1 /* $NetBSD: nfs_socket.c,v 1.21 1995/08/13 00:00:01 mycroft Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)nfs_socket.c 8.3 (Berkeley) 1/12/94 39 */ 40 41 /* 42 * Socket operations for use by nfs 43 */ 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/proc.h> 48 #include <sys/mount.h> 49 #include <sys/kernel.h> 50 #include <sys/mbuf.h> 51 #include <sys/vnode.h> 52 #include <sys/domain.h> 53 #include <sys/protosw.h> 54 #include <sys/socket.h> 55 #include <sys/socketvar.h> 56 #include <sys/syslog.h> 57 #include <sys/tprintf.h> 58 59 #include <netinet/in.h> 60 #include <netinet/tcp.h> 61 #include <nfs/rpcv2.h> 62 #include <nfs/nfsv2.h> 63 #include <nfs/nfs.h> 64 #include <nfs/xdr_subs.h> 65 #include <nfs/nfsm_subs.h> 66 #include <nfs/nfsmount.h> 67 #include <nfs/nfsnode.h> 68 #include <nfs/nfsrtt.h> 69 #include <nfs/nqnfs.h> 70 71 #define TRUE 1 72 #define FALSE 0 73 74 /* 75 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 76 * Use the mean and mean deviation of rtt for the appropriate type of rpc 77 * for the frequent rpcs and a default for the others. 78 * The justification for doing "other" this way is that these rpcs 79 * happen so infrequently that timer est. would probably be stale. 80 * Also, since many of these rpcs are 81 * non-idempotent, a conservative timeout is desired. 82 * getattr, lookup - A+2D 83 * read, write - A+4D 84 * other - nm_timeo 85 */ 86 #define NFS_RTO(n, t) \ 87 ((t) == 0 ? (n)->nm_timeo : \ 88 ((t) < 3 ? \ 89 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 90 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 91 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 92 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 93 /* 94 * External data, mostly RPC constants in XDR form 95 */ 96 extern u_long rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, rpc_auth_unix, 97 rpc_msgaccepted, rpc_call, rpc_autherr, rpc_rejectedcred, 98 rpc_auth_kerb; 99 extern u_long nfs_prog, nfs_vers, nqnfs_prog, nqnfs_vers; 100 extern time_t nqnfsstarttime; 101 extern int nonidempotent[NFS_NPROCS]; 102 103 /* 104 * Maps errno values to nfs error numbers. 105 * Use NFSERR_IO as the catch all for ones not specifically defined in 106 * RFC 1094. 107 */ 108 static int nfsrv_errmap[ELAST] = { 109 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, 110 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 111 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, 112 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR, 113 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 114 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, 115 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 116 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 117 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 118 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 119 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 120 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 121 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO, 122 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, 123 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 124 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 125 NFSERR_IO, 126 }; 127 128 /* 129 * Defines which timer to use for the procnum. 130 * 0 - default 131 * 1 - getattr 132 * 2 - lookup 133 * 3 - read 134 * 4 - write 135 */ 136 static int proct[NFS_NPROCS] = { 137 0, 1, 0, 0, 2, 3, 3, 0, 4, 0, 0, 0, 0, 0, 0, 0, 3, 0, 3, 0, 0, 0, 0, 138 }; 139 140 /* 141 * There is a congestion window for outstanding rpcs maintained per mount 142 * point. The cwnd size is adjusted in roughly the way that: 143 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 144 * SIGCOMM '88". ACM, August 1988. 145 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 146 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 147 * of rpcs is in progress. 148 * (The sent count and cwnd are scaled for integer arith.) 149 * Variants of "slow start" were tried and were found to be too much of a 150 * performance hit (ave. rtt 3 times larger), 151 * I suspect due to the large rtt that nfs rpcs have. 152 */ 153 #define NFS_CWNDSCALE 256 154 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 155 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 156 int nfs_sbwait(); 157 void nfs_disconnect(), nfs_realign(), nfsrv_wakenfsd(), nfs_sndunlock(); 158 void nfs_rcvunlock(), nqnfs_serverd(), nqnfs_clientlease(); 159 struct mbuf *nfsm_rpchead(); 160 int nfsrtton = 0; 161 struct nfsrtt nfsrtt; 162 163 /* 164 * Initialize sockets and congestion for a new NFS connection. 165 * We do not free the sockaddr if error. 166 */ 167 nfs_connect(nmp, rep) 168 register struct nfsmount *nmp; 169 struct nfsreq *rep; 170 { 171 register struct socket *so; 172 int s, error, rcvreserve, sndreserve; 173 struct sockaddr *saddr; 174 struct sockaddr_in *sin; 175 struct mbuf *m; 176 u_short tport; 177 178 nmp->nm_so = (struct socket *)0; 179 saddr = mtod(nmp->nm_nam, struct sockaddr *); 180 if (error = socreate(saddr->sa_family, 181 &nmp->nm_so, nmp->nm_sotype, nmp->nm_soproto)) 182 goto bad; 183 so = nmp->nm_so; 184 nmp->nm_soflags = so->so_proto->pr_flags; 185 186 /* 187 * Some servers require that the client port be a reserved port number. 188 */ 189 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 190 MGET(m, M_WAIT, MT_SONAME); 191 sin = mtod(m, struct sockaddr_in *); 192 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 193 sin->sin_family = AF_INET; 194 sin->sin_addr.s_addr = INADDR_ANY; 195 tport = IPPORT_RESERVED - 1; 196 sin->sin_port = htons(tport); 197 while ((error = sobind(so, m)) == EADDRINUSE && 198 --tport > IPPORT_RESERVED / 2) 199 sin->sin_port = htons(tport); 200 m_freem(m); 201 if (error) 202 goto bad; 203 } 204 205 /* 206 * Protocols that do not require connections may be optionally left 207 * unconnected for servers that reply from a port other than NFS_PORT. 208 */ 209 if (nmp->nm_flag & NFSMNT_NOCONN) { 210 if (nmp->nm_soflags & PR_CONNREQUIRED) { 211 error = ENOTCONN; 212 goto bad; 213 } 214 } else { 215 if (error = soconnect(so, nmp->nm_nam)) 216 goto bad; 217 218 /* 219 * Wait for the connection to complete. Cribbed from the 220 * connect system call but with the wait timing out so 221 * that interruptible mounts don't hang here for a long time. 222 */ 223 s = splsoftnet(); 224 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 225 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 226 "nfscon", 2 * hz); 227 if ((so->so_state & SS_ISCONNECTING) && 228 so->so_error == 0 && rep && 229 (error = nfs_sigintr(nmp, rep, rep->r_procp))) { 230 so->so_state &= ~SS_ISCONNECTING; 231 