nfs_socket.c revision 1.66.2.1
1 /* $NetBSD: nfs_socket.c,v 1.66.2.1 2001/03/05 22:49:59 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 log(LOG_INFO, "nfsd send error %d\n", error); 470 471 /* 472 * Handle any recoverable (soft) socket errors here. (? ? ?) 473 */ 474 if (error != EINTR && error != ERESTART && 475 error != EWOULDBLOCK && error != EPIPE) 476 error = 0; 477 } 478 return (error); 479 } 480 481 #ifdef NFS 482 /* 483 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 484 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 485 * Mark and consolidate the data into a new mbuf list. 486 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 487 * small mbufs. 488 * For SOCK_STREAM we must be very careful to read an entire record once 489 * we have read any of it, even if the system call has been interrupted. 490 */ 491 int 492 nfs_receive(rep, aname, mp) 493 struct nfsreq *rep; 494 struct mbuf **aname; 495 struct mbuf **mp; 496 { 497 struct socket *so; 498 struct uio auio; 499 struct iovec aio; 500 struct mbuf *m; 501 struct mbuf *control; 502 u_int32_t len; 503 struct mbuf **getnam; 504 int error, sotype, rcvflg; 505 struct proc *p = curproc->l_proc; /* XXX */ 506 507 /* 508 * Set up arguments for soreceive() 509 */ 510 *mp = (struct mbuf *)0; 511 *aname = (struct mbuf *)0; 512 sotype = rep->r_nmp->nm_sotype; 513 514 /* 515 * For reliable protocols, lock against other senders/receivers 516 * in case a reconnect is necessary. 517 * For SOCK_STREAM, first get the Record Mark to find out how much 518 * more there is to get. 519 * We must lock the socket against other receivers 520 * until we have an entire rpc request/reply. 521 */ 522 if (sotype != SOCK_DGRAM) { 523 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 524 if (error) 525 return (error); 526 tryagain: 527 /* 528 * Check for fatal errors and resending request. 529 */ 530 /* 531 * Ugh: If a reconnect attempt just happened, nm_so 532 * would have changed. NULL indicates a failed 533 * attempt that has essentially shut down this 534 * mount point. 535 */ 536 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 537 nfs_sndunlock(&rep->r_nmp->nm_iflag); 538 return (EINTR); 539 } 540 so = rep->r_nmp->nm_so; 541 if (!so) { 542 error = nfs_reconnect(rep); 543 if (error) { 544 nfs_sndunlock(&rep->r_nmp->nm_iflag); 545 return (error); 546 } 547 goto tryagain; 548 } 549 while (rep->r_flags & R_MUSTRESEND) { 550 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 551 nfsstats.rpcretries++; 552 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 553 if (error) { 554 if (error == EINTR || error == ERESTART || 555 (error = nfs_reconnect(rep)) != 0) { 556 nfs_sndunlock(&rep->r_nmp->nm_iflag); 557 return (error); 558 } 559 goto tryagain; 560 } 561 } 562 nfs_sndunlock(&rep->r_nmp->nm_iflag); 563 if (sotype == SOCK_STREAM) { 564 aio.iov_base = (caddr_t) &len; 565 aio.iov_len = sizeof(u_int32_t); 566 auio.uio_iov = &aio; 567 auio.uio_iovcnt = 1; 568 auio.uio_segflg = UIO_SYSSPACE; 569 auio.uio_rw = UIO_READ; 570 auio.uio_offset = 0; 571 auio.uio_resid = sizeof(u_int32_t); 572 auio.uio_procp = p; 573 do { 574 rcvflg = MSG_WAITALL; 575 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 576 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 577 if (error == EWOULDBLOCK && rep) { 578 if (rep->r_flags & R_SOFTTERM) 579 return (EINTR); 580 } 581 } while (error == EWOULDBLOCK); 582 if (!error && auio.uio_resid > 0) { 583 /* 584 * Don't log a 0 byte receive; it means 585 * that the socket has been closed, and 586 * can happen during normal operation 587 * (forcible unmount or Solaris server). 588 */ 589 if (auio.uio_resid != sizeof (u_int32_t)) 590 log(LOG_INFO, 591 "short receive (%lu/%lu) from nfs server %s\n", 592 (u_long)sizeof(u_int32_t) - auio.uio_resid, 593 (u_long)sizeof(u_int32_t), 594 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 595 error = EPIPE; 596 } 597 if (error) 598 goto errout; 599 len = ntohl(len) & ~0x80000000; 600 /* 601 * This is SERIOUS! We are out of sync with the sender 602 * and forcing a disconnect/reconnect is all I can do. 603 */ 604 if (len > NFS_MAXPACKET) { 605 log(LOG_ERR, "%s (%d) from nfs server %s\n", 606 "impossible packet length", 607 len, 608 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 609 error = EFBIG; 610 goto errout; 611 } 612 auio.uio_resid = len; 613 do { 614 rcvflg = MSG_WAITALL; 615 error = (*so->so_receive)(so, (struct mbuf **)0, 616 &auio, mp, (struct mbuf **)0, &rcvflg); 617 } while (error == EWOULDBLOCK || error == EINTR || 618 error == ERESTART); 619 if (!error && auio.uio_resid > 0) { 620 if (len != auio.uio_resid) 621 log(LOG_INFO, 622 "short receive (%lu/%d) from nfs server %s\n", 623 (u_long)len - auio.uio_resid, len, 624 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 625 error = EPIPE; 626 } 627 } else { 628 /* 629 * NB: Since uio_resid is big, MSG_WAITALL is ignored 630 * and soreceive() will return when it has either a 631 * control msg or a data msg. 632 * We have no use for control msg., but must grab them 633 * and then throw them away so we know what is going 634 * on. 635 */ 636 auio.uio_resid = len = 100000000; /* Anything Big */ 637 auio.uio_procp = p; 638 do { 639 rcvflg = 0; 640 error = (*so->so_receive)(so, (struct mbuf **)0, 641 &auio, mp, &control, &rcvflg); 642 if (control) 643 m_freem(control); 644 if (error == EWOULDBLOCK && rep) { 645 if (rep->r_flags & R_SOFTTERM) 646 return (EINTR); 647 } 648 } while (error == EWOULDBLOCK || 649 (!error && *mp == NULL && control)); 650 if ((rcvflg & MSG_EOR) == 0) 651 printf("Egad!!