1 /* $OpenBSD: nfs_socket.c,v 1.144 2023/08/03 09:49:09 mvs 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/protosw.h> 50 #include <sys/signalvar.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/nfs_var.h> 69 70 /* External data, mostly RPC constants in XDR form. */ 71 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 72 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr; 73 extern u_int32_t nfs_prog; 74 extern struct nfsstats nfsstats; 75 extern int nfsv3_procid[NFS_NPROCS]; 76 extern int nfs_ticks; 77 78 extern struct pool nfsrv_descript_pl; 79 80 /* 81 * There is a congestion window for outstanding rpcs maintained per mount 82 * point. The cwnd size is adjusted in roughly the way that: 83 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 84 * SIGCOMM '88". ACM, August 1988. 85 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 86 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 87 * of rpcs is in progress. 88 * (The sent count and cwnd are scaled for integer arith.) 89 * Variants of "slow start" were tried and were found to be too much of a 90 * performance hit (ave. rtt 3 times larger), 91 * I suspect due to the large rtt that nfs rpcs have. 92 */ 93 #define NFS_CWNDSCALE 256 94 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 95 int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256 }; 96 97 /* RTT estimator */ 98 enum nfs_rto_timers nfs_ptimers[NFS_NPROCS] = { 99 NFS_DEFAULT_TIMER, /* NULL */ 100 NFS_GETATTR_TIMER, /* GETATTR */ 101 NFS_DEFAULT_TIMER, /* SETATTR */ 102 NFS_LOOKUP_TIMER, /* LOOKUP */ 103 NFS_GETATTR_TIMER, /* ACCESS */ 104 NFS_READ_TIMER, /* READLINK */ 105 NFS_READ_TIMER, /* READ */ 106 NFS_WRITE_TIMER, /* WRITE */ 107 NFS_DEFAULT_TIMER, /* CREATE */ 108 NFS_DEFAULT_TIMER, /* MKDIR */ 109 NFS_DEFAULT_TIMER, /* SYMLINK */ 110 NFS_DEFAULT_TIMER, /* MKNOD */ 111 NFS_DEFAULT_TIMER, /* REMOVE */ 112 NFS_DEFAULT_TIMER, /* RMDIR */ 113 NFS_DEFAULT_TIMER, /* RENAME */ 114 NFS_DEFAULT_TIMER, /* LINK */ 115 NFS_READ_TIMER, /* READDIR */ 116 NFS_READ_TIMER, /* READDIRPLUS */ 117 NFS_DEFAULT_TIMER, /* FSSTAT */ 118 NFS_DEFAULT_TIMER, /* FSINFO */ 119 NFS_DEFAULT_TIMER, /* PATHCONF */ 120 NFS_DEFAULT_TIMER, /* COMMIT */ 121 NFS_DEFAULT_TIMER, /* NOOP */ 122 }; 123 124 void nfs_init_rtt(struct nfsmount *); 125 void nfs_update_rtt(struct nfsreq *); 126 int nfs_estimate_rto(struct nfsmount *, u_int32_t procnum); 127 128 void nfs_realign(struct mbuf **, int); 129 void nfs_realign_fixup(struct mbuf *, struct mbuf *, unsigned int *); 130 131 int nfs_rcvlock(struct nfsreq *); 132 int nfs_receive(struct nfsreq *, struct mbuf **, struct mbuf **); 133 int nfs_reconnect(struct nfsreq *); 134 int nfs_reply(struct nfsreq *); 135 void nfs_msg(struct nfsreq *, char *); 136 void nfs_rcvunlock(int *); 137 138 int nfsrv_getstream(struct nfssvc_sock *, int); 139 140 unsigned int nfs_realign_test = 0; 141 unsigned int nfs_realign_count = 0; 142 143 /* Initialize the RTT estimator state for a new mount point. */ 144 void 145 nfs_init_rtt(struct nfsmount *nmp) 146 { 147 int i; 148 149 for (i = 0; i < NFS_MAX_TIMER; i++) 150 nmp->nm_srtt[i] = NFS_INITRTT; 151 for (i = 0; i < NFS_MAX_TIMER; i++) 152 nmp->nm_sdrtt[i] = 0; 153 } 154 155 /* 156 * Update a mount point's RTT estimator state using data from the 157 * passed-in request. 158 * 159 * Use a gain of 0.125 on the mean and a gain of 0.25 on the deviation. 160 * 161 * NB: Since the timer resolution of NFS_HZ is so course, it can often 162 * result in r_rtt == 0. Since r_rtt == N means that the actual RTT is 163 * between N + dt and N + 2 - dt ticks, add 1 before calculating the 164 * update values. 165 */ 166 void 167 nfs_update_rtt(struct nfsreq *rep) 168 { 169 int t1 = rep->r_rtt + 1; 170 int index = nfs_ptimers[rep->r_procnum] - 1; 171 int *srtt = &rep->r_nmp->nm_srtt[index]; 172 int *sdrtt = &rep->r_nmp->nm_sdrtt[index]; 173 174 t1 -= *srtt >> 3; 175 *srtt += t1; 176 if (t1 < 0) 177 t1 = -t1; 178 t1 -= *sdrtt >> 2; 179 *sdrtt += t1; 180 } 181 182 /* 183 * Estimate RTO for an NFS RPC sent via an unreliable datagram. 184 * 185 * Use the mean and mean deviation of RTT for the appropriate type 186 * of RPC for the frequent RPCs and a default for the others. 187 * The justification for doing "other" this way is that these RPCs 188 * happen so infrequently that timer est. would probably be stale. 189 * Also, since many of these RPCs are non-idempotent, a conservative 190 * timeout is desired. 191 * 192 * getattr, lookup - A+2D 193 * read, write - A+4D 194 * other - nm_timeo 195 */ 196 int 197 nfs_estimate_rto(struct nfsmount *nmp, u_int32_t procnum) 198 { 199 enum nfs_rto_timers timer = nfs_ptimers[procnum]; 200 int index = timer - 1; 201 int rto; 202 203 switch (timer) { 204 case NFS_GETATTR_TIMER: 205 case NFS_LOOKUP_TIMER: 206 rto = ((nmp->nm_srtt[index] + 3) >> 2) + 207 ((nmp->nm_sdrtt[index] + 1) >> 1); 208 break; 209 case NFS_READ_TIMER: 210 case NFS_WRITE_TIMER: 211 rto = ((nmp->nm_srtt[index] + 7) >> 3) + 212 (nmp->nm_sdrtt[index] + 1); 213 break; 214 default: 215 rto = nmp->nm_timeo; 216 return (rto); 217 } 218 219 if (rto < NFS_MINRTO) 220 rto = NFS_MINRTO; 221 else if (rto > NFS_MAXRTO) 222 rto = NFS_MAXRTO; 223 224 return (rto); 225 } 226 227 228 229 /* 230 * Initialize sockets and congestion for a new NFS connection. 231 * We do not free the sockaddr if error. 232 */ 233 int 234 nfs_connect(struct nfsmount *nmp, struct nfsreq *rep) 235 { 236 struct socket *so; 237 int error, rcvreserve, sndreserve; 238 struct sockaddr *saddr; 239 struct sockaddr_in *sin; 240 struct mbuf *nam = NULL, *mopt = NULL; 241 242 if (!(nmp->nm_sotype == SOCK_DGRAM || nmp->nm_sotype == SOCK_STREAM)) 243 return (EINVAL); 244 245 nmp->nm_so = NULL; 246 saddr = mtod(nmp->nm_nam, struct sockaddr *); 247 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 248 nmp->nm_soproto); 249 if (error) { 250 nfs_disconnect(nmp); 251 return (error); 252 } 253 254 /* Allocate mbufs possibly waiting before grabbing the socket lock. */ 255 if (nmp->nm_sotype == SOCK_STREAM || saddr->sa_family == AF_INET) 256 MGET(mopt, M_WAIT, MT_SOOPTS); 257 if (saddr->sa_family == AF_INET) 258 MGET(nam, M_WAIT, MT_SONAME); 259 260 so = nmp->nm_so; 261 nmp->nm_soflags = so->so_proto->pr_flags; 262 263 /* 264 * Some servers require that the client port be a reserved port number. 