1 /* $NetBSD: nfs_socket.c,v 1.122 2006/01/03 12:30:01 yamt 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 1.122 2006/01/03 12:30:01 yamt Exp $"); 43 44 #include "fs_nfs.h" 45 #include "opt_nfs.h" 46 #include "opt_nfsserver.h" 47 #include "opt_mbuftrace.h" 48 #include "opt_inet.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/callout.h> 53 #include <sys/proc.h> 54 #include <sys/mount.h> 55 #include <sys/kernel.h> 56 #include <sys/mbuf.h> 57 #include <sys/vnode.h> 58 #include <sys/domain.h> 59 #include <sys/protosw.h> 60 #include <sys/socket.h> 61 #include <sys/socketvar.h> 62 #include <sys/syslog.h> 63 #include <sys/tprintf.h> 64 #include <sys/namei.h> 65 #include <sys/signal.h> 66 #include <sys/signalvar.h> 67 68 #include <netinet/in.h> 69 #include <netinet/tcp.h> 70 71 #include <nfs/rpcv2.h> 72 #include <nfs/nfsproto.h> 73 #include <nfs/nfs.h> 74 #include <nfs/xdr_subs.h> 75 #include <nfs/nfsm_subs.h> 76 #include <nfs/nfsmount.h> 77 #include <nfs/nfsnode.h> 78 #include <nfs/nfsrtt.h> 79 #include <nfs/nqnfs.h> 80 #include <nfs/nfs_var.h> 81 82 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header"); 83 #ifdef MBUFTRACE 84 struct mowner nfs_mowner = { "nfs" }; 85 #endif 86 87 /* 88 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 89 * Use the mean and mean deviation of rtt for the appropriate type of rpc 90 * for the frequent rpcs and a default for the others. 91 * The justification for doing "other" this way is that these rpcs 92 * happen so infrequently that timer est. would probably be stale. 93 * Also, since many of these rpcs are 94 * non-idempotent, a conservative timeout is desired. 95 * getattr, lookup - A+2D 96 * read, write - A+4D 97 * other - nm_timeo 98 */ 99 #define NFS_RTO(n, t) \ 100 ((t) == 0 ? (n)->nm_timeo : \ 101 ((t) < 3 ? \ 102 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 103 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 104 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 105 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 106 /* 107 * External data, mostly RPC constants in XDR form 108 */ 109 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 110 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, 111 rpc_auth_kerb; 112 extern u_int32_t nfs_prog, nqnfs_prog; 113 extern time_t nqnfsstarttime; 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 [NQNFSPROC_GETLEASE] = 0, 149 [NQNFSPROC_VACATED] = 0, 150 [NQNFSPROC_EVICTED] = 0, 151 [NFSPROC_NOOP] = 0, 152 }; 153 154 /* 155 * There is a congestion window for outstanding rpcs maintained per mount 156 * point. The cwnd size is adjusted in roughly the way that: 157 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 158 * SIGCOMM '88". ACM, August 1988. 159 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 160 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 161 * of rpcs is in progress. 162 * (The sent count and cwnd are scaled for integer arith.) 163 * Variants of "slow start" were tried and were found to be too much of a 164 * performance hit (ave. rtt 3 times larger), 165 * I suspect due to the large rtt that nfs rpcs have. 166 */ 167 #define NFS_CWNDSCALE 256 168 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 169 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 170 int nfsrtton = 0; 171 struct nfsrtt nfsrtt; 172 struct nfsreqhead nfs_reqq; 173 174 struct callout nfs_timer_ch = CALLOUT_INITIALIZER_SETFUNC(nfs_timer, NULL); 175 176 /* 177 * Initialize sockets and congestion for a new NFS connection. 178 * We do not free the sockaddr if error. 179 */ 180 int 181 nfs_connect(nmp, rep, l) 182 struct nfsmount *nmp; 183 struct nfsreq *rep; 184 struct lwp *l; 185 { 186 struct socket *so; 187 int s, error, rcvreserve, sndreserve; 188 struct sockaddr *saddr; 189 struct sockaddr_in *sin; 190 #ifdef INET6 191 struct sockaddr_in6 *sin6; 192 #endif 193 struct mbuf *m; 194 195 nmp->nm_so = (struct socket *)0; 196 saddr = mtod(nmp->nm_nam, struct sockaddr *); 197 error = socreate(saddr->sa_family, &nmp->nm_so, 198 nmp->nm_sotype, nmp->nm_soproto, l); 199 if (error) 200 goto bad; 201 so = nmp->nm_so; 202 #ifdef MBUFTRACE 203 so->so_mowner = &nfs_mowner; 204 so->so_rcv.sb_mowner = &nfs_mowner; 205 so->so_snd.sb_mowner = &nfs_mowner; 206 #endif 207 nmp->nm_soflags = so->so_proto->pr_flags; 208 209 /* 210 * Some servers require that the client port be a reserved port number. 211 */ 212 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 213 m = m_get(M_WAIT, MT_SOOPTS); 214 MCLAIM(m, so->so_mowner); 215 *mtod(m, int32_t *) = IP_PORTRANGE_LOW; 216 m->m_len = sizeof(int32_t); 217 if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m))) 218 goto bad; 219 m = m_get(M_WAIT, MT_SONAME); 220 MCLAIM(m, so->so_mowner); 221 sin = mtod(m, struct sockaddr_in *); 222 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 223 sin->sin_family = AF_INET; 224 sin->sin_addr.s_addr = INADDR_ANY; 225 sin->sin_port = 0; 226 error = sobind(so, m, &lwp0); 227 m_freem(m); 228 if (error) 229 goto bad; 230 } 231 #ifdef INET6 232 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) { 233 m = m_get(M_WAIT, MT_SOOPTS); 234 MCLAIM(m, so->so_mowner); 235 *mtod(m, int32_t *) = IPV6_PORTRANGE_LOW; 236 m->m_len = sizeof(int32_t); 237 if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m))) 238 goto bad; 239 m = m_get(M_WAIT, MT_SONAME); 240 MCLAIM(m, so->so_mowner); 241 sin6 = mtod(m, struct sockaddr_in6 *); 242 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6); 243 sin6->sin6_family = AF_INET6; 244 sin6->sin6_addr = in6addr_any; 245 sin6->sin6_port = 0; 246 error = sobind(so, m, &lwp0); 247 m_freem(m); 248 if (error) 249 goto bad; 250 } 251 #endif 252 253 /* 254 * Protocols that do not require connections may be optionally left 255 * unconnected for servers that reply from a port other than NFS_PORT. 256 */ 257 if (nmp->nm_flag & NFSMNT_NOCONN) { 258 if (nmp->nm_soflags & PR_CONNREQUIRED) { 259 error = ENOTCONN; 260 goto bad; 261 } 262 } else { 263 error = soconnect(so, nmp->nm_nam, l); 264 if (error) 265 goto bad; 266 267 /* 268 * Wait for the connection to complete. Cribbed from the 269 * connect system call but with the wait timing out so 270 * that interruptible mounts don't hang here for a long time. 271 */ 272 s = splsoftnet(); 273 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 274 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 275 "nfscn1", 2 * hz); 276 if ((so->so_state & SS_ISCONNECTING) && 277 so->so_error == 0 && rep && 278 (error = nfs_sigintr(nmp, rep, rep->r_lwp)) != 0){ 279 so->so_state &= ~SS_ISCONNECTING; 280 splx(s); 281 goto bad; 282 } 283 } 284 if (so->so_error) { 285 error = so->so_error; 286 so->so_error = 0; 287 splx(s); 288 goto bad; 289 } 290 splx(s); 291 } 292 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 293 so->so_rcv.sb_timeo = (5 * hz); 294 so->so_snd.sb_timeo = (5 * hz); 295 } else { 296 /* 297 * enable receive timeout to detect server crash and reconnect. 298 * otherwise, we can be stuck in soreceive forever. 299 */ 300 so->so_rcv.sb_timeo = (5 * hz); 301 so->so_snd.sb_timeo = 0; 302 } 303 if (nmp->nm_sotype == SOCK_DGRAM) { 304 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 305 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 306 NFS_MAXPKTHDR) * 2; 307 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 308 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 309 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 310 NFS_MAXPKTHDR) * 2; 311 } else { 312 if (nmp->nm_sotype != SOCK_STREAM) 313 panic("nfscon sotype"); 314 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 315 m = m_get(M_WAIT, MT_SOOPTS); 316 MCLAIM(m, so->so_mowner); 317 *mtod(m, int32_t *) = 1; 318 m->m_len = sizeof(int32_t); 319 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 320 } 321 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 322 m = m_get(M_WAIT, MT_SOOPTS); 323 MCLAIM(m, so->so_mowner); 324 *mtod(m, int32_t *) = 1; 325 m->m_len = sizeof(int32_t); 326 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 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 } 333 error = soreserve(so, sndreserve, rcvreserve); 334 if (error) 335 goto bad; 336 so->so_rcv.sb_flags |= SB_NOINTR; 337 so->so_snd.sb_flags |= SB_NOINTR; 338 339 /* Initialize other non-zero congestion variables */ 340 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 341 NFS_TIMEO << 3; 342 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 343 nmp->nm_sdrtt[3] = 0; 344 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 345 nmp->nm_sent = 0; 346 nmp->nm_timeouts = 0; 347 return (0); 348 349 bad: 350 nfs_disconnect(nmp); 351 return (error); 352 } 353 354 /* 355 * Reconnect routine: 356 * Called when a connection is broken on a reliable protocol. 