1 /* $NetBSD: nfs_socket.c,v 1.48 1998/11/12 22:31:02 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 "fs_nfs.h" 46 #include "opt_nfsserver.h" 47 48 #include <sys/param.h> 49 #include <sys/systm.h> 50 #include <sys/proc.h> 51 #include <sys/mount.h> 52 #include <sys/kernel.h> 53 #include <sys/mbuf.h> 54 #include <sys/vnode.h> 55 #include <sys/domain.h> 56 #include <sys/protosw.h> 57 #include <sys/socket.h> 58 #include <sys/socketvar.h> 59 #include <sys/syslog.h> 60 #include <sys/tprintf.h> 61 #include <sys/namei.h> 62 #include <sys/signal.h> 63 #include <sys/signalvar.h> 64 65 #include <netinet/in.h> 66 #include <netinet/tcp.h> 67 68 #include <nfs/rpcv2.h> 69 #include <nfs/nfsproto.h> 70 #include <nfs/nfs.h> 71 #include <nfs/xdr_subs.h> 72 #include <nfs/nfsm_subs.h> 73 #include <nfs/nfsmount.h> 74 #include <nfs/nfsnode.h> 75 #include <nfs/nfsrtt.h> 76 #include <nfs/nqnfs.h> 77 #include <nfs/nfs_var.h> 78 79 #define TRUE 1 80 #define FALSE 0 81 82 /* 83 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 84 * Use the mean and mean deviation of rtt for the appropriate type of rpc 85 * for the frequent rpcs and a default for the others. 86 * The justification for doing "other" this way is that these rpcs 87 * happen so infrequently that timer est. would probably be stale. 88 * Also, since many of these rpcs are 89 * non-idempotent, a conservative timeout is desired. 90 * getattr, lookup - A+2D 91 * read, write - A+4D 92 * other - nm_timeo 93 */ 94 #define NFS_RTO(n, t) \ 95 ((t) == 0 ? (n)->nm_timeo : \ 96 ((t) < 3 ? \ 97 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 98 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 99 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 100 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 101 /* 102 * External data, mostly RPC constants in XDR form 103 */ 104 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 105 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, 106 rpc_auth_kerb; 107 extern u_int32_t nfs_prog, nqnfs_prog; 108 extern time_t nqnfsstarttime; 109 extern struct nfsstats nfsstats; 110 extern int nfsv3_procid[NFS_NPROCS]; 111 extern int nfs_ticks; 112 113 /* 114 * Defines which timer to use for the procnum. 115 * 0 - default 116 * 1 - getattr 117 * 2 - lookup 118 * 3 - read 119 * 4 - write 120 */ 121 static int proct[NFS_NPROCS] = { 122 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0, 123 0, 0, 0, 124 }; 125 126 /* 127 * There is a congestion window for outstanding rpcs maintained per mount 128 * point. The cwnd size is adjusted in roughly the way that: 129 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 130 * SIGCOMM '88". ACM, August 1988. 131 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 132 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 133 * of rpcs is in progress. 134 * (The sent count and cwnd are scaled for integer arith.) 135 * Variants of "slow start" were tried and were found to be too much of a 136 * performance hit (ave. rtt 3 times larger), 137 * I suspect due to the large rtt that nfs rpcs have. 138 */ 139 #define NFS_CWNDSCALE 256 140 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 141 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 142 int nfsrtton = 0; 143 struct nfsrtt nfsrtt; 144 145 /* 146 * Initialize sockets and congestion for a new NFS connection. 147 * We do not free the sockaddr if error. 148 */ 149 int 150 nfs_connect(nmp, rep) 151 register struct nfsmount *nmp; 152 struct nfsreq *rep; 153 { 154 register struct socket *so; 155 int s, error, rcvreserve, sndreserve; 156 struct sockaddr *saddr; 157 struct sockaddr_in *sin; 158 struct mbuf *m; 159 u_int16_t tport; 160 161 nmp->nm_so = (struct socket *)0; 162 saddr = mtod(nmp->nm_nam, struct sockaddr *); 163 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 164 nmp->nm_soproto); 165 if (error) 166 goto bad; 167 so = nmp->nm_so; 168 nmp->nm_soflags = so->so_proto->pr_flags; 169 170 /* 171 * Some servers require that the client port be a reserved port number. 172 */ 173 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 174 MGET(m, M_WAIT, MT_SONAME); 175 sin = mtod(m, struct sockaddr_in *); 176 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 177 sin->sin_family = AF_INET; 178 sin->sin_addr.s_addr = INADDR_ANY; 179 tport = IPPORT_RESERVED - 1; 180 sin->sin_port = htons(tport); 181 while ((error = sobind(so, m)) == EADDRINUSE && 182 --tport > IPPORT_RESERVED / 2) 183 sin->sin_port = htons(tport); 184 m_freem(m); 185 if (error) 186 goto bad; 187 } 188 189 /* 190 * Protocols that do not require connections may be optionally left 191 * unconnected for servers that reply from a port other than NFS_PORT. 192 */ 193 if (nmp->nm_flag & NFSMNT_NOCONN) { 194 if (nmp->nm_soflags & PR_CONNREQUIRED) { 195 error = ENOTCONN; 196 goto bad; 197 } 198 } else { 199 error = soconnect(so, nmp->nm_nam); 200 if (error) 201 goto bad; 202 203 /* 204 * Wait for the connection to complete. Cribbed from the 205 * connect system call but with the wait timing out so 206 * that interruptible mounts don't hang here for a long time. 207 */ 208 s = splsoftnet(); 209 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 210 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 211 "nfscn1", 2 * hz); 212 if ((so->so_state & SS_ISCONNECTING) && 213 so->so_error == 0 && rep && 214 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ 215 so->so_state &= ~SS_ISCONNECTING; 216 splx(s); 217 goto bad; 218 } 219 } 220 if (so->so_error) { 221 error = so->so_error; 222 so->so_error = 0; 223 splx(s); 224 goto bad; 225 } 226 splx(s); 227 } 228 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 229 so->so_rcv.sb_timeo = (5 * hz); 230 so->so_snd.sb_timeo = (5 * hz); 231 } else { 232 so->so_rcv.sb_timeo = 0; 233 so->so_snd.sb_timeo = 0; 234 } 235 if (nmp->nm_sotype == SOCK_DGRAM) { 236 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 237 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 238 NFS_MAXPKTHDR) * 2; 239 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 240 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 241 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 242 NFS_MAXPKTHDR) * 2; 243 } else { 244 if (nmp->nm_sotype != SOCK_STREAM) 245 panic("nfscon sotype"); 246 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 247 MGET(m, M_WAIT, MT_SOOPTS); 248 *mtod(m, int32_t *) = 1; 249 m->m_len = sizeof(int32_t); 250 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 251 } 252 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 253 MGET(m, M_WAIT, MT_SOOPTS); 254 *mtod(m, int32_t *) = 1; 255 m->m_len = sizeof(int32_t); 256 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 257 } 258 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 259 sizeof (u_int32_t)) * 2; 260 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 261 sizeof (u_int32_t)) * 2; 262 } 263 error = soreserve(so, sndreserve, rcvreserve); 264 if (error) 265 goto bad; 266 so->so_rcv.sb_flags |= SB_NOINTR; 267 so->so_snd.sb_flags |= SB_NOINTR; 268 269 /* Initialize other non-zero congestion variables */ 270 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 271 nmp->nm_srtt[4] = (NFS_TIMEO << 3); 272 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 273 nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0; 274 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 275 nmp->nm_sent = 0; 276 nmp->nm_timeouts = 0; 277 return (0); 278 279 bad: 280 nfs_disconnect(nmp); 281 return (error); 282 } 283 284 /* 285 * Reconnect routine: 286 * Called when a connection is broken on a reliable protocol. 