nfs_socket.c revision 1.24
1 /* $NetBSD: nfs_socket.c,v 1.24 1996/02/18 11:53:48 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 = nmp->nm_rsize + NFS_MAXPKTHDR; 233 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 234 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 235 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR) * 2; 236 } else { 237 if (nmp->nm_sotype != SOCK_STREAM) 238 panic("nfscon sotype"); 239 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 240 MGET(m, M_WAIT, MT_SOOPTS); 241 *mtod(m, int32_t *) = 1; 242 m->m_len = sizeof(int32_t); 243 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 244 } 245 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 246 MGET(m, M_WAIT, MT_SOOPTS); 247 *mtod(m, int32_t *) = 1; 248 m->m_len = sizeof(int32_t); 249 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 250 } 251 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 252 sizeof (u_int32_t)) * 2; 253 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 254 sizeof (u_int32_t)) * 2; 255 } 256 error = soreserve(so, sndreserve, rcvreserve); 257 if (error) 258 goto bad; 259 so->so_rcv.sb_flags |= SB_NOINTR; 260 so->so_snd.sb_flags |= SB_NOINTR; 261 262 /* Initialize other non-zero congestion variables */ 263 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 264 nmp->nm_srtt[4] = (NFS_TIMEO << 3); 265 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 266 nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0; 267 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 268 nmp->nm_sent = 0; 269 nmp->nm_timeouts = 0; 270 return (0); 271 272 bad: 273 nfs_disconnect(nmp); 274 return (error); 275 } 276 277 /* 278 * Reconnect routine: 279 * Called when a connection is broken on a reliable protocol. 280 * - clean up the old socket 281 * - nfs_connect() again 282 * - set R_MUSTRESEND for all outstanding requests on mount point 283 * If this fails the mount point is DEAD! 284 * nb: Must be called with the nfs_sndlock() set on the mount point. 285 */ 286 int 287 nfs_reconnect(rep) 288 register struct nfsreq *rep; 289 { 290 register struct nfsreq *rp; 291 register struct nfsmount *nmp = rep->r_nmp; 292 int error; 293 294 nfs_disconnect(nmp); 295 while ((error = nfs_connect(nmp, rep)) != 0) { 296 if (error == EINTR || error == ERESTART) 297 return (EINTR); 298 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); 299 } 300 301 /* 302 * Loop through outstanding request list and fix up all requests 303 * on old socket. 304 */ 305 for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) { 306 if (rp->r_nmp == nmp) 307 rp->r_flags |= R_MUSTRESEND; 308 } 309 return (0); 310 } 311 312 /* 313 * NFS disconnect. Clean up and unlink. 314 */ 315 void 316 nfs_disconnect(nmp) 317 register struct nfsmount *nmp; 318 { 319 register struct socket *so; 320 321 if (nmp->nm_so) { 322 so = nmp->nm_so; 323 nmp->nm_so = (struct socket *)0; 324 soshutdown(so, 2); 325 soclose(so); 326 } 327 } 328 329 /* 330 * This is the nfs send routine. For connection based socket types, it 331 * must be called with an nfs_sndlock() on the socket. 332 * "rep == NULL" indicates that it has been called from a server. 333 * For the client side: 334 * - return EINTR if the RPC is terminated, 0 otherwise 335 * - set R_MUSTRESEND if the send fails for any reason 336 * - do any cleanup required by recoverable socket errors (???) 337 * For the server side: 338 * - return EINTR or ERESTART if interrupted by a signal 339 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 340 * - do any cleanup required by recoverable socket errors (???) 341 */ 342 int 343 nfs_send(so, nam, top, rep) 344 register struct socket *so; 345 struct mbuf *nam; 346 register struct mbuf *top; 347 struct nfsreq *rep; 348 { 349 struct mbuf *sendnam; 350 int error, soflags, flags; 351 352 if (rep) { 353 if (rep->r_flags & R_SOFTTERM) { 354 m_freem(top); 355 return (EINTR); 356 } 357 if ((so = rep->r_nmp->nm_so) == NULL) { 358 rep->r_flags |= R_MUSTRESEND; 359 m_freem(top); 360 return (0); 361 } 362 rep->r_flags &= ~R_MUSTRESEND; 363 soflags = rep->r_nmp->nm_soflags; 364 } else 365 soflags = so->so_proto->pr_flags; 366 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 367 sendnam = (struct mbuf *)0; 368 else 369 sendnam = nam; 370 if (so->so_type == SOCK_SEQPACKET) 371 flags = MSG_EOR; 372 else 373 flags = 0; 374 375 error = sosend(so, sendnam, (struct uio *)0, top, 376 (struct mbuf *)0, flags); 377 if (error) { 378 if (rep) { 379 log(LOG_INFO, "nfs send error %d for server %s\n",error, 380 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 381 /* 382 * Deal with errors for the client side. 383 */ 384 if (rep->r_flags & R_SOFTTERM) 385 error = EINTR; 386 else 387 rep->r_flags |= R_MUSTRESEND; 388 } else 389 log(LOG_INFO, "nfsd send error %d\n", error); 390 391 /* 392 * Handle any recoverable (soft) socket errors here. (???) 393 */ 394 if (error != EINTR && error != ERESTART && 395 error != EWOULDBLOCK && error != EPIPE) 396 error = 0; 397 } 398 return (error); 399 } 400 401 #ifdef NFSCLIENT 402 /* 403 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 404 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 405 * Mark and consolidate the data into a new mbuf list. 406 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 407 * small mbufs. 408 * For SOCK_STREAM we must be very careful to read an entire record once 409 * we have read any of it, even if the system call has been interrupted. 410 */ 411 int 412 nfs_receive(rep, aname, mp) 413 register struct nfsreq *rep; 414 struct mbuf **aname; 415 struct mbuf **mp; 416 { 417 register struct socket *so; 418 struct uio auio; 419 struct iovec aio; 420 register struct mbuf *m; 421 struct mbuf *control; 422 u_int32_t len; 423 struct mbuf **getnam; 424 int error, sotype, rcvflg; 425 struct proc *p = curproc; /* XXX */ 426 427 /* 428 * Set up arguments for soreceive() 429 */ 430 *mp = (struct mbuf *)0; 431 *aname = (struct mbuf *)0; 432 sotype = rep->r_nmp->nm_sotype; 433 434 /* 435 * For reliable protocols, lock against other senders/receivers 436 * in case a reconnect is necessary. 437 * For SOCK_STREAM, first get the Record Mark to find out how much 438 * more there is to get. 439 * We must lock the socket against other receivers 440 * until we have an entire rpc request/reply. 441 */ 442 if (sotype != SOCK_DGRAM) { 443 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 444 if (error) 445 return (error); 446 tryagain: 447 /* 448 * Check for fatal errors and resending request. 449 */ 450 /* 451 * Ugh: If a reconnect attempt just happened, nm_so 452 * would have changed. NULL indicates a failed 453 * attempt that has essentially shut down this 454 * mount point. 