1 /* $OpenBSD: nfs_socket.c,v 1.116 2017/05/17 08:59:05 mpi 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 *m; 242 243 nmp->nm_so = NULL; 244 saddr = mtod(nmp->nm_nam, struct sockaddr *); 245 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 246 nmp->nm_soproto); 247 if (error) 248 goto bad; 249 so = nmp->nm_so; 250 nmp->nm_soflags = so->so_proto->pr_flags; 251 252 /* 253 * Some servers require that the client port be a reserved port number. 254 * We always allocate a reserved port, as this prevents filehandle 255 * disclosure through UDP port capture. 256 */ 257 if (saddr->sa_family == AF_INET) { 258 struct mbuf *mopt; 259 int *ip; 260 261 MGET(mopt, M_WAIT, MT_SOOPTS); 262 mopt->m_len = sizeof(int); 263 ip = mtod(mopt, int *); 264 *ip = IP_PORTRANGE_LOW; 265 error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt); 266 if (error) 267 goto bad; 268 269 MGET(m, M_WAIT, MT_SONAME); 270 sin = mtod(m, struct sockaddr_in *); 271 memset(sin, 0, sizeof(*sin)); 272 sin->sin_len = m->m_len = sizeof(struct sockaddr_in); 273 sin->sin_family = AF_INET; 274 sin->sin_addr.s_addr = INADDR_ANY; 275 sin->sin_port = htons(0); 276 error = sobind(so, m, &proc0); 277 m_freem(m); 278 if (error) 279 goto bad; 280 281 MGET(mopt, M_WAIT, MT_SOOPTS); 282 mopt->m_len = sizeof(int); 283 ip = mtod(mopt, int *); 284 *ip = IP_PORTRANGE_DEFAULT; 285 error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt); 286 if (error) 287 goto bad; 288 } 289 290 /* 291 * Protocols that do not require connections may be optionally left 292 * unconnected for servers that reply from a port other than NFS_PORT. 293 */ 294 if (nmp->nm_flag & NFSMNT_NOCONN) { 295 if (nmp->nm_soflags & PR_CONNREQUIRED) { 296 error = ENOTCONN; 297 goto bad; 298 } 299 } else { 300 error = soconnect(so, nmp->nm_nam); 301 if (error) 302 goto bad; 303 304 /* 305 * Wait for the connection to complete. Cribbed from the 306 * connect system call but with the wait timing out so 307 * that interruptible mounts don't hang here for a long time. 308 */ 309 s = solock(so); 310 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 311 sosleep(so, &so->so_timeo, PSOCK, "nfscon", 2 * hz); 312 if ((so->so_state & SS_ISCONNECTING) && 313 so->so_error == 0 && rep && 314 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ 315 so->so_state &= ~SS_ISCONNECTING; 316 sounlock(s); 317 goto bad; 318 } 319 } 320 if (so->so_error) { 321 error = so->so_error; 322 so->so_error = 0; 323 sounlock(s); 324 goto bad; 325 } 326 sounlock(s); 327 } 328 /* 329 * Always set receive timeout to detect server crash and reconnect. 330 * Otherwise, we can get stuck in soreceive forever. 331 */ 332 so->so_rcv.sb_timeo = (5 * hz); 333 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) 334 so->so_snd.sb_timeo = (5 * hz); 335 else 336 so->so_snd.sb_timeo = 0; 337 if (nmp->nm_sotype == SOCK_DGRAM) { 338 sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR; 339 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 340 NFS_MAXPKTHDR) * 2; 341 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 342 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 343 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 344 NFS_MAXPKTHDR) * 2; 345 } else { 346 if (nmp->nm_sotype != SOCK_STREAM) 347 panic("nfscon sotype"); 348 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 349 MGET(m, M_WAIT, MT_SOOPTS); 350 *mtod(m, int32_t *) = 1; 351 m->m_len = sizeof(int32_t); 352 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 353 } 354 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 355 MGET(m, M_WAIT, MT_SOOPTS); 356 *mtod(m, int32_t *) = 1; 357 m->m_len = sizeof(int32_t); 358 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 359 } 360 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 361 sizeof (u_int32_t)) * 2; 362 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 363 sizeof (u_int32_t)) * 2; 364 } 365 error = soreserve(so, sndreserve, rcvreserve); 366 if (error) 367 goto bad; 368 so->so_rcv.sb_flags |= SB_NOINTR; 369 so->so_snd.sb_flags |= SB_NOINTR; 370 371 /* Initialize other non-zero congestion variables */ 372 nfs_init_rtt(nmp); 373 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 374 nmp->nm_sent = 0; 375 nmp->nm_timeouts = 0; 376 return (0); 377 378 bad: 379 nfs_disconnect(nmp); 380 return (error); 381 } 382 383 /* 384 * Reconnect routine: 385 * Called when a connection is broken on a reliable protocol. 386 * - clean up the old socket 387 * - nfs_connect() again 388 * - set R_MUSTRESEND for all outstanding requests on mount point 389 * If this fails the mount point is DEAD! 390 * nb: Must be called with the nfs_sndlock() set on the mount point. 