Commit | Line | Data |
---|---|---|
b688fc87 WJ |
1 | /* |
2 | * Copyright (c) 1989, 1991 The Regents of the University of California. | |
3 | * 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 7.23 (Berkeley) 4/20/91 | |
37 | */ | |
38 | ||
39 | /* | |
40 | * Socket operations for use by nfs | |
41 | */ | |
42 | ||
43 | #include "param.h" | |
44 | #include "proc.h" | |
45 | #include "mount.h" | |
46 | #include "kernel.h" | |
47 | #include "malloc.h" | |
48 | #include "mbuf.h" | |
49 | #include "namei.h" | |
50 | #include "vnode.h" | |
51 | #include "domain.h" | |
52 | #include "protosw.h" | |
53 | #include "socket.h" | |
54 | #include "socketvar.h" | |
55 | #include "syslog.h" | |
56 | #include "tprintf.h" | |
57 | #include "../netinet/in.h" | |
58 | #include "../netinet/tcp.h" | |
59 | ||
60 | #include "rpcv2.h" | |
61 | #include "nfsv2.h" | |
62 | #include "nfs.h" | |
63 | #include "xdr_subs.h" | |
64 | #include "nfsm_subs.h" | |
65 | #include "nfsmount.h" | |
66 | ||
67 | #define TRUE 1 | |
68 | #define FALSE 0 | |
69 | ||
70 | /* | |
71 | * External data, mostly RPC constants in XDR form | |
72 | */ | |
73 | extern u_long rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, rpc_auth_unix, | |
74 | rpc_msgaccepted, rpc_call; | |
75 | extern u_long nfs_prog, nfs_vers; | |
76 | /* Maybe these should be bits in a u_long ?? */ | |
77 | extern int nonidempotent[NFS_NPROCS]; | |
78 | static int compressrequest[NFS_NPROCS] = { | |
79 | FALSE, | |
80 | TRUE, | |
81 | TRUE, | |
82 | FALSE, | |
83 | TRUE, | |
84 | TRUE, | |
85 | TRUE, | |
86 | FALSE, | |
87 | FALSE, | |
88 | TRUE, | |
89 | TRUE, | |
90 | TRUE, | |
91 | TRUE, | |
92 | TRUE, | |
93 | TRUE, | |
94 | TRUE, | |
95 | TRUE, | |
96 | TRUE, | |
97 | }; | |
98 | int nfs_sbwait(); | |
99 | void nfs_disconnect(); | |
100 | struct mbuf *nfs_compress(), *nfs_uncompress(); | |
101 | ||
102 | int nfsrv_null(), | |
103 | nfsrv_getattr(), | |
104 | nfsrv_setattr(), | |
105 | nfsrv_lookup(), | |
106 | nfsrv_readlink(), | |
107 | nfsrv_read(), | |
108 | nfsrv_write(), | |
109 | nfsrv_create(), | |
110 | nfsrv_remove(), | |
111 | nfsrv_rename(), | |
112 | nfsrv_link(), | |
113 | nfsrv_symlink(), | |
114 | nfsrv_mkdir(), | |
115 | nfsrv_rmdir(), | |
116 | nfsrv_readdir(), | |
117 | nfsrv_statfs(), | |
118 | nfsrv_noop(); | |
119 | ||
120 | int (*nfsrv_procs[NFS_NPROCS])() = { | |
121 | nfsrv_null, | |
122 | nfsrv_getattr, | |
123 | nfsrv_setattr, | |
124 | nfsrv_noop, | |
125 | nfsrv_lookup, | |
126 | nfsrv_readlink, | |
127 | nfsrv_read, | |
128 | nfsrv_noop, | |
129 | nfsrv_write, | |
130 | nfsrv_create, | |
131 | nfsrv_remove, | |
132 | nfsrv_rename, | |
133 | nfsrv_link, | |
134 | nfsrv_symlink, | |
135 | nfsrv_mkdir, | |
136 | nfsrv_rmdir, | |
137 | nfsrv_readdir, | |
138 | nfsrv_statfs, | |
139 | }; | |
140 | ||
141 | struct nfsreq nfsreqh; | |
142 | int nfsrexmtthresh = NFS_FISHY; | |
143 | int nfs_tcpnodelay = 1; | |
144 | ||
145 | /* | |
146 | * Initialize sockets and congestion for a new NFS connection. | |
147 | * We do not free the sockaddr if error. | |
148 | */ | |
149 | nfs_connect(nmp) | |
150 | register struct nfsmount *nmp; | |
151 | { | |
152 | register struct socket *so; | |
153 | int s, error, bufsize; | |
154 | struct mbuf *m; | |
155 | ||
156 | nmp->nm_so = (struct socket *)0; | |
157 | if (error = socreate(mtod(nmp->nm_nam, struct sockaddr *)->sa_family, | |
158 | &nmp->nm_so, nmp->nm_sotype, nmp->nm_soproto)) | |
159 | goto bad; | |
160 | so = nmp->nm_so; | |
161 | nmp->nm_soflags = so->so_proto->pr_flags; | |
162 | ||
163 | if (nmp->nm_sotype == SOCK_DGRAM) | |
164 | bufsize = min(4 * (nmp->nm_wsize + NFS_MAXPKTHDR), | |
165 | NFS_MAXPACKET); | |
166 | else | |
167 | bufsize = min(4 * (nmp->nm_wsize + NFS_MAXPKTHDR + sizeof(u_long)), | |
168 | NFS_MAXPACKET + sizeof(u_long)); | |
169 | if (error = soreserve(so, bufsize, bufsize)) | |
170 | goto bad; | |
171 | ||
172 | /* | |
173 | * Protocols that do not require connections may be optionally left | |
174 | * unconnected for servers that reply from a port other than NFS_PORT. | |
175 | */ | |
176 | if (nmp->nm_flag & NFSMNT_NOCONN) { | |
177 | if (nmp->nm_soflags & PR_CONNREQUIRED) { | |
178 | error = ENOTCONN; | |
179 | goto bad; | |
180 | } | |
181 | } else { | |
182 | if (error = soconnect(so, nmp->nm_nam)) | |
183 | goto bad; | |
184 | ||
185 | /* | |
186 | * Wait for the connection to complete. Cribbed from the | |
187 | * connect system call but with the wait at negative prio. | |
188 | */ | |
189 | s = splnet(); | |
190 | while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) | |
191 | (void) tsleep((caddr_t)&so->so_timeo, PSOCK, "nfscon", 0); | |
192 | splx(s); | |
193 | if (so->so_error) { | |
194 | error = so->so_error; | |
195 | goto bad; | |
196 | } | |
197 | } | |
198 | if (nmp->nm_sotype == SOCK_DGRAM) { | |
199 | if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_SPONGY | NFSMNT_INT)) { | |
200 | so->so_rcv.sb_timeo = (5 * hz); | |
201 | so->so_snd.sb_timeo = (5 * hz); | |
202 | } else { | |
203 | so->so_rcv.sb_timeo = 0; | |
204 | so->so_snd.sb_timeo = 0; | |
205 | } | |
206 | nmp->nm_rto = NFS_TIMEO; | |
207 | } else { | |
208 | if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_SPONGY | NFSMNT_INT)) { | |
209 | so->so_rcv.sb_timeo = (5 * hz); | |
210 | so->so_snd.sb_timeo = (5 * hz); | |
211 | } else { | |
212 | so->so_rcv.sb_timeo = 0; | |
213 | so->so_snd.sb_timeo = 0; | |
214 | } | |
215 | if (so->so_proto->pr_flags & PR_CONNREQUIRED) { | |
216 | MGET(m, M_WAIT, MT_SOOPTS); | |
217 | *mtod(m, int *) = 1; | |
218 | m->m_len = sizeof(int); | |
219 | sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); | |
220 | } | |
