nfs_socket.c revision 1.92.2.7
1 /* $NetBSD: nfs_socket.c,v 1.92.2.7 2004/09/21 13:38:38 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 1.92.2.7 2004/09/21 13:38:38 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 if (error == NFSERR_ACCES && retry_cred) { 1218 m_freem(mrep); 1219 m_freem(rep->r_mreq); 1220 FREE(rep, M_NFSREQ); 1221 use_opencred = !use_opencred; 1222 if (mrest_backup == NULL) 1223 return ENOMEM; /* m_copym failure */ 1224 mrest = mrest_backup; 1225 mrest_backup = NULL; 1226 cred = origcred; 1227 error = 0; 1228 retry_cred = FALSE; 1229 goto tryagain_cred; 1230 } 1231 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1232 error == NFSERR_TRYLATER) { 1233 m_freem(mrep); 1234 error = 0; 1235 waituntil = time.tv_sec + trylater_delay; 1236 while (time.tv_sec < waituntil) 1237 (void) tsleep((caddr_t)&lbolt, 1238 PSOCK, "nqnfstry", 0); 1239 trylater_delay *= NFS_TRYLATERDELMUL; 1240 if (trylater_delay > NFS_TRYLATERDELMAX) 1241 trylater_delay = NFS_TRYLATERDELMAX; 1242 /* 1243 * RFC1813: 1244 * The client should wait and then try 1245 * the request with a new RPC transaction ID. 1246 */ 1247 nfs_renewxid(rep); 1248 goto tryagain; 1249 } 1250 1251 /* 1252 * If the File Handle was stale, invalidate the 1253 * lookup cache, just in case. 1254 */ 1255 if (error == ESTALE) 1256 cache_purge(NFSTOV(np)); 1257 if (nmp->nm_flag & NFSMNT_NFSV3) { 1258 *mrp = mrep; 1259 *mdp = md; 1260 *dposp = dpos; 1261 error |= NFSERR_RETERR; 1262 } else 1263 m_freem(mrep); 1264 goto nfsmout; 1265 } 1266 1267 /* 1268 * note which credential worked to minimize number of retries. 1269 */ 1270 if (use_opencred) 1271 np->n_flag |= NUSEOPENCRED; 1272 else 1273 np->n_flag &= ~NUSEOPENCRED; 1274 1275 #ifndef NFS_V2_ONLY 1276 /* 1277 * For nqnfs, get any lease in reply 1278 */ 1279 if (nmp->nm_flag & NFSMNT_NQNFS) { 1280 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1281 if (*tl) { 1282 nqlflag = fxdr_unsigned(int, *tl); 1283 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1284 cachable = fxdr_unsigned(int, *tl++); 1285 reqtime += fxdr_unsigned(int, *tl++); 1286 if (reqtime > time.tv_sec) { 1287 frev = fxdr_hyper(tl); 1288 nqnfs_clientlease(nmp, np, nqlflag, 1289 cachable, reqtime, frev); 1290 } 1291 } 1292 } 1293 #endif 1294 *mrp = mrep; 1295 *mdp = md; 1296 *dposp = dpos; 1297 1298 KASSERT(error == 0); 1299 goto nfsmout; 1300 } 1301 m_freem(mrep); 1302 error = EPROTONOSUPPORT; 1303 nfsmout: 1304 m_freem(rep->r_mreq); 1305 free((caddr_t)rep, M_NFSREQ); 1306 m_freem(mrest_backup); 1307 return (error); 1308 } 1309 #endif /* NFS */ 1310 1311 /* 1312 * Generate the rpc reply header 1313 * siz arg. is used to decide if adding a cluster is worthwhile 1314 */ 1315 int 1316 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1317 int siz; 1318 struct nfsrv_descript *nd; 1319 struct nfssvc_sock *slp; 1320 int err; 1321 int cache; 1322 u_quad_t *frev; 1323 struct mbuf **mrq; 1324 struct mbuf **mbp; 1325 caddr_t *bposp; 1326 { 1327 u_int32_t *tl; 1328 struct mbuf *mreq; 1329 caddr_t bpos; 1330 struct mbuf *mb; 1331 1332 mreq = m_gethdr(M_WAIT, MT_DATA); 1333 MCLAIM(mreq, &nfs_mowner); 1334 mb = mreq; 1335 /* 1336 * If this is a big reply, use a cluster else 1337 * try and leave leading space for the lower level headers. 