In contrast to FreeBSD 4 and 5, our slab allocator does hand out cross-page
[dragonfly.git] / sys / vfs / nfs / nfs_socket.c
CommitLineData
984263bc
MD
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 $
b2eb81cd 38 * $DragonFly: src/sys/vfs/nfs/nfs_socket.c,v 1.15 2004/04/07 05:15:48 dillon Exp $
984263bc
MD
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/protosw.h>
e4700d00 54#include <sys/resourcevar.h>
984263bc
MD
55#include <sys/socket.h>
56#include <sys/socketvar.h>
6b6e0885 57#include <sys/socketops.h>
984263bc 58#include <sys/syslog.h>
e4700d00 59#include <sys/thread.h>
984263bc
MD
60#include <sys/tprintf.h>
61#include <sys/sysctl.h>
62#include <sys/signalvar.h>
63
64#include <netinet/in.h>
65#include <netinet/tcp.h>
54938b92 66#include <sys/thread2.h>
984263bc 67
1f2de5d4
MD
68#include "rpcv2.h"
69#include "nfsproto.h"
70#include "nfs.h"
71#include "xdr_subs.h"
72#include "nfsm_subs.h"
73#include "nfsmount.h"
74#include "nfsnode.h"
75#include "nfsrtt.h"
76#include "nqnfs.h"
984263bc
MD
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 */
103extern 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;
106extern u_int32_t nfs_prog, nqnfs_prog;
107extern time_t nqnfsstarttime;
108extern struct nfsstats nfsstats;
109extern int nfsv3_procid[NFS_NPROCS];
110extern 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 */
120static 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
125static int nfs_realign_test;
126static int nfs_realign_count;
127static int nfs_bufpackets = 4;
128
129SYSCTL_DECL(_vfs_nfs);
130
131SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_test, CTLFLAG_RW, &nfs_realign_test, 0, "");
132SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_count, CTLFLAG_RW, &nfs_realign_count, 0, "");
133SYSCTL_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)
151static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
152int nfsrtton = 0;
153struct nfsrtt nfsrtt;
154struct callout_handle nfs_timer_handle;
155
a6ee311a
RG
156static int nfs_msg (struct thread *,char *,char *);
157static int nfs_rcvlock (struct nfsreq *);
158static void nfs_rcvunlock (struct nfsreq *);
159static void nfs_realign (struct mbuf **pm, int hsiz);
160static int nfs_receive (struct nfsreq *rep, struct sockaddr **aname,
161 struct mbuf **mp);
162static void nfs_softterm (struct nfsreq *rep);
163static int nfs_reconnect (struct nfsreq *rep);
984263bc 164#ifndef NFS_NOSERVER
a6ee311a 165static int nfsrv_getstream (struct nfssvc_sock *,int);
984263bc 166
a6ee311a 167int (*nfsrv3_procs[NFS_NPROCS]) (struct nfsrv_descript *nd,
984263bc 168 struct nfssvc_sock *slp,
dadab5e9 169 struct thread *td,
a6ee311a 170 struct mbuf **mreqp) = {
984263bc
MD
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 nqnfsrv_getlease,
194 nqnfsrv_vacated,
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 */
204int
dadab5e9 205nfs_connect(struct nfsmount *nmp, struct nfsreq *rep)
984263bc 206{
dadab5e9 207 struct socket *so;
984263bc
MD
208 int s, error, rcvreserve, sndreserve;
209 int pktscale;
210 struct sockaddr *saddr;
211 struct sockaddr_in *sin;
dadab5e9 212 struct thread *td = &thread0; /* only used for socreate and sobind */
984263bc
MD
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,
dadab5e9 217 nmp->nm_soproto, td);
984263bc
MD
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_dir = SOPT_SET;
234 sopt.sopt_level = IPPROTO_IP;
235 sopt.sopt_name = IP_PORTRANGE;
236 sopt.sopt_val = (void *)&ip;
237 sopt.sopt_valsize = sizeof(ip);
dadab5e9 238 sopt.sopt_td = NULL;
984263bc
MD
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);
dadab5e9 248 error = sobind(so, (struct sockaddr *)sin, td);
984263bc
MD
249 if (error)
250 goto bad;
251 bzero(&sopt, sizeof sopt);
252 ip = IP_PORTRANGE_DEFAULT;
253 sopt.sopt_dir = SOPT_SET;
254 sopt.sopt_level = IPPROTO_IP;
255 sopt.sopt_name = IP_PORTRANGE;
256 sopt.sopt_val = (void *)&ip;
257 sopt.sopt_valsize = sizeof(ip);
dadab5e9 258 sopt.sopt_td = NULL;
984263bc
MD
259 error = sosetopt(so, &sopt);
260 if (error)
261 goto bad;
262 }
263
264 /*
265 * Protocols that do not require connections may be optionally left
266 * unconnected for servers that reply from a port other than NFS_PORT.
267 */
268 if (nmp->nm_flag & NFSMNT_NOCONN) {
269 if (nmp->nm_soflags & PR_CONNREQUIRED) {
270 error = ENOTCONN;
271 goto bad;
272 }
273 } else {
dadab5e9 274 error = soconnect(so, nmp->nm_nam, td);
984263bc
MD
275 if (error)
276 goto bad;
277
278 /*
279 * Wait for the connection to complete. Cribbed from the
280 * connect system call but with the wait timing out so
281 * that interruptible mounts don't hang here for a long time.
282 */
283 s = splnet();
284 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
377d4740 285 (void) tsleep((caddr_t)&so->so_timeo, 0,
984263bc
MD
286 "nfscon", 2 * hz);
287 if ((so->so_state & SS_ISCONNECTING) &&
288 so->so_error == 0 && rep &&
dadab5e9 289 (error = nfs_sigintr(nmp, rep, rep->r_td)) != 0){
984263bc
MD
290 so->so_state &= ~SS_ISCONNECTING;
291 splx(s);
292 goto bad;
293 }
294 }
295 if (so->so_error) {
296 error = so->so_error;
297 so->so_error = 0;
298 splx(s);
299 goto bad;
300 }
301 splx(s);
302 }
303 so->so_rcv.sb_timeo = (5 * hz);
304 so->so_snd.sb_timeo = (5 * hz);
305
306 /*
307 * Get buffer reservation size from sysctl, but impose reasonable
308 * limits.
309 */
310 pktscale = nfs_bufpackets;
311 if (pktscale < 2)
312 pktscale = 2;
313 if (pktscale > 64)
314 pktscale = 64;
315
316 if (nmp->nm_sotype == SOCK_DGRAM) {
317 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
318 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
319 NFS_MAXPKTHDR) * pktscale;
320 } else if (nmp->nm_sotype == SOCK_SEQPACKET) {
321 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
322 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
323 NFS_MAXPKTHDR) * pktscale;
324 } else {
325 if (nmp->nm_sotype != SOCK_STREAM)
326 panic("nfscon sotype");
327 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
328 struct sockopt sopt;
329 int val;
330
331 bzero(&sopt, sizeof sopt);
332 sopt.sopt_level = SOL_SOCKET;
333 sopt.sopt_name = SO_KEEPALIVE;
334 sopt.sopt_val = &val;
335 sopt.sopt_valsize = sizeof val;
336 val = 1;
337 sosetopt(so, &sopt);
338 }
339 if (so->so_proto->pr_protocol == IPPROTO_TCP) {
340 struct sockopt sopt;
341 int val;
342
343 bzero(&sopt, sizeof sopt);
344 sopt.sopt_level = IPPROTO_TCP;
345 sopt.sopt_name = TCP_NODELAY;
346 sopt.sopt_val = &val;
347 sopt.sopt_valsize = sizeof val;
348 val = 1;
349 sosetopt(so, &sopt);
350 }
351 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
352 sizeof (u_int32_t)) * pktscale;
353 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
354 sizeof (u_int32_t)) * pktscale;
355 }
e4700d00
JH
356 error = soreserve(so, sndreserve, rcvreserve,
357 &td->td_proc->p_rlimit[RLIMIT_SBSIZE]);
984263bc
MD
358 if (error)
359 goto bad;
360 so->so_rcv.sb_flags |= SB_NOINTR;
361 so->so_snd.sb_flags |= SB_NOINTR;
362
363 /* Initialize other non-zero congestion variables */
364 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] =
365 nmp->nm_srtt[3] = (NFS_TIMEO << 3);
366 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
367 nmp->nm_sdrtt[3] = 0;
368 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */
369 nmp->nm_sent = 0;
370 nmp->nm_timeouts = 0;
371 return (0);
372
373bad:
374 nfs_disconnect(nmp);
375 return (error);
376}
377
378/*
379 * Reconnect routine:
380 * Called when a connection is broken on a reliable protocol.
381 * - clean up the old socket
382 * - nfs_connect() again
383 * - set R_MUSTRESEND for all outstanding requests on mount point
384 * If this fails the mount point is DEAD!
385 * nb: Must be called with the nfs_sndlock() set on the mount point.
386 */
387static int
388nfs_reconnect(rep)
40393ded 389 struct nfsreq *rep;
984263bc 390{
40393ded
RG
391 struct nfsreq *rp;
392 struct nfsmount *nmp = rep->r_nmp;
984263bc
MD
393 int error;
394
395 nfs_disconnect(nmp);
396 while ((error = nfs_connect(nmp, rep)) != 0) {
397 if (error == EINTR || error == ERESTART)
398 return (EINTR);
377d4740 399 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
984263bc
MD
400 }
401
402 /*
403 * Loop through outstanding request list and fix up all requests
404 * on old socket.
405 */
406 for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) {
407 if (rp->r_nmp == nmp)
408 rp->r_flags |= R_MUSTRESEND;
409 }
410 return (0);
411}
412
413/*
414 * NFS disconnect. Clean up and unlink.
