Commit | Line | Data |
---|---|---|
a2907882 | 1 | /* |
f777974b | 2 | * Copyright (c) 1989, 1991 The Regents of the University of California. |
a2907882 KM |
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 | * | |
dbf0c423 | 8 | * %sccs.include.redist.c% |
a2907882 | 9 | * |
79993818 | 10 | * @(#)nfs_socket.c 7.21 (Berkeley) %G% |
a2907882 KM |
11 | */ |
12 | ||
13 | /* | |
f0f1cbaa | 14 | * Socket operations for use by nfs |
a2907882 KM |
15 | */ |
16 | ||
a2907882 | 17 | #include "param.h" |
2f08b65a | 18 | #include "proc.h" |
a2907882 KM |
19 | #include "mount.h" |
20 | #include "kernel.h" | |
21 | #include "malloc.h" | |
22 | #include "mbuf.h" | |
23 | #include "vnode.h" | |
24 | #include "domain.h" | |
25 | #include "protosw.h" | |
26 | #include "socket.h" | |
27 | #include "socketvar.h" | |
f777974b | 28 | #include "syslog.h" |
79993818 | 29 | #include "tprintf.h" |
37ced908 KM |
30 | #include "../netinet/in.h" |
31 | #include "../netinet/tcp.h" | |
f777974b | 32 | |
a2907882 KM |
33 | #include "rpcv2.h" |
34 | #include "nfsv2.h" | |
35 | #include "nfs.h" | |
36 | #include "xdr_subs.h" | |
37 | #include "nfsm_subs.h" | |
38 | #include "nfsmount.h" | |
39 | ||
40 | #define TRUE 1 | |
170bfd05 | 41 | #define FALSE 0 |
a2907882 | 42 | |
a2907882 KM |
43 | /* |
44 | * External data, mostly RPC constants in XDR form | |
45 | */ | |
46 | extern u_long rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, rpc_auth_unix, | |
47 | rpc_msgaccepted, rpc_call; | |
48 | extern u_long nfs_prog, nfs_vers; | |
170bfd05 | 49 | /* Maybe these should be bits in a u_long ?? */ |
f0f1cbaa | 50 | extern int nonidempotent[NFS_NPROCS]; |
958df9fb KM |
51 | static int compressrequest[NFS_NPROCS] = { |
52 | FALSE, | |
53 | TRUE, | |
54 | TRUE, | |
55 | FALSE, | |
56 | TRUE, | |
57 | TRUE, | |
58 | TRUE, | |
59 | FALSE, | |
60 | FALSE, | |
61 | TRUE, | |
62 | TRUE, | |
63 | TRUE, | |
64 | TRUE, | |
65 | TRUE, | |
66 | TRUE, | |
67 | TRUE, | |
68 | TRUE, | |
69 | TRUE, | |
70 | }; | |
f0f1cbaa KM |
71 | int nfs_sbwait(); |
72 | void nfs_disconnect(); | |
958df9fb | 73 | struct mbuf *nfs_compress(), *nfs_uncompress(); |
f0f1cbaa | 74 | |
a2907882 KM |
75 | int nfsrv_null(), |
76 | nfsrv_getattr(), | |
77 | nfsrv_setattr(), | |
78 | nfsrv_lookup(), | |
79 | nfsrv_readlink(), | |
80 | nfsrv_read(), | |
81 | nfsrv_write(), | |
82 | nfsrv_create(), | |
83 | nfsrv_remove(), | |
84 | nfsrv_rename(), | |
85 | nfsrv_link(), | |
86 | nfsrv_symlink(), | |
87 | nfsrv_mkdir(), | |
88 | nfsrv_rmdir(), | |
89 | nfsrv_readdir(), | |
90 | nfsrv_statfs(), | |
91 | nfsrv_noop(); | |
92 | ||
93 | int (*nfsrv_procs[NFS_NPROCS])() = { | |
94 | nfsrv_null, | |
95 | nfsrv_getattr, | |
96 | nfsrv_setattr, | |
97 | nfsrv_noop, | |
98 | nfsrv_lookup, | |
99 | nfsrv_readlink, | |
100 | nfsrv_read, | |
101 | nfsrv_noop, | |
102 | nfsrv_write, | |
103 | nfsrv_create, | |
104 | nfsrv_remove, | |
105 | nfsrv_rename, | |
106 | nfsrv_link, | |
107 | nfsrv_symlink, | |
108 | nfsrv_mkdir, | |
109 | nfsrv_rmdir, | |
110 | nfsrv_readdir, | |
111 | nfsrv_statfs, | |
112 | }; | |
113 | ||
2f08b65a KM |
114 | struct nfsreq nfsreqh; |
115 | int nfsrexmtthresh = NFS_FISHY; | |
f0f1cbaa | 116 | int nfs_tcpnodelay = 1; |
2f08b65a KM |
117 | |
118 | /* | |
f0f1cbaa | 119 | * Initialize sockets and congestion for a new NFS connection. |
2f08b65a KM |
120 | * We do not free the sockaddr if error. |
121 | */ | |
f0f1cbaa | 122 | nfs_connect(nmp) |
2f08b65a | 123 | register struct nfsmount *nmp; |
2f08b65a | 124 | { |
f0f1cbaa KM |
125 | register struct socket *so; |
126 | int s, error; | |
2f08b65a | 127 | struct mbuf *m; |
2f08b65a | 128 | |
f0f1cbaa KM |
129 | nmp->nm_so = (struct socket *)0; |
130 | if (error = socreate(mtod(nmp->nm_nam, struct sockaddr *)->sa_family, | |
131 | &nmp->nm_so, nmp->nm_sotype, nmp->nm_soproto)) | |
2f08b65a | 132 | goto bad; |
f0f1cbaa KM |
133 | so = nmp->nm_so; |
134 | nmp->nm_soflags = so->so_proto->pr_flags; | |
2f08b65a | 135 | |
f0f1cbaa KM |
136 | /* |
137 | * Protocols that do not require connections may be optionally left | |
138 | * unconnected for servers that reply from a port other than NFS_PORT. | |
139 | */ | |
140 | if (nmp->nm_flag & NFSMNT_NOCONN) { | |
141 | if (nmp->nm_soflags & PR_CONNREQUIRED) { | |
142 | error = ENOTCONN; | |
2f08b65a KM |
143 | goto bad; |
144 | } | |
f0f1cbaa KM |
145 | } else { |
146 | if (error = soconnect(so, nmp->nm_nam)) | |
2f08b65a | 147 | goto bad; |
f0f1cbaa KM |
148 | |
149 | /* | |
150 | * Wait for the connection to complete. Cribbed from the | |
151 | * connect system call but with the wait at negative prio. | |
152 | */ | |
153 | s = splnet(); | |
154 | while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) | |
170bfd05 | 155 | (void) tsleep((caddr_t)&so->so_timeo, PSOCK, "nfscon", 0); |
f0f1cbaa KM |
156 | splx(s); |
157 | if (so->so_error) { | |
158 | error = so->so_error; | |
159 | goto bad; | |
160 | } | |
2f08b65a | 161 | } |
f0f1cbaa | 162 | if (nmp->nm_sotype == SOCK_DGRAM) { |
170bfd05 | 163 | if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_SPONGY | NFSMNT_INT)) { |
f0f1cbaa KM |
164 | so->so_rcv.sb_timeo = (5 * hz); |
165 | so->so_snd.sb_timeo = (5 * hz); | |
166 | } else { | |
167 | so->so_rcv.sb_timeo = 0; | |
168 | so->so_snd.sb_timeo = 0; | |
169 | } | |
f777974b MK |
170 | if (error = soreserve(so, |
171 | min(4 * (nmp->nm_wsize + NFS_MAXPKTHDR), NFS_MAXPACKET), | |
172 | min(4 * (nmp->nm_rsize + NFS_MAXPKTHDR), NFS_MAXPACKET))) | |
f0f1cbaa KM |
173 | goto bad; |
174 | } else { | |
170bfd05 | 175 | if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_SPONGY | NFSMNT_INT)) { |
f0f1cbaa KM |
176 | so->so_rcv.sb_timeo = (5 * hz); |
177 | so->so_snd.sb_timeo = (5 * hz); | |
178 | } else { | |
179 | so->so_rcv.sb_timeo = 0; | |
180 | so->so_snd.sb_timeo = 0; | |
181 | } | |
182 | if (so->so_proto->pr_flags & PR_CONNREQUIRED) { | |
183 | MGET(m, M_WAIT, MT_SOOPTS); | |
184 | *mtod(m, int *) = 1; | |
185 | m->m_len = sizeof(int); | |
186 | sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); | |
187 | } | |
188 | if (so->so_proto->pr_domain->dom_family == AF_INET && | |
189 | so->so_proto->pr_protocol == IPPROTO_TCP && | |
