Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/nfs/nfs_socket.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /* $NetBSD: nfs_socket.c,v 1.102.2.4 2005/01/11 06:39:49 jmc 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 1.102.2.4 2005/01/11 06:39:49 jmc 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) 158 struct nfsmount *nmp; 159 struct nfsreq *rep; 160 { 161 struct socket *so; 162 int s, error, rcvreserve, sndreserve; 163 struct sockaddr *saddr; 164 struct sockaddr_in *sin; 165 #ifdef INET6 166 struct sockaddr_in6 *sin6; 167 #endif 168 struct mbuf *m; 169 170 nmp->nm_so = (struct socket *)0; 171 saddr = mtod(nmp->nm_nam, struct sockaddr *); 172 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 173 nmp->nm_soproto); 174 if (error) 175 goto bad; 176 so = nmp->nm_so; 177 #ifdef MBUFTRACE 178 so->so_mowner = &nfs_mowner; 179 so->so_rcv.sb_mowner = &nfs_mowner; 180 so->so_snd.sb_mowner = &nfs_mowner; 181 #endif 182 nmp->nm_soflags = so->so_proto->pr_flags; 183 184 /* 185 * Some servers require that the client port be a reserved port number. 186 */ 187 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 188 m = m_get(M_WAIT, MT_SOOPTS); 189 MCLAIM(m, so->so_mowner); 190 *mtod(m, int32_t *) = IP_PORTRANGE_LOW; 191 m->m_len = sizeof(int32_t); 192 if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m))) 193 goto bad; 194 m = m_get(M_WAIT, MT_SONAME); 195 MCLAIM(m, so->so_mowner); 196 sin = mtod(m, struct sockaddr_in *); 197 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 198 sin->sin_family = AF_INET; 199 sin->sin_addr.s_addr = INADDR_ANY; 200 sin->sin_port = 0; 201 error = sobind(so, m, &proc0); 202 m_freem(m); 203 if (error) 204 goto bad; 205 } 206 #ifdef INET6 207 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) { 208 m = m_get(M_WAIT, MT_SOOPTS); 209 MCLAIM(m, so->so_mowner); 210 *mtod(m, int32_t *) = IPV6_PORTRANGE_LOW; 211 m->m_len = sizeof(int32_t); 212 if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m))) 213 goto bad; 214 m = m_get(M_WAIT, MT_SONAME); 215 MCLAIM(m, so->so_mowner); 216 sin6 = mtod(m, struct sockaddr_in6 *); 217 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6); 218 sin6->sin6_family = AF_INET6; 219 sin6->sin6_addr = in6addr_any; 220 sin6->sin6_port = 0; 221 error = sobind(so, m, &proc0); 222 m_freem(m); 223 if (error) 224 goto bad; 225 } 226 #endif 227 228 /* 229 * Protocols that do not require connections may be optionally left 230 * unconnected for servers that reply from a port other than NFS_PORT. 231 */ 232 if (nmp->nm_flag & NFSMNT_NOCONN) { 233 if (nmp->nm_soflags & PR_CONNREQUIRED) { 234 error = ENOTCONN; 235 goto bad; 236 } 237 } else { 238 error = soconnect(so, nmp->nm_nam); 239 if (error) 240 goto bad; 241 242 /* 243 * Wait for the connection to complete. Cribbed from the 244 * connect system call but with the wait timing out so 245 * that interruptible mounts don't hang here for a long time. 246 */ 247 s = splsoftnet(); 248 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 249 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 250 "nfscn1", 2 * hz); 251 if ((so->so_state & SS_ISCONNECTING) && 252 so->so_error == 0 && rep && 253 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ 254 so->so_state &= ~SS_ISCONNECTING; 255 splx(s); 256 goto bad; 257 } 258 } 259 if (so->so_error) { 260 error = so->so_error; 261 so->so_error = 0; 262 splx(s); 263 goto bad; 264 } 265 splx(s); 266 } 267 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 268 so->so_rcv.sb_timeo = (5 * hz); 269 so->so_snd.sb_timeo = (5 * hz); 270 } else { 271 /* 272 * enable receive timeout to detect server crash and reconnect. 273 * otherwise, we can be stuck in soreceive forever. 274 */ 275 so->so_rcv.sb_timeo = (5 * hz); 276 so->so_snd.sb_timeo = 0; 277 } 278 if (nmp->nm_sotype == SOCK_DGRAM) { 279 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 280 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 281 NFS_MAXPKTHDR) * 2; 282 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 283 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 284 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 285 NFS_MAXPKTHDR) * 2; 286 } else { 287 if (nmp->nm_sotype != SOCK_STREAM) 288 panic("nfscon sotype"); 289 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 290 m = m_get(M_WAIT, MT_SOOPTS); 291 MCLAIM(m, so->so_mowner); 292 *mtod(m, int32_t *) = 1; 293 m->m_len = sizeof(int32_t); 294 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 295 } 296 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 297 m = m_get(M_WAIT, MT_SOOPTS); 298 MCLAIM(m, so->so_mowner); 299 *mtod(m, int32_t *) = 1; 300 m->m_len = sizeof(int32_t); 301 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 302 } 303 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 304 sizeof (u_int32_t)) * 2; 305 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 306 sizeof (u_int32_t)) * 2; 307 } 308 error = soreserve(so, sndreserve, rcvreserve); 309 if (error) 310 goto bad; 311 so->so_rcv.sb_flags |= SB_NOINTR; 312 so->so_snd.sb_flags |= SB_NOINTR; 313 314 /* Initialize other non-zero congestion variables */ 315 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 316 NFS_TIMEO << 3; 317 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 318 nmp->nm_sdrtt[3] = 0; 319 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 320 nmp->nm_sent = 0; 321 nmp->nm_timeouts = 0; 322 return (0); 323 324 bad: 325 nfs_disconnect(nmp); 326 return (error); 327 } 328 329 /* 330 * Reconnect routine: 331 * Called when a connection is broken on a reliable protocol. 332 * - clean up the old socket 333 * - nfs_connect() again 334 * - set R_MUSTRESEND for all outstanding requests on mount point 335 * If this fails the mount point is DEAD! 336 * nb: Must be called with the nfs_sndlock() set on the mount point. 337 */ 338 int 339 nfs_reconnect(rep) 340 struct nfsreq *rep; 341 { 342 struct nfsreq *rp; 343 struct nfsmount *nmp = rep->r_nmp; 344 int error; 345 346 nfs_disconnect(nmp); 347 while ((error = nfs_connect(nmp, rep)) != 0) { 348 if (error == EINTR || error == ERESTART) 349 return (EINTR); 350 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0); 351 } 352 353 /* 354 * Loop through outstanding request list and fix up all requests 355 * on old socket. 356 */ 357 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 358 if (rp->r_nmp == nmp) { 359 if ((rp->r_flags & R_MUSTRESEND) == 0) 360 rp->r_flags |= R_MUSTRESEND | R_REXMITTED; 361 rp->r_rexmit = 0; 362 } 363 } 364 return (0); 365 } 366 367 /* 368 * NFS disconnect. Clean up and unlink. 369 */ 370 void 371 nfs_disconnect(nmp) 372 struct nfsmount *nmp; 373 { 374 struct socket *so; 375 int drain = 0; 376 377 if (nmp->nm_so) { 378 so = nmp->nm_so; 379 nmp->nm_so = (struct socket *)0; 380 soshutdown(so, 2); 381 drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0; 382 if (drain) { 383 /* 384 * soshutdown() above should wake up the current 385 * listener. 386 * Now wake up those waiting for the receive lock, and 387 * wait for them to go away unhappy, to prevent *nmp 388 * from evaporating while they're sleeping. 389 */ 390 while (nmp->nm_waiters > 0) { 391 wakeup (&nmp->nm_iflag); 392 (void) tsleep(&nmp->nm_waiters, PVFS, 393 "nfsdis", 0); 394 } 395 } 396 soclose(so); 397 } 398 #ifdef DIAGNOSTIC 399 if (drain && (nmp->nm_waiters > 0)) 400 panic("nfs_disconnect: waiters left after drain?"); 401 #endif 402 } 403 404 void 405 nfs_safedisconnect(nmp) 406 struct nfsmount *nmp; 407 { 408 struct nfsreq dummyreq; 409 410 memset(&dummyreq, 0, sizeof(dummyreq)); 411 dummyreq.r_nmp = nmp; 412 nfs_rcvlock(&dummyreq); /* XXX ignored error return */ 413 nfs_disconnect(nmp); 414 nfs_rcvunlock(nmp); 415 } 416 417 /* 418 * This is the nfs send routine. For connection based socket types, it 419 * must be called with an nfs_sndlock() on the socket. 