nfs_socket.c revision 1.34
1 /* $NetBSD: nfs_socket.c,v 1.34 1997/02/09 21:19:05 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 /* 609 * Search for any mbufs that are not a multiple of 4 bytes long 610 * or with m_data not longword aligned. 611 * These could cause pointer alignment problems, so copy them to 612 * well aligned mbufs. 613 */ 614 nfs_realign(*mp, 5 * NFSX_UNSIGNED); 615 return (error); 616 } 617 618 /* 619 * Implement receipt of reply on a socket. 620 * We must search through the list of received datagrams matching them 621 * with outstanding requests using the xid, until ours is found. 622 */ 623 /* ARGSUSED */ 624 int 625 nfs_reply(myrep) 626 struct nfsreq *myrep; 627 { 628 register struct nfsreq *rep; 629 register struct nfsmount *nmp = myrep->r_nmp; 630 register int32_t t1; 631 struct mbuf *mrep, *nam, *md; 632 u_int32_t rxid, *tl; 633 caddr_t dpos, cp2; 634 int error; 635 636 /* 637 * Loop around until we get our own reply 638 */ 639 for (;;) { 640 /* 641 * Lock against other receivers so that I don't get stuck in 642 * sbwait() after someone else has received my reply for me. 643 * Also necessary for connection based protocols to avoid 644 * race conditions during a reconnect. 645 */ 646 error = nfs_rcvlock(myrep); 647 if (error) 648 return (error); 649 /* Already received, bye bye */ 650 if (myrep->r_mrep != NULL) { 651 nfs_rcvunlock(&nmp->nm_flag); 652 return (0); 653 } 654 /* 655 * Get the next Rpc reply off the socket 656 */ 657 error = nfs_receive(myrep, &nam, &mrep); 658 nfs_rcvunlock(&nmp->nm_flag); 659 if (error) { 660 661 /* 662 * Ignore routing errors on connectionless protocols?? 663 */ 664 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 665 nmp->nm_so->so_error = 0; 666 printf("nfs_reply: ignoring error %d\n", error); 667 if (myrep->r_flags & R_GETONEREP) 668 return (0); 669 continue; 670 } 671 return (error); 672 } 673 if (nam) 674 m_freem(nam); 675 676 /* 677 * Get the xid and check that it is an rpc reply 678 */ 679 md = mrep; 680 dpos = mtod(md, caddr_t); 681 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 682 rxid = *tl++; 683 if (*tl != rpc_reply) { 684 if (nmp->nm_flag & NFSMNT_NQNFS) { 685 if (nqnfs_callback(nmp, mrep, md, dpos)) 686 nfsstats.rpcinvalid++; 687 } else { 688 nfsstats.rpcinvalid++; 689 m_freem(mrep); 690 } 691 nfsmout: 692 if (myrep->r_flags & R_GETONEREP) 693 return (0); 694 continue; 695 } 696 697 /* 698 * Loop through the request list to match up the reply 699 * Iff no match, just drop the datagram 700 */ 701 for (rep = nfs_reqq.tqh_first; rep != 0; 702 rep = rep->r_chain.tqe_next) { 703 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 704 /* Found it.. */ 705 rep->r_mrep = mrep; 706 rep->r_md = md; 707 rep->r_dpos = dpos; 708 if (nfsrtton) { 709 struct rttl *rt; 710 711 rt = &nfsrtt.rttl[nfsrtt.pos]; 712 rt->proc = rep->r_procnum; 713 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 714 rt->sent = nmp->nm_sent; 715 rt->cwnd = nmp->nm_cwnd; 716 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 717 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 718 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 719 rt->tstamp = time; 720 if (rep->r_flags & R_TIMING) 721 rt->rtt = rep->r_rtt; 722 else 723 rt->rtt = 1000000; 724 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 725 } 726 /* 727 * Update congestion window. 728 * Do the additive increase of 729 * one rpc/rtt. 730 */ 731 if (nmp->nm_cwnd <= nmp->nm_sent) { 732 nmp->nm_cwnd += 733 (NFS_CWNDSCALE * NFS_CWNDSCALE + 734 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 735 if (nmp->nm_cwnd > NFS_MAXCWND) 736 nmp->nm_cwnd = NFS_MAXCWND; 737 } 738 rep->r_flags &= ~R_SENT; 739 nmp->nm_sent -= NFS_CWNDSCALE; 740 /* 741 * Update rtt using a gain of 0.125 on the mean 742 * and a gain of 0.25 on the deviation. 743 */ 744 if (rep->r_flags & R_TIMING) { 745 /* 746 * Since the timer resolution of 747 * NFS_HZ is so course, it can often 748 * result in r_rtt == 0. Since 749 * r_rtt == N means that the actual 750 * rtt is between N+dt and N+2-dt ticks, 751 * add 1. 752 */ 753 t1 = rep->r_rtt + 1; 754 t1 -= (NFS_SRTT(rep) >> 3); 755 NFS_SRTT(rep) += t1; 756 if (t1 < 0) 757 t1 = -t1; 758 t1 -= (NFS_SDRTT(rep) >> 2); 759 NFS_SDRTT(rep) += t1; 760 } 761 nmp->nm_timeouts = 0; 762 break; 763 } 764 } 765 /* 766 * If not matched to a request, drop it. 767 * If it's mine, get out. 768 */ 769 if (rep == 0) { 770 nfsstats.rpcunexpected++; 771 m_freem(mrep); 772 } else if (rep == myrep) { 773 if (rep->r_mrep == NULL) 774 panic("nfsreply nil"); 775 return (0); 776 } 777 if (myrep->r_flags & R_GETONEREP) 778 return (0); 779 } 780 } 781 782 /* 783 * nfs_request - goes something like this 784 * - fill in request struct 785 * - links it into list 786 * - calls nfs_send() for first transmit 787 * - calls nfs_receive() to get reply 788 * - break down rpc header and return with nfs reply pointed to 789 * by mrep or error 790 * nb: always frees up mreq mbuf list 791 */ 792 int 793 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 794 struct vnode *vp; 795 struct mbuf *mrest; 796 int procnum; 797 struct proc *procp; 798 struct ucred *cred; 799 struct mbuf **mrp; 800 struct mbuf **mdp; 801 caddr_t *dposp; 802 { 803 register struct mbuf *m, *mrep; 804 