391 */ 392 int 393 nfs_reconnect(struct nfsreq *rep) 394 { 395 struct nfsreq *rp; 396 struct nfsmount *nmp = rep->r_nmp; 397 int error; 398 399 nfs_disconnect(nmp); 400 while ((error = nfs_connect(nmp, rep)) != 0) { 401 if (error == EINTR || error == ERESTART) 402 return (EINTR); 403 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfsrecon", 0); 404 } 405 406 /* 407 * Loop through outstanding request list and fix up all requests 408 * on old socket. 409 */ 410 TAILQ_FOREACH(rp, &nmp->nm_reqsq, r_chain) { 411 rp->r_flags |= R_MUSTRESEND; 412 rp->r_rexmit = 0; 413 } 414 return (0); 415 } 416 417 /* 418 * NFS disconnect. Clean up and unlink. 419 */ 420 void 421 nfs_disconnect(struct nfsmount *nmp) 422 { 423 struct socket *so; 424 425 if (nmp->nm_so) { 426 so = nmp->nm_so; 427 nmp->nm_so = NULL; 428 soshutdown(so, SHUT_RDWR); 429 soclose(so); 430 } 431 } 432 433 /* 434 * This is the nfs send routine. For connection based socket types, it 435 * must be called with an nfs_sndlock() on the socket. 436 * "rep == NULL" indicates that it has been called from a server. 437 * For the client side: 438 * - return EINTR if the RPC is terminated, 0 otherwise 439 * - set R_MUSTRESEND if the send fails for any reason 440 * - do any cleanup required by recoverable socket errors (???) 441 * For the server side: 442 * - return EINTR or ERESTART if interrupted by a signal 443 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 444 * - do any cleanup required by recoverable socket errors (???) 445 */ 446 int 447 nfs_send(struct socket *so, struct mbuf *nam, struct mbuf *top, 448 struct nfsreq *rep) 449 { 450 struct mbuf *sendnam; 451 int error, soflags, flags; 452 453 if (rep) { 454 if (rep->r_flags & R_SOFTTERM) { 455 m_freem(top); 456 return (EINTR); 457 } 458 if ((so = rep->r_nmp->nm_so) == NULL) { 459 rep->r_flags |= R_MUSTRESEND; 460 m_freem(top); 461 return (0); 462 } 463 rep->r_flags &= ~R_MUSTRESEND; 464 soflags = rep->r_nmp->nm_soflags; 465 } else 466 soflags = so->so_proto->pr_flags; 467 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 468 sendnam = NULL; 469 else 470 sendnam = nam; 471 if (so->so_type == SOCK_SEQPACKET) 472 flags = MSG_EOR; 473 else 474 flags = 0; 475 476 error = sosend(so, sendnam, NULL, top, NULL, flags); 477 if (error) { 478 if (rep) { 479 /* 480 * Deal with errors for the client side. 481 */ 482 if (rep->r_flags & R_SOFTTERM) 483 error = EINTR; 484 else 485 rep->r_flags |= R_MUSTRESEND; 486 } 487 488 /* 489 * Handle any recoverable (soft) socket errors here. (???) 490 */ 491 if (error != EINTR && error != ERESTART && 492 error != EWOULDBLOCK && error != EPIPE) 493 error = 0; 494 } 495 return (error); 496 } 497 498 #ifdef NFSCLIENT 499 /* 500 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 501 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 502 * Mark and consolidate the data into a new mbuf list. 503 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 504 * small mbufs. 505 * For SOCK_STREAM we must be very careful to read an entire record once 506 * we have read any of it, even if the system call has been interrupted. 507 */ 508 int 509 nfs_receive(struct nfsreq *rep, struct mbuf **aname, struct mbuf **mp) 510 { 511 struct socket *so; 512 struct uio auio; 513 struct iovec aio; 514 struct mbuf *m; 515 struct mbuf *control; 516 u_int32_t len; 517 struct mbuf **getnam; 518 int error, sotype, rcvflg; 519 struct proc *p = curproc; /* XXX */ 520 521 /* 522 * Set up arguments for soreceive() 523 */ 524 *mp = NULL; 525 *aname = NULL; 526 sotype = rep->r_nmp->nm_sotype; 527 528 /* 529 * For reliable protocols, lock against other senders/receivers 530 * in case a reconnect is necessary. 531 * For SOCK_STREAM, first get the Record Mark to find out how much 532 * more there is to get. 533 * We must lock the socket against other receivers 534 * until we have an entire rpc request/reply. 535 */ 536 if (sotype != SOCK_DGRAM) { 537 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 538 if (error) 539 return (error); 540 tryagain: 541 /* 542 * Check for fatal errors and resending request. 543 */ 544 /* 545 * Ugh: If a reconnect attempt just happened, nm_so 546 * would have changed. NULL indicates a failed 547 * attempt that has essentially shut down this 548 * mount point. 