nfs_socket.c revision 1.57.2.3
1 /* $NetBSD: nfs_socket.c,v 1.57.2.3 2002/09/04 01:15:54 itojun Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1991, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 39 */ 40 41 /* 42 * Socket operations for use by nfs 43 */ 44 45 #include "fs_nfs.h" 46 #include "opt_nfsserver.h" 47 #include "opt_inet.h" 48 49 #include <sys/param.h> 50 #include <sys/systm.h> 51 #include <sys/callout.h> 52 #include <sys/proc.h> 53 #include <sys/mount.h> 54 #include <sys/kernel.h> 55 #include <sys/mbuf.h> 56 #include <sys/vnode.h> 57 #include <sys/domain.h> 58 #include <sys/protosw.h> 59 #include <sys/socket.h> 60 #include <sys/socketvar.h> 61 #include <sys/syslog.h> 62 #include <sys/tprintf.h> 63 #include <sys/namei.h> 64 #include <sys/signal.h> 65 #include <sys/signalvar.h> 66 67 #include <netinet/in.h> 68 #include <netinet/tcp.h> 69 70 #include <nfs/rpcv2.h> 71 #include <nfs/nfsproto.h> 72 #include <nfs/nfs.h> 73 #include <nfs/xdr_subs.h> 74 #include <nfs/nfsm_subs.h> 75 #include <nfs/nfsmount.h> 76 #include <nfs/nfsnode.h> 77 #include <nfs/nfsrtt.h> 78 #include <nfs/nqnfs.h> 79 #include <nfs/nfs_var.h> 80 81 #define TRUE 1 82 #define FALSE 0 83 84 /* 85 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 86 * Use the mean and mean deviation of rtt for the appropriate type of rpc 87 * for the frequent rpcs and a default for the others. 88 * The justification for doing "other" this way is that these rpcs 89 * happen so infrequently that timer est. would probably be stale. 90 * Also, since many of these rpcs are 91 * non-idempotent, a conservative timeout is desired. 92 * getattr, lookup - A+2D 93 * read, write - A+4D 94 * other - nm_timeo 95 */ 96 #define NFS_RTO(n, t) \ 97 ((t) == 0 ? (n)->nm_timeo : \ 98 ((t) < 3 ? \ 99 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 100 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 101 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 102 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 103 /* 104 * External data, mostly RPC constants in XDR form 105 */ 106 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 107 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, 108 rpc_auth_kerb; 109 extern u_int32_t nfs_prog, nqnfs_prog; 110 extern time_t nqnfsstarttime; 111 extern struct nfsstats nfsstats; 112 extern int nfsv3_procid[NFS_NPROCS]; 113 extern int nfs_ticks; 114 115 /* 116 * Defines which timer to use for the procnum. 117 * 0 - default 118 * 1 - getattr 119 * 2 - lookup 120 * 3 - read 121 * 4 - write 122 */ 123 static int proct[NFS_NPROCS] = { 124 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0, 125 0, 0, 0, 126 }; 127 128 /* 129 * There is a congestion window for outstanding rpcs maintained per mount 130 * point. The cwnd size is adjusted in roughly the way that: 131 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 132 * SIGCOMM '88". ACM, August 1988. 133 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 134 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 135 * of rpcs is in progress. 136 * (The sent count and cwnd are scaled for integer arith.) 137 * Variants of "slow start" were tried and were found to be too much of a 138 * performance hit (ave. rtt 3 times larger), 139 * I suspect due to the large rtt that nfs rpcs have. 140 */ 141 #define NFS_CWNDSCALE 256 142 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 143 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 144 int nfsrtton = 0; 145 struct nfsrtt nfsrtt; 146 147 struct callout nfs_timer_ch = CALLOUT_INITIALIZER; 148 149 /* 150 * Initialize sockets and congestion for a new NFS connection. 151 * We do not free the sockaddr if error. 152 */ 153 int 154 nfs_connect(nmp, rep) 155 struct nfsmount *nmp; 156 struct nfsreq *rep; 157 { 158 struct socket *so; 159 int s, error, rcvreserve, sndreserve; 160 struct sockaddr *saddr; 161 struct sockaddr_in *sin; 162 #ifdef INET6 163 struct sockaddr_in6 *sin6; 164 #endif 165 struct mbuf *m; 166 u_int16_t tport; 167 168 nmp->nm_so = (struct socket *)0; 169 saddr = mtod(nmp->nm_nam, struct sockaddr *); 170 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 171 nmp->nm_soproto); 172 if (error) 173 goto bad; 174 so = nmp->nm_so; 175 nmp->nm_soflags = so->so_proto->pr_flags; 176 177 /* 178 * Some servers require that the client port be a reserved port number. 179 */ 180 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 181 MGET(m, M_WAIT, MT_SONAME); 182 sin = mtod(m, struct sockaddr_in *); 183 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 184 sin->sin_family = AF_INET; 185 sin->sin_addr.s_addr = INADDR_ANY; 186 tport = IPPORT_RESERVED - 1; 187 sin->sin_port = htons(tport); 188 while ((error = sobind(so, m, &proc0)) == EADDRINUSE && 189 --tport > IPPORT_RESERVED / 2) 190 sin->sin_port = htons(tport); 191 m_freem(m); 192 if (error) 193 goto bad; 194 } 195 #ifdef INET6 196 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) { 197 MGET(m, M_WAIT, MT_SONAME); 198 sin6 = mtod(m, struct sockaddr_in6 *); 199 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6); 200 sin6->sin6_family = AF_INET6; 201 sin6->sin6_addr = in6addr_any; 202 tport = IPV6PORT_RESERVED - 1; 203 sin6->sin6_port = htons(tport); 204 while ((error = sobind(so, m, &proc0)) == EADDRINUSE && 205 --tport > IPV6PORT_RESERVED / 2) 206 sin6->sin6_port = htons(tport); 207 m_freem(m); 208 if (error) 209 goto bad; 210 } 211 #endif 212 213 /* 214 * Protocols that do not require connections may be optionally left 215 * unconnected for servers that reply from a port other than NFS_PORT. 216 */ 217 if (nmp->nm_flag & NFSMNT_NOCONN) { 218 if (nmp->nm_soflags & PR_CONNREQUIRED) { 219 error = ENOTCONN; 220 goto bad; 221 } 222 } else { 223 error = soconnect(so, nmp->nm_nam); 224 if (error) 225 goto bad; 226 227 /* 228 * Wait for the connection to complete. Cribbed from the 229 * connect system call but with the wait timing out so 230 * that interruptible mounts don't hang here for a long time. 231 */ 232 s = splsoftnet(); 233 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 234 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 235 "nfscn1", 2 * hz); 236 if ((so->so_state & SS_ISCONNECTING) && 237 so->so_error == 0 && rep && 238 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ 239 so->so_state &= ~SS_ISCONNECTING; 240 splx(s); 241 goto bad; 242 } 243 } 244 if (so->so_error) { 245 error = so->so_error; 246 so->so_error = 0; 247 splx(s); 248 goto bad; 249 } 250 splx(s); 251 } 252 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 253 so->so_rcv.sb_timeo = (5 * hz); 254 so->so_snd.sb_timeo = (5 * hz); 255 } else { 256 so->so_rcv.sb_timeo = 0; 257 so->so_snd.sb_timeo = 0; 258 } 259 if (nmp->nm_sotype == SOCK_DGRAM) { 260 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 261 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 262 NFS_MAXPKTHDR) * 2; 263 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 264 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 265 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 266 NFS_MAXPKTHDR) * 2; 267 } else { 268 if (nmp->nm_sotype != SOCK_STREAM) 269 panic("nfscon sotype"); 270 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 271 MGET(m, M_WAIT, MT_SOOPTS); 272 *mtod(m, int32_t *) = 1; 273 m->m_len = sizeof(int32_t); 274 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 275 } 276 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 277 MGET(m, M_WAIT, MT_SOOPTS); 278 *mtod(m, int32_t *) = 1; 279 m->m_len = sizeof(int32_t); 280 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 281 } 282 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 283 sizeof (u_int32_t)) * 2; 284 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 285 sizeof (u_int32_t)) * 2; 286 } 287 error = soreserve(so, sndreserve, rcvreserve); 288 if (error) 289 goto bad; 290 so->so_rcv.sb_flags |= SB_NOINTR; 291 so->so_snd.sb_flags |= SB_NOINTR; 292 293 /* Initialize other non-zero congestion variables */ 294 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 295 nmp->nm_srtt[4] = (NFS_TIMEO << 3); 296 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 297 nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0; 298 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 299 nmp->nm_sent = 0; 300 nmp->nm_timeouts = 0; 301 return (0); 302 303 bad: 304 nfs_disconnect(nmp); 305 return (error); 306 } 307 308 /* 309 * Reconnect routine: 310 * Called when a connection is broken on a reliable protocol. 