420 * "rep == NULL" indicates that it has been called from a server. 421 * For the client side: 422 * - return EINTR if the RPC is terminated, 0 otherwise 423 * - set R_MUSTRESEND if the send fails for any reason 424 * - do any cleanup required by recoverable socket errors (? ? ?) 425 * For the server side: 426 * - return EINTR or ERESTART if interrupted by a signal 427 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 428 * - do any cleanup required by recoverable socket errors (? ? ?) 429 */ 430 int 431 nfs_send(so, nam, top, rep) 432 struct socket *so; 433 struct mbuf *nam; 434 struct mbuf *top; 435 struct nfsreq *rep; 436 { 437 struct mbuf *sendnam; 438 int error, soflags, flags; 439 440 if (rep) { 441 if (rep->r_flags & R_SOFTTERM) { 442 m_freem(top); 443 return (EINTR); 444 } 445 if ((so = rep->r_nmp->nm_so) == NULL) { 446 rep->r_flags |= R_MUSTRESEND; 447 m_freem(top); 448 return (0); 449 } 450 rep->r_flags &= ~R_MUSTRESEND; 451 soflags = rep->r_nmp->nm_soflags; 452 } else 453 soflags = so->so_proto->pr_flags; 454 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 455 sendnam = (struct mbuf *)0; 456 else 457 sendnam = nam; 458 if (so->so_type == SOCK_SEQPACKET) 459 flags = MSG_EOR; 460 else 461 flags = 0; 462 463 error = (*so->so_send)(so, sendnam, (struct uio *)0, top, 464 (struct mbuf *)0, flags); 465 if (error) { 466 if (rep) { 467 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 468 /* 469 * We're too fast for the network/driver, 470 * and UDP isn't flowcontrolled. 471 * We need to resend. This is not fatal, 472 * just try again. 473 * 474 * Could be smarter here by doing some sort 475 * of a backoff, but this is rare. 476 */ 477 rep->r_flags |= R_MUSTRESEND; 478 } else { 479 if (error != EPIPE) 480 log(LOG_INFO, 481 "nfs send error %d for %s\n", 482 error, 483 rep->r_nmp->nm_mountp-> 484 mnt_stat.f_mntfromname); 485 /* 486 * Deal with errors for the client side. 487 */ 488 if (rep->r_flags & R_SOFTTERM) 489 error = EINTR; 490 else 491 rep->r_flags |= R_MUSTRESEND; 492 } 493 } else { 494 /* 495 * See above. This error can happen under normal 496 * circumstances and the log is too noisy. 497 * The error will still show up in nfsstat. 498 */ 499 if (error != ENOBUFS || so->so_type != SOCK_DGRAM) 500 log(LOG_INFO, "nfsd send error %d\n", error); 501 } 502 503 /* 504 * Handle any recoverable (soft) socket errors here. (? ? ?) 505 */ 506 if (error != EINTR && error != ERESTART && 507 error != EWOULDBLOCK && error != EPIPE) 508 error = 0; 509 } 510 return (error); 511 } 512 513 #ifdef NFS 514 /* 515 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 516 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 517 * Mark and consolidate the data into a new mbuf list. 518 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 519 * small mbufs. 520 * For SOCK_STREAM we must be very careful to read an entire record once 521 * we have read any of it, even if the system call has been interrupted. 522 */ 523 int 524 nfs_receive(rep, aname, mp) 525 struct nfsreq *rep; 526 struct mbuf **aname; 527 struct mbuf **mp; 528 { 529 struct socket *so; 530 struct uio auio; 531 struct iovec aio; 532 struct mbuf *m; 533 struct mbuf *control; 534 u_int32_t len; 535 struct mbuf **getnam; 536 int error, sotype, rcvflg; 537 struct proc *p = curproc; /* XXX */ 538 539 /* 540 * Set up arguments for soreceive() 541 */ 542 *mp = (struct mbuf *)0; 543 *aname = (struct mbuf *)0; 544 sotype = rep->r_nmp->nm_sotype; 545 546 /* 547 * For reliable protocols, lock against other senders/receivers 548 * in case a reconnect is necessary. 549 * For SOCK_STREAM, first get the Record Mark to find out how much 550 * more there is to get. 551 * We must lock the socket against other receivers 552 * until we have an entire rpc request/reply. 553 */ 554 if (sotype != SOCK_DGRAM) { 555 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 556 if (error) 557 return (error); 558 tryagain: 559 /* 560 * Check for fatal errors and resending request. 561 */ 562 /* 563 * Ugh: If a reconnect attempt just happened, nm_so 564 * would have changed. NULL indicates a failed 565 * attempt that has essentially shut down this 566 * mount point. 567 */ 568 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 569 nfs_sndunlock(&rep->r_nmp->nm_iflag); 570 return (EINTR); 571 } 572 so = rep->r_nmp->nm_so; 573 if (!so) { 574 error = nfs_reconnect(rep); 575 if (error) { 576 nfs_sndunlock(&rep->r_nmp->nm_iflag); 577 return (error); 578 } 579 goto tryagain; 580 } 581 while (rep->r_flags & R_MUSTRESEND) { 582 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 583 nfsstats.rpcretries++; 584 rep->r_rtt = 0; 585 rep->r_flags &= ~R_TIMING; 586 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 587 if (error) { 588 if (error == EINTR || error == ERESTART || 589 (error = nfs_reconnect(rep)) != 0) { 590 nfs_sndunlock(&rep->r_nmp->nm_iflag); 591 return (error); 592 } 593 goto tryagain; 594 } 595 } 596 nfs_sndunlock(&rep->r_nmp->nm_iflag); 597 if (sotype == SOCK_STREAM) { 598 aio.iov_base = (caddr_t) &len; 599 aio.iov_len = sizeof(u_int32_t); 600 auio.uio_iov = &aio; 601 auio.uio_iovcnt = 1; 602 auio.uio_segflg = UIO_SYSSPACE; 603 auio.uio_rw = UIO_READ; 604 auio.uio_offset = 0; 605 auio.uio_resid = sizeof(u_int32_t); 606 auio.uio_procp = p; 607 do { 608 rcvflg = MSG_WAITALL; 609 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 610 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 611 if (error == EWOULDBLOCK && rep) { 612 if (rep->r_flags & R_SOFTTERM) 613 return (EINTR); 614 /* 615 * if it seems that the server died after it 616 * received our request, set EPIPE so that 617 * we'll reconnect and retransmit requests. 618 */ 619 if (rep->r_rexmit >= rep->r_nmp->nm_retry) { 620 nfsstats.rpctimeouts++; 621 error = EPIPE; 622 } 623 } 624 } while (error == EWOULDBLOCK); 625 if (!error && auio.uio_resid > 0) { 626 /* 627 * Don't log a 0 byte receive; it means 628 * that the socket has been closed, and 629 * can happen during normal operation 630 * (forcible unmount or Solaris server). 631 */ 632 if (auio.uio_resid != sizeof (u_int32_t)) 633 log(LOG_INFO, 634 "short receive (%lu/%lu) from nfs server %s\n", 635 (u_long)sizeof(u_int32_t) - auio.uio_resid, 636 (u_long)sizeof(u_int32_t), 637 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 638 error = EPIPE; 639 } 640 if (error) 641 goto errout; 642 len = ntohl(len) & ~0x80000000; 643 /* 644 * This is SERIOUS! We are out of sync with the sender 645 * and forcing a disconnect/reconnect is all I can do. 646 */ 647 if (len > NFS_MAXPACKET) { 648 log(LOG_ERR, "%s (%d) from nfs server %s\n", 649 "impossible packet length", 650 len, 651 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 652 error = EFBIG; 653 goto errout; 654 } 655 auio.uio_resid = len; 656 do { 657 rcvflg = MSG_WAITALL; 658 error = (*so->so_receive)(so, (struct mbuf **)0, 659 &auio, mp, (struct mbuf **)0, &rcvflg); 660 } while (error == EWOULDBLOCK || error == EINTR || 661 error == ERESTART); 662 if (!error && auio.uio_resid > 0) { 663 if (len != auio.uio_resid) 664 log(LOG_INFO, 665 "short receive (%lu/%d) from nfs server %s\n", 666 (u_long)len - auio.uio_resid, len, 667 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 668 error = EPIPE; 669 } 670 } else { 671 /* 672 * NB: Since uio_resid is big, MSG_WAITALL is ignored 673 * and soreceive() will return when it has either a 674 * control msg or a data msg. 675 * We have no use for control msg., but must grab them 676 * and then throw them away so we know what is going 677 * on. 678 */ 679 auio.uio_resid = len = 100000000; /* Anything Big */ 680 auio.uio_procp = p; 681 do { 682 rcvflg = 0; 683 error = (*so->so_receive)(so, (struct mbuf **)0, 684 &auio, mp, &control, &rcvflg); 685 if (control) 686 m_freem(control); 687 if (error == EWOULDBLOCK && rep) { 688 if (rep->r_flags & R_SOFTTERM) 689 return (EINTR); 690 } 691 } while (error == EWOULDBLOCK || 692 (!error && *mp == NULL && control)); 693 if ((rcvflg & MSG_EOR) == 0) 694 printf("Egad!!