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 /* 2 * Copyright (c) 1989, 1991, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 37 * $FreeBSD$ 38 */ 39 40 /* 41 * Socket operations for use by nfs 42 */ 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/proc.h> 47 #include <sys/malloc.h> 48 #include <sys/mount.h> 49 #include <sys/kernel.h> 50 #include <sys/mbuf.h> 51 #include <sys/vnode.h> 52 #include <sys/protosw.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/syslog.h> 56 #include <sys/tprintf.h> 57 58 #include <netinet/in.h> 59 #include <netinet/tcp.h> 60 61 #include <nfs/rpcv2.h> 62 #include <nfs/nfsproto.h> 63 #include <nfs/nfs.h> 64 #include <nfs/xdr_subs.h> 65 #include <nfs/nfsm_subs.h> 66 #include <nfs/nfsmount.h> 67 #include <nfs/nfsnode.h> 68 #include <nfs/nfsrtt.h> 69 #include <nfs/nqnfs.h> 70 71 #define TRUE 1 72 #define FALSE 0 73 74 /* 75 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 76 * Use the mean and mean deviation of rtt for the appropriate type of rpc 77 * for the frequent rpcs and a default for the others. 78 * The justification for doing "other" this way is that these rpcs 79 * happen so infrequently that timer est. would probably be stale. 80 * Also, since many of these rpcs are 81 * non-idempotent, a conservative timeout is desired. 82 * getattr, lookup - A+2D 83 * read, write - A+4D 84 * other - nm_timeo 85 */ 86 #define NFS_RTO(n, t) \ 87 ((t) == 0 ? (n)->nm_timeo : \ 88 ((t) < 3 ? \ 89 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 90 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 91 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 92 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 93 /* 94 * External data, mostly RPC constants in XDR form 95 */ 96 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 97 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, 98 rpc_auth_kerb; 99 extern u_int32_t nfs_prog, nqnfs_prog; 100 extern time_t nqnfsstarttime; 101 extern struct nfsstats nfsstats; 102 extern int nfsv3_procid[NFS_NPROCS]; 103 extern int nfs_ticks; 104 105 /* 106 * Defines which timer to use for the procnum. 107 * 0 - default 108 * 1 - getattr 109 * 2 - lookup 110 * 3 - read 111 * 4 - write 112 */ 113 static int proct[NFS_NPROCS] = { 114 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0, 115 0, 0, 0, 116 }; 117 118 /* 119 * There is a congestion window for outstanding rpcs maintained per mount 120 * point. The cwnd size is adjusted in roughly the way that: 121 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 122 * SIGCOMM '88". ACM, August 1988. 123 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 124 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 125 * of rpcs is in progress. 126 * (The sent count and cwnd are scaled for integer arith.) 127 * Variants of "slow start" were tried and were found to be too much of a 128 * performance hit (ave. rtt 3 times larger), 129 * I suspect due to the large rtt that nfs rpcs have. 130 */ 131 #define NFS_CWNDSCALE 256 132 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 133 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 134 int nfsrtton = 0; 135 struct nfsrtt nfsrtt; 136 struct callout_handle nfs_timer_handle; 137 138 static int nfs_msg __P((struct proc *,char *,char *)); 139 static int nfs_rcvlock __P((struct nfsreq *)); 140 static void nfs_rcvunlock __P((int *flagp, int *statep)); 141 static void nfs_realign __P((struct mbuf **pm, int hsiz)); 142 static int nfs_receive __P((struct nfsreq *rep, struct sockaddr **aname, 143 struct mbuf **mp)); 144 static void nfs_softterm __P((struct nfsreq *rep)); 145 static int nfs_reconnect __P((struct nfsreq *rep)); 146 #ifndef NFS_NOSERVER 147 static int nfsrv_getstream __P((struct nfssvc_sock *,int)); 148 149 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *nd, 150 struct nfssvc_sock *slp, 151 struct proc *procp, 152 struct mbuf **mreqp)) = { 153 nfsrv_null, 154 nfsrv_getattr, 155 nfsrv_setattr, 156 nfsrv_lookup, 157 nfsrv3_access, 158 nfsrv_readlink, 159 nfsrv_read, 160 nfsrv_write, 161 nfsrv_create, 162 nfsrv_mkdir, 163 nfsrv_symlink, 164 nfsrv_mknod, 165 nfsrv_remove, 166 nfsrv_rmdir, 167 nfsrv_rename, 168 nfsrv_link, 169 nfsrv_readdir, 170 nfsrv_readdirplus, 171 nfsrv_statfs, 172 nfsrv_fsinfo, 173 nfsrv_pathconf, 174 nfsrv_commit, 175 nqnfsrv_getlease, 176 nqnfsrv_vacated, 177 nfsrv_noop, 178 nfsrv_noop 179 }; 180 #endif /* NFS_NOSERVER */ 181 182 /* 183 * Initialize sockets and congestion for a new NFS connection. 184 * We do not free the sockaddr if error. 185 */ 186 int 187 nfs_connect(nmp, rep) 188 register struct nfsmount *nmp; 189 struct nfsreq *rep; 190 { 191 register struct socket *so; 192 int s, error, rcvreserve, sndreserve; 193 struct sockaddr *saddr; 194 struct sockaddr_in *sin; 195 struct proc *p = &proc0; /* only used for socreate and sobind */ 196 197 nmp->nm_so = (struct socket *)0; 198 saddr = nmp->nm_nam; 199 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 200 nmp->nm_soproto, p); 201 if (error) 202 goto bad; 203 so = nmp->nm_so; 204 nmp->nm_soflags = so->so_proto->pr_flags; 205 206 /* 207 * Some servers require that the client port be a reserved port number. 208 */ 209 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 210 struct sockopt sopt; 211 int ip; 212 struct sockaddr_in ssin; 213 214 bzero(&sopt, sizeof sopt); 215 ip = IP_PORTRANGE_LOW; 216 sopt.sopt_dir = SOPT_SET; 217 sopt.sopt_level = IPPROTO_IP; 218 sopt.sopt_name = IP_PORTRANGE; 219 sopt.sopt_val = (void *)&ip; 220 sopt.sopt_valsize = sizeof(ip); 221 sopt.sopt_p = NULL; 222 error = sosetopt(so, &sopt); 223 if (error) 224 goto bad; 225 bzero(&ssin, sizeof ssin); 226 sin = &ssin; 227 sin->sin_len = sizeof (struct sockaddr_in); 228 sin->sin_family = AF_INET; 229 sin->sin_addr.s_addr = INADDR_ANY; 230 sin->sin_port = htons(0); 231 error = sobind(so, (struct sockaddr *)sin, p); 232 if (error) 233 goto bad; 234 bzero(&sopt, sizeof sopt); 235 ip = IP_PORTRANGE_DEFAULT; 236 sopt.sopt_dir = SOPT_SET; 237 sopt.sopt_level = IPPROTO_IP; 238 sopt.sopt_name = IP_PORTRANGE; 239 sopt.sopt_val = (void *)&ip; 240 sopt.sopt_valsize = sizeof(ip); 241 sopt.sopt_p = NULL; 242 error = sosetopt(so, &sopt); 243 if (error) 244 goto bad; 245 } 246 247 /* 248 * Protocols that do not require connections may be optionally left 249 * unconnected for servers that reply from a port other than NFS_PORT. 250 */ 251 if (nmp->nm_flag & NFSMNT_NOCONN) { 252 if (nmp->nm_soflags & PR_CONNREQUIRED) { 253 error = ENOTCONN; 254 goto bad; 255 } 256 } else { 257 /* XXX should not use mbuf */ 258 error = soconnect(so, nmp->nm_nam, p); 259 if (error) 260 goto bad; 261 262 /* 263 * Wait for the connection to complete. Cribbed from the 264 * connect system call but with the wait timing out so 265 * that interruptible mounts don't hang here for a long time. 266 */ 267 s = splnet(); 268 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 269 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 270 "nfscon", 2 * hz); 271 if ((so->so_state & SS_ISCONNECTING) && 272 so->so_error == 0 && rep && 273 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ 274 so->so_state &= ~SS_ISCONNECTING; 275 splx(s); 276 goto bad; 277 } 278 } 279 if (so->so_error) { 280 error = so->so_error; 281 so->so_error = 0; 282 splx(s); 283 goto bad; 284 } 285 splx(s); 286 } 287 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 288 so->so_rcv.sb_timeo = (5 * hz); 289 so->so_snd.sb_timeo = (5 * hz); 290 } else { 291 so->so_rcv.sb_timeo = 0; 292 so->so_snd.sb_timeo = 0; 293 } 294 if (nmp->nm_sotype == SOCK_DGRAM) { 295 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 296 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 297 NFS_MAXPKTHDR) * 2; 298 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 299 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 300 