1338 */ 1339 siz += RPC_REPLYSIZ; 1340 if (siz >= max_datalen) { 1341 m_clget(mreq, M_WAIT); 1342 } else 1343 mreq->m_data += max_hdr; 1344 tl = mtod(mreq, u_int32_t *); 1345 mreq->m_len = 6 * NFSX_UNSIGNED; 1346 bpos = ((caddr_t)tl) + mreq->m_len; 1347 *tl++ = txdr_unsigned(nd->nd_retxid); 1348 *tl++ = rpc_reply; 1349 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1350 *tl++ = rpc_msgdenied; 1351 if (err & NFSERR_AUTHERR) { 1352 *tl++ = rpc_autherr; 1353 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1354 mreq->m_len -= NFSX_UNSIGNED; 1355 bpos -= NFSX_UNSIGNED; 1356 } else { 1357 *tl++ = rpc_mismatch; 1358 *tl++ = txdr_unsigned(RPC_VER2); 1359 *tl = txdr_unsigned(RPC_VER2); 1360 } 1361 } else { 1362 *tl++ = rpc_msgaccepted; 1363 1364 /* 1365 * For Kerberos authentication, we must send the nickname 1366 * verifier back, otherwise just RPCAUTH_NULL. 1367 */ 1368 if (nd->nd_flag & ND_KERBFULL) { 1369 struct nfsuid *nuidp; 1370 struct timeval ktvin, ktvout; 1371 1372 LIST_FOREACH(nuidp, NUIDHASH(slp, nd->nd_cr.cr_uid), 1373 nu_hash) { 1374 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1375 (!nd->nd_nam2 || netaddr_match( 1376 NU_NETFAM(nuidp), &nuidp->nu_haddr, 1377 nd->nd_nam2))) 1378 break; 1379 } 1380 if (nuidp) { 1381 ktvin.tv_sec = 1382 txdr_unsigned(nuidp->nu_timestamp.tv_sec 1383 - 1); 1384 ktvin.tv_usec = 1385 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1386 1387 /* 1388 * Encrypt the timestamp in ecb mode using the 1389 * session key. 1390 */ 1391 #ifdef NFSKERB 1392 XXX 1393 #endif 1394 1395 *tl++ = rpc_auth_kerb; 1396 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1397 *tl = ktvout.tv_sec; 1398 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1399 *tl++ = ktvout.tv_usec; 1400 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1401 } else { 1402 *tl++ = 0; 1403 *tl++ = 0; 1404 } 1405 } else { 1406 *tl++ = 0; 1407 *tl++ = 0; 1408 } 1409 switch (err) { 1410 case EPROGUNAVAIL: 1411 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1412 break; 1413 case EPROGMISMATCH: 1414 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1415 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1416 if (nd->nd_flag & ND_NQNFS) { 1417 *tl++ = txdr_unsigned(3); 1418 *tl = txdr_unsigned(3); 1419 } else { 1420 *tl++ = txdr_unsigned(2); 1421 *tl = txdr_unsigned(3); 1422 } 1423 break; 1424 case EPROCUNAVAIL: 1425 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1426 break; 1427 case EBADRPC: 1428 *tl = txdr_unsigned(RPC_GARBAGE); 1429 break; 1430 default: 1431 *tl = 0; 1432 if (err != NFSERR_RETVOID) { 1433 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1434 if (err) 1435 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1436 else 1437 *tl = 0; 1438 } 1439 break; 1440 }; 1441 } 1442 1443 /* 1444 * For nqnfs, piggyback lease as requested. 1445 */ 1446 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1447 if (nd->nd_flag & ND_LEASE) { 1448 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1449 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1450 *tl++ = txdr_unsigned(cache); 1451 *tl++ = txdr_unsigned(nd->nd_duration); 1452 txdr_hyper(*frev, tl); 1453 } else { 1454 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1455 *tl = 0; 1456 } 1457 } 1458 if (mrq != NULL) 1459 *mrq = mreq; 1460 *mbp = mb; 1461 *bposp = bpos; 1462 if (err != 0 && err != NFSERR_RETVOID) 1463 nfsstats.srvrpc_errs++; 1464 return (0); 1465 } 1466 1467 /* 1468 * Nfs timer routine 1469 * Scan the nfsreq list and retranmit any requests that have timed out 1470 * To avoid retransmission attempts on STREAM sockets (in the future) make 1471 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1472 */ 1473 void 1474 nfs_timer(arg) 1475 void *arg; /* never used */ 1476 { 1477 struct nfsreq *rep; 1478 struct mbuf *m; 1479 struct socket *so; 1480 struct nfsmount *nmp; 1481 int timeo; 1482 int s, error; 1483 #ifdef NFSSERVER 1484 struct nfssvc_sock *slp; 1485 static long lasttime = 0; 1486 u_quad_t cur_usec; 1487 #endif 1488 1489 s = splsoftnet(); 1490 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1491 nmp = rep->r_nmp; 1492 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1493 continue; 1494 if (nfs_sigintr(nmp, rep, rep->r_lwp)) { 1495 rep->r_flags |= R_SOFTTERM; 1496 continue; 1497 } 1498 if (rep->r_rtt >= 0) { 1499 rep->r_rtt++; 1500 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1501 timeo = nmp->nm_timeo; 