455 */ 456 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 457 nfs_sndunlock(&rep->r_nmp->nm_flag); 458 return (EINTR); 459 } 460 so = rep->r_nmp->nm_so; 461 if (!so) { 462 error = nfs_reconnect(rep); 463 if (error) { 464 nfs_sndunlock(&rep->r_nmp->nm_flag); 465 return (error); 466 } 467 goto tryagain; 468 } 469 while (rep->r_flags & R_MUSTRESEND) { 470 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 471 nfsstats.rpcretries++; 472 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 473 if (error) { 474 if (error == EINTR || error == ERESTART || 475 (error = nfs_reconnect(rep)) != 0) { 476 nfs_sndunlock(&rep->r_nmp->nm_flag); 477 return (error); 478 } 479 goto tryagain; 480 } 481 } 482 nfs_sndunlock(&rep->r_nmp->nm_flag); 483 if (sotype == SOCK_STREAM) { 484 aio.iov_base = (caddr_t) &len; 485 aio.iov_len = sizeof(u_int32_t); 486 auio.uio_iov = &aio; 487 auio.uio_iovcnt = 1; 488 auio.uio_segflg = UIO_SYSSPACE; 489 auio.uio_rw = UIO_READ; 490 auio.uio_offset = 0; 491 auio.uio_resid = sizeof(u_int32_t); 492 auio.uio_procp = p; 493 do { 494 rcvflg = MSG_WAITALL; 495 error = soreceive(so, (struct mbuf **)0, &auio, 496 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 497 if (error == EWOULDBLOCK && rep) { 498 if (rep->r_flags & R_SOFTTERM) 499 return (EINTR); 500 } 501 } while (error == EWOULDBLOCK); 502 if (!error && auio.uio_resid > 0) { 503 log(LOG_INFO, 504 "short receive (%d/%d) from nfs server %s\n", 505 sizeof(u_int32_t) - auio.uio_resid, 506 sizeof(u_int32_t), 507 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 508 error = EPIPE; 509 } 510 if (error) 511 goto errout; 512 len = ntohl(len) & ~0x80000000; 513 /* 514 * This is SERIOUS! We are out of sync with the sender 515 * and forcing a disconnect/reconnect is all I can do. 516 */ 517 if (len > NFS_MAXPACKET) { 518 log(LOG_ERR, "%s (%d) from nfs server %s\n", 519 "impossible packet length", 520 len, 521 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 522 error = EFBIG; 523 goto errout; 524 } 525 auio.uio_resid = len; 526 do { 527 rcvflg = MSG_WAITALL; 528 error = soreceive(so, (struct mbuf **)0, 529 &auio, mp, (struct mbuf **)0, &rcvflg); 530 } while (error == EWOULDBLOCK || error == EINTR || 531 error == ERESTART); 532 if (!error && auio.uio_resid > 0) { 533 log(LOG_INFO, 534 "short receive (%d/%d) from nfs server %s\n", 535 len - auio.uio_resid, len, 536 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 537 error = EPIPE; 538 } 539 } else { 540 /* 541 * NB: Since uio_resid is big, MSG_WAITALL is ignored 542 * and soreceive() will return when it has either a 543 * control msg or a data msg. 544 * We have no use for control msg., but must grab them 545 * and then throw them away so we know what is going 546 * on. 547 */ 548 auio.uio_resid = len = 100000000; /* Anything Big */ 549 auio.uio_procp = p; 550 do { 551 rcvflg = 0; 552 error = soreceive(so, (struct mbuf **)0, 553 &auio, mp, &control, &rcvflg); 554 if (control) 555 m_freem(control); 556 if (error == EWOULDBLOCK && rep) { 557 if (rep->r_flags & R_SOFTTERM) 558 return (EINTR); 559 } 560 } while (error == EWOULDBLOCK || 561 (!error && *mp == NULL && control)); 562 if ((rcvflg & MSG_EOR) == 0) 563 printf("Egad!!\n"); 564 if (!error && *mp == NULL) 565 error = EPIPE; 566 len -= auio.uio_resid; 567 } 568 errout: 569 if (error && error != EINTR && error != ERESTART) { 570 m_freem(*mp); 571 *mp = (struct mbuf *)0; 572 if (error != EPIPE) 573 log(LOG_INFO, 574 "receive error %d from nfs server %s\n", 575 error, 576 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 577 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 578 if (!error) 579 error = nfs_reconnect(rep); 580 if (!error) 581 goto tryagain; 582 } 583 } else { 584 if ((so = rep->r_nmp->nm_so) == NULL) 585 return (EACCES); 586 if (so->so_state & SS_ISCONNECTED) 587 getnam = (struct mbuf **)0; 588 else 589 getnam = aname; 590 auio.uio_resid = len = 1000000; 591 auio.uio_procp = p; 592 do { 593 rcvflg = 0; 594 error = soreceive(so, getnam, &auio, mp, 595 (struct mbuf **)0, &rcvflg); 596 if (error == EWOULDBLOCK && 597 (rep->r_flags & R_SOFTTERM)) 598 return (EINTR); 599 } while (error == EWOULDBLOCK); 600 len -= auio.uio_resid; 601 } 602 if (error) { 603 m_freem(*mp); 604 *mp = (struct mbuf *)0; 605 } 606 /* 607 * Search for any mbufs that are not a multiple of 4 bytes long 608 * or with m_data not longword aligned. 609 * These could cause pointer alignment problems, so copy them to 610 * well aligned mbufs. 611 */ 612 nfs_realign(*mp, 5 * NFSX_UNSIGNED); 613 return (error); 614 } 615 616 /* 617 * Implement receipt of reply on a socket. 618 * We must search through the list of received datagrams matching them 619 * with outstanding requests using the xid, until ours is found. 620 */ 621 /* ARGSUSED */ 622 int 623 nfs_reply(myrep) 624 struct nfsreq *myrep; 625 { 626 register struct nfsreq *rep; 627 register struct nfsmount *nmp = myrep->r_nmp; 628 register int32_t t1; 629 struct mbuf *mrep, *nam, *md; 630 u_int32_t rxid, *tl; 631 caddr_t dpos, cp2; 632 int error; 633 634 /* 635 * Loop around until we get our own reply 636 */ 637 for (;;) { 638 /* 639 * Lock against other receivers so that I don't get stuck in 640 * sbwait() after someone else has received my reply for me. 641 * Also necessary for connection based protocols to avoid 642 * race conditions during a reconnect. 643 */ 644 error = nfs_rcvlock(myrep); 645 if (error) 646 return (error); 647 /* Already received, bye bye */ 648 if (myrep->r_mrep != NULL) { 649 nfs_rcvunlock(&nmp->nm_flag); 650 return (0); 651 } 652 /* 653 * Get the next Rpc reply off the socket 654 */ 655 error = nfs_receive(myrep, &nam, &mrep); 656 nfs_rcvunlock(&nmp->nm_flag); 657 if (error) { 658 659 /* 660 * Ignore routing errors on connectionless protocols?? 