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 301 NFS_MAXPKTHDR) * 2; 302 } else { 303 if (nmp->nm_sotype != SOCK_STREAM) 304 panic("nfscon sotype"); 305 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 306 struct sockopt sopt; 307 int val; 308 309 bzero(&sopt, sizeof sopt); 310 sopt.sopt_level = SOL_SOCKET; 311 sopt.sopt_name = SO_KEEPALIVE; 312 sopt.sopt_val = &val; 313 sopt.sopt_valsize = sizeof val; 314 val = 1; 315 sosetopt(so, &sopt); 316 } 317 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 318 struct sockopt sopt; 319 int val; 320 321 bzero(&sopt, sizeof sopt); 322 sopt.sopt_level = IPPROTO_TCP; 323 sopt.sopt_name = TCP_NODELAY; 324 sopt.sopt_val = &val; 325 sopt.sopt_valsize = sizeof val; 326 val = 1; 327 sosetopt(so, &sopt); 328 } 329 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 330 sizeof (u_int32_t)) * 2; 331 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 332 sizeof (u_int32_t)) * 2; 333 } 334 error = soreserve(so, sndreserve, rcvreserve); 335 if (error) 336 goto bad; 337 so->so_rcv.sb_flags |= SB_NOINTR; 338 so->so_snd.sb_flags |= SB_NOINTR; 339 340 /* Initialize other non-zero congestion variables */ 341 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 342 nmp->nm_srtt[4] = (NFS_TIMEO << 3); 343 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 344 nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0; 345 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 346 nmp->nm_sent = 0; 347 nmp->nm_timeouts = 0; 348 return (0); 349 350 bad: 351 nfs_disconnect(nmp); 352 return (error); 353 } 354 355 /* 356 * Reconnect routine: 357 * Called when a connection is broken on a reliable protocol. 358 * - clean up the old socket 359 * - nfs_connect() again 360 * - set R_MUSTRESEND for all outstanding requests on mount point 361 * If this fails the mount point is DEAD! 362 * nb: Must be called with the nfs_sndlock() set on the mount point. 363 */ 364 static int 365 nfs_reconnect(rep) 366 register struct nfsreq *rep; 367 { 368 register struct nfsreq *rp; 369 register struct nfsmount *nmp = rep->r_nmp; 370 int error; 371 372 nfs_disconnect(nmp); 373 while ((error = nfs_connect(nmp, rep)) != 0) { 374 if (error == EINTR || error == ERESTART) 375 return (EINTR); 376 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); 377 } 378 379 /* 380 * Loop through outstanding request list and fix up all requests 381 * on old socket. 382 */ 383 for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) { 384 if (rp->r_nmp == nmp) 385 rp->r_flags |= R_MUSTRESEND; 386 } 387 return (0); 388 } 389 390 /* 391 * NFS disconnect. Clean up and unlink. 392 */ 393 void 394 nfs_disconnect(nmp) 395 register struct nfsmount *nmp; 396 { 397 register struct socket *so; 398 399 if (nmp->nm_so) { 400 so = nmp->nm_so; 401 nmp->nm_so = (struct socket *)0; 402 soshutdown(so, 2); 403 soclose(so); 404 } 405 } 406 407 void 408 nfs_safedisconnect(nmp) 409 struct nfsmount *nmp; 410 { 411 struct nfsreq dummyreq; 412 413 bzero(&dummyreq, sizeof(dummyreq)); 414 dummyreq.r_nmp = nmp; 415 nfs_rcvlock(&dummyreq); 416 nfs_disconnect(nmp); 417 nfs_rcvunlock(&nmp->nm_flag, &nmp->nm_state); 418 } 419 420 /* 421 * This is the nfs send routine. For connection based socket types, it 422 * must be called with an nfs_sndlock() on the socket. 423 * "rep == NULL" indicates that it has been called from a server. 424 * For the client side: 425 * - return EINTR if the RPC is terminated, 0 otherwise 426 * - set R_MUSTRESEND if the send fails for any reason 427 * - do any cleanup required by recoverable socket errors (?) 428 * For the server side: 429 * - return EINTR or ERESTART if interrupted by a signal 430 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 431 * - do any cleanup required by recoverable socket errors (?) 432 */ 433 int 434 nfs_send(so, nam, top, rep) 435 register struct socket *so; 436 struct sockaddr *nam; 437 register struct mbuf *top; 438 struct nfsreq *rep; 439 { 440 struct sockaddr *sendnam; 441 int error, soflags, flags; 442 443 if (rep) { 444 if (rep->r_flags & R_SOFTTERM) { 445 m_freem(top); 446 return (EINTR); 447 } 448 if ((so = rep->r_nmp->nm_so) == NULL) { 449 rep->r_flags |= R_MUSTRESEND; 450 m_freem(top); 451 return (0); 452 } 453 rep->r_flags &= ~R_MUSTRESEND; 454 soflags = rep->r_nmp->nm_soflags; 455 } else 456 soflags = so->so_proto->pr_flags; 457 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 458 sendnam = (struct sockaddr *)0; 459 else 460 sendnam = nam; 461 if (so->so_type == SOCK_SEQPACKET) 462 flags = MSG_EOR; 463 else 464 flags = 0; 465 466 error = so->so_proto->pr_usrreqs->pru_sosend(so, sendnam, 0, top, 0, 467 flags, curproc /*XXX*/); 468 /* 469 * ENOBUFS for dgram sockets is transient and non fatal. 470 * No need to log, and no need to break a soft mount. 471 */ 472 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 473 error = 0; 474 if (rep) /* do backoff retransmit on client */ 475 rep->r_flags |= R_MUSTRESEND; 476 } 477 478 if (error) { 479 if (rep) { 480 log(LOG_INFO, "nfs send error %d for server %s\n",error, 481 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 482 /* 483 * Deal with errors for the client side. 484 */ 485 if (rep->r_flags & R_SOFTTERM) 486 error = EINTR; 487 else 488 rep->r_flags |= R_MUSTRESEND; 489 } else 490 log(LOG_INFO, "nfsd send error %d\n", error); 491 492 /* 493 * Handle any recoverable (soft) socket errors here. (?) 494 */ 495 if (error != EINTR && error != ERESTART && 496 error != EWOULDBLOCK && error != EPIPE) 497 error = 0; 498 } 499 return (error); 500 } 501 502 /* 503 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 504 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 505 * Mark and consolidate the data into a new mbuf list. 506 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 507 * small mbufs. 508 * For SOCK_STREAM we must be very careful to read an entire record once 509 * we have read any of it, even if the system call has been interrupted. 510 */ 511 static int 512 nfs_receive(rep, aname, mp) 513 register struct nfsreq *rep; 514 struct sockaddr **aname; 515 struct mbuf **mp; 516 { 517 register struct socket *so; 518 struct uio auio; 519 struct iovec aio; 520 register struct mbuf *m; 521 struct mbuf *control; 522 u_int32_t len; 523 struct sockaddr **getnam; 524 int error, sotype, rcvflg; 525 struct proc *p = curproc; /* XXX */ 526 527 /* 528 * Set up arguments for soreceive() 529 */ 530 *mp = (struct mbuf *)0; 531 *aname = (struct sockaddr *)0; 532 sotype = rep->r_nmp->nm_sotype; 533 534 /* 535 * For reliable protocols, lock against other senders/receivers 536 * in case a reconnect is necessary. 537 * For SOCK_STREAM, first get the Record Mark to find out how much 538 * more there is to get. 539 * We must lock the socket against other receivers 540 * until we have an entire rpc request/reply. 541 */ 542 if (sotype != SOCK_DGRAM) { 543 error = nfs_sndlock(&rep->r_nmp->nm_flag, &rep->r_nmp->nm_state, 544 rep); 545 if (error) 546 return (error); 547 tryagain: 548 /* 549 * Check for fatal errors and resending request. 550 */ 551 /* 552 * Ugh: If a reconnect attempt just happened, nm_so 553 * would have changed. NULL indicates a failed 554 * attempt that has essentially shut down this 555 * mount point. 