nfs_socket.c revision 1.66.2.2
1 /* $NetBSD: nfs_socket.c,v 1.66.2.2 2001/06/21 20:09:34 nathanw 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_nfs.h" 47 #include "opt_nfsserver.h" 48 #include "opt_inet.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/callout.h> 53 #include <sys/lwp.h> 54 #include <sys/proc.h> 55 #include <sys/mount.h> 56 #include <sys/kernel.h> 57 #include <sys/mbuf.h> 58 #include <sys/vnode.h> 59 #include <sys/domain.h> 60 #include <sys/protosw.h> 61 #include <sys/socket.h> 62 #include <sys/socketvar.h> 63 #include <sys/syslog.h> 64 #include <sys/tprintf.h> 65 #include <sys/namei.h> 66 #include <sys/signal.h> 67 #include <sys/signalvar.h> 68 69 #include <netinet/in.h> 70 #include <netinet/tcp.h> 71 72 #include <nfs/rpcv2.h> 73 #include <nfs/nfsproto.h> 74 #include <nfs/nfs.h> 75 #include <nfs/xdr_subs.h> 76 #include <nfs/nfsm_subs.h> 77 #include <nfs/nfsmount.h> 78 #include <nfs/nfsnode.h> 79 #include <nfs/nfsrtt.h> 80 #include <nfs/nqnfs.h> 81 #include <nfs/nfs_var.h> 82 83 #define TRUE 1 84 #define FALSE 0 85 86 /* 87 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 88 * Use the mean and mean deviation of rtt for the appropriate type of rpc 89 * for the frequent rpcs and a default for the others. 90 * The justification for doing "other" this way is that these rpcs 91 * happen so infrequently that timer est. would probably be stale. 92 * Also, since many of these rpcs are 93 * non-idempotent, a conservative timeout is desired. 94 * getattr, lookup - A+2D 95 * read, write - A+4D 96 * other - nm_timeo 97 */ 98 #define NFS_RTO(n, t) \ 99 ((t) == 0 ? (n)->nm_timeo : \ 100 ((t) < 3 ? \ 101 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 102 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 103 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 104 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 105 /* 106 * External data, mostly RPC constants in XDR form 107 */ 108 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 109 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, 110 rpc_auth_kerb; 111 extern u_int32_t nfs_prog, nqnfs_prog; 112 extern time_t nqnfsstarttime; 113 extern struct nfsstats nfsstats; 114 extern int nfsv3_procid[NFS_NPROCS]; 115 extern int nfs_ticks; 116 117 /* 118 * Defines which timer to use for the procnum. 119 * 0 - default 120 * 1 - getattr 121 * 2 - lookup 122 * 3 - read 123 * 4 - write 124 */ 125 static const int proct[NFS_NPROCS] = { 126 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0, 127 0, 0, 0, 128 }; 129 130 /* 131 * There is a congestion window for outstanding rpcs maintained per mount 132 * point. The cwnd size is adjusted in roughly the way that: 133 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 134 * SIGCOMM '88". ACM, August 1988. 135 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 136 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 137 * of rpcs is in progress. 138 * (The sent count and cwnd are scaled for integer arith.) 139 * Variants of "slow start" were tried and were found to be too much of a 140 * performance hit (ave. rtt 3 times larger), 141 * I suspect due to the large rtt that nfs rpcs have. 142 */ 143 #define NFS_CWNDSCALE 256 144 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 145 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 146 int nfsrtton = 0; 147 struct nfsrtt nfsrtt; 148 149 struct callout nfs_timer_ch = CALLOUT_INITIALIZER; 150 151 /* 152 * Initialize sockets and congestion for a new NFS connection. 153 * We do not free the sockaddr if error. 154 */ 155 int 156 nfs_connect(nmp, rep) 157 struct nfsmount *nmp; 158 struct nfsreq *rep; 159 { 160 struct socket *so; 161 int s, error, rcvreserve, sndreserve; 162 struct sockaddr *saddr; 163 struct sockaddr_in *sin; 164 #ifdef INET6 165 struct sockaddr_in6 *sin6; 166 #endif 167 struct mbuf *m; 168 u_int16_t tport; 169 170 nmp->nm_so = (struct socket *)0; 171 saddr = mtod(nmp->nm_nam, struct sockaddr *); 172 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 173 nmp->nm_soproto); 174 if (error) 175 goto bad; 176 so = nmp->nm_so; 177 nmp->nm_soflags = so->so_proto->pr_flags; 178 179 /* 180 * Some servers require that the client port be a reserved port number. 181 */ 182 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 183 MGET(m, M_WAIT, MT_SONAME); 184 sin = mtod(m, struct sockaddr_in *); 185 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 186 sin->sin_family = AF_INET; 187 sin->sin_addr.s_addr = INADDR_ANY; 188 tport = IPPORT_RESERVED - 1; 189 sin->sin_port = htons(tport); 190 while ((error = sobind(so, m, &proc0)) == EADDRINUSE && 191 --tport > IPPORT_RESERVED / 2) 192 sin->sin_port = htons(tport); 193 m_freem(m); 194 if (error) 195 goto bad; 196 } 197 #ifdef INET6 198 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) { 199 MGET(m, M_WAIT, MT_SONAME); 200 sin6 = mtod(m, struct sockaddr_in6 *); 201 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6); 202 sin6->sin6_family = AF_INET6; 203 sin6->sin6_addr = in6addr_any; 204 tport = IPV6PORT_RESERVED - 1; 205 sin6->sin6_port = htons(tport); 206 while ((error = sobind(so, m, &proc0)) == EADDRINUSE && 207 --tport > IPV6PORT_RESERVED / 2) 208 sin6->sin6_port = htons(tport); 209 m_freem(m); 210 if (error) 211 goto bad; 212 } 213 #endif 214 215 /* 216 * Protocols that do not require connections may be optionally left 217 * unconnected for servers that reply from a port other than NFS_PORT. 218 */ 219 if (nmp->nm_flag & NFSMNT_NOCONN) { 220 if (nmp->nm_soflags & PR_CONNREQUIRED) { 221 error = ENOTCONN; 222 goto bad; 223 } 224 } else { 225 error = soconnect(so, nmp->nm_nam); 226 if (error) 227 goto bad; 228 229 /* 230 * Wait for the connection to complete. Cribbed from the 231 * connect system call but with the wait timing out so 232 * that interruptible mounts don't hang here for a long time. 233 */ 234 s = splsoftnet(); 235 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 236 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 237 "nfscn1", 2 * hz); 238 if ((so->so_state & SS_ISCONNECTING) && 239 so->so_error == 0 && rep && 240 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ 241 so->so_state &= ~SS_ISCONNECTING; 242 splx(s); 243 goto bad; 244 } 245 } 246 if (so->so_error) { 247 error = so->so_error; 248 so->so_error = 0; 249 splx(s); 250 goto bad; 251 } 252 splx(s); 253 } 254 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 255 so->so_rcv.sb_timeo = (5 * hz); 256 so->so_snd.sb_timeo = (5 * hz); 257 } else { 258 so->so_rcv.sb_timeo = 0; 259 so->so_snd.sb_timeo = 0; 260 } 261 if (nmp->nm_sotype == SOCK_DGRAM) { 262 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 263 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 264 NFS_MAXPKTHDR) * 2; 265 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 266 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 267 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 268 NFS_MAXPKTHDR) * 2; 269 } else { 270 if (nmp->nm_sotype != SOCK_STREAM) 271 panic("nfscon sotype"); 272 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 273 MGET(m, M_WAIT, MT_SOOPTS); 274 *mtod(m, int32_t *) = 1; 275 m->m_len = sizeof(int32_t); 276 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 277 } 278 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 279 MGET(m, M_WAIT, MT_SOOPTS); 280 *mtod(m, int32_t *) = 1; 281 m->m_len = sizeof(int32_t); 282 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 283 } 284 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 285 sizeof (u_int32_t)) * 2; 286 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 287 sizeof (u_int32_t)) * 2; 288 } 289 error = soreserve(so, sndreserve, rcvreserve); 290 if (error) 291 goto bad; 292 so->so_rcv.sb_flags |= SB_NOINTR; 293 so->so_snd.sb_flags |= SB_NOINTR; 294 295 /* Initialize other non-zero congestion variables */ 296 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 297 nmp->nm_srtt[4] = (NFS_TIMEO << 3); 298 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 299 nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0; 300 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 301 nmp->nm_sent = 0; 302 nmp->nm_timeouts = 0; 303 return (0); 304 305 bad: 306 nfs_disconnect(nmp); 307 return (error); 308 } 309 310 /* 311 * Reconnect routine: 312 * Called when a connection is broken on a reliable protocol. 