nfs_socket.c revision 1.174
1 /* $NetBSD: nfs_socket.c,v 1.174 2008/11/19 18:36:09 ad Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1991, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 35 */ 36 37 /* 38 * Socket operations for use by nfs 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: nfs_socket.c,v 1.174 2008/11/19 18:36:09 ad Exp $"); 43 44 #ifdef _KERNEL_OPT 45 #include "fs_nfs.h" 46 #include "opt_nfs.h" 47 #include "opt_mbuftrace.h" 48 #endif 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/evcnt.h> 53 #include <sys/callout.h> 54 #include <sys/proc.h> 55 #include <sys/mount.h> 56 #include <sys/kernel.h> 57 #include <sys/kmem.h> 58 #include <sys/mbuf.h> 59 #include <sys/vnode.h> 60 #include <sys/domain.h> 61 #include <sys/protosw.h> 62 #include <sys/socket.h> 63 #include <sys/socketvar.h> 64 #include <sys/syslog.h> 65 #include <sys/tprintf.h> 66 #include <sys/namei.h> 67 #include <sys/signal.h> 68 #include <sys/signalvar.h> 69 #include <sys/kauth.h> 70 71 #include <netinet/in.h> 72 #include <netinet/tcp.h> 73 74 #include <nfs/rpcv2.h> 75 #include <nfs/nfsproto.h> 76 #include <nfs/nfs.h> 77 #include <nfs/xdr_subs.h> 78 #include <nfs/nfsm_subs.h> 79 #include <nfs/nfsmount.h> 80 #include <nfs/nfsnode.h> 81 #include <nfs/nfsrtt.h> 82 #include <nfs/nfs_var.h> 83 84 #ifdef MBUFTRACE 85 struct mowner nfs_mowner = MOWNER_INIT("nfs",""); 86 #endif 87 88 /* 89 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 90 * Use the mean and mean deviation of rtt for the appropriate type of rpc 91 * for the frequent rpcs and a default for the others. 92 * The justification for doing "other" this way is that these rpcs 93 * happen so infrequently that timer est. would probably be stale. 94 * Also, since many of these rpcs are 95 * non-idempotent, a conservative timeout is desired. 96 * getattr, lookup - A+2D 97 * read, write - A+4D 98 * other - nm_timeo 99 */ 100 #define NFS_RTO(n, t) \ 101 ((t) == 0 ? (n)->nm_timeo : \ 102 ((t) < 3 ? \ 103 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 104 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 105 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 106 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 107 /* 108 * External data, mostly RPC constants in XDR form 109 */ 110 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 111 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, 112 rpc_auth_kerb; 113 extern u_int32_t nfs_prog; 114 extern const int nfsv3_procid[NFS_NPROCS]; 115 extern int nfs_ticks; 116 117 /* 118 * Defines which timer to use for the procnum. 119 * 0 - default 120 * 1 - getattr 121 * 2 - lookup 122 * 3 - read 123 * 4 - write 124 */ 125 static const int proct[NFS_NPROCS] = { 126 [NFSPROC_NULL] = 0, 127 [NFSPROC_GETATTR] = 1, 128 [NFSPROC_SETATTR] = 0, 129 [NFSPROC_LOOKUP] = 2, 130 [NFSPROC_ACCESS] = 1, 131 [NFSPROC_READLINK] = 3, 132 [NFSPROC_READ] = 3, 133 [NFSPROC_WRITE] = 4, 134 [NFSPROC_CREATE] = 0, 135 [NFSPROC_MKDIR] = 0, 136 [NFSPROC_SYMLINK] = 0, 137 [NFSPROC_MKNOD] = 0, 138 [NFSPROC_REMOVE] = 0, 139 [NFSPROC_RMDIR] = 0, 140 [NFSPROC_RENAME] = 0, 141 [NFSPROC_LINK] = 0, 142 [NFSPROC_READDIR] = 3, 143 [NFSPROC_READDIRPLUS] = 3, 144 [NFSPROC_FSSTAT] = 0, 145 [NFSPROC_FSINFO] = 0, 146 [NFSPROC_PATHCONF] = 0, 147 [NFSPROC_COMMIT] = 0, 148 [NFSPROC_NOOP] = 0, 149 }; 150 151 /* 152 * There is a congestion window for outstanding rpcs maintained per mount 153 * point. The cwnd size is adjusted in roughly the way that: 154 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 155 * SIGCOMM '88". ACM, August 1988. 156 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 157 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 158 * of rpcs is in progress. 159 * (The sent count and cwnd are scaled for integer arith.) 160 * Variants of "slow start" were tried and were found to be too much of a 161 * performance hit (ave. rtt 3 times larger), 162 * I suspect due to the large rtt that nfs rpcs have. 163 */ 164 #define NFS_CWNDSCALE 256 165 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 166 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 167 int nfsrtton = 0; 168 struct nfsrtt nfsrtt; 169 struct nfsreqhead nfs_reqq; 170 static callout_t nfs_timer_ch; 171 static struct evcnt nfs_timer_ev; 172 static struct evcnt nfs_timer_start_ev; 173 static struct evcnt nfs_timer_stop_ev; 174 static kmutex_t nfs_timer_lock; 175 static bool (*nfs_timer_srvvec)(void); 176 177 #ifdef NFS 178 static int nfs_sndlock(struct nfsmount *, struct nfsreq *); 179 static void nfs_sndunlock(struct nfsmount *); 180 #endif 181 static int nfs_rcvlock(struct nfsmount *, struct nfsreq *); 182 static void nfs_rcvunlock(struct nfsmount *); 183 184 /* 185 * Initialize sockets and congestion for a new NFS connection. 186 * We do not free the sockaddr if error. 187 */ 188 int 189 nfs_connect(nmp, rep, l) 190 struct nfsmount *nmp; 191 struct nfsreq *rep; 192 struct lwp *l; 193 { 194 struct socket *so; 195 int error, rcvreserve, sndreserve; 196 struct sockaddr *saddr; 197 struct sockaddr_in *sin; 198 struct sockaddr_in6 *sin6; 199 struct mbuf *m; 200 int val; 201 202 nmp->nm_so = (struct socket *)0; 203 saddr = mtod(nmp->nm_nam, struct sockaddr *); 204 error = socreate(saddr->sa_family, &nmp->nm_so, 205 nmp->nm_sotype, nmp->nm_soproto, l, NULL); 206 if (error) 207 goto bad; 208 so = nmp->nm_so; 209 #ifdef MBUFTRACE 210 so->so_mowner = &nfs_mowner; 211 so->so_rcv.sb_mowner = &nfs_mowner; 212 so->so_snd.sb_mowner = &nfs_mowner; 213 #endif 214 nmp->nm_soflags = so->so_proto->pr_flags; 215 216 /* 217 * Some servers require that the client port be a reserved port number. 218 */ 219 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 220 val = IP_PORTRANGE_LOW; 221 222 if ((error = so_setsockopt(NULL, so, IPPROTO_IP, IP_PORTRANGE, 223 &val, sizeof(val)))) 224 goto bad; 225 m = m_get(M_WAIT, MT_SONAME); 226 MCLAIM(m, so->so_mowner); 227 sin = mtod(m, struct sockaddr_in *); 228 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 229 sin->sin_family = AF_INET; 230 sin->sin_addr.s_addr = INADDR_ANY; 231 sin->sin_port = 0; 232 error = sobind(so, m, &lwp0); 233 m_freem(m); 234 if (error) 235 goto bad; 236 } 237 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) { 238 val = IPV6_PORTRANGE_LOW; 239 240 if ((error = so_setsockopt(NULL, so, IPPROTO_IPV6, 241 IPV6_PORTRANGE, &val, sizeof(val)))) 242 goto bad; 243 m = m_get(M_WAIT, MT_SONAME); 244 MCLAIM(m, so->so_mowner); 245 sin6 = mtod(m, struct sockaddr_in6 *); 246 memset(sin6, 0, sizeof(*sin6)); 247 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6); 248 sin6->sin6_family = AF_INET6; 249 error = sobind(so, m, &lwp0); 250 m_freem(m); 251 if (error) 252 goto bad; 253 } 254 255 /* 256 * Protocols that do not require connections may be optionally left 257 * unconnected for servers that reply from a port other than NFS_PORT. 258 */ 259 solock(so); 260 if (nmp->nm_flag & NFSMNT_NOCONN) { 261 if (nmp->nm_soflags & PR_CONNREQUIRED) { 262 sounlock(so); 263 error = ENOTCONN; 264 goto bad; 265 } 266 } else { 267 error = soconnect(so, nmp->nm_nam, l); 268 if (error) { 269 sounlock(so); 270 goto bad; 271 } 272 273 /* 274 * Wait for the connection to complete. Cribbed from the 275 * connect system call but with the wait timing out so 276 * that interruptible mounts don't hang here for a long time. 277 */ 278 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 279 (void)sowait(so, 2 * hz); 280 if ((so->so_state & SS_ISCONNECTING) && 281 so->so_error == 0 && rep && 282 (error = nfs_sigintr(nmp, rep, rep->r_lwp)) != 0){ 283 so->so_state &= ~SS_ISCONNECTING; 284 sounlock(so); 285 goto bad; 286 } 287 } 288 if (so->so_error) { 289 error = so->so_error; 290 so->so_error = 0; 291 sounlock(so); 292 goto bad; 293 } 294 } 295 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 296 so->so_rcv.sb_timeo = (5 * hz); 297 so->so_snd.sb_timeo = (5 * hz); 298 } else { 299 /* 300 * enable receive timeout to detect server crash and reconnect. 