1502 else 1503 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1504 if (nmp->nm_timeouts > 0) 1505 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1506 if (rep->r_rtt <= timeo) 1507 continue; 1508 if (nmp->nm_timeouts < 1509 (sizeof(nfs_backoff) / sizeof(nfs_backoff[0]))) 1510 nmp->nm_timeouts++; 1511 } 1512 /* 1513 * Check for server not responding 1514 */ 1515 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1516 rep->r_rexmit > nmp->nm_deadthresh) { 1517 nfs_msg(rep->r_lwp, 1518 nmp->nm_mountp->mnt_stat.f_mntfromname, 1519 "not responding"); 1520 rep->r_flags |= R_TPRINTFMSG; 1521 } 1522 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1523 nfsstats.rpctimeouts++; 1524 rep->r_flags |= R_SOFTTERM; 1525 continue; 1526 } 1527 if (nmp->nm_sotype != SOCK_DGRAM) { 1528 if (++rep->r_rexmit > NFS_MAXREXMIT) 1529 rep->r_rexmit = NFS_MAXREXMIT; 1530 continue; 1531 } 1532 if ((so = nmp->nm_so) == NULL) 1533 continue; 1534 1535 /* 1536 * If there is enough space and the window allows.. 1537 * Resend it 1538 * Set r_rtt to -1 in case we fail to send it now. 1539 */ 1540 rep->r_rtt = -1; 1541 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1542 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1543 (rep->r_flags & R_SENT) || 1544 nmp->nm_sent < nmp->nm_cwnd) && 1545 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1546 if (so->so_state & SS_ISCONNECTED) 1547 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1548 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1549 else 1550 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1551 nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0); 1552 if (error) { 1553 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1554 #ifdef DEBUG 1555 printf("nfs_timer: ignoring error %d\n", 1556 error); 1557 #endif 1558 so->so_error = 0; 1559 } 1560 } else { 1561 /* 1562 * Iff first send, start timing 1563 * else turn timing off, backoff timer 1564 * and divide congestion window by 2. 1565 */ 1566 if (rep->r_flags & R_SENT) { 1567 rep->r_flags &= ~R_TIMING; 1568 if (++rep->r_rexmit > NFS_MAXREXMIT) 1569 rep->r_rexmit = NFS_MAXREXMIT; 1570 nmp->nm_cwnd >>= 1; 1571 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1572 nmp->nm_cwnd = NFS_CWNDSCALE; 1573 nfsstats.rpcretries++; 1574 } else { 1575 rep->r_flags |= R_SENT; 1576 nmp->nm_sent += NFS_CWNDSCALE; 1577 } 1578 rep->r_rtt = 0; 1579 } 1580 } 1581 } 1582 1583 #ifdef NFSSERVER 1584 /* 1585 * Call the nqnfs server timer once a second to handle leases. 1586 */ 1587 if (lasttime != time.tv_sec) { 1588 lasttime = time.tv_sec; 1589 nqnfs_serverd(); 1590 } 1591 1592 /* 1593 * Scan the write gathering queues for writes that need to be 1594 * completed now. 1595 */ 1596 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1597 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1598 if (LIST_FIRST(&slp->ns_tq) && 1599 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) 1600 nfsrv_wakenfsd(slp); 1601 } 1602 #endif /* NFSSERVER */ 1603 splx(s); 1604 callout_schedule(&nfs_timer_ch, nfs_ticks); 1605 } 1606 1607 /*ARGSUSED*/ 1608 void 1609 nfs_exit(p, v) 1610 struct proc *p; 1611 void *v; 1612 { 1613 struct nfsreq *rp; 1614 int s = splsoftnet(); 1615 1616 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 1617 if (rp->r_lwp->l_proc == p) 1618 TAILQ_REMOVE(&nfs_reqq, rp, r_chain); 1619 } 1620 splx(s); 1621 } 1622 1623 /* 1624 * Test for a termination condition pending on the process. 1625 * This is used for NFSMNT_INT mounts. 1626 */ 1627 int 1628 nfs_sigintr(nmp, rep, l) 1629 struct nfsmount *nmp; 1630 struct nfsreq *rep; 1631 struct lwp *l; 1632 { 1633 sigset_t ss; 1634 1635 if (rep && (rep->r_flags & R_SOFTTERM)) 1636 return (EINTR); 1637 if (!