1 /* $NetBSD: nfs_socket.c,v 1.126 2006/03/01 21:31:00 rpaulo 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 1.126 2006/03/01 21:31:00 rpaulo 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_rw = UIO_READ; 634 auio.uio_offset = 0; 635 auio.uio_resid = sizeof(u_int32_t); 636 UIO_SETUP_SYSSPACE(&auio); 637 do { 638 rcvflg = MSG_WAITALL; 639 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 640 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 641 if (error == EWOULDBLOCK && rep) { 642 if (rep->r_flags & R_SOFTTERM) 643 return (EINTR); 644 /* 645 * if it seems that the server died after it 646 * received our request, set EPIPE so that 647 * we'll reconnect and retransmit requests. 648 */ 649 if (rep->r_rexmit >= rep->r_nmp->nm_retry) { 650 nfsstats.rpctimeouts++; 651 error = EPIPE; 652 } 653 } 654 } while (error == EWOULDBLOCK); 655 if (!error && auio.uio_resid > 0) { 656 /* 657 * Don't log a 0 byte receive; it means 658 * that the socket has been closed, and 659 * can happen during normal operation 660 * (forcible unmount or Solaris server). 661 */ 662 if (auio.uio_resid != sizeof (u_int32_t)) 663 log(LOG_INFO, 664 "short receive (%lu/%lu) from nfs server %s\n", 665 (u_long)sizeof(u_int32_t) - auio.uio_resid, 666 (u_long)sizeof(u_int32_t), 667 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 668 error = EPIPE; 669 } 670 if (error) 671 goto errout; 672 len = ntohl(len) & ~0x80000000; 673 /* 674 * This is SERIOUS! We are out of sync with the sender 675 * and forcing a disconnect/reconnect is all I can do. 676 */ 677 if (len > NFS_MAXPACKET) { 678 log(LOG_ERR, "%s (%d) from nfs server %s\n", 679 "impossible packet length", 680 len, 681 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 682 error = EFBIG; 683 goto errout; 684 } 685 auio.uio_resid = len; 686 do { 687 rcvflg = MSG_WAITALL; 688 error = (*so->so_receive)(so, (struct mbuf **)0, 689 &auio, mp, (struct mbuf **)0, &rcvflg); 690 } while (error == EWOULDBLOCK || error == EINTR || 691 error == ERESTART); 692 if (!error && auio.uio_resid > 0) { 693 if (len != auio.uio_resid) 694 log(LOG_INFO, 695 "short receive (%lu/%d) from nfs server %s\n", 696 (u_long)len - auio.uio_resid, len, 697 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 698 error = EPIPE; 699 } 700 } else { 701 /* 702 * NB: Since uio_resid is big, MSG_WAITALL is ignored 703 * and soreceive() will return when it has either a 704 * control msg or a data msg. 705 * We have no use for control msg., but must grab them 706 * and then throw them away so we know what is going 707 * on. 708 */ 709 auio.uio_resid = len = 100000000; /* Anything Big */ 710 /* not need to setup uio_vmspace */ 711 do { 712 rcvflg = 0; 713 error = (*so->so_receive)(so, (struct mbuf **)0, 714 &auio, mp, &control, &rcvflg); 715 if (control) 716 m_freem(control); 717 if (error == EWOULDBLOCK && rep) { 718 if (rep->r_flags & R_SOFTTERM) 719 return (EINTR); 720 } 721 } while (error == EWOULDBLOCK || 722 (!error && *mp == NULL && control)); 723 if ((rcvflg & MSG_EOR) == 0) 724 printf("Egad!!\n"); 725 if (!error && *mp == NULL) 726 error = EPIPE; 727 len -= auio.uio_resid; 728 } 729 errout: 730 if (error && error != EINTR && error != ERESTART) { 731 m_freem(*mp); 732 *mp = (struct mbuf *)0; 733 if (error != EPIPE) 734 log(LOG_INFO, 735 "receive error %d from nfs server %s\n", 736 error, 737 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 738 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 739 if (!error) 740 error = nfs_reconnect(rep, l); 741 if (!error) 742 goto tryagain; 743 else 744 nfs_sndunlock(&rep->r_nmp->nm_iflag); 745 } 746 } else { 747 if ((so = rep->r_nmp->nm_so) == NULL) 748 return (EACCES); 749 if (so->so_state & SS_ISCONNECTED) 750 getnam = (struct mbuf **)0; 751 else 752 getnam = aname; 753 auio.uio_resid = len = 1000000; 754 /* not need to setup uio_vmspace */ 755 do { 756 rcvflg = 0; 757 error = (*so->so_receive)(so, getnam, &auio, mp, 758 (struct mbuf **)0, &rcvflg); 759 if (error == EWOULDBLOCK && 760 (rep->r_flags & R_SOFTTERM)) 761 return (EINTR); 762 } while (error == EWOULDBLOCK); 763 len -= auio.uio_resid; 764 if (!error && *mp == NULL) 765 error = EPIPE; 766 } 767 if (error) { 768 m_freem(*mp); 769 *mp = (struct mbuf *)0; 770 } 771 return (error); 772 } 773 774 /* 775 * Implement receipt of reply on a socket. 776 * We must search through the list of received datagrams matching them 777 * with outstanding requests using the xid, until ours is found. 778 */ 779 /* ARGSUSED */ 780 int 781 nfs_reply(myrep, lwp) 782 struct nfsreq *myrep; 783 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 caddr_t 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(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 nmp->nm_waiters++; 812 error = nfs_receive(myrep, &nam, &mrep, lwp); 813 nfs_rcvunlock(nmp); 814 if (error) { 815 816 if (nmp->nm_iflag & NFSMNT_DISMNT) { 817 /* 818 * Oops, we're going away now.. 819 */ 820 nmp->nm_waiters--; 821 wakeup (&nmp->nm_waiters); 822 return error; 823 } 824 nmp->nm_waiters--; 825 /* 826 * Ignore routing errors on connectionless protocols? ? 827 */ 828 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 829 nmp->nm_so->so_error = 0; 830 #ifdef DEBUG 831 printf("nfs_reply: ignoring error %d\n", error); 832 #endif 833 if (myrep->r_flags & R_GETONEREP) 834 return (0); 835 continue; 836 } 837 return (error); 838 } 839 nmp->nm_waiters--; 840 if (nam) 841 m_freem(nam); 842 843 /* 844 * Get the xid and check that it is an rpc reply 845 */ 846 md = mrep; 847 dpos = mtod(md, caddr_t); 848 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 849 rxid = *tl++; 850 if (*tl != rpc_reply) { 851 #ifndef NFS_V2_ONLY 852 if (nmp->nm_flag & NFSMNT_NQNFS) { 853 if (nqnfs_callback(nmp, mrep, md, dpos, 854 myrep->r_lwp)) 855 nfsstats.rpcinvalid++; 856 } else 857 #endif 858 { 859 nfsstats.rpcinvalid++; 860 m_freem(mrep); 861 } 862 nfsmout: 863 if (myrep->r_flags & R_GETONEREP) 864 return (0); 865 continue; 866 } 867 868 /* 869 * Loop through the request list to match up the reply 870 * Iff no match, just drop the datagram 871 */ 872 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 873 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 874 /* Found it.. */ 875 rep->r_mrep = mrep; 876 rep->r_md = md; 877 rep->r_dpos = dpos; 878 if (nfsrtton) { 879 struct rttl *rt; 880 881 rt = &nfsrtt.rttl[nfsrtt.pos]; 882 rt->proc = rep->r_procnum; 883 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 884 rt->sent = nmp->nm_sent; 885 rt->cwnd = nmp->nm_cwnd; 886 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 887 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 888 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx; 889 rt->tstamp = time; 890 if (rep->r_flags & R_TIMING) 891 rt->rtt = rep->r_rtt; 892 else 893 rt->rtt = 1000000; 894 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 895 } 896 /* 897 * Update congestion window. 