nfs_socket.c revision 1.126.4.10
1 /* $NetBSD: nfs_socket.c,v 1.126.4.10 2006/04/19 05:06:37 elad 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 1.126.4.10 2006/04/19 05:06:37 elad 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 kauth_cred_t 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 kauth_cred_t acred; 990 #ifndef NFS_V2_ONLY 991 int nqlflag, cachable; 992 u_quad_t frev; 993 #endif 994 struct mbuf *mrest_backup = NULL; 995 kauth_cred_t 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 acred = kauth_cred_alloc(); 1003 kauth_cred_hold(acred); /* Just to be safe.. */ 1004 1005 tryagain_cred: 1006 KASSERT(cred != NULL); 1007 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 1008 rep->r_nmp = nmp; 1009 rep->r_lwp = lwp; 1010 rep->r_procnum = procnum; 1011 i = 0; 1012 m = mrest; 1013 while (m) { 1014 i += m->m_len; 1015 m = m->m_next; 1016 } 1017 mrest_len = i; 1018 1019 /* 1020 * Get the RPC header with authorization. 1021 */ 1022 kerbauth: 1023 verf_str = auth_str = (char *)0; 1024 if (nmp->nm_flag & NFSMNT_KERB) { 1025 verf_str = nickv; 1026 verf_len = sizeof (nickv); 1027 auth_type = RPCAUTH_KERB4; 1028 memset((caddr_t)key, 0, sizeof (key)); 1029 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 1030 &auth_len, verf_str, verf_len)) { 1031 error = nfs_getauth(nmp, rep, cred, &auth_str, 1032 &auth_len, verf_str, &verf_len, key); 1033 if (error) { 1034 free((caddr_t)rep, M_NFSREQ); 1035 m_freem(mrest); 1036 kauth_cred_destroy(acred); 1037 return (error); 1038 } 1039 } 1040 retry_cred = FALSE; 1041 } else { 1042 /* AUTH_UNIX */ 1043 uid_t uid; 1044 gid_t gid; 1045 1046 /* 1047 * on the most unix filesystems, permission checks are 1048 * done when the file is open(2)'ed. 1049 * ie. once a file is successfully open'ed, 1050 * following i/o operations never fail with EACCES. 1051 * we try to follow the semantics as far as possible. 1052 * 1053 * note that we expect that the nfs server always grant 1054 * accesses by the file's owner. 1055 */ 1056 origcred = cred; 1057 switch (procnum) { 1058 case NFSPROC_READ: 1059 case NFSPROC_WRITE: 1060 case NFSPROC_COMMIT: 1061 uid = np->n_vattr->va_uid; 1062 gid = np->n_vattr->va_gid; 1063 if (kauth_cred_geteuid(cred) == uid && 1064 kauth_cred_getegid(cred) == gid) { 1065 retry_cred = FALSE; 1066 break; 1067 } 1068 if (use_opencred) 1069 break; 1070 kauth_cred_setuid(acred, uid); 1071 kauth_cred_seteuid(acred, uid); 1072 kauth_cred_setsvuid(acred, uid); 1073 kauth_cred_setgid(acred, gid); 1074 kauth_cred_setegid(acred, gid); 1075 kauth_cred_setsvgid(acred, gid); 1076 cred = acred; 1077 break; 1078 default: 1079 retry_cred = FALSE; 1080 break; 1081 } 1082 /* 1083 * backup mbuf chain if we can need it later to retry. 1084 * 1085 * XXX maybe we can keep a direct reference to 1086 * mrest without doing m_copym, but it's ...ugly. 1087 */ 1088 if (retry_cred) 1089 mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT); 1090 auth_type = RPCAUTH_UNIX; 1091 /* XXX elad - ngroups */ 1092 auth_len = (((kauth_cred_ngroups(cred) > nmp->nm_numgrps) ? 1093 nmp->nm_numgrps : kauth_cred_ngroups(cred)) << 2) + 1094 5 * NFSX_UNSIGNED; 1095 } 1096 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 1097 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 1098 if (auth_str) 1099 free(auth_str, M_TEMP); 1100 1101 /* 1102 * For stream protocols, insert a Sun RPC Record Mark. 1103 */ 1104 if (nmp->nm_sotype == SOCK_STREAM) { 1105 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 1106 *mtod(m, u_int32_t *) = htonl(0x80000000 | 1107 (m->m_pkthdr.len - NFSX_UNSIGNED)); 1108 } 1109 rep->r_mreq = m; 1110 rep->r_xid = xid; 1111 tryagain: 1112 if (nmp->nm_flag & NFSMNT_SOFT) 1113 rep->r_retry = nmp->nm_retry; 1114 else 1115 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1116 rep->r_rtt = rep->r_rexmit = 0; 1117 if (proct[procnum] > 0) 1118 rep->r_flags = R_TIMING; 1119 else 1120 rep->r_flags = 0; 1121 rep->r_mrep = NULL; 1122 1123 /* 1124 * Do the client side RPC. 1125 */ 1126 nfsstats.rpcrequests++; 1127 /* 1128 * Chain request into list of outstanding requests. Be sure 1129 * to put it LAST so timer finds oldest requests first. 1130 */ 1131 s = splsoftnet(); 1132 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1133 1134 /* Get send time for nqnfs */ 1135 reqtime = time.tv_sec; 1136 1137 /* 1138 * If backing off another request or avoiding congestion, don't 1139 * send this one now but let timer do it. If not timing a request, 1140 * do it now. 1141 */ 1142 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1143 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1144 nmp->nm_sent < nmp->nm_cwnd)) { 1145 splx(s); 1146 if (nmp->nm_soflags & PR_CONNREQUIRED) 1147 error = nfs_sndlock(&nmp->nm_iflag, rep); 1148 if (!error) { 1149 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1150 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp); 1151 if (nmp->nm_soflags & PR_CONNREQUIRED) 1152 nfs_sndunlock(&nmp->nm_iflag); 1153 } 1154 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1155 nmp->nm_sent += NFS_CWNDSCALE; 1156 rep->r_flags |= R_SENT; 1157 } 1158 } else { 1159 splx(s); 1160 rep->r_rtt = -1; 1161 } 1162 1163 /* 1164 * Wait for the reply from our send or the timer's. 1165 */ 1166 if (!error || error == EPIPE) 1167 error = nfs_reply(rep, lwp); 1168 1169 /* 1170 * RPC done, unlink the request. 