311 * - clean up the old socket 312 * - nfs_connect() again 313 * - set R_MUSTRESEND for all outstanding requests on mount point 314 * If this fails the mount point is DEAD! 315 * nb: Must be called with the nfs_sndlock() set on the mount point. 316 */ 317 int 318 nfs_reconnect(rep) 319 struct nfsreq *rep; 320 { 321 struct nfsreq *rp; 322 struct nfsmount *nmp = rep->r_nmp; 323 int error; 324 325 nfs_disconnect(nmp); 326 while ((error = nfs_connect(nmp, rep)) != 0) { 327 if (error == EINTR || error == ERESTART) 328 return (EINTR); 329 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0); 330 } 331 332 /* 333 * Loop through outstanding request list and fix up all requests 334 * on old socket. 335 */ 336 for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) { 337 if (rp->r_nmp == nmp) 338 rp->r_flags |= R_MUSTRESEND; 339 } 340 return (0); 341 } 342 343 /* 344 * NFS disconnect. Clean up and unlink. 345 */ 346 void 347 nfs_disconnect(nmp) 348 struct nfsmount *nmp; 349 { 350 struct socket *so; 351 int drain = 0; 352 353 if (nmp->nm_so) { 354 so = nmp->nm_so; 355 nmp->nm_so = (struct socket *)0; 356 soshutdown(so, 2); 357 drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0; 358 if (drain) { 359 /* 360 * soshutdown() above should wake up the current 361 * listener. 362 * Now wake up those waiting for the recive lock, and 363 * wait for them to go away unhappy, to prevent *nmp 364 * from evaporating while they're sleeping. 365 */ 366 while (nmp->nm_waiters > 0) { 367 wakeup (&nmp->nm_iflag); 368 (void) tsleep(&nmp->nm_waiters, PVFS, 369 "nfsdis", 0); 370 } 371 } 372 soclose(so); 373 } 374 #ifdef DIAGNOSTIC 375 if (drain && (nmp->nm_waiters > 0)) 376 panic("nfs_disconnect: waiters left after drain?\n"); 377 #endif 378 } 379 380 void 381 nfs_safedisconnect(nmp) 382 struct nfsmount *nmp; 383 { 384 struct nfsreq dummyreq; 385 386 memset(&dummyreq, 0, sizeof(dummyreq)); 387 dummyreq.r_nmp = nmp; 388 nfs_rcvlock(&dummyreq); /* XXX ignored error return */ 389 nfs_disconnect(nmp); 390 nfs_rcvunlock(&nmp->nm_iflag); 391 } 392 393 /* 394 * This is the nfs send routine. For connection based socket types, it 395 * must be called with an nfs_sndlock() on the socket. 396 * "rep == NULL" indicates that it has been called from a server. 397 * For the client side: 398 * - return EINTR if the RPC is terminated, 0 otherwise 399 * - set R_MUSTRESEND if the send fails for any reason 400 * - do any cleanup required by recoverable socket errors (???) 401 * For the server side: 402 * - return EINTR or ERESTART if interrupted by a signal 403 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 404 * - do any cleanup required by recoverable socket errors (???) 405 */ 406 int 407 nfs_send(so, nam, top, rep) 408 struct socket *so; 409 struct mbuf *nam; 410 struct mbuf *top; 411 struct nfsreq *rep; 412 { 413 struct mbuf *sendnam; 414 int error, soflags, flags; 415 416 if (rep) { 417 if (rep->r_flags & R_SOFTTERM) { 418 m_freem(top); 419 return (EINTR); 420 } 421 if ((so = rep->r_nmp->nm_so) == NULL) { 422 rep->r_flags |= R_MUSTRESEND; 423 m_freem(top); 424 return (0); 425 } 426 rep->r_flags &= ~R_MUSTRESEND; 427 soflags = rep->r_nmp->nm_soflags; 428 } else 429 soflags = so->so_proto->pr_flags; 430 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 431 sendnam = (struct mbuf *)0; 432 else 433 sendnam = nam; 434 if (so->so_type == SOCK_SEQPACKET) 435 flags = MSG_EOR; 436 else 437 flags = 0; 438 439 error = (*so->so_send)(so, sendnam, (struct uio *)0, top, 440 (struct mbuf *)0, flags); 441 if (error) { 442 if (rep) { 443 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 444 /* 445 * We're too fast for the network/driver, 446 * and UDP isn't flowcontrolled. 447 * We need to resend. This is not fatal, 448 * just try again. 449 * 450 * Could be smarter here by doing some sort 451 * of a backoff, but this is rare. 452 */ 453 rep->r_flags |= R_MUSTRESEND; 454 } else { 455 log(LOG_INFO, "nfs send error %d for %s\n", 456 error, 457 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 458 /* 459 * Deal with errors for the client side. 460 */ 461 if (rep->r_flags & R_SOFTTERM) 462 error = EINTR; 463 else 464 rep->r_flags |= R_MUSTRESEND; 465 } 466 } else 467 log(LOG_INFO, "nfsd send error %d\n", error); 468 469 /* 470 * Handle any recoverable (soft) socket errors here. (???) 471 */ 472 if (error != EINTR && error != ERESTART && 473 error != EWOULDBLOCK && error != EPIPE) 474 error = 0; 475 } 476 return (error); 477 } 478 479 #ifdef NFS 480 /* 481 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 482 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 483 * Mark and consolidate the data into a new mbuf list. 484 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 485 * small mbufs. 486 * For SOCK_STREAM we must be very careful to read an entire record once 487 * we have read any of it, even if the system call has been interrupted. 488 */ 489 int 490 nfs_receive(rep, aname, mp) 491 struct nfsreq *rep; 492 struct mbuf **aname; 493 struct mbuf **mp; 494 { 495 struct socket *so; 496 struct uio auio; 497 struct iovec aio; 498 struct mbuf *m; 499 struct mbuf *control; 500 u_int32_t len; 501 struct mbuf **getnam; 502 int error, sotype, rcvflg; 503 struct proc *p = curproc; /* XXX */ 504 505 /* 506 * Set up arguments for soreceive() 507 */ 508 *mp = (struct mbuf *)0; 509 *aname = (struct mbuf *)0; 510 sotype = rep->r_nmp->nm_sotype; 511 512 /* 513 * For reliable protocols, lock against other senders/receivers 514 * in case a reconnect is necessary. 515 * For SOCK_STREAM, first get the Record Mark to find out how much 516 * more there is to get. 517 * We must lock the socket against other receivers 518 * until we have an entire rpc request/reply. 519 */ 520 if (sotype != SOCK_DGRAM) { 521 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 522 if (error) 523 return (error); 524 tryagain: 525 /* 526 * Check for fatal errors and resending request. 527 */ 528 /* 529 * Ugh: If a reconnect attempt just happened, nm_so 530 * would have changed. NULL indicates a failed 531 * attempt that has essentially shut down this 532 * mount point. 