313 * - clean up the old socket 314 * - nfs_connect() again 315 * - set R_MUSTRESEND for all outstanding requests on mount point 316 * If this fails the mount point is DEAD! 317 * nb: Must be called with the nfs_sndlock() set on the mount point. 318 */ 319 int 320 nfs_reconnect(rep) 321 struct nfsreq *rep; 322 { 323 struct nfsreq *rp; 324 struct nfsmount *nmp = rep->r_nmp; 325 int error; 326 327 nfs_disconnect(nmp); 328 while ((error = nfs_connect(nmp, rep)) != 0) { 329 if (error == EINTR || error == ERESTART) 330 return (EINTR); 331 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0); 332 } 333 334 /* 335 * Loop through outstanding request list and fix up all requests 336 * on old socket. 337 */ 338 for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) { 339 if (rp->r_nmp == nmp) 340 rp->r_flags |= R_MUSTRESEND; 341 } 342 return (0); 343 } 344 345 /* 346 * NFS disconnect. Clean up and unlink. 347 */ 348 void 349 nfs_disconnect(nmp) 350 struct nfsmount *nmp; 351 { 352 struct socket *so; 353 int drain = 0; 354 355 if (nmp->nm_so) { 356 so = nmp->nm_so; 357 nmp->nm_so = (struct socket *)0; 358 soshutdown(so, 2); 359 drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0; 360 if (drain) { 361 /* 362 * soshutdown() above should wake up the current 363 * listener. 364 * Now wake up those waiting for the recive lock, and 365 * wait for them to go away unhappy, to prevent *nmp 366 * from evaporating while they're sleeping. 367 */ 368 while (nmp->nm_waiters > 0) { 369 wakeup (&nmp->nm_iflag); 370 (void) tsleep(&nmp->nm_waiters, PVFS, 371 "nfsdis", 0); 372 } 373 } 374 soclose(so); 375 } 376 #ifdef DIAGNOSTIC 377 if (drain && (nmp->nm_waiters > 0)) 378 panic("nfs_disconnect: waiters left after drain?\n"); 379 #endif 380 } 381 382 void 383 nfs_safedisconnect(nmp) 384 struct nfsmount *nmp; 385 { 386 struct nfsreq dummyreq; 387 388 memset(&dummyreq, 0, sizeof(dummyreq)); 389 dummyreq.r_nmp = nmp; 390 nfs_rcvlock(&dummyreq); /* XXX ignored error return */ 391 nfs_disconnect(nmp); 392 nfs_rcvunlock(&nmp->nm_iflag); 393 } 394 395 /* 396 * This is the nfs send routine. For connection based socket types, it 397 * must be called with an nfs_sndlock() on the socket. 398 * "rep == NULL" indicates that it has been called from a server. 399 * For the client side: 400 * - return EINTR if the RPC is terminated, 0 otherwise 401 * - set R_MUSTRESEND if the send fails for any reason 402 * - do any cleanup required by recoverable socket errors (? ? ?) 403 * For the server side: 404 * - return EINTR or ERESTART if interrupted by a signal 405 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 406 * - do any cleanup required by recoverable socket errors (? ? ?) 407 */ 408 int 409 nfs_send(so, nam, top, rep) 410 struct socket *so; 411 struct mbuf *nam; 412 struct mbuf *top; 413 struct nfsreq *rep; 414 { 415 struct mbuf *sendnam; 416 int error, soflags, flags; 417 418 if (rep) { 419 if (rep->r_flags & R_SOFTTERM) { 420 m_freem(top); 421 return (EINTR); 422 } 423 if ((so = rep->r_nmp->nm_so) == NULL) { 424 rep->r_flags |= R_MUSTRESEND; 425 m_freem(top); 426 return (0); 427 } 428 rep->r_flags &= ~R_MUSTRESEND; 429 soflags = rep->r_nmp->nm_soflags; 430 } else 431 soflags = so->so_proto->pr_flags; 432 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 433 sendnam = (struct mbuf *)0; 434 else 435 sendnam = nam; 436 if (so->so_type == SOCK_SEQPACKET) 437 flags = MSG_EOR; 438 else 439 flags = 0; 440 441 error = (*so->so_send)(so, sendnam, (struct uio *)0, top, 442 (struct mbuf *)0, flags); 443 if (error) { 444 if (rep) { 445 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 446 /* 447 * We're too fast for the network/driver, 448 * and UDP isn't flowcontrolled. 449 * We need to resend. This is not fatal, 450 * just try again. 451 * 452 * Could be smarter here by doing some sort 453 * of a backoff, but this is rare. 454 */ 455 rep->r_flags |= R_MUSTRESEND; 456 } else { 457 log(LOG_INFO, "nfs send error %d for %s\n", 458 error, 459 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 460 /* 461 * Deal with errors for the client side. 462 */ 463 if (rep->r_flags & R_SOFTTERM) 464 error = EINTR; 465 else 466 rep->r_flags |= R_MUSTRESEND; 467 } 468 } else { 469 /* 470 * See above. This error can happen under normal 471 * circumstances and the log is too noisy. 472 * The error will still show up in nfsstat. 473 */ 474 if (error != ENOBUFS || so->so_type != SOCK_DGRAM) 475 log(LOG_INFO, "nfsd send error %d\n", error); 476 } 477 478 /* 479 * Handle any recoverable (soft) socket errors here. (? ? ?) 480 */ 481 if (error != EINTR && error != ERESTART && 482 error != EWOULDBLOCK && error != EPIPE) 483 error = 0; 484 } 485 return (error); 486 } 487 488 #ifdef NFS 489 /* 490 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 491 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 492 * Mark and consolidate the data into a new mbuf list. 493 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 494 * small mbufs. 495 * For SOCK_STREAM we must be very careful to read an entire record once 496 * we have read any of it, even if the system call has been interrupted. 497 */ 498 int 499 nfs_receive(rep, aname, mp) 500 struct nfsreq *rep; 501 struct mbuf **aname; 502 struct mbuf **mp; 503 { 504 struct socket *so; 505 struct uio auio; 506 struct iovec aio; 507 struct mbuf *m; 508 struct mbuf *control; 509 u_int32_t len; 510 struct mbuf **getnam; 511 int error, sotype, rcvflg; 512 struct proc *p = curproc->l_proc; /* XXX */ 513 514 /* 515 * Set up arguments for soreceive() 516 */ 517 *mp = (struct mbuf *)0; 518 *aname = (struct mbuf *)0; 519 sotype = rep->r_nmp->nm_sotype; 520 521 /* 522 * For reliable protocols, lock against other senders/receivers 523 * in case a reconnect is necessary. 524 * For SOCK_STREAM, first get the Record Mark to find out how much 525 * more there is to get. 526 * We must lock the socket against other receivers 527 * until we have an entire rpc request/reply. 528 */ 529 if (sotype != SOCK_DGRAM) { 530 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 531 if (error) 532 return (error); 533 tryagain: 534 /* 535 * Check for fatal errors and resending request. 536 */ 537 /* 538 * Ugh: If a reconnect attempt just happened, nm_so 539 * would have changed. NULL indicates a failed 540 * attempt that has essentially shut down this 541 * mount point. 