265 * We always allocate a reserved port, as this prevents filehandle 266 * disclosure through UDP port capture. 267 */ 268 if (saddr->sa_family == AF_INET) { 269 int *ip; 270 271 mopt->m_len = sizeof(int); 272 ip = mtod(mopt, int *); 273 *ip = IP_PORTRANGE_LOW; 274 error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt); 275 if (error) 276 goto bad; 277 278 sin = mtod(nam, struct sockaddr_in *); 279 memset(sin, 0, sizeof(*sin)); 280 sin->sin_len = nam->m_len = sizeof(struct sockaddr_in); 281 sin->sin_family = AF_INET; 282 sin->sin_addr.s_addr = INADDR_ANY; 283 sin->sin_port = htons(0); 284 solock(so); 285 error = sobind(so, nam, &proc0); 286 sounlock(so); 287 if (error) 288 goto bad; 289 290 mopt->m_len = sizeof(int); 291 ip = mtod(mopt, int *); 292 *ip = IP_PORTRANGE_DEFAULT; 293 error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt); 294 if (error) 295 goto bad; 296 } 297 298 solock(so); 299 /* 300 * Protocols that do not require connections may be optionally left 301 * unconnected for servers that reply from a port other than NFS_PORT. 302 */ 303 if (nmp->nm_flag & NFSMNT_NOCONN) { 304 if (nmp->nm_soflags & PR_CONNREQUIRED) { 305 error = ENOTCONN; 306 goto bad_locked; 307 } 308 } else { 309 error = soconnect(so, nmp->nm_nam); 310 if (error) 311 goto bad_locked; 312 313 /* 314 * Wait for the connection to complete. Cribbed from the 315 * connect system call but with the wait timing out so 316 * that interruptible mounts don't hang here for a long time. 317 */ 318 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 319 sosleep_nsec(so, &so->so_timeo, PSOCK, "nfscon", 320 SEC_TO_NSEC(2)); 321 if ((so->so_state & SS_ISCONNECTING) && 322 so->so_error == 0 && rep && 323 (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ 324 so->so_state &= ~SS_ISCONNECTING; 325 goto bad_locked; 326 } 327 } 328 if (so->so_error) { 329 error = so->so_error; 330 so->so_error = 0; 331 goto bad_locked; 332 } 333 } 334 /* 335 * Always set receive timeout to detect server crash and reconnect. 336 * Otherwise, we can get stuck in soreceive forever. 337 */ 338 so->so_rcv.sb_timeo_nsecs = SEC_TO_NSEC(5); 339 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) 340 so->so_snd.sb_timeo_nsecs = SEC_TO_NSEC(5); 341 else 342 so->so_snd.sb_timeo_nsecs = INFSLP; 343 sounlock(so); 344 if (nmp->nm_sotype == SOCK_DGRAM) { 345 sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR; 346 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 347 NFS_MAXPKTHDR) * 2; 348 } else if (nmp->nm_sotype == SOCK_STREAM) { 349 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 350 *mtod(mopt, int32_t *) = 1; 351 mopt->m_len = sizeof(int32_t); 352 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, mopt); 353 } 354 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 355 *mtod(mopt, int32_t *) = 1; 356 mopt->m_len = sizeof(int32_t); 357 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, mopt); 358 } 359 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 360 sizeof (u_int32_t)) * 2; 361 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 362 sizeof (u_int32_t)) * 2; 363 } else { 364 panic("%s: nm_sotype %d", __func__, nmp->nm_sotype); 365 } 366 solock(so); 367 error = soreserve(so, sndreserve, rcvreserve); 368 if (error) 369 goto bad_locked; 370 so->so_rcv.sb_flags |= SB_NOINTR; 371 so->so_snd.sb_flags |= SB_NOINTR; 372 sounlock(so); 373 374 m_freem(mopt); 375 m_freem(nam); 376 377 /* Initialize other non-zero congestion variables */ 378 nfs_init_rtt(nmp); 379 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 380 nmp->nm_sent = 0; 381 nmp->nm_timeouts = 0; 382 return (0); 383 384 bad_locked: 385 sounlock(so); 386 bad: 387 388 m_freem(mopt); 389 m_freem(nam); 390 391 nfs_disconnect(nmp); 392 return (error); 393 } 394 395 /* 396 * Reconnect routine: 397 * Called when a connection is broken on a reliable protocol. 398 * - clean up the old socket 399 * - nfs_connect() again 400 * - set R_MUSTRESEND for all outstanding requests on mount point 401 * If this fails the mount point is DEAD! 402 * nb: Must be called with the nfs_sndlock() set on the mount point. 403 */ 404 int 405 nfs_reconnect(struct nfsreq *rep) 406 { 407 struct nfsreq *rp; 408 struct nfsmount *nmp = rep->r_nmp; 409 int error; 410 411 nfs_disconnect(nmp); 412 while ((error = nfs_connect(nmp, rep)) != 0) { 413 if (error == EINTR || error == ERESTART) 414 return (EINTR); 415 tsleep_nsec(&nowake, PSOCK, "nfsrecon", SEC_TO_NSEC(1)); 416 } 417 418 /* 419 * Loop through outstanding request list and fix up all requests 420 * on old socket. 421 */ 422 TAILQ_FOREACH(rp, &nmp->nm_reqsq, r_chain) { 423 rp->r_flags |= R_MUSTRESEND; 424 rp->r_rexmit = 0; 425 } 426 return (0); 427 } 428 429 /* 430 * NFS disconnect. Clean up and unlink. 431 */ 432 void 433 nfs_disconnect(struct nfsmount *nmp) 434 { 435 struct socket *so; 436 437 if (nmp->nm_so) { 438 so = nmp->nm_so; 439 nmp->nm_so = NULL; 440 soshutdown(so, SHUT_RDWR); 441 soclose(so, 0); 442 } 443 } 444 445 /* 446 * This is the nfs send routine. For connection based socket types, it 447 * must be called with an nfs_sndlock() on the socket. 448 * "rep == NULL" indicates that it has been called from a server. 449 * For the client side: 450 * - return EINTR if the RPC is terminated, 0 otherwise 451 * - set R_MUSTRESEND if the send fails for any reason 452 * - do any cleanup required by recoverable socket errors (???) 453 * For the server side: 454 * - return EINTR or ERESTART if interrupted by a signal 455 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 456 * - do any cleanup required by recoverable socket errors (???) 