221 | if (so->so_proto->pr_domain->dom_family == AF_INET && | |
222 | so->so_proto->pr_protocol == IPPROTO_TCP && | |
223 | nfs_tcpnodelay) { | |
224 | MGET(m, M_WAIT, MT_SOOPTS); | |
225 | *mtod(m, int *) = 1; | |
226 | m->m_len = sizeof(int); | |
227 | sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); | |
228 | } | |
229 | nmp->nm_rto = 10 * NFS_TIMEO; /* XXX */ | |
230 | } | |
231 | so->so_rcv.sb_flags |= SB_NOINTR; | |
232 | so->so_snd.sb_flags |= SB_NOINTR; | |
233 | ||
234 | /* Initialize other non-zero congestion variables */ | |
235 | nmp->nm_window = 2; /* Initial send window */ | |
236 | nmp->nm_ssthresh = NFS_MAXWINDOW; /* Slowstart threshold */ | |
237 | nmp->nm_rttvar = nmp->nm_rto << 1; | |
238 | nmp->nm_sent = 0; | |
239 | nmp->nm_currexmit = 0; | |
240 | return (0); | |
241 | ||
242 | bad: | |
243 | nfs_disconnect(nmp); | |
244 | return (error); | |
245 | } | |
246 | ||
247 | /* | |
248 | * Reconnect routine: | |
249 | * Called when a connection is broken on a reliable protocol. | |
250 | * - clean up the old socket | |
251 | * - nfs_connect() again | |
252 | * - set R_MUSTRESEND for all outstanding requests on mount point | |
253 | * If this fails the mount point is DEAD! | |
254 | * nb: Must be called with the nfs_solock() set on the mount point. | |
255 | */ | |
256 | nfs_reconnect(rep, nmp) | |
257 | register struct nfsreq *rep; | |
258 | register struct nfsmount *nmp; | |
259 | { | |
260 | register struct nfsreq *rp; | |
261 | int error; | |
262 | ||
263 | nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, | |
264 | "trying reconnect"); | |
265 | while (error = nfs_connect(nmp)) { | |
266 | #ifdef lint | |
267 | error = error; | |
268 | #endif /* lint */ | |
269 | if ((nmp->nm_flag & NFSMNT_INT) && nfs_sigintr(rep->r_procp)) | |
270 | return (EINTR); | |
271 | (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); | |
272 | } | |
273 | nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, | |
274 | "reconnected"); | |
275 | ||
276 | /* | |
277 | * Loop through outstanding request list and fix up all requests | |
278 | * on old socket. | |
279 | */ | |
280 | rp = nfsreqh.r_next; | |
281 | while (rp != &nfsreqh) { | |
282 | if (rp->r_nmp == nmp) | |
283 | rp->r_flags |= R_MUSTRESEND; | |
284 | rp = rp->r_next; | |
285 | } | |
286 | return (0); | |
287 | } | |
288 | ||
289 | /* | |
290 | * NFS disconnect. Clean up and unlink. | |
291 | */ | |
292 | void | |
293 | nfs_disconnect(nmp) | |
294 | register struct nfsmount *nmp; | |
295 | { | |
296 | register struct socket *so; | |
297 | ||
298 | if (nmp->nm_so) { | |
299 | so = nmp->nm_so; | |
300 | nmp->nm_so = (struct socket *)0; | |
301 | soshutdown(so, 2); | |
302 | soclose(so); | |
303 | } | |
304 | } | |
305 | ||
306 | /* | |
307 | * This is the nfs send routine. For connection based socket types, it | |
308 | * must be called with an nfs_solock() on the socket. | |
309 | * "rep == NULL" indicates that it has been called from a server. | |
310 | */ | |
311 | nfs_send(so, nam, top, rep) | |
312 | register struct socket *so; | |
313 | struct mbuf *nam; | |
314 | register struct mbuf *top; | |
315 | struct nfsreq *rep; | |
316 | { | |
317 | struct mbuf *sendnam; | |
318 | int error, soflags; | |
319 | ||
320 | if (rep) { | |
321 | if (rep->r_flags & R_SOFTTERM) { | |
322 | m_freem(top); | |
323 | return (EINTR); | |
324 | } | |
325 | if (rep->r_nmp->nm_so == NULL && | |
326 | (error = nfs_reconnect(rep, rep->r_nmp))) | |
327 | return (error); | |
328 | rep->r_flags &= ~R_MUSTRESEND; | |
329 | so = rep->r_nmp->nm_so; | |
330 | soflags = rep->r_nmp->nm_soflags; | |
331 | } else | |
332 | soflags = so->so_proto->pr_flags; | |
333 | if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) | |
334 | sendnam = (struct mbuf *)0; | |
335 | else | |
336 | sendnam = nam; | |
337 | ||
338 | error = sosend(so, sendnam, (struct uio *)0, top, | |
339 | (struct mbuf *)0, 0); | |
340 | if (error == EWOULDBLOCK && rep) { | |
341 | if (rep->r_flags & R_SOFTTERM) | |
342 | error = EINTR; | |
343 | else { | |
344 | rep->r_flags |= R_MUSTRESEND; | |
345 | error = 0; | |
346 | } | |
347 | } | |
348 | /* | |
349 | * Ignore socket errors?? | |
350 | */ | |
351 | if (error && error != EINTR && error != ERESTART) | |
352 | error = 0; | |
353 | return (error); | |
354 | } | |
355 | ||
356 | /* | |
357 | * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all | |
358 | * done by soreceive(), but for SOCK_STREAM we must deal with the Record | |
359 | * Mark and consolidate the data into a new mbuf list. | |
360 | * nb: Sometimes TCP passes the data up to soreceive() in long lists of | |
361 | * small mbufs. | |
362 | * For SOCK_STREAM we must be very careful to read an entire record once | |
363 | * we have read any of it, even if the system call has been interrupted. | |
364 | */ | |
365 | nfs_receive(so, aname, mp, rep) | |
366 | register struct socket *so; | |
367 | struct mbuf **aname; | |
368 | struct mbuf **mp; | |
369 | register struct nfsreq *rep; | |
370 | { | |
371 | struct uio auio; | |
372 | struct iovec aio; | |
373 | register struct mbuf *m; | |
374 | struct mbuf *m2, *mnew, **mbp; | |
375 | caddr_t fcp, tcp; | |
376 | u_long len; | |
377 | struct mbuf **getnam; | |
378 | int error, siz, mlen, soflags, rcvflg; | |
379 | ||
380 | /* | |
381 | * Set up arguments for soreceive() | |
382 | */ | |
383 | *mp = (struct mbuf *)0; | |
384 | *aname = (struct mbuf *)0; | |
385 | if (rep) | |
386 | soflags = rep->r_nmp->nm_soflags; | |
387 | else | |
388 | soflags = so->so_proto->pr_flags; | |
389 | ||
390 | /* | |
391 | * For reliable protocols, lock against other senders/receivers | |
392 | * in case a reconnect is necessary. | |