357 * - clean up the old socket 358 * - nfs_connect() again 359 * - set R_MUSTRESEND for all outstanding requests on mount point 360 * If this fails the mount point is DEAD! 361 * nb: Must be called with the nfs_sndlock() set on the mount point. 362 */ 363 int 364 nfs_reconnect(rep, l) 365 struct nfsreq *rep; 366 struct lwp *l; 367 { 368 struct nfsreq *rp; 369 struct nfsmount *nmp = rep->r_nmp; 370 int error; 371 372 nfs_disconnect(nmp); 373 while ((error = nfs_connect(nmp, rep, l)) != 0) { 374 if (error == EINTR || error == ERESTART) 375 return (EINTR); 376 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0); 377 } 378 379 /* 380 * Loop through outstanding request list and fix up all requests 381 * on old socket. 382 */ 383 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 384 if (rp->r_nmp == nmp) { 385 if ((rp->r_flags & R_MUSTRESEND) == 0) 386 rp->r_flags |= R_MUSTRESEND | R_REXMITTED; 387 rp->r_rexmit = 0; 388 } 389 } 390 return (0); 391 } 392 393 /* 394 * NFS disconnect. Clean up and unlink. 395 */ 396 void 397 nfs_disconnect(nmp) 398 struct nfsmount *nmp; 399 { 400 struct socket *so; 401 int drain = 0; 402 403 if (nmp->nm_so) { 404 so = nmp->nm_so; 405 nmp->nm_so = (struct socket *)0; 406 soshutdown(so, 2); 407 drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0; 408 if (drain) { 409 /* 410 * soshutdown() above should wake up the current 411 * listener. 412 * Now wake up those waiting for the receive lock, and 413 * wait for them to go away unhappy, to prevent *nmp 414 * from evaporating while they're sleeping. 415 */ 416 while (nmp->nm_waiters > 0) { 417 wakeup (&nmp->nm_iflag); 418 (void) tsleep(&nmp->nm_waiters, PVFS, 419 "nfsdis", 0); 420 } 421 } 422 soclose(so); 423 } 424 #ifdef DIAGNOSTIC 425 if (drain && (nmp->nm_waiters > 0)) 426 panic("nfs_disconnect: waiters left after drain?"); 427 #endif 428 } 429 430 void 431 nfs_safedisconnect(nmp) 432 struct nfsmount *nmp; 433 { 434 struct nfsreq dummyreq; 435 436 memset(&dummyreq, 0, sizeof(dummyreq)); 437 dummyreq.r_nmp = nmp; 438 nfs_rcvlock(&dummyreq); /* XXX ignored error return */ 439 nfs_disconnect(nmp); 440 nfs_rcvunlock(nmp); 441 } 442 443 /* 444 * This is the nfs send routine. For connection based socket types, it 445 * must be called with an nfs_sndlock() on the socket. 446 * "rep == NULL" indicates that it has been called from a server. 447 * For the client side: 448 * - return EINTR if the RPC is terminated, 0 otherwise 449 * - set R_MUSTRESEND if the send fails for any reason 450 * - do any cleanup required by recoverable socket errors (? ? ?) 451 * For the server side: 452 * - return EINTR or ERESTART if interrupted by a signal 453 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 454 * - do any cleanup required by recoverable socket errors (? ? ?) 455 */ 456 int 457 nfs_send(so, nam, top, rep, l) 458 struct socket *so; 459 struct mbuf *nam; 460 struct mbuf *top; 461 struct nfsreq *rep; 462 struct lwp *l; 463 { 464 struct mbuf *sendnam; 465 int error, soflags, flags; 466 467 /* XXX nfs_doio()/nfs_request() calls with rep->r_lwp == NULL */ 468 if (l == NULL && rep->r_lwp == NULL) 469 l = curlwp; 470 471 if (rep) { 472 if (rep->r_flags & R_SOFTTERM) { 473 m_freem(top); 474 return (EINTR); 475 } 476 if ((so = rep->r_nmp->nm_so) == NULL) { 477 rep->r_flags |= R_MUSTRESEND; 478 m_freem(top); 479 return (0); 480 } 481 rep->r_flags &= ~R_MUSTRESEND; 482 soflags = rep->r_nmp->nm_soflags; 483 } else 484 soflags = so->so_proto->pr_flags; 485 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 486 sendnam = (struct mbuf *)0; 487 else 488 sendnam = nam; 489 if (so->so_type == SOCK_SEQPACKET) 490 flags = MSG_EOR; 491 else 492 flags = 0; 493 494 error = (*so->so_send)(so, sendnam, (struct uio *)0, top, 495 (struct mbuf *)0, flags, l); 496 if (error) { 497 if (rep) { 498 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 499 /* 500 * We're too fast for the network/driver, 501 * and UDP isn't flowcontrolled. 502 * We need to resend. This is not fatal, 503 * just try again. 504 * 505 * Could be smarter here by doing some sort 506 * of a backoff, but this is rare. 507 */ 508 rep->r_flags |= R_MUSTRESEND; 509 } else { 510 if (error != EPIPE) 511 log(LOG_INFO, 512 "nfs send error %d for %s\n", 513 error, 514 rep->r_nmp->nm_mountp-> 515 mnt_stat.f_mntfromname); 516 /* 517 * Deal with errors for the client side. 518 */ 519 if (rep->r_flags & R_SOFTTERM) 520 error = EINTR; 521 else 522 rep->r_flags |= R_MUSTRESEND; 523 } 524 } else { 525 /* 526 * See above. This error can happen under normal 527 * circumstances and the log is too noisy. 528 * The error will still show up in nfsstat. 529 */ 530 if (error != ENOBUFS || so->so_type != SOCK_DGRAM) 531 log(LOG_INFO, "nfsd send error %d\n", error); 532 } 533 534 /* 535 * Handle any recoverable (soft) socket errors here. (? ? ?) 536 */ 537 if (error != EINTR && error != ERESTART && 538 error != EWOULDBLOCK && error != EPIPE) 539 error = 0; 540 } 541 return (error); 542 } 543 544 #ifdef NFS 545 /* 546 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 547 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 548 * Mark and consolidate the data into a new mbuf list. 549 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 550 * small mbufs. 551 * For SOCK_STREAM we must be very careful to read an entire record once 552 * we have read any of it, even if the system call has been interrupted. 553 */ 554 int 555 nfs_receive(rep, aname, mp, l) 556 struct nfsreq *rep; 557 struct mbuf **aname; 558 struct mbuf **mp; 559 struct lwp *l; 560 { 561 struct socket *so; 562 struct uio auio; 563 struct iovec aio; 564 struct mbuf *m; 565 struct mbuf *control; 566 u_int32_t len; 567 struct mbuf **getnam; 568 int error, sotype, rcvflg; 569 570 /* 571 * Set up arguments for soreceive() 572 */ 573 *mp = (struct mbuf *)0; 574 *aname = (struct mbuf *)0; 575 sotype = rep->r_nmp->nm_sotype; 576 577 /* 578 * For reliable protocols, lock against other senders/receivers 579 * in case a reconnect is necessary. 580 * For SOCK_STREAM, first get the Record Mark to find out how much 581 * more there is to get. 582 * We must lock the socket against other receivers 583 * until we have an entire rpc request/reply. 584 */ 585 if (sotype != SOCK_DGRAM) { 586 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 587 if (error) 588 return (error); 589 tryagain: 590 /* 591 * Check for fatal errors and resending request. 592 */ 593 /* 594 * Ugh: If a reconnect attempt just happened, nm_so 595 * would have changed. NULL indicates a failed 596 * attempt that has essentially shut down this 597 * mount point. 598 */ 599 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 600 nfs_sndunlock(&rep->r_nmp->nm_iflag); 601 return (EINTR); 602 } 603 so = rep->r_nmp->nm_so; 604 if (!so) { 605 error = nfs_reconnect(rep, l); 606 if (error) { 607 nfs_sndunlock(&rep->r_nmp->nm_iflag); 608 return (error); 609 } 610 goto tryagain; 611 } 612 while (rep->r_flags & R_MUSTRESEND) { 613 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 614 nfsstats.rpcretries++; 615 rep->r_rtt = 0; 616 rep->r_flags &= ~R_TIMING; 617 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, l); 618 if (error) { 619 if (error == EINTR || error == ERESTART || 620 (error = nfs_reconnect(rep, l)) != 0) { 621 nfs_sndunlock(&rep->r_nmp->nm_iflag); 622 return (error); 623 } 624 goto tryagain; 625 } 626 } 627 nfs_sndunlock(&rep->r_nmp->nm_iflag); 628 if (sotype == SOCK_STREAM) { 629 aio.iov_base = (caddr_t) &len; 630 aio.iov_len = sizeof(u_int32_t); 631 auio.uio_iov = &aio; 632 auio.uio_iovcnt = 1; 633 auio.uio_segflg = UIO_SYSSPACE; 634 auio.uio_rw = UIO_READ; 635 auio.uio_offset = 0; 636 auio.uio_resid = sizeof(u_int32_t); 637 auio.uio_lwp = NULL; 638 do { 639 rcvflg = MSG_WAITALL; 640 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 641 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 642 if (error == EWOULDBLOCK && rep) { 643 if (rep->r_flags & R_SOFTTERM) 644 return (EINTR); 645 /* 646 * if it seems that the server died after it 647 * received our request, set EPIPE so that 648 * we'll reconnect and retransmit requests. 