nfs_socket.c revision 1.16
1 /* $NetBSD: nfs_socket.c,v 1.16 1994/08/17 11:41:42 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 struct nfsd nfsd_head; 163 164 /* 165 * Initialize sockets and congestion for a new NFS connection. 166 * We do not free the sockaddr if error. 167 */ 168 nfs_connect(nmp, rep) 169 register struct nfsmount *nmp; 170 struct nfsreq *rep; 171 { 172 register struct socket *so; 173 int s, error, rcvreserve, sndreserve; 174 struct sockaddr *saddr; 175 struct sockaddr_in *sin; 176 struct mbuf *m; 177 u_short tport; 178 179 nmp->nm_so = (struct socket *)0; 180 saddr = mtod(nmp->nm_nam, struct sockaddr *); 181 if (error = socreate(saddr->sa_family, 182 &nmp->nm_so, nmp->nm_sotype, nmp->nm_soproto)) 183 goto bad; 184 so = nmp->nm_so; 185 nmp->nm_soflags = so->so_proto->pr_flags; 186 187 /* 188 * Some servers require that the client port be a reserved port number. 189 */ 190 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 191 MGET(m, M_WAIT, MT_SONAME); 192 sin = mtod(m, struct sockaddr_in *); 193 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 194 sin->sin_family = AF_INET; 195 sin->sin_addr.s_addr = INADDR_ANY; 196 tport = IPPORT_RESERVED - 1; 197 sin->sin_port = htons(tport); 198 while ((error = sobind(so, m)) == EADDRINUSE && 199 --tport > IPPORT_RESERVED / 2) 200 sin->sin_port = htons(tport); 201 m_freem(m); 202 if (error) 203 goto bad; 204 } 205 206 /* 207 * Protocols that do not require connections may be optionally left 208 * unconnected for servers that reply from a port other than NFS_PORT. 209 */ 210 if (nmp->nm_flag & NFSMNT_NOCONN) { 211 if (nmp->nm_soflags & PR_CONNREQUIRED) { 212 error = ENOTCONN; 213 goto bad; 214 } 215 } else { 216 if (error = soconnect(so, nmp->nm_nam)) 217 goto bad; 218 219 /* 220 * Wait for the connection to complete. Cribbed from the 221 * connect system call but with the wait timing out so 222 * that interruptible mounts don't hang here for a long time. 223 */ 224 s = splnet(); 225 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 226 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 227 "nfscon", 2 * hz); 228 if ((so->so_state & SS_ISCONNECTING) && 229 so->so_error == 0 && rep && 230 (error = nfs_sigintr(nmp, rep, rep->r_procp))) { 231 so->so_state &= ~SS_ISCONNECTING; 232 splx(s); 233 goto bad; 234 } 235 } 236 if (so->so_error) { 237 error = so->so_error; 238 so->so_error = 0; 239 splx(s); 240 goto bad; 241 } 242 splx(s); 243 } 244 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 245 so->so_rcv.sb_timeo = (5 * hz); 246 so->so_snd.sb_timeo = (5 * hz); 247 } else { 248 so->so_rcv.sb_timeo = 0; 249 so->so_snd.sb_timeo = 0; 250 } 251 if (nmp->nm_sotype == SOCK_DGRAM) { 252 sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR; 253 rcvreserve = nmp->nm_rsize + NFS_MAXPKTHDR; 254 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 255 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 256 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR) * 2; 257 } else { 258 if (nmp->nm_sotype != SOCK_STREAM) 259 panic("nfscon sotype"); 260 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 261 MGET(m, M_WAIT, MT_SOOPTS); 262 *mtod(m, int *) = 1; 263 m->m_len = sizeof(int); 264 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 265 } 266 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 267 MGET(m, M_WAIT, MT_SOOPTS); 268 *mtod(m, int *) = 1; 269 m->m_len = sizeof(int); 270 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 271 } 272 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + sizeof (u_long)) 273 * 2; 274 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + sizeof (u_long)) 275 * 2; 276 } 277 if (error = soreserve(so, sndreserve, rcvreserve)) 278 goto bad; 279 so->so_rcv.sb_flags |= SB_NOINTR; 280 so->so_snd.sb_flags |= SB_NOINTR; 281 282 /* Initialize other non-zero congestion variables */ 283 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 284 nmp->nm_srtt[4] = (NFS_TIMEO << 3); 285 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 286 nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0; 287 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 288 nmp->nm_sent = 0; 289 nmp->nm_timeouts = 0; 290 return (0); 291 292 bad: 293 nfs_disconnect(nmp); 294 return (error); 295 } 296 297 /* 298 * Reconnect routine: 299 * Called when a connection is broken on a reliable protocol. 300 * - clean up the old socket 301 * - nfs_connect() again 302 * - set R_MUSTRESEND for all outstanding requests on mount point 303 * If this fails the mount point is DEAD! 304 * nb: Must be called with the nfs_sndlock() set on the mount point. 305 */ 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)) { 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 nfs_send(so, nam, top, rep) 362 register struct socket *so; 363 struct mbuf *nam; 364 register struct mbuf *top; 365 struct nfsreq *rep; 366 { 367 struct mbuf *sendnam; 368 int error, soflags, flags; 369 370 if (rep) { 371 if (rep->r_flags & R_SOFTTERM) { 372 m_freem(top); 373 return (EINTR); 374 } 375 if ((so = rep->r_nmp->nm_so) == NULL) { 376 rep->r_flags |= R_MUSTRESEND; 377 m_freem(top); 378 return (0); 379 } 380 rep->r_flags &= ~R_MUSTRESEND; 381 soflags = rep->r_nmp->nm_soflags; 382 } else 383 soflags = so->so_proto->pr_flags; 384 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 385 sendnam = (struct mbuf *)0; 386 else 387 sendnam = nam; 388 if (so->so_type == SOCK_SEQPACKET) 389 flags = MSG_EOR; 390 else 391 flags = 0; 392 393 error = sosend(so, sendnam, (struct uio *)0, top, 394 (struct mbuf *)0, flags); 395 if (error) { 396 if (rep) { 397 log(LOG_INFO, "nfs send error %d for server %s\n",error, 398 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 399 /* 400 * Deal with errors for the client side. 401 */ 402 if (rep->r_flags & R_SOFTTERM) 403 error = EINTR; 404 else 405 rep->r_flags |= R_MUSTRESEND; 406 } else 407 log(LOG_INFO, "nfsd send error %d\n", error); 408 409 /* 410 * Handle any recoverable (soft) socket errors here. (???) 