1 /* $OpenBSD: nfs_socket.c,v 1.128 2017/09/07 11:35:34 bluhm 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/signalvar.h> 52 #include <sys/socket.h> 53 #include <sys/socketvar.h> 54 #include <sys/syslog.h> 55 #include <sys/tprintf.h> 56 #include <sys/namei.h> 57 #include <sys/pool.h> 58 #include <sys/queue.h> 59 60 #include <netinet/in.h> 61 #include <netinet/tcp.h> 62 63 #include <nfs/rpcv2.h> 64 #include <nfs/nfsproto.h> 65 #include <nfs/nfs.h> 66 #include <nfs/xdr_subs.h> 67 #include <nfs/nfsm_subs.h> 68 #include <nfs/nfsmount.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 *); 131 132 int nfs_rcvlock(struct nfsreq *); 133 int nfs_receive(struct nfsreq *, struct mbuf **, struct mbuf **); 134 int nfs_reconnect(struct nfsreq *); 135 int nfs_reply(struct nfsreq *); 136 void nfs_msg(struct nfsreq *, char *); 137 void nfs_rcvunlock(int *); 138 139 int nfsrv_getstream(struct nfssvc_sock *, int); 140 141 unsigned int nfs_realign_test = 0; 142 unsigned int nfs_realign_count = 0; 143 144 /* Initialize the RTT estimator state for a new mount point. */ 145 void 146 nfs_init_rtt(struct nfsmount *nmp) 147 { 148 int i; 149 150 for (i = 0; i < NFS_MAX_TIMER; i++) 151 nmp->nm_srtt[i] = NFS_INITRTT; 152 for (i = 0; i < NFS_MAX_TIMER; i++) 153 nmp->nm_sdrtt[i] = 0; 154 } 155 156 /* 157 * Update a mount point's RTT estimator state using data from the 158 * passed-in request. 159 * 160 * Use a gain of 0.125 on the mean and a gain of 0.25 on the deviation. 161 * 162 * NB: Since the timer resolution of NFS_HZ is so course, it can often 163 * result in r_rtt == 0. Since r_rtt == N means that the actual RTT is 164 * between N + dt and N + 2 - dt ticks, add 1 before calculating the 165 * update values. 166 */ 167 void 168 nfs_update_rtt(struct nfsreq *rep) 169 { 170 int t1 = rep->r_rtt + 1; 171 int index = nfs_ptimers[rep->r_procnum] - 1; 172 int *srtt = &rep->r_nmp->nm_srtt[index]; 173 int *sdrtt = &rep->r_nmp->nm_sdrtt[index]; 174 175 t1 -= *srtt >> 3; 176 *srtt += t1; 177 if (t1 < 0) 178 t1 = -t1; 179 t1 -= *sdrtt >> 2; 180 *sdrtt += t1; 181 } 182 183 /* 184 * Estimate RTO for an NFS RPC sent via an unreliable datagram. 185 * 186 * Use the mean and mean deviation of RTT for the appropriate type 187 * of RPC for the frequent RPCs and a default for the others. 188 * The justification for doing "other" this way is that these RPCs 189 * happen so infrequently that timer est. would probably be stale. 190 * Also, since many of these RPCs are non-idempotent, a conservative 191 * timeout is desired. 192 * 193 * getattr, lookup - A+2D 194 * read, write - A+4D 195 * other - nm_timeo 196 */ 197 int 198 nfs_estimate_rto(struct nfsmount *nmp, u_int32_t procnum) 199 { 200 enum nfs_rto_timers timer = nfs_ptimers[procnum]; 201 int index = timer - 1; 202 int rto; 203 204 switch (timer) { 205 case NFS_GETATTR_TIMER: 206 case NFS_LOOKUP_TIMER: 207 rto = ((nmp->nm_srtt[index] + 3) >> 2) + 208 ((nmp->nm_sdrtt[index] + 1) >> 1); 209 break; 210 case NFS_READ_TIMER: 211 case NFS_WRITE_TIMER: 212 rto = ((nmp->nm_srtt[index] + 7) >> 3) + 213 (nmp->nm_sdrtt[index] + 1); 214 break; 215 default: 216 rto = nmp->nm_timeo; 217 return (rto); 218 } 219 220 if (rto < NFS_MINRTO) 221 rto = NFS_MINRTO; 222 else if (rto > NFS_MAXRTO) 223 rto = NFS_MAXRTO; 224 225 return (rto); 226 } 227 228 229 230 /* 231 * Initialize sockets and congestion for a new NFS connection. 232 * We do not free the sockaddr if error. 233 */ 234 int 235 nfs_connect(struct nfsmount *nmp, struct nfsreq *rep) 236 { 237 struct socket *so; 238 int s, error, rcvreserve, sndreserve; 239 struct sockaddr *saddr; 240 struct sockaddr_in *sin; 241 struct mbuf *nam = NULL, *mopt = NULL; 242 243 if (!(nmp->nm_sotype == SOCK_DGRAM || nmp->nm_sotype == SOCK_STREAM)) 244 return (EINVAL); 245 246 nmp->nm_so = NULL; 247 saddr = mtod(nmp->nm_nam, struct sockaddr *); 248 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 249 nmp->nm_soproto); 250 if (error) { 251 nfs_disconnect(nmp); 252 return (error); 253 } 254 255 /* Allocate mbufs possibly waiting before grabbing the socket lock. */ 256 if (nmp->nm_sotype == SOCK_STREAM || saddr->sa_family == AF_INET) 257 MGET(mopt, M_WAIT, MT_SOOPTS); 258 if (saddr->sa_family == AF_INET) 259 MGET(nam, M_WAIT, MT_SONAME); 260 261 so = nmp->nm_so; 262 s = solock(so); 263 nmp->nm_soflags = so->so_proto->pr_flags; 264 265 /* 266 * Some servers require that the client port be a reserved port number. 