649 */ 650 if (rep->r_rexmit >= rep->r_nmp->nm_retry) { 651 nfsstats.rpctimeouts++; 652 error = EPIPE; 653 } 654 } 655 } while (error == EWOULDBLOCK); 656 if (!error && auio.uio_resid > 0) { 657 /* 658 * Don't log a 0 byte receive; it means 659 * that the socket has been closed, and 660 * can happen during normal operation 661 * (forcible unmount or Solaris server). 662 */ 663 if (auio.uio_resid != sizeof (u_int32_t)) 664 log(LOG_INFO, 665 "short receive (%lu/%lu) from nfs server %s\n", 666 (u_long)sizeof(u_int32_t) - auio.uio_resid, 667 (u_long)sizeof(u_int32_t), 668 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 669 error = EPIPE; 670 } 671 if (error) 672 goto errout; 673 len = ntohl(len) & ~0x80000000; 674 /* 675 * This is SERIOUS! We are out of sync with the sender 676 * and forcing a disconnect/reconnect is all I can do. 677 */ 678 if (len > NFS_MAXPACKET) { 679 log(LOG_ERR, "%s (%d) from nfs server %s\n", 680 "impossible packet length", 681 len, 682 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 683 error = EFBIG; 684 goto errout; 685 } 686 auio.uio_resid = len; 687 do { 688 rcvflg = MSG_WAITALL; 689 error = (*so->so_receive)(so, (struct mbuf **)0, 690 &auio, mp, (struct mbuf **)0, &rcvflg); 691 } while (error == EWOULDBLOCK || error == EINTR || 692 error == ERESTART); 693 if (!error && auio.uio_resid > 0) { 694 if (len != auio.uio_resid) 695 log(LOG_INFO, 696 "short receive (%lu/%d) from nfs server %s\n", 697 (u_long)len - auio.uio_resid, len, 698 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 699 error = EPIPE; 700 } 701 } else { 702 /* 703 * NB: Since uio_resid is big, MSG_WAITALL is ignored 704 * and soreceive() will return when it has either a 705 * control msg or a data msg. 706 * We have no use for control msg., but must grab them 707 * and then throw them away so we know what is going 708 * on. 709 */ 710 auio.uio_resid = len = 100000000; /* Anything Big */ 711 auio.uio_lwp = l; 712 do { 713 rcvflg = 0; 714 error = (*so->so_receive)(so, (struct mbuf **)0, 715 &auio, mp, &control, &rcvflg); 716 if (control) 717 m_freem(control); 718 if (error == EWOULDBLOCK && rep) { 719 if (rep->r_flags & R_SOFTTERM) 720 return (EINTR); 721 } 722 } while (error == EWOULDBLOCK || 723 (!error && *mp == NULL && control)); 724 if ((rcvflg & MSG_EOR) == 0) 725 printf("Egad!!\n"); 726 if (!error && *mp == NULL) 727 error = EPIPE; 728 len -= auio.uio_resid; 729 } 730 errout: 731 if (error && error != EINTR && error != ERESTART) { 732 m_freem(*mp); 733 *mp = (struct mbuf *)0; 734 if (error != EPIPE) 735 log(LOG_INFO, 736 "receive error %d from nfs server %s\n", 737 error, 738 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 739 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 740 if (!error) 741 error = nfs_reconnect(rep, l); 742 if (!error) 743 goto tryagain; 744 else 745 nfs_sndunlock(&rep->r_nmp->nm_iflag); 746 } 747 } else { 748 if ((so = rep->r_nmp->nm_so) == NULL) 749 return (EACCES); 750 if (so->so_state & SS_ISCONNECTED) 751 getnam = (struct mbuf **)0; 752 else 753 getnam = aname; 754 auio.uio_resid = len = 1000000; 755 auio.uio_lwp = l; 756 do { 757 rcvflg = 0; 758 error = (*so->so_receive)(so, getnam, &auio, mp, 759 (struct mbuf **)0, &rcvflg); 760 if (error == EWOULDBLOCK && 761 (rep->r_flags & R_SOFTTERM)) 762 return (EINTR); 763 } while (error == EWOULDBLOCK); 764 len -= auio.uio_resid; 765 if (!error && *mp == NULL) 766 error = EPIPE; 767 } 768 if (error) { 769 m_freem(*mp); 770 *mp = (struct mbuf *)0; 771 } 772 return (error); 773 } 774 775 /* 776 * Implement receipt of reply on a socket. 777 * We must search through the list of received datagrams matching them 778 * with outstanding requests using the xid, until ours is found. 779 */ 780 /* ARGSUSED */ 781 int 782 nfs_reply(myrep, lwp) 783 struct nfsreq *myrep; 784 struct lwp *lwp; 785 { 786 struct nfsreq *rep; 787 struct nfsmount *nmp = myrep->r_nmp; 788 int32_t t1; 789 struct mbuf *mrep, *nam, *md; 790 u_int32_t rxid, *tl; 791 caddr_t dpos, cp2; 792 int error; 793 794 /* 795 * Loop around until we get our own reply 796 */ 797 for (;;) { 798 /* 799 * Lock against other receivers so that I don't get stuck in 800 * sbwait() after someone else has received my reply for me. 801 * Also necessary for connection based protocols to avoid 802 * race conditions during a reconnect. 803 */ 804 error = nfs_rcvlock(myrep); 805 if (error == EALREADY) 806 return (0); 807 if (error) 808 return (error); 809 /* 810 * Get the next Rpc reply off the socket 811 */ 812 nmp->nm_waiters++; 813 error = nfs_receive(myrep, &nam, &mrep, lwp); 814 nfs_rcvunlock(nmp); 815 if (error) { 816 817 if (nmp->nm_iflag & NFSMNT_DISMNT) { 818 /* 819 * Oops, we're going away now.. 820 */ 821 nmp->nm_waiters--; 822 wakeup (&nmp->nm_waiters); 823 return error; 824 } 825 nmp->nm_waiters--; 826 /* 827 * Ignore routing errors on connectionless protocols? ? 828 */ 829 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 830 nmp->nm_so->so_error = 0; 831 #ifdef DEBUG 832 printf("nfs_reply: ignoring error %d\n", error); 833 #endif 834 if (myrep->r_flags & R_GETONEREP) 835 return (0); 836 continue; 837 } 838 return (error); 839 } 840 nmp->nm_waiters--; 841 if (nam) 842 m_freem(nam); 843 844 /* 845 * Get the xid and check that it is an rpc reply 846 */ 847 md = mrep; 848 dpos = mtod(md, caddr_t); 849 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 850 rxid = *tl++; 851 if (*tl != rpc_reply) { 852 #ifndef NFS_V2_ONLY 853 if (nmp->nm_flag & NFSMNT_NQNFS) { 854 if (nqnfs_callback(nmp, mrep, md, dpos, 855 myrep->r_lwp)) 856 nfsstats.rpcinvalid++; 857 } else 858 #endif 859 { 860 nfsstats.rpcinvalid++; 861 m_freem(mrep); 862 } 863 nfsmout: 864 if (myrep->r_flags & R_GETONEREP) 865 return (0); 866 continue; 867 } 868 869 /* 870 * Loop through the request list to match up the reply 871 * Iff no match, just drop the datagram 872 */ 873 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 874 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 875 /* Found it.. */ 876 rep->r_mrep = mrep; 877 rep->r_md = md; 878 rep->r_dpos = dpos; 879 if (nfsrtton) { 880 struct rttl *rt; 881 882 rt = &nfsrtt.rttl[nfsrtt.pos]; 883 rt->proc = rep->r_procnum; 884 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 885 rt->sent = nmp->nm_sent; 886 rt->cwnd = nmp->nm_cwnd; 887 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 888 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 889 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx; 890 rt->tstamp = time; 891 if (rep->r_flags & R_TIMING) 892 rt->rtt = rep->r_rtt; 893 else 894 rt->rtt = 1000000; 895 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 896 } 897 /* 898 * Update congestion window. 899 * Do the additive increase of 900 * one rpc/rtt. 901 */ 902 if (nmp->nm_cwnd <= nmp->nm_sent) { 903 nmp->nm_cwnd += 904 (NFS_CWNDSCALE * NFS_CWNDSCALE + 905 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 906 if (nmp->nm_cwnd > NFS_MAXCWND) 907 nmp->nm_cwnd = NFS_MAXCWND; 908 } 909 rep->r_flags &= ~R_SENT; 910 nmp->nm_sent -= NFS_CWNDSCALE; 911 /* 912 * Update rtt using a gain of 0.125 on the mean 913 * and a gain of 0.25 on the deviation. 914 */ 915 if (rep->r_flags & R_TIMING) { 916 /* 917 * Since the timer resolution of 918 * NFS_HZ is so course, it can often 919 * result in r_rtt == 0. Since 920 * r_rtt == N means that the actual 921 * rtt is between N+dt and N+2-dt ticks, 922 * add 1. 