\n"); 695 if (!error && *mp == NULL) 696 error = EPIPE; 697 len -= auio.uio_resid; 698 } 699 errout: 700 if (error && error != EINTR && error != ERESTART) { 701 m_freem(*mp); 702 *mp = (struct mbuf *)0; 703 if (error != EPIPE) 704 log(LOG_INFO, 705 "receive error %d from nfs server %s\n", 706 error, 707 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 708 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 709 if (!error) 710 error = nfs_reconnect(rep); 711 if (!error) 712 goto tryagain; 713 else 714 nfs_sndunlock(&rep->r_nmp->nm_iflag); 715 } 716 } else { 717 if ((so = rep->r_nmp->nm_so) == NULL) 718 return (EACCES); 719 if (so->so_state & SS_ISCONNECTED) 720 getnam = (struct mbuf **)0; 721 else 722 getnam = aname; 723 auio.uio_resid = len = 1000000; 724 auio.uio_procp = p; 725 do { 726 rcvflg = 0; 727 error = (*so->so_receive)(so, getnam, &auio, mp, 728 (struct mbuf **)0, &rcvflg); 729 if (error == EWOULDBLOCK && 730 (rep->r_flags & R_SOFTTERM)) 731 return (EINTR); 732 } while (error == EWOULDBLOCK); 733 len -= auio.uio_resid; 734 if (!error && *mp == NULL) 735 error = EPIPE; 736 } 737 if (error) { 738 m_freem(*mp); 739 *mp = (struct mbuf *)0; 740 } 741 return (error); 742 } 743 744 /* 745 * Implement receipt of reply on a socket. 746 * We must search through the list of received datagrams matching them 747 * with outstanding requests using the xid, until ours is found. 748 */ 749 /* ARGSUSED */ 750 int 751 nfs_reply(myrep) 752 struct nfsreq *myrep; 753 { 754 struct nfsreq *rep; 755 struct nfsmount *nmp = myrep->r_nmp; 756 int32_t t1; 757 struct mbuf *mrep, *nam, *md; 758 u_int32_t rxid, *tl; 759 caddr_t dpos, cp2; 760 int error; 761 762 /* 763 * Loop around until we get our own reply 764 */ 765 for (;;) { 766 /* 767 * Lock against other receivers so that I don't get stuck in 768 * sbwait() after someone else has received my reply for me. 769 * Also necessary for connection based protocols to avoid 770 * race conditions during a reconnect. 771 */ 772 error = nfs_rcvlock(myrep); 773 if (error == EALREADY) 774 return (0); 775 if (error) 776 return (error); 777 /* 778 * Get the next Rpc reply off the socket 779 */ 780 nmp->nm_waiters++; 781 error = nfs_receive(myrep, &nam, &mrep); 782 nfs_rcvunlock(nmp); 783 if (error) { 784 785 if (nmp->nm_iflag & NFSMNT_DISMNT) { 786 /* 787 * Oops, we're going away now.. 788 */ 789 nmp->nm_waiters--; 790 wakeup (&nmp->nm_waiters); 791 return error; 792 } 793 nmp->nm_waiters--; 794 /* 795 * Ignore routing errors on connectionless protocols? ? 796 */ 797 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 798 nmp->nm_so->so_error = 0; 799 #ifdef DEBUG 800 printf("nfs_reply: ignoring error %d\n", error); 801 #endif 802 if (myrep->r_flags & R_GETONEREP) 803 return (0); 804 continue; 805 } 806 return (error); 807 } 808 nmp->nm_waiters--; 809 if (nam) 810 m_freem(nam); 811 812 /* 813 * Get the xid and check that it is an rpc reply 814 */ 815 md = mrep; 816 dpos = mtod(md, caddr_t); 817 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 818 rxid = *tl++; 819 if (*tl != rpc_reply) { 820 #ifndef NFS_V2_ONLY 821 if (nmp->nm_flag & NFSMNT_NQNFS) { 822 if (nqnfs_callback(nmp, mrep, md, dpos)) 823 nfsstats.rpcinvalid++; 824 } else 825 #endif 826 { 827 nfsstats.rpcinvalid++; 828 m_freem(mrep); 829 } 830 nfsmout: 831 if (myrep->r_flags & R_GETONEREP) 832 return (0); 833 continue; 834 } 835 836 /* 837 * Loop through the request list to match up the reply 838 * Iff no match, just drop the datagram 839 */ 840 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 841 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 842 /* Found it.. */ 843 rep->r_mrep = mrep; 844 rep->r_md = md; 845 rep->r_dpos = dpos; 846 if (nfsrtton) { 847 struct rttl *rt; 848 849 rt = &nfsrtt.rttl[nfsrtt.pos]; 850 rt->proc = rep->r_procnum; 851 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 852 rt->sent = nmp->nm_sent; 853 rt->cwnd = nmp->nm_cwnd; 854 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 855 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 856 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 857 rt->tstamp = time; 858 if (rep->r_flags & R_TIMING) 859 rt->rtt = rep->r_rtt; 860 else 861 rt->rtt = 1000000; 862 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 863 } 864 /* 865 * Update congestion window. 866 * Do the additive increase of 867 * one rpc/rtt. 868 */ 869 if (nmp->nm_cwnd <= nmp->nm_sent) { 870 nmp->nm_cwnd += 871 (NFS_CWNDSCALE * NFS_CWNDSCALE + 872 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 873 if (nmp->nm_cwnd > NFS_MAXCWND) 874 nmp->nm_cwnd = NFS_MAXCWND; 875 } 876 rep->r_flags &= ~R_SENT; 877 nmp->nm_sent -= NFS_CWNDSCALE; 878 /* 879 * Update rtt using a gain of 0.125 on the mean 880 * and a gain of 0.25 on the deviation. 881 */ 882 if (rep->r_flags & R_TIMING) { 883 /* 884 * Since the timer resolution of 885 * NFS_HZ is so course, it can often 886 * result in r_rtt == 0. Since 887 * r_rtt == N means that the actual 888 * rtt is between N+dt and N+2-dt ticks, 889 * add 1. 890 */ 891 t1 = rep->r_rtt + 1; 892 t1 -= (NFS_SRTT(rep) >> 3); 893 NFS_SRTT(rep) += t1; 894 if (t1 < 0) 895 t1 = -t1; 896 t1 -= (NFS_SDRTT(rep) >> 2); 897 NFS_SDRTT(rep) += t1; 898 } 899 nmp->nm_timeouts = 0; 900 break; 901 } 902 } 903 /* 904 * If not matched to a request, drop it. 905 * If it's mine, get out. 906 */ 907 if (rep == 0) { 908 nfsstats.rpcunexpected++; 909 m_freem(mrep); 910 } else if (rep == myrep) { 911 if (rep->r_mrep == NULL) 912 panic("nfsreply nil"); 913 return (0); 914 } 915 if (myrep->r_flags & R_GETONEREP) 916 return (0); 917 } 918 } 919 920 /* 921 * nfs_request - goes something like this 922 * - fill in request struct 923 * - links it into list 924 * - calls nfs_send() for first transmit 925 * - calls nfs_receive() to get reply 926 * - break down rpc header and return with nfs reply pointed to 927 * by mrep or error 928 * nb: always frees up mreq mbuf list 929 */ 930 int 931 nfs_request(np, mrest, procnum, procp, cred, mrp, mdp, dposp, rexmitp) 932 struct nfsnode *np; 933 struct mbuf *mrest; 934 int procnum; 935 struct proc *procp; 936 struct ucred *cred; 937 struct mbuf **mrp; 938 struct mbuf **mdp; 939 caddr_t *dposp; 940 int *rexmitp; 941 { 942 struct mbuf *m, *mrep; 943 struct nfsreq *rep; 944 u_int32_t *tl; 945 int i; 946 struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount); 947 struct mbuf *md, *mheadend; 948 char nickv[RPCX_NICKVERF]; 949 time_t reqtime, waituntil; 950 caddr_t dpos, cp2; 951 int t1, s, error = 0, mrest_len, auth_len, auth_type; 952 int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0; 953 int verf_len, verf_type; 954 u_int32_t xid; 955 char *auth_str, *verf_str; 956 NFSKERBKEY_T key; /* save session key */ 957 struct ucred acred; 958 #ifndef NFS_V2_ONLY 959 int nqlflag, cachable; 960 u_quad_t frev; 961 #endif 962 struct mbuf *mrest_backup = NULL; 963 struct ucred *origcred = NULL; /* XXX: gcc */ 964 boolean_t retry_cred = TRUE; 965 boolean_t use_opencred = (np->n_flag & NUSEOPENCRED) != 0; 966 967 if (rexmitp != NULL) 968 *rexmitp = 0; 969 970 tryagain_cred: 971 KASSERT(cred != NULL); 972 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 973 rep->r_nmp = nmp; 974 rep->r_procp = procp; 975 rep->r_procnum = procnum; 976 i = 0; 977 m = mrest; 978 while (m) { 979 i += m->m_len; 980 m = m->m_next; 981 } 982 mrest_len = i; 983 984 /* 985 * Get the RPC header with authorization. 