1 /* $NetBSD: nfs_socket.c,v 1.170 2008/04/24 11:38:39 ad Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1991, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 35 */ 36 37 /* 38 * Socket operations for use by nfs 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: nfs_socket.c,v 1.170 2008/04/24 11:38:39 ad Exp $"); 43 44 #include "fs_nfs.h" 45 #include "opt_nfs.h" 46 #include "opt_nfsserver.h" 47 #include "opt_mbuftrace.h" 48 #include "opt_inet.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/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 static int nfs_sndlock(struct nfsmount *, struct nfsreq *); 176 static void nfs_sndunlock(struct nfsmount *); 177 static int nfs_rcvlock(struct nfsmount *, struct nfsreq *); 178 static void nfs_rcvunlock(struct nfsmount *); 179 180 #if defined(NFSSERVER) 181 static void nfsrv_wakenfsd_locked(struct nfssvc_sock *); 182 #endif /* defined(NFSSERVER) */ 183 184 /* 185 * Initialize sockets and congestion for a new NFS connection. 186 * We do not free the sockaddr if error. 187 */ 188 int 189 nfs_connect(nmp, rep, l) 190 struct nfsmount *nmp; 191 struct nfsreq *rep; 192 struct lwp *l; 193 { 194 struct socket *so; 195 int error, rcvreserve, sndreserve; 196 struct sockaddr *saddr; 197 struct sockaddr_in *sin; 198 #ifdef INET6 199 struct sockaddr_in6 *sin6; 200 #endif 201 struct mbuf *m; 202 203 nmp->nm_so = (struct socket *)0; 204 saddr = mtod(nmp->nm_nam, struct sockaddr *); 205 error = socreate(saddr->sa_family, &nmp->nm_so, 206 nmp->nm_sotype, nmp->nm_soproto, l, NULL); 207 if (error) 208 goto bad; 209 so = nmp->nm_so; 210 #ifdef MBUFTRACE 211 so->so_mowner = &nfs_mowner; 212 so->so_rcv.sb_mowner = &nfs_mowner; 213 so->so_snd.sb_mowner = &nfs_mowner; 214 #endif 215 nmp->nm_soflags = so->so_proto->pr_flags; 216 217 /* 218 * Some servers require that the client port be a reserved port number. 219 */ 220 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 221 m = m_get(M_WAIT, MT_SOOPTS); 222 MCLAIM(m, so->so_mowner); 223 *mtod(m, int32_t *) = IP_PORTRANGE_LOW; 224 m->m_len = sizeof(int32_t); 225 if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m))) 226 goto bad; 227 m = m_get(M_WAIT, MT_SONAME); 228 MCLAIM(m, so->so_mowner); 229 sin = mtod(m, struct sockaddr_in *); 230 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 231 sin->sin_family = AF_INET; 232 sin->sin_addr.s_addr = INADDR_ANY; 233 sin->sin_port = 0; 234 error = sobind(so, m, &lwp0); 235 m_freem(m); 236 if (error) 237 goto bad; 238 } 239 #ifdef INET6 240 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) { 241 m = m_get(M_WAIT, MT_SOOPTS); 242 MCLAIM(m, so->so_mowner); 243 *mtod(m, int32_t *) = IPV6_PORTRANGE_LOW; 244 m->m_len = sizeof(int32_t); 245 if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m))) 246 goto bad; 247 m = m_get(M_WAIT, MT_SONAME); 248 MCLAIM(m, so->so_mowner); 249 sin6 = mtod(m, struct sockaddr_in6 *); 250 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6); 251 sin6->sin6_family = AF_INET6; 252 sin6->sin6_addr = in6addr_any; 253 sin6->sin6_port = 0; 254 error = sobind(so, m, &lwp0); 255 m_freem(m); 256 if (error) 257 goto bad; 258 } 259 #endif 260 261 /* 262 * Protocols that do not require connections may be optionally left 263 * unconnected for servers that reply from a port other than NFS_PORT. 264 */ 265 solock(so); 266 if (nmp->nm_flag & NFSMNT_NOCONN) { 267 if (nmp->nm_soflags & PR_CONNREQUIRED) { 268 sounlock(so); 269 error = ENOTCONN; 270 goto bad; 271 } 272 } else { 273 error = soconnect(so, nmp->nm_nam, l); 274 if (error) { 275 sounlock(so); 276 goto bad; 277 } 278 279 /* 280 * Wait for the connection to complete. Cribbed from the 281 * connect system call but with the wait timing out so 282 * that interruptible mounts don't hang here for a long time. 283 */ 284 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 285 (void)sowait(so, 2 * hz); 286 if ((so->so_state & SS_ISCONNECTING) && 287 so->so_error == 0 && rep && 288 (error = nfs_sigintr(nmp, rep, rep->r_lwp)) != 0){ 289 so->so_state &= ~SS_ISCONNECTING; 290 sounlock(so); 291 goto bad; 292 } 293 } 294 if (so->so_error) { 295 error = so->so_error; 296 so->so_error = 0; 297 sounlock(so); 298 goto bad; 299 } 300 } 301 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 302 so->so_rcv.sb_timeo = (5 * hz); 303 so->so_snd.sb_timeo = (5 * hz); 304 } else { 305 /* 306 * enable receive timeout to detect server crash and reconnect. 307 * otherwise, we can be stuck in soreceive forever. 308 */ 309 so->so_rcv.sb_timeo = (5 * hz); 310 so->so_snd.sb_timeo = 0; 311 } 312 if (nmp->nm_sotype == SOCK_DGRAM) { 313 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 314 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 315 NFS_MAXPKTHDR) * 2; 316 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 317 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 318 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 319 NFS_MAXPKTHDR) * 2; 320 } else { 321 sounlock(so); 322 if (nmp->nm_sotype != SOCK_STREAM) 323 panic("nfscon sotype"); 324 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 325 m = m_get(M_WAIT, MT_SOOPTS); 326 MCLAIM(m, so->so_mowner); 327 *mtod(m, int32_t *) = 1; 328 m->m_len = sizeof(int32_t); 329 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 330 } 331 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 332 m = m_get(M_WAIT, MT_SOOPTS); 333 MCLAIM(m, so->so_mowner); 334 *mtod(m, int32_t *) = 1; 335 m->m_len = sizeof(int32_t); 336 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 337 } 338 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 339 sizeof (u_int32_t)) * 2; 340 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 341 sizeof (u_int32_t)) * 2; 342 solock(so); 343 } 344 error = soreserve(so, sndreserve, rcvreserve); 345 if (error) { 346 sounlock(so); 347 goto bad; 348 } 349 so->so_rcv.sb_flags |= SB_NOINTR; 350 so->so_snd.sb_flags |= SB_NOINTR; 351 sounlock(so); 352 353 /* Initialize other non-zero congestion variables */ 354 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 355 NFS_TIMEO << 3; 356 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 357 nmp->nm_sdrtt[3] = 0; 358 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 359 nmp->nm_sent = 0; 360 nmp->nm_timeouts = 0; 361 return (0); 362 363 bad: 364 nfs_disconnect(nmp); 365 return (error); 366 } 367 368 /* 369 * Reconnect routine: 370 * Called when a connection is broken on a reliable protocol. 371 * - clean up the old socket 372 * - nfs_connect() again 373 * - set R_MUSTRESEND for all outstanding requests on mount point 374 * If this fails the mount point is DEAD! 375 * nb: Must be called with the nfs_sndlock() set on the mount point. 376 */ 377 int 378 nfs_reconnect(struct nfsreq *rep) 379 { 380 struct nfsreq *rp; 381 struct nfsmount *nmp = rep->r_nmp; 382 int error; 383 384 nfs_disconnect(nmp); 385 while ((error = nfs_connect(nmp, rep, &lwp0)) != 0) { 386 if (error == EINTR || error == ERESTART) 387 return (EINTR); 388 kpause("nfscn2", false, hz, NULL); 389 } 390 391 /* 392 * Loop through outstanding request list and fix up all requests 393 * on old socket. 394 */ 395 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 396 if (rp->r_nmp == nmp) { 397 if ((rp->r_flags & R_MUSTRESEND) == 0) 398 rp->r_flags |= R_MUSTRESEND | R_REXMITTED; 399 rp->r_rexmit = 0; 400 } 401 } 402 return (0); 403 } 404 405 /* 406 * NFS disconnect. Clean up and unlink. 407 */ 408 void 409 nfs_disconnect(nmp) 410 struct nfsmount *nmp; 411 { 412 struct socket *so; 413 int drain = 0; 414 415 if (nmp->nm_so) { 416 so = nmp->nm_so; 417 nmp->nm_so = (struct socket *)0; 418 solock(so); 419 soshutdown(so, SHUT_RDWR); 420 sounlock(so); 421 drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0; 422 if (drain) { 423 /* 424 * soshutdown() above should wake up the current 425 * listener. 426 * Now wake up those waiting for the receive lock, and 427 * wait for them to go away unhappy, to prevent *nmp 428 * from evaporating while they're sleeping. 