533 */ 534 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 535 nfs_sndunlock(&rep->r_nmp->nm_iflag); 536 return (EINTR); 537 } 538 so = rep->r_nmp->nm_so; 539 if (!so) { 540 error = nfs_reconnect(rep); 541 if (error) { 542 nfs_sndunlock(&rep->r_nmp->nm_iflag); 543 return (error); 544 } 545 goto tryagain; 546 } 547 while (rep->r_flags & R_MUSTRESEND) { 548 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 549 nfsstats.rpcretries++; 550 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 551 if (error) { 552 if (error == EINTR || error == ERESTART || 553 (error = nfs_reconnect(rep)) != 0) { 554 nfs_sndunlock(&rep->r_nmp->nm_iflag); 555 return (error); 556 } 557 goto tryagain; 558 } 559 } 560 nfs_sndunlock(&rep->r_nmp->nm_iflag); 561 if (sotype == SOCK_STREAM) { 562 aio.iov_base = (caddr_t) &len; 563 aio.iov_len = sizeof(u_int32_t); 564 auio.uio_iov = &aio; 565 auio.uio_iovcnt = 1; 566 auio.uio_segflg = UIO_SYSSPACE; 567 auio.uio_rw = UIO_READ; 568 auio.uio_offset = 0; 569 auio.uio_resid = sizeof(u_int32_t); 570 auio.uio_procp = p; 571 do { 572 rcvflg = MSG_WAITALL; 573 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 574 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 575 if (error == EWOULDBLOCK && rep) { 576 if (rep->r_flags & R_SOFTTERM) 577 return (EINTR); 578 } 579 } while (error == EWOULDBLOCK); 580 if (!error && auio.uio_resid > 0) { 581 /* 582 * Don't log a 0 byte receive; it means 583 * that the socket has been closed, and 584 * can happen during normal operation 585 * (forcible unmount or Solaris server). 586 */ 587 if (auio.uio_resid != sizeof (u_int32_t)) 588 log(LOG_INFO, 589 "short receive (%lu/%lu) from nfs server %s\n", 590 (u_long)sizeof(u_int32_t) - auio.uio_resid, 591 (u_long)sizeof(u_int32_t), 592 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 593 error = EPIPE; 594 } 595 if (error) 596 goto errout; 597 len = ntohl(len) & ~0x80000000; 598 /* 599 * This is SERIOUS! We are out of sync with the sender 600 * and forcing a disconnect/reconnect is all I can do. 601 */ 602 if (len > NFS_MAXPACKET) { 603 log(LOG_ERR, "%s (%d) from nfs server %s\n", 604 "impossible packet length", 605 len, 606 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 607 error = EFBIG; 608 goto errout; 609 } 610 auio.uio_resid = len; 611 do { 612 rcvflg = MSG_WAITALL; 613 error = (*so->so_receive)(so, (struct mbuf **)0, 614 &auio, mp, (struct mbuf **)0, &rcvflg); 615 } while (error == EWOULDBLOCK || error == EINTR || 616 error == ERESTART); 617 if (!error && auio.uio_resid > 0) { 618 if (len != auio.uio_resid) 619 log(LOG_INFO, 620 "short receive (%lu/%d) from nfs server %s\n", 621 (u_long)len - auio.uio_resid, len, 622 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 623 error = EPIPE; 624 } 625 } else { 626 /* 627 * NB: Since uio_resid is big, MSG_WAITALL is ignored 628 * and soreceive() will return when it has either a 629 * control msg or a data msg. 630 * We have no use for control msg., but must grab them 631 * and then throw them away so we know what is going 632 * on. 633 */ 634 auio.uio_resid = len = 100000000; /* Anything Big */ 635 auio.uio_procp = p; 636 do { 637 rcvflg = 0; 638 error = (*so->so_receive)(so, (struct mbuf **)0, 639 &auio, mp, &control, &rcvflg); 640 if (control) 641 m_freem(control); 642 if (error == EWOULDBLOCK && rep) { 643 if (rep->r_flags & R_SOFTTERM) 644 return (EINTR); 645 } 646 } while (error == EWOULDBLOCK || 647 (!error && *mp == NULL && control)); 648 if ((rcvflg & MSG_EOR) == 0) 649 printf("Egad!!\n"); 650 if (!error && *mp == NULL) 651 error = EPIPE; 652 len -= auio.uio_resid; 653 } 654 errout: 655 if (error && error != EINTR && error != ERESTART) { 656 m_freem(*mp); 657 *mp = (struct mbuf *)0; 658 if (error != EPIPE) 659 log(LOG_INFO, 660 "receive error %d from nfs server %s\n", 661 error, 662 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 663 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 664 if (!error) 665 error = nfs_reconnect(rep); 666 if (!error) 667 goto tryagain; 668 else 669 nfs_sndunlock(&rep->r_nmp->nm_iflag); 670 } 671 } else { 672 if ((so = rep->r_nmp->nm_so) == NULL) 673 return (EACCES); 674 if (so->so_state & SS_ISCONNECTED) 675 getnam = (struct mbuf **)0; 676 else 677 getnam = aname; 678 auio.uio_resid = len = 1000000; 679 auio.uio_procp = p; 680 do { 681 rcvflg = 0; 682 error = (*so->so_receive)(so, getnam, &auio, mp, 683 (struct mbuf **)0, &rcvflg); 684 if (error == EWOULDBLOCK && 685 (rep->r_flags & R_SOFTTERM)) 686 return (EINTR); 687 } while (error == EWOULDBLOCK); 688 len -= auio.uio_resid; 689 if (!error && *mp == NULL) 690 error = EPIPE; 691 } 692 if (error) { 693 m_freem(*mp); 694 *mp = (struct mbuf *)0; 695 } 696 return (error); 697 } 698 699 /* 700 * Implement receipt of reply on a socket. 701 * We must search through the list of received datagrams matching them 702 * with outstanding requests using the xid, until ours is found. 703 */ 704 /* ARGSUSED */ 705 int 706 nfs_reply(myrep) 707 struct nfsreq *myrep; 708 { 709 struct nfsreq *rep; 710 struct nfsmount *nmp = myrep->r_nmp; 711 int32_t t1; 712 struct mbuf *mrep, *nam, *md; 713 u_int32_t rxid, *tl; 714 caddr_t dpos, cp2; 715 int error; 716 717 /* 718 * Loop around until we get our own reply 719 */ 720 for (;;) { 721 /* 722 * Lock against other receivers so that I don't get stuck in 723 * sbwait() after someone else has received my reply for me. 724 * Also necessary for connection based protocols to avoid 725 * race conditions during a reconnect. 726 */ 727 error = nfs_rcvlock(myrep); 728 if (error == EALREADY) 729 return (0); 730 if (error) 731 return (error); 732 /* 733 * Get the next Rpc reply off the socket 734 */ 735 nmp->nm_waiters++; 736 error = nfs_receive(myrep, &nam, &mrep); 737 nfs_rcvunlock(&nmp->nm_iflag); 738 if (error) { 739 740 if (nmp->nm_iflag & NFSMNT_DISMNT) { 741 /* 742 * Oops, we're going away now.. 743 */ 744 nmp->nm_waiters--; 745 wakeup (&nmp->nm_waiters); 746 return error; 747 } 748 nmp->nm_waiters--; 749 /* 750 * Ignore routing errors on connectionless protocols?? 751 */ 752 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 753 nmp->nm_so->so_error = 0; 754 #ifdef DEBUG 755 printf("nfs_reply: ignoring error %d\n", error); 756 #endif 757 if (myrep->r_flags & R_GETONEREP) 758 return (0); 759 continue; 760 } 761 return (error); 762 } 763 nmp->nm_waiters--; 764 if (nam) 765 m_freem(nam); 766 767 /* 768 * Get the xid and check that it is an rpc reply 769 */ 770 md = mrep; 771 dpos = mtod(md, caddr_t); 772 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 773 rxid = *tl++; 774 if (*tl != rpc_reply) { 775 if (nmp->nm_flag & NFSMNT_NQNFS) { 776 if (nqnfs_callback(nmp, mrep, md, dpos)) 777 nfsstats.rpcinvalid++; 778 } else { 779 nfsstats.rpcinvalid++; 780 m_freem(mrep); 781 } 782 nfsmout: 783 if (myrep->r_flags & R_GETONEREP) 784 return (0); 785 continue; 786 } 787 788 /* 789 * Loop through the request list to match up the reply 790 * Iff no match, just drop the datagram 791 */ 792 for (rep = nfs_reqq.tqh_first; rep != 0; 793 rep = rep->r_chain.tqe_next) { 794 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 795 /* Found it.. */ 796 rep->r_mrep = mrep; 797 rep->r_md = md; 798 rep->r_dpos = dpos; 799 if (nfsrtton) { 800 struct rttl *rt; 801 802 rt = &nfsrtt.rttl[nfsrtt.pos]; 803 rt->proc = rep->r_procnum; 804 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 805 rt->sent = nmp->nm_sent; 806 rt->cwnd = nmp->nm_cwnd; 807 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 808 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 809 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 810 rt->tstamp = time; 811 if (rep->r_flags & R_TIMING) 812 rt->rtt = rep->r_rtt; 813 else 814 rt->rtt = 1000000; 815 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 816 } 817 /* 818 * Update congestion window. 