nfs_socket.c revision 1.23
1 /* $NetBSD: nfs_socket.c,v 1.23 1996/02/09 21:48:29 christos 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 #include <sys/namei.h> 59 60 #include <netinet/in.h> 61 #include <netinet/tcp.h> 62 #include <nfs/rpcv2.h> 63 #include <nfs/nfsv2.h> 64 #include <nfs/nfs.h> 65 #include <nfs/xdr_subs.h> 66 #include <nfs/nfsm_subs.h> 67 #include <nfs/nfsmount.h> 68 #include <nfs/nfsnode.h> 69 #include <nfs/nfsrtt.h> 70 #include <nfs/nqnfs.h> 71 #include <nfs/nfs_var.h> 72 73 #define TRUE 1 74 #define FALSE 0 75 76 /* 77 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 78 * Use the mean and mean deviation of rtt for the appropriate type of rpc 79 * for the frequent rpcs and a default for the others. 80 * The justification for doing "other" this way is that these rpcs 81 * happen so infrequently that timer est. would probably be stale. 82 * Also, since many of these rpcs are 83 * non-idempotent, a conservative timeout is desired. 84 * getattr, lookup - A+2D 85 * read, write - A+4D 86 * other - nm_timeo 87 */ 88 #define NFS_RTO(n, t) \ 89 ((t) == 0 ? (n)->nm_timeo : \ 90 ((t) < 3 ? \ 91 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 92 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 93 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 94 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 95 /* 96 * External data, mostly RPC constants in XDR form 97 */ 98 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 99 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, rpc_rejectedcred, 100 rpc_auth_kerb; 101 extern u_int32_t nfs_prog, nfs_vers, nqnfs_prog, nqnfs_vers; 102 extern time_t nqnfsstarttime; 103 extern int nonidempotent[NFS_NPROCS]; 104 105 /* 106 * Maps errno values to nfs error numbers. 107 * Use NFSERR_IO as the catch all for ones not specifically defined in 108 * RFC 1094. 109 */ 110 static int nfsrv_errmap[ELAST] = { 111 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, 112 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 113 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, 114 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR, 115 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 116 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, 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_IO, NFSERR_IO, NFSERR_IO, 122 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 123 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO, 124 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, 125 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 126 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 127 NFSERR_IO, 128 }; 129 130 /* 131 * Defines which timer to use for the procnum. 132 * 0 - default 133 * 1 - getattr 134 * 2 - lookup 135 * 3 - read 136 * 4 - write 137 */ 138 static int proct[NFS_NPROCS] = { 139 0, 1, 0, 0, 2, 3, 3, 0, 4, 0, 0, 0, 0, 0, 0, 0, 3, 0, 3, 0, 0, 0, 0, 140 }; 141 142 /* 143 * There is a congestion window for outstanding rpcs maintained per mount 144 * point. The cwnd size is adjusted in roughly the way that: 145 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 146 * SIGCOMM '88". ACM, August 1988. 147 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 148 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 149 * of rpcs is in progress. 150 * (The sent count and cwnd are scaled for integer arith.) 151 * Variants of "slow start" were tried and were found to be too much of a 152 * performance hit (ave. rtt 3 times larger), 153 * I suspect due to the large rtt that nfs rpcs have. 154 */ 155 #define NFS_CWNDSCALE 256 156 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 157 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 158 int nfsrtton = 0; 159 struct nfsrtt nfsrtt; 160 161 /* 162 * Initialize sockets and congestion for a new NFS connection. 163 * We do not free the sockaddr if error. 164 */ 165 int 166 nfs_connect(nmp, rep) 167 register struct nfsmount *nmp; 168 struct nfsreq *rep; 169 { 170 register struct socket *so; 171 int s, error, rcvreserve, sndreserve; 172 struct sockaddr *saddr; 173 struct sockaddr_in *sin; 174 struct mbuf *m; 175 u_int16_t tport; 176 177 nmp->nm_so = (struct socket *)0; 178 saddr = mtod(nmp->nm_nam, struct sockaddr *); 179 error = socreate(saddr->sa_family, 180 &nmp->nm_so, nmp->nm_sotype, nmp->nm_soproto); 181 if (error) 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)) != 0) 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, int32_t *) = 1; 262 m->m_len = sizeof(int32_t); 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, int32_t *) = 1; 268 m->m_len = sizeof(int32_t); 269 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 270 } 271 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 272 sizeof (u_int32_t)) * 2; 273 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 274 sizeof (u_int32_t)) * 2; 275 } 276 if ((error = soreserve(so, sndreserve, rcvreserve)) != 0) 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 int 306 nfs_reconnect(rep) 307 register struct nfsreq *rep; 308 { 309 register struct nfsreq *rp; 310 register struct nfsmount *nmp = rep->r_nmp; 311 int error; 312 313 nfs_disconnect(nmp); 314 while ((error = nfs_connect(nmp, rep)) != 0) { 315 if (error == EINTR || error == ERESTART) 316 return (EINTR); 317 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); 318 } 319 320 /* 321 * Loop through outstanding request list and fix up all requests 322 * on old socket. 323 */ 324 for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) { 325 if (rp->r_nmp == nmp) 326 rp->r_flags |= R_MUSTRESEND; 327 } 328 return (0); 329 } 330 331 /* 332 * NFS disconnect. Clean up and unlink. 333 */ 334 void 335 nfs_disconnect(nmp) 336 register struct nfsmount *nmp; 337 { 338 register struct socket *so; 339 340 if (nmp->nm_so) { 341 so = nmp->nm_so; 342 nmp->nm_so = (struct socket *)0; 343 soshutdown(so, 2); 344 soclose(so); 345 } 346 } 347 348 /* 349 * This is the nfs send routine. For connection based socket types, it 350 * must be called with an nfs_sndlock() on the socket. 351 * "rep == NULL" indicates that it has been called from a server. 