898 * Do the additive increase of 899 * one rpc/rtt. 900 */ 901 if (nmp->nm_cwnd <= nmp->nm_sent) { 902 nmp->nm_cwnd += 903 (NFS_CWNDSCALE * NFS_CWNDSCALE + 904 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 905 if (nmp->nm_cwnd > NFS_MAXCWND) 906 nmp->nm_cwnd = NFS_MAXCWND; 907 } 908 rep->r_flags &= ~R_SENT; 909 nmp->nm_sent -= NFS_CWNDSCALE; 910 /* 911 * Update rtt using a gain of 0.125 on the mean 912 * and a gain of 0.25 on the deviation. 913 */ 914 if (rep->r_flags & R_TIMING) { 915 /* 916 * Since the timer resolution of 917 * NFS_HZ is so course, it can often 918 * result in r_rtt == 0. Since 919 * r_rtt == N means that the actual 920 * rtt is between N+dt and N+2-dt ticks, 921 * add 1. 922 */ 923 t1 = rep->r_rtt + 1; 924 t1 -= (NFS_SRTT(rep) >> 3); 925 NFS_SRTT(rep) += t1; 926 if (t1 < 0) 927 t1 = -t1; 928 t1 -= (NFS_SDRTT(rep) >> 2); 929 NFS_SDRTT(rep) += t1; 930 } 931 nmp->nm_timeouts = 0; 932 break; 933 } 934 } 935 /* 936 * If not matched to a request, drop it. 937 * If it's mine, get out. 938 */ 939 if (rep == 0) { 940 nfsstats.rpcunexpected++; 941 m_freem(mrep); 942 } else if (rep == myrep) { 943 if (rep->r_mrep == NULL) 944 panic("nfsreply nil"); 945 return (0); 946 } 947 if (myrep->r_flags & R_GETONEREP) 948 return (0); 949 } 950 } 951 952 /* 953 * nfs_request - goes something like this 954 * - fill in request struct 955 * - links it into list 956 * - calls nfs_send() for first transmit 957 * - calls nfs_receive() to get reply 958 * - break down rpc header and return with nfs reply pointed to 959 * by mrep or error 960 * nb: always frees up mreq mbuf list 961 */ 962 int 963 nfs_request(np, mrest, procnum, lwp, cred, mrp, mdp, dposp, rexmitp) 964 struct nfsnode *np; 965 struct mbuf *mrest; 966 int procnum; 967 struct lwp *lwp; 968 struct ucred *cred; 969 struct mbuf **mrp; 970 struct mbuf **mdp; 971 caddr_t *dposp; 972 int *rexmitp; 973 { 974 struct mbuf *m, *mrep; 975 struct nfsreq *rep; 976 u_int32_t *tl; 977 int i; 978 struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount); 979 struct mbuf *md, *mheadend; 980 char nickv[RPCX_NICKVERF]; 981 time_t reqtime, waituntil; 982 caddr_t dpos, cp2; 983 int t1, s, error = 0, mrest_len, auth_len, auth_type; 984 int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0; 985 int verf_len, verf_type; 986 u_int32_t xid; 987 char *auth_str, *verf_str; 988 NFSKERBKEY_T key; /* save session key */ 989 struct ucred acred; 990 #ifndef NFS_V2_ONLY 991 int nqlflag, cachable; 992 u_quad_t frev; 993 #endif 994 struct mbuf *mrest_backup = NULL; 995 struct ucred *origcred = NULL; /* XXX: gcc */ 996 boolean_t retry_cred = TRUE; 997 boolean_t use_opencred = (np->n_flag & NUSEOPENCRED) != 0; 998 999 if (rexmitp != NULL) 1000 *rexmitp = 0; 1001 1002 tryagain_cred: 1003 KASSERT(cred != NULL); 1004 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 1005 rep->r_nmp = nmp; 1006 rep->r_lwp = lwp; 1007 rep->r_procnum = procnum; 1008 i = 0; 1009 m = mrest; 1010 while (m) { 1011 i += m->m_len; 1012 m = m->m_next; 1013 } 1014 mrest_len = i; 1015 1016 /* 1017 * Get the RPC header with authorization. 1018 */ 1019 kerbauth: 1020 verf_str = auth_str = (char *)0; 1021 if (nmp->nm_flag & NFSMNT_KERB) { 1022 verf_str = nickv; 1023 verf_len = sizeof (nickv); 1024 auth_type = RPCAUTH_KERB4; 1025 memset((caddr_t)key, 0, sizeof (key)); 1026 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 1027 &auth_len, verf_str, verf_len)) { 1028 error = nfs_getauth(nmp, rep, cred, &auth_str, 1029 &auth_len, verf_str, &verf_len, key); 1030 if (error) { 1031 free((caddr_t)rep, M_NFSREQ); 1032 m_freem(mrest); 1033 return (error); 1034 } 1035 } 1036 retry_cred = FALSE; 1037 } else { 1038 /* AUTH_UNIX */ 1039 uid_t uid; 1040 gid_t gid; 1041 1042 /* 1043 * on the most unix filesystems, permission checks are 1044 * done when the file is open(2)'ed. 1045 * ie. once a file is successfully open'ed, 1046 * following i/o operations never fail with EACCES. 1047 * we try to follow the semantics as far as possible. 1048 * 1049 * note that we expect that the nfs server always grant 1050 * accesses by the file's owner. 1051 */ 1052 origcred = cred; 1053 switch (procnum) { 1054 case NFSPROC_READ: 1055 case NFSPROC_WRITE: 1056 case NFSPROC_COMMIT: 1057 uid = np->n_vattr->va_uid; 1058 gid = np->n_vattr->va_gid; 1059 if (cred->cr_uid == uid && cred->cr_gid == gid) { 1060 retry_cred = FALSE; 1061 break; 1062 } 1063 if (use_opencred) 1064 break; 1065 acred.cr_uid = uid; 1066 acred.cr_gid = gid; 1067 acred.cr_ngroups = 0; 1068 acred.cr_ref = 2; /* Just to be safe.. */ 1069 cred = &acred; 1070 break; 1071 default: 1072 retry_cred = FALSE; 1073 break; 1074 } 1075 /* 1076 * backup mbuf chain if we can need it later to retry. 1077 * 1078 * XXX maybe we can keep a direct reference to 1079 * mrest without doing m_copym, but it's ...ugly. 1080 */ 1081 if (retry_cred) 1082 mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT); 1083 auth_type = RPCAUTH_UNIX; 1084 auth_len = (((cred->cr_ngroups > nmp->nm_numgrps) ? 1085 nmp->nm_numgrps : cred->cr_ngroups) << 2) + 1086 5 * NFSX_UNSIGNED; 1087 } 1088 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 1089 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 1090 if (auth_str) 1091 free(auth_str, M_TEMP); 1092 1093 /* 1094 * For stream protocols, insert a Sun RPC Record Mark. 1095 */ 1096 if (nmp->nm_sotype == SOCK_STREAM) { 1097 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 1098 *mtod(m, u_int32_t *) = htonl(0x80000000 | 1099 (m->m_pkthdr.len - NFSX_UNSIGNED)); 1100 } 1101 rep->r_mreq = m; 1102 rep->r_xid = xid; 1103 tryagain: 1104 if (nmp->nm_flag & NFSMNT_SOFT) 1105 rep->r_retry = nmp->nm_retry; 1106 else 1107 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1108 rep->r_rtt = rep->r_rexmit = 0; 1109 if (proct[procnum] > 0) 1110 rep->r_flags = R_TIMING; 1111 else 1112 rep->r_flags = 0; 1113 rep->r_mrep = NULL; 1114 1115 /* 1116 * Do the client side RPC. 1117 */ 1118 nfsstats.rpcrequests++; 1119 /* 1120 * Chain request into list of outstanding requests. Be sure 1121 * to put it LAST so timer finds oldest requests first. 