457 */ 458 int 459 nfs_send(struct socket *so, struct mbuf *nam, struct mbuf *top, 460 struct nfsreq *rep) 461 { 462 struct mbuf *sendnam; 463 int error, soflags, flags; 464 465 if (rep) { 466 if (rep->r_flags & R_SOFTTERM) { 467 m_freem(top); 468 return (EINTR); 469 } 470 if ((so = rep->r_nmp->nm_so) == NULL) { 471 rep->r_flags |= R_MUSTRESEND; 472 m_freem(top); 473 return (0); 474 } 475 rep->r_flags &= ~R_MUSTRESEND; 476 soflags = rep->r_nmp->nm_soflags; 477 } else 478 soflags = so->so_proto->pr_flags; 479 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 480 sendnam = NULL; 481 else 482 sendnam = nam; 483 flags = 0; 484 485 error = sosend(so, sendnam, NULL, top, NULL, flags); 486 if (error) { 487 if (rep) { 488 /* 489 * Deal with errors for the client side. 490 */ 491 if (rep->r_flags & R_SOFTTERM) 492 error = EINTR; 493 else 494 rep->r_flags |= R_MUSTRESEND; 495 } 496 497 /* 498 * Handle any recoverable (soft) socket errors here. (???) 499 */ 500 if (error != EINTR && error != ERESTART && 501 error != EWOULDBLOCK && error != EPIPE) 502 error = 0; 503 } 504 return (error); 505 } 506 507 #ifdef NFSCLIENT 508 /* 509 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 510 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 511 * Mark and consolidate the data into a new mbuf list. 512 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 513 * small mbufs. 514 * For SOCK_STREAM we must be very careful to read an entire record once 515 * we have read any of it, even if the system call has been interrupted. 516 */ 517 int 518 nfs_receive(struct nfsreq *rep, struct mbuf **aname, struct mbuf **mp) 519 { 520 struct socket *so; 521 struct uio auio; 522 struct iovec aio; 523 struct mbuf *m; 524 struct mbuf *control; 525 u_int32_t len; 526 struct mbuf **getnam; 527 int error, sotype, rcvflg; 528 struct proc *p = curproc; /* XXX */ 529 530 /* 531 * Set up arguments for soreceive() 532 */ 533 *mp = NULL; 534 *aname = NULL; 535 sotype = rep->r_nmp->nm_sotype; 536 537 /* 538 * For reliable protocols, lock against other senders/receivers 539 * in case a reconnect is necessary. 540 * For SOCK_STREAM, first get the Record Mark to find out how much 541 * more there is to get. 542 * We must lock the socket against other receivers 543 * until we have an entire rpc request/reply. 544 */ 545 if (sotype != SOCK_DGRAM) { 546 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 547 if (error) 548 return (error); 549 tryagain: 550 /* 551 * Check for fatal errors and resending request. 552 */ 553 /* 554 * Ugh: If a reconnect attempt just happened, nm_so 555 * would have changed. NULL indicates a failed 556 * attempt that has essentially shut down this 557 * mount point. 558 */ 559 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 560 nfs_sndunlock(&rep->r_nmp->nm_flag); 561 return (EINTR); 562 } 563 so = rep->r_nmp->nm_so; 564 if (!so) { 565 error = nfs_reconnect(rep); 566 if (error) { 567 nfs_sndunlock(&rep->r_nmp->nm_flag); 568 return (error); 569 } 570 goto tryagain; 571 } 572 while (rep->r_flags & R_MUSTRESEND) { 573 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 574 nfsstats.rpcretries++; 575 rep->r_rtt = 0; 576 rep->r_flags &= ~R_TIMING; 577 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 578 if (error) { 579 if (error == EINTR || error == ERESTART || 580 (error = nfs_reconnect(rep)) != 0) { 581 nfs_sndunlock(&rep->r_nmp->nm_flag); 582 return (error); 583 } 584 goto tryagain; 585 } 586 } 587 nfs_sndunlock(&rep->r_nmp->nm_flag); 588 if (sotype == SOCK_STREAM) { 589 aio.iov_base = (caddr_t) &len; 590 aio.iov_len = sizeof(u_int32_t); 591 auio.uio_iov = &aio; 592 auio.uio_iovcnt = 1; 593 auio.uio_segflg = UIO_SYSSPACE; 594 auio.uio_rw = UIO_READ; 595 auio.uio_offset = 0; 596 auio.uio_resid = sizeof(u_int32_t); 597 auio.uio_procp = p; 598 do { 599 rcvflg = MSG_WAITALL; 600 error = soreceive(so, NULL, &auio, NULL, NULL, 601 &rcvflg, 0); 602 if (error == EWOULDBLOCK && rep) { 603 if (rep->r_flags & R_SOFTTERM) 604 return (EINTR); 605 /* 606 * looks like the server died after it 607 * received the request, make sure 608 * that we will retransmit and we 609 * don't get stuck here forever. 610 */ 611 if (rep->r_rexmit >= 612 rep->r_nmp->nm_retry) { 613 nfsstats.rpctimeouts++; 614 error = EPIPE; 615 } 616 } 617 } while (error == EWOULDBLOCK); 618 if (!error && auio.uio_resid > 0) { 619 log(LOG_INFO, 620 "short receive (%zu/%zu) from nfs server %s\n", 621 sizeof(u_int32_t) - auio.uio_resid, 622 sizeof(u_int32_t), 623 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 624 error = EPIPE; 625 } 626 if (error) 627 goto errout; 628 629 len = ntohl(len) & ~0x80000000; 630 /* 631 * This is SERIOUS! We are out of sync with the sender 632 * and forcing a disconnect/reconnect is all I can do. 633 */ 634 if (len > NFS_MAXPACKET) { 635 log(LOG_ERR, "%s (%u) from nfs server %s\n", 636 "impossible packet length", 637 len, 638 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 639 error = EFBIG; 640 goto errout; 641 } 642 auio.uio_resid = len; 643 do { 644 rcvflg = MSG_WAITALL; 645 error = soreceive(so, NULL, &auio, mp, NULL, 646 &rcvflg, 0); 647 } while (error == EWOULDBLOCK || error == EINTR || 648 error == ERESTART); 649 if (!error && auio.uio_resid > 0) { 650 log(LOG_INFO, "short receive (%zu/%u) from " 651 "nfs server %s\n", len - auio.uio_resid, 652 len, rep->r_nmp->nm_mountp-> 653 mnt_stat.f_mntfromname); 654 error = EPIPE; 655 } 656 } else { 657 /* 658 * NB: Since uio_resid is big, MSG_WAITALL is ignored 659 * and soreceive() will return when it has either a 660 * control msg or a data msg. 661 * We have no use for control msg., but must grab them 662 * and then throw them away so we know what is going 663 * on. 664 */ 665 auio.uio_resid = len = 100000000; /* Anything Big */ 666 auio.uio_procp = p; 667 do { 668 rcvflg = 0; 669 error = soreceive(so, NULL, &auio, mp, &control, 670 &rcvflg, 0); 671 m_freem(control); 672 if (error == EWOULDBLOCK && rep) { 673 if (rep->r_flags & R_SOFTTERM) 674 return (EINTR); 675 } 676 } while (error == EWOULDBLOCK || 677 (!error && *mp == NULL && control)); 678 if ((rcvflg & MSG_EOR) == 0) 679 printf("Egad!!