429 */ 430 mutex_enter(&nmp->nm_lock); 431 while (nmp->nm_waiters > 0) { 432 cv_broadcast(&nmp->nm_rcvcv); 433 cv_broadcast(&nmp->nm_sndcv); 434 cv_wait(&nmp->nm_disconcv, &nmp->nm_lock); 435 } 436 mutex_exit(&nmp->nm_lock); 437 } 438 soclose(so); 439 } 440 #ifdef DIAGNOSTIC 441 if (drain && (nmp->nm_waiters > 0)) 442 panic("nfs_disconnect: waiters left after drain?"); 443 #endif 444 } 445 446 void 447 nfs_safedisconnect(nmp) 448 struct nfsmount *nmp; 449 { 450 struct nfsreq dummyreq; 451 452 memset(&dummyreq, 0, sizeof(dummyreq)); 453 dummyreq.r_nmp = nmp; 454 nfs_rcvlock(nmp, &dummyreq); /* XXX ignored error return */ 455 nfs_disconnect(nmp); 456 nfs_rcvunlock(nmp); 457 } 458 459 /* 460 * This is the nfs send routine. For connection based socket types, it 461 * must be called with an nfs_sndlock() on the socket. 462 * "rep == NULL" indicates that it has been called from a server. 463 * For the client side: 464 * - return EINTR if the RPC is terminated, 0 otherwise 465 * - set R_MUSTRESEND if the send fails for any reason 466 * - do any cleanup required by recoverable socket errors (? ? ?) 467 * For the server side: 468 * - return EINTR or ERESTART if interrupted by a signal 469 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 470 * - do any cleanup required by recoverable socket errors (? ? ?) 471 */ 472 int 473 nfs_send(so, nam, top, rep, l) 474 struct socket *so; 475 struct mbuf *nam; 476 struct mbuf *top; 477 struct nfsreq *rep; 478 struct lwp *l; 479 { 480 struct mbuf *sendnam; 481 int error, soflags, flags; 482 483 /* XXX nfs_doio()/nfs_request() calls with rep->r_lwp == NULL */ 484 if (l == NULL && rep->r_lwp == NULL) 485 l = curlwp; 486 487 if (rep) { 488 if (rep->r_flags & R_SOFTTERM) { 489 m_freem(top); 490 return (EINTR); 491 } 492 if ((so = rep->r_nmp->nm_so) == NULL) { 493 rep->r_flags |= R_MUSTRESEND; 494 m_freem(top); 495 return (0); 496 } 497 rep->r_flags &= ~R_MUSTRESEND; 498 soflags = rep->r_nmp->nm_soflags; 499 } else 500 soflags = so->so_proto->pr_flags; 501 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 502 sendnam = (struct mbuf *)0; 503 else 504 sendnam = nam; 505 if (so->so_type == SOCK_SEQPACKET) 506 flags = MSG_EOR; 507 else 508 flags = 0; 509 510 error = (*so->so_send)(so, sendnam, NULL, top, NULL, flags, l); 511 if (error) { 512 if (rep) { 513 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 514 /* 515 * We're too fast for the network/driver, 516 * and UDP isn't flowcontrolled. 517 * We need to resend. This is not fatal, 518 * just try again. 519 * 520 * Could be smarter here by doing some sort 521 * of a backoff, but this is rare. 522 */ 523 rep->r_flags |= R_MUSTRESEND; 524 } else { 525 if (error != EPIPE) 526 log(LOG_INFO, 527 "nfs send error %d for %s\n", 528 error, 529 rep->r_nmp->nm_mountp-> 530 mnt_stat.f_mntfromname); 531 /* 532 * Deal with errors for the client side. 533 */ 534 if (rep->r_flags & R_SOFTTERM) 535 error = EINTR; 536 else 537 rep->r_flags |= R_MUSTRESEND; 538 } 539 } else { 540 /* 541 * See above. This error can happen under normal 542 * circumstances and the log is too noisy. 543 * The error will still show up in nfsstat. 544 */ 545 if (error != ENOBUFS || so->so_type != SOCK_DGRAM) 546 log(LOG_INFO, "nfsd send error %d\n", error); 547 } 548 549 /* 550 * Handle any recoverable (soft) socket errors here. (? ? ?) 551 */ 552 if (error != EINTR && error != ERESTART && 553 error != EWOULDBLOCK && error != EPIPE) 554 error = 0; 555 } 556 return (error); 557 } 558 559 #ifdef NFS 560 /* 561 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 562 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 563 * Mark and consolidate the data into a new mbuf list. 564 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 565 * small mbufs. 566 * For SOCK_STREAM we must be very careful to read an entire record once 567 * we have read any of it, even if the system call has been interrupted. 568 */ 569 static int 570 nfs_receive(struct nfsreq *rep, struct mbuf **aname, struct mbuf **mp, 571 struct lwp *l) 572 { 573 struct socket *so; 574 struct uio auio; 575 struct iovec aio; 576 struct mbuf *m; 577 struct mbuf *control; 578 u_int32_t len; 579 struct mbuf **getnam; 580 int error, sotype, rcvflg; 581 582 /* 583 * Set up arguments for soreceive() 584 */ 585 *mp = (struct mbuf *)0; 586 *aname = (struct mbuf *)0; 587 sotype = rep->r_nmp->nm_sotype; 588 589 /* 590 * For reliable protocols, lock against other senders/receivers 591 * in case a reconnect is necessary. 592 * For SOCK_STREAM, first get the Record Mark to find out how much 593 * more there is to get. 594 * We must lock the socket against other receivers 595 * until we have an entire rpc request/reply. 596 */ 597 if (sotype != SOCK_DGRAM) { 598 error = nfs_sndlock(rep->r_nmp, rep); 599 if (error) 600 return (error); 601 tryagain: 602 /* 603 * Check for fatal errors and resending request. 604 */ 605 /* 606 * Ugh: If a reconnect attempt just happened, nm_so 607 * would have changed. NULL indicates a failed 608 * attempt that has essentially shut down this 609 * mount point. 610 */ 611 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 612 nfs_sndunlock(rep->r_nmp); 613 return (EINTR); 614 } 615 so = rep->r_nmp->nm_so; 616 if (!so) { 617 error = nfs_reconnect(rep); 618 if (error) { 619 nfs_sndunlock(rep->r_nmp); 620 return (error); 621 } 622 goto tryagain; 623 } 624 while (rep->r_flags & R_MUSTRESEND) { 625 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 626 nfsstats.rpcretries++; 627 rep->r_rtt = 0; 628 rep->r_flags &= ~R_TIMING; 629 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, l); 630 if (error) { 631 if (error == EINTR || error == ERESTART || 632 (error = nfs_reconnect(rep)) != 0) { 633 nfs_sndunlock(rep->r_nmp); 634 return (error); 635 } 636 goto tryagain; 637 } 638 } 639 nfs_sndunlock(rep->r_nmp); 640 if (sotype == SOCK_STREAM) { 641 aio.iov_base = (void *) &len; 642 aio.iov_len = sizeof(u_int32_t); 643 auio.uio_iov = &aio; 644 auio.uio_iovcnt = 1; 645 auio.uio_rw = UIO_READ; 646 auio.uio_offset = 0; 647 auio.uio_resid = sizeof(u_int32_t); 648 UIO_SETUP_SYSSPACE(&auio); 649 do { 650 rcvflg = MSG_WAITALL; 651 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 652 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 653 if (error == EWOULDBLOCK && rep) { 654 if (rep->r_flags & R_SOFTTERM) 655 return (EINTR); 656 /* 657 * if it seems that the server died after it 658 * received our request, set EPIPE so that 659 * we'll reconnect and retransmit requests. 660 */ 661 if (rep->r_rexmit >= rep->r_nmp->nm_retry) { 662 nfsstats.rpctimeouts++; 663 error = EPIPE; 664 } 665 } 666 } while (error == EWOULDBLOCK); 667 if (!error && auio.uio_resid > 0) { 668 /* 669 * Don't log a 0 byte receive; it means 670 * that the socket has been closed, and 671 * can happen during normal operation 672 * (forcible unmount or Solaris server). 673 */ 674 if (auio.uio_resid != sizeof (u_int32_t)) 675 log(LOG_INFO, 676 "short receive (%lu/%lu) from nfs server %s\n", 677 (u_long)sizeof(u_int32_t) - auio.uio_resid, 678 (u_long)sizeof(u_int32_t), 679 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 680 error = EPIPE; 681 } 682 if (error) 683 goto errout; 684 len = ntohl(len) & ~0x80000000; 685 /* 686 * This is SERIOUS! We are out of sync with the sender 687 * and forcing a disconnect/reconnect is all I can do. 688 */ 689 if (len > NFS_MAXPACKET) { 690 log(LOG_ERR, "%s (%d) from nfs server %s\n", 691 "impossible packet length", 692 len, 693 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 694 error = EFBIG; 695 goto errout; 696 } 697 auio.uio_resid = len; 698 do { 699 rcvflg = MSG_WAITALL; 700 error = (*so->so_receive)(so, (struct mbuf **)0, 701 &auio, mp, (struct mbuf **)0, &rcvflg); 702 } while (error == EWOULDBLOCK || error == EINTR || 703 error == ERESTART); 704 if (!error && auio.uio_resid > 0) { 705 if (len != auio.uio_resid) 706 log(LOG_INFO, 707 "short receive (%lu/%d) from nfs server %s\n", 708 (u_long)len - auio.uio_resid, len, 709 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 710 error = EPIPE; 711 } 712 } else { 713 /* 714 * NB: Since uio_resid is big, MSG_WAITALL is ignored 715 * and soreceive() will return when it has either a 716 * control msg or a data msg. 717 * We have no use for control msg., but must grab them 718 * and then throw them away so we know what is going 719 * on. 720 */ 721 auio.uio_resid = len = 100000000; /* Anything Big */ 722 /* not need to setup uio_vmspace */ 723 do { 724 rcvflg = 0; 725 error = (*so->so_receive)(so, (struct mbuf **)0, 726 &auio, mp, &control, &rcvflg); 727 if (control) 728 m_freem(control); 729 if (error == EWOULDBLOCK && rep) { 730 if (rep->r_flags & R_SOFTTERM) 731 return (EINTR); 732 } 733 } while (error == EWOULDBLOCK || 734 (!error && *mp == NULL && control)); 735 if ((rcvflg & MSG_EOR) == 0) 736 printf("Egad!!