661 */ 662 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 663 nmp->nm_so->so_error = 0; 664 if (myrep->r_flags & R_GETONEREP) 665 return (0); 666 continue; 667 } 668 return (error); 669 } 670 if (nam) 671 m_freem(nam); 672 673 /* 674 * Get the xid and check that it is an rpc reply 675 */ 676 md = mrep; 677 dpos = mtod(md, caddr_t); 678 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 679 rxid = *tl++; 680 if (*tl != rpc_reply) { 681 if (nmp->nm_flag & NFSMNT_NQNFS) { 682 if (nqnfs_callback(nmp, mrep, md, dpos)) 683 nfsstats.rpcinvalid++; 684 } else { 685 nfsstats.rpcinvalid++; 686 m_freem(mrep); 687 } 688 nfsmout: 689 if (myrep->r_flags & R_GETONEREP) 690 return (0); 691 continue; 692 } 693 694 /* 695 * Loop through the request list to match up the reply 696 * Iff no match, just drop the datagram 697 */ 698 for (rep = nfs_reqq.tqh_first; rep != 0; 699 rep = rep->r_chain.tqe_next) { 700 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 701 /* Found it.. */ 702 rep->r_mrep = mrep; 703 rep->r_md = md; 704 rep->r_dpos = dpos; 705 if (nfsrtton) { 706 struct rttl *rt; 707 708 rt = &nfsrtt.rttl[nfsrtt.pos]; 709 rt->proc = rep->r_procnum; 710 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 711 rt->sent = nmp->nm_sent; 712 rt->cwnd = nmp->nm_cwnd; 713 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 714 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 715 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 716 rt->tstamp = time; 717 if (rep->r_flags & R_TIMING) 718 rt->rtt = rep->r_rtt; 719 else 720 rt->rtt = 1000000; 721 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 722 } 723 /* 724 * Update congestion window. 725 * Do the additive increase of 726 * one rpc/rtt. 727 */ 728 if (nmp->nm_cwnd <= nmp->nm_sent) { 729 nmp->nm_cwnd += 730 (NFS_CWNDSCALE * NFS_CWNDSCALE + 731 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 732 if (nmp->nm_cwnd > NFS_MAXCWND) 733 nmp->nm_cwnd = NFS_MAXCWND; 734 } 735 rep->r_flags &= ~R_SENT; 736 nmp->nm_sent -= NFS_CWNDSCALE; 737 /* 738 * Update rtt using a gain of 0.125 on the mean 739 * and a gain of 0.25 on the deviation. 740 */ 741 if (rep->r_flags & R_TIMING) { 742 /* 743 * Since the timer resolution of 744 * NFS_HZ is so course, it can often 745 * result in r_rtt == 0. Since 746 * r_rtt == N means that the actual 747 * rtt is between N+dt and N+2-dt ticks, 748 * add 1. 749 */ 750 t1 = rep->r_rtt + 1; 751 t1 -= (NFS_SRTT(rep) >> 3); 752 NFS_SRTT(rep) += t1; 753 if (t1 < 0) 754 t1 = -t1; 755 t1 -= (NFS_SDRTT(rep) >> 2); 756 NFS_SDRTT(rep) += t1; 757 } 758 nmp->nm_timeouts = 0; 759 break; 760 } 761 } 762 /* 763 * If not matched to a request, drop it. 764 * If it's mine, get out. 765 */ 766 if (rep == 0) { 767 nfsstats.rpcunexpected++; 768 m_freem(mrep); 769 } else if (rep == myrep) { 770 if (rep->r_mrep == NULL) 771 panic("nfsreply nil"); 772 return (0); 773 } 774 if (myrep->r_flags & R_GETONEREP) 775 return (0); 776 } 777 } 778 779 /* 780 * nfs_request - goes something like this 781 * - fill in request struct 782 * - links it into list 783 * - calls nfs_send() for first transmit 784 * - calls nfs_receive() to get reply 785 * - break down rpc header and return with nfs reply pointed to 786 * by mrep or error 787 * nb: always frees up mreq mbuf list 788 */ 789 int 790 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 791 struct vnode *vp; 792 struct mbuf *mrest; 793 int procnum; 794 struct proc *procp; 795 struct ucred *cred; 796 struct mbuf **mrp; 797 struct mbuf **mdp; 798 caddr_t *dposp; 799 { 800 register struct mbuf *m, *mrep; 801 register struct nfsreq *rep; 802 register u_int32_t *tl; 803 register int i; 804 struct nfsmount *nmp; 805 struct mbuf *md, *mheadend; 806 struct nfsnode *np; 807 char nickv[RPCX_NICKVERF]; 808 time_t reqtime, waituntil; 809 caddr_t dpos, cp2; 810 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; 811 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 812 int verf_len, verf_type; 813 u_int32_t xid; 814 u_quad_t frev; 815 char *auth_str, *verf_str; 816 NFSKERBKEY_T key; /* save session key */ 817 818 nmp = VFSTONFS(vp->v_mount); 819 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 820 rep->r_nmp = nmp; 821 rep->r_vp = vp; 822 rep->r_procp = procp; 823 rep->r_procnum = procnum; 824 i = 0; 825 m = mrest; 826 while (m) { 827 i += m->m_len; 828 m = m->m_next; 829 } 830 mrest_len = i; 831 832 /* 833 * Get the RPC header with authorization. 834 */ 835 kerbauth: 836 verf_str = auth_str = (char *)0; 837 if (nmp->nm_flag & NFSMNT_KERB) { 838 verf_str = nickv; 839 verf_len = sizeof (nickv); 840 auth_type = RPCAUTH_KERB4; 841 bzero((caddr_t)key, sizeof (key)); 842 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 843 &auth_len, verf_str, verf_len)) { 844 error = nfs_getauth(nmp, rep, cred, &auth_str, 845 &auth_len, verf_str, &verf_len, key); 846 if (error) { 847 free((caddr_t)rep, M_NFSREQ); 848 m_freem(mrest); 849 return (error); 850 } 851 } 852 } else { 853 auth_type = RPCAUTH_UNIX; 854 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 855 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 856 5 * NFSX_UNSIGNED; 857 } 858 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 859 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 860 if (auth_str) 861 free(auth_str, M_TEMP); 862 863 /* 864 * For stream protocols, insert a Sun RPC Record Mark. 865 */ 866 if (nmp->nm_sotype == SOCK_STREAM) { 867 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 868 *mtod(m, u_int32_t *) = htonl(0x80000000 | 869 (m->m_pkthdr.len - NFSX_UNSIGNED)); 870 } 871 rep->r_mreq = m; 872 rep->r_xid = xid; 873 tryagain: 874 if (nmp->nm_flag & NFSMNT_SOFT) 875 rep->r_retry = nmp->nm_retry; 876 else 877 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 878 rep->r_rtt = rep->r_rexmit = 0; 879 if (proct[procnum] > 0) 880 rep->r_flags = R_TIMING; 881 else 882 rep->r_flags = 0; 883 rep->r_mrep = NULL; 884 885 /* 886 * Do the client side RPC. 887 */ 888 nfsstats.rpcrequests++; 889 /* 890 * Chain request into list of outstanding requests. Be sure 891 * to put it LAST so timer finds oldest requests first. 892 */ 893 s = splsoftclock(); 894 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 895 896 /* Get send time for nqnfs */ 897 reqtime = time.tv_sec; 898 899 /* 900 * If backing off another request or avoiding congestion, don't 901 * send this one now but let timer do it. If not timing a request, 902 * do it now. 903 */ 904 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 905 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 906 nmp->nm_sent < nmp->nm_cwnd)) { 907 splx(s); 908 if (nmp->nm_soflags & PR_CONNREQUIRED) 909 error = nfs_sndlock(&nmp->nm_flag, rep); 910 if (!error) { 911 m = m_copym(m, 0, M_COPYALL, M_WAIT); 912 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 913 if (nmp->nm_soflags & PR_CONNREQUIRED) 914 nfs_sndunlock(&nmp->nm_flag); 915 } 916 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 917 nmp->nm_sent += NFS_CWNDSCALE; 918 rep->r_flags |= R_SENT; 919 } 920 } else { 921 splx(s); 922 rep->r_rtt = -1; 923 } 924 925 /* 926 * Wait for the reply from our send or the timer's. 927 */ 928 if (!error || error == EPIPE) 929 error = nfs_reply(rep); 930 931 /* 932 * RPC done, unlink the request. 933 */ 934 s = splsoftclock(); 935 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 936 splx(s); 937 938 /* 939 * Decrement the outstanding request count. 