nfs_socket.c revision 1.92.2.11
1 /* $NetBSD: nfs_socket.c,v 1.92.2.11 2005/11/10 14:11:55 skrll 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. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 35 */ 36 37 /* 38 * Socket operations for use by nfs 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: nfs_socket.c,v 1.92.2.11 2005/11/10 14:11:55 skrll Exp $"); 43 44 #include "fs_nfs.h" 45 #include "opt_nfs.h" 46 #include "opt_nfsserver.h" 47 #include "opt_mbuftrace.h" 48 #include "opt_inet.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/callout.h> 53 #include <sys/proc.h> 54 #include <sys/mount.h> 55 #include <sys/kernel.h> 56 #include <sys/mbuf.h> 57 #include <sys/vnode.h> 58 #include <sys/domain.h> 59 #include <sys/protosw.h> 60 #include <sys/socket.h> 61 #include <sys/socketvar.h> 62 #include <sys/syslog.h> 63 #include <sys/tprintf.h> 64 #include <sys/namei.h> 65 #include <sys/signal.h> 66 #include <sys/signalvar.h> 67 68 #include <netinet/in.h> 69 #include <netinet/tcp.h> 70 71 #include <nfs/rpcv2.h> 72 #include <nfs/nfsproto.h> 73 #include <nfs/nfs.h> 74 #include <nfs/xdr_subs.h> 75 #include <nfs/nfsm_subs.h> 76 #include <nfs/nfsmount.h> 77 #include <nfs/nfsnode.h> 78 #include <nfs/nfsrtt.h> 79 #include <nfs/nqnfs.h> 80 #include <nfs/nfs_var.h> 81 82 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header"); 83 #ifdef MBUFTRACE 84 struct mowner nfs_mowner = { "nfs" }; 85 #endif 86 87 /* 88 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 89 * Use the mean and mean deviation of rtt for the appropriate type of rpc 90 * for the frequent rpcs and a default for the others. 91 * The justification for doing "other" this way is that these rpcs 92 * happen so infrequently that timer est. would probably be stale. 93 * Also, since many of these rpcs are 94 * non-idempotent, a conservative timeout is desired. 95 * getattr, lookup - A+2D 96 * read, write - A+4D 97 * other - nm_timeo 98 */ 99 #define NFS_RTO(n, t) \ 100 ((t) == 0 ? (n)->nm_timeo : \ 101 ((t) < 3 ? \ 102 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 103 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 104 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 105 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 106 /* 107 * External data, mostly RPC constants in XDR form 108 */ 109 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 110 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, 111 rpc_auth_kerb; 112 extern u_int32_t nfs_prog, nqnfs_prog; 113 extern time_t nqnfsstarttime; 114 extern const int nfsv3_procid[NFS_NPROCS]; 115 extern int nfs_ticks; 116 117 /* 118 * Defines which timer to use for the procnum. 119 * 0 - default 120 * 1 - getattr 121 * 2 - lookup 122 * 3 - read 123 * 4 - write 124 */ 125 static const int proct[NFS_NPROCS] = { 126 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0, 127 0, 0, 0, 128 }; 129 130 /* 131 * There is a congestion window for outstanding rpcs maintained per mount 132 * point. The cwnd size is adjusted in roughly the way that: 133 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 134 * SIGCOMM '88". ACM, August 1988. 135 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 136 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 137 * of rpcs is in progress. 138 * (The sent count and cwnd are scaled for integer arith.) 139 * Variants of "slow start" were tried and were found to be too much of a 140 * performance hit (ave. rtt 3 times larger), 141 * I suspect due to the large rtt that nfs rpcs have. 142 */ 143 #define NFS_CWNDSCALE 256 144 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 145 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 146 int nfsrtton = 0; 147 struct nfsrtt nfsrtt; 148 struct nfsreqhead nfs_reqq; 149 150 struct callout nfs_timer_ch = CALLOUT_INITIALIZER_SETFUNC(nfs_timer, NULL); 151 152 /* 153 * Initialize sockets and congestion for a new NFS connection. 154 * We do not free the sockaddr if error. 155 */ 156 int 157 nfs_connect(nmp, rep, l) 158 struct nfsmount *nmp; 159 struct nfsreq *rep; 160 struct lwp *l; 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, 174 nmp->nm_sotype, nmp->nm_soproto, l); 175 if (error) 176 goto bad; 177 so = nmp->nm_so; 178 #ifdef MBUFTRACE 179 so->so_mowner = &nfs_mowner; 180 so->so_rcv.sb_mowner = &nfs_mowner; 181 so->so_snd.sb_mowner = &nfs_mowner; 182 #endif 183 nmp->nm_soflags = so->so_proto->pr_flags; 184 185 /* 186 * Some servers require that the client port be a reserved port number. 187 */ 188 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 189 m = m_get(M_WAIT, MT_SOOPTS); 190 MCLAIM(m, so->so_mowner); 191 *mtod(m, int32_t *) = IP_PORTRANGE_LOW; 192 m->m_len = sizeof(int32_t); 193 if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m))) 194 goto bad; 195 m = m_get(M_WAIT, MT_SONAME); 196 MCLAIM(m, so->so_mowner); 197 sin = mtod(m, struct sockaddr_in *); 198 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 199 sin->sin_family = AF_INET; 200 sin->sin_addr.s_addr = INADDR_ANY; 201 sin->sin_port = 0; 202 error = sobind(so, m, &lwp0); 203 m_freem(m); 204 if (error) 205 goto bad; 206 } 207 #ifdef INET6 208 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) { 209 m = m_get(M_WAIT, MT_SOOPTS); 210 MCLAIM(m, so->so_mowner); 211 *mtod(m, int32_t *) = IPV6_PORTRANGE_LOW; 212 m->m_len = sizeof(int32_t); 213 if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m))) 214 goto bad; 215 m = m_get(M_WAIT, MT_SONAME); 216 MCLAIM(m, so->so_mowner); 217 sin6 = mtod(m, struct sockaddr_in6 *); 218 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6); 219 sin6->sin6_family = AF_INET6; 220 sin6->sin6_addr = in6addr_any; 221 sin6->sin6_port = 0; 222 error = sobind(so, m, &lwp0); 223 m_freem(m); 224 if (error) 225 goto bad; 226 } 227 #endif 228 229 /* 230 * Protocols that do not require connections may be optionally left 231 * unconnected for servers that reply from a port other than NFS_PORT. 232 */ 233 if (nmp->nm_flag & NFSMNT_NOCONN) { 234 if (nmp->nm_soflags & PR_CONNREQUIRED) { 235 error = ENOTCONN; 236 goto bad; 237 } 238 } else { 239 error = soconnect(so, nmp->nm_nam, l); 240 if (error) 241 goto bad; 242 243 /* 244 * Wait for the connection to complete. Cribbed from the 245 * connect system call but with the wait timing out so 246 * that interruptible mounts don't hang here for a long time. 247 */ 248 s = splsoftnet(); 249 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 250 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 251 "nfscn1", 2 * hz); 252 if ((so->so_state & SS_ISCONNECTING) && 253 so->so_error == 0 && rep && 254 (error = nfs_sigintr(nmp, rep, rep->r_lwp)) != 0){ 255 so->so_state &= ~SS_ISCONNECTING; 256 splx(s); 257 goto bad; 258 } 259 } 260 if (so->so_error) { 261 error = so->so_error; 262 so->so_error = 0; 263 splx(s); 264 goto bad; 265 } 266 splx(s); 267 } 268 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 269 so->so_rcv.sb_timeo = (5 * hz); 270 so->so_snd.sb_timeo = (5 * hz); 271 } else { 272 /* 273 * enable receive timeout to detect server crash and reconnect. 274 * otherwise, we can be stuck in soreceive forever. 275 */ 276 so->so_rcv.sb_timeo = (5 * hz); 277 so->so_snd.sb_timeo = 0; 278 } 279 if (nmp->nm_sotype == SOCK_DGRAM) { 280 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 281 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 282 NFS_MAXPKTHDR) * 2; 283 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 284 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 285 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 286 NFS_MAXPKTHDR) * 2; 287 } else { 288 if (nmp->nm_sotype != SOCK_STREAM) 289 panic("nfscon sotype"); 290 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 291 m = m_get(M_WAIT, MT_SOOPTS); 292 MCLAIM(m, so->so_mowner); 293 *mtod(m, int32_t *) = 1; 294 m->m_len = sizeof(int32_t); 295 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 296 } 297 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 298 m = m_get(M_WAIT, MT_SOOPTS); 299 MCLAIM(m, so->so_mowner); 300 *mtod(m, int32_t *) = 1; 301 m->m_len = sizeof(int32_t); 302 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 303 } 304 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 305 sizeof (u_int32_t)) * 2; 306 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 307 sizeof (u_int32_t)) * 2; 308 } 309 error = soreserve(so, sndreserve, rcvreserve); 310 if (error) 311 goto bad; 312 so->so_rcv.sb_flags |= SB_NOINTR; 313 so->so_snd.sb_flags |= SB_NOINTR; 314 315 /* Initialize other non-zero congestion variables */ 316 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 317 NFS_TIMEO << 3; 318 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 319 nmp->nm_sdrtt[3] = 0; 320 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 321 nmp->nm_sent = 0; 322 nmp->nm_timeouts = 0; 323 return (0); 324 325 bad: 326 nfs_disconnect(nmp); 327 return (error); 328 } 329 330 /* 331 * Reconnect routine: 332 * Called when a connection is broken on a reliable protocol. 