\n"); 680 if (!error && *mp == NULL) 681 error = EPIPE; 682 len -= auio.uio_resid; 683 } 684 errout: 685 if (error && error != EINTR && error != ERESTART) { 686 m_freemp(mp); 687 if (error != EPIPE) 688 log(LOG_INFO, 689 "receive error %d from nfs server %s\n", 690 error, 691 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 692 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 693 if (!error) { 694 error = nfs_reconnect(rep); 695 if (!error) 696 goto tryagain; 697 nfs_sndunlock(&rep->r_nmp->nm_flag); 698 } 699 } 700 } else { 701 if ((so = rep->r_nmp->nm_so) == NULL) 702 return (EACCES); 703 if (so->so_state & SS_ISCONNECTED) 704 getnam = NULL; 705 else 706 getnam = aname; 707 auio.uio_resid = len = 1000000; 708 auio.uio_procp = p; 709 do { 710 rcvflg = 0; 711 error = soreceive(so, getnam, &auio, mp, NULL, 712 &rcvflg, 0); 713 if (error == EWOULDBLOCK && 714 (rep->r_flags & R_SOFTTERM)) 715 return (EINTR); 716 } while (error == EWOULDBLOCK); 717 len -= auio.uio_resid; 718 } 719 if (error) 720 m_freemp(mp); 721 /* 722 * Search for any mbufs that are not a multiple of 4 bytes long 723 * or with m_data not longword aligned. 724 * These could cause pointer alignment problems, so copy them to 725 * well aligned mbufs. 726 */ 727 nfs_realign(mp, 5 * NFSX_UNSIGNED); 728 return (error); 729 } 730 731 /* 732 * Implement receipt of reply on a socket. 733 * We must search through the list of received datagrams matching them 734 * with outstanding requests using the xid, until ours is found. 735 */ 736 int 737 nfs_reply(struct nfsreq *myrep) 738 { 739 struct nfsreq *rep; 740 struct nfsmount *nmp = myrep->r_nmp; 741 struct nfsm_info info; 742 struct mbuf *nam; 743 u_int32_t rxid, *tl, t1; 744 caddr_t cp2; 745 int error; 746 747 /* 748 * Loop around until we get our own reply 749 */ 750 for (;;) { 751 /* 752 * Lock against other receivers so that I don't get stuck in 753 * sbwait() after someone else has received my reply for me. 754 * Also necessary for connection based protocols to avoid 755 * race conditions during a reconnect. 756 */ 757 error = nfs_rcvlock(myrep); 758 if (error) 759 return (error == EALREADY ? 0 : error); 760 761 /* 762 * Get the next Rpc reply off the socket 763 */ 764 error = nfs_receive(myrep, &nam, &info.nmi_mrep); 765 nfs_rcvunlock(&nmp->nm_flag); 766 if (error) { 767 768 /* 769 * Ignore routing errors on connectionless protocols?? 770 */ 771 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 772 if (nmp->nm_so) 773 nmp->nm_so->so_error = 0; 774 continue; 775 } 776 return (error); 777 } 778 m_freem(nam); 779 780 /* 781 * Get the xid and check that it is an rpc reply 782 */ 783 info.nmi_md = info.nmi_mrep; 784 info.nmi_dpos = mtod(info.nmi_md, caddr_t); 785 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 786 rxid = *tl++; 787 if (*tl != rpc_reply) { 788 nfsstats.rpcinvalid++; 789 m_freem(info.nmi_mrep); 790 nfsmout: 791 continue; 792 } 793 794 /* 795 * Loop through the request list to match up the reply 796 * Iff no match, just drop the datagram 797 */ 798 TAILQ_FOREACH(rep, &nmp->nm_reqsq, r_chain) { 799 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 800 /* Found it.. */ 801 rep->r_mrep = info.nmi_mrep; 802 rep->r_md = info.nmi_md; 803 rep->r_dpos = info.nmi_dpos; 804 805 /* 806 * Update congestion window. 807 * Do the additive increase of 808 * one rpc/rtt. 809 */ 810 if (nmp->nm_cwnd <= nmp->nm_sent) { 811 nmp->nm_cwnd += 812 (NFS_CWNDSCALE * NFS_CWNDSCALE + 813 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 814 if (nmp->nm_cwnd > NFS_MAXCWND) 815 nmp->nm_cwnd = NFS_MAXCWND; 816 } 817 rep->r_flags &= ~R_SENT; 818 nmp->nm_sent -= NFS_CWNDSCALE; 819 820 if (rep->r_flags & R_TIMING) 821 nfs_update_rtt(rep); 822 823 nmp->nm_timeouts = 0; 824 break; 825 } 826 } 827 /* 828 * If not matched to a request, drop it. 829 * If it's mine, get out. 830 */ 831 if (rep == 0) { 832 nfsstats.rpcunexpected++; 833 m_freem(info.nmi_mrep); 834 } else if (rep == myrep) { 835 if (rep->r_mrep == NULL) 836 panic("nfsreply nil"); 837 return (0); 838 } 839 } 840 } 841 842 /* 843 * nfs_request - goes something like this 844 * - fill in request struct 845 * - links it into list 846 * - calls nfs_send() for first transmit 847 * - calls nfs_receive() to get reply 848 * - break down rpc header and return with nfs reply pointed to 849 * by mrep or error 850 * nb: always frees up mreq mbuf list 851 */ 852 int 853 nfs_request(struct vnode *vp, int procnum, struct nfsm_info *infop) 854 { 855 struct mbuf *m; 856 u_int32_t *tl; 857 struct nfsmount *nmp; 858 caddr_t cp2; 859 int t1, i, error = 0; 860 int trylater_delay; 861 struct nfsreq *rep; 862 struct nfsm_info info; 863 864 rep = pool_get(&nfsreqpl, PR_WAITOK); 865 rep->r_nmp = VFSTONFS(vp->v_mount); 866 rep->r_vp = vp; 867 rep->r_procp = infop->nmi_procp; 868 rep->r_procnum = procnum; 869 870 /* empty mbuf for AUTH_UNIX header */ 871 rep->r_mreq = m_gethdr(M_WAIT, MT_DATA); 872 rep->r_mreq->m_next = infop->nmi_mreq; 873 rep->r_mreq->m_len = 0; 874 m_calchdrlen(rep->r_mreq); 875 876 trylater_delay = NFS_MINTIMEO; 877 878 nmp = rep->r_nmp; 879 880 /* Get the RPC header with authorization. */ 881 nfsm_rpchead(rep, infop->nmi_cred, RPCAUTH_UNIX); 882 m = rep->r_mreq; 883 884 /* 885 * For stream protocols, insert a Sun RPC Record Mark. 886 */ 887 if (nmp->nm_sotype == SOCK_STREAM) { 888 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 889 *mtod(m, u_int32_t *) = htonl(0x80000000 | 890 (m->m_pkthdr.len - NFSX_UNSIGNED)); 891 } 892 893 tryagain: 894 rep->r_rtt = rep->r_rexmit = 0; 895 if (nfs_ptimers[rep->r_procnum] != NFS_DEFAULT_TIMER) 896 rep->r_flags = R_TIMING; 897 else 898 rep->r_flags = 0; 899 rep->r_mrep = NULL; 900 901 /* 902 * Do the client side RPC. 903 */ 904 nfsstats.rpcrequests++; 905 /* 906 * Chain request into list of outstanding requests. Be sure 907 * to put it LAST so timer finds oldest requests first. 