190 | nfs_tcpnodelay) { | |
191 | MGET(m, M_WAIT, MT_SOOPTS); | |
192 | *mtod(m, int *) = 1; | |
193 | m->m_len = sizeof(int); | |
194 | sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); | |
195 | } | |
196 | if (error = soreserve(so, | |
f777974b MK |
197 | min(4 * (nmp->nm_wsize + NFS_MAXPKTHDR + sizeof(u_long)), |
198 | NFS_MAXPACKET + sizeof(u_long)), | |
199 | min(4 * (nmp->nm_rsize + NFS_MAXPKTHDR + sizeof(u_long)), | |
200 | NFS_MAXPACKET + sizeof(u_long)))) | |
f0f1cbaa KM |
201 | goto bad; |
202 | } | |
203 | so->so_rcv.sb_flags |= SB_NOINTR; | |
204 | so->so_snd.sb_flags |= SB_NOINTR; | |
2f08b65a | 205 | |
f0f1cbaa KM |
206 | /* Initialize other non-zero congestion variables */ |
207 | nmp->nm_rto = NFS_TIMEO; | |
208 | nmp->nm_window = 2; /* Initial send window */ | |
209 | nmp->nm_ssthresh = NFS_MAXWINDOW; /* Slowstart threshold */ | |
210 | nmp->nm_rttvar = nmp->nm_rto << 1; | |
211 | nmp->nm_sent = 0; | |
212 | nmp->nm_currexmit = 0; | |
213 | return (0); | |
2f08b65a | 214 | |
f0f1cbaa KM |
215 | bad: |
216 | nfs_disconnect(nmp); | |
217 | return (error); | |
218 | } | |
2f08b65a | 219 | |
f0f1cbaa KM |
220 | /* |
221 | * Reconnect routine: | |
222 | * Called when a connection is broken on a reliable protocol. | |
223 | * - clean up the old socket | |
224 | * - nfs_connect() again | |
225 | * - set R_MUSTRESEND for all outstanding requests on mount point | |
226 | * If this fails the mount point is DEAD! | |
227 | * nb: Must be called with the nfs_solock() set on the mount point. | |
228 | */ | |
229 | nfs_reconnect(rep, nmp) | |
230 | register struct nfsreq *rep; | |
231 | register struct nfsmount *nmp; | |
232 | { | |
233 | register struct nfsreq *rp; | |
f0f1cbaa | 234 | int error; |
2f08b65a | 235 | |
79993818 MK |
236 | nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, |
237 | "trying reconnect"); | |
f0f1cbaa | 238 | while (error = nfs_connect(nmp)) { |
d4e5799e KM |
239 | #ifdef lint |
240 | error = error; | |
241 | #endif /* lint */ | |
f0f1cbaa KM |
242 | if ((nmp->nm_flag & NFSMNT_INT) && nfs_sigintr(rep->r_procp)) |
243 | return (EINTR); | |
170bfd05 | 244 | (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); |
2f08b65a | 245 | } |
79993818 MK |
246 | nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, |
247 | "reconnected"); | |
f0f1cbaa KM |
248 | |
249 | /* | |
250 | * Loop through outstanding request list and fix up all requests | |
251 | * on old socket. | |
252 | */ | |
253 | rp = nfsreqh.r_next; | |
254 | while (rp != &nfsreqh) { | |
255 | if (rp->r_nmp == nmp) | |
256 | rp->r_flags |= R_MUSTRESEND; | |
257 | rp = rp->r_next; | |
2f08b65a KM |
258 | } |
259 | return (0); | |
2f08b65a KM |
260 | } |
261 | ||
262 | /* | |
263 | * NFS disconnect. Clean up and unlink. | |
264 | */ | |
f0f1cbaa | 265 | void |
2f08b65a KM |
266 | nfs_disconnect(nmp) |
267 | register struct nfsmount *nmp; | |
268 | { | |
f0f1cbaa | 269 | register struct socket *so; |
2f08b65a | 270 | |
f0f1cbaa KM |
271 | if (nmp->nm_so) { |
272 | so = nmp->nm_so; | |
273 | nmp->nm_so = (struct socket *)0; | |
274 | soshutdown(so, 2); | |
275 | soclose(so); | |
2f08b65a KM |
276 | } |
277 | } | |
a2907882 KM |
278 | |
279 | /* | |
f0f1cbaa KM |
280 | * This is the nfs send routine. For connection based socket types, it |
281 | * must be called with an nfs_solock() on the socket. | |
282 | * "rep == NULL" indicates that it has been called from a server. | |
a2907882 | 283 | */ |
f0f1cbaa | 284 | nfs_send(so, nam, top, rep) |
a2907882 KM |
285 | register struct socket *so; |
286 | struct mbuf *nam; | |
f0f1cbaa KM |
287 | register struct mbuf *top; |
288 | struct nfsreq *rep; | |
a2907882 | 289 | { |
f0f1cbaa KM |
290 | struct mbuf *sendnam; |
291 | int error, soflags; | |
a2907882 | 292 | |
f0f1cbaa KM |
293 | if (rep) { |
294 | if (rep->r_flags & R_SOFTTERM) { | |
2f08b65a | 295 | m_freem(top); |
f0f1cbaa | 296 | return (EINTR); |
2f08b65a | 297 | } |
5044b7a3 | 298 | if (rep->r_nmp->nm_so == NULL && |
f0f1cbaa KM |
299 | (error = nfs_reconnect(rep, rep->r_nmp))) |
300 | return (error); | |
301 | rep->r_flags &= ~R_MUSTRESEND; | |
5044b7a3 | 302 | so = rep->r_nmp->nm_so; |
f0f1cbaa KM |
303 | soflags = rep->r_nmp->nm_soflags; |
304 | } else | |
305 | soflags = so->so_proto->pr_flags; | |
306 | if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) | |
307 | sendnam = (struct mbuf *)0; | |
308 | else | |
309 | sendnam = nam; | |
310 | ||
311 | error = sosend(so, sendnam, (struct uio *)0, top, | |
312 | (struct mbuf *)0, 0); | |
313 | if (error == EWOULDBLOCK && rep) { | |
314 | if (rep->r_flags & R_SOFTTERM) | |
315 | error = EINTR; | |
316 | else { | |
317 | rep->r_flags |= R_MUSTRESEND; | |
318 | error = 0; | |
2f08b65a | 319 | } |
a2907882 | 320 | } |
f0f1cbaa KM |
321 | /* |
322 | * Ignore socket errors?? | |
323 | */ | |
324 | if (error && error != EINTR && error != ERESTART) | |
325 | error = 0; | |
a2907882 KM |
326 | return (error); |
327 | } | |
328 | ||
329 | /* | |
f0f1cbaa KM |
330 | * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all |
331 | * done by soreceive(), but for SOCK_STREAM we must deal with the Record | |
332 | * Mark and consolidate the data into a new mbuf list. | |
333 | * nb: Sometimes TCP passes the data up to soreceive() in long lists of | |
334 | * small mbufs. | |
335 | * For SOCK_STREAM we must be very careful to read an entire record once | |
336 | * we have read any of it, even if the system call has been interrupted. | |
a2907882 | 337 | */ |
f0f1cbaa | 338 | nfs_receive(so, aname, mp, rep) |
a2907882 KM |
339 | register struct socket *so; |
340 | struct mbuf **aname; | |
341 | struct mbuf **mp; | |
f0f1cbaa | 342 | register struct nfsreq *rep; |
a2907882 | 343 | { |
f0f1cbaa KM |
344 | struct uio auio; |
345 | struct iovec aio; | |
a2907882 | 346 | register struct mbuf *m; |
958df9fb | 347 | struct mbuf *m2, *mnew, **mbp; |
f0f1cbaa KM |
348 | caddr_t fcp, tcp; |
349 | u_long len; | |
350 | struct mbuf **getnam; | |
79993818 | 351 | int error, siz, mlen, soflags, rcvflg; |
a2907882 | 352 | |
f0f1cbaa KM |
353 | /* |
354 | * Set up arguments for soreceive() | |
355 | */ | |
356 | *mp = (struct mbuf *)0; | |