549 */ 550 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 551 nfs_sndunlock(&rep->r_nmp->nm_flag); 552 return (EINTR); 553 } 554 so = rep->r_nmp->nm_so; 555 if (!so) { 556 error = nfs_reconnect(rep); 557 if (error) { 558 nfs_sndunlock(&rep->r_nmp->nm_flag); 559 return (error); 560 } 561 goto tryagain; 562 } 563 while (rep->r_flags & R_MUSTRESEND) { 564 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 565 nfsstats.rpcretries++; 566 rep->r_rtt = 0; 567 rep->r_flags &= ~R_TIMING; 568 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 569 if (error) { 570 if (error == EINTR || error == ERESTART || 571 (error = nfs_reconnect(rep)) != 0) { 572 nfs_sndunlock(&rep->r_nmp->nm_flag); 573 return (error); 574 } 575 goto tryagain; 576 } 577 } 578 nfs_sndunlock(&rep->r_nmp->nm_flag); 579 if (sotype == SOCK_STREAM) { 580 aio.iov_base = (caddr_t) &len; 581 aio.iov_len = sizeof(u_int32_t); 582 auio.uio_iov = &aio; 583 auio.uio_iovcnt = 1; 584 auio.uio_segflg = UIO_SYSSPACE; 585 auio.uio_rw = UIO_READ; 586 auio.uio_offset = 0; 587 auio.uio_resid = sizeof(u_int32_t); 588 auio.uio_procp = p; 589 do { 590 rcvflg = MSG_WAITALL; 591 error = soreceive(so, NULL, &auio, NULL, NULL, 592 &rcvflg, 0); 593 if (error == EWOULDBLOCK && rep) { 594 if (rep->r_flags & R_SOFTTERM) 595 return (EINTR); 596 /* 597 * looks like the server died after it 598 * received the request, make sure 599 * that we will retransmit and we 600 * don't get stuck here forever. 601 */ 602 if (rep->r_rexmit >= rep->r_nmp->nm_retry) { 603 nfsstats.rpctimeouts++; 604 error = EPIPE; 605 } 606 } 607 } while (error == EWOULDBLOCK); 608 if (!error && auio.uio_resid > 0) { 609 log(LOG_INFO, 610 "short receive (%zu/%zu) from nfs server %s\n", 611 sizeof(u_int32_t) - auio.uio_resid, 612 sizeof(u_int32_t), 613 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 614 error = EPIPE; 615 } 616 if (error) 617 goto errout; 618 619 len = ntohl(len) & ~0x80000000; 620 /* 621 * This is SERIOUS! We are out of sync with the sender 622 * and forcing a disconnect/reconnect is all I can do. 623 */ 624 if (len > NFS_MAXPACKET) { 625 log(LOG_ERR, "%s (%u) from nfs server %s\n", 626 "impossible packet length", 627 len, 628 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 629 error = EFBIG; 630 goto errout; 631 } 632 auio.uio_resid = len; 633 do { 634 rcvflg = MSG_WAITALL; 635 error = soreceive(so, NULL, &auio, mp, NULL, 636 &rcvflg, 0); 637 } while (error == EWOULDBLOCK || error == EINTR || 638 error == ERESTART); 639 if (!error && auio.uio_resid > 0) { 640 log(LOG_INFO, 641 "short receive (%zu/%u) from nfs server %s\n", 642 len - auio.uio_resid, len, 643 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 644 error = EPIPE; 645 } 646 } else { 647 /* 648 * NB: Since uio_resid is big, MSG_WAITALL is ignored 649 * and soreceive() will return when it has either a 650 * control msg or a data msg. 651 * We have no use for control msg., but must grab them 652 * and then throw them away so we know what is going 653 * on. 654 */ 655 auio.uio_resid = len = 100000000; /* Anything Big */ 656 auio.uio_procp = p; 657 do { 658 rcvflg = 0; 659 error = soreceive(so, NULL, &auio, mp, &control, 660 &rcvflg, 0); 661 m_freem(control); 662 if (error == EWOULDBLOCK && rep) { 663 if (rep->r_flags & R_SOFTTERM) 664 return (EINTR); 665 } 666 } while (error == EWOULDBLOCK || 667 (!error && *mp == NULL && control)); 668 if ((rcvflg & MSG_EOR) == 0) 669 printf("Egad!!\n"); 670 if (!error && *mp == NULL) 671 error = EPIPE; 672 len -= auio.uio_resid; 673 } 674 errout: 675 if (error && error != EINTR && error != ERESTART) { 676 m_freem(*mp); 677 *mp = NULL; 678 if (error != EPIPE) 679 log(LOG_INFO, 680 "receive error %d from nfs server %s\n", 681 error, 682 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 683 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 684 if (!error) { 685 error = nfs_reconnect(rep); 686 if (!error) 687 goto tryagain; 688 nfs_sndunlock(&rep->r_nmp->nm_flag); 689 } 690 } 691 } else { 692 if ((so = rep->r_nmp->nm_so) == NULL) 693 return (EACCES); 694 if (so->so_state & SS_ISCONNECTED) 695 getnam = NULL; 696 else 697 getnam = aname; 698 auio.uio_resid = len = 1000000; 699 auio.uio_procp = p; 700 do { 701 rcvflg = 0; 702 error = soreceive(so, getnam, &auio, mp, NULL, 703 &rcvflg, 0); 704 if (error == EWOULDBLOCK && 705 (rep->r_flags & R_SOFTTERM)) 706 return (EINTR); 707 } while (error == EWOULDBLOCK); 708 len -= auio.uio_resid; 709 } 710 if (error) { 711 m_freem(*mp); 712 *mp = NULL; 713 } 714 /* 715 * Search for any mbufs that are not a multiple of 4 bytes long 716 * or with m_data not longword aligned. 717 * These could cause pointer alignment problems, so copy them to 718 * well aligned mbufs. 719 */ 720 nfs_realign(mp, 5 * NFSX_UNSIGNED); 721 return (error); 722 } 723 724 /* 725 * Implement receipt of reply on a socket. 726 * We must search through the list of received datagrams matching them 727 * with outstanding requests using the xid, until ours is found. 728 */ 729 int 730 nfs_reply(struct nfsreq *myrep) 731 { 732 struct nfsreq *rep; 733 struct nfsmount *nmp = myrep->r_nmp; 734 struct nfsm_info info; 735 struct mbuf *nam; 736 u_int32_t rxid, *tl, t1; 737 caddr_t cp2; 738 int error; 739 740 /* 741 * Loop around until we get our own reply 742 */ 743 for (;;) { 744 /* 745 * Lock against other receivers so that I don't get stuck in 746 * sbwait() after someone else has received my reply for me. 747 * Also necessary for connection based protocols to avoid 748 * race conditions during a reconnect. 