\n"); 652 if (!error && *mp == NULL) 653 error = EPIPE; 654 len -= auio.uio_resid; 655 } 656 errout: 657 if (error && error != EINTR && error != ERESTART) { 658 m_freem(*mp); 659 *mp = (struct mbuf *)0; 660 if (error != EPIPE) 661 log(LOG_INFO, 662 "receive error %d from nfs server %s\n", 663 error, 664 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 665 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 666 if (!error) 667 error = nfs_reconnect(rep); 668 if (!error) 669 goto tryagain; 670 else 671 nfs_sndunlock(&rep->r_nmp->nm_iflag); 672 } 673 } else { 674 if ((so = rep->r_nmp->nm_so) == NULL) 675 return (EACCES); 676 if (so->so_state & SS_ISCONNECTED) 677 getnam = (struct mbuf **)0; 678 else 679 getnam = aname; 680 auio.uio_resid = len = 1000000; 681 auio.uio_procp = p; 682 do { 683 rcvflg = 0; 684 error = (*so->so_receive)(so, getnam, &auio, mp, 685 (struct mbuf **)0, &rcvflg); 686 if (error == EWOULDBLOCK && 687 (rep->r_flags & R_SOFTTERM)) 688 return (EINTR); 689 } while (error == EWOULDBLOCK); 690 len -= auio.uio_resid; 691 if (!error && *mp == NULL) 692 error = EPIPE; 693 } 694 if (error) { 695 m_freem(*mp); 696 *mp = (struct mbuf *)0; 697 } 698 return (error); 699 } 700 701 /* 702 * Implement receipt of reply on a socket. 703 * We must search through the list of received datagrams matching them 704 * with outstanding requests using the xid, until ours is found. 705 */ 706 /* ARGSUSED */ 707 int 708 nfs_reply(myrep) 709 struct nfsreq *myrep; 710 { 711 struct nfsreq *rep; 712 struct nfsmount *nmp = myrep->r_nmp; 713 int32_t t1; 714 struct mbuf *mrep, *nam, *md; 715 u_int32_t rxid, *tl; 716 caddr_t dpos, cp2; 717 int error; 718 719 /* 720 * Loop around until we get our own reply 721 */ 722 for (;;) { 723 /* 724 * Lock against other receivers so that I don't get stuck in 725 * sbwait() after someone else has received my reply for me. 726 * Also necessary for connection based protocols to avoid 727 * race conditions during a reconnect. 728 */ 729 error = nfs_rcvlock(myrep); 730 if (error == EALREADY) 731 return (0); 732 if (error) 733 return (error); 734 /* 735 * Get the next Rpc reply off the socket 736 */ 737 nmp->nm_waiters++; 738 error = nfs_receive(myrep, &nam, &mrep); 739 nfs_rcvunlock(&nmp->nm_iflag); 740 if (error) { 741 742 if (nmp->nm_iflag & NFSMNT_DISMNT) { 743 /* 744 * Oops, we're going away now.. 745 */ 746 nmp->nm_waiters--; 747 wakeup (&nmp->nm_waiters); 748 return error; 749 } 750 nmp->nm_waiters--; 751 /* 752 * Ignore routing errors on connectionless protocols? ? 753 */ 754 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 755 nmp->nm_so->so_error = 0; 756 #ifdef DEBUG 757 printf("nfs_reply: ignoring error %d\n", error); 758 #endif 759 if (myrep->r_flags & R_GETONEREP) 760 return (0); 761 continue; 762 } 763 return (error); 764 } 765 nmp->nm_waiters--; 766 if (nam) 767 m_freem(nam); 768 769 /* 770 * Get the xid and check that it is an rpc reply 771 */ 772 md = mrep; 773 dpos = mtod(md, caddr_t); 774 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 775 rxid = *tl++; 776 if (*tl != rpc_reply) { 777 #ifndef NFS_V2_ONLY 778 if (nmp->nm_flag & NFSMNT_NQNFS) { 779 if (nqnfs_callback(nmp, mrep, md, dpos)) 780 nfsstats.rpcinvalid++; 781 } else 782 #endif 783 { 784 nfsstats.rpcinvalid++; 785 m_freem(mrep); 786 } 787 nfsmout: 788 if (myrep->r_flags & R_GETONEREP) 789 return (0); 790 continue; 791 } 792 793 /* 794 * Loop through the request list to match up the reply 795 * Iff no match, just drop the datagram 796 */ 797 for (rep = nfs_reqq.tqh_first; rep != 0; 798 rep = rep->r_chain.tqe_next) { 799 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 800 /* Found it.. */ 801 rep->r_mrep = mrep; 802 rep->r_md = md; 803 rep->r_dpos = dpos; 804 if (nfsrtton) { 805 struct rttl *rt; 806 807 rt = &nfsrtt.rttl[nfsrtt.pos]; 808 rt->proc = rep->r_procnum; 809 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 810 rt->sent = nmp->nm_sent; 811 rt->cwnd = nmp->nm_cwnd; 812 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 813 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 814 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 815 rt->tstamp = time; 816 if (rep->r_flags & R_TIMING) 817 rt->rtt = rep->r_rtt; 818 else 819 rt->rtt = 1000000; 820 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 821 } 822 /* 823 * Update congestion window. 824 * Do the additive increase of 825 * one rpc/rtt. 826 */ 827 if (nmp->nm_cwnd <= nmp->nm_sent) { 828 nmp->nm_cwnd += 829 (NFS_CWNDSCALE * NFS_CWNDSCALE + 830 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 831 if (nmp->nm_cwnd > NFS_MAXCWND) 832 nmp->nm_cwnd = NFS_MAXCWND; 833 } 834 rep->r_flags &= ~R_SENT; 835 nmp->nm_sent -= NFS_CWNDSCALE; 836 /* 837 * Update rtt using a gain of 0.125 on the mean 838 * and a gain of 0.25 on the deviation. 839 */ 840 if (rep->r_flags & R_TIMING) { 841 /* 842 * Since the timer resolution of 843 * NFS_HZ is so course, it can often 844 * result in r_rtt == 0. Since 845 * r_rtt == N means that the actual 846 * rtt is between N+dt and N+2-dt ticks, 847 * add 1. 848 */ 849 t1 = rep->r_rtt + 1; 850 t1 -= (NFS_SRTT(rep) >> 3); 851 NFS_SRTT(rep) += t1; 852 if (t1 < 0) 853 t1 = -t1; 854 t1 -= (NFS_SDRTT(rep) >> 2); 855 NFS_SDRTT(rep) += t1; 856 } 857 nmp->nm_timeouts = 0; 858 break; 859 } 860 } 861 /* 862 * If not matched to a request, drop it. 863 * If it's mine, get out. 864 */ 865 if (rep == 0) { 866 nfsstats.rpcunexpected++; 867 m_freem(mrep); 868 } else if (rep == myrep) { 869 if (rep->r_mrep == NULL) 870 panic("nfsreply nil"); 871 return (0); 872 } 873 if (myrep->r_flags & R_GETONEREP) 874 return (0); 875 } 876 } 877 878 /* 879 * nfs_request - goes something like this 880 * - fill in request struct 881 * - links it into list 882 * - calls nfs_send() for first transmit 883 * - calls nfs_receive() to get reply 884 * - break down rpc header and return with nfs reply pointed to 885 * by mrep or error 886 * nb: always frees up mreq mbuf list 887 */ 888 int 889 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 890 struct vnode *vp; 891 struct mbuf *mrest; 892 int procnum; 893 struct proc *procp; 894 struct ucred *cred; 895 struct mbuf **mrp; 896 struct mbuf **mdp; 897 caddr_t *dposp; 898 { 899 struct mbuf *m, *mrep; 900 struct nfsreq *rep; 901 u_int32_t *tl; 902 int i; 903 struct nfsmount *nmp; 904 struct mbuf *md, *mheadend; 905 char nickv[RPCX_NICKVERF]; 906 time_t reqtime, waituntil; 907 caddr_t dpos, cp2; 908 int t1, s, error = 0, mrest_len, auth_len, auth_type; 909 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 910 int verf_len, verf_type; 911 u_int32_t xid; 912 char *auth_str, *verf_str; 913 NFSKERBKEY_T key; /* save session key */ 914 #ifndef NFS_V2_ONLY 915 int nqlflag, cachable; 916 u_quad_t frev; 917 struct nfsnode *np; 918 #endif 919 920 KASSERT(cred != NULL); 921 nmp = VFSTONFS(vp->v_mount); 922 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 923 rep->r_nmp = nmp; 924 rep->r_vp = vp; 925 rep->r_procp = procp; 926 rep->r_procnum = procnum; 927 i = 0; 928 m = mrest; 929 while (m) { 930 i += m->m_len; 931 m = m->m_next; 932 } 933 mrest_len = i; 934 935 /* 936 * Get the RPC header with authorization. 