Commit | Line | Data |
---|---|---|
a2907882 KM |
1 | /* |
2 | * Copyright (c) 1989 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 | * | |
dbf0c423 | 8 | * %sccs.include.redist.c% |
a2907882 | 9 | * |
958df9fb | 10 | * @(#)nfs_socket.c 7.19 (Berkeley) %G% |
a2907882 KM |
11 | */ |
12 | ||
13 | /* | |
f0f1cbaa | 14 | * Socket operations for use by nfs |
a2907882 KM |
15 | */ |
16 | ||
17 | #include "types.h" | |
18 | #include "param.h" | |
19 | #include "uio.h" | |
20 | #include "user.h" | |
2f08b65a KM |
21 | #include "proc.h" |
22 | #include "signal.h" | |
a2907882 KM |
23 | #include "mount.h" |
24 | #include "kernel.h" | |
25 | #include "malloc.h" | |
26 | #include "mbuf.h" | |
27 | #include "vnode.h" | |
28 | #include "domain.h" | |
29 | #include "protosw.h" | |
30 | #include "socket.h" | |
31 | #include "socketvar.h" | |
37ced908 KM |
32 | #include "../netinet/in.h" |
33 | #include "../netinet/tcp.h" | |
a2907882 KM |
34 | #include "rpcv2.h" |
35 | #include "nfsv2.h" | |
36 | #include "nfs.h" | |
37 | #include "xdr_subs.h" | |
38 | #include "nfsm_subs.h" | |
39 | #include "nfsmount.h" | |
40 | ||
2f08b65a | 41 | #include "syslog.h" |
2f08b65a | 42 | |
a2907882 | 43 | #define TRUE 1 |
170bfd05 | 44 | #define FALSE 0 |
a2907882 | 45 | |
a2907882 KM |
46 | /* |
47 | * External data, mostly RPC constants in XDR form | |
48 | */ | |
49 | extern u_long rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, rpc_auth_unix, | |
50 | rpc_msgaccepted, rpc_call; | |
51 | extern u_long nfs_prog, nfs_vers; | |
170bfd05 | 52 | /* Maybe these should be bits in a u_long ?? */ |
f0f1cbaa | 53 | extern int nonidempotent[NFS_NPROCS]; |
958df9fb KM |
54 | static int compressrequest[NFS_NPROCS] = { |
55 | FALSE, | |
56 | TRUE, | |
57 | TRUE, | |
58 | FALSE, | |
59 | TRUE, | |
60 | TRUE, | |
61 | TRUE, | |
62 | FALSE, | |
63 | FALSE, | |
64 | TRUE, | |
65 | TRUE, | |
66 | TRUE, | |
67 | TRUE, | |
68 | TRUE, | |
69 | TRUE, | |
70 | TRUE, | |
71 | TRUE, | |
72 | TRUE, | |
73 | }; | |
f0f1cbaa KM |
74 | int nfs_sbwait(); |
75 | void nfs_disconnect(); | |
958df9fb | 76 | struct mbuf *nfs_compress(), *nfs_uncompress(); |
f0f1cbaa | 77 | |
a2907882 KM |
78 | int nfsrv_null(), |
79 | nfsrv_getattr(), | |
80 | nfsrv_setattr(), | |
81 | nfsrv_lookup(), | |
82 | nfsrv_readlink(), | |
83 | nfsrv_read(), | |
84 | nfsrv_write(), | |
85 | nfsrv_create(), | |
86 | nfsrv_remove(), | |
87 | nfsrv_rename(), | |
88 | nfsrv_link(), | |
89 | nfsrv_symlink(), | |
90 | nfsrv_mkdir(), | |
91 | nfsrv_rmdir(), | |
92 | nfsrv_readdir(), | |
93 | nfsrv_statfs(), | |
94 | nfsrv_noop(); | |
95 | ||
96 | int (*nfsrv_procs[NFS_NPROCS])() = { | |
97 | nfsrv_null, | |
98 | nfsrv_getattr, | |
99 | nfsrv_setattr, | |
100 | nfsrv_noop, | |
101 | nfsrv_lookup, | |
102 | nfsrv_readlink, | |
103 | nfsrv_read, | |
104 | nfsrv_noop, | |
105 | nfsrv_write, | |
106 | nfsrv_create, | |
107 | nfsrv_remove, | |
108 | nfsrv_rename, | |
109 | nfsrv_link, | |
110 | nfsrv_symlink, | |
111 | nfsrv_mkdir, | |
112 | nfsrv_rmdir, | |
113 | nfsrv_readdir, | |
114 | nfsrv_statfs, | |
115 | }; | |
116 | ||
2f08b65a KM |
117 | struct nfsreq nfsreqh; |
118 | int nfsrexmtthresh = NFS_FISHY; | |
f0f1cbaa | 119 | int nfs_tcpnodelay = 1; |
2f08b65a KM |
120 | |
121 | /* | |
f0f1cbaa | 122 | * Initialize sockets and congestion for a new NFS connection. |
2f08b65a KM |
123 | * We do not free the sockaddr if error. |
124 | */ | |
f0f1cbaa | 125 | nfs_connect(nmp) |
2f08b65a | 126 | register struct nfsmount *nmp; |
2f08b65a | 127 | { |
f0f1cbaa KM |
128 | register struct socket *so; |
129 | int s, error; | |
2f08b65a | 130 | struct mbuf *m; |
2f08b65a | 131 | |
f0f1cbaa KM |
132 | nmp->nm_so = (struct socket *)0; |
133 | if (error = socreate(mtod(nmp->nm_nam, struct sockaddr *)->sa_family, | |
134 | &nmp->nm_so, nmp->nm_sotype, nmp->nm_soproto)) | |
2f08b65a | 135 | goto bad; |
f0f1cbaa KM |
136 | so = nmp->nm_so; |
137 | nmp->nm_soflags = so->so_proto->pr_flags; | |
2f08b65a | 138 | |
f0f1cbaa KM |
139 | /* |
140 | * Protocols that do not require connections may be optionally left | |
141 | * unconnected for servers that reply from a port other than NFS_PORT. | |
142 | */ | |
143 | if (nmp->nm_flag & NFSMNT_NOCONN) { | |
144 | if (nmp->nm_soflags & PR_CONNREQUIRED) { | |
145 | error = ENOTCONN; | |
2f08b65a KM |
146 | goto bad; |
147 | } | |
f0f1cbaa KM |
148 | } else { |
149 | if (error = soconnect(so, nmp->nm_nam)) | |
2f08b65a | 150 | goto bad; |
f0f1cbaa KM |
151 | |
152 | /* | |
153 | * Wait for the connection to complete. Cribbed from the | |
154 | * connect system call but with the wait at negative prio. | |
155 | */ | |
156 | s = splnet(); | |
157 | while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) | |
170bfd05 | 158 | (void) tsleep((caddr_t)&so->so_timeo, PSOCK, "nfscon", 0); |
f0f1cbaa KM |
159 | splx(s); |
160 | if (so->so_error) { | |
161 | error = so->so_error; | |
162 | goto bad; | |
163 | } | |
2f08b65a | 164 | } |
f0f1cbaa | 165 | if (nmp->nm_sotype == SOCK_DGRAM) { |
170bfd05 | 166 | if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_SPONGY | NFSMNT_INT)) { |
f0f1cbaa KM |
167 | so->so_rcv.sb_timeo = (5 * hz); |
168 | so->so_snd.sb_timeo = (5 * hz); | |
169 | } else { | |
170 | so->so_rcv.sb_timeo = 0; | |
171 | so->so_snd.sb_timeo = 0; | |
172 | } | |
173 | if (error = soreserve(so, nmp->nm_wsize + NFS_MAXPKTHDR, | |
170bfd05 | 174 | nmp->nm_rsize + NFS_MAXPKTHDR)) |
f0f1cbaa KM |
175 | goto bad; |
176 | } else { | |
170bfd05 | 177 | if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_SPONGY | NFSMNT_INT)) { |
f0f1cbaa KM |
178 | so->so_rcv.sb_timeo = (5 * hz); |
179 | so->so_snd.sb_timeo = (5 * hz); | |
180 | } else { | |
181 | so->so_rcv.sb_timeo = 0; | |
182 | so->so_snd.sb_timeo = 0; | |
183 | } | |
184 | if (so->so_proto->pr_flags & PR_CONNREQUIRED) { | |
185 | MGET(m, M_WAIT, MT_SOOPTS); | |
186 | *mtod(m, int *) = 1; | |
187 | m->m_len = sizeof(int); | |