415 */
416void
417nfs_disconnect(nmp)
40393ded 418 struct nfsmount *nmp;
984263bc 419{
40393ded 420 struct socket *so;
984263bc
MD
421
422 if (nmp->nm_so) {
423 so = nmp->nm_so;
424 nmp->nm_so = (struct socket *)0;
425 soshutdown(so, 2);
426 soclose(so);
427 }
428}
429
430void
431nfs_safedisconnect(nmp)
432 struct nfsmount *nmp;
433{
434 struct nfsreq dummyreq;
435
436 bzero(&dummyreq, sizeof(dummyreq));
437 dummyreq.r_nmp = nmp;
cd990953 438 dummyreq.r_td = NULL;
984263bc
MD
439 nfs_rcvlock(&dummyreq);
440 nfs_disconnect(nmp);
441 nfs_rcvunlock(&dummyreq);
442}
443
444/*
445 * This is the nfs send routine. For connection based socket types, it
446 * must be called with an nfs_sndlock() on the socket.
447 * "rep == NULL" indicates that it has been called from a server.
448 * For the client side:
449 * - return EINTR if the RPC is terminated, 0 otherwise
450 * - set R_MUSTRESEND if the send fails for any reason
451 * - do any cleanup required by recoverable socket errors (?)
452 * For the server side:
453 * - return EINTR or ERESTART if interrupted by a signal
454 * - return EPIPE if a connection is lost for connection based sockets (TCP...)
455 * - do any cleanup required by recoverable socket errors (?)
456 */
457int
458nfs_send(so, nam, top, rep)
40393ded 459 struct socket *so;
984263bc 460 struct sockaddr *nam;
40393ded 461 struct mbuf *top;
984263bc
MD
462 struct nfsreq *rep;
463{
464 struct sockaddr *sendnam;
465 int error, soflags, flags;
466
467 if (rep) {
468 if (rep->r_flags & R_SOFTTERM) {
469 m_freem(top);
470 return (EINTR);
471 }
472 if ((so = rep->r_nmp->nm_so) == NULL) {
473 rep->r_flags |= R_MUSTRESEND;
474 m_freem(top);
475 return (0);
476 }
477 rep->r_flags &= ~R_MUSTRESEND;
478 soflags = rep->r_nmp->nm_soflags;
479 } else
480 soflags = so->so_proto->pr_flags;
481 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
482 sendnam = (struct sockaddr *)0;
483 else
484 sendnam = nam;
485 if (so->so_type == SOCK_SEQPACKET)
486 flags = MSG_EOR;
487 else
488 flags = 0;
489
6b6e0885
JH
490 error = so_pru_sosend(so, sendnam, NULL, top, NULL, flags,
491 curthread /*XXX*/);
984263bc
MD
492 /*
493 * ENOBUFS for dgram sockets is transient and non fatal.
494 * No need to log, and no need to break a soft mount.
495 */
496 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
497 error = 0;
498 if (rep) /* do backoff retransmit on client */
499 rep->r_flags |= R_MUSTRESEND;
500 }
501
502 if (error) {
503 if (rep) {
504 log(LOG_INFO, "nfs send error %d for server %s\n",error,
505 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
506 /*
507 * Deal with errors for the client side.
508 */
509 if (rep->r_flags & R_SOFTTERM)
510 error = EINTR;
511 else
512 rep->r_flags |= R_MUSTRESEND;
513 } else
514 log(LOG_INFO, "nfsd send error %d\n", error);
515
516 /*
517 * Handle any recoverable (soft) socket errors here. (?)
518 */
519 if (error != EINTR && error != ERESTART &&
520 error != EWOULDBLOCK && error != EPIPE)
521 error = 0;
522 }
523 return (error);
524}
525
526/*
527 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
528 * done by soreceive(), but for SOCK_STREAM we must deal with the Record
529 * Mark and consolidate the data into a new mbuf list.
530 * nb: Sometimes TCP passes the data up to soreceive() in long lists of
531 * small mbufs.
532 * For SOCK_STREAM we must be very careful to read an entire record once
533 * we have read any of it, even if the system call has been interrupted.
534 */
535static int
dadab5e9 536nfs_receive(struct nfsreq *rep, struct sockaddr **aname, struct mbuf **mp)
984263bc 537{
40393ded 538 struct socket *so;
984263bc
MD
539 struct uio auio;
540 struct iovec aio;
40393ded 541 struct mbuf *m;
984263bc
MD
542 struct mbuf *control;
543 u_int32_t len;
544 struct sockaddr **getnam;
545 int error, sotype, rcvflg;
dadab5e9 546 struct thread *td = curthread; /* XXX */
984263bc
MD
547
548 /*
549 * Set up arguments for soreceive()
550 */
551 *mp = (struct mbuf *)0;
552 *aname = (struct sockaddr *)0;
553 sotype = rep->r_nmp->nm_sotype;
554
555 /*
556 * For reliable protocols, lock against other senders/receivers
557 * in case a reconnect is necessary.
558 * For SOCK_STREAM, first get the Record Mark to find out how much
559 * more there is to get.
560 * We must lock the socket against other receivers
561 * until we have an entire rpc request/reply.
562 */
563 if (sotype != SOCK_DGRAM) {
564 error = nfs_sndlock(rep);
565 if (error)
566 return (error);
567tryagain:
568 /*
569 * Check for fatal errors and resending request.
570 */
571 /*
572 * Ugh: If a reconnect attempt just happened, nm_so
573 * would have changed. NULL indicates a failed
574 * attempt that has essentially shut down this
575 * mount point.
576 */
577 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
578 nfs_sndunlock(rep);
579 return (EINTR);
580 }
581 so = rep->r_nmp->nm_so;
582 if (!so) {
583 error = nfs_reconnect(rep);
584 if (error) {
585 nfs_sndunlock(rep);
586 return (error);
587 }
588 goto tryagain;
589 }
590 while (rep->r_flags & R_MUSTRESEND) {
591 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
592 nfsstats.rpcretries++;
593 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
594 if (error) {
595 if (error == EINTR || error == ERESTART ||
596 (error = nfs_reconnect(rep)) != 0) {
597 nfs_sndunlock(rep);
598 return (error);
599 }
600 goto tryagain;
601 }
602 }
603 nfs_sndunlock(rep);
604 if (sotype == SOCK_STREAM) {
605 aio.iov_base = (caddr_t) &len;
606 aio.iov_len = sizeof(u_int32_t);
607 auio.uio_iov = &aio;
608 auio.uio_iovcnt = 1;
609 auio.uio_segflg = UIO_SYSSPACE;
610 auio.uio_rw = UIO_READ;
611 auio.uio_offset = 0;
612 auio.uio_resid = sizeof(u_int32_t);
dadab5e9 613 auio.uio_td = td;
984263bc
MD
614 do {
615 rcvflg = MSG_WAITALL;
6b6e0885
JH
616 error = so_pru_soreceive(so, NULL, &auio, NULL,
617 NULL, &rcvflg);
984263bc
MD
618 if (error == EWOULDBLOCK && rep) {
619 if (rep->r_flags & R_SOFTTERM)
620 return (EINTR);
621 }
622 } while (error == EWOULDBLOCK);
623 if (!error && auio.uio_resid > 0) {
624 /*
625 * Don't log a 0 byte receive; it means
626 * that the socket has been closed, and
627 * can happen during normal operation
628 * (forcible unmount or Solaris server).
629 */
630 if (auio.uio_resid != sizeof (u_int32_t))
631 log(LOG_INFO,
632 "short receive (%d/%d) from nfs server %s\n",
633 (int)(sizeof(u_int32_t) - auio.uio_resid),
634 (int)sizeof(u_int32_t),
635 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
636 error = EPIPE;
637 }
638 if (error)
639 goto errout;
640 len = ntohl(len) & ~0x80000000;
641 /*
642 * This is SERIOUS! We are out of sync with the sender
643 * and forcing a disconnect/reconnect is all I can do.
644 */
645 if (len > NFS_MAXPACKET) {
646 log(LOG_ERR, "%s (%d) from nfs server %s\n",
647 "impossible packet length",
648 len,
649 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
650 error = EFBIG;
651 goto errout;
652 }
653 auio.uio_resid = len;
654 do {
655 rcvflg = MSG_WAITALL;
6b6e0885
JH
656 error = so_pru_soreceive(so, NULL, &auio, mp,
657 NULL, &rcvflg);
984263bc
MD
658 } while (error == EWOULDBLOCK || error == EINTR ||
659 error == ERESTART);
660 if (!error && auio.uio_resid > 0) {
661 if (len != auio.uio_resid)
662 log(LOG_INFO,
663 "short receive (%d/%d) from nfs server %s\n",
664 len - auio.uio_resid, len,
665 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
666 error = EPIPE;
667 }
668 } else {
669 /*
670 * NB: Since uio_resid is big, MSG_WAITALL is ignored
671 * and soreceive() will return when it has either a
672 * control msg or a data msg.
673 * We have no use for control msg., but must grab them
674 * and then throw them away so we know what is going
675 * on.