937 */ 938 kerbauth: 939 verf_str = auth_str = (char *)0; 940 if (nmp->nm_flag & NFSMNT_KERB) { 941 verf_str = nickv; 942 verf_len = sizeof (nickv); 943 auth_type = RPCAUTH_KERB4; 944 memset((caddr_t)key, 0, sizeof (key)); 945 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 946 &auth_len, verf_str, verf_len)) { 947 error = nfs_getauth(nmp, rep, cred, &auth_str, 948 &auth_len, verf_str, &verf_len, key); 949 if (error) { 950 free((caddr_t)rep, M_NFSREQ); 951 m_freem(mrest); 952 return (error); 953 } 954 } 955 } else { 956 auth_type = RPCAUTH_UNIX; 957 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 958 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 959 5 * NFSX_UNSIGNED; 960 } 961 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 962 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 963 if (auth_str) 964 free(auth_str, M_TEMP); 965 966 /* 967 * For stream protocols, insert a Sun RPC Record Mark. 968 */ 969 if (nmp->nm_sotype == SOCK_STREAM) { 970 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 971 *mtod(m, u_int32_t *) = htonl(0x80000000 | 972 (m->m_pkthdr.len - NFSX_UNSIGNED)); 973 } 974 rep->r_mreq = m; 975 rep->r_xid = xid; 976 tryagain: 977 if (nmp->nm_flag & NFSMNT_SOFT) 978 rep->r_retry = nmp->nm_retry; 979 else 980 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 981 rep->r_rtt = rep->r_rexmit = 0; 982 if (proct[procnum] > 0) 983 rep->r_flags = R_TIMING; 984 else 985 rep->r_flags = 0; 986 rep->r_mrep = NULL; 987 988 /* 989 * Do the client side RPC. 990 */ 991 nfsstats.rpcrequests++; 992 /* 993 * Chain request into list of outstanding requests. Be sure 994 * to put it LAST so timer finds oldest requests first. 995 */ 996 s = splsoftnet(); 997 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 998 999 /* Get send time for nqnfs */ 1000 reqtime = time.tv_sec; 1001 1002 /* 1003 * If backing off another request or avoiding congestion, don't 1004 * send this one now but let timer do it. If not timing a request, 1005 * do it now. 1006 */ 1007 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1008 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1009 nmp->nm_sent < nmp->nm_cwnd)) { 1010 splx(s); 1011 if (nmp->nm_soflags & PR_CONNREQUIRED) 1012 error = nfs_sndlock(&nmp->nm_iflag, rep); 1013 if (!error) { 1014 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1015 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 1016 if (nmp->nm_soflags & PR_CONNREQUIRED) 1017 nfs_sndunlock(&nmp->nm_iflag); 1018 } 1019 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1020 nmp->nm_sent += NFS_CWNDSCALE; 1021 rep->r_flags |= R_SENT; 1022 } 1023 } else { 1024 splx(s); 1025 rep->r_rtt = -1; 1026 } 1027 1028 /* 1029 * Wait for the reply from our send or the timer's. 1030 */ 1031 if (!error || error == EPIPE) 1032 error = nfs_reply(rep); 1033 1034 /* 1035 * RPC done, unlink the request. 1036 */ 1037 s = splsoftnet(); 1038 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1039 splx(s); 1040 1041 /* 1042 * Decrement the outstanding request count. 1043 */ 1044 if (rep->r_flags & R_SENT) { 1045 rep->r_flags &= ~R_SENT; /* paranoia */ 1046 nmp->nm_sent -= NFS_CWNDSCALE; 1047 } 1048 1049 /* 1050 * If there was a successful reply and a tprintf msg. 1051 * tprintf a response. 1052 */ 1053 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1054 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1055 "is alive again"); 1056 mrep = rep->r_mrep; 1057 md = rep->r_md; 1058 dpos = rep->r_dpos; 1059 if (error) { 1060 m_freem(rep->r_mreq); 1061 free((caddr_t)rep, M_NFSREQ); 1062 return (error); 1063 } 1064 1065 /* 1066 * break down the rpc header and check if ok 1067 */ 1068 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1069 if (*tl++ == rpc_msgdenied) { 1070 if (*tl == rpc_mismatch) 1071 error = EOPNOTSUPP; 1072 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1073 if (!failed_auth) { 1074 failed_auth++; 1075 mheadend->m_next = (struct mbuf *)0; 1076 m_freem(mrep); 1077 m_freem(rep->r_mreq); 1078 goto kerbauth; 1079 } else 1080 error = EAUTH; 1081 } else 1082 error = EACCES; 1083 m_freem(mrep); 1084 m_freem(rep->r_mreq); 1085 free((caddr_t)rep, M_NFSREQ); 1086 return (error); 1087 } 1088 1089 /* 1090 * Grab any Kerberos verifier, otherwise just throw it away. 1091 */ 1092 verf_type = fxdr_unsigned(int, *tl++); 1093 i = fxdr_unsigned(int32_t, *tl); 1094 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1095 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1096 if (error) 1097 goto nfsmout; 1098 } else if (i > 0) 1099 nfsm_adv(nfsm_rndup(i)); 1100 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1101 /* 0 == ok */ 1102 if (*tl == 0) { 1103 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1104 if (*tl != 0) { 1105 error = fxdr_unsigned(int, *tl); 1106 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1107 error == NFSERR_TRYLATER) { 1108 m_freem(mrep); 1109 error = 0; 1110 waituntil = time.tv_sec + trylater_delay; 1111 while (time.tv_sec < waituntil) 1112 (void) tsleep((caddr_t)&lbolt, 1113 PSOCK, "nqnfstry", 0); 1114 