986 */ 987 kerbauth: 988 verf_str = auth_str = (char *)0; 989 if (nmp->nm_flag & NFSMNT_KERB) { 990 verf_str = nickv; 991 verf_len = sizeof (nickv); 992 auth_type = RPCAUTH_KERB4; 993 memset((caddr_t)key, 0, sizeof (key)); 994 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 995 &auth_len, verf_str, verf_len)) { 996 error = nfs_getauth(nmp, rep, cred, &auth_str, 997 &auth_len, verf_str, &verf_len, key); 998 if (error) { 999 free((caddr_t)rep, M_NFSREQ); 1000 m_freem(mrest); 1001 return (error); 1002 } 1003 } 1004 retry_cred = FALSE; 1005 } else { 1006 /* AUTH_UNIX */ 1007 uid_t uid; 1008 gid_t gid; 1009 1010 /* 1011 * on the most unix filesystems, permission checks are 1012 * done when the file is open(2)'ed. 1013 * ie. once a file is successfully open'ed, 1014 * following i/o operations never fail with EACCES. 1015 * we try to follow the semantics as far as possible. 1016 * 1017 * note that we expect that the nfs server always grant 1018 * accesses by the file's owner. 1019 */ 1020 origcred = cred; 1021 switch (procnum) { 1022 case NFSPROC_READ: 1023 case NFSPROC_WRITE: 1024 case NFSPROC_COMMIT: 1025 uid = np->n_vattr->va_uid; 1026 gid = np->n_vattr->va_gid; 1027 if (cred->cr_uid == uid && cred->cr_gid == gid) { 1028 retry_cred = FALSE; 1029 break; 1030 } 1031 if (use_opencred) 1032 break; 1033 acred.cr_uid = uid; 1034 acred.cr_gid = gid; 1035 acred.cr_ngroups = 0; 1036 acred.cr_ref = 2; /* Just to be safe.. */ 1037 cred = &acred; 1038 break; 1039 default: 1040 retry_cred = FALSE; 1041 break; 1042 } 1043 /* 1044 * backup mbuf chain if we can need it later to retry. 1045 * 1046 * XXX maybe we can keep a direct reference to 1047 * mrest without doing m_copym, but it's ...ugly. 1048 */ 1049 if (retry_cred) 1050 mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT); 1051 auth_type = RPCAUTH_UNIX; 1052 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 1053 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 1054 5 * NFSX_UNSIGNED; 1055 } 1056 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 1057 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 1058 if (auth_str) 1059 free(auth_str, M_TEMP); 1060 1061 /* 1062 * For stream protocols, insert a Sun RPC Record Mark. 1063 */ 1064 if (nmp->nm_sotype == SOCK_STREAM) { 1065 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 1066 *mtod(m, u_int32_t *) = htonl(0x80000000 | 1067 (m->m_pkthdr.len - NFSX_UNSIGNED)); 1068 } 1069 rep->r_mreq = m; 1070 rep->r_xid = xid; 1071 tryagain: 1072 if (nmp->nm_flag & NFSMNT_SOFT) 1073 rep->r_retry = nmp->nm_retry; 1074 else 1075 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1076 rep->r_rtt = rep->r_rexmit = 0; 1077 if (proct[procnum] > 0) 1078 rep->r_flags = R_TIMING; 1079 else 1080 rep->r_flags = 0; 1081 rep->r_mrep = NULL; 1082 1083 /* 1084 * Do the client side RPC. 1085 */ 1086 nfsstats.rpcrequests++; 1087 /* 1088 * Chain request into list of outstanding requests. Be sure 1089 * to put it LAST so timer finds oldest requests first. 1090 */ 1091 s = splsoftnet(); 1092 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1093 1094 /* Get send time for nqnfs */ 1095 reqtime = time.tv_sec; 1096 1097 /* 1098 * If backing off another request or avoiding congestion, don't 1099 * send this one now but let timer do it. If not timing a request, 1100 * do it now. 1101 */ 1102 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1103 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1104 nmp->nm_sent < nmp->nm_cwnd)) { 1105 splx(s); 1106 if (nmp->nm_soflags & PR_CONNREQUIRED) 1107 error = nfs_sndlock(&nmp->nm_iflag, rep); 1108 if (!error) { 1109 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1110 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 1111 if (nmp->nm_soflags & PR_CONNREQUIRED) 1112 nfs_sndunlock(&nmp->nm_iflag); 1113 } 1114 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1115 nmp->nm_sent += NFS_CWNDSCALE; 1116 rep->r_flags |= R_SENT; 1117 } 1118 } else { 1119 splx(s); 1120 rep->r_rtt = -1; 1121 } 1122 1123 /* 1124 * Wait for the reply from our send or the timer's. 1125 */ 1126 if (!error || error == EPIPE) 1127 error = nfs_reply(rep); 1128 1129 /* 1130 * RPC done, unlink the request. 1131 */ 1132 s = splsoftnet(); 1133 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1134 splx(s); 1135 1136 /* 1137 * Decrement the outstanding request count. 1138 */ 1139 if (rep->r_flags & R_SENT) { 1140 rep->r_flags &= ~R_SENT; /* paranoia */ 1141 nmp->nm_sent -= NFS_CWNDSCALE; 1142 } 1143 1144 if (rexmitp != NULL) { 1145 int rexmit; 1146 1147 if (nmp->nm_sotype != SOCK_DGRAM) 1148 rexmit = (rep->r_flags & R_REXMITTED) != 0; 1149 else 1150 rexmit = rep->r_rexmit; 1151 *rexmitp = rexmit; 1152 } 1153 1154 /* 1155 * If there was a successful reply and a tprintf msg. 1156 * tprintf a response. 1157 */ 1158 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1159 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1160 "is alive again"); 1161 mrep = rep->r_mrep; 1162 md = rep->r_md; 1163 dpos = rep->r_dpos; 1164 if (error) 1165 goto nfsmout; 1166 1167 /* 1168 * break down the rpc header and check if ok 1169 */ 1170 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1171 if (*tl++ == rpc_msgdenied) { 1172 if (*tl == rpc_mismatch) 1173 error = EOPNOTSUPP; 1174 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1175 if (!failed_auth) { 1176 failed_auth++; 1177 mheadend->m_next = (struct mbuf *)0; 1178 m_freem(mrep); 1179 m_freem(rep->r_mreq); 1180 goto kerbauth; 1181 } else 1182 error = EAUTH; 1183 } else 1184 error = EACCES; 1185 m_freem(mrep); 1186 goto nfsmout; 1187 } 1188 1189 /* 1190 * Grab any Kerberos verifier, otherwise just throw it away. 1191 */ 1192 verf_type = fxdr_unsigned(int, *tl++); 1193 i = fxdr_unsigned(int32_t, *tl); 1194 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1195 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1196 if (error) 1197 goto nfsmout; 1198 } else if (i > 0) 1199 nfsm_adv(nfsm_rndup(i)); 1200 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1201 /* 0 == ok */ 1202 if (*tl == 0) { 1203 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1204 if (*tl != 0) { 1205 error = fxdr_unsigned(int, *tl); 1206 if (error == NFSERR_ACCES && retry_cred) { 1207 m_freem(mrep); 1208 m_freem(rep->r_mreq); 1209 FREE(rep, M_NFSREQ); 1210 use_opencred = !use_opencred; 1211 if (mrest_backup == NULL) 1212 return ENOMEM; /* m_copym failure */ 1213 mrest = mrest_backup; 1214 mrest_backup = NULL; 1215 cred = origcred; 1216 error = 0; 1217 retry_cred = FALSE; 1218 goto tryagain_cred; 1219 } 1220 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1221 error == NFSERR_TRYLATER) { 1222 m_freem(mrep); 1223 error = 0; 1224 waituntil = time.tv_sec + trylater_delay; 1225 while (time.tv_sec < waituntil) 1226 (void) tsleep((caddr_t)&lbolt, 1227 PSOCK, "nqnfstry", 0); 1228 trylater_delay *= NFS_TRYLATERDELMUL; 1229 if (trylater_delay > NFS_TRYLATERDELMAX) 1230 trylater_delay = NFS_TRYLATERDELMAX; 1231 /* 1232 * RFC1813: 1233 * The client should wait and then try 1234 * the request with a new RPC transaction ID. 1235 */ 1236 nfs_renewxid(rep); 1237 goto tryagain; 1238 } 1239 1240 /* 1241 * If the File Handle was stale, invalidate the 1242 * lookup cache, just in case. 1243 */ 1244 if (error == ESTALE) 1245 cache_purge(NFSTOV(np)); 1246 if (nmp->nm_flag & NFSMNT_NFSV3) { 1247 *mrp = mrep; 1248 *mdp = md; 1249 *dposp = dpos; 1250 error |= NFSERR_RETERR; 1251 } else 1252 m_freem(mrep); 1253 goto nfsmout; 1254 } 1255 1256 /* 1257 * note which credential worked to minimize number of retries. 