1 /* 2 * Copyright (c) 1989, 1991, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 37 * $FreeBSD: src/sys/nfs/nfs_socket.c,v 1.60.2.6 2003/03/26 01:44:46 alfred Exp $ 38 * $DragonFly: src/sys/vfs/nfs/nfs_socket.c,v 1.30 2005/09/17 07:43:12 dillon Exp $ 39 */ 40 41 /* 42 * Socket operations for use by nfs 43 */ 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/proc.h> 48 #include <sys/malloc.h> 49 #include <sys/mount.h> 50 #include <sys/kernel.h> 51 #include <sys/mbuf.h> 52 #include <sys/vnode.h> 53 #include <sys/protosw.h> 54 #include <sys/resourcevar.h> 55 #include <sys/socket.h> 56 #include <sys/socketvar.h> 57 #include <sys/socketops.h> 58 #include <sys/syslog.h> 59 #include <sys/thread.h> 60 #include <sys/tprintf.h> 61 #include <sys/sysctl.h> 62 #include <sys/signalvar.h> 63 64 #include <netinet/in.h> 65 #include <netinet/tcp.h> 66 #include <sys/thread2.h> 67 68 #include "rpcv2.h" 69 #include "nfsproto.h" 70 #include "nfs.h" 71 #include "xdr_subs.h" 72 #include "nfsm_subs.h" 73 #include "nfsmount.h" 74 #include "nfsnode.h" 75 #include "nfsrtt.h" 76 #include "nqnfs.h" 77 78 #define TRUE 1 79 #define FALSE 0 80 81 /* 82 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 83 * Use the mean and mean deviation of rtt for the appropriate type of rpc 84 * for the frequent rpcs and a default for the others. 85 * The justification for doing "other" this way is that these rpcs 86 * happen so infrequently that timer est. would probably be stale. 87 * Also, since many of these rpcs are 88 * non-idempotent, a conservative timeout is desired. 89 * getattr, lookup - A+2D 90 * read, write - A+4D 91 * other - nm_timeo 92 */ 93 #define NFS_RTO(n, t) \ 94 ((t) == 0 ? (n)->nm_timeo : \ 95 ((t) < 3 ? \ 96 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 97 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 98 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 99 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 100 /* 101 * External data, mostly RPC constants in XDR form 102 */ 103 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 104 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, 105 rpc_auth_kerb; 106 extern u_int32_t nfs_prog, nqnfs_prog; 107 extern time_t nqnfsstarttime; 108 extern struct nfsstats nfsstats; 109 extern int nfsv3_procid[NFS_NPROCS]; 110 extern int nfs_ticks; 111 112 /* 113 * Defines which timer to use for the procnum. 114 * 0 - default 115 * 1 - getattr 116 * 2 - lookup 117 * 3 - read 118 * 4 - write 119 */ 120 static int proct[NFS_NPROCS] = { 121 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0, 122 0, 0, 0, 123 }; 124 125 static int nfs_realign_test; 126 static int nfs_realign_count; 127 static int nfs_bufpackets = 4; 128 129 SYSCTL_DECL(_vfs_nfs); 130 131 SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_test, CTLFLAG_RW, &nfs_realign_test, 0, ""); 132 SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_count, CTLFLAG_RW, &nfs_realign_count, 0, ""); 133 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0, ""); 134 135 136 /* 137 * There is a congestion window for outstanding rpcs maintained per mount 138 * point. The cwnd size is adjusted in roughly the way that: 139 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 140 * SIGCOMM '88". ACM, August 1988. 141 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 142 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 143 * of rpcs is in progress. 144 * (The sent count and cwnd are scaled for integer arith.) 145 * Variants of "slow start" were tried and were found to be too much of a 146 * performance hit (ave. rtt 3 times larger), 147 * I suspect due to the large rtt that nfs rpcs have. 148 */ 149 #define NFS_CWNDSCALE 256 150 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 151 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 152 int nfsrtton = 0; 153 struct nfsrtt nfsrtt; 154 struct callout nfs_timer_handle; 155 156 static int nfs_msg (struct thread *,char *,char *); 157 static int nfs_rcvlock (struct nfsreq *); 158 static void nfs_rcvunlock (struct nfsreq *); 159 static void nfs_realign (struct mbuf **pm, int hsiz); 160 static int nfs_receive (struct nfsreq *rep, struct sockaddr **aname, 161 struct mbuf **mp); 162 static void nfs_softterm (struct nfsreq *rep); 163 static int nfs_reconnect (struct nfsreq *rep); 164 #ifndef NFS_NOSERVER 165 static int nfsrv_getstream (struct nfssvc_sock *, int, int *); 166 167 int (*nfsrv3_procs[NFS_NPROCS]) (struct nfsrv_descript *nd, 168 struct nfssvc_sock *slp, 169 struct thread *td, 170 struct mbuf **mreqp) = { 171 nfsrv_null, 172 nfsrv_getattr, 173 nfsrv_setattr, 174 nfsrv_lookup, 175 nfsrv3_access, 176 nfsrv_readlink, 177 nfsrv_read, 178 nfsrv_write, 179 nfsrv_create, 180 nfsrv_mkdir, 181 nfsrv_symlink, 182 nfsrv_mknod, 183 nfsrv_remove, 184 nfsrv_rmdir, 185 nfsrv_rename, 186 nfsrv_link, 187 nfsrv_readdir, 188 nfsrv_readdirplus, 189 nfsrv_statfs, 190 nfsrv_fsinfo, 191 nfsrv_pathconf, 192 nfsrv_commit, 193 nqnfsrv_getlease, 194 nqnfsrv_vacated, 195 nfsrv_noop, 196 nfsrv_noop 197 }; 198 #endif /* NFS_NOSERVER */ 199 200 /* 201 * Initialize sockets and congestion for a new NFS connection. 202 * We do not free the sockaddr if error. 203 */ 204 int 205 nfs_connect(struct nfsmount *nmp, struct nfsreq *rep) 206 { 207 struct socket *so; 208 int error, rcvreserve, sndreserve; 209 int pktscale; 210 struct sockaddr *saddr; 211 struct sockaddr_in *sin; 212 struct thread *td = &thread0; /* only used for socreate and sobind */ 213 214 nmp->nm_so = (struct socket *)0; 215 saddr = nmp->nm_nam; 216 error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, 217 nmp->nm_soproto, td); 218 if (error) 219 goto bad; 220 so = nmp->nm_so; 221 nmp->nm_soflags = so->so_proto->pr_flags; 222 223 /* 224 * Some servers require that the client port be a reserved port number. 225 */ 226 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 227 struct sockopt sopt; 228 int ip; 229 struct sockaddr_in ssin; 230 231 bzero(&sopt, sizeof sopt); 232 ip = IP_PORTRANGE_LOW; 233 sopt.sopt_level = IPPROTO_IP; 234 sopt.sopt_name = IP_PORTRANGE; 235 sopt.sopt_val = (void *)&ip; 236 sopt.sopt_valsize = sizeof(ip); 237 sopt.sopt_td = NULL; 238 error = sosetopt(so, &sopt); 239 if (error) 240 goto bad; 241 bzero(&ssin, sizeof ssin); 242 sin = &ssin; 243 sin->sin_len = sizeof (struct sockaddr_in); 244 sin->sin_family = AF_INET; 245 sin->sin_addr.s_addr = INADDR_ANY; 246 sin->sin_port = htons(0); 247 error = sobind(so, (struct sockaddr *)sin, td); 248 if (error) 249 goto bad; 250 bzero(&sopt, sizeof sopt); 251 ip = IP_PORTRANGE_DEFAULT; 252 sopt.sopt_level = IPPROTO_IP; 253 sopt.sopt_name = IP_PORTRANGE; 254 sopt.sopt_val = (void *)&ip; 255 sopt.sopt_valsize = sizeof(ip); 256 sopt.sopt_td = NULL; 257 error = sosetopt(so, &sopt); 258 if (error) 259 goto bad; 260 } 261 262 /* 263 * Protocols that do not require connections may be optionally left 264 * unconnected for servers that reply from a port other than NFS_PORT. 265 */ 266 if (nmp->nm_flag & NFSMNT_NOCONN) { 267 if (nmp->nm_soflags & PR_CONNREQUIRED) { 268 error = ENOTCONN; 269 goto bad; 270 } 271 } else { 272 error = soconnect(so, nmp->nm_nam, td); 273 if (error) 274 goto bad; 275 276 /* 277 * Wait for the connection to complete. Cribbed from the 278 * connect system call but with the wait timing out so 279 * that interruptible mounts don't hang here for a long time. 280 */ 281 crit_enter(); 282 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 283 (void) tsleep((caddr_t)&so->so_timeo, 0, 284 "nfscon", 2 * hz); 285 if ((so->so_state & SS_ISCONNECTING) && 286 so->so_error == 0 && rep && 287 (error = nfs_sigintr(nmp, rep, rep->r_td)) != 0){ 288 so->so_state &= ~SS_ISCONNECTING; 289 crit_exit(); 290 goto bad; 291 } 292 } 293 if (so->so_error) { 294 error = so->so_error; 295 so->so_error = 0; 296 crit_exit(); 297 goto bad; 298 } 299 crit_exit(); 300 } 301 so->so_rcv.sb_timeo = (5 * hz); 302 so->so_snd.sb_timeo = (5 * hz); 303 304 /* 305 * Get buffer reservation size from sysctl, but impose reasonable 306 * limits. 