nfs_socket.c revision 1.35
1 /* $NetBSD: nfs_socket.c,v 1.35 1997/02/22 02:59:08 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) 641 return (error); 642 /* Already received, bye bye */ 643 if (myrep->r_mrep != NULL) { 644 nfs_rcvunlock(&nmp->nm_flag); 645 return (0); 646 } 647 /* 648 * Get the next Rpc reply off the socket 649 */ 650 error = nfs_receive(myrep, &nam, &mrep); 651 nfs_rcvunlock(&nmp->nm_flag); 652 if (error) { 653 654 /* 655 * Ignore routing errors on connectionless protocols?? 656 */ 657 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 658 nmp->nm_so->so_error = 0; 659 printf("nfs_reply: ignoring error %d\n", error); 660 if (myrep->r_flags & R_GETONEREP) 661 return (0); 662 continue; 663 } 664 return (error); 665 } 666 if (nam) 667 m_freem(nam); 668 669 /* 670 * Get the xid and check that it is an rpc reply 671 */ 672 md = mrep; 673 dpos = mtod(md, caddr_t); 674 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 675 rxid = *tl++; 676 if (*tl != rpc_reply) { 677 if (nmp->nm_flag & NFSMNT_NQNFS) { 678 if (nqnfs_callback(nmp, mrep, md, dpos)) 679 nfsstats.rpcinvalid++; 680 } else { 681 nfsstats.rpcinvalid++; 682 m_freem(mrep); 683 } 684 nfsmout: 685 if (myrep->r_flags & R_GETONEREP) 686 return (0); 687 continue; 688 } 689 690 /* 691 * Loop through the request list to match up the reply 692 * Iff no match, just drop the datagram 693 */ 694 for (rep = nfs_reqq.tqh_first; rep != 0; 695 rep = rep->r_chain.tqe_next) { 696 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 697 /* Found it.. */ 698 rep->r_mrep = mrep; 699 rep->r_md = md; 700 rep->r_dpos = dpos; 701 if (nfsrtton) { 702 struct rttl *rt; 703 704 rt = &nfsrtt.rttl[nfsrtt.pos]; 705 rt->proc = rep->r_procnum; 706 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 707 rt->sent = nmp->nm_sent; 708 rt->cwnd = nmp->nm_cwnd; 709 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 710 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 711 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 712 rt->tstamp = time; 713 if (rep->r_flags & R_TIMING) 714 rt->rtt = rep->r_rtt; 715 else 716 rt->rtt = 1000000; 717 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 718 } 719 /* 720 * Update congestion window. 721 * Do the additive increase of 722 * one rpc/rtt. 723 */ 724 if (nmp->nm_cwnd <= nmp->nm_sent) { 725 nmp->nm_cwnd += 726 (NFS_CWNDSCALE * NFS_CWNDSCALE + 727 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 728 if (nmp->nm_cwnd > NFS_MAXCWND) 729 nmp->nm_cwnd = NFS_MAXCWND; 730 } 731 rep->r_flags &= ~R_SENT; 732 nmp->nm_sent -= NFS_CWNDSCALE; 733 /* 734 * Update rtt using a gain of 0.125 on the mean 735 * and a gain of 0.25 on the deviation. 736 */ 737 if (rep->r_flags & R_TIMING) { 738 /* 739 * Since the timer resolution of 740 * NFS_HZ is so course, it can often 741 * result in r_rtt == 0. Since 742 * r_rtt == N means that the actual 743 * rtt is between N+dt and N+2-dt ticks, 744 * add 1. 745 */ 746 t1 = rep->r_rtt + 1; 747 t1 -= (NFS_SRTT(rep) >> 3); 748 NFS_SRTT(rep) += t1; 749 if (t1 < 0) 750 t1 = -t1; 751 t1 -= (NFS_SDRTT(rep) >> 2); 752 NFS_SDRTT(rep) += t1; 753 } 754 nmp->nm_timeouts = 0; 755 break; 756 } 757 } 758 /* 759 * If not matched to a request, drop it. 760 * If it's mine, get out. 761 */ 762 if (rep == 0) { 763 nfsstats.rpcunexpected++; 764 m_freem(mrep); 765 } else if (rep == myrep) { 766 if (rep->r_mrep == NULL) 767 panic("nfsreply nil"); 768 return (0); 769 } 770 if (myrep->r_flags & R_GETONEREP) 771 return (0); 772 } 773 } 774 775 /* 776 * nfs_request - goes something like this 777 * - fill in request struct 778 * - links it into list 779 * - calls nfs_send() for first transmit 780 * - calls nfs_receive() to get reply 781 * - break down rpc header and return with nfs reply pointed to 782 * by mrep or error 783 * nb: always frees up mreq mbuf list 784 */ 785 int 786 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 787 struct vnode *vp; 788 struct mbuf *mrest; 789 int procnum; 790 struct proc *procp; 791 struct ucred *cred; 792 struct mbuf **mrp; 793 struct mbuf **mdp; 794 caddr_t *dposp; 795 { 796 register struct mbuf *m, *mrep; 797 register struct nfsreq *rep; 798 register u_int32_t *tl; 799 register int i; 800 struct nfsmount *nmp; 801 struct mbuf *md, *mheadend; 802 struct nfsnode *np; 803 char nickv[RPCX_NICKVERF]; 804 time_t reqtime, waituntil; 805 caddr_t dpos, cp2; 806 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; 807 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 808 int verf_len, verf_type; 809 u_int32_t xid; 810 u_quad_t frev; 811 char *auth_str, *verf_str; 812 NFSKERBKEY_T key; /* save session key */ 813 814 nmp = VFSTONFS(vp->v_mount); 815 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 816 rep->r_nmp = nmp; 817 rep->r_vp = vp; 818 rep->r_procp = procp; 819 rep->r_procnum = procnum; 820 i = 0; 821 m = mrest; 822 while (m) { 823 i += m->m_len; 824 m = m->m_next; 825 } 826 mrest_len = i; 827 828 /* 829 * Get the RPC header with authorization. 