1258 */ 1259 if (use_opencred) 1260 np->n_flag |= NUSEOPENCRED; 1261 else 1262 np->n_flag &= ~NUSEOPENCRED; 1263 1264 #ifndef NFS_V2_ONLY 1265 /* 1266 * For nqnfs, get any lease in reply 1267 */ 1268 if (nmp->nm_flag & NFSMNT_NQNFS) { 1269 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1270 if (*tl) { 1271 nqlflag = fxdr_unsigned(int, *tl); 1272 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1273 cachable = fxdr_unsigned(int, *tl++); 1274 reqtime += fxdr_unsigned(int, *tl++); 1275 if (reqtime > time.tv_sec) { 1276 frev = fxdr_hyper(tl); 1277 nqnfs_clientlease(nmp, np, nqlflag, 1278 cachable, reqtime, frev); 1279 } 1280 } 1281 } 1282 #endif 1283 *mrp = mrep; 1284 *mdp = md; 1285 *dposp = dpos; 1286 1287 KASSERT(error == 0); 1288 goto nfsmout; 1289 } 1290 m_freem(mrep); 1291 error = EPROTONOSUPPORT; 1292 nfsmout: 1293 m_freem(rep->r_mreq); 1294 free((caddr_t)rep, M_NFSREQ); 1295 m_freem(mrest_backup); 1296 return (error); 1297 } 1298 #endif /* NFS */ 1299 1300 /* 1301 * Generate the rpc reply header 1302 * siz arg. is used to decide if adding a cluster is worthwhile 1303 */ 1304 int 1305 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1306 int siz; 1307 struct nfsrv_descript *nd; 1308 struct nfssvc_sock *slp; 1309 int err; 1310 int cache; 1311 u_quad_t *frev; 1312 struct mbuf **mrq; 1313 struct mbuf **mbp; 1314 caddr_t *bposp; 1315 { 1316 u_int32_t *tl; 1317 struct mbuf *mreq; 1318 caddr_t bpos; 1319 struct mbuf *mb; 1320 1321 mreq = m_gethdr(M_WAIT, MT_DATA); 1322 MCLAIM(mreq, &nfs_mowner); 1323 mb = mreq; 1324 /* 1325 * If this is a big reply, use a cluster else 1326 * try and leave leading space for the lower level headers. 1327 */ 1328 siz += RPC_REPLYSIZ; 1329 if (siz >= max_datalen) { 1330 m_clget(mreq, M_WAIT); 1331 } else 1332 mreq->m_data += max_hdr; 1333 tl = mtod(mreq, u_int32_t *); 1334 mreq->m_len = 6 * NFSX_UNSIGNED; 1335 bpos = ((caddr_t)tl) + mreq->m_len; 1336 *tl++ = txdr_unsigned(nd->nd_retxid); 1337 *tl++ = rpc_reply; 1338 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1339 *tl++ = rpc_msgdenied; 1340 if (err & NFSERR_AUTHERR) { 1341 *tl++ = rpc_autherr; 1342 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1343 mreq->m_len -= NFSX_UNSIGNED; 1344 bpos -= NFSX_UNSIGNED; 1345 } else { 1346 *tl++ = rpc_mismatch; 1347 *tl++ = txdr_unsigned(RPC_VER2); 1348 *tl = txdr_unsigned(RPC_VER2); 1349 } 1350 } else { 1351 *tl++ = rpc_msgaccepted; 1352 1353 /* 1354 * For Kerberos authentication, we must send the nickname 1355 * verifier back, otherwise just RPCAUTH_NULL. 1356 */ 1357 if (nd->nd_flag & ND_KERBFULL) { 1358 struct nfsuid *nuidp; 1359 struct timeval ktvin, ktvout; 1360 1361 LIST_FOREACH(nuidp, NUIDHASH(slp, nd->nd_cr.cr_uid), 1362 nu_hash) { 1363 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1364 (!nd->nd_nam2 || netaddr_match( 1365 NU_NETFAM(nuidp), &nuidp->nu_haddr, 1366 nd->nd_nam2))) 1367 break; 1368 } 1369 if (nuidp) { 1370 ktvin.tv_sec = 1371 txdr_unsigned(nuidp->nu_timestamp.tv_sec 1372 - 1); 1373 ktvin.tv_usec = 1374 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1375 1376 /* 1377 * Encrypt the timestamp in ecb mode using the 1378 * session key. 1379 */ 1380 #ifdef NFSKERB 1381 XXX 1382 #endif 1383 1384 *tl++ = rpc_auth_kerb; 1385 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1386 *tl = ktvout.tv_sec; 1387 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1388 *tl++ = ktvout.tv_usec; 1389 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1390 } else { 1391 *tl++ = 0; 1392 *tl++ = 0; 1393 } 1394 } else { 1395 *tl++ = 0; 1396 *tl++ = 0; 1397 } 1398 switch (err) { 1399 case EPROGUNAVAIL: 1400 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1401 break; 1402 case EPROGMISMATCH: 1403 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1404 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1405 if (nd->nd_flag & ND_NQNFS) { 1406 *tl++ = txdr_unsigned(3); 1407 *tl = txdr_unsigned(3); 1408 } else { 1409 *tl++ = txdr_unsigned(2); 1410 *tl = txdr_unsigned(3); 1411 } 1412 break; 1413 case EPROCUNAVAIL: 1414 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1415 break; 1416 case EBADRPC: 1417 *tl = txdr_unsigned(RPC_GARBAGE); 1418 break; 1419 default: 1420 *tl = 0; 1421 if (err != NFSERR_RETVOID) { 1422 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1423 if (err) 1424 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1425 else 1426 *tl = 0; 1427 } 1428 break; 1429 }; 1430 } 1431 1432 /* 1433 * For nqnfs, piggyback lease as requested. 1434 */ 1435 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1436 if (nd->nd_flag & ND_LEASE) { 1437 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1438 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1439 *tl++ = txdr_unsigned(cache); 1440 *tl++ = txdr_unsigned(nd->nd_duration); 1441 txdr_hyper(*frev, tl); 1442 } else { 1443 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1444 *tl = 0; 1445 } 1446 } 1447 if (mrq != NULL) 1448 *mrq = mreq; 1449 *mbp = mb; 1450 *bposp = bpos; 1451 if (err != 0 && err != NFSERR_RETVOID) 1452 nfsstats.srvrpc_errs++; 1453 return (0); 1454 } 1455 1456 /* 1457 * Nfs timer routine 1458 * Scan the nfsreq list and retranmit any requests that have timed out 1459 * To avoid retransmission attempts on STREAM sockets (in the future) make 1460 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1461 */ 1462 void 1463 nfs_timer(arg) 1464 void *arg; /* never used */ 1465 { 1466 struct nfsreq *rep; 1467 struct mbuf *m; 1468 struct socket *so; 1469 struct nfsmount *nmp; 1470 int timeo; 1471 int s, error; 1472 #ifdef NFSSERVER 1473 struct nfssvc_sock *slp; 1474 static long lasttime = 0; 1475 u_quad_t cur_usec; 1476 #endif 1477 1478 s = splsoftnet(); 1479 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1480 nmp = rep->r_nmp; 1481 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1482 continue; 1483 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1484 rep->r_flags |= R_SOFTTERM; 1485 continue; 1486 } 1487 if (rep->r_rtt >= 0) { 1488 rep->r_rtt++; 1489 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1490 timeo = nmp->nm_timeo; 1491 else 1492 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1493 if (nmp->nm_timeouts > 0) 1494 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1495 if (rep->r_rtt <= timeo) 1496 continue; 1497 if (nmp->nm_timeouts < 1498 (sizeof(nfs_backoff) / sizeof(nfs_backoff[0]))) 1499 nmp->nm_timeouts++; 1500 } 1501 /* 1502 * Check for server not responding 1503 */ 1504 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1505 rep->r_rexmit > nmp->nm_deadthresh) { 1506 nfs_msg(rep->r_procp, 1507 nmp->nm_mountp->mnt_stat.f_mntfromname, 1508 "not responding"); 1509 rep->r_flags |= R_TPRINTFMSG; 1510 } 1511 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1512 nfsstats.rpctimeouts++; 1513 rep->r_flags |= R_SOFTTERM; 1514 continue; 1515 } 1516 if (nmp->nm_sotype != SOCK_DGRAM) { 1517 if (++rep->r_rexmit > NFS_MAXREXMIT) 1518 rep->r_rexmit = NFS_MAXREXMIT; 1519 continue; 1520 } 1521 if ((so = nmp->nm_so) == NULL) 1522 continue; 1523 1524 /* 1525 * If there is enough space and the window allows.. 