nfs_socket.c revision 1.52
1 /* $NetBSD: nfs_socket.c,v 1.52 1999/07/30 09:46:24 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 if (nmp->nm_iflag & NFSMNT_DISMNT) { 333 /* 334 * soshutdown() above should wake up the current 335 * listener. 336 * Now wake up those waiting for the recive lock, and 337 * wait for them to go away unhappy, to prevent *nmp 338 * from evaporating while they're sleeping. 339 */ 340 while (nmp->nm_waiters > 0) { 341 wakeup (&nmp->nm_iflag); 342 sleep(&nmp->nm_waiters, PVFS); 343 } 344 } 345 soclose(so); 346 } 347 #ifdef DIAGNOSTIC 348 if (nmp->nm_waiters > 0) 349 panic("nfs_disconnect: waiters left after drain?\n"); 350 #endif 351 } 352 353 void 354 nfs_safedisconnect(nmp) 355 struct nfsmount *nmp; 356 { 357 struct nfsreq dummyreq; 358 359 memset(&dummyreq, 0, sizeof(dummyreq)); 360 dummyreq.r_nmp = nmp; 361 nfs_rcvlock(&dummyreq); /* XXX ignored error return */ 362 nfs_disconnect(nmp); 363 nfs_rcvunlock(&nmp->nm_iflag); 364 } 365 366 /* 367 * This is the nfs send routine. For connection based socket types, it 368 * must be called with an nfs_sndlock() on the socket. 369 * "rep == NULL" indicates that it has been called from a server. 370 * For the client side: 371 * - return EINTR if the RPC is terminated, 0 otherwise 372 * - set R_MUSTRESEND if the send fails for any reason 373 * - do any cleanup required by recoverable socket errors (???) 374 * For the server side: 375 * - return EINTR or ERESTART if interrupted by a signal 376 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 377 * - do any cleanup required by recoverable socket errors (???) 378 */ 379 int 380 nfs_send(so, nam, top, rep) 381 register struct socket *so; 382 struct mbuf *nam; 383 register struct mbuf *top; 384 struct nfsreq *rep; 385 { 386 struct mbuf *sendnam; 387 int error, soflags, flags; 388 389 if (rep) { 390 if (rep->r_flags & R_SOFTTERM) { 391 m_freem(top); 392 return (EINTR); 393 } 394 if ((so = rep->r_nmp->nm_so) == NULL) { 395 rep->r_flags |= R_MUSTRESEND; 396 m_freem(top); 397 return (0); 398 } 399 rep->r_flags &= ~R_MUSTRESEND; 400 soflags = rep->r_nmp->nm_soflags; 401 } else 402 soflags = so->so_proto->pr_flags; 403 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 404 sendnam = (struct mbuf *)0; 405 else 406 sendnam = nam; 407 if (so->so_type == SOCK_SEQPACKET) 408 flags = MSG_EOR; 409 else 410 flags = 0; 411 412 error = (*so->so_send)(so, sendnam, (struct uio *)0, top, 413 (struct mbuf *)0, flags); 414 if (error) { 415 if (rep) { 416 log(LOG_INFO, "nfs send error %d for server %s\n",error, 417 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 418 /* 419 * Deal with errors for the client side. 420 */ 421 if (rep->r_flags & R_SOFTTERM) 422 error = EINTR; 423 else 424 rep->r_flags |= R_MUSTRESEND; 425 } else 426 log(LOG_INFO, "nfsd send error %d\n", error); 427 428 /* 429 * Handle any recoverable (soft) socket errors here. (???) 430 */ 431 if (error != EINTR && error != ERESTART && 432 error != EWOULDBLOCK && error != EPIPE) 433 error = 0; 434 } 435 return (error); 436 } 437 438 #ifdef NFS 439 /* 440 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 441 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 442 * Mark and consolidate the data into a new mbuf list. 443 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 444 * small mbufs. 445 * For SOCK_STREAM we must be very careful to read an entire record once 446 * we have read any of it, even if the system call has been interrupted. 447 */ 448 int 449 nfs_receive(rep, aname, mp) 450 register struct nfsreq *rep; 451 struct mbuf **aname; 452 struct mbuf **mp; 453 { 454 register struct socket *so; 455 struct uio auio; 456 struct iovec aio; 457 register struct mbuf *m; 458 struct mbuf *control; 459 u_int32_t len; 460 struct mbuf **getnam; 461 int error, sotype, rcvflg; 462 struct proc *p = curproc; /* XXX */ 463 464 /* 465 * Set up arguments for soreceive() 466 */ 467 *mp = (struct mbuf *)0; 468 *aname = (struct mbuf *)0; 469 sotype = rep->r_nmp->nm_sotype; 470 471 /* 472 * For reliable protocols, lock against other senders/receivers 473 * in case a reconnect is necessary. 474 * For SOCK_STREAM, first get the Record Mark to find out how much 475 * more there is to get. 476 * We must lock the socket against other receivers 477 * until we have an entire rpc request/reply. 478 */ 479 if (sotype != SOCK_DGRAM) { 480 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 481 if (error) 482 return (error); 483 tryagain: 484 /* 485 * Check for fatal errors and resending request. 486 */ 487 /* 488 * Ugh: If a reconnect attempt just happened, nm_so 489 * would have changed. NULL indicates a failed 490 * attempt that has essentially shut down this 491 * mount point. 492 */ 493 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 494 nfs_sndunlock(&rep->r_nmp->nm_iflag); 495 return (EINTR); 496 } 497 so = rep->r_nmp->nm_so; 498 if (!so) { 499 error = nfs_reconnect(rep); 500 if (error) { 501 nfs_sndunlock(&rep->r_nmp->nm_iflag); 502 return (error); 503 } 504 goto tryagain; 505 } 506 while (rep->r_flags & R_MUSTRESEND) { 507 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 508 nfsstats.rpcretries++; 509 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 510 if (error) { 511 if (error == EINTR || error == ERESTART || 512 (error = nfs_reconnect(rep)) != 0) { 513 nfs_sndunlock(&rep->r_nmp->nm_iflag); 514 return (error); 515 } 516 goto tryagain; 517 } 518 } 519 nfs_sndunlock(&rep->r_nmp->nm_iflag); 520 if (sotype == SOCK_STREAM) { 521 aio.iov_base = (caddr_t) &len; 522 aio.iov_len = sizeof(u_int32_t); 523 auio.uio_iov = &aio; 524 auio.uio_iovcnt = 1; 525 auio.uio_segflg = UIO_SYSSPACE; 526 auio.uio_rw = UIO_READ; 527 auio.uio_offset = 0; 528 auio.uio_resid = sizeof(u_int32_t); 529 auio.uio_procp = p; 530 do { 531 rcvflg = MSG_WAITALL; 532 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 533 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 534 if (error == EWOULDBLOCK && rep) { 535 if (rep->r_flags & R_SOFTTERM) 536 return (EINTR); 537 } 538 } while (error == EWOULDBLOCK); 539 if (!error && auio.uio_resid > 0) { 540 /* 541 * Don't log a 0 byte receive; it means 542 * that the socket has been closed, and 543 * can happen during normal operation 544 * (forcible unmount or Solaris server). 