301 * otherwise, we can be stuck in soreceive forever. 302 */ 303 so->so_rcv.sb_timeo = (5 * hz); 304 so->so_snd.sb_timeo = 0; 305 } 306 if (nmp->nm_sotype == SOCK_DGRAM) { 307 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 308 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 309 NFS_MAXPKTHDR) * 2; 310 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 311 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 312 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 313 NFS_MAXPKTHDR) * 2; 314 } else { 315 sounlock(so); 316 if (nmp->nm_sotype != SOCK_STREAM) 317 panic("nfscon sotype"); 318 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 319 val = 1; 320 so_setsockopt(NULL, so, SOL_SOCKET, SO_KEEPALIVE, &val, 321 sizeof(val)); 322 } 323 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 324 val = 1; 325 so_setsockopt(NULL, so, IPPROTO_TCP, TCP_NODELAY, &val, 326 sizeof(val)); 327 } 328 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 329 sizeof (u_int32_t)) * 2; 330 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 331 sizeof (u_int32_t)) * 2; 332 solock(so); 333 } 334 error = soreserve(so, sndreserve, rcvreserve); 335 if (error) { 336 sounlock(so); 337 goto bad; 338 } 339 so->so_rcv.sb_flags |= SB_NOINTR; 340 so->so_snd.sb_flags |= SB_NOINTR; 341 sounlock(so); 342 343 /* Initialize other non-zero congestion variables */ 344 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 345 NFS_TIMEO << 3; 346 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 347 nmp->nm_sdrtt[3] = 0; 348 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 349 nmp->nm_sent = 0; 350 nmp->nm_timeouts = 0; 351 return (0); 352 353 bad: 354 nfs_disconnect(nmp); 355 return (error); 356 } 357 358 /* 359 * Reconnect routine: 360 * Called when a connection is broken on a reliable protocol. 361 * - clean up the old socket 362 * - nfs_connect() again 363 * - set R_MUSTRESEND for all outstanding requests on mount point 364 * If this fails the mount point is DEAD! 365 * nb: Must be called with the nfs_sndlock() set on the mount point. 366 */ 367 int 368 nfs_reconnect(struct nfsreq *rep) 369 { 370 struct nfsreq *rp; 371 struct nfsmount *nmp = rep->r_nmp; 372 int error; 373 374 nfs_disconnect(nmp); 375 while ((error = nfs_connect(nmp, rep, &lwp0)) != 0) { 376 if (error == EINTR || error == ERESTART) 377 return (EINTR); 378 kpause("nfscn2", false, hz, NULL); 379 } 380 381 /* 382 * Loop through outstanding request list and fix up all requests 383 * on old socket. 384 */ 385 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 386 if (rp->r_nmp == nmp) { 387 if ((rp->r_flags & R_MUSTRESEND) == 0) 388 rp->r_flags |= R_MUSTRESEND | R_REXMITTED; 389 rp->r_rexmit = 0; 390 } 391 } 392 return (0); 393 } 394 395 /* 396 * NFS disconnect. Clean up and unlink. 397 */ 398 void 399 nfs_disconnect(nmp) 400 struct nfsmount *nmp; 401 { 402 struct socket *so; 403 int drain = 0; 404 405 if (nmp->nm_so) { 406 so = nmp->nm_so; 407 nmp->nm_so = (struct socket *)0; 408 solock(so); 409 soshutdown(so, SHUT_RDWR); 410 sounlock(so); 411 drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0; 412 if (drain) { 413 /* 414 * soshutdown() above should wake up the current 415 * listener. 416 * Now wake up those waiting for the receive lock, and 417 * wait for them to go away unhappy, to prevent *nmp 418 * from evaporating while they're sleeping. 419 */ 420 mutex_enter(&nmp->nm_lock); 421 while (nmp->nm_waiters > 0) { 422 cv_broadcast(&nmp->nm_rcvcv); 423 cv_broadcast(&nmp->nm_sndcv); 424 cv_wait(&nmp->nm_disconcv, &nmp->nm_lock); 425 } 426 mutex_exit(&nmp->nm_lock); 427 } 428 soclose(so); 429 } 430 #ifdef DIAGNOSTIC 431 if (drain && (nmp->nm_waiters > 0)) 432 panic("nfs_disconnect: waiters left after drain?"); 433 #endif 434 } 435 436 void 437 nfs_safedisconnect(nmp) 438 struct nfsmount *nmp; 439 { 440 struct nfsreq dummyreq; 441 442 memset(&dummyreq, 0, sizeof(dummyreq)); 443 dummyreq.r_nmp = nmp; 444 nfs_rcvlock(nmp, &dummyreq); /* XXX ignored error return */ 445 nfs_disconnect(nmp); 446 nfs_rcvunlock(nmp); 447 } 448 449 /* 450 * This is the nfs send routine. For connection based socket types, it 451 * must be called with an nfs_sndlock() on the socket. 452 * "rep == NULL" indicates that it has been called from a server. 453 * For the client side: 454 * - return EINTR if the RPC is terminated, 0 otherwise 455 * - set R_MUSTRESEND if the send fails for any reason 456 * - do any cleanup required by recoverable socket errors (? ? ?) 457 * For the server side: 458 * - return EINTR or ERESTART if interrupted by a signal 459 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 460 * - do any cleanup required by recoverable socket errors (? ? ?) 461 */ 462 int 463 nfs_send(so, nam, top, rep, l) 464 struct socket *so; 465 struct mbuf *nam; 466 struct mbuf *top; 467 struct nfsreq *rep; 468 struct lwp *l; 469 { 470 struct mbuf *sendnam; 471 int error, soflags, flags; 472 473 /* XXX nfs_doio()/nfs_request() calls with rep->r_lwp == NULL */ 474 if (l == NULL && rep->r_lwp == NULL) 475 l = curlwp; 476 477 if (rep) { 478 if (rep->r_flags & R_SOFTTERM) { 479 m_freem(top); 480 return (EINTR); 481 } 482 if ((so = rep->r_nmp->nm_so) == NULL) { 483 rep->r_flags |= R_MUSTRESEND; 484 m_freem(top); 485 return (0); 486 } 487 rep->r_flags &= ~R_MUSTRESEND; 488 soflags = rep->r_nmp->nm_soflags; 489 } else 490 soflags = so->so_proto->pr_flags; 491 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 492 sendnam = (struct mbuf *)0; 493 else 494 sendnam = nam; 495 if (so->so_type == SOCK_SEQPACKET) 496 flags = MSG_EOR; 497 else 498 flags = 0; 499 500 error = (*so->so_send)(so, sendnam, NULL, top, NULL, flags, l); 501 if (error) { 502 if (rep) { 503 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 504 /* 505 * We're too fast for the network/driver, 506 * and UDP isn't flowcontrolled. 507 * We need to resend. This is not fatal, 508 * just try again. 509 * 510 * Could be smarter here by doing some sort 511 * of a backoff, but this is rare. 512 */ 513 rep->r_flags |= R_MUSTRESEND; 514 } else { 515 if (error != EPIPE) 516 log(LOG_INFO, 517 "nfs send error %d for %s\n", 518 error, 519 rep->r_nmp->nm_mountp-> 520 mnt_stat.f_mntfromname); 521 /* 522 * Deal with errors for the client side. 523 */ 524 if (rep->r_flags & R_SOFTTERM) 525 error = EINTR; 526 else 527 rep->r_flags |= R_MUSTRESEND; 528 } 529 } else { 530 /* 531 * See above. This error can happen under normal 532 * circumstances and the log is too noisy. 533 * The error will still show up in nfsstat. 534 */ 535 if (error != ENOBUFS || so->so_type != SOCK_DGRAM) 536 log(LOG_INFO, "nfsd send error %d\n", error); 537 } 538 539 /* 540 * Handle any recoverable (soft) socket errors here. (? ? ?) 