749 */ 750 error = nfs_rcvlock(myrep); 751 if (error) 752 return (error == EALREADY ? 0 : error); 753 754 /* 755 * Get the next Rpc reply off the socket 756 */ 757 error = nfs_receive(myrep, &nam, &info.nmi_mrep); 758 nfs_rcvunlock(&nmp->nm_flag); 759 if (error) { 760 761 /* 762 * Ignore routing errors on connectionless protocols?? 763 */ 764 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 765 if (nmp->nm_so) 766 nmp->nm_so->so_error = 0; 767 continue; 768 } 769 return (error); 770 } 771 m_freem(nam); 772 773 /* 774 * Get the xid and check that it is an rpc reply 775 */ 776 info.nmi_md = info.nmi_mrep; 777 info.nmi_dpos = mtod(info.nmi_md, caddr_t); 778 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 779 rxid = *tl++; 780 if (*tl != rpc_reply) { 781 nfsstats.rpcinvalid++; 782 m_freem(info.nmi_mrep); 783 nfsmout: 784 continue; 785 } 786 787 /* 788 * Loop through the request list to match up the reply 789 * Iff no match, just drop the datagram 790 */ 791 TAILQ_FOREACH(rep, &nmp->nm_reqsq, r_chain) { 792 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 793 /* Found it.. */ 794 rep->r_mrep = info.nmi_mrep; 795 rep->r_md = info.nmi_md; 796 rep->r_dpos = info.nmi_dpos; 797 798 /* 799 * Update congestion window. 800 * Do the additive increase of 801 * one rpc/rtt. 802 */ 803 if (nmp->nm_cwnd <= nmp->nm_sent) { 804 nmp->nm_cwnd += 805 (NFS_CWNDSCALE * NFS_CWNDSCALE + 806 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 807 if (nmp->nm_cwnd > NFS_MAXCWND) 808 nmp->nm_cwnd = NFS_MAXCWND; 809 } 810 rep->r_flags &= ~R_SENT; 811 nmp->nm_sent -= NFS_CWNDSCALE; 812 813 if (rep->r_flags & R_TIMING) 814 nfs_update_rtt(rep); 815 816 nmp->nm_timeouts = 0; 817 break; 818 } 819 } 820 /* 821 * If not matched to a request, drop it. 822 * If it's mine, get out. 823 */ 824 if (rep == 0) { 825 nfsstats.rpcunexpected++; 826 m_freem(info.nmi_mrep); 827 } else if (rep == myrep) { 828 if (rep->r_mrep == NULL) 829 panic("nfsreply nil"); 830 return (0); 831 } 832 } 833 } 834 835 /* 836 * nfs_request - goes something like this 837 * - fill in request struct 838 * - links it into list 839 * - calls nfs_send() for first transmit 840 * - calls nfs_receive() to get reply 841 * - break down rpc header and return with nfs reply pointed to 842 * by mrep or error 843 * nb: always frees up mreq mbuf list 844 */ 845 int 846 nfs_request(struct vnode *vp, int procnum, struct nfsm_info *infop) 847 { 848 struct mbuf *m; 849 u_int32_t *tl; 850 struct nfsmount *nmp; 851 struct timeval tv; 852 caddr_t cp2; 853 int t1, i, error = 0; 854 int trylater_delay; 855 struct nfsreq *rep; 856 int mrest_len; 857 struct nfsm_info info; 858 859 rep = pool_get(&nfsreqpl, PR_WAITOK); 860 rep->r_nmp = VFSTONFS(vp->v_mount); 861 rep->r_vp = vp; 862 rep->r_procp = infop->nmi_procp; 863 rep->r_procnum = procnum; 864 865 mrest_len = 0; 866 m = infop->nmi_mreq; 867 while (m) { 868 mrest_len += m->m_len; 869 m = m->m_next; 870 } 871 872 /* empty mbuf for AUTH_UNIX header */ 873 rep->r_mreq = m_gethdr(M_WAIT, MT_DATA); 874 rep->r_mreq->m_next = infop->nmi_mreq; 875 rep->r_mreq->m_pkthdr.len = mrest_len; 876 877 trylater_delay = NFS_MINTIMEO; 878 879 nmp = rep->r_nmp; 880 881 /* Get the RPC header with authorization. */ 882 nfsm_rpchead(rep, infop->nmi_cred, RPCAUTH_UNIX); 883 m = rep->r_mreq; 884 885 /* 886 * For stream protocols, insert a Sun RPC Record Mark. 887 */ 888 if (nmp->nm_sotype == SOCK_STREAM) { 889 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 890 *mtod(m, u_int32_t *) = htonl(0x80000000 | 891 (m->m_pkthdr.len - NFSX_UNSIGNED)); 892 } 893 894 tryagain: 895 rep->r_rtt = rep->r_rexmit = 0; 896 if (nfs_ptimers[rep->r_procnum] != NFS_DEFAULT_TIMER) 897 rep->r_flags = R_TIMING; 898 else 899 rep->r_flags = 0; 900 rep->r_mrep = NULL; 901 902 /* 903 * Do the client side RPC. 904 */ 905 nfsstats.rpcrequests++; 906 /* 907 * Chain request into list of outstanding requests. Be sure 908 * to put it LAST so timer finds oldest requests first. 909 */ 910 if (TAILQ_EMPTY(&nmp->nm_reqsq)) 911 timeout_add(&nmp->nm_rtimeout, nfs_ticks); 912 TAILQ_INSERT_TAIL(&nmp->nm_reqsq, rep, r_chain); 913 914 /* 915 * If backing off another request or avoiding congestion, don't 916 * send this one now but let timer do it. If not timing a request, 917 * do it now. 918 */ 919 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 920 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 921 nmp->nm_sent < nmp->nm_cwnd)) { 922 if (nmp->nm_soflags & PR_CONNREQUIRED) 923 error = nfs_sndlock(&nmp->nm_flag, rep); 924 if (!error) { 925 error = nfs_send(nmp->nm_so, nmp->nm_nam, 926 m_copym(m, 0, M_COPYALL, M_WAIT), 927 rep); 928 if (nmp->nm_soflags & PR_CONNREQUIRED) 929 nfs_sndunlock(&nmp->nm_flag); 930 } 931 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 932 nmp->nm_sent += NFS_CWNDSCALE; 933 rep->r_flags |= R_SENT; 934 } 935 } else { 936 rep->r_rtt = -1; 937 } 938 939 /* 940 * Wait for the reply from our send or the timer's. 941 */ 942 if (!error || error == EPIPE) 943 error = nfs_reply(rep); 944 945 /* 946 * RPC done, unlink the request. 