545 */ 546 if (auio.uio_resid != sizeof (u_int32_t)) 547 log(LOG_INFO, 548 "short receive (%lu/%lu) from nfs server %s\n", 549 (u_long)sizeof(u_int32_t) - auio.uio_resid, 550 (u_long)sizeof(u_int32_t), 551 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 552 error = EPIPE; 553 } 554 if (error) 555 goto errout; 556 len = ntohl(len) & ~0x80000000; 557 /* 558 * This is SERIOUS! We are out of sync with the sender 559 * and forcing a disconnect/reconnect is all I can do. 560 */ 561 if (len > NFS_MAXPACKET) { 562 log(LOG_ERR, "%s (%d) from nfs server %s\n", 563 "impossible packet length", 564 len, 565 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 566 error = EFBIG; 567 goto errout; 568 } 569 auio.uio_resid = len; 570 do { 571 rcvflg = MSG_WAITALL; 572 error = (*so->so_receive)(so, (struct mbuf **)0, 573 &auio, mp, (struct mbuf **)0, &rcvflg); 574 } while (error == EWOULDBLOCK || error == EINTR || 575 error == ERESTART); 576 if (!error && auio.uio_resid > 0) { 577 if (len != auio.uio_resid) 578 log(LOG_INFO, 579 "short receive (%lu/%d) from nfs server %s\n", 580 (u_long)len - auio.uio_resid, len, 581 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 582 error = EPIPE; 583 } 584 } else { 585 /* 586 * NB: Since uio_resid is big, MSG_WAITALL is ignored 587 * and soreceive() will return when it has either a 588 * control msg or a data msg. 589 * We have no use for control msg., but must grab them 590 * and then throw them away so we know what is going 591 * on. 592 */ 593 auio.uio_resid = len = 100000000; /* Anything Big */ 594 auio.uio_procp = p; 595 do { 596 rcvflg = 0; 597 error = (*so->so_receive)(so, (struct mbuf **)0, 598 &auio, mp, &control, &rcvflg); 599 if (control) 600 m_freem(control); 601 if (error == EWOULDBLOCK && rep) { 602 if (rep->r_flags & R_SOFTTERM) 603 return (EINTR); 604 } 605 } while (error == EWOULDBLOCK || 606 (!error && *mp == NULL && control)); 607 if ((rcvflg & MSG_EOR) == 0) 608 printf("Egad!!\n"); 609 if (!error && *mp == NULL) 610 error = EPIPE; 611 len -= auio.uio_resid; 612 } 613 errout: 614 if (error && error != EINTR && error != ERESTART) { 615 m_freem(*mp); 616 *mp = (struct mbuf *)0; 617 if (error != EPIPE) 618 log(LOG_INFO, 619 "receive error %d from nfs server %s\n", 620 error, 621 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 622 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 623 if (!error) 624 error = nfs_reconnect(rep); 625 if (!error) 626 goto tryagain; 627 else 628 nfs_sndunlock(&rep->r_nmp->nm_iflag); 629 } 630 } else { 631 if ((so = rep->r_nmp->nm_so) == NULL) 632 return (EACCES); 633 if (so->so_state & SS_ISCONNECTED) 634 getnam = (struct mbuf **)0; 635 else 636 getnam = aname; 637 auio.uio_resid = len = 1000000; 638 auio.uio_procp = p; 639 do { 640 rcvflg = 0; 641 error = (*so->so_receive)(so, getnam, &auio, mp, 642 (struct mbuf **)0, &rcvflg); 643 if (error == EWOULDBLOCK && 644 (rep->r_flags & R_SOFTTERM)) 645 return (EINTR); 646 } while (error == EWOULDBLOCK); 647 len -= auio.uio_resid; 648 if (!error && *mp == NULL) 649 error = EPIPE; 650 } 651 if (error) { 652 m_freem(*mp); 653 *mp = (struct mbuf *)0; 654 } 655 return (error); 656 } 657 658 /* 659 * Implement receipt of reply on a socket. 660 * We must search through the list of received datagrams matching them 661 * with outstanding requests using the xid, until ours is found. 662 */ 663 /* ARGSUSED */ 664 int 665 nfs_reply(myrep) 666 struct nfsreq *myrep; 667 { 668 register struct nfsreq *rep; 669 register struct nfsmount *nmp = myrep->r_nmp; 670 register int32_t t1; 671 struct mbuf *mrep, *nam, *md; 672 u_int32_t rxid, *tl; 673 caddr_t dpos, cp2; 674 int error; 675 676 /* 677 * Loop around until we get our own reply 678 */ 679 for (;;) { 680 /* 681 * Lock against other receivers so that I don't get stuck in 682 * sbwait() after someone else has received my reply for me. 683 * Also necessary for connection based protocols to avoid 684 * race conditions during a reconnect. 685 */ 686 error = nfs_rcvlock(myrep); 687 if (error == EALREADY) 688 return (0); 689 if (error) 690 return (error); 691 /* 692 * Get the next Rpc reply off the socket 693 */ 694 nmp->nm_waiters++; 695 error = nfs_receive(myrep, &nam, &mrep); 696 nfs_rcvunlock(&nmp->nm_iflag); 697 if (error) { 698 699 if (nmp->nm_iflag & NFSMNT_DISMNT) { 700 /* 701 * Oops, we're going away now.. 702 */ 703 nmp->nm_waiters--; 704 wakeup (&nmp->nm_waiters); 705 return error; 706 } 707 nmp->nm_waiters--; 708 /* 709 * Ignore routing errors on connectionless protocols?? 710 */ 711 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 712 nmp->nm_so->so_error = 0; 713 #ifdef DEBUG 714 printf("nfs_reply: ignoring error %d\n", error); 715 #endif 716 if (myrep->r_flags & R_GETONEREP) 717 return (0); 718 continue; 719 } 720 return (error); 721 } 722 nmp->nm_waiters--; 723 if (nam) 724 m_freem(nam); 725 726 /* 727 * Get the xid and check that it is an rpc reply 728 */ 729 md = mrep; 730 dpos = mtod(md, caddr_t); 731 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 732 rxid = *tl++; 733 if (*tl != rpc_reply) { 734 if (nmp->nm_flag & NFSMNT_NQNFS) { 735 if (nqnfs_callback(nmp, mrep, md, dpos)) 736 nfsstats.rpcinvalid++; 737 } else { 738 nfsstats.rpcinvalid++; 739 m_freem(mrep); 740 } 741 nfsmout: 742 if (myrep->r_flags & R_GETONEREP) 743 return (0); 744 continue; 745 } 746 747 /* 748 * Loop through the request list to match up the reply 749 * Iff no match, just drop the datagram 750 */ 751 for (rep = nfs_reqq.tqh_first; rep != 0; 752 rep = rep->r_chain.tqe_next) { 753 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 754 /* Found it.. */ 755 rep->r_mrep = mrep; 756 rep->r_md = md; 757 rep->r_dpos = dpos; 758 if (nfsrtton) { 759 struct rttl *rt; 760 761 rt = &nfsrtt.rttl[nfsrtt.pos]; 762 rt->proc = rep->r_procnum; 763 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 764 rt->sent = nmp->nm_sent; 765 rt->cwnd = nmp->nm_cwnd; 766 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 767 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 768 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 769 rt->tstamp = time; 770 if (rep->r_flags & R_TIMING) 771 rt->rtt = rep->r_rtt; 772 else 773 rt->rtt = 1000000; 774 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 775 } 776 /* 777 * Update congestion window. 