1 /* $OpenBSD: nfs_socket.c,v 1.68 2008/09/12 15:41:40 thib Exp $ */ 2 /* $NetBSD: nfs_socket.c,v 1.27 1996/04/15 20:20:00 thorpej Exp $ */ 3 4 /* 5 * Copyright (c) 1989, 1991, 1993, 1995 6 * The Regents of the University of California. All rights reserved. 7 * 8 * This code is derived from software contributed to Berkeley by 9 * Rick Macklem at The University of Guelph. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 36 */ 37 38 /* 39 * Socket operations for use by nfs 40 */ 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/proc.h> 45 #include <sys/mount.h> 46 #include <sys/kernel.h> 47 #include <sys/mbuf.h> 48 #include <sys/vnode.h> 49 #include <sys/domain.h> 50 #include <sys/protosw.h> 51 #include <sys/socket.h> 52 #include <sys/socketvar.h> 53 #include <sys/syslog.h> 54 #include <sys/tprintf.h> 55 #include <sys/namei.h> 56 #include <sys/pool.h> 57 #include <sys/queue.h> 58 59 #include <netinet/in.h> 60 #include <netinet/tcp.h> 61 62 #include <nfs/rpcv2.h> 63 #include <nfs/nfsproto.h> 64 #include <nfs/nfs.h> 65 #include <nfs/xdr_subs.h> 66 #include <nfs/nfsm_subs.h> 67 #include <nfs/nfsmount.h> 68 #include <nfs/nfsnode.h> 69 #include <nfs/nfsrtt.h> 70 #include <nfs/nfs_var.h> 71 72 #define TRUE 1 73 #define FALSE 0 74 75 /* 76 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 77 * Use the mean and mean deviation of rtt for the appropriate type of rpc 78 * for the frequent rpcs and a default for the others. 79 * The justification for doing "other" this way is that these rpcs 80 * happen so infrequently that timer est. would probably be stale. 81 * Also, since many of these rpcs are 82 * non-idempotent, a conservative timeout is desired. 83 * getattr, lookup - A+2D 84 * read, write - A+4D 85 * other - nm_timeo 86 */ 87 #define NFS_RTO(n, t) \ 88 ((t) == 0 ? (n)->nm_timeo : \ 89 ((t) < 3 ? \ 90 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 91 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 92 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 93 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 94 /* 95 * External data, mostly RPC constants in XDR form 96 */ 97 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 98 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr; 99 extern u_int32_t nfs_prog; 100 extern struct nfsstats nfsstats; 101 extern int nfsv3_procid[NFS_NPROCS]; 102 extern int nfs_ticks; 103 104 /* 105 * Defines which timer to use for the procnum. 106 * 0 - default 107 * 1 - getattr 108 * 2 - lookup 109 * 3 - read 110 * 4 - write 111 */ 112 static int proct[NFS_NPROCS] = { 113 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0, 114 0, 0, 0, 115 }; 116 117 /* 118 * There is a congestion window for outstanding rpcs maintained per mount 119 * point. The cwnd size is adjusted in roughly the way that: 120 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 121 * SIGCOMM '88". ACM, August 1988. 122 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 123 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 124 * of rpcs is in progress. 125 * (The sent count and cwnd are scaled for integer arith.) 126 * Variants of "slow start" were tried and were found to be too much of a 127 * performance hit (ave. rtt 3 times larger), 128 * I suspect due to the large rtt that nfs rpcs have. 129 */ 130 #define NFS_CWNDSCALE 256 131 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 132 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 133 int nfsrtton = 0; 134 struct nfsrtt nfsrtt; 135 136 void nfs_realign(struct mbuf **, int); 137 void nfs_realign_fixup(struct mbuf *, struct mbuf *, unsigned int *); 138 unsigned int nfs_realign_test = 0; 139 unsigned int nfs_realign_count = 0; 140 141 struct nfsreqhead nfs_reqq; 142 143 /* 144 * Initialize sockets and congestion for a new NFS connection. 145 * We do not free the sockaddr if error. 146 */ 147 int 148 nfs_connect(nmp, rep) 149 struct nfsmount *nmp; 150 struct nfsreq *rep; 151 { 152 struct socket *so; 153 int s, error, rcvreserve, sndreserve; 154 struct sockaddr *saddr; 155 struct sockaddr_in *sin; 156 struct mbuf *m; 157 158 nmp->nm_so = (struct socket *)0; 159 saddr = mtod(nmp->nm_nam, struct sockaddr *); 160 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 161 nmp->nm_soproto); 162 if (error) 163 goto bad; 164 so = nmp->nm_so; 165 nmp->nm_soflags = so->so_proto->pr_flags; 166 167 /* 168 * Some servers require that the client port be a reserved port number. 169 * We always allocate a reserved port, as this prevents filehandle 170 * disclosure through UDP port capture. 171 */ 172 if (saddr->sa_family == AF_INET) { 173 struct mbuf *mopt; 174 int *ip; 175 176 MGET(mopt, M_WAIT, MT_SOOPTS); 177 mopt->m_len = sizeof(int); 178 ip = mtod(mopt, int *); 179 *ip = IP_PORTRANGE_LOW; 180 error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt); 181 if (error) 182 goto bad; 183 184 MGET(m, M_WAIT, MT_SONAME); 185 sin = mtod(m, struct sockaddr_in *); 186 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 187 sin->sin_family = AF_INET; 188 sin->sin_addr.s_addr = INADDR_ANY; 189 sin->sin_port = htons(0); 190 error = sobind(so, m, &proc0); 191 m_freem(m); 192 if (error) 193 goto bad; 194 195 MGET(mopt, M_WAIT, MT_SOOPTS); 196 mopt->m_len = sizeof(int); 197 ip = mtod(mopt, int *); 198 *ip = IP_PORTRANGE_DEFAULT; 199 error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt); 200 if (error) 201 goto bad; 202 } 203 204 /* 205 * Protocols that do not require connections may be optionally left 206 * unconnected for servers that reply from a port other than NFS_PORT. 207 */ 208 if (nmp->nm_flag & NFSMNT_NOCONN) { 209 if (nmp->nm_soflags & PR_CONNREQUIRED) { 210 error = ENOTCONN; 211 goto bad; 212 } 213 } else { 214 error = soconnect(so, nmp->nm_nam); 215 if (error) 216 goto bad; 217 218 /* 219 * Wait for the connection to complete. Cribbed from the 220 * connect system call but with the wait timing out so 221 * that interruptible mounts don't hang here for a long time. 222 */ 223 s = splsoftnet(); 224 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 225 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 226 "nfscon", 2 * hz); 227 if ((so->so_state & SS_ISCONNECTING) && 228 so->so_error == 0 && rep && 229 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ 230 so->so_state &= ~SS_ISCONNECTING; 231 splx(s); 232 goto bad; 233 } 234 } 235 if (so->so_error) { 236 error = so->so_error; 237 so->so_error = 0; 238 splx(s); 239 goto bad; 240 } 241 splx(s); 242 } 243 /* 244 * Always set receive timeout to detect server crash and reconnect. 