1 /* $NetBSD: nfs_socket.c,v 1.173 2008/10/07 01:20:37 pooka 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 1.173 2008/10/07 01:20:37 pooka 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/evcnt.h> 53 #include <sys/callout.h> 54 #include <sys/proc.h> 55 #include <sys/mount.h> 56 #include <sys/kernel.h> 57 #include <sys/kmem.h> 58 #include <sys/mbuf.h> 59 #include <sys/vnode.h> 60 #include <sys/domain.h> 61 #include <sys/protosw.h> 62 #include <sys/socket.h> 63 #include <sys/socketvar.h> 64 #include <sys/syslog.h> 65 #include <sys/tprintf.h> 66 #include <sys/namei.h> 67 #include <sys/signal.h> 68 #include <sys/signalvar.h> 69 #include <sys/kauth.h> 70 71 #include <netinet/in.h> 72 #include <netinet/tcp.h> 73 74 #include <nfs/rpcv2.h> 75 #include <nfs/nfsproto.h> 76 #include <nfs/nfs.h> 77 #include <nfs/xdr_subs.h> 78 #include <nfs/nfsm_subs.h> 79 #include <nfs/nfsmount.h> 80 #include <nfs/nfsnode.h> 81 #include <nfs/nfsrtt.h> 82 #include <nfs/nfs_var.h> 83 84 #ifdef MBUFTRACE 85 struct mowner nfs_mowner = MOWNER_INIT("nfs",""); 86 #endif 87 88 /* 89 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 90 * Use the mean and mean deviation of rtt for the appropriate type of rpc 91 * for the frequent rpcs and a default for the others. 92 * The justification for doing "other" this way is that these rpcs 93 * happen so infrequently that timer est. would probably be stale. 94 * Also, since many of these rpcs are 95 * non-idempotent, a conservative timeout is desired. 96 * getattr, lookup - A+2D 97 * read, write - A+4D 98 * other - nm_timeo 99 */ 100 #define NFS_RTO(n, t) \ 101 ((t) == 0 ? (n)->nm_timeo : \ 102 ((t) < 3 ? \ 103 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 104 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 105 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 106 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 107 /* 108 * External data, mostly RPC constants in XDR form 109 */ 110 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, 111 rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, 112 rpc_auth_kerb; 113 extern u_int32_t nfs_prog; 114 extern const int nfsv3_procid[NFS_NPROCS]; 115 extern int nfs_ticks; 116 117 /* 118 * Defines which timer to use for the procnum. 119 * 0 - default 120 * 1 - getattr 121 * 2 - lookup 122 * 3 - read 123 * 4 - write 124 */ 125 static const int proct[NFS_NPROCS] = { 126 [NFSPROC_NULL] = 0, 127 [NFSPROC_GETATTR] = 1, 128 [NFSPROC_SETATTR] = 0, 129 [NFSPROC_LOOKUP] = 2, 130 [NFSPROC_ACCESS] = 1, 131 [NFSPROC_READLINK] = 3, 132 [NFSPROC_READ] = 3, 133 [NFSPROC_WRITE] = 4, 134 [NFSPROC_CREATE] = 0, 135 [NFSPROC_MKDIR] = 0, 136 [NFSPROC_SYMLINK] = 0, 137 [NFSPROC_MKNOD] = 0, 138 [NFSPROC_REMOVE] = 0, 139 [NFSPROC_RMDIR] = 0, 140 [NFSPROC_RENAME] = 0, 141 [NFSPROC_LINK] = 0, 142 [NFSPROC_READDIR] = 3, 143 [NFSPROC_READDIRPLUS] = 3, 144 [NFSPROC_FSSTAT] = 0, 145 [NFSPROC_FSINFO] = 0, 146 [NFSPROC_PATHCONF] = 0, 147 [NFSPROC_COMMIT] = 0, 148 [NFSPROC_NOOP] = 0, 149 }; 150 151 /* 152 * There is a congestion window for outstanding rpcs maintained per mount 153 * point. The cwnd size is adjusted in roughly the way that: 154 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 155 * SIGCOMM '88". ACM, August 1988. 156 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 157 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 158 * of rpcs is in progress. 159 * (The sent count and cwnd are scaled for integer arith.) 160 * Variants of "slow start" were tried and were found to be too much of a 161 * performance hit (ave. rtt 3 times larger), 162 * I suspect due to the large rtt that nfs rpcs have. 163 */ 164 #define NFS_CWNDSCALE 256 165 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 166 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 167 int nfsrtton = 0; 168 struct nfsrtt nfsrtt; 169 struct nfsreqhead nfs_reqq; 170 static callout_t nfs_timer_ch; 171 static struct evcnt nfs_timer_ev; 172 static struct evcnt nfs_timer_start_ev; 173 static struct evcnt nfs_timer_stop_ev; 174 175 #ifdef NFS 176 static int nfs_sndlock(struct nfsmount *, struct nfsreq *); 177 static void nfs_sndunlock(struct nfsmount *); 178 #endif 179 static int nfs_rcvlock(struct nfsmount *, struct nfsreq *); 180 static void nfs_rcvunlock(struct nfsmount *); 181 182 #if defined(NFSSERVER) 183 static void nfsrv_wakenfsd_locked(struct nfssvc_sock *); 184 #endif /* defined(NFSSERVER) */ 185 186 /* 187 * Initialize sockets and congestion for a new NFS connection. 188 * We do not free the sockaddr if error. 189 */ 190 int 191 nfs_connect(nmp, rep, l) 192 struct nfsmount *nmp; 193 struct nfsreq *rep; 194 struct lwp *l; 195 { 196 struct socket *so; 197 int error, rcvreserve, sndreserve; 198 struct sockaddr *saddr; 199 struct sockaddr_in *sin; 200 #ifdef INET6 201 struct sockaddr_in6 *sin6; 202 #endif 203 struct mbuf *m; 204 int val; 205 206 nmp->nm_so = (struct socket *)0; 207 saddr = mtod(nmp->nm_nam, struct sockaddr *); 208 error = socreate(saddr->sa_family, &nmp->nm_so, 209 nmp->nm_sotype, nmp->nm_soproto, l, NULL); 210 if (error) 211 goto bad; 212 so = nmp->nm_so; 213 #ifdef MBUFTRACE 214 so->so_mowner = &nfs_mowner; 215 so->so_rcv.sb_mowner = &nfs_mowner; 216 so->so_snd.sb_mowner = &nfs_mowner; 217 #endif 218 nmp->nm_soflags = so->so_proto->pr_flags; 219 220 /* 221 * Some servers require that the client port be a reserved port number. 222 */ 223 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 224 val = IP_PORTRANGE_LOW; 225 226 if ((error = so_setsockopt(NULL, so, IPPROTO_IP, IP_PORTRANGE, 227 &val, sizeof(val)))) 228 goto bad; 229 m = m_get(M_WAIT, MT_SONAME); 230 MCLAIM(m, so->so_mowner); 231 sin = mtod(m, struct sockaddr_in *); 232 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 233 sin->sin_family = AF_INET; 234 sin->sin_addr.s_addr = INADDR_ANY; 235 sin->sin_port = 0; 236 error = sobind(so, m, &lwp0); 237 m_freem(m); 238 if (error) 239 goto bad; 240 } 241 #ifdef INET6 242 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) { 243 val = IPV6_PORTRANGE_LOW; 244 245 if ((error = so_setsockopt(NULL, so, IPPROTO_IPV6, 246 IPV6_PORTRANGE, &val, sizeof(val)))) 247 goto bad; 248 m = m_get(M_WAIT, MT_SONAME); 249 MCLAIM(m, so->so_mowner); 250 sin6 = mtod(m, struct sockaddr_in6 *); 251 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6); 252 sin6->sin6_family = AF_INET6; 253 sin6->sin6_addr = in6addr_any; 254 sin6->sin6_port = 0; 255 error = sobind(so, m, &lwp0); 256 m_freem(m); 257 if (error) 258 goto bad; 259 } 260 #endif 261 262 /* 263 * Protocols that do not require connections may be optionally left 264 * unconnected for servers that reply from a port other than NFS_PORT. 265 */ 266 solock(so); 267 if (nmp->nm_flag & NFSMNT_NOCONN) { 268 if (nmp->nm_soflags & PR_CONNREQUIRED) { 269 sounlock(so); 270 error = ENOTCONN; 271 goto bad; 272 } 273 } else { 274 error = soconnect(so, nmp->nm_nam, l); 275 if (error) { 276 sounlock(so); 277 goto bad; 278 } 279 280 /* 281 * Wait for the connection to complete. Cribbed from the 282 * connect system call but with the wait timing out so 283 * that interruptible mounts don't hang here for a long time. 284 */ 285 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 286 (void)sowait(so, 2 * hz); 287 if ((so->so_state & SS_ISCONNECTING) && 288 so->so_error == 0 && rep && 289 (error = nfs_sigintr(nmp, rep, rep->r_lwp)) != 0){ 290 so->so_state &= ~SS_ISCONNECTING; 291 sounlock(so); 292 goto bad; 293 } 294 } 295 if (so->so_error) { 296 error = so->so_error; 297 so->so_error = 0; 298 sounlock(so); 299 goto bad; 300 } 301 } 302 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 303 so->so_rcv.sb_timeo = (5 * hz); 304 so->so_snd.sb_timeo = (5 * hz); 305 } else { 306 /* 307 * enable receive timeout to detect server crash and reconnect. 308 * otherwise, we can be stuck in soreceive forever. 