556 */ 557 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 558 nfs_sndunlock(&rep->r_nmp->nm_flag, 559 &rep->r_nmp->nm_state); 560 return (EINTR); 561 } 562 so = rep->r_nmp->nm_so; 563 if (!so) { 564 error = nfs_reconnect(rep); 565 if (error) { 566 nfs_sndunlock(&rep->r_nmp->nm_flag, 567 &rep->r_nmp->nm_state); 568 return (error); 569 } 570 goto tryagain; 571 } 572 while (rep->r_flags & R_MUSTRESEND) { 573 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 574 nfsstats.rpcretries++; 575 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 576 if (error) { 577 if (error == EINTR || error == ERESTART || 578 (error = nfs_reconnect(rep)) != 0) { 579 nfs_sndunlock(&rep->r_nmp->nm_flag, 580 &rep->r_nmp->nm_state); 581 return (error); 582 } 583 goto tryagain; 584 } 585 } 586 nfs_sndunlock(&rep->r_nmp->nm_flag, &rep->r_nmp->nm_state); 587 if (sotype == SOCK_STREAM) { 588 aio.iov_base = (caddr_t) &len; 589 aio.iov_len = sizeof(u_int32_t); 590 auio.uio_iov = &aio; 591 auio.uio_iovcnt = 1; 592 auio.uio_segflg = UIO_SYSSPACE; 593 auio.uio_rw = UIO_READ; 594 auio.uio_offset = 0; 595 auio.uio_resid = sizeof(u_int32_t); 596 auio.uio_procp = p; 597 do { 598 rcvflg = MSG_WAITALL; 599 error = so->so_proto->pr_usrreqs->pru_soreceive 600 (so, (struct sockaddr **)0, &auio, 601 (struct mbuf **)0, (struct mbuf **)0, 602 &rcvflg); 603 if (error == EWOULDBLOCK && rep) { 604 if (rep->r_flags & R_SOFTTERM) 605 return (EINTR); 606 } 607 } while (error == EWOULDBLOCK); 608 if (!error && auio.uio_resid > 0) { 609 /* 610 * Don't log a 0 byte receive; it means 611 * that the socket has been closed, and 612 * can happen during normal operation 613 * (forcible unmount or Solaris server). 614 */ 615 if (auio.uio_resid != sizeof (u_int32_t)) 616 log(LOG_INFO, 617 "short receive (%d/%d) from nfs server %s\n", 618 sizeof(u_int32_t) - auio.uio_resid, 619 sizeof(u_int32_t), 620 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 621 error = EPIPE; 622 } 623 if (error) 624 goto errout; 625 len = ntohl(len) & ~0x80000000; 626 /* 627 * This is SERIOUS! We are out of sync with the sender 628 * and forcing a disconnect/reconnect is all I can do. 629 */ 630 if (len > NFS_MAXPACKET) { 631 log(LOG_ERR, "%s (%d) from nfs server %s\n", 632 "impossible packet length", 633 len, 634 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 635 error = EFBIG; 636 goto errout; 637 } 638 auio.uio_resid = len; 639 do { 640 rcvflg = MSG_WAITALL; 641 error = so->so_proto->pr_usrreqs->pru_soreceive 642 (so, (struct sockaddr **)0, 643 &auio, mp, (struct mbuf **)0, &rcvflg); 644 } while (error == EWOULDBLOCK || error == EINTR || 645 error == ERESTART); 646 if (!error && auio.uio_resid > 0) { 647 if (len != auio.uio_resid) 648 log(LOG_INFO, 649 "short receive (%d/%d) from nfs server %s\n", 650 len - auio.uio_resid, len, 651 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 652 error = EPIPE; 653 } 654 } else { 655 /* 656 * NB: Since uio_resid is big, MSG_WAITALL is ignored 657 * and soreceive() will return when it has either a 658 * control msg or a data msg. 659 * We have no use for control msg., but must grab them 660 * and then throw them away so we know what is going 661 * on. 662 */ 663 auio.uio_resid = len = 100000000; /* Anything Big */ 664 auio.uio_procp = p; 665 do { 666 rcvflg = 0; 667 error = so->so_proto->pr_usrreqs->pru_soreceive 668 (so, (struct sockaddr **)0, 669 &auio, mp, &control, &rcvflg); 670 if (control) 671 m_freem(control); 672 if (error == EWOULDBLOCK && rep) { 673 if (rep->r_flags & R_SOFTTERM) 674 return (EINTR); 675 } 676 } while (error == EWOULDBLOCK || 677 (!error && *mp == NULL && control)); 678 if ((rcvflg & MSG_EOR) == 0) 679 printf("Egad!!\n"); 680 if (!error && *mp == NULL) 681 error = EPIPE; 682 len -= auio.uio_resid; 683 } 684 errout: 685 if (error && error != EINTR && error != ERESTART) { 686 m_freem(*mp); 687 *mp = (struct mbuf *)0; 688 if (error != EPIPE) 689 log(LOG_INFO, 690 "receive error %d from nfs server %s\n", 691 error, 692 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 693 error = nfs_sndlock(&rep->r_nmp->nm_flag, 694 &rep->r_nmp->nm_state, rep); 695 if (!error) 696 error = nfs_reconnect(rep); 697 if (!error) 698 goto tryagain; 699 else 700 nfs_sndunlock(&rep->r_nmp->nm_flag, 701 &rep->r_nmp->nm_state); 702 } 703 } else { 704 if ((so = rep->r_nmp->nm_so) == NULL) 705 return (EACCES); 706 if (so->so_state & SS_ISCONNECTED) 707 getnam = (struct sockaddr **)0; 708 else 709 getnam = aname; 710 auio.uio_resid = len = 1000000; 711 auio.uio_procp = p; 712 do { 713 rcvflg = 0; 714 error = so->so_proto->pr_usrreqs->pru_soreceive 715 (so, getnam, &auio, mp, 716 (struct mbuf **)0, &rcvflg); 717 if (error == EWOULDBLOCK && 718 (rep->r_flags & R_SOFTTERM)) 719 return (EINTR); 720 } while (error == EWOULDBLOCK); 721 len -= auio.uio_resid; 722 } 723 if (error) { 724 m_freem(*mp); 725 *mp = (struct mbuf *)0; 726 } 727 /* 728 * Search for any mbufs that are not a multiple of 4 bytes long 729 * or with m_data not longword aligned. 730 * These could cause pointer alignment problems, so copy them to 731 * well aligned mbufs. 732 */ 733 nfs_realign(mp, 5 * NFSX_UNSIGNED); 734 return (error); 735 } 736 737 /* 738 * Implement receipt of reply on a socket. 739 * We must search through the list of received datagrams matching them 740 * with outstanding requests using the xid, until ours is found. 741 */ 742 /* ARGSUSED */ 743 int 744 nfs_reply(myrep) 745 struct nfsreq *myrep; 746 { 747 register struct nfsreq *rep; 748 register struct nfsmount *nmp = myrep->r_nmp; 749 register int32_t t1; 750 struct mbuf *mrep, *md; 751 struct sockaddr *nam; 752 u_int32_t rxid, *tl; 753 caddr_t dpos, cp2; 754 int error; 755 756 /* 757 * Loop around until we get our own reply 758 */ 759 for (;;) { 760 /* 761 * Lock against other receivers so that I don't get stuck in 762 * sbwait() after someone else has received my reply for me. 763 * Also necessary for connection based protocols to avoid 764 * race conditions during a reconnect. 765 * If nfs_rcvlock() returns EALREADY, that means that 766 * the reply has already been recieved by another 767 * process and we can return immediately. In this 768 * case, the lock is not taken to avoid races with 769 * other processes. 770 */ 771 error = nfs_rcvlock(myrep); 772 if (error == EALREADY) 773 return (0); 774 if (error) 775 return (error); 776 /* 777 * Get the next Rpc reply off the socket 778 */ 779 error = nfs_receive(myrep, &nam, &mrep); 780 nfs_rcvunlock(&nmp->nm_flag, &nmp->nm_state); 781 if (error) { 782 783 /* 784 * Ignore routing errors on connectionless protocols?? 785 */ 786 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 787 nmp->nm_so->so_error = 0; 788 if (myrep->r_flags & R_GETONEREP) 789 return (0); 790 continue; 791 } 792 return (error); 793 } 794 if (nam) 795 FREE(nam, M_SONAME); 796 797 /* 798 * Get the xid and check that it is an rpc reply 799 */ 800 md = mrep; 801 dpos = mtod(md, caddr_t); 802 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 803 rxid = *tl++; 804 if (*tl != rpc_reply) { 805 #ifndef NFS_NOSERVER 806 if (nmp->nm_flag & NFSMNT_NQNFS) { 807 if (nqnfs_callback(nmp, mrep, md, dpos)) 808 nfsstats.rpcinvalid++; 809 } else { 810 nfsstats.rpcinvalid++; 811 m_freem(mrep); 812 } 813 #else 814 nfsstats.rpcinvalid++; 815 m_freem(mrep); 816 #endif 817 nfsmout: 818 if (myrep->r_flags & R_GETONEREP) 819 return (0); 820 continue; 821 } 822 823 /* 824 * Loop through the request list to match up the reply 825 * Iff no match, just drop the datagram 826 */ 827 for (rep = nfs_reqq.tqh_first; rep != 0; 828 rep = rep->r_chain.tqe_next) { 829 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 830 /* Found it.. */ 831 rep->r_mrep = mrep; 832 rep->r_md = md; 833 rep->r_dpos = dpos; 834 if (nfsrtton) { 835 struct rttl *rt; 836 837 rt = &nfsrtt.rttl[nfsrtt.pos]; 838 rt->proc = rep->r_procnum; 839 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 840 rt->sent = nmp->nm_sent; 841 rt->cwnd = nmp->nm_cwnd; 842 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 843 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 844 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 845 getmicrotime(&rt->tstamp); 846 if (rep->r_flags & R_TIMING) 847 rt->rtt = rep->r_rtt; 848 else 849 rt->rtt = 1000000; 850 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 851 } 852 /* 853 * Update congestion window. 