splx(s); 232 goto bad; 233 } 234 } 235 if (so->so_error) { 236 error = so->so_error; 237 so->so_error = 0; 238 splx(s); 239 goto bad; 240 } 241 splx(s); 242 } 243 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 244 so->so_rcv.sb_timeo = (5 * hz); 245 so->so_snd.sb_timeo = (5 * hz); 246 } else { 247 so->so_rcv.sb_timeo = 0; 248 so->so_snd.sb_timeo = 0; 249 } 250 if (nmp->nm_sotype == SOCK_DGRAM) { 251 sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR; 252 rcvreserve = nmp->nm_rsize + NFS_MAXPKTHDR; 253 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 254 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 255 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR) * 2; 256 } else { 257 if (nmp->nm_sotype != SOCK_STREAM) 258 panic("nfscon sotype"); 259 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 260 MGET(m, M_WAIT, MT_SOOPTS); 261 *mtod(m, int *) = 1; 262 m->m_len = sizeof(int); 263 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 264 } 265 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 266 MGET(m, M_WAIT, MT_SOOPTS); 267 *mtod(m, int *) = 1; 268 m->m_len = sizeof(int); 269 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 270 } 271 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + sizeof (u_long)) 272 * 2; 273 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + sizeof (u_long)) 274 * 2; 275 } 276 if (error = soreserve(so, sndreserve, rcvreserve)) 277 goto bad; 278 so->so_rcv.sb_flags |= SB_NOINTR; 279 so->so_snd.sb_flags |= SB_NOINTR; 280 281 /* Initialize other non-zero congestion variables */ 282 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 283 nmp->nm_srtt[4] = (NFS_TIMEO << 3); 284 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 285 nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0; 286 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 287 nmp->nm_sent = 0; 288 nmp->nm_timeouts = 0; 289 return (0); 290 291 bad: 292 nfs_disconnect(nmp); 293 return (error); 294 } 295 296 /* 297 * Reconnect routine: 298 * Called when a connection is broken on a reliable protocol. 299 * - clean up the old socket 300 * - nfs_connect() again 301 * - set R_MUSTRESEND for all outstanding requests on mount point 302 * If this fails the mount point is DEAD! 303 * nb: Must be called with the nfs_sndlock() set on the mount point. 304 */ 305 nfs_reconnect(rep) 306 register struct nfsreq *rep; 307 { 308 register struct nfsreq *rp; 309 register struct nfsmount *nmp = rep->r_nmp; 310 int error; 311 312 nfs_disconnect(nmp); 313 while (error = nfs_connect(nmp, rep)) { 314 if (error == EINTR || error == ERESTART) 315 return (EINTR); 316 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); 317 } 318 319 /* 320 * Loop through outstanding request list and fix up all requests 321 * on old socket. 322 */ 323 for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) { 324 if (rp->r_nmp == nmp) 325 rp->r_flags |= R_MUSTRESEND; 326 } 327 return (0); 328 } 329 330 /* 331 * NFS disconnect. Clean up and unlink. 332 */ 333 void 334 nfs_disconnect(nmp) 335 register struct nfsmount *nmp; 336 { 337 register struct socket *so; 338 339 if (nmp->nm_so) { 340 so = nmp->nm_so; 341 nmp->nm_so = (struct socket *)0; 342 soshutdown(so, 2); 343 soclose(so); 344 } 345 } 346 347 /* 348 * This is the nfs send routine. For connection based socket types, it 349 * must be called with an nfs_sndlock() on the socket. 350 * "rep == NULL" indicates that it has been called from a server. 351 * For the client side: 352 * - return EINTR if the RPC is terminated, 0 otherwise 353 * - set R_MUSTRESEND if the send fails for any reason 354 * - do any cleanup required by recoverable socket errors (???) 355 * For the server side: 356 * - return EINTR or ERESTART if interrupted by a signal 357 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 358 * - do any cleanup required by recoverable socket errors (???) 359 */ 360 nfs_send(so, nam, top, rep) 361 register struct socket *so; 362 struct mbuf *nam; 363 register struct mbuf *top; 364 struct nfsreq *rep; 365 { 366 struct mbuf *sendnam; 367 int error, soflags, flags; 368 369 if (rep) { 370 if (rep->r_flags & R_SOFTTERM) { 371 m_freem(top); 372 return (EINTR); 373 } 374 if ((so = rep->r_nmp->nm_so) == NULL) { 375 rep->r_flags |= R_MUSTRESEND; 376 m_freem(top); 377 return (0); 378 } 379 rep->r_flags &= ~R_MUSTRESEND; 380 soflags = rep->r_nmp->nm_soflags; 381 } else 382 soflags = so->so_proto->pr_flags; 383 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 384 sendnam = (struct mbuf *)0; 385 else 386 sendnam = nam; 387 if (so->so_type == SOCK_SEQPACKET) 388 flags = MSG_EOR; 389 else 390 flags = 0; 391 392 error = sosend(so, sendnam, (struct uio *)0, top, 393 (struct mbuf *)0, flags); 394 if (error) { 395 if (rep) { 396 log(LOG_INFO, "nfs send error %d for server %s\n",error, 397 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 398 /* 399 * Deal with errors for the client side. 400 */ 401 if (rep->r_flags & R_SOFTTERM) 402 error = EINTR; 403 else 404 rep->r_flags |= R_MUSTRESEND; 405 } else 406 log(LOG_INFO, "nfsd send error %d\n", error); 407 408 /* 409 * Handle any recoverable (soft) socket errors here. (???) 410 */ 411 if (error != EINTR && error != ERESTART && 412 error != EWOULDBLOCK && error != EPIPE) 413 error = 0; 414 } 415 return (error); 416 } 417 418 #ifdef NFSCLIENT 419 /* 420 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 421 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 422 * Mark and consolidate the data into a new mbuf list. 423 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 424 * small mbufs. 425 * For SOCK_STREAM we must be very careful to read an entire record once 426 * we have read any of it, even if the system call has been interrupted. 427 */ 428 nfs_receive(rep, aname, mp) 429 register struct nfsreq *rep; 430 struct mbuf **aname; 431 struct mbuf **mp; 432 { 433 register struct socket *so; 434 struct uio auio; 435 struct iovec aio; 436 register struct mbuf *m; 437 struct mbuf *control; 438 u_long len; 439 struct mbuf **getnam; 440 int error, sotype, rcvflg; 441 struct proc *p = curproc; /* XXX */ 442 443 /* 444 * Set up arguments for soreceive() 445 */ 446 *mp = (struct mbuf *)0; 447 *aname = (struct mbuf *)0; 448 sotype = rep->r_nmp->nm_sotype; 449 450 /* 451 * For reliable protocols, lock against other senders/receivers 452 * in case a reconnect is necessary. 453 * For SOCK_STREAM, first get the Record Mark to find out how much 454 * more there is to get. 455 * We must lock the socket against other receivers 456 * until we have an entire rpc request/reply. 