541 */ 542 if (error != EINTR && error != ERESTART && 543 error != EWOULDBLOCK && error != EPIPE) 544 error = 0; 545 } 546 return (error); 547 } 548 549 #ifdef NFS 550 /* 551 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 552 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 553 * Mark and consolidate the data into a new mbuf list. 554 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 555 * small mbufs. 556 * For SOCK_STREAM we must be very careful to read an entire record once 557 * we have read any of it, even if the system call has been interrupted. 558 */ 559 static int 560 nfs_receive(struct nfsreq *rep, struct mbuf **aname, struct mbuf **mp, 561 struct lwp *l) 562 { 563 struct socket *so; 564 struct uio auio; 565 struct iovec aio; 566 struct mbuf *m; 567 struct mbuf *control; 568 u_int32_t len; 569 struct mbuf **getnam; 570 int error, sotype, rcvflg; 571 572 /* 573 * Set up arguments for soreceive() 574 */ 575 *mp = (struct mbuf *)0; 576 *aname = (struct mbuf *)0; 577 sotype = rep->r_nmp->nm_sotype; 578 579 /* 580 * For reliable protocols, lock against other senders/receivers 581 * in case a reconnect is necessary. 582 * For SOCK_STREAM, first get the Record Mark to find out how much 583 * more there is to get. 584 * We must lock the socket against other receivers 585 * until we have an entire rpc request/reply. 586 */ 587 if (sotype != SOCK_DGRAM) { 588 error = nfs_sndlock(rep->r_nmp, rep); 589 if (error) 590 return (error); 591 tryagain: 592 /* 593 * Check for fatal errors and resending request. 594 */ 595 /* 596 * Ugh: If a reconnect attempt just happened, nm_so 597 * would have changed. NULL indicates a failed 598 * attempt that has essentially shut down this 599 * mount point. 600 */ 601 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 602 nfs_sndunlock(rep->r_nmp); 603 return (EINTR); 604 } 605 so = rep->r_nmp->nm_so; 606 if (!so) { 607 error = nfs_reconnect(rep); 608 if (error) { 609 nfs_sndunlock(rep->r_nmp); 610 return (error); 611 } 612 goto tryagain; 613 } 614 while (rep->r_flags & R_MUSTRESEND) { 615 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 616 nfsstats.rpcretries++; 617 rep->r_rtt = 0; 618 rep->r_flags &= ~R_TIMING; 619 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, l); 620 if (error) { 621 if (error == EINTR || error == ERESTART || 622 (error = nfs_reconnect(rep)) != 0) { 623 nfs_sndunlock(rep->r_nmp); 624 return (error); 625 } 626 goto tryagain; 627 } 628 } 629 nfs_sndunlock(rep->r_nmp); 630 if (sotype == SOCK_STREAM) { 631 aio.iov_base = (void *) &len; 632 aio.iov_len = sizeof(u_int32_t); 633 auio.uio_iov = &aio; 634 auio.uio_iovcnt = 1; 635 auio.uio_rw = UIO_READ; 636 auio.uio_offset = 0; 637 auio.uio_resid = sizeof(u_int32_t); 638 UIO_SETUP_SYSSPACE(&auio); 639 do { 640 rcvflg = MSG_WAITALL; 641 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 642 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 643 if (error == EWOULDBLOCK && rep) { 644 if (rep->r_flags & R_SOFTTERM) 645 return (EINTR); 646 /* 647 * if it seems that the server died after it 648 * received our request, set EPIPE so that 649 * we'll reconnect and retransmit requests. 650 */ 651 if (rep->r_rexmit >= rep->r_nmp->nm_retry) { 652 nfsstats.rpctimeouts++; 653 error = EPIPE; 654 } 655 } 656 } while (error == EWOULDBLOCK); 657 if (!error && auio.uio_resid > 0) { 658 /* 659 * Don't log a 0 byte receive; it means 660 * that the socket has been closed, and 661 * can happen during normal operation 662 * (forcible unmount or Solaris server). 663 */ 664 if (auio.uio_resid != sizeof (u_int32_t)) 665 log(LOG_INFO, 666 "short receive (%lu/%lu) from nfs server %s\n", 667 (u_long)sizeof(u_int32_t) - auio.uio_resid, 668 (u_long)sizeof(u_int32_t), 669 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 670 error = EPIPE; 671 } 672 if (error) 673 goto errout; 674 len = ntohl(len) & ~0x80000000; 675 /* 676 * This is SERIOUS! We are out of sync with the sender 677 * and forcing a disconnect/reconnect is all I can do. 678 */ 679 if (len > NFS_MAXPACKET) { 680 log(LOG_ERR, "%s (%d) from nfs server %s\n", 681 "impossible packet length", 682 len, 683 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 684 error = EFBIG; 685 goto errout; 686 } 687 auio.uio_resid = len; 688 do { 689 rcvflg = MSG_WAITALL; 690 error = (*so->so_receive)(so, (struct mbuf **)0, 691 &auio, mp, (struct mbuf **)0, &rcvflg); 692 } while (error == EWOULDBLOCK || error == EINTR || 693 error == ERESTART); 694 if (!error && auio.uio_resid > 0) { 695 if (len != auio.uio_resid) 696 log(LOG_INFO, 697 "short receive (%lu/%d) from nfs server %s\n", 698 (u_long)len - auio.uio_resid, len, 699 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 700 error = EPIPE; 701 } 702 } else { 703 /* 704 * NB: Since uio_resid is big, MSG_WAITALL is ignored 705 * and soreceive() will return when it has either a 706 * control msg or a data msg. 707 * We have no use for control msg., but must grab them 708 * and then throw them away so we know what is going 709 * on. 710 */ 711 auio.uio_resid = len = 100000000; /* Anything Big */ 712 /* not need to setup uio_vmspace */ 713 do { 714 rcvflg = 0; 715 error = (*so->so_receive)(so, (struct mbuf **)0, 716 &auio, mp, &control, &rcvflg); 717 if (control) 718 m_freem(control); 719 if (error == EWOULDBLOCK && rep) { 720 if (rep->r_flags & R_SOFTTERM) 721 return (EINTR); 722 } 723 } while (error == EWOULDBLOCK || 724 (!error && *mp == NULL && control)); 725 if ((rcvflg & MSG_EOR) == 0) 726 printf("Egad!!\n"); 727 if (!error && *mp == NULL) 728 error = EPIPE; 729 len -= auio.uio_resid; 730 } 731 errout: 732 if (error && error != EINTR && error != ERESTART) { 733 m_freem(*mp); 734 *mp = (struct mbuf *)0; 735 if (error != EPIPE) 736 log(LOG_INFO, 737 "receive error %d from nfs server %s\n", 738 error, 739 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 740 error = nfs_sndlock(rep->r_nmp, rep); 741 if (!error) 742 error = nfs_reconnect(rep); 743 if (!error) 744 goto tryagain; 745 else 746 nfs_sndunlock(rep->r_nmp); 747 } 748 } else { 749 if ((so = rep->r_nmp->nm_so) == NULL) 750 return (EACCES); 751 if (so->so_state & SS_ISCONNECTED) 752 getnam = (struct mbuf **)0; 753 else 754 getnam = aname; 755 auio.uio_resid = len = 1000000; 756 /* not need to setup uio_vmspace */ 757 do { 758 rcvflg = 0; 759 error = (*so->so_receive)(so, getnam, &auio, mp, 760 (struct mbuf **)0, &rcvflg); 761 if (error == EWOULDBLOCK && 762 (rep->r_flags & R_SOFTTERM)) 763 return (EINTR); 764 } while (error == EWOULDBLOCK); 765 len -= auio.uio_resid; 766 if (!error && *mp == NULL) 767 error = EPIPE; 768 } 769 if (error) { 770 m_freem(*mp); 771 *mp = (struct mbuf *)0; 772 } 773 return (error); 774 } 775 776 /* 777 * Implement receipt of reply on a socket. 778 * We must search through the list of received datagrams matching them 779 * with outstanding requests using the xid, until ours is found. 780 */ 781 /* ARGSUSED */ 782 static int 783 nfs_reply(struct nfsreq *myrep, struct lwp *lwp) 784 { 785 struct nfsreq *rep; 786 struct nfsmount *nmp = myrep->r_nmp; 787 int32_t t1; 788 struct mbuf *mrep, *nam, *md; 789 u_int32_t rxid, *tl; 790 char *dpos, *cp2; 791 int error; 792 793 /* 794 * Loop around until we get our own reply 795 */ 796 for (;;) { 797 /* 798 * Lock against other receivers so that I don't get stuck in 799 * sbwait() after someone else has received my reply for me. 800 * Also necessary for connection based protocols to avoid 801 * race conditions during a reconnect. 