188 | sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); | |
189 | } | |
190 | if (so->so_proto->pr_domain->dom_family == AF_INET && | |
191 | so->so_proto->pr_protocol == IPPROTO_TCP && | |
192 | nfs_tcpnodelay) { | |
193 | MGET(m, M_WAIT, MT_SOOPTS); | |
194 | *mtod(m, int *) = 1; | |
195 | m->m_len = sizeof(int); | |
196 | sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); | |
197 | } | |
198 | if (error = soreserve(so, | |
170bfd05 | 199 | nmp->nm_wsize + NFS_MAXPKTHDR + sizeof(u_long), |
f0f1cbaa KM |
200 | nmp->nm_rsize + NFS_MAXPKTHDR + sizeof(u_long))) |
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 | |
f0f1cbaa | 236 | if (rep->r_procp) |
5580a343 | 237 | tprintf(rep->r_procp->p_session, |
f0f1cbaa KM |
238 | "Nfs server %s, trying reconnect\n", |
239 | nmp->nm_mountp->mnt_stat.f_mntfromname); | |
240 | else | |
5580a343 | 241 | tprintf(NULL, "Nfs server %s, trying a reconnect\n", |
f0f1cbaa KM |
242 | nmp->nm_mountp->mnt_stat.f_mntfromname); |
243 | while (error = nfs_connect(nmp)) { | |
d4e5799e KM |
244 | #ifdef lint |
245 | error = error; | |
246 | #endif /* lint */ | |
f0f1cbaa KM |
247 | if ((nmp->nm_flag & NFSMNT_INT) && nfs_sigintr(rep->r_procp)) |
248 | return (EINTR); | |
170bfd05 | 249 | (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); |
2f08b65a | 250 | } |
f0f1cbaa | 251 | if (rep->r_procp) |
5580a343 | 252 | tprintf(rep->r_procp->p_session, |
f0f1cbaa KM |
253 | "Nfs server %s, reconnected\n", |
254 | nmp->nm_mountp->mnt_stat.f_mntfromname); | |
255 | else | |
5580a343 | 256 | tprintf(NULL, "Nfs server %s, reconnected\n", |
f0f1cbaa KM |
257 | nmp->nm_mountp->mnt_stat.f_mntfromname); |
258 | ||
259 | /* | |
260 | * Loop through outstanding request list and fix up all requests | |
261 | * on old socket. | |
262 | */ | |
263 | rp = nfsreqh.r_next; | |
264 | while (rp != &nfsreqh) { | |
265 | if (rp->r_nmp == nmp) | |
266 | rp->r_flags |= R_MUSTRESEND; | |
267 | rp = rp->r_next; | |
2f08b65a KM |
268 | } |
269 | return (0); | |
2f08b65a KM |
270 | } |
271 | ||
272 | /* | |
273 | * NFS disconnect. Clean up and unlink. | |
274 | */ | |
f0f1cbaa | 275 | void |
2f08b65a KM |
276 | nfs_disconnect(nmp) |
277 | register struct nfsmount *nmp; | |
278 | { | |
f0f1cbaa | 279 | register struct socket *so; |
2f08b65a | 280 | |
f0f1cbaa KM |
281 | if (nmp->nm_so) { |
282 | so = nmp->nm_so; | |
283 | nmp->nm_so = (struct socket *)0; | |
284 | soshutdown(so, 2); | |
285 | soclose(so); | |
2f08b65a KM |
286 | } |
287 | } | |
a2907882 KM |
288 | |
289 | /* | |
f0f1cbaa KM |
290 | * This is the nfs send routine. For connection based socket types, it |
291 | * must be called with an nfs_solock() on the socket. | |
292 | * "rep == NULL" indicates that it has been called from a server. | |
a2907882 | 293 | */ |
f0f1cbaa | 294 | nfs_send(so, nam, top, rep) |
a2907882 KM |
295 | register struct socket *so; |
296 | struct mbuf *nam; | |
f0f1cbaa KM |
297 | register struct mbuf *top; |
298 | struct nfsreq *rep; | |
a2907882 | 299 | { |
f0f1cbaa KM |
300 | struct mbuf *sendnam; |
301 | int error, soflags; | |
a2907882 | 302 | |
f0f1cbaa KM |
303 | if (rep) { |
304 | if (rep->r_flags & R_SOFTTERM) { | |
2f08b65a | 305 | m_freem(top); |
f0f1cbaa | 306 | return (EINTR); |
2f08b65a | 307 | } |
5044b7a3 | 308 | if (rep->r_nmp->nm_so == NULL && |
f0f1cbaa KM |
309 | (error = nfs_reconnect(rep, rep->r_nmp))) |
310 | return (error); | |
311 | rep->r_flags &= ~R_MUSTRESEND; | |
5044b7a3 | 312 | so = rep->r_nmp->nm_so; |
f0f1cbaa KM |
313 | soflags = rep->r_nmp->nm_soflags; |
314 | } else | |
315 | soflags = so->so_proto->pr_flags; | |
316 | if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) | |
317 | sendnam = (struct mbuf *)0; | |
318 | else | |
319 | sendnam = nam; | |
320 | ||
321 | error = sosend(so, sendnam, (struct uio *)0, top, | |
322 | (struct mbuf *)0, 0); | |
323 | if (error == EWOULDBLOCK && rep) { | |
324 | if (rep->r_flags & R_SOFTTERM) | |
325 | error = EINTR; | |
326 | else { | |
327 | rep->r_flags |= R_MUSTRESEND; | |
328 | error = 0; | |
2f08b65a | 329 | } |
a2907882 | 330 | } |
f0f1cbaa KM |
331 | /* |
332 | * Ignore socket errors?? | |
333 | */ | |
334 | if (error && error != EINTR && error != ERESTART) | |
335 | error = 0; | |
a2907882 KM |
336 | return (error); |
337 | } | |
338 | ||
339 | /* | |
f0f1cbaa KM |
340 | * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all |
341 | * done by soreceive(), but for SOCK_STREAM we must deal with the Record | |
342 | * Mark and consolidate the data into a new mbuf list. | |
343 | * nb: Sometimes TCP passes the data up to soreceive() in long lists of | |
344 | * small mbufs. | |
345 | * For SOCK_STREAM we must be very careful to read an entire record once | |
346 | * we have read any of it, even if the system call has been interrupted. | |
a2907882 | 347 | */ |
f0f1cbaa | 348 | nfs_receive(so, aname, mp, rep) |
a2907882 KM |
349 | register struct socket *so; |
350 | struct mbuf **aname; | |
351 | struct mbuf **mp; | |
f0f1cbaa | 352 | register struct nfsreq *rep; |
a2907882 | 353 | { |
f0f1cbaa KM |
354 | struct uio auio; |
355 | struct iovec aio; | |
a2907882 | 356 | register struct mbuf *m; |
958df9fb | 357 | struct mbuf *m2, *mnew, **mbp; |
f0f1cbaa KM |
358 | caddr_t fcp, tcp; |
359 | u_long len; | |
360 | struct mbuf **getnam; | |
361 | int error, siz, mlen, soflags, rcvflg = MSG_WAITALL; | |
a2907882 | 362 | |
f0f1cbaa KM |
363 | /* |
364 | * Set up arguments for soreceive() | |
365 | */ | |
366 | *mp = (struct mbuf *)0; | |
367 | *aname = (struct mbuf *)0; | |
368 | if (rep) | |
369 | soflags = rep->r_nmp->nm_soflags; | |
370 | else | |
371 | soflags = so->so_proto->pr_flags; | |
a2907882 | 372 | |
f0f1cbaa KM |
373 | /* |
374 | * For reliable protocols, lock against other senders/receivers | |
375 | * in case a reconnect is necessary. | |
376 | * For SOCK_STREAM, first get the Record Mark to find out how much | |
377 | * more there is to get. | |
