nfs_socket.c revision 1.36
1 /* $NetBSD: nfs_socket.c,v 1.36 1997/04/08 17:57:16 fvdl Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1991, 1993, 1995 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 8.5 (Berkeley) 3/30/95 39 */ 40 41 /* 42 * Socket operations for use by nfs 43 */ 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/proc.h> 48 #include <sys/mount.h> 49 #include <sys/kernel.h> 50 #include <sys/mbuf.h> 51 #include <sys/vnode.h> 52 #include <sys/domain.h> 53 #include <sys/protosw.h> 54 #include <sys/socket.h> 55 #include <sys/socketvar.h> 56 #include <sys/syslog.h> 57 #include <sys/tprintf.h> 58 #include <sys/namei.h> 59 60 #include <netinet/in.h> 61 #include <netinet/tcp.h> 62 63 #include <nfs/rpcv2.h> 64 #include <nfs/nfsproto.h> 65 #include <nfs/nfs.h> 66 #include <nfs/xdr_subs.h> 67 #include <nfs/nfsm_subs.h> 68 #include <nfs/nfsmount.h> 69 #include <nfs/nfsnode.h> 70 #include <nfs/nfsrtt.h> 71 #include <nfs/nqnfs.h> 72 #include <nfs/nfs_var.h> 73 74 #define TRUE 1 75 #define FALSE 0 76 77 /* 78 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 79 * Use the mean and mean deviation of rtt for the appropriate type of rpc 80 * for the frequent rpcs and a default for the others. 81 * The justification for doing "other" this way is that these rpcs 82 * happen so infrequently that timer est. would probably be stale. 83 * Also, since many of these rpcs are 84 * non-idempotent, a conservative timeout is desired. 85 * getattr, lookup - A+2D 86 * read, write - A+4D 87 * other - nm_timeo 88 */ 89 #define NFS_RTO(n, t) \ 90 ((t) == 0 ? (n)->nm_timeo : \ 91 ((t) < 3 ? \ 92 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 93 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 94 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 95 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 96 /* 97 * External data, mostly RPC constants in XDR form 98 */ 99 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 100 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, 101 rpc_auth_kerb; 102 extern u_int32_t nfs_prog, nqnfs_prog; 103 extern time_t nqnfsstarttime; 104 extern struct nfsstats nfsstats; 105 extern int nfsv3_procid[NFS_NPROCS]; 106 extern int nfs_ticks; 107 108 /* 109 * Defines which timer to use for the procnum. 110 * 0 - default 111 * 1 - getattr 112 * 2 - lookup 113 * 3 - read 114 * 4 - write 115 */ 116 static int proct[NFS_NPROCS] = { 117 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0, 118 0, 0, 0, 119 }; 120 121 /* 122 * There is a congestion window for outstanding rpcs maintained per mount 123 * point. The cwnd size is adjusted in roughly the way that: 124 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 125 * SIGCOMM '88". ACM, August 1988. 126 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 127 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 128 * of rpcs is in progress. 129 * (The sent count and cwnd are scaled for integer arith.) 130 * Variants of "slow start" were tried and were found to be too much of a 131 * performance hit (ave. rtt 3 times larger), 132 * I suspect due to the large rtt that nfs rpcs have. 133 */ 134 #define NFS_CWNDSCALE 256 135 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 136 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 137 int nfsrtton = 0; 138 struct nfsrtt nfsrtt; 139 140 /* 141 * Initialize sockets and congestion for a new NFS connection. 142 * We do not free the sockaddr if error. 143 */ 144 int 145 nfs_connect(nmp, rep) 146 register struct nfsmount *nmp; 147 struct nfsreq *rep; 148 { 149 register struct socket *so; 150 int s, error, rcvreserve, sndreserve; 151 struct sockaddr *saddr; 152 struct sockaddr_in *sin; 153 struct mbuf *m; 154 u_int16_t tport; 155 156 nmp->nm_so = (struct socket *)0; 157 saddr = mtod(nmp->nm_nam, struct sockaddr *); 158 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 159 nmp->nm_soproto); 160 if (error) 161 goto bad; 162 so = nmp->nm_so; 163 nmp->nm_soflags = so->so_proto->pr_flags; 164 165 /* 166 * Some servers require that the client port be a reserved port number. 167 */ 168 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 169 MGET(m, M_WAIT, MT_SONAME); 170 sin = mtod(m, struct sockaddr_in *); 171 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 172 sin->sin_family = AF_INET; 173 sin->sin_addr.s_addr = INADDR_ANY; 174 tport = IPPORT_RESERVED - 1; 175 sin->sin_port = htons(tport); 176 while ((error = sobind(so, m)) == EADDRINUSE && 177 --tport > IPPORT_RESERVED / 2) 178 sin->sin_port = htons(tport); 179 m_freem(m); 180 if (error) 181 goto bad; 182 } 183 184 /* 185 * Protocols that do not require connections may be optionally left 186 * unconnected for servers that reply from a port other than NFS_PORT. 187 */ 188 if (nmp->nm_flag & NFSMNT_NOCONN) { 189 if (nmp->nm_soflags & PR_CONNREQUIRED) { 190 error = ENOTCONN; 191 goto bad; 192 } 193 } else { 194 error = soconnect(so, nmp->nm_nam); 195 if (error) 196 goto bad; 197 198 /* 199 * Wait for the connection to complete. Cribbed from the 200 * connect system call but with the wait timing out so 201 * that interruptible mounts don't hang here for a long time. 202 */ 203 s = splsoftnet(); 204 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 205 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 206 "nfscon", 2 * hz); 207 if ((so->so_state & SS_ISCONNECTING) && 208 so->so_error == 0 && rep && 209 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ 210 so->so_state &= ~SS_ISCONNECTING; 211 splx(s); 212 goto bad; 213 } 214 } 215 if (so->so_error) { 216 error = so->so_error; 217 so->so_error = 0; 218 splx(s); 219 goto bad; 220 } 221 splx(s); 222 } 223 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 224 so->so_rcv.sb_timeo = (5 * hz); 225 so->so_snd.sb_timeo = (5 * hz); 226 } else { 227 so->so_rcv.sb_timeo = 0; 228 so->so_snd.sb_timeo = 0; 229 } 230 if (nmp->nm_sotype == SOCK_DGRAM) { 231 sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR; 232 rcvreserve = max(nmp->nm_rsize, nmp->nm_readdirsize) + 233 NFS_MAXPKTHDR; 234 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 235 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 236 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 237 NFS_MAXPKTHDR) * 2; 238 } else { 239 if (nmp->nm_sotype != SOCK_STREAM) 240 panic("nfscon sotype"); 241 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 242 MGET(m, M_WAIT, MT_SOOPTS); 243 *mtod(m, int32_t *) = 1; 244 m->m_len = sizeof(int32_t); 245 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 246 } 247 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 248 MGET(m, M_WAIT, MT_SOOPTS); 249 *mtod(m, int32_t *) = 1; 250 m->m_len = sizeof(int32_t); 251 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 252 } 253 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 254 sizeof (u_int32_t)) * 2; 255 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 256 sizeof (u_int32_t)) * 2; 257 } 258 error = soreserve(so, sndreserve, rcvreserve); 259 if (error) 260 goto bad; 261 so->so_rcv.sb_flags |= SB_NOINTR; 262 so->so_snd.sb_flags |= SB_NOINTR; 263 264 /* Initialize other non-zero congestion variables */ 265 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 266 nmp->nm_srtt[4] = (NFS_TIMEO << 3); 267 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 268 nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0; 269 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 270 nmp->nm_sent = 0; 271 nmp->nm_timeouts = 0; 272 return (0); 273 274 bad: 275 nfs_disconnect(nmp); 276 return (error); 277 } 278 279 /* 280 * Reconnect routine: 281 * Called when a connection is broken on a reliable protocol. 