676 */
677 auio.uio_resid = len = 100000000; /* Anything Big */
dadab5e9 678 auio.uio_td = td;
984263bc
MD
679 do {
680 rcvflg = 0;
6b6e0885
JH
681 error = so_pru_soreceive(so, NULL, &auio, mp,
682 &control, &rcvflg);
984263bc
MD
683 if (control)
684 m_freem(control);
685 if (error == EWOULDBLOCK && rep) {
686 if (rep->r_flags & R_SOFTTERM)
687 return (EINTR);
688 }
689 } while (error == EWOULDBLOCK ||
690 (!error && *mp == NULL && control));
691 if ((rcvflg & MSG_EOR) == 0)
692 printf("Egad!!\n");
693 if (!error && *mp == NULL)
694 error = EPIPE;
695 len -= auio.uio_resid;
696 }
697errout:
698 if (error && error != EINTR && error != ERESTART) {
699 m_freem(*mp);
700 *mp = (struct mbuf *)0;
701 if (error != EPIPE)
702 log(LOG_INFO,
703 "receive error %d from nfs server %s\n",
704 error,
705 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
706 error = nfs_sndlock(rep);
707 if (!error) {
708 error = nfs_reconnect(rep);
709 if (!error)
710 goto tryagain;
711 else
712 nfs_sndunlock(rep);
713 }
714 }
715 } else {
716 if ((so = rep->r_nmp->nm_so) == NULL)
717 return (EACCES);
718 if (so->so_state & SS_ISCONNECTED)
719 getnam = (struct sockaddr **)0;
720 else
721 getnam = aname;
722 auio.uio_resid = len = 1000000;
dadab5e9 723 auio.uio_td = td;
984263bc
MD
724 do {
725 rcvflg = 0;
6b6e0885
JH
726 error = so_pru_soreceive(so, getnam, &auio, mp, NULL,
727 &rcvflg);
984263bc
MD
728 if (error == EWOULDBLOCK &&
729 (rep->r_flags & R_SOFTTERM))
730 return (EINTR);
731 } while (error == EWOULDBLOCK);
732 len -= auio.uio_resid;
733 }
734 if (error) {
735 m_freem(*mp);
736 *mp = (struct mbuf *)0;
737 }
738 /*
739 * Search for any mbufs that are not a multiple of 4 bytes long
740 * or with m_data not longword aligned.
741 * These could cause pointer alignment problems, so copy them to
742 * well aligned mbufs.
743 */
744 nfs_realign(mp, 5 * NFSX_UNSIGNED);
745 return (error);
746}
747
748/*
749 * Implement receipt of reply on a socket.
750 * We must search through the list of received datagrams matching them
751 * with outstanding requests using the xid, until ours is found.
752 */
753/* ARGSUSED */
754int
755nfs_reply(myrep)
756 struct nfsreq *myrep;
757{
40393ded
RG
758 struct nfsreq *rep;
759 struct nfsmount *nmp = myrep->r_nmp;
760 int32_t t1;
984263bc
MD
761 struct mbuf *mrep, *md;
762 struct sockaddr *nam;
763 u_int32_t rxid, *tl;
764 caddr_t dpos, cp2;
765 int error;
766
767 /*
768 * Loop around until we get our own reply
769 */
770 for (;;) {
771 /*
772 * Lock against other receivers so that I don't get stuck in
773 * sbwait() after someone else has received my reply for me.
774 * Also necessary for connection based protocols to avoid
775 * race conditions during a reconnect.
776 * If nfs_rcvlock() returns EALREADY, that means that
777 * the reply has already been recieved by another
778 * process and we can return immediately. In this
779 * case, the lock is not taken to avoid races with
780 * other processes.
781 */
782 error = nfs_rcvlock(myrep);
783 if (error == EALREADY)
784 return (0);
785 if (error)
786 return (error);
787 /*
788 * Get the next Rpc reply off the socket
789 */
790 error = nfs_receive(myrep, &nam, &mrep);
791 nfs_rcvunlock(myrep);
792 if (error) {
793
794 /*
795 * Ignore routing errors on connectionless protocols??
796 */
797 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
798 nmp->nm_so->so_error = 0;
799 if (myrep->r_flags & R_GETONEREP)
800 return (0);
801 continue;
802 }
803 return (error);
804 }
805 if (nam)
806 FREE(nam, M_SONAME);
807
808 /*
809 * Get the xid and check that it is an rpc reply
810 */
811 md = mrep;
812 dpos = mtod(md, caddr_t);
813 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
814 rxid = *tl++;
815 if (*tl != rpc_reply) {
816#ifndef NFS_NOSERVER
817 if (nmp->nm_flag & NFSMNT_NQNFS) {
818 if (nqnfs_callback(nmp, mrep, md, dpos))
819 nfsstats.rpcinvalid++;
820 } else {
821 nfsstats.rpcinvalid++;
822 m_freem(mrep);
823 }
824#else
825 nfsstats.rpcinvalid++;
826 m_freem(mrep);
827#endif
828nfsmout:
829 if (myrep->r_flags & R_GETONEREP)
830 return (0);
831 continue;
832 }
833
834 /*
835 * Loop through the request list to match up the reply
836 * Iff no match, just drop the datagram
837 */
838 for (rep = nfs_reqq.tqh_first; rep != 0;
839 rep = rep->r_chain.tqe_next) {
840 if (rep->r_mrep == NULL && rxid == rep->r_xid) {
841 /* Found it.. */
842 rep->r_mrep = mrep;
843 rep->r_md = md;
844 rep->r_dpos = dpos;
845 if (nfsrtton) {
846 struct rttl *rt;
847
848 rt = &nfsrtt.rttl[nfsrtt.pos];
849 rt->proc = rep->r_procnum;
850 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
851 rt->sent = nmp->nm_sent;
852 rt->cwnd = nmp->nm_cwnd;
853 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
854 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
855 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
856 getmicrotime(&rt->tstamp);
857 if (rep->r_flags & R_TIMING)
858 rt->rtt = rep->r_rtt;
859 else
860 rt->rtt = 1000000;
861 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
862 }
863 /*
864 * Update congestion window.
865 * Do the additive increase of
866 * one rpc/rtt.
867 */
868 if (nmp->nm_cwnd <= nmp->nm_sent) {
869 nmp->nm_cwnd +=
870 (NFS_CWNDSCALE * NFS_CWNDSCALE +
871 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
872 if (nmp->nm_cwnd > NFS_MAXCWND)
873 nmp->nm_cwnd = NFS_MAXCWND;
874 }
54938b92 875 crit_enter(); /* nfs_timer interlock*/
984263bc
MD
876 if (rep->r_flags & R_SENT) {
877 rep->r_flags &= ~R_SENT;
878 nmp->nm_sent -= NFS_CWNDSCALE;
879 }
54938b92 880 crit_exit();
984263bc
MD
881 /*
882 * Update rtt using a gain of 0.125 on the mean
883 * and a gain of 0.25 on the deviation.
884 */
885 if (rep->r_flags & R_TIMING) {
886 /*
887 * Since the timer resolution of
888 * NFS_HZ is so course, it can often
889 * result in r_rtt == 0. Since
890 * r_rtt == N means that the actual
891 * rtt is between N+dt and N+2-dt ticks,
892 * add 1.
893 */
894 t1 = rep->r_rtt + 1;
895 t1 -= (NFS_SRTT(rep) >> 3);
896 NFS_SRTT(rep) += t1;
897 if (t1 < 0)
898 t1 = -t1;
899 t1 -= (NFS_SDRTT(rep) >> 2);
900 NFS_SDRTT(rep) += t1;
901 }
902 nmp->nm_timeouts = 0;
903 break;
904 }
905 }
906 /*
907 * If not matched to a request, drop it.
908 * If it's mine, get out.
909 */
910 if (rep == 0) {
911 nfsstats.rpcunexpected++;
912 m_freem(mrep);
913 } else if (rep == myrep) {
914 if (rep->r_mrep == NULL)
915 panic("nfsreply nil");
916 return (0);
917 }
918 if (myrep->r_flags & R_GETONEREP)
919 return (0);
920 }
921}
922
923/*
924 * nfs_request - goes something like this
925 * - fill in request struct
926 * - links it into list
927 * - calls nfs_send() for first transmit
928 * - calls nfs_receive() to get reply
929 * - break down rpc header and return with nfs reply pointed to
930 * by mrep or error
931 * nb: always frees up mreq mbuf list
932 */
933int
dadab5e9 934nfs_request(vp, mrest, procnum, td, cred, mrp, mdp, dposp)
984263bc
MD
935 struct vnode *vp;
936 struct mbuf *mrest;
937 int procnum;
dadab5e9 938 struct thread *td;
984263bc
MD
939 struct ucred *cred;
940 struct mbuf **mrp;
941 struct mbuf **mdp;
942 caddr_t *dposp;
943{
40393ded
RG
944 struct mbuf *mrep, *m2;
945 struct nfsreq *rep;
946 u_int32_t *tl;
947 int i;
984263bc
MD
948 struct nfsmount *nmp;
949 struct mbuf *m, *md, *mheadend;
950 struct nfsnode *np;
951 char nickv[RPCX_NICKVERF];
952 time_t reqtime, waituntil;
953 caddr_t dpos, cp2;
954 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type;
955 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0;
956 int verf_len, verf_type;
957 u_int32_t xid;
958 u_quad_t frev;
959 char *auth_str, *verf_str;
960 NFSKERBKEY_T key; /* save session key */
961
962 /* Reject requests while attempting a forced unmount. */
963 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) {
964 m_freem(mrest);
965 return (ESTALE);
966 }
967 nmp = VFSTONFS(vp->v_mount);
968 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
969 rep->r_nmp = nmp;
970 rep->r_vp = vp;
dadab5e9 971 rep->r_td = td;
984263bc
MD
972 rep->r_procnum = procnum;
973 i = 0;
974 m = mrest;
975 while (m) {
976 i += m->m_len;
977 m = m->m_next;
978 }
979 mrest_len = i;
980
981 /*
982 * Get the RPC header with authorization.
983 */
984kerbauth:
985 verf_str = auth_str = (char *)0;
986 if (nmp->nm_flag & NFSMNT_KERB) {
987 verf_str = nickv;
988 verf_len = sizeof (nickv);
989 auth_type = RPCAUTH_KERB4;
990 bzero((caddr_t)key, sizeof (key));
991 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
992 &auth_len, verf_str, verf_len)) {
993 error = nfs_getauth(nmp, rep, cred, &auth_str,
994 &auth_len, verf_str, &verf_len, key);
995 if (error) {
996 free((caddr_t)rep, M_NFSREQ);
997 m_freem(mrest);
998 return (error);
999 }
1000 }
1001 } else {
1002 auth_type = RPCAUTH_UNIX;
1003 if (cred->cr_ngroups < 1)
1004 panic("nfsreq nogrps");
1005 auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ?
1006 nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) +
1007 5 * NFSX_UNSIGNED;
1008 }
1009 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
1010 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
1011 if (auth_str)
1012 free(auth_str, M_TEMP);
1013
1014 /*
1015 * For stream protocols, insert a Sun RPC Record Mark.