245 * Otherwise, we can get stuck in soreceive forever. 246 */ 247 so->so_rcv.sb_timeo = (5 * hz); 248 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) 249 so->so_snd.sb_timeo = (5 * hz); 250 else 251 so->so_snd.sb_timeo = 0; 252 if (nmp->nm_sotype == SOCK_DGRAM) { 253 sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR; 254 rcvreserve = max(nmp->nm_rsize, nmp->nm_readdirsize) + 255 NFS_MAXPKTHDR; 256 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 257 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 258 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 259 NFS_MAXPKTHDR) * 2; 260 } else { 261 if (nmp->nm_sotype != SOCK_STREAM) 262 panic("nfscon sotype"); 263 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 264 MGET(m, M_WAIT, MT_SOOPTS); 265 *mtod(m, int32_t *) = 1; 266 m->m_len = sizeof(int32_t); 267 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 268 } 269 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 270 MGET(m, M_WAIT, MT_SOOPTS); 271 *mtod(m, int32_t *) = 1; 272 m->m_len = sizeof(int32_t); 273 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 274 } 275 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 276 sizeof (u_int32_t)) * 2; 277 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 278 sizeof (u_int32_t)) * 2; 279 } 280 error = soreserve(so, sndreserve, rcvreserve); 281 if (error) 282 goto bad; 283 so->so_rcv.sb_flags |= SB_NOINTR; 284 so->so_snd.sb_flags |= SB_NOINTR; 285 286 /* Initialize other non-zero congestion variables */ 287 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = 288 nmp->nm_srtt[3] = (NFS_TIMEO << 3); 289 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 290 nmp->nm_sdrtt[3] = 0; 291 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 292 nmp->nm_sent = 0; 293 nmp->nm_timeouts = 0; 294 return (0); 295 296 bad: 297 nfs_disconnect(nmp); 298 return (error); 299 } 300 301 /* 302 * Reconnect routine: 303 * Called when a connection is broken on a reliable protocol. 304 * - clean up the old socket 305 * - nfs_connect() again 306 * - set R_MUSTRESEND for all outstanding requests on mount point 307 * If this fails the mount point is DEAD! 308 * nb: Must be called with the nfs_sndlock() set on the mount point. 309 */ 310 int 311 nfs_reconnect(rep) 312 struct nfsreq *rep; 313 { 314 struct nfsreq *rp; 315 struct nfsmount *nmp = rep->r_nmp; 316 int error; 317 318 nfs_disconnect(nmp); 319 while ((error = nfs_connect(nmp, rep)) != 0) { 320 if (error == EINTR || error == ERESTART) 321 return (EINTR); 322 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); 323 } 324 325 /* 326 * Loop through outstanding request list and fix up all requests 327 * on old socket. 328 */ 329 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 330 if (rp->r_nmp == nmp) { 331 rp->r_flags |= R_MUSTRESEND; 332 rp->r_rexmit = 0; 333 } 334 } 335 return (0); 336 } 337 338 /* 339 * NFS disconnect. Clean up and unlink. 340 */ 341 void 342 nfs_disconnect(nmp) 343 struct nfsmount *nmp; 344 { 345 struct socket *so; 346 347 if (nmp->nm_so) { 348 so = nmp->nm_so; 349 nmp->nm_so = (struct socket *)0; 350 soshutdown(so, SHUT_RDWR); 351 soclose(so); 352 } 353 } 354 355 /* 356 * This is the nfs send routine. For connection based socket types, it 357 * must be called with an nfs_sndlock() on the socket. 358 * "rep == NULL" indicates that it has been called from a server. 359 * For the client side: 360 * - return EINTR if the RPC is terminated, 0 otherwise 361 * - set R_MUSTRESEND if the send fails for any reason 362 * - do any cleanup required by recoverable socket errors (???) 363 * For the server side: 364 * - return EINTR or ERESTART if interrupted by a signal 365 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 366 * - do any cleanup required by recoverable socket errors (???) 367 */ 368 int 369 nfs_send(so, nam, top, rep) 370 struct socket *so; 371 struct mbuf *nam; 372 struct mbuf *top; 373 struct nfsreq *rep; 374 { 375 struct mbuf *sendnam; 376 int error, soflags, flags; 377 378 if (rep) { 379 if (rep->r_flags & R_SOFTTERM) { 380 m_freem(top); 381 return (EINTR); 382 } 383 if ((so = rep->r_nmp->nm_so) == NULL) { 384 rep->r_flags |= R_MUSTRESEND; 385 m_freem(top); 386 return (0); 387 } 388 rep->r_flags &= ~R_MUSTRESEND; 389 soflags = rep->r_nmp->nm_soflags; 390 } else 391 soflags = so->so_proto->pr_flags; 392 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 393 sendnam = (struct mbuf *)0; 394 else 395 sendnam = nam; 396 if (so->so_type == SOCK_SEQPACKET) 397 flags = MSG_EOR; 398 else 399 flags = 0; 400 401 error = sosend(so, sendnam, (struct uio *)0, top, 402 (struct mbuf *)0, flags); 403 if (error) { 404 if (rep) { 405 /* 406 * Deal with errors for the client side. 407 */ 408 if (rep->r_flags & R_SOFTTERM) 409 error = EINTR; 410 else 411 rep->r_flags |= R_MUSTRESEND; 412 } 413 414 /* 415 * Handle any recoverable (soft) socket errors here. (???) 416 */ 417 if (error != EINTR && error != ERESTART && 418 error != EWOULDBLOCK && error != EPIPE) 419 error = 0; 420 } 421 return (error); 422 } 423 424 #ifdef NFSCLIENT 425 /* 426 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 427 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 428 * Mark and consolidate the data into a new mbuf list. 429 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 430 * small mbufs. 431 * For SOCK_STREAM we must be very careful to read an entire record once 432 * we have read any of it, even if the system call has been interrupted. 433 */ 434 int 435 nfs_receive(rep, aname, mp) 436 struct nfsreq *rep; 437 struct mbuf **aname; 438 struct mbuf **mp; 439 { 440 struct socket *so; 441 struct uio auio; 442 struct iovec aio; 443 struct mbuf *m; 444 struct mbuf *control; 445 u_int32_t len; 446 struct mbuf **getnam; 447 int error, sotype, rcvflg; 448 struct proc *p = curproc; /* XXX */ 449 450 /* 451 * Set up arguments for soreceive() 452 */ 453 *mp = (struct mbuf *)0; 454 *aname = (struct mbuf *)0; 455 sotype = rep->r_nmp->nm_sotype; 456 457 /* 458 * For reliable protocols, lock against other senders/receivers 459 * in case a reconnect is necessary. 