1171 */ 1172 s = splsoftnet(); 1173 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1174 splx(s); 1175 1176 /* 1177 * Decrement the outstanding request count. 1178 */ 1179 if (rep->r_flags & R_SENT) { 1180 rep->r_flags &= ~R_SENT; /* paranoia */ 1181 nmp->nm_sent -= NFS_CWNDSCALE; 1182 } 1183 1184 if (rexmitp != NULL) { 1185 int rexmit; 1186 1187 if (nmp->nm_sotype != SOCK_DGRAM) 1188 rexmit = (rep->r_flags & R_REXMITTED) != 0; 1189 else 1190 rexmit = rep->r_rexmit; 1191 *rexmitp = rexmit; 1192 } 1193 1194 /* 1195 * If there was a successful reply and a tprintf msg. 1196 * tprintf a response. 1197 */ 1198 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1199 nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1200 "is alive again"); 1201 mrep = rep->r_mrep; 1202 md = rep->r_md; 1203 dpos = rep->r_dpos; 1204 if (error) 1205 goto nfsmout; 1206 1207 /* 1208 * break down the rpc header and check if ok 1209 */ 1210 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1211 if (*tl++ == rpc_msgdenied) { 1212 if (*tl == rpc_mismatch) 1213 error = EOPNOTSUPP; 1214 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1215 if (!failed_auth) { 1216 failed_auth++; 1217 mheadend->m_next = (struct mbuf *)0; 1218 m_freem(mrep); 1219 m_freem(rep->r_mreq); 1220 goto kerbauth; 1221 } else 1222 error = EAUTH; 1223 } else 1224 error = EACCES; 1225 m_freem(mrep); 1226 goto nfsmout; 1227 } 1228 1229 /* 1230 * Grab any Kerberos verifier, otherwise just throw it away. 1231 */ 1232 verf_type = fxdr_unsigned(int, *tl++); 1233 i = fxdr_unsigned(int32_t, *tl); 1234 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1235 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1236 if (error) 1237 goto nfsmout; 1238 } else if (i > 0) 1239 nfsm_adv(nfsm_rndup(i)); 1240 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1241 /* 0 == ok */ 1242 if (*tl == 0) { 1243 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1244 if (*tl != 0) { 1245 error = fxdr_unsigned(int, *tl); 1246 switch (error) { 1247 case NFSERR_PERM: 1248 error = EPERM; 1249 break; 1250 1251 case NFSERR_NOENT: 1252 error = ENOENT; 1253 break; 1254 1255 case NFSERR_IO: 1256 error = EIO; 1257 break; 1258 1259 case NFSERR_NXIO: 1260 error = ENXIO; 1261 break; 1262 1263 case NFSERR_ACCES: 1264 error = EACCES; 1265 if (!retry_cred) 1266 break; 1267 m_freem(mrep); 1268 m_freem(rep->r_mreq); 1269 FREE(rep, M_NFSREQ); 1270 use_opencred = !use_opencred; 1271 if (mrest_backup == NULL) { 1272 /* m_copym failure */ 1273 kauth_cred_destroy(acred); 1274 return ENOMEM; 1275 } 1276 mrest = mrest_backup; 1277 mrest_backup = NULL; 1278 cred = origcred; 1279 error = 0; 1280 retry_cred = FALSE; 1281 goto tryagain_cred; 1282 1283 case NFSERR_EXIST: 1284 error = EEXIST; 1285 break; 1286 1287 case NFSERR_XDEV: 1288 error = EXDEV; 1289 break; 1290 1291 case NFSERR_NODEV: 1292 error = ENODEV; 1293 break; 1294 1295 case NFSERR_NOTDIR: 1296 error = ENOTDIR; 1297 break; 1298 1299 case NFSERR_ISDIR: 1300 error = EISDIR; 1301 break; 1302 1303 case NFSERR_INVAL: 1304 error = EINVAL; 1305 break; 1306 1307 case NFSERR_FBIG: 1308 error = EFBIG; 1309 break; 1310 1311 case NFSERR_NOSPC: 1312 error = ENOSPC; 1313 break; 1314 1315 case NFSERR_ROFS: 1316 error = EROFS; 1317 break; 1318 1319 case NFSERR_MLINK: 1320 error = EMLINK; 1321 break; 1322 1323 case NFSERR_TIMEDOUT: 1324 error = ETIMEDOUT; 1325 break; 1326 1327 case NFSERR_NAMETOL: 1328 error = ENAMETOOLONG; 1329 break; 1330 1331 case NFSERR_NOTEMPTY: 1332 error = ENOTEMPTY; 1333 break; 1334 1335 case NFSERR_DQUOT: 1336 error = EDQUOT; 1337 break; 1338 1339 case NFSERR_STALE: 1340 /* 1341 * If the File Handle was stale, invalidate the 1342 * lookup cache, just in case. 1343 */ 1344 error = ESTALE; 1345 cache_purge(NFSTOV(np)); 1346 break; 1347 1348 case NFSERR_REMOTE: 1349 error = EREMOTE; 1350 break; 1351 1352 case NFSERR_WFLUSH: 1353 case NFSERR_BADHANDLE: 1354 case NFSERR_NOT_SYNC: 1355 case NFSERR_BAD_COOKIE: 1356 error = EINVAL; 1357 break; 1358 1359 case NFSERR_NOTSUPP: 1360 error = ENOTSUP; 1361 break; 1362 1363 case NFSERR_TOOSMALL: 1364 case NFSERR_SERVERFAULT: 1365 case NFSERR_BADTYPE: 1366 error = EINVAL; 1367 break; 1368 1369 case NFSERR_TRYLATER: 1370 if ((nmp->nm_flag & NFSMNT_NFSV3) == 0) 1371 break; 1372 m_freem(mrep); 1373 error = 0; 1374 waituntil = time.tv_sec + trylater_delay; 1375 while (time.tv_sec < waituntil) 1376 (void) tsleep((caddr_t)&lbolt, 1377 PSOCK, "nqnfstry", 0); 1378 trylater_delay *= NFS_TRYLATERDELMUL; 1379 if (trylater_delay > NFS_TRYLATERDELMAX) 1380 trylater_delay = NFS_TRYLATERDELMAX; 1381 /* 1382 * RFC1813: 1383 * The client should wait and then try 1384 * the request with a new RPC transaction ID. 1385 */ 1386 nfs_renewxid(rep); 1387 goto tryagain; 1388 1389 case NFSERR_STALEWRITEVERF: 1390 error = EINVAL; 1391 break; 1392 1393 default: 1394 #ifdef DIAGNOSTIC 1395 printf("Invalid rpc error code %d\n", error); 1396 #endif 1397 error = EINVAL; 1398 break; 1399 } 1400 1401 if (nmp->nm_flag & NFSMNT_NFSV3) { 1402 *mrp = mrep; 1403 *mdp = md; 1404 *dposp = dpos; 1405 error |= NFSERR_RETERR; 1406 } else 1407 m_freem(mrep); 1408 goto nfsmout; 1409 } 1410 1411 /* 1412 * note which credential worked to minimize number of retries. 