\n"); 737 if (!error && *mp == NULL) 738 error = EPIPE; 739 len -= auio.uio_resid; 740 } 741 errout: 742 if (error && error != EINTR && error != ERESTART) { 743 m_freem(*mp); 744 *mp = (struct mbuf *)0; 745 if (error != EPIPE) 746 log(LOG_INFO, 747 "receive error %d from nfs server %s\n", 748 error, 749 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 750 error = nfs_sndlock(rep->r_nmp, rep); 751 if (!error) 752 error = nfs_reconnect(rep); 753 if (!error) 754 goto tryagain; 755 else 756 nfs_sndunlock(rep->r_nmp); 757 } 758 } else { 759 if ((so = rep->r_nmp->nm_so) == NULL) 760 return (EACCES); 761 if (so->so_state & SS_ISCONNECTED) 762 getnam = (struct mbuf **)0; 763 else 764 getnam = aname; 765 auio.uio_resid = len = 1000000; 766 /* not need to setup uio_vmspace */ 767 do { 768 rcvflg = 0; 769 error = (*so->so_receive)(so, getnam, &auio, mp, 770 (struct mbuf **)0, &rcvflg); 771 if (error == EWOULDBLOCK && 772 (rep->r_flags & R_SOFTTERM)) 773 return (EINTR); 774 } while (error == EWOULDBLOCK); 775 len -= auio.uio_resid; 776 if (!error && *mp == NULL) 777 error = EPIPE; 778 } 779 if (error) { 780 m_freem(*mp); 781 *mp = (struct mbuf *)0; 782 } 783 return (error); 784 } 785 786 /* 787 * Implement receipt of reply on a socket. 788 * We must search through the list of received datagrams matching them 789 * with outstanding requests using the xid, until ours is found. 790 */ 791 /* ARGSUSED */ 792 static int 793 nfs_reply(struct nfsreq *myrep, struct lwp *lwp) 794 { 795 struct nfsreq *rep; 796 struct nfsmount *nmp = myrep->r_nmp; 797 int32_t t1; 798 struct mbuf *mrep, *nam, *md; 799 u_int32_t rxid, *tl; 800 char *dpos, *cp2; 801 int error; 802 803 /* 804 * Loop around until we get our own reply 805 */ 806 for (;;) { 807 /* 808 * Lock against other receivers so that I don't get stuck in 809 * sbwait() after someone else has received my reply for me. 810 * Also necessary for connection based protocols to avoid 811 * race conditions during a reconnect. 812 */ 813 error = nfs_rcvlock(nmp, myrep); 814 if (error == EALREADY) 815 return (0); 816 if (error) 817 return (error); 818 /* 819 * Get the next Rpc reply off the socket 820 */ 821 822 mutex_enter(&nmp->nm_lock); 823 nmp->nm_waiters++; 824 mutex_exit(&nmp->nm_lock); 825 826 error = nfs_receive(myrep, &nam, &mrep, lwp); 827 828 mutex_enter(&nmp->nm_lock); 829 nmp->nm_waiters--; 830 cv_signal(&nmp->nm_disconcv); 831 mutex_exit(&nmp->nm_lock); 832 833 if (error) { 834 nfs_rcvunlock(nmp); 835 836 if (nmp->nm_iflag & NFSMNT_DISMNT) { 837 /* 838 * Oops, we're going away now.. 839 */ 840 return error; 841 } 842 /* 843 * Ignore routing errors on connectionless protocols? ? 844 */ 845 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 846 nmp->nm_so->so_error = 0; 847 #ifdef DEBUG 848 printf("nfs_reply: ignoring error %d\n", error); 849 #endif 850 continue; 851 } 852 return (error); 853 } 854 if (nam) 855 m_freem(nam); 856 857 /* 858 * Get the xid and check that it is an rpc reply 859 */ 860 md = mrep; 861 dpos = mtod(md, void *); 862 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 863 rxid = *tl++; 864 if (*tl != rpc_reply) { 865 nfsstats.rpcinvalid++; 866 m_freem(mrep); 867 nfsmout: 868 nfs_rcvunlock(nmp); 869 continue; 870 } 871 872 /* 873 * Loop through the request list to match up the reply 874 * Iff no match, just drop the datagram 875 */ 876 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 877 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 878 /* Found it.. */ 879 rep->r_mrep = mrep; 880 rep->r_md = md; 881 rep->r_dpos = dpos; 882 if (nfsrtton) { 883 struct rttl *rt; 884 885 rt = &nfsrtt.rttl[nfsrtt.pos]; 886 rt->proc = rep->r_procnum; 887 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 888 rt->sent = nmp->nm_sent; 889 rt->cwnd = nmp->nm_cwnd; 890 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 891 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 892 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx; 893 getmicrotime(&rt->tstamp); 894 if (rep->r_flags & R_TIMING) 895 rt->rtt = rep->r_rtt; 896 else 897 rt->rtt = 1000000; 898 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 899 } 900 /* 901 * Update congestion window. 902 * Do the additive increase of 903 * one rpc/rtt. 904 */ 905 if (nmp->nm_cwnd <= nmp->nm_sent) { 906 nmp->nm_cwnd += 907 (NFS_CWNDSCALE * NFS_CWNDSCALE + 908 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 909 if (nmp->nm_cwnd > NFS_MAXCWND) 910 nmp->nm_cwnd = NFS_MAXCWND; 911 } 912 rep->r_flags &= ~R_SENT; 913 nmp->nm_sent -= NFS_CWNDSCALE; 914 /* 915 * Update rtt using a gain of 0.125 on the mean 916 * and a gain of 0.25 on the deviation. 917 */ 918 if (rep->r_flags & R_TIMING) { 919 /* 920 * Since the timer resolution of 921 * NFS_HZ is so course, it can often 922 * result in r_rtt == 0. Since 923 * r_rtt == N means that the actual 924 * rtt is between N+dt and N+2-dt ticks, 925 * add 1. 926 */ 927 t1 = rep->r_rtt + 1; 928 t1 -= (NFS_SRTT(rep) >> 3); 929 NFS_SRTT(rep) += t1; 930 if (t1 < 0) 931 t1 = -t1; 932 t1 -= (NFS_SDRTT(rep) >> 2); 933 NFS_SDRTT(rep) += t1; 934 } 935 nmp->nm_timeouts = 0; 936 break; 937 } 938 } 939 nfs_rcvunlock(nmp); 940 /* 941 * If not matched to a request, drop it. 942 * If it's mine, get out. 943 */ 944 if (rep == 0) { 945 nfsstats.rpcunexpected++; 946 m_freem(mrep); 947 } else if (rep == myrep) { 948 if (rep->r_mrep == NULL) 949 panic("nfsreply nil"); 950 return (0); 951 } 952 } 953 } 954 955 /* 956 * nfs_request - goes something like this 957 * - fill in request struct 958 * - links it into list 959 * - calls nfs_send() for first transmit 960 * - calls nfs_receive() to get reply 961 * - break down rpc header and return with nfs reply pointed to 962 * by mrep or error 963 * nb: always frees up mreq mbuf list 964 */ 965 int 966 nfs_request(np, mrest, procnum, lwp, cred, mrp, mdp, dposp, rexmitp) 967 struct nfsnode *np; 968 struct mbuf *mrest; 969 int procnum; 970 struct lwp *lwp; 971 kauth_cred_t cred; 972 struct mbuf **mrp; 973 struct mbuf **mdp; 974 char **dposp; 975 int *rexmitp; 976 { 977 struct mbuf *m, *mrep; 978 struct nfsreq *rep; 979 u_int32_t *tl; 980 int i; 981 struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount); 982 struct mbuf *md, *mheadend; 983 char nickv[RPCX_NICKVERF]; 984 time_t waituntil; 985 char *dpos, *cp2; 986 int t1, s, error = 0, mrest_len, auth_len, auth_type; 987 int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0; 988 int verf_len, verf_type; 989 u_int32_t xid; 990 char *auth_str, *verf_str; 991 NFSKERBKEY_T key; /* save session key */ 992 kauth_cred_t acred; 993 struct mbuf *mrest_backup = NULL; 994 kauth_cred_t origcred = NULL; /* XXX: gcc */ 995 bool retry_cred = true; 996 bool use_opencred = (np->n_flag & NUSEOPENCRED) != 0; 997 998 if (rexmitp != NULL) 999 *rexmitp = 0; 1000 1001 acred = kauth_cred_alloc(); 1002 1003 tryagain_cred: 1004 KASSERT(cred != NULL); 1005 rep = kmem_alloc(sizeof(*rep), KM_SLEEP); 1006 rep->r_nmp = nmp; 1007 KASSERT(lwp == NULL || lwp == curlwp); 1008 rep->r_lwp = lwp; 1009 rep->r_procnum = procnum; 1010 i = 0; 1011 m = mrest; 1012 while (m) { 1013 i += m->m_len; 1014 m = m->m_next; 1015 } 1016 mrest_len = i; 1017 1018 /* 1019 * Get the RPC header with authorization. 1020 */ 1021 kerbauth: 1022 verf_str = auth_str = (char *)0; 1023 if (nmp->nm_flag & NFSMNT_KERB) { 1024 verf_str = nickv; 1025 verf_len = sizeof (nickv); 1026 auth_type = RPCAUTH_KERB4; 1027 memset((void *)key, 0, sizeof (key)); 1028 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 1029 &auth_len, verf_str, verf_len)) { 1030 error = nfs_getauth(nmp, rep, cred, &auth_str, 1031 &auth_len, verf_str, &verf_len, key); 1032 if (error) { 1033 kmem_free(rep, sizeof(*rep)); 1034 m_freem(mrest); 1035 KASSERT(kauth_cred_getrefcnt(acred) == 1); 1036 kauth_cred_free(acred); 1037 return (error); 1038 } 1039 } 1040 retry_cred = false; 1041 } else { 1042 /* AUTH_UNIX */ 1043 uid_t uid; 1044 gid_t gid; 1045 1046 /* 1047 * on the most unix filesystems, permission checks are 1048 * done when the file is open(2)'ed. 1049 * ie. once a file is successfully open'ed, 1050 * following i/o operations never fail with EACCES. 