nfs_socket.c revision 1.37
1 /* $NetBSD: nfs_socket.c,v 1.37 1997/05/12 23:40:22 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) * 2; 232 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 233 NFS_MAXPKTHDR) * 2; 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 else 585 nfs_sndunlock(&rep->r_nmp->nm_flag); 586 } 587 } else { 588 if ((so = rep->r_nmp->nm_so) == NULL) 589 return (EACCES); 590 if (so->so_state & SS_ISCONNECTED) 591 getnam = (struct mbuf **)0; 592 else 593 getnam = aname; 594 auio.uio_resid = len = 1000000; 595 auio.uio_procp = p; 596 do { 597 rcvflg = 0; 598 error = soreceive(so, getnam, &auio, mp, 599 (struct mbuf **)0, &rcvflg); 600 if (error == EWOULDBLOCK && 601 (rep->r_flags & R_SOFTTERM)) 602 return (EINTR); 603 } while (error == EWOULDBLOCK); 604 len -= auio.uio_resid; 605 } 606 if (error) { 607 m_freem(*mp); 608 *mp = (struct mbuf *)0; 609 } 610 return (error); 611 } 612 613 /* 614 * Implement receipt of reply on a socket. 615 * We must search through the list of received datagrams matching them 616 * with outstanding requests using the xid, until ours is found. 617 */ 618 /* ARGSUSED */ 619 int 620 nfs_reply(myrep) 621 struct nfsreq *myrep; 622 { 623 register struct nfsreq *rep; 624 register struct nfsmount *nmp = myrep->r_nmp; 625 register int32_t t1; 626 struct mbuf *mrep, *nam, *md; 627 u_int32_t rxid, *tl; 628 caddr_t dpos, cp2; 629 int error; 630 631 /* 632 * Loop around until we get our own reply 633 */ 634 for (;;) { 635 /* 636 * Lock against other receivers so that I don't get stuck in 637 * sbwait() after someone else has received my reply for me. 638 * Also necessary for connection based protocols to avoid 639 * race conditions during a reconnect. 640 */ 641 error = nfs_rcvlock(myrep); 642 if (error == EALREADY) 643 return (0); 644 if (error) 645 return (error); 646 /* 647 * Get the next Rpc reply off the socket 648 */ 649 error = nfs_receive(myrep, &nam, &mrep); 650 nfs_rcvunlock(&nmp->nm_flag); 651 if (error) { 652 653 /* 654 * Ignore routing errors on connectionless protocols?? 655 */ 656 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 657 nmp->nm_so->so_error = 0; 658 #ifdef DEBUG 659 printf("nfs_reply: ignoring error %d\n", error); 660 #endif 661 if (myrep->r_flags & R_GETONEREP) 662 return (0); 663 continue; 664 } 665 return (error); 666 } 667 if (nam) 668 m_freem(nam); 669 670 /* 671 * Get the xid and check that it is an rpc reply 672 */ 673 md = mrep; 674 dpos = mtod(md, caddr_t); 675 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 676 rxid = *tl++; 677 if (*tl != rpc_reply) { 678 if (nmp->nm_flag & NFSMNT_NQNFS) { 679 if (nqnfs_callback(nmp, mrep, md, dpos)) 680 nfsstats.rpcinvalid++; 681 } else { 682 nfsstats.rpcinvalid++; 683 m_freem(mrep); 684 } 685 nfsmout: 686 if (myrep->r_flags & R_GETONEREP) 687 return (0); 688 continue; 689 } 690 691 /* 692 * Loop through the request list to match up the reply 693 * Iff no match, just drop the datagram 694 */ 695 for (rep = nfs_reqq.tqh_first; rep != 0; 696 rep = rep->r_chain.tqe_next) { 697 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 698 /* Found it.. */ 699 rep->r_mrep = mrep; 700 rep->r_md = md; 701 rep->r_dpos = dpos; 702 if (nfsrtton) { 703 struct rttl *rt; 704 705 rt = &nfsrtt.rttl[nfsrtt.pos]; 706 rt->proc = rep->r_procnum; 707 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 708 rt->sent = nmp->nm_sent; 709 rt->cwnd = nmp->nm_cwnd; 710 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 711 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 712 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 713 rt->tstamp = time; 714 if (rep->r_flags & R_TIMING) 715 rt->rtt = rep->r_rtt; 716 else 717 rt->rtt = 1000000; 718 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 719 } 720 /* 721 * Update congestion window. 722 * Do the additive increase of 723 * one rpc/rtt. 724 */ 725 if (nmp->nm_cwnd <= nmp->nm_sent) { 726 nmp->nm_cwnd += 727 (NFS_CWNDSCALE * NFS_CWNDSCALE + 728 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 729 if (nmp->nm_cwnd > NFS_MAXCWND) 730 nmp->nm_cwnd = NFS_MAXCWND; 731 } 732 rep->r_flags &= ~R_SENT; 733 nmp->nm_sent -= NFS_CWNDSCALE; 734 /* 735 * Update rtt using a gain of 0.125 on the mean 736 * and a gain of 0.25 on the deviation. 737 */ 738 if (rep->r_flags & R_TIMING) { 739 /* 740 * Since the timer resolution of 741 * NFS_HZ is so course, it can often 742 * result in r_rtt == 0. Since 743 * r_rtt == N means that the actual 744 * rtt is between N+dt and N+2-dt ticks, 745 * add 1. 