830 */ 831 kerbauth: 832 verf_str = auth_str = (char *)0; 833 if (nmp->nm_flag & NFSMNT_KERB) { 834 verf_str = nickv; 835 verf_len = sizeof (nickv); 836 auth_type = RPCAUTH_KERB4; 837 bzero((caddr_t)key, sizeof (key)); 838 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 839 &auth_len, verf_str, verf_len)) { 840 error = nfs_getauth(nmp, rep, cred, &auth_str, 841 &auth_len, verf_str, &verf_len, key); 842 if (error) { 843 free((caddr_t)rep, M_NFSREQ); 844 m_freem(mrest); 845 return (error); 846 } 847 } 848 } else { 849 auth_type = RPCAUTH_UNIX; 850 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 851 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 852 5 * NFSX_UNSIGNED; 853 } 854 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 855 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 856 if (auth_str) 857 free(auth_str, M_TEMP); 858 859 /* 860 * For stream protocols, insert a Sun RPC Record Mark. 861 */ 862 if (nmp->nm_sotype == SOCK_STREAM) { 863 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 864 *mtod(m, u_int32_t *) = htonl(0x80000000 | 865 (m->m_pkthdr.len - NFSX_UNSIGNED)); 866 } 867 rep->r_mreq = m; 868 rep->r_xid = xid; 869 tryagain: 870 if (nmp->nm_flag & NFSMNT_SOFT) 871 rep->r_retry = nmp->nm_retry; 872 else 873 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 874 rep->r_rtt = rep->r_rexmit = 0; 875 if (proct[procnum] > 0) 876 rep->r_flags = R_TIMING; 877 else 878 rep->r_flags = 0; 879 rep->r_mrep = NULL; 880 881 /* 882 * Do the client side RPC. 883 */ 884 nfsstats.rpcrequests++; 885 /* 886 * Chain request into list of outstanding requests. Be sure 887 * to put it LAST so timer finds oldest requests first. 888 */ 889 s = splsoftnet(); 890 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 891 892 /* Get send time for nqnfs */ 893 reqtime = time.tv_sec; 894 895 /* 896 * If backing off another request or avoiding congestion, don't 897 * send this one now but let timer do it. If not timing a request, 898 * do it now. 899 */ 900 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 901 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 902 nmp->nm_sent < nmp->nm_cwnd)) { 903 splx(s); 904 if (nmp->nm_soflags & PR_CONNREQUIRED) 905 error = nfs_sndlock(&nmp->nm_flag, rep); 906 if (!error) { 907 m = m_copym(m, 0, M_COPYALL, M_WAIT); 908 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 909 if (nmp->nm_soflags & PR_CONNREQUIRED) 910 nfs_sndunlock(&nmp->nm_flag); 911 } 912 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 913 nmp->nm_sent += NFS_CWNDSCALE; 914 rep->r_flags |= R_SENT; 915 } 916 } else { 917 splx(s); 918 rep->r_rtt = -1; 919 } 920 921 /* 922 * Wait for the reply from our send or the timer's. 923 */ 924 if (!error || error == EPIPE) 925 error = nfs_reply(rep); 926 927 /* 928 * RPC done, unlink the request. 929 */ 930 s = splsoftnet(); 931 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 932 splx(s); 933 934 /* 935 * Decrement the outstanding request count. 936 */ 937 if (rep->r_flags & R_SENT) { 938 rep->r_flags &= ~R_SENT; /* paranoia */ 939 nmp->nm_sent -= NFS_CWNDSCALE; 940 } 941 942 /* 943 * If there was a successful reply and a tprintf msg. 944 * tprintf a response. 945 */ 946 if (!error && (rep->r_flags & R_TPRINTFMSG)) 947 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 948 "is alive again"); 949 mrep = rep->r_mrep; 950 md = rep->r_md; 951 dpos = rep->r_dpos; 952 if (error) { 953 m_freem(rep->r_mreq); 954 free((caddr_t)rep, M_NFSREQ); 955 return (error); 956 } 957 958 /* 959 * break down the rpc header and check if ok 960 */ 961 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 962 if (*tl++ == rpc_msgdenied) { 963 if (*tl == rpc_mismatch) 964 error = EOPNOTSUPP; 965 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 966 if (!failed_auth) { 967 failed_auth++; 968 mheadend->m_next = (struct mbuf *)0; 969 m_freem(mrep); 970 m_freem(rep->r_mreq); 971 goto kerbauth; 972 } else 973 error = EAUTH; 974 } else 975 error = EACCES; 976 m_freem(mrep); 977 m_freem(rep->r_mreq); 978 free((caddr_t)rep, M_NFSREQ); 979 return (error); 980 } 981 982 /* 983 * Grab any Kerberos verifier, otherwise just throw it away. 984 */ 985 verf_type = fxdr_unsigned(int, *tl++); 986 i = fxdr_unsigned(int32_t, *tl); 987 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 988 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 989 if (error) 990 goto nfsmout; 991 } else if (i > 0) 992 nfsm_adv(nfsm_rndup(i)); 993 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 994 /* 0 == ok */ 995 if (*tl == 0) { 996 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 997 if (*tl != 0) { 998 error = fxdr_unsigned(int, *tl); 999 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1000 error == NFSERR_TRYLATER) { 1001 m_freem(mrep); 1002 error = 0; 1003 waituntil = time.tv_sec + trylater_delay; 1004 while (time.tv_sec < waituntil) 1005 (void) tsleep((caddr_t)&lbolt, 1006 PSOCK, "nqnfstry", 0); 1007 trylater_delay *= nfs_backoff[trylater_cnt]; 1008 if (trylater_cnt < 7) 1009 trylater_cnt++; 1010 goto tryagain; 1011 } 1012 1013 /* 1014 * If the File Handle was stale, invalidate the 1015 * lookup cache, just in case. 1016 */ 1017 if (error == ESTALE) 1018 cache_purge(vp); 1019 if (nmp->nm_flag & NFSMNT_NFSV3) { 1020 *mrp = mrep; 1021 *mdp = md; 1022 *dposp = dpos; 1023 error |= NFSERR_RETERR; 1024 } else 1025 m_freem(mrep); 1026 m_freem(rep->r_mreq); 1027 free((caddr_t)rep, M_NFSREQ); 1028 return (error); 1029 } 1030 1031 /* 1032 * For nqnfs, get any lease in reply 1033 */ 1034 if (nmp->nm_flag & NFSMNT_NQNFS) { 1035 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1036 if (*tl) { 1037 np = VTONFS(vp); 1038 nqlflag = fxdr_unsigned(int, *tl); 1039 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1040 cachable = fxdr_unsigned(int, *tl++); 1041 reqtime += fxdr_unsigned(int, *tl++); 1042 if (reqtime > time.tv_sec) { 1043 fxdr_hyper(tl, &frev); 1044 nqnfs_clientlease(nmp, np, nqlflag, 1045 cachable, reqtime, frev); 1046 } 1047 } 1048 } 1049 *mrp = mrep; 1050 *mdp = md; 1051 *dposp = dpos; 1052 m_freem(rep->r_mreq); 1053 FREE((caddr_t)rep, M_NFSREQ); 1054 return (0); 1055 } 1056 m_freem(mrep); 1057 error = EPROTONOSUPPORT; 1058 nfsmout: 1059 m_freem(rep->r_mreq); 1060 free((caddr_t)rep, M_NFSREQ); 1061 return (error); 1062 } 1063 #endif /* NFS */ 1064 1065 /* 1066 * Generate the rpc reply header 1067 * siz arg. is used to decide if adding a cluster is worthwhile 1068 */ 1069 int 1070 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1071 int siz; 1072 struct nfsrv_descript *nd; 1073 struct nfssvc_sock *slp; 1074 int err; 1075 int cache; 1076 u_quad_t *frev; 1077 struct mbuf **mrq; 1078 struct mbuf **mbp; 1079 caddr_t *bposp; 1080 { 1081 register u_int32_t *tl; 1082 register struct mbuf *mreq; 1083 caddr_t bpos; 1084 struct mbuf *mb, *mb2; 1085 1086 MGETHDR(mreq, M_WAIT, MT_DATA); 1087 mb = mreq; 1088 /* 1089 * If this is a big reply, use a cluster else 1090 * try and leave leading space for the lower level headers. 1091 */ 1092 siz += RPC_REPLYSIZ; 1093 if (siz >= MINCLSIZE) { 1094 MCLGET(mreq, M_WAIT); 1095 } else 1096 mreq->m_data += max_hdr; 1097 tl = mtod(mreq, u_int32_t *); 1098 mreq->m_len = 6 * NFSX_UNSIGNED; 1099 bpos = ((caddr_t)tl) + mreq->m_len; 1100 *tl++ = txdr_unsigned(nd->nd_retxid); 1101 *tl++ = rpc_reply; 1102 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1103 *tl++ = rpc_msgdenied; 1104 if (err & NFSERR_AUTHERR) { 1105 *tl++ = rpc_autherr; 1106 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1107 mreq->m_len -= NFSX_UNSIGNED; 1108 bpos -= NFSX_UNSIGNED; 1109 } else { 1110 *tl++ = rpc_mismatch; 1111 *tl++ = txdr_unsigned(RPC_VER2); 1112 *tl = txdr_unsigned(RPC_VER2); 1113 } 1114 } else { 1115 *tl++ = rpc_msgaccepted; 1116 1117 /* 1118 * For Kerberos authentication, we must send the nickname 1119 * verifier back, otherwise just RPCAUTH_NULL. 1120 */ 1121 if (nd->nd_flag & ND_KERBFULL) { 1122 register struct nfsuid *nuidp; 1123 struct timeval ktvin, ktvout; 1124 1125 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; 1126 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1127 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1128 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), 1129 &nuidp->nu_haddr, nd->nd_nam2))) 1130 break; 1131 } 1132 if (nuidp) { 1133 ktvin.tv_sec = 1134 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); 1135 ktvin.tv_usec = 1136 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1137 1138 /* 1139 * Encrypt the timestamp in ecb mode using the 1140 * session key. 1141 */ 1142 #ifdef NFSKERB 1143 XXX 1144 #endif 1145 1146 *tl++ = rpc_auth_kerb; 1147 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1148 *tl = ktvout.tv_sec; 1149 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1150 *tl++ = ktvout.tv_usec; 1151 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1152 } else { 1153 *tl++ = 0; 1154 *tl++ = 0; 1155 } 1156 } else { 1157 *tl++ = 0; 1158 *tl++ = 0; 1159 } 1160 switch (err) { 1161 case EPROGUNAVAIL: 1162 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1163 break; 1164 case EPROGMISMATCH: 1165 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1166 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1167 if (nd->nd_flag & ND_NQNFS) { 1168 *tl++ = txdr_unsigned(3); 1169 *tl = txdr_unsigned(3); 1170 } else { 1171 *tl++ = txdr_unsigned(2); 1172 *tl = txdr_unsigned(3); 1173 } 1174 break; 1175 case EPROCUNAVAIL: 1176 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1177 break; 1178 case EBADRPC: 1179 *tl = txdr_unsigned(RPC_GARBAGE); 1180 break; 1181 default: 1182 *tl = 0; 1183 if (err != NFSERR_RETVOID) { 1184 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1185 if (err) 1186 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1187 else 1188 *tl = 0; 1189 } 1190 break; 1191 }; 1192 } 1193 1194 /* 1195 * For nqnfs, piggyback lease as requested. 