923 */ 924 t1 = rep->r_rtt + 1; 925 t1 -= (NFS_SRTT(rep) >> 3); 926 NFS_SRTT(rep) += t1; 927 if (t1 < 0) 928 t1 = -t1; 929 t1 -= (NFS_SDRTT(rep) >> 2); 930 NFS_SDRTT(rep) += t1; 931 } 932 nmp->nm_timeouts = 0; 933 break; 934 } 935 } 936 /* 937 * If not matched to a request, drop it. 938 * If it's mine, get out. 939 */ 940 if (rep == 0) { 941 nfsstats.rpcunexpected++; 942 m_freem(mrep); 943 } else if (rep == myrep) { 944 if (rep->r_mrep == NULL) 945 panic("nfsreply nil"); 946 return (0); 947 } 948 if (myrep->r_flags & R_GETONEREP) 949 return (0); 950 } 951 } 952 953 /* 954 * nfs_request - goes something like this 955 * - fill in request struct 956 * - links it into list 957 * - calls nfs_send() for first transmit 958 * - calls nfs_receive() to get reply 959 * - break down rpc header and return with nfs reply pointed to 960 * by mrep or error 961 * nb: always frees up mreq mbuf list 962 */ 963 int 964 nfs_request(np, mrest, procnum, lwp, cred, mrp, mdp, dposp, rexmitp) 965 struct nfsnode *np; 966 struct mbuf *mrest; 967 int procnum; 968 struct lwp *lwp; 969 struct ucred *cred; 970 struct mbuf **mrp; 971 struct mbuf **mdp; 972 caddr_t *dposp; 973 int *rexmitp; 974 { 975 struct mbuf *m, *mrep; 976 struct nfsreq *rep; 977 u_int32_t *tl; 978 int i; 979 struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount); 980 struct mbuf *md, *mheadend; 981 char nickv[RPCX_NICKVERF]; 982 time_t reqtime, waituntil; 983 caddr_t dpos, cp2; 984 int t1, s, error = 0, mrest_len, auth_len, auth_type; 985 int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0; 986 int verf_len, verf_type; 987 u_int32_t xid; 988 char *auth_str, *verf_str; 989 NFSKERBKEY_T key; /* save session key */ 990 struct ucred acred; 991 #ifndef NFS_V2_ONLY 992 int nqlflag, cachable; 993 u_quad_t frev; 994 #endif 995 struct mbuf *mrest_backup = NULL; 996 struct ucred *origcred = NULL; /* XXX: gcc */ 997 boolean_t retry_cred = TRUE; 998 boolean_t use_opencred = (np->n_flag & NUSEOPENCRED) != 0; 999 1000 if (rexmitp != NULL) 1001 *rexmitp = 0; 1002 1003 tryagain_cred: 1004 KASSERT(cred != NULL); 1005 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 1006 rep->r_nmp = nmp; 1007 rep->r_lwp = lwp; 1008 rep->r_procnum = procnum; 1009 i = 0; 1010 m = mrest; 1011 while (m) { 1012 i += m->m_len; 1013 m = m->m_next; 1014 } 1015 mrest_len = i; 1016 1017 /* 1018 * Get the RPC header with authorization. 1019 */ 1020 kerbauth: 1021 verf_str = auth_str = (char *)0; 1022 if (nmp->nm_flag & NFSMNT_KERB) { 1023 verf_str = nickv; 1024 verf_len = sizeof (nickv); 1025 auth_type = RPCAUTH_KERB4; 1026 memset((caddr_t)key, 0, sizeof (key)); 1027 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 1028 &auth_len, verf_str, verf_len)) { 1029 error = nfs_getauth(nmp, rep, cred, &auth_str, 1030 &auth_len, verf_str, &verf_len, key); 1031 if (error) { 1032 free((caddr_t)rep, M_NFSREQ); 1033 m_freem(mrest); 1034 return (error); 1035 } 1036 } 1037 retry_cred = FALSE; 1038 } else { 1039 /* AUTH_UNIX */ 1040 uid_t uid; 1041 gid_t gid; 1042 1043 /* 1044 * on the most unix filesystems, permission checks are 1045 * done when the file is open(2)'ed. 1046 * ie. once a file is successfully open'ed, 1047 * following i/o operations never fail with EACCES. 1048 * we try to follow the semantics as far as possible. 1049 * 1050 * note that we expect that the nfs server always grant 1051 * accesses by the file's owner. 1052 */ 1053 origcred = cred; 1054 switch (procnum) { 1055 case NFSPROC_READ: 1056 case NFSPROC_WRITE: 1057 case NFSPROC_COMMIT: 1058 uid = np->n_vattr->va_uid; 1059 gid = np->n_vattr->va_gid; 1060 if (cred->cr_uid == uid && cred->cr_gid == gid) { 1061 retry_cred = FALSE; 1062 break; 1063 } 1064 if (use_opencred) 1065 break; 1066 acred.cr_uid = uid; 1067 acred.cr_gid = gid; 1068 acred.cr_ngroups = 0; 1069 acred.cr_ref = 2; /* Just to be safe.. */ 1070 cred = &acred; 1071 break; 1072 default: 1073 retry_cred = FALSE; 1074 break; 1075 } 1076 /* 1077 * backup mbuf chain if we can need it later to retry. 1078 * 1079 * XXX maybe we can keep a direct reference to 1080 * mrest without doing m_copym, but it's ...ugly. 1081 */ 1082 if (retry_cred) 1083 mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT); 1084 auth_type = RPCAUTH_UNIX; 1085 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 1086 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 1087 5 * NFSX_UNSIGNED; 1088 } 1089 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 1090 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 1091 if (auth_str) 1092 free(auth_str, M_TEMP); 1093 1094 /* 1095 * For stream protocols, insert a Sun RPC Record Mark. 1096 */ 1097 if (nmp->nm_sotype == SOCK_STREAM) { 1098 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 1099 *mtod(m, u_int32_t *) = htonl(0x80000000 | 1100 (m->m_pkthdr.len - NFSX_UNSIGNED)); 1101 } 1102 rep->r_mreq = m; 1103 rep->r_xid = xid; 1104 tryagain: 1105 if (nmp->nm_flag & NFSMNT_SOFT) 1106 rep->r_retry = nmp->nm_retry; 1107 else 1108 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1109 rep->r_rtt = rep->r_rexmit = 0; 1110 if (proct[procnum] > 0) 1111 rep->r_flags = R_TIMING; 1112 else 1113 rep->r_flags = 0; 1114 rep->r_mrep = NULL; 1115 1116 /* 1117 * Do the client side RPC. 1118 */ 1119 nfsstats.rpcrequests++; 1120 /* 1121 * Chain request into list of outstanding requests. Be sure 1122 * to put it LAST so timer finds oldest requests first. 1123 */ 1124 s = splsoftnet(); 1125 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1126 1127 /* Get send time for nqnfs */ 1128 reqtime = time.tv_sec; 1129 1130 /* 1131 * If backing off another request or avoiding congestion, don't 1132 * send this one now but let timer do it. If not timing a request, 1133 * do it now. 1134 */ 1135 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1136 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1137 nmp->nm_sent < nmp->nm_cwnd)) { 1138 splx(s); 1139 if (nmp->nm_soflags & PR_CONNREQUIRED) 1140 error = nfs_sndlock(&nmp->nm_iflag, rep); 1141 if (!error) { 1142 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1143 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp); 1144 if (nmp->nm_soflags & PR_CONNREQUIRED) 1145 nfs_sndunlock(&nmp->nm_iflag); 1146 } 1147 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1148 nmp->nm_sent += NFS_CWNDSCALE; 1149 rep->r_flags |= R_SENT; 1150 } 1151 } else { 1152 splx(s); 1153 rep->r_rtt = -1; 1154 } 1155 1156 /* 1157 * Wait for the reply from our send or the timer's. 1158 */ 1159 if (!error || error == EPIPE) 1160 error = nfs_reply(rep, lwp); 1161 1162 /* 1163 * RPC done, unlink the request. 1164 */ 1165 s = splsoftnet(); 1166 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1167 splx(s); 1168 1169 /* 1170 * Decrement the outstanding request count. 1171 */ 1172 if (rep->r_flags & R_SENT) { 1173 rep->r_flags &= ~R_SENT; /* paranoia */ 1174 nmp->nm_sent -= NFS_CWNDSCALE; 1175 } 1176 1177 if (rexmitp != NULL) { 1178 int rexmit; 1179 1180 if (nmp->nm_sotype != SOCK_DGRAM) 1181 rexmit = (rep->r_flags & R_REXMITTED) != 0; 1182 else 1183 rexmit = rep->r_rexmit; 1184 *rexmitp = rexmit; 1185 } 1186 1187 /* 1188 * If there was a successful reply and a tprintf msg. 1189 * tprintf a response. 1190 */ 1191 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1192 nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1193 "is alive again"); 1194 mrep = rep->r_mrep; 1195 md = rep->r_md; 1196 dpos = rep->r_dpos; 1197 if (error) 1198 goto nfsmout; 1199 1200 /* 1201 * break down the rpc header and check if ok 1202 */ 1203 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1204 if (*tl++ == rpc_msgdenied) { 1205 if (*tl == rpc_mismatch) 1206 error = EOPNOTSUPP; 1207 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1208 if (!failed_auth) { 1209 failed_auth++; 1210 mheadend->m_next = (struct mbuf *)0; 1211 m_freem(mrep); 1212 m_freem(rep->r_mreq); 1213 goto kerbauth; 1214 } else 1215 error = EAUTH; 1216 } else 1217 error = EACCES; 1218 m_freem(mrep); 1219 goto nfsmout; 1220 } 1221 1222 /* 1223 * Grab any Kerberos verifier, otherwise just throw it away. 1224 */ 1225 verf_type = fxdr_unsigned(int, *tl++); 1226 i = fxdr_unsigned(int32_t, *tl); 1227 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1228 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1229 if (error) 1230 goto nfsmout; 1231 } else if (i > 0) 1232 nfsm_adv(nfsm_rndup(i)); 1233 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1234 /* 0 == ok */ 1235 if (*tl == 0) { 1236 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1237 if (*tl != 0) { 1238 error = fxdr_unsigned(int, *tl); 1239 switch (error) { 1240 case NFSERR_PERM: 1241 error = EPERM; 1242 break; 1243 1244 case NFSERR_NOENT: 1245 error = ENOENT; 1246 break; 1247 1248 case NFSERR_IO: 1249 error = EIO; 1250 break; 1251 1252 case NFSERR_NXIO: 1253 error = ENXIO; 1254 break; 1255 1256 case NFSERR_ACCES: 1257 error = EACCES; 1258 if (!retry_cred) 1259 break; 1260 m_freem(mrep); 1261 m_freem(rep->r_mreq); 1262 FREE(rep, M_NFSREQ); 1263 use_opencred = !use_opencred; 1264 if (mrest_backup == NULL) 1265 /* m_copym failure */ 1266 return ENOMEM; 1267 mrest = mrest_backup; 1268 mrest_backup = NULL; 1269 cred = origcred; 1270 error = 0; 1271 retry_cred = FALSE; 1272 goto tryagain_cred; 1273 1274 case NFSERR_EXIST: 1275 error = EEXIST; 1276 break; 1277 1278 case NFSERR_XDEV: 1279 error = EXDEV; 1280 break; 1281 1282 case NFSERR_NODEV: 1283 error = ENODEV; 1284 break; 1285 1286 case NFSERR_NOTDIR: 1287 error = ENOTDIR; 1288 break; 1289 1290 case NFSERR_ISDIR: 1291 error = EISDIR; 1292 break; 1293 1294 case NFSERR_INVAL: 1295 error = EINVAL; 1296 break; 1297 1298 case NFSERR_FBIG: 