1122 */ 1123 s = splsoftnet(); 1124 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1125 1126 /* Get send time for nqnfs */ 1127 reqtime = time.tv_sec; 1128 1129 /* 1130 * If backing off another request or avoiding congestion, don't 1131 * send this one now but let timer do it. If not timing a request, 1132 * do it now. 1133 */ 1134 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1135 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1136 nmp->nm_sent < nmp->nm_cwnd)) { 1137 splx(s); 1138 if (nmp->nm_soflags & PR_CONNREQUIRED) 1139 error = nfs_sndlock(&nmp->nm_iflag, rep); 1140 if (!error) { 1141 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1142 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp); 1143 if (nmp->nm_soflags & PR_CONNREQUIRED) 1144 nfs_sndunlock(&nmp->nm_iflag); 1145 } 1146 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1147 nmp->nm_sent += NFS_CWNDSCALE; 1148 rep->r_flags |= R_SENT; 1149 } 1150 } else { 1151 splx(s); 1152 rep->r_rtt = -1; 1153 } 1154 1155 /* 1156 * Wait for the reply from our send or the timer's. 1157 */ 1158 if (!error || error == EPIPE) 1159 error = nfs_reply(rep, lwp); 1160 1161 /* 1162 * RPC done, unlink the request. 1163 */ 1164 s = splsoftnet(); 1165 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1166 splx(s); 1167 1168 /* 1169 * Decrement the outstanding request count. 1170 */ 1171 if (rep->r_flags & R_SENT) { 1172 rep->r_flags &= ~R_SENT; /* paranoia */ 1173 nmp->nm_sent -= NFS_CWNDSCALE; 1174 } 1175 1176 if (rexmitp != NULL) { 1177 int rexmit; 1178 1179 if (nmp->nm_sotype != SOCK_DGRAM) 1180 rexmit = (rep->r_flags & R_REXMITTED) != 0; 1181 else 1182 rexmit = rep->r_rexmit; 1183 *rexmitp = rexmit; 1184 } 1185 1186 /* 1187 * If there was a successful reply and a tprintf msg. 1188 * tprintf a response. 1189 */ 1190 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1191 nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1192 "is alive again"); 1193 mrep = rep->r_mrep; 1194 md = rep->r_md; 1195 dpos = rep->r_dpos; 1196 if (error) 1197 goto nfsmout; 1198 1199 /* 1200 * break down the rpc header and check if ok 1201 */ 1202 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1203 if (*tl++ == rpc_msgdenied) { 1204 if (*tl == rpc_mismatch) 1205 error = EOPNOTSUPP; 1206 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1207 if (!failed_auth) { 1208 failed_auth++; 1209 mheadend->m_next = (struct mbuf *)0; 1210 m_freem(mrep); 1211 m_freem(rep->r_mreq); 1212 goto kerbauth; 1213 } else 1214 error = EAUTH; 1215 } else 1216 error = EACCES; 1217 m_freem(mrep); 1218 goto nfsmout; 1219 } 1220 1221 /* 1222 * Grab any Kerberos verifier, otherwise just throw it away. 1223 */ 1224 verf_type = fxdr_unsigned(int, *tl++); 1225 i = fxdr_unsigned(int32_t, *tl); 1226 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1227 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1228 if (error) 1229 goto nfsmout; 1230 } else if (i > 0) 1231 nfsm_adv(nfsm_rndup(i)); 1232 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1233 /* 0 == ok */ 1234 if (*tl == 0) { 1235 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1236 if (*tl != 0) { 1237 error = fxdr_unsigned(int, *tl); 1238 switch (error) { 1239 case NFSERR_PERM: 1240 error = EPERM; 1241 break; 1242 1243 case NFSERR_NOENT: 1244 error = ENOENT; 1245 break; 1246 1247 case NFSERR_IO: 1248 error = EIO; 1249 break; 1250 1251 case NFSERR_NXIO: 1252 error = ENXIO; 1253 break; 1254 1255 case NFSERR_ACCES: 1256 error = EACCES; 1257 if (!retry_cred) 1258 break; 1259 m_freem(mrep); 1260 m_freem(rep->r_mreq); 1261 FREE(rep, M_NFSREQ); 1262 use_opencred = !use_opencred; 1263 if (mrest_backup == NULL) 1264 /* m_copym failure */ 1265 return ENOMEM; 1266 mrest = mrest_backup; 1267 mrest_backup = NULL; 1268 cred = origcred; 1269 error = 0; 1270 retry_cred = FALSE; 1271 goto tryagain_cred; 1272 1273 case NFSERR_EXIST: 1274 error = EEXIST; 1275 break; 1276 1277 case NFSERR_XDEV: 1278 error = EXDEV; 1279 break; 1280 1281 case NFSERR_NODEV: 1282 error = ENODEV; 1283 break; 1284 1285 case NFSERR_NOTDIR: 1286 error = ENOTDIR; 1287 break; 1288 1289 case NFSERR_ISDIR: 1290 error = EISDIR; 1291 break; 1292 1293 case NFSERR_INVAL: 1294 error = EINVAL; 1295 break; 1296 1297 case NFSERR_FBIG: 1298 error = EFBIG; 1299 break; 1300 1301 case NFSERR_NOSPC: 1302 error = ENOSPC; 1303 break; 1304 1305 case NFSERR_ROFS: 1306 error = EROFS; 1307 break; 1308 1309 case NFSERR_MLINK: 1310 error = EMLINK; 1311 break; 1312 1313 case NFSERR_TIMEDOUT: 1314 error = ETIMEDOUT; 1315 break; 1316 1317 case NFSERR_NAMETOL: 1318 error = ENAMETOOLONG; 1319 break; 1320 1321 case NFSERR_NOTEMPTY: 1322 error = ENOTEMPTY; 1323 break; 1324 1325 case NFSERR_DQUOT: 1326 error = EDQUOT; 1327 break; 1328 1329 case NFSERR_STALE: 1330 /* 1331 * If the File Handle was stale, invalidate the 1332 * lookup cache, just in case. 1333 */ 1334 error = ESTALE; 1335 cache_purge(NFSTOV(np)); 1336 break; 1337 1338 case NFSERR_REMOTE: 1339 error = EREMOTE; 1340 break; 1341 1342 case NFSERR_WFLUSH: 1343 case NFSERR_BADHANDLE: 1344 case NFSERR_NOT_SYNC: 1345 case NFSERR_BAD_COOKIE: 1346 error = EINVAL; 1347 break; 1348 1349 case NFSERR_NOTSUPP: 1350 error = ENOTSUP; 1351 break; 1352 1353 case NFSERR_TOOSMALL: 1354 case NFSERR_SERVERFAULT: 1355 case NFSERR_BADTYPE: 1356 error = EINVAL; 1357 break; 1358 1359 case NFSERR_TRYLATER: 1360 if ((nmp->nm_flag & NFSMNT_NFSV3) == 0) 1361 break; 1362 m_freem(mrep); 1363 error = 0; 1364 waituntil = time.tv_sec + trylater_delay; 1365 while (time.tv_sec < waituntil) 1366 (void) tsleep((caddr_t)&lbolt, 1367 PSOCK, "nqnfstry", 0); 1368 trylater_delay *= NFS_TRYLATERDELMUL; 1369 if (trylater_delay > NFS_TRYLATERDELMAX) 1370 trylater_delay = NFS_TRYLATERDELMAX; 1371 /* 1372 * RFC1813: 1373 * The client should wait and then try 1374 * the request with a new RPC transaction ID. 1375 */ 1376 nfs_renewxid(rep); 1377 goto tryagain; 1378 1379 case NFSERR_STALEWRITEVERF: 1380 error = EINVAL; 1381 break; 1382 1383 default: 1384 #ifdef DIAGNOSTIC 1385 printf("Invalid rpc error code %d\n", error); 1386 #endif 1387 error = EINVAL; 1388 break; 1389 } 1390 1391 if (nmp->nm_flag & NFSMNT_NFSV3) { 1392 *mrp = mrep; 1393 *mdp = md; 1394 *dposp = dpos; 1395 error |= NFSERR_RETERR; 1396 } else 1397 m_freem(mrep); 1398 goto nfsmout; 1399 } 1400 1401 /* 1402 * note which credential worked to minimize number of retries. 