register struct nfsreq *rep; 805 register u_int32_t *tl; 806 register int i; 807 struct nfsmount *nmp; 808 struct mbuf *md, *mheadend; 809 struct nfsnode *np; 810 char nickv[RPCX_NICKVERF]; 811 time_t reqtime, waituntil; 812 caddr_t dpos, cp2; 813 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; 814 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 815 int verf_len, verf_type; 816 u_int32_t xid; 817 u_quad_t frev; 818 char *auth_str, *verf_str; 819 NFSKERBKEY_T key; /* save session key */ 820 821 nmp = VFSTONFS(vp->v_mount); 822 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 823 rep->r_nmp = nmp; 824 rep->r_vp = vp; 825 rep->r_procp = procp; 826 rep->r_procnum = procnum; 827 i = 0; 828 m = mrest; 829 while (m) { 830 i += m->m_len; 831 m = m->m_next; 832 } 833 mrest_len = i; 834 835 /* 836 * Get the RPC header with authorization. 837 */ 838 kerbauth: 839 verf_str = auth_str = (char *)0; 840 if (nmp->nm_flag & NFSMNT_KERB) { 841 verf_str = nickv; 842 verf_len = sizeof (nickv); 843 auth_type = RPCAUTH_KERB4; 844 bzero((caddr_t)key, sizeof (key)); 845 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 846 &auth_len, verf_str, verf_len)) { 847 error = nfs_getauth(nmp, rep, cred, &auth_str, 848 &auth_len, verf_str, &verf_len, key); 849 if (error) { 850 free((caddr_t)rep, M_NFSREQ); 851 m_freem(mrest); 852 return (error); 853 } 854 } 855 } else { 856 auth_type = RPCAUTH_UNIX; 857 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 858 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 859 5 * NFSX_UNSIGNED; 860 } 861 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 862 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 863 if (auth_str) 864 free(auth_str, M_TEMP); 865 866 /* 867 * For stream protocols, insert a Sun RPC Record Mark. 868 */ 869 if (nmp->nm_sotype == SOCK_STREAM) { 870 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 871 *mtod(m, u_int32_t *) = htonl(0x80000000 | 872 (m->m_pkthdr.len - NFSX_UNSIGNED)); 873 } 874 rep->r_mreq = m; 875 rep->r_xid = xid; 876 tryagain: 877 if (nmp->nm_flag & NFSMNT_SOFT) 878 rep->r_retry = nmp->nm_retry; 879 else 880 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 881 rep->r_rtt = rep->r_rexmit = 0; 882 if (proct[procnum] > 0) 883 rep->r_flags = R_TIMING; 884 else 885 rep->r_flags = 0; 886 rep->r_mrep = NULL; 887 888 /* 889 * Do the client side RPC. 890 */ 891 nfsstats.rpcrequests++; 892 /* 893 * Chain request into list of outstanding requests. Be sure 894 * to put it LAST so timer finds oldest requests first. 895 */ 896 s = splsoftclock(); 897 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 898 899 /* Get send time for nqnfs */ 900 reqtime = time.tv_sec; 901 902 /* 903 * If backing off another request or avoiding congestion, don't 904 * send this one now but let timer do it. If not timing a request, 905 * do it now. 906 */ 907 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 908 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 909 nmp->nm_sent < nmp->nm_cwnd)) { 910 splx(s); 911 if (nmp->nm_soflags & PR_CONNREQUIRED) 912 error = nfs_sndlock(&nmp->nm_flag, rep); 913 if (!error) { 914 m = m_copym(m, 0, M_COPYALL, M_WAIT); 915 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 916 if (nmp->nm_soflags & PR_CONNREQUIRED) 917 nfs_sndunlock(&nmp->nm_flag); 918 } 919 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 920 nmp->nm_sent += NFS_CWNDSCALE; 921 rep->r_flags |= R_SENT; 922 } 923 } else { 924 splx(s); 925 rep->r_rtt = -1; 926 } 927 928 /* 929 * Wait for the reply from our send or the timer's. 930 */ 931 if (!error || error == EPIPE) 932 error = nfs_reply(rep); 933 934 /* 935 * RPC done, unlink the request. 936 */ 937 s = splsoftclock(); 938 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 939 splx(s); 940 941 /* 942 * Decrement the outstanding request count. 943 */ 944 if (rep->r_flags & R_SENT) { 945 rep->r_flags &= ~R_SENT; /* paranoia */ 946 nmp->nm_sent -= NFS_CWNDSCALE; 947 } 948 949 /* 950 * If there was a successful reply and a tprintf msg. 951 * tprintf a response. 952 */ 953 if (!error && (rep->r_flags & R_TPRINTFMSG)) 954 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 955 "is alive again"); 956 mrep = rep->r_mrep; 957 md = rep->r_md; 958 dpos = rep->r_dpos; 959 if (error) { 960 m_freem(rep->r_mreq); 961 free((caddr_t)rep, M_NFSREQ); 962 return (error); 963 } 964 965 /* 966 * break down the rpc header and check if ok 967 */ 968 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 969 if (*tl++ == rpc_msgdenied) { 970 if (*tl == rpc_mismatch) 971 error = EOPNOTSUPP; 972 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 973 if (!failed_auth) { 974 failed_auth++; 975 mheadend->m_next = (struct mbuf *)0; 976 m_freem(mrep); 977 m_freem(rep->r_mreq); 978 goto kerbauth; 979 } else 980 error = EAUTH; 981 } else 982 error = EACCES; 983 m_freem(mrep); 984 m_freem(rep->r_mreq); 985 free((caddr_t)rep, M_NFSREQ); 986 return (error); 987 } 988 989 /* 990 * Grab any Kerberos verifier, otherwise just throw it away. 991 */ 992 verf_type = fxdr_unsigned(int, *tl++); 993 i = fxdr_unsigned(int32_t, *tl); 994 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 995 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 996 if (error) 997 goto nfsmout; 998 } else if (i > 0) 999 nfsm_adv(nfsm_rndup(i)); 1000 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1001 /* 0 == ok */ 1002 if (*tl == 0) { 1003 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1004 if (*tl != 0) { 1005 error = fxdr_unsigned(int, *tl); 1006 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1007 error == NFSERR_TRYLATER) { 1008 m_freem(mrep); 1009 error = 0; 1010 waituntil = time.tv_sec + trylater_delay; 1011 while (time.tv_sec < waituntil) 1012 (void) tsleep((caddr_t)&lbolt, 1013 PSOCK, "nqnfstry", 0); 1014 trylater_delay *= nfs_backoff[trylater_cnt]; 1015 if (trylater_cnt < 7) 1016 trylater_cnt++; 1017 goto tryagain; 1018 } 1019 1020 /* 1021 * If the File Handle was stale, invalidate the 1022 * lookup cache, just in case. 1023 */ 1024 if (error == ESTALE) 1025 cache_purge(vp); 1026 if (nmp->nm_flag & NFSMNT_NFSV3) { 1027 *mrp = mrep; 1028 *mdp = md; 1029 *dposp = dpos; 1030 error |= NFSERR_RETERR; 1031 } else 1032 m_freem(mrep); 1033 m_freem(rep->r_mreq); 1034 free((caddr_t)rep, M_NFSREQ); 1035 return (error); 1036 } 1037 1038 /* 1039 * For nqnfs, get any lease in reply 1040 */ 1041 if (nmp->nm_flag & NFSMNT_NQNFS) { 1042 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1043 if (*tl) { 1044 np = VTONFS(vp); 1045 nqlflag = fxdr_unsigned(int, *tl); 1046 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1047 cachable = fxdr_unsigned(int, *tl++); 1048 reqtime += fxdr_unsigned(int, *tl++); 1049 if (reqtime > time.tv_sec) { 1050 fxdr_hyper(tl, &frev); 1051 nqnfs_clientlease(nmp, np, nqlflag, 1052 cachable, reqtime, frev); 1053 } 1054 } 1055 } 1056 *mrp = mrep; 1057 *mdp = md; 1058 *dposp = dpos; 1059 m_freem(rep->r_mreq); 1060 FREE((caddr_t)rep, M_NFSREQ); 1061 return (0); 1062 } 1063 m_freem(mrep); 1064 error = EPROTONOSUPPORT; 1065 nfsmout: 1066 m_freem(rep->r_mreq); 1067 free((caddr_t)rep, M_NFSREQ); 1068 return (error); 1069 } 1070 #endif /* NFS */ 1071 1072 /* 1073 * Generate the rpc reply header 1074 * siz arg. is used to decide if adding a cluster is worthwhile 1075 */ 1076 int 1077 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1078 int siz; 1079 struct nfsrv_descript *nd; 1080 struct nfssvc_sock *slp; 1081 int err; 1082 int cache; 1083 u_quad_t *frev; 1084 struct mbuf **mrq; 1085 struct mbuf **mbp; 1086 caddr_t *bposp; 1087 { 1088 register u_int32_t *tl; 1089 register struct mbuf *mreq; 1090 caddr_t bpos; 1091 struct mbuf *mb, *mb2; 1092 1093 MGETHDR(mreq, M_WAIT, MT_DATA); 1094 mb = mreq; 1095 /* 1096 * If this is a big reply, use a cluster else 1097 * try and leave leading space for the lower level headers. 1098 */ 1099 siz += RPC_REPLYSIZ; 1100 if (siz >= MINCLSIZE) { 1101 MCLGET(mreq, M_WAIT); 1102 } else 1103 mreq->m_data += max_hdr; 1104 tl = mtod(mreq, u_int32_t *); 1105 mreq->m_len = 6 * NFSX_UNSIGNED; 1106 bpos = ((caddr_t)tl) + mreq->m_len; 1107 *tl++ = txdr_unsigned(nd->nd_retxid); 1108 *tl++ = rpc_reply; 1109 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1110 *tl++ = rpc_msgdenied; 1111 if (err & NFSERR_AUTHERR) { 1112 *tl++ = rpc_autherr; 1113 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1114 mreq->m_len -= NFSX_UNSIGNED; 1115 bpos -= NFSX_UNSIGNED; 1116 } else { 1117 *tl++ = rpc_mismatch; 1118 *tl++ = txdr_unsigned(RPC_VER2); 1119 *tl = txdr_unsigned(RPC_VER2); 1120 } 1121 } else { 1122 *tl++ = rpc_msgaccepted; 1123 1124 /* 1125 * For Kerberos authentication, we must send the nickname 1126 * verifier back, otherwise just RPCAUTH_NULL. 1127 */ 1128 if (nd->nd_flag & ND_KERBFULL) { 1129 register struct nfsuid *nuidp; 1130 struct timeval ktvin, ktvout; 1131 1132 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; 1133 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1134 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1135 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), 1136 &nuidp->nu_haddr, nd->nd_nam2))) 1137 break; 1138 } 1139 if (nuidp) { 1140 ktvin.tv_sec = 1141 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); 1142 ktvin.tv_usec = 1143 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1144 1145 /* 1146 * Encrypt the timestamp in ecb mode using the 1147 * session key. 1148 */ 1149 #ifdef NFSKERB 1150 XXX 1151 #endif 1152 1153 *tl++ = rpc_auth_kerb; 1154 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1155 *tl = ktvout.tv_sec; 1156 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1157 *tl++ = ktvout.tv_usec; 1158 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1159 } else { 1160 *tl++ = 0; 1161 *tl++ = 0; 1162 } 1163 } else { 1164 *tl++ = 0; 1165 *tl++ = 0; 1166 } 1167 switch (err) { 1168 case EPROGUNAVAIL: 1169 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1170 break; 1171 case EPROGMISMATCH: 1172 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1173 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1174 if (nd->nd_flag & ND_NQNFS) { 1175 *tl++ = txdr_unsigned(3); 1176 *tl = txdr_unsigned(3); 1177 } else { 1178 *tl++ = txdr_unsigned(2); 1179 *tl = txdr_unsigned(3); 1180 } 1181 break; 1182 case EPROCUNAVAIL: 1183 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1184 break; 1185 case EBADRPC: 1186 *tl = txdr_unsigned(RPC_GARBAGE); 1187 break; 1188 default: 1189 *tl = 0; 1190 if (err != NFSERR_RETVOID) { 1191 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1192 if (err) 1193 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1194 else 1195 *tl = 0; 1196 } 1197 break; 1198 }; 1199 } 1200 1201 /* 1202 * For nqnfs, piggyback lease as requested. 