542 */ 543 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 544 nfs_sndunlock(&rep->r_nmp->nm_iflag); 545 return (EINTR); 546 } 547 so = rep->r_nmp->nm_so; 548 if (!so) { 549 error = nfs_reconnect(rep); 550 if (error) { 551 nfs_sndunlock(&rep->r_nmp->nm_iflag); 552 return (error); 553 } 554 goto tryagain; 555 } 556 while (rep->r_flags & R_MUSTRESEND) { 557 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 558 nfsstats.rpcretries++; 559 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 560 if (error) { 561 if (error == EINTR || error == ERESTART || 562 (error = nfs_reconnect(rep)) != 0) { 563 nfs_sndunlock(&rep->r_nmp->nm_iflag); 564 return (error); 565 } 566 goto tryagain; 567 } 568 } 569 nfs_sndunlock(&rep->r_nmp->nm_iflag); 570 if (sotype == SOCK_STREAM) { 571 aio.iov_base = (caddr_t) &len; 572 aio.iov_len = sizeof(u_int32_t); 573 auio.uio_iov = &aio; 574 auio.uio_iovcnt = 1; 575 auio.uio_segflg = UIO_SYSSPACE; 576 auio.uio_rw = UIO_READ; 577 auio.uio_offset = 0; 578 auio.uio_resid = sizeof(u_int32_t); 579 auio.uio_procp = p; 580 do { 581 rcvflg = MSG_WAITALL; 582 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 583 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 584 if (error == EWOULDBLOCK && rep) { 585 if (rep->r_flags & R_SOFTTERM) 586 return (EINTR); 587 } 588 } while (error == EWOULDBLOCK); 589 if (!error && auio.uio_resid > 0) { 590 /* 591 * Don't log a 0 byte receive; it means 592 * that the socket has been closed, and 593 * can happen during normal operation 594 * (forcible unmount or Solaris server). 595 */ 596 if (auio.uio_resid != sizeof (u_int32_t)) 597 log(LOG_INFO, 598 "short receive (%lu/%lu) from nfs server %s\n", 599 (u_long)sizeof(u_int32_t) - auio.uio_resid, 600 (u_long)sizeof(u_int32_t), 601 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 602 error = EPIPE; 603 } 604 if (error) 605 goto errout; 606 len = ntohl(len) & ~0x80000000; 607 /* 608 * This is SERIOUS! We are out of sync with the sender 609 * and forcing a disconnect/reconnect is all I can do. 610 */ 611 if (len > NFS_MAXPACKET) { 612 log(LOG_ERR, "%s (%d) from nfs server %s\n", 613 "impossible packet length", 614 len, 615 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 616 error = EFBIG; 617 goto errout; 618 } 619 auio.uio_resid = len; 620 do { 621 rcvflg = MSG_WAITALL; 622 error = (*so->so_receive)(so, (struct mbuf **)0, 623 &auio, mp, (struct mbuf **)0, &rcvflg); 624 } while (error == EWOULDBLOCK || error == EINTR || 625 error == ERESTART); 626 if (!error && auio.uio_resid > 0) { 627 if (len != auio.uio_resid) 628 log(LOG_INFO, 629 "short receive (%lu/%d) from nfs server %s\n", 630 (u_long)len - auio.uio_resid, len, 631 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 632 error = EPIPE; 633 } 634 } else { 635 /* 636 * NB: Since uio_resid is big, MSG_WAITALL is ignored 637 * and soreceive() will return when it has either a 638 * control msg or a data msg. 639 * We have no use for control msg., but must grab them 640 * and then throw them away so we know what is going 641 * on. 642 */ 643 auio.uio_resid = len = 100000000; /* Anything Big */ 644 auio.uio_procp = p; 645 do { 646 rcvflg = 0; 647 error = (*so->so_receive)(so, (struct mbuf **)0, 648 &auio, mp, &control, &rcvflg); 649 if (control) 650 m_freem(control); 651 if (error == EWOULDBLOCK && rep) { 652 if (rep->r_flags & R_SOFTTERM) 653 return (EINTR); 654 } 655 } while (error == EWOULDBLOCK || 656 (!error && *mp == NULL && control)); 657 if ((rcvflg & MSG_EOR) == 0) 658 printf("Egad!!\n"); 659 if (!error && *mp == NULL) 660 error = EPIPE; 661 len -= auio.uio_resid; 662 } 663 errout: 664 if (error && error != EINTR && error != ERESTART) { 665 m_freem(*mp); 666 *mp = (struct mbuf *)0; 667 if (error != EPIPE) 668 log(LOG_INFO, 669 "receive error %d from nfs server %s\n", 670 error, 671 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 672 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 673 if (!error) 674 error = nfs_reconnect(rep); 675 if (!error) 676 goto tryagain; 677 else 678 nfs_sndunlock(&rep->r_nmp->nm_iflag); 679 } 680 } else { 681 if ((so = rep->r_nmp->nm_so) == NULL) 682 return (EACCES); 683 if (so->so_state & SS_ISCONNECTED) 684 getnam = (struct mbuf **)0; 685 else 686 getnam = aname; 687 auio.uio_resid = len = 1000000; 688 auio.uio_procp = p; 689 do { 690 rcvflg = 0; 691 error = (*so->so_receive)(so, getnam, &auio, mp, 692 (struct mbuf **)0, &rcvflg); 693 if (error == EWOULDBLOCK && 694 (rep->r_flags & R_SOFTTERM)) 695 return (EINTR); 696 } while (error == EWOULDBLOCK); 697 len -= auio.uio_resid; 698 if (!error && *mp == NULL) 699 error = EPIPE; 700 } 701 if (error) { 702 m_freem(*mp); 703 *mp = (struct mbuf *)0; 704 } 705 return (error); 706 } 707 708 /* 709 * Implement receipt of reply on a socket. 710 * We must search through the list of received datagrams matching them 711 * with outstanding requests using the xid, until ours is found. 712 */ 713 /* ARGSUSED */ 714 int 715 nfs_reply(myrep) 716 struct nfsreq *myrep; 717 { 718 struct nfsreq *rep; 719 struct nfsmount *nmp = myrep->r_nmp; 720 int32_t t1; 721 struct mbuf *mrep, *nam, *md; 722 u_int32_t rxid, *tl; 723 caddr_t dpos, cp2; 724 int error; 725 726 /* 727 * Loop around until we get our own reply 728 */ 729 for (;;) { 730 /* 731 * Lock against other receivers so that I don't get stuck in 732 * sbwait() after someone else has received my reply for me. 733 * Also necessary for connection based protocols to avoid 734 * race conditions during a reconnect. 735 */ 736 error = nfs_rcvlock(myrep); 737 if (error == EALREADY) 738 return (0); 739 if (error) 740 return (error); 741 /* 742 * Get the next Rpc reply off the socket 743 */ 744 nmp->nm_waiters++; 745 error = nfs_receive(myrep, &nam, &mrep); 746 nfs_rcvunlock(&nmp->nm_iflag); 747 if (error) { 748 749 if (nmp->nm_iflag & NFSMNT_DISMNT) { 750 /* 751 * Oops, we're going away now.. 752 */ 753 nmp->nm_waiters--; 754 wakeup (&nmp->nm_waiters); 755 return error; 756 } 757 nmp->nm_waiters--; 758 /* 759 * Ignore routing errors on connectionless protocols? ? 760 */ 761 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 762 nmp->nm_so->so_error = 0; 763 #ifdef DEBUG 764 printf("nfs_reply: ignoring error %d\n", error); 765 #endif 766 if (myrep->r_flags & R_GETONEREP) 767 return (0); 768 continue; 769 } 770 return (error); 771 } 772 nmp->nm_waiters--; 773 if (nam) 774 m_freem(nam); 775 776 /* 777 * Get the xid and check that it is an rpc reply 778 */ 779 md = mrep; 780 dpos = mtod(md, caddr_t); 781 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 782 rxid = *tl++; 783 if (*tl != rpc_reply) { 784 #ifndef NFS_V2_ONLY 785 if (nmp->nm_flag & NFSMNT_NQNFS) { 786 if (nqnfs_callback(nmp, mrep, md, dpos)) 787 nfsstats.rpcinvalid++; 788 } else 789 #endif 790 { 791 nfsstats.rpcinvalid++; 792 m_freem(mrep); 793 } 794 nfsmout: 795 if (myrep->r_flags & R_GETONEREP) 796 return (0); 797 continue; 798 } 799 800 /* 801 * Loop through the request list to match up the reply 802 * Iff no match, just drop the datagram 803 */ 804 for (rep = nfs_reqq.tqh_first; rep != 0; 805 rep = rep->r_chain.tqe_next) { 806 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 807 /* Found it.. */ 808 rep->r_mrep = mrep; 809 rep->r_md = md; 810 rep->r_dpos = dpos; 811 if (nfsrtton) { 812 struct rttl *rt; 813 814 rt = &nfsrtt.rttl[nfsrtt.pos]; 815 rt->proc = rep->r_procnum; 816 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 817 rt->sent = nmp->nm_sent; 818 rt->cwnd = nmp->nm_cwnd; 819 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 820 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 821 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 822 rt->tstamp = time; 823 if (rep->r_flags & R_TIMING) 824 rt->rtt = rep->r_rtt; 825 else 826 rt->rtt = 1000000; 827 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 828 } 829 /* 830 * Update congestion window. 