1403 */ 1404 if (use_opencred) 1405 np->n_flag |= NUSEOPENCRED; 1406 else 1407 np->n_flag &= ~NUSEOPENCRED; 1408 1409 #ifndef NFS_V2_ONLY 1410 /* 1411 * For nqnfs, get any lease in reply 1412 */ 1413 if (nmp->nm_flag & NFSMNT_NQNFS) { 1414 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1415 if (*tl) { 1416 nqlflag = fxdr_unsigned(int, *tl); 1417 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1418 cachable = fxdr_unsigned(int, *tl++); 1419 reqtime += fxdr_unsigned(int, *tl++); 1420 if (reqtime > time.tv_sec) { 1421 frev = fxdr_hyper(tl); 1422 nqnfs_clientlease(nmp, np, nqlflag, 1423 cachable, reqtime, frev); 1424 } 1425 } 1426 } 1427 #endif 1428 *mrp = mrep; 1429 *mdp = md; 1430 *dposp = dpos; 1431 1432 KASSERT(error == 0); 1433 goto nfsmout; 1434 } 1435 m_freem(mrep); 1436 error = EPROTONOSUPPORT; 1437 nfsmout: 1438 m_freem(rep->r_mreq); 1439 free((caddr_t)rep, M_NFSREQ); 1440 m_freem(mrest_backup); 1441 return (error); 1442 } 1443 #endif /* NFS */ 1444 1445 /* 1446 * Generate the rpc reply header 1447 * siz arg. is used to decide if adding a cluster is worthwhile 1448 */ 1449 int 1450 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1451 int siz; 1452 struct nfsrv_descript *nd; 1453 struct nfssvc_sock *slp; 1454 int err; 1455 int cache; 1456 u_quad_t *frev; 1457 struct mbuf **mrq; 1458 struct mbuf **mbp; 1459 caddr_t *bposp; 1460 { 1461 u_int32_t *tl; 1462 struct mbuf *mreq; 1463 caddr_t bpos; 1464 struct mbuf *mb; 1465 1466 mreq = m_gethdr(M_WAIT, MT_DATA); 1467 MCLAIM(mreq, &nfs_mowner); 1468 mb = mreq; 1469 /* 1470 * If this is a big reply, use a cluster else 1471 * try and leave leading space for the lower level headers. 1472 */ 1473 siz += RPC_REPLYSIZ; 1474 if (siz >= max_datalen) { 1475 m_clget(mreq, M_WAIT); 1476 } else 1477 mreq->m_data += max_hdr; 1478 tl = mtod(mreq, u_int32_t *); 1479 mreq->m_len = 6 * NFSX_UNSIGNED; 1480 bpos = ((caddr_t)tl) + mreq->m_len; 1481 *tl++ = txdr_unsigned(nd->nd_retxid); 1482 *tl++ = rpc_reply; 1483 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1484 *tl++ = rpc_msgdenied; 1485 if (err & NFSERR_AUTHERR) { 1486 *tl++ = rpc_autherr; 1487 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1488 mreq->m_len -= NFSX_UNSIGNED; 1489 bpos -= NFSX_UNSIGNED; 1490 } else { 1491 *tl++ = rpc_mismatch; 1492 *tl++ = txdr_unsigned(RPC_VER2); 1493 *tl = txdr_unsigned(RPC_VER2); 1494 } 1495 } else { 1496 *tl++ = rpc_msgaccepted; 1497 1498 /* 1499 * For Kerberos authentication, we must send the nickname 1500 * verifier back, otherwise just RPCAUTH_NULL. 1501 */ 1502 if (nd->nd_flag & ND_KERBFULL) { 1503 struct nfsuid *nuidp; 1504 struct timeval ktvin, ktvout; 1505 1506 LIST_FOREACH(nuidp, NUIDHASH(slp, nd->nd_cr.cr_uid), 1507 nu_hash) { 1508 if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && 1509 (!nd->nd_nam2 || netaddr_match( 1510 NU_NETFAM(nuidp), &nuidp->nu_haddr, 1511 nd->nd_nam2))) 1512 break; 1513 } 1514 if (nuidp) { 1515 ktvin.tv_sec = 1516 txdr_unsigned(nuidp->nu_timestamp.tv_sec 1517 - 1); 1518 ktvin.tv_usec = 1519 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1520 1521 /* 1522 * Encrypt the timestamp in ecb mode using the 1523 * session key. 1524 */ 1525 #ifdef NFSKERB 1526 XXX 1527 #endif 1528 1529 *tl++ = rpc_auth_kerb; 1530 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1531 *tl = ktvout.tv_sec; 1532 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1533 *tl++ = ktvout.tv_usec; 1534 *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); 1535 } else { 1536 *tl++ = 0; 1537 *tl++ = 0; 1538 } 1539 } else { 1540 *tl++ = 0; 1541 *tl++ = 0; 1542 } 1543 switch (err) { 1544 case EPROGUNAVAIL: 1545 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1546 break; 1547 case EPROGMISMATCH: 1548 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1549 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1550 if (nd->nd_flag & ND_NQNFS) { 1551 *tl++ = txdr_unsigned(3); 1552 *tl = txdr_unsigned(3); 1553 } else { 1554 *tl++ = txdr_unsigned(2); 1555 *tl = txdr_unsigned(3); 1556 } 1557 break; 1558 case EPROCUNAVAIL: 1559 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1560 break; 1561 case EBADRPC: 1562 *tl = txdr_unsigned(RPC_GARBAGE); 1563 break; 1564 default: 1565 *tl = 0; 1566 if (err != NFSERR_RETVOID) { 1567 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1568 if (err) 1569 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1570 else 1571 *tl = 0; 1572 } 1573 break; 1574 }; 1575 } 1576 1577 /* 1578 * For nqnfs, piggyback lease as requested. 1579 */ 1580 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1581 if (nd->nd_flag & ND_LEASE) { 1582 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1583 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1584 *tl++ = txdr_unsigned(cache); 1585 *tl++ = txdr_unsigned(nd->nd_duration); 1586 txdr_hyper(*frev, tl); 1587 } else { 1588 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1589 *tl = 0; 1590 } 1591 } 1592 if (mrq != NULL) 1593 *mrq = mreq; 1594 *mbp = mb; 1595 *bposp = bpos; 1596 if (err != 0 && err != NFSERR_RETVOID) 1597 nfsstats.srvrpc_errs++; 1598 return (0); 1599 } 1600 1601 /* 1602 * Nfs timer routine 1603 * Scan the nfsreq list and retranmit any requests that have timed out 1604 * To avoid retransmission attempts on STREAM sockets (in the future) make 1605 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1606 */ 1607 void 1608 nfs_timer(arg) 1609 void *arg; /* never used */ 1610 { 1611 struct nfsreq *rep; 1612 struct mbuf *m; 1613 struct socket *so; 1614 struct nfsmount *nmp; 1615 int timeo; 1616 int s, error; 1617 #ifdef NFSSERVER 1618 struct nfssvc_sock *slp; 1619 static long lasttime = 0; 1620 u_quad_t cur_usec; 1621 #endif 1622 1623 s = splsoftnet(); 1624 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1625 nmp = rep->r_nmp; 1626 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1627 continue; 1628 if (nfs_sigintr(nmp, rep, rep->r_lwp)) { 1629 rep->r_flags |= R_SOFTTERM; 1630 continue; 1631 } 1632 if (rep->r_rtt >= 0) { 