1 /* $NetBSD: nfs_socket.c,v 1.92 2003/06/29 22:32:17 fvdl Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1991, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 39 */ 40 41 /* 42 * Socket operations for use by nfs 43 */ 44 45 #include <sys/cdefs.h> 46 __KERNEL_RCSID(0, "$NetBSD: nfs_socket.c,v 1.92 2003/06/29 22:32:17 fvdl Exp $"); 47 48 #include "fs_nfs.h" 49 #include "opt_nfs.h" 50 #include "opt_nfsserver.h" 51 #include "opt_mbuftrace.h" 52 #include "opt_inet.h" 53 54 #include <sys/param.h> 55 #include <sys/systm.h> 56 #include <sys/callout.h> 57 #include <sys/proc.h> 58 #include <sys/mount.h> 59 #include <sys/kernel.h> 60 #include <sys/mbuf.h> 61 #include <sys/vnode.h> 62 #include <sys/domain.h> 63 #include <sys/protosw.h> 64 #include <sys/socket.h> 65 #include <sys/socketvar.h> 66 #include <sys/syslog.h> 67 #include <sys/tprintf.h> 68 #include <sys/namei.h> 69 #include <sys/signal.h> 70 #include <sys/signalvar.h> 71 72 #include <netinet/in.h> 73 #include <netinet/tcp.h> 74 75 #include <nfs/rpcv2.h> 76 #include <nfs/nfsproto.h> 77 #include <nfs/nfs.h> 78 #include <nfs/xdr_subs.h> 79 #include <nfs/nfsm_subs.h> 80 #include <nfs/nfsmount.h> 81 #include <nfs/nfsnode.h> 82 #include <nfs/nfsrtt.h> 83 #include <nfs/nqnfs.h> 84 #include <nfs/nfs_var.h> 85 86 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header"); 87 #ifdef MBUFTRACE 88 struct mowner nfs_mowner = { "nfs" }; 89 #endif 90 91 /* 92 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 93 * Use the mean and mean deviation of rtt for the appropriate type of rpc 94 * for the frequent rpcs and a default for the others. 95 * The justification for doing "other" this way is that these rpcs 96 * happen so infrequently that timer est. would probably be stale. 97 * Also, since many of these rpcs are 98 * non-idempotent, a conservative timeout is desired. 99 * getattr, lookup - A+2D 100 * read, write - A+4D 101 * other - nm_timeo 102 */ 103 #define NFS_RTO(n, t) \ 104 ((t) == 0 ? (n)->nm_timeo : \ 105 ((t) < 3 ? \ 106 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 107 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 108 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 109 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 110 /* 111 * External data, mostly RPC constants in XDR form 112 */ 113 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 114 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, 115 rpc_auth_kerb; 116 extern u_int32_t nfs_prog, nqnfs_prog; 117 extern time_t nqnfsstarttime; 118 extern const int nfsv3_procid[NFS_NPROCS]; 119 extern int nfs_ticks; 120 121 /* 122 * Defines which timer to use for the procnum. 123 * 0 - default 124 * 1 - getattr 125 * 2 - lookup 126 * 3 - read 127 * 4 - write 128 */ 129 static const int proct[NFS_NPROCS] = { 130 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0, 131 0, 0, 0, 132 }; 133 134 /* 135 * There is a congestion window for outstanding rpcs maintained per mount 136 * point. The cwnd size is adjusted in roughly the way that: 137 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 138 * SIGCOMM '88". ACM, August 1988. 139 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 140 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 141 * of rpcs is in progress. 142 * (The sent count and cwnd are scaled for integer arith.) 143 * Variants of "slow start" were tried and were found to be too much of a 144 * performance hit (ave. rtt 3 times larger), 145 * I suspect due to the large rtt that nfs rpcs have. 146 */ 147 #define NFS_CWNDSCALE 256 148 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 149 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 150 int nfsrtton = 0; 151 struct nfsrtt nfsrtt; 152 struct nfsreqhead nfs_reqq; 153 154 struct callout nfs_timer_ch = CALLOUT_INITIALIZER; 155 156 /* 157 * Initialize sockets and congestion for a new NFS connection. 158 * We do not free the sockaddr if error. 159 */ 160 int 161 nfs_connect(nmp, rep) 162 struct nfsmount *nmp; 163 struct nfsreq *rep; 164 { 165 struct socket *so; 166 int s, error, rcvreserve, sndreserve; 167 struct sockaddr *saddr; 168 struct sockaddr_in *sin; 169 #ifdef INET6 170 struct sockaddr_in6 *sin6; 171 #endif 172 struct mbuf *m; 173 174 nmp->nm_so = (struct socket *)0; 175 saddr = mtod(nmp->nm_nam, struct sockaddr *); 176 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 177 nmp->nm_soproto); 178 if (error) 179 goto bad; 180 so = nmp->nm_so; 181 #ifdef MBUFTRACE 182 so->so_mowner = &nfs_mowner; 183 so->so_rcv.sb_mowner = &nfs_mowner; 184 so->so_snd.sb_mowner = &nfs_mowner; 185 #endif 186 nmp->nm_soflags = so->so_proto->pr_flags; 187 188 /* 189 * Some servers require that the client port be a reserved port number. 190 */ 191 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 192 m = m_get(M_WAIT, MT_SOOPTS); 193 MCLAIM(m, so->so_mowner); 194 *mtod(m, int32_t *) = IP_PORTRANGE_LOW; 195 m->m_len = sizeof(int32_t); 196 if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m))) 197 goto bad; 198 m = m_get(M_WAIT, MT_SONAME); 199 MCLAIM(m, so->so_mowner); 200 sin = mtod(m, struct sockaddr_in *); 201 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 202 sin->sin_family = AF_INET; 203 sin->sin_addr.s_addr = INADDR_ANY; 204 sin->sin_port = 0; 205 error = sobind(so, m, &proc0); 206 m_freem(m); 207 if (error) 208 goto bad; 209 } 210 #ifdef INET6 211 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) { 212 m = m_get(M_WAIT, MT_SOOPTS); 213 MCLAIM(m, so->so_mowner); 214 *mtod(m, int32_t *) = IPV6_PORTRANGE_LOW; 215 m->m_len = sizeof(int32_t); 216 if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m))) 217 goto bad; 218 m = m_get(M_WAIT, MT_SONAME); 219 MCLAIM(m, so->so_mowner); 220 sin6 = mtod(m, struct sockaddr_in6 *); 221 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6); 222 sin6->sin6_family = AF_INET6; 223 sin6->sin6_addr = in6addr_any; 224 sin6->sin6_port = 0; 225 error = sobind(so, m, &proc0); 226 m_freem(m); 227 if (error) 228 goto bad; 229 } 230 #endif 231 232 /* 233 * Protocols that do not require connections may be optionally left 234 * unconnected for servers that reply from a port other than NFS_PORT. 235 */ 236 if (nmp->nm_flag & NFSMNT_NOCONN) { 237 if (nmp->nm_soflags & PR_CONNREQUIRED) { 238 error = ENOTCONN; 239 goto bad; 240 } 241 } else { 242 error = soconnect(so, nmp->nm_nam); 243 if (error) 244 goto bad; 245 246 /* 247 * Wait for the connection to complete. Cribbed from the 248 * connect system call but with the wait timing out so 249 * that interruptible mounts don't hang here for a long time. 250 */ 251 s = splsoftnet(); 252 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 253 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 254 "nfscn1", 2 * hz); 255 if ((so->so_state & SS_ISCONNECTING) && 256 so->so_error == 0 && rep && 257 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ 258 so->so_state &= ~SS_ISCONNECTING; 259 splx(s); 260 goto bad; 261 } 262 } 263 if (so->so_error) { 264 error = so->so_error; 265 so->so_error = 0; 266 splx(s); 267 goto bad; 268 } 269 splx(s); 270 } 271 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 272 so->so_rcv.sb_timeo = (5 * hz); 273 so->so_snd.sb_timeo = (5 * hz); 274 } else { 275 so->so_rcv.sb_timeo = 0; 276 so->so_snd.sb_timeo = 0; 277 } 278 if (nmp->nm_sotype == SOCK_DGRAM) { 279 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 280 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 281 NFS_MAXPKTHDR) * 2; 282 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 283 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 