393 | * For SOCK_STREAM, first get the Record Mark to find out how much | |
394 | * more there is to get. | |
395 | * We must lock the socket against other receivers | |
396 | * until we have an entire rpc request/reply. | |
397 | */ | |
398 | if (soflags & PR_CONNREQUIRED) { | |
399 | tryagain: | |
400 | /* | |
401 | * Check for fatal errors and resending request. | |
402 | */ | |
403 | if (rep) { | |
404 | /* | |
405 | * Ugh: If a reconnect attempt just happened, nm_so | |
406 | * would have changed. NULL indicates a failed | |
407 | * attempt that has essentially shut down this | |
408 | * mount point. | |
409 | */ | |
410 | if (rep->r_mrep || (so = rep->r_nmp->nm_so) == NULL || | |
411 | (rep->r_flags & R_SOFTTERM)) | |
412 | return (EINTR); | |
413 | while (rep->r_flags & R_MUSTRESEND) { | |
414 | m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); | |
415 | nfsstats.rpcretries++; | |
416 | if (error = nfs_send(so, rep->r_nmp->nm_nam, m, | |
417 | rep)) | |
418 | goto errout; | |
419 | } | |
420 | } | |
421 | if ((soflags & PR_ATOMIC) == 0) { | |
422 | aio.iov_base = (caddr_t) &len; | |
423 | aio.iov_len = sizeof(u_long); | |
424 | auio.uio_iov = &aio; | |
425 | auio.uio_iovcnt = 1; | |
426 | auio.uio_segflg = UIO_SYSSPACE; | |
427 | auio.uio_rw = UIO_READ; | |
428 | auio.uio_procp = (struct proc *)0; | |
429 | auio.uio_offset = 0; | |
430 | auio.uio_resid = sizeof(u_long); | |
431 | do { | |
432 | rcvflg = MSG_WAITALL; | |
433 | error = soreceive(so, (struct mbuf **)0, &auio, | |
434 | (struct mbuf **)0, (struct mbuf **)0, &rcvflg); | |
435 | if (error == EWOULDBLOCK && rep) { | |
436 | if (rep->r_flags & R_SOFTTERM) | |
437 | return (EINTR); | |
438 | if (rep->r_flags & R_MUSTRESEND) | |
439 | goto tryagain; | |
440 | } | |
441 | } while (error == EWOULDBLOCK); | |
442 | if (!error && auio.uio_resid > 0) { | |
443 | if (rep) | |
444 | log(LOG_INFO, | |
445 | "short receive (%d/%d) from nfs server %s\n", | |
446 | sizeof(u_long) - auio.uio_resid, | |
447 | sizeof(u_long), | |
448 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); | |
449 | error = EPIPE; | |
450 | } | |
451 | if (error) | |
452 | goto errout; | |
453 | len = ntohl(len) & ~0x80000000; | |
454 | /* | |
455 | * This is SERIOUS! We are out of sync with the sender | |
456 | * and forcing a disconnect/reconnect is all I can do. | |
457 | */ | |
458 | if (len > NFS_MAXPACKET) { | |
459 | if (rep) | |
460 | log(LOG_ERR, "%s (%d) from nfs server %s\n", | |
461 | "impossible packet length", | |
462 | len, | |
463 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); | |
464 | error = EFBIG; | |
465 | goto errout; | |
466 | } | |
467 | auio.uio_resid = len; | |
468 | do { | |
469 | rcvflg = MSG_WAITALL; | |
470 | error = soreceive(so, (struct mbuf **)0, | |
471 | &auio, mp, (struct mbuf **)0, &rcvflg); | |
472 | } while (error == EWOULDBLOCK || error == EINTR || | |
473 | error == ERESTART); | |
474 | if (!error && auio.uio_resid > 0) { | |
475 | if (rep) | |
476 | log(LOG_INFO, | |
477 | "short receive (%d/%d) from nfs server %s\n", | |
478 | len - auio.uio_resid, len, | |
479 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); | |
480 | error = EPIPE; | |
481 | } | |
482 | } else { | |
483 | auio.uio_resid = len = 1000000; /* Anything Big */ | |
484 | do { | |
485 | rcvflg = 0; | |
486 | error = soreceive(so, (struct mbuf **)0, | |
487 | &auio, mp, (struct mbuf **)0, &rcvflg); | |
488 | if (error == EWOULDBLOCK && rep) { | |
489 | if (rep->r_flags & R_SOFTTERM) | |
490 | return (EINTR); | |
491 | if (rep->r_flags & R_MUSTRESEND) | |
492 | goto tryagain; | |
493 | } | |
494 | } while (error == EWOULDBLOCK); | |
495 | if (!error && *mp == NULL) | |
496 | error = EPIPE; | |
497 | len -= auio.uio_resid; | |
498 | } | |
499 | errout: | |
500 | if (error && rep && error != EINTR && error != ERESTART) { | |
501 | m_freem(*mp); | |
502 | *mp = (struct mbuf *)0; | |
503 | if (error != EPIPE && rep) | |
504 | log(LOG_INFO, | |
505 | "receive error %d from nfs server %s\n", | |
506 | error, | |
507 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); | |
508 | nfs_disconnect(rep->r_nmp); | |
509 | error = nfs_reconnect(rep, rep->r_nmp); | |
510 | if (!error) | |
511 | goto tryagain; | |
512 | } | |
513 | } else { | |
514 | if (so->so_state & SS_ISCONNECTED) | |
515 | getnam = (struct mbuf **)0; | |
516 | else | |
517 | getnam = aname; | |
518 | auio.uio_resid = len = 1000000; | |
519 | do { | |
520 | rcvflg = 0; | |
521 | error = soreceive(so, getnam, &auio, mp, | |
522 | (struct mbuf **)0, &rcvflg); | |
523 | if (error == EWOULDBLOCK && rep && | |
524 | (rep->r_flags & R_SOFTTERM)) | |
525 | return (EINTR); | |
526 | } while (error == EWOULDBLOCK); | |
527 | len -= auio.uio_resid; | |
528 | } | |
529 | if (error) { | |
530 | m_freem(*mp); | |
531 | *mp = (struct mbuf *)0; | |
532 | } | |
533 | /* | |
534 | * Search for any mbufs that are not a multiple of 4 bytes long. | |
535 | * These could cause pointer alignment problems, so copy them to | |
536 | * well aligned mbufs. | |
537 | */ | |
538 | m = *mp; | |
539 | mbp = mp; | |
540 | while (m) { | |
541 | /* | |
542 | * All this for something that may never happen. | |
543 | */ | |
544 | if (m->m_next && (m->m_len & 0x3)) { | |
545 | printf("nfs_rcv odd length!\n"); | |
546 | mlen = 0; | |
547 | while (m) { | |
548 | fcp = mtod(m, caddr_t); | |
549 | while (m->m_len > 0) { | |
550 | if (mlen == 0) { | |
551 | MGET(m2, M_WAIT, MT_DATA); | |
552 | if (len >= MINCLSIZE) | |
553 | MCLGET(m2, M_WAIT); | |
554 | m2->m_len = 0; | |
555 | mlen = M_TRAILINGSPACE(m2); | |
556 | tcp = mtod(m2, caddr_t); | |
557 | *mbp = m2; | |
558 | mbp = &m2->m_next; | |