460 * For SOCK_STREAM, first get the Record Mark to find out how much 461 * more there is to get. 462 * We must lock the socket against other receivers 463 * until we have an entire rpc request/reply. 464 */ 465 if (sotype != SOCK_DGRAM) { 466 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 467 if (error) 468 return (error); 469 tryagain: 470 /* 471 * Check for fatal errors and resending request. 472 */ 473 /* 474 * Ugh: If a reconnect attempt just happened, nm_so 475 * would have changed. NULL indicates a failed 476 * attempt that has essentially shut down this 477 * mount point. 478 */ 479 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 480 nfs_sndunlock(&rep->r_nmp->nm_flag); 481 return (EINTR); 482 } 483 so = rep->r_nmp->nm_so; 484 if (!so) { 485 error = nfs_reconnect(rep); 486 if (error) { 487 nfs_sndunlock(&rep->r_nmp->nm_flag); 488 return (error); 489 } 490 goto tryagain; 491 } 492 while (rep->r_flags & R_MUSTRESEND) { 493 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 494 nfsstats.rpcretries++; 495 rep->r_rtt = 0; 496 rep->r_flags &= ~R_TIMING; 497 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 498 if (error) { 499 if (error == EINTR || error == ERESTART || 500 (error = nfs_reconnect(rep)) != 0) { 501 nfs_sndunlock(&rep->r_nmp->nm_flag); 502 return (error); 503 } 504 goto tryagain; 505 } 506 } 507 nfs_sndunlock(&rep->r_nmp->nm_flag); 508 if (sotype == SOCK_STREAM) { 509 aio.iov_base = (caddr_t) &len; 510 aio.iov_len = sizeof(u_int32_t); 511 auio.uio_iov = &aio; 512 auio.uio_iovcnt = 1; 513 auio.uio_segflg = UIO_SYSSPACE; 514 auio.uio_rw = UIO_READ; 515 auio.uio_offset = 0; 516 auio.uio_resid = sizeof(u_int32_t); 517 auio.uio_procp = p; 518 do { 519 rcvflg = MSG_WAITALL; 520 error = soreceive(so, (struct mbuf **)0, &auio, 521 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 522 if (error == EWOULDBLOCK && rep) { 523 if (rep->r_flags & R_SOFTTERM) 524 return (EINTR); 525 /* 526 * looks like the server died after it 527 * received the request, make sure 528 * that we will retransmit and we 529 * don't get stuck here forever. 530 */ 531 if (rep->r_rexmit >= rep->r_nmp->nm_retry) { 532 nfsstats.rpctimeouts++; 533 error = EPIPE; 534 } 535 } 536 } while (error == EWOULDBLOCK); 537 if (!error && auio.uio_resid > 0) { 538 log(LOG_INFO, 539 "short receive (%d/%d) from nfs server %s\n", 540 sizeof(u_int32_t) - auio.uio_resid, 541 sizeof(u_int32_t), 542 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 543 error = EPIPE; 544 } 545 if (error) 546 goto errout; 547 548 len = ntohl(len) & ~0x80000000; 549 /* 550 * This is SERIOUS! We are out of sync with the sender 551 * and forcing a disconnect/reconnect is all I can do. 552 */ 553 if (len > NFS_MAXPACKET) { 554 log(LOG_ERR, "%s (%d) from nfs server %s\n", 555 "impossible packet length", 556 len, 557 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 558 error = EFBIG; 559 goto errout; 560 } 561 auio.uio_resid = len; 562 do { 563 rcvflg = MSG_WAITALL; 564 error = soreceive(so, (struct mbuf **)0, 565 &auio, mp, (struct mbuf **)0, &rcvflg); 566 } while (error == EWOULDBLOCK || error == EINTR || 567 error == ERESTART); 568 if (!error && auio.uio_resid > 0) { 569 log(LOG_INFO, 570 "short receive (%d/%d) from nfs server %s\n", 571 len - auio.uio_resid, len, 572 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 573 error = EPIPE; 574 } 575 } else { 576 /* 577 * NB: Since uio_resid is big, MSG_WAITALL is ignored 578 * and soreceive() will return when it has either a 579 * control msg or a data msg. 580 * We have no use for control msg., but must grab them 581 * and then throw them away so we know what is going 582 * on. 583 */ 584 auio.uio_resid = len = 100000000; /* Anything Big */ 585 auio.uio_procp = p; 586 do { 587 rcvflg = 0; 588 error = soreceive(so, (struct mbuf **)0, 589 &auio, mp, &control, &rcvflg); 590 if (control) 591 m_freem(control); 592 if (error == EWOULDBLOCK && rep) { 593 if (rep->r_flags & R_SOFTTERM) 594 return (EINTR); 595 } 596 } while (error == EWOULDBLOCK || 597 (!error && *mp == NULL && control)); 598 if ((rcvflg & MSG_EOR) == 0) 599 printf("Egad!!\n"); 600 if (!error && *mp == NULL) 601 error = EPIPE; 602 len -= auio.uio_resid; 603 } 604 errout: 605 if (error && error != EINTR && error != ERESTART) { 606 m_freem(*mp); 607 *mp = (struct mbuf *)0; 608 if (error != EPIPE) 609 log(LOG_INFO, 610 "receive error %d from nfs server %s\n", 611 error, 612 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 613 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 614 if (!error) { 615 error = nfs_reconnect(rep); 616 if (!error) 617 goto tryagain; 618 nfs_sndunlock(&rep->r_nmp->nm_flag); 619 } 620 } 621 } else { 622 if ((so = rep->r_nmp->nm_so) == NULL) 623 return (EACCES); 624 if (so->so_state & SS_ISCONNECTED) 625 getnam = (struct mbuf **)0; 626 else 627 getnam = aname; 628 auio.uio_resid = len = 1000000; 629 auio.uio_procp = p; 630 do { 631 rcvflg = 0; 632 error = soreceive(so, getnam, &auio, mp, 633 (struct mbuf **)0, &rcvflg); 634 if (error == EWOULDBLOCK && 635 (rep->r_flags & R_SOFTTERM)) 636 return (EINTR); 637 } while (error == EWOULDBLOCK); 638 len -= auio.uio_resid; 639 } 640 if (error) { 641 m_freem(*mp); 642 *mp = (struct mbuf *)0; 643 } 644 /* 645 * Search for any mbufs that are not a multiple of 4 bytes long 646 * or with m_data not longword aligned. 647 * These could cause pointer alignment problems, so copy them to 648 * well aligned mbufs. 649 */ 650 nfs_realign(mp, 5 * NFSX_UNSIGNED); 651 return (error); 652 } 653 654 /* 655 * Implement receipt of reply on a socket. 656 * We must search through the list of received datagrams matching them 657 * with outstanding requests using the xid, until ours is found. 658 */ 659 int 660 nfs_reply(myrep) 661 struct nfsreq *myrep; 662 { 663 struct nfsreq *rep; 664 struct nfsmount *nmp = myrep->r_nmp; 665 int32_t t1; 666 struct mbuf *mrep, *nam, *md; 667 u_int32_t rxid, *tl; 668 caddr_t dpos, cp2; 669 int error; 670 671 /* 672 * Loop around until we get our own reply 673 */ 674 for (;;) { 675 /* 676 * Lock against other receivers so that I don't get stuck in 677 * sbwait() after someone else has received my reply for me. 678 * Also necessary for connection based protocols to avoid 679 * race conditions during a reconnect. 