309 */ 310 so->so_rcv.sb_timeo = (5 * hz); 311 so->so_snd.sb_timeo = 0; 312 } 313 if (nmp->nm_sotype == SOCK_DGRAM) { 314 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 315 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 316 NFS_MAXPKTHDR) * 2; 317 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 318 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 319 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 320 NFS_MAXPKTHDR) * 2; 321 } else { 322 sounlock(so); 323 if (nmp->nm_sotype != SOCK_STREAM) 324 panic("nfscon sotype"); 325 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 326 val = 1; 327 so_setsockopt(NULL, so, SOL_SOCKET, SO_KEEPALIVE, &val, 328 sizeof(val)); 329 } 330 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 331 val = 1; 332 so_setsockopt(NULL, so, IPPROTO_TCP, TCP_NODELAY, &val, 333 sizeof(val)); 334 } 335 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 336 sizeof (u_int32_t)) * 2; 337 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 338 sizeof (u_int32_t)) * 2; 339 solock(so); 340 } 341 error = soreserve(so, sndreserve, rcvreserve); 342 if (error) { 343 sounlock(so); 344 goto bad; 345 } 346 so->so_rcv.sb_flags |= SB_NOINTR; 347 so->so_snd.sb_flags |= SB_NOINTR; 348 sounlock(so); 349 350 /* Initialize other non-zero congestion variables */ 351 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 352 NFS_TIMEO << 3; 353 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 354 nmp->nm_sdrtt[3] = 0; 355 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 356 nmp->nm_sent = 0; 357 nmp->nm_timeouts = 0; 358 return (0); 359 360 bad: 361 nfs_disconnect(nmp); 362 return (error); 363 } 364 365 /* 366 * Reconnect routine: 367 * Called when a connection is broken on a reliable protocol. 368 * - clean up the old socket 369 * - nfs_connect() again 370 * - set R_MUSTRESEND for all outstanding requests on mount point 371 * If this fails the mount point is DEAD! 372 * nb: Must be called with the nfs_sndlock() set on the mount point. 373 */ 374 int 375 nfs_reconnect(struct nfsreq *rep) 376 { 377 struct nfsreq *rp; 378 struct nfsmount *nmp = rep->r_nmp; 379 int error; 380 381 nfs_disconnect(nmp); 382 while ((error = nfs_connect(nmp, rep, &lwp0)) != 0) { 383 if (error == EINTR || error == ERESTART) 384 return (EINTR); 385 kpause("nfscn2", false, hz, NULL); 386 } 387 388 /* 389 * Loop through outstanding request list and fix up all requests 390 * on old socket. 391 */ 392 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 393 if (rp->r_nmp == nmp) { 394 if ((rp->r_flags & R_MUSTRESEND) == 0) 395 rp->r_flags |= R_MUSTRESEND | R_REXMITTED; 396 rp->r_rexmit = 0; 397 } 398 } 399 return (0); 400 } 401 402 /* 403 * NFS disconnect. Clean up and unlink. 404 */ 405 void 406 nfs_disconnect(nmp) 407 struct nfsmount *nmp; 408 { 409 struct socket *so; 410 int drain = 0; 411 412 if (nmp->nm_so) { 413 so = nmp->nm_so; 414 nmp->nm_so = (struct socket *)0; 415 solock(so); 416 soshutdown(so, SHUT_RDWR); 417 sounlock(so); 418 drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0; 419 if (drain) { 420 /* 421 * soshutdown() above should wake up the current 422 * listener. 423 * Now wake up those waiting for the receive lock, and 424 * wait for them to go away unhappy, to prevent *nmp 425 * from evaporating while they're sleeping. 426 */ 427 mutex_enter(&nmp->nm_lock); 428 while (nmp->nm_waiters > 0) { 429 cv_broadcast(&nmp->nm_rcvcv); 430 cv_broadcast(&nmp->nm_sndcv); 431 cv_wait(&nmp->nm_disconcv, &nmp->nm_lock); 432 } 433 mutex_exit(&nmp->nm_lock); 434 } 435 soclose(so); 436 } 437 #ifdef DIAGNOSTIC 438 if (drain && (nmp->nm_waiters > 0)) 439 panic("nfs_disconnect: waiters left after drain?"); 440 #endif 441 } 442 443 void 444 nfs_safedisconnect(nmp) 445 struct nfsmount *nmp; 446 { 447 struct nfsreq dummyreq; 448 449 memset(&dummyreq, 0, sizeof(dummyreq)); 450 dummyreq.r_nmp = nmp; 451 nfs_rcvlock(nmp, &dummyreq); /* XXX ignored error return */ 452 nfs_disconnect(nmp); 453 nfs_rcvunlock(nmp); 454 } 455 456 /* 457 * This is the nfs send routine. For connection based socket types, it 458 * must be called with an nfs_sndlock() on the socket. 459 * "rep == NULL" indicates that it has been called from a server. 460 * For the client side: 461 * - return EINTR if the RPC is terminated, 0 otherwise 462 * - set R_MUSTRESEND if the send fails for any reason 463 * - do any cleanup required by recoverable socket errors (? ? ?) 464 * For the server side: 465 * - return EINTR or ERESTART if interrupted by a signal 466 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 467 * - do any cleanup required by recoverable socket errors (? ? ?) 468 */ 469 int 470 nfs_send(so, nam, top, rep, l) 471 struct socket *so; 472 struct mbuf *nam; 473 struct mbuf *top; 474 struct nfsreq *rep; 475 struct lwp *l; 476 { 477 struct mbuf *sendnam; 478 int error, soflags, flags; 479 480 /* XXX nfs_doio()/nfs_request() calls with rep->r_lwp == NULL */ 481 if (l == NULL && rep->r_lwp == NULL) 482 l = curlwp; 483 484 if (rep) { 485 if (rep->r_flags & R_SOFTTERM) { 486 m_freem(top); 487 return (EINTR); 488 } 489 if ((so = rep->r_nmp->nm_so) == NULL) { 490 rep->r_flags |= R_MUSTRESEND; 491 m_freem(top); 492 return (0); 493 } 494 rep->r_flags &= ~R_MUSTRESEND; 495 soflags = rep->r_nmp->nm_soflags; 496 } else 497 soflags = so->so_proto->pr_flags; 498 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 499 sendnam = (struct mbuf *)0; 500 else 501 sendnam = nam; 502 if (so->so_type == SOCK_SEQPACKET) 503 flags = MSG_EOR; 504 else 505 flags = 0; 506 507 error = (*so->so_send)(so, sendnam, NULL, top, NULL, flags, l); 508 if (error) { 509 if (rep) { 510 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 511 /* 512 * We're too fast for the network/driver, 513 * and UDP isn't flowcontrolled. 514 * We need to resend. This is not fatal, 515 * just try again. 516 * 517 * Could be smarter here by doing some sort 518 * of a backoff, but this is rare. 519 */ 520 rep->r_flags |= R_MUSTRESEND; 521 } else { 522 if (error != EPIPE) 523 log(LOG_INFO, 524 "nfs send error %d for %s\n", 525 error, 526 rep->r_nmp->nm_mountp-> 527 mnt_stat.f_mntfromname); 528 /* 529 * Deal with errors for the client side. 530 */ 531 if (rep->r_flags & R_SOFTTERM) 532 error = EINTR; 533 else 534 rep->r_flags |= R_MUSTRESEND; 535 } 536 } else { 537 /* 538 * See above. This error can happen under normal 539 * circumstances and the log is too noisy. 540 * The error will still show up in nfsstat. 541 */ 542 if (error != ENOBUFS || so->so_type != SOCK_DGRAM) 543 log(LOG_INFO, "nfsd send error %d\n", error); 544 } 545 546 /* 547 * Handle any recoverable (soft) socket errors here. (? ? ?) 548 */ 549 if (error != EINTR && error != ERESTART && 550 error != EWOULDBLOCK && error != EPIPE) 551 error = 0; 552 } 553 return (error); 554 } 555 556 #ifdef NFS 557 /* 558 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 559 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 560 * Mark and consolidate the data into a new mbuf list. 561 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 562 * small mbufs. 563 * For SOCK_STREAM we must be very careful to read an entire record once 564 * we have read any of it, even if the system call has been interrupted. 565 */ 566 static int 567 nfs_receive(struct nfsreq *rep, struct mbuf **aname, struct mbuf **mp, 568 struct lwp *l) 569 { 570 struct socket *so; 571 struct uio auio; 572 struct iovec aio; 573 struct mbuf *m; 574 struct mbuf *control; 575 u_int32_t len; 576 struct mbuf **getnam; 577 int error, sotype, rcvflg; 578 579 /* 580 * Set up arguments for soreceive() 581 */ 582 *mp = (struct mbuf *)0; 583 *aname = (struct mbuf *)0; 584 sotype = rep->r_nmp->nm_sotype; 585 586 /* 587 * For reliable protocols, lock against other senders/receivers 588 * in case a reconnect is necessary. 589 * For SOCK_STREAM, first get the Record Mark to find out how much 590 * more there is to get. 591 * We must lock the socket against other receivers 592 * until we have an entire rpc request/reply. 593 */ 594 if (sotype != SOCK_DGRAM) { 595 error = nfs_sndlock(rep->r_nmp, rep); 596 if (error) 597 return (error); 598 tryagain: 599 /* 600 * Check for fatal errors and resending request. 601 */ 602 /* 603 * Ugh: If a reconnect attempt just happened, nm_so 604 * would have changed. NULL indicates a failed 605 * attempt that has essentially shut down this 606 * mount point. 