746 */ 747 t1 = rep->r_rtt + 1; 748 t1 -= (NFS_SRTT(rep) >> 3); 749 NFS_SRTT(rep) += t1; 750 if (t1 < 0) 751 t1 = -t1; 752 t1 -= (NFS_SDRTT(rep) >> 2); 753 NFS_SDRTT(rep) += t1; 754 } 755 nmp->nm_timeouts = 0; 756 break; 757 } 758 } 759 /* 760 * If not matched to a request, drop it. 761 * If it's mine, get out. 762 */ 763 if (rep == 0) { 764 nfsstats.rpcunexpected++; 765 m_freem(mrep); 766 } else if (rep == myrep) { 767 if (rep->r_mrep == NULL) 768 panic("nfsreply nil"); 769 return (0); 770 } 771 if (myrep->r_flags & R_GETONEREP) 772 return (0); 773 } 774 } 775 776 /* 777 * nfs_request - goes something like this 778 * - fill in request struct 779 * - links it into list 780 * - calls nfs_send() for first transmit 781 * - calls nfs_receive() to get reply 782 * - break down rpc header and return with nfs reply pointed to 783 * by mrep or error 784 * nb: always frees up mreq mbuf list 785 */ 786 int 787 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 788 struct vnode *vp; 789 struct mbuf *mrest; 790 int procnum; 791 struct proc *procp; 792 struct ucred *cred; 793 struct mbuf **mrp; 794 struct mbuf **mdp; 795 caddr_t *dposp; 796 { 797 register struct mbuf *m, *mrep; 798 register struct nfsreq *rep; 799 register u_int32_t *tl; 800 register int i; 801 struct nfsmount *nmp; 802 struct mbuf *md, *mheadend; 803 struct nfsnode *np; 804 char nickv[RPCX_NICKVERF]; 805 time_t reqtime, waituntil; 806 caddr_t dpos, cp2; 807 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; 808 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 809 int verf_len, verf_type; 810 u_int32_t xid; 811 u_quad_t frev; 812 char *auth_str, *verf_str; 813 NFSKERBKEY_T key; /* save session key */ 814 815 nmp = VFSTONFS(vp->v_mount); 816 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 817 rep->r_nmp = nmp; 818 rep->r_vp = vp; 819 rep->r_procp = procp; 820 rep->r_procnum = procnum; 821 i = 0; 822 m = mrest; 823 while (m) { 824 i += m->m_len; 825 m = m->m_next; 826 } 827 mrest_len = i; 828 829 /* 830 * Get the RPC header with authorization. 831 */ 832 kerbauth: 833 verf_str = auth_str = (char *)0; 834 if (nmp->nm_flag & NFSMNT_KERB) { 835 verf_str = nickv; 836 verf_len = sizeof (nickv); 837 auth_type = RPCAUTH_KERB4; 838 bzero((caddr_t)key, sizeof (key)); 839 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 840 &auth_len, verf_str, verf_len)) { 841 error = nfs_getauth(nmp, rep, cred, &auth_str, 842 &auth_len, verf_str, &verf_len, key); 843 if (error) { 844 free((caddr_t)rep, M_NFSREQ); 845 m_freem(mrest); 846 return (error); 847 } 848 } 849 } else { 850 auth_type = RPCAUTH_UNIX; 851 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 852 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 853 5 * NFSX_UNSIGNED; 854 } 855 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 856 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 857 if (auth_str) 858 free(auth_str, M_TEMP); 859 860 /* 861 * For stream protocols, insert a Sun RPC Record Mark. 862 */ 863 if (nmp->nm_sotype == SOCK_STREAM) { 864 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 865 *mtod(m, u_int32_t *) = htonl(0x80000000 | 866 (m->m_pkthdr.len - NFSX_UNSIGNED)); 867 } 868 rep->r_mreq = m; 869 rep->r_xid = xid; 870 tryagain: 871 if (nmp->nm_flag & NFSMNT_SOFT) 872 rep->r_retry = nmp->nm_retry; 873 else 874 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 875 rep->r_rtt = rep->r_rexmit = 0; 876 if (proct[procnum] > 0) 877 rep->r_flags = R_TIMING; 878 else 879 rep->r_flags = 0; 880 rep->r_mrep = NULL; 881 882 /* 883 * Do the client side RPC. 884 */ 885 nfsstats.rpcrequests++; 886 /* 887 * Chain request into list of outstanding requests. Be sure 888 * to put it LAST so timer finds oldest requests first. 889 */ 890 s = splsoftnet(); 891 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 892 893 /* Get send time for nqnfs */ 894 reqtime = time.tv_sec; 895 896 /* 897 * If backing off another request or avoiding congestion, don't 898 * send this one now but let timer do it. If not timing a request, 899 * do it now. 900 */ 901 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 902 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 903 nmp->nm_sent < nmp->nm_cwnd)) { 904 splx(s); 905 if (nmp->nm_soflags & PR_CONNREQUIRED) 906 error = nfs_sndlock(&nmp->nm_flag, rep); 907 if (!error) { 908 m = m_copym(m, 0, M_COPYALL, M_WAIT); 909 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 910 if (nmp->nm_soflags & PR_CONNREQUIRED) 911 nfs_sndunlock(&nmp->nm_flag); 912 } 913 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 914 nmp->nm_sent += NFS_CWNDSCALE; 915 rep->r_flags |= R_SENT; 916 } 917 } else { 918 splx(s); 919 rep->r_rtt = -1; 920 } 921 922 /* 923 * Wait for the reply from our send or the timer's. 924 */ 925 if (!error || error == EPIPE) 926 error = nfs_reply(rep); 927 928 /* 929 * RPC done, unlink the request. 930 */ 931 s = splsoftnet(); 932 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 933 splx(s); 934 935 /* 936 * Decrement the outstanding request count. 937 */ 938 if (rep->r_flags & R_SENT) { 939 rep->r_flags &= ~R_SENT; /* paranoia */ 940 nmp->nm_sent -= NFS_CWNDSCALE; 941 } 942 943 /* 944 * If there was a successful reply and a tprintf msg. 945 * tprintf a response. 