307 */ 308 pktscale = nfs_bufpackets; 309 if (pktscale < 2) 310 pktscale = 2; 311 if (pktscale > 64) 312 pktscale = 64; 313 314 if (nmp->nm_sotype == SOCK_DGRAM) { 315 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale; 316 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 317 NFS_MAXPKTHDR) * pktscale; 318 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 319 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale; 320 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 321 NFS_MAXPKTHDR) * pktscale; 322 } else { 323 if (nmp->nm_sotype != SOCK_STREAM) 324 panic("nfscon sotype"); 325 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 326 struct sockopt sopt; 327 int val; 328 329 bzero(&sopt, sizeof sopt); 330 sopt.sopt_level = SOL_SOCKET; 331 sopt.sopt_name = SO_KEEPALIVE; 332 sopt.sopt_val = &val; 333 sopt.sopt_valsize = sizeof val; 334 val = 1; 335 sosetopt(so, &sopt); 336 } 337 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 338 struct sockopt sopt; 339 int val; 340 341 bzero(&sopt, sizeof sopt); 342 sopt.sopt_level = IPPROTO_TCP; 343 sopt.sopt_name = TCP_NODELAY; 344 sopt.sopt_val = &val; 345 sopt.sopt_valsize = sizeof val; 346 val = 1; 347 sosetopt(so, &sopt); 348 } 349 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 350 sizeof (u_int32_t)) * pktscale; 351 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 352 sizeof (u_int32_t)) * pktscale; 353 } 354 error = soreserve(so, sndreserve, rcvreserve, 355 &td->td_proc->p_rlimit[RLIMIT_SBSIZE]); 356 if (error) 357 goto bad; 358 so->so_rcv.sb_flags |= SB_NOINTR; 359 so->so_snd.sb_flags |= SB_NOINTR; 360 361 /* Initialize other non-zero congestion variables */ 362 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = 363 nmp->nm_srtt[3] = (NFS_TIMEO << 3); 364 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 365 nmp->nm_sdrtt[3] = 0; 366 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 367 nmp->nm_sent = 0; 368 nmp->nm_timeouts = 0; 369 return (0); 370 371 bad: 372 nfs_disconnect(nmp); 373 return (error); 374 } 375 376 /* 377 * Reconnect routine: 378 * Called when a connection is broken on a reliable protocol. 379 * - clean up the old socket 380 * - nfs_connect() again 381 * - set R_MUSTRESEND for all outstanding requests on mount point 382 * If this fails the mount point is DEAD! 383 * nb: Must be called with the nfs_sndlock() set on the mount point. 384 */ 385 static int 386 nfs_reconnect(struct nfsreq *rep) 387 { 388 struct nfsreq *rp; 389 struct nfsmount *nmp = rep->r_nmp; 390 int error; 391 392 nfs_disconnect(nmp); 393 while ((error = nfs_connect(nmp, rep)) != 0) { 394 if (error == EINTR || error == ERESTART) 395 return (EINTR); 396 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0); 397 } 398 399 /* 400 * Loop through outstanding request list and fix up all requests 401 * on old socket. 402 */ 403 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 404 if (rp->r_nmp == nmp) 405 rp->r_flags |= R_MUSTRESEND; 406 } 407 return (0); 408 } 409 410 /* 411 * NFS disconnect. Clean up and unlink. 412 */ 413 void 414 nfs_disconnect(struct nfsmount *nmp) 415 { 416 struct socket *so; 417 418 if (nmp->nm_so) { 419 so = nmp->nm_so; 420 nmp->nm_so = (struct socket *)0; 421 soshutdown(so, 2); 422 soclose(so); 423 } 424 } 425 426 void 427 nfs_safedisconnect(struct nfsmount *nmp) 428 { 429 struct nfsreq dummyreq; 430 431 bzero(&dummyreq, sizeof(dummyreq)); 432 dummyreq.r_nmp = nmp; 433 dummyreq.r_td = NULL; 434 nfs_rcvlock(&dummyreq); 435 nfs_disconnect(nmp); 436 nfs_rcvunlock(&dummyreq); 437 } 438 439 /* 440 * This is the nfs send routine. For connection based socket types, it 441 * must be called with an nfs_sndlock() on the socket. 442 * "rep == NULL" indicates that it has been called from a server. 443 * For the client side: 444 * - return EINTR if the RPC is terminated, 0 otherwise 445 * - set R_MUSTRESEND if the send fails for any reason 446 * - do any cleanup required by recoverable socket errors (?) 447 * For the server side: 448 * - return EINTR or ERESTART if interrupted by a signal 449 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 450 * - do any cleanup required by recoverable socket errors (?) 451 */ 452 int 453 nfs_send(struct socket *so, struct sockaddr *nam, struct mbuf *top, 454 struct nfsreq *rep) 455 { 456 struct sockaddr *sendnam; 457 int error, soflags, flags; 458 459 if (rep) { 460 if (rep->r_flags & R_SOFTTERM) { 461 m_freem(top); 462 return (EINTR); 463 } 464 if ((so = rep->r_nmp->nm_so) == NULL) { 465 rep->r_flags |= R_MUSTRESEND; 466 m_freem(top); 467 return (0); 468 } 469 rep->r_flags &= ~R_MUSTRESEND; 470 soflags = rep->r_nmp->nm_soflags; 471 } else 472 soflags = so->so_proto->pr_flags; 473 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 474 sendnam = (struct sockaddr *)0; 475 else 476 sendnam = nam; 477 if (so->so_type == SOCK_SEQPACKET) 478 flags = MSG_EOR; 479 else 480 flags = 0; 481 482 error = so_pru_sosend(so, sendnam, NULL, top, NULL, flags, 483 curthread /*XXX*/); 484 /* 485 * ENOBUFS for dgram sockets is transient and non fatal. 486 * No need to log, and no need to break a soft mount. 487 */ 488 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 489 error = 0; 490 if (rep) /* do backoff retransmit on client */ 491 rep->r_flags |= R_MUSTRESEND; 492 } 493 494 if (error) { 495 if (rep) { 496 log(LOG_INFO, "nfs send error %d for server %s\n",error, 497 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 498 /* 499 * Deal with errors for the client side. 500 */ 501 if (rep->r_flags & R_SOFTTERM) 502 error = EINTR; 503 else 504 rep->r_flags |= R_MUSTRESEND; 505 } else 506 log(LOG_INFO, "nfsd send error %d\n", error); 507 508 /* 509 * Handle any recoverable (soft) socket errors here. (?) 510 */ 511 if (error != EINTR && error != ERESTART && 512 error != EWOULDBLOCK && error != EPIPE) 513 error = 0; 514 } 515 return (error); 516 } 517 518 /* 519 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 520 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 521 * Mark and consolidate the data into a new mbuf list. 522 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 523 * small mbufs. 524 * For SOCK_STREAM we must be very careful to read an entire record once 525 * we have read any of it, even if the system call has been interrupted. 526 */ 527 static int 528 nfs_receive(struct nfsreq *rep, struct sockaddr **aname, struct mbuf **mp) 529 { 530 struct socket *so; 531 struct uio auio; 532 struct iovec aio; 533 struct mbuf *m; 534 struct mbuf *control; 535 u_int32_t len; 536 struct sockaddr **getnam; 537 int error, sotype, rcvflg; 538 struct thread *td = curthread; /* XXX */ 539 540 /* 541 * Set up arguments for soreceive() 542 */ 543 *mp = (struct mbuf *)0; 544 *aname = (struct sockaddr *)0; 545 sotype = rep->r_nmp->nm_sotype; 546 547 /* 548 * For reliable protocols, lock against other senders/receivers 549 * in case a reconnect is necessary. 550 * For SOCK_STREAM, first get the Record Mark to find out how much 551 * more there is to get. 552 * We must lock the socket against other receivers 553 * until we have an entire rpc request/reply. 554 */ 555 if (sotype != SOCK_DGRAM) { 556 error = nfs_sndlock(rep); 557 if (error) 558 return (error); 559 tryagain: 560 /* 561 * Check for fatal errors and resending request. 562 */ 563 /* 564 * Ugh: If a reconnect attempt just happened, nm_so 565 * would have changed. NULL indicates a failed 566 * attempt that has essentially shut down this 567 * mount point. 568 */ 569 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 570 nfs_sndunlock(rep); 571 return (EINTR); 572 } 573 so = rep->r_nmp->nm_so; 574 if (!so) { 575 error = nfs_reconnect(rep); 576 if (error) { 577 nfs_sndunlock(rep); 578 return (error); 579 } 580 goto tryagain; 581 } 582 while (rep->r_flags & R_MUSTRESEND) { 583 m = m_copym(rep->r_mreq, 0, M_COPYALL, MB_WAIT); 584 nfsstats.rpcretries++; 585 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); 586 if (error) { 587 if (error == EINTR || error == ERESTART || 588 (error = nfs_reconnect(rep)) != 0) { 589 nfs_sndunlock(rep); 590 return (error); 591 } 592 goto tryagain; 593 } 594 } 595 nfs_sndunlock(rep); 596 if (sotype == SOCK_STREAM) { 597 aio.iov_base = (caddr_t) &len; 598 aio.iov_len = sizeof(u_int32_t); 599 auio.uio_iov = &aio; 600 auio.uio_iovcnt = 1; 601 auio.uio_segflg = UIO_SYSSPACE; 602 auio.uio_rw = UIO_READ; 603 auio.uio_offset = 0; 604 auio.uio_resid = sizeof(u_int32_t); 605 auio.uio_td = td; 606 do { 607 rcvflg = MSG_WAITALL; 608 error = so_pru_soreceive(so, NULL, &auio, NULL, 609 NULL, &rcvflg); 610 if (error == EWOULDBLOCK && rep) { 611 if (rep->r_flags & R_SOFTTERM) 612 return (EINTR); 613 } 614 } while (error == EWOULDBLOCK); 615 if (!error && auio.uio_resid > 0) { 616 /* 617 * Don't log a 0 byte receive; it means 618 * that the socket has been closed, and 619 * can happen during normal operation 620 * (forcible unmount or Solaris server). 