1196 */ 1197 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1198 if (nd->nd_flag & ND_LEASE) { 1199 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1200 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1201 *tl++ = txdr_unsigned(cache); 1202 *tl++ = txdr_unsigned(nd->nd_duration); 1203 txdr_hyper(frev, tl); 1204 } else { 1205 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1206 *tl = 0; 1207 } 1208 } 1209 if (mrq != NULL) 1210 *mrq = mreq; 1211 *mbp = mb; 1212 *bposp = bpos; 1213 if (err != 0 && err != NFSERR_RETVOID) 1214 nfsstats.srvrpc_errs++; 1215 return (0); 1216 } 1217 1218 /* 1219 * Nfs timer routine 1220 * Scan the nfsreq list and retranmit any requests that have timed out 1221 * To avoid retransmission attempts on STREAM sockets (in the future) make 1222 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1223 */ 1224 void 1225 nfs_timer(arg) 1226 void *arg; /* never used */ 1227 { 1228 register struct nfsreq *rep; 1229 register struct mbuf *m; 1230 register struct socket *so; 1231 register struct nfsmount *nmp; 1232 register int timeo; 1233 int s, error; 1234 #ifdef NFSSERVER 1235 register struct nfssvc_sock *slp; 1236 static long lasttime = 0; 1237 u_quad_t cur_usec; 1238 #endif 1239 1240 s = splsoftnet(); 1241 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1242 nmp = rep->r_nmp; 1243 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1244 continue; 1245 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1246 rep->r_flags |= R_SOFTTERM; 1247 continue; 1248 } 1249 if (rep->r_rtt >= 0) { 1250 rep->r_rtt++; 1251 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1252 timeo = nmp->nm_timeo; 1253 else 1254 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1255 if (nmp->nm_timeouts > 0) 1256 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1257 if (rep->r_rtt <= timeo) 1258 continue; 1259 if (nmp->nm_timeouts < 8) 1260 nmp->nm_timeouts++; 1261 } 1262 /* 1263 * Check for server not responding 1264 */ 1265 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1266 rep->r_rexmit > nmp->nm_deadthresh) { 1267 nfs_msg(rep->r_procp, 1268 nmp->nm_mountp->mnt_stat.f_mntfromname, 1269 "not responding"); 1270 rep->r_flags |= R_TPRINTFMSG; 1271 } 1272 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1273 nfsstats.rpctimeouts++; 1274 rep->r_flags |= R_SOFTTERM; 1275 continue; 1276 } 1277 if (nmp->nm_sotype != SOCK_DGRAM) { 1278 if (++rep->r_rexmit > NFS_MAXREXMIT) 1279 rep->r_rexmit = NFS_MAXREXMIT; 1280 continue; 1281 } 1282 if ((so = nmp->nm_so) == NULL) 1283 continue; 1284 1285 /* 1286 * If there is enough space and the window allows.. 1287 * Resend it 1288 * Set r_rtt to -1 in case we fail to send it now. 1289 */ 1290 rep->r_rtt = -1; 1291 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1292 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1293 (rep->r_flags & R_SENT) || 1294 nmp->nm_sent < nmp->nm_cwnd) && 1295 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1296 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1297 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1298 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1299 else 1300 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1301 nmp->nm_nam, (struct mbuf *)0, (struct proc *)0); 1302 if (error) { 1303 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1304 printf("nfs_timer: ignoring error %d\n", 1305 error); 1306 so->so_error = 0; 1307 } 1308 } else { 1309 /* 1310 * Iff first send, start timing 1311 * else turn timing off, backoff timer 1312 * and divide congestion window by 2. 1313 */ 1314 if (rep->r_flags & R_SENT) { 1315 rep->r_flags &= ~R_TIMING; 1316 if (++rep->r_rexmit > NFS_MAXREXMIT) 1317 rep->r_rexmit = NFS_MAXREXMIT; 1318 nmp->nm_cwnd >>= 1; 1319 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1320 nmp->nm_cwnd = NFS_CWNDSCALE; 1321 nfsstats.rpcretries++; 1322 } else { 1323 rep->r_flags |= R_SENT; 1324 nmp->nm_sent += NFS_CWNDSCALE; 1325 } 1326 rep->r_rtt = 0; 1327 } 1328 } 1329 } 1330 1331 #ifdef NFSSERVER 1332 /* 1333 * Call the nqnfs server timer once a second to handle leases. 1334 */ 1335 if (lasttime != time.tv_sec) { 1336 lasttime = time.tv_sec; 1337 nqnfs_serverd(); 1338 } 1339 1340 /* 1341 * Scan the write gathering queues for writes that need to be 1342 * completed now. 1343 */ 1344 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1345 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 1346 slp = slp->ns_chain.tqe_next) { 1347 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec) 1348 nfsrv_wakenfsd(slp); 1349 } 1350 #endif /* NFSSERVER */ 1351 splx(s); 1352 timeout(nfs_timer, (void *)0, nfs_ticks); 1353 } 1354 1355 /* 1356 * Test for a termination condition pending on the process. 