333 * - clean up the old socket 334 * - nfs_connect() again 335 * - set R_MUSTRESEND for all outstanding requests on mount point 336 * If this fails the mount point is DEAD! 337 * nb: Must be called with the nfs_sndlock() set on the mount point. 338 */ 339 int 340 nfs_reconnect(rep, l) 341 struct nfsreq *rep; 342 struct lwp *l; 343 { 344 struct nfsreq *rp; 345 struct nfsmount *nmp = rep->r_nmp; 346 int error; 347 348 nfs_disconnect(nmp); 349 while ((error = nfs_connect(nmp, rep, l)) != 0) { 350 if (error == EINTR || error == ERESTART) 351 return (EINTR); 352 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0); 353 } 354 355 /* 356 * Loop through outstanding request list and fix up all requests 357 * on old socket. 358 */ 359 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 360 if (rp->r_nmp == nmp) { 361 if ((rp->r_flags & R_MUSTRESEND) == 0) 362 rp->r_flags |= R_MUSTRESEND | R_REXMITTED; 363 rp->r_rexmit = 0; 364 } 365 } 366 return (0); 367 } 368 369 /* 370 * NFS disconnect. Clean up and unlink. 371 */ 372 void 373 nfs_disconnect(nmp) 374 struct nfsmount *nmp; 375 { 376 struct socket *so; 377 int drain = 0; 378 379 if (nmp->nm_so) { 380 so = nmp->nm_so; 381 nmp->nm_so = (struct socket *)0; 382 soshutdown(so, 2); 383 drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0; 384 if (drain) { 385 /* 386 * soshutdown() above should wake up the current 387 * listener. 388 * Now wake up those waiting for the receive lock, and 389 * wait for them to go away unhappy, to prevent *nmp 390 * from evaporating while they're sleeping. 391 */ 392 while (nmp->nm_waiters > 0) { 393 wakeup (&nmp->nm_iflag); 394 (void) tsleep(&nmp->nm_waiters, PVFS, 395 "nfsdis", 0); 396 } 397 } 398 soclose(so); 399 } 400 #ifdef DIAGNOSTIC 401 if (drain && (nmp->nm_waiters > 0)) 402 panic("nfs_disconnect: waiters left after drain?"); 403 #endif 404 } 405 406 void 407 nfs_safedisconnect(nmp) 408 struct nfsmount *nmp; 409 { 410 struct nfsreq dummyreq; 411 412 memset(&dummyreq, 0, sizeof(dummyreq)); 413 dummyreq.r_nmp = nmp; 414 nfs_rcvlock(&dummyreq); /* XXX ignored error return */ 415 nfs_disconnect(nmp); 416 nfs_rcvunlock(nmp); 417 } 418 419 /* 420 * This is the nfs send routine. For connection based socket types, it 421 * must be called with an nfs_sndlock() on the socket. 422 * "rep == NULL" indicates that it has been called from a server. 423 * For the client side: 424 * - return EINTR if the RPC is terminated, 0 otherwise 425 * - set R_MUSTRESEND if the send fails for any reason 426 * - do any cleanup required by recoverable socket errors (? ? ?) 427 * For the server side: 428 * - return EINTR or ERESTART if interrupted by a signal 429 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 430 * - do any cleanup required by recoverable socket errors (? ? ?) 431 */ 432 int 433 nfs_send(so, nam, top, rep, l) 434 struct socket *so; 435 struct mbuf *nam; 436 struct mbuf *top; 437 struct nfsreq *rep; 438 struct lwp *l; 439 { 440 struct mbuf *sendnam; 441 int error, soflags, flags; 442 443 /* XXX nfs_doio()/nfs_request() calls with rep->r_lwp == NULL */ 444 if (l == NULL && rep->r_lwp == NULL) 445 l = curlwp; 446 447 if (rep) { 448 if (rep->r_flags & R_SOFTTERM) { 449 m_freem(top); 450 return (EINTR); 451 } 452 if ((so = rep->r_nmp->nm_so) == NULL) { 453 rep->r_flags |= R_MUSTRESEND; 454 m_freem(top); 455 return (0); 456 } 457 rep->r_flags &= ~R_MUSTRESEND; 458 soflags = rep->r_nmp->nm_soflags; 459 } else 460 soflags = so->so_proto->pr_flags; 461 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 462 sendnam = (struct mbuf *)0; 463 else 464 sendnam = nam; 465 if (so->so_type == SOCK_SEQPACKET) 466 flags = MSG_EOR; 467 else 468 flags = 0; 469 470 error = (*so->so_send)(so, sendnam, (struct uio *)0, top, 471 (struct mbuf *)0, flags, l); 472 if (error) { 473 if (rep) { 474 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 475 /* 476 * We're too fast for the network/driver, 477 * and UDP isn't flowcontrolled. 478 * We need to resend. This is not fatal, 479 * just try again. 480 * 481 * Could be smarter here by doing some sort 482 * of a backoff, but this is rare. 483 */ 484 rep->r_flags |= R_MUSTRESEND; 485 } else { 486 if (error != EPIPE) 487 log(LOG_INFO, 488 "nfs send error %d for %s\n", 489 error, 490 rep->r_nmp->nm_mountp-> 491 mnt_stat.f_mntfromname); 492 /* 493 * Deal with errors for the client side. 494 */ 495 if (rep->r_flags & R_SOFTTERM) 496 error = EINTR; 497 else 498 rep->r_flags |= R_MUSTRESEND; 499 } 500 } else { 501 /* 502 * See above. This error can happen under normal 503 * circumstances and the log is too noisy. 504 * The error will still show up in nfsstat. 505 */ 506 if (error != ENOBUFS || so->so_type != SOCK_DGRAM) 507 log(LOG_INFO, "nfsd send error %d\n", error); 508 } 509 510 /* 511 * Handle any recoverable (soft) socket errors here. (? ? ?) 512 */ 513 if (error != EINTR && error != ERESTART && 514 error != EWOULDBLOCK && error != EPIPE) 515 error = 0; 516 } 517 return (error); 518 } 519 520 #ifdef NFS 521 /* 522 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 523 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 524 * Mark and consolidate the data into a new mbuf list. 525 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 526 * small mbufs. 527 * For SOCK_STREAM we must be very careful to read an entire record once 528 * we have read any of it, even if the system call has been interrupted. 529 */ 530 int 531 nfs_receive(rep, aname, mp, l) 532 struct nfsreq *rep; 533 struct mbuf **aname; 534 struct mbuf **mp; 535 struct lwp *l; 536 { 537 struct proc *p; 538 struct socket *so; 539 struct uio auio; 540 struct iovec aio; 541 struct mbuf *m; 542 struct mbuf *control; 543 u_int32_t len; 544 struct mbuf **getnam; 545 int error, sotype, rcvflg; 546 547 p = l->l_proc; 548 /* 549 * Set up arguments for soreceive() 550 */ 551 *mp = (struct mbuf *)0; 552 *aname = (struct mbuf *)0; 553 sotype = rep->r_nmp->nm_sotype; 554 555 /* 556 * For reliable protocols, lock against other senders/receivers 557 * in case a reconnect is necessary. 558 * For SOCK_STREAM, first get the Record Mark to find out how much 559 * more there is to get. 560 * We must lock the socket against other receivers 561 * until we have an entire rpc request/reply. 562 */ 563 if (sotype != SOCK_DGRAM) { 564 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 565 if (error) 566 return (error); 567 tryagain: 568 /* 569 * Check for fatal errors and resending request. 570 */ 571 /* 572 * Ugh: If a reconnect attempt just happened, nm_so 573 * would have changed. NULL indicates a failed 574 * attempt that has essentially shut down this 575 * mount point. 