1 /* 2 * Copyright (c) 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * from: @(#)nfs_socket.c 8.3 (Berkeley) 1/12/94 37 * $Id: nfs_socket.c,v 1.14 1994/06/08 11:36:57 mycroft Exp $ 38 */ 39 40 /* 41 * Socket operations for use by nfs 42 */ 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/proc.h> 47 #include <sys/mount.h> 48 #include <sys/kernel.h> 49 #include <sys/mbuf.h> 50 #include <sys/vnode.h> 51 #include <sys/domain.h> 52 #include <sys/protosw.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/syslog.h> 56 #include <sys/tprintf.h> 57 58 #include <netinet/in.h> 59 #include <netinet/tcp.h> 60 #include <nfs/rpcv2.h> 61 #include <nfs/nfsv2.h> 62 #include <nfs/nfs.h> 63 #include <nfs/xdr_subs.h> 64 #include <nfs/nfsm_subs.h> 65 #include <nfs/nfsmount.h> 66 #include <nfs/nfsnode.h> 67 #include <nfs/nfsrtt.h> 68 #include <nfs/nqnfs.h> 69 70 #define TRUE 1 71 #define FALSE 0 72 73 /* 74 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 75 * Use the mean and mean deviation of rtt for the appropriate type of rpc 76 * for the frequent rpcs and a default for the others. 77 * The justification for doing "other" this way is that these rpcs 78 * happen so infrequently that timer est. would probably be stale. 79 * Also, since many of these rpcs are 80 * non-idempotent, a conservative timeout is desired. 81 * getattr, lookup - A+2D 82 * read, write - A+4D 83 * other - nm_timeo 84 */ 85 #define NFS_RTO(n, t) \ 86 ((t) == 0 ? (n)->nm_timeo : \ 87 ((t) < 3 ? \ 88 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 89 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 90 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 91 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 92 /* 93 * External data, mostly RPC constants in XDR form 94 */ 95 extern u_long rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, rpc_auth_unix, 96 rpc_msgaccepted, rpc_call, rpc_autherr, rpc_rejectedcred, 97 rpc_auth_kerb; 98 extern u_long nfs_prog, nfs_vers, nqnfs_prog, nqnfs_vers; 99 extern time_t nqnfsstarttime; 100 extern int nonidempotent[NFS_NPROCS]; 101 102 /* 103 * Maps errno values to nfs error numbers. 104 * Use NFSERR_IO as the catch all for ones not specifically defined in 105 * RFC 1094. 106 */ 107 static int nfsrv_errmap[ELAST] = { 108 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, 109 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 110 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, 111 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR, 112 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 113 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, 114 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 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_NAMETOL, NFSERR_IO, NFSERR_IO, 121 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, 122 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 123 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 124 NFSERR_IO, 125 }; 126 127 /* 128 * Defines which timer to use for the procnum. 129 * 0 - default 130 * 1 - getattr 131 * 2 - lookup 132 * 3 - read 133 * 4 - write 134 */ 135 static int proct[NFS_NPROCS] = { 136 0, 1, 0, 0, 2, 3, 3, 0, 4, 0, 0, 0, 0, 0, 0, 0, 3, 0, 3, 0, 0, 0, 0, 137 }; 138 139 /* 140 * There is a congestion window for outstanding rpcs maintained per mount 141 * point. The cwnd size is adjusted in roughly the way that: 142 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 143 * SIGCOMM '88". ACM, August 1988. 144 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 145 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 146 * of rpcs is in progress. 147 * (The sent count and cwnd are scaled for integer arith.) 148 * Variants of "slow start" were tried and were found to be too much of a 149 * performance hit (ave. rtt 3 times larger), 150 * I suspect due to the large rtt that nfs rpcs have. 151 */ 152 #define NFS_CWNDSCALE 256 153 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 154 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 155 int nfs_sbwait(); 156 void nfs_disconnect(), nfs_realign(), nfsrv_wakenfsd(), nfs_sndunlock(); 157 void nfs_rcvunlock(), nqnfs_serverd(), nqnfs_clientlease(); 158 struct mbuf *nfsm_rpchead(); 159 int nfsrtton = 0; 160 struct nfsrtt nfsrtt; 161 struct nfsd nfsd_head; 162 163 struct nfsreq nfsreqh; 164 165 /* 166 * Initialize sockets and congestion for a new NFS connection. 167 * We do not free the sockaddr if error. 168 */ 169 nfs_connect(nmp, rep) 170 register struct nfsmount *nmp; 171 struct nfsreq *rep; 172 { 173 register struct socket *so; 174 int s, error, rcvreserve, sndreserve; 175 struct sockaddr *saddr; 176 struct sockaddr_in *sin; 177 struct mbuf *m; 178 u_short tport; 179 180 nmp->nm_so = (struct socket *)0; 181 saddr = mtod(nmp->nm_nam, struct sockaddr *); 182 if (error = socreate(saddr->sa_family, 183 &nmp->nm_so, nmp->nm_sotype, nmp->nm_soproto)) 184 goto bad; 185 so = nmp->nm_so; 186 nmp->nm_soflags = so->so_proto->pr_flags; 187 188 /* 189 * Some servers require that the client port be a reserved port number. 