1016 */
1017 if (nmp->nm_sotype == SOCK_STREAM) {
1018 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
1cee5817
HP
1019 if (m == NULL)
1020 return (ENOBUFS);
984263bc
MD
1021 *mtod(m, u_int32_t *) = htonl(0x80000000 |
1022 (m->m_pkthdr.len - NFSX_UNSIGNED));
1023 }
1024 rep->r_mreq = m;
1025 rep->r_xid = xid;
1026tryagain:
1027 if (nmp->nm_flag & NFSMNT_SOFT)
1028 rep->r_retry = nmp->nm_retry;
1029 else
1030 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */
1031 rep->r_rtt = rep->r_rexmit = 0;
1032 if (proct[procnum] > 0)
54938b92 1033 rep->r_flags = R_TIMING | R_MASKTIMER;
984263bc 1034 else
54938b92 1035 rep->r_flags = R_MASKTIMER;
984263bc
MD
1036 rep->r_mrep = NULL;
1037
1038 /*
1039 * Do the client side RPC.
1040 */
1041 nfsstats.rpcrequests++;
54938b92 1042
984263bc
MD
1043 /*
1044 * Chain request into list of outstanding requests. Be sure
54938b92
MD
1045 * to put it LAST so timer finds oldest requests first. Note
1046 * that R_MASKTIMER is set at the moment to prevent any timer
1047 * action on this request while we are still doing processing on
1048 * it below. splsoftclock() primarily protects nm_sent. Note
1049 * that we may block in this code so there is no atomicy guarentee.
984263bc
MD
1050 */
1051 s = splsoftclock();
1052 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1053
1054 /* Get send time for nqnfs */
1055 reqtime = time_second;
1056
1057 /*
1058 * If backing off another request or avoiding congestion, don't
1059 * send this one now but let timer do it. If not timing a request,
1060 * do it now.
1061 */
1062 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1063 (nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1064 nmp->nm_sent < nmp->nm_cwnd)) {
984263bc
MD
1065 if (nmp->nm_soflags & PR_CONNREQUIRED)
1066 error = nfs_sndlock(rep);
1067 if (!error) {
1068 m2 = m_copym(m, 0, M_COPYALL, M_WAIT);
1069 error = nfs_send(nmp->nm_so, nmp->nm_nam, m2, rep);
1070 if (nmp->nm_soflags & PR_CONNREQUIRED)
1071 nfs_sndunlock(rep);
1072 }
1073 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1074 nmp->nm_sent += NFS_CWNDSCALE;
1075 rep->r_flags |= R_SENT;
1076 }
1077 } else {
984263bc
MD
1078 rep->r_rtt = -1;
1079 }
1080
1081 /*
54938b92
MD
1082 * Let the timer do what it will with the request, then
1083 * wait for the reply from our send or the timer's.
984263bc 1084 */
54938b92
MD
1085 rep->r_flags &= ~R_MASKTIMER;
1086 splx(s);
984263bc
MD
1087 if (!error || error == EPIPE)
1088 error = nfs_reply(rep);
1089
1090 /*
1091 * RPC done, unlink the request.
1092 */
1093 s = splsoftclock();
1094 TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
984263bc
MD
1095
1096 /*
1097 * Decrement the outstanding request count.
1098 */
1099 if (rep->r_flags & R_SENT) {
54938b92 1100 rep->r_flags &= ~R_SENT;
984263bc
MD
1101 nmp->nm_sent -= NFS_CWNDSCALE;
1102 }
54938b92 1103 splx(s);
984263bc
MD
1104
1105 /*
1106 * If there was a successful reply and a tprintf msg.
1107 * tprintf a response.
1108 */
1109 if (!error && (rep->r_flags & R_TPRINTFMSG))
dadab5e9 1110 nfs_msg(rep->r_td, nmp->nm_mountp->mnt_stat.f_mntfromname,
984263bc
MD
1111 "is alive again");
1112 mrep = rep->r_mrep;
1113 md = rep->r_md;
1114 dpos = rep->r_dpos;
1115 if (error) {
1116 m_freem(rep->r_mreq);
1117 free((caddr_t)rep, M_NFSREQ);
1118 return (error);
1119 }
1120
1121 /*
1122 * break down the rpc header and check if ok
1123 */
1124 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1125 if (*tl++ == rpc_msgdenied) {
1126 if (*tl == rpc_mismatch)
1127 error = EOPNOTSUPP;
1128 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1129 if (!failed_auth) {
1130 failed_auth++;
1131 mheadend->m_next = (struct mbuf *)0;
1132 m_freem(mrep);
1133 m_freem(rep->r_mreq);
1134 goto kerbauth;
1135 } else
1136 error = EAUTH;
1137 } else
1138 error = EACCES;
1139 m_freem(mrep);
1140 m_freem(rep->r_mreq);
1141 free((caddr_t)rep, M_NFSREQ);
1142 return (error);
1143 }
1144
1145 /*
1146 * Grab any Kerberos verifier, otherwise just throw it away.
1147 */
1148 verf_type = fxdr_unsigned(int, *tl++);
1149 i = fxdr_unsigned(int32_t, *tl);
1150 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1151 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1152 if (error)
1153 goto nfsmout;
1154 } else if (i > 0)
1155 nfsm_adv(nfsm_rndup(i));
1156 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1157 /* 0 == ok */
1158 if (*tl == 0) {
1159 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1160 if (*tl != 0) {
1161 error = fxdr_unsigned(int, *tl);
1162 if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1163 error == NFSERR_TRYLATER) {
1164 m_freem(mrep);
1165 error = 0;
1166 waituntil = time_second + trylater_delay;
1167 while (time_second < waituntil)
1168 (void) tsleep((caddr_t)&lbolt,
377d4740 1169 0, "nqnfstry", 0);
984263bc
MD
1170 trylater_delay *= nfs_backoff[trylater_cnt];
1171 if (trylater_cnt < 7)
1172 trylater_cnt++;
1173 goto tryagain;
1174 }
1175
1176 /*
1177 * If the File Handle was stale, invalidate the
1178 * lookup cache, just in case.
1179 */
1180 if (error == ESTALE)
1181 cache_purge(vp);
1182 if (nmp->nm_flag & NFSMNT_NFSV3) {
1183 *mrp = mrep;
1184 *mdp = md;
1185 *dposp = dpos;
1186 error |= NFSERR_RETERR;
1187 } else
1188 m_freem(mrep);
1189 m_freem(rep->r_mreq);
1190 free((caddr_t)rep, M_NFSREQ);
1191 return (error);
1192 }
1193
1194 /*
1195 * For nqnfs, get any lease in reply
1196 */
1197 if (nmp->nm_flag & NFSMNT_NQNFS) {
1198 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1199 if (*tl) {
1200 np = VTONFS(vp);
1201 nqlflag = fxdr_unsigned(int, *tl);
1202 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1203 cachable = fxdr_unsigned(int, *tl++);
1204 reqtime += fxdr_unsigned(int, *tl++);
1205 if (reqtime > time_second) {
1206 frev = fxdr_hyper(tl);
1207 nqnfs_clientlease(nmp, np, nqlflag,
1208 cachable, reqtime, frev);
1209 }
1210 }
1211 }
1212 *mrp = mrep;
1213 *mdp = md;
1214 *dposp = dpos;
1215 m_freem(rep->r_mreq);
1216 FREE((caddr_t)rep, M_NFSREQ);
1217 return (0);
1218 }
1219 m_freem(mrep);
1220 error = EPROTONOSUPPORT;
1221nfsmout:
1222 m_freem(rep->r_mreq);
1223 free((caddr_t)rep, M_NFSREQ);
1224 return (error);
1225}
1226
1227#ifndef NFS_NOSERVER
1228/*
1229 * Generate the rpc reply header
1230 * siz arg. is used to decide if adding a cluster is worthwhile
1231 */
1232int
1233nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1234 int siz;
1235 struct nfsrv_descript *nd;
1236 struct nfssvc_sock *slp;
1237 int err;
1238 int cache;
1239 u_quad_t *frev;
1240 struct mbuf **mrq;
1241 struct mbuf **mbp;
1242 caddr_t *bposp;
1243{
40393ded
RG
1244 u_int32_t *tl;
1245 struct mbuf *mreq;
984263bc
MD
1246 caddr_t bpos;
1247 struct mbuf *mb, *mb2;
1248
1249 MGETHDR(mreq, M_WAIT, MT_DATA);
1250 mb = mreq;
1251 /*
1252 * If this is a big reply, use a cluster else
1253 * try and leave leading space for the lower level headers.
1254 */
1255 siz += RPC_REPLYSIZ;
1256 if ((max_hdr + siz) >= MINCLSIZE) {
1257 MCLGET(mreq, M_WAIT);
1258 } else
1259 mreq->m_data += max_hdr;
1260 tl = mtod(mreq, u_int32_t *);
1261 mreq->m_len = 6 * NFSX_UNSIGNED;
1262 bpos = ((caddr_t)tl) + mreq->m_len;
1263 *tl++ = txdr_unsigned(nd->nd_retxid);
1264 *tl++ = rpc_reply;
1265 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1266 *tl++ = rpc_msgdenied;
1267 if (err & NFSERR_AUTHERR) {
1268 *tl++ = rpc_autherr;
1269 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1270 mreq->m_len -= NFSX_UNSIGNED;
1271 bpos -= NFSX_UNSIGNED;
1272 } else {
1273 *tl++ = rpc_mismatch;
1274 *tl++ = txdr_unsigned(RPC_VER2);
1275 *tl = txdr_unsigned(RPC_VER2);
1276 }
1277 } else {
1278 *tl++ = rpc_msgaccepted;
1279
1280 /*
1281 * For Kerberos authentication, we must send the nickname
1282 * verifier back, otherwise just RPCAUTH_NULL.