576 */ 577 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 578 nfs_sndunlock(&rep->r_nmp->nm_iflag); 579 return (EINTR); 580 } 581 so = rep->r_nmp->nm_so; 582 if (!so) { 583 error = nfs_reconnect(rep, l); 584 if (error) { 585 nfs_sndunlock(&rep->r_nmp->nm_iflag); 586 return (error); 587 } 588 goto tryagain; 589 } 590 while (rep->r_flags & R_MUSTRESEND) { 591 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 592 nfsstats.rpcretries++; 593 rep->r_rtt = 0; 594 rep->r_flags &= ~R_TIMING; 595 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, l); 596 if (error) { 597 if (error == EINTR || error == ERESTART || 598 (error = nfs_reconnect(rep, l)) != 0) { 599 nfs_sndunlock(&rep->r_nmp->nm_iflag); 600 return (error); 601 } 602 goto tryagain; 603 } 604 } 605 nfs_sndunlock(&rep->r_nmp->nm_iflag); 606 if (sotype == SOCK_STREAM) { 607 aio.iov_base = (caddr_t) &len; 608 aio.iov_len = sizeof(u_int32_t); 609 auio.uio_iov = &aio; 610 auio.uio_iovcnt = 1; 611 auio.uio_segflg = UIO_SYSSPACE; 612 auio.uio_rw = UIO_READ; 613 auio.uio_offset = 0; 614 auio.uio_resid = sizeof(u_int32_t); 615 auio.uio_lwp = NULL; 616 do { 617 rcvflg = MSG_WAITALL; 618 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 619 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 620 if (error == EWOULDBLOCK && rep) { 621 if (rep->r_flags & R_SOFTTERM) 622 return (EINTR); 623 /* 624 * if it seems that the server died after it 625 * received our request, set EPIPE so that 626 * we'll reconnect and retransmit requests. 627 */ 628 if (rep->r_rexmit >= rep->r_nmp->nm_retry) { 629 nfsstats.rpctimeouts++; 630 error = EPIPE; 631 } 632 } 633 } while (error == EWOULDBLOCK); 634 if (!error && auio.uio_resid > 0) { 635 /* 636 * Don't log a 0 byte receive; it means 637 * that the socket has been closed, and 638 * can happen during normal operation 639 * (forcible unmount or Solaris server). 640 */ 641 if (auio.uio_resid != sizeof (u_int32_t)) 642 log(LOG_INFO, 643 "short receive (%lu/%lu) from nfs server %s\n", 644 (u_long)sizeof(u_int32_t) - auio.uio_resid, 645 (u_long)sizeof(u_int32_t), 646 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 647 error = EPIPE; 648 } 649 if (error) 650 goto errout; 651 len = ntohl(len) & ~0x80000000; 652 /* 653 * This is SERIOUS! We are out of sync with the sender 654 * and forcing a disconnect/reconnect is all I can do. 655 */ 656 if (len > NFS_MAXPACKET) { 657 log(LOG_ERR, "%s (%d) from nfs server %s\n", 658 "impossible packet length", 659 len, 660 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 661 error = EFBIG; 662 goto errout; 663 } 664 auio.uio_resid = len; 665 do { 666 rcvflg = MSG_WAITALL; 667 error = (*so->so_receive)(so, (struct mbuf **)0, 668 &auio, mp, (struct mbuf **)0, &rcvflg); 669 } while (error == EWOULDBLOCK || error == EINTR || 670 error == ERESTART); 671 if (!error && auio.uio_resid > 0) { 672 if (len != auio.uio_resid) 673 log(LOG_INFO, 674 "short receive (%lu/%d) from nfs server %s\n", 675 (u_long)len - auio.uio_resid, len, 676 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 677 error = EPIPE; 678 } 679 } else { 680 /* 681 * NB: Since uio_resid is big, MSG_WAITALL is ignored 682 * and soreceive() will return when it has either a 683 * control msg or a data msg. 684 * We have no use for control msg., but must grab them 685 * and then throw them away so we know what is going 686 * on. 687 */ 688 auio.uio_resid = len = 100000000; /* Anything Big */ 689 auio.uio_lwp = l; 690 do { 691 rcvflg = 0; 692 error = (*so->so_receive)(so, (struct mbuf **)0, 693 &auio, mp, &control, &rcvflg); 694 if (control) 695 m_freem(control); 696 if (error == EWOULDBLOCK && rep) { 697 if (rep->r_flags & R_SOFTTERM) 698 return (EINTR); 699 } 700 } while (error == EWOULDBLOCK || 701 (!error && *mp == NULL && control)); 702 if ((rcvflg & MSG_EOR) == 0) 703 printf("Egad!!\n"); 704 if (!error && *mp == NULL) 705 error = EPIPE; 706 len -= auio.uio_resid; 707 } 708 errout: 709 if (error && error != EINTR && error != ERESTART) { 710 m_freem(*mp); 711 *mp = (struct mbuf *)0; 712 if (error != EPIPE) 713 log(LOG_INFO, 714 "receive error %d from nfs server %s\n", 715 error, 716 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 717 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 718 if (!error) 719 error = nfs_reconnect(rep, l); 720 if (!error) 721 goto tryagain; 722 else 723 nfs_sndunlock(&rep->r_nmp->nm_iflag); 724 } 725 } else { 726 if ((so = rep->r_nmp->nm_so) == NULL) 727 return (EACCES); 728 if (so->so_state & SS_ISCONNECTED) 729 getnam = (struct mbuf **)0; 730 else 731 getnam = aname; 732 auio.uio_resid = len = 1000000; 733 auio.uio_lwp = l; 734 do { 735 rcvflg = 0; 736 error = (*so->so_receive)(so, getnam, &auio, mp, 737 (struct mbuf **)0, &rcvflg); 738 if (error == EWOULDBLOCK && 739 (rep->r_flags & R_SOFTTERM)) 740 return (EINTR); 741 } while (error == EWOULDBLOCK); 742 len -= auio.uio_resid; 743 if (!error && *mp == NULL) 744 error = EPIPE; 745 } 746 if (error) { 747 m_freem(*mp); 748 *mp = (struct mbuf *)0; 749 } 750 return (error); 751 } 752 753 /* 754 * Implement receipt of reply on a socket. 755 * We must search through the list of received datagrams matching them 756 * with outstanding requests using the xid, until ours is found. 757 */ 758 /* ARGSUSED */ 759 int 760 nfs_reply(myrep, lwp) 761 struct nfsreq *myrep; 762 struct lwp *lwp; 763 { 764 struct nfsreq *rep; 765 struct nfsmount *nmp = myrep->r_nmp; 766 int32_t t1; 767 struct mbuf *mrep, *nam, *md; 768 u_int32_t rxid, *tl; 769 caddr_t dpos, cp2; 770 int error; 771 772 /* 773 * Loop around until we get our own reply 774 */ 775 for (;;) { 776 /* 777 * Lock against other receivers so that I don't get stuck in 778 * sbwait() after someone else has received my reply for me. 779 * Also necessary for connection based protocols to avoid 780 * race conditions during a reconnect. 781 */ 782 error = nfs_rcvlock(myrep); 783 if (error == EALREADY) 784 return (0); 785 if (error) 786 return (error); 787 /* 788 * Get the next Rpc reply off the socket 789 */ 790 nmp->nm_waiters++; 791 error = nfs_receive(myrep, &nam, &mrep, lwp); 792 nfs_rcvunlock(nmp); 793 if (error) { 794 795 if (nmp->nm_iflag & NFSMNT_DISMNT) { 796 /* 797 * Oops, we're going away now.. 798 */ 799 nmp->nm_waiters--; 800 wakeup (&nmp->nm_waiters); 801 return error; 802 } 803 nmp->nm_waiters--; 804 /* 805 * Ignore routing errors on connectionless protocols? ? 806 */ 807 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 808 nmp->nm_so->so_error = 0; 809 #ifdef DEBUG 810 printf("nfs_reply: ignoring error %d\n", error); 811 #endif 812 if (myrep->r_flags & R_GETONEREP) 813 return (0); 814 continue; 815 } 816 return (error); 817 } 818 nmp->nm_waiters--; 819 if (nam) 820 m_freem(nam); 821 822 /* 823 * Get the xid and check that it is an rpc reply 824 */ 825 md = mrep; 826 dpos = mtod(md, caddr_t); 827 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 828 rxid = *tl++; 829 if (*tl != rpc_reply) { 830 #ifndef NFS_V2_ONLY 831 if (nmp->nm_flag & NFSMNT_NQNFS) { 832 if (nqnfs_callback(nmp, mrep, md, dpos, 833 myrep->r_lwp)) 834 nfsstats.rpcinvalid++; 835 } else 836 #endif 837 { 838 nfsstats.rpcinvalid++; 839 m_freem(mrep); 840 } 841 nfsmout: 842 if (myrep->r_flags & R_GETONEREP) 843 return (0); 844 continue; 845 } 846 847 /* 848 * Loop through the request list to match up the reply 849 * Iff no match, just drop the datagram 850 */ 851 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 852 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 853 /* Found it.. */ 854 rep->r_mrep = mrep; 855 rep->r_md = md; 856 rep->r_dpos = dpos; 857 if (nfsrtton) { 858 struct rttl *rt; 859 860 rt = &nfsrtt.rttl[nfsrtt.pos]; 861 rt->proc = rep->r_procnum; 862 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 863 rt->sent = nmp->nm_sent; 864 rt->cwnd = nmp->nm_cwnd; 865 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 866 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 867 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx; 868 rt->tstamp = time; 869 if (rep->r_flags & R_TIMING) 870 rt->rtt = rep->r_rtt; 871 else 872 rt->rtt = 1000000; 873 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 874 } 875 /* 876 * Update congestion window. 877 * Do the additive increase of 878 * one rpc/rtt. 879 */ 880 if (nmp->nm_cwnd <= nmp->nm_sent) { 881 nmp->nm_cwnd += 882 (NFS_CWNDSCALE * NFS_CWNDSCALE + 883 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 884 if (nmp->nm_cwnd > NFS_MAXCWND) 885 nmp->nm_cwnd = NFS_MAXCWND; 886 } 887 rep->r_flags &= ~R_SENT; 888 nmp->nm_sent -= NFS_CWNDSCALE; 889 /* 890 * Update rtt using a gain of 0.125 on the mean 891 * and a gain of 0.25 on the deviation. 892 */ 893 if (rep->r_flags & R_TIMING) { 894 /* 895 * Since the timer resolution of 896 * NFS_HZ is so course, it can often 897 * result in r_rtt == 0. Since 898 * r_rtt == N means that the actual 899 * rtt is between N+dt and N+2-dt ticks, 900 * add 1. 