td, struct ucred *cred, struct mbuf **mrp,
927 struct mbuf **mdp, caddr_t *dposp)
928 {
929 struct mbuf *mrep, *m2;
930 struct nfsreq *rep;
931 u_int32_t *tl;
932 int i;
933 struct nfsmount *nmp;
934 struct mbuf *m, *md, *mheadend;
935 char nickv[RPCX_NICKVERF];
936 time_t waituntil;
937 caddr_t dpos, cp2;
938 int t1, error = 0, mrest_len, auth_len, auth_type;
939 int trylater_delay = 15, trylater_cnt = 0, failed_auth = 0;
940 int verf_len, verf_type;
941 u_int32_t xid;
942 char *auth_str, *verf_str;
943 NFSKERBKEY_T key; /* save session key */
944
945 /* Reject requests while attempting a forced unmount. */
946 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) {
947 m_freem(mrest);
948 return (ESTALE);
949 }
950 nmp = VFSTONFS(vp->v_mount);
951 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
952 rep->r_nmp = nmp;
953 rep->r_vp = vp;
954 rep->r_td = td;
955 rep->r_procnum = procnum;
956 rep->r_mreq = NULL;
957 i = 0;
958 m = mrest;
959 while (m) {
960 i += m->m_len;
961 m = m->m_next;
962 }
963 mrest_len = i;
964
965 /*
966 * Get the RPC header with authorization.
967 */
968 kerbauth:
969 verf_str = auth_str = (char *)0;
970 if (nmp->nm_flag & NFSMNT_KERB) {
971 verf_str = nickv;
972 verf_len = sizeof (nickv);
973 auth_type = RPCAUTH_KERB4;
974 bzero((caddr_t)key, sizeof (key));
975 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
976 &auth_len, verf_str, verf_len)) {
977 error = nfs_getauth(nmp, rep, cred, &auth_str,
978 &auth_len, verf_str, &verf_len, key);
979 if (error) {
980 kfree((caddr_t)rep, M_NFSREQ);
981 m_freem(mrest);
982 return (error);
983 }
984 }
985 } else {
986 auth_type = RPCAUTH_UNIX;
987 if (cred->cr_ngroups < 1)
988 panic("nfsreq nogrps");
989 auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ?
990 nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) +
991 5 * NFSX_UNSIGNED;
992 }
993 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
994 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
995 if (auth_str)
996 kfree(auth_str, M_TEMP);
997
998 /*
999 * For stream protocols, insert a Sun RPC Record Mark.
1000 */
1001 if (nmp->nm_sotype == SOCK_STREAM) {
1002 M_PREPEND(m, NFSX_UNSIGNED, MB_WAIT);
1003 if (m == NULL) {
1004 kfree(rep, M_NFSREQ);
1005 return (ENOBUFS);
1006 }
1007 *mtod(m, u_int32_t *) = htonl(0x80000000 |
1008 (m->m_pkthdr.len - NFSX_UNSIGNED));
1009 }
1010 rep->r_mreq = m;
1011 rep->r_xid = xid;
1012 tryagain:
1013 if (nmp->nm_flag & NFSMNT_SOFT)
1014 rep->r_retry = nmp->nm_retry;
1015 else
1016 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */
1017 rep->r_rtt = rep->r_rexmit = 0;
1018 if (proct[procnum] > 0)
1019 rep->r_flags = R_TIMING | R_MASKTIMER;
1020 else
1021 rep->r_flags = R_MASKTIMER;
1022 rep->r_mrep = NULL;
1023
1024 /*
1025 * Do the client side RPC.
1026 */
1027 nfsstats.rpcrequests++;
1028
1029 /*
1030 * Chain request into list of outstanding requests. Be sure
1031 * to put it LAST so timer finds oldest requests first. Note
1032 * that R_MASKTIMER is set at the moment to prevent any timer
1033 * action on this request while we are still doing processing on
1034 * it below. splsoftclock() primarily protects nm_sent. Note
1035 * that we may block in this code so there is no atomicy guarentee.
1036 */
1037 crit_enter();
1038 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1039
1040 /*
1041 * If backing off another request or avoiding congestion, don't
1042 * send this one now but let timer do it. If not timing a request,
1043 * do it now.
1044 *
1045 * Even though the timer will not mess with our request there is
1046 * still the possibility that we will race a reply (which clears
1047 * R_SENT), especially on localhost connections, so be very careful
1048 * when setting R_SENT. We could set R_SENT prior to calling
1049 * nfs_send() but why bother if the response occurs that quickly?
1050 */
1051 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1052 (nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1053 nmp->nm_sent < nmp->nm_cwnd)) {
1054 if (nmp->nm_soflags & PR_CONNREQUIRED)
1055 error = nfs_sndlock(rep);
1056 if (!error) {
1057 m2 = m_copym(m, 0, M_COPYALL, MB_WAIT);
1058 error = nfs_send(nmp->nm_so, nmp->nm_nam, m2, rep);
1059 if (nmp->nm_soflags & PR_CONNREQUIRED)
1060 nfs_sndunlock(rep);
1061 }
1062 if (!error && (rep->r_flags & R_MUSTRESEND) == 0 &&
1063 rep->r_mrep == NULL) {
1064 KASSERT((rep->r_flags & R_SENT) == 0,
1065 ("R_SENT ASSERT %p", rep));
1066 nmp->nm_sent += NFS_CWNDSCALE;
1067 rep->r_flags |= R_SENT;
1068 }
1069 } else {
1070 rep->r_rtt = -1;
1071 }
1072
1073 /*
1074 * Let the timer do what it will with the request, then
1075 * wait for the reply from our send or the timer's.
1076 */
1077 if (!error || error == EPIPE) {
1078 rep->r_flags &= ~R_MASKTIMER;
1079 crit_exit();
1080 error = nfs_reply(rep);
1081 crit_enter();
1082 }
1083
1084 /*
1085 * RPC done, unlink the request, but don't rip it out from under
1086 * the callout timer.
1087 */
1088 while (rep->r_flags & R_LOCKED) {
1089 nfs_timer_raced = 1;
1090 tsleep(&nfs_timer_raced, 0, "nfstrac", 0);
1091 }
1092 TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1093
1094 /*
1095 * Decrement the outstanding request count.
1096 */
1097 if (rep->r_flags & R_SENT) {
1098 rep->r_flags &= ~R_SENT;
1099 nmp->nm_sent -= NFS_CWNDSCALE;
1100 }
1101 crit_exit();
1102
1103 /*
1104 * If there was a successful reply and a tprintf msg.