(nmp->nm_flag & NFSMNT_INT)) 1638 return (0); 1639 if (l) { 1640 sigpending1(l->l_proc, &ss); 1641 #if 0 1642 sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss); 1643 #endif 1644 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1645 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1646 sigismember(&ss, SIGQUIT)) 1647 return (EINTR); 1648 } 1649 return (0); 1650 } 1651 1652 /* 1653 * Lock a socket against others. 1654 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1655 * and also to avoid race conditions between the processes with nfs requests 1656 * in progress when a reconnect is necessary. 1657 */ 1658 int 1659 nfs_sndlock(flagp, rep) 1660 int *flagp; 1661 struct nfsreq *rep; 1662 { 1663 struct lwp *l; 1664 int slpflag = 0, slptimeo = 0; 1665 1666 if (rep) { 1667 l = rep->r_lwp; 1668 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1669 slpflag = PCATCH; 1670 } else 1671 l = (struct lwp *)0; 1672 while (*flagp & NFSMNT_SNDLOCK) { 1673 if (rep && nfs_sigintr(rep->r_nmp, rep, l)) 1674 return (EINTR); 1675 *flagp |= NFSMNT_WANTSND; 1676 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1677 slptimeo); 1678 if (slpflag == PCATCH) { 1679 slpflag = 0; 1680 slptimeo = 2 * hz; 1681 } 1682 } 1683 *flagp |= NFSMNT_SNDLOCK; 1684 return (0); 1685 } 1686 1687 /* 1688 * Unlock the stream socket for others. 1689 */ 1690 void 1691 nfs_sndunlock(flagp) 1692 int *flagp; 1693 { 1694 1695 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1696 panic("nfs sndunlock"); 1697 *flagp &= ~NFSMNT_SNDLOCK; 1698 if (*flagp & NFSMNT_WANTSND) { 1699 *flagp &= ~NFSMNT_WANTSND; 1700 wakeup((caddr_t)flagp); 1701 } 1702 } 1703 1704 int 1705 nfs_rcvlock(rep) 1706 struct nfsreq *rep; 1707 { 1708 struct nfsmount *nmp = rep->r_nmp; 1709 int *flagp = &nmp->nm_iflag; 1710 int slpflag, slptimeo = 0; 1711 int error = 0; 1712 1713 if (*flagp & NFSMNT_DISMNT) 1714 return EIO; 1715 1716 if (*flagp & NFSMNT_INT) 1717 slpflag = PCATCH; 1718 else 1719 slpflag = 0; 1720 simple_lock(&nmp->nm_slock); 1721 while (*flagp & NFSMNT_RCVLOCK) { 1722 if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) { 1723 error = EINTR; 1724 goto quit; 1725 } 1726 *flagp |= NFSMNT_WANTRCV; 1727 nmp->nm_waiters++; 1728 (void) ltsleep(flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1729 slptimeo, &nmp->nm_slock); 1730 nmp->nm_waiters--; 1731 if (*flagp & NFSMNT_DISMNT) { 1732 wakeup(&nmp->nm_waiters); 1733 error = EIO; 1734 goto quit; 1735 } 1736 /* If our reply was received while we were sleeping, 1737 * then just return without taking the lock to avoid a 1738 * situation where a single iod could 'capture' the 1739 * receive lock. 1740 */ 1741 if (rep->r_mrep != NULL) { 1742 error = EALREADY; 1743 goto quit; 1744 } 1745 if (slpflag == PCATCH) { 1746 slpflag = 0; 1747 slptimeo = 2 * hz; 1748 } 1749 } 1750 *flagp |= NFSMNT_RCVLOCK; 1751 quit: 1752 simple_unlock(&nmp->nm_slock); 1753 return error; 1754 } 1755 1756 /* 1757 * Unlock the stream socket for others. 1758 */ 1759 void 1760 nfs_rcvunlock(nmp) 1761 struct nfsmount *nmp; 1762 { 1763 int *flagp = &nmp->nm_iflag; 1764 1765 simple_lock(&nmp->nm_slock); 1766 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1767 panic("nfs rcvunlock"); 1768 *flagp &= ~NFSMNT_RCVLOCK; 1769 if (*flagp & NFSMNT_WANTRCV) { 1770 *flagp &= ~NFSMNT_WANTRCV; 1771 wakeup((caddr_t)flagp); 1772 } 1773 simple_unlock(&nmp->nm_slock); 1774 } 1775 1776 /* 1777 * Parse an RPC request 1778 * - verify it 1779 * - fill in the cred struct. 