559 | } | |
560 | siz = MIN(mlen, m->m_len); | |
561 | bcopy(fcp, tcp, siz); | |
562 | m2->m_len += siz; | |
563 | mlen -= siz; | |
564 | len -= siz; | |
565 | tcp += siz; | |
566 | m->m_len -= siz; | |
567 | fcp += siz; | |
568 | } | |
569 | MFREE(m, mnew); | |
570 | m = mnew; | |
571 | } | |
572 | break; | |
573 | } | |
574 | len -= m->m_len; | |
575 | mbp = &m->m_next; | |
576 | m = m->m_next; | |
577 | } | |
578 | return (error); | |
579 | } | |
580 | ||
581 | /* | |
582 | * Implement receipt of reply on a socket. | |
583 | * We must search through the list of received datagrams matching them | |
584 | * with outstanding requests using the xid, until ours is found. | |
585 | */ | |
586 | /* ARGSUSED */ | |
587 | nfs_reply(nmp, myrep) | |
588 | struct nfsmount *nmp; | |
589 | struct nfsreq *myrep; | |
590 | { | |
591 | register struct mbuf *m; | |
592 | register struct nfsreq *rep; | |
593 | register int error = 0; | |
594 | u_long rxid; | |
595 | struct mbuf *mp, *nam; | |
596 | char *cp; | |
597 | int cnt, xfer; | |
598 | ||
599 | /* | |
600 | * Loop around until we get our own reply | |
601 | */ | |
602 | for (;;) { | |
603 | /* | |
604 | * Lock against other receivers so that I don't get stuck in | |
605 | * sbwait() after someone else has received my reply for me. | |
606 | * Also necessary for connection based protocols to avoid | |
607 | * race conditions during a reconnect. | |
608 | */ | |
609 | nfs_solock(&nmp->nm_flag); | |
610 | /* Already received, bye bye */ | |
611 | if (myrep->r_mrep != NULL) { | |
612 | nfs_sounlock(&nmp->nm_flag); | |
613 | return (0); | |
614 | } | |
615 | /* | |
616 | * Get the next Rpc reply off the socket | |
617 | */ | |
618 | if (error = nfs_receive(nmp->nm_so, &nam, &mp, myrep)) { | |
619 | nfs_sounlock(&nmp->nm_flag); | |
620 | ||
621 | /* | |
622 | * Ignore routing errors on connectionless protocols?? | |
623 | */ | |
624 | if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { | |
625 | nmp->nm_so->so_error = 0; | |
626 | continue; | |
627 | } | |
628 | ||
629 | /* | |
630 | * Otherwise cleanup and return a fatal error. | |
631 | */ | |
632 | if (myrep->r_flags & R_TIMING) { | |
633 | myrep->r_flags &= ~R_TIMING; | |
634 | nmp->nm_rtt = -1; | |
635 | } | |
636 | if (myrep->r_flags & R_SENT) { | |
637 | myrep->r_flags &= ~R_SENT; | |
638 | nmp->nm_sent--; | |
639 | } | |
640 | return (error); | |
641 | } | |
642 | ||
643 | /* | |
644 | * Get the xid and check that it is an rpc reply | |
645 | */ | |
646 | m = mp; | |
647 | while (m && m->m_len == 0) | |
648 | m = m->m_next; | |
649 | if (m == NULL) { | |
650 | nfsstats.rpcinvalid++; | |
651 | m_freem(mp); | |
652 | nfs_sounlock(&nmp->nm_flag); | |
653 | continue; | |
654 | } | |
655 | bcopy(mtod(m, caddr_t), (caddr_t)&rxid, NFSX_UNSIGNED); | |
656 | /* | |
657 | * Loop through the request list to match up the reply | |
658 | * Iff no match, just drop the datagram | |
659 | */ | |
660 | m = mp; | |
661 | rep = nfsreqh.r_next; | |
662 | while (rep != &nfsreqh) { | |
663 | if (rep->r_mrep == NULL && rxid == rep->r_xid) { | |
664 | /* Found it.. */ | |
665 | rep->r_mrep = m; | |
666 | /* | |
667 | * Update timing | |
668 | */ | |
669 | if (rep->r_flags & R_TIMING) { | |
670 | nfs_updatetimer(rep->r_nmp); | |
671 | rep->r_flags &= ~R_TIMING; | |
672 | rep->r_nmp->nm_rtt = -1; | |
673 | } | |
674 | if (rep->r_flags & R_SENT) { | |
675 | rep->r_flags &= ~R_SENT; | |
676 | rep->r_nmp->nm_sent--; | |
677 | } | |
678 | break; | |
679 | } | |
680 | rep = rep->r_next; | |
681 | } | |
682 | nfs_sounlock(&nmp->nm_flag); | |
683 | if (nam) | |
684 | m_freem(nam); | |
685 | /* | |
686 | * If not matched to a request, drop it. | |
687 | * If it's mine, get out. | |
688 | */ | |
689 | if (rep == &nfsreqh) { | |
690 | nfsstats.rpcunexpected++; | |
691 | m_freem(m); | |
692 | } else if (rep == myrep) | |
693 | return (0); | |
694 | } | |
695 | } | |
696 | ||
697 | /* | |
698 | * nfs_request - goes something like this | |
699 | * - fill in request struct | |
700 | * - links it into list | |
701 | * - calls nfs_send() for first transmit | |
702 | * - calls nfs_receive() to get reply | |
703 | * - break down rpc header and return with nfs reply pointed to | |
704 | * by mrep or error | |
705 | * nb: always frees up mreq mbuf list | |
706 | */ | |
707 | nfs_request(vp, mreq, xid, procnum, procp, tryhard, mp, mrp, mdp, dposp) | |
708 | struct vnode *vp; | |
709 | struct mbuf *mreq; | |
710 | u_long xid; | |
711 | int procnum; | |
712 | struct proc *procp; | |
713 | int tryhard; | |
714 | struct mount *mp; | |
715 | struct mbuf **mrp; | |
716 | struct mbuf **mdp; | |
717 | caddr_t *dposp; | |
718 | { | |
719 | register struct mbuf *m, *mrep; | |
720 | register struct nfsreq *rep; | |
721 | register u_long *tl; | |
722 | register int len; | |
723 | struct nfsmount *nmp; | |
724 | struct mbuf *md; | |
725 | struct nfsreq *reph; | |
726 | caddr_t dpos; | |
727 | char *cp2; | |
728 | int t1; | |
729 | int s, compressed; | |
730 | int error = 0; | |
731 | ||
732 | nmp = VFSTONFS(mp); | |
733 | m = mreq; | |
734 | MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); | |
735 | rep->r_xid = xid; | |
736 | rep->r_nmp = nmp; | |
737 | rep->r_vp = vp; | |
738 | rep->r_procp = procp; | |
739 | if ((nmp->nm_flag & NFSMNT_SOFT) || | |
740 | ((nmp->nm_flag & NFSMNT_SPONGY) && !tryhard)) | |
741 | rep->r_retry = nmp->nm_retry; | |
742 | else | |
743 | rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ | |
744 | rep->r_flags = rep->r_rexmit = 0; | |
745 | /* | |
746 | * Three cases: | |
747 | * - non-idempotent requests on SOCK_DGRAM use NFS_MINIDEMTIMEO | |
748 | * - idempotent requests on SOCK_DGRAM use 0 | |