1105 * tprintf a response.
1106 */
1107 if (!error && (rep->r_flags & R_TPRINTFMSG))
1108 nfs_msg(rep->r_td, nmp->nm_mountp->mnt_stat.f_mntfromname,
1109 "is alive again");
1110 mrep = rep->r_mrep;
1111 md = rep->r_md;
1112 dpos = rep->r_dpos;
1113 if (error) {
1114 m_freem(rep->r_mreq);
1115 kfree((caddr_t)rep, M_NFSREQ);
1116 return (error);
1117 }
1118
1119 /*
1120 * break down the rpc header and check if ok
1121 */
1122 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1123 if (*tl++ == rpc_msgdenied) {
1124 if (*tl == rpc_mismatch)
1125 error = EOPNOTSUPP;
1126 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1127 if (!failed_auth) {
1128 failed_auth++;
1129 mheadend->m_next = (struct mbuf *)0;
1130 m_freem(mrep);
1131 m_freem(rep->r_mreq);
1132 goto kerbauth;
1133 } else
1134 error = EAUTH;
1135 } else
1136 error = EACCES;
1137 m_freem(mrep);
1138 m_freem(rep->r_mreq);
1139 kfree((caddr_t)rep, M_NFSREQ);
1140 return (error);
1141 }
1142
1143 /*
1144 * Grab any Kerberos verifier, otherwise just throw it away.
1145 */
1146 verf_type = fxdr_unsigned(int, *tl++);
1147 i = fxdr_unsigned(int32_t, *tl);
1148 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1149 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1150 if (error)
1151 goto nfsmout;
1152 } else if (i > 0)
1153 nfsm_adv(nfsm_rndup(i));
1154 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1155 /* 0 == ok */
1156 if (*tl == 0) {
1157 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1158 if (*tl != 0) {
1159 error = fxdr_unsigned(int, *tl);
1160 if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1161 error == NFSERR_TRYLATER) {
1162 m_freem(mrep);
1163 error = 0;
1164 waituntil = time_second + trylater_delay;
1165 while (time_second < waituntil)
1166 (void) tsleep((caddr_t)&lbolt,
1167 0, "nqnfstry", 0);
1168 trylater_delay *= nfs_backoff[trylater_cnt];
1169 if (trylater_cnt < 7)
1170 trylater_cnt++;
1171 goto tryagain;
1172 }
1173
1174 /*
1175 * If the File Handle was stale, invalidate the
1176 * lookup cache, just in case.
1177 *
1178 * To avoid namecache<->vnode deadlocks we must
1179 * release the vnode lock if we hold it.
1180 */
1181 if (error == ESTALE) {
1182 int ltype;
1183
1184 ltype = lockstatus(&vp->v_lock, curthread);
1185 if (ltype == LK_EXCLUSIVE || ltype == LK_SHARED)
1186 lockmgr(&vp->v_lock, LK_RELEASE);
1187 cache_inval_vp(vp, CINV_CHILDREN);
1188 if (ltype == LK_EXCLUSIVE || ltype == LK_SHARED)
1189 lockmgr(&vp->v_lock, ltype);
1190 }
1191 if (nmp->nm_flag & NFSMNT_NFSV3) {
1192 *mrp = mrep;
1193 *mdp = md;
1194 *dposp = dpos;
1195 error |= NFSERR_RETERR;
1196 } else
1197 m_freem(mrep);
1198 m_freem(rep->r_mreq);
1199 kfree((caddr_t)rep, M_NFSREQ);
1200 return (error);
1201 }
1202
1203 *mrp = mrep;
1204 *mdp = md;
1205 *dposp = dpos;
1206 m_freem(rep->r_mreq);
1207 FREE((caddr_t)rep, M_NFSREQ);
1208 return (0);
1209 }
1210 m_freem(mrep);
1211 error = EPROTONOSUPPORT;
1212 nfsmout:
1213 m_freem(rep->r_mreq);
1214 kfree((caddr_t)rep, M_NFSREQ);
1215 return (error);
1216 }
1217
1218 #ifndef NFS_NOSERVER
1219 /*
1220 * Generate the rpc reply header
1221 * siz arg. is used to decide if adding a cluster is worthwhile
1222 */
1223 int
1224 nfs_rephead(int siz, struct nfsrv_descript *nd, struct nfssvc_sock *slp,
1225 int err, struct mbuf **mrq, struct mbuf **mbp, caddr_t *bposp)
1226 {
1227 u_int32_t *tl;
1228 struct mbuf *mreq;
1229 caddr_t bpos;
1230 struct mbuf *mb, *mb2;
1231
1232 siz += RPC_REPLYSIZ;
1233 mb = mreq = m_getl(max_hdr + siz, MB_WAIT, MT_DATA, M_PKTHDR, NULL);
1234 mreq->m_pkthdr.len = 0;
1235 /*
1236 * If this is not a cluster, try and leave leading space
1237 * for the lower level headers.
1238 */
1239 if ((max_hdr + siz) < MINCLSIZE)
1240 mreq->m_data += max_hdr;
1241 tl = mtod(mreq, u_int32_t *);
1242 mreq->m_len = 6 * NFSX_UNSIGNED;
1243 bpos = ((caddr_t)tl) + mreq->m_len;
1244 *tl++ = txdr_unsigned(nd->nd_retxid);
1245 *tl++ = rpc_reply;
1246 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1247 *tl++ = rpc_msgdenied;
1248 if (err & NFSERR_AUTHERR) {
1249 *tl++ = rpc_autherr;
1250 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1251 mreq->m_len -= NFSX_UNSIGNED;
1252 bpos -= NFSX_UNSIGNED;
1253 } else {
1254 *tl++ = rpc_mismatch;
1255 *tl++ = txdr_unsigned(RPC_VER2);
1256 *tl = txdr_unsigned(RPC_VER2);
1257 }
1258 } else {
1259 *tl++ = rpc_msgaccepted;
1260
1261 /*
1262 * For Kerberos authentication, we must send the nickname
1263 * verifier back, otherwise just RPCAUTH_NULL.
1264 */
1265 if (nd->nd_flag & ND_KERBFULL) {
1266 struct nfsuid *nuidp;
1267 struct timeval ktvin, ktvout;
1268
1269 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1270 nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1271 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1272 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1273 &nuidp->nu_haddr, nd->nd_nam2)))
1274 break;
1275 }
1276 if (nuidp) {
1277 ktvin.tv_sec =
1278 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1279 ktvin.tv_usec =
1280 txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1281
1282 /*
1283 * Encrypt the timestamp in ecb mode using the
1284 * session key.
