1 /* $OpenBSD: nfs_socket.c,v 1.96 2009/10/19 22:24:18 jsg Exp $ */ 2 /* $NetBSD: nfs_socket.c,v 1.27 1996/04/15 20:20:00 thorpej Exp $ */ 3 4 /* 5 * Copyright (c) 1989, 1991, 1993, 1995 6 * The Regents of the University of California. All rights reserved. 7 * 8 * This code is derived from software contributed to Berkeley by 9 * Rick Macklem at The University of Guelph. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 36 */ 37 38 /* 39 * Socket operations for use by nfs 40 */ 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/proc.h> 45 #include <sys/mount.h> 46 #include <sys/kernel.h> 47 #include <sys/mbuf.h> 48 #include <sys/vnode.h> 49 #include <sys/domain.h> 50 #include <sys/protosw.h> 51 #include <sys/socket.h> 52 #include <sys/socketvar.h> 53 #include <sys/syslog.h> 54 #include <sys/tprintf.h> 55 #include <sys/namei.h> 56 #include <sys/pool.h> 57 #include <sys/queue.h> 58 59 #include <netinet/in.h> 60 #include <netinet/tcp.h> 61 62 #include <nfs/rpcv2.h> 63 #include <nfs/nfsproto.h> 64 #include <nfs/nfs.h> 65 #include <nfs/xdr_subs.h> 66 #include <nfs/nfsm_subs.h> 67 #include <nfs/nfsmount.h> 68 #include <nfs/nfsnode.h> 69 #include <nfs/nfs_var.h> 70 71 /* External data, mostly RPC constants in XDR form. */ 72 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 73 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr; 74 extern u_int32_t nfs_prog; 75 extern struct nfsstats nfsstats; 76 extern int nfsv3_procid[NFS_NPROCS]; 77 extern int nfs_ticks; 78 79 extern struct pool nfsrv_descript_pl; 80 81 /* 82 * There is a congestion window for outstanding rpcs maintained per mount 83 * point. The cwnd size is adjusted in roughly the way that: 84 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 85 * SIGCOMM '88". ACM, August 1988. 86 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 87 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 88 * of rpcs is in progress. 89 * (The sent count and cwnd are scaled for integer arith.) 90 * Variants of "slow start" were tried and were found to be too much of a 91 * performance hit (ave. rtt 3 times larger), 92 * I suspect due to the large rtt that nfs rpcs have. 93 */ 94 #define NFS_CWNDSCALE 256 95 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 96 int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256 }; 97 98 /* RTT estimator */ 99 enum nfs_rto_timers nfs_ptimers[NFS_NPROCS] = { 100 NFS_DEFAULT_TIMER, /* NULL */ 101 NFS_GETATTR_TIMER, /* GETATTR */ 102 NFS_DEFAULT_TIMER, /* SETATTR */ 103 NFS_LOOKUP_TIMER, /* LOOKUP */ 104 NFS_GETATTR_TIMER, /* ACCESS */ 105 NFS_READ_TIMER, /* READLINK */ 106 NFS_READ_TIMER, /* READ */ 107 NFS_WRITE_TIMER, /* WRITE */ 108 NFS_DEFAULT_TIMER, /* CREATE */ 109 NFS_DEFAULT_TIMER, /* MKDIR */ 110 NFS_DEFAULT_TIMER, /* SYMLINK */ 111 NFS_DEFAULT_TIMER, /* MKNOD */ 112 NFS_DEFAULT_TIMER, /* REMOVE */ 113 NFS_DEFAULT_TIMER, /* RMDIR */ 114 NFS_DEFAULT_TIMER, /* RENAME */ 115 NFS_DEFAULT_TIMER, /* LINK */ 116 NFS_READ_TIMER, /* READDIR */ 117 NFS_READ_TIMER, /* READDIRPLUS */ 118 NFS_DEFAULT_TIMER, /* FSSTAT */ 119 NFS_DEFAULT_TIMER, /* FSINFO */ 120 NFS_DEFAULT_TIMER, /* PATHCONF */ 121 NFS_DEFAULT_TIMER, /* COMMIT */ 122 NFS_DEFAULT_TIMER, /* NOOP */ 123 }; 124 125 void nfs_init_rtt(struct nfsmount *); 126 void nfs_update_rtt(struct nfsreq *); 127 int nfs_estimate_rto(struct nfsmount *, u_int32_t procnum); 128 129 void nfs_realign(struct mbuf **, int); 130 void nfs_realign_fixup(struct mbuf *, struct mbuf *, unsigned int *); 131 unsigned int nfs_realign_test = 0; 132 unsigned int nfs_realign_count = 0; 133 134 /* Initialize the RTT estimator state for a new mount point. */ 135 void 136 nfs_init_rtt(struct nfsmount *nmp) 137 { 138 int i; 139 140 for (i = 0; i < NFS_MAX_TIMER; i++) 141 nmp->nm_srtt[i] = NFS_INITRTT; 142 for (i = 0; i < NFS_MAX_TIMER; i++) 143 nmp->nm_sdrtt[i] = 0; 144 } 145 146 /* 147 * Update a mount point's RTT estimator state using data from the 148 * passed-in request. 149 * 150 * Use a gain of 0.125 on the mean and a gain of 0.25 on the deviation. 151 * 152 * NB: Since the timer resolution of NFS_HZ is so course, it can often 153 * result in r_rtt == 0. Since r_rtt == N means that the actual RTT is 154 * between N + dt and N + 2 - dt ticks, add 1 before calculating the 155 * update values. 156 */ 157 void 158 nfs_update_rtt(struct nfsreq *rep) 159 { 160 int t1 = rep->r_rtt + 1; 161 int index = nfs_ptimers[rep->r_procnum] - 1; 162 int *srtt = &rep->r_nmp->nm_srtt[index]; 163 int *sdrtt = &rep->r_nmp->nm_sdrtt[index]; 164 165 t1 -= *srtt >> 3; 166 *srtt += t1; 167 if (t1 < 0) 168 t1 = -t1; 169 t1 -= *sdrtt >> 2; 170 *sdrtt += t1; 171 } 172 173 /* 174 * Estimate RTO for an NFS RPC sent via an unreliable datagram. 175 * 176 * Use the mean and mean deviation of RTT for the appropriate type 177 * of RPC for the frequent RPCs and a default for the others. 178 * The justification for doing "other" this way is that these RPCs 179 * happen so infrequently that timer est. would probably be stale. 180 * Also, since many of these RPCs are non-idempotent, a conservative 181 * timeout is desired. 182 * 183 * getattr, lookup - A+2D 184 * read, write - A+4D 185 * other - nm_timeo 186 */ 187 int 188 nfs_estimate_rto(struct nfsmount *nmp, u_int32_t procnum) 189 { 190 enum nfs_rto_timers timer = nfs_ptimers[procnum]; 191 int index = timer - 1; 192 int rto; 193 194 switch (timer) { 195 case NFS_GETATTR_TIMER: 196 case NFS_LOOKUP_TIMER: 197 rto = ((nmp->nm_srtt[index] + 3) >> 2) + 198 ((nmp->nm_sdrtt[index] + 1) >> 1); 199 break; 200 case NFS_READ_TIMER: 201 case NFS_WRITE_TIMER: 202 rto = ((nmp->nm_srtt[index] + 7) >> 3) + 203 (nmp->nm_sdrtt[index] + 1); 204 break; 205 default: 206 rto = nmp->nm_timeo; 207 return (rto); 208 } 209 210 if (rto < NFS_MINRTO) 211 rto = NFS_MINRTO; 212 else if (rto > NFS_MAXRTO) 213 rto = NFS_MAXRTO; 214 215 return (rto); 216 } 217 218 219 220 /* 221 * Initialize sockets and congestion for a new NFS connection. 222 * We do not free the sockaddr if error. 