621 */ 622 if (auio.uio_resid != sizeof (u_int32_t)) 623 log(LOG_INFO, 624 "short receive (%d/%d) from nfs server %s\n", 625 (int)(sizeof(u_int32_t) - auio.uio_resid), 626 (int)sizeof(u_int32_t), 627 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 628 error = EPIPE; 629 } 630 if (error) 631 goto errout; 632 len = ntohl(len) & ~0x80000000; 633 /* 634 * This is SERIOUS! We are out of sync with the sender 635 * and forcing a disconnect/reconnect is all I can do. 636 */ 637 if (len > NFS_MAXPACKET) { 638 log(LOG_ERR, "%s (%d) from nfs server %s\n", 639 "impossible packet length", 640 len, 641 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 642 error = EFBIG; 643 goto errout; 644 } 645 auio.uio_resid = len; 646 do { 647 rcvflg = MSG_WAITALL; 648 error = so_pru_soreceive(so, NULL, &auio, mp, 649 NULL, &rcvflg); 650 } while (error == EWOULDBLOCK || error == EINTR || 651 error == ERESTART); 652 if (!error && auio.uio_resid > 0) { 653 if (len != auio.uio_resid) 654 log(LOG_INFO, 655 "short receive (%d/%d) from nfs server %s\n", 656 len - auio.uio_resid, len, 657 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 658 error = EPIPE; 659 } 660 } else { 661 /* 662 * NB: Since uio_resid is big, MSG_WAITALL is ignored 663 * and soreceive() will return when it has either a 664 * control msg or a data msg. 665 * We have no use for control msg., but must grab them 666 * and then throw them away so we know what is going 667 * on. 668 */ 669 auio.uio_resid = len = 100000000; /* Anything Big */ 670 auio.uio_td = td; 671 do { 672 rcvflg = 0; 673 error = so_pru_soreceive(so, NULL, &auio, mp, 674 &control, &rcvflg); 675 if (control) 676 m_freem(control); 677 if (error == EWOULDBLOCK && rep) { 678 if (rep->r_flags & R_SOFTTERM) 679 return (EINTR); 680 } 681 } while (error == EWOULDBLOCK || 682 (!error && *mp == NULL && control)); 683 if ((rcvflg & MSG_EOR) == 0) 684 printf("Egad!!\n"); 685 if (!error && *mp == NULL) 686 error = EPIPE; 687 len -= auio.uio_resid; 688 } 689 errout: 690 if (error && error != EINTR && error != ERESTART) { 691 m_freem(*mp); 692 *mp = (struct mbuf *)0; 693 if (error != EPIPE) 694 log(LOG_INFO, 695 "receive error %d from nfs server %s\n", 696 error, 697 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 698 error = nfs_sndlock(rep); 699 if (!error) { 700 error = nfs_reconnect(rep); 701 if (!error) 702 goto tryagain; 703 else 704 nfs_sndunlock(rep); 705 } 706 } 707 } else { 708 if ((so = rep->r_nmp->nm_so) == NULL) 709 return (EACCES); 710 if (so->so_state & SS_ISCONNECTED) 711 getnam = (struct sockaddr **)0; 712 else 713 getnam = aname; 714 auio.uio_resid = len = 1000000; 715 auio.uio_td = td; 716 do { 717 rcvflg = 0; 718 error = so_pru_soreceive(so, getnam, &auio, mp, NULL, 719 &rcvflg); 720 if (error == EWOULDBLOCK && 721 (rep->r_flags & R_SOFTTERM)) 722 return (EINTR); 723 } while (error == EWOULDBLOCK); 724 len -= auio.uio_resid; 725 } 726 if (error) { 727 m_freem(*mp); 728 *mp = (struct mbuf *)0; 729 } 730 /* 731 * Search for any mbufs that are not a multiple of 4 bytes long 732 * or with m_data not longword aligned. 733 * These could cause pointer alignment problems, so copy them to 734 * well aligned mbufs. 735 */ 736 nfs_realign(mp, 5 * NFSX_UNSIGNED); 737 return (error); 738 } 739 740 /* 741 * Implement receipt of reply on a socket. 742 * We must search through the list of received datagrams matching them 743 * with outstanding requests using the xid, until ours is found. 744 */ 745 /* ARGSUSED */ 746 int 747 nfs_reply(struct nfsreq *myrep) 748 { 749 struct nfsreq *rep; 750 struct nfsmount *nmp = myrep->r_nmp; 751 int32_t t1; 752 struct mbuf *mrep, *md; 753 struct sockaddr *nam; 754 u_int32_t rxid, *tl; 755 caddr_t dpos, cp2; 756 int error; 757 758 /* 759 * Loop around until we get our own reply 760 */ 761 for (;;) { 762 /* 763 * Lock against other receivers so that I don't get stuck in 764 * sbwait() after someone else has received my reply for me. 765 * Also necessary for connection based protocols to avoid 766 * race conditions during a reconnect. 767 * If nfs_rcvlock() returns EALREADY, that means that 768 * the reply has already been recieved by another 769 * process and we can return immediately. In this 770 * case, the lock is not taken to avoid races with 771 * other processes. 772 */ 773 error = nfs_rcvlock(myrep); 774 if (error == EALREADY) 775 return (0); 776 if (error) 777 return (error); 778 /* 779 * Get the next Rpc reply off the socket 780 */ 781 error = nfs_receive(myrep, &nam, &mrep); 782 nfs_rcvunlock(myrep); 783 if (error) { 784 785 /* 786 * Ignore routing errors on connectionless protocols?? 787 */ 788 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 789 nmp->nm_so->so_error = 0; 790 if (myrep->r_flags & R_GETONEREP) 791 return (0); 792 continue; 793 } 794 return (error); 795 } 796 if (nam) 797 FREE(nam, M_SONAME); 798 799 /* 800 * Get the xid and check that it is an rpc reply 801 */ 802 md = mrep; 803 dpos = mtod(md, caddr_t); 804 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 805 rxid = *tl++; 806 if (*tl != rpc_reply) { 807 #ifndef NFS_NOSERVER 808 if (nmp->nm_flag & NFSMNT_NQNFS) { 809 if (nqnfs_callback(nmp, mrep, md, dpos)) 810 nfsstats.rpcinvalid++; 811 } else { 812 nfsstats.rpcinvalid++; 813 m_freem(mrep); 814 } 815 #else 816 nfsstats.rpcinvalid++; 817 m_freem(mrep); 818 #endif 819 nfsmout: 820 if (myrep->r_flags & R_GETONEREP) 821 return (0); 822 continue; 823 } 824 825 /* 826 * Loop through the request list to match up the reply 827 * Iff no match, just drop the datagram 828 */ 829 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 830 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 831 /* Found it.. */ 832 rep->r_mrep = mrep; 833 rep->r_md = md; 834 rep->r_dpos = dpos; 835 if (nfsrtton) { 836 struct rttl *rt; 837 838 rt = &nfsrtt.rttl[nfsrtt.pos]; 839 rt->proc = rep->r_procnum; 840 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 841 rt->sent = nmp->nm_sent; 842 rt->cwnd = nmp->nm_cwnd; 843 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 844 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 845 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 846 getmicrotime(&rt->tstamp); 847 if (rep->r_flags & R_TIMING) 848 rt->rtt = rep->r_rtt; 849 else 850 rt->rtt = 1000000; 851 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 852 } 853 /* 854 * Update congestion window. 855 * Do the additive increase of 856 * one rpc/rtt. 857 */ 858 if (nmp->nm_cwnd <= nmp->nm_sent) { 859 nmp->nm_cwnd += 860 (NFS_CWNDSCALE * NFS_CWNDSCALE + 861 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 862 if (nmp->nm_cwnd > NFS_MAXCWND) 863 nmp->nm_cwnd = NFS_MAXCWND; 864 } 865 crit_enter(); /* nfs_timer interlock*/ 866 if (rep->r_flags & R_SENT) { 867 rep->r_flags &= ~R_SENT; 868 nmp->nm_sent -= NFS_CWNDSCALE; 869 } 870 crit_exit(); 871 /* 872 * Update rtt using a gain of 0.125 on the mean 873 * and a gain of 0.25 on the deviation. 874 */ 875 if (rep->r_flags & R_TIMING) { 876 /* 877 * Since the timer resolution of 878 * NFS_HZ is so course, it can often 879 * result in r_rtt == 0. Since 880 * r_rtt == N means that the actual 881 * rtt is between N+dt and N+2-dt ticks, 882 * add 1. 