1299 error = EFBIG; 1300 break; 1301 1302 case NFSERR_NOSPC: 1303 error = ENOSPC; 1304 break; 1305 1306 case NFSERR_ROFS: 1307 error = EROFS; 1308 break; 1309 1310 case NFSERR_MLINK: 1311 error = EMLINK; 1312 break; 1313 1314 case NFSERR_TIMEDOUT: 1315 error = ETIMEDOUT; 1316 break; 1317 1318 case NFSERR_NAMETOL: 1319 error = ENAMETOOLONG; 1320 break; 1321 1322 case NFSERR_NOTEMPTY: 1323 error = ENOTEMPTY; 1324 break; 1325 1326 case NFSERR_DQUOT: 1327 error = EDQUOT; 1328 break; 1329 1330 case NFSERR_STALE: 1331 /* 1332 * If the File Handle was stale, invalidate the 1333 * lookup cache, just in case. 1334 */ 1335 error = ESTALE; 1336 cache_purge(NFSTOV(np)); 1337 break; 1338 1339 case NFSERR_REMOTE: 1340 error = EREMOTE; 1341 break; 1342 1343 case NFSERR_WFLUSH: 1344 case NFSERR_BADHANDLE: 1345 case NFSERR_NOT_SYNC: 1346 case NFSERR_BAD_COOKIE: 1347 error = EINVAL; 1348 break; 1349 1350 case NFSERR_NOTSUPP: 1351 error = ENOTSUP; 1352 break; 1353 1354 case NFSERR_TOOSMALL: 1355 case NFSERR_SERVERFAULT: 1356 case NFSERR_BADTYPE: 1357 error = EINVAL; 1358 break; 1359 1360 case NFSERR_TRYLATER: 1361 if ((nmp->nm_flag & NFSMNT_NFSV3) == 0) 1362 break; 1363 m_freem(mrep); 1364 error = 0; 1365 waituntil = time.tv_sec + trylater_delay; 1366 while (time.tv_sec < waituntil) 1367 (void) tsleep((caddr_t)&lbolt, 1368 PSOCK, "nqnfstry", 0); 1369 trylater_delay *= NFS_TRYLATERDELMUL; 1370 if (trylater_delay > NFS_TRYLATERDELMAX) 1371 trylater_delay = NFS_TRYLATERDELMAX; 1372 /* 1373 * RFC1813: 1374 * The client should wait and then try 1375 * the request with a new RPC transaction ID. 1376 */ 1377 nfs_renewxid(rep); 1378 goto tryagain; 1379 1380 case NFSERR_STALEWRITEVERF: 1381 error = EINVAL; 1382 break; 1383 1384 default: 1385 #ifdef DIAGNOSTIC 1386 printf("Invalid rpc error code %d\n", error); 1387 #endif 1388 error = EINVAL; 1389 break; 1390 } 1391 1392 if (nmp->nm_flag & NFSMNT_NFSV3) { 1393 *mrp = mrep; 1394 *mdp = md; 1395 *dposp = dpos; 1396 error |= NFSERR_RETERR; 1397 } else 1398 m_freem(mrep); 1399 goto nfsmout; 1400 } 1401 1402 /* 1403 * note which credential worked to minimize number of retries. 1404 */ 1405 if (use_opencred) 1406 np->n_flag |= NUSEOPENCRED; 1407 else 1408 np->n_flag &= ~NUSEOPENCRED; 1409 1410 #ifndef NFS_V2_ONLY 1411 /* 1412 * For nqnfs, get any lease in reply 1413 */ 1414 if (nmp->nm_flag & NFSMNT_NQNFS) { 1415 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1416 if (*tl) { 1417 nqlflag = fxdr_unsigned(int, *tl); 1418 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1419 cachable = fxdr_unsigned(int, *tl++); 1420 reqtime += fxdr_unsigned(int, *tl++); 1421 if (reqtime > time.tv_sec) { 1422 frev = fxdr_hyper(tl); 1423 nqnfs_clientlease(nmp, np, nqlflag, 1424 cachable, reqtime, frev); 1425 } 1426 } 1427 } 1428 #endif 1429 *mrp = mrep; 1430 *mdp = md; 1431 *dposp = dpos; 1432 1433 KASSERT(error == 0); 1434 goto nfsmout; 1435 } 1436 m_freem(mrep); 1437 error = EPROTONOSUPPORT; 1438 nfsmout: 1439 m_freem(rep->r_mreq); 1440 free((caddr_t)rep, M_NFSREQ); 1441 m_freem(mrest_backup); 1442 return (error); 1443 } 1444 #endif /* NFS */ 1445 1446 /* 1447 * Generate the rpc reply header 1448 * siz arg. is used to decide if adding a cluster is worthwhile 1449 */ 1450 int 1451 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1452 int siz; 1453 struct nfsrv_descript *nd; 1454 struct nfssvc_sock *slp; 1455 int err; 1456 int cache; 1457 u_quad_t *frev; 1458 struct mbuf **mrq; 1459 struct mbuf **mbp; 1460 caddr_t *bposp; 1461 { 1462 u_int32_t *tl; 1463 struct mbuf *mreq; 1464 caddr_t bpos; 1465 struct mbuf *mb; 1466 1467 mreq = m_gethdr(M_WAIT, MT_DATA); 1468 MCLAIM(mreq, &nfs_mowner); 1469 mb = mreq; 1470 /* 1471 * If this is a big reply, use a cluster else 1472 * try and leave leading space for the lower level headers. 1473 */ 1474 siz += RPC_REPLYSIZ; 1475 if (siz >= max_datalen) { 1476 m_clget(mreq, M_WAIT); 1477 } else 1478 mreq->m_data += max_hdr; 1479 tl = mtod(mreq, u_int32_t *); 1480 mreq->m_len = 6 * NFSX_UNSIGNED; 1481 bpos = ((caddr_t)tl) + mreq->m_len; 1482 *tl++ = txdr_unsigned(nd->nd_retxid); 1483 *tl++ = rpc_reply; 1484 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1485 *tl++ = rpc_msgdenied; 1486 if (err & NFSERR_AUTHERR) { 1487 *tl++ = rpc_autherr; 1488 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1489 mreq->m_len -= NFSX_UNSIGNED; 1490 bpos -= NFSX_UNSIGNED; 1491 } else { 1492 *tl++ = rpc_mismatch; 1493 *tl++ = txdr_unsigned(RPC_VER2); 1494 *tl = txdr_unsigned(RPC_VER2); 1495 } 1496 } else { 1497 *tl++ = rpc_msgaccepted; 1498 1499 /* 1500 * For Kerberos authentication, we must send the nickname 1501 * verifier back, otherwise just RPCAUTH_NULL. 1502 */ 1503 if (nd->nd_flag & ND_KERBFULL) { 1504 struct nfsuid *nuidp; 1505 struct timeval ktvin, ktvout; 1506 1507 LIST_FOREACH(nuidp, NUIDHASH(slp, nd->nd_cr.cr_uid), 1508 nu_hash) { 1509 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1510 (!nd->nd_nam2 || netaddr_match( 1511 NU_NETFAM(nuidp), &nuidp->nu_haddr, 1512 nd->nd_nam2))) 1513 break; 1514 } 1515 if (nuidp) { 1516 ktvin.tv_sec = 1517 txdr_unsigned(nuidp->nu_timestamp.tv_sec 1518 - 1); 1519 ktvin.tv_usec = 1520 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1521 1522 /* 1523 * Encrypt the timestamp in ecb mode using the 1524 * session key. 1525 */ 1526 #ifdef NFSKERB 1527 XXX 1528 #endif 1529 1530 *tl++ = rpc_auth_kerb; 1531 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1532 *tl = ktvout.tv_sec; 1533 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1534 *tl++ = ktvout.tv_usec; 1535 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1536 } else { 1537 *tl++ = 0; 1538 *tl++ = 0; 1539 } 1540 } else { 1541 *tl++ = 0; 1542 *tl++ = 0; 1543 } 1544 switch (err) { 1545 case EPROGUNAVAIL: 1546 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1547 break; 1548 case EPROGMISMATCH: 1549 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1550 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1551 if (nd->nd_flag & ND_NQNFS) { 1552 *tl++ = txdr_unsigned(3); 1553 *tl = txdr_unsigned(3); 1554 } else { 1555 *tl++ = txdr_unsigned(2); 1556 *tl = txdr_unsigned(3); 1557 } 1558 break; 1559 case EPROCUNAVAIL: 1560 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1561 break; 1562 case EBADRPC: 1563 *tl = txdr_unsigned(RPC_GARBAGE); 1564 break; 1565 default: 1566 *tl = 0; 1567 if (err != NFSERR_RETVOID) { 1568 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1569 if (err) 1570 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1571 else 1572 *tl = 0; 1573 } 1574 break; 1575 }; 1576 } 1577 1578 /* 1579 * For nqnfs, piggyback lease as requested. 1580 */ 1581 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1582 if (nd->nd_flag & ND_LEASE) { 1583 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1584 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1585 *tl++ = txdr_unsigned(cache); 1586 *tl++ = txdr_unsigned(nd->nd_duration); 1587 txdr_hyper(*frev, tl); 1588 } else { 1589 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1590 *tl = 0; 1591 } 1592 } 1593 if (mrq != NULL) 1594 *mrq = mreq; 1595 *mbp = mb; 1596 *bposp = bpos; 1597 if (err != 0 && err != NFSERR_RETVOID) 1598 nfsstats.srvrpc_errs++; 1599 return (0); 1600 } 1601 1602 /* 1603 * Nfs timer routine 1604 * Scan the nfsreq list and retranmit any requests that have timed out 1605 * To avoid retransmission attempts on STREAM sockets (in the future) make 1606 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1607 */ 1608 void 1609 nfs_timer(arg) 1610 void *arg; /* never used */ 1611 { 1612 struct nfsreq *rep; 1613 struct mbuf *m; 1614 struct socket *so; 1615 struct nfsmount *nmp; 1616 int timeo; 1617 int s, error; 1618 #ifdef NFSSERVER 1619 struct nfssvc_sock *slp; 1620 static long lasttime = 0; 1621 u_quad_t cur_usec; 1622 #endif 1623 1624 s = splsoftnet(); 1625 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1626 nmp = rep->r_nmp; 1627 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1628 continue; 1629 if (nfs_sigintr(nmp, rep, rep->r_lwp)) { 1630 rep->r_flags |= R_SOFTTERM; 1631 continue; 1632 } 1633 if (rep->r_rtt >= 0) { 1634 rep->r_rtt++; 1635 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1636 timeo = nmp->nm_timeo; 1637 else 1638 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1639 if (nmp->nm_timeouts > 0) 1640 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1641 if (rep->r_rtt <= timeo) 1642 continue; 1643 if (nmp->nm_timeouts < 1644 (sizeof(nfs_backoff) / sizeof(nfs_backoff[0]))) 1645 nmp->nm_timeouts++; 1646 } 1647 /* 1648 * Check for server not responding 1649 */ 1650 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1651 rep->r_rexmit > nmp->nm_deadthresh) { 1652 nfs_msg(rep->r_lwp, 1653 nmp->nm_mountp->mnt_stat.f_mntfromname, 1654 "not responding"); 1655 rep->r_flags |= R_TPRINTFMSG; 1656 } 1657 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1658 nfsstats.rpctimeouts++; 1659 rep->r_flags |= R_SOFTTERM; 1660 continue; 1661 } 1662 if (nmp->nm_sotype != SOCK_DGRAM) { 1663 if (++rep->r_rexmit > NFS_MAXREXMIT) 1664 rep->r_rexmit = NFS_MAXREXMIT; 1665 continue; 1666 } 1667 if ((so = nmp->nm_so) == NULL) 1668 continue; 1669 1670 /* 1671 * If there is enough space and the window allows.. 