223 */ 224 int 225 nfs_connect(struct nfsmount *nmp, struct nfsreq *rep) 226 { 227 struct socket *so; 228 int s, error, rcvreserve, sndreserve; 229 struct sockaddr *saddr; 230 struct sockaddr_in *sin; 231 struct mbuf *m; 232 233 nmp->nm_so = NULL; 234 saddr = mtod(nmp->nm_nam, struct sockaddr *); 235 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 236 nmp->nm_soproto); 237 if (error) 238 goto bad; 239 so = nmp->nm_so; 240 nmp->nm_soflags = so->so_proto->pr_flags; 241 242 /* 243 * Some servers require that the client port be a reserved port number. 244 * We always allocate a reserved port, as this prevents filehandle 245 * disclosure through UDP port capture. 246 */ 247 if (saddr->sa_family == AF_INET) { 248 struct mbuf *mopt; 249 int *ip; 250 251 MGET(mopt, M_WAIT, MT_SOOPTS); 252 mopt->m_len = sizeof(int); 253 ip = mtod(mopt, int *); 254 *ip = IP_PORTRANGE_LOW; 255 error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt); 256 if (error) 257 goto bad; 258 259 MGET(m, M_WAIT, MT_SONAME); 260 sin = mtod(m, struct sockaddr_in *); 261 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 262 sin->sin_family = AF_INET; 263 sin->sin_addr.s_addr = INADDR_ANY; 264 sin->sin_port = htons(0); 265 error = sobind(so, m, &proc0); 266 m_freem(m); 267 if (error) 268 goto bad; 269 270 MGET(mopt, M_WAIT, MT_SOOPTS); 271 mopt->m_len = sizeof(int); 272 ip = mtod(mopt, int *); 273 *ip = IP_PORTRANGE_DEFAULT; 274 error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt); 275 if (error) 276 goto bad; 277 } 278 279 /* 280 * Protocols that do not require connections may be optionally left 281 * unconnected for servers that reply from a port other than NFS_PORT. 282 */ 283 if (nmp->nm_flag & NFSMNT_NOCONN) { 284 if (nmp->nm_soflags & PR_CONNREQUIRED) { 285 error = ENOTCONN; 286 goto bad; 287 } 288 } else { 289 error = soconnect(so, nmp->nm_nam); 290 if (error) 291 goto bad; 292 293 /* 294 * Wait for the connection to complete. Cribbed from the 295 * connect system call but with the wait timing out so 296 * that interruptible mounts don't hang here for a long time. 297 */ 298 s = splsoftnet(); 299 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 300 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 301 "nfscon", 2 * hz); 302 if ((so->so_state & SS_ISCONNECTING) && 303 so->so_error == 0 && rep && 304 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ 305 so->so_state &= ~SS_ISCONNECTING; 306 splx(s); 307 goto bad; 308 } 309 } 310 if (so->so_error) { 311 error = so->so_error; 312 so->so_error = 0; 313 splx(s); 314 goto bad; 315 } 316 splx(s); 317 } 318 /* 319 * Always set receive timeout to detect server crash and reconnect. 320 * Otherwise, we can get stuck in soreceive forever. 321 */ 322 so->so_rcv.sb_timeo = (5 * hz); 323 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) 324 so->so_snd.sb_timeo = (5 * hz); 325 else 326 so->so_snd.sb_timeo = 0; 327 if (nmp->nm_sotype == SOCK_DGRAM) { 328 sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR; 329 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 330 NFS_MAXPKTHDR) * 2; 331 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 332 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 333 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 334 NFS_MAXPKTHDR) * 2; 335 } else { 336 if (nmp->nm_sotype != SOCK_STREAM) 337 panic("nfscon sotype"); 338 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 339 MGET(m, M_WAIT, MT_SOOPTS); 340 *mtod(m, int32_t *) = 1; 341 m->m_len = sizeof(int32_t); 342 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 343 } 344 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 345 MGET(m, M_WAIT, MT_SOOPTS); 346 *mtod(m, int32_t *) = 1; 347 m->m_len = sizeof(int32_t); 348 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 349 } 350 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 351 sizeof (u_int32_t)) * 2; 352 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 353 sizeof (u_int32_t)) * 2; 354 } 355 error = soreserve(so, sndreserve, rcvreserve); 356 if (error) 357 goto bad; 358 so->so_rcv.sb_flags |= SB_NOINTR; 359 so->so_snd.sb_flags |= SB_NOINTR; 360 361 /* Initialize other non-zero congestion variables */ 362 nfs_init_rtt(nmp); 363 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 364 nmp->nm_sent = 0; 365 nmp->nm_timeouts = 0; 366 return (0); 367 368 bad: 369 nfs_disconnect(nmp); 370 return (error); 371 } 372 373 /* 374 * Reconnect routine: 375 * Called when a connection is broken on a reliable protocol. 376 * - clean up the old socket 377 * - nfs_connect() again 378 * - set R_MUSTRESEND for all outstanding requests on mount point 379 * If this fails the mount point is DEAD! 380 * nb: Must be called with the nfs_sndlock() set on the mount point. 381 */ 382 int 383 nfs_reconnect(struct nfsreq *rep) 384 { 385 struct nfsreq *rp; 386 struct nfsmount *nmp = rep->r_nmp; 387 int s, error; 388 389 nfs_disconnect(nmp); 390 while ((error = nfs_connect(nmp, rep)) != 0) { 391 if (error == EINTR || error == ERESTART) 392 return (EINTR); 393 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfsrecon", 0); 394 } 395 396 /* 397 * Loop through outstanding request list and fix up all requests 398 * on old socket. 399 */ 400 s = splsoftnet(); 401 TAILQ_FOREACH(rp, &nmp->nm_reqsq, r_chain) { 402 rp->r_flags |= R_MUSTRESEND; 403 rp->r_rexmit = 0; 404 } 405 splx(s); 406 return (0); 407 } 408 409 /* 410 * NFS disconnect. Clean up and unlink. 411 */ 412 void 413 nfs_disconnect(struct nfsmount *nmp) 414 { 415 struct socket *so; 416 417 if (nmp->nm_so) { 418 so = nmp->nm_so; 419 nmp->nm_so = NULL; 420 soshutdown(so, SHUT_RDWR); 421 soclose(so); 422 } 423 } 424 425 /* 426 * This is the nfs send routine. For connection based socket types, it 427 * must be called with an nfs_sndlock() on the socket. 428 * "rep == NULL" indicates that it has been called from a server. 429 * For the client side: 430 * - return EINTR if the RPC is terminated, 0 otherwise 431 * - set R_MUSTRESEND if the send fails for any reason 432 * - do any cleanup required by recoverable socket errors (???) 433 * For the server side: 434 * - return EINTR or ERESTART if interrupted by a signal 435 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 436 * - do any cleanup required by recoverable socket errors (???) 