1413 */ 1414 if (use_opencred) 1415 np->n_flag |= NUSEOPENCRED; 1416 else 1417 np->n_flag &= ~NUSEOPENCRED; 1418 1419 #ifndef NFS_V2_ONLY 1420 /* 1421 * For nqnfs, get any lease in reply 1422 */ 1423 if (nmp->nm_flag & NFSMNT_NQNFS) { 1424 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1425 if (*tl) { 1426 nqlflag = fxdr_unsigned(int, *tl); 1427 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1428 cachable = fxdr_unsigned(int, *tl++); 1429 reqtime += fxdr_unsigned(int, *tl++); 1430 if (reqtime > time.tv_sec) { 1431 frev = fxdr_hyper(tl); 1432 nqnfs_clientlease(nmp, np, nqlflag, 1433 cachable, reqtime, frev); 1434 } 1435 } 1436 } 1437 #endif 1438 *mrp = mrep; 1439 *mdp = md; 1440 *dposp = dpos; 1441 1442 KASSERT(error == 0); 1443 goto nfsmout; 1444 } 1445 m_freem(mrep); 1446 error = EPROTONOSUPPORT; 1447 nfsmout: 1448 kauth_cred_destroy(acred); 1449 m_freem(rep->r_mreq); 1450 free((caddr_t)rep, M_NFSREQ); 1451 m_freem(mrest_backup); 1452 return (error); 1453 } 1454 #endif /* NFS */ 1455 1456 /* 1457 * Generate the rpc reply header 1458 * siz arg. is used to decide if adding a cluster is worthwhile 1459 */ 1460 int 1461 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1462 int siz; 1463 struct nfsrv_descript *nd; 1464 struct nfssvc_sock *slp; 1465 int err; 1466 int cache; 1467 u_quad_t *frev; 1468 struct mbuf **mrq; 1469 struct mbuf **mbp; 1470 caddr_t *bposp; 1471 { 1472 u_int32_t *tl; 1473 struct mbuf *mreq; 1474 caddr_t bpos; 1475 struct mbuf *mb; 1476 1477 mreq = m_gethdr(M_WAIT, MT_DATA); 1478 MCLAIM(mreq, &nfs_mowner); 1479 mb = mreq; 1480 /* 1481 * If this is a big reply, use a cluster else 1482 * try and leave leading space for the lower level headers. 1483 */ 1484 siz += RPC_REPLYSIZ; 1485 if (siz >= max_datalen) { 1486 m_clget(mreq, M_WAIT); 1487 } else 1488 mreq->m_data += max_hdr; 1489 tl = mtod(mreq, u_int32_t *); 1490 mreq->m_len = 6 * NFSX_UNSIGNED; 1491 bpos = ((caddr_t)tl) + mreq->m_len; 1492 *tl++ = txdr_unsigned(nd->nd_retxid); 1493 *tl++ = rpc_reply; 1494 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1495 *tl++ = rpc_msgdenied; 1496 if (err & NFSERR_AUTHERR) { 1497 *tl++ = rpc_autherr; 1498 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1499 mreq->m_len -= NFSX_UNSIGNED; 1500 bpos -= NFSX_UNSIGNED; 1501 } else { 1502 *tl++ = rpc_mismatch; 1503 *tl++ = txdr_unsigned(RPC_VER2); 1504 *tl = txdr_unsigned(RPC_VER2); 1505 } 1506 } else { 1507 *tl++ = rpc_msgaccepted; 1508 1509 /* 1510 * For Kerberos authentication, we must send the nickname 1511 * verifier back, otherwise just RPCAUTH_NULL. 1512 */ 1513 if (nd->nd_flag & ND_KERBFULL) { 1514 struct nfsuid *nuidp; 1515 struct timeval ktvin, ktvout; 1516 1517 LIST_FOREACH(nuidp, NUIDHASH(slp, kauth_cred_geteuid(nd->nd_cr)), 1518 nu_hash) { 1519 if (kauth_cred_geteuid(nuidp->nu_cr) == kauth_cred_geteuid(nd->nd_cr) && 1520 (!nd->nd_nam2 || netaddr_match( 1521 NU_NETFAM(nuidp), &nuidp->nu_haddr, 1522 nd->nd_nam2))) 1523 break; 1524 } 1525 if (nuidp) { 1526 ktvin.tv_sec = 1527 txdr_unsigned(nuidp->nu_timestamp.tv_sec 1528 - 1); 1529 ktvin.tv_usec = 1530 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1531 1532 /* 1533 * Encrypt the timestamp in ecb mode using the 1534 * session key. 1535 */ 1536 #ifdef NFSKERB 1537 XXX 1538 #endif 1539 1540 *tl++ = rpc_auth_kerb; 1541 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1542 *tl = ktvout.tv_sec; 1543 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1544 *tl++ = ktvout.tv_usec; 1545 *tl++ = txdr_unsigned(kauth_cred_geteuid(nuidp->nu_cr)); 1546 } else { 1547 *tl++ = 0; 1548 *tl++ = 0; 1549 } 1550 } else { 1551 *tl++ = 0; 1552 *tl++ = 0; 1553 } 1554 switch (err) { 1555 case EPROGUNAVAIL: 1556 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1557 break; 1558 case EPROGMISMATCH: 1559 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1560 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1561 if (nd->nd_flag & ND_NQNFS) { 1562 *tl++ = txdr_unsigned(3); 1563 *tl = txdr_unsigned(3); 1564 } else { 1565 *tl++ = txdr_unsigned(2); 1566 *tl = txdr_unsigned(3); 1567 } 1568 break; 1569 case EPROCUNAVAIL: 1570 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1571 break; 1572 case EBADRPC: 1573 *tl = txdr_unsigned(RPC_GARBAGE); 1574 break; 1575 default: 1576 *tl = 0; 1577 if (err != NFSERR_RETVOID) { 1578 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1579 if (err) 1580 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1581 else 1582 *tl = 0; 1583 } 1584 break; 1585 }; 1586 } 1587 1588 /* 1589 * For nqnfs, piggyback lease as requested. 1590 */ 1591 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1592 if (nd->nd_flag & ND_LEASE) { 1593 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1594 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1595 *tl++ = txdr_unsigned(cache); 1596 *tl++ = txdr_unsigned(nd->nd_duration); 1597 txdr_hyper(*frev, tl); 1598 } else { 1599 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1600 *tl = 0; 1601 } 1602 } 1603 if (mrq != NULL) 1604 *mrq = mreq; 1605 *mbp = mb; 1606 *bposp = bpos; 1607 if (err != 0 && err != NFSERR_RETVOID) 1608 nfsstats.srvrpc_errs++; 1609 return (0); 1610 } 1611 1612 /* 1613 * Nfs timer routine 1614 * Scan the nfsreq list and retranmit any requests that have timed out 1615 * To avoid retransmission attempts on STREAM sockets (in the future) make 1616 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1617 */ 1618 void 1619 nfs_timer(arg) 1620 void *arg; /* never used */ 1621 { 