854 * Do the additive increase of 855 * one rpc/rtt. 856 */ 857 if (nmp->nm_cwnd <= nmp->nm_sent) { 858 nmp->nm_cwnd += 859 (NFS_CWNDSCALE * NFS_CWNDSCALE + 860 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 861 if (nmp->nm_cwnd > NFS_MAXCWND) 862 nmp->nm_cwnd = NFS_MAXCWND; 863 } 864 if (rep->r_flags & R_SENT) { 865 rep->r_flags &= ~R_SENT; 866 nmp->nm_sent -= NFS_CWNDSCALE; 867 } 868 /* 869 * Update rtt using a gain of 0.125 on the mean 870 * and a gain of 0.25 on the deviation. 871 */ 872 if (rep->r_flags & R_TIMING) { 873 /* 874 * Since the timer resolution of 875 * NFS_HZ is so course, it can often 876 * result in r_rtt == 0. Since 877 * r_rtt == N means that the actual 878 * rtt is between N+dt and N+2-dt ticks, 879 * add 1. 880 */ 881 t1 = rep->r_rtt + 1; 882 t1 -= (NFS_SRTT(rep) >> 3); 883 NFS_SRTT(rep) += t1; 884 if (t1 < 0) 885 t1 = -t1; 886 t1 -= (NFS_SDRTT(rep) >> 2); 887 NFS_SDRTT(rep) += t1; 888 } 889 nmp->nm_timeouts = 0; 890 break; 891 } 892 } 893 /* 894 * If not matched to a request, drop it. 895 * If it's mine, get out. 896 */ 897 if (rep == 0) { 898 nfsstats.rpcunexpected++; 899 m_freem(mrep); 900 } else if (rep == myrep) { 901 if (rep->r_mrep == NULL) 902 panic("nfsreply nil"); 903 return (0); 904 } 905 if (myrep->r_flags & R_GETONEREP) 906 return (0); 907 } 908 } 909 910 /* 911 * nfs_request - goes something like this 912 * - fill in request struct 913 * - links it into list 914 * - calls nfs_send() for first transmit 915 * - calls nfs_receive() to get reply 916 * - break down rpc header and return with nfs reply pointed to 917 * by mrep or error 918 * nb: always frees up mreq mbuf list 919 */ 920 int 921 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 922 struct vnode *vp; 923 struct mbuf *mrest; 924 int procnum; 925 struct proc *procp; 926 struct ucred *cred; 927 struct mbuf **mrp; 928 struct mbuf **mdp; 929 caddr_t *dposp; 930 { 931 register struct mbuf *m, *mrep, *m2; 932 register struct nfsreq *rep; 933 register u_int32_t *tl; 934 register int i; 935 struct nfsmount *nmp; 936 struct mbuf *md, *mheadend; 937 struct nfsnode *np; 938 char nickv[RPCX_NICKVERF]; 939 time_t reqtime, waituntil; 940 caddr_t dpos, cp2; 941 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; 942 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 943 int verf_len, verf_type; 944 u_int32_t xid; 945 u_quad_t frev; 946 char *auth_str, *verf_str; 947 NFSKERBKEY_T key; /* save session key */ 948 949 nmp = VFSTONFS(vp->v_mount); 950 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 951 rep->r_nmp = nmp; 952 rep->r_vp = vp; 953 rep->r_procp = procp; 954 rep->r_procnum = procnum; 955 i = 0; 956 m = mrest; 957 while (m) { 958 i += m->m_len; 959 m = m->m_next; 960 } 961 mrest_len = i; 962 963 /* 964 * Get the RPC header with authorization. 965 */ 966 kerbauth: 967 verf_str = auth_str = (char *)0; 968 if (nmp->nm_flag & NFSMNT_KERB) { 969 verf_str = nickv; 970 verf_len = sizeof (nickv); 971 auth_type = RPCAUTH_KERB4; 972 bzero((caddr_t)key, sizeof (key)); 973 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 974 &auth_len, verf_str, verf_len)) { 975 error = nfs_getauth(nmp, rep, cred, &auth_str, 976 &auth_len, verf_str, &verf_len, key); 977 if (error) { 978 free((caddr_t)rep, M_NFSREQ); 979 m_freem(mrest); 980 return (error); 981 } 982 } 983 } else { 984 auth_type = RPCAUTH_UNIX; 985 if (cred->cr_ngroups < 1) 986 panic("nfsreq nogrps"); 987 auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ? 988 nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) + 989 5 * NFSX_UNSIGNED; 990 } 991 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 992 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 993 if (auth_str) 994 free(auth_str, M_TEMP); 995 996 /* 997 * For stream protocols, insert a Sun RPC Record Mark. 998 */ 999 if (nmp->nm_sotype == SOCK_STREAM) { 1000 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 1001 *mtod(m, u_int32_t *) = htonl(0x80000000 | 1002 (m->m_pkthdr.len - NFSX_UNSIGNED)); 1003 } 1004 rep->r_mreq = m; 1005 rep->r_xid = xid; 1006 tryagain: 1007 if (nmp->nm_flag & NFSMNT_SOFT) 1008 rep->r_retry = nmp->nm_retry; 1009 else 1010 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1011 rep->r_rtt = rep->r_rexmit = 0; 1012 if (proct[procnum] > 0) 1013 rep->r_flags = R_TIMING; 1014 else 1015 rep->r_flags = 0; 1016 rep->r_mrep = NULL; 1017 1018 /* 1019 * Do the client side RPC. 1020 */ 1021 nfsstats.rpcrequests++; 1022 /* 1023 * Chain request into list of outstanding requests. Be sure 1024 * to put it LAST so timer finds oldest requests first. 1025 */ 1026 s = splsoftclock(); 1027 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1028 1029 /* Get send time for nqnfs */ 1030 reqtime = time_second; 1031 1032 /* 1033 * If backing off another request or avoiding congestion, don't 1034 * send this one now but let timer do it. If not timing a request, 1035 * do it now. 1036 */ 1037 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1038 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1039 nmp->nm_sent < nmp->nm_cwnd)) { 1040 splx(s); 1041 if (nmp->nm_soflags & PR_CONNREQUIRED) 1042 error = nfs_sndlock(&nmp->nm_flag, &nmp->nm_state, rep); 1043 if (!error) { 1044 m2 = m_copym(m, 0, M_COPYALL, M_WAIT); 1045 error = nfs_send(nmp->nm_so, nmp->nm_nam, m2, rep); 1046 if (nmp->nm_soflags & PR_CONNREQUIRED) 1047 nfs_sndunlock(&nmp->nm_flag, &nmp->nm_state); 1048 } 1049 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1050 nmp->nm_sent += NFS_CWNDSCALE; 1051 rep->r_flags |= R_SENT; 1052 } 1053 } else { 1054 splx(s); 1055 rep->r_rtt = -1; 1056 } 1057 1058 /* 1059 * Wait for the reply from our send or the timer's. 1060 */ 1061 if (!error || error == EPIPE) 1062 error = nfs_reply(rep); 1063 1064 /* 1065 * RPC done, unlink the request. 1066 */ 1067 s = splsoftclock(); 1068 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1069 splx(s); 1070 1071 /* 1072 * Decrement the outstanding request count. 1073 */ 1074 if (rep->r_flags & R_SENT) { 1075 rep->r_flags &= ~R_SENT; /* paranoia */ 1076 nmp->nm_sent -= NFS_CWNDSCALE; 1077 } 1078 1079 /* 1080 * If there was a successful reply and a tprintf msg. 1081 * tprintf a response. 1082 */ 1083 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1084 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1085 "is alive again"); 1086 mrep = rep->r_mrep; 1087 md = rep->r_md; 1088 dpos = rep->r_dpos; 1089 if (error) { 1090 m_freem(rep->r_mreq); 1091 free((caddr_t)rep, M_NFSREQ); 1092 return (error); 1093 } 1094 1095 /* 1096 * break down the rpc header and check if ok 1097 */ 1098 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1099 if (*tl++ == rpc_msgdenied) { 1100 if (*tl == rpc_mismatch) 1101 error = EOPNOTSUPP; 1102 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1103 if (!failed_auth) { 1104 failed_auth++; 1105 mheadend->m_next = (struct mbuf *)0; 1106 m_freem(mrep); 1107 m_freem(rep->r_mreq); 1108 goto kerbauth; 1109 } else 1110 error = EAUTH; 1111 } else 1112 error = EACCES; 1113 m_freem(mrep); 1114 m_freem(rep->r_mreq); 1115 free((caddr_t)rep, M_NFSREQ); 1116 return (error); 1117 } 1118 1119 /* 1120 * Grab any Kerberos verifier, otherwise just throw it away. 1121 */ 1122 verf_type = fxdr_unsigned(int, *tl++); 1123 i = fxdr_unsigned(int32_t, *tl); 1124 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1125 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1126 if (error) 1127 goto nfsmout; 1128 } else if (i > 0) 1129 nfsm_adv(nfsm_rndup(i)); 1130 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1131 /* 0 == ok */ 1132 if (*tl == 0) { 1133 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1134 if (*tl != 0) { 1135 error = fxdr_unsigned(int, *tl); 1136 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1137 error == NFSERR_TRYLATER) { 1138 m_freem(mrep); 1139 error = 0; 1140 waituntil = time_second + trylater_delay; 1141 while (time_second < waituntil) 1142 (void) tsleep((caddr_t)&lbolt, 1143 PSOCK, "nqnfstry", 