607 */ 608 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 609 nfs_sndunlock(rep->r_nmp); 610 return (EINTR); 611 } 612 so = rep->r_nmp->nm_so; 613 if (!so) { 614 error = nfs_reconnect(rep); 615 if (error) { 616 nfs_sndunlock(rep->r_nmp); 617 return (error); 618 } 619 goto tryagain; 620 } 621 while (rep->r_flags & R_MUSTRESEND) { 622 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 623 nfsstats.rpcretries++; 624 rep->r_rtt = 0; 625 rep->r_flags &= ~R_TIMING; 626 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, l); 627 if (error) { 628 if (error == EINTR || error == ERESTART || 629 (error = nfs_reconnect(rep)) != 0) { 630 nfs_sndunlock(rep->r_nmp); 631 return (error); 632 } 633 goto tryagain; 634 } 635 } 636 nfs_sndunlock(rep->r_nmp); 637 if (sotype == SOCK_STREAM) { 638 aio.iov_base = (void *) &len; 639 aio.iov_len = sizeof(u_int32_t); 640 auio.uio_iov = &aio; 641 auio.uio_iovcnt = 1; 642 auio.uio_rw = UIO_READ; 643 auio.uio_offset = 0; 644 auio.uio_resid = sizeof(u_int32_t); 645 UIO_SETUP_SYSSPACE(&auio); 646 do { 647 rcvflg = MSG_WAITALL; 648 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 649 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 650 if (error == EWOULDBLOCK && rep) { 651 if (rep->r_flags & R_SOFTTERM) 652 return (EINTR); 653 /* 654 * if it seems that the server died after it 655 * received our request, set EPIPE so that 656 * we'll reconnect and retransmit requests. 657 */ 658 if (rep->r_rexmit >= rep->r_nmp->nm_retry) { 659 nfsstats.rpctimeouts++; 660 error = EPIPE; 661 } 662 } 663 } while (error == EWOULDBLOCK); 664 if (!error && auio.uio_resid > 0) { 665 /* 666 * Don't log a 0 byte receive; it means 667 * that the socket has been closed, and 668 * can happen during normal operation 669 * (forcible unmount or Solaris server). 670 */ 671 if (auio.uio_resid != sizeof (u_int32_t)) 672 log(LOG_INFO, 673 "short receive (%lu/%lu) from nfs server %s\n", 674 (u_long)sizeof(u_int32_t) - auio.uio_resid, 675 (u_long)sizeof(u_int32_t), 676 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 677 error = EPIPE; 678 } 679 if (error) 680 goto errout; 681 len = ntohl(len) & ~0x80000000; 682 /* 683 * This is SERIOUS! We are out of sync with the sender 684 * and forcing a disconnect/reconnect is all I can do. 685 */ 686 if (len > NFS_MAXPACKET) { 687 log(LOG_ERR, "%s (%d) from nfs server %s\n", 688 "impossible packet length", 689 len, 690 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 691 error = EFBIG; 692 goto errout; 693 } 694 auio.uio_resid = len; 695 do { 696 rcvflg = MSG_WAITALL; 697 error = (*so->so_receive)(so, (struct mbuf **)0, 698 &auio, mp, (struct mbuf **)0, &rcvflg); 699 } while (error == EWOULDBLOCK || error == EINTR || 700 error == ERESTART); 701 if (!error && auio.uio_resid > 0) { 702 if (len != auio.uio_resid) 703 log(LOG_INFO, 704 "short receive (%lu/%d) from nfs server %s\n", 705 (u_long)len - auio.uio_resid, len, 706 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 707 error = EPIPE; 708 } 709 } else { 710 /* 711 * NB: Since uio_resid is big, MSG_WAITALL is ignored 712 * and soreceive() will return when it has either a 713 * control msg or a data msg. 714 * We have no use for control msg., but must grab them 715 * and then throw them away so we know what is going 716 * on. 717 */ 718 auio.uio_resid = len = 100000000; /* Anything Big */ 719 /* not need to setup uio_vmspace */ 720 do { 721 rcvflg = 0; 722 error = (*so->so_receive)(so, (struct mbuf **)0, 723 &auio, mp, &control, &rcvflg); 724 if (control) 725 m_freem(control); 726 if (error == EWOULDBLOCK && rep) { 727 if (rep->r_flags & R_SOFTTERM) 728 return (EINTR); 729 } 730 } while (error == EWOULDBLOCK || 731 (!error && *mp == NULL && control)); 732 if ((rcvflg & MSG_EOR) == 0) 733 printf("Egad!!\n"); 734 if (!error && *mp == NULL) 735 error = EPIPE; 736 len -= auio.uio_resid; 737 } 738 errout: 739 if (error && error != EINTR && error != ERESTART) { 740 m_freem(*mp); 741 *mp = (struct mbuf *)0; 742 if (error != EPIPE) 743 log(LOG_INFO, 744 "receive error %d from nfs server %s\n", 745 error, 746 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 747 error = nfs_sndlock(rep->r_nmp, rep); 748 if (!error) 749 error = nfs_reconnect(rep); 750 if (!error) 751 goto tryagain; 752 else 753 nfs_sndunlock(rep->r_nmp); 754 } 755 } else { 756 if ((so = rep->r_nmp->nm_so) == NULL) 757 return (EACCES); 758 if (so->so_state & SS_ISCONNECTED) 759 getnam = (struct mbuf **)0; 760 else 761 getnam = aname; 762 auio.uio_resid = len = 1000000; 763 /* not need to setup uio_vmspace */ 764 do { 765 rcvflg = 0; 766 error = (*so->so_receive)(so, getnam, &auio, mp, 767 (struct mbuf **)0, &rcvflg); 768 if (error == EWOULDBLOCK && 769 (rep->r_flags & R_SOFTTERM)) 770 return (EINTR); 771 } while (error == EWOULDBLOCK); 772 len -= auio.uio_resid; 773 if (!error && *mp == NULL) 774 error = EPIPE; 775 } 776 if (error) { 777 m_freem(*mp); 778 *mp = (struct mbuf *)0; 779 } 780 return (error); 781 } 782 783 /* 784 * Implement receipt of reply on a socket. 785 * We must search through the list of received datagrams matching them 786 * with outstanding requests using the xid, until ours is found. 787 */ 788 /* ARGSUSED */ 789 static int 790 nfs_reply(struct nfsreq *myrep, struct lwp *lwp) 791 { 792 struct nfsreq *rep; 793 struct nfsmount *nmp = myrep->r_nmp; 794 int32_t t1; 795 struct mbuf *mrep, *nam, *md; 796 u_int32_t rxid, *tl; 797 char *dpos, *cp2; 798 int error; 799 800 /* 801 * Loop around until we get our own reply 802 */ 803 for (;;) { 804 /* 805 * Lock against other receivers so that I don't get stuck in 806 * sbwait() after someone else has received my reply for me. 807 * Also necessary for connection based protocols to avoid 808 * race conditions during a reconnect. 809 */ 810 error = nfs_rcvlock(nmp, myrep); 811 if (error == EALREADY) 812 return (0); 813 if (error) 814 return (error); 815 /* 816 * Get the next Rpc reply off the socket 817 */ 818 819 mutex_enter(&nmp->nm_lock); 820 nmp->nm_waiters++; 821 mutex_exit(&nmp->nm_lock); 822 823 error = nfs_receive(myrep, &nam, &mrep, lwp); 824 825 mutex_enter(&nmp->nm_lock); 826 nmp->nm_waiters--; 827 cv_signal(&nmp->nm_disconcv); 828 mutex_exit(&nmp->nm_lock); 829 830 if (error) { 831 nfs_rcvunlock(nmp); 832 833 if (nmp->nm_iflag & NFSMNT_DISMNT) { 834 /* 835 * Oops, we're going away now.. 836 */ 837 return error; 838 } 839 /* 840 * Ignore routing errors on connectionless protocols? ? 841 */ 842 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 843 nmp->nm_so->so_error = 0; 844 #ifdef DEBUG 845 printf("nfs_reply: ignoring error %d\n", error); 846 #endif 847 continue; 848 } 849 return (error); 850 } 851 if (nam) 852 m_freem(nam); 853 854 /* 855 * Get the xid and check that it is an rpc reply 856 */ 857 md = mrep; 858 dpos = mtod(md, void *); 859 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 860 rxid = *tl++; 861 if (*tl != rpc_reply) { 862 nfsstats.rpcinvalid++; 863 m_freem(mrep); 864 nfsmout: 865 nfs_rcvunlock(nmp); 866 continue; 867 } 868 869 /* 870 * Loop through the request list to match up the reply 871 * Iff no match, just drop the datagram 872 */ 873 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 874 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 875 /* Found it.. */ 876 rep->r_mrep = mrep; 877 rep->r_md = md; 878 rep->r_dpos = dpos; 879 if (nfsrtton) { 880 struct rttl *rt; 881 882 rt = &nfsrtt.rttl[nfsrtt.pos]; 883 rt->proc = rep->r_procnum; 884 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 885 rt->sent = nmp->nm_sent; 886 rt->cwnd = nmp->nm_cwnd; 887 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 888 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 889 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx; 890 getmicrotime(&rt->tstamp); 891 if (rep->r_flags & R_TIMING) 892 rt->rtt = rep->r_rtt; 893 else 894 rt->rtt = 1000000; 895 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 896 } 897 /* 898 * Update congestion window. 899 * Do the additive increase of 900 * one rpc/rtt. 901 */ 902 if (nmp->nm_cwnd <= nmp->nm_sent) { 903 nmp->nm_cwnd += 904 (NFS_CWNDSCALE * NFS_CWNDSCALE + 905 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 906 if (nmp->nm_cwnd > NFS_MAXCWND) 907 nmp->nm_cwnd = NFS_MAXCWND; 908 } 909 rep->r_flags &= ~R_SENT; 910 nmp->nm_sent -= NFS_CWNDSCALE; 911 /* 912 * Update rtt using a gain of 0.125 on the mean 913 * and a gain of 0.25 on the deviation. 914 */ 915 if (rep->r_flags & R_TIMING) { 916 /* 917 * Since the timer resolution of 918 * NFS_HZ is so course, it can often 919 * result in r_rtt == 0. Since 920 * r_rtt == N means that the actual 921 * rtt is between N+dt and N+2-dt ticks, 922 * add 1. 