946 */ 947 if (!error && (rep->r_flags & R_TPRINTFMSG)) 948 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 949 "is alive again"); 950 mrep = rep->r_mrep; 951 md = rep->r_md; 952 dpos = rep->r_dpos; 953 if (error) { 954 m_freem(rep->r_mreq); 955 free((caddr_t)rep, M_NFSREQ); 956 return (error); 957 } 958 959 /* 960 * break down the rpc header and check if ok 961 */ 962 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 963 if (*tl++ == rpc_msgdenied) { 964 if (*tl == rpc_mismatch) 965 error = EOPNOTSUPP; 966 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 967 if (!failed_auth) { 968 failed_auth++; 969 mheadend->m_next = (struct mbuf *)0; 970 m_freem(mrep); 971 m_freem(rep->r_mreq); 972 goto kerbauth; 973 } else 974 error = EAUTH; 975 } else 976 error = EACCES; 977 m_freem(mrep); 978 m_freem(rep->r_mreq); 979 free((caddr_t)rep, M_NFSREQ); 980 return (error); 981 } 982 983 /* 984 * Grab any Kerberos verifier, otherwise just throw it away. 985 */ 986 verf_type = fxdr_unsigned(int, *tl++); 987 i = fxdr_unsigned(int32_t, *tl); 988 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 989 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 990 if (error) 991 goto nfsmout; 992 } else if (i > 0) 993 nfsm_adv(nfsm_rndup(i)); 994 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 995 /* 0 == ok */ 996 if (*tl == 0) { 997 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 998 if (*tl != 0) { 999 error = fxdr_unsigned(int, *tl); 1000 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1001 error == NFSERR_TRYLATER) { 1002 m_freem(mrep); 1003 error = 0; 1004 waituntil = time.tv_sec + trylater_delay; 1005 while (time.tv_sec < waituntil) 1006 (void) tsleep((caddr_t)&lbolt, 1007 PSOCK, "nqnfstry", 0); 1008 trylater_delay *= nfs_backoff[trylater_cnt]; 1009 if (trylater_cnt < 7) 1010 trylater_cnt++; 1011 goto tryagain; 1012 } 1013 1014 /* 1015 * If the File Handle was stale, invalidate the 1016 * lookup cache, just in case. 1017 */ 1018 if (error == ESTALE) 1019 cache_purge(vp); 1020 if (nmp->nm_flag & NFSMNT_NFSV3) { 1021 *mrp = mrep; 1022 *mdp = md; 1023 *dposp = dpos; 1024 error |= NFSERR_RETERR; 1025 } else 1026 m_freem(mrep); 1027 m_freem(rep->r_mreq); 1028 free((caddr_t)rep, M_NFSREQ); 1029 return (error); 1030 } 1031 1032 /* 1033 * For nqnfs, get any lease in reply 1034 */ 1035 if (nmp->nm_flag & NFSMNT_NQNFS) { 1036 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1037 if (*tl) { 1038 np = VTONFS(vp); 1039 nqlflag = fxdr_unsigned(int, *tl); 1040 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1041 cachable = fxdr_unsigned(int, *tl++); 1042 reqtime += fxdr_unsigned(int, *tl++); 1043 if (reqtime > time.tv_sec) { 1044 fxdr_hyper(tl, &frev); 1045 nqnfs_clientlease(nmp, np, nqlflag, 1046 cachable, reqtime, frev); 1047 } 1048 } 1049 } 1050 *mrp = mrep; 1051 *mdp = md; 1052 *dposp = dpos; 1053 m_freem(rep->r_mreq); 1054 FREE((caddr_t)rep, M_NFSREQ); 1055 return (0); 1056 } 1057 m_freem(mrep); 1058 error = EPROTONOSUPPORT; 1059 nfsmout: 1060 m_freem(rep->r_mreq); 1061 free((caddr_t)rep, M_NFSREQ); 1062 return (error); 1063 } 1064 #endif /* NFS */ 1065 1066 /* 1067 * Generate the rpc reply header 1068 * siz arg. is used to decide if adding a cluster is worthwhile 1069 */ 1070 int 1071 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1072 int siz; 1073 struct nfsrv_descript *nd; 1074 struct nfssvc_sock *slp; 1075 int err; 1076 int cache; 1077 u_quad_t *frev; 1078 struct mbuf **mrq; 1079 struct mbuf **mbp; 1080 caddr_t *bposp; 1081 { 1082 register u_int32_t *tl; 1083 register struct mbuf *mreq; 1084 caddr_t bpos; 1085 struct mbuf *mb, *mb2; 1086 1087 MGETHDR(mreq, M_WAIT, MT_DATA); 1088 mb = mreq; 1089 /* 1090 * If this is a big reply, use a cluster else 1091 * try and leave leading space for the lower level headers. 1092 */ 1093 siz += RPC_REPLYSIZ; 1094 if (siz >= MINCLSIZE) { 1095 MCLGET(mreq, M_WAIT); 1096 } else 1097 mreq->m_data += max_hdr; 1098 tl = mtod(mreq, u_int32_t *); 1099 mreq->m_len = 6 * NFSX_UNSIGNED; 1100 bpos = ((caddr_t)tl) + mreq->m_len; 1101 *tl++ = txdr_unsigned(nd->nd_retxid); 1102 *tl++ = rpc_reply; 1103 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1104 *tl++ = rpc_msgdenied; 1105 if (err & NFSERR_AUTHERR) { 1106 *tl++ = rpc_autherr; 1107 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1108 mreq->m_len -= NFSX_UNSIGNED; 1109 bpos -= NFSX_UNSIGNED; 1110 } else { 1111 *tl++ = rpc_mismatch; 1112 *tl++ = txdr_unsigned(RPC_VER2); 1113 *tl = txdr_unsigned(RPC_VER2); 1114 } 1115 } else { 1116 *tl++ = rpc_msgaccepted; 1117 1118 /* 1119 * For Kerberos authentication, we must send the nickname 1120 * verifier back, otherwise just RPCAUTH_NULL. 1121 */ 1122 if (nd->nd_flag & ND_KERBFULL) { 1123 register struct nfsuid *nuidp; 1124 struct timeval ktvin, ktvout; 1125 1126 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; 1127 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1128 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1129 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), 1130 &nuidp->nu_haddr, nd->nd_nam2))) 1131 break; 1132 } 1133 if (nuidp) { 1134 ktvin.tv_sec = 1135 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); 1136 ktvin.tv_usec = 1137 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1138 1139 /* 1140 * Encrypt the timestamp in ecb mode using the 1141 * session key. 1142 */ 1143 #ifdef NFSKERB 1144 XXX 1145 #endif 1146 1147 *tl++ = rpc_auth_kerb; 1148 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1149 *tl = ktvout.tv_sec; 1150 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1151 *tl++ = ktvout.tv_usec; 1152 