1051 * we try to follow the semantics as far as possible. 1052 * 1053 * note that we expect that the nfs server always grant 1054 * accesses by the file's owner. 1055 */ 1056 origcred = cred; 1057 switch (procnum) { 1058 case NFSPROC_READ: 1059 case NFSPROC_WRITE: 1060 case NFSPROC_COMMIT: 1061 uid = np->n_vattr->va_uid; 1062 gid = np->n_vattr->va_gid; 1063 if (kauth_cred_geteuid(cred) == uid && 1064 kauth_cred_getegid(cred) == gid) { 1065 retry_cred = false; 1066 break; 1067 } 1068 if (use_opencred) 1069 break; 1070 kauth_cred_setuid(acred, uid); 1071 kauth_cred_seteuid(acred, uid); 1072 kauth_cred_setsvuid(acred, uid); 1073 kauth_cred_setgid(acred, gid); 1074 kauth_cred_setegid(acred, gid); 1075 kauth_cred_setsvgid(acred, gid); 1076 cred = acred; 1077 break; 1078 default: 1079 retry_cred = false; 1080 break; 1081 } 1082 /* 1083 * backup mbuf chain if we can need it later to retry. 1084 * 1085 * XXX maybe we can keep a direct reference to 1086 * mrest without doing m_copym, but it's ...ugly. 1087 */ 1088 if (retry_cred) 1089 mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT); 1090 auth_type = RPCAUTH_UNIX; 1091 /* XXX elad - ngroups */ 1092 auth_len = (((kauth_cred_ngroups(cred) > nmp->nm_numgrps) ? 1093 nmp->nm_numgrps : kauth_cred_ngroups(cred)) << 2) + 1094 5 * NFSX_UNSIGNED; 1095 } 1096 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 1097 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 1098 if (auth_str) 1099 free(auth_str, M_TEMP); 1100 1101 /* 1102 * For stream protocols, insert a Sun RPC Record Mark. 1103 */ 1104 if (nmp->nm_sotype == SOCK_STREAM) { 1105 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 1106 *mtod(m, u_int32_t *) = htonl(0x80000000 | 1107 (m->m_pkthdr.len - NFSX_UNSIGNED)); 1108 } 1109 rep->r_mreq = m; 1110 rep->r_xid = xid; 1111 tryagain: 1112 if (nmp->nm_flag & NFSMNT_SOFT) 1113 rep->r_retry = nmp->nm_retry; 1114 else 1115 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1116 rep->r_rtt = rep->r_rexmit = 0; 1117 if (proct[procnum] > 0) 1118 rep->r_flags = R_TIMING; 1119 else 1120 rep->r_flags = 0; 1121 rep->r_mrep = NULL; 1122 1123 /* 1124 * Do the client side RPC. 1125 */ 1126 nfsstats.rpcrequests++; 1127 /* 1128 * Chain request into list of outstanding requests. Be sure 1129 * to put it LAST so timer finds oldest requests first. 1130 */ 1131 s = splsoftnet(); 1132 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1133 nfs_timer_start(); 1134 1135 /* 1136 * If backing off another request or avoiding congestion, don't 1137 * send this one now but let timer do it. If not timing a request, 1138 * do it now. 1139 */ 1140 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1141 (nmp->nm_flag & NFSMNT_DUMBTIMR) || nmp->nm_sent < nmp->nm_cwnd)) { 1142 splx(s); 1143 if (nmp->nm_soflags & PR_CONNREQUIRED) 1144 error = nfs_sndlock(nmp, rep); 1145 if (!error) { 1146 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1147 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp); 1148 if (nmp->nm_soflags & PR_CONNREQUIRED) 1149 nfs_sndunlock(nmp); 1150 } 1151 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1152 nmp->nm_sent += NFS_CWNDSCALE; 1153 rep->r_flags |= R_SENT; 1154 } 1155 } else { 1156 splx(s); 1157 rep->r_rtt = -1; 1158 } 1159 1160 /* 1161 * Wait for the reply from our send or the timer's. 1162 */ 1163 if (!error || error == EPIPE) 1164 error = nfs_reply(rep, lwp); 1165 1166 /* 1167 * RPC done, unlink the request. 1168 */ 1169 s = splsoftnet(); 1170 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1171 splx(s); 1172 1173 /* 1174 * Decrement the outstanding request count. 1175 */ 1176 if (rep->r_flags & R_SENT) { 1177 rep->r_flags &= ~R_SENT; /* paranoia */ 1178 nmp->nm_sent -= NFS_CWNDSCALE; 1179 } 1180 1181 if (rexmitp != NULL) { 1182 int rexmit; 1183 1184 if (nmp->nm_sotype != SOCK_DGRAM) 1185 rexmit = (rep->r_flags & R_REXMITTED) != 0; 1186 else 1187 rexmit = rep->r_rexmit; 1188 *rexmitp = rexmit; 1189 } 1190 1191 /* 1192 * If there was a successful reply and a tprintf msg. 1193 * tprintf a response. 1194 */ 1195 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1196 nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1197 "is alive again"); 1198 mrep = rep->r_mrep; 1199 md = rep->r_md; 1200 dpos = rep->r_dpos; 1201 if (error) 1202 goto nfsmout; 1203 1204 /* 1205 * break down the rpc header and check if ok 1206 */ 1207 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1208 if (*tl++ == rpc_msgdenied) { 1209 if (*tl == rpc_mismatch) 1210 error = EOPNOTSUPP; 1211 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1212 if (!failed_auth) { 1213 failed_auth++; 1214 mheadend->m_next = (struct mbuf *)0; 1215 m_freem(mrep); 1216 m_freem(rep->r_mreq); 1217 goto kerbauth; 1218 } else 1219 error = EAUTH; 1220 } else 1221 error = EACCES; 1222 m_freem(mrep); 1223 goto nfsmout; 1224 } 1225 1226 /* 1227 * Grab any Kerberos verifier, otherwise just throw it away. 1228 */ 1229 verf_type = fxdr_unsigned(int, *tl++); 1230 i = fxdr_unsigned(int32_t, *tl); 1231 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1232 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1233 if (error) 1234 goto nfsmout; 1235 } else if (i > 0) 1236 nfsm_adv(nfsm_rndup(i)); 1237 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1238 /* 0 == ok */ 1239 if (*tl == 0) { 1240 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1241 if (*tl != 0) { 1242 error = fxdr_unsigned(int, *tl); 1243 switch (error) { 1244 case NFSERR_PERM: 1245 error = EPERM; 1246 break; 1247 1248 case NFSERR_NOENT: 1249 error = ENOENT; 1250 break; 1251 1252 case NFSERR_IO: 1253 error = EIO; 1254 break; 1255 1256 case NFSERR_NXIO: 1257 error = ENXIO; 1258 break; 1259 1260 case NFSERR_ACCES: 1261 error = EACCES; 1262 if (!retry_cred) 1263 break; 1264 m_freem(mrep); 1265 m_freem(rep->r_mreq); 1266 kmem_free(rep, sizeof(*rep)); 1267 use_opencred = !use_opencred; 1268 if (mrest_backup == NULL) { 1269 /* m_copym failure */ 1270 KASSERT( 1271 kauth_cred_getrefcnt(acred) == 1); 1272 kauth_cred_free(acred); 1273 return ENOMEM; 1274 } 1275 mrest = mrest_backup; 1276 mrest_backup = NULL; 1277 cred = origcred; 1278 error = 0; 1279 retry_cred = false; 1280 goto tryagain_cred; 1281 1282 case NFSERR_EXIST: 1283 error = EEXIST; 1284 break; 1285 1286 case NFSERR_XDEV: 1287 error = EXDEV; 1288 break; 1289 1290 case NFSERR_NODEV: 1291 error = ENODEV; 1292 break; 1293 1294 case NFSERR_NOTDIR: 1295 error = ENOTDIR; 1296 break; 1297 1298 case NFSERR_ISDIR: 1299 error = EISDIR; 1300 break; 1301 1302 case NFSERR_INVAL: 1303 error = EINVAL; 1304 break; 1305 1306 case NFSERR_FBIG: 1307 error = EFBIG; 1308 break; 1309 1310 case NFSERR_NOSPC: 1311 error = ENOSPC; 1312 break; 1313 1314 case NFSERR_ROFS: 1315 error = EROFS; 1316 break; 1317 1318 case NFSERR_MLINK: 1319 error = EMLINK; 1320 break; 1321 1322 case NFSERR_TIMEDOUT: 1323 error = ETIMEDOUT; 1324 break; 1325 1326 case NFSERR_NAMETOL: 1327 error = ENAMETOOLONG; 1328 break; 1329 1330 case NFSERR_NOTEMPTY: 1331 error = ENOTEMPTY; 1332 break; 1333 1334 case NFSERR_DQUOT: 1335 error = EDQUOT; 1336 break; 1337 1338 case NFSERR_STALE: 1339 /* 1340 * If the File Handle was stale, invalidate the 1341 * lookup cache, just in case. 1342 */ 1343 error = ESTALE; 1344 cache_purge(NFSTOV(np)); 1345 break; 1346 1347 case NFSERR_REMOTE: 1348 error = EREMOTE; 1349 break; 1350 1351 case NFSERR_WFLUSH: 1352 case NFSERR_BADHANDLE: 1353 case NFSERR_NOT_SYNC: 1354 case NFSERR_BAD_COOKIE: 1355 error = EINVAL; 1356 break; 1357 1358 case NFSERR_NOTSUPP: 1359 error = ENOTSUP; 1360 break; 1361 1362 case NFSERR_TOOSMALL: 1363 case NFSERR_SERVERFAULT: 1364 case NFSERR_BADTYPE: 1365 error = EINVAL; 1366 break; 1367 1368 case NFSERR_TRYLATER: 1369 if ((nmp->nm_flag & NFSMNT_NFSV3) == 0) 1370 break; 1371 m_freem(mrep); 1372 error = 0; 1373 waituntil = time_second + trylater_delay; 1374 while (time_second < waituntil) { 1375 kpause("nfstrylater", false, hz, NULL); 1376 } 1377 trylater_delay *= NFS_TRYLATERDELMUL; 1378 if (trylater_delay > NFS_TRYLATERDELMAX) 1379 trylater_delay = NFS_TRYLATERDELMAX; 1380 /* 1381 * RFC1813: 1382 * The client should wait and then try 1383 * the request with a new RPC transaction ID. 1384 */ 1385 nfs_renewxid(rep); 1386 goto tryagain; 1387 1388 default: 1389 #ifdef DIAGNOSTIC 1390 printf("Invalid rpc error code %d\n", error); 1391 #endif 1392 error = EINVAL; 1393 break; 1394 } 1395 1396 if (nmp->nm_flag & NFSMNT_NFSV3) { 1397 *mrp = mrep; 1398 *mdp = md; 1399 *dposp = dpos; 1400 error |= NFSERR_RETERR; 1401 } else 1402 m_freem(mrep); 1403 goto nfsmout; 1404 } 1405 1406 /* 1407 * note which credential worked to minimize number of retries. 