1633 rep->r_rtt++; 1634 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1635 timeo = nmp->nm_timeo; 1636 else 1637 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1638 if (nmp->nm_timeouts > 0) 1639 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1640 if (rep->r_rtt <= timeo) 1641 continue; 1642 if (nmp->nm_timeouts < 1643 (sizeof(nfs_backoff) / sizeof(nfs_backoff[0]))) 1644 nmp->nm_timeouts++; 1645 } 1646 /* 1647 * Check for server not responding 1648 */ 1649 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1650 rep->r_rexmit > nmp->nm_deadthresh) { 1651 nfs_msg(rep->r_lwp, 1652 nmp->nm_mountp->mnt_stat.f_mntfromname, 1653 "not responding"); 1654 rep->r_flags |= R_TPRINTFMSG; 1655 } 1656 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1657 nfsstats.rpctimeouts++; 1658 rep->r_flags |= R_SOFTTERM; 1659 continue; 1660 } 1661 if (nmp->nm_sotype != SOCK_DGRAM) { 1662 if (++rep->r_rexmit > NFS_MAXREXMIT) 1663 rep->r_rexmit = NFS_MAXREXMIT; 1664 continue; 1665 } 1666 if ((so = nmp->nm_so) == NULL) 1667 continue; 1668 1669 /* 1670 * If there is enough space and the window allows.. 1671 * Resend it 1672 * Set r_rtt to -1 in case we fail to send it now. 1673 */ 1674 rep->r_rtt = -1; 1675 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1676 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1677 (rep->r_flags & R_SENT) || 1678 nmp->nm_sent < nmp->nm_cwnd) && 1679 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1680 if (so->so_state & SS_ISCONNECTED) 1681 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1682 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1683 else 1684 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1685 nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0); 1686 if (error) { 1687 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1688 #ifdef DEBUG 1689 printf("nfs_timer: ignoring error %d\n", 1690 error); 1691 #endif 1692 so->so_error = 0; 1693 } 1694 } else { 1695 /* 1696 * Iff first send, start timing 1697 * else turn timing off, backoff timer 1698 * and divide congestion window by 2. 1699 */ 1700 if (rep->r_flags & R_SENT) { 1701 rep->r_flags &= ~R_TIMING; 1702 if (++rep->r_rexmit > NFS_MAXREXMIT) 1703 rep->r_rexmit = NFS_MAXREXMIT; 1704 nmp->nm_cwnd >>= 1; 1705 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1706 nmp->nm_cwnd = NFS_CWNDSCALE; 1707 nfsstats.rpcretries++; 1708 } else { 1709 rep->r_flags |= R_SENT; 1710 nmp->nm_sent += NFS_CWNDSCALE; 1711 } 1712 rep->r_rtt = 0; 1713 } 1714 } 1715 } 1716 1717 #ifdef NFSSERVER 1718 /* 1719 * Call the nqnfs server timer once a second to handle leases. 1720 */ 1721 if (lasttime != time.tv_sec) { 1722 lasttime = time.tv_sec; 1723 nqnfs_serverd(); 1724 } 1725 1726 /* 1727 * Scan the write gathering queues for writes that need to be 1728 * completed now. 1729 */ 1730 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1731 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1732 if (LIST_FIRST(&slp->ns_tq) && 1733 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) 1734 nfsrv_wakenfsd(slp); 1735 } 1736 #endif /* NFSSERVER */ 1737 splx(s); 1738 callout_schedule(&nfs_timer_ch, nfs_ticks); 1739 } 1740 1741 /*ARGSUSED*/ 1742 void 1743 nfs_exit(p, v) 1744 struct proc *p; 1745 void *v; 1746 { 1747 struct nfsreq *rp; 1748 int s = splsoftnet(); 1749 1750 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 1751 if (rp->r_lwp && rp->r_lwp->l_proc == p) 1752 TAILQ_REMOVE(&nfs_reqq, rp, r_chain); 1753 } 1754 splx(s); 1755 } 1756 1757 /* 1758 * Test for a termination condition pending on the process. 1759 * This is used for NFSMNT_INT mounts. 1760 */ 1761 int 1762 nfs_sigintr(nmp, rep, l) 1763 struct nfsmount *nmp; 1764 struct nfsreq *rep; 1765 struct lwp *l; 1766 { 1767 sigset_t ss; 1768 1769 if (rep && (rep->r_flags & R_SOFTTERM)) 1770 return (EINTR); 1771 if (!(nmp->nm_flag & NFSMNT_INT)) 1772 return (0); 1773 if (l) { 1774 sigpending1(l->l_proc, &ss); 1775 #if 0 1776 sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss); 1777 #endif 1778 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1779 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1780 sigismember(&ss, SIGQUIT)) 1781 return (EINTR); 1782 } 1783 return (0); 1784 } 1785 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 int 1793 nfs_sndlock(flagp, rep) 1794 int *flagp; 1795 struct nfsreq *rep; 1796 { 1797 struct lwp *l; 1798 int slpflag = 0, slptimeo = 0; 1799 1800 if (rep) { 1801 l = rep->r_lwp; 1802 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1803 slpflag = PCATCH; 1804 } else 1805 l = (struct lwp *)0; 1806 while (*flagp & NFSMNT_SNDLOCK) { 1807 if (rep && nfs_sigintr(rep->r_nmp, rep, l)) 1808 return (EINTR); 1809 *flagp |= NFSMNT_WANTSND; 1810 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1811 slptimeo); 1812 if (slpflag == PCATCH) { 1813 slpflag = 0; 1814 slptimeo = 2 * hz; 1815 } 1816 } 1817 *flagp |= NFSMNT_SNDLOCK; 1818 return (0); 1819 } 1820 1821 /* 1822 * Unlock the stream socket for others. 1823 */ 1824 void 1825 nfs_sndunlock(flagp) 1826 int *flagp; 1827 { 1828 1829 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1830 panic("nfs sndunlock"); 1831 *flagp &= ~NFSMNT_SNDLOCK; 1832 if (*flagp & NFSMNT_WANTSND) { 1833 *flagp &= ~NFSMNT_WANTSND; 1834 wakeup((caddr_t)flagp); 1835 } 1836 } 1837 1838 int 1839 nfs_rcvlock(rep) 1840 struct nfsreq *rep; 1841 { 1842 struct nfsmount *nmp = rep->r_nmp; 1843 int *flagp = &nmp->nm_iflag; 1844 int slpflag, slptimeo = 0; 1845 int error = 0; 1846 1847 if (*flagp & NFSMNT_DISMNT) 1848 return EIO; 1849 1850 if (*flagp & NFSMNT_INT) 1851 slpflag = PCATCH; 1852 else 1853 slpflag = 0; 1854 simple_lock(&nmp->nm_slock); 1855 while (*flagp & NFSMNT_RCVLOCK) { 1856 if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) { 1857 error = EINTR; 1858 goto quit; 1859 } 1860 *flagp |= NFSMNT_WANTRCV; 1861 nmp->nm_waiters++; 1862 (void) ltsleep(flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1863 slptimeo, &nmp->nm_slock); 1864 nmp->nm_waiters--; 1865 if (*flagp & NFSMNT_DISMNT) { 1866 wakeup(&nmp->nm_waiters); 1867 error = EIO; 1868 goto quit; 1869 } 1870 /* If our reply was received while we were sleeping, 1871 * then just return without taking the lock to avoid a 1872 * situation where a single iod could 'capture' the 1873 * receive lock. 1874 */ 1875 if (rep->r_mrep != NULL) { 1876 error = EALREADY; 1877 goto quit; 1878 } 1879 if (slpflag == PCATCH) { 1880 slpflag = 0; 1881 slptimeo = 2 * hz; 1882 } 1883 } 1884 *flagp |= NFSMNT_RCVLOCK; 1885 quit: 1886 simple_unlock(&nmp->nm_slock); 1887 return error; 1888 } 1889 1890 /* 1891 * Unlock the stream socket for others. 