190 */ 191 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 192 MGET(m, M_WAIT, MT_SONAME); 193 sin = mtod(m, struct sockaddr_in *); 194 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 195 sin->sin_family = AF_INET; 196 sin->sin_addr.s_addr = INADDR_ANY; 197 tport = IPPORT_RESERVED - 1; 198 sin->sin_port = htons(tport); 199 while ((error = sobind(so, m)) == EADDRINUSE && 200 --tport > IPPORT_RESERVED / 2) 201 sin->sin_port = htons(tport); 202 m_freem(m); 203 if (error) 204 goto bad; 205 } 206 207 /* 208 * Protocols that do not require connections may be optionally left 209 * unconnected for servers that reply from a port other than NFS_PORT. 210 */ 211 if (nmp->nm_flag & NFSMNT_NOCONN) { 212 if (nmp->nm_soflags & PR_CONNREQUIRED) { 213 error = ENOTCONN; 214 goto bad; 215 } 216 } else { 217 if (error = soconnect(so, nmp->nm_nam)) 218 goto bad; 219 220 /* 221 * Wait for the connection to complete. Cribbed from the 222 * connect system call but with the wait timing out so 223 * that interruptible mounts don't hang here for a long time. 224 */ 225 s = splnet(); 226 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 227 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 228 "nfscon", 2 * hz); 229 if ((so->so_state & SS_ISCONNECTING) && 230 so->so_error == 0 && rep && 231 (error = nfs_sigintr(nmp, rep, rep->r_procp))) { 232 so->so_state &= ~SS_ISCONNECTING; 233 splx(s); 234 goto bad; 235 } 236 } 237 if (so->so_error) { 238 error = so->so_error; 239 so->so_error = 0; 240 splx(s); 241 goto bad; 242 } 243 splx(s); 244 } 245 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 246 so->so_rcv.sb_timeo = (5 * hz); 247 so->so_snd.sb_timeo = (5 * hz); 248 } else { 249 so->so_rcv.sb_timeo = 0; 250 so->so_snd.sb_timeo = 0; 251 } 252 if (nmp->nm_sotype == SOCK_DGRAM) { 253 sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR; 254 rcvreserve = nmp->nm_rsize + NFS_MAXPKTHDR; 255 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 256 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 257 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR) * 2; 258 } else { 259 if (nmp->nm_sotype != SOCK_STREAM) 260 panic("nfscon sotype"); 261 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 262 MGET(m, M_WAIT, MT_SOOPTS); 263 *mtod(m, int *) = 1; 264 m->m_len = sizeof(int); 265 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 266 } 267 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 268 MGET(m, M_WAIT, MT_SOOPTS); 269 *mtod(m, int *) = 1; 270 m->m_len = sizeof(int); 271 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 272 } 273 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + sizeof (u_long)) 274 * 2; 275 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + sizeof (u_long)) 276 * 2; 277 } 278 if (error = soreserve(so, sndreserve, rcvreserve)) 279 goto bad; 280 so->so_rcv.sb_flags |= SB_NOINTR; 281 so->so_snd.sb_flags |= SB_NOINTR; 282 283 /* Initialize other non-zero congestion variables */ 284 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 285 nmp->nm_srtt[4] = (NFS_TIMEO << 3); 286 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 287 nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0; 288 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 289 nmp->nm_sent = 0; 290 nmp->nm_timeouts = 0; 291 return (0); 292 293 bad: 294 nfs_disconnect(nmp); 295 return (error); 296 } 297 298 /* 299 * Reconnect routine: 300 * Called when a connection is broken on a reliable protocol. 301 * - clean up the old socket 302 * - nfs_connect() again 303 * - set R_MUSTRESEND for all outstanding requests on mount point 304 * If this fails the mount point is DEAD! 305 * nb: Must be called with the nfs_sndlock() set on the mount point. 306 */ 307 nfs_reconnect(rep) 308 register struct nfsreq *rep; 309 { 310 register struct nfsreq *rp; 311 register struct nfsmount *nmp = rep->r_nmp; 312 int error; 313 314 nfs_disconnect(nmp); 315 while (error = nfs_connect(nmp, rep)) { 316 if (error == EINTR || error == ERESTART) 317 return (EINTR); 318 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); 319 } 320 321 /* 322 * Loop through outstanding request list and fix up all requests 323 * on old socket. 324 */ 325 rp = nfsreqh.r_next; 326 while (rp != &nfsreqh) { 327 if (rp->r_nmp == nmp) 328 rp->r_flags |= R_MUSTRESEND; 329 rp = rp->r_next; 330 } 331 return (0); 332 } 333 334 /* 335 * NFS disconnect. Clean up and unlink. 336 */ 337 void 338 nfs_disconnect(nmp) 339 register struct nfsmount *nmp; 340 { 341 register struct socket *so; 342 343 if (nmp->nm_so) { 344 so = nmp->nm_so; 345 nmp->nm_so = (struct socket *)0; 346 soshutdown(so, 2); 347 soclose(so); 348 } 349 } 350 351 /* 352 * This is the nfs send routine. For connection based socket types, it 353 * must be called with an nfs_sndlock() on the socket. 354 * "rep == NULL" indicates that it has been called from a server. 355 * For the client side: 356 * - return EINTR if the RPC is terminated, 0 otherwise 357 * - set R_MUSTRESEND if the send fails for any reason 358 * - do any cleanup required by recoverable socket errors (???) 359 * For the server side: 360 * - return EINTR or ERESTART if interrupted by a signal 361 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 362 * - do any cleanup required by recoverable socket errors (???) 363 */ 364 nfs_send(so, nam, top, rep) 365 register struct socket *so; 366 struct mbuf *nam; 367 register struct mbuf *top; 368 struct nfsreq *rep; 369 { 370 struct mbuf *sendnam; 371 int error, soflags, flags; 372 373 if (rep) { 374 if (rep->r_flags & R_SOFTTERM) { 375 m_freem(top); 376 return (EINTR); 377 } 378 if ((so = rep->r_nmp->nm_so) == NULL) { 379 rep->r_flags |= R_MUSTRESEND; 380 m_freem(top); 381 return (0); 382 } 383 rep->r_flags &= ~R_MUSTRESEND; 384 soflags = rep->r_nmp->nm_soflags; 385 } else 386 soflags = so->so_proto->pr_flags; 387 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 388 sendnam = (struct mbuf *)0; 389 else 390 sendnam = nam; 391 if (so->so_type == SOCK_SEQPACKET) 392 flags = MSG_EOR; 393 else 394 flags = 0; 395 396 error = sosend(so, sendnam, (struct uio *)0, top, 397 (struct mbuf *)0, flags); 398 if (error) { 399 if (rep) { 400 log(LOG_INFO, "nfs send error %d for server %s\n",error, 401 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 402 /* 403 * Deal with errors for the client side. 404 */ 405 if (rep->r_flags & R_SOFTTERM) 406 error = EINTR; 407 else 408 rep->r_flags |= R_MUSTRESEND; 409 } else 410 log(LOG_INFO, "nfsd send error %d\n", error); 411 412 /* 413 * Handle any recoverable (soft) socket errors here. (???) 