437 */ 438 int 439 nfs_send(struct socket *so, struct mbuf *nam, struct mbuf *top, 440 struct nfsreq *rep) 441 { 442 struct mbuf *sendnam; 443 int error, soflags, flags; 444 445 if (rep) { 446 if (rep->r_flags & R_SOFTTERM) { 447 m_freem(top); 448 return (EINTR); 449 } 450 if ((so = rep->r_nmp->nm_so) == NULL) { 451 rep->r_flags |= R_MUSTRESEND; 452 m_freem(top); 453 return (0); 454 } 455 rep->r_flags &= ~R_MUSTRESEND; 456 soflags = rep->r_nmp->nm_soflags; 457 } else 458 soflags = so->so_proto->pr_flags; 459 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 460 sendnam = NULL; 461 else 462 sendnam = nam; 463 if (so->so_type == SOCK_SEQPACKET) 464 flags = MSG_EOR; 465 else 466 flags = 0; 467 468 error = sosend(so, sendnam, NULL, top, NULL, flags); 469 if (error) { 470 if (rep) { 471 /* 472 * Deal with errors for the client side. 473 */ 474 if (rep->r_flags & R_SOFTTERM) 475 error = EINTR; 476 else 477 rep->r_flags |= R_MUSTRESEND; 478 } 479 480 /* 481 * Handle any recoverable (soft) socket errors here. (???) 482 */ 483 if (error != EINTR && error != ERESTART && 484 error != EWOULDBLOCK && error != EPIPE) 485 error = 0; 486 } 487 return (error); 488 } 489 490 #ifdef NFSCLIENT 491 /* 492 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 493 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 494 * Mark and consolidate the data into a new mbuf list. 495 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 496 * small mbufs. 497 * For SOCK_STREAM we must be very careful to read an entire record once 498 * we have read any of it, even if the system call has been interrupted. 499 */ 500 int 501 nfs_receive(struct nfsreq *rep, struct mbuf **aname, struct mbuf **mp) 502 { 503 struct socket *so; 504 struct uio auio; 505 struct iovec aio; 506 struct mbuf *m; 507 struct mbuf *control; 508 u_int32_t len; 509 struct mbuf **getnam; 510 int error, sotype, rcvflg; 511 struct proc *p = curproc; /* XXX */ 512 513 /* 514 * Set up arguments for soreceive() 515 */ 516 *mp = NULL; 517 *aname = NULL; 518 sotype = rep->r_nmp->nm_sotype; 519 520 /* 521 * For reliable protocols, lock against other senders/receivers 522 * in case a reconnect is necessary. 523 * For SOCK_STREAM, first get the Record Mark to find out how much 524 * more there is to get. 525 * We must lock the socket against other receivers 526 * until we have an entire rpc request/reply. 527 */ 528 if (sotype != SOCK_DGRAM) { 529 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 530 if (error) 531 return (error); 532 tryagain: 533 /* 534 * Check for fatal errors and resending request. 535 */ 536 /* 537 * Ugh: If a reconnect attempt just happened, nm_so 538 * would have changed. NULL indicates a failed 539 * attempt that has essentially shut down this 540 * mount point. 541 */ 542 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 543 nfs_sndunlock(&rep->r_nmp->nm_flag); 544 return (EINTR); 545 } 546 so = rep->r_nmp->nm_so; 547 if (!so) { 548 error = nfs_reconnect(rep); 549 if (error) { 550 nfs_sndunlock(&rep->r_nmp->nm_flag); 551 return (error); 552 } 553 goto tryagain; 554 } 555 while (rep->r_flags & R_MUSTRESEND) { 556 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 557 nfsstats.rpcretries++; 558 rep->r_rtt = 0; 559 rep->r_flags &= ~R_TIMING; 560 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 561 if (error) { 562 if (error == EINTR || error == ERESTART || 563 (error = nfs_reconnect(rep)) != 0) { 564 nfs_sndunlock(&rep->r_nmp->nm_flag); 565 return (error); 566 } 567 goto tryagain; 568 } 569 } 570 nfs_sndunlock(&rep->r_nmp->nm_flag); 571 if (sotype == SOCK_STREAM) { 572 aio.iov_base = (caddr_t) &len; 573 aio.iov_len = sizeof(u_int32_t); 574 auio.uio_iov = &aio; 575 auio.uio_iovcnt = 1; 576 auio.uio_segflg = UIO_SYSSPACE; 577 auio.uio_rw = UIO_READ; 578 auio.uio_offset = 0; 579 auio.uio_resid = sizeof(u_int32_t); 580 auio.uio_procp = p; 581 do { 582 rcvflg = MSG_WAITALL; 583 error = soreceive(so, NULL, &auio, NULL, NULL, 584 &rcvflg, 0); 585 if (error == EWOULDBLOCK && rep) { 586 if (rep->r_flags & R_SOFTTERM) 587 return (EINTR); 588 /* 589 * looks like the server died after it 590 * received the request, make sure 591 * that we will retransmit and we 592 * don't get stuck here forever. 593 */ 594 if (rep->r_rexmit >= rep->r_nmp->nm_retry) { 595 nfsstats.rpctimeouts++; 596 error = EPIPE; 597 } 598 } 599 } while (error == EWOULDBLOCK); 600 if (!error && auio.uio_resid > 0) { 601 log(LOG_INFO, 602 "short receive (%d/%d) from nfs server %s\n", 603 sizeof(u_int32_t) - auio.uio_resid, 604 sizeof(u_int32_t), 605 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 606 error = EPIPE; 607 } 608 if (error) 609 goto errout; 610 611 len = ntohl(len) & ~0x80000000; 612 /* 613 * This is SERIOUS! We are out of sync with the sender 614 * and forcing a disconnect/reconnect is all I can do. 615 */ 616 if (len > NFS_MAXPACKET) { 617 log(LOG_ERR, "%s (%d) from nfs server %s\n", 618 "impossible packet length", 619 len, 620 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 621 error = EFBIG; 622 goto errout; 623 } 624 auio.uio_resid = len; 625 do { 626 rcvflg = MSG_WAITALL; 627 error = soreceive(so, NULL, &auio, mp, NULL, 628 &rcvflg, 0); 629 } while (error == EWOULDBLOCK || error == EINTR || 630 error == ERESTART); 631 if (!error && auio.uio_resid > 0) { 632 log(LOG_INFO, 633 "short receive (%d/%d) from nfs server %s\n", 634 len - auio.uio_resid, len, 635 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 636 error = EPIPE; 637 } 638 } else { 639 /* 640 * NB: Since uio_resid is big, MSG_WAITALL is ignored 641 * and soreceive() will return when it has either a 642 * control msg or a data msg. 643 * We have no use for control msg., but must grab them 644 * and then throw them away so we know what is going 645 * on. 646 */ 647 auio.uio_resid = len = 100000000; /* Anything Big */ 648 auio.uio_procp = p; 649 do { 650 rcvflg = 0; 651 error = soreceive(so, NULL, &auio, mp, &control, 652 &rcvflg, 0); 653 if (control) 654 m_freem(control); 655 if (error == EWOULDBLOCK && rep) { 656 if (rep->r_flags & R_SOFTTERM) 657 return (EINTR); 658 } 659 } while (error == EWOULDBLOCK || 660 (!error && *mp == NULL && control)); 661 if ((rcvflg & MSG_EOR) == 0) 662 printf("Egad!!