nfs_socket.c revision 1.145
1 /* $NetBSD: nfs_socket.c,v 1.145 2007/02/09 21:55:36 ad Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1991, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 35 */ 36 37 /* 38 * Socket operations for use by nfs 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: nfs_socket.c,v 1.145 2007/02/09 21:55:36 ad 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 #include <sys/kauth.h> 68 69 #include <netinet/in.h> 70 #include <netinet/tcp.h> 71 72 #include <nfs/rpcv2.h> 73 #include <nfs/nfsproto.h> 74 #include <nfs/nfs.h> 75 #include <nfs/xdr_subs.h> 76 #include <nfs/nfsm_subs.h> 77 #include <nfs/nfsmount.h> 78 #include <nfs/nfsnode.h> 79 #include <nfs/nfsrtt.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 = MOWNER_INIT("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; 113 extern const int nfsv3_procid[NFS_NPROCS]; 114 extern int nfs_ticks; 115 116 /* 117 * Defines which timer to use for the procnum. 118 * 0 - default 119 * 1 - getattr 120 * 2 - lookup 121 * 3 - read 122 * 4 - write 123 */ 124 static const int proct[NFS_NPROCS] = { 125 [NFSPROC_NULL] = 0, 126 [NFSPROC_GETATTR] = 1, 127 [NFSPROC_SETATTR] = 0, 128 [NFSPROC_LOOKUP] = 2, 129 [NFSPROC_ACCESS] = 1, 130 [NFSPROC_READLINK] = 3, 131 [NFSPROC_READ] = 3, 132 [NFSPROC_WRITE] = 4, 133 [NFSPROC_CREATE] = 0, 134 [NFSPROC_MKDIR] = 0, 135 [NFSPROC_SYMLINK] = 0, 136 [NFSPROC_MKNOD] = 0, 137 [NFSPROC_REMOVE] = 0, 138 [NFSPROC_RMDIR] = 0, 139 [NFSPROC_RENAME] = 0, 140 [NFSPROC_LINK] = 0, 141 [NFSPROC_READDIR] = 3, 142 [NFSPROC_READDIRPLUS] = 3, 143 [NFSPROC_FSSTAT] = 0, 144 [NFSPROC_FSINFO] = 0, 145 [NFSPROC_PATHCONF] = 0, 146 [NFSPROC_COMMIT] = 0, 147 [NFSPROC_NOOP] = 0, 148 }; 149 150 /* 151 * There is a congestion window for outstanding rpcs maintained per mount 152 * point. The cwnd size is adjusted in roughly the way that: 153 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 154 * SIGCOMM '88". ACM, August 1988. 155 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 156 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 157 * of rpcs is in progress. 158 * (The sent count and cwnd are scaled for integer arith.) 159 * Variants of "slow start" were tried and were found to be too much of a 160 * performance hit (ave. rtt 3 times larger), 161 * I suspect due to the large rtt that nfs rpcs have. 162 */ 163 #define NFS_CWNDSCALE 256 164 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 165 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 166 int nfsrtton = 0; 167 struct nfsrtt nfsrtt; 168 struct nfsreqhead nfs_reqq; 169 170 struct callout nfs_timer_ch = CALLOUT_INITIALIZER_SETFUNC(nfs_timer, NULL); 171 172 /* 173 * Initialize sockets and congestion for a new NFS connection. 174 * We do not free the sockaddr if error. 175 */ 176 int 177 nfs_connect(nmp, rep, l) 178 struct nfsmount *nmp; 179 struct nfsreq *rep; 180 struct lwp *l; 181 { 182 struct socket *so; 183 int s, error, rcvreserve, sndreserve; 184 struct sockaddr *saddr; 185 struct sockaddr_in *sin; 186 #ifdef INET6 187 struct sockaddr_in6 *sin6; 188 #endif 189 struct mbuf *m; 190 191 nmp->nm_so = (struct socket *)0; 192 saddr = mtod(nmp->nm_nam, struct sockaddr *); 193 error = socreate(saddr->sa_family, &nmp->nm_so, 194 nmp->nm_sotype, nmp->nm_soproto, l); 195 if (error) 196 goto bad; 197 so = nmp->nm_so; 198 #ifdef MBUFTRACE 199 so->so_mowner = &nfs_mowner; 200 so->so_rcv.sb_mowner = &nfs_mowner; 201 so->so_snd.sb_mowner = &nfs_mowner; 202 #endif 203 nmp->nm_soflags = so->so_proto->pr_flags; 204 205 /* 206 * Some servers require that the client port be a reserved port number. 207 */ 208 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 209 m = m_get(M_WAIT, MT_SOOPTS); 210 MCLAIM(m, so->so_mowner); 211 *mtod(m, int32_t *) = IP_PORTRANGE_LOW; 212 m->m_len = sizeof(int32_t); 213 if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m))) 214 goto bad; 215 m = m_get(M_WAIT, MT_SONAME); 216 MCLAIM(m, so->so_mowner); 217 sin = mtod(m, struct sockaddr_in *); 218 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 219 sin->sin_family = AF_INET; 220 sin->sin_addr.s_addr = INADDR_ANY; 221 sin->sin_port = 0; 222 error = sobind(so, m, &lwp0); 223 m_freem(m); 224 if (error) 225 goto bad; 226 } 227 #ifdef INET6 228 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) { 229 m = m_get(M_WAIT, MT_SOOPTS); 230 MCLAIM(m, so->so_mowner); 231 *mtod(m, int32_t *) = IPV6_PORTRANGE_LOW; 232 m->m_len = sizeof(int32_t); 233 if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m))) 234 goto bad; 235 m = m_get(M_WAIT, MT_SONAME); 236 MCLAIM(m, so->so_mowner); 237 sin6 = mtod(m, struct sockaddr_in6 *); 238 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6); 239 sin6->sin6_family = AF_INET6; 240 sin6->sin6_addr = in6addr_any; 241 sin6->sin6_port = 0; 242 error = sobind(so, m, &lwp0); 243 m_freem(m); 244 if (error) 245 goto bad; 246 } 247 #endif 248 249 /* 250 * Protocols that do not require connections may be optionally left 251 * unconnected for servers that reply from a port other than NFS_PORT. 252 */ 253 if (nmp->nm_flag & NFSMNT_NOCONN) { 254 if (nmp->nm_soflags & PR_CONNREQUIRED) { 255 error = ENOTCONN; 256 goto bad; 257 } 258 } else { 259 error = soconnect(so, nmp->nm_nam, l); 260 if (error) 261 goto bad; 262 263 /* 264 * Wait for the connection to complete. Cribbed from the 265 * connect system call but with the wait timing out so 266 * that interruptible mounts don't hang here for a long time. 267 */ 268 s = splsoftnet(); 269 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 270 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 271 "nfscn1", 2 * hz); 272 if ((so->so_state & SS_ISCONNECTING) && 273 so->so_error == 0 && rep && 274 (error = nfs_sigintr(nmp, rep, rep->r_lwp)) != 0){ 275 so->so_state &= ~SS_ISCONNECTING; 276 splx(s); 277 goto bad; 278 } 279 } 280 if (so->so_error) { 281 error = so->so_error; 282 so->so_error = 0; 283 splx(s); 284 goto bad; 285 } 286 splx(s); 287 } 288 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 289 so->so_rcv.sb_timeo = (5 * hz); 290 so->so_snd.sb_timeo = (5 * hz); 291 } else { 292 /* 293 * enable receive timeout to detect server crash and reconnect. 294 * otherwise, we can be stuck in soreceive forever. 295 */ 296 so->so_rcv.sb_timeo = (5 * hz); 297 so->so_snd.sb_timeo = 0; 298 } 299 if (nmp->nm_sotype == SOCK_DGRAM) { 300 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 301 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 302 NFS_MAXPKTHDR) * 2; 303 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 304 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 305 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 306 NFS_MAXPKTHDR) * 2; 307 } else { 308 if (nmp->nm_sotype != SOCK_STREAM) 309 panic("nfscon sotype"); 310 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 311 m = m_get(M_WAIT, MT_SOOPTS); 312 MCLAIM(m, so->so_mowner); 313 *mtod(m, int32_t *) = 1; 314 m->m_len = sizeof(int32_t); 315 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 316 } 317 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 318 m = m_get(M_WAIT, MT_SOOPTS); 319 MCLAIM(m, so->so_mowner); 320 *mtod(m, int32_t *) = 1; 321 m->m_len = sizeof(int32_t); 322 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 323 } 324 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 325 sizeof (u_int32_t)) * 2; 326 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 327 sizeof (u_int32_t)) * 2; 328 } 329 error = soreserve(so, sndreserve, rcvreserve); 330 if (error) 331 goto bad; 332 so->so_rcv.sb_flags |= SB_NOINTR; 333 so->so_snd.sb_flags |= SB_NOINTR; 334 335 /* Initialize other non-zero congestion variables */ 336 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 337 NFS_TIMEO << 3; 338 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 339 nmp->nm_sdrtt[3] = 0; 340 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 341 nmp->nm_sent = 0; 342 nmp->nm_timeouts = 0; 343 return (0); 344 345 bad: 346 nfs_disconnect(nmp); 347 return (error); 348 } 349 350 /* 351 * Reconnect routine: 352 * Called when a connection is broken on a reliable protocol. 353 * - clean up the old socket 354 * - nfs_connect() again 355 * - set R_MUSTRESEND for all outstanding requests on mount point 356 * If this fails the mount point is DEAD! 357 * nb: Must be called with the nfs_sndlock() set on the mount point. 