trylater_delay *= nfs_backoff[trylater_cnt]; 1115 if (trylater_cnt < 7) 1116 trylater_cnt++; 1117 goto tryagain; 1118 } 1119 1120 /* 1121 * If the File Handle was stale, invalidate the 1122 * lookup cache, just in case. 1123 */ 1124 if (error == ESTALE) 1125 cache_purge(vp); 1126 if (nmp->nm_flag & NFSMNT_NFSV3) { 1127 *mrp = mrep; 1128 *mdp = md; 1129 *dposp = dpos; 1130 error |= NFSERR_RETERR; 1131 } else 1132 m_freem(mrep); 1133 m_freem(rep->r_mreq); 1134 free((caddr_t)rep, M_NFSREQ); 1135 return (error); 1136 } 1137 1138 #ifndef NFS_V2_ONLY 1139 /* 1140 * For nqnfs, get any lease in reply 1141 */ 1142 if (nmp->nm_flag & NFSMNT_NQNFS) { 1143 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1144 if (*tl) { 1145 np = VTONFS(vp); 1146 nqlflag = fxdr_unsigned(int, *tl); 1147 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1148 cachable = fxdr_unsigned(int, *tl++); 1149 reqtime += fxdr_unsigned(int, *tl++); 1150 if (reqtime > time.tv_sec) { 1151 frev = fxdr_hyper(tl); 1152 nqnfs_clientlease(nmp, np, nqlflag, 1153 cachable, reqtime, frev); 1154 } 1155 } 1156 } 1157 #endif 1158 *mrp = mrep; 1159 *mdp = md; 1160 *dposp = dpos; 1161 m_freem(rep->r_mreq); 1162 FREE((caddr_t)rep, M_NFSREQ); 1163 return (0); 1164 } 1165 m_freem(mrep); 1166 error = EPROTONOSUPPORT; 1167 nfsmout: 1168 m_freem(rep->r_mreq); 1169 free((caddr_t)rep, M_NFSREQ); 1170 return (error); 1171 } 1172 #endif /* NFS */ 1173 1174 /* 1175 * Generate the rpc reply header 1176 * siz arg. is used to decide if adding a cluster is worthwhile 1177 */ 1178 int 1179 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1180 int siz; 1181 struct nfsrv_descript *nd; 1182 struct nfssvc_sock *slp; 1183 int err; 1184 int cache; 1185 u_quad_t *frev; 1186 struct mbuf **mrq; 1187 struct mbuf **mbp; 1188 caddr_t *bposp; 1189 { 1190 u_int32_t *tl; 1191 struct mbuf *mreq; 1192 caddr_t bpos; 1193 struct mbuf *mb, *mb2; 1194 1195 MGETHDR(mreq, M_WAIT, MT_DATA); 1196 mb = mreq; 1197 /* 1198 * If this is a big reply, use a cluster else 1199 * try and leave leading space for the lower level headers. 1200 */ 1201 siz += RPC_REPLYSIZ; 1202 if (siz >= max_datalen) { 1203 MCLGET(mreq, M_WAIT); 1204 } else 1205 mreq->m_data += max_hdr; 1206 tl = mtod(mreq, u_int32_t *); 1207 mreq->m_len = 6 * NFSX_UNSIGNED; 1208 bpos = ((caddr_t)tl) + mreq->m_len; 1209 *tl++ = txdr_unsigned(nd->nd_retxid); 1210 *tl++ = rpc_reply; 1211 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1212 *tl++ = rpc_msgdenied; 1213 if (err & NFSERR_AUTHERR) { 1214 *tl++ = rpc_autherr; 1215 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1216 mreq->m_len -= NFSX_UNSIGNED; 1217 bpos -= NFSX_UNSIGNED; 1218 } else { 1219 *tl++ = rpc_mismatch; 1220 *tl++ = txdr_unsigned(RPC_VER2); 1221 *tl = txdr_unsigned(RPC_VER2); 1222 } 1223 } else { 1224 *tl++ = rpc_msgaccepted; 1225 1226 /* 1227 * For Kerberos authentication, we must send the nickname 1228 * verifier back, otherwise just RPCAUTH_NULL. 1229 */ 1230 if (nd->nd_flag & ND_KERBFULL) { 1231 struct nfsuid *nuidp; 1232 struct timeval ktvin, ktvout; 1233 1234 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; 1235 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1236 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1237 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), 1238 &nuidp->nu_haddr, nd->nd_nam2))) 1239 break; 1240 } 1241 if (nuidp) { 1242 ktvin.tv_sec = 1243 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); 1244 ktvin.tv_usec = 1245 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1246 1247 /* 1248 * Encrypt the timestamp in ecb mode using the 1249 * session key. 1250 */ 1251 #ifdef NFSKERB 1252 XXX 1253 #endif 1254 1255 *tl++ = rpc_auth_kerb; 1256 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1257 *tl = ktvout.tv_sec; 1258 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1259 *tl++ = ktvout.tv_usec; 1260 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1261 } else { 1262 *tl++ = 0; 1263 *tl++ = 0; 1264 } 1265 } else { 1266 *tl++ = 0; 1267 *tl++ = 0; 1268 } 1269 switch (err) { 1270 case EPROGUNAVAIL: 1271 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1272 break; 1273 case EPROGMISMATCH: 1274 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1275 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1276 if (nd->nd_flag & ND_NQNFS) { 1277 *tl++ = txdr_unsigned(3); 1278 *tl = txdr_unsigned(3); 1279 } else { 1280 *tl++ = txdr_unsigned(2); 1281 *tl = txdr_unsigned(3); 1282 } 1283 break; 1284 case EPROCUNAVAIL: 1285 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1286 break; 1287 case EBADRPC: 1288 *tl = txdr_unsigned(RPC_GARBAGE); 1289 break; 1290 default: 1291 *tl = 0; 1292 if (err != NFSERR_RETVOID) { 1293 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1294 if (err) 1295 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1296 else 1297 *tl = 0; 1298 } 1299 break; 1300 }; 1301 } 1302 1303 /* 1304 * For nqnfs, piggyback lease as requested. 1305 */ 1306 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1307 if (nd->nd_flag & ND_LEASE) { 1308 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1309 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1310 *tl++ = txdr_unsigned(cache); 1311 *tl++ = txdr_unsigned(nd->nd_duration); 1312 txdr_hyper(*frev, tl); 1313 } else { 1314 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1315 *tl = 0; 1316 } 1317 } 1318 if (mrq != NULL) 1319 *mrq = mreq; 1320 *mbp = mb; 1321 *bposp = bpos; 1322 if (err != 0 && err != NFSERR_RETVOID) 1323 nfsstats.srvrpc_errs++; 1324 return (0); 1325 } 1326 1327 /* 1328 * Nfs timer routine 1329 * Scan the nfsreq list and retranmit any requests that have timed out 1330 * To avoid retransmission attempts on STREAM sockets (in the future) make 1331 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1332 */ 1333 void 1334 nfs_timer(arg) 1335 void *arg; /* never used */ 1336 { 1337 struct nfsreq *rep; 1338 struct mbuf *m; 1339 struct socket *so; 1340 struct nfsmount *nmp; 1341 int timeo; 1342 int s, error; 1343 #ifdef NFSSERVER 1344 struct nfssvc_sock *slp; 1345 static long lasttime = 0; 1346 u_quad_t cur_usec; 1347 #endif 1348 1349 s = splsoftnet(); 1350 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1351 nmp = rep->r_nmp; 1352 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1353 continue; 1354 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1355 rep->r_flags |= R_SOFTTERM; 1356 continue; 1357 } 1358 if (rep->r_rtt >= 0) { 1359 rep->r_rtt++; 1360 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1361 timeo = nmp->nm_timeo; 1362 else 1363 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1364 if (nmp->nm_timeouts > 0) 1365 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1366 if (rep->r_rtt <= timeo) 1367 continue; 1368 if (nmp->nm_timeouts < 8) 1369 nmp->nm_timeouts++; 1370 } 1371 /* 1372 * Check for server not responding 1373 */ 1374 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1375 rep->r_rexmit > nmp->nm_deadthresh) { 1376 nfs_msg(rep->r_procp, 1377 nmp->nm_mountp->mnt_stat.f_mntfromname, 1378 "not responding"); 1379 rep->r_flags |= R_TPRINTFMSG; 1380 } 1381 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1382 nfsstats.rpctimeouts++; 1383 rep->r_flags |= R_SOFTTERM; 1384 continue; 1385 } 1386 if (nmp->nm_sotype != SOCK_DGRAM) { 1387 if (++rep->r_rexmit > NFS_MAXREXMIT) 1388 rep->r_rexmit = NFS_MAXREXMIT; 1389 continue; 1390 } 1391 if ((so = nmp->nm_so) == NULL) 1392 continue; 1393 1394 /* 1395 * If there is enough space and the window allows.. 1396 * Resend it 1397 * Set r_rtt to -1 in case we fail to send it now. 1398 */ 1399 rep->r_rtt = -1; 1400 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1401 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1402 (rep->r_flags & R_SENT) || 1403 nmp->nm_sent < nmp->nm_cwnd) && 1404 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1405 if (so->so_state & SS_ISCONNECTED) 1406 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1407 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1408 else 1409 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1410 nmp->nm_nam, (struct mbuf *)0, (struct proc *)0); 1411 if (error) { 1412 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1413 #ifdef DEBUG 1414 printf("nfs_timer: ignoring error %d\n", 1415 error); 1416 #endif 1417 so->so_error = 0; 1418 } 1419 } else { 1420 /* 1421 * Iff first send, start timing 1422 * else turn timing off, backoff timer 1423 * and divide congestion window by 2. 1424 */ 1425 if (rep->r_flags & R_SENT) { 1426 rep->r_flags &= ~R_TIMING; 1427 if (++rep->r_rexmit > NFS_MAXREXMIT) 1428 rep->r_rexmit = NFS_MAXREXMIT; 1429 nmp->nm_cwnd >>= 1; 1430 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1431 nmp->nm_cwnd = NFS_CWNDSCALE; 1432 nfsstats.rpcretries++; 1433 } else { 1434 rep->r_flags |= R_SENT; 1435 nmp->nm_sent += NFS_CWNDSCALE; 1436 } 1437 rep->r_rtt = 0; 1438 } 1439 } 1440 } 1441 1442 #ifdef NFSSERVER 1443 /* 1444 * Call the nqnfs server timer once a second to handle leases. 1445 */ 1446 if (lasttime != time.tv_sec) { 1447 lasttime = time.tv_sec; 1448 nqnfs_serverd(); 1449 } 1450 1451 /* 1452 * Scan the write gathering queues for writes that need to be 1453 * completed now. 1454 */ 1455 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1456 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 1457 slp = slp->ns_chain.tqe_next) { 1458 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec) 1459 nfsrv_wakenfsd(slp); 1460 } 1461 #endif /* NFSSERVER */ 1462 splx(s); 1463 callout_reset(&nfs_timer_ch, nfs_ticks, nfs_timer, NULL); 1464 } 1465 1466 /* 1467 * Test for a termination condition pending on the process. 1468 * This is used for NFSMNT_INT mounts. 1469 */ 1470 int 1471 nfs_sigintr(nmp, rep, p) 1472 struct nfsmount *nmp; 1473 struct nfsreq *rep; 1474 struct proc *p; 1475 { 1476 sigset_t ss; 1477 1478 if (rep && (rep->r_flags & R_SOFTTERM)) 1479 return (EINTR); 1480 if (!(nmp->nm_flag & NFSMNT_INT)) 1481 return (0); 1482 if (p) { 1483 sigpending1(p, &ss); 1484 #if 0 1485 sigminusset(&p->p_sigctx.ps_sigignore, &ss); 1486 #endif 1487 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1488 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1489 sigismember(&ss, SIGQUIT)) 1490 return (EINTR); 1491 } 1492 return (0); 1493 } 1494 1495 /* 1496 * Lock a socket against others. 1497 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1498 * and also to avoid race conditions between the processes with nfs requests 1499 * in progress when a reconnect is necessary. 1500 */ 1501 int 1502 nfs_sndlock(flagp, rep) 1503 int *flagp; 1504 struct nfsreq *rep; 1505 { 1506 struct proc *p; 1507 int slpflag = 0, slptimeo = 0; 1508 1509 if (rep) { 1510 p = rep->r_procp; 1511 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1512 slpflag = PCATCH; 1513 } else 1514 p = (struct proc *)0; 1515 while (*flagp & NFSMNT_SNDLOCK) { 1516 if (nfs_sigintr(rep->r_nmp, rep, p)) 1517 return (EINTR); 1518 *flagp |= NFSMNT_WANTSND; 1519 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1520 slptimeo); 1521 if (slpflag == PCATCH) { 1522 slpflag = 0; 1523 slptimeo = 2 * hz; 1524 } 1525 } 1526 *flagp |= NFSMNT_SNDLOCK; 1527 return (0); 1528 } 1529 1530 /* 1531 * Unlock the stream socket for others. 