1283 */
1284 if (nd->nd_flag & ND_KERBFULL) {
40393ded 1285 struct nfsuid *nuidp;
984263bc
MD
1286 struct timeval ktvin, ktvout;
1287
1288 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1289 nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1290 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1291 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1292 &nuidp->nu_haddr, nd->nd_nam2)))
1293 break;
1294 }
1295 if (nuidp) {
1296 ktvin.tv_sec =
1297 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1298 ktvin.tv_usec =
1299 txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1300
1301 /*
1302 * Encrypt the timestamp in ecb mode using the
1303 * session key.
1304 */
1305#ifdef NFSKERB
1306 XXX
1307#endif
1308
1309 *tl++ = rpc_auth_kerb;
1310 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1311 *tl = ktvout.tv_sec;
1312 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1313 *tl++ = ktvout.tv_usec;
1314 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1315 } else {
1316 *tl++ = 0;
1317 *tl++ = 0;
1318 }
1319 } else {
1320 *tl++ = 0;
1321 *tl++ = 0;
1322 }
1323 switch (err) {
1324 case EPROGUNAVAIL:
1325 *tl = txdr_unsigned(RPC_PROGUNAVAIL);
1326 break;
1327 case EPROGMISMATCH:
1328 *tl = txdr_unsigned(RPC_PROGMISMATCH);
1329 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1330 if (nd->nd_flag & ND_NQNFS) {
1331 *tl++ = txdr_unsigned(3);
1332 *tl = txdr_unsigned(3);
1333 } else {
1334 *tl++ = txdr_unsigned(2);
1335 *tl = txdr_unsigned(3);
1336 }
1337 break;
1338 case EPROCUNAVAIL:
1339 *tl = txdr_unsigned(RPC_PROCUNAVAIL);
1340 break;
1341 case EBADRPC:
1342 *tl = txdr_unsigned(RPC_GARBAGE);
1343 break;
1344 default:
1345 *tl = 0;
1346 if (err != NFSERR_RETVOID) {
1347 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1348 if (err)
1349 *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1350 else
1351 *tl = 0;
1352 }
1353 break;
1354 };
1355 }
1356
1357 /*
1358 * For nqnfs, piggyback lease as requested.
1359 */
1360 if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1361 if (nd->nd_flag & ND_LEASE) {
1362 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1363 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1364 *tl++ = txdr_unsigned(cache);
1365 *tl++ = txdr_unsigned(nd->nd_duration);
1366 txdr_hyper(*frev, tl);
1367 } else {
1368 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1369 *tl = 0;
1370 }
1371 }
1372 if (mrq != NULL)
1373 *mrq = mreq;
1374 *mbp = mb;
1375 *bposp = bpos;
1376 if (err != 0 && err != NFSERR_RETVOID)
1377 nfsstats.srvrpc_errs++;
1378 return (0);
1379}
1380
1381
1382#endif /* NFS_NOSERVER */
1383/*
1384 * Nfs timer routine
1385 * Scan the nfsreq list and retranmit any requests that have timed out
1386 * To avoid retransmission attempts on STREAM sockets (in the future) make
1387 * sure to set the r_retry field to 0 (implies nm_retry == 0).
1388 */
1389void
1390nfs_timer(arg)
1391 void *arg; /* never used */
1392{
40393ded
RG
1393 struct nfsreq *rep;
1394 struct mbuf *m;
1395 struct socket *so;
1396 struct nfsmount *nmp;
1397 int timeo;
984263bc
MD
1398 int s, error;
1399#ifndef NFS_NOSERVER
1400 static long lasttime = 0;
40393ded 1401 struct nfssvc_sock *slp;
984263bc
MD
1402 u_quad_t cur_usec;
1403#endif /* NFS_NOSERVER */
dadab5e9 1404 struct thread *td = &thread0; /* XXX for credentials, will break if sleep */
984263bc
MD
1405
1406 s = splnet();
1407 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) {
1408 nmp = rep->r_nmp;
54938b92 1409 if (rep->r_mrep || (rep->r_flags & (R_SOFTTERM|R_MASKTIMER)))
984263bc 1410 continue;
dadab5e9 1411 if (nfs_sigintr(nmp, rep, rep->r_td)) {
984263bc
MD
1412 nfs_softterm(rep);
1413 continue;
1414 }
1415 if (rep->r_rtt >= 0) {
1416 rep->r_rtt++;
1417 if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1418 timeo = nmp->nm_timeo;
1419 else
1420 timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1421 if (nmp->nm_timeouts > 0)
1422 timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1423 if (rep->r_rtt <= timeo)
1424 continue;
1425 if (nmp->nm_timeouts < 8)
1426 nmp->nm_timeouts++;
1427 }
1428 /*
1429 * Check for server not responding
1430 */
1431 if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1432 rep->r_rexmit > nmp->nm_deadthresh) {
dadab5e9 1433 nfs_msg(rep->r_td,
984263bc
MD
1434 nmp->nm_mountp->mnt_stat.f_mntfromname,
1435 "not responding");
1436 rep->r_flags |= R_TPRINTFMSG;
1437 }
1438 if (rep->r_rexmit >= rep->r_retry) { /* too many */
1439 nfsstats.rpctimeouts++;
1440 nfs_softterm(rep);
1441 continue;
1442 }
1443 if (nmp->nm_sotype != SOCK_DGRAM) {
1444 if (++rep->r_rexmit > NFS_MAXREXMIT)
1445 rep->r_rexmit = NFS_MAXREXMIT;
1446 continue;
1447 }
1448 if ((so = nmp->nm_so) == NULL)
1449 continue;
1450
1451 /*
1452 * If there is enough space and the window allows..
1453 * Resend it
1454 * Set r_rtt to -1 in case we fail to send it now.
1455 */
1456 rep->r_rtt = -1;
1457 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1458 ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1459 (rep->r_flags & R_SENT) ||
1460 nmp->nm_sent < nmp->nm_cwnd) &&
1461 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1462 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
6b6e0885 1463 error = so_pru_send(so, 0, m, (struct sockaddr *)0,
dadab5e9 1464 (struct mbuf *)0, td);
984263bc 1465 else
6b6e0885
JH
1466 error = so_pru_send(so, 0, m, nmp->nm_nam,
1467 (struct mbuf *)0, td);
984263bc
MD
1468 if (error) {
1469 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
1470 so->so_error = 0;
1471 } else {
1472 /*
1473 * Iff first send, start timing
1474 * else turn timing off, backoff timer
1475 * and divide congestion window by 2.
1476 */
1477 if (rep->r_flags & R_SENT) {
1478 rep->r_flags &= ~R_TIMING;
1479 if (++rep->r_rexmit > NFS_MAXREXMIT)
1480 rep->r_rexmit = NFS_MAXREXMIT;
1481 nmp->nm_cwnd >>= 1;
1482 if (nmp->nm_cwnd < NFS_CWNDSCALE)
1483 nmp->nm_cwnd = NFS_CWNDSCALE;
1484 nfsstats.rpcretries++;
1485 } else {
1486 rep->r_flags |= R_SENT;
1487 nmp->nm_sent += NFS_CWNDSCALE;
1488 }
1489 rep->r_rtt = 0;
1490 }
1491 }
1492 }
1493#ifndef NFS_NOSERVER
1494 /*
1495 * Call the nqnfs server timer once a second to handle leases.
1496 */
1497 if (lasttime != time_second) {
1498 lasttime = time_second;
1499 nqnfs_serverd();
1500 }
1501
1502 /*
1503 * Scan the write gathering queues for writes that need to be
1504 * completed now.
1505 */
1506 cur_usec = nfs_curusec();
1507 for (slp = nfssvc_sockhead.tqh_first; slp != 0;
1508 slp = slp->ns_chain.tqe_next) {
1509 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1510 nfsrv_wakenfsd(slp);
1511 }
1512#endif /* NFS_NOSERVER */
1513 splx(s);
1514 nfs_timer_handle = timeout(nfs_timer, (void *)0, nfs_ticks);
1515}
1516
1517/*
1518 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1519 * wait for all requests to complete. This is used by forced unmounts
1520 * to terminate any outstanding RPCs.
1521 */
1522int
1523nfs_nmcancelreqs(nmp)
1524 struct nfsmount *nmp;
1525{
1526 struct nfsreq *req;
54938b92 1527 int i, s1, s2;
984263bc 1528
54938b92
MD
1529 s1 = splnet();
1530 s2 = splsoftclock();
984263bc
MD
1531 TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
1532 if (nmp != req->r_nmp || req->r_mrep != NULL ||
1533 (req->r_flags & R_SOFTTERM))
1534 continue;
1535 nfs_softterm(req);
1536 }
54938b92
MD
1537 splx(s2);
1538 splx(s1);
984263bc
MD
1539
1540 for (i = 0; i < 30; i++) {
54938b92 1541 int s = splnet();
984263bc
MD
1542 TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
1543 if (nmp == req->r_nmp)
1544 break;
1545 }
1546 splx(s);
1547 if (req == NULL)
1548 return (0);
377d4740 1549 tsleep(&lbolt, 0, "nfscancel", 0);
984263bc
MD
1550 }
1551 return (EBUSY);
1552}
1553
1554/*
1555 * Flag a request as being about to terminate (due to NFSMNT_INT/NFSMNT_SOFT).
1556 * The nm_send count is decremented now to avoid deadlocks when the process in
1557 * soreceive() hasn't yet managed to send its own request.
54938b92
MD
1558 *
1559 * This routine must be called at splsoftclock() to protect r_flags and
1560 * nm_sent.
984263bc
MD
1561 */
1562
1563static void
1564nfs_softterm(rep)
1565 struct nfsreq *rep;
1566{
1567 rep->r_flags |= R_SOFTTERM;
1568
1569 if (rep->r_flags & R_SENT) {
1570 rep->r_nmp->nm_sent -= NFS_CWNDSCALE;
1571 rep->r_flags &= ~R_SENT;
1572 }
1573}
1574
1575/*
1576 * Test for a termination condition pending on the process.