819 * Do the additive increase of 820 * one rpc/rtt. 821 */ 822 if (nmp->nm_cwnd <= nmp->nm_sent) { 823 nmp->nm_cwnd += 824 (NFS_CWNDSCALE * NFS_CWNDSCALE + 825 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 826 if (nmp->nm_cwnd > NFS_MAXCWND) 827 nmp->nm_cwnd = NFS_MAXCWND; 828 } 829 rep->r_flags &= ~R_SENT; 830 nmp->nm_sent -= NFS_CWNDSCALE; 831 /* 832 * Update rtt using a gain of 0.125 on the mean 833 * and a gain of 0.25 on the deviation. 834 */ 835 if (rep->r_flags & R_TIMING) { 836 /* 837 * Since the timer resolution of 838 * NFS_HZ is so course, it can often 839 * result in r_rtt == 0. Since 840 * r_rtt == N means that the actual 841 * rtt is between N+dt and N+2-dt ticks, 842 * add 1. 843 */ 844 t1 = rep->r_rtt + 1; 845 t1 -= (NFS_SRTT(rep) >> 3); 846 NFS_SRTT(rep) += t1; 847 if (t1 < 0) 848 t1 = -t1; 849 t1 -= (NFS_SDRTT(rep) >> 2); 850 NFS_SDRTT(rep) += t1; 851 } 852 nmp->nm_timeouts = 0; 853 break; 854 } 855 } 856 /* 857 * If not matched to a request, drop it. 858 * If it's mine, get out. 859 */ 860 if (rep == 0) { 861 nfsstats.rpcunexpected++; 862 m_freem(mrep); 863 } else if (rep == myrep) { 864 if (rep->r_mrep == NULL) 865 panic("nfsreply nil"); 866 return (0); 867 } 868 if (myrep->r_flags & R_GETONEREP) 869 return (0); 870 } 871 } 872 873 /* 874 * nfs_request - goes something like this 875 * - fill in request struct 876 * - links it into list 877 * - calls nfs_send() for first transmit 878 * - calls nfs_receive() to get reply 879 * - break down rpc header and return with nfs reply pointed to 880 * by mrep or error 881 * nb: always frees up mreq mbuf list 882 */ 883 int 884 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 885 struct vnode *vp; 886 struct mbuf *mrest; 887 int procnum; 888 struct proc *procp; 889 struct ucred *cred; 890 struct mbuf **mrp; 891 struct mbuf **mdp; 892 caddr_t *dposp; 893 { 894 struct mbuf *m, *mrep; 895 struct nfsreq *rep; 896 u_int32_t *tl; 897 int i; 898 struct nfsmount *nmp; 899 struct mbuf *md, *mheadend; 900 struct nfsnode *np; 901 char nickv[RPCX_NICKVERF]; 902 time_t reqtime, waituntil; 903 caddr_t dpos, cp2; 904 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; 905 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 906 int verf_len, verf_type; 907 u_int32_t xid; 908 u_quad_t frev; 909 char *auth_str, *verf_str; 910 NFSKERBKEY_T key; /* save session key */ 911 912 nmp = VFSTONFS(vp->v_mount); 913 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 914 rep->r_nmp = nmp; 915 rep->r_vp = vp; 916 rep->r_procp = procp; 917 rep->r_procnum = procnum; 918 i = 0; 919 m = mrest; 920 while (m) { 921 i += m->m_len; 922 m = m->m_next; 923 } 924 mrest_len = i; 925 926 /* 927 * Get the RPC header with authorization. 928 */ 929 kerbauth: 930 verf_str = auth_str = (char *)0; 931 if (nmp->nm_flag & NFSMNT_KERB) { 932 verf_str = nickv; 933 verf_len = sizeof (nickv); 934 auth_type = RPCAUTH_KERB4; 935 memset((caddr_t)key, 0, sizeof (key)); 936 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 937 &auth_len, verf_str, verf_len)) { 938 error = nfs_getauth(nmp, rep, cred, &auth_str, 939 &auth_len, verf_str, &verf_len, key); 940 if (error) { 941 free((caddr_t)rep, M_NFSREQ); 942 m_freem(mrest); 943 return (error); 944 } 945 } 946 } else { 947 auth_type = RPCAUTH_UNIX; 948 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 949 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 950 5 * NFSX_UNSIGNED; 951 } 952 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 953 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 954 if (auth_str) 955 free(auth_str, M_TEMP); 956 957 /* 958 * For stream protocols, insert a Sun RPC Record Mark. 959 */ 960 if (nmp->nm_sotype == SOCK_STREAM) { 961 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 962 *mtod(m, u_int32_t *) = htonl(0x80000000 | 963 (m->m_pkthdr.len - NFSX_UNSIGNED)); 964 } 965 rep->r_mreq = m; 966 rep->r_xid = xid; 967 tryagain: 968 if (nmp->nm_flag & NFSMNT_SOFT) 969 rep->r_retry = nmp->nm_retry; 970 else 971 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 972 rep->r_rtt = rep->r_rexmit = 0; 973 if (proct[procnum] > 0) 974 rep->r_flags = R_TIMING; 975 else 976 rep->r_flags = 0; 977 rep->r_mrep = NULL; 978 979 /* 980 * Do the client side RPC. 981 */ 982 nfsstats.rpcrequests++; 983 /* 984 * Chain request into list of outstanding requests. Be sure 985 * to put it LAST so timer finds oldest requests first. 986 */ 987 s = splsoftnet(); 988 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 989 990 /* Get send time for nqnfs */ 991 reqtime = time.tv_sec; 992 993 /* 994 * If backing off another request or avoiding congestion, don't 995 * send this one now but let timer do it. If not timing a request, 996 * do it now. 997 */ 998 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 999 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1000 nmp->nm_sent < nmp->nm_cwnd)) { 1001 splx(s); 1002 if (nmp->nm_soflags & PR_CONNREQUIRED) 1003 error = nfs_sndlock(&nmp->nm_iflag, rep); 1004 if (!error) { 1005 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1006 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 1007 if (nmp->nm_soflags & PR_CONNREQUIRED) 1008 nfs_sndunlock(&nmp->nm_iflag); 1009 } 1010 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1011 nmp->nm_sent += NFS_CWNDSCALE; 1012 rep->r_flags |= R_SENT; 1013 } 1014 } else { 1015 splx(s); 1016 rep->r_rtt = -1; 1017 } 1018 1019 /* 1020 * Wait for the reply from our send or the timer's. 1021 */ 1022 if (!error || error == EPIPE) 1023 error = nfs_reply(rep); 1024 1025 /* 1026 * RPC done, unlink the request. 1027 */ 1028 s = splsoftnet(); 1029 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1030 splx(s); 1031 1032 /* 1033 * Decrement the outstanding request count. 1034 */ 1035 if (rep->r_flags & R_SENT) { 1036 rep->r_flags &= ~R_SENT; /* paranoia */ 1037 nmp->nm_sent -= NFS_CWNDSCALE; 1038 } 1039 1040 /* 1041 * If there was a successful reply and a tprintf msg. 1042 * tprintf a response. 1043 */ 1044 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1045 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1046 "is alive again"); 1047 mrep = rep->r_mrep; 1048 md = rep->r_md; 1049 dpos = rep->r_dpos; 1050 if (error) { 1051 m_freem(rep->r_mreq); 1052 free((caddr_t)rep, M_NFSREQ); 1053 return (error); 1054 } 1055 1056 /* 1057 * break down the rpc header and check if ok 1058 */ 1059 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1060 if (*tl++ == rpc_msgdenied) { 1061 if (*tl == rpc_mismatch) 1062 error = EOPNOTSUPP; 1063 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1064 if (!failed_auth) { 1065 failed_auth++; 1066 mheadend->m_next = (struct mbuf *)0; 1067 m_freem(mrep); 1068 m_freem(rep->r_mreq); 1069 goto kerbauth; 1070 } else 1071 error = EAUTH; 1072 } else 1073 error = EACCES; 1074 m_freem(mrep); 1075 m_freem(rep->r_mreq); 1076 free((caddr_t)rep, M_NFSREQ); 1077 return (error); 1078 } 1079 1080 /* 1081 * Grab any Kerberos verifier, otherwise just throw it away. 1082 */ 1083 verf_type = fxdr_unsigned(int, *tl++); 1084 i = fxdr_unsigned(int32_t, *tl); 1085 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1086 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1087 if (error) 1088 goto nfsmout; 1089 } else if (i > 0) 1090 nfsm_adv(nfsm_rndup(i)); 1091 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1092 /* 0 == ok */ 1093 if (*tl == 0) { 1094 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1095 if (*tl != 0) { 1096 error = fxdr_unsigned(int, *tl); 1097 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1098 