940 */ 941 if (rep->r_flags & R_SENT) { 942 rep->r_flags &= ~R_SENT; /* paranoia */ 943 nmp->nm_sent -= NFS_CWNDSCALE; 944 } 945 946 /* 947 * If there was a successful reply and a tprintf msg. 948 * tprintf a response. 949 */ 950 if (!error && (rep->r_flags & R_TPRINTFMSG)) 951 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 952 "is alive again"); 953 mrep = rep->r_mrep; 954 md = rep->r_md; 955 dpos = rep->r_dpos; 956 if (error) { 957 m_freem(rep->r_mreq); 958 free((caddr_t)rep, M_NFSREQ); 959 return (error); 960 } 961 962 /* 963 * break down the rpc header and check if ok 964 */ 965 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 966 if (*tl++ == rpc_msgdenied) { 967 if (*tl == rpc_mismatch) 968 error = EOPNOTSUPP; 969 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 970 if (!failed_auth) { 971 failed_auth++; 972 mheadend->m_next = (struct mbuf *)0; 973 m_freem(mrep); 974 m_freem(rep->r_mreq); 975 goto kerbauth; 976 } else 977 error = EAUTH; 978 } else 979 error = EACCES; 980 m_freem(mrep); 981 m_freem(rep->r_mreq); 982 free((caddr_t)rep, M_NFSREQ); 983 return (error); 984 } 985 986 /* 987 * Grab any Kerberos verifier, otherwise just throw it away. 988 */ 989 verf_type = fxdr_unsigned(int, *tl++); 990 i = fxdr_unsigned(int32_t, *tl); 991 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 992 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 993 if (error) 994 goto nfsmout; 995 } else if (i > 0) 996 nfsm_adv(nfsm_rndup(i)); 997 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 998 /* 0 == ok */ 999 if (*tl == 0) { 1000 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1001 if (*tl != 0) { 1002 error = fxdr_unsigned(int, *tl); 1003 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1004 error == NFSERR_TRYLATER) { 1005 m_freem(mrep); 1006 error = 0; 1007 waituntil = time.tv_sec + trylater_delay; 1008 while (time.tv_sec < waituntil) 1009 (void) tsleep((caddr_t)&lbolt, 1010 PSOCK, "nqnfstry", 0); 1011 trylater_delay *= nfs_backoff[trylater_cnt]; 1012 if (trylater_cnt < 7) 1013 trylater_cnt++; 1014 goto tryagain; 1015 } 1016 1017 /* 1018 * If the File Handle was stale, invalidate the 1019 * lookup cache, just in case. 1020 */ 1021 if (error == ESTALE) 1022 cache_purge(vp); 1023 if (nmp->nm_flag & NFSMNT_NFSV3) { 1024 *mrp = mrep; 1025 *mdp = md; 1026 *dposp = dpos; 1027 error |= NFSERR_RETERR; 1028 } else 1029 m_freem(mrep); 1030 m_freem(rep->r_mreq); 1031 free((caddr_t)rep, M_NFSREQ); 1032 return (error); 1033 } 1034 1035 /* 1036 * For nqnfs, get any lease in reply 1037 */ 1038 if (nmp->nm_flag & NFSMNT_NQNFS) { 1039 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1040 if (*tl) { 1041 np = VTONFS(vp); 1042 nqlflag = fxdr_unsigned(int, *tl); 1043 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1044 cachable = fxdr_unsigned(int, *tl++); 1045 reqtime += fxdr_unsigned(int, *tl++); 1046 if (reqtime > time.tv_sec) { 1047 fxdr_hyper(tl, &frev); 1048 nqnfs_clientlease(nmp, np, nqlflag, 1049 cachable, reqtime, frev); 1050 } 1051 } 1052 } 1053 *mrp = mrep; 1054 *mdp = md; 1055 *dposp = dpos; 1056 m_freem(rep->r_mreq); 1057 FREE((caddr_t)rep, M_NFSREQ); 1058 return (0); 1059 } 1060 m_freem(mrep); 1061 error = EPROTONOSUPPORT; 1062 nfsmout: 1063 m_freem(rep->r_mreq); 1064 free((caddr_t)rep, M_NFSREQ); 1065 return (error); 1066 } 1067 #endif /* NFSCLIENT */ 1068 1069 /* 1070 * Generate the rpc reply header 1071 * siz arg. is used to decide if adding a cluster is worthwhile 1072 */ 1073 int 1074 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1075 int siz; 1076 struct nfsrv_descript *nd; 1077 struct nfssvc_sock *slp; 1078 int err; 1079 int cache; 1080 u_quad_t *frev; 1081 struct mbuf **mrq; 1082 struct mbuf **mbp; 1083 caddr_t *bposp; 1084 { 1085 register u_int32_t *tl; 1086 register struct mbuf *mreq; 1087 caddr_t bpos; 1088 struct mbuf *mb, *mb2; 1089 1090 MGETHDR(mreq, M_WAIT, MT_DATA); 1091 mb = mreq; 1092 /* 1093 * If this is a big reply, use a cluster else 1094 * try and leave leading space for the lower level headers. 1095 */ 1096 siz += RPC_REPLYSIZ; 1097 if (siz >= MINCLSIZE) { 1098 MCLGET(mreq, M_WAIT); 1099 } else 1100 mreq->m_data += max_hdr; 1101 tl = mtod(mreq, u_int32_t *); 1102 mreq->m_len = 6 * NFSX_UNSIGNED; 1103 bpos = ((caddr_t)tl) + mreq->m_len; 1104 *tl++ = txdr_unsigned(nd->nd_retxid); 1105 *tl++ = rpc_reply; 1106 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1107 *tl++ = rpc_msgdenied; 1108 if (err & NFSERR_AUTHERR) { 1109 *tl++ = rpc_autherr; 1110 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1111 mreq->m_len -= NFSX_UNSIGNED; 1112 bpos -= NFSX_UNSIGNED; 1113 } else { 1114 *tl++ = rpc_mismatch; 1115 *tl++ = txdr_unsigned(RPC_VER2); 1116 *tl = txdr_unsigned(RPC_VER2); 1117 } 1118 } else { 1119 *tl++ = rpc_msgaccepted; 1120 1121 /* 1122 * For Kerberos authentication, we must send the nickname 1123 * verifier back, otherwise just RPCAUTH_NULL. 1124 */ 1125 if (nd->nd_flag & ND_KERBFULL) { 1126 register struct nfsuid *nuidp; 1127 struct timeval ktvin, ktvout; 1128 1129 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; 1130 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1131 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1132 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), 1133 &nuidp->nu_haddr, nd->nd_nam2))) 1134 break; 1135 } 1136 if (nuidp) { 1137 ktvin.tv_sec = 1138 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); 1139 ktvin.tv_usec = 1140 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1141 1142 /* 1143 * Encrypt the timestamp in ecb mode using the 1144 * session key. 1145 */ 1146 #ifdef NFSKERB 1147 XXX 1148 #endif 1149 1150 *tl++ = rpc_auth_kerb; 1151 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1152 *tl = ktvout.tv_sec; 1153 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1154 *tl++ = ktvout.tv_usec; 1155 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1156 } else { 1157 *tl++ = 0; 1158 *tl++ = 0; 1159 } 1160 } else { 1161 *tl++ = 0; 1162 *tl++ = 0; 1163 } 1164 switch (err) { 1165 case EPROGUNAVAIL: 1166 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1167 break; 1168 case EPROGMISMATCH: 1169 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1170 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1171 if (nd->nd_flag & ND_NQNFS) { 1172 *tl++ = txdr_unsigned(3); 1173 *tl = txdr_unsigned(3); 1174 } else { 1175 *tl++ = txdr_unsigned(2); 1176 *tl = txdr_unsigned(3); 1177 } 1178 break; 1179 case EPROCUNAVAIL: 1180 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1181 break; 1182 case EBADRPC: 1183 *tl = txdr_unsigned(RPC_GARBAGE); 1184 break; 1185 default: 1186 *tl = 0; 1187 if (err != NFSERR_RETVOID) { 1188 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1189 if (err) 1190 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1191 else 1192 *tl = 0; 1193 } 1194 break; 1195 }; 1196 } 1197 1198 /* 1199 * For nqnfs, piggyback lease as requested. 