287 * - clean up the old socket 288 * - nfs_connect() again 289 * - set R_MUSTRESEND for all outstanding requests on mount point 290 * If this fails the mount point is DEAD! 291 * nb: Must be called with the nfs_sndlock() set on the mount point. 292 */ 293 int 294 nfs_reconnect(rep) 295 register struct nfsreq *rep; 296 { 297 register struct nfsreq *rp; 298 register struct nfsmount *nmp = rep->r_nmp; 299 int error; 300 301 nfs_disconnect(nmp); 302 while ((error = nfs_connect(nmp, rep)) != 0) { 303 if (error == EINTR || error == ERESTART) 304 return (EINTR); 305 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0); 306 } 307 308 /* 309 * Loop through outstanding request list and fix up all requests 310 * on old socket. 311 */ 312 for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) { 313 if (rp->r_nmp == nmp) 314 rp->r_flags |= R_MUSTRESEND; 315 } 316 return (0); 317 } 318 319 /* 320 * NFS disconnect. Clean up and unlink. 321 */ 322 void 323 nfs_disconnect(nmp) 324 register struct nfsmount *nmp; 325 { 326 register struct socket *so; 327 328 if (nmp->nm_so) { 329 so = nmp->nm_so; 330 nmp->nm_so = (struct socket *)0; 331 soshutdown(so, 2); 332 soclose(so); 333 } 334 } 335 336 void 337 nfs_safedisconnect(nmp) 338 struct nfsmount *nmp; 339 { 340 struct nfsreq dummyreq; 341 342 memset(&dummyreq, 0, sizeof(dummyreq)); 343 dummyreq.r_nmp = nmp; 344 nfs_rcvlock(&dummyreq); 345 nfs_disconnect(nmp); 346 nfs_rcvunlock(&nmp->nm_iflag); 347 } 348 349 /* 350 * This is the nfs send routine. For connection based socket types, it 351 * must be called with an nfs_sndlock() on the socket. 352 * "rep == NULL" indicates that it has been called from a server. 353 * For the client side: 354 * - return EINTR if the RPC is terminated, 0 otherwise 355 * - set R_MUSTRESEND if the send fails for any reason 356 * - do any cleanup required by recoverable socket errors (???) 357 * For the server side: 358 * - return EINTR or ERESTART if interrupted by a signal 359 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 360 * - do any cleanup required by recoverable socket errors (???) 361 */ 362 int 363 nfs_send(so, nam, top, rep) 364 register struct socket *so; 365 struct mbuf *nam; 366 register struct mbuf *top; 367 struct nfsreq *rep; 368 { 369 struct mbuf *sendnam; 370 int error, soflags, flags; 371 372 if (rep) { 373 if (rep->r_flags & R_SOFTTERM) { 374 m_freem(top); 375 return (EINTR); 376 } 377 if ((so = rep->r_nmp->nm_so) == NULL) { 378 rep->r_flags |= R_MUSTRESEND; 379 m_freem(top); 380 return (0); 381 } 382 rep->r_flags &= ~R_MUSTRESEND; 383 soflags = rep->r_nmp->nm_soflags; 384 } else 385 soflags = so->so_proto->pr_flags; 386 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 387 sendnam = (struct mbuf *)0; 388 else 389 sendnam = nam; 390 if (so->so_type == SOCK_SEQPACKET) 391 flags = MSG_EOR; 392 else 393 flags = 0; 394 395 error = (*so->so_send)(so, sendnam, (struct uio *)0, top, 396 (struct mbuf *)0, flags); 397 if (error) { 398 if (rep) { 399 log(LOG_INFO, "nfs send error %d for server %s\n",error, 400 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 401 /* 402 * Deal with errors for the client side. 403 */ 404 if (rep->r_flags & R_SOFTTERM) 405 error = EINTR; 406 else 407 rep->r_flags |= R_MUSTRESEND; 408 } else 409 log(LOG_INFO, "nfsd send error %d\n", error); 410 411 /* 412 * Handle any recoverable (soft) socket errors here. (???) 413 */ 414 if (error != EINTR && error != ERESTART && 415 error != EWOULDBLOCK && error != EPIPE) 416 error = 0; 417 } 418 return (error); 419 } 420 421 #ifdef NFS 422 /* 423 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 424 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 425 * Mark and consolidate the data into a new mbuf list. 426 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 427 * small mbufs. 428 * For SOCK_STREAM we must be very careful to read an entire record once 429 * we have read any of it, even if the system call has been interrupted. 430 */ 431 int 432 nfs_receive(rep, aname, mp) 433 register struct nfsreq *rep; 434 struct mbuf **aname; 435 struct mbuf **mp; 436 { 437 register struct socket *so; 438 struct uio auio; 439 struct iovec aio; 440 register struct mbuf *m; 441 struct mbuf *control; 442 u_int32_t len; 443 struct mbuf **getnam; 444 int error, sotype, rcvflg; 445 struct proc *p = curproc; /* XXX */ 446 447 /* 448 * Set up arguments for soreceive() 449 */ 450 *mp = (struct mbuf *)0; 451 *aname = (struct mbuf *)0; 452 sotype = rep->r_nmp->nm_sotype; 453 454 /* 455 * For reliable protocols, lock against other senders/receivers 456 * in case a reconnect is necessary. 457 * For SOCK_STREAM, first get the Record Mark to find out how much 458 * more there is to get. 459 * We must lock the socket against other receivers 460 * until we have an entire rpc request/reply. 461 */ 462 if (sotype != SOCK_DGRAM) { 463 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 464 if (error) 465 return (error); 466 tryagain: 467 /* 468 * Check for fatal errors and resending request. 469 */ 470 /* 471 * Ugh: If a reconnect attempt just happened, nm_so 472 * would have changed. NULL indicates a failed 473 * attempt that has essentially shut down this 474 * mount point. 475 */ 476 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 477 nfs_sndunlock(&rep->r_nmp->nm_iflag); 478 return (EINTR); 479 } 480 so = rep->r_nmp->nm_so; 481 if (!so) { 482 error = nfs_reconnect(rep); 483 if (error) { 484 nfs_sndunlock(&rep->r_nmp->nm_iflag); 485 return (error); 486 } 487 goto tryagain; 488 } 489 while (rep->r_flags & R_MUSTRESEND) { 490 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 491 nfsstats.rpcretries++; 492 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 493 if (error) { 494 if (error == EINTR || error == ERESTART || 495 (error = nfs_reconnect(rep)) != 0) { 496 nfs_sndunlock(&rep->r_nmp->nm_iflag); 497 return (error); 498 } 499 goto tryagain; 500 } 501 } 502 nfs_sndunlock(&rep->r_nmp->nm_iflag); 503 if (sotype == SOCK_STREAM) { 504 aio.iov_base = (caddr_t) &len; 505 aio.iov_len = sizeof(u_int32_t); 506 auio.uio_iov = &aio; 507 auio.uio_iovcnt = 1; 508 auio.uio_segflg = UIO_SYSSPACE; 509 auio.uio_rw = UIO_READ; 510 auio.uio_offset = 0; 511 auio.uio_resid = sizeof(u_int32_t); 512 auio.uio_procp = p; 513 do { 514 rcvflg = MSG_WAITALL; 515 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 516 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 517 if (error == EWOULDBLOCK && rep) { 518 if (rep->r_flags & R_SOFTTERM) 519 return (EINTR); 520 } 521 } while (error == EWOULDBLOCK); 522 if (!error && auio.uio_resid > 0) { 523 /* 524 * Don't log a 0 byte receive; it means 525 * that the socket has been closed, and 526 * can happen during normal operation 527 * (forcible unmount or Solaris server). 