802 */ 803 error = nfs_rcvlock(nmp, myrep); 804 if (error == EALREADY) 805 return (0); 806 if (error) 807 return (error); 808 /* 809 * Get the next Rpc reply off the socket 810 */ 811 812 mutex_enter(&nmp->nm_lock); 813 nmp->nm_waiters++; 814 mutex_exit(&nmp->nm_lock); 815 816 error = nfs_receive(myrep, &nam, &mrep, lwp); 817 818 mutex_enter(&nmp->nm_lock); 819 nmp->nm_waiters--; 820 cv_signal(&nmp->nm_disconcv); 821 mutex_exit(&nmp->nm_lock); 822 823 if (error) { 824 nfs_rcvunlock(nmp); 825 826 if (nmp->nm_iflag & NFSMNT_DISMNT) { 827 /* 828 * Oops, we're going away now.. 829 */ 830 return error; 831 } 832 /* 833 * Ignore routing errors on connectionless protocols? ? 834 */ 835 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 836 nmp->nm_so->so_error = 0; 837 #ifdef DEBUG 838 printf("nfs_reply: ignoring error %d\n", error); 839 #endif 840 continue; 841 } 842 return (error); 843 } 844 if (nam) 845 m_freem(nam); 846 847 /* 848 * Get the xid and check that it is an rpc reply 849 */ 850 md = mrep; 851 dpos = mtod(md, void *); 852 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 853 rxid = *tl++; 854 if (*tl != rpc_reply) { 855 nfsstats.rpcinvalid++; 856 m_freem(mrep); 857 nfsmout: 858 nfs_rcvunlock(nmp); 859 continue; 860 } 861 862 /* 863 * Loop through the request list to match up the reply 864 * Iff no match, just drop the datagram 865 */ 866 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 867 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 868 /* Found it.. */ 869 rep->r_mrep = mrep; 870 rep->r_md = md; 871 rep->r_dpos = dpos; 872 if (nfsrtton) { 873 struct rttl *rt; 874 875 rt = &nfsrtt.rttl[nfsrtt.pos]; 876 rt->proc = rep->r_procnum; 877 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 878 rt->sent = nmp->nm_sent; 879 rt->cwnd = nmp->nm_cwnd; 880 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 881 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 882 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx; 883 getmicrotime(&rt->tstamp); 884 if (rep->r_flags & R_TIMING) 885 rt->rtt = rep->r_rtt; 886 else 887 rt->rtt = 1000000; 888 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 889 } 890 /* 891 * Update congestion window. 892 * Do the additive increase of 893 * one rpc/rtt. 894 */ 895 if (nmp->nm_cwnd <= nmp->nm_sent) { 896 nmp->nm_cwnd += 897 (NFS_CWNDSCALE * NFS_CWNDSCALE + 898 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 899 if (nmp->nm_cwnd > NFS_MAXCWND) 900 nmp->nm_cwnd = NFS_MAXCWND; 901 } 902 rep->r_flags &= ~R_SENT; 903 nmp->nm_sent -= NFS_CWNDSCALE; 904 /* 905 * Update rtt using a gain of 0.125 on the mean 906 * and a gain of 0.25 on the deviation. 907 */ 908 if (rep->r_flags & R_TIMING) { 909 /* 910 * Since the timer resolution of 911 * NFS_HZ is so course, it can often 912 * result in r_rtt == 0. Since 913 * r_rtt == N means that the actual 914 * rtt is between N+dt and N+2-dt ticks, 915 * add 1. 916 */ 917 t1 = rep->r_rtt + 1; 918 t1 -= (NFS_SRTT(rep) >> 3); 919 NFS_SRTT(rep) += t1; 920 if (t1 < 0) 921 t1 = -t1; 922 t1 -= (NFS_SDRTT(rep) >> 2); 923 NFS_SDRTT(rep) += t1; 924 } 925 nmp->nm_timeouts = 0; 926 break; 927 } 928 } 929 nfs_rcvunlock(nmp); 930 /* 931 * If not matched to a request, drop it. 932 * If it's mine, get out. 933 */ 934 if (rep == 0) { 935 nfsstats.rpcunexpected++; 936 m_freem(mrep); 937 } else if (rep == myrep) { 938 if (rep->r_mrep == NULL) 939 panic("nfsreply nil"); 940 return (0); 941 } 942 } 943 } 944 945 /* 946 * nfs_request - goes something like this 947 * - fill in request struct 948 * - links it into list 949 * - calls nfs_send() for first transmit 950 * - calls nfs_receive() to get reply 951 * - break down rpc header and return with nfs reply pointed to 952 * by mrep or error 953 * nb: always frees up mreq mbuf list 954 */ 955 int 956 nfs_request(np, mrest, procnum, lwp, cred, mrp, mdp, dposp, rexmitp) 957 struct nfsnode *np; 958 struct mbuf *mrest; 959 int procnum; 960 struct lwp *lwp; 961 kauth_cred_t cred; 962 struct mbuf **mrp; 963 struct mbuf **mdp; 964 char **dposp; 965 int *rexmitp; 966 { 967 struct mbuf *m, *mrep; 968 struct nfsreq *rep; 969 u_int32_t *tl; 970 int i; 971 struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount); 972 struct mbuf *md, *mheadend; 973 char nickv[RPCX_NICKVERF]; 974 time_t waituntil; 975 char *dpos, *cp2; 976 int t1, s, error = 0, mrest_len, auth_len, auth_type; 977 int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0; 978 int verf_len, verf_type; 979 u_int32_t xid; 980 char *auth_str, *verf_str; 981 NFSKERBKEY_T key; /* save session key */ 982 kauth_cred_t acred; 983 struct mbuf *mrest_backup = NULL; 984 kauth_cred_t origcred = NULL; /* XXX: gcc */ 985 bool retry_cred = true; 986 bool use_opencred = (np->n_flag & NUSEOPENCRED) != 0; 987 988 if (rexmitp != NULL) 989 *rexmitp = 0; 990 991 acred = kauth_cred_alloc(); 992 993 tryagain_cred: 994 KASSERT(cred != NULL); 995 rep = kmem_alloc(sizeof(*rep), KM_SLEEP); 996 rep->r_nmp = nmp; 997 KASSERT(lwp == NULL || lwp == curlwp); 998 rep->r_lwp = lwp; 999 rep->r_procnum = procnum; 1000 i = 0; 1001 m = mrest; 1002 while (m) { 1003 i += m->m_len; 1004 m = m->m_next; 1005 } 1006 mrest_len = i; 1007 1008 /* 1009 * Get the RPC header with authorization. 1010 */ 1011 kerbauth: 1012 verf_str = auth_str = (char *)0; 1013 if (nmp->nm_flag & NFSMNT_KERB) { 1014 verf_str = nickv; 1015 verf_len = sizeof (nickv); 1016 auth_type = RPCAUTH_KERB4; 1017 memset((void *)key, 0, sizeof (key)); 1018 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 1019 &auth_len, verf_str, verf_len)) { 1020 error = nfs_getauth(nmp, rep, cred, &auth_str, 1021 &auth_len, verf_str, &verf_len, key); 1022 if (error) { 1023 kmem_free(rep, sizeof(*rep)); 1024 m_freem(mrest); 1025 KASSERT(kauth_cred_getrefcnt(acred) == 1); 1026 kauth_cred_free(acred); 1027 return (error); 1028 } 1029 } 1030 retry_cred = false; 1031 } else { 1032 /* AUTH_UNIX */ 1033 uid_t uid; 1034 gid_t gid; 1035 1036 /* 1037 * on the most unix filesystems, permission checks are 1038 * done when the file is open(2)'ed. 1039 * ie. once a file is successfully open'ed, 1040 * following i/o operations never fail with EACCES. 1041 * we try to follow the semantics as far as possible. 1042 * 1043 * note that we expect that the nfs server always grant 1044 * accesses by the file's owner. 1045 */ 1046 origcred = cred; 1047 switch (procnum) { 1048 case NFSPROC_READ: 1049 case NFSPROC_WRITE: 1050 case NFSPROC_COMMIT: 1051 uid = np->n_vattr->va_uid; 1052 gid = np->n_vattr->va_gid; 1053 if (kauth_cred_geteuid(cred) == uid && 1054 kauth_cred_getegid(cred) == gid) { 1055 retry_cred = false; 1056 break; 1057 } 1058 if (use_opencred) 1059 break; 1060 kauth_cred_setuid(acred, uid); 1061 kauth_cred_seteuid(acred, uid); 1062 kauth_cred_setsvuid(acred, uid); 1063 kauth_cred_setgid(acred, gid); 1064 kauth_cred_setegid(acred, gid); 1065 kauth_cred_setsvgid(acred, gid); 1066 cred = acred; 1067 break; 1068 default: 1069 retry_cred = false; 1070 break; 1071 } 1072 /* 1073 * backup mbuf chain if we can need it later to retry. 