457 */ 458 if (sotype != SOCK_DGRAM) { 459 if (error = nfs_sndlock(&rep->r_nmp->nm_flag, rep)) 460 return (error); 461 tryagain: 462 /* 463 * Check for fatal errors and resending request. 464 */ 465 /* 466 * Ugh: If a reconnect attempt just happened, nm_so 467 * would have changed. NULL indicates a failed 468 * attempt that has essentially shut down this 469 * mount point. 470 */ 471 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 472 nfs_sndunlock(&rep->r_nmp->nm_flag); 473 return (EINTR); 474 } 475 if ((so = rep->r_nmp->nm_so) == NULL) { 476 if (error = nfs_reconnect(rep)) { 477 nfs_sndunlock(&rep->r_nmp->nm_flag); 478 return (error); 479 } 480 goto tryagain; 481 } 482 while (rep->r_flags & R_MUSTRESEND) { 483 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 484 nfsstats.rpcretries++; 485 if (error = nfs_send(so, rep->r_nmp->nm_nam, m, rep)) { 486 if (error == EINTR || error == ERESTART || 487 (error = nfs_reconnect(rep))) { 488 nfs_sndunlock(&rep->r_nmp->nm_flag); 489 return (error); 490 } 491 goto tryagain; 492 } 493 } 494 nfs_sndunlock(&rep->r_nmp->nm_flag); 495 if (sotype == SOCK_STREAM) { 496 aio.iov_base = (caddr_t) &len; 497 aio.iov_len = sizeof(u_long); 498 auio.uio_iov = &aio; 499 auio.uio_iovcnt = 1; 500 auio.uio_segflg = UIO_SYSSPACE; 501 auio.uio_rw = UIO_READ; 502 auio.uio_offset = 0; 503 auio.uio_resid = sizeof(u_long); 504 auio.uio_procp = p; 505 do { 506 rcvflg = MSG_WAITALL; 507 error = soreceive(so, (struct mbuf **)0, &auio, 508 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 509 if (error == EWOULDBLOCK && rep) { 510 if (rep->r_flags & R_SOFTTERM) 511 return (EINTR); 512 } 513 } while (error == EWOULDBLOCK); 514 if (!error && auio.uio_resid > 0) { 515 log(LOG_INFO, 516 "short receive (%d/%d) from nfs server %s\n", 517 sizeof(u_long) - auio.uio_resid, 518 sizeof(u_long), 519 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 520 error = EPIPE; 521 } 522 if (error) 523 goto errout; 524 len = ntohl(len) & ~0x80000000; 525 /* 526 * This is SERIOUS! We are out of sync with the sender 527 * and forcing a disconnect/reconnect is all I can do. 528 */ 529 if (len > NFS_MAXPACKET) { 530 log(LOG_ERR, "%s (%d) from nfs server %s\n", 531 "impossible packet length", 532 len, 533 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 534 error = EFBIG; 535 goto errout; 536 } 537 auio.uio_resid = len; 538 do { 539 rcvflg = MSG_WAITALL; 540 error = soreceive(so, (struct mbuf **)0, 541 &auio, mp, (struct mbuf **)0, &rcvflg); 542 } while (error == EWOULDBLOCK || error == EINTR || 543 error == ERESTART); 544 if (!error && auio.uio_resid > 0) { 545 log(LOG_INFO, 546 "short receive (%d/%d) from nfs server %s\n", 547 len - auio.uio_resid, len, 548 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 549 error = EPIPE; 550 } 551 } else { 552 /* 553 * NB: Since uio_resid is big, MSG_WAITALL is ignored 554 * and soreceive() will return when it has either a 555 * control msg or a data msg. 556 * We have no use for control msg., but must grab them 557 * and then throw them away so we know what is going 558 * on. 559 */ 560 auio.uio_resid = len = 100000000; /* Anything Big */ 561 auio.uio_procp = p; 562 do { 563 rcvflg = 0; 564 error = soreceive(so, (struct mbuf **)0, 565 &auio, mp, &control, &rcvflg); 566 if (control) 567 m_freem(control); 568 if (error == EWOULDBLOCK && rep) { 569 if (rep->r_flags & R_SOFTTERM) 570 return (EINTR); 571 } 572 } while (error == EWOULDBLOCK || 573 (!error && *mp == NULL && control)); 574 if ((rcvflg & MSG_EOR) == 0) 575 printf("Egad!!\n"); 576 if (!error && *mp == NULL) 577 error = EPIPE; 578 len -= auio.uio_resid; 579 } 580 errout: 581 if (error && error != EINTR && error != ERESTART) { 582 m_freem(*mp); 583 *mp = (struct mbuf *)0; 584 if (error != EPIPE) 585 log(LOG_INFO, 586 "receive error %d from nfs server %s\n", 587 error, 588 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 589 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 590 if (!error) 591 error = nfs_reconnect(rep); 592 if (!error) 593 goto tryagain; 594 } 595 } else { 596 if ((so = rep->r_nmp->nm_so) == NULL) 597 return (EACCES); 598 if (so->so_state & SS_ISCONNECTED) 599 getnam = (struct mbuf **)0; 600 else 601 getnam = aname; 602 auio.uio_resid = len = 1000000; 603 auio.uio_procp = p; 604 do { 605 rcvflg = 0; 606 error = soreceive(so, getnam, &auio, mp, 607 (struct mbuf **)0, &rcvflg); 608 if (error == EWOULDBLOCK && 609 (rep->r_flags & R_SOFTTERM)) 610 return (EINTR); 611 } while (error == EWOULDBLOCK); 612 len -= auio.uio_resid; 613 } 614 if (error) { 615 m_freem(*mp); 616 *mp = (struct mbuf *)0; 617 } 618 /* 619 * Search for any mbufs that are not a multiple of 4 bytes long 620 * or with m_data not longword aligned. 621 * These could cause pointer alignment problems, so copy them to 622 * well aligned mbufs. 623 */ 624 nfs_realign(*mp, 5 * NFSX_UNSIGNED); 625 return (error); 626 } 627 628 /* 629 * Implement receipt of reply on a socket. 630 * We must search through the list of received datagrams matching them 631 * with outstanding requests using the xid, until ours is found. 632 */ 633 /* ARGSUSED */ 634 nfs_reply(myrep) 635 struct nfsreq *myrep; 636 { 637 register struct nfsreq *rep; 638 register struct nfsmount *nmp = myrep->r_nmp; 639 register long t1; 640 struct mbuf *mrep, *nam, *md; 641 u_long rxid, *tl; 642 caddr_t dpos, cp2; 643 int error; 644 645 /* 646 * Loop around until we get our own reply 647 */ 648 for (;;) { 649 /* 650 * Lock against other receivers so that I don't get stuck in 651 * sbwait() after someone else has received my reply for me. 652 * Also necessary for connection based protocols to avoid 653 * race conditions during a reconnect. 654 */ 655 if (error = nfs_rcvlock(myrep)) 656 return (error); 657 /* Already received, bye bye */ 658 if (myrep->r_mrep != NULL) { 659 nfs_rcvunlock(&nmp->nm_flag); 660 return (0); 661 } 662 /* 663 * Get the next Rpc reply off the socket 664 */ 665 error = nfs_receive(myrep, &nam, &mrep); 666 nfs_rcvunlock(&nmp->nm_flag); 667 if (error) { 668 669 /* 670 * Ignore routing errors on connectionless protocols?? 