nfs_socket.c revision 1.38.6.1
1 /* $NetBSD: nfs_socket.c,v 1.38.6.1 1997/09/08 23:15:11 thorpej 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/signalvar.h> 57 #include <sys/syslog.h> 58 #include <sys/tprintf.h> 59 #include <sys/namei.h> 60 61 #include <netinet/in.h> 62 #include <netinet/tcp.h> 63 64 #include <nfs/rpcv2.h> 65 #include <nfs/nfsproto.h> 66 #include <nfs/nfs.h> 67 #include <nfs/xdr_subs.h> 68 #include <nfs/nfsm_subs.h> 69 #include <nfs/nfsmount.h> 70 #include <nfs/nfsnode.h> 71 #include <nfs/nfsrtt.h> 72 #include <nfs/nqnfs.h> 73 #include <nfs/nfs_var.h> 74 75 #define TRUE 1 76 #define FALSE 0 77 78 /* 79 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 80 * Use the mean and mean deviation of rtt for the appropriate type of rpc 81 * for the frequent rpcs and a default for the others. 82 * The justification for doing "other" this way is that these rpcs 83 * happen so infrequently that timer est. would probably be stale. 84 * Also, since many of these rpcs are 85 * non-idempotent, a conservative timeout is desired. 86 * getattr, lookup - A+2D 87 * read, write - A+4D 88 * other - nm_timeo 89 */ 90 #define NFS_RTO(n, t) \ 91 ((t) == 0 ? (n)->nm_timeo : \ 92 ((t) < 3 ? \ 93 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 94 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 95 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 96 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 97 /* 98 * External data, mostly RPC constants in XDR form 99 */ 100 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 101 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, 102 rpc_auth_kerb; 103 extern u_int32_t nfs_prog, nqnfs_prog; 104 extern time_t nqnfsstarttime; 105 extern struct nfsstats nfsstats; 106 extern int nfsv3_procid[NFS_NPROCS]; 107 extern int nfs_ticks; 108 109 /* 110 * Defines which timer to use for the procnum. 111 * 0 - default 112 * 1 - getattr 113 * 2 - lookup 114 * 3 - read 115 * 4 - write 116 */ 117 static int proct[NFS_NPROCS] = { 118 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0, 119 0, 0, 0, 120 }; 121 122 /* 123 * There is a congestion window for outstanding rpcs maintained per mount 124 * point. The cwnd size is adjusted in roughly the way that: 125 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 126 * SIGCOMM '88". ACM, August 1988. 127 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 128 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 129 * of rpcs is in progress. 130 * (The sent count and cwnd are scaled for integer arith.) 131 * Variants of "slow start" were tried and were found to be too much of a 132 * performance hit (ave. rtt 3 times larger), 133 * I suspect due to the large rtt that nfs rpcs have. 134 */ 135 #define NFS_CWNDSCALE 256 136 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 137 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 138 int nfsrtton = 0; 139 struct nfsrtt nfsrtt; 140 141 /* 142 * Initialize sockets and congestion for a new NFS connection. 143 * We do not free the sockaddr if error. 144 */ 145 int 146 nfs_connect(nmp, rep) 147 register struct nfsmount *nmp; 148 struct nfsreq *rep; 149 { 150 register struct socket *so; 151 int s, error, rcvreserve, sndreserve; 152 struct sockaddr *saddr; 153 struct sockaddr_in *sin; 154 struct mbuf *m; 155 u_int16_t tport; 156 157 nmp->nm_so = (struct socket *)0; 158 saddr = mtod(nmp->nm_nam, struct sockaddr *); 159 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 160 nmp->nm_soproto); 161 if (error) 162 goto bad; 163 so = nmp->nm_so; 164 nmp->nm_soflags = so->so_proto->pr_flags; 165 166 /* 167 * Some servers require that the client port be a reserved port number. 168 */ 169 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 170 MGET(m, M_WAIT, MT_SONAME); 171 sin = mtod(m, struct sockaddr_in *); 172 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 173 sin->sin_family = AF_INET; 174 sin->sin_addr.s_addr = INADDR_ANY; 175 tport = IPPORT_RESERVED - 1; 176 sin->sin_port = htons(tport); 177 while ((error = sobind(so, m)) == EADDRINUSE && 178 --tport > IPPORT_RESERVED / 2) 179 sin->sin_port = htons(tport); 180 m_freem(m); 181 if (error) 182 goto bad; 183 } 184 185 /* 186 * Protocols that do not require connections may be optionally left 187 * unconnected for servers that reply from a port other than NFS_PORT. 188 */ 189 if (nmp->nm_flag & NFSMNT_NOCONN) { 190 if (nmp->nm_soflags & PR_CONNREQUIRED) { 191 error = ENOTCONN; 192 goto bad; 193 } 194 } else { 195 error = soconnect(so, nmp->nm_nam); 196 if (error) 197 goto bad; 198 199 /* 200 * Wait for the connection to complete. Cribbed from the 201 * connect system call but with the wait timing out so 202 * that interruptible mounts don't hang here for a long time. 203 */ 204 s = splsoftnet(); 205 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 206 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 207 "nfscon", 2 * hz); 208 if ((so->so_state & SS_ISCONNECTING) && 209 so->so_error == 0 && rep && 210 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ 211 so->so_state &= ~SS_ISCONNECTING; 212 splx(s); 213 goto bad; 214 } 215 } 216 if (so->so_error) { 217 error = so->so_error; 218 so->so_error = 0; 219 splx(s); 220 goto bad; 221 } 222 splx(s); 223 } 224 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 225 so->so_rcv.sb_timeo = (5 * hz); 226 so->so_snd.sb_timeo = (5 * hz); 227 } else { 228 so->so_rcv.sb_timeo = 0; 229 so->so_snd.sb_timeo = 0; 230 } 231 if (nmp->nm_sotype == SOCK_DGRAM) { 232 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 233 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 234 NFS_MAXPKTHDR) * 2; 235 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 236 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 237 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 238 NFS_MAXPKTHDR) * 2; 239 } else { 240 if (nmp->nm_sotype != SOCK_STREAM) 241 panic("nfscon sotype"); 242 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 243 MGET(m, M_WAIT, MT_SOOPTS); 244 *mtod(m, int32_t *) = 1; 245 m->m_len = sizeof(int32_t); 246 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 247 } 248 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 249 MGET(m, M_WAIT, MT_SOOPTS); 250 *mtod(m, int32_t *) = 1; 251 m->m_len = sizeof(int32_t); 252 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 253 } 254 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 255 sizeof (u_int32_t)) * 2; 256 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 257 sizeof (u_int32_t)) * 2; 258 } 259 error = soreserve(so, sndreserve, rcvreserve); 260 if (error) 261 goto bad; 262 so->so_rcv.sb_flags |= SB_NOINTR; 263 so->so_snd.sb_flags |= SB_NOINTR; 264 265 /* Initialize other non-zero congestion variables */ 266 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 267 nmp->nm_srtt[4] = (NFS_TIMEO << 3); 268 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 269 nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0; 270 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 271 nmp->nm_sent = 0; 272 nmp->nm_timeouts = 0; 273 return (0); 274 275 bad: 276 nfs_disconnect(nmp); 277 return (error); 278 } 279 280 /* 281 * Reconnect routine: 282 * Called when a connection is broken on a reliable protocol. 