1780 */ 1781 int 1782 nfs_getreq(nd, nfsd, has_header) 1783 struct nfsrv_descript *nd; 1784 struct nfsd *nfsd; 1785 int has_header; 1786 { 1787 int len, i; 1788 u_int32_t *tl; 1789 int32_t t1; 1790 struct uio uio; 1791 struct iovec iov; 1792 caddr_t dpos, cp2, cp; 1793 u_int32_t nfsvers, auth_type; 1794 uid_t nickuid; 1795 int error = 0, nqnfs = 0, ticklen; 1796 struct mbuf *mrep, *md; 1797 struct nfsuid *nuidp; 1798 struct timeval tvin, tvout; 1799 1800 mrep = nd->nd_mrep; 1801 md = nd->nd_md; 1802 dpos = nd->nd_dpos; 1803 if (has_header) { 1804 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1805 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1806 if (*tl++ != rpc_call) { 1807 m_freem(mrep); 1808 return (EBADRPC); 1809 } 1810 } else 1811 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1812 nd->nd_repstat = 0; 1813 nd->nd_flag = 0; 1814 if (*tl++ != rpc_vers) { 1815 nd->nd_repstat = ERPCMISMATCH; 1816 nd->nd_procnum = NFSPROC_NOOP; 1817 return (0); 1818 } 1819 if (*tl != nfs_prog) { 1820 if (*tl == nqnfs_prog) 1821 nqnfs++; 1822 else { 1823 nd->nd_repstat = EPROGUNAVAIL; 1824 nd->nd_procnum = NFSPROC_NOOP; 1825 return (0); 1826 } 1827 } 1828 tl++; 1829 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1830 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1831 (nfsvers != NQNFS_VER3 && nqnfs)) { 1832 nd->nd_repstat = EPROGMISMATCH; 1833 nd->nd_procnum = NFSPROC_NOOP; 1834 return (0); 1835 } 1836 if (nqnfs) 1837 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1838 else if (nfsvers == NFS_VER3) 1839 nd->nd_flag = ND_NFSV3; 1840 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1841 if (nd->nd_procnum == NFSPROC_NULL) 1842 return (0); 1843 if (nd->nd_procnum >= NFS_NPROCS || 1844 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1845 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1846 nd->nd_repstat = EPROCUNAVAIL; 1847 nd->nd_procnum = NFSPROC_NOOP; 1848 return (0); 1849 } 1850 if ((nd->nd_flag & ND_NFSV3) == 0) 1851 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1852 auth_type = *tl++; 1853 len = fxdr_unsigned(int, *tl++); 1854 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1855 m_freem(mrep); 1856 return (EBADRPC); 1857 } 1858 1859 nd->nd_flag &= ~ND_KERBAUTH; 1860 /* 1861 * Handle auth_unix or auth_kerb. 1862 */ 1863 if (auth_type == rpc_auth_unix) { 1864 len = fxdr_unsigned(int, *++tl); 1865 if (len < 0 || len > NFS_MAXNAMLEN) { 1866 m_freem(mrep); 1867 return (EBADRPC); 1868 } 1869 nfsm_adv(nfsm_rndup(len)); 1870 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1871 memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred)); 1872 nd->nd_cr.cr_ref = 1; 1873 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1874 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1875 len = fxdr_unsigned(int, *tl); 1876 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1877 m_freem(mrep); 1878 return (EBADRPC); 1879 } 1880 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1881 for (i = 0; i < len; i++) 1882 if (i < NGROUPS) 1883 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1884 else 1885 tl++; 1886 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1887 if (nd->nd_cr.cr_ngroups > 1) 1888 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1889 len = fxdr_unsigned(int, *++tl); 1890 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1891 m_freem(mrep); 1892 return (EBADRPC); 1893 } 1894 if (len > 0) 1895 nfsm_adv(nfsm_rndup(len)); 1896 } else if (auth_type == rpc_auth_kerb) { 1897 switch (fxdr_unsigned(int, *tl++)) { 1898 case RPCAKN_FULLNAME: 1899 ticklen = fxdr_unsigned(int, *tl); 1900 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1901 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1902 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1903 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1904 m_freem(mrep); 1905 return (EBADRPC); 1906 } 1907 uio.uio_offset = 0; 1908 uio.uio_iov = &iov; 1909 uio.uio_iovcnt = 1; 1910 uio.uio_segflg = UIO_SYSSPACE; 1911 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1912 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1913 nfsm_mtouio(&uio, uio.uio_resid); 1914 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1915 if (*tl++ != rpc_auth_kerb || 1916 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1917 printf("Bad kerb