\n"); 663 if (!error && *mp == NULL) 664 error = EPIPE; 665 len -= auio.uio_resid; 666 } 667 errout: 668 if (error && error != EINTR && error != ERESTART) { 669 m_freem(*mp); 670 *mp = NULL; 671 if (error != EPIPE) 672 log(LOG_INFO, 673 "receive error %d from nfs server %s\n", 674 error, 675 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 676 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 677 if (!error) { 678 error = nfs_reconnect(rep); 679 if (!error) 680 goto tryagain; 681 nfs_sndunlock(&rep->r_nmp->nm_flag); 682 } 683 } 684 } else { 685 if ((so = rep->r_nmp->nm_so) == NULL) 686 return (EACCES); 687 if (so->so_state & SS_ISCONNECTED) 688 getnam = NULL; 689 else 690 getnam = aname; 691 auio.uio_resid = len = 1000000; 692 auio.uio_procp = p; 693 do { 694 rcvflg = 0; 695 error = soreceive(so, getnam, &auio, mp, NULL, 696 &rcvflg, 0); 697 if (error == EWOULDBLOCK && 698 (rep->r_flags & R_SOFTTERM)) 699 return (EINTR); 700 } while (error == EWOULDBLOCK); 701 len -= auio.uio_resid; 702 } 703 if (error) { 704 m_freem(*mp); 705 *mp = NULL; 706 } 707 /* 708 * Search for any mbufs that are not a multiple of 4 bytes long 709 * or with m_data not longword aligned. 710 * These could cause pointer alignment problems, so copy them to 711 * well aligned mbufs. 712 */ 713 nfs_realign(mp, 5 * NFSX_UNSIGNED); 714 return (error); 715 } 716 717 /* 718 * Implement receipt of reply on a socket. 719 * We must search through the list of received datagrams matching them 720 * with outstanding requests using the xid, until ours is found. 721 */ 722 int 723 nfs_reply(struct nfsreq *myrep) 724 { 725 struct nfsreq *rep; 726 struct nfsmount *nmp = myrep->r_nmp; 727 struct nfsm_info info; 728 struct mbuf *nam; 729 u_int32_t rxid, *tl, t1; 730 caddr_t cp2; 731 int s, error; 732 733 /* 734 * Loop around until we get our own reply 735 */ 736 for (;;) { 737 /* 738 * Lock against other receivers so that I don't get stuck in 739 * sbwait() after someone else has received my reply for me. 740 * Also necessary for connection based protocols to avoid 741 * race conditions during a reconnect. 742 */ 743 error = nfs_rcvlock(myrep); 744 if (error) 745 return (error == EALREADY ? 0 : error); 746 747 /* 748 * Get the next Rpc reply off the socket 749 */ 750 error = nfs_receive(myrep, &nam, &info.nmi_mrep); 751 nfs_rcvunlock(&nmp->nm_flag); 752 if (error) { 753 754 /* 755 * Ignore routing errors on connectionless protocols?? 756 */ 757 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 758 if (nmp->nm_so) 759 nmp->nm_so->so_error = 0; 760 continue; 761 } 762 return (error); 763 } 764 if (nam) 765 m_freem(nam); 766 767 /* 768 * Get the xid and check that it is an rpc reply 769 */ 770 info.nmi_md = info.nmi_mrep; 771 info.nmi_dpos = mtod(info.nmi_md, caddr_t); 772 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 773 rxid = *tl++; 774 if (*tl != rpc_reply) { 775 nfsstats.rpcinvalid++; 776 m_freem(info.nmi_mrep); 777 nfsmout: 778 continue; 779 } 780 781 /* 782 * Loop through the request list to match up the reply 783 * Iff no match, just drop the datagram 784 */ 785 s = splsoftnet(); 786 TAILQ_FOREACH(rep, &nmp->nm_reqsq, r_chain) { 787 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 788 /* Found it.. */ 789 rep->r_mrep = info.nmi_mrep; 790 rep->r_md = info.nmi_md; 791 rep->r_dpos = info.nmi_dpos; 792 793 /* 794 * Update congestion window. 795 * Do the additive increase of 796 * one rpc/rtt. 797 */ 798 if (nmp->nm_cwnd <= nmp->nm_sent) { 799 nmp->nm_cwnd += 800 (NFS_CWNDSCALE * NFS_CWNDSCALE + 801 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 802 if (nmp->nm_cwnd > NFS_MAXCWND) 803 nmp->nm_cwnd = NFS_MAXCWND; 804 } 805 rep->r_flags &= ~R_SENT; 806 nmp->nm_sent -= NFS_CWNDSCALE; 807 808 if (rep->r_flags & R_TIMING) 809 nfs_update_rtt(rep); 810 811 nmp->nm_timeouts = 0; 812 break; 813 } 814 } 815 splx(s); 816 /* 817 * If not matched to a request, drop it. 818 * If it's mine, get out. 819 */ 820 if (rep == 0) { 821 nfsstats.rpcunexpected++; 822 m_freem(info.nmi_mrep); 823 } else if (rep == myrep) { 824 if (rep->r_mrep == NULL) 825 panic("nfsreply nil"); 826 return (0); 827 } 828 } 829 } 830 831 /* 832 * nfs_request - goes something like this 833 * - fill in request struct 834 * - links it into list 835 * - calls nfs_send() for first transmit 836 * - calls nfs_receive() to get reply 837 * - break down rpc header and return with nfs reply pointed to 838 * by mrep or error 839 * nb: always frees up mreq mbuf list 840 */ 841 int 842 nfs_request(struct vnode *vp, int procnum, struct nfsm_info *infop) 843 { 844 struct mbuf *m; 845 u_int32_t *tl; 846 struct nfsmount *nmp; 847 struct timeval tv; 848 caddr_t cp2; 849 int t1, i, s, error = 0; 850 int trylater_delay; 851 struct nfsreq *rep; 852 int mrest_len; 853 struct nfsm_info info; 854 855 rep = pool_get(&nfsreqpl, PR_WAITOK); 856 rep->r_nmp = VFSTONFS(vp->v_mount); 857 rep->r_vp = vp; 858 rep->r_procp = infop->nmi_procp; 859 rep->r_procnum = procnum; 860 861 mrest_len = 0; 862 m = infop->nmi_mreq; 863 while (m) { 864 mrest_len += m->m_len; 865 m = m->m_next; 866 } 867 868 /* empty mbuf for AUTH_UNIX header */ 869 rep->r_mreq = m_gethdr(M_WAIT, MT_DATA); 870 rep->r_mreq->m_next = infop->nmi_mreq; 871 rep->r_mreq->m_pkthdr.len = mrest_len; 872 873 trylater_delay = NFS_MINTIMEO; 874 875 nmp = rep->r_nmp; 876 877 /* Get the RPC header with authorization. */ 878 nfsm_rpchead(rep, infop->nmi_cred, RPCAUTH_UNIX); 879 m = rep->r_mreq; 880 881 /* 882 * For stream protocols, insert a Sun RPC Record Mark. 883 */ 884 if (nmp->nm_sotype == SOCK_STREAM) { 885 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 886 *mtod(m, u_int32_t *) = htonl(0x80000000 | 887 (m->m_pkthdr.len - NFSX_UNSIGNED)); 888 } 889 890 tryagain: 891 rep->r_rtt = rep->r_rexmit = 0; 892 if (nfs_ptimers[rep->r_procnum] != NFS_DEFAULT_TIMER) 893 rep->r_flags = R_TIMING; 894 else 895 rep->r_flags = 0; 896 rep->r_mrep = NULL; 897 898 /* 899 * Do the client side RPC. 900 */ 901 nfsstats.rpcrequests++; 902 /* 903 * Chain request into list of outstanding requests. Be sure 904 * to put it LAST so timer finds oldest requests first. 