749 | * - Reliable transports, NFS_RELIABLETIMEO | |
750 | * Timeouts are still done on reliable transports to ensure detection | |
751 | * of excessive connection delay. | |
752 | */ | |
753 | if (nmp->nm_sotype != SOCK_DGRAM) | |
754 | rep->r_timerinit = -NFS_RELIABLETIMEO; | |
755 | else if (nonidempotent[procnum]) | |
756 | rep->r_timerinit = -NFS_MINIDEMTIMEO; | |
757 | else | |
758 | rep->r_timerinit = 0; | |
759 | rep->r_timer = rep->r_timerinit; | |
760 | rep->r_mrep = NULL; | |
761 | len = 0; | |
762 | while (m) { | |
763 | len += m->m_len; | |
764 | m = m->m_next; | |
765 | } | |
766 | mreq->m_pkthdr.len = len; | |
767 | mreq->m_pkthdr.rcvif = (struct ifnet *)0; | |
768 | compressed = 0; | |
769 | m = mreq; | |
770 | if ((nmp->nm_flag & NFSMNT_COMPRESS) && compressrequest[procnum]) { | |
771 | mreq = nfs_compress(mreq); | |
772 | if (mreq != m) { | |
773 | len = mreq->m_pkthdr.len; | |
774 | compressed++; | |
775 | } | |
776 | } | |
777 | /* | |
778 | * For non-atomic protocols, insert a Sun RPC Record Mark. | |
779 | */ | |
780 | if ((nmp->nm_soflags & PR_ATOMIC) == 0) { | |
781 | M_PREPEND(mreq, sizeof(u_long), M_WAIT); | |
782 | *mtod(mreq, u_long *) = htonl(0x80000000 | len); | |
783 | } | |
784 | rep->r_mreq = mreq; | |
785 | ||
786 | /* | |
787 | * Do the client side RPC. | |
788 | */ | |
789 | nfsstats.rpcrequests++; | |
790 | /* | |
791 | * Chain request into list of outstanding requests. Be sure | |
792 | * to put it LAST so timer finds oldest requests first. | |
793 | */ | |
794 | s = splnet(); | |
795 | reph = &nfsreqh; | |
796 | reph->r_prev->r_next = rep; | |
797 | rep->r_prev = reph->r_prev; | |
798 | reph->r_prev = rep; | |
799 | rep->r_next = reph; | |
800 | /* | |
801 | * If backing off another request or avoiding congestion, don't | |
802 | * send this one now but let timer do it. If not timing a request, | |
803 | * do it now. | |
804 | */ | |
805 | if (nmp->nm_sent <= 0 || nmp->nm_sotype != SOCK_DGRAM || | |
806 | (nmp->nm_currexmit == 0 && nmp->nm_sent < nmp->nm_window)) { | |
807 | nmp->nm_sent++; | |
808 | rep->r_flags |= R_SENT; | |
809 | if (nmp->nm_rtt == -1) { | |
810 | nmp->nm_rtt = 0; | |
811 | rep->r_flags |= R_TIMING; | |
812 | } | |
813 | splx(s); | |
814 | m = m_copym(mreq, 0, M_COPYALL, M_WAIT); | |
815 | if (nmp->nm_soflags & PR_CONNREQUIRED) | |
816 | nfs_solock(&nmp->nm_flag); | |
817 | error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); | |
818 | if (nmp->nm_soflags & PR_CONNREQUIRED) | |
819 | nfs_sounlock(&nmp->nm_flag); | |
820 | if (error && NFSIGNORE_SOERROR(nmp->nm_soflags, error)) | |
821 | nmp->nm_so->so_error = error = 0; | |
822 | } else | |
823 | splx(s); | |
824 | ||
825 | /* | |
826 | * Wait for the reply from our send or the timer's. | |
827 | */ | |
828 | if (!error) | |
829 | error = nfs_reply(nmp, rep); | |
830 | ||
831 | /* | |
832 | * RPC done, unlink the request. | |
833 | */ | |
834 | s = splnet(); | |
835 | rep->r_prev->r_next = rep->r_next; | |
836 | rep->r_next->r_prev = rep->r_prev; | |
837 | splx(s); | |
838 | ||
839 | /* | |
840 | * If there was a successful reply and a tprintf msg. | |
841 | * tprintf a response. | |
842 | */ | |
843 | if (!error && (rep->r_flags & R_TPRINTFMSG)) | |
844 | nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, | |
845 | "is alive again"); | |
846 | m_freem(rep->r_mreq); | |
847 | mrep = rep->r_mrep; | |
848 | FREE((caddr_t)rep, M_NFSREQ); | |
849 | if (error) | |
850 | return (error); | |
851 | ||
852 | if (compressed) | |
853 | mrep = nfs_uncompress(mrep); | |
854 | md = mrep; | |
855 | /* | |
856 | * break down the rpc header and check if ok | |
857 | */ | |
858 | dpos = mtod(md, caddr_t); | |
859 | nfsm_disect(tl, u_long *, 5*NFSX_UNSIGNED); | |
860 | tl += 2; | |
861 | if (*tl++ == rpc_msgdenied) { | |
862 | if (*tl == rpc_mismatch) | |
863 | error = EOPNOTSUPP; | |
864 | else | |
865 | error = EACCES; | |
866 | m_freem(mrep); | |
867 | return (error); | |
868 | } | |
869 | /* | |
870 | * skip over the auth_verf, someday we may want to cache auth_short's | |
871 | * for nfs_reqhead(), but for now just dump it | |
872 | */ | |
873 | if (*++tl != 0) { | |
874 | len = nfsm_rndup(fxdr_unsigned(long, *tl)); | |
875 | nfsm_adv(len); | |
876 | } | |
877 | nfsm_disect(tl, u_long *, NFSX_UNSIGNED); | |
878 | /* 0 == ok */ | |
879 | if (*tl == 0) { | |
880 | nfsm_disect(tl, u_long *, NFSX_UNSIGNED); | |
881 | if (*tl != 0) { | |
882 | error = fxdr_unsigned(int, *tl); | |
883 | m_freem(mrep); | |
884 | return (error); | |
885 | } | |
886 | *mrp = mrep; | |
887 | *mdp = md; | |
888 | *dposp = dpos; | |
889 | return (0); | |
890 | } | |
891 | m_freem(mrep); | |
892 | return (EPROTONOSUPPORT); | |
893 | nfsmout: | |
894 | return (error); | |
895 | } | |
896 | ||
897 | /* | |
898 | * Get a request for the server main loop | |
899 | * - receive a request via. nfs_soreceive() | |
900 | * - verify it | |
901 | * - fill in the cred struct. | |
902 | */ | |
903 | nfs_getreq(so, prog, vers, maxproc, nam, mrp, mdp, dposp, retxid, procnum, cr, | |
904 | msk, mtch, wascomp) | |
905 | struct socket *so; | |
906 | u_long prog; | |
907 | u_long vers; | |
908 | int maxproc; | |
909 | struct mbuf **nam; | |
910 | struct mbuf **mrp; | |
911 | struct mbuf **mdp; | |
912 | caddr_t *dposp; | |
913 | u_long *retxid; | |
914 | u_long *procnum; | |
915 | register struct ucred *cr; | |
916 | struct mbuf *msk, *mtch; | |
917 | int *wascomp; | |
918 | { | |
919 | register int i; | |
920 | register u_long *tl; | |
921 | register long t1; | |
922 | caddr_t dpos, cp2; | |
923 | int error = 0; | |
924 | struct mbuf *mrep, *md; | |
925 | int len; | |
926 | ||