357 | *aname = (struct mbuf *)0; | |
358 | if (rep) | |
359 | soflags = rep->r_nmp->nm_soflags; | |
360 | else | |
361 | soflags = so->so_proto->pr_flags; | |
a2907882 | 362 | |
f0f1cbaa KM |
363 | /* |
364 | * For reliable protocols, lock against other senders/receivers | |
365 | * in case a reconnect is necessary. | |
366 | * For SOCK_STREAM, first get the Record Mark to find out how much | |
367 | * more there is to get. | |
368 | * We must lock the socket against other receivers | |
369 | * until we have an entire rpc request/reply. | |
370 | */ | |
371 | if (soflags & PR_CONNREQUIRED) { | |
372 | tryagain: | |
373 | /* | |
374 | * Check for fatal errors and resending request. | |
375 | */ | |
376 | if (rep) { | |
377 | /* | |
378 | * Ugh: If a reconnect attempt just happened, nm_so | |
379 | * would have changed. NULL indicates a failed | |
380 | * attempt that has essentially shut down this | |
381 | * mount point. | |
382 | */ | |
383 | if (rep->r_mrep || (so = rep->r_nmp->nm_so) == NULL || | |
384 | (rep->r_flags & R_SOFTTERM)) | |
385 | return (EINTR); | |
386 | while (rep->r_flags & R_MUSTRESEND) { | |
387 | m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); | |
388 | nfsstats.rpcretries++; | |
389 | if (error = nfs_send(so, rep->r_nmp->nm_nam, m, | |
390 | rep)) | |
391 | goto errout; | |
2f08b65a | 392 | } |
e8540f59 | 393 | } |
f0f1cbaa KM |
394 | if ((soflags & PR_ATOMIC) == 0) { |
395 | aio.iov_base = (caddr_t) &len; | |
396 | aio.iov_len = sizeof(u_long); | |
397 | auio.uio_iov = &aio; | |
398 | auio.uio_iovcnt = 1; | |
399 | auio.uio_segflg = UIO_SYSSPACE; | |
400 | auio.uio_rw = UIO_READ; | |
401 | auio.uio_offset = 0; | |
402 | auio.uio_resid = sizeof(u_long); | |
403 | do { | |
79993818 MK |
404 | rcvflg = MSG_WAITALL; |
405 | error = soreceive(so, (struct mbuf **)0, &auio, | |
f0f1cbaa | 406 | (struct mbuf **)0, (struct mbuf **)0, &rcvflg); |
79993818 | 407 | if (error == EWOULDBLOCK && rep) { |
f0f1cbaa KM |
408 | if (rep->r_flags & R_SOFTTERM) |
409 | return (EINTR); | |
410 | if (rep->r_flags & R_MUSTRESEND) | |
411 | goto tryagain; | |
79993818 | 412 | } |
f0f1cbaa | 413 | } while (error == EWOULDBLOCK); |
79993818 MK |
414 | if (!error && auio.uio_resid > 0) { |
415 | if (rep) | |
416 | log(LOG_INFO, | |
417 | "short receive (%d/%d) from nfs server %s\n", | |
418 | sizeof(u_long) - auio.uio_resid, | |
419 | sizeof(u_long), | |
420 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); | |
421 | error = EPIPE; | |
422 | } | |
f0f1cbaa KM |
423 | if (error) |
424 | goto errout; | |
425 | len = ntohl(len) & ~0x80000000; | |
426 | /* | |
427 | * This is SERIOUS! We are out of sync with the sender | |
428 | * and forcing a disconnect/reconnect is all I can do. | |
429 | */ | |
430 | if (len > NFS_MAXPACKET) { | |
79993818 MK |
431 | if (rep) |
432 | log(LOG_ERR, "%s (%d) from nfs server %s\n", | |
433 | "impossible packet length", | |
434 | len, | |
435 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); | |
436 | error = EFBIG; | |
437 | goto errout; | |
f0f1cbaa KM |
438 | } |
439 | auio.uio_resid = len; | |
440 | do { | |
79993818 | 441 | rcvflg = MSG_WAITALL; |
f0f1cbaa KM |
442 | error = soreceive(so, (struct mbuf **)0, |
443 | &auio, mp, (struct mbuf **)0, &rcvflg); | |
444 | } while (error == EWOULDBLOCK || error == EINTR || | |
445 | error == ERESTART); | |
79993818 MK |
446 | if (!error && auio.uio_resid > 0) { |
447 | if (rep) | |
448 | log(LOG_INFO, | |
449 | "short receive (%d/%d) from nfs server %s\n", | |
450 | len - auio.uio_resid, len, | |
451 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); | |
452 | error = EPIPE; | |
453 | } | |
2f08b65a | 454 | } else { |
f0f1cbaa KM |
455 | auio.uio_resid = len = 1000000; /* Anything Big */ |
456 | do { | |
79993818 | 457 | rcvflg = 0; |
f0f1cbaa KM |
458 | error = soreceive(so, (struct mbuf **)0, |
459 | &auio, mp, (struct mbuf **)0, &rcvflg); | |
460 | if (error == EWOULDBLOCK && rep) { | |
461 | if (rep->r_flags & R_SOFTTERM) | |
462 | return (EINTR); | |
463 | if (rep->r_flags & R_MUSTRESEND) | |
464 | goto tryagain; | |
465 | } | |
466 | } while (error == EWOULDBLOCK); | |
467 | if (!error && *mp == NULL) | |
468 | error = EPIPE; | |
469 | len -= auio.uio_resid; | |
2f08b65a | 470 | } |
f0f1cbaa KM |
471 | errout: |
472 | if (error && rep && error != EINTR && error != ERESTART) { | |
473 | m_freem(*mp); | |
474 | *mp = (struct mbuf *)0; | |
79993818 MK |
475 | if (error != EPIPE && rep) |
476 | log(LOG_INFO, | |
477 | "receive error %d from nfs server %s\n", | |
478 | error, | |
479 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); | |
f0f1cbaa KM |
480 | nfs_disconnect(rep->r_nmp); |
481 | error = nfs_reconnect(rep, rep->r_nmp); | |
482 | if (!error) | |
483 | goto tryagain; | |
2f08b65a | 484 | } |
f0f1cbaa KM |
485 | } else { |
486 | if (so->so_state & SS_ISCONNECTED) | |
487 | getnam = (struct mbuf **)0; | |
488 | else | |
489 | getnam = aname; | |
490 | auio.uio_resid = len = 1000000; | |
491 | do { | |
79993818 | 492 | rcvflg = 0; |
f0f1cbaa KM |
493 | error = soreceive(so, getnam, &auio, mp, |
494 | (struct mbuf **)0, &rcvflg); | |
495 | if (error == EWOULDBLOCK && rep && | |
496 | (rep->r_flags & R_SOFTTERM)) | |
497 | return (EINTR); | |
498 | } while (error == EWOULDBLOCK); | |
499 | len -= auio.uio_resid; | |
500 | } | |
501 | if (error) { | |
502 | m_freem(*mp); | |
503 | *mp = (struct mbuf *)0; | |
504 | } | |
505 | /* | |
506 | * Search for any mbufs that are not a multiple of 4 bytes long. | |
507 | * These could cause pointer alignment problems, so copy them to | |
508 | * well aligned mbufs. | |
509 | */ | |
510 | m = *mp; | |
511 | mbp = mp; | |
512 | while (m) { | |
513 | /* | |
514 | * All this for something that may never happen. | |
515 | */ | |
958df9fb | 516 | if (m->m_next && (m->m_len & 0x3)) { |
f0f1cbaa | 517 | printf("nfs_rcv odd length!\n"); |
d4e5799e | 518 | mlen = 0; |
f0f1cbaa | 519 | while (m) { |
958df9fb KM |
520 | fcp = mtod(m, caddr_t); |
521 | while (m->m_len > 0) { | |
522 | if (mlen == 0) { | |
523 | MGET(m2, M_WAIT, MT_DATA); | |
524 | if (len >= MINCLSIZE) | |
525 | MCLGET(m2, M_WAIT); | |
526 | m2->m_len = 0; | |