282 * - clean up the old socket 283 * - nfs_connect() again 284 * - set R_MUSTRESEND for all outstanding requests on mount point 285 * If this fails the mount point is DEAD! 286 * nb: Must be called with the nfs_sndlock() set on the mount point. 287 */ 288 int 289 nfs_reconnect(rep) 290 register struct nfsreq *rep; 291 { 292 register struct nfsreq *rp; 293 register struct nfsmount *nmp = rep->r_nmp; 294 int error; 295 296 nfs_disconnect(nmp); 297 while ((error = nfs_connect(nmp, rep)) != 0) { 298 if (error == EINTR || error == ERESTART) 299 return (EINTR); 300 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); 301 } 302 303 /* 304 * Loop through outstanding request list and fix up all requests 305 * on old socket. 306 */ 307 for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) { 308 if (rp->r_nmp == nmp) 309 rp->r_flags |= R_MUSTRESEND; 310 } 311 return (0); 312 } 313 314 /* 315 * NFS disconnect. Clean up and unlink. 316 */ 317 void 318 nfs_disconnect(nmp) 319 register struct nfsmount *nmp; 320 { 321 register struct socket *so; 322 323 if (nmp->nm_so) { 324 so = nmp->nm_so; 325 nmp->nm_so = (struct socket *)0; 326 soshutdown(so, 2); 327 soclose(so); 328 } 329 } 330 331 /* 332 * This is the nfs send routine. For connection based socket types, it 333 * must be called with an nfs_sndlock() on the socket. 334 * "rep == NULL" indicates that it has been called from a server. 335 * For the client side: 336 * - return EINTR if the RPC is terminated, 0 otherwise 337 * - set R_MUSTRESEND if the send fails for any reason 338 * - do any cleanup required by recoverable socket errors (???) 339 * For the server side: 340 * - return EINTR or ERESTART if interrupted by a signal 341 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 342 * - do any cleanup required by recoverable socket errors (???) 343 */ 344 int 345 nfs_send(so, nam, top, rep) 346 register struct socket *so; 347 struct mbuf *nam; 348 register struct mbuf *top; 349 struct nfsreq *rep; 350 { 351 struct mbuf *sendnam; 352 int error, soflags, flags; 353 354 if (rep) { 355 if (rep->r_flags & R_SOFTTERM) { 356 m_freem(top); 357 return (EINTR); 358 } 359 if ((so = rep->r_nmp->nm_so) == NULL) { 360 rep->r_flags |= R_MUSTRESEND; 361 m_freem(top); 362 return (0); 363 } 364 rep->r_flags &= ~R_MUSTRESEND; 365 soflags = rep->r_nmp->nm_soflags; 366 } else 367 soflags = so->so_proto->pr_flags; 368 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 369 sendnam = (struct mbuf *)0; 370 else 371 sendnam = nam; 372 if (so->so_type == SOCK_SEQPACKET) 373 flags = MSG_EOR; 374 else 375 flags = 0; 376 377 error = sosend(so, sendnam, (struct uio *)0, top, 378 (struct mbuf *)0, flags); 379 if (error) { 380 if (rep) { 381 log(LOG_INFO, "nfs send error %d for server %s\n",error, 382 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 383 /* 384 * Deal with errors for the client side. 385 */ 386 if (rep->r_flags & R_SOFTTERM) 387 error = EINTR; 388 else 389 rep->r_flags |= R_MUSTRESEND; 390 } else 391 log(LOG_INFO, "nfsd send error %d\n", error); 392 393 /* 394 * Handle any recoverable (soft) socket errors here. (???) 395 */ 396 if (error != EINTR && error != ERESTART && 397 error != EWOULDBLOCK && error != EPIPE) 398 error = 0; 399 } 400 return (error); 401 } 402 403 #ifdef NFS 404 /* 405 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 406 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 407 * Mark and consolidate the data into a new mbuf list. 408 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 409 * small mbufs. 410 * For SOCK_STREAM we must be very careful to read an entire record once 411 * we have read any of it, even if the system call has been interrupted. 412 */ 413 int 414 nfs_receive(rep, aname, mp) 415 register struct nfsreq *rep; 416 struct mbuf **aname; 417 struct mbuf **mp; 418 { 419 register struct socket *so; 420 struct uio auio; 421 struct iovec aio; 422 register struct mbuf *m; 423 struct mbuf *control; 424 u_int32_t len; 425 struct mbuf **getnam; 426 int error, sotype, rcvflg; 427 struct proc *p = curproc; /* XXX */ 428 429 /* 430 * Set up arguments for soreceive() 431 */ 432 *mp = (struct mbuf *)0; 433 *aname = (struct mbuf *)0; 434 sotype = rep->r_nmp->nm_sotype; 435 436 /* 437 * For reliable protocols, lock against other senders/receivers 438 * in case a reconnect is necessary. 439 * For SOCK_STREAM, first get the Record Mark to find out how much 440 * more there is to get. 441 * We must lock the socket against other receivers 442 * until we have an entire rpc request/reply. 443 */ 444 if (sotype != SOCK_DGRAM) { 445 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 446 if (error) 447 return (error); 448 tryagain: 449 /* 450 * Check for fatal errors and resending request. 451 */ 452 /* 453 * Ugh: If a reconnect attempt just happened, nm_so 454 * would have changed. NULL indicates a failed 455 * attempt that has essentially shut down this 456 * mount point. 457 */ 458 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 459 nfs_sndunlock(&rep->r_nmp->nm_flag); 460 return (EINTR); 461 } 462 so = rep->r_nmp->nm_so; 463 if (!so) { 464 error = nfs_reconnect(rep); 465 if (error) { 466 nfs_sndunlock(&rep->r_nmp->nm_flag); 467 return (error); 468 } 469 goto tryagain; 470 } 471 while (rep->r_flags & R_MUSTRESEND) { 472 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 473 nfsstats.rpcretries++; 474 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 475 if (error) { 476 if (error == EINTR || error == ERESTART || 477 (error = nfs_reconnect(rep)) != 0) { 478 nfs_sndunlock(&rep->r_nmp->nm_flag); 479 return (error); 480 } 481 goto tryagain; 482 } 483 } 484 nfs_sndunlock(&rep->r_nmp->nm_flag); 485 if (sotype == SOCK_STREAM) { 486 aio.iov_base = (caddr_t) &len; 487 aio.iov_len = sizeof(u_int32_t); 488 auio.uio_iov = &aio; 489 auio.uio_iovcnt = 1; 490 auio.uio_segflg = UIO_SYSSPACE; 491 auio.uio_rw = UIO_READ; 492 auio.uio_offset = 0; 493 auio.uio_resid = sizeof(u_int32_t); 494 auio.uio_procp = p; 495 do { 496 rcvflg = MSG_WAITALL; 497 error = soreceive(so, (struct mbuf **)0, &auio, 498 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 499 if (error == EWOULDBLOCK && rep) { 500 if (rep->r_flags & R_SOFTTERM) 501 return (EINTR); 502 } 503 } while (error == EWOULDBLOCK); 504 if (!error && auio.uio_resid > 0) { 505 log(LOG_INFO, 506 "short receive (%d/%d) from nfs server %s\n", 507 sizeof(u_int32_t) - auio.uio_resid, 508 sizeof(u_int32_t), 509 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 510 error = EPIPE; 511 } 512 if (error) 513 goto errout; 514 len = ntohl(len) & ~0x80000000; 