411 */ 412 if (error != EINTR && error != ERESTART && 413 error != EWOULDBLOCK && error != EPIPE) 414 error = 0; 415 } 416 return (error); 417 } 418 419 #ifdef NFSCLIENT 420 /* 421 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 422 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 423 * Mark and consolidate the data into a new mbuf list. 424 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 425 * small mbufs. 426 * For SOCK_STREAM we must be very careful to read an entire record once 427 * we have read any of it, even if the system call has been interrupted. 428 */ 429 nfs_receive(rep, aname, mp) 430 register struct nfsreq *rep; 431 struct mbuf **aname; 432 struct mbuf **mp; 433 { 434 register struct socket *so; 435 struct uio auio; 436 struct iovec aio; 437 register struct mbuf *m; 438 struct mbuf *control; 439 u_long len; 440 struct mbuf **getnam; 441 int error, sotype, rcvflg; 442 struct proc *p = curproc; /* XXX */ 443 444 /* 445 * Set up arguments for soreceive() 446 */ 447 *mp = (struct mbuf *)0; 448 *aname = (struct mbuf *)0; 449 sotype = rep->r_nmp->nm_sotype; 450 451 /* 452 * For reliable protocols, lock against other senders/receivers 453 * in case a reconnect is necessary. 454 * For SOCK_STREAM, first get the Record Mark to find out how much 455 * more there is to get. 456 * We must lock the socket against other receivers 457 * until we have an entire rpc request/reply. 458 */ 459 if (sotype != SOCK_DGRAM) { 460 if (error = nfs_sndlock(&rep->r_nmp->nm_flag, rep)) 461 return (error); 462 tryagain: 463 /* 464 * Check for fatal errors and resending request. 465 */ 466 /* 467 * Ugh: If a reconnect attempt just happened, nm_so 468 * would have changed. NULL indicates a failed 469 * attempt that has essentially shut down this 470 * mount point. 471 */ 472 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 473 nfs_sndunlock(&rep->r_nmp->nm_flag); 474 return (EINTR); 475 } 476 if ((so = rep->r_nmp->nm_so) == NULL) { 477 if (error = nfs_reconnect(rep)) { 478 nfs_sndunlock(&rep->r_nmp->nm_flag); 479 return (error); 480 } 481 goto tryagain; 482 } 483 while (rep->r_flags & R_MUSTRESEND) { 484 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 485 nfsstats.rpcretries++; 486 if (error = nfs_send(so, rep->r_nmp->nm_nam, m, rep)) { 487 if (error == EINTR || error == ERESTART || 488 (error = nfs_reconnect(rep))) { 489 nfs_sndunlock(&rep->r_nmp->nm_flag); 490 return (error); 491 } 492 goto tryagain; 493 } 494 } 495 nfs_sndunlock(&rep->r_nmp->nm_flag); 496 if (sotype == SOCK_STREAM) { 497 aio.iov_base = (caddr_t) &len; 498 aio.iov_len = sizeof(u_long); 499 auio.uio_iov = &aio; 500 auio.uio_iovcnt = 1; 501 auio.uio_segflg = UIO_SYSSPACE; 502 auio.uio_rw = UIO_READ; 503 auio.uio_offset = 0; 504 auio.uio_resid = sizeof(u_long); 505 auio.uio_procp = p; 506 do { 507 rcvflg = MSG_WAITALL; 508 error = soreceive(so, (struct mbuf **)0, &auio, 509 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 510 if (error == EWOULDBLOCK && rep) { 511 if (rep->r_flags & R_SOFTTERM) 512 return (EINTR); 513 } 514 } while (error == EWOULDBLOCK); 515 if (!error && auio.uio_resid > 0) { 516 log(LOG_INFO, 517 "short receive (%d/%d) from nfs server %s\n", 518 sizeof(u_long) - auio.uio_resid, 519 sizeof(u_long), 520 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 521 error = EPIPE; 522 } 523 if (error) 524 goto errout; 525 len = ntohl(len) & ~0x80000000; 526 /* 527 * This is SERIOUS! We are out of sync with the sender 528 * and forcing a disconnect/reconnect is all I can do. 529 */ 530 if (len > NFS_MAXPACKET) { 531 log(LOG_ERR, "%s (%d) from nfs server %s\n", 532 "impossible packet length", 533 len, 534 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 535 error = EFBIG; 536 goto errout; 537 } 538 auio.uio_resid = len; 539 do { 540 rcvflg = MSG_WAITALL; 541 error = soreceive(so, (struct mbuf **)0, 542 &auio, mp, (struct mbuf **)0, &rcvflg); 543 } while (error == EWOULDBLOCK || error == EINTR || 544 error == ERESTART); 545 if (!error && auio.uio_resid > 0) { 546 log(LOG_INFO, 547 "short receive (%d/%d) from nfs server %s\n", 548 len - auio.uio_resid, len, 549 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 550 error = EPIPE; 551 } 552 } else { 553 /* 554 * NB: Since uio_resid is big, MSG_WAITALL is ignored 555 * and soreceive() will return when it has either a 556 * control msg or a data msg. 557 * We have no use for control msg., but must grab them 558 * and then throw them away so we know what is going 559 * on. 560 */ 561 auio.uio_resid = len = 100000000; /* Anything Big */ 562 auio.uio_procp = p; 563 do { 564 rcvflg = 0; 565 error = soreceive(so, (struct mbuf **)0, 566 &auio, mp, &control, &rcvflg); 567 if (control) 568 m_freem(control); 569 if (error == EWOULDBLOCK && rep) { 570 if (rep->r_flags & R_SOFTTERM) 571 return (EINTR); 572 } 573 } while (error == EWOULDBLOCK || 574 (!error && *mp == NULL && control)); 575 if ((rcvflg & MSG_EOR) == 0) 576 printf("Egad!!\n"); 577 if (!error && *mp == NULL) 578 error = EPIPE; 579 len -= auio.uio_resid; 580 } 581 errout: 582 if (error && error != EINTR && error != ERESTART) { 583 m_freem(*mp); 584 *mp = (struct mbuf *)0; 585 if (error != EPIPE) 586 log(LOG_INFO, 587 "receive error %d from nfs server %s\n", 588 error, 589 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 590 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 591 if (!error) 592 error = nfs_reconnect(rep); 593 if (!error) 594 goto tryagain; 595 } 596 } else { 597 if ((so = rep->r_nmp->nm_so) == NULL) 598 return (EACCES); 599 if (so->so_state & SS_ISCONNECTED) 600 getnam = (struct mbuf **)0; 601 else 602 getnam = aname; 603 auio.uio_resid = len = 1000000; 604 auio.uio_procp = p; 605 do { 606 rcvflg = 0; 607 error = soreceive(so, getnam, &auio, mp, 608 (struct mbuf **)0, &rcvflg); 609 if (error == EWOULDBLOCK && 610 (rep->r_flags & R_SOFTTERM)) 611 return (EINTR); 612 } while (error == EWOULDBLOCK); 613 len -= auio.uio_resid; 614 } 615 if (error) { 616 m_freem(*mp); 617 *mp = (struct mbuf *)0; 618 } 619 /* 620 * Search for any mbufs that are not a multiple of 4 bytes long 621 * or with m_data not longword aligned. 