908 */ 909 if (TAILQ_EMPTY(&nmp->nm_reqsq)) 910 timeout_add(&nmp->nm_rtimeout, nfs_ticks); 911 TAILQ_INSERT_TAIL(&nmp->nm_reqsq, rep, r_chain); 912 913 /* 914 * If backing off another request or avoiding congestion, don't 915 * send this one now but let timer do it. If not timing a request, 916 * do it now. 917 */ 918 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 919 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 920 nmp->nm_sent < nmp->nm_cwnd)) { 921 if (nmp->nm_soflags & PR_CONNREQUIRED) 922 error = nfs_sndlock(&nmp->nm_flag, rep); 923 if (!error) { 924 error = nfs_send(nmp->nm_so, nmp->nm_nam, 925 m_copym(m, 0, M_COPYALL, M_WAIT), rep); 926 if (nmp->nm_soflags & PR_CONNREQUIRED) 927 nfs_sndunlock(&nmp->nm_flag); 928 } 929 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 930 nmp->nm_sent += NFS_CWNDSCALE; 931 rep->r_flags |= R_SENT; 932 } 933 } else { 934 rep->r_rtt = -1; 935 } 936 937 /* 938 * Wait for the reply from our send or the timer's. 939 */ 940 if (!error || error == EPIPE) 941 error = nfs_reply(rep); 942 943 /* 944 * RPC done, unlink the request. 945 */ 946 TAILQ_REMOVE(&nmp->nm_reqsq, rep, r_chain); 947 if (TAILQ_EMPTY(&nmp->nm_reqsq)) 948 timeout_del(&nmp->nm_rtimeout); 949 950 /* 951 * Decrement the outstanding request count. 952 */ 953 if (rep->r_flags & R_SENT) { 954 rep->r_flags &= ~R_SENT; /* paranoia */ 955 nmp->nm_sent -= NFS_CWNDSCALE; 956 } 957 958 /* 959 * If there was a successful reply and a tprintf msg. 960 * tprintf a response. 961 */ 962 if (!error && (rep->r_flags & R_TPRINTFMSG)) 963 nfs_msg(rep, "is alive again"); 964 info.nmi_mrep = rep->r_mrep; 965 info.nmi_md = rep->r_md; 966 info.nmi_dpos = rep->r_dpos; 967 if (error) { 968 infop->nmi_mrep = NULL; 969 goto nfsmout1; 970 } 971 972 /* 973 * break down the rpc header and check if ok 974 */ 975 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 976 if (*tl++ == rpc_msgdenied) { 977 if (*tl == rpc_mismatch) 978 error = EOPNOTSUPP; 979 else 980 error = EACCES; /* Should be EAUTH. */ 981 infop->nmi_mrep = NULL; 982 goto nfsmout1; 983 } 984 985 /* 986 * Since we only support RPCAUTH_UNIX atm we step over the 987 * reply verifer type, and in the (error) case that there really 988 * is any data in it, we advance over it. 989 */ 990 tl++; /* Step over verifer type */ 991 i = fxdr_unsigned(int32_t, *tl); 992 if (i > 0) 993 nfsm_adv(nfsm_rndup(i)); /* Should not happen */ 994 995 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 996 /* 0 == ok */ 997 if (*tl == 0) { 998 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 999 if (*tl != 0) { 1000 error = fxdr_unsigned(int, *tl); 1001 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1002 error == NFSERR_TRYLATER) { 1003 m_freem(info.nmi_mrep); 1004 error = 0; 1005 tsleep_nsec(&nowake, PSOCK, "nfsretry", 1006 SEC_TO_NSEC(trylater_delay)); 1007 trylater_delay *= NFS_TIMEOUTMUL; 1008 if (trylater_delay > NFS_MAXTIMEO) 1009 trylater_delay = NFS_MAXTIMEO; 1010 1011 goto tryagain; 1012 } 1013 1014 /* 1015 * If the File Handle was stale, invalidate the 1016 * lookup cache, just in case. 1017 */ 1018 if (error == ESTALE) 1019 cache_purge(rep->r_vp); 1020 } 1021 goto nfsmout; 1022 } 1023 1024 error = EPROTONOSUPPORT; 1025 1026 nfsmout: 1027 infop->nmi_mrep = info.nmi_mrep; 1028 infop->nmi_md = info.nmi_md; 1029 infop->nmi_dpos = info.nmi_dpos; 1030 nfsmout1: 1031 m_freem(rep->r_mreq); 1032 pool_put(&nfsreqpl, rep); 1033 return (error); 1034 } 1035 #endif /* NFSCLIENT */ 1036 1037 /* 1038 * Generate the rpc reply header 1039 * siz arg. is used to decide if adding a cluster is worthwhile 1040 */ 1041 int 1042 nfs_rephead(int siz, struct nfsrv_descript *nd, struct nfssvc_sock *slp, 1043 int err, struct mbuf **mrq, struct mbuf **mbp) 1044 { 1045 u_int32_t *tl; 1046 struct mbuf *mreq; 1047 struct mbuf *mb; 1048 1049 MGETHDR(mreq, M_WAIT, MT_DATA); 1050 mb = mreq; 1051 /* 1052 * If this is a big reply, use a cluster else 1053 * try and leave leading space for the lower level headers. 1054 */ 1055 siz += RPC_REPLYSIZ; 1056 if (siz >= MHLEN - max_hdr) { 1057 MCLGET(mreq, M_WAIT); 1058 } else 1059 mreq->m_data += max_hdr; 1060 tl = mtod(mreq, u_int32_t *); 1061 mreq->m_len = 6 * NFSX_UNSIGNED; 1062 *tl++ = txdr_unsigned(nd->nd_retxid); 1063 *tl++ = rpc_reply; 1064 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1065 *tl++ = rpc_msgdenied; 1066 if (err & NFSERR_AUTHERR) { 1067 *tl++ = rpc_autherr; 1068 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1069 mreq->m_len -= NFSX_UNSIGNED; 1070 } else { 1071 *tl++ = rpc_mismatch; 1072 *tl++ = txdr_unsigned(RPC_VER2); 1073 *tl = txdr_unsigned(RPC_VER2); 1074 } 1075 } else { 1076 *tl++ = rpc_msgaccepted; 1077 1078 /* AUTH_UNIX requires RPCAUTH_NULL. */ 1079 *tl++ = 0; 1080 *tl++ = 0; 1081 1082 switch (err) { 1083 case EPROGUNAVAIL: 1084 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1085 break; 1086 case EPROGMISMATCH: 1087 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1088 tl = nfsm_build(&mb, 2 * NFSX_UNSIGNED); 1089 *tl++ = txdr_unsigned(NFS_VER2); 1090 *tl = txdr_unsigned(NFS_VER3); 1091 break; 1092 case EPROCUNAVAIL: 1093 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1094 break; 1095 case EBADRPC: 1096 *tl = txdr_unsigned(RPC_GARBAGE); 1097 break; 1098 default: 1099 *tl = 0; 1100 if (err != NFSERR_RETVOID) { 1101 tl = nfsm_build(&mb, NFSX_UNSIGNED); 1102 if (err) 1103 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1104 else 1105 *tl = 0; 1106 } 1107 break; 1108 }; 1109 } 1110 1111 *mrq = mreq; 1112 if (mbp != NULL) 1113 *mbp = mb; 1114 if (err != 0 && err != NFSERR_RETVOID) 1115 nfsstats.srvrpc_errs++; 1116 return (0); 1117 } 1118 1119 /* 1120 * nfs timer routine 1121 * Scan the nfsreq list and retransmit any requests that have timed out. 1122 */ 1123 void 1124 nfs_timer(void *arg) 1125 { 1126 struct nfsmount *nmp = arg; 1127 struct nfsreq *rep; 1128 struct mbuf *m; 1129 struct socket *so; 1130 int timeo, error; 1131 1132 NET_LOCK(); 1133 TAILQ_FOREACH(rep, &nmp->nm_reqsq, r_chain) { 1134 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1135 continue; 1136 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1137 rep->r_flags |= R_SOFTTERM; 1138 continue; 1139 } 1140 if (rep->r_rtt >= 0) { 1141 rep->r_rtt++; 1142 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1143 timeo = nmp->nm_timeo; 1144 else 1145 timeo = nfs_estimate_rto(nmp, rep->r_procnum); 1146 if (nmp->nm_timeouts > 0) 1147 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1148 if (rep->r_rtt <= timeo) 1149 continue; 1150 if (nmp->nm_timeouts < nitems(nfs_backoff)) 1151 nmp->nm_timeouts++; 1152 } 1153 1154 /* Check for server not responding. */ 1155 if ((rep->r_flags & R_TPRINTFMSG) == 0 && rep->r_rexmit > 4) { 1156 nfs_msg(rep, "not responding"); 1157 rep->r_flags |= R_TPRINTFMSG; 1158 } 1159 if (rep->r_rexmit >= nmp->nm_retry) { /* too many */ 1160 nfsstats.rpctimeouts++; 1161 rep->r_flags |= R_SOFTTERM; 1162 continue; 1163 } 1164 if (nmp->nm_sotype != SOCK_DGRAM) { 1165 if (++rep->r_rexmit > NFS_MAXREXMIT) 1166 rep->r_rexmit = NFS_MAXREXMIT; 1167 continue; 1168 } 1169 1170 if ((so = nmp->nm_so) == NULL) 1171 continue; 1172 1173 /* 1174 * If there is enough space and the window allows.. 