error == NFSERR_TRYLATER) { 1099 m_freem(mrep); 1100 error = 0; 1101 waituntil = time.tv_sec + trylater_delay; 1102 while (time.tv_sec < waituntil) 1103 (void) tsleep((caddr_t)&lbolt, 1104 PSOCK, "nqnfstry", 0); 1105 trylater_delay *= nfs_backoff[trylater_cnt]; 1106 if (trylater_cnt < 7) 1107 trylater_cnt++; 1108 goto tryagain; 1109 } 1110 1111 /* 1112 * If the File Handle was stale, invalidate the 1113 * lookup cache, just in case. 1114 */ 1115 if (error == ESTALE) 1116 cache_purge(vp); 1117 if (nmp->nm_flag & NFSMNT_NFSV3) { 1118 *mrp = mrep; 1119 *mdp = md; 1120 *dposp = dpos; 1121 error |= NFSERR_RETERR; 1122 } else 1123 m_freem(mrep); 1124 m_freem(rep->r_mreq); 1125 free((caddr_t)rep, M_NFSREQ); 1126 return (error); 1127 } 1128 1129 /* 1130 * For nqnfs, get any lease in reply 1131 */ 1132 if (nmp->nm_flag & NFSMNT_NQNFS) { 1133 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1134 if (*tl) { 1135 np = VTONFS(vp); 1136 nqlflag = fxdr_unsigned(int, *tl); 1137 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1138 cachable = fxdr_unsigned(int, *tl++); 1139 reqtime += fxdr_unsigned(int, *tl++); 1140 if (reqtime > time.tv_sec) { 1141 frev = fxdr_hyper(tl); 1142 nqnfs_clientlease(nmp, np, nqlflag, 1143 cachable, reqtime, frev); 1144 } 1145 } 1146 } 1147 *mrp = mrep; 1148 *mdp = md; 1149 *dposp = dpos; 1150 m_freem(rep->r_mreq); 1151 FREE((caddr_t)rep, M_NFSREQ); 1152 return (0); 1153 } 1154 m_freem(mrep); 1155 error = EPROTONOSUPPORT; 1156 nfsmout: 1157 m_freem(rep->r_mreq); 1158 free((caddr_t)rep, M_NFSREQ); 1159 return (error); 1160 } 1161 #endif /* NFS */ 1162 1163 /* 1164 * Generate the rpc reply header 1165 * siz arg. is used to decide if adding a cluster is worthwhile 1166 */ 1167 int 1168 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1169 int siz; 1170 struct nfsrv_descript *nd; 1171 struct nfssvc_sock *slp; 1172 int err; 1173 int cache; 1174 u_quad_t *frev; 1175 struct mbuf **mrq; 1176 struct mbuf **mbp; 1177 caddr_t *bposp; 1178 { 1179 u_int32_t *tl; 1180 struct mbuf *mreq; 1181 caddr_t bpos; 1182 struct mbuf *mb, *mb2; 1183 1184 MGETHDR(mreq, M_WAIT, MT_DATA); 1185 mb = mreq; 1186 /* 1187 * If this is a big reply, use a cluster else 1188 * try and leave leading space for the lower level headers. 1189 */ 1190 siz += RPC_REPLYSIZ; 1191 if (siz >= max_datalen) { 1192 MCLGET(mreq, M_WAIT); 1193 } else 1194 mreq->m_data += max_hdr; 1195 tl = mtod(mreq, u_int32_t *); 1196 mreq->m_len = 6 * NFSX_UNSIGNED; 1197 bpos = ((caddr_t)tl) + mreq->m_len; 1198 *tl++ = txdr_unsigned(nd->nd_retxid); 1199 *tl++ = rpc_reply; 1200 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1201 *tl++ = rpc_msgdenied; 1202 if (err & NFSERR_AUTHERR) { 1203 *tl++ = rpc_autherr; 1204 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1205 mreq->m_len -= NFSX_UNSIGNED; 1206 bpos -= NFSX_UNSIGNED; 1207 } else { 1208 *tl++ = rpc_mismatch; 1209 *tl++ = txdr_unsigned(RPC_VER2); 1210 *tl = txdr_unsigned(RPC_VER2); 1211 } 1212 } else { 1213 *tl++ = rpc_msgaccepted; 1214 1215 /* 1216 * For Kerberos authentication, we must send the nickname 1217 * verifier back, otherwise just RPCAUTH_NULL. 1218 */ 1219 if (nd->nd_flag & ND_KERBFULL) { 1220 struct nfsuid *nuidp; 1221 struct timeval ktvin, ktvout; 1222 1223 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; 1224 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1225 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1226 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), 1227 &nuidp->nu_haddr, nd->nd_nam2))) 1228 break; 1229 } 1230 if (nuidp) { 1231 ktvin.tv_sec = 1232 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); 1233 ktvin.tv_usec = 1234 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1235 1236 /* 1237 * Encrypt the timestamp in ecb mode using the 1238 * session key. 1239 */ 1240 #ifdef NFSKERB 1241 XXX 1242 #endif 1243 1244 *tl++ = rpc_auth_kerb; 1245 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1246 *tl = ktvout.tv_sec; 1247 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1248 *tl++ = ktvout.tv_usec; 1249 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1250 } else { 1251 *tl++ = 0; 1252 *tl++ = 0; 1253 } 1254 } else { 1255 *tl++ = 0; 1256 *tl++ = 0; 1257 } 1258 switch (err) { 1259 case EPROGUNAVAIL: 1260 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1261 break; 1262 case EPROGMISMATCH: 1263 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1264 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1265 if (nd->nd_flag & ND_NQNFS) { 1266 *tl++ = txdr_unsigned(3); 1267 *tl = txdr_unsigned(3); 1268 } else { 1269 *tl++ = txdr_unsigned(2); 1270 *tl = txdr_unsigned(3); 1271 } 1272 break; 1273 case EPROCUNAVAIL: 1274 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1275 break; 1276 case EBADRPC: 1277 *tl = txdr_unsigned(RPC_GARBAGE); 1278 break; 1279 default: 1280 *tl = 0; 1281 if (err != NFSERR_RETVOID) { 1282 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1283 if (err) 1284 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1285 else 1286 *tl = 0; 1287 } 1288 break; 1289 }; 1290 } 1291 1292 /* 1293 * For nqnfs, piggyback lease as requested. 1294 */ 1295 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1296 if (nd->nd_flag & ND_LEASE) { 1297 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1298 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1299 *tl++ = txdr_unsigned(cache); 1300 *tl++ = txdr_unsigned(nd->nd_duration); 1301 txdr_hyper(*frev, tl); 1302 } else { 1303 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1304 *tl = 0; 1305 } 1306 } 1307 if (mrq != NULL) 1308 *mrq = mreq; 1309 *mbp = mb; 1310 *bposp = bpos; 1311 if (err != 0 && err != NFSERR_RETVOID) 1312 nfsstats.srvrpc_errs++; 1313 return (0); 1314 } 1315 1316 /* 1317 * Nfs timer routine 1318 * Scan the nfsreq list and retranmit any requests that have timed out 1319 * To avoid retransmission attempts on STREAM sockets (in the future) make 1320 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1321 */ 1322 void 1323 nfs_timer(arg) 1324 void *arg; /* never used */ 1325 { 1326 struct nfsreq *rep; 1327 struct mbuf *m; 1328 struct socket *so; 1329 struct nfsmount *nmp; 1330 int timeo; 1331 int s, error; 1332 #ifdef NFSSERVER 1333 struct nfssvc_sock *slp; 1334 static long lasttime = 0; 1335 u_quad_t cur_usec; 1336 #endif 1337 1338 s = splsoftnet(); 1339 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1340 nmp = rep->r_nmp; 1341 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1342 continue; 1343 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1344 rep->r_flags |= R_SOFTTERM; 1345 continue; 1346 } 1347 if (rep->r_rtt >= 0) { 1348 rep->r_rtt++; 1349 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1350 timeo = nmp->nm_timeo; 1351 else 1352 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1353 if (nmp->nm_timeouts > 0) 1354 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1355 if (rep->r_rtt <= timeo) 1356 continue; 1357 if (nmp->nm_timeouts < 8) 1358 nmp->nm_timeouts++; 1359 } 1360 /* 1361 * Check for server not responding 1362 */ 1363 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1364 rep->r_rexmit > nmp->nm_deadthresh) { 1365 nfs_msg(rep->r_procp, 1366 nmp->nm_mountp->mnt_stat.f_mntfromname, 1367 "not responding"); 1368 rep->r_flags |= R_TPRINTFMSG; 1369 } 1370 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1371 nfsstats.rpctimeouts++; 1372 rep->r_flags |= R_SOFTTERM; 1373 continue; 1374 } 1375 if (nmp->nm_sotype != SOCK_DGRAM) { 1376 if (++rep->r_rexmit > NFS_MAXREXMIT) 1377 rep->r_rexmit = NFS_MAXREXMIT; 1378 continue; 1379 } 1380 if ((so = nmp->nm_so) == NULL) 1381 continue; 1382 1383 /* 1384 * If there is enough space and the window allows.. 