622 * These could cause pointer alignment problems, so copy them to 623 * well aligned mbufs. 624 */ 625 nfs_realign(*mp, 5 * NFSX_UNSIGNED); 626 return (error); 627 } 628 629 /* 630 * Implement receipt of reply on a socket. 631 * We must search through the list of received datagrams matching them 632 * with outstanding requests using the xid, until ours is found. 633 */ 634 /* ARGSUSED */ 635 nfs_reply(myrep) 636 struct nfsreq *myrep; 637 { 638 register struct nfsreq *rep; 639 register struct nfsmount *nmp = myrep->r_nmp; 640 register long t1; 641 struct mbuf *mrep, *nam, *md; 642 u_long rxid, *tl; 643 caddr_t dpos, cp2; 644 int error; 645 646 /* 647 * Loop around until we get our own reply 648 */ 649 for (;;) { 650 /* 651 * Lock against other receivers so that I don't get stuck in 652 * sbwait() after someone else has received my reply for me. 653 * Also necessary for connection based protocols to avoid 654 * race conditions during a reconnect. 655 */ 656 if (error = nfs_rcvlock(myrep)) 657 return (error); 658 /* Already received, bye bye */ 659 if (myrep->r_mrep != NULL) { 660 nfs_rcvunlock(&nmp->nm_flag); 661 return (0); 662 } 663 /* 664 * Get the next Rpc reply off the socket 665 */ 666 error = nfs_receive(myrep, &nam, &mrep); 667 nfs_rcvunlock(&nmp->nm_flag); 668 if (error) { 669 670 /* 671 * Ignore routing errors on connectionless protocols?? 672 */ 673 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 674 nmp->nm_so->so_error = 0; 675 if (myrep->r_flags & R_GETONEREP) 676 return (0); 677 continue; 678 } 679 return (error); 680 } 681 if (nam) 682 m_freem(nam); 683 684 /* 685 * Get the xid and check that it is an rpc reply 686 */ 687 md = mrep; 688 dpos = mtod(md, caddr_t); 689 nfsm_dissect(tl, u_long *, 2*NFSX_UNSIGNED); 690 rxid = *tl++; 691 if (*tl != rpc_reply) { 692 if (nmp->nm_flag & NFSMNT_NQNFS) { 693 if (nqnfs_callback(nmp, mrep, md, dpos)) 694 nfsstats.rpcinvalid++; 695 } else { 696 nfsstats.rpcinvalid++; 697 m_freem(mrep); 698 } 699 nfsmout: 700 if (myrep->r_flags & R_GETONEREP) 701 return (0); 702 continue; 703 } 704 705 /* 706 * Loop through the request list to match up the reply 707 * Iff no match, just drop the datagram 708 */ 709 for (rep = nfs_reqq.tqh_first; rep != 0; 710 rep = rep->r_chain.tqe_next) { 711 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 712 /* Found it.. */ 713 rep->r_mrep = mrep; 714 rep->r_md = md; 715 rep->r_dpos = dpos; 716 if (nfsrtton) { 717 struct rttl *rt; 718 719 rt = &nfsrtt.rttl[nfsrtt.pos]; 720 rt->proc = rep->r_procnum; 721 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 722 rt->sent = nmp->nm_sent; 723 rt->cwnd = nmp->nm_cwnd; 724 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 725 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 726 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 727 rt->tstamp = time; 728 if (rep->r_flags & R_TIMING) 729 rt->rtt = rep->r_rtt; 730 else 731 rt->rtt = 1000000; 732 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 733 } 734 /* 735 * Update congestion window. 736 * Do the additive increase of 737 * one rpc/rtt. 738 */ 739 if (nmp->nm_cwnd <= nmp->nm_sent) { 740 nmp->nm_cwnd += 741 (NFS_CWNDSCALE * NFS_CWNDSCALE + 742 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 743 if (nmp->nm_cwnd > NFS_MAXCWND) 744 nmp->nm_cwnd = NFS_MAXCWND; 745 } 746 rep->r_flags &= ~R_SENT; 747 nmp->nm_sent -= NFS_CWNDSCALE; 748 /* 749 * Update rtt using a gain of 0.125 on the mean 750 * and a gain of 0.25 on the deviation. 751 */ 752 if (rep->r_flags & R_TIMING) { 753 /* 754 * Since the timer resolution of 755 * NFS_HZ is so course, it can often 756 * result in r_rtt == 0. Since 757 * r_rtt == N means that the actual 758 * rtt is between N+dt and N+2-dt ticks, 759 * add 1. 760 */ 761 t1 = rep->r_rtt + 1; 762 t1 -= (NFS_SRTT(rep) >> 3); 763 NFS_SRTT(rep) += t1; 764 if (t1 < 0) 765 t1 = -t1; 766 t1 -= (NFS_SDRTT(rep) >> 2); 767 NFS_SDRTT(rep) += t1; 768 } 769 nmp->nm_timeouts = 0; 770 break; 771 } 772 } 773 /* 774 * If not matched to a request, drop it. 775 * If it's mine, get out. 776 */ 777 if (rep == 0) { 778 nfsstats.rpcunexpected++; 779 m_freem(mrep); 780 } else if (rep == myrep) { 781 if (rep->r_mrep == NULL) 782 panic("nfsreply nil"); 783 return (0); 784 } 785 if (myrep->r_flags & R_GETONEREP) 786 return (0); 787 } 788 } 789 790 /* 791 * nfs_request - goes something like this 792 * - fill in request struct 793 * - links it into list 794 * - calls nfs_send() for first transmit 795 * - calls nfs_receive() to get reply 796 * - break down rpc header and return with nfs reply pointed to 797 * by mrep or error 798 * nb: always frees up mreq mbuf list 799 */ 800 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 801 struct vnode *vp; 802 struct mbuf *mrest; 803 int procnum; 804 struct proc *procp; 805 struct ucred *cred; 806 struct mbuf **mrp; 807 struct mbuf **mdp; 808 caddr_t *dposp; 809 { 810 register struct mbuf *m, *mrep; 811 register struct nfsreq *rep; 812 register u_long *tl; 813 register int i; 814 struct nfsmount *nmp; 815 struct mbuf *md, *mheadend; 816 struct nfsreq *reph; 817 struct nfsnode *np; 818 time_t reqtime, waituntil; 819 caddr_t dpos, cp2; 820 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; 821 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 822 u_long xid; 823 u_quad_t frev; 824 char *auth_str; 825 826 nmp = VFSTONFS(vp->v_mount); 827 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 828 rep->r_nmp = nmp; 829 rep->r_vp = vp; 830 rep->r_procp = procp; 831 rep->r_procnum = procnum; 832 i = 0; 833 m = mrest; 834 while (m) { 835 i += m->m_len; 836 m = m->m_next; 837 } 838 mrest_len = i; 839 840 /* 841 * Get the RPC header with authorization. 