1203 */ 1204 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1205 if (nd->nd_flag & ND_LEASE) { 1206 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1207 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1208 *tl++ = txdr_unsigned(cache); 1209 *tl++ = txdr_unsigned(nd->nd_duration); 1210 txdr_hyper(frev, tl); 1211 } else { 1212 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1213 *tl = 0; 1214 } 1215 } 1216 if (mrq != NULL) 1217 *mrq = mreq; 1218 *mbp = mb; 1219 *bposp = bpos; 1220 if (err != 0 && err != NFSERR_RETVOID) 1221 nfsstats.srvrpc_errs++; 1222 return (0); 1223 } 1224 1225 /* 1226 * Nfs timer routine 1227 * Scan the nfsreq list and retranmit any requests that have timed out 1228 * To avoid retransmission attempts on STREAM sockets (in the future) make 1229 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1230 */ 1231 void 1232 nfs_timer(arg) 1233 void *arg; /* never used */ 1234 { 1235 register struct nfsreq *rep; 1236 register struct mbuf *m; 1237 register struct socket *so; 1238 register struct nfsmount *nmp; 1239 register int timeo; 1240 int s, error; 1241 #ifdef NFSSERVER 1242 register struct nfssvc_sock *slp; 1243 static long lasttime = 0; 1244 u_quad_t cur_usec; 1245 #endif 1246 1247 s = splsoftnet(); 1248 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1249 nmp = rep->r_nmp; 1250 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1251 continue; 1252 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1253 rep->r_flags |= R_SOFTTERM; 1254 continue; 1255 } 1256 if (rep->r_rtt >= 0) { 1257 rep->r_rtt++; 1258 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1259 timeo = nmp->nm_timeo; 1260 else 1261 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1262 if (nmp->nm_timeouts > 0) 1263 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1264 if (rep->r_rtt <= timeo) 1265 continue; 1266 if (nmp->nm_timeouts < 8) 1267 nmp->nm_timeouts++; 1268 } 1269 /* 1270 * Check for server not responding 1271 */ 1272 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1273 rep->r_rexmit > nmp->nm_deadthresh) { 1274 nfs_msg(rep->r_procp, 1275 nmp->nm_mountp->mnt_stat.f_mntfromname, 1276 "not responding"); 1277 rep->r_flags |= R_TPRINTFMSG; 1278 } 1279 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1280 nfsstats.rpctimeouts++; 1281 rep->r_flags |= R_SOFTTERM; 1282 continue; 1283 } 1284 if (nmp->nm_sotype != SOCK_DGRAM) { 1285 if (++rep->r_rexmit > NFS_MAXREXMIT) 1286 rep->r_rexmit = NFS_MAXREXMIT; 1287 continue; 1288 } 1289 if ((so = nmp->nm_so) == NULL) 1290 continue; 1291 1292 /* 1293 * If there is enough space and the window allows.. 1294 * Resend it 1295 * Set r_rtt to -1 in case we fail to send it now. 1296 */ 1297 rep->r_rtt = -1; 1298 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1299 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1300 (rep->r_flags & R_SENT) || 1301 nmp->nm_sent < nmp->nm_cwnd) && 1302 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1303 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1304 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1305 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1306 else 1307 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1308 nmp->nm_nam, (struct mbuf *)0, (struct proc *)0); 1309 if (error) { 1310 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1311 printf("nfs_timer: ignoring error %d\n", 1312 error); 1313 so->so_error = 0; 1314 } 1315 } else { 1316 /* 1317 * Iff first send, start timing 1318 * else turn timing off, backoff timer 1319 * and divide congestion window by 2. 1320 */ 1321 if (rep->r_flags & R_SENT) { 1322 rep->r_flags &= ~R_TIMING; 1323 if (++rep->r_rexmit > NFS_MAXREXMIT) 1324 rep->r_rexmit = NFS_MAXREXMIT; 1325 nmp->nm_cwnd >>= 1; 1326 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1327 nmp->nm_cwnd = NFS_CWNDSCALE; 1328 nfsstats.rpcretries++; 1329 } else { 1330 rep->r_flags |= R_SENT; 1331 nmp->nm_sent += NFS_CWNDSCALE; 1332 } 1333 rep->r_rtt = 0; 1334 } 1335 } 1336 } 1337 1338 #ifdef NFSSERVER 1339 /* 1340 * Call the nqnfs server timer once a second to handle leases. 1341 */ 1342 if (lasttime != time.tv_sec) { 1343 lasttime = time.tv_sec; 1344 nqnfs_serverd(); 1345 } 1346 1347 /* 1348 * Scan the write gathering queues for writes that need to be 1349 * completed now. 1350 */ 1351 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1352 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 1353 slp = slp->ns_chain.tqe_next) { 1354 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec) 1355 nfsrv_wakenfsd(slp); 1356 } 1357 #endif /* NFSSERVER */ 1358 splx(s); 1359 timeout(nfs_timer, (void *)0, nfs_ticks); 1360 } 1361 1362 /* 1363 * Test for a termination condition pending on the process. 1364 * This is used for NFSMNT_INT mounts. 1365 */ 1366 int 1367 nfs_sigintr(nmp, rep, p) 1368 struct nfsmount *nmp; 1369 struct nfsreq *rep; 1370 register struct proc *p; 1371 { 1372 1373 if (rep && (rep->r_flags & R_SOFTTERM)) 1374 return (EINTR); 1375 if (!