905 */ 906 s = splsoftnet(); 907 if (TAILQ_EMPTY(&nmp->nm_reqsq)) 908 timeout_add(&nmp->nm_rtimeout, nfs_ticks); 909 TAILQ_INSERT_TAIL(&nmp->nm_reqsq, rep, r_chain); 910 911 /* 912 * If backing off another request or avoiding congestion, don't 913 * send this one now but let timer do it. If not timing a request, 914 * do it now. 915 */ 916 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 917 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 918 nmp->nm_sent < nmp->nm_cwnd)) { 919 splx(s); 920 if (nmp->nm_soflags & PR_CONNREQUIRED) 921 error = nfs_sndlock(&nmp->nm_flag, rep); 922 if (!error) { 923 error = nfs_send(nmp->nm_so, nmp->nm_nam, 924 m_copym(m, 0, M_COPYALL, M_WAIT), 925 rep); 926 if (nmp->nm_soflags & PR_CONNREQUIRED) 927 nfs_sndunlock(&nmp->nm_flag); 928 } 929 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 930 nmp->nm_sent += NFS_CWNDSCALE; 931 rep->r_flags |= R_SENT; 932 } 933 } else { 934 splx(s); 935 rep->r_rtt = -1; 936 } 937 938 /* 939 * Wait for the reply from our send or the timer's. 940 */ 941 if (!error || error == EPIPE) 942 error = nfs_reply(rep); 943 944 /* 945 * RPC done, unlink the request. 946 */ 947 s = splsoftnet(); 948 TAILQ_REMOVE(&nmp->nm_reqsq, rep, r_chain); 949 if (TAILQ_EMPTY(&nmp->nm_reqsq)) 950 timeout_del(&nmp->nm_rtimeout); 951 splx(s); 952 953 /* 954 * Decrement the outstanding request count. 955 */ 956 if (rep->r_flags & R_SENT) { 957 rep->r_flags &= ~R_SENT; /* paranoia */ 958 nmp->nm_sent -= NFS_CWNDSCALE; 959 } 960 961 /* 962 * If there was a successful reply and a tprintf msg. 963 * tprintf a response. 964 */ 965 if (!error && (rep->r_flags & R_TPRINTFMSG)) 966 nfs_msg(rep, "is alive again"); 967 info.nmi_mrep = rep->r_mrep; 968 info.nmi_md = rep->r_md; 969 info.nmi_dpos = rep->r_dpos; 970 if (error) { 971 infop->nmi_mrep = NULL; 972 goto nfsmout1; 973 } 974 975 /* 976 * break down the rpc header and check if ok 977 */ 978 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 979 if (*tl++ == rpc_msgdenied) { 980 if (*tl == rpc_mismatch) 981 error = EOPNOTSUPP; 982 else 983 error = EACCES; /* Should be EAUTH. */ 984 infop->nmi_mrep = NULL; 985 goto nfsmout1; 986 } 987 988 /* 989 * Since we only support RPCAUTH_UNIX atm we step over the 990 * reply verifer type, and in the (error) case that there really 991 * is any data in it, we advance over it. 992 */ 993 tl++; /* Step over verifer type */ 994 i = fxdr_unsigned(int32_t, *tl); 995 if (i > 0) 996 nfsm_adv(nfsm_rndup(i)); /* Should not happen */ 997 998 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 999 /* 0 == ok */ 1000 if (*tl == 0) { 1001 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1002 if (*tl != 0) { 1003 error = fxdr_unsigned(int, *tl); 1004 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1005 error == NFSERR_TRYLATER) { 1006 m_freem(info.nmi_mrep); 1007 error = 0; 1008 tv.tv_sec = time_second + trylater_delay; 1009 tv.tv_usec = 0; 1010 tsleep(&tv, PSOCK, "nfsretry", hzto(&tv)); 1011 trylater_delay *= NFS_TIMEOUTMUL; 1012 if (trylater_delay > NFS_MAXTIMEO) 1013 trylater_delay = NFS_MAXTIMEO; 1014 1015 goto tryagain; 1016 } 1017 1018 /* 1019 * If the File Handle was stale, invalidate the 1020 * lookup cache, just in case. 1021 */ 1022 if (error == ESTALE) 1023 cache_purge(rep->r_vp); 1024 } 1025 goto nfsmout; 1026 } 1027 1028 error = EPROTONOSUPPORT; 1029 1030 nfsmout: 1031 infop->nmi_mrep = info.nmi_mrep; 1032 infop->nmi_md = info.nmi_md; 1033 infop->nmi_dpos = info.nmi_dpos; 1034 nfsmout1: 1035 m_freem(rep->r_mreq); 1036 pool_put(&nfsreqpl, rep); 1037 return (error); 1038 } 1039 #endif /* NFSCLIENT */ 1040 1041 /* 1042 * Generate the rpc reply header 1043 * siz arg. is used to decide if adding a cluster is worthwhile 1044 */ 1045 int 1046 nfs_rephead(int siz, struct nfsrv_descript *nd, struct nfssvc_sock *slp, 1047 int err, struct mbuf **mrq, struct mbuf **mbp) 1048 { 1049 u_int32_t *tl; 1050 struct mbuf *mreq; 1051 struct mbuf *mb; 1052 1053 MGETHDR(mreq, M_WAIT, MT_DATA); 1054 mb = mreq; 1055 /* 1056 * If this is a big reply, use a cluster else 1057 * try and leave leading space for the lower level headers. 1058 */ 1059 siz += RPC_REPLYSIZ; 1060 if (siz >= max_datalen) { 1061 MCLGET(mreq, M_WAIT); 1062 } else 1063 mreq->m_data += max_hdr; 1064 tl = mtod(mreq, u_int32_t *); 1065 mreq->m_len = 6 * NFSX_UNSIGNED; 1066 *tl++ = txdr_unsigned(nd->nd_retxid); 1067 *tl++ = rpc_reply; 1068 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1069 *tl++ = rpc_msgdenied; 1070 if (err & NFSERR_AUTHERR) { 1071 *tl++ = rpc_autherr; 1072 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1073 mreq->m_len -= NFSX_UNSIGNED; 1074 } else { 1075 *tl++ = rpc_mismatch; 1076 *tl++ = txdr_unsigned(RPC_VER2); 1077 *tl = txdr_unsigned(RPC_VER2); 1078 } 1079 } else { 1080 *tl++ = rpc_msgaccepted; 1081 1082 /* AUTH_UNIX requires RPCAUTH_NULL. */ 1083 *tl++ = 0; 1084 *tl++ = 0; 1085 1086 switch (err) { 1087 case EPROGUNAVAIL: 1088 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1089 break; 1090 case EPROGMISMATCH: 1091 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1092 tl = nfsm_build(&mb, 2 * NFSX_UNSIGNED); 1093 *tl++ = txdr_unsigned(NFS_VER2); 1094 *tl = txdr_unsigned(NFS_VER3); 1095 break; 1096 case EPROCUNAVAIL: 1097 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1098 break; 1099 case EBADRPC: 1100 *tl = txdr_unsigned(RPC_GARBAGE); 1101 break; 1102 default: 1103 *tl = 0; 1104 if (err != NFSERR_RETVOID) { 1105 tl = nfsm_build(&mb, NFSX_UNSIGNED); 1106 if (err) 1107 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1108 else 1109 *tl = 0; 1110 } 1111 break; 1112 }; 1113 } 1114 1115 *mrq = mreq; 1116 if (mbp != NULL) 1117 *mbp = mb; 1118 if (err != 0 && err != NFSERR_RETVOID) 1119 nfsstats.srvrpc_errs++; 1120 return (0); 1121 } 1122 1123 /* 1124 * nfs timer routine 1125 * Scan the nfsreq list and retranmit any requests that have timed out. 1126 */ 1127 void 1128 nfs_timer(void *arg) 1129 { 1130 struct nfsmount *nmp = arg; 1131 struct nfsreq *rep; 1132 struct mbuf *m; 1133 struct socket *so; 1134 int timeo, s, error; 1135 1136 s = splsoftnet(); 1137 TAILQ_FOREACH(rep, &nmp->nm_reqsq, r_chain) { 1138 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1139 continue; 1140 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1141 rep->r_flags |= R_SOFTTERM; 1142 continue; 1143 } 1144 if (rep->r_rtt >= 0) { 1145 rep->r_rtt++; 1146 