842 */ 843 kerbauth: 844 auth_str = (char *)0; 845 if (nmp->nm_flag & NFSMNT_KERB) { 846 if (failed_auth) { 847 error = nfs_getauth(nmp, rep, cred, &auth_type, 848 &auth_str, &auth_len); 849 if (error) { 850 free((caddr_t)rep, M_NFSREQ); 851 m_freem(mrest); 852 return (error); 853 } 854 } else { 855 auth_type = RPCAUTH_UNIX; 856 auth_len = 5 * NFSX_UNSIGNED; 857 } 858 } else { 859 auth_type = RPCAUTH_UNIX; 860 if (cred->cr_ngroups < 1) 861 panic("nfsreq nogrps"); 862 auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ? 863 nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) + 864 5 * NFSX_UNSIGNED; 865 } 866 m = nfsm_rpchead(cred, (nmp->nm_flag & NFSMNT_NQNFS), procnum, 867 auth_type, auth_len, auth_str, mrest, mrest_len, &mheadend, &xid); 868 if (auth_str) 869 free(auth_str, M_TEMP); 870 871 /* 872 * For stream protocols, insert a Sun RPC Record Mark. 873 */ 874 if (nmp->nm_sotype == SOCK_STREAM) { 875 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 876 *mtod(m, u_long *) = htonl(0x80000000 | 877 (m->m_pkthdr.len - NFSX_UNSIGNED)); 878 } 879 rep->r_mreq = m; 880 rep->r_xid = xid; 881 tryagain: 882 if (nmp->nm_flag & NFSMNT_SOFT) 883 rep->r_retry = nmp->nm_retry; 884 else 885 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 886 rep->r_rtt = rep->r_rexmit = 0; 887 if (proct[procnum] > 0) 888 rep->r_flags = R_TIMING; 889 else 890 rep->r_flags = 0; 891 rep->r_mrep = NULL; 892 893 /* 894 * Do the client side RPC. 895 */ 896 nfsstats.rpcrequests++; 897 /* 898 * Chain request into list of outstanding requests. Be sure 899 * to put it LAST so timer finds oldest requests first. 900 */ 901 s = splsoftclock(); 902 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 903 904 /* Get send time for nqnfs */ 905 reqtime = time.tv_sec; 906 907 /* 908 * If backing off another request or avoiding congestion, don't 909 * send this one now but let timer do it. If not timing a request, 910 * do it now. 911 */ 912 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 913 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 914 nmp->nm_sent < nmp->nm_cwnd)) { 915 splx(s); 916 if (nmp->nm_soflags & PR_CONNREQUIRED) 917 error = nfs_sndlock(&nmp->nm_flag, rep); 918 if (!error) { 919 m = m_copym(m, 0, M_COPYALL, M_WAIT); 920 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 921 if (nmp->nm_soflags & PR_CONNREQUIRED) 922 nfs_sndunlock(&nmp->nm_flag); 923 } 924 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 925 nmp->nm_sent += NFS_CWNDSCALE; 926 rep->r_flags |= R_SENT; 927 } 928 } else { 929 splx(s); 930 rep->r_rtt = -1; 931 } 932 933 /* 934 * Wait for the reply from our send or the timer's. 935 */ 936 if (!error || error == EPIPE) 937 error = nfs_reply(rep); 938 939 /* 940 * RPC done, unlink the request. 941 */ 942 s = splsoftclock(); 943 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 944 splx(s); 945 946 /* 947 * Decrement the outstanding request count. 948 */ 949 if (rep->r_flags & R_SENT) { 950 rep->r_flags &= ~R_SENT; /* paranoia */ 951 nmp->nm_sent -= NFS_CWNDSCALE; 952 } 953 954 /* 955 * If there was a successful reply and a tprintf msg. 956 * tprintf a response. 957 */ 958 if (!error && (rep->r_flags & R_TPRINTFMSG)) 959 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 960 "is alive again"); 961 mrep = rep->r_mrep; 962 md = rep->r_md; 963 dpos = rep->r_dpos; 964 if (error) { 965 m_freem(rep->r_mreq); 966 free((caddr_t)rep, M_NFSREQ); 967 return (error); 968 } 969 970 /* 971 * break down the rpc header and check if ok 972 */ 973 nfsm_dissect(tl, u_long *, 3*NFSX_UNSIGNED); 974 if (*tl++ == rpc_msgdenied) { 975 if (*tl == rpc_mismatch) 976 error = EOPNOTSUPP; 977 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 978 if (*tl == rpc_rejectedcred && failed_auth == 0) { 979 failed_auth++; 980 mheadend->m_next = (struct mbuf *)0; 981 m_freem(mrep); 982 m_freem(rep->r_mreq); 983 goto kerbauth; 984 } else 985 error = EAUTH; 986 } else 987 error = EACCES; 988 m_freem(mrep); 989 m_freem(rep->r_mreq); 990 free((caddr_t)rep, M_NFSREQ); 991 return (error); 992 } 993 994 /* 995 * skip over the auth_verf, someday we may want to cache auth_short's 996 * for nfs_reqhead(), but for now just dump it 997 */ 998 if (*++tl != 0) { 999 i = nfsm_rndup(fxdr_unsigned(long, *tl)); 1000 nfsm_adv(i); 1001 } 1002 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1003 /* 0 == ok */ 1004 if (*tl == 0) { 1005 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1006 if (*tl != 0) { 1007 error = fxdr_unsigned(int, *tl); 1008 m_freem(mrep); 1009 if ((nmp->nm_flag & NFSMNT_NQNFS) && 1010 error == NQNFS_TRYLATER) { 1011 error = 0; 1012 waituntil = time.tv_sec + trylater_delay; 1013 while (time.tv_sec < waituntil) 1014 (void) tsleep((caddr_t)&lbolt, 1015 PSOCK, "nqnfstry", 0); 1016 trylater_delay *= nfs_backoff[trylater_cnt]; 1017 if (trylater_cnt < 7) 1018 trylater_cnt++; 1019 goto tryagain; 1020 } 1021 1022 /* 1023 * If the File Handle was stale, invalidate the 1024 * lookup cache, just in case. 1025 */ 1026 if (error == ESTALE) 1027 cache_purge(vp); 1028 m_freem(rep->r_mreq); 1029 free((caddr_t)rep, M_NFSREQ); 1030 return (error); 1031 } 1032 1033 /* 1034 * For nqnfs, get any lease in reply 1035 */ 1036 if (nmp->nm_flag & NFSMNT_NQNFS) { 1037 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1038 if (*tl) { 1039 np = VTONFS(vp); 1040 nqlflag = fxdr_unsigned(int, *tl); 1041 nfsm_dissect(tl, u_long *, 4*NFSX_UNSIGNED); 1042 cachable = fxdr_unsigned(int, *tl++); 1043 reqtime += fxdr_unsigned(int, *tl++); 1044 if (reqtime > time.tv_sec) { 1045 fxdr_hyper(tl, &frev); 1046 nqnfs_clientlease(nmp, np, nqlflag, 1047 cachable, reqtime, frev); 1048 } 1049 } 1050 } 1051 *mrp = mrep; 1052 *mdp = md; 1053 *dposp = dpos; 1054 m_freem(rep->r_mreq); 1055 FREE((caddr_t)rep, M_NFSREQ); 1056 return (0); 1057 } 1058 m_freem(mrep); 1059 m_freem(rep->r_mreq); 1060 free((caddr_t)rep, M_NFSREQ); 1061 error = EPROTONOSUPPORT; 1062 nfsmout: 1063 return (error); 1064 } 1065 #endif /* NFSCLIENT */ 1066 1067 /* 1068 * Generate the rpc reply header 1069 * siz arg. is used to decide if adding a cluster is worthwhile 1070 */ 1071 nfs_rephead(siz, nd, err, cache, frev, mrq, mbp, bposp) 1072 int siz; 1073 struct nfsd *nd; 1074 int err; 1075 int cache; 1076 u_quad_t *frev; 1077 struct mbuf **mrq; 1078 struct mbuf **mbp; 1079 caddr_t *bposp; 1080 { 1081 register u_long *tl; 1082 register struct mbuf *mreq; 1083 caddr_t bpos; 1084 struct mbuf *mb, *mb2; 1085 1086 MGETHDR(mreq, M_WAIT, MT_DATA); 1087 mb = mreq; 1088 /* 1089 * If this is a big reply, use a cluster else 1090 * try and leave leading space for the lower level headers. 