1074 * 1075 * XXX maybe we can keep a direct reference to 1076 * mrest without doing m_copym, but it's ...ugly. 1077 */ 1078 if (retry_cred) 1079 mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT); 1080 auth_type = RPCAUTH_UNIX; 1081 /* XXX elad - ngroups */ 1082 auth_len = (((kauth_cred_ngroups(cred) > nmp->nm_numgrps) ? 1083 nmp->nm_numgrps : kauth_cred_ngroups(cred)) << 2) + 1084 5 * NFSX_UNSIGNED; 1085 } 1086 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 1087 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 1088 if (auth_str) 1089 free(auth_str, M_TEMP); 1090 1091 /* 1092 * For stream protocols, insert a Sun RPC Record Mark. 1093 */ 1094 if (nmp->nm_sotype == SOCK_STREAM) { 1095 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 1096 *mtod(m, u_int32_t *) = htonl(0x80000000 | 1097 (m->m_pkthdr.len - NFSX_UNSIGNED)); 1098 } 1099 rep->r_mreq = m; 1100 rep->r_xid = xid; 1101 tryagain: 1102 if (nmp->nm_flag & NFSMNT_SOFT) 1103 rep->r_retry = nmp->nm_retry; 1104 else 1105 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1106 rep->r_rtt = rep->r_rexmit = 0; 1107 if (proct[procnum] > 0) 1108 rep->r_flags = R_TIMING; 1109 else 1110 rep->r_flags = 0; 1111 rep->r_mrep = NULL; 1112 1113 /* 1114 * Do the client side RPC. 1115 */ 1116 nfsstats.rpcrequests++; 1117 /* 1118 * Chain request into list of outstanding requests. Be sure 1119 * to put it LAST so timer finds oldest requests first. 1120 */ 1121 s = splsoftnet(); 1122 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1123 nfs_timer_start(); 1124 1125 /* 1126 * If backing off another request or avoiding congestion, don't 1127 * send this one now but let timer do it. If not timing a request, 1128 * do it now. 1129 */ 1130 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1131 (nmp->nm_flag & NFSMNT_DUMBTIMR) || nmp->nm_sent < nmp->nm_cwnd)) { 1132 splx(s); 1133 if (nmp->nm_soflags & PR_CONNREQUIRED) 1134 error = nfs_sndlock(nmp, rep); 1135 if (!error) { 1136 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1137 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp); 1138 if (nmp->nm_soflags & PR_CONNREQUIRED) 1139 nfs_sndunlock(nmp); 1140 } 1141 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1142 nmp->nm_sent += NFS_CWNDSCALE; 1143 rep->r_flags |= R_SENT; 1144 } 1145 } else { 1146 splx(s); 1147 rep->r_rtt = -1; 1148 } 1149 1150 /* 1151 * Wait for the reply from our send or the timer's. 1152 */ 1153 if (!error || error == EPIPE) 1154 error = nfs_reply(rep, lwp); 1155 1156 /* 1157 * RPC done, unlink the request. 1158 */ 1159 s = splsoftnet(); 1160 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1161 splx(s); 1162 1163 /* 1164 * Decrement the outstanding request count. 1165 */ 1166 if (rep->r_flags & R_SENT) { 1167 rep->r_flags &= ~R_SENT; /* paranoia */ 1168 nmp->nm_sent -= NFS_CWNDSCALE; 1169 } 1170 1171 if (rexmitp != NULL) { 1172 int rexmit; 1173 1174 if (nmp->nm_sotype != SOCK_DGRAM) 1175 rexmit = (rep->r_flags & R_REXMITTED) != 0; 1176 else 1177 rexmit = rep->r_rexmit; 1178 *rexmitp = rexmit; 1179 } 1180 1181 /* 1182 * If there was a successful reply and a tprintf msg. 1183 * tprintf a response. 1184 */ 1185 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1186 nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1187 "is alive again"); 1188 mrep = rep->r_mrep; 1189 md = rep->r_md; 1190 dpos = rep->r_dpos; 1191 if (error) 1192 goto nfsmout; 1193 1194 /* 1195 * break down the rpc header and check if ok 1196 */ 1197 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1198 if (*tl++ == rpc_msgdenied) { 1199 if (*tl == rpc_mismatch) 1200 error = EOPNOTSUPP; 1201 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1202 if (!failed_auth) { 1203 failed_auth++; 1204 mheadend->m_next = (struct mbuf *)0; 1205 m_freem(mrep); 1206 m_freem(rep->r_mreq); 1207 goto kerbauth; 1208 } else 1209 error = EAUTH; 1210 } else 1211 error = EACCES; 1212 m_freem(mrep); 1213 goto nfsmout; 1214 } 1215 1216 /* 1217 * Grab any Kerberos verifier, otherwise just throw it away. 1218 */ 1219 verf_type = fxdr_unsigned(int, *tl++); 1220 i = fxdr_unsigned(int32_t, *tl); 1221 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1222 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1223 if (error) 1224 goto nfsmout; 1225 } else if (i > 0) 1226 nfsm_adv(nfsm_rndup(i)); 1227 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1228 /* 0 == ok */ 1229 if (*tl == 0) { 1230 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1231 if (*tl != 0) { 1232 error = fxdr_unsigned(int, *tl); 1233 switch (error) { 1234 case NFSERR_PERM: 1235 error = EPERM; 1236 break; 1237 1238 case NFSERR_NOENT: 1239 error = ENOENT; 1240 break; 1241 1242 case NFSERR_IO: 1243 error = EIO; 1244 break; 1245 1246 case NFSERR_NXIO: 1247 error = ENXIO; 1248 break; 1249 1250 case NFSERR_ACCES: 1251 error = EACCES; 1252 if (!retry_cred) 1253 break; 1254 m_freem(mrep); 1255 m_freem(rep->r_mreq); 1256 kmem_free(rep, sizeof(*rep)); 1257 use_opencred = !use_opencred; 1258 if (mrest_backup == NULL) { 1259 /* m_copym failure */ 1260 KASSERT( 1261 kauth_cred_getrefcnt(acred) == 1); 1262 kauth_cred_free(acred); 1263 return ENOMEM; 1264 } 1265 mrest = mrest_backup; 1266 mrest_backup = NULL; 1267 cred = origcred; 1268 error = 0; 1269 retry_cred = false; 1270 goto tryagain_cred; 1271 1272 case NFSERR_EXIST: 1273 error = EEXIST; 1274 break; 1275 1276 case NFSERR_XDEV: 1277 error = EXDEV; 1278 break; 1279 1280 case NFSERR_NODEV: 1281 error = ENODEV; 1282 break; 1283 1284 case NFSERR_NOTDIR: 1285 error = ENOTDIR; 1286 break; 1287 1288 case NFSERR_ISDIR: 1289 error = EISDIR; 1290 break; 1291 1292 case NFSERR_INVAL: 1293 error = EINVAL; 1294 break; 1295 1296 case NFSERR_FBIG: 1297 error = EFBIG; 1298 break; 1299 1300 case NFSERR_NOSPC: 1301 error = ENOSPC; 1302 break; 1303 1304 case NFSERR_ROFS: 1305 error = EROFS; 1306 break; 1307 1308 case NFSERR_MLINK: 1309 error = EMLINK; 1310 break; 1311 1312 case NFSERR_TIMEDOUT: 1313 error = ETIMEDOUT; 1314 break; 1315 1316 case NFSERR_NAMETOL: 1317 error = ENAMETOOLONG; 1318 break; 1319 1320 case NFSERR_NOTEMPTY: 1321 error = ENOTEMPTY; 1322 break; 1323 1324 case NFSERR_DQUOT: 1325 error = EDQUOT; 1326 break; 1327 1328 case NFSERR_STALE: 1329 /* 1330 * If the File Handle was stale, invalidate the 1331 * lookup cache, just in case. 1332 */ 1333 error = ESTALE; 1334 cache_purge(NFSTOV(np)); 1335 break; 1336 1337 case NFSERR_REMOTE: 1338 error = EREMOTE; 1339 break; 1340 1341 case NFSERR_WFLUSH: 1342 case NFSERR_BADHANDLE: 1343 case NFSERR_NOT_SYNC: 1344 case NFSERR_BAD_COOKIE: 1345 error = EINVAL; 1346 break; 1347 1348 case NFSERR_NOTSUPP: 1349 error = ENOTSUP; 1350 break; 1351 1352 case NFSERR_TOOSMALL: 1353 case NFSERR_SERVERFAULT: 1354 case NFSERR_BADTYPE: 1355 error = EINVAL; 1356 break; 1357 1358 case NFSERR_TRYLATER: 1359 if ((nmp->nm_flag & NFSMNT_NFSV3) == 0) 1360 break; 1361 m_freem(mrep); 1362 error = 0; 1363 waituntil = time_second + trylater_delay; 1364 while (time_second < waituntil) { 1365 kpause("nfstrylater", false, hz, NULL); 1366 } 1367 trylater_delay *= NFS_TRYLATERDELMUL; 1368 if (trylater_delay > NFS_TRYLATERDELMAX) 1369 trylater_delay = NFS_TRYLATERDELMAX; 1370 /* 1371 * RFC1813: 1372 * The client should wait and then try 1373 * the request with a new RPC transaction ID. 1374 */ 1375 nfs_renewxid(rep); 1376 goto tryagain; 1377 1378 default: 1379 #ifdef DIAGNOSTIC 1380 printf("Invalid rpc error code %d\n", error); 1381 #endif 1382 error = EINVAL; 1383 break; 1384 } 1385 1386 if (nmp->nm_flag & NFSMNT_NFSV3) { 1387 *mrp = mrep; 1388 *mdp = md; 1389 *dposp = dpos; 1390 error |= NFSERR_RETERR; 1391 } else 1392 m_freem(mrep); 1393 goto nfsmout; 1394 } 1395 1396 /* 1397 * note which credential worked to minimize number of retries. 