1408 */ 1409 if (use_opencred) 1410 np->n_flag |= NUSEOPENCRED; 1411 else 1412 np->n_flag &= ~NUSEOPENCRED; 1413 1414 *mrp = mrep; 1415 *mdp = md; 1416 *dposp = dpos; 1417 1418 KASSERT(error == 0); 1419 goto nfsmout; 1420 } 1421 m_freem(mrep); 1422 error = EPROTONOSUPPORT; 1423 nfsmout: 1424 KASSERT(kauth_cred_getrefcnt(acred) == 1); 1425 kauth_cred_free(acred); 1426 m_freem(rep->r_mreq); 1427 kmem_free(rep, sizeof(*rep)); 1428 m_freem(mrest_backup); 1429 return (error); 1430 } 1431 #endif /* NFS */ 1432 1433 /* 1434 * Generate the rpc reply header 1435 * siz arg. is used to decide if adding a cluster is worthwhile 1436 */ 1437 int 1438 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1439 int siz; 1440 struct nfsrv_descript *nd; 1441 struct nfssvc_sock *slp; 1442 int err; 1443 int cache; 1444 u_quad_t *frev; 1445 struct mbuf **mrq; 1446 struct mbuf **mbp; 1447 char **bposp; 1448 { 1449 u_int32_t *tl; 1450 struct mbuf *mreq; 1451 char *bpos; 1452 struct mbuf *mb; 1453 1454 mreq = m_gethdr(M_WAIT, MT_DATA); 1455 MCLAIM(mreq, &nfs_mowner); 1456 mb = mreq; 1457 /* 1458 * If this is a big reply, use a cluster else 1459 * try and leave leading space for the lower level headers. 1460 */ 1461 siz += RPC_REPLYSIZ; 1462 if (siz >= max_datalen) { 1463 m_clget(mreq, M_WAIT); 1464 } else 1465 mreq->m_data += max_hdr; 1466 tl = mtod(mreq, u_int32_t *); 1467 mreq->m_len = 6 * NFSX_UNSIGNED; 1468 bpos = ((char *)tl) + mreq->m_len; 1469 *tl++ = txdr_unsigned(nd->nd_retxid); 1470 *tl++ = rpc_reply; 1471 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1472 *tl++ = rpc_msgdenied; 1473 if (err & NFSERR_AUTHERR) { 1474 *tl++ = rpc_autherr; 1475 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1476 mreq->m_len -= NFSX_UNSIGNED; 1477 bpos -= NFSX_UNSIGNED; 1478 } else { 1479 *tl++ = rpc_mismatch; 1480 *tl++ = txdr_unsigned(RPC_VER2); 1481 *tl = txdr_unsigned(RPC_VER2); 1482 } 1483 } else { 1484 *tl++ = rpc_msgaccepted; 1485 1486 /* 1487 * For Kerberos authentication, we must send the nickname 1488 * verifier back, otherwise just RPCAUTH_NULL. 1489 */ 1490 if (nd->nd_flag & ND_KERBFULL) { 1491 struct nfsuid *nuidp; 1492 struct timeval ktvin, ktvout; 1493 1494 memset(&ktvout, 0, sizeof ktvout); /* XXX gcc */ 1495 1496 LIST_FOREACH(nuidp, 1497 NUIDHASH(slp, kauth_cred_geteuid(nd->nd_cr)), 1498 nu_hash) { 1499 if (kauth_cred_geteuid(nuidp->nu_cr) == 1500 kauth_cred_geteuid(nd->nd_cr) && 1501 (!nd->nd_nam2 || netaddr_match( 1502 NU_NETFAM(nuidp), &nuidp->nu_haddr, 1503 nd->nd_nam2))) 1504 break; 1505 } 1506 if (nuidp) { 1507 ktvin.tv_sec = 1508 txdr_unsigned(nuidp->nu_timestamp.tv_sec 1509 - 1); 1510 ktvin.tv_usec = 1511 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1512 1513 /* 1514 * Encrypt the timestamp in ecb mode using the 1515 * session key. 1516 */ 1517 #ifdef NFSKERB 1518 XXX 1519 #endif 1520 1521 *tl++ = rpc_auth_kerb; 1522 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1523 *tl = ktvout.tv_sec; 1524 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1525 *tl++ = ktvout.tv_usec; 1526 *tl++ = txdr_unsigned( 1527 kauth_cred_geteuid(nuidp->nu_cr)); 1528 } else { 1529 *tl++ = 0; 1530 *tl++ = 0; 1531 } 1532 } else { 1533 *tl++ = 0; 1534 *tl++ = 0; 1535 } 1536 switch (err) { 1537 case EPROGUNAVAIL: 1538 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1539 break; 1540 case EPROGMISMATCH: 1541 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1542 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1543 *tl++ = txdr_unsigned(2); 1544 *tl = txdr_unsigned(3); 1545 break; 1546 case EPROCUNAVAIL: 1547 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1548 break; 1549 case EBADRPC: 1550 *tl = txdr_unsigned(RPC_GARBAGE); 1551 break; 1552 default: 1553 *tl = 0; 1554 if (err != NFSERR_RETVOID) { 1555 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1556 if (err) 1557 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1558 else 1559 *tl = 0; 1560 } 1561 break; 1562 }; 1563 } 1564 1565 if (mrq != NULL) 1566 *mrq = mreq; 1567 *mbp = mb; 1568 *bposp = bpos; 1569 if (err != 0 && err != NFSERR_RETVOID) 1570 nfsstats.srvrpc_errs++; 1571 return (0); 1572 } 1573 1574 static void 1575 nfs_timer_schedule(void) 1576 { 1577 1578 callout_schedule(&nfs_timer_ch, nfs_ticks); 1579 } 1580 1581 void 1582 nfs_timer_start(void) 1583 { 1584 1585 if (callout_pending(&nfs_timer_ch)) 1586 return; 1587 1588 nfs_timer_start_ev.ev_count++; 1589 nfs_timer_schedule(); 1590 } 1591 1592 void 1593 nfs_timer_init(void) 1594 { 1595 1596 callout_init(&nfs_timer_ch, 0); 1597 callout_setfunc(&nfs_timer_ch, nfs_timer, NULL); 1598 evcnt_attach_dynamic(&nfs_timer_ev, EVCNT_TYPE_MISC, NULL, 1599 "nfs", "timer"); 1600 evcnt_attach_dynamic(&nfs_timer_start_ev, EVCNT_TYPE_MISC, NULL, 1601 "nfs", "timer start"); 1602 evcnt_attach_dynamic(&nfs_timer_stop_ev, EVCNT_TYPE_MISC, NULL, 1603 "nfs", "timer stop"); 1604 } 1605 1606 /* 1607 * Nfs timer routine 1608 * Scan the nfsreq list and retranmit any requests that have timed out 1609 * To avoid retransmission attempts on STREAM sockets (in the future) make 1610 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1611 * A non-NULL argument means 'initialize'. 1612 */ 1613 void 1614 nfs_timer(void *arg) 1615 { 1616 struct nfsreq *rep; 1617 struct mbuf *m; 1618 struct socket *so; 1619 struct nfsmount *nmp; 1620 int timeo; 1621 int s, error; 1622 bool more = false; 1623 #ifdef NFSSERVER 1624 struct timeval tv; 1625 struct nfssvc_sock *slp; 1626 u_quad_t cur_usec; 1627 #endif 1628 1629 nfs_timer_ev.ev_count++; 1630 1631 s = splsoftnet(); 1632 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1633 more = true; 1634 nmp = rep->r_nmp; 1635 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1636 continue; 1637 if (nfs_sigintr(nmp, rep, rep->r_lwp)) { 1638 rep->r_flags |= R_SOFTTERM; 1639 continue; 1640 } 1641 if (rep->r_rtt >= 0) { 1642 rep->r_rtt++; 1643 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1644 timeo = nmp->nm_timeo; 1645 else 1646 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1647 if (nmp->nm_timeouts > 0) 1648 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1649 if (rep->r_rtt <= timeo) 1650 continue; 1651 if (nmp->nm_timeouts < 1652 (sizeof(nfs_backoff) / sizeof(nfs_backoff[0]))) 1653 nmp->nm_timeouts++; 1654 } 1655 /* 1656 * Check for server not responding 1657 */ 1658 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1659 rep->r_rexmit > nmp->nm_deadthresh) { 1660 nfs_msg(rep->r_lwp, 1661 nmp->nm_mountp->mnt_stat.f_mntfromname, 1662 "not responding"); 1663 rep->r_flags |= R_TPRINTFMSG; 1664 } 1665 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1666 nfsstats.rpctimeouts++; 1667 rep->r_flags |= R_SOFTTERM; 1668 continue; 1669 } 1670 if (nmp->nm_sotype != SOCK_DGRAM) { 1671 if (++rep->r_rexmit > NFS_MAXREXMIT) 1672 rep->r_rexmit = NFS_MAXREXMIT; 1673 continue; 1674 } 1675 if ((so = nmp->nm_so) == NULL) 1676 continue; 1677 1678 /* 1679 * If there is enough space and the window allows.. 