1091 */ 1092 siz += RPC_REPLYSIZ; 1093 if (siz >= MINCLSIZE) { 1094 MCLGET(mreq, M_WAIT); 1095 } else 1096 mreq->m_data += max_hdr; 1097 tl = mtod(mreq, u_long *); 1098 mreq->m_len = 6*NFSX_UNSIGNED; 1099 bpos = ((caddr_t)tl)+mreq->m_len; 1100 *tl++ = txdr_unsigned(nd->nd_retxid); 1101 *tl++ = rpc_reply; 1102 if (err == ERPCMISMATCH || err == NQNFS_AUTHERR) { 1103 *tl++ = rpc_msgdenied; 1104 if (err == NQNFS_AUTHERR) { 1105 *tl++ = rpc_autherr; 1106 *tl = rpc_rejectedcred; 1107 mreq->m_len -= NFSX_UNSIGNED; 1108 bpos -= NFSX_UNSIGNED; 1109 } else { 1110 *tl++ = rpc_mismatch; 1111 *tl++ = txdr_unsigned(2); 1112 *tl = txdr_unsigned(2); 1113 } 1114 } else { 1115 *tl++ = rpc_msgaccepted; 1116 *tl++ = 0; 1117 *tl++ = 0; 1118 switch (err) { 1119 case EPROGUNAVAIL: 1120 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1121 break; 1122 case EPROGMISMATCH: 1123 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1124 nfsm_build(tl, u_long *, 2*NFSX_UNSIGNED); 1125 *tl++ = txdr_unsigned(2); 1126 *tl = txdr_unsigned(2); /* someday 3 */ 1127 break; 1128 case EPROCUNAVAIL: 1129 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1130 break; 1131 default: 1132 *tl = 0; 1133 if (err != VNOVAL) { 1134 nfsm_build(tl, u_long *, NFSX_UNSIGNED); 1135 if (err) 1136 *tl = txdr_unsigned(nfsrv_errmap[err - 1]); 1137 else 1138 *tl = 0; 1139 } 1140 break; 1141 }; 1142 } 1143 1144 /* 1145 * For nqnfs, piggyback lease as requested. 1146 */ 1147 if (nd->nd_nqlflag != NQL_NOVAL && err == 0) { 1148 if (nd->nd_nqlflag) { 1149 nfsm_build(tl, u_long *, 5*NFSX_UNSIGNED); 1150 *tl++ = txdr_unsigned(nd->nd_nqlflag); 1151 *tl++ = txdr_unsigned(cache); 1152 *tl++ = txdr_unsigned(nd->nd_duration); 1153 txdr_hyper(frev, tl); 1154 } else { 1155 if (nd->nd_nqlflag != 0) 1156 panic("nqreph"); 1157 nfsm_build(tl, u_long *, NFSX_UNSIGNED); 1158 *tl = 0; 1159 } 1160 } 1161 *mrq = mreq; 1162 *mbp = mb; 1163 *bposp = bpos; 1164 if (err != 0 && err != VNOVAL) 1165 nfsstats.srvrpc_errs++; 1166 return (0); 1167 } 1168 1169 /* 1170 * Nfs timer routine 1171 * Scan the nfsreq list and retranmit any requests that have timed out 1172 * To avoid retransmission attempts on STREAM sockets (in the future) make 1173 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1174 */ 1175 void 1176 nfs_timer(arg) 1177 void *arg; 1178 { 1179 register struct nfsreq *rep; 1180 register struct mbuf *m; 1181 register struct socket *so; 1182 register struct nfsmount *nmp; 1183 register int timeo; 1184 static long lasttime = 0; 1185 int s, error; 1186 1187 s = splnet(); 1188 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1189 nmp = rep->r_nmp; 1190 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1191 continue; 1192 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1193 rep->r_flags |= R_SOFTTERM; 1194 continue; 1195 } 1196 if (rep->r_rtt >= 0) { 1197 rep->r_rtt++; 1198 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1199 timeo = nmp->nm_timeo; 1200 else 1201 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1202 if (nmp->nm_timeouts > 0) 1203 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1204 if (rep->r_rtt <= timeo) 1205 continue; 1206 if (nmp->nm_timeouts < 8) 1207 nmp->nm_timeouts++; 1208 } 1209 /* 1210 * Check for server not responding 1211 */ 1212 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1213 rep->r_rexmit > nmp->nm_deadthresh) { 1214 nfs_msg(rep->r_procp, 1215 nmp->nm_mountp->mnt_stat.f_mntfromname, 1216 "not responding"); 1217 rep->r_flags |= R_TPRINTFMSG; 1218 } 1219 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1220 nfsstats.rpctimeouts++; 1221 rep->r_flags |= R_SOFTTERM; 1222 continue; 1223 } 1224 if (nmp->nm_sotype != SOCK_DGRAM) { 1225 if (++rep->r_rexmit > NFS_MAXREXMIT) 1226 rep->r_rexmit = NFS_MAXREXMIT; 1227 continue; 1228 } 1229 if ((so = nmp->nm_so) == NULL) 1230 continue; 1231 1232 /* 1233 * If there is enough space and the window allows.. 1234 * Resend it 1235 * Set r_rtt to -1 in case we fail to send it now. 1236 */ 1237 rep->r_rtt = -1; 1238 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1239 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1240 (rep->r_flags & R_SENT) || 1241 nmp->nm_sent < nmp->nm_cwnd) && 1242 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1243 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1244 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1245 (struct mbuf *)0, (struct mbuf *)0); 1246 else 1247 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1248 nmp->nm_nam, (struct mbuf *)0); 1249 if (error) { 1250 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) 1251 so->so_error = 0; 1252 } else { 1253 /* 1254 * Iff first send, start timing 1255 * else turn timing off, backoff timer 1256 * and divide congestion window by 2. 1257 */ 1258 if (rep->r_flags & R_SENT) { 1259 rep->r_flags &= ~R_TIMING; 1260 if (++rep->r_rexmit > NFS_MAXREXMIT) 1261 rep->r_rexmit = NFS_MAXREXMIT; 1262 nmp->nm_cwnd >>= 1; 1263 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1264 nmp->nm_cwnd = NFS_CWNDSCALE; 1265 nfsstats.rpcretries++; 1266 } else { 1267 rep->r_flags |= R_SENT; 1268 nmp->nm_sent += NFS_CWNDSCALE; 1269 } 1270 rep->r_rtt = 0; 1271 } 1272 } 1273 } 1274 1275 #ifdef NFSSERVER 1276 /* 1277 * Call the nqnfs server timer once a second to handle leases. 