1398 */ 1399 if (use_opencred) 1400 np->n_flag |= NUSEOPENCRED; 1401 else 1402 np->n_flag &= ~NUSEOPENCRED; 1403 1404 *mrp = mrep; 1405 *mdp = md; 1406 *dposp = dpos; 1407 1408 KASSERT(error == 0); 1409 goto nfsmout; 1410 } 1411 m_freem(mrep); 1412 error = EPROTONOSUPPORT; 1413 nfsmout: 1414 KASSERT(kauth_cred_getrefcnt(acred) == 1); 1415 kauth_cred_free(acred); 1416 m_freem(rep->r_mreq); 1417 kmem_free(rep, sizeof(*rep)); 1418 m_freem(mrest_backup); 1419 return (error); 1420 } 1421 #endif /* NFS */ 1422 1423 /* 1424 * Generate the rpc reply header 1425 * siz arg. is used to decide if adding a cluster is worthwhile 1426 */ 1427 int 1428 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1429 int siz; 1430 struct nfsrv_descript *nd; 1431 struct nfssvc_sock *slp; 1432 int err; 1433 int cache; 1434 u_quad_t *frev; 1435 struct mbuf **mrq; 1436 struct mbuf **mbp; 1437 char **bposp; 1438 { 1439 u_int32_t *tl; 1440 struct mbuf *mreq; 1441 char *bpos; 1442 struct mbuf *mb; 1443 1444 mreq = m_gethdr(M_WAIT, MT_DATA); 1445 MCLAIM(mreq, &nfs_mowner); 1446 mb = mreq; 1447 /* 1448 * If this is a big reply, use a cluster else 1449 * try and leave leading space for the lower level headers. 1450 */ 1451 siz += RPC_REPLYSIZ; 1452 if (siz >= max_datalen) { 1453 m_clget(mreq, M_WAIT); 1454 } else 1455 mreq->m_data += max_hdr; 1456 tl = mtod(mreq, u_int32_t *); 1457 mreq->m_len = 6 * NFSX_UNSIGNED; 1458 bpos = ((char *)tl) + mreq->m_len; 1459 *tl++ = txdr_unsigned(nd->nd_retxid); 1460 *tl++ = rpc_reply; 1461 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1462 *tl++ = rpc_msgdenied; 1463 if (err & NFSERR_AUTHERR) { 1464 *tl++ = rpc_autherr; 1465 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1466 mreq->m_len -= NFSX_UNSIGNED; 1467 bpos -= NFSX_UNSIGNED; 1468 } else { 1469 *tl++ = rpc_mismatch; 1470 *tl++ = txdr_unsigned(RPC_VER2); 1471 *tl = txdr_unsigned(RPC_VER2); 1472 } 1473 } else { 1474 *tl++ = rpc_msgaccepted; 1475 1476 /* 1477 * For Kerberos authentication, we must send the nickname 1478 * verifier back, otherwise just RPCAUTH_NULL. 1479 */ 1480 if (nd->nd_flag & ND_KERBFULL) { 1481 struct nfsuid *nuidp; 1482 struct timeval ktvin, ktvout; 1483 1484 memset(&ktvout, 0, sizeof ktvout); /* XXX gcc */ 1485 1486 LIST_FOREACH(nuidp, 1487 NUIDHASH(slp, kauth_cred_geteuid(nd->nd_cr)), 1488 nu_hash) { 1489 if (kauth_cred_geteuid(nuidp->nu_cr) == 1490 kauth_cred_geteuid(nd->nd_cr) && 1491 (!nd->nd_nam2 || netaddr_match( 1492 NU_NETFAM(nuidp), &nuidp->nu_haddr, 1493 nd->nd_nam2))) 1494 break; 1495 } 1496 if (nuidp) { 1497 ktvin.tv_sec = 1498 txdr_unsigned(nuidp->nu_timestamp.tv_sec 1499 - 1); 1500 ktvin.tv_usec = 1501 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1502 1503 /* 1504 * Encrypt the timestamp in ecb mode using the 1505 * session key. 1506 */ 1507 #ifdef NFSKERB 1508 XXX 1509 #endif 1510 1511 *tl++ = rpc_auth_kerb; 1512 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1513 *tl = ktvout.tv_sec; 1514 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1515 *tl++ = ktvout.tv_usec; 1516 *tl++ = txdr_unsigned( 1517 kauth_cred_geteuid(nuidp->nu_cr)); 1518 } else { 1519 *tl++ = 0; 1520 *tl++ = 0; 1521 } 1522 } else { 1523 *tl++ = 0; 1524 *tl++ = 0; 1525 } 1526 switch (err) { 1527 case EPROGUNAVAIL: 1528 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1529 break; 1530 case EPROGMISMATCH: 1531 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1532 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1533 *tl++ = txdr_unsigned(2); 1534 *tl = txdr_unsigned(3); 1535 break; 1536 case EPROCUNAVAIL: 1537 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1538 break; 1539 case EBADRPC: 1540 *tl = txdr_unsigned(RPC_GARBAGE); 1541 break; 1542 default: 1543 *tl = 0; 1544 if (err != NFSERR_RETVOID) { 1545 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1546 if (err) 1547 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1548 else 1549 *tl = 0; 1550 } 1551 break; 1552 }; 1553 } 1554 1555 if (mrq != NULL) 1556 *mrq = mreq; 1557 *mbp = mb; 1558 *bposp = bpos; 1559 if (err != 0 && err != NFSERR_RETVOID) 1560 nfsstats.srvrpc_errs++; 1561 return (0); 1562 } 1563 1564 static void 1565 nfs_timer_schedule(void) 1566 { 1567 1568 callout_schedule(&nfs_timer_ch, nfs_ticks); 1569 } 1570 1571 void 1572 nfs_timer_start(void) 1573 { 1574 1575 if (callout_pending(&nfs_timer_ch)) 1576 return; 1577 1578 nfs_timer_start_ev.ev_count++; 1579 nfs_timer_schedule(); 1580 } 1581 1582 void 1583 nfs_timer_init(void) 1584 { 1585 1586 mutex_init(&nfs_timer_lock, MUTEX_DEFAULT, IPL_NONE); 1587 callout_init(&nfs_timer_ch, 0); 1588 callout_setfunc(&nfs_timer_ch, nfs_timer, NULL); 1589 evcnt_attach_dynamic(&nfs_timer_ev, EVCNT_TYPE_MISC, NULL, 1590 "nfs", "timer"); 1591 evcnt_attach_dynamic(&nfs_timer_start_ev, EVCNT_TYPE_MISC, NULL, 1592 "nfs", "timer start"); 1593 evcnt_attach_dynamic(&nfs_timer_stop_ev, EVCNT_TYPE_MISC, NULL, 1594 "nfs", "timer stop"); 1595 } 1596 1597 void 1598 nfs_timer_fini(void) 1599 { 1600 1601 callout_halt(&nfs_timer_ch, NULL); 1602 callout_destroy(&nfs_timer_ch); 1603 mutex_destroy(&nfs_timer_lock); 1604 evcnt_detach(&nfs_timer_ev); 1605 evcnt_detach(&nfs_timer_start_ev); 1606 evcnt_detach(&nfs_timer_stop_ev); 1607 } 1608 1609 void 1610 nfs_timer_srvinit(bool (*func)(void)) 1611 { 1612 1613 nfs_timer_srvvec = func; 1614 } 1615 1616 void 1617 nfs_timer_srvfini(void) 1618 { 1619 1620 mutex_enter(&nfs_timer_lock); 1621 nfs_timer_srvvec = NULL; 1622 mutex_exit(&nfs_timer_lock); 1623 } 1624 1625 1626 /* 1627 * Nfs timer routine 1628 * Scan the nfsreq list and retranmit any requests that have timed out 1629 * To avoid retransmission attempts on STREAM sockets (in the future) make 1630 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1631 */ 1632 void 1633 nfs_timer(void *arg) 1634 { 1635 struct nfsreq *rep; 1636 struct mbuf *m; 1637 struct socket *so; 1638 struct nfsmount *nmp; 1639 int timeo; 1640 int s, error; 1641 bool more = false; 1642 1643 nfs_timer_ev.ev_count++; 1644 1645 s = splsoftnet(); 1646 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1647 more = true; 1648 nmp = rep->r_nmp; 1649 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1650 continue; 1651 if (nfs_sigintr(nmp, rep, rep->r_lwp)) { 1652 rep->r_flags |= R_SOFTTERM; 1653 continue; 1654 } 1655 if (rep->r_rtt >= 0) { 1656 rep->r_rtt++; 1657 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1658 timeo = nmp->nm_timeo; 1659 else 1660 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1661 if (nmp->nm_timeouts > 0) 1662 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1663 if (rep->r_rtt <= timeo) 1664 continue; 1665 if (nmp->nm_timeouts < 1666 (sizeof(nfs_backoff) / sizeof(nfs_backoff[0]))) 1667 nmp->nm_timeouts++; 1668 } 1669 /* 1670 * Check for server not responding 1671 */ 1672 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1673 rep->r_rexmit > nmp->nm_deadthresh) { 1674 nfs_msg(rep->r_lwp, 1675 nmp->nm_mountp->mnt_stat.f_mntfromname, 1676 "not responding"); 1677 rep->r_flags |= R_TPRINTFMSG; 1678 } 1679 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1680 nfsstats.rpctimeouts++; 1681 rep->r_flags |= R_SOFTTERM; 1682 continue; 1683 } 1684 if (nmp->nm_sotype != SOCK_DGRAM) { 1685 if (++rep->r_rexmit > NFS_MAXREXMIT) 1686 rep->r_rexmit = NFS_MAXREXMIT; 1687 continue; 1688 } 1689 if ((so = nmp->nm_so) == NULL) 1690 continue; 1691 1692 /* 1693 * If there is enough space and the window allows.. 