283 * - clean up the old socket 284 * - nfs_connect() again 285 * - set R_MUSTRESEND for all outstanding requests on mount point 286 * If this fails the mount point is DEAD! 287 * nb: Must be called with the nfs_sndlock() set on the mount point. 288 */ 289 int 290 nfs_reconnect(rep) 291 register struct nfsreq *rep; 292 { 293 register struct nfsreq *rp; 294 register struct nfsmount *nmp = rep->r_nmp; 295 int error; 296 297 nfs_disconnect(nmp); 298 while ((error = nfs_connect(nmp, rep)) != 0) { 299 if (error == EINTR || error == ERESTART) 300 return (EINTR); 301 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); 302 } 303 304 /* 305 * Loop through outstanding request list and fix up all requests 306 * on old socket. 307 */ 308 for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) { 309 if (rp->r_nmp == nmp) 310 rp->r_flags |= R_MUSTRESEND; 311 } 312 return (0); 313 } 314 315 /* 316 * NFS disconnect. Clean up and unlink. 317 */ 318 void 319 nfs_disconnect(nmp) 320 register struct nfsmount *nmp; 321 { 322 register struct socket *so; 323 324 if (nmp->nm_so) { 325 so = nmp->nm_so; 326 nmp->nm_so = (struct socket *)0; 327 soshutdown(so, 2); 328 soclose(so); 329 } 330 } 331 332 /* 333 * This is the nfs send routine. For connection based socket types, it 334 * must be called with an nfs_sndlock() on the socket. 335 * "rep == NULL" indicates that it has been called from a server. 336 * For the client side: 337 * - return EINTR if the RPC is terminated, 0 otherwise 338 * - set R_MUSTRESEND if the send fails for any reason 339 * - do any cleanup required by recoverable socket errors (???) 340 * For the server side: 341 * - return EINTR or ERESTART if interrupted by a signal 342 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 343 * - do any cleanup required by recoverable socket errors (???) 344 */ 345 int 346 nfs_send(so, nam, top, rep) 347 register struct socket *so; 348 struct mbuf *nam; 349 register struct mbuf *top; 350 struct nfsreq *rep; 351 { 352 struct mbuf *sendnam; 353 int error, soflags, flags; 354 355 if (rep) { 356 if (rep->r_flags & R_SOFTTERM) { 357 m_freem(top); 358 return (EINTR); 359 } 360 if ((so = rep->r_nmp->nm_so) == NULL) { 361 rep->r_flags |= R_MUSTRESEND; 362 m_freem(top); 363 return (0); 364 } 365 rep->r_flags &= ~R_MUSTRESEND; 366 soflags = rep->r_nmp->nm_soflags; 367 } else 368 soflags = so->so_proto->pr_flags; 369 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 370 sendnam = (struct mbuf *)0; 371 else 372 sendnam = nam; 373 if (so->so_type == SOCK_SEQPACKET) 374 flags = MSG_EOR; 375 else 376 flags = 0; 377 378 error = sosend(so, sendnam, (struct uio *)0, top, 379 (struct mbuf *)0, flags); 380 if (error) { 381 if (rep) { 382 log(LOG_INFO, "nfs send error %d for server %s\n",error, 383 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 384 /* 385 * Deal with errors for the client side. 386 */ 387 if (rep->r_flags & R_SOFTTERM) 388 error = EINTR; 389 else 390 rep->r_flags |= R_MUSTRESEND; 391 } else 392 log(LOG_INFO, "nfsd send error %d\n", error); 393 394 /* 395 * Handle any recoverable (soft) socket errors here. (???) 396 */ 397 if (error != EINTR && error != ERESTART && 398 error != EWOULDBLOCK && error != EPIPE) 399 error = 0; 400 } 401 return (error); 402 } 403 404 #ifdef NFS 405 /* 406 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 407 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 408 * Mark and consolidate the data into a new mbuf list. 409 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 410 * small mbufs. 411 * For SOCK_STREAM we must be very careful to read an entire record once 412 * we have read any of it, even if the system call has been interrupted. 413 */ 414 int 415 nfs_receive(rep, aname, mp) 416 register struct nfsreq *rep; 417 struct mbuf **aname; 418 struct mbuf **mp; 419 { 420 register struct socket *so; 421 struct uio auio; 422 struct iovec aio; 423 register struct mbuf *m; 424 struct mbuf *control; 425 u_int32_t len; 426 struct mbuf **getnam; 427 int error, sotype, rcvflg; 428 struct proc *p = curproc; /* XXX */ 429 430 /* 431 * Set up arguments for soreceive() 432 */ 433 *mp = (struct mbuf *)0; 434 *aname = (struct mbuf *)0; 435 sotype = rep->r_nmp->nm_sotype; 436 437 /* 438 * For reliable protocols, lock against other senders/receivers 439 * in case a reconnect is necessary. 440 * For SOCK_STREAM, first get the Record Mark to find out how much 441 * more there is to get. 442 * We must lock the socket against other receivers 443 * until we have an entire rpc request/reply. 444 */ 445 if (sotype != SOCK_DGRAM) { 446 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 447 if (error) 448 return (error); 449 tryagain: 450 /* 451 * Check for fatal errors and resending request. 452 */ 453 /* 454 * Ugh: If a reconnect attempt just happened, nm_so 455 * would have changed. NULL indicates a failed 456 * attempt that has essentially shut down this 457 * mount point. 458 */ 459 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 460 nfs_sndunlock(&rep->r_nmp->nm_flag); 461 return (EINTR); 462 } 463 so = rep->r_nmp->nm_so; 464 if (!so) { 465 error = nfs_reconnect(rep); 466 if (error) { 467 nfs_sndunlock(&rep->r_nmp->nm_flag); 468 return (error); 469 } 470 goto tryagain; 471 } 472 while (rep->r_flags & R_MUSTRESEND) { 473 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 474 nfsstats.rpcretries++; 475 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 476 if (error) { 477 if (error == EINTR || error == ERESTART || 478 (error = nfs_reconnect(rep)) != 0) { 479 nfs_sndunlock(&rep->r_nmp->nm_flag); 480 return (error); 481 } 482 goto tryagain; 483 } 484 } 485 nfs_sndunlock(&rep->r_nmp->nm_flag); 486 if (sotype == SOCK_STREAM) { 487 aio.iov_base = (caddr_t) &len; 488 aio.iov_len = sizeof(u_int32_t); 489 auio.uio_iov = &aio; 490 auio.uio_iovcnt = 1; 491 auio.uio_segflg = UIO_SYSSPACE; 492 auio.uio_rw = UIO_READ; 493 auio.uio_offset = 0; 494 auio.uio_resid = sizeof(u_int32_t); 495 auio.uio_procp = p; 496 do { 497 rcvflg = MSG_WAITALL; 498 error = soreceive(so, (struct mbuf **)0, &auio, 499 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 500 if (error == EWOULDBLOCK && rep) { 501 if (rep->r_flags & R_SOFTTERM) 502 return (EINTR); 503 } 504 } while (error == EWOULDBLOCK); 505 if (!error && auio.uio_resid > 0) { 506 log(LOG_INFO, 507 "short receive (%d/%d) from nfs server %s\n", 508 sizeof(u_int32_t) - auio.uio_resid, 509 sizeof(u_int32_t), 510 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 511 error = EPIPE; 512 } 513 if (error) 514 goto errout; 515 len = ntohl(len) & ~0x80000000; 516 /* 517 * This is SERIOUS! We are out of sync with the sender 518 * and forcing a disconnect/reconnect is all I can do. 519 */ 520 if (len > NFS_MAXPACKET) { 521 log(LOG_ERR, "%s (%d) from nfs server %s\n", 522 "impossible packet length", 523 len, 524 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 525 error = EFBIG; 526 goto errout; 527 } 528 auio.uio_resid = len; 529 do { 530 rcvflg = MSG_WAITALL; 531 error = soreceive(so, (struct mbuf **)0, 532 &auio, mp, (struct mbuf **)0, &rcvflg); 533 } while (error == EWOULDBLOCK || error == EINTR || 534 error == ERESTART); 535 if (!error && auio.uio_resid > 0) { 536 log(LOG_INFO, 537 "short receive (%d/%d) from nfs server %s\n", 538 len - auio.uio_resid, len, 539 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 540 error = EPIPE; 541 } 542 } else { 543 /* 544 * NB: Since uio_resid is big, MSG_WAITALL is ignored 545 * and soreceive() will return when it has either a 546 * control msg or a data msg. 547 * We have no use for control msg., but must grab them 548 * and then throw them away so we know what is going 549 * on. 550 */ 551 auio.uio_resid = len = 100000000; /* Anything Big */ 552 auio.uio_procp = p; 553 do { 554 rcvflg = 0; 555 error = soreceive(so, (struct mbuf **)0, 556 &auio, mp, &control, &rcvflg); 557 if (control) 558 m_freem(control); 559 if (error == EWOULDBLOCK && rep) { 560 if (rep->r_flags & R_SOFTTERM) 561 return (EINTR); 562 } 563 } while (error == EWOULDBLOCK || 564 (!error && *mp == NULL && control)); 565 if ((rcvflg & MSG_EOR) == 0) 566 printf("Egad!!\n"); 567 if (!error && *mp == NULL) 568 error = EPIPE; 569 len -= auio.uio_resid; 570 } 571 errout: 572 if (error && error != EINTR && error != ERESTART) { 573 m_freem(*mp); 574 *mp = (struct mbuf *)0; 575 if (error != EPIPE) 576 log(LOG_INFO, 577 "receive error %d from nfs server %s\n", 578 error, 579 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 580 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 581 if (!error) 582 error = nfs_reconnect(rep); 583 if (!error) 584 goto tryagain; 585 else 586 nfs_sndunlock(&rep->r_nmp->nm_flag); 587 } 588 } else { 589 if ((so = rep->r_nmp->nm_so) == NULL) 590 return (EACCES); 591 if (so->so_state & SS_ISCONNECTED) 592 getnam = (struct mbuf **)0; 593 else 594 getnam = aname; 595 auio.uio_resid = len = 1000000; 596 auio.uio_procp = p; 597 do { 598 rcvflg = 0; 599 error = soreceive(so, getnam, &auio, mp, 600 (struct mbuf **)0, &rcvflg); 601 if (error == EWOULDBLOCK && 602 (rep->r_flags & R_SOFTTERM)) 603 return (EINTR); 604 } while (error == EWOULDBLOCK); 605 len -= auio.uio_resid; 606 } 607 if (error) { 608 m_freem(*mp); 609 *mp = (struct mbuf *)0; 610 } 611 return (error); 612 } 613 614 /* 615 * Implement receipt of reply on a socket. 616 * We must search through the list of received datagrams matching them 617 * with outstanding requests using the xid, until ours is found. 618 */ 619 /* ARGSUSED */ 620 int 621 nfs_reply(myrep) 622 struct nfsreq *myrep; 623 { 624 register struct nfsreq *rep; 625 register struct nfsmount *nmp = myrep->r_nmp; 626 register int32_t t1; 627 struct mbuf *mrep, *nam, *md; 628 u_int32_t rxid, *tl; 629 caddr_t dpos, cp2; 630 int error; 631 632 /* 633 * Loop around until we get our own reply 634 */ 635 for (;;) { 636 /* 637 * Lock against other receivers so that I don't get stuck in 638 * sbwait() after someone else has received my reply for me. 639 * Also necessary for connection based protocols to avoid 640 * race conditions during a reconnect. 641 */ 642 error = nfs_rcvlock(myrep); 643 if (error == EALREADY) 644 return (0); 645 if (error) 646 return (error); 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 #ifdef DEBUG 660 printf("nfs_reply: ignoring error %d\n", error); 661 #endif 662 if (myrep->r_flags & R_GETONEREP) 663 return (0); 664 continue; 665 } 666 return (error); 667 } 668 if (nam) 669 m_freem(nam); 670 671 /* 672 * XXX: Temporary work-around for unexplained lossage with a 673 * XXX: netmask that's not byte-aligned, i.e. 255.255.255.192. 674 * XXX: See PR kern/3579 for details. 675 */ 676 if (mrep == 0) { 677 printf("nfs_reply: null mbuf from nfs_receive()\n"); 678 continue; 679 } 680 681 /* 682 * Get the xid and check that it is an rpc reply 683 */ 684 md = mrep; 685 dpos = mtod(md, caddr_t); 686 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 687 rxid = *tl++; 688 if (*tl != rpc_reply) { 689 if (nmp->nm_flag & NFSMNT_NQNFS) { 690 if (nqnfs_callback(nmp, mrep, md, dpos)) 691 nfsstats.rpcinvalid++; 692 } else { 693 nfsstats.rpcinvalid++; 694 m_freem(mrep); 695 } 696 nfsmout: 697 if (myrep->r_flags & R_GETONEREP) 698 return (0); 699 continue; 700 } 701 702 /* 703 * Loop through the request list to match up the reply 704 * Iff no match, just drop the datagram 705 */ 706 for (rep = nfs_reqq.tqh_first; rep != 0; 707 rep = rep->r_chain.tqe_next) { 708 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 709 /* Found it.. */ 710 rep->r_mrep = mrep; 711 rep->r_md = md; 712 rep->r_dpos = dpos; 713 if (nfsrtton) { 714 struct rttl *rt; 715 716 rt = &nfsrtt.rttl[nfsrtt.pos]; 717 rt->proc = rep->r_procnum; 718 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 719 rt->sent = nmp->nm_sent; 720 rt->cwnd = nmp->nm_cwnd; 721 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 722 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 723 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 724 rt->tstamp = time; 725 if (rep->r_flags & R_TIMING) 726 rt->rtt = rep->r_rtt; 727 else 728 rt->rtt = 1000000; 729 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 730 } 731 /* 732 * Update congestion window. 733 * Do the additive increase of 734 * one rpc/rtt. 735 */ 736 if (nmp->nm_cwnd <= nmp->nm_sent) { 737 nmp->nm_cwnd += 738 (NFS_CWNDSCALE * NFS_CWNDSCALE + 739 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 740 if (nmp->nm_cwnd > NFS_MAXCWND) 741 nmp->nm_cwnd = NFS_MAXCWND; 742 } 743 rep->r_flags &= ~R_SENT; 744 nmp->nm_sent -= NFS_CWNDSCALE; 745 /* 746 * Update rtt using a gain of 0.125 on the mean 747 * and a gain of 0.25 on the deviation. 748 */ 749 if (rep->r_flags & R_TIMING) { 750 /* 751 * Since the timer resolution of 752 * NFS_HZ is so course, it can often 753 * result in r_rtt == 0. Since 754 * r_rtt == N means that the actual 755 * rtt is between N+dt and N+2-dt ticks, 756 * add 1. 757 */ 758 t1 = rep->r_rtt + 1; 759 t1 -= (NFS_SRTT(rep) >> 3); 760 NFS_SRTT(rep) += t1; 761 if (t1 < 0) 762 t1 = -t1; 763 t1 -= (NFS_SDRTT(rep) >> 2); 764 NFS_SDRTT(rep) += t1; 765 } 766 nmp->nm_timeouts = 0; 767 break; 768 } 769 } 770 /* 771 * If not matched to a request, drop it. 772 * If it's mine, get out. 773 */ 774 if (rep == 0) { 775 nfsstats.rpcunexpected++; 776 m_freem(mrep); 777 } else if (rep == myrep) { 778 if (rep->r_mrep == NULL) 779 panic("nfsreply nil"); 780 return (0); 781 } 782 if (myrep->r_flags & R_GETONEREP) 783 return (0); 784 } 785 } 786 787 /* 788 * nfs_request - goes something like this 789 * - fill in request struct 790 * - links it into list 791 * - calls nfs_send() for first transmit 792 * - calls nfs_receive() to get reply 793 * - break down rpc header and return with nfs reply pointed to 794 * by mrep or error 795 * nb: always frees up mreq mbuf list 796 */ 797 int 798 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 799 struct vnode *vp; 800 struct mbuf *mrest; 801 int procnum; 802 struct proc *procp; 803 struct ucred *cred; 804 struct mbuf **mrp; 805 struct mbuf **mdp; 806 caddr_t *dposp; 807 { 808 register struct mbuf *m, *mrep; 809 register struct nfsreq *rep; 810 register u_int32_t *tl; 811 register int i; 812 struct nfsmount *nmp; 813 struct mbuf *md, *mheadend; 814 struct nfsnode *np; 815 char nickv[RPCX_NICKVERF]; 816 time_t reqtime, waituntil; 817 caddr_t dpos, cp2; 818 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; 819 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 820 int verf_len, verf_type; 821 u_int32_t xid; 822 u_quad_t frev; 823 char *auth_str, *verf_str; 824 NFSKERBKEY_T key; /* save session key */ 825 826 nmp = VFSTONFS(vp->v_mount); 827 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 828 rep->r_nmp = nmp; 829 rep->r_vp = vp; 830 rep->r_procp = procp; 831 rep->r_procnum = procnum; 832 i = 0; 833 m = mrest; 834 while (m) { 835 i += m->m_len; 836 m = m->m_next; 837 } 838 mrest_len = i; 839 840 /* 841 * Get the RPC header with authorization. 