1278 */ 1279 if (lasttime != time.tv_sec) { 1280 lasttime = time.tv_sec; 1281 nqnfs_serverd(); 1282 } 1283 #endif /* NFSSERVER */ 1284 splx(s); 1285 timeout(nfs_timer, (void *)0, hz / NFS_HZ); 1286 } 1287 1288 /* 1289 * Test for a termination condition pending on the process. 1290 * This is used for NFSMNT_INT mounts. 1291 */ 1292 nfs_sigintr(nmp, rep, p) 1293 struct nfsmount *nmp; 1294 struct nfsreq *rep; 1295 register struct proc *p; 1296 { 1297 1298 if (rep && (rep->r_flags & R_SOFTTERM)) 1299 return (EINTR); 1300 if (!(nmp->nm_flag & NFSMNT_INT)) 1301 return (0); 1302 if (p && p->p_siglist && 1303 (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) & 1304 NFSINT_SIGMASK)) 1305 return (EINTR); 1306 return (0); 1307 } 1308 1309 /* 1310 * Lock a socket against others. 1311 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1312 * and also to avoid race conditions between the processes with nfs requests 1313 * in progress when a reconnect is necessary. 1314 */ 1315 nfs_sndlock(flagp, rep) 1316 register int *flagp; 1317 struct nfsreq *rep; 1318 { 1319 struct proc *p; 1320 int slpflag = 0, slptimeo = 0; 1321 1322 if (rep) { 1323 p = rep->r_procp; 1324 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1325 slpflag = PCATCH; 1326 } else 1327 p = (struct proc *)0; 1328 while (*flagp & NFSMNT_SNDLOCK) { 1329 if (nfs_sigintr(rep->r_nmp, rep, p)) 1330 return (EINTR); 1331 *flagp |= NFSMNT_WANTSND; 1332 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1333 slptimeo); 1334 if (slpflag == PCATCH) { 1335 slpflag = 0; 1336 slptimeo = 2 * hz; 1337 } 1338 } 1339 *flagp |= NFSMNT_SNDLOCK; 1340 return (0); 1341 } 1342 1343 /* 1344 * Unlock the stream socket for others. 1345 */ 1346 void 1347 nfs_sndunlock(flagp) 1348 register int *flagp; 1349 { 1350 1351 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1352 panic("nfs sndunlock"); 1353 *flagp &= ~NFSMNT_SNDLOCK; 1354 if (*flagp & NFSMNT_WANTSND) { 1355 *flagp &= ~NFSMNT_WANTSND; 1356 wakeup((caddr_t)flagp); 1357 } 1358 } 1359 1360 nfs_rcvlock(rep) 1361 register struct nfsreq *rep; 1362 { 1363 register int *flagp = &rep->r_nmp->nm_flag; 1364 int slpflag, slptimeo = 0; 1365 1366 if (*flagp & NFSMNT_INT) 1367 slpflag = PCATCH; 1368 else 1369 slpflag = 0; 1370 while (*flagp & NFSMNT_RCVLOCK) { 1371 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1372 return (EINTR); 1373 *flagp |= NFSMNT_WANTRCV; 1374 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1375 slptimeo); 1376 if (slpflag == PCATCH) { 1377 slpflag = 0; 1378 slptimeo = 2 * hz; 1379 } 1380 } 1381 *flagp |= NFSMNT_RCVLOCK; 1382 return (0); 1383 } 1384 1385 /* 1386 * Unlock the stream socket for others. 1387 */ 1388 void 1389 nfs_rcvunlock(flagp) 1390 register int *flagp; 1391 { 1392 1393 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1394 panic("nfs rcvunlock"); 1395 *flagp &= ~NFSMNT_RCVLOCK; 1396 if (*flagp & NFSMNT_WANTRCV) { 1397 *flagp &= ~NFSMNT_WANTRCV; 1398 wakeup((caddr_t)flagp); 1399 } 1400 } 1401 1402 /* 1403 * Check for badly aligned mbuf data areas and 1404 * realign data in an mbuf list by copying the data areas up, as required. 1405 */ 1406 void 1407 nfs_realign(m, hsiz) 1408 register struct mbuf *m; 1409 int hsiz; 1410 { 1411 register struct mbuf *m2; 1412 register int siz, mlen, olen; 1413 register caddr_t tcp, fcp; 1414 struct mbuf *mnew; 1415 1416 while (m) { 1417 /* 1418 * This never happens for UDP, rarely happens for TCP 1419 * but frequently happens for iso transport. 1420 */ 1421 if ((m->m_len & 0x3) || (mtod(m, int) & 0x3)) { 1422 olen = m->m_len; 1423 fcp = mtod(m, caddr_t); 1424 if ((int)fcp & 0x3) { 1425 m->m_flags &= ~M_PKTHDR; 1426 if (m->m_flags & M_EXT) 1427 m->m_data = m->m_ext.ext_buf + 1428 ((m->m_ext.ext_size - olen) & ~0x3); 1429 else 1430 m->m_data = m->m_dat; 1431 } 1432 m->m_len = 0; 1433 tcp = mtod(m, caddr_t); 1434 mnew = m; 1435 m2 = m->m_next; 1436 1437 /* 1438 * If possible, only put the first invariant part 1439 * of the RPC header in the first mbuf. 1440 */ 1441 mlen = M_TRAILINGSPACE(m); 1442 if (olen <= hsiz && mlen > hsiz) 1443 mlen = hsiz; 1444 1445 /* 1446 * Loop through the mbuf list consolidating data. 1447 */ 1448 while (m) { 1449 while (olen > 0) { 1450 if (mlen == 0) { 1451 m2->m_flags &= ~M_PKTHDR; 1452 if (m2->m_flags & M_EXT) 1453 m2->m_data = m2->m_ext.ext_buf; 1454 else 1455 m2->m_data = m2->m_dat; 1456 m2->m_len = 0; 1457 mlen = M_TRAILINGSPACE(m2); 1458 tcp = mtod(m2, caddr_t); 1459 mnew = m2; 1460 m2 = m2->m_next; 1461 } 1462 siz = min(mlen, olen); 1463 if (tcp != fcp) 1464 bcopy(fcp, tcp, siz); 1465 mnew->m_len += siz; 1466 mlen -= siz; 1467 olen -= siz; 1468 tcp += siz; 1469 fcp += siz; 1470 } 1471 m = m->m_next; 1472 if (m) { 1473 olen = m->m_len; 1474 fcp = mtod(m, caddr_t); 1475 } 1476 } 1477 1478 /* 1479 * Finally, set m_len == 0 for any trailing mbufs that have 1480 * been copied out of. 1481 */ 1482 while (m2) { 1483 m2->m_len = 0; 1484 m2 = m2->m_next; 1485 } 1486 return; 1487 } 1488 m = m->m_next; 1489 } 1490 } 1491 1492 /* 1493 * Parse an RPC request 1494 * - verify it 1495 * - fill in the cred struct. 1496 */ 1497 nfs_getreq(nd, has_header) 1498 register struct nfsd *nd; 1499 int has_header; 1500 { 1501 register int len, i; 1502 register u_long *tl; 1503 register long t1; 1504 struct uio uio; 1505 struct iovec iov; 1506 caddr_t dpos, cp2; 1507 u_long nfsvers, auth_type; 1508 int error = 0, nqnfs = 0; 1509 struct mbuf *mrep, *md; 1510 1511 mrep = nd->nd_mrep; 1512 md = nd->nd_md; 1513 dpos = nd->nd_dpos; 1514 if (has_header) { 1515 nfsm_dissect(tl, u_long *, 10*NFSX_UNSIGNED); 1516 nd->nd_retxid = fxdr_unsigned(u_long, *tl++); 1517 if (*tl++ != rpc_call) { 1518 m_freem(mrep); 1519 return (EBADRPC); 1520 } 1521 } else { 1522 nfsm_dissect(tl, u_long *, 8*NFSX_UNSIGNED); 1523 } 1524 nd->nd_repstat = 0; 1525 if (*tl++ != rpc_vers) { 1526 nd->nd_repstat = ERPCMISMATCH; 1527 nd->nd_procnum = NFSPROC_NOOP; 1528 return (0); 1529 } 1530 nfsvers = nfs_vers; 1531 if (*tl != nfs_prog) { 1532 if (*tl == nqnfs_prog) { 