1357 * This is used for NFSMNT_INT mounts. 1358 */ 1359 int 1360 nfs_sigintr(nmp, rep, p) 1361 struct nfsmount *nmp; 1362 struct nfsreq *rep; 1363 register struct proc *p; 1364 { 1365 1366 if (rep && (rep->r_flags & R_SOFTTERM)) 1367 return (EINTR); 1368 if (!(nmp->nm_flag & NFSMNT_INT)) 1369 return (0); 1370 if (p && p->p_siglist && 1371 (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) & 1372 NFSINT_SIGMASK)) 1373 return (EINTR); 1374 return (0); 1375 } 1376 1377 /* 1378 * Lock a socket against others. 1379 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1380 * and also to avoid race conditions between the processes with nfs requests 1381 * in progress when a reconnect is necessary. 1382 */ 1383 int 1384 nfs_sndlock(flagp, rep) 1385 register int *flagp; 1386 struct nfsreq *rep; 1387 { 1388 struct proc *p; 1389 int slpflag = 0, slptimeo = 0; 1390 1391 if (rep) { 1392 p = rep->r_procp; 1393 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1394 slpflag = PCATCH; 1395 } else 1396 p = (struct proc *)0; 1397 while (*flagp & NFSMNT_SNDLOCK) { 1398 if (nfs_sigintr(rep->r_nmp, rep, p)) 1399 return (EINTR); 1400 *flagp |= NFSMNT_WANTSND; 1401 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1402 slptimeo); 1403 if (slpflag == PCATCH) { 1404 slpflag = 0; 1405 slptimeo = 2 * hz; 1406 } 1407 } 1408 *flagp |= NFSMNT_SNDLOCK; 1409 return (0); 1410 } 1411 1412 /* 1413 * Unlock the stream socket for others. 1414 */ 1415 void 1416 nfs_sndunlock(flagp) 1417 register int *flagp; 1418 { 1419 1420 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1421 panic("nfs sndunlock"); 1422 *flagp &= ~NFSMNT_SNDLOCK; 1423 if (*flagp & NFSMNT_WANTSND) { 1424 *flagp &= ~NFSMNT_WANTSND; 1425 wakeup((caddr_t)flagp); 1426 } 1427 } 1428 1429 int 1430 nfs_rcvlock(rep) 1431 register struct nfsreq *rep; 1432 { 1433 register int *flagp = &rep->r_nmp->nm_flag; 1434 int slpflag, slptimeo = 0; 1435 1436 if (*flagp & NFSMNT_INT) 1437 slpflag = PCATCH; 1438 else 1439 slpflag = 0; 1440 while (*flagp & NFSMNT_RCVLOCK) { 1441 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1442 return (EINTR); 1443 *flagp |= NFSMNT_WANTRCV; 1444 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1445 slptimeo); 1446 if (slpflag == PCATCH) { 1447 slpflag = 0; 1448 slptimeo = 2 * hz; 1449 } 1450 } 1451 *flagp |= NFSMNT_RCVLOCK; 1452 return (0); 1453 } 1454 1455 /* 1456 * Unlock the stream socket for others. 1457 */ 1458 void 1459 nfs_rcvunlock(flagp) 1460 register int *flagp; 1461 { 1462 1463 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1464 panic("nfs rcvunlock"); 1465 *flagp &= ~NFSMNT_RCVLOCK; 1466 if (*flagp & NFSMNT_WANTRCV) { 1467 *flagp &= ~NFSMNT_WANTRCV; 1468 wakeup((caddr_t)flagp); 1469 } 1470 } 1471 1472 /* 1473 * Parse an RPC request 1474 * - verify it 1475 * - fill in the cred struct. 1476 */ 1477 int 1478 nfs_getreq(nd, nfsd, has_header) 1479 register struct nfsrv_descript *nd; 1480 struct nfsd *nfsd; 1481 int has_header; 1482 { 1483 register int len, i; 1484 register u_int32_t *tl; 1485 register int32_t t1; 1486 struct uio uio; 1487 struct iovec iov; 1488 caddr_t dpos, cp2, cp; 1489 u_int32_t nfsvers, auth_type; 1490 uid_t nickuid; 1491 int error = 0, nqnfs = 0, ticklen; 1492 struct mbuf *mrep, *md; 1493 register struct nfsuid *nuidp; 1494 struct timeval tvin, tvout; 1495 1496 mrep = nd->nd_mrep; 1497 md = nd->nd_md; 1498 dpos = nd->nd_dpos; 1499 if (has_header) { 1500 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1501 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1502 if (*tl++ != rpc_call) { 1503 m_freem(mrep); 1504 return (EBADRPC); 1505 } 1506 } else 1507 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1508 nd->nd_repstat = 0; 1509 nd->nd_flag = 0; 1510 if (*tl++ != rpc_vers) { 1511 nd->nd_repstat = ERPCMISMATCH; 1512 nd->nd_procnum = NFSPROC_NOOP; 1513 return (0); 1514 } 1515 if (*tl != nfs_prog) { 1516 if (*tl == nqnfs_prog) 1517 nqnfs++; 1518 else { 1519 nd->nd_repstat = EPROGUNAVAIL; 1520 nd->nd_procnum = NFSPROC_NOOP; 1521 return (0); 1522 } 1523 } 1524 tl++; 1525 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1526 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1527 (nfsvers != NQNFS_VER3 && nqnfs)) { 1528 nd->nd_repstat = EPROGMISMATCH; 1529 nd->nd_procnum = NFSPROC_NOOP; 1530 return (0); 1531 } 1532 if (nqnfs) 1533 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1534 else if (nfsvers == NFS_VER3) 1535 nd->nd_flag = ND_NFSV3; 1536 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1537 if (nd->nd_procnum == NFSPROC_NULL) 1538 return (0); 1539 if (nd->nd_procnum >= NFS_NPROCS || 1540 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1541 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1542 nd->nd_repstat = EPROCUNAVAIL; 1543 nd->nd_procnum = NFSPROC_NOOP; 1544 return (0); 1545 } 1546 if ((nd->nd_flag & ND_NFSV3) == 0) 1547 