1 /* $NetBSD: nfs_socket.c,v 1.74 2002/05/12 23:04:36 matt Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1991, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 39 */ 40 41 /* 42 * Socket operations for use by nfs 43 */ 44 45 #include <sys/cdefs.h> 46 __KERNEL_RCSID(0, "$NetBSD: nfs_socket.c,v 1.74 2002/05/12 23:04:36 matt Exp $"); 47 48 #include "fs_nfs.h" 49 #include "opt_nfs.h" 50 #include "opt_nfsserver.h" 51 #include "opt_inet.h" 52 53 #include <sys/param.h> 54 #include <sys/systm.h> 55 #include <sys/callout.h> 56 #include <sys/proc.h> 57 #include <sys/mount.h> 58 #include <sys/kernel.h> 59 #include <sys/mbuf.h> 60 #include <sys/vnode.h> 61 #include <sys/domain.h> 62 #include <sys/protosw.h> 63 #include <sys/socket.h> 64 #include <sys/socketvar.h> 65 #include <sys/syslog.h> 66 #include <sys/tprintf.h> 67 #include <sys/namei.h> 68 #include <sys/signal.h> 69 #include <sys/signalvar.h> 70 71 #include <netinet/in.h> 72 #include <netinet/tcp.h> 73 74 #include <nfs/rpcv2.h> 75 #include <nfs/nfsproto.h> 76 #include <nfs/nfs.h> 77 #include <nfs/xdr_subs.h> 78 #include <nfs/nfsm_subs.h> 79 #include <nfs/nfsmount.h> 80 #include <nfs/nfsnode.h> 81 #include <nfs/nfsrtt.h> 82 #include <nfs/nqnfs.h> 83 #include <nfs/nfs_var.h> 84 85 #define TRUE 1 86 #define FALSE 0 87 88 /* 89 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 90 * Use the mean and mean deviation of rtt for the appropriate type of rpc 91 * for the frequent rpcs and a default for the others. 92 * The justification for doing "other" this way is that these rpcs 93 * happen so infrequently that timer est. would probably be stale. 94 * Also, since many of these rpcs are 95 * non-idempotent, a conservative timeout is desired. 96 * getattr, lookup - A+2D 97 * read, write - A+4D 98 * other - nm_timeo 99 */ 100 #define NFS_RTO(n, t) \ 101 ((t) == 0 ? (n)->nm_timeo : \ 102 ((t) < 3 ? \ 103 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 104 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 105 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 106 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 107 /* 108 * External data, mostly RPC constants in XDR form 109 */ 110 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 111 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, 112 rpc_auth_kerb; 113 extern u_int32_t nfs_prog, nqnfs_prog; 114 extern time_t nqnfsstarttime; 115 extern int nfsv3_procid[NFS_NPROCS]; 116 extern int nfs_ticks; 117 118 /* 119 * Defines which timer to use for the procnum. 120 * 0 - default 121 * 1 - getattr 122 * 2 - lookup 123 * 3 - read 124 * 4 - write 125 */ 126 static const int proct[NFS_NPROCS] = { 127 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0, 128 0, 0, 0, 129 }; 130 131 /* 132 * There is a congestion window for outstanding rpcs maintained per mount 133 * point. The cwnd size is adjusted in roughly the way that: 134 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 135 * SIGCOMM '88". ACM, August 1988. 136 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 137 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 138 * of rpcs is in progress. 139 * (The sent count and cwnd are scaled for integer arith.) 140 * Variants of "slow start" were tried and were found to be too much of a 141 * performance hit (ave. rtt 3 times larger), 142 * I suspect due to the large rtt that nfs rpcs have. 143 */ 144 #define NFS_CWNDSCALE 256 145 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 146 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 147 int nfsrtton = 0; 148 struct nfsrtt nfsrtt; 149 struct nfsreqhead nfs_reqq; 150 151 struct callout nfs_timer_ch = CALLOUT_INITIALIZER; 152 153 /* 154 * Initialize sockets and congestion for a new NFS connection. 155 * We do not free the sockaddr if error. 156 */ 157 int 158 nfs_connect(nmp, rep) 159 struct nfsmount *nmp; 160 struct nfsreq *rep; 161 { 162 struct socket *so; 163 int s, error, rcvreserve, sndreserve; 164 struct sockaddr *saddr; 165 struct sockaddr_in *sin; 166 #ifdef INET6 167 struct sockaddr_in6 *sin6; 168 #endif 169 struct mbuf *m; 170 171 nmp->nm_so = (struct socket *)0; 172 saddr = mtod(nmp->nm_nam, struct sockaddr *); 173 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 174 nmp->nm_soproto); 175 if (error) 176 goto bad; 177 so = nmp->nm_so; 178 nmp->nm_soflags = so->so_proto->pr_flags; 179 180 /* 181 * Some servers require that the client port be a reserved port number. 182 */ 183 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 184 MGET(m, M_WAIT, MT_SOOPTS); 185 *mtod(m, int32_t *) = IP_PORTRANGE_LOW; 186 m->m_len = sizeof(int32_t); 187 if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m))) 188 goto bad; 189 MGET(m, M_WAIT, MT_SONAME); 190 sin = mtod(m, struct sockaddr_in *); 191 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 192 sin->sin_family = AF_INET; 193 sin->sin_addr.s_addr = INADDR_ANY; 194 sin->sin_port = 0; 195 error = sobind(so, m, &proc0); 196 m_freem(m); 197 if (error) 198 goto bad; 199 } 200 #ifdef INET6 201 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) { 202 MGET(m, M_WAIT, MT_SOOPTS); 203 *mtod(m, int32_t *) = IPV6_PORTRANGE_LOW; 204 m->m_len = sizeof(int32_t); 205 if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m))) 206 goto bad; 207 MGET(m, M_WAIT, MT_SONAME); 208 sin6 = mtod(m, struct sockaddr_in6 *); 209 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6); 210 sin6->sin6_family = AF_INET6; 211 sin6->sin6_addr = in6addr_any; 212 sin6->sin6_port = 0; 213 error = sobind(so, m, &proc0); 214 m_freem(m); 215 if (error) 216 goto bad; 217 } 218 #endif 219 220 /* 221 * Protocols that do not require connections may be optionally left 222 * unconnected for servers that reply from a port other than NFS_PORT. 223 */ 224 if (nmp->nm_flag & NFSMNT_NOCONN) { 225 if (nmp->nm_soflags & PR_CONNREQUIRED) { 226 error = ENOTCONN; 227 goto bad; 228 } 229 } else { 230 error = soconnect(so, nmp->nm_nam); 231 if (error) 232 goto bad; 233 234 /* 235 * Wait for the connection to complete. Cribbed from the 236 * connect system call but with the wait timing out so 237 * that interruptible mounts don't hang here for a long time. 238 */ 239 s = splsoftnet(); 240 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 241 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 242 "nfscn1", 2 * hz); 243 if ((so->so_state & SS_ISCONNECTING) && 244 so->so_error == 0 && rep && 245 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ 246 so->so_state &= ~SS_ISCONNECTING; 247 splx(s); 248 goto bad; 249 } 250 } 251 if (so->so_error) { 252 error = so->so_error; 253 so->so_error = 0; 254 splx(s); 255 goto bad; 256 } 257 splx(s); 258 } 259 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 260 so->so_rcv.sb_timeo = (5 * hz); 261 so->so_snd.sb_timeo = (5 * hz); 262 } else { 263 so->so_rcv.sb_timeo = 0; 264 so->so_snd.sb_timeo = 0; 265 } 266 if (nmp->nm_sotype == SOCK_DGRAM) { 267 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 268 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 269 NFS_MAXPKTHDR) * 2; 270 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 271 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 272 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 273 NFS_MAXPKTHDR) * 2; 274 } else { 275 if (nmp->nm_sotype != SOCK_STREAM) 276 panic("nfscon sotype"); 277 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 278 MGET(m, M_WAIT, MT_SOOPTS); 279 *mtod(m, int32_t *) = 1; 280 m->m_len = sizeof(int32_t); 281 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 282 } 283 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 284 MGET(m, M_WAIT, MT_SOOPTS); 285 *mtod(m, int32_t *) = 1; 286 m->m_len = sizeof(int32_t); 287 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 288 } 289 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 290 sizeof (u_int32_t)) * 2; 291 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 292 sizeof (u_int32_t)) * 2; 293 } 294 error = soreserve(so, sndreserve, rcvreserve); 295 if (error) 296 goto bad; 297 so->so_rcv.sb_flags |= SB_NOINTR; 298 so->so_snd.sb_flags |= SB_NOINTR; 299 300 /* Initialize other non-zero congestion variables */ 301 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 302 NFS_TIMEO << 3; 303 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 304 nmp->nm_sdrtt[3] = 0; 305 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 306 nmp->nm_sent = 0; 307 nmp->nm_timeouts = 0; 308 return (0); 309 310 bad: 311 nfs_disconnect(nmp); 312 return (error); 313 } 314 315 /* 316 * Reconnect routine: 317 * Called when a connection is broken on a reliable protocol. 