831 * Do the additive increase of 832 * one rpc/rtt. 833 */ 834 if (nmp->nm_cwnd <= nmp->nm_sent) { 835 nmp->nm_cwnd += 836 (NFS_CWNDSCALE * NFS_CWNDSCALE + 837 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 838 if (nmp->nm_cwnd > NFS_MAXCWND) 839 nmp->nm_cwnd = NFS_MAXCWND; 840 } 841 rep->r_flags &= ~R_SENT; 842 nmp->nm_sent -= NFS_CWNDSCALE; 843 /* 844 * Update rtt using a gain of 0.125 on the mean 845 * and a gain of 0.25 on the deviation. 846 */ 847 if (rep->r_flags & R_TIMING) { 848 /* 849 * Since the timer resolution of 850 * NFS_HZ is so course, it can often 851 * result in r_rtt == 0. Since 852 * r_rtt == N means that the actual 853 * rtt is between N+dt and N+2-dt ticks, 854 * add 1. 855 */ 856 t1 = rep->r_rtt + 1; 857 t1 -= (NFS_SRTT(rep) >> 3); 858 NFS_SRTT(rep) += t1; 859 if (t1 < 0) 860 t1 = -t1; 861 t1 -= (NFS_SDRTT(rep) >> 2); 862 NFS_SDRTT(rep) += t1; 863 } 864 nmp->nm_timeouts = 0; 865 break; 866 } 867 } 868 /* 869 * If not matched to a request, drop it. 870 * If it's mine, get out. 871 */ 872 if (rep == 0) { 873 nfsstats.rpcunexpected++; 874 m_freem(mrep); 875 } else if (rep == myrep) { 876 if (rep->r_mrep == NULL) 877 panic("nfsreply nil"); 878 return (0); 879 } 880 if (myrep->r_flags & R_GETONEREP) 881 return (0); 882 } 883 } 884 885 /* 886 * nfs_request - goes something like this 887 * - fill in request struct 888 * - links it into list 889 * - calls nfs_send() for first transmit 890 * - calls nfs_receive() to get reply 891 * - break down rpc header and return with nfs reply pointed to 892 * by mrep or error 893 * nb: always frees up mreq mbuf list 894 */ 895 int 896 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 897 struct vnode *vp; 898 struct mbuf *mrest; 899 int procnum; 900 struct proc *procp; 901 struct ucred *cred; 902 struct mbuf **mrp; 903 struct mbuf **mdp; 904 caddr_t *dposp; 905 { 906 struct mbuf *m, *mrep; 907 struct nfsreq *rep; 908 u_int32_t *tl; 909 int i; 910 struct nfsmount *nmp; 911 struct mbuf *md, *mheadend; 912 char nickv[RPCX_NICKVERF]; 913 time_t reqtime, waituntil; 914 caddr_t dpos, cp2; 915 int t1, s, error = 0, mrest_len, auth_len, auth_type; 916 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 917 int verf_len, verf_type; 918 u_int32_t xid; 919 char *auth_str, *verf_str; 920 NFSKERBKEY_T key; /* save session key */ 921 #ifndef NFS_V2_ONLY 922 int nqlflag, cachable; 923 u_quad_t frev; 924 struct nfsnode *np; 925 #endif 926 927 KASSERT(cred != NULL); 928 nmp = VFSTONFS(vp->v_mount); 929 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 930 rep->r_nmp = nmp; 931 rep->r_vp = vp; 932 rep->r_procp = procp; 933 rep->r_procnum = procnum; 934 i = 0; 935 m = mrest; 936 while (m) { 937 i += m->m_len; 938 m = m->m_next; 939 } 940 mrest_len = i; 941 942 /* 943 * Get the RPC header with authorization. 944 */ 945 kerbauth: 946 verf_str = auth_str = (char *)0; 947 if (nmp->nm_flag & NFSMNT_KERB) { 948 verf_str = nickv; 949 verf_len = sizeof (nickv); 950 auth_type = RPCAUTH_KERB4; 951 memset((caddr_t)key, 0, sizeof (key)); 952 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 953 &auth_len, verf_str, verf_len)) { 954 error = nfs_getauth(nmp, rep, cred, &auth_str, 955 &auth_len, verf_str, &verf_len, key); 956 if (error) { 957 free((caddr_t)rep, M_NFSREQ); 958 m_freem(mrest); 959 return (error); 960 } 961 } 962 } else { 963 auth_type = RPCAUTH_UNIX; 964 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 965 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 966 5 * NFSX_UNSIGNED; 967 } 968 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 969 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 970 if (auth_str) 971 free(auth_str, M_TEMP); 972 973 /* 974 * For stream protocols, insert a Sun RPC Record Mark. 975 */ 976 if (nmp->nm_sotype == SOCK_STREAM) { 977 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 978 *mtod(m, u_int32_t *) = htonl(0x80000000 | 979 (m->m_pkthdr.len - NFSX_UNSIGNED)); 980 } 981 rep->r_mreq = m; 982 rep->r_xid = xid; 983 tryagain: 984 if (nmp->nm_flag & NFSMNT_SOFT) 985 rep->r_retry = nmp->nm_retry; 986 else 987 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 988 rep->r_rtt = rep->r_rexmit = 0; 989 if (proct[procnum] > 0) 990 rep->r_flags = R_TIMING; 991 else 992 rep->r_flags = 0; 993 rep->r_mrep = NULL; 994 995 /* 996 * Do the client side RPC. 997 */ 998 nfsstats.rpcrequests++; 999 /* 1000 * Chain request into list of outstanding requests. Be sure 1001 * to put it LAST so timer finds oldest requests first. 1002 */ 1003 s = splsoftnet(); 1004 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1005 1006 /* Get send time for nqnfs */ 1007 reqtime = time.tv_sec; 1008 1009 /* 1010 * If backing off another request or avoiding congestion, don't 1011 * send this one now but let timer do it. If not timing a request, 1012 * do it now. 1013 */ 1014 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1015 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1016 nmp->nm_sent < nmp->nm_cwnd)) { 1017 splx(s); 1018 if (nmp->nm_soflags & PR_CONNREQUIRED) 1019 error = nfs_sndlock(&nmp->nm_iflag, rep); 1020 if (!error) { 1021 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1022 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 1023 if (nmp->nm_soflags & PR_CONNREQUIRED) 1024 nfs_sndunlock(&nmp->nm_iflag); 1025 } 1026 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1027 nmp->nm_sent += NFS_CWNDSCALE; 1028 rep->r_flags |= R_SENT; 1029 } 1030 } else { 1031 splx(s); 1032 rep->r_rtt = -1; 1033 } 1034 1035 /* 1036 * Wait for the reply from our send or the timer's. 1037 */ 1038 if (!error || error == EPIPE) 1039 error = nfs_reply(rep); 1040 1041 /* 1042 * RPC done, unlink the request. 1043 */ 1044 s = splsoftnet(); 1045 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1046 splx(s); 1047 1048 /* 1049 * Decrement the outstanding request count. 1050 */ 1051 if (rep->r_flags & R_SENT) { 1052 rep->r_flags &= ~R_SENT; /* paranoia */ 1053 nmp->nm_sent -= NFS_CWNDSCALE; 1054 } 1055 1056 /* 1057 * If there was a successful reply and a tprintf msg. 1058 * tprintf a response. 1059 */ 1060 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1061 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1062 "is alive again"); 1063 mrep = rep->r_mrep; 1064 md = rep->r_md; 1065 dpos = rep->r_dpos; 1066 if (error) { 1067 m_freem(rep->r_mreq); 1068 free((caddr_t)rep, M_NFSREQ); 1069 return (error); 1070 } 1071 1072 /* 1073 * break down the rpc header and check if ok 1074 */ 1075 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1076 if (*tl++ == rpc_msgdenied) { 1077 if (*tl == rpc_mismatch) 1078 error = EOPNOTSUPP; 1079 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1080 if (!failed_auth) { 1081 failed_auth++; 1082 mheadend->m_next = (struct mbuf *)0; 1083 m_freem(mrep); 1084 m_freem(rep->r_mreq); 1085 goto kerbauth; 1086 } else 1087 error = EAUTH; 1088 } else 1089 error = EACCES; 1090 m_freem(mrep); 1091 m_freem(rep->r_mreq); 1092 free((caddr_t)rep, M_NFSREQ); 1093 return (error); 1094 } 1095 1096 /* 1097 * Grab any Kerberos verifier, otherwise just throw it away. 1098 */ 1099 verf_type = fxdr_unsigned(int, *tl++); 1100 i = fxdr_unsigned(int32_t, *tl); 1101 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1102 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1103 if (error) 1104 goto nfsmout; 1105 } else if (i > 0) 1106 nfsm_adv(nfsm_rndup(i)); 1107 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1108 /* 0 == ok */ 1109 if (*tl == 0) { 1110 