358 */ 359 int 360 nfs_reconnect(rep, l) 361 struct nfsreq *rep; 362 struct lwp *l; 363 { 364 struct nfsreq *rp; 365 struct nfsmount *nmp = rep->r_nmp; 366 int error; 367 368 nfs_disconnect(nmp); 369 while ((error = nfs_connect(nmp, rep, l)) != 0) { 370 if (error == EINTR || error == ERESTART) 371 return (EINTR); 372 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0); 373 } 374 375 /* 376 * Loop through outstanding request list and fix up all requests 377 * on old socket. 378 */ 379 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 380 if (rp->r_nmp == nmp) { 381 if ((rp->r_flags & R_MUSTRESEND) == 0) 382 rp->r_flags |= R_MUSTRESEND | R_REXMITTED; 383 rp->r_rexmit = 0; 384 } 385 } 386 return (0); 387 } 388 389 /* 390 * NFS disconnect. Clean up and unlink. 391 */ 392 void 393 nfs_disconnect(nmp) 394 struct nfsmount *nmp; 395 { 396 struct socket *so; 397 int drain = 0; 398 399 if (nmp->nm_so) { 400 so = nmp->nm_so; 401 nmp->nm_so = (struct socket *)0; 402 soshutdown(so, SHUT_RDWR); 403 drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0; 404 if (drain) { 405 /* 406 * soshutdown() above should wake up the current 407 * listener. 408 * Now wake up those waiting for the receive lock, and 409 * wait for them to go away unhappy, to prevent *nmp 410 * from evaporating while they're sleeping. 411 */ 412 while (nmp->nm_waiters > 0) { 413 wakeup (&nmp->nm_iflag); 414 (void) tsleep(&nmp->nm_waiters, PVFS, 415 "nfsdis", 0); 416 } 417 } 418 soclose(so); 419 } 420 #ifdef DIAGNOSTIC 421 if (drain && (nmp->nm_waiters > 0)) 422 panic("nfs_disconnect: waiters left after drain?"); 423 #endif 424 } 425 426 void 427 nfs_safedisconnect(nmp) 428 struct nfsmount *nmp; 429 { 430 struct nfsreq dummyreq; 431 432 memset(&dummyreq, 0, sizeof(dummyreq)); 433 dummyreq.r_nmp = nmp; 434 nfs_rcvlock(&dummyreq); /* XXX ignored error return */ 435 nfs_disconnect(nmp); 436 nfs_rcvunlock(nmp); 437 } 438 439 /* 440 * This is the nfs send routine. For connection based socket types, it 441 * must be called with an nfs_sndlock() on the socket. 442 * "rep == NULL" indicates that it has been called from a server. 443 * For the client side: 444 * - return EINTR if the RPC is terminated, 0 otherwise 445 * - set R_MUSTRESEND if the send fails for any reason 446 * - do any cleanup required by recoverable socket errors (? ? ?) 447 * For the server side: 448 * - return EINTR or ERESTART if interrupted by a signal 449 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 450 * - do any cleanup required by recoverable socket errors (? ? ?) 451 */ 452 int 453 nfs_send(so, nam, top, rep, l) 454 struct socket *so; 455 struct mbuf *nam; 456 struct mbuf *top; 457 struct nfsreq *rep; 458 struct lwp *l; 459 { 460 struct mbuf *sendnam; 461 int error, soflags, flags; 462 463 /* XXX nfs_doio()/nfs_request() calls with rep->r_lwp == NULL */ 464 if (l == NULL && rep->r_lwp == NULL) 465 l = curlwp; 466 467 if (rep) { 468 if (rep->r_flags & R_SOFTTERM) { 469 m_freem(top); 470 return (EINTR); 471 } 472 if ((so = rep->r_nmp->nm_so) == NULL) { 473 rep->r_flags |= R_MUSTRESEND; 474 m_freem(top); 475 return (0); 476 } 477 rep->r_flags &= ~R_MUSTRESEND; 478 soflags = rep->r_nmp->nm_soflags; 479 } else 480 soflags = so->so_proto->pr_flags; 481 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 482 sendnam = (struct mbuf *)0; 483 else 484 sendnam = nam; 485 if (so->so_type == SOCK_SEQPACKET) 486 flags = MSG_EOR; 487 else 488 flags = 0; 489 490 error = (*so->so_send)(so, sendnam, (struct uio *)0, top, 491 (struct mbuf *)0, flags, l); 492 if (error) { 493 if (rep) { 494 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 495 /* 496 * We're too fast for the network/driver, 497 * and UDP isn't flowcontrolled. 498 * We need to resend. This is not fatal, 499 * just try again. 500 * 501 * Could be smarter here by doing some sort 502 * of a backoff, but this is rare. 503 */ 504 rep->r_flags |= R_MUSTRESEND; 505 } else { 506 if (error != EPIPE) 507 log(LOG_INFO, 508 "nfs send error %d for %s\n", 509 error, 510 rep->r_nmp->nm_mountp-> 511 mnt_stat.f_mntfromname); 512 /* 513 * Deal with errors for the client side. 514 */ 515 if (rep->r_flags & R_SOFTTERM) 516 error = EINTR; 517 else 518 rep->r_flags |= R_MUSTRESEND; 519 } 520 } else { 521 /* 522 * See above. This error can happen under normal 523 * circumstances and the log is too noisy. 524 * The error will still show up in nfsstat. 525 */ 526 if (error != ENOBUFS || so->so_type != SOCK_DGRAM) 527 log(LOG_INFO, "nfsd send error %d\n", error); 528 } 529 530 /* 531 * Handle any recoverable (soft) socket errors here. (? ? ?) 532 */ 533 if (error != EINTR && error != ERESTART && 534 error != EWOULDBLOCK && error != EPIPE) 535 error = 0; 536 } 537 return (error); 538 } 539 540 #ifdef NFS 541 /* 542 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 543 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 544 * Mark and consolidate the data into a new mbuf list. 545 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 546 * small mbufs. 547 * For SOCK_STREAM we must be very careful to read an entire record once 548 * we have read any of it, even if the system call has been interrupted. 549 */ 550 int 551 nfs_receive(rep, aname, mp, l) 552 struct nfsreq *rep; 553 struct mbuf **aname; 554 struct mbuf **mp; 555 struct lwp *l; 556 { 557 struct socket *so; 558 struct uio auio; 559 struct iovec aio; 560 struct mbuf *m; 561 struct mbuf *control; 562 u_int32_t len; 563 struct mbuf **getnam; 564 int error, sotype, rcvflg; 565 566 /* 567 * Set up arguments for soreceive() 568 */ 569 *mp = (struct mbuf *)0; 570 *aname = (struct mbuf *)0; 571 sotype = rep->r_nmp->nm_sotype; 572 573 /* 574 * For reliable protocols, lock against other senders/receivers 575 * in case a reconnect is necessary. 576 * For SOCK_STREAM, first get the Record Mark to find out how much 577 * more there is to get. 578 * We must lock the socket against other receivers 579 * until we have an entire rpc request/reply. 580 */ 581 if (sotype != SOCK_DGRAM) { 582 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 583 if (error) 584 return (error); 585 tryagain: 586 /* 587 * Check for fatal errors and resending request. 588 */ 589 /* 590 * Ugh: If a reconnect attempt just happened, nm_so 591 * would have changed. NULL indicates a failed 592 * attempt that has essentially shut down this 593 * mount point. 594 */ 595 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 596 nfs_sndunlock(&rep->r_nmp->nm_iflag); 597 return (EINTR); 598 } 599 so = rep->r_nmp->nm_so; 600 if (!so) { 601 error = nfs_reconnect(rep, l); 602 if (error) { 603 nfs_sndunlock(&rep->r_nmp->nm_iflag); 604 return (error); 605 } 606 goto tryagain; 607 } 608 while (rep->r_flags & R_MUSTRESEND) { 609 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 610 nfsstats.rpcretries++; 611 rep->r_rtt = 0; 612 rep->r_flags &= ~R_TIMING; 613 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, l); 614 if (error) { 615 if (error == EINTR || error == ERESTART || 616 (error = nfs_reconnect(rep, l)) != 0) { 617 nfs_sndunlock(&rep->r_nmp->nm_iflag); 618 return (error); 619 } 620 goto tryagain; 621 } 622 } 623 nfs_sndunlock(&rep->r_nmp->nm_iflag); 624 if (sotype == SOCK_STREAM) { 625 aio.iov_base = (caddr_t) &len; 626 aio.iov_len = sizeof(u_int32_t); 627 auio.uio_iov = &aio; 628 auio.uio_iovcnt = 1; 629 auio.uio_rw = UIO_READ; 630 auio.uio_offset = 0; 631 auio.uio_resid = sizeof(u_int32_t); 632 UIO_SETUP_SYSSPACE(&auio); 633 do { 634 rcvflg = MSG_WAITALL; 635 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 636 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 637 if (error == EWOULDBLOCK && rep) { 638 if (rep->r_flags & R_SOFTTERM) 639 return (EINTR); 640 /* 641 * if it seems that the server died after it 642 * received our request, set EPIPE so that 643 * we'll reconnect and retransmit requests. 644 */ 645 if (rep->r_rexmit >= rep->r_nmp->nm_retry) { 646 nfsstats.rpctimeouts++; 647 error = EPIPE; 648 } 649 } 650 } while (error == EWOULDBLOCK); 651 if (!error && auio.uio_resid > 0) { 652 /* 653 * Don't log a 0 byte receive; it means 654 * that the socket has been closed, and 655 * can happen during normal operation 656 * (forcible unmount or Solaris server). 