378 | * We must lock the socket against other receivers | |
379 | * until we have an entire rpc request/reply. | |
380 | */ | |
381 | if (soflags & PR_CONNREQUIRED) { | |
382 | tryagain: | |
383 | /* | |
384 | * Check for fatal errors and resending request. | |
385 | */ | |
386 | if (rep) { | |
387 | /* | |
388 | * Ugh: If a reconnect attempt just happened, nm_so | |
389 | * would have changed. NULL indicates a failed | |
390 | * attempt that has essentially shut down this | |
391 | * mount point. | |
392 | */ | |
393 | if (rep->r_mrep || (so = rep->r_nmp->nm_so) == NULL || | |
394 | (rep->r_flags & R_SOFTTERM)) | |
395 | return (EINTR); | |
396 | while (rep->r_flags & R_MUSTRESEND) { | |
397 | m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); | |
398 | nfsstats.rpcretries++; | |
399 | if (error = nfs_send(so, rep->r_nmp->nm_nam, m, | |
400 | rep)) | |
401 | goto errout; | |
2f08b65a | 402 | } |
e8540f59 | 403 | } |
f0f1cbaa KM |
404 | if ((soflags & PR_ATOMIC) == 0) { |
405 | aio.iov_base = (caddr_t) &len; | |
406 | aio.iov_len = sizeof(u_long); | |
407 | auio.uio_iov = &aio; | |
408 | auio.uio_iovcnt = 1; | |
409 | auio.uio_segflg = UIO_SYSSPACE; | |
410 | auio.uio_rw = UIO_READ; | |
411 | auio.uio_offset = 0; | |
412 | auio.uio_resid = sizeof(u_long); | |
413 | do { | |
414 | error = soreceive(so, (struct mbuf **)0, &auio, | |
415 | (struct mbuf **)0, (struct mbuf **)0, &rcvflg); | |
416 | if (error == EWOULDBLOCK && rep) { | |
417 | if (rep->r_flags & R_SOFTTERM) | |
418 | return (EINTR); | |
419 | if (rep->r_flags & R_MUSTRESEND) | |
420 | goto tryagain; | |
421 | } | |
422 | } while (error == EWOULDBLOCK); | |
423 | if (!error && auio.uio_resid > 0) | |
424 | error = EPIPE; | |
425 | if (error) | |
426 | goto errout; | |
427 | len = ntohl(len) & ~0x80000000; | |
428 | /* | |
429 | * This is SERIOUS! We are out of sync with the sender | |
430 | * and forcing a disconnect/reconnect is all I can do. | |
431 | */ | |
432 | if (len > NFS_MAXPACKET) { | |
433 | error = EFBIG; | |
434 | goto errout; | |
435 | } | |
436 | auio.uio_resid = len; | |
437 | do { | |
438 | error = soreceive(so, (struct mbuf **)0, | |
439 | &auio, mp, (struct mbuf **)0, &rcvflg); | |
440 | } while (error == EWOULDBLOCK || error == EINTR || | |
441 | error == ERESTART); | |
442 | if (!error && auio.uio_resid > 0) | |
443 | error = EPIPE; | |
2f08b65a | 444 | } else { |
f0f1cbaa KM |
445 | auio.uio_resid = len = 1000000; /* Anything Big */ |
446 | do { | |
447 | error = soreceive(so, (struct mbuf **)0, | |
448 | &auio, mp, (struct mbuf **)0, &rcvflg); | |
449 | if (error == EWOULDBLOCK && rep) { | |
450 | if (rep->r_flags & R_SOFTTERM) | |
451 | return (EINTR); | |
452 | if (rep->r_flags & R_MUSTRESEND) | |
453 | goto tryagain; | |
454 | } | |
455 | } while (error == EWOULDBLOCK); | |
456 | if (!error && *mp == NULL) | |
457 | error = EPIPE; | |
458 | len -= auio.uio_resid; | |
2f08b65a | 459 | } |
f0f1cbaa KM |
460 | errout: |
461 | if (error && rep && error != EINTR && error != ERESTART) { | |
462 | m_freem(*mp); | |
463 | *mp = (struct mbuf *)0; | |
464 | nfs_disconnect(rep->r_nmp); | |
465 | error = nfs_reconnect(rep, rep->r_nmp); | |
466 | if (!error) | |
467 | goto tryagain; | |
2f08b65a | 468 | } |
f0f1cbaa KM |
469 | } else { |
470 | if (so->so_state & SS_ISCONNECTED) | |
471 | getnam = (struct mbuf **)0; | |
472 | else | |
473 | getnam = aname; | |
474 | auio.uio_resid = len = 1000000; | |
475 | do { | |
476 | error = soreceive(so, getnam, &auio, mp, | |
477 | (struct mbuf **)0, &rcvflg); | |
478 | if (error == EWOULDBLOCK && rep && | |
479 | (rep->r_flags & R_SOFTTERM)) | |
480 | return (EINTR); | |
481 | } while (error == EWOULDBLOCK); | |
482 | len -= auio.uio_resid; | |
483 | } | |
484 | if (error) { | |
485 | m_freem(*mp); | |
486 | *mp = (struct mbuf *)0; | |
487 | } | |
488 | /* | |
489 | * Search for any mbufs that are not a multiple of 4 bytes long. | |
490 | * These could cause pointer alignment problems, so copy them to | |
491 | * well aligned mbufs. | |
492 | */ | |
493 | m = *mp; | |
494 | mbp = mp; | |
495 | while (m) { | |
496 | /* | |
497 | * All this for something that may never happen. | |
498 | */ | |
958df9fb | 499 | if (m->m_next && (m->m_len & 0x3)) { |
f0f1cbaa | 500 | printf("nfs_rcv odd length!\n"); |
d4e5799e | 501 | mlen = 0; |
f0f1cbaa | 502 | while (m) { |
958df9fb KM |
503 | fcp = mtod(m, caddr_t); |
504 | while (m->m_len > 0) { | |
505 | if (mlen == 0) { | |
506 | MGET(m2, M_WAIT, MT_DATA); | |
507 | if (len >= MINCLSIZE) | |
508 | MCLGET(m2, M_WAIT); | |
509 | m2->m_len = 0; | |
510 | mlen = M_TRAILINGSPACE(m2); | |
511 | tcp = mtod(m2, caddr_t); | |
512 | *mbp = m2; | |
513 | mbp = &m2->m_next; | |
514 | } | |
515 | siz = MIN(mlen, m->m_len); | |
516 | bcopy(fcp, tcp, siz); | |
517 | m2->m_len += siz; | |
518 | mlen -= siz; | |
519 | len -= siz; | |
520 | tcp += siz; | |
521 | m->m_len -= siz; | |
522 | fcp += siz; | |
f0f1cbaa | 523 | } |
958df9fb KM |
524 | MFREE(m, mnew); |
525 | m = mnew; | |
f0f1cbaa | 526 | } |
f0f1cbaa | 527 | break; |
2f08b65a | 528 | } |
f0f1cbaa KM |
529 | len -= m->m_len; |
530 | mbp = &m->m_next; | |
531 | m = m->m_next; | |
a2907882 | 532 | } |
a2907882 KM |
533 | return (error); |
534 | } | |
535 | ||
a2907882 | 536 | /* |
f0f1cbaa | 537 | * Implement receipt of reply on a socket. |
a2907882 KM |
538 | * We must search through the list of received datagrams matching them |
539 | * with outstanding requests using the xid, until ours is found. | |
540 | */ | |
f0f1cbaa KM |
541 | /* ARGSUSED */ |
542 | nfs_reply(nmp, myrep) | |
543 | struct nfsmount *nmp; | |
ffe6f482 | 544 | struct nfsreq *myrep; |
a2907882 KM |
545 | { |
546 | register struct mbuf *m; | |
547 | register struct nfsreq *rep; | |
f0f1cbaa | 548 | register int error = 0; |
958df9fb | 549 | u_long rxid; |
f0f1cbaa KM |
550 | struct mbuf *mp, *nam; |
551 | char *cp; | |
552 | int cnt, xfer; | |
a2907882 KM |
553 | |