352 * For the client side: 353 * - return EINTR if the RPC is terminated, 0 otherwise 354 * - set R_MUSTRESEND if the send fails for any reason 355 * - do any cleanup required by recoverable socket errors (???) 356 * For the server side: 357 * - return EINTR or ERESTART if interrupted by a signal 358 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 359 * - do any cleanup required by recoverable socket errors (???) 360 */ 361 int 362 nfs_send(so, nam, top, rep) 363 register struct socket *so; 364 struct mbuf *nam; 365 register struct mbuf *top; 366 struct nfsreq *rep; 367 { 368 struct mbuf *sendnam; 369 int error, soflags, flags; 370 371 if (rep) { 372 if (rep->r_flags & R_SOFTTERM) { 373 m_freem(top); 374 return (EINTR); 375 } 376 if ((so = rep->r_nmp->nm_so) == NULL) { 377 rep->r_flags |= R_MUSTRESEND; 378 m_freem(top); 379 return (0); 380 } 381 rep->r_flags &= ~R_MUSTRESEND; 382 soflags = rep->r_nmp->nm_soflags; 383 } else 384 soflags = so->so_proto->pr_flags; 385 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 386 sendnam = (struct mbuf *)0; 387 else 388 sendnam = nam; 389 if (so->so_type == SOCK_SEQPACKET) 390 flags = MSG_EOR; 391 else 392 flags = 0; 393 394 error = sosend(so, sendnam, (struct uio *)0, top, 395 (struct mbuf *)0, flags); 396 if (error) { 397 if (rep) { 398 log(LOG_INFO, "nfs send error %d for server %s\n",error, 399 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 400 /* 401 * Deal with errors for the client side. 402 */ 403 if (rep->r_flags & R_SOFTTERM) 404 error = EINTR; 405 else 406 rep->r_flags |= R_MUSTRESEND; 407 } else 408 log(LOG_INFO, "nfsd send error %d\n", error); 409 410 /* 411 * Handle any recoverable (soft) socket errors here. (???) 412 */ 413 if (error != EINTR && error != ERESTART && 414 error != EWOULDBLOCK && error != EPIPE) 415 error = 0; 416 } 417 return (error); 418 } 419 420 #ifdef NFSCLIENT 421 /* 422 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 423 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 424 * Mark and consolidate the data into a new mbuf list. 425 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 426 * small mbufs. 427 * For SOCK_STREAM we must be very careful to read an entire record once 428 * we have read any of it, even if the system call has been interrupted. 429 */ 430 int 431 nfs_receive(rep, aname, mp) 432 register struct nfsreq *rep; 433 struct mbuf **aname; 434 struct mbuf **mp; 435 { 436 register struct socket *so; 437 struct uio auio; 438 struct iovec aio; 439 register struct mbuf *m; 440 struct mbuf *control; 441 u_int32_t len; 442 struct mbuf **getnam; 443 int error, sotype, rcvflg; 444 struct proc *p = curproc; /* XXX */ 445 446 /* 447 * Set up arguments for soreceive() 448 */ 449 *mp = (struct mbuf *)0; 450 *aname = (struct mbuf *)0; 451 sotype = rep->r_nmp->nm_sotype; 452 453 /* 454 * For reliable protocols, lock against other senders/receivers 455 * in case a reconnect is necessary. 456 * For SOCK_STREAM, first get the Record Mark to find out how much 457 * more there is to get. 458 * We must lock the socket against other receivers 459 * until we have an entire rpc request/reply. 460 */ 461 if (sotype != SOCK_DGRAM) { 462 if ((error = nfs_sndlock(&rep->r_nmp->nm_flag, rep)) != 0) 463 return (error); 464 tryagain: 465 /* 466 * Check for fatal errors and resending request. 467 */ 468 /* 469 * Ugh: If a reconnect attempt just happened, nm_so 470 * would have changed. NULL indicates a failed 471 * attempt that has essentially shut down this 472 * mount point. 473 */ 474 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 475 nfs_sndunlock(&rep->r_nmp->nm_flag); 476 return (EINTR); 477 } 478 if ((so = rep->r_nmp->nm_so) == NULL) { 479 if ((error = nfs_reconnect(rep)) != 0) { 480 nfs_sndunlock(&rep->r_nmp->nm_flag); 481 return (error); 482 } 483 goto tryagain; 484 } 485 while (rep->r_flags & R_MUSTRESEND) { 486 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 487 nfsstats.rpcretries++; 488 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 489 if (error) { 490 if (error == EINTR || error == ERESTART || 491 (error = nfs_reconnect(rep))) { 492 nfs_sndunlock(&rep->r_nmp->nm_flag); 493 return (error); 494 } 495 goto tryagain; 496 } 497 } 498 nfs_sndunlock(&rep->r_nmp->nm_flag); 499 if (sotype == SOCK_STREAM) { 500 aio.iov_base = (caddr_t) &len; 501 aio.iov_len = sizeof(u_int32_t); 502 auio.uio_iov = &aio; 503 auio.uio_iovcnt = 1; 504 auio.uio_segflg = UIO_SYSSPACE; 505 auio.uio_rw = UIO_READ; 506 auio.uio_offset = 0; 507 auio.uio_resid = sizeof(u_int32_t); 508 auio.uio_procp = p; 509 do { 510 rcvflg = MSG_WAITALL; 511 error = soreceive(so, (struct mbuf **)0, &auio, 512 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 513 if (error == EWOULDBLOCK && rep) { 514 if (rep->r_flags & R_SOFTTERM) 515 return (EINTR); 516 } 517 } while (error == EWOULDBLOCK); 518 if (!error && auio.uio_resid > 0) { 519 log(LOG_INFO, 520 "short receive (%d/%d) from nfs server %s\n", 521 sizeof(u_int32_t) - auio.uio_resid, 522 sizeof(u_int32_t), 523 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 524 error = EPIPE; 525 } 526 if (error) 527 goto errout; 528 len = ntohl(len) & ~0x80000000; 529 /* 530 * This is SERIOUS! We are out of sync with the sender 531 * and forcing a disconnect/reconnect is all I can do. 532 */ 533 if (len > NFS_MAXPACKET) { 534 log(LOG_ERR, "%s (%d) from nfs server %s\n", 535 "impossible packet length", 536 len, 537 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 538 error = EFBIG; 539 goto errout; 540 } 541 auio.uio_resid = len; 542 do { 543 rcvflg = MSG_WAITALL; 544 error = soreceive(so, (struct mbuf **)0, 545 &auio, mp, (struct mbuf **)0, &rcvflg); 546 } while (error == EWOULDBLOCK || error == EINTR || 547 error == ERESTART); 548 if (!error && auio.uio_resid > 0) { 549 log(LOG_INFO, 550 "short receive (%d/%d) from nfs server %s\n", 551 len - auio.uio_resid, len, 552 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 553 error = EPIPE; 554 } 555 } else { 556 /* 557 * NB: Since uio_resid is big, MSG_WAITALL is ignored 558 * and soreceive() will return when it has either a 559 * control msg or a data msg. 560 * We have no use for control msg., but must grab them 561 * and then throw them away so we know what is going 562 * on. 563 */ 564 auio.uio_resid = len = 100000000; /* Anything Big */ 565 auio.uio_procp = p; 566 do { 567 rcvflg = 0; 568 error = soreceive(so, (struct mbuf **)0, 569 &auio, mp, &control, &rcvflg); 570 if (control) 571 m_freem(control); 572 if (error == EWOULDBLOCK && rep) { 573 if (rep->r_flags & R_SOFTTERM) 574 return (EINTR); 575 } 576 } while (error == EWOULDBLOCK || 577 (!error && *mp == NULL && control)); 578 if ((rcvflg & MSG_EOR) == 0) 579 printf("Egad!!