284 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 285 NFS_MAXPKTHDR) * 2; 286 } else { 287 if (nmp->nm_sotype != SOCK_STREAM) 288 panic("nfscon sotype"); 289 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 290 m = m_get(M_WAIT, MT_SOOPTS); 291 MCLAIM(m, so->so_mowner); 292 *mtod(m, int32_t *) = 1; 293 m->m_len = sizeof(int32_t); 294 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 295 } 296 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 297 m = m_get(M_WAIT, MT_SOOPTS); 298 MCLAIM(m, so->so_mowner); 299 *mtod(m, int32_t *) = 1; 300 m->m_len = sizeof(int32_t); 301 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 302 } 303 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 304 sizeof (u_int32_t)) * 2; 305 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 306 sizeof (u_int32_t)) * 2; 307 } 308 error = soreserve(so, sndreserve, rcvreserve); 309 if (error) 310 goto bad; 311 so->so_rcv.sb_flags |= SB_NOINTR; 312 so->so_snd.sb_flags |= SB_NOINTR; 313 314 /* Initialize other non-zero congestion variables */ 315 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 316 NFS_TIMEO << 3; 317 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 318 nmp->nm_sdrtt[3] = 0; 319 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 320 nmp->nm_sent = 0; 321 nmp->nm_timeouts = 0; 322 return (0); 323 324 bad: 325 nfs_disconnect(nmp); 326 return (error); 327 } 328 329 /* 330 * Reconnect routine: 331 * Called when a connection is broken on a reliable protocol. 332 * - clean up the old socket 333 * - nfs_connect() again 334 * - set R_MUSTRESEND for all outstanding requests on mount point 335 * If this fails the mount point is DEAD! 336 * nb: Must be called with the nfs_sndlock() set on the mount point. 337 */ 338 int 339 nfs_reconnect(rep) 340 struct nfsreq *rep; 341 { 342 struct nfsreq *rp; 343 struct nfsmount *nmp = rep->r_nmp; 344 int error; 345 346 nfs_disconnect(nmp); 347 while ((error = nfs_connect(nmp, rep)) != 0) { 348 if (error == EINTR || error == ERESTART) 349 return (EINTR); 350 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0); 351 } 352 353 /* 354 * Loop through outstanding request list and fix up all requests 355 * on old socket. 356 */ 357 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 358 if (rp->r_nmp == nmp) 359 rp->r_flags |= R_MUSTRESEND; 360 } 361 return (0); 362 } 363 364 /* 365 * NFS disconnect. Clean up and unlink. 366 */ 367 void 368 nfs_disconnect(nmp) 369 struct nfsmount *nmp; 370 { 371 struct socket *so; 372 int drain = 0; 373 374 if (nmp->nm_so) { 375 so = nmp->nm_so; 376 nmp->nm_so = (struct socket *)0; 377 soshutdown(so, 2); 378 drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0; 379 if (drain) { 380 /* 381 * soshutdown() above should wake up the current 382 * listener. 383 * Now wake up those waiting for the receive lock, and 384 * wait for them to go away unhappy, to prevent *nmp 385 * from evaporating while they're sleeping. 386 */ 387 while (nmp->nm_waiters > 0) { 388 wakeup (&nmp->nm_iflag); 389 (void) tsleep(&nmp->nm_waiters, PVFS, 390 "nfsdis", 0); 391 } 392 } 393 soclose(so); 394 } 395 #ifdef DIAGNOSTIC 396 if (drain && (nmp->nm_waiters > 0)) 397 panic("nfs_disconnect: waiters left after drain?"); 398 #endif 399 } 400 401 void 402 nfs_safedisconnect(nmp) 403 struct nfsmount *nmp; 404 { 405 struct nfsreq dummyreq; 406 407 memset(&dummyreq, 0, sizeof(dummyreq)); 408 dummyreq.r_nmp = nmp; 409 nfs_rcvlock(&dummyreq); /* XXX ignored error return */ 410 nfs_disconnect(nmp); 411 nfs_rcvunlock(nmp); 412 } 413 414 /* 415 * This is the nfs send routine. For connection based socket types, it 416 * must be called with an nfs_sndlock() on the socket. 417 * "rep == NULL" indicates that it has been called from a server. 418 * For the client side: 419 * - return EINTR if the RPC is terminated, 0 otherwise 420 * - set R_MUSTRESEND if the send fails for any reason 421 * - do any cleanup required by recoverable socket errors (? ? ?) 422 * For the server side: 423 * - return EINTR or ERESTART if interrupted by a signal 424 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 425 * - do any cleanup required by recoverable socket errors (? ? ?) 426 */ 427 int 428 nfs_send(so, nam, top, rep) 429 struct socket *so; 430 struct mbuf *nam; 431 struct mbuf *top; 432 struct nfsreq *rep; 433 { 434 struct mbuf *sendnam; 435 int error, soflags, flags; 436 437 if (rep) { 438 if (rep->r_flags & R_SOFTTERM) { 439 m_freem(top); 440 return (EINTR); 441 } 442 if ((so = rep->r_nmp->nm_so) == NULL) { 443 rep->r_flags |= R_MUSTRESEND; 444 m_freem(top); 445 return (0); 446 } 447 rep->r_flags &= ~R_MUSTRESEND; 448 soflags = rep->r_nmp->nm_soflags; 449 } else 450 soflags = so->so_proto->pr_flags; 451 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 452 sendnam = (struct mbuf *)0; 453 else 454 sendnam = nam; 455 if (so->so_type == SOCK_SEQPACKET) 456 flags = MSG_EOR; 457 else 458 flags = 0; 459 460 error = (*so->so_send)(so, sendnam, (struct uio *)0, top, 461 (struct mbuf *)0, flags); 462 if (error) { 463 if (rep) { 464 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 465 /* 466 * We're too fast for the network/driver, 467 * and UDP isn't flowcontrolled. 468 * We need to resend. This is not fatal, 469 * just try again. 470 * 471 * Could be smarter here by doing some sort 472 * of a backoff, but this is rare. 473 */ 474 rep->r_flags |= R_MUSTRESEND; 475 } else { 476 log(LOG_INFO, "nfs send error %d for %s\n", 477 error, 478 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 479 /* 480 * Deal with errors for the client side. 481 */ 482 if (rep->r_flags & R_SOFTTERM) 483 error = EINTR; 484 else 485 rep->r_flags |= R_MUSTRESEND; 486 } 487 } else { 488 /* 489 * See above. This error can happen under normal 490 * circumstances and the log is too noisy. 491 * The error will still show up in nfsstat. 492 */ 493 if (error != ENOBUFS || so->so_type != SOCK_DGRAM) 494 log(LOG_INFO, "nfsd send error %d\n", error); 495 } 496 497 /* 498 * Handle any recoverable (soft) socket errors here. (? ? ?) 499 */ 500 if (error != EINTR && error != ERESTART && 501 error != EWOULDBLOCK && error != EPIPE) 502 error = 0; 503 } 504 return (error); 505 } 506 507 #ifdef NFS 508 /* 509 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 510 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 511 * Mark and consolidate the data into a new mbuf list. 512 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 513 * small mbufs. 514 * For SOCK_STREAM we must be very careful to read an entire record once 515 * we have read any of it, even if the system call has been interrupted. 516 */ 517 int 518 nfs_receive(rep, aname, mp) 519 struct nfsreq *rep; 520 struct mbuf **aname; 521 struct mbuf **mp; 522 { 523 struct socket *so; 524 struct uio auio; 525 struct iovec aio; 526 struct mbuf *m; 527 struct mbuf *control; 528 u_int32_t len; 529 struct mbuf **getnam; 530 int error, sotype, rcvflg; 531 struct proc *p = curproc; /* XXX */ 532 533 /* 534 * Set up arguments for soreceive() 535 */ 536 *mp = (struct mbuf *)0; 537 *aname = (struct mbuf *)0; 538 sotype = rep->r_nmp->nm_sotype; 539 540 /* 541 * For reliable protocols, lock against other senders/receivers 542 * in case a reconnect is necessary. 543 * For SOCK_STREAM, first get the Record Mark to find out how much 544 * more there is to get. 545 * We must lock the socket against other receivers 546 * until we have an entire rpc request/reply. 547 */ 548 if (sotype != SOCK_DGRAM) { 549 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 550 if (error) 551 return (error); 552 tryagain: 553 /* 554 * Check for fatal errors and resending request. 