nfs_socket.c revision 1.151
1 /* $NetBSD: nfs_socket.c,v 1.151 2007/04/29 14:56:59 yamt Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1991, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 35 */ 36 37 /* 38 * Socket operations for use by nfs 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: nfs_socket.c,v 1.151 2007/04/29 14:56:59 yamt Exp $"); 43 44 #include "fs_nfs.h" 45 #include "opt_nfs.h" 46 #include "opt_nfsserver.h" 47 #include "opt_mbuftrace.h" 48 #include "opt_inet.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/callout.h> 53 #include <sys/proc.h> 54 #include <sys/mount.h> 55 #include <sys/kernel.h> 56 #include <sys/mbuf.h> 57 #include <sys/vnode.h> 58 #include <sys/domain.h> 59 #include <sys/protosw.h> 60 #include <sys/socket.h> 61 #include <sys/socketvar.h> 62 #include <sys/syslog.h> 63 #include <sys/tprintf.h> 64 #include <sys/namei.h> 65 #include <sys/signal.h> 66 #include <sys/signalvar.h> 67 #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 static int nfs_sndlock(struct nfsmount *, struct nfsreq *); 173 static void nfs_sndunlock(struct nfsmount *); 174 static int nfs_rcvlock(struct nfsmount *, struct nfsreq *); 175 static void nfs_rcvunlock(struct nfsmount *); 176 177 /* 178 * Initialize sockets and congestion for a new NFS connection. 179 * We do not free the sockaddr if error. 180 */ 181 int 182 nfs_connect(nmp, rep, l) 183 struct nfsmount *nmp; 184 struct nfsreq *rep; 185 struct lwp *l; 186 { 187 struct socket *so; 188 int s, error, rcvreserve, sndreserve; 189 struct sockaddr *saddr; 190 struct sockaddr_in *sin; 191 #ifdef INET6 192 struct sockaddr_in6 *sin6; 193 #endif 194 struct mbuf *m; 195 196 nmp->nm_so = (struct socket *)0; 197 saddr = mtod(nmp->nm_nam, struct sockaddr *); 198 error = socreate(saddr->sa_family, &nmp->nm_so, 199 nmp->nm_sotype, nmp->nm_soproto, l); 200 if (error) 201 goto bad; 202 so = nmp->nm_so; 203 #ifdef MBUFTRACE 204 so->so_mowner = &nfs_mowner; 205 so->so_rcv.sb_mowner = &nfs_mowner; 206 so->so_snd.sb_mowner = &nfs_mowner; 207 #endif 208 nmp->nm_soflags = so->so_proto->pr_flags; 209 210 /* 211 * Some servers require that the client port be a reserved port number. 212 */ 213 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 214 m = m_get(M_WAIT, MT_SOOPTS); 215 MCLAIM(m, so->so_mowner); 216 *mtod(m, int32_t *) = IP_PORTRANGE_LOW; 217 m->m_len = sizeof(int32_t); 218 if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m))) 219 goto bad; 220 m = m_get(M_WAIT, MT_SONAME); 221 MCLAIM(m, so->so_mowner); 222 sin = mtod(m, struct sockaddr_in *); 223 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 224 sin->sin_family = AF_INET; 225 sin->sin_addr.s_addr = INADDR_ANY; 226 sin->sin_port = 0; 227 error = sobind(so, m, &lwp0); 228 m_freem(m); 229 if (error) 230 goto bad; 231 } 232 #ifdef INET6 233 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) { 234 m = m_get(M_WAIT, MT_SOOPTS); 235 MCLAIM(m, so->so_mowner); 236 *mtod(m, int32_t *) = IPV6_PORTRANGE_LOW; 237 m->m_len = sizeof(int32_t); 238 if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m))) 239 goto bad; 240 m = m_get(M_WAIT, MT_SONAME); 241 MCLAIM(m, so->so_mowner); 242 sin6 = mtod(m, struct sockaddr_in6 *); 243 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6); 244 sin6->sin6_family = AF_INET6; 245 sin6->sin6_addr = in6addr_any; 246 sin6->sin6_port = 0; 247 error = sobind(so, m, &lwp0); 248 m_freem(m); 249 if (error) 250 goto bad; 251 } 252 #endif 253 254 /* 255 * Protocols that do not require connections may be optionally left 256 * unconnected for servers that reply from a port other than NFS_PORT. 257 */ 258 if (nmp->nm_flag & NFSMNT_NOCONN) { 259 if (nmp->nm_soflags & PR_CONNREQUIRED) { 260 error = ENOTCONN; 261 goto bad; 262 } 263 } else { 264 error = soconnect(so, nmp->nm_nam, l); 265 if (error) 266 goto bad; 267 268 /* 269 * Wait for the connection to complete. Cribbed from the 270 * connect system call but with the wait timing out so 271 * that interruptible mounts don't hang here for a long time. 272 */ 273 s = splsoftnet(); 274 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 275 (void) tsleep((void *)&so->so_timeo, PSOCK, 276 "nfscn1", 2 * hz); 277 if ((so->so_state & SS_ISCONNECTING) && 278 so->so_error == 0 && rep && 279 (error = nfs_sigintr(nmp, rep, rep->r_lwp)) != 0){ 280 so->so_state &= ~SS_ISCONNECTING; 281 splx(s); 282 goto bad; 283 } 284 } 285 if (so->so_error) { 286 error = so->so_error; 287 so->so_error = 0; 288 splx(s); 289 goto bad; 290 } 291 splx(s); 292 } 293 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 294 so->so_rcv.sb_timeo = (5 * hz); 295 so->so_snd.sb_timeo = (5 * hz); 296 } else { 297 /* 298 * enable receive timeout to detect server crash and reconnect. 299 * otherwise, we can be stuck in soreceive forever. 300 */ 301 so->so_rcv.sb_timeo = (5 * hz); 302 so->so_snd.sb_timeo = 0; 303 } 304 if (nmp->nm_sotype == SOCK_DGRAM) { 305 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 306 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 307 NFS_MAXPKTHDR) * 2; 308 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 309 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 310 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 311 NFS_MAXPKTHDR) * 2; 312 } else { 313 if (nmp->nm_sotype != SOCK_STREAM) 314 panic("nfscon sotype"); 315 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 316 m = m_get(M_WAIT, MT_SOOPTS); 317 MCLAIM(m, so->so_mowner); 318 *mtod(m, int32_t *) = 1; 319 m->m_len = sizeof(int32_t); 320 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 321 } 322 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 323 m = m_get(M_WAIT, MT_SOOPTS); 324 MCLAIM(m, so->so_mowner); 325 *mtod(m, int32_t *) = 1; 326 m->m_len = sizeof(int32_t); 327 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 328 } 329 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 330 sizeof (u_int32_t)) * 2; 331 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 332 sizeof (u_int32_t)) * 2; 333 } 334 error = soreserve(so, sndreserve, rcvreserve); 335 if (error) 336 goto bad; 337 so->so_rcv.sb_flags |= SB_NOINTR; 338 so->so_snd.sb_flags |= SB_NOINTR; 339 340 /* Initialize other non-zero congestion variables */ 341 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 342 NFS_TIMEO << 3; 343 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 344 nmp->nm_sdrtt[3] = 0; 345 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 346 nmp->nm_sent = 0; 347 nmp->nm_timeouts = 0; 348 return (0); 349 350 bad: 351 nfs_disconnect(nmp); 352 return (error); 353 } 354 355 /* 356 * Reconnect routine: 357 * Called when a connection is broken on a reliable protocol. 358 * - clean up the old socket 359 * - nfs_connect() again 360 * - set R_MUSTRESEND for all outstanding requests on mount point 361 * If this fails the mount point is DEAD! 362 * nb: Must be called with the nfs_sndlock() set on the mount point. 363 */ 364 int 365 nfs_reconnect(rep, l) 366 struct nfsreq *rep; 367 struct lwp *l; 368 { 369 struct nfsreq *rp; 370 struct nfsmount *nmp = rep->r_nmp; 371 int error; 372 373 nfs_disconnect(nmp); 374 while ((error = nfs_connect(nmp, rep, l)) != 0) { 375 if (error == EINTR || error == ERESTART) 376 return (EINTR); 377 (void) tsleep((void *)&lbolt, PSOCK, "nfscn2", 0); 378 } 379 380 /* 381 * Loop through outstanding request list and fix up all requests 382 * on old socket. 383 */ 384 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 385 if (rp->r_nmp == nmp) { 386 if ((rp->r_flags & R_MUSTRESEND) == 0) 387 rp->r_flags |= R_MUSTRESEND | R_REXMITTED; 388 rp->r_rexmit = 0; 389 } 390 } 391 return (0); 392 } 393 394 /* 395 * NFS disconnect. Clean up and unlink. 