657 */ 658 if (auio.uio_resid != sizeof (u_int32_t)) 659 log(LOG_INFO, 660 "short receive (%lu/%lu) from nfs server %s\n", 661 (u_long)sizeof(u_int32_t) - auio.uio_resid, 662 (u_long)sizeof(u_int32_t), 663 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 664 error = EPIPE; 665 } 666 if (error) 667 goto errout; 668 len = ntohl(len) & ~0x80000000; 669 /* 670 * This is SERIOUS! We are out of sync with the sender 671 * and forcing a disconnect/reconnect is all I can do. 672 */ 673 if (len > NFS_MAXPACKET) { 674 log(LOG_ERR, "%s (%d) from nfs server %s\n", 675 "impossible packet length", 676 len, 677 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 678 error = EFBIG; 679 goto errout; 680 } 681 auio.uio_resid = len; 682 do { 683 rcvflg = MSG_WAITALL; 684 error = (*so->so_receive)(so, (struct mbuf **)0, 685 &auio, mp, (struct mbuf **)0, &rcvflg); 686 } while (error == EWOULDBLOCK || error == EINTR || 687 error == ERESTART); 688 if (!error && auio.uio_resid > 0) { 689 if (len != auio.uio_resid) 690 log(LOG_INFO, 691 "short receive (%lu/%d) from nfs server %s\n", 692 (u_long)len - auio.uio_resid, len, 693 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 694 error = EPIPE; 695 } 696 } else { 697 /* 698 * NB: Since uio_resid is big, MSG_WAITALL is ignored 699 * and soreceive() will return when it has either a 700 * control msg or a data msg. 701 * We have no use for control msg., but must grab them 702 * and then throw them away so we know what is going 703 * on. 704 */ 705 auio.uio_resid = len = 100000000; /* Anything Big */ 706 /* not need to setup uio_vmspace */ 707 do { 708 rcvflg = 0; 709 error = (*so->so_receive)(so, (struct mbuf **)0, 710 &auio, mp, &control, &rcvflg); 711 if (control) 712 m_freem(control); 713 if (error == EWOULDBLOCK && rep) { 714 if (rep->r_flags & R_SOFTTERM) 715 return (EINTR); 716 } 717 } while (error == EWOULDBLOCK || 718 (!error && *mp == NULL && control)); 719 if ((rcvflg & MSG_EOR) == 0) 720 printf("Egad!!\n"); 721 if (!error && *mp == NULL) 722 error = EPIPE; 723 len -= auio.uio_resid; 724 } 725 errout: 726 if (error && error != EINTR && error != ERESTART) { 727 m_freem(*mp); 728 *mp = (struct mbuf *)0; 729 if (error != EPIPE) 730 log(LOG_INFO, 731 "receive error %d from nfs server %s\n", 732 error, 733 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 734 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 735 if (!error) 736 error = nfs_reconnect(rep, l); 737 if (!error) 738 goto tryagain; 739 else 740 nfs_sndunlock(&rep->r_nmp->nm_iflag); 741 } 742 } else { 743 if ((so = rep->r_nmp->nm_so) == NULL) 744 return (EACCES); 745 if (so->so_state & SS_ISCONNECTED) 746 getnam = (struct mbuf **)0; 747 else 748 getnam = aname; 749 auio.uio_resid = len = 1000000; 750 /* not need to setup uio_vmspace */ 751 do { 752 rcvflg = 0; 753 error = (*so->so_receive)(so, getnam, &auio, mp, 754 (struct mbuf **)0, &rcvflg); 755 if (error == EWOULDBLOCK && 756 (rep->r_flags & R_SOFTTERM)) 757 return (EINTR); 758 } while (error == EWOULDBLOCK); 759 len -= auio.uio_resid; 760 if (!error && *mp == NULL) 761 error = EPIPE; 762 } 763 if (error) { 764 m_freem(*mp); 765 *mp = (struct mbuf *)0; 766 } 767 return (error); 768 } 769 770 /* 771 * Implement receipt of reply on a socket. 772 * We must search through the list of received datagrams matching them 773 * with outstanding requests using the xid, until ours is found. 774 */ 775 /* ARGSUSED */ 776 int 777 nfs_reply(myrep, lwp) 778 struct nfsreq *myrep; 779 struct lwp *lwp; 780 { 781 struct nfsreq *rep; 782 struct nfsmount *nmp = myrep->r_nmp; 783 int32_t t1; 784 struct mbuf *mrep, *nam, *md; 785 u_int32_t rxid, *tl; 786 caddr_t dpos, cp2; 787 int error; 788 789 /* 790 * Loop around until we get our own reply 791 */ 792 for (;;) { 793 /* 794 * Lock against other receivers so that I don't get stuck in 795 * sbwait() after someone else has received my reply for me. 796 * Also necessary for connection based protocols to avoid 797 * race conditions during a reconnect. 798 */ 799 error = nfs_rcvlock(myrep); 800 if (error == EALREADY) 801 return (0); 802 if (error) 803 return (error); 804 /* 805 * Get the next Rpc reply off the socket 806 */ 807 nmp->nm_waiters++; 808 error = nfs_receive(myrep, &nam, &mrep, lwp); 809 nfs_rcvunlock(nmp); 810 if (error) { 811 812 if (nmp->nm_iflag & NFSMNT_DISMNT) { 813 /* 814 * Oops, we're going away now.. 815 */ 816 nmp->nm_waiters--; 817 wakeup (&nmp->nm_waiters); 818 return error; 819 } 820 nmp->nm_waiters--; 821 /* 822 * Ignore routing errors on connectionless protocols? ? 823 */ 824 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 825 nmp->nm_so->so_error = 0; 826 #ifdef DEBUG 827 printf("nfs_reply: ignoring error %d\n", error); 828 #endif 829 if (myrep->r_flags & R_GETONEREP) 830 return (0); 831 continue; 832 } 833 return (error); 834 } 835 nmp->nm_waiters--; 836 if (nam) 837 m_freem(nam); 838 839 /* 840 * Get the xid and check that it is an rpc reply 841 */ 842 md = mrep; 843 dpos = mtod(md, caddr_t); 844 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 845 rxid = *tl++; 846 if (*tl != rpc_reply) { 847 nfsstats.rpcinvalid++; 848 m_freem(mrep); 849 nfsmout: 850 if (myrep->r_flags & R_GETONEREP) 851 return (0); 852 continue; 853 } 854 855 /* 856 * Loop through the request list to match up the reply 857 * Iff no match, just drop the datagram 858 */ 859 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 860 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 861 /* Found it.. */ 862 rep->r_mrep = mrep; 863 rep->r_md = md; 864 rep->r_dpos = dpos; 865 if (nfsrtton) { 866 struct rttl *rt; 867 868 rt = &nfsrtt.rttl[nfsrtt.pos]; 869 rt->proc = rep->r_procnum; 870 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 871 rt->sent = nmp->nm_sent; 872 rt->cwnd = nmp->nm_cwnd; 873 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 874 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 875 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx; 876 getmicrotime(&rt->tstamp); 877 if (rep->r_flags & R_TIMING) 878 rt->rtt = rep->r_rtt; 879 else 880 rt->rtt = 1000000; 881 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 882 } 883 /* 884 * Update congestion window. 885 * Do the additive increase of 886 * one rpc/rtt. 887 */ 888 if (nmp->nm_cwnd <= nmp->nm_sent) { 889 nmp->nm_cwnd += 890 (NFS_CWNDSCALE * NFS_CWNDSCALE + 891 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 892 if (nmp->nm_cwnd > NFS_MAXCWND) 893 nmp->nm_cwnd = NFS_MAXCWND; 894 } 895 rep->r_flags &= ~R_SENT; 896 nmp->nm_sent -= NFS_CWNDSCALE; 897 /* 898 * Update rtt using a gain of 0.125 on the mean 899 * and a gain of 0.25 on the deviation. 900 */ 901 if (rep->r_flags & R_TIMING) { 902 /* 903 * Since the timer resolution of 904 * NFS_HZ is so course, it can often 905 * result in r_rtt == 0. Since 906 * r_rtt == N means that the actual 907 * rtt is between N+dt and N+2-dt ticks, 908 * add 1. 909 */ 910 t1 = rep->r_rtt + 1; 911 t1 -= (NFS_SRTT(rep) >> 3); 912 NFS_SRTT(rep) += t1; 913 if (t1 < 0) 914 t1 = -t1; 915 t1 -= (NFS_SDRTT(rep) >> 2); 916 NFS_SDRTT(rep) += t1; 917 } 918 nmp->nm_timeouts = 0; 919 break; 920 } 921 } 922 /* 923 * If not matched to a request, drop it. 924 * If it's mine, get out. 925 */ 926 if (rep == 0) { 927 nfsstats.rpcunexpected++; 928 m_freem(mrep); 929 } else if (rep == myrep) { 930 if (rep->r_mrep == NULL) 931 panic("nfsreply nil"); 932 return (0); 933 } 934 if (myrep->r_flags & R_GETONEREP) 935 return (0); 936 } 937 } 938 939 /* 940 * nfs_request - goes something like this 941 * - fill in request struct 942 * - links it into list 943 * - calls nfs_send() for first transmit 944 * - calls nfs_receive() to get reply 945 * - break down rpc header and return with nfs reply pointed to 946 * by mrep or error 947 * nb: always frees up mreq mbuf list 948 */ 949 int 950 nfs_request(np, mrest, procnum, lwp, cred, mrp, mdp, dposp, rexmitp) 951 struct nfsnode *np; 952 struct mbuf *mrest; 953 int procnum; 954 struct lwp *lwp; 955 kauth_cred_t cred; 956 struct mbuf **mrp; 957 struct mbuf **mdp; 958 caddr_t *dposp; 959 int *rexmitp; 960 { 961 struct mbuf *m, *mrep; 962 struct nfsreq *rep; 963 u_int32_t *tl; 964 int i; 965 struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount); 966 struct mbuf *md, *mheadend; 967 char nickv[RPCX_NICKVERF]; 968 time_t waituntil; 969 caddr_t dpos, cp2; 970 int t1, s, error = 0, mrest_len, auth_len, auth_type; 971 int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0; 972 int verf_len, verf_type; 973 u_int32_t xid; 974 char *auth_str, *verf_str; 975 NFSKERBKEY_T key; /* save session key */ 976 kauth_cred_t acred; 977 struct mbuf *mrest_backup = NULL; 978 kauth_cred_t origcred = NULL; /* XXX: gcc */ 979 boolean_t retry_cred = TRUE; 980 boolean_t use_opencred = (np->n_flag & NUSEOPENCRED) != 0; 981 982 if (rexmitp != NULL) 983 *rexmitp = 0; 984 985 acred = kauth_cred_alloc(); 986 987 tryagain_cred: 988 KASSERT(cred != NULL); 989 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 990 rep->r_nmp = nmp; 991 rep->r_lwp = lwp; 992 rep->r_procnum = procnum; 993 i = 0; 994 m = mrest; 995 while (m) { 996 i += m->m_len; 997 m = m->m_next; 998 } 999 mrest_len = i; 1000 1001 /* 1002 * Get the RPC header with authorization. 