1175 * Resend it 1176 * Set r_rtt to -1 in case we fail to send it now. 1177 */ 1178 rep->r_rtt = -1; 1179 if (sbspace(so, &so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1180 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1181 (rep->r_flags & R_SENT) || 1182 nmp->nm_sent < nmp->nm_cwnd) && 1183 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1184 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1185 error = pru_send(so, m, NULL, NULL); 1186 else 1187 error = pru_send(so, m, nmp->nm_nam, NULL); 1188 if (error) { 1189 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) 1190 so->so_error = 0; 1191 } else { 1192 /* 1193 * Iff first send, start timing 1194 * else turn timing off, backoff timer 1195 * and divide congestion window by 2. 1196 */ 1197 if (rep->r_flags & R_SENT) { 1198 rep->r_flags &= ~R_TIMING; 1199 if (++rep->r_rexmit > NFS_MAXREXMIT) 1200 rep->r_rexmit = NFS_MAXREXMIT; 1201 nmp->nm_cwnd >>= 1; 1202 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1203 nmp->nm_cwnd = NFS_CWNDSCALE; 1204 nfsstats.rpcretries++; 1205 } else { 1206 rep->r_flags |= R_SENT; 1207 nmp->nm_sent += NFS_CWNDSCALE; 1208 } 1209 rep->r_rtt = 0; 1210 } 1211 } 1212 } 1213 NET_UNLOCK(); 1214 timeout_add(&nmp->nm_rtimeout, nfs_ticks); 1215 } 1216 1217 /* 1218 * Test for a termination condition pending on the process. 1219 * This is used for NFSMNT_INT mounts. 1220 */ 1221 int 1222 nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct proc *p) 1223 { 1224 1225 if (rep && (rep->r_flags & R_SOFTTERM)) 1226 return (EINTR); 1227 if (!(nmp->nm_flag & NFSMNT_INT)) 1228 return (0); 1229 if (p && (SIGPENDING(p) & ~p->p_p->ps_sigacts->ps_sigignore & 1230 NFSINT_SIGMASK)) 1231 return (EINTR); 1232 return (0); 1233 } 1234 1235 /* 1236 * Lock a socket against others. 1237 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1238 * and also to avoid race conditions between the processes with nfs requests 1239 * in progress when a reconnect is necessary. 1240 */ 1241 int 1242 nfs_sndlock(int *flagp, struct nfsreq *rep) 1243 { 1244 uint64_t slptimeo = INFSLP; 1245 struct proc *p; 1246 int slpflag = 0; 1247 1248 if (rep) { 1249 p = rep->r_procp; 1250 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1251 slpflag = PCATCH; 1252 } else 1253 p = NULL; 1254 while (*flagp & NFSMNT_SNDLOCK) { 1255 if (rep && nfs_sigintr(rep->r_nmp, rep, p)) 1256 return (EINTR); 1257 *flagp |= NFSMNT_WANTSND; 1258 tsleep_nsec(flagp, slpflag | (PZERO - 1), "nfsndlck", slptimeo); 1259 if (slpflag == PCATCH) { 1260 slpflag = 0; 1261 slptimeo = SEC_TO_NSEC(2); 1262 } 1263 } 1264 *flagp |= NFSMNT_SNDLOCK; 1265 return (0); 1266 } 1267 1268 /* 1269 * Unlock the stream socket for others. 1270 */ 1271 void 1272 nfs_sndunlock(int *flagp) 1273 { 1274 1275 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1276 panic("nfs sndunlock"); 1277 *flagp &= ~NFSMNT_SNDLOCK; 1278 if (*flagp & NFSMNT_WANTSND) { 1279 *flagp &= ~NFSMNT_WANTSND; 1280 wakeup((caddr_t)flagp); 1281 } 1282 } 1283 1284 int 1285 nfs_rcvlock(struct nfsreq *rep) 1286 { 1287 uint64_t slptimeo = INFSLP; 1288 int *flagp = &rep->r_nmp->nm_flag; 1289 int slpflag; 1290 1291 if (*flagp & NFSMNT_INT) 1292 slpflag = PCATCH; 1293 else 1294 slpflag = 0; 1295 1296 while (*flagp & NFSMNT_RCVLOCK) { 1297 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1298 return (EINTR); 1299 *flagp |= NFSMNT_WANTRCV; 1300 tsleep_nsec(flagp, slpflag | (PZERO - 1), "nfsrcvlk", slptimeo); 1301 if (rep->r_mrep != NULL) { 1302 /* 1303 * Don't take the lock if our reply has been received 1304 * while we where sleeping. 1305 */ 1306 return (EALREADY); 1307 } 1308 if (slpflag == PCATCH) { 1309 slpflag = 0; 1310 slptimeo = SEC_TO_NSEC(2); 1311 } 1312 } 1313 *flagp |= NFSMNT_RCVLOCK; 1314 return (0); 1315 } 1316 1317 /* 1318 * Unlock the stream socket for others. 1319 */ 1320 void 1321 nfs_rcvunlock(int *flagp) 1322 { 1323 1324 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1325 panic("nfs rcvunlock"); 1326 *flagp &= ~NFSMNT_RCVLOCK; 1327 if (*flagp & NFSMNT_WANTRCV) { 1328 *flagp &= ~NFSMNT_WANTRCV; 1329 wakeup(flagp); 1330 } 1331 } 1332 1333 /* 1334 * Auxiliary routine to align the length of mbuf copies made with m_copyback(). 1335 */ 1336 void 1337 nfs_realign_fixup(struct mbuf *m, struct mbuf *n, unsigned int *off) 1338 { 1339 size_t padding; 1340 1341 /* 1342 * The maximum number of bytes that m_copyback() places in a mbuf is 1343 * always an aligned quantity, so realign happens at the chain's tail. 1344 */ 1345 while (n->m_next != NULL) 1346 n = n->m_next; 1347 1348 /* 1349 * Pad from the next elements in the source chain. Loop until the 1350 * destination chain is aligned, or the end of the source is reached. 1351 */ 1352 do { 1353 m = m->m_next; 1354 if (m == NULL) 1355 return; 1356 1357 padding = min(ALIGN(n->m_len) - n->m_len, m->m_len); 1358 if (padding > m_trailingspace(n)) 1359 panic("nfs_realign_fixup: no memory to pad to"); 1360 1361 bcopy(mtod(m, void *), mtod(n, char *) + n->m_len, padding); 1362 1363 n->m_len += padding; 1364 m_adj(m, padding); 1365 *off += padding; 1366 1367 } while (!ALIGNED_POINTER(n->m_len, void *)); 1368 } 1369 1370 /* 1371 * The NFS RPC parsing code uses the data address and the length of mbuf 1372 * structures to calculate on-memory addresses. This function makes sure these 1373 * parameters are correctly aligned. 