1285 */
1286 #ifdef NFSKERB
1287 XXX
1288 #endif
1289
1290 *tl++ = rpc_auth_kerb;
1291 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1292 *tl = ktvout.tv_sec;
1293 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1294 *tl++ = ktvout.tv_usec;
1295 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1296 } else {
1297 *tl++ = 0;
1298 *tl++ = 0;
1299 }
1300 } else {
1301 *tl++ = 0;
1302 *tl++ = 0;
1303 }
1304 switch (err) {
1305 case EPROGUNAVAIL:
1306 *tl = txdr_unsigned(RPC_PROGUNAVAIL);
1307 break;
1308 case EPROGMISMATCH:
1309 *tl = txdr_unsigned(RPC_PROGMISMATCH);
1310 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1311 *tl++ = txdr_unsigned(2);
1312 *tl = txdr_unsigned(3);
1313 break;
1314 case EPROCUNAVAIL:
1315 *tl = txdr_unsigned(RPC_PROCUNAVAIL);
1316 break;
1317 case EBADRPC:
1318 *tl = txdr_unsigned(RPC_GARBAGE);
1319 break;
1320 default:
1321 *tl = 0;
1322 if (err != NFSERR_RETVOID) {
1323 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1324 if (err)
1325 *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1326 else
1327 *tl = 0;
1328 }
1329 break;
1330 };
1331 }
1332
1333 if (mrq != NULL)
1334 *mrq = mreq;
1335 *mbp = mb;
1336 *bposp = bpos;
1337 if (err != 0 && err != NFSERR_RETVOID)
1338 nfsstats.srvrpc_errs++;
1339 return (0);
1340 }
1341
1342
1343 #endif /* NFS_NOSERVER */
1344 /*
1345 * Nfs timer routine
1346 * Scan the nfsreq list and retranmit any requests that have timed out
1347 * To avoid retransmission attempts on STREAM sockets (in the future) make
1348 * sure to set the r_retry field to 0 (implies nm_retry == 0).
1349 */
1350 void
1351 nfs_timer(void *arg /* never used */)
1352 {
1353 struct nfsreq *rep;
1354 struct mbuf *m;
1355 struct socket *so;
1356 struct nfsmount *nmp;
1357 int timeo;
1358 int error;
1359 #ifndef NFS_NOSERVER
1360 struct nfssvc_sock *slp;
1361 u_quad_t cur_usec;
1362 #endif /* NFS_NOSERVER */
1363 struct thread *td = &thread0; /* XXX for credentials, will break if sleep */
1364
1365 crit_enter();
1366 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
1367 nmp = rep->r_nmp;
1368 if (rep->r_mrep || (rep->r_flags & (R_SOFTTERM|R_MASKTIMER)))
1369 continue;
1370 rep->r_flags |= R_LOCKED;
1371 if (nfs_sigintr(nmp, rep, rep->r_td)) {
1372 nfs_softterm(rep);
1373 goto skip;
1374 }
1375 if (rep->r_rtt >= 0) {
1376 rep->r_rtt++;
1377 if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1378 timeo = nmp->nm_timeo;
1379 else
1380 timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1381 if (nmp->nm_timeouts > 0)
1382 timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1383 if (rep->r_rtt <= timeo)
1384 goto skip;
1385 if (nmp->nm_timeouts < 8)
1386 nmp->nm_timeouts++;
1387 }
1388 /*
1389 * Check for server not responding
1390 */
1391 if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1392 rep->r_rexmit > nmp->nm_deadthresh) {
1393 nfs_msg(rep->r_td,
1394 nmp->nm_mountp->mnt_stat.f_mntfromname,
1395 "not responding");
1396 rep->r_flags |= R_TPRINTFMSG;
1397 }
1398 if (rep->r_rexmit >= rep->r_retry) { /* too many */
1399 nfsstats.rpctimeouts++;
1400 nfs_softterm(rep);
1401 goto skip;
1402 }
1403 if (nmp->nm_sotype != SOCK_DGRAM) {
1404 if (++rep->r_rexmit > NFS_MAXREXMIT)
1405 rep->r_rexmit = NFS_MAXREXMIT;
1406 goto skip;
1407 }
1408 if ((so = nmp->nm_so) == NULL)
1409 goto skip;
1410
1411 /*
1412 * If there is enough space and the window allows..
1413 * Resend it
1414 * Set r_rtt to -1 in case we fail to send it now.
1415 */
1416 rep->r_rtt = -1;
1417 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1418 ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1419 (rep->r_flags & R_SENT) ||
1420 nmp->nm_sent < nmp->nm_cwnd) &&
1421 (m = m_copym(rep->r_mreq, 0, M_COPYALL, MB_DONTWAIT))){
1422 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
1423 error = so_pru_send(so, 0, m, (struct sockaddr *)0,
1424 (struct mbuf *)0, td);
1425 else
1426 error = so_pru_send(so, 0, m, nmp->nm_nam,
1427 (struct mbuf *)0, td);
1428 if (error) {
1429 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
1430 so->so_error = 0;
1431 } else if (rep->r_mrep == NULL) {
1432 /*
1433 * Iff first send, start timing
1434 * else turn timing off, backoff timer
1435 * and divide congestion window by 2.
1436 *
1437 * It is possible for the so_pru_send() to
1438 * block and for us to race a reply so we
1439 * only do this if the reply field has not
1440 * been filled in. R_LOCKED will prevent
1441 * the request from being ripped out from under
1442 * us entirely.
1443 */
1444 if (rep->r_flags & R_SENT) {
1445 rep->r_flags &= ~R_TIMING;
1446 if (++rep->r_rexmit > NFS_MAXREXMIT)
1447 rep->r_rexmit = NFS_MAXREXMIT;
1448 nmp->nm_cwnd >>= 1;
1449 if (nmp->nm_cwnd < NFS_CWNDSCALE)
1450 nmp->nm_cwnd = NFS_CWNDSCALE;
1451 nfsstats.rpcretries++;
1452 } else {
1453 rep->r_flags |= R_SENT;
1454 nmp->nm_sent += NFS_CWNDSCALE;
1455 }
1456 rep->r_rtt = 0;
1457 }
1458 }
1459 skip:
1460 rep->r_flags &= ~R_LOCKED;
1461 }
1462 #ifndef NFS_NOSERVER
1463
1464 /*
1465 * Scan the write gathering queues for writes that need to be
1466 * completed now.