554 | /* | |
f0f1cbaa | 555 | * Loop around until we get our own reply |
a2907882 | 556 | */ |
f0f1cbaa KM |
557 | for (;;) { |
558 | /* | |
559 | * Lock against other receivers so that I don't get stuck in | |
560 | * sbwait() after someone else has received my reply for me. | |
561 | * Also necessary for connection based protocols to avoid | |
562 | * race conditions during a reconnect. | |
563 | */ | |
170bfd05 | 564 | nfs_solock(&nmp->nm_flag); |
f0f1cbaa KM |
565 | /* Already received, bye bye */ |
566 | if (myrep->r_mrep != NULL) { | |
567 | nfs_sounlock(&nmp->nm_flag); | |
568 | return (0); | |
569 | } | |
570 | /* | |
571 | * Get the next Rpc reply off the socket | |
572 | */ | |
573 | if (error = nfs_receive(nmp->nm_so, &nam, &mp, myrep)) { | |
574 | nfs_sounlock(&nmp->nm_flag); | |
a2907882 | 575 | |
f0f1cbaa KM |
576 | /* |
577 | * Ignore routing errors on connectionless protocols?? | |
578 | */ | |
579 | if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { | |
580 | nmp->nm_so->so_error = 0; | |
581 | continue; | |
a2907882 | 582 | } |
f0f1cbaa KM |
583 | |
584 | /* | |
585 | * Otherwise cleanup and return a fatal error. | |
586 | */ | |
587 | if (myrep->r_flags & R_TIMING) { | |
588 | myrep->r_flags &= ~R_TIMING; | |
589 | nmp->nm_rtt = -1; | |
a2907882 | 590 | } |
f0f1cbaa KM |
591 | if (myrep->r_flags & R_SENT) { |
592 | myrep->r_flags &= ~R_SENT; | |
593 | nmp->nm_sent--; | |
2f08b65a | 594 | } |
f0f1cbaa KM |
595 | return (error); |
596 | } | |
597 | ||
598 | /* | |
599 | * Get the xid and check that it is an rpc reply | |
600 | */ | |
601 | m = mp; | |
958df9fb KM |
602 | while (m && m->m_len == 0) |
603 | m = m->m_next; | |
604 | if (m == NULL) { | |
f0f1cbaa KM |
605 | nfsstats.rpcinvalid++; |
606 | m_freem(mp); | |
607 | nfs_sounlock(&nmp->nm_flag); | |
608 | continue; | |
609 | } | |
958df9fb | 610 | bcopy(mtod(m, caddr_t), (caddr_t)&rxid, NFSX_UNSIGNED); |
f0f1cbaa KM |
611 | /* |
612 | * Loop through the request list to match up the reply | |
613 | * Iff no match, just drop the datagram | |
614 | */ | |
615 | m = mp; | |
616 | rep = nfsreqh.r_next; | |
617 | while (rep != &nfsreqh) { | |
958df9fb | 618 | if (rep->r_mrep == NULL && rxid == rep->r_xid) { |
f0f1cbaa KM |
619 | /* Found it.. */ |
620 | rep->r_mrep = m; | |
621 | /* | |
622 | * Update timing | |
623 | */ | |
624 | if (rep->r_flags & R_TIMING) { | |
625 | nfs_updatetimer(rep->r_nmp); | |
626 | rep->r_flags &= ~R_TIMING; | |
627 | rep->r_nmp->nm_rtt = -1; | |
628 | } | |
629 | if (rep->r_flags & R_SENT) { | |
630 | rep->r_flags &= ~R_SENT; | |
631 | rep->r_nmp->nm_sent--; | |
632 | } | |
633 | break; | |
2f08b65a | 634 | } |
f0f1cbaa | 635 | rep = rep->r_next; |
a2907882 | 636 | } |
f0f1cbaa KM |
637 | nfs_sounlock(&nmp->nm_flag); |
638 | if (nam) | |
639 | m_freem(nam); | |
640 | /* | |
641 | * If not matched to a request, drop it. | |
642 | * If it's mine, get out. | |
643 | */ | |
644 | if (rep == &nfsreqh) { | |
645 | nfsstats.rpcunexpected++; | |
646 | m_freem(m); | |
647 | } else if (rep == myrep) | |
648 | return (0); | |
a2907882 | 649 | } |
a2907882 KM |
650 | } |
651 | ||
652 | /* | |
653 | * nfs_request - goes something like this | |
654 | * - fill in request struct | |
655 | * - links it into list | |
f0f1cbaa KM |
656 | * - calls nfs_send() for first transmit |
657 | * - calls nfs_receive() to get reply | |
a2907882 KM |
658 | * - break down rpc header and return with nfs reply pointed to |
659 | * by mrep or error | |
660 | * nb: always frees up mreq mbuf list | |
661 | */ | |
170bfd05 | 662 | nfs_request(vp, mreq, xid, procnum, procp, tryhard, mp, mrp, mdp, dposp) |
a2907882 KM |
663 | struct vnode *vp; |
664 | struct mbuf *mreq; | |
665 | u_long xid; | |
f0f1cbaa KM |
666 | int procnum; |
667 | struct proc *procp; | |
170bfd05 | 668 | int tryhard; |
a2907882 KM |
669 | struct mount *mp; |
670 | struct mbuf **mrp; | |
671 | struct mbuf **mdp; | |
672 | caddr_t *dposp; | |
673 | { | |
674 | register struct mbuf *m, *mrep; | |
675 | register struct nfsreq *rep; | |
676 | register u_long *p; | |
677 | register int len; | |
f0f1cbaa | 678 | struct nfsmount *nmp; |
a2907882 | 679 | struct mbuf *md; |
ffe6f482 | 680 | struct nfsreq *reph; |
a2907882 KM |
681 | caddr_t dpos; |
682 | char *cp2; | |
683 | int t1; | |
958df9fb | 684 | int s, compressed; |
f0f1cbaa | 685 | int error = 0; |
a2907882 | 686 | |
f0f1cbaa | 687 | nmp = VFSTONFS(mp); |
a2907882 KM |
688 | m = mreq; |
689 | MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); | |
690 | rep->r_xid = xid; | |
f0f1cbaa | 691 | rep->r_nmp = nmp; |
a2907882 | 692 | rep->r_vp = vp; |
f0f1cbaa | 693 | rep->r_procp = procp; |
170bfd05 KM |
694 | if ((nmp->nm_flag & NFSMNT_SOFT) || |
695 | ((nmp->nm_flag & NFSMNT_SPONGY) && !tryhard)) | |
f0f1cbaa | 696 | rep->r_retry = nmp->nm_retry; |
a2907882 | 697 | else |
2f08b65a KM |
698 | rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ |
699 | rep->r_flags = rep->r_rexmit = 0; | |
f0f1cbaa KM |
700 | /* |
701 | * Three cases: | |
702 | * - non-idempotent requests on SOCK_DGRAM use NFS_MINIDEMTIMEO | |
703 | * - idempotent requests on SOCK_DGRAM use 0 | |
704 | * - Reliable transports, NFS_RELIABLETIMEO | |
705 | * Timeouts are still done on reliable transports to ensure detection | |
170bfd05 | 706 | * of excessive connection delay. |
f0f1cbaa KM |
707 | */ |
708 | if (nmp->nm_sotype != SOCK_DGRAM) | |
709 | rep->r_timerinit = -NFS_RELIABLETIMEO; | |
710 | else if (nonidempotent[procnum]) | |
711 | rep->r_timerinit = -NFS_MINIDEMTIMEO; | |
712 | else | |
713 | rep->r_timerinit = 0; | |
714 | rep->r_timer = rep->r_timerinit; | |
a2907882 | 715 | rep->r_mrep = NULL; |
a2907882 KM |
716 | len = 0; |
717 | while (m) { | |
718 | len += m->m_len; | |
719 | m = m->m_next; | |
720 | } | |
f0f1cbaa KM |
721 | mreq->m_pkthdr.len = len; |
722 | mreq->m_pkthdr.rcvif = (struct ifnet *)0; | |
958df9fb KM |
723 | compressed = 0; |
724 | m = mreq; | |