927 | if (so->so_proto->pr_flags & PR_CONNREQUIRED) { | |
928 | error = nfs_receive(so, nam, &mrep, (struct nfsreq *)0); | |
929 | } else { | |
930 | mrep = (struct mbuf *)0; | |
931 | do { | |
932 | if (mrep) { | |
933 | m_freem(*nam); | |
934 | m_freem(mrep); | |
935 | } | |
936 | error = nfs_receive(so, nam, &mrep, (struct nfsreq *)0); | |
937 | } while (!error && nfs_badnam(*nam, msk, mtch)); | |
938 | } | |
939 | if (error) | |
940 | return (error); | |
941 | md = mrep; | |
942 | mrep = nfs_uncompress(mrep); | |
943 | if (mrep != md) { | |
944 | *wascomp = 1; | |
945 | md = mrep; | |
946 | } else | |
947 | *wascomp = 0; | |
948 | dpos = mtod(mrep, caddr_t); | |
949 | nfsm_disect(tl, u_long *, 10*NFSX_UNSIGNED); | |
950 | *retxid = *tl++; | |
951 | if (*tl++ != rpc_call) { | |
952 | m_freem(mrep); | |
953 | return (ERPCMISMATCH); | |
954 | } | |
955 | if (*tl++ != rpc_vers) { | |
956 | m_freem(mrep); | |
957 | return (ERPCMISMATCH); | |
958 | } | |
959 | if (*tl++ != prog) { | |
960 | m_freem(mrep); | |
961 | return (EPROGUNAVAIL); | |
962 | } | |
963 | if (*tl++ != vers) { | |
964 | m_freem(mrep); | |
965 | return (EPROGMISMATCH); | |
966 | } | |
967 | *procnum = fxdr_unsigned(u_long, *tl++); | |
968 | if (*procnum == NFSPROC_NULL) { | |
969 | *mrp = mrep; | |
970 | return (0); | |
971 | } | |
972 | if (*procnum > maxproc || *tl++ != rpc_auth_unix) { | |
973 | m_freem(mrep); | |
974 | return (EPROCUNAVAIL); | |
975 | } | |
976 | len = fxdr_unsigned(int, *tl++); | |
977 | if (len < 0 || len > RPCAUTH_MAXSIZ) { | |
978 | m_freem(mrep); | |
979 | return (EBADRPC); | |
980 | } | |
981 | len = fxdr_unsigned(int, *++tl); | |
982 | if (len < 0 || len > NFS_MAXNAMLEN) { | |
983 | m_freem(mrep); | |
984 | return (EBADRPC); | |
985 | } | |
986 | nfsm_adv(nfsm_rndup(len)); | |
987 | nfsm_disect(tl, u_long *, 3*NFSX_UNSIGNED); | |
988 | cr->cr_uid = fxdr_unsigned(uid_t, *tl++); | |
989 | cr->cr_gid = fxdr_unsigned(gid_t, *tl++); | |
990 | len = fxdr_unsigned(int, *tl); | |
991 | if (len < 0 || len > RPCAUTH_UNIXGIDS) { | |
992 | m_freem(mrep); | |
993 | return (EBADRPC); | |
994 | } | |
995 | nfsm_disect(tl, u_long *, (len + 2)*NFSX_UNSIGNED); | |
996 | for (i = 1; i <= len; i++) | |
997 | if (i < NGROUPS) | |
998 | cr->cr_groups[i] = fxdr_unsigned(gid_t, *tl++); | |
999 | else | |
1000 | tl++; | |
1001 | cr->cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1); | |
1002 | /* | |
1003 | * Do we have any use for the verifier. | |
1004 | * According to the "Remote Procedure Call Protocol Spec." it | |
1005 | * should be AUTH_NULL, but some clients make it AUTH_UNIX? | |
1006 | * For now, just skip over it | |
1007 | */ | |
1008 | len = fxdr_unsigned(int, *++tl); | |
1009 | if (len < 0 || len > RPCAUTH_MAXSIZ) { | |
1010 | m_freem(mrep); | |
1011 | return (EBADRPC); | |
1012 | } | |
1013 | if (len > 0) | |
1014 | nfsm_adv(nfsm_rndup(len)); | |
1015 | *mrp = mrep; | |
1016 | *mdp = md; | |
1017 | *dposp = dpos; | |
1018 | return (0); | |
1019 | nfsmout: | |
1020 | return (error); | |
1021 | } | |
1022 | ||
1023 | /* | |
1024 | * Generate the rpc reply header | |
1025 | * siz arg. is used to decide if adding a cluster is worthwhile | |
1026 | */ | |
1027 | nfs_rephead(siz, retxid, err, mrq, mbp, bposp) | |
1028 | int siz; | |
1029 | u_long retxid; | |
1030 | int err; | |
1031 | struct mbuf **mrq; | |
1032 | struct mbuf **mbp; | |
1033 | caddr_t *bposp; | |
1034 | { | |
1035 | register u_long *tl; | |
1036 | register long t1; | |
1037 | caddr_t bpos; | |
1038 | struct mbuf *mreq, *mb, *mb2; | |
1039 | ||
1040 | NFSMGETHDR(mreq); | |
1041 | mb = mreq; | |
1042 | if ((siz+RPC_REPLYSIZ) > MHLEN) | |
1043 | MCLGET(mreq, M_WAIT); | |
1044 | tl = mtod(mreq, u_long *); | |
1045 | mreq->m_len = 6*NFSX_UNSIGNED; | |
1046 | bpos = ((caddr_t)tl)+mreq->m_len; | |
1047 | *tl++ = retxid; | |
1048 | *tl++ = rpc_reply; | |
1049 | if (err == ERPCMISMATCH) { | |
1050 | *tl++ = rpc_msgdenied; | |
1051 | *tl++ = rpc_mismatch; | |
1052 | *tl++ = txdr_unsigned(2); | |
1053 | *tl = txdr_unsigned(2); | |
1054 | } else { | |
1055 | *tl++ = rpc_msgaccepted; | |
1056 | *tl++ = 0; | |
1057 | *tl++ = 0; | |
1058 | switch (err) { | |
1059 | case EPROGUNAVAIL: | |
1060 | *tl = txdr_unsigned(RPC_PROGUNAVAIL); | |
1061 | break; | |
1062 | case EPROGMISMATCH: | |
1063 | *tl = txdr_unsigned(RPC_PROGMISMATCH); | |
1064 | nfsm_build(tl, u_long *, 2*NFSX_UNSIGNED); | |
1065 | *tl++ = txdr_unsigned(2); | |
1066 | *tl = txdr_unsigned(2); /* someday 3 */ | |
1067 | break; | |
1068 | case EPROCUNAVAIL: | |
1069 | *tl = txdr_unsigned(RPC_PROCUNAVAIL); | |
1070 | break; | |
1071 | default: | |
1072 | *tl = 0; | |
1073 | if (err != VNOVAL) { | |
1074 | nfsm_build(tl, u_long *, NFSX_UNSIGNED); | |
1075 | *tl = txdr_unsigned(err); | |
1076 | } | |
1077 | break; | |
1078 | }; | |
1079 | } | |
1080 | *mrq = mreq; | |
1081 | *mbp = mb; | |
1082 | *bposp = bpos; | |
1083 | if (err != 0 && err != VNOVAL) | |
1084 | nfsstats.srvrpc_errs++; | |
1085 | return (0); | |
1086 | } | |
1087 | ||
1088 | /* | |
1089 | * Nfs timer routine | |
1090 | * Scan the nfsreq list and retranmit any requests that have timed out | |
1091 | * To avoid retransmission attempts on STREAM sockets (in the future) make | |
1092 | * sure to set the r_retry field to 0 (implies nm_retry == 0). | |
1093 | */ | |
1094 | nfs_timer() | |
1095 | { | |
1096 | register struct nfsreq *rep; | |
1097 | register struct mbuf *m; | |
1098 | register struct socket *so; | |
1099 | register struct nfsmount *nmp; | |
1100 | int s, error; | |
1101 | ||
1102 | s = splnet(); | |
1103 | for (rep = nfsreqh.r_next; rep != &nfsreqh; rep = rep->r_next) { | |
1104 | nmp = rep->r_nmp; | |