1003 */ 1004 kerbauth: 1005 verf_str = auth_str = (char *)0; 1006 if (nmp->nm_flag & NFSMNT_KERB) { 1007 verf_str = nickv; 1008 verf_len = sizeof (nickv); 1009 auth_type = RPCAUTH_KERB4; 1010 memset((caddr_t)key, 0, sizeof (key)); 1011 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 1012 &auth_len, verf_str, verf_len)) { 1013 error = nfs_getauth(nmp, rep, cred, &auth_str, 1014 &auth_len, verf_str, &verf_len, key); 1015 if (error) { 1016 free((caddr_t)rep, M_NFSREQ); 1017 m_freem(mrest); 1018 KASSERT(kauth_cred_getrefcnt(acred) == 1); 1019 kauth_cred_free(acred); 1020 return (error); 1021 } 1022 } 1023 retry_cred = FALSE; 1024 } else { 1025 /* AUTH_UNIX */ 1026 uid_t uid; 1027 gid_t gid; 1028 1029 /* 1030 * on the most unix filesystems, permission checks are 1031 * done when the file is open(2)'ed. 1032 * ie. once a file is successfully open'ed, 1033 * following i/o operations never fail with EACCES. 1034 * we try to follow the semantics as far as possible. 1035 * 1036 * note that we expect that the nfs server always grant 1037 * accesses by the file's owner. 1038 */ 1039 origcred = cred; 1040 switch (procnum) { 1041 case NFSPROC_READ: 1042 case NFSPROC_WRITE: 1043 case NFSPROC_COMMIT: 1044 uid = np->n_vattr->va_uid; 1045 gid = np->n_vattr->va_gid; 1046 if (kauth_cred_geteuid(cred) == uid && 1047 kauth_cred_getegid(cred) == gid) { 1048 retry_cred = FALSE; 1049 break; 1050 } 1051 if (use_opencred) 1052 break; 1053 kauth_cred_setuid(acred, uid); 1054 kauth_cred_seteuid(acred, uid); 1055 kauth_cred_setsvuid(acred, uid); 1056 kauth_cred_setgid(acred, gid); 1057 kauth_cred_setegid(acred, gid); 1058 kauth_cred_setsvgid(acred, gid); 1059 cred = acred; 1060 break; 1061 default: 1062 retry_cred = FALSE; 1063 break; 1064 } 1065 /* 1066 * backup mbuf chain if we can need it later to retry. 1067 * 1068 * XXX maybe we can keep a direct reference to 1069 * mrest without doing m_copym, but it's ...ugly. 1070 */ 1071 if (retry_cred) 1072 mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT); 1073 auth_type = RPCAUTH_UNIX; 1074 /* XXX elad - ngroups */ 1075 auth_len = (((kauth_cred_ngroups(cred) > nmp->nm_numgrps) ? 1076 nmp->nm_numgrps : kauth_cred_ngroups(cred)) << 2) + 1077 5 * NFSX_UNSIGNED; 1078 } 1079 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 1080 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 1081 if (auth_str) 1082 free(auth_str, M_TEMP); 1083 1084 /* 1085 * For stream protocols, insert a Sun RPC Record Mark. 1086 */ 1087 if (nmp->nm_sotype == SOCK_STREAM) { 1088 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 1089 *mtod(m, u_int32_t *) = htonl(0x80000000 | 1090 (m->m_pkthdr.len - NFSX_UNSIGNED)); 1091 } 1092 rep->r_mreq = m; 1093 rep->r_xid = xid; 1094 tryagain: 1095 if (nmp->nm_flag & NFSMNT_SOFT) 1096 rep->r_retry = nmp->nm_retry; 1097 else 1098 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1099 rep->r_rtt = rep->r_rexmit = 0; 1100 if (proct[procnum] > 0) 1101 rep->r_flags = R_TIMING; 1102 else 1103 rep->r_flags = 0; 1104 rep->r_mrep = NULL; 1105 1106 /* 1107 * Do the client side RPC. 1108 */ 1109 nfsstats.rpcrequests++; 1110 /* 1111 * Chain request into list of outstanding requests. Be sure 1112 * to put it LAST so timer finds oldest requests first. 1113 */ 1114 s = splsoftnet(); 1115 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1116 1117 /* 1118 * If backing off another request or avoiding congestion, don't 1119 * send this one now but let timer do it. If not timing a request, 1120 * do it now. 1121 */ 1122 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1123 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1124 nmp->nm_sent < nmp->nm_cwnd)) { 1125 splx(s); 1126 if (nmp->nm_soflags & PR_CONNREQUIRED) 1127 error = nfs_sndlock(&nmp->nm_iflag, rep); 1128 if (!error) { 1129 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1130 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp); 1131 if (nmp->nm_soflags & PR_CONNREQUIRED) 1132 nfs_sndunlock(&nmp->nm_iflag); 1133 } 1134 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1135 nmp->nm_sent += NFS_CWNDSCALE; 1136 rep->r_flags |= R_SENT; 1137 } 1138 } else { 1139 splx(s); 1140 rep->r_rtt = -1; 1141 } 1142 1143 /* 1144 * Wait for the reply from our send or the timer's. 1145 */ 1146 if (!error || error == EPIPE) 1147 error = nfs_reply(rep, lwp); 1148 1149 /* 1150 * RPC done, unlink the request. 1151 */ 1152 s = splsoftnet(); 1153 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1154 splx(s); 1155 1156 /* 1157 * Decrement the outstanding request count. 1158 */ 1159 if (rep->r_flags & R_SENT) { 1160 rep->r_flags &= ~R_SENT; /* paranoia */ 1161 nmp->nm_sent -= NFS_CWNDSCALE; 1162 } 1163 1164 if (rexmitp != NULL) { 1165 int rexmit; 1166 1167 if (nmp->nm_sotype != SOCK_DGRAM) 1168 rexmit = (rep->r_flags & R_REXMITTED) != 0; 1169 else 1170 rexmit = rep->r_rexmit; 1171 *rexmitp = rexmit; 1172 } 1173 1174 /* 1175 * If there was a successful reply and a tprintf msg. 1176 * tprintf a response. 1177 */ 1178 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1179 nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1180 "is alive again"); 1181 mrep = rep->r_mrep; 1182 md = rep->r_md; 1183 dpos = rep->r_dpos; 1184 if (error) 1185 goto nfsmout; 1186 1187 /* 1188 * break down the rpc header and check if ok 1189 */ 1190 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1191 if (*tl++ == rpc_msgdenied) { 1192 if (*tl == rpc_mismatch) 1193 error = EOPNOTSUPP; 1194 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1195 if (!failed_auth) { 1196 failed_auth++; 1197 mheadend->m_next = (struct mbuf *)0; 1198 m_freem(mrep); 1199 m_freem(rep->r_mreq); 1200 goto kerbauth; 1201 } else 1202 error = EAUTH; 1203 } else 1204 error = EACCES; 1205 m_freem(mrep); 1206 goto nfsmout; 1207 } 1208 1209 /* 1210 * Grab any Kerberos verifier, otherwise just throw it away. 1211 */ 1212 verf_type = fxdr_unsigned(int, *tl++); 1213 i = fxdr_unsigned(int32_t, *tl); 1214 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1215 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1216 if (error) 1217 goto nfsmout; 1218 } else if (i > 0) 1219 nfsm_adv(nfsm_rndup(i)); 1220 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1221 /* 0 == ok */ 1222 if (*tl == 0) { 1223 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1224 if (*tl != 0) { 1225 error = fxdr_unsigned(int, *tl); 1226 switch (error) { 1227 case NFSERR_PERM: 1228 error = EPERM; 1229 break; 1230 1231 case NFSERR_NOENT: 1232 error = ENOENT; 1233 break; 1234 1235 case NFSERR_IO: 1236 error = EIO; 1237 break; 1238 1239 case NFSERR_NXIO: 1240 error = ENXIO; 1241 break; 1242 1243 case NFSERR_ACCES: 1244 error = EACCES; 1245 if (!retry_cred) 1246 break; 1247 m_freem(mrep); 1248 m_freem(rep->r_mreq); 1249 FREE(rep, M_NFSREQ); 1250 use_opencred = !use_opencred; 1251 if (mrest_backup == NULL) { 1252 /* m_copym failure */ 1253 KASSERT( 1254 kauth_cred_getrefcnt(acred) == 1); 1255 kauth_cred_free(acred); 1256 return ENOMEM; 1257 } 1258 mrest = mrest_backup; 1259 mrest_backup = NULL; 1260 cred = origcred; 1261 error = 0; 1262 retry_cred = FALSE; 1263 goto tryagain_cred; 1264 1265 case NFSERR_EXIST: 1266 error = EEXIST; 1267 break; 1268 1269 case NFSERR_XDEV: 1270 error = EXDEV; 1271 break; 1272 1273 case NFSERR_NODEV: 1274 error = ENODEV; 1275 break; 1276 1277 case NFSERR_NOTDIR: 1278 error = ENOTDIR; 1279 break; 1280 1281 case NFSERR_ISDIR: 1282 error = EISDIR; 1283 break; 1284 1285 case NFSERR_INVAL: 1286 error = EINVAL; 1287 break; 1288 1289 case NFSERR_FBIG: 1290 error = EFBIG; 1291 break; 1292 1293 case NFSERR_NOSPC: 1294 error = ENOSPC; 1295 break; 