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1548 auth_type = *tl++; 1549 len = fxdr_unsigned(int, *tl++); 1550 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1551 m_freem(mrep); 1552 return (EBADRPC); 1553 } 1554 1555 nd->nd_flag &= ~ND_KERBAUTH; 1556 /* 1557 * Handle auth_unix or auth_kerb. 1558 */ 1559 if (auth_type == rpc_auth_unix) { 1560 len = fxdr_unsigned(int, *++tl); 1561 if (len < 0 || len > NFS_MAXNAMLEN) { 1562 m_freem(mrep); 1563 return (EBADRPC); 1564 } 1565 nfsm_adv(nfsm_rndup(len)); 1566 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1567 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred)); 1568 nd->nd_cr.cr_ref = 1; 1569 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1570 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1571 len = fxdr_unsigned(int, *tl); 1572 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1573 m_freem(mrep); 1574 return (EBADRPC); 1575 } 1576 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1577 for (i = 0; i < len; i++) 1578 if (i < NGROUPS) 1579 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1580 else 1581 tl++; 1582 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1583 if (nd->nd_cr.cr_ngroups > 1) 1584 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1585 len = fxdr_unsigned(int, *++tl); 1586 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1587 m_freem(mrep); 1588 return (EBADRPC); 1589 } 1590 if (len > 0) 1591 nfsm_adv(nfsm_rndup(len)); 1592 } else if (auth_type == rpc_auth_kerb) { 1593 switch (fxdr_unsigned(int, *tl++)) { 1594 case RPCAKN_FULLNAME: 1595 ticklen = fxdr_unsigned(int, *tl); 1596 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1597 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1598 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1599 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1600 m_freem(mrep); 1601 return (EBADRPC); 1602 } 1603 uio.uio_offset = 0; 1604 uio.uio_iov = &iov; 1605 uio.uio_iovcnt = 1; 1606 uio.uio_segflg = UIO_SYSSPACE; 1607 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1608 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1609 nfsm_mtouio(&uio, uio.uio_resid); 1610 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1611 if (*tl++ != rpc_auth_kerb || 1612 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1613 printf("Bad kerb verifier\n"); 1614 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1615 nd->nd_procnum = NFSPROC_NOOP; 1616 return (0); 1617 } 1618 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1619 tl = (u_int32_t *)cp; 1620 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1621 printf("Not fullname kerb verifier\n"); 1622 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1623 nd->nd_procnum = NFSPROC_NOOP; 1624 return (0); 1625 } 1626 cp += NFSX_UNSIGNED; 1627 bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED); 1628 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1629 nd->nd_flag |= ND_KERBFULL; 1630 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1631 break; 1632 case RPCAKN_NICKNAME: 1633 if (len != 2 * NFSX_UNSIGNED) { 1634 printf("Kerb nickname short\n"); 1635 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1636 nd->nd_procnum = NFSPROC_NOOP; 1637 return (0); 1638 } 1639 nickuid = fxdr_unsigned(uid_t, *tl); 1640 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1641 if (*tl++ != rpc_auth_kerb || 1642 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1643 printf("Kerb nick verifier bad\n"); 1644 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1645 nd->nd_procnum = NFSPROC_NOOP; 1646 return (0); 1647 } 1648 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1649 tvin.tv_sec = *tl++; 1650 tvin.tv_usec = *tl; 1651 1652 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; 1653 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1654 if (nuidp->nu_cr.cr_uid == nickuid && 1655 (!nd->nd_nam2 || 1656 netaddr_match(NU_NETFAM(nuidp), 1657 &nuidp->nu_haddr, nd->nd_nam2))) 1658 break; 1659 } 1660 if (!nuidp) { 1661 nd->nd_repstat = 1662 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1663 nd->nd_procnum = NFSPROC_NOOP; 1664 return (0); 1665 } 1666 1667 /* 1668 * Now, decrypt the timestamp using the session key 1669 * and validate it. 1670 */ 1671 #ifdef NFSKERB 1672 XXX 1673 #endif 1674 1675 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1676 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1677 if (nuidp->nu_expire < time.tv_sec || 1678 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1679 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1680 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1681 nuidp->nu_expire = 0; 1682 nd->nd_repstat = 1683 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1684 nd->nd_procnum = NFSPROC_NOOP; 1685 return (0); 1686 } 1687 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1688 nd->nd_flag |= ND_KERBNICK; 1689 }; 1690 } else { 1691 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1692 nd->nd_procnum = NFSPROC_NOOP; 1693 return (0); 1694 } 1695 1696 /* 1697 * For nqnfs, get piggybacked lease request. 