1622 struct nfsreq *rep; 1623 struct mbuf *m; 1624 struct socket *so; 1625 struct nfsmount *nmp; 1626 int timeo; 1627 int s, error; 1628 #ifdef NFSSERVER 1629 struct nfssvc_sock *slp; 1630 static long lasttime = 0; 1631 u_quad_t cur_usec; 1632 #endif 1633 1634 s = splsoftnet(); 1635 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1636 nmp = rep->r_nmp; 1637 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1638 continue; 1639 if (nfs_sigintr(nmp, rep, rep->r_lwp)) { 1640 rep->r_flags |= R_SOFTTERM; 1641 continue; 1642 } 1643 if (rep->r_rtt >= 0) { 1644 rep->r_rtt++; 1645 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1646 timeo = nmp->nm_timeo; 1647 else 1648 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1649 if (nmp->nm_timeouts > 0) 1650 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1651 if (rep->r_rtt <= timeo) 1652 continue; 1653 if (nmp->nm_timeouts < 1654 (sizeof(nfs_backoff) / sizeof(nfs_backoff[0]))) 1655 nmp->nm_timeouts++; 1656 } 1657 /* 1658 * Check for server not responding 1659 */ 1660 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1661 rep->r_rexmit > nmp->nm_deadthresh) { 1662 nfs_msg(rep->r_lwp, 1663 nmp->nm_mountp->mnt_stat.f_mntfromname, 1664 "not responding"); 1665 rep->r_flags |= R_TPRINTFMSG; 1666 } 1667 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1668 nfsstats.rpctimeouts++; 1669 rep->r_flags |= R_SOFTTERM; 1670 continue; 1671 } 1672 if (nmp->nm_sotype != SOCK_DGRAM) { 1673 if (++rep->r_rexmit > NFS_MAXREXMIT) 1674 rep->r_rexmit = NFS_MAXREXMIT; 1675 continue; 1676 } 1677 if ((so = nmp->nm_so) == NULL) 1678 continue; 1679 1680 /* 1681 * If there is enough space and the window allows.. 1682 * Resend it 1683 * Set r_rtt to -1 in case we fail to send it now. 1684 */ 1685 rep->r_rtt = -1; 1686 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1687 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1688 (rep->r_flags & R_SENT) || 1689 nmp->nm_sent < nmp->nm_cwnd) && 1690 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1691 if (so->so_state & SS_ISCONNECTED) 1692 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1693 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1694 else 1695 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1696 nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0); 1697 if (error) { 1698 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1699 #ifdef DEBUG 1700 printf("nfs_timer: ignoring error %d\n", 1701 error); 1702 #endif 1703 so->so_error = 0; 1704 } 1705 } else { 1706 /* 1707 * Iff first send, start timing 1708 * else turn timing off, backoff timer 1709 * and divide congestion window by 2. 1710 */ 1711 if (rep->r_flags & R_SENT) { 1712 rep->r_flags &= ~R_TIMING; 1713 if (++rep->r_rexmit > NFS_MAXREXMIT) 1714 rep->r_rexmit = NFS_MAXREXMIT; 1715 nmp->nm_cwnd >>= 1; 1716 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1717 nmp->nm_cwnd = NFS_CWNDSCALE; 1718 nfsstats.rpcretries++; 1719 } else { 1720 rep->r_flags |= R_SENT; 1721 nmp->nm_sent += NFS_CWNDSCALE; 1722 } 1723 rep->r_rtt = 0; 1724 } 1725 } 1726 } 1727 1728 #ifdef NFSSERVER 1729 /* 1730 * Call the nqnfs server timer once a second to handle leases. 1731 */ 1732 if (lasttime != time.tv_sec) { 1733 lasttime = time.tv_sec; 1734 nqnfs_serverd(); 1735 } 1736 1737 /* 1738 * Scan the write gathering queues for writes that need to be 1739 * completed now. 1740 */ 1741 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1742 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1743 if (LIST_FIRST(&slp->ns_tq) && 1744 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) 1745 nfsrv_wakenfsd(slp); 1746 } 1747 #endif /* NFSSERVER */ 1748 splx(s); 1749 callout_schedule(&nfs_timer_ch, nfs_ticks); 1750 } 1751 1752 /*ARGSUSED*/ 1753 void 1754 nfs_exit(p, v) 1755 struct proc *p; 1756 void *v; 1757 { 1758 struct nfsreq *rp; 1759 int s = splsoftnet(); 1760 1761 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 1762 if (rp->r_lwp && rp->r_lwp->l_proc == p) 1763 TAILQ_REMOVE(&nfs_reqq, rp, r_chain); 1764 } 1765 splx(s); 1766 } 1767 1768 /* 1769 * Test for a termination condition pending on the process. 1770 * This is used for NFSMNT_INT mounts. 1771 */ 1772 int 1773 nfs_sigintr(nmp, rep, l) 1774 struct nfsmount *nmp; 1775 struct nfsreq *rep; 1776 struct lwp *l; 1777 { 1778 sigset_t ss; 1779 1780 if (rep && (rep->r_flags & R_SOFTTERM)) 1781 return (EINTR); 1782 if (!(nmp->nm_flag & NFSMNT_INT)) 1783 return (0); 1784 if (l) { 1785 sigpending1(l->l_proc, &ss); 1786 #if 0 1787 sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss); 1788 #endif 1789 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1790 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1791 sigismember(&ss, SIGQUIT)) 1792 return (EINTR); 1793 } 1794 return (0); 1795 } 1796 1797 /* 1798 * Lock a socket against others. 1799 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1800 * and also to avoid race conditions between the processes with nfs requests 1801 * in progress when a reconnect is necessary. 