1385 * Resend it 1386 * Set r_rtt to -1 in case we fail to send it now. 1387 */ 1388 rep->r_rtt = -1; 1389 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1390 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1391 (rep->r_flags & R_SENT) || 1392 nmp->nm_sent < nmp->nm_cwnd) && 1393 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1394 if (so->so_state & SS_ISCONNECTED) 1395 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1396 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1397 else 1398 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1399 nmp->nm_nam, (struct mbuf *)0, (struct proc *)0); 1400 if (error) { 1401 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1402 #ifdef DEBUG 1403 printf("nfs_timer: ignoring error %d\n", 1404 error); 1405 #endif 1406 so->so_error = 0; 1407 } 1408 } else { 1409 /* 1410 * Iff first send, start timing 1411 * else turn timing off, backoff timer 1412 * and divide congestion window by 2. 1413 */ 1414 if (rep->r_flags & R_SENT) { 1415 rep->r_flags &= ~R_TIMING; 1416 if (++rep->r_rexmit > NFS_MAXREXMIT) 1417 rep->r_rexmit = NFS_MAXREXMIT; 1418 nmp->nm_cwnd >>= 1; 1419 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1420 nmp->nm_cwnd = NFS_CWNDSCALE; 1421 nfsstats.rpcretries++; 1422 } else { 1423 rep->r_flags |= R_SENT; 1424 nmp->nm_sent += NFS_CWNDSCALE; 1425 } 1426 rep->r_rtt = 0; 1427 } 1428 } 1429 } 1430 1431 #ifdef NFSSERVER 1432 /* 1433 * Call the nqnfs server timer once a second to handle leases. 1434 */ 1435 if (lasttime != time.tv_sec) { 1436 lasttime = time.tv_sec; 1437 nqnfs_serverd(); 1438 } 1439 1440 /* 1441 * Scan the write gathering queues for writes that need to be 1442 * completed now. 1443 */ 1444 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1445 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 1446 slp = slp->ns_chain.tqe_next) { 1447 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec) 1448 nfsrv_wakenfsd(slp); 1449 } 1450 #endif /* NFSSERVER */ 1451 splx(s); 1452 callout_reset(&nfs_timer_ch, nfs_ticks, nfs_timer, NULL); 1453 } 1454 1455 /* 1456 * Test for a termination condition pending on the process. 1457 * This is used for NFSMNT_INT mounts. 1458 */ 1459 int 1460 nfs_sigintr(nmp, rep, p) 1461 struct nfsmount *nmp; 1462 struct nfsreq *rep; 1463 struct proc *p; 1464 { 1465 sigset_t ss; 1466 1467 if (rep && (rep->r_flags & R_SOFTTERM)) 1468 return (EINTR); 1469 if (!(nmp->nm_flag & NFSMNT_INT)) 1470 return (0); 1471 if (p) { 1472 sigpending1(p, &ss); 1473 #if 0 1474 sigminusset(&p->p_sigignore, &ss); 1475 #endif 1476 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1477 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1478 sigismember(&ss, SIGQUIT)) 1479 return (EINTR); 1480 } 1481 return (0); 1482 } 1483 1484 /* 1485 * Lock a socket against others. 1486 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1487 * and also to avoid race conditions between the processes with nfs requests 1488 * in progress when a reconnect is necessary. 1489 */ 1490 int 1491 nfs_sndlock(flagp, rep) 1492 int *flagp; 1493 struct nfsreq *rep; 1494 { 1495 struct proc *p; 1496 int slpflag = 0, slptimeo = 0; 1497 1498 if (rep) { 1499 p = rep->r_procp; 1500 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1501 slpflag = PCATCH; 1502 } else 1503 p = (struct proc *)0; 1504 while (*flagp & NFSMNT_SNDLOCK) { 1505 if (nfs_sigintr(rep->r_nmp, rep, p)) 1506 return (EINTR); 1507 *flagp |= NFSMNT_WANTSND; 1508 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1509 slptimeo); 1510 if (slpflag == PCATCH) { 1511 slpflag = 0; 1512 slptimeo = 2 * hz; 1513 } 1514 } 1515 *flagp |= NFSMNT_SNDLOCK; 1516 return (0); 1517 } 1518 1519 /* 1520 * Unlock the stream socket for others. 1521 */ 1522 void 1523 nfs_sndunlock(flagp) 1524 int *flagp; 1525 { 1526 1527 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1528 panic("nfs sndunlock"); 1529 *flagp &= ~NFSMNT_SNDLOCK; 1530 if (*flagp & NFSMNT_WANTSND) { 1531 *flagp &= ~NFSMNT_WANTSND; 1532 wakeup((caddr_t)flagp); 1533 } 1534 } 1535 1536 int 1537 nfs_rcvlock(rep) 1538 struct nfsreq *rep; 1539 { 1540 struct nfsmount *nmp = rep->r_nmp; 1541 int *flagp = &nmp->nm_iflag; 1542 int slpflag, slptimeo = 0; 1543 1544 if (*flagp & NFSMNT_DISMNT) 1545 return EIO; 1546 1547 if (*flagp & NFSMNT_INT) 1548 slpflag = PCATCH; 1549 else 1550 slpflag = 0; 1551 while (*flagp & NFSMNT_RCVLOCK) { 1552 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1553 return (EINTR); 1554 *flagp |= NFSMNT_WANTRCV; 1555 nmp->nm_waiters++; 1556 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1557 slptimeo); 1558 nmp->nm_waiters--; 1559 if (*flagp & NFSMNT_DISMNT) { 1560 wakeup(&nmp->nm_waiters); 1561 return EIO; 1562 } 1563 /* If our reply was received while we were sleeping, 1564 * then just return without taking the lock to avoid a 1565 * situation where a single iod could 'capture' the 1566 * receive lock. 1567 */ 1568 if (rep->r_mrep != NULL) 1569 return (EALREADY); 1570 if (slpflag == PCATCH) { 1571 slpflag = 0; 1572 slptimeo = 2 * hz; 1573 } 1574 } 1575 *flagp |= NFSMNT_RCVLOCK; 1576 return (0); 1577 } 1578 1579 /* 1580 * Unlock the stream socket for others. 1581 */ 1582 void 1583 nfs_rcvunlock(flagp) 1584 int *flagp; 1585 { 1586 1587 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1588 panic("nfs rcvunlock"); 1589 *flagp &= ~NFSMNT_RCVLOCK; 1590 if (*flagp & NFSMNT_WANTRCV) { 1591 *flagp &= ~NFSMNT_WANTRCV; 1592 wakeup((caddr_t)flagp); 1593 } 1594 } 1595 1596 /* 1597 * Parse an RPC request 1598 * - verify it 1599 * - fill in the cred struct. 1600 */ 1601 int 1602 nfs_getreq(nd, nfsd, has_header) 1603 struct nfsrv_descript *nd; 1604 struct nfsd *nfsd; 1605 int has_header; 1606 { 1607 int len, i; 1608 u_int32_t *tl; 1609 int32_t t1; 1610 struct uio uio; 1611 struct iovec iov; 1612 caddr_t dpos, cp2, cp; 1613 u_int32_t nfsvers, auth_type; 1614 uid_t nickuid; 1615 int error = 0, nqnfs = 0, ticklen; 1616 struct mbuf *mrep, *md; 1617 struct nfsuid *nuidp; 1618 struct timeval tvin, tvout; 1619 1620 mrep = nd->nd_mrep; 1621 md = nd->nd_md; 1622 dpos = nd->nd_dpos; 1623 if (has_header) { 1624 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1625 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1626 if (*tl++ != rpc_call) { 1627 m_freem(mrep); 1628 return (EBADRPC); 1629 } 1630 } else 1631 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1632 nd->nd_repstat = 0; 1633 nd->nd_flag = 0; 1634 if (*tl++ != rpc_vers) { 1635 nd->nd_repstat = ERPCMISMATCH; 1636 nd->nd_procnum = NFSPROC_NOOP; 1637 return (0); 1638 } 1639 if (*tl != nfs_prog) { 1640 if (*tl == nqnfs_prog) 1641 nqnfs++; 1642 else { 1643 nd->nd_repstat = EPROGUNAVAIL; 1644 nd->nd_procnum = NFSPROC_NOOP; 1645 return (0); 1646 } 1647 } 1648 tl++; 1649 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1650 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1651 (nfsvers != NQNFS_VER3 && nqnfs)) { 1652 nd->nd_repstat = EPROGMISMATCH; 1653 nd->nd_procnum = NFSPROC_NOOP; 1654 return (0); 1655 } 1656 if (nqnfs) 1657 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1658 else if (nfsvers == NFS_VER3) 1659 nd->nd_flag = ND_NFSV3; 1660 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1661 if (nd->nd_procnum == NFSPROC_NULL) 1662 return (0); 1663 if (nd->nd_procnum >= NFS_NPROCS || 1664 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1665 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1666 nd->nd_repstat = EPROCUNAVAIL; 1667 nd->nd_procnum = NFSPROC_NOOP; 1668 return (0); 1669 } 1670 if ((nd->nd_flag & ND_NFSV3) == 0) 1671 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1672 auth_type = *tl++; 1673 len = fxdr_unsigned(int, *tl++); 1674 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1675 m_freem(mrep); 1676 return (EBADRPC); 1677 } 1678 1679 nd->nd_flag &= ~ND_KERBAUTH; 1680 /* 1681 * Handle auth_unix or auth_kerb. 1682 */ 1683 if (auth_type == rpc_auth_unix) { 1684 len = fxdr_unsigned(int, *++tl); 1685 if (len < 0 || len > NFS_MAXNAMLEN) { 1686 m_freem(mrep); 1687 return (EBADRPC); 1688 } 1689 nfsm_adv(nfsm_rndup(len)); 1690 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1691 memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred)); 1692 nd->nd_cr.cr_ref = 1; 1693 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1694 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1695 len = fxdr_unsigned(int, *tl); 1696 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1697 m_freem(mrep); 1698 return (EBADRPC); 1699 } 1700 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1701 for (i = 0; i < len; i++) 1702 if (i < NGROUPS) 1703 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1704 else 1705 tl++; 1706 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1707 if (nd->nd_cr.cr_ngroups > 1) 1708 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1709 len = fxdr_unsigned(int, *++tl); 1710 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1711 m_freem(mrep); 1712 return (EBADRPC); 1713 } 1714 if (len > 0) 1715 nfsm_adv(nfsm_rndup(len)); 1716 } else if (auth_type == rpc_auth_kerb) { 1717 switch (fxdr_unsigned(int, *tl++)) { 1718 case RPCAKN_FULLNAME: 1719 ticklen = fxdr_unsigned(int, *tl); 1720 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1721 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1722 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1723 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1724 m_freem(mrep); 1725 return (EBADRPC); 1726 } 1727 uio.uio_offset = 0; 1728 uio.uio_iov = &iov; 1729 uio.uio_iovcnt = 1; 1730 uio.uio_segflg = UIO_SYSSPACE; 1731 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1732 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1733 nfsm_mtouio(&uio, uio.uio_resid); 1734 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1735 if (*tl++ != rpc_auth_kerb || 1736 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1737 printf("Bad kerb verifier\n"); 1738 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1739 nd->nd_procnum = NFSPROC_NOOP; 1740 return (0); 1741 } 1742 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1743 tl = (u_int32_t *)cp; 1744 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1745 printf("Not fullname kerb verifier\n"); 1746 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1747 nd->nd_procnum = NFSPROC_NOOP; 1748 return (0); 1749 } 1750 cp += NFSX_UNSIGNED; 1751 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 1752 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1753 nd->nd_flag |= ND_KERBFULL; 1754 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1755 break; 1756 case RPCAKN_NICKNAME: 1757 if (len != 2 * NFSX_UNSIGNED) { 1758 printf("Kerb nickname short\n"); 1759 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1760 nd->nd_procnum = NFSPROC_NOOP; 1761 return (0); 1762 } 1763 nickuid = fxdr_unsigned(uid_t, *tl); 1764 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1765 if (*tl++ != rpc_auth_kerb || 1766 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1767 printf("Kerb nick verifier bad\n"); 1768 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1769 nd->nd_procnum = NFSPROC_NOOP; 1770 return (0); 1771 } 1772 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1773 tvin.tv_sec = *tl++; 1774 tvin.tv_usec = *tl; 1775 1776 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; 1777 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1778 if (nuidp->nu_cr.cr_uid == nickuid && 1779 (!nd->nd_nam2 || 1780 netaddr_match(NU_NETFAM(nuidp), 1781 &nuidp->nu_haddr, nd->nd_nam2))) 1782 break; 1783 } 1784 if (!nuidp) { 1785 nd->nd_repstat = 1786 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1787 nd->nd_procnum = NFSPROC_NOOP; 1788 return (0); 1789 } 1790 1791 /* 1792 * Now, decrypt the timestamp using the session key 1793 * and validate it. 1794 */ 1795 #ifdef NFSKERB 1796 XXX 1797 #endif 1798 1799 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1800 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1801 if (nuidp->nu_expire < time.tv_sec || 1802 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1803 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1804 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1805 nuidp->nu_expire = 0; 1806 nd->nd_repstat = 1807 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1808 nd->nd_procnum = NFSPROC_NOOP; 1809 return (0); 1810 } 1811 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1812 nd->nd_flag |= ND_KERBNICK; 1813 }; 1814 } else { 1815 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1816 nd->nd_procnum = NFSPROC_NOOP; 1817 return (0); 1818 } 1819 1820 /* 1821 * For nqnfs, get piggybacked lease request. 1822 */ 1823 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1824 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1825 nd->nd_flag |= fxdr_unsigned(int, *tl); 1826 if (nd->nd_flag & ND_LEASE) { 1827 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1828 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 1829 } else 1830 nd->nd_duration = NQ_MINLEASE; 1831 } else 1832 nd->nd_duration = NQ_MINLEASE; 1833 nd->nd_md = md; 1834 nd->nd_dpos = dpos; 1835 return (0); 1836 nfsmout: 1837 return (error); 1838 } 1839 1840 int 1841 nfs_msg(p, server, msg) 1842 struct proc *p; 1843 char *server, *msg; 1844 { 1845 tpr_t tpr; 1846 1847 if (p) 1848 tpr = tprintf_open(p); 1849 else 1850 tpr = NULL; 1851 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1852 tprintf_close(tpr); 1853 return (0); 1854 } 1855 1856 #ifdef NFSSERVER 1857 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 1858 struct nfssvc_sock *, struct proc *, 1859 struct mbuf **)) = { 1860 nfsrv_null, 1861 nfsrv_getattr, 1862 nfsrv_setattr, 1863 nfsrv_lookup, 1864 nfsrv3_access, 1865 nfsrv_readlink, 1866 nfsrv_read, 1867 nfsrv_write, 1868 nfsrv_create, 1869 nfsrv_mkdir, 1870 nfsrv_symlink, 1871 nfsrv_mknod, 1872 nfsrv_remove, 1873 nfsrv_rmdir, 1874 nfsrv_rename, 1875 nfsrv_link, 1876 nfsrv_readdir, 1877 nfsrv_readdirplus, 1878 nfsrv_statfs, 1879 nfsrv_fsinfo, 1880 nfsrv_pathconf, 1881 nfsrv_commit, 1882 nqnfsrv_getlease, 1883 nqnfsrv_vacated, 1884 nfsrv_noop, 1885 nfsrv_noop 1886 }; 1887 1888 /* 1889 * Socket upcall routine for the nfsd sockets. 1890 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1891 * Essentially do as much as possible non-blocking, else punt and it will 1892 * be called with M_WAIT from an nfsd. 1893 */ 1894 void 1895 nfsrv_rcv(so, arg, waitflag) 1896 struct socket *so; 1897 caddr_t arg; 1898 int waitflag; 1899 { 1900 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1901 struct mbuf *m; 1902 struct mbuf *mp, *nam; 1903 struct uio auio; 1904 int flags, error; 1905 1906 if ((slp->ns_flag & SLP_VALID) == 0) 1907 return; 1908 #ifdef notdef 1909 /* 1910 * Define this to test for nfsds handling this under heavy load. 1911 */ 1912 if (waitflag == M_DONTWAIT) { 1913 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1914 } 1915 #endif 1916 auio.uio_procp = NULL; 1917 if (so->so_type == SOCK_STREAM) { 1918 /* 1919 * If there are already records on the queue, defer soreceive() 1920 * to an nfsd so that there is feedback to the TCP layer that 1921 * the nfs servers are heavily loaded. 