527 | mlen = M_TRAILINGSPACE(m2); | |
528 | tcp = mtod(m2, caddr_t); | |
529 | *mbp = m2; | |
530 | mbp = &m2->m_next; | |
531 | } | |
532 | siz = MIN(mlen, m->m_len); | |
533 | bcopy(fcp, tcp, siz); | |
534 | m2->m_len += siz; | |
535 | mlen -= siz; | |
536 | len -= siz; | |
537 | tcp += siz; | |
538 | m->m_len -= siz; | |
539 | fcp += siz; | |
f0f1cbaa | 540 | } |
958df9fb KM |
541 | MFREE(m, mnew); |
542 | m = mnew; | |
f0f1cbaa | 543 | } |
f0f1cbaa | 544 | break; |
2f08b65a | 545 | } |
f0f1cbaa KM |
546 | len -= m->m_len; |
547 | mbp = &m->m_next; | |
548 | m = m->m_next; | |
a2907882 | 549 | } |
a2907882 KM |
550 | return (error); |
551 | } | |
552 | ||
a2907882 | 553 | /* |
f0f1cbaa | 554 | * Implement receipt of reply on a socket. |
a2907882 KM |
555 | * We must search through the list of received datagrams matching them |
556 | * with outstanding requests using the xid, until ours is found. | |
557 | */ | |
f0f1cbaa KM |
558 | /* ARGSUSED */ |
559 | nfs_reply(nmp, myrep) | |
560 | struct nfsmount *nmp; | |
ffe6f482 | 561 | struct nfsreq *myrep; |
a2907882 KM |
562 | { |
563 | register struct mbuf *m; | |
564 | register struct nfsreq *rep; | |
f0f1cbaa | 565 | register int error = 0; |
958df9fb | 566 | u_long rxid; |
f0f1cbaa KM |
567 | struct mbuf *mp, *nam; |
568 | char *cp; | |
569 | int cnt, xfer; | |
a2907882 KM |
570 | |
571 | /* | |
f0f1cbaa | 572 | * Loop around until we get our own reply |
a2907882 | 573 | */ |
f0f1cbaa KM |
574 | for (;;) { |
575 | /* | |
576 | * Lock against other receivers so that I don't get stuck in | |
577 | * sbwait() after someone else has received my reply for me. | |
578 | * Also necessary for connection based protocols to avoid | |
579 | * race conditions during a reconnect. | |
580 | */ | |
170bfd05 | 581 | nfs_solock(&nmp->nm_flag); |
f0f1cbaa KM |
582 | /* Already received, bye bye */ |
583 | if (myrep->r_mrep != NULL) { | |
584 | nfs_sounlock(&nmp->nm_flag); | |
585 | return (0); | |
586 | } | |
587 | /* | |
588 | * Get the next Rpc reply off the socket | |
589 | */ | |
590 | if (error = nfs_receive(nmp->nm_so, &nam, &mp, myrep)) { | |
591 | nfs_sounlock(&nmp->nm_flag); | |
a2907882 | 592 | |
f0f1cbaa KM |
593 | /* |
594 | * Ignore routing errors on connectionless protocols?? | |
595 | */ | |
596 | if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { | |
597 | nmp->nm_so->so_error = 0; | |
598 | continue; | |
a2907882 | 599 | } |
f0f1cbaa KM |
600 | |
601 | /* | |
602 | * Otherwise cleanup and return a fatal error. | |
603 | */ | |
604 | if (myrep->r_flags & R_TIMING) { | |
605 | myrep->r_flags &= ~R_TIMING; | |
606 | nmp->nm_rtt = -1; | |
a2907882 | 607 | } |
f0f1cbaa KM |
608 | if (myrep->r_flags & R_SENT) { |
609 | myrep->r_flags &= ~R_SENT; | |
610 | nmp->nm_sent--; | |
2f08b65a | 611 | } |
f0f1cbaa KM |
612 | return (error); |
613 | } | |
614 | ||
615 | /* | |
616 | * Get the xid and check that it is an rpc reply | |
617 | */ | |
618 | m = mp; | |
958df9fb KM |
619 | while (m && m->m_len == 0) |
620 | m = m->m_next; | |
621 | if (m == NULL) { | |
f0f1cbaa KM |
622 | nfsstats.rpcinvalid++; |
623 | m_freem(mp); | |
624 | nfs_sounlock(&nmp->nm_flag); | |
625 | continue; | |
626 | } | |
958df9fb | 627 | bcopy(mtod(m, caddr_t), (caddr_t)&rxid, NFSX_UNSIGNED); |
f0f1cbaa KM |
628 | /* |
629 | * Loop through the request list to match up the reply | |
630 | * Iff no match, just drop the datagram | |
631 | */ | |
632 | m = mp; | |
633 | rep = nfsreqh.r_next; | |
634 | while (rep != &nfsreqh) { | |
958df9fb | 635 | if (rep->r_mrep == NULL && rxid == rep->r_xid) { |
f0f1cbaa KM |
636 | /* Found it.. */ |
637 | rep->r_mrep = m; | |
638 | /* | |
639 | * Update timing | |
640 | */ | |
641 | if (rep->r_flags & R_TIMING) { | |
642 | nfs_updatetimer(rep->r_nmp); | |
643 | rep->r_flags &= ~R_TIMING; | |
644 | rep->r_nmp->nm_rtt = -1; | |
645 | } | |
646 | if (rep->r_flags & R_SENT) { | |
647 | rep->r_flags &= ~R_SENT; | |
648 | rep->r_nmp->nm_sent--; | |
649 | } | |
650 | break; | |
2f08b65a | 651 | } |
f0f1cbaa | 652 | rep = rep->r_next; |
a2907882 | 653 | } |
f0f1cbaa KM |
654 | nfs_sounlock(&nmp->nm_flag); |
655 | if (nam) | |
656 | m_freem(nam); | |
657 | /* | |
658 | * If not matched to a request, drop it. | |
659 | * If it's mine, get out. | |
660 | */ | |
661 | if (rep == &nfsreqh) { | |
662 | nfsstats.rpcunexpected++; | |
663 | m_freem(m); | |
664 | } else if (rep == myrep) | |
665 | return (0); | |
a2907882 | 666 | } |
a2907882 KM |
667 | } |
668 | ||
669 | /* | |
670 | * nfs_request - goes something like this | |
671 | * - fill in request struct | |
672 | * - links it into list | |
f0f1cbaa KM |
673 | * - calls nfs_send() for first transmit |
674 | * - calls nfs_receive() to get reply | |
a2907882 KM |
675 | * - break down rpc header and return with nfs reply pointed to |
676 | * by mrep or error | |
677 | * nb: always frees up mreq mbuf list | |
678 | */ | |
170bfd05 | 679 | nfs_request(vp, mreq, xid, procnum, procp, tryhard, mp, mrp, mdp, dposp) |
a2907882 KM |
680 | struct vnode *vp; |
681 | struct mbuf *mreq; | |
682 | u_long xid; | |
f0f1cbaa KM |
683 | int procnum; |
684 | struct proc *procp; | |
170bfd05 | 685 | int tryhard; |
a2907882 KM |
686 | struct mount *mp; |
687 | struct mbuf **mrp; | |
688 | struct mbuf **mdp; | |
689 | caddr_t *dposp; | |
690 | { | |
691 | register struct mbuf *m, *mrep; | |
692 | register struct nfsreq *rep; | |
693 | register u_long *p; | |
694 | register int len; | |
f0f1cbaa | 695 | struct nfsmount *nmp; |
a2907882 | 696 | struct mbuf *md; |
ffe6f482 | 697 | struct nfsreq *reph; |
a2907882 KM |
698 | caddr_t dpos; |
699 | char *cp2; | |
700 | int t1; | |
958df9fb | 701 | int s, compressed; |
f0f1cbaa | 702 | int error = 0; |
a2907882 | 703 | |
f0f1cbaa | 704 | nmp = VFSTONFS(mp); |
a2907882 KM |
705 | m = mreq; |
706 | MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); | |
707 | rep->r_xid = xid; | |
f0f1cbaa | 708 | rep->r_nmp = nmp; |
a2907882 | 709 | rep->r_vp = vp; |
f0f1cbaa | 710 | rep->r_procp = procp; |
170bfd05 KM |
711 | if ((nmp->nm_flag & NFSMNT_SOFT) || |
712 | ((nmp->nm_flag & NFSMNT_SPONGY) && !tryhard)) | |
f0f1cbaa | 713 | rep->r_retry = nmp->nm_retry; |