923 */ 924 t1 = rep->r_rtt + 1; 925 t1 -= (NFS_SRTT(rep) >> 3); 926 NFS_SRTT(rep) += t1; 927 if (t1 < 0) 928 t1 = -t1; 929 t1 -= (NFS_SDRTT(rep) >> 2); 930 NFS_SDRTT(rep) += t1; 931 } 932 nmp->nm_timeouts = 0; 933 break; 934 } 935 } 936 nfs_rcvunlock(nmp); 937 /* 938 * If not matched to a request, drop it. 939 * If it's mine, get out. 940 */ 941 if (rep == 0) { 942 nfsstats.rpcunexpected++; 943 m_freem(mrep); 944 } else if (rep == myrep) { 945 if (rep->r_mrep == NULL) 946 panic("nfsreply nil"); 947 return (0); 948 } 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 rep = kmem_alloc(sizeof(*rep), KM_SLEEP); 1003 rep->r_nmp = nmp; 1004 KASSERT(lwp == NULL || lwp == curlwp); 1005 rep->r_lwp = lwp; 1006 rep->r_procnum = procnum; 1007 i = 0; 1008 m = mrest; 1009 while (m) { 1010 i += m->m_len; 1011 m = m->m_next; 1012 } 1013 mrest_len = i; 1014 1015 /* 1016 * Get the RPC header with authorization. 1017 */ 1018 kerbauth: 1019 verf_str = auth_str = (char *)0; 1020 if (nmp->nm_flag & NFSMNT_KERB) { 1021 verf_str = nickv; 1022 verf_len = sizeof (nickv); 1023 auth_type = RPCAUTH_KERB4; 1024 memset((void *)key, 0, sizeof (key)); 1025 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 1026 &auth_len, verf_str, verf_len)) { 1027 error = nfs_getauth(nmp, rep, cred, &auth_str, 1028 &auth_len, verf_str, &verf_len, key); 1029 if (error) { 1030 kmem_free(rep, sizeof(*rep)); 1031 m_freem(mrest); 1032 KASSERT(kauth_cred_getrefcnt(acred) == 1); 1033 kauth_cred_free(acred); 1034 return (error); 1035 } 1036 } 1037 retry_cred = false; 1038 } else { 1039 /* AUTH_UNIX */ 1040 uid_t uid; 1041 gid_t gid; 1042 1043 /* 1044 * on the most unix filesystems, permission checks are 1045 * done when the file is open(2)'ed. 1046 * ie. once a file is successfully open'ed, 1047 * following i/o operations never fail with EACCES. 1048 * we try to follow the semantics as far as possible. 1049 * 1050 * note that we expect that the nfs server always grant 1051 * accesses by the file's owner. 1052 */ 1053 origcred = cred; 1054 switch (procnum) { 1055 case NFSPROC_READ: 1056 case NFSPROC_WRITE: 1057 case NFSPROC_COMMIT: 1058 uid = np->n_vattr->va_uid; 1059 gid = np->n_vattr->va_gid; 1060 if (kauth_cred_geteuid(cred) == uid && 1061 kauth_cred_getegid(cred) == gid) { 1062 retry_cred = false; 1063 break; 1064 } 1065 if (use_opencred) 1066 break; 1067 kauth_cred_setuid(acred, uid); 1068 kauth_cred_seteuid(acred, uid); 1069 kauth_cred_setsvuid(acred, uid); 1070 kauth_cred_setgid(acred, gid); 1071 kauth_cred_setegid(acred, gid); 1072 kauth_cred_setsvgid(acred, gid); 1073 cred = acred; 1074 break; 1075 default: 1076 retry_cred = false; 1077 break; 1078 } 1079 /* 1080 * backup mbuf chain if we can need it later to retry. 1081 * 1082 * XXX maybe we can keep a direct reference to 1083 * mrest without doing m_copym, but it's ...ugly. 1084 */ 1085 if (retry_cred) 1086 mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT); 1087 auth_type = RPCAUTH_UNIX; 1088 /* XXX elad - ngroups */ 1089 auth_len = (((kauth_cred_ngroups(cred) > nmp->nm_numgrps) ? 1090 nmp->nm_numgrps : kauth_cred_ngroups(cred)) << 2) + 1091 5 * NFSX_UNSIGNED; 1092 } 1093 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 1094 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 1095 if (auth_str) 1096 free(auth_str, M_TEMP); 1097 1098 /* 1099 * For stream protocols, insert a Sun RPC Record Mark. 1100 */ 1101 if (nmp->nm_sotype == SOCK_STREAM) { 1102 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 1103 *mtod(m, u_int32_t *) = htonl(0x80000000 | 1104 (m->m_pkthdr.len - NFSX_UNSIGNED)); 1105 } 1106 rep->r_mreq = m; 1107 rep->r_xid = xid; 1108 tryagain: 1109 if (nmp->nm_flag & NFSMNT_SOFT) 1110 rep->r_retry = nmp->nm_retry; 1111 else 1112 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1113 rep->r_rtt = rep->r_rexmit = 0; 1114 if (proct[procnum] > 0) 1115 rep->r_flags = R_TIMING; 1116 else 1117 rep->r_flags = 0; 1118 rep->r_mrep = NULL; 1119 1120 /* 1121 * Do the client side RPC. 1122 */ 1123 nfsstats.rpcrequests++; 1124 /* 1125 * Chain request into list of outstanding requests. Be sure 1126 * to put it LAST so timer finds oldest requests first. 1127 */ 1128 s = splsoftnet(); 1129 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1130 nfs_timer_start(); 1131 1132 /* 1133 * If backing off another request or avoiding congestion, don't 1134 * send this one now but let timer do it. If not timing a request, 1135 * do it now. 1136 */ 1137 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1138 (nmp->nm_flag & NFSMNT_DUMBTIMR) || nmp->nm_sent < nmp->nm_cwnd)) { 1139 splx(s); 1140 if (nmp->nm_soflags & PR_CONNREQUIRED) 1141 error = nfs_sndlock(nmp, rep); 1142 if (!error) { 1143 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1144 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp); 1145 if (nmp->nm_soflags & PR_CONNREQUIRED) 1146 nfs_sndunlock(nmp); 1147 } 1148 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1149 nmp->nm_sent += NFS_CWNDSCALE; 1150 rep->r_flags |= R_SENT; 1151 } 1152 } else { 1153 splx(s); 1154 rep->r_rtt = -1; 1155 } 1156 1157 /* 1158 * Wait for the reply from our send or the timer's. 1159 */ 1160 if (!error || error == EPIPE) 1161 error = nfs_reply(rep, lwp); 1162 1163 /* 1164 * RPC done, unlink the request. 1165 */ 1166 s = splsoftnet(); 1167 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1168 splx(s); 1169 1170 /* 1171 * Decrement the outstanding request count. 1172 */ 1173 if (rep->r_flags & R_SENT) { 1174 rep->r_flags &= ~R_SENT; /* paranoia */ 1175 nmp->nm_sent -= NFS_CWNDSCALE; 1176 } 1177 1178 if (rexmitp != NULL) { 1179 int rexmit; 1180 1181 if (nmp->nm_sotype != SOCK_DGRAM) 1182 rexmit = (rep->r_flags & R_REXMITTED) != 0; 1183 else 1184 rexmit = rep->r_rexmit; 1185 *rexmitp = rexmit; 1186 } 1187 1188 /* 1189 * If there was a successful reply and a tprintf msg. 1190 * tprintf a response. 1191 */ 1192 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1193 nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1194 "is alive again"); 1195 mrep = rep->r_mrep; 1196 md = rep->r_md; 1197 dpos = rep->r_dpos; 1198 if (error) 1199 goto nfsmout; 1200 1201 /* 1202 * break down the rpc header and check if ok 1203 */ 1204 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1205 if (*tl++ == rpc_msgdenied) { 1206 if (*tl == rpc_mismatch) 1207 error = EOPNOTSUPP; 1208 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1209 if (!failed_auth) { 1210 failed_auth++; 1211 mheadend->m_next = (struct mbuf *)0; 1212 m_freem(mrep); 1213 m_freem(rep->r_mreq); 1214 goto kerbauth; 1215 } else 1216 error = EAUTH; 1217 } else 1218 error = EACCES; 1219 m_freem(mrep); 1220 goto nfsmout; 1221 } 1222 1223 /* 1224 * Grab any Kerberos verifier, otherwise just throw it away. 1225 */ 1226 verf_type = fxdr_unsigned(int, *tl++); 1227 i = fxdr_unsigned(int32_t, *tl); 1228 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1229 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1230 if (error) 1231 goto nfsmout; 1232 } else if (i > 0) 1233 nfsm_adv(nfsm_rndup(i)); 1234 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1235 /* 0 == ok */ 1236 if (*tl == 0) { 1237 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1238 if (*tl != 0) { 1239 error = fxdr_unsigned(int, *tl); 1240 switch (error) { 1241 case NFSERR_PERM: 1242 error = EPERM; 1243 break; 1244 1245 case NFSERR_NOENT: 1246 error = ENOENT; 1247 break; 1248 1249 case NFSERR_IO: 1250 error = EIO; 1251 break; 1252 1253 case NFSERR_NXIO: 1254 error = ENXIO; 1255 break; 1256 1257 case NFSERR_ACCES: 1258 error = EACCES; 1259 if (!retry_cred) 1260 break; 1261 m_freem(mrep); 1262 m_freem(rep->r_mreq); 1263 kmem_free(rep, sizeof(*rep)); 1264 use_opencred = !use_opencred; 1265 if (mrest_backup == NULL) { 1266 /* m_copym failure */ 1267 KASSERT( 1268 kauth_cred_getrefcnt(acred) == 1); 1269 kauth_cred_free(acred); 1270 return ENOMEM; 1271 } 1272 mrest = mrest_backup; 1273 mrest_backup = NULL; 1274 cred = origcred; 1275 error = 0; 1276 retry_cred = false; 1277 goto tryagain_cred; 1278 1279 case NFSERR_EXIST: 1280 error = EEXIST; 1281 break; 1282 1283 case NFSERR_XDEV: 1284 error = EXDEV; 1285 break; 1286 1287 case NFSERR_NODEV: 1288 error = ENODEV; 1289 break; 1290 1291 case NFSERR_NOTDIR: 1292 error = ENOTDIR; 1293 break; 1294 1295 case NFSERR_ISDIR: 1296 error = EISDIR; 1297 break; 1298 1299 case NFSERR_INVAL: 1300 error = EINVAL; 1301 break; 1302 1303 case NFSERR_FBIG: 1304 error = EFBIG; 1305 break; 1306 1307 case NFSERR_NOSPC: 1308 error = ENOSPC; 1309 break; 1310 1311 case NFSERR_ROFS: 1312 error = EROFS; 1313 break; 1314 1315 case NFSERR_MLINK: 1316 error = EMLINK; 1317 break; 1318 