555 */ 556 /* 557 * Ugh: If a reconnect attempt just happened, nm_so 558 * would have changed. NULL indicates a failed 559 * attempt that has essentially shut down this 560 * mount point. 561 */ 562 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 563 nfs_sndunlock(&rep->r_nmp->nm_iflag); 564 return (EINTR); 565 } 566 so = rep->r_nmp->nm_so; 567 if (!so) { 568 error = nfs_reconnect(rep); 569 if (error) { 570 nfs_sndunlock(&rep->r_nmp->nm_iflag); 571 return (error); 572 } 573 goto tryagain; 574 } 575 while (rep->r_flags & R_MUSTRESEND) { 576 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 577 nfsstats.rpcretries++; 578 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 579 if (error) { 580 if (error == EINTR || error == ERESTART || 581 (error = nfs_reconnect(rep)) != 0) { 582 nfs_sndunlock(&rep->r_nmp->nm_iflag); 583 return (error); 584 } 585 goto tryagain; 586 } 587 } 588 nfs_sndunlock(&rep->r_nmp->nm_iflag); 589 if (sotype == SOCK_STREAM) { 590 aio.iov_base = (caddr_t) &len; 591 aio.iov_len = sizeof(u_int32_t); 592 auio.uio_iov = &aio; 593 auio.uio_iovcnt = 1; 594 auio.uio_segflg = UIO_SYSSPACE; 595 auio.uio_rw = UIO_READ; 596 auio.uio_offset = 0; 597 auio.uio_resid = sizeof(u_int32_t); 598 auio.uio_procp = p; 599 do { 600 rcvflg = MSG_WAITALL; 601 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 602 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 603 if (error == EWOULDBLOCK && rep) { 604 if (rep->r_flags & R_SOFTTERM) 605 return (EINTR); 606 } 607 } while (error == EWOULDBLOCK); 608 if (!error && auio.uio_resid > 0) { 609 /* 610 * Don't log a 0 byte receive; it means 611 * that the socket has been closed, and 612 * can happen during normal operation 613 * (forcible unmount or Solaris server). 614 */ 615 if (auio.uio_resid != sizeof (u_int32_t)) 616 log(LOG_INFO, 617 "short receive (%lu/%lu) from nfs server %s\n", 618 (u_long)sizeof(u_int32_t) - auio.uio_resid, 619 (u_long)sizeof(u_int32_t), 620 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 621 error = EPIPE; 622 } 623 if (error) 624 goto errout; 625 len = ntohl(len) & ~0x80000000; 626 /* 627 * This is SERIOUS! We are out of sync with the sender 628 * and forcing a disconnect/reconnect is all I can do. 629 */ 630 if (len > NFS_MAXPACKET) { 631 log(LOG_ERR, "%s (%d) from nfs server %s\n", 632 "impossible packet length", 633 len, 634 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 635 error = EFBIG; 636 goto errout; 637 } 638 auio.uio_resid = len; 639 do { 640 rcvflg = MSG_WAITALL; 641 error = (*so->so_receive)(so, (struct mbuf **)0, 642 &auio, mp, (struct mbuf **)0, &rcvflg); 643 } while (error == EWOULDBLOCK || error == EINTR || 644 error == ERESTART); 645 if (!error && auio.uio_resid > 0) { 646 if (len != auio.uio_resid) 647 log(LOG_INFO, 648 "short receive (%lu/%d) from nfs server %s\n", 649 (u_long)len - auio.uio_resid, len, 650 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 651 error = EPIPE; 652 } 653 } else { 654 /* 655 * NB: Since uio_resid is big, MSG_WAITALL is ignored 656 * and soreceive() will return when it has either a 657 * control msg or a data msg. 658 * We have no use for control msg., but must grab them 659 * and then throw them away so we know what is going 660 * on. 661 */ 662 auio.uio_resid = len = 100000000; /* Anything Big */ 663 auio.uio_procp = p; 664 do { 665 rcvflg = 0; 666 error = (*so->so_receive)(so, (struct mbuf **)0, 667 &auio, mp, &control, &rcvflg); 668 if (control) 669 m_freem(control); 670 if (error == EWOULDBLOCK && rep) { 671 if (rep->r_flags & R_SOFTTERM) 672 return (EINTR); 673 } 674 } while (error == EWOULDBLOCK || 675 (!error && *mp == NULL && control)); 676 if ((rcvflg & MSG_EOR) == 0) 677 printf("Egad!!\n"); 678 if (!error && *mp == NULL) 679 error = EPIPE; 680 len -= auio.uio_resid; 681 } 682 errout: 683 if (error && error != EINTR && error != ERESTART) { 684 m_freem(*mp); 685 *mp = (struct mbuf *)0; 686 if (error != EPIPE) 687 log(LOG_INFO, 688 "receive error %d from nfs server %s\n", 689 error, 690 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 691 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 692 if (!error) 693 error = nfs_reconnect(rep); 694 if (!error) 695 goto tryagain; 696 else 697 nfs_sndunlock(&rep->r_nmp->nm_iflag); 698 } 699 } else { 700 if ((so = rep->r_nmp->nm_so) == NULL) 701 return (EACCES); 702 if (so->so_state & SS_ISCONNECTED) 703 getnam = (struct mbuf **)0; 704 else 705 getnam = aname; 706 auio.uio_resid = len = 1000000; 707 auio.uio_procp = p; 708 do { 709 rcvflg = 0; 710 error = (*so->so_receive)(so, getnam, &auio, mp, 711 (struct mbuf **)0, &rcvflg); 712 if (error == EWOULDBLOCK && 713 (rep->r_flags & R_SOFTTERM)) 714 return (EINTR); 715 } while (error == EWOULDBLOCK); 716 len -= auio.uio_resid; 717 if (!error && *mp == NULL) 718 error = EPIPE; 719 } 720 if (error) { 721 m_freem(*mp); 722 *mp = (struct mbuf *)0; 723 } 724 return (error); 725 } 726 727 /* 728 * Implement receipt of reply on a socket. 729 * We must search through the list of received datagrams matching them 730 * with outstanding requests using the xid, until ours is found. 731 */ 732 /* ARGSUSED */ 733 int 734 nfs_reply(myrep) 735 struct nfsreq *myrep; 736 { 737 struct nfsreq *rep; 738 struct nfsmount *nmp = myrep->r_nmp; 739 int32_t t1; 740 struct mbuf *mrep, *nam, *md; 741 u_int32_t rxid, *tl; 742 caddr_t dpos, cp2; 743 int error; 744 745 /* 746 * Loop around until we get our own reply 747 */ 748 for (;;) { 749 /* 750 * Lock against other receivers so that I don't get stuck in 751 * sbwait() after someone else has received my reply for me. 752 * Also necessary for connection based protocols to avoid 753 * race conditions during a reconnect. 754 */ 755 error = nfs_rcvlock(myrep); 756 if (error == EALREADY) 757 return (0); 758 if (error) 759 return (error); 760 /* 761 * Get the next Rpc reply off the socket 762 */ 763 nmp->nm_waiters++; 764 error = nfs_receive(myrep, &nam, &mrep); 765 nfs_rcvunlock(nmp); 766 if (error) { 767 768 if (nmp->nm_iflag & NFSMNT_DISMNT) { 769 /* 770 * Oops, we're going away now.. 771 */ 772 nmp->nm_waiters--; 773 wakeup (&nmp->nm_waiters); 774 return error; 775 } 776 nmp->nm_waiters--; 777 /* 778 * Ignore routing errors on connectionless protocols? ? 779 */ 780 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 781 nmp->nm_so->so_error = 0; 782 #ifdef DEBUG 783 printf("nfs_reply: ignoring error %d\n", error); 784 #endif 785 if (myrep->r_flags & R_GETONEREP) 786 return (0); 787 continue; 788 } 789 return (error); 790 } 791 nmp->nm_waiters--; 792 if (nam) 793 m_freem(nam); 794 795 /* 796 * Get the xid and check that it is an rpc reply 797 */ 798 md = mrep; 799 dpos = mtod(md, caddr_t); 800 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 801 rxid = *tl++; 802 if (*tl != rpc_reply) { 803 #ifndef NFS_V2_ONLY 804 if (nmp->nm_flag & NFSMNT_NQNFS) { 805 if (nqnfs_callback(nmp, mrep, md, dpos)) 806 nfsstats.rpcinvalid++; 807 } else 808 #endif 809 { 810 nfsstats.rpcinvalid++; 811 m_freem(mrep); 812 } 813 nfsmout: 814 if (myrep->r_flags & R_GETONEREP) 815 return (0); 816 continue; 817 } 818 819 /* 820 * Loop through the request list to match up the reply 821 * Iff no match, just drop the datagram 822 */ 823 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 824 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 825 /* Found it.. */ 826 rep->r_mrep = mrep; 827 rep->r_md = md; 828 rep->r_dpos = dpos; 829 if (nfsrtton) { 830 struct rttl *rt; 831 832 rt = &nfsrtt.rttl[nfsrtt.pos]; 833 rt->proc = rep->r_procnum; 834 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 835 rt->sent = nmp->nm_sent; 836 rt->cwnd = nmp->nm_cwnd; 837 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 838 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 839 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 840 rt->tstamp = time; 841 if (rep->r_flags & R_TIMING) 842 rt->rtt = rep->r_rtt; 843 else 844 rt->rtt = 1000000; 845 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 846 } 847 /* 848 * Update congestion window. 