a2907882 | 714 | else |
2f08b65a KM |
715 | rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ |
716 | rep->r_flags = rep->r_rexmit = 0; | |
f0f1cbaa KM |
717 | /* |
718 | * Three cases: | |
719 | * - non-idempotent requests on SOCK_DGRAM use NFS_MINIDEMTIMEO | |
720 | * - idempotent requests on SOCK_DGRAM use 0 | |
721 | * - Reliable transports, NFS_RELIABLETIMEO | |
722 | * Timeouts are still done on reliable transports to ensure detection | |
170bfd05 | 723 | * of excessive connection delay. |
f0f1cbaa KM |
724 | */ |
725 | if (nmp->nm_sotype != SOCK_DGRAM) | |
726 | rep->r_timerinit = -NFS_RELIABLETIMEO; | |
727 | else if (nonidempotent[procnum]) | |
728 | rep->r_timerinit = -NFS_MINIDEMTIMEO; | |
729 | else | |
730 | rep->r_timerinit = 0; | |
731 | rep->r_timer = rep->r_timerinit; | |
a2907882 | 732 | rep->r_mrep = NULL; |
a2907882 KM |
733 | len = 0; |
734 | while (m) { | |
735 | len += m->m_len; | |
736 | m = m->m_next; | |
737 | } | |
f0f1cbaa KM |
738 | mreq->m_pkthdr.len = len; |
739 | mreq->m_pkthdr.rcvif = (struct ifnet *)0; | |
958df9fb KM |
740 | compressed = 0; |
741 | m = mreq; | |
742 | if ((nmp->nm_flag & NFSMNT_COMPRESS) && compressrequest[procnum]) { | |
743 | mreq = nfs_compress(mreq); | |
744 | if (mreq != m) { | |
745 | len = mreq->m_pkthdr.len; | |
746 | compressed++; | |
747 | } | |
748 | } | |
f0f1cbaa KM |
749 | /* |
750 | * For non-atomic protocols, insert a Sun RPC Record Mark. | |
751 | */ | |
752 | if ((nmp->nm_soflags & PR_ATOMIC) == 0) { | |
753 | M_PREPEND(mreq, sizeof(u_long), M_WAIT); | |
754 | *mtod(mreq, u_long *) = htonl(0x80000000 | len); | |
755 | } | |
756 | rep->r_mreq = mreq; | |
a2907882 | 757 | |
2f08b65a KM |
758 | /* |
759 | * Do the client side RPC. | |
760 | */ | |
761 | nfsstats.rpcrequests++; | |
f0f1cbaa KM |
762 | /* |
763 | * Chain request into list of outstanding requests. Be sure | |
764 | * to put it LAST so timer finds oldest requests first. | |
765 | */ | |
a2907882 | 766 | s = splnet(); |
2f08b65a | 767 | reph = &nfsreqh; |
f0f1cbaa KM |
768 | reph->r_prev->r_next = rep; |
769 | rep->r_prev = reph->r_prev; | |
ffe6f482 KM |
770 | reph->r_prev = rep; |
771 | rep->r_next = reph; | |
2f08b65a KM |
772 | /* |
773 | * If backing off another request or avoiding congestion, don't | |
774 | * send this one now but let timer do it. If not timing a request, | |
775 | * do it now. | |
776 | */ | |
f0f1cbaa KM |
777 | if (nmp->nm_sent <= 0 || nmp->nm_sotype != SOCK_DGRAM || |
778 | (nmp->nm_currexmit == 0 && nmp->nm_sent < nmp->nm_window)) { | |
779 | nmp->nm_sent++; | |
780 | rep->r_flags |= R_SENT; | |
781 | if (nmp->nm_rtt == -1) { | |
782 | nmp->nm_rtt = 0; | |
783 | rep->r_flags |= R_TIMING; | |
784 | } | |
785 | splx(s); | |
786 | m = m_copym(mreq, 0, M_COPYALL, M_WAIT); | |
787 | if (nmp->nm_soflags & PR_CONNREQUIRED) | |
170bfd05 | 788 | nfs_solock(&nmp->nm_flag); |
f0f1cbaa KM |
789 | error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); |
790 | if (nmp->nm_soflags & PR_CONNREQUIRED) | |
791 | nfs_sounlock(&nmp->nm_flag); | |
792 | if (error && NFSIGNORE_SOERROR(nmp->nm_soflags, error)) | |
793 | nmp->nm_so->so_error = error = 0; | |
794 | } else | |
2f08b65a | 795 | splx(s); |
a2907882 | 796 | |
2f08b65a KM |
797 | /* |
798 | * Wait for the reply from our send or the timer's. | |
799 | */ | |
f0f1cbaa KM |
800 | if (!error) |
801 | error = nfs_reply(nmp, rep); | |
a2907882 | 802 | |
2f08b65a KM |
803 | /* |
804 | * RPC done, unlink the request. | |
805 | */ | |
a2907882 KM |
806 | s = splnet(); |
807 | rep->r_prev->r_next = rep->r_next; | |
ffe6f482 | 808 | rep->r_next->r_prev = rep->r_prev; |
a2907882 | 809 | splx(s); |
f0f1cbaa KM |
810 | |
811 | /* | |
812 | * If there was a successful reply and a tprintf msg. | |
813 | * tprintf a response. | |
814 | */ | |
79993818 MK |
815 | if (!error && (rep->r_flags & R_TPRINTFMSG)) |
816 | nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, | |
817 | "is alive again"); | |
a2907882 | 818 | m_freem(rep->r_mreq); |
958df9fb | 819 | mrep = rep->r_mrep; |
a2907882 KM |
820 | FREE((caddr_t)rep, M_NFSREQ); |
821 | if (error) | |
822 | return (error); | |
823 | ||
958df9fb KM |
824 | if (compressed) |
825 | mrep = nfs_uncompress(mrep); | |
826 | md = mrep; | |
a2907882 KM |
827 | /* |
828 | * break down the rpc header and check if ok | |
829 | */ | |
830 | dpos = mtod(md, caddr_t); | |
831 | nfsm_disect(p, u_long *, 5*NFSX_UNSIGNED); | |
832 | p += 2; | |
833 | if (*p++ == rpc_msgdenied) { | |
834 | if (*p == rpc_mismatch) | |
835 | error = EOPNOTSUPP; | |
836 | else | |
837 | error = EACCES; | |
838 | m_freem(mrep); | |
839 | return (error); | |
840 | } | |
841 | /* | |
842 | * skip over the auth_verf, someday we may want to cache auth_short's | |
843 | * for nfs_reqhead(), but for now just dump it | |
844 | */ | |
845 | if (*++p != 0) { | |
846 | len = nfsm_rndup(fxdr_unsigned(long, *p)); | |
847 | nfsm_adv(len); | |
848 | } | |
849 | nfsm_disect(p, u_long *, NFSX_UNSIGNED); | |
850 | /* 0 == ok */ | |
851 | if (*p == 0) { | |
852 | nfsm_disect(p, u_long *, NFSX_UNSIGNED); | |
853 | if (*p != 0) { | |
854 | error = fxdr_unsigned(int, *p); | |
855 | m_freem(mrep); | |
856 | return (error); | |
857 | } | |
858 | *mrp = mrep; | |
859 | *mdp = md; | |
860 | *dposp = dpos; | |
861 | return (0); | |
862 | } | |
863 | m_freem(mrep); | |
864 | return (EPROTONOSUPPORT); | |
865 | nfsmout: | |
866 | return (error); | |
867 | } | |
868 | ||
869 | /* | |
870 | * Get a request for the server main loop | |
871 | * - receive a request via. nfs_soreceive() | |
872 | * - verify it | |
873 | * - fill in the cred struct. | |
874 | */ | |
d4e5799e | 875 | nfs_getreq(so, prog, vers, maxproc, nam, mrp, mdp, dposp, retxid, procnum, cr, |
958df9fb | 876 | msk, mtch, wascomp) |
a2907882 KM |
877 | struct socket *so; |
878 | u_long prog; | |
879 | u_long vers; | |
880 | int maxproc; | |
881 | struct mbuf **nam; | |
882 | struct mbuf **mrp; | |
883 | struct mbuf **mdp; | |
884 | caddr_t *dposp; | |
885 | u_long *retxid; | |
d4e5799e | 886 | u_long *procnum; |
a2907882 | 887 | register struct ucred *cr; |
f0f1cbaa | 888 | struct mbuf *msk, *mtch; |
958df9fb | 889 | int *wascomp; |