414 */ 415 if (error != EINTR && error != ERESTART && 416 error != EWOULDBLOCK && error != EPIPE) 417 error = 0; 418 } 419 return (error); 420 } 421 422 #ifdef NFSCLIENT 423 /* 424 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 425 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 426 * Mark and consolidate the data into a new mbuf list. 427 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 428 * small mbufs. 429 * For SOCK_STREAM we must be very careful to read an entire record once 430 * we have read any of it, even if the system call has been interrupted. 431 */ 432 nfs_receive(rep, aname, mp) 433 register struct nfsreq *rep; 434 struct mbuf **aname; 435 struct mbuf **mp; 436 { 437 register struct socket *so; 438 struct uio auio; 439 struct iovec aio; 440 register struct mbuf *m; 441 struct mbuf *control; 442 u_long len; 443 struct mbuf **getnam; 444 int error, sotype, rcvflg; 445 struct proc *p = curproc; /* XXX */ 446 447 /* 448 * Set up arguments for soreceive() 449 */ 450 *mp = (struct mbuf *)0; 451 *aname = (struct mbuf *)0; 452 sotype = rep->r_nmp->nm_sotype; 453 454 /* 455 * For reliable protocols, lock against other senders/receivers 456 * in case a reconnect is necessary. 457 * For SOCK_STREAM, first get the Record Mark to find out how much 458 * more there is to get. 459 * We must lock the socket against other receivers 460 * until we have an entire rpc request/reply. 461 */ 462 if (sotype != SOCK_DGRAM) { 463 if (error = nfs_sndlock(&rep->r_nmp->nm_flag, rep)) 464 return (error); 465 tryagain: 466 /* 467 * Check for fatal errors and resending request. 468 */ 469 /* 470 * Ugh: If a reconnect attempt just happened, nm_so 471 * would have changed. NULL indicates a failed 472 * attempt that has essentially shut down this 473 * mount point. 474 */ 475 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 476 nfs_sndunlock(&rep->r_nmp->nm_flag); 477 return (EINTR); 478 } 479 if ((so = rep->r_nmp->nm_so) == NULL) { 480 if (error = nfs_reconnect(rep)) { 481 nfs_sndunlock(&rep->r_nmp->nm_flag); 482 return (error); 483 } 484 goto tryagain; 485 } 486 while (rep->r_flags & R_MUSTRESEND) { 487 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 488 nfsstats.rpcretries++; 489 if (error = nfs_send(so, rep->r_nmp->nm_nam, m, rep)) { 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_long); 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_long); 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_long) - auio.uio_resid, 522 sizeof(u_long), 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 nfs_reply(myrep) 639 struct nfsreq *myrep; 640 { 641 register struct nfsreq *rep; 642 register struct nfsmount *nmp = myrep->r_nmp; 643 register long t1; 644 struct mbuf *mrep, *nam, *md; 645 u_long rxid, *tl; 646 caddr_t dpos, cp2; 647 int error; 648 649 /* 650 * Loop around until we get our own reply 651 */ 652 for (;;) { 653 /* 654 * Lock against other receivers so that I don't get stuck in 655 * sbwait() after someone else has received my reply for me. 656 * Also necessary for connection based protocols to avoid 657 * race conditions during a reconnect. 658 */ 659 if (error = nfs_rcvlock(myrep)) 660 return (error); 661 /* Already received, bye bye */ 662 if (myrep->r_mrep != NULL) { 663 nfs_rcvunlock(&nmp->nm_flag); 664 return (0); 665 } 666 /* 667 * Get the next Rpc reply off the socket 668 */ 669 error = nfs_receive(myrep, &nam, &mrep); 670 nfs_rcvunlock(&nmp->nm_flag); 671 if (error) { 672 673 /* 674 * Ignore routing errors on connectionless protocols?? 675 */ 676 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 677 nmp->nm_so->so_error = 0; 678 if (myrep->r_flags & R_GETONEREP) 679 return (0); 680 continue; 681 } 682 return (error); 683 } 684 if (nam) 685 m_freem(nam); 686 687 /* 688 * Get the xid and check that it is an rpc reply 689 */ 690 md = mrep; 691 dpos = mtod(md, caddr_t); 692 nfsm_dissect(tl, u_long *, 2*NFSX_UNSIGNED); 693 rxid = *tl++; 694 if (*tl != rpc_reply) { 695 if (nmp->nm_flag & NFSMNT_NQNFS) { 696 if (nqnfs_callback(nmp, mrep, md, dpos)) 697 nfsstats.rpcinvalid++; 698 } else { 699 nfsstats.rpcinvalid++; 700 m_freem(mrep); 701 } 702 nfsmout: 703 if (myrep->r_flags & R_GETONEREP) 704 return (0); 705 continue; 706 } 707 708 /* 709 * Loop through the request list to match up the reply 710 * Iff no match, just drop the datagram 711 */ 712 rep = nfsreqh.r_next; 713 while (rep != &nfsreqh) { 714 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 715 /* Found it.. */ 716 rep->r_mrep = mrep; 717 rep->r_md = md; 718 rep->r_dpos = dpos; 719 if (nfsrtton) { 720 struct rttl *rt; 721 722 rt = &nfsrtt.rttl[nfsrtt.pos]; 723 rt->proc = rep->r_procnum; 724 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 725 rt->sent = nmp->nm_sent; 726 rt->cwnd = nmp->nm_cwnd; 727 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 728 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 729 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 730 rt->tstamp = time; 731 if (rep->r_flags & R_TIMING) 732 rt->rtt = rep->r_rtt; 733 else 734 rt->rtt = 1000000; 735 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 736 } 737 /* 738 * Update congestion window. 