883 */ 884 t1 = rep->r_rtt + 1; 885 t1 -= (NFS_SRTT(rep) >> 3); 886 NFS_SRTT(rep) += t1; 887 if (t1 < 0) 888 t1 = -t1; 889 t1 -= (NFS_SDRTT(rep) >> 2); 890 NFS_SDRTT(rep) += t1; 891 } 892 nmp->nm_timeouts = 0; 893 break; 894 } 895 } 896 /* 897 * If not matched to a request, drop it. 898 * If it's mine, get out. 899 */ 900 if (rep == 0) { 901 nfsstats.rpcunexpected++; 902 m_freem(mrep); 903 } else if (rep == myrep) { 904 if (rep->r_mrep == NULL) 905 panic("nfsreply nil"); 906 return (0); 907 } 908 if (myrep->r_flags & R_GETONEREP) 909 return (0); 910 } 911 } 912 913 /* 914 * nfs_request - goes something like this 915 * - fill in request struct 916 * - links it into list 917 * - calls nfs_send() for first transmit 918 * - calls nfs_receive() to get reply 919 * - break down rpc header and return with nfs reply pointed to 920 * by mrep or error 921 * nb: always frees up mreq mbuf list 922 */ 923 int 924 nfs_request(struct vnode *vp, struct mbuf *mrest, int procnum, 925 struct thread *td, struct ucred *cred, struct mbuf **mrp, 926 struct mbuf **mdp, caddr_t *dposp) 927 { 928 struct mbuf *mrep, *m2; 929 struct nfsreq *rep; 930 u_int32_t *tl; 931 int i; 932 struct nfsmount *nmp; 933 struct mbuf *m, *md, *mheadend; 934 struct nfsnode *np; 935 char nickv[RPCX_NICKVERF]; 936 time_t reqtime, waituntil; 937 caddr_t dpos, cp2; 938 int t1, nqlflag, cachable, error = 0, mrest_len, auth_len, auth_type; 939 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 940 int verf_len, verf_type; 941 int retdummy; 942 u_int32_t xid; 943 u_quad_t frev; 944 char *auth_str, *verf_str; 945 NFSKERBKEY_T key; /* save session key */ 946 947 /* Reject requests while attempting a forced unmount. */ 948 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) { 949 m_freem(mrest); 950 return (ESTALE); 951 } 952 nmp = VFSTONFS(vp->v_mount); 953 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 954 rep->r_nmp = nmp; 955 rep->r_vp = vp; 956 rep->r_td = td; 957 rep->r_procnum = procnum; 958 i = 0; 959 m = mrest; 960 while (m) { 961 i += m->m_len; 962 m = m->m_next; 963 } 964 mrest_len = i; 965 966 /* 967 * Get the RPC header with authorization. 968 */ 969 kerbauth: 970 verf_str = auth_str = (char *)0; 971 if (nmp->nm_flag & NFSMNT_KERB) { 972 verf_str = nickv; 973 verf_len = sizeof (nickv); 974 auth_type = RPCAUTH_KERB4; 975 bzero((caddr_t)key, sizeof (key)); 976 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 977 &auth_len, verf_str, verf_len)) { 978 error = nfs_getauth(nmp, rep, cred, &auth_str, 979 &auth_len, verf_str, &verf_len, key); 980 if (error) { 981 free((caddr_t)rep, M_NFSREQ); 982 m_freem(mrest); 983 return (error); 984 } 985 } 986 } else { 987 auth_type = RPCAUTH_UNIX; 988 if (cred->cr_ngroups < 1) 989 panic("nfsreq nogrps"); 990 auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ? 991 nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) + 992 5 * NFSX_UNSIGNED; 993 } 994 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 995 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 996 if (auth_str) 997 free(auth_str, M_TEMP); 998 999 /* 1000 * For stream protocols, insert a Sun RPC Record Mark. 1001 */ 1002 if (nmp->nm_sotype == SOCK_STREAM) { 1003 M_PREPEND(m, NFSX_UNSIGNED, MB_WAIT); 1004 if (m == NULL) { 1005 free(rep, M_NFSREQ); 1006 return (ENOBUFS); 1007 } 1008 *mtod(m, u_int32_t *) = htonl(0x80000000 | 1009 (m->m_pkthdr.len - NFSX_UNSIGNED)); 1010 } 1011 rep->r_mreq = m; 1012 rep->r_xid = xid; 1013 tryagain: 1014 if (nmp->nm_flag & NFSMNT_SOFT) 1015 rep->r_retry = nmp->nm_retry; 1016 else 1017 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1018 rep->r_rtt = rep->r_rexmit = 0; 1019 if (proct[procnum] > 0) 1020 rep->r_flags = R_TIMING | R_MASKTIMER; 1021 else 1022 rep->r_flags = R_MASKTIMER; 1023 rep->r_mrep = NULL; 1024 1025 /* 1026 * Do the client side RPC. 1027 */ 1028 nfsstats.rpcrequests++; 1029 1030 /* 1031 * Chain request into list of outstanding requests. Be sure 1032 * to put it LAST so timer finds oldest requests first. Note 1033 * that R_MASKTIMER is set at the moment to prevent any timer 1034 * action on this request while we are still doing processing on 1035 * it below. splsoftclock() primarily protects nm_sent. Note 1036 * that we may block in this code so there is no atomicy guarentee. 1037 */ 1038 crit_enter(); 1039 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1040 1041 /* Get send time for nqnfs */ 1042 reqtime = time_second; 1043 1044 /* 1045 * If backing off another request or avoiding congestion, don't 1046 * send this one now but let timer do it. If not timing a request, 1047 * do it now. 1048 */ 1049 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1050 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1051 nmp->nm_sent < nmp->nm_cwnd)) { 1052 if (nmp->nm_soflags & PR_CONNREQUIRED) 1053 error = nfs_sndlock(rep); 1054 if (!error) { 1055 m2 = m_copym(m, 0, M_COPYALL, MB_WAIT); 1056 error = nfs_send(nmp->nm_so, nmp->nm_nam, m2, rep); 1057 if (nmp->nm_soflags & PR_CONNREQUIRED) 1058 nfs_sndunlock(rep); 1059 } 1060 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1061 nmp->nm_sent += NFS_CWNDSCALE; 1062 rep->r_flags |= R_SENT; 1063 } 1064 } else { 1065 rep->r_rtt = -1; 1066 } 1067 1068 /* 1069 * Let the timer do what it will with the request, then 1070 * wait for the reply from our send or the timer's. 1071 */ 1072 rep->r_flags &= ~R_MASKTIMER; 1073 crit_exit(); 1074 if (!error || error == EPIPE) 1075 error = nfs_reply(rep); 1076 1077 /* 1078 * RPC done, unlink the request. 1079 */ 1080 crit_enter(); 1081 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1082 1083 /* 1084 * Decrement the outstanding request count. 1085 */ 1086 if (rep->r_flags & R_SENT) { 1087 rep->r_flags &= ~R_SENT; 1088 nmp->nm_sent -= NFS_CWNDSCALE; 1089 } 1090 crit_exit(); 1091 1092 /* 1093 * If there was a successful reply and a tprintf msg. 1094 * tprintf a response. 1095 */ 1096 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1097 nfs_msg(rep->r_td, nmp->nm_mountp->mnt_stat.f_mntfromname, 1098 "is alive again"); 1099 mrep = rep->r_mrep; 1100 md = rep->r_md; 1101 dpos = rep->r_dpos; 1102 if (error) { 1103 m_freem(rep->r_mreq); 1104 free((caddr_t)rep, M_NFSREQ); 1105 return (error); 1106 } 1107 1108 /* 1109 * break down the rpc header and check if ok 1110 */ 1111 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1112 if (*tl++ == rpc_msgdenied) { 1113 if (*tl == rpc_mismatch) 1114 error = EOPNOTSUPP; 1115 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1116 if (!failed_auth) { 1117 failed_auth++; 1118 mheadend->m_next = (struct mbuf *)0; 1119 m_freem(mrep); 1120 m_freem(rep->r_mreq); 1121 goto kerbauth; 1122 } else 1123 error = EAUTH; 1124 } else 1125 error = EACCES; 1126 m_freem(mrep); 1127 m_freem(rep->r_mreq); 1128 free((caddr_t)rep, M_NFSREQ); 1129 return (error); 1130 } 1131 1132 /* 1133 * Grab any Kerberos verifier, otherwise just throw it away. 1134 */ 1135 verf_type = fxdr_unsigned(int, *tl++); 1136 i = fxdr_unsigned(int32_t, *tl); 1137 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1138 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1139 if (error) 1140 goto nfsmout; 1141 } else if (i > 0) 1142 nfsm_adv(nfsm_rndup(i)); 1143 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1144 /* 0 == ok */ 1145 if (*tl == 0) { 1146 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1147 if (*tl != 0) { 1148 error = fxdr_unsigned(int, *tl); 1149 if ((nmp->nm_flag & NFSMNT_NFSV3) && 1150 error == NFSERR_TRYLATER) { 1151 m_freem(mrep); 1152 error = 0; 1153 waituntil = time_second + trylater_delay; 1154 while (time_second < waituntil) 1155 (void) tsleep((caddr_t)&lbolt, 1156 0, "nqnfstry", 0); 1157 trylater_delay *= nfs_backoff[trylater_cnt]; 1158 if (trylater_cnt < 7) 1159 trylater_cnt++; 1160 goto tryagain; 1161 } 1162 1163 /* 1164 * If the File Handle was stale, invalidate the 1165 * lookup cache, just in case. 1166 */ 1167 if (error == ESTALE) { 1168 retdummy = 0; 1169 cache_inval_vp(vp, CINV_CHILDREN, &retdummy); 1170 } 1171 if (nmp->nm_flag & NFSMNT_NFSV3) { 1172 *mrp = mrep; 1173 *mdp = md; 1174 *dposp = dpos; 1175 error |= NFSERR_RETERR; 1176 } else 1177 m_freem(mrep); 1178 m_freem(rep->r_mreq); 1179 free((caddr_t)rep, M_NFSREQ); 1180 return (error); 1181 } 1182 1183 /* 1184 * For nqnfs, get any lease in reply 1185 */ 1186 if (nmp->nm_flag & NFSMNT_NQNFS) { 1187 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1188 if (*tl) { 1189 np = VTONFS(vp); 1190 nqlflag = fxdr_unsigned(int, *tl); 1191 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1192 cachable = fxdr_unsigned(int, *tl++); 1193 reqtime += fxdr_unsigned(int, *tl++); 1194 if (reqtime > time_second) { 1195 frev = fxdr_hyper(tl); 1196 nqnfs_clientlease(nmp, np, nqlflag, 1197 cachable, reqtime, frev); 1198 } 1199 } 1200 } 1201 *mrp = mrep; 1202 *mdp = md; 1203 *dposp = dpos; 1204 m_freem(rep->r_mreq); 1205 FREE((caddr_t)rep, M_NFSREQ); 1206 return (0); 1207 } 1208 m_freem(mrep); 1209 error = EPROTONOSUPPORT; 1210 nfsmout: 1211 m_freem(rep->r_mreq); 1212 free((caddr_t)rep, M_NFSREQ); 1213 return (error); 1214 } 1215 1216 #ifndef NFS_NOSERVER 1217 /* 1218 * Generate the rpc reply header 1219 * siz arg. is used to decide if adding a cluster is worthwhile 1220 */ 1221 int 1222 nfs_rephead(int siz, struct nfsrv_descript *nd, struct nfssvc_sock *slp, 1223 int err, int cache, u_quad_t *frev, struct mbuf **mrq, 1224 struct mbuf **mbp, caddr_t *bposp) 1225 { 1226 u_int32_t *tl; 1227 struct mbuf *mreq; 1228 caddr_t bpos; 1229 struct mbuf *mb, *mb2; 1230 1231 siz += RPC_REPLYSIZ; 1232 mb = mreq = m_getl(max_hdr + siz, MB_WAIT, MT_DATA, M_PKTHDR, NULL); 1233 mreq->m_pkthdr.len = 0; 1234 /* 1235 * If this is not a cluster, try and leave leading space 1236 * for the lower level headers. 