verifier\n"); 1918 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1919 nd->nd_procnum = NFSPROC_NOOP; 1920 return (0); 1921 } 1922 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1923 tl = (u_int32_t *)cp; 1924 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1925 printf("Not fullname kerb verifier\n"); 1926 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1927 nd->nd_procnum = NFSPROC_NOOP; 1928 return (0); 1929 } 1930 cp += NFSX_UNSIGNED; 1931 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 1932 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1933 nd->nd_flag |= ND_KERBFULL; 1934 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1935 break; 1936 case RPCAKN_NICKNAME: 1937 if (len != 2 * NFSX_UNSIGNED) { 1938 printf("Kerb nickname short\n"); 1939 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1940 nd->nd_procnum = NFSPROC_NOOP; 1941 return (0); 1942 } 1943 nickuid = fxdr_unsigned(uid_t, *tl); 1944 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1945 if (*tl++ != rpc_auth_kerb || 1946 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1947 printf("Kerb nick verifier bad\n"); 1948 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1949 nd->nd_procnum = NFSPROC_NOOP; 1950 return (0); 1951 } 1952 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1953 tvin.tv_sec = *tl++; 1954 tvin.tv_usec = *tl; 1955 1956 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), 1957 nu_hash) { 1958 if (nuidp->nu_cr.cr_uid == nickuid && 1959 (!nd->nd_nam2 || 1960 netaddr_match(NU_NETFAM(nuidp), 1961 &nuidp->nu_haddr, nd->nd_nam2))) 1962 break; 1963 } 1964 if (!nuidp) { 1965 nd->nd_repstat = 1966 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1967 nd->nd_procnum = NFSPROC_NOOP; 1968 return (0); 1969 } 1970 1971 /* 1972 * Now, decrypt the timestamp using the session key 1973 * and validate it. 1974 */ 1975 #ifdef NFSKERB 1976 XXX 1977 #endif 1978 1979 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1980 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1981 if (nuidp->nu_expire < time.tv_sec || 1982 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1983 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1984 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1985 nuidp->nu_expire = 0; 1986 nd->nd_repstat = 1987 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1988 nd->nd_procnum = NFSPROC_NOOP; 1989 return (0); 1990 } 1991 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1992 nd->nd_flag |= ND_KERBNICK; 1993 }; 1994 } else { 1995 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1996 nd->nd_procnum = NFSPROC_NOOP; 1997 return (0); 1998 } 1999 2000 /* 2001 * For nqnfs, get piggybacked lease request. 2002 */ 2003 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 2004 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2005 nd->nd_flag |= fxdr_unsigned(int, *tl); 2006 if (nd->nd_flag & ND_LEASE) { 2007 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2008 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 2009 } else 2010 nd->nd_duration = NQ_MINLEASE; 2011 } else 2012 nd->nd_duration = NQ_MINLEASE; 2013 nd->nd_md = md; 2014 nd->nd_dpos = dpos; 2015 return (0); 2016 nfsmout: 2017 return (error); 2018 } 2019 2020 int 2021 nfs_msg(l, server, msg) 2022 struct lwp *l; 2023 char *server, *msg; 2024 { 2025 tpr_t tpr; 2026 2027 if (l) 2028 tpr = tprintf_open(l->l_proc); 2029 else 2030 tpr = NULL; 2031 tprintf(tpr, "nfs server %s: %s\n", server, msg); 2032 tprintf_close(tpr); 2033 return (0); 2034 } 2035 2036 #ifdef NFSSERVER 2037 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 2038 struct nfssvc_sock *, struct lwp *, 2039 struct mbuf **)) = { 2040 nfsrv_null, 2041 nfsrv_getattr, 2042 nfsrv_setattr, 2043 nfsrv_lookup, 2044 nfsrv3_access, 2045 nfsrv_readlink, 2046 nfsrv_read, 2047 nfsrv_write, 2048 nfsrv_create, 2049 nfsrv_mkdir, 2050 nfsrv_symlink, 2051 nfsrv_mknod, 2052 nfsrv_remove, 2053 nfsrv_rmdir, 2054 nfsrv_rename, 2055 nfsrv_link, 2056 nfsrv_readdir, 2057 nfsrv_readdirplus, 2058 nfsrv_statfs, 2059 nfsrv_fsinfo, 2060 nfsrv_pathconf, 2061 nfsrv_commit, 2062 nqnfsrv_getlease, 2063 nqnfsrv_vacated, 2064 nfsrv_noop, 2065 nfsrv_noop 2066 }; 2067 2068 /* 2069 * Socket upcall routine for the nfsd sockets. 