671 */ 672 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 673 nmp->nm_so->so_error = 0; 674 if (myrep->r_flags & R_GETONEREP) 675 return (0); 676 continue; 677 } 678 return (error); 679 } 680 if (nam) 681 m_freem(nam); 682 683 /* 684 * Get the xid and check that it is an rpc reply 685 */ 686 md = mrep; 687 dpos = mtod(md, caddr_t); 688 nfsm_dissect(tl, u_long *, 2*NFSX_UNSIGNED); 689 rxid = *tl++; 690 if (*tl != rpc_reply) { 691 if (nmp->nm_flag & NFSMNT_NQNFS) { 692 if (nqnfs_callback(nmp, mrep, md, dpos)) 693 nfsstats.rpcinvalid++; 694 } else { 695 nfsstats.rpcinvalid++; 696 m_freem(mrep); 697 } 698 nfsmout: 699 if (myrep->r_flags & R_GETONEREP) 700 return (0); 701 continue; 702 } 703 704 /* 705 * Loop through the request list to match up the reply 706 * Iff no match, just drop the datagram 707 */ 708 for (rep = nfs_reqq.tqh_first; rep != 0; 709 rep = rep->r_chain.tqe_next) { 710 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 711 /* Found it.. */ 712 rep->r_mrep = mrep; 713 rep->r_md = md; 714 rep->r_dpos = dpos; 715 if (nfsrtton) { 716 struct rttl *rt; 717 718 rt = &nfsrtt.rttl[nfsrtt.pos]; 719 rt->proc = rep->r_procnum; 720 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 721 rt->sent = nmp->nm_sent; 722 rt->cwnd = nmp->nm_cwnd; 723 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 724 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 725 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 726 rt->tstamp = time; 727 if (rep->r_flags & R_TIMING) 728 rt->rtt = rep->r_rtt; 729 else 730 rt->rtt = 1000000; 731 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 732 } 733 /* 734 * Update congestion window. 735 * Do the additive increase of 736 * one rpc/rtt. 737 */ 738 if (nmp->nm_cwnd <= nmp->nm_sent) { 739 nmp->nm_cwnd += 740 (NFS_CWNDSCALE * NFS_CWNDSCALE + 741 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 742 if (nmp->nm_cwnd > NFS_MAXCWND) 743 nmp->nm_cwnd = NFS_MAXCWND; 744 } 745 rep->r_flags &= ~R_SENT; 746 nmp->nm_sent -= NFS_CWNDSCALE; 747 /* 748 * Update rtt using a gain of 0.125 on the mean 749 * and a gain of 0.25 on the deviation. 750 */ 751 if (rep->r_flags & R_TIMING) { 752 /* 753 * Since the timer resolution of 754 * NFS_HZ is so course, it can often 755 * result in r_rtt == 0. Since 756 * r_rtt == N means that the actual 757 * rtt is between N+dt and N+2-dt ticks, 758 * add 1. 759 */ 760 t1 = rep->r_rtt + 1; 761 t1 -= (NFS_SRTT(rep) >> 3); 762 NFS_SRTT(rep) += t1; 763 if (t1 < 0) 764 t1 = -t1; 765 t1 -= (NFS_SDRTT(rep) >> 2); 766 NFS_SDRTT(rep) += t1; 767 } 768 nmp->nm_timeouts = 0; 769 break; 770 } 771 } 772 /* 773 * If not matched to a request, drop it. 774 * If it's mine, get out. 775 */ 776 if (rep == 0) { 777 nfsstats.rpcunexpected++; 778 m_freem(mrep); 779 } else if (rep == myrep) { 780 if (rep->r_mrep == NULL) 781 panic("nfsreply nil"); 782 return (0); 783 } 784 if (myrep->r_flags & R_GETONEREP) 785 return (0); 786 } 787 } 788 789 /* 790 * nfs_request - goes something like this 791 * - fill in request struct 792 * - links it into list 793 * - calls nfs_send() for first transmit 794 * - calls nfs_receive() to get reply 795 * - break down rpc header and return with nfs reply pointed to 796 * by mrep or error 797 * nb: always frees up mreq mbuf list 798 */ 799 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 800 struct vnode *vp; 801 struct mbuf *mrest; 802 int procnum; 803 struct proc *procp; 804 struct ucred *cred; 805 struct mbuf **mrp; 806 struct mbuf **mdp; 807 caddr_t *dposp; 808 { 809 register struct mbuf *m, *mrep; 810 register struct nfsreq *rep; 811 register u_long *tl; 812 register int i; 813 struct nfsmount *nmp; 814 struct mbuf *md, *mheadend; 815 struct nfsreq *reph; 816 struct nfsnode *np; 817 time_t reqtime, waituntil; 818 caddr_t dpos, cp2; 819 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; 820 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 821 u_long xid; 822 u_quad_t frev; 823 char *auth_str; 824 825 nmp = VFSTONFS(vp->v_mount); 826 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 827 rep->r_nmp = nmp; 828 rep->r_vp = vp; 829 rep->r_procp = procp; 830 rep->r_procnum = procnum; 831 i = 0; 832 m = mrest; 833 while (m) { 834 i += m->m_len; 835 m = m->m_next; 836 } 837 mrest_len = i; 838 839 /* 840 * Get the RPC header with authorization. 841 */ 842 kerbauth: 843 auth_str = (char *)0; 844 if (nmp->nm_flag & NFSMNT_KERB) { 845 if (failed_auth) { 846 error = nfs_getauth(nmp, rep, cred, &auth_type, 847 &auth_str, &auth_len); 848 if (error) { 849 free((caddr_t)rep, M_NFSREQ); 850 m_freem(mrest); 851 return (error); 852 } 853 } else { 854 auth_type = RPCAUTH_UNIX; 855 auth_len = 5 * NFSX_UNSIGNED; 856 } 857 } else { 858 auth_type = RPCAUTH_UNIX; 859 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 860 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 861 5 * NFSX_UNSIGNED; 862 } 863 m = nfsm_rpchead(cred, (nmp->nm_flag & NFSMNT_NQNFS), procnum, 864 auth_type, auth_len, auth_str, mrest, mrest_len, &mheadend, &xid); 865 if (auth_str) 866 free(auth_str, M_TEMP); 867 868 /* 869 * For stream protocols, insert a Sun RPC Record Mark. 870 */ 871 if (nmp->nm_sotype == SOCK_STREAM) { 872 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 873 *mtod(m, u_long *) = htonl(0x80000000 | 874 (m->m_pkthdr.len - NFSX_UNSIGNED)); 875 } 876 rep->r_mreq = m; 877 rep->r_xid = xid; 878 tryagain: 879 if (nmp->nm_flag & NFSMNT_SOFT) 880 rep->r_retry = nmp->nm_retry; 881 else 882 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 883 rep->r_rtt = rep->r_rexmit = 0; 884 if (proct[procnum] > 0) 885 rep->r_flags = R_TIMING; 886 else 887 rep->r_flags = 0; 888 rep->r_mrep = NULL; 889 890 /* 891 * Do the client side RPC. 892 */ 893 nfsstats.rpcrequests++; 894 /* 895 * Chain request into list of outstanding requests. Be sure 896 * to put it LAST so timer finds oldest requests first. 897 */ 898 s = splsoftclock(); 899 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 900 901 /* Get send time for nqnfs */ 902 reqtime = time.tv_sec; 903 904 /* 905 * If backing off another request or avoiding congestion, don't 906 * send this one now but let timer do it. If not timing a request, 907 * do it now. 908 */ 909 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 910 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 911 nmp->nm_sent < nmp->nm_cwnd)) { 912 splx(s); 913 if (nmp->nm_soflags & PR_CONNREQUIRED) 914 error = nfs_sndlock(&nmp->nm_flag, rep); 915 if (!error) { 916 m = m_copym(m, 0, M_COPYALL, M_WAIT); 917 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 918 if (nmp->nm_soflags & PR_CONNREQUIRED) 919 nfs_sndunlock(&nmp->nm_flag); 920 } 921 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 922 nmp->nm_sent += NFS_CWNDSCALE; 923 rep->r_flags |= R_SENT; 924 } 925 } else { 926 splx(s); 927 rep->r_rtt = -1; 928 } 929 930 /* 931 * Wait for the reply from our send or the timer's. 932 */ 933 if (!error || error == EPIPE) 934 error = nfs_reply(rep); 935 936 /* 937 * RPC done, unlink the request. 938 */ 939 s = splsoftclock(); 940 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 941 splx(s); 942 943 /* 944 * Decrement the outstanding request count. 945 */ 946 if (rep->r_flags & R_SENT) { 947 rep->r_flags &= ~R_SENT; /* paranoia */ 948 nmp->nm_sent -= NFS_CWNDSCALE; 949 } 950 951 /* 952 * If there was a successful reply and a tprintf msg. 953 * tprintf a response. 