0); 1144 trylater_delay *= nfs_backoff[trylater_cnt]; 1145 if (trylater_cnt < 7) 1146 trylater_cnt++; 1147 goto tryagain; 1148 } 1149 1150 /* 1151 * If the File Handle was stale, invalidate the 1152 * lookup cache, just in case. 1153 */ 1154 if (error == ESTALE) 1155 cache_purge(vp); 1156 if (nmp->nm_flag & NFSMNT_NFSV3) { 1157 *mrp = mrep; 1158 *mdp = md; 1159 *dposp = dpos; 1160 error |= NFSERR_RETERR; 1161 } else 1162 m_freem(mrep); 1163 m_freem(rep->r_mreq); 1164 free((caddr_t)rep, M_NFSREQ); 1165 return (error); 1166 } 1167 1168 /* 1169 * For nqnfs, get any lease in reply 1170 */ 1171 if (nmp->nm_flag & NFSMNT_NQNFS) { 1172 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1173 if (*tl) { 1174 np = VTONFS(vp); 1175 nqlflag = fxdr_unsigned(int, *tl); 1176 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1177 cachable = fxdr_unsigned(int, *tl++); 1178 reqtime += fxdr_unsigned(int, *tl++); 1179 if (reqtime > time_second) { 1180 fxdr_hyper(tl, &frev); 1181 nqnfs_clientlease(nmp, np, nqlflag, 1182 cachable, reqtime, frev); 1183 } 1184 } 1185 } 1186 *mrp = mrep; 1187 *mdp = md; 1188 *dposp = dpos; 1189 m_freem(rep->r_mreq); 1190 FREE((caddr_t)rep, M_NFSREQ); 1191 return (0); 1192 } 1193 m_freem(mrep); 1194 error = EPROTONOSUPPORT; 1195 nfsmout: 1196 m_freem(rep->r_mreq); 1197 free((caddr_t)rep, M_NFSREQ); 1198 return (error); 1199 } 1200 1201 #ifndef NFS_NOSERVER 1202 /* 1203 * Generate the rpc reply header 1204 * siz arg. is used to decide if adding a cluster is worthwhile 1205 */ 1206 int 1207 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1208 int siz; 1209 struct nfsrv_descript *nd; 1210 struct nfssvc_sock *slp; 1211 int err; 1212 int cache; 1213 u_quad_t *frev; 1214 struct mbuf **mrq; 1215 struct mbuf **mbp; 1216 caddr_t *bposp; 1217 { 1218 register u_int32_t *tl; 1219 register struct mbuf *mreq; 1220 caddr_t bpos; 1221 struct mbuf *mb, *mb2; 1222 1223 MGETHDR(mreq, M_WAIT, MT_DATA); 1224 mb = mreq; 1225 /* 1226 * If this is a big reply, use a cluster else 1227 * try and leave leading space for the lower level headers. 1228 */ 1229 siz += RPC_REPLYSIZ; 1230 if (siz >= MINCLSIZE) { 1231 MCLGET(mreq, M_WAIT); 1232 } else 1233 mreq->m_data += max_hdr; 1234 tl = mtod(mreq, u_int32_t *); 1235 mreq->m_len = 6 * NFSX_UNSIGNED; 1236 bpos = ((caddr_t)tl) + mreq->m_len; 1237 *tl++ = txdr_unsigned(nd->nd_retxid); 1238 *tl++ = rpc_reply; 1239 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1240 *tl++ = rpc_msgdenied; 1241 if (err & NFSERR_AUTHERR) { 1242 *tl++ = rpc_autherr; 1243 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1244 mreq->m_len -= NFSX_UNSIGNED; 1245 bpos -= NFSX_UNSIGNED; 1246 } else { 1247 *tl++ = rpc_mismatch; 1248 *tl++ = txdr_unsigned(RPC_VER2); 1249 *tl = txdr_unsigned(RPC_VER2); 1250 } 1251 } else { 1252 *tl++ = rpc_msgaccepted; 1253 1254 /* 1255 * For Kerberos authentication, we must send the nickname 1256 * verifier back, otherwise just RPCAUTH_NULL. 1257 */ 1258 if (nd->nd_flag & ND_KERBFULL) { 1259 register struct nfsuid *nuidp; 1260 struct timeval ktvin, ktvout; 1261 1262 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; 1263 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1264 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1265 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), 1266 &nuidp->nu_haddr, nd->nd_nam2))) 1267 break; 1268 } 1269 if (nuidp) { 1270 ktvin.tv_sec = 1271 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); 1272 ktvin.tv_usec = 1273 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1274 1275 /* 1276 * Encrypt the timestamp in ecb mode using the 1277 * session key. 1278 */ 1279 #ifdef NFSKERB 1280 XXX 1281 #endif 1282 1283 *tl++ = rpc_auth_kerb; 1284 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1285 *tl = ktvout.tv_sec; 1286 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1287 *tl++ = ktvout.tv_usec; 1288 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1289 } else { 1290 *tl++ = 0; 1291 *tl++ = 0; 1292 } 1293 } else { 1294 *tl++ = 0; 1295 *tl++ = 0; 1296 } 1297 switch (err) { 1298 case EPROGUNAVAIL: 1299 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1300 break; 1301 case EPROGMISMATCH: 1302 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1303 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1304 if (nd->nd_flag & ND_NQNFS) { 1305 *tl++ = txdr_unsigned(3); 1306 *tl = txdr_unsigned(3); 1307 } else { 1308 *tl++ = txdr_unsigned(2); 1309 *tl = txdr_unsigned(3); 1310 } 1311 break; 1312 case EPROCUNAVAIL: 1313 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1314 break; 1315 case EBADRPC: 1316 *tl = txdr_unsigned(RPC_GARBAGE); 1317 break; 1318 default: 1319 *tl = 0; 1320 if (err != NFSERR_RETVOID) { 1321 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1322 if (err) 1323 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1324 else 1325 *tl = 0; 1326 } 1327 break; 1328 }; 1329 } 1330 1331 /* 1332 * For nqnfs, piggyback lease as requested. 1333 */ 1334 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1335 if (nd->nd_flag & ND_LEASE) { 1336 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1337 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1338 *tl++ = txdr_unsigned(cache); 1339 *tl++ = txdr_unsigned(nd->nd_duration); 1340 txdr_hyper(frev, tl); 1341 } else { 1342 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1343 *tl = 0; 1344 } 1345 } 1346 if (mrq != NULL) 1347 *mrq = mreq; 1348 *mbp = mb; 1349 *bposp = bpos; 1350 if (err != 0 && err != NFSERR_RETVOID) 1351 nfsstats.srvrpc_errs++; 1352 return (0); 1353 } 1354 1355 1356 #endif /* NFS_NOSERVER */ 1357 /* 1358 * Nfs timer routine 1359 * Scan the nfsreq list and retranmit any requests that have timed out 1360 * To avoid retransmission attempts on STREAM sockets (in the future) make 1361 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1362 */ 1363 void 1364 nfs_timer(arg) 1365 void *arg; /* never used */ 1366 { 1367 register struct nfsreq *rep; 1368 register struct mbuf *m; 1369 register struct socket *so; 1370 register struct nfsmount *nmp; 1371 register int timeo; 1372 int s, error; 1373 #ifndef NFS_NOSERVER 1374 static long lasttime = 0; 1375 register struct nfssvc_sock *slp; 1376 u_quad_t cur_usec; 1377 #endif /* NFS_NOSERVER */ 1378 struct proc *p = &proc0; /* XXX for credentials, will break if sleep */ 1379 1380 s = splnet(); 1381 for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) { 1382 nmp = rep->r_nmp; 1383 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1384 continue; 1385 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1386 nfs_softterm(rep); 1387 continue; 1388 } 1389 if (rep->r_rtt >= 0) { 1390 rep->r_rtt++; 1391 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1392 timeo = nmp->nm_timeo; 1393 else 1394 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1395 if (nmp->nm_timeouts > 0) 1396 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1397 if (rep->r_rtt <= timeo) 1398 continue; 1399 if (nmp->nm_timeouts < 8) 1400 nmp->nm_timeouts++; 1401 } 1402 /* 1403 * Check for server not responding 1404 */ 1405 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1406 rep->r_rexmit > nmp->nm_deadthresh) { 1407 nfs_msg(rep->r_procp, 1408 nmp->nm_mountp->mnt_stat.f_mntfromname, 1409 "not responding"); 1410 rep->r_flags |= R_TPRINTFMSG; 1411 } 1412 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1413 nfsstats.rpctimeouts++; 1414 nfs_softterm(rep); 1415 continue; 1416 } 1417 if (nmp->nm_sotype != SOCK_DGRAM) { 1418 if (++rep->r_rexmit > NFS_MAXREXMIT) 1419 rep->r_rexmit = NFS_MAXREXMIT; 1420 continue; 1421 } 1422 if ((so = nmp->nm_so) == NULL) 1423 continue; 1424 1425 /* 1426 * If there is enough space and the window allows.. 