396 */ 397 void 398 nfs_disconnect(nmp) 399 struct nfsmount *nmp; 400 { 401 struct socket *so; 402 int drain = 0; 403 404 if (nmp->nm_so) { 405 so = nmp->nm_so; 406 nmp->nm_so = (struct socket *)0; 407 soshutdown(so, SHUT_RDWR); 408 drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0; 409 if (drain) { 410 /* 411 * soshutdown() above should wake up the current 412 * listener. 413 * Now wake up those waiting for the receive lock, and 414 * wait for them to go away unhappy, to prevent *nmp 415 * from evaporating while they're sleeping. 416 */ 417 mutex_enter(&nmp->nm_lock); 418 while (nmp->nm_waiters > 0) { 419 cv_broadcast(&nmp->nm_rcvcv); 420 cv_broadcast(&nmp->nm_sndcv); 421 cv_wait(&nmp->nm_disconcv, &nmp->nm_lock); 422 } 423 mutex_exit(&nmp->nm_lock); 424 } 425 soclose(so); 426 } 427 #ifdef DIAGNOSTIC 428 if (drain && (nmp->nm_waiters > 0)) 429 panic("nfs_disconnect: waiters left after drain?"); 430 #endif 431 } 432 433 void 434 nfs_safedisconnect(nmp) 435 struct nfsmount *nmp; 436 { 437 struct nfsreq dummyreq; 438 439 memset(&dummyreq, 0, sizeof(dummyreq)); 440 dummyreq.r_nmp = nmp; 441 nfs_rcvlock(nmp, &dummyreq); /* XXX ignored error return */ 442 nfs_disconnect(nmp); 443 nfs_rcvunlock(nmp); 444 } 445 446 /* 447 * This is the nfs send routine. For connection based socket types, it 448 * must be called with an nfs_sndlock() on the socket. 449 * "rep == NULL" indicates that it has been called from a server. 450 * For the client side: 451 * - return EINTR if the RPC is terminated, 0 otherwise 452 * - set R_MUSTRESEND if the send fails for any reason 453 * - do any cleanup required by recoverable socket errors (? ? ?) 454 * For the server side: 455 * - return EINTR or ERESTART if interrupted by a signal 456 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 457 * - do any cleanup required by recoverable socket errors (? ? ?) 458 */ 459 int 460 nfs_send(so, nam, top, rep, l) 461 struct socket *so; 462 struct mbuf *nam; 463 struct mbuf *top; 464 struct nfsreq *rep; 465 struct lwp *l; 466 { 467 struct mbuf *sendnam; 468 int error, soflags, flags; 469 470 /* XXX nfs_doio()/nfs_request() calls with rep->r_lwp == NULL */ 471 if (l == NULL && rep->r_lwp == NULL) 472 l = curlwp; 473 474 if (rep) { 475 if (rep->r_flags & R_SOFTTERM) { 476 m_freem(top); 477 return (EINTR); 478 } 479 if ((so = rep->r_nmp->nm_so) == NULL) { 480 rep->r_flags |= R_MUSTRESEND; 481 m_freem(top); 482 return (0); 483 } 484 rep->r_flags &= ~R_MUSTRESEND; 485 soflags = rep->r_nmp->nm_soflags; 486 } else 487 soflags = so->so_proto->pr_flags; 488 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 489 sendnam = (struct mbuf *)0; 490 else 491 sendnam = nam; 492 if (so->so_type == SOCK_SEQPACKET) 493 flags = MSG_EOR; 494 else 495 flags = 0; 496 497 error = (*so->so_send)(so, sendnam, (struct uio *)0, top, 498 (struct mbuf *)0, flags, l); 499 if (error) { 500 if (rep) { 501 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 502 /* 503 * We're too fast for the network/driver, 504 * and UDP isn't flowcontrolled. 505 * We need to resend. This is not fatal, 506 * just try again. 507 * 508 * Could be smarter here by doing some sort 509 * of a backoff, but this is rare. 510 */ 511 rep->r_flags |= R_MUSTRESEND; 512 } else { 513 if (error != EPIPE) 514 log(LOG_INFO, 515 "nfs send error %d for %s\n", 516 error, 517 rep->r_nmp->nm_mountp-> 518 mnt_stat.f_mntfromname); 519 /* 520 * Deal with errors for the client side. 521 */ 522 if (rep->r_flags & R_SOFTTERM) 523 error = EINTR; 524 else 525 rep->r_flags |= R_MUSTRESEND; 526 } 527 } else { 528 /* 529 * See above. This error can happen under normal 530 * circumstances and the log is too noisy. 531 * The error will still show up in nfsstat. 532 */ 533 if (error != ENOBUFS || so->so_type != SOCK_DGRAM) 534 log(LOG_INFO, "nfsd send error %d\n", error); 535 } 536 537 /* 538 * Handle any recoverable (soft) socket errors here. (? ? ?) 539 */ 540 if (error != EINTR && error != ERESTART && 541 error != EWOULDBLOCK && error != EPIPE) 542 error = 0; 543 } 544 return (error); 545 } 546 547 #ifdef NFS 548 /* 549 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 550 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 551 * Mark and consolidate the data into a new mbuf list. 552 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 553 * small mbufs. 554 * For SOCK_STREAM we must be very careful to read an entire record once 555 * we have read any of it, even if the system call has been interrupted. 556 */ 557 int 558 nfs_receive(rep, aname, mp, l) 559 struct nfsreq *rep; 560 struct mbuf **aname; 561 struct mbuf **mp; 562 struct lwp *l; 563 { 564 struct socket *so; 565 struct uio auio; 566 struct iovec aio; 567 struct mbuf *m; 568 struct mbuf *control; 569 u_int32_t len; 570 struct mbuf **getnam; 571 int error, sotype, rcvflg; 572 573 /* 574 * Set up arguments for soreceive() 575 */ 576 *mp = (struct mbuf *)0; 577 *aname = (struct mbuf *)0; 578 sotype = rep->r_nmp->nm_sotype; 579 580 /* 581 * For reliable protocols, lock against other senders/receivers 582 * in case a reconnect is necessary. 583 * For SOCK_STREAM, first get the Record Mark to find out how much 584 * more there is to get. 585 * We must lock the socket against other receivers 586 * until we have an entire rpc request/reply. 587 */ 588 if (sotype != SOCK_DGRAM) { 589 error = nfs_sndlock(rep->r_nmp, rep); 590 if (error) 591 return (error); 592 tryagain: 593 /* 594 * Check for fatal errors and resending request. 595 */ 596 /* 597 * Ugh: If a reconnect attempt just happened, nm_so 598 * would have changed. NULL indicates a failed 599 * attempt that has essentially shut down this 600 * mount point. 601 */ 602 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 603 nfs_sndunlock(rep->r_nmp); 604 return (EINTR); 605 } 606 so = rep->r_nmp->nm_so; 607 if (!so) { 608 error = nfs_reconnect(rep, l); 609 if (error) { 610 nfs_sndunlock(rep->r_nmp); 611 return (error); 612 } 613 goto tryagain; 614 } 615 while (rep->r_flags & R_MUSTRESEND) { 616 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 617 nfsstats.rpcretries++; 618 rep->r_rtt = 0; 619 rep->r_flags &= ~R_TIMING; 620 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, l); 621 if (error) { 622 if (error == EINTR || error == ERESTART || 623 (error = nfs_reconnect(rep, l)) != 0) { 624 nfs_sndunlock(rep->r_nmp); 625 return (error); 626 } 627 goto tryagain; 628 } 629 } 630 nfs_sndunlock(rep->r_nmp); 631 if (sotype == SOCK_STREAM) { 632 aio.iov_base = (void *) &len; 633 aio.iov_len = sizeof(u_int32_t); 634 auio.uio_iov = &aio; 635 auio.uio_iovcnt = 1; 636 auio.uio_rw = UIO_READ; 637 auio.uio_offset = 0; 638 auio.uio_resid = sizeof(u_int32_t); 639 UIO_SETUP_SYSSPACE(&auio); 640 do { 641 rcvflg = MSG_WAITALL; 642 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 643 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 644 if (error == EWOULDBLOCK && rep) { 645 if (rep->r_flags & R_SOFTTERM) 646 return (EINTR); 647 /* 648 * if it seems that the server died after it 649 * received our request, set EPIPE so that 650 * we'll reconnect and retransmit requests. 651 */ 652 if (rep->r_rexmit >= rep->r_nmp->nm_retry) { 653 nfsstats.rpctimeouts++; 654 error = EPIPE; 655 } 656 } 657 } while (error == EWOULDBLOCK); 658 if (!error && auio.uio_resid > 0) { 659 /* 660 * Don't log a 0 byte receive; it means 661 * that the socket has been closed, and 662 * can happen during normal operation 663 * (forcible unmount or Solaris server). 