515 /* 516 * This is SERIOUS! We are out of sync with the sender 517 * and forcing a disconnect/reconnect is all I can do. 518 */ 519 if (len > NFS_MAXPACKET) { 520 log(LOG_ERR, "%s (%d) from nfs server %s\n", 521 "impossible packet length", 522 len, 523 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 524 error = EFBIG; 525 goto errout; 526 } 527 auio.uio_resid = len; 528 do { 529 rcvflg = MSG_WAITALL; 530 error = soreceive(so, (struct mbuf **)0, 531 &auio, mp, (struct mbuf **)0, &rcvflg); 532 } while (error == EWOULDBLOCK || error == EINTR || 533 error == ERESTART); 534 if (!error && auio.uio_resid > 0) { 535 log(LOG_INFO, 536 "short receive (%d/%d) from nfs server %s\n", 537 len - auio.uio_resid, len, 538 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 539 error = EPIPE; 540 } 541 } else { 542 /* 543 * NB: Since uio_resid is big, MSG_WAITALL is ignored 544 * and soreceive() will return when it has either a 545 * control msg or a data msg. 546 * We have no use for control msg., but must grab them 547 * and then throw them away so we know what is going 548 * on. 549 */ 550 auio.uio_resid = len = 100000000; /* Anything Big */ 551 auio.uio_procp = p; 552 do { 553 rcvflg = 0; 554 error = soreceive(so, (struct mbuf **)0, 555 &auio, mp, &control, &rcvflg); 556 if (control) 557 m_freem(control); 558 if (error == EWOULDBLOCK && rep) { 559 if (rep->r_flags & R_SOFTTERM) 560 return (EINTR); 561 } 562 } while (error == EWOULDBLOCK || 563 (!error && *mp == NULL && control)); 564 if ((rcvflg & MSG_EOR) == 0) 565 printf("Egad!!\n"); 566 if (!error && *mp == NULL) 567 error = EPIPE; 568 len -= auio.uio_resid; 569 } 570 errout: 571 if (error && error != EINTR && error != ERESTART) { 572 m_freem(*mp); 573 *mp = (struct mbuf *)0; 574 if (error != EPIPE) 575 log(LOG_INFO, 576 "receive error %d from nfs server %s\n", 577 error, 578 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 579 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 580 if (!error) 581 error = nfs_reconnect(rep); 582 if (!error) 583 goto tryagain; 584 } 585 } else { 586 if ((so = rep->r_nmp->nm_so) == NULL) 587 return (EACCES); 588 if (so->so_state & SS_ISCONNECTED) 589 getnam = (struct mbuf **)0; 590 else 591 getnam = aname; 592 auio.uio_resid = len = 1000000; 593 auio.uio_procp = p; 594 do { 595 rcvflg = 0; 596 error = soreceive(so, getnam, &auio, mp, 597 (struct mbuf **)0, &rcvflg); 598 if (error == EWOULDBLOCK && 599 (rep->r_flags & R_SOFTTERM)) 600 return (EINTR); 601 } while (error == EWOULDBLOCK); 602 len -= auio.uio_resid; 603 } 604 if (error) { 605 m_freem(*mp); 606 *mp = (struct mbuf *)0; 607 } 608 return (error); 609 } 610 611 /* 612 * Implement receipt of reply on a socket. 613 * We must search through the list of received datagrams matching them 614 * with outstanding requests using the xid, until ours is found. 615 */ 616 /* ARGSUSED */ 617 int 618 nfs_reply(myrep) 619 struct nfsreq *myrep; 620 { 621 register struct nfsreq *rep; 622 register struct nfsmount *nmp = myrep->r_nmp; 623 register int32_t t1; 624 struct mbuf *mrep, *nam, *md; 625 u_int32_t rxid, *tl; 626 caddr_t dpos, cp2; 627 int error; 628 629 /* 630 * Loop around until we get our own reply 631 */ 632 for (;;) { 633 /* 634 * Lock against other receivers so that I don't get stuck in 635 * sbwait() after someone else has received my reply for me. 636 * Also necessary for connection based protocols to avoid 637 * race conditions during a reconnect. 638 */ 639 error = nfs_rcvlock(myrep); 640 if (error == EALREADY) 641 return (0); 642 if (error) 643 return (error); 644 /* 645 * Get the next Rpc reply off the socket 646 */ 647 error = nfs_receive(myrep, &nam, &mrep); 648 nfs_rcvunlock(&nmp->nm_flag); 649 if (error) { 650 651 /* 652 * Ignore routing errors on connectionless protocols?? 653 */ 654 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 655 nmp->nm_so->so_error = 0; 656 printf("nfs_reply: ignoring error %d\n", error); 657 if (myrep->r_flags & R_GETONEREP) 658 return (0); 659 continue; 660 } 661 return (error); 662 } 663 if (nam) 664 m_freem(nam); 665 666 /* 667 * Get the xid and check that it is an rpc reply 668 */ 669 md = mrep; 670 dpos = mtod(md, caddr_t); 671 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 672 rxid = *tl++; 673 if (*tl != rpc_reply) { 674 if (nmp->nm_flag & NFSMNT_NQNFS) { 675 if (nqnfs_callback(nmp, mrep, md, dpos)) 676 nfsstats.rpcinvalid++; 677 } else { 678 nfsstats.rpcinvalid++; 679 m_freem(mrep); 680 } 681 nfsmout: 682 if (myrep->r_flags & R_GETONEREP) 683 return (0); 684 continue; 685 } 686 687 /* 688 * Loop through the request list to match up the reply 689 * Iff no match, just drop the datagram 690 */ 691 for (rep = nfs_reqq.tqh_first; rep != 0; 692 rep = rep->r_chain.tqe_next) { 693 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 694 /* Found it.. */ 695 rep->r_mrep = mrep; 696 rep->r_md = md; 697 rep->r_dpos = dpos; 698 if (nfsrtton) { 699 struct rttl *rt; 700 701 rt = &nfsrtt.rttl[nfsrtt.pos]; 702 rt->proc = rep->r_procnum; 703 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 704 rt->sent = nmp->nm_sent; 705 rt->cwnd = nmp->nm_cwnd; 706 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 707 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 708 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 709 rt->tstamp = time; 710 if (rep->r_flags & R_TIMING) 711 rt->rtt = rep->r_rtt; 712 else 713 rt->rtt = 1000000; 714 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 715 } 716 /* 717 * Update congestion window. 718 * Do the additive increase of 719 * one rpc/rtt. 720 */ 721 if (nmp->nm_cwnd <= nmp->nm_sent) { 722 nmp->nm_cwnd += 723 (NFS_CWNDSCALE * NFS_CWNDSCALE + 724 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 725 if (nmp->nm_cwnd > NFS_MAXCWND) 726 nmp->nm_cwnd = NFS_MAXCWND; 727 } 728 rep->r_flags &= ~R_SENT; 729 nmp->nm_sent -= NFS_CWNDSCALE; 730 /* 731 * Update rtt using a gain of 0.125 on the mean 732 * and a gain of 0.25 on the deviation. 733 */ 734 if (rep->r_flags & R_TIMING) { 735 /* 736 * Since the timer resolution of 737 * NFS_HZ is so course, it can often 738 * result in r_rtt == 0. Since 739 * r_rtt == N means that the actual 740 * rtt is between N+dt and N+2-dt ticks, 741 * add 1. 