1698 */ 1699 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1700 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1701 nd->nd_flag |= fxdr_unsigned(int, *tl); 1702 if (nd->nd_flag & ND_LEASE) { 1703 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1704 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 1705 } else 1706 nd->nd_duration = NQ_MINLEASE; 1707 } else 1708 nd->nd_duration = NQ_MINLEASE; 1709 nd->nd_md = md; 1710 nd->nd_dpos = dpos; 1711 return (0); 1712 nfsmout: 1713 return (error); 1714 } 1715 1716 int 1717 nfs_msg(p, server, msg) 1718 struct proc *p; 1719 char *server, *msg; 1720 { 1721 tpr_t tpr; 1722 1723 if (p) 1724 tpr = tprintf_open(p); 1725 else 1726 tpr = NULL; 1727 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1728 tprintf_close(tpr); 1729 return (0); 1730 } 1731 1732 #ifdef NFSSERVER 1733 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 1734 struct nfssvc_sock *, struct proc *, 1735 struct mbuf **)) = { 1736 nfsrv_null, 1737 nfsrv_getattr, 1738 nfsrv_setattr, 1739 nfsrv_lookup, 1740 nfsrv3_access, 1741 nfsrv_readlink, 1742 nfsrv_read, 1743 nfsrv_write, 1744 nfsrv_create, 1745 nfsrv_mkdir, 1746 nfsrv_symlink, 1747 nfsrv_mknod, 1748 nfsrv_remove, 1749 nfsrv_rmdir, 1750 nfsrv_rename, 1751 nfsrv_link, 1752 nfsrv_readdir, 1753 nfsrv_readdirplus, 1754 nfsrv_statfs, 1755 nfsrv_fsinfo, 1756 nfsrv_pathconf, 1757 nfsrv_commit, 1758 nqnfsrv_getlease, 1759 nqnfsrv_vacated, 1760 nfsrv_noop, 1761 nfsrv_noop 1762 }; 1763 1764 /* 1765 * Socket upcall routine for the nfsd sockets. 1766 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1767 * Essentially do as much as possible non-blocking, else punt and it will 1768 * be called with M_WAIT from an nfsd. 1769 */ 1770 void 1771 nfsrv_rcv(so, arg, waitflag) 1772 struct socket *so; 1773 caddr_t arg; 1774 int waitflag; 1775 { 1776 register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1777 register struct mbuf *m; 1778 struct mbuf *mp, *nam; 1779 struct uio auio; 1780 int flags, error; 1781 1782 if ((slp->ns_flag & SLP_VALID) == 0) 1783 return; 1784 #ifdef notdef 1785 /* 1786 * Define this to test for nfsds handling this under heavy load. 1787 */ 1788 if (waitflag == M_DONTWAIT) { 1789 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1790 } 1791 #endif 1792 auio.uio_procp = NULL; 1793 if (so->so_type == SOCK_STREAM) { 1794 /* 1795 * If there are already records on the queue, defer soreceive() 1796 * to an nfsd so that there is feedback to the TCP layer that 1797 * the nfs servers are heavily loaded. 1798 */ 1799 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1800 slp->ns_flag |= SLP_NEEDQ; 1801 goto dorecs; 1802 } 1803 1804 /* 1805 * Do soreceive(). 1806 */ 1807 auio.uio_resid = 1000000000; 1808 flags = MSG_DONTWAIT; 1809 error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1810 if (error || mp == (struct mbuf *)0) { 1811 if (error == EWOULDBLOCK) 1812 slp->ns_flag |= SLP_NEEDQ; 1813 else 1814 slp->ns_flag |= SLP_DISCONN; 1815 goto dorecs; 1816 } 1817 m = mp; 1818 if (slp->ns_rawend) { 1819 slp->ns_rawend->m_next = m; 1820 slp->ns_cc += 1000000000 - auio.uio_resid; 1821 } else { 1822 slp->ns_raw = m; 1823 slp->ns_cc = 1000000000 - auio.uio_resid; 1824 } 1825 while (m->m_next) 1826 m = m->m_next; 1827 slp->ns_rawend = m; 1828 1829 /* 1830 * Now try and parse record(s) out of the raw stream data. 1831 */ 1832 error = nfsrv_getstream(slp, waitflag); 1833 if (error) { 1834 if (error == EPERM) 1835 slp->ns_flag |= SLP_DISCONN; 1836 else 1837 slp->ns_flag |= SLP_NEEDQ; 1838 } 1839 } else { 1840 do { 1841 auio.uio_resid = 1000000000; 1842 flags = MSG_DONTWAIT; 1843 error = soreceive(so, &nam, &auio, &mp, 1844 (struct mbuf **)0, &flags); 1845 if (mp) { 1846 if (nam) { 1847 m = nam; 1848 m->m_next = mp; 1849 } else 1850 m = mp; 1851 if (slp->ns_recend) 1852 slp->ns_recend->m_nextpkt = m; 1853 else 1854 slp->ns_rec = m; 1855 slp->ns_recend = m; 1856 m->m_nextpkt = (struct mbuf *)0; 1857 } 1858 if (error) { 1859 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1860 && error != EWOULDBLOCK) { 1861 slp->ns_flag |= SLP_DISCONN; 1862 goto dorecs; 1863 } 1864 } 1865 } while (mp); 1866 } 1867 1868 /* 1869 * Now try and process the request records, non-blocking. 1870 */ 1871 dorecs: 1872 if (waitflag == M_DONTWAIT && 1873 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1874 nfsrv_wakenfsd(slp); 1875 } 1876 1877 /* 1878 * Try and extract an RPC request from the mbuf data list received on a 1879 * stream socket. The "waitflag" argument indicates whether or not it 1880 * can sleep. 