1200 */ 1201 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1202 if (nd->nd_flag & ND_LEASE) { 1203 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1204 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1205 *tl++ = txdr_unsigned(cache); 1206 *tl++ = txdr_unsigned(nd->nd_duration); 1207 txdr_hyper(frev, tl); 1208 } else { 1209 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1210 *tl = 0; 1211 } 1212 } 1213 *mrq = mreq; 1214 *mbp = mb; 1215 *bposp = bpos; 1216 if (err != 0 && err != NFSERR_RETVOID) 1217 nfsstats.srvrpc_errs++; 1218 return (0); 1219 } 1220 1221 /* 1222 * Nfs timer routine 1223 * Scan the nfsreq list and retranmit any requests that have timed out 1224 * To avoid retransmission attempts on STREAM sockets (in the future) make 1225 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1226 */ 1227 void 1228 nfs_timer(arg) 1229 void *arg; /* never used */ 1230 { 1231 register struct nfsreq *rep; 1232 register struct mbuf *m; 1233 register struct socket *so; 1234 register struct nfsmount *nmp; 1235 register int timeo; 1236 register struct nfssvc_sock *slp; 1237 #ifdef NFSSERVER 1238 static long lasttime = 0; 1239 #endif 1240 int s, error; 1241 u_quad_t cur_usec; 1242 1243 s = splsoftnet(); 1244 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1245 nmp = rep->r_nmp; 1246 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1247 continue; 1248 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1249 rep->r_flags |= R_SOFTTERM; 1250 continue; 1251 } 1252 if (rep->r_rtt >= 0) { 1253 rep->r_rtt++; 1254 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1255 timeo = nmp->nm_timeo; 1256 else 1257 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1258 if (nmp->nm_timeouts > 0) 1259 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1260 if (rep->r_rtt <= timeo) 1261 continue; 1262 if (nmp->nm_timeouts < 8) 1263 nmp->nm_timeouts++; 1264 } 1265 /* 1266 * Check for server not responding 1267 */ 1268 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1269 rep->r_rexmit > nmp->nm_deadthresh) { 1270 nfs_msg(rep->r_procp, 1271 nmp->nm_mountp->mnt_stat.f_mntfromname, 1272 "not responding"); 1273 rep->r_flags |= R_TPRINTFMSG; 1274 } 1275 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1276 nfsstats.rpctimeouts++; 1277 rep->r_flags |= R_SOFTTERM; 1278 continue; 1279 } 1280 if (nmp->nm_sotype != SOCK_DGRAM) { 1281 if (++rep->r_rexmit > NFS_MAXREXMIT) 1282 rep->r_rexmit = NFS_MAXREXMIT; 1283 continue; 1284 } 1285 if ((so = nmp->nm_so) == NULL) 1286 continue; 1287 1288 /* 1289 * If there is enough space and the window allows.. 1290 * Resend it 1291 * Set r_rtt to -1 in case we fail to send it now. 1292 */ 1293 rep->r_rtt = -1; 1294 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1295 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1296 (rep->r_flags & R_SENT) || 1297 nmp->nm_sent < nmp->nm_cwnd) && 1298 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1299 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1300 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1301 (struct mbuf *)0, (struct mbuf *)0); 1302 else 1303 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1304 nmp->nm_nam, (struct mbuf *)0); 1305 if (error) { 1306 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) 1307 so->so_error = 0; 1308 } else { 1309 /* 1310 * Iff first send, start timing 1311 * else turn timing off, backoff timer 1312 * and divide congestion window by 2. 1313 */ 1314 if (rep->r_flags & R_SENT) { 1315 rep->r_flags &= ~R_TIMING; 1316 if (++rep->r_rexmit > NFS_MAXREXMIT) 1317 rep->r_rexmit = NFS_MAXREXMIT; 1318 nmp->nm_cwnd >>= 1; 1319 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1320 nmp->nm_cwnd = NFS_CWNDSCALE; 1321 nfsstats.rpcretries++; 1322 } else { 1323 rep->r_flags |= R_SENT; 1324 nmp->nm_sent += NFS_CWNDSCALE; 1325 } 1326 rep->r_rtt = 0; 1327 } 1328 } 1329 } 1330 1331 #ifdef NFSSERVER 1332 /* 1333 * Call the nqnfs server timer once a second to handle leases. 1334 */ 1335 if (lasttime != time.tv_sec) { 1336 lasttime = time.tv_sec; 1337 nqnfs_serverd(); 1338 } 1339 1340 /* 1341 * Scan the write gathering queues for writes that need to be 1342 * completed now. 1343 */ 1344 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1345 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 1346 slp = slp->ns_chain.tqe_next) { 1347 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec) 1348 nfsrv_wakenfsd(slp); 1349 } 1350 #endif /* NFSSERVER */ 1351 splx(s); 1352 timeout(nfs_timer, (void *)0, nfs_ticks); 1353 } 1354 1355 /* 1356 * Test for a termination condition pending on the process. 1357 * This is used for NFSMNT_INT mounts. 1358 */ 1359 int 1360 nfs_sigintr(nmp, rep, p) 1361 struct nfsmount *nmp; 1362 struct nfsreq *rep; 1363 register struct proc *p; 1364 { 1365 1366 if (rep && (rep->r_flags & R_SOFTTERM)) 1367 return (EINTR); 1368 if (!(nmp->nm_flag & NFSMNT_INT)) 1369 return (0); 1370 if (p && p->p_siglist && 1371 (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) & 1372 NFSINT_SIGMASK)) 1373 return (EINTR); 1374 return (0); 1375 } 1376 1377 /* 1378 * Lock a socket against others. 1379 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1380 * and also to avoid race conditions between the processes with nfs requests 1381 * in progress when a reconnect is necessary. 1382 */ 1383 int 1384 nfs_sndlock(flagp, rep) 1385 register int *flagp; 1386 struct nfsreq *rep; 1387 { 1388 struct proc *p; 1389 int slpflag = 0, slptimeo = 0; 1390 1391 if (rep) { 1392 p = rep->r_procp; 1393 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1394 slpflag = PCATCH; 1395 } else 1396 p = (struct proc *)0; 1397 while (*flagp & NFSMNT_SNDLOCK) { 1398 if (nfs_sigintr(rep->r_nmp, rep, p)) 1399 return (EINTR); 1400 *flagp |= NFSMNT_WANTSND; 1401 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1402 slptimeo); 1403 if (slpflag == PCATCH) { 1404 slpflag = 0; 1405 slptimeo = 2 * hz; 1406 } 1407 } 1408 *flagp |= NFSMNT_SNDLOCK; 1409 return (0); 1410 } 1411 1412 /* 1413 * Unlock the stream socket for others. 