318 * - clean up the old socket 319 * - nfs_connect() again 320 * - set R_MUSTRESEND for all outstanding requests on mount point 321 * If this fails the mount point is DEAD! 322 * nb: Must be called with the nfs_sndlock() set on the mount point. 323 */ 324 int 325 nfs_reconnect(rep) 326 struct nfsreq *rep; 327 { 328 struct nfsreq *rp; 329 struct nfsmount *nmp = rep->r_nmp; 330 int error; 331 332 nfs_disconnect(nmp); 333 while ((error = nfs_connect(nmp, rep)) != 0) { 334 if (error == EINTR || error == ERESTART) 335 return (EINTR); 336 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0); 337 } 338 339 /* 340 * Loop through outstanding request list and fix up all requests 341 * on old socket. 342 */ 343 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 344 if (rp->r_nmp == nmp) 345 rp->r_flags |= R_MUSTRESEND; 346 } 347 return (0); 348 } 349 350 /* 351 * NFS disconnect. Clean up and unlink. 352 */ 353 void 354 nfs_disconnect(nmp) 355 struct nfsmount *nmp; 356 { 357 struct socket *so; 358 int drain = 0; 359 360 if (nmp->nm_so) { 361 so = nmp->nm_so; 362 nmp->nm_so = (struct socket *)0; 363 soshutdown(so, 2); 364 drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0; 365 if (drain) { 366 /* 367 * soshutdown() above should wake up the current 368 * listener. 369 * Now wake up those waiting for the receive lock, and 370 * wait for them to go away unhappy, to prevent *nmp 371 * from evaporating while they're sleeping. 372 */ 373 while (nmp->nm_waiters > 0) { 374 wakeup (&nmp->nm_iflag); 375 (void) tsleep(&nmp->nm_waiters, PVFS, 376 "nfsdis", 0); 377 } 378 } 379 soclose(so); 380 } 381 #ifdef DIAGNOSTIC 382 if (drain && (nmp->nm_waiters > 0)) 383 panic("nfs_disconnect: waiters left after drain?\n"); 384 #endif 385 } 386 387 void 388 nfs_safedisconnect(nmp) 389 struct nfsmount *nmp; 390 { 391 struct nfsreq dummyreq; 392 393 memset(&dummyreq, 0, sizeof(dummyreq)); 394 dummyreq.r_nmp = nmp; 395 nfs_rcvlock(&dummyreq); /* XXX ignored error return */ 396 nfs_disconnect(nmp); 397 nfs_rcvunlock(&nmp->nm_iflag); 398 } 399 400 /* 401 * This is the nfs send routine. For connection based socket types, it 402 * must be called with an nfs_sndlock() on the socket. 403 * "rep == NULL" indicates that it has been called from a server. 404 * For the client side: 405 * - return EINTR if the RPC is terminated, 0 otherwise 406 * - set R_MUSTRESEND if the send fails for any reason 407 * - do any cleanup required by recoverable socket errors (? ? ?) 408 * For the server side: 409 * - return EINTR or ERESTART if interrupted by a signal 410 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 411 * - do any cleanup required by recoverable socket errors (? ? ?) 412 */ 413 int 414 nfs_send(so, nam, top, rep) 415 struct socket *so; 416 struct mbuf *nam; 417 struct mbuf *top; 418 struct nfsreq *rep; 419 { 420 struct mbuf *sendnam; 421 int error, soflags, flags; 422 423 if (rep) { 424 if (rep->r_flags & R_SOFTTERM) { 425 m_freem(top); 426 return (EINTR); 427 } 428 if ((so = rep->r_nmp->nm_so) == NULL) { 429 rep->r_flags |= R_MUSTRESEND; 430 m_freem(top); 431 return (0); 432 } 433 rep->r_flags &= ~R_MUSTRESEND; 434 soflags = rep->r_nmp->nm_soflags; 435 } else 436 soflags = so->so_proto->pr_flags; 437 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 438 sendnam = (struct mbuf *)0; 439 else 440 sendnam = nam; 441 if (so->so_type == SOCK_SEQPACKET) 442 flags = MSG_EOR; 443 else 444 flags = 0; 445 446 error = (*so->so_send)(so, sendnam, (struct uio *)0, top, 447 (struct mbuf *)0, flags); 448 if (error) { 449 if (rep) { 450 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 451 /* 452 * We're too fast for the network/driver, 453 * and UDP isn't flowcontrolled. 454 * We need to resend. This is not fatal, 455 * just try again. 456 * 457 * Could be smarter here by doing some sort 458 * of a backoff, but this is rare. 459 */ 460 rep->r_flags |= R_MUSTRESEND; 461 } else { 462 log(LOG_INFO, "nfs send error %d for %s\n", 463 error, 464 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 465 /* 466 * Deal with errors for the client side. 467 */ 468 if (rep->r_flags & R_SOFTTERM) 469 error = EINTR; 470 else 471 rep->r_flags |= R_MUSTRESEND; 472 } 473 } else { 474 /* 475 * See above. This error can happen under normal 476 * circumstances and the log is too noisy. 477 * The error will still show up in nfsstat. 478 */ 479 if (error != ENOBUFS || so->so_type != SOCK_DGRAM) 480 log(LOG_INFO, "nfsd send error %d\n", error); 481 } 482 483 /* 484 * Handle any recoverable (soft) socket errors here. (? ? ?) 485 */ 486 if (error != EINTR && error != ERESTART && 487 error != EWOULDBLOCK && error != EPIPE) 488 error = 0; 489 } 490 return (error); 491 } 492 493 #ifdef NFS 494 /* 495 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 496 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 497 * Mark and consolidate the data into a new mbuf list. 498 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 499 * small mbufs. 500 * For SOCK_STREAM we must be very careful to read an entire record once 501 * we have read any of it, even if the system call has been interrupted. 502 */ 503 int 504 nfs_receive(rep, aname, mp) 505 struct nfsreq *rep; 506 struct mbuf **aname; 507 struct mbuf **mp; 508 { 509 struct socket *so; 510 struct uio auio; 511 struct iovec aio; 512 struct mbuf *m; 513 struct mbuf *control; 514 u_int32_t len; 515 struct mbuf **getnam; 516 int error, sotype, rcvflg; 517 struct proc *p = curproc; /* XXX */ 518 519 /* 520 * Set up arguments for soreceive() 521 */ 522 *mp = (struct mbuf *)0; 523 *aname = (struct mbuf *)0; 524 sotype = rep->r_nmp->nm_sotype; 525 526 /* 527 * For reliable protocols, lock against other senders/receivers 528 * in case a reconnect is necessary. 529 * For SOCK_STREAM, first get the Record Mark to find out how much 530 * more there is to get. 531 * We must lock the socket against other receivers 532 * until we have an entire rpc request/reply. 533 */ 534 if (sotype != SOCK_DGRAM) { 535 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 536 if (error) 537 return (error); 538 tryagain: 539 /* 540 * Check for fatal errors and resending request. 541 */ 542 /* 543 * Ugh: If a reconnect attempt just happened, nm_so 544 * would have changed. NULL indicates a failed 545 * attempt that has essentially shut down this 546 * mount point. 