(nmp->nm_flag & NFSMNT_INT)) 1376 return (0); 1377 if (p && p->p_siglist && 1378 (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) & 1379 NFSINT_SIGMASK)) 1380 return (EINTR); 1381 return (0); 1382 } 1383 1384 /* 1385 * Lock a socket against others. 1386 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1387 * and also to avoid race conditions between the processes with nfs requests 1388 * in progress when a reconnect is necessary. 1389 */ 1390 int 1391 nfs_sndlock(flagp, rep) 1392 register int *flagp; 1393 struct nfsreq *rep; 1394 { 1395 struct proc *p; 1396 int slpflag = 0, slptimeo = 0; 1397 1398 if (rep) { 1399 p = rep->r_procp; 1400 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1401 slpflag = PCATCH; 1402 } else 1403 p = (struct proc *)0; 1404 while (*flagp & NFSMNT_SNDLOCK) { 1405 if (nfs_sigintr(rep->r_nmp, rep, p)) 1406 return (EINTR); 1407 *flagp |= NFSMNT_WANTSND; 1408 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1409 slptimeo); 1410 if (slpflag == PCATCH) { 1411 slpflag = 0; 1412 slptimeo = 2 * hz; 1413 } 1414 } 1415 *flagp |= NFSMNT_SNDLOCK; 1416 return (0); 1417 } 1418 1419 /* 1420 * Unlock the stream socket for others. 1421 */ 1422 void 1423 nfs_sndunlock(flagp) 1424 register int *flagp; 1425 { 1426 1427 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1428 panic("nfs sndunlock"); 1429 *flagp &= ~NFSMNT_SNDLOCK; 1430 if (*flagp & NFSMNT_WANTSND) { 1431 *flagp &= ~NFSMNT_WANTSND; 1432 wakeup((caddr_t)flagp); 1433 } 1434 } 1435 1436 int 1437 nfs_rcvlock(rep) 1438 register struct nfsreq *rep; 1439 { 1440 register int *flagp = &rep->r_nmp->nm_flag; 1441 int slpflag, slptimeo = 0; 1442 1443 if (*flagp & NFSMNT_INT) 1444 slpflag = PCATCH; 1445 else 1446 slpflag = 0; 1447 while (*flagp & NFSMNT_RCVLOCK) { 1448 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1449 return (EINTR); 1450 *flagp |= NFSMNT_WANTRCV; 1451 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1452 slptimeo); 1453 if (slpflag == PCATCH) { 1454 slpflag = 0; 1455 slptimeo = 2 * hz; 1456 } 1457 } 1458 *flagp |= NFSMNT_RCVLOCK; 1459 return (0); 1460 } 1461 1462 /* 1463 * Unlock the stream socket for others. 1464 */ 1465 void 1466 nfs_rcvunlock(flagp) 1467 register int *flagp; 1468 { 1469 1470 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1471 panic("nfs rcvunlock"); 1472 *flagp &= ~NFSMNT_RCVLOCK; 1473 if (*flagp & NFSMNT_WANTRCV) { 1474 *flagp &= ~NFSMNT_WANTRCV; 1475 wakeup((caddr_t)flagp); 1476 } 1477 } 1478 1479 /* 1480 * Check for badly aligned mbuf data areas and 1481 * realign data in an mbuf list by copying the data areas up, as required. 1482 */ 1483 void 1484 nfs_realign(m, hsiz) 1485 register struct mbuf *m; 1486 int hsiz; 1487 { 1488 register struct mbuf *m2; 1489 register int siz, mlen, olen; 1490 register caddr_t tcp, fcp; 1491 struct mbuf *mnew; 1492 1493 while (m) { 1494 /* 1495 * This never happens for UDP, rarely happens for TCP 1496 * but frequently happens for iso transport. 1497 */ 1498 if ((m->m_len & 0x3) || (mtod(m, long) & 0x3)) { 1499 olen = m->m_len; 1500 fcp = mtod(m, caddr_t); 1501 if ((long)fcp & 0x3) { 1502 m->m_flags &= ~M_PKTHDR; 1503 if (m->m_flags & M_EXT) 1504 m->m_data = m->m_ext.ext_buf + 1505 ((m->m_ext.ext_size - olen) & ~0x3); 1506 else 1507 m->m_data = m->m_dat; 1508 } 1509 m->m_len = 0; 1510 tcp = mtod(m, caddr_t); 1511 mnew = m; 1512 m2 = m->m_next; 1513 1514 /* 1515 * If possible, only put the first invariant part 1516 * of the RPC header in the first mbuf. 1517 */ 1518 mlen = M_TRAILINGSPACE(m); 1519 if (olen <= hsiz && mlen > hsiz) 1520 mlen = hsiz; 1521 1522 /* 1523 * Loop through the mbuf list consolidating data. 1524 */ 1525 while (m) { 1526 while (olen > 0) { 1527 if (mlen == 0) { 1528 m2->m_flags &= ~M_PKTHDR; 1529 if (m2->m_flags & M_EXT) 1530 m2->m_data = m2->m_ext.ext_buf; 1531 else 1532 m2->m_data = m2->m_dat; 1533 m2->m_len = 0; 1534 mlen = M_TRAILINGSPACE(m2); 1535 tcp = mtod(m2, caddr_t); 1536 mnew = m2; 1537 m2 = m2->m_next; 1538 } 1539 siz = min(mlen, olen); 1540 if (tcp != fcp) 1541 bcopy(fcp, tcp, siz); 1542 mnew->m_len += siz; 1543 mlen -= siz; 1544 olen -= siz; 1545 tcp += siz; 1546 fcp += siz; 1547 } 1548 m = m->m_next; 1549 if (m) { 1550 olen = m->m_len; 1551 fcp = mtod(m, caddr_t); 1552 } 1553 } 1554 1555 /* 1556 * Finally, set m_len == 0 for any trailing mbufs that have 1557 * been copied out of. 1558 */ 1559 while (m2) { 1560 m2->m_len = 0; 1561 m2 = m2->m_next; 1562 } 1563 return; 1564 } 1565 m = m->m_next; 1566 } 1567 } 1568 1569 /* 1570 * Parse an RPC request 1571 * - verify it 1572 * - fill in the cred struct. 