1526 * Resend it 1527 * Set r_rtt to -1 in case we fail to send it now. 1528 */ 1529 rep->r_rtt = -1; 1530 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1531 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1532 (rep->r_flags & R_SENT) || 1533 nmp->nm_sent < nmp->nm_cwnd) && 1534 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1535 if (so->so_state & SS_ISCONNECTED) 1536 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1537 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1538 else 1539 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1540 nmp->nm_nam, (struct mbuf *)0, (struct proc *)0); 1541 if (error) { 1542 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1543 #ifdef DEBUG 1544 printf("nfs_timer: ignoring error %d\n", 1545 error); 1546 #endif 1547 so->so_error = 0; 1548 } 1549 } else { 1550 /* 1551 * Iff first send, start timing 1552 * else turn timing off, backoff timer 1553 * and divide congestion window by 2. 1554 */ 1555 if (rep->r_flags & R_SENT) { 1556 rep->r_flags &= ~R_TIMING; 1557 if (++rep->r_rexmit > NFS_MAXREXMIT) 1558 rep->r_rexmit = NFS_MAXREXMIT; 1559 nmp->nm_cwnd >>= 1; 1560 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1561 nmp->nm_cwnd = NFS_CWNDSCALE; 1562 nfsstats.rpcretries++; 1563 } else { 1564 rep->r_flags |= R_SENT; 1565 nmp->nm_sent += NFS_CWNDSCALE; 1566 } 1567 rep->r_rtt = 0; 1568 } 1569 } 1570 } 1571 1572 #ifdef NFSSERVER 1573 /* 1574 * Call the nqnfs server timer once a second to handle leases. 1575 */ 1576 if (lasttime != time.tv_sec) { 1577 lasttime = time.tv_sec; 1578 nqnfs_serverd(); 1579 } 1580 1581 /* 1582 * Scan the write gathering queues for writes that need to be 1583 * completed now. 1584 */ 1585 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1586 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1587 if (LIST_FIRST(&slp->ns_tq) && 1588 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) 1589 nfsrv_wakenfsd(slp); 1590 } 1591 #endif /* NFSSERVER */ 1592 splx(s); 1593 callout_schedule(&nfs_timer_ch, nfs_ticks); 1594 } 1595 1596 /*ARGSUSED*/ 1597 void 1598 nfs_exit(p, v) 1599 struct proc *p; 1600 void *v; 1601 { 1602 struct nfsreq *rp; 1603 int s = splsoftnet(); 1604 1605 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 1606 if (rp->r_procp == p) 1607 TAILQ_REMOVE(&nfs_reqq, rp, r_chain); 1608 } 1609 splx(s); 1610 } 1611 1612 /* 1613 * Test for a termination condition pending on the process. 1614 * This is used for NFSMNT_INT mounts. 1615 */ 1616 int 1617 nfs_sigintr(nmp, rep, p) 1618 struct nfsmount *nmp; 1619 struct nfsreq *rep; 1620 struct proc *p; 1621 { 1622 sigset_t ss; 1623 1624 if (rep && (rep->r_flags & R_SOFTTERM)) 1625 return (EINTR); 1626 if (!(nmp->nm_flag & NFSMNT_INT)) 1627 return (0); 1628 if (p) { 1629 sigpending1(p, &ss); 1630 #if 0 1631 sigminusset(&p->p_sigctx.ps_sigignore, &ss); 1632 #endif 1633 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1634 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1635 sigismember(&ss, SIGQUIT)) 1636 return (EINTR); 1637 } 1638 return (0); 1639 } 1640 1641 /* 1642 * Lock a socket against others. 1643 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1644 * and also to avoid race conditions between the processes with nfs requests 1645 * in progress when a reconnect is necessary. 1646 */ 1647 int 1648 nfs_sndlock(flagp, rep) 1649 int *flagp; 1650 struct nfsreq *rep; 1651 { 1652 struct proc *p; 1653 int slpflag = 0, slptimeo = 0; 1654 1655 if (rep) { 1656 p = rep->r_procp; 1657 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1658 slpflag = PCATCH; 1659 } else 1660 p = (struct proc *)0; 1661 while (*flagp & NFSMNT_SNDLOCK) { 1662 if (rep && nfs_sigintr(rep->r_nmp, rep, p)) 1663 return (EINTR); 1664 *flagp |= NFSMNT_WANTSND; 1665 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1666 slptimeo); 1667 if (slpflag == PCATCH) { 1668 slpflag = 0; 1669 slptimeo = 2 * hz; 1670 } 1671 } 1672 *flagp |= NFSMNT_SNDLOCK; 1673 return (0); 1674 } 1675 1676 /* 1677 * Unlock the stream socket for others. 1678 */ 1679 void 1680 nfs_sndunlock(flagp) 1681 int *flagp; 1682 { 1683 1684 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1685 panic("nfs sndunlock"); 1686 *flagp &= ~NFSMNT_SNDLOCK; 1687 if (*flagp & NFSMNT_WANTSND) { 1688 *flagp &= ~NFSMNT_WANTSND; 1689 wakeup((caddr_t)flagp); 1690 } 1691 } 1692 1693 int 1694 nfs_rcvlock(rep) 1695 struct nfsreq *rep; 1696 { 1697 struct nfsmount *nmp = rep->r_nmp; 1698 int *flagp = &nmp->nm_iflag; 1699 int slpflag, slptimeo = 0; 1700 int error = 0; 1701 1702 if (*flagp & NFSMNT_DISMNT) 1703 return EIO; 1704 1705 if (*flagp & NFSMNT_INT) 1706 slpflag = PCATCH; 1707 else 1708 slpflag = 0; 1709 simple_lock(&nmp->nm_slock); 1710 while (*flagp & NFSMNT_RCVLOCK) { 1711 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) { 1712 error = EINTR; 1713 goto quit; 1714 } 1715 *flagp |= NFSMNT_WANTRCV; 1716 nmp->nm_waiters++; 1717 (void) ltsleep(flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1718 slptimeo, &nmp->nm_slock); 1719 nmp->nm_waiters--; 1720 if (*flagp & NFSMNT_DISMNT) { 1721 wakeup(&nmp->nm_waiters); 1722 error = EIO; 1723 goto quit; 1724 } 1725 /* If our reply was received while we were sleeping, 1726 * then just return without taking the lock to avoid a 1727 * situation where a single iod could 'capture' the 1728 * receive lock. 1729 */ 1730 if (rep->r_mrep != NULL) { 1731 error = EALREADY; 1732 goto quit; 1733 } 1734 if (slpflag == PCATCH) { 1735 slpflag = 0; 1736 slptimeo = 2 * hz; 1737 } 1738 } 1739 *flagp |= NFSMNT_RCVLOCK; 1740 quit: 1741 simple_unlock(&nmp->nm_slock); 1742 return error; 1743 } 1744 1745 /* 1746 * Unlock the stream socket for others. 1747 */ 1748 void 1749 nfs_rcvunlock(nmp) 1750 struct nfsmount *nmp; 1751 { 1752 int *flagp = &nmp->nm_iflag; 1753 1754 simple_lock(&nmp->nm_slock); 1755 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1756 panic("nfs rcvunlock"); 1757 *flagp &= ~NFSMNT_RCVLOCK; 1758 if (*flagp & NFSMNT_WANTRCV) { 1759 *flagp &= ~NFSMNT_WANTRCV; 1760 wakeup((caddr_t)flagp); 1761 } 1762 simple_unlock(&nmp->nm_slock); 1763 } 1764 1765 /* 1766 * Parse an RPC request 1767 * - verify it 1768 * - fill in the cred struct. 