1577 * This is used for NFSMNT_INT mounts.
1578 */
1579int
dadab5e9 1580nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct thread *td)
984263bc
MD
1581{
1582 sigset_t tmpset;
dadab5e9 1583 struct proc *p;
984263bc
MD
1584
1585 if (rep && (rep->r_flags & R_SOFTTERM))
1586 return (EINTR);
1587 /* Terminate all requests while attempting a forced unmount. */
1588 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1589 return (EINTR);
1590 if (!(nmp->nm_flag & NFSMNT_INT))
1591 return (0);
cd990953
MD
1592 /* td might be NULL YYY */
1593 if (td == NULL || (p = td->td_proc) == NULL)
984263bc
MD
1594 return (0);
1595
1596 tmpset = p->p_siglist;
1597 SIGSETNAND(tmpset, p->p_sigmask);
1598 SIGSETNAND(tmpset, p->p_sigignore);
1599 if (SIGNOTEMPTY(p->p_siglist) && NFSINT_SIGMASK(tmpset))
1600 return (EINTR);
1601
1602 return (0);
1603}
1604
1605/*
1606 * Lock a socket against others.
1607 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1608 * and also to avoid race conditions between the processes with nfs requests
1609 * in progress when a reconnect is necessary.
1610 */
1611int
dadab5e9 1612nfs_sndlock(struct nfsreq *rep)
984263bc 1613{
dadab5e9
MD
1614 int *statep = &rep->r_nmp->nm_state;
1615 struct thread *td;
b2eb81cd
MD
1616 int slptimeo;
1617 int slpflag;
1618 int error;
984263bc 1619
b2eb81cd
MD
1620 slpflag = 0;
1621 slptimeo = 0;
dadab5e9 1622 td = rep->r_td;
984263bc
MD
1623 if (rep->r_nmp->nm_flag & NFSMNT_INT)
1624 slpflag = PCATCH;
b2eb81cd
MD
1625
1626 error = 0;
1627 crit_enter();
984263bc 1628 while (*statep & NFSSTA_SNDLOCK) {
54938b92 1629 *statep |= NFSSTA_WANTSND;
b2eb81cd
MD
1630 if (nfs_sigintr(rep->r_nmp, rep, td)) {
1631 error = EINTR;
1632 break;
1633 }
1634 tsleep((caddr_t)statep, slpflag, "nfsndlck", slptimeo);
984263bc
MD
1635 if (slpflag == PCATCH) {
1636 slpflag = 0;
1637 slptimeo = 2 * hz;
1638 }
1639 }
1640 /* Always fail if our request has been cancelled. */
1641 if ((rep->r_flags & R_SOFTTERM))
b2eb81cd
MD
1642 error = EINTR;
1643 if (error == 0)
1644 *statep |= NFSSTA_SNDLOCK;
1645 crit_exit();
1646 return (error);
984263bc
MD
1647}
1648
1649/*
1650 * Unlock the stream socket for others.
1651 */
1652void
1653nfs_sndunlock(rep)
1654 struct nfsreq *rep;
1655{
40393ded 1656 int *statep = &rep->r_nmp->nm_state;
984263bc
MD
1657
1658 if ((*statep & NFSSTA_SNDLOCK) == 0)
1659 panic("nfs sndunlock");
b2eb81cd 1660 crit_enter();
984263bc
MD
1661 *statep &= ~NFSSTA_SNDLOCK;
1662 if (*statep & NFSSTA_WANTSND) {
1663 *statep &= ~NFSSTA_WANTSND;
1664 wakeup((caddr_t)statep);
1665 }
b2eb81cd 1666 crit_exit();
984263bc
MD
1667}
1668
1669static int
1670nfs_rcvlock(rep)
40393ded 1671 struct nfsreq *rep;
984263bc 1672{
40393ded 1673 int *statep = &rep->r_nmp->nm_state;
b2eb81cd
MD
1674 int slpflag;
1675 int slptimeo;
1676 int error;
984263bc 1677
54938b92
MD
1678 /*
1679 * Unconditionally check for completion in case another nfsiod
1680 * get the packet while the caller was blocked, before the caller
1681 * called us. Packet reception is handled by mainline code which
1682 * is protected by the BGL at the moment.
b2eb81cd
MD
1683 *
1684 * We do not strictly need the second check just before the
1685 * tsleep(), but it's good defensive programming.
54938b92
MD
1686 */
1687 if (rep->r_mrep != NULL)
1688 return (EALREADY);
1689
984263bc
MD
1690 if (rep->r_nmp->nm_flag & NFSMNT_INT)
1691 slpflag = PCATCH;
1692 else
1693 slpflag = 0;
b2eb81cd
MD
1694 slptimeo = 0;
1695 error = 0;
1696 crit_enter();
984263bc 1697 while (*statep & NFSSTA_RCVLOCK) {
b2eb81cd
MD
1698 if (nfs_sigintr(rep->r_nmp, rep, rep->r_td)) {
1699 error = EINTR;
1700 break;
1701 }
1702 if (rep->r_mrep != NULL) {
1703 error = EALREADY;
1704 break;
1705 }
984263bc 1706 *statep |= NFSSTA_WANTRCV;
b2eb81cd 1707 tsleep((caddr_t)statep, slpflag, "nfsrcvlk", slptimeo);
984263bc
MD
1708 /*
1709 * If our reply was recieved while we were sleeping,
1710 * then just return without taking the lock to avoid a
1711 * situation where a single iod could 'capture' the
1712 * recieve lock.
1713 */
b2eb81cd
MD
1714 if (rep->r_mrep != NULL) {
1715 error = EALREADY;
1716 break;
1717 }
984263bc
MD
1718 if (slpflag == PCATCH) {
1719 slpflag = 0;
1720 slptimeo = 2 * hz;
1721 }
1722 }
b2eb81cd
MD
1723 if (error == 0) {
1724 *statep |= NFSSTA_RCVLOCK;
1725 rep->r_nmp->nm_rcvlock_td = curthread; /* DEBUGGING */
1726 }
1727 crit_exit();
1728 return (error);
984263bc
MD
1729}
1730
1731/*
1732 * Unlock the stream socket for others.
1733 */
1734static void
1735nfs_rcvunlock(rep)
40393ded 1736 struct nfsreq *rep;
984263bc 1737{
40393ded 1738 int *statep = &rep->r_nmp->nm_state;
984263bc
MD
1739
1740 if ((*statep & NFSSTA_RCVLOCK) == 0)
1741 panic("nfs rcvunlock");
b2eb81cd
MD
1742 crit_enter();
1743 rep->r_nmp->nm_rcvlock_td = (void *)-1; /* DEBUGGING */
984263bc
MD
1744 *statep &= ~NFSSTA_RCVLOCK;
1745 if (*statep & NFSSTA_WANTRCV) {
1746 *statep &= ~NFSSTA_WANTRCV;
1747 wakeup((caddr_t)statep);
1748 }
b2eb81cd 1749 crit_exit();
984263bc
MD
1750}
1751
1752/*
1753 * nfs_realign:
1754 *
1755 * Check for badly aligned mbuf data and realign by copying the unaligned
1756 * portion of the data into a new mbuf chain and freeing the portions
1757 * of the old chain that were replaced.
1758 *
1759 * We cannot simply realign the data within the existing mbuf chain
1760 * because the underlying buffers may contain other rpc commands and
1761 * we cannot afford to overwrite them.
1762 *
1763 * We would prefer to avoid this situation entirely. The situation does
1764 * not occur with NFS/UDP and is supposed to only occassionally occur
1765 * with TCP. Use vfs.nfs.realign_count and realign_test to check this.
1766 */
1767static void
1768nfs_realign(pm, hsiz)
40393ded 1769 struct mbuf **pm;
984263bc
MD
1770 int hsiz;
1771{
1772 struct mbuf *m;
1773 struct mbuf *n = NULL;
1774 int off = 0;
1775
1776 ++nfs_realign_test;
1777
1778 while ((m = *pm) != NULL) {
1779 if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) {
1780 MGET(n, M_WAIT, MT_DATA);
1781 if (m->m_len >= MINCLSIZE) {
1782 MCLGET(n, M_WAIT);
1783 }
1784 n->m_len = 0;
1785 break;
1786 }
1787 pm = &m->m_next;
1788 }
1789
1790 /*
1791 * If n is non-NULL, loop on m copying data, then replace the
1792 * portion of the chain that had to be realigned.
1793 */
1794 if (n != NULL) {
1795 ++nfs_realign_count;
1796 while (m) {
1797 m_copyback(n, off, m->m_len, mtod(m, caddr_t));
1798 off += m->m_len;
1799 m = m->m_next;
1800 }
1801 m_freem(*pm);
1802 *pm = n;
1803 }
1804}
1805
1806#ifndef NFS_NOSERVER
1807
1808/*
1809 * Parse an RPC request
1810 * - verify it
1811 * - fill in the cred struct.