1892 */ 1893 void 1894 nfs_rcvunlock(nmp) 1895 struct nfsmount *nmp; 1896 { 1897 int *flagp = &nmp->nm_iflag; 1898 1899 simple_lock(&nmp->nm_slock); 1900 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1901 panic("nfs rcvunlock"); 1902 *flagp &= ~NFSMNT_RCVLOCK; 1903 if (*flagp & NFSMNT_WANTRCV) { 1904 *flagp &= ~NFSMNT_WANTRCV; 1905 wakeup((caddr_t)flagp); 1906 } 1907 simple_unlock(&nmp->nm_slock); 1908 } 1909 1910 /* 1911 * Parse an RPC request 1912 * - verify it 1913 * - fill in the cred struct. 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 caddr_t dpos, cp2, cp; 1927 u_int32_t nfsvers, auth_type; 1928 uid_t nickuid; 1929 int error = 0, nqnfs = 0, ticklen; 1930 struct mbuf *mrep, *md; 1931 struct nfsuid *nuidp; 1932 struct timeval tvin, tvout; 1933 1934 mrep = nd->nd_mrep; 1935 md = nd->nd_md; 1936 dpos = nd->nd_dpos; 1937 if (has_header) { 1938 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1939 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1940 if (*tl++ != rpc_call) { 1941 m_freem(mrep); 1942 return (EBADRPC); 1943 } 1944 } else 1945 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1946 nd->nd_repstat = 0; 1947 nd->nd_flag = 0; 1948 if (*tl++ != rpc_vers) { 1949 nd->nd_repstat = ERPCMISMATCH; 1950 nd->nd_procnum = NFSPROC_NOOP; 1951 return (0); 1952 } 1953 if (*tl != nfs_prog) { 1954 if (*tl == nqnfs_prog) 1955 nqnfs++; 1956 else { 1957 nd->nd_repstat = EPROGUNAVAIL; 1958 nd->nd_procnum = NFSPROC_NOOP; 1959 return (0); 1960 } 1961 } 1962 tl++; 1963 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1964 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1965 (nfsvers != NQNFS_VER3 && nqnfs)) { 1966 nd->nd_repstat = EPROGMISMATCH; 1967 nd->nd_procnum = NFSPROC_NOOP; 1968 return (0); 1969 } 1970 if (nqnfs) 1971 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1972 else if (nfsvers == NFS_VER3) 1973 nd->nd_flag = ND_NFSV3; 1974 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1975 if (nd->nd_procnum == NFSPROC_NULL) 1976 return (0); 1977 if (nd->nd_procnum >= NFS_NPROCS || 1978 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1979 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1980 nd->nd_repstat = EPROCUNAVAIL; 1981 nd->nd_procnum = NFSPROC_NOOP; 1982 return (0); 1983 } 1984 if ((nd->nd_flag & ND_NFSV3) == 0) 1985 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1986 auth_type = *tl++; 1987 len = fxdr_unsigned(int, *tl++); 1988 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1989 m_freem(mrep); 1990 return (EBADRPC); 1991 } 1992 1993 nd->nd_flag &= ~ND_KERBAUTH; 1994 /* 1995 * Handle auth_unix or auth_kerb. 1996 */ 1997 if (auth_type == rpc_auth_unix) { 1998 len = fxdr_unsigned(int, *++tl); 1999 if (len < 0 || len > NFS_MAXNAMLEN) { 2000 m_freem(mrep); 2001 return (EBADRPC); 2002 } 2003 nfsm_adv(nfsm_rndup(len)); 2004 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2005 memset((caddr_t)&nd->nd_cr, 0, sizeof (struct ucred)); 2006 nd->nd_cr.cr_ref = 1; 2007 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 2008 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 2009 len = fxdr_unsigned(int, *tl); 2010 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 2011 m_freem(mrep); 2012 return (EBADRPC); 2013 } 2014 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 2015 for (i = 0; i < len; i++) 2016 if (i < NGROUPS) 2017 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 2018 else 2019 tl++; 2020 nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; 2021 if (nd->nd_cr.cr_ngroups > 1) 2022 nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); 2023 len = fxdr_unsigned(int, *++tl); 2024 if (len < 0 || len > RPCAUTH_MAXSIZ) { 2025 m_freem(mrep); 2026 return (EBADRPC); 2027 } 2028 if (len > 0) 2029 nfsm_adv(nfsm_rndup(len)); 2030 } else if (auth_type == rpc_auth_kerb) { 2031 switch (fxdr_unsigned(int, *tl++)) { 2032 case RPCAKN_FULLNAME: 2033 ticklen = fxdr_unsigned(int, *tl); 2034 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 2035 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 2036 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 2037 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 2038 m_freem(mrep); 2039 return (EBADRPC); 2040 } 2041 uio.uio_offset = 0; 2042 uio.uio_iov = &iov; 2043 uio.uio_iovcnt = 1; 2044 UIO_SETUP_SYSSPACE(&uio); 2045 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 2046 iov.iov_len = RPCAUTH_MAXSIZ - 4; 2047 nfsm_mtouio(&uio, uio.uio_resid); 2048 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2049 if (*tl++ != rpc_auth_kerb || 2050 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 2051 printf("Bad kerb verifier\n"); 2052 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2053 nd->nd_procnum = NFSPROC_NOOP; 2054 return (0); 2055 } 2056 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 2057 tl = (u_int32_t *)cp; 2058 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 2059 printf("Not fullname kerb verifier\n"); 2060 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2061 nd->nd_procnum = NFSPROC_NOOP; 2062 return (0); 2063 } 2064 cp += NFSX_UNSIGNED; 2065 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 2066 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 2067 nd->nd_flag |= ND_KERBFULL; 2068 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 2069 break; 2070 case RPCAKN_NICKNAME: 2071 if (len != 2 * NFSX_UNSIGNED) { 2072 printf("Kerb nickname short\n"); 2073 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 2074 nd->nd_procnum = NFSPROC_NOOP; 2075 return (0); 2076 } 2077 nickuid = fxdr_unsigned(uid_t, *tl); 2078 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2079 if (*tl++ != rpc_auth_kerb || 2080 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 2081 printf("Kerb nick verifier bad\n"); 2082 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2083 nd->nd_procnum = NFSPROC_NOOP; 2084 return (0); 2085 } 2086 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2087 tvin.tv_sec = *tl++; 2088 tvin.tv_usec = *tl; 2089 2090 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), 2091 