1467 */
1468 cur_usec = nfs_curusec();
1469 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
1470 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1471 nfsrv_wakenfsd(slp, 1);
1472 }
1473 #endif /* NFS_NOSERVER */
1474
1475 /*
1476 * Due to possible blocking, a client operation may be waiting for
1477 * us to finish processing this request so it can remove it.
1478 */
1479 if (nfs_timer_raced) {
1480 nfs_timer_raced = 0;
1481 wakeup(&nfs_timer_raced);
1482 }
1483 crit_exit();
1484 callout_reset(&nfs_timer_handle, nfs_ticks, nfs_timer, NULL);
1485 }
1486
1487 /*
1488 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1489 * wait for all requests to complete. This is used by forced unmounts
1490 * to terminate any outstanding RPCs.
1491 */
1492 int
1493 nfs_nmcancelreqs(struct nfsmount *nmp)
1494 {
1495 struct nfsreq *req;
1496 int i;
1497
1498 crit_enter();
1499 TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
1500 if (nmp != req->r_nmp || req->r_mrep != NULL ||
1501 (req->r_flags & R_SOFTTERM)) {
1502 continue;
1503 }
1504 nfs_softterm(req);
1505 }
1506 crit_exit();
1507
1508 for (i = 0; i < 30; i++) {
1509 crit_enter();
1510 TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
1511 if (nmp == req->r_nmp)
1512 break;
1513 }
1514 crit_exit();
1515 if (req == NULL)
1516 return (0);
1517 tsleep(&lbolt, 0, "nfscancel", 0);
1518 }
1519 return (EBUSY);
1520 }
1521
1522 /*
1523 * Flag a request as being about to terminate (due to NFSMNT_INT/NFSMNT_SOFT).
1524 * The nm_send count is decremented now to avoid deadlocks when the process in
1525 * soreceive() hasn't yet managed to send its own request.
1526 *
1527 * This routine must be called at splsoftclock() to protect r_flags and
1528 * nm_sent.
1529 */
1530
1531 static void
1532 nfs_softterm(struct nfsreq *rep)
1533 {
1534 rep->r_flags |= R_SOFTTERM;
1535
1536 if (rep->r_flags & R_SENT) {
1537 rep->r_nmp->nm_sent -= NFS_CWNDSCALE;
1538 rep->r_flags &= ~R_SENT;
1539 }
1540 }
1541
1542 /*
1543 * Test for a termination condition pending on the process.
1544 * This is used for NFSMNT_INT mounts.
1545 */
1546 int
1547 nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct thread *td)
1548 {
1549 sigset_t tmpset;
1550 struct proc *p;
1551 struct lwp *lp;
1552
1553 if (rep && (rep->r_flags & R_SOFTTERM))
1554 return (EINTR);
1555 /* Terminate all requests while attempting a forced unmount. */
1556 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1557 return (EINTR);
1558 if (!(nmp->nm_flag & NFSMNT_INT))
1559 return (0);
1560 /* td might be NULL YYY */
1561 if (td == NULL || (p = td->td_proc) == NULL)
1562 return (0);
1563
1564 lp = td->td_lwp;
1565 tmpset = lwp_sigpend(lp);
1566 SIGSETNAND(tmpset, lp->lwp_sigmask);
1567 SIGSETNAND(tmpset, p->p_sigignore);
1568 if (SIGNOTEMPTY(tmpset) && NFSINT_SIGMASK(tmpset))
1569 return (EINTR);
1570
1571 return (0);
1572 }
1573
1574 /*
1575 * Lock a socket against others.
1576 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1577 * and also to avoid race conditions between the processes with nfs requests
1578 * in progress when a reconnect is necessary.
1579 */
1580 int
1581 nfs_sndlock(struct nfsreq *rep)
1582 {
1583 int *statep = &rep->r_nmp->nm_state;
1584 struct thread *td;
1585 int slptimeo;
1586 int slpflag;
1587 int error;
1588
1589 slpflag = 0;
1590 slptimeo = 0;
1591 td = rep->r_td;
1592 if (rep->r_nmp->nm_flag & NFSMNT_INT)
1593 slpflag = PCATCH;
1594
1595 error = 0;
1596 crit_enter();
1597 while (*statep & NFSSTA_SNDLOCK) {
1598 *statep |= NFSSTA_WANTSND;
1599 if (nfs_sigintr(rep->r_nmp, rep, td)) {
1600 error = EINTR;
1601 break;
1602 }
1603 tsleep((caddr_t)statep, slpflag, "nfsndlck", slptimeo);
1604 if (slpflag == PCATCH) {
1605 slpflag = 0;
1606 slptimeo = 2 * hz;
1607 }
1608 }
1609 /* Always fail if our request has been cancelled. */
1610 if ((rep->r_flags & R_SOFTTERM))
1611 error = EINTR;
1612 if (error == 0)
1613 *statep |= NFSSTA_SNDLOCK;
1614 crit_exit();
1615 return (error);
1616 }
1617
1618 /*
1619 * Unlock the stream socket for others.
1620 */
1621 void
1622 nfs_sndunlock(struct nfsreq *rep)
1623 {
1624 int *statep = &rep->r_nmp->nm_state;
1625
1626 if ((*statep & NFSSTA_SNDLOCK) == 0)
1627 panic("nfs sndunlock");
1628 crit_enter();
1629 *statep &= ~NFSSTA_SNDLOCK;
1630 if (*statep & NFSSTA_WANTSND) {
1631 *statep &= ~NFSSTA_WANTSND;
1632 wakeup((caddr_t)statep);
1633 }
1634 crit_exit();
1635 }
1636
1637 static int
1638 nfs_rcvlock(struct nfsreq *rep)
1639 {
1640 int *statep = &rep->r_nmp->nm_state;
1641 int slpflag;
1642 int slptimeo;
1643 int error;
1644
1645 /*
1646 * Unconditionally check for completion in case another nfsiod
1647 * get the packet while the caller was blocked, before the caller
1648 * called us. Packet reception is handled by mainline code which
1649 * is protected by the BGL at the moment.
1650 *
1651 * We do not strictly need the second check just before the
1652 * tsleep(), but it's good defensive programming.