1237 */ 1238 if ((max_hdr + siz) < MINCLSIZE) 1239 mreq->m_data += max_hdr; 1240 tl = mtod(mreq, u_int32_t *); 1241 mreq->m_len = 6 * NFSX_UNSIGNED; 1242 bpos = ((caddr_t)tl) + mreq->m_len; 1243 *tl++ = txdr_unsigned(nd->nd_retxid); 1244 *tl++ = rpc_reply; 1245 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1246 *tl++ = rpc_msgdenied; 1247 if (err & NFSERR_AUTHERR) { 1248 *tl++ = rpc_autherr; 1249 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1250 mreq->m_len -= NFSX_UNSIGNED; 1251 bpos -= NFSX_UNSIGNED; 1252 } else { 1253 *tl++ = rpc_mismatch; 1254 *tl++ = txdr_unsigned(RPC_VER2); 1255 *tl = txdr_unsigned(RPC_VER2); 1256 } 1257 } else { 1258 *tl++ = rpc_msgaccepted; 1259 1260 /* 1261 * For Kerberos authentication, we must send the nickname 1262 * verifier back, otherwise just RPCAUTH_NULL. 1263 */ 1264 if (nd->nd_flag & ND_KERBFULL) { 1265 struct nfsuid *nuidp; 1266 struct timeval ktvin, ktvout; 1267 1268 for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; 1269 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1270 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1271 (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), 1272 &nuidp->nu_haddr, nd->nd_nam2))) 1273 break; 1274 } 1275 if (nuidp) { 1276 ktvin.tv_sec = 1277 txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); 1278 ktvin.tv_usec = 1279 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1280 1281 /* 1282 * Encrypt the timestamp in ecb mode using the 1283 * session key. 1284 */ 1285 #ifdef NFSKERB 1286 XXX 1287 #endif 1288 1289 *tl++ = rpc_auth_kerb; 1290 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1291 *tl = ktvout.tv_sec; 1292 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1293 *tl++ = ktvout.tv_usec; 1294 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1295 } else { 1296 *tl++ = 0; 1297 *tl++ = 0; 1298 } 1299 } else { 1300 *tl++ = 0; 1301 *tl++ = 0; 1302 } 1303 switch (err) { 1304 case EPROGUNAVAIL: 1305 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1306 break; 1307 case EPROGMISMATCH: 1308 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1309 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1310 if (nd->nd_flag & ND_NQNFS) { 1311 *tl++ = txdr_unsigned(3); 1312 *tl = txdr_unsigned(3); 1313 } else { 1314 *tl++ = txdr_unsigned(2); 1315 *tl = txdr_unsigned(3); 1316 } 1317 break; 1318 case EPROCUNAVAIL: 1319 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1320 break; 1321 case EBADRPC: 1322 *tl = txdr_unsigned(RPC_GARBAGE); 1323 break; 1324 default: 1325 *tl = 0; 1326 if (err != NFSERR_RETVOID) { 1327 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1328 if (err) 1329 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1330 else 1331 *tl = 0; 1332 } 1333 break; 1334 }; 1335 } 1336 1337 /* 1338 * For nqnfs, piggyback lease as requested. 1339 */ 1340 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1341 if (nd->nd_flag & ND_LEASE) { 1342 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1343 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1344 *tl++ = txdr_unsigned(cache); 1345 *tl++ = txdr_unsigned(nd->nd_duration); 1346 txdr_hyper(*frev, tl); 1347 } else { 1348 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1349 *tl = 0; 1350 } 1351 } 1352 if (mrq != NULL) 1353 *mrq = mreq; 1354 *mbp = mb; 1355 *bposp = bpos; 1356 if (err != 0 && err != NFSERR_RETVOID) 1357 nfsstats.srvrpc_errs++; 1358 return (0); 1359 } 1360 1361 1362 #endif /* NFS_NOSERVER */ 1363 /* 1364 * Nfs timer routine 1365 * Scan the nfsreq list and retranmit any requests that have timed out 1366 * To avoid retransmission attempts on STREAM sockets (in the future) make 1367 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1368 */ 1369 void 1370 nfs_timer(void *arg /* never used */) 1371 { 1372 struct nfsreq *rep; 1373 struct mbuf *m; 1374 struct socket *so; 1375 struct nfsmount *nmp; 1376 int timeo; 1377 int error; 1378 #ifndef NFS_NOSERVER 1379 static long lasttime = 0; 1380 struct nfssvc_sock *slp; 1381 u_quad_t cur_usec; 1382 #endif /* NFS_NOSERVER */ 1383 struct thread *td = &thread0; /* XXX for credentials, will break if sleep */ 1384 1385 crit_enter(); 1386 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1387 nmp = rep->r_nmp; 1388 if (rep->r_mrep || (rep->r_flags & (R_SOFTTERM|R_MASKTIMER))) 1389 continue; 1390 if (nfs_sigintr(nmp, rep, rep->r_td)) { 1391 nfs_softterm(rep); 1392 continue; 1393 } 1394 if (rep->r_rtt >= 0) { 1395 rep->r_rtt++; 1396 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1397 timeo = nmp->nm_timeo; 1398 else 1399 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1400 if (nmp->nm_timeouts > 0) 1401 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1402 if (rep->r_rtt <= timeo) 1403 continue; 1404 if (nmp->nm_timeouts < 8) 1405 nmp->nm_timeouts++; 1406 } 1407 /* 1408 * Check for server not responding 1409 */ 1410 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1411 rep->r_rexmit > nmp->nm_deadthresh) { 1412 nfs_msg(rep->r_td, 1413 nmp->nm_mountp->mnt_stat.f_mntfromname, 1414 "not responding"); 1415 rep->r_flags |= R_TPRINTFMSG; 1416 } 1417 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1418 nfsstats.rpctimeouts++; 1419 nfs_softterm(rep); 1420 continue; 1421 } 1422 if (nmp->nm_sotype != SOCK_DGRAM) { 1423 if (++rep->r_rexmit > NFS_MAXREXMIT) 1424 rep->r_rexmit = NFS_MAXREXMIT; 1425 continue; 1426 } 1427 if ((so = nmp->nm_so) == NULL) 1428 continue; 1429 1430 /* 1431 * If there is enough space and the window allows.. 1432 * Resend it 1433 * Set r_rtt to -1 in case we fail to send it now. 1434 */ 1435 rep->r_rtt = -1; 1436 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1437 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1438 (rep->r_flags & R_SENT) || 1439 nmp->nm_sent < nmp->nm_cwnd) && 1440 (m = m_copym(rep->r_mreq, 0, M_COPYALL, MB_DONTWAIT))){ 1441 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1442 error = so_pru_send(so, 0, m, (struct sockaddr *)0, 1443 (struct mbuf *)0, td); 1444 else 1445 error = so_pru_send(so, 0, m, nmp->nm_nam, 1446 (struct mbuf *)0, td); 1447 if (error) { 1448 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) 1449 so->so_error = 0; 1450 } else { 1451 /* 1452 * Iff first send, start timing 1453 * else turn timing off, backoff timer 1454 * and divide congestion window by 2. 1455 */ 1456 if (rep->r_flags & R_SENT) { 1457 rep->r_flags &= ~R_TIMING; 1458 if (++rep->r_rexmit > NFS_MAXREXMIT) 1459 rep->r_rexmit = NFS_MAXREXMIT; 1460 nmp->nm_cwnd >>= 1; 1461 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1462 nmp->nm_cwnd = NFS_CWNDSCALE; 1463 nfsstats.rpcretries++; 1464 } else { 1465 rep->r_flags |= R_SENT; 1466 nmp->nm_sent += NFS_CWNDSCALE; 1467 } 1468 rep->r_rtt = 0; 1469 } 1470 } 1471 } 1472 #ifndef NFS_NOSERVER 1473 /* 1474 * Call the nqnfs server timer once a second to handle leases. 1475 */ 1476 if (lasttime != time_second) { 1477 lasttime = time_second; 1478 nqnfs_serverd(); 1479 } 1480 1481 /* 1482 * Scan the write gathering queues for writes that need to be 1483 * completed now. 1484 */ 1485 cur_usec = nfs_curusec(); 1486 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1487 if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec) 1488 nfsrv_wakenfsd(slp, 1); 1489 } 1490 #endif /* NFS_NOSERVER */ 1491 crit_exit(); 1492 callout_reset(&nfs_timer_handle, nfs_ticks, nfs_timer, NULL); 1493 } 1494 1495 /* 1496 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and 1497 * wait for all requests to complete. This is used by forced unmounts 1498 * to terminate any outstanding RPCs. 1499 */ 1500 int 1501 nfs_nmcancelreqs(struct nfsmount *nmp) 1502 { 1503 struct nfsreq *req; 1504 int i; 1505 1506 crit_enter(); 1507 TAILQ_FOREACH(req, &nfs_reqq, r_chain) { 1508 if (nmp != req->r_nmp || req->r_mrep != NULL || 1509 (req->r_flags & R_SOFTTERM)) 1510 continue; 1511 nfs_softterm(req); 1512 } 1513 crit_exit(); 1514 1515 for (i = 0; i < 30; i++) { 1516 crit_enter(); 1517 TAILQ_FOREACH(req, &nfs_reqq, r_chain) { 1518 if (nmp == req->r_nmp) 1519 break; 1520 } 1521 crit_exit(); 1522 if (req == NULL) 1523 return (0); 1524 tsleep(&lbolt, 0, "nfscancel", 0); 1525 } 1526 return (EBUSY); 1527 } 1528 1529 /* 1530 * Flag a request as being about to terminate (due to NFSMNT_INT/NFSMNT_SOFT). 