742 */ 743 t1 = rep->r_rtt + 1; 744 t1 -= (NFS_SRTT(rep) >> 3); 745 NFS_SRTT(rep) += t1; 746 if (t1 < 0) 747 t1 = -t1; 748 t1 -= (NFS_SDRTT(rep) >> 2); 749 NFS_SDRTT(rep) += t1; 750 } 751 nmp->nm_timeouts = 0; 752 break; 753 } 754 } 755 /* 756 * If not matched to a request, drop it. 757 * If it's mine, get out. 758 */ 759 if (rep == 0) { 760 nfsstats.rpcunexpected++; 761 m_freem(mrep); 762 } else if (rep == myrep) { 763 if (rep->r_mrep == NULL) 764 panic("nfsreply nil"); 765 return (0); 766 } 767 if (myrep->r_flags & R_GETONEREP) 768 return (0); 769 } 770 } 771 772 /* 773 * nfs_request - goes something like this 774 * - fill in request struct 775 * - links it into list 776 * - calls nfs_send() for first transmit 777 * - calls nfs_receive() to get reply 778 * - break down rpc header and return with nfs reply pointed to 779 * by mrep or error 780 * nb: always frees up mreq mbuf list 781 */ 782 int 783 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 784 struct vnode *vp; 785 struct mbuf *mrest; 786 int procnum; 787 struct proc *procp; 788 struct ucred *cred; 789 struct mbuf **mrp; 790 struct mbuf **mdp; 791 caddr_t *dposp; 792 { 793 register struct mbuf *m, *mrep; 794 register struct nfsreq *rep; 795 register u_int32_t *tl; 796 register int i; 797 struct nfsmount *nmp; 798 struct mbuf *md, *mheadend; 799 struct nfsnode *np; 800 char nickv[RPCX_NICKVERF]; 801 time_t reqtime, waituntil; 802 caddr_t dpos, cp2; 803 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; 804 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 805 int verf_len, verf_type; 806 u_int32_t xid; 807 u_quad_t frev; 808 char *auth_str, *verf_str; 809 NFSKERBKEY_T key; /* save session key */ 810 811 nmp = VFSTONFS(vp->v_mount); 812 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 813 rep->r_nmp = nmp; 814 rep->r_vp = vp; 815 rep->r_procp = procp; 816 rep->r_procnum = procnum; 817 i = 0; 818 m = mrest; 819 while (m) { 820 i += m->m_len; 821 m = m->m_next; 822 } 823 mrest_len = i; 824 825 /* 826 * Get the RPC header with authorization. 827 */ 828 kerbauth: 829 verf_str = auth_str = (char *)0; 830 if (nmp->nm_flag & NFSMNT_KERB) { 831 verf_str = nickv; 832 verf_len = sizeof (nickv); 833 auth_type = RPCAUTH_KERB4; 834 bzero((caddr_t)key, sizeof (key)); 835 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 836 &auth_len, verf_str, verf_len)) { 837 error = nfs_getauth(nmp, rep, cred, &auth_str, 838 &auth_len, verf_str, &verf_len, key); 839 if (error) { 840 free((caddr_t)rep, M_NFSREQ); 841 m_freem(mrest); 842 return (error); 843 } 844 } 845 } else { 846 auth_type = RPCAUTH_UNIX; 847 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 848 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 849 5 * NFSX_UNSIGNED; 850 } 851 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 852 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 853 if (auth_str) 854 free(auth_str, M_TEMP); 855 856 /* 857 * For stream protocols, insert a Sun RPC Record Mark. 858 */ 859 if (nmp->nm_sotype == SOCK_STREAM) { 860 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 861 *mtod(m, u_int32_t *) = htonl(0x80000000 | 862 (m->m_pkthdr.len - NFSX_UNSIGNED)); 863 } 864 rep->r_mreq = m; 865 rep->r_xid = xid; 866 tryagain: 867 if (nmp->nm_flag & NFSMNT_SOFT) 868 rep->r_retry = nmp->nm_retry; 869 else 870 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 871 rep->r_rtt = rep->r_rexmit = 0; 872 if (proct[procnum] > 0) 873 rep->r_flags = R_TIMING; 874 else 875 rep->r_flags = 0; 876 rep->r_mrep = NULL; 877 878 /* 879 * Do the client side RPC. 880 */ 881 nfsstats.rpcrequests++; 882 /* 883 * Chain request into list of outstanding requests. Be sure 884 * to put it LAST so timer finds oldest requests first. 885 */ 886 s = splsoftnet(); 887 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 888 889 /* Get send time for nqnfs */ 890 reqtime = time.tv_sec; 891 892 /* 893 * If backing off another request or avoiding congestion, don't 894 * send this one now but let timer do it. If not timing a request, 895 * do it now. 896 */ 897 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 898 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 899 nmp->nm_sent < nmp->nm_cwnd)) { 900 splx(s); 901 if (nmp->nm_soflags & PR_CONNREQUIRED) 902 error = nfs_sndlock(&nmp->nm_flag, rep); 903 if (!error) { 904 m = m_copym(m, 0, M_COPYALL, M_WAIT); 905 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 906 if (nmp->nm_soflags & PR_CONNREQUIRED) 907 nfs_sndunlock(&nmp->nm_flag); 908 } 909 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 910 nmp->nm_sent += NFS_CWNDSCALE; 911 rep->r_flags |= R_SENT; 912 } 913 } else { 914 splx(s); 915 rep->r_rtt = -1; 916 } 917 918 /* 919 * Wait for the reply from our send or the timer's. 920 */ 921 if (!error || error == EPIPE) 922 error = nfs_reply(rep); 923 924 /* 925 * RPC done, unlink the request. 926 */ 927 s = splsoftnet(); 928 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 929 splx(s); 930 931 /* 932 * Decrement the outstanding request count. 933 */ 934 if (rep->r_flags & R_SENT) { 935 rep->r_flags &= ~R_SENT; /* paranoia */ 936 nmp->nm_sent -= NFS_CWNDSCALE; 937 } 938 939 /* 940 * If there was a successful reply and a tprintf msg. 941 * tprintf a response. 942 */ 943 if (!error && (rep->r_flags & R_TPRINTFMSG)) 944 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 945 "is alive again"); 946 mrep = rep->r_mrep; 947 md = rep->r_md; 948 dpos = rep->r_dpos; 949 if (error) { 950 m_freem(rep->r_mreq); 951 free((caddr_t)rep, M_NFSREQ); 952 return (error); 953 } 954 955 /* 956 * break down the rpc header and check if ok 957 */ 958 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 959 if (*tl++ == rpc_msgdenied) { 960 if (*tl == rpc_mismatch) 961 error = EOPNOTSUPP; 962 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 963 if (!failed_auth) { 964 failed_auth++; 965 mheadend->m_next = (struct mbuf *)0; 966 m_freem(mrep); 967 m_freem(rep->r_mreq); 968 goto kerbauth; 969 } else 970 error = EAUTH; 971 } else 972 error = EACCES; 973 m_freem(mrep); 974 m_freem(rep->r_mreq); 975 free((caddr_t)rep, M_NFSREQ); 976 return (error); 977 } 978 979 /* 980 * Grab any Kerberos verifier, otherwise just throw it away. 981 */ 982 verf_type = fxdr_unsigned(int, *tl++); 983 i = fxdr_unsigned(int32_t, *tl); 984 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 985 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 986 if (error) 987 goto nfsmout; 988 } else if (i > 0) 989 nfsm_adv(nfsm_rndup(i)); 990 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 991 /* 0 == ok */ 992 if (*tl == 0) { 993 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 994 if (*tl != 0) { 995 error = fxdr_unsigned(int, *tl); 996 if ((nmp->nm_flag & NFSMNT_NFSV3) && 997 error == NFSERR_TRYLATER) { 998 m_freem(mrep); 999 error = 0; 1000 waituntil = time.tv_sec + trylater_delay; 1001 while (time.tv_sec < waituntil) 1002 (void) tsleep((caddr_t)&lbolt, 1003 PSOCK, "nqnfstry", 0); 1004 trylater_delay *= nfs_backoff[trylater_cnt]; 1005 if (trylater_cnt < 7) 1006 trylater_cnt++; 1007 goto tryagain; 1008 } 1009 1010 /* 1011 * If the File Handle was stale, invalidate the 1012 * lookup cache, just in case. 1013 */ 1014 if (error == ESTALE) 1015 cache_purge(vp); 1016 if (nmp->nm_flag & NFSMNT_NFSV3) { 1017 *mrp = mrep; 1018 *mdp = md; 1019 *dposp = dpos; 1020 error |= NFSERR_RETERR; 1021 } else 1022 m_freem(mrep); 1023 m_freem(rep->r_mreq); 1024 free((caddr_t)rep, M_NFSREQ); 1025 return (error); 1026 } 1027 1028 /* 1029 * For nqnfs, get any lease in reply 1030 */ 1031 if (nmp->nm_flag & NFSMNT_NQNFS) { 1032 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1033 if (*tl) { 1034 np = VTONFS(vp); 1035 nqlflag = fxdr_unsigned(int, *tl); 1036 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1037 cachable = fxdr_unsigned(int, *tl++); 1038 reqtime += fxdr_unsigned(int, *tl++); 1039 if (reqtime > time.tv_sec) { 1040 fxdr_hyper(tl, &frev); 1041 nqnfs_clientlease(nmp, np, nqlflag, 1042 cachable, reqtime, frev); 1043 } 1044 } 1045 } 1046 *mrp = mrep; 1047 *mdp = md; 1048 *dposp = dpos; 1049 m_freem(rep->r_mreq); 1050 FREE((caddr_t)rep, M_NFSREQ); 1051 return (0); 1052 } 1053 m_freem(mrep); 1054 error = EPROTONOSUPPORT; 1055 nfsmout: 1056 m_freem(rep->r_mreq); 1057 free((caddr_t)rep, M_NFSREQ); 1058 return (error); 1059 } 1060 #endif /* NFS */ 1061 1062 /* 1063 * Generate the rpc reply header 1064 * siz arg. is used to decide if adding a cluster is worthwhile 1065 */ 1066 int 1067 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1068 int siz; 1069 struct nfsrv_descript *nd; 1070 struct nfssvc_sock *slp; 1071 int err; 1072 int cache; 1073 u_quad_t *frev; 1074 struct mbuf **mrq; 1075 struct mbuf **mbp; 1076 caddr_t *bposp; 1077 { 1078 register u_int32_t *tl; 1079 register struct mbuf *mreq; 1080 caddr_t bpos; 1081 struct mbuf *mb, *mb2; 1082 1083 MGETHDR(mreq, M_WAIT, MT_DATA); 1084 mb = mreq; 1085 /* 1086 * If this is a big reply, use a cluster else 1087 * try and leave leading space for the lower level headers. 