1319 case NFSERR_TIMEDOUT: 1320 error = ETIMEDOUT; 1321 break; 1322 1323 case NFSERR_NAMETOL: 1324 error = ENAMETOOLONG; 1325 break; 1326 1327 case NFSERR_NOTEMPTY: 1328 error = ENOTEMPTY; 1329 break; 1330 1331 case NFSERR_DQUOT: 1332 error = EDQUOT; 1333 break; 1334 1335 case NFSERR_STALE: 1336 /* 1337 * If the File Handle was stale, invalidate the 1338 * lookup cache, just in case. 1339 */ 1340 error = ESTALE; 1341 cache_purge(NFSTOV(np)); 1342 break; 1343 1344 case NFSERR_REMOTE: 1345 error = EREMOTE; 1346 break; 1347 1348 case NFSERR_WFLUSH: 1349 case NFSERR_BADHANDLE: 1350 case NFSERR_NOT_SYNC: 1351 case NFSERR_BAD_COOKIE: 1352 error = EINVAL; 1353 break; 1354 1355 case NFSERR_NOTSUPP: 1356 error = ENOTSUP; 1357 break; 1358 1359 case NFSERR_TOOSMALL: 1360 case NFSERR_SERVERFAULT: 1361 case NFSERR_BADTYPE: 1362 error = EINVAL; 1363 break; 1364 1365 case NFSERR_TRYLATER: 1366 if ((nmp->nm_flag & NFSMNT_NFSV3) == 0) 1367 break; 1368 m_freem(mrep); 1369 error = 0; 1370 waituntil = time_second + trylater_delay; 1371 while (time_second < waituntil) { 1372 kpause("nfstrylater", false, hz, NULL); 1373 } 1374 trylater_delay *= NFS_TRYLATERDELMUL; 1375 if (trylater_delay > NFS_TRYLATERDELMAX) 1376 trylater_delay = NFS_TRYLATERDELMAX; 1377 /* 1378 * RFC1813: 1379 * The client should wait and then try 1380 * the request with a new RPC transaction ID. 1381 */ 1382 nfs_renewxid(rep); 1383 goto tryagain; 1384 1385 default: 1386 #ifdef DIAGNOSTIC 1387 printf("Invalid rpc error code %d\n", error); 1388 #endif 1389 error = EINVAL; 1390 break; 1391 } 1392 1393 if (nmp->nm_flag & NFSMNT_NFSV3) { 1394 *mrp = mrep; 1395 *mdp = md; 1396 *dposp = dpos; 1397 error |= NFSERR_RETERR; 1398 } else 1399 m_freem(mrep); 1400 goto nfsmout; 1401 } 1402 1403 /* 1404 * note which credential worked to minimize number of retries. 1405 */ 1406 if (use_opencred) 1407 np->n_flag |= NUSEOPENCRED; 1408 else 1409 np->n_flag &= ~NUSEOPENCRED; 1410 1411 *mrp = mrep; 1412 *mdp = md; 1413 *dposp = dpos; 1414 1415 KASSERT(error == 0); 1416 goto nfsmout; 1417 } 1418 m_freem(mrep); 1419 error = EPROTONOSUPPORT; 1420 nfsmout: 1421 KASSERT(kauth_cred_getrefcnt(acred) == 1); 1422 kauth_cred_free(acred); 1423 m_freem(rep->r_mreq); 1424 kmem_free(rep, sizeof(*rep)); 1425 m_freem(mrest_backup); 1426 return (error); 1427 } 1428 #endif /* NFS */ 1429 1430 /* 1431 * Generate the rpc reply header 1432 * siz arg. is used to decide if adding a cluster is worthwhile 1433 */ 1434 int 1435 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1436 int siz; 1437 struct nfsrv_descript *nd; 1438 struct nfssvc_sock *slp; 1439 int err; 1440 int cache; 1441 u_quad_t *frev; 1442 struct mbuf **mrq; 1443 struct mbuf **mbp; 1444 char **bposp; 1445 { 1446 u_int32_t *tl; 1447 struct mbuf *mreq; 1448 char *bpos; 1449 struct mbuf *mb; 1450 1451 mreq = m_gethdr(M_WAIT, MT_DATA); 1452 MCLAIM(mreq, &nfs_mowner); 1453 mb = mreq; 1454 /* 1455 * If this is a big reply, use a cluster else 1456 * try and leave leading space for the lower level headers. 1457 */ 1458 siz += RPC_REPLYSIZ; 1459 if (siz >= max_datalen) { 1460 m_clget(mreq, M_WAIT); 1461 } else 1462 mreq->m_data += max_hdr; 1463 tl = mtod(mreq, u_int32_t *); 1464 mreq->m_len = 6 * NFSX_UNSIGNED; 1465 bpos = ((char *)tl) + mreq->m_len; 1466 *tl++ = txdr_unsigned(nd->nd_retxid); 1467 *tl++ = rpc_reply; 1468 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1469 *tl++ = rpc_msgdenied; 1470 if (err & NFSERR_AUTHERR) { 1471 *tl++ = rpc_autherr; 1472 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1473 mreq->m_len -= NFSX_UNSIGNED; 1474 bpos -= NFSX_UNSIGNED; 1475 } else { 1476 *tl++ = rpc_mismatch; 1477 *tl++ = txdr_unsigned(RPC_VER2); 1478 *tl = txdr_unsigned(RPC_VER2); 1479 } 1480 } else { 1481 *tl++ = rpc_msgaccepted; 1482 1483 /* 1484 * For Kerberos authentication, we must send the nickname 1485 * verifier back, otherwise just RPCAUTH_NULL. 1486 */ 1487 if (nd->nd_flag & ND_KERBFULL) { 1488 struct nfsuid *nuidp; 1489 struct timeval ktvin, ktvout; 1490 1491 memset(&ktvout, 0, sizeof ktvout); /* XXX gcc */ 1492 1493 LIST_FOREACH(nuidp, 1494 NUIDHASH(slp, kauth_cred_geteuid(nd->nd_cr)), 1495 nu_hash) { 1496 if (kauth_cred_geteuid(nuidp->nu_cr) == 1497 kauth_cred_geteuid(nd->nd_cr) && 1498 (!nd->nd_nam2 || netaddr_match( 1499 NU_NETFAM(nuidp), &nuidp->nu_haddr, 1500 nd->nd_nam2))) 1501 break; 1502 } 1503 if (nuidp) { 1504 ktvin.tv_sec = 1505 txdr_unsigned(nuidp->nu_timestamp.tv_sec 1506 - 1); 1507 ktvin.tv_usec = 1508 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1509 1510 /* 1511 * Encrypt the timestamp in ecb mode using the 1512 * session key. 1513 */ 1514 #ifdef NFSKERB 1515 XXX 1516 #endif 1517 1518 *tl++ = rpc_auth_kerb; 1519 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1520 *tl = ktvout.tv_sec; 1521 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1522 *tl++ = ktvout.tv_usec; 1523 *tl++ = txdr_unsigned( 1524 kauth_cred_geteuid(nuidp->nu_cr)); 1525 } else { 1526 *tl++ = 0; 1527 *tl++ = 0; 1528 } 1529 } else { 1530 *tl++ = 0; 1531 *tl++ = 0; 1532 } 1533 switch (err) { 1534 case EPROGUNAVAIL: 1535 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1536 break; 1537 case EPROGMISMATCH: 1538 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1539 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1540 *tl++ = txdr_unsigned(2); 1541 *tl = txdr_unsigned(3); 1542 break; 1543 case EPROCUNAVAIL: 1544 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1545 break; 1546 case EBADRPC: 1547 *tl = txdr_unsigned(RPC_GARBAGE); 1548 break; 1549 default: 1550 *tl = 0; 1551 if (err != NFSERR_RETVOID) { 1552 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1553 if (err) 1554 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1555 else 1556 *tl = 0; 1557 } 1558 break; 1559 }; 1560 } 1561 1562 if (mrq != NULL) 1563 *mrq = mreq; 1564 *mbp = mb; 1565 *bposp = bpos; 1566 if (err != 0 && err != NFSERR_RETVOID) 1567 nfsstats.srvrpc_errs++; 1568 return (0); 1569 } 1570 1571 static void 1572 nfs_timer_schedule(void) 1573 { 1574 1575 callout_schedule(&nfs_timer_ch, nfs_ticks); 1576 } 1577 1578 void 1579 nfs_timer_start(void) 1580 { 1581 1582 if (callout_pending(&nfs_timer_ch)) 1583 return; 1584 1585 nfs_timer_start_ev.ev_count++; 1586 nfs_timer_schedule(); 1587 } 1588 1589 void 1590 nfs_timer_init(void) 1591 { 1592 1593 callout_init(&nfs_timer_ch, 0); 1594 callout_setfunc(&nfs_timer_ch, nfs_timer, NULL); 1595 evcnt_attach_dynamic(&nfs_timer_ev, EVCNT_TYPE_MISC, NULL, 1596 "nfs", "timer"); 1597 evcnt_attach_dynamic(&nfs_timer_start_ev, EVCNT_TYPE_MISC, NULL, 1598 "nfs", "timer start"); 1599 evcnt_attach_dynamic(&nfs_timer_stop_ev, EVCNT_TYPE_MISC, NULL, 1600 "nfs", "timer stop"); 1601 } 1602 1603 /* 1604 * Nfs timer routine 1605 * Scan the nfsreq list and retranmit any requests that have timed out 1606 * To avoid retransmission attempts on STREAM sockets (in the future) make 1607 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1608 */ 1609 void 1610 nfs_timer(void *arg) 1611 { 1612 struct nfsreq *rep; 1613 struct mbuf *m; 1614 struct socket *so; 1615 struct nfsmount *nmp; 1616 int timeo; 1617 int s, error; 1618 bool more = false; 1619 #ifdef NFSSERVER 1620 struct timeval tv; 1621 struct nfssvc_sock *slp; 1622 u_quad_t cur_usec; 1623 #endif 1624 1625 nfs_timer_ev.ev_count++; 1626 1627 s = splsoftnet(); 1628 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1629 more = true; 1630 nmp = rep->r_nmp; 1631 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1632 continue; 1633 if (nfs_sigintr(nmp, rep, rep->r_lwp)) { 1634 rep->r_flags |= R_SOFTTERM; 1635 continue; 1636 } 1637 if (rep->r_rtt >= 0) { 1638 rep->r_rtt++; 1639 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1640 timeo = nmp->nm_timeo; 1641 else 1642 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1643 if (nmp->nm_timeouts > 0) 1644 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1645 if (rep->r_rtt <= timeo) 1646 continue; 1647 if (nmp->nm_timeouts < 1648 (sizeof(nfs_backoff) / sizeof(nfs_backoff[0]))) 1649 nmp->nm_timeouts++; 1650 } 1651 /* 1652 * Check for server not responding 1653 */ 1654 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1655 rep->r_rexmit > nmp->nm_deadthresh) { 1656 nfs_msg(rep->r_lwp, 1657 nmp->nm_mountp->mnt_stat.f_mntfromname, 1658 "not responding"); 1659 rep->r_flags |= R_TPRINTFMSG; 1660 } 1661 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1662 nfsstats.rpctimeouts++; 1663 rep->r_flags |= R_SOFTTERM; 1664 continue; 1665 } 1666 if (nmp->nm_sotype != SOCK_DGRAM) { 1667 if (++rep->r_rexmit > NFS_MAXREXMIT) 1668 rep->r_rexmit = NFS_MAXREXMIT; 1669 continue; 1670 } 1671 if ((so = nmp->nm_so) == NULL) 1672 continue; 1673 1674 /* 1675 * If there is enough space and the window allows.. 