1531 * The nm_send count is decremented now to avoid deadlocks when the process in 1532 * soreceive() hasn't yet managed to send its own request. 1533 * 1534 * This routine must be called at splsoftclock() to protect r_flags and 1535 * nm_sent. 1536 */ 1537 1538 static void 1539 nfs_softterm(struct nfsreq *rep) 1540 { 1541 rep->r_flags |= R_SOFTTERM; 1542 1543 if (rep->r_flags & R_SENT) { 1544 rep->r_nmp->nm_sent -= NFS_CWNDSCALE; 1545 rep->r_flags &= ~R_SENT; 1546 } 1547 } 1548 1549 /* 1550 * Test for a termination condition pending on the process. 1551 * This is used for NFSMNT_INT mounts. 1552 */ 1553 int 1554 nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct thread *td) 1555 { 1556 sigset_t tmpset; 1557 struct proc *p; 1558 1559 if (rep && (rep->r_flags & R_SOFTTERM)) 1560 return (EINTR); 1561 /* Terminate all requests while attempting a forced unmount. */ 1562 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) 1563 return (EINTR); 1564 if (!(nmp->nm_flag & NFSMNT_INT)) 1565 return (0); 1566 /* td might be NULL YYY */ 1567 if (td == NULL || (p = td->td_proc) == NULL) 1568 return (0); 1569 1570 tmpset = p->p_siglist; 1571 SIGSETNAND(tmpset, p->p_sigmask); 1572 SIGSETNAND(tmpset, p->p_sigignore); 1573 if (SIGNOTEMPTY(p->p_siglist) && NFSINT_SIGMASK(tmpset)) 1574 return (EINTR); 1575 1576 return (0); 1577 } 1578 1579 /* 1580 * Lock a socket against others. 1581 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1582 * and also to avoid race conditions between the processes with nfs requests 1583 * in progress when a reconnect is necessary. 1584 */ 1585 int 1586 nfs_sndlock(struct nfsreq *rep) 1587 { 1588 int *statep = &rep->r_nmp->nm_state; 1589 struct thread *td; 1590 int slptimeo; 1591 int slpflag; 1592 int error; 1593 1594 slpflag = 0; 1595 slptimeo = 0; 1596 td = rep->r_td; 1597 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1598 slpflag = PCATCH; 1599 1600 error = 0; 1601 crit_enter(); 1602 while (*statep & NFSSTA_SNDLOCK) { 1603 *statep |= NFSSTA_WANTSND; 1604 if (nfs_sigintr(rep->r_nmp, rep, td)) { 1605 error = EINTR; 1606 break; 1607 } 1608 tsleep((caddr_t)statep, slpflag, "nfsndlck", slptimeo); 1609 if (slpflag == PCATCH) { 1610 slpflag = 0; 1611 slptimeo = 2 * hz; 1612 } 1613 } 1614 /* Always fail if our request has been cancelled. */ 1615 if ((rep->r_flags & R_SOFTTERM)) 1616 error = EINTR; 1617 if (error == 0) 1618 *statep |= NFSSTA_SNDLOCK; 1619 crit_exit(); 1620 return (error); 1621 } 1622 1623 /* 1624 * Unlock the stream socket for others. 1625 */ 1626 void 1627 nfs_sndunlock(struct nfsreq *rep) 1628 { 1629 int *statep = &rep->r_nmp->nm_state; 1630 1631 if ((*statep & NFSSTA_SNDLOCK) == 0) 1632 panic("nfs sndunlock"); 1633 crit_enter(); 1634 *statep &= ~NFSSTA_SNDLOCK; 1635 if (*statep & NFSSTA_WANTSND) { 1636 *statep &= ~NFSSTA_WANTSND; 1637 wakeup((caddr_t)statep); 1638 } 1639 crit_exit(); 1640 } 1641 1642 static int 1643 nfs_rcvlock(struct nfsreq *rep) 1644 { 1645 int *statep = &rep->r_nmp->nm_state; 1646 int slpflag; 1647 int slptimeo; 1648 int error; 1649 1650 /* 1651 * Unconditionally check for completion in case another nfsiod 1652 * get the packet while the caller was blocked, before the caller 1653 * called us. Packet reception is handled by mainline code which 1654 * is protected by the BGL at the moment. 1655 * 1656 * We do not strictly need the second check just before the 1657 * tsleep(), but it's good defensive programming. 1658 */ 1659 if (rep->r_mrep != NULL) 1660 return (EALREADY); 1661 1662 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1663 slpflag = PCATCH; 1664 else 1665 slpflag = 0; 1666 slptimeo = 0; 1667 error = 0; 1668 crit_enter(); 1669 while (*statep & NFSSTA_RCVLOCK) { 1670 if (nfs_sigintr(rep->r_nmp, rep, rep->r_td)) { 1671 error = EINTR; 1672 break; 1673 } 1674 if (rep->r_mrep != NULL) { 1675 error = EALREADY; 1676 break; 1677 } 1678 *statep |= NFSSTA_WANTRCV; 1679 tsleep((caddr_t)statep, slpflag, "nfsrcvlk", slptimeo); 1680 /* 1681 * If our reply was recieved while we were sleeping, 1682 * then just return without taking the lock to avoid a 1683 * situation where a single iod could 'capture' the 1684 * recieve lock. 1685 */ 1686 if (rep->r_mrep != NULL) { 1687 error = EALREADY; 1688 break; 1689 } 1690 if (slpflag == PCATCH) { 1691 slpflag = 0; 1692 slptimeo = 2 * hz; 1693 } 1694 } 1695 if (error == 0) { 1696 *statep |= NFSSTA_RCVLOCK; 1697 rep->r_nmp->nm_rcvlock_td = curthread; /* DEBUGGING */ 1698 } 1699 crit_exit(); 1700 return (error); 1701 } 1702 1703 /* 1704 * Unlock the stream socket for others. 1705 */ 1706 static void 1707 nfs_rcvunlock(struct nfsreq *rep) 1708 { 1709 int *statep = &rep->r_nmp->nm_state; 1710 1711 if ((*statep & NFSSTA_RCVLOCK) == 0) 1712 panic("nfs rcvunlock"); 1713 crit_enter(); 1714 rep->r_nmp->nm_rcvlock_td = (void *)-1; /* DEBUGGING */ 1715 *statep &= ~NFSSTA_RCVLOCK; 1716 if (*statep & NFSSTA_WANTRCV) { 1717 *statep &= ~NFSSTA_WANTRCV; 1718 wakeup((caddr_t)statep); 1719 } 1720 crit_exit(); 1721 } 1722 1723 /* 1724 * nfs_realign: 1725 * 1726 * Check for badly aligned mbuf data and realign by copying the unaligned 1727 * portion of the data into a new mbuf chain and freeing the portions 1728 * of the old chain that were replaced. 1729 * 1730 * We cannot simply realign the data within the existing mbuf chain 1731 * because the underlying buffers may contain other rpc commands and 1732 * we cannot afford to overwrite them. 1733 * 1734 * We would prefer to avoid this situation entirely. The situation does 1735 * not occur with NFS/UDP and is supposed to only occassionally occur 1736 * with TCP. Use vfs.nfs.realign_count and realign_test to check this. 1737 */ 1738 static void 1739 nfs_realign(struct mbuf **pm, int hsiz) 1740 { 1741 struct mbuf *m; 1742 struct mbuf *n = NULL; 1743 int off = 0; 1744 1745 ++nfs_realign_test; 1746 1747 while ((m = *pm) != NULL) { 1748 if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) { 1749 n = m_getl(m->m_len, MB_WAIT, MT_DATA, 0, NULL); 1750 n->m_len = 0; 1751 break; 1752 } 1753 pm = &m->m_next; 1754 } 1755 1756 /* 1757 * If n is non-NULL, loop on m copying data, then replace the 1758 * portion of the chain that had to be realigned. 1759 */ 1760 if (n != NULL) { 1761 ++nfs_realign_count; 1762 while (m) { 1763 m_copyback(n, off, m->m_len, mtod(m, caddr_t)); 1764 off += m->m_len; 1765 m = m->m_next; 1766 } 1767 m_freem(*pm); 1768 *pm = n; 1769 } 1770 } 1771 1772 #ifndef NFS_NOSERVER 1773 1774 /* 1775 * Parse an RPC request 1776 * - verify it 1777 * - fill in the cred struct. 1778 */ 1779 int 1780 nfs_getreq(struct nfsrv_descript *nd, struct nfsd *nfsd, int has_header) 1781 { 1782 int len, i; 1783 u_int32_t *tl; 1784 int32_t t1; 1785 struct uio uio; 1786 struct iovec iov; 1787 caddr_t dpos, cp2, cp; 1788 u_int32_t nfsvers, auth_type; 1789 uid_t nickuid; 1790 int error = 0, nqnfs = 0, ticklen; 1791 struct mbuf *mrep, *md; 1792 struct nfsuid *nuidp; 1793 struct timeval tvin, tvout; 1794 #if 0 /* until encrypted keys are implemented */ 1795 NFSKERBKEYSCHED_T keys; /* stores key schedule */ 1796 #endif 1797 1798 mrep = nd->nd_mrep; 1799 md = nd->nd_md; 1800 dpos = nd->nd_dpos; 1801 if (has_header) { 1802 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1803 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1804 if (*tl++ != rpc_call) { 1805 m_freem(mrep); 1806 return (EBADRPC); 1807 } 1808 } else 1809 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1810 nd->nd_repstat = 0; 1811 nd->nd_flag = 0; 1812 if (*tl++ != rpc_vers) { 1813 nd->nd_repstat = ERPCMISMATCH; 1814 nd->nd_procnum = NFSPROC_NOOP; 1815 return (0); 1816 } 1817 if (*tl != nfs_prog) { 1818 if (*tl == nqnfs_prog) 1819 nqnfs++; 1820 else { 1821 nd->nd_repstat = EPROGUNAVAIL; 1822 nd->nd_procnum = NFSPROC_NOOP; 1823 return (0); 1824 } 1825 } 1826 tl++; 1827 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1828 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1829 (nfsvers != NQNFS_VER3 && nqnfs)) { 1830 nd->nd_repstat = EPROGMISMATCH; 1831 nd->nd_procnum = NFSPROC_NOOP; 1832 return (0); 1833 } 1834 if (nqnfs) 1835 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1836 else if (nfsvers == NFS_VER3) 1837 