1088 */ 1089 siz += RPC_REPLYSIZ; 1090 if (siz >= MINCLSIZE) { 1091 MCLGET(mreq, M_WAIT); 1092 } else 1093 mreq->m_data += max_hdr; 1094 tl = mtod(mreq, u_int32_t *); 1095 mreq->m_len = 6 * NFSX_UNSIGNED; 1096 bpos = ((caddr_t)tl) + mreq->m_len; 1097 *tl++ = txdr_unsigned(nd->nd_retxid); 1098 *tl++ = rpc_reply; 1099 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1100 *tl++ = rpc_msgdenied; 1101 if (err & NFSERR_AUTHERR) { 1102 *tl++ = rpc_autherr; 1103 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1104 mreq->m_len -= NFSX_UNSIGNED; 1105 bpos -= NFSX_UNSIGNED; 1106 } else { 1107 *tl++ = rpc_mismatch; 1108 *tl++ = txdr_unsigned(RPC_VER2); 1109 *tl = txdr_unsigned(RPC_VER2); 1110 } 1111 } else { 1112 *tl++ = rpc_msgaccepted; 1113 1114 /* 1115 * For Kerberos authentication, we must send the nickname 1116 * verifier back, otherwise just RPCAUTH_NULL. 1117 */ 1118 if (nd->nd_flag & ND_KERBFULL) { 1119 register struct nfsuid *nuidp; 1120 struct timeval ktvin, ktvout; 1121 1122 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; 1123 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1124 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1125 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), 1126 &nuidp->nu_haddr, nd->nd_nam2))) 1127 break; 1128 } 1129 if (nuidp) { 1130 ktvin.tv_sec = 1131 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); 1132 ktvin.tv_usec = 1133 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1134 1135 /* 1136 * Encrypt the timestamp in ecb mode using the 1137 * session key. 1138 */ 1139 #ifdef NFSKERB 1140 XXX 1141 #endif 1142 1143 *tl++ = rpc_auth_kerb; 1144 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1145 *tl = ktvout.tv_sec; 1146 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1147 *tl++ = ktvout.tv_usec; 1148 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1149 } else { 1150 *tl++ = 0; 1151 *tl++ = 0; 1152 } 1153 } else { 1154 *tl++ = 0; 1155 *tl++ = 0; 1156 } 1157 switch (err) { 1158 case EPROGUNAVAIL: 1159 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1160 break; 1161 case EPROGMISMATCH: 1162 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1163 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1164 if (nd->nd_flag & ND_NQNFS) { 1165 *tl++ = txdr_unsigned(3); 1166 *tl = txdr_unsigned(3); 1167 } else { 1168 *tl++ = txdr_unsigned(2); 1169 *tl = txdr_unsigned(3); 1170 } 1171 break; 1172 case EPROCUNAVAIL: 1173 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1174 break; 1175 case EBADRPC: 1176 *tl = txdr_unsigned(RPC_GARBAGE); 1177 break; 1178 default: 1179 *tl = 0; 1180 if (err != NFSERR_RETVOID) { 1181 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1182 if (err) 1183 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1184 else 1185 *tl = 0; 1186 } 1187 break; 1188 }; 1189 } 1190 1191 /* 1192 * For nqnfs, piggyback lease as requested. 1193 */ 1194 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1195 if (nd->nd_flag & ND_LEASE) { 1196 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1197 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1198 *tl++ = txdr_unsigned(cache); 1199 *tl++ = txdr_unsigned(nd->nd_duration); 1200 txdr_hyper(frev, tl); 1201 } else { 1202 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1203 *tl = 0; 1204 } 1205 } 1206 if (mrq != NULL) 1207 *mrq = mreq; 1208 *mbp = mb; 1209 *bposp = bpos; 1210 if (err != 0 && err != NFSERR_RETVOID) 1211 nfsstats.srvrpc_errs++; 1212 return (0); 1213 } 1214 1215 /* 1216 * Nfs timer routine 1217 * Scan the nfsreq list and retranmit any requests that have timed out 1218 * To avoid retransmission attempts on STREAM sockets (in the future) make 1219 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1220 */ 1221 void 1222 nfs_timer(arg) 1223 void *arg; /* never used */ 1224 { 1225 register struct nfsreq *rep; 1226 register struct mbuf *m; 1227 register struct socket *so; 1228 register struct nfsmount *nmp; 1229 register int timeo; 1230 int s, error; 1231 #ifdef NFSSERVER 1232 register struct nfssvc_sock *slp; 1233 static long lasttime = 0; 1234 u_quad_t cur_usec; 1235 #endif 1236 1237 s = splsoftnet(); 1238 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1239 nmp = rep->r_nmp; 1240 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1241 continue; 1242 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1243 rep->r_flags |= R_SOFTTERM; 1244 continue; 1245 } 1246 if (rep->r_rtt >= 0) { 1247 rep->r_rtt++; 1248 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1249 timeo = nmp->nm_timeo; 1250 else 1251 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1252 if (nmp->nm_timeouts > 0) 1253 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1254 if (rep->r_rtt <= timeo) 1255 continue; 1256 if (nmp->nm_timeouts < 8) 1257 nmp->nm_timeouts++; 1258 } 1259 /* 1260 * Check for server not responding 1261 */ 1262 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1263 rep->r_rexmit > nmp->nm_deadthresh) { 1264 nfs_msg(rep->r_procp, 1265 nmp->nm_mountp->mnt_stat.f_mntfromname, 1266 "not responding"); 1267 rep->r_flags |= R_TPRINTFMSG; 1268 } 1269 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1270 nfsstats.rpctimeouts++; 1271 rep->r_flags |= R_SOFTTERM; 1272 continue; 1273 } 1274 if (nmp->nm_sotype != SOCK_DGRAM) { 1275 if (++rep->r_rexmit > NFS_MAXREXMIT) 1276 rep->r_rexmit = NFS_MAXREXMIT; 1277 continue; 1278 } 1279 if ((so = nmp->nm_so) == NULL) 1280 continue; 1281 1282 /* 1283 * If there is enough space and the window allows.. 1284 * Resend it 1285 * Set r_rtt to -1 in case we fail to send it now. 1286 */ 1287 rep->r_rtt = -1; 1288 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1289 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1290 (rep->r_flags & R_SENT) || 1291 nmp->nm_sent < nmp->nm_cwnd) && 1292 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1293 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1294 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1295 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1296 else 1297 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1298 nmp->nm_nam, (struct mbuf *)0, (struct proc *)0); 1299 if (error) { 1300 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1301 printf("nfs_timer: ignoring error %d\n", 1302 error); 1303 so->so_error = 0; 1304 } 1305 } else { 1306 /* 1307 * Iff first send, start timing 1308 * else turn timing off, backoff timer 1309 * and divide congestion window by 2. 