725 | if ((nmp->nm_flag & NFSMNT_COMPRESS) && compressrequest[procnum]) { | |
726 | mreq = nfs_compress(mreq); | |
727 | if (mreq != m) { | |
728 | len = mreq->m_pkthdr.len; | |
729 | compressed++; | |
730 | } | |
731 | } | |
f0f1cbaa KM |
732 | /* |
733 | * For non-atomic protocols, insert a Sun RPC Record Mark. | |
734 | */ | |
735 | if ((nmp->nm_soflags & PR_ATOMIC) == 0) { | |
736 | M_PREPEND(mreq, sizeof(u_long), M_WAIT); | |
737 | *mtod(mreq, u_long *) = htonl(0x80000000 | len); | |
738 | } | |
739 | rep->r_mreq = mreq; | |
a2907882 | 740 | |
2f08b65a KM |
741 | /* |
742 | * Do the client side RPC. | |
743 | */ | |
744 | nfsstats.rpcrequests++; | |
f0f1cbaa KM |
745 | /* |
746 | * Chain request into list of outstanding requests. Be sure | |
747 | * to put it LAST so timer finds oldest requests first. | |
748 | */ | |
a2907882 | 749 | s = splnet(); |
2f08b65a | 750 | reph = &nfsreqh; |
f0f1cbaa KM |
751 | reph->r_prev->r_next = rep; |
752 | rep->r_prev = reph->r_prev; | |
ffe6f482 KM |
753 | reph->r_prev = rep; |
754 | rep->r_next = reph; | |
2f08b65a KM |
755 | /* |
756 | * If backing off another request or avoiding congestion, don't | |
757 | * send this one now but let timer do it. If not timing a request, | |
758 | * do it now. | |
759 | */ | |
f0f1cbaa KM |
760 | if (nmp->nm_sent <= 0 || nmp->nm_sotype != SOCK_DGRAM || |
761 | (nmp->nm_currexmit == 0 && nmp->nm_sent < nmp->nm_window)) { | |
762 | nmp->nm_sent++; | |
763 | rep->r_flags |= R_SENT; | |
764 | if (nmp->nm_rtt == -1) { | |
765 | nmp->nm_rtt = 0; | |
766 | rep->r_flags |= R_TIMING; | |
767 | } | |
768 | splx(s); | |
769 | m = m_copym(mreq, 0, M_COPYALL, M_WAIT); | |
770 | if (nmp->nm_soflags & PR_CONNREQUIRED) | |
170bfd05 | 771 | nfs_solock(&nmp->nm_flag); |
f0f1cbaa KM |
772 | error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); |
773 | if (nmp->nm_soflags & PR_CONNREQUIRED) | |
774 | nfs_sounlock(&nmp->nm_flag); | |
775 | if (error && NFSIGNORE_SOERROR(nmp->nm_soflags, error)) | |
776 | nmp->nm_so->so_error = error = 0; | |
777 | } else | |
2f08b65a | 778 | splx(s); |
a2907882 | 779 | |
2f08b65a KM |
780 | /* |
781 | * Wait for the reply from our send or the timer's. | |
782 | */ | |
f0f1cbaa KM |
783 | if (!error) |
784 | error = nfs_reply(nmp, rep); | |
a2907882 | 785 | |
2f08b65a KM |
786 | /* |
787 | * RPC done, unlink the request. | |
788 | */ | |
a2907882 KM |
789 | s = splnet(); |
790 | rep->r_prev->r_next = rep->r_next; | |
ffe6f482 | 791 | rep->r_next->r_prev = rep->r_prev; |
a2907882 | 792 | splx(s); |
f0f1cbaa KM |
793 | |
794 | /* | |
795 | * If there was a successful reply and a tprintf msg. | |
796 | * tprintf a response. | |
797 | */ | |
798 | if (!error && (rep->r_flags & R_TPRINTFMSG)) { | |
799 | if (rep->r_procp) | |
5580a343 | 800 | tprintf(rep->r_procp->p_session, |
f0f1cbaa KM |
801 | "Nfs server %s, is alive again\n", |
802 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); | |
803 | else | |
5580a343 | 804 | tprintf(NULL, "Nfs server %s, is alive again\n", |
f0f1cbaa KM |
805 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); |
806 | } | |
a2907882 | 807 | m_freem(rep->r_mreq); |
958df9fb | 808 | mrep = rep->r_mrep; |
a2907882 KM |
809 | FREE((caddr_t)rep, M_NFSREQ); |
810 | if (error) | |
811 | return (error); | |
812 | ||
958df9fb KM |
813 | if (compressed) |
814 | mrep = nfs_uncompress(mrep); | |
815 | md = mrep; | |
a2907882 KM |
816 | /* |
817 | * break down the rpc header and check if ok | |
818 | */ | |
819 | dpos = mtod(md, caddr_t); | |
820 | nfsm_disect(p, u_long *, 5*NFSX_UNSIGNED); | |
821 | p += 2; | |
822 | if (*p++ == rpc_msgdenied) { | |
823 | if (*p == rpc_mismatch) | |
824 | error = EOPNOTSUPP; | |
825 | else | |
826 | error = EACCES; | |
827 | m_freem(mrep); | |
828 | return (error); | |
829 | } | |
830 | /* | |
831 | * skip over the auth_verf, someday we may want to cache auth_short's | |
832 | * for nfs_reqhead(), but for now just dump it | |
833 | */ | |
834 | if (*++p != 0) { | |
835 | len = nfsm_rndup(fxdr_unsigned(long, *p)); | |
836 | nfsm_adv(len); | |
837 | } | |
838 | nfsm_disect(p, u_long *, NFSX_UNSIGNED); | |
839 | /* 0 == ok */ | |
840 | if (*p == 0) { | |
841 | nfsm_disect(p, u_long *, NFSX_UNSIGNED); | |
842 | if (*p != 0) { | |
843 | error = fxdr_unsigned(int, *p); | |
844 | m_freem(mrep); | |
845 | return (error); | |
846 | } | |
847 | *mrp = mrep; | |
848 | *mdp = md; | |
849 | *dposp = dpos; | |
850 | return (0); | |
851 | } | |
852 | m_freem(mrep); | |
853 | return (EPROTONOSUPPORT); | |
854 | nfsmout: | |
855 | return (error); | |
856 | } | |
857 | ||
858 | /* | |
859 | * Get a request for the server main loop | |
860 | * - receive a request via. nfs_soreceive() | |
861 | * - verify it | |
862 | * - fill in the cred struct. | |
863 | */ | |
d4e5799e | 864 | nfs_getreq(so, prog, vers, maxproc, nam, mrp, mdp, dposp, retxid, procnum, cr, |
958df9fb | 865 | msk, mtch, wascomp) |
a2907882 KM |
866 | struct socket *so; |
867 | u_long prog; | |
868 | u_long vers; | |
869 | int maxproc; | |
870 | struct mbuf **nam; | |
871 | struct mbuf **mrp; | |
872 | struct mbuf **mdp; | |
873 | caddr_t *dposp; | |
874 | u_long *retxid; | |
d4e5799e | 875 | u_long *procnum; |
a2907882 | 876 | register struct ucred *cr; |
f0f1cbaa | 877 | struct mbuf *msk, *mtch; |
958df9fb | 878 | int *wascomp; |
a2907882 KM |
879 | { |
880 | register int i; | |
0bd503ad KM |
881 | register u_long *p; |
882 | register long t1; | |
883 | caddr_t dpos, cp2; | |
884 | int error = 0; | |
885 | struct mbuf *mrep, *md; | |
886 | int len; | |
a2907882 | 887 | |
f0f1cbaa | 888 | if (so->so_proto->pr_flags & PR_CONNREQUIRED) { |
f0f1cbaa | 889 | error = nfs_receive(so, nam, &mrep, (struct nfsreq *)0); |
f0f1cbaa KM |
890 | } else { |
891 | mrep = (struct mbuf *)0; | |
892 | do { | |
893 | if (mrep) { | |
894 | m_freem(*nam); | |
895 | m_freem(mrep); | |
896 | } | |
897 | error = nfs_receive(so, nam, &mrep, (struct nfsreq *)0); | |