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1153 } else { 1154 *tl++ = 0; 1155 *tl++ = 0; 1156 } 1157 } else { 1158 *tl++ = 0; 1159 *tl++ = 0; 1160 } 1161 switch (err) { 1162 case EPROGUNAVAIL: 1163 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1164 break; 1165 case EPROGMISMATCH: 1166 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1167 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1168 if (nd->nd_flag & ND_NQNFS) { 1169 *tl++ = txdr_unsigned(3); 1170 *tl = txdr_unsigned(3); 1171 } else { 1172 *tl++ = txdr_unsigned(2); 1173 *tl = txdr_unsigned(3); 1174 } 1175 break; 1176 case EPROCUNAVAIL: 1177 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1178 break; 1179 case EBADRPC: 1180 *tl = txdr_unsigned(RPC_GARBAGE); 1181 break; 1182 default: 1183 *tl = 0; 1184 if (err != NFSERR_RETVOID) { 1185 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1186 if (err) 1187 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1188 else 1189 *tl = 0; 1190 } 1191 break; 1192 }; 1193 } 1194 1195 /* 1196 * For nqnfs, piggyback lease as requested. 1197 */ 1198 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1199 if (nd->nd_flag & ND_LEASE) { 1200 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1201 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1202 *tl++ = txdr_unsigned(cache); 1203 *tl++ = txdr_unsigned(nd->nd_duration); 1204 txdr_hyper(frev, tl); 1205 } else { 1206 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1207 *tl = 0; 1208 } 1209 } 1210 if (mrq != NULL) 1211 *mrq = mreq; 1212 *mbp = mb; 1213 *bposp = bpos; 1214 if (err != 0 && err != NFSERR_RETVOID) 1215 nfsstats.srvrpc_errs++; 1216 return (0); 1217 } 1218 1219 /* 1220 * Nfs timer routine 1221 * Scan the nfsreq list and retranmit any requests that have timed out 1222 * To avoid retransmission attempts on STREAM sockets (in the future) make 1223 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1224 */ 1225 void 1226 nfs_timer(arg) 1227 void *arg; /* never used */ 1228 { 1229 register struct nfsreq *rep; 1230 register struct mbuf *m; 1231 register struct socket *so; 1232 register struct nfsmount *nmp; 1233 register int timeo; 1234 int s, error; 1235 #ifdef NFSSERVER 1236 register struct nfssvc_sock *slp; 1237 static long lasttime = 0; 1238 u_quad_t cur_usec; 1239 #endif 1240 1241 s = splsoftnet(); 1242 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1243 nmp = rep->r_nmp; 1244 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1245 continue; 1246 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1247 rep->r_flags |= R_SOFTTERM; 1248 continue; 1249 } 1250 if (rep->r_rtt >= 0) { 1251 rep->r_rtt++; 1252 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1253 timeo = nmp->nm_timeo; 1254 else 1255 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1256 if (nmp->nm_timeouts > 0) 1257 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1258 if (rep->r_rtt <= timeo) 1259 continue; 1260 if (nmp->nm_timeouts < 8) 1261 nmp->nm_timeouts++; 1262 } 1263 /* 1264 * Check for server not responding 1265 */ 1266 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1267 rep->r_rexmit > nmp->nm_deadthresh) { 1268 nfs_msg(rep->r_procp, 1269 nmp->nm_mountp->mnt_stat.f_mntfromname, 1270 "not responding"); 1271 rep->r_flags |= R_TPRINTFMSG; 1272 } 1273 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1274 nfsstats.rpctimeouts++; 1275 rep->r_flags |= R_SOFTTERM; 1276 continue; 1277 } 1278 if (nmp->nm_sotype != SOCK_DGRAM) { 1279 if (++rep->r_rexmit > NFS_MAXREXMIT) 1280 rep->r_rexmit = NFS_MAXREXMIT; 1281 continue; 1282 } 1283 if ((so = nmp->nm_so) == NULL) 1284 continue; 1285 1286 /* 1287 * If there is enough space and the window allows.. 1288 * Resend it 1289 * Set r_rtt to -1 in case we fail to send it now. 1290 */ 1291 rep->r_rtt = -1; 1292 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1293 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1294 (rep->r_flags & R_SENT) || 1295 nmp->nm_sent < nmp->nm_cwnd) && 1296 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1297 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1298 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1299 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1300 else 1301 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1302 nmp->nm_nam, (struct mbuf *)0, (struct proc *)0); 1303 if (error) { 1304 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1305 #ifdef DEBUG 1306 printf("nfs_timer: ignoring error %d\n", 1307 error); 1308 #endif 1309 so->so_error = 0; 1310 } 1311 } else { 1312 /* 1313 * Iff first send, start timing 1314 * else turn timing off, backoff timer 1315 * and divide congestion window by 2. 1316 */ 1317 if (rep->r_flags & R_SENT) { 1318 rep->r_flags &= ~R_TIMING; 1319 if (++rep->r_rexmit > NFS_MAXREXMIT) 1320 rep->r_rexmit = NFS_MAXREXMIT; 1321 nmp->nm_cwnd >>= 1; 1322 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1323 nmp->nm_cwnd = NFS_CWNDSCALE; 1324 nfsstats.rpcretries++; 1325 } else { 1326 rep->r_flags |= R_SENT; 1327 nmp->nm_sent += NFS_CWNDSCALE; 1328 } 1329 rep->r_rtt = 0; 1330 } 1331 } 1332 } 1333 1334 #ifdef NFSSERVER 1335 /* 1336 * Call the nqnfs server timer once a second to handle leases. 