\n"); 580 if (!error && *mp == NULL) 581 error = EPIPE; 582 len -= auio.uio_resid; 583 } 584 errout: 585 if (error && error != EINTR && error != ERESTART) { 586 m_freem(*mp); 587 *mp = (struct mbuf *)0; 588 if (error != EPIPE) 589 log(LOG_INFO, 590 "receive error %d from nfs server %s\n", 591 error, 592 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 593 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 594 if (!error) 595 error = nfs_reconnect(rep); 596 if (!error) 597 goto tryagain; 598 } 599 } else { 600 if ((so = rep->r_nmp->nm_so) == NULL) 601 return (EACCES); 602 if (so->so_state & SS_ISCONNECTED) 603 getnam = (struct mbuf **)0; 604 else 605 getnam = aname; 606 auio.uio_resid = len = 1000000; 607 auio.uio_procp = p; 608 do { 609 rcvflg = 0; 610 error = soreceive(so, getnam, &auio, mp, 611 (struct mbuf **)0, &rcvflg); 612 if (error == EWOULDBLOCK && 613 (rep->r_flags & R_SOFTTERM)) 614 return (EINTR); 615 } while (error == EWOULDBLOCK); 616 len -= auio.uio_resid; 617 } 618 if (error) { 619 m_freem(*mp); 620 *mp = (struct mbuf *)0; 621 } 622 /* 623 * Search for any mbufs that are not a multiple of 4 bytes long 624 * or with m_data not longword aligned. 625 * These could cause pointer alignment problems, so copy them to 626 * well aligned mbufs. 627 */ 628 nfs_realign(*mp, 5 * NFSX_UNSIGNED); 629 return (error); 630 } 631 632 /* 633 * Implement receipt of reply on a socket. 634 * We must search through the list of received datagrams matching them 635 * with outstanding requests using the xid, until ours is found. 636 */ 637 /* ARGSUSED */ 638 int 639 nfs_reply(myrep) 640 struct nfsreq *myrep; 641 { 642 register struct nfsreq *rep; 643 register struct nfsmount *nmp = myrep->r_nmp; 644 register int32_t t1; 645 struct mbuf *mrep, *nam, *md; 646 u_int32_t rxid, *tl; 647 caddr_t dpos, cp2; 648 int error; 649 650 /* 651 * Loop around until we get our own reply 652 */ 653 for (;;) { 654 /* 655 * Lock against other receivers so that I don't get stuck in 656 * sbwait() after someone else has received my reply for me. 657 * Also necessary for connection based protocols to avoid 658 * race conditions during a reconnect. 659 */ 660 if ((error = nfs_rcvlock(myrep)) != 0) 661 return (error); 662 /* Already received, bye bye */ 663 if (myrep->r_mrep != NULL) { 664 nfs_rcvunlock(&nmp->nm_flag); 665 return (0); 666 } 667 /* 668 * Get the next Rpc reply off the socket 669 */ 670 error = nfs_receive(myrep, &nam, &mrep); 671 nfs_rcvunlock(&nmp->nm_flag); 672 if (error) { 673 674 /* 675 * Ignore routing errors on connectionless protocols?? 676 */ 677 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 678 nmp->nm_so->so_error = 0; 679 if (myrep->r_flags & R_GETONEREP) 680 return (0); 681 continue; 682 } 683 return (error); 684 } 685 if (nam) 686 m_freem(nam); 687 688 /* 689 * Get the xid and check that it is an rpc reply 690 */ 691 md = mrep; 692 dpos = mtod(md, caddr_t); 693 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 694 rxid = *tl++; 695 if (*tl != rpc_reply) { 696 if (nmp->nm_flag & NFSMNT_NQNFS) { 697 if (nqnfs_callback(nmp, mrep, md, dpos)) 698 nfsstats.rpcinvalid++; 699 } else { 700 nfsstats.rpcinvalid++; 701 m_freem(mrep); 702 } 703 nfsmout: 704 if (myrep->r_flags & R_GETONEREP) 705 return (0); 706 continue; 707 } 708 709 /* 710 * Loop through the request list to match up the reply 711 * Iff no match, just drop the datagram 712 */ 713 for (rep = nfs_reqq.tqh_first; rep != 0; 714 rep = rep->r_chain.tqe_next) { 715 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 716 /* Found it.. */ 717 rep->r_mrep = mrep; 718 rep->r_md = md; 719 rep->r_dpos = dpos; 720 if (nfsrtton) { 721 struct rttl *rt; 722 723 rt = &nfsrtt.rttl[nfsrtt.pos]; 724 rt->proc = rep->r_procnum; 725 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 726 rt->sent = nmp->nm_sent; 727 rt->cwnd = nmp->nm_cwnd; 728 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 729 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 730 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 731 rt->tstamp = time; 732 if (rep->r_flags & R_TIMING) 733 rt->rtt = rep->r_rtt; 734 else 735 rt->rtt = 1000000; 736 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 737 } 738 /* 739 * Update congestion window. 740 * Do the additive increase of 741 * one rpc/rtt. 742 */ 743 if (nmp->nm_cwnd <= nmp->nm_sent) { 744 nmp->nm_cwnd += 745 (NFS_CWNDSCALE * NFS_CWNDSCALE + 746 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 747 if (nmp->nm_cwnd > NFS_MAXCWND) 748 nmp->nm_cwnd = NFS_MAXCWND; 749 } 750 rep->r_flags &= ~R_SENT; 751 nmp->nm_sent -= NFS_CWNDSCALE; 752 /* 753 * Update rtt using a gain of 0.125 on the mean 754 * and a gain of 0.25 on the deviation. 755 */ 756 if (rep->r_flags & R_TIMING) { 757 /* 758 * Since the timer resolution of 759 * NFS_HZ is so course, it can often 760 * result in r_rtt == 0. Since 761 * r_rtt == N means that the actual 762 * rtt is between N+dt and N+2-dt ticks, 763 * add 1. 