849 * Do the additive increase of 850 * one rpc/rtt. 851 */ 852 if (nmp->nm_cwnd <= nmp->nm_sent) { 853 nmp->nm_cwnd += 854 (NFS_CWNDSCALE * NFS_CWNDSCALE + 855 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 856 if (nmp->nm_cwnd > NFS_MAXCWND) 857 nmp->nm_cwnd = NFS_MAXCWND; 858 } 859 rep->r_flags &= ~R_SENT; 860 nmp->nm_sent -= NFS_CWNDSCALE; 861 /* 862 * Update rtt using a gain of 0.125 on the mean 863 * and a gain of 0.25 on the deviation. 864 */ 865 if (rep->r_flags & R_TIMING) { 866 /* 867 * Since the timer resolution of 868 * NFS_HZ is so course, it can often 869 * result in r_rtt == 0. Since 870 * r_rtt == N means that the actual 871 * rtt is between N+dt and N+2-dt ticks, 872 * add 1. 873 */ 874 t1 = rep->r_rtt + 1; 875 t1 -= (NFS_SRTT(rep) >> 3); 876 NFS_SRTT(rep) += t1; 877 if (t1 < 0) 878 t1 = -t1; 879 t1 -= (NFS_SDRTT(rep) >> 2); 880 NFS_SDRTT(rep) += t1; 881 } 882 nmp->nm_timeouts = 0; 883 break; 884 } 885 } 886 /* 887 * If not matched to a request, drop it. 888 * If it's mine, get out. 889 */ 890 if (rep == 0) { 891 nfsstats.rpcunexpected++; 892 m_freem(mrep); 893 } else if (rep == myrep) { 894 if (rep->r_mrep == NULL) 895 panic("nfsreply nil"); 896 return (0); 897 } 898 if (myrep->r_flags & R_GETONEREP) 899 return (0); 900 } 901 } 902 903 /* 904 * nfs_request - goes something like this 905 * - fill in request struct 906 * - links it into list 907 * - calls nfs_send() for first transmit 908 * - calls nfs_receive() to get reply 909 * - break down rpc header and return with nfs reply pointed to 910 * by mrep or error 911 * nb: always frees up mreq mbuf list 912 */ 913 int 914 nfs_request(np, mrest, procnum, procp, cred, mrp, mdp, dposp) 915 struct nfsnode *np; 916 struct mbuf *mrest; 917 int procnum; 918 struct proc *procp; 919 struct ucred *cred; 920 struct mbuf **mrp; 921 struct mbuf **mdp; 922 caddr_t *dposp; 923 { 924 struct mbuf *m, *mrep; 925 struct nfsreq *rep; 926 u_int32_t *tl; 927 int i; 928 struct nfsmount *nmp; 929 struct mbuf *md, *mheadend; 930 char nickv[RPCX_NICKVERF]; 931 time_t reqtime, waituntil; 932 caddr_t dpos, cp2; 933 int t1, s, error = 0, mrest_len, auth_len, auth_type; 934 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 935 int verf_len, verf_type; 936 u_int32_t xid; 937 char *auth_str, *verf_str; 938 NFSKERBKEY_T key; /* save session key */ 939 #ifndef NFS_V2_ONLY 940 int nqlflag, cachable; 941 u_quad_t frev; 942 #endif 943 944 KASSERT(cred != NULL); 945 nmp = VFSTONFS(np->n_vnode->v_mount); 946 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 947 rep->r_nmp = nmp; 948 rep->r_procp = procp; 949 rep->r_procnum = procnum; 950 i = 0; 951 m = mrest; 952 while (m) { 953 i += m->m_len; 954 m = m->m_next; 955 } 956 mrest_len = i; 957 958 /* 959 * Get the RPC header with authorization. 960 */ 961 kerbauth: 962 verf_str = auth_str = (char *)0; 963 if (nmp->nm_flag & NFSMNT_KERB) { 964 verf_str = nickv; 965 verf_len = sizeof (nickv); 966 auth_type = RPCAUTH_KERB4; 967 memset((caddr_t)key, 0, sizeof (key)); 968 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 969 &auth_len, verf_str, verf_len)) { 970 error = nfs_getauth(nmp, rep, cred, &auth_str, 971 &auth_len, verf_str, &verf_len, key); 972 if (error) { 973 free((caddr_t)rep, M_NFSREQ); 974 m_freem(mrest); 975 return (error); 976 } 977 } 978 } else { 979 auth_type = RPCAUTH_UNIX; 980 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 981 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 982 5 * NFSX_UNSIGNED; 983 } 984 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 985 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 986 if (auth_str) 987 free(auth_str, M_TEMP); 988 989 /* 990 * For stream protocols, insert a Sun RPC Record Mark. 991 */ 992 if (nmp->nm_sotype == SOCK_STREAM) { 993 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 994 *mtod(m, u_int32_t *) = htonl(0x80000000 | 995 (m->m_pkthdr.len - NFSX_UNSIGNED)); 996 } 997 rep->r_mreq = m; 998 rep->r_xid = xid; 999 tryagain: 1000 if (nmp->nm_flag & NFSMNT_SOFT) 1001 rep->r_retry = nmp->nm_retry; 1002 else 1003 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1004 rep->r_rtt = rep->r_rexmit = 0; 1005 if (proct[procnum] > 0) 1006 rep->r_flags = R_TIMING; 1007 else 1008 rep->r_flags = 0; 1009 rep->r_mrep = NULL; 1010 1011 /* 1012 * Do the client side RPC. 1013 */ 1014 nfsstats.rpcrequests++; 1015 /* 1016 * Chain request into list of outstanding requests. Be sure 1017 * to put it LAST so timer finds oldest requests first. 1018 */ 1019 s = splsoftnet(); 1020 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1021 1022 /* Get send time for nqnfs */ 1023 reqtime = time.tv_sec; 1024 1025 /* 1026 * If backing off another request or avoiding congestion, don't 1027 * send this one now but let timer do it. If not timing a request, 1028 * do it now. 1029 */ 1030 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1031 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1032 nmp->nm_sent < nmp->nm_cwnd)) { 1033 splx(s); 1034 if (nmp->nm_soflags & PR_CONNREQUIRED) 1035 error = nfs_sndlock(&nmp->nm_iflag, rep); 1036 if (!error) { 1037 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1038 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 1039 if (nmp->nm_soflags & PR_CONNREQUIRED) 1040 nfs_sndunlock(&nmp->nm_iflag); 1041 } 1042 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1043 nmp->nm_sent += NFS_CWNDSCALE; 1044 rep->r_flags |= R_SENT; 1045 } 1046 } else { 1047 splx(s); 1048 rep->r_rtt = -1; 1049 } 1050 1051 /* 1052 * Wait for the reply from our send or the timer's. 1053 */ 1054 if (!error || error == EPIPE) 1055 error = nfs_reply(rep); 1056 1057 /* 1058 * RPC done, unlink the request. 1059 */ 1060 s = splsoftnet(); 1061 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1062 splx(s); 1063 1064 /* 1065 * Decrement the outstanding request count. 1066 */ 1067 if (rep->r_flags & R_SENT) { 1068 rep->r_flags &= ~R_SENT; /* paranoia */ 1069 nmp->nm_sent -= NFS_CWNDSCALE; 1070 } 1071 1072 /* 1073 * If there was a successful reply and a tprintf msg. 1074 * tprintf a response. 