a2907882 KM |
890 | { |
891 | register int i; | |
0bd503ad KM |
892 | register u_long *p; |
893 | register long t1; | |
894 | caddr_t dpos, cp2; | |
895 | int error = 0; | |
896 | struct mbuf *mrep, *md; | |
897 | int len; | |
a2907882 | 898 | |
f0f1cbaa | 899 | if (so->so_proto->pr_flags & PR_CONNREQUIRED) { |
f0f1cbaa | 900 | error = nfs_receive(so, nam, &mrep, (struct nfsreq *)0); |
f0f1cbaa KM |
901 | } else { |
902 | mrep = (struct mbuf *)0; | |
903 | do { | |
904 | if (mrep) { | |
905 | m_freem(*nam); | |
906 | m_freem(mrep); | |
907 | } | |
908 | error = nfs_receive(so, nam, &mrep, (struct nfsreq *)0); | |
909 | } while (!error && nfs_badnam(*nam, msk, mtch)); | |
910 | } | |
911 | if (error) | |
a2907882 KM |
912 | return (error); |
913 | md = mrep; | |
958df9fb KM |
914 | mrep = nfs_uncompress(mrep); |
915 | if (mrep != md) { | |
916 | *wascomp = 1; | |
917 | md = mrep; | |
918 | } else | |
919 | *wascomp = 0; | |
a2907882 KM |
920 | dpos = mtod(mrep, caddr_t); |
921 | nfsm_disect(p, u_long *, 10*NFSX_UNSIGNED); | |
922 | *retxid = *p++; | |
923 | if (*p++ != rpc_call) { | |
924 | m_freem(mrep); | |
925 | return (ERPCMISMATCH); | |
926 | } | |
927 | if (*p++ != rpc_vers) { | |
928 | m_freem(mrep); | |
929 | return (ERPCMISMATCH); | |
930 | } | |
931 | if (*p++ != prog) { | |
932 | m_freem(mrep); | |
933 | return (EPROGUNAVAIL); | |
934 | } | |
935 | if (*p++ != vers) { | |
936 | m_freem(mrep); | |
937 | return (EPROGMISMATCH); | |
938 | } | |
d4e5799e KM |
939 | *procnum = fxdr_unsigned(u_long, *p++); |
940 | if (*procnum == NFSPROC_NULL) { | |
a2907882 KM |
941 | *mrp = mrep; |
942 | return (0); | |
943 | } | |
d4e5799e | 944 | if (*procnum > maxproc || *p++ != rpc_auth_unix) { |
a2907882 KM |
945 | m_freem(mrep); |
946 | return (EPROCUNAVAIL); | |
947 | } | |
f0f1cbaa KM |
948 | len = fxdr_unsigned(int, *p++); |
949 | if (len < 0 || len > RPCAUTH_MAXSIZ) { | |
950 | m_freem(mrep); | |
951 | return (EBADRPC); | |
952 | } | |
0bd503ad | 953 | len = fxdr_unsigned(int, *++p); |
f0f1cbaa KM |
954 | if (len < 0 || len > NFS_MAXNAMLEN) { |
955 | m_freem(mrep); | |
956 | return (EBADRPC); | |
957 | } | |
0bd503ad | 958 | nfsm_adv(nfsm_rndup(len)); |
a2907882 KM |
959 | nfsm_disect(p, u_long *, 3*NFSX_UNSIGNED); |
960 | cr->cr_uid = fxdr_unsigned(uid_t, *p++); | |
961 | cr->cr_gid = fxdr_unsigned(gid_t, *p++); | |
0bd503ad | 962 | len = fxdr_unsigned(int, *p); |
f0f1cbaa | 963 | if (len < 0 || len > RPCAUTH_UNIXGIDS) { |
a2907882 KM |
964 | m_freem(mrep); |
965 | return (EBADRPC); | |
966 | } | |
0bd503ad KM |
967 | nfsm_disect(p, u_long *, (len + 2)*NFSX_UNSIGNED); |
968 | for (i = 1; i <= len; i++) | |
f0f1cbaa KM |
969 | if (i < NGROUPS) |
970 | cr->cr_groups[i] = fxdr_unsigned(gid_t, *p++); | |
971 | else | |
972 | p++; | |
973 | cr->cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1); | |
a2907882 KM |
974 | /* |
975 | * Do we have any use for the verifier. | |
976 | * According to the "Remote Procedure Call Protocol Spec." it | |
977 | * should be AUTH_NULL, but some clients make it AUTH_UNIX? | |
978 | * For now, just skip over it | |
979 | */ | |
0bd503ad | 980 | len = fxdr_unsigned(int, *++p); |
f0f1cbaa KM |
981 | if (len < 0 || len > RPCAUTH_MAXSIZ) { |
982 | m_freem(mrep); | |
983 | return (EBADRPC); | |
984 | } | |
0bd503ad KM |
985 | if (len > 0) |
986 | nfsm_adv(nfsm_rndup(len)); | |
a2907882 KM |
987 | *mrp = mrep; |
988 | *mdp = md; | |
989 | *dposp = dpos; | |
990 | return (0); | |
991 | nfsmout: | |
992 | return (error); | |
993 | } | |
994 | ||
995 | /* | |
996 | * Generate the rpc reply header | |
997 | * siz arg. is used to decide if adding a cluster is worthwhile | |
998 | */ | |
999 | nfs_rephead(siz, retxid, err, mrq, mbp, bposp) | |
1000 | int siz; | |
1001 | u_long retxid; | |
1002 | int err; | |
1003 | struct mbuf **mrq; | |
1004 | struct mbuf **mbp; | |
1005 | caddr_t *bposp; | |
1006 | { | |
0bd503ad KM |
1007 | register u_long *p; |
1008 | register long t1; | |
1009 | caddr_t bpos; | |
1010 | struct mbuf *mreq, *mb, *mb2; | |
a2907882 KM |
1011 | |
1012 | NFSMGETHDR(mreq); | |
1013 | mb = mreq; | |
1014 | if ((siz+RPC_REPLYSIZ) > MHLEN) | |
f0f1cbaa | 1015 | MCLGET(mreq, M_WAIT); |
a2907882 KM |
1016 | p = mtod(mreq, u_long *); |
1017 | mreq->m_len = 6*NFSX_UNSIGNED; | |
1018 | bpos = ((caddr_t)p)+mreq->m_len; | |
1019 | *p++ = retxid; | |
1020 | *p++ = rpc_reply; | |
1021 | if (err == ERPCMISMATCH) { | |
1022 | *p++ = rpc_msgdenied; | |
1023 | *p++ = rpc_mismatch; | |
1024 | *p++ = txdr_unsigned(2); | |
1025 | *p = txdr_unsigned(2); | |
1026 | } else { | |
1027 | *p++ = rpc_msgaccepted; | |
1028 | *p++ = 0; | |
1029 | *p++ = 0; | |
1030 | switch (err) { | |
1031 | case EPROGUNAVAIL: | |
1032 | *p = txdr_unsigned(RPC_PROGUNAVAIL); | |
1033 | break; | |
1034 | case EPROGMISMATCH: | |
1035 | *p = txdr_unsigned(RPC_PROGMISMATCH); | |
1036 | nfsm_build(p, u_long *, 2*NFSX_UNSIGNED); | |
1037 | *p++ = txdr_unsigned(2); | |
1038 | *p = txdr_unsigned(2); /* someday 3 */ | |
1039 | break; | |
1040 | case EPROCUNAVAIL: | |
1041 | *p = txdr_unsigned(RPC_PROCUNAVAIL); | |
1042 | break; | |
1043 | default: | |
1044 | *p = 0; | |
1045 | if (err != VNOVAL) { | |
1046 | nfsm_build(p, u_long *, NFSX_UNSIGNED); | |
1047 | *p = txdr_unsigned(err); | |
1048 | } | |
1049 | break; | |
1050 | }; | |
1051 | } | |
1052 | *mrq = mreq; | |
1053 | *mbp = mb; | |
1054 | *bposp = bpos; | |
1055 | if (err != 0 && err != VNOVAL) | |
1056 | nfsstats.srvrpc_errs++; | |
1057 | return (0); | |
1058 | } | |
1059 | ||
1060 | /* | |
1061 | * Nfs timer routine | |
1062 | * Scan the nfsreq list and retranmit any requests that have timed out | |
1063 | * To avoid retransmission attempts on STREAM sockets (in the future) make | |
2f08b65a | 1064 | * sure to set the r_retry field to 0 (implies nm_retry == 0). |
a2907882 KM |
1065 | */ |
1066 | nfs_timer() | |
1067 | { | |
1068 | register struct nfsreq *rep; | |
1069 | register struct mbuf *m; | |
1070 | register struct socket *so; | |
f0f1cbaa | 1071 | register struct nfsmount *nmp; |
2f08b65a | 1072 | int s, error; |
a2907882 KM |
1073 | |
1074 | s = splnet(); | |
f0f1cbaa KM |
1075 | for (rep = nfsreqh.r_next; rep != &nfsreqh; rep = rep->r_next) { |