739 * Do the additive increase of 740 * one rpc/rtt. 741 */ 742 if (nmp->nm_cwnd <= nmp->nm_sent) { 743 nmp->nm_cwnd += 744 (NFS_CWNDSCALE * NFS_CWNDSCALE + 745 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 746 if (nmp->nm_cwnd > NFS_MAXCWND) 747 nmp->nm_cwnd = NFS_MAXCWND; 748 } 749 rep->r_flags &= ~R_SENT; 750 nmp->nm_sent -= NFS_CWNDSCALE; 751 /* 752 * Update rtt using a gain of 0.125 on the mean 753 * and a gain of 0.25 on the deviation. 754 */ 755 if (rep->r_flags & R_TIMING) { 756 /* 757 * Since the timer resolution of 758 * NFS_HZ is so course, it can often 759 * result in r_rtt == 0. Since 760 * r_rtt == N means that the actual 761 * rtt is between N+dt and N+2-dt ticks, 762 * add 1. 763 */ 764 t1 = rep->r_rtt + 1; 765 t1 -= (NFS_SRTT(rep) >> 3); 766 NFS_SRTT(rep) += t1; 767 if (t1 < 0) 768 t1 = -t1; 769 t1 -= (NFS_SDRTT(rep) >> 2); 770 NFS_SDRTT(rep) += t1; 771 } 772 nmp->nm_timeouts = 0; 773 break; 774 } 775 rep = rep->r_next; 776 } 777 /* 778 * If not matched to a request, drop it. 779 * If it's mine, get out. 780 */ 781 if (rep == &nfsreqh) { 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 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 805 struct vnode *vp; 806 struct mbuf *mrest; 807 int procnum; 808 struct proc *procp; 809 struct ucred *cred; 810 struct mbuf **mrp; 811 struct mbuf **mdp; 812 caddr_t *dposp; 813 { 814 register struct mbuf *m, *mrep; 815 register struct nfsreq *rep; 816 register u_long *tl; 817 register int i; 818 struct nfsmount *nmp; 819 struct mbuf *md, *mheadend; 820 struct nfsreq *reph; 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_long 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 if (cred->cr_ngroups < 1) 865 panic("nfsreq nogrps"); 866 auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ? 867 nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) + 868 5 * NFSX_UNSIGNED; 869 } 870 m = nfsm_rpchead(cred, (nmp->nm_flag & NFSMNT_NQNFS), procnum, 871 auth_type, auth_len, auth_str, mrest, mrest_len, &mheadend, &xid); 872 if (auth_str) 873 free(auth_str, M_TEMP); 874 875 /* 876 * For stream protocols, insert a Sun RPC Record Mark. 877 */ 878 if (nmp->nm_sotype == SOCK_STREAM) { 879 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 880 *mtod(m, u_long *) = htonl(0x80000000 | 881 (m->m_pkthdr.len - NFSX_UNSIGNED)); 882 } 883 rep->r_mreq = m; 884 rep->r_xid = xid; 885 tryagain: 886 if (nmp->nm_flag & NFSMNT_SOFT) 887 rep->r_retry = nmp->nm_retry; 888 else 889 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 890 rep->r_rtt = rep->r_rexmit = 0; 891 if (proct[procnum] > 0) 892 rep->r_flags = R_TIMING; 893 else 894 rep->r_flags = 0; 895 rep->r_mrep = NULL; 896 897 /* 898 * Do the client side RPC. 899 */ 900 nfsstats.rpcrequests++; 901 /* 902 * Chain request into list of outstanding requests. Be sure 903 * to put it LAST so timer finds oldest requests first. 904 */ 905 s = splsoftclock(); 906 reph = &nfsreqh; 907 reph->r_prev->r_next = rep; 908 rep->r_prev = reph->r_prev; 909 reph->r_prev = rep; 910 rep->r_next = reph; 911 912 /* Get send time for nqnfs */ 913 reqtime = time.tv_sec; 914 915 /* 916 * If backing off another request or avoiding congestion, don't 917 * send this one now but let timer do it. If not timing a request, 918 * do it now. 919 */ 920 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 921 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 922 nmp->nm_sent < nmp->nm_cwnd)) { 923 splx(s); 924 if (nmp->nm_soflags & PR_CONNREQUIRED) 925 error = nfs_sndlock(&nmp->nm_flag, rep); 926 if (!error) { 927 m = m_copym(m, 0, M_COPYALL, M_WAIT); 928 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 929 if (nmp->nm_soflags & PR_CONNREQUIRED) 930 nfs_sndunlock(&nmp->nm_flag); 931 } 932 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 933 nmp->nm_sent += NFS_CWNDSCALE; 934 rep->r_flags |= R_SENT; 935 } 936 } else { 937 splx(s); 938 rep->r_rtt = -1; 939 } 940 941 /* 942 * Wait for the reply from our send or the timer's. 943 */ 944 if (!error || error == EPIPE) 945 error = nfs_reply(rep); 946 947 /* 948 * RPC done, unlink the request. 949 */ 950 s = splsoftclock(); 951 rep->r_prev->r_next = rep->r_next; 952 rep->r_next->r_prev = rep->r_prev; 953 splx(s); 954 955 /* 956 * Decrement the outstanding request count. 957 */ 958 if (rep->r_flags & R_SENT) { 959 rep->r_flags &= ~R_SENT; /* paranoia */ 960 nmp->nm_sent -= NFS_CWNDSCALE; 961 } 962 963 /* 964 * If there was a successful reply and a tprintf msg. 965 * tprintf a response. 966 */ 967 if (!error && (rep->r_flags & R_TPRINTFMSG)) 968 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 969 "is alive again"); 970 mrep = rep->r_mrep; 971 md = rep->r_md; 972 dpos = rep->r_dpos; 973 if (error) { 974 m_freem(rep->r_mreq); 975 free((caddr_t)rep, M_NFSREQ); 976 return (error); 977 } 978 979 /* 980 * break down the rpc header and check if ok 981 */ 982 nfsm_dissect(tl, u_long *, 3*NFSX_UNSIGNED); 983 if (*tl++ == rpc_msgdenied) { 984 if (*tl == rpc_mismatch) 985 error = EOPNOTSUPP; 986 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 987 if (*tl == rpc_rejectedcred && failed_auth == 0) { 988 failed_auth++; 989 mheadend->m_next = (struct mbuf *)0; 990 m_freem(mrep); 991 m_freem(rep->r_mreq); 992 goto kerbauth; 993 } else 994 error = EAUTH; 995 } else 996 error = EACCES; 997 m_freem(mrep); 998 m_freem(rep->r_mreq); 999 free((caddr_t)rep, M_NFSREQ); 1000 return (error); 1001 } 1002 1003 /* 1004 * skip over the auth_verf, someday we may want to cache auth_short's 1005 * for nfs_reqhead(), but for now just dump it 1006 */ 1007 if (*++tl != 0) { 1008 