1310 */ 1311 if (rep->r_flags & R_SENT) { 1312 rep->r_flags &= ~R_TIMING; 1313 if (++rep->r_rexmit > NFS_MAXREXMIT) 1314 rep->r_rexmit = NFS_MAXREXMIT; 1315 nmp->nm_cwnd >>= 1; 1316 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1317 nmp->nm_cwnd = NFS_CWNDSCALE; 1318 nfsstats.rpcretries++; 1319 } else { 1320 rep->r_flags |= R_SENT; 1321 nmp->nm_sent += NFS_CWNDSCALE; 1322 } 1323 rep->r_rtt = 0; 1324 } 1325 } 1326 } 1327 1328 #ifdef NFSSERVER 1329 /* 1330 * Call the nqnfs server timer once a second to handle leases. 1331 */ 1332 if (lasttime != time.tv_sec) { 1333 lasttime = time.tv_sec; 1334 nqnfs_serverd(); 1335 } 1336 1337 /* 1338 * Scan the write gathering queues for writes that need to be 1339 * completed now. 1340 */ 1341 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1342 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 1343 slp = slp->ns_chain.tqe_next) { 1344 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec) 1345 nfsrv_wakenfsd(slp); 1346 } 1347 #endif /* NFSSERVER */ 1348 splx(s); 1349 timeout(nfs_timer, (void *)0, nfs_ticks); 1350 } 1351 1352 /* 1353 * Test for a termination condition pending on the process. 1354 * This is used for NFSMNT_INT mounts. 1355 */ 1356 int 1357 nfs_sigintr(nmp, rep, p) 1358 struct nfsmount *nmp; 1359 struct nfsreq *rep; 1360 register struct proc *p; 1361 { 1362 1363 if (rep && (rep->r_flags & R_SOFTTERM)) 1364 return (EINTR); 1365 if (!(nmp->nm_flag & NFSMNT_INT)) 1366 return (0); 1367 if (p && p->p_siglist && 1368 (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) & 1369 NFSINT_SIGMASK)) 1370 return (EINTR); 1371 return (0); 1372 } 1373 1374 /* 1375 * Lock a socket against others. 1376 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1377 * and also to avoid race conditions between the processes with nfs requests 1378 * in progress when a reconnect is necessary. 1379 */ 1380 int 1381 nfs_sndlock(flagp, rep) 1382 register int *flagp; 1383 struct nfsreq *rep; 1384 { 1385 struct proc *p; 1386 int slpflag = 0, slptimeo = 0; 1387 1388 if (rep) { 1389 p = rep->r_procp; 1390 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1391 slpflag = PCATCH; 1392 } else 1393 p = (struct proc *)0; 1394 while (*flagp & NFSMNT_SNDLOCK) { 1395 if (nfs_sigintr(rep->r_nmp, rep, p)) 1396 return (EINTR); 1397 *flagp |= NFSMNT_WANTSND; 1398 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1399 slptimeo); 1400 if (slpflag == PCATCH) { 1401 slpflag = 0; 1402 slptimeo = 2 * hz; 1403 } 1404 } 1405 *flagp |= NFSMNT_SNDLOCK; 1406 return (0); 1407 } 1408 1409 /* 1410 * Unlock the stream socket for others. 1411 */ 1412 void 1413 nfs_sndunlock(flagp) 1414 register int *flagp; 1415 { 1416 1417 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1418 panic("nfs sndunlock"); 1419 *flagp &= ~NFSMNT_SNDLOCK; 1420 if (*flagp & NFSMNT_WANTSND) { 1421 *flagp &= ~NFSMNT_WANTSND; 1422 wakeup((caddr_t)flagp); 1423 } 1424 } 1425 1426 int 1427 nfs_rcvlock(rep) 1428 register struct nfsreq *rep; 1429 { 1430 register int *flagp = &rep->r_nmp->nm_flag; 1431 int slpflag, slptimeo = 0; 1432 1433 if (*flagp & NFSMNT_INT) 1434 slpflag = PCATCH; 1435 else 1436 slpflag = 0; 1437 while (*flagp & NFSMNT_RCVLOCK) { 1438 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1439 return (EINTR); 1440 *flagp |= NFSMNT_WANTRCV; 1441 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1442 slptimeo); 1443 /* If our reply was received while we were sleeping, 1444 * then just return without taking the lock to avoid a 1445 * situation where a single iod could 'capture' the 1446 * receive lock. 1447 */ 1448 if (rep->r_mrep != NULL) 1449 return (EALREADY); 1450 if (slpflag == PCATCH) { 1451 slpflag = 0; 1452 slptimeo = 2 * hz; 1453 } 1454 } 1455 *flagp |= NFSMNT_RCVLOCK; 1456 return (0); 1457 } 1458 1459 /* 1460 * Unlock the stream socket for others. 1461 */ 1462 void 1463 nfs_rcvunlock(flagp) 1464 register int *flagp; 1465 { 1466 1467 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1468 panic("nfs rcvunlock"); 1469 *flagp &= ~NFSMNT_RCVLOCK; 1470 if (*flagp & NFSMNT_WANTRCV) { 1471 *flagp &= ~NFSMNT_WANTRCV; 1472 wakeup((caddr_t)flagp); 1473 } 1474 } 1475 1476 /* 1477 * Parse an RPC request 1478 * - verify it 1479 * - fill in the cred struct. 1480 */ 1481 int 1482 nfs_getreq(nd, nfsd, has_header) 1483 register struct nfsrv_descript *nd; 1484 struct nfsd *nfsd; 1485 int has_header; 1486 { 1487 register int len, i; 1488 register u_int32_t *tl; 1489 register int32_t t1; 1490 struct uio uio; 1491 struct iovec iov; 1492 caddr_t dpos, cp2, cp; 1493 u_int32_t nfsvers, auth_type; 1494 uid_t nickuid; 1495 int error = 0, nqnfs = 0, ticklen; 1496 struct mbuf *mrep, *md; 1497 register struct nfsuid *nuidp; 1498 struct timeval tvin, tvout; 1499 1500 mrep = nd->nd_mrep; 1501 md = nd->nd_md; 1502 dpos = nd->nd_dpos; 1503 if (has_header) { 1504 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1505 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1506 if (*tl++ != rpc_call) { 1507 m_freem(mrep); 1508 return (EBADRPC); 1509 } 1510 } else 1511 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1512 nd->nd_repstat = 0; 1513 nd->nd_flag = 0; 1514 if (*tl++ != rpc_vers) { 1515 nd->nd_repstat = ERPCMISMATCH; 1516 nd->nd_procnum = NFSPROC_NOOP; 1517 return (0); 1518 } 1519 if (*tl != nfs_prog) { 1520 if (*tl == nqnfs_prog) 1521 nqnfs++; 1522 else { 1523 nd->nd_repstat = EPROGUNAVAIL; 1524 nd->nd_procnum = NFSPROC_NOOP; 1525 return (0); 1526 } 1527 } 1528 tl++; 1529 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1530 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1531 (nfsvers != NQNFS_VER3 && nqnfs)) { 1532 nd->nd_repstat = EPROGMISMATCH; 1533 nd->nd_procnum = NFSPROC_NOOP; 1534 return (0); 1535 } 1536 if (nqnfs) 1537 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1538 else if (nfsvers == NFS_VER3) 1539 nd->nd_flag = ND_NFSV3; 1540 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1541 if (nd->nd_procnum == NFSPROC_NULL) 1542 return (0); 1543 if (nd->nd_procnum >= NFS_NPROCS || 1544 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1545 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1546 nd->nd_repstat = EPROCUNAVAIL; 1547 nd->nd_procnum = NFSPROC_NOOP; 1548 return (0); 1549 } 1550 if ((nd->nd_flag & ND_NFSV3) == 0) 1551 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1552 auth_type = *tl++; 1553 len = fxdr_unsigned(int, *tl++); 1554 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1555 m_freem(mrep); 1556 return (EBADRPC); 1557 } 1558 1559 nd->nd_flag &= ~ND_KERBAUTH; 1560 /* 1561 * Handle auth_unix or auth_kerb. 1562 */ 1563 if (auth_type == rpc_auth_unix) { 1564 len = fxdr_unsigned(int, *++tl); 1565 if (len < 0 || len > NFS_MAXNAMLEN) { 1566 m_freem(mrep); 1567 return (EBADRPC); 