901 */ 902 t1 = rep->r_rtt + 1; 903 t1 -= (NFS_SRTT(rep) >> 3); 904 NFS_SRTT(rep) += t1; 905 if (t1 < 0) 906 t1 = -t1; 907 t1 -= (NFS_SDRTT(rep) >> 2); 908 NFS_SDRTT(rep) += t1; 909 } 910 nmp->nm_timeouts = 0; 911 break; 912 } 913 } 914 /* 915 * If not matched to a request, drop it. 916 * If it's mine, get out. 917 */ 918 if (rep == 0) { 919 nfsstats.rpcunexpected++; 920 m_freem(mrep); 921 } else if (rep == myrep) { 922 if (rep->r_mrep == NULL) 923 panic("nfsreply nil"); 924 return (0); 925 } 926 if (myrep->r_flags & R_GETONEREP) 927 return (0); 928 } 929 } 930 931 /* 932 * nfs_request - goes something like this 933 * - fill in request struct 934 * - links it into list 935 * - calls nfs_send() for first transmit 936 * - calls nfs_receive() to get reply 937 * - break down rpc header and return with nfs reply pointed to 938 * by mrep or error 939 * nb: always frees up mreq mbuf list 940 */ 941 int 942 nfs_request(np, mrest, procnum, lwp, cred, mrp, mdp, dposp, rexmitp) 943 struct nfsnode *np; 944 struct mbuf *mrest; 945 int procnum; 946 struct lwp *lwp; 947 struct ucred *cred; 948 struct mbuf **mrp; 949 struct mbuf **mdp; 950 caddr_t *dposp; 951 int *rexmitp; 952 { 953 struct mbuf *m, *mrep; 954 struct nfsreq *rep; 955 u_int32_t *tl; 956 int i; 957 struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount); 958 struct mbuf *md, *mheadend; 959 char nickv[RPCX_NICKVERF]; 960 time_t reqtime, waituntil; 961 caddr_t dpos, cp2; 962 int t1, s, error = 0, mrest_len, auth_len, auth_type; 963 int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0; 964 int verf_len, verf_type; 965 u_int32_t xid; 966 char *auth_str, *verf_str; 967 NFSKERBKEY_T key; /* save session key */ 968 struct ucred acred; 969 #ifndef NFS_V2_ONLY 970 int nqlflag, cachable; 971 u_quad_t frev; 972 #endif 973 struct mbuf *mrest_backup = NULL; 974 struct ucred *origcred = NULL; /* XXX: gcc */ 975 boolean_t retry_cred = TRUE; 976 boolean_t use_opencred = (np->n_flag & NUSEOPENCRED) != 0; 977 978 if (rexmitp != NULL) 979 *rexmitp = 0; 980 981 tryagain_cred: 982 KASSERT(cred != NULL); 983 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 984 rep->r_nmp = nmp; 985 rep->r_lwp = lwp; 986 rep->r_procnum = procnum; 987 i = 0; 988 m = mrest; 989 while (m) { 990 i += m->m_len; 991 m = m->m_next; 992 } 993 mrest_len = i; 994 995 /* 996 * Get the RPC header with authorization. 997 */ 998 kerbauth: 999 verf_str = auth_str = (char *)0; 1000 if (nmp->nm_flag & NFSMNT_KERB) { 1001 verf_str = nickv; 1002 verf_len = sizeof (nickv); 1003 auth_type = RPCAUTH_KERB4; 1004 memset((caddr_t)key, 0, sizeof (key)); 1005 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 1006 &auth_len, verf_str, verf_len)) { 1007 error = nfs_getauth(nmp, rep, cred, &auth_str, 1008 &auth_len, verf_str, &verf_len, key); 1009 if (error) { 1010 free((caddr_t)rep, M_NFSREQ); 1011 m_freem(mrest); 1012 return (error); 1013 } 1014 } 1015 retry_cred = FALSE; 1016 } else { 1017 /* AUTH_UNIX */ 1018 uid_t uid; 1019 gid_t gid; 1020 1021 /* 1022 * on the most unix filesystems, permission checks are 1023 * done when the file is open(2)'ed. 1024 * ie. once a file is successfully open'ed, 1025 * following i/o operations never fail with EACCES. 1026 * we try to follow the semantics as far as possible. 1027 * 1028 * note that we expect that the nfs server always grant 1029 * accesses by the file's owner. 1030 */ 1031 origcred = cred; 1032 switch (procnum) { 1033 case NFSPROC_READ: 1034 case NFSPROC_WRITE: 1035 case NFSPROC_COMMIT: 1036 uid = np->n_vattr->va_uid; 1037 gid = np->n_vattr->va_gid; 1038 if (cred->cr_uid == uid && cred->cr_gid == gid) { 1039 retry_cred = FALSE; 1040 break; 1041 } 1042 if (use_opencred) 1043 break; 1044 acred.cr_uid = uid; 1045 acred.cr_gid = gid; 1046 acred.cr_ngroups = 0; 1047 acred.cr_ref = 2; /* Just to be safe.. */ 1048 cred = &acred; 1049 break; 1050 default: 1051 retry_cred = FALSE; 1052 break; 1053 } 1054 /* 1055 * backup mbuf chain if we can need it later to retry. 1056 * 1057 * XXX maybe we can keep a direct reference to 1058 * mrest without doing m_copym, but it's ...ugly. 1059 */ 1060 if (retry_cred) 1061 mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT); 1062 auth_type = RPCAUTH_UNIX; 1063 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 1064 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 1065 5 * NFSX_UNSIGNED; 1066 } 1067 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 1068 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 1069 if (auth_str) 1070 free(auth_str, M_TEMP); 1071 1072 /* 1073 * For stream protocols, insert a Sun RPC Record Mark. 1074 */ 1075 if (nmp->nm_sotype == SOCK_STREAM) { 1076 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 1077 *mtod(m, u_int32_t *) = htonl(0x80000000 | 1078 (m->m_pkthdr.len - NFSX_UNSIGNED)); 1079 } 1080 rep->r_mreq = m; 1081 rep->r_xid = xid; 1082 tryagain: 1083 if (nmp->nm_flag & NFSMNT_SOFT) 1084 rep->r_retry = nmp->nm_retry; 1085 else 1086 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1087 rep->r_rtt = rep->r_rexmit = 0; 1088 if (proct[procnum] > 0) 1089 rep->r_flags = R_TIMING; 1090 else 1091 rep->r_flags = 0; 1092 rep->r_mrep = NULL; 1093 1094 /* 1095 * Do the client side RPC. 1096 */ 1097 nfsstats.rpcrequests++; 1098 /* 1099 * Chain request into list of outstanding requests. Be sure 1100 * to put it LAST so timer finds oldest requests first. 1101 */ 1102 s = splsoftnet(); 1103 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1104 1105 /* Get send time for nqnfs */ 1106 reqtime = time.tv_sec; 1107 1108 /* 1109 * If backing off another request or avoiding congestion, don't 1110 * send this one now but let timer do it. If not timing a request, 1111 * do it now. 1112 */ 1113 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1114 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1115 nmp->nm_sent < nmp->nm_cwnd)) { 1116 splx(s); 1117 if (nmp->nm_soflags & PR_CONNREQUIRED) 1118 error = nfs_sndlock(&nmp->nm_iflag, rep); 1119 if (!error) { 1120 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1121 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp); 1122 if (nmp->nm_soflags & PR_CONNREQUIRED) 1123 nfs_sndunlock(&nmp->nm_iflag); 1124 } 1125 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1126 nmp->nm_sent += NFS_CWNDSCALE; 1127 rep->r_flags |= R_SENT; 1128 } 1129 } else { 1130 splx(s); 1131 rep->r_rtt = -1; 1132 } 1133 1134 /* 1135 * Wait for the reply from our send or the timer's. 1136 */ 1137 if (!error || error == EPIPE) 1138 error = nfs_reply(rep, lwp); 1139 1140 /* 1141 * RPC done, unlink the request. 1142 */ 1143 s = splsoftnet(); 1144 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1145 splx(s); 1146 1147 /* 1148 * Decrement the outstanding request count. 1149 */ 1150 if (rep->r_flags & R_SENT) { 1151 rep->r_flags &= ~R_SENT; /* paranoia */ 1152 nmp->nm_sent -= NFS_CWNDSCALE; 1153 } 1154 1155 if (rexmitp != NULL) { 1156 int rexmit; 1157 1158 if (nmp->nm_sotype != SOCK_DGRAM) 1159 rexmit = (rep->r_flags & R_REXMITTED) != 0; 1160 else 1161 rexmit = rep->r_rexmit; 1162 *rexmitp = rexmit; 1163 } 1164 1165 /* 1166 * If there was a successful reply and a tprintf msg. 1167 * tprintf a response. 