954 */ 955 if (!error && (rep->r_flags & R_TPRINTFMSG)) 956 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 957 "is alive again"); 958 mrep = rep->r_mrep; 959 md = rep->r_md; 960 dpos = rep->r_dpos; 961 if (error) { 962 m_freem(rep->r_mreq); 963 free((caddr_t)rep, M_NFSREQ); 964 return (error); 965 } 966 967 /* 968 * break down the rpc header and check if ok 969 */ 970 nfsm_dissect(tl, u_long *, 3*NFSX_UNSIGNED); 971 if (*tl++ == rpc_msgdenied) { 972 if (*tl == rpc_mismatch) 973 error = EOPNOTSUPP; 974 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 975 if (*tl == rpc_rejectedcred && failed_auth == 0) { 976 failed_auth++; 977 mheadend->m_next = (struct mbuf *)0; 978 m_freem(mrep); 979 m_freem(rep->r_mreq); 980 goto kerbauth; 981 } else 982 error = EAUTH; 983 } else 984 error = EACCES; 985 m_freem(mrep); 986 m_freem(rep->r_mreq); 987 free((caddr_t)rep, M_NFSREQ); 988 return (error); 989 } 990 991 /* 992 * skip over the auth_verf, someday we may want to cache auth_short's 993 * for nfs_reqhead(), but for now just dump it 994 */ 995 if (*++tl != 0) { 996 i = nfsm_rndup(fxdr_unsigned(long, *tl)); 997 nfsm_adv(i); 998 } 999 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1000 /* 0 == ok */ 1001 if (*tl == 0) { 1002 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1003 if (*tl != 0) { 1004 error = fxdr_unsigned(int, *tl); 1005 m_freem(mrep); 1006 if ((nmp->nm_flag & NFSMNT_NQNFS) && 1007 error == NQNFS_TRYLATER) { 1008 error = 0; 1009 waituntil = time.tv_sec + trylater_delay; 1010 while (time.tv_sec < waituntil) 1011 (void) tsleep((caddr_t)&lbolt, 1012 PSOCK, "nqnfstry", 0); 1013 trylater_delay *= nfs_backoff[trylater_cnt]; 1014 if (trylater_cnt < 7) 1015 trylater_cnt++; 1016 goto tryagain; 1017 } 1018 1019 /* 1020 * If the File Handle was stale, invalidate the 1021 * lookup cache, just in case. 1022 */ 1023 if (error == ESTALE) 1024 cache_purge(vp); 1025 m_freem(rep->r_mreq); 1026 free((caddr_t)rep, M_NFSREQ); 1027 return (error); 1028 } 1029 1030 /* 1031 * For nqnfs, get any lease in reply 1032 */ 1033 if (nmp->nm_flag & NFSMNT_NQNFS) { 1034 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1035 if (*tl) { 1036 np = VTONFS(vp); 1037 nqlflag = fxdr_unsigned(int, *tl); 1038 nfsm_dissect(tl, u_long *, 4*NFSX_UNSIGNED); 1039 cachable = fxdr_unsigned(int, *tl++); 1040 reqtime += fxdr_unsigned(int, *tl++); 1041 if (reqtime > time.tv_sec) { 1042 fxdr_hyper(tl, &frev); 1043 nqnfs_clientlease(nmp, np, nqlflag, 1044 cachable, reqtime, frev); 1045 } 1046 } 1047 } 1048 *mrp = mrep; 1049 *mdp = md; 1050 *dposp = dpos; 1051 m_freem(rep->r_mreq); 1052 FREE((caddr_t)rep, M_NFSREQ); 1053 return (0); 1054 } 1055 m_freem(mrep); 1056 m_freem(rep->r_mreq); 1057 free((caddr_t)rep, M_NFSREQ); 1058 error = EPROTONOSUPPORT; 1059 nfsmout: 1060 return (error); 1061 } 1062 #endif /* NFSCLIENT */ 1063 1064 /* 1065 * Generate the rpc reply header 1066 * siz arg. is used to decide if adding a cluster is worthwhile 1067 */ 1068 nfs_rephead(siz, nd, err, cache, frev, mrq, mbp, bposp) 1069 int siz; 1070 struct nfsd *nd; 1071 int err; 1072 int cache; 1073 u_quad_t *frev; 1074 struct mbuf **mrq; 1075 struct mbuf **mbp; 1076 caddr_t *bposp; 1077 { 1078 register u_long *tl; 1079 register struct mbuf *mreq; 1080 caddr_t bpos; 1081 struct mbuf *mb, *mb2; 1082 1083 MGETHDR(mreq, M_WAIT, MT_DATA); 1084 mb = mreq; 1085 /* 1086 * If this is a big reply, use a cluster else 1087 * try and leave leading space for the lower level headers. 1088 */ 1089 siz += RPC_REPLYSIZ; 1090 if (siz >= MINCLSIZE) { 1091 MCLGET(mreq, M_WAIT); 1092 } else 1093 mreq->m_data += max_hdr; 1094 tl = mtod(mreq, u_long *); 1095 mreq->m_len = 6*NFSX_UNSIGNED; 1096 bpos = ((caddr_t)tl)+mreq->m_len; 1097 *tl++ = txdr_unsigned(nd->nd_retxid); 1098 *tl++ = rpc_reply; 1099 if (err == ERPCMISMATCH || err == NQNFS_AUTHERR) { 1100 *tl++ = rpc_msgdenied; 1101 if (err == NQNFS_AUTHERR) { 1102 *tl++ = rpc_autherr; 1103 *tl = rpc_rejectedcred; 1104 mreq->m_len -= NFSX_UNSIGNED; 1105 bpos -= NFSX_UNSIGNED; 1106 } else { 1107 *tl++ = rpc_mismatch; 1108 *tl++ = txdr_unsigned(2); 1109 *tl = txdr_unsigned(2); 1110 } 1111 } else { 1112 *tl++ = rpc_msgaccepted; 1113 *tl++ = 0; 1114 *tl++ = 0; 1115 switch (err) { 1116 case EPROGUNAVAIL: 1117 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1118 break; 1119 case EPROGMISMATCH: 1120 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1121 nfsm_build(tl, u_long *, 2*NFSX_UNSIGNED); 1122 *tl++ = txdr_unsigned(2); 1123 *tl = txdr_unsigned(2); /* someday 3 */ 1124 break; 1125 case EPROCUNAVAIL: 1126 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1127 break; 1128 default: 1129 *tl = 0; 1130 if (err != VNOVAL) { 1131 nfsm_build(tl, u_long *, NFSX_UNSIGNED); 1132 if (err) 1133 *tl = txdr_unsigned(nfsrv_errmap[err - 1]); 1134 else 1135 *tl = 0; 1136 } 1137 break; 1138 }; 1139 } 1140 1141 /* 1142 * For nqnfs, piggyback lease as requested. 1143 */ 1144 if (nd->nd_nqlflag != NQL_NOVAL && err == 0) { 1145 if (nd->nd_nqlflag) { 1146 nfsm_build(tl, u_long *, 5*NFSX_UNSIGNED); 1147 *tl++ = txdr_unsigned(nd->nd_nqlflag); 1148 *tl++ = txdr_unsigned(cache); 1149 *tl++ = txdr_unsigned(nd->nd_duration); 1150 txdr_hyper(frev, tl); 1151 } else { 1152 if (nd->nd_nqlflag != 0) 1153 panic("nqreph"); 1154 nfsm_build(tl, u_long *, NFSX_UNSIGNED); 1155 *tl = 0; 1156 } 1157 } 1158 *mrq = mreq; 1159 *mbp = mb; 1160 *bposp = bpos; 1161 if (err != 0 && err != VNOVAL) 1162 nfsstats.srvrpc_errs++; 1163 return (0); 1164 } 1165 1166 /* 1167 * Nfs timer routine 1168 * Scan the nfsreq list and retranmit any requests that have timed out 1169 * To avoid retransmission attempts on STREAM sockets (in the future) make 1170 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1171 */ 1172 void 1173 nfs_timer(arg) 1174 void *arg; 1175 { 1176 register struct nfsreq *rep; 1177 register struct mbuf *m; 1178 register struct socket *so; 1179 register struct nfsmount *nmp; 1180 register int timeo; 1181 static long lasttime = 0; 1182 int s, error; 1183 1184 s = splsoftnet(); 1185 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1186 nmp = rep->r_nmp; 1187 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1188 continue; 1189 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1190 rep->r_flags |= R_SOFTTERM; 1191 continue; 1192 } 1193 if (rep->r_rtt >= 0) { 1194 rep->r_rtt++; 1195 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1196 timeo = nmp->nm_timeo; 1197 else 1198 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1199 if (nmp->nm_timeouts > 0) 1200 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1201 if (rep->r_rtt <= timeo) 1202 continue; 1203 if (nmp->nm_timeouts < 8) 1204 nmp->nm_timeouts++; 1205 } 1206 /* 1207 * Check for server not responding 1208 */ 1209 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1210 rep->r_rexmit > nmp->nm_deadthresh) { 1211 nfs_msg(rep->r_procp, 1212 nmp->nm_mountp->mnt_stat.f_mntfromname, 1213 "not responding"); 1214 rep->r_flags |= R_TPRINTFMSG; 1215 } 1216 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1217 nfsstats.rpctimeouts++; 1218 rep->r_flags |= R_SOFTTERM; 1219 continue; 1220 } 1221 if (nmp->nm_sotype != SOCK_DGRAM) { 1222 if (++rep->r_rexmit > NFS_MAXREXMIT) 1223 rep->r_rexmit = NFS_MAXREXMIT; 1224 continue; 1225 } 1226 if ((so = nmp->nm_so) == NULL) 1227 continue; 1228 1229 /* 1230 * If there is enough space and the window allows.. 