1812 */
1813int
1814nfs_getreq(nd, nfsd, has_header)
40393ded 1815 struct nfsrv_descript *nd;
984263bc
MD
1816 struct nfsd *nfsd;
1817 int has_header;
1818{
40393ded
RG
1819 int len, i;
1820 u_int32_t *tl;
1821 int32_t t1;
984263bc
MD
1822 struct uio uio;
1823 struct iovec iov;
1824 caddr_t dpos, cp2, cp;
1825 u_int32_t nfsvers, auth_type;
1826 uid_t nickuid;
1827 int error = 0, nqnfs = 0, ticklen;
1828 struct mbuf *mrep, *md;
40393ded 1829 struct nfsuid *nuidp;
984263bc
MD
1830 struct timeval tvin, tvout;
1831#if 0 /* until encrypted keys are implemented */
1832 NFSKERBKEYSCHED_T keys; /* stores key schedule */
1833#endif
1834
1835 mrep = nd->nd_mrep;
1836 md = nd->nd_md;
1837 dpos = nd->nd_dpos;
1838 if (has_header) {
1839 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1840 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1841 if (*tl++ != rpc_call) {
1842 m_freem(mrep);
1843 return (EBADRPC);
1844 }
1845 } else
1846 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1847 nd->nd_repstat = 0;
1848 nd->nd_flag = 0;
1849 if (*tl++ != rpc_vers) {
1850 nd->nd_repstat = ERPCMISMATCH;
1851 nd->nd_procnum = NFSPROC_NOOP;
1852 return (0);
1853 }
1854 if (*tl != nfs_prog) {
1855 if (*tl == nqnfs_prog)
1856 nqnfs++;
1857 else {
1858 nd->nd_repstat = EPROGUNAVAIL;
1859 nd->nd_procnum = NFSPROC_NOOP;
1860 return (0);
1861 }
1862 }
1863 tl++;
1864 nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1865 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1866 (nfsvers != NQNFS_VER3 && nqnfs)) {
1867 nd->nd_repstat = EPROGMISMATCH;
1868 nd->nd_procnum = NFSPROC_NOOP;
1869 return (0);
1870 }
1871 if (nqnfs)
1872 nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1873 else if (nfsvers == NFS_VER3)
1874 nd->nd_flag = ND_NFSV3;
1875 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1876 if (nd->nd_procnum == NFSPROC_NULL)
1877 return (0);
1878 if (nd->nd_procnum >= NFS_NPROCS ||
1879 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1880 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1881 nd->nd_repstat = EPROCUNAVAIL;
1882 nd->nd_procnum = NFSPROC_NOOP;
1883 return (0);
1884 }
1885 if ((nd->nd_flag & ND_NFSV3) == 0)
1886 nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1887 auth_type = *tl++;
1888 len = fxdr_unsigned(int, *tl++);
1889 if (len < 0 || len > RPCAUTH_MAXSIZ) {
1890 m_freem(mrep);
1891 return (EBADRPC);
1892 }
1893
1894 nd->nd_flag &= ~ND_KERBAUTH;
1895 /*
1896 * Handle auth_unix or auth_kerb.
1897 */
1898 if (auth_type == rpc_auth_unix) {
1899 len = fxdr_unsigned(int, *++tl);
1900 if (len < 0 || len > NFS_MAXNAMLEN) {
1901 m_freem(mrep);
1902 return (EBADRPC);
1903 }
1904 nfsm_adv(nfsm_rndup(len));
1905 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1906 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred));
1907 nd->nd_cr.cr_ref = 1;
1908 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1909 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1910 len = fxdr_unsigned(int, *tl);
1911 if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1912 m_freem(mrep);
1913 return (EBADRPC);
1914 }
1915 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1916 for (i = 1; i <= len; i++)
1917 if (i < NGROUPS)
1918 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1919 else
1920 tl++;
1921 nd->nd_cr.cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1);
1922 if (nd->nd_cr.cr_ngroups > 1)
1923 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1924 len = fxdr_unsigned(int, *++tl);
1925 if (len < 0 || len > RPCAUTH_MAXSIZ) {
1926 m_freem(mrep);
1927 return (EBADRPC);
1928 }
1929 if (len > 0)
1930 nfsm_adv(nfsm_rndup(len));
1931 } else if (auth_type == rpc_auth_kerb) {
1932 switch (fxdr_unsigned(int, *tl++)) {
1933 case RPCAKN_FULLNAME:
1934 ticklen = fxdr_unsigned(int, *tl);
1935 *((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1936 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1937 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1938 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1939 m_freem(mrep);
1940 return (EBADRPC);
1941 }
1942 uio.uio_offset = 0;
1943 uio.uio_iov = &iov;
1944 uio.uio_iovcnt = 1;
1945 uio.uio_segflg = UIO_SYSSPACE;
1946 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1947 iov.iov_len = RPCAUTH_MAXSIZ - 4;
1948 nfsm_mtouio(&uio, uio.uio_resid);
1949 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1950 if (*tl++ != rpc_auth_kerb ||
1951 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1952 printf("Bad kerb verifier\n");
1953 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1954 nd->nd_procnum = NFSPROC_NOOP;
1955 return (0);
1956 }
1957 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1958 tl = (u_int32_t *)cp;
1959 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1960 printf("Not fullname kerb verifier\n");
1961 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1962 nd->nd_procnum = NFSPROC_NOOP;
1963 return (0);
1964 }
1965 cp += NFSX_UNSIGNED;
1966 bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED);
1967 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1968 nd->nd_flag |= ND_KERBFULL;
1969 nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1970 break;
1971 case RPCAKN_NICKNAME:
1972 if (len != 2 * NFSX_UNSIGNED) {
1973 printf("Kerb nickname short\n");
1974 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1975 nd->nd_procnum = NFSPROC_NOOP;
1976 return (0);
1977 }
1978 nickuid = fxdr_unsigned(uid_t, *tl);
1979 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1980 if (*tl++ != rpc_auth_kerb ||
1981 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1982 printf("Kerb nick verifier bad\n");
1983 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1984 nd->nd_procnum = NFSPROC_NOOP;
1985 return (0);
1986 }
1987 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1988 tvin.tv_sec = *tl++;
1989 tvin.tv_usec = *tl;
1990
1991 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1992 nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1993 if (nuidp->nu_cr.cr_uid == nickuid &&
1994 (!nd->nd_nam2 ||
1995 netaddr_match(NU_NETFAM(nuidp),
1996 &nuidp->nu_haddr, nd->nd_nam2)))
1997 break;
1998 }
1999 if (!nuidp) {
2000 nd->nd_repstat =
2001 (NFSERR_AUTHERR|AUTH_REJECTCRED);
2002 nd->nd_procnum = NFSPROC_NOOP;
2003 return (0);
2004 }
2005
2006 /*
2007 * Now, decrypt the timestamp using the session key
2008 * and validate it.
2009 */
2010#ifdef NFSKERB
2011 XXX
2012#endif
2013
2014 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
2015 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
2016 if (nuidp->nu_expire < time_second ||
2017 nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
2018 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
2019 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
2020 nuidp->nu_expire = 0;
2021 nd->nd_repstat =
2022 (NFSERR_AUTHERR|AUTH_REJECTVERF);
2023 nd->nd_procnum = NFSPROC_NOOP;
2024 return (0);
2025 }
2026 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
2027 nd->nd_flag |= ND_KERBNICK;
2028 };
2029 } else {
2030 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
2031 nd->nd_procnum = NFSPROC_NOOP;
2032 return (0);
2033 }
2034
2035 /*
2036 * For nqnfs, get piggybacked lease request.
2037 */
2038 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
2039 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2040 nd->nd_flag |= fxdr_unsigned(int, *tl);
2041 if (nd->nd_flag & ND_LEASE) {
2042 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2043 nd->nd_duration = fxdr_unsigned(int32_t, *tl);
2044 } else
2045 nd->nd_duration = NQ_MINLEASE;
2046 } else
2047 nd->nd_duration = NQ_MINLEASE;
2048 nd->nd_md = md;
2049 nd->nd_dpos = dpos;
2050 return (0);
2051nfsmout:
2052 return (error);
2053}
2054
2055#endif
2056
b70ddfbf
MD
2057/*
2058 * Send a message to the originating process's terminal. The thread and/or
2059 * process may be NULL. YYY the thread should not be NULL but there may
2060 * still be some uio_td's that are still being passed as NULL through to
2061 * nfsm_request().
2062 */
984263bc 2063static int
dadab5e9 2064nfs_msg(struct thread *td, char *server, char *msg)
984263bc
MD
2065{
2066 tpr_t tpr;
2067
b70ddfbf 2068 if (td && td->td_proc)
dadab5e9 2069 tpr = tprintf_open(td->td_proc);
984263bc
MD
2070 else
2071 tpr = NULL;
2072 tprintf(tpr, "nfs server %s: %s\n", server, msg);
2073 tprintf_close(tpr);
2074 return (0);
2075}
2076
2077#ifndef NFS_NOSERVER
2078/*
2079 * Socket upcall routine for the nfsd sockets.
2080 * The caddr_t arg is a pointer to the "struct nfssvc_sock".
2081 * Essentially do as much as possible non-blocking, else punt and it will
2082 * be called with M_WAIT from an nfsd.
2083 */
2084void
2085nfsrv_rcv(so, arg, waitflag)
2086 struct socket *so;
2087 void *arg;
2088 int waitflag;
2089{
40393ded
RG
2090 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
2091 struct mbuf *m;
984263bc
MD
2092 struct mbuf *mp;
2093 struct sockaddr *nam;
2094 struct uio auio;
2095 int flags, error;
2096
2097 if ((slp->ns_flag & SLP_VALID) == 0)
2098 return;
2099#ifdef notdef
2100 /*
2101 * Define this to test for nfsds handling this under heavy load.
2102 */
2103 if (waitflag == M_DONTWAIT) {
2104 slp->ns_flag |= SLP_NEEDQ; goto dorecs;
2105 }
2106#endif
dadab5e9 2107 auio.uio_td = NULL;
984263bc
MD
2108 if (so->so_type == SOCK_STREAM) {
2109 /*
2110 * If there are already records on the queue, defer soreceive()
2111 * to an nfsd so that there is feedback to the TCP layer that
2112 * the nfs servers are heavily loaded.
2113 */
2114 if (STAILQ_FIRST(&slp->ns_rec) && waitflag == M_DONTWAIT) {
2115 slp->ns_flag |= SLP_NEEDQ;
2116 goto dorecs;
2117 }
2118
2119 /*
2120 * Do soreceive().