267 * We always allocate a reserved port, as this prevents filehandle 268 * disclosure through UDP port capture. 269 */ 270 if (saddr->sa_family == AF_INET) { 271 int *ip; 272 273 mopt->m_len = sizeof(int); 274 ip = mtod(mopt, int *); 275 *ip = IP_PORTRANGE_LOW; 276 error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt); 277 if (error) 278 goto bad; 279 280 sin = mtod(nam, struct sockaddr_in *); 281 memset(sin, 0, sizeof(*sin)); 282 sin->sin_len = nam->m_len = sizeof(struct sockaddr_in); 283 sin->sin_family = AF_INET; 284 sin->sin_addr.s_addr = INADDR_ANY; 285 sin->sin_port = htons(0); 286 error = sobind(so, nam, &proc0); 287 if (error) 288 goto bad; 289 290 mopt->m_len = sizeof(int); 291 ip = mtod(mopt, int *); 292 *ip = IP_PORTRANGE_DEFAULT; 293 error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt); 294 if (error) 295 goto bad; 296 } 297 298 /* 299 * Protocols that do not require connections may be optionally left 300 * unconnected for servers that reply from a port other than NFS_PORT. 301 */ 302 if (nmp->nm_flag & NFSMNT_NOCONN) { 303 if (nmp->nm_soflags & PR_CONNREQUIRED) { 304 error = ENOTCONN; 305 goto bad; 306 } 307 } else { 308 error = soconnect(so, nmp->nm_nam); 309 if (error) 310 goto bad; 311 312 /* 313 * Wait for the connection to complete. Cribbed from the 314 * connect system call but with the wait timing out so 315 * that interruptible mounts don't hang here for a long time. 316 */ 317 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 318 sosleep(so, &so->so_timeo, PSOCK, "nfscon", 2 * hz); 319 if ((so->so_state & SS_ISCONNECTING) && 320 so->so_error == 0 && rep && 321 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ 322 so->so_state &= ~SS_ISCONNECTING; 323 goto bad; 324 } 325 } 326 if (so->so_error) { 327 error = so->so_error; 328 so->so_error = 0; 329 goto bad; 330 } 331 } 332 /* 333 * Always set receive timeout to detect server crash and reconnect. 334 * Otherwise, we can get stuck in soreceive forever. 335 */ 336 so->so_rcv.sb_timeo = (5 * hz); 337 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) 338 so->so_snd.sb_timeo = (5 * hz); 339 else 340 so->so_snd.sb_timeo = 0; 341 if (nmp->nm_sotype == SOCK_DGRAM) { 342 sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR; 343 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 344 NFS_MAXPKTHDR) * 2; 345 } else if (nmp->nm_sotype == SOCK_STREAM) { 346 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 347 *mtod(mopt, int32_t *) = 1; 348 mopt->m_len = sizeof(int32_t); 349 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, mopt); 350 } 351 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 352 *mtod(mopt, int32_t *) = 1; 353 mopt->m_len = sizeof(int32_t); 354 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, mopt); 355 } 356 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 357 sizeof (u_int32_t)) * 2; 358 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 359 sizeof (u_int32_t)) * 2; 360 } else { 361 panic("%s: nm_sotype %d", __func__, nmp->nm_sotype); 362 } 363 error = soreserve(so, sndreserve, rcvreserve); 364 if (error) 365 goto bad; 366 so->so_rcv.sb_flags |= SB_NOINTR; 367 so->so_snd.sb_flags |= SB_NOINTR; 368 sounlock(s); 369 370 m_freem(mopt); 371 m_freem(nam); 372 373 /* Initialize other non-zero congestion variables */ 374 nfs_init_rtt(nmp); 375 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 376 nmp->nm_sent = 0; 377 nmp->nm_timeouts = 0; 378 return (0); 379 380 bad: 381 sounlock(s); 382 383 m_freem(mopt); 384 m_freem(nam); 385 386 nfs_disconnect(nmp); 387 return (error); 388 } 389 390 /* 391 * Reconnect routine: 392 * Called when a connection is broken on a reliable protocol. 393 * - clean up the old socket 394 * - nfs_connect() again 395 * - set R_MUSTRESEND for all outstanding requests on mount point 396 * If this fails the mount point is DEAD! 397 * nb: Must be called with the nfs_sndlock() set on the mount point. 398 */ 399 int 400 nfs_reconnect(struct nfsreq *rep) 401 { 402 struct nfsreq *rp; 403 struct nfsmount *nmp = rep->r_nmp; 404 int error; 405 406 nfs_disconnect(nmp); 407 while ((error = nfs_connect(nmp, rep)) != 0) { 408 if (error == EINTR || error == ERESTART) 409 return (EINTR); 410 (void)tsleep((caddr_t)&lbolt, PSOCK, "nfsrecon", 0); 411 } 412 413 /* 414 * Loop through outstanding request list and fix up all requests 415 * on old socket. 