778 * Do the additive increase of 779 * one rpc/rtt. 780 */ 781 if (nmp->nm_cwnd <= nmp->nm_sent) { 782 nmp->nm_cwnd += 783 (NFS_CWNDSCALE * NFS_CWNDSCALE + 784 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 785 if (nmp->nm_cwnd > NFS_MAXCWND) 786 nmp->nm_cwnd = NFS_MAXCWND; 787 } 788 rep->r_flags &= ~R_SENT; 789 nmp->nm_sent -= NFS_CWNDSCALE; 790 /* 791 * Update rtt using a gain of 0.125 on the mean 792 * and a gain of 0.25 on the deviation. 793 */ 794 if (rep->r_flags & R_TIMING) { 795 /* 796 * Since the timer resolution of 797 * NFS_HZ is so course, it can often 798 * result in r_rtt == 0. Since 799 * r_rtt == N means that the actual 800 * rtt is between N+dt and N+2-dt ticks, 801 * add 1. 802 */ 803 t1 = rep->r_rtt + 1; 804 t1 -= (NFS_SRTT(rep) >> 3); 805 NFS_SRTT(rep) += t1; 806 if (t1 < 0) 807 t1 = -t1; 808 t1 -= (NFS_SDRTT(rep) >> 2); 809 NFS_SDRTT(rep) += t1; 810 } 811 nmp->nm_timeouts = 0; 812 break; 813 } 814 } 815 /* 816 * If not matched to a request, drop it. 817 * If it's mine, get out. 818 */ 819 if (rep == 0) { 820 nfsstats.rpcunexpected++; 821 m_freem(mrep); 822 } else if (rep == myrep) { 823 if (rep->r_mrep == NULL) 824 panic("nfsreply nil"); 825 return (0); 826 } 827 if (myrep->r_flags & R_GETONEREP) 828 return (0); 829 } 830 } 831 832 /* 833 * nfs_request - goes something like this 834 * - fill in request struct 835 * - links it into list 836 * - calls nfs_send() for first transmit 837 * - calls nfs_receive() to get reply 838 * - break down rpc header and return with nfs reply pointed to 839 * by mrep or error 840 * nb: always frees up mreq mbuf list 841 */ 842 int 843 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 844 struct vnode *vp; 845 struct mbuf *mrest; 846 int procnum; 847 struct proc *procp; 848 struct ucred *cred; 849 struct mbuf **mrp; 850 struct mbuf **mdp; 851 caddr_t *dposp; 852 { 853 register struct mbuf *m, *mrep; 854 register struct nfsreq *rep; 855 register u_int32_t *tl; 856 register int i; 857 struct nfsmount *nmp; 858 struct mbuf *md, *mheadend; 859 struct nfsnode *np; 860 char nickv[RPCX_NICKVERF]; 861 time_t reqtime, waituntil; 862 caddr_t dpos, cp2; 863 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; 864 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 865 int verf_len, verf_type; 866 u_int32_t xid; 867 u_quad_t frev; 868 char *auth_str, *verf_str; 869 NFSKERBKEY_T key; /* save session key */ 870 871 nmp = VFSTONFS(vp->v_mount); 872 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 873 rep->r_nmp = nmp; 874 rep->r_vp = vp; 875 rep->r_procp = procp; 876 rep->r_procnum = procnum; 877 i = 0; 878 m = mrest; 879 while (m) { 880 i += m->m_len; 881 m = m->m_next; 882 } 883 mrest_len = i; 884 885 /* 886 * Get the RPC header with authorization. 887 */ 888 kerbauth: 889 verf_str = auth_str = (char *)0; 890 if (nmp->nm_flag & NFSMNT_KERB) { 891 verf_str = nickv; 892 verf_len = sizeof (nickv); 893 auth_type = RPCAUTH_KERB4; 894 memset((caddr_t)key, 0, sizeof (key)); 895 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 896 &auth_len, verf_str, verf_len)) { 897 error = nfs_getauth(nmp, rep, cred, &auth_str, 898 &auth_len, verf_str, &verf_len, key); 899 if (error) { 900 free((caddr_t)rep, M_NFSREQ); 901 m_freem(mrest); 902 return (error); 903 } 904 } 905 } else { 906 auth_type = RPCAUTH_UNIX; 907 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 908 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 909 5 * NFSX_UNSIGNED; 910 } 911 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 912 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 913 if (auth_str) 914 free(auth_str, M_TEMP); 915 916 /* 917 * For stream protocols, insert a Sun RPC Record Mark. 918 */ 919 if (nmp->nm_sotype == SOCK_STREAM) { 920 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 921 *mtod(m, u_int32_t *) = htonl(0x80000000 | 922 (m->m_pkthdr.len - NFSX_UNSIGNED)); 923 } 924 rep->r_mreq = m; 925 rep->r_xid = xid; 926 tryagain: 927 if (nmp->nm_flag & NFSMNT_SOFT) 928 rep->r_retry = nmp->nm_retry; 929 else 930 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 931 rep->r_rtt = rep->r_rexmit = 0; 932 if (proct[procnum] > 0) 933 rep->r_flags = R_TIMING; 934 else 935 rep->r_flags = 0; 936 rep->r_mrep = NULL; 937 938 /* 939 * Do the client side RPC. 940 */ 941 nfsstats.rpcrequests++; 942 /* 943 * Chain request into list of outstanding requests. Be sure 944 * to put it LAST so timer finds oldest requests first. 945 */ 946 s = splsoftnet(); 947 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 948 949 /* Get send time for nqnfs */ 950 reqtime = time.tv_sec; 951 952 /* 953 * If backing off another request or avoiding congestion, don't 954 * send this one now but let timer do it. If not timing a request, 955 * do it now. 956 */ 957 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 958 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 959 nmp->nm_sent < nmp->nm_cwnd)) { 960 splx(s); 961 if (nmp->nm_soflags & PR_CONNREQUIRED) 962 error = nfs_sndlock(&nmp->nm_iflag, rep); 963 if (!error) { 964 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 965 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 966 if (nmp->nm_soflags & PR_CONNREQUIRED) 967 nfs_sndunlock(&nmp->nm_iflag); 968 } 969 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 970 nmp->nm_sent += NFS_CWNDSCALE; 971 rep->r_flags |= R_SENT; 972 } 973 } else { 974 splx(s); 975 rep->r_rtt = -1; 976 } 977 978 /* 979 * Wait for the reply from our send or the timer's. 980 */ 981 if (!error || error == EPIPE) 982 error = nfs_reply(rep); 983 984 /* 985 * RPC done, unlink the request. 986 */ 987 s = splsoftnet(); 988 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 989 splx(s); 990 991 /* 992 * Decrement the outstanding request count. 993 */ 994 if (rep->r_flags & R_SENT) { 995 rep->r_flags &= ~R_SENT; /* paranoia */ 996 nmp->nm_sent -= NFS_CWNDSCALE; 997 } 998 999 /* 1000 * If there was a successful reply and a tprintf msg. 1001 * tprintf a response. 