i = nfsm_rndup(fxdr_unsigned(long, *tl)); 1009 nfsm_adv(i); 1010 } 1011 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1012 /* 0 == ok */ 1013 if (*tl == 0) { 1014 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1015 if (*tl != 0) { 1016 error = fxdr_unsigned(int, *tl); 1017 m_freem(mrep); 1018 if ((nmp->nm_flag & NFSMNT_NQNFS) && 1019 error == NQNFS_TRYLATER) { 1020 error = 0; 1021 waituntil = time.tv_sec + trylater_delay; 1022 while (time.tv_sec < waituntil) 1023 (void) tsleep((caddr_t)&lbolt, 1024 PSOCK, "nqnfstry", 0); 1025 trylater_delay *= nfs_backoff[trylater_cnt]; 1026 if (trylater_cnt < 7) 1027 trylater_cnt++; 1028 goto tryagain; 1029 } 1030 1031 /* 1032 * If the File Handle was stale, invalidate the 1033 * lookup cache, just in case. 1034 */ 1035 if (error == ESTALE) 1036 cache_purge(vp); 1037 m_freem(rep->r_mreq); 1038 free((caddr_t)rep, M_NFSREQ); 1039 return (error); 1040 } 1041 1042 /* 1043 * For nqnfs, get any lease in reply 1044 */ 1045 if (nmp->nm_flag & NFSMNT_NQNFS) { 1046 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1047 if (*tl) { 1048 np = VTONFS(vp); 1049 nqlflag = fxdr_unsigned(int, *tl); 1050 nfsm_dissect(tl, u_long *, 4*NFSX_UNSIGNED); 1051 cachable = fxdr_unsigned(int, *tl++); 1052 reqtime += fxdr_unsigned(int, *tl++); 1053 if (reqtime > time.tv_sec) { 1054 fxdr_hyper(tl, &frev); 1055 nqnfs_clientlease(nmp, np, nqlflag, 1056 cachable, reqtime, frev); 1057 } 1058 } 1059 } 1060 *mrp = mrep; 1061 *mdp = md; 1062 *dposp = dpos; 1063 m_freem(rep->r_mreq); 1064 FREE((caddr_t)rep, M_NFSREQ); 1065 return (0); 1066 } 1067 m_freem(mrep); 1068 m_freem(rep->r_mreq); 1069 free((caddr_t)rep, M_NFSREQ); 1070 error = EPROTONOSUPPORT; 1071 nfsmout: 1072 return (error); 1073 } 1074 #endif /* NFSCLIENT */ 1075 1076 /* 1077 * Generate the rpc reply header 1078 * siz arg. is used to decide if adding a cluster is worthwhile 1079 */ 1080 nfs_rephead(siz, nd, err, cache, frev, mrq, mbp, bposp) 1081 int siz; 1082 struct nfsd *nd; 1083 int err; 1084 int cache; 1085 u_quad_t *frev; 1086 struct mbuf **mrq; 1087 struct mbuf **mbp; 1088 caddr_t *bposp; 1089 { 1090 register u_long *tl; 1091 register struct mbuf *mreq; 1092 caddr_t bpos; 1093 struct mbuf *mb, *mb2; 1094 1095 MGETHDR(mreq, M_WAIT, MT_DATA); 1096 mb = mreq; 1097 /* 1098 * If this is a big reply, use a cluster else 1099 * try and leave leading space for the lower level headers. 1100 */ 1101 siz += RPC_REPLYSIZ; 1102 if (siz >= MINCLSIZE) { 1103 MCLGET(mreq, M_WAIT); 1104 } else 1105 mreq->m_data += max_hdr; 1106 tl = mtod(mreq, u_long *); 1107 mreq->m_len = 6*NFSX_UNSIGNED; 1108 bpos = ((caddr_t)tl)+mreq->m_len; 1109 *tl++ = txdr_unsigned(nd->nd_retxid); 1110 *tl++ = rpc_reply; 1111 if (err == ERPCMISMATCH || err == NQNFS_AUTHERR) { 1112 *tl++ = rpc_msgdenied; 1113 if (err == NQNFS_AUTHERR) { 1114 *tl++ = rpc_autherr; 1115 *tl = rpc_rejectedcred; 1116 mreq->m_len -= NFSX_UNSIGNED; 1117 bpos -= NFSX_UNSIGNED; 1118 } else { 1119 *tl++ = rpc_mismatch; 1120 *tl++ = txdr_unsigned(2); 1121 *tl = txdr_unsigned(2); 1122 } 1123 } else { 1124 *tl++ = rpc_msgaccepted; 1125 *tl++ = 0; 1126 *tl++ = 0; 1127 switch (err) { 1128 case EPROGUNAVAIL: 1129 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1130 break; 1131 case EPROGMISMATCH: 1132 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1133 nfsm_build(tl, u_long *, 2*NFSX_UNSIGNED); 1134 *tl++ = txdr_unsigned(2); 1135 *tl = txdr_unsigned(2); /* someday 3 */ 1136 break; 1137 case EPROCUNAVAIL: 1138 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1139 break; 1140 default: 1141 *tl = 0; 1142 if (err != VNOVAL) { 1143 nfsm_build(tl, u_long *, NFSX_UNSIGNED); 1144 if (err) 1145 *tl = txdr_unsigned(nfsrv_errmap[err - 1]); 1146 else 1147 *tl = 0; 1148 } 1149 break; 1150 }; 1151 } 1152 1153 /* 1154 * For nqnfs, piggyback lease as requested. 1155 */ 1156 if (nd->nd_nqlflag != NQL_NOVAL && err == 0) { 1157 if (nd->nd_nqlflag) { 1158 nfsm_build(tl, u_long *, 5*NFSX_UNSIGNED); 1159 *tl++ = txdr_unsigned(nd->nd_nqlflag); 1160 *tl++ = txdr_unsigned(cache); 1161 *tl++ = txdr_unsigned(nd->nd_duration); 1162 txdr_hyper(frev, tl); 1163 } else { 1164 if (nd->nd_nqlflag != 0) 1165 panic("nqreph"); 1166 nfsm_build(tl, u_long *, NFSX_UNSIGNED); 1167 *tl = 0; 1168 } 1169 } 1170 *mrq = mreq; 1171 *mbp = mb; 1172 *bposp = bpos; 1173 if (err != 0 && err != VNOVAL) 1174 nfsstats.srvrpc_errs++; 1175 return (0); 1176 } 1177 1178 /* 1179 * Nfs timer routine 1180 * Scan the nfsreq list and retranmit any requests that have timed out 1181 * To avoid retransmission attempts on STREAM sockets (in the future) make 1182 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1183 */ 1184 void 1185 nfs_timer(arg) 1186 void *arg; 1187 { 1188 register struct nfsreq *rep; 1189 register struct mbuf *m; 1190 register struct socket *so; 1191 register struct nfsmount *nmp; 1192 register int timeo; 1193 static long lasttime = 0; 1194 int s, error; 1195 1196 s = splnet(); 1197 for (rep = nfsreqh.r_next; rep != &nfsreqh; rep = rep->r_next) { 1198 nmp = rep->r_nmp; 1199 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1200 continue; 1201 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1202 rep->r_flags |= R_SOFTTERM; 1203 continue; 1204 } 1205 if (rep->r_rtt >= 0) { 1206 rep->r_rtt++; 1207 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1208 timeo = nmp->nm_timeo; 1209 else 1210 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1211 if (nmp->nm_timeouts > 0) 1212 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1213 if (rep->r_rtt <= timeo) 1214 continue; 1215 if (nmp->nm_timeouts < 8) 1216 nmp->nm_timeouts++; 1217 } 1218 /* 1219 * Check for server not responding 1220 */ 1221 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1222 rep->r_rexmit > nmp->nm_deadthresh) { 1223 nfs_msg(rep->r_procp, 1224 nmp->nm_mountp->mnt_stat.f_mntfromname, 1225 "not responding"); 1226 rep->r_flags |= R_TPRINTFMSG; 1227 } 1228 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1229 nfsstats.rpctimeouts++; 1230 rep->r_flags |= R_SOFTTERM; 1231 continue; 1232 } 1233 if (nmp->nm_sotype != SOCK_DGRAM) { 1234 if (++rep->r_rexmit > NFS_MAXREXMIT) 1235 rep->r_rexmit = NFS_MAXREXMIT; 1236 continue; 1237 } 1238 if ((so = nmp->nm_so) == NULL) 1239 continue; 1240 1241 /* 1242 * If there is enough space and the window allows.. 