2070 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 2071 * Essentially do as much as possible non-blocking, else punt and it will 2072 * be called with M_WAIT from an nfsd. 2073 */ 2074 void 2075 nfsrv_rcv(so, arg, waitflag) 2076 struct socket *so; 2077 caddr_t arg; 2078 int waitflag; 2079 { 2080 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 2081 struct mbuf *m; 2082 struct mbuf *mp, *nam; 2083 struct uio auio; 2084 int flags, error; 2085 2086 if ((slp->ns_flag & SLP_VALID) == 0) 2087 return; 2088 #ifdef notdef 2089 /* 2090 * Define this to test for nfsds handling this under heavy load. 2091 */ 2092 if (waitflag == M_DONTWAIT) { 2093 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 2094 } 2095 #endif 2096 /* XXX: was NULL, soreceive() requires non-NULL uio->uio_lwp */ 2097 auio.uio_lwp = curlwp; /* XXX curlwp */ 2098 if (so->so_type == SOCK_STREAM) { 2099 /* 2100 * If there are already records on the queue, defer soreceive() 2101 * to an nfsd so that there is feedback to the TCP layer that 2102 * the nfs servers are heavily loaded. 2103 */ 2104 if (slp->ns_rec && waitflag == M_DONTWAIT) { 2105 slp->ns_flag |= SLP_NEEDQ; 2106 goto dorecs; 2107 } 2108 2109 /* 2110 * Do soreceive(). 2111 */ 2112 auio.uio_resid = 1000000000; 2113 flags = MSG_DONTWAIT; 2114 error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 2115 if (error || mp == (struct mbuf *)0) { 2116 if (error == EWOULDBLOCK) 2117 slp->ns_flag |= SLP_NEEDQ; 2118 else 2119 slp->ns_flag |= SLP_DISCONN; 2120 goto dorecs; 2121 } 2122 m = mp; 2123 if (slp->ns_rawend) { 2124 slp->ns_rawend->m_next = m; 2125 slp->ns_cc += 1000000000 - auio.uio_resid; 2126 } else { 2127 slp->ns_raw = m; 2128 slp->ns_cc = 1000000000 - auio.uio_resid; 2129 } 2130 while (m->m_next) 2131 m = m->m_next; 2132 slp->ns_rawend = m; 2133 2134 /* 2135 * Now try and parse record(s) out of the raw stream data. 2136 */ 2137 error = nfsrv_getstream(slp, waitflag); 2138 if (error) { 2139 if (error == EPERM) 2140 slp->ns_flag |= SLP_DISCONN; 2141 else 2142 slp->ns_flag |= SLP_NEEDQ; 2143 } 2144 } else { 2145 do { 2146 auio.uio_resid = 1000000000; 2147 flags = MSG_DONTWAIT; 2148 error = (*so->so_receive)(so, &nam, &auio, &mp, 2149 (struct mbuf **)0, &flags); 2150 if (mp) { 2151 if (nam) { 2152 m = nam; 2153 m->m_next = mp; 2154 } else 2155 m = mp; 2156 if (slp->ns_recend) 2157 slp->ns_recend->m_nextpkt = m; 2158 else 2159 slp->ns_rec = m; 2160 slp->ns_recend = m; 2161 m->m_nextpkt = (struct mbuf *)0; 2162 } 2163 if (error) { 2164 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2165 && error != EWOULDBLOCK) { 2166 slp->ns_flag |= SLP_DISCONN; 2167 goto dorecs; 2168 } 2169 } 2170 } while (mp); 2171 } 2172 2173 /* 2174 * Now try and process the request records, non-blocking. 2175 */ 2176 dorecs: 2177 if (waitflag == M_DONTWAIT && 2178 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 2179 nfsrv_wakenfsd(slp); 2180 } 2181 2182 /* 2183 * Try and extract an RPC request from the mbuf data list received on a 2184 * stream socket. The "waitflag" argument indicates whether or not it 2185 * can sleep. 