1672 * Resend it 1673 * Set r_rtt to -1 in case we fail to send it now. 1674 */ 1675 rep->r_rtt = -1; 1676 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1677 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1678 (rep->r_flags & R_SENT) || 1679 nmp->nm_sent < nmp->nm_cwnd) && 1680 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1681 if (so->so_state & SS_ISCONNECTED) 1682 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1683 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1684 else 1685 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1686 nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0); 1687 if (error) { 1688 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1689 #ifdef DEBUG 1690 printf("nfs_timer: ignoring error %d\n", 1691 error); 1692 #endif 1693 so->so_error = 0; 1694 } 1695 } else { 1696 /* 1697 * Iff first send, start timing 1698 * else turn timing off, backoff timer 1699 * and divide congestion window by 2. 1700 */ 1701 if (rep->r_flags & R_SENT) { 1702 rep->r_flags &= ~R_TIMING; 1703 if (++rep->r_rexmit > NFS_MAXREXMIT) 1704 rep->r_rexmit = NFS_MAXREXMIT; 1705 nmp->nm_cwnd >>= 1; 1706 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1707 nmp->nm_cwnd = NFS_CWNDSCALE; 1708 nfsstats.rpcretries++; 1709 } else { 1710 rep->r_flags |= R_SENT; 1711 nmp->nm_sent += NFS_CWNDSCALE; 1712 } 1713 rep->r_rtt = 0; 1714 } 1715 } 1716 } 1717 1718 #ifdef NFSSERVER 1719 /* 1720 * Call the nqnfs server timer once a second to handle leases. 1721 */ 1722 if (lasttime != time.tv_sec) { 1723 lasttime = time.tv_sec; 1724 nqnfs_serverd(); 1725 } 1726 1727 /* 1728 * Scan the write gathering queues for writes that need to be 1729 * completed now. 1730 */ 1731 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1732 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1733 if (LIST_FIRST(&slp->ns_tq) && 1734 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) 1735 nfsrv_wakenfsd(slp); 1736 } 1737 #endif /* NFSSERVER */ 1738 splx(s); 1739 callout_schedule(&nfs_timer_ch, nfs_ticks); 1740 } 1741 1742 /*ARGSUSED*/ 1743 void 1744 nfs_exit(p, v) 1745 struct proc *p; 1746 void *v; 1747 { 1748 struct nfsreq *rp; 1749 int s = splsoftnet(); 1750 1751 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 1752 if (rp->r_lwp && rp->r_lwp->l_proc == p) 1753 TAILQ_REMOVE(&nfs_reqq, rp, r_chain); 1754 } 1755 splx(s); 1756 } 1757 1758 /* 1759 * Test for a termination condition pending on the process. 1760 * This is used for NFSMNT_INT mounts. 1761 */ 1762 int 1763 nfs_sigintr(nmp, rep, l) 1764 struct nfsmount *nmp; 1765 struct nfsreq *rep; 1766 struct lwp *l; 1767 { 1768 sigset_t ss; 1769 1770 if (rep && (rep->r_flags & R_SOFTTERM)) 1771 return (EINTR); 1772 if (!(nmp->nm_flag & NFSMNT_INT)) 1773 return (0); 1774 if (l) { 1775 sigpending1(l->l_proc, &ss); 1776 #if 0 1777 sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss); 1778 #endif 1779 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1780 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1781 sigismember(&ss, SIGQUIT)) 1782 return (EINTR); 1783 } 1784 return (0); 1785 } 1786 1787 /* 1788 * Lock a socket against others. 1789 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1790 * and also to avoid race conditions between the processes with nfs requests 1791 * in progress when a reconnect is necessary. 1792 */ 1793 int 1794 nfs_sndlock(flagp, rep) 1795 int *flagp; 1796 struct nfsreq *rep; 1797 { 1798 struct lwp *l; 1799 int slpflag = 0, slptimeo = 0; 1800 1801 if (rep) { 1802 l = rep->r_lwp; 1803 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1804 slpflag = PCATCH; 1805 } else 1806 l = (struct lwp *)0; 1807 while (*flagp & NFSMNT_SNDLOCK) { 1808 if (rep && nfs_sigintr(rep->r_nmp, rep, l)) 1809 return (EINTR); 1810 *flagp |= NFSMNT_WANTSND; 1811 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1812 slptimeo); 1813 if (slpflag == PCATCH) { 1814 slpflag = 0; 1815 slptimeo = 2 * hz; 1816 } 1817 } 1818 *flagp |= NFSMNT_SNDLOCK; 1819 return (0); 1820 } 1821 1822 /* 1823 * Unlock the stream socket for others. 1824 */ 1825 void 1826 nfs_sndunlock(flagp) 1827 int *flagp; 1828 { 1829 1830 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1831 panic("nfs sndunlock"); 1832 *flagp &= ~NFSMNT_SNDLOCK; 1833 if (*flagp & NFSMNT_WANTSND) { 1834 *flagp &= ~NFSMNT_WANTSND; 1835 wakeup((caddr_t)flagp); 1836 } 1837 } 1838 1839 int 1840 nfs_rcvlock(rep) 1841 struct nfsreq *rep; 1842 { 1843 struct nfsmount *nmp = rep->r_nmp; 1844 int *flagp = &nmp->nm_iflag; 1845 int slpflag, slptimeo = 0; 1846 int error = 0; 1847 1848 if (*flagp & NFSMNT_DISMNT) 1849 return EIO; 1850 1851 if (*flagp & NFSMNT_INT) 1852 slpflag = PCATCH; 1853 else 1854 slpflag = 0; 1855 simple_lock(&nmp->nm_slock); 1856 while (*flagp & NFSMNT_RCVLOCK) { 1857 if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) { 1858 error = EINTR; 1859 goto quit; 1860 } 1861 *flagp |= NFSMNT_WANTRCV; 1862 nmp->nm_waiters++; 1863 (void) ltsleep(flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1864 slptimeo, &nmp->nm_slock); 1865 nmp->nm_waiters--; 1866 if (*flagp & NFSMNT_DISMNT) { 1867 wakeup(&nmp->nm_waiters); 1868 error = EIO; 1869 goto quit; 1870 } 1871 /* If our reply was received while we were sleeping, 1872 * then just return without taking the lock to avoid a 1873 * situation where a single iod could 'capture' the 1874 * receive lock. 1875 */ 1876 if (rep->r_mrep != NULL) { 1877 error = EALREADY; 1878 goto quit; 1879 } 1880 if (slpflag == PCATCH) { 1881 slpflag = 0; 1882 slptimeo = 2 * hz; 1883 } 1884 } 1885 *flagp |= NFSMNT_RCVLOCK; 1886 quit: 1887 simple_unlock(&nmp->nm_slock); 1888 return error; 1889 } 1890 1891 /* 1892 * Unlock the stream socket for others. 1893 */ 1894 void 1895 nfs_rcvunlock(nmp) 1896 struct nfsmount *nmp; 1897 { 1898 int *flagp = &nmp->nm_iflag; 1899 1900 simple_lock(&nmp->nm_slock); 1901 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1902 panic("nfs rcvunlock"); 1903 *flagp &= ~NFSMNT_RCVLOCK; 1904 if (*flagp & NFSMNT_WANTRCV) { 1905 *flagp &= ~NFSMNT_WANTRCV; 1906 wakeup((caddr_t)flagp); 1907 } 1908 simple_unlock(&nmp->nm_slock); 1909 } 1910 1911 /* 1912 * Parse an RPC request 1913 * - verify it 1914 * - fill in the cred struct. 1915 */ 1916 int 1917 nfs_getreq(nd, nfsd, has_header) 1918 struct nfsrv_descript *nd; 1919 struct nfsd *nfsd; 1920 int has_header; 1921 { 1922 int len, i; 1923 u_int32_t *tl; 1924 int32_t t1; 1925 struct uio uio; 1926 struct iovec iov; 1927 caddr_t dpos, cp2, cp; 1928 u_int32_t nfsvers, auth_type; 1929 uid_t nickuid; 1930 int error = 0, nqnfs = 0, ticklen; 1931 struct mbuf *mrep, *md; 1932 struct nfsuid *nuidp; 1933 struct timeval tvin, tvout; 1934 1935 mrep = nd->nd_mrep; 1936 md = nd->nd_md; 1937 dpos = nd->nd_dpos; 1938 if (has_header) { 1939 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1940 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1941 if (*tl++ != rpc_call) { 1942 m_freem(mrep); 1943 return (EBADRPC); 1944 } 1945 } else 1946 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1947 nd->nd_repstat = 0; 1948 nd->nd_flag = 0; 1949 if (*tl++ != rpc_vers) { 1950 nd->nd_repstat = ERPCMISMATCH; 1951 nd->nd_procnum = NFSPROC_NOOP; 1952 return (0); 1953 } 1954 if (*tl != nfs_prog) { 1955 if (*tl == nqnfs_prog) 1956 nqnfs++; 1957 else { 1958 nd->nd_repstat = EPROGUNAVAIL; 1959 nd->nd_procnum = NFSPROC_NOOP; 1960 return (0); 1961 } 1962 } 1963 tl++; 1964 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1965 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1966 (nfsvers != NQNFS_VER3 && nqnfs)) { 1967 nd->nd_repstat = EPROGMISMATCH; 1968 nd->nd_procnum = NFSPROC_NOOP; 1969 return (0); 1970 } 1971 if (nqnfs) 1972 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1973 