1802 */ 1803 int 1804 nfs_sndlock(flagp, rep) 1805 int *flagp; 1806 struct nfsreq *rep; 1807 { 1808 struct lwp *l; 1809 int slpflag = 0, slptimeo = 0; 1810 1811 if (rep) { 1812 l = rep->r_lwp; 1813 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1814 slpflag = PCATCH; 1815 } else 1816 l = (struct lwp *)0; 1817 while (*flagp & NFSMNT_SNDLOCK) { 1818 if (rep && nfs_sigintr(rep->r_nmp, rep, l)) 1819 return (EINTR); 1820 *flagp |= NFSMNT_WANTSND; 1821 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1822 slptimeo); 1823 if (slpflag == PCATCH) { 1824 slpflag = 0; 1825 slptimeo = 2 * hz; 1826 } 1827 } 1828 *flagp |= NFSMNT_SNDLOCK; 1829 return (0); 1830 } 1831 1832 /* 1833 * Unlock the stream socket for others. 1834 */ 1835 void 1836 nfs_sndunlock(flagp) 1837 int *flagp; 1838 { 1839 1840 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1841 panic("nfs sndunlock"); 1842 *flagp &= ~NFSMNT_SNDLOCK; 1843 if (*flagp & NFSMNT_WANTSND) { 1844 *flagp &= ~NFSMNT_WANTSND; 1845 wakeup((caddr_t)flagp); 1846 } 1847 } 1848 1849 int 1850 nfs_rcvlock(rep) 1851 struct nfsreq *rep; 1852 { 1853 struct nfsmount *nmp = rep->r_nmp; 1854 int *flagp = &nmp->nm_iflag; 1855 int slpflag, slptimeo = 0; 1856 int error = 0; 1857 1858 if (*flagp & NFSMNT_DISMNT) 1859 return EIO; 1860 1861 if (*flagp & NFSMNT_INT) 1862 slpflag = PCATCH; 1863 else 1864 slpflag = 0; 1865 simple_lock(&nmp->nm_slock); 1866 while (*flagp & NFSMNT_RCVLOCK) { 1867 if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) { 1868 error = EINTR; 1869 goto quit; 1870 } 1871 *flagp |= NFSMNT_WANTRCV; 1872 nmp->nm_waiters++; 1873 (void) ltsleep(flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1874 slptimeo, &nmp->nm_slock); 1875 nmp->nm_waiters--; 1876 if (*flagp & NFSMNT_DISMNT) { 1877 wakeup(&nmp->nm_waiters); 1878 error = EIO; 1879 goto quit; 1880 } 1881 /* If our reply was received while we were sleeping, 1882 * then just return without taking the lock to avoid a 1883 * situation where a single iod could 'capture' the 1884 * receive lock. 1885 */ 1886 if (rep->r_mrep != NULL) { 1887 error = EALREADY; 1888 goto quit; 1889 } 1890 if (slpflag == PCATCH) { 1891 slpflag = 0; 1892 slptimeo = 2 * hz; 1893 } 1894 } 1895 *flagp |= NFSMNT_RCVLOCK; 1896 quit: 1897 simple_unlock(&nmp->nm_slock); 1898 return error; 1899 } 1900 1901 /* 1902 * Unlock the stream socket for others. 1903 */ 1904 void 1905 nfs_rcvunlock(nmp) 1906 struct nfsmount *nmp; 1907 { 1908 int *flagp = &nmp->nm_iflag; 1909 1910 simple_lock(&nmp->nm_slock); 1911 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1912 panic("nfs rcvunlock"); 1913 *flagp &= ~NFSMNT_RCVLOCK; 1914 if (*flagp & NFSMNT_WANTRCV) { 1915 *flagp &= ~NFSMNT_WANTRCV; 1916 wakeup((caddr_t)flagp); 1917 } 1918 simple_unlock(&nmp->nm_slock); 1919 } 1920 1921 /* 1922 * Parse an RPC request 1923 * - verify it 1924 * - fill in the cred struct. 1925 */ 1926 int 1927 nfs_getreq(nd, nfsd, has_header) 1928 struct nfsrv_descript *nd; 1929 struct nfsd *nfsd; 1930 int has_header; 1931 { 1932 int len, i; 1933 u_int32_t *tl; 1934 int32_t t1; 1935 struct uio uio; 1936 struct iovec iov; 1937 caddr_t dpos, cp2, cp; 1938 u_int32_t nfsvers, auth_type; 1939 uid_t nickuid; 1940 int error = 0, nqnfs = 0, ticklen; 1941 struct mbuf *mrep, *md; 1942 struct nfsuid *nuidp; 1943 struct timeval tvin, tvout; 1944 1945 mrep = nd->nd_mrep; 1946 md = nd->nd_md; 1947 dpos = nd->nd_dpos; 1948 if (has_header) { 1949 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1950 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1951 if (*tl++ != rpc_call) { 1952 m_freem(mrep); 1953 return (EBADRPC); 1954 } 1955 } else 1956 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1957 nd->nd_repstat = 0; 1958 nd->nd_flag = 0; 1959 if (*tl++ != rpc_vers) { 1960 nd->nd_repstat = ERPCMISMATCH; 1961 nd->nd_procnum = NFSPROC_NOOP; 1962 return (0); 1963 } 1964 if (*tl != nfs_prog) { 1965 if (*tl == nqnfs_prog) 1966 nqnfs++; 1967 else { 1968 nd->nd_repstat = EPROGUNAVAIL; 1969 nd->nd_procnum = NFSPROC_NOOP; 1970 return (0); 1971 } 1972 } 1973 tl++; 1974 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1975 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1976 (nfsvers != NQNFS_VER3 && nqnfs)) { 1977 nd->nd_repstat = EPROGMISMATCH; 1978 nd->nd_procnum = NFSPROC_NOOP; 1979 return (0); 1980 } 1981 if (nqnfs) 1982 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1983 else if (nfsvers == NFS_VER3) 1984 nd->nd_flag = ND_NFSV3; 1985 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1986 if (nd->nd_procnum == NFSPROC_NULL) 1987 return (0); 1988 if (nd->nd_procnum >= NFS_NPROCS || 1989 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 1990 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1991 nd->nd_repstat = EPROCUNAVAIL; 1992 nd->nd_procnum = NFSPROC_NOOP; 1993 return (0); 1994 } 1995 if ((nd->nd_flag & ND_NFSV3) == 0) 1996 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1997 auth_type = *tl++; 1998 len = fxdr_unsigned(int, *tl++); 1999 if (len < 0 || len > RPCAUTH_MAXSIZ) { 2000 m_freem(mrep); 2001 return (EBADRPC); 2002 } 2003 2004 nd->nd_flag &= ~ND_KERBAUTH; 2005 /* 2006 * Handle auth_unix or auth_kerb. 2007 */ 2008 if (auth_type == rpc_auth_unix) { 2009 uid_t uid; 2010 gid_t gid, *grbuf; 2011 2012 len = fxdr_unsigned(int, *++tl); 2013 if (len < 0 || len > NFS_MAXNAMLEN) { 2014 m_freem(mrep); 2015 return (EBADRPC); 2016 } 2017 nfsm_adv(nfsm_rndup(len)); 2018 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2019 2020 nd->nd_cr = kauth_cred_alloc(); 2021 2022 uid = fxdr_unsigned(uid_t, *tl++); 2023 gid = fxdr_unsigned(gid_t, *tl++); 2024 kauth_cred_setuid(nd->nd_cr, uid); 2025 kauth_cred_setgid(nd->nd_cr, gid); 2026 kauth_cred_seteuid(nd->nd_cr, uid); 