1414 */ 1415 void 1416 nfs_sndunlock(flagp) 1417 register int *flagp; 1418 { 1419 1420 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1421 panic("nfs sndunlock"); 1422 *flagp &= ~NFSMNT_SNDLOCK; 1423 if (*flagp & NFSMNT_WANTSND) { 1424 *flagp &= ~NFSMNT_WANTSND; 1425 wakeup((caddr_t)flagp); 1426 } 1427 } 1428 1429 int 1430 nfs_rcvlock(rep) 1431 register struct nfsreq *rep; 1432 { 1433 register int *flagp = &rep->r_nmp->nm_flag; 1434 int slpflag, slptimeo = 0; 1435 1436 if (*flagp & NFSMNT_INT) 1437 slpflag = PCATCH; 1438 else 1439 slpflag = 0; 1440 while (*flagp & NFSMNT_RCVLOCK) { 1441 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1442 return (EINTR); 1443 *flagp |= NFSMNT_WANTRCV; 1444 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1445 slptimeo); 1446 if (slpflag == PCATCH) { 1447 slpflag = 0; 1448 slptimeo = 2 * hz; 1449 } 1450 } 1451 *flagp |= NFSMNT_RCVLOCK; 1452 return (0); 1453 } 1454 1455 /* 1456 * Unlock the stream socket for others. 1457 */ 1458 void 1459 nfs_rcvunlock(flagp) 1460 register int *flagp; 1461 { 1462 1463 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1464 panic("nfs rcvunlock"); 1465 *flagp &= ~NFSMNT_RCVLOCK; 1466 if (*flagp & NFSMNT_WANTRCV) { 1467 *flagp &= ~NFSMNT_WANTRCV; 1468 wakeup((caddr_t)flagp); 1469 } 1470 } 1471 1472 /* 1473 * Check for badly aligned mbuf data areas and 1474 * realign data in an mbuf list by copying the data areas up, as required. 1475 */ 1476 void 1477 nfs_realign(m, hsiz) 1478 register struct mbuf *m; 1479 int hsiz; 1480 { 1481 register struct mbuf *m2; 1482 register int siz, mlen, olen; 1483 register caddr_t tcp, fcp; 1484 struct mbuf *mnew; 1485 1486 while (m) { 1487 /* 1488 * This never happens for UDP, rarely happens for TCP 1489 * but frequently happens for iso transport. 1490 */ 1491 if ((m->m_len & 0x3) || (mtod(m, long) & 0x3)) { 1492 olen = m->m_len; 1493 fcp = mtod(m, caddr_t); 1494 if ((long)fcp & 0x3) { 1495 m->m_flags &= ~M_PKTHDR; 1496 if (m->m_flags & M_EXT) 1497 m->m_data = m->m_ext.ext_buf + 1498 ((m->m_ext.ext_size - olen) & ~0x3); 1499 else 1500 m->m_data = m->m_dat; 1501 } 1502 m->m_len = 0; 1503 tcp = mtod(m, caddr_t); 1504 mnew = m; 1505 m2 = m->m_next; 1506 1507 /* 1508 * If possible, only put the first invariant part 1509 * of the RPC header in the first mbuf. 1510 */ 1511 mlen = M_TRAILINGSPACE(m); 1512 if (olen <= hsiz && mlen > hsiz) 1513 mlen = hsiz; 1514 1515 /* 1516 * Loop through the mbuf list consolidating data. 1517 */ 1518 while (m) { 1519 while (olen > 0) { 1520 if (mlen == 0) { 1521 m2->m_flags &= ~M_PKTHDR; 1522 if (m2->m_flags & M_EXT) 1523 m2->m_data = m2->m_ext.ext_buf; 1524 else 1525 m2->m_data = m2->m_dat; 1526 m2->m_len = 0; 1527 mlen = M_TRAILINGSPACE(m2); 1528 tcp = mtod(m2, caddr_t); 1529 mnew = m2; 1530 m2 = m2->m_next; 1531 } 1532 siz = min(mlen, olen); 1533 if (tcp != fcp) 1534 bcopy(fcp, tcp, siz); 1535 mnew->m_len += siz; 1536 mlen -= siz; 1537 olen -= siz; 1538 tcp += siz; 1539 fcp += siz; 1540 } 1541 m = m->m_next; 1542 if (m) { 1543 olen = m->m_len; 1544 fcp = mtod(m, caddr_t); 1545 } 1546 } 1547 1548 /* 1549 * Finally, set m_len == 0 for any trailing mbufs that have 1550 * been copied out of. 1551 */ 1552 while (m2) { 1553 m2->m_len = 0; 1554 m2 = m2->m_next; 1555 } 1556 return; 1557 } 1558 m = m->m_next; 1559 } 1560 } 1561 1562 /* 1563 * Parse an RPC request 1564 * - verify it 1565 * - fill in the cred struct. 1566 */ 1567 int 1568 nfs_getreq(nd, nfsd, has_header) 1569 register struct nfsrv_descript *nd; 1570 struct nfsd *nfsd; 1571 int has_header; 1572 { 1573 register int len, i; 1574 register u_int32_t *tl; 1575 register int32_t t1; 1576 struct uio uio; 1577 struct iovec iov; 1578 caddr_t dpos, cp2, cp; 1579 u_int32_t nfsvers, auth_type; 1580 uid_t nickuid; 1581 int error = 0, nqnfs = 0, ticklen; 1582 struct mbuf *mrep, *md; 1583 register struct nfsuid *nuidp; 1584 struct timeval tvin, tvout; 1585 1586 mrep = nd->nd_mrep; 1587 md = nd->nd_md; 1588 dpos = nd->nd_dpos; 1589 if (has_header) { 1590 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1591 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1592 if (*tl++ != rpc_call) { 1593 m_freem(mrep); 1594 return (EBADRPC); 1595 } 1596 } else 1597 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1598 nd->nd_repstat = 0; 1599 nd->nd_flag = 0; 1600 if (*tl++ != rpc_vers) { 1601 nd->nd_repstat = ERPCMISMATCH; 1602 nd->nd_procnum = NFSPROC_NOOP; 1603 return (0); 1604 } 1605 if (*tl != nfs_prog) { 1606 if (*tl == nqnfs_prog) 1607 nqnfs++; 1608 else { 1609 nd->nd_repstat = EPROGUNAVAIL; 1610 nd->nd_procnum = NFSPROC_NOOP; 1611 return (0); 1612 } 1613 } 1614 tl++; 1615 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1616 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1617 (nfsvers != NQNFS_VER3 && nqnfs)) { 1618 nd->nd_repstat = EPROGMISMATCH; 1619 nd->nd_procnum = NFSPROC_NOOP; 1620 return (0); 1621 } 1622 if (nqnfs) 1623 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1624 else if (nfsvers == NFS_VER3) 1625 nd->nd_flag = ND_NFSV3; 1626 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1627 if (nd->nd_procnum == NFSPROC_NULL) 1628 return (0); 1629 if (nd->nd_procnum >= NFS_NPROCS || 1630 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1631 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1632 nd->nd_repstat = EPROCUNAVAIL; 1633 nd->nd_procnum = NFSPROC_NOOP; 1634 return (0); 1635 } 1636 if ((nd->nd_flag & ND_NFSV3) == 0) 1637 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1638 auth_type = *tl++; 1639 len = fxdr_unsigned(int, *tl++); 1640 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1641 m_freem(mrep); 1642 return (EBADRPC); 1643 } 1644 1645 nd->nd_flag &= ~ND_KERBAUTH; 1646 /* 1647 * Handle auth_unix or auth_kerb. 1648 */ 1649 if (auth_type == rpc_auth_unix) { 1650 len = fxdr_unsigned(int, *++tl); 1651 if (len < 0 || len > NFS_MAXNAMLEN) { 1652 m_freem(mrep); 1653 return (EBADRPC); 1654 } 1655 nfsm_adv(nfsm_rndup(len)); 1656 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1657 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred)); 1658 nd->nd_cr.cr_ref = 1; 1659 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1660 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1661 len = fxdr_unsigned(int, *tl); 1662 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1663 m_freem(mrep); 1664 return (EBADRPC); 1665 } 1666 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1667 for (i = 0; i < len; i++) 1668 if (i < NGROUPS) 1669 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1670 else 1671 tl++; 1672 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1673 if (nd->nd_cr.cr_ngroups > 1) 1674 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1675 len = fxdr_unsigned(int, *++tl); 1676 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1677 