842 */ 843 kerbauth: 844 verf_str = auth_str = (char *)0; 845 if (nmp->nm_flag & NFSMNT_KERB) { 846 verf_str = nickv; 847 verf_len = sizeof (nickv); 848 auth_type = RPCAUTH_KERB4; 849 bzero((caddr_t)key, sizeof (key)); 850 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 851 &auth_len, verf_str, verf_len)) { 852 error = nfs_getauth(nmp, rep, cred, &auth_str, 853 &auth_len, verf_str, &verf_len, key); 854 if (error) { 855 free((caddr_t)rep, M_NFSREQ); 856 m_freem(mrest); 857 return (error); 858 } 859 } 860 } else { 861 auth_type = RPCAUTH_UNIX; 862 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 863 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 864 5 * NFSX_UNSIGNED; 865 } 866 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 867 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 868 if (auth_str) 869 free(auth_str, M_TEMP); 870 871 /* 872 * For stream protocols, insert a Sun RPC Record Mark. 873 */ 874 if (nmp->nm_sotype == SOCK_STREAM) { 875 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 876 *mtod(m, u_int32_t *) = htonl(0x80000000 | 877 (m->m_pkthdr.len - NFSX_UNSIGNED)); 878 } 879 rep->r_mreq = m; 880 rep->r_xid = xid; 881 tryagain: 882 if (nmp->nm_flag & NFSMNT_SOFT) 883 rep->r_retry = nmp->nm_retry; 884 else 885 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 886 rep->r_rtt = rep->r_rexmit = 0; 887 if (proct[procnum] > 0) 888 rep->r_flags = R_TIMING; 889 else 890 rep->r_flags = 0; 891 rep->r_mrep = NULL; 892 893 /* 894 * Do the client side RPC. 895 */ 896 nfsstats.rpcrequests++; 897 /* 898 * Chain request into list of outstanding requests. Be sure 899 * to put it LAST so timer finds oldest requests first. 900 */ 901 s = splsoftnet(); 902 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 903 904 /* Get send time for nqnfs */ 905 reqtime = time.tv_sec; 906 907 /* 908 * If backing off another request or avoiding congestion, don't 909 * send this one now but let timer do it. If not timing a request, 910 * do it now. 911 */ 912 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 913 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 914 nmp->nm_sent < nmp->nm_cwnd)) { 915 splx(s); 916 if (nmp->nm_soflags & PR_CONNREQUIRED) 917 error = nfs_sndlock(&nmp->nm_flag, rep); 918 if (!error) { 919 m = m_copym(m, 0, M_COPYALL, M_WAIT); 920 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 921 if (nmp->nm_soflags & PR_CONNREQUIRED) 922 nfs_sndunlock(&nmp->nm_flag); 923 } 924 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 925 nmp->nm_sent += NFS_CWNDSCALE; 926 rep->r_flags |= R_SENT; 927 } 928 } else { 929 splx(s); 930 rep->r_rtt = -1; 931 } 932 933 /* 934 * Wait for the reply from our send or the timer's. 935 */ 936 if (!error || error == EPIPE) 937 error = nfs_reply(rep); 938 939 /* 940 * RPC done, unlink the request. 941 */ 942 s = splsoftnet(); 943 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 944 splx(s); 945 946 /* 947 * Decrement the outstanding request count. 948 */ 949 if (rep->r_flags & R_SENT) { 950 rep->r_flags &= ~R_SENT; /* paranoia */ 951 nmp->nm_sent -= NFS_CWNDSCALE; 952 } 953 954 /* 955 * If there was a successful reply and a tprintf msg. 956 * tprintf a response. 957 */ 958 if (!error && (rep->r_flags & R_TPRINTFMSG)) 959 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 960 "is alive again"); 961 mrep = rep->r_mrep; 962 md = rep->r_md; 963 dpos = rep->r_dpos; 964 if (error) { 965 m_freem(rep->r_mreq); 966 free((caddr_t)rep, M_NFSREQ); 967 return (error); 968 } 969 970 /* 971 * break down the rpc header and check if ok 972 */ 973 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 974 if (*tl++ == rpc_msgdenied) { 975 if (*tl == rpc_mismatch) 976 error = EOPNOTSUPP; 977 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 978 if (!failed_auth) { 979 failed_auth++; 980 mheadend->m_next = (struct mbuf *)0; 981 m_freem(mrep); 982 m_freem(rep->r_mreq); 983 goto kerbauth; 984 } else 985 error = EAUTH; 986 } else 987 error = EACCES; 988 m_freem(mrep); 989 m_freem(rep->r_mreq); 990 free((caddr_t)rep, M_NFSREQ); 991 return (error); 992 } 993 994 /* 995 * Grab any Kerberos verifier, otherwise just throw it away. 996 */ 997 verf_type = fxdr_unsigned(int, *tl++); 998 i = fxdr_unsigned(int32_t, *tl); 999 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1000 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1001 if (error) 1002 goto nfsmout; 1003 } else if (i > 0) 1004 nfsm_adv(nfsm_rndup(i)); 1005 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1006 /* 0 == ok */ 1007 if (*tl == 0) { 1008 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1009 if (*tl != 0) { 1010 error = fxdr_unsigned(int, *tl); 1011 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1012 error == NFSERR_TRYLATER) { 1013 m_freem(mrep); 1014 error = 0; 1015 waituntil = time.tv_sec + trylater_delay; 1016 while (time.tv_sec < waituntil) 1017 (void) tsleep((caddr_t)&lbolt, 1018 PSOCK, "nqnfstry", 0); 1019 trylater_delay *= nfs_backoff[trylater_cnt]; 1020 if (trylater_cnt < 7) 1021 trylater_cnt++; 1022 goto tryagain; 1023 } 1024 1025 /* 1026 * If the File Handle was stale, invalidate the 1027 * lookup cache, just in case. 1028 */ 1029 if (error == ESTALE) 1030 cache_purge(vp); 1031 if (nmp->nm_flag & NFSMNT_NFSV3) { 1032 *mrp = mrep; 1033 *mdp = md; 1034 *dposp = dpos; 1035 error |= NFSERR_RETERR; 1036 } else 1037 m_freem(mrep); 1038 m_freem(rep->r_mreq); 1039 free((caddr_t)rep, M_NFSREQ); 1040 return (error); 1041 } 1042 1043 /* 1044 * For nqnfs, get any lease in reply 1045 */ 1046 if (nmp->nm_flag & NFSMNT_NQNFS) { 1047 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1048 if (*tl) { 1049 np = VTONFS(vp); 1050 nqlflag = fxdr_unsigned(int, *tl); 1051 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1052 cachable = fxdr_unsigned(int, *tl++); 1053 reqtime += fxdr_unsigned(int, *tl++); 1054 if (reqtime > time.tv_sec) { 1055 fxdr_hyper(tl, &frev); 1056 nqnfs_clientlease(nmp, np, nqlflag, 1057 cachable, reqtime, frev); 1058 } 1059 } 1060 } 1061 *mrp = mrep; 1062 *mdp = md; 1063 *dposp = dpos; 1064 m_freem(rep->r_mreq); 1065 FREE((caddr_t)rep, M_NFSREQ); 1066 return (0); 1067 } 1068 m_freem(mrep); 1069 error = EPROTONOSUPPORT; 1070 nfsmout: 1071 m_freem(rep->r_mreq); 1072 free((caddr_t)rep, M_NFSREQ); 1073 return (error); 1074 } 1075 #endif /* NFS */ 1076 1077 /* 1078 * Generate the rpc reply header 1079 * siz arg. is used to decide if adding a cluster is worthwhile 1080 */ 1081 int 1082 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1083 int siz; 1084 struct nfsrv_descript *nd; 1085 struct nfssvc_sock *slp; 1086 int err; 1087 int cache; 1088 u_quad_t *frev; 1089 struct mbuf **mrq; 1090 struct mbuf **mbp; 1091 caddr_t *bposp; 1092 { 1093 register u_int32_t *tl; 1094 register struct mbuf *mreq; 1095 caddr_t bpos; 1096 struct mbuf *mb, *mb2; 1097 1098 MGETHDR(mreq, M_WAIT, MT_DATA); 1099 mb = mreq; 1100 /* 1101 * If this is a big reply, use a cluster else 1102 * try and leave leading space for the lower level headers. 