1568 } 1569 nfsm_adv(nfsm_rndup(len)); 1570 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1571 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred)); 1572 nd->nd_cr.cr_ref = 1; 1573 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1574 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1575 len = fxdr_unsigned(int, *tl); 1576 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1577 m_freem(mrep); 1578 return (EBADRPC); 1579 } 1580 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1581 for (i = 0; i < len; i++) 1582 if (i < NGROUPS) 1583 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1584 else 1585 tl++; 1586 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1587 if (nd->nd_cr.cr_ngroups > 1) 1588 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1589 len = fxdr_unsigned(int, *++tl); 1590 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1591 m_freem(mrep); 1592 return (EBADRPC); 1593 } 1594 if (len > 0) 1595 nfsm_adv(nfsm_rndup(len)); 1596 } else if (auth_type == rpc_auth_kerb) { 1597 switch (fxdr_unsigned(int, *tl++)) { 1598 case RPCAKN_FULLNAME: 1599 ticklen = fxdr_unsigned(int, *tl); 1600 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1601 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1602 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1603 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1604 m_freem(mrep); 1605 return (EBADRPC); 1606 } 1607 uio.uio_offset = 0; 1608 uio.uio_iov = &iov; 1609 uio.uio_iovcnt = 1; 1610 uio.uio_segflg = UIO_SYSSPACE; 1611 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1612 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1613 nfsm_mtouio(&uio, uio.uio_resid); 1614 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1615 if (*tl++ != rpc_auth_kerb || 1616 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1617 printf("Bad kerb verifier\n"); 1618 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1619 nd->nd_procnum = NFSPROC_NOOP; 1620 return (0); 1621 } 1622 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1623 tl = (u_int32_t *)cp; 1624 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1625 printf("Not fullname kerb verifier\n"); 1626 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1627 nd->nd_procnum = NFSPROC_NOOP; 1628 return (0); 1629 } 1630 cp += NFSX_UNSIGNED; 1631 bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED); 1632 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1633 nd->nd_flag |= ND_KERBFULL; 1634 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1635 break; 1636 case RPCAKN_NICKNAME: 1637 if (len != 2 * NFSX_UNSIGNED) { 1638 printf("Kerb nickname short\n"); 1639 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1640 nd->nd_procnum = NFSPROC_NOOP; 1641 return (0); 1642 } 1643 nickuid = fxdr_unsigned(uid_t, *tl); 1644 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1645 if (*tl++ != rpc_auth_kerb || 1646 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1647 printf("Kerb nick verifier bad\n"); 1648 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1649 nd->nd_procnum = NFSPROC_NOOP; 1650 return (0); 1651 } 1652 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1653 tvin.tv_sec = *tl++; 1654 tvin.tv_usec = *tl; 1655 1656 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; 1657 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1658 if (nuidp->nu_cr.cr_uid == nickuid && 1659 (!nd->nd_nam2 || 1660 netaddr_match(NU_NETFAM(nuidp), 1661 &nuidp->nu_haddr, nd->nd_nam2))) 1662 break; 1663 } 1664 if (!nuidp) { 1665 nd->nd_repstat = 1666 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1667 nd->nd_procnum = NFSPROC_NOOP; 1668 return (0); 1669 } 1670 1671 /* 1672 * Now, decrypt the timestamp using the session key 1673 * and validate it. 1674 */ 1675 #ifdef NFSKERB 1676 XXX 1677 #endif 1678 1679 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1680 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1681 if (nuidp->nu_expire < time.tv_sec || 1682 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1683 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1684 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1685 nuidp->nu_expire = 0; 1686 nd->nd_repstat = 1687 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1688 nd->nd_procnum = NFSPROC_NOOP; 1689 return (0); 1690 } 1691 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1692 nd->nd_flag |= ND_KERBNICK; 1693 }; 1694 } else { 1695 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1696 nd->nd_procnum = NFSPROC_NOOP; 1697 return (0); 1698 } 1699 1700 /* 1701 * For nqnfs, get piggybacked lease request. 1702 */ 1703 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1704 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1705 nd->nd_flag |= fxdr_unsigned(int, *tl); 1706 if (nd->nd_flag & ND_LEASE) { 1707 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1708 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 1709 } else 1710 nd->nd_duration = NQ_MINLEASE; 1711 } else 1712 nd->nd_duration = NQ_MINLEASE; 1713 nd->nd_md = md; 1714 nd->nd_dpos = dpos; 1715 return (0); 1716 nfsmout: 1717 return (error); 1718 } 1719 1720 int 1721 nfs_msg(p, server, msg) 1722 struct proc *p; 1723 char *server, *msg; 1724 { 1725 tpr_t tpr; 1726 1727 if (p) 1728 tpr = tprintf_open(p); 1729 else 1730 tpr = NULL; 1731 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1732 tprintf_close(tpr); 1733 return (0); 1734 } 1735 1736 #ifdef NFSSERVER 1737 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 1738 struct nfssvc_sock *, struct proc *, 1739 struct mbuf **)) = { 1740 nfsrv_null, 1741 nfsrv_getattr, 1742 nfsrv_setattr, 1743 nfsrv_lookup, 1744 nfsrv3_access, 1745 nfsrv_readlink, 1746 nfsrv_read, 1747 nfsrv_write, 1748 nfsrv_create, 1749 nfsrv_mkdir, 1750 nfsrv_symlink, 1751 nfsrv_mknod, 1752 nfsrv_remove, 1753 nfsrv_rmdir, 1754 nfsrv_rename, 1755 nfsrv_link, 1756 nfsrv_readdir, 1757 nfsrv_readdirplus, 1758 nfsrv_statfs, 1759 nfsrv_fsinfo, 1760 nfsrv_pathconf, 1761 nfsrv_commit, 1762 nqnfsrv_getlease, 1763 nqnfsrv_vacated, 1764 nfsrv_noop, 1765 nfsrv_noop 1766 }; 1767 1768 /* 1769 * Socket upcall routine for the nfsd sockets. 1770 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1771 * Essentially do as much as possible non-blocking, else punt and it will 1772 * be called with M_WAIT from an nfsd. 1773 */ 1774 void 1775 nfsrv_rcv(so, arg, waitflag) 1776 struct socket *so; 1777 caddr_t arg; 1778 int waitflag; 1779 { 1780 register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1781 register struct mbuf *m; 1782 struct mbuf *mp, *nam; 1783 struct uio auio; 1784 int flags, error; 1785 1786 if ((slp->ns_flag & SLP_VALID) == 0) 1787 return; 1788 #ifdef notdef 1789 /* 1790 * Define this to test for nfsds handling this under heavy load. 1791 */ 1792 if (waitflag == M_DONTWAIT) { 1793 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1794 } 1795 #endif 1796 auio.uio_procp = NULL; 1797 if (so->so_type == SOCK_STREAM) { 1798 /* 1799 * If there are already records on the queue, defer soreceive() 1800 * to an nfsd so that there is feedback to the TCP layer that 1801 * the nfs servers are heavily loaded. 