898 | } while (!error && nfs_badnam(*nam, msk, mtch)); | |
899 | } | |
900 | if (error) | |
a2907882 KM |
901 | return (error); |
902 | md = mrep; | |
958df9fb KM |
903 | mrep = nfs_uncompress(mrep); |
904 | if (mrep != md) { | |
905 | *wascomp = 1; | |
906 | md = mrep; | |
907 | } else | |
908 | *wascomp = 0; | |
a2907882 KM |
909 | dpos = mtod(mrep, caddr_t); |
910 | nfsm_disect(p, u_long *, 10*NFSX_UNSIGNED); | |
911 | *retxid = *p++; | |
912 | if (*p++ != rpc_call) { | |
913 | m_freem(mrep); | |
914 | return (ERPCMISMATCH); | |
915 | } | |
916 | if (*p++ != rpc_vers) { | |
917 | m_freem(mrep); | |
918 | return (ERPCMISMATCH); | |
919 | } | |
920 | if (*p++ != prog) { | |
921 | m_freem(mrep); | |
922 | return (EPROGUNAVAIL); | |
923 | } | |
924 | if (*p++ != vers) { | |
925 | m_freem(mrep); | |
926 | return (EPROGMISMATCH); | |
927 | } | |
d4e5799e KM |
928 | *procnum = fxdr_unsigned(u_long, *p++); |
929 | if (*procnum == NFSPROC_NULL) { | |
a2907882 KM |
930 | *mrp = mrep; |
931 | return (0); | |
932 | } | |
d4e5799e | 933 | if (*procnum > maxproc || *p++ != rpc_auth_unix) { |
a2907882 KM |
934 | m_freem(mrep); |
935 | return (EPROCUNAVAIL); | |
936 | } | |
f0f1cbaa KM |
937 | len = fxdr_unsigned(int, *p++); |
938 | if (len < 0 || len > RPCAUTH_MAXSIZ) { | |
939 | m_freem(mrep); | |
940 | return (EBADRPC); | |
941 | } | |
0bd503ad | 942 | len = fxdr_unsigned(int, *++p); |
f0f1cbaa KM |
943 | if (len < 0 || len > NFS_MAXNAMLEN) { |
944 | m_freem(mrep); | |
945 | return (EBADRPC); | |
946 | } | |
0bd503ad | 947 | nfsm_adv(nfsm_rndup(len)); |
a2907882 KM |
948 | nfsm_disect(p, u_long *, 3*NFSX_UNSIGNED); |
949 | cr->cr_uid = fxdr_unsigned(uid_t, *p++); | |
950 | cr->cr_gid = fxdr_unsigned(gid_t, *p++); | |
0bd503ad | 951 | len = fxdr_unsigned(int, *p); |
f0f1cbaa | 952 | if (len < 0 || len > RPCAUTH_UNIXGIDS) { |
a2907882 KM |
953 | m_freem(mrep); |
954 | return (EBADRPC); | |
955 | } | |
0bd503ad KM |
956 | nfsm_disect(p, u_long *, (len + 2)*NFSX_UNSIGNED); |
957 | for (i = 1; i <= len; i++) | |
f0f1cbaa KM |
958 | if (i < NGROUPS) |
959 | cr->cr_groups[i] = fxdr_unsigned(gid_t, *p++); | |
960 | else | |
961 | p++; | |
962 | cr->cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1); | |
a2907882 KM |
963 | /* |
964 | * Do we have any use for the verifier. | |
965 | * According to the "Remote Procedure Call Protocol Spec." it | |
966 | * should be AUTH_NULL, but some clients make it AUTH_UNIX? | |
967 | * For now, just skip over it | |
968 | */ | |
0bd503ad | 969 | len = fxdr_unsigned(int, *++p); |
f0f1cbaa KM |
970 | if (len < 0 || len > RPCAUTH_MAXSIZ) { |
971 | m_freem(mrep); | |
972 | return (EBADRPC); | |
973 | } | |
0bd503ad KM |
974 | if (len > 0) |
975 | nfsm_adv(nfsm_rndup(len)); | |
a2907882 KM |
976 | *mrp = mrep; |
977 | *mdp = md; | |
978 | *dposp = dpos; | |
979 | return (0); | |
980 | nfsmout: | |
981 | return (error); | |
982 | } | |
983 | ||
984 | /* | |
985 | * Generate the rpc reply header | |
986 | * siz arg. is used to decide if adding a cluster is worthwhile | |
987 | */ | |
988 | nfs_rephead(siz, retxid, err, mrq, mbp, bposp) | |
989 | int siz; | |
990 | u_long retxid; | |
991 | int err; | |
992 | struct mbuf **mrq; | |
993 | struct mbuf **mbp; | |
994 | caddr_t *bposp; | |
995 | { | |
0bd503ad KM |
996 | register u_long *p; |
997 | register long t1; | |
998 | caddr_t bpos; | |
999 | struct mbuf *mreq, *mb, *mb2; | |
a2907882 KM |
1000 | |
1001 | NFSMGETHDR(mreq); | |
1002 | mb = mreq; | |
1003 | if ((siz+RPC_REPLYSIZ) > MHLEN) | |
f0f1cbaa | 1004 | MCLGET(mreq, M_WAIT); |
a2907882 KM |
1005 | p = mtod(mreq, u_long *); |
1006 | mreq->m_len = 6*NFSX_UNSIGNED; | |
1007 | bpos = ((caddr_t)p)+mreq->m_len; | |
1008 | *p++ = retxid; | |
1009 | *p++ = rpc_reply; | |
1010 | if (err == ERPCMISMATCH) { | |
1011 | *p++ = rpc_msgdenied; | |
1012 | *p++ = rpc_mismatch; | |
1013 | *p++ = txdr_unsigned(2); | |
1014 | *p = txdr_unsigned(2); | |
1015 | } else { | |
1016 | *p++ = rpc_msgaccepted; | |
1017 | *p++ = 0; | |
1018 | *p++ = 0; | |
1019 | switch (err) { | |
1020 | case EPROGUNAVAIL: | |
1021 | *p = txdr_unsigned(RPC_PROGUNAVAIL); | |
1022 | break; | |
1023 | case EPROGMISMATCH: | |
1024 | *p = txdr_unsigned(RPC_PROGMISMATCH); | |
1025 | nfsm_build(p, u_long *, 2*NFSX_UNSIGNED); | |
1026 | *p++ = txdr_unsigned(2); | |
1027 | *p = txdr_unsigned(2); /* someday 3 */ | |
1028 | break; | |
1029 | case EPROCUNAVAIL: | |
1030 | *p = txdr_unsigned(RPC_PROCUNAVAIL); | |
1031 | break; | |
1032 | default: | |
1033 | *p = 0; | |
1034 | if (err != VNOVAL) { | |
1035 | nfsm_build(p, u_long *, NFSX_UNSIGNED); | |
1036 | *p = txdr_unsigned(err); | |
1037 | } | |
1038 | break; | |
1039 | }; | |
1040 | } | |
1041 | *mrq = mreq; | |
1042 | *mbp = mb; | |
1043 | *bposp = bpos; | |
1044 | if (err != 0 && err != VNOVAL) | |
1045 | nfsstats.srvrpc_errs++; | |
1046 | return (0); | |
1047 | } | |
1048 | ||
1049 | /* | |
1050 | * Nfs timer routine | |
1051 | * Scan the nfsreq list and retranmit any requests that have timed out | |
1052 | * To avoid retransmission attempts on STREAM sockets (in the future) make | |
2f08b65a | 1053 | * sure to set the r_retry field to 0 (implies nm_retry == 0). |
a2907882 KM |
1054 | */ |
1055 | nfs_timer() | |
1056 | { | |
1057 | register struct nfsreq *rep; | |
1058 | register struct mbuf *m; | |
1059 | register struct socket *so; | |
f0f1cbaa | 1060 | register struct nfsmount *nmp; |
2f08b65a | 1061 | int s, error; |
a2907882 KM |
1062 | |
1063 | s = splnet(); | |
f0f1cbaa KM |
1064 | for (rep = nfsreqh.r_next; rep != &nfsreqh; rep = rep->r_next) { |
1065 | nmp = rep->r_nmp; | |
1066 | if (rep->r_mrep || (rep->r_flags & R_SOFTTERM) || | |
1067 | (so = nmp->nm_so) == NULL) | |
1068 | continue; | |
1069 | if ((nmp->nm_flag & NFSMNT_INT) && nfs_sigintr(rep->r_procp)) { | |
1070 | rep->r_flags |= R_SOFTTERM; | |
1071 | continue; | |
1072 | } | |
2f08b65a | 1073 | if (rep->r_flags & R_TIMING) /* update rtt in mount */ |