1103 */ 1104 siz += RPC_REPLYSIZ; 1105 if (siz >= MINCLSIZE) { 1106 MCLGET(mreq, M_WAIT); 1107 } else 1108 mreq->m_data += max_hdr; 1109 tl = mtod(mreq, u_int32_t *); 1110 mreq->m_len = 6 * NFSX_UNSIGNED; 1111 bpos = ((caddr_t)tl) + mreq->m_len; 1112 *tl++ = txdr_unsigned(nd->nd_retxid); 1113 *tl++ = rpc_reply; 1114 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1115 *tl++ = rpc_msgdenied; 1116 if (err & NFSERR_AUTHERR) { 1117 *tl++ = rpc_autherr; 1118 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1119 mreq->m_len -= NFSX_UNSIGNED; 1120 bpos -= NFSX_UNSIGNED; 1121 } else { 1122 *tl++ = rpc_mismatch; 1123 *tl++ = txdr_unsigned(RPC_VER2); 1124 *tl = txdr_unsigned(RPC_VER2); 1125 } 1126 } else { 1127 *tl++ = rpc_msgaccepted; 1128 1129 /* 1130 * For Kerberos authentication, we must send the nickname 1131 * verifier back, otherwise just RPCAUTH_NULL. 1132 */ 1133 if (nd->nd_flag & ND_KERBFULL) { 1134 register struct nfsuid *nuidp; 1135 struct timeval ktvin, ktvout; 1136 1137 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; 1138 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1139 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1140 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), 1141 &nuidp->nu_haddr, nd->nd_nam2))) 1142 break; 1143 } 1144 if (nuidp) { 1145 ktvin.tv_sec = 1146 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); 1147 ktvin.tv_usec = 1148 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1149 1150 /* 1151 * Encrypt the timestamp in ecb mode using the 1152 * session key. 1153 */ 1154 #ifdef NFSKERB 1155 XXX 1156 #endif 1157 1158 *tl++ = rpc_auth_kerb; 1159 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1160 *tl = ktvout.tv_sec; 1161 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1162 *tl++ = ktvout.tv_usec; 1163 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1164 } else { 1165 *tl++ = 0; 1166 *tl++ = 0; 1167 } 1168 } else { 1169 *tl++ = 0; 1170 *tl++ = 0; 1171 } 1172 switch (err) { 1173 case EPROGUNAVAIL: 1174 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1175 break; 1176 case EPROGMISMATCH: 1177 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1178 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1179 if (nd->nd_flag & ND_NQNFS) { 1180 *tl++ = txdr_unsigned(3); 1181 *tl = txdr_unsigned(3); 1182 } else { 1183 *tl++ = txdr_unsigned(2); 1184 *tl = txdr_unsigned(3); 1185 } 1186 break; 1187 case EPROCUNAVAIL: 1188 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1189 break; 1190 case EBADRPC: 1191 *tl = txdr_unsigned(RPC_GARBAGE); 1192 break; 1193 default: 1194 *tl = 0; 1195 if (err != NFSERR_RETVOID) { 1196 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1197 if (err) 1198 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1199 else 1200 *tl = 0; 1201 } 1202 break; 1203 }; 1204 } 1205 1206 /* 1207 * For nqnfs, piggyback lease as requested. 1208 */ 1209 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1210 if (nd->nd_flag & ND_LEASE) { 1211 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1212 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1213 *tl++ = txdr_unsigned(cache); 1214 *tl++ = txdr_unsigned(nd->nd_duration); 1215 txdr_hyper(frev, tl); 1216 } else { 1217 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1218 *tl = 0; 1219 } 1220 } 1221 if (mrq != NULL) 1222 *mrq = mreq; 1223 *mbp = mb; 1224 *bposp = bpos; 1225 if (err != 0 && err != NFSERR_RETVOID) 1226 nfsstats.srvrpc_errs++; 1227 return (0); 1228 } 1229 1230 /* 1231 * Nfs timer routine 1232 * Scan the nfsreq list and retranmit any requests that have timed out 1233 * To avoid retransmission attempts on STREAM sockets (in the future) make 1234 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1235 */ 1236 void 1237 nfs_timer(arg) 1238 void *arg; /* never used */ 1239 { 1240 register struct nfsreq *rep; 1241 register struct mbuf *m; 1242 register struct socket *so; 1243 register struct nfsmount *nmp; 1244 register int timeo; 1245 int s, error; 1246 #ifdef NFSSERVER 1247 register struct nfssvc_sock *slp; 1248 static long lasttime = 0; 1249 u_quad_t cur_usec; 1250 #endif 1251 1252 s = splsoftnet(); 1253 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1254 nmp = rep->r_nmp; 1255 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1256 continue; 1257 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1258 rep->r_flags |= R_SOFTTERM; 1259 continue; 1260 } 1261 if (rep->r_rtt >= 0) { 1262 rep->r_rtt++; 1263 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1264 timeo = nmp->nm_timeo; 1265 else 1266 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1267 if (nmp->nm_timeouts > 0) 1268 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1269 if (rep->r_rtt <= timeo) 1270 continue; 1271 if (nmp->nm_timeouts < 8) 1272 nmp->nm_timeouts++; 1273 } 1274 /* 1275 * Check for server not responding 1276 */ 1277 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1278 rep->r_rexmit > nmp->nm_deadthresh) { 1279 nfs_msg(rep->r_procp, 1280 nmp->nm_mountp->mnt_stat.f_mntfromname, 1281 "not responding"); 1282 rep->r_flags |= R_TPRINTFMSG; 1283 } 1284 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1285 nfsstats.rpctimeouts++; 1286 rep->r_flags |= R_SOFTTERM; 1287 continue; 1288 } 1289 if (nmp->nm_sotype != SOCK_DGRAM) { 1290 if (++rep->r_rexmit > NFS_MAXREXMIT) 1291 rep->r_rexmit = NFS_MAXREXMIT; 1292 continue; 1293 } 1294 if ((so = nmp->nm_so) == NULL) 1295 continue; 1296 1297 /* 1298 * If there is enough space and the window allows.. 1299 * Resend it 1300 * Set r_rtt to -1 in case we fail to send it now. 1301 */ 1302 rep->r_rtt = -1; 1303 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1304 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1305 (rep->r_flags & R_SENT) || 1306 nmp->nm_sent < nmp->nm_cwnd) && 1307 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1308 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1309 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1310 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1311 else 1312 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1313 nmp->nm_nam, (struct mbuf *)0, (struct proc *)0); 1314 if (error) { 1315 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1316 #ifdef DEBUG 1317 printf("nfs_timer: ignoring error %d\n", 1318 error); 1319 #endif 1320 so->so_error = 0; 1321 } 1322 } else { 1323 /* 1324 * Iff first send, start timing 1325 * else turn timing off, backoff timer 1326 * and divide congestion window by 2. 