1075 */ 1076 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1077 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1078 "is alive again"); 1079 mrep = rep->r_mrep; 1080 md = rep->r_md; 1081 dpos = rep->r_dpos; 1082 if (error) { 1083 m_freem(rep->r_mreq); 1084 free((caddr_t)rep, M_NFSREQ); 1085 return (error); 1086 } 1087 1088 /* 1089 * break down the rpc header and check if ok 1090 */ 1091 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1092 if (*tl++ == rpc_msgdenied) { 1093 if (*tl == rpc_mismatch) 1094 error = EOPNOTSUPP; 1095 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1096 if (!failed_auth) { 1097 failed_auth++; 1098 mheadend->m_next = (struct mbuf *)0; 1099 m_freem(mrep); 1100 m_freem(rep->r_mreq); 1101 goto kerbauth; 1102 } else 1103 error = EAUTH; 1104 } else 1105 error = EACCES; 1106 m_freem(mrep); 1107 m_freem(rep->r_mreq); 1108 free((caddr_t)rep, M_NFSREQ); 1109 return (error); 1110 } 1111 1112 /* 1113 * Grab any Kerberos verifier, otherwise just throw it away. 1114 */ 1115 verf_type = fxdr_unsigned(int, *tl++); 1116 i = fxdr_unsigned(int32_t, *tl); 1117 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1118 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1119 if (error) 1120 goto nfsmout; 1121 } else if (i > 0) 1122 nfsm_adv(nfsm_rndup(i)); 1123 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1124 /* 0 == ok */ 1125 if (*tl == 0) { 1126 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1127 if (*tl != 0) { 1128 error = fxdr_unsigned(int, *tl); 1129 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1130 error == NFSERR_TRYLATER) { 1131 m_freem(mrep); 1132 error = 0; 1133 waituntil = time.tv_sec + trylater_delay; 1134 while (time.tv_sec < waituntil) 1135 (void) tsleep((caddr_t)&lbolt, 1136 PSOCK, "nqnfstry", 0); 1137 trylater_delay *= nfs_backoff[trylater_cnt]; 1138 if (trylater_cnt < 7) 1139 trylater_cnt++; 1140 goto tryagain; 1141 } 1142 1143 /* 1144 * If the File Handle was stale, invalidate the 1145 * lookup cache, just in case. 1146 */ 1147 if (error == ESTALE) 1148 cache_purge(NFSTOV(np)); 1149 if (nmp->nm_flag & NFSMNT_NFSV3) { 1150 *mrp = mrep; 1151 *mdp = md; 1152 *dposp = dpos; 1153 error |= NFSERR_RETERR; 1154 } else 1155 m_freem(mrep); 1156 m_freem(rep->r_mreq); 1157 free((caddr_t)rep, M_NFSREQ); 1158 return (error); 1159 } 1160 1161 #ifndef NFS_V2_ONLY 1162 /* 1163 * For nqnfs, get any lease in reply 1164 */ 1165 if (nmp->nm_flag & NFSMNT_NQNFS) { 1166 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1167 if (*tl) { 1168 nqlflag = fxdr_unsigned(int, *tl); 1169 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1170 cachable = fxdr_unsigned(int, *tl++); 1171 reqtime += fxdr_unsigned(int, *tl++); 1172 if (reqtime > time.tv_sec) { 1173 frev = fxdr_hyper(tl); 1174 nqnfs_clientlease(nmp, np, nqlflag, 1175 cachable, reqtime, frev); 1176 } 1177 } 1178 } 1179 #endif 1180 *mrp = mrep; 1181 *mdp = md; 1182 *dposp = dpos; 1183 m_freem(rep->r_mreq); 1184 FREE((caddr_t)rep, M_NFSREQ); 1185 return (0); 1186 } 1187 m_freem(mrep); 1188 error = EPROTONOSUPPORT; 1189 nfsmout: 1190 m_freem(rep->r_mreq); 1191 free((caddr_t)rep, M_NFSREQ); 1192 return (error); 1193 } 1194 #endif /* NFS */ 1195 1196 /* 1197 * Generate the rpc reply header 1198 * siz arg. is used to decide if adding a cluster is worthwhile 1199 */ 1200 int 1201 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1202 int siz; 1203 struct nfsrv_descript *nd; 1204 struct nfssvc_sock *slp; 1205 int err; 1206 int cache; 1207 u_quad_t *frev; 1208 struct mbuf **mrq; 1209 struct mbuf **mbp; 1210 caddr_t *bposp; 1211 { 1212 u_int32_t *tl; 1213 struct mbuf *mreq; 1214 caddr_t bpos; 1215 struct mbuf *mb; 1216 1217 mreq = m_gethdr(M_WAIT, MT_DATA); 1218 MCLAIM(mreq, &nfs_mowner); 1219 mb = mreq; 1220 /* 1221 * If this is a big reply, use a cluster else 1222 * try and leave leading space for the lower level headers. 1223 */ 1224 siz += RPC_REPLYSIZ; 1225 if (siz >= max_datalen) { 1226 m_clget(mreq, M_WAIT); 1227 } else 1228 mreq->m_data += max_hdr; 1229 tl = mtod(mreq, u_int32_t *); 1230 mreq->m_len = 6 * NFSX_UNSIGNED; 1231 bpos = ((caddr_t)tl) + mreq->m_len; 1232 *tl++ = txdr_unsigned(nd->nd_retxid); 1233 *tl++ = rpc_reply; 1234 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1235 *tl++ = rpc_msgdenied; 1236 if (err & NFSERR_AUTHERR) { 1237 *tl++ = rpc_autherr; 1238 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1239 mreq->m_len -= NFSX_UNSIGNED; 1240 bpos -= NFSX_UNSIGNED; 1241 } else { 1242 *tl++ = rpc_mismatch; 1243 *tl++ = txdr_unsigned(RPC_VER2); 1244 *tl = txdr_unsigned(RPC_VER2); 1245 } 1246 } else { 1247 *tl++ = rpc_msgaccepted; 1248 1249 /* 1250 * For Kerberos authentication, we must send the nickname 1251 * verifier back, otherwise just RPCAUTH_NULL. 1252 */ 1253 if (nd->nd_flag & ND_KERBFULL) { 1254 struct nfsuid *nuidp; 1255 struct timeval ktvin, ktvout; 1256 1257 LIST_FOREACH(nuidp, NUIDHASH(slp, nd->nd_cr.cr_uid), 1258 nu_hash) { 1259 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1260 (!nd->nd_nam2 || netaddr_match( 1261 NU_NETFAM(nuidp), &nuidp->nu_haddr, 1262 nd->nd_nam2))) 1263 break; 1264 } 1265 if (nuidp) { 1266 ktvin.tv_sec = 1267 txdr_unsigned(nuidp->nu_timestamp.tv_sec 1268 - 1); 1269 ktvin.tv_usec = 1270 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1271 1272 /* 1273 * Encrypt the timestamp in ecb mode using the 1274 * session key. 1275 */ 1276 #ifdef NFSKERB 1277 XXX 1278 #endif 1279 1280 *tl++ = rpc_auth_kerb; 1281 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1282 *tl = ktvout.tv_sec; 1283 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1284 *tl++ = ktvout.tv_usec; 1285 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1286 } else { 1287 *tl++ = 0; 1288 *tl++ = 0; 1289 } 1290 } else { 1291 *tl++ = 0; 1292 *tl++ = 0; 1293 } 1294 switch (err) { 1295 case EPROGUNAVAIL: 1296 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1297 break; 1298 case EPROGMISMATCH: 1299 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1300 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1301 if (nd->nd_flag & ND_NQNFS) { 1302 *tl++ = txdr_unsigned(3); 1303 *tl = txdr_unsigned(3); 1304 } else { 1305 *tl++ = txdr_unsigned(2); 1306 *tl = txdr_unsigned(3); 1307 } 1308 break; 1309 case EPROCUNAVAIL: 1310 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1311 break; 1312 case EBADRPC: 1313 *tl = txdr_unsigned(RPC_GARBAGE); 1314 break; 1315 default: 1316 *tl = 0; 1317 if (err != NFSERR_RETVOID) { 1318 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1319 if (err) 1320 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1321 else 1322 *tl = 0; 1323 } 1324 break; 1325 }; 1326 } 1327 1328 /* 1329 * For nqnfs, piggyback lease as requested. 1330 */ 1331 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1332 if (nd->nd_flag & ND_LEASE) { 1333 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1334 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1335 *tl++ = txdr_unsigned(cache); 1336 *tl++ = txdr_unsigned(nd->nd_duration); 1337 txdr_hyper(*frev, tl); 1338 } else { 1339 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1340 *tl = 0; 1341 } 1342 } 1343 if (mrq != NULL) 1344 *mrq = mreq; 1345 *mbp = mb; 1346 *bposp = bpos; 1347 if (err != 0 && err != NFSERR_RETVOID) 1348 nfsstats.srvrpc_errs++; 1349 return (0); 1350 } 1351 1352 /* 1353 * Nfs timer routine 1354 * Scan the nfsreq list and retranmit any requests that have timed out 1355 * To avoid retransmission attempts on STREAM sockets (in the future) make 1356 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1357 */ 1358 void 1359 nfs_timer(arg) 1360 void *arg; /* never used */ 1361 { 1362 struct nfsreq *rep; 1363 struct mbuf *m; 1364 struct socket *so; 1365 struct nfsmount *nmp; 1366 int timeo; 1367 int s, error; 1368 #ifdef NFSSERVER 1369 struct nfssvc_sock *slp; 1370 static long lasttime = 0; 1371 u_quad_t cur_usec; 1372 #endif 1373 1374 s = splsoftnet(); 1375 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1376 nmp = rep->r_nmp; 1377 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1378 continue; 1379 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1380 rep->r_flags |= R_SOFTTERM; 1381 continue; 1382 } 1383 if (rep->r_rtt >= 0) { 1384 rep->r_rtt++; 1385 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1386 timeo = nmp->nm_timeo; 1387 else 1388 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1389 if (nmp->nm_timeouts > 0) 1390 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1391 if (rep->r_rtt <= timeo) 1392 continue; 1393 if (nmp->nm_timeouts < 8) 1394 nmp->nm_timeouts++; 1395 } 1396 /* 1397 * Check for server not responding 1398 */ 1399 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1400 rep->r_rexmit > nmp->nm_deadthresh) { 1401 nfs_msg(rep->r_procp, 1402 nmp->nm_mountp->mnt_stat.f_mntfromname, 1403 "not responding"); 1404 rep->r_flags |= R_TPRINTFMSG; 1405 } 1406 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1407 nfsstats.rpctimeouts++; 1408 rep->r_flags |= R_SOFTTERM; 1409 continue; 1410 } 1411 if (nmp->nm_sotype != SOCK_DGRAM) { 1412 if (++rep->r_rexmit > NFS_MAXREXMIT) 1413 rep->r_rexmit = NFS_MAXREXMIT; 1414 continue; 1415 } 1416 if ((so = nmp->nm_so) == NULL) 1417 continue; 1418 1419 /* 1420 * If there is enough space and the window allows.. 