1881 */ 1882 int 1883 nfsrv_getstream(slp, waitflag) 1884 register struct nfssvc_sock *slp; 1885 int waitflag; 1886 { 1887 register struct mbuf *m, **mpp; 1888 register char *cp1, *cp2; 1889 register int len; 1890 struct mbuf *om, *m2, *recm = NULL; 1891 u_int32_t recmark; 1892 1893 if (slp->ns_flag & SLP_GETSTREAM) 1894 panic("nfs getstream"); 1895 slp->ns_flag |= SLP_GETSTREAM; 1896 for (;;) { 1897 if (slp->ns_reclen == 0) { 1898 if (slp->ns_cc < NFSX_UNSIGNED) { 1899 slp->ns_flag &= ~SLP_GETSTREAM; 1900 return (0); 1901 } 1902 m = slp->ns_raw; 1903 if (m->m_len >= NFSX_UNSIGNED) { 1904 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 1905 m->m_data += NFSX_UNSIGNED; 1906 m->m_len -= NFSX_UNSIGNED; 1907 } else { 1908 cp1 = (caddr_t)&recmark; 1909 cp2 = mtod(m, caddr_t); 1910 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1911 while (m->m_len == 0) { 1912 m = m->m_next; 1913 cp2 = mtod(m, caddr_t); 1914 } 1915 *cp1++ = *cp2++; 1916 m->m_data++; 1917 m->m_len--; 1918 } 1919 } 1920 slp->ns_cc -= NFSX_UNSIGNED; 1921 recmark = ntohl(recmark); 1922 slp->ns_reclen = recmark & ~0x80000000; 1923 if (recmark & 0x80000000) 1924 slp->ns_flag |= SLP_LASTFRAG; 1925 else 1926 slp->ns_flag &= ~SLP_LASTFRAG; 1927 if (slp->ns_reclen > NFS_MAXPACKET) { 1928 slp->ns_flag &= ~SLP_GETSTREAM; 1929 return (EPERM); 1930 } 1931 } 1932 1933 /* 1934 * Now get the record part. 1935 */ 1936 if (slp->ns_cc == slp->ns_reclen) { 1937 recm = slp->ns_raw; 1938 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 1939 slp->ns_cc = slp->ns_reclen = 0; 1940 } else if (slp->ns_cc > slp->ns_reclen) { 1941 len = 0; 1942 m = slp->ns_raw; 1943 om = (struct mbuf *)0; 1944 while (len < slp->ns_reclen) { 1945 if ((len + m->m_len) > slp->ns_reclen) { 1946 size_t left = slp->ns_reclen - len; 1947 1948 MGETHDR(m2, waitflag, m->m_type); 1949 if (m2 == NULL) { 1950 slp->ns_flag &= ~SLP_GETSTREAM; 1951 return (EWOULDBLOCK); 1952 } 1953 if (left > MHLEN) { 1954 MCLGET(m2, waitflag); 1955 if (!(m2->m_flags & M_EXT)) { 1956 m_freem(m2); 1957 slp->ns_flag &= ~SLP_GETSTREAM; 1958 return (EWOULDBLOCK); 1959 } 1960 } 1961 bcopy(mtod(m, caddr_t), mtod(m2, caddr_t), 1962 left); 1963 m2->m_len = left; 1964 m->m_data += left; 1965 m->m_len -= left; 1966 if (om) { 1967 om->m_next = m2; 1968 recm = slp->ns_raw; 1969 } else 1970 recm = m2; 1971 len = slp->ns_reclen; 1972 } else if ((len + m->m_len) == slp->ns_reclen) { 1973 om = m; 1974 len += m->m_len; 1975 m = m->m_next; 1976 recm = slp->ns_raw; 1977 om->m_next = (struct mbuf *)0; 1978 } else { 1979 om = m; 1980 len += m->m_len; 1981 m = m->m_next; 1982 } 1983 } 1984 slp->ns_raw = m; 1985 slp->ns_cc -= len; 1986 slp->ns_reclen = 0; 1987 } else { 1988 slp->ns_flag &= ~SLP_GETSTREAM; 1989 return (0); 1990 } 1991 1992 /* 1993 * Accumulate the fragments into a record. 1994 */ 1995 mpp = &slp->ns_frag; 1996 while (*mpp) 1997 mpp = &((*mpp)->m_next); 1998 *mpp = recm; 1999 if (slp->ns_flag & SLP_LASTFRAG) { 2000 if (slp->ns_recend) 2001 slp->ns_recend->m_nextpkt = slp->ns_frag; 2002 else 2003 slp->ns_rec = slp->ns_frag; 2004 slp->ns_recend = slp->ns_frag; 2005 slp->ns_frag = (struct mbuf *)0; 2006 } 2007 } 2008 } 2009 2010 /* 2011 * Parse an RPC header. 2012 */ 2013 int 2014 nfsrv_dorec(slp, nfsd, ndp) 2015 register struct nfssvc_sock *slp; 2016 struct nfsd *nfsd; 2017 struct nfsrv_descript **ndp; 2018 { 2019 register struct mbuf *m, *nam; 2020 register struct nfsrv_descript *nd; 2021 int error; 2022 2023 *ndp = NULL; 2024 if ((slp->ns_flag & SLP_VALID) == 0 || 2025 (m = slp->ns_rec) == (struct mbuf *)0) 2026 return (ENOBUFS); 2027 slp->ns_rec = m->m_nextpkt; 2028 if (slp->ns_rec) 2029 m->m_nextpkt = (struct mbuf *)0; 2030 else 2031 slp->ns_recend = (struct mbuf *)0; 2032 if (m->m_type == MT_SONAME) { 2033 nam = m; 2034 m = m->m_next; 2035 nam->m_next = NULL; 2036 } else 2037 nam = NULL; 2038 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), 2039 M_NFSRVDESC, M_WAITOK); 2040 nd->nd_md = nd->nd_mrep = m; 2041 nd->nd_nam2 = nam; 2042 nd->nd_dpos = mtod(m, caddr_t); 2043 error = nfs_getreq(nd, nfsd, TRUE); 2044 if (error) { 2045 m_freem(nam); 2046 free((caddr_t)nd, M_NFSRVDESC); 2047 return (error); 2048 } 2049 *ndp = nd; 2050 nfsd->nfsd_nd = nd; 2051 return (0); 2052 } 2053 2054 2055 /* 2056 * Search for a sleeping nfsd and wake it up. 2057 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2058 * running nfsds will go look for the work in the nfssvc_sock list. 2059 */ 2060 void 2061 nfsrv_wakenfsd(slp) 2062 struct nfssvc_sock *slp; 2063 { 2064 register struct nfsd *nd; 2065 2066 if ((slp->ns_flag & SLP_VALID) == 0) 2067 return; 2068 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) { 2069 if (nd->nfsd_flag & NFSD_WAITING) { 2070 nd->nfsd_flag &= ~NFSD_WAITING; 2071 if (nd->nfsd_slp) 2072 panic("nfsd wakeup"); 2073 slp->ns_sref++; 2074 nd->nfsd_slp = slp; 2075 wakeup((caddr_t)nd); 2076 return; 2077 } 2078 } 2079 slp->ns_flag |= SLP_DOREC; 2080 nfsd_head_flag |= NFSD_CHECKSLP; 2081 } 2082 #endif /* NFSSERVER */ 2083
Indexes created Sun Sep 21 20:09:37 GMT 2025