680 */ 681 error = nfs_rcvlock(myrep); 682 if (error) 683 return (error == EALREADY ? 0 : error); 684 685 /* 686 * Get the next Rpc reply off the socket 687 */ 688 error = nfs_receive(myrep, &nam, &mrep); 689 nfs_rcvunlock(&nmp->nm_flag); 690 if (error) { 691 692 /* 693 * Ignore routing errors on connectionless protocols?? 694 */ 695 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 696 if (nmp->nm_so) 697 nmp->nm_so->so_error = 0; 698 continue; 699 } 700 return (error); 701 } 702 if (nam) 703 m_freem(nam); 704 705 /* 706 * Get the xid and check that it is an rpc reply 707 */ 708 md = mrep; 709 dpos = mtod(md, caddr_t); 710 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 711 rxid = *tl++; 712 if (*tl != rpc_reply) { 713 nfsstats.rpcinvalid++; 714 m_freem(mrep); 715 nfsmout: 716 continue; 717 } 718 719 /* 720 * Loop through the request list to match up the reply 721 * Iff no match, just drop the datagram 722 */ 723 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 724 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 725 /* Found it.. */ 726 rep->r_mrep = mrep; 727 rep->r_md = md; 728 rep->r_dpos = dpos; 729 if (nfsrtton) { 730 struct rttl *rt; 731 732 rt = &nfsrtt.rttl[nfsrtt.pos]; 733 rt->proc = rep->r_procnum; 734 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 735 rt->sent = nmp->nm_sent; 736 rt->cwnd = nmp->nm_cwnd; 737 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 738 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 739 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 740 getmicrotime(&rt->tstamp); 741 if (rep->r_flags & R_TIMING) 742 rt->rtt = rep->r_rtt; 743 else 744 rt->rtt = 1000000; 745 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 746 } 747 /* 748 * Update congestion window. 749 * Do the additive increase of 750 * one rpc/rtt. 751 */ 752 if (nmp->nm_cwnd <= nmp->nm_sent) { 753 nmp->nm_cwnd += 754 (NFS_CWNDSCALE * NFS_CWNDSCALE + 755 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 756 if (nmp->nm_cwnd > NFS_MAXCWND) 757 nmp->nm_cwnd = NFS_MAXCWND; 758 } 759 rep->r_flags &= ~R_SENT; 760 nmp->nm_sent -= NFS_CWNDSCALE; 761 /* 762 * Update rtt using a gain of 0.125 on the mean 763 * and a gain of 0.25 on the deviation. 764 */ 765 if (rep->r_flags & R_TIMING) { 766 /* 767 * Since the timer resolution of 768 * NFS_HZ is so course, it can often 769 * result in r_rtt == 0. Since 770 * r_rtt == N means that the actual 771 * rtt is between N+dt and N+2-dt ticks, 772 * add 1. 773 */ 774 t1 = rep->r_rtt + 1; 775 t1 -= (NFS_SRTT(rep) >> 3); 776 NFS_SRTT(rep) += t1; 777 if (t1 < 0) 778 t1 = -t1; 779 t1 -= (NFS_SDRTT(rep) >> 2); 780 NFS_SDRTT(rep) += t1; 781 } 782 nmp->nm_timeouts = 0; 783 break; 784 } 785 } 786 /* 787 * If not matched to a request, drop it. 788 * If it's mine, get out. 789 */ 790 if (rep == 0) { 791 nfsstats.rpcunexpected++; 792 m_freem(mrep); 793 } else if (rep == myrep) { 794 if (rep->r_mrep == NULL) 795 panic("nfsreply nil"); 796 return (0); 797 } 798 } 799 } 800 801 /* 802 * nfs_request - goes something like this 803 * - fill in request struct 804 * - links it into list 805 * - calls nfs_send() for first transmit 806 * - calls nfs_receive() to get reply 807 * - break down rpc header and return with nfs reply pointed to 808 * by mrep or error 809 * nb: always frees up mreq mbuf list 810 */ 811 int 812 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 813 struct vnode *vp; 814 struct mbuf *mrest; 815 int procnum; 816 struct proc *procp; 817 struct ucred *cred; 818 struct mbuf **mrp; 819 struct mbuf **mdp; 820 caddr_t *dposp; 821 { 822 struct mbuf *m, *mrep; 823 struct nfsreq *rep; 824 u_int32_t *tl; 825 int i; 826 struct nfsmount *nmp; 827 struct mbuf *md; 828 time_t waituntil; 829 caddr_t dpos, cp2; 830 int t1, s, error = 0, mrest_len; 831 int trylater_delay; 832 833 trylater_delay = NFS_MINTIMEO; 834 835 nmp = VFSTONFS(vp->v_mount); 836 rep = pool_get(&nfsreqpl, PR_WAITOK); 837 rep->r_nmp = nmp; 838 rep->r_vp = vp; 839 rep->r_procp = procp; 840 rep->r_procnum = procnum; 841 i = 0; 842 m = mrest; 843 while (m) { 844 i += m->m_len; 845 m = m->m_next; 846 } 847 mrest_len = i; 848 849 /* Get the RPC header with authorization. */ 850 nfsm_rpchead(rep, cred, RPCAUTH_UNIX, mrest, mrest_len); 851 m = rep->r_mreq; 852 853 /* 854 * For stream protocols, insert a Sun RPC Record Mark. 855 */ 856 if (nmp->nm_sotype == SOCK_STREAM) { 857 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 858 *mtod(m, u_int32_t *) = htonl(0x80000000 | 859 (m->m_pkthdr.len - NFSX_UNSIGNED)); 860 } 861 862 tryagain: 863 if (nmp->nm_flag & NFSMNT_SOFT) 864 rep->r_retry = nmp->nm_retry; 865 else 866 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 867 rep->r_rtt = rep->r_rexmit = 0; 868 if (proct[procnum] > 0) 869 rep->r_flags = R_TIMING; 870 else 871 rep->r_flags = 0; 872 rep->r_mrep = NULL; 873 874 /* 875 * Do the client side RPC. 876 */ 877 nfsstats.rpcrequests++; 878 /* 879 * Chain request into list of outstanding requests. Be sure 880 * to put it LAST so timer finds oldest requests first. 881 */ 882 s = splsoftnet(); 883 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 884 885 /* 886 * If backing off another request or avoiding congestion, don't 887 * send this one now but let timer do it. If not timing a request, 888 * do it now. 889 */ 890 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 891 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 892 nmp->nm_sent < nmp->nm_cwnd)) { 893 splx(s); 894 if (nmp->nm_soflags & PR_CONNREQUIRED) 895 error = nfs_sndlock(&nmp->nm_flag, rep); 896 if (!error) { 897 error = nfs_send(nmp->nm_so, nmp->nm_nam, 898 m_copym(m, 0, M_COPYALL, M_WAIT), 899 rep); 900 if (nmp->nm_soflags & PR_CONNREQUIRED) 901 nfs_sndunlock(&nmp->nm_flag); 902 } 903 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 904 nmp->nm_sent += NFS_CWNDSCALE; 905 rep->r_flags |= R_SENT; 906 } 907 } else { 908 splx(s); 909 rep->r_rtt = -1; 910 } 911 912 /* 913 * Wait for the reply from our send or the timer's. 914 */ 915 if (!error || error == EPIPE) 916 error = nfs_reply(rep); 917 918 /* 919 * RPC done, unlink the request. 920 */ 921 s = splsoftnet(); 922 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 923 splx(s); 924 925 /* 926 * Decrement the outstanding request count. 