1427 * Resend it 1428 * Set r_rtt to -1 in case we fail to send it now. 1429 */ 1430 rep->r_rtt = -1; 1431 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1432 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1433 (rep->r_flags & R_SENT) || 1434 nmp->nm_sent < nmp->nm_cwnd) && 1435 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1436 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1437 error = (*so->so_proto->pr_usrreqs->pru_send) 1438 (so, 0, m, (struct sockaddr *)0, 1439 (struct mbuf *)0, p); 1440 else 1441 error = (*so->so_proto->pr_usrreqs->pru_send) 1442 (so, 0, m, nmp->nm_nam, (struct mbuf *)0, 1443 p); 1444 if (error) { 1445 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) 1446 so->so_error = 0; 1447 } else { 1448 /* 1449 * Iff first send, start timing 1450 * else turn timing off, backoff timer 1451 * and divide congestion window by 2. 1452 */ 1453 if (rep->r_flags & R_SENT) { 1454 rep->r_flags &= ~R_TIMING; 1455 if (++rep->r_rexmit > NFS_MAXREXMIT) 1456 rep->r_rexmit = NFS_MAXREXMIT; 1457 nmp->nm_cwnd >>= 1; 1458 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1459 nmp->nm_cwnd = NFS_CWNDSCALE; 1460 nfsstats.rpcretries++; 1461 } else { 1462 rep->r_flags |= R_SENT; 1463 nmp->nm_sent += NFS_CWNDSCALE; 1464 } 1465 rep->r_rtt = 0; 1466 } 1467 } 1468 } 1469 #ifndef NFS_NOSERVER 1470 /* 1471 * Call the nqnfs server timer once a second to handle leases. 1472 */ 1473 if (lasttime != time_second) { 1474 lasttime = time_second; 1475 nqnfs_serverd(); 1476 } 1477 1478 /* 1479 * Scan the write gathering queues for writes that need to be 1480 * completed now. 1481 */ 1482 cur_usec = nfs_curusec(); 1483 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 1484 slp = slp->ns_chain.tqe_next) { 1485 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec) 1486 nfsrv_wakenfsd(slp); 1487 } 1488 #endif /* NFS_NOSERVER */ 1489 splx(s); 1490 nfs_timer_handle = timeout(nfs_timer, (void *)0, nfs_ticks); 1491 } 1492 1493 /* 1494 * Flag a request as being about to terminate (due to NFSMNT_INT/NFSMNT_SOFT). 1495 * The nm_send count is decremented now to avoid deadlocks when the process in 1496 * soreceive() hasn't yet managed to send its own request. 1497 */ 1498 1499 static void 1500 nfs_softterm(rep) 1501 struct nfsreq *rep; 1502 { 1503 rep->r_flags |= R_SOFTTERM; 1504 1505 if (rep->r_flags & R_SENT) { 1506 rep->r_nmp->nm_sent -= NFS_CWNDSCALE; 1507 rep->r_flags &= ~R_SENT; 1508 } 1509 } 1510 1511 /* 1512 * Test for a termination condition pending on the process. 1513 * This is used for NFSMNT_INT mounts. 1514 */ 1515 int 1516 nfs_sigintr(nmp, rep, p) 1517 struct nfsmount *nmp; 1518 struct nfsreq *rep; 1519 register struct proc *p; 1520 { 1521 1522 if (rep && (rep->r_flags & R_SOFTTERM)) 1523 return (EINTR); 1524 if (!(nmp->nm_flag & NFSMNT_INT)) 1525 return (0); 1526 if (p && p->p_siglist && 1527 (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) & 1528 NFSINT_SIGMASK)) 1529 return (EINTR); 1530 return (0); 1531 } 1532 1533 /* 1534 * Lock a socket against others. 1535 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1536 * and also to avoid race conditions between the processes with nfs requests 1537 * in progress when a reconnect is necessary. 1538 */ 1539 int 1540 nfs_sndlock(flagp, statep, rep) 1541 register int *flagp; 1542 register int *statep; 1543 struct nfsreq *rep; 1544 { 1545 struct proc *p; 1546 int slpflag = 0, slptimeo = 0; 1547 1548 if (rep) { 1549 p = rep->r_procp; 1550 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1551 slpflag = PCATCH; 1552 } else 1553 p = (struct proc *)0; 1554 while (*statep & NFSSTA_SNDLOCK) { 1555 if (nfs_sigintr(rep->r_nmp, rep, p)) 1556 return (EINTR); 1557 *statep |= NFSSTA_WANTSND; 1558 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), 1559 "nfsndlck", slptimeo); 1560 if (slpflag == PCATCH) { 1561 slpflag = 0; 1562 slptimeo = 2 * hz; 1563 } 1564 } 1565 *statep |= NFSSTA_SNDLOCK; 1566 return (0); 1567 } 1568 1569 /* 1570 * Unlock the stream socket for others. 1571 */ 1572 void 1573 nfs_sndunlock(flagp, statep) 1574 register int *flagp; 1575 register int *statep; 1576 { 1577 1578 if ((*statep & NFSSTA_SNDLOCK) == 0) 1579 panic("nfs sndunlock"); 1580 *statep &= ~NFSSTA_SNDLOCK; 1581 if (*statep & NFSSTA_WANTSND) { 1582 *statep &= ~NFSSTA_WANTSND; 1583 wakeup((caddr_t)flagp); 1584 } 1585 } 1586 1587 static int 1588 nfs_rcvlock(rep) 1589 register struct nfsreq *rep; 1590 { 1591 register int *flagp = &rep->r_nmp->nm_flag; 1592 register int *statep = &rep->r_nmp->nm_state; 1593 int slpflag, slptimeo = 0; 1594 1595 if (*flagp & NFSMNT_INT) 1596 slpflag = PCATCH; 1597 else 1598 slpflag = 0; 1599 while (*statep & NFSSTA_RCVLOCK) { 1600 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1601 return (EINTR); 1602 *statep |= NFSSTA_WANTRCV; 1603 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1604 slptimeo); 1605 /* 1606 * If our reply was recieved while we were sleeping, 1607 * then just return without taking the lock to avoid a 1608 * situation where a single iod could 'capture' the 1609 * recieve lock. 1610 */ 1611 if (rep->r_mrep != NULL) 1612 return (EALREADY); 1613 if (slpflag == PCATCH) { 1614 slpflag = 0; 1615 slptimeo = 2 * hz; 1616 } 1617 } 1618 *statep |= NFSSTA_RCVLOCK; 1619 return (0); 1620 } 1621 1622 /* 1623 * Unlock the stream socket for others. 1624 */ 1625 static void 1626 nfs_rcvunlock(flagp, statep) 1627 register int *flagp; 1628 register int *statep; 1629 { 1630 1631 if ((*statep & NFSSTA_RCVLOCK) == 0) 1632 panic("nfs rcvunlock"); 1633 *statep &= ~NFSSTA_RCVLOCK; 1634 if (*statep & NFSSTA_WANTRCV) { 1635 *statep &= ~NFSSTA_WANTRCV; 1636 wakeup((caddr_t)flagp); 1637 } 1638 } 1639 1640 /* 1641 * nfs_realign: 1642 * 1643 * Check for badly aligned mbuf data and realign by copying the unaligned 1644 * portion of the data into a new mbuf chain and freeing the portions 1645 * of the old chain that were replaced. 1646 * 1647 * We cannot simply realign the data within the existing mbuf chain 1648 * because the underlying buffers may contain other rpc commands and 1649 * we cannot afford to overwrite them. 1650 * 1651 * We would prefer to avoid this situation entirely. The situation does 1652 * not occur with NFS/UDP and is supposed to only occasionally occur 1653 * with TCP. 1654 */ 1655 static void 1656 nfs_realign(pm, hsiz) 1657 register struct mbuf **pm; 1658 int hsiz; 1659 { 1660 struct mbuf *m; 1661 struct mbuf *n = NULL; 1662 int off = 0; 1663 1664 while ((m = *pm) != NULL) { 1665 if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) { 1666 MGET(n, M_WAIT, MT_DATA); 1667 if (m->m_len >= MINCLSIZE) { 1668 MCLGET(n, M_WAIT); 1669 } 1670 n->m_len = 0; 1671 break; 1672 } 1673 pm = &m->m_next; 1674 } 1675 1676 /* 1677 * If n is non-NULL, loop on m copying data, then replace the 1678 * portion of the chain that had to be realigned. 1679 */ 1680 if (n != NULL) { 1681 while (m) { 1682 m_copyback(n, off, m->m_len, mtod(m, caddr_t)); 1683 off += m->m_len; 1684 m = m->m_next; 1685 } 1686 m_freem(*pm); 1687 *pm = n; 1688 } 1689 } 1690 1691 #ifndef NFS_NOSERVER 1692 1693 /* 1694 * Parse an RPC request 1695 * - verify it 1696 * - fill in the cred struct. 