947 */ 948 TAILQ_REMOVE(&nmp->nm_reqsq, rep, r_chain); 949 if (TAILQ_EMPTY(&nmp->nm_reqsq)) 950 timeout_del(&nmp->nm_rtimeout); 951 952 /* 953 * Decrement the outstanding request count. 954 */ 955 if (rep->r_flags & R_SENT) { 956 rep->r_flags &= ~R_SENT; /* paranoia */ 957 nmp->nm_sent -= NFS_CWNDSCALE; 958 } 959 960 /* 961 * If there was a successful reply and a tprintf msg. 962 * tprintf a response. 963 */ 964 if (!error && (rep->r_flags & R_TPRINTFMSG)) 965 nfs_msg(rep, "is alive again"); 966 info.nmi_mrep = rep->r_mrep; 967 info.nmi_md = rep->r_md; 968 info.nmi_dpos = rep->r_dpos; 969 if (error) { 970 infop->nmi_mrep = NULL; 971 goto nfsmout1; 972 } 973 974 /* 975 * break down the rpc header and check if ok 976 */ 977 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 978 if (*tl++ == rpc_msgdenied) { 979 if (*tl == rpc_mismatch) 980 error = EOPNOTSUPP; 981 else 982 error = EACCES; /* Should be EAUTH. */ 983 infop->nmi_mrep = NULL; 984 goto nfsmout1; 985 } 986 987 /* 988 * Since we only support RPCAUTH_UNIX atm we step over the 989 * reply verifer type, and in the (error) case that there really 990 * is any data in it, we advance over it. 991 */ 992 tl++; /* Step over verifer type */ 993 i = fxdr_unsigned(int32_t, *tl); 994 if (i > 0) 995 nfsm_adv(nfsm_rndup(i)); /* Should not happen */ 996 997 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 998 /* 0 == ok */ 999 if (*tl == 0) { 1000 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1001 if (*tl != 0) { 1002 error = fxdr_unsigned(int, *tl); 1003 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1004 error == NFSERR_TRYLATER) { 1005 m_freem(info.nmi_mrep); 1006 error = 0; 1007 tv.tv_sec = trylater_delay; 1008 tv.tv_usec = 0; 1009 tsleep(&tv, PSOCK, "nfsretry", tvtohz(&tv)); 1010 trylater_delay *= NFS_TIMEOUTMUL; 1011 if (trylater_delay > NFS_MAXTIMEO) 1012 trylater_delay = NFS_MAXTIMEO; 1013 1014 goto tryagain; 1015 } 1016 1017 /* 1018 * If the File Handle was stale, invalidate the 1019 * lookup cache, just in case. 1020 */ 1021 if (error == ESTALE) 1022 cache_purge(rep->r_vp); 1023 } 1024 goto nfsmout; 1025 } 1026 1027 error = EPROTONOSUPPORT; 1028 1029 nfsmout: 1030 infop->nmi_mrep = info.nmi_mrep; 1031 infop->nmi_md = info.nmi_md; 1032 infop->nmi_dpos = info.nmi_dpos; 1033 nfsmout1: 1034 m_freem(rep->r_mreq); 1035 pool_put(&nfsreqpl, rep); 1036 return (error); 1037 } 1038 #endif /* NFSCLIENT */ 1039 1040 /* 1041 * Generate the rpc reply header 1042 * siz arg. is used to decide if adding a cluster is worthwhile 1043 */ 1044 int 1045 nfs_rephead(int siz, struct nfsrv_descript *nd, struct nfssvc_sock *slp, 1046 int err, struct mbuf **mrq, struct mbuf **mbp) 1047 { 1048 u_int32_t *tl; 1049 struct mbuf *mreq; 1050 struct mbuf *mb; 1051 1052 MGETHDR(mreq, M_WAIT, MT_DATA); 1053 mb = mreq; 1054 /* 1055 * If this is a big reply, use a cluster else 1056 * try and leave leading space for the lower level headers. 1057 */ 1058 siz += RPC_REPLYSIZ; 1059 if (siz >= MHLEN - max_hdr) { 1060 MCLGET(mreq, M_WAIT); 1061 } else 1062 mreq->m_data += max_hdr; 1063 tl = mtod(mreq, u_int32_t *); 1064 mreq->m_len = 6 * NFSX_UNSIGNED; 1065 *tl++ = txdr_unsigned(nd->nd_retxid); 1066 *tl++ = rpc_reply; 1067 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1068 *tl++ = rpc_msgdenied; 1069 if (err & NFSERR_AUTHERR) { 1070 *tl++ = rpc_autherr; 1071 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1072 mreq->m_len -= NFSX_UNSIGNED; 1073 } else { 1074 *tl++ = rpc_mismatch; 1075 *tl++ = txdr_unsigned(RPC_VER2); 1076 *tl = txdr_unsigned(RPC_VER2); 1077 } 1078 } else { 1079 *tl++ = rpc_msgaccepted; 1080 1081 /* AUTH_UNIX requires RPCAUTH_NULL. */ 1082 *tl++ = 0; 1083 *tl++ = 0; 1084 1085 switch (err) { 1086 case EPROGUNAVAIL: 1087 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1088 break; 1089 case EPROGMISMATCH: 1090 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1091 tl = nfsm_build(&mb, 2 * NFSX_UNSIGNED); 1092 *tl++ = txdr_unsigned(NFS_VER2); 1093 *tl = txdr_unsigned(NFS_VER3); 1094 break; 1095 case EPROCUNAVAIL: 1096 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1097 break; 1098 case EBADRPC: 1099 *tl = txdr_unsigned(RPC_GARBAGE); 1100 break; 1101 default: 1102 *tl = 0; 1103 if (err != NFSERR_RETVOID) { 1104 tl = nfsm_build(&mb, NFSX_UNSIGNED); 1105 if (err) 1106 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1107 else 1108 *tl = 0; 1109 } 1110 break; 1111 }; 1112 } 1113 1114 *mrq = mreq; 1115 if (mbp != NULL) 1116 *mbp = mb; 1117 if (err != 0 && err != NFSERR_RETVOID) 1118 nfsstats.srvrpc_errs++; 