nu_hash) { 2092 if (nuidp->nu_cr.cr_uid == nickuid && 2093 (!nd->nd_nam2 || 2094 netaddr_match(NU_NETFAM(nuidp), 2095 &nuidp->nu_haddr, nd->nd_nam2))) 2096 break; 2097 } 2098 if (!nuidp) { 2099 nd->nd_repstat = 2100 (NFSERR_AUTHERR|AUTH_REJECTCRED); 2101 nd->nd_procnum = NFSPROC_NOOP; 2102 return (0); 2103 } 2104 2105 /* 2106 * Now, decrypt the timestamp using the session key 2107 * and validate it. 2108 */ 2109 #ifdef NFSKERB 2110 XXX 2111 #endif 2112 2113 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 2114 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 2115 if (nuidp->nu_expire < time.tv_sec || 2116 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 2117 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 2118 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 2119 nuidp->nu_expire = 0; 2120 nd->nd_repstat = 2121 (NFSERR_AUTHERR|AUTH_REJECTVERF); 2122 nd->nd_procnum = NFSPROC_NOOP; 2123 return (0); 2124 } 2125 nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); 2126 nd->nd_flag |= ND_KERBNICK; 2127 }; 2128 } else { 2129 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 2130 nd->nd_procnum = NFSPROC_NOOP; 2131 return (0); 2132 } 2133 2134 /* 2135 * For nqnfs, get piggybacked lease request. 2136 */ 2137 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 2138 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2139 nd->nd_flag |= fxdr_unsigned(int, *tl); 2140 if (nd->nd_flag & ND_LEASE) { 2141 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2142 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 2143 } else 2144 nd->nd_duration = NQ_MINLEASE; 2145 } else 2146 nd->nd_duration = NQ_MINLEASE; 2147 nd->nd_md = md; 2148 nd->nd_dpos = dpos; 2149 return (0); 2150 nfsmout: 2151 return (error); 2152 } 2153 2154 int 2155 nfs_msg(l, server, msg) 2156 struct lwp *l; 2157 const char *server, *msg; 2158 { 2159 tpr_t tpr; 2160 2161 if (l) 2162 tpr = tprintf_open(l->l_proc); 2163 else 2164 tpr = NULL; 2165 tprintf(tpr, "nfs server %s: %s\n", server, msg); 2166 tprintf_close(tpr); 2167 return (0); 2168 } 2169 2170 #ifdef NFSSERVER 2171 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 2172 struct nfssvc_sock *, struct lwp *, 2173 struct mbuf **)) = { 2174 nfsrv_null, 2175 nfsrv_getattr, 2176 nfsrv_setattr, 2177 nfsrv_lookup, 2178 nfsrv3_access, 2179 nfsrv_readlink, 2180 nfsrv_read, 2181 nfsrv_write, 2182 nfsrv_create, 2183 nfsrv_mkdir, 2184 nfsrv_symlink, 2185 nfsrv_mknod, 2186 nfsrv_remove, 2187 nfsrv_rmdir, 2188 nfsrv_rename, 2189 nfsrv_link, 2190 nfsrv_readdir, 2191 nfsrv_readdirplus, 2192 nfsrv_statfs, 2193 nfsrv_fsinfo, 2194 nfsrv_pathconf, 2195 nfsrv_commit, 2196 nqnfsrv_getlease, 2197 nqnfsrv_vacated, 2198 nfsrv_noop, 2199 nfsrv_noop 2200 }; 2201 2202 /* 2203 * Socket upcall routine for the nfsd sockets. 2204 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 2205 * Essentially do as much as possible non-blocking, else punt and it will 2206 * be called with M_WAIT from an nfsd. 2207 */ 2208 void 2209 nfsrv_rcv(so, arg, waitflag) 2210 struct socket *so; 2211 caddr_t arg; 2212 int waitflag; 2213 { 2214 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 2215 struct mbuf *m; 2216 struct mbuf *mp, *nam; 2217 struct uio auio; 2218 int flags, error; 2219 int setflags = 0; 2220 2221 error = nfsdsock_lock(slp, (waitflag != M_DONTWAIT)); 2222 if (error) { 2223 setflags |= SLP_NEEDQ; 2224 goto dorecs_unlocked; 2225 } 2226 2227 KASSERT(so == slp->ns_so); 2228 #if 1 2229 /* 2230 * Define this to test for nfsds handling this under heavy load. 2231 * 2232 * XXX it isn't safe to call so_receive from so_upcall context. 2233 */ 2234 if (waitflag == M_DONTWAIT) { 2235 setflags |= SLP_NEEDQ; 2236 goto dorecs; 2237 } 2238 #endif 2239 simple_lock(&slp->ns_lock); 2240 slp->ns_flag &= ~SLP_NEEDQ; 2241 simple_unlock(&slp->ns_lock); 2242 if (so->so_type == SOCK_STREAM) { 2243 /* 2244 * If there are already records on the queue, defer soreceive() 2245 * to an nfsd so that there is feedback to the TCP layer that 2246 * the nfs servers are heavily loaded. 2247 */ 2248 if (slp->ns_rec && waitflag == M_DONTWAIT) { 2249 setflags |= SLP_NEEDQ; 2250 goto dorecs; 2251 } 2252 2253 /* 2254 * Do soreceive(). 2255 */ 2256 auio.uio_resid = 1000000000; 2257 /* not need to setup uio_vmspace */ 2258 flags = MSG_DONTWAIT; 2259 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, &flags); 2260 if (error || mp == NULL) { 2261 if (error == EWOULDBLOCK) 2262 setflags |= SLP_NEEDQ; 2263 else 2264 setflags |= SLP_DISCONN; 2265 goto dorecs; 2266 } 2267 m = mp; 2268 if (slp->ns_rawend) { 2269 slp->ns_rawend->m_next = m; 2270 slp->ns_cc += 1000000000 - auio.uio_resid; 2271 } else { 2272 slp->ns_raw = m; 2273 slp->ns_cc = 1000000000 - auio.uio_resid; 2274 } 2275 while (m->m_next) 2276 m = m->m_next; 2277 slp->ns_rawend = m; 2278 2279 /* 2280 * Now try and parse record(s) out of the raw stream data. 2281 */ 2282 error = nfsrv_getstream(slp, waitflag); 2283 if (error) { 2284 if (error == EPERM) 2285 setflags |= SLP_DISCONN; 2286 else 2287 setflags |= SLP_NEEDQ; 2288 } 2289 } else { 2290 do { 2291 auio.uio_resid = 1000000000; 2292 /* not need to setup uio_vmspace */ 2293 flags = MSG_DONTWAIT; 2294 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, 2295 &flags); 2296 if (mp) { 2297 if (nam) { 2298 m = nam; 2299 m->m_next = mp; 2300 } else 2301 m = mp; 2302 if (slp->ns_recend) 2303 slp->ns_recend->m_nextpkt = m; 2304 else 2305 slp->ns_rec = m; 2306 slp->ns_recend = m; 2307 m->m_nextpkt = (struct mbuf *)0; 2308 } 2309 if (error) { 2310 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2311 && error != EWOULDBLOCK) { 2312 setflags |= SLP_DISCONN; 2313 goto dorecs; 2314 } 2315 } 2316 } while (mp); 2317 } 2318 dorecs: 2319 nfsdsock_unlock(slp); 2320 2321 dorecs_unlocked: 2322 /* 2323 * Now try and process the request records, non-blocking. 