1076 | nmp = rep->r_nmp; | |
1077 | if (rep->r_mrep || (rep->r_flags & R_SOFTTERM) || | |
1078 | (so = nmp->nm_so) == NULL) | |
1079 | continue; | |
1080 | if ((nmp->nm_flag & NFSMNT_INT) && nfs_sigintr(rep->r_procp)) { | |
1081 | rep->r_flags |= R_SOFTTERM; | |
1082 | continue; | |
1083 | } | |
2f08b65a | 1084 | if (rep->r_flags & R_TIMING) /* update rtt in mount */ |
f0f1cbaa | 1085 | nmp->nm_rtt++; |
f0f1cbaa KM |
1086 | /* If not timed out */ |
1087 | if (++rep->r_timer < nmp->nm_rto) | |
2f08b65a KM |
1088 | continue; |
1089 | /* Do backoff and save new timeout in mount */ | |
1090 | if (rep->r_flags & R_TIMING) { | |
f0f1cbaa | 1091 | nfs_backofftimer(nmp); |
2f08b65a | 1092 | rep->r_flags &= ~R_TIMING; |
f0f1cbaa | 1093 | nmp->nm_rtt = -1; |
2f08b65a KM |
1094 | } |
1095 | if (rep->r_flags & R_SENT) { | |
1096 | rep->r_flags &= ~R_SENT; | |
f0f1cbaa | 1097 | nmp->nm_sent--; |
2f08b65a | 1098 | } |
f0f1cbaa KM |
1099 | |
1100 | /* | |
1101 | * Check for too many retries on soft mount. | |
1102 | * nb: For hard mounts, r_retry == NFS_MAXREXMIT+1 | |
1103 | */ | |
1104 | if (++rep->r_rexmit > NFS_MAXREXMIT) | |
2f08b65a | 1105 | rep->r_rexmit = NFS_MAXREXMIT; |
2f08b65a | 1106 | |
f0f1cbaa KM |
1107 | /* |
1108 | * Check for server not responding | |
1109 | */ | |
1110 | if ((rep->r_flags & R_TPRINTFMSG) == 0 && | |
170bfd05 | 1111 | rep->r_rexmit > NFS_FISHY) { |
79993818 MK |
1112 | nfs_msg(rep->r_procp, |
1113 | nmp->nm_mountp->mnt_stat.f_mntfromname, | |
1114 | "not responding"); | |
f0f1cbaa KM |
1115 | rep->r_flags |= R_TPRINTFMSG; |
1116 | } | |
170bfd05 | 1117 | if (rep->r_rexmit >= rep->r_retry) { /* too many */ |
f0f1cbaa KM |
1118 | nfsstats.rpctimeouts++; |
1119 | rep->r_flags |= R_SOFTTERM; | |
1120 | continue; | |
1121 | } | |
170bfd05 KM |
1122 | if (nmp->nm_sotype != SOCK_DGRAM) |
1123 | continue; | |
f0f1cbaa KM |
1124 | |
1125 | /* | |
1126 | * If there is enough space and the window allows.. | |
1127 | * Resend it | |
1128 | */ | |
1129 | if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && | |
1130 | nmp->nm_sent < nmp->nm_window && | |
1131 | (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ | |
1132 | nfsstats.rpcretries++; | |
1133 | if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) | |
1134 | error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, | |
1135 | (caddr_t)0, (struct mbuf *)0, (struct mbuf *)0); | |
1136 | else | |
1137 | error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, | |
1138 | nmp->nm_nam, (struct mbuf *)0, (struct mbuf *)0); | |
1139 | if (error) { | |
1140 | if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) | |
1141 | so->so_error = 0; | |
1142 | } else { | |
1143 | /* | |
1144 | * We need to time the request even though we | |
1145 | * are retransmitting. | |
1146 | */ | |
1147 | nmp->nm_rtt = 0; | |
1148 | nmp->nm_sent++; | |
1149 | rep->r_flags |= (R_SENT|R_TIMING); | |
1150 | rep->r_timer = rep->r_timerinit; | |
1151 | } | |
1152 | } | |
2f08b65a KM |
1153 | } |
1154 | splx(s); | |
1155 | timeout(nfs_timer, (caddr_t)0, hz/NFS_HZ); | |
1156 | } | |
1157 | ||
1158 | /* | |
1159 | * NFS timer update and backoff. The "Jacobson/Karels/Karn" scheme is | |
1160 | * used here. The timer state is held in the nfsmount structure and | |
1161 | * a single request is used to clock the response. When successful | |
1162 | * the rtt smoothing in nfs_updatetimer is used, when failed the backoff | |
1163 | * is done by nfs_backofftimer. We also log failure messages in these | |
1164 | * routines. | |
1165 | * | |
1166 | * Congestion variables are held in the nfshost structure which | |
1167 | * is referenced by nfsmounts and shared per-server. This separation | |
1168 | * makes it possible to do per-mount timing which allows varying disk | |
1169 | * access times to be dealt with, while preserving a network oriented | |
1170 | * congestion control scheme. | |
1171 | * | |
1172 | * The windowing implements the Jacobson/Karels slowstart algorithm | |
1173 | * with adjusted scaling factors. We start with one request, then send | |
1174 | * 4 more after each success until the ssthresh limit is reached, then | |
1175 | * we increment at a rate proportional to the window. On failure, we | |
1176 | * remember 3/4 the current window and clamp the send limit to 1. Note | |
1177 | * ICMP source quench is not reflected in so->so_error so we ignore that | |
1178 | * for now. | |
1179 | * | |
1180 | * NFS behaves much more like a transport protocol with these changes, | |
1181 | * shedding the teenage pedal-to-the-metal tendencies of "other" | |
1182 | * implementations. | |
1183 | * | |
1184 | * Timers and congestion avoidance by Tom Talpey, Open Software Foundation. | |
1185 | */ | |
1186 | ||
1187 | /* | |
1188 | * The TCP algorithm was not forgiving enough. Because the NFS server | |
1189 | * responds only after performing lookups/diskio/etc, we have to be | |
1190 | * more prepared to accept a spiky variance. The TCP algorithm is: | |
f0f1cbaa | 1191 | * TCP_RTO(nmp) ((((nmp)->nm_srtt >> 2) + (nmp)->nm_rttvar) >> 1) |
2f08b65a | 1192 | */ |
f0f1cbaa | 1193 | #define NFS_RTO(nmp) (((nmp)->nm_srtt >> 3) + (nmp)->nm_rttvar) |
2f08b65a | 1194 | |
f0f1cbaa KM |
1195 | nfs_updatetimer(nmp) |
1196 | register struct nfsmount *nmp; | |
2f08b65a | 1197 | { |
2f08b65a KM |
1198 | |
1199 | /* If retransmitted, clear and return */ | |
f0f1cbaa KM |
1200 | if (nmp->nm_rexmit || nmp->nm_currexmit) { |
1201 | nmp->nm_rexmit = nmp->nm_currexmit = 0; | |
2f08b65a KM |
1202 | return; |
1203 | } | |
1204 | /* If have a measurement, do smoothing */ | |
f0f1cbaa | 1205 | if (nmp->nm_srtt) { |
2f08b65a | 1206 | register short delta; |
f0f1cbaa KM |
1207 | delta = nmp->nm_rtt - (nmp->nm_srtt >> 3); |
1208 | if ((nmp->nm_srtt += delta) <= 0) | |
1209 | nmp->nm_srtt = 1; | |
2f08b65a KM |
1210 | if (delta < 0) |
1211 | delta = -delta; | |
f0f1cbaa KM |
1212 | delta -= (nmp->nm_rttvar >> 2); |
1213 | if ((nmp->nm_rttvar += delta) <= 0) | |
1214 | nmp->nm_rttvar = 1; | |
2f08b65a KM |
1215 | /* Else initialize */ |
1216 | } else { | |
f0f1cbaa KM |
1217 | nmp->nm_rttvar = nmp->nm_rtt << 1; |