1337 */ 1338 if (lasttime != time.tv_sec) { 1339 lasttime = time.tv_sec; 1340 nqnfs_serverd(); 1341 } 1342 1343 /* 1344 * Scan the write gathering queues for writes that need to be 1345 * completed now. 1346 */ 1347 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1348 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 1349 slp = slp->ns_chain.tqe_next) { 1350 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec) 1351 nfsrv_wakenfsd(slp); 1352 } 1353 #endif /* NFSSERVER */ 1354 splx(s); 1355 timeout(nfs_timer, (void *)0, nfs_ticks); 1356 } 1357 1358 /* 1359 * Test for a termination condition pending on the process. 1360 * This is used for NFSMNT_INT mounts. 1361 */ 1362 int 1363 nfs_sigintr(nmp, rep, p) 1364 struct nfsmount *nmp; 1365 struct nfsreq *rep; 1366 register struct proc *p; 1367 { 1368 1369 if (rep && (rep->r_flags & R_SOFTTERM)) 1370 return (EINTR); 1371 if (!(nmp->nm_flag & NFSMNT_INT)) 1372 return (0); 1373 if (p && p->p_siglist && 1374 (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) & 1375 NFSINT_SIGMASK)) 1376 return (EINTR); 1377 return (0); 1378 } 1379 1380 /* 1381 * Lock a socket against others. 1382 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1383 * and also to avoid race conditions between the processes with nfs requests 1384 * in progress when a reconnect is necessary. 1385 */ 1386 int 1387 nfs_sndlock(flagp, rep) 1388 register int *flagp; 1389 struct nfsreq *rep; 1390 { 1391 struct proc *p; 1392 int slpflag = 0, slptimeo = 0; 1393 1394 if (rep) { 1395 p = rep->r_procp; 1396 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1397 slpflag = PCATCH; 1398 } else 1399 p = (struct proc *)0; 1400 while (*flagp & NFSMNT_SNDLOCK) { 1401 if (nfs_sigintr(rep->r_nmp, rep, p)) 1402 return (EINTR); 1403 *flagp |= NFSMNT_WANTSND; 1404 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1405 slptimeo); 1406 if (slpflag == PCATCH) { 1407 slpflag = 0; 1408 slptimeo = 2 * hz; 1409 } 1410 } 1411 *flagp |= NFSMNT_SNDLOCK; 1412 return (0); 1413 } 1414 1415 /* 1416 * Unlock the stream socket for others. 1417 */ 1418 void 1419 nfs_sndunlock(flagp) 1420 register int *flagp; 1421 { 1422 1423 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1424 panic("nfs sndunlock"); 1425 *flagp &= ~NFSMNT_SNDLOCK; 1426 if (*flagp & NFSMNT_WANTSND) { 1427 *flagp &= ~NFSMNT_WANTSND; 1428 wakeup((caddr_t)flagp); 1429 } 1430 } 1431 1432 int 1433 nfs_rcvlock(rep) 1434 register struct nfsreq *rep; 1435 { 1436 register int *flagp = &rep->r_nmp->nm_flag; 1437 int slpflag, slptimeo = 0; 1438 1439 if (*flagp & NFSMNT_INT) 1440 slpflag = PCATCH; 1441 else 1442 slpflag = 0; 1443 while (*flagp & NFSMNT_RCVLOCK) { 1444 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1445 return (EINTR); 1446 *flagp |= NFSMNT_WANTRCV; 1447 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1448 slptimeo); 1449 /* If our reply was received while we were sleeping, 1450 * then just return without taking the lock to avoid a 1451 * situation where a single iod could 'capture' the 1452 * receive lock. 1453 */ 1454 if (rep->r_mrep != NULL) 1455 return (EALREADY); 1456 if (slpflag == PCATCH) { 1457 slpflag = 0; 1458 slptimeo = 2 * hz; 1459 } 1460 } 1461 *flagp |= NFSMNT_RCVLOCK; 1462 return (0); 1463 } 1464 1465 /* 1466 * Unlock the stream socket for others. 1467 */ 1468 void 1469 nfs_rcvunlock(flagp) 1470 register int *flagp; 1471 { 1472 1473 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1474 panic("nfs rcvunlock"); 1475 *flagp &= ~NFSMNT_RCVLOCK; 1476 if (*flagp & NFSMNT_WANTRCV) { 1477 *flagp &= ~NFSMNT_WANTRCV; 1478 wakeup((caddr_t)flagp); 1479 } 1480 } 1481 1482 /* 1483 * Parse an RPC request 1484 * - verify it 1485 * - fill in the cred struct. 1486 */ 1487 int 1488 nfs_getreq(nd, nfsd, has_header) 1489 register struct nfsrv_descript *nd; 1490 struct nfsd *nfsd; 1491 int has_header; 1492 { 1493 register int len, i; 1494 register u_int32_t *tl; 1495 register int32_t t1; 1496 struct uio uio; 1497 struct iovec iov; 1498 caddr_t dpos, cp2, cp; 1499 u_int32_t nfsvers, auth_type; 1500 uid_t nickuid; 1501 int error = 0, nqnfs = 0, ticklen; 1502 struct mbuf *mrep, *md; 1503 register struct nfsuid *nuidp; 1504 struct timeval tvin, tvout; 1505 1506 mrep = nd->nd_mrep; 1507 md = nd->nd_md; 1508 dpos = nd->nd_dpos; 1509 if (has_header) { 1510 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1511 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1512 if (*tl++ != rpc_call) { 1513 m_freem(mrep); 1514 return (EBADRPC); 1515 } 1516 } else 1517 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1518 nd->nd_repstat = 0; 1519 nd->nd_flag = 0; 1520 if (*tl++ != rpc_vers) { 1521 nd->nd_repstat = ERPCMISMATCH; 1522 nd->nd_procnum = NFSPROC_NOOP; 1523 return (0); 1524 } 1525 if (*tl != nfs_prog) { 1526 if (*tl == nqnfs_prog) 1527 nqnfs++; 1528 else { 1529 nd->nd_repstat = EPROGUNAVAIL; 1530 nd->nd_procnum = NFSPROC_NOOP; 1531 return (0); 1532 } 1533 } 1534 tl++; 1535 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1536 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1537 (nfsvers != NQNFS_VER3 && nqnfs)) { 1538 nd->nd_repstat = EPROGMISMATCH; 1539 nd->nd_procnum = NFSPROC_NOOP; 1540 return (0); 1541 } 1542 if (nqnfs) 1543 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1544 else if (nfsvers == NFS_VER3) 1545 nd->nd_flag = ND_NFSV3; 1546 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1547 if (nd->nd_procnum == NFSPROC_NULL) 1548 return (0); 1549 if (nd->nd_procnum >= NFS_NPROCS || 1550 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1551 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1552 nd->nd_repstat = EPROCUNAVAIL; 1553 nd->nd_procnum = NFSPROC_NOOP; 1554 return (0); 1555 } 1556 if ((nd->nd_flag & ND_NFSV3) == 0) 1557 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1558 auth_type = *tl++; 1559 len = fxdr_unsigned(int, *tl++); 1560 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1561 m_freem(mrep); 1562 return (EBADRPC); 1563 } 1564 1565 nd->nd_flag &= ~ND_KERBAUTH; 1566 /* 1567 * Handle auth_unix or auth_kerb. 1568 */ 1569 if (auth_type == rpc_auth_unix) { 1570 len = fxdr_unsigned(int, *++tl); 1571 if (len < 0 || len > NFS_MAXNAMLEN) { 1572 m_freem(mrep); 1573 return (EBADRPC); 1574 } 1575 nfsm_adv(nfsm_rndup(len)); 1576 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1577 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred)); 1578 nd->nd_cr.cr_ref = 1; 1579 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1580 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1581 len = fxdr_unsigned(int, *tl); 1582 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1583 m_freem(mrep); 1584 return (EBADRPC); 1585 } 1586 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1587 for (i = 0; i < len; i++) 1588 if (i < NGROUPS) 1589 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1590 else 1591 tl++; 1592 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1593 if (nd->nd_cr.cr_ngroups > 1) 1594 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1595 len = fxdr_unsigned(int, *++tl); 1596 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1597 m_freem(mrep); 1598 return (EBADRPC); 1599 } 1600 if (len > 0) 1601 nfsm_adv(nfsm_rndup(len)); 1602 } else if (auth_type == rpc_auth_kerb) { 1603 switch (fxdr_unsigned(int, *tl++)) { 1604 case RPCAKN_FULLNAME: 1605 ticklen = fxdr_unsigned(int, *tl); 1606 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1607 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1608 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1609 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1610 m_freem(mrep); 1611 return (EBADRPC); 1612 } 1613 uio.uio_offset = 0; 1614 uio.uio_iov = &iov; 1615 uio.uio_iovcnt = 1; 1616 uio.uio_segflg = UIO_SYSSPACE; 1617 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1618 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1619 nfsm_mtouio(&uio, uio.uio_resid); 1620 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1621 if (*tl++ != rpc_auth_kerb || 1622 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1623 printf("Bad kerb verifier\n"); 1624 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1625 nd->nd_procnum = NFSPROC_NOOP; 1626 return (0); 1627 } 1628 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1629 tl = (u_int32_t *)cp; 1630 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1631 printf("Not fullname kerb verifier\n"); 1632 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1633 nd->nd_procnum = NFSPROC_NOOP; 1634 return (0); 1635 } 1636 cp += NFSX_UNSIGNED; 1637 bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED); 1638 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1639 nd->nd_flag |= ND_KERBFULL; 1640 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1641 break; 1642 case RPCAKN_NICKNAME: 1643 if (len != 2 * NFSX_UNSIGNED) { 1644 printf("Kerb nickname short\n"); 1645 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1646 nd->nd_procnum = NFSPROC_NOOP; 1647 return (0); 1648 } 1649 nickuid = fxdr_unsigned(uid_t, *tl); 1650 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1651 if (*tl++ != rpc_auth_kerb || 1652 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1653 printf("Kerb nick verifier bad\n"); 1654 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1655 nd->nd_procnum = NFSPROC_NOOP; 1656 return (0); 1657 } 1658 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1659 tvin.tv_sec = *tl++; 1660 tvin.tv_usec = *tl; 1661 1662 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; 1663 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1664 if (nuidp->nu_cr.cr_uid == nickuid && 1665 (!nd->nd_nam2 || 1666 netaddr_match(NU_NETFAM(nuidp), 1667 &nuidp->nu_haddr, nd->nd_nam2))) 1668 break; 1669 } 1670 if (!nuidp) { 1671 nd->nd_repstat = 1672 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1673 nd->nd_procnum = NFSPROC_NOOP; 1674 return (0); 1675 } 1676 1677 /* 1678 * Now, decrypt the timestamp using the session key 1679 * and validate it. 1680 */ 1681 #ifdef NFSKERB 1682 XXX 1683 #endif 1684 1685 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1686 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1687 if (nuidp->nu_expire < time.tv_sec || 1688 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1689 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1690 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1691 nuidp->nu_expire = 0; 1692 nd->nd_repstat = 1693 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1694 nd->nd_procnum = NFSPROC_NOOP; 1695 return (0); 1696 } 1697 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1698 nd->nd_flag |= ND_KERBNICK; 1699 }; 1700 } else { 1701 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1702 nd->nd_procnum = NFSPROC_NOOP; 1703 return (0); 1704 } 1705 1706 /* 1707 * For nqnfs, get piggybacked lease request. 