547 */ 548 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 549 nfs_sndunlock(&rep->r_nmp->nm_iflag); 550 return (EINTR); 551 } 552 so = rep->r_nmp->nm_so; 553 if (!so) { 554 error = nfs_reconnect(rep); 555 if (error) { 556 nfs_sndunlock(&rep->r_nmp->nm_iflag); 557 return (error); 558 } 559 goto tryagain; 560 } 561 while (rep->r_flags & R_MUSTRESEND) { 562 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 563 nfsstats.rpcretries++; 564 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 565 if (error) { 566 if (error == EINTR || error == ERESTART || 567 (error = nfs_reconnect(rep)) != 0) { 568 nfs_sndunlock(&rep->r_nmp->nm_iflag); 569 return (error); 570 } 571 goto tryagain; 572 } 573 } 574 nfs_sndunlock(&rep->r_nmp->nm_iflag); 575 if (sotype == SOCK_STREAM) { 576 aio.iov_base = (caddr_t) &len; 577 aio.iov_len = sizeof(u_int32_t); 578 auio.uio_iov = &aio; 579 auio.uio_iovcnt = 1; 580 auio.uio_segflg = UIO_SYSSPACE; 581 auio.uio_rw = UIO_READ; 582 auio.uio_offset = 0; 583 auio.uio_resid = sizeof(u_int32_t); 584 auio.uio_procp = p; 585 do { 586 rcvflg = MSG_WAITALL; 587 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 588 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 589 if (error == EWOULDBLOCK && rep) { 590 if (rep->r_flags & R_SOFTTERM) 591 return (EINTR); 592 } 593 } while (error == EWOULDBLOCK); 594 if (!error && auio.uio_resid > 0) { 595 /* 596 * Don't log a 0 byte receive; it means 597 * that the socket has been closed, and 598 * can happen during normal operation 599 * (forcible unmount or Solaris server). 600 */ 601 if (auio.uio_resid != sizeof (u_int32_t)) 602 log(LOG_INFO, 603 "short receive (%lu/%lu) from nfs server %s\n", 604 (u_long)sizeof(u_int32_t) - auio.uio_resid, 605 (u_long)sizeof(u_int32_t), 606 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 607 error = EPIPE; 608 } 609 if (error) 610 goto errout; 611 len = ntohl(len) & ~0x80000000; 612 /* 613 * This is SERIOUS! We are out of sync with the sender 614 * and forcing a disconnect/reconnect is all I can do. 615 */ 616 if (len > NFS_MAXPACKET) { 617 log(LOG_ERR, "%s (%d) from nfs server %s\n", 618 "impossible packet length", 619 len, 620 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 621 error = EFBIG; 622 goto errout; 623 } 624 auio.uio_resid = len; 625 do { 626 rcvflg = MSG_WAITALL; 627 error = (*so->so_receive)(so, (struct mbuf **)0, 628 &auio, mp, (struct mbuf **)0, &rcvflg); 629 } while (error == EWOULDBLOCK || error == EINTR || 630 error == ERESTART); 631 if (!error && auio.uio_resid > 0) { 632 if (len != auio.uio_resid) 633 log(LOG_INFO, 634 "short receive (%lu/%d) from nfs server %s\n", 635 (u_long)len - auio.uio_resid, len, 636 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 637 error = EPIPE; 638 } 639 } else { 640 /* 641 * NB: Since uio_resid is big, MSG_WAITALL is ignored 642 * and soreceive() will return when it has either a 643 * control msg or a data msg. 644 * We have no use for control msg., but must grab them 645 * and then throw them away so we know what is going 646 * on. 647 */ 648 auio.uio_resid = len = 100000000; /* Anything Big */ 649 auio.uio_procp = p; 650 do { 651 rcvflg = 0; 652 error = (*so->so_receive)(so, (struct mbuf **)0, 653 &auio, mp, &control, &rcvflg); 654 if (control) 655 m_freem(control); 656 if (error == EWOULDBLOCK && rep) { 657 if (rep->r_flags & R_SOFTTERM) 658 return (EINTR); 659 } 660 } while (error == EWOULDBLOCK || 661 (!error && *mp == NULL && control)); 662 if ((rcvflg & MSG_EOR) == 0) 663 printf("Egad!!\n"); 664 if (!error && *mp == NULL) 665 error = EPIPE; 666 len -= auio.uio_resid; 667 } 668 errout: 669 if (error && error != EINTR && error != ERESTART) { 670 m_freem(*mp); 671 *mp = (struct mbuf *)0; 672 if (error != EPIPE) 673 log(LOG_INFO, 674 "receive error %d from nfs server %s\n", 675 error, 676 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 677 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 678 if (!error) 679 error = nfs_reconnect(rep); 680 if (!error) 681 goto tryagain; 682 else 683 nfs_sndunlock(&rep->r_nmp->nm_iflag); 684 } 685 } else { 686 if ((so = rep->r_nmp->nm_so) == NULL) 687 return (EACCES); 688 if (so->so_state & SS_ISCONNECTED) 689 getnam = (struct mbuf **)0; 690 else 691 getnam = aname; 692 auio.uio_resid = len = 1000000; 693 auio.uio_procp = p; 694 do { 695 rcvflg = 0; 696 error = (*so->so_receive)(so, getnam, &auio, mp, 697 (struct mbuf **)0, &rcvflg); 698 if (error == EWOULDBLOCK && 699 (rep->r_flags & R_SOFTTERM)) 700 return (EINTR); 701 } while (error == EWOULDBLOCK); 702 len -= auio.uio_resid; 703 if (!error && *mp == NULL) 704 error = EPIPE; 705 } 706 if (error) { 707 m_freem(*mp); 708 *mp = (struct mbuf *)0; 709 } 710 return (error); 711 } 712 713 /* 714 * Implement receipt of reply on a socket. 715 * We must search through the list of received datagrams matching them 716 * with outstanding requests using the xid, until ours is found. 717 */ 718 /* ARGSUSED */ 719 int 720 nfs_reply(myrep) 721 struct nfsreq *myrep; 722 { 723 struct nfsreq *rep; 724 struct nfsmount *nmp = myrep->r_nmp; 725 int32_t t1; 726 struct mbuf *mrep, *nam, *md; 727 u_int32_t rxid, *tl; 728 caddr_t dpos, cp2; 729 int error; 730 731 /* 732 * Loop around until we get our own reply 733 */ 734 for (;;) { 735 /* 736 * Lock against other receivers so that I don't get stuck in 737 * sbwait() after someone else has received my reply for me. 738 * Also necessary for connection based protocols to avoid 739 * race conditions during a reconnect. 740 */ 741 error = nfs_rcvlock(myrep); 742 if (error == EALREADY) 743 return (0); 744 if (error) 745 return (error); 746 /* 747 * Get the next Rpc reply off the socket 748 */ 749 nmp->nm_waiters++; 750 error = nfs_receive(myrep, &nam, &mrep); 751 nfs_rcvunlock(&nmp->nm_iflag); 752 if (error) { 753 754 if (nmp->nm_iflag & NFSMNT_DISMNT) { 755 /* 756 * Oops, we're going away now.. 757 */ 758 nmp->nm_waiters--; 759 wakeup (&nmp->nm_waiters); 760 return error; 761 } 762 nmp->nm_waiters--; 763 /* 764 * Ignore routing errors on connectionless protocols? ? 765 */ 766 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 767 nmp->nm_so->so_error = 0; 768 #ifdef DEBUG 769 printf("nfs_reply: ignoring error %d\n", error); 770 #endif 771 if (myrep->r_flags & R_GETONEREP) 772 return (0); 773 continue; 774 } 775 return (error); 776 } 777 nmp->nm_waiters--; 778 if (nam) 779 m_freem(nam); 780 781 /* 782 * Get the xid and check that it is an rpc reply 783 */ 784 md = mrep; 785 dpos = mtod(md, caddr_t); 786 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 787 rxid = *tl++; 788 if (*tl != rpc_reply) { 789 #ifndef NFS_V2_ONLY 790 if (nmp->nm_flag & NFSMNT_NQNFS) { 791 if (nqnfs_callback(nmp, mrep, md, dpos)) 792 nfsstats.rpcinvalid++; 793 } else 794 #endif 795 { 796 nfsstats.rpcinvalid++; 797 m_freem(mrep); 798 } 799 nfsmout: 800 if (myrep->r_flags & R_GETONEREP) 801 return (0); 802 continue; 803 } 804 805 /* 806 * Loop through the request list to match up the reply 807 * Iff no match, just drop the datagram 808 */ 809 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 810 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 811 /* Found it.. */ 812 rep->r_mrep = mrep; 813 rep->r_md = md; 814 rep->r_dpos = dpos; 815 if (nfsrtton) { 816 struct rttl *rt; 817 818 rt = &nfsrtt.rttl[nfsrtt.pos]; 819 rt->proc = rep->r_procnum; 820 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 821 rt->sent = nmp->nm_sent; 822 rt->cwnd = nmp->nm_cwnd; 823 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 824 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 825 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 826 rt->tstamp = time; 827 if (rep->r_flags & R_TIMING) 828 rt->rtt = rep->r_rtt; 829 else 830 rt->rtt = 1000000; 831 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 832 } 833 /* 834 * Update congestion window. 