927 */ 928 if (rep->r_flags & R_SENT) { 929 rep->r_flags &= ~R_SENT; /* paranoia */ 930 nmp->nm_sent -= NFS_CWNDSCALE; 931 } 932 933 /* 934 * If there was a successful reply and a tprintf msg. 935 * tprintf a response. 936 */ 937 if (!error && (rep->r_flags & R_TPRINTFMSG)) 938 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 939 "is alive again"); 940 mrep = rep->r_mrep; 941 md = rep->r_md; 942 dpos = rep->r_dpos; 943 if (error) { 944 m_freem(rep->r_mreq); 945 pool_put(&nfsreqpl, rep); 946 return (error); 947 } 948 949 /* 950 * break down the rpc header and check if ok 951 */ 952 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 953 if (*tl++ == rpc_msgdenied) { 954 if (*tl == rpc_mismatch) 955 error = EOPNOTSUPP; 956 else 957 error = EACCES; 958 m_freem(mrep); 959 m_freem(rep->r_mreq); 960 pool_put(&nfsreqpl, rep); 961 return (error); 962 } 963 964 /* 965 * Since we only support RPCAUTH_UNIX atm we step over the 966 * reply verifer type, and if the (error) case that there really 967 * is any data init, we advance over it. 968 */ 969 tl++; /* Step over verifer type */ 970 i = fxdr_unsigned(int32_t, *tl); 971 if (i > 0) 972 nfsm_adv(nfsm_rndup(i)); /* Should not happen */ 973 974 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 975 /* 0 == ok */ 976 if (*tl == 0) { 977 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 978 if (*tl != 0) { 979 error = fxdr_unsigned(int, *tl); 980 if ((nmp->nm_flag & NFSMNT_NFSV3) && 981 error == NFSERR_TRYLATER) { 982 m_freem(mrep); 983 error = 0; 984 waituntil = time_second + trylater_delay; 985 while (time_second < waituntil) 986 (void) tsleep((caddr_t)&lbolt, 987 PSOCK, "nqnfstry", 0); 988 trylater_delay *= NFS_TIMEOUTMUL; 989 if (trylater_delay > NFS_MAXTIMEO) 990 trylater_delay = NFS_MAXTIMEO; 991 992 goto tryagain; 993 } 994 995 /* 996 * If the File Handle was stale, invalidate the 997 * lookup cache, just in case. 998 */ 999 if (error == ESTALE) 1000 cache_purge(vp); 1001 1002 if (nmp->nm_flag & NFSMNT_NFSV3 || error == ESTALE) { 1003 *mrp = mrep; 1004 *mdp = md; 1005 *dposp = dpos; 1006 error |= NFSERR_RETERR; 1007 } else 1008 m_freem(mrep); 1009 m_freem(rep->r_mreq); 1010 pool_put(&nfsreqpl, rep); 1011 return (error); 1012 } 1013 1014 *mrp = mrep; 1015 *mdp = md; 1016 *dposp = dpos; 1017 m_freem(rep->r_mreq); 1018 pool_put(&nfsreqpl, rep); 1019 return (0); 1020 } 1021 m_freem(mrep); 1022 error = EPROTONOSUPPORT; 1023 nfsmout: 1024 m_freem(rep->r_mreq); 1025 pool_put(&nfsreqpl, rep); 1026 return (error); 1027 } 1028 #endif /* NFSCLIENT */ 1029 1030 /* 1031 * Generate the rpc reply header 1032 * siz arg. is used to decide if adding a cluster is worthwhile 1033 */ 1034 int 1035 nfs_rephead(siz, nd, slp, err, mrq, mbp) 1036 int siz; 1037 struct nfsrv_descript *nd; 1038 struct nfssvc_sock *slp; 1039 int err; 1040 struct mbuf **mrq; 1041 struct mbuf **mbp; 1042 { 1043 u_int32_t *tl; 1044 struct mbuf *mreq; 1045 struct mbuf *mb; 1046 1047 MGETHDR(mreq, M_WAIT, MT_DATA); 1048 mb = mreq; 1049 /* 1050 * If this is a big reply, use a cluster else 1051 * try and leave leading space for the lower level headers. 1052 */ 1053 siz += RPC_REPLYSIZ; 1054 if (siz >= max_datalen) { 1055 MCLGET(mreq, M_WAIT); 1056 } else 1057 mreq->m_data += max_hdr; 1058 tl = mtod(mreq, u_int32_t *); 1059 mreq->m_len = 6 * NFSX_UNSIGNED; 1060 *tl++ = txdr_unsigned(nd->nd_retxid); 1061 *tl++ = rpc_reply; 1062 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1063 *tl++ = rpc_msgdenied; 1064 if (err & NFSERR_AUTHERR) { 1065 *tl++ = rpc_autherr; 1066 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1067 mreq->m_len -= NFSX_UNSIGNED; 1068 } else { 1069 *tl++ = rpc_mismatch; 1070 *tl++ = txdr_unsigned(RPC_VER2); 1071 *tl = txdr_unsigned(RPC_VER2); 1072 } 1073 } else { 1074 *tl++ = rpc_msgaccepted; 1075 1076 /* AUTH_UNIX requires RPCAUTH_NULL. */ 1077 *tl++ = 0; 1078 *tl++ = 0; 1079 1080 switch (err) { 1081 case EPROGUNAVAIL: 1082 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1083 break; 1084 case EPROGMISMATCH: 1085 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1086 tl = nfsm_build(&mb, 2 * NFSX_UNSIGNED); 1087 *tl++ = txdr_unsigned(NFS_VER2); 1088 *tl = txdr_unsigned(NFS_VER3); 1089 break; 1090 case EPROCUNAVAIL: 1091 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1092 break; 1093 case EBADRPC: 1094 *tl = txdr_unsigned(RPC_GARBAGE); 1095 break; 1096 default: 1097 *tl = 0; 1098 if (err != NFSERR_RETVOID) { 1099 tl = nfsm_build(&mb, NFSX_UNSIGNED); 1100 if (err) 1101 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1102 else 1103 *tl = 0; 1104 } 1105 break; 1106 }; 1107 } 1108 1109 *mrq = mreq; 1110 if (mbp != NULL) 1111 *mbp = mb; 1112 if (err != 0 && err != NFSERR_RETVOID) 1113 nfsstats.srvrpc_errs++; 1114 return (0); 1115 } 1116 1117 /* 1118 * Nfs timer routine 1119 * Scan the nfsreq list and retranmit any requests that have timed out 1120 * To avoid retransmission attempts on STREAM sockets (in the future) make 1121 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1122 */ 1123 void 1124 nfs_timer(arg) 1125 void *arg; 1126 { 1127 struct timeout *to = (struct timeout *)arg; 1128 struct nfsreq *rep; 1129 struct mbuf *m; 1130 struct socket *so; 1131 struct nfsmount *nmp; 1132 int timeo; 1133 int s, error; 1134 #ifdef NFSSERVER 1135 struct nfssvc_sock *slp; 1136 struct timeval tv; 1137 u_quad_t cur_usec; 1138 #endif 1139 1140 s = splsoftnet(); 1141 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1142 nmp = rep->r_nmp; 1143 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1144 continue; 1145 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1146 rep->r_flags |= R_SOFTTERM; 1147 continue; 1148 } 1149 if (rep->r_rtt >= 0) { 1150 rep->r_rtt++; 1151 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1152 timeo = nmp->nm_timeo; 1153 else 1154 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1155 if (nmp->nm_timeouts > 0) 1156 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1157 if (rep->r_rtt <= timeo) 1158 continue; 1159 if (nmp->nm_timeouts < 8) 1160 nmp->nm_timeouts++; 1161 } 1162 /* 1163 * Check for server not responding 1164 */ 1165 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1166 rep->r_rexmit > nmp->nm_deadthresh) { 1167 nfs_msg(rep->r_procp, 1168 nmp->nm_mountp->mnt_stat.f_mntfromname, 1169 "not responding"); 1170 rep->r_flags |= R_TPRINTFMSG; 1171 } 1172 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1173 nfsstats.rpctimeouts++; 1174 rep->r_flags |= R_SOFTTERM; 1175 continue; 1176 } 1177 if (nmp->nm_sotype != SOCK_DGRAM) { 1178 if (++rep->r_rexmit > NFS_MAXREXMIT) 1179 rep->r_rexmit = NFS_MAXREXMIT; 1180 continue; 1181 } 1182 if ((so = nmp->nm_so) == NULL) 1183 continue; 1184 1185 /* 1186 * If there is enough space and the window allows.. 