416 */ 417 TAILQ_FOREACH(rp, &nmp->nm_reqsq, r_chain) { 418 rp->r_flags |= R_MUSTRESEND; 419 rp->r_rexmit = 0; 420 } 421 return (0); 422 } 423 424 /* 425 * NFS disconnect. Clean up and unlink. 426 */ 427 void 428 nfs_disconnect(struct nfsmount *nmp) 429 { 430 struct socket *so; 431 432 if (nmp->nm_so) { 433 so = nmp->nm_so; 434 nmp->nm_so = NULL; 435 soshutdown(so, SHUT_RDWR); 436 soclose(so); 437 } 438 } 439 440 /* 441 * This is the nfs send routine. For connection based socket types, it 442 * must be called with an nfs_sndlock() on the socket. 443 * "rep == NULL" indicates that it has been called from a server. 444 * For the client side: 445 * - return EINTR if the RPC is terminated, 0 otherwise 446 * - set R_MUSTRESEND if the send fails for any reason 447 * - do any cleanup required by recoverable socket errors (???) 448 * For the server side: 449 * - return EINTR or ERESTART if interrupted by a signal 450 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 451 * - do any cleanup required by recoverable socket errors (???) 452 */ 453 int 454 nfs_send(struct socket *so, struct mbuf *nam, struct mbuf *top, 455 struct nfsreq *rep) 456 { 457 struct mbuf *sendnam; 458 int error, soflags, flags; 459 460 if (rep) { 461 if (rep->r_flags & R_SOFTTERM) { 462 m_freem(top); 463 return (EINTR); 464 } 465 if ((so = rep->r_nmp->nm_so) == NULL) { 466 rep->r_flags |= R_MUSTRESEND; 467 m_freem(top); 468 return (0); 469 } 470 rep->r_flags &= ~R_MUSTRESEND; 471 soflags = rep->r_nmp->nm_soflags; 472 } else 473 soflags = so->so_proto->pr_flags; 474 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 475 sendnam = NULL; 476 else 477 sendnam = nam; 478 flags = 0; 479 480 error = sosend(so, sendnam, NULL, top, NULL, flags); 481 if (error) { 482 if (rep) { 483 /* 484 * Deal with errors for the client side. 485 */ 486 if (rep->r_flags & R_SOFTTERM) 487 error = EINTR; 488 else 489 rep->r_flags |= R_MUSTRESEND; 490 } 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 #ifdef NFSCLIENT 503 /* 504 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 505 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 506 * Mark and consolidate the data into a new mbuf list. 507 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 508 * small mbufs. 509 * For SOCK_STREAM we must be very careful to read an entire record once 510 * we have read any of it, even if the system call has been interrupted. 511 */ 512 int 513 nfs_receive(struct nfsreq *rep, struct mbuf **aname, struct mbuf **mp) 514 { 515 struct socket *so; 516 struct uio auio; 517 struct iovec aio; 518 struct mbuf *m; 519 struct mbuf *control; 520 u_int32_t len; 521 struct mbuf **getnam; 522 int error, sotype, rcvflg; 523 struct proc *p = curproc; /* XXX */ 524 525 /* 526 * Set up arguments for soreceive() 527 */ 528 *mp = NULL; 529 *aname = NULL; 530 sotype = rep->r_nmp->nm_sotype; 531 532 /* 533 * For reliable protocols, lock against other senders/receivers 534 * in case a reconnect is necessary. 535 * For SOCK_STREAM, first get the Record Mark to find out how much 536 * more there is to get. 537 * We must lock the socket against other receivers 538 * until we have an entire rpc request/reply. 539 */ 540 if (sotype != SOCK_DGRAM) { 541 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 542 if (error) 543 return (error); 544 tryagain: 545 /* 546 * Check for fatal errors and resending request. 547 */ 548 /* 549 * Ugh: If a reconnect attempt just happened, nm_so 550 * would have changed. NULL indicates a failed 551 * attempt that has essentially shut down this 552 * mount point. 553 */ 554 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 555 nfs_sndunlock(&rep->r_nmp->nm_flag); 556 return (EINTR); 557 } 558 so = rep->r_nmp->nm_so; 559 if (!so) { 560 error = nfs_reconnect(rep); 561 if (error) { 562 nfs_sndunlock(&rep->r_nmp->nm_flag); 563 return (error); 564 } 565 goto tryagain; 566 } 567 while (rep->r_flags & R_MUSTRESEND) { 568 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 569 nfsstats.rpcretries++; 570 rep->r_rtt = 0; 571 rep->r_flags &= ~R_TIMING; 572 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 573 if (error) { 574 if (error == EINTR || error == ERESTART || 575 (error = nfs_reconnect(rep)) != 0) { 576 nfs_sndunlock(&rep->r_nmp->nm_flag); 577 return (error); 578 } 579 goto tryagain; 580 } 581 } 582 nfs_sndunlock(&rep->r_nmp->nm_flag); 583 if (sotype == SOCK_STREAM) { 584 aio.iov_base = (caddr_t) &len; 585 aio.iov_len = sizeof(u_int32_t); 586 auio.uio_iov = &aio; 587 auio.uio_iovcnt = 1; 588 auio.uio_segflg = UIO_SYSSPACE; 589 auio.uio_rw = UIO_READ; 590 auio.uio_offset = 0; 591 auio.uio_resid = sizeof(u_int32_t); 592 auio.uio_procp = p; 593 do { 594 rcvflg = MSG_WAITALL; 595 error = soreceive(so, NULL, &auio, NULL, NULL, 596 &rcvflg, 0); 597 if (error == EWOULDBLOCK && rep) { 598 if (rep->r_flags & R_SOFTTERM) 599 return (EINTR); 600 /* 601 * looks like the server died after it 602 * received the request, make sure 603 * that we will retransmit and we 604 * don't get stuck here forever. 