2027 kauth_cred_setsvuid(nd->nd_cr, gid); 2028 kauth_cred_setegid(nd->nd_cr, uid); 2029 kauth_cred_setsvgid(nd->nd_cr, gid); 2030 2031 len = fxdr_unsigned(int, *tl); 2032 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 2033 kauth_cred_free(nd->nd_cr); 2034 nd->nd_cr = NULL; 2035 m_freem(mrep); 2036 return (EBADRPC); 2037 } 2038 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 2039 2040 grbuf = malloc(len * sizeof(gid_t), M_TEMP, M_WAITOK); 2041 for (i = 0; i < len; i++) { 2042 if (i < NGROUPS) /* XXX elad */ 2043 grbuf[i] = fxdr_unsigned(gid_t, *tl++); 2044 else 2045 tl++; 2046 } 2047 kauth_cred_setgroups(nd->nd_cr, grbuf, min(len, NGROUPS), -1); 2048 free(grbuf, M_TEMP); 2049 2050 len = fxdr_unsigned(int, *++tl); 2051 if (len < 0 || len > RPCAUTH_MAXSIZ) { 2052 kauth_cred_free(nd->nd_cr); 2053 kauth_cred_free(nd->nd_cr); 2054 m_freem(mrep); 2055 return (EBADRPC); 2056 } 2057 if (len > 0) 2058 nfsm_adv(nfsm_rndup(len)); 2059 } else if (auth_type == rpc_auth_kerb) { 2060 switch (fxdr_unsigned(int, *tl++)) { 2061 case RPCAKN_FULLNAME: 2062 ticklen = fxdr_unsigned(int, *tl); 2063 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 2064 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 2065 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 2066 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 2067 m_freem(mrep); 2068 return (EBADRPC); 2069 } 2070 uio.uio_offset = 0; 2071 uio.uio_iov = &iov; 2072 uio.uio_iovcnt = 1; 2073 UIO_SETUP_SYSSPACE(&uio); 2074 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 2075 iov.iov_len = RPCAUTH_MAXSIZ - 4; 2076 nfsm_mtouio(&uio, uio.uio_resid); 2077 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2078 if (*tl++ != rpc_auth_kerb || 2079 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 2080 printf("Bad kerb verifier\n"); 2081 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2082 nd->nd_procnum = NFSPROC_NOOP; 2083 return (0); 2084 } 2085 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 2086 tl = (u_int32_t *)cp; 2087 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 2088 printf("Not fullname kerb verifier\n"); 2089 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2090 nd->nd_procnum = NFSPROC_NOOP; 2091 return (0); 2092 } 2093 cp += NFSX_UNSIGNED; 2094 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 2095 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 2096 nd->nd_flag |= ND_KERBFULL; 2097 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 2098 break; 2099 case RPCAKN_NICKNAME: 2100 if (len != 2 * NFSX_UNSIGNED) { 2101 printf("Kerb nickname short\n"); 2102 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 2103 nd->nd_procnum = NFSPROC_NOOP; 2104 return (0); 2105 } 2106 nickuid = fxdr_unsigned(uid_t, *tl); 2107 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2108 if (*tl++ != rpc_auth_kerb || 2109 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 2110 printf("Kerb nick verifier bad\n"); 2111 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2112 nd->nd_procnum = NFSPROC_NOOP; 2113 return (0); 2114 } 2115 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2116 tvin.tv_sec = *tl++; 2117 tvin.tv_usec = *tl; 2118 2119 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), 2120 nu_hash) { 2121 if (kauth_cred_geteuid(nuidp->nu_cr) == nickuid && 2122 (!nd->nd_nam2 || 2123 netaddr_match(NU_NETFAM(nuidp), 2124 &nuidp->nu_haddr, nd->nd_nam2))) 2125 break; 2126 } 2127 if (!nuidp) { 2128 nd->nd_repstat = 2129 (NFSERR_AUTHERR|AUTH_REJECTCRED); 2130 nd->nd_procnum = NFSPROC_NOOP; 2131 return (0); 2132 } 2133 2134 /* 2135 * Now, decrypt the timestamp using the session key 2136 * and validate it. 2137 */ 2138 #ifdef NFSKERB 2139 XXX 2140 #endif 2141 2142 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 2143 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 2144 if (nuidp->nu_expire < time.tv_sec || 2145 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 2146 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 2147 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 2148 nuidp->nu_expire = 0; 2149 nd->nd_repstat = 2150 (NFSERR_AUTHERR|AUTH_REJECTVERF); 2151 nd->nd_procnum = NFSPROC_NOOP; 2152 return (0); 2153 } 2154 nfsrv_setcred(nuidp->nu_cr, &nd->nd_cr); 2155 nd->nd_flag |= ND_KERBNICK; 2156 }; 2157 } else { 2158 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 2159 nd->nd_procnum = NFSPROC_NOOP; 2160 return (0); 2161 } 2162 2163 /* 2164 * For nqnfs, get piggybacked lease request. 2165 */ 2166 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 2167 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2168 nd->nd_flag |= fxdr_unsigned(int, *tl); 2169 if (nd->nd_flag & ND_LEASE) { 2170 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2171 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 2172 } else 2173 nd->nd_duration = NQ_MINLEASE; 2174 } else 2175 nd->nd_duration = NQ_MINLEASE; 2176 nd->nd_md = md; 2177 nd->nd_dpos = dpos; 2178 return (0); 2179 nfsmout: 2180 return (error); 2181 } 2182 2183 int 2184 nfs_msg(l, server, msg) 2185 struct lwp *l; 2186 const char *server, *msg; 2187 { 2188 tpr_t tpr; 2189 2190 if (l) 2191 tpr = tprintf_open(l->l_proc); 2192 else 2193 tpr = NULL; 2194 tprintf(tpr, "nfs server %s: %s\n", server, msg); 2195 tprintf_close(tpr); 2196 return (0); 2197 } 2198 2199 #ifdef NFSSERVER 2200 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 2201 struct nfssvc_sock *, struct lwp *, 2202 struct mbuf **)) = { 2203 nfsrv_null, 2204 nfsrv_getattr, 2205 nfsrv_setattr, 2206 nfsrv_lookup, 2207 nfsrv3_access, 2208 nfsrv_readlink, 2209 nfsrv_read, 2210 nfsrv_write, 2211 nfsrv_create, 2212 nfsrv_mkdir, 2213 nfsrv_symlink, 2214 nfsrv_mknod, 2215 nfsrv_remove, 2216 nfsrv_rmdir, 2217 nfsrv_rename, 2218 nfsrv_link, 2219 nfsrv_readdir, 2220 nfsrv_readdirplus, 2221 nfsrv_statfs, 2222 nfsrv_fsinfo, 2223 nfsrv_pathconf, 2224 nfsrv_commit, 2225 nqnfsrv_getlease, 2226 nqnfsrv_vacated, 2227 nfsrv_noop, 2228 nfsrv_noop 2229 }; 2230 2231 /* 2232 * Socket upcall routine for the nfsd sockets. 