1573 */ 1574 int 1575 nfs_getreq(nd, nfsd, has_header) 1576 register struct nfsrv_descript *nd; 1577 struct nfsd *nfsd; 1578 int has_header; 1579 { 1580 register int len, i; 1581 register u_int32_t *tl; 1582 register int32_t t1; 1583 struct uio uio; 1584 struct iovec iov; 1585 caddr_t dpos, cp2, cp; 1586 u_int32_t nfsvers, auth_type; 1587 uid_t nickuid; 1588 int error = 0, nqnfs = 0, ticklen; 1589 struct mbuf *mrep, *md; 1590 register struct nfsuid *nuidp; 1591 struct timeval tvin, tvout; 1592 1593 mrep = nd->nd_mrep; 1594 md = nd->nd_md; 1595 dpos = nd->nd_dpos; 1596 if (has_header) { 1597 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1598 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1599 if (*tl++ != rpc_call) { 1600 m_freem(mrep); 1601 return (EBADRPC); 1602 } 1603 } else 1604 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1605 nd->nd_repstat = 0; 1606 nd->nd_flag = 0; 1607 if (*tl++ != rpc_vers) { 1608 nd->nd_repstat = ERPCMISMATCH; 1609 nd->nd_procnum = NFSPROC_NOOP; 1610 return (0); 1611 } 1612 if (*tl != nfs_prog) { 1613 if (*tl == nqnfs_prog) 1614 nqnfs++; 1615 else { 1616 nd->nd_repstat = EPROGUNAVAIL; 1617 nd->nd_procnum = NFSPROC_NOOP; 1618 return (0); 1619 } 1620 } 1621 tl++; 1622 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1623 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1624 (nfsvers != NQNFS_VER3 && nqnfs)) { 1625 nd->nd_repstat = EPROGMISMATCH; 1626 nd->nd_procnum = NFSPROC_NOOP; 1627 return (0); 1628 } 1629 if (nqnfs) 1630 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1631 else if (nfsvers == NFS_VER3) 1632 nd->nd_flag = ND_NFSV3; 1633 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1634 if (nd->nd_procnum == NFSPROC_NULL) 1635 return (0); 1636 if (nd->nd_procnum >= NFS_NPROCS || 1637 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1638 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1639 nd->nd_repstat = EPROCUNAVAIL; 1640 nd->nd_procnum = NFSPROC_NOOP; 1641 return (0); 1642 } 1643 if ((nd->nd_flag & ND_NFSV3) == 0) 1644 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1645 auth_type = *tl++; 1646 len = fxdr_unsigned(int, *tl++); 1647 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1648 m_freem(mrep); 1649 return (EBADRPC); 1650 } 1651 1652 nd->nd_flag &= ~ND_KERBAUTH; 1653 /* 1654 * Handle auth_unix or auth_kerb. 1655 */ 1656 if (auth_type == rpc_auth_unix) { 1657 len = fxdr_unsigned(int, *++tl); 1658 if (len < 0 || len > NFS_MAXNAMLEN) { 1659 m_freem(mrep); 1660 return (EBADRPC); 1661 } 1662 nfsm_adv(nfsm_rndup(len)); 1663 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1664 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred)); 1665 nd->nd_cr.cr_ref = 1; 1666 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1667 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1668 len = fxdr_unsigned(int, *tl); 1669 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1670 m_freem(mrep); 1671 return (EBADRPC); 1672 } 1673 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1674 for (i = 0; i < len; i++) 1675 if (i < NGROUPS) 1676 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1677 else 1678 tl++; 1679 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1680 if (nd->nd_cr.cr_ngroups > 1) 1681 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1682 len = fxdr_unsigned(int, *++tl); 1683 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1684 m_freem(mrep); 1685 return (EBADRPC); 1686 } 1687 if (len > 0) 1688 nfsm_adv(nfsm_rndup(len)); 1689 } else if (auth_type == rpc_auth_kerb) { 1690 switch (fxdr_unsigned(int, *tl++)) { 1691 case RPCAKN_FULLNAME: 1692 ticklen = fxdr_unsigned(int, *tl); 1693 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1694 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1695 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1696 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1697 m_freem(mrep); 1698 return (EBADRPC); 1699 } 1700 uio.uio_offset = 0; 1701 uio.uio_iov = &iov; 1702 uio.uio_iovcnt = 1; 1703 uio.uio_segflg = UIO_SYSSPACE; 1704 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1705 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1706 nfsm_mtouio(&uio, uio.uio_resid); 1707 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1708 if (*tl++ != rpc_auth_kerb || 1709 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1710 printf("Bad kerb verifier\n"); 1711 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1712 nd->nd_procnum = NFSPROC_NOOP; 1713 return (0); 1714 } 1715 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1716 tl = (u_int32_t *)cp; 1717 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1718 printf("Not fullname kerb verifier\n"); 1719 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1720 nd->nd_procnum = NFSPROC_NOOP; 1721 return (0); 1722 } 1723 cp += NFSX_UNSIGNED; 1724 bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED); 1725 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1726 nd->nd_flag |= ND_KERBFULL; 1727 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1728 break; 1729 case RPCAKN_NICKNAME: 1730 if (len != 2 * NFSX_UNSIGNED) { 1731 printf("Kerb nickname short\n"); 1732 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1733 nd->nd_procnum = NFSPROC_NOOP; 1734 return (0); 1735 } 1736 nickuid = fxdr_unsigned(uid_t, *tl); 1737 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1738 if (*tl++ != rpc_auth_kerb || 1739 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1740 printf("Kerb nick verifier bad\n"); 1741 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1742 nd->nd_procnum = NFSPROC_NOOP; 1743 return (0); 1744 } 1745 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1746 tvin.tv_sec = *tl++; 1747 tvin.tv_usec = *tl; 1748 1749 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; 1750 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1751 if (nuidp->nu_cr.cr_uid == nickuid && 1752 (!nd->nd_nam2 || 1753 netaddr_match(NU_NETFAM(nuidp), 1754 &nuidp->nu_haddr, nd->nd_nam2))) 1755 break; 1756 } 1757 if (!nuidp) { 