1296 1297 case NFSERR_ROFS: 1298 error = EROFS; 1299 break; 1300 1301 case NFSERR_MLINK: 1302 error = EMLINK; 1303 break; 1304 1305 case NFSERR_TIMEDOUT: 1306 error = ETIMEDOUT; 1307 break; 1308 1309 case NFSERR_NAMETOL: 1310 error = ENAMETOOLONG; 1311 break; 1312 1313 case NFSERR_NOTEMPTY: 1314 error = ENOTEMPTY; 1315 break; 1316 1317 case NFSERR_DQUOT: 1318 error = EDQUOT; 1319 break; 1320 1321 case NFSERR_STALE: 1322 /* 1323 * If the File Handle was stale, invalidate the 1324 * lookup cache, just in case. 1325 */ 1326 error = ESTALE; 1327 cache_purge(NFSTOV(np)); 1328 break; 1329 1330 case NFSERR_REMOTE: 1331 error = EREMOTE; 1332 break; 1333 1334 case NFSERR_WFLUSH: 1335 case NFSERR_BADHANDLE: 1336 case NFSERR_NOT_SYNC: 1337 case NFSERR_BAD_COOKIE: 1338 error = EINVAL; 1339 break; 1340 1341 case NFSERR_NOTSUPP: 1342 error = ENOTSUP; 1343 break; 1344 1345 case NFSERR_TOOSMALL: 1346 case NFSERR_SERVERFAULT: 1347 case NFSERR_BADTYPE: 1348 error = EINVAL; 1349 break; 1350 1351 case NFSERR_TRYLATER: 1352 if ((nmp->nm_flag & NFSMNT_NFSV3) == 0) 1353 break; 1354 m_freem(mrep); 1355 error = 0; 1356 waituntil = time_second + trylater_delay; 1357 while (time_second < waituntil) { 1358 tsleep(&lbolt, PSOCK, "nfstrylater", 0); 1359 } 1360 trylater_delay *= NFS_TRYLATERDELMUL; 1361 if (trylater_delay > NFS_TRYLATERDELMAX) 1362 trylater_delay = NFS_TRYLATERDELMAX; 1363 /* 1364 * RFC1813: 1365 * The client should wait and then try 1366 * the request with a new RPC transaction ID. 1367 */ 1368 nfs_renewxid(rep); 1369 goto tryagain; 1370 1371 default: 1372 #ifdef DIAGNOSTIC 1373 printf("Invalid rpc error code %d\n", error); 1374 #endif 1375 error = EINVAL; 1376 break; 1377 } 1378 1379 if (nmp->nm_flag & NFSMNT_NFSV3) { 1380 *mrp = mrep; 1381 *mdp = md; 1382 *dposp = dpos; 1383 error |= NFSERR_RETERR; 1384 } else 1385 m_freem(mrep); 1386 goto nfsmout; 1387 } 1388 1389 /* 1390 * note which credential worked to minimize number of retries. 1391 */ 1392 if (use_opencred) 1393 np->n_flag |= NUSEOPENCRED; 1394 else 1395 np->n_flag &= ~NUSEOPENCRED; 1396 1397 *mrp = mrep; 1398 *mdp = md; 1399 *dposp = dpos; 1400 1401 KASSERT(error == 0); 1402 goto nfsmout; 1403 } 1404 m_freem(mrep); 1405 error = EPROTONOSUPPORT; 1406 nfsmout: 1407 KASSERT(kauth_cred_getrefcnt(acred) == 1); 1408 kauth_cred_free(acred); 1409 m_freem(rep->r_mreq); 1410 free((caddr_t)rep, M_NFSREQ); 1411 m_freem(mrest_backup); 1412 return (error); 1413 } 1414 #endif /* NFS */ 1415 1416 /* 1417 * Generate the rpc reply header 1418 * siz arg. is used to decide if adding a cluster is worthwhile 1419 */ 1420 int 1421 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1422 int siz; 1423 struct nfsrv_descript *nd; 1424 struct nfssvc_sock *slp; 1425 int err; 1426 int cache; 1427 u_quad_t *frev; 1428 struct mbuf **mrq; 1429 struct mbuf **mbp; 1430 caddr_t *bposp; 1431 { 1432 u_int32_t *tl; 1433 struct mbuf *mreq; 1434 caddr_t bpos; 1435 struct mbuf *mb; 1436 1437 mreq = m_gethdr(M_WAIT, MT_DATA); 1438 MCLAIM(mreq, &nfs_mowner); 1439 mb = mreq; 1440 /* 1441 * If this is a big reply, use a cluster else 1442 * try and leave leading space for the lower level headers. 1443 */ 1444 siz += RPC_REPLYSIZ; 1445 if (siz >= max_datalen) { 1446 m_clget(mreq, M_WAIT); 1447 } else 1448 mreq->m_data += max_hdr; 1449 tl = mtod(mreq, u_int32_t *); 1450 mreq->m_len = 6 * NFSX_UNSIGNED; 1451 bpos = ((caddr_t)tl) + mreq->m_len; 1452 *tl++ = txdr_unsigned(nd->nd_retxid); 1453 *tl++ = rpc_reply; 1454 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1455 *tl++ = rpc_msgdenied; 1456 if (err & NFSERR_AUTHERR) { 1457 *tl++ = rpc_autherr; 1458 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1459 mreq->m_len -= NFSX_UNSIGNED; 1460 bpos -= NFSX_UNSIGNED; 1461 } else { 1462 *tl++ = rpc_mismatch; 1463 *tl++ = txdr_unsigned(RPC_VER2); 1464 *tl = txdr_unsigned(RPC_VER2); 1465 } 1466 } else { 1467 *tl++ = rpc_msgaccepted; 1468 1469 /* 1470 * For Kerberos authentication, we must send the nickname 1471 * verifier back, otherwise just RPCAUTH_NULL. 1472 */ 1473 if (nd->nd_flag & ND_KERBFULL) { 1474 struct nfsuid *nuidp; 1475 struct timeval ktvin, ktvout; 1476 1477 memset(&ktvout, 0, sizeof ktvout); /* XXX gcc */ 1478 1479 LIST_FOREACH(nuidp, 1480 NUIDHASH(slp, kauth_cred_geteuid(nd->nd_cr)), 1481 nu_hash) { 1482 if (kauth_cred_geteuid(nuidp->nu_cr) == 1483 kauth_cred_geteuid(nd->nd_cr) && 1484 (!nd->nd_nam2 || netaddr_match( 1485 NU_NETFAM(nuidp), &nuidp->nu_haddr, 1486 nd->nd_nam2))) 1487 break; 1488 } 1489 if (nuidp) { 1490 ktvin.tv_sec = 1491 txdr_unsigned(nuidp->nu_timestamp.tv_sec 1492 - 1); 1493 ktvin.tv_usec = 1494 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1495 1496 /* 1497 * Encrypt the timestamp in ecb mode using the 1498 * session key. 1499 */ 1500 #ifdef NFSKERB 1501 XXX 1502 #endif 1503 1504 *tl++ = rpc_auth_kerb; 1505 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1506 *tl = ktvout.tv_sec; 1507 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1508 *tl++ = ktvout.tv_usec; 1509 *tl++ = txdr_unsigned( 1510 kauth_cred_geteuid(nuidp->nu_cr)); 1511 } else { 1512 *tl++ = 0; 1513 *tl++ = 0; 1514 } 1515 } else { 1516 *tl++ = 0; 1517 *tl++ = 0; 1518 } 1519 switch (err) { 1520 case EPROGUNAVAIL: 1521 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1522 break; 1523 case EPROGMISMATCH: 1524 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1525 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1526 *tl++ = txdr_unsigned(2); 1527 *tl = txdr_unsigned(3); 1528 break; 1529 case EPROCUNAVAIL: 1530 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1531 break; 1532 case EBADRPC: 1533 *tl = txdr_unsigned(RPC_GARBAGE); 1534 break; 1535 default: 1536 *tl = 0; 1537 if (err != NFSERR_RETVOID) { 1538 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1539 if (err) 1540 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1541 else 1542 *tl = 0; 1543 } 1544 break; 1545 }; 1546 } 1547 1548 if (mrq != NULL) 1549 *mrq = mreq; 1550 *mbp = mb; 1551 *bposp = bpos; 1552 if (err != 0 && err != NFSERR_RETVOID) 1553 nfsstats.srvrpc_errs++; 1554 return (0); 1555 } 1556 1557 /* 1558 * Nfs timer routine 1559 * Scan the nfsreq list and retranmit any requests that have timed out 1560 * To avoid retransmission attempts on STREAM sockets (in the future) make 1561 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1562 */ 1563 void 1564 nfs_timer(void *arg) 1565 { 1566 struct nfsreq *rep; 1567 struct mbuf *m; 1568 struct socket *so; 1569 struct nfsmount *nmp; 1570 int timeo; 1571 int s, error; 1572 #ifdef NFSSERVER 1573 struct timeval tv; 1574 struct nfssvc_sock *slp; 1575 u_quad_t cur_usec; 1576 #endif 1577 1578 s = splsoftnet(); 1579 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1580 nmp = rep->r_nmp; 1581 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1582 continue; 1583 if (nfs_sigintr(nmp, rep, rep->r_lwp)) { 1584 rep->r_flags |= R_SOFTTERM; 1585 continue; 1586 } 1587 if (rep->r_rtt >= 0) { 1588 rep->r_rtt++; 1589 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1590 timeo = nmp->nm_timeo; 1591 else 1592 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1593 if (nmp->nm_timeouts > 0) 1594 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1595 if (rep->r_rtt <= timeo) 1596 continue; 1597 if (nmp->nm_timeouts < 1598 (sizeof(nfs_backoff) / sizeof(nfs_backoff[0]))) 1599 nmp->nm_timeouts++; 1600 } 1601 /* 1602 * Check for server not responding 1603 */ 1604 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1605 rep->r_rexmit > nmp->nm_deadthresh) { 1606 nfs_msg(rep->r_lwp, 1607 nmp->nm_mountp->mnt_stat.f_mntfromname, 1608 "not responding"); 1609 rep->r_flags |= R_TPRINTFMSG; 1610 } 1611 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1612 nfsstats.rpctimeouts++; 1613 rep->r_flags |= R_SOFTTERM; 1614 continue; 1615 } 1616 if (nmp->nm_sotype != SOCK_DGRAM) { 1617 if (++rep->r_rexmit > NFS_MAXREXMIT) 1618 rep->r_rexmit = NFS_MAXREXMIT; 1619 continue; 1620 } 1621 if ((so = nmp->nm_so) == NULL) 1622 continue; 1623 1624 /* 1625 * If there is enough space and the window allows.. 