2324 */ 2325 if (setflags) { 2326 simple_lock(&slp->ns_lock); 2327 slp->ns_flag |= setflags; 2328 simple_unlock(&slp->ns_lock); 2329 } 2330 if (waitflag == M_DONTWAIT && 2331 (slp->ns_rec || (slp->ns_flag & (SLP_DISCONN | SLP_NEEDQ)) != 0)) { 2332 nfsrv_wakenfsd(slp); 2333 } 2334 } 2335 2336 int 2337 nfsdsock_lock(struct nfssvc_sock *slp, boolean_t waitok) 2338 { 2339 2340 simple_lock(&slp->ns_lock); 2341 while ((slp->ns_flag & (SLP_BUSY|SLP_VALID)) == SLP_BUSY) { 2342 if (!waitok) { 2343 simple_unlock(&slp->ns_lock); 2344 return EWOULDBLOCK; 2345 } 2346 slp->ns_flag |= SLP_WANT; 2347 ltsleep(&slp->ns_flag, PSOCK, "nslock", 0, &slp->ns_lock); 2348 } 2349 if ((slp->ns_flag & SLP_VALID) == 0) { 2350 simple_unlock(&slp->ns_lock); 2351 return EINVAL; 2352 } 2353 slp->ns_flag |= SLP_BUSY; 2354 simple_unlock(&slp->ns_lock); 2355 2356 return 0; 2357 } 2358 2359 void 2360 nfsdsock_unlock(struct nfssvc_sock *slp) 2361 { 2362 2363 KASSERT((slp->ns_flag & SLP_BUSY) != 0); 2364 2365 simple_lock(&slp->ns_lock); 2366 if ((slp->ns_flag & SLP_WANT) != 0) { 2367 wakeup(&slp->ns_flag); 2368 } 2369 slp->ns_flag &= ~(SLP_BUSY|SLP_WANT); 2370 simple_unlock(&slp->ns_lock); 2371 } 2372 2373 int 2374 nfsdsock_drain(struct nfssvc_sock *slp) 2375 { 2376 int error = 0; 2377 2378 simple_lock(&slp->ns_lock); 2379 if ((slp->ns_flag & SLP_VALID) == 0) { 2380 error = EINVAL; 2381 goto done; 2382 } 2383 slp->ns_flag &= ~SLP_VALID; 2384 while ((slp->ns_flag & SLP_BUSY) != 0) { 2385 slp->ns_flag |= SLP_WANT; 2386 ltsleep(&slp->ns_flag, PSOCK, "nsdrain", 0, &slp->ns_lock); 2387 } 2388 done: 2389 simple_unlock(&slp->ns_lock); 2390 2391 return error; 2392 } 2393 2394 /* 2395 * Try and extract an RPC request from the mbuf data list received on a 2396 * stream socket. The "waitflag" argument indicates whether or not it 2397 * can sleep. 2398 */ 2399 int 2400 nfsrv_getstream(slp, waitflag) 2401 struct nfssvc_sock *slp; 2402 int waitflag; 2403 { 2404 struct mbuf *m, **mpp; 2405 struct mbuf *recm; 2406 u_int32_t recmark; 2407 int error = 0; 2408 2409 for (;;) { 2410 if (slp->ns_reclen == 0) { 2411 if (slp->ns_cc < NFSX_UNSIGNED) { 2412 break; 2413 } 2414 m = slp->ns_raw; 2415 m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark); 2416 m_adj(m, NFSX_UNSIGNED); 2417 slp->ns_cc -= NFSX_UNSIGNED; 2418 recmark = ntohl(recmark); 2419 slp->ns_reclen = recmark & ~0x80000000; 2420 if (recmark & 0x80000000) 2421 slp->ns_flag |= SLP_LASTFRAG; 2422 else 2423 slp->ns_flag &= ~SLP_LASTFRAG; 2424 if (slp->ns_reclen > NFS_MAXPACKET) { 2425 error = EPERM; 2426 break; 2427 } 2428 } 2429 2430 /* 2431 * Now get the record part. 2432 * 2433 * Note that slp->ns_reclen may be 0. Linux sometimes 2434 * generates 0-length records. 2435 */ 2436 if (slp->ns_cc == slp->ns_reclen) { 2437 recm = slp->ns_raw; 2438 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2439 slp->ns_cc = slp->ns_reclen = 0; 2440 } else if (slp->ns_cc > slp->ns_reclen) { 2441 recm = slp->ns_raw; 2442 m = m_split(recm, slp->ns_reclen, waitflag); 2443 if (m == NULL) { 2444 error = EWOULDBLOCK; 2445 break; 2446 } 2447 m_claimm(recm, &nfs_mowner); 2448 slp->ns_raw = m; 2449 if (m->m_next == NULL) 2450 slp->ns_rawend = m; 2451 slp->ns_cc -= slp->ns_reclen; 2452 slp->ns_reclen = 0; 2453 } else { 2454 break; 2455 } 2456 2457 /* 2458 * Accumulate the fragments into a record. 2459 */ 2460 mpp = &slp->ns_frag; 2461 while (*mpp) 2462 mpp = &((*mpp)->m_next); 2463 *mpp = recm; 2464 if (slp->ns_flag & SLP_LASTFRAG) { 2465 if (slp->ns_recend) 2466 slp->ns_recend->m_nextpkt = slp->ns_frag; 2467 else 2468 slp->ns_rec = slp->ns_frag; 2469 slp->ns_recend = slp->ns_frag; 2470 slp->ns_frag = (struct mbuf *)0; 2471 } 2472 } 2473 2474 return error; 2475 } 2476 2477 /* 2478 * Parse an RPC header. 2479 */ 2480 int 2481 nfsrv_dorec(slp, nfsd, ndp) 2482 struct nfssvc_sock *slp; 2483 struct nfsd *nfsd; 2484 struct nfsrv_descript **ndp; 2485 { 2486 struct mbuf *m, *nam; 2487 struct nfsrv_descript *nd; 2488 int error; 2489 2490 *ndp = NULL; 2491 2492 if (nfsdsock_lock(slp, TRUE)) { 2493 return ENOBUFS; 2494 } 2495 m = slp->ns_rec; 2496 if (m == NULL) { 2497 nfsdsock_unlock(slp); 2498 return ENOBUFS; 2499 } 2500 slp->ns_rec = m->m_nextpkt; 2501 if (slp->ns_rec) 2502 m->m_nextpkt = NULL; 2503 else 2504 slp->ns_recend = NULL; 2505 nfsdsock_unlock(slp); 2506 2507 if (m->m_type == MT_SONAME) { 2508 nam = m; 2509 m = m->m_next; 2510 nam->m_next = NULL; 2511 } else 2512 nam = NULL; 2513 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK); 2514 nd->nd_md = nd->nd_mrep = m; 2515 nd->nd_nam2 = nam; 2516 nd->nd_dpos = mtod(m, caddr_t); 2517 error = nfs_getreq(nd, nfsd, TRUE); 2518 if (error) { 2519 m_freem(nam); 2520 pool_put(&nfs_srvdesc_pool, nd); 2521 return (error); 2522 } 2523 *ndp = nd; 2524 nfsd->nfsd_nd = nd; 2525 return (0); 2526 } 2527 2528 2529 /* 2530 * Search for a sleeping nfsd and wake it up. 2531 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2532 * running nfsds will go look for the work in the nfssvc_sock list. 2533 */ 2534 void 2535 nfsrv_wakenfsd(slp) 2536 struct nfssvc_sock *slp; 2537 { 2538 struct nfsd *nd; 2539 2540 if ((slp->ns_flag & SLP_VALID) == 0) 2541 return; 2542 simple_lock(&nfsd_slock); 2543 if (slp->ns_flag & SLP_DOREC) { 2544 simple_unlock(&nfsd_slock); 2545 return; 2546 } 2547 nd = SLIST_FIRST(&nfsd_idle_head); 2548 if (nd) { 2549 SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle); 2550 simple_unlock(&nfsd_slock); 2551 2552 if (nd->nfsd_slp) 2553 panic("nfsd wakeup"); 2554 slp->ns_sref++; 2555 nd->nfsd_slp = slp; 2556 wakeup(nd); 2557 return; 2558 } 2559 slp->ns_flag |= SLP_DOREC; 2560 nfsd_head_flag |= NFSD_CHECKSLP; 2561 TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending); 2562 simple_unlock(&nfsd_slock); 2563 } 2564 2565 int 2566 nfsdsock_sendreply(struct nfssvc_sock *slp, struct nfsrv_descript *nd) 2567 { 2568 int error; 2569 2570 if (nd->nd_mrep != NULL) { 2571 m_freem(nd->nd_mrep); 2572 nd->nd_mrep = NULL; 2573 } 2574 2575 simple_lock(&slp->ns_lock); 2576 if ((slp->ns_flag & SLP_SENDING) != 0) { 2577 SIMPLEQ_INSERT_TAIL(&slp->ns_sendq, nd, nd_sendq); 2578 simple_unlock(&slp->ns_lock); 2579 return 0; 2580 } 2581 KASSERT(SIMPLEQ_EMPTY(&slp->ns_sendq)); 2582 slp->ns_flag |= SLP_SENDING; 2583 simple_unlock(&slp->ns_lock); 2584 2585 again: 2586 error = nfs_send(slp->ns_so, nd->nd_nam2, nd->nd_mreq, NULL, curlwp); 2587 if (nd->nd_nam2) { 2588 m_free(nd->nd_nam2); 2589 } 2590 pool_put(&nfs_srvdesc_pool, nd); 2591 2592 simple_lock(&slp->ns_lock); 2593 KASSERT((slp->ns_flag & SLP_SENDING) != 0); 2594 nd = SIMPLEQ_FIRST(&slp->ns_sendq); 2595 if (nd != NULL) { 2596 SIMPLEQ_REMOVE_HEAD(&slp->ns_sendq, nd_sendq); 2597 simple_unlock(&slp->ns_lock); 2598 goto again; 2599 } 2600 slp->ns_flag &= ~SLP_SENDING; 2601 simple_unlock(&slp->ns_lock); 2602 2603 return error; 2604 } 2605 #endif /* NFSSERVER */ 2606