m_freem(mrep); 1678 return (EBADRPC); 1679 } 1680 if (len > 0) 1681 nfsm_adv(nfsm_rndup(len)); 1682 } else if (auth_type == rpc_auth_kerb) { 1683 switch (fxdr_unsigned(int, *tl++)) { 1684 case RPCAKN_FULLNAME: 1685 ticklen = fxdr_unsigned(int, *tl); 1686 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1687 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1688 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1689 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1690 m_freem(mrep); 1691 return (EBADRPC); 1692 } 1693 uio.uio_offset = 0; 1694 uio.uio_iov = &iov; 1695 uio.uio_iovcnt = 1; 1696 uio.uio_segflg = UIO_SYSSPACE; 1697 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1698 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1699 nfsm_mtouio(&uio, uio.uio_resid); 1700 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1701 if (*tl++ != rpc_auth_kerb || 1702 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1703 printf("Bad kerb verifier\n"); 1704 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1705 nd->nd_procnum = NFSPROC_NOOP; 1706 return (0); 1707 } 1708 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1709 tl = (u_int32_t *)cp; 1710 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1711 printf("Not fullname kerb verifier\n"); 1712 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1713 nd->nd_procnum = NFSPROC_NOOP; 1714 return (0); 1715 } 1716 cp += NFSX_UNSIGNED; 1717 bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED); 1718 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1719 nd->nd_flag |= ND_KERBFULL; 1720 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1721 break; 1722 case RPCAKN_NICKNAME: 1723 if (len != 2 * NFSX_UNSIGNED) { 1724 printf("Kerb nickname short\n"); 1725 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1726 nd->nd_procnum = NFSPROC_NOOP; 1727 return (0); 1728 } 1729 nickuid = fxdr_unsigned(uid_t, *tl); 1730 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1731 if (*tl++ != rpc_auth_kerb || 1732 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1733 printf("Kerb nick verifier bad\n"); 1734 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1735 nd->nd_procnum = NFSPROC_NOOP; 1736 return (0); 1737 } 1738 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1739 tvin.tv_sec = *tl++; 1740 tvin.tv_usec = *tl; 1741 1742 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; 1743 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1744 if (nuidp->nu_cr.cr_uid == nickuid && 1745 (!nd->nd_nam2 || 1746 netaddr_match(NU_NETFAM(nuidp), 1747 &nuidp->nu_haddr, nd->nd_nam2))) 1748 break; 1749 } 1750 if (!nuidp) { 1751 nd->nd_repstat = 1752 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1753 nd->nd_procnum = NFSPROC_NOOP; 1754 return (0); 1755 } 1756 1757 /* 1758 * Now, decrypt the timestamp using the session key 1759 * and validate it. 1760 */ 1761 #ifdef NFSKERB 1762 XXX 1763 #endif 1764 1765 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1766 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1767 if (nuidp->nu_expire < time.tv_sec || 1768 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1769 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1770 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1771 nuidp->nu_expire = 0; 1772 nd->nd_repstat = 1773 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1774 nd->nd_procnum = NFSPROC_NOOP; 1775 return (0); 1776 } 1777 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1778 nd->nd_flag |= ND_KERBNICK; 1779 }; 1780 } else { 1781 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1782 nd->nd_procnum = NFSPROC_NOOP; 1783 return (0); 1784 } 1785 1786 /* 1787 * For nqnfs, get piggybacked lease request. 1788 */ 1789 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1790 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1791 nd->nd_flag |= fxdr_unsigned(int, *tl); 1792 if (nd->nd_flag & ND_LEASE) { 1793 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1794 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 1795 } else 1796 nd->nd_duration = NQ_MINLEASE; 1797 } else 1798 nd->nd_duration = NQ_MINLEASE; 1799 nd->nd_md = md; 1800 nd->nd_dpos = dpos; 1801 return (0); 1802 nfsmout: 1803 return (error); 1804 } 1805 1806 int 1807 nfs_msg(p, server, msg) 1808 struct proc *p; 1809 char *server, *msg; 1810 { 1811 tpr_t tpr; 1812 1813 if (p) 1814 tpr = tprintf_open(p); 1815 else 1816 tpr = NULL; 1817 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1818 tprintf_close(tpr); 1819 return (0); 1820 } 1821 1822 #ifdef NFSSERVER 1823 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 1824 struct nfssvc_sock *, struct proc *, 1825 struct mbuf **)) = { 1826 nfsrv_null, 1827 nfsrv_getattr, 1828 nfsrv_setattr, 1829 nfsrv_lookup, 1830 nfsrv3_access, 1831 nfsrv_readlink, 1832 nfsrv_read, 1833 nfsrv_write, 1834 nfsrv_create, 1835 nfsrv_mkdir, 1836 nfsrv_symlink, 1837 nfsrv_mknod, 1838 nfsrv_remove, 1839 nfsrv_rmdir, 1840 nfsrv_rename, 1841 nfsrv_link, 1842 nfsrv_readdir, 1843 nfsrv_readdirplus, 1844 nfsrv_statfs, 1845 nfsrv_fsinfo, 1846 nfsrv_pathconf, 1847 nfsrv_commit, 1848 nqnfsrv_getlease, 1849 nqnfsrv_vacated, 1850 nfsrv_noop, 1851 nfsrv_noop 1852 }; 1853 1854 /* 1855 * Socket upcall routine for the nfsd sockets. 1856 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1857 * Essentially do as much as possible non-blocking, else punt and it will 1858 * be called with M_WAIT from an nfsd. 1859 */ 1860 void 1861 nfsrv_rcv(so, arg, waitflag) 1862 struct socket *so; 1863 caddr_t arg; 1864 int waitflag; 1865 { 1866 register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1867 register struct mbuf *m; 1868 struct mbuf *mp, *nam; 1869 struct uio auio; 1870 int flags, error; 1871 1872 if ((slp->ns_flag & SLP_VALID) == 0) 1873 return; 1874 #ifdef notdef 1875 /* 1876 * Define this to test for nfsds handling this under heavy load. 1877 */ 1878 if (waitflag == M_DONTWAIT) { 1879 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1880 } 1881 #endif 1882 auio.uio_procp = NULL; 1883 if (so->so_type == SOCK_STREAM) { 1884 /* 1885 * If there are already records on the queue, defer soreceive() 1886 * to an nfsd so that there is feedback to the TCP layer that 1887 * the nfs servers are heavily loaded. 1888 */ 1889 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1890 slp->ns_flag |= SLP_NEEDQ; 1891 goto dorecs; 1892 } 1893 1894 /* 1895 * Do soreceive(). 