1653 */
1654 if (rep->r_mrep != NULL)
1655 return (EALREADY);
1656
1657 if (rep->r_nmp->nm_flag & NFSMNT_INT)
1658 slpflag = PCATCH;
1659 else
1660 slpflag = 0;
1661 slptimeo = 0;
1662 error = 0;
1663 crit_enter();
1664 while (*statep & NFSSTA_RCVLOCK) {
1665 if (nfs_sigintr(rep->r_nmp, rep, rep->r_td)) {
1666 error = EINTR;
1667 break;
1668 }
1669 if (rep->r_mrep != NULL) {
1670 error = EALREADY;
1671 break;
1672 }
1673 *statep |= NFSSTA_WANTRCV;
1674 tsleep((caddr_t)statep, slpflag, "nfsrcvlk", slptimeo);
1675 /*
1676 * If our reply was recieved while we were sleeping,
1677 * then just return without taking the lock to avoid a
1678 * situation where a single iod could 'capture' the
1679 * recieve lock.
1680 */
1681 if (rep->r_mrep != NULL) {
1682 error = EALREADY;
1683 break;
1684 }
1685 if (slpflag == PCATCH) {
1686 slpflag = 0;
1687 slptimeo = 2 * hz;
1688 }
1689 }
1690 if (error == 0) {
1691 *statep |= NFSSTA_RCVLOCK;
1692 rep->r_nmp->nm_rcvlock_td = curthread; /* DEBUGGING */
1693 }
1694 crit_exit();
1695 return (error);
1696 }
1697
1698 /*
1699 * Unlock the stream socket for others.
1700 */
1701 static void
1702 nfs_rcvunlock(struct nfsreq *rep)
1703 {
1704 int *statep = &rep->r_nmp->nm_state;
1705
1706 if ((*statep & NFSSTA_RCVLOCK) == 0)
1707 panic("nfs rcvunlock");
1708 crit_enter();
1709 rep->r_nmp->nm_rcvlock_td = (void *)-1; /* DEBUGGING */
1710 *statep &= ~NFSSTA_RCVLOCK;
1711 if (*statep & NFSSTA_WANTRCV) {
1712 *statep &= ~NFSSTA_WANTRCV;
1713 wakeup((caddr_t)statep);
1714 }
1715 crit_exit();
1716 }
1717
1718 /*
1719 * nfs_realign:
1720 *
1721 * Check for badly aligned mbuf data and realign by copying the unaligned
1722 * portion of the data into a new mbuf chain and freeing the portions
1723 * of the old chain that were replaced.
1724 *
1725 * We cannot simply realign the data within the existing mbuf chain
1726 * because the underlying buffers may contain other rpc commands and
1727 * we cannot afford to overwrite them.
1728 *
1729 * We would prefer to avoid this situation entirely. The situation does
1730 * not occur with NFS/UDP and is supposed to only occassionally occur
1731 * with TCP. Use vfs.nfs.realign_count and realign_test to check this.
1732 */
1733 static void
1734 nfs_realign(struct mbuf **pm, int hsiz)
1735 {
1736 struct mbuf *m;
1737 struct mbuf *n = NULL;
1738 int off = 0;
1739
1740 ++nfs_realign_test;
1741
1742 while ((m = *pm) != NULL) {
1743 if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) {
1744 n = m_getl(m->m_len, MB_WAIT, MT_DATA, 0, NULL);
1745 n->m_len = 0;
1746 break;
1747 }
1748 pm = &m->m_next;
1749 }
1750
1751 /*
1752 * If n is non-NULL, loop on m copying data, then replace the
1753 * portion of the chain that had to be realigned.
1754 */
1755 if (n != NULL) {
1756 ++nfs_realign_count;
1757 while (m) {
1758 m_copyback(n, off, m->m_len, mtod(m, caddr_t));
1759 off += m->m_len;
1760 m = m->m_next;
1761 }
1762 m_freem(*pm);
1763 *pm = n;
1764 }
1765 }
1766
1767 #ifndef NFS_NOSERVER
1768
1769 /*
1770 * Parse an RPC request
1771 * - verify it
1772 * - fill in the cred struct.
1773 */
1774 int
1775 nfs_getreq(struct nfsrv_descript *nd, struct nfsd *nfsd, int has_header)
1776 {
1777 int len, i;
1778 u_int32_t *tl;
1779 int32_t t1;
1780 struct uio uio;
1781 struct iovec iov;
1782 caddr_t dpos, cp2, cp;
1783 u_int32_t nfsvers, auth_type;
1784 uid_t nickuid;
1785 int error = 0, ticklen;
1786 struct mbuf *mrep, *md;
1787 struct nfsuid *nuidp;
1788 struct timeval tvin, tvout;
1789 #if 0 /* until encrypted keys are implemented */
1790 NFSKERBKEYSCHED_T keys; /* stores key schedule */
1791 #endif
1792
1793 mrep = nd->nd_mrep;
1794 md = nd->nd_md;
1795 dpos = nd->nd_dpos;
1796 if (has_header) {
1797 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1798 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1799 if (*tl++ != rpc_call) {
1800 m_freem(mrep);
1801 return (EBADRPC);
1802 }
1803 } else
1804 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1805 nd->nd_repstat = 0;
1806 nd->nd_flag = 0;
1807 if (*tl++ != rpc_vers) {
1808 nd->nd_repstat = ERPCMISMATCH;
1809 nd->nd_procnum = NFSPROC_NOOP;
1810 return (0);
1811 }
1812 if (*tl != nfs_prog) {
1813 nd->nd_repstat = EPROGUNAVAIL;
1814 nd->nd_procnum = NFSPROC_NOOP;
1815 return (0);
1816 }
1817 tl++;
1818 nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1819 if (nfsvers < NFS_VER2 || nfsvers > NFS_VER3) {
1820 nd->nd_repstat = EPROGMISMATCH;
1821 nd->nd_procnum = NFSPROC_NOOP;
1822 return (0);
1823 }
1824 if (nfsvers == NFS_VER3)
1825 nd->nd_flag = ND_NFSV3;
1826 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1827 if (nd->nd_procnum == NFSPROC_NULL)
1828 return (0);
1829 if (nd->nd_procnum >= NFS_NPROCS ||
1830 (nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1831 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1832 nd->nd_repstat = EPROCUNAVAIL;
1833 nd->nd_procnum = NFSPROC_NOOP;
1834 return (0);
1835 }
1836 if ((nd->nd_flag & ND_NFSV3) == 0)
1837 nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1838 auth_type = *tl++;
1839 len = fxdr_unsigned(int, *tl++);
1840 if (len < 0 || len > RPCAUTH_MAXSIZ) {
1841 m_freem(mrep);
1842 return (EBADRPC);
1843 }
1844
1845 nd->nd_flag &= ~ND_KERBAUTH;
1846 /*
1847 * Handle auth_unix or auth_kerb.