1922 */ 1923 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1924 slp->ns_flag |= SLP_NEEDQ; 1925 goto dorecs; 1926 } 1927 1928 /* 1929 * Do soreceive(). 1930 */ 1931 auio.uio_resid = 1000000000; 1932 flags = MSG_DONTWAIT; 1933 error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1934 if (error || mp == (struct mbuf *)0) { 1935 if (error == EWOULDBLOCK) 1936 slp->ns_flag |= SLP_NEEDQ; 1937 else 1938 slp->ns_flag |= SLP_DISCONN; 1939 goto dorecs; 1940 } 1941 m = mp; 1942 if (slp->ns_rawend) { 1943 slp->ns_rawend->m_next = m; 1944 slp->ns_cc += 1000000000 - auio.uio_resid; 1945 } else { 1946 slp->ns_raw = m; 1947 slp->ns_cc = 1000000000 - auio.uio_resid; 1948 } 1949 while (m->m_next) 1950 m = m->m_next; 1951 slp->ns_rawend = m; 1952 1953 /* 1954 * Now try and parse record(s) out of the raw stream data. 1955 */ 1956 error = nfsrv_getstream(slp, waitflag); 1957 if (error) { 1958 if (error == EPERM) 1959 slp->ns_flag |= SLP_DISCONN; 1960 else 1961 slp->ns_flag |= SLP_NEEDQ; 1962 } 1963 } else { 1964 do { 1965 auio.uio_resid = 1000000000; 1966 flags = MSG_DONTWAIT; 1967 error = (*so->so_receive)(so, &nam, &auio, &mp, 1968 (struct mbuf **)0, &flags); 1969 if (mp) { 1970 if (nam) { 1971 m = nam; 1972 m->m_next = mp; 1973 } else 1974 m = mp; 1975 if (slp->ns_recend) 1976 slp->ns_recend->m_nextpkt = m; 1977 else 1978 slp->ns_rec = m; 1979 slp->ns_recend = m; 1980 m->m_nextpkt = (struct mbuf *)0; 1981 } 1982 if (error) { 1983 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1984 && error != EWOULDBLOCK) { 1985 slp->ns_flag |= SLP_DISCONN; 1986 goto dorecs; 1987 } 1988 } 1989 } while (mp); 1990 } 1991 1992 /* 1993 * Now try and process the request records, non-blocking. 1994 */ 1995 dorecs: 1996 if (waitflag == M_DONTWAIT && 1997 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1998 nfsrv_wakenfsd(slp); 1999 } 2000 2001 /* 2002 * Try and extract an RPC request from the mbuf data list received on a 2003 * stream socket. The "waitflag" argument indicates whether or not it 2004 * can sleep. 2005 */ 2006 int 2007 nfsrv_getstream(slp, waitflag) 2008 struct nfssvc_sock *slp; 2009 int waitflag; 2010 { 2011 struct mbuf *m, **mpp; 2012 char *cp1, *cp2; 2013 int len; 2014 struct mbuf *om, *m2, *recm; 2015 u_int32_t recmark; 2016 2017 if (slp->ns_flag & SLP_GETSTREAM) 2018 panic("nfs getstream"); 2019 slp->ns_flag |= SLP_GETSTREAM; 2020 for (;;) { 2021 if (slp->ns_reclen == 0) { 2022 if (slp->ns_cc < NFSX_UNSIGNED) { 2023 slp->ns_flag &= ~SLP_GETSTREAM; 2024 return (0); 2025 } 2026 m = slp->ns_raw; 2027 if (m->m_len >= NFSX_UNSIGNED) { 2028 memcpy((caddr_t)&recmark, mtod(m, caddr_t), NFSX_UNSIGNED); 2029 m->m_data += NFSX_UNSIGNED; 2030 m->m_len -= NFSX_UNSIGNED; 2031 } else { 2032 cp1 = (caddr_t)&recmark; 2033 cp2 = mtod(m, caddr_t); 2034 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 2035 while (m->m_len == 0) { 2036 m = m->m_next; 2037 cp2 = mtod(m, caddr_t); 2038 } 2039 *cp1++ = *cp2++; 2040 m->m_data++; 2041 m->m_len--; 2042 } 2043 } 2044 slp->ns_cc -= NFSX_UNSIGNED; 2045 recmark = ntohl(recmark); 2046 slp->ns_reclen = recmark & ~0x80000000; 2047 if (recmark & 0x80000000) 2048 slp->ns_flag |= SLP_LASTFRAG; 2049 else 2050 slp->ns_flag &= ~SLP_LASTFRAG; 2051 if (slp->ns_reclen > NFS_MAXPACKET) { 2052 slp->ns_flag &= ~SLP_GETSTREAM; 2053 return (EPERM); 2054 } 2055 } 2056 2057 /* 2058 * Now get the record part. 2059 * 2060 * Note that slp->ns_reclen may be 0. Linux sometimes 2061 * generates 0-length records. 2062 */ 2063 if (slp->ns_cc == slp->ns_reclen) { 2064 recm = slp->ns_raw; 2065 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2066 slp->ns_cc = slp->ns_reclen = 0; 2067 } else if (slp->ns_cc > slp->ns_reclen) { 2068 len = 0; 2069 m = slp->ns_raw; 2070 recm = om = NULL; 2071 2072 while (len < slp->ns_reclen) { 2073 if ((len + m->m_len) > slp->ns_reclen) { 2074 size_t left = slp->ns_reclen - len; 2075 2076 MGETHDR(m2, waitflag, m->m_type); 2077 if (m2 == NULL) { 2078 slp->ns_flag &= ~SLP_GETSTREAM; 2079 return (EWOULDBLOCK); 2080 } 2081 if (left > MHLEN) { 2082 MCLGET(m2, waitflag); 2083 if (!(m2->m_flags & M_EXT)) { 2084 m_freem(m2); 2085 slp->ns_flag &= ~SLP_GETSTREAM; 2086 return (EWOULDBLOCK); 2087 } 2088 } 2089 memcpy(mtod(m2, caddr_t), mtod(m, caddr_t), 2090 left); 2091 m2->m_len = left; 2092 m->m_data += left; 2093 m->m_len -= left; 2094 if (om) { 2095 om->m_next = m2; 2096 recm = slp->ns_raw; 2097 } else 2098 recm = m2; 2099 len = slp->ns_reclen; 2100 } else if ((len + m->m_len) == slp->ns_reclen) { 2101 om = m; 2102 len += m->m_len; 2103 m = m->m_next; 2104 recm = slp->ns_raw; 2105 om->m_next = (struct mbuf *)0; 2106 } else { 2107 om = m; 2108 len += m->m_len; 2109 m = m->m_next; 2110 } 2111 } 2112 slp->ns_raw = m; 2113 slp->ns_cc -= len; 2114 slp->ns_reclen = 0; 2115 } else { 2116 slp->ns_flag &= ~SLP_GETSTREAM; 2117 return (0); 2118 } 2119 2120 /* 2121 * Accumulate the fragments into a record. 2122 */ 2123 mpp = &slp->ns_frag; 2124 while (*mpp) 2125 mpp = &((*mpp)->m_next); 2126 *mpp = recm; 2127 if (slp->ns_flag & SLP_LASTFRAG) { 2128 if (slp->ns_recend) 2129 slp->ns_recend->m_nextpkt = slp->ns_frag; 2130 else 2131 slp->ns_rec = slp->ns_frag; 2132 slp->ns_recend = slp->ns_frag; 2133 slp->ns_frag = (struct mbuf *)0; 2134 } 2135 } 2136 } 2137 2138 /* 2139 * Parse an RPC header. 2140 */ 2141 int 2142 nfsrv_dorec(slp, nfsd, ndp) 2143 struct nfssvc_sock *slp; 2144 struct nfsd *nfsd; 2145 struct nfsrv_descript **ndp; 2146 { 2147 struct mbuf *m, *nam; 2148 struct nfsrv_descript *nd; 2149 int error; 2150 2151 *ndp = NULL; 2152 if ((slp->ns_flag & SLP_VALID) == 0 || 2153 (m = slp->ns_rec) == (struct mbuf *)0) 2154 return (ENOBUFS); 2155 slp->ns_rec = m->m_nextpkt; 2156 if (slp->ns_rec) 2157 m->m_nextpkt = (struct mbuf *)0; 2158 else 2159 slp->ns_recend = (struct mbuf *)0; 2160 if (m->m_type == MT_SONAME) { 2161 nam = m; 2162 m = m->m_next; 2163 nam->m_next = NULL; 2164 } else 2165 nam = NULL; 2166 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), 2167 M_NFSRVDESC, M_WAITOK); 2168 nd->nd_md = nd->nd_mrep = m; 2169 nd->nd_nam2 = nam; 2170 nd->nd_dpos = mtod(m, caddr_t); 2171 error = nfs_getreq(nd, nfsd, TRUE); 2172 if (error) { 2173 m_freem(nam); 2174 free((caddr_t)nd, M_NFSRVDESC); 2175 return (error); 2176 } 2177 *ndp = nd; 2178 nfsd->nfsd_nd = nd; 2179 return (0); 2180 } 2181 2182 2183 /* 2184 * Search for a sleeping nfsd and wake it up. 2185 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2186 * running nfsds will go look for the work in the nfssvc_sock list. 2187 */ 2188 void 2189 nfsrv_wakenfsd(slp) 2190 struct nfssvc_sock *slp; 2191 { 2192 struct nfsd *nd; 2193 2194 if ((slp->ns_flag & SLP_VALID) == 0) 2195 return; 2196 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) { 2197 if (nd->nfsd_flag & NFSD_WAITING) { 2198 nd->nfsd_flag &= ~NFSD_WAITING; 2199 if (nd->nfsd_slp) 2200 panic("nfsd wakeup"); 2201 slp->ns_sref++; 2202 nd->nfsd_slp = slp; 2203 wakeup((caddr_t)nd); 2204 return; 2205 } 2206 } 2207 slp->ns_flag |= SLP_DOREC; 2208 nfsd_head_flag |= NFSD_CHECKSLP; 2209 } 2210 #endif /* NFSSERVER */ 2211
Indexes created Tue Sep 30 11:09:46 GMT 2025