1002 */ 1003 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1004 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1005 "is alive again"); 1006 mrep = rep->r_mrep; 1007 md = rep->r_md; 1008 dpos = rep->r_dpos; 1009 if (error) { 1010 m_freem(rep->r_mreq); 1011 free((caddr_t)rep, M_NFSREQ); 1012 return (error); 1013 } 1014 1015 /* 1016 * break down the rpc header and check if ok 1017 */ 1018 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1019 if (*tl++ == rpc_msgdenied) { 1020 if (*tl == rpc_mismatch) 1021 error = EOPNOTSUPP; 1022 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1023 if (!failed_auth) { 1024 failed_auth++; 1025 mheadend->m_next = (struct mbuf *)0; 1026 m_freem(mrep); 1027 m_freem(rep->r_mreq); 1028 goto kerbauth; 1029 } else 1030 error = EAUTH; 1031 } else 1032 error = EACCES; 1033 m_freem(mrep); 1034 m_freem(rep->r_mreq); 1035 free((caddr_t)rep, M_NFSREQ); 1036 return (error); 1037 } 1038 1039 /* 1040 * Grab any Kerberos verifier, otherwise just throw it away. 1041 */ 1042 verf_type = fxdr_unsigned(int, *tl++); 1043 i = fxdr_unsigned(int32_t, *tl); 1044 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1045 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1046 if (error) 1047 goto nfsmout; 1048 } else if (i > 0) 1049 nfsm_adv(nfsm_rndup(i)); 1050 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1051 /* 0 == ok */ 1052 if (*tl == 0) { 1053 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1054 if (*tl != 0) { 1055 error = fxdr_unsigned(int, *tl); 1056 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1057 error == NFSERR_TRYLATER) { 1058 m_freem(mrep); 1059 error = 0; 1060 waituntil = time.tv_sec + trylater_delay; 1061 while (time.tv_sec < waituntil) 1062 (void) tsleep((caddr_t)&lbolt, 1063 PSOCK, "nqnfstry", 0); 1064 trylater_delay *= nfs_backoff[trylater_cnt]; 1065 if (trylater_cnt < 7) 1066 trylater_cnt++; 1067 goto tryagain; 1068 } 1069 1070 /* 1071 * If the File Handle was stale, invalidate the 1072 * lookup cache, just in case. 1073 */ 1074 if (error == ESTALE) 1075 cache_purge(vp); 1076 if (nmp->nm_flag & NFSMNT_NFSV3) { 1077 *mrp = mrep; 1078 *mdp = md; 1079 *dposp = dpos; 1080 error |= NFSERR_RETERR; 1081 } else 1082 m_freem(mrep); 1083 m_freem(rep->r_mreq); 1084 free((caddr_t)rep, M_NFSREQ); 1085 return (error); 1086 } 1087 1088 /* 1089 * For nqnfs, get any lease in reply 1090 */ 1091 if (nmp->nm_flag & NFSMNT_NQNFS) { 1092 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1093 if (*tl) { 1094 np = VTONFS(vp); 1095 nqlflag = fxdr_unsigned(int, *tl); 1096 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1097 cachable = fxdr_unsigned(int, *tl++); 1098 reqtime += fxdr_unsigned(int, *tl++); 1099 if (reqtime > time.tv_sec) { 1100 frev = fxdr_hyper(tl); 1101 nqnfs_clientlease(nmp, np, nqlflag, 1102 cachable, reqtime, frev); 1103 } 1104 } 1105 } 1106 *mrp = mrep; 1107 *mdp = md; 1108 *dposp = dpos; 1109 m_freem(rep->r_mreq); 1110 FREE((caddr_t)rep, M_NFSREQ); 1111 return (0); 1112 } 1113 m_freem(mrep); 1114 error = EPROTONOSUPPORT; 1115 nfsmout: 1116 m_freem(rep->r_mreq); 1117 free((caddr_t)rep, M_NFSREQ); 1118 return (error); 1119 } 1120 #endif /* NFS */ 1121 1122 /* 1123 * Generate the rpc reply header 1124 * siz arg. is used to decide if adding a cluster is worthwhile 1125 */ 1126 int 1127 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1128 int siz; 1129 struct nfsrv_descript *nd; 1130 struct nfssvc_sock *slp; 1131 int err; 1132 int cache; 1133 u_quad_t *frev; 1134 struct mbuf **mrq; 1135 struct mbuf **mbp; 1136 caddr_t *bposp; 1137 { 1138 register u_int32_t *tl; 1139 register struct mbuf *mreq; 1140 caddr_t bpos; 1141 struct mbuf *mb, *mb2; 1142 1143 MGETHDR(mreq, M_WAIT, MT_DATA); 1144 mb = mreq; 1145 /* 1146 * If this is a big reply, use a cluster else 1147 * try and leave leading space for the lower level headers. 1148 */ 1149 siz += RPC_REPLYSIZ; 1150 if (siz >= max_datalen) { 1151 MCLGET(mreq, M_WAIT); 1152 } else 1153 mreq->m_data += max_hdr; 1154 tl = mtod(mreq, u_int32_t *); 1155 mreq->m_len = 6 * NFSX_UNSIGNED; 1156 bpos = ((caddr_t)tl) + mreq->m_len; 1157 *tl++ = txdr_unsigned(nd->nd_retxid); 1158 *tl++ = rpc_reply; 1159 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1160 *tl++ = rpc_msgdenied; 1161 if (err & NFSERR_AUTHERR) { 1162 *tl++ = rpc_autherr; 1163 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1164 mreq->m_len -= NFSX_UNSIGNED; 1165 bpos -= NFSX_UNSIGNED; 1166 } else { 1167 *tl++ = rpc_mismatch; 1168 *tl++ = txdr_unsigned(RPC_VER2); 1169 *tl = txdr_unsigned(RPC_VER2); 1170 } 1171 } else { 1172 *tl++ = rpc_msgaccepted; 1173 1174 /* 1175 * For Kerberos authentication, we must send the nickname 1176 * verifier back, otherwise just RPCAUTH_NULL. 1177 */ 1178 if (nd->nd_flag & ND_KERBFULL) { 1179 register struct nfsuid *nuidp; 1180 struct timeval ktvin, ktvout; 1181 1182 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; 1183 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1184 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1185 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), 1186 &nuidp->nu_haddr, nd->nd_nam2))) 1187 break; 1188 } 1189 if (nuidp) { 1190 ktvin.tv_sec = 1191 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); 1192 ktvin.tv_usec = 1193 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1194 1195 /* 1196 * Encrypt the timestamp in ecb mode using the 1197 * session key. 1198 */ 1199 #ifdef NFSKERB 1200 XXX 1201 #endif 1202 1203 *tl++ = rpc_auth_kerb; 1204 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1205 *tl = ktvout.tv_sec; 1206 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1207 *tl++ = ktvout.tv_usec; 1208 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1209 } else { 1210 *tl++ = 0; 1211 *tl++ = 0; 1212 } 1213 } else { 1214 *tl++ = 0; 1215 *tl++ = 0; 1216 } 1217 switch (err) { 1218 case EPROGUNAVAIL: 1219 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1220 break; 1221 case EPROGMISMATCH: 1222 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1223 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1224 if (nd->nd_flag & ND_NQNFS) { 1225 *tl++ = txdr_unsigned(3); 1226 *tl = txdr_unsigned(3); 1227 } else { 1228 *tl++ = txdr_unsigned(2); 1229 *tl = txdr_unsigned(3); 1230 } 1231 break; 1232 case EPROCUNAVAIL: 1233 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1234 break; 1235 case EBADRPC: 1236 *tl = txdr_unsigned(RPC_GARBAGE); 1237 break; 1238 default: 1239 *tl = 0; 1240 if (err != NFSERR_RETVOID) { 1241 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1242 if (err) 1243 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1244 else 1245 *tl = 0; 1246 } 1247 break; 1248 }; 1249 } 1250 1251 /* 1252 * For nqnfs, piggyback lease as requested. 