else if (nfsvers == NFS_VER3) 1974 nd->nd_flag = ND_NFSV3; 1975 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1976 if (nd->nd_procnum == NFSPROC_NULL) 1977 return (0); 1978 if (nd->nd_procnum >= NFS_NPROCS || 1979 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1980 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1981 nd->nd_repstat = EPROCUNAVAIL; 1982 nd->nd_procnum = NFSPROC_NOOP; 1983 return (0); 1984 } 1985 if ((nd->nd_flag & ND_NFSV3) == 0) 1986 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1987 auth_type = *tl++; 1988 len = fxdr_unsigned(int, *tl++); 1989 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1990 m_freem(mrep); 1991 return (EBADRPC); 1992 } 1993 1994 nd->nd_flag &= ~ND_KERBAUTH; 1995 /* 1996 * Handle auth_unix or auth_kerb. 1997 */ 1998 if (auth_type == rpc_auth_unix) { 1999 len = fxdr_unsigned(int, *++tl); 2000 if (len < 0 || len > NFS_MAXNAMLEN) { 2001 m_freem(mrep); 2002 return (EBADRPC); 2003 } 2004 nfsm_adv(nfsm_rndup(len)); 2005 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2006 memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred)); 2007 nd->nd_cr.cr_ref = 1; 2008 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 2009 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 2010 len = fxdr_unsigned(int, *tl); 2011 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 2012 m_freem(mrep); 2013 return (EBADRPC); 2014 } 2015 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 2016 for (i = 0; i < len; i++) 2017 if (i < NGROUPS) 2018 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 2019 else 2020 tl++; 2021 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 2022 if (nd->nd_cr.cr_ngroups > 1) 2023 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 2024 len = fxdr_unsigned(int, *++tl); 2025 if (len < 0 || len > RPCAUTH_MAXSIZ) { 2026 m_freem(mrep); 2027 return (EBADRPC); 2028 } 2029 if (len > 0) 2030 nfsm_adv(nfsm_rndup(len)); 2031 } else if (auth_type == rpc_auth_kerb) { 2032 switch (fxdr_unsigned(int, *tl++)) { 2033 case RPCAKN_FULLNAME: 2034 ticklen = fxdr_unsigned(int, *tl); 2035 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 2036 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 2037 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 2038 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 2039 m_freem(mrep); 2040 return (EBADRPC); 2041 } 2042 uio.uio_offset = 0; 2043 uio.uio_iov = &iov; 2044 uio.uio_iovcnt = 1; 2045 uio.uio_segflg = UIO_SYSSPACE; 2046 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 2047 iov.iov_len = RPCAUTH_MAXSIZ - 4; 2048 nfsm_mtouio(&uio, uio.uio_resid); 2049 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2050 if (*tl++ != rpc_auth_kerb || 2051 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 2052 printf("Bad kerb verifier\n"); 2053 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2054 nd->nd_procnum = NFSPROC_NOOP; 2055 return (0); 2056 } 2057 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 2058 tl = (u_int32_t *)cp; 2059 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 2060 printf("Not fullname kerb verifier\n"); 2061 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2062 nd->nd_procnum = NFSPROC_NOOP; 2063 return (0); 2064 } 2065 cp += NFSX_UNSIGNED; 2066 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 2067 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 2068 nd->nd_flag |= ND_KERBFULL; 2069 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 2070 break; 2071 case RPCAKN_NICKNAME: 2072 if (len != 2 * NFSX_UNSIGNED) { 2073 printf("Kerb nickname short\n"); 2074 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 2075 nd->nd_procnum = NFSPROC_NOOP; 2076 return (0); 2077 } 2078 nickuid = fxdr_unsigned(uid_t, *tl); 2079 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2080 if (*tl++ != rpc_auth_kerb || 2081 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 2082 printf("Kerb nick verifier bad\n"); 2083 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2084 nd->nd_procnum = NFSPROC_NOOP; 2085 return (0); 2086 } 2087 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2088 tvin.tv_sec = *tl++; 2089 tvin.tv_usec = *tl; 2090 2091 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), 2092 nu_hash) { 2093 if (nuidp->nu_cr.cr_uid == nickuid && 2094 (!nd->nd_nam2 || 2095 netaddr_match(NU_NETFAM(nuidp), 2096 &nuidp->nu_haddr, nd->nd_nam2))) 2097 break; 2098 } 2099 if (!nuidp) { 2100 nd->nd_repstat = 2101 (NFSERR_AUTHERR|AUTH_REJECTCRED); 2102 nd->nd_procnum = NFSPROC_NOOP; 2103 return (0); 2104 } 2105 2106 /* 2107 * Now, decrypt the timestamp using the session key 2108 * and validate it. 2109 */ 2110 #ifdef NFSKERB 2111 XXX 2112 #endif 2113 2114 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 2115 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 2116 if (nuidp->nu_expire < time.tv_sec || 2117 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 2118 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 2119 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 2120 nuidp->nu_expire = 0; 2121 nd->nd_repstat = 2122 (NFSERR_AUTHERR|AUTH_REJECTVERF); 2123 nd->nd_procnum = NFSPROC_NOOP; 2124 return (0); 2125 } 2126 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 2127 nd->nd_flag |= ND_KERBNICK; 2128 }; 2129 } else { 2130 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 2131 nd->nd_procnum = NFSPROC_NOOP; 2132 return (0); 2133 } 2134 2135 /* 2136 * For nqnfs, get piggybacked lease request. 2137 */ 2138 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 2139 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2140 nd->nd_flag |= fxdr_unsigned(int, *tl); 2141 if (nd->nd_flag & ND_LEASE) { 2142 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2143 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 2144 } else 2145 nd->nd_duration = NQ_MINLEASE; 2146 } else 2147 nd->nd_duration = NQ_MINLEASE; 2148 nd->nd_md = md; 2149 nd->nd_dpos = dpos; 2150 return (0); 2151 nfsmout: 2152 return (error); 2153 } 2154 2155 int 2156 nfs_msg(l, server, msg) 2157 struct lwp *l; 2158 const char *server, *msg; 2159 { 2160 tpr_t tpr; 2161 2162 if (l) 2163 tpr = tprintf_open(l->l_proc); 2164 else 2165 tpr = NULL; 2166 tprintf(tpr, "nfs server %s: %s\n", server, msg); 2167 tprintf_close(tpr); 2168 return (0); 2169 } 2170 2171 #ifdef NFSSERVER 2172 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 2173 struct nfssvc_sock *, struct lwp *, 2174 struct mbuf **)) = { 2175 nfsrv_null, 2176 nfsrv_getattr, 2177 nfsrv_setattr, 2178 nfsrv_lookup, 2179 nfsrv3_access, 2180 nfsrv_readlink, 2181 nfsrv_read, 2182 nfsrv_write, 2183 nfsrv_create, 2184 nfsrv_mkdir, 2185 nfsrv_symlink, 2186 nfsrv_mknod, 2187 nfsrv_remove, 2188 nfsrv_rmdir, 2189 nfsrv_rename, 2190 nfsrv_link, 2191 nfsrv_readdir, 2192 nfsrv_readdirplus, 2193 nfsrv_statfs, 2194 nfsrv_fsinfo, 2195 nfsrv_pathconf, 2196 nfsrv_commit, 2197 nqnfsrv_getlease, 2198 nqnfsrv_vacated, 2199 nfsrv_noop, 2200 nfsrv_noop 2201 }; 2202 2203 /* 2204 * Socket upcall routine for the nfsd sockets. 2205 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 2206 * Essentially do as much as possible non-blocking, else punt and it will 2207 * be called with M_WAIT from an nfsd. 2208 */ 2209 void 2210 nfsrv_rcv(so, arg, waitflag) 2211 struct socket *so; 2212 caddr_t arg; 2213 int waitflag; 2214 { 2215 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 2216 struct mbuf *m; 2217 struct mbuf *mp, *nam; 2218 struct uio auio; 2219 int flags, error; 2220 int setflags = 0; 2221 2222 error = nfsdsock_lock(slp, (waitflag != M_DONTWAIT)); 2223 if (error) { 2224 setflags |= SLP_NEEDQ; 2225 goto dorecs_unlocked; 2226 } 2227 2228 KASSERT(so == slp->ns_so); 2229 #if 1 2230 /* 2231 * Define this to test for nfsds handling this under heavy load. 2232 * 2233 * XXX it isn't safe to call so_receive from so_upcall context. 2234 */ 2235 if (waitflag == M_DONTWAIT) { 2236 setflags |= SLP_NEEDQ; 2237 goto dorecs; 2238 } 2239 #endif 2240 /* XXX: was NULL, soreceive() requires non-NULL uio->uio_lwp */ 2241 auio.uio_lwp = curlwp; /* XXX curlwp */ 2242 simple_lock(&slp->ns_lock); 2243 slp->ns_flag &= ~SLP_NEEDQ; 2244 simple_unlock(&slp->ns_lock); 2245 if (so->so_type == SOCK_STREAM) { 2246 /* 2247 * If there are already records on the queue, defer soreceive() 2248 * to an nfsd so that there is feedback to the TCP layer that 2249 * the nfs servers are heavily loaded. 2250 */ 2251 if (slp->ns_rec && waitflag == M_DONTWAIT) { 2252 setflags |= SLP_NEEDQ; 2253 goto dorecs; 2254 } 2255 2256 /* 2257 * Do soreceive(). 