1680 * Resend it 1681 * Set r_rtt to -1 in case we fail to send it now. 1682 */ 1683 solock(so); 1684 rep->r_rtt = -1; 1685 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1686 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1687 (rep->r_flags & R_SENT) || 1688 nmp->nm_sent < nmp->nm_cwnd) && 1689 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1690 if (so->so_state & SS_ISCONNECTED) 1691 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1692 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1693 else 1694 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1695 nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0); 1696 if (error) { 1697 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1698 #ifdef DEBUG 1699 printf("nfs_timer: ignoring error %d\n", 1700 error); 1701 #endif 1702 so->so_error = 0; 1703 } 1704 } else { 1705 /* 1706 * Iff first send, start timing 1707 * else turn timing off, backoff timer 1708 * and divide congestion window by 2. 1709 */ 1710 if (rep->r_flags & R_SENT) { 1711 rep->r_flags &= ~R_TIMING; 1712 if (++rep->r_rexmit > NFS_MAXREXMIT) 1713 rep->r_rexmit = NFS_MAXREXMIT; 1714 nmp->nm_cwnd >>= 1; 1715 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1716 nmp->nm_cwnd = NFS_CWNDSCALE; 1717 nfsstats.rpcretries++; 1718 } else { 1719 rep->r_flags |= R_SENT; 1720 nmp->nm_sent += NFS_CWNDSCALE; 1721 } 1722 rep->r_rtt = 0; 1723 } 1724 } 1725 sounlock(so); 1726 } 1727 splx(s); 1728 1729 #ifdef NFSSERVER 1730 /* 1731 * Scan the write gathering queues for writes that need to be 1732 * completed now. 1733 */ 1734 getmicrotime(&tv); 1735 cur_usec = (u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec; 1736 mutex_enter(&nfsd_lock); 1737 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1738 struct nfsrv_descript *nd; 1739 1740 nd = LIST_FIRST(&slp->ns_tq); 1741 if (nd != NULL) { 1742 if (nd->nd_time <= cur_usec) { 1743 nfsrv_wakenfsd_locked(slp); 1744 } 1745 more = true; 1746 } 1747 } 1748 mutex_exit(&nfsd_lock); 1749 #endif /* NFSSERVER */ 1750 if (more) { 1751 nfs_timer_schedule(); 1752 } else { 1753 nfs_timer_stop_ev.ev_count++; 1754 } 1755 } 1756 1757 /* 1758 * Test for a termination condition pending on the process. 1759 * This is used for NFSMNT_INT mounts. 1760 */ 1761 int 1762 nfs_sigintr(nmp, rep, l) 1763 struct nfsmount *nmp; 1764 struct nfsreq *rep; 1765 struct lwp *l; 1766 { 1767 sigset_t ss; 1768 1769 if (rep && (rep->r_flags & R_SOFTTERM)) 1770 return (EINTR); 1771 if (!(nmp->nm_flag & NFSMNT_INT)) 1772 return (0); 1773 if (l) { 1774 sigpending1(l, &ss); 1775 #if 0 1776 sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss); 1777 #endif 1778 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1779 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1780 sigismember(&ss, SIGQUIT)) 1781 return (EINTR); 1782 } 1783 return (0); 1784 } 1785 1786 /* 1787 * Lock a socket against others. 1788 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1789 * and also to avoid race conditions between the processes with nfs requests 1790 * in progress when a reconnect is necessary. 1791 */ 1792 static int 1793 nfs_sndlock(struct nfsmount *nmp, struct nfsreq *rep) 1794 { 1795 struct lwp *l; 1796 int timeo = 0; 1797 bool catch = false; 1798 int error = 0; 1799 1800 if (rep) { 1801 l = rep->r_lwp; 1802 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1803 catch = true; 1804 } else 1805 l = NULL; 1806 mutex_enter(&nmp->nm_lock); 1807 while ((nmp->nm_iflag & NFSMNT_SNDLOCK) != 0) { 1808 if (rep && nfs_sigintr(rep->r_nmp, rep, l)) { 1809 error = EINTR; 1810 goto quit; 1811 } 1812 if (catch) { 1813 cv_timedwait_sig(&nmp->nm_sndcv, &nmp->nm_lock, timeo); 1814 } else { 1815 cv_timedwait(&nmp->nm_sndcv, &nmp->nm_lock, timeo); 1816 } 1817 if (catch) { 1818 catch = false; 1819 timeo = 2 * hz; 1820 } 1821 } 1822 nmp->nm_iflag |= NFSMNT_SNDLOCK; 1823 quit: 1824 mutex_exit(&nmp->nm_lock); 1825 return error; 1826 } 1827 1828 /* 1829 * Unlock the stream socket for others. 1830 */ 1831 static void 1832 nfs_sndunlock(struct nfsmount *nmp) 1833 { 1834 1835 mutex_enter(&nmp->nm_lock); 1836 if ((nmp->nm_iflag & NFSMNT_SNDLOCK) == 0) 1837 panic("nfs sndunlock"); 1838 nmp->nm_iflag &= ~NFSMNT_SNDLOCK; 1839 cv_signal(&nmp->nm_sndcv); 1840 mutex_exit(&nmp->nm_lock); 1841 } 1842 1843 static int 1844 nfs_rcvlock(struct nfsmount *nmp, struct nfsreq *rep) 1845 { 1846 int *flagp = &nmp->nm_iflag; 1847 int slptimeo = 0; 1848 bool catch; 1849 int error = 0; 1850 1851 KASSERT(nmp == rep->r_nmp); 1852 1853 catch = (nmp->nm_flag & NFSMNT_INT) != 0; 1854 mutex_enter(&nmp->nm_lock); 1855 while (/* CONSTCOND */ true) { 1856 if (*flagp & NFSMNT_DISMNT) { 1857 cv_signal(&nmp->nm_disconcv); 1858 error = EIO; 1859 break; 1860 } 1861 /* If our reply was received while we were sleeping, 1862 * then just return without taking the lock to avoid a 1863 * situation where a single iod could 'capture' the 1864 * receive lock. 1865 */ 1866 if (rep->r_mrep != NULL) { 1867 error = EALREADY; 1868 break; 1869 } 1870 if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) { 1871 error = EINTR; 1872 break; 1873 } 1874 if ((*flagp & NFSMNT_RCVLOCK) == 0) { 1875 *flagp |= NFSMNT_RCVLOCK; 1876 break; 1877 } 1878 if (catch) { 1879 cv_timedwait_sig(&nmp->nm_rcvcv, &nmp->nm_lock, 1880 slptimeo); 1881 } else { 1882 cv_timedwait(&nmp->nm_rcvcv, &nmp->nm_lock, 1883 slptimeo); 1884 } 1885 if (catch) { 1886 catch = false; 1887 slptimeo = 2 * hz; 1888 } 1889 } 1890 mutex_exit(&nmp->nm_lock); 1891 return error; 1892 } 1893 1894 /* 1895 * Unlock the stream socket for others. 1896 */ 1897 static void 1898 nfs_rcvunlock(struct nfsmount *nmp) 1899 { 1900 1901 mutex_enter(&nmp->nm_lock); 1902 if ((nmp->nm_iflag & NFSMNT_RCVLOCK) == 0) 1903 panic("nfs rcvunlock"); 1904 nmp->nm_iflag &= ~NFSMNT_RCVLOCK; 1905 cv_broadcast(&nmp->nm_rcvcv); 1906 mutex_exit(&nmp->nm_lock); 1907 } 1908 1909 /* 1910 * Parse an RPC request 1911 * - verify it 1912 * - allocate and fill in the cred. 1913 */ 1914 int 1915 nfs_getreq(nd, nfsd, has_header) 1916 struct nfsrv_descript *nd; 1917 struct nfsd *nfsd; 1918 int has_header; 1919 { 1920 int len, i; 1921 u_int32_t *tl; 1922 int32_t t1; 1923 struct uio uio; 1924 struct iovec iov; 1925 char *dpos, *cp2, *cp; 1926 u_int32_t nfsvers, auth_type; 1927 uid_t nickuid; 1928 int error = 0, ticklen; 1929 struct mbuf *mrep, *md; 1930 struct nfsuid *nuidp; 1931 struct timeval tvin, tvout; 1932 1933 memset(&tvout, 0, sizeof tvout); /* XXX gcc */ 1934 1935 KASSERT(nd->nd_cr == NULL); 1936 mrep = nd->nd_mrep; 1937 md = nd->nd_md; 1938 dpos = nd->nd_dpos; 1939 if (has_header) { 1940 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1941 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1942 if (*tl++ != rpc_call) { 1943 m_freem(mrep); 1944 return (EBADRPC); 1945 } 1946 } else 1947 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1948 nd->nd_repstat = 0; 1949 nd->nd_flag = 0; 1950 if (*tl++ != rpc_vers) { 1951 nd->nd_repstat = ERPCMISMATCH; 1952 nd->nd_procnum = NFSPROC_NOOP; 1953 return (0); 1954 } 1955 if (*tl != nfs_prog) { 1956 nd->nd_repstat = EPROGUNAVAIL; 1957 nd->nd_procnum = NFSPROC_NOOP; 1958 return (0); 1959 } 1960 tl++; 1961 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1962 if (nfsvers < NFS_VER2 || nfsvers > NFS_VER3) { 1963 nd->nd_repstat = EPROGMISMATCH; 1964 nd->nd_procnum = NFSPROC_NOOP; 1965 return (0); 1966 } 1967 if (nfsvers == NFS_VER3) 1968 nd->nd_flag = ND_NFSV3; 1969 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1970 if (nd->nd_procnum == NFSPROC_NULL) 1971 return (0); 1972 if (nd->nd_procnum > NFSPROC_COMMIT || 1973 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 1974 nd->nd_repstat = EPROCUNAVAIL; 1975 nd->nd_procnum = NFSPROC_NOOP; 1976 return (0); 1977 } 1978 if ((nd->nd_flag & ND_NFSV3) == 0) 1979 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 1980 auth_type = *tl++; 1981 len = fxdr_unsigned(int, *tl++); 1982 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1983 m_freem(mrep); 1984 return (EBADRPC); 1985 } 1986 1987 nd->nd_flag &= ~ND_KERBAUTH; 1988 /* 1989 * Handle auth_unix or auth_kerb. 1990 */ 1991 if (auth_type == rpc_auth_unix) { 1992 uid_t uid; 1993 gid_t gid; 1994 1995 nd->nd_cr = kauth_cred_alloc(); 1996 len = fxdr_unsigned(int, *++tl); 1997 if (len < 0 || len > NFS_MAXNAMLEN) { 1998 m_freem(mrep); 1999 error = EBADRPC; 2000 goto errout; 2001 } 2002 nfsm_adv(nfsm_rndup(len)); 2003 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2004 2005 uid = fxdr_unsigned(uid_t, *tl++); 