1802 */ 1803 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1804 slp->ns_flag |= SLP_NEEDQ; 1805 goto dorecs; 1806 } 1807 1808 /* 1809 * Do soreceive(). 1810 */ 1811 auio.uio_resid = 1000000000; 1812 flags = MSG_DONTWAIT; 1813 error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1814 if (error || mp == (struct mbuf *)0) { 1815 if (error == EWOULDBLOCK) 1816 slp->ns_flag |= SLP_NEEDQ; 1817 else 1818 slp->ns_flag |= SLP_DISCONN; 1819 goto dorecs; 1820 } 1821 m = mp; 1822 if (slp->ns_rawend) { 1823 slp->ns_rawend->m_next = m; 1824 slp->ns_cc += 1000000000 - auio.uio_resid; 1825 } else { 1826 slp->ns_raw = m; 1827 slp->ns_cc = 1000000000 - auio.uio_resid; 1828 } 1829 while (m->m_next) 1830 m = m->m_next; 1831 slp->ns_rawend = m; 1832 1833 /* 1834 * Now try and parse record(s) out of the raw stream data. 1835 */ 1836 error = nfsrv_getstream(slp, waitflag); 1837 if (error) { 1838 if (error == EPERM) 1839 slp->ns_flag |= SLP_DISCONN; 1840 else 1841 slp->ns_flag |= SLP_NEEDQ; 1842 } 1843 } else { 1844 do { 1845 auio.uio_resid = 1000000000; 1846 flags = MSG_DONTWAIT; 1847 error = soreceive(so, &nam, &auio, &mp, 1848 (struct mbuf **)0, &flags); 1849 if (mp) { 1850 if (nam) { 1851 m = nam; 1852 m->m_next = mp; 1853 } else 1854 m = mp; 1855 if (slp->ns_recend) 1856 slp->ns_recend->m_nextpkt = m; 1857 else 1858 slp->ns_rec = m; 1859 slp->ns_recend = m; 1860 m->m_nextpkt = (struct mbuf *)0; 1861 } 1862 if (error) { 1863 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1864 && error != EWOULDBLOCK) { 1865 slp->ns_flag |= SLP_DISCONN; 1866 goto dorecs; 1867 } 1868 } 1869 } while (mp); 1870 } 1871 1872 /* 1873 * Now try and process the request records, non-blocking. 1874 */ 1875 dorecs: 1876 if (waitflag == M_DONTWAIT && 1877 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1878 nfsrv_wakenfsd(slp); 1879 } 1880 1881 /* 1882 * Try and extract an RPC request from the mbuf data list received on a 1883 * stream socket. The "waitflag" argument indicates whether or not it 1884 * can sleep. 1885 */ 1886 int 1887 nfsrv_getstream(slp, waitflag) 1888 register struct nfssvc_sock *slp; 1889 int waitflag; 1890 { 1891 register struct mbuf *m, **mpp; 1892 register char *cp1, *cp2; 1893 register int len; 1894 struct mbuf *om, *m2, *recm = NULL; 1895 u_int32_t recmark; 1896 1897 if (slp->ns_flag & SLP_GETSTREAM) 1898 panic("nfs getstream"); 1899 slp->ns_flag |= SLP_GETSTREAM; 1900 for (;;) { 1901 if (slp->ns_reclen == 0) { 1902 if (slp->ns_cc < NFSX_UNSIGNED) { 1903 slp->ns_flag &= ~SLP_GETSTREAM; 1904 return (0); 1905 } 1906 m = slp->ns_raw; 1907 if (m->m_len >= NFSX_UNSIGNED) { 1908 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 1909 m->m_data += NFSX_UNSIGNED; 1910 m->m_len -= NFSX_UNSIGNED; 1911 } else { 1912 cp1 = (caddr_t)&recmark; 1913 cp2 = mtod(m, caddr_t); 1914 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1915 while (m->m_len == 0) { 1916 m = m->m_next; 1917 cp2 = mtod(m, caddr_t); 1918 } 1919 *cp1++ = *cp2++; 1920 m->m_data++; 1921 m->m_len--; 1922 } 1923 } 1924 slp->ns_cc -= NFSX_UNSIGNED; 1925 recmark = ntohl(recmark); 1926 slp->ns_reclen = recmark & ~0x80000000; 1927 if (recmark & 0x80000000) 1928 slp->ns_flag |= SLP_LASTFRAG; 1929 else 1930 slp->ns_flag &= ~SLP_LASTFRAG; 1931 if (slp->ns_reclen > NFS_MAXPACKET) { 1932 slp->ns_flag &= ~SLP_GETSTREAM; 1933 return (EPERM); 1934 } 1935 } 1936 1937 /* 1938 * Now get the record part. 1939 */ 1940 if (slp->ns_cc == slp->ns_reclen) { 1941 recm = slp->ns_raw; 1942 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 1943 slp->ns_cc = slp->ns_reclen = 0; 1944 } else if (slp->ns_cc > slp->ns_reclen) { 1945 len = 0; 1946 m = slp->ns_raw; 1947 om = (struct mbuf *)0; 1948 while (len < slp->ns_reclen) { 1949 if ((len + m->m_len) > slp->ns_reclen) { 1950 size_t left = slp->ns_reclen - len; 1951 1952 MGETHDR(m2, waitflag, m->m_type); 1953 if (m2 == NULL) { 1954 slp->ns_flag &= ~SLP_GETSTREAM; 1955 return (EWOULDBLOCK); 1956 } 1957 if (left > MHLEN) { 1958 MCLGET(m2, waitflag); 1959 if (!(m2->m_flags & M_EXT)) { 1960 m_freem(m2); 1961 slp->ns_flag &= ~SLP_GETSTREAM; 1962 return (EWOULDBLOCK); 1963 } 1964 } 1965 bcopy(mtod(m, caddr_t), mtod(m2, caddr_t), 1966 left); 1967 m2->m_len = left; 1968 m->m_data += left; 1969 m->m_len -= left; 1970 if (om) { 1971 om->m_next = m2; 1972 recm = slp->ns_raw; 1973 } else 1974 recm = m2; 1975 len = slp->ns_reclen; 1976 } else if ((len + m->m_len) == slp->ns_reclen) { 1977 om = m; 1978 len += m->m_len; 1979 m = m->m_next; 1980 recm = slp->ns_raw; 1981 om->m_next = (struct mbuf *)0; 1982 } else { 1983 om = m; 1984 len += m->m_len; 1985 m = m->m_next; 1986 } 1987 } 1988 slp->ns_raw = m; 1989 slp->ns_cc -= len; 1990 slp->ns_reclen = 0; 1991 } else { 1992 slp->ns_flag &= ~SLP_GETSTREAM; 1993 return (0); 1994 } 1995 1996 /* 1997 * Accumulate the fragments into a record. 1998 */ 1999 mpp = &slp->ns_frag; 2000 while (*mpp) 2001 mpp = &((*mpp)->m_next); 2002 *mpp = recm; 2003 if (slp->ns_flag & SLP_LASTFRAG) { 2004 if (slp->ns_recend) 2005 slp->ns_recend->m_nextpkt = slp->ns_frag; 2006 else 2007 slp->ns_rec = slp->ns_frag; 2008 slp->ns_recend = slp->ns_frag; 2009 slp->ns_frag = (struct mbuf *)0; 2010 } 2011 } 2012 } 2013 2014 /* 2015 * Parse an RPC header. 2016 */ 2017 int 2018 nfsrv_dorec(slp, nfsd, ndp) 2019 register struct nfssvc_sock *slp; 2020 struct nfsd *nfsd; 2021 struct nfsrv_descript **ndp; 2022 { 2023 register struct mbuf *m, *nam; 2024 register struct nfsrv_descript *nd; 2025 int error; 2026 2027 *ndp = NULL; 2028 if ((slp->ns_flag & SLP_VALID) == 0 || 2029 (m = slp->ns_rec) == (struct mbuf *)0) 2030 return (ENOBUFS); 2031 slp->ns_rec = m->m_nextpkt; 2032 if (slp->ns_rec) 2033 m->m_nextpkt = (struct mbuf *)0; 2034 else 2035 slp->ns_recend = (struct mbuf *)0; 2036 if (m->m_type == MT_SONAME) { 2037 nam = m; 2038 m = m->m_next; 2039 nam->m_next = NULL; 2040 } else 2041 nam = NULL; 2042 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), 2043 M_NFSRVDESC, M_WAITOK); 2044 nd->nd_md = nd->nd_mrep = m; 2045 nd->nd_nam2 = nam; 2046 nd->nd_dpos = mtod(m, caddr_t); 2047 error = nfs_getreq(nd, nfsd, TRUE); 2048 if (error) { 2049 m_freem(nam); 2050 free((caddr_t)nd, M_NFSRVDESC); 2051 return (error); 2052 } 2053 *ndp = nd; 2054 nfsd->nfsd_nd = nd; 2055 return (0); 2056 } 2057 2058 2059 /* 2060 * Search for a sleeping nfsd and wake it up. 2061 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2062 * running nfsds will go look for the work in the nfssvc_sock list. 2063 */ 2064 void 2065 nfsrv_wakenfsd(slp) 2066 struct nfssvc_sock *slp; 2067 { 2068 register struct nfsd *nd; 2069 2070 if ((slp->ns_flag & SLP_VALID) == 0) 2071 return; 2072 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) { 2073 if (nd->nfsd_flag & NFSD_WAITING) { 2074 nd->nfsd_flag &= ~NFSD_WAITING; 2075 if (nd->nfsd_slp) 2076 panic("nfsd wakeup"); 2077 slp->ns_sref++; 2078 nd->nfsd_slp = slp; 2079 wakeup((caddr_t)nd); 2080 return; 2081 } 2082 } 2083 slp->ns_flag |= SLP_DOREC; 2084 nfsd_head_flag |= NFSD_CHECKSLP; 2085 } 2086 #endif /* NFSSERVER */ 2087
Indexes created Sun Sep 21 20:09:37 GMT 2025