664 */ 665 if (auio.uio_resid != sizeof (u_int32_t)) 666 log(LOG_INFO, 667 "short receive (%lu/%lu) from nfs server %s\n", 668 (u_long)sizeof(u_int32_t) - auio.uio_resid, 669 (u_long)sizeof(u_int32_t), 670 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 671 error = EPIPE; 672 } 673 if (error) 674 goto errout; 675 len = ntohl(len) & ~0x80000000; 676 /* 677 * This is SERIOUS! We are out of sync with the sender 678 * and forcing a disconnect/reconnect is all I can do. 679 */ 680 if (len > NFS_MAXPACKET) { 681 log(LOG_ERR, "%s (%d) from nfs server %s\n", 682 "impossible packet length", 683 len, 684 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 685 error = EFBIG; 686 goto errout; 687 } 688 auio.uio_resid = len; 689 do { 690 rcvflg = MSG_WAITALL; 691 error = (*so->so_receive)(so, (struct mbuf **)0, 692 &auio, mp, (struct mbuf **)0, &rcvflg); 693 } while (error == EWOULDBLOCK || error == EINTR || 694 error == ERESTART); 695 if (!error && auio.uio_resid > 0) { 696 if (len != auio.uio_resid) 697 log(LOG_INFO, 698 "short receive (%lu/%d) from nfs server %s\n", 699 (u_long)len - auio.uio_resid, len, 700 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 701 error = EPIPE; 702 } 703 } else { 704 /* 705 * NB: Since uio_resid is big, MSG_WAITALL is ignored 706 * and soreceive() will return when it has either a 707 * control msg or a data msg. 708 * We have no use for control msg., but must grab them 709 * and then throw them away so we know what is going 710 * on. 711 */ 712 auio.uio_resid = len = 100000000; /* Anything Big */ 713 /* not need to setup uio_vmspace */ 714 do { 715 rcvflg = 0; 716 error = (*so->so_receive)(so, (struct mbuf **)0, 717 &auio, mp, &control, &rcvflg); 718 if (control) 719 m_freem(control); 720 if (error == EWOULDBLOCK && rep) { 721 if (rep->r_flags & R_SOFTTERM) 722 return (EINTR); 723 } 724 } while (error == EWOULDBLOCK || 725 (!error && *mp == NULL && control)); 726 if ((rcvflg & MSG_EOR) == 0) 727 printf("Egad!!\n"); 728 if (!error && *mp == NULL) 729 error = EPIPE; 730 len -= auio.uio_resid; 731 } 732 errout: 733 if (error && error != EINTR && error != ERESTART) { 734 m_freem(*mp); 735 *mp = (struct mbuf *)0; 736 if (error != EPIPE) 737 log(LOG_INFO, 738 "receive error %d from nfs server %s\n", 739 error, 740 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 741 error = nfs_sndlock(rep->r_nmp, rep); 742 if (!error) 743 error = nfs_reconnect(rep, l); 744 if (!error) 745 goto tryagain; 746 else 747 nfs_sndunlock(rep->r_nmp); 748 } 749 } else { 750 if ((so = rep->r_nmp->nm_so) == NULL) 751 return (EACCES); 752 if (so->so_state & SS_ISCONNECTED) 753 getnam = (struct mbuf **)0; 754 else 755 getnam = aname; 756 auio.uio_resid = len = 1000000; 757 /* not need to setup uio_vmspace */ 758 do { 759 rcvflg = 0; 760 error = (*so->so_receive)(so, getnam, &auio, mp, 761 (struct mbuf **)0, &rcvflg); 762 if (error == EWOULDBLOCK && 763 (rep->r_flags & R_SOFTTERM)) 764 return (EINTR); 765 } while (error == EWOULDBLOCK); 766 len -= auio.uio_resid; 767 if (!error && *mp == NULL) 768 error = EPIPE; 769 } 770 if (error) { 771 m_freem(*mp); 772 *mp = (struct mbuf *)0; 773 } 774 return (error); 775 } 776 777 /* 778 * Implement receipt of reply on a socket. 779 * We must search through the list of received datagrams matching them 780 * with outstanding requests using the xid, until ours is found. 781 */ 782 /* ARGSUSED */ 783 int 784 nfs_reply(myrep, lwp) 785 struct nfsreq *myrep; 786 struct lwp *lwp; 787 { 788 struct nfsreq *rep; 789 struct nfsmount *nmp = myrep->r_nmp; 790 int32_t t1; 791 struct mbuf *mrep, *nam, *md; 792 u_int32_t rxid, *tl; 793 char *dpos, *cp2; 794 int error; 795 796 /* 797 * Loop around until we get our own reply 798 */ 799 for (;;) { 800 /* 801 * Lock against other receivers so that I don't get stuck in 802 * sbwait() after someone else has received my reply for me. 803 * Also necessary for connection based protocols to avoid 804 * race conditions during a reconnect. 805 */ 806 error = nfs_rcvlock(nmp, myrep); 807 if (error == EALREADY) 808 return (0); 809 if (error) 810 return (error); 811 /* 812 * Get the next Rpc reply off the socket 813 */ 814 815 mutex_enter(&nmp->nm_lock); 816 nmp->nm_waiters++; 817 mutex_exit(&nmp->nm_lock); 818 819 error = nfs_receive(myrep, &nam, &mrep, lwp); 820 nfs_rcvunlock(nmp); 821 822 mutex_enter(&nmp->nm_lock); 823 nmp->nm_waiters--; 824 cv_signal(&nmp->nm_disconcv); 825 mutex_exit(&nmp->nm_lock); 826 827 if (error) { 828 829 if (nmp->nm_iflag & NFSMNT_DISMNT) { 830 /* 831 * Oops, we're going away now.. 832 */ 833 return error; 834 } 835 /* 836 * Ignore routing errors on connectionless protocols? ? 837 */ 838 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 839 nmp->nm_so->so_error = 0; 840 #ifdef DEBUG 841 printf("nfs_reply: ignoring error %d\n", error); 842 #endif 843 if (myrep->r_flags & R_GETONEREP) 844 return (0); 845 continue; 846 } 847 return (error); 848 } 849 if (nam) 850 m_freem(nam); 851 852 /* 853 * Get the xid and check that it is an rpc reply 854 */ 855 md = mrep; 856 dpos = mtod(md, void *); 857 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 858 rxid = *tl++; 859 if (*tl != rpc_reply) { 860 nfsstats.rpcinvalid++; 861 m_freem(mrep); 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 getmicrotime(&rt->tstamp); 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 char **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 waituntil; 982 char *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 struct mbuf *mrest_backup = NULL; 991 kauth_cred_t origcred = NULL; /* XXX: gcc */ 992 bool retry_cred = true; 993 bool use_opencred = (np->n_flag & NUSEOPENCRED) != 0; 994 995 if (rexmitp != NULL) 996 *rexmitp = 0; 997 998 acred = kauth_cred_alloc(); 999 1000 tryagain_cred: 1001 KASSERT(cred != NULL); 1002 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 1003 rep->r_nmp = nmp; 1004 rep->r_lwp = lwp; 1005 rep->r_procnum = procnum; 1006 i = 0; 1007 m = mrest; 1008 while (m) { 1009 i += m->m_len; 1010 m = m->m_next; 1011 } 1012 mrest_len = i; 1013 1014 /* 1015 * Get the RPC header with authorization. 1016 */ 1017 kerbauth: 1018 verf_str = auth_str = (char *)0; 1019 if (nmp->nm_flag & NFSMNT_KERB) { 1020 verf_str = nickv; 1021 verf_len = sizeof (nickv); 1022 auth_type = RPCAUTH_KERB4; 1023 memset((void *)key, 0, sizeof (key)); 1024 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 1025 &auth_len, verf_str, verf_len)) { 1026 error = nfs_getauth(nmp, rep, cred, &auth_str, 1027 &auth_len, verf_str, &verf_len, key); 1028 if (error) { 1029 free((void *)rep, M_NFSREQ); 1030 m_freem(mrest); 1031 KASSERT(kauth_cred_getrefcnt(acred) == 1); 1032 kauth_cred_free(acred); 1033 return (error); 1034 } 1035 } 1036 retry_cred = false; 1037 } else { 1038 /* AUTH_UNIX */ 1039 uid_t uid; 1040 gid_t gid; 1041 1042 /* 1043 * on the most unix filesystems, permission checks are 1044 * done when the file is open(2)'ed. 1045 * ie. once a file is successfully open'ed, 1046 * following i/o operations never fail with EACCES. 1047 * we try to follow the semantics as far as possible. 1048 * 1049 * note that we expect that the nfs server always grant 1050 * accesses by the file's owner. 1051 */ 1052 origcred = cred; 1053 switch (procnum) { 1054 case NFSPROC_READ: 1055 case NFSPROC_WRITE: 1056 case NFSPROC_COMMIT: 1057 uid = np->n_vattr->va_uid; 1058 gid = np->n_vattr->va_gid; 1059 if (kauth_cred_geteuid(cred) == uid && 1060 kauth_cred_getegid(cred) == gid) { 1061 retry_cred = false; 1062 break; 1063 } 1064 if (use_opencred) 1065 break; 1066 kauth_cred_setuid(acred, uid); 1067 kauth_cred_seteuid(acred, uid); 1068 kauth_cred_setsvuid(acred, uid); 1069 kauth_cred_setgid(acred, gid); 1070 kauth_cred_setegid(acred, gid); 1071 kauth_cred_setsvgid(acred, gid); 1072 cred = acred; 1073 break; 1074 default: 1075 retry_cred = false; 1076 break; 1077 } 1078 /* 1079 * backup mbuf chain if we can need it later to retry. 1080 * 1081 * XXX maybe we can keep a direct reference to 1082 * mrest without doing m_copym, but it's ...ugly. 1083 */ 1084 if (retry_cred) 1085 mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT); 1086 auth_type = RPCAUTH_UNIX; 1087 /* XXX elad - ngroups */ 1088 auth_len = (((kauth_cred_ngroups(cred) > nmp->nm_numgrps) ? 