2006 gid = fxdr_unsigned(gid_t, *tl++); 2007 kauth_cred_setuid(nd->nd_cr, uid); 2008 kauth_cred_seteuid(nd->nd_cr, uid); 2009 kauth_cred_setsvuid(nd->nd_cr, uid); 2010 kauth_cred_setgid(nd->nd_cr, gid); 2011 kauth_cred_setegid(nd->nd_cr, gid); 2012 kauth_cred_setsvgid(nd->nd_cr, gid); 2013 2014 len = fxdr_unsigned(int, *tl); 2015 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 2016 m_freem(mrep); 2017 error = EBADRPC; 2018 goto errout; 2019 } 2020 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 2021 2022 if (len > 0) { 2023 size_t grbuf_size = min(len, NGROUPS) * sizeof(gid_t); 2024 gid_t *grbuf = kmem_alloc(grbuf_size, KM_SLEEP); 2025 2026 for (i = 0; i < len; i++) { 2027 if (i < NGROUPS) /* XXX elad */ 2028 grbuf[i] = fxdr_unsigned(gid_t, *tl++); 2029 else 2030 tl++; 2031 } 2032 kauth_cred_setgroups(nd->nd_cr, grbuf, 2033 min(len, NGROUPS), -1, UIO_SYSSPACE); 2034 kmem_free(grbuf, grbuf_size); 2035 } 2036 2037 len = fxdr_unsigned(int, *++tl); 2038 if (len < 0 || len > RPCAUTH_MAXSIZ) { 2039 m_freem(mrep); 2040 error = EBADRPC; 2041 goto errout; 2042 } 2043 if (len > 0) 2044 nfsm_adv(nfsm_rndup(len)); 2045 } else if (auth_type == rpc_auth_kerb) { 2046 switch (fxdr_unsigned(int, *tl++)) { 2047 case RPCAKN_FULLNAME: 2048 ticklen = fxdr_unsigned(int, *tl); 2049 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 2050 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 2051 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 2052 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 2053 m_freem(mrep); 2054 error = EBADRPC; 2055 goto errout; 2056 } 2057 uio.uio_offset = 0; 2058 uio.uio_iov = &iov; 2059 uio.uio_iovcnt = 1; 2060 UIO_SETUP_SYSSPACE(&uio); 2061 iov.iov_base = (void *)&nfsd->nfsd_authstr[4]; 2062 iov.iov_len = RPCAUTH_MAXSIZ - 4; 2063 nfsm_mtouio(&uio, uio.uio_resid); 2064 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2065 if (*tl++ != rpc_auth_kerb || 2066 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 2067 printf("Bad kerb verifier\n"); 2068 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2069 nd->nd_procnum = NFSPROC_NOOP; 2070 return (0); 2071 } 2072 nfsm_dissect(cp, void *, 4 * NFSX_UNSIGNED); 2073 tl = (u_int32_t *)cp; 2074 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 2075 printf("Not fullname kerb verifier\n"); 2076 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2077 nd->nd_procnum = NFSPROC_NOOP; 2078 return (0); 2079 } 2080 cp += NFSX_UNSIGNED; 2081 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 2082 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 2083 nd->nd_flag |= ND_KERBFULL; 2084 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 2085 break; 2086 case RPCAKN_NICKNAME: 2087 if (len != 2 * NFSX_UNSIGNED) { 2088 printf("Kerb nickname short\n"); 2089 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 2090 nd->nd_procnum = NFSPROC_NOOP; 2091 return (0); 2092 } 2093 nickuid = fxdr_unsigned(uid_t, *tl); 2094 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2095 if (*tl++ != rpc_auth_kerb || 2096 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 2097 printf("Kerb nick verifier bad\n"); 2098 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2099 nd->nd_procnum = NFSPROC_NOOP; 2100 return (0); 2101 } 2102 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2103 tvin.tv_sec = *tl++; 2104 tvin.tv_usec = *tl; 2105 2106 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), 2107 nu_hash) { 2108 if (kauth_cred_geteuid(nuidp->nu_cr) == nickuid && 2109 (!nd->nd_nam2 || 2110 netaddr_match(NU_NETFAM(nuidp), 2111 &nuidp->nu_haddr, nd->nd_nam2))) 2112 break; 2113 } 2114 if (!nuidp) { 2115 nd->nd_repstat = 2116 (NFSERR_AUTHERR|AUTH_REJECTCRED); 2117 nd->nd_procnum = NFSPROC_NOOP; 2118 return (0); 2119 } 2120 2121 /* 2122 * Now, decrypt the timestamp using the session key 2123 * and validate it. 2124 */ 2125 #ifdef NFSKERB 2126 XXX 2127 #endif 2128 2129 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 2130 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 2131 if (nuidp->nu_expire < time_second || 2132 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 2133 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 2134 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 2135 nuidp->nu_expire = 0; 2136 nd->nd_repstat = 2137 (NFSERR_AUTHERR|AUTH_REJECTVERF); 2138 nd->nd_procnum = NFSPROC_NOOP; 2139 return (0); 2140 } 2141 kauth_cred_hold(nuidp->nu_cr); 2142 nd->nd_cr = nuidp->nu_cr; 2143 nd->nd_flag |= ND_KERBNICK; 2144 } 2145 } else { 2146 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 2147 nd->nd_procnum = NFSPROC_NOOP; 2148 return (0); 2149 } 2150 2151 nd->nd_md = md; 2152 nd->nd_dpos = dpos; 2153 KASSERT((nd->nd_cr == NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) != 0) 2154 || (nd->nd_cr != NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) == 0)); 2155 return (0); 2156 nfsmout: 2157 errout: 2158 KASSERT(error != 0); 2159 if (nd->nd_cr != NULL) { 2160 kauth_cred_free(nd->nd_cr); 2161 nd->nd_cr = NULL; 2162 } 2163 return (error); 2164 } 2165 2166 int 2167 nfs_msg(l, server, msg) 2168 struct lwp *l; 2169 const char *server, *msg; 2170 { 2171 tpr_t tpr; 2172 2173 if (l) 2174 tpr = tprintf_open(l->l_proc); 2175 else 2176 tpr = NULL; 2177 tprintf(tpr, "nfs server %s: %s\n", server, msg); 2178 tprintf_close(tpr); 2179 return (0); 2180 } 2181 2182 #ifdef NFSSERVER 2183 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 2184 struct nfssvc_sock *, struct lwp *, 2185 struct mbuf **)) = { 2186 nfsrv_null, 2187 nfsrv_getattr, 2188 nfsrv_setattr, 2189 nfsrv_lookup, 2190 nfsrv3_access, 2191 nfsrv_readlink, 2192 nfsrv_read, 2193 nfsrv_write, 2194 nfsrv_create, 2195 nfsrv_mkdir, 2196 nfsrv_symlink, 2197 nfsrv_mknod, 2198 nfsrv_remove, 2199 nfsrv_rmdir, 2200 nfsrv_rename, 2201 nfsrv_link, 2202 nfsrv_readdir, 2203 nfsrv_readdirplus, 2204 nfsrv_statfs, 2205 nfsrv_fsinfo, 2206 nfsrv_pathconf, 2207 nfsrv_commit, 2208 nfsrv_noop 2209 }; 2210 2211 /* 2212 * Socket upcall routine for the nfsd sockets. 2213 * The void *arg is a pointer to the "struct nfssvc_sock". 2214 */ 2215 void 2216 nfsrv_soupcall(struct socket *so, void *arg, int waitflag) 2217 { 2218 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 2219 2220 nfsdsock_setbits(slp, SLP_A_NEEDQ); 2221 nfsrv_wakenfsd(slp); 2222 } 2223 2224 void 2225 nfsrv_rcv(struct nfssvc_sock *slp) 2226 { 2227 struct socket *so; 2228 struct mbuf *m; 2229 struct mbuf *mp, *nam; 2230 struct uio auio; 2231 int flags; 2232 int error; 2233 int setflags = 0; 2234 2235 error = nfsdsock_lock(slp, true); 2236 if (error) { 2237 setflags |= SLP_A_NEEDQ; 2238 goto dorecs_unlocked; 2239 } 2240 2241 nfsdsock_clearbits(slp, SLP_A_NEEDQ); 2242 2243 so = slp->ns_so; 2244 if (so->so_type == SOCK_STREAM) { 2245 /* 2246 * Do soreceive(). 2247 */ 2248 auio.uio_resid = 1000000000; 2249 /* not need to setup uio_vmspace */ 2250 flags = MSG_DONTWAIT; 2251 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, &flags); 2252 if (error || mp == NULL) { 2253 if (error == EWOULDBLOCK) 2254 setflags |= SLP_A_NEEDQ; 2255 else 2256 setflags |= SLP_A_DISCONN; 2257 goto dorecs; 2258 } 2259 m = mp; 2260 m_claimm(m, &nfs_mowner); 2261 if (slp->ns_rawend) { 2262 slp->ns_rawend->m_next = m; 2263 slp->ns_cc += 1000000000 - auio.uio_resid; 2264 } else { 2265 slp->ns_raw = m; 2266 slp->ns_cc = 1000000000 - auio.uio_resid; 2267 } 2268 while (m->m_next) 2269 m = m->m_next; 2270 slp->ns_rawend = m; 2271 2272 /* 2273 * Now try and parse record(s) out of the raw stream data. 2274 */ 2275 error = nfsrv_getstream(slp, M_WAIT); 2276 if (error) { 2277 if (error == EPERM) 2278 setflags |= SLP_A_DISCONN; 2279 else 2280 setflags |= SLP_A_NEEDQ; 2281 } 2282 } else { 2283 do { 2284 auio.uio_resid = 1000000000; 2285 /* not need to setup uio_vmspace */ 2286 flags = MSG_DONTWAIT; 2287 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, 2288 &flags); 2289 if (mp) { 2290 if (nam) { 2291 m = nam; 2292 m->m_next = mp; 2293 } else 2294 m = mp; 2295 m_claimm(m, &nfs_mowner); 2296 if (slp->ns_recend) 2297 slp->ns_recend->m_nextpkt = m; 2298 else 2299 slp->ns_rec = m; 2300 slp->ns_recend = m; 2301 m->m_nextpkt = (struct mbuf *)0; 2302 } 2303 if (error) { 