1532 */ 1533 void 1534 nfs_sndunlock(flagp) 1535 int *flagp; 1536 { 1537 1538 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1539 panic("nfs sndunlock"); 1540 *flagp &= ~NFSMNT_SNDLOCK; 1541 if (*flagp & NFSMNT_WANTSND) { 1542 *flagp &= ~NFSMNT_WANTSND; 1543 wakeup((caddr_t)flagp); 1544 } 1545 } 1546 1547 int 1548 nfs_rcvlock(rep) 1549 struct nfsreq *rep; 1550 { 1551 struct nfsmount *nmp = rep->r_nmp; 1552 int *flagp = &nmp->nm_iflag; 1553 int slpflag, slptimeo = 0; 1554 1555 if (*flagp & NFSMNT_DISMNT) 1556 return EIO; 1557 1558 if (*flagp & NFSMNT_INT) 1559 slpflag = PCATCH; 1560 else 1561 slpflag = 0; 1562 while (*flagp & NFSMNT_RCVLOCK) { 1563 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1564 return (EINTR); 1565 *flagp |= NFSMNT_WANTRCV; 1566 nmp->nm_waiters++; 1567 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1568 slptimeo); 1569 nmp->nm_waiters--; 1570 if (*flagp & NFSMNT_DISMNT) { 1571 wakeup(&nmp->nm_waiters); 1572 return EIO; 1573 } 1574 /* If our reply was received while we were sleeping, 1575 * then just return without taking the lock to avoid a 1576 * situation where a single iod could 'capture' the 1577 * receive lock. 1578 */ 1579 if (rep->r_mrep != NULL) 1580 return (EALREADY); 1581 if (slpflag == PCATCH) { 1582 slpflag = 0; 1583 slptimeo = 2 * hz; 1584 } 1585 } 1586 *flagp |= NFSMNT_RCVLOCK; 1587 return (0); 1588 } 1589 1590 /* 1591 * Unlock the stream socket for others. 1592 */ 1593 void 1594 nfs_rcvunlock(flagp) 1595 int *flagp; 1596 { 1597 1598 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1599 panic("nfs rcvunlock"); 1600 *flagp &= ~NFSMNT_RCVLOCK; 1601 if (*flagp & NFSMNT_WANTRCV) { 1602 *flagp &= ~NFSMNT_WANTRCV; 1603 wakeup((caddr_t)flagp); 1604 } 1605 } 1606 1607 /* 1608 * Parse an RPC request 1609 * - verify it 1610 * - fill in the cred struct. 1611 */ 1612 int 1613 nfs_getreq(nd, nfsd, has_header) 1614 struct nfsrv_descript *nd; 1615 struct nfsd *nfsd; 1616 int has_header; 1617 { 1618 int len, i; 1619 u_int32_t *tl; 1620 int32_t t1; 1621 struct uio uio; 1622 struct iovec iov; 1623 caddr_t dpos, cp2, cp; 1624 u_int32_t nfsvers, auth_type; 1625 uid_t nickuid; 1626 int error = 0, nqnfs = 0, ticklen; 1627 struct mbuf *mrep, *md; 1628 struct nfsuid *nuidp; 1629 struct timeval tvin, tvout; 1630 1631 mrep = nd->nd_mrep; 1632 md = nd->nd_md; 1633 dpos = nd->nd_dpos; 1634 if (has_header) { 1635 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1636 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1637 if (*tl++ != rpc_call) { 1638 m_freem(mrep); 1639 return (EBADRPC); 1640 } 1641 } else 1642 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1643 nd->nd_repstat = 0; 1644 nd->nd_flag = 0; 1645 if (*tl++ != rpc_vers) { 1646 nd->nd_repstat = ERPCMISMATCH; 1647 nd->nd_procnum = NFSPROC_NOOP; 1648 return (0); 1649 } 1650 if (*tl != nfs_prog) { 1651 if (*tl == nqnfs_prog) 1652 nqnfs++; 1653 else { 1654 nd->nd_repstat = EPROGUNAVAIL; 1655 nd->nd_procnum = NFSPROC_NOOP; 1656 return (0); 1657 } 1658 } 1659 tl++; 1660 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1661 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1662 (nfsvers != NQNFS_VER3 && nqnfs)) { 1663 nd->nd_repstat = EPROGMISMATCH; 1664 nd->nd_procnum = NFSPROC_NOOP; 1665 return (0); 1666 } 1667 if (nqnfs) 1668 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1669 else if (nfsvers == NFS_VER3) 1670 nd->nd_flag = ND_NFSV3; 1671 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1672 if (nd->nd_procnum == NFSPROC_NULL) 1673 return (0); 1674 if (nd->nd_procnum >= NFS_NPROCS || 1675 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1676 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1677 nd->nd_repstat = EPROCUNAVAIL; 1678 nd->nd_procnum = NFSPROC_NOOP; 1679 return (0); 1680 } 1681 if ((nd->nd_flag & ND_NFSV3) == 0) 1682 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1683 auth_type = *tl++; 1684 len = fxdr_unsigned(int, *tl++); 1685 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1686 m_freem(mrep); 1687 return (EBADRPC); 1688 } 1689 1690 nd->nd_flag &= ~ND_KERBAUTH; 1691 /* 1692 * Handle auth_unix or auth_kerb. 1693 */ 1694 if (auth_type == rpc_auth_unix) { 1695 len = fxdr_unsigned(int, *++tl); 1696 if (len < 0 || len > NFS_MAXNAMLEN) { 1697 m_freem(mrep); 1698 return (EBADRPC); 1699 } 1700 nfsm_adv(nfsm_rndup(len)); 1701 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1702 memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred)); 1703 nd->nd_cr.cr_ref = 1; 1704 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1705 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1706 len = fxdr_unsigned(int, *tl); 1707 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1708 m_freem(mrep); 1709 return (EBADRPC); 1710 } 1711 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1712 for (i = 0; i < len; i++) 1713 if (i < NGROUPS) 1714 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1715 else 1716 tl++; 1717 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1718 if (nd->nd_cr.cr_ngroups > 1) 1719 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1720 len = fxdr_unsigned(int, *++tl); 1721 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1722 m_freem(mrep); 1723 return (EBADRPC); 1724 } 1725 if (len > 0) 1726 nfsm_adv(nfsm_rndup(len)); 1727 } else if (auth_type == rpc_auth_kerb) { 1728 switch (fxdr_unsigned(int, *tl++)) { 1729 case RPCAKN_FULLNAME: 1730 ticklen = fxdr_unsigned(int, *tl); 1731 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1732 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1733 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1734 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1735 m_freem(mrep); 1736 return (EBADRPC); 1737 } 1738 uio.uio_offset = 0; 1739 uio.uio_iov = &iov; 1740 uio.uio_iovcnt = 1; 1741 uio.uio_segflg = UIO_SYSSPACE; 1742 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1743 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1744 nfsm_mtouio(&uio, uio.uio_resid); 1745 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1746 if (*tl++ != rpc_auth_kerb || 1747 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1748 printf("Bad kerb verifier\n"); 1749 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1750 nd->nd_procnum = NFSPROC_NOOP; 1751 return (0); 1752 } 1753 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1754 tl = (u_int32_t *)cp; 1755 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1756 printf("Not fullname kerb verifier\n"); 1757 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1758 nd->nd_procnum = NFSPROC_NOOP; 