1758 nd->nd_repstat = 1759 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1760 nd->nd_procnum = NFSPROC_NOOP; 1761 return (0); 1762 } 1763 1764 /* 1765 * Now, decrypt the timestamp using the session key 1766 * and validate it. 1767 */ 1768 #ifdef NFSKERB 1769 XXX 1770 #endif 1771 1772 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1773 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1774 if (nuidp->nu_expire < time.tv_sec || 1775 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1776 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1777 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1778 nuidp->nu_expire = 0; 1779 nd->nd_repstat = 1780 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1781 nd->nd_procnum = NFSPROC_NOOP; 1782 return (0); 1783 } 1784 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1785 nd->nd_flag |= ND_KERBNICK; 1786 }; 1787 } else { 1788 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1789 nd->nd_procnum = NFSPROC_NOOP; 1790 return (0); 1791 } 1792 1793 /* 1794 * For nqnfs, get piggybacked lease request. 1795 */ 1796 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1797 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1798 nd->nd_flag |= fxdr_unsigned(int, *tl); 1799 if (nd->nd_flag & ND_LEASE) { 1800 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1801 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 1802 } else 1803 nd->nd_duration = NQ_MINLEASE; 1804 } else 1805 nd->nd_duration = NQ_MINLEASE; 1806 nd->nd_md = md; 1807 nd->nd_dpos = dpos; 1808 return (0); 1809 nfsmout: 1810 return (error); 1811 } 1812 1813 int 1814 nfs_msg(p, server, msg) 1815 struct proc *p; 1816 char *server, *msg; 1817 { 1818 tpr_t tpr; 1819 1820 if (p) 1821 tpr = tprintf_open(p); 1822 else 1823 tpr = NULL; 1824 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1825 tprintf_close(tpr); 1826 return (0); 1827 } 1828 1829 #ifdef NFSSERVER 1830 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 1831 struct nfssvc_sock *, struct proc *, 1832 struct mbuf **)) = { 1833 nfsrv_null, 1834 nfsrv_getattr, 1835 nfsrv_setattr, 1836 nfsrv_lookup, 1837 nfsrv3_access, 1838 nfsrv_readlink, 1839 nfsrv_read, 1840 nfsrv_write, 1841 nfsrv_create, 1842 nfsrv_mkdir, 1843 nfsrv_symlink, 1844 nfsrv_mknod, 1845 nfsrv_remove, 1846 nfsrv_rmdir, 1847 nfsrv_rename, 1848 nfsrv_link, 1849 nfsrv_readdir, 1850 nfsrv_readdirplus, 1851 nfsrv_statfs, 1852 nfsrv_fsinfo, 1853 nfsrv_pathconf, 1854 nfsrv_commit, 1855 nqnfsrv_getlease, 1856 nqnfsrv_vacated, 1857 nfsrv_noop, 1858 nfsrv_noop 1859 }; 1860 1861 /* 1862 * Socket upcall routine for the nfsd sockets. 1863 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1864 * Essentially do as much as possible non-blocking, else punt and it will 1865 * be called with M_WAIT from an nfsd. 1866 */ 1867 void 1868 nfsrv_rcv(so, arg, waitflag) 1869 struct socket *so; 1870 caddr_t arg; 1871 int waitflag; 1872 { 1873 register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1874 register struct mbuf *m; 1875 struct mbuf *mp, *nam; 1876 struct uio auio; 1877 int flags, error; 1878 1879 if ((slp->ns_flag & SLP_VALID) == 0) 1880 return; 1881 #ifdef notdef 1882 /* 1883 * Define this to test for nfsds handling this under heavy load. 1884 */ 1885 if (waitflag == M_DONTWAIT) { 1886 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1887 } 1888 #endif 1889 auio.uio_procp = NULL; 1890 if (so->so_type == SOCK_STREAM) { 1891 /* 1892 * If there are already records on the queue, defer soreceive() 1893 * to an nfsd so that there is feedback to the TCP layer that 1894 * the nfs servers are heavily loaded. 1895 */ 1896 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1897 slp->ns_flag |= SLP_NEEDQ; 1898 goto dorecs; 1899 } 1900 1901 /* 1902 * Do soreceive(). 1903 */ 1904 auio.uio_resid = 1000000000; 1905 flags = MSG_DONTWAIT; 1906 error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1907 if (error || mp == (struct mbuf *)0) { 1908 if (error == EWOULDBLOCK) 1909 slp->ns_flag |= SLP_NEEDQ; 1910 else 1911 slp->ns_flag |= SLP_DISCONN; 1912 goto dorecs; 1913 } 1914 m = mp; 1915 if (slp->ns_rawend) { 1916 slp->ns_rawend->m_next = m; 1917 slp->ns_cc += 1000000000 - auio.uio_resid; 1918 } else { 1919 slp->ns_raw = m; 1920 slp->ns_cc = 1000000000 - auio.uio_resid; 1921 } 1922 while (m->m_next) 1923 m = m->m_next; 1924 slp->ns_rawend = m; 1925 1926 /* 1927 * Now try and parse record(s) out of the raw stream data. 1928 */ 1929 error = nfsrv_getstream(slp, waitflag); 1930 if (error) { 1931 if (error == EPERM) 1932 slp->ns_flag |= SLP_DISCONN; 1933 else 1934 slp->ns_flag |= SLP_NEEDQ; 1935 } 1936 } else { 1937 do { 1938 auio.uio_resid = 1000000000; 1939 flags = MSG_DONTWAIT; 1940 error = soreceive(so, &nam, &auio, &mp, 1941 (struct mbuf **)0, &flags); 1942 if (mp) { 1943 nfs_realign(mp, 10 * NFSX_UNSIGNED); 1944 if (nam) { 1945 m = nam; 1946 m->m_next = mp; 1947 } else 1948 m = mp; 1949 if (slp->ns_recend) 1950 slp->ns_recend->m_nextpkt = m; 1951 else 1952 slp->ns_rec = m; 1953 slp->ns_recend = m; 1954 m->m_nextpkt = (struct mbuf *)0; 1955 } 1956 if (error) { 1957 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1958 && error != EWOULDBLOCK) { 1959 slp->ns_flag |= SLP_DISCONN; 1960 goto dorecs; 1961 } 1962 } 1963 } while (mp); 1964 } 1965 1966 /* 1967 * Now try and process the request records, non-blocking. 