1089 nmp->nm_numgrps : kauth_cred_ngroups(cred)) << 2) + 1090 5 * NFSX_UNSIGNED; 1091 } 1092 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 1093 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 1094 if (auth_str) 1095 free(auth_str, M_TEMP); 1096 1097 /* 1098 * For stream protocols, insert a Sun RPC Record Mark. 1099 */ 1100 if (nmp->nm_sotype == SOCK_STREAM) { 1101 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 1102 *mtod(m, u_int32_t *) = htonl(0x80000000 | 1103 (m->m_pkthdr.len - NFSX_UNSIGNED)); 1104 } 1105 rep->r_mreq = m; 1106 rep->r_xid = xid; 1107 tryagain: 1108 if (nmp->nm_flag & NFSMNT_SOFT) 1109 rep->r_retry = nmp->nm_retry; 1110 else 1111 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1112 rep->r_rtt = rep->r_rexmit = 0; 1113 if (proct[procnum] > 0) 1114 rep->r_flags = R_TIMING; 1115 else 1116 rep->r_flags = 0; 1117 rep->r_mrep = NULL; 1118 1119 /* 1120 * Do the client side RPC. 1121 */ 1122 nfsstats.rpcrequests++; 1123 /* 1124 * Chain request into list of outstanding requests. Be sure 1125 * to put it LAST so timer finds oldest requests first. 1126 */ 1127 s = splsoftnet(); 1128 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1129 1130 /* 1131 * If backing off another request or avoiding congestion, don't 1132 * send this one now but let timer do it. If not timing a request, 1133 * do it now. 1134 */ 1135 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1136 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1137 nmp->nm_sent < nmp->nm_cwnd)) { 1138 splx(s); 1139 if (nmp->nm_soflags & PR_CONNREQUIRED) 1140 error = nfs_sndlock(nmp, rep); 1141 if (!error) { 1142 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1143 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp); 1144 if (nmp->nm_soflags & PR_CONNREQUIRED) 1145 nfs_sndunlock(nmp); 1146 } 1147 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1148 nmp->nm_sent += NFS_CWNDSCALE; 1149 rep->r_flags |= R_SENT; 1150 } 1151 } else { 1152 splx(s); 1153 rep->r_rtt = -1; 1154 } 1155 1156 /* 1157 * Wait for the reply from our send or the timer's. 1158 */ 1159 if (!error || error == EPIPE) 1160 error = nfs_reply(rep, lwp); 1161 1162 /* 1163 * RPC done, unlink the request. 1164 */ 1165 s = splsoftnet(); 1166 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1167 splx(s); 1168 1169 /* 1170 * Decrement the outstanding request count. 1171 */ 1172 if (rep->r_flags & R_SENT) { 1173 rep->r_flags &= ~R_SENT; /* paranoia */ 1174 nmp->nm_sent -= NFS_CWNDSCALE; 1175 } 1176 1177 if (rexmitp != NULL) { 1178 int rexmit; 1179 1180 if (nmp->nm_sotype != SOCK_DGRAM) 1181 rexmit = (rep->r_flags & R_REXMITTED) != 0; 1182 else 1183 rexmit = rep->r_rexmit; 1184 *rexmitp = rexmit; 1185 } 1186 1187 /* 1188 * If there was a successful reply and a tprintf msg. 1189 * tprintf a response. 1190 */ 1191 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1192 nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1193 "is alive again"); 1194 mrep = rep->r_mrep; 1195 md = rep->r_md; 1196 dpos = rep->r_dpos; 1197 if (error) 1198 goto nfsmout; 1199 1200 /* 1201 * break down the rpc header and check if ok 1202 */ 1203 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1204 if (*tl++ == rpc_msgdenied) { 1205 if (*tl == rpc_mismatch) 1206 error = EOPNOTSUPP; 1207 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1208 if (!failed_auth) { 1209 failed_auth++; 1210 mheadend->m_next = (struct mbuf *)0; 1211 m_freem(mrep); 1212 m_freem(rep->r_mreq); 1213 goto kerbauth; 1214 } else 1215 error = EAUTH; 1216 } else 1217 error = EACCES; 1218 m_freem(mrep); 1219 goto nfsmout; 1220 } 1221 1222 /* 1223 * Grab any Kerberos verifier, otherwise just throw it away. 1224 */ 1225 verf_type = fxdr_unsigned(int, *tl++); 1226 i = fxdr_unsigned(int32_t, *tl); 1227 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1228 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1229 if (error) 1230 goto nfsmout; 1231 } else if (i > 0) 1232 nfsm_adv(nfsm_rndup(i)); 1233 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1234 /* 0 == ok */ 1235 if (*tl == 0) { 1236 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1237 if (*tl != 0) { 1238 error = fxdr_unsigned(int, *tl); 1239 switch (error) { 1240 case NFSERR_PERM: 1241 error = EPERM; 1242 break; 1243 1244 case NFSERR_NOENT: 1245 error = ENOENT; 1246 break; 1247 1248 case NFSERR_IO: 1249 error = EIO; 1250 break; 1251 1252 case NFSERR_NXIO: 1253 error = ENXIO; 1254 break; 1255 1256 case NFSERR_ACCES: 1257 error = EACCES; 1258 if (!retry_cred) 1259 break; 1260 m_freem(mrep); 1261 m_freem(rep->r_mreq); 1262 FREE(rep, M_NFSREQ); 1263 use_opencred = !use_opencred; 1264 if (mrest_backup == NULL) { 1265 /* m_copym failure */ 1266 KASSERT( 1267 kauth_cred_getrefcnt(acred) == 1); 1268 kauth_cred_free(acred); 1269 return ENOMEM; 1270 } 1271 mrest = mrest_backup; 1272 mrest_backup = NULL; 1273 cred = origcred; 1274 error = 0; 1275 retry_cred = false; 1276 goto tryagain_cred; 1277 1278 case NFSERR_EXIST: 1279 error = EEXIST; 1280 break; 1281 1282 case NFSERR_XDEV: 1283 error = EXDEV; 1284 break; 1285 1286 case NFSERR_NODEV: 1287 error = ENODEV; 1288 break; 1289 1290 case NFSERR_NOTDIR: 1291 error = ENOTDIR; 1292 break; 1293 1294 case NFSERR_ISDIR: 1295 error = EISDIR; 1296 break; 1297 1298 case NFSERR_INVAL: 1299 error = EINVAL; 1300 break; 1301 1302 case NFSERR_FBIG: 1303 error = EFBIG; 1304 break; 1305 1306 case NFSERR_NOSPC: 1307 error = ENOSPC; 1308 break; 1309 1310 case NFSERR_ROFS: 1311 error = EROFS; 1312 break; 1313 1314 case NFSERR_MLINK: 1315 error = EMLINK; 1316 break; 1317 1318 case NFSERR_TIMEDOUT: 1319 error = ETIMEDOUT; 1320 break; 1321 1322 case NFSERR_NAMETOL: 1323 error = ENAMETOOLONG; 1324 break; 1325 1326 case NFSERR_NOTEMPTY: 1327 error = ENOTEMPTY; 1328 break; 1329 1330 case NFSERR_DQUOT: 1331 error = EDQUOT; 1332 break; 1333 1334 case NFSERR_STALE: 1335 /* 1336 * If the File Handle was stale, invalidate the 1337 * lookup cache, just in case. 1338 */ 1339 error = ESTALE; 1340 cache_purge(NFSTOV(np)); 1341 break; 1342 1343 case NFSERR_REMOTE: 1344 error = EREMOTE; 1345 break; 1346 1347 case NFSERR_WFLUSH: 1348 case NFSERR_BADHANDLE: 1349 case NFSERR_NOT_SYNC: 1350 case NFSERR_BAD_COOKIE: 1351 error = EINVAL; 1352 break; 1353 1354 case NFSERR_NOTSUPP: 1355 error = ENOTSUP; 1356 break; 1357 1358 case NFSERR_TOOSMALL: 1359 case NFSERR_SERVERFAULT: 1360 case NFSERR_BADTYPE: 1361 error = EINVAL; 1362 break; 1363 1364 case NFSERR_TRYLATER: 1365 if ((nmp->nm_flag & NFSMNT_NFSV3) == 0) 1366 break; 1367 m_freem(mrep); 1368 error = 0; 1369 waituntil = time_second + trylater_delay; 1370 while (time_second < waituntil) { 1371 tsleep(&lbolt, PSOCK, "nfstrylater", 0); 1372 } 1373 trylater_delay *= NFS_TRYLATERDELMUL; 1374 if (trylater_delay > NFS_TRYLATERDELMAX) 1375 trylater_delay = NFS_TRYLATERDELMAX; 1376 /* 1377 * RFC1813: 1378 * The client should wait and then try 1379 * the request with a new RPC transaction ID. 1380 */ 1381 nfs_renewxid(rep); 1382 goto tryagain; 1383 1384 default: 1385 #ifdef DIAGNOSTIC 1386 printf("Invalid rpc error code %d\n", error); 1387 #endif 1388 error = EINVAL; 1389 break; 1390 } 1391 1392 if (nmp->nm_flag & NFSMNT_NFSV3) { 1393 *mrp = mrep; 1394 *mdp = md; 1395 *dposp = dpos; 1396 error |= NFSERR_RETERR; 1397 } else 1398 m_freem(mrep); 1399 goto nfsmout; 1400 } 1401 1402 /* 1403 * note which credential worked to minimize number of retries. 