2233 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 2234 * Essentially do as much as possible non-blocking, else punt and it will 2235 * be called with M_WAIT from an nfsd. 2236 */ 2237 void 2238 nfsrv_rcv(so, arg, waitflag) 2239 struct socket *so; 2240 caddr_t arg; 2241 int waitflag; 2242 { 2243 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 2244 struct mbuf *m; 2245 struct mbuf *mp, *nam; 2246 struct uio auio; 2247 int flags, error; 2248 int setflags = 0; 2249 2250 error = nfsdsock_lock(slp, (waitflag != M_DONTWAIT)); 2251 if (error) { 2252 setflags |= SLP_NEEDQ; 2253 goto dorecs_unlocked; 2254 } 2255 2256 KASSERT(so == slp->ns_so); 2257 #define NFS_TEST_HEAVY 2258 #ifdef NFS_TEST_HEAVY 2259 /* 2260 * Define this to test for nfsds handling this under heavy load. 2261 * 2262 * XXX it isn't safe to call so_receive from so_upcall context. 2263 */ 2264 if (waitflag == M_DONTWAIT) { 2265 setflags |= SLP_NEEDQ; 2266 goto dorecs; 2267 } 2268 #endif 2269 simple_lock(&slp->ns_lock); 2270 slp->ns_flag &= ~SLP_NEEDQ; 2271 simple_unlock(&slp->ns_lock); 2272 if (so->so_type == SOCK_STREAM) { 2273 #ifndef NFS_TEST_HEAVY 2274 /* 2275 * If there are already records on the queue, defer soreceive() 2276 * to an nfsd so that there is feedback to the TCP layer that 2277 * the nfs servers are heavily loaded. 2278 */ 2279 if (slp->ns_rec && waitflag == M_DONTWAIT) { 2280 setflags |= SLP_NEEDQ; 2281 goto dorecs; 2282 } 2283 #endif 2284 2285 /* 2286 * Do soreceive(). 2287 */ 2288 auio.uio_resid = 1000000000; 2289 /* not need to setup uio_vmspace */ 2290 flags = MSG_DONTWAIT; 2291 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, &flags); 2292 if (error || mp == NULL) { 2293 if (error == EWOULDBLOCK) 2294 setflags |= SLP_NEEDQ; 2295 else 2296 setflags |= SLP_DISCONN; 2297 goto dorecs; 2298 } 2299 m = mp; 2300 if (slp->ns_rawend) { 2301 slp->ns_rawend->m_next = m; 2302 slp->ns_cc += 1000000000 - auio.uio_resid; 2303 } else { 2304 slp->ns_raw = m; 2305 slp->ns_cc = 1000000000 - auio.uio_resid; 2306 } 2307 while (m->m_next) 2308 m = m->m_next; 2309 slp->ns_rawend = m; 2310 2311 /* 2312 * Now try and parse record(s) out of the raw stream data. 2313 */ 2314 error = nfsrv_getstream(slp, waitflag); 2315 if (error) { 2316 if (error == EPERM) 2317 setflags |= SLP_DISCONN; 2318 else 2319 setflags |= SLP_NEEDQ; 2320 } 2321 } else { 2322 do { 2323 auio.uio_resid = 1000000000; 2324 /* not need to setup uio_vmspace */ 2325 flags = MSG_DONTWAIT; 2326 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, 2327 &flags); 2328 if (mp) { 2329 if (nam) { 2330 m = nam; 2331 m->m_next = mp; 2332 } else 2333 m = mp; 2334 if (slp->ns_recend) 2335 slp->ns_recend->m_nextpkt = m; 2336 else 2337 slp->ns_rec = m; 2338 slp->ns_recend = m; 2339 m->m_nextpkt = (struct mbuf *)0; 2340 } 2341 if (error) { 2342 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2343 && error != EWOULDBLOCK) { 2344 setflags |= SLP_DISCONN; 2345 goto dorecs; 2346 } 2347 } 2348 } while (mp); 2349 } 2350 dorecs: 2351 nfsdsock_unlock(slp); 2352 2353 dorecs_unlocked: 2354 /* 2355 * Now try and process the request records, non-blocking. 2356 */ 2357 if (setflags) { 2358 simple_lock(&slp->ns_lock); 2359 slp->ns_flag |= setflags; 2360 simple_unlock(&slp->ns_lock); 2361 } 2362 if (waitflag == M_DONTWAIT && 2363 (slp->ns_rec || (slp->ns_flag & (SLP_DISCONN | SLP_NEEDQ)) != 0)) { 2364 nfsrv_wakenfsd(slp); 2365 } 2366 } 2367 2368 int 2369 nfsdsock_lock(struct nfssvc_sock *slp, boolean_t waitok) 2370 { 2371 2372 simple_lock(&slp->ns_lock); 2373 while ((slp->ns_flag & (SLP_BUSY|SLP_VALID)) == SLP_BUSY) { 2374 if (!waitok) { 2375 simple_unlock(&slp->ns_lock); 2376 return EWOULDBLOCK; 2377 } 2378 slp->ns_flag |= SLP_WANT; 2379 ltsleep(&slp->ns_flag, PSOCK, "nslock", 0, &slp->ns_lock); 2380 } 2381 if ((slp->ns_flag & SLP_VALID) == 0) { 2382 simple_unlock(&slp->ns_lock); 2383 return EINVAL; 2384 } 2385 slp->ns_flag |= SLP_BUSY; 2386 simple_unlock(&slp->ns_lock); 2387 2388 return 0; 2389 } 2390 2391 void 2392 nfsdsock_unlock(struct nfssvc_sock *slp) 2393 { 2394 2395 KASSERT((slp->ns_flag & SLP_BUSY) != 0); 2396 2397 simple_lock(&slp->ns_lock); 2398 if ((slp->ns_flag & SLP_WANT) != 0) { 2399 wakeup(&slp->ns_flag); 2400 } 2401 slp->ns_flag &= ~(SLP_BUSY|SLP_WANT); 2402 simple_unlock(&slp->ns_lock); 2403 } 2404 2405 int 2406 nfsdsock_drain(struct nfssvc_sock *slp) 2407 { 2408 int error = 0; 2409 2410 simple_lock(&slp->ns_lock); 2411 if ((slp->ns_flag & SLP_VALID) == 0) { 2412 error = EINVAL; 2413 goto done; 2414 } 2415 slp->ns_flag &= ~SLP_VALID; 2416 while ((slp->ns_flag & SLP_BUSY) != 0) { 2417 slp->ns_flag |= SLP_WANT; 2418 ltsleep(&slp->ns_flag, PSOCK, "nsdrain", 0, &slp->ns_lock); 2419 } 2420 done: 2421 simple_unlock(&slp->ns_lock); 2422 2423 return error; 2424 } 2425 2426 /* 2427 * Try and extract an RPC request from the mbuf data list received on a 2428 * stream socket. The "waitflag" argument indicates whether or not it 2429 * can sleep. 