1187 * Resend it 1188 * Set r_rtt to -1 in case we fail to send it now. 1189 */ 1190 rep->r_rtt = -1; 1191 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1192 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1193 (rep->r_flags & R_SENT) || 1194 nmp->nm_sent < nmp->nm_cwnd) && 1195 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1196 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1197 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1198 (struct mbuf *)0, (struct mbuf *)0, curproc); 1199 else 1200 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1201 nmp->nm_nam, (struct mbuf *)0, curproc); 1202 if (error) { 1203 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) 1204 so->so_error = 0; 1205 } else { 1206 /* 1207 * Iff first send, start timing 1208 * else turn timing off, backoff timer 1209 * and divide congestion window by 2. 1210 */ 1211 if (rep->r_flags & R_SENT) { 1212 rep->r_flags &= ~R_TIMING; 1213 if (++rep->r_rexmit > NFS_MAXREXMIT) 1214 rep->r_rexmit = NFS_MAXREXMIT; 1215 nmp->nm_cwnd >>= 1; 1216 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1217 nmp->nm_cwnd = NFS_CWNDSCALE; 1218 nfsstats.rpcretries++; 1219 } else { 1220 rep->r_flags |= R_SENT; 1221 nmp->nm_sent += NFS_CWNDSCALE; 1222 } 1223 rep->r_rtt = 0; 1224 } 1225 } 1226 } 1227 1228 #ifdef NFSSERVER 1229 /* 1230 * Scan the write gathering queues for writes that need to be 1231 * completed now. 1232 */ 1233 getmicrotime(&tv); 1234 cur_usec = (u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec; 1235 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1236 if (LIST_FIRST(&slp->ns_tq) && 1237 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) 1238 nfsrv_wakenfsd(slp); 1239 } 1240 #endif /* NFSSERVER */ 1241 splx(s); 1242 timeout_add(to, nfs_ticks); 1243 } 1244 1245 /* 1246 * Test for a termination condition pending on the process. 1247 * This is used for NFSMNT_INT mounts. 1248 */ 1249 int 1250 nfs_sigintr(nmp, rep, p) 1251 struct nfsmount *nmp; 1252 struct nfsreq *rep; 1253 struct proc *p; 1254 { 1255 1256 if (rep && (rep->r_flags & R_SOFTTERM)) 1257 return (EINTR); 1258 if (!(nmp->nm_flag & NFSMNT_INT)) 1259 return (0); 1260 if (p && p->p_siglist && 1261 (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) & 1262 NFSINT_SIGMASK)) 1263 return (EINTR); 1264 return (0); 1265 } 1266 1267 /* 1268 * Lock a socket against others. 1269 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1270 * and also to avoid race conditions between the processes with nfs requests 1271 * in progress when a reconnect is necessary. 1272 */ 1273 int 1274 nfs_sndlock(flagp, rep) 1275 int *flagp; 1276 struct nfsreq *rep; 1277 { 1278 struct proc *p; 1279 int slpflag = 0, slptimeo = 0; 1280 1281 if (rep) { 1282 p = rep->r_procp; 1283 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1284 slpflag = PCATCH; 1285 } else 1286 p = (struct proc *)0; 1287 while (*flagp & NFSMNT_SNDLOCK) { 1288 if (rep && nfs_sigintr(rep->r_nmp, rep, p)) 1289 return (EINTR); 1290 *flagp |= NFSMNT_WANTSND; 1291 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1292 slptimeo); 1293 if (slpflag == PCATCH) { 1294 slpflag = 0; 1295 slptimeo = 2 * hz; 1296 } 1297 } 1298 *flagp |= NFSMNT_SNDLOCK; 1299 return (0); 1300 } 1301 1302 /* 1303 * Unlock the stream socket for others. 1304 */ 1305 void 1306 nfs_sndunlock(flagp) 1307 int *flagp; 1308 { 1309 1310 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1311 panic("nfs sndunlock"); 1312 *flagp &= ~NFSMNT_SNDLOCK; 1313 if (*flagp & NFSMNT_WANTSND) { 1314 *flagp &= ~NFSMNT_WANTSND; 1315 wakeup((caddr_t)flagp); 1316 } 1317 } 1318 1319 int 1320 nfs_rcvlock(rep) 1321 struct nfsreq *rep; 1322 { 1323 int *flagp = &rep->r_nmp->nm_flag; 1324 int slpflag, slptimeo = 0; 1325 1326 if (*flagp & NFSMNT_INT) 1327 slpflag = PCATCH; 1328 else 1329 slpflag = 0; 1330 1331 while (*flagp & NFSMNT_RCVLOCK) { 1332 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1333 return (EINTR); 1334 *flagp |= NFSMNT_WANTRCV; 1335 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1336 slptimeo); 1337 if (rep->r_mrep != NULL) { 1338 /* 1339 * Don't take the lock if our reply has been received 1340 * while we where sleeping. 1341 */ 1342 return (EALREADY); 1343 } 1344 if (slpflag == PCATCH) { 1345 slpflag = 0; 1346 slptimeo = 2 * hz; 1347 } 1348 } 1349 *flagp |= NFSMNT_RCVLOCK; 1350 return (0); 1351 } 1352 1353 /* 1354 * Unlock the stream socket for others. 1355 */ 1356 void 1357 nfs_rcvunlock(flagp) 1358 int *flagp; 1359 { 1360 1361 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1362 panic("nfs rcvunlock"); 1363 *flagp &= ~NFSMNT_RCVLOCK; 1364 if (*flagp & NFSMNT_WANTRCV) { 1365 *flagp &= ~NFSMNT_WANTRCV; 1366 wakeup((caddr_t)flagp); 1367 } 1368 } 1369 1370 /* 1371 * Auxiliary routine to align the length of mbuf copies made with m_copyback(). 1372 */ 1373 void 1374 nfs_realign_fixup(struct mbuf *m, struct mbuf *n, unsigned int *off) 1375 { 1376 size_t padding; 1377 1378 /* 1379 * The maximum number of bytes that m_copyback() places in a mbuf is 1380 * always an aligned quantity, so realign happens at the chain's tail. 1381 */ 1382 while (n->m_next != NULL) 1383 n = n->m_next; 1384 1385 /* 1386 * Pad from the next elements in the source chain. Loop until the 1387 * destination chain is aligned, or the end of the source is reached. 1388 */ 1389 do { 1390 m = m->m_next; 1391 if (m == NULL) 1392 return; 1393 1394 padding = min(ALIGN(n->m_len) - n->m_len, m->m_len); 1395 if (padding > M_TRAILINGSPACE(n)) 1396 panic("nfs_realign_fixup: no memory to pad to"); 1397 1398 bcopy(mtod(m, void *), mtod(n, char *) + n->m_len, padding); 1399 1400 n->m_len += padding; 1401 m_adj(m, padding); 1402 *off += padding; 1403 1404 } while (!ALIGNED_POINTER(n->m_len, void *)); 1405 } 1406 1407 /* 1408 * The NFS RPC parsing code uses the data address and the length of mbuf 1409 * structures to calculate on-memory addresses. This function makes sure these 1410 * parameters are correctly aligned. 