1404 */ 1405 if (use_opencred) 1406 np->n_flag |= NUSEOPENCRED; 1407 else 1408 np->n_flag &= ~NUSEOPENCRED; 1409 1410 *mrp = mrep; 1411 *mdp = md; 1412 *dposp = dpos; 1413 1414 KASSERT(error == 0); 1415 goto nfsmout; 1416 } 1417 m_freem(mrep); 1418 error = EPROTONOSUPPORT; 1419 nfsmout: 1420 KASSERT(kauth_cred_getrefcnt(acred) == 1); 1421 kauth_cred_free(acred); 1422 m_freem(rep->r_mreq); 1423 free((void *)rep, M_NFSREQ); 1424 m_freem(mrest_backup); 1425 return (error); 1426 } 1427 #endif /* NFS */ 1428 1429 /* 1430 * Generate the rpc reply header 1431 * siz arg. is used to decide if adding a cluster is worthwhile 1432 */ 1433 int 1434 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1435 int siz; 1436 struct nfsrv_descript *nd; 1437 struct nfssvc_sock *slp; 1438 int err; 1439 int cache; 1440 u_quad_t *frev; 1441 struct mbuf **mrq; 1442 struct mbuf **mbp; 1443 char **bposp; 1444 { 1445 u_int32_t *tl; 1446 struct mbuf *mreq; 1447 char *bpos; 1448 struct mbuf *mb; 1449 1450 mreq = m_gethdr(M_WAIT, MT_DATA); 1451 MCLAIM(mreq, &nfs_mowner); 1452 mb = mreq; 1453 /* 1454 * If this is a big reply, use a cluster else 1455 * try and leave leading space for the lower level headers. 1456 */ 1457 siz += RPC_REPLYSIZ; 1458 if (siz >= max_datalen) { 1459 m_clget(mreq, M_WAIT); 1460 } else 1461 mreq->m_data += max_hdr; 1462 tl = mtod(mreq, u_int32_t *); 1463 mreq->m_len = 6 * NFSX_UNSIGNED; 1464 bpos = ((char *)tl) + mreq->m_len; 1465 *tl++ = txdr_unsigned(nd->nd_retxid); 1466 *tl++ = rpc_reply; 1467 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1468 *tl++ = rpc_msgdenied; 1469 if (err & NFSERR_AUTHERR) { 1470 *tl++ = rpc_autherr; 1471 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1472 mreq->m_len -= NFSX_UNSIGNED; 1473 bpos -= NFSX_UNSIGNED; 1474 } else { 1475 *tl++ = rpc_mismatch; 1476 *tl++ = txdr_unsigned(RPC_VER2); 1477 *tl = txdr_unsigned(RPC_VER2); 1478 } 1479 } else { 1480 *tl++ = rpc_msgaccepted; 1481 1482 /* 1483 * For Kerberos authentication, we must send the nickname 1484 * verifier back, otherwise just RPCAUTH_NULL. 1485 */ 1486 if (nd->nd_flag & ND_KERBFULL) { 1487 struct nfsuid *nuidp; 1488 struct timeval ktvin, ktvout; 1489 1490 memset(&ktvout, 0, sizeof ktvout); /* XXX gcc */ 1491 1492 LIST_FOREACH(nuidp, 1493 NUIDHASH(slp, kauth_cred_geteuid(nd->nd_cr)), 1494 nu_hash) { 1495 if (kauth_cred_geteuid(nuidp->nu_cr) == 1496 kauth_cred_geteuid(nd->nd_cr) && 1497 (!nd->nd_nam2 || netaddr_match( 1498 NU_NETFAM(nuidp), &nuidp->nu_haddr, 1499 nd->nd_nam2))) 1500 break; 1501 } 1502 if (nuidp) { 1503 ktvin.tv_sec = 1504 txdr_unsigned(nuidp->nu_timestamp.tv_sec 1505 - 1); 1506 ktvin.tv_usec = 1507 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1508 1509 /* 1510 * Encrypt the timestamp in ecb mode using the 1511 * session key. 1512 */ 1513 #ifdef NFSKERB 1514 XXX 1515 #endif 1516 1517 *tl++ = rpc_auth_kerb; 1518 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1519 *tl = ktvout.tv_sec; 1520 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1521 *tl++ = ktvout.tv_usec; 1522 *tl++ = txdr_unsigned( 1523 kauth_cred_geteuid(nuidp->nu_cr)); 1524 } else { 1525 *tl++ = 0; 1526 *tl++ = 0; 1527 } 1528 } else { 1529 *tl++ = 0; 1530 *tl++ = 0; 1531 } 1532 switch (err) { 1533 case EPROGUNAVAIL: 1534 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1535 break; 1536 case EPROGMISMATCH: 1537 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1538 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1539 *tl++ = txdr_unsigned(2); 1540 *tl = txdr_unsigned(3); 1541 break; 1542 case EPROCUNAVAIL: 1543 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1544 break; 1545 case EBADRPC: 1546 *tl = txdr_unsigned(RPC_GARBAGE); 1547 break; 1548 default: 1549 *tl = 0; 1550 if (err != NFSERR_RETVOID) { 1551 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1552 if (err) 1553 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1554 else 1555 *tl = 0; 1556 } 1557 break; 1558 }; 1559 } 1560 1561 if (mrq != NULL) 1562 *mrq = mreq; 1563 *mbp = mb; 1564 *bposp = bpos; 1565 if (err != 0 && err != NFSERR_RETVOID) 1566 nfsstats.srvrpc_errs++; 1567 return (0); 1568 } 1569 1570 /* 1571 * Nfs timer routine 1572 * Scan the nfsreq list and retranmit any requests that have timed out 1573 * To avoid retransmission attempts on STREAM sockets (in the future) make 1574 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1575 */ 1576 void 1577 nfs_timer(void *arg) 1578 { 1579 struct nfsreq *rep; 1580 struct mbuf *m; 1581 struct socket *so; 1582 struct nfsmount *nmp; 1583 int timeo; 1584 int s, error; 1585 #ifdef NFSSERVER 1586 struct timeval tv; 1587 struct nfssvc_sock *slp; 1588 u_quad_t cur_usec; 1589 #endif 1590 1591 s = splsoftnet(); 1592 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1593 nmp = rep->r_nmp; 1594 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1595 continue; 1596 if (nfs_sigintr(nmp, rep, rep->r_lwp)) { 1597 rep->r_flags |= R_SOFTTERM; 1598 continue; 1599 } 1600 if (rep->r_rtt >= 0) { 1601 rep->r_rtt++; 1602 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1603 timeo = nmp->nm_timeo; 1604 else 1605 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1606 if (nmp->nm_timeouts > 0) 1607 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1608 if (rep->r_rtt <= timeo) 1609 continue; 1610 if (nmp->nm_timeouts < 1611 (sizeof(nfs_backoff) / sizeof(nfs_backoff[0]))) 1612 nmp->nm_timeouts++; 1613 } 1614 /* 1615 * Check for server not responding 1616 */ 1617 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1618 rep->r_rexmit > nmp->nm_deadthresh) { 1619 nfs_msg(rep->r_lwp, 1620 nmp->nm_mountp->mnt_stat.f_mntfromname, 1621 "not responding"); 1622 rep->r_flags |= R_TPRINTFMSG; 1623 } 1624 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1625 nfsstats.rpctimeouts++; 1626 rep->r_flags |= R_SOFTTERM; 1627 continue; 1628 } 1629 if (nmp->nm_sotype != SOCK_DGRAM) { 1630 if (++rep->r_rexmit > NFS_MAXREXMIT) 1631 rep->r_rexmit = NFS_MAXREXMIT; 1632 continue; 1633 } 1634 if ((so = nmp->nm_so) == NULL) 1635 continue; 1636 1637 /* 1638 * If there is enough space and the window allows.. 1639 * Resend it 1640 * Set r_rtt to -1 in case we fail to send it now. 1641 */ 1642 rep->r_rtt = -1; 1643 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1644 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1645 (rep->r_flags & R_SENT) || 1646 nmp->nm_sent < nmp->nm_cwnd) && 1647 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1648 if (so->so_state & SS_ISCONNECTED) 1649 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1650 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1651 else 1652 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1653 nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0); 1654 if (error) { 1655 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1656 #ifdef DEBUG 1657 printf("nfs_timer: ignoring error %d\n", 1658 error); 1659 #endif 1660 so->so_error = 0; 1661 } 1662 } else { 1663 /* 1664 * Iff first send, start timing 1665 * else turn timing off, backoff timer 1666 * and divide congestion window by 2. 1667 */ 1668 if (rep->r_flags & R_SENT) { 1669 rep->r_flags &= ~R_TIMING; 1670 if (++rep->r_rexmit > NFS_MAXREXMIT) 1671 rep->r_rexmit = NFS_MAXREXMIT; 1672 nmp->nm_cwnd >>= 1; 1673 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1674 nmp->nm_cwnd = NFS_CWNDSCALE; 1675 nfsstats.rpcretries++; 1676 } else { 1677 rep->r_flags |= R_SENT; 1678 nmp->nm_sent += NFS_CWNDSCALE; 1679 } 1680 rep->r_rtt = 0; 1681 } 1682 } 1683 } 1684 1685 #ifdef NFSSERVER 1686 /* 1687 * Scan the write gathering queues for writes that need to be 1688 * completed now. 1689 */ 1690 getmicrotime(&tv); 1691 cur_usec = (u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec; 1692 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1693 if (LIST_FIRST(&slp->ns_tq) && 1694 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) 1695 nfsrv_wakenfsd(slp); 1696 } 1697 #endif /* NFSSERVER */ 1698 splx(s); 1699 callout_schedule(&nfs_timer_ch, nfs_ticks); 1700 } 1701 1702 /*ARGSUSED*/ 1703 void 1704 nfs_exit(struct proc *p, void *v) 1705 { 1706 struct nfsreq *rp; 1707 int s = splsoftnet(); 1708 1709 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 1710 if (rp->r_lwp && rp->r_lwp->l_proc == p) 1711 TAILQ_REMOVE(&nfs_reqq, rp, r_chain); 1712 } 1713 splx(s); 1714 } 1715 1716 /* 1717 * Test for a termination condition pending on the process. 