1327 */ 1328 if (rep->r_flags & R_SENT) { 1329 rep->r_flags &= ~R_TIMING; 1330 if (++rep->r_rexmit > NFS_MAXREXMIT) 1331 rep->r_rexmit = NFS_MAXREXMIT; 1332 nmp->nm_cwnd >>= 1; 1333 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1334 nmp->nm_cwnd = NFS_CWNDSCALE; 1335 nfsstats.rpcretries++; 1336 } else { 1337 rep->r_flags |= R_SENT; 1338 nmp->nm_sent += NFS_CWNDSCALE; 1339 } 1340 rep->r_rtt = 0; 1341 } 1342 } 1343 } 1344 1345 #ifdef NFSSERVER 1346 /* 1347 * Call the nqnfs server timer once a second to handle leases. 1348 */ 1349 if (lasttime != time.tv_sec) { 1350 lasttime = time.tv_sec; 1351 nqnfs_serverd(); 1352 } 1353 1354 /* 1355 * Scan the write gathering queues for writes that need to be 1356 * completed now. 1357 */ 1358 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1359 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 1360 slp = slp->ns_chain.tqe_next) { 1361 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec) 1362 nfsrv_wakenfsd(slp); 1363 } 1364 #endif /* NFSSERVER */ 1365 splx(s); 1366 timeout(nfs_timer, (void *)0, nfs_ticks); 1367 } 1368 1369 static const int nfsint_siglist[] = NFSINT_SIGLIST; 1370 1371 /* 1372 * Test for a termination condition pending on the process. 1373 * This is used for NFSMNT_INT mounts. 1374 */ 1375 int 1376 nfs_sigintr(nmp, rep, p) 1377 struct nfsmount *nmp; 1378 struct nfsreq *rep; 1379 register struct proc *p; 1380 { 1381 int i, bit; 1382 1383 if (rep && (rep->r_flags & R_SOFTTERM)) 1384 return (EINTR); 1385 if (!(nmp->nm_flag & NFSMNT_INT)) 1386 return (0); 1387 if (p == NULL || p->p_sigacts == NULL) 1388 return (0); 1389 for (i = 0; i < (sizeof(nfsint_siglist) / sizeof(nfsint_siglist[0])); 1390 i++) { 1391 if (SIGIGNORE(p, nfsint_siglist[i])) 1392 continue; /* signal is ignored */ 1393 bit = sigmask(nfsint_siglist[i]); 1394 if ((p->p_siglist & bit) == 0) 1395 continue; /* signal is not pending */ 1396 if (p->p_sigmask & bit) 1397 continue; /* signal is masked */ 1398 return (EINTR); 1399 } 1400 return (0); 1401 } 1402 1403 /* 1404 * Lock a socket against others. 1405 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1406 * and also to avoid race conditions between the processes with nfs requests 1407 * in progress when a reconnect is necessary. 1408 */ 1409 int 1410 nfs_sndlock(flagp, rep) 1411 register int *flagp; 1412 struct nfsreq *rep; 1413 { 1414 struct proc *p; 1415 int slpflag = 0, slptimeo = 0; 1416 1417 if (rep) { 1418 p = rep->r_procp; 1419 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1420 slpflag = PCATCH; 1421 } else 1422 p = (struct proc *)0; 1423 while (*flagp & NFSMNT_SNDLOCK) { 1424 if (nfs_sigintr(rep->r_nmp, rep, p)) 1425 return (EINTR); 1426 *flagp |= NFSMNT_WANTSND; 1427 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1428 slptimeo); 1429 if (slpflag == PCATCH) { 1430 slpflag = 0; 1431 slptimeo = 2 * hz; 1432 } 1433 } 1434 *flagp |= NFSMNT_SNDLOCK; 1435 return (0); 1436 } 1437 1438 /* 1439 * Unlock the stream socket for others. 1440 */ 1441 void 1442 nfs_sndunlock(flagp) 1443 register int *flagp; 1444 { 1445 1446 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1447 panic("nfs sndunlock"); 1448 *flagp &= ~NFSMNT_SNDLOCK; 1449 if (*flagp & NFSMNT_WANTSND) { 1450 *flagp &= ~NFSMNT_WANTSND; 1451 wakeup((caddr_t)flagp); 1452 } 1453 } 1454 1455 int 1456 nfs_rcvlock(rep) 1457 register struct nfsreq *rep; 1458 { 1459 register int *flagp = &rep->r_nmp->nm_flag; 1460 int slpflag, slptimeo = 0; 1461 1462 if (*flagp & NFSMNT_INT) 1463 slpflag = PCATCH; 1464 else 1465 slpflag = 0; 1466 while (*flagp & NFSMNT_RCVLOCK) { 1467 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1468 return (EINTR); 1469 *flagp |= NFSMNT_WANTRCV; 1470 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1471 slptimeo); 1472 /* If our reply was received while we were sleeping, 1473 * then just return without taking the lock to avoid a 1474 * situation where a single iod could 'capture' the 1475 * receive lock. 1476 */ 1477 if (rep->r_mrep != NULL) 1478 return (EALREADY); 1479 if (slpflag == PCATCH) { 1480 slpflag = 0; 1481 slptimeo = 2 * hz; 1482 } 1483 } 1484 *flagp |= NFSMNT_RCVLOCK; 1485 return (0); 1486 } 1487 1488 /* 1489 * Unlock the stream socket for others. 1490 */ 1491 void 1492 nfs_rcvunlock(flagp) 1493 register int *flagp; 1494 { 1495 1496 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1497 panic("nfs rcvunlock"); 1498 *flagp &= ~NFSMNT_RCVLOCK; 1499 if (*flagp & NFSMNT_WANTRCV) { 1500 *flagp &= ~NFSMNT_WANTRCV; 1501 wakeup((caddr_t)flagp); 1502 } 1503 } 1504 1505 /* 1506 * Parse an RPC request 1507 * - verify it 1508 * - fill in the cred struct. 1509 */ 1510 int 1511 nfs_getreq(nd, nfsd, has_header) 1512 register struct nfsrv_descript *nd; 1513 struct nfsd *nfsd; 1514 int has_header; 1515 { 1516 register int len, i; 1517 register u_int32_t *tl; 1518 register int32_t t1; 1519 struct uio uio; 1520 struct iovec iov; 1521 caddr_t dpos, cp2, cp; 1522 u_int32_t nfsvers, auth_type; 1523 uid_t nickuid; 1524 int error = 0, nqnfs = 0, ticklen; 1525 struct mbuf *mrep, *md; 1526 register struct nfsuid *nuidp; 1527 struct timeval tvin, tvout; 1528 1529 mrep = nd->nd_mrep; 1530 md = nd->nd_md; 1531 dpos = nd->nd_dpos; 1532 if (has_header) { 1533 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1534 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1535 if (*tl++ != rpc_call) { 1536 m_freem(mrep); 1537 return (EBADRPC); 1538 } 1539 } else 1540 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1541 nd->nd_repstat = 0; 1542 nd->nd_flag = 0; 1543 if (*tl++ != rpc_vers) { 1544 nd->nd_repstat = ERPCMISMATCH; 1545 nd->nd_procnum = NFSPROC_NOOP; 1546 return (0); 1547 } 1548 if (*tl != nfs_prog) { 1549 if (*tl == nqnfs_prog) 1550 nqnfs++; 1551 else { 1552 nd->nd_repstat = EPROGUNAVAIL; 1553 nd->nd_procnum = NFSPROC_NOOP; 1554 return (0); 1555 } 1556 } 1557 tl++; 1558 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1559 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1560 (nfsvers != NQNFS_VER3 && nqnfs)) { 1561 nd->nd_repstat = EPROGMISMATCH; 1562 nd->nd_procnum = NFSPROC_NOOP; 1563 return (0); 1564 } 1565 if (nqnfs) 1566 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1567 else if (nfsvers == NFS_VER3) 1568 nd->nd_flag = ND_NFSV3; 1569 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1570 if (nd->nd_procnum == NFSPROC_NULL) 1571 return (0); 1572 if (nd->nd_procnum >= NFS_NPROCS || 1573 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1574 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1575 nd->nd_repstat = EPROCUNAVAIL; 1576 nd->nd_procnum = NFSPROC_NOOP; 1577 return (0); 1578 } 1579 if ((nd->nd_flag & ND_NFSV3) == 0) 1580 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1581 auth_type = *tl++; 1582 len = fxdr_unsigned(int, *tl++); 1583 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1584 m_freem(mrep); 1585 return (EBADRPC); 1586 } 1587 1588 nd->nd_flag &= ~ND_KERBAUTH; 1589 /* 1590 * Handle auth_unix or auth_kerb. 1591 */ 1592 if (auth_type == rpc_auth_unix) { 1593 len = fxdr_unsigned(int, *++tl); 1594 if (len < 0 || len > NFS_MAXNAMLEN) { 1595 