605 */ 606 if (rep->r_rexmit >= 607 rep->r_nmp->nm_retry) { 608 nfsstats.rpctimeouts++; 609 error = EPIPE; 610 } 611 } 612 } while (error == EWOULDBLOCK); 613 if (!error && auio.uio_resid > 0) { 614 log(LOG_INFO, 615 "short receive (%zu/%zu) from nfs server %s\n", 616 sizeof(u_int32_t) - auio.uio_resid, 617 sizeof(u_int32_t), 618 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 619 error = EPIPE; 620 } 621 if (error) 622 goto errout; 623 624 len = ntohl(len) & ~0x80000000; 625 /* 626 * This is SERIOUS! We are out of sync with the sender 627 * and forcing a disconnect/reconnect is all I can do. 628 */ 629 if (len > NFS_MAXPACKET) { 630 log(LOG_ERR, "%s (%u) from nfs server %s\n", 631 "impossible packet length", 632 len, 633 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 634 error = EFBIG; 635 goto errout; 636 } 637 auio.uio_resid = len; 638 do { 639 rcvflg = MSG_WAITALL; 640 error = soreceive(so, NULL, &auio, mp, NULL, 641 &rcvflg, 0); 642 } while (error == EWOULDBLOCK || error == EINTR || 643 error == ERESTART); 644 if (!error && auio.uio_resid > 0) { 645 log(LOG_INFO, "short receive (%zu/%u) from " 646 "nfs server %s\n", len - auio.uio_resid, 647 len, rep->r_nmp->nm_mountp-> 648 mnt_stat.f_mntfromname); 649 error = EPIPE; 650 } 651 } else { 652 /* 653 * NB: Since uio_resid is big, MSG_WAITALL is ignored 654 * and soreceive() will return when it has either a 655 * control msg or a data msg. 656 * We have no use for control msg., but must grab them 657 * and then throw them away so we know what is going 658 * on. 659 */ 660 auio.uio_resid = len = 100000000; /* Anything Big */ 661 auio.uio_procp = p; 662 do { 663 rcvflg = 0; 664 error = soreceive(so, NULL, &auio, mp, &control, 665 &rcvflg, 0); 666 m_freem(control); 667 if (error == EWOULDBLOCK && rep) { 668 if (rep->r_flags & R_SOFTTERM) 669 return (EINTR); 670 } 671 } while (error == EWOULDBLOCK || 672 (!error && *mp == NULL && control)); 673 if ((rcvflg & MSG_EOR) == 0) 674 printf("Egad!!\n"); 675 if (!error && *mp == NULL) 676 error = EPIPE; 677 len -= auio.uio_resid; 678 } 679 errout: 680 if (error && error != EINTR && error != ERESTART) { 681 m_freemp(mp); 682 if (error != EPIPE) 683 log(LOG_INFO, 684 "receive error %d from nfs server %s\n", 685 error, 686 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 687 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 688 if (!error) { 689 error = nfs_reconnect(rep); 690 if (!error) 691 goto tryagain; 692 nfs_sndunlock(&rep->r_nmp->nm_flag); 693 } 694 } 695 } else { 696 if ((so = rep->r_nmp->nm_so) == NULL) 697 return (EACCES); 698 if (so->so_state & SS_ISCONNECTED) 699 getnam = NULL; 700 else 701 getnam = aname; 702 auio.uio_resid = len = 1000000; 703 auio.uio_procp = p; 704 do { 705 rcvflg = 0; 706 error = soreceive(so, getnam, &auio, mp, NULL, 707 &rcvflg, 0); 708 if (error == EWOULDBLOCK && 709 (rep->r_flags & R_SOFTTERM)) 710 return (EINTR); 711 } while (error == EWOULDBLOCK); 712 len -= auio.uio_resid; 713 } 714 if (error) 715 m_freemp(mp); 716 /* 717 * Search for any mbufs that are not a multiple of 4 bytes long 718 * or with m_data not longword aligned. 719 * These could cause pointer alignment problems, so copy them to 720 * well aligned mbufs. 721 */ 722 nfs_realign(mp, 5 * NFSX_UNSIGNED); 723 return (error); 724 } 725 726 /* 727 * Implement receipt of reply on a socket. 728 * We must search through the list of received datagrams matching them 729 * with outstanding requests using the xid, until ours is found. 730 */ 731 int 732 nfs_reply(struct nfsreq *myrep) 733 { 734 struct nfsreq *rep; 735 struct nfsmount *nmp = myrep->r_nmp; 736 struct nfsm_info info; 737 struct mbuf *nam; 738 u_int32_t rxid, *tl, t1; 739 caddr_t cp2; 740 int error; 741 742 /* 743 * Loop around until we get our own reply 744 */ 745 for (;;) { 746 /* 747 * Lock against other receivers so that I don't get stuck in 748 * sbwait() after someone else has received my reply for me. 749 * Also necessary for connection based protocols to avoid 750 * race conditions during a reconnect. 