528 */ 529 if (auio.uio_resid != sizeof (u_int32_t)) 530 log(LOG_INFO, 531 "short receive (%d/%d) from nfs server %s\n", 532 sizeof(u_int32_t) - auio.uio_resid, 533 sizeof(u_int32_t), 534 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 535 error = EPIPE; 536 } 537 if (error) 538 goto errout; 539 len = ntohl(len) & ~0x80000000; 540 /* 541 * This is SERIOUS! We are out of sync with the sender 542 * and forcing a disconnect/reconnect is all I can do. 543 */ 544 if (len > NFS_MAXPACKET) { 545 log(LOG_ERR, "%s (%d) from nfs server %s\n", 546 "impossible packet length", 547 len, 548 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 549 error = EFBIG; 550 goto errout; 551 } 552 auio.uio_resid = len; 553 do { 554 rcvflg = MSG_WAITALL; 555 error = (*so->so_receive)(so, (struct mbuf **)0, 556 &auio, mp, (struct mbuf **)0, &rcvflg); 557 } while (error == EWOULDBLOCK || error == EINTR || 558 error == ERESTART); 559 if (!error && auio.uio_resid > 0) { 560 if (len != auio.uio_resid) 561 log(LOG_INFO, 562 "short receive (%d/%d) from nfs server %s\n", 563 len - auio.uio_resid, len, 564 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 565 error = EPIPE; 566 } 567 } else { 568 /* 569 * NB: Since uio_resid is big, MSG_WAITALL is ignored 570 * and soreceive() will return when it has either a 571 * control msg or a data msg. 572 * We have no use for control msg., but must grab them 573 * and then throw them away so we know what is going 574 * on. 575 */ 576 auio.uio_resid = len = 100000000; /* Anything Big */ 577 auio.uio_procp = p; 578 do { 579 rcvflg = 0; 580 error = (*so->so_receive)(so, (struct mbuf **)0, 581 &auio, mp, &control, &rcvflg); 582 if (control) 583 m_freem(control); 584 if (error == EWOULDBLOCK && rep) { 585 if (rep->r_flags & R_SOFTTERM) 586 return (EINTR); 587 } 588 } while (error == EWOULDBLOCK || 589 (!error && *mp == NULL && control)); 590 if ((rcvflg & MSG_EOR) == 0) 591 printf("Egad!!\n"); 592 if (!error && *mp == NULL) 593 error = EPIPE; 594 len -= auio.uio_resid; 595 } 596 errout: 597 if (error && error != EINTR && error != ERESTART) { 598 m_freem(*mp); 599 *mp = (struct mbuf *)0; 600 if (error != EPIPE) 601 log(LOG_INFO, 602 "receive error %d from nfs server %s\n", 603 error, 604 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 605 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 606 if (!error) 607 error = nfs_reconnect(rep); 608 if (!error) 609 goto tryagain; 610 else 611 nfs_sndunlock(&rep->r_nmp->nm_iflag); 612 } 613 } else { 614 if ((so = rep->r_nmp->nm_so) == NULL) 615 return (EACCES); 616 if (so->so_state & SS_ISCONNECTED) 617 getnam = (struct mbuf **)0; 618 else 619 getnam = aname; 620 auio.uio_resid = len = 1000000; 621 auio.uio_procp = p; 622 do { 623 rcvflg = 0; 624 error = (*so->so_receive)(so, getnam, &auio, mp, 625 (struct mbuf **)0, &rcvflg); 626 if (error == EWOULDBLOCK && 627 (rep->r_flags & R_SOFTTERM)) 628 return (EINTR); 629 } while (error == EWOULDBLOCK); 630 len -= auio.uio_resid; 631 } 632 if (error) { 633 m_freem(*mp); 634 *mp = (struct mbuf *)0; 635 } 636 return (error); 637 } 638 639 /* 640 * Implement receipt of reply on a socket. 641 * We must search through the list of received datagrams matching them 642 * with outstanding requests using the xid, until ours is found. 643 */ 644 /* ARGSUSED */ 645 int 646 nfs_reply(myrep) 647 struct nfsreq *myrep; 648 { 649 register struct nfsreq *rep; 650 register struct nfsmount *nmp = myrep->r_nmp; 651 register int32_t t1; 652 struct mbuf *mrep, *nam, *md; 653 u_int32_t rxid, *tl; 654 caddr_t dpos, cp2; 655 int error; 656 657 /* 658 * Loop around until we get our own reply 659 */ 660 for (;;) { 661 /* 662 * Lock against other receivers so that I don't get stuck in 663 * sbwait() after someone else has received my reply for me. 664 * Also necessary for connection based protocols to avoid 665 * race conditions during a reconnect. 666 */ 667 error = nfs_rcvlock(myrep); 668 if (error == EALREADY) 669 return (0); 670 if (error) 671 return (error); 672 /* 673 * Get the next Rpc reply off the socket 674 */ 675 error = nfs_receive(myrep, &nam, &mrep); 676 nfs_rcvunlock(&nmp->nm_iflag); 677 if (error) { 678 679 /* 680 * Ignore routing errors on connectionless protocols?? 681 */ 682 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 683 nmp->nm_so->so_error = 0; 684 #ifdef DEBUG 685 printf("nfs_reply: ignoring error %d\n", error); 686 #endif 687 if (myrep->r_flags & R_GETONEREP) 688 return (0); 689 continue; 690 } 691 return (error); 692 } 693 if (nam) 694 m_freem(nam); 695 696 /* 697 * Get the xid and check that it is an rpc reply 698 */ 699 md = mrep; 700 dpos = mtod(md, caddr_t); 701 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 702 rxid = *tl++; 703 if (*tl != rpc_reply) { 704 if (nmp->nm_flag & NFSMNT_NQNFS) { 705 if (nqnfs_callback(nmp, mrep, md, dpos)) 706 nfsstats.rpcinvalid++; 707 } else { 708 nfsstats.rpcinvalid++; 709 m_freem(mrep); 710 } 711 nfsmout: 712 if (myrep->r_flags & R_GETONEREP) 713 return (0); 714 continue; 715 } 716 717 /* 718 * Loop through the request list to match up the reply 719 * Iff no match, just drop the datagram 720 */ 721 for (rep = nfs_reqq.tqh_first; rep != 0; 722 rep = rep->r_chain.tqe_next) { 723 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 724 /* Found it.. */ 725 rep->r_mrep = mrep; 726 rep->r_md = md; 727 rep->r_dpos = dpos; 728 if (nfsrtton) { 729 struct rttl *rt; 730 731 rt = &nfsrtt.rttl[nfsrtt.pos]; 732 rt->proc = rep->r_procnum; 733 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 734 rt->sent = nmp->nm_sent; 735 rt->cwnd = nmp->nm_cwnd; 736 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 737 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 738 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 739 rt->tstamp = time; 740 if (rep->r_flags & R_TIMING) 741 rt->rtt = rep->r_rtt; 742 else 743 rt->rtt = 1000000; 744 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 745 } 746 /* 747 * Update congestion window. 748 * Do the additive increase of 749 * one rpc/rtt. 750 */ 751 if (nmp->nm_cwnd <= nmp->nm_sent) { 752 nmp->nm_cwnd += 753 (NFS_CWNDSCALE * NFS_CWNDSCALE + 754 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 755 if (nmp->nm_cwnd > NFS_MAXCWND) 756 nmp->nm_cwnd = NFS_MAXCWND; 757 } 758 rep->r_flags &= ~R_SENT; 759 nmp->nm_sent -= NFS_CWNDSCALE; 760 /* 761 * Update rtt using a gain of 0.125 on the mean 762 * and a gain of 0.25 on the deviation. 763 */ 764 if (rep->r_flags & R_TIMING) { 765 /* 766 * Since the timer resolution of 767 * NFS_HZ is so course, it can often 768 * result in r_rtt == 0. Since 769 * r_rtt == N means that the actual 770 * rtt is between N+dt and N+2-dt ticks, 771 * add 1. 