1253 */ 1254 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1255 if (nd->nd_flag & ND_LEASE) { 1256 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1257 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1258 *tl++ = txdr_unsigned(cache); 1259 *tl++ = txdr_unsigned(nd->nd_duration); 1260 txdr_hyper(*frev, tl); 1261 } else { 1262 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1263 *tl = 0; 1264 } 1265 } 1266 if (mrq != NULL) 1267 *mrq = mreq; 1268 *mbp = mb; 1269 *bposp = bpos; 1270 if (err != 0 && err != NFSERR_RETVOID) 1271 nfsstats.srvrpc_errs++; 1272 return (0); 1273 } 1274 1275 /* 1276 * Nfs timer routine 1277 * Scan the nfsreq list and retranmit any requests that have timed out 1278 * To avoid retransmission attempts on STREAM sockets (in the future) make 1279 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1280 */ 1281 void 1282 nfs_timer(arg) 1283 void *arg; /* never used */ 1284 { 1285 register struct nfsreq *rep; 1286 register struct mbuf *m; 1287 register struct socket *so; 1288 register struct nfsmount *nmp; 1289 register int timeo; 1290 int s, error; 1291 #ifdef NFSSERVER 1292 register struct nfssvc_sock *slp; 1293 static long lasttime = 0; 1294 u_quad_t cur_usec; 1295 #endif 1296 1297 s = splsoftnet(); 1298 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1299 nmp = rep->r_nmp; 1300 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1301 continue; 1302 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1303 rep->r_flags |= R_SOFTTERM; 1304 continue; 1305 } 1306 if (rep->r_rtt >= 0) { 1307 rep->r_rtt++; 1308 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1309 timeo = nmp->nm_timeo; 1310 else 1311 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1312 if (nmp->nm_timeouts > 0) 1313 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1314 if (rep->r_rtt <= timeo) 1315 continue; 1316 if (nmp->nm_timeouts < 8) 1317 nmp->nm_timeouts++; 1318 } 1319 /* 1320 * Check for server not responding 1321 */ 1322 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1323 rep->r_rexmit > nmp->nm_deadthresh) { 1324 nfs_msg(rep->r_procp, 1325 nmp->nm_mountp->mnt_stat.f_mntfromname, 1326 "not responding"); 1327 rep->r_flags |= R_TPRINTFMSG; 1328 } 1329 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1330 nfsstats.rpctimeouts++; 1331 rep->r_flags |= R_SOFTTERM; 1332 continue; 1333 } 1334 if (nmp->nm_sotype != SOCK_DGRAM) { 1335 if (++rep->r_rexmit > NFS_MAXREXMIT) 1336 rep->r_rexmit = NFS_MAXREXMIT; 1337 continue; 1338 } 1339 if ((so = nmp->nm_so) == NULL) 1340 continue; 1341 1342 /* 1343 * If there is enough space and the window allows.. 1344 * Resend it 1345 * Set r_rtt to -1 in case we fail to send it now. 1346 */ 1347 rep->r_rtt = -1; 1348 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1349 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1350 (rep->r_flags & R_SENT) || 1351 nmp->nm_sent < nmp->nm_cwnd) && 1352 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1353 if (so->so_state & SS_ISCONNECTED) 1354 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1355 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1356 else 1357 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1358 nmp->nm_nam, (struct mbuf *)0, (struct proc *)0); 1359 if (error) { 1360 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1361 #ifdef DEBUG 1362 printf("nfs_timer: ignoring error %d\n", 1363 error); 1364 #endif 1365 so->so_error = 0; 1366 } 1367 } else { 1368 /* 1369 * Iff first send, start timing 1370 * else turn timing off, backoff timer 1371 * and divide congestion window by 2. 1372 */ 1373 if (rep->r_flags & R_SENT) { 1374 rep->r_flags &= ~R_TIMING; 1375 if (++rep->r_rexmit > NFS_MAXREXMIT) 1376 rep->r_rexmit = NFS_MAXREXMIT; 1377 nmp->nm_cwnd >>= 1; 1378 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1379 nmp->nm_cwnd = NFS_CWNDSCALE; 1380 nfsstats.rpcretries++; 1381 } else { 1382 rep->r_flags |= R_SENT; 1383 nmp->nm_sent += NFS_CWNDSCALE; 1384 } 1385 rep->r_rtt = 0; 1386 } 1387 } 1388 } 1389 1390 #ifdef NFSSERVER 1391 /* 1392 * Call the nqnfs server timer once a second to handle leases. 1393 */ 1394 if (lasttime != time.tv_sec) { 1395 lasttime = time.tv_sec; 1396 nqnfs_serverd(); 1397 } 1398 1399 /* 1400 * Scan the write gathering queues for writes that need to be 1401 * completed now. 1402 */ 1403 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1404 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 1405 slp = slp->ns_chain.tqe_next) { 1406 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec) 1407 nfsrv_wakenfsd(slp); 1408 } 1409 #endif /* NFSSERVER */ 1410 splx(s); 1411 timeout(nfs_timer, (void *)0, nfs_ticks); 1412 } 1413 1414 /* 1415 * Test for a termination condition pending on the process. 1416 * This is used for NFSMNT_INT mounts. 1417 */ 1418 int 1419 nfs_sigintr(nmp, rep, p) 1420 struct nfsmount *nmp; 1421 struct nfsreq *rep; 1422 register struct proc *p; 1423 { 1424 sigset_t ss; 1425 1426 if (rep && (rep->r_flags & R_SOFTTERM)) 1427 return (EINTR); 1428 if (!(nmp->nm_flag & NFSMNT_INT)) 1429 return (0); 1430 if (p) { 1431 sigpending1(p, &ss); 1432 #if 0 1433 sigminusset(&p->p_sigignore, &ss); 1434 #endif 1435 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1436 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1437 sigismember(&ss, SIGQUIT)) 1438 return (EINTR); 1439 } 1440 return (0); 1441 } 1442 1443 /* 1444 * Lock a socket against others. 1445 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1446 * and also to avoid race conditions between the processes with nfs requests 1447 * in progress when a reconnect is necessary. 1448 */ 1449 int 1450 nfs_sndlock(flagp, rep) 1451 register int *flagp; 1452 struct nfsreq *rep; 1453 { 1454 struct proc *p; 1455 int slpflag = 0, slptimeo = 0; 1456 1457 if (rep) { 1458 p = rep->r_procp; 1459 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1460 slpflag = PCATCH; 1461 } else 1462 p = (struct proc *)0; 1463 while (*flagp & NFSMNT_SNDLOCK) { 1464 if (nfs_sigintr(rep->r_nmp, rep, p)) 1465 return (EINTR); 1466 *flagp |= NFSMNT_WANTSND; 1467 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1468 slptimeo); 1469 if (slpflag == PCATCH) { 1470 slpflag = 0; 1471 slptimeo = 2 * hz; 1472 } 1473 } 1474 *flagp |= NFSMNT_SNDLOCK; 1475 return (0); 1476 } 1477 1478 /* 1479 * Unlock the stream socket for others. 