1708 */ 1709 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1710 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1711 nd->nd_flag |= fxdr_unsigned(int, *tl); 1712 if (nd->nd_flag & ND_LEASE) { 1713 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1714 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 1715 } else 1716 nd->nd_duration = NQ_MINLEASE; 1717 } else 1718 nd->nd_duration = NQ_MINLEASE; 1719 nd->nd_md = md; 1720 nd->nd_dpos = dpos; 1721 return (0); 1722 nfsmout: 1723 return (error); 1724 } 1725 1726 int 1727 nfs_msg(p, server, msg) 1728 struct proc *p; 1729 char *server, *msg; 1730 { 1731 tpr_t tpr; 1732 1733 if (p) 1734 tpr = tprintf_open(p); 1735 else 1736 tpr = NULL; 1737 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1738 tprintf_close(tpr); 1739 return (0); 1740 } 1741 1742 #ifdef NFSSERVER 1743 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 1744 struct nfssvc_sock *, struct proc *, 1745 struct mbuf **)) = { 1746 nfsrv_null, 1747 nfsrv_getattr, 1748 nfsrv_setattr, 1749 nfsrv_lookup, 1750 nfsrv3_access, 1751 nfsrv_readlink, 1752 nfsrv_read, 1753 nfsrv_write, 1754 nfsrv_create, 1755 nfsrv_mkdir, 1756 nfsrv_symlink, 1757 nfsrv_mknod, 1758 nfsrv_remove, 1759 nfsrv_rmdir, 1760 nfsrv_rename, 1761 nfsrv_link, 1762 nfsrv_readdir, 1763 nfsrv_readdirplus, 1764 nfsrv_statfs, 1765 nfsrv_fsinfo, 1766 nfsrv_pathconf, 1767 nfsrv_commit, 1768 nqnfsrv_getlease, 1769 nqnfsrv_vacated, 1770 nfsrv_noop, 1771 nfsrv_noop 1772 }; 1773 1774 /* 1775 * Socket upcall routine for the nfsd sockets. 1776 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1777 * Essentially do as much as possible non-blocking, else punt and it will 1778 * be called with M_WAIT from an nfsd. 1779 */ 1780 void 1781 nfsrv_rcv(so, arg, waitflag) 1782 struct socket *so; 1783 caddr_t arg; 1784 int waitflag; 1785 { 1786 register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1787 register struct mbuf *m; 1788 struct mbuf *mp, *nam; 1789 struct uio auio; 1790 int flags, error; 1791 1792 if ((slp->ns_flag & SLP_VALID) == 0) 1793 return; 1794 #ifdef notdef 1795 /* 1796 * Define this to test for nfsds handling this under heavy load. 1797 */ 1798 if (waitflag == M_DONTWAIT) { 1799 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1800 } 1801 #endif 1802 auio.uio_procp = NULL; 1803 if (so->so_type == SOCK_STREAM) { 1804 /* 1805 * If there are already records on the queue, defer soreceive() 1806 * to an nfsd so that there is feedback to the TCP layer that 1807 * the nfs servers are heavily loaded. 1808 */ 1809 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1810 slp->ns_flag |= SLP_NEEDQ; 1811 goto dorecs; 1812 } 1813 1814 /* 1815 * Do soreceive(). 1816 */ 1817 auio.uio_resid = 1000000000; 1818 flags = MSG_DONTWAIT; 1819 error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1820 if (error || mp == (struct mbuf *)0) { 1821 if (error == EWOULDBLOCK) 1822 slp->ns_flag |= SLP_NEEDQ; 1823 else 1824 slp->ns_flag |= SLP_DISCONN; 1825 goto dorecs; 1826 } 1827 m = mp; 1828 if (slp->ns_rawend) { 1829 slp->ns_rawend->m_next = m; 1830 slp->ns_cc += 1000000000 - auio.uio_resid; 1831 } else { 1832 slp->ns_raw = m; 1833 slp->ns_cc = 1000000000 - auio.uio_resid; 1834 } 1835 while (m->m_next) 1836 m = m->m_next; 1837 slp->ns_rawend = m; 1838 1839 /* 1840 * Now try and parse record(s) out of the raw stream data. 1841 */ 1842 error = nfsrv_getstream(slp, waitflag); 1843 if (error) { 1844 if (error == EPERM) 1845 slp->ns_flag |= SLP_DISCONN; 1846 else 1847 slp->ns_flag |= SLP_NEEDQ; 1848 } 1849 } else { 1850 do { 1851 auio.uio_resid = 1000000000; 1852 flags = MSG_DONTWAIT; 1853 error = soreceive(so, &nam, &auio, &mp, 1854 (struct mbuf **)0, &flags); 1855 if (mp) { 1856 if (nam) { 1857 m = nam; 1858 m->m_next = mp; 1859 } else 1860 m = mp; 1861 if (slp->ns_recend) 1862 slp->ns_recend->m_nextpkt = m; 1863 else 1864 slp->ns_rec = m; 1865 slp->ns_recend = m; 1866 m->m_nextpkt = (struct mbuf *)0; 1867 } 1868 if (error) { 1869 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1870 && error != EWOULDBLOCK) { 1871 slp->ns_flag |= SLP_DISCONN; 1872 goto dorecs; 1873 } 1874 } 1875 } while (mp); 1876 } 1877 1878 /* 1879 * Now try and process the request records, non-blocking. 1880 */ 1881 dorecs: 1882 if (waitflag == M_DONTWAIT && 1883 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1884 nfsrv_wakenfsd(slp); 1885 } 1886 1887 /* 1888 * Try and extract an RPC request from the mbuf data list received on a 1889 * stream socket. The "waitflag" argument indicates whether or not it 1890 * can sleep. 