1119 return (0); 1120 } 1121 1122 /* 1123 * nfs timer routine 1124 * Scan the nfsreq list and retranmit any requests that have timed out. 1125 */ 1126 void 1127 nfs_timer(void *arg) 1128 { 1129 struct nfsmount *nmp = arg; 1130 struct nfsreq *rep; 1131 struct mbuf *m; 1132 struct socket *so; 1133 int timeo, s, error; 1134 1135 NET_LOCK(s); 1136 TAILQ_FOREACH(rep, &nmp->nm_reqsq, r_chain) { 1137 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1138 continue; 1139 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1140 rep->r_flags |= R_SOFTTERM; 1141 continue; 1142 } 1143 if (rep->r_rtt >= 0) { 1144 rep->r_rtt++; 1145 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1146 timeo = nmp->nm_timeo; 1147 else 1148 timeo = nfs_estimate_rto(nmp, rep->r_procnum); 1149 if (nmp->nm_timeouts > 0) 1150 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1151 if (rep->r_rtt <= timeo) 1152 continue; 1153 if (nmp->nm_timeouts < nitems(nfs_backoff)) 1154 nmp->nm_timeouts++; 1155 } 1156 1157 /* Check for server not responding. */ 1158 if ((rep->r_flags & R_TPRINTFMSG) == 0 && rep->r_rexmit > 4) { 1159 nfs_msg(rep, "not responding"); 1160 rep->r_flags |= R_TPRINTFMSG; 1161 } 1162 if (rep->r_rexmit >= nmp->nm_retry) { /* too many */ 1163 nfsstats.rpctimeouts++; 1164 rep->r_flags |= R_SOFTTERM; 1165 continue; 1166 } 1167 if (nmp->nm_sotype != SOCK_DGRAM) { 1168 if (++rep->r_rexmit > NFS_MAXREXMIT) 1169 rep->r_rexmit = NFS_MAXREXMIT; 1170 continue; 1171 } 1172 1173 if ((so = nmp->nm_so) == NULL) 1174 continue; 1175 1176 /* 1177 * If there is enough space and the window allows.. 1178 * Resend it 1179 * Set r_rtt to -1 in case we fail to send it now. 1180 */ 1181 rep->r_rtt = -1; 1182 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1183 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1184 (rep->r_flags & R_SENT) || 1185 nmp->nm_sent < nmp->nm_cwnd) && 1186 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1187 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1188 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1189 NULL, NULL, curproc); 1190 else 1191 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1192 nmp->nm_nam, NULL, curproc); 1193 if (error) { 1194 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) 1195 so->so_error = 0; 1196 } else { 1197 /* 1198 * Iff first send, start timing 1199 * else turn timing off, backoff timer 1200 * and divide congestion window by 2. 1201 */ 1202 if (rep->r_flags & R_SENT) { 1203 rep->r_flags &= ~R_TIMING; 1204 if (++rep->r_rexmit > NFS_MAXREXMIT) 1205 rep->r_rexmit = NFS_MAXREXMIT; 1206 nmp->nm_cwnd >>= 1; 1207 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1208 nmp->nm_cwnd = NFS_CWNDSCALE; 1209 nfsstats.rpcretries++; 1210 } else { 1211 rep->r_flags |= R_SENT; 1212 nmp->nm_sent += NFS_CWNDSCALE; 1213 } 1214 rep->r_rtt = 0; 1215 } 1216 } 1217 } 1218 NET_UNLOCK(s); 1219 timeout_add(&nmp->nm_rtimeout, nfs_ticks); 1220 } 1221 1222 /* 1223 * Test for a termination condition pending on the process. 1224 * This is used for NFSMNT_INT mounts. 1225 */ 1226 int 1227 nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct proc *p) 1228 { 1229 1230 if (rep && (rep->r_flags & R_SOFTTERM)) 1231 return (EINTR); 1232 if (!(nmp->nm_flag & NFSMNT_INT)) 1233 return (0); 1234 if (p && p->p_siglist && 1235 (((p->p_siglist & ~p->p_sigmask) & 1236 ~p->p_p->ps_sigacts->ps_sigignore) & NFSINT_SIGMASK)) 1237 return (EINTR); 1238 return (0); 1239 } 1240 1241 /* 1242 * Lock a socket against others. 1243 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1244 * and also to avoid race conditions between the processes with nfs requests 1245 * in progress when a reconnect is necessary. 1246 */ 1247 int 1248 nfs_sndlock(int *flagp, struct nfsreq *rep) 1249 { 1250 struct proc *p; 1251 int slpflag = 0, slptimeo = 0; 1252 1253 if (rep) { 1254 p = rep->r_procp; 1255 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1256 slpflag = PCATCH; 1257 } else 1258 p = NULL; 1259 while (*flagp & NFSMNT_SNDLOCK) { 1260 if (rep && nfs_sigintr(rep->r_nmp, rep, p)) 1261 return (EINTR); 1262 *flagp |= NFSMNT_WANTSND; 1263 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1264 slptimeo); 1265 if (slpflag == PCATCH) { 1266 slpflag = 0; 1267 slptimeo = 2 * hz; 1268 } 1269 } 1270 *flagp |= NFSMNT_SNDLOCK; 1271 return (0); 1272 } 1273 1274 /* 1275 * Unlock the stream socket for others. 1276 */ 1277 void 1278 nfs_sndunlock(int *flagp) 1279 { 1280 1281 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1282 panic("nfs sndunlock"); 1283 *flagp &= ~NFSMNT_SNDLOCK; 1284 if (*flagp & NFSMNT_WANTSND) { 1285 *flagp &= ~NFSMNT_WANTSND; 1286 wakeup((caddr_t)flagp); 1287 } 1288 } 1289 1290 int 1291 nfs_rcvlock(struct nfsreq *rep) 1292 { 1293 int *flagp = &rep->r_nmp->nm_flag; 1294 int slpflag, slptimeo = 0; 1295 1296 if (*flagp & NFSMNT_INT) 1297 slpflag = PCATCH; 1298 else 1299 slpflag = 0; 1300 1301 while (*flagp & NFSMNT_RCVLOCK) { 1302 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1303 return (EINTR); 1304 *flagp |= NFSMNT_WANTRCV; 1305 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1306 slptimeo); 1307 if (rep->r_mrep != NULL) { 1308 /* 1309 * Don't take the lock if our reply has been received 1310 * while we where sleeping. 