1105 | if (rep->r_mrep || (rep->r_flags & R_SOFTTERM) || | |
1106 | (so = nmp->nm_so) == NULL) | |
1107 | continue; | |
1108 | if ((nmp->nm_flag & NFSMNT_INT) && nfs_sigintr(rep->r_procp)) { | |
1109 | rep->r_flags |= R_SOFTTERM; | |
1110 | continue; | |
1111 | } | |
1112 | if (rep->r_flags & R_TIMING) /* update rtt in mount */ | |
1113 | nmp->nm_rtt++; | |
1114 | /* If not timed out */ | |
1115 | if (++rep->r_timer < nmp->nm_rto) | |
1116 | continue; | |
1117 | /* Do backoff and save new timeout in mount */ | |
1118 | if (rep->r_flags & R_TIMING) { | |
1119 | nfs_backofftimer(nmp); | |
1120 | rep->r_flags &= ~R_TIMING; | |
1121 | nmp->nm_rtt = -1; | |
1122 | } | |
1123 | if (rep->r_flags & R_SENT) { | |
1124 | rep->r_flags &= ~R_SENT; | |
1125 | nmp->nm_sent--; | |
1126 | } | |
1127 | ||
1128 | /* | |
1129 | * Check for too many retries on soft mount. | |
1130 | * nb: For hard mounts, r_retry == NFS_MAXREXMIT+1 | |
1131 | */ | |
1132 | if (++rep->r_rexmit > NFS_MAXREXMIT) | |
1133 | rep->r_rexmit = NFS_MAXREXMIT; | |
1134 | ||
1135 | /* | |
1136 | * Check for server not responding | |
1137 | */ | |
1138 | if ((rep->r_flags & R_TPRINTFMSG) == 0 && | |
1139 | rep->r_rexmit > NFS_FISHY) { | |
1140 | nfs_msg(rep->r_procp, | |
1141 | nmp->nm_mountp->mnt_stat.f_mntfromname, | |
1142 | "not responding"); | |
1143 | rep->r_flags |= R_TPRINTFMSG; | |
1144 | } | |
1145 | if (rep->r_rexmit >= rep->r_retry) { /* too many */ | |
1146 | nfsstats.rpctimeouts++; | |
1147 | rep->r_flags |= R_SOFTTERM; | |
1148 | continue; | |
1149 | } | |
1150 | if (nmp->nm_sotype != SOCK_DGRAM) | |
1151 | continue; | |
1152 | ||
1153 | /* | |
1154 | * If there is enough space and the window allows.. | |
1155 | * Resend it | |
1156 | */ | |
1157 | if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && | |
1158 | nmp->nm_sent < nmp->nm_window && | |
1159 | (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ | |
1160 | nfsstats.rpcretries++; | |
1161 | if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) | |
1162 | error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, | |
1163 | (caddr_t)0, (struct mbuf *)0, (struct mbuf *)0); | |
1164 | else | |
1165 | error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, | |
1166 | nmp->nm_nam, (struct mbuf *)0, (struct mbuf *)0); | |
1167 | if (error) { | |
1168 | if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) | |
1169 | so->so_error = 0; | |
1170 | } else { | |
1171 | /* | |
1172 | * We need to time the request even though we | |
1173 | * are retransmitting. | |
1174 | */ | |
1175 | nmp->nm_rtt = 0; | |
1176 | nmp->nm_sent++; | |
1177 | rep->r_flags |= (R_SENT|R_TIMING); | |
1178 | rep->r_timer = rep->r_timerinit; | |
1179 | } | |
1180 | } | |
1181 | } | |
1182 | splx(s); | |
1183 | timeout(nfs_timer, (caddr_t)0, hz/NFS_HZ); | |
1184 | } | |
1185 | ||
1186 | /* | |
1187 | * NFS timer update and backoff. The "Jacobson/Karels/Karn" scheme is | |
1188 | * used here. The timer state is held in the nfsmount structure and | |
1189 | * a single request is used to clock the response. When successful | |
1190 | * the rtt smoothing in nfs_updatetimer is used, when failed the backoff | |
1191 | * is done by nfs_backofftimer. We also log failure messages in these | |
1192 | * routines. | |
1193 | * | |
1194 | * Congestion variables are held in the nfshost structure which | |
1195 | * is referenced by nfsmounts and shared per-server. This separation | |
1196 | * makes it possible to do per-mount timing which allows varying disk | |
1197 | * access times to be dealt with, while preserving a network oriented | |
1198 | * congestion control scheme. | |
1199 | * | |
1200 | * The windowing implements the Jacobson/Karels slowstart algorithm | |
1201 | * with adjusted scaling factors. We start with one request, then send | |
1202 | * 4 more after each success until the ssthresh limit is reached, then | |
1203 | * we increment at a rate proportional to the window. On failure, we | |
1204 | * remember 3/4 the current window and clamp the send limit to 1. Note | |
1205 | * ICMP source quench is not reflected in so->so_error so we ignore that | |
1206 | * for now. | |
1207 | * | |
1208 | * NFS behaves much more like a transport protocol with these changes, | |
1209 | * shedding the teenage pedal-to-the-metal tendencies of "other" | |
1210 | * implementations. | |
1211 | * | |
1212 | * Timers and congestion avoidance by Tom Talpey, Open Software Foundation. | |
1213 | */ | |
1214 | ||
1215 | /* | |
1216 | * The TCP algorithm was not forgiving enough. Because the NFS server | |
1217 | * responds only after performing lookups/diskio/etc, we have to be | |
1218 | * more prepared to accept a spiky variance. The TCP algorithm is: | |
1219 | * TCP_RTO(nmp) ((((nmp)->nm_srtt >> 2) + (nmp)->nm_rttvar) >> 1) | |
1220 | */ | |
1221 | #define NFS_RTO(nmp) (((nmp)->nm_srtt >> 3) + (nmp)->nm_rttvar) | |
1222 | ||
1223 | nfs_updatetimer(nmp) | |
1224 | register struct nfsmount *nmp; | |
1225 | { | |
1226 | ||
1227 | /* If retransmitted, clear and return */ | |
1228 | if (nmp->nm_rexmit || nmp->nm_currexmit) { | |
1229 | nmp->nm_rexmit = nmp->nm_currexmit = 0; | |
1230 | return; | |
1231 | } | |
1232 | /* If have a measurement, do smoothing */ | |
1233 | if (nmp->nm_srtt) { | |
1234 | register short delta; | |
1235 | delta = nmp->nm_rtt - (nmp->nm_srtt >> 3); | |
1236 | if ((nmp->nm_srtt += delta) <= 0) | |
1237 | nmp->nm_srtt = 1; | |
1238 | if (delta < 0) | |
1239 | delta = -delta; | |
1240 | delta -= (nmp->nm_rttvar >> 2); | |
1241 | if ((nmp->nm_rttvar += delta) <= 0) | |
1242 | nmp->nm_rttvar = 1; | |
1243 | /* Else initialize */ | |
1244 | } else { | |
1245 | nmp->nm_rttvar = nmp->nm_rtt << 1; | |