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1111 if (*tl != 0) { 1112 error = fxdr_unsigned(int, *tl); 1113 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1114 error == NFSERR_TRYLATER) { 1115 m_freem(mrep); 1116 error = 0; 1117 waituntil = time.tv_sec + trylater_delay; 1118 while (time.tv_sec < waituntil) 1119 (void) tsleep((caddr_t)&lbolt, 1120 PSOCK, "nqnfstry", 0); 1121 trylater_delay *= nfs_backoff[trylater_cnt]; 1122 if (trylater_cnt < 7) 1123 trylater_cnt++; 1124 goto tryagain; 1125 } 1126 1127 /* 1128 * If the File Handle was stale, invalidate the 1129 * lookup cache, just in case. 1130 */ 1131 if (error == ESTALE) 1132 cache_purge(vp); 1133 if (nmp->nm_flag & NFSMNT_NFSV3) { 1134 *mrp = mrep; 1135 *mdp = md; 1136 *dposp = dpos; 1137 error |= NFSERR_RETERR; 1138 } else 1139 m_freem(mrep); 1140 m_freem(rep->r_mreq); 1141 free((caddr_t)rep, M_NFSREQ); 1142 return (error); 1143 } 1144 1145 #ifndef NFS_V2_ONLY 1146 /* 1147 * For nqnfs, get any lease in reply 1148 */ 1149 if (nmp->nm_flag & NFSMNT_NQNFS) { 1150 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1151 if (*tl) { 1152 np = VTONFS(vp); 1153 nqlflag = fxdr_unsigned(int, *tl); 1154 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1155 cachable = fxdr_unsigned(int, *tl++); 1156 reqtime += fxdr_unsigned(int, *tl++); 1157 if (reqtime > time.tv_sec) { 1158 frev = fxdr_hyper(tl); 1159 nqnfs_clientlease(nmp, np, nqlflag, 1160 cachable, reqtime, frev); 1161 } 1162 } 1163 } 1164 #endif 1165 *mrp = mrep; 1166 *mdp = md; 1167 *dposp = dpos; 1168 m_freem(rep->r_mreq); 1169 FREE((caddr_t)rep, M_NFSREQ); 1170 return (0); 1171 } 1172 m_freem(mrep); 1173 error = EPROTONOSUPPORT; 1174 nfsmout: 1175 m_freem(rep->r_mreq); 1176 free((caddr_t)rep, M_NFSREQ); 1177 return (error); 1178 } 1179 #endif /* NFS */ 1180 1181 /* 1182 * Generate the rpc reply header 1183 * siz arg. is used to decide if adding a cluster is worthwhile 1184 */ 1185 int 1186 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1187 int siz; 1188 struct nfsrv_descript *nd; 1189 struct nfssvc_sock *slp; 1190 int err; 1191 int cache; 1192 u_quad_t *frev; 1193 struct mbuf **mrq; 1194 struct mbuf **mbp; 1195 caddr_t *bposp; 1196 { 1197 u_int32_t *tl; 1198 struct mbuf *mreq; 1199 caddr_t bpos; 1200 struct mbuf *mb, *mb2; 1201 1202 MGETHDR(mreq, M_WAIT, MT_DATA); 1203 mb = mreq; 1204 /* 1205 * If this is a big reply, use a cluster else 1206 * try and leave leading space for the lower level headers. 1207 */ 1208 siz += RPC_REPLYSIZ; 1209 if (siz >= max_datalen) { 1210 MCLGET(mreq, M_WAIT); 1211 } else 1212 mreq->m_data += max_hdr; 1213 tl = mtod(mreq, u_int32_t *); 1214 mreq->m_len = 6 * NFSX_UNSIGNED; 1215 bpos = ((caddr_t)tl) + mreq->m_len; 1216 *tl++ = txdr_unsigned(nd->nd_retxid); 1217 *tl++ = rpc_reply; 1218 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1219 *tl++ = rpc_msgdenied; 1220 if (err & NFSERR_AUTHERR) { 1221 *tl++ = rpc_autherr; 1222 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1223 mreq->m_len -= NFSX_UNSIGNED; 1224 bpos -= NFSX_UNSIGNED; 1225 } else { 1226 *tl++ = rpc_mismatch; 1227 *tl++ = txdr_unsigned(RPC_VER2); 1228 *tl = txdr_unsigned(RPC_VER2); 1229 } 1230 } else { 1231 *tl++ = rpc_msgaccepted; 1232 1233 /* 1234 * For Kerberos authentication, we must send the nickname 1235 * verifier back, otherwise just RPCAUTH_NULL. 1236 */ 1237 if (nd->nd_flag & ND_KERBFULL) { 1238 struct nfsuid *nuidp; 1239 struct timeval ktvin, ktvout; 1240 1241 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; 1242 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1243 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1244 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), 1245 &nuidp->nu_haddr, nd->nd_nam2))) 1246 break; 1247 } 1248 if (nuidp) { 1249 ktvin.tv_sec = 1250 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); 1251 ktvin.tv_usec = 1252 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1253 1254 /* 1255 * Encrypt the timestamp in ecb mode using the 1256 * session key. 1257 */ 1258 #ifdef NFSKERB 1259 XXX 1260 #endif 1261 1262 *tl++ = rpc_auth_kerb; 1263 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1264 *tl = ktvout.tv_sec; 1265 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1266 *tl++ = ktvout.tv_usec; 1267 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1268 } else { 1269 *tl++ = 0; 1270 *tl++ = 0; 1271 } 1272 } else { 1273 *tl++ = 0; 1274 *tl++ = 0; 1275 } 1276 switch (err) { 1277 case EPROGUNAVAIL: 1278 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1279 break; 1280 case EPROGMISMATCH: 1281 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1282 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1283 if (nd->nd_flag & ND_NQNFS) { 1284 *tl++ = txdr_unsigned(3); 1285 *tl = txdr_unsigned(3); 1286 } else { 1287 *tl++ = txdr_unsigned(2); 1288 *tl = txdr_unsigned(3); 1289 } 1290 break; 1291 case EPROCUNAVAIL: 1292 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1293 break; 1294 case EBADRPC: 1295 *tl = txdr_unsigned(RPC_GARBAGE); 1296 break; 1297 default: 1298 *tl = 0; 1299 if (err != NFSERR_RETVOID) { 1300 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1301 if (err) 1302 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1303 else 1304 *tl = 0; 1305 } 1306 break; 1307 }; 1308 } 1309 1310 /* 1311 * For nqnfs, piggyback lease as requested. 1312 */ 1313 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1314 if (nd->nd_flag & ND_LEASE) { 1315 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1316 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1317 *tl++ = txdr_unsigned(cache); 1318 *tl++ = txdr_unsigned(nd->nd_duration); 1319 txdr_hyper(*frev, tl); 1320 } else { 1321 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1322 *tl = 0; 1323 } 1324 } 1325 if (mrq != NULL) 1326 *mrq = mreq; 1327 *mbp = mb; 1328 *bposp = bpos; 1329 if (err != 0 && err != NFSERR_RETVOID) 1330 nfsstats.srvrpc_errs++; 1331 return (0); 1332 } 1333 1334 /* 1335 * Nfs timer routine 1336 * Scan the nfsreq list and retranmit any requests that have timed out 1337 * To avoid retransmission attempts on STREAM sockets (in the future) make 1338 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1339 */ 1340 void 1341 nfs_timer(arg) 1342 void *arg; /* never used */ 1343 { 1344 struct nfsreq *rep; 1345 struct mbuf *m; 1346 struct socket *so; 1347 struct nfsmount *nmp; 1348 int timeo; 1349 int s, error; 1350 #ifdef NFSSERVER 1351 struct nfssvc_sock *slp; 1352 static long lasttime = 0; 1353 u_quad_t cur_usec; 1354 #endif 1355 1356 s = splsoftnet(); 1357 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1358 nmp = rep->r_nmp; 1359 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1360 continue; 1361 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1362 rep->r_flags |= R_SOFTTERM; 1363 continue; 1364 } 1365 if (rep->r_rtt >= 0) { 1366 rep->r_rtt++; 1367 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1368 timeo = nmp->nm_timeo; 1369 else 1370 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1371 if (nmp->nm_timeouts > 0) 1372 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1373 if (rep->r_rtt <= timeo) 1374 continue; 1375 if (nmp->nm_timeouts < 8) 1376 nmp->nm_timeouts++; 1377 } 1378 /* 1379 * Check for server not responding 1380 */ 1381 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1382 rep->r_rexmit > nmp->nm_deadthresh) { 1383 nfs_msg(rep->r_procp, 1384 nmp->nm_mountp->mnt_stat.f_mntfromname, 1385 "not responding"); 1386 rep->r_flags |= R_TPRINTFMSG; 1387 } 1388 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1389 nfsstats.rpctimeouts++; 1390 rep->r_flags |= R_SOFTTERM; 1391 continue; 1392 } 1393 if (nmp->nm_sotype != SOCK_DGRAM) { 1394 if (++rep->r_rexmit > NFS_MAXREXMIT) 1395 rep->r_rexmit = NFS_MAXREXMIT; 1396 continue; 1397 } 1398 if ((so = nmp->nm_so) == NULL) 1399 continue; 1400 1401 /* 1402 * If there is enough space and the window allows.. 