1697 */ 1698 int 1699 nfs_getreq(nd, nfsd, has_header) 1700 register struct nfsrv_descript *nd; 1701 struct nfsd *nfsd; 1702 int has_header; 1703 { 1704 register int len, i; 1705 register u_int32_t *tl; 1706 register int32_t t1; 1707 struct uio uio; 1708 struct iovec iov; 1709 caddr_t dpos, cp2, cp; 1710 u_int32_t nfsvers, auth_type; 1711 uid_t nickuid; 1712 int error = 0, nqnfs = 0, ticklen; 1713 struct mbuf *mrep, *md; 1714 register struct nfsuid *nuidp; 1715 struct timeval tvin, tvout; 1716 #if 0 /* until encrypted keys are implemented */ 1717 NFSKERBKEYSCHED_T keys; /* stores key schedule */ 1718 #endif 1719 1720 mrep = nd->nd_mrep; 1721 md = nd->nd_md; 1722 dpos = nd->nd_dpos; 1723 if (has_header) { 1724 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1725 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1726 if (*tl++ != rpc_call) { 1727 m_freem(mrep); 1728 return (EBADRPC); 1729 } 1730 } else 1731 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1732 nd->nd_repstat = 0; 1733 nd->nd_flag = 0; 1734 if (*tl++ != rpc_vers) { 1735 nd->nd_repstat = ERPCMISMATCH; 1736 nd->nd_procnum = NFSPROC_NOOP; 1737 return (0); 1738 } 1739 if (*tl != nfs_prog) { 1740 if (*tl == nqnfs_prog) 1741 nqnfs++; 1742 else { 1743 nd->nd_repstat = EPROGUNAVAIL; 1744 nd->nd_procnum = NFSPROC_NOOP; 1745 return (0); 1746 } 1747 } 1748 tl++; 1749 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1750 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1751 (nfsvers != NQNFS_VER3 && nqnfs)) { 1752 nd->nd_repstat = EPROGMISMATCH; 1753 nd->nd_procnum = NFSPROC_NOOP; 1754 return (0); 1755 } 1756 if (nqnfs) 1757 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1758 else if (nfsvers == NFS_VER3) 1759 nd->nd_flag = ND_NFSV3; 1760 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1761 if (nd->nd_procnum == NFSPROC_NULL) 1762 return (0); 1763 if (nd->nd_procnum >= NFS_NPROCS || 1764 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1765 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1766 nd->nd_repstat = EPROCUNAVAIL; 1767 nd->nd_procnum = NFSPROC_NOOP; 1768 return (0); 1769 } 1770 if ((nd->nd_flag & ND_NFSV3) == 0) 1771 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1772 auth_type = *tl++; 1773 len = fxdr_unsigned(int, *tl++); 1774 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1775 m_freem(mrep); 1776 return (EBADRPC); 1777 } 1778 1779 nd->nd_flag &= ~ND_KERBAUTH; 1780 /* 1781 * Handle auth_unix or auth_kerb. 1782 */ 1783 if (auth_type == rpc_auth_unix) { 1784 len = fxdr_unsigned(int, *++tl); 1785 if (len < 0 || len > NFS_MAXNAMLEN) { 1786 m_freem(mrep); 1787 return (EBADRPC); 1788 } 1789 nfsm_adv(nfsm_rndup(len)); 1790 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1791 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred)); 1792 nd->nd_cr.cr_ref = 1; 1793 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1794 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1795 len = fxdr_unsigned(int, *tl); 1796 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1797 m_freem(mrep); 1798 return (EBADRPC); 1799 } 1800 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1801 for (i = 1; i <= len; i++) 1802 if (i < NGROUPS) 1803 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1804 else 1805 tl++; 1806 nd->nd_cr.cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1); 1807 if (nd->nd_cr.cr_ngroups > 1) 1808 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1809 len = fxdr_unsigned(int, *++tl); 1810 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1811 m_freem(mrep); 1812 return (EBADRPC); 1813 } 1814 if (len > 0) 1815 nfsm_adv(nfsm_rndup(len)); 1816 } else if (auth_type == rpc_auth_kerb) { 1817 switch (fxdr_unsigned(int, *tl++)) { 1818 case RPCAKN_FULLNAME: 1819 ticklen = fxdr_unsigned(int, *tl); 1820 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1821 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1822 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1823 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1824 m_freem(mrep); 1825 return (EBADRPC); 1826 } 1827 uio.uio_offset = 0; 1828 uio.uio_iov = &iov; 1829 uio.uio_iovcnt = 1; 1830 uio.uio_segflg = UIO_SYSSPACE; 1831 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1832 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1833 nfsm_mtouio(&uio, uio.uio_resid); 1834 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1835 if (*tl++ != rpc_auth_kerb || 1836 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1837 printf("Bad kerb verifier\n"); 1838 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1839 nd->nd_procnum = NFSPROC_NOOP; 1840 return (0); 1841 } 1842 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1843 tl = (u_int32_t *)cp; 1844 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1845 printf("Not fullname kerb verifier\n"); 1846 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1847 nd->nd_procnum = NFSPROC_NOOP; 1848 return (0); 1849 } 1850 cp += NFSX_UNSIGNED; 1851 bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED); 1852 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1853 nd->nd_flag |= ND_KERBFULL; 1854 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1855 break; 1856 case RPCAKN_NICKNAME: 1857 if (len != 2 * NFSX_UNSIGNED) { 1858 printf("Kerb nickname short\n"); 1859 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1860 nd->nd_procnum = NFSPROC_NOOP; 1861 return (0); 1862 } 1863 nickuid = fxdr_unsigned(uid_t, *tl); 1864 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1865 if (*tl++ != rpc_auth_kerb || 1866 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1867 printf("Kerb nick verifier bad\n"); 1868 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1869 nd->nd_procnum = NFSPROC_NOOP; 1870 return (0); 1871 } 1872 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1873 tvin.tv_sec = *tl++; 1874 tvin.tv_usec = *tl; 1875 1876 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; 1877 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1878 if (nuidp->nu_cr.cr_uid == nickuid && 1879 (!nd->nd_nam2 || 1880 netaddr_match(NU_NETFAM(nuidp), 1881 &nuidp->nu_haddr, nd->nd_nam2))) 1882 break; 1883 } 1884 if (!nuidp) { 1885 nd->nd_repstat = 1886 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1887 nd->nd_procnum = NFSPROC_NOOP; 1888 return (0); 1889 } 1890 1891 /* 1892 * Now, decrypt the timestamp using the session key 1893 * and validate it. 1894 */ 1895 #ifdef NFSKERB 1896 XXX 1897 #endif 1898 1899 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1900 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1901 if (nuidp->nu_expire < time_second || 1902 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1903 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1904 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1905 nuidp->nu_expire = 0; 1906 nd->nd_repstat = 1907 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1908 nd->nd_procnum = NFSPROC_NOOP; 1909 return (0); 1910 } 1911 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1912 nd->nd_flag |= ND_KERBNICK; 1913 }; 1914 } else { 1915 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1916 nd->nd_procnum = NFSPROC_NOOP; 1917 return (0); 1918 } 1919 1920 /* 1921 * For nqnfs, get piggybacked lease request. 1922 */ 1923 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1924 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1925 nd->nd_flag |= fxdr_unsigned(int, *tl); 1926 if (nd->nd_flag & ND_LEASE) { 1927 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1928 nd->nd_duration = fxdr_unsigned(int32_t, *tl); 1929 } else 1930 nd->nd_duration = NQ_MINLEASE; 1931 } else 1932 nd->nd_duration = NQ_MINLEASE; 1933 nd->nd_md = md; 1934 nd->nd_dpos = dpos; 1935 return (0); 1936 nfsmout: 1937 return (error); 1938 } 1939 1940 #endif 1941 1942 static int 1943 nfs_msg(p, server, msg) 1944 struct proc *p; 1945 char *server, *msg; 1946 { 1947 tpr_t tpr; 1948 1949 if (p) 1950 tpr = tprintf_open(p); 1951 else 1952 tpr = NULL; 1953 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1954 tprintf_close(tpr); 1955 return (0); 1956 } 1957 1958 #ifndef NFS_NOSERVER 1959 /* 1960 * Socket upcall routine for the nfsd sockets. 1961 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1962 * Essentially do as much as possible non-blocking, else punt and it will 1963 * be called with M_WAIT from an nfsd. 1964 */ 1965 void 1966 nfsrv_rcv(so, arg, waitflag) 1967 struct socket *so; 1968 void *arg; 1969 int waitflag; 1970 { 1971 register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1972 register struct mbuf *m; 1973 struct mbuf *mp; 1974 struct sockaddr *nam; 1975 struct uio auio; 1976 int flags, error; 1977 1978 if ((slp->ns_flag & SLP_VALID) == 0) 1979 return; 1980 #ifdef notdef 1981 /* 1982 * Define this to test for nfsds handling this under heavy load. 1983 */ 1984 if (waitflag == M_DONTWAIT) { 1985 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1986 } 1987 #endif 1988 auio.uio_procp = NULL; 1989 if (so->so_type == SOCK_STREAM) { 1990 /* 1991 * If there are already records on the queue, defer soreceive() 1992 * to an nfsd so that there is feedback to the TCP layer that 1993 * the nfs servers are heavily loaded. 