835 * Do the additive increase of 836 * one rpc/rtt. 837 */ 838 if (nmp->nm_cwnd <= nmp->nm_sent) { 839 nmp->nm_cwnd += 840 (NFS_CWNDSCALE * NFS_CWNDSCALE + 841 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 842 if (nmp->nm_cwnd > NFS_MAXCWND) 843 nmp->nm_cwnd = NFS_MAXCWND; 844 } 845 rep->r_flags &= ~R_SENT; 846 nmp->nm_sent -= NFS_CWNDSCALE; 847 /* 848 * Update rtt using a gain of 0.125 on the mean 849 * and a gain of 0.25 on the deviation. 850 */ 851 if (rep->r_flags & R_TIMING) { 852 /* 853 * Since the timer resolution of 854 * NFS_HZ is so course, it can often 855 * result in r_rtt == 0. Since 856 * r_rtt == N means that the actual 857 * rtt is between N+dt and N+2-dt ticks, 858 * add 1. 859 */ 860 t1 = rep->r_rtt + 1; 861 t1 -= (NFS_SRTT(rep) >> 3); 862 NFS_SRTT(rep) += t1; 863 if (t1 < 0) 864 t1 = -t1; 865 t1 -= (NFS_SDRTT(rep) >> 2); 866 NFS_SDRTT(rep) += t1; 867 } 868 nmp->nm_timeouts = 0; 869 break; 870 } 871 } 872 /* 873 * If not matched to a request, drop it. 874 * If it's mine, get out. 875 */ 876 if (rep == 0) { 877 nfsstats.rpcunexpected++; 878 m_freem(mrep); 879 } else if (rep == myrep) { 880 if (rep->r_mrep == NULL) 881 panic("nfsreply nil"); 882 return (0); 883 } 884 if (myrep->r_flags & R_GETONEREP) 885 return (0); 886 } 887 } 888 889 /* 890 * nfs_request - goes something like this 891 * - fill in request struct 892 * - links it into list 893 * - calls nfs_send() for first transmit 894 * - calls nfs_receive() to get reply 895 * - break down rpc header and return with nfs reply pointed to 896 * by mrep or error 897 * nb: always frees up mreq mbuf list 898 */ 899 int 900 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 901 struct vnode *vp; 902 struct mbuf *mrest; 903 int procnum; 904 struct proc *procp; 905 struct ucred *cred; 906 struct mbuf **mrp; 907 struct mbuf **mdp; 908 caddr_t *dposp; 909 { 910 struct mbuf *m, *mrep; 911 struct nfsreq *rep; 912 u_int32_t *tl; 913 int i; 914 struct nfsmount *nmp; 915 struct mbuf *md, *mheadend; 916 char nickv[RPCX_NICKVERF]; 917 time_t reqtime, waituntil; 918 caddr_t dpos, cp2; 919 int t1, s, error = 0, mrest_len, auth_len, auth_type; 920 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 921 int verf_len, verf_type; 922 u_int32_t xid; 923 char *auth_str, *verf_str; 924 NFSKERBKEY_T key; /* save session key */ 925 #ifndef NFS_V2_ONLY 926 int nqlflag, cachable; 927 u_quad_t frev; 928 struct nfsnode *np; 929 #endif 930 931 KASSERT(cred != NULL); 932 nmp = VFSTONFS(vp->v_mount); 933 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 934 rep->r_nmp = nmp; 935 rep->r_vp = vp; 936 rep->r_procp = procp; 937 rep->r_procnum = procnum; 938 i = 0; 939 m = mrest; 940 while (m) { 941 i += m->m_len; 942 m = m->m_next; 943 } 944 mrest_len = i; 945 946 /* 947 * Get the RPC header with authorization. 948 */ 949 kerbauth: 950 verf_str = auth_str = (char *)0; 951 if (nmp->nm_flag & NFSMNT_KERB) { 952 verf_str = nickv; 953 verf_len = sizeof (nickv); 954 auth_type = RPCAUTH_KERB4; 955 memset((caddr_t)key, 0, sizeof (key)); 956 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 957 &auth_len, verf_str, verf_len)) { 958 error = nfs_getauth(nmp, rep, cred, &auth_str, 959 &auth_len, verf_str, &verf_len, key); 960 if (error) { 961 free((caddr_t)rep, M_NFSREQ); 962 m_freem(mrest); 963 return (error); 964 } 965 } 966 } else { 967 auth_type = RPCAUTH_UNIX; 968 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 969 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 970 5 * NFSX_UNSIGNED; 971 } 972 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 973 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 974 if (auth_str) 975 free(auth_str, M_TEMP); 976 977 /* 978 * For stream protocols, insert a Sun RPC Record Mark. 979 */ 980 if (nmp->nm_sotype == SOCK_STREAM) { 981 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 982 *mtod(m, u_int32_t *) = htonl(0x80000000 | 983 (m->m_pkthdr.len - NFSX_UNSIGNED)); 984 } 985 rep->r_mreq = m; 986 rep->r_xid = xid; 987 tryagain: 988 if (nmp->nm_flag & NFSMNT_SOFT) 989 rep->r_retry = nmp->nm_retry; 990 else 991 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 992 rep->r_rtt = rep->r_rexmit = 0; 993 if (proct[procnum] > 0) 994 rep->r_flags = R_TIMING; 995 else 996 rep->r_flags = 0; 997 rep->r_mrep = NULL; 998 999 /* 1000 * Do the client side RPC. 1001 */ 1002 nfsstats.rpcrequests++; 1003 /* 1004 * Chain request into list of outstanding requests. Be sure 1005 * to put it LAST so timer finds oldest requests first. 1006 */ 1007 s = splsoftnet(); 1008 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1009 1010 /* Get send time for nqnfs */ 1011 reqtime = time.tv_sec; 1012 1013 /* 1014 * If backing off another request or avoiding congestion, don't 1015 * send this one now but let timer do it. If not timing a request, 1016 * do it now. 1017 */ 1018 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1019 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1020 nmp->nm_sent < nmp->nm_cwnd)) { 1021 splx(s); 1022 if (nmp->nm_soflags & PR_CONNREQUIRED) 1023 error = nfs_sndlock(&nmp->nm_iflag, rep); 1024 if (!error) { 1025 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1026 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 1027 if (nmp->nm_soflags & PR_CONNREQUIRED) 1028 nfs_sndunlock(&nmp->nm_iflag); 1029 } 1030 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1031 nmp->nm_sent += NFS_CWNDSCALE; 1032 rep->r_flags |= R_SENT; 1033 } 1034 } else { 1035 splx(s); 1036 rep->r_rtt = -1; 1037 } 1038 1039 /* 1040 * Wait for the reply from our send or the timer's. 1041 */ 1042 if (!error || error == EPIPE) 1043 error = nfs_reply(rep); 1044 1045 /* 1046 * RPC done, unlink the request. 1047 */ 1048 s = splsoftnet(); 1049 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1050 splx(s); 1051 1052 /* 1053 * Decrement the outstanding request count. 1054 */ 1055 if (rep->r_flags & R_SENT) { 1056 rep->r_flags &= ~R_SENT; /* paranoia */ 1057 nmp->nm_sent -= NFS_CWNDSCALE; 1058 } 1059 1060 /* 1061 * If there was a successful reply and a tprintf msg. 1062 * tprintf a response. 1063 */ 1064 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1065 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1066 "is alive again"); 1067 mrep = rep->r_mrep; 1068 md = rep->r_md; 1069 dpos = rep->r_dpos; 1070 if (error) { 1071 m_freem(rep->r_mreq); 1072 free((caddr_t)rep, M_NFSREQ); 1073 return (error); 1074 } 1075 1076 /* 1077 * break down the rpc header and check if ok 1078 */ 1079 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1080 if (*tl++ == rpc_msgdenied) { 1081 if (*tl == rpc_mismatch) 1082 error = EOPNOTSUPP; 1083 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1084 if (!failed_auth) { 1085 failed_auth++; 1086 mheadend->m_next = (struct mbuf *)0; 1087 m_freem(mrep); 1088 m_freem(rep->r_mreq); 1089 goto kerbauth; 1090 } else 1091 error = EAUTH; 1092 } else 1093 error = EACCES; 1094 m_freem(mrep); 1095 m_freem(rep->r_mreq); 1096 free((caddr_t)rep, M_NFSREQ); 1097 return (error); 1098 } 1099 1100 /* 1101 * Grab any Kerberos verifier, otherwise just throw it away. 1102 */ 1103 verf_type = fxdr_unsigned(int, *tl++); 1104 i = fxdr_unsigned(int32_t, *tl); 1105 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1106 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1107 if (error) 1108 goto nfsmout; 1109 } else if (i > 0) 1110 nfsm_adv(nfsm_rndup(i)); 1111 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1112 /* 0 == ok */ 1113 if (*tl == 0) { 1114 