1694 * Resend it 1695 * Set r_rtt to -1 in case we fail to send it now. 1696 */ 1697 solock(so); 1698 rep->r_rtt = -1; 1699 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1700 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1701 (rep->r_flags & R_SENT) || 1702 nmp->nm_sent < nmp->nm_cwnd) && 1703 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1704 if (so->so_state & SS_ISCONNECTED) 1705 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1706 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1707 else 1708 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1709 nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0); 1710 if (error) { 1711 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1712 #ifdef DEBUG 1713 printf("nfs_timer: ignoring error %d\n", 1714 error); 1715 #endif 1716 so->so_error = 0; 1717 } 1718 } else { 1719 /* 1720 * Iff first send, start timing 1721 * else turn timing off, backoff timer 1722 * and divide congestion window by 2. 1723 */ 1724 if (rep->r_flags & R_SENT) { 1725 rep->r_flags &= ~R_TIMING; 1726 if (++rep->r_rexmit > NFS_MAXREXMIT) 1727 rep->r_rexmit = NFS_MAXREXMIT; 1728 nmp->nm_cwnd >>= 1; 1729 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1730 nmp->nm_cwnd = NFS_CWNDSCALE; 1731 nfsstats.rpcretries++; 1732 } else { 1733 rep->r_flags |= R_SENT; 1734 nmp->nm_sent += NFS_CWNDSCALE; 1735 } 1736 rep->r_rtt = 0; 1737 } 1738 } 1739 sounlock(so); 1740 } 1741 splx(s); 1742 1743 mutex_enter(&nfs_timer_lock); 1744 if (nfs_timer_srvvec != NULL) { 1745 more |= (*nfs_timer_srvvec)(); 1746 } 1747 mutex_exit(&nfs_timer_lock); 1748 1749 if (more) { 1750 nfs_timer_schedule(); 1751 } else { 1752 nfs_timer_stop_ev.ev_count++; 1753 } 1754 } 1755 1756 /* 1757 * Test for a termination condition pending on the process. 1758 * This is used for NFSMNT_INT mounts. 1759 */ 1760 int 1761 nfs_sigintr(nmp, rep, l) 1762 struct nfsmount *nmp; 1763 struct nfsreq *rep; 1764 struct lwp *l; 1765 { 1766 sigset_t ss; 1767 1768 if (rep && (rep->r_flags & R_SOFTTERM)) 1769 return (EINTR); 1770 if (!(nmp->nm_flag & NFSMNT_INT)) 1771 return (0); 1772 if (l) { 1773 sigpending1(l, &ss); 1774 #if 0 1775 sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss); 1776 #endif 1777 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1778 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1779 sigismember(&ss, SIGQUIT)) 1780 return (EINTR); 1781 } 1782 return (0); 1783 } 1784 1785 #ifdef NFS 1786 /* 1787 * Lock a socket against others. 1788 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1789 * and also to avoid race conditions between the processes with nfs requests 1790 * in progress when a reconnect is necessary. 1791 */ 1792 static int 1793 nfs_sndlock(struct nfsmount *nmp, struct nfsreq *rep) 1794 { 1795 struct lwp *l; 1796 int timeo = 0; 1797 bool catch = false; 1798 int error = 0; 1799 1800 if (rep) { 1801 l = rep->r_lwp; 1802 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1803 catch = true; 1804 } else 1805 l = NULL; 1806 mutex_enter(&nmp->nm_lock); 1807 while ((nmp->nm_iflag & NFSMNT_SNDLOCK) != 0) { 1808 if (rep && nfs_sigintr(rep->r_nmp, rep, l)) { 1809 error = EINTR; 1810 goto quit; 1811 } 1812 if (catch) { 1813 cv_timedwait_sig(&nmp->nm_sndcv, &nmp->nm_lock, timeo); 1814 } else { 1815 cv_timedwait(&nmp->nm_sndcv, &nmp->nm_lock, timeo); 1816 } 1817 if (catch) { 1818 catch = false; 1819 timeo = 2 * hz; 1820 } 1821 } 1822 nmp->nm_iflag |= NFSMNT_SNDLOCK; 1823 quit: 1824 mutex_exit(&nmp->nm_lock); 1825 return error; 1826 } 1827 1828 /* 1829 * Unlock the stream socket for others. 1830 */ 1831 static void 1832 nfs_sndunlock(struct nfsmount *nmp) 1833 { 1834 1835 mutex_enter(&nmp->nm_lock); 1836 if ((nmp->nm_iflag & NFSMNT_SNDLOCK) == 0) 1837 panic("nfs sndunlock"); 1838 nmp->nm_iflag &= ~NFSMNT_SNDLOCK; 1839 cv_signal(&nmp->nm_sndcv); 1840 mutex_exit(&nmp->nm_lock); 1841 } 1842 #endif /* NFS */ 1843 1844 static int 1845 nfs_rcvlock(struct nfsmount *nmp, struct nfsreq *rep) 1846 { 1847 int *flagp = &nmp->nm_iflag; 1848 int slptimeo = 0; 1849 bool catch; 1850 int error = 0; 1851 1852 KASSERT(nmp == rep->r_nmp); 1853 1854 catch = (nmp->nm_flag & NFSMNT_INT) != 0; 1855 mutex_enter(&nmp->nm_lock); 1856 while (/* CONSTCOND */ true) { 1857 if (*flagp & NFSMNT_DISMNT) { 1858 cv_signal(&nmp->nm_disconcv); 1859 error = EIO; 1860 break; 1861 } 1862 /* If our reply was received while we were sleeping, 1863 * then just return without taking the lock to avoid a 1864 * situation where a single iod could 'capture' the 1865 * receive lock. 1866 */ 1867 if (rep->r_mrep != NULL) { 1868 error = EALREADY; 1869 break; 1870 } 1871 if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) { 1872 error = EINTR; 1873 break; 1874 } 1875 if ((*flagp & NFSMNT_RCVLOCK) == 0) { 1876 *flagp |= NFSMNT_RCVLOCK; 1877 break; 1878 } 1879 if (catch) { 1880 cv_timedwait_sig(&nmp->nm_rcvcv, &nmp->nm_lock, 1881 slptimeo); 1882 } else { 1883 cv_timedwait(&nmp->nm_rcvcv, &nmp->nm_lock, 1884 slptimeo); 1885 } 1886 if (catch) { 1887 catch = false; 1888 slptimeo = 2 * hz; 1889 } 1890 } 1891 mutex_exit(&nmp->nm_lock); 1892 return error; 1893 } 1894 1895 /* 1896 * Unlock the stream socket for others. 1897 */ 1898 static void 1899 nfs_rcvunlock(struct nfsmount *nmp) 1900 { 1901 1902 mutex_enter(&nmp->nm_lock); 1903 if ((nmp->nm_iflag & NFSMNT_RCVLOCK) == 0) 1904 panic("nfs rcvunlock"); 1905 nmp->nm_iflag &= ~NFSMNT_RCVLOCK; 1906 cv_broadcast(&nmp->nm_rcvcv); 1907 mutex_exit(&nmp->nm_lock); 1908 } 1909 1910 /* 1911 * Parse an RPC request 1912 * - verify it 1913 * - allocate and fill in the cred. 1914 */ 1915 int 1916 nfs_getreq(nd, nfsd, has_header) 1917 struct nfsrv_descript *nd; 1918 struct nfsd *nfsd; 1919 int has_header; 1920 { 1921 int len, i; 1922 u_int32_t *tl; 1923 int32_t t1; 1924 struct uio uio; 1925 struct iovec iov; 1926 char *dpos, *cp2, *cp; 1927 u_int32_t nfsvers, auth_type; 1928 uid_t nickuid; 1929 int error = 0, ticklen; 1930 struct mbuf *mrep, *md; 1931 struct nfsuid *nuidp; 1932 struct timeval tvin, tvout; 1933 1934 memset(&tvout, 0, sizeof tvout); /* XXX gcc */ 1935 1936 KASSERT(nd->nd_cr == NULL); 1937 mrep = nd->nd_mrep; 1938 md = nd->nd_md; 1939 dpos = nd->nd_dpos; 1940 if (has_header) { 1941 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1942 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1943 if (*tl++ != rpc_call) { 1944 m_freem(mrep); 1945 return (EBADRPC); 1946 } 1947 } else 1948 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1949 nd->nd_repstat = 0; 1950 nd->nd_flag = 0; 1951 if (*tl++ != rpc_vers) { 1952 nd->nd_repstat = ERPCMISMATCH; 1953 nd->nd_procnum = NFSPROC_NOOP; 1954 return (0); 1955 } 1956 if (*tl != nfs_prog) { 1957 nd->nd_repstat = EPROGUNAVAIL; 1958 nd->nd_procnum = NFSPROC_NOOP; 1959 return (0); 1960 } 1961 tl++; 1962 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1963 if (nfsvers < NFS_VER2 || nfsvers > NFS_VER3) { 1964 nd->nd_repstat = EPROGMISMATCH; 1965 nd->nd_procnum = NFSPROC_NOOP; 1966 return (0); 1967 } 1968 if (nfsvers == NFS_VER3) 1969 nd->nd_flag = ND_NFSV3; 1970 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1971 if (nd->nd_procnum == NFSPROC_NULL) 1972 return (0); 1973 if (nd->nd_procnum > NFSPROC_COMMIT || 1974 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1975 nd->nd_repstat = EPROCUNAVAIL; 1976 nd->nd_procnum = NFSPROC_NOOP; 1977 return (0); 1978 } 1979 if ((nd->nd_flag & ND_NFSV3) == 0) 1980 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1981 auth_type = *tl++; 1982 len = fxdr_unsigned(int, *tl++); 1983 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1984 m_freem(mrep); 1985 return (EBADRPC); 1986 } 1987 1988 nd->nd_flag &= ~ND_KERBAUTH; 1989 /* 1990 * Handle auth_unix or auth_kerb. 1991 */ 1992 if (auth_type == rpc_auth_unix) { 1993 uid_t uid; 1994 gid_t gid; 1995 1996 nd->nd_cr = kauth_cred_alloc(); 1997 len = fxdr_unsigned(int, *++tl); 1998 if (len < 0 || len > NFS_MAXNAMLEN) { 1999 m_freem(mrep); 2000 error = EBADRPC; 2001 goto errout; 2002 } 2003 nfsm_adv(nfsm_rndup(len)); 2004 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2005 2006 uid = fxdr_unsigned(uid_t, *tl++); 2007 gid = fxdr_unsigned(gid_t, *tl++); 2008 kauth_cred_setuid(nd->nd_cr, uid); 2009 kauth_cred_seteuid(nd->nd_cr, uid); 2010 kauth_cred_setsvuid(nd->nd_cr, uid); 2011 kauth_cred_setgid(nd->nd_cr, gid); 2012 kauth_cred_setegid(nd->nd_cr, gid); 2013 kauth_cred_setsvgid(nd->nd_cr, gid); 2014 2015 len = fxdr_unsigned(int, *tl); 2016 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 2017 m_freem(mrep); 2018 error = EBADRPC; 2019 goto errout; 2020 } 2021 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 2022 2023 if (len > 0) { 2024 size_t grbuf_size = min(len, NGROUPS) * sizeof(gid_t); 2025 gid_t *grbuf = kmem_alloc(grbuf_size, KM_SLEEP); 2026 2027 for (i = 0; i < len; i++) { 2028 if (i < NGROUPS) /* XXX elad */ 2029 grbuf[i] = fxdr_unsigned(gid_t, *tl++); 2030 else 2031 tl++; 2032 } 2033 kauth_cred_setgroups(nd->nd_cr, grbuf, 2034 min(len, NGROUPS), -1, UIO_SYSSPACE); 2035 kmem_free(grbuf, grbuf_size); 2036 } 2037 2038 len = fxdr_unsigned(int, *++tl); 2039 if (len < 0 || len > RPCAUTH_MAXSIZ) { 2040 m_freem(mrep); 2041 error = EBADRPC; 2042 goto errout; 2043 } 2044 if (len > 0) 2045 nfsm_adv(nfsm_rndup(len)); 2046 } else if (auth_type == rpc_auth_kerb) { 2047 switch (fxdr_unsigned(int, *tl++)) { 2048 case RPCAKN_FULLNAME: 2049 ticklen = fxdr_unsigned(int, *tl); 2050 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 2051 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 2052 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 2053 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 2054 m_freem(mrep); 2055 error = EBADRPC; 2056 goto errout; 2057 } 2058 uio.uio_offset = 0; 2059 uio.uio_iov = &iov; 2060 uio.uio_iovcnt = 1; 2061 UIO_SETUP_SYSSPACE(&uio); 2062 iov.iov_base = (void *)&nfsd->nfsd_authstr[4]; 2063 iov.iov_len = RPCAUTH_MAXSIZ - 4; 2064 nfsm_mtouio(&uio, uio.uio_resid); 2065 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2066 if (*tl++ != rpc_auth_kerb || 2067 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 2068 printf("Bad kerb verifier\n"); 2069 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2070 nd->nd_procnum = NFSPROC_NOOP; 2071 return (0); 2072 } 2073 nfsm_dissect(cp, void *, 4 * NFSX_UNSIGNED); 2074 tl = (u_int32_t *)cp; 2075 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 2076 printf("Not fullname kerb verifier\n"); 2077 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2078 nd->nd_procnum = NFSPROC_NOOP; 2079 return (0); 2080 } 2081 cp += NFSX_UNSIGNED; 2082 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 2083 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 2084 nd->nd_flag |= ND_KERBFULL; 2085 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 2086 break; 2087 case RPCAKN_NICKNAME: 2088 if (len != 2 * NFSX_UNSIGNED) { 2089 printf("Kerb nickname short\n"); 2090 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 2091 nd->nd_procnum = NFSPROC_NOOP; 2092 return (0); 2093 } 2094 nickuid = fxdr_unsigned(uid_t, *tl); 2095 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2096 if (*tl++ != rpc_auth_kerb || 2097 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 2098 printf("Kerb nick verifier bad\n"); 2099 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2100 nd->nd_procnum = NFSPROC_NOOP; 2101 return (0); 2102 } 2103 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2104 tvin.tv_sec = *tl++; 2105 tvin.tv_usec = *tl; 2106 2107 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), 2108 nu_hash) { 2109 if (kauth_cred_geteuid(nuidp->nu_cr) == nickuid && 2110 (!nd->nd_nam2 || 2111 netaddr_match(NU_NETFAM(nuidp), 2112 &nuidp->nu_haddr, nd->nd_nam2))) 2113 break; 2114 } 2115 if (!nuidp) { 2116 nd->nd_repstat = 2117 (NFSERR_AUTHERR|AUTH_REJECTCRED); 2118 nd->nd_procnum = NFSPROC_NOOP; 2119 return (0); 2120 } 2121 2122 /* 2123 * Now, decrypt the timestamp using the session key 2124 * and validate it. 2125 */ 2126 #ifdef NFSKERB 2127 XXX 2128 #endif 2129 2130 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 2131 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 2132 if (nuidp->nu_expire < time_second || 2133 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 2134 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 2135 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 2136 nuidp->nu_expire = 0; 2137 nd->nd_repstat = 2138 (NFSERR_AUTHERR|AUTH_REJECTVERF); 2139 nd->nd_procnum = NFSPROC_NOOP; 2140 return (0); 2141 } 2142 kauth_cred_hold(nuidp->nu_cr); 2143 nd->nd_cr = nuidp->nu_cr; 2144 nd->nd_flag |= ND_KERBNICK; 2145 } 2146 } else { 2147 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 2148 nd->nd_procnum = NFSPROC_NOOP; 2149 return (0); 2150 } 2151 2152 nd->nd_md = md; 2153 nd->nd_dpos = dpos; 2154 KASSERT((nd->nd_cr == NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) != 0) 2155 || (nd->nd_cr != NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) == 0)); 2156 return (0); 2157 nfsmout: 2158 errout: 2159 KASSERT(error != 0); 2160 if (nd->nd_cr != NULL) { 2161 kauth_cred_free(nd->nd_cr); 2162 nd->nd_cr = NULL; 2163 } 2164 return (error); 2165 } 2166 2167 int 2168 nfs_msg(l, server, msg) 2169 struct lwp *l; 2170 const char *server, *msg; 2171 { 2172 tpr_t tpr; 2173 2174 if (l) 2175 tpr = tprintf_open(l->l_proc); 2176 else 2177 tpr = NULL; 2178 tprintf(tpr, "nfs server %s: %s\n", server, msg); 2179 tprintf_close(tpr); 2180 return (0); 2181 } 2182 2183 static struct pool nfs_srvdesc_pool; 2184 2185 void 2186 nfsdreq_init(void) 2187 { 2188 2189 pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript), 2190 0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr, IPL_NONE); 2191 } 2192 2193 void 2194 nfsdreq_fini(void) 2195 { 2196 2197 pool_destroy(&nfs_srvdesc_pool); 2198 } 2199 2200 struct nfsrv_descript * 2201 nfsdreq_alloc(void) 2202 { 2203 struct nfsrv_descript *nd; 2204 2205 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK); 2206 nd->nd_cr = NULL; 2207 return nd; 2208 } 2209 2210 void 2211 nfsdreq_free(struct nfsrv_descript *nd) 2212 { 2213 kauth_cred_t cr; 2214 2215 cr = nd->nd_cr; 2216 if (cr != NULL) { 2217 kauth_cred_free(cr); 2218 } 2219 pool_put(&nfs_srvdesc_pool, nd); 2220 } 2221
Indexes created Sat Sep 20 22:09:52 GMT 2025