772 */ 773 t1 = rep->r_rtt + 1; 774 t1 -= (NFS_SRTT(rep) >> 3); 775 NFS_SRTT(rep) += t1; 776 if (t1 < 0) 777 t1 = -t1; 778 t1 -= (NFS_SDRTT(rep) >> 2); 779 NFS_SDRTT(rep) += t1; 780 } 781 nmp->nm_timeouts = 0; 782 break; 783 } 784 } 785 /* 786 * If not matched to a request, drop it. 787 * If it's mine, get out. 788 */ 789 if (rep == 0) { 790 nfsstats.rpcunexpected++; 791 m_freem(mrep); 792 } else if (rep == myrep) { 793 if (rep->r_mrep == NULL) 794 panic("nfsreply nil"); 795 return (0); 796 } 797 if (myrep->r_flags & R_GETONEREP) 798 return (0); 799 } 800 } 801 802 /* 803 * nfs_request - goes something like this 804 * - fill in request struct 805 * - links it into list 806 * - calls nfs_send() for first transmit 807 * - calls nfs_receive() to get reply 808 * - break down rpc header and return with nfs reply pointed to 809 * by mrep or error 810 * nb: always frees up mreq mbuf list 811 */ 812 int 813 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 814 struct vnode *vp; 815 struct mbuf *mrest; 816 int procnum; 817 struct proc *procp; 818 struct ucred *cred; 819 struct mbuf **mrp; 820 struct mbuf **mdp; 821 caddr_t *dposp; 822 { 823 register struct mbuf *m, *mrep; 824 register struct nfsreq *rep; 825 register u_int32_t *tl; 826 register int i; 827 struct nfsmount *nmp; 828 struct mbuf *md, *mheadend; 829 struct nfsnode *np; 830 char nickv[RPCX_NICKVERF]; 831 time_t reqtime, waituntil; 832 caddr_t dpos, cp2; 833 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; 834 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 835 int verf_len, verf_type; 836 u_int32_t xid; 837 u_quad_t frev; 838 char *auth_str, *verf_str; 839 NFSKERBKEY_T key; /* save session key */ 840 841 nmp = VFSTONFS(vp->v_mount); 842 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 843 rep->r_nmp = nmp; 844 rep->r_vp = vp; 845 rep->r_procp = procp; 846 rep->r_procnum = procnum; 847 i = 0; 848 m = mrest; 849 while (m) { 850 i += m->m_len; 851 m = m->m_next; 852 } 853 mrest_len = i; 854 855 /* 856 * Get the RPC header with authorization. 857 */ 858 kerbauth: 859 verf_str = auth_str = (char *)0; 860 if (nmp->nm_flag & NFSMNT_KERB) { 861 verf_str = nickv; 862 verf_len = sizeof (nickv); 863 auth_type = RPCAUTH_KERB4; 864 memset((caddr_t)key, 0, sizeof (key)); 865 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 866 &auth_len, verf_str, verf_len)) { 867 error = nfs_getauth(nmp, rep, cred, &auth_str, 868 &auth_len, verf_str, &verf_len, key); 869 if (error) { 870 free((caddr_t)rep, M_NFSREQ); 871 m_freem(mrest); 872 return (error); 873 } 874 } 875 } else { 876 auth_type = RPCAUTH_UNIX; 877 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 878 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 879 5 * NFSX_UNSIGNED; 880 } 881 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 882 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 883 if (auth_str) 884 free(auth_str, M_TEMP); 885 886 /* 887 * For stream protocols, insert a Sun RPC Record Mark. 888 */ 889 if (nmp->nm_sotype == SOCK_STREAM) { 890 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 891 *mtod(m, u_int32_t *) = htonl(0x80000000 | 892 (m->m_pkthdr.len - NFSX_UNSIGNED)); 893 } 894 rep->r_mreq = m; 895 rep->r_xid = xid; 896 tryagain: 897 if (nmp->nm_flag & NFSMNT_SOFT) 898 rep->r_retry = nmp->nm_retry; 899 else 900 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 901 rep->r_rtt = rep->r_rexmit = 0; 902 if (proct[procnum] > 0) 903 rep->r_flags = R_TIMING; 904 else 905 rep->r_flags = 0; 906 rep->r_mrep = NULL; 907 908 /* 909 * Do the client side RPC. 910 */ 911 nfsstats.rpcrequests++; 912 /* 913 * Chain request into list of outstanding requests. Be sure 914 * to put it LAST so timer finds oldest requests first. 915 */ 916 s = splsoftnet(); 917 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 918 919 /* Get send time for nqnfs */ 920 reqtime = time.tv_sec; 921 922 /* 923 * If backing off another request or avoiding congestion, don't 924 * send this one now but let timer do it. If not timing a request, 925 * do it now. 926 */ 927 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 928 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 929 nmp->nm_sent < nmp->nm_cwnd)) { 930 splx(s); 931 if (nmp->nm_soflags & PR_CONNREQUIRED) 932 error = nfs_sndlock(&nmp->nm_iflag, rep); 933 if (!error) { 934 m = m_copym(m, 0, M_COPYALL, M_WAIT); 935 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 936 if (nmp->nm_soflags & PR_CONNREQUIRED) 937 nfs_sndunlock(&nmp->nm_iflag); 938 } 939 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 940 nmp->nm_sent += NFS_CWNDSCALE; 941 rep->r_flags |= R_SENT; 942 } 943 } else { 944 splx(s); 945 rep->r_rtt = -1; 946 } 947 948 /* 949 * Wait for the reply from our send or the timer's. 950 */ 951 if (!error || error == EPIPE) 952 error = nfs_reply(rep); 953 954 /* 955 * RPC done, unlink the request. 956 */ 957 s = splsoftnet(); 958 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 959 splx(s); 960 961 /* 962 * Decrement the outstanding request count. 963 */ 964 if (rep->r_flags & R_SENT) { 965 rep->r_flags &= ~R_SENT; /* paranoia */ 966 nmp->nm_sent -= NFS_CWNDSCALE; 967 } 968 969 /* 970 * If there was a successful reply and a tprintf msg. 971 * tprintf a response. 972 */ 973 if (!error && (rep->r_flags & R_TPRINTFMSG)) 974 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 975 "is alive again"); 976 mrep = rep->r_mrep; 977 md = rep->r_md; 978 dpos = rep->r_dpos; 979 if (error) { 980 m_freem(rep->r_mreq); 981 free((caddr_t)rep, M_NFSREQ); 982 return (error); 983 } 984 985 /* 986 * break down the rpc header and check if ok 987 */ 988 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 989 if (*tl++ == rpc_msgdenied) { 990 if (*tl == rpc_mismatch) 991 error = EOPNOTSUPP; 992 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 993 if (!failed_auth) { 994 failed_auth++; 995 mheadend->m_next = (struct mbuf *)0; 996 m_freem(mrep); 997 m_freem(rep->r_mreq); 998 goto kerbauth; 999 } else 1000 error = EAUTH; 1001 } else 1002 error = EACCES; 1003 m_freem(mrep); 1004 m_freem(rep->r_mreq); 1005 free((caddr_t)rep, M_NFSREQ); 1006 return (error); 1007 } 1008 1009 /* 1010 * Grab any Kerberos verifier, otherwise just throw it away. 1011 */ 1012 verf_type = fxdr_unsigned(int, *tl++); 1013 i = fxdr_unsigned(int32_t, *tl); 1014 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1015 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1016 if (error) 1017 goto nfsmout; 1018 } else if (i > 0) 1019 nfsm_adv(nfsm_rndup(i)); 1020 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1021 /* 0 == ok */ 1022 if (*tl == 0) { 1023 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1024 if (*tl != 0) { 1025 error = fxdr_unsigned(int, *tl); 1026 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1027 error == NFSERR_TRYLATER) { 1028 m_freem(mrep); 1029 error = 0; 1030 waituntil = time.tv_sec + trylater_delay; 1031 while (time.tv_sec < waituntil) 1032 (void) tsleep((caddr_t)&lbolt, 1033 PSOCK, "nqnfstry", 0); 1034 trylater_delay *= nfs_backoff[trylater_cnt]; 1035 if (trylater_cnt < 7) 1036 trylater_cnt++; 1037 goto tryagain; 1038 } 1039 1040 /* 1041 * If the File Handle was stale, invalidate the 1042 * lookup cache, just in case. 1043 */ 1044 if (error == ESTALE) 1045 cache_purge(vp); 1046 if (nmp->nm_flag & NFSMNT_NFSV3) { 1047 *mrp = mrep; 1048 *mdp = md; 1049 *dposp = dpos; 1050 error |= NFSERR_RETERR; 1051 } else 1052 m_freem(mrep); 1053 m_freem(rep->r_mreq); 1054 free((caddr_t)rep, M_NFSREQ); 1055 return (error); 1056 } 1057 1058 /* 1059 * For nqnfs, get any lease in reply 1060 */ 1061 if (nmp->nm_flag & NFSMNT_NQNFS) { 1062 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1063 if (*tl) { 1064 np = VTONFS(vp); 1065 nqlflag = fxdr_unsigned(int, *tl); 1066 