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1147 timeo = nmp->nm_timeo; 1148 else 1149 timeo = nfs_estimate_rto(nmp, rep->r_procnum); 1150 if (nmp->nm_timeouts > 0) 1151 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1152 if (rep->r_rtt <= timeo) 1153 continue; 1154 if (nmp->nm_timeouts < nitems(nfs_backoff)) 1155 nmp->nm_timeouts++; 1156 } 1157 1158 /* Check for server not responding. */ 1159 if ((rep->r_flags & R_TPRINTFMSG) == 0 && rep->r_rexmit > 4) { 1160 nfs_msg(rep, "not responding"); 1161 rep->r_flags |= R_TPRINTFMSG; 1162 } 1163 if (rep->r_rexmit >= nmp->nm_retry) { /* too many */ 1164 nfsstats.rpctimeouts++; 1165 rep->r_flags |= R_SOFTTERM; 1166 continue; 1167 } 1168 if (nmp->nm_sotype != SOCK_DGRAM) { 1169 if (++rep->r_rexmit > NFS_MAXREXMIT) 1170 rep->r_rexmit = NFS_MAXREXMIT; 1171 continue; 1172 } 1173 1174 if ((so = nmp->nm_so) == NULL) 1175 continue; 1176 1177 /* 1178 * If there is enough space and the window allows.. 1179 * Resend it 1180 * Set r_rtt to -1 in case we fail to send it now. 1181 */ 1182 rep->r_rtt = -1; 1183 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1184 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1185 (rep->r_flags & R_SENT) || 1186 nmp->nm_sent < nmp->nm_cwnd) && 1187 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1188 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1189 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1190 NULL, NULL, curproc); 1191 else 1192 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1193 nmp->nm_nam, NULL, curproc); 1194 if (error) { 1195 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) 1196 so->so_error = 0; 1197 } else { 1198 /* 1199 * Iff first send, start timing 1200 * else turn timing off, backoff timer 1201 * and divide congestion window by 2. 1202 */ 1203 if (rep->r_flags & R_SENT) { 1204 rep->r_flags &= ~R_TIMING; 1205 if (++rep->r_rexmit > NFS_MAXREXMIT) 1206 rep->r_rexmit = NFS_MAXREXMIT; 1207 nmp->nm_cwnd >>= 1; 1208 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1209 nmp->nm_cwnd = NFS_CWNDSCALE; 1210 nfsstats.rpcretries++; 1211 } else { 1212 rep->r_flags |= R_SENT; 1213 nmp->nm_sent += NFS_CWNDSCALE; 1214 } 1215 rep->r_rtt = 0; 1216 } 1217 } 1218 } 1219 splx(s); 1220 timeout_add(&nmp->nm_rtimeout, nfs_ticks); 1221 } 1222 1223 /* 1224 * Test for a termination condition pending on the process. 1225 * This is used for NFSMNT_INT mounts. 1226 */ 1227 int 1228 nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct proc *p) 1229 { 1230 1231 if (rep && (rep->r_flags & R_SOFTTERM)) 1232 return (EINTR); 1233 if (!(nmp->nm_flag & NFSMNT_INT)) 1234 return (0); 1235 if (p && p->p_siglist && 1236 (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) & 1237 NFSINT_SIGMASK)) 1238 return (EINTR); 1239 return (0); 1240 } 1241 1242 /* 1243 * Lock a socket against others. 1244 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1245 * and also to avoid race conditions between the processes with nfs requests 1246 * in progress when a reconnect is necessary. 1247 */ 1248 int 1249 nfs_sndlock(int *flagp, struct nfsreq *rep) 1250 { 1251 struct proc *p; 1252 int slpflag = 0, slptimeo = 0; 1253 1254 if (rep) { 1255 p = rep->r_procp; 1256 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1257 slpflag = PCATCH; 1258 } else 1259 p = NULL; 1260 while (*flagp & NFSMNT_SNDLOCK) { 1261 if (rep && nfs_sigintr(rep->r_nmp, rep, p)) 1262 return (EINTR); 1263 *flagp |= NFSMNT_WANTSND; 1264 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1265 slptimeo); 1266 if (slpflag == PCATCH) { 1267 slpflag = 0; 1268 slptimeo = 2 * hz; 1269 } 1270 } 1271 *flagp |= NFSMNT_SNDLOCK; 1272 return (0); 1273 } 1274 1275 /* 1276 * Unlock the stream socket for others. 1277 */ 1278 void 1279 nfs_sndunlock(int *flagp) 1280 { 1281 1282 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1283 panic("nfs sndunlock"); 1284 *flagp &= ~NFSMNT_SNDLOCK; 1285 if (*flagp & NFSMNT_WANTSND) { 1286 *flagp &= ~NFSMNT_WANTSND; 1287 wakeup((caddr_t)flagp); 1288 } 1289 } 1290 1291 int 1292 nfs_rcvlock(struct nfsreq *rep) 1293 { 1294 int *flagp = &rep->r_nmp->nm_flag; 1295 int slpflag, slptimeo = 0; 1296 1297 if (*flagp & NFSMNT_INT) 1298 slpflag = PCATCH; 1299 else 1300 slpflag = 0; 1301 1302 while (*flagp & NFSMNT_RCVLOCK) { 1303 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1304 return (EINTR); 1305 *flagp |= NFSMNT_WANTRCV; 1306 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1307 slptimeo); 1308 if (rep->r_mrep != NULL) { 1309 /* 1310 * Don't take the lock if our reply has been received 1311 * while we where sleeping. 1312 */ 1313 return (EALREADY); 1314 } 1315 if (slpflag == PCATCH) { 1316 slpflag = 0; 1317 slptimeo = 2 * hz; 1318 } 1319 } 1320 *flagp |= NFSMNT_RCVLOCK; 1321 return (0); 1322 } 1323 1324 /* 1325 * Unlock the stream socket for others. 1326 */ 1327 void 1328 nfs_rcvunlock(int *flagp) 1329 { 1330 1331 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1332 panic("nfs rcvunlock"); 1333 *flagp &= ~NFSMNT_RCVLOCK; 1334 if (*flagp & NFSMNT_WANTRCV) { 1335 *flagp &= ~NFSMNT_WANTRCV; 1336 wakeup((caddr_t)flagp); 1337 } 1338 } 1339 1340 /* 1341 * Auxiliary routine to align the length of mbuf copies made with m_copyback(). 1342 */ 1343 void 1344 nfs_realign_fixup(struct mbuf *m, struct mbuf *n, unsigned int *off) 1345 { 1346 size_t padding; 1347 1348 /* 1349 * The maximum number of bytes that m_copyback() places in a mbuf is 1350 * always an aligned quantity, so realign happens at the chain's tail. 1351 */ 1352 while (n->m_next != NULL) 1353 n = n->m_next; 1354 1355 /* 1356 * Pad from the next elements in the source chain. Loop until the 1357 * destination chain is aligned, or the end of the source is reached. 1358 */ 1359 do { 1360 m = m->m_next; 1361 if (m == NULL) 1362 return; 1363 1364 padding = min(ALIGN(n->m_len) - n->m_len, m->m_len); 1365 if (padding > M_TRAILINGSPACE(n)) 1366 panic("nfs_realign_fixup: no memory to pad to"); 1367 1368 bcopy(mtod(m, void *), mtod(n, char *) + n->m_len, padding); 1369 1370 n->m_len += padding; 1371 m_adj(m, padding); 1372 *off += padding; 1373 1374 } while (!ALIGNED_POINTER(n->m_len, void *)); 1375 } 1376 1377 /* 1378 * The NFS RPC parsing code uses the data address and the length of mbuf 1379 * structures to calculate on-memory addresses. This function makes sure these 1380 * parameters are correctly aligned. 