1421 * Resend it 1422 * Set r_rtt to -1 in case we fail to send it now. 1423 */ 1424 rep->r_rtt = -1; 1425 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1426 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1427 (rep->r_flags & R_SENT) || 1428 nmp->nm_sent < nmp->nm_cwnd) && 1429 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1430 if (so->so_state & SS_ISCONNECTED) 1431 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1432 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1433 else 1434 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1435 nmp->nm_nam, (struct mbuf *)0, (struct proc *)0); 1436 if (error) { 1437 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1438 #ifdef DEBUG 1439 printf("nfs_timer: ignoring error %d\n", 1440 error); 1441 #endif 1442 so->so_error = 0; 1443 } 1444 } else { 1445 /* 1446 * Iff first send, start timing 1447 * else turn timing off, backoff timer 1448 * and divide congestion window by 2. 1449 */ 1450 if (rep->r_flags & R_SENT) { 1451 rep->r_flags &= ~R_TIMING; 1452 if (++rep->r_rexmit > NFS_MAXREXMIT) 1453 rep->r_rexmit = NFS_MAXREXMIT; 1454 nmp->nm_cwnd >>= 1; 1455 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1456 nmp->nm_cwnd = NFS_CWNDSCALE; 1457 nfsstats.rpcretries++; 1458 } else { 1459 rep->r_flags |= R_SENT; 1460 nmp->nm_sent += NFS_CWNDSCALE; 1461 } 1462 rep->r_rtt = 0; 1463 } 1464 } 1465 } 1466 1467 #ifdef NFSSERVER 1468 /* 1469 * Call the nqnfs server timer once a second to handle leases. 1470 */ 1471 if (lasttime != time.tv_sec) { 1472 lasttime = time.tv_sec; 1473 nqnfs_serverd(); 1474 } 1475 1476 /* 1477 * Scan the write gathering queues for writes that need to be 1478 * completed now. 1479 */ 1480 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1481 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1482 if (LIST_FIRST(&slp->ns_tq) && 1483 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) 1484 nfsrv_wakenfsd(slp); 1485 } 1486 #endif /* NFSSERVER */ 1487 splx(s); 1488 callout_reset(&nfs_timer_ch, nfs_ticks, nfs_timer, NULL); 1489 } 1490 1491 /*ARGSUSED*/ 1492 void 1493 nfs_exit(p, v) 1494 struct proc *p; 1495 void *v; 1496 { 1497 struct nfsreq *rp; 1498 int s = splsoftnet(); 1499 1500 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 1501 if (rp->r_procp == p) 1502 TAILQ_REMOVE(&nfs_reqq, rp, r_chain); 1503 } 1504 splx(s); 1505 } 1506 1507 /* 1508 * Test for a termination condition pending on the process. 1509 * This is used for NFSMNT_INT mounts. 1510 */ 1511 int 1512 nfs_sigintr(nmp, rep, p) 1513 struct nfsmount *nmp; 1514 struct nfsreq *rep; 1515 struct proc *p; 1516 { 1517 sigset_t ss; 1518 1519 if (rep && (rep->r_flags & R_SOFTTERM)) 1520 return (EINTR); 1521 if (!(nmp->nm_flag & NFSMNT_INT)) 1522 return (0); 1523 if (p) { 1524 sigpending1(p, &ss); 1525 #if 0 1526 sigminusset(&p->p_sigctx.ps_sigignore, &ss); 1527 #endif 1528 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1529 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1530 sigismember(&ss, SIGQUIT)) 1531 return (EINTR); 1532 } 1533 return (0); 1534 } 1535 1536 /* 1537 * Lock a socket against others. 1538 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1539 * and also to avoid race conditions between the processes with nfs requests 1540 * in progress when a reconnect is necessary. 1541 */ 1542 int 1543 nfs_sndlock(flagp, rep) 1544 int *flagp; 1545 struct nfsreq *rep; 1546 { 1547 struct proc *p; 1548 int slpflag = 0, slptimeo = 0; 1549 1550 if (rep) { 1551 p = rep->r_procp; 1552 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1553 slpflag = PCATCH; 1554 } else 1555 p = (struct proc *)0; 1556 while (*flagp & NFSMNT_SNDLOCK) { 1557 if (nfs_sigintr(rep->r_nmp, rep, p)) 1558 return (EINTR); 1559 *flagp |= NFSMNT_WANTSND; 1560 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1561 slptimeo); 1562 if (slpflag == PCATCH) { 1563 slpflag = 0; 1564 slptimeo = 2 * hz; 1565 } 1566 } 1567 *flagp |= NFSMNT_SNDLOCK; 1568 return (0); 1569 } 1570 1571 /* 1572 * Unlock the stream socket for others. 1573 */ 1574 void 1575 nfs_sndunlock(flagp) 1576 int *flagp; 1577 { 1578 1579 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1580 panic("nfs sndunlock"); 1581 *flagp &= ~NFSMNT_SNDLOCK; 1582 if (*flagp & NFSMNT_WANTSND) { 1583 *flagp &= ~NFSMNT_WANTSND; 1584 wakeup((caddr_t)flagp); 1585 } 1586 } 1587 1588 int 1589 nfs_rcvlock(rep) 1590 struct nfsreq *rep; 1591 { 1592 struct nfsmount *nmp = rep->r_nmp; 1593 int *flagp = &nmp->nm_iflag; 1594 int slpflag, slptimeo = 0; 1595 int error = 0; 1596 1597 if (*flagp & NFSMNT_DISMNT) 1598 return EIO; 1599 1600 if (*flagp & NFSMNT_INT) 1601 slpflag = PCATCH; 1602 else 1603 slpflag = 0; 1604 simple_lock(&nmp->nm_slock); 1605 while (*flagp & NFSMNT_RCVLOCK) { 1606 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) { 1607 error = EINTR; 1608 goto quit; 1609 } 1610 *flagp |= NFSMNT_WANTRCV; 1611 nmp->nm_waiters++; 1612 (void) ltsleep(flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1613 slptimeo, &nmp->nm_slock); 1614 nmp->nm_waiters--; 1615 if (*flagp & NFSMNT_DISMNT) { 1616 wakeup(&nmp->nm_waiters); 1617 error = EIO; 1618 goto quit; 1619 } 1620 /* If our reply was received while we were sleeping, 1621 * then just return without taking the lock to avoid a 1622 * situation where a single iod could 'capture' the 1623 * receive lock. 1624 */ 1625 if (rep->r_mrep != NULL) { 1626 error = EALREADY; 1627 goto quit; 1628 } 1629 if (slpflag == PCATCH) { 1630 slpflag = 0; 1631 slptimeo = 2 * hz; 1632 } 1633 } 1634 *flagp |= NFSMNT_RCVLOCK; 1635 quit: 1636 simple_unlock(&nmp->nm_slock); 1637 return error; 1638 } 1639 1640 /* 1641 * Unlock the stream socket for others. 1642 */ 1643 void 1644 nfs_rcvunlock(nmp) 1645 struct nfsmount *nmp; 1646 { 1647 int *flagp = &nmp->nm_iflag; 1648 1649 simple_lock(&nmp->nm_slock); 1650 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1651 panic("nfs rcvunlock"); 1652 *flagp &= ~NFSMNT_RCVLOCK; 1653 if (*flagp & NFSMNT_WANTRCV) { 1654 *flagp &= ~NFSMNT_WANTRCV; 1655 wakeup((caddr_t)flagp); 1656 } 1657 simple_unlock(&nmp->nm_slock); 1658 } 1659 1660 /* 1661 * Parse an RPC request 1662 * - verify it 1663 * - fill in the cred struct. 