751 */ 752 error = nfs_rcvlock(myrep); 753 if (error) 754 return (error == EALREADY ? 0 : error); 755 756 /* 757 * Get the next Rpc reply off the socket 758 */ 759 error = nfs_receive(myrep, &nam, &info.nmi_mrep); 760 nfs_rcvunlock(&nmp->nm_flag); 761 if (error) { 762 763 /* 764 * Ignore routing errors on connectionless protocols?? 765 */ 766 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 767 if (nmp->nm_so) 768 nmp->nm_so->so_error = 0; 769 continue; 770 } 771 return (error); 772 } 773 m_freem(nam); 774 775 /* 776 * Get the xid and check that it is an rpc reply 777 */ 778 info.nmi_md = info.nmi_mrep; 779 info.nmi_dpos = mtod(info.nmi_md, caddr_t); 780 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 781 rxid = *tl++; 782 if (*tl != rpc_reply) { 783 nfsstats.rpcinvalid++; 784 m_freem(info.nmi_mrep); 785 nfsmout: 786 continue; 787 } 788 789 /* 790 * Loop through the request list to match up the reply 791 * Iff no match, just drop the datagram 792 */ 793 TAILQ_FOREACH(rep, &nmp->nm_reqsq, r_chain) { 794 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 795 /* Found it.. */ 796 rep->r_mrep = info.nmi_mrep; 797 rep->r_md = info.nmi_md; 798 rep->r_dpos = info.nmi_dpos; 799 800 /* 801 * Update congestion window. 802 * Do the additive increase of 803 * one rpc/rtt. 804 */ 805 if (nmp->nm_cwnd <= nmp->nm_sent) { 806 nmp->nm_cwnd += 807 (NFS_CWNDSCALE * NFS_CWNDSCALE + 808 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 809 if (nmp->nm_cwnd > NFS_MAXCWND) 810 nmp->nm_cwnd = NFS_MAXCWND; 811 } 812 rep->r_flags &= ~R_SENT; 813 nmp->nm_sent -= NFS_CWNDSCALE; 814 815 if (rep->r_flags & R_TIMING) 816 nfs_update_rtt(rep); 817 818 nmp->nm_timeouts = 0; 819 break; 820 } 821 } 822 /* 823 * If not matched to a request, drop it. 824 * If it's mine, get out. 825 */ 826 if (rep == 0) { 827 nfsstats.rpcunexpected++; 828 m_freem(info.nmi_mrep); 829 } else if (rep == myrep) { 830 if (rep->r_mrep == NULL) 831 panic("nfsreply nil"); 832 return (0); 833 } 834 } 835 } 836 837 /* 838 * nfs_request - goes something like this 839 * - fill in request struct 840 * - links it into list 841 * - calls nfs_send() for first transmit 842 * - calls nfs_receive() to get reply 843 * - break down rpc header and return with nfs reply pointed to 844 * by mrep or error 845 * nb: always frees up mreq mbuf list 846 */ 847 int 848 nfs_request(struct vnode *vp, int procnum, struct nfsm_info *infop) 849 { 850 struct mbuf *m; 851 u_int32_t *tl; 852 struct nfsmount *nmp; 853 struct timeval tv; 854 caddr_t cp2; 855 int t1, i, error = 0; 856 int trylater_delay; 857 struct nfsreq *rep; 858 int mrest_len; 859 struct nfsm_info info; 860 861 rep = pool_get(&nfsreqpl, PR_WAITOK); 862 rep->r_nmp = VFSTONFS(vp->v_mount); 863 rep->r_vp = vp; 864 rep->r_procp = infop->nmi_procp; 865 rep->r_procnum = procnum; 866 867 mrest_len = 0; 868 m = infop->nmi_mreq; 869 while (m) { 870 mrest_len += m->m_len; 871 m = m->m_next; 872 } 873 874 /* empty mbuf for AUTH_UNIX header */ 875 rep->r_mreq = m_gethdr(M_WAIT, MT_DATA); 876 rep->r_mreq->m_next = infop->nmi_mreq; 877 rep->r_mreq->m_pkthdr.len = mrest_len; 878 879 trylater_delay = NFS_MINTIMEO; 880 881 nmp = rep->r_nmp; 882 883 /* Get the RPC header with authorization. */ 884 nfsm_rpchead(rep, infop->nmi_cred, RPCAUTH_UNIX); 885 m = rep->r_mreq; 886 887 /* 888 * For stream protocols, insert a Sun RPC Record Mark. 889 */ 890 if (nmp->nm_sotype == SOCK_STREAM) { 891 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 892 *mtod(m, u_int32_t *) = htonl(0x80000000 | 893 (m->m_pkthdr.len - NFSX_UNSIGNED)); 894 } 895 896 tryagain: 897 rep->r_rtt = rep->r_rexmit = 0; 898 if (nfs_ptimers[rep->r_procnum] != NFS_DEFAULT_TIMER) 899 rep->r_flags = R_TIMING; 900 else 901 rep->r_flags = 0; 902 rep->r_mrep = NULL; 903 904 /* 905 * Do the client side RPC. 906 */ 907 nfsstats.rpcrequests++; 908 /* 909 * Chain request into list of outstanding requests. Be sure 910 * to put it LAST so timer finds oldest requests first. 