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1115 if (*tl != 0) { 1116 error = fxdr_unsigned(int, *tl); 1117 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1118 error == NFSERR_TRYLATER) { 1119 m_freem(mrep); 1120 error = 0; 1121 waituntil = time.tv_sec + trylater_delay; 1122 while (time.tv_sec < waituntil) 1123 (void) tsleep((caddr_t)&lbolt, 1124 PSOCK, "nqnfstry", 0); 1125 trylater_delay *= nfs_backoff[trylater_cnt]; 1126 if (trylater_cnt < 7) 1127 trylater_cnt++; 1128 goto tryagain; 1129 } 1130 1131 /* 1132 * If the File Handle was stale, invalidate the 1133 * lookup cache, just in case. 1134 */ 1135 if (error == ESTALE) 1136 cache_purge(vp); 1137 if (nmp->nm_flag & NFSMNT_NFSV3) { 1138 *mrp = mrep; 1139 *mdp = md; 1140 *dposp = dpos; 1141 error |= NFSERR_RETERR; 1142 } else 1143 m_freem(mrep); 1144 m_freem(rep->r_mreq); 1145 free((caddr_t)rep, M_NFSREQ); 1146 return (error); 1147 } 1148 1149 #ifndef NFS_V2_ONLY 1150 /* 1151 * For nqnfs, get any lease in reply 1152 */ 1153 if (nmp->nm_flag & NFSMNT_NQNFS) { 1154 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1155 if (*tl) { 1156 np = VTONFS(vp); 1157 nqlflag = fxdr_unsigned(int, *tl); 1158 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1159 cachable = fxdr_unsigned(int, *tl++); 1160 reqtime += fxdr_unsigned(int, *tl++); 1161 if (reqtime > time.tv_sec) { 1162 frev = fxdr_hyper(tl); 1163 nqnfs_clientlease(nmp, np, nqlflag, 1164 cachable, reqtime, frev); 1165 } 1166 } 1167 } 1168 #endif 1169 *mrp = mrep; 1170 *mdp = md; 1171 *dposp = dpos; 1172 m_freem(rep->r_mreq); 1173 FREE((caddr_t)rep, M_NFSREQ); 1174 return (0); 1175 } 1176 m_freem(mrep); 1177 error = EPROTONOSUPPORT; 1178 nfsmout: 1179 m_freem(rep->r_mreq); 1180 free((caddr_t)rep, M_NFSREQ); 1181 return (error); 1182 } 1183 #endif /* NFS */ 1184 1185 /* 1186 * Generate the rpc reply header 1187 * siz arg. is used to decide if adding a cluster is worthwhile 1188 */ 1189 int 1190 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1191 int siz; 1192 struct nfsrv_descript *nd; 1193 struct nfssvc_sock *slp; 1194 int err; 1195 int cache; 1196 u_quad_t *frev; 1197 struct mbuf **mrq; 1198 struct mbuf **mbp; 1199 caddr_t *bposp; 1200 { 1201 u_int32_t *tl; 1202 struct mbuf *mreq; 1203 caddr_t bpos; 1204 struct mbuf *mb, *mb2; 1205 1206 MGETHDR(mreq, M_WAIT, MT_DATA); 1207 mb = mreq; 1208 /* 1209 * If this is a big reply, use a cluster else 1210 * try and leave leading space for the lower level headers. 1211 */ 1212 siz += RPC_REPLYSIZ; 1213 if (siz >= max_datalen) { 1214 MCLGET(mreq, M_WAIT); 1215 } else 1216 mreq->m_data += max_hdr; 1217 tl = mtod(mreq, u_int32_t *); 1218 mreq->m_len = 6 * NFSX_UNSIGNED; 1219 bpos = ((caddr_t)tl) + mreq->m_len; 1220 *tl++ = txdr_unsigned(nd->nd_retxid); 1221 *tl++ = rpc_reply; 1222 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1223 *tl++ = rpc_msgdenied; 1224 if (err & NFSERR_AUTHERR) { 1225 *tl++ = rpc_autherr; 1226 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1227 mreq->m_len -= NFSX_UNSIGNED; 1228 bpos -= NFSX_UNSIGNED; 1229 } else { 1230 *tl++ = rpc_mismatch; 1231 *tl++ = txdr_unsigned(RPC_VER2); 1232 *tl = txdr_unsigned(RPC_VER2); 1233 } 1234 } else { 1235 *tl++ = rpc_msgaccepted; 1236 1237 /* 1238 * For Kerberos authentication, we must send the nickname 1239 * verifier back, otherwise just RPCAUTH_NULL. 1240 */ 1241 if (nd->nd_flag & ND_KERBFULL) { 1242 struct nfsuid *nuidp; 1243 struct timeval ktvin, ktvout; 1244 1245 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; 1246 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1247 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1248 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), 1249 &nuidp->nu_haddr, nd->nd_nam2))) 1250 break; 1251 } 1252 if (nuidp) { 1253 ktvin.tv_sec = 1254 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); 1255 ktvin.tv_usec = 1256 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1257 1258 /* 1259 * Encrypt the timestamp in ecb mode using the 1260 * session key. 1261 */ 1262 #ifdef NFSKERB 1263 XXX 1264 #endif 1265 1266 *tl++ = rpc_auth_kerb; 1267 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1268 *tl = ktvout.tv_sec; 1269 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1270 *tl++ = ktvout.tv_usec; 1271 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1272 } else { 1273 *tl++ = 0; 1274 *tl++ = 0; 1275 } 1276 } else { 1277 *tl++ = 0; 1278 *tl++ = 0; 1279 } 1280 switch (err) { 1281 case EPROGUNAVAIL: 1282 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1283 break; 1284 case EPROGMISMATCH: 1285 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1286 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1287 if (nd->nd_flag & ND_NQNFS) { 1288 *tl++ = txdr_unsigned(3); 1289 *tl = txdr_unsigned(3); 1290 } else { 1291 *tl++ = txdr_unsigned(2); 1292 *tl = txdr_unsigned(3); 1293 } 1294 break; 1295 case EPROCUNAVAIL: 1296 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1297 break; 1298 case EBADRPC: 1299 *tl = txdr_unsigned(RPC_GARBAGE); 1300 break; 1301 default: 1302 *tl = 0; 1303 if (err != NFSERR_RETVOID) { 1304 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1305 if (err) 1306 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1307 else 1308 *tl = 0; 1309 } 1310 break; 1311 }; 1312 } 1313 1314 /* 1315 * For nqnfs, piggyback lease as requested. 1316 */ 1317 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1318 if (nd->nd_flag & ND_LEASE) { 1319 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1320 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1321 *tl++ = txdr_unsigned(cache); 1322 *tl++ = txdr_unsigned(nd->nd_duration); 1323 txdr_hyper(*frev, tl); 1324 } else { 1325 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1326 *tl = 0; 1327 } 1328 } 1329 if (mrq != NULL) 1330 *mrq = mreq; 1331 *mbp = mb; 1332 *bposp = bpos; 1333 if (err != 0 && err != NFSERR_RETVOID) 1334 nfsstats.srvrpc_errs++; 1335 return (0); 1336 } 1337 1338 /* 1339 * Nfs timer routine 1340 * Scan the nfsreq list and retranmit any requests that have timed out 1341 * To avoid retransmission attempts on STREAM sockets (in the future) make 1342 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1343 */ 1344 void 1345 nfs_timer(arg) 1346 void *arg; /* never used */ 1347 { 1348 struct nfsreq *rep; 1349 struct mbuf *m; 1350 struct socket *so; 1351 struct nfsmount *nmp; 1352 int timeo; 1353 int s, error; 1354 #ifdef NFSSERVER 1355 struct nfssvc_sock *slp; 1356 static long lasttime = 0; 1357 u_quad_t cur_usec; 1358 #endif 1359 1360 s = splsoftnet(); 1361 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1362 nmp = rep->r_nmp; 1363 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1364 continue; 1365 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1366 rep->r_flags |= R_SOFTTERM; 1367 continue; 1368 } 1369 if (rep->r_rtt >= 0) { 1370 rep->r_rtt++; 1371 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1372 timeo = nmp->nm_timeo; 1373 else 1374 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1375 if (nmp->nm_timeouts > 0) 1376 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1377 if (rep->r_rtt <= timeo) 1378 continue; 1379 if (nmp->nm_timeouts < 8) 1380 nmp->nm_timeouts++; 1381 } 1382 /* 1383 * Check for server not responding 1384 */ 1385 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1386 rep->r_rexmit > nmp->nm_deadthresh) { 1387 nfs_msg(rep->r_procp, 1388 nmp->nm_mountp->mnt_stat.f_mntfromname, 1389 "not responding"); 1390 rep->r_flags |= R_TPRINTFMSG; 1391 } 1392 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1393 nfsstats.rpctimeouts++; 1394 rep->r_flags |= R_SOFTTERM; 1395 continue; 1396 } 1397 if (nmp->nm_sotype != SOCK_DGRAM) { 1398 if (++rep->r_rexmit > NFS_MAXREXMIT) 1399 rep->r_rexmit = NFS_MAXREXMIT; 1400 continue; 1401 } 1402 if ((so = nmp->nm_so) == NULL) 1403 continue; 1404 1405 /* 1406 * If there is enough space and the window allows.. 