1533 nqnfs++; 1534 nfsvers = nqnfs_vers; 1535 } else { 1536 nd->nd_repstat = EPROGUNAVAIL; 1537 nd->nd_procnum = NFSPROC_NOOP; 1538 return (0); 1539 } 1540 } 1541 tl++; 1542 if (*tl++ != nfsvers) { 1543 nd->nd_repstat = EPROGMISMATCH; 1544 nd->nd_procnum = NFSPROC_NOOP; 1545 return (0); 1546 } 1547 nd->nd_procnum = fxdr_unsigned(u_long, *tl++); 1548 if (nd->nd_procnum == NFSPROC_NULL) 1549 return (0); 1550 if (nd->nd_procnum >= NFS_NPROCS || 1551 (!nqnfs && nd->nd_procnum > NFSPROC_STATFS) || 1552 (*tl != rpc_auth_unix && *tl != rpc_auth_kerb)) { 1553 nd->nd_repstat = EPROCUNAVAIL; 1554 nd->nd_procnum = NFSPROC_NOOP; 1555 return (0); 1556 } 1557 auth_type = *tl++; 1558 len = fxdr_unsigned(int, *tl++); 1559 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1560 m_freem(mrep); 1561 return (EBADRPC); 1562 } 1563 1564 /* 1565 * Handle auth_unix or auth_kerb. 1566 */ 1567 if (auth_type == rpc_auth_unix) { 1568 len = fxdr_unsigned(int, *++tl); 1569 if (len < 0 || len > NFS_MAXNAMLEN) { 1570 m_freem(mrep); 1571 return (EBADRPC); 1572 } 1573 nfsm_adv(nfsm_rndup(len)); 1574 nfsm_dissect(tl, u_long *, 3*NFSX_UNSIGNED); 1575 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1576 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1577 len = fxdr_unsigned(int, *tl); 1578 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1579 m_freem(mrep); 1580 return (EBADRPC); 1581 } 1582 nfsm_dissect(tl, u_long *, (len + 2)*NFSX_UNSIGNED); 1583 for (i = 1; i <= len; i++) 1584 if (i < NGROUPS) 1585 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1586 else 1587 tl++; 1588 nd->nd_cr.cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1); 1589 } else if (auth_type == rpc_auth_kerb) { 1590 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1591 nd->nd_authlen = fxdr_unsigned(int, *tl); 1592 uio.uio_resid = nfsm_rndup(nd->nd_authlen); 1593 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1594 m_freem(mrep); 1595 return (EBADRPC); 1596 } 1597 uio.uio_offset = 0; 1598 uio.uio_iov = &iov; 1599 uio.uio_iovcnt = 1; 1600 uio.uio_segflg = UIO_SYSSPACE; 1601 iov.iov_base = (caddr_t)nd->nd_authstr; 1602 iov.iov_len = RPCAUTH_MAXSIZ; 1603 nfsm_mtouio(&uio, uio.uio_resid); 1604 nfsm_dissect(tl, u_long *, 2 * NFSX_UNSIGNED); 1605 nd->nd_flag |= NFSD_NEEDAUTH; 1606 } 1607 1608 /* 1609 * Do we have any use for the verifier. 1610 * According to the "Remote Procedure Call Protocol Spec." it 1611 * should be AUTH_NULL, but some clients make it AUTH_UNIX? 1612 * For now, just skip over it 1613 */ 1614 len = fxdr_unsigned(int, *++tl); 1615 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1616 m_freem(mrep); 1617 return (EBADRPC); 1618 } 1619 if (len > 0) { 1620 nfsm_adv(nfsm_rndup(len)); 1621 } 1622 1623 /* 1624 * For nqnfs, get piggybacked lease request. 1625 */ 1626 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1627 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1628 nd->nd_nqlflag = fxdr_unsigned(int, *tl); 1629 if (nd->nd_nqlflag) { 1630 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1631 nd->nd_duration = fxdr_unsigned(int, *tl); 1632 } else 1633 nd->nd_duration = NQ_MINLEASE; 1634 } else { 1635 nd->nd_nqlflag = NQL_NOVAL; 1636 nd->nd_duration = NQ_MINLEASE; 1637 } 1638 nd->nd_md = md; 1639 nd->nd_dpos = dpos; 1640 return (0); 1641 nfsmout: 1642 return (error); 1643 } 1644 1645 nfs_msg(p, server, msg) 1646 struct proc *p; 1647 char *server, *msg; 1648 { 1649 tpr_t tpr; 1650 1651 if (p) 1652 tpr = tprintf_open(p); 1653 else 1654 tpr = NULL; 1655 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1656 tprintf_close(tpr); 1657 } 1658 1659 #ifdef NFSSERVER 1660 int nfsrv_null(), 1661 nfsrv_getattr(), 1662 nfsrv_setattr(), 1663 nfsrv_lookup(), 1664 nfsrv_readlink(), 1665 nfsrv_read(), 1666 nfsrv_write(), 1667 nfsrv_create(), 1668 nfsrv_remove(), 1669 nfsrv_rename(), 1670 nfsrv_link(), 1671 nfsrv_symlink(), 1672 nfsrv_mkdir(), 1673 nfsrv_rmdir(), 1674 nfsrv_readdir(), 1675 nfsrv_statfs(), 1676 nfsrv_noop(), 1677 nqnfsrv_readdirlook(), 1678 nqnfsrv_getlease(), 1679 nqnfsrv_vacated(), 1680 nqnfsrv_access(); 1681 1682 int (*nfsrv_procs[NFS_NPROCS])() = { 1683 nfsrv_null, 1684 nfsrv_getattr, 1685 nfsrv_setattr, 1686 nfsrv_noop, 1687 nfsrv_lookup, 1688 nfsrv_readlink, 1689 nfsrv_read, 1690 nfsrv_noop, 1691 nfsrv_write, 1692 nfsrv_create, 1693 nfsrv_remove, 1694 nfsrv_rename, 1695 nfsrv_link, 1696 nfsrv_symlink, 1697 nfsrv_mkdir, 1698 nfsrv_rmdir, 1699 nfsrv_readdir, 1700 nfsrv_statfs, 1701 nqnfsrv_readdirlook, 1702 nqnfsrv_getlease, 1703 nqnfsrv_vacated, 1704 nfsrv_noop, 1705 nqnfsrv_access, 1706 }; 1707 1708 /* 1709 * Socket upcall routine for the nfsd sockets. 1710 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1711 * Essentially do as much as possible non-blocking, else punt and it will 1712 * be called with M_WAIT from an nfsd. 1713 */ 1714 void 1715 nfsrv_rcv(so, arg, waitflag) 1716 struct socket *so; 1717 caddr_t arg; 1718 int waitflag; 1719 { 1720 register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1721 register struct mbuf *m; 1722 struct mbuf *mp, *nam; 1723 struct uio auio; 1724 int flags, error; 1725 1726 if ((slp->ns_flag & SLP_VALID) == 0) 1727 return; 1728 #ifdef notdef 1729 /* 1730 * Define this to test for nfsds handling this under heavy load. 1731 */ 1732 if (waitflag == M_DONTWAIT) { 1733 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1734 } 1735 #endif 1736 auio.uio_procp = NULL; 1737 if (so->so_type == SOCK_STREAM) { 1738 /* 1739 * If there are already records on the queue, defer soreceive() 1740 * to an nfsd so that there is feedback to the TCP layer that 1741 * the nfs servers are heavily loaded. 1742 */ 1743 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1744 slp->ns_flag |= SLP_NEEDQ; 1745 goto dorecs; 1746 } 1747 1748 /* 1749 * Do soreceive(). 