911 */ 912 if (TAILQ_EMPTY(&nmp->nm_reqsq)) 913 timeout_add(&nmp->nm_rtimeout, nfs_ticks); 914 TAILQ_INSERT_TAIL(&nmp->nm_reqsq, rep, r_chain); 915 916 /* 917 * If backing off another request or avoiding congestion, don't 918 * send this one now but let timer do it. If not timing a request, 919 * do it now. 920 */ 921 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 922 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 923 nmp->nm_sent < nmp->nm_cwnd)) { 924 if (nmp->nm_soflags & PR_CONNREQUIRED) 925 error = nfs_sndlock(&nmp->nm_flag, rep); 926 if (!error) { 927 error = nfs_send(nmp->nm_so, nmp->nm_nam, 928 m_copym(m, 0, M_COPYALL, M_WAIT), rep); 929 if (nmp->nm_soflags & PR_CONNREQUIRED) 930 nfs_sndunlock(&nmp->nm_flag); 931 } 932 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 933 nmp->nm_sent += NFS_CWNDSCALE; 934 rep->r_flags |= R_SENT; 935 } 936 } else { 937 rep->r_rtt = -1; 938 } 939 940 /* 941 * Wait for the reply from our send or the timer's. 942 */ 943 if (!error || error == EPIPE) 944 error = nfs_reply(rep); 945 946 /* 947 * RPC done, unlink the request. 948 */ 949 TAILQ_REMOVE(&nmp->nm_reqsq, rep, r_chain); 950 if (TAILQ_EMPTY(&nmp->nm_reqsq)) 951 timeout_del(&nmp->nm_rtimeout); 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 = trylater_delay; 1009 tv.tv_usec = 0; 1010 tsleep(&tv, PSOCK, "nfsretry", tvtohz(&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 >= MHLEN - max_hdr) { 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, error; 1135 1136 NET_LOCK(); 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->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, 1190 m, NULL, NULL, curproc); 1191 else 1192 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, 1193 m, 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 NET_UNLOCK(); 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) & 1237 ~p->p_p->ps_sigacts->ps_sigignore) & 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(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 #define ALIGN_POINTER(n) ((u_int)(((n) + sizeof(void *)) & ~sizeof(void *))) 1395 if (ALIGN_POINTER(m->m_len) >= MINCLSIZE) { 1396 MCLGET(n, M_WAIT); 1397 } 1398 n->m_len = 0; 1399 break; 1400 } 1401 pm = &m->m_next; 1402 } 1403 /* 1404 * If n is non-NULL, loop on m copying data, then replace the 1405 * portion of the chain that had to be realigned. 1406 */ 1407 if (n != NULL) { 1408 ++nfs_realign_count; 1409 while (m) { 1410 m_copyback(n, off, m->m_len, mtod(m, caddr_t), M_WAIT); 1411 1412 /* 1413 * If an unaligned amount of memory was copied, fix up 1414 * the last mbuf created by m_copyback(). 1415 */ 1416 if (!ALIGNED_POINTER(m->m_len, void *)) 1417 nfs_realign_fixup(m, n, &off); 1418 1419 off += m->m_len; 1420 m = m->m_next; 1421 } 1422 m_freemp(pm); 1423 *pm = n; 1424 } 1425 } 1426 1427 1428 /* 1429 * Parse an RPC request 1430 * - verify it 1431 * - fill in the cred struct. 1432 */ 1433 int 1434 nfs_getreq(struct nfsrv_descript *nd, struct nfsd *nfsd, int has_header) 1435 { 1436 int len, i; 1437 u_int32_t *tl; 1438 int32_t t1; 1439 caddr_t cp2; 1440 u_int32_t nfsvers, auth_type; 1441 int error = 0; 1442 struct nfsm_info info; 1443 1444 info.nmi_mrep = nd->nd_mrep; 1445 info.nmi_md = nd->nd_md; 1446 info.nmi_dpos = nd->nd_dpos; 1447 if (has_header) { 1448 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1449 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1450 if (*tl++ != rpc_call) { 1451 m_freem(info.nmi_mrep); 1452 return (EBADRPC); 1453 } 1454 } else 1455 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1456 nd->nd_repstat = 0; 1457 nd->nd_flag = 0; 1458 if (*tl++ != rpc_vers) { 1459 nd->nd_repstat = ERPCMISMATCH; 1460 nd->nd_procnum = NFSPROC_NOOP; 1461 return (0); 1462 } 1463 if (*tl != nfs_prog) { 1464 nd->nd_repstat = EPROGUNAVAIL; 1465 nd->nd_procnum = NFSPROC_NOOP; 1466 return (0); 1467 } 1468 tl++; 1469 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1470 if (nfsvers != NFS_VER2 && nfsvers != NFS_VER3) { 1471 nd->nd_repstat = EPROGMISMATCH; 1472 nd->nd_procnum = NFSPROC_NOOP; 1473 return (0); 1474 } 1475 if (nfsvers == NFS_VER3) 1476 nd->nd_flag = ND_NFSV3; 1477 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1478 if (nd->nd_procnum == NFSPROC_NULL) 1479 return (0); 1480 if (nd->nd_procnum >= NFS_NPROCS || 1481 (nd->nd_procnum > NFSPROC_COMMIT) || 1482 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1483 nd->nd_repstat = EPROCUNAVAIL; 1484 nd->nd_procnum = NFSPROC_NOOP; 1485 return (0); 1486 } 1487 if ((nd->nd_flag & ND_NFSV3) == 0) 1488 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1489 auth_type = *tl++; 1490 len = fxdr_unsigned(int, *tl++); 1491 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1492 m_freem(info.nmi_mrep); 1493 return (EBADRPC); 1494 } 1495 1496 /* Handle