1407 * Resend it 1408 * Set r_rtt to -1 in case we fail to send it now. 1409 */ 1410 rep->r_rtt = -1; 1411 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1412 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1413 (rep->r_flags & R_SENT) || 1414 nmp->nm_sent < nmp->nm_cwnd) && 1415 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1416 if (so->so_state & SS_ISCONNECTED) 1417 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1418 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1419 else 1420 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1421 nmp->nm_nam, (struct mbuf *)0, (struct proc *)0); 1422 if (error) { 1423 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1424 #ifdef DEBUG 1425 printf("nfs_timer: ignoring error %d\n", 1426 error); 1427 #endif 1428 so->so_error = 0; 1429 } 1430 } else { 1431 /* 1432 * Iff first send, start timing 1433 * else turn timing off, backoff timer 1434 * and divide congestion window by 2. 1435 */ 1436 if (rep->r_flags & R_SENT) { 1437 rep->r_flags &= ~R_TIMING; 1438 if (++rep->r_rexmit > NFS_MAXREXMIT) 1439 rep->r_rexmit = NFS_MAXREXMIT; 1440 nmp->nm_cwnd >>= 1; 1441 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1442 nmp->nm_cwnd = NFS_CWNDSCALE; 1443 nfsstats.rpcretries++; 1444 } else { 1445 rep->r_flags |= R_SENT; 1446 nmp->nm_sent += NFS_CWNDSCALE; 1447 } 1448 rep->r_rtt = 0; 1449 } 1450 } 1451 } 1452 1453 #ifdef NFSSERVER 1454 /* 1455 * Call the nqnfs server timer once a second to handle leases. 1456 */ 1457 if (lasttime != time.tv_sec) { 1458 lasttime = time.tv_sec; 1459 nqnfs_serverd(); 1460 } 1461 1462 /* 1463 * Scan the write gathering queues for writes that need to be 1464 * completed now. 1465 */ 1466 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1467 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1468 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec) 1469 nfsrv_wakenfsd(slp); 1470 } 1471 #endif /* NFSSERVER */ 1472 splx(s); 1473 callout_reset(&nfs_timer_ch, nfs_ticks, nfs_timer, NULL); 1474 } 1475 1476 /*ARGSUSED*/ 1477 void 1478 nfs_exit(p, v) 1479 struct proc *p; 1480 void *v; 1481 { 1482 struct nfsreq *rp; 1483 int s = splsoftnet(); 1484 1485 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 1486 if (rp->r_procp == p) 1487 TAILQ_REMOVE(&nfs_reqq, rp, r_chain); 1488 } 1489 splx(s); 1490 } 1491 1492 /* 1493 * Test for a termination condition pending on the process. 1494 * This is used for NFSMNT_INT mounts. 1495 */ 1496 int 1497 nfs_sigintr(nmp, rep, p) 1498 struct nfsmount *nmp; 1499 struct nfsreq *rep; 1500 struct proc *p; 1501 { 1502 sigset_t ss; 1503 1504 if (rep && (rep->r_flags & R_SOFTTERM)) 1505 return (EINTR); 1506 if (!(nmp->nm_flag & NFSMNT_INT)) 1507 return (0); 1508 if (p) { 1509 sigpending1(p, &ss); 1510 #if 0 1511 sigminusset(&p->p_sigctx.ps_sigignore, &ss); 1512 #endif 1513 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1514 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1515 sigismember(&ss, SIGQUIT)) 1516 return (EINTR); 1517 } 1518 return (0); 1519 } 1520 1521 /* 1522 * Lock a socket against others. 1523 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1524 * and also to avoid race conditions between the processes with nfs requests 1525 * in progress when a reconnect is necessary. 1526 */ 1527 int 1528 nfs_sndlock(flagp, rep) 1529 int *flagp; 1530 struct nfsreq *rep; 1531 { 1532 struct proc *p; 1533 int slpflag = 0, slptimeo = 0; 1534 1535 if (rep) { 1536 p = rep->r_procp; 1537 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1538 slpflag = PCATCH; 1539 } else 1540 p = (struct proc *)0; 1541 while (*flagp & NFSMNT_SNDLOCK) { 1542 if (nfs_sigintr(rep->r_nmp, rep, p)) 1543 return (EINTR); 1544 *flagp |= NFSMNT_WANTSND; 1545 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1546 slptimeo); 1547 if (slpflag == PCATCH) { 1548 slpflag = 0; 1549 slptimeo = 2 * hz; 1550 } 1551 } 1552 *flagp |= NFSMNT_SNDLOCK; 1553 return (0); 1554 } 1555 1556 /* 1557 * Unlock the stream socket for others. 1558 */ 1559 void 1560 nfs_sndunlock(flagp) 1561 int *flagp; 1562 { 1563 1564 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1565 panic("nfs sndunlock"); 1566 *flagp &= ~NFSMNT_SNDLOCK; 1567 if (*flagp & NFSMNT_WANTSND) { 1568 *flagp &= ~NFSMNT_WANTSND; 1569 wakeup((caddr_t)flagp); 1570 } 1571 } 1572 1573 int 1574 nfs_rcvlock(rep) 1575 struct nfsreq *rep; 1576 { 1577 struct nfsmount *nmp = rep->r_nmp; 1578 int *flagp = &nmp->nm_iflag; 1579 int slpflag, slptimeo = 0; 1580 1581 if (*flagp & NFSMNT_DISMNT) 1582 return EIO; 1583 1584 if (*flagp & NFSMNT_INT) 1585 slpflag = PCATCH; 1586 else 1587 slpflag = 0; 1588 while (*flagp & NFSMNT_RCVLOCK) { 1589 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1590 return (EINTR); 1591 *flagp |= NFSMNT_WANTRCV; 1592 nmp->nm_waiters++; 1593 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1594 slptimeo); 1595 nmp->nm_waiters--; 1596 if (*flagp & NFSMNT_DISMNT) { 1597 wakeup(&nmp->nm_waiters); 1598 return EIO; 1599 } 1600 /* If our reply was received while we were sleeping, 1601 * then just return without taking the lock to avoid a 1602 * situation where a single iod could 'capture' the 1603 * receive lock. 1604 */ 1605 if (rep->r_mrep != NULL) 1606 return (EALREADY); 1607 if (slpflag == PCATCH) { 1608 slpflag = 0; 1609 slptimeo = 2 * hz; 1610 } 1611 } 1612 *flagp |= NFSMNT_RCVLOCK; 1613 return (0); 1614 } 1615 1616 /* 1617 * Unlock the stream socket for others. 1618 */ 1619 void 1620 nfs_rcvunlock(flagp) 1621 int *flagp; 1622 { 1623 1624 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1625 panic("nfs rcvunlock"); 1626 *flagp &= ~NFSMNT_RCVLOCK; 1627 if (*flagp & NFSMNT_WANTRCV) { 1628 *flagp &= ~NFSMNT_WANTRCV; 1629 wakeup((caddr_t)flagp); 1630 } 1631 } 1632 1633 /* 1634 * Parse an RPC request 1635 * - verify it 1636 * - fill in the cred struct. 1637 */ 1638 int 1639 nfs_getreq(nd, nfsd, has_header) 1640 struct nfsrv_descript *nd; 1641 struct nfsd *nfsd; 1642 int has_header; 1643 { 1644 int len, i; 1645 u_int32_t *tl; 1646 int32_t t1; 1647 struct uio uio; 1648 struct iovec iov; 1649 caddr_t dpos, cp2, cp; 1650 u_int32_t nfsvers, auth_type; 1651 uid_t nickuid; 1652 int error = 0, nqnfs = 0, ticklen; 1653 struct mbuf *mrep, *md; 1654 struct nfsuid *nuidp; 1655 struct timeval tvin, tvout; 1656 1657 mrep = nd->nd_mrep; 1658 md = nd->nd_md; 1659 dpos = nd->nd_dpos; 1660 if (has_header) { 1661 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1662 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1663 if (*tl++ != rpc_call) { 1664 m_freem(mrep); 1665 return (EBADRPC); 1666 } 1667 } else 1668 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1669 nd->nd_repstat = 0; 1670 nd->nd_flag = 0; 1671 if (*tl++ != rpc_vers) { 1672 nd->nd_repstat = ERPCMISMATCH; 1673 nd->nd_procnum = NFSPROC_NOOP; 1674 return (0); 1675 } 1676 if (*tl != nfs_prog) { 1677 if (*tl == nqnfs_prog) 1678 nqnfs++; 1679 else { 1680 nd->nd_repstat = EPROGUNAVAIL; 1681 nd->nd_procnum = NFSPROC_NOOP; 1682 return (0); 1683 } 1684 } 1685 tl++; 1686 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1687 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1688 (nfsvers != NQNFS_VER3 && nqnfs)) { 1689 nd->nd_repstat = EPROGMISMATCH; 1690 nd->nd_procnum = NFSPROC_NOOP; 1691 