1231 * Resend it 1232 * Set r_rtt to -1 in case we fail to send it now. 1233 */ 1234 rep->r_rtt = -1; 1235 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1236 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1237 (rep->r_flags & R_SENT) || 1238 nmp->nm_sent < nmp->nm_cwnd) && 1239 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1240 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1241 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1242 (struct mbuf *)0, (struct mbuf *)0); 1243 else 1244 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1245 nmp->nm_nam, (struct mbuf *)0); 1246 if (error) { 1247 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) 1248 so->so_error = 0; 1249 } else { 1250 /* 1251 * Iff first send, start timing 1252 * else turn timing off, backoff timer 1253 * and divide congestion window by 2. 1254 */ 1255 if (rep->r_flags & R_SENT) { 1256 rep->r_flags &= ~R_TIMING; 1257 if (++rep->r_rexmit > NFS_MAXREXMIT) 1258 rep->r_rexmit = NFS_MAXREXMIT; 1259 nmp->nm_cwnd >>= 1; 1260 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1261 nmp->nm_cwnd = NFS_CWNDSCALE; 1262 nfsstats.rpcretries++; 1263 } else { 1264 rep->r_flags |= R_SENT; 1265 nmp->nm_sent += NFS_CWNDSCALE; 1266 } 1267 rep->r_rtt = 0; 1268 } 1269 } 1270 } 1271 1272 #ifdef NFSSERVER 1273 /* 1274 * Call the nqnfs server timer once a second to handle leases. 1275 */ 1276 if (lasttime != time.tv_sec) { 1277 lasttime = time.tv_sec; 1278 nqnfs_serverd(); 1279 } 1280 #endif /* NFSSERVER */ 1281 splx(s); 1282 timeout(nfs_timer, (void *)0, hz / NFS_HZ); 1283 } 1284 1285 /* 1286 * Test for a termination condition pending on the process. 1287 * This is used for NFSMNT_INT mounts. 1288 */ 1289 nfs_sigintr(nmp, rep, p) 1290 struct nfsmount *nmp; 1291 struct nfsreq *rep; 1292 register struct proc *p; 1293 { 1294 1295 if (rep && (rep->r_flags & R_SOFTTERM)) 1296 return (EINTR); 1297 if (!(nmp->nm_flag & NFSMNT_INT)) 1298 return (0); 1299 if (p && p->p_siglist && 1300 (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) & 1301 NFSINT_SIGMASK)) 1302 return (EINTR); 1303 return (0); 1304 } 1305 1306 /* 1307 * Lock a socket against others. 1308 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1309 * and also to avoid race conditions between the processes with nfs requests 1310 * in progress when a reconnect is necessary. 1311 */ 1312 nfs_sndlock(flagp, rep) 1313 register int *flagp; 1314 struct nfsreq *rep; 1315 { 1316 struct proc *p; 1317 int slpflag = 0, slptimeo = 0; 1318 1319 if (rep) { 1320 p = rep->r_procp; 1321 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1322 slpflag = PCATCH; 1323 } else 1324 p = (struct proc *)0; 1325 while (*flagp & NFSMNT_SNDLOCK) { 1326 if (nfs_sigintr(rep->r_nmp, rep, p)) 1327 return (EINTR); 1328 *flagp |= NFSMNT_WANTSND; 1329 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1330 slptimeo); 1331 if (slpflag == PCATCH) { 1332 slpflag = 0; 1333 slptimeo = 2 * hz; 1334 } 1335 } 1336 *flagp |= NFSMNT_SNDLOCK; 1337 return (0); 1338 } 1339 1340 /* 1341 * Unlock the stream socket for others. 1342 */ 1343 void 1344 nfs_sndunlock(flagp) 1345 register int *flagp; 1346 { 1347 1348 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1349 panic("nfs sndunlock"); 1350 *flagp &= ~NFSMNT_SNDLOCK; 1351 if (*flagp & NFSMNT_WANTSND) { 1352 *flagp &= ~NFSMNT_WANTSND; 1353 wakeup((caddr_t)flagp); 1354 } 1355 } 1356 1357 nfs_rcvlock(rep) 1358 register struct nfsreq *rep; 1359 { 1360 register int *flagp = &rep->r_nmp->nm_flag; 1361 int slpflag, slptimeo = 0; 1362 1363 if (*flagp & NFSMNT_INT) 1364 slpflag = PCATCH; 1365 else 1366 slpflag = 0; 1367 while (*flagp & NFSMNT_RCVLOCK) { 1368 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1369 return (EINTR); 1370 *flagp |= NFSMNT_WANTRCV; 1371 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1372 slptimeo); 1373 if (slpflag == PCATCH) { 1374 slpflag = 0; 1375 slptimeo = 2 * hz; 1376 } 1377 } 1378 *flagp |= NFSMNT_RCVLOCK; 1379 return (0); 1380 } 1381 1382 /* 1383 * Unlock the stream socket for others. 1384 */ 1385 void 1386 nfs_rcvunlock(flagp) 1387 register int *flagp; 1388 { 1389 1390 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1391 panic("nfs rcvunlock"); 1392 *flagp &= ~NFSMNT_RCVLOCK; 1393 if (*flagp & NFSMNT_WANTRCV) { 1394 *flagp &= ~NFSMNT_WANTRCV; 1395 wakeup((caddr_t)flagp); 1396 } 1397 } 1398 1399 /* 1400 * Check for badly aligned mbuf data areas and 1401 * realign data in an mbuf list by copying the data areas up, as required. 1402 */ 1403 void 1404 nfs_realign(m, hsiz) 1405 register struct mbuf *m; 1406 int hsiz; 1407 { 1408 register struct mbuf *m2; 1409 register int siz, mlen, olen; 1410 register caddr_t tcp, fcp; 1411 struct mbuf *mnew; 1412 1413 while (m) { 1414 /* 1415 * This never happens for UDP, rarely happens for TCP 1416 * but frequently happens for iso transport. 1417 */ 1418 if ((m->m_len & 0x3) || (mtod(m, int) & 0x3)) { 1419 olen = m->m_len; 1420 fcp = mtod(m, caddr_t); 1421 if ((int)fcp & 0x3) { 1422 m->m_flags &= ~M_PKTHDR; 1423 if (m->m_flags & M_EXT) 1424 m->m_data = m->m_ext.ext_buf + 1425 ((m->m_ext.ext_size - olen) & ~0x3); 1426 else 1427 m->m_data = m->m_dat; 1428 } 1429 m->m_len = 0; 1430 tcp = mtod(m, caddr_t); 1431 mnew = m; 1432 m2 = m->m_next; 1433 1434 /* 1435 * If possible, only put the first invariant part 1436 * of the RPC header in the first mbuf. 1437 */ 1438 mlen = M_TRAILINGSPACE(m); 1439 if (olen <= hsiz && mlen > hsiz) 1440 mlen = hsiz; 1441 1442 /* 1443 * Loop through the mbuf list consolidating data. 1444 */ 1445 while (m) { 1446 while (olen > 0) { 1447 if (mlen == 0) { 1448 m2->m_flags &= ~M_PKTHDR; 1449 if (m2->m_flags & M_EXT) 1450 m2->m_data = m2->m_ext.ext_buf; 1451 else 1452 m2->m_data = m2->m_dat; 1453 m2->m_len = 0; 1454 mlen = M_TRAILINGSPACE(m2); 1455 tcp = mtod(m2, caddr_t); 1456 mnew = m2; 1457 m2 = m2->m_next; 1458 } 1459 siz = min(mlen, olen); 1460 if (tcp != fcp) 1461 bcopy(fcp, tcp, siz); 1462 mnew->m_len += siz; 1463 mlen -= siz; 1464 olen -= siz; 1465 tcp += siz; 1466 fcp += siz; 1467 } 1468 m = m->m_next; 1469 if (m) { 1470 olen = m->m_len; 1471 fcp = mtod(m, caddr_t); 1472 } 1473 } 1474 1475 /* 1476 * Finally, set m_len == 0 for any trailing mbufs that have 1477 * been copied out of. 1478 */ 1479 while (m2) { 1480 m2->m_len = 0; 1481 m2 = m2->m_next; 1482 } 1483 return; 1484 } 1485 m = m->m_next; 1486 } 1487 } 1488 1489 /* 1490 * Parse an RPC request 1491 * - verify it 1492 * - fill in the cred struct. 