nfs_socket.c revision 1.134
1 /* $NetBSD: nfs_socket.c,v 1.134 2006/05/28 13:23:08 blymn 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 1.134 2006/05/28 13:23:08 blymn Exp $"); 43 44 #include "fs_nfs.h" 45 #include "opt_nfs.h" 46 #include "opt_nfsserver.h" 47 #include "opt_mbuftrace.h" 48 #include "opt_inet.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/callout.h> 53 #include <sys/proc.h> 54 #include <sys/mount.h> 55 #include <sys/kernel.h> 56 #include <sys/mbuf.h> 57 #include <sys/vnode.h> 58 #include <sys/domain.h> 59 #include <sys/protosw.h> 60 #include <sys/socket.h> 61 #include <sys/socketvar.h> 62 #include <sys/syslog.h> 63 #include <sys/tprintf.h> 64 #include <sys/namei.h> 65 #include <sys/signal.h> 66 #include <sys/signalvar.h> 67 #include <sys/kauth.h> 68 69 #include <netinet/in.h> 70 #include <netinet/tcp.h> 71 72 #include <nfs/rpcv2.h> 73 #include <nfs/nfsproto.h> 74 #include <nfs/nfs.h> 75 #include <nfs/xdr_subs.h> 76 #include <nfs/nfsm_subs.h> 77 #include <nfs/nfsmount.h> 78 #include <nfs/nfsnode.h> 79 #include <nfs/nfsrtt.h> 80 #include <nfs/nqnfs.h> 81 #include <nfs/nfs_var.h> 82 83 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header"); 84 #ifdef MBUFTRACE 85 struct mowner nfs_mowner = { "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, nqnfs_prog; 114 extern time_t nqnfsstarttime; 115 extern const int nfsv3_procid[NFS_NPROCS]; 116 extern int nfs_ticks; 117 118 /* 119 * Defines which timer to use for the procnum. 120 * 0 - default 121 * 1 - getattr 122 * 2 - lookup 123 * 3 - read 124 * 4 - write 125 */ 126 static const int proct[NFS_NPROCS] = { 127 [NFSPROC_NULL] = 0, 128 [NFSPROC_GETATTR] = 1, 129 [NFSPROC_SETATTR] = 0, 130 [NFSPROC_LOOKUP] = 2, 131 [NFSPROC_ACCESS] = 1, 132 [NFSPROC_READLINK] = 3, 133 [NFSPROC_READ] = 3, 134 [NFSPROC_WRITE] = 4, 135 [NFSPROC_CREATE] = 0, 136 [NFSPROC_MKDIR] = 0, 137 [NFSPROC_SYMLINK] = 0, 138 [NFSPROC_MKNOD] = 0, 139 [NFSPROC_REMOVE] = 0, 140 [NFSPROC_RMDIR] = 0, 141 [NFSPROC_RENAME] = 0, 142 [NFSPROC_LINK] = 0, 143 [NFSPROC_READDIR] = 3, 144 [NFSPROC_READDIRPLUS] = 3, 145 [NFSPROC_FSSTAT] = 0, 146 [NFSPROC_FSINFO] = 0, 147 [NFSPROC_PATHCONF] = 0, 148 [NFSPROC_COMMIT] = 0, 149 [NQNFSPROC_GETLEASE] = 0, 150 [NQNFSPROC_VACATED] = 0, 151 [NQNFSPROC_EVICTED] = 0, 152 [NFSPROC_NOOP] = 0, 153 }; 154 155 /* 156 * There is a congestion window for outstanding rpcs maintained per mount 157 * point. The cwnd size is adjusted in roughly the way that: 158 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 159 * SIGCOMM '88". ACM, August 1988. 160 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 161 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 162 * of rpcs is in progress. 163 * (The sent count and cwnd are scaled for integer arith.) 164 * Variants of "slow start" were tried and were found to be too much of a 165 * performance hit (ave. rtt 3 times larger), 166 * I suspect due to the large rtt that nfs rpcs have. 167 */ 168 #define NFS_CWNDSCALE 256 169 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 170 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 171 int nfsrtton = 0; 172 struct nfsrtt nfsrtt; 173 struct nfsreqhead nfs_reqq; 174 175 struct callout nfs_timer_ch = CALLOUT_INITIALIZER_SETFUNC(nfs_timer, NULL); 176 177 /* 178 * Initialize sockets and congestion for a new NFS connection. 179 * We do not free the sockaddr if error. 180 */ 181 int 182 nfs_connect(nmp, rep, l) 183 struct nfsmount *nmp; 184 struct nfsreq *rep; 185 struct lwp *l; 186 { 187 struct socket *so; 188 int s, error, rcvreserve, sndreserve; 189 struct sockaddr *saddr; 190 struct sockaddr_in *sin; 191 #ifdef INET6 192 struct sockaddr_in6 *sin6; 193 #endif 194 struct mbuf *m; 195 196 nmp->nm_so = (struct socket *)0; 197 saddr = mtod(nmp->nm_nam, struct sockaddr *); 198 error = socreate(saddr->sa_family, &nmp->nm_so, 199 nmp->nm_sotype, nmp->nm_soproto, l); 200 if (error) 201 goto bad; 202 so = nmp->nm_so; 203 #ifdef MBUFTRACE 204 so->so_mowner = &nfs_mowner; 205 so->so_rcv.sb_mowner = &nfs_mowner; 206 so->so_snd.sb_mowner = &nfs_mowner; 207 #endif 208 nmp->nm_soflags = so->so_proto->pr_flags; 209 210 /* 211 * Some servers require that the client port be a reserved port number. 