1243 * Resend it 1244 * Set r_rtt to -1 in case we fail to send it now. 1245 */ 1246 rep->r_rtt = -1; 1247 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1248 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1249 (rep->r_flags & R_SENT) || 1250 nmp->nm_sent < nmp->nm_cwnd) && 1251 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1252 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1253 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1254 (struct mbuf *)0, (struct mbuf *)0); 1255 else 1256 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1257 nmp->nm_nam, (struct mbuf *)0); 1258 if (error) { 1259 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) 1260 so->so_error = 0; 1261 } else { 1262 /* 1263 * Iff first send, start timing 1264 * else turn timing off, backoff timer 1265 * and divide congestion window by 2. 1266 */ 1267 if (rep->r_flags & R_SENT) { 1268 rep->r_flags &= ~R_TIMING; 1269 if (++rep->r_rexmit > NFS_MAXREXMIT) 1270 rep->r_rexmit = NFS_MAXREXMIT; 1271 nmp->nm_cwnd >>= 1; 1272 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1273 nmp->nm_cwnd = NFS_CWNDSCALE; 1274 nfsstats.rpcretries++; 1275 } else { 1276 rep->r_flags |= R_SENT; 1277 nmp->nm_sent += NFS_CWNDSCALE; 1278 } 1279 rep->r_rtt = 0; 1280 } 1281 } 1282 } 1283 1284 #ifdef NFSSERVER 1285 /* 1286 * Call the nqnfs server timer once a second to handle leases. 1287 */ 1288 if (lasttime != time.tv_sec) { 1289 lasttime = time.tv_sec; 1290 nqnfs_serverd(); 1291 } 1292 #endif /* NFSSERVER */ 1293 splx(s); 1294 timeout(nfs_timer, (void *)0, hz / NFS_HZ); 1295 } 1296 1297 /* 1298 * Test for a termination condition pending on the process. 1299 * This is used for NFSMNT_INT mounts. 1300 */ 1301 nfs_sigintr(nmp, rep, p) 1302 struct nfsmount *nmp; 1303 struct nfsreq *rep; 1304 register struct proc *p; 1305 { 1306 1307 if (rep && (rep->r_flags & R_SOFTTERM)) 1308 return (EINTR); 1309 if (!(nmp->nm_flag & NFSMNT_INT)) 1310 return (0); 1311 if (p && p->p_siglist && 1312 (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) & 1313 NFSINT_SIGMASK)) 1314 return (EINTR); 1315 return (0); 1316 } 1317 1318 /* 1319 * Lock a socket against others. 1320 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1321 * and also to avoid race conditions between the processes with nfs requests 1322 * in progress when a reconnect is necessary. 1323 */ 1324 nfs_sndlock(flagp, rep) 1325 register int *flagp; 1326 struct nfsreq *rep; 1327 { 1328 struct proc *p; 1329 int slpflag = 0, slptimeo = 0; 1330 1331 if (rep) { 1332 p = rep->r_procp; 1333 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1334 slpflag = PCATCH; 1335 } else 1336 p = (struct proc *)0; 1337 while (*flagp & NFSMNT_SNDLOCK) { 1338 if (nfs_sigintr(rep->r_nmp, rep, p)) 1339 return (EINTR); 1340 *flagp |= NFSMNT_WANTSND; 1341 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1342 slptimeo); 1343 if (slpflag == PCATCH) { 1344 slpflag = 0; 1345 slptimeo = 2 * hz; 1346 } 1347 } 1348 *flagp |= NFSMNT_SNDLOCK; 1349 return (0); 1350 } 1351 1352 /* 1353 * Unlock the stream socket for others. 1354 */ 1355 void 1356 nfs_sndunlock(flagp) 1357 register int *flagp; 1358 { 1359 1360 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1361 panic("nfs sndunlock"); 1362 *flagp &= ~NFSMNT_SNDLOCK; 1363 if (*flagp & NFSMNT_WANTSND) { 1364 *flagp &= ~NFSMNT_WANTSND; 1365 wakeup((caddr_t)flagp); 1366 } 1367 } 1368 1369 nfs_rcvlock(rep) 1370 register struct nfsreq *rep; 1371 { 1372 register int *flagp = &rep->r_nmp->nm_flag; 1373 int slpflag, slptimeo = 0; 1374 1375 if (*flagp & NFSMNT_INT) 1376 slpflag = PCATCH; 1377 else 1378 slpflag = 0; 1379 while (*flagp & NFSMNT_RCVLOCK) { 1380 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1381 return (EINTR); 1382 *flagp |= NFSMNT_WANTRCV; 1383 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1384 slptimeo); 1385 if (slpflag == PCATCH) { 1386 slpflag = 0; 1387 slptimeo = 2 * hz; 1388 } 1389 } 1390 *flagp |= NFSMNT_RCVLOCK; 1391 return (0); 1392 } 1393 1394 /* 1395 * Unlock the stream socket for others. 1396 */ 1397 void 1398 nfs_rcvunlock(flagp) 1399 register int *flagp; 1400 { 1401 1402 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1403 panic("nfs rcvunlock"); 1404 *flagp &= ~NFSMNT_RCVLOCK; 1405 if (*flagp & NFSMNT_WANTRCV) { 1406 *flagp &= ~NFSMNT_WANTRCV; 1407 wakeup((caddr_t)flagp); 1408 } 1409 } 1410 1411 /* 1412 * Check for badly aligned mbuf data areas and 1413 * realign data in an mbuf list by copying the data areas up, as required. 1414 */ 1415 void 1416 nfs_realign(m, hsiz) 1417 register struct mbuf *m; 1418 int hsiz; 1419 { 1420 register struct mbuf *m2; 1421 register int siz, mlen, olen; 1422 register caddr_t tcp, fcp; 1423 struct mbuf *mnew; 1424 1425 while (m) { 1426 /* 1427 * This never happens for UDP, rarely happens for TCP 1428 * but frequently happens for iso transport. 