1480 */ 1481 void 1482 nfs_sndunlock(flagp) 1483 register int *flagp; 1484 { 1485 1486 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1487 panic("nfs sndunlock"); 1488 *flagp &= ~NFSMNT_SNDLOCK; 1489 if (*flagp & NFSMNT_WANTSND) { 1490 *flagp &= ~NFSMNT_WANTSND; 1491 wakeup((caddr_t)flagp); 1492 } 1493 } 1494 1495 int 1496 nfs_rcvlock(rep) 1497 register struct nfsreq *rep; 1498 { 1499 struct nfsmount *nmp = rep->r_nmp; 1500 register int *flagp = &nmp->nm_iflag; 1501 int slpflag, slptimeo = 0; 1502 1503 if (*flagp & NFSMNT_DISMNT) 1504 return EIO; 1505 1506 if (*flagp & NFSMNT_INT) 1507 slpflag = PCATCH; 1508 else 1509 slpflag = 0; 1510 while (*flagp & NFSMNT_RCVLOCK) { 1511 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1512 return (EINTR); 1513 *flagp |= NFSMNT_WANTRCV; 1514 nmp->nm_waiters++; 1515 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1516 slptimeo); 1517 nmp->nm_waiters--; 1518 if (*flagp & NFSMNT_DISMNT) { 1519 wakeup(&nmp->nm_waiters); 1520 return EIO; 1521 } 1522 /* If our reply was received while we were sleeping, 1523 * then just return without taking the lock to avoid a 1524 * situation where a single iod could 'capture' the 1525 * receive lock. 1526 */ 1527 if (rep->r_mrep != NULL) 1528 return (EALREADY); 1529 if (slpflag == PCATCH) { 1530 slpflag = 0; 1531 slptimeo = 2 * hz; 1532 } 1533 } 1534 *flagp |= NFSMNT_RCVLOCK; 1535 return (0); 1536 } 1537 1538 /* 1539 * Unlock the stream socket for others. 1540 */ 1541 void 1542 nfs_rcvunlock(flagp) 1543 register int *flagp; 1544 { 1545 1546 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1547 panic("nfs rcvunlock"); 1548 *flagp &= ~NFSMNT_RCVLOCK; 1549 if (*flagp & NFSMNT_WANTRCV) { 1550 *flagp &= ~NFSMNT_WANTRCV; 1551 wakeup((caddr_t)flagp); 1552 } 1553 } 1554 1555 /* 1556 * Parse an RPC request 1557 * - verify it 1558 * - fill in the cred struct. 1559 */ 1560 int 1561 nfs_getreq(nd, nfsd, has_header) 1562 register struct nfsrv_descript *nd; 1563 struct nfsd *nfsd; 1564 int has_header; 1565 { 1566 register int len, i; 1567 register u_int32_t *tl; 1568 register int32_t t1; 1569 struct uio uio; 1570 struct iovec iov; 1571 caddr_t dpos, cp2, cp; 1572 u_int32_t nfsvers, auth_type; 1573 uid_t nickuid; 1574 int error = 0, nqnfs = 0, ticklen; 1575 struct mbuf *mrep, *md; 1576 register struct nfsuid *nuidp; 1577 struct timeval tvin, tvout; 1578 1579 mrep = nd->nd_mrep; 1580 md = nd->nd_md; 1581 dpos = nd->nd_dpos; 1582 if (has_header) { 1583 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1584 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1585 if (*tl++ != rpc_call) { 1586 m_freem(mrep); 1587 return (EBADRPC); 1588 } 1589 } else 1590 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1591 nd->nd_repstat = 0; 1592 nd->nd_flag = 0; 1593 if (*tl++ != rpc_vers) { 1594 nd->nd_repstat = ERPCMISMATCH; 1595 nd->nd_procnum = NFSPROC_NOOP; 1596 return (0); 1597 } 1598 if (*tl != nfs_prog) { 1599 if (*tl == nqnfs_prog) 1600 nqnfs++; 1601 else { 1602 nd->nd_repstat = EPROGUNAVAIL; 1603 nd->nd_procnum = NFSPROC_NOOP; 1604 return (0); 1605 } 1606 } 1607 tl++; 1608 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1609 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1610 (nfsvers != NQNFS_VER3 && nqnfs)) { 1611 nd->nd_repstat = EPROGMISMATCH; 1612 nd->nd_procnum = NFSPROC_NOOP; 1613 return (0); 1614 } 1615 if (nqnfs) 1616 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1617 else if (nfsvers == NFS_VER3) 1618 nd->nd_flag = ND_NFSV3; 1619 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1620 if (nd->nd_procnum == NFSPROC_NULL) 1621 return (0); 1622 if (nd->nd_procnum >= NFS_NPROCS || 1623 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1624 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1625 nd->nd_repstat = EPROCUNAVAIL; 1626 nd->nd_procnum = NFSPROC_NOOP; 1627 return (0); 1628 } 1629 if ((nd->nd_flag & ND_NFSV3) == 0) 1630 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1631 auth_type = *tl++; 1632 len = fxdr_unsigned(int, *tl++); 1633 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1634 m_freem(mrep); 1635 return (EBADRPC); 1636 } 1637 1638 nd->nd_flag &= ~ND_KERBAUTH; 1639 /* 1640 * Handle auth_unix or auth_kerb. 1641 */ 1642 if (auth_type == rpc_auth_unix) { 1643 len = fxdr_unsigned(int, *++tl); 1644 if (len < 0 || len > NFS_MAXNAMLEN) { 1645 m_freem(mrep); 1646 return (EBADRPC); 1647 } 1648 nfsm_adv(nfsm_rndup(len)); 1649 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1650 memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred)); 1651 nd->nd_cr.cr_ref = 1; 1652 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1653 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1654 len = fxdr_unsigned(int, *tl); 1655 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1656 m_freem(mrep); 1657 return (EBADRPC); 1658 } 1659 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1660 for (i = 0; i < len; i++) 1661 if (i < NGROUPS) 1662 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1663 else 1664 tl++; 1665 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1666 if (nd->nd_cr.cr_ngroups > 1) 1667 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1668 len = fxdr_unsigned(int, *++tl); 1669 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1670 m_freem(mrep); 1671 return (EBADRPC); 1672 } 1673 if (len > 0) 1674 nfsm_adv(nfsm_rndup(len)); 1675 } else if (auth_type == rpc_auth_kerb) { 1676 switch (fxdr_unsigned(int, *tl++)) { 1677 case RPCAKN_FULLNAME: 1678 ticklen = fxdr_unsigned(int, *tl); 1679 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1680 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1681 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1682 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1683 m_freem(mrep); 1684 return (EBADRPC); 1685 } 1686 uio.uio_offset = 0; 1687 uio.uio_iov = &iov; 1688 uio.uio_iovcnt = 1; 1689 uio.uio_segflg = UIO_SYSSPACE; 1690 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1691 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1692 