1218 | if (nmp->nm_rttvar == 0) nmp->nm_rttvar = 2; | |
1219 | nmp->nm_srtt = nmp->nm_rttvar << 2; | |
2f08b65a KM |
1220 | } |
1221 | /* Compute new Retransmission TimeOut and clip */ | |
f0f1cbaa KM |
1222 | nmp->nm_rto = NFS_RTO(nmp); |
1223 | if (nmp->nm_rto < NFS_MINTIMEO) | |
1224 | nmp->nm_rto = NFS_MINTIMEO; | |
1225 | else if (nmp->nm_rto > NFS_MAXTIMEO) | |
1226 | nmp->nm_rto = NFS_MAXTIMEO; | |
2f08b65a KM |
1227 | |
1228 | /* Update window estimate */ | |
f0f1cbaa KM |
1229 | if (nmp->nm_window < nmp->nm_ssthresh) /* quickly */ |
1230 | nmp->nm_window += 4; | |
2f08b65a | 1231 | else { /* slowly */ |
f0f1cbaa KM |
1232 | register long incr = ++nmp->nm_winext; |
1233 | incr = (incr * incr) / nmp->nm_window; | |
2f08b65a | 1234 | if (incr > 0) { |
f0f1cbaa KM |
1235 | nmp->nm_winext = 0; |
1236 | ++nmp->nm_window; | |
2f08b65a KM |
1237 | } |
1238 | } | |
f0f1cbaa KM |
1239 | if (nmp->nm_window > NFS_MAXWINDOW) |
1240 | nmp->nm_window = NFS_MAXWINDOW; | |
2f08b65a KM |
1241 | } |
1242 | ||
f0f1cbaa KM |
1243 | nfs_backofftimer(nmp) |
1244 | register struct nfsmount *nmp; | |
2f08b65a | 1245 | { |
2f08b65a KM |
1246 | register unsigned long newrto; |
1247 | ||
1248 | /* Clip shift count */ | |
f0f1cbaa KM |
1249 | if (++nmp->nm_rexmit > 8 * sizeof nmp->nm_rto) |
1250 | nmp->nm_rexmit = 8 * sizeof nmp->nm_rto; | |
2f08b65a | 1251 | /* Back off RTO exponentially */ |
f0f1cbaa KM |
1252 | newrto = NFS_RTO(nmp); |
1253 | newrto <<= (nmp->nm_rexmit - 1); | |
2f08b65a KM |
1254 | if (newrto == 0 || newrto > NFS_MAXTIMEO) |
1255 | newrto = NFS_MAXTIMEO; | |
f0f1cbaa | 1256 | nmp->nm_rto = newrto; |
2f08b65a KM |
1257 | |
1258 | /* If too many retries, message, assume a bogus RTT and re-measure */ | |
f0f1cbaa KM |
1259 | if (nmp->nm_currexmit < nmp->nm_rexmit) { |
1260 | nmp->nm_currexmit = nmp->nm_rexmit; | |
1261 | if (nmp->nm_currexmit >= nfsrexmtthresh) { | |
1262 | if (nmp->nm_currexmit == nfsrexmtthresh) { | |
1263 | nmp->nm_rttvar += (nmp->nm_srtt >> 2); | |
1264 | nmp->nm_srtt = 0; | |
a2907882 KM |
1265 | } |
1266 | } | |
a2907882 | 1267 | } |
2f08b65a | 1268 | /* Close down window but remember this point (3/4 current) for later */ |
f0f1cbaa KM |
1269 | nmp->nm_ssthresh = ((nmp->nm_window << 1) + nmp->nm_window) >> 2; |
1270 | nmp->nm_window = 1; | |
1271 | nmp->nm_winext = 0; | |
a2907882 KM |
1272 | } |
1273 | ||
1274 | /* | |
f0f1cbaa KM |
1275 | * Test for a termination signal pending on procp. |
1276 | * This is used for NFSMNT_INT mounts. | |
a2907882 | 1277 | */ |
f0f1cbaa KM |
1278 | nfs_sigintr(p) |
1279 | register struct proc *p; | |
1280 | { | |
1281 | if (p && p->p_sig && (((p->p_sig &~ p->p_sigmask) &~ p->p_sigignore) & | |
1282 | NFSINT_SIGMASK)) | |
1283 | return (1); | |
1284 | else | |
1285 | return (0); | |
1286 | } | |
2f08b65a | 1287 | |
79993818 MK |
1288 | nfs_msg(p, server, msg) |
1289 | struct proc *p; | |
1290 | char *server, *msg; | |
1291 | { | |
1292 | tpr_t tpr; | |
1293 | ||
1294 | if (p) | |
1295 | tpr = tprintf_open(p); | |
1296 | else | |
1297 | tpr = NULL; | |
1298 | tprintf(tpr, "nfs server %s: %s\n", server, msg); | |
1299 | tprintf_close(tpr); | |
1300 | } | |
1301 | ||
f0f1cbaa KM |
1302 | /* |
1303 | * Lock a socket against others. | |
1304 | * Necessary for STREAM sockets to ensure you get an entire rpc request/reply | |
1305 | * and also to avoid race conditions between the processes with nfs requests | |
1306 | * in progress when a reconnect is necessary. | |
1307 | */ | |
170bfd05 KM |
1308 | nfs_solock(flagp) |
1309 | register int *flagp; | |
a2907882 | 1310 | { |
2f08b65a | 1311 | |
f0f1cbaa KM |
1312 | while (*flagp & NFSMNT_SCKLOCK) { |
1313 | *flagp |= NFSMNT_WANTSCK; | |
170bfd05 | 1314 | (void) tsleep((caddr_t)flagp, PZERO-1, "nfsolck", 0); |
2f08b65a | 1315 | } |
f0f1cbaa KM |
1316 | *flagp |= NFSMNT_SCKLOCK; |
1317 | } | |
2f08b65a | 1318 | |
f0f1cbaa KM |
1319 | /* |
1320 | * Unlock the stream socket for others. | |
1321 | */ | |
1322 | nfs_sounlock(flagp) | |
170bfd05 | 1323 | register int *flagp; |
f0f1cbaa KM |
1324 | { |
1325 | ||
1326 | if ((*flagp & NFSMNT_SCKLOCK) == 0) | |
1327 | panic("nfs sounlock"); | |
1328 | *flagp &= ~NFSMNT_SCKLOCK; | |
1329 | if (*flagp & NFSMNT_WANTSCK) { | |
1330 | *flagp &= ~NFSMNT_WANTSCK; | |
1331 | wakeup((caddr_t)flagp); | |
2f08b65a | 1332 | } |
f0f1cbaa KM |
1333 | } |
1334 | ||
1335 | /* | |
1336 | * This function compares two net addresses by family and returns TRUE | |
1337 | * if they are the same. | |
1338 | * If there is any doubt, return FALSE. | |
1339 | */ | |
1340 | nfs_netaddr_match(nam1, nam2) | |
1341 | struct mbuf *nam1, *nam2; | |
1342 | { | |
1343 | register struct sockaddr *saddr1, *saddr2; | |
1344 | ||
1345 | saddr1 = mtod(nam1, struct sockaddr *); | |
1346 | saddr2 = mtod(nam2, struct sockaddr *); | |
1347 | if (saddr1->sa_family != saddr2->sa_family) | |
1348 | return (0); | |
1349 | ||
1350 | /* | |
1351 | * Must do each address family separately since unused fields | |
1352 | * are undefined values and not always zeroed. | |
1353 | */ | |
1354 | switch (saddr1->sa_family) { | |
1355 | case AF_INET: | |
1356 | if (((struct sockaddr_in *)saddr1)->sin_addr.s_addr == | |
1357 | ((struct sockaddr_in *)saddr2)->sin_addr.s_addr) | |
1358 | return (1); | |
1359 | break; | |
1360 | default: | |
1361 | break; | |
1362 | }; | |
1363 | return (0); | |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * Check the hostname fields for nfsd's mask and match fields. | |
1368 | * By address family: | |
1369 | * - Bitwise AND the mask with the host address field | |
1370 | * - Compare for == with match | |
1371 | * return TRUE if not equal | |
1372 | */ | |
1373 | nfs_badnam(nam, msk, mtch) | |
1374 | register struct mbuf *nam, *msk, *mtch; | |
1375 | { | |
1376 | switch (mtod(nam, struct sockaddr *)->sa_family) { | |
1377 | case AF_INET: | |
1378 | return ((mtod(nam, struct sockaddr_in *)->sin_addr.s_addr & | |
1379 | mtod(msk, struct sockaddr_in *)->sin_addr.s_addr) != | |
1380 | mtod(mtch, struct sockaddr_in *)->sin_addr.s_addr); | |
1381 | default: | |
1382 | printf("nfs_badmatch, unknown sa_family\n"); | |
1383 | return (0); | |
1384 | }; | |
a2907882 | 1385 | } |