nd->nd_flag = ND_NFSV3; 1838 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1839 if (nd->nd_procnum == NFSPROC_NULL) 1840 return (0); 1841 if (nd->nd_procnum >= NFS_NPROCS || 1842 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1843 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1844 nd->nd_repstat = EPROCUNAVAIL; 1845 nd->nd_procnum = NFSPROC_NOOP; 1846 return (0); 1847 } 1848 if ((nd->nd_flag & ND_NFSV3) == 0) 1849 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1850 auth_type = *tl++; 1851 len = fxdr_unsigned(int, *tl++); 1852 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1853 m_freem(mrep); 1854 return (EBADRPC); 1855 } 1856 1857 nd->nd_flag &= ~ND_KERBAUTH; 1858 /* 1859 * Handle auth_unix or auth_kerb. 1860 */ 1861 if (auth_type == rpc_auth_unix) { 1862 len = fxdr_unsigned(int, *++tl); 1863 if (len < 0 || len > NFS_MAXNAMLEN) { 1864 m_freem(mrep); 1865 return (EBADRPC); 1866 } 1867 nfsm_adv(nfsm_rndup(len)); 1868 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1869 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred)); 1870 nd->nd_cr.cr_ref = 1; 1871 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1872 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1873 len = fxdr_unsigned(int, *tl); 1874 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1875 m_freem(mrep); 1876 return (EBADRPC); 1877 } 1878 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1879 for (i = 1; i <= len; i++) 1880 if (i < NGROUPS) 1881 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1882 else 1883 tl++; 1884 nd->nd_cr.cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1); 1885 if (nd->nd_cr.cr_ngroups > 1) 1886 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 1887 len = fxdr_unsigned(int, *++tl); 1888 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1889 m_freem(mrep); 1890 return (EBADRPC); 1891 } 1892 if (len > 0) 1893 nfsm_adv(nfsm_rndup(len)); 1894 } else if (auth_type == rpc_auth_kerb) { 1895 switch (fxdr_unsigned(int, *tl++)) { 1896 case RPCAKN_FULLNAME: 1897 ticklen = fxdr_unsigned(int, *tl); 1898 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1899 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1900 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1901 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1902 m_freem(mrep); 1903 return (EBADRPC); 1904 } 1905 uio.uio_offset = 0; 1906 uio.uio_iov = &iov; 1907 uio.uio_iovcnt = 1; 1908 uio.uio_segflg = UIO_SYSSPACE; 1909 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 1910 iov.iov_len = RPCAUTH_MAXSIZ - 4; 1911 nfsm_mtouio(&uio, uio.uio_resid); 1912 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1913 if (*tl++ != rpc_auth_kerb || 1914 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 1915 printf("Bad kerb verifier\n"); 1916 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1917 nd->nd_procnum = NFSPROC_NOOP; 1918 return (0); 1919 } 1920 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 1921 tl = (u_int32_t *)cp; 1922 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 1923 printf("Not fullname kerb verifier\n"); 1924 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1925 nd->nd_procnum = NFSPROC_NOOP; 1926 return (0); 1927 } 1928 cp += NFSX_UNSIGNED; 1929 bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED); 1930 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 1931 nd->nd_flag |= ND_KERBFULL; 1932 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 1933 break; 1934 case RPCAKN_NICKNAME: 1935 if (len != 2 * NFSX_UNSIGNED) { 1936 printf("Kerb nickname short\n"); 1937 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 1938 nd->nd_procnum = NFSPROC_NOOP; 1939 return (0); 1940 } 1941 nickuid = fxdr_unsigned(uid_t, *tl); 1942 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1943 if (*tl++ != rpc_auth_kerb || 1944 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 1945 printf("Kerb nick verifier bad\n"); 1946 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 1947 nd->nd_procnum = NFSPROC_NOOP; 1948 return (0); 1949 } 1950 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1951 tvin.tv_sec = *tl++; 1952 tvin.tv_usec = *tl; 1953 1954 for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; 1955 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1956 if (nuidp->nu_cr.cr_uid == nickuid && 1957 (!nd->nd_nam2 || 1958 netaddr_match(NU_NETFAM(nuidp), 1959 &nuidp->nu_haddr, nd->nd_nam2))) 1960 break; 1961 } 1962 if (!nuidp) { 1963 nd->nd_repstat = 1964 (NFSERR_AUTHERR|AUTH_REJECTCRED); 1965 nd->nd_procnum = NFSPROC_NOOP; 1966 return (0); 1967 } 1968 1969 /* 1970 * Now, decrypt the timestamp using the session key 1971 * and validate it. 1972 */ 1973 #ifdef NFSKERB 1974 XXX 1975 #endif 1976 1977 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 1978 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 1979 if (nuidp->nu_expire < time_second || 1980 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 1981 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 1982 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 1983 nuidp->nu_expire = 0; 1984 nd->nd_repstat = 1985 (NFSERR_AUTHERR|AUTH_REJECTVERF); 1986 nd->nd_procnum = NFSPROC_NOOP; 1987 return (0); 1988 } 1989 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 1990 nd->nd_flag |= ND_KERBNICK; 1991 }; 1992 } else { 1993 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 1994 nd->nd_procnum = NFSPROC_NOOP; 1995 return (0); 1996 } 1997 1998 /* 1999 * For nqnfs, get piggybacked lease request. 2000 */ 2001 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 2002 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2003 nd->nd_flag |= fxdr_unsigned(int, *tl); 2004 if (nd->nd_flag & ND_LEASE) { 2005 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2006 nd->nd_duration = fxdr_unsigned(int32_t, *tl); 2007 } else 2008 nd->nd_duration = NQ_MINLEASE; 2009 } else 2010 nd->nd_duration = NQ_MINLEASE; 2011 nd->nd_md = md; 2012 nd->nd_dpos = dpos; 2013 return (0); 2014 nfsmout: 2015 return (error); 2016 } 2017 2018 #endif 2019 2020 /* 2021 * Send a message to the originating process's terminal. The thread and/or 2022 * process may be NULL. YYY the thread should not be NULL but there may 2023 * still be some uio_td's that are still being passed as NULL through to 2024 * nfsm_request(). 2025 */ 2026 static int 2027 nfs_msg(struct thread *td, char *server, char *msg) 2028 { 2029 tpr_t tpr; 2030 2031 if (td && td->td_proc) 2032 tpr = tprintf_open(td->td_proc); 2033 else 2034 tpr = NULL; 2035 tprintf(tpr, "nfs server %s: %s\n", server, msg); 2036 tprintf_close(tpr); 2037 return (0); 2038 } 2039 2040 #ifndef NFS_NOSERVER 2041 /* 2042 * Socket upcall routine for the nfsd sockets. 2043 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 2044 * Essentially do as much as possible non-blocking, else punt and it will 2045 * be called with MB_WAIT from an nfsd. 2046 */ 2047 void 2048 nfsrv_rcv(struct socket *so, void *arg, int waitflag) 2049 { 2050 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 2051 struct mbuf *m; 2052 struct mbuf *mp; 2053 struct sockaddr *nam; 2054 struct uio auio; 2055 int flags, error; 2056 int nparallel_wakeup = 0; 2057 2058 if ((slp->ns_flag & SLP_VALID) == 0) 2059 return; 2060 2061 /* 2062 * Do not allow an infinite number of completed RPC records to build 2063 * up before we stop reading data from the socket. Otherwise we could 2064 * end up holding onto an unreasonable number of mbufs for requests 2065 * waiting for service. 2066 * 2067 * This should give pretty good feedback to the TCP 2068 * layer and prevents a memory crunch for other protocols. 2069 * 2070 * Note that the same service socket can be dispatched to several 2071 * nfs servers simultaniously. 2072 * 2073 * the tcp protocol callback calls us with MB_DONTWAIT. 2074 * nfsd calls us with MB_WAIT (typically). 2075 */ 2076 if (waitflag == MB_DONTWAIT && slp->ns_numrec >= nfsd_waiting / 2 + 1) { 2077 slp->ns_flag |= SLP_NEEDQ; 2078 goto dorecs; 2079 } 2080 2081 /* 2082 * Handle protocol specifics to parse an RPC request. We always 2083 * pull from the socket using non-blocking I/O. 2084 */ 2085 auio.uio_td = NULL; 2086 if (so->so_type == SOCK_STREAM) { 2087 /* 2088 * The data has to be read in an orderly fashion from a TCP 2089 * stream, unlike a UDP socket. It is possible for soreceive 2090 * and/or nfsrv_getstream() to block, so make sure only one 2091 * entity is messing around with the TCP stream at any given 2092 * moment. The receive sockbuf's lock in soreceive is not 2093 * sufficient. 