2258 */ 2259 auio.uio_resid = 1000000000; 2260 flags = MSG_DONTWAIT; 2261 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, &flags); 2262 if (error || mp == NULL) { 2263 if (error == EWOULDBLOCK) 2264 setflags |= SLP_NEEDQ; 2265 else 2266 setflags |= SLP_DISCONN; 2267 goto dorecs; 2268 } 2269 m = mp; 2270 if (slp->ns_rawend) { 2271 slp->ns_rawend->m_next = m; 2272 slp->ns_cc += 1000000000 - auio.uio_resid; 2273 } else { 2274 slp->ns_raw = m; 2275 slp->ns_cc = 1000000000 - auio.uio_resid; 2276 } 2277 while (m->m_next) 2278 m = m->m_next; 2279 slp->ns_rawend = m; 2280 2281 /* 2282 * Now try and parse record(s) out of the raw stream data. 2283 */ 2284 error = nfsrv_getstream(slp, waitflag); 2285 if (error) { 2286 if (error == EPERM) 2287 setflags |= SLP_DISCONN; 2288 else 2289 setflags |= SLP_NEEDQ; 2290 } 2291 } else { 2292 do { 2293 auio.uio_resid = 1000000000; 2294 flags = MSG_DONTWAIT; 2295 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, 2296 &flags); 2297 if (mp) { 2298 if (nam) { 2299 m = nam; 2300 m->m_next = mp; 2301 } else 2302 m = mp; 2303 if (slp->ns_recend) 2304 slp->ns_recend->m_nextpkt = m; 2305 else 2306 slp->ns_rec = m; 2307 slp->ns_recend = m; 2308 m->m_nextpkt = (struct mbuf *)0; 2309 } 2310 if (error) { 2311 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2312 && error != EWOULDBLOCK) { 2313 setflags |= SLP_DISCONN; 2314 goto dorecs; 2315 } 2316 } 2317 } while (mp); 2318 } 2319 dorecs: 2320 nfsdsock_unlock(slp); 2321 2322 dorecs_unlocked: 2323 /* 2324 * Now try and process the request records, non-blocking. 2325 */ 2326 if (setflags) { 2327 simple_lock(&slp->ns_lock); 2328 slp->ns_flag |= setflags; 2329 simple_unlock(&slp->ns_lock); 2330 } 2331 if (waitflag == M_DONTWAIT && 2332 (slp->ns_rec || (slp->ns_flag & (SLP_DISCONN | SLP_NEEDQ)) != 0)) { 2333 nfsrv_wakenfsd(slp); 2334 } 2335 } 2336 2337 int 2338 nfsdsock_lock(struct nfssvc_sock *slp, boolean_t waitok) 2339 { 2340 2341 simple_lock(&slp->ns_lock); 2342 while ((slp->ns_flag & (SLP_BUSY|SLP_VALID)) == SLP_BUSY) { 2343 if (!waitok) { 2344 simple_unlock(&slp->ns_lock); 2345 return EWOULDBLOCK; 2346 } 2347 slp->ns_flag |= SLP_WANT; 2348 ltsleep(&slp->ns_flag, PSOCK, "nslock", 0, &slp->ns_lock); 2349 } 2350 if ((slp->ns_flag & SLP_VALID) == 0) { 2351 simple_unlock(&slp->ns_lock); 2352 return EINVAL; 2353 } 2354 slp->ns_flag |= SLP_BUSY; 2355 simple_unlock(&slp->ns_lock); 2356 2357 return 0; 2358 } 2359 2360 void 2361 nfsdsock_unlock(struct nfssvc_sock *slp) 2362 { 2363 2364 KASSERT((slp->ns_flag & SLP_BUSY) != 0); 2365 2366 simple_lock(&slp->ns_lock); 2367 if ((slp->ns_flag & SLP_WANT) != 0) { 2368 wakeup(&slp->ns_flag); 2369 } 2370 slp->ns_flag &= ~(SLP_BUSY|SLP_WANT); 2371 simple_unlock(&slp->ns_lock); 2372 } 2373 2374 int 2375 nfsdsock_drain(struct nfssvc_sock *slp) 2376 { 2377 int error = 0; 2378 2379 simple_lock(&slp->ns_lock); 2380 if ((slp->ns_flag & SLP_VALID) == 0) { 2381 error = EINVAL; 2382 goto done; 2383 } 2384 slp->ns_flag &= ~SLP_VALID; 2385 while ((slp->ns_flag & SLP_BUSY) != 0) { 2386 slp->ns_flag |= SLP_WANT; 2387 ltsleep(&slp->ns_flag, PSOCK, "nsdrain", 0, &slp->ns_lock); 2388 } 2389 done: 2390 simple_unlock(&slp->ns_lock); 2391 2392 return error; 2393 } 2394 2395 /* 2396 * Try and extract an RPC request from the mbuf data list received on a 2397 * stream socket. The "waitflag" argument indicates whether or not it 2398 * can sleep. 2399 */ 2400 int 2401 nfsrv_getstream(slp, waitflag) 2402 struct nfssvc_sock *slp; 2403 int waitflag; 2404 { 2405 struct mbuf *m, **mpp; 2406 struct mbuf *recm; 2407 u_int32_t recmark; 2408 int error = 0; 2409 2410 for (;;) { 2411 if (slp->ns_reclen == 0) { 2412 if (slp->ns_cc < NFSX_UNSIGNED) { 2413 break; 2414 } 2415 m = slp->ns_raw; 2416 m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark); 2417 m_adj(m, NFSX_UNSIGNED); 2418 slp->ns_cc -= NFSX_UNSIGNED; 2419 recmark = ntohl(recmark); 2420 slp->ns_reclen = recmark & ~0x80000000; 2421 if (recmark & 0x80000000) 2422 slp->ns_flag |= SLP_LASTFRAG; 2423 else 2424 slp->ns_flag &= ~SLP_LASTFRAG; 2425 if (slp->ns_reclen > NFS_MAXPACKET) { 2426 error = EPERM; 2427 break; 2428 } 2429 } 2430 2431 /* 2432 * Now get the record part. 2433 * 2434 * Note that slp->ns_reclen may be 0. Linux sometimes 2435 * generates 0-length records. 2436 */ 2437 if (slp->ns_cc == slp->ns_reclen) { 2438 recm = slp->ns_raw; 2439 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2440 slp->ns_cc = slp->ns_reclen = 0; 2441 } else if (slp->ns_cc > slp->ns_reclen) { 2442 recm = slp->ns_raw; 2443 m = m_split(recm, slp->ns_reclen, waitflag); 2444 if (m == NULL) { 2445 error = EWOULDBLOCK; 2446 break; 2447 } 2448 m_claimm(recm, &nfs_mowner); 2449 slp->ns_raw = m; 2450 if (m->m_next == NULL) 2451 slp->ns_rawend = m; 2452 slp->ns_cc -= slp->ns_reclen; 2453 slp->ns_reclen = 0; 2454 } else { 2455 break; 2456 } 2457 2458 /* 2459 * Accumulate the fragments into a record. 2460 */ 2461 mpp = &slp->ns_frag; 2462 while (*mpp) 2463 mpp = &((*mpp)->m_next); 2464 *mpp = recm; 2465 if (slp->ns_flag & SLP_LASTFRAG) { 2466 if (slp->ns_recend) 2467 slp->ns_recend->m_nextpkt = slp->ns_frag; 2468 else 2469 slp->ns_rec = slp->ns_frag; 2470 slp->ns_recend = slp->ns_frag; 2471 slp->ns_frag = (struct mbuf *)0; 2472 } 2473 } 2474 2475 return error; 2476 } 2477 2478 /* 2479 * Parse an RPC header. 2480 */ 2481 int 2482 nfsrv_dorec(slp, nfsd, ndp) 2483 struct nfssvc_sock *slp; 2484 struct nfsd *nfsd; 2485 struct nfsrv_descript **ndp; 2486 { 2487 struct mbuf *m, *nam; 2488 struct nfsrv_descript *nd; 2489 int error; 2490 2491 *ndp = NULL; 2492 2493 if (nfsdsock_lock(slp, TRUE)) { 2494 return ENOBUFS; 2495 } 2496 m = slp->ns_rec; 2497 if (m == NULL) { 2498 nfsdsock_unlock(slp); 2499 return ENOBUFS; 2500 } 2501 slp->ns_rec = m->m_nextpkt; 2502 if (slp->ns_rec) 2503 m->m_nextpkt = NULL; 2504 else 2505 slp->ns_recend = NULL; 2506 nfsdsock_unlock(slp); 2507 2508 if (m->m_type == MT_SONAME) { 2509 nam = m; 2510 m = m->m_next; 2511 nam->m_next = NULL; 2512 } else 2513 nam = NULL; 2514 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK); 2515 nd->nd_md = nd->nd_mrep = m; 2516 nd->nd_nam2 = nam; 2517 nd->nd_dpos = mtod(m, caddr_t); 2518 error = nfs_getreq(nd, nfsd, TRUE); 2519 if (error) { 2520 m_freem(nam); 2521 pool_put(&nfs_srvdesc_pool, nd); 2522 return (error); 2523 } 2524 *ndp = nd; 2525 nfsd->nfsd_nd = nd; 2526 return (0); 2527 } 2528 2529 2530 /* 2531 * Search for a sleeping nfsd and wake it up. 2532 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2533 * running nfsds will go look for the work in the nfssvc_sock list. 2534 */ 2535 void 2536 nfsrv_wakenfsd(slp) 2537 struct nfssvc_sock *slp; 2538 { 2539 struct nfsd *nd; 2540 2541 if ((slp->ns_flag & SLP_VALID) == 0) 2542 return; 2543 simple_lock(&nfsd_slock); 2544 if (slp->ns_flag & SLP_DOREC) { 2545 simple_unlock(&nfsd_slock); 2546 return; 2547 } 2548 nd = SLIST_FIRST(&nfsd_idle_head); 2549 if (nd) { 2550 SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle); 2551 simple_unlock(&nfsd_slock); 2552 2553 if (nd->nfsd_slp) 2554 panic("nfsd wakeup"); 2555 slp->ns_sref++; 2556 nd->nfsd_slp = slp; 2557 wakeup(nd); 2558 return; 2559 } 2560 slp->ns_flag |= SLP_DOREC; 2561 nfsd_head_flag |= NFSD_CHECKSLP; 2562 TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending); 2563 simple_unlock(&nfsd_slock); 2564 } 2565 2566 int 2567 nfsdsock_sendreply(struct nfssvc_sock *slp, struct nfsrv_descript *nd) 2568 { 2569 int error; 2570 2571 if (nd->nd_mrep != NULL) { 2572 m_freem(nd->nd_mrep); 2573 nd->nd_mrep = NULL; 2574 } 2575 2576 simple_lock(&slp->ns_lock); 2577 if ((slp->ns_flag & SLP_SENDING) != 0) { 2578 SIMPLEQ_INSERT_TAIL(&slp->ns_sendq, nd, nd_sendq); 2579 simple_unlock(&slp->ns_lock); 2580 return 0; 2581 } 2582 KASSERT(SIMPLEQ_EMPTY(&slp->ns_sendq)); 2583 slp->ns_flag |= SLP_SENDING; 2584 simple_unlock(&slp->ns_lock); 2585 2586 again: 2587 error = nfs_send(slp->ns_so, nd->nd_nam2, nd->nd_mreq, NULL, curlwp); 2588 if (nd->nd_nam2) { 2589 m_free(nd->nd_nam2); 2590 } 2591 pool_put(&nfs_srvdesc_pool, nd); 2592 2593 simple_lock(&slp->ns_lock); 2594 KASSERT((slp->ns_flag & SLP_SENDING) != 0); 2595 nd = SIMPLEQ_FIRST(&slp->ns_sendq); 2596 if (nd != NULL) { 2597 SIMPLEQ_REMOVE_HEAD(&slp->ns_sendq, nd_sendq); 2598 simple_unlock(&slp->ns_lock); 2599 goto again; 2600 } 2601 slp->ns_flag &= ~SLP_SENDING; 2602 simple_unlock(&slp->ns_lock); 2603 2604 return error; 2605 } 2606 #endif /* NFSSERVER */ 2607