1676 * Resend it 1677 * Set r_rtt to -1 in case we fail to send it now. 1678 */ 1679 solock(so); 1680 rep->r_rtt = -1; 1681 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1682 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1683 (rep->r_flags & R_SENT) || 1684 nmp->nm_sent < nmp->nm_cwnd) && 1685 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1686 if (so->so_state & SS_ISCONNECTED) 1687 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1688 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1689 else 1690 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1691 nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0); 1692 if (error) { 1693 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1694 #ifdef DEBUG 1695 printf("nfs_timer: ignoring error %d\n", 1696 error); 1697 #endif 1698 so->so_error = 0; 1699 } 1700 } else { 1701 /* 1702 * Iff first send, start timing 1703 * else turn timing off, backoff timer 1704 * and divide congestion window by 2. 1705 */ 1706 if (rep->r_flags & R_SENT) { 1707 rep->r_flags &= ~R_TIMING; 1708 if (++rep->r_rexmit > NFS_MAXREXMIT) 1709 rep->r_rexmit = NFS_MAXREXMIT; 1710 nmp->nm_cwnd >>= 1; 1711 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1712 nmp->nm_cwnd = NFS_CWNDSCALE; 1713 nfsstats.rpcretries++; 1714 } else { 1715 rep->r_flags |= R_SENT; 1716 nmp->nm_sent += NFS_CWNDSCALE; 1717 } 1718 rep->r_rtt = 0; 1719 } 1720 } 1721 sounlock(so); 1722 } 1723 splx(s); 1724 1725 #ifdef NFSSERVER 1726 /* 1727 * Scan the write gathering queues for writes that need to be 1728 * completed now. 1729 */ 1730 getmicrotime(&tv); 1731 cur_usec = (u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec; 1732 mutex_enter(&nfsd_lock); 1733 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1734 struct nfsrv_descript *nd; 1735 1736 nd = LIST_FIRST(&slp->ns_tq); 1737 if (nd != NULL) { 1738 if (nd->nd_time <= cur_usec) { 1739 nfsrv_wakenfsd_locked(slp); 1740 } 1741 more = true; 1742 } 1743 } 1744 mutex_exit(&nfsd_lock); 1745 #endif /* NFSSERVER */ 1746 if (more) { 1747 nfs_timer_schedule(); 1748 } else { 1749 nfs_timer_stop_ev.ev_count++; 1750 } 1751 } 1752 1753 /* 1754 * Test for a termination condition pending on the process. 1755 * This is used for NFSMNT_INT mounts. 1756 */ 1757 int 1758 nfs_sigintr(nmp, rep, l) 1759 struct nfsmount *nmp; 1760 struct nfsreq *rep; 1761 struct lwp *l; 1762 { 1763 sigset_t ss; 1764 1765 if (rep && (rep->r_flags & R_SOFTTERM)) 1766 return (EINTR); 1767 if (!(nmp->nm_flag & NFSMNT_INT)) 1768 return (0); 1769 if (l) { 1770 sigpending1(l, &ss); 1771 #if 0 1772 sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss); 1773 #endif 1774 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1775 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1776 sigismember(&ss, SIGQUIT)) 1777 return (EINTR); 1778 } 1779 return (0); 1780 } 1781 1782 #ifdef NFS 1783 /* 1784 * Lock a socket against others. 1785 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1786 * and also to avoid race conditions between the processes with nfs requests 1787 * in progress when a reconnect is necessary. 1788 */ 1789 static int 1790 nfs_sndlock(struct nfsmount *nmp, struct nfsreq *rep) 1791 { 1792 struct lwp *l; 1793 int timeo = 0; 1794 bool catch = false; 1795 int error = 0; 1796 1797 if (rep) { 1798 l = rep->r_lwp; 1799 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1800 catch = true; 1801 } else 1802 l = NULL; 1803 mutex_enter(&nmp->nm_lock); 1804 while ((nmp->nm_iflag & NFSMNT_SNDLOCK) != 0) { 1805 if (rep && nfs_sigintr(rep->r_nmp, rep, l)) { 1806 error = EINTR; 1807 goto quit; 1808 } 1809 if (catch) { 1810 cv_timedwait_sig(&nmp->nm_sndcv, &nmp->nm_lock, timeo); 1811 } else { 1812 cv_timedwait(&nmp->nm_sndcv, &nmp->nm_lock, timeo); 1813 } 1814 if (catch) { 1815 catch = false; 1816 timeo = 2 * hz; 1817 } 1818 } 1819 nmp->nm_iflag |= NFSMNT_SNDLOCK; 1820 quit: 1821 mutex_exit(&nmp->nm_lock); 1822 return error; 1823 } 1824 1825 /* 1826 * Unlock the stream socket for others. 1827 */ 1828 static void 1829 nfs_sndunlock(struct nfsmount *nmp) 1830 { 1831 1832 mutex_enter(&nmp->nm_lock); 1833 if ((nmp->nm_iflag & NFSMNT_SNDLOCK) == 0) 1834 panic("nfs sndunlock"); 1835 nmp->nm_iflag &= ~NFSMNT_SNDLOCK; 1836 cv_signal(&nmp->nm_sndcv); 1837 mutex_exit(&nmp->nm_lock); 1838 } 1839 #endif /* NFS */ 1840 1841 static int 1842 nfs_rcvlock(struct nfsmount *nmp, struct nfsreq *rep) 1843 { 1844 int *flagp = &nmp->nm_iflag; 1845 int slptimeo = 0; 1846 bool catch; 1847 int error = 0; 1848 1849 KASSERT(nmp == rep->r_nmp); 1850 1851 catch = (nmp->nm_flag & NFSMNT_INT) != 0; 1852 mutex_enter(&nmp->nm_lock); 1853 while (/* CONSTCOND */ true) { 1854 if (*flagp & NFSMNT_DISMNT) { 1855 cv_signal(&nmp->nm_disconcv); 1856 error = EIO; 1857 break; 1858 } 1859 /* If our reply was received while we were sleeping, 1860 * then just return without taking the lock to avoid a 1861 * situation where a single iod could 'capture' the 1862 * receive lock. 1863 */ 1864 if (rep->r_mrep != NULL) { 1865 error = EALREADY; 1866 break; 1867 } 1868 if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) { 1869 error = EINTR; 1870 break; 1871 } 1872 if ((*flagp & NFSMNT_RCVLOCK) == 0) { 1873 *flagp |= NFSMNT_RCVLOCK; 1874 break; 1875 } 1876 if (catch) { 1877 cv_timedwait_sig(&nmp->nm_rcvcv, &nmp->nm_lock, 1878 slptimeo); 1879 } else { 1880 cv_timedwait(&nmp->nm_rcvcv, &nmp->nm_lock, 1881 slptimeo); 1882 } 1883 if (catch) { 1884 catch = false; 1885 slptimeo = 2 * hz; 1886 } 1887 } 1888 mutex_exit(&nmp->nm_lock); 1889 return error; 1890 } 1891 1892 /* 1893 * Unlock the stream socket for others. 1894 */ 1895 static void 1896 nfs_rcvunlock(struct nfsmount *nmp) 1897 { 1898 1899 mutex_enter(&nmp->nm_lock); 1900 if ((nmp->nm_iflag & NFSMNT_RCVLOCK) == 0) 1901 panic("nfs rcvunlock"); 1902 nmp->nm_iflag &= ~NFSMNT_RCVLOCK; 1903 cv_broadcast(&nmp->nm_rcvcv); 1904 mutex_exit(&nmp->nm_lock); 1905 } 1906 1907 /* 1908 * Parse an RPC request 1909 * - verify it 1910 * - allocate and fill in the cred. 1911 */ 1912 int 1913 nfs_getreq(nd, nfsd, has_header) 1914 struct nfsrv_descript *nd; 1915 struct nfsd *nfsd; 1916 int has_header; 1917 { 1918 int len, i; 1919 u_int32_t *tl; 1920 int32_t t1; 1921 struct uio uio; 1922 struct iovec iov; 1923 char *dpos, *cp2, *cp; 1924 u_int32_t nfsvers, auth_type; 1925 uid_t nickuid; 1926 int error = 0, ticklen; 1927 struct mbuf *mrep, *md; 1928 struct nfsuid *nuidp; 1929 struct timeval tvin, tvout; 1930 1931 memset(&tvout, 0, sizeof tvout); /* XXX gcc */ 1932 1933 KASSERT(nd->nd_cr == NULL); 1934 mrep = nd->nd_mrep; 1935 md = nd->nd_md; 1936 dpos = nd->nd_dpos; 1937 if (has_header) { 1938 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1939 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1940 if (*tl++ != rpc_call) { 1941 m_freem(mrep); 1942 return (EBADRPC); 1943 } 1944 } else 1945 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1946 nd->nd_repstat = 0; 1947 nd->nd_flag = 0; 1948 if (*tl++ != rpc_vers) { 1949 nd->nd_repstat = ERPCMISMATCH; 1950 nd->nd_procnum = NFSPROC_NOOP; 1951 return (0); 1952 } 1953 if (*tl != nfs_prog) { 1954 nd->nd_repstat = EPROGUNAVAIL; 1955 nd->nd_procnum = NFSPROC_NOOP; 1956 return (0); 1957 } 1958 tl++; 1959 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1960 if (nfsvers < NFS_VER2 || nfsvers > NFS_VER3) { 1961 nd->nd_repstat = EPROGMISMATCH; 1962 nd->nd_procnum = NFSPROC_NOOP; 1963 return (0); 1964 } 1965 if (nfsvers == NFS_VER3) 1966 nd->nd_flag = ND_NFSV3; 1967 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1968 if (nd->nd_procnum == NFSPROC_NULL) 1969 return (0); 1970 if (nd->nd_procnum > NFSPROC_COMMIT || 1971 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1972 nd->nd_repstat = EPROCUNAVAIL; 1973 nd->nd_procnum = NFSPROC_NOOP; 1974 return (0); 1975 } 1976 if ((nd->nd_flag & ND_NFSV3) == 0) 1977 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1978 auth_type = *tl++; 1979 len = fxdr_unsigned(int, *tl++); 1980 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1981 m_freem(mrep); 