1381 */ 1382 void 1383 nfs_realign(struct mbuf **pm, int hsiz) 1384 { 1385 struct mbuf *m; 1386 struct mbuf *n = NULL; 1387 unsigned int off = 0; 1388 1389 ++nfs_realign_test; 1390 while ((m = *pm) != NULL) { 1391 if (!ALIGNED_POINTER(m->m_data, void *) || 1392 !ALIGNED_POINTER(m->m_len, void *)) { 1393 MGET(n, M_WAIT, MT_DATA); 1394 if (ALIGN(m->m_len) >= MINCLSIZE) { 1395 MCLGET(n, M_WAIT); 1396 } 1397 n->m_len = 0; 1398 break; 1399 } 1400 pm = &m->m_next; 1401 } 1402 /* 1403 * If n is non-NULL, loop on m copying data, then replace the 1404 * portion of the chain that had to be realigned. 1405 */ 1406 if (n != NULL) { 1407 ++nfs_realign_count; 1408 while (m) { 1409 m_copyback(n, off, m->m_len, mtod(m, caddr_t)); 1410 1411 /* 1412 * If an unaligned amount of memory was copied, fix up 1413 * the last mbuf created by m_copyback(). 1414 */ 1415 if (!ALIGNED_POINTER(m->m_len, void *)) 1416 nfs_realign_fixup(m, n, &off); 1417 1418 off += m->m_len; 1419 m = m->m_next; 1420 } 1421 m_freem(*pm); 1422 *pm = n; 1423 } 1424 } 1425 1426 1427 /* 1428 * Parse an RPC request 1429 * - verify it 1430 * - fill in the cred struct. 1431 */ 1432 int 1433 nfs_getreq(struct nfsrv_descript *nd, struct nfsd *nfsd, int has_header) 1434 { 1435 int len, i; 1436 u_int32_t *tl; 1437 int32_t t1; 1438 caddr_t cp2; 1439 u_int32_t nfsvers, auth_type; 1440 int error = 0; 1441 struct nfsm_info info; 1442 1443 info.nmi_mrep = nd->nd_mrep; 1444 info.nmi_md = nd->nd_md; 1445 info.nmi_dpos = nd->nd_dpos; 1446 if (has_header) { 1447 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1448 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1449 if (*tl++ != rpc_call) { 1450 m_freem(info.nmi_mrep); 1451 return (EBADRPC); 1452 } 1453 } else 1454 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1455 nd->nd_repstat = 0; 1456 nd->nd_flag = 0; 1457 if (*tl++ != rpc_vers) { 1458 nd->nd_repstat = ERPCMISMATCH; 1459 nd->nd_procnum = NFSPROC_NOOP; 1460 return (0); 1461 } 1462 if (*tl != nfs_prog) { 1463 nd->nd_repstat = EPROGUNAVAIL; 1464 nd->nd_procnum = NFSPROC_NOOP; 1465 return (0); 1466 } 1467 tl++; 1468 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1469 if (nfsvers != NFS_VER2 && nfsvers != NFS_VER3) { 1470 nd->nd_repstat = EPROGMISMATCH; 1471 nd->nd_procnum = NFSPROC_NOOP; 1472 return (0); 1473 } 1474 if (nfsvers == NFS_VER3) 1475 nd->nd_flag = ND_NFSV3; 1476 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1477 if (nd->nd_procnum == NFSPROC_NULL) 1478 return (0); 1479 if (nd->nd_procnum >= NFS_NPROCS || 1480 (nd->nd_procnum > NFSPROC_COMMIT) || 1481 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1482 nd->nd_repstat = EPROCUNAVAIL; 1483 nd->nd_procnum = NFSPROC_NOOP; 1484 return (0); 1485 } 1486 if ((nd->nd_flag & ND_NFSV3) == 0) 1487 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1488 auth_type = *tl++; 1489 len = fxdr_unsigned(int, *tl++); 1490 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1491 m_freem(info.nmi_mrep); 1492 return (EBADRPC); 1493 } 1494 1495 /* Handle auth_unix */ 1496 if (auth_type == rpc_auth_unix) { 1497 len = fxdr_unsigned(int, *++tl); 1498 if (len < 0 || len > NFS_MAXNAMLEN) { 1499 m_freem(info.nmi_mrep); 1500 return (EBADRPC); 1501 } 1502 nfsm_adv(nfsm_rndup(len)); 1503 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1504 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred)); 1505 nd->nd_cr.cr_ref = 1; 1506 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1507 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1508 len = fxdr_unsigned(int, *tl); 1509 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1510 m_freem(info.nmi_mrep); 1511 return (EBADRPC); 1512 } 1513 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1514 for (i = 0; i < len; i++) 1515 if (i < NGROUPS) 1516 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1517 else 1518 tl++; 1519 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1520 len = fxdr_unsigned(int, *++tl); 1521 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1522 m_freem(info.nmi_mrep); 1523 return (EBADRPC); 1524 } 1525 if (len > 0) 1526 nfsm_adv(nfsm_rndup(len)); 1527 } else { 1528 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1529 nd->nd_procnum = NFSPROC_NOOP; 1530 return (0); 1531 } 1532 1533 nd->nd_md = info.nmi_md; 1534 nd->nd_dpos = info.nmi_dpos; 1535 return (0); 1536 nfsmout: 1537 return (error); 1538 } 1539 1540 void 1541 nfs_msg(struct nfsreq *rep, char *msg) 1542 { 1543 tpr_t tpr; 1544 1545 if (rep->r_procp) 1546 tpr = tprintf_open(rep->r_procp); 1547 else 1548 tpr = NULL; 1549 1550 tprintf(tpr, "nfs server %s: %s\n", 1551 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname, msg); 1552 tprintf_close(tpr); 1553 } 1554 1555 #ifdef NFSSERVER 1556 /* 1557 * Socket upcall routine for the nfsd sockets. 1558 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1559 * Essentially do as much as possible non-blocking, else punt and it will 1560 * be called with M_WAIT from an nfsd. 1561 */ 1562 void 1563 nfsrv_rcv(struct socket *so, caddr_t arg, int waitflag) 1564 { 1565 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1566 struct mbuf *m; 1567 struct mbuf *mp, *nam; 1568 struct uio auio; 1569 int flags, error; 1570 1571 if ((slp->ns_flag & SLP_VALID) == 0) 1572 return; 1573 #ifdef notdef 1574 /* 1575 * Define this to test for nfsds handling this under heavy load. 1576 */ 1577 if (waitflag == M_DONTWAIT) { 1578 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1579 } 1580 #endif 1581 auio.uio_procp = NULL; 1582 if (so->so_type == SOCK_STREAM) { 1583 /* 1584 * If there are already records on the queue, defer soreceive() 1585 * to an nfsd so that there is feedback to the TCP layer that 1586 * the nfs servers are heavily loaded. 1587 */ 1588 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1589 slp->ns_flag |= SLP_NEEDQ; 1590 goto dorecs; 1591 } 1592 1593 /* 1594 * Do soreceive(). 