1759 return (0); 1760 } 1761 cp += NFSX_UNSIGNED; 1762 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 1763 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1764 nd->nd_flag |= ND_KERBFULL; 1765 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1766 break; 1767 case RPCAKN_NICKNAME: 1768 if (len != 2 * NFSX_UNSIGNED) { 1769 printf("Kerb nickname short\n"); 1770 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1771 nd->nd_procnum = NFSPROC_NOOP; 1772 return (0); 1773 } 1774 nickuid = fxdr_unsigned(uid_t, *tl); 1775 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1776 if (*tl++ != rpc_auth_kerb || 1777 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1778 printf("Kerb nick verifier bad\n"); 1779 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1780 nd->nd_procnum = NFSPROC_NOOP; 1781 return (0); 1782 } 1783 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1784 tvin.tv_sec = *tl++; 1785 tvin.tv_usec = *tl; 1786 1787 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; 1788 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1789 if (nuidp->nu_cr.cr_uid == nickuid && 1790 (!nd->nd_nam2 || 1791 netaddr_match(NU_NETFAM(nuidp), 1792 &nuidp->nu_haddr, nd->nd_nam2))) 1793 break; 1794 } 1795 if (!nuidp) { 1796 nd->nd_repstat = 1797 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1798 nd->nd_procnum = NFSPROC_NOOP; 1799 return (0); 1800 } 1801 1802 /* 1803 * Now, decrypt the timestamp using the session key 1804 * and validate it. 1805 */ 1806 #ifdef NFSKERB 1807 XXX 1808 #endif 1809 1810 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1811 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1812 if (nuidp->nu_expire < time.tv_sec || 1813 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1814 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1815 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1816 nuidp->nu_expire = 0; 1817 nd->nd_repstat = 1818 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1819 nd->nd_procnum = NFSPROC_NOOP; 1820 return (0); 1821 } 1822 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1823 nd->nd_flag |= ND_KERBNICK; 1824 }; 1825 } else { 1826 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1827 nd->nd_procnum = NFSPROC_NOOP; 1828 return (0); 1829 } 1830 1831 /* 1832 * For nqnfs, get piggybacked lease request. 1833 */ 1834 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1835 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1836 nd->nd_flag |= fxdr_unsigned(int, *tl); 1837 if (nd->nd_flag & ND_LEASE) { 1838 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1839 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 1840 } else 1841 nd->nd_duration = NQ_MINLEASE; 1842 } else 1843 nd->nd_duration = NQ_MINLEASE; 1844 nd->nd_md = md; 1845 nd->nd_dpos = dpos; 1846 return (0); 1847 nfsmout: 1848 return (error); 1849 } 1850 1851 int 1852 nfs_msg(p, server, msg) 1853 struct proc *p; 1854 char *server, *msg; 1855 { 1856 tpr_t tpr; 1857 1858 if (p) 1859 tpr = tprintf_open(p); 1860 else 1861 tpr = NULL; 1862 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1863 tprintf_close(tpr); 1864 return (0); 1865 } 1866 1867 #ifdef NFSSERVER 1868 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 1869 struct nfssvc_sock *, struct proc *, 1870 struct mbuf **)) = { 1871 nfsrv_null, 1872 nfsrv_getattr, 1873 nfsrv_setattr, 1874 nfsrv_lookup, 1875 nfsrv3_access, 1876 nfsrv_readlink, 1877 nfsrv_read, 1878 nfsrv_write, 1879 nfsrv_create, 1880 nfsrv_mkdir, 1881 nfsrv_symlink, 1882 nfsrv_mknod, 1883 nfsrv_remove, 1884 nfsrv_rmdir, 1885 nfsrv_rename, 1886 nfsrv_link, 1887 nfsrv_readdir, 1888 nfsrv_readdirplus, 1889 nfsrv_statfs, 1890 nfsrv_fsinfo, 1891 nfsrv_pathconf, 1892 nfsrv_commit, 1893 nqnfsrv_getlease, 1894 nqnfsrv_vacated, 1895 nfsrv_noop, 1896 nfsrv_noop 1897 }; 1898 1899 /* 1900 * Socket upcall routine for the nfsd sockets. 1901 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1902 * Essentially do as much as possible non-blocking, else punt and it will 1903 * be called with M_WAIT from an nfsd. 1904 */ 1905 void 1906 nfsrv_rcv(so, arg, waitflag) 1907 struct socket *so; 1908 caddr_t arg; 1909 int waitflag; 1910 { 1911 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1912 struct mbuf *m; 1913 struct mbuf *mp, *nam; 1914 struct uio auio; 1915 int flags, error; 1916 1917 if ((slp->ns_flag & SLP_VALID) == 0) 1918 return; 1919 #ifdef notdef 1920 /* 1921 * Define this to test for nfsds handling this under heavy load. 1922 */ 1923 if (waitflag == M_DONTWAIT) { 1924 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1925 } 1926 #endif 1927 auio.uio_procp = NULL; 1928 if (so->so_type == SOCK_STREAM) { 1929 /* 1930 * If there are already records on the queue, defer soreceive() 1931 * to an nfsd so that there is feedback to the TCP layer that 1932 * the nfs servers are heavily loaded. 1933 */ 1934 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1935 slp->ns_flag |= SLP_NEEDQ; 1936 goto dorecs; 1937 } 1938 1939 /* 1940 * Do soreceive(). 1941 */ 1942 auio.uio_resid = 1000000000; 1943 flags = MSG_DONTWAIT; 1944 error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1945 if (error || mp == (struct mbuf *)0) { 1946 if (error == EWOULDBLOCK) 1947 slp->ns_flag |= SLP_NEEDQ; 1948 else 1949 slp->ns_flag |= SLP_DISCONN; 1950 goto dorecs; 1951 } 1952 m = mp; 1953 if (slp->ns_rawend) { 1954 slp->ns_rawend->m_next = m; 1955 slp->ns_cc += 1000000000 - auio.uio_resid; 1956 } else { 1957 slp->ns_raw = m; 1958 slp->ns_cc = 1000000000 - auio.uio_resid; 1959 } 1960 while (m->m_next) 1961 m = m->m_next; 1962 slp->ns_rawend = m; 1963 1964 /* 1965 * Now try and parse record(s) out of the raw stream data. 1966 */ 1967 error = nfsrv_getstream(slp, waitflag); 1968 if (error) { 1969 if (error == EPERM) 1970 slp->ns_flag |= SLP_DISCONN; 1971 else 1972 slp->ns_flag |= SLP_NEEDQ; 1973 } 1974 } else { 1975 do { 1976 auio.uio_resid = 1000000000; 1977 flags = MSG_DONTWAIT; 1978 error = (*so->so_receive)(so, &nam, &auio, &mp, 1979 (struct mbuf **)0, &flags); 1980 if (mp) { 1981 if (nam) { 1982 m = nam; 1983 m->m_next = mp; 1984 } else 1985 m = mp; 1986 if (slp->ns_recend) 1987 slp->ns_recend->m_nextpkt = m; 1988 else 1989 slp->ns_rec = m; 1990 slp->ns_recend = m; 1991 m->m_nextpkt = (struct mbuf *)0; 1992 } 1993 if (error) { 1994 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1995 && error != EWOULDBLOCK) { 1996 slp->ns_flag |= SLP_DISCONN; 1997 goto dorecs; 1998 } 1999 } 2000 } while (mp); 2001 } 2002 2003 /* 2004 * Now try and process the request records, non-blocking. 