1626 * Resend it 1627 * Set r_rtt to -1 in case we fail to send it now. 1628 */ 1629 rep->r_rtt = -1; 1630 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1631 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1632 (rep->r_flags & R_SENT) || 1633 nmp->nm_sent < nmp->nm_cwnd) && 1634 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1635 if (so->so_state & SS_ISCONNECTED) 1636 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1637 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1638 else 1639 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1640 nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0); 1641 if (error) { 1642 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1643 #ifdef DEBUG 1644 printf("nfs_timer: ignoring error %d\n", 1645 error); 1646 #endif 1647 so->so_error = 0; 1648 } 1649 } else { 1650 /* 1651 * Iff first send, start timing 1652 * else turn timing off, backoff timer 1653 * and divide congestion window by 2. 1654 */ 1655 if (rep->r_flags & R_SENT) { 1656 rep->r_flags &= ~R_TIMING; 1657 if (++rep->r_rexmit > NFS_MAXREXMIT) 1658 rep->r_rexmit = NFS_MAXREXMIT; 1659 nmp->nm_cwnd >>= 1; 1660 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1661 nmp->nm_cwnd = NFS_CWNDSCALE; 1662 nfsstats.rpcretries++; 1663 } else { 1664 rep->r_flags |= R_SENT; 1665 nmp->nm_sent += NFS_CWNDSCALE; 1666 } 1667 rep->r_rtt = 0; 1668 } 1669 } 1670 } 1671 1672 #ifdef NFSSERVER 1673 /* 1674 * Scan the write gathering queues for writes that need to be 1675 * completed now. 1676 */ 1677 getmicrotime(&tv); 1678 cur_usec = (u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec; 1679 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1680 if (LIST_FIRST(&slp->ns_tq) && 1681 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) 1682 nfsrv_wakenfsd(slp); 1683 } 1684 #endif /* NFSSERVER */ 1685 splx(s); 1686 callout_schedule(&nfs_timer_ch, nfs_ticks); 1687 } 1688 1689 /*ARGSUSED*/ 1690 void 1691 nfs_exit(struct proc *p, void *v) 1692 { 1693 struct nfsreq *rp; 1694 int s = splsoftnet(); 1695 1696 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 1697 if (rp->r_lwp && rp->r_lwp->l_proc == p) 1698 TAILQ_REMOVE(&nfs_reqq, rp, r_chain); 1699 } 1700 splx(s); 1701 } 1702 1703 /* 1704 * Test for a termination condition pending on the process. 1705 * This is used for NFSMNT_INT mounts. 1706 */ 1707 int 1708 nfs_sigintr(nmp, rep, l) 1709 struct nfsmount *nmp; 1710 struct nfsreq *rep; 1711 struct lwp *l; 1712 { 1713 sigset_t ss; 1714 1715 if (rep && (rep->r_flags & R_SOFTTERM)) 1716 return (EINTR); 1717 if (!(nmp->nm_flag & NFSMNT_INT)) 1718 return (0); 1719 if (l) { 1720 sigpending1(l, &ss); 1721 #if 0 1722 sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss); 1723 #endif 1724 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1725 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1726 sigismember(&ss, SIGQUIT)) 1727 return (EINTR); 1728 } 1729 return (0); 1730 } 1731 1732 /* 1733 * Lock a socket against others. 1734 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1735 * and also to avoid race conditions between the processes with nfs requests 1736 * in progress when a reconnect is necessary. 1737 */ 1738 int 1739 nfs_sndlock(flagp, rep) 1740 int *flagp; 1741 struct nfsreq *rep; 1742 { 1743 struct lwp *l; 1744 int slpflag = 0, slptimeo = 0; 1745 1746 if (rep) { 1747 l = rep->r_lwp; 1748 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1749 slpflag = PCATCH; 1750 } else 1751 l = (struct lwp *)0; 1752 while (*flagp & NFSMNT_SNDLOCK) { 1753 if (rep && nfs_sigintr(rep->r_nmp, rep, l)) 1754 return (EINTR); 1755 *flagp |= NFSMNT_WANTSND; 1756 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1757 slptimeo); 1758 if (slpflag == PCATCH) { 1759 slpflag = 0; 1760 slptimeo = 2 * hz; 1761 } 1762 } 1763 *flagp |= NFSMNT_SNDLOCK; 1764 return (0); 1765 } 1766 1767 /* 1768 * Unlock the stream socket for others. 1769 */ 1770 void 1771 nfs_sndunlock(flagp) 1772 int *flagp; 1773 { 1774 1775 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1776 panic("nfs sndunlock"); 1777 *flagp &= ~NFSMNT_SNDLOCK; 1778 if (*flagp & NFSMNT_WANTSND) { 1779 *flagp &= ~NFSMNT_WANTSND; 1780 wakeup((caddr_t)flagp); 1781 } 1782 } 1783 1784 int 1785 nfs_rcvlock(rep) 1786 struct nfsreq *rep; 1787 { 1788 struct nfsmount *nmp = rep->r_nmp; 1789 int *flagp = &nmp->nm_iflag; 1790 int slpflag, slptimeo = 0; 1791 int error = 0; 1792 1793 if (*flagp & NFSMNT_DISMNT) 1794 return EIO; 1795 1796 if (*flagp & NFSMNT_INT) 1797 slpflag = PCATCH; 1798 else 1799 slpflag = 0; 1800 simple_lock(&nmp->nm_slock); 1801 while (*flagp & NFSMNT_RCVLOCK) { 1802 if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) { 1803 error = EINTR; 1804 goto quit; 1805 } 1806 *flagp |= NFSMNT_WANTRCV; 1807 nmp->nm_waiters++; 1808 (void) ltsleep(flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1809 slptimeo, &nmp->nm_slock); 1810 nmp->nm_waiters--; 1811 if (*flagp & NFSMNT_DISMNT) { 1812 wakeup(&nmp->nm_waiters); 1813 error = EIO; 1814 goto quit; 1815 } 1816 /* If our reply was received while we were sleeping, 1817 * then just return without taking the lock to avoid a 1818 * situation where a single iod could 'capture' the 1819 * receive lock. 1820 */ 1821 if (rep->r_mrep != NULL) { 1822 error = EALREADY; 1823 goto quit; 1824 } 1825 if (slpflag == PCATCH) { 1826 slpflag = 0; 1827 slptimeo = 2 * hz; 1828 } 1829 } 1830 *flagp |= NFSMNT_RCVLOCK; 1831 quit: 1832 simple_unlock(&nmp->nm_slock); 1833 return error; 1834 } 1835 1836 /* 1837 * Unlock the stream socket for others. 1838 */ 1839 void 1840 nfs_rcvunlock(nmp) 1841 struct nfsmount *nmp; 1842 { 1843 int *flagp = &nmp->nm_iflag; 1844 1845 simple_lock(&nmp->nm_slock); 1846 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1847 panic("nfs rcvunlock"); 1848 *flagp &= ~NFSMNT_RCVLOCK; 1849 if (*flagp & NFSMNT_WANTRCV) { 1850 *flagp &= ~NFSMNT_WANTRCV; 1851 wakeup((caddr_t)flagp); 1852 } 1853 simple_unlock(&nmp->nm_slock); 1854 } 1855 1856 /* 1857 * Parse an RPC request 1858 * - verify it 1859 * - allocate and fill in the cred. 1860 */ 1861 int 1862 nfs_getreq(nd, nfsd, has_header) 1863 struct nfsrv_descript *nd; 1864 struct nfsd *nfsd; 1865 int has_header; 1866 { 1867 int len, i; 1868 u_int32_t *tl; 1869 int32_t t1; 1870 struct uio uio; 1871 struct iovec iov; 1872 caddr_t dpos, cp2, cp; 1873 u_int32_t nfsvers, auth_type; 1874 uid_t nickuid; 1875 int error = 0, ticklen; 1876 struct mbuf *mrep, *md; 1877 struct nfsuid *nuidp; 1878 struct timeval tvin, tvout; 1879 1880 memset(&tvout, 0, sizeof tvout); /* XXX gcc */ 1881 1882 KASSERT(nd->nd_cr == NULL); 1883 mrep = nd->nd_mrep; 1884 md = nd->nd_md; 1885 dpos = nd->nd_dpos; 1886 if (has_header) { 1887 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1888 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1889 if (*tl++ != rpc_call) { 1890 m_freem(mrep); 1891 return (EBADRPC); 1892 } 1893 } else 1894 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1895 nd->nd_repstat = 0; 1896 nd->nd_flag = 0; 1897 if (*tl++ != rpc_vers) { 1898 nd->nd_repstat = ERPCMISMATCH; 1899 nd->nd_procnum = NFSPROC_NOOP; 1900 return (0); 1901 } 1902 if (*tl != nfs_prog) { 1903 nd->nd_repstat = EPROGUNAVAIL; 1904 nd->nd_procnum = NFSPROC_NOOP; 1905 return (0); 1906 } 1907 tl++; 1908 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1909 if (nfsvers < NFS_VER2 || nfsvers > NFS_VER3) { 1910 nd->nd_repstat = EPROGMISMATCH; 1911 nd->nd_procnum = NFSPROC_NOOP; 1912 return (0); 1913 } 1914 if (nfsvers == NFS_VER3) 1915 nd->nd_flag = ND_NFSV3; 1916 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1917 if (nd->nd_procnum == NFSPROC_NULL) 1918 return (0); 1919 if (nd->nd_procnum > NFSPROC_COMMIT || 1920 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1921 nd->nd_repstat = EPROCUNAVAIL; 1922 nd->nd_procnum = NFSPROC_NOOP; 1923 return (0); 1924 } 1925 if ((nd->nd_flag & ND_NFSV3) == 0) 1926 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1927 auth_type = *tl++; 1928 len = fxdr_unsigned(int, *tl++); 1929 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1930 m_freem(mrep); 1931 return (EBADRPC); 1932 } 1933 1934 nd->nd_flag &= ~ND_KERBAUTH; 1935 /* 1936 * Handle auth_unix or auth_kerb. 1937 */ 1938 if (auth_type == rpc_auth_unix) { 1939 uid_t uid; 1940 gid_t gid, *grbuf; 1941 1942 nd->nd_cr = kauth_cred_alloc(); 1943 len = fxdr_unsigned(int, *++tl); 1944 if (len < 0 || len > NFS_MAXNAMLEN) { 1945 m_freem(mrep); 1946 error = EBADRPC; 1947 goto errout; 1948 } 1949 nfsm_adv(nfsm_rndup(len)); 1950 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1951 1952 uid = fxdr_unsigned(uid_t, *tl++); 1953 gid = fxdr_unsigned(gid_t, *tl++); 