1968 */ 1969 dorecs: 1970 if (waitflag == M_DONTWAIT && 1971 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1972 nfsrv_wakenfsd(slp); 1973 } 1974 1975 /* 1976 * Try and extract an RPC request from the mbuf data list received on a 1977 * stream socket. The "waitflag" argument indicates whether or not it 1978 * can sleep. 1979 */ 1980 int 1981 nfsrv_getstream(slp, waitflag) 1982 register struct nfssvc_sock *slp; 1983 int waitflag; 1984 { 1985 register struct mbuf *m, **mpp; 1986 register char *cp1, *cp2; 1987 register int len; 1988 struct mbuf *om, *m2, *recm = NULL; 1989 u_int32_t recmark; 1990 1991 if (slp->ns_flag & SLP_GETSTREAM) 1992 panic("nfs getstream"); 1993 slp->ns_flag |= SLP_GETSTREAM; 1994 for (;;) { 1995 if (slp->ns_reclen == 0) { 1996 if (slp->ns_cc < NFSX_UNSIGNED) { 1997 slp->ns_flag &= ~SLP_GETSTREAM; 1998 return (0); 1999 } 2000 m = slp->ns_raw; 2001 if (m->m_len >= NFSX_UNSIGNED) { 2002 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 2003 m->m_data += NFSX_UNSIGNED; 2004 m->m_len -= NFSX_UNSIGNED; 2005 } else { 2006 cp1 = (caddr_t)&recmark; 2007 cp2 = mtod(m, caddr_t); 2008 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 2009 while (m->m_len == 0) { 2010 m = m->m_next; 2011 cp2 = mtod(m, caddr_t); 2012 } 2013 *cp1++ = *cp2++; 2014 m->m_data++; 2015 m->m_len--; 2016 } 2017 } 2018 slp->ns_cc -= NFSX_UNSIGNED; 2019 recmark = ntohl(recmark); 2020 slp->ns_reclen = recmark & ~0x80000000; 2021 if (recmark & 0x80000000) 2022 slp->ns_flag |= SLP_LASTFRAG; 2023 else 2024 slp->ns_flag &= ~SLP_LASTFRAG; 2025 if (slp->ns_reclen > NFS_MAXPACKET) { 2026 slp->ns_flag &= ~SLP_GETSTREAM; 2027 return (EPERM); 2028 } 2029 } 2030 2031 /* 2032 * Now get the record part. 2033 */ 2034 if (slp->ns_cc == slp->ns_reclen) { 2035 recm = slp->ns_raw; 2036 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2037 slp->ns_cc = slp->ns_reclen = 0; 2038 } else if (slp->ns_cc > slp->ns_reclen) { 2039 len = 0; 2040 m = slp->ns_raw; 2041 om = (struct mbuf *)0; 2042 while (len < slp->ns_reclen) { 2043 if ((len + m->m_len) > slp->ns_reclen) { 2044 m2 = m_copym(m, 0, slp->ns_reclen - len, 2045 waitflag); 2046 if (m2) { 2047 if (om) { 2048 om->m_next = m2; 2049 recm = slp->ns_raw; 2050 } else 2051 recm = m2; 2052 m->m_data += slp->ns_reclen - len; 2053 m->m_len -= slp->ns_reclen - len; 2054 len = slp->ns_reclen; 2055 } else { 2056 slp->ns_flag &= ~SLP_GETSTREAM; 2057 return (EWOULDBLOCK); 2058 } 2059 } else if ((len + m->m_len) == slp->ns_reclen) { 2060 om = m; 2061 len += m->m_len; 2062 m = m->m_next; 2063 recm = slp->ns_raw; 2064 om->m_next = (struct mbuf *)0; 2065 } else { 2066 om = m; 2067 len += m->m_len; 2068 m = m->m_next; 2069 } 2070 } 2071 slp->ns_raw = m; 2072 slp->ns_cc -= len; 2073 slp->ns_reclen = 0; 2074 } else { 2075 slp->ns_flag &= ~SLP_GETSTREAM; 2076 return (0); 2077 } 2078 2079 /* 2080 * Accumulate the fragments into a record. 2081 */ 2082 mpp = &slp->ns_frag; 2083 while (*mpp) 2084 mpp = &((*mpp)->m_next); 2085 *mpp = recm; 2086 if (slp->ns_flag & SLP_LASTFRAG) { 2087 nfs_realign(slp->ns_frag, 10 * NFSX_UNSIGNED); 2088 if (slp->ns_recend) 2089 slp->ns_recend->m_nextpkt = slp->ns_frag; 2090 else 2091 slp->ns_rec = slp->ns_frag; 2092 slp->ns_recend = slp->ns_frag; 2093 slp->ns_frag = (struct mbuf *)0; 2094 } 2095 } 2096 } 2097 2098 /* 2099 * Parse an RPC header. 2100 */ 2101 int 2102 nfsrv_dorec(slp, nfsd, ndp) 2103 register struct nfssvc_sock *slp; 2104 struct nfsd *nfsd; 2105 struct nfsrv_descript **ndp; 2106 { 2107 register struct mbuf *m, *nam; 2108 register struct nfsrv_descript *nd; 2109 int error; 2110 2111 *ndp = NULL; 2112 if ((slp->ns_flag & SLP_VALID) == 0 || 2113 (m = slp->ns_rec) == (struct mbuf *)0) 2114 return (ENOBUFS); 2115 slp->ns_rec = m->m_nextpkt; 2116 if (slp->ns_rec) 2117 m->m_nextpkt = (struct mbuf *)0; 2118 else 2119 slp->ns_recend = (struct mbuf *)0; 2120 if (m->m_type == MT_SONAME) { 2121 nam = m; 2122 m = m->m_next; 2123 nam->m_next = NULL; 2124 } else 2125 nam = NULL; 2126 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), 2127 M_NFSRVDESC, M_WAITOK); 2128 nd->nd_md = nd->nd_mrep = m; 2129 nd->nd_nam2 = nam; 2130 nd->nd_dpos = mtod(m, caddr_t); 2131 error = nfs_getreq(nd, nfsd, TRUE); 2132 if (error) { 2133 m_freem(nam); 2134 free((caddr_t)nd, M_NFSRVDESC); 2135 return (error); 2136 } 2137 *ndp = nd; 2138 nfsd->nfsd_nd = nd; 2139 return (0); 2140 } 2141 2142 2143 /* 2144 * Search for a sleeping nfsd and wake it up. 2145 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2146 * running nfsds will go look for the work in the nfssvc_sock list. 2147 */ 2148 void 2149 nfsrv_wakenfsd(slp) 2150 struct nfssvc_sock *slp; 2151 { 2152 register struct nfsd *nd; 2153 2154 if ((slp->ns_flag & SLP_VALID) == 0) 2155 return; 2156 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) { 2157 if (nd->nfsd_flag & NFSD_WAITING) { 2158 nd->nfsd_flag &= ~NFSD_WAITING; 2159 if (nd->nfsd_slp) 2160 panic("nfsd wakeup"); 2161 slp->ns_sref++; 2162 nd->nfsd_slp = slp; 2163 wakeup((caddr_t)nd); 2164 return; 2165 } 2166 } 2167 slp->ns_flag |= SLP_DOREC; 2168 nfsd_head_flag |= NFSD_CHECKSLP; 2169 } 2170 #endif /* NFSSERVER */ 2171
Indexes created Tue Sep 30 11:09:46 GMT 2025