764 */ 765 t1 = rep->r_rtt + 1; 766 t1 -= (NFS_SRTT(rep) >> 3); 767 NFS_SRTT(rep) += t1; 768 if (t1 < 0) 769 t1 = -t1; 770 t1 -= (NFS_SDRTT(rep) >> 2); 771 NFS_SDRTT(rep) += t1; 772 } 773 nmp->nm_timeouts = 0; 774 break; 775 } 776 } 777 /* 778 * If not matched to a request, drop it. 779 * If it's mine, get out. 780 */ 781 if (rep == 0) { 782 nfsstats.rpcunexpected++; 783 m_freem(mrep); 784 } else if (rep == myrep) { 785 if (rep->r_mrep == NULL) 786 panic("nfsreply nil"); 787 return (0); 788 } 789 if (myrep->r_flags & R_GETONEREP) 790 return (0); 791 } 792 } 793 794 /* 795 * nfs_request - goes something like this 796 * - fill in request struct 797 * - links it into list 798 * - calls nfs_send() for first transmit 799 * - calls nfs_receive() to get reply 800 * - break down rpc header and return with nfs reply pointed to 801 * by mrep or error 802 * nb: always frees up mreq mbuf list 803 */ 804 int 805 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 806 struct vnode *vp; 807 struct mbuf *mrest; 808 int procnum; 809 struct proc *procp; 810 struct ucred *cred; 811 struct mbuf **mrp; 812 struct mbuf **mdp; 813 caddr_t *dposp; 814 { 815 register struct mbuf *m, *mrep; 816 register struct nfsreq *rep; 817 register u_int32_t *tl; 818 register int i; 819 struct nfsmount *nmp; 820 struct mbuf *md, *mheadend; 821 struct nfsnode *np; 822 time_t reqtime, waituntil; 823 caddr_t dpos, cp2; 824 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; 825 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 826 u_int32_t xid; 827 u_quad_t frev; 828 char *auth_str; 829 830 nmp = VFSTONFS(vp->v_mount); 831 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 832 rep->r_nmp = nmp; 833 rep->r_vp = vp; 834 rep->r_procp = procp; 835 rep->r_procnum = procnum; 836 i = 0; 837 m = mrest; 838 while (m) { 839 i += m->m_len; 840 m = m->m_next; 841 } 842 mrest_len = i; 843 844 /* 845 * Get the RPC header with authorization. 846 */ 847 kerbauth: 848 auth_str = (char *)0; 849 if (nmp->nm_flag & NFSMNT_KERB) { 850 if (failed_auth) { 851 error = nfs_getauth(nmp, rep, cred, &auth_type, 852 &auth_str, &auth_len); 853 if (error) { 854 free((caddr_t)rep, M_NFSREQ); 855 m_freem(mrest); 856 return (error); 857 } 858 } else { 859 auth_type = RPCAUTH_UNIX; 860 auth_len = 5 * NFSX_UNSIGNED; 861 } 862 } else { 863 auth_type = RPCAUTH_UNIX; 864 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 865 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 866 5 * NFSX_UNSIGNED; 867 } 868 m = nfsm_rpchead(cred, (nmp->nm_flag & NFSMNT_NQNFS), procnum, 869 auth_type, auth_len, auth_str, mrest, mrest_len, &mheadend, &xid); 870 if (auth_str) 871 free(auth_str, M_TEMP); 872 873 /* 874 * For stream protocols, insert a Sun RPC Record Mark. 875 */ 876 if (nmp->nm_sotype == SOCK_STREAM) { 877 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 878 *mtod(m, u_int32_t *) = htonl(0x80000000 | 879 (m->m_pkthdr.len - NFSX_UNSIGNED)); 880 } 881 rep->r_mreq = m; 882 rep->r_xid = xid; 883 tryagain: 884 if (nmp->nm_flag & NFSMNT_SOFT) 885 rep->r_retry = nmp->nm_retry; 886 else 887 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 888 rep->r_rtt = rep->r_rexmit = 0; 889 if (proct[procnum] > 0) 890 rep->r_flags = R_TIMING; 891 else 892 rep->r_flags = 0; 893 rep->r_mrep = NULL; 894 895 /* 896 * Do the client side RPC. 897 */ 898 nfsstats.rpcrequests++; 899 /* 900 * Chain request into list of outstanding requests. Be sure 901 * to put it LAST so timer finds oldest requests first. 902 */ 903 s = splsoftclock(); 904 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 905 906 /* Get send time for nqnfs */ 907 reqtime = time.tv_sec; 908 909 /* 910 * If backing off another request or avoiding congestion, don't 911 * send this one now but let timer do it. If not timing a request, 912 * do it now. 913 */ 914 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 915 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 916 nmp->nm_sent < nmp->nm_cwnd)) { 917 splx(s); 918 if (nmp->nm_soflags & PR_CONNREQUIRED) 919 error = nfs_sndlock(&nmp->nm_flag, rep); 920 if (!error) { 921 m = m_copym(m, 0, M_COPYALL, M_WAIT); 922 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 923 if (nmp->nm_soflags & PR_CONNREQUIRED) 924 nfs_sndunlock(&nmp->nm_flag); 925 } 926 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 927 nmp->nm_sent += NFS_CWNDSCALE; 928 rep->r_flags |= R_SENT; 929 } 930 } else { 931 splx(s); 932 rep->r_rtt = -1; 933 } 934 935 /* 936 * Wait for the reply from our send or the timer's. 937 */ 938 if (!error || error == EPIPE) 939 error = nfs_reply(rep); 940 941 /* 942 * RPC done, unlink the request. 943 */ 944 s = splsoftclock(); 945 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 946 splx(s); 947 948 /* 949 * Decrement the outstanding request count. 950 */ 951 if (rep->r_flags & R_SENT) { 952 rep->r_flags &= ~R_SENT; /* paranoia */ 953 nmp->nm_sent -= NFS_CWNDSCALE; 954 } 955 956 /* 957 * If there was a successful reply and a tprintf msg. 958 * tprintf a response. 959 */ 960 if (!error && (rep->r_flags & R_TPRINTFMSG)) 961 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 962 "is alive again"); 963 mrep = rep->r_mrep; 964 md = rep->r_md; 965 dpos = rep->r_dpos; 966 if (error) { 967 m_freem(rep->r_mreq); 968 free((caddr_t)rep, M_NFSREQ); 969 return (error); 970 } 971 972 /* 973 * break down the rpc header and check if ok 974 */ 975 nfsm_dissect(tl, u_int32_t *, 3*NFSX_UNSIGNED); 976 if (*tl++ == rpc_msgdenied) { 977 if (*tl == rpc_mismatch) 978 error = EOPNOTSUPP; 979 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 980 if (*tl == rpc_rejectedcred && failed_auth == 0) { 981 failed_auth++; 982 mheadend->m_next = (struct mbuf *)0; 983 m_freem(mrep); 984 m_freem(rep->r_mreq); 985 goto kerbauth; 986 } else 987 error = EAUTH; 988 } else 989 error = EACCES; 990 m_freem(mrep); 991 m_freem(rep->r_mreq); 992 free((caddr_t)rep, M_NFSREQ); 993 return (error); 994 } 995 996 /* 997 * skip over the auth_verf, someday we may want to cache auth_short's 998 * for nfs_reqhead(), but for now just dump it 999 */ 1000 if (*++tl != 0) { 1001 i = nfsm_rndup(fxdr_unsigned(int32_t, *tl)); 1002 nfsm_adv(i); 1003 } 1004 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1005 /* 0 == ok */ 1006 if (*tl == 0) { 1007 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1008 if (*tl != 0) { 1009 error = fxdr_unsigned(int, *tl); 1010 m_freem(mrep); 1011 if ((nmp->nm_flag & NFSMNT_NQNFS) && 1012 error == NQNFS_TRYLATER) { 1013 error = 0; 1014 waituntil = time.tv_sec + trylater_delay; 1015 while (time.tv_sec < waituntil) 1016 (void) tsleep((caddr_t)&lbolt, 1017 PSOCK, "nqnfstry", 0); 1018 trylater_delay *= nfs_backoff[trylater_cnt]; 1019 if (trylater_cnt < 7) 1020 trylater_cnt++; 1021 goto tryagain; 1022 } 1023 1024 /* 1025 * If the File Handle was stale, invalidate the 1026 * lookup cache, just in case. 1027 */ 1028 if (error == ESTALE) 1029 cache_purge(vp); 1030 m_freem(rep->r_mreq); 1031 free((caddr_t)rep, M_NFSREQ); 1032 return (error); 1033 } 1034 1035 /* 1036 * For nqnfs, get any lease in reply 1037 */ 1038 if (nmp->nm_flag & NFSMNT_NQNFS) { 1039 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1040 if (*tl) { 1041 np = VTONFS(vp); 1042 nqlflag = fxdr_unsigned(int, *tl); 1043 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1044 cachable = fxdr_unsigned(int, *tl++); 1045 reqtime += fxdr_unsigned(int, *tl++); 1046 if (reqtime > time.tv_sec) { 1047 fxdr_hyper(tl, &frev); 1048 nqnfs_clientlease(nmp, np, nqlflag, 1049 cachable, reqtime, frev); 1050 } 1051 } 1052 } 1053 *mrp = mrep; 1054 *mdp = md; 1055 *dposp = dpos; 1056 m_freem(rep->r_mreq); 1057 FREE((caddr_t)rep, M_NFSREQ); 1058 return (0); 1059 } 1060 m_freem(mrep); 1061 m_freem(rep->r_mreq); 1062 free((caddr_t)rep, M_NFSREQ); 1063 error = EPROTONOSUPPORT; 1064 nfsmout: 1065 return (error); 1066 } 1067 #endif /* NFSCLIENT */ 1068 1069 /* 1070 * Generate the rpc reply header 1071 * siz arg. is used to decide if adding a cluster is worthwhile 1072 */ 1073 int 1074 nfs_rephead(siz, nd, err, cache, frev, mrq, mbp, bposp) 1075 int siz; 1076 struct nfsd *nd; 1077 int err; 1078 int cache; 1079 u_quad_t *frev; 1080 struct mbuf **mrq; 1081 struct mbuf **mbp; 1082 caddr_t *bposp; 1083 { 1084 register u_int32_t *tl; 1085 register struct mbuf *mreq; 1086 caddr_t bpos; 1087 struct mbuf *mb, *mb2; 1088 1089 MGETHDR(mreq, M_WAIT, MT_DATA); 1090 mb = mreq; 1091 /* 1092 * If this is a big reply, use a cluster else 1093 * try and leave leading space for the lower level headers. 1094 */ 1095 siz += RPC_REPLYSIZ; 1096 if (siz >= MINCLSIZE) { 1097 MCLGET(mreq, M_WAIT); 1098 } else 1099 mreq->m_data += max_hdr; 1100 tl = mtod(mreq, u_int32_t *); 1101 mreq->m_len = 6*NFSX_UNSIGNED; 1102 bpos = ((caddr_t)tl)+mreq->m_len; 1103 *tl++ = txdr_unsigned(nd->nd_retxid); 1104 *tl++ = rpc_reply; 1105 if (err == ERPCMISMATCH || err == NQNFS_AUTHERR) { 1106 *tl++ = rpc_msgdenied; 1107 if (err == NQNFS_AUTHERR) { 1108 *tl++ = rpc_autherr; 1109 *tl = rpc_rejectedcred; 1110 mreq->m_len -= NFSX_UNSIGNED; 1111 bpos -= NFSX_UNSIGNED; 1112 } else { 1113 *tl++ = rpc_mismatch; 1114 *tl++ = txdr_unsigned(2); 1115 *tl = txdr_unsigned(2); 1116 } 1117 } else { 1118 *tl++ = rpc_msgaccepted; 1119 *tl++ = 0; 1120 *tl++ = 0; 1121 switch (err) { 1122 case EPROGUNAVAIL: 1123 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1124 break; 1125 case EPROGMISMATCH: 1126 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1127 nfsm_build(tl, u_int32_t *, 2*NFSX_UNSIGNED); 1128 *tl++ = txdr_unsigned(2); 1129 *tl = txdr_unsigned(2); /* someday 3 */ 1130 break; 1131 case EPROCUNAVAIL: 1132 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1133 break; 1134 default: 1135 *tl = 0; 1136 if (err != VNOVAL) { 1137 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1138 if (err) 1139 *tl = txdr_unsigned(nfsrv_errmap[err - 1]); 1140 else 1141 *tl = 0; 1142 } 1143 break; 1144 }; 1145 } 1146 1147 /* 1148 * For nqnfs, piggyback lease as requested. 1149 */ 1150 if (nd->nd_nqlflag != NQL_NOVAL && err == 0) { 1151 if (nd->nd_nqlflag) { 1152 nfsm_build(tl, u_int32_t *, 5*NFSX_UNSIGNED); 1153 *tl++ = txdr_unsigned(nd->nd_nqlflag); 1154 *tl++ = txdr_unsigned(cache); 1155 *tl++ = txdr_unsigned(nd->nd_duration); 1156 txdr_hyper(frev, tl); 1157 } else { 1158 if (nd->nd_nqlflag != 0) 1159 panic("nqreph"); 1160 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1161 *tl = 0; 1162 } 1163 } 1164 *mrq = mreq; 1165 *mbp = mb; 1166 *bposp = bpos; 1167 if (err != 0 && err != VNOVAL) 1168 nfsstats.srvrpc_errs++; 1169 return (0); 1170 } 1171 1172 /* 1173 * Nfs timer routine 1174 * Scan the nfsreq list and retranmit any requests that have timed out 1175 * To avoid retransmission attempts on STREAM sockets (in the future) make 1176 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1177 */ 1178 void 1179 nfs_timer(arg) 1180 void *arg; 1181 { 1182 register struct nfsreq *rep; 1183 register struct mbuf *m; 1184 register struct socket *so; 1185 register struct nfsmount *nmp; 1186 register int timeo; 1187 #ifdef NFSSERVER 1188 static long lasttime = 0; 1189 #endif 1190 int s, error; 1191 1192 s = splsoftnet(); 1193 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1194 nmp = rep->r_nmp; 1195 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1196 continue; 1197 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1198 rep->r_flags |= R_SOFTTERM; 1199 continue; 1200 } 1201 if (rep->r_rtt >= 0) { 1202 rep->r_rtt++; 1203 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1204 timeo = nmp->nm_timeo; 1205 else 1206 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1207 if (nmp->nm_timeouts > 0) 1208 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1209 if (rep->r_rtt <= timeo) 1210 continue; 1211 if (nmp->nm_timeouts < 8) 1212 nmp->nm_timeouts++; 1213 } 1214 /* 1215 * Check for server not responding 1216 */ 1217 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1218 rep->r_rexmit > nmp->nm_deadthresh) { 1219 nfs_msg(rep->r_procp, 1220 nmp->nm_mountp->mnt_stat.f_mntfromname, 1221 "not responding"); 1222 rep->r_flags |= R_TPRINTFMSG; 1223 } 1224 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1225 nfsstats.rpctimeouts++; 1226 rep->r_flags |= R_SOFTTERM; 1227 continue; 1228 } 1229 if (nmp->nm_sotype != SOCK_DGRAM) { 1230 if (++rep->r_rexmit > NFS_MAXREXMIT) 1231 rep->r_rexmit = NFS_MAXREXMIT; 1232 continue; 1233 } 1234 if ((so = nmp->nm_so) == NULL) 1235 continue; 1236 1237 /* 1238 * If there is enough space and the window allows.. 1239 * Resend it 1240 * Set r_rtt to -1 in case we fail to send it now. 1241 */ 1242 rep->r_rtt = -1; 1243 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1244 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1245 (rep->r_flags & R_SENT) || 1246 nmp->nm_sent < nmp->nm_cwnd) && 1247 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1248 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1249 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1250 (struct mbuf *)0, (struct mbuf *)0); 1251 else 1252 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1253 nmp->nm_nam, (struct mbuf *)0); 1254 if (error) { 1255 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) 1256 so->so_error = 0; 1257 } else { 1258 /* 1259 * Iff first send, start timing 1260 * else turn timing off, backoff timer 1261 * and divide congestion window by 2. 