1891 */ 1892 int 1893 nfsrv_getstream(slp, waitflag) 1894 register struct nfssvc_sock *slp; 1895 int waitflag; 1896 { 1897 register struct mbuf *m, **mpp; 1898 register char *cp1, *cp2; 1899 register int len; 1900 struct mbuf *om, *m2, *recm = NULL; 1901 u_int32_t recmark; 1902 1903 if (slp->ns_flag & SLP_GETSTREAM) 1904 panic("nfs getstream"); 1905 slp->ns_flag |= SLP_GETSTREAM; 1906 for (;;) { 1907 if (slp->ns_reclen == 0) { 1908 if (slp->ns_cc < NFSX_UNSIGNED) { 1909 slp->ns_flag &= ~SLP_GETSTREAM; 1910 return (0); 1911 } 1912 m = slp->ns_raw; 1913 if (m->m_len >= NFSX_UNSIGNED) { 1914 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 1915 m->m_data += NFSX_UNSIGNED; 1916 m->m_len -= NFSX_UNSIGNED; 1917 } else { 1918 cp1 = (caddr_t)&recmark; 1919 cp2 = mtod(m, caddr_t); 1920 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1921 while (m->m_len == 0) { 1922 m = m->m_next; 1923 cp2 = mtod(m, caddr_t); 1924 } 1925 *cp1++ = *cp2++; 1926 m->m_data++; 1927 m->m_len--; 1928 } 1929 } 1930 slp->ns_cc -= NFSX_UNSIGNED; 1931 recmark = ntohl(recmark); 1932 slp->ns_reclen = recmark & ~0x80000000; 1933 if (recmark & 0x80000000) 1934 slp->ns_flag |= SLP_LASTFRAG; 1935 else 1936 slp->ns_flag &= ~SLP_LASTFRAG; 1937 if (slp->ns_reclen > NFS_MAXPACKET) { 1938 slp->ns_flag &= ~SLP_GETSTREAM; 1939 return (EPERM); 1940 } 1941 } 1942 1943 /* 1944 * Now get the record part. 1945 */ 1946 if (slp->ns_cc == slp->ns_reclen) { 1947 recm = slp->ns_raw; 1948 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 1949 slp->ns_cc = slp->ns_reclen = 0; 1950 } else if (slp->ns_cc > slp->ns_reclen) { 1951 len = 0; 1952 m = slp->ns_raw; 1953 om = (struct mbuf *)0; 1954 while (len < slp->ns_reclen) { 1955 if ((len + m->m_len) > slp->ns_reclen) { 1956 size_t left = slp->ns_reclen - len; 1957 1958 MGETHDR(m2, waitflag, m->m_type); 1959 if (m2 == NULL) { 1960 slp->ns_flag &= ~SLP_GETSTREAM; 1961 return (EWOULDBLOCK); 1962 } 1963 if (left > MHLEN) { 1964 MCLGET(m2, waitflag); 1965 if (!(m2->m_flags & M_EXT)) { 1966 m_freem(m2); 1967 slp->ns_flag &= ~SLP_GETSTREAM; 1968 return (EWOULDBLOCK); 1969 } 1970 } 1971 bcopy(mtod(m, caddr_t), mtod(m2, caddr_t), 1972 left); 1973 m2->m_len = left; 1974 m->m_data += left; 1975 m->m_len -= left; 1976 if (om) { 1977 om->m_next = m2; 1978 recm = slp->ns_raw; 1979 } else 1980 recm = m2; 1981 len = slp->ns_reclen; 1982 } else if ((len + m->m_len) == slp->ns_reclen) { 1983 om = m; 1984 len += m->m_len; 1985 m = m->m_next; 1986 recm = slp->ns_raw; 1987 om->m_next = (struct mbuf *)0; 1988 } else { 1989 om = m; 1990 len += m->m_len; 1991 m = m->m_next; 1992 } 1993 } 1994 slp->ns_raw = m; 1995 slp->ns_cc -= len; 1996 slp->ns_reclen = 0; 1997 } else { 1998 slp->ns_flag &= ~SLP_GETSTREAM; 1999 return (0); 2000 } 2001 2002 /* 2003 * Accumulate the fragments into a record. 2004 */ 2005 mpp = &slp->ns_frag; 2006 while (*mpp) 2007 mpp = &((*mpp)->m_next); 2008 *mpp = recm; 2009 if (slp->ns_flag & SLP_LASTFRAG) { 2010 if (slp->ns_recend) 2011 slp->ns_recend->m_nextpkt = slp->ns_frag; 2012 else 2013 slp->ns_rec = slp->ns_frag; 2014 slp->ns_recend = slp->ns_frag; 2015 slp->ns_frag = (struct mbuf *)0; 2016 } 2017 } 2018 } 2019 2020 /* 2021 * Parse an RPC header. 2022 */ 2023 int 2024 nfsrv_dorec(slp, nfsd, ndp) 2025 register struct nfssvc_sock *slp; 2026 struct nfsd *nfsd; 2027 struct nfsrv_descript **ndp; 2028 { 2029 register struct mbuf *m, *nam; 2030 register struct nfsrv_descript *nd; 2031 int error; 2032 2033 *ndp = NULL; 2034 if ((slp->ns_flag & SLP_VALID) == 0 || 2035 (m = slp->ns_rec) == (struct mbuf *)0) 2036 return (ENOBUFS); 2037 slp->ns_rec = m->m_nextpkt; 2038 if (slp->ns_rec) 2039 m->m_nextpkt = (struct mbuf *)0; 2040 else 2041 slp->ns_recend = (struct mbuf *)0; 2042 if (m->m_type == MT_SONAME) { 2043 nam = m; 2044 m = m->m_next; 2045 nam->m_next = NULL; 2046 } else 2047 nam = NULL; 2048 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), 2049 M_NFSRVDESC, M_WAITOK); 2050 nd->nd_md = nd->nd_mrep = m; 2051 nd->nd_nam2 = nam; 2052 nd->nd_dpos = mtod(m, caddr_t); 2053 error = nfs_getreq(nd, nfsd, TRUE); 2054 if (error) { 2055 m_freem(nam); 2056 free((caddr_t)nd, M_NFSRVDESC); 2057 return (error); 2058 } 2059 *ndp = nd; 2060 nfsd->nfsd_nd = nd; 2061 return (0); 2062 } 2063 2064 2065 /* 2066 * Search for a sleeping nfsd and wake it up. 2067 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2068 * running nfsds will go look for the work in the nfssvc_sock list. 2069 */ 2070 void 2071 nfsrv_wakenfsd(slp) 2072 struct nfssvc_sock *slp; 2073 { 2074 register struct nfsd *nd; 2075 2076 if ((slp->ns_flag & SLP_VALID) == 0) 2077 return; 2078 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) { 2079 if (nd->nfsd_flag & NFSD_WAITING) { 2080 nd->nfsd_flag &= ~NFSD_WAITING; 2081 if (nd->nfsd_slp) 2082 panic("nfsd wakeup"); 2083 slp->ns_sref++; 2084 nd->nfsd_slp = slp; 2085 wakeup((caddr_t)nd); 2086 return; 2087 } 2088 } 2089 slp->ns_flag |= SLP_DOREC; 2090 nfsd_head_flag |= NFSD_CHECKSLP; 2091 } 2092 #endif /* NFSSERVER */ 2093
Indexes created Sun Sep 21 20:09:37 GMT 2025