1403 * Resend it 1404 * Set r_rtt to -1 in case we fail to send it now. 1405 */ 1406 rep->r_rtt = -1; 1407 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1408 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1409 (rep->r_flags & R_SENT) || 1410 nmp->nm_sent < nmp->nm_cwnd) && 1411 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1412 if (so->so_state & SS_ISCONNECTED) 1413 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1414 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1415 else 1416 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1417 nmp->nm_nam, (struct mbuf *)0, (struct proc *)0); 1418 if (error) { 1419 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1420 #ifdef DEBUG 1421 printf("nfs_timer: ignoring error %d\n", 1422 error); 1423 #endif 1424 so->so_error = 0; 1425 } 1426 } else { 1427 /* 1428 * Iff first send, start timing 1429 * else turn timing off, backoff timer 1430 * and divide congestion window by 2. 1431 */ 1432 if (rep->r_flags & R_SENT) { 1433 rep->r_flags &= ~R_TIMING; 1434 if (++rep->r_rexmit > NFS_MAXREXMIT) 1435 rep->r_rexmit = NFS_MAXREXMIT; 1436 nmp->nm_cwnd >>= 1; 1437 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1438 nmp->nm_cwnd = NFS_CWNDSCALE; 1439 nfsstats.rpcretries++; 1440 } else { 1441 rep->r_flags |= R_SENT; 1442 nmp->nm_sent += NFS_CWNDSCALE; 1443 } 1444 rep->r_rtt = 0; 1445 } 1446 } 1447 } 1448 1449 #ifdef NFSSERVER 1450 /* 1451 * Call the nqnfs server timer once a second to handle leases. 1452 */ 1453 if (lasttime != time.tv_sec) { 1454 lasttime = time.tv_sec; 1455 nqnfs_serverd(); 1456 } 1457 1458 /* 1459 * Scan the write gathering queues for writes that need to be 1460 * completed now. 1461 */ 1462 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1463 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 1464 slp = slp->ns_chain.tqe_next) { 1465 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec) 1466 nfsrv_wakenfsd(slp); 1467 } 1468 #endif /* NFSSERVER */ 1469 splx(s); 1470 callout_reset(&nfs_timer_ch, nfs_ticks, nfs_timer, NULL); 1471 } 1472 1473 /* 1474 * Test for a termination condition pending on the process. 1475 * This is used for NFSMNT_INT mounts. 1476 */ 1477 int 1478 nfs_sigintr(nmp, rep, p) 1479 struct nfsmount *nmp; 1480 struct nfsreq *rep; 1481 struct proc *p; 1482 { 1483 sigset_t ss; 1484 1485 if (rep && (rep->r_flags & R_SOFTTERM)) 1486 return (EINTR); 1487 if (!(nmp->nm_flag & NFSMNT_INT)) 1488 return (0); 1489 if (p) { 1490 sigpending1(p, &ss); 1491 #if 0 1492 sigminusset(&p->p_sigctx.ps_sigignore, &ss); 1493 #endif 1494 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1495 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1496 sigismember(&ss, SIGQUIT)) 1497 return (EINTR); 1498 } 1499 return (0); 1500 } 1501 1502 /* 1503 * Lock a socket against others. 1504 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1505 * and also to avoid race conditions between the processes with nfs requests 1506 * in progress when a reconnect is necessary. 1507 */ 1508 int 1509 nfs_sndlock(flagp, rep) 1510 int *flagp; 1511 struct nfsreq *rep; 1512 { 1513 struct proc *p; 1514 int slpflag = 0, slptimeo = 0; 1515 1516 if (rep) { 1517 p = rep->r_procp; 1518 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1519 slpflag = PCATCH; 1520 } else 1521 p = (struct proc *)0; 1522 while (*flagp & NFSMNT_SNDLOCK) { 1523 if (nfs_sigintr(rep->r_nmp, rep, p)) 1524 return (EINTR); 1525 *flagp |= NFSMNT_WANTSND; 1526 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1527 slptimeo); 1528 if (slpflag == PCATCH) { 1529 slpflag = 0; 1530 slptimeo = 2 * hz; 1531 } 1532 } 1533 *flagp |= NFSMNT_SNDLOCK; 1534 return (0); 1535 } 1536 1537 /* 1538 * Unlock the stream socket for others. 1539 */ 1540 void 1541 nfs_sndunlock(flagp) 1542 int *flagp; 1543 { 1544 1545 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1546 panic("nfs sndunlock"); 1547 *flagp &= ~NFSMNT_SNDLOCK; 1548 if (*flagp & NFSMNT_WANTSND) { 1549 *flagp &= ~NFSMNT_WANTSND; 1550 wakeup((caddr_t)flagp); 1551 } 1552 } 1553 1554 int 1555 nfs_rcvlock(rep) 1556 struct nfsreq *rep; 1557 { 1558 struct nfsmount *nmp = rep->r_nmp; 1559 int *flagp = &nmp->nm_iflag; 1560 int slpflag, slptimeo = 0; 1561 1562 if (*flagp & NFSMNT_DISMNT) 1563 return EIO; 1564 1565 if (*flagp & NFSMNT_INT) 1566 slpflag = PCATCH; 1567 else 1568 slpflag = 0; 1569 while (*flagp & NFSMNT_RCVLOCK) { 1570 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1571 return (EINTR); 1572 *flagp |= NFSMNT_WANTRCV; 1573 nmp->nm_waiters++; 1574 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1575 slptimeo); 1576 nmp->nm_waiters--; 1577 if (*flagp & NFSMNT_DISMNT) { 1578 wakeup(&nmp->nm_waiters); 1579 return EIO; 1580 } 1581 /* If our reply was received while we were sleeping, 1582 * then just return without taking the lock to avoid a 1583 * situation where a single iod could 'capture' the 1584 * receive lock. 1585 */ 1586 if (rep->r_mrep != NULL) 1587 return (EALREADY); 1588 if (slpflag == PCATCH) { 1589 slpflag = 0; 1590 slptimeo = 2 * hz; 1591 } 1592 } 1593 *flagp |= NFSMNT_RCVLOCK; 1594 return (0); 1595 } 1596 1597 /* 1598 * Unlock the stream socket for others. 1599 */ 1600 void 1601 nfs_rcvunlock(flagp) 1602 int *flagp; 1603 { 1604 1605 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1606 panic("nfs rcvunlock"); 1607 *flagp &= ~NFSMNT_RCVLOCK; 1608 if (*flagp & NFSMNT_WANTRCV) { 1609 *flagp &= ~NFSMNT_WANTRCV; 1610 wakeup((caddr_t)flagp); 1611 } 1612 } 1613 1614 /* 1615 * Parse an RPC request 1616 * - verify it 1617 * - fill in the cred struct. 1618 */ 1619 int 1620 nfs_getreq(nd, nfsd, has_header) 1621 struct nfsrv_descript *nd; 1622 struct nfsd *nfsd; 1623 int has_header; 1624 { 1625 int len, i; 1626 u_int32_t *tl; 1627 int32_t t1; 1628 struct uio uio; 1629 struct iovec iov; 1630 caddr_t dpos, cp2, cp; 1631 u_int32_t nfsvers, auth_type; 1632 uid_t nickuid; 1633 int error = 0, nqnfs = 0, ticklen; 1634 struct mbuf *mrep, *md; 1635 struct nfsuid *nuidp; 1636 struct timeval tvin, tvout; 1637 1638 mrep = nd->nd_mrep; 1639 md = nd->nd_md; 1640 dpos = nd->nd_dpos; 1641 if (has_header) { 1642 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1643 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1644 if (*tl++ != rpc_call) { 1645 m_freem(mrep); 1646 return (EBADRPC); 1647 } 1648 } else 1649 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1650 nd->nd_repstat = 0; 1651 nd->nd_flag = 0; 1652 if (*tl++ != rpc_vers) { 1653 nd->nd_repstat = ERPCMISMATCH; 1654 nd->nd_procnum = NFSPROC_NOOP; 1655 return (0); 1656 } 1657 if (*tl != nfs_prog) { 1658 if (*tl == nqnfs_prog) 1659 nqnfs++; 1660 else { 1661 nd->nd_repstat = EPROGUNAVAIL; 1662 nd->nd_procnum = NFSPROC_NOOP; 1663 return (0); 1664 } 1665 } 1666 tl++; 1667 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1668 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1669 (nfsvers != NQNFS_VER3 && nqnfs)) { 1670 nd->nd_repstat = EPROGMISMATCH; 1671 nd->nd_procnum = NFSPROC_NOOP; 1672 return (0); 1673 } 1674 if (nqnfs) 1675 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1676 else if (nfsvers == NFS_VER3) 1677 nd->nd_flag = ND_NFSV3; 1678 