1168 */ 1169 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1170 nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1171 "is alive again"); 1172 mrep = rep->r_mrep; 1173 md = rep->r_md; 1174 dpos = rep->r_dpos; 1175 if (error) 1176 goto nfsmout; 1177 1178 /* 1179 * break down the rpc header and check if ok 1180 */ 1181 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1182 if (*tl++ == rpc_msgdenied) { 1183 if (*tl == rpc_mismatch) 1184 error = EOPNOTSUPP; 1185 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1186 if (!failed_auth) { 1187 failed_auth++; 1188 mheadend->m_next = (struct mbuf *)0; 1189 m_freem(mrep); 1190 m_freem(rep->r_mreq); 1191 goto kerbauth; 1192 } else 1193 error = EAUTH; 1194 } else 1195 error = EACCES; 1196 m_freem(mrep); 1197 goto nfsmout; 1198 } 1199 1200 /* 1201 * Grab any Kerberos verifier, otherwise just throw it away. 1202 */ 1203 verf_type = fxdr_unsigned(int, *tl++); 1204 i = fxdr_unsigned(int32_t, *tl); 1205 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1206 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1207 if (error) 1208 goto nfsmout; 1209 } else if (i > 0) 1210 nfsm_adv(nfsm_rndup(i)); 1211 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1212 /* 0 == ok */ 1213 if (*tl == 0) { 1214 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1215 if (*tl != 0) { 1216 error = fxdr_unsigned(int, *tl); 1217 switch (error) { 1218 case NFSERR_PERM: 1219 error = EPERM; 1220 break; 1221 1222 case NFSERR_NOENT: 1223 error = ENOENT; 1224 break; 1225 1226 case NFSERR_IO: 1227 error = EIO; 1228 break; 1229 1230 case NFSERR_NXIO: 1231 error = ENXIO; 1232 break; 1233 1234 case NFSERR_ACCES: 1235 error = EACCES; 1236 if (!retry_cred) 1237 break; 1238 m_freem(mrep); 1239 m_freem(rep->r_mreq); 1240 FREE(rep, M_NFSREQ); 1241 use_opencred = !use_opencred; 1242 if (mrest_backup == NULL) 1243 /* m_copym failure */ 1244 return ENOMEM; 1245 mrest = mrest_backup; 1246 mrest_backup = NULL; 1247 cred = origcred; 1248 error = 0; 1249 retry_cred = FALSE; 1250 goto tryagain_cred; 1251 1252 case NFSERR_EXIST: 1253 error = EEXIST; 1254 break; 1255 1256 case NFSERR_XDEV: 1257 error = EXDEV; 1258 break; 1259 1260 case NFSERR_NODEV: 1261 error = ENODEV; 1262 break; 1263 1264 case NFSERR_NOTDIR: 1265 error = ENOTDIR; 1266 break; 1267 1268 case NFSERR_ISDIR: 1269 error = EISDIR; 1270 break; 1271 1272 case NFSERR_INVAL: 1273 error = EINVAL; 1274 break; 1275 1276 case NFSERR_FBIG: 1277 error = EFBIG; 1278 break; 1279 1280 case NFSERR_NOSPC: 1281 error = ENOSPC; 1282 break; 1283 1284 case NFSERR_ROFS: 1285 error = EROFS; 1286 break; 1287 1288 case NFSERR_MLINK: 1289 error = EMLINK; 1290 break; 1291 1292 case NFSERR_TIMEDOUT: 1293 error = ETIMEDOUT; 1294 break; 1295 1296 case NFSERR_NAMETOL: 1297 error = ENAMETOOLONG; 1298 break; 1299 1300 case NFSERR_NOTEMPTY: 1301 error = ENOTEMPTY; 1302 break; 1303 1304 case NFSERR_DQUOT: 1305 error = EDQUOT; 1306 break; 1307 1308 case NFSERR_STALE: 1309 /* 1310 * If the File Handle was stale, invalidate the 1311 * lookup cache, just in case. 1312 */ 1313 error = ESTALE; 1314 cache_purge(NFSTOV(np)); 1315 break; 1316 1317 case NFSERR_REMOTE: 1318 error = EREMOTE; 1319 break; 1320 1321 case NFSERR_WFLUSH: 1322 case NFSERR_BADHANDLE: 1323 case NFSERR_NOT_SYNC: 1324 case NFSERR_BAD_COOKIE: 1325 error = EINVAL; 1326 break; 1327 1328 case NFSERR_NOTSUPP: 1329 error = ENOTSUP; 1330 break; 1331 1332 case NFSERR_TOOSMALL: 1333 case NFSERR_SERVERFAULT: 1334 case NFSERR_BADTYPE: 1335 error = EINVAL; 1336 break; 1337 1338 case NFSERR_TRYLATER: 1339 if ((nmp->nm_flag & NFSMNT_NFSV3) == 0) 1340 break; 1341 m_freem(mrep); 1342 error = 0; 1343 waituntil = time.tv_sec + trylater_delay; 1344 while (time.tv_sec < waituntil) 1345 (void) tsleep((caddr_t)&lbolt, 1346 PSOCK, "nqnfstry", 0); 1347 trylater_delay *= NFS_TRYLATERDELMUL; 1348 if (trylater_delay > NFS_TRYLATERDELMAX) 1349 trylater_delay = NFS_TRYLATERDELMAX; 1350 /* 1351 * RFC1813: 1352 * The client should wait and then try 1353 * the request with a new RPC transaction ID. 1354 */ 1355 nfs_renewxid(rep); 1356 goto tryagain; 1357 1358 case NFSERR_STALEWRITEVERF: 1359 error = EINVAL; 1360 break; 1361 1362 default: 1363 #ifdef DIAGNOSTIC 1364 printf("Invalid rpc error code %d\n", error); 1365 #endif 1366 error = EINVAL; 1367 break; 1368 } 1369 1370 if (nmp->nm_flag & NFSMNT_NFSV3) { 1371 *mrp = mrep; 1372 *mdp = md; 1373 *dposp = dpos; 1374 error |= NFSERR_RETERR; 1375 } else 1376 m_freem(mrep); 1377 goto nfsmout; 1378 } 1379 1380 /* 1381 * note which credential worked to minimize number of retries. 1382 */ 1383 if (use_opencred) 1384 np->n_flag |= NUSEOPENCRED; 1385 else 1386 np->n_flag &= ~NUSEOPENCRED; 1387 1388 #ifndef NFS_V2_ONLY 1389 /* 1390 * For nqnfs, get any lease in reply 1391 */ 1392 if (nmp->nm_flag & NFSMNT_NQNFS) { 1393 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1394 if (*tl) { 1395 nqlflag = fxdr_unsigned(int, *tl); 1396 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1397 cachable = fxdr_unsigned(int, *tl++); 1398 reqtime += fxdr_unsigned(int, *tl++); 1399 if (reqtime > time.tv_sec) { 1400 frev = fxdr_hyper(tl); 1401 nqnfs_clientlease(nmp, np, nqlflag, 1402 cachable, reqtime, frev); 1403 } 1404 } 1405 } 1406 #endif 1407 *mrp = mrep; 1408 *mdp = md; 1409 *dposp = dpos; 1410 1411 KASSERT(error == 0); 1412 goto nfsmout; 1413 } 1414 m_freem(mrep); 1415 error = EPROTONOSUPPORT; 1416 nfsmout: 1417 m_freem(rep->r_mreq); 1418 free((caddr_t)rep, M_NFSREQ); 1419 m_freem(mrest_backup); 1420 return (error); 1421 } 1422 #endif /* NFS */ 1423 1424 /* 1425 * Generate the rpc reply header 1426 * siz arg. is used to decide if adding a cluster is worthwhile 1427 */ 1428 int 1429 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1430 int siz; 1431 struct nfsrv_descript *nd; 1432 struct nfssvc_sock *slp; 1433 int err; 1434 int cache; 1435 u_quad_t *frev; 1436 struct mbuf **mrq; 1437 struct mbuf **mbp; 1438 caddr_t *bposp; 1439 { 1440 u_int32_t *tl; 1441 struct mbuf *mreq; 1442 caddr_t bpos; 1443 struct mbuf *mb; 1444 1445 mreq = m_gethdr(M_WAIT, MT_DATA); 1446 MCLAIM(mreq, &nfs_mowner); 1447 mb = mreq; 1448 /* 1449 * If this is a big reply, use a cluster else 1450 * try and leave leading space for the lower level headers. 1451 */ 1452 siz += RPC_REPLYSIZ; 1453 if (siz >= max_datalen) { 1454 m_clget(mreq, M_WAIT); 1455 } else 1456 mreq->m_data += max_hdr; 1457 tl = mtod(mreq, u_int32_t *); 1458 mreq->m_len = 6 * NFSX_UNSIGNED; 1459 bpos = ((caddr_t)tl) + mreq->m_len; 1460 *tl++ = txdr_unsigned(nd->nd_retxid); 1461 *tl++ = rpc_reply; 1462 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1463 *tl++ = rpc_msgdenied; 1464 if (err & NFSERR_AUTHERR) { 1465 *tl++ = rpc_autherr; 1466 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1467 mreq->m_len -= NFSX_UNSIGNED; 1468 bpos -= NFSX_UNSIGNED; 1469 } else { 1470 *tl++ = rpc_mismatch; 1471 *tl++ = txdr_unsigned(RPC_VER2); 1472 *tl = txdr_unsigned(RPC_VER2); 1473 } 1474 } else { 1475 *tl++ = rpc_msgaccepted; 1476 1477 /* 1478 * For Kerberos authentication, we must send the nickname 1479 * verifier back, otherwise just RPCAUTH_NULL. 1480 */ 1481 if (nd->nd_flag & ND_KERBFULL) { 1482 struct nfsuid *nuidp; 1483 struct timeval ktvin, ktvout; 1484 1485 LIST_FOREACH(nuidp, NUIDHASH(slp, nd->nd_cr.cr_uid), 1486 nu_hash) { 1487 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1488 (!nd->nd_nam2 || netaddr_match( 1489 NU_NETFAM(nuidp), &nuidp->nu_haddr, 1490 nd->nd_nam2))) 1491 break; 1492 } 1493 if (nuidp) { 1494 ktvin.tv_sec = 1495 txdr_unsigned(nuidp->nu_timestamp.tv_sec 1496 - 1); 1497 ktvin.tv_usec = 1498 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1499 1500 /* 1501 * Encrypt the timestamp in ecb mode using the 1502 * session key. 1503 */ 1504 #ifdef NFSKERB 1505 XXX 1506 #endif 1507 1508 *tl++ = rpc_auth_kerb; 1509 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1510 *tl = ktvout.tv_sec; 1511 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1512 *tl++ = ktvout.tv_usec; 1513 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1514 } else { 1515 *tl++ = 0; 1516 *tl++ = 0; 1517 } 1518 } else { 1519 *tl++ = 0; 1520 *tl++ = 0; 1521 } 1522 switch (err) { 1523 case EPROGUNAVAIL: 1524 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1525 break; 1526 case EPROGMISMATCH: 1527 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1528 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1529 if (nd->nd_flag & ND_NQNFS) { 1530 *tl++ = txdr_unsigned(3); 1531 *tl = txdr_unsigned(3); 1532 } else { 1533 *tl++ = txdr_unsigned(2); 1534 *tl = txdr_unsigned(3); 1535 } 1536 break; 1537 case EPROCUNAVAIL: 1538 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1539 break; 1540 case EBADRPC: 1541 *tl = txdr_unsigned(RPC_GARBAGE); 1542 break; 1543 default: 1544 *tl = 0; 1545 if (err != NFSERR_RETVOID) { 1546 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1547 if (err) 1548 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1549 else 1550 *tl = 0; 1551 } 1552 break; 1553 }; 1554 } 1555 1556 /* 1557 * For nqnfs, piggyback lease as requested. 