1718 * This is used for NFSMNT_INT mounts. 1719 */ 1720 int 1721 nfs_sigintr(nmp, rep, l) 1722 struct nfsmount *nmp; 1723 struct nfsreq *rep; 1724 struct lwp *l; 1725 { 1726 sigset_t ss; 1727 1728 if (rep && (rep->r_flags & R_SOFTTERM)) 1729 return (EINTR); 1730 if (!(nmp->nm_flag & NFSMNT_INT)) 1731 return (0); 1732 if (l) { 1733 sigpending1(l, &ss); 1734 #if 0 1735 sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss); 1736 #endif 1737 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1738 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1739 sigismember(&ss, SIGQUIT)) 1740 return (EINTR); 1741 } 1742 return (0); 1743 } 1744 1745 /* 1746 * Lock a socket against others. 1747 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1748 * and also to avoid race conditions between the processes with nfs requests 1749 * in progress when a reconnect is necessary. 1750 */ 1751 static int 1752 nfs_sndlock(struct nfsmount *nmp, struct nfsreq *rep) 1753 { 1754 struct lwp *l; 1755 int timeo = 0; 1756 bool catch = false; 1757 int error = 0; 1758 1759 if (rep) { 1760 l = rep->r_lwp; 1761 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1762 catch = true; 1763 } else 1764 l = NULL; 1765 mutex_enter(&nmp->nm_lock); 1766 while ((nmp->nm_iflag & NFSMNT_SNDLOCK) != 0) { 1767 if (rep && nfs_sigintr(rep->r_nmp, rep, l)) { 1768 error = EINTR; 1769 goto quit; 1770 } 1771 if (catch) { 1772 cv_timedwait_sig(&nmp->nm_sndcv, &nmp->nm_lock, timeo); 1773 } else { 1774 cv_timedwait(&nmp->nm_sndcv, &nmp->nm_lock, timeo); 1775 } 1776 if (catch) { 1777 catch = false; 1778 timeo = 2 * hz; 1779 } 1780 } 1781 nmp->nm_iflag |= NFSMNT_SNDLOCK; 1782 quit: 1783 mutex_exit(&nmp->nm_lock); 1784 return error; 1785 } 1786 1787 /* 1788 * Unlock the stream socket for others. 1789 */ 1790 static void 1791 nfs_sndunlock(struct nfsmount *nmp) 1792 { 1793 1794 mutex_enter(&nmp->nm_lock); 1795 if ((nmp->nm_iflag & NFSMNT_SNDLOCK) == 0) 1796 panic("nfs sndunlock"); 1797 nmp->nm_iflag &= ~NFSMNT_SNDLOCK; 1798 cv_signal(&nmp->nm_sndcv); 1799 mutex_exit(&nmp->nm_lock); 1800 } 1801 1802 static int 1803 nfs_rcvlock(struct nfsmount *nmp, struct nfsreq *rep) 1804 { 1805 int *flagp = &nmp->nm_iflag; 1806 int slptimeo = 0; 1807 bool catch; 1808 int error = 0; 1809 1810 KASSERT(nmp == rep->r_nmp); 1811 1812 if (*flagp & NFSMNT_DISMNT) 1813 return EIO; 1814 1815 if (*flagp & NFSMNT_INT) 1816 catch = true; 1817 else 1818 catch = false; 1819 mutex_enter(&nmp->nm_lock); 1820 while (*flagp & NFSMNT_RCVLOCK) { 1821 if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) { 1822 error = EINTR; 1823 goto quit; 1824 } 1825 if (catch) { 1826 cv_timedwait_sig(&nmp->nm_rcvcv, &nmp->nm_lock, 1827 slptimeo); 1828 } else { 1829 cv_timedwait(&nmp->nm_rcvcv, &nmp->nm_lock, 1830 slptimeo); 1831 } 1832 if (*flagp & NFSMNT_DISMNT) { 1833 cv_signal(&nmp->nm_disconcv); 1834 error = EIO; 1835 goto quit; 1836 } 1837 /* If our reply was received while we were sleeping, 1838 * then just return without taking the lock to avoid a 1839 * situation where a single iod could 'capture' the 1840 * receive lock. 1841 */ 1842 if (rep->r_mrep != NULL) { 1843 cv_signal(&nmp->nm_rcvcv); 1844 error = EALREADY; 1845 goto quit; 1846 } 1847 if (catch) { 1848 catch = false; 1849 slptimeo = 2 * hz; 1850 } 1851 } 1852 *flagp |= NFSMNT_RCVLOCK; 1853 quit: 1854 mutex_exit(&nmp->nm_lock); 1855 return error; 1856 } 1857 1858 /* 1859 * Unlock the stream socket for others. 1860 */ 1861 static void 1862 nfs_rcvunlock(struct nfsmount *nmp) 1863 { 1864 1865 mutex_enter(&nmp->nm_lock); 1866 if ((nmp->nm_iflag & NFSMNT_RCVLOCK) == 0) 1867 panic("nfs rcvunlock"); 1868 nmp->nm_iflag &= ~NFSMNT_RCVLOCK; 1869 cv_signal(&nmp->nm_rcvcv); 1870 mutex_exit(&nmp->nm_lock); 1871 } 1872 1873 /* 1874 * Parse an RPC request 1875 * - verify it 1876 * - allocate and fill in the cred. 1877 */ 1878 int 1879 nfs_getreq(nd, nfsd, has_header) 1880 struct nfsrv_descript *nd; 1881 struct nfsd *nfsd; 1882 int has_header; 1883 { 1884 int len, i; 1885 u_int32_t *tl; 1886 int32_t t1; 1887 struct uio uio; 1888 struct iovec iov; 1889 char *dpos, *cp2, *cp; 1890 u_int32_t nfsvers, auth_type; 1891 uid_t nickuid; 1892 int error = 0, ticklen; 1893 struct mbuf *mrep, *md; 1894 struct nfsuid *nuidp; 1895 struct timeval tvin, tvout; 1896 1897 memset(&tvout, 0, sizeof tvout); /* XXX gcc */ 1898 1899 KASSERT(nd->nd_cr == NULL); 1900 mrep = nd->nd_mrep; 1901 md = nd->nd_md; 1902 dpos = nd->nd_dpos; 1903 if (has_header) { 1904 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1905 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1906 if (*tl++ != rpc_call) { 1907 m_freem(mrep); 1908 return (EBADRPC); 1909 } 1910 } else 1911 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1912 nd->nd_repstat = 0; 1913 nd->nd_flag = 0; 1914 if (*tl++ != rpc_vers) { 1915 nd->nd_repstat = ERPCMISMATCH; 1916 nd->nd_procnum = NFSPROC_NOOP; 1917 return (0); 1918 } 1919 if (*tl != nfs_prog) { 1920 nd->nd_repstat = EPROGUNAVAIL; 1921 nd->nd_procnum = NFSPROC_NOOP; 1922 return (0); 1923 } 1924 tl++; 1925 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1926 if (nfsvers < NFS_VER2 || nfsvers > NFS_VER3) { 1927 nd->nd_repstat = EPROGMISMATCH; 1928 nd->nd_procnum = NFSPROC_NOOP; 1929 return (0); 1930 } 1931 if (nfsvers == NFS_VER3) 1932 nd->nd_flag = ND_NFSV3; 1933 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1934 if (nd->nd_procnum == NFSPROC_NULL) 1935 return (0); 1936 if (nd->nd_procnum > NFSPROC_COMMIT || 1937 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1938 nd->nd_repstat = EPROCUNAVAIL; 1939 nd->nd_procnum = NFSPROC_NOOP; 1940 return (0); 1941 } 1942 if ((nd->nd_flag & ND_NFSV3) == 0) 1943 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1944 auth_type = *tl++; 1945 len = fxdr_unsigned(int, *tl++); 1946 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1947 m_freem(mrep); 1948 return (EBADRPC); 1949 } 1950 1951 nd->nd_flag &= ~ND_KERBAUTH; 1952 /* 1953 * Handle auth_unix or auth_kerb. 1954 */ 1955 if (auth_type == rpc_auth_unix) { 1956 uid_t uid; 1957 gid_t gid, *grbuf; 1958 1959 nd->nd_cr = kauth_cred_alloc(); 1960 len = fxdr_unsigned(int, *++tl); 1961 if (len < 0 || len > NFS_MAXNAMLEN) { 1962 m_freem(mrep); 1963 error = EBADRPC; 1964 goto errout; 1965 } 1966 nfsm_adv(nfsm_rndup(len)); 1967 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1968 1969 uid = fxdr_unsigned(uid_t, *tl++); 1970 gid = fxdr_unsigned(gid_t, *tl++); 1971 kauth_cred_setuid(nd->nd_cr, uid); 1972 kauth_cred_seteuid(nd->nd_cr, uid); 1973 kauth_cred_setsvuid(nd->nd_cr, uid); 1974 kauth_cred_setgid(nd->nd_cr, gid); 1975 kauth_cred_setegid(nd->nd_cr, gid); 1976 kauth_cred_setsvgid(nd->nd_cr, gid); 1977 1978 len = fxdr_unsigned(int, *tl); 1979 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1980 m_freem(mrep); 1981 error = EBADRPC; 1982 goto errout; 1983 } 1984 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 1985 1986 grbuf = malloc(len * sizeof(gid_t), M_TEMP, M_WAITOK); 1987 for (i = 0; i < len; i++) { 1988 if (i < NGROUPS) /* XXX elad */ 1989 grbuf[i] = fxdr_unsigned(gid_t, *tl++); 1990 else 1991 tl++; 1992 } 1993 kauth_cred_setgroups(nd->nd_cr, grbuf, min(len, NGROUPS), -1); 1994 free(grbuf, M_TEMP); 1995 1996 len = fxdr_unsigned(int, *++tl); 1997 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1998 m_freem(mrep); 1999 error = EBADRPC; 2000 goto errout; 2001 } 2002 if (len > 0) 2003 nfsm_adv(nfsm_rndup(len)); 2004 } else if (auth_type == rpc_auth_kerb) { 2005 switch (fxdr_unsigned(int, *tl++)) { 2006 case