1493 */ 1494 nfs_getreq(nd, has_header) 1495 register struct nfsd *nd; 1496 int has_header; 1497 { 1498 register int len, i; 1499 register u_long *tl; 1500 register long t1; 1501 struct uio uio; 1502 struct iovec iov; 1503 caddr_t dpos, cp2; 1504 u_long nfsvers, auth_type; 1505 int error = 0, nqnfs = 0; 1506 struct mbuf *mrep, *md; 1507 1508 mrep = nd->nd_mrep; 1509 md = nd->nd_md; 1510 dpos = nd->nd_dpos; 1511 if (has_header) { 1512 nfsm_dissect(tl, u_long *, 10*NFSX_UNSIGNED); 1513 nd->nd_retxid = fxdr_unsigned(u_long, *tl++); 1514 if (*tl++ != rpc_call) { 1515 m_freem(mrep); 1516 return (EBADRPC); 1517 } 1518 } else { 1519 nfsm_dissect(tl, u_long *, 8*NFSX_UNSIGNED); 1520 } 1521 nd->nd_repstat = 0; 1522 if (*tl++ != rpc_vers) { 1523 nd->nd_repstat = ERPCMISMATCH; 1524 nd->nd_procnum = NFSPROC_NOOP; 1525 return (0); 1526 } 1527 nfsvers = nfs_vers; 1528 if (*tl != nfs_prog) { 1529 if (*tl == nqnfs_prog) { 1530 nqnfs++; 1531 nfsvers = nqnfs_vers; 1532 } else { 1533 nd->nd_repstat = EPROGUNAVAIL; 1534 nd->nd_procnum = NFSPROC_NOOP; 1535 return (0); 1536 } 1537 } 1538 tl++; 1539 if (*tl++ != nfsvers) { 1540 nd->nd_repstat = EPROGMISMATCH; 1541 nd->nd_procnum = NFSPROC_NOOP; 1542 return (0); 1543 } 1544 nd->nd_procnum = fxdr_unsigned(u_long, *tl++); 1545 if (nd->nd_procnum == NFSPROC_NULL) 1546 return (0); 1547 if (nd->nd_procnum >= NFS_NPROCS || 1548 (!nqnfs && nd->nd_procnum > NFSPROC_STATFS) || 1549 (*tl != rpc_auth_unix && *tl != rpc_auth_kerb)) { 1550 nd->nd_repstat = EPROCUNAVAIL; 1551 nd->nd_procnum = NFSPROC_NOOP; 1552 return (0); 1553 } 1554 auth_type = *tl++; 1555 len = fxdr_unsigned(int, *tl++); 1556 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1557 m_freem(mrep); 1558 return (EBADRPC); 1559 } 1560 1561 /* 1562 * Handle auth_unix or auth_kerb. 1563 */ 1564 if (auth_type == rpc_auth_unix) { 1565 len = fxdr_unsigned(int, *++tl); 1566 if (len < 0 || len > NFS_MAXNAMLEN) { 1567 m_freem(mrep); 1568 return (EBADRPC); 1569 } 1570 nfsm_adv(nfsm_rndup(len)); 1571 nfsm_dissect(tl, u_long *, 3*NFSX_UNSIGNED); 1572 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1573 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1574 len = fxdr_unsigned(int, *tl); 1575 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1576 m_freem(mrep); 1577 return (EBADRPC); 1578 } 1579 nfsm_dissect(tl, u_long *, (len + 2)*NFSX_UNSIGNED); 1580 for (i = 0; i < len; i++) 1581 if (i < NGROUPS) 1582 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1583 else 1584 tl++; 1585 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1586 } else if (auth_type == rpc_auth_kerb) { 1587 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1588 nd->nd_authlen = fxdr_unsigned(int, *tl); 1589 uio.uio_resid = nfsm_rndup(nd->nd_authlen); 1590 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1591 m_freem(mrep); 1592 return (EBADRPC); 1593 } 1594 uio.uio_offset = 0; 1595 uio.uio_iov = &iov; 1596 uio.uio_iovcnt = 1; 1597 uio.uio_segflg = UIO_SYSSPACE; 1598 iov.iov_base = (caddr_t)nd->nd_authstr; 1599 iov.iov_len = RPCAUTH_MAXSIZ; 1600 nfsm_mtouio(&uio, uio.uio_resid); 1601 nfsm_dissect(tl, u_long *, 2 * NFSX_UNSIGNED); 1602 nd->nd_flag |= NFSD_NEEDAUTH; 1603 } 1604 1605 /* 1606 * Do we have any use for the verifier. 1607 * According to the "Remote Procedure Call Protocol Spec." it 1608 * should be AUTH_NULL, but some clients make it AUTH_UNIX? 1609 * For now, just skip over it 1610 */ 1611 len = fxdr_unsigned(int, *++tl); 1612 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1613 m_freem(mrep); 1614 return (EBADRPC); 1615 } 1616 if (len > 0) { 1617 nfsm_adv(nfsm_rndup(len)); 1618 } 1619 1620 /* 1621 * For nqnfs, get piggybacked lease request. 1622 */ 1623 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1624 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1625 nd->nd_nqlflag = fxdr_unsigned(int, *tl); 1626 if (nd->nd_nqlflag) { 1627 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1628 nd->nd_duration = fxdr_unsigned(int, *tl); 1629 } else 1630 nd->nd_duration = NQ_MINLEASE; 1631 } else { 1632 nd->nd_nqlflag = NQL_NOVAL; 1633 nd->nd_duration = NQ_MINLEASE; 1634 } 1635 nd->nd_md = md; 1636 nd->nd_dpos = dpos; 1637 return (0); 1638 nfsmout: 1639 return (error); 1640 } 1641 1642 nfs_msg(p, server, msg) 1643 struct proc *p; 1644 char *server, *msg; 1645 { 1646 tpr_t tpr; 1647 1648 if (p) 1649 tpr = tprintf_open(p); 1650 else 1651 tpr = NULL; 1652 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1653 tprintf_close(tpr); 1654 } 1655 1656 #ifdef NFSSERVER 1657 int nfsrv_null(), 1658 nfsrv_getattr(), 1659 nfsrv_setattr(), 1660 nfsrv_lookup(), 1661 nfsrv_readlink(), 1662 nfsrv_read(), 1663 nfsrv_write(), 1664 nfsrv_create(), 1665 nfsrv_remove(), 1666 nfsrv_rename(), 1667 nfsrv_link(), 1668 nfsrv_symlink(), 1669 nfsrv_mkdir(), 1670 nfsrv_rmdir(), 1671 nfsrv_readdir(), 1672 nfsrv_statfs(), 1673 nfsrv_noop(), 1674 nqnfsrv_readdirlook(), 1675 nqnfsrv_getlease(), 1676 nqnfsrv_vacated(), 1677 nqnfsrv_access(); 1678 1679 int (*nfsrv_procs[NFS_NPROCS])() = { 1680 nfsrv_null, 1681 nfsrv_getattr, 1682 nfsrv_setattr, 1683 nfsrv_noop, 1684 nfsrv_lookup, 1685 nfsrv_readlink, 1686 nfsrv_read, 1687 nfsrv_noop, 1688 nfsrv_write, 1689 nfsrv_create, 1690 nfsrv_remove, 1691 nfsrv_rename, 1692 nfsrv_link, 1693 nfsrv_symlink, 1694 nfsrv_mkdir, 1695 nfsrv_rmdir, 1696 nfsrv_readdir, 1697 nfsrv_statfs, 1698 nqnfsrv_readdirlook, 1699 nqnfsrv_getlease, 1700 nqnfsrv_vacated, 1701 nfsrv_noop, 1702 nqnfsrv_access, 1703 }; 1704 1705 /* 1706 * Socket upcall routine for the nfsd sockets. 1707 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1708 * Essentially do as much as possible non-blocking, else punt and it will 1709 * be called with M_WAIT from an nfsd. 1710 */ 1711 void 1712 nfsrv_rcv(so, arg, waitflag) 1713 struct socket *so; 1714 caddr_t arg; 1715 int waitflag; 1716 { 1717 register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1718 register struct mbuf *m; 1719 struct mbuf *mp, *nam; 1720 struct uio auio; 1721 int flags, error; 1722 1723 if ((slp->ns_flag & SLP_VALID) == 0) 1724 return; 1725 #ifdef notdef 1726 /* 1727 * Define this to test for nfsds handling this under heavy load. 1728 */ 1729 if (waitflag == M_DONTWAIT) { 1730 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1731 } 1732 #endif 1733 auio.uio_procp = NULL; 1734 if (so->so_type == SOCK_STREAM) { 1735 /* 1736 * If there are already records on the queue, defer soreceive() 1737 * to an nfsd so that there is feedback to the TCP layer that 1738 * the nfs servers are heavily loaded. 