1411 */ 1412 void 1413 nfs_realign(struct mbuf **pm, int hsiz) 1414 { 1415 struct mbuf *m; 1416 struct mbuf *n = NULL; 1417 unsigned int off = 0; 1418 1419 ++nfs_realign_test; 1420 while ((m = *pm) != NULL) { 1421 if (!ALIGNED_POINTER(m->m_data, void *) || 1422 !ALIGNED_POINTER(m->m_len, void *)) { 1423 MGET(n, M_WAIT, MT_DATA); 1424 if (ALIGN(m->m_len) >= MINCLSIZE) { 1425 MCLGET(n, M_WAIT); 1426 } 1427 n->m_len = 0; 1428 break; 1429 } 1430 pm = &m->m_next; 1431 } 1432 /* 1433 * If n is non-NULL, loop on m copying data, then replace the 1434 * portion of the chain that had to be realigned. 1435 */ 1436 if (n != NULL) { 1437 ++nfs_realign_count; 1438 while (m) { 1439 m_copyback(n, off, m->m_len, mtod(m, caddr_t)); 1440 1441 /* 1442 * If an unaligned amount of memory was copied, fix up 1443 * the last mbuf created by m_copyback(). 1444 */ 1445 if (!ALIGNED_POINTER(m->m_len, void *)) 1446 nfs_realign_fixup(m, n, &off); 1447 1448 off += m->m_len; 1449 m = m->m_next; 1450 } 1451 m_freem(*pm); 1452 *pm = n; 1453 } 1454 } 1455 1456 1457 /* 1458 * Parse an RPC request 1459 * - verify it 1460 * - fill in the cred struct. 1461 */ 1462 int 1463 nfs_getreq(nd, nfsd, has_header) 1464 struct nfsrv_descript *nd; 1465 struct nfsd *nfsd; 1466 int has_header; 1467 { 1468 int len, i; 1469 u_int32_t *tl; 1470 int32_t t1; 1471 caddr_t dpos, cp2; 1472 u_int32_t nfsvers, auth_type; 1473 int error = 0; 1474 struct mbuf *mrep, *md; 1475 1476 mrep = nd->nd_mrep; 1477 md = nd->nd_md; 1478 dpos = nd->nd_dpos; 1479 if (has_header) { 1480 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1481 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1482 if (*tl++ != rpc_call) { 1483 m_freem(mrep); 1484 return (EBADRPC); 1485 } 1486 } else 1487 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1488 nd->nd_repstat = 0; 1489 nd->nd_flag = 0; 1490 if (*tl++ != rpc_vers) { 1491 nd->nd_repstat = ERPCMISMATCH; 1492 nd->nd_procnum = NFSPROC_NOOP; 1493 return (0); 1494 } 1495 if (*tl != nfs_prog) { 1496 nd->nd_repstat = EPROGUNAVAIL; 1497 nd->nd_procnum = NFSPROC_NOOP; 1498 return (0); 1499 } 1500 tl++; 1501 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1502 if (nfsvers != NFS_VER2 && nfsvers != NFS_VER3) { 1503 nd->nd_repstat = EPROGMISMATCH; 1504 nd->nd_procnum = NFSPROC_NOOP; 1505 return (0); 1506 } 1507 if (nfsvers == NFS_VER3) 1508 nd->nd_flag = ND_NFSV3; 1509 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1510 if (nd->nd_procnum == NFSPROC_NULL) 1511 return (0); 1512 if (nd->nd_procnum >= NFS_NPROCS || 1513 (nd->nd_procnum > NFSPROC_COMMIT) || 1514 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1515 nd->nd_repstat = EPROCUNAVAIL; 1516 nd->nd_procnum = NFSPROC_NOOP; 1517 return (0); 1518 } 1519 if ((nd->nd_flag & ND_NFSV3) == 0) 1520 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1521 auth_type = *tl++; 1522 len = fxdr_unsigned(int, *tl++); 1523 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1524 m_freem(mrep); 1525 return (EBADRPC); 1526 } 1527 1528 /* Handle auth_unix */ 1529 if (auth_type == rpc_auth_unix) { 1530 len = fxdr_unsigned(int, *++tl); 1531 if (len < 0 || len > NFS_MAXNAMLEN) { 1532 m_freem(mrep); 1533 return (EBADRPC); 1534 } 1535 nfsm_adv(nfsm_rndup(len)); 1536 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1537 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred)); 1538 nd->nd_cr.cr_ref = 1; 1539 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1540 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1541 len = fxdr_unsigned(int, *tl); 1542 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1543 m_freem(mrep); 1544 return (EBADRPC); 1545 } 1546 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1547 for (i = 0; i < len; i++) 1548 if (i < NGROUPS) 1549 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1550 else 1551 tl++; 1552 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 1553 if (nd->nd_cr.cr_ngroups > 1) 1554 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1555 len = fxdr_unsigned(int, *++tl); 1556 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1557 m_freem(mrep); 1558 return (EBADRPC); 1559 } 1560 if (len > 0) 1561 nfsm_adv(nfsm_rndup(len)); 1562 } else { 1563 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1564 nd->nd_procnum = NFSPROC_NOOP; 1565 return (0); 1566 } 1567 1568 nd->nd_md = md; 1569 nd->nd_dpos = dpos; 1570 return (0); 1571 nfsmout: 1572 return (error); 1573 } 1574 1575 int 1576 nfs_msg(p, server, msg) 1577 struct proc *p; 1578 char *server, *msg; 1579 { 1580 tpr_t tpr; 1581 1582 if (p) 1583 tpr = tprintf_open(p); 1584 else 1585 tpr = NULL; 1586 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1587 tprintf_close(tpr); 1588 return (0); 1589 } 1590 1591 #ifdef NFSSERVER 1592 /* 1593 * Socket upcall routine for the nfsd sockets. 1594 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1595 * Essentially do as much as possible non-blocking, else punt and it will 1596 * be called with M_WAIT from an nfsd. 1597 */ 1598 void 1599 nfsrv_rcv(so, arg, waitflag) 1600 struct socket *so; 1601 caddr_t arg; 1602 int waitflag; 1603 { 1604 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1605 struct mbuf *m; 1606 struct mbuf *mp, *nam; 1607 struct uio auio; 1608 int flags, error; 1609 1610 if ((slp->ns_flag & SLP_VALID) == 0) 1611 return; 1612 #ifdef notdef 1613 /* 1614 * Define this to test for nfsds handling this under heavy load. 1615 */ 1616 if (waitflag == M_DONTWAIT) { 1617 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1618 } 1619 #endif 1620 auio.uio_procp = NULL; 1621 if (so->so_type == SOCK_STREAM) { 1622 /* 1623 * If there are already records on the queue, defer soreceive() 1624 * to an nfsd so that there is feedback to the TCP layer that 1625 * the nfs servers are heavily loaded. 1626 */ 1627 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1628 slp->ns_flag |= SLP_NEEDQ; 1629 goto dorecs; 1630 } 1631 1632 /* 1633 * Do soreceive(). 