1246 | if (nmp->nm_rttvar == 0) nmp->nm_rttvar = 2; | |
1247 | nmp->nm_srtt = nmp->nm_rttvar << 2; | |
1248 | } | |
1249 | /* Compute new Retransmission TimeOut and clip */ | |
1250 | nmp->nm_rto = NFS_RTO(nmp); | |
1251 | if (nmp->nm_rto < NFS_MINTIMEO) | |
1252 | nmp->nm_rto = NFS_MINTIMEO; | |
1253 | else if (nmp->nm_rto > NFS_MAXTIMEO) | |
1254 | nmp->nm_rto = NFS_MAXTIMEO; | |
1255 | ||
1256 | /* Update window estimate */ | |
1257 | if (nmp->nm_window < nmp->nm_ssthresh) /* quickly */ | |
1258 | nmp->nm_window += 4; | |
1259 | else { /* slowly */ | |
1260 | register long incr = ++nmp->nm_winext; | |
1261 | incr = (incr * incr) / nmp->nm_window; | |
1262 | if (incr > 0) { | |
1263 | nmp->nm_winext = 0; | |
1264 | ++nmp->nm_window; | |
1265 | } | |
1266 | } | |
1267 | if (nmp->nm_window > NFS_MAXWINDOW) | |
1268 | nmp->nm_window = NFS_MAXWINDOW; | |
1269 | } | |
1270 | ||
1271 | nfs_backofftimer(nmp) | |
1272 | register struct nfsmount *nmp; | |
1273 | { | |
1274 | register unsigned long newrto; | |
1275 | ||
1276 | /* Clip shift count */ | |
1277 | if (++nmp->nm_rexmit > 8 * sizeof nmp->nm_rto) | |
1278 | nmp->nm_rexmit = 8 * sizeof nmp->nm_rto; | |
1279 | /* Back off RTO exponentially */ | |
1280 | newrto = NFS_RTO(nmp); | |
1281 | newrto <<= (nmp->nm_rexmit - 1); | |
1282 | if (newrto == 0 || newrto > NFS_MAXTIMEO) | |
1283 | newrto = NFS_MAXTIMEO; | |
1284 | nmp->nm_rto = newrto; | |
1285 | ||
1286 | /* If too many retries, message, assume a bogus RTT and re-measure */ | |
1287 | if (nmp->nm_currexmit < nmp->nm_rexmit) { | |
1288 | nmp->nm_currexmit = nmp->nm_rexmit; | |
1289 | if (nmp->nm_currexmit >= nfsrexmtthresh) { | |
1290 | if (nmp->nm_currexmit == nfsrexmtthresh) { | |
1291 | nmp->nm_rttvar += (nmp->nm_srtt >> 2); | |
1292 | nmp->nm_srtt = 0; | |
1293 | } | |
1294 | } | |
1295 | } | |
1296 | /* Close down window but remember this point (3/4 current) for later */ | |
1297 | nmp->nm_ssthresh = ((nmp->nm_window << 1) + nmp->nm_window) >> 2; | |
1298 | nmp->nm_window = 1; | |
1299 | nmp->nm_winext = 0; | |
1300 | } | |
1301 | ||
1302 | /* | |
1303 | * Test for a termination signal pending on procp. | |
1304 | * This is used for NFSMNT_INT mounts. | |
1305 | */ | |
1306 | nfs_sigintr(p) | |
1307 | register struct proc *p; | |
1308 | { | |
1309 | if (p && p->p_sig && (((p->p_sig &~ p->p_sigmask) &~ p->p_sigignore) & | |
1310 | NFSINT_SIGMASK)) | |
1311 | return (1); | |
1312 | else | |
1313 | return (0); | |
1314 | } | |
1315 | ||
1316 | nfs_msg(p, server, msg) | |
1317 | struct proc *p; | |
1318 | char *server, *msg; | |
1319 | { | |
1320 | tpr_t tpr; | |
1321 | ||
1322 | if (p) | |
1323 | tpr = tprintf_open(p); | |
1324 | else | |
1325 | tpr = NULL; | |
1326 | tprintf(tpr, "nfs server %s: %s\n", server, msg); | |
1327 | tprintf_close(tpr); | |
1328 | } | |
1329 | ||
1330 | /* | |
1331 | * Lock a socket against others. | |
1332 | * Necessary for STREAM sockets to ensure you get an entire rpc request/reply | |
1333 | * and also to avoid race conditions between the processes with nfs requests | |
1334 | * in progress when a reconnect is necessary. | |
1335 | */ | |
1336 | nfs_solock(flagp) | |
1337 | register int *flagp; | |
1338 | { | |
1339 | ||
1340 | while (*flagp & NFSMNT_SCKLOCK) { | |
1341 | *flagp |= NFSMNT_WANTSCK; | |
1342 | (void) tsleep((caddr_t)flagp, PZERO-1, "nfsolck", 0); | |
1343 | } | |
1344 | *flagp |= NFSMNT_SCKLOCK; | |
1345 | } | |
1346 | ||
1347 | /* | |
1348 | * Unlock the stream socket for others. | |
1349 | */ | |
1350 | nfs_sounlock(flagp) | |
1351 | register int *flagp; | |
1352 | { | |
1353 | ||
1354 | if ((*flagp & NFSMNT_SCKLOCK) == 0) | |
1355 | panic("nfs sounlock"); | |
1356 | *flagp &= ~NFSMNT_SCKLOCK; | |
1357 | if (*flagp & NFSMNT_WANTSCK) { | |
1358 | *flagp &= ~NFSMNT_WANTSCK; | |
1359 | wakeup((caddr_t)flagp); | |
1360 | } | |
1361 | } | |
1362 | ||
1363 | /* | |
1364 | * This function compares two net addresses by family and returns TRUE | |
1365 | * if they are the same. | |
1366 | * If there is any doubt, return FALSE. | |
1367 | */ | |
1368 | nfs_netaddr_match(nam1, nam2) | |
1369 | struct mbuf *nam1, *nam2; | |
1370 | { | |
1371 | register struct sockaddr *saddr1, *saddr2; | |
1372 | ||
1373 | saddr1 = mtod(nam1, struct sockaddr *); | |
1374 | saddr2 = mtod(nam2, struct sockaddr *); | |
1375 | if (saddr1->sa_family != saddr2->sa_family) | |
1376 | return (0); | |
1377 | ||
1378 | /* | |
1379 | * Must do each address family separately since unused fields | |
1380 | * are undefined values and not always zeroed. | |
1381 | */ | |
1382 | switch (saddr1->sa_family) { | |
1383 | case AF_INET: | |
1384 | if (((struct sockaddr_in *)saddr1)->sin_addr.s_addr == | |
1385 | ((struct sockaddr_in *)saddr2)->sin_addr.s_addr) | |
1386 | return (1); | |
1387 | break; | |
1388 | default: | |
1389 | break; | |
1390 | }; | |
1391 | return (0); | |
1392 | } | |
1393 | ||
1394 | /* | |
1395 | * Check the hostname fields for nfsd's mask and match fields. | |
1396 | * By address family: | |
1397 | * - Bitwise AND the mask with the host address field | |
1398 | * - Compare for == with match | |
1399 | * return TRUE if not equal | |
1400 | */ | |
1401 | nfs_badnam(nam, msk, mtch) | |
1402 | register struct mbuf *nam, *msk, *mtch; | |
1403 | { | |
1404 | switch (mtod(nam, struct sockaddr *)->sa_family) { | |
1405 | case AF_INET: | |
1406 | return ((mtod(nam, struct sockaddr_in *)->sin_addr.s_addr & | |
1407 | mtod(msk, struct sockaddr_in *)->sin_addr.s_addr) != | |
1408 | mtod(mtch, struct sockaddr_in *)->sin_addr.s_addr); | |
1409 | default: | |
1410 | printf("nfs_badmatch, unknown sa_family\n"); | |
1411 | return (0); | |
1412 | }; | |
1413 | } |