1374 */ 1375 void 1376 nfs_realign(struct mbuf **pm, int hsiz) 1377 { 1378 struct mbuf *m; 1379 struct mbuf *n = NULL; 1380 unsigned int off = 0; 1381 1382 ++nfs_realign_test; 1383 while ((m = *pm) != NULL) { 1384 if (!ALIGNED_POINTER(m->m_data, void *) || 1385 !ALIGNED_POINTER(m->m_len, void *)) { 1386 MGET(n, M_WAIT, MT_DATA); 1387 #define ALIGN_POINTER(n) ((u_int)(((n) + sizeof(void *)) & ~sizeof(void *))) 1388 if (ALIGN_POINTER(m->m_len) >= MINCLSIZE) { 1389 MCLGET(n, M_WAIT); 1390 } 1391 n->m_len = 0; 1392 break; 1393 } 1394 pm = &m->m_next; 1395 } 1396 /* 1397 * If n is non-NULL, loop on m copying data, then replace the 1398 * portion of the chain that had to be realigned. 1399 */ 1400 if (n != NULL) { 1401 ++nfs_realign_count; 1402 while (m) { 1403 m_copyback(n, off, m->m_len, mtod(m, caddr_t), M_WAIT); 1404 1405 /* 1406 * If an unaligned amount of memory was copied, fix up 1407 * the last mbuf created by m_copyback(). 1408 */ 1409 if (!ALIGNED_POINTER(m->m_len, void *)) 1410 nfs_realign_fixup(m, n, &off); 1411 1412 off += m->m_len; 1413 m = m->m_next; 1414 } 1415 m_freemp(pm); 1416 *pm = n; 1417 } 1418 } 1419 1420 1421 /* 1422 * Parse an RPC request 1423 * - verify it 1424 * - fill in the cred struct. 1425 */ 1426 int 1427 nfs_getreq(struct nfsrv_descript *nd, struct nfsd *nfsd, int has_header) 1428 { 1429 int len, i; 1430 u_int32_t *tl; 1431 int32_t t1; 1432 caddr_t cp2; 1433 u_int32_t nfsvers, auth_type; 1434 int error = 0; 1435 struct nfsm_info info; 1436 1437 info.nmi_mrep = nd->nd_mrep; 1438 info.nmi_md = nd->nd_md; 1439 info.nmi_dpos = nd->nd_dpos; 1440 if (has_header) { 1441 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1442 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1443 if (*tl++ != rpc_call) { 1444 m_freem(info.nmi_mrep); 1445 return (EBADRPC); 1446 } 1447 } else 1448 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1449 nd->nd_repstat = 0; 1450 nd->nd_flag = 0; 1451 if (*tl++ != rpc_vers) { 1452 nd->nd_repstat = ERPCMISMATCH; 1453 nd->nd_procnum = NFSPROC_NOOP; 1454 return (0); 1455 } 1456 if (*tl != nfs_prog) { 1457 nd->nd_repstat = EPROGUNAVAIL; 1458 nd->nd_procnum = NFSPROC_NOOP; 1459 return (0); 1460 } 1461 tl++; 1462 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1463 if (nfsvers != NFS_VER2 && nfsvers != NFS_VER3) { 1464 nd->nd_repstat = EPROGMISMATCH; 1465 nd->nd_procnum = NFSPROC_NOOP; 1466 return (0); 1467 } 1468 if (nfsvers == NFS_VER3) 1469 nd->nd_flag = ND_NFSV3; 1470 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1471 if (nd->nd_procnum == NFSPROC_NULL) 1472 return (0); 1473 if (nd->nd_procnum >= NFS_NPROCS || 1474 (nd->nd_procnum > NFSPROC_COMMIT) || 1475 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1476 nd->nd_repstat = EPROCUNAVAIL; 1477 nd->nd_procnum = NFSPROC_NOOP; 1478 return (0); 1479 } 1480 if ((nd->nd_flag & ND_NFSV3) == 0) 1481 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1482 auth_type = *tl++; 1483 len = fxdr_unsigned(int, *tl++); 1484 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1485 m_freem(info.nmi_mrep); 1486 return (EBADRPC); 1487 } 1488 1489 /* Handle auth_unix */ 1490 if (auth_type == rpc_auth_unix) { 1491 len = fxdr_unsigned(int, *++tl); 1492 if (len < 0 || len > NFS_MAXNAMLEN) { 1493 m_freem(info.nmi_mrep); 1494 return (EBADRPC); 1495 } 1496 nfsm_adv(nfsm_rndup(len)); 1497 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1498 memset(&nd->nd_cr, 0, sizeof (struct ucred)); 1499 refcnt_init(&nd->nd_cr.cr_refcnt); 1500 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1501 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1502 len = fxdr_unsigned(int, *tl); 1503 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1504 m_freem(info.nmi_mrep); 1505 return (EBADRPC); 1506 } 1507 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1508 for (i = 0; i < len; i++) { 1509 if (i < NGROUPS_MAX) 1510 nd->nd_cr.cr_groups[i] = 1511 fxdr_unsigned(gid_t, *tl++); 1512 else 1513 tl++; 1514 } 1515 nd->nd_cr.cr_ngroups = (len > NGROUPS_MAX) ? NGROUPS_MAX : len; 1516 len = fxdr_unsigned(int, *++tl); 1517 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1518 m_freem(info.nmi_mrep); 1519 return (EBADRPC); 1520 } 1521 if (len > 0) 1522 nfsm_adv(nfsm_rndup(len)); 1523 } else { 1524 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1525 nd->nd_procnum = NFSPROC_NOOP; 1526 return (0); 1527 } 1528 1529 nd->nd_md = info.nmi_md; 1530 nd->nd_dpos = info.nmi_dpos; 1531 return (0); 1532 nfsmout: 1533 return (error); 1534 } 1535 1536 void 1537 nfs_msg(struct nfsreq *rep, char *msg) 1538 { 1539 tpr_t tpr; 1540 1541 if (rep->r_procp) 1542 tpr = tprintf_open(rep->r_procp); 1543 else 1544 tpr = NULL; 1545 1546 tprintf(tpr, "nfs server %s: %s\n", 1547 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname, msg); 1548 tprintf_close(tpr); 1549 } 1550 1551 #ifdef NFSSERVER 1552 /* 1553 * Socket upcall routine for the nfsd sockets. 1554 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1555 * Essentially do as much as possible non-blocking, else punt and it will 1556 * be called with M_WAIT from an nfsd. 1557 */ 1558 void 1559 nfsrv_rcv(struct socket *so, caddr_t arg, int waitflag) 1560 { 1561 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1562 struct mbuf *m; 1563 struct mbuf *mp, *nam; 1564 struct uio auio; 1565 int flags, error; 1566 1567 KERNEL_LOCK(); 1568 1569 if ((slp->ns_flag & SLP_VALID) == 0) 1570 goto out; 1571 1572 /* Defer soreceive() to an nfsd. */ 1573 if (waitflag == M_DONTWAIT) { 1574 slp->ns_flag |= SLP_NEEDQ; 1575 goto dorecs; 1576 } 1577 1578 auio.uio_procp = NULL; 1579 if (so->so_type == SOCK_STREAM) { 1580 /* 1581 * Do soreceive(). 