nfsm_mtouio(&uio, uio.uio_resid); 1693 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1694 if (*tl++ != rpc_auth_kerb || 1695 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1696 printf("Bad kerb verifier\n"); 1697 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1698 nd->nd_procnum = NFSPROC_NOOP; 1699 return (0); 1700 } 1701 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1702 tl = (u_int32_t *)cp; 1703 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1704 printf("Not fullname kerb verifier\n"); 1705 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1706 nd->nd_procnum = NFSPROC_NOOP; 1707 return (0); 1708 } 1709 cp += NFSX_UNSIGNED; 1710 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 1711 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1712 nd->nd_flag |= ND_KERBFULL; 1713 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1714 break; 1715 case RPCAKN_NICKNAME: 1716 if (len != 2 * NFSX_UNSIGNED) { 1717 printf("Kerb nickname short\n"); 1718 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1719 nd->nd_procnum = NFSPROC_NOOP; 1720 return (0); 1721 } 1722 nickuid = fxdr_unsigned(uid_t, *tl); 1723 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1724 if (*tl++ != rpc_auth_kerb || 1725 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1726 printf("Kerb nick verifier bad\n"); 1727 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1728 nd->nd_procnum = NFSPROC_NOOP; 1729 return (0); 1730 } 1731 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1732 tvin.tv_sec = *tl++; 1733 tvin.tv_usec = *tl; 1734 1735 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; 1736 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1737 if (nuidp->nu_cr.cr_uid == nickuid && 1738 (!nd->nd_nam2 || 1739 netaddr_match(NU_NETFAM(nuidp), 1740 &nuidp->nu_haddr, nd->nd_nam2))) 1741 break; 1742 } 1743 if (!nuidp) { 1744 nd->nd_repstat = 1745 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1746 nd->nd_procnum = NFSPROC_NOOP; 1747 return (0); 1748 } 1749 1750 /* 1751 * Now, decrypt the timestamp using the session key 1752 * and validate it. 1753 */ 1754 #ifdef NFSKERB 1755 XXX 1756 #endif 1757 1758 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1759 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1760 if (nuidp->nu_expire < time.tv_sec || 1761 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1762 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1763 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1764 nuidp->nu_expire = 0; 1765 nd->nd_repstat = 1766 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1767 nd->nd_procnum = NFSPROC_NOOP; 1768 return (0); 1769 } 1770 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1771 nd->nd_flag |= ND_KERBNICK; 1772 }; 1773 } else { 1774 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1775 nd->nd_procnum = NFSPROC_NOOP; 1776 return (0); 1777 } 1778 1779 /* 1780 * For nqnfs, get piggybacked lease request. 1781 */ 1782 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1783 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1784 nd->nd_flag |= fxdr_unsigned(int, *tl); 1785 if (nd->nd_flag & ND_LEASE) { 1786 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1787 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 1788 } else 1789 nd->nd_duration = NQ_MINLEASE; 1790 } else 1791 nd->nd_duration = NQ_MINLEASE; 1792 nd->nd_md = md; 1793 nd->nd_dpos = dpos; 1794 return (0); 1795 nfsmout: 1796 return (error); 1797 } 1798 1799 int 1800 nfs_msg(p, server, msg) 1801 struct proc *p; 1802 char *server, *msg; 1803 { 1804 tpr_t tpr; 1805 1806 if (p) 1807 tpr = tprintf_open(p); 1808 else 1809 tpr = NULL; 1810 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1811 tprintf_close(tpr); 1812 return (0); 1813 } 1814 1815 #ifdef NFSSERVER 1816 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 1817 struct nfssvc_sock *, struct proc *, 1818 struct mbuf **)) = { 1819 nfsrv_null, 1820 nfsrv_getattr, 1821 nfsrv_setattr, 1822 nfsrv_lookup, 1823 nfsrv3_access, 1824 nfsrv_readlink, 1825 nfsrv_read, 1826 nfsrv_write, 1827 nfsrv_create, 1828 nfsrv_mkdir, 1829 nfsrv_symlink, 1830 nfsrv_mknod, 1831 nfsrv_remove, 1832 nfsrv_rmdir, 1833 nfsrv_rename, 1834 nfsrv_link, 1835 nfsrv_readdir, 1836 nfsrv_readdirplus, 1837 nfsrv_statfs, 1838 nfsrv_fsinfo, 1839 nfsrv_pathconf, 1840 nfsrv_commit, 1841 nqnfsrv_getlease, 1842 nqnfsrv_vacated, 1843 nfsrv_noop, 1844 nfsrv_noop 1845 }; 1846 1847 /* 1848 * Socket upcall routine for the nfsd sockets. 1849 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1850 * Essentially do as much as possible non-blocking, else punt and it will 1851 * be called with M_WAIT from an nfsd. 1852 */ 1853 void 1854 nfsrv_rcv(so, arg, waitflag) 1855 struct socket *so; 1856 caddr_t arg; 1857 int waitflag; 1858 { 1859 register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1860 register struct mbuf *m; 1861 struct mbuf *mp, *nam; 1862 struct uio auio; 1863 int flags, error; 1864 1865 if ((slp->ns_flag & SLP_VALID) == 0) 1866 return; 1867 #ifdef notdef 1868 /* 1869 * Define this to test for nfsds handling this under heavy load. 1870 */ 1871 if (waitflag == M_DONTWAIT) { 1872 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1873 } 1874 #endif 1875 auio.uio_procp = NULL; 1876 if (so->so_type == SOCK_STREAM) { 1877 /* 1878 * If there are already records on the queue, defer soreceive() 1879 * to an nfsd so that there is feedback to the TCP layer that 1880 * the nfs servers are heavily loaded. 1881 */ 1882 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1883 slp->ns_flag |= SLP_NEEDQ; 1884 goto dorecs; 1885 } 1886 1887 /* 1888 * Do soreceive(). 