1750 */ 1751 auio.uio_resid = 1000000000; 1752 flags = MSG_DONTWAIT; 1753 error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1754 if (error || mp == (struct mbuf *)0) { 1755 if (error == EWOULDBLOCK) 1756 slp->ns_flag |= SLP_NEEDQ; 1757 else 1758 slp->ns_flag |= SLP_DISCONN; 1759 goto dorecs; 1760 } 1761 m = mp; 1762 if (slp->ns_rawend) { 1763 slp->ns_rawend->m_next = m; 1764 slp->ns_cc += 1000000000 - auio.uio_resid; 1765 } else { 1766 slp->ns_raw = m; 1767 slp->ns_cc = 1000000000 - auio.uio_resid; 1768 } 1769 while (m->m_next) 1770 m = m->m_next; 1771 slp->ns_rawend = m; 1772 1773 /* 1774 * Now try and parse record(s) out of the raw stream data. 1775 */ 1776 if (error = nfsrv_getstream(slp, waitflag)) { 1777 if (error == EPERM) 1778 slp->ns_flag |= SLP_DISCONN; 1779 else 1780 slp->ns_flag |= SLP_NEEDQ; 1781 } 1782 } else { 1783 do { 1784 auio.uio_resid = 1000000000; 1785 flags = MSG_DONTWAIT; 1786 error = soreceive(so, &nam, &auio, &mp, 1787 (struct mbuf **)0, &flags); 1788 if (mp) { 1789 nfs_realign(mp, 10 * NFSX_UNSIGNED); 1790 if (nam) { 1791 m = nam; 1792 m->m_next = mp; 1793 } else 1794 m = mp; 1795 if (slp->ns_recend) 1796 slp->ns_recend->m_nextpkt = m; 1797 else 1798 slp->ns_rec = m; 1799 slp->ns_recend = m; 1800 m->m_nextpkt = (struct mbuf *)0; 1801 } 1802 if (error) { 1803 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1804 && error != EWOULDBLOCK) { 1805 slp->ns_flag |= SLP_DISCONN; 1806 goto dorecs; 1807 } 1808 } 1809 } while (mp); 1810 } 1811 1812 /* 1813 * Now try and process the request records, non-blocking. 1814 */ 1815 dorecs: 1816 if (waitflag == M_DONTWAIT && 1817 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1818 nfsrv_wakenfsd(slp); 1819 } 1820 1821 /* 1822 * Try and extract an RPC request from the mbuf data list received on a 1823 * stream socket. The "waitflag" argument indicates whether or not it 1824 * can sleep. 1825 */ 1826 nfsrv_getstream(slp, waitflag) 1827 register struct nfssvc_sock *slp; 1828 int waitflag; 1829 { 1830 register struct mbuf *m; 1831 register char *cp1, *cp2; 1832 register int len; 1833 struct mbuf *om, *m2, *recm; 1834 u_long recmark; 1835 1836 if (slp->ns_flag & SLP_GETSTREAM) 1837 panic("nfs getstream"); 1838 slp->ns_flag |= SLP_GETSTREAM; 1839 for (;;) { 1840 if (slp->ns_reclen == 0) { 1841 if (slp->ns_cc < NFSX_UNSIGNED) { 1842 slp->ns_flag &= ~SLP_GETSTREAM; 1843 return (0); 1844 } 1845 m = slp->ns_raw; 1846 if (m->m_len >= NFSX_UNSIGNED) { 1847 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 1848 m->m_data += NFSX_UNSIGNED; 1849 m->m_len -= NFSX_UNSIGNED; 1850 } else { 1851 cp1 = (caddr_t)&recmark; 1852 cp2 = mtod(m, caddr_t); 1853 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1854 while (m->m_len == 0) { 1855 m = m->m_next; 1856 cp2 = mtod(m, caddr_t); 1857 } 1858 *cp1++ = *cp2++; 1859 m->m_data++; 1860 m->m_len--; 1861 } 1862 } 1863 slp->ns_cc -= NFSX_UNSIGNED; 1864 slp->ns_reclen = ntohl(recmark) & ~0x80000000; 1865 if (slp->ns_reclen < NFS_MINPACKET || slp->ns_reclen > NFS_MAXPACKET) { 1866 slp->ns_flag &= ~SLP_GETSTREAM; 1867 return (EPERM); 1868 } 1869 } 1870 1871 /* 1872 * Now get the record part. 1873 */ 1874 if (slp->ns_cc == slp->ns_reclen) { 1875 recm = slp->ns_raw; 1876 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 1877 slp->ns_cc = slp->ns_reclen = 0; 1878 } else if (slp->ns_cc > slp->ns_reclen) { 1879 len = 0; 1880 m = slp->ns_raw; 1881 om = (struct mbuf *)0; 1882 while (len < slp->ns_reclen) { 1883 if ((len + m->m_len) > slp->ns_reclen) { 1884 m2 = m_copym(m, 0, slp->ns_reclen - len, 1885 waitflag); 1886 if (m2) { 1887 if (om) { 1888 om->m_next = m2; 1889 recm = slp->ns_raw; 1890 } else 1891 recm = m2; 1892 m->m_data += slp->ns_reclen - len; 1893 m->m_len -= slp->ns_reclen - len; 1894 len = slp->ns_reclen; 1895 } else { 1896 slp->ns_flag &= ~SLP_GETSTREAM; 1897 return (EWOULDBLOCK); 1898 } 1899 } else if ((len + m->m_len) == slp->ns_reclen) { 1900 om = m; 1901 len += m->m_len; 1902 m = m->m_next; 1903 recm = slp->ns_raw; 1904 om->m_next = (struct mbuf *)0; 1905 } else { 1906 om = m; 1907 len += m->m_len; 1908 m = m->m_next; 1909 } 1910 } 1911 slp->ns_raw = m; 1912 slp->ns_cc -= len; 1913 slp->ns_reclen = 0; 1914 } else { 1915 slp->ns_flag &= ~SLP_GETSTREAM; 1916 return (0); 1917 } 1918 nfs_realign(recm, 10 * NFSX_UNSIGNED); 1919 if (slp->ns_recend) 1920 slp->ns_recend->m_nextpkt = recm; 1921 else 1922 slp->ns_rec = recm; 1923 slp->ns_recend = recm; 1924 } 1925 } 1926 1927 /* 1928 * Parse an RPC header. 1929 */ 1930 nfsrv_dorec(slp, nd) 1931 register struct nfssvc_sock *slp; 1932 register struct nfsd *nd; 1933 { 1934 register struct mbuf *m; 1935 int error; 1936 1937 if ((slp->ns_flag & SLP_VALID) == 0 || 1938 (m = slp->ns_rec) == (struct mbuf *)0) 1939 return (ENOBUFS); 1940 if (slp->ns_rec = m->m_nextpkt) 1941 m->m_nextpkt = (struct mbuf *)0; 1942 else 1943 slp->ns_recend = (struct mbuf *)0; 1944 if (m->m_type == MT_SONAME) { 1945 nd->nd_nam = m; 1946 nd->nd_md = nd->nd_mrep = m->m_next; 1947 m->m_next = (struct mbuf *)0; 1948 } else { 1949 nd->nd_nam = (struct mbuf *)0; 1950 nd->nd_md = nd->nd_mrep = m; 1951 } 1952 nd->nd_dpos = mtod(nd->nd_md, caddr_t); 1953 if (error = nfs_getreq(nd, TRUE)) { 1954 m_freem(nd->nd_nam); 1955 return (error); 1956 } 1957 return (0); 1958 } 1959 1960 /* 1961 * Search for a sleeping nfsd and wake it up. 1962 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 1963 * running nfsds will go look for the work in the nfssvc_sock list. 1964 */ 1965 void 1966 nfsrv_wakenfsd(slp) 1967 struct nfssvc_sock *slp; 1968 { 1969 register struct nfsd *nd = nfsd_head.nd_next; 1970 1971 if ((slp->ns_flag & SLP_VALID) == 0) 1972 return; 1973 while (nd != (struct nfsd *)&nfsd_head) { 1974 if (nd->nd_flag & NFSD_WAITING) { 1975 nd->nd_flag &= ~NFSD_WAITING; 1976 if (nd->nd_slp) 1977 panic("nfsd wakeup"); 1978 slp->ns_sref++; 1979 nd->nd_slp = slp; 1980 wakeup((caddr_t)nd); 1981 return; 1982 } 1983 nd = nd->nd_next; 1984 } 1985 slp->ns_flag |= SLP_DOREC; 1986 nfsd_head.nd_flag |= NFSD_CHECKSLP; 1987 } 1988 #endif /* NFSSERVER */ 1989
Indexes created Sun Sep 21 18:09:52 GMT 2025