1739 */ 1740 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1741 slp->ns_flag |= SLP_NEEDQ; 1742 goto dorecs; 1743 } 1744 1745 /* 1746 * Do soreceive(). 1747 */ 1748 auio.uio_resid = 1000000000; 1749 flags = MSG_DONTWAIT; 1750 error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1751 if (error || mp == (struct mbuf *)0) { 1752 if (error == EWOULDBLOCK) 1753 slp->ns_flag |= SLP_NEEDQ; 1754 else 1755 slp->ns_flag |= SLP_DISCONN; 1756 goto dorecs; 1757 } 1758 m = mp; 1759 if (slp->ns_rawend) { 1760 slp->ns_rawend->m_next = m; 1761 slp->ns_cc += 1000000000 - auio.uio_resid; 1762 } else { 1763 slp->ns_raw = m; 1764 slp->ns_cc = 1000000000 - auio.uio_resid; 1765 } 1766 while (m->m_next) 1767 m = m->m_next; 1768 slp->ns_rawend = m; 1769 1770 /* 1771 * Now try and parse record(s) out of the raw stream data. 1772 */ 1773 if (error = nfsrv_getstream(slp, waitflag)) { 1774 if (error == EPERM) 1775 slp->ns_flag |= SLP_DISCONN; 1776 else 1777 slp->ns_flag |= SLP_NEEDQ; 1778 } 1779 } else { 1780 do { 1781 auio.uio_resid = 1000000000; 1782 flags = MSG_DONTWAIT; 1783 error = soreceive(so, &nam, &auio, &mp, 1784 (struct mbuf **)0, &flags); 1785 if (mp) { 1786 nfs_realign(mp, 10 * NFSX_UNSIGNED); 1787 if (nam) { 1788 m = nam; 1789 m->m_next = mp; 1790 } else 1791 m = mp; 1792 if (slp->ns_recend) 1793 slp->ns_recend->m_nextpkt = m; 1794 else 1795 slp->ns_rec = m; 1796 slp->ns_recend = m; 1797 m->m_nextpkt = (struct mbuf *)0; 1798 } 1799 if (error) { 1800 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1801 && error != EWOULDBLOCK) { 1802 slp->ns_flag |= SLP_DISCONN; 1803 goto dorecs; 1804 } 1805 } 1806 } while (mp); 1807 } 1808 1809 /* 1810 * Now try and process the request records, non-blocking. 1811 */ 1812 dorecs: 1813 if (waitflag == M_DONTWAIT && 1814 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1815 nfsrv_wakenfsd(slp); 1816 } 1817 1818 /* 1819 * Try and extract an RPC request from the mbuf data list received on a 1820 * stream socket. The "waitflag" argument indicates whether or not it 1821 * can sleep. 1822 */ 1823 nfsrv_getstream(slp, waitflag) 1824 register struct nfssvc_sock *slp; 1825 int waitflag; 1826 { 1827 register struct mbuf *m; 1828 register char *cp1, *cp2; 1829 register int len; 1830 struct mbuf *om, *m2, *recm; 1831 u_long recmark; 1832 1833 if (slp->ns_flag & SLP_GETSTREAM) 1834 panic("nfs getstream"); 1835 slp->ns_flag |= SLP_GETSTREAM; 1836 for (;;) { 1837 if (slp->ns_reclen == 0) { 1838 if (slp->ns_cc < NFSX_UNSIGNED) { 1839 slp->ns_flag &= ~SLP_GETSTREAM; 1840 return (0); 1841 } 1842 m = slp->ns_raw; 1843 if (m->m_len >= NFSX_UNSIGNED) { 1844 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 1845 m->m_data += NFSX_UNSIGNED; 1846 m->m_len -= NFSX_UNSIGNED; 1847 } else { 1848 cp1 = (caddr_t)&recmark; 1849 cp2 = mtod(m, caddr_t); 1850 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1851 while (m->m_len == 0) { 1852 m = m->m_next; 1853 cp2 = mtod(m, caddr_t); 1854 } 1855 *cp1++ = *cp2++; 1856 m->m_data++; 1857 m->m_len--; 1858 } 1859 } 1860 slp->ns_cc -= NFSX_UNSIGNED; 1861 slp->ns_reclen = ntohl(recmark) & ~0x80000000; 1862 if (slp->ns_reclen < NFS_MINPACKET || slp->ns_reclen > NFS_MAXPACKET) { 1863 slp->ns_flag &= ~SLP_GETSTREAM; 1864 return (EPERM); 1865 } 1866 } 1867 1868 /* 1869 * Now get the record part. 1870 */ 1871 if (slp->ns_cc == slp->ns_reclen) { 1872 recm = slp->ns_raw; 1873 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 1874 slp->ns_cc = slp->ns_reclen = 0; 1875 } else if (slp->ns_cc > slp->ns_reclen) { 1876 len = 0; 1877 m = slp->ns_raw; 1878 om = (struct mbuf *)0; 1879 while (len < slp->ns_reclen) { 1880 if ((len + m->m_len) > slp->ns_reclen) { 1881 m2 = m_copym(m, 0, slp->ns_reclen - len, 1882 waitflag); 1883 if (m2) { 1884 if (om) { 1885 om->m_next = m2; 1886 recm = slp->ns_raw; 1887 } else 1888 recm = m2; 1889 m->m_data += slp->ns_reclen - len; 1890 m->m_len -= slp->ns_reclen - len; 1891 len = slp->ns_reclen; 1892 } else { 1893 slp->ns_flag &= ~SLP_GETSTREAM; 1894 return (EWOULDBLOCK); 1895 } 1896 } else if ((len + m->m_len) == slp->ns_reclen) { 1897 om = m; 1898 len += m->m_len; 1899 m = m->m_next; 1900 recm = slp->ns_raw; 1901 om->m_next = (struct mbuf *)0; 1902 } else { 1903 om = m; 1904 len += m->m_len; 1905 m = m->m_next; 1906 } 1907 } 1908 slp->ns_raw = m; 1909 slp->ns_cc -= len; 1910 slp->ns_reclen = 0; 1911 } else { 1912 slp->ns_flag &= ~SLP_GETSTREAM; 1913 return (0); 1914 } 1915 nfs_realign(recm, 10 * NFSX_UNSIGNED); 1916 if (slp->ns_recend) 1917 slp->ns_recend->m_nextpkt = recm; 1918 else 1919 slp->ns_rec = recm; 1920 slp->ns_recend = recm; 1921 } 1922 } 1923 1924 /* 1925 * Parse an RPC header. 1926 */ 1927 nfsrv_dorec(slp, nd) 1928 register struct nfssvc_sock *slp; 1929 register struct nfsd *nd; 1930 { 1931 register struct mbuf *m; 1932 int error; 1933 1934 if ((slp->ns_flag & SLP_VALID) == 0 || 1935 (m = slp->ns_rec) == (struct mbuf *)0) 1936 return (ENOBUFS); 1937 if (slp->ns_rec = m->m_nextpkt) 1938 m->m_nextpkt = (struct mbuf *)0; 1939 else 1940 slp->ns_recend = (struct mbuf *)0; 1941 if (m->m_type == MT_SONAME) { 1942 nd->nd_nam = m; 1943 nd->nd_md = nd->nd_mrep = m->m_next; 1944 m->m_next = (struct mbuf *)0; 1945 } else { 1946 nd->nd_nam = (struct mbuf *)0; 1947 nd->nd_md = nd->nd_mrep = m; 1948 } 1949 nd->nd_dpos = mtod(nd->nd_md, caddr_t); 1950 if (error = nfs_getreq(nd, TRUE)) { 1951 m_freem(nd->nd_nam); 1952 return (error); 1953 } 1954 return (0); 1955 } 1956 1957 /* 1958 * Search for a sleeping nfsd and wake it up. 1959 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 1960 * running nfsds will go look for the work in the nfssvc_sock list. 1961 */ 1962 void 1963 nfsrv_wakenfsd(slp) 1964 struct nfssvc_sock *slp; 1965 { 1966 register struct nfsd *nd; 1967 1968 if ((slp->ns_flag & SLP_VALID) == 0) 1969 return; 1970 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nd_chain.tqe_next) { 1971 if (nd->nd_flag & NFSD_WAITING) { 1972 nd->nd_flag &= ~NFSD_WAITING; 1973 if (nd->nd_slp) 1974 panic("nfsd wakeup"); 1975 slp->ns_sref++; 1976 nd->nd_slp = slp; 1977 wakeup((caddr_t)nd); 1978 return; 1979 } 1980 } 1981 slp->ns_flag |= SLP_DOREC; 1982 nfsd_head_flag |= NFSD_CHECKSLP; 1983 } 1984 #endif /* NFSSERVER */ 1985