1262 */ 1263 if (rep->r_flags & R_SENT) { 1264 rep->r_flags &= ~R_TIMING; 1265 if (++rep->r_rexmit > NFS_MAXREXMIT) 1266 rep->r_rexmit = NFS_MAXREXMIT; 1267 nmp->nm_cwnd >>= 1; 1268 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1269 nmp->nm_cwnd = NFS_CWNDSCALE; 1270 nfsstats.rpcretries++; 1271 } else { 1272 rep->r_flags |= R_SENT; 1273 nmp->nm_sent += NFS_CWNDSCALE; 1274 } 1275 rep->r_rtt = 0; 1276 } 1277 } 1278 } 1279 1280 #ifdef NFSSERVER 1281 /* 1282 * Call the nqnfs server timer once a second to handle leases. 1283 */ 1284 if (lasttime != time.tv_sec) { 1285 lasttime = time.tv_sec; 1286 nqnfs_serverd(); 1287 } 1288 #endif /* NFSSERVER */ 1289 splx(s); 1290 timeout(nfs_timer, (void *)0, hz / NFS_HZ); 1291 } 1292 1293 /* 1294 * Test for a termination condition pending on the process. 1295 * This is used for NFSMNT_INT mounts. 1296 */ 1297 int 1298 nfs_sigintr(nmp, rep, p) 1299 struct nfsmount *nmp; 1300 struct nfsreq *rep; 1301 register struct proc *p; 1302 { 1303 1304 if (rep && (rep->r_flags & R_SOFTTERM)) 1305 return (EINTR); 1306 if (!(nmp->nm_flag & NFSMNT_INT)) 1307 return (0); 1308 if (p && p->p_siglist && 1309 (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) & 1310 NFSINT_SIGMASK)) 1311 return (EINTR); 1312 return (0); 1313 } 1314 1315 /* 1316 * Lock a socket against others. 1317 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1318 * and also to avoid race conditions between the processes with nfs requests 1319 * in progress when a reconnect is necessary. 1320 */ 1321 int 1322 nfs_sndlock(flagp, rep) 1323 register int *flagp; 1324 struct nfsreq *rep; 1325 { 1326 struct proc *p; 1327 int slpflag = 0, slptimeo = 0; 1328 1329 if (rep) { 1330 p = rep->r_procp; 1331 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1332 slpflag = PCATCH; 1333 } else 1334 p = (struct proc *)0; 1335 while (*flagp & NFSMNT_SNDLOCK) { 1336 if (nfs_sigintr(rep->r_nmp, rep, p)) 1337 return (EINTR); 1338 *flagp |= NFSMNT_WANTSND; 1339 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1340 slptimeo); 1341 if (slpflag == PCATCH) { 1342 slpflag = 0; 1343 slptimeo = 2 * hz; 1344 } 1345 } 1346 *flagp |= NFSMNT_SNDLOCK; 1347 return (0); 1348 } 1349 1350 /* 1351 * Unlock the stream socket for others. 1352 */ 1353 void 1354 nfs_sndunlock(flagp) 1355 register int *flagp; 1356 { 1357 1358 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1359 panic("nfs sndunlock"); 1360 *flagp &= ~NFSMNT_SNDLOCK; 1361 if (*flagp & NFSMNT_WANTSND) { 1362 *flagp &= ~NFSMNT_WANTSND; 1363 wakeup((caddr_t)flagp); 1364 } 1365 } 1366 1367 int 1368 nfs_rcvlock(rep) 1369 register struct nfsreq *rep; 1370 { 1371 register int *flagp = &rep->r_nmp->nm_flag; 1372 int slpflag, slptimeo = 0; 1373 1374 if (*flagp & NFSMNT_INT) 1375 slpflag = PCATCH; 1376 else 1377 slpflag = 0; 1378 while (*flagp & NFSMNT_RCVLOCK) { 1379 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1380 return (EINTR); 1381 *flagp |= NFSMNT_WANTRCV; 1382 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1383 slptimeo); 1384 if (slpflag == PCATCH) { 1385 slpflag = 0; 1386 slptimeo = 2 * hz; 1387 } 1388 } 1389 *flagp |= NFSMNT_RCVLOCK; 1390 return (0); 1391 } 1392 1393 /* 1394 * Unlock the stream socket for others. 1395 */ 1396 void 1397 nfs_rcvunlock(flagp) 1398 register int *flagp; 1399 { 1400 1401 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1402 panic("nfs rcvunlock"); 1403 *flagp &= ~NFSMNT_RCVLOCK; 1404 if (*flagp & NFSMNT_WANTRCV) { 1405 *flagp &= ~NFSMNT_WANTRCV; 1406 wakeup((caddr_t)flagp); 1407 } 1408 } 1409 1410 /* 1411 * Check for badly aligned mbuf data areas and 1412 * realign data in an mbuf list by copying the data areas up, as required. 1413 */ 1414 void 1415 nfs_realign(m, hsiz) 1416 register struct mbuf *m; 1417 int hsiz; 1418 { 1419 register struct mbuf *m2; 1420 register int siz, mlen, olen; 1421 register caddr_t tcp, fcp; 1422 struct mbuf *mnew; 1423 1424 while (m) { 1425 /* 1426 * This never happens for UDP, rarely happens for TCP 1427 * but frequently happens for iso transport. 1428 */ 1429 if ((m->m_len & 0x3) || (mtod(m, long) & 0x3)) { 1430 olen = m->m_len; 1431 fcp = mtod(m, caddr_t); 1432 if ((long)fcp & 0x3) { 1433 m->m_flags &= ~M_PKTHDR; 1434 if (m->m_flags & M_EXT) 1435 m->m_data = m->m_ext.ext_buf + 1436 ((m->m_ext.ext_size - olen) & ~0x3); 1437 else 1438 m->m_data = m->m_dat; 1439 } 1440 m->m_len = 0; 1441 tcp = mtod(m, caddr_t); 1442 mnew = m; 1443 m2 = m->m_next; 1444 1445 /* 1446 * If possible, only put the first invariant part 1447 * of the RPC header in the first mbuf. 1448 */ 1449 mlen = M_TRAILINGSPACE(m); 1450 if (olen <= hsiz && mlen > hsiz) 1451 mlen = hsiz; 1452 1453 /* 1454 * Loop through the mbuf list consolidating data. 1455 */ 1456 while (m) { 1457 while (olen > 0) { 1458 if (mlen == 0) { 1459 m2->m_flags &= ~M_PKTHDR; 1460 if (m2->m_flags & M_EXT) 1461 m2->m_data = m2->m_ext.ext_buf; 1462 else 1463 m2->m_data = m2->m_dat; 1464 m2->m_len = 0; 1465 mlen = M_TRAILINGSPACE(m2); 1466 tcp = mtod(m2, caddr_t); 1467 mnew = m2; 1468 m2 = m2->m_next; 1469 } 1470 siz = min(mlen, olen); 1471 if (tcp != fcp) 1472 bcopy(fcp, tcp, siz); 1473 mnew->m_len += siz; 1474 mlen -= siz; 1475 olen -= siz; 1476 tcp += siz; 1477 fcp += siz; 1478 } 1479 m = m->m_next; 1480 if (m) { 1481 olen = m->m_len; 1482 fcp = mtod(m, caddr_t); 1483 } 1484 } 1485 1486 /* 1487 * Finally, set m_len == 0 for any trailing mbufs that have 1488 * been copied out of. 1489 */ 1490 while (m2) { 1491 m2->m_len = 0; 1492 m2 = m2->m_next; 1493 } 1494 return; 1495 } 1496 m = m->m_next; 1497 } 1498 } 1499 1500 /* 1501 * Parse an RPC request 1502 * - verify it 1503 * - fill in the cred struct. 