RPCAKN_FULLNAME: 2007 ticklen = fxdr_unsigned(int, *tl); 2008 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 2009 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 2010 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 2011 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 2012 m_freem(mrep); 2013 error = EBADRPC; 2014 goto errout; 2015 } 2016 uio.uio_offset = 0; 2017 uio.uio_iov = &iov; 2018 uio.uio_iovcnt = 1; 2019 UIO_SETUP_SYSSPACE(&uio); 2020 iov.iov_base = (void *)&nfsd->nfsd_authstr[4]; 2021 iov.iov_len = RPCAUTH_MAXSIZ - 4; 2022 nfsm_mtouio(&uio, uio.uio_resid); 2023 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2024 if (*tl++ != rpc_auth_kerb || 2025 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 2026 printf("Bad kerb verifier\n"); 2027 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2028 nd->nd_procnum = NFSPROC_NOOP; 2029 return (0); 2030 } 2031 nfsm_dissect(cp, void *, 4 * NFSX_UNSIGNED); 2032 tl = (u_int32_t *)cp; 2033 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 2034 printf("Not fullname kerb verifier\n"); 2035 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2036 nd->nd_procnum = NFSPROC_NOOP; 2037 return (0); 2038 } 2039 cp += NFSX_UNSIGNED; 2040 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 2041 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 2042 nd->nd_flag |= ND_KERBFULL; 2043 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 2044 break; 2045 case RPCAKN_NICKNAME: 2046 if (len != 2 * NFSX_UNSIGNED) { 2047 printf("Kerb nickname short\n"); 2048 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 2049 nd->nd_procnum = NFSPROC_NOOP; 2050 return (0); 2051 } 2052 nickuid = fxdr_unsigned(uid_t, *tl); 2053 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2054 if (*tl++ != rpc_auth_kerb || 2055 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 2056 printf("Kerb nick verifier bad\n"); 2057 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2058 nd->nd_procnum = NFSPROC_NOOP; 2059 return (0); 2060 } 2061 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2062 tvin.tv_sec = *tl++; 2063 tvin.tv_usec = *tl; 2064 2065 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), 2066 nu_hash) { 2067 if (kauth_cred_geteuid(nuidp->nu_cr) == nickuid && 2068 (!nd->nd_nam2 || 2069 netaddr_match(NU_NETFAM(nuidp), 2070 &nuidp->nu_haddr, nd->nd_nam2))) 2071 break; 2072 } 2073 if (!nuidp) { 2074 nd->nd_repstat = 2075 (NFSERR_AUTHERR|AUTH_REJECTCRED); 2076 nd->nd_procnum = NFSPROC_NOOP; 2077 return (0); 2078 } 2079 2080 /* 2081 * Now, decrypt the timestamp using the session key 2082 * and validate it. 2083 */ 2084 #ifdef NFSKERB 2085 XXX 2086 #endif 2087 2088 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 2089 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 2090 if (nuidp->nu_expire < time_second || 2091 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 2092 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 2093 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 2094 nuidp->nu_expire = 0; 2095 nd->nd_repstat = 2096 (NFSERR_AUTHERR|AUTH_REJECTVERF); 2097 nd->nd_procnum = NFSPROC_NOOP; 2098 return (0); 2099 } 2100 kauth_cred_hold(nuidp->nu_cr); 2101 nd->nd_cr = nuidp->nu_cr; 2102 nd->nd_flag |= ND_KERBNICK; 2103 } 2104 } else { 2105 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 2106 nd->nd_procnum = NFSPROC_NOOP; 2107 return (0); 2108 } 2109 2110 nd->nd_md = md; 2111 nd->nd_dpos = dpos; 2112 KASSERT((nd->nd_cr == NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) != 0) 2113 || (nd->nd_cr != NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) == 0)); 2114 return (0); 2115 nfsmout: 2116 errout: 2117 KASSERT(error != 0); 2118 if (nd->nd_cr != NULL) { 2119 kauth_cred_free(nd->nd_cr); 2120 nd->nd_cr = NULL; 2121 } 2122 return (error); 2123 } 2124 2125 int 2126 nfs_msg(l, server, msg) 2127 struct lwp *l; 2128 const char *server, *msg; 2129 { 2130 tpr_t tpr; 2131 2132 if (l) 2133 tpr = tprintf_open(l->l_proc); 2134 else 2135 tpr = NULL; 2136 tprintf(tpr, "nfs server %s: %s\n", server, msg); 2137 tprintf_close(tpr); 2138 return (0); 2139 } 2140 2141 #ifdef NFSSERVER 2142 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 2143 struct nfssvc_sock *, struct lwp *, 2144 struct mbuf **)) = { 2145 nfsrv_null, 2146 nfsrv_getattr, 2147 nfsrv_setattr, 2148 nfsrv_lookup, 2149 nfsrv3_access, 2150 nfsrv_readlink, 2151 nfsrv_read, 2152 nfsrv_write, 2153 nfsrv_create, 2154 nfsrv_mkdir, 2155 nfsrv_symlink, 2156 nfsrv_mknod, 2157 nfsrv_remove, 2158 nfsrv_rmdir, 2159 nfsrv_rename, 2160 nfsrv_link, 2161 nfsrv_readdir, 2162 nfsrv_readdirplus, 2163 nfsrv_statfs, 2164 nfsrv_fsinfo, 2165 nfsrv_pathconf, 2166 nfsrv_commit, 2167 nfsrv_noop 2168 }; 2169 2170 /* 2171 * Socket upcall routine for the nfsd sockets. 2172 * The void *arg is a pointer to the "struct nfssvc_sock". 2173 * Essentially do as much as possible non-blocking, else punt and it will 2174 * be called with M_WAIT from an nfsd. 2175 */ 2176 void 2177 nfsrv_rcv(so, arg, waitflag) 2178 struct socket *so; 2179 void *arg; 2180 int waitflag; 2181 { 2182 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 2183 struct mbuf *m; 2184 struct mbuf *mp, *nam; 2185 struct uio auio; 2186 int flags, error; 2187 int setflags = 0; 2188 2189 error = nfsdsock_lock(slp, (waitflag != M_DONTWAIT)); 2190 if (error) { 2191 setflags |= SLP_NEEDQ; 2192 goto dorecs_unlocked; 2193 } 2194 2195 KASSERT(so == slp->ns_so); 2196 #define NFS_TEST_HEAVY 2197 #ifdef NFS_TEST_HEAVY 2198 /* 2199 * Define this to test for nfsds handling this under heavy load. 2200 * 2201 * XXX it isn't safe to call so_receive from so_upcall context. 2202 */ 2203 if (waitflag == M_DONTWAIT) { 2204 setflags |= SLP_NEEDQ; 2205 goto dorecs; 2206 } 2207 #endif 2208 simple_lock(&slp->ns_lock); 2209 slp->ns_flag &= ~SLP_NEEDQ; 2210 simple_unlock(&slp->ns_lock); 2211 if (so->so_type == SOCK_STREAM) { 2212 #ifndef NFS_TEST_HEAVY 2213 /* 2214 * If there are already records on the queue, defer soreceive() 2215 * to an nfsd so that there is feedback to the TCP layer that 2216 * the nfs servers are heavily loaded. 2217 */ 2218 if (slp->ns_rec && waitflag == M_DONTWAIT) { 2219 setflags |= SLP_NEEDQ; 2220 goto dorecs; 2221 } 2222 #endif 2223 2224 /* 2225 * Do soreceive(). 2226 */ 2227 auio.uio_resid = 1000000000; 2228 /* not need to setup uio_vmspace */ 2229 flags = MSG_DONTWAIT; 2230 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, &flags); 2231 if (error || mp == NULL) { 2232 if (error == EWOULDBLOCK) 2233 setflags |= SLP_NEEDQ; 2234 else 2235 setflags |= SLP_DISCONN; 2236 goto dorecs; 2237 } 2238 m = mp; 2239 m_claimm(m, &nfs_mowner); 2240 if (slp->ns_rawend) { 2241 slp->ns_rawend->m_next = m; 2242 slp->ns_cc += 1000000000 - auio.uio_resid; 2243 } else { 2244 slp->ns_raw = m; 2245 slp->ns_cc = 1000000000 - auio.uio_resid; 2246 } 2247 while (m->m_next) 2248 m = m->m_next; 2249 slp->ns_rawend = m; 2250 2251 /* 2252 * Now try and parse record(s) out of the raw stream data. 2253 */ 2254 error = nfsrv_getstream(slp, waitflag); 2255 if (error) { 2256 if (error == EPERM) 2257 setflags |= SLP_DISCONN; 2258 else 2259 setflags |= SLP_NEEDQ; 2260 } 2261 } else { 2262 do { 2263 auio.uio_resid = 1000000000; 2264 /* not need to setup uio_vmspace */ 2265 flags = MSG_DONTWAIT; 2266 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, 2267 &flags); 2268 if (mp) { 2269 if (nam) { 2270 m = nam; 2271 m->m_next = mp; 2272 } else 2273 m = mp; 2274 m_claimm(m, &nfs_mowner); 2275 if (slp->ns_recend) 2276 slp->ns_recend->m_nextpkt = m; 2277 else 2278 slp->ns_rec = m; 2279 slp->ns_recend = m; 2280 m->m_nextpkt = (struct mbuf *)0; 2281 } 2282 if (error) { 2283 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2284 && error != EWOULDBLOCK) { 2285 setflags |= SLP_DISCONN; 2286 goto dorecs; 2287 } 2288 } 2289 } while (mp); 2290 } 2291 dorecs: 2292 nfsdsock_unlock(slp); 2293 2294 dorecs_unlocked: 2295 /* 2296 * Now try and process the request records, non-blocking. 