2094 * 2095 * Note that this procedure can be called from any number of 2096 * NFS severs *OR* can be upcalled directly from a TCP 2097 * protocol thread. 2098 */ 2099 if (slp->ns_flag & SLP_GETSTREAM) { 2100 slp->ns_flag |= SLP_NEEDQ; 2101 goto dorecs; 2102 } 2103 slp->ns_flag |= SLP_GETSTREAM; 2104 2105 /* 2106 * Do soreceive(). 2107 */ 2108 auio.uio_resid = 1000000000; 2109 flags = MSG_DONTWAIT; 2110 error = so_pru_soreceive(so, &nam, &auio, &mp, NULL, &flags); 2111 if (error || mp == (struct mbuf *)0) { 2112 if (error == EWOULDBLOCK) 2113 slp->ns_flag |= SLP_NEEDQ; 2114 else 2115 slp->ns_flag |= SLP_DISCONN; 2116 slp->ns_flag &= ~SLP_GETSTREAM; 2117 goto dorecs; 2118 } 2119 m = mp; 2120 if (slp->ns_rawend) { 2121 slp->ns_rawend->m_next = m; 2122 slp->ns_cc += 1000000000 - auio.uio_resid; 2123 } else { 2124 slp->ns_raw = m; 2125 slp->ns_cc = 1000000000 - auio.uio_resid; 2126 } 2127 while (m->m_next) 2128 m = m->m_next; 2129 slp->ns_rawend = m; 2130 2131 /* 2132 * Now try and parse as many record(s) as we can out of the 2133 * raw stream data. 2134 */ 2135 error = nfsrv_getstream(slp, waitflag, &nparallel_wakeup); 2136 if (error) { 2137 if (error == EPERM) 2138 slp->ns_flag |= SLP_DISCONN; 2139 else 2140 slp->ns_flag |= SLP_NEEDQ; 2141 } 2142 slp->ns_flag &= ~SLP_GETSTREAM; 2143 } else { 2144 /* 2145 * For UDP soreceive typically pulls just one packet, loop 2146 * to get the whole batch. 2147 */ 2148 do { 2149 auio.uio_resid = 1000000000; 2150 flags = MSG_DONTWAIT; 2151 error = so_pru_soreceive(so, &nam, &auio, &mp, NULL, 2152 &flags); 2153 if (mp) { 2154 struct nfsrv_rec *rec; 2155 int mf = (waitflag & MB_DONTWAIT) ? 2156 M_NOWAIT : M_WAITOK; 2157 rec = malloc(sizeof(struct nfsrv_rec), 2158 M_NFSRVDESC, mf); 2159 if (!rec) { 2160 if (nam) 2161 FREE(nam, M_SONAME); 2162 m_freem(mp); 2163 continue; 2164 } 2165 nfs_realign(&mp, 10 * NFSX_UNSIGNED); 2166 rec->nr_address = nam; 2167 rec->nr_packet = mp; 2168 STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link); 2169 ++slp->ns_numrec; 2170 ++nparallel_wakeup; 2171 } 2172 if (error) { 2173 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2174 && error != EWOULDBLOCK) { 2175 slp->ns_flag |= SLP_DISCONN; 2176 goto dorecs; 2177 } 2178 } 2179 } while (mp); 2180 } 2181 2182 /* 2183 * If we were upcalled from the tcp protocol layer and we have 2184 * fully parsed records ready to go, or there is new data pending, 2185 * or something went wrong, try to wake up an nfsd thread to deal 2186 * with it. 2187 */ 2188 dorecs: 2189 if (waitflag == MB_DONTWAIT && (slp->ns_numrec > 0 2190 || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) { 2191 nfsrv_wakenfsd(slp, nparallel_wakeup); 2192 } 2193 } 2194 2195 /* 2196 * Try and extract an RPC request from the mbuf data list received on a 2197 * stream socket. The "waitflag" argument indicates whether or not it 2198 * can sleep. 2199 */ 2200 static int 2201 nfsrv_getstream(struct nfssvc_sock *slp, int waitflag, int *countp) 2202 { 2203 struct mbuf *m, **mpp; 2204 char *cp1, *cp2; 2205 int len; 2206 struct mbuf *om, *m2, *recm; 2207 u_int32_t recmark; 2208 2209 for (;;) { 2210 if (slp->ns_reclen == 0) { 2211 if (slp->ns_cc < NFSX_UNSIGNED) 2212 return (0); 2213 m = slp->ns_raw; 2214 if (m->m_len >= NFSX_UNSIGNED) { 2215 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 2216 m->m_data += NFSX_UNSIGNED; 2217 m->m_len -= NFSX_UNSIGNED; 2218 } else { 2219 cp1 = (caddr_t)&recmark; 2220 cp2 = mtod(m, caddr_t); 2221 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 2222 while (m->m_len == 0) { 2223 m = m->m_next; 2224 cp2 = mtod(m, caddr_t); 2225 } 2226 *cp1++ = *cp2++; 2227 m->m_data++; 2228 m->m_len--; 2229 } 2230 } 2231 slp->ns_cc -= NFSX_UNSIGNED; 2232 recmark = ntohl(recmark); 2233 slp->ns_reclen = recmark & ~0x80000000; 2234 if (recmark & 0x80000000) 2235 slp->ns_flag |= SLP_LASTFRAG; 2236 else 2237 slp->ns_flag &= ~SLP_LASTFRAG; 2238 if (slp->ns_reclen > NFS_MAXPACKET) { 2239 log(LOG_ERR, "%s (%d) from nfs client\n", 2240 "impossible packet length", 2241 slp->ns_reclen); 2242 return (EPERM); 2243 } 2244 } 2245 2246 /* 2247 * Now get the record part. 2248 * 2249 * Note that slp->ns_reclen may be 0. Linux sometimes 2250 * generates 0-length RPCs 2251 */ 2252 recm = NULL; 2253 if (slp->ns_cc == slp->ns_reclen) { 2254 recm = slp->ns_raw; 2255 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2256 slp->ns_cc = slp->ns_reclen = 0; 2257 } else if (slp->ns_cc > slp->ns_reclen) { 2258 len = 0; 2259 m = slp->ns_raw; 2260 om = (struct mbuf *)0; 2261 2262 while (len < slp->ns_reclen) { 2263 if ((len + m->m_len) > slp->ns_reclen) { 2264 m2 = m_copym(m, 0, slp->ns_reclen - len, 2265 waitflag); 2266 if (m2) { 2267 if (om) { 2268 om->m_next = m2; 2269 recm = slp->ns_raw; 2270 } else 2271 recm = m2; 2272 m->m_data += slp->ns_reclen - len; 2273 m->m_len -= slp->ns_reclen - len; 2274 len = slp->ns_reclen; 2275 } else { 2276 return (EWOULDBLOCK); 2277 } 2278 } else if ((len + m->m_len) == slp->ns_reclen) { 2279 om = m; 2280 len += m->m_len; 2281 m = m->m_next; 2282 recm = slp->ns_raw; 2283 om->m_next = (struct mbuf *)0; 2284 } else { 2285 om = m; 2286 len += m->m_len; 2287 m = m->m_next; 2288 } 2289 } 2290 slp->ns_raw = m; 2291 slp->ns_cc -= len; 2292 slp->ns_reclen = 0; 2293 } else { 2294 return (0); 2295 } 2296 2297 /* 2298 * Accumulate the fragments into a record. 2299 */ 2300 mpp = &slp->ns_frag; 2301 while (*mpp) 2302 mpp = &((*mpp)->m_next); 2303 *mpp = recm; 2304 if (slp->ns_flag & SLP_LASTFRAG) { 2305 struct nfsrv_rec *rec; 2306 int mf = (waitflag & MB_DONTWAIT) ? M_NOWAIT : M_WAITOK; 2307 rec = malloc(sizeof(struct nfsrv_rec), M_NFSRVDESC, mf); 2308 if (!rec) { 2309 m_freem(slp->ns_frag); 2310 } else { 2311 nfs_realign(&slp->ns_frag, 10 * NFSX_UNSIGNED); 2312 rec->nr_address = (struct sockaddr *)0; 2313 rec->nr_packet = slp->ns_frag; 2314 STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link); 2315 ++slp->ns_numrec; 2316 ++*countp; 2317 } 2318 slp->ns_frag = (struct mbuf *)0; 2319 } 2320 } 2321 } 2322 2323 /* 2324 * Parse an RPC header. 2325 */ 2326 int 2327 nfsrv_dorec(struct nfssvc_sock *slp, struct nfsd *nfsd, 2328 struct nfsrv_descript **ndp) 2329 { 2330 struct nfsrv_rec *rec; 2331 struct mbuf *m; 2332 struct sockaddr *nam; 2333 struct nfsrv_descript *nd; 2334 int error; 2335 2336 *ndp = NULL; 2337 if ((slp->ns_flag & SLP_VALID) == 0 || !STAILQ_FIRST(&slp->ns_rec)) 2338 return (ENOBUFS); 2339 rec = STAILQ_FIRST(&slp->ns_rec); 2340 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link); 2341 KKASSERT(slp->ns_numrec > 0); 2342 --slp->ns_numrec; 2343 nam = rec->nr_address; 2344 m = rec->nr_packet; 2345 free(rec, M_NFSRVDESC); 2346 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), 2347 M_NFSRVDESC, M_WAITOK); 2348 nd->nd_md = nd->nd_mrep = m; 2349 nd->nd_nam2 = nam; 2350 nd->nd_dpos = mtod(m, caddr_t); 2351 error = nfs_getreq(nd, nfsd, TRUE); 2352 if (error) { 2353 if (nam) { 2354 FREE(nam, M_SONAME); 2355 } 2356 free((caddr_t)nd, M_NFSRVDESC); 2357 return (error); 2358 } 2359 *ndp = nd; 2360 nfsd->nfsd_nd = nd; 2361 return (0); 2362 } 2363 2364 /* 2365 * Try to assign service sockets to nfsd threads based on the number 2366 * of new rpc requests that have been queued on the service socket. 2367 * 2368 * If no nfsd's are available or additonal requests are pending, set the 2369 * NFSD_CHECKSLP flag so that one of the running nfsds will go look for 2370 * the work in the nfssvc_sock list when it is finished processing its 2371 * current work. This flag is only cleared when an nfsd can not find 2372 * any new work to perform. 2373 */ 2374 void 2375 nfsrv_wakenfsd(struct nfssvc_sock *slp, int nparallel) 2376 { 2377 struct nfsd *nd; 2378 2379 if ((slp->ns_flag & SLP_VALID) == 0) 2380 return; 2381 if (nparallel <= 1) 2382 nparallel = 1; 2383 TAILQ_FOREACH(nd, &nfsd_head, nfsd_chain) { 2384 if (nd->nfsd_flag & NFSD_WAITING) { 2385 nd->nfsd_flag &= ~NFSD_WAITING; 2386 if (nd->nfsd_slp) 2387 panic("nfsd wakeup"); 2388 slp->ns_sref++; 2389 nd->nfsd_slp = slp; 2390 wakeup((caddr_t)nd); 2391 if (--nparallel == 0) 2392 break; 2393 } 2394 } 2395 if (nparallel) { 2396 slp->ns_flag |= SLP_DOREC; 2397 nfsd_head_flag |= NFSD_CHECKSLP; 2398 } 2399 } 2400 #endif /* NFS_NOSERVER */ 2401