1769 */ 1770 int 1771 nfs_getreq(nd, nfsd, has_header) 1772 struct nfsrv_descript *nd; 1773 struct nfsd *nfsd; 1774 int has_header; 1775 { 1776 int len, i; 1777 u_int32_t *tl; 1778 int32_t t1; 1779 struct uio uio; 1780 struct iovec iov; 1781 caddr_t dpos, cp2, cp; 1782 u_int32_t nfsvers, auth_type; 1783 uid_t nickuid; 1784 int error = 0, nqnfs = 0, ticklen; 1785 struct mbuf *mrep, *md; 1786 struct nfsuid *nuidp; 1787 struct timeval tvin, tvout; 1788 1789 mrep = nd->nd_mrep; 1790 md = nd->nd_md; 1791 dpos = nd->nd_dpos; 1792 if (has_header) { 1793 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1794 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1795 if (*tl++ != rpc_call) { 1796 m_freem(mrep); 1797 return (EBADRPC); 1798 } 1799 } else 1800 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1801 nd->nd_repstat = 0; 1802 nd->nd_flag = 0; 1803 if (*tl++ != rpc_vers) { 1804 nd->nd_repstat = ERPCMISMATCH; 1805 nd->nd_procnum = NFSPROC_NOOP; 1806 return (0); 1807 } 1808 if (*tl != nfs_prog) { 1809 if (*tl == nqnfs_prog) 1810 nqnfs++; 1811 else { 1812 nd->nd_repstat = EPROGUNAVAIL; 1813 nd->nd_procnum = NFSPROC_NOOP; 1814 return (0); 1815 } 1816 } 1817 tl++; 1818 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1819 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1820 (nfsvers != NQNFS_VER3 && nqnfs)) { 1821 nd->nd_repstat = EPROGMISMATCH; 1822 nd->nd_procnum = NFSPROC_NOOP; 1823 return (0); 1824 } 1825 if (nqnfs) 1826 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1827 else if (nfsvers == NFS_VER3) 1828 nd->nd_flag = ND_NFSV3; 1829 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1830 if (nd->nd_procnum == NFSPROC_NULL) 1831 return (0); 1832 if (nd->nd_procnum >= NFS_NPROCS || 1833 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1834 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1835 nd->nd_repstat = EPROCUNAVAIL; 1836 nd->nd_procnum = NFSPROC_NOOP; 1837 return (0); 1838 } 1839 if ((nd->nd_flag & ND_NFSV3) == 0) 1840 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1841 auth_type = *tl++; 1842 len = fxdr_unsigned(int, *tl++); 1843 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1844 m_freem(mrep); 1845 return (EBADRPC); 1846 } 1847 1848 nd->nd_flag &= ~ND_KERBAUTH; 1849 /* 1850 * Handle auth_unix or auth_kerb. 1851 */ 1852 if (auth_type == rpc_auth_unix) { 1853 len = fxdr_unsigned(int, *++tl); 1854 if (len < 0 || len > NFS_MAXNAMLEN) { 1855 m_freem(mrep); 1856 return (EBADRPC); 1857 } 1858 nfsm_adv(nfsm_rndup(len)); 1859 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1860 memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred)); 1861 nd->nd_cr.cr_ref = 1; 1862 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1863 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1864 len = fxdr_unsigned(int, *tl); 1865 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1866 m_freem(mrep); 1867 return (EBADRPC); 1868 } 1869 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1870 for (i = 0; i < len; i++) 1871 if (i < NGROUPS) 1872 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1873 else 1874 tl++; 1875 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1876 if (nd->nd_cr.cr_ngroups > 1) 1877 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1878 len = fxdr_unsigned(int, *++tl); 1879 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1880 m_freem(mrep); 1881 return (EBADRPC); 1882 } 1883 if (len > 0) 1884 nfsm_adv(nfsm_rndup(len)); 1885 } else if (auth_type == rpc_auth_kerb) { 1886 switch (fxdr_unsigned(int, *tl++)) { 1887 case RPCAKN_FULLNAME: 1888 ticklen = fxdr_unsigned(int, *tl); 1889 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1890 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1891 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1892 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1893 m_freem(mrep); 1894 return (EBADRPC); 1895 } 1896 uio.uio_offset = 0; 1897 uio.uio_iov = &iov; 1898 uio.uio_iovcnt = 1; 1899 uio.uio_segflg = UIO_SYSSPACE; 1900 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1901 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1902 nfsm_mtouio(&uio, uio.uio_resid); 1903 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1904 if (*tl++ != rpc_auth_kerb || 1905 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1906 printf("Bad kerb verifier\n"); 1907 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1908 nd->nd_procnum = NFSPROC_NOOP; 1909 return (0); 1910 } 1911 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1912 tl = (u_int32_t *)cp; 1913 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1914 printf("Not fullname kerb verifier\n"); 1915 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1916 nd->nd_procnum = NFSPROC_NOOP; 1917 return (0); 1918 } 1919 cp += NFSX_UNSIGNED; 1920 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 1921 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1922 nd->nd_flag |= ND_KERBFULL; 1923 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1924 break; 1925 case RPCAKN_NICKNAME: 1926 if (len != 2 * NFSX_UNSIGNED) { 1927 printf("Kerb nickname short\n"); 1928 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1929 nd->nd_procnum = NFSPROC_NOOP; 1930 return (0); 1931 } 1932 nickuid = fxdr_unsigned(uid_t, *tl); 1933 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1934 if (*tl++ != rpc_auth_kerb || 1935 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1936 printf("Kerb nick verifier bad\n"); 1937 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1938 nd->nd_procnum = NFSPROC_NOOP; 1939 return (0); 1940 } 1941 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1942 tvin.tv_sec = *tl++; 1943 tvin.tv_usec = *tl; 1944 1945 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), 1946 nu_hash) { 1947 if (nuidp->nu_cr.cr_uid == nickuid && 1948 (!nd->nd_nam2 || 1949 netaddr_match(NU_NETFAM(nuidp), 1950 &nuidp->nu_haddr, nd->nd_nam2))) 1951 break; 1952 } 1953 if (!nuidp) { 1954 nd->nd_repstat = 1955 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1956 nd->nd_procnum = NFSPROC_NOOP; 1957 return (0); 1958 } 1959 1960 /* 1961 * Now, decrypt the timestamp using the session key 1962 * and validate it. 1963 */ 1964 #ifdef NFSKERB 1965 XXX 1966 #endif 1967 1968 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1969 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1970 if (nuidp->nu_expire < time.tv_sec || 1971 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1972 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1973 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1974 nuidp->nu_expire = 0; 1975 nd->nd_repstat = 1976 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1977 nd->nd_procnum = NFSPROC_NOOP; 1978 return (0); 1979 } 1980 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1981 nd->nd_flag |= ND_KERBNICK; 1982 }; 1983 } else { 1984 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1985 nd->nd_procnum = NFSPROC_NOOP; 1986 return (0); 1987 } 1988 1989 /* 1990 * For nqnfs, get piggybacked lease request. 1991 */ 1992 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1993 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1994 nd->nd_flag |= fxdr_unsigned(int, *tl); 1995 if (nd->nd_flag & ND_LEASE) { 1996 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1997 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 1998 } else 1999 nd->nd_duration = NQ_MINLEASE; 2000 } else 2001 nd->nd_duration = NQ_MINLEASE; 2002 nd->nd_md = md; 2003 nd->nd_dpos = dpos; 2004 return (0); 2005 nfsmout: 2006 return (error); 2007 } 2008 2009 int 2010 nfs_msg(p, server, msg) 2011 struct proc *p; 2012 char *server, *msg; 2013 { 2014 tpr_t tpr; 2015 2016 if (p) 2017 tpr = tprintf_open(p); 2018 else 2019 tpr = NULL; 2020 tprintf(tpr, "nfs server %s: %s\n", server, msg); 2021 tprintf_close(tpr); 2022 return (0); 2023 } 2024 2025 #ifdef NFSSERVER 2026 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 2027 struct nfssvc_sock *, struct proc *, 2028 struct mbuf **)) = { 2029 nfsrv_null, 2030 nfsrv_getattr, 2031 nfsrv_setattr, 2032 nfsrv_lookup, 2033 nfsrv3_access, 2034 nfsrv_readlink, 2035 nfsrv_read, 2036 nfsrv_write, 2037 nfsrv_create, 2038 nfsrv_mkdir, 2039 nfsrv_symlink, 2040 nfsrv_mknod, 2041 nfsrv_remove, 2042 nfsrv_rmdir, 2043 nfsrv_rename, 2044 nfsrv_link, 2045 nfsrv_readdir, 2046 nfsrv_readdirplus, 2047 nfsrv_statfs, 2048 nfsrv_fsinfo, 2049 nfsrv_pathconf, 2050 nfsrv_commit, 2051 nqnfsrv_getlease, 2052 nqnfsrv_vacated, 2053 nfsrv_noop, 2054 nfsrv_noop 2055 }; 2056 2057 /* 2058 * Socket upcall routine for the nfsd sockets. 