1582 */ 1583 auio.uio_resid = 1000000000; 1584 flags = MSG_DONTWAIT; 1585 error = soreceive(so, &nam, &auio, &mp, NULL, 1586 &flags, 0); 1587 if (error || mp == NULL) { 1588 if (error == EWOULDBLOCK) 1589 slp->ns_flag |= SLP_NEEDQ; 1590 else 1591 slp->ns_flag |= SLP_DISCONN; 1592 goto dorecs; 1593 } 1594 m = mp; 1595 if (slp->ns_rawend) { 1596 slp->ns_rawend->m_next = m; 1597 slp->ns_cc += 1000000000 - auio.uio_resid; 1598 } else { 1599 slp->ns_raw = m; 1600 slp->ns_cc = 1000000000 - auio.uio_resid; 1601 } 1602 while (m->m_next) 1603 m = m->m_next; 1604 slp->ns_rawend = m; 1605 1606 /* 1607 * Now try and parse record(s) out of the raw stream data. 1608 */ 1609 error = nfsrv_getstream(slp, waitflag); 1610 if (error) { 1611 if (error == EPERM) 1612 slp->ns_flag |= SLP_DISCONN; 1613 else 1614 slp->ns_flag |= SLP_NEEDQ; 1615 } 1616 } else { 1617 do { 1618 auio.uio_resid = 1000000000; 1619 flags = MSG_DONTWAIT; 1620 error = soreceive(so, &nam, &auio, &mp, 1621 NULL, &flags, 0); 1622 if (mp) { 1623 if (nam) { 1624 m = nam; 1625 m->m_next = mp; 1626 } else 1627 m = mp; 1628 if (slp->ns_recend) 1629 slp->ns_recend->m_nextpkt = m; 1630 else 1631 slp->ns_rec = m; 1632 slp->ns_recend = m; 1633 m->m_nextpkt = NULL; 1634 } 1635 if (error) { 1636 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1637 && error != EWOULDBLOCK) { 1638 slp->ns_flag |= SLP_DISCONN; 1639 goto dorecs; 1640 } 1641 } 1642 } while (mp); 1643 } 1644 1645 /* 1646 * Now try and process the request records, non-blocking. 1647 */ 1648 dorecs: 1649 if (waitflag == M_DONTWAIT && 1650 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1651 nfsrv_wakenfsd(slp); 1652 1653 out: 1654 KERNEL_UNLOCK(); 1655 } 1656 1657 /* 1658 * Try and extract an RPC request from the mbuf data list received on a 1659 * stream socket. The "waitflag" argument indicates whether or not it 1660 * can sleep. 1661 */ 1662 int 1663 nfsrv_getstream(struct nfssvc_sock *slp, int waitflag) 1664 { 1665 struct mbuf *m, **mpp; 1666 char *cp1, *cp2; 1667 int len; 1668 struct mbuf *om, *m2, *recm; 1669 u_int32_t recmark; 1670 1671 if (slp->ns_flag & SLP_GETSTREAM) 1672 return (0); 1673 slp->ns_flag |= SLP_GETSTREAM; 1674 for (;;) { 1675 if (slp->ns_reclen == 0) { 1676 if (slp->ns_cc < NFSX_UNSIGNED) { 1677 slp->ns_flag &= ~SLP_GETSTREAM; 1678 return (0); 1679 } 1680 m = slp->ns_raw; 1681 if (m->m_len >= NFSX_UNSIGNED) { 1682 bcopy(mtod(m, caddr_t), &recmark, 1683 NFSX_UNSIGNED); 1684 m->m_data += NFSX_UNSIGNED; 1685 m->m_len -= NFSX_UNSIGNED; 1686 } else { 1687 cp1 = (caddr_t)&recmark; 1688 cp2 = mtod(m, caddr_t); 1689 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1690 while (m->m_len == 0) { 1691 m = m->m_next; 1692 cp2 = mtod(m, caddr_t); 1693 } 1694 *cp1++ = *cp2++; 1695 m->m_data++; 1696 m->m_len--; 1697 } 1698 } 1699 slp->ns_cc -= NFSX_UNSIGNED; 1700 recmark = ntohl(recmark); 1701 slp->ns_reclen = recmark & ~0x80000000; 1702 if (recmark & 0x80000000) 1703 slp->ns_flag |= SLP_LASTFRAG; 1704 else 1705 slp->ns_flag &= ~SLP_LASTFRAG; 1706 if (slp->ns_reclen > NFS_MAXPACKET) { 1707 slp->ns_flag &= ~SLP_GETSTREAM; 1708 return (EPERM); 1709 } 1710 } 1711 1712 /* 1713 * Now get the record part. 1714 */ 1715 recm = NULL; 1716 if (slp->ns_cc == slp->ns_reclen) { 1717 recm = slp->ns_raw; 1718 slp->ns_raw = slp->ns_rawend = NULL; 1719 slp->ns_cc = slp->ns_reclen = 0; 1720 } else if (slp->ns_cc > slp->ns_reclen) { 1721 len = 0; 1722 m = slp->ns_raw; 1723 om = NULL; 1724 while (len < slp->ns_reclen) { 1725 if ((len + m->m_len) > slp->ns_reclen) { 1726 m2 = m_copym(m, 0, slp->ns_reclen - len, 1727 waitflag); 1728 if (m2) { 1729 if (om) { 1730 om->m_next = m2; 1731 recm = slp->ns_raw; 1732 } else 1733 recm = m2; 1734 m->m_data += slp->ns_reclen-len; 1735 m->m_len -= slp->ns_reclen-len; 1736 len = slp->ns_reclen; 1737 } else { 1738 slp->ns_flag &= ~SLP_GETSTREAM; 1739 return (EWOULDBLOCK); 1740 } 1741 } else if ((len + m->m_len) == slp->ns_reclen) { 1742 om = m; 1743 len += m->m_len; 1744 m = m->m_next; 1745 recm = slp->ns_raw; 1746 om->m_next = NULL; 1747 } else { 1748 om = m; 1749 len += m->m_len; 1750 m = m->m_next; 1751 } 1752 } 1753 slp->ns_raw = m; 1754 slp->ns_cc -= len; 1755 slp->ns_reclen = 0; 1756 } else { 1757 slp->ns_flag &= ~SLP_GETSTREAM; 1758 return (0); 1759 } 1760 1761 /* 1762 * Accumulate the fragments into a record. 1763 */ 1764 mpp = &slp->ns_frag; 1765 while (*mpp) 1766 mpp = &((*mpp)->m_next); 1767 *mpp = recm; 1768 if (slp->ns_flag & SLP_LASTFRAG) { 1769 if (slp->ns_recend) 1770 slp->ns_recend->m_nextpkt = slp->ns_frag; 1771 else 1772 slp->ns_rec = slp->ns_frag; 1773 slp->ns_recend = slp->ns_frag; 1774 slp->ns_frag = NULL; 1775 } 1776 } 1777 } 1778 1779 /* 1780 * Parse an RPC header. 1781 */ 1782 int 1783 nfsrv_dorec(struct nfssvc_sock *slp, struct nfsd *nfsd, 1784 struct nfsrv_descript **ndp) 1785 { 1786 struct mbuf *m, *nam; 1787 struct nfsrv_descript *nd; 1788 int error; 1789 1790 *ndp = NULL; 1791 if ((slp->ns_flag & SLP_VALID) == 0 || 1792 (m = slp->ns_rec) == NULL) 1793 return (ENOBUFS); 1794 slp->ns_rec = m->m_nextpkt; 1795 if (slp->ns_rec) 1796 m->m_nextpkt = NULL; 1797 else 1798 slp->ns_recend = NULL; 1799 if (m->m_type == MT_SONAME) { 1800 nam = m; 1801 m = m->m_next; 1802 nam->m_next = NULL; 1803 } else 1804 nam = NULL; 1805 nd = pool_get(&nfsrv_descript_pl, PR_WAITOK); 1806 nfs_realign(&m, 10 * NFSX_UNSIGNED); 1807 nd->nd_md = nd->nd_mrep = m; 1808 nd->nd_nam2 = nam; 1809 nd->nd_dpos = mtod(m, caddr_t); 1810 error = nfs_getreq(nd, nfsd, 1); 1811 if (error) { 1812 m_freem(nam); 1813 pool_put(&nfsrv_descript_pl, nd); 1814 return (error); 1815 } 1816 *ndp = nd; 1817 nfsd->nfsd_nd = nd; 1818 return (0); 1819 } 1820 1821 1822 /* 1823 * Search for a sleeping nfsd and wake it up. 1824 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 1825 * running nfsds will go look for the work in the nfssvc_sock list. 1826 */ 1827 void 1828 nfsrv_wakenfsd(struct nfssvc_sock *slp) 1829 { 1830 struct nfsd *nfsd; 1831 1832 if ((slp->ns_flag & SLP_VALID) == 0) 1833 return; 1834 1835 TAILQ_FOREACH(nfsd, &nfsd_head, nfsd_chain) { 1836 if (nfsd->nfsd_flag & NFSD_WAITING) { 1837 nfsd->nfsd_flag &= ~NFSD_WAITING; 1838 if (nfsd->nfsd_slp) 1839 panic("nfsd wakeup"); 1840 slp->ns_sref++; 1841 nfsd->nfsd_slp = slp; 1842 wakeup_one(nfsd); 1843 return; 1844 } 1845 } 1846 1847 slp->ns_flag |= SLP_DOREC; 1848 nfsd_head_flag |= NFSD_CHECKSLP; 1849 } 1850 #endif /* NFSSERVER */ 1851