auth_unix */ 1497 if (auth_type == rpc_auth_unix) { 1498 len = fxdr_unsigned(int, *++tl); 1499 if (len < 0 || len > NFS_MAXNAMLEN) { 1500 m_freem(info.nmi_mrep); 1501 return (EBADRPC); 1502 } 1503 nfsm_adv(nfsm_rndup(len)); 1504 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1505 memset(&nd->nd_cr, 0, sizeof (struct ucred)); 1506 nd->nd_cr.cr_ref = 1; 1507 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1508 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1509 len = fxdr_unsigned(int, *tl); 1510 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1511 m_freem(info.nmi_mrep); 1512 return (EBADRPC); 1513 } 1514 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1515 for (i = 0; i < len; i++) { 1516 if (i < NGROUPS_MAX) 1517 nd->nd_cr.cr_groups[i] = 1518 fxdr_unsigned(gid_t, *tl++); 1519 else 1520 tl++; 1521 } 1522 nd->nd_cr.cr_ngroups = (len > NGROUPS_MAX) ? NGROUPS_MAX : len; 1523 len = fxdr_unsigned(int, *++tl); 1524 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1525 m_freem(info.nmi_mrep); 1526 return (EBADRPC); 1527 } 1528 if (len > 0) 1529 nfsm_adv(nfsm_rndup(len)); 1530 } else { 1531 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1532 nd->nd_procnum = NFSPROC_NOOP; 1533 return (0); 1534 } 1535 1536 nd->nd_md = info.nmi_md; 1537 nd->nd_dpos = info.nmi_dpos; 1538 return (0); 1539 nfsmout: 1540 return (error); 1541 } 1542 1543 void 1544 nfs_msg(struct nfsreq *rep, char *msg) 1545 { 1546 tpr_t tpr; 1547 1548 if (rep->r_procp) 1549 tpr = tprintf_open(rep->r_procp); 1550 else 1551 tpr = NULL; 1552 1553 tprintf(tpr, "nfs server %s: %s\n", 1554 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname, msg); 1555 tprintf_close(tpr); 1556 } 1557 1558 #ifdef NFSSERVER 1559 /* 1560 * Socket upcall routine for the nfsd sockets. 1561 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1562 * Essentially do as much as possible non-blocking, else punt and it will 1563 * be called with M_WAIT from an nfsd. 1564 */ 1565 void 1566 nfsrv_rcv(struct socket *so, caddr_t arg, int waitflag) 1567 { 1568 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1569 struct mbuf *m; 1570 struct mbuf *mp, *nam; 1571 struct uio auio; 1572 int flags, error; 1573 1574 if ((slp->ns_flag & SLP_VALID) == 0) 1575 return; 1576 1577 /* Defer soreceive() to an nfsd. */ 1578 if (waitflag == M_DONTWAIT) { 1579 slp->ns_flag |= SLP_NEEDQ; 1580 goto dorecs; 1581 } 1582 1583 auio.uio_procp = NULL; 1584 if (so->so_type == SOCK_STREAM) { 1585 /* 1586 * Do soreceive(). 1587 */ 1588 auio.uio_resid = 1000000000; 1589 flags = MSG_DONTWAIT; 1590 error = soreceive(so, &nam, &auio, &mp, NULL, 1591 &flags, 0); 1592 if (error || mp == NULL) { 1593 if (error == EWOULDBLOCK) 1594 slp->ns_flag |= SLP_NEEDQ; 1595 else 1596 slp->ns_flag |= SLP_DISCONN; 1597 goto dorecs; 1598 } 1599 m = mp; 1600 if (slp->ns_rawend) { 1601 slp->ns_rawend->m_next = m; 1602 slp->ns_cc += 1000000000 - auio.uio_resid; 1603 } else { 1604 slp->ns_raw = m; 1605 slp->ns_cc = 1000000000 - auio.uio_resid; 1606 } 1607 while (m->m_next) 1608 m = m->m_next; 1609 slp->ns_rawend = m; 1610 1611 /* 1612 * Now try and parse record(s) out of the raw stream data. 1613 */ 1614 error = nfsrv_getstream(slp, waitflag); 1615 if (error) { 1616 if (error == EPERM) 1617 slp->ns_flag |= SLP_DISCONN; 1618 else 1619 slp->ns_flag |= SLP_NEEDQ; 1620 } 1621 } else { 1622 do { 1623 auio.uio_resid = 1000000000; 1624 flags = MSG_DONTWAIT; 1625 error = soreceive(so, &nam, &auio, &mp, 1626 NULL, &flags, 0); 1627 if (mp) { 1628 if (nam) { 1629 m = nam; 1630 m->m_next = mp; 1631 } else 1632 m = mp; 1633 if (slp->ns_recend) 1634 slp->ns_recend->m_nextpkt = m; 1635 else 1636 slp->ns_rec = m; 1637 slp->ns_recend = m; 1638 m->m_nextpkt = NULL; 1639 } 1640 if (error) { 1641 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1642 && error != EWOULDBLOCK) { 1643 slp->ns_flag |= SLP_DISCONN; 1644 goto dorecs; 1645 } 1646 } 1647 } while (mp); 1648 } 1649 1650 /* 1651 * Now try and process the request records, non-blocking. 1652 */ 1653 dorecs: 1654 if (waitflag == M_DONTWAIT && 1655 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1656 nfsrv_wakenfsd(slp); 1657 } 1658 1659 /* 1660 * Try and extract an RPC request from the mbuf data list received on a 1661 * stream socket. The "waitflag" argument indicates whether or not it 1662 * can sleep. 