1634 */ 1635 auio.uio_resid = 1000000000; 1636 flags = MSG_DONTWAIT; 1637 error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1638 if (error || mp == (struct mbuf *)0) { 1639 if (error == EWOULDBLOCK) 1640 slp->ns_flag |= SLP_NEEDQ; 1641 else 1642 slp->ns_flag |= SLP_DISCONN; 1643 goto dorecs; 1644 } 1645 m = mp; 1646 if (slp->ns_rawend) { 1647 slp->ns_rawend->m_next = m; 1648 slp->ns_cc += 1000000000 - auio.uio_resid; 1649 } else { 1650 slp->ns_raw = m; 1651 slp->ns_cc = 1000000000 - auio.uio_resid; 1652 } 1653 while (m->m_next) 1654 m = m->m_next; 1655 slp->ns_rawend = m; 1656 1657 /* 1658 * Now try and parse record(s) out of the raw stream data. 1659 */ 1660 error = nfsrv_getstream(slp, waitflag); 1661 if (error) { 1662 if (error == EPERM) 1663 slp->ns_flag |= SLP_DISCONN; 1664 else 1665 slp->ns_flag |= SLP_NEEDQ; 1666 } 1667 } else { 1668 do { 1669 auio.uio_resid = 1000000000; 1670 flags = MSG_DONTWAIT; 1671 error = soreceive(so, &nam, &auio, &mp, 1672 (struct mbuf **)0, &flags); 1673 if (mp) { 1674 if (nam) { 1675 m = nam; 1676 m->m_next = mp; 1677 } else 1678 m = mp; 1679 if (slp->ns_recend) 1680 slp->ns_recend->m_nextpkt = m; 1681 else 1682 slp->ns_rec = m; 1683 slp->ns_recend = m; 1684 m->m_nextpkt = (struct mbuf *)0; 1685 } 1686 if (error) { 1687 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1688 && error != EWOULDBLOCK) { 1689 slp->ns_flag |= SLP_DISCONN; 1690 goto dorecs; 1691 } 1692 } 1693 } while (mp); 1694 } 1695 1696 /* 1697 * Now try and process the request records, non-blocking. 1698 */ 1699 dorecs: 1700 if (waitflag == M_DONTWAIT && 1701 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1702 nfsrv_wakenfsd(slp); 1703 } 1704 1705 /* 1706 * Try and extract an RPC request from the mbuf data list received on a 1707 * stream socket. The "waitflag" argument indicates whether or not it 1708 * can sleep. 1709 */ 1710 int 1711 nfsrv_getstream(slp, waitflag) 1712 struct nfssvc_sock *slp; 1713 int waitflag; 1714 { 1715 struct mbuf *m, **mpp; 1716 char *cp1, *cp2; 1717 int len; 1718 struct mbuf *om, *m2, *recm; 1719 u_int32_t recmark; 1720 1721 if (slp->ns_flag & SLP_GETSTREAM) 1722 panic("nfs getstream"); 1723 slp->ns_flag |= SLP_GETSTREAM; 1724 for (;;) { 1725 if (slp->ns_reclen == 0) { 1726 if (slp->ns_cc < NFSX_UNSIGNED) { 1727 slp->ns_flag &= ~SLP_GETSTREAM; 1728 return (0); 1729 } 1730 m = slp->ns_raw; 1731 if (m->m_len >= NFSX_UNSIGNED) { 1732 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 1733 m->m_data += NFSX_UNSIGNED; 1734 m->m_len -= NFSX_UNSIGNED; 1735 } else { 1736 cp1 = (caddr_t)&recmark; 1737 cp2 = mtod(m, caddr_t); 1738 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1739 while (m->m_len == 0) { 1740 m = m->m_next; 1741 cp2 = mtod(m, caddr_t); 1742 } 1743 *cp1++ = *cp2++; 1744 m->m_data++; 1745 m->m_len--; 1746 } 1747 } 1748 slp->ns_cc -= NFSX_UNSIGNED; 1749 recmark = ntohl(recmark); 1750 slp->ns_reclen = recmark & ~0x80000000; 1751 if (recmark & 0x80000000) 1752 slp->ns_flag |= SLP_LASTFRAG; 1753 else 1754 slp->ns_flag &= ~SLP_LASTFRAG; 1755 if (slp->ns_reclen > NFS_MAXPACKET) { 1756 slp->ns_flag &= ~SLP_GETSTREAM; 1757 return (EPERM); 1758 } 1759 } 1760 1761 /* 1762 * Now get the record part. 1763 */ 1764 recm = NULL; 1765 if (slp->ns_cc == slp->ns_reclen) { 1766 recm = slp->ns_raw; 1767 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 1768 slp->ns_cc = slp->ns_reclen = 0; 1769 } else if (slp->ns_cc > slp->ns_reclen) { 1770 len = 0; 1771 m = slp->ns_raw; 1772 om = (struct mbuf *)0; 1773 while (len < slp->ns_reclen) { 1774 if ((len + m->m_len) > slp->ns_reclen) { 1775 m2 = m_copym(m, 0, slp->ns_reclen - len, 1776 waitflag); 1777 if (m2) { 1778 if (om) { 1779 om->m_next = m2; 1780 recm = slp->ns_raw; 1781 } else 1782 recm = m2; 1783 m->m_data += slp->ns_reclen - len; 1784 m->m_len -= slp->ns_reclen - len; 1785 len = slp->ns_reclen; 1786 } else { 1787 slp->ns_flag &= ~SLP_GETSTREAM; 1788 return (EWOULDBLOCK); 1789 } 1790 } else if ((len + m->m_len) == slp->ns_reclen) { 1791 om = m; 1792 len += m->m_len; 1793 m = m->m_next; 1794 recm = slp->ns_raw; 1795 om->m_next = (struct mbuf *)0; 1796 } else { 1797 om = m; 1798 len += m->m_len; 1799 m = m->m_next; 1800 } 1801 } 1802 slp->ns_raw = m; 1803 slp->ns_cc -= len; 1804 slp->ns_reclen = 0; 1805 } else { 1806 slp->ns_flag &= ~SLP_GETSTREAM; 1807 return (0); 1808 } 1809 1810 /* 1811 * Accumulate the fragments into a record. 1812 */ 1813 mpp = &slp->ns_frag; 1814 while (*mpp) 1815 mpp = &((*mpp)->m_next); 1816 *mpp = recm; 1817 if (slp->ns_flag & SLP_LASTFRAG) { 1818 if (slp->ns_recend) 1819 slp->ns_recend->m_nextpkt = slp->ns_frag; 1820 else 1821 slp->ns_rec = slp->ns_frag; 1822 slp->ns_recend = slp->ns_frag; 1823 slp->ns_frag = (struct mbuf *)0; 1824 } 1825 } 1826 } 1827 1828 /* 1829 * Parse an RPC header. 1830 */ 1831 int 1832 nfsrv_dorec(slp, nfsd, ndp) 1833 struct nfssvc_sock *slp; 1834 struct nfsd *nfsd; 1835 struct nfsrv_descript **ndp; 1836 { 1837 struct mbuf *m, *nam; 1838 struct nfsrv_descript *nd; 1839 int error; 1840 1841 *ndp = NULL; 1842 if ((slp->ns_flag & SLP_VALID) == 0 || 1843 (m = slp->ns_rec) == (struct mbuf *)0) 1844 return (ENOBUFS); 1845 slp->ns_rec = m->m_nextpkt; 1846 if (slp->ns_rec) 1847 m->m_nextpkt = (struct mbuf *)0; 1848 else 1849 slp->ns_recend = (struct mbuf *)0; 1850 if (m->m_type == MT_SONAME) { 1851 nam = m; 1852 m = m->m_next; 1853 nam->m_next = NULL; 1854 } else 1855 nam = NULL; 1856 nd = malloc(sizeof(struct nfsrv_descript), M_NFSRVDESC, M_WAITOK); 1857 nfs_realign(&m, 10 * NFSX_UNSIGNED); 1858 nd->nd_md = nd->nd_mrep = m; 1859 nd->nd_nam2 = nam; 1860 nd->nd_dpos = mtod(m, caddr_t); 1861 error = nfs_getreq(nd, nfsd, TRUE); 1862 if (error) { 1863 m_freem(nam); 1864 free(nd, M_NFSRVDESC); 1865 return (error); 1866 } 1867 *ndp = nd; 1868 nfsd->nfsd_nd = nd; 1869 return (0); 1870 } 1871 1872 1873 /* 1874 * Search for a sleeping nfsd and wake it up. 1875 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 1876 * running nfsds will go look for the work in the nfssvc_sock list. 1877 */ 1878 void 1879 nfsrv_wakenfsd(struct nfssvc_sock *slp) 1880 { 1881 struct nfsd *nfsd; 1882 1883 if ((slp->ns_flag & SLP_VALID) == 0) 1884 return; 1885 1886 TAILQ_FOREACH(nfsd, &nfsd_head, nfsd_chain) { 1887 if (nfsd->nfsd_flag & NFSD_WAITING) { 1888 nfsd->nfsd_flag &= ~NFSD_WAITING; 1889 if (nfsd->nfsd_slp) 1890 panic("nfsd wakeup"); 1891 slp->ns_sref++; 1892 nfsd->nfsd_slp = slp; 1893 wakeup_one(nfsd); 1894 return; 1895 } 1896 } 1897 1898 slp->ns_flag |= SLP_DOREC; 1899 nfsd_head_flag |= NFSD_CHECKSLP; 1900 } 1901 #endif /* NFSSERVER */ 1902