m_freem(mrep); 1596 return (EBADRPC); 1597 } 1598 nfsm_adv(nfsm_rndup(len)); 1599 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1600 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred)); 1601 nd->nd_cr.cr_ref = 1; 1602 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1603 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1604 len = fxdr_unsigned(int, *tl); 1605 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1606 m_freem(mrep); 1607 return (EBADRPC); 1608 } 1609 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1610 for (i = 0; i < len; i++) 1611 if (i < NGROUPS) 1612 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1613 else 1614 tl++; 1615 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1616 if (nd->nd_cr.cr_ngroups > 1) 1617 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1618 len = fxdr_unsigned(int, *++tl); 1619 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1620 m_freem(mrep); 1621 return (EBADRPC); 1622 } 1623 if (len > 0) 1624 nfsm_adv(nfsm_rndup(len)); 1625 } else if (auth_type == rpc_auth_kerb) { 1626 switch (fxdr_unsigned(int, *tl++)) { 1627 case RPCAKN_FULLNAME: 1628 ticklen = fxdr_unsigned(int, *tl); 1629 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1630 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1631 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1632 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1633 m_freem(mrep); 1634 return (EBADRPC); 1635 } 1636 uio.uio_offset = 0; 1637 uio.uio_iov = &iov; 1638 uio.uio_iovcnt = 1; 1639 uio.uio_segflg = UIO_SYSSPACE; 1640 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1641 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1642 nfsm_mtouio(&uio, uio.uio_resid); 1643 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1644 if (*tl++ != rpc_auth_kerb || 1645 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1646 printf("Bad kerb verifier\n"); 1647 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1648 nd->nd_procnum = NFSPROC_NOOP; 1649 return (0); 1650 } 1651 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1652 tl = (u_int32_t *)cp; 1653 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1654 printf("Not fullname kerb verifier\n"); 1655 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1656 nd->nd_procnum = NFSPROC_NOOP; 1657 return (0); 1658 } 1659 cp += NFSX_UNSIGNED; 1660 bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED); 1661 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1662 nd->nd_flag |= ND_KERBFULL; 1663 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1664 break; 1665 case RPCAKN_NICKNAME: 1666 if (len != 2 * NFSX_UNSIGNED) { 1667 printf("Kerb nickname short\n"); 1668 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1669 nd->nd_procnum = NFSPROC_NOOP; 1670 return (0); 1671 } 1672 nickuid = fxdr_unsigned(uid_t, *tl); 1673 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1674 if (*tl++ != rpc_auth_kerb || 1675 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1676 printf("Kerb nick verifier bad\n"); 1677 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1678 nd->nd_procnum = NFSPROC_NOOP; 1679 return (0); 1680 } 1681 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1682 tvin.tv_sec = *tl++; 1683 tvin.tv_usec = *tl; 1684 1685 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; 1686 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1687 if (nuidp->nu_cr.cr_uid == nickuid && 1688 (!nd->nd_nam2 || 1689 netaddr_match(NU_NETFAM(nuidp), 1690 &nuidp->nu_haddr, nd->nd_nam2))) 1691 break; 1692 } 1693 if (!nuidp) { 1694 nd->nd_repstat = 1695 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1696 nd->nd_procnum = NFSPROC_NOOP; 1697 return (0); 1698 } 1699 1700 /* 1701 * Now, decrypt the timestamp using the session key 1702 * and validate it. 1703 */ 1704 #ifdef NFSKERB 1705 XXX 1706 #endif 1707 1708 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1709 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1710 if (nuidp->nu_expire < time.tv_sec || 1711 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1712 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1713 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1714 nuidp->nu_expire = 0; 1715 nd->nd_repstat = 1716 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1717 nd->nd_procnum = NFSPROC_NOOP; 1718 return (0); 1719 } 1720 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1721 nd->nd_flag |= ND_KERBNICK; 1722 }; 1723 } else { 1724 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1725 nd->nd_procnum = NFSPROC_NOOP; 1726 return (0); 1727 } 1728 1729 /* 1730 * For nqnfs, get piggybacked lease request. 1731 */ 1732 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1733 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1734 nd->nd_flag |= fxdr_unsigned(int, *tl); 1735 if (nd->nd_flag & ND_LEASE) { 1736 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1737 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 1738 } else 1739 nd->nd_duration = NQ_MINLEASE; 1740 } else 1741 nd->nd_duration = NQ_MINLEASE; 1742 nd->nd_md = md; 1743 nd->nd_dpos = dpos; 1744 return (0); 1745 nfsmout: 1746 return (error); 1747 } 1748 1749 int 1750 nfs_msg(p, server, msg) 1751 struct proc *p; 1752 char *server, *msg; 1753 { 1754 tpr_t tpr; 1755 1756 if (p) 1757 tpr = tprintf_open(p); 1758 else 1759 tpr = NULL; 1760 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1761 tprintf_close(tpr); 1762 return (0); 1763 } 1764 1765 #ifdef NFSSERVER 1766 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 1767 struct nfssvc_sock *, struct proc *, 1768 struct mbuf **)) = { 1769 nfsrv_null, 1770 nfsrv_getattr, 1771 nfsrv_setattr, 1772 nfsrv_lookup, 1773 nfsrv3_access, 1774 nfsrv_readlink, 1775 nfsrv_read, 1776 nfsrv_write, 1777 nfsrv_create, 1778 nfsrv_mkdir, 1779 nfsrv_symlink, 1780 nfsrv_mknod, 1781 nfsrv_remove, 1782 nfsrv_rmdir, 1783 nfsrv_rename, 1784 nfsrv_link, 1785 nfsrv_readdir, 1786 nfsrv_readdirplus, 1787 nfsrv_statfs, 1788 nfsrv_fsinfo, 1789 nfsrv_pathconf, 1790 nfsrv_commit, 1791 nqnfsrv_getlease, 1792 nqnfsrv_vacated, 1793 nfsrv_noop, 1794 nfsrv_noop 1795 }; 1796 1797 /* 1798 * Socket upcall routine for the nfsd sockets. 1799 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1800 * Essentially do as much as possible non-blocking, else punt and it will 1801 * be called with M_WAIT from an nfsd. 1802 */ 1803 void 1804 nfsrv_rcv(so, arg, waitflag) 1805 struct socket *so; 1806 caddr_t arg; 1807 int waitflag; 1808 { 1809 register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1810 register struct mbuf *m; 1811 struct mbuf *mp, *nam; 1812 struct uio auio; 1813 int flags, error; 1814 1815 if ((slp->ns_flag & SLP_VALID) == 0) 1816 return; 1817 #ifdef notdef 1818 /* 1819 * Define this to test for nfsds handling this under heavy load. 1820 */ 1821 if (waitflag == M_DONTWAIT) { 1822 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1823 } 1824 #endif 1825 auio.uio_procp = NULL; 1826 if (so->so_type == SOCK_STREAM) { 1827 /* 1828 * If there are already records on the queue, defer soreceive() 1829 * to an nfsd so that there is feedback to the TCP layer that 1830 * the nfs servers are heavily loaded. 