return (0); 1692 } 1693 if (nqnfs) 1694 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1695 else if (nfsvers == NFS_VER3) 1696 nd->nd_flag = ND_NFSV3; 1697 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1698 if (nd->nd_procnum == NFSPROC_NULL) 1699 return (0); 1700 if (nd->nd_procnum >= NFS_NPROCS || 1701 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1702 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1703 nd->nd_repstat = EPROCUNAVAIL; 1704 nd->nd_procnum = NFSPROC_NOOP; 1705 return (0); 1706 } 1707 if ((nd->nd_flag & ND_NFSV3) == 0) 1708 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1709 auth_type = *tl++; 1710 len = fxdr_unsigned(int, *tl++); 1711 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1712 m_freem(mrep); 1713 return (EBADRPC); 1714 } 1715 1716 nd->nd_flag &= ~ND_KERBAUTH; 1717 /* 1718 * Handle auth_unix or auth_kerb. 1719 */ 1720 if (auth_type == rpc_auth_unix) { 1721 len = fxdr_unsigned(int, *++tl); 1722 if (len < 0 || len > NFS_MAXNAMLEN) { 1723 m_freem(mrep); 1724 return (EBADRPC); 1725 } 1726 nfsm_adv(nfsm_rndup(len)); 1727 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1728 memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred)); 1729 nd->nd_cr.cr_ref = 1; 1730 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1731 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1732 len = fxdr_unsigned(int, *tl); 1733 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1734 m_freem(mrep); 1735 return (EBADRPC); 1736 } 1737 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1738 for (i = 0; i < len; i++) 1739 if (i < NGROUPS) 1740 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1741 else 1742 tl++; 1743 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1744 if (nd->nd_cr.cr_ngroups > 1) 1745 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1746 len = fxdr_unsigned(int, *++tl); 1747 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1748 m_freem(mrep); 1749 return (EBADRPC); 1750 } 1751 if (len > 0) 1752 nfsm_adv(nfsm_rndup(len)); 1753 } else if (auth_type == rpc_auth_kerb) { 1754 switch (fxdr_unsigned(int, *tl++)) { 1755 case RPCAKN_FULLNAME: 1756 ticklen = fxdr_unsigned(int, *tl); 1757 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1758 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1759 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1760 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1761 m_freem(mrep); 1762 return (EBADRPC); 1763 } 1764 uio.uio_offset = 0; 1765 uio.uio_iov = &iov; 1766 uio.uio_iovcnt = 1; 1767 uio.uio_segflg = UIO_SYSSPACE; 1768 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1769 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1770 nfsm_mtouio(&uio, uio.uio_resid); 1771 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1772 if (*tl++ != rpc_auth_kerb || 1773 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1774 printf("Bad kerb verifier\n"); 1775 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1776 nd->nd_procnum = NFSPROC_NOOP; 1777 return (0); 1778 } 1779 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1780 tl = (u_int32_t *)cp; 1781 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1782 printf("Not fullname kerb verifier\n"); 1783 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1784 nd->nd_procnum = NFSPROC_NOOP; 1785 return (0); 1786 } 1787 cp += NFSX_UNSIGNED; 1788 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 1789 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1790 nd->nd_flag |= ND_KERBFULL; 1791 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1792 break; 1793 case RPCAKN_NICKNAME: 1794 if (len != 2 * NFSX_UNSIGNED) { 1795 printf("Kerb nickname short\n"); 1796 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1797 nd->nd_procnum = NFSPROC_NOOP; 1798 return (0); 1799 } 1800 nickuid = fxdr_unsigned(uid_t, *tl); 1801 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1802 if (*tl++ != rpc_auth_kerb || 1803 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1804 printf("Kerb nick verifier bad\n"); 1805 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1806 nd->nd_procnum = NFSPROC_NOOP; 1807 return (0); 1808 } 1809 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1810 tvin.tv_sec = *tl++; 1811 tvin.tv_usec = *tl; 1812 1813 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; 1814 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1815 if (nuidp->nu_cr.cr_uid == nickuid && 1816 (!nd->nd_nam2 || 1817 netaddr_match(NU_NETFAM(nuidp), 1818 &nuidp->nu_haddr, nd->nd_nam2))) 1819 break; 1820 } 1821 if (!nuidp) { 1822 nd->nd_repstat = 1823 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1824 nd->nd_procnum = NFSPROC_NOOP; 1825 return (0); 1826 } 1827 1828 /* 1829 * Now, decrypt the timestamp using the session key 1830 * and validate it. 1831 */ 1832 #ifdef NFSKERB 1833 XXX 1834 #endif 1835 1836 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1837 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1838 if (nuidp->nu_expire < time.tv_sec || 1839 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1840 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1841 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1842 nuidp->nu_expire = 0; 1843 nd->nd_repstat = 1844 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1845 nd->nd_procnum = NFSPROC_NOOP; 1846 return (0); 1847 } 1848 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1849 nd->nd_flag |= ND_KERBNICK; 1850 }; 1851 } else { 1852 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1853 nd->nd_procnum = NFSPROC_NOOP; 1854 return (0); 1855 } 1856 1857 /* 1858 * For nqnfs, get piggybacked lease request. 1859 */ 1860 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1861 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1862 nd->nd_flag |= fxdr_unsigned(int, *tl); 1863 if (nd->nd_flag & ND_LEASE) { 1864 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1865 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 1866 } else 1867 nd->nd_duration = NQ_MINLEASE; 1868 } else 1869 nd->nd_duration = NQ_MINLEASE; 1870 nd->nd_md = md; 1871 nd->nd_dpos = dpos; 1872 return (0); 1873 nfsmout: 1874 return (error); 1875 } 1876 1877 int 1878 nfs_msg(p, server, msg) 1879 struct proc *p; 1880 char *server, *msg; 1881 { 1882 tpr_t tpr; 1883 1884 if (p) 1885 tpr = tprintf_open(p); 1886 else 1887 tpr = NULL; 1888 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1889 tprintf_close(tpr); 1890 return (0); 1891 } 1892 1893 #ifdef NFSSERVER 1894 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 1895 struct nfssvc_sock *, struct proc *, 1896 struct mbuf **)) = { 1897 nfsrv_null, 1898 nfsrv_getattr, 1899 nfsrv_setattr, 1900 nfsrv_lookup, 1901 nfsrv3_access, 1902 nfsrv_readlink, 1903 nfsrv_read, 1904 nfsrv_write, 1905 nfsrv_create, 1906 nfsrv_mkdir, 1907 nfsrv_symlink, 1908 nfsrv_mknod, 1909 nfsrv_remove, 1910 nfsrv_rmdir, 1911 nfsrv_rename, 1912 nfsrv_link, 1913 nfsrv_readdir, 1914 nfsrv_readdirplus, 1915 nfsrv_statfs, 1916 nfsrv_fsinfo, 1917 nfsrv_pathconf, 1918 nfsrv_commit, 1919 nqnfsrv_getlease, 1920 nqnfsrv_vacated, 1921 nfsrv_noop, 1922 nfsrv_noop 1923 }; 1924 1925 /* 1926 * Socket upcall routine for the nfsd sockets. 