1848 */
1849 if (auth_type == rpc_auth_unix) {
1850 len = fxdr_unsigned(int, *++tl);
1851 if (len < 0 || len > NFS_MAXNAMLEN) {
1852 m_freem(mrep);
1853 return (EBADRPC);
1854 }
1855 nfsm_adv(nfsm_rndup(len));
1856 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1857 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred));
1858 nd->nd_cr.cr_ref = 1;
1859 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1860 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1861 len = fxdr_unsigned(int, *tl);
1862 if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1863 m_freem(mrep);
1864 return (EBADRPC);
1865 }
1866 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1867 for (i = 1; i <= len; i++)
1868 if (i < NGROUPS)
1869 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1870 else
1871 tl++;
1872 nd->nd_cr.cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1);
1873 if (nd->nd_cr.cr_ngroups > 1)
1874 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1875 len = fxdr_unsigned(int, *++tl);
1876 if (len < 0 || len > RPCAUTH_MAXSIZ) {
1877 m_freem(mrep);
1878 return (EBADRPC);
1879 }
1880 if (len > 0)
1881 nfsm_adv(nfsm_rndup(len));
1882 } else if (auth_type == rpc_auth_kerb) {
1883 switch (fxdr_unsigned(int, *tl++)) {
1884 case RPCAKN_FULLNAME:
1885 ticklen = fxdr_unsigned(int, *tl);
1886 *((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1887 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1888 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1889 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1890 m_freem(mrep);
1891 return (EBADRPC);
1892 }
1893 uio.uio_offset = 0;
1894 uio.uio_iov = &iov;
1895 uio.uio_iovcnt = 1;
1896 uio.uio_segflg = UIO_SYSSPACE;
1897 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1898 iov.iov_len = RPCAUTH_MAXSIZ - 4;
1899 nfsm_mtouio(&uio, uio.uio_resid);
1900 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1901 if (*tl++ != rpc_auth_kerb ||
1902 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1903 kprintf("Bad kerb verifier\n");
1904 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1905 nd->nd_procnum = NFSPROC_NOOP;
1906 return (0);
1907 }
1908 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1909 tl = (u_int32_t *)cp;
1910 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1911 kprintf("Not fullname kerb verifier\n");
1912 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1913 nd->nd_procnum = NFSPROC_NOOP;
1914 return (0);
1915 }
1916 cp += NFSX_UNSIGNED;
1917 bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED);
1918 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1919 nd->nd_flag |= ND_KERBFULL;
1920 nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1921 break;
1922 case RPCAKN_NICKNAME:
1923 if (len != 2 * NFSX_UNSIGNED) {
1924 kprintf("Kerb nickname short\n");
1925 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1926 nd->nd_procnum = NFSPROC_NOOP;
1927 return (0);
1928 }
1929 nickuid = fxdr_unsigned(uid_t, *tl);
1930 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1931 if (*tl++ != rpc_auth_kerb ||
1932 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1933 kprintf("Kerb nick verifier bad\n");
1934 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1935 nd->nd_procnum = NFSPROC_NOOP;
1936 return (0);
1937 }
1938 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1939 tvin.tv_sec = *tl++;
1940 tvin.tv_usec = *tl;
1941
1942 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1943 nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1944 if (nuidp->nu_cr.cr_uid == nickuid &&
1945 (!nd->nd_nam2 ||
1946 netaddr_match(NU_NETFAM(nuidp),
1947 &nuidp->nu_haddr, nd->nd_nam2)))
1948 break;
1949 }
1950 if (!nuidp) {
1951 nd->nd_repstat =
1952 (NFSERR_AUTHERR|AUTH_REJECTCRED);
1953 nd->nd_procnum = NFSPROC_NOOP;
1954 return (0);
1955 }
1956
1957 /*
1958 * Now, decrypt the timestamp using the session key
1959 * and validate it.
1960 */
1961 #ifdef NFSKERB
1962 XXX
1963 #endif
1964
1965 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1966 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1967 if (nuidp->nu_expire < time_second ||
1968 nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1969 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1970 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1971 nuidp->nu_expire = 0;
1972 nd->nd_repstat =
1973 (NFSERR_AUTHERR|AUTH_REJECTVERF);
1974 nd->nd_procnum = NFSPROC_NOOP;
1975 return (0);
1976 }
1977 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1978 nd->nd_flag |= ND_KERBNICK;
1979 };
1980 } else {
1981 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1982 nd->nd_procnum = NFSPROC_NOOP;
1983 return (0);
1984 }
1985
1986 nd->nd_md = md;
1987 nd->nd_dpos = dpos;
1988 return (0);
1989 nfsmout:
1990 return (error);
1991 }
1992
1993 #endif
1994
1995 /*
1996 * Send a message to the originating process's terminal. The thread and/or
1997 * process may be NULL. YYY the thread should not be NULL but there may
1998 * still be some uio_td's that are still being passed as NULL through to
1999 * nfsm_request().
2000 */
2001 static int
2002 nfs_msg(struct thread *td, char *server, char *msg)
2003 {
2004 tpr_t tpr;
2005
2006 if (td && td->td_proc)
2007 tpr = tprintf_open(td->td_proc);
2008 else
2009 tpr = NULL;
2010 tprintf(tpr, "nfs server %s: %s\n", server, msg);
2011 tprintf_close(tpr);
2012 return (0);
2013 }
2014
2015 #ifndef NFS_NOSERVER
2016 /*
2017 * Socket upcall routine for the nfsd sockets.
2018 * The caddr_t arg is a pointer to the "struct nfssvc_sock".
2019 * Essentially do as much as possible non-blocking, else punt and it will
2020 * be called with MB_WAIT from an nfsd.
2021 */
2022 void
2023 nfsrv_rcv(struct socket *so, void *arg, int waitflag)
2024 {
2025 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
2026 struct mbuf *m;
2027 struct mbuf *mp;
2028 struct sockaddr *nam;
2029 struct uio auio;
2030 int flags, error;
2031 int nparallel_wakeup = 0;
2032
2033 if ((slp->ns_flag & SLP_VALID) == 0)
2034 return;
2035
2036 /*
2037 * Do not allow an infinite number of completed RPC records to build
2038 * up before we stop reading data from the socket. Otherwise we could
2039 * end up holding onto an unreasonable number of mbufs for requests
2040 * waiting for service.