2430 */ 2431 int 2432 nfsrv_getstream(slp, waitflag) 2433 struct nfssvc_sock *slp; 2434 int waitflag; 2435 { 2436 struct mbuf *m, **mpp; 2437 struct mbuf *recm; 2438 u_int32_t recmark; 2439 int error = 0; 2440 2441 for (;;) { 2442 if (slp->ns_reclen == 0) { 2443 if (slp->ns_cc < NFSX_UNSIGNED) { 2444 break; 2445 } 2446 m = slp->ns_raw; 2447 m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark); 2448 m_adj(m, NFSX_UNSIGNED); 2449 slp->ns_cc -= NFSX_UNSIGNED; 2450 recmark = ntohl(recmark); 2451 slp->ns_reclen = recmark & ~0x80000000; 2452 if (recmark & 0x80000000) 2453 slp->ns_flag |= SLP_LASTFRAG; 2454 else 2455 slp->ns_flag &= ~SLP_LASTFRAG; 2456 if (slp->ns_reclen > NFS_MAXPACKET) { 2457 error = EPERM; 2458 break; 2459 } 2460 } 2461 2462 /* 2463 * Now get the record part. 2464 * 2465 * Note that slp->ns_reclen may be 0. Linux sometimes 2466 * generates 0-length records. 2467 */ 2468 if (slp->ns_cc == slp->ns_reclen) { 2469 recm = slp->ns_raw; 2470 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2471 slp->ns_cc = slp->ns_reclen = 0; 2472 } else if (slp->ns_cc > slp->ns_reclen) { 2473 recm = slp->ns_raw; 2474 m = m_split(recm, slp->ns_reclen, waitflag); 2475 if (m == NULL) { 2476 error = EWOULDBLOCK; 2477 break; 2478 } 2479 m_claimm(recm, &nfs_mowner); 2480 slp->ns_raw = m; 2481 if (m->m_next == NULL) 2482 slp->ns_rawend = m; 2483 slp->ns_cc -= slp->ns_reclen; 2484 slp->ns_reclen = 0; 2485 } else { 2486 break; 2487 } 2488 2489 /* 2490 * Accumulate the fragments into a record. 2491 */ 2492 mpp = &slp->ns_frag; 2493 while (*mpp) 2494 mpp = &((*mpp)->m_next); 2495 *mpp = recm; 2496 if (slp->ns_flag & SLP_LASTFRAG) { 2497 if (slp->ns_recend) 2498 slp->ns_recend->m_nextpkt = slp->ns_frag; 2499 else 2500 slp->ns_rec = slp->ns_frag; 2501 slp->ns_recend = slp->ns_frag; 2502 slp->ns_frag = (struct mbuf *)0; 2503 } 2504 } 2505 2506 return error; 2507 } 2508 2509 /* 2510 * Parse an RPC header. 2511 */ 2512 int 2513 nfsrv_dorec(slp, nfsd, ndp) 2514 struct nfssvc_sock *slp; 2515 struct nfsd *nfsd; 2516 struct nfsrv_descript **ndp; 2517 { 2518 struct mbuf *m, *nam; 2519 struct nfsrv_descript *nd; 2520 int error; 2521 2522 *ndp = NULL; 2523 2524 if (nfsdsock_lock(slp, TRUE)) { 2525 return ENOBUFS; 2526 } 2527 m = slp->ns_rec; 2528 if (m == NULL) { 2529 nfsdsock_unlock(slp); 2530 return ENOBUFS; 2531 } 2532 slp->ns_rec = m->m_nextpkt; 2533 if (slp->ns_rec) 2534 m->m_nextpkt = NULL; 2535 else 2536 slp->ns_recend = NULL; 2537 nfsdsock_unlock(slp); 2538 2539 if (m->m_type == MT_SONAME) { 2540 nam = m; 2541 m = m->m_next; 2542 nam->m_next = NULL; 2543 } else 2544 nam = NULL; 2545 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK); 2546 nd->nd_md = nd->nd_mrep = m; 2547 nd->nd_nam2 = nam; 2548 nd->nd_dpos = mtod(m, caddr_t); 2549 error = nfs_getreq(nd, nfsd, TRUE); 2550 if (error) { 2551 m_freem(nam); 2552 pool_put(&nfs_srvdesc_pool, nd); 2553 return (error); 2554 } 2555 *ndp = nd; 2556 nfsd->nfsd_nd = nd; 2557 return (0); 2558 } 2559 2560 2561 /* 2562 * Search for a sleeping nfsd and wake it up. 2563 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2564 * running nfsds will go look for the work in the nfssvc_sock list. 2565 */ 2566 void 2567 nfsrv_wakenfsd(slp) 2568 struct nfssvc_sock *slp; 2569 { 2570 struct nfsd *nd; 2571 2572 if ((slp->ns_flag & SLP_VALID) == 0) 2573 return; 2574 simple_lock(&nfsd_slock); 2575 if (slp->ns_flag & SLP_DOREC) { 2576 simple_unlock(&nfsd_slock); 2577 return; 2578 } 2579 nd = SLIST_FIRST(&nfsd_idle_head); 2580 if (nd) { 2581 SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle); 2582 simple_unlock(&nfsd_slock); 2583 2584 if (nd->nfsd_slp) 2585 panic("nfsd wakeup"); 2586 slp->ns_sref++; 2587 nd->nfsd_slp = slp; 2588 wakeup(nd); 2589 return; 2590 } 2591 slp->ns_flag |= SLP_DOREC; 2592 nfsd_head_flag |= NFSD_CHECKSLP; 2593 TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending); 2594 simple_unlock(&nfsd_slock); 2595 } 2596 2597 int 2598 nfsdsock_sendreply(struct nfssvc_sock *slp, struct nfsrv_descript *nd) 2599 { 2600 int error; 2601 2602 if (nd->nd_mrep != NULL) { 2603 m_freem(nd->nd_mrep); 2604 nd->nd_mrep = NULL; 2605 } 2606 2607 simple_lock(&slp->ns_lock); 2608 if ((slp->ns_flag & SLP_SENDING) != 0) { 2609 SIMPLEQ_INSERT_TAIL(&slp->ns_sendq, nd, nd_sendq); 2610 simple_unlock(&slp->ns_lock); 2611 return 0; 2612 } 2613 KASSERT(SIMPLEQ_EMPTY(&slp->ns_sendq)); 2614 slp->ns_flag |= SLP_SENDING; 2615 simple_unlock(&slp->ns_lock); 2616 2617 again: 2618 error = nfs_send(slp->ns_so, nd->nd_nam2, nd->nd_mreq, NULL, curlwp); 2619 if (nd->nd_nam2) { 2620 m_free(nd->nd_nam2); 2621 } 2622 pool_put(&nfs_srvdesc_pool, nd); 2623 2624 simple_lock(&slp->ns_lock); 2625 KASSERT((slp->ns_flag & SLP_SENDING) != 0); 2626 nd = SIMPLEQ_FIRST(&slp->ns_sendq); 2627 if (nd != NULL) { 2628 SIMPLEQ_REMOVE_HEAD(&slp->ns_sendq, nd_sendq); 2629 simple_unlock(&slp->ns_lock); 2630 goto again; 2631 } 2632 slp->ns_flag &= ~SLP_SENDING; 2633 simple_unlock(&slp->ns_lock); 2634 2635 return error; 2636 } 2637 #endif /* NFSSERVER */ 2638
Indexes created Tue Sep 30 11:09:46 GMT 2025