1927 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1928 * Essentially do as much as possible non-blocking, else punt and it will 1929 * be called with M_WAIT from an nfsd. 1930 */ 1931 void 1932 nfsrv_rcv(so, arg, waitflag) 1933 struct socket *so; 1934 caddr_t arg; 1935 int waitflag; 1936 { 1937 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1938 struct mbuf *m; 1939 struct mbuf *mp, *nam; 1940 struct uio auio; 1941 int flags, error; 1942 1943 if ((slp->ns_flag & SLP_VALID) == 0) 1944 return; 1945 #ifdef notdef 1946 /* 1947 * Define this to test for nfsds handling this under heavy load. 1948 */ 1949 if (waitflag == M_DONTWAIT) { 1950 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1951 } 1952 #endif 1953 auio.uio_procp = NULL; 1954 if (so->so_type == SOCK_STREAM) { 1955 /* 1956 * If there are already records on the queue, defer soreceive() 1957 * to an nfsd so that there is feedback to the TCP layer that 1958 * the nfs servers are heavily loaded. 1959 */ 1960 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1961 slp->ns_flag |= SLP_NEEDQ; 1962 goto dorecs; 1963 } 1964 1965 /* 1966 * Do soreceive(). 1967 */ 1968 auio.uio_resid = 1000000000; 1969 flags = MSG_DONTWAIT; 1970 error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1971 if (error || mp == (struct mbuf *)0) { 1972 if (error == EWOULDBLOCK) 1973 slp->ns_flag |= SLP_NEEDQ; 1974 else 1975 slp->ns_flag |= SLP_DISCONN; 1976 goto dorecs; 1977 } 1978 m = mp; 1979 if (slp->ns_rawend) { 1980 slp->ns_rawend->m_next = m; 1981 slp->ns_cc += 1000000000 - auio.uio_resid; 1982 } else { 1983 slp->ns_raw = m; 1984 slp->ns_cc = 1000000000 - auio.uio_resid; 1985 } 1986 while (m->m_next) 1987 m = m->m_next; 1988 slp->ns_rawend = m; 1989 1990 /* 1991 * Now try and parse record(s) out of the raw stream data. 1992 */ 1993 error = nfsrv_getstream(slp, waitflag); 1994 if (error) { 1995 if (error == EPERM) 1996 slp->ns_flag |= SLP_DISCONN; 1997 else 1998 slp->ns_flag |= SLP_NEEDQ; 1999 } 2000 } else { 2001 do { 2002 auio.uio_resid = 1000000000; 2003 flags = MSG_DONTWAIT; 2004 error = (*so->so_receive)(so, &nam, &auio, &mp, 2005 (struct mbuf **)0, &flags); 2006 if (mp) { 2007 if (nam) { 2008 m = nam; 2009 m->m_next = mp; 2010 } else 2011 m = mp; 2012 if (slp->ns_recend) 2013 slp->ns_recend->m_nextpkt = m; 2014 else 2015 slp->ns_rec = m; 2016 slp->ns_recend = m; 2017 m->m_nextpkt = (struct mbuf *)0; 2018 } 2019 if (error) { 2020 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2021 && error != EWOULDBLOCK) { 2022 slp->ns_flag |= SLP_DISCONN; 2023 goto dorecs; 2024 } 2025 } 2026 } while (mp); 2027 } 2028 2029 /* 2030 * Now try and process the request records, non-blocking. 2031 */ 2032 dorecs: 2033 if (waitflag == M_DONTWAIT && 2034 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 2035 nfsrv_wakenfsd(slp); 2036 } 2037 2038 /* 2039 * Try and extract an RPC request from the mbuf data list received on a 2040 * stream socket. The "waitflag" argument indicates whether or not it 2041 * can sleep. 2042 */ 2043 int 2044 nfsrv_getstream(slp, waitflag) 2045 struct nfssvc_sock *slp; 2046 int waitflag; 2047 { 2048 struct mbuf *m, **mpp; 2049 char *cp1, *cp2; 2050 int len; 2051 struct mbuf *om, *m2, *recm = NULL; 2052 u_int32_t recmark; 2053 2054 if (slp->ns_flag & SLP_GETSTREAM) 2055 panic("nfs getstream"); 2056 slp->ns_flag |= SLP_GETSTREAM; 2057 for (;;) { 2058 if (slp->ns_reclen == 0) { 2059 if (slp->ns_cc < NFSX_UNSIGNED) { 2060 slp->ns_flag &= ~SLP_GETSTREAM; 2061 return (0); 2062 } 2063 m = slp->ns_raw; 2064 if (m->m_len >= NFSX_UNSIGNED) { 2065 memcpy((caddr_t)&recmark, mtod(m, caddr_t), NFSX_UNSIGNED); 2066 m->m_data += NFSX_UNSIGNED; 2067 m->m_len -= NFSX_UNSIGNED; 2068 } else { 2069 cp1 = (caddr_t)&recmark; 2070 cp2 = mtod(m, caddr_t); 2071 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 2072 while (m->m_len == 0) { 2073 m = m->m_next; 2074 cp2 = mtod(m, caddr_t); 2075 } 2076 *cp1++ = *cp2++; 2077 m->m_data++; 2078 m->m_len--; 2079 } 2080 } 2081 slp->ns_cc -= NFSX_UNSIGNED; 2082 recmark = ntohl(recmark); 2083 slp->ns_reclen = recmark & ~0x80000000; 2084 if (recmark & 0x80000000) 2085 slp->ns_flag |= SLP_LASTFRAG; 2086 else 2087 slp->ns_flag &= ~SLP_LASTFRAG; 2088 if (slp->ns_reclen > NFS_MAXPACKET) { 2089 slp->ns_flag &= ~SLP_GETSTREAM; 2090 return (EPERM); 2091 } 2092 } 2093 2094 /* 2095 * Now get the record part. 2096 */ 2097 if (slp->ns_cc == slp->ns_reclen) { 2098 recm = slp->ns_raw; 2099 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2100 slp->ns_cc = slp->ns_reclen = 0; 2101 } else if (slp->ns_cc > slp->ns_reclen) { 2102 len = 0; 2103 m = slp->ns_raw; 2104 om = (struct mbuf *)0; 2105 while (len < slp->ns_reclen) { 2106 if ((len + m->m_len) > slp->ns_reclen) { 2107 size_t left = slp->ns_reclen - len; 2108 2109 MGETHDR(m2, waitflag, m->m_type); 2110 if (m2 == NULL) { 2111 slp->ns_flag &= ~SLP_GETSTREAM; 2112 return (EWOULDBLOCK); 2113 } 2114 if (left > MHLEN) { 2115 MCLGET(m2, waitflag); 2116 if (!(m2->m_flags & M_EXT)) { 2117 m_freem(m2); 2118 slp->ns_flag &= ~SLP_GETSTREAM; 2119 return (EWOULDBLOCK); 2120 } 2121 } 2122 memcpy(mtod(m2, caddr_t), mtod(m, caddr_t), 2123 left); 2124 m2->m_len = left; 2125 m->m_data += left; 2126 m->m_len -= left; 2127 if (om) { 2128 om->m_next = m2; 2129 recm = slp->ns_raw; 2130 } else 2131 recm = m2; 2132 len = slp->ns_reclen; 2133 } else if ((len + m->m_len) == slp->ns_reclen) { 2134 om = m; 2135 len += m->m_len; 2136 m = m->m_next; 2137 recm = slp->ns_raw; 2138 om->m_next = (struct mbuf *)0; 2139 } else { 2140 om = m; 2141 len += m->m_len; 2142 m = m->m_next; 2143 } 2144 } 2145 slp->ns_raw = m; 2146 slp->ns_cc -= len; 2147 slp->ns_reclen = 0; 2148 } else { 2149 slp->ns_flag &= ~SLP_GETSTREAM; 2150 return (0); 2151 } 2152 2153 /* 2154 * Accumulate the fragments into a record. 2155 */ 2156 mpp = &slp->ns_frag; 2157 while (*mpp) 2158 mpp = &((*mpp)->m_next); 2159 *mpp = recm; 2160 if (slp->ns_flag & SLP_LASTFRAG) { 2161 if (slp->ns_recend) 2162 slp->ns_recend->m_nextpkt = slp->ns_frag; 2163 else 2164 slp->ns_rec = slp->ns_frag; 2165 slp->ns_recend = slp->ns_frag; 2166 slp->ns_frag = (struct mbuf *)0; 2167 } 2168 } 2169 } 2170 2171 /* 2172 * Parse an RPC header. 2173 */ 2174 int 2175 nfsrv_dorec(slp, nfsd, ndp) 2176 struct nfssvc_sock *slp; 2177 struct nfsd *nfsd; 2178 struct nfsrv_descript **ndp; 2179 { 2180 struct mbuf *m, *nam; 2181 struct nfsrv_descript *nd; 2182 int error; 2183 2184 *ndp = NULL; 2185 if ((slp->ns_flag & SLP_VALID) == 0 || 2186 (m = slp->ns_rec) == (struct mbuf *)0) 2187 return (ENOBUFS); 2188 slp->ns_rec = m->m_nextpkt; 2189 if (slp->ns_rec) 2190 m->m_nextpkt = (struct mbuf *)0; 2191 else 2192 slp->ns_recend = (struct mbuf *)0; 2193 if (m->m_type == MT_SONAME) { 2194 nam = m; 2195 m = m->m_next; 2196 nam->m_next = NULL; 2197 } else 2198 nam = NULL; 2199 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), 2200 M_NFSRVDESC, M_WAITOK); 2201 nd->nd_md = nd->nd_mrep = m; 2202 nd->nd_nam2 = nam; 2203 nd->nd_dpos = mtod(m, caddr_t); 2204 error = nfs_getreq(nd, nfsd, TRUE); 2205 if (error) { 2206 m_freem(nam); 2207 free((caddr_t)nd, M_NFSRVDESC); 2208 return (error); 2209 } 2210 *ndp = nd; 2211 nfsd->nfsd_nd = nd; 2212 return (0); 2213 } 2214 2215 2216 /* 2217 * Search for a sleeping nfsd and wake it up. 2218 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2219 * running nfsds will go look for the work in the nfssvc_sock list. 2220 */ 2221 void 2222 nfsrv_wakenfsd(slp) 2223 struct nfssvc_sock *slp; 2224 { 2225 struct nfsd *nd; 2226 2227 if ((slp->ns_flag & SLP_VALID) == 0) 2228 return; 2229 TAILQ_FOREACH(nd, &nfsd_head, nfsd_chain) { 2230 if (nd->nfsd_flag & NFSD_WAITING) { 2231 nd->nfsd_flag &= ~NFSD_WAITING; 2232 if (nd->nfsd_slp) 2233 panic("nfsd wakeup"); 2234 slp->ns_sref++; 2235 nd->nfsd_slp = slp; 2236 wakeup((caddr_t)nd); 2237 return; 2238 } 2239 } 2240 slp->ns_flag |= SLP_DOREC; 2241 nfsd_head_flag |= NFSD_CHECKSLP; 2242 } 2243 #endif /* NFSSERVER */ 2244