1954 kauth_cred_setuid(nd->nd_cr, uid); 1955 kauth_cred_seteuid(nd->nd_cr, uid); 1956 kauth_cred_setsvuid(nd->nd_cr, uid); 1957 kauth_cred_setgid(nd->nd_cr, gid); 1958 kauth_cred_setegid(nd->nd_cr, gid); 1959 kauth_cred_setsvgid(nd->nd_cr, gid); 1960 1961 len = fxdr_unsigned(int, *tl); 1962 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1963 m_freem(mrep); 1964 error = EBADRPC; 1965 goto errout; 1966 } 1967 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1968 1969 grbuf = malloc(len * sizeof(gid_t), M_TEMP, M_WAITOK); 1970 for (i = 0; i < len; i++) { 1971 if (i < NGROUPS) /* XXX elad */ 1972 grbuf[i] = fxdr_unsigned(gid_t, *tl++); 1973 else 1974 tl++; 1975 } 1976 kauth_cred_setgroups(nd->nd_cr, grbuf, min(len, NGROUPS), -1); 1977 free(grbuf, M_TEMP); 1978 1979 len = fxdr_unsigned(int, *++tl); 1980 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1981 m_freem(mrep); 1982 error = EBADRPC; 1983 goto errout; 1984 } 1985 if (len > 0) 1986 nfsm_adv(nfsm_rndup(len)); 1987 } else if (auth_type == rpc_auth_kerb) { 1988 switch (fxdr_unsigned(int, *tl++)) { 1989 case RPCAKN_FULLNAME: 1990 ticklen = fxdr_unsigned(int, *tl); 1991 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 1992 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 1993 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 1994 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1995 m_freem(mrep); 1996 error = EBADRPC; 1997 goto errout; 1998 } 1999 uio.uio_offset = 0; 2000 uio.uio_iov = &iov; 2001 uio.uio_iovcnt = 1; 2002 UIO_SETUP_SYSSPACE(&uio); 2003 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 2004 iov.iov_len = RPCAUTH_MAXSIZ - 4; 2005 nfsm_mtouio(&uio, uio.uio_resid); 2006 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2007 if (*tl++ != rpc_auth_kerb || 2008 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 2009 printf("Bad kerb verifier\n"); 2010 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2011 nd->nd_procnum = NFSPROC_NOOP; 2012 return (0); 2013 } 2014 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 2015 tl = (u_int32_t *)cp; 2016 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 2017 printf("Not fullname kerb verifier\n"); 2018 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2019 nd->nd_procnum = NFSPROC_NOOP; 2020 return (0); 2021 } 2022 cp += NFSX_UNSIGNED; 2023 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 2024 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 2025 nd->nd_flag |= ND_KERBFULL; 2026 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 2027 break; 2028 case RPCAKN_NICKNAME: 2029 if (len != 2 * NFSX_UNSIGNED) { 2030 printf("Kerb nickname short\n"); 2031 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 2032 nd->nd_procnum = NFSPROC_NOOP; 2033 return (0); 2034 } 2035 nickuid = fxdr_unsigned(uid_t, *tl); 2036 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2037 if (*tl++ != rpc_auth_kerb || 2038 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 2039 printf("Kerb nick verifier bad\n"); 2040 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2041 nd->nd_procnum = NFSPROC_NOOP; 2042 return (0); 2043 } 2044 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2045 tvin.tv_sec = *tl++; 2046 tvin.tv_usec = *tl; 2047 2048 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), 2049 nu_hash) { 2050 if (kauth_cred_geteuid(nuidp->nu_cr) == nickuid && 2051 (!nd->nd_nam2 || 2052 netaddr_match(NU_NETFAM(nuidp), 2053 &nuidp->nu_haddr, nd->nd_nam2))) 2054 break; 2055 } 2056 if (!nuidp) { 2057 nd->nd_repstat = 2058 (NFSERR_AUTHERR|AUTH_REJECTCRED); 2059 nd->nd_procnum = NFSPROC_NOOP; 2060 return (0); 2061 } 2062 2063 /* 2064 * Now, decrypt the timestamp using the session key 2065 * and validate it. 2066 */ 2067 #ifdef NFSKERB 2068 XXX 2069 #endif 2070 2071 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 2072 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 2073 if (nuidp->nu_expire < time_second || 2074 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 2075 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 2076 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 2077 nuidp->nu_expire = 0; 2078 nd->nd_repstat = 2079 (NFSERR_AUTHERR|AUTH_REJECTVERF); 2080 nd->nd_procnum = NFSPROC_NOOP; 2081 return (0); 2082 } 2083 kauth_cred_hold(nuidp->nu_cr); 2084 nd->nd_cr = nuidp->nu_cr; 2085 nd->nd_flag |= ND_KERBNICK; 2086 } 2087 } else { 2088 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 2089 nd->nd_procnum = NFSPROC_NOOP; 2090 return (0); 2091 } 2092 2093 nd->nd_md = md; 2094 nd->nd_dpos = dpos; 2095 KASSERT((nd->nd_cr == NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) != 0) 2096 || (nd->nd_cr != NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) == 0)); 2097 return (0); 2098 nfsmout: 2099 errout: 2100 KASSERT(error != 0); 2101 if (nd->nd_cr != NULL) { 2102 kauth_cred_free(nd->nd_cr); 2103 nd->nd_cr = NULL; 2104 } 2105 return (error); 2106 } 2107 2108 int 2109 nfs_msg(l, server, msg) 2110 struct lwp *l; 2111 const char *server, *msg; 2112 { 2113 tpr_t tpr; 2114 2115 if (l) 2116 tpr = tprintf_open(l->l_proc); 2117 else 2118 tpr = NULL; 2119 tprintf(tpr, "nfs server %s: %s\n", server, msg); 2120 tprintf_close(tpr); 2121 return (0); 2122 } 2123 2124 #ifdef NFSSERVER 2125 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 2126 struct nfssvc_sock *, struct lwp *, 2127 struct mbuf **)) = { 2128 nfsrv_null, 2129 nfsrv_getattr, 2130 nfsrv_setattr, 2131 nfsrv_lookup, 2132 nfsrv3_access, 2133 nfsrv_readlink, 2134 nfsrv_read, 2135 nfsrv_write, 2136 nfsrv_create, 2137 nfsrv_mkdir, 2138 nfsrv_symlink, 2139 nfsrv_mknod, 2140 nfsrv_remove, 2141 nfsrv_rmdir, 2142 nfsrv_rename, 2143 nfsrv_link, 2144 nfsrv_readdir, 2145 nfsrv_readdirplus, 2146 nfsrv_statfs, 2147 nfsrv_fsinfo, 2148 nfsrv_pathconf, 2149 nfsrv_commit, 2150 nfsrv_noop 2151 }; 2152 2153 /* 2154 * Socket upcall routine for the nfsd sockets. 2155 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 2156 * Essentially do as much as possible non-blocking, else punt and it will 2157 * be called with M_WAIT from an nfsd. 2158 */ 2159 void 2160 nfsrv_rcv(so, arg, waitflag) 2161 struct socket *so; 2162 caddr_t arg; 2163 int waitflag; 2164 { 2165 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 2166 struct mbuf *m; 2167 struct mbuf *mp, *nam; 2168 struct uio auio; 2169 int flags, error; 2170 int setflags = 0; 2171 2172 error = nfsdsock_lock(slp, (waitflag != M_DONTWAIT)); 2173 if (error) { 2174 setflags |= SLP_NEEDQ; 2175 goto dorecs_unlocked; 2176 } 2177 2178 KASSERT(so == slp->ns_so); 2179 #define NFS_TEST_HEAVY 2180 #ifdef NFS_TEST_HEAVY 2181 /* 2182 * Define this to test for nfsds handling this under heavy load. 2183 * 2184 * XXX it isn't safe to call so_receive from so_upcall context. 2185 */ 2186 if (waitflag == M_DONTWAIT) { 2187 setflags |= SLP_NEEDQ; 2188 goto dorecs; 2189 } 2190 #endif 2191 simple_lock(&slp->ns_lock); 2192 slp->ns_flag &= ~SLP_NEEDQ; 2193 simple_unlock(&slp->ns_lock); 2194 if (so->so_type == SOCK_STREAM) { 2195 #ifndef NFS_TEST_HEAVY 2196 /* 2197 * If there are already records on the queue, defer soreceive() 2198 * to an nfsd so that there is feedback to the TCP layer that 2199 * the nfs servers are heavily loaded. 2200 */ 2201 if (slp->ns_rec && waitflag == M_DONTWAIT) { 2202 setflags |= SLP_NEEDQ; 2203 goto dorecs; 2204 } 2205 #endif 2206 2207 /* 2208 * Do soreceive(). 2209 */ 2210 auio.uio_resid = 1000000000; 2211 /* not need to setup uio_vmspace */ 2212 flags = MSG_DONTWAIT; 2213 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, &flags); 2214 if (error || mp == NULL) { 2215 if (error == EWOULDBLOCK) 2216 setflags |= SLP_NEEDQ; 2217 else 2218 setflags |= SLP_DISCONN; 2219 goto dorecs; 2220 } 2221 m = mp; 2222 m_claimm(m, &nfs_mowner); 2223 if (slp->ns_rawend) { 2224 slp->ns_rawend->m_next = m; 2225 slp->ns_cc += 1000000000 - auio.uio_resid; 2226 } else { 2227 slp->ns_raw = m; 2228 slp->ns_cc = 1000000000 - auio.uio_resid; 2229 } 2230 while (m->m_next) 2231 m = m->m_next; 2232 slp->ns_rawend = m; 2233 2234 /* 2235 * Now try and parse record(s) out of the raw stream data. 2236 */ 2237 error = nfsrv_getstream(slp, waitflag); 2238 if (error) { 2239 if (error == EPERM) 2240 setflags |= SLP_DISCONN; 2241 else 2242 setflags |= SLP_NEEDQ; 2243 } 2244 } else { 2245 do { 2246 auio.uio_resid = 1000000000; 2247 /* not need to setup uio_vmspace */ 2248 flags = MSG_DONTWAIT; 2249 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, 2250 &flags); 2251 if (mp) { 2252 if (nam) { 2253 m = nam; 2254 m->m_next = mp; 2255 } else 2256 m = mp; 2257 m_claimm(m, &nfs_mowner); 2258 if (slp->ns_recend) 2259 slp->ns_recend->m_nextpkt = m; 2260 else 2261 slp->ns_rec = m; 2262 slp->ns_recend = m; 2263 m->m_nextpkt = (struct mbuf *)0; 2264 } 2265 if (error) { 2266 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2267 && error != EWOULDBLOCK) { 2268 setflags |= SLP_DISCONN; 2269 goto dorecs; 2270 } 2271 } 2272 } while (mp); 2273 } 2274 dorecs: 2275 nfsdsock_unlock(slp); 2276 2277 dorecs_unlocked: 2278 /* 2279 * Now try and process the request records, non-blocking. 