1831 */ 1832 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1833 slp->ns_flag |= SLP_NEEDQ; 1834 goto dorecs; 1835 } 1836 1837 /* 1838 * Do soreceive(). 1839 */ 1840 auio.uio_resid = 1000000000; 1841 flags = MSG_DONTWAIT; 1842 error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1843 if (error || mp == (struct mbuf *)0) { 1844 if (error == EWOULDBLOCK) 1845 slp->ns_flag |= SLP_NEEDQ; 1846 else 1847 slp->ns_flag |= SLP_DISCONN; 1848 goto dorecs; 1849 } 1850 m = mp; 1851 if (slp->ns_rawend) { 1852 slp->ns_rawend->m_next = m; 1853 slp->ns_cc += 1000000000 - auio.uio_resid; 1854 } else { 1855 slp->ns_raw = m; 1856 slp->ns_cc = 1000000000 - auio.uio_resid; 1857 } 1858 while (m->m_next) 1859 m = m->m_next; 1860 slp->ns_rawend = m; 1861 1862 /* 1863 * Now try and parse record(s) out of the raw stream data. 1864 */ 1865 error = nfsrv_getstream(slp, waitflag); 1866 if (error) { 1867 if (error == EPERM) 1868 slp->ns_flag |= SLP_DISCONN; 1869 else 1870 slp->ns_flag |= SLP_NEEDQ; 1871 } 1872 } else { 1873 do { 1874 auio.uio_resid = 1000000000; 1875 flags = MSG_DONTWAIT; 1876 error = soreceive(so, &nam, &auio, &mp, 1877 (struct mbuf **)0, &flags); 1878 if (mp) { 1879 if (nam) { 1880 m = nam; 1881 m->m_next = mp; 1882 } else 1883 m = mp; 1884 if (slp->ns_recend) 1885 slp->ns_recend->m_nextpkt = m; 1886 else 1887 slp->ns_rec = m; 1888 slp->ns_recend = m; 1889 m->m_nextpkt = (struct mbuf *)0; 1890 } 1891 if (error) { 1892 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1893 && error != EWOULDBLOCK) { 1894 slp->ns_flag |= SLP_DISCONN; 1895 goto dorecs; 1896 } 1897 } 1898 } while (mp); 1899 } 1900 1901 /* 1902 * Now try and process the request records, non-blocking. 1903 */ 1904 dorecs: 1905 if (waitflag == M_DONTWAIT && 1906 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1907 nfsrv_wakenfsd(slp); 1908 } 1909 1910 /* 1911 * Try and extract an RPC request from the mbuf data list received on a 1912 * stream socket. The "waitflag" argument indicates whether or not it 1913 * can sleep. 1914 */ 1915 int 1916 nfsrv_getstream(slp, waitflag) 1917 register struct nfssvc_sock *slp; 1918 int waitflag; 1919 { 1920 register struct mbuf *m, **mpp; 1921 register char *cp1, *cp2; 1922 register int len; 1923 struct mbuf *om, *m2, *recm = NULL; 1924 u_int32_t recmark; 1925 1926 if (slp->ns_flag & SLP_GETSTREAM) 1927 panic("nfs getstream"); 1928 slp->ns_flag |= SLP_GETSTREAM; 1929 for (;;) { 1930 if (slp->ns_reclen == 0) { 1931 if (slp->ns_cc < NFSX_UNSIGNED) { 1932 slp->ns_flag &= ~SLP_GETSTREAM; 1933 return (0); 1934 } 1935 m = slp->ns_raw; 1936 if (m->m_len >= NFSX_UNSIGNED) { 1937 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 1938 m->m_data += NFSX_UNSIGNED; 1939 m->m_len -= NFSX_UNSIGNED; 1940 } else { 1941 cp1 = (caddr_t)&recmark; 1942 cp2 = mtod(m, caddr_t); 1943 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1944 while (m->m_len == 0) { 1945 m = m->m_next; 1946 cp2 = mtod(m, caddr_t); 1947 } 1948 *cp1++ = *cp2++; 1949 m->m_data++; 1950 m->m_len--; 1951 } 1952 } 1953 slp->ns_cc -= NFSX_UNSIGNED; 1954 recmark = ntohl(recmark); 1955 slp->ns_reclen = recmark & ~0x80000000; 1956 if (recmark & 0x80000000) 1957 slp->ns_flag |= SLP_LASTFRAG; 1958 else 1959 slp->ns_flag &= ~SLP_LASTFRAG; 1960 if (slp->ns_reclen > NFS_MAXPACKET) { 1961 slp->ns_flag &= ~SLP_GETSTREAM; 1962 return (EPERM); 1963 } 1964 } 1965 1966 /* 1967 * Now get the record part. 1968 */ 1969 if (slp->ns_cc == slp->ns_reclen) { 1970 recm = slp->ns_raw; 1971 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 1972 slp->ns_cc = slp->ns_reclen = 0; 1973 } else if (slp->ns_cc > slp->ns_reclen) { 1974 len = 0; 1975 m = slp->ns_raw; 1976 om = (struct mbuf *)0; 1977 while (len < slp->ns_reclen) { 1978 if ((len + m->m_len) > slp->ns_reclen) { 1979 size_t left = slp->ns_reclen - len; 1980 1981 MGETHDR(m2, waitflag, m->m_type); 1982 if (m2 == NULL) { 1983 slp->ns_flag &= ~SLP_GETSTREAM; 1984 return (EWOULDBLOCK); 1985 } 1986 if (left > MHLEN) { 1987 MCLGET(m2, waitflag); 1988 if (!(m2->m_flags & M_EXT)) { 1989 m_freem(m2); 1990 slp->ns_flag &= ~SLP_GETSTREAM; 1991 return (EWOULDBLOCK); 1992 } 1993 } 1994 bcopy(mtod(m, caddr_t), mtod(m2, caddr_t), 1995 left); 1996 m2->m_len = left; 1997 m->m_data += left; 1998 m->m_len -= left; 1999 if (om) { 2000 om->m_next = m2; 2001 recm = slp->ns_raw; 2002 } else 2003 recm = m2; 2004 len = slp->ns_reclen; 2005 } else if ((len + m->m_len) == slp->ns_reclen) { 2006 om = m; 2007 len += m->m_len; 2008 m = m->m_next; 2009 recm = slp->ns_raw; 2010 om->m_next = (struct mbuf *)0; 2011 } else { 2012 om = m; 2013 len += m->m_len; 2014 m = m->m_next; 2015 } 2016 } 2017 slp->ns_raw = m; 2018 slp->ns_cc -= len; 2019 slp->ns_reclen = 0; 2020 } else { 2021 slp->ns_flag &= ~SLP_GETSTREAM; 2022 return (0); 2023 } 2024 2025 /* 2026 * Accumulate the fragments into a record. 2027 */ 2028 mpp = &slp->ns_frag; 2029 while (*mpp) 2030 mpp = &((*mpp)->m_next); 2031 *mpp = recm; 2032 if (slp->ns_flag & SLP_LASTFRAG) { 2033 if (slp->ns_recend) 2034 slp->ns_recend->m_nextpkt = slp->ns_frag; 2035 else 2036 slp->ns_rec = slp->ns_frag; 2037 slp->ns_recend = slp->ns_frag; 2038 slp->ns_frag = (struct mbuf *)0; 2039 } 2040 } 2041 } 2042 2043 /* 2044 * Parse an RPC header. 2045 */ 2046 int 2047 nfsrv_dorec(slp, nfsd, ndp) 2048 register struct nfssvc_sock *slp; 2049 struct nfsd *nfsd; 2050 struct nfsrv_descript **ndp; 2051 { 2052 register struct mbuf *m, *nam; 2053 register struct nfsrv_descript *nd; 2054 int error; 2055 2056 *ndp = NULL; 2057 if ((slp->ns_flag & SLP_VALID) == 0 || 2058 (m = slp->ns_rec) == (struct mbuf *)0) 2059 return (ENOBUFS); 2060 slp->ns_rec = m->m_nextpkt; 2061 if (slp->ns_rec) 2062 m->m_nextpkt = (struct mbuf *)0; 2063 else 2064 slp->ns_recend = (struct mbuf *)0; 2065 if (m->m_type == MT_SONAME) { 2066 nam = m; 2067 m = m->m_next; 2068 nam->m_next = NULL; 2069 } else 2070 nam = NULL; 2071 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), 2072 M_NFSRVDESC, M_WAITOK); 2073 nd->nd_md = nd->nd_mrep = m; 2074 nd->nd_nam2 = nam; 2075 nd->nd_dpos = mtod(m, caddr_t); 2076 error = nfs_getreq(nd, nfsd, TRUE); 2077 if (error) { 2078 m_freem(nam); 2079 free((caddr_t)nd, M_NFSRVDESC); 2080 return (error); 2081 } 2082 *ndp = nd; 2083 nfsd->nfsd_nd = nd; 2084 return (0); 2085 } 2086 2087 2088 /* 2089 * Search for a sleeping nfsd and wake it up. 2090 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2091 * running nfsds will go look for the work in the nfssvc_sock list. 2092 */ 2093 void 2094 nfsrv_wakenfsd(slp) 2095 struct nfssvc_sock *slp; 2096 { 2097 register struct nfsd *nd; 2098 2099 if ((slp->ns_flag & SLP_VALID) == 0) 2100 return; 2101 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) { 2102 if (nd->nfsd_flag & NFSD_WAITING) { 2103 nd->nfsd_flag &= ~NFSD_WAITING; 2104 if (nd->nfsd_slp) 2105 panic("nfsd wakeup"); 2106 slp->ns_sref++; 2107 nd->nfsd_slp = slp; 2108 wakeup((caddr_t)nd); 2109 return; 2110 } 2111 } 2112 slp->ns_flag |= SLP_DOREC; 2113 nfsd_head_flag |= NFSD_CHECKSLP; 2114 } 2115 #endif /* NFSSERVER */ 2116
Indexes created Tue Sep 30 11:09:46 GMT 2025