1504 */ 1505 int 1506 nfs_getreq(nd, has_header) 1507 register struct nfsd *nd; 1508 int has_header; 1509 { 1510 register int len, i; 1511 register u_int32_t *tl; 1512 register int32_t t1; 1513 struct uio uio; 1514 struct iovec iov; 1515 caddr_t dpos, cp2; 1516 u_int32_t nfsvers, auth_type; 1517 int error = 0, nqnfs = 0; 1518 struct mbuf *mrep, *md; 1519 1520 mrep = nd->nd_mrep; 1521 md = nd->nd_md; 1522 dpos = nd->nd_dpos; 1523 if (has_header) { 1524 nfsm_dissect(tl, u_int32_t *, 10*NFSX_UNSIGNED); 1525 nd->nd_retxid = fxdr_unsigned(u_int32_t , *tl++); 1526 if (*tl++ != rpc_call) { 1527 m_freem(mrep); 1528 return (EBADRPC); 1529 } 1530 } else { 1531 nfsm_dissect(tl, u_int32_t *, 8*NFSX_UNSIGNED); 1532 } 1533 nd->nd_repstat = 0; 1534 if (*tl++ != rpc_vers) { 1535 nd->nd_repstat = ERPCMISMATCH; 1536 nd->nd_procnum = NFSPROC_NOOP; 1537 return (0); 1538 } 1539 nfsvers = nfs_vers; 1540 if (*tl != nfs_prog) { 1541 if (*tl == nqnfs_prog) { 1542 nqnfs++; 1543 nfsvers = nqnfs_vers; 1544 } else { 1545 nd->nd_repstat = EPROGUNAVAIL; 1546 nd->nd_procnum = NFSPROC_NOOP; 1547 return (0); 1548 } 1549 } 1550 tl++; 1551 if (*tl++ != nfsvers) { 1552 nd->nd_repstat = EPROGMISMATCH; 1553 nd->nd_procnum = NFSPROC_NOOP; 1554 return (0); 1555 } 1556 nd->nd_procnum = fxdr_unsigned(u_int32_t , *tl++); 1557 if (nd->nd_procnum == NFSPROC_NULL) 1558 return (0); 1559 if (nd->nd_procnum >= NFS_NPROCS || 1560 (!nqnfs && nd->nd_procnum > NFSPROC_STATFS) || 1561 (*tl != rpc_auth_unix && *tl != rpc_auth_kerb)) { 1562 nd->nd_repstat = EPROCUNAVAIL; 1563 nd->nd_procnum = NFSPROC_NOOP; 1564 return (0); 1565 } 1566 auth_type = *tl++; 1567 len = fxdr_unsigned(int, *tl++); 1568 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1569 m_freem(mrep); 1570 return (EBADRPC); 1571 } 1572 1573 /* 1574 * Handle auth_unix or auth_kerb. 1575 */ 1576 if (auth_type == rpc_auth_unix) { 1577 len = fxdr_unsigned(int, *++tl); 1578 if (len < 0 || len > NFS_MAXNAMLEN) { 1579 m_freem(mrep); 1580 return (EBADRPC); 1581 } 1582 nfsm_adv(nfsm_rndup(len)); 1583 nfsm_dissect(tl, u_int32_t *, 3*NFSX_UNSIGNED); 1584 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1585 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1586 len = fxdr_unsigned(int, *tl); 1587 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1588 m_freem(mrep); 1589 return (EBADRPC); 1590 } 1591 nfsm_dissect(tl, u_int32_t *, (len + 2)*NFSX_UNSIGNED); 1592 for (i = 0; i < len; i++) 1593 if (i < NGROUPS) 1594 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1595 else 1596 tl++; 1597 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1598 } else if (auth_type == rpc_auth_kerb) { 1599 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1600 nd->nd_authlen = fxdr_unsigned(int, *tl); 1601 uio.uio_resid = nfsm_rndup(nd->nd_authlen); 1602 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1603 m_freem(mrep); 1604 return (EBADRPC); 1605 } 1606 uio.uio_offset = 0; 1607 uio.uio_iov = &iov; 1608 uio.uio_iovcnt = 1; 1609 uio.uio_segflg = UIO_SYSSPACE; 1610 iov.iov_base = (caddr_t)nd->nd_authstr; 1611 iov.iov_len = RPCAUTH_MAXSIZ; 1612 nfsm_mtouio(&uio, uio.uio_resid); 1613 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1614 nd->nd_flag |= NFSD_NEEDAUTH; 1615 } 1616 1617 /* 1618 * Do we have any use for the verifier. 1619 * According to the "Remote Procedure Call Protocol Spec." it 1620 * should be AUTH_NULL, but some clients make it AUTH_UNIX? 1621 * For now, just skip over it 1622 */ 1623 len = fxdr_unsigned(int, *++tl); 1624 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1625 m_freem(mrep); 1626 return (EBADRPC); 1627 } 1628 if (len > 0) { 1629 nfsm_adv(nfsm_rndup(len)); 1630 } 1631 1632 /* 1633 * For nqnfs, get piggybacked lease request. 1634 */ 1635 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1636 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1637 nd->nd_nqlflag = fxdr_unsigned(int, *tl); 1638 if (nd->nd_nqlflag) { 1639 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1640 nd->nd_duration = fxdr_unsigned(int, *tl); 1641 } else 1642 nd->nd_duration = NQ_MINLEASE; 1643 } else { 1644 nd->nd_nqlflag = NQL_NOVAL; 1645 nd->nd_duration = NQ_MINLEASE; 1646 } 1647 nd->nd_md = md; 1648 nd->nd_dpos = dpos; 1649 return (0); 1650 nfsmout: 1651 return (error); 1652 } 1653 1654 void 1655 nfs_msg(p, server, msg) 1656 struct proc *p; 1657 char *server, *msg; 1658 { 1659 tpr_t tpr; 1660 1661 if (p) 1662 tpr = tprintf_open(p); 1663 else 1664 tpr = NULL; 1665 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1666 tprintf_close(tpr); 1667 } 1668 1669 #ifdef NFSSERVER 1670 int (*nfsrv_procs[NFS_NPROCS]) __P((struct nfsd *, struct mbuf *, struct mbuf *, 1671 caddr_t, struct ucred *, struct mbuf *, 1672 struct mbuf **)) = { 1673 nfsrv_null, 1674 nfsrv_getattr, 1675 nfsrv_setattr, 1676 nfsrv_noop, 1677 nfsrv_lookup, 1678 nfsrv_readlink, 1679 nfsrv_read, 1680 nfsrv_noop, 1681 nfsrv_write, 1682 nfsrv_create, 1683 nfsrv_remove, 1684 nfsrv_rename, 1685 nfsrv_link, 1686 nfsrv_symlink, 1687 nfsrv_mkdir, 1688 nfsrv_rmdir, 1689 nfsrv_readdir, 1690 nfsrv_statfs, 1691 nqnfsrv_readdirlook, 1692 nqnfsrv_getlease, 1693 nqnfsrv_vacated, 1694 nfsrv_noop, 1695 nqnfsrv_access, 1696 }; 1697 1698 /* 1699 * Socket upcall routine for the nfsd sockets. 1700 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1701 * Essentially do as much as possible non-blocking, else punt and it will 1702 * be called with M_WAIT from an nfsd. 1703 */ 1704 void 1705 nfsrv_rcv(so, arg, waitflag) 1706 struct socket *so; 1707 caddr_t arg; 1708 int waitflag; 1709 { 1710 register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1711 register struct mbuf *m; 1712 struct mbuf *mp, *nam; 1713 struct uio auio; 1714 int flags, error; 1715 1716 if ((slp->ns_flag & SLP_VALID) == 0) 1717 return; 1718 #ifdef notdef 1719 /* 1720 * Define this to test for nfsds handling this under heavy load. 1721 */ 1722 if (waitflag == M_DONTWAIT) { 1723 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1724 } 1725 #endif 1726 auio.uio_procp = NULL; 1727 if (so->so_type == SOCK_STREAM) { 1728 /* 1729 * If there are already records on the queue, defer soreceive() 1730 * to an nfsd so that there is feedback to the TCP layer that 1731 * the nfs servers are heavily loaded. 1732 */ 1733 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1734 slp->ns_flag |= SLP_NEEDQ; 1735 goto dorecs; 1736 } 1737 1738 /* 1739 * Do soreceive(). 