1889 */ 1890 auio.uio_resid = 1000000000; 1891 flags = MSG_DONTWAIT; 1892 error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1893 if (error || mp == (struct mbuf *)0) { 1894 if (error == EWOULDBLOCK) 1895 slp->ns_flag |= SLP_NEEDQ; 1896 else 1897 slp->ns_flag |= SLP_DISCONN; 1898 goto dorecs; 1899 } 1900 m = mp; 1901 if (slp->ns_rawend) { 1902 slp->ns_rawend->m_next = m; 1903 slp->ns_cc += 1000000000 - auio.uio_resid; 1904 } else { 1905 slp->ns_raw = m; 1906 slp->ns_cc = 1000000000 - auio.uio_resid; 1907 } 1908 while (m->m_next) 1909 m = m->m_next; 1910 slp->ns_rawend = m; 1911 1912 /* 1913 * Now try and parse record(s) out of the raw stream data. 1914 */ 1915 error = nfsrv_getstream(slp, waitflag); 1916 if (error) { 1917 if (error == EPERM) 1918 slp->ns_flag |= SLP_DISCONN; 1919 else 1920 slp->ns_flag |= SLP_NEEDQ; 1921 } 1922 } else { 1923 do { 1924 auio.uio_resid = 1000000000; 1925 flags = MSG_DONTWAIT; 1926 error = (*so->so_receive)(so, &nam, &auio, &mp, 1927 (struct mbuf **)0, &flags); 1928 if (mp) { 1929 if (nam) { 1930 m = nam; 1931 m->m_next = mp; 1932 } else 1933 m = mp; 1934 if (slp->ns_recend) 1935 slp->ns_recend->m_nextpkt = m; 1936 else 1937 slp->ns_rec = m; 1938 slp->ns_recend = m; 1939 m->m_nextpkt = (struct mbuf *)0; 1940 } 1941 if (error) { 1942 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1943 && error != EWOULDBLOCK) { 1944 slp->ns_flag |= SLP_DISCONN; 1945 goto dorecs; 1946 } 1947 } 1948 } while (mp); 1949 } 1950 1951 /* 1952 * Now try and process the request records, non-blocking. 1953 */ 1954 dorecs: 1955 if (waitflag == M_DONTWAIT && 1956 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1957 nfsrv_wakenfsd(slp); 1958 } 1959 1960 /* 1961 * Try and extract an RPC request from the mbuf data list received on a 1962 * stream socket. The "waitflag" argument indicates whether or not it 1963 * can sleep. 1964 */ 1965 int 1966 nfsrv_getstream(slp, waitflag) 1967 register struct nfssvc_sock *slp; 1968 int waitflag; 1969 { 1970 register struct mbuf *m, **mpp; 1971 register char *cp1, *cp2; 1972 register int len; 1973 struct mbuf *om, *m2, *recm = NULL; 1974 u_int32_t recmark; 1975 1976 if (slp->ns_flag & SLP_GETSTREAM) 1977 panic("nfs getstream"); 1978 slp->ns_flag |= SLP_GETSTREAM; 1979 for (;;) { 1980 if (slp->ns_reclen == 0) { 1981 if (slp->ns_cc < NFSX_UNSIGNED) { 1982 slp->ns_flag &= ~SLP_GETSTREAM; 1983 return (0); 1984 } 1985 m = slp->ns_raw; 1986 if (m->m_len >= NFSX_UNSIGNED) { 1987 memcpy((caddr_t)&recmark, mtod(m, caddr_t), NFSX_UNSIGNED); 1988 m->m_data += NFSX_UNSIGNED; 1989 m->m_len -= NFSX_UNSIGNED; 1990 } else { 1991 cp1 = (caddr_t)&recmark; 1992 cp2 = mtod(m, caddr_t); 1993 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1994 while (m->m_len == 0) { 1995 m = m->m_next; 1996 cp2 = mtod(m, caddr_t); 1997 } 1998 *cp1++ = *cp2++; 1999 m->m_data++; 2000 m->m_len--; 2001 } 2002 } 2003 slp->ns_cc -= NFSX_UNSIGNED; 2004 recmark = ntohl(recmark); 2005 slp->ns_reclen = recmark & ~0x80000000; 2006 if (recmark & 0x80000000) 2007 slp->ns_flag |= SLP_LASTFRAG; 2008 else 2009 slp->ns_flag &= ~SLP_LASTFRAG; 2010 if (slp->ns_reclen > NFS_MAXPACKET) { 2011 slp->ns_flag &= ~SLP_GETSTREAM; 2012 return (EPERM); 2013 } 2014 } 2015 2016 /* 2017 * Now get the record part. 2018 */ 2019 if (slp->ns_cc == slp->ns_reclen) { 2020 recm = slp->ns_raw; 2021 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2022 slp->ns_cc = slp->ns_reclen = 0; 2023 } else if (slp->ns_cc > slp->ns_reclen) { 2024 len = 0; 2025 m = slp->ns_raw; 2026 om = (struct mbuf *)0; 2027 while (len < slp->ns_reclen) { 2028 if ((len + m->m_len) > slp->ns_reclen) { 2029 size_t left = slp->ns_reclen - len; 2030 2031 MGETHDR(m2, waitflag, m->m_type); 2032 if (m2 == NULL) { 2033 slp->ns_flag &= ~SLP_GETSTREAM; 2034 return (EWOULDBLOCK); 2035 } 2036 if (left > MHLEN) { 2037 MCLGET(m2, waitflag); 2038 if (!(m2->m_flags & M_EXT)) { 2039 m_freem(m2); 2040 slp->ns_flag &= ~SLP_GETSTREAM; 2041 return (EWOULDBLOCK); 2042 } 2043 } 2044 memcpy(mtod(m2, caddr_t), mtod(m, caddr_t), 2045 left); 2046 m2->m_len = left; 2047 m->m_data += left; 2048 m->m_len -= left; 2049 if (om) { 2050 om->m_next = m2; 2051 recm = slp->ns_raw; 2052 } else 2053 recm = m2; 2054 len = slp->ns_reclen; 2055 } else if ((len + m->m_len) == slp->ns_reclen) { 2056 om = m; 2057 len += m->m_len; 2058 m = m->m_next; 2059 recm = slp->ns_raw; 2060 om->m_next = (struct mbuf *)0; 2061 } else { 2062 om = m; 2063 len += m->m_len; 2064 m = m->m_next; 2065 } 2066 } 2067 slp->ns_raw = m; 2068 slp->ns_cc -= len; 2069 slp->ns_reclen = 0; 2070 } else { 2071 slp->ns_flag &= ~SLP_GETSTREAM; 2072 return (0); 2073 } 2074 2075 /* 2076 * Accumulate the fragments into a record. 2077 */ 2078 mpp = &slp->ns_frag; 2079 while (*mpp) 2080 mpp = &((*mpp)->m_next); 2081 *mpp = recm; 2082 if (slp->ns_flag & SLP_LASTFRAG) { 2083 if (slp->ns_recend) 2084 slp->ns_recend->m_nextpkt = slp->ns_frag; 2085 else 2086 slp->ns_rec = slp->ns_frag; 2087 slp->ns_recend = slp->ns_frag; 2088 slp->ns_frag = (struct mbuf *)0; 2089 } 2090 } 2091 } 2092 2093 /* 2094 * Parse an RPC header. 2095 */ 2096 int 2097 nfsrv_dorec(slp, nfsd, ndp) 2098 register struct nfssvc_sock *slp; 2099 struct nfsd *nfsd; 2100 struct nfsrv_descript **ndp; 2101 { 2102 register struct mbuf *m, *nam; 2103 register struct nfsrv_descript *nd; 2104 int error; 2105 2106 *ndp = NULL; 2107 if ((slp->ns_flag & SLP_VALID) == 0 || 2108 (m = slp->ns_rec) == (struct mbuf *)0) 2109 return (ENOBUFS); 2110 slp->ns_rec = m->m_nextpkt; 2111 if (slp->ns_rec) 2112 m->m_nextpkt = (struct mbuf *)0; 2113 else 2114 slp->ns_recend = (struct mbuf *)0; 2115 if (m->m_type == MT_SONAME) { 2116 nam = m; 2117 m = m->m_next; 2118 nam->m_next = NULL; 2119 } else 2120 nam = NULL; 2121 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), 2122 M_NFSRVDESC, M_WAITOK); 2123 nd->nd_md = nd->nd_mrep = m; 2124 nd->nd_nam2 = nam; 2125 nd->nd_dpos = mtod(m, caddr_t); 2126 error = nfs_getreq(nd, nfsd, TRUE); 2127 if (error) { 2128 m_freem(nam); 2129 free((caddr_t)nd, M_NFSRVDESC); 2130 return (error); 2131 } 2132 *ndp = nd; 2133 nfsd->nfsd_nd = nd; 2134 return (0); 2135 } 2136 2137 2138 /* 2139 * Search for a sleeping nfsd and wake it up. 2140 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2141 * running nfsds will go look for the work in the nfssvc_sock list. 2142 */ 2143 void 2144 nfsrv_wakenfsd(slp) 2145 struct nfssvc_sock *slp; 2146 { 2147 register struct nfsd *nd; 2148 2149 if ((slp->ns_flag & SLP_VALID) == 0) 2150 return; 2151 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) { 2152 if (nd->nfsd_flag & NFSD_WAITING) { 2153 nd->nfsd_flag &= ~NFSD_WAITING; 2154 if (nd->nfsd_slp) 2155 panic("nfsd wakeup"); 2156 slp->ns_sref++; 2157 nd->nfsd_slp = slp; 2158 wakeup((caddr_t)nd); 2159 return; 2160 } 2161 } 2162 slp->ns_flag |= SLP_DOREC; 2163 nfsd_head_flag |= NFSD_CHECKSLP; 2164 } 2165 #endif /* NFSSERVER */ 2166
Indexes created Mon Sep 22 13:09:51 GMT 2025