2059 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 2060 * Essentially do as much as possible non-blocking, else punt and it will 2061 * be called with M_WAIT from an nfsd. 2062 */ 2063 void 2064 nfsrv_rcv(so, arg, waitflag) 2065 struct socket *so; 2066 caddr_t arg; 2067 int waitflag; 2068 { 2069 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 2070 struct mbuf *m; 2071 struct mbuf *mp, *nam; 2072 struct uio auio; 2073 int flags, error; 2074 2075 if ((slp->ns_flag & SLP_VALID) == 0) 2076 return; 2077 #ifdef notdef 2078 /* 2079 * Define this to test for nfsds handling this under heavy load. 2080 */ 2081 if (waitflag == M_DONTWAIT) { 2082 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 2083 } 2084 #endif 2085 auio.uio_procp = NULL; 2086 if (so->so_type == SOCK_STREAM) { 2087 /* 2088 * If there are already records on the queue, defer soreceive() 2089 * to an nfsd so that there is feedback to the TCP layer that 2090 * the nfs servers are heavily loaded. 2091 */ 2092 if (slp->ns_rec && waitflag == M_DONTWAIT) { 2093 slp->ns_flag |= SLP_NEEDQ; 2094 goto dorecs; 2095 } 2096 2097 /* 2098 * Do soreceive(). 2099 */ 2100 auio.uio_resid = 1000000000; 2101 flags = MSG_DONTWAIT; 2102 error = (*so->so_receive)(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 2103 if (error || mp == (struct mbuf *)0) { 2104 if (error == EWOULDBLOCK) 2105 slp->ns_flag |= SLP_NEEDQ; 2106 else 2107 slp->ns_flag |= SLP_DISCONN; 2108 goto dorecs; 2109 } 2110 m = mp; 2111 if (slp->ns_rawend) { 2112 slp->ns_rawend->m_next = m; 2113 slp->ns_cc += 1000000000 - auio.uio_resid; 2114 } else { 2115 slp->ns_raw = m; 2116 slp->ns_cc = 1000000000 - auio.uio_resid; 2117 } 2118 while (m->m_next) 2119 m = m->m_next; 2120 slp->ns_rawend = m; 2121 2122 /* 2123 * Now try and parse record(s) out of the raw stream data. 2124 */ 2125 error = nfsrv_getstream(slp, waitflag); 2126 if (error) { 2127 if (error == EPERM) 2128 slp->ns_flag |= SLP_DISCONN; 2129 else 2130 slp->ns_flag |= SLP_NEEDQ; 2131 } 2132 } else { 2133 do { 2134 auio.uio_resid = 1000000000; 2135 flags = MSG_DONTWAIT; 2136 error = (*so->so_receive)(so, &nam, &auio, &mp, 2137 (struct mbuf **)0, &flags); 2138 if (mp) { 2139 if (nam) { 2140 m = nam; 2141 m->m_next = mp; 2142 } else 2143 m = mp; 2144 if (slp->ns_recend) 2145 slp->ns_recend->m_nextpkt = m; 2146 else 2147 slp->ns_rec = m; 2148 slp->ns_recend = m; 2149 m->m_nextpkt = (struct mbuf *)0; 2150 } 2151 if (error) { 2152 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2153 && error != EWOULDBLOCK) { 2154 slp->ns_flag |= SLP_DISCONN; 2155 goto dorecs; 2156 } 2157 } 2158 } while (mp); 2159 } 2160 2161 /* 2162 * Now try and process the request records, non-blocking. 2163 */ 2164 dorecs: 2165 if (waitflag == M_DONTWAIT && 2166 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 2167 nfsrv_wakenfsd(slp); 2168 } 2169 2170 /* 2171 * Try and extract an RPC request from the mbuf data list received on a 2172 * stream socket. The "waitflag" argument indicates whether or not it 2173 * can sleep. 2174 */ 2175 int 2176 nfsrv_getstream(slp, waitflag) 2177 struct nfssvc_sock *slp; 2178 int waitflag; 2179 { 2180 struct mbuf *m, **mpp; 2181 struct mbuf *recm; 2182 u_int32_t recmark; 2183 2184 if (slp->ns_flag & SLP_GETSTREAM) 2185 panic("nfs getstream"); 2186 slp->ns_flag |= SLP_GETSTREAM; 2187 for (;;) { 2188 if (slp->ns_reclen == 0) { 2189 if (slp->ns_cc < NFSX_UNSIGNED) { 2190 slp->ns_flag &= ~SLP_GETSTREAM; 2191 return (0); 2192 } 2193 m = slp->ns_raw; 2194 m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark); 2195 m_adj(m, NFSX_UNSIGNED); 2196 slp->ns_cc -= NFSX_UNSIGNED; 2197 recmark = ntohl(recmark); 2198 slp->ns_reclen = recmark & ~0x80000000; 2199 if (recmark & 0x80000000) 2200 slp->ns_flag |= SLP_LASTFRAG; 2201 else 2202 slp->ns_flag &= ~SLP_LASTFRAG; 2203 if (slp->ns_reclen > NFS_MAXPACKET) { 2204 slp->ns_flag &= ~SLP_GETSTREAM; 2205 return (EPERM); 2206 } 2207 } 2208 2209 /* 2210 * Now get the record part. 2211 * 2212 * Note that slp->ns_reclen may be 0. Linux sometimes 2213 * generates 0-length records. 2214 */ 2215 if (slp->ns_cc == slp->ns_reclen) { 2216 recm = slp->ns_raw; 2217 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2218 slp->ns_cc = slp->ns_reclen = 0; 2219 } else if (slp->ns_cc > slp->ns_reclen) { 2220 recm = slp->ns_raw; 2221 m = m_split(recm, slp->ns_reclen, waitflag); 2222 if (m == NULL) { 2223 slp->ns_flag &= ~SLP_GETSTREAM; 2224 return (EWOULDBLOCK); 2225 } 2226 m_claimm(recm, &nfs_mowner); 2227 slp->ns_raw = m; 2228 if (m->m_next == NULL) 2229 slp->ns_rawend = m; 2230 slp->ns_cc -= slp->ns_reclen; 2231 slp->ns_reclen = 0; 2232 } else { 2233 slp->ns_flag &= ~SLP_GETSTREAM; 2234 return (0); 2235 } 2236 2237 /* 2238 * Accumulate the fragments into a record. 2239 */ 2240 mpp = &slp->ns_frag; 2241 while (*mpp) 2242 mpp = &((*mpp)->m_next); 2243 *mpp = recm; 2244 if (slp->ns_flag & SLP_LASTFRAG) { 2245 if (slp->ns_recend) 2246 slp->ns_recend->m_nextpkt = slp->ns_frag; 2247 else 2248 slp->ns_rec = slp->ns_frag; 2249 slp->ns_recend = slp->ns_frag; 2250 slp->ns_frag = (struct mbuf *)0; 2251 } 2252 } 2253 } 2254 2255 /* 2256 * Parse an RPC header. 2257 */ 2258 int 2259 nfsrv_dorec(slp, nfsd, ndp) 2260 struct nfssvc_sock *slp; 2261 struct nfsd *nfsd; 2262 struct nfsrv_descript **ndp; 2263 { 2264 struct mbuf *m, *nam; 2265 struct nfsrv_descript *nd; 2266 int error; 2267 2268 *ndp = NULL; 2269 if ((slp->ns_flag & SLP_VALID) == 0 || 2270 (m = slp->ns_rec) == (struct mbuf *)0) 2271 return (ENOBUFS); 2272 slp->ns_rec = m->m_nextpkt; 2273 if (slp->ns_rec) 2274 m->m_nextpkt = (struct mbuf *)0; 2275 else 2276 slp->ns_recend = (struct mbuf *)0; 2277 if (m->m_type == MT_SONAME) { 2278 nam = m; 2279 m = m->m_next; 2280 nam->m_next = NULL; 2281 } else 2282 nam = NULL; 2283 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK); 2284 nd->nd_md = nd->nd_mrep = m; 2285 nd->nd_nam2 = nam; 2286 nd->nd_dpos = mtod(m, caddr_t); 2287 error = nfs_getreq(nd, nfsd, TRUE); 2288 if (error) { 2289 m_freem(nam); 2290 pool_put(&nfs_srvdesc_pool, nd); 2291 return (error); 2292 } 2293 *ndp = nd; 2294 nfsd->nfsd_nd = nd; 2295 return (0); 2296 } 2297 2298 2299 /* 2300 * Search for a sleeping nfsd and wake it up. 2301 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2302 * running nfsds will go look for the work in the nfssvc_sock list. 2303 */ 2304 void 2305 nfsrv_wakenfsd(slp) 2306 struct nfssvc_sock *slp; 2307 { 2308 struct nfsd *nd; 2309 2310 if ((slp->ns_flag & SLP_VALID) == 0) 2311 return; 2312 simple_lock(&nfsd_slock); 2313 if (slp->ns_flag & SLP_DOREC) { 2314 simple_unlock(&nfsd_slock); 2315 return; 2316 } 2317 nd = SLIST_FIRST(&nfsd_idle_head); 2318 if (nd) { 2319 SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle); 2320 simple_unlock(&nfsd_slock); 2321 2322 KASSERT(nd->nfsd_flag & NFSD_WAITING); 2323 nd->nfsd_flag &= ~NFSD_WAITING; 2324 if (nd->nfsd_slp) 2325 panic("nfsd wakeup"); 2326 slp->ns_sref++; 2327 nd->nfsd_slp = slp; 2328 wakeup(nd); 2329 return; 2330 } 2331 slp->ns_flag |= SLP_DOREC; 2332 nfsd_head_flag |= NFSD_CHECKSLP; 2333 TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending); 2334 simple_unlock(&nfsd_slock); 2335 } 2336 #endif /* NFSSERVER */
Cache object: b956c66009b9c8bf05caa266194a5c1d
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]