1740 */ 1741 auio.uio_resid = 1000000000; 1742 flags = MSG_DONTWAIT; 1743 error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1744 if (error || mp == (struct mbuf *)0) { 1745 if (error == EWOULDBLOCK) 1746 slp->ns_flag |= SLP_NEEDQ; 1747 else 1748 slp->ns_flag |= SLP_DISCONN; 1749 goto dorecs; 1750 } 1751 m = mp; 1752 if (slp->ns_rawend) { 1753 slp->ns_rawend->m_next = m; 1754 slp->ns_cc += 1000000000 - auio.uio_resid; 1755 } else { 1756 slp->ns_raw = m; 1757 slp->ns_cc = 1000000000 - auio.uio_resid; 1758 } 1759 while (m->m_next) 1760 m = m->m_next; 1761 slp->ns_rawend = m; 1762 1763 /* 1764 * Now try and parse record(s) out of the raw stream data. 1765 */ 1766 if ((error = nfsrv_getstream(slp, waitflag)) != 0) { 1767 if (error == EPERM) 1768 slp->ns_flag |= SLP_DISCONN; 1769 else 1770 slp->ns_flag |= SLP_NEEDQ; 1771 } 1772 } else { 1773 do { 1774 auio.uio_resid = 1000000000; 1775 flags = MSG_DONTWAIT; 1776 error = soreceive(so, &nam, &auio, &mp, 1777 (struct mbuf **)0, &flags); 1778 if (mp) { 1779 nfs_realign(mp, 10 * NFSX_UNSIGNED); 1780 if (nam) { 1781 m = nam; 1782 m->m_next = mp; 1783 } else 1784 m = mp; 1785 if (slp->ns_recend) 1786 slp->ns_recend->m_nextpkt = m; 1787 else 1788 slp->ns_rec = m; 1789 slp->ns_recend = m; 1790 m->m_nextpkt = (struct mbuf *)0; 1791 } 1792 if (error) { 1793 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1794 && error != EWOULDBLOCK) { 1795 slp->ns_flag |= SLP_DISCONN; 1796 goto dorecs; 1797 } 1798 } 1799 } while (mp); 1800 } 1801 1802 /* 1803 * Now try and process the request records, non-blocking. 1804 */ 1805 dorecs: 1806 if (waitflag == M_DONTWAIT && 1807 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1808 nfsrv_wakenfsd(slp); 1809 } 1810 1811 /* 1812 * Try and extract an RPC request from the mbuf data list received on a 1813 * stream socket. The "waitflag" argument indicates whether or not it 1814 * can sleep. 1815 */ 1816 int 1817 nfsrv_getstream(slp, waitflag) 1818 register struct nfssvc_sock *slp; 1819 int waitflag; 1820 { 1821 register struct mbuf *m; 1822 register char *cp1, *cp2; 1823 register int len; 1824 struct mbuf *om, *m2, *recm = NULL; 1825 u_long recmark; 1826 1827 if (slp->ns_flag & SLP_GETSTREAM) 1828 panic("nfs getstream"); 1829 slp->ns_flag |= SLP_GETSTREAM; 1830 for (;;) { 1831 if (slp->ns_reclen == 0) { 1832 if (slp->ns_cc < NFSX_UNSIGNED) { 1833 slp->ns_flag &= ~SLP_GETSTREAM; 1834 return (0); 1835 } 1836 m = slp->ns_raw; 1837 if (m->m_len >= NFSX_UNSIGNED) { 1838 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 1839 m->m_data += NFSX_UNSIGNED; 1840 m->m_len -= NFSX_UNSIGNED; 1841 } else { 1842 cp1 = (caddr_t)&recmark; 1843 cp2 = mtod(m, caddr_t); 1844 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1845 while (m->m_len == 0) { 1846 m = m->m_next; 1847 cp2 = mtod(m, caddr_t); 1848 } 1849 *cp1++ = *cp2++; 1850 m->m_data++; 1851 m->m_len--; 1852 } 1853 } 1854 slp->ns_cc -= NFSX_UNSIGNED; 1855 slp->ns_reclen = ntohl(recmark) & ~0x80000000; 1856 if (slp->ns_reclen < NFS_MINPACKET || slp->ns_reclen > NFS_MAXPACKET) { 1857 slp->ns_flag &= ~SLP_GETSTREAM; 1858 return (EPERM); 1859 } 1860 } 1861 1862 /* 1863 * Now get the record part. 1864 */ 1865 if (slp->ns_cc == slp->ns_reclen) { 1866 recm = slp->ns_raw; 1867 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 1868 slp->ns_cc = slp->ns_reclen = 0; 1869 } else if (slp->ns_cc > slp->ns_reclen) { 1870 len = 0; 1871 m = slp->ns_raw; 1872 om = (struct mbuf *)0; 1873 while (len < slp->ns_reclen) { 1874 if ((len + m->m_len) > slp->ns_reclen) { 1875 m2 = m_copym(m, 0, slp->ns_reclen - len, 1876 waitflag); 1877 if (m2) { 1878 if (om) { 1879 om->m_next = m2; 1880 recm = slp->ns_raw; 1881 } else 1882 recm = m2; 1883 m->m_data += slp->ns_reclen - len; 1884 m->m_len -= slp->ns_reclen - len; 1885 len = slp->ns_reclen; 1886 } else { 1887 slp->ns_flag &= ~SLP_GETSTREAM; 1888 return (EWOULDBLOCK); 1889 } 1890 } else if ((len + m->m_len) == slp->ns_reclen) { 1891 om = m; 1892 len += m->m_len; 1893 m = m->m_next; 1894 recm = slp->ns_raw; 1895 om->m_next = (struct mbuf *)0; 1896 } else { 1897 om = m; 1898 len += m->m_len; 1899 m = m->m_next; 1900 } 1901 } 1902 slp->ns_raw = m; 1903 slp->ns_cc -= len; 1904 slp->ns_reclen = 0; 1905 } else { 1906 slp->ns_flag &= ~SLP_GETSTREAM; 1907 return (0); 1908 } 1909 nfs_realign(recm, 10 * NFSX_UNSIGNED); 1910 if (slp->ns_recend) 1911 slp->ns_recend->m_nextpkt = recm; 1912 else 1913 slp->ns_rec = recm; 1914 slp->ns_recend = recm; 1915 } 1916 } 1917 1918 /* 1919 * Parse an RPC header. 1920 */ 1921 int 1922 nfsrv_dorec(slp, nd) 1923 register struct nfssvc_sock *slp; 1924 register struct nfsd *nd; 1925 { 1926 register struct mbuf *m; 1927 int error; 1928 1929 if ((slp->ns_flag & SLP_VALID) == 0 || 1930 (m = slp->ns_rec) == (struct mbuf *)0) 1931 return (ENOBUFS); 1932 if ((slp->ns_rec = m->m_nextpkt) != NULL) 1933 m->m_nextpkt = (struct mbuf *)0; 1934 else 1935 slp->ns_recend = (struct mbuf *)0; 1936 if (m->m_type == MT_SONAME) { 1937 nd->nd_nam = m; 1938 nd->nd_md = nd->nd_mrep = m->m_next; 1939 m->m_next = (struct mbuf *)0; 1940 } else { 1941 nd->nd_nam = (struct mbuf *)0; 1942 nd->nd_md = nd->nd_mrep = m; 1943 } 1944 nd->nd_dpos = mtod(nd->nd_md, caddr_t); 1945 if ((error = nfs_getreq(nd, TRUE)) != 0) { 1946 m_freem(nd->nd_nam); 1947 return (error); 1948 } 1949 return (0); 1950 } 1951 1952 /* 1953 * Search for a sleeping nfsd and wake it up. 1954 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 1955 * running nfsds will go look for the work in the nfssvc_sock list. 1956 */ 1957 void 1958 nfsrv_wakenfsd(slp) 1959 struct nfssvc_sock *slp; 1960 { 1961 register struct nfsd *nd; 1962 1963 if ((slp->ns_flag & SLP_VALID) == 0) 1964 return; 1965 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nd_chain.tqe_next) { 1966 if (nd->nd_flag & NFSD_WAITING) { 1967 nd->nd_flag &= ~NFSD_WAITING; 1968 if (nd->nd_slp) 1969 panic("nfsd wakeup"); 1970 slp->ns_sref++; 1971 nd->nd_slp = slp; 1972 wakeup((caddr_t)nd); 1973 return; 1974 } 1975 } 1976 slp->ns_flag |= SLP_DOREC; 1977 nfsd_head_flag |= NFSD_CHECKSLP; 1978 } 1979 #endif /* NFSSERVER */ 1980
Indexes created Sun Sep 21 20:09:37 GMT 2025