1429 */ 1430 if ((m->m_len & 0x3) || (mtod(m, int) & 0x3)) { 1431 olen = m->m_len; 1432 fcp = mtod(m, caddr_t); 1433 if ((int)fcp & 0x3) { 1434 m->m_flags &= ~M_PKTHDR; 1435 if (m->m_flags & M_EXT) 1436 m->m_data = m->m_ext.ext_buf + 1437 ((m->m_ext.ext_size - olen) & ~0x3); 1438 else 1439 m->m_data = m->m_dat; 1440 } 1441 m->m_len = 0; 1442 tcp = mtod(m, caddr_t); 1443 mnew = m; 1444 m2 = m->m_next; 1445 1446 /* 1447 * If possible, only put the first invariant part 1448 * of the RPC header in the first mbuf. 1449 */ 1450 mlen = M_TRAILINGSPACE(m); 1451 if (olen <= hsiz && mlen > hsiz) 1452 mlen = hsiz; 1453 1454 /* 1455 * Loop through the mbuf list consolidating data. 1456 */ 1457 while (m) { 1458 while (olen > 0) { 1459 if (mlen == 0) { 1460 m2->m_flags &= ~M_PKTHDR; 1461 if (m2->m_flags & M_EXT) 1462 m2->m_data = m2->m_ext.ext_buf; 1463 else 1464 m2->m_data = m2->m_dat; 1465 m2->m_len = 0; 1466 mlen = M_TRAILINGSPACE(m2); 1467 tcp = mtod(m2, caddr_t); 1468 mnew = m2; 1469 m2 = m2->m_next; 1470 } 1471 siz = min(mlen, olen); 1472 if (tcp != fcp) 1473 bcopy(fcp, tcp, siz); 1474 mnew->m_len += siz; 1475 mlen -= siz; 1476 olen -= siz; 1477 tcp += siz; 1478 fcp += siz; 1479 } 1480 m = m->m_next; 1481 if (m) { 1482 olen = m->m_len; 1483 fcp = mtod(m, caddr_t); 1484 } 1485 } 1486 1487 /* 1488 * Finally, set m_len == 0 for any trailing mbufs that have 1489 * been copied out of. 1490 */ 1491 while (m2) { 1492 m2->m_len = 0; 1493 m2 = m2->m_next; 1494 } 1495 return; 1496 } 1497 m = m->m_next; 1498 } 1499 } 1500 1501 /* 1502 * Parse an RPC request 1503 * - verify it 1504 * - fill in the cred struct. 1505 */ 1506 nfs_getreq(nd, has_header) 1507 register struct nfsd *nd; 1508 int has_header; 1509 { 1510 register int len, i; 1511 register u_long *tl; 1512 register long t1; 1513 struct uio uio; 1514 struct iovec iov; 1515 caddr_t dpos, cp2; 1516 u_long 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_long *, 10*NFSX_UNSIGNED); 1525 nd->nd_retxid = fxdr_unsigned(u_long, *tl++); 1526 if (*tl++ != rpc_call) { 1527 m_freem(mrep); 1528 return (EBADRPC); 1529 } 1530 } else { 1531 nfsm_dissect(tl, u_long *, 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_long, *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_long *, 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_long *, (len + 2)*NFSX_UNSIGNED); 1592 for (i = 1; 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 + 1); 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_long *, 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_long *, NFSX_UNSIGNED); 1637 nd->nd_nqlflag = fxdr_unsigned(int, *tl); 1638 if (nd->nd_nqlflag) { 1639 nfsm_dissect(tl, u_long *, 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 nfs_msg(p, server, msg) 1655 struct proc *p; 1656 char *server, *msg; 1657 { 1658 tpr_t tpr; 1659 1660 if (p) 1661 tpr = tprintf_open(p); 1662 else 1663 tpr = NULL; 1664 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1665 tprintf_close(tpr); 1666 } 1667 1668 #ifdef NFSSERVER 1669 int nfsrv_null(), 1670 nfsrv_getattr(), 1671 nfsrv_setattr(), 1672 nfsrv_lookup(), 1673 nfsrv_readlink(), 1674 nfsrv_read(), 1675 nfsrv_write(), 1676 nfsrv_create(), 1677 nfsrv_remove(), 1678 nfsrv_rename(), 1679 nfsrv_link(), 1680 nfsrv_symlink(), 1681 nfsrv_mkdir(), 1682 nfsrv_rmdir(), 1683 nfsrv_readdir(), 1684 nfsrv_statfs(), 1685 nfsrv_noop(), 1686 nqnfsrv_readdirlook(), 1687 nqnfsrv_getlease(), 1688 nqnfsrv_vacated(), 1689 nqnfsrv_access(); 1690 1691 int (*nfsrv_procs[NFS_NPROCS])() = { 1692 nfsrv_null, 1693 nfsrv_getattr, 1694 nfsrv_setattr, 1695 nfsrv_noop, 1696 nfsrv_lookup, 1697 nfsrv_readlink, 1698 nfsrv_read, 1699 nfsrv_noop, 1700 nfsrv_write, 1701 nfsrv_create, 1702 nfsrv_remove, 1703 nfsrv_rename, 1704 nfsrv_link, 1705 nfsrv_symlink, 1706 nfsrv_mkdir, 1707 nfsrv_rmdir, 1708 nfsrv_readdir, 1709 nfsrv_statfs, 1710 nqnfsrv_readdirlook, 1711 nqnfsrv_getlease, 1712 nqnfsrv_vacated, 1713 nfsrv_noop, 1714 nqnfsrv_access, 1715 }; 1716 1717 /* 1718 * Socket upcall routine for the nfsd sockets. 1719 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1720 * Essentially do as much as possible non-blocking, else punt and it will 1721 * be called with M_WAIT from an nfsd. 1722 */ 1723 void 1724 nfsrv_rcv(so, arg, waitflag) 1725 struct socket *so; 1726 caddr_t arg; 1727 int waitflag; 1728 { 1729 register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1730 register struct mbuf *m; 1731 struct mbuf *mp, *nam; 1732 struct uio auio; 1733 int flags, error; 1734 1735 if ((slp->ns_flag & SLP_VALID) == 0) 1736 return; 1737 #ifdef notdef 1738 /* 1739 * Define this to test for nfsds handling this under heavy load. 1740 */ 1741 if (waitflag == M_DONTWAIT) { 1742 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1743 } 1744 #endif 1745 auio.uio_procp = NULL; 1746 if (so->so_type == SOCK_STREAM) { 1747 /* 1748 * If there are already records on the queue, defer soreceive() 1749 * to an nfsd so that there is feedback to the TCP layer that 1750 * the nfs servers are heavily loaded. 1751 */ 1752 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1753 slp->ns_flag |= SLP_NEEDQ; 1754 goto dorecs; 1755 } 1756 1757 /* 1758 * Do soreceive(). 