1558 */ 1559 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1560 if (nd->nd_flag & ND_LEASE) { 1561 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1562 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1563 *tl++ = txdr_unsigned(cache); 1564 *tl++ = txdr_unsigned(nd->nd_duration); 1565 txdr_hyper(*frev, tl); 1566 } else { 1567 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1568 *tl = 0; 1569 } 1570 } 1571 if (mrq != NULL) 1572 *mrq = mreq; 1573 *mbp = mb; 1574 *bposp = bpos; 1575 if (err != 0 && err != NFSERR_RETVOID) 1576 nfsstats.srvrpc_errs++; 1577 return (0); 1578 } 1579 1580 /* 1581 * Nfs timer routine 1582 * Scan the nfsreq list and retranmit any requests that have timed out 1583 * To avoid retransmission attempts on STREAM sockets (in the future) make 1584 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1585 */ 1586 void 1587 nfs_timer(arg) 1588 void *arg; /* never used */ 1589 { 1590 struct nfsreq *rep; 1591 struct mbuf *m; 1592 struct socket *so; 1593 struct nfsmount *nmp; 1594 int timeo; 1595 int s, error; 1596 #ifdef NFSSERVER 1597 struct nfssvc_sock *slp; 1598 static long lasttime = 0; 1599 u_quad_t cur_usec; 1600 #endif 1601 1602 s = splsoftnet(); 1603 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1604 nmp = rep->r_nmp; 1605 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1606 continue; 1607 if (nfs_sigintr(nmp, rep, rep->r_lwp)) { 1608 rep->r_flags |= R_SOFTTERM; 1609 continue; 1610 } 1611 if (rep->r_rtt >= 0) { 1612 rep->r_rtt++; 1613 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1614 timeo = nmp->nm_timeo; 1615 else 1616 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1617 if (nmp->nm_timeouts > 0) 1618 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1619 if (rep->r_rtt <= timeo) 1620 continue; 1621 if (nmp->nm_timeouts < 1622 (sizeof(nfs_backoff) / sizeof(nfs_backoff[0]))) 1623 nmp->nm_timeouts++; 1624 } 1625 /* 1626 * Check for server not responding 1627 */ 1628 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1629 rep->r_rexmit > nmp->nm_deadthresh) { 1630 nfs_msg(rep->r_lwp, 1631 nmp->nm_mountp->mnt_stat.f_mntfromname, 1632 "not responding"); 1633 rep->r_flags |= R_TPRINTFMSG; 1634 } 1635 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1636 nfsstats.rpctimeouts++; 1637 rep->r_flags |= R_SOFTTERM; 1638 continue; 1639 } 1640 if (nmp->nm_sotype != SOCK_DGRAM) { 1641 if (++rep->r_rexmit > NFS_MAXREXMIT) 1642 rep->r_rexmit = NFS_MAXREXMIT; 1643 continue; 1644 } 1645 if ((so = nmp->nm_so) == NULL) 1646 continue; 1647 1648 /* 1649 * If there is enough space and the window allows.. 1650 * Resend it 1651 * Set r_rtt to -1 in case we fail to send it now. 1652 */ 1653 rep->r_rtt = -1; 1654 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1655 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1656 (rep->r_flags & R_SENT) || 1657 nmp->nm_sent < nmp->nm_cwnd) && 1658 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1659 if (so->so_state & SS_ISCONNECTED) 1660 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1661 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1662 else 1663 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1664 nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0); 1665 if (error) { 1666 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1667 #ifdef DEBUG 1668 printf("nfs_timer: ignoring error %d\n", 1669 error); 1670 #endif 1671 so->so_error = 0; 1672 } 1673 } else { 1674 /* 1675 * Iff first send, start timing 1676 * else turn timing off, backoff timer 1677 * and divide congestion window by 2. 1678 */ 1679 if (rep->r_flags & R_SENT) { 1680 rep->r_flags &= ~R_TIMING; 1681 if (++rep->r_rexmit > NFS_MAXREXMIT) 1682 rep->r_rexmit = NFS_MAXREXMIT; 1683 nmp->nm_cwnd >>= 1; 1684 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1685 nmp->nm_cwnd = NFS_CWNDSCALE; 1686 nfsstats.rpcretries++; 1687 } else { 1688 rep->r_flags |= R_SENT; 1689 nmp->nm_sent += NFS_CWNDSCALE; 1690 } 1691 rep->r_rtt = 0; 1692 } 1693 } 1694 } 1695 1696 #ifdef NFSSERVER 1697 /* 1698 * Call the nqnfs server timer once a second to handle leases. 1699 */ 1700 if (lasttime != time.tv_sec) { 1701 lasttime = time.tv_sec; 1702 nqnfs_serverd(); 1703 } 1704 1705 /* 1706 * Scan the write gathering queues for writes that need to be 1707 * completed now. 1708 */ 1709 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1710 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1711 if (LIST_FIRST(&slp->ns_tq) && 1712 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) 1713 nfsrv_wakenfsd(slp); 1714 } 1715 #endif /* NFSSERVER */ 1716 splx(s); 1717 callout_schedule(&nfs_timer_ch, nfs_ticks); 1718 } 1719 1720 /*ARGSUSED*/ 1721 void 1722 nfs_exit(p, v) 1723 struct proc *p; 1724 void *v; 1725 { 1726 struct nfsreq *rp; 1727 int s = splsoftnet(); 1728 1729 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 1730 if (rp->r_lwp && rp->r_lwp->l_proc == p) 1731 TAILQ_REMOVE(&nfs_reqq, rp, r_chain); 1732 } 1733 splx(s); 1734 } 1735 1736 /* 1737 * Test for a termination condition pending on the process. 1738 * This is used for NFSMNT_INT mounts. 1739 */ 1740 int 1741 nfs_sigintr(nmp, rep, l) 1742 struct nfsmount *nmp; 1743 struct nfsreq *rep; 1744 struct lwp *l; 1745 { 1746 sigset_t ss; 1747 1748 if (rep && (rep->r_flags & R_SOFTTERM)) 1749 return (EINTR); 1750 if (!(nmp->nm_flag & NFSMNT_INT)) 1751 return (0); 1752 if (l) { 1753 sigpending1(l->l_proc, &ss); 1754 #if 0 1755 sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss); 1756 #endif 1757 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1758 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1759 sigismember(&ss, SIGQUIT)) 1760 return (EINTR); 1761 } 1762 return (0); 1763 } 1764 1765 /* 1766 * Lock a socket against others. 1767 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1768 * and also to avoid race conditions between the processes with nfs requests 1769 * in progress when a reconnect is necessary. 1770 */ 1771 int 1772 nfs_sndlock(flagp, rep) 1773 int *flagp; 1774 struct nfsreq *rep; 1775 { 1776 struct lwp *l; 1777 int slpflag = 0, slptimeo = 0; 1778 1779 if (rep) { 1780 l = rep->r_lwp; 1781 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1782 slpflag = PCATCH; 1783 } else 1784 l = (struct lwp *)0; 1785 while (*flagp & NFSMNT_SNDLOCK) { 1786 if (rep && nfs_sigintr(rep->r_nmp, rep, l)) 1787 return (EINTR); 1788 *flagp |= NFSMNT_WANTSND; 1789 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1790 slptimeo); 1791 if (slpflag == PCATCH) { 1792 slpflag = 0; 1793 slptimeo = 2 * hz; 1794 } 1795 } 1796 *flagp |= NFSMNT_SNDLOCK; 1797 return (0); 1798 } 1799 1800 /* 1801 * Unlock the stream socket for others. 1802 */ 1803 void 1804 nfs_sndunlock(flagp) 1805 int *flagp; 1806 { 1807 1808 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1809 panic("nfs sndunlock"); 1810 *flagp &= ~NFSMNT_SNDLOCK; 1811 if (*flagp & NFSMNT_WANTSND) { 1812 *flagp &= ~NFSMNT_WANTSND; 1813 wakeup((caddr_t)flagp); 1814 } 1815 } 1816 1817 int 1818 nfs_rcvlock(rep) 1819 struct nfsreq *rep; 1820 { 1821 struct nfsmount *nmp = rep->r_nmp; 1822 int *flagp = &nmp->nm_iflag; 1823 int slpflag, slptimeo = 0; 1824 int error = 0; 1825 1826 if (*flagp & NFSMNT_DISMNT) 1827 return EIO; 1828 1829 if (*flagp & NFSMNT_INT) 1830 slpflag = PCATCH; 1831 else 1832 slpflag = 0; 1833 simple_lock(&nmp->nm_slock); 1834 while (*flagp & NFSMNT_RCVLOCK) { 1835 if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) { 1836 error = EINTR; 1837 goto quit; 1838 } 1839 *flagp |= NFSMNT_WANTRCV; 1840 nmp->nm_waiters++; 1841 (void) ltsleep(flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1842 slptimeo, &nmp->nm_slock); 1843 nmp->nm_waiters--; 1844 if (*flagp & NFSMNT_DISMNT) { 1845 wakeup(&nmp->nm_waiters); 1846 error = EIO; 1847 goto quit; 1848 } 1849 /* If our reply was received while we were sleeping, 1850 * then just return without taking the lock to avoid a 1851 * situation where a single iod could 'capture' the 1852 * receive lock. 1853 */ 1854 if (rep->r_mrep != NULL) { 1855 error = EALREADY; 1856 goto quit; 1857 } 1858 if (slpflag == PCATCH) { 1859 slpflag = 0; 1860 slptimeo = 2 * hz; 1861 } 1862 } 1863 *flagp |= NFSMNT_RCVLOCK; 1864 quit: 1865 simple_unlock(&nmp->nm_slock); 1866 return error; 1867 } 1868 1869 /* 1870 * Unlock the stream socket for others. 