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1067 cachable = fxdr_unsigned(int, *tl++); 1068 reqtime += fxdr_unsigned(int, *tl++); 1069 if (reqtime > time.tv_sec) { 1070 fxdr_hyper(tl, &frev); 1071 nqnfs_clientlease(nmp, np, nqlflag, 1072 cachable, reqtime, frev); 1073 } 1074 } 1075 } 1076 *mrp = mrep; 1077 *mdp = md; 1078 *dposp = dpos; 1079 m_freem(rep->r_mreq); 1080 FREE((caddr_t)rep, M_NFSREQ); 1081 return (0); 1082 } 1083 m_freem(mrep); 1084 error = EPROTONOSUPPORT; 1085 nfsmout: 1086 m_freem(rep->r_mreq); 1087 free((caddr_t)rep, M_NFSREQ); 1088 return (error); 1089 } 1090 #endif /* NFS */ 1091 1092 /* 1093 * Generate the rpc reply header 1094 * siz arg. is used to decide if adding a cluster is worthwhile 1095 */ 1096 int 1097 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1098 int siz; 1099 struct nfsrv_descript *nd; 1100 struct nfssvc_sock *slp; 1101 int err; 1102 int cache; 1103 u_quad_t *frev; 1104 struct mbuf **mrq; 1105 struct mbuf **mbp; 1106 caddr_t *bposp; 1107 { 1108 register u_int32_t *tl; 1109 register struct mbuf *mreq; 1110 caddr_t bpos; 1111 struct mbuf *mb, *mb2; 1112 1113 MGETHDR(mreq, M_WAIT, MT_DATA); 1114 mb = mreq; 1115 /* 1116 * If this is a big reply, use a cluster else 1117 * try and leave leading space for the lower level headers. 1118 */ 1119 siz += RPC_REPLYSIZ; 1120 if (siz >= max_datalen) { 1121 MCLGET(mreq, M_WAIT); 1122 } else 1123 mreq->m_data += max_hdr; 1124 tl = mtod(mreq, u_int32_t *); 1125 mreq->m_len = 6 * NFSX_UNSIGNED; 1126 bpos = ((caddr_t)tl) + mreq->m_len; 1127 *tl++ = txdr_unsigned(nd->nd_retxid); 1128 *tl++ = rpc_reply; 1129 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1130 *tl++ = rpc_msgdenied; 1131 if (err & NFSERR_AUTHERR) { 1132 *tl++ = rpc_autherr; 1133 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1134 mreq->m_len -= NFSX_UNSIGNED; 1135 bpos -= NFSX_UNSIGNED; 1136 } else { 1137 *tl++ = rpc_mismatch; 1138 *tl++ = txdr_unsigned(RPC_VER2); 1139 *tl = txdr_unsigned(RPC_VER2); 1140 } 1141 } else { 1142 *tl++ = rpc_msgaccepted; 1143 1144 /* 1145 * For Kerberos authentication, we must send the nickname 1146 * verifier back, otherwise just RPCAUTH_NULL. 1147 */ 1148 if (nd->nd_flag & ND_KERBFULL) { 1149 register struct nfsuid *nuidp; 1150 struct timeval ktvin, ktvout; 1151 1152 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; 1153 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1154 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1155 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), 1156 &nuidp->nu_haddr, nd->nd_nam2))) 1157 break; 1158 } 1159 if (nuidp) { 1160 ktvin.tv_sec = 1161 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); 1162 ktvin.tv_usec = 1163 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1164 1165 /* 1166 * Encrypt the timestamp in ecb mode using the 1167 * session key. 1168 */ 1169 #ifdef NFSKERB 1170 XXX 1171 #endif 1172 1173 *tl++ = rpc_auth_kerb; 1174 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1175 *tl = ktvout.tv_sec; 1176 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1177 *tl++ = ktvout.tv_usec; 1178 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1179 } else { 1180 *tl++ = 0; 1181 *tl++ = 0; 1182 } 1183 } else { 1184 *tl++ = 0; 1185 *tl++ = 0; 1186 } 1187 switch (err) { 1188 case EPROGUNAVAIL: 1189 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1190 break; 1191 case EPROGMISMATCH: 1192 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1193 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1194 if (nd->nd_flag & ND_NQNFS) { 1195 *tl++ = txdr_unsigned(3); 1196 *tl = txdr_unsigned(3); 1197 } else { 1198 *tl++ = txdr_unsigned(2); 1199 *tl = txdr_unsigned(3); 1200 } 1201 break; 1202 case EPROCUNAVAIL: 1203 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1204 break; 1205 case EBADRPC: 1206 *tl = txdr_unsigned(RPC_GARBAGE); 1207 break; 1208 default: 1209 *tl = 0; 1210 if (err != NFSERR_RETVOID) { 1211 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1212 if (err) 1213 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1214 else 1215 *tl = 0; 1216 } 1217 break; 1218 }; 1219 } 1220 1221 /* 1222 * For nqnfs, piggyback lease as requested. 1223 */ 1224 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1225 if (nd->nd_flag & ND_LEASE) { 1226 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1227 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1228 *tl++ = txdr_unsigned(cache); 1229 *tl++ = txdr_unsigned(nd->nd_duration); 1230 txdr_hyper(frev, tl); 1231 } else { 1232 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1233 *tl = 0; 1234 } 1235 } 1236 if (mrq != NULL) 1237 *mrq = mreq; 1238 *mbp = mb; 1239 *bposp = bpos; 1240 if (err != 0 && err != NFSERR_RETVOID) 1241 nfsstats.srvrpc_errs++; 1242 return (0); 1243 } 1244 1245 /* 1246 * Nfs timer routine 1247 * Scan the nfsreq list and retranmit any requests that have timed out 1248 * To avoid retransmission attempts on STREAM sockets (in the future) make 1249 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1250 */ 1251 void 1252 nfs_timer(arg) 1253 void *arg; /* never used */ 1254 { 1255 register struct nfsreq *rep; 1256 register struct mbuf *m; 1257 register struct socket *so; 1258 register struct nfsmount *nmp; 1259 register int timeo; 1260 int s, error; 1261 #ifdef NFSSERVER 1262 register struct nfssvc_sock *slp; 1263 static long lasttime = 0; 1264 u_quad_t cur_usec; 1265 #endif 1266 1267 s = splsoftnet(); 1268 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1269 nmp = rep->r_nmp; 1270 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1271 continue; 1272 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1273 rep->r_flags |= R_SOFTTERM; 1274 continue; 1275 } 1276 if (rep->r_rtt >= 0) { 1277 rep->r_rtt++; 1278 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1279 timeo = nmp->nm_timeo; 1280 else 1281 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1282 if (nmp->nm_timeouts > 0) 1283 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1284 if (rep->r_rtt <= timeo) 1285 continue; 1286 if (nmp->nm_timeouts < 8) 1287 nmp->nm_timeouts++; 1288 } 1289 /* 1290 * Check for server not responding 1291 */ 1292 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1293 rep->r_rexmit > nmp->nm_deadthresh) { 1294 nfs_msg(rep->r_procp, 1295 nmp->nm_mountp->mnt_stat.f_mntfromname, 1296 "not responding"); 1297 rep->r_flags |= R_TPRINTFMSG; 1298 } 1299 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1300 nfsstats.rpctimeouts++; 1301 rep->r_flags |= R_SOFTTERM; 1302 continue; 1303 } 1304 if (nmp->nm_sotype != SOCK_DGRAM) { 1305 if (++rep->r_rexmit > NFS_MAXREXMIT) 1306 rep->r_rexmit = NFS_MAXREXMIT; 1307 continue; 1308 } 1309 if ((so = nmp->nm_so) == NULL) 1310 continue; 1311 1312 /* 1313 * If there is enough space and the window allows.. 