2041 *
2042 * This should give pretty good feedback to the TCP
2043 * layer and prevents a memory crunch for other protocols.
2044 *
2045 * Note that the same service socket can be dispatched to several
2046 * nfs servers simultaniously.
2047 *
2048 * the tcp protocol callback calls us with MB_DONTWAIT.
2049 * nfsd calls us with MB_WAIT (typically).
2050 */
2051 if (waitflag == MB_DONTWAIT && slp->ns_numrec >= nfsd_waiting / 2 + 1) {
2052 slp->ns_flag |= SLP_NEEDQ;
2053 goto dorecs;
2054 }
2055
2056 /*
2057 * Handle protocol specifics to parse an RPC request. We always
2058 * pull from the socket using non-blocking I/O.
2059 */
2060 auio.uio_td = NULL;
2061 if (so->so_type == SOCK_STREAM) {
2062 /*
2063 * The data has to be read in an orderly fashion from a TCP
2064 * stream, unlike a UDP socket. It is possible for soreceive
2065 * and/or nfsrv_getstream() to block, so make sure only one
2066 * entity is messing around with the TCP stream at any given
2067 * moment. The receive sockbuf's lock in soreceive is not
2068 * sufficient.
2069 *
2070 * Note that this procedure can be called from any number of
2071 * NFS severs *OR* can be upcalled directly from a TCP
2072 * protocol thread.
2073 */
2074 if (slp->ns_flag & SLP_GETSTREAM) {
2075 slp->ns_flag |= SLP_NEEDQ;
2076 goto dorecs;
2077 }
2078 slp->ns_flag |= SLP_GETSTREAM;
2079
2080 /*
2081 * Do soreceive().
2082 */
2083 auio.uio_resid = 1000000000;
2084 flags = MSG_DONTWAIT;
2085 error = so_pru_soreceive(so, &nam, &auio, &mp, NULL, &flags);
2086 if (error || mp == (struct mbuf *)0) {
2087 if (error == EWOULDBLOCK)
2088 slp->ns_flag |= SLP_NEEDQ;
2089 else
2090 slp->ns_flag |= SLP_DISCONN;
2091 slp->ns_flag &= ~SLP_GETSTREAM;
2092 goto dorecs;
2093 }
2094 m = mp;
2095 if (slp->ns_rawend) {
2096 slp->ns_rawend->m_next = m;
2097 slp->ns_cc += 1000000000 - auio.uio_resid;
2098 } else {
2099 slp->ns_raw = m;
2100 slp->ns_cc = 1000000000 - auio.uio_resid;
2101 }
2102 while (m->m_next)
2103 m = m->m_next;
2104 slp->ns_rawend = m;
2105
2106 /*
2107 * Now try and parse as many record(s) as we can out of the
2108 * raw stream data.
2109 */
2110 error = nfsrv_getstream(slp, waitflag, &nparallel_wakeup);
2111 if (error) {
2112 if (error == EPERM)
2113 slp->ns_flag |= SLP_DISCONN;
2114 else
2115 slp->ns_flag |= SLP_NEEDQ;
2116 }
2117 slp->ns_flag &= ~SLP_GETSTREAM;
2118 } else {
2119 /*
2120 * For UDP soreceive typically pulls just one packet, loop
2121 * to get the whole batch.
2122 */
2123 do {
2124 auio.uio_resid = 1000000000;
2125 flags = MSG_DONTWAIT;
2126 error = so_pru_soreceive(so, &nam, &auio, &mp, NULL,
2127 &flags);
2128 if (mp) {
2129 struct nfsrv_rec *rec;
2130 int mf = (waitflag & MB_DONTWAIT) ?
2131 M_NOWAIT : M_WAITOK;
2132 rec = kmalloc(sizeof(struct nfsrv_rec),
2133 M_NFSRVDESC, mf);
2134 if (!rec) {
2135 if (nam)
2136 FREE(nam, M_SONAME);
2137 m_freem(mp);
2138 continue;
2139 }
2140 nfs_realign(&mp, 10 * NFSX_UNSIGNED);
2141 rec->nr_address = nam;
2142 rec->nr_packet = mp;
2143 STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
2144 ++slp->ns_numrec;
2145 ++nparallel_wakeup;
2146 }
2147 if (error) {
2148 if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2149 && error != EWOULDBLOCK) {
2150 slp->ns_flag |= SLP_DISCONN;
2151 goto dorecs;
2152 }
2153 }
2154 } while (mp);
2155 }
2156
2157 /*
2158 * If we were upcalled from the tcp protocol layer and we have
2159 * fully parsed records ready to go, or there is new data pending,
2160 * or something went wrong, try to wake up an nfsd thread to deal
2161 * with it.
2162 */
2163 dorecs:
2164 if (waitflag == MB_DONTWAIT && (slp->ns_numrec > 0
2165 || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) {
2166 nfsrv_wakenfsd(slp, nparallel_wakeup);
2167 }
2168 }
2169
2170 /*
2171 * Try and extract an RPC request from the mbuf data list received on a
2172 * stream socket. The "waitflag" argument indicates whether or not it
2173 * can sleep.
2174 */
2175 static int
2176 nfsrv_getstream(struct nfssvc_sock *slp, int waitflag, int *countp)
2177 {
2178 struct mbuf *m, **mpp;
2179 char *cp1, *cp2;
2180 int len;
2181 struct mbuf *om, *m2, *recm;
2182 u_int32_t recmark;
2183
2184 for (;;) {
2185 if (slp->ns_reclen == 0) {
2186 if (slp->ns_cc < NFSX_UNSIGNED)
2187 return (0);
2188 m = slp->ns_raw;
2189 if (m->m_len >= NFSX_UNSIGNED) {
2190 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
2191 m->m_data += NFSX_UNSIGNED;
2192 m->m_len -= NFSX_UNSIGNED;
2193 } else {
2194 cp1 = (caddr_t)&recmark;
2195 cp2 = mtod(m, caddr_t);
2196 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
2197 while (m->m_len == 0) {
2198 m = m->m_next;
2199 cp2 = mtod(m, caddr_t);
2200 }
2201 *cp1++ = *cp2++;
2202 m->m_data++;
2203 m->m_len--;
2204 }
2205 }
2206 slp->ns_cc -= NFSX_UNSIGNED;
2207 recmark = ntohl(recmark);
2208 slp->ns_reclen = recmark & ~0x80000000;
2209 if (recmark & 0x80000000)
2210 slp->ns_flag |= SLP_LASTFRAG;
2211 else
2212 slp->ns_flag &= ~SLP_LASTFRAG;
2213 if (slp->ns_reclen > NFS_MAXPACKET || slp->ns_reclen <= 0) {
2214 log(LOG_ERR, "%s (%d) from nfs client\n",
2215 "impossible packet length",
2216 slp->ns_reclen);
2217 return (EPERM);
2218 }
2219 }
2220
2221 /*
2222 * Now get the record part.