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1679 if (nd->nd_procnum == NFSPROC_NULL) 1680 return (0); 1681 if (nd->nd_procnum >= NFS_NPROCS || 1682 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1683 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1684 nd->nd_repstat = EPROCUNAVAIL; 1685 nd->nd_procnum = NFSPROC_NOOP; 1686 return (0); 1687 } 1688 if ((nd->nd_flag & ND_NFSV3) == 0) 1689 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1690 auth_type = *tl++; 1691 len = fxdr_unsigned(int, *tl++); 1692 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1693 m_freem(mrep); 1694 return (EBADRPC); 1695 } 1696 1697 nd->nd_flag &= ~ND_KERBAUTH; 1698 /* 1699 * Handle auth_unix or auth_kerb. 1700 */ 1701 if (auth_type == rpc_auth_unix) { 1702 len = fxdr_unsigned(int, *++tl); 1703 if (len < 0 || len > NFS_MAXNAMLEN) { 1704 m_freem(mrep); 1705 return (EBADRPC); 1706 } 1707 nfsm_adv(nfsm_rndup(len)); 1708 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1709 memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred)); 1710 nd->nd_cr.cr_ref = 1; 1711 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1712 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1713 len = fxdr_unsigned(int, *tl); 1714 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1715 m_freem(mrep); 1716 return (EBADRPC); 1717 } 1718 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1719 for (i = 0; i < len; i++) 1720 if (i < NGROUPS) 1721 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1722 else 1723 tl++; 1724 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1725 if (nd->nd_cr.cr_ngroups > 1) 1726 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1727 len = fxdr_unsigned(int, *++tl); 1728 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1729 m_freem(mrep); 1730 return (EBADRPC); 1731 } 1732 if (len > 0) 1733 nfsm_adv(nfsm_rndup(len)); 1734 } else if (auth_type == rpc_auth_kerb) { 1735 switch (fxdr_unsigned(int, *tl++)) { 1736 case RPCAKN_FULLNAME: 1737 ticklen = fxdr_unsigned(int, *tl); 1738 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1739 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1740 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1741 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1742 m_freem(mrep); 1743 return (EBADRPC); 1744 } 1745 uio.uio_offset = 0; 1746 uio.uio_iov = &iov; 1747 uio.uio_iovcnt = 1; 1748 uio.uio_segflg = UIO_SYSSPACE; 1749 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1750 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1751 nfsm_mtouio(&uio, uio.uio_resid); 1752 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1753 if (*tl++ != rpc_auth_kerb || 1754 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1755 printf("Bad kerb verifier\n"); 1756 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1757 nd->nd_procnum = NFSPROC_NOOP; 1758 return (0); 1759 } 1760 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1761 tl = (u_int32_t *)cp; 1762 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1763 printf("Not fullname kerb verifier\n"); 1764 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1765 nd->nd_procnum = NFSPROC_NOOP; 1766 return (0); 1767 } 1768 cp += NFSX_UNSIGNED; 1769 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 1770 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1771 nd->nd_flag |= ND_KERBFULL; 1772 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1773 break; 1774 case RPCAKN_NICKNAME: 1775 if (len != 2 * NFSX_UNSIGNED) { 1776 printf("Kerb nickname short\n"); 1777 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1778 nd->nd_procnum = NFSPROC_NOOP; 1779 return (0); 1780 } 1781 nickuid = fxdr_unsigned(uid_t, *tl); 1782 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1783 if (*tl++ != rpc_auth_kerb || 1784 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1785 printf("Kerb nick verifier bad\n"); 1786 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1787 nd->nd_procnum = NFSPROC_NOOP; 1788 return (0); 1789 } 1790 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1791 tvin.tv_sec = *tl++; 1792 tvin.tv_usec = *tl; 1793 1794 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; 1795 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1796 if (nuidp->nu_cr.cr_uid == nickuid && 1797 (!nd->nd_nam2 || 1798 netaddr_match(NU_NETFAM(nuidp), 1799 &nuidp->nu_haddr, nd->nd_nam2))) 1800 break; 1801 } 1802 if (!nuidp) { 1803 nd->nd_repstat = 1804 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1805 nd->nd_procnum = NFSPROC_NOOP; 1806 return (0); 1807 } 1808 1809 /* 1810 * Now, decrypt the timestamp using the session key 1811 * and validate it. 1812 */ 1813 #ifdef NFSKERB 1814 XXX 1815 #endif 1816 1817 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1818 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1819 if (nuidp->nu_expire < time.tv_sec || 1820 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1821 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1822 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1823 nuidp->nu_expire = 0; 1824 nd->nd_repstat = 1825 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1826 nd->nd_procnum = NFSPROC_NOOP; 1827 return (0); 1828 } 1829 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1830 nd->nd_flag |= ND_KERBNICK; 1831 }; 1832 } else { 1833 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1834 nd->nd_procnum = NFSPROC_NOOP; 1835 return (0); 1836 } 1837 1838 /* 1839 * For nqnfs, get piggybacked lease request. 1840 */ 1841 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1842 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1843 nd->nd_flag |= fxdr_unsigned(int, *tl); 1844 if (nd->nd_flag & ND_LEASE) { 1845 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1846 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 1847 } else 1848 nd->nd_duration = NQ_MINLEASE; 1849 } else 1850 nd->nd_duration = NQ_MINLEASE; 1851 nd->nd_md = md; 1852 nd->nd_dpos = dpos; 1853 return (0); 1854 nfsmout: 1855 return (error); 1856 } 1857 1858 int 1859 nfs_msg(p, server, msg) 1860 struct proc *p; 1861 char *server, *msg; 1862 { 1863 tpr_t tpr; 1864 1865 if (p) 1866 tpr = tprintf_open(p); 1867 else 1868 tpr = NULL; 1869 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1870 tprintf_close(tpr); 1871 return (0); 1872 } 1873 1874 #ifdef NFSSERVER 1875 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 1876 struct nfssvc_sock *, struct proc *, 1877 struct mbuf **)) = { 1878 nfsrv_null, 1879 nfsrv_getattr, 1880 nfsrv_setattr, 1881 nfsrv_lookup, 1882 nfsrv3_access, 1883 nfsrv_readlink, 1884 nfsrv_read, 1885 nfsrv_write, 1886 nfsrv_create, 1887 nfsrv_mkdir, 1888 nfsrv_symlink, 1889 nfsrv_mknod, 1890 nfsrv_remove, 1891 nfsrv_rmdir, 1892 nfsrv_rename, 1893 nfsrv_link, 1894 nfsrv_readdir, 1895 nfsrv_readdirplus, 1896 nfsrv_statfs, 1897 nfsrv_fsinfo, 1898 nfsrv_pathconf, 1899 nfsrv_commit, 1900 nqnfsrv_getlease, 1901 nqnfsrv_vacated, 1902 nfsrv_noop, 1903 nfsrv_noop 1904 }; 1905 1906 /* 1907 * Socket upcall routine for the nfsd sockets. 