1982 return (EBADRPC); 1983 } 1984 1985 nd->nd_flag &= ~ND_KERBAUTH; 1986 /* 1987 * Handle auth_unix or auth_kerb. 1988 */ 1989 if (auth_type == rpc_auth_unix) { 1990 uid_t uid; 1991 gid_t gid; 1992 1993 nd->nd_cr = kauth_cred_alloc(); 1994 len = fxdr_unsigned(int, *++tl); 1995 if (len < 0 || len > NFS_MAXNAMLEN) { 1996 m_freem(mrep); 1997 error = EBADRPC; 1998 goto errout; 1999 } 2000 nfsm_adv(nfsm_rndup(len)); 2001 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2002 2003 uid = fxdr_unsigned(uid_t, *tl++); 2004 gid = fxdr_unsigned(gid_t, *tl++); 2005 kauth_cred_setuid(nd->nd_cr, uid); 2006 kauth_cred_seteuid(nd->nd_cr, uid); 2007 kauth_cred_setsvuid(nd->nd_cr, uid); 2008 kauth_cred_setgid(nd->nd_cr, gid); 2009 kauth_cred_setegid(nd->nd_cr, gid); 2010 kauth_cred_setsvgid(nd->nd_cr, gid); 2011 2012 len = fxdr_unsigned(int, *tl); 2013 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 2014 m_freem(mrep); 2015 error = EBADRPC; 2016 goto errout; 2017 } 2018 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 2019 2020 if (len > 0) { 2021 size_t grbuf_size = min(len, NGROUPS) * sizeof(gid_t); 2022 gid_t *grbuf = kmem_alloc(grbuf_size, KM_SLEEP); 2023 2024 for (i = 0; i < len; i++) { 2025 if (i < NGROUPS) /* XXX elad */ 2026 grbuf[i] = fxdr_unsigned(gid_t, *tl++); 2027 else 2028 tl++; 2029 } 2030 kauth_cred_setgroups(nd->nd_cr, grbuf, 2031 min(len, NGROUPS), -1, UIO_SYSSPACE); 2032 kmem_free(grbuf, grbuf_size); 2033 } 2034 2035 len = fxdr_unsigned(int, *++tl); 2036 if (len < 0 || len > RPCAUTH_MAXSIZ) { 2037 m_freem(mrep); 2038 error = EBADRPC; 2039 goto errout; 2040 } 2041 if (len > 0) 2042 nfsm_adv(nfsm_rndup(len)); 2043 } else if (auth_type == rpc_auth_kerb) { 2044 switch (fxdr_unsigned(int, *tl++)) { 2045 case RPCAKN_FULLNAME: 2046 ticklen = fxdr_unsigned(int, *tl); 2047 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 2048 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 2049 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 2050 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 2051 m_freem(mrep); 2052 error = EBADRPC; 2053 goto errout; 2054 } 2055 uio.uio_offset = 0; 2056 uio.uio_iov = &iov; 2057 uio.uio_iovcnt = 1; 2058 UIO_SETUP_SYSSPACE(&uio); 2059 iov.iov_base = (void *)&nfsd->nfsd_authstr[4]; 2060 iov.iov_len = RPCAUTH_MAXSIZ - 4; 2061 nfsm_mtouio(&uio, uio.uio_resid); 2062 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2063 if (*tl++ != rpc_auth_kerb || 2064 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 2065 printf("Bad kerb verifier\n"); 2066 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2067 nd->nd_procnum = NFSPROC_NOOP; 2068 return (0); 2069 } 2070 nfsm_dissect(cp, void *, 4 * NFSX_UNSIGNED); 2071 tl = (u_int32_t *)cp; 2072 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 2073 printf("Not fullname kerb verifier\n"); 2074 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2075 nd->nd_procnum = NFSPROC_NOOP; 2076 return (0); 2077 } 2078 cp += NFSX_UNSIGNED; 2079 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 2080 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 2081 nd->nd_flag |= ND_KERBFULL; 2082 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 2083 break; 2084 case RPCAKN_NICKNAME: 2085 if (len != 2 * NFSX_UNSIGNED) { 2086 printf("Kerb nickname short\n"); 2087 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 2088 nd->nd_procnum = NFSPROC_NOOP; 2089 return (0); 2090 } 2091 nickuid = fxdr_unsigned(uid_t, *tl); 2092 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2093 if (*tl++ != rpc_auth_kerb || 2094 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 2095 printf("Kerb nick verifier bad\n"); 2096 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2097 nd->nd_procnum = NFSPROC_NOOP; 2098 return (0); 2099 } 2100 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2101 tvin.tv_sec = *tl++; 2102 tvin.tv_usec = *tl; 2103 2104 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), 2105 nu_hash) { 2106 if (kauth_cred_geteuid(nuidp->nu_cr) == nickuid && 2107 (!nd->nd_nam2 || 2108 netaddr_match(NU_NETFAM(nuidp), 2109 &nuidp->nu_haddr, nd->nd_nam2))) 2110 break; 2111 } 2112 if (!nuidp) { 2113 nd->nd_repstat = 2114 (NFSERR_AUTHERR|AUTH_REJECTCRED); 2115 nd->nd_procnum = NFSPROC_NOOP; 2116 return (0); 2117 } 2118 2119 /* 2120 * Now, decrypt the timestamp using the session key 2121 * and validate it. 2122 */ 2123 #ifdef NFSKERB 2124 XXX 2125 #endif 2126 2127 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 2128 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 2129 if (nuidp->nu_expire < time_second || 2130 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 2131 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 2132 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 2133 nuidp->nu_expire = 0; 2134 nd->nd_repstat = 2135 (NFSERR_AUTHERR|AUTH_REJECTVERF); 2136 nd->nd_procnum = NFSPROC_NOOP; 2137 return (0); 2138 } 2139 kauth_cred_hold(nuidp->nu_cr); 2140 nd->nd_cr = nuidp->nu_cr; 2141 nd->nd_flag |= ND_KERBNICK; 2142 } 2143 } else { 2144 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 2145 nd->nd_procnum = NFSPROC_NOOP; 2146 return (0); 2147 } 2148 2149 nd->nd_md = md; 2150 nd->nd_dpos = dpos; 2151 KASSERT((nd->nd_cr == NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) != 0) 2152 || (nd->nd_cr != NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) == 0)); 2153 return (0); 2154 nfsmout: 2155 errout: 2156 KASSERT(error != 0); 2157 if (nd->nd_cr != NULL) { 2158 kauth_cred_free(nd->nd_cr); 2159 nd->nd_cr = NULL; 2160 } 2161 return (error); 2162 } 2163 2164 int 2165 nfs_msg(l, server, msg) 2166 struct lwp *l; 2167 const char *server, *msg; 2168 { 2169 tpr_t tpr; 2170 2171 if (l) 2172 tpr = tprintf_open(l->l_proc); 2173 else 2174 tpr = NULL; 2175 tprintf(tpr, "nfs server %s: %s\n", server, msg); 2176 tprintf_close(tpr); 2177 return (0); 2178 } 2179 2180 #ifdef NFSSERVER 2181 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 2182 struct nfssvc_sock *, struct lwp *, 2183 struct mbuf **)) = { 2184 nfsrv_null, 2185 nfsrv_getattr, 2186 nfsrv_setattr, 2187 nfsrv_lookup, 2188 nfsrv3_access, 2189 nfsrv_readlink, 2190 nfsrv_read, 2191 nfsrv_write, 2192 nfsrv_create, 2193 nfsrv_mkdir, 2194 nfsrv_symlink, 2195 nfsrv_mknod, 2196 nfsrv_remove, 2197 nfsrv_rmdir, 2198 nfsrv_rename, 2199 nfsrv_link, 2200 nfsrv_readdir, 2201 nfsrv_readdirplus, 2202 nfsrv_statfs, 2203 nfsrv_fsinfo, 2204 nfsrv_pathconf, 2205 nfsrv_commit, 2206 nfsrv_noop 2207 }; 2208 2209 /* 2210 * Socket upcall routine for the nfsd sockets. 2211 * The void *arg is a pointer to the "struct nfssvc_sock". 2212 */ 2213 void 2214 nfsrv_soupcall(struct socket *so, void *arg, int waitflag) 2215 { 2216 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 2217 2218 nfsdsock_setbits(slp, SLP_A_NEEDQ); 2219 nfsrv_wakenfsd(slp); 2220 } 2221 2222 void 2223 nfsrv_rcv(struct nfssvc_sock *slp) 2224 { 2225 struct socket *so; 2226 struct mbuf *m; 2227 struct mbuf *mp, *nam; 2228 struct uio auio; 2229 int flags; 2230 int error; 2231 int setflags = 0; 2232 2233 error = nfsdsock_lock(slp, true); 2234 if (error) { 2235 setflags |= SLP_A_NEEDQ; 2236 goto dorecs_unlocked; 2237 } 2238 2239 nfsdsock_clearbits(slp, SLP_A_NEEDQ); 2240 2241 so = slp->ns_so; 2242 if (so->so_type == SOCK_STREAM) { 2243 /* 2244 * Do soreceive(). 