1314 * Resend it 1315 * Set r_rtt to -1 in case we fail to send it now. 1316 */ 1317 rep->r_rtt = -1; 1318 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1319 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1320 (rep->r_flags & R_SENT) || 1321 nmp->nm_sent < nmp->nm_cwnd) && 1322 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1323 if (so->so_state & SS_ISCONNECTED) 1324 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1325 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1326 else 1327 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1328 nmp->nm_nam, (struct mbuf *)0, (struct proc *)0); 1329 if (error) { 1330 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1331 #ifdef DEBUG 1332 printf("nfs_timer: ignoring error %d\n", 1333 error); 1334 #endif 1335 so->so_error = 0; 1336 } 1337 } else { 1338 /* 1339 * Iff first send, start timing 1340 * else turn timing off, backoff timer 1341 * and divide congestion window by 2. 1342 */ 1343 if (rep->r_flags & R_SENT) { 1344 rep->r_flags &= ~R_TIMING; 1345 if (++rep->r_rexmit > NFS_MAXREXMIT) 1346 rep->r_rexmit = NFS_MAXREXMIT; 1347 nmp->nm_cwnd >>= 1; 1348 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1349 nmp->nm_cwnd = NFS_CWNDSCALE; 1350 nfsstats.rpcretries++; 1351 } else { 1352 rep->r_flags |= R_SENT; 1353 nmp->nm_sent += NFS_CWNDSCALE; 1354 } 1355 rep->r_rtt = 0; 1356 } 1357 } 1358 } 1359 1360 #ifdef NFSSERVER 1361 /* 1362 * Call the nqnfs server timer once a second to handle leases. 1363 */ 1364 if (lasttime != time.tv_sec) { 1365 lasttime = time.tv_sec; 1366 nqnfs_serverd(); 1367 } 1368 1369 /* 1370 * Scan the write gathering queues for writes that need to be 1371 * completed now. 1372 */ 1373 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1374 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 1375 slp = slp->ns_chain.tqe_next) { 1376 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec) 1377 nfsrv_wakenfsd(slp); 1378 } 1379 #endif /* NFSSERVER */ 1380 splx(s); 1381 timeout(nfs_timer, (void *)0, nfs_ticks); 1382 } 1383 1384 /* 1385 * Test for a termination condition pending on the process. 1386 * This is used for NFSMNT_INT mounts. 1387 */ 1388 int 1389 nfs_sigintr(nmp, rep, p) 1390 struct nfsmount *nmp; 1391 struct nfsreq *rep; 1392 register struct proc *p; 1393 { 1394 sigset_t ss; 1395 1396 if (rep && (rep->r_flags & R_SOFTTERM)) 1397 return (EINTR); 1398 if (!(nmp->nm_flag & NFSMNT_INT)) 1399 return (0); 1400 if (p) { 1401 sigpending1(p, &ss); 1402 #if 0 1403 sigminusset(&p->p_sigignore, &ss); 1404 #endif 1405 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1406 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1407 sigismember(&ss, SIGQUIT)) 1408 return (EINTR); 1409 } 1410 return (0); 1411 } 1412 1413 /* 1414 * Lock a socket against others. 1415 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1416 * and also to avoid race conditions between the processes with nfs requests 1417 * in progress when a reconnect is necessary. 1418 */ 1419 int 1420 nfs_sndlock(flagp, rep) 1421 register int *flagp; 1422 struct nfsreq *rep; 1423 { 1424 struct proc *p; 1425 int slpflag = 0, slptimeo = 0; 1426 1427 if (rep) { 1428 p = rep->r_procp; 1429 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1430 slpflag = PCATCH; 1431 } else 1432 p = (struct proc *)0; 1433 while (*flagp & NFSMNT_SNDLOCK) { 1434 if (nfs_sigintr(rep->r_nmp, rep, p)) 1435 return (EINTR); 1436 *flagp |= NFSMNT_WANTSND; 1437 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1438 slptimeo); 1439 if (slpflag == PCATCH) { 1440 slpflag = 0; 1441 slptimeo = 2 * hz; 1442 } 1443 } 1444 *flagp |= NFSMNT_SNDLOCK; 1445 return (0); 1446 } 1447 1448 /* 1449 * Unlock the stream socket for others. 1450 */ 1451 void 1452 nfs_sndunlock(flagp) 1453 register int *flagp; 1454 { 1455 1456 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1457 panic("nfs sndunlock"); 1458 *flagp &= ~NFSMNT_SNDLOCK; 1459 if (*flagp & NFSMNT_WANTSND) { 1460 *flagp &= ~NFSMNT_WANTSND; 1461 wakeup((caddr_t)flagp); 1462 } 1463 } 1464 1465 int 1466 nfs_rcvlock(rep) 1467 register struct nfsreq *rep; 1468 { 1469 register int *flagp = &rep->r_nmp->nm_iflag; 1470 int slpflag, slptimeo = 0; 1471 1472 if (*flagp & NFSMNT_INT) 1473 slpflag = PCATCH; 1474 else 1475 slpflag = 0; 1476 while (*flagp & NFSMNT_RCVLOCK) { 1477 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1478 return (EINTR); 1479 *flagp |= NFSMNT_WANTRCV; 1480 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1481 slptimeo); 1482 /* If our reply was received while we were sleeping, 1483 * then just return without taking the lock to avoid a 1484 * situation where a single iod could 'capture' the 1485 * receive lock. 1486 */ 1487 if (rep->r_mrep != NULL) 1488 return (EALREADY); 1489 if (slpflag == PCATCH) { 1490 slpflag = 0; 1491 slptimeo = 2 * hz; 1492 } 1493 } 1494 *flagp |= NFSMNT_RCVLOCK; 1495 return (0); 1496 } 1497 1498 /* 1499 * Unlock the stream socket for others. 1500 */ 1501 void 1502 nfs_rcvunlock(flagp) 1503 register int *flagp; 1504 { 1505 1506 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1507 panic("nfs rcvunlock"); 1508 *flagp &= ~NFSMNT_RCVLOCK; 1509 if (*flagp & NFSMNT_WANTRCV) { 1510 *flagp &= ~NFSMNT_WANTRCV; 1511 wakeup((caddr_t)flagp); 1512 } 1513 } 1514 1515 /* 1516 * Parse an RPC request 1517 * - verify it 1518 * - fill in the cred struct. 1519 */ 1520 int 1521 nfs_getreq(nd, nfsd, has_header) 1522 register struct nfsrv_descript *nd; 1523 struct nfsd *nfsd; 1524 int has_header; 1525 { 1526 register int len, i; 1527 register u_int32_t *tl; 1528 register int32_t t1; 1529 struct uio uio; 1530 struct iovec iov; 1531 caddr_t dpos, cp2, cp; 1532 u_int32_t nfsvers, auth_type; 1533 uid_t nickuid; 1534 int error = 0, nqnfs = 0, ticklen; 1535 struct mbuf *mrep, *md; 1536 register struct nfsuid *nuidp; 1537 struct timeval tvin, tvout; 1538 1539 mrep = nd->nd_mrep; 1540 md = nd->nd_md; 1541 dpos = nd->nd_dpos; 1542 if (has_header) { 1543 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1544 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1545 if (*tl++ != rpc_call) { 1546 m_freem(mrep); 1547 return (EBADRPC); 1548 } 1549 } else 1550 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1551 nd->nd_repstat = 0; 1552 nd->nd_flag = 0; 1553 if (*tl++ != rpc_vers) { 1554 nd->nd_repstat = ERPCMISMATCH; 1555 nd->nd_procnum = NFSPROC_NOOP; 1556 return (0); 1557 } 1558 if (*tl != nfs_prog) { 1559 if (*tl == nqnfs_prog) 1560 nqnfs++; 1561 else { 1562 nd->nd_repstat = EPROGUNAVAIL; 1563 nd->nd_procnum = NFSPROC_NOOP; 1564 return (0); 1565 } 1566 } 1567 tl++; 1568 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1569 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1570 (nfsvers != NQNFS_VER3 && nqnfs)) { 1571 nd->nd_repstat = EPROGMISMATCH; 1572 nd->nd_procnum = NFSPROC_NOOP; 1573 return (0); 1574 } 1575 if (nqnfs) 1576 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1577 else if (nfsvers == NFS_VER3) 1578 nd->nd_flag = ND_NFSV3; 1579 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1580 if (nd->nd_procnum == NFSPROC_NULL) 1581 return (0); 1582 if (nd->nd_procnum >= NFS_NPROCS || 1583 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1584 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1585 nd->nd_repstat = EPROCUNAVAIL; 1586 nd->nd_procnum = NFSPROC_NOOP; 1587 return (0); 1588 } 1589 if ((nd->nd_flag & ND_NFSV3) == 0) 1590 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1591 auth_type = *tl++; 1592 len = fxdr_unsigned(int, *tl++); 1593 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1594 m_freem(mrep); 1595 return (EBADRPC); 1596 } 1597 1598 nd->nd_flag &= ~ND_KERBAUTH; 1599 /* 1600 * Handle auth_unix or auth_kerb. 1601 */ 1602 if (auth_type == rpc_auth_unix) { 1603 len = fxdr_unsigned(int, *++tl); 1604 if (len < 0 || len > NFS_MAXNAMLEN) { 1605 m_freem(mrep); 1606 return (EBADRPC); 1607 } 1608 nfsm_adv(nfsm_rndup(len)); 1609 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1610 memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred)); 1611 nd->nd_cr.cr_ref = 1; 1612 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1613 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1614 len = fxdr_unsigned(int, *tl); 1615 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1616 m_freem(mrep); 1617 return (EBADRPC); 1618 } 1619 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1620 for (i = 0; i < len; i++) 1621 if (i < NGROUPS) 1622 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1623 else 1624 tl++; 1625 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1626 if (nd->nd_cr.cr_ngroups > 1) 1627 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1628 len = fxdr_unsigned(int, *++tl); 1629 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1630 m_freem(mrep); 1631 return (EBADRPC); 1632 } 1633 if (len > 0) 1634 nfsm_adv(nfsm_rndup(len)); 1635 } else if (auth_type == rpc_auth_kerb) { 1636 switch (fxdr_unsigned(int, *tl++)) { 1637 case RPCAKN_FULLNAME: 1638 ticklen = fxdr_unsigned(int, *tl); 1639 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1640 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1641 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1642 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1643 m_freem(mrep); 1644 return (EBADRPC); 1645 } 1646 uio.uio_offset = 0; 1647 uio.uio_iov = &iov; 1648 uio.uio_iovcnt = 1; 1649 uio.uio_segflg = UIO_SYSSPACE; 1650 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1651 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1652 nfsm_mtouio(&uio, uio.uio_resid); 1653 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1654 if (*tl++ != rpc_auth_kerb || 1655 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1656 printf("Bad kerb verifier\n"); 1657 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1658 nd->nd_procnum = NFSPROC_NOOP; 1659 return (0); 1660 } 1661 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1662 tl = (u_int32_t *)cp; 1663 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1664 printf("Not fullname kerb verifier\n"); 1665 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1666 nd->nd_procnum = NFSPROC_NOOP; 1667 return (0); 1668 } 1669 cp += NFSX_UNSIGNED; 1670 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 1671 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1672 nd->nd_flag |= ND_KERBFULL; 1673 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1674 break; 1675 case RPCAKN_NICKNAME: 1676 if (len != 2 * NFSX_UNSIGNED) { 1677 printf("Kerb nickname short\n"); 1678 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1679 nd->nd_procnum = NFSPROC_NOOP; 1680 return (0); 1681 } 1682 nickuid = fxdr_unsigned(uid_t, *tl); 1683 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1684 if (*tl++ != rpc_auth_kerb || 1685 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1686 printf("Kerb nick verifier bad\n"); 1687 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1688 nd->nd_procnum = NFSPROC_NOOP; 1689 return (0); 1690 } 1691 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1692 tvin.tv_sec = *tl++; 1693 tvin.tv_usec = *tl; 1694 1695 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; 1696 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1697 if (nuidp->nu_cr.cr_uid == nickuid && 1698 (!nd->nd_nam2 || 1699 netaddr_match(NU_NETFAM(nuidp), 1700 &nuidp->nu_haddr, nd->nd_nam2))) 1701 break; 1702 } 1703 if (!nuidp) { 1704 nd->nd_repstat = 1705 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1706 nd->nd_procnum = NFSPROC_NOOP; 1707 return (0); 1708 } 1709 1710 /* 1711 * Now, decrypt the timestamp using the session key 1712 * and validate it. 1713 */ 1714 #ifdef NFSKERB 1715 XXX 1716 #endif 1717 1718 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1719 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1720 if (nuidp->nu_expire < time.tv_sec || 1721 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1722 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1723 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1724 nuidp->nu_expire = 0; 1725 nd->nd_repstat = 1726 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1727 nd->nd_procnum = NFSPROC_NOOP; 1728 return (0); 1729 } 1730 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1731 nd->nd_flag |= ND_KERBNICK; 1732 }; 1733 } else { 1734 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1735 nd->nd_procnum = NFSPROC_NOOP; 1736 return (0); 1737 } 1738 1739 /* 1740 * For nqnfs, get piggybacked lease request. 1741 */ 1742 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1743 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1744 nd->nd_flag |= fxdr_unsigned(int, *tl); 1745 if (nd->nd_flag & ND_LEASE) { 1746 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1747 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 1748 } else 1749 nd->nd_duration = NQ_MINLEASE; 1750 } else 1751 nd->nd_duration = NQ_MINLEASE; 1752 nd->nd_md = md; 1753 nd->nd_dpos = dpos; 1754 return (0); 1755 nfsmout: 1756 return (error); 1757 } 1758 1759 int 1760 nfs_msg(p, server, msg) 1761 struct proc *p; 1762 char *server, *msg; 1763 { 1764 tpr_t tpr; 1765 1766 if (p) 1767 tpr = tprintf_open(p); 1768 else 1769 tpr = NULL; 1770 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1771 tprintf_close(tpr); 1772 return (0); 1773 } 1774 1775 #ifdef NFSSERVER 1776 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 1777 struct nfssvc_sock *, struct proc *, 1778 struct mbuf **)) = { 1779 nfsrv_null, 1780 nfsrv_getattr, 1781 nfsrv_setattr, 1782 nfsrv_lookup, 1783 nfsrv3_access, 1784 nfsrv_readlink, 1785 nfsrv_read, 1786 nfsrv_write, 1787 nfsrv_create, 1788 nfsrv_mkdir, 1789 nfsrv_symlink, 1790 nfsrv_mknod, 1791 nfsrv_remove, 1792 nfsrv_rmdir, 1793 nfsrv_rename, 1794 nfsrv_link, 1795 nfsrv_readdir, 1796 nfsrv_readdirplus, 1797 nfsrv_statfs, 1798 nfsrv_fsinfo, 1799 nfsrv_pathconf, 1800 nfsrv_commit, 1801 nqnfsrv_getlease, 1802 nqnfsrv_vacated, 1803 nfsrv_noop, 1804 nfsrv_noop 1805 }; 1806 1807 /* 1808 * Socket upcall routine for the nfsd sockets. 1809 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1810 * Essentially do as much as possible non-blocking, else punt and it will 1811 * be called with M_WAIT from an nfsd. 1812 */ 1813 void 1814 nfsrv_rcv(so, arg, waitflag) 1815 struct socket *so; 1816 caddr_t arg; 1817 int waitflag; 1818 { 1819 register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1820 register struct mbuf *m; 1821 struct mbuf *mp, *nam; 1822 struct uio auio; 1823 int flags, error; 1824 1825 if ((slp->ns_flag & SLP_VALID) == 0) 1826 return; 1827 #ifdef notdef 1828 /* 1829 * Define this to test for nfsds handling this under heavy load. 1830 */ 1831 if (waitflag == M_DONTWAIT) { 1832 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1833 } 1834 #endif 1835 auio.uio_procp = NULL; 1836 if (so->so_type == SOCK_STREAM) { 1837 /* 1838 * If there are already records on the queue, defer soreceive() 1839 * to an nfsd so that there is feedback to the TCP layer that 1840 * the nfs servers are heavily loaded. 