1871 */ 1872 void 1873 nfs_rcvunlock(nmp) 1874 struct nfsmount *nmp; 1875 { 1876 int *flagp = &nmp->nm_iflag; 1877 1878 simple_lock(&nmp->nm_slock); 1879 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1880 panic("nfs rcvunlock"); 1881 *flagp &= ~NFSMNT_RCVLOCK; 1882 if (*flagp & NFSMNT_WANTRCV) { 1883 *flagp &= ~NFSMNT_WANTRCV; 1884 wakeup((caddr_t)flagp); 1885 } 1886 simple_unlock(&nmp->nm_slock); 1887 } 1888 1889 /* 1890 * Parse an RPC request 1891 * - verify it 1892 * - fill in the cred struct. 1893 */ 1894 int 1895 nfs_getreq(nd, nfsd, has_header) 1896 struct nfsrv_descript *nd; 1897 struct nfsd *nfsd; 1898 int has_header; 1899 { 1900 int len, i; 1901 u_int32_t *tl; 1902 int32_t t1; 1903 struct uio uio; 1904 struct iovec iov; 1905 caddr_t dpos, cp2, cp; 1906 u_int32_t nfsvers, auth_type; 1907 uid_t nickuid; 1908 int error = 0, nqnfs = 0, ticklen; 1909 struct mbuf *mrep, *md; 1910 struct nfsuid *nuidp; 1911 struct timeval tvin, tvout; 1912 1913 mrep = nd->nd_mrep; 1914 md = nd->nd_md; 1915 dpos = nd->nd_dpos; 1916 if (has_header) { 1917 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1918 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1919 if (*tl++ != rpc_call) { 1920 m_freem(mrep); 1921 return (EBADRPC); 1922 } 1923 } else 1924 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1925 nd->nd_repstat = 0; 1926 nd->nd_flag = 0; 1927 if (*tl++ != rpc_vers) { 1928 nd->nd_repstat = ERPCMISMATCH; 1929 nd->nd_procnum = NFSPROC_NOOP; 1930 return (0); 1931 } 1932 if (*tl != nfs_prog) { 1933 if (*tl == nqnfs_prog) 1934 nqnfs++; 1935 else { 1936 nd->nd_repstat = EPROGUNAVAIL; 1937 nd->nd_procnum = NFSPROC_NOOP; 1938 return (0); 1939 } 1940 } 1941 tl++; 1942 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1943 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1944 (nfsvers != NQNFS_VER3 && nqnfs)) { 1945 nd->nd_repstat = EPROGMISMATCH; 1946 nd->nd_procnum = NFSPROC_NOOP; 1947 return (0); 1948 } 1949 if (nqnfs) 1950 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1951 else if (nfsvers == NFS_VER3) 1952 nd->nd_flag = ND_NFSV3; 1953 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1954 if (nd->nd_procnum == NFSPROC_NULL) 1955 return (0); 1956 if (nd->nd_procnum >= NFS_NPROCS || 1957 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1958 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1959 nd->nd_repstat = EPROCUNAVAIL; 1960 nd->nd_procnum = NFSPROC_NOOP; 1961 return (0); 1962 } 1963 if ((nd->nd_flag & ND_NFSV3) == 0) 1964 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1965 auth_type = *tl++; 1966 len = fxdr_unsigned(int, *tl++); 1967 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1968 m_freem(mrep); 1969 return (EBADRPC); 1970 } 1971 1972 nd->nd_flag &= ~ND_KERBAUTH; 1973 /* 1974 * Handle auth_unix or auth_kerb. 1975 */ 1976 if (auth_type == rpc_auth_unix) { 1977 len = fxdr_unsigned(int, *++tl); 1978 if (len < 0 || len > NFS_MAXNAMLEN) { 1979 m_freem(mrep); 1980 return (EBADRPC); 1981 } 1982 nfsm_adv(nfsm_rndup(len)); 1983 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1984 memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred)); 1985 nd->nd_cr.cr_ref = 1; 1986 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1987 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1988 len = fxdr_unsigned(int, *tl); 1989 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1990 m_freem(mrep); 1991 return (EBADRPC); 1992 } 1993 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1994 for (i = 0; i < len; i++) 1995 if (i < NGROUPS) 1996 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1997 else 1998 tl++; 1999 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 2000 if (nd->nd_cr.cr_ngroups > 1) 2001 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 2002 len = fxdr_unsigned(int, *++tl); 2003 if (len < 0 || len > RPCAUTH_MAXSIZ) { 2004 m_freem(mrep); 2005 return (EBADRPC); 2006 } 2007 if (len > 0) 2008 nfsm_adv(nfsm_rndup(len)); 2009 } else if (auth_type == rpc_auth_kerb) { 2010 switch (fxdr_unsigned(int, *tl++)) { 2011 case RPCAKN_FULLNAME: 2012 ticklen = fxdr_unsigned(int, *tl); 2013 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 2014 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 2015 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 2016 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 2017 m_freem(mrep); 2018 return (EBADRPC); 2019 } 2020 uio.uio_offset = 0; 2021 uio.uio_iov = &iov; 2022 uio.uio_iovcnt = 1; 2023 uio.uio_segflg = UIO_SYSSPACE; 2024 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 2025 iov.iov_len = RPCAUTH_MAXSIZ - 4; 2026 nfsm_mtouio(&uio, uio.uio_resid); 2027 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2028 if (*tl++ != rpc_auth_kerb || 2029 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 2030 printf("Bad kerb verifier\n"); 2031 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2032 nd->nd_procnum = NFSPROC_NOOP; 2033 return (0); 2034 } 2035 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 2036 tl = (u_int32_t *)cp; 2037 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 2038 printf("Not fullname kerb verifier\n"); 2039 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2040 nd->nd_procnum = NFSPROC_NOOP; 2041 return (0); 2042 } 2043 cp += NFSX_UNSIGNED; 2044 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 2045 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 2046 nd->nd_flag |= ND_KERBFULL; 2047 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 2048 break; 2049 case RPCAKN_NICKNAME: 2050 if (len != 2 * NFSX_UNSIGNED) { 2051 printf("Kerb nickname short\n"); 2052 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 2053 nd->nd_procnum = NFSPROC_NOOP; 2054 return (0); 2055 } 2056 nickuid = fxdr_unsigned(uid_t, *tl); 2057 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2058 if (*tl++ != rpc_auth_kerb || 2059 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 2060 printf("Kerb nick verifier bad\n"); 2061 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2062 nd->nd_procnum = NFSPROC_NOOP; 2063 return (0); 2064 } 2065 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2066 tvin.tv_sec = *tl++; 2067 tvin.tv_usec = *tl; 2068 2069 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), 2070 nu_hash) { 2071 if (nuidp->nu_cr.cr_uid == nickuid && 2072 (!nd->nd_nam2 || 2073 netaddr_match(NU_NETFAM(nuidp), 2074 &nuidp->nu_haddr, nd->nd_nam2))) 2075 break; 2076 } 2077 if (!nuidp) { 2078 nd->nd_repstat = 2079 (NFSERR_AUTHERR|AUTH_REJECTCRED); 2080 nd->nd_procnum = NFSPROC_NOOP; 2081 return (0); 2082 } 2083 2084 /* 2085 * Now, decrypt the timestamp using the session key 2086 * and validate it. 2087 */ 2088 #ifdef NFSKERB 2089 XXX 2090 #endif 2091 2092 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 2093 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 2094 if (nuidp->nu_expire < time.tv_sec || 2095 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 2096 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 2097 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 2098 nuidp->nu_expire = 0; 2099 nd->nd_repstat = 2100 (NFSERR_AUTHERR|AUTH_REJECTVERF); 2101 nd->nd_procnum = NFSPROC_NOOP; 2102 return (0); 2103 } 2104 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 2105 nd->nd_flag |= ND_KERBNICK; 2106 }; 2107 } else { 2108 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 2109 nd->nd_procnum = NFSPROC_NOOP; 2110 return (0); 2111 } 2112 2113 /* 2114 * For nqnfs, get piggybacked lease request. 