2245 */ 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, &flags); 2250 if (error || mp == NULL) { 2251 if (error == EWOULDBLOCK) 2252 setflags |= SLP_A_NEEDQ; 2253 else 2254 setflags |= SLP_A_DISCONN; 2255 goto dorecs; 2256 } 2257 m = mp; 2258 m_claimm(m, &nfs_mowner); 2259 if (slp->ns_rawend) { 2260 slp->ns_rawend->m_next = m; 2261 slp->ns_cc += 1000000000 - auio.uio_resid; 2262 } else { 2263 slp->ns_raw = m; 2264 slp->ns_cc = 1000000000 - auio.uio_resid; 2265 } 2266 while (m->m_next) 2267 m = m->m_next; 2268 slp->ns_rawend = m; 2269 2270 /* 2271 * Now try and parse record(s) out of the raw stream data. 2272 */ 2273 error = nfsrv_getstream(slp, M_WAIT); 2274 if (error) { 2275 if (error == EPERM) 2276 setflags |= SLP_A_DISCONN; 2277 else 2278 setflags |= SLP_A_NEEDQ; 2279 } 2280 } else { 2281 do { 2282 auio.uio_resid = 1000000000; 2283 /* not need to setup uio_vmspace */ 2284 flags = MSG_DONTWAIT; 2285 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, 2286 &flags); 2287 if (mp) { 2288 if (nam) { 2289 m = nam; 2290 m->m_next = mp; 2291 } else 2292 m = mp; 2293 m_claimm(m, &nfs_mowner); 2294 if (slp->ns_recend) 2295 slp->ns_recend->m_nextpkt = m; 2296 else 2297 slp->ns_rec = m; 2298 slp->ns_recend = m; 2299 m->m_nextpkt = (struct mbuf *)0; 2300 } 2301 if (error) { 2302 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2303 && error != EWOULDBLOCK) { 2304 setflags |= SLP_A_DISCONN; 2305 goto dorecs; 2306 } 2307 } 2308 } while (mp); 2309 } 2310 dorecs: 2311 nfsdsock_unlock(slp); 2312 2313 dorecs_unlocked: 2314 if (setflags) { 2315 nfsdsock_setbits(slp, setflags); 2316 } 2317 } 2318 2319 int 2320 nfsdsock_lock(struct nfssvc_sock *slp, bool waitok) 2321 { 2322 2323 mutex_enter(&slp->ns_lock); 2324 while ((~slp->ns_flags & (SLP_BUSY|SLP_VALID)) == 0) { 2325 if (!waitok) { 2326 mutex_exit(&slp->ns_lock); 2327 return EWOULDBLOCK; 2328 } 2329 cv_wait(&slp->ns_cv, &slp->ns_lock); 2330 } 2331 if ((slp->ns_flags & SLP_VALID) == 0) { 2332 mutex_exit(&slp->ns_lock); 2333 return EINVAL; 2334 } 2335 KASSERT((slp->ns_flags & SLP_BUSY) == 0); 2336 slp->ns_flags |= SLP_BUSY; 2337 mutex_exit(&slp->ns_lock); 2338 2339 return 0; 2340 } 2341 2342 void 2343 nfsdsock_unlock(struct nfssvc_sock *slp) 2344 { 2345 2346 mutex_enter(&slp->ns_lock); 2347 KASSERT((slp->ns_flags & SLP_BUSY) != 0); 2348 cv_broadcast(&slp->ns_cv); 2349 slp->ns_flags &= ~SLP_BUSY; 2350 mutex_exit(&slp->ns_lock); 2351 } 2352 2353 int 2354 nfsdsock_drain(struct nfssvc_sock *slp) 2355 { 2356 int error = 0; 2357 2358 mutex_enter(&slp->ns_lock); 2359 if ((slp->ns_flags & SLP_VALID) == 0) { 2360 error = EINVAL; 2361 goto done; 2362 } 2363 slp->ns_flags &= ~SLP_VALID; 2364 while ((slp->ns_flags & SLP_BUSY) != 0) { 2365 cv_wait(&slp->ns_cv, &slp->ns_lock); 2366 } 2367 done: 2368 mutex_exit(&slp->ns_lock); 2369 2370 return error; 2371 } 2372 2373 /* 2374 * Try and extract an RPC request from the mbuf data list received on a 2375 * stream socket. The "waitflag" argument indicates whether or not it 2376 * can sleep. 2377 */ 2378 int 2379 nfsrv_getstream(slp, waitflag) 2380 struct nfssvc_sock *slp; 2381 int waitflag; 2382 { 2383 struct mbuf *m, **mpp; 2384 struct mbuf *recm; 2385 u_int32_t recmark; 2386 int error = 0; 2387 2388 KASSERT((slp->ns_flags & SLP_BUSY) != 0); 2389 for (;;) { 2390 if (slp->ns_reclen == 0) { 2391 if (slp->ns_cc < NFSX_UNSIGNED) { 2392 break; 2393 } 2394 m = slp->ns_raw; 2395 m_copydata(m, 0, NFSX_UNSIGNED, (void *)&recmark); 2396 m_adj(m, NFSX_UNSIGNED); 2397 slp->ns_cc -= NFSX_UNSIGNED; 2398 recmark = ntohl(recmark); 2399 slp->ns_reclen = recmark & ~0x80000000; 2400 if (recmark & 0x80000000) 2401 slp->ns_sflags |= SLP_S_LASTFRAG; 2402 else 2403 slp->ns_sflags &= ~SLP_S_LASTFRAG; 2404 if (slp->ns_reclen > NFS_MAXPACKET) { 2405 error = EPERM; 2406 break; 2407 } 2408 } 2409 2410 /* 2411 * Now get the record part. 2412 * 2413 * Note that slp->ns_reclen may be 0. Linux sometimes 2414 * generates 0-length records. 2415 */ 2416 if (slp->ns_cc == slp->ns_reclen) { 2417 recm = slp->ns_raw; 2418 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2419 slp->ns_cc = slp->ns_reclen = 0; 2420 } else if (slp->ns_cc > slp->ns_reclen) { 2421 recm = slp->ns_raw; 2422 m = m_split(recm, slp->ns_reclen, waitflag); 2423 if (m == NULL) { 2424 error = EWOULDBLOCK; 2425 break; 2426 } 2427 m_claimm(recm, &nfs_mowner); 2428 slp->ns_raw = m; 2429 if (m->m_next == NULL) 2430 slp->ns_rawend = m; 2431 slp->ns_cc -= slp->ns_reclen; 2432 slp->ns_reclen = 0; 2433 } else { 2434 break; 2435 } 2436 2437 /* 2438 * Accumulate the fragments into a record. 2439 */ 2440 mpp = &slp->ns_frag; 2441 while (*mpp) 2442 mpp = &((*mpp)->m_next); 2443 *mpp = recm; 2444 if (slp->ns_sflags & SLP_S_LASTFRAG) { 2445 if (slp->ns_recend) 2446 slp->ns_recend->m_nextpkt = slp->ns_frag; 2447 else 2448 slp->ns_rec = slp->ns_frag; 2449 slp->ns_recend = slp->ns_frag; 2450 slp->ns_frag = NULL; 2451 } 2452 } 2453 2454 return error; 2455 } 2456 2457 /* 2458 * Parse an RPC header. 2459 */ 2460 int 2461 nfsrv_dorec(struct nfssvc_sock *slp, struct nfsd *nfsd, 2462 struct nfsrv_descript **ndp, bool *more) 2463 { 2464 struct mbuf *m, *nam; 2465 struct nfsrv_descript *nd; 2466 int error; 2467 2468 *ndp = NULL; 2469 *more = false; 2470 2471 if (nfsdsock_lock(slp, true)) { 2472 return ENOBUFS; 2473 } 2474 m = slp->ns_rec; 2475 if (m == NULL) { 2476 nfsdsock_unlock(slp); 2477 return ENOBUFS; 2478 } 2479 slp->ns_rec = m->m_nextpkt; 2480 if (slp->ns_rec) { 2481 m->m_nextpkt = NULL; 2482 *more = true; 2483 } else { 2484 slp->ns_recend = NULL; 2485 } 2486 nfsdsock_unlock(slp); 2487 2488 if (m->m_type == MT_SONAME) { 2489 nam = m; 2490 m = m->m_next; 2491 nam->m_next = NULL; 2492 } else 2493 nam = NULL; 2494 nd = nfsdreq_alloc(); 2495 nd->nd_md = nd->nd_mrep = m; 2496 nd->nd_nam2 = nam; 2497 nd->nd_dpos = mtod(m, void *); 2498 error = nfs_getreq(nd, nfsd, true); 2499 if (error) { 2500 m_freem(nam); 2501 nfsdreq_free(nd); 2502 return (error); 2503 } 2504 *ndp = nd; 2505 nfsd->nfsd_nd = nd; 2506 return (0); 2507 } 2508 2509 /* 2510 * Search for a sleeping nfsd and wake it up. 2511 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2512 * running nfsds will go look for the work in the nfssvc_sock list. 2513 */ 2514 static void 2515 nfsrv_wakenfsd_locked(struct nfssvc_sock *slp) 2516 { 2517 struct nfsd *nd; 2518 2519 KASSERT(mutex_owned(&nfsd_lock)); 2520 2521 if ((slp->ns_flags & SLP_VALID) == 0) 2522 return; 2523 if (slp->ns_gflags & SLP_G_DOREC) 2524 return; 2525 nd = SLIST_FIRST(&nfsd_idle_head); 2526 if (nd) { 2527 SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle); 2528 if (nd->nfsd_slp) 2529 panic("nfsd wakeup"); 2530 slp->ns_sref++; 2531 KASSERT(slp->ns_sref > 0); 2532 nd->nfsd_slp = slp; 2533 cv_signal(&nd->nfsd_cv); 2534 } else { 2535 slp->ns_gflags |= SLP_G_DOREC; 2536 nfsd_head_flag |= NFSD_CHECKSLP; 2537 TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending); 2538 } 2539 } 2540 2541 void 2542 nfsrv_wakenfsd(struct nfssvc_sock *slp) 2543 { 2544 2545 mutex_enter(&nfsd_lock); 2546 nfsrv_wakenfsd_locked(slp); 2547 mutex_exit(&nfsd_lock); 2548 } 2549 2550 int 2551 nfsdsock_sendreply(struct nfssvc_sock *slp, struct nfsrv_descript *nd) 2552 { 2553 int error; 2554 2555 if (nd->nd_mrep != NULL) { 2556 m_freem(nd->nd_mrep); 2557 nd->nd_mrep = NULL; 2558 } 2559 2560 mutex_enter(&slp->ns_lock); 2561 if ((slp->ns_flags & SLP_SENDING) != 0) { 2562 SIMPLEQ_INSERT_TAIL(&slp->ns_sendq, nd, nd_sendq); 2563 mutex_exit(&slp->ns_lock); 2564 return 0; 2565 } 2566 KASSERT(SIMPLEQ_EMPTY(&slp->ns_sendq)); 2567 slp->ns_flags |= SLP_SENDING; 2568 mutex_exit(&slp->ns_lock); 2569 2570 again: 2571 error = nfs_send(slp->ns_so, nd->nd_nam2, nd->nd_mreq, NULL, curlwp); 2572 if (nd->nd_nam2) { 2573 m_free(nd->nd_nam2); 2574 } 2575 nfsdreq_free(nd); 2576 2577 mutex_enter(&slp->ns_lock); 2578 KASSERT((slp->ns_flags & SLP_SENDING) != 0); 2579 nd = SIMPLEQ_FIRST(&slp->ns_sendq); 2580 if (nd != NULL) { 2581 SIMPLEQ_REMOVE_HEAD(&slp->ns_sendq, nd_sendq); 2582 mutex_exit(&slp->ns_lock); 2583 goto again; 2584 } 2585 slp->ns_flags &= ~SLP_SENDING; 2586 mutex_exit(&slp->ns_lock); 2587 2588 return error; 2589 } 2590 2591 void 2592 nfsdsock_setbits(struct nfssvc_sock *slp, int bits) 2593 { 2594 2595 mutex_enter(&slp->ns_alock); 2596 slp->ns_aflags |= bits; 2597 mutex_exit(&slp->ns_alock); 2598 } 2599 2600 void 2601 nfsdsock_clearbits(struct nfssvc_sock *slp, int bits) 2602 { 2603 2604 mutex_enter(&slp->ns_alock); 2605 slp->ns_aflags &= ~bits; 2606 mutex_exit(&slp->ns_alock); 2607 } 2608 2609 bool 2610 nfsdsock_testbits(struct nfssvc_sock *slp, int bits) 2611 { 2612 2613 return (slp->ns_aflags & bits); 2614 } 2615 #endif /* NFSSERVER */ 2616 2617 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) 2618 static struct pool nfs_srvdesc_pool; 2619 2620 void 2621 nfsdreq_init(void) 2622 { 2623 2624 pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript), 2625 0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr, IPL_NONE); 2626 } 2627 2628 struct nfsrv_descript * 2629 nfsdreq_alloc(void) 2630 { 2631 struct nfsrv_descript *nd; 2632 2633 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK); 2634 nd->nd_cr = NULL; 2635 return nd; 2636 } 2637 2638 void 2639 nfsdreq_free(struct nfsrv_descript *nd) 2640 { 2641 kauth_cred_t cr; 2642 2643 cr = nd->nd_cr; 2644 if (cr != NULL) { 2645 kauth_cred_free(cr); 2646 } 2647 pool_put(&nfs_srvdesc_pool, nd); 2648 } 2649 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */ 2650