1759 */ 1760 auio.uio_resid = 1000000000; 1761 flags = MSG_DONTWAIT; 1762 error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1763 if (error || mp == (struct mbuf *)0) { 1764 if (error == EWOULDBLOCK) 1765 slp->ns_flag |= SLP_NEEDQ; 1766 else 1767 slp->ns_flag |= SLP_DISCONN; 1768 goto dorecs; 1769 } 1770 m = mp; 1771 if (slp->ns_rawend) { 1772 slp->ns_rawend->m_next = m; 1773 slp->ns_cc += 1000000000 - auio.uio_resid; 1774 } else { 1775 slp->ns_raw = m; 1776 slp->ns_cc = 1000000000 - auio.uio_resid; 1777 } 1778 while (m->m_next) 1779 m = m->m_next; 1780 slp->ns_rawend = m; 1781 1782 /* 1783 * Now try and parse record(s) out of the raw stream data. 1784 */ 1785 if (error = nfsrv_getstream(slp, waitflag)) { 1786 if (error == EPERM) 1787 slp->ns_flag |= SLP_DISCONN; 1788 else 1789 slp->ns_flag |= SLP_NEEDQ; 1790 } 1791 } else { 1792 do { 1793 auio.uio_resid = 1000000000; 1794 flags = MSG_DONTWAIT; 1795 error = soreceive(so, &nam, &auio, &mp, 1796 (struct mbuf **)0, &flags); 1797 if (mp) { 1798 nfs_realign(mp, 10 * NFSX_UNSIGNED); 1799 if (nam) { 1800 m = nam; 1801 m->m_next = mp; 1802 } else 1803 m = mp; 1804 if (slp->ns_recend) 1805 slp->ns_recend->m_nextpkt = m; 1806 else 1807 slp->ns_rec = m; 1808 slp->ns_recend = m; 1809 m->m_nextpkt = (struct mbuf *)0; 1810 } 1811 if (error) { 1812 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1813 && error != EWOULDBLOCK) { 1814 slp->ns_flag |= SLP_DISCONN; 1815 goto dorecs; 1816 } 1817 } 1818 } while (mp); 1819 } 1820 1821 /* 1822 * Now try and process the request records, non-blocking. 1823 */ 1824 dorecs: 1825 if (waitflag == M_DONTWAIT && 1826 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1827 nfsrv_wakenfsd(slp); 1828 } 1829 1830 /* 1831 * Try and extract an RPC request from the mbuf data list received on a 1832 * stream socket. The "waitflag" argument indicates whether or not it 1833 * can sleep. 1834 */ 1835 nfsrv_getstream(slp, waitflag) 1836 register struct nfssvc_sock *slp; 1837 int waitflag; 1838 { 1839 register struct mbuf *m; 1840 register char *cp1, *cp2; 1841 register int len; 1842 struct mbuf *om, *m2, *recm; 1843 u_long recmark; 1844 1845 if (slp->ns_flag & SLP_GETSTREAM) 1846 panic("nfs getstream"); 1847 slp->ns_flag |= SLP_GETSTREAM; 1848 for (;;) { 1849 if (slp->ns_reclen == 0) { 1850 if (slp->ns_cc < NFSX_UNSIGNED) { 1851 slp->ns_flag &= ~SLP_GETSTREAM; 1852 return (0); 1853 } 1854 m = slp->ns_raw; 1855 if (m->m_len >= NFSX_UNSIGNED) { 1856 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 1857 m->m_data += NFSX_UNSIGNED; 1858 m->m_len -= NFSX_UNSIGNED; 1859 } else { 1860 cp1 = (caddr_t)&recmark; 1861 cp2 = mtod(m, caddr_t); 1862 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1863 while (m->m_len == 0) { 1864 m = m->m_next; 1865 cp2 = mtod(m, caddr_t); 1866 } 1867 *cp1++ = *cp2++; 1868 m->m_data++; 1869 m->m_len--; 1870 } 1871 } 1872 slp->ns_cc -= NFSX_UNSIGNED; 1873 slp->ns_reclen = ntohl(recmark) & ~0x80000000; 1874 if (slp->ns_reclen < NFS_MINPACKET || slp->ns_reclen > NFS_MAXPACKET) { 1875 slp->ns_flag &= ~SLP_GETSTREAM; 1876 return (EPERM); 1877 } 1878 } 1879 1880 /* 1881 * Now get the record part. 1882 */ 1883 if (slp->ns_cc == slp->ns_reclen) { 1884 recm = slp->ns_raw; 1885 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 1886 slp->ns_cc = slp->ns_reclen = 0; 1887 } else if (slp->ns_cc > slp->ns_reclen) { 1888 len = 0; 1889 m = slp->ns_raw; 1890 om = (struct mbuf *)0; 1891 while (len < slp->ns_reclen) { 1892 if ((len + m->m_len) > slp->ns_reclen) { 1893 m2 = m_copym(m, 0, slp->ns_reclen - len, 1894 waitflag); 1895 if (m2) { 1896 if (om) { 1897 om->m_next = m2; 1898 recm = slp->ns_raw; 1899 } else 1900 recm = m2; 1901 m->m_data += slp->ns_reclen - len; 1902 m->m_len -= slp->ns_reclen - len; 1903 len = slp->ns_reclen; 1904 } else { 1905 slp->ns_flag &= ~SLP_GETSTREAM; 1906 return (EWOULDBLOCK); 1907 } 1908 } else if ((len + m->m_len) == slp->ns_reclen) { 1909 om = m; 1910 len += m->m_len; 1911 m = m->m_next; 1912 recm = slp->ns_raw; 1913 om->m_next = (struct mbuf *)0; 1914 } else { 1915 om = m; 1916 len += m->m_len; 1917 m = m->m_next; 1918 } 1919 } 1920 slp->ns_raw = m; 1921 slp->ns_cc -= len; 1922 slp->ns_reclen = 0; 1923 } else { 1924 slp->ns_flag &= ~SLP_GETSTREAM; 1925 return (0); 1926 } 1927 nfs_realign(recm, 10 * NFSX_UNSIGNED); 1928 if (slp->ns_recend) 1929 slp->ns_recend->m_nextpkt = recm; 1930 else 1931 slp->ns_rec = recm; 1932 slp->ns_recend = recm; 1933 } 1934 } 1935 1936 /* 1937 * Parse an RPC header. 1938 */ 1939 nfsrv_dorec(slp, nd) 1940 register struct nfssvc_sock *slp; 1941 register struct nfsd *nd; 1942 { 1943 register struct mbuf *m; 1944 int error; 1945 1946 if ((slp->ns_flag & SLP_VALID) == 0 || 1947 (m = slp->ns_rec) == (struct mbuf *)0) 1948 return (ENOBUFS); 1949 if (slp->ns_rec = m->m_nextpkt) 1950 m->m_nextpkt = (struct mbuf *)0; 1951 else 1952 slp->ns_recend = (struct mbuf *)0; 1953 if (m->m_type == MT_SONAME) { 1954 nd->nd_nam = m; 1955 nd->nd_md = nd->nd_mrep = m->m_next; 1956 m->m_next = (struct mbuf *)0; 1957 } else { 1958 nd->nd_nam = (struct mbuf *)0; 1959 nd->nd_md = nd->nd_mrep = m; 1960 } 1961 nd->nd_dpos = mtod(nd->nd_md, caddr_t); 1962 if (error = nfs_getreq(nd, TRUE)) { 1963 m_freem(nd->nd_nam); 1964 return (error); 1965 } 1966 return (0); 1967 } 1968 1969 /* 1970 * Search for a sleeping nfsd and wake it up. 1971 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 1972 * running nfsds will go look for the work in the nfssvc_sock list. 1973 */ 1974 void 1975 nfsrv_wakenfsd(slp) 1976 struct nfssvc_sock *slp; 1977 { 1978 register struct nfsd *nd = nfsd_head.nd_next; 1979 1980 if ((slp->ns_flag & SLP_VALID) == 0) 1981 return; 1982 while (nd != (struct nfsd *)&nfsd_head) { 1983 if (nd->nd_flag & NFSD_WAITING) { 1984 nd->nd_flag &= ~NFSD_WAITING; 1985 if (nd->nd_slp) 1986 panic("nfsd wakeup"); 1987 slp->ns_sref++; 1988 nd->nd_slp = slp; 1989 wakeup((caddr_t)nd); 1990 return; 1991 } 1992 nd = nd->nd_next; 1993 } 1994 slp->ns_flag |= SLP_DOREC; 1995 nfsd_head.nd_flag |= NFSD_CHECKSLP; 1996 } 1997 #endif /* NFSSERVER */ 1998