1664 */ 1665 int 1666 nfs_getreq(nd, nfsd, has_header) 1667 struct nfsrv_descript *nd; 1668 struct nfsd *nfsd; 1669 int has_header; 1670 { 1671 int len, i; 1672 u_int32_t *tl; 1673 int32_t t1; 1674 struct uio uio; 1675 struct iovec iov; 1676 caddr_t dpos, cp2, cp; 1677 u_int32_t nfsvers, auth_type; 1678 uid_t nickuid; 1679 int error = 0, nqnfs = 0, ticklen; 1680 struct mbuf *mrep, *md; 1681 struct nfsuid *nuidp; 1682 struct timeval tvin, tvout; 1683 1684 mrep = nd->nd_mrep; 1685 md = nd->nd_md; 1686 dpos = nd->nd_dpos; 1687 if (has_header) { 1688 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1689 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1690 if (*tl++ != rpc_call) { 1691 m_freem(mrep); 1692 return (EBADRPC); 1693 } 1694 } else 1695 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1696 nd->nd_repstat = 0; 1697 nd->nd_flag = 0; 1698 if (*tl++ != rpc_vers) { 1699 nd->nd_repstat = ERPCMISMATCH; 1700 nd->nd_procnum = NFSPROC_NOOP; 1701 return (0); 1702 } 1703 if (*tl != nfs_prog) { 1704 if (*tl == nqnfs_prog) 1705 nqnfs++; 1706 else { 1707 nd->nd_repstat = EPROGUNAVAIL; 1708 nd->nd_procnum = NFSPROC_NOOP; 1709 return (0); 1710 } 1711 } 1712 tl++; 1713 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1714 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1715 (nfsvers != NQNFS_VER3 && nqnfs)) { 1716 nd->nd_repstat = EPROGMISMATCH; 1717 nd->nd_procnum = NFSPROC_NOOP; 1718 return (0); 1719 } 1720 if (nqnfs) 1721 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1722 else if (nfsvers == NFS_VER3) 1723 nd->nd_flag = ND_NFSV3; 1724 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1725 if (nd->nd_procnum == NFSPROC_NULL) 1726 return (0); 1727 if (nd->nd_procnum >= NFS_NPROCS || 1728 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1729 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1730 nd->nd_repstat = EPROCUNAVAIL; 1731 nd->nd_procnum = NFSPROC_NOOP; 1732 return (0); 1733 } 1734 if ((nd->nd_flag & ND_NFSV3) == 0) 1735 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1736 auth_type = *tl++; 1737 len = fxdr_unsigned(int, *tl++); 1738 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1739 m_freem(mrep); 1740 return (EBADRPC); 1741 } 1742 1743 nd->nd_flag &= ~ND_KERBAUTH; 1744 /* 1745 * Handle auth_unix or auth_kerb. 1746 */ 1747 if (auth_type == rpc_auth_unix) { 1748 len = fxdr_unsigned(int, *++tl); 1749 if (len < 0 || len > NFS_MAXNAMLEN) { 1750 m_freem(mrep); 1751 return (EBADRPC); 1752 } 1753 nfsm_adv(nfsm_rndup(len)); 1754 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1755 memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred)); 1756 nd->nd_cr.cr_ref = 1; 1757 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1758 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1759 len = fxdr_unsigned(int, *tl); 1760 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1761 m_freem(mrep); 1762 return (EBADRPC); 1763 } 1764 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1765 for (i = 0; i < len; i++) 1766 if (i < NGROUPS) 1767 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1768 else 1769 tl++; 1770 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1771 if (nd->nd_cr.cr_ngroups > 1) 1772 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1773 len = fxdr_unsigned(int, *++tl); 1774 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1775 m_freem(mrep); 1776 return (EBADRPC); 1777 } 1778 if (len > 0) 1779 nfsm_adv(nfsm_rndup(len)); 1780 } else if (auth_type == rpc_auth_kerb) { 1781 switch (fxdr_unsigned(int, *tl++)) { 1782 case RPCAKN_FULLNAME: 1783 ticklen = fxdr_unsigned(int, *tl); 1784 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1785 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1786 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1787 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1788 m_freem(mrep); 1789 return (EBADRPC); 1790 } 1791 uio.uio_offset = 0; 1792 uio.uio_iov = &iov; 1793 uio.uio_iovcnt = 1; 1794 uio.uio_segflg = UIO_SYSSPACE; 1795 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1796 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1797 nfsm_mtouio(&uio, uio.uio_resid); 1798 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1799 if (*tl++ != rpc_auth_kerb || 1800 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1801 printf("Bad kerb verifier\n"); 1802 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1803 nd->nd_procnum = NFSPROC_NOOP; 1804 return (0); 1805 } 1806 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1807 tl = (u_int32_t *)cp; 1808 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1809 printf("Not fullname kerb verifier\n"); 1810 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1811 nd->nd_procnum = NFSPROC_NOOP; 1812 return (0); 1813 } 1814 cp += NFSX_UNSIGNED; 1815 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 1816 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1817 nd->nd_flag |= ND_KERBFULL; 1818 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1819 break; 1820 case RPCAKN_NICKNAME: 1821 if (len != 2 * NFSX_UNSIGNED) { 1822 printf("Kerb nickname short\n"); 1823 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1824 nd->nd_procnum = NFSPROC_NOOP; 1825 return (0); 1826 } 1827 nickuid = fxdr_unsigned(uid_t, *tl); 1828 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1829 if (*tl++ != rpc_auth_kerb || 1830 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1831 printf("Kerb nick verifier bad\n"); 1832 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1833 nd->nd_procnum = NFSPROC_NOOP; 1834 return (0); 1835 } 1836 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1837 tvin.tv_sec = *tl++; 1838 tvin.tv_usec = *tl; 1839 1840 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), 1841 nu_hash) { 1842 if (nuidp->nu_cr.cr_uid == nickuid && 1843 (!nd->nd_nam2 || 1844 netaddr_match(NU_NETFAM(nuidp), 1845 &nuidp->nu_haddr, nd->nd_nam2))) 1846 break; 1847 } 1848 if (!nuidp) { 1849 nd->nd_repstat = 1850 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1851 nd->nd_procnum = NFSPROC_NOOP; 1852 return (0); 1853 } 1854 1855 /* 1856 * Now, decrypt the timestamp using the session key 1857 * and validate it. 1858 */ 1859 #ifdef NFSKERB 1860 XXX 1861 #endif 1862 1863 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1864 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1865 if (nuidp->nu_expire < time.tv_sec || 1866 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1867 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1868 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1869 nuidp->nu_expire = 0; 1870 nd->nd_repstat = 1871 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1872 nd->nd_procnum = NFSPROC_NOOP; 1873 return (0); 1874 } 1875 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1876 nd->nd_flag |= ND_KERBNICK; 1877 }; 1878 } else { 1879 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1880 nd->nd_procnum = NFSPROC_NOOP; 1881 return (0); 1882 } 1883 1884 /* 1885 * For nqnfs, get piggybacked lease request. 1886 */ 1887 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1888 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1889 nd->nd_flag |= fxdr_unsigned(int, *tl); 1890 if (nd->nd_flag & ND_LEASE) { 1891 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1892 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 1893 } else 1894 nd->nd_duration = NQ_MINLEASE; 1895 } else 1896 nd->nd_duration = NQ_MINLEASE; 1897 nd->nd_md = md; 1898 nd->nd_dpos = dpos; 1899 return (0); 1900 nfsmout: 1901 return (error); 1902 } 1903 1904 int 1905 nfs_msg(p, server, msg) 1906 struct proc *p; 1907 char *server, *msg; 1908 { 1909 tpr_t tpr; 1910 1911 if (p) 1912 tpr = tprintf_open(p); 1913 else 1914 tpr = NULL; 1915 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1916 tprintf_close(tpr); 1917 return (0); 1918 } 1919 1920 #ifdef NFSSERVER 1921 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 1922 struct nfssvc_sock *, struct proc *, 1923 struct mbuf **)) = { 1924 nfsrv_null, 1925 nfsrv_getattr, 1926 nfsrv_setattr, 1927 nfsrv_lookup, 1928 nfsrv3_access, 1929 nfsrv_readlink, 1930 nfsrv_read, 1931 nfsrv_write, 1932 nfsrv_create, 1933 nfsrv_mkdir, 1934 nfsrv_symlink, 1935 nfsrv_mknod, 1936 nfsrv_remove, 1937 nfsrv_rmdir, 1938 nfsrv_rename, 1939 nfsrv_link, 1940 nfsrv_readdir, 1941 nfsrv_readdirplus, 1942 nfsrv_statfs, 1943 nfsrv_fsinfo, 1944 nfsrv_pathconf, 1945 nfsrv_commit, 1946 nqnfsrv_getlease, 1947 nqnfsrv_vacated, 1948 nfsrv_noop, 1949 nfsrv_noop 1950 }; 1951 1952 /* 1953 * Socket upcall routine for the nfsd sockets. 