1841 */ 1842 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1843 slp->ns_flag |= SLP_NEEDQ; 1844 goto dorecs; 1845 } 1846 1847 /* 1848 * Do soreceive(). 1849 */ 1850 auio.uio_resid = 1000000000; 1851 flags = MSG_DONTWAIT; 1852 error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1853 if (error || mp == (struct mbuf *)0) { 1854 if (error == EWOULDBLOCK) 1855 slp->ns_flag |= SLP_NEEDQ; 1856 else 1857 slp->ns_flag |= SLP_DISCONN; 1858 goto dorecs; 1859 } 1860 m = mp; 1861 if (slp->ns_rawend) { 1862 slp->ns_rawend->m_next = m; 1863 slp->ns_cc += 1000000000 - auio.uio_resid; 1864 } else { 1865 slp->ns_raw = m; 1866 slp->ns_cc = 1000000000 - auio.uio_resid; 1867 } 1868 while (m->m_next) 1869 m = m->m_next; 1870 slp->ns_rawend = m; 1871 1872 /* 1873 * Now try and parse record(s) out of the raw stream data. 1874 */ 1875 error = nfsrv_getstream(slp, waitflag); 1876 if (error) { 1877 if (error == EPERM) 1878 slp->ns_flag |= SLP_DISCONN; 1879 else 1880 slp->ns_flag |= SLP_NEEDQ; 1881 } 1882 } else { 1883 do { 1884 auio.uio_resid = 1000000000; 1885 flags = MSG_DONTWAIT; 1886 error = (*so->so_receive)(so, &nam, &auio, &mp, 1887 (struct mbuf **)0, &flags); 1888 if (mp) { 1889 if (nam) { 1890 m = nam; 1891 m->m_next = mp; 1892 } else 1893 m = mp; 1894 if (slp->ns_recend) 1895 slp->ns_recend->m_nextpkt = m; 1896 else 1897 slp->ns_rec = m; 1898 slp->ns_recend = m; 1899 m->m_nextpkt = (struct mbuf *)0; 1900 } 1901 if (error) { 1902 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1903 && error != EWOULDBLOCK) { 1904 slp->ns_flag |= SLP_DISCONN; 1905 goto dorecs; 1906 } 1907 } 1908 } while (mp); 1909 } 1910 1911 /* 1912 * Now try and process the request records, non-blocking. 1913 */ 1914 dorecs: 1915 if (waitflag == M_DONTWAIT && 1916 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1917 nfsrv_wakenfsd(slp); 1918 } 1919 1920 /* 1921 * Try and extract an RPC request from the mbuf data list received on a 1922 * stream socket. The "waitflag" argument indicates whether or not it 1923 * can sleep. 1924 */ 1925 int 1926 nfsrv_getstream(slp, waitflag) 1927 register struct nfssvc_sock *slp; 1928 int waitflag; 1929 { 1930 register struct mbuf *m, **mpp; 1931 register char *cp1, *cp2; 1932 register int len; 1933 struct mbuf *om, *m2, *recm = NULL; 1934 u_int32_t recmark; 1935 1936 if (slp->ns_flag & SLP_GETSTREAM) 1937 panic("nfs getstream"); 1938 slp->ns_flag |= SLP_GETSTREAM; 1939 for (;;) { 1940 if (slp->ns_reclen == 0) { 1941 if (slp->ns_cc < NFSX_UNSIGNED) { 1942 slp->ns_flag &= ~SLP_GETSTREAM; 1943 return (0); 1944 } 1945 m = slp->ns_raw; 1946 if (m->m_len >= NFSX_UNSIGNED) { 1947 memcpy((caddr_t)&recmark, mtod(m, caddr_t), NFSX_UNSIGNED); 1948 m->m_data += NFSX_UNSIGNED; 1949 m->m_len -= NFSX_UNSIGNED; 1950 } else { 1951 cp1 = (caddr_t)&recmark; 1952 cp2 = mtod(m, caddr_t); 1953 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1954 while (m->m_len == 0) { 1955 m = m->m_next; 1956 cp2 = mtod(m, caddr_t); 1957 } 1958 *cp1++ = *cp2++; 1959 m->m_data++; 1960 m->m_len--; 1961 } 1962 } 1963 slp->ns_cc -= NFSX_UNSIGNED; 1964 recmark = ntohl(recmark); 1965 slp->ns_reclen = recmark & ~0x80000000; 1966 if (recmark & 0x80000000) 1967 slp->ns_flag |= SLP_LASTFRAG; 1968 else 1969 slp->ns_flag &= ~SLP_LASTFRAG; 1970 if (slp->ns_reclen > NFS_MAXPACKET) { 1971 slp->ns_flag &= ~SLP_GETSTREAM; 1972 return (EPERM); 1973 } 1974 } 1975 1976 /* 1977 * Now get the record part. 1978 */ 1979 if (slp->ns_cc == slp->ns_reclen) { 1980 recm = slp->ns_raw; 1981 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 1982 slp->ns_cc = slp->ns_reclen = 0; 1983 } else if (slp->ns_cc > slp->ns_reclen) { 1984 len = 0; 1985 m = slp->ns_raw; 1986 om = (struct mbuf *)0; 1987 while (len < slp->ns_reclen) { 1988 if ((len + m->m_len) > slp->ns_reclen) { 1989 size_t left = slp->ns_reclen - len; 1990 1991 MGETHDR(m2, waitflag, m->m_type); 1992 if (m2 == NULL) { 1993 slp->ns_flag &= ~SLP_GETSTREAM; 1994 return (EWOULDBLOCK); 1995 } 1996 if (left > MHLEN) { 1997 MCLGET(m2, waitflag); 1998 if (!(m2->m_flags & M_EXT)) { 1999 m_freem(m2); 2000 slp->ns_flag &= ~SLP_GETSTREAM; 2001 return (EWOULDBLOCK); 2002 } 2003 } 2004 memcpy(mtod(m2, caddr_t), mtod(m, caddr_t), 2005 left); 2006 m2->m_len = left; 2007 m->m_data += left; 2008 m->m_len -= left; 2009 if (om) { 2010 om->m_next = m2; 2011 recm = slp->ns_raw; 2012 } else 2013 recm = m2; 2014 len = slp->ns_reclen; 2015 } else if ((len + m->m_len) == slp->ns_reclen) { 2016 om = m; 2017 len += m->m_len; 2018 m = m->m_next; 2019 recm = slp->ns_raw; 2020 om->m_next = (struct mbuf *)0; 2021 } else { 2022 om = m; 2023 len += m->m_len; 2024 m = m->m_next; 2025 } 2026 } 2027 slp->ns_raw = m; 2028 slp->ns_cc -= len; 2029 slp->ns_reclen = 0; 2030 } else { 2031 slp->ns_flag &= ~SLP_GETSTREAM; 2032 return (0); 2033 } 2034 2035 /* 2036 * Accumulate the fragments into a record. 2037 */ 2038 mpp = &slp->ns_frag; 2039 while (*mpp) 2040 mpp = &((*mpp)->m_next); 2041 *mpp = recm; 2042 if (slp->ns_flag & SLP_LASTFRAG) { 2043 if (slp->ns_recend) 2044 slp->ns_recend->m_nextpkt = slp->ns_frag; 2045 else 2046 slp->ns_rec = slp->ns_frag; 2047 slp->ns_recend = slp->ns_frag; 2048 slp->ns_frag = (struct mbuf *)0; 2049 } 2050 } 2051 } 2052 2053 /* 2054 * Parse an RPC header. 2055 */ 2056 int 2057 nfsrv_dorec(slp, nfsd, ndp) 2058 register struct nfssvc_sock *slp; 2059 struct nfsd *nfsd; 2060 struct nfsrv_descript **ndp; 2061 { 2062 register struct mbuf *m, *nam; 2063 register struct nfsrv_descript *nd; 2064 int error; 2065 2066 *ndp = NULL; 2067 if ((slp->ns_flag & SLP_VALID) == 0 || 2068 (m = slp->ns_rec) == (struct mbuf *)0) 2069 return (ENOBUFS); 2070 slp->ns_rec = m->m_nextpkt; 2071 if (slp->ns_rec) 2072 m->m_nextpkt = (struct mbuf *)0; 2073 else 2074 slp->ns_recend = (struct mbuf *)0; 2075 if (m->m_type == MT_SONAME) { 2076 nam = m; 2077 m = m->m_next; 2078 nam->m_next = NULL; 2079 } else 2080 nam = NULL; 2081 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), 2082 M_NFSRVDESC, M_WAITOK); 2083 nd->nd_md = nd->nd_mrep = m; 2084 nd->nd_nam2 = nam; 2085 nd->nd_dpos = mtod(m, caddr_t); 2086 error = nfs_getreq(nd, nfsd, TRUE); 2087 if (error) { 2088 m_freem(nam); 2089 free((caddr_t)nd, M_NFSRVDESC); 2090 return (error); 2091 } 2092 *ndp = nd; 2093 nfsd->nfsd_nd = nd; 2094 return (0); 2095 } 2096 2097 2098 /* 2099 * Search for a sleeping nfsd and wake it up. 2100 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2101 * running nfsds will go look for the work in the nfssvc_sock list. 2102 */ 2103 void 2104 nfsrv_wakenfsd(slp) 2105 struct nfssvc_sock *slp; 2106 { 2107 register struct nfsd *nd; 2108 2109 if ((slp->ns_flag & SLP_VALID) == 0) 2110 return; 2111 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) { 2112 if (nd->nfsd_flag & NFSD_WAITING) { 2113 nd->nfsd_flag &= ~NFSD_WAITING; 2114 if (nd->nfsd_slp) 2115 panic("nfsd wakeup"); 2116 slp->ns_sref++; 2117 nd->nfsd_slp = slp; 2118 wakeup((caddr_t)nd); 2119 return; 2120 } 2121 } 2122 slp->ns_flag |= SLP_DOREC; 2123 nfsd_head_flag |= NFSD_CHECKSLP; 2124 } 2125 #endif /* NFSSERVER */ 2126