1896 */ 1897 auio.uio_resid = 1000000000; 1898 flags = MSG_DONTWAIT; 1899 error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1900 if (error || mp == (struct mbuf *)0) { 1901 if (error == EWOULDBLOCK) 1902 slp->ns_flag |= SLP_NEEDQ; 1903 else 1904 slp->ns_flag |= SLP_DISCONN; 1905 goto dorecs; 1906 } 1907 m = mp; 1908 if (slp->ns_rawend) { 1909 slp->ns_rawend->m_next = m; 1910 slp->ns_cc += 1000000000 - auio.uio_resid; 1911 } else { 1912 slp->ns_raw = m; 1913 slp->ns_cc = 1000000000 - auio.uio_resid; 1914 } 1915 while (m->m_next) 1916 m = m->m_next; 1917 slp->ns_rawend = m; 1918 1919 /* 1920 * Now try and parse record(s) out of the raw stream data. 1921 */ 1922 error = nfsrv_getstream(slp, waitflag); 1923 if (error) { 1924 if (error == EPERM) 1925 slp->ns_flag |= SLP_DISCONN; 1926 else 1927 slp->ns_flag |= SLP_NEEDQ; 1928 } 1929 } else { 1930 do { 1931 auio.uio_resid = 1000000000; 1932 flags = MSG_DONTWAIT; 1933 error = soreceive(so, &nam, &auio, &mp, 1934 (struct mbuf **)0, &flags); 1935 if (mp) { 1936 nfs_realign(mp, 10 * NFSX_UNSIGNED); 1937 if (nam) { 1938 m = nam; 1939 m->m_next = mp; 1940 } else 1941 m = mp; 1942 if (slp->ns_recend) 1943 slp->ns_recend->m_nextpkt = m; 1944 else 1945 slp->ns_rec = m; 1946 slp->ns_recend = m; 1947 m->m_nextpkt = (struct mbuf *)0; 1948 } 1949 if (error) { 1950 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1951 && error != EWOULDBLOCK) { 1952 slp->ns_flag |= SLP_DISCONN; 1953 goto dorecs; 1954 } 1955 } 1956 } while (mp); 1957 } 1958 1959 /* 1960 * Now try and process the request records, non-blocking. 1961 */ 1962 dorecs: 1963 if (waitflag == M_DONTWAIT && 1964 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1965 nfsrv_wakenfsd(slp); 1966 } 1967 1968 /* 1969 * Try and extract an RPC request from the mbuf data list received on a 1970 * stream socket. The "waitflag" argument indicates whether or not it 1971 * can sleep. 1972 */ 1973 int 1974 nfsrv_getstream(slp, waitflag) 1975 register struct nfssvc_sock *slp; 1976 int waitflag; 1977 { 1978 register struct mbuf *m, **mpp; 1979 register char *cp1, *cp2; 1980 register int len; 1981 struct mbuf *om, *m2, *recm = NULL; 1982 u_int32_t recmark; 1983 1984 if (slp->ns_flag & SLP_GETSTREAM) 1985 panic("nfs getstream"); 1986 slp->ns_flag |= SLP_GETSTREAM; 1987 for (;;) { 1988 if (slp->ns_reclen == 0) { 1989 if (slp->ns_cc < NFSX_UNSIGNED) { 1990 slp->ns_flag &= ~SLP_GETSTREAM; 1991 return (0); 1992 } 1993 m = slp->ns_raw; 1994 if (m->m_len >= NFSX_UNSIGNED) { 1995 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 1996 m->m_data += NFSX_UNSIGNED; 1997 m->m_len -= NFSX_UNSIGNED; 1998 } else { 1999 cp1 = (caddr_t)&recmark; 2000 cp2 = mtod(m, caddr_t); 2001 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 2002 while (m->m_len == 0) { 2003 m = m->m_next; 2004 cp2 = mtod(m, caddr_t); 2005 } 2006 *cp1++ = *cp2++; 2007 m->m_data++; 2008 m->m_len--; 2009 } 2010 } 2011 slp->ns_cc -= NFSX_UNSIGNED; 2012 recmark = ntohl(recmark); 2013 slp->ns_reclen = recmark & ~0x80000000; 2014 if (recmark & 0x80000000) 2015 slp->ns_flag |= SLP_LASTFRAG; 2016 else 2017 slp->ns_flag &= ~SLP_LASTFRAG; 2018 if (slp->ns_reclen < NFS_MINPACKET || slp->ns_reclen > NFS_MAXPACKET) { 2019 slp->ns_flag &= ~SLP_GETSTREAM; 2020 return (EPERM); 2021 } 2022 } 2023 2024 /* 2025 * Now get the record part. 2026 */ 2027 if (slp->ns_cc == slp->ns_reclen) { 2028 recm = slp->ns_raw; 2029 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2030 slp->ns_cc = slp->ns_reclen = 0; 2031 } else if (slp->ns_cc > slp->ns_reclen) { 2032 len = 0; 2033 m = slp->ns_raw; 2034 om = (struct mbuf *)0; 2035 while (len < slp->ns_reclen) { 2036 if ((len + m->m_len) > slp->ns_reclen) { 2037 m2 = m_copym(m, 0, slp->ns_reclen - len, 2038 waitflag); 2039 if (m2) { 2040 if (om) { 2041 om->m_next = m2; 2042 recm = slp->ns_raw; 2043 } else 2044 recm = m2; 2045 m->m_data += slp->ns_reclen - len; 2046 m->m_len -= slp->ns_reclen - len; 2047 len = slp->ns_reclen; 2048 } else { 2049 slp->ns_flag &= ~SLP_GETSTREAM; 2050 return (EWOULDBLOCK); 2051 } 2052 } else if ((len + m->m_len) == slp->ns_reclen) { 2053 om = m; 2054 len += m->m_len; 2055 m = m->m_next; 2056 recm = slp->ns_raw; 2057 om->m_next = (struct mbuf *)0; 2058 } else { 2059 om = m; 2060 len += m->m_len; 2061 m = m->m_next; 2062 } 2063 } 2064 slp->ns_raw = m; 2065 slp->ns_cc -= len; 2066 slp->ns_reclen = 0; 2067 } else { 2068 slp->ns_flag &= ~SLP_GETSTREAM; 2069 return (0); 2070 } 2071 2072 /* 2073 * Accumulate the fragments into a record. 2074 */ 2075 mpp = &slp->ns_frag; 2076 while (*mpp) 2077 mpp = &((*mpp)->m_next); 2078 *mpp = recm; 2079 if (slp->ns_flag & SLP_LASTFRAG) { 2080 nfs_realign(slp->ns_frag, 10 * NFSX_UNSIGNED); 2081 if (slp->ns_recend) 2082 slp->ns_recend->m_nextpkt = slp->ns_frag; 2083 else 2084 slp->ns_rec = slp->ns_frag; 2085 slp->ns_recend = slp->ns_frag; 2086 slp->ns_frag = (struct mbuf *)0; 2087 } 2088 } 2089 } 2090 2091 /* 2092 * Parse an RPC header. 2093 */ 2094 int 2095 nfsrv_dorec(slp, nfsd, ndp) 2096 register struct nfssvc_sock *slp; 2097 struct nfsd *nfsd; 2098 struct nfsrv_descript **ndp; 2099 { 2100 register struct mbuf *m, *nam; 2101 register struct nfsrv_descript *nd; 2102 int error; 2103 2104 *ndp = NULL; 2105 if ((slp->ns_flag & SLP_VALID) == 0 || 2106 (m = slp->ns_rec) == (struct mbuf *)0) 2107 return (ENOBUFS); 2108 slp->ns_rec = m->m_nextpkt; 2109 if (slp->ns_rec) 2110 m->m_nextpkt = (struct mbuf *)0; 2111 else 2112 slp->ns_recend = (struct mbuf *)0; 2113 if (m->m_type == MT_SONAME) { 2114 nam = m; 2115 m = m->m_next; 2116 nam->m_next = NULL; 2117 } else 2118 nam = NULL; 2119 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), 2120 M_NFSRVDESC, M_WAITOK); 2121 nd->nd_md = nd->nd_mrep = m; 2122 nd->nd_nam2 = nam; 2123 nd->nd_dpos = mtod(m, caddr_t); 2124 error = nfs_getreq(nd, nfsd, TRUE); 2125 if (error) { 2126 m_freem(nam); 2127 free((caddr_t)nd, M_NFSRVDESC); 2128 return (error); 2129 } 2130 *ndp = nd; 2131 nfsd->nfsd_nd = nd; 2132 return (0); 2133 } 2134 2135 2136 /* 2137 * Search for a sleeping nfsd and wake it up. 2138 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2139 * running nfsds will go look for the work in the nfssvc_sock list. 2140 */ 2141 void 2142 nfsrv_wakenfsd(slp) 2143 struct nfssvc_sock *slp; 2144 { 2145 register struct nfsd *nd; 2146 2147 if ((slp->ns_flag & SLP_VALID) == 0) 2148 return; 2149 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) { 2150 if (nd->nfsd_flag & NFSD_WAITING) { 2151 nd->nfsd_flag &= ~NFSD_WAITING; 2152 if (nd->nfsd_slp) 2153 panic("nfsd wakeup"); 2154 slp->ns_sref++; 2155 nd->nfsd_slp = slp; 2156 wakeup((caddr_t)nd); 2157 return; 2158 } 2159 } 2160 slp->ns_flag |= SLP_DOREC; 2161 nfsd_head_flag |= NFSD_CHECKSLP; 2162 } 2163 #endif /* NFSSERVER */ 2164
Indexes created Sun Sep 21 20:09:37 GMT 2025