1311 */ 1312 return (EALREADY); 1313 } 1314 if (slpflag == PCATCH) { 1315 slpflag = 0; 1316 slptimeo = 2 * hz; 1317 } 1318 } 1319 *flagp |= NFSMNT_RCVLOCK; 1320 return (0); 1321 } 1322 1323 /* 1324 * Unlock the stream socket for others. 1325 */ 1326 void 1327 nfs_rcvunlock(int *flagp) 1328 { 1329 1330 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1331 panic("nfs rcvunlock"); 1332 *flagp &= ~NFSMNT_RCVLOCK; 1333 if (*flagp & NFSMNT_WANTRCV) { 1334 *flagp &= ~NFSMNT_WANTRCV; 1335 wakeup((caddr_t)flagp); 1336 } 1337 } 1338 1339 /* 1340 * Auxiliary routine to align the length of mbuf copies made with m_copyback(). 1341 */ 1342 void 1343 nfs_realign_fixup(struct mbuf *m, struct mbuf *n, unsigned int *off) 1344 { 1345 size_t padding; 1346 1347 /* 1348 * The maximum number of bytes that m_copyback() places in a mbuf is 1349 * always an aligned quantity, so realign happens at the chain's tail. 1350 */ 1351 while (n->m_next != NULL) 1352 n = n->m_next; 1353 1354 /* 1355 * Pad from the next elements in the source chain. Loop until the 1356 * destination chain is aligned, or the end of the source is reached. 1357 */ 1358 do { 1359 m = m->m_next; 1360 if (m == NULL) 1361 return; 1362 1363 padding = min(ALIGN(n->m_len) - n->m_len, m->m_len); 1364 if (padding > M_TRAILINGSPACE(n)) 1365 panic("nfs_realign_fixup: no memory to pad to"); 1366 1367 bcopy(mtod(m, void *), mtod(n, char *) + n->m_len, padding); 1368 1369 n->m_len += padding; 1370 m_adj(m, padding); 1371 *off += padding; 1372 1373 } while (!ALIGNED_POINTER(n->m_len, void *)); 1374 } 1375 1376 /* 1377 * The NFS RPC parsing code uses the data address and the length of mbuf 1378 * structures to calculate on-memory addresses. This function makes sure these 1379 * parameters are correctly aligned. 1380 */ 1381 void 1382 nfs_realign(struct mbuf **pm, int hsiz) 1383 { 1384 struct mbuf *m; 1385 struct mbuf *n = NULL; 1386 unsigned int off = 0; 1387 1388 ++nfs_realign_test; 1389 while ((m = *pm) != NULL) { 1390 if (!ALIGNED_POINTER(m->m_data, void *) || 1391 !ALIGNED_POINTER(m->m_len, void *)) { 1392 MGET(n, M_WAIT, MT_DATA); 1393 #define ALIGN_POINTER(n) ((u_int)(((n) + sizeof(void *)) & ~sizeof(void *))) 1394 if (ALIGN_POINTER(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), M_WAIT); 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 memset(&nd->nd_cr, 0, 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_MAX) 1516 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1517 else 1518 tl++; 1519 nd->nd_cr.cr_ngroups = (len > NGROUPS_MAX) ? NGROUPS_MAX : 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 1574 /* Defer soreceive() to an nfsd. */ 1575 if (waitflag == M_DONTWAIT) { 1576 slp->ns_flag |= SLP_NEEDQ; 1577 goto dorecs; 1578 } 1579 1580 auio.uio_procp = NULL; 1581 if (so->so_type == SOCK_STREAM) { 1582 /* 1583 * Do soreceive(). 1584 */ 1585 auio.uio_resid = 1000000000; 1586 flags = MSG_DONTWAIT; 1587 error = soreceive(so, &nam, &auio, &mp, NULL, 1588 &flags, 0); 1589 if (error || mp == NULL) { 1590 if (error == EWOULDBLOCK) 1591 slp->ns_flag |= SLP_NEEDQ; 1592 else 1593 slp->ns_flag |= SLP_DISCONN; 1594 goto dorecs; 1595 } 1596 m = mp; 1597 if (slp->ns_rawend) { 1598 slp->ns_rawend->m_next = m; 1599 slp->ns_cc += 1000000000 - auio.uio_resid; 1600 } else { 1601 slp->ns_raw = m; 1602 slp->ns_cc = 1000000000 - auio.uio_resid; 1603 } 1604 while (m->m_next) 1605 m = m->m_next; 1606 slp->ns_rawend = m; 1607 1608 /* 1609 * Now try and parse record(s) out of the raw stream data. 1610 */ 1611 error = nfsrv_getstream(slp, waitflag); 1612 if (error) { 1613 if (error == EPERM) 1614 slp->ns_flag |= SLP_DISCONN; 1615 else 1616 slp->ns_flag |= SLP_NEEDQ; 1617 } 1618 } else { 1619 do { 1620 auio.uio_resid = 1000000000; 1621 flags = MSG_DONTWAIT; 1622 error = soreceive(so, &nam, &auio, &mp, 1623 NULL, &flags, 0); 1624 if (mp) { 1625 if (nam) { 1626 m = nam; 1627 m->m_next = mp; 1628 } else 1629 m = mp; 1630 if (slp->ns_recend) 1631 slp->ns_recend->m_nextpkt = m; 1632 else 1633 slp->ns_rec = m; 1634 slp->ns_recend = m; 1635 m->m_nextpkt = NULL; 1636 } 1637 if (error) { 1638 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1639 && error != EWOULDBLOCK) { 1640 slp->ns_flag |= SLP_DISCONN; 1641 goto dorecs; 1642 } 1643 } 1644 } while (mp); 1645 } 1646 1647 /* 1648 * Now try and process the request records, non-blocking. 