1595 */ 1596 auio.uio_resid = 1000000000; 1597 flags = MSG_DONTWAIT; 1598 error = soreceive(so, &nam, &auio, &mp, NULL, 1599 &flags, 0); 1600 if (error || mp == NULL) { 1601 if (error == EWOULDBLOCK) 1602 slp->ns_flag |= SLP_NEEDQ; 1603 else 1604 slp->ns_flag |= SLP_DISCONN; 1605 goto dorecs; 1606 } 1607 m = mp; 1608 if (slp->ns_rawend) { 1609 slp->ns_rawend->m_next = m; 1610 slp->ns_cc += 1000000000 - auio.uio_resid; 1611 } else { 1612 slp->ns_raw = m; 1613 slp->ns_cc = 1000000000 - auio.uio_resid; 1614 } 1615 while (m->m_next) 1616 m = m->m_next; 1617 slp->ns_rawend = m; 1618 1619 /* 1620 * Now try and parse record(s) out of the raw stream data. 1621 */ 1622 error = nfsrv_getstream(slp, waitflag); 1623 if (error) { 1624 if (error == EPERM) 1625 slp->ns_flag |= SLP_DISCONN; 1626 else 1627 slp->ns_flag |= SLP_NEEDQ; 1628 } 1629 } else { 1630 do { 1631 auio.uio_resid = 1000000000; 1632 flags = MSG_DONTWAIT; 1633 error = soreceive(so, &nam, &auio, &mp, 1634 NULL, &flags, 0); 1635 if (mp) { 1636 if (nam) { 1637 m = nam; 1638 m->m_next = mp; 1639 } else 1640 m = mp; 1641 if (slp->ns_recend) 1642 slp->ns_recend->m_nextpkt = m; 1643 else 1644 slp->ns_rec = m; 1645 slp->ns_recend = m; 1646 m->m_nextpkt = NULL; 1647 } 1648 if (error) { 1649 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1650 && error != EWOULDBLOCK) { 1651 slp->ns_flag |= SLP_DISCONN; 1652 goto dorecs; 1653 } 1654 } 1655 } while (mp); 1656 } 1657 1658 /* 1659 * Now try and process the request records, non-blocking. 1660 */ 1661 dorecs: 1662 if (waitflag == M_DONTWAIT && 1663 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1664 nfsrv_wakenfsd(slp); 1665 } 1666 1667 /* 1668 * Try and extract an RPC request from the mbuf data list received on a 1669 * stream socket. The "waitflag" argument indicates whether or not it 1670 * can sleep. 1671 */ 1672 int 1673 nfsrv_getstream(struct nfssvc_sock *slp, int waitflag) 1674 { 1675 struct mbuf *m, **mpp; 1676 char *cp1, *cp2; 1677 int len; 1678 struct mbuf *om, *m2, *recm; 1679 u_int32_t recmark; 1680 1681 if (slp->ns_flag & SLP_GETSTREAM) 1682 panic("nfs getstream"); 1683 slp->ns_flag |= SLP_GETSTREAM; 1684 for (;;) { 1685 if (slp->ns_reclen == 0) { 1686 if (slp->ns_cc < NFSX_UNSIGNED) { 1687 slp->ns_flag &= ~SLP_GETSTREAM; 1688 return (0); 1689 } 1690 m = slp->ns_raw; 1691 if (m->m_len >= NFSX_UNSIGNED) { 1692 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 1693 m->m_data += NFSX_UNSIGNED; 1694 m->m_len -= NFSX_UNSIGNED; 1695 } else { 1696 cp1 = (caddr_t)&recmark; 1697 cp2 = mtod(m, caddr_t); 1698 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1699 while (m->m_len == 0) { 1700 m = m->m_next; 1701 cp2 = mtod(m, caddr_t); 1702 } 1703 *cp1++ = *cp2++; 1704 m->m_data++; 1705 m->m_len--; 1706 } 1707 } 1708 slp->ns_cc -= NFSX_UNSIGNED; 1709 recmark = ntohl(recmark); 1710 slp->ns_reclen = recmark & ~0x80000000; 1711 if (recmark & 0x80000000) 1712 slp->ns_flag |= SLP_LASTFRAG; 1713 else 1714 slp->ns_flag &= ~SLP_LASTFRAG; 1715 if (slp->ns_reclen > NFS_MAXPACKET) { 1716 slp->ns_flag &= ~SLP_GETSTREAM; 1717 return (EPERM); 1718 } 1719 } 1720 1721 /* 1722 * Now get the record part. 1723 */ 1724 recm = NULL; 1725 if (slp->ns_cc == slp->ns_reclen) { 1726 recm = slp->ns_raw; 1727 slp->ns_raw = slp->ns_rawend = NULL; 1728 slp->ns_cc = slp->ns_reclen = 0; 1729 } else if (slp->ns_cc > slp->ns_reclen) { 1730 len = 0; 1731 m = slp->ns_raw; 1732 om = NULL; 1733 while (len < slp->ns_reclen) { 1734 if ((len + m->m_len) > slp->ns_reclen) { 1735 m2 = m_copym(m, 0, slp->ns_reclen - len, 1736 waitflag); 1737 if (m2) { 1738 if (om) { 1739 om->m_next = m2; 1740 recm = slp->ns_raw; 1741 } else 1742 recm = m2; 1743 m->m_data += slp->ns_reclen - len; 1744 m->m_len -= slp->ns_reclen - len; 1745 len = slp->ns_reclen; 1746 } else { 1747 slp->ns_flag &= ~SLP_GETSTREAM; 1748 return (EWOULDBLOCK); 1749 } 1750 } else if ((len + m->m_len) == slp->ns_reclen) { 1751 om = m; 1752 len += m->m_len; 1753 m = m->m_next; 1754 recm = slp->ns_raw; 1755 om->m_next = NULL; 1756 } else { 1757 om = m; 1758 len += m->m_len; 1759 m = m->m_next; 1760 } 1761 } 1762 slp->ns_raw = m; 1763 slp->ns_cc -= len; 1764 slp->ns_reclen = 0; 1765 } else { 1766 slp->ns_flag &= ~SLP_GETSTREAM; 1767 return (0); 1768 } 1769 1770 /* 1771 * Accumulate the fragments into a record. 1772 */ 1773 mpp = &slp->ns_frag; 1774 while (*mpp) 1775 mpp = &((*mpp)->m_next); 1776 *mpp = recm; 1777 if (slp->ns_flag & SLP_LASTFRAG) { 1778 if (slp->ns_recend) 1779 slp->ns_recend->m_nextpkt = slp->ns_frag; 1780 else 1781 slp->ns_rec = slp->ns_frag; 1782 slp->ns_recend = slp->ns_frag; 1783 slp->ns_frag = NULL; 1784 } 1785 } 1786 } 1787 1788 /* 1789 * Parse an RPC header. 1790 */ 1791 int 1792 nfsrv_dorec(struct nfssvc_sock *slp, struct nfsd *nfsd, 1793 struct nfsrv_descript **ndp) 1794 { 1795 struct mbuf *m, *nam; 1796 struct nfsrv_descript *nd; 1797 int error; 1798 1799 *ndp = NULL; 1800 if ((slp->ns_flag & SLP_VALID) == 0 || 1801 (m = slp->ns_rec) == NULL) 1802 return (ENOBUFS); 1803 slp->ns_rec = m->m_nextpkt; 1804 if (slp->ns_rec) 1805 m->m_nextpkt = NULL; 1806 else 1807 slp->ns_recend = NULL; 1808 if (m->m_type == MT_SONAME) { 1809 nam = m; 1810 m = m->m_next; 1811 nam->m_next = NULL; 1812 } else 1813 nam = NULL; 1814 nd = pool_get(&nfsrv_descript_pl, PR_WAITOK); 1815 nfs_realign(&m, 10 * NFSX_UNSIGNED); 1816 nd->nd_md = nd->nd_mrep = m; 1817 nd->nd_nam2 = nam; 1818 nd->nd_dpos = mtod(m, caddr_t); 1819 error = nfs_getreq(nd, nfsd, 1); 1820 if (error) { 1821 m_freem(nam); 1822 pool_put(&nfsrv_descript_pl, nd); 1823 return (error); 1824 } 1825 *ndp = nd; 1826 nfsd->nfsd_nd = nd; 1827 return (0); 1828 } 1829 1830 1831 /* 1832 * Search for a sleeping nfsd and wake it up. 1833 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 1834 * running nfsds will go look for the work in the nfssvc_sock list. 1835 */ 1836 void 1837 nfsrv_wakenfsd(struct nfssvc_sock *slp) 1838 { 1839 struct nfsd *nfsd; 1840 1841 if ((slp->ns_flag & SLP_VALID) == 0) 1842 return; 1843 1844 TAILQ_FOREACH(nfsd, &nfsd_head, nfsd_chain) { 1845 if (nfsd->nfsd_flag & NFSD_WAITING) { 1846 nfsd->nfsd_flag &= ~NFSD_WAITING; 1847 if (nfsd->nfsd_slp) 1848 panic("nfsd wakeup"); 1849 slp->ns_sref++; 1850 nfsd->nfsd_slp = slp; 1851 wakeup_one(nfsd); 1852 return; 1853 } 1854 } 1855 1856 slp->ns_flag |= SLP_DOREC; 1857 nfsd_head_flag |= NFSD_CHECKSLP; 1858 } 1859 #endif /* NFSSERVER */ 1860