1994 */ 1995 if (STAILQ_FIRST(&slp->ns_rec) && waitflag == M_DONTWAIT) { 1996 slp->ns_flag |= SLP_NEEDQ; 1997 goto dorecs; 1998 } 1999 2000 /* 2001 * Do soreceive(). 2002 */ 2003 auio.uio_resid = 1000000000; 2004 flags = MSG_DONTWAIT; 2005 error = so->so_proto->pr_usrreqs->pru_soreceive 2006 (so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 2007 if (error || mp == (struct mbuf *)0) { 2008 if (error == EWOULDBLOCK) 2009 slp->ns_flag |= SLP_NEEDQ; 2010 else 2011 slp->ns_flag |= SLP_DISCONN; 2012 goto dorecs; 2013 } 2014 m = mp; 2015 if (slp->ns_rawend) { 2016 slp->ns_rawend->m_next = m; 2017 slp->ns_cc += 1000000000 - auio.uio_resid; 2018 } else { 2019 slp->ns_raw = m; 2020 slp->ns_cc = 1000000000 - auio.uio_resid; 2021 } 2022 while (m->m_next) 2023 m = m->m_next; 2024 slp->ns_rawend = m; 2025 2026 /* 2027 * Now try and parse record(s) out of the raw stream data. 2028 */ 2029 error = nfsrv_getstream(slp, waitflag); 2030 if (error) { 2031 if (error == EPERM) 2032 slp->ns_flag |= SLP_DISCONN; 2033 else 2034 slp->ns_flag |= SLP_NEEDQ; 2035 } 2036 } else { 2037 do { 2038 auio.uio_resid = 1000000000; 2039 flags = MSG_DONTWAIT; 2040 error = so->so_proto->pr_usrreqs->pru_soreceive 2041 (so, &nam, &auio, &mp, 2042 (struct mbuf **)0, &flags); 2043 if (mp) { 2044 struct nfsrv_rec *rec; 2045 rec = malloc(sizeof(struct nfsrv_rec), 2046 M_NFSRVDESC, waitflag); 2047 if (!rec) { 2048 if (nam) 2049 FREE(nam, M_SONAME); 2050 m_freem(mp); 2051 continue; 2052 } 2053 nfs_realign(&mp, 10 * NFSX_UNSIGNED); 2054 rec->nr_address = nam; 2055 rec->nr_packet = mp; 2056 STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link); 2057 } 2058 if (error) { 2059 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2060 && error != EWOULDBLOCK) { 2061 slp->ns_flag |= SLP_DISCONN; 2062 goto dorecs; 2063 } 2064 } 2065 } while (mp); 2066 } 2067 2068 /* 2069 * Now try and process the request records, non-blocking. 2070 */ 2071 dorecs: 2072 if (waitflag == M_DONTWAIT && 2073 (STAILQ_FIRST(&slp->ns_rec) 2074 || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 2075 nfsrv_wakenfsd(slp); 2076 } 2077 2078 /* 2079 * Try and extract an RPC request from the mbuf data list received on a 2080 * stream socket. The "waitflag" argument indicates whether or not it 2081 * can sleep. 2082 */ 2083 static int 2084 nfsrv_getstream(slp, waitflag) 2085 register struct nfssvc_sock *slp; 2086 int waitflag; 2087 { 2088 register struct mbuf *m, **mpp; 2089 register char *cp1, *cp2; 2090 register int len; 2091 struct mbuf *om, *m2, *recm = NULL; 2092 u_int32_t recmark; 2093 2094 if (slp->ns_flag & SLP_GETSTREAM) 2095 panic("nfs getstream"); 2096 slp->ns_flag |= SLP_GETSTREAM; 2097 for (;;) { 2098 if (slp->ns_reclen == 0) { 2099 if (slp->ns_cc < NFSX_UNSIGNED) { 2100 slp->ns_flag &= ~SLP_GETSTREAM; 2101 return (0); 2102 } 2103 m = slp->ns_raw; 2104 if (m->m_len >= NFSX_UNSIGNED) { 2105 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 2106 m->m_data += NFSX_UNSIGNED; 2107 m->m_len -= NFSX_UNSIGNED; 2108 } else { 2109 cp1 = (caddr_t)&recmark; 2110 cp2 = mtod(m, caddr_t); 2111 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 2112 while (m->m_len == 0) { 2113 m = m->m_next; 2114 cp2 = mtod(m, caddr_t); 2115 } 2116 *cp1++ = *cp2++; 2117 m->m_data++; 2118 m->m_len--; 2119 } 2120 } 2121 slp->ns_cc -= NFSX_UNSIGNED; 2122 recmark = ntohl(recmark); 2123 slp->ns_reclen = recmark & ~0x80000000; 2124 if (recmark & 0x80000000) 2125 slp->ns_flag |= SLP_LASTFRAG; 2126 else 2127 slp->ns_flag &= ~SLP_LASTFRAG; 2128 if (slp->ns_reclen > NFS_MAXPACKET) { 2129 slp->ns_flag &= ~SLP_GETSTREAM; 2130 return (EPERM); 2131 } 2132 } 2133 2134 /* 2135 * Now get the record part. 2136 */ 2137 if (slp->ns_cc == slp->ns_reclen) { 2138 recm = slp->ns_raw; 2139 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2140 slp->ns_cc = slp->ns_reclen = 0; 2141 } else if (slp->ns_cc > slp->ns_reclen) { 2142 len = 0; 2143 m = slp->ns_raw; 2144 om = (struct mbuf *)0; 2145 while (len < slp->ns_reclen) { 2146 if ((len + m->m_len) > slp->ns_reclen) { 2147 m2 = m_copym(m, 0, slp->ns_reclen - len, 2148 waitflag); 2149 if (m2) { 2150 if (om) { 2151 om->m_next = m2; 2152 recm = slp->ns_raw; 2153 } else 2154 recm = m2; 2155 m->m_data += slp->ns_reclen - len; 2156 m->m_len -= slp->ns_reclen - len; 2157 len = slp->ns_reclen; 2158 } else { 2159 slp->ns_flag &= ~SLP_GETSTREAM; 2160 return (EWOULDBLOCK); 2161 } 2162 } else if ((len + m->m_len) == slp->ns_reclen) { 2163 om = m; 2164 len += m->m_len; 2165 m = m->m_next; 2166 recm = slp->ns_raw; 2167 om->m_next = (struct mbuf *)0; 2168 } else { 2169 om = m; 2170 len += m->m_len; 2171 m = m->m_next; 2172 } 2173 } 2174 slp->ns_raw = m; 2175 slp->ns_cc -= len; 2176 slp->ns_reclen = 0; 2177 } else { 2178 slp->ns_flag &= ~SLP_GETSTREAM; 2179 return (0); 2180 } 2181 2182 /* 2183 * Accumulate the fragments into a record. 2184 */ 2185 mpp = &slp->ns_frag; 2186 while (*mpp) 2187 mpp = &((*mpp)->m_next); 2188 *mpp = recm; 2189 if (slp->ns_flag & SLP_LASTFRAG) { 2190 struct nfsrv_rec *rec; 2191 rec = malloc(sizeof(struct nfsrv_rec), M_NFSRVDESC, waitflag); 2192 if (!rec) { 2193 m_freem(slp->ns_frag); 2194 } else { 2195 nfs_realign(&slp->ns_frag, 10 * NFSX_UNSIGNED); 2196 rec->nr_address = (struct sockaddr *)0; 2197 rec->nr_packet = slp->ns_frag; 2198 STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link); 2199 } 2200 slp->ns_frag = (struct mbuf *)0; 2201 } 2202 } 2203 } 2204 2205 /* 2206 * Parse an RPC header. 2207 */ 2208 int 2209 nfsrv_dorec(slp, nfsd, ndp) 2210 register struct nfssvc_sock *slp; 2211 struct nfsd *nfsd; 2212 struct nfsrv_descript **ndp; 2213 { 2214 struct nfsrv_rec *rec; 2215 register struct mbuf *m; 2216 struct sockaddr *nam; 2217 register struct nfsrv_descript *nd; 2218 int error; 2219 2220 *ndp = NULL; 2221 if ((slp->ns_flag & SLP_VALID) == 0 || !STAILQ_FIRST(&slp->ns_rec)) 2222 return (ENOBUFS); 2223 rec = STAILQ_FIRST(&slp->ns_rec); 2224 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link); 2225 nam = rec->nr_address; 2226 m = rec->nr_packet; 2227 free(rec, M_NFSRVDESC); 2228 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), 2229 M_NFSRVDESC, M_WAITOK); 2230 nd->nd_md = nd->nd_mrep = m; 2231 nd->nd_nam2 = nam; 2232 nd->nd_dpos = mtod(m, caddr_t); 2233 error = nfs_getreq(nd, nfsd, TRUE); 2234 if (error) { 2235 if (nam) { 2236 FREE(nam, M_SONAME); 2237 } 2238 free((caddr_t)nd, M_NFSRVDESC); 2239 return (error); 2240 } 2241 *ndp = nd; 2242 nfsd->nfsd_nd = nd; 2243 return (0); 2244 } 2245 2246 /* 2247 * Search for a sleeping nfsd and wake it up. 2248 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2249 * running nfsds will go look for the work in the nfssvc_sock list. 2250 */ 2251 void 2252 nfsrv_wakenfsd(slp) 2253 struct nfssvc_sock *slp; 2254 { 2255 register struct nfsd *nd; 2256 2257 if ((slp->ns_flag & SLP_VALID) == 0) 2258 return; 2259 for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) { 2260 if (nd->nfsd_flag & NFSD_WAITING) { 2261 nd->nfsd_flag &= ~NFSD_WAITING; 2262 if (nd->nfsd_slp) 2263 panic("nfsd wakeup"); 2264 slp->ns_sref++; 2265 nd->nfsd_slp = slp; 2266 wakeup((caddr_t)nd); 2267 return; 2268 } 2269 } 2270 slp->ns_flag |= SLP_DOREC; 2271 nfsd_head_flag |= NFSD_CHECKSLP; 2272 } 2273 #endif /* NFS_NOSERVER */
Cache object: 7904dc73127de27450906e7da0536d33
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]