2297 */ 2298 if (setflags) { 2299 simple_lock(&slp->ns_lock); 2300 slp->ns_flag |= setflags; 2301 simple_unlock(&slp->ns_lock); 2302 } 2303 if (waitflag == M_DONTWAIT && 2304 (slp->ns_rec || (slp->ns_flag & (SLP_DISCONN | SLP_NEEDQ)) != 0)) { 2305 nfsrv_wakenfsd(slp); 2306 } 2307 } 2308 2309 int 2310 nfsdsock_lock(struct nfssvc_sock *slp, bool waitok) 2311 { 2312 2313 simple_lock(&slp->ns_lock); 2314 while ((slp->ns_flag & (SLP_BUSY|SLP_VALID)) == SLP_BUSY) { 2315 if (!waitok) { 2316 simple_unlock(&slp->ns_lock); 2317 return EWOULDBLOCK; 2318 } 2319 slp->ns_flag |= SLP_WANT; 2320 ltsleep(&slp->ns_flag, PSOCK, "nslock", 0, &slp->ns_lock); 2321 } 2322 if ((slp->ns_flag & SLP_VALID) == 0) { 2323 simple_unlock(&slp->ns_lock); 2324 return EINVAL; 2325 } 2326 slp->ns_flag |= SLP_BUSY; 2327 simple_unlock(&slp->ns_lock); 2328 2329 return 0; 2330 } 2331 2332 void 2333 nfsdsock_unlock(struct nfssvc_sock *slp) 2334 { 2335 2336 KASSERT((slp->ns_flag & SLP_BUSY) != 0); 2337 2338 simple_lock(&slp->ns_lock); 2339 if ((slp->ns_flag & SLP_WANT) != 0) { 2340 wakeup(&slp->ns_flag); 2341 } 2342 slp->ns_flag &= ~(SLP_BUSY|SLP_WANT); 2343 simple_unlock(&slp->ns_lock); 2344 } 2345 2346 int 2347 nfsdsock_drain(struct nfssvc_sock *slp) 2348 { 2349 int error = 0; 2350 2351 simple_lock(&slp->ns_lock); 2352 if ((slp->ns_flag & SLP_VALID) == 0) { 2353 error = EINVAL; 2354 goto done; 2355 } 2356 slp->ns_flag &= ~SLP_VALID; 2357 while ((slp->ns_flag & SLP_BUSY) != 0) { 2358 slp->ns_flag |= SLP_WANT; 2359 ltsleep(&slp->ns_flag, PSOCK, "nsdrain", 0, &slp->ns_lock); 2360 } 2361 done: 2362 simple_unlock(&slp->ns_lock); 2363 2364 return error; 2365 } 2366 2367 /* 2368 * Try and extract an RPC request from the mbuf data list received on a 2369 * stream socket. The "waitflag" argument indicates whether or not it 2370 * can sleep. 2371 */ 2372 int 2373 nfsrv_getstream(slp, waitflag) 2374 struct nfssvc_sock *slp; 2375 int waitflag; 2376 { 2377 struct mbuf *m, **mpp; 2378 struct mbuf *recm; 2379 u_int32_t recmark; 2380 int error = 0; 2381 2382 for (;;) { 2383 if (slp->ns_reclen == 0) { 2384 if (slp->ns_cc < NFSX_UNSIGNED) { 2385 break; 2386 } 2387 m = slp->ns_raw; 2388 m_copydata(m, 0, NFSX_UNSIGNED, (void *)&recmark); 2389 m_adj(m, NFSX_UNSIGNED); 2390 slp->ns_cc -= NFSX_UNSIGNED; 2391 recmark = ntohl(recmark); 2392 slp->ns_reclen = recmark & ~0x80000000; 2393 if (recmark & 0x80000000) 2394 slp->ns_flag |= SLP_LASTFRAG; 2395 else 2396 slp->ns_flag &= ~SLP_LASTFRAG; 2397 if (slp->ns_reclen > NFS_MAXPACKET) { 2398 error = EPERM; 2399 break; 2400 } 2401 } 2402 2403 /* 2404 * Now get the record part. 2405 * 2406 * Note that slp->ns_reclen may be 0. Linux sometimes 2407 * generates 0-length records. 2408 */ 2409 if (slp->ns_cc == slp->ns_reclen) { 2410 recm = slp->ns_raw; 2411 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2412 slp->ns_cc = slp->ns_reclen = 0; 2413 } else if (slp->ns_cc > slp->ns_reclen) { 2414 recm = slp->ns_raw; 2415 m = m_split(recm, slp->ns_reclen, waitflag); 2416 if (m == NULL) { 2417 error = EWOULDBLOCK; 2418 break; 2419 } 2420 m_claimm(recm, &nfs_mowner); 2421 slp->ns_raw = m; 2422 if (m->m_next == NULL) 2423 slp->ns_rawend = m; 2424 slp->ns_cc -= slp->ns_reclen; 2425 slp->ns_reclen = 0; 2426 } else { 2427 break; 2428 } 2429 2430 /* 2431 * Accumulate the fragments into a record. 2432 */ 2433 mpp = &slp->ns_frag; 2434 while (*mpp) 2435 mpp = &((*mpp)->m_next); 2436 *mpp = recm; 2437 if (slp->ns_flag & SLP_LASTFRAG) { 2438 if (slp->ns_recend) 2439 slp->ns_recend->m_nextpkt = slp->ns_frag; 2440 else 2441 slp->ns_rec = slp->ns_frag; 2442 slp->ns_recend = slp->ns_frag; 2443 slp->ns_frag = (struct mbuf *)0; 2444 } 2445 } 2446 2447 return error; 2448 } 2449 2450 /* 2451 * Parse an RPC header. 2452 */ 2453 int 2454 nfsrv_dorec(slp, nfsd, ndp) 2455 struct nfssvc_sock *slp; 2456 struct nfsd *nfsd; 2457 struct nfsrv_descript **ndp; 2458 { 2459 struct mbuf *m, *nam; 2460 struct nfsrv_descript *nd; 2461 int error; 2462 2463 *ndp = NULL; 2464 2465 if (nfsdsock_lock(slp, true)) { 2466 return ENOBUFS; 2467 } 2468 m = slp->ns_rec; 2469 if (m == NULL) { 2470 nfsdsock_unlock(slp); 2471 return ENOBUFS; 2472 } 2473 slp->ns_rec = m->m_nextpkt; 2474 if (slp->ns_rec) 2475 m->m_nextpkt = NULL; 2476 else 2477 slp->ns_recend = NULL; 2478 nfsdsock_unlock(slp); 2479 2480 if (m->m_type == MT_SONAME) { 2481 nam = m; 2482 m = m->m_next; 2483 nam->m_next = NULL; 2484 } else 2485 nam = NULL; 2486 nd = nfsdreq_alloc(); 2487 nd->nd_md = nd->nd_mrep = m; 2488 nd->nd_nam2 = nam; 2489 nd->nd_dpos = mtod(m, void *); 2490 error = nfs_getreq(nd, nfsd, true); 2491 if (error) { 2492 m_freem(nam); 2493 nfsdreq_free(nd); 2494 return (error); 2495 } 2496 *ndp = nd; 2497 nfsd->nfsd_nd = nd; 2498 return (0); 2499 } 2500 2501 /* 2502 * Search for a sleeping nfsd and wake it up. 2503 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2504 * running nfsds will go look for the work in the nfssvc_sock list. 2505 */ 2506 void 2507 nfsrv_wakenfsd(slp) 2508 struct nfssvc_sock *slp; 2509 { 2510 struct nfsd *nd; 2511 2512 if ((slp->ns_flag & SLP_VALID) == 0) 2513 return; 2514 simple_lock(&nfsd_slock); 2515 if (slp->ns_flag & SLP_DOREC) { 2516 simple_unlock(&nfsd_slock); 2517 return; 2518 } 2519 nd = SLIST_FIRST(&nfsd_idle_head); 2520 if (nd) { 2521 SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle); 2522 simple_unlock(&nfsd_slock); 2523 2524 if (nd->nfsd_slp) 2525 panic("nfsd wakeup"); 2526 slp->ns_sref++; 2527 nd->nfsd_slp = slp; 2528 wakeup(nd); 2529 return; 2530 } 2531 slp->ns_flag |= SLP_DOREC; 2532 nfsd_head_flag |= NFSD_CHECKSLP; 2533 TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending); 2534 simple_unlock(&nfsd_slock); 2535 } 2536 2537 int 2538 nfsdsock_sendreply(struct nfssvc_sock *slp, struct nfsrv_descript *nd) 2539 { 2540 int error; 2541 2542 if (nd->nd_mrep != NULL) { 2543 m_freem(nd->nd_mrep); 2544 nd->nd_mrep = NULL; 2545 } 2546 2547 simple_lock(&slp->ns_lock); 2548 if ((slp->ns_flag & SLP_SENDING) != 0) { 2549 SIMPLEQ_INSERT_TAIL(&slp->ns_sendq, nd, nd_sendq); 2550 simple_unlock(&slp->ns_lock); 2551 return 0; 2552 } 2553 KASSERT(SIMPLEQ_EMPTY(&slp->ns_sendq)); 2554 slp->ns_flag |= SLP_SENDING; 2555 simple_unlock(&slp->ns_lock); 2556 2557 again: 2558 error = nfs_send(slp->ns_so, nd->nd_nam2, nd->nd_mreq, NULL, curlwp); 2559 if (nd->nd_nam2) { 2560 m_free(nd->nd_nam2); 2561 } 2562 nfsdreq_free(nd); 2563 2564 simple_lock(&slp->ns_lock); 2565 KASSERT((slp->ns_flag & SLP_SENDING) != 0); 2566 nd = SIMPLEQ_FIRST(&slp->ns_sendq); 2567 if (nd != NULL) { 2568 SIMPLEQ_REMOVE_HEAD(&slp->ns_sendq, nd_sendq); 2569 simple_unlock(&slp->ns_lock); 2570 goto again; 2571 } 2572 slp->ns_flag &= ~SLP_SENDING; 2573 simple_unlock(&slp->ns_lock); 2574 2575 return error; 2576 } 2577 #endif /* NFSSERVER */ 2578 2579 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) 2580 static struct pool nfs_srvdesc_pool; 2581 2582 void 2583 nfsdreq_init(void) 2584 { 2585 2586 pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript), 2587 0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr, IPL_NONE); 2588 } 2589 2590 struct nfsrv_descript * 2591 nfsdreq_alloc(void) 2592 { 2593 struct nfsrv_descript *nd; 2594 2595 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK); 2596 nd->nd_cr = NULL; 2597 return nd; 2598 } 2599 2600 void 2601 nfsdreq_free(struct nfsrv_descript *nd) 2602 { 2603 kauth_cred_t cr; 2604 2605 cr = nd->nd_cr; 2606 if (cr != NULL) { 2607 kauth_cred_free(cr); 2608 } 2609 pool_put(&nfs_srvdesc_pool, nd); 2610 } 2611 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */ 2612
Indexes created Sun Sep 21 19:09:51 GMT 2025