2223 *
2224 * Note that slp->ns_reclen may be 0. Linux sometimes
2225 * generates 0-length RPCs
2226 */
2227 recm = NULL;
2228 if (slp->ns_cc == slp->ns_reclen) {
2229 recm = slp->ns_raw;
2230 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2231 slp->ns_cc = slp->ns_reclen = 0;
2232 } else if (slp->ns_cc > slp->ns_reclen) {
2233 len = 0;
2234 m = slp->ns_raw;
2235 om = (struct mbuf *)0;
2236
2237 while (len < slp->ns_reclen) {
2238 if ((len + m->m_len) > slp->ns_reclen) {
2239 m2 = m_copym(m, 0, slp->ns_reclen - len,
2240 waitflag);
2241 if (m2) {
2242 if (om) {
2243 om->m_next = m2;
2244 recm = slp->ns_raw;
2245 } else
2246 recm = m2;
2247 m->m_data += slp->ns_reclen - len;
2248 m->m_len -= slp->ns_reclen - len;
2249 len = slp->ns_reclen;
2250 } else {
2251 return (EWOULDBLOCK);
2252 }
2253 } else if ((len + m->m_len) == slp->ns_reclen) {
2254 om = m;
2255 len += m->m_len;
2256 m = m->m_next;
2257 recm = slp->ns_raw;
2258 om->m_next = (struct mbuf *)0;
2259 } else {
2260 om = m;
2261 len += m->m_len;
2262 m = m->m_next;
2263 }
2264 }
2265 slp->ns_raw = m;
2266 slp->ns_cc -= len;
2267 slp->ns_reclen = 0;
2268 } else {
2269 return (0);
2270 }
2271
2272 /*
2273 * Accumulate the fragments into a record.
2274 */
2275 mpp = &slp->ns_frag;
2276 while (*mpp)
2277 mpp = &((*mpp)->m_next);
2278 *mpp = recm;
2279 if (slp->ns_flag & SLP_LASTFRAG) {
2280 struct nfsrv_rec *rec;
2281 int mf = (waitflag & MB_DONTWAIT) ? M_NOWAIT : M_WAITOK;
2282 rec = kmalloc(sizeof(struct nfsrv_rec), M_NFSRVDESC, mf);
2283 if (!rec) {
2284 m_freem(slp->ns_frag);
2285 } else {
2286 nfs_realign(&slp->ns_frag, 10 * NFSX_UNSIGNED);
2287 rec->nr_address = (struct sockaddr *)0;
2288 rec->nr_packet = slp->ns_frag;
2289 STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
2290 ++slp->ns_numrec;
2291 ++*countp;
2292 }
2293 slp->ns_frag = (struct mbuf *)0;
2294 }
2295 }
2296 }
2297
2298 /*
2299 * Parse an RPC header.
2300 */
2301 int
2302 nfsrv_dorec(struct nfssvc_sock *slp, struct nfsd *nfsd,
2303 struct nfsrv_descript **ndp)
2304 {
2305 struct nfsrv_rec *rec;
2306 struct mbuf *m;
2307 struct sockaddr *nam;
2308 struct nfsrv_descript *nd;
2309 int error;
2310
2311 *ndp = NULL;
2312 if ((slp->ns_flag & SLP_VALID) == 0 || !STAILQ_FIRST(&slp->ns_rec))
2313 return (ENOBUFS);
2314 rec = STAILQ_FIRST(&slp->ns_rec);
2315 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
2316 KKASSERT(slp->ns_numrec > 0);
2317 --slp->ns_numrec;
2318 nam = rec->nr_address;
2319 m = rec->nr_packet;
2320 kfree(rec, M_NFSRVDESC);
2321 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2322 M_NFSRVDESC, M_WAITOK);
2323 nd->nd_md = nd->nd_mrep = m;
2324 nd->nd_nam2 = nam;
2325 nd->nd_dpos = mtod(m, caddr_t);
2326 error = nfs_getreq(nd, nfsd, TRUE);
2327 if (error) {
2328 if (nam) {
2329 FREE(nam, M_SONAME);
2330 }
2331 kfree((caddr_t)nd, M_NFSRVDESC);
2332 return (error);
2333 }
2334 *ndp = nd;
2335 nfsd->nfsd_nd = nd;
2336 return (0);
2337 }
2338
2339 /*
2340 * Try to assign service sockets to nfsd threads based on the number
2341 * of new rpc requests that have been queued on the service socket.
2342 *
2343 * If no nfsd's are available or additonal requests are pending, set the
2344 * NFSD_CHECKSLP flag so that one of the running nfsds will go look for
2345 * the work in the nfssvc_sock list when it is finished processing its
2346 * current work. This flag is only cleared when an nfsd can not find
2347 * any new work to perform.
2348 */
2349 void
2350 nfsrv_wakenfsd(struct nfssvc_sock *slp, int nparallel)
2351 {
2352 struct nfsd *nd;
2353
2354 if ((slp->ns_flag & SLP_VALID) == 0)
2355 return;
2356 if (nparallel <= 1)
2357 nparallel = 1;
2358 TAILQ_FOREACH(nd, &nfsd_head, nfsd_chain) {
2359 if (nd->nfsd_flag & NFSD_WAITING) {
2360 nd->nfsd_flag &= ~NFSD_WAITING;
2361 if (nd->nfsd_slp)
2362 panic("nfsd wakeup");
2363 slp->ns_sref++;
2364 nd->nfsd_slp = slp;
2365 wakeup((caddr_t)nd);
2366 if (--nparallel == 0)
2367 break;
2368 }
2369 }
2370 if (nparallel) {
2371 slp->ns_flag |= SLP_DOREC;
2372 nfsd_head_flag |= NFSD_CHECKSLP;
2373 }
2374 }
2375 #endif /* NFS_NOSERVER */
DragonFly BSD