1954 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1955 * Essentially do as much as possible non-blocking, else punt and it will 1956 * be called with M_WAIT from an nfsd. 1957 */ 1958 void 1959 nfsrv_rcv(so, arg, waitflag) 1960 struct socket *so; 1961 caddr_t arg; 1962 int waitflag; 1963 { 1964 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1965 struct mbuf *m; 1966 struct mbuf *mp, *nam; 1967 struct uio auio; 1968 int flags, error; 1969 1970 if ((slp->ns_flag & SLP_VALID) == 0) 1971 return; 1972 #ifdef notdef 1973 /* 1974 * Define this to test for nfsds handling this under heavy load. 1975 */ 1976 if (waitflag == M_DONTWAIT) { 1977 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1978 } 1979 #endif 1980 auio.uio_procp = NULL; 1981 if (so->so_type == SOCK_STREAM) { 1982 /* 1983 * If there are already records on the queue, defer soreceive() 1984 * to an nfsd so that there is feedback to the TCP layer that 1985 * the nfs servers are heavily loaded. 1986 */ 1987 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1988 slp->ns_flag |= SLP_NEEDQ; 1989 goto dorecs; 1990 } 1991 1992 /* 1993 * Do soreceive(). 1994 */ 1995 auio.uio_resid = 1000000000; 1996 flags = MSG_DONTWAIT; 1997 error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1998 if (error || mp == (struct mbuf *)0) { 1999 if (error == EWOULDBLOCK) 2000 slp->ns_flag |= SLP_NEEDQ; 2001 else 2002 slp->ns_flag |= SLP_DISCONN; 2003 goto dorecs; 2004 } 2005 m = mp; 2006 if (slp->ns_rawend) { 2007 slp->ns_rawend->m_next = m; 2008 slp->ns_cc += 1000000000 - auio.uio_resid; 2009 } else { 2010 slp->ns_raw = m; 2011 slp->ns_cc = 1000000000 - auio.uio_resid; 2012 } 2013 while (m->m_next) 2014 m = m->m_next; 2015 slp->ns_rawend = m; 2016 2017 /* 2018 * Now try and parse record(s) out of the raw stream data. 2019 */ 2020 error = nfsrv_getstream(slp, waitflag); 2021 if (error) { 2022 if (error == EPERM) 2023 slp->ns_flag |= SLP_DISCONN; 2024 else 2025 slp->ns_flag |= SLP_NEEDQ; 2026 } 2027 } else { 2028 do { 2029 auio.uio_resid = 1000000000; 2030 flags = MSG_DONTWAIT; 2031 error = (*so->so_receive)(so, &nam, &auio, &mp, 2032 (struct mbuf **)0, &flags); 2033 if (mp) { 2034 if (nam) { 2035 m = nam; 2036 m->m_next = mp; 2037 } else 2038 m = mp; 2039 if (slp->ns_recend) 2040 slp->ns_recend->m_nextpkt = m; 2041 else 2042 slp->ns_rec = m; 2043 slp->ns_recend = m; 2044 m->m_nextpkt = (struct mbuf *)0; 2045 } 2046 if (error) { 2047 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2048 && error != EWOULDBLOCK) { 2049 slp->ns_flag |= SLP_DISCONN; 2050 goto dorecs; 2051 } 2052 } 2053 } while (mp); 2054 } 2055 2056 /* 2057 * Now try and process the request records, non-blocking. 2058 */ 2059 dorecs: 2060 if (waitflag == M_DONTWAIT && 2061 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 2062 nfsrv_wakenfsd(slp); 2063 } 2064 2065 /* 2066 * Try and extract an RPC request from the mbuf data list received on a 2067 * stream socket. The "waitflag" argument indicates whether or not it 2068 * can sleep. 2069 */ 2070 int 2071 nfsrv_getstream(slp, waitflag) 2072 struct nfssvc_sock *slp; 2073 int waitflag; 2074 { 2075 struct mbuf *m, **mpp; 2076 struct mbuf *recm; 2077 u_int32_t recmark; 2078 2079 if (slp->ns_flag & SLP_GETSTREAM) 2080 panic("nfs getstream"); 2081 slp->ns_flag |= SLP_GETSTREAM; 2082 for (;;) { 2083 if (slp->ns_reclen == 0) { 2084 if (slp->ns_cc < NFSX_UNSIGNED) { 2085 slp->ns_flag &= ~SLP_GETSTREAM; 2086 return (0); 2087 } 2088 m = slp->ns_raw; 2089 m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark); 2090 m_adj(m, NFSX_UNSIGNED); 2091 slp->ns_cc -= NFSX_UNSIGNED; 2092 recmark = ntohl(recmark); 2093 slp->ns_reclen = recmark & ~0x80000000; 2094 if (recmark & 0x80000000) 2095 slp->ns_flag |= SLP_LASTFRAG; 2096 else 2097 slp->ns_flag &= ~SLP_LASTFRAG; 2098 if (slp->ns_reclen > NFS_MAXPACKET) { 2099 slp->ns_flag &= ~SLP_GETSTREAM; 2100 return (EPERM); 2101 } 2102 } 2103 2104 /* 2105 * Now get the record part. 2106 * 2107 * Note that slp->ns_reclen may be 0. Linux sometimes 2108 * generates 0-length records. 2109 */ 2110 if (slp->ns_cc == slp->ns_reclen) { 2111 recm = slp->ns_raw; 2112 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2113 slp->ns_cc = slp->ns_reclen = 0; 2114 } else if (slp->ns_cc > slp->ns_reclen) { 2115 recm = slp->ns_raw; 2116 m = m_split(recm, slp->ns_reclen, waitflag); 2117 if (m == NULL) { 2118 slp->ns_flag &= ~SLP_GETSTREAM; 2119 return (EWOULDBLOCK); 2120 } 2121 m_claim(recm, &nfs_mowner); 2122 slp->ns_raw = m; 2123 if (m->m_next == NULL) 2124 slp->ns_rawend = m; 2125 slp->ns_cc -= slp->ns_reclen; 2126 slp->ns_reclen = 0; 2127 } else { 2128 slp->ns_flag &= ~SLP_GETSTREAM; 2129 return (0); 2130 } 2131 2132 /* 2133 * Accumulate the fragments into a record. 2134 */ 2135 mpp = &slp->ns_frag; 2136 while (*mpp) 2137 mpp = &((*mpp)->m_next); 2138 *mpp = recm; 2139 if (slp->ns_flag & SLP_LASTFRAG) { 2140 if (slp->ns_recend) 2141 slp->ns_recend->m_nextpkt = slp->ns_frag; 2142 else 2143 slp->ns_rec = slp->ns_frag; 2144 slp->ns_recend = slp->ns_frag; 2145 slp->ns_frag = (struct mbuf *)0; 2146 } 2147 } 2148 } 2149 2150 /* 2151 * Parse an RPC header. 2152 */ 2153 int 2154 nfsrv_dorec(slp, nfsd, ndp) 2155 struct nfssvc_sock *slp; 2156 struct nfsd *nfsd; 2157 struct nfsrv_descript **ndp; 2158 { 2159 struct mbuf *m, *nam; 2160 struct nfsrv_descript *nd; 2161 int error; 2162 2163 *ndp = NULL; 2164 if ((slp->ns_flag & SLP_VALID) == 0 || 2165 (m = slp->ns_rec) == (struct mbuf *)0) 2166 return (ENOBUFS); 2167 slp->ns_rec = m->m_nextpkt; 2168 if (slp->ns_rec) 2169 m->m_nextpkt = (struct mbuf *)0; 2170 else 2171 slp->ns_recend = (struct mbuf *)0; 2172 if (m->m_type == MT_SONAME) { 2173 nam = m; 2174 m = m->m_next; 2175 nam->m_next = NULL; 2176 } else 2177 nam = NULL; 2178 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK); 2179 nd->nd_md = nd->nd_mrep = m; 2180 nd->nd_nam2 = nam; 2181 nd->nd_dpos = mtod(m, caddr_t); 2182 error = nfs_getreq(nd, nfsd, TRUE); 2183 if (error) { 2184 m_freem(nam); 2185 pool_put(&nfs_srvdesc_pool, nd); 2186 return (error); 2187 } 2188 *ndp = nd; 2189 nfsd->nfsd_nd = nd; 2190 return (0); 2191 } 2192 2193 2194 /* 2195 * Search for a sleeping nfsd and wake it up. 2196 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2197 * running nfsds will go look for the work in the nfssvc_sock list. 2198 */ 2199 void 2200 nfsrv_wakenfsd(slp) 2201 struct nfssvc_sock *slp; 2202 { 2203 struct nfsd *nd; 2204 2205 if ((slp->ns_flag & SLP_VALID) == 0) 2206 return; 2207 simple_lock(&nfsd_slock); 2208 if (slp->ns_flag & SLP_DOREC) { 2209 simple_unlock(&nfsd_slock); 2210 return; 2211 } 2212 nd = SLIST_FIRST(&nfsd_idle_head); 2213 if (nd) { 2214 SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle); 2215 simple_unlock(&nfsd_slock); 2216 2217 KASSERT(nd->nfsd_flag & NFSD_WAITING); 2218 nd->nfsd_flag &= ~NFSD_WAITING; 2219 if (nd->nfsd_slp) 2220 panic("nfsd wakeup"); 2221 slp->ns_sref++; 2222 nd->nfsd_slp = slp; 2223 wakeup(nd); 2224 return; 2225 } 2226 slp->ns_flag |= SLP_DOREC; 2227 nfsd_head_flag |= NFSD_CHECKSLP; 2228 TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending); 2229 simple_unlock(&nfsd_slock); 2230 } 2231 #endif /* NFSSERVER */ 2232