2186 */ 2187 int 2188 nfsrv_getstream(slp, waitflag) 2189 struct nfssvc_sock *slp; 2190 int waitflag; 2191 { 2192 struct mbuf *m, **mpp; 2193 struct mbuf *recm; 2194 u_int32_t recmark; 2195 2196 if (slp->ns_flag & SLP_GETSTREAM) 2197 panic("nfs getstream"); 2198 slp->ns_flag |= SLP_GETSTREAM; 2199 for (;;) { 2200 if (slp->ns_reclen == 0) { 2201 if (slp->ns_cc < NFSX_UNSIGNED) { 2202 slp->ns_flag &= ~SLP_GETSTREAM; 2203 return (0); 2204 } 2205 m = slp->ns_raw; 2206 m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark); 2207 m_adj(m, NFSX_UNSIGNED); 2208 slp->ns_cc -= NFSX_UNSIGNED; 2209 recmark = ntohl(recmark); 2210 slp->ns_reclen = recmark & ~0x80000000; 2211 if (recmark & 0x80000000) 2212 slp->ns_flag |= SLP_LASTFRAG; 2213 else 2214 slp->ns_flag &= ~SLP_LASTFRAG; 2215 if (slp->ns_reclen > NFS_MAXPACKET) { 2216 slp->ns_flag &= ~SLP_GETSTREAM; 2217 return (EPERM); 2218 } 2219 } 2220 2221 /* 2222 * Now get the record part. 2223 * 2224 * Note that slp->ns_reclen may be 0. Linux sometimes 2225 * generates 0-length records. 2226 */ 2227 if (slp->ns_cc == slp->ns_reclen) { 2228 recm = slp->ns_raw; 2229 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2230 slp->ns_cc = slp->ns_reclen = 0; 2231 } else if (slp->ns_cc > slp->ns_reclen) { 2232 recm = slp->ns_raw; 2233 m = m_split(recm, slp->ns_reclen, waitflag); 2234 if (m == NULL) { 2235 slp->ns_flag &= ~SLP_GETSTREAM; 2236 return (EWOULDBLOCK); 2237 } 2238 m_claimm(recm, &nfs_mowner); 2239 slp->ns_raw = m; 2240 if (m->m_next == NULL) 2241 slp->ns_rawend = m; 2242 slp->ns_cc -= slp->ns_reclen; 2243 slp->ns_reclen = 0; 2244 } else { 2245 slp->ns_flag &= ~SLP_GETSTREAM; 2246 return (0); 2247 } 2248 2249 /* 2250 * Accumulate the fragments into a record. 2251 */ 2252 mpp = &slp->ns_frag; 2253 while (*mpp) 2254 mpp = &((*mpp)->m_next); 2255 *mpp = recm; 2256 if (slp->ns_flag & SLP_LASTFRAG) { 2257 if (slp->ns_recend) 2258 slp->ns_recend->m_nextpkt = slp->ns_frag; 2259 else 2260 slp->ns_rec = slp->ns_frag; 2261 slp->ns_recend = slp->ns_frag; 2262 slp->ns_frag = (struct mbuf *)0; 2263 } 2264 } 2265 } 2266 2267 /* 2268 * Parse an RPC header. 2269 */ 2270 int 2271 nfsrv_dorec(slp, nfsd, ndp) 2272 struct nfssvc_sock *slp; 2273 struct nfsd *nfsd; 2274 struct nfsrv_descript **ndp; 2275 { 2276 struct mbuf *m, *nam; 2277 struct nfsrv_descript *nd; 2278 int error; 2279 2280 *ndp = NULL; 2281 if ((slp->ns_flag & SLP_VALID) == 0 || 2282 (m = slp->ns_rec) == (struct mbuf *)0) 2283 return (ENOBUFS); 2284 slp->ns_rec = m->m_nextpkt; 2285 if (slp->ns_rec) 2286 m->m_nextpkt = (struct mbuf *)0; 2287 else 2288 slp->ns_recend = (struct mbuf *)0; 2289 if (m->m_type == MT_SONAME) { 2290 nam = m; 2291 m = m->m_next; 2292 nam->m_next = NULL; 2293 } else 2294 nam = NULL; 2295 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK); 2296 nd->nd_md = nd->nd_mrep = m; 2297 nd->nd_nam2 = nam; 2298 nd->nd_dpos = mtod(m, caddr_t); 2299 error = nfs_getreq(nd, nfsd, TRUE); 2300 if (error) { 2301 m_freem(nam); 2302 pool_put(&nfs_srvdesc_pool, nd); 2303 return (error); 2304 } 2305 *ndp = nd; 2306 nfsd->nfsd_nd = nd; 2307 return (0); 2308 } 2309 2310 2311 /* 2312 * Search for a sleeping nfsd and wake it up. 2313 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2314 * running nfsds will go look for the work in the nfssvc_sock list. 2315 */ 2316 void 2317 nfsrv_wakenfsd(slp) 2318 struct nfssvc_sock *slp; 2319 { 2320 struct nfsd *nd; 2321 2322 if ((slp->ns_flag & SLP_VALID) == 0) 2323 return; 2324 simple_lock(&nfsd_slock); 2325 if (slp->ns_flag & SLP_DOREC) { 2326 simple_unlock(&nfsd_slock); 2327 return; 2328 } 2329 nd = SLIST_FIRST(&nfsd_idle_head); 2330 if (nd) { 2331 SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle); 2332 simple_unlock(&nfsd_slock); 2333 2334 KASSERT(nd->nfsd_flag & NFSD_WAITING); 2335 nd->nfsd_flag &= ~NFSD_WAITING; 2336 if (nd->nfsd_slp) 2337 panic("nfsd wakeup"); 2338 slp->ns_sref++; 2339 nd->nfsd_slp = slp; 2340 wakeup(nd); 2341 return; 2342 } 2343 slp->ns_flag |= SLP_DOREC; 2344 nfsd_head_flag |= NFSD_CHECKSLP; 2345 TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending); 2346 simple_unlock(&nfsd_slock); 2347 } 2348 #endif /* NFSSERVER */ 2349
Indexes created Tue Sep 30 11:09:46 GMT 2025