1649 */ 1650 dorecs: 1651 if (waitflag == M_DONTWAIT && 1652 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1653 nfsrv_wakenfsd(slp); 1654 } 1655 1656 /* 1657 * Try and extract an RPC request from the mbuf data list received on a 1658 * stream socket. The "waitflag" argument indicates whether or not it 1659 * can sleep. 1660 */ 1661 int 1662 nfsrv_getstream(struct nfssvc_sock *slp, int waitflag) 1663 { 1664 struct mbuf *m, **mpp; 1665 char *cp1, *cp2; 1666 int len; 1667 struct mbuf *om, *m2, *recm; 1668 u_int32_t recmark; 1669 1670 if (slp->ns_flag & SLP_GETSTREAM) 1671 return (0); 1672 slp->ns_flag |= SLP_GETSTREAM; 1673 for (;;) { 1674 if (slp->ns_reclen == 0) { 1675 if (slp->ns_cc < NFSX_UNSIGNED) { 1676 slp->ns_flag &= ~SLP_GETSTREAM; 1677 return (0); 1678 } 1679 m = slp->ns_raw; 1680 if (m->m_len >= NFSX_UNSIGNED) { 1681 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 1682 m->m_data += NFSX_UNSIGNED; 1683 m->m_len -= NFSX_UNSIGNED; 1684 } else { 1685 cp1 = (caddr_t)&recmark; 1686 cp2 = mtod(m, caddr_t); 1687 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1688 while (m->m_len == 0) { 1689 m = m->m_next; 1690 cp2 = mtod(m, caddr_t); 1691 } 1692 *cp1++ = *cp2++; 1693 m->m_data++; 1694 m->m_len--; 1695 } 1696 } 1697 slp->ns_cc -= NFSX_UNSIGNED; 1698 recmark = ntohl(recmark); 1699 slp->ns_reclen = recmark & ~0x80000000; 1700 if (recmark & 0x80000000) 1701 slp->ns_flag |= SLP_LASTFRAG; 1702 else 1703 slp->ns_flag &= ~SLP_LASTFRAG; 1704 if (slp->ns_reclen > NFS_MAXPACKET) { 1705 slp->ns_flag &= ~SLP_GETSTREAM; 1706 return (EPERM); 1707 } 1708 } 1709 1710 /* 1711 * Now get the record part. 1712 */ 1713 recm = NULL; 1714 if (slp->ns_cc == slp->ns_reclen) { 1715 recm = slp->ns_raw; 1716 slp->ns_raw = slp->ns_rawend = NULL; 1717 slp->ns_cc = slp->ns_reclen = 0; 1718 } else if (slp->ns_cc > slp->ns_reclen) { 1719 len = 0; 1720 m = slp->ns_raw; 1721 om = NULL; 1722 while (len < slp->ns_reclen) { 1723 if ((len + m->m_len) > slp->ns_reclen) { 1724 m2 = m_copym(m, 0, slp->ns_reclen - len, 1725 waitflag); 1726 if (m2) { 1727 if (om) { 1728 om->m_next = m2; 1729 recm = slp->ns_raw; 1730 } else 1731 recm = m2; 1732 m->m_data += slp->ns_reclen - len; 1733 m->m_len -= slp->ns_reclen - len; 1734 len = slp->ns_reclen; 1735 } else { 1736 slp->ns_flag &= ~SLP_GETSTREAM; 1737 return (EWOULDBLOCK); 1738 } 1739 } else if ((len + m->m_len) == slp->ns_reclen) { 1740 om = m; 1741 len += m->m_len; 1742 m = m->m_next; 1743 recm = slp->ns_raw; 1744 om->m_next = NULL; 1745 } else { 1746 om = m; 1747 len += m->m_len; 1748 m = m->m_next; 1749 } 1750 } 1751 slp->ns_raw = m; 1752 slp->ns_cc -= len; 1753 slp->ns_reclen = 0; 1754 } else { 1755 slp->ns_flag &= ~SLP_GETSTREAM; 1756 return (0); 1757 } 1758 1759 /* 1760 * Accumulate the fragments into a record. 1761 */ 1762 mpp = &slp->ns_frag; 1763 while (*mpp) 1764 mpp = &((*mpp)->m_next); 1765 *mpp = recm; 1766 if (slp->ns_flag & SLP_LASTFRAG) { 1767 if (slp->ns_recend) 1768 slp->ns_recend->m_nextpkt = slp->ns_frag; 1769 else 1770 slp->ns_rec = slp->ns_frag; 1771 slp->ns_recend = slp->ns_frag; 1772 slp->ns_frag = NULL; 1773 } 1774 } 1775 } 1776 1777 /* 1778 * Parse an RPC header. 1779 */ 1780 int 1781 nfsrv_dorec(struct nfssvc_sock *slp, struct nfsd *nfsd, 1782 struct nfsrv_descript **ndp) 1783 { 1784 struct mbuf *m, *nam; 1785 struct nfsrv_descript *nd; 1786 int error; 1787 1788 *ndp = NULL; 1789 if ((slp->ns_flag & SLP_VALID) == 0 || 1790 (m = slp->ns_rec) == NULL) 1791 return (ENOBUFS); 1792 slp->ns_rec = m->m_nextpkt; 1793 if (slp->ns_rec) 1794 m->m_nextpkt = NULL; 1795 else 1796 slp->ns_recend = NULL; 1797 if (m->m_type == MT_SONAME) { 1798 nam = m; 1799 m = m->m_next; 1800 nam->m_next = NULL; 1801 } else 1802 nam = NULL; 1803 nd = pool_get(&nfsrv_descript_pl, PR_WAITOK); 1804 nfs_realign(&m, 10 * NFSX_UNSIGNED); 1805 nd->nd_md = nd->nd_mrep = m; 1806 nd->nd_nam2 = nam; 1807 nd->nd_dpos = mtod(m, caddr_t); 1808 error = nfs_getreq(nd, nfsd, 1); 1809 if (error) { 1810 m_freem(nam); 1811 pool_put(&nfsrv_descript_pl, nd); 1812 return (error); 1813 } 1814 *ndp = nd; 1815 nfsd->nfsd_nd = nd; 1816 return (0); 1817 } 1818 1819 1820 /* 1821 * Search for a sleeping nfsd and wake it up. 1822 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 1823 * running nfsds will go look for the work in the nfssvc_sock list. 1824 */ 1825 void 1826 nfsrv_wakenfsd(struct nfssvc_sock *slp) 1827 { 1828 struct nfsd *nfsd; 1829 1830 if ((slp->ns_flag & SLP_VALID) == 0) 1831 return; 1832 1833 TAILQ_FOREACH(nfsd, &nfsd_head, nfsd_chain) { 1834 if (nfsd->nfsd_flag & NFSD_WAITING) { 1835 nfsd->nfsd_flag &= ~NFSD_WAITING; 1836 if (nfsd->nfsd_slp) 1837 panic("nfsd wakeup"); 1838 slp->ns_sref++; 1839 nfsd->nfsd_slp = slp; 1840 wakeup_one(nfsd); 1841 return; 1842 } 1843 } 1844 1845 slp->ns_flag |= SLP_DOREC; 1846 nfsd_head_flag |= NFSD_CHECKSLP; 1847 } 1848 #endif /* NFSSERVER */ 1849