2005 */ 2006 dorecs: 2007 if (waitflag == M_DONTWAIT && 2008 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 2009 nfsrv_wakenfsd(slp); 2010 } 2011 2012 /* 2013 * Try and extract an RPC request from the mbuf data list received on a 2014 * stream socket. The "waitflag" argument indicates whether or not it 2015 * can sleep. 2016 */ 2017 int 2018 nfsrv_getstream(slp, waitflag) 2019 struct nfssvc_sock *slp; 2020 int waitflag; 2021 { 2022 struct mbuf *m, **mpp; 2023 char *cp1, *cp2; 2024 int len; 2025 struct mbuf *om, *m2, *recm = NULL; 2026 u_int32_t recmark; 2027 2028 if (slp->ns_flag & SLP_GETSTREAM) 2029 panic("nfs getstream"); 2030 slp->ns_flag |= SLP_GETSTREAM; 2031 for (;;) { 2032 if (slp->ns_reclen == 0) { 2033 if (slp->ns_cc < NFSX_UNSIGNED) { 2034 slp->ns_flag &= ~SLP_GETSTREAM; 2035 return (0); 2036 } 2037 m = slp->ns_raw; 2038 if (m->m_len >= NFSX_UNSIGNED) { 2039 memcpy((caddr_t)&recmark, mtod(m, caddr_t), NFSX_UNSIGNED); 2040 m->m_data += NFSX_UNSIGNED; 2041 m->m_len -= NFSX_UNSIGNED; 2042 } else { 2043 cp1 = (caddr_t)&recmark; 2044 cp2 = mtod(m, caddr_t); 2045 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 2046 while (m->m_len == 0) { 2047 m = m->m_next; 2048 cp2 = mtod(m, caddr_t); 2049 } 2050 *cp1++ = *cp2++; 2051 m->m_data++; 2052 m->m_len--; 2053 } 2054 } 2055 slp->ns_cc -= NFSX_UNSIGNED; 2056 recmark = ntohl(recmark); 2057 slp->ns_reclen = recmark & ~0x80000000; 2058 if (recmark & 0x80000000) 2059 slp->ns_flag |= SLP_LASTFRAG; 2060 else 2061 slp->ns_flag &= ~SLP_LASTFRAG; 2062 if (slp->ns_reclen > NFS_MAXPACKET) { 2063 slp->ns_flag &= ~SLP_GETSTREAM; 2064 return (EPERM); 2065 } 2066 } 2067 2068 /* 2069 * Now get the record part. 2070 */ 2071 if (slp->ns_cc == slp->ns_reclen) { 2072 recm = slp->ns_raw; 2073 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2074 slp->ns_cc = slp->ns_reclen = 0; 2075 } else if (slp->ns_cc > slp->ns_reclen) { 2076 len = 0; 2077 m = slp->ns_raw; 2078 om = (struct mbuf *)0; 2079 while (len < slp->ns_reclen) { 2080 if ((len + m->m_len) > slp->ns_reclen) { 2081 size_t left = slp->ns_reclen - len; 2082 2083 MGETHDR(m2, waitflag, m->m_type); 2084 if (m2 == NULL) { 2085 slp->ns_flag &= ~SLP_GETSTREAM; 2086 return (EWOULDBLOCK); 2087 } 2088 if (left > MHLEN) { 2089 MCLGET(m2, waitflag); 2090 if (!(m2->m_flags & M_EXT)) { 2091 m_freem(m2); 2092 slp->ns_flag &= ~SLP_GETSTREAM; 2093 return (EWOULDBLOCK); 2094 } 2095 } 2096 memcpy(mtod(m2, caddr_t), mtod(m, caddr_t), 2097 left); 2098 m2->m_len = left; 2099 m->m_data += left; 2100 m->m_len -= left; 2101 if (om) { 2102 om->m_next = m2; 2103 recm = slp->ns_raw; 2104 } else 2105 recm = m2; 2106 len = slp->ns_reclen; 2107 } else if ((len + m->m_len) == slp->ns_reclen) { 2108 om = m; 2109 len += m->m_len; 2110 m = m->m_next; 2111 recm = slp->ns_raw; 2112 om->m_next = (struct mbuf *)0; 2113 } else { 2114 om = m; 2115 len += m->m_len; 2116 m = m->m_next; 2117 } 2118 } 2119 slp->ns_raw = m; 2120 slp->ns_cc -= len; 2121 slp->ns_reclen = 0; 2122 } else { 2123 slp->ns_flag &= ~SLP_GETSTREAM; 2124 return (0); 2125 } 2126 2127 /* 2128 * Accumulate the fragments into a record. 2129 */ 2130 mpp = &slp->ns_frag; 2131 while (*mpp) 2132 mpp = &((*mpp)->m_next); 2133 *mpp = recm; 2134 if (slp->ns_flag & SLP_LASTFRAG) { 2135 if (slp->ns_recend) 2136 slp->ns_recend->m_nextpkt = slp->ns_frag; 2137 else 2138 slp->ns_rec = slp->ns_frag; 2139 slp->ns_recend = slp->ns_frag; 2140 slp->ns_frag = (struct mbuf *)0; 2141 } 2142 } 2143 } 2144 2145 /* 2146 * Parse an RPC header. 2147 */ 2148 int 2149 nfsrv_dorec(slp, nfsd, ndp) 2150 struct nfssvc_sock *slp; 2151 struct nfsd *nfsd; 2152 struct nfsrv_descript **ndp; 2153 { 2154 struct mbuf *m, *nam; 2155 struct nfsrv_descript *nd; 2156 int error; 2157 2158 *ndp = NULL; 2159 if ((slp->ns_flag & SLP_VALID) == 0 || 2160 (m = slp->ns_rec) == (struct mbuf *)0) 2161 return (ENOBUFS); 2162 slp->ns_rec = m->m_nextpkt; 2163 if (slp->ns_rec) 2164 m->m_nextpkt = (struct mbuf *)0; 2165 else 2166 slp->ns_recend = (struct mbuf *)0; 2167 if (m->m_type == MT_SONAME) { 2168 nam = m; 2169 m = m->m_next; 2170 nam->m_next = NULL; 2171 } else 2172 nam = NULL; 2173 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), 2174 M_NFSRVDESC, M_WAITOK); 2175 nd->nd_md = nd->nd_mrep = m; 2176 nd->nd_nam2 = nam; 2177 nd->nd_dpos = mtod(m, caddr_t); 2178 error = nfs_getreq(nd, nfsd, TRUE); 2179 if (error) { 2180 m_freem(nam); 2181 free((caddr_t)nd, M_NFSRVDESC); 2182 return (error); 2183 } 2184 *ndp = nd; 2185 nfsd->nfsd_nd = nd; 2186 return (0); 2187 } 2188 2189 2190 /* 2191 * Search for a sleeping nfsd and wake it up. 2192 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2193 * running nfsds will go look for the work in the nfssvc_sock list. 2194 */ 2195 void 2196 nfsrv_wakenfsd(slp) 2197 struct nfssvc_sock *slp; 2198 { 2199 struct nfsd *nd; 2200 2201 if ((slp->ns_flag & SLP_VALID) == 0) 2202 return; 2203 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) { 2204 if (nd->nfsd_flag & NFSD_WAITING) { 2205 nd->nfsd_flag &= ~NFSD_WAITING; 2206 if (nd->nfsd_slp) 2207 panic("nfsd wakeup"); 2208 slp->ns_sref++; 2209 nd->nfsd_slp = slp; 2210 wakeup((caddr_t)nd); 2211 return; 2212 } 2213 } 2214 slp->ns_flag |= SLP_DOREC; 2215 nfsd_head_flag |= NFSD_CHECKSLP; 2216 } 2217 #endif /* NFSSERVER */ 2218
Indexes created Tue Sep 30 11:09:46 GMT 2025