1908 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1909 * Essentially do as much as possible non-blocking, else punt and it will 1910 * be called with M_WAIT from an nfsd. 1911 */ 1912 void 1913 nfsrv_rcv(so, arg, waitflag) 1914 struct socket *so; 1915 caddr_t arg; 1916 int waitflag; 1917 { 1918 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1919 struct mbuf *m; 1920 struct mbuf *mp, *nam; 1921 struct uio auio; 1922 int flags, error; 1923 1924 if ((slp->ns_flag & SLP_VALID) == 0) 1925 return; 1926 #ifdef notdef 1927 /* 1928 * Define this to test for nfsds handling this under heavy load. 1929 */ 1930 if (waitflag == M_DONTWAIT) { 1931 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1932 } 1933 #endif 1934 auio.uio_procp = NULL; 1935 if (so->so_type == SOCK_STREAM) { 1936 /* 1937 * If there are already records on the queue, defer soreceive() 1938 * to an nfsd so that there is feedback to the TCP layer that 1939 * the nfs servers are heavily loaded. 1940 */ 1941 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1942 slp->ns_flag |= SLP_NEEDQ; 1943 goto dorecs; 1944 } 1945 1946 /* 1947 * Do soreceive(). 1948 */ 1949 auio.uio_resid = 1000000000; 1950 flags = MSG_DONTWAIT; 1951 error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1952 if (error || mp == (struct mbuf *)0) { 1953 if (error == EWOULDBLOCK) 1954 slp->ns_flag |= SLP_NEEDQ; 1955 else 1956 slp->ns_flag |= SLP_DISCONN; 1957 goto dorecs; 1958 } 1959 m = mp; 1960 if (slp->ns_rawend) { 1961 slp->ns_rawend->m_next = m; 1962 slp->ns_cc += 1000000000 - auio.uio_resid; 1963 } else { 1964 slp->ns_raw = m; 1965 slp->ns_cc = 1000000000 - auio.uio_resid; 1966 } 1967 while (m->m_next) 1968 m = m->m_next; 1969 slp->ns_rawend = m; 1970 1971 /* 1972 * Now try and parse record(s) out of the raw stream data. 1973 */ 1974 error = nfsrv_getstream(slp, waitflag); 1975 if (error) { 1976 if (error == EPERM) 1977 slp->ns_flag |= SLP_DISCONN; 1978 else 1979 slp->ns_flag |= SLP_NEEDQ; 1980 } 1981 } else { 1982 do { 1983 auio.uio_resid = 1000000000; 1984 flags = MSG_DONTWAIT; 1985 error = (*so->so_receive)(so, &nam, &auio, &mp, 1986 (struct mbuf **)0, &flags); 1987 if (mp) { 1988 if (nam) { 1989 m = nam; 1990 m->m_next = mp; 1991 } else 1992 m = mp; 1993 if (slp->ns_recend) 1994 slp->ns_recend->m_nextpkt = m; 1995 else 1996 slp->ns_rec = m; 1997 slp->ns_recend = m; 1998 m->m_nextpkt = (struct mbuf *)0; 1999 } 2000 if (error) { 2001 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2002 && error != EWOULDBLOCK) { 2003 slp->ns_flag |= SLP_DISCONN; 2004 goto dorecs; 2005 } 2006 } 2007 } while (mp); 2008 } 2009 2010 /* 2011 * Now try and process the request records, non-blocking. 2012 */ 2013 dorecs: 2014 if (waitflag == M_DONTWAIT && 2015 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 2016 nfsrv_wakenfsd(slp); 2017 } 2018 2019 /* 2020 * Try and extract an RPC request from the mbuf data list received on a 2021 * stream socket. The "waitflag" argument indicates whether or not it 2022 * can sleep. 2023 */ 2024 int 2025 nfsrv_getstream(slp, waitflag) 2026 struct nfssvc_sock *slp; 2027 int waitflag; 2028 { 2029 struct mbuf *m, **mpp; 2030 char *cp1, *cp2; 2031 int len; 2032 struct mbuf *om, *m2, *recm = NULL; 2033 u_int32_t recmark; 2034 2035 if (slp->ns_flag & SLP_GETSTREAM) 2036 panic("nfs getstream"); 2037 slp->ns_flag |= SLP_GETSTREAM; 2038 for (;;) { 2039 if (slp->ns_reclen == 0) { 2040 if (slp->ns_cc < NFSX_UNSIGNED) { 2041 slp->ns_flag &= ~SLP_GETSTREAM; 2042 return (0); 2043 } 2044 m = slp->ns_raw; 2045 if (m->m_len >= NFSX_UNSIGNED) { 2046 memcpy((caddr_t)&recmark, mtod(m, caddr_t), NFSX_UNSIGNED); 2047 m->m_data += NFSX_UNSIGNED; 2048 m->m_len -= NFSX_UNSIGNED; 2049 } else { 2050 cp1 = (caddr_t)&recmark; 2051 cp2 = mtod(m, caddr_t); 2052 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 2053 while (m->m_len == 0) { 2054 m = m->m_next; 2055 cp2 = mtod(m, caddr_t); 2056 } 2057 *cp1++ = *cp2++; 2058 m->m_data++; 2059 m->m_len--; 2060 } 2061 } 2062 slp->ns_cc -= NFSX_UNSIGNED; 2063 recmark = ntohl(recmark); 2064 slp->ns_reclen = recmark & ~0x80000000; 2065 if (recmark & 0x80000000) 2066 slp->ns_flag |= SLP_LASTFRAG; 2067 else 2068 slp->ns_flag &= ~SLP_LASTFRAG; 2069 if (slp->ns_reclen > NFS_MAXPACKET) { 2070 slp->ns_flag &= ~SLP_GETSTREAM; 2071 return (EPERM); 2072 } 2073 } 2074 2075 /* 2076 * Now get the record part. 2077 */ 2078 if (slp->ns_cc == slp->ns_reclen) { 2079 recm = slp->ns_raw; 2080 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2081 slp->ns_cc = slp->ns_reclen = 0; 2082 } else if (slp->ns_cc > slp->ns_reclen) { 2083 len = 0; 2084 m = slp->ns_raw; 2085 om = (struct mbuf *)0; 2086 while (len < slp->ns_reclen) { 2087 if ((len + m->m_len) > slp->ns_reclen) { 2088 size_t left = slp->ns_reclen - len; 2089 2090 MGETHDR(m2, waitflag, m->m_type); 2091 if (m2 == NULL) { 2092 slp->ns_flag &= ~SLP_GETSTREAM; 2093 return (EWOULDBLOCK); 2094 } 2095 if (left > MHLEN) { 2096 MCLGET(m2, waitflag); 2097 if (!(m2->m_flags & M_EXT)) { 2098 m_freem(m2); 2099 slp->ns_flag &= ~SLP_GETSTREAM; 2100 return (EWOULDBLOCK); 2101 } 2102 } 2103 memcpy(mtod(m2, caddr_t), mtod(m, caddr_t), 2104 left); 2105 m2->m_len = left; 2106 m->m_data += left; 2107 m->m_len -= left; 2108 if (om) { 2109 om->m_next = m2; 2110 recm = slp->ns_raw; 2111 } else 2112 recm = m2; 2113 len = slp->ns_reclen; 2114 } else if ((len + m->m_len) == slp->ns_reclen) { 2115 om = m; 2116 len += m->m_len; 2117 m = m->m_next; 2118 recm = slp->ns_raw; 2119 om->m_next = (struct mbuf *)0; 2120 } else { 2121 om = m; 2122 len += m->m_len; 2123 m = m->m_next; 2124 } 2125 } 2126 slp->ns_raw = m; 2127 slp->ns_cc -= len; 2128 slp->ns_reclen = 0; 2129 } else { 2130 slp->ns_flag &= ~SLP_GETSTREAM; 2131 return (0); 2132 } 2133 2134 /* 2135 * Accumulate the fragments into a record. 2136 */ 2137 mpp = &slp->ns_frag; 2138 while (*mpp) 2139 mpp = &((*mpp)->m_next); 2140 *mpp = recm; 2141 if (slp->ns_flag & SLP_LASTFRAG) { 2142 if (slp->ns_recend) 2143 slp->ns_recend->m_nextpkt = slp->ns_frag; 2144 else 2145 slp->ns_rec = slp->ns_frag; 2146 slp->ns_recend = slp->ns_frag; 2147 slp->ns_frag = (struct mbuf *)0; 2148 } 2149 } 2150 } 2151 2152 /* 2153 * Parse an RPC header. 2154 */ 2155 int 2156 nfsrv_dorec(slp, nfsd, ndp) 2157 struct nfssvc_sock *slp; 2158 struct nfsd *nfsd; 2159 struct nfsrv_descript **ndp; 2160 { 2161 struct mbuf *m, *nam; 2162 struct nfsrv_descript *nd; 2163 int error; 2164 2165 *ndp = NULL; 2166 if ((slp->ns_flag & SLP_VALID) == 0 || 2167 (m = slp->ns_rec) == (struct mbuf *)0) 2168 return (ENOBUFS); 2169 slp->ns_rec = m->m_nextpkt; 2170 if (slp->ns_rec) 2171 m->m_nextpkt = (struct mbuf *)0; 2172 else 2173 slp->ns_recend = (struct mbuf *)0; 2174 if (m->m_type == MT_SONAME) { 2175 nam = m; 2176 m = m->m_next; 2177 nam->m_next = NULL; 2178 } else 2179 nam = NULL; 2180 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), 2181 M_NFSRVDESC, M_WAITOK); 2182 nd->nd_md = nd->nd_mrep = m; 2183 nd->nd_nam2 = nam; 2184 nd->nd_dpos = mtod(m, caddr_t); 2185 error = nfs_getreq(nd, nfsd, TRUE); 2186 if (error) { 2187 m_freem(nam); 2188 free((caddr_t)nd, M_NFSRVDESC); 2189 return (error); 2190 } 2191 *ndp = nd; 2192 nfsd->nfsd_nd = nd; 2193 return (0); 2194 } 2195 2196 2197 /* 2198 * Search for a sleeping nfsd and wake it up. 2199 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2200 * running nfsds will go look for the work in the nfssvc_sock list. 2201 */ 2202 void 2203 nfsrv_wakenfsd(slp) 2204 struct nfssvc_sock *slp; 2205 { 2206 struct nfsd *nd; 2207 2208 if ((slp->ns_flag & SLP_VALID) == 0) 2209 return; 2210 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) { 2211 if (nd->nfsd_flag & NFSD_WAITING) { 2212 nd->nfsd_flag &= ~NFSD_WAITING; 2213 if (nd->nfsd_slp) 2214 panic("nfsd wakeup"); 2215 slp->ns_sref++; 2216 nd->nfsd_slp = slp; 2217 wakeup((caddr_t)nd); 2218 return; 2219 } 2220 } 2221 slp->ns_flag |= SLP_DOREC; 2222 nfsd_head_flag |= NFSD_CHECKSLP; 2223 } 2224 #endif /* NFSSERVER */ 2225
Indexes created Tue Sep 30 11:09:46 GMT 2025