f0f1cbaa | 1074 | nmp->nm_rtt++; |
f0f1cbaa KM |
1075 | /* If not timed out */ |
1076 | if (++rep->r_timer < nmp->nm_rto) | |
2f08b65a KM |
1077 | continue; |
1078 | /* Do backoff and save new timeout in mount */ | |
1079 | if (rep->r_flags & R_TIMING) { | |
f0f1cbaa | 1080 | nfs_backofftimer(nmp); |
2f08b65a | 1081 | rep->r_flags &= ~R_TIMING; |
f0f1cbaa | 1082 | nmp->nm_rtt = -1; |
2f08b65a KM |
1083 | } |
1084 | if (rep->r_flags & R_SENT) { | |
1085 | rep->r_flags &= ~R_SENT; | |
f0f1cbaa | 1086 | nmp->nm_sent--; |
2f08b65a | 1087 | } |
f0f1cbaa KM |
1088 | |
1089 | /* | |
1090 | * Check for too many retries on soft mount. | |
1091 | * nb: For hard mounts, r_retry == NFS_MAXREXMIT+1 | |
1092 | */ | |
1093 | if (++rep->r_rexmit > NFS_MAXREXMIT) | |
2f08b65a | 1094 | rep->r_rexmit = NFS_MAXREXMIT; |
2f08b65a | 1095 | |
f0f1cbaa KM |
1096 | /* |
1097 | * Check for server not responding | |
1098 | */ | |
1099 | if ((rep->r_flags & R_TPRINTFMSG) == 0 && | |
170bfd05 | 1100 | rep->r_rexmit > NFS_FISHY) { |
f0f1cbaa | 1101 | if (rep->r_procp && rep->r_procp->p_session) |
5580a343 | 1102 | tprintf(rep->r_procp->p_session, |
f0f1cbaa KM |
1103 | "Nfs server %s, not responding\n", |
1104 | nmp->nm_mountp->mnt_stat.f_mntfromname); | |
1105 | else | |
5580a343 | 1106 | tprintf(NULL, |
f0f1cbaa KM |
1107 | "Nfs server %s, not responding\n", |
1108 | nmp->nm_mountp->mnt_stat.f_mntfromname); | |
1109 | rep->r_flags |= R_TPRINTFMSG; | |
1110 | } | |
170bfd05 | 1111 | if (rep->r_rexmit >= rep->r_retry) { /* too many */ |
f0f1cbaa KM |
1112 | nfsstats.rpctimeouts++; |
1113 | rep->r_flags |= R_SOFTTERM; | |
1114 | continue; | |
1115 | } | |
170bfd05 KM |
1116 | if (nmp->nm_sotype != SOCK_DGRAM) |
1117 | continue; | |
f0f1cbaa KM |
1118 | |
1119 | /* | |
1120 | * If there is enough space and the window allows.. | |
1121 | * Resend it | |
1122 | */ | |
1123 | if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && | |
1124 | nmp->nm_sent < nmp->nm_window && | |
1125 | (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ | |
1126 | nfsstats.rpcretries++; | |
1127 | if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) | |
1128 | error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, | |
1129 | (caddr_t)0, (struct mbuf *)0, (struct mbuf *)0); | |
1130 | else | |
1131 | error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, | |
1132 | nmp->nm_nam, (struct mbuf *)0, (struct mbuf *)0); | |
1133 | if (error) { | |
1134 | if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) | |
1135 | so->so_error = 0; | |
1136 | } else { | |
1137 | /* | |
1138 | * We need to time the request even though we | |
1139 | * are retransmitting. | |
1140 | */ | |
1141 | nmp->nm_rtt = 0; | |
1142 | nmp->nm_sent++; | |
1143 | rep->r_flags |= (R_SENT|R_TIMING); | |
1144 | rep->r_timer = rep->r_timerinit; | |
1145 | } | |
1146 | } | |
2f08b65a KM |
1147 | } |
1148 | splx(s); | |
1149 | timeout(nfs_timer, (caddr_t)0, hz/NFS_HZ); | |
1150 | } | |
1151 | ||
1152 | /* | |
1153 | * NFS timer update and backoff. The "Jacobson/Karels/Karn" scheme is | |
1154 | * used here. The timer state is held in the nfsmount structure and | |
1155 | * a single request is used to clock the response. When successful | |
1156 | * the rtt smoothing in nfs_updatetimer is used, when failed the backoff | |
1157 | * is done by nfs_backofftimer. We also log failure messages in these | |
1158 | * routines. | |
1159 | * | |
1160 | * Congestion variables are held in the nfshost structure which | |
1161 | * is referenced by nfsmounts and shared per-server. This separation | |
1162 | * makes it possible to do per-mount timing which allows varying disk | |
1163 | * access times to be dealt with, while preserving a network oriented | |
1164 | * congestion control scheme. | |
1165 | * | |
1166 | * The windowing implements the Jacobson/Karels slowstart algorithm | |
1167 | * with adjusted scaling factors. We start with one request, then send | |
1168 | * 4 more after each success until the ssthresh limit is reached, then | |
1169 | * we increment at a rate proportional to the window. On failure, we | |
1170 | * remember 3/4 the current window and clamp the send limit to 1. Note | |
1171 | * ICMP source quench is not reflected in so->so_error so we ignore that | |
1172 | * for now. | |
1173 | * | |
1174 | * NFS behaves much more like a transport protocol with these changes, | |
1175 | * shedding the teenage pedal-to-the-metal tendencies of "other" | |
1176 | * implementations. | |
1177 | * | |
1178 | * Timers and congestion avoidance by Tom Talpey, Open Software Foundation. | |
1179 | */ | |
1180 | ||
1181 | /* | |
1182 | * The TCP algorithm was not forgiving enough. Because the NFS server | |
1183 | * responds only after performing lookups/diskio/etc, we have to be | |
1184 | * more prepared to accept a spiky variance. The TCP algorithm is: | |
f0f1cbaa | 1185 | * TCP_RTO(nmp) ((((nmp)->nm_srtt >> 2) + (nmp)->nm_rttvar) >> 1) |
2f08b65a | 1186 | */ |
f0f1cbaa | 1187 | #define NFS_RTO(nmp) (((nmp)->nm_srtt >> 3) + (nmp)->nm_rttvar) |
2f08b65a | 1188 | |
f0f1cbaa KM |
1189 | nfs_updatetimer(nmp) |
1190 | register struct nfsmount *nmp; | |
2f08b65a | 1191 | { |
2f08b65a KM |
1192 | |
1193 | /* If retransmitted, clear and return */ | |
f0f1cbaa KM |
1194 | if (nmp->nm_rexmit || nmp->nm_currexmit) { |
1195 | nmp->nm_rexmit = nmp->nm_currexmit = 0; | |
2f08b65a KM |
1196 | return; |
1197 | } | |
1198 | /* If have a measurement, do smoothing */ | |
f0f1cbaa | 1199 | if (nmp->nm_srtt) { |
2f08b65a | 1200 | register short delta; |
f0f1cbaa KM |
1201 | delta = nmp->nm_rtt - (nmp->nm_srtt >> 3); |
1202 | if ((nmp->nm_srtt += delta) <= 0) | |
1203 | nmp->nm_srtt = 1; | |
2f08b65a KM |
1204 | if (delta < 0) |
1205 | delta = -delta; | |
f0f1cbaa KM |
1206 | delta -= (nmp->nm_rttvar >> 2); |
1207 | if ((nmp->nm_rttvar += delta) <= 0) | |
1208 | nmp->nm_rttvar = 1; | |
2f08b65a KM |