2304 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2305 && error != EWOULDBLOCK) { 2306 setflags |= SLP_A_DISCONN; 2307 goto dorecs; 2308 } 2309 } 2310 } while (mp); 2311 } 2312 dorecs: 2313 nfsdsock_unlock(slp); 2314 2315 dorecs_unlocked: 2316 if (setflags) { 2317 nfsdsock_setbits(slp, setflags); 2318 } 2319 } 2320 2321 int 2322 nfsdsock_lock(struct nfssvc_sock *slp, bool waitok) 2323 { 2324 2325 mutex_enter(&slp->ns_lock); 2326 while ((~slp->ns_flags & (SLP_BUSY|SLP_VALID)) == 0) { 2327 if (!waitok) { 2328 mutex_exit(&slp->ns_lock); 2329 return EWOULDBLOCK; 2330 } 2331 cv_wait(&slp->ns_cv, &slp->ns_lock); 2332 } 2333 if ((slp->ns_flags & SLP_VALID) == 0) { 2334 mutex_exit(&slp->ns_lock); 2335 return EINVAL; 2336 } 2337 KASSERT((slp->ns_flags & SLP_BUSY) == 0); 2338 slp->ns_flags |= SLP_BUSY; 2339 mutex_exit(&slp->ns_lock); 2340 2341 return 0; 2342 } 2343 2344 void 2345 nfsdsock_unlock(struct nfssvc_sock *slp) 2346 { 2347 2348 mutex_enter(&slp->ns_lock); 2349 KASSERT((slp->ns_flags & SLP_BUSY) != 0); 2350 cv_broadcast(&slp->ns_cv); 2351 slp->ns_flags &= ~SLP_BUSY; 2352 mutex_exit(&slp->ns_lock); 2353 } 2354 2355 int 2356 nfsdsock_drain(struct nfssvc_sock *slp) 2357 { 2358 int error = 0; 2359 2360 mutex_enter(&slp->ns_lock); 2361 if ((slp->ns_flags & SLP_VALID) == 0) { 2362 error = EINVAL; 2363 goto done; 2364 } 2365 slp->ns_flags &= ~SLP_VALID; 2366 while ((slp->ns_flags & SLP_BUSY) != 0) { 2367 cv_wait(&slp->ns_cv, &slp->ns_lock); 2368 } 2369 done: 2370 mutex_exit(&slp->ns_lock); 2371 2372 return error; 2373 } 2374 2375 /* 2376 * Try and extract an RPC request from the mbuf data list received on a 2377 * stream socket. The "waitflag" argument indicates whether or not it 2378 * can sleep. 2379 */ 2380 int 2381 nfsrv_getstream(slp, waitflag) 2382 struct nfssvc_sock *slp; 2383 int waitflag; 2384 { 2385 struct mbuf *m, **mpp; 2386 struct mbuf *recm; 2387 u_int32_t recmark; 2388 int error = 0; 2389 2390 KASSERT((slp->ns_flags & SLP_BUSY) != 0); 2391 for (;;) { 2392 if (slp->ns_reclen == 0) { 2393 if (slp->ns_cc < NFSX_UNSIGNED) { 2394 break; 2395 } 2396 m = slp->ns_raw; 2397 m_copydata(m, 0, NFSX_UNSIGNED, (void *)&recmark); 2398 m_adj(m, NFSX_UNSIGNED); 2399 slp->ns_cc -= NFSX_UNSIGNED; 2400 recmark = ntohl(recmark); 2401 slp->ns_reclen = recmark & ~0x80000000; 2402 if (recmark & 0x80000000) 2403 slp->ns_sflags |= SLP_S_LASTFRAG; 2404 else 2405 slp->ns_sflags &= ~SLP_S_LASTFRAG; 2406 if (slp->ns_reclen > NFS_MAXPACKET) { 2407 error = EPERM; 2408 break; 2409 } 2410 } 2411 2412 /* 2413 * Now get the record part. 2414 * 2415 * Note that slp->ns_reclen may be 0. Linux sometimes 2416 * generates 0-length records. 2417 */ 2418 if (slp->ns_cc == slp->ns_reclen) { 2419 recm = slp->ns_raw; 2420 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2421 slp->ns_cc = slp->ns_reclen = 0; 2422 } else if (slp->ns_cc > slp->ns_reclen) { 2423 recm = slp->ns_raw; 2424 m = m_split(recm, slp->ns_reclen, waitflag); 2425 if (m == NULL) { 2426 error = EWOULDBLOCK; 2427 break; 2428 } 2429 m_claimm(recm, &nfs_mowner); 2430 slp->ns_raw = m; 2431 if (m->m_next == NULL) 2432 slp->ns_rawend = m; 2433 slp->ns_cc -= slp->ns_reclen; 2434 slp->ns_reclen = 0; 2435 } else { 2436 break; 2437 } 2438 2439 /* 2440 * Accumulate the fragments into a record. 2441 */ 2442 mpp = &slp->ns_frag; 2443 while (*mpp) 2444 mpp = &((*mpp)->m_next); 2445 *mpp = recm; 2446 if (slp->ns_sflags & SLP_S_LASTFRAG) { 2447 if (slp->ns_recend) 2448 slp->ns_recend->m_nextpkt = slp->ns_frag; 2449 else 2450 slp->ns_rec = slp->ns_frag; 2451 slp->ns_recend = slp->ns_frag; 2452 slp->ns_frag = NULL; 2453 } 2454 } 2455 2456 return error; 2457 } 2458 2459 /* 2460 * Parse an RPC header. 2461 */ 2462 int 2463 nfsrv_dorec(struct nfssvc_sock *slp, struct nfsd *nfsd, 2464 struct nfsrv_descript **ndp, bool *more) 2465 { 2466 struct mbuf *m, *nam; 2467 struct nfsrv_descript *nd; 2468 int error; 2469 2470 *ndp = NULL; 2471 *more = false; 2472 2473 if (nfsdsock_lock(slp, true)) { 2474 return ENOBUFS; 2475 } 2476 m = slp->ns_rec; 2477 if (m == NULL) { 2478 nfsdsock_unlock(slp); 2479 return ENOBUFS; 2480 } 2481 slp->ns_rec = m->m_nextpkt; 2482 if (slp->ns_rec) { 2483 m->m_nextpkt = NULL; 2484 *more = true; 2485 } else { 2486 slp->ns_recend = NULL; 2487 } 2488 nfsdsock_unlock(slp); 2489 2490 if (m->m_type == MT_SONAME) { 2491 nam = m; 2492 m = m->m_next; 2493 nam->m_next = NULL; 2494 } else 2495 nam = NULL; 2496 nd = nfsdreq_alloc(); 2497 nd->nd_md = nd->nd_mrep = m; 2498 nd->nd_nam2 = nam; 2499 nd->nd_dpos = mtod(m, void *); 2500 error = nfs_getreq(nd, nfsd, true); 2501 if (error) { 2502 m_freem(nam); 2503 nfsdreq_free(nd); 2504 return (error); 2505 } 2506 *ndp = nd; 2507 nfsd->nfsd_nd = nd; 2508 return (0); 2509 } 2510 2511 /* 2512 * Search for a sleeping nfsd and wake it up. 2513 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2514 * running nfsds will go look for the work in the nfssvc_sock list. 2515 */ 2516 static void 2517 nfsrv_wakenfsd_locked(struct nfssvc_sock *slp) 2518 { 2519 struct nfsd *nd; 2520 2521 KASSERT(mutex_owned(&nfsd_lock)); 2522 2523 if ((slp->ns_flags & SLP_VALID) == 0) 2524 return; 2525 if (slp->ns_gflags & SLP_G_DOREC) 2526 return; 2527 nd = SLIST_FIRST(&nfsd_idle_head); 2528 if (nd) { 2529 SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle); 2530 if (nd->nfsd_slp) 2531 panic("nfsd wakeup"); 2532 slp->ns_sref++; 2533 KASSERT(slp->ns_sref > 0); 2534 nd->nfsd_slp = slp; 2535 cv_signal(&nd->nfsd_cv); 2536 } else { 2537 slp->ns_gflags |= SLP_G_DOREC; 2538 nfsd_head_flag |= NFSD_CHECKSLP; 2539 TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending); 2540 } 2541 } 2542 2543 void 2544 nfsrv_wakenfsd(struct nfssvc_sock *slp) 2545 { 2546 2547 mutex_enter(&nfsd_lock); 2548 nfsrv_wakenfsd_locked(slp); 2549 mutex_exit(&nfsd_lock); 2550 } 2551 2552 int 2553 nfsdsock_sendreply(struct nfssvc_sock *slp, struct nfsrv_descript *nd) 2554 { 2555 int error; 2556 2557 if (nd->nd_mrep != NULL) { 2558 m_freem(nd->nd_mrep); 2559 nd->nd_mrep = NULL; 2560 } 2561 2562 mutex_enter(&slp->ns_lock); 2563 if ((slp->ns_flags & SLP_SENDING) != 0) { 2564 SIMPLEQ_INSERT_TAIL(&slp->ns_sendq, nd, nd_sendq); 2565 mutex_exit(&slp->ns_lock); 2566 return 0; 2567 } 2568 KASSERT(SIMPLEQ_EMPTY(&slp->ns_sendq)); 2569 slp->ns_flags |= SLP_SENDING; 2570 mutex_exit(&slp->ns_lock); 2571 2572 again: 2573 error = nfs_send(slp->ns_so, nd->nd_nam2, nd->nd_mreq, NULL, curlwp); 2574 if (nd->nd_nam2) { 2575 m_free(nd->nd_nam2); 2576 } 2577 nfsdreq_free(nd); 2578 2579 mutex_enter(&slp->ns_lock); 2580 KASSERT((slp->ns_flags & SLP_SENDING) != 0); 2581 nd = SIMPLEQ_FIRST(&slp->ns_sendq); 2582 if (nd != NULL) { 2583 SIMPLEQ_REMOVE_HEAD(&slp->ns_sendq, nd_sendq); 2584 mutex_exit(&slp->ns_lock); 2585 goto again; 2586 } 2587 slp->ns_flags &= ~SLP_SENDING; 2588 mutex_exit(&slp->ns_lock); 2589 2590 return error; 2591 } 2592 2593 void 2594 nfsdsock_setbits(struct nfssvc_sock *slp, int bits) 2595 { 2596 2597 mutex_enter(&slp->ns_alock); 2598 slp->ns_aflags |= bits; 2599 mutex_exit(&slp->ns_alock); 2600 } 2601 2602 void 2603 nfsdsock_clearbits(struct nfssvc_sock *slp, int bits) 2604 { 2605 2606 mutex_enter(&slp->ns_alock); 2607 slp->ns_aflags &= ~bits; 2608 mutex_exit(&slp->ns_alock); 2609 } 2610 2611 bool 2612 nfsdsock_testbits(struct nfssvc_sock *slp, int bits) 2613 { 2614 2615 return (slp->ns_aflags & bits); 2616 } 2617 #endif /* NFSSERVER */ 2618 2619 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) 2620 static struct pool nfs_srvdesc_pool; 2621 2622 void 2623 nfsdreq_init(void) 2624 { 2625 2626 pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript), 2627 0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr, IPL_NONE); 2628 } 2629 2630 struct nfsrv_descript * 2631 nfsdreq_alloc(void) 2632 { 2633 struct nfsrv_descript *nd; 2634 2635 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK); 2636 nd->nd_cr = NULL; 2637 return nd; 2638 } 2639 2640 void 2641 nfsdreq_free(struct nfsrv_descript *nd) 2642 { 2643 kauth_cred_t cr; 2644 2645 cr = nd->nd_cr; 2646 if (cr != NULL) { 2647 kauth_cred_free(cr); 2648 } 2649 pool_put(&nfs_srvdesc_pool, nd); 2650 } 2651 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */ 2652