2280 */ 2281 if (setflags) { 2282 simple_lock(&slp->ns_lock); 2283 slp->ns_flag |= setflags; 2284 simple_unlock(&slp->ns_lock); 2285 } 2286 if (waitflag == M_DONTWAIT && 2287 (slp->ns_rec || (slp->ns_flag & (SLP_DISCONN | SLP_NEEDQ)) != 0)) { 2288 nfsrv_wakenfsd(slp); 2289 } 2290 } 2291 2292 int 2293 nfsdsock_lock(struct nfssvc_sock *slp, boolean_t waitok) 2294 { 2295 2296 simple_lock(&slp->ns_lock); 2297 while ((slp->ns_flag & (SLP_BUSY|SLP_VALID)) == SLP_BUSY) { 2298 if (!waitok) { 2299 simple_unlock(&slp->ns_lock); 2300 return EWOULDBLOCK; 2301 } 2302 slp->ns_flag |= SLP_WANT; 2303 ltsleep(&slp->ns_flag, PSOCK, "nslock", 0, &slp->ns_lock); 2304 } 2305 if ((slp->ns_flag & SLP_VALID) == 0) { 2306 simple_unlock(&slp->ns_lock); 2307 return EINVAL; 2308 } 2309 slp->ns_flag |= SLP_BUSY; 2310 simple_unlock(&slp->ns_lock); 2311 2312 return 0; 2313 } 2314 2315 void 2316 nfsdsock_unlock(struct nfssvc_sock *slp) 2317 { 2318 2319 KASSERT((slp->ns_flag & SLP_BUSY) != 0); 2320 2321 simple_lock(&slp->ns_lock); 2322 if ((slp->ns_flag & SLP_WANT) != 0) { 2323 wakeup(&slp->ns_flag); 2324 } 2325 slp->ns_flag &= ~(SLP_BUSY|SLP_WANT); 2326 simple_unlock(&slp->ns_lock); 2327 } 2328 2329 int 2330 nfsdsock_drain(struct nfssvc_sock *slp) 2331 { 2332 int error = 0; 2333 2334 simple_lock(&slp->ns_lock); 2335 if ((slp->ns_flag & SLP_VALID) == 0) { 2336 error = EINVAL; 2337 goto done; 2338 } 2339 slp->ns_flag &= ~SLP_VALID; 2340 while ((slp->ns_flag & SLP_BUSY) != 0) { 2341 slp->ns_flag |= SLP_WANT; 2342 ltsleep(&slp->ns_flag, PSOCK, "nsdrain", 0, &slp->ns_lock); 2343 } 2344 done: 2345 simple_unlock(&slp->ns_lock); 2346 2347 return error; 2348 } 2349 2350 /* 2351 * Try and extract an RPC request from the mbuf data list received on a 2352 * stream socket. The "waitflag" argument indicates whether or not it 2353 * can sleep. 2354 */ 2355 int 2356 nfsrv_getstream(slp, waitflag) 2357 struct nfssvc_sock *slp; 2358 int waitflag; 2359 { 2360 struct mbuf *m, **mpp; 2361 struct mbuf *recm; 2362 u_int32_t recmark; 2363 int error = 0; 2364 2365 for (;;) { 2366 if (slp->ns_reclen == 0) { 2367 if (slp->ns_cc < NFSX_UNSIGNED) { 2368 break; 2369 } 2370 m = slp->ns_raw; 2371 m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark); 2372 m_adj(m, NFSX_UNSIGNED); 2373 slp->ns_cc -= NFSX_UNSIGNED; 2374 recmark = ntohl(recmark); 2375 slp->ns_reclen = recmark & ~0x80000000; 2376 if (recmark & 0x80000000) 2377 slp->ns_flag |= SLP_LASTFRAG; 2378 else 2379 slp->ns_flag &= ~SLP_LASTFRAG; 2380 if (slp->ns_reclen > NFS_MAXPACKET) { 2381 error = EPERM; 2382 break; 2383 } 2384 } 2385 2386 /* 2387 * Now get the record part. 2388 * 2389 * Note that slp->ns_reclen may be 0. Linux sometimes 2390 * generates 0-length records. 2391 */ 2392 if (slp->ns_cc == slp->ns_reclen) { 2393 recm = slp->ns_raw; 2394 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2395 slp->ns_cc = slp->ns_reclen = 0; 2396 } else if (slp->ns_cc > slp->ns_reclen) { 2397 recm = slp->ns_raw; 2398 m = m_split(recm, slp->ns_reclen, waitflag); 2399 if (m == NULL) { 2400 error = EWOULDBLOCK; 2401 break; 2402 } 2403 m_claimm(recm, &nfs_mowner); 2404 slp->ns_raw = m; 2405 if (m->m_next == NULL) 2406 slp->ns_rawend = m; 2407 slp->ns_cc -= slp->ns_reclen; 2408 slp->ns_reclen = 0; 2409 } else { 2410 break; 2411 } 2412 2413 /* 2414 * Accumulate the fragments into a record. 2415 */ 2416 mpp = &slp->ns_frag; 2417 while (*mpp) 2418 mpp = &((*mpp)->m_next); 2419 *mpp = recm; 2420 if (slp->ns_flag & SLP_LASTFRAG) { 2421 if (slp->ns_recend) 2422 slp->ns_recend->m_nextpkt = slp->ns_frag; 2423 else 2424 slp->ns_rec = slp->ns_frag; 2425 slp->ns_recend = slp->ns_frag; 2426 slp->ns_frag = (struct mbuf *)0; 2427 } 2428 } 2429 2430 return error; 2431 } 2432 2433 /* 2434 * Parse an RPC header. 2435 */ 2436 int 2437 nfsrv_dorec(slp, nfsd, ndp) 2438 struct nfssvc_sock *slp; 2439 struct nfsd *nfsd; 2440 struct nfsrv_descript **ndp; 2441 { 2442 struct mbuf *m, *nam; 2443 struct nfsrv_descript *nd; 2444 int error; 2445 2446 *ndp = NULL; 2447 2448 if (nfsdsock_lock(slp, TRUE)) { 2449 return ENOBUFS; 2450 } 2451 m = slp->ns_rec; 2452 if (m == NULL) { 2453 nfsdsock_unlock(slp); 2454 return ENOBUFS; 2455 } 2456 slp->ns_rec = m->m_nextpkt; 2457 if (slp->ns_rec) 2458 m->m_nextpkt = NULL; 2459 else 2460 slp->ns_recend = NULL; 2461 nfsdsock_unlock(slp); 2462 2463 if (m->m_type == MT_SONAME) { 2464 nam = m; 2465 m = m->m_next; 2466 nam->m_next = NULL; 2467 } else 2468 nam = NULL; 2469 nd = nfsdreq_alloc(); 2470 nd->nd_md = nd->nd_mrep = m; 2471 nd->nd_nam2 = nam; 2472 nd->nd_dpos = mtod(m, caddr_t); 2473 error = nfs_getreq(nd, nfsd, TRUE); 2474 if (error) { 2475 m_freem(nam); 2476 nfsdreq_free(nd); 2477 return (error); 2478 } 2479 *ndp = nd; 2480 nfsd->nfsd_nd = nd; 2481 return (0); 2482 } 2483 2484 /* 2485 * Search for a sleeping nfsd and wake it up. 2486 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2487 * running nfsds will go look for the work in the nfssvc_sock list. 2488 */ 2489 void 2490 nfsrv_wakenfsd(slp) 2491 struct nfssvc_sock *slp; 2492 { 2493 struct nfsd *nd; 2494 2495 if ((slp->ns_flag & SLP_VALID) == 0) 2496 return; 2497 simple_lock(&nfsd_slock); 2498 if (slp->ns_flag & SLP_DOREC) { 2499 simple_unlock(&nfsd_slock); 2500 return; 2501 } 2502 nd = SLIST_FIRST(&nfsd_idle_head); 2503 if (nd) { 2504 SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle); 2505 simple_unlock(&nfsd_slock); 2506 2507 if (nd->nfsd_slp) 2508 panic("nfsd wakeup"); 2509 slp->ns_sref++; 2510 nd->nfsd_slp = slp; 2511 wakeup(nd); 2512 return; 2513 } 2514 slp->ns_flag |= SLP_DOREC; 2515 nfsd_head_flag |= NFSD_CHECKSLP; 2516 TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending); 2517 simple_unlock(&nfsd_slock); 2518 } 2519 2520 int 2521 nfsdsock_sendreply(struct nfssvc_sock *slp, struct nfsrv_descript *nd) 2522 { 2523 int error; 2524 2525 if (nd->nd_mrep != NULL) { 2526 m_freem(nd->nd_mrep); 2527 nd->nd_mrep = NULL; 2528 } 2529 2530 simple_lock(&slp->ns_lock); 2531 if ((slp->ns_flag & SLP_SENDING) != 0) { 2532 SIMPLEQ_INSERT_TAIL(&slp->ns_sendq, nd, nd_sendq); 2533 simple_unlock(&slp->ns_lock); 2534 return 0; 2535 } 2536 KASSERT(SIMPLEQ_EMPTY(&slp->ns_sendq)); 2537 slp->ns_flag |= SLP_SENDING; 2538 simple_unlock(&slp->ns_lock); 2539 2540 again: 2541 error = nfs_send(slp->ns_so, nd->nd_nam2, nd->nd_mreq, NULL, curlwp); 2542 if (nd->nd_nam2) { 2543 m_free(nd->nd_nam2); 2544 } 2545 nfsdreq_free(nd); 2546 2547 simple_lock(&slp->ns_lock); 2548 KASSERT((slp->ns_flag & SLP_SENDING) != 0); 2549 nd = SIMPLEQ_FIRST(&slp->ns_sendq); 2550 if (nd != NULL) { 2551 SIMPLEQ_REMOVE_HEAD(&slp->ns_sendq, nd_sendq); 2552 simple_unlock(&slp->ns_lock); 2553 goto again; 2554 } 2555 slp->ns_flag &= ~SLP_SENDING; 2556 simple_unlock(&slp->ns_lock); 2557 2558 return error; 2559 } 2560 #endif /* NFSSERVER */ 2561 2562 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) 2563 static struct pool nfs_srvdesc_pool; 2564 2565 void 2566 nfsdreq_init(void) 2567 { 2568 2569 pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript), 2570 0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr); 2571 } 2572 2573 struct nfsrv_descript * 2574 nfsdreq_alloc(void) 2575 { 2576 struct nfsrv_descript *nd; 2577 2578 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK); 2579 nd->nd_cr = NULL; 2580 return nd; 2581 } 2582 2583 void 2584 nfsdreq_free(struct nfsrv_descript *nd) 2585 { 2586 kauth_cred_t cr; 2587 2588 cr = nd->nd_cr; 2589 if (cr != NULL) { 2590 kauth_cred_free(cr); 2591 } 2592 pool_put(&nfs_srvdesc_pool, nd); 2593 } 2594 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */ 2595
Indexes created Sun Sep 21 15:09:59 GMT 2025