1663 */ 1664 int 1665 nfsrv_getstream(struct nfssvc_sock *slp, int waitflag) 1666 { 1667 struct mbuf *m, **mpp; 1668 char *cp1, *cp2; 1669 int len; 1670 struct mbuf *om, *m2, *recm; 1671 u_int32_t recmark; 1672 1673 if (slp->ns_flag & SLP_GETSTREAM) 1674 return (0); 1675 slp->ns_flag |= SLP_GETSTREAM; 1676 for (;;) { 1677 if (slp->ns_reclen == 0) { 1678 if (slp->ns_cc < NFSX_UNSIGNED) { 1679 slp->ns_flag &= ~SLP_GETSTREAM; 1680 return (0); 1681 } 1682 m = slp->ns_raw; 1683 if (m->m_len >= NFSX_UNSIGNED) { 1684 bcopy(mtod(m, caddr_t), &recmark, 1685 NFSX_UNSIGNED); 1686 m->m_data += NFSX_UNSIGNED; 1687 m->m_len -= NFSX_UNSIGNED; 1688 } else { 1689 cp1 = (caddr_t)&recmark; 1690 cp2 = mtod(m, caddr_t); 1691 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1692 while (m->m_len == 0) { 1693 m = m->m_next; 1694 cp2 = mtod(m, caddr_t); 1695 } 1696 *cp1++ = *cp2++; 1697 m->m_data++; 1698 m->m_len--; 1699 } 1700 } 1701 slp->ns_cc -= NFSX_UNSIGNED; 1702 recmark = ntohl(recmark); 1703 slp->ns_reclen = recmark & ~0x80000000; 1704 if (recmark & 0x80000000) 1705 slp->ns_flag |= SLP_LASTFRAG; 1706 else 1707 slp->ns_flag &= ~SLP_LASTFRAG; 1708 if (slp->ns_reclen > NFS_MAXPACKET) { 1709 slp->ns_flag &= ~SLP_GETSTREAM; 1710 return (EPERM); 1711 } 1712 } 1713 1714 /* 1715 * Now get the record part. 1716 */ 1717 recm = NULL; 1718 if (slp->ns_cc == slp->ns_reclen) { 1719 recm = slp->ns_raw; 1720 slp->ns_raw = slp->ns_rawend = NULL; 1721 slp->ns_cc = slp->ns_reclen = 0; 1722 } else if (slp->ns_cc > slp->ns_reclen) { 1723 len = 0; 1724 m = slp->ns_raw; 1725 om = NULL; 1726 while (len < slp->ns_reclen) { 1727 if ((len + m->m_len) > slp->ns_reclen) { 1728 m2 = m_copym(m, 0, slp->ns_reclen - len, 1729 waitflag); 1730 if (m2) { 1731 if (om) { 1732 om->m_next = m2; 1733 recm = slp->ns_raw; 1734 } else 1735 recm = m2; 1736 m->m_data += slp->ns_reclen-len; 1737 m->m_len -= slp->ns_reclen-len; 1738 len = slp->ns_reclen; 1739 } else { 1740 slp->ns_flag &= ~SLP_GETSTREAM; 1741 return (EWOULDBLOCK); 1742 } 1743 } else if ((len + m->m_len) == slp->ns_reclen) { 1744 om = m; 1745 len += m->m_len; 1746 m = m->m_next; 1747 recm = slp->ns_raw; 1748 om->m_next = NULL; 1749 } else { 1750 om = m; 1751 len += m->m_len; 1752 m = m->m_next; 1753 } 1754 } 1755 slp->ns_raw = m; 1756 slp->ns_cc -= len; 1757 slp->ns_reclen = 0; 1758 } else { 1759 slp->ns_flag &= ~SLP_GETSTREAM; 1760 return (0); 1761 } 1762 1763 /* 1764 * Accumulate the fragments into a record. 1765 */ 1766 mpp = &slp->ns_frag; 1767 while (*mpp) 1768 mpp = &((*mpp)->m_next); 1769 *mpp = recm; 1770 if (slp->ns_flag & SLP_LASTFRAG) { 1771 if (slp->ns_recend) 1772 slp->ns_recend->m_nextpkt = slp->ns_frag; 1773 else 1774 slp->ns_rec = slp->ns_frag; 1775 slp->ns_recend = slp->ns_frag; 1776 slp->ns_frag = NULL; 1777 } 1778 } 1779 } 1780 1781 /* 1782 * Parse an RPC header. 1783 */ 1784 int 1785 nfsrv_dorec(struct nfssvc_sock *slp, struct nfsd *nfsd, 1786 struct nfsrv_descript **ndp) 1787 { 1788 struct mbuf *m, *nam; 1789 struct nfsrv_descript *nd; 1790 int error; 1791 1792 *ndp = NULL; 1793 if ((slp->ns_flag & SLP_VALID) == 0 || 1794 (m = slp->ns_rec) == NULL) 1795 return (ENOBUFS); 1796 slp->ns_rec = m->m_nextpkt; 1797 if (slp->ns_rec) 1798 m->m_nextpkt = NULL; 1799 else 1800 slp->ns_recend = NULL; 1801 if (m->m_type == MT_SONAME) { 1802 nam = m; 1803 m = m->m_next; 1804 nam->m_next = NULL; 1805 } else 1806 nam = NULL; 1807 nd = pool_get(&nfsrv_descript_pl, PR_WAITOK); 1808 nfs_realign(&m, 10 * NFSX_UNSIGNED); 1809 nd->nd_md = nd->nd_mrep = m; 1810 nd->nd_nam2 = nam; 1811 nd->nd_dpos = mtod(m, caddr_t); 1812 error = nfs_getreq(nd, nfsd, 1); 1813 if (error) { 1814 m_freem(nam); 1815 pool_put(&nfsrv_descript_pl, nd); 1816 return (error); 1817 } 1818 *ndp = nd; 1819 nfsd->nfsd_nd = nd; 1820 return (0); 1821 } 1822 1823 1824 /* 1825 * Search for a sleeping nfsd and wake it up. 1826 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 1827 * running nfsds will go look for the work in the nfssvc_sock list. 1828 */ 1829 void 1830 nfsrv_wakenfsd(struct nfssvc_sock *slp) 1831 { 1832 struct nfsd *nfsd; 1833 1834 if ((slp->ns_flag & SLP_VALID) == 0) 1835 return; 1836 1837 TAILQ_FOREACH(nfsd, &nfsd_head, nfsd_chain) { 1838 if (nfsd->nfsd_flag & NFSD_WAITING) { 1839 nfsd->nfsd_flag &= ~NFSD_WAITING; 1840 if (nfsd->nfsd_slp) 1841 panic("nfsd wakeup"); 1842 slp->ns_sref++; 1843 nfsd->nfsd_slp = slp; 1844 wakeup_one(nfsd); 1845 return; 1846 } 1847 } 1848 1849 slp->ns_flag |= SLP_DOREC; 1850 nfsd_head_flag |= NFSD_CHECKSLP; 1851 } 1852 #endif /* NFSSERVER */ 1853