2115 */ 2116 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 2117 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2118 nd->nd_flag |= fxdr_unsigned(int, *tl); 2119 if (nd->nd_flag & ND_LEASE) { 2120 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2121 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 2122 } else 2123 nd->nd_duration = NQ_MINLEASE; 2124 } else 2125 nd->nd_duration = NQ_MINLEASE; 2126 nd->nd_md = md; 2127 nd->nd_dpos = dpos; 2128 return (0); 2129 nfsmout: 2130 return (error); 2131 } 2132 2133 int 2134 nfs_msg(l, server, msg) 2135 struct lwp *l; 2136 const char *server, *msg; 2137 { 2138 tpr_t tpr; 2139 2140 if (l) 2141 tpr = tprintf_open(l->l_proc); 2142 else 2143 tpr = NULL; 2144 tprintf(tpr, "nfs server %s: %s\n", server, msg); 2145 tprintf_close(tpr); 2146 return (0); 2147 } 2148 2149 #ifdef NFSSERVER 2150 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 2151 struct nfssvc_sock *, struct lwp *, 2152 struct mbuf **)) = { 2153 nfsrv_null, 2154 nfsrv_getattr, 2155 nfsrv_setattr, 2156 nfsrv_lookup, 2157 nfsrv3_access, 2158 nfsrv_readlink, 2159 nfsrv_read, 2160 nfsrv_write, 2161 nfsrv_create, 2162 nfsrv_mkdir, 2163 nfsrv_symlink, 2164 nfsrv_mknod, 2165 nfsrv_remove, 2166 nfsrv_rmdir, 2167 nfsrv_rename, 2168 nfsrv_link, 2169 nfsrv_readdir, 2170 nfsrv_readdirplus, 2171 nfsrv_statfs, 2172 nfsrv_fsinfo, 2173 nfsrv_pathconf, 2174 nfsrv_commit, 2175 nqnfsrv_getlease, 2176 nqnfsrv_vacated, 2177 nfsrv_noop, 2178 nfsrv_noop 2179 }; 2180 2181 /* 2182 * Socket upcall routine for the nfsd sockets. 2183 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 2184 * Essentially do as much as possible non-blocking, else punt and it will 2185 * be called with M_WAIT from an nfsd. 2186 */ 2187 void 2188 nfsrv_rcv(so, arg, waitflag) 2189 struct socket *so; 2190 caddr_t arg; 2191 int waitflag; 2192 { 2193 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 2194 struct mbuf *m; 2195 struct mbuf *mp, *nam; 2196 struct uio auio; 2197 int flags, error; 2198 2199 if ((slp->ns_flag & SLP_VALID) == 0) 2200 return; 2201 #if 1 2202 /* 2203 * Define this to test for nfsds handling this under heavy load. 2204 */ 2205 if (waitflag == M_DONTWAIT) { 2206 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 2207 } 2208 #endif 2209 /* XXX: was NULL, soreceive() requires non-NULL uio->uio_lwp */ 2210 auio.uio_lwp = curlwp; /* XXX curlwp */ 2211 if (so->so_type == SOCK_STREAM) { 2212 /* 2213 * If there are already records on the queue, defer soreceive() 2214 * to an nfsd so that there is feedback to the TCP layer that 2215 * the nfs servers are heavily loaded. 2216 */ 2217 if (slp->ns_rec && waitflag == M_DONTWAIT) { 2218 slp->ns_flag |= SLP_NEEDQ; 2219 goto dorecs; 2220 } 2221 2222 /* 2223 * Do soreceive(). 2224 */ 2225 auio.uio_resid = 1000000000; 2226 flags = MSG_DONTWAIT; 2227 error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 2228 if (error || mp == (struct mbuf *)0) { 2229 if (error == EWOULDBLOCK) 2230 slp->ns_flag |= SLP_NEEDQ; 2231 else 2232 slp->ns_flag |= SLP_DISCONN; 2233 goto dorecs; 2234 } 2235 m = mp; 2236 if (slp->ns_rawend) { 2237 slp->ns_rawend->m_next = m; 2238 slp->ns_cc += 1000000000 - auio.uio_resid; 2239 } else { 2240 slp->ns_raw = m; 2241 slp->ns_cc = 1000000000 - auio.uio_resid; 2242 } 2243 while (m->m_next) 2244 m = m->m_next; 2245 slp->ns_rawend = m; 2246 2247 /* 2248 * Now try and parse record(s) out of the raw stream data. 2249 */ 2250 error = nfsrv_getstream(slp, waitflag); 2251 if (error) { 2252 if (error == EPERM) 2253 slp->ns_flag |= SLP_DISCONN; 2254 else 2255 slp->ns_flag |= SLP_NEEDQ; 2256 } 2257 } else { 2258 do { 2259 auio.uio_resid = 1000000000; 2260 flags = MSG_DONTWAIT; 2261 error = (*so->so_receive)(so, &nam, &auio, &mp, 2262 (struct mbuf **)0, &flags); 2263 if (mp) { 2264 if (nam) { 2265 m = nam; 2266 m->m_next = mp; 2267 } else 2268 m = mp; 2269 if (slp->ns_recend) 2270 slp->ns_recend->m_nextpkt = m; 2271 else 2272 slp->ns_rec = m; 2273 slp->ns_recend = m; 2274 m->m_nextpkt = (struct mbuf *)0; 2275 } 2276 if (error) { 2277 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2278 && error != EWOULDBLOCK) { 2279 slp->ns_flag |= SLP_DISCONN; 2280 goto dorecs; 2281 } 2282 } 2283 } while (mp); 2284 } 2285 2286 /* 2287 * Now try and process the request records, non-blocking. 2288 */ 2289 dorecs: 2290 if (waitflag == M_DONTWAIT && 2291 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 2292 nfsrv_wakenfsd(slp); 2293 } 2294 2295 /* 2296 * Try and extract an RPC request from the mbuf data list received on a 2297 * stream socket. The "waitflag" argument indicates whether or not it 2298 * can sleep. 2299 */ 2300 int 2301 nfsrv_getstream(slp, waitflag) 2302 struct nfssvc_sock *slp; 2303 int waitflag; 2304 { 2305 struct mbuf *m, **mpp; 2306 struct mbuf *recm; 2307 u_int32_t recmark; 2308 2309 if (slp->ns_flag & SLP_GETSTREAM) 2310 panic("nfs getstream"); 2311 slp->ns_flag |= SLP_GETSTREAM; 2312 for (;;) { 2313 if (slp->ns_reclen == 0) { 2314 if (slp->ns_cc < NFSX_UNSIGNED) { 2315 slp->ns_flag &= ~SLP_GETSTREAM; 2316 return (0); 2317 } 2318 m = slp->ns_raw; 2319 m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark); 2320 m_adj(m, NFSX_UNSIGNED); 2321 slp->ns_cc -= NFSX_UNSIGNED; 2322 recmark = ntohl(recmark); 2323 slp->ns_reclen = recmark & ~0x80000000; 2324 if (recmark & 0x80000000) 2325 slp->ns_flag |= SLP_LASTFRAG; 2326 else 2327 slp->ns_flag &= ~SLP_LASTFRAG; 2328 if (slp->ns_reclen > NFS_MAXPACKET) { 2329 slp->ns_flag &= ~SLP_GETSTREAM; 2330 return (EPERM); 2331 } 2332 } 2333 2334 /* 2335 * Now get the record part. 2336 * 2337 * Note that slp->ns_reclen may be 0. Linux sometimes 2338 * generates 0-length records. 2339 */ 2340 if (slp->ns_cc == slp->ns_reclen) { 2341 recm = slp->ns_raw; 2342 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2343 slp->ns_cc = slp->ns_reclen = 0; 2344 } else if (slp->ns_cc > slp->ns_reclen) { 2345 recm = slp->ns_raw; 2346 m = m_split(recm, slp->ns_reclen, waitflag); 2347 if (m == NULL) { 2348 slp->ns_flag &= ~SLP_GETSTREAM; 2349 return (EWOULDBLOCK); 2350 } 2351 m_claimm(recm, &nfs_mowner); 2352 slp->ns_raw = m; 2353 if (m->m_next == NULL) 2354 slp->ns_rawend = m; 2355 slp->ns_cc -= slp->ns_reclen; 2356 slp->ns_reclen = 0; 2357 } else { 2358 slp->ns_flag &= ~SLP_GETSTREAM; 2359 return (0); 2360 } 2361 2362 /* 2363 * Accumulate the fragments into a record. 2364 */ 2365 mpp = &slp->ns_frag; 2366 while (*mpp) 2367 mpp = &((*mpp)->m_next); 2368 *mpp = recm; 2369 if (slp->ns_flag & SLP_LASTFRAG) { 2370 if (slp->ns_recend) 2371 slp->ns_recend->m_nextpkt = slp->ns_frag; 2372 else 2373 slp->ns_rec = slp->ns_frag; 2374 slp->ns_recend = slp->ns_frag; 2375 slp->ns_frag = (struct mbuf *)0; 2376 } 2377 } 2378 } 2379 2380 /* 2381 * Parse an RPC header. 2382 */ 2383 int 2384 nfsrv_dorec(slp, nfsd, ndp) 2385 struct nfssvc_sock *slp; 2386 struct nfsd *nfsd; 2387 struct nfsrv_descript **ndp; 2388 { 2389 struct mbuf *m, *nam; 2390 struct nfsrv_descript *nd; 2391 int error; 2392 2393 *ndp = NULL; 2394 if ((slp->ns_flag & SLP_VALID) == 0 || 2395 (m = slp->ns_rec) == (struct mbuf *)0) 2396 return (ENOBUFS); 2397 slp->ns_rec = m->m_nextpkt; 2398 if (slp->ns_rec) 2399 m->m_nextpkt = (struct mbuf *)0; 2400 else 2401 slp->ns_recend = (struct mbuf *)0; 2402 if (m->m_type == MT_SONAME) { 2403 nam = m; 2404 m = m->m_next; 2405 nam->m_next = NULL; 2406 } else 2407 nam = NULL; 2408 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK); 2409 nd->nd_md = nd->nd_mrep = m; 2410 nd->nd_nam2 = nam; 2411 nd->nd_dpos = mtod(m, caddr_t); 2412 error = nfs_getreq(nd, nfsd, TRUE); 2413 if (error) { 2414 m_freem(nam); 2415 pool_put(&nfs_srvdesc_pool, nd); 2416 return (error); 2417 } 2418 *ndp = nd; 2419 nfsd->nfsd_nd = nd; 2420 return (0); 2421 } 2422 2423 2424 /* 2425 * Search for a sleeping nfsd and wake it up. 2426 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2427 * running nfsds will go look for the work in the nfssvc_sock list. 2428 */ 2429 void 2430 nfsrv_wakenfsd(slp) 2431 struct nfssvc_sock *slp; 2432 { 2433 struct nfsd *nd; 2434 2435 if ((slp->ns_flag & SLP_VALID) == 0) 2436 return; 2437 simple_lock(&nfsd_slock); 2438 if (slp->ns_flag & SLP_DOREC) { 2439 simple_unlock(&nfsd_slock); 2440 return; 2441 } 2442 nd = SLIST_FIRST(&nfsd_idle_head); 2443 if (nd) { 2444 SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle); 2445 simple_unlock(&nfsd_slock); 2446 2447 KASSERT(nd->nfsd_flag & NFSD_WAITING); 2448 nd->nfsd_flag &= ~NFSD_WAITING; 2449 if (nd->nfsd_slp) 2450 panic("nfsd wakeup"); 2451 slp->ns_sref++; 2452 nd->nfsd_slp = slp; 2453 wakeup(nd); 2454 return; 2455 } 2456 slp->ns_flag |= SLP_DOREC; 2457 nfsd_head_flag |= NFSD_CHECKSLP; 2458 TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending); 2459 simple_unlock(&nfsd_slock); 2460 } 2461 #endif /* NFSSERVER */ 2462
Indexes created Tue Sep 30 11:09:46 GMT 2025