1209 | /* Else initialize */ |
1210 | } else { | |
f0f1cbaa KM |
1211 | nmp->nm_rttvar = nmp->nm_rtt << 1; |
1212 | if (nmp->nm_rttvar == 0) nmp->nm_rttvar = 2; | |
1213 | nmp->nm_srtt = nmp->nm_rttvar << 2; | |
2f08b65a KM |
1214 | } |
1215 | /* Compute new Retransmission TimeOut and clip */ | |
f0f1cbaa KM |
1216 | nmp->nm_rto = NFS_RTO(nmp); |
1217 | if (nmp->nm_rto < NFS_MINTIMEO) | |
1218 | nmp->nm_rto = NFS_MINTIMEO; | |
1219 | else if (nmp->nm_rto > NFS_MAXTIMEO) | |
1220 | nmp->nm_rto = NFS_MAXTIMEO; | |
2f08b65a KM |
1221 | |
1222 | /* Update window estimate */ | |
f0f1cbaa KM |
1223 | if (nmp->nm_window < nmp->nm_ssthresh) /* quickly */ |
1224 | nmp->nm_window += 4; | |
2f08b65a | 1225 | else { /* slowly */ |
f0f1cbaa KM |
1226 | register long incr = ++nmp->nm_winext; |
1227 | incr = (incr * incr) / nmp->nm_window; | |
2f08b65a | 1228 | if (incr > 0) { |
f0f1cbaa KM |
1229 | nmp->nm_winext = 0; |
1230 | ++nmp->nm_window; | |
2f08b65a KM |
1231 | } |
1232 | } | |
f0f1cbaa KM |
1233 | if (nmp->nm_window > NFS_MAXWINDOW) |
1234 | nmp->nm_window = NFS_MAXWINDOW; | |
2f08b65a KM |
1235 | } |
1236 | ||
f0f1cbaa KM |
1237 | nfs_backofftimer(nmp) |
1238 | register struct nfsmount *nmp; | |
2f08b65a | 1239 | { |
2f08b65a KM |
1240 | register unsigned long newrto; |
1241 | ||
1242 | /* Clip shift count */ | |
f0f1cbaa KM |
1243 | if (++nmp->nm_rexmit > 8 * sizeof nmp->nm_rto) |
1244 | nmp->nm_rexmit = 8 * sizeof nmp->nm_rto; | |
2f08b65a | 1245 | /* Back off RTO exponentially */ |
f0f1cbaa KM |
1246 | newrto = NFS_RTO(nmp); |
1247 | newrto <<= (nmp->nm_rexmit - 1); | |
2f08b65a KM |
1248 | if (newrto == 0 || newrto > NFS_MAXTIMEO) |
1249 | newrto = NFS_MAXTIMEO; | |
f0f1cbaa | 1250 | nmp->nm_rto = newrto; |
2f08b65a KM |
1251 | |
1252 | /* If too many retries, message, assume a bogus RTT and re-measure */ | |
f0f1cbaa KM |
1253 | if (nmp->nm_currexmit < nmp->nm_rexmit) { |
1254 | nmp->nm_currexmit = nmp->nm_rexmit; | |
1255 | if (nmp->nm_currexmit >= nfsrexmtthresh) { | |
1256 | if (nmp->nm_currexmit == nfsrexmtthresh) { | |
1257 | nmp->nm_rttvar += (nmp->nm_srtt >> 2); | |
1258 | nmp->nm_srtt = 0; | |
a2907882 KM |
1259 | } |
1260 | } | |
a2907882 | 1261 | } |
2f08b65a | 1262 | /* Close down window but remember this point (3/4 current) for later */ |
f0f1cbaa KM |
1263 | nmp->nm_ssthresh = ((nmp->nm_window << 1) + nmp->nm_window) >> 2; |
1264 | nmp->nm_window = 1; | |
1265 | nmp->nm_winext = 0; | |
a2907882 KM |
1266 | } |
1267 | ||
1268 | /* | |
f0f1cbaa KM |
1269 | * Test for a termination signal pending on procp. |
1270 | * This is used for NFSMNT_INT mounts. | |
a2907882 | 1271 | */ |
f0f1cbaa KM |
1272 | nfs_sigintr(p) |
1273 | register struct proc *p; | |
1274 | { | |
1275 | if (p && p->p_sig && (((p->p_sig &~ p->p_sigmask) &~ p->p_sigignore) & | |
1276 | NFSINT_SIGMASK)) | |
1277 | return (1); | |
1278 | else | |
1279 | return (0); | |
1280 | } | |
2f08b65a | 1281 | |
f0f1cbaa KM |
1282 | /* |
1283 | * Lock a socket against others. | |
1284 | * Necessary for STREAM sockets to ensure you get an entire rpc request/reply | |
1285 | * and also to avoid race conditions between the processes with nfs requests | |
1286 | * in progress when a reconnect is necessary. | |
1287 | */ | |
170bfd05 KM |
1288 | nfs_solock(flagp) |
1289 | register int *flagp; | |
a2907882 | 1290 | { |
2f08b65a | 1291 | |
f0f1cbaa KM |
1292 | while (*flagp & NFSMNT_SCKLOCK) { |
1293 | *flagp |= NFSMNT_WANTSCK; | |
170bfd05 | 1294 | (void) tsleep((caddr_t)flagp, PZERO-1, "nfsolck", 0); |
2f08b65a | 1295 | } |
f0f1cbaa KM |
1296 | *flagp |= NFSMNT_SCKLOCK; |
1297 | } | |
2f08b65a | 1298 | |
f0f1cbaa KM |
1299 | /* |
1300 | * Unlock the stream socket for others. | |
1301 | */ | |
1302 | nfs_sounlock(flagp) | |
170bfd05 | 1303 | register int *flagp; |
f0f1cbaa KM |
1304 | { |
1305 | ||
1306 | if ((*flagp & NFSMNT_SCKLOCK) == 0) | |
1307 | panic("nfs sounlock"); | |
1308 | *flagp &= ~NFSMNT_SCKLOCK; | |
1309 | if (*flagp & NFSMNT_WANTSCK) { | |
1310 | *flagp &= ~NFSMNT_WANTSCK; | |
1311 | wakeup((caddr_t)flagp); | |
2f08b65a | 1312 | } |
f0f1cbaa KM |
1313 | } |
1314 | ||
1315 | /* | |
1316 | * This function compares two net addresses by family and returns TRUE | |
1317 | * if they are the same. | |
1318 | * If there is any doubt, return FALSE. | |
1319 | */ | |
1320 | nfs_netaddr_match(nam1, nam2) | |
1321 | struct mbuf *nam1, *nam2; | |
1322 | { | |
1323 | register struct sockaddr *saddr1, *saddr2; | |
1324 | ||
1325 | saddr1 = mtod(nam1, struct sockaddr *); | |
1326 | saddr2 = mtod(nam2, struct sockaddr *); | |
1327 | if (saddr1->sa_family != saddr2->sa_family) | |
1328 | return (0); | |
1329 | ||
1330 | /* | |
1331 | * Must do each address family separately since unused fields | |
1332 | * are undefined values and not always zeroed. | |
1333 | */ | |
1334 | switch (saddr1->sa_family) { | |
1335 | case AF_INET: | |
1336 | if (((struct sockaddr_in *)saddr1)->sin_addr.s_addr == | |
1337 | ((struct sockaddr_in *)saddr2)->sin_addr.s_addr) | |
1338 | return (1); | |
1339 | break; | |
1340 | default: | |
1341 | break; | |
1342 | }; | |
1343 | return (0); | |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * Check the hostname fields for nfsd's mask and match fields. | |
1348 | * By address family: | |
1349 | * - Bitwise AND the mask with the host address field | |
1350 | * - Compare for == with match | |
1351 | * return TRUE if not equal | |
1352 | */ | |
1353 | nfs_badnam(nam, msk, mtch) | |
1354 | register struct mbuf *nam, *msk, *mtch; | |
1355 | { | |
1356 | switch (mtod(nam, struct sockaddr *)->sa_family) { | |
1357 | case AF_INET: | |
1358 | return ((mtod(nam, struct sockaddr_in *)->sin_addr.s_addr & | |
1359 | mtod(msk, struct sockaddr_in *)->sin_addr.s_addr) != | |
1360 | mtod(mtch, struct sockaddr_in *)->sin_addr.s_addr); | |
1361 | default: | |
1362 | printf("nfs_badmatch, unknown sa_family\n"); | |
1363 | return (0); | |
1364 | }; | |
a2907882 | 1365 | } |