2121 */
2122 auio.uio_resid = 1000000000;
2123 flags = MSG_DONTWAIT;
6b6e0885 2124 error = so_pru_soreceive(so, &nam, &auio, &mp, NULL, &flags);
984263bc
MD
2125 if (error || mp == (struct mbuf *)0) {
2126 if (error == EWOULDBLOCK)
2127 slp->ns_flag |= SLP_NEEDQ;
2128 else
2129 slp->ns_flag |= SLP_DISCONN;
2130 goto dorecs;
2131 }
2132 m = mp;
2133 if (slp->ns_rawend) {
2134 slp->ns_rawend->m_next = m;
2135 slp->ns_cc += 1000000000 - auio.uio_resid;
2136 } else {
2137 slp->ns_raw = m;
2138 slp->ns_cc = 1000000000 - auio.uio_resid;
2139 }
2140 while (m->m_next)
2141 m = m->m_next;
2142 slp->ns_rawend = m;
2143
2144 /*
2145 * Now try and parse record(s) out of the raw stream data.
2146 */
2147 error = nfsrv_getstream(slp, waitflag);
2148 if (error) {
2149 if (error == EPERM)
2150 slp->ns_flag |= SLP_DISCONN;
2151 else
2152 slp->ns_flag |= SLP_NEEDQ;
2153 }
2154 } else {
2155 do {
2156 auio.uio_resid = 1000000000;
2157 flags = MSG_DONTWAIT;
6b6e0885
JH
2158 error = so_pru_soreceive(so, &nam, &auio, &mp, NULL,
2159 &flags);
984263bc
MD
2160 if (mp) {
2161 struct nfsrv_rec *rec;
da004c97
MD
2162 int mf = (waitflag & M_DONTWAIT) ?
2163 M_NOWAIT : M_WAITOK;
984263bc 2164 rec = malloc(sizeof(struct nfsrv_rec),
da004c97 2165 M_NFSRVDESC, mf);
984263bc
MD
2166 if (!rec) {
2167 if (nam)
2168 FREE(nam, M_SONAME);
2169 m_freem(mp);
2170 continue;
2171 }
2172 nfs_realign(&mp, 10 * NFSX_UNSIGNED);
2173 rec->nr_address = nam;
2174 rec->nr_packet = mp;
2175 STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
2176 }
2177 if (error) {
2178 if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2179 && error != EWOULDBLOCK) {
2180 slp->ns_flag |= SLP_DISCONN;
2181 goto dorecs;
2182 }
2183 }
2184 } while (mp);
2185 }
2186
2187 /*
2188 * Now try and process the request records, non-blocking.
2189 */
2190dorecs:
2191 if (waitflag == M_DONTWAIT &&
2192 (STAILQ_FIRST(&slp->ns_rec)
2193 || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
2194 nfsrv_wakenfsd(slp);
2195}
2196
2197/*
2198 * Try and extract an RPC request from the mbuf data list received on a
2199 * stream socket. The "waitflag" argument indicates whether or not it
2200 * can sleep.
2201 */
2202static int
2203nfsrv_getstream(slp, waitflag)
40393ded 2204 struct nfssvc_sock *slp;
984263bc
MD
2205 int waitflag;
2206{
40393ded
RG
2207 struct mbuf *m, **mpp;
2208 char *cp1, *cp2;
2209 int len;
984263bc
MD
2210 struct mbuf *om, *m2, *recm;
2211 u_int32_t recmark;
2212
2213 if (slp->ns_flag & SLP_GETSTREAM)
2214 panic("nfs getstream");
2215 slp->ns_flag |= SLP_GETSTREAM;
2216 for (;;) {
2217 if (slp->ns_reclen == 0) {
2218 if (slp->ns_cc < NFSX_UNSIGNED) {
2219 slp->ns_flag &= ~SLP_GETSTREAM;
2220 return (0);
2221 }
2222 m = slp->ns_raw;
2223 if (m->m_len >= NFSX_UNSIGNED) {
2224 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
2225 m->m_data += NFSX_UNSIGNED;
2226 m->m_len -= NFSX_UNSIGNED;
2227 } else {
2228 cp1 = (caddr_t)&recmark;
2229 cp2 = mtod(m, caddr_t);
2230 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
2231 while (m->m_len == 0) {
2232 m = m->m_next;
2233 cp2 = mtod(m, caddr_t);
2234 }
2235 *cp1++ = *cp2++;
2236 m->m_data++;
2237 m->m_len--;
2238 }
2239 }
2240 slp->ns_cc -= NFSX_UNSIGNED;
2241 recmark = ntohl(recmark);
2242 slp->ns_reclen = recmark & ~0x80000000;
2243 if (recmark & 0x80000000)
2244 slp->ns_flag |= SLP_LASTFRAG;
2245 else
2246 slp->ns_flag &= ~SLP_LASTFRAG;
2247 if (slp->ns_reclen > NFS_MAXPACKET) {
2248 slp->ns_flag &= ~SLP_GETSTREAM;
2249 return (EPERM);
2250 }
2251 }
2252
2253 /*
2254 * Now get the record part.
2255 *
2256 * Note that slp->ns_reclen may be 0. Linux sometimes
2257 * generates 0-length RPCs
2258 */
2259 recm = NULL;
2260 if (slp->ns_cc == slp->ns_reclen) {
2261 recm = slp->ns_raw;
2262 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2263 slp->ns_cc = slp->ns_reclen = 0;
2264 } else if (slp->ns_cc > slp->ns_reclen) {
2265 len = 0;
2266 m = slp->ns_raw;
2267 om = (struct mbuf *)0;
2268
2269 while (len < slp->ns_reclen) {
2270 if ((len + m->m_len) > slp->ns_reclen) {
2271 m2 = m_copym(m, 0, slp->ns_reclen - len,
2272 waitflag);
2273 if (m2) {
2274 if (om) {
2275 om->m_next = m2;
2276 recm = slp->ns_raw;
2277 } else
2278 recm = m2;
2279 m->m_data += slp->ns_reclen - len;
2280 m->m_len -= slp->ns_reclen - len;
2281 len = slp->ns_reclen;
2282 } else {
2283 slp->ns_flag &= ~SLP_GETSTREAM;
2284 return (EWOULDBLOCK);
2285 }
2286 } else if ((len + m->m_len) == slp->ns_reclen) {
2287 om = m;
2288 len += m->m_len;
2289 m = m->m_next;
2290 recm = slp->ns_raw;
2291 om->m_next = (struct mbuf *)0;
2292 } else {
2293 om = m;
2294 len += m->m_len;
2295 m = m->m_next;
2296 }
2297 }
2298 slp->ns_raw = m;
2299 slp->ns_cc -= len;
2300 slp->ns_reclen = 0;
2301 } else {
2302 slp->ns_flag &= ~SLP_GETSTREAM;
2303 return (0);
2304 }
2305
2306 /*
2307 * Accumulate the fragments into a record.
2308 */
2309 mpp = &slp->ns_frag;
2310 while (*mpp)
2311 mpp = &((*mpp)->m_next);
2312 *mpp = recm;
2313 if (slp->ns_flag & SLP_LASTFRAG) {
2314 struct nfsrv_rec *rec;
da004c97
MD
2315 int mf = (waitflag & M_DONTWAIT) ? M_NOWAIT : M_WAITOK;
2316 rec = malloc(sizeof(struct nfsrv_rec), M_NFSRVDESC, mf);
984263bc
MD
2317 if (!rec) {
2318 m_freem(slp->ns_frag);
2319 } else {
2320 nfs_realign(&slp->ns_frag, 10 * NFSX_UNSIGNED);
2321 rec->nr_address = (struct sockaddr *)0;
2322 rec->nr_packet = slp->ns_frag;
2323 STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
2324 }
2325 slp->ns_frag = (struct mbuf *)0;
2326 }
2327 }
2328}
2329
2330/*
2331 * Parse an RPC header.
2332 */
2333int
2334nfsrv_dorec(slp, nfsd, ndp)
40393ded 2335 struct nfssvc_sock *slp;
984263bc
MD
2336 struct nfsd *nfsd;
2337 struct nfsrv_descript **ndp;
2338{
2339 struct nfsrv_rec *rec;
40393ded 2340 struct mbuf *m;
984263bc 2341 struct sockaddr *nam;
40393ded 2342 struct nfsrv_descript *nd;
984263bc
MD
2343 int error;
2344
2345 *ndp = NULL;
2346 if ((slp->ns_flag & SLP_VALID) == 0 || !STAILQ_FIRST(&slp->ns_rec))
2347 return (ENOBUFS);
2348 rec = STAILQ_FIRST(&slp->ns_rec);
2349 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
2350 nam = rec->nr_address;
2351 m = rec->nr_packet;
2352 free(rec, M_NFSRVDESC);
2353 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2354 M_NFSRVDESC, M_WAITOK);
2355 nd->nd_md = nd->nd_mrep = m;
2356 nd->nd_nam2 = nam;
2357 nd->nd_dpos = mtod(m, caddr_t);
2358 error = nfs_getreq(nd, nfsd, TRUE);
2359 if (error) {
2360 if (nam) {
2361 FREE(nam, M_SONAME);
2362 }
2363 free((caddr_t)nd, M_NFSRVDESC);
2364 return (error);
2365 }
2366 *ndp = nd;
2367 nfsd->nfsd_nd = nd;
2368 return (0);
2369}
2370
2371/*
2372 * Search for a sleeping nfsd and wake it up.
2373 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2374 * running nfsds will go look for the work in the nfssvc_sock list.
2375 */
2376void
2377nfsrv_wakenfsd(slp)
2378 struct nfssvc_sock *slp;
2379{
40393ded 2380 struct nfsd *nd;
984263bc
MD
2381
2382 if ((slp->ns_flag & SLP_VALID) == 0)
2383 return;
2384 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) {
2385 if (nd->nfsd_flag & NFSD_WAITING) {
2386 nd->nfsd_flag &= ~NFSD_WAITING;
2387 if (nd->nfsd_slp)
2388 panic("nfsd wakeup");
2389 slp->ns_sref++;
2390 nd->nfsd_slp = slp;
2391 wakeup((caddr_t)nd);
2392 return;
2393 }
2394 }
2395 slp->ns_flag |= SLP_DOREC;
2396 nfsd_head_flag |= NFSD_CHECKSLP;
2397}
2398#endif /* NFS_NOSERVER */