212 */ 213 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 214 m = m_get(M_WAIT, MT_SOOPTS); 215 MCLAIM(m, so->so_mowner); 216 *mtod(m, int32_t *) = IP_PORTRANGE_LOW; 217 m->m_len = sizeof(int32_t); 218 if ((error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, m))) 219 goto bad; 220 m = m_get(M_WAIT, MT_SONAME); 221 MCLAIM(m, so->so_mowner); 222 sin = mtod(m, struct sockaddr_in *); 223 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 224 sin->sin_family = AF_INET; 225 sin->sin_addr.s_addr = INADDR_ANY; 226 sin->sin_port = 0; 227 error = sobind(so, m, &lwp0); 228 m_freem(m); 229 if (error) 230 goto bad; 231 } 232 #ifdef INET6 233 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) { 234 m = m_get(M_WAIT, MT_SOOPTS); 235 MCLAIM(m, so->so_mowner); 236 *mtod(m, int32_t *) = IPV6_PORTRANGE_LOW; 237 m->m_len = sizeof(int32_t); 238 if ((error = sosetopt(so, IPPROTO_IPV6, IPV6_PORTRANGE, m))) 239 goto bad; 240 m = m_get(M_WAIT, MT_SONAME); 241 MCLAIM(m, so->so_mowner); 242 sin6 = mtod(m, struct sockaddr_in6 *); 243 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6); 244 sin6->sin6_family = AF_INET6; 245 sin6->sin6_addr = in6addr_any; 246 sin6->sin6_port = 0; 247 error = sobind(so, m, &lwp0); 248 m_freem(m); 249 if (error) 250 goto bad; 251 } 252 #endif 253 254 /* 255 * Protocols that do not require connections may be optionally left 256 * unconnected for servers that reply from a port other than NFS_PORT. 257 */ 258 if (nmp->nm_flag & NFSMNT_NOCONN) { 259 if (nmp->nm_soflags & PR_CONNREQUIRED) { 260 error = ENOTCONN; 261 goto bad; 262 } 263 } else { 264 error = soconnect(so, nmp->nm_nam, l); 265 if (error) 266 goto bad; 267 268 /* 269 * Wait for the connection to complete. Cribbed from the 270 * connect system call but with the wait timing out so 271 * that interruptible mounts don't hang here for a long time. 272 */ 273 s = splsoftnet(); 274 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 275 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 276 "nfscn1", 2 * hz); 277 if ((so->so_state & SS_ISCONNECTING) && 278 so->so_error == 0 && rep && 279 (error = nfs_sigintr(nmp, rep, rep->r_lwp)) != 0){ 280 so->so_state &= ~SS_ISCONNECTING; 281 splx(s); 282 goto bad; 283 } 284 } 285 if (so->so_error) { 286 error = so->so_error; 287 so->so_error = 0; 288 splx(s); 289 goto bad; 290 } 291 splx(s); 292 } 293 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 294 so->so_rcv.sb_timeo = (5 * hz); 295 so->so_snd.sb_timeo = (5 * hz); 296 } else { 297 /* 298 * enable receive timeout to detect server crash and reconnect. 299 * otherwise, we can be stuck in soreceive forever. 300 */ 301 so->so_rcv.sb_timeo = (5 * hz); 302 so->so_snd.sb_timeo = 0; 303 } 304 if (nmp->nm_sotype == SOCK_DGRAM) { 305 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 306 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 307 NFS_MAXPKTHDR) * 2; 308 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 309 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 310 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 311 NFS_MAXPKTHDR) * 2; 312 } else { 313 if (nmp->nm_sotype != SOCK_STREAM) 314 panic("nfscon sotype"); 315 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 316 m = m_get(M_WAIT, MT_SOOPTS); 317 MCLAIM(m, so->so_mowner); 318 *mtod(m, int32_t *) = 1; 319 m->m_len = sizeof(int32_t); 320 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 321 } 322 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 323 m = m_get(M_WAIT, MT_SOOPTS); 324 MCLAIM(m, so->so_mowner); 325 *mtod(m, int32_t *) = 1; 326 m->m_len = sizeof(int32_t); 327 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 328 } 329 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 330 sizeof (u_int32_t)) * 2; 331 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 332 sizeof (u_int32_t)) * 2; 333 } 334 error = soreserve(so, sndreserve, rcvreserve); 335 if (error) 336 goto bad; 337 so->so_rcv.sb_flags |= SB_NOINTR; 338 so->so_snd.sb_flags |= SB_NOINTR; 339 340 /* Initialize other non-zero congestion variables */ 341 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 342 NFS_TIMEO << 3; 343 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 344 nmp->nm_sdrtt[3] = 0; 345 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 346 nmp->nm_sent = 0; 347 nmp->nm_timeouts = 0; 348 return (0); 349 350 bad: 351 nfs_disconnect(nmp); 352 return (error); 353 } 354 355 /* 356 * Reconnect routine: 357 * Called when a connection is broken on a reliable protocol. 358 * - clean up the old socket 359 * - nfs_connect() again 360 * - set R_MUSTRESEND for all outstanding requests on mount point 361 * If this fails the mount point is DEAD! 362 * nb: Must be called with the nfs_sndlock() set on the mount point. 363 */ 364 int 365 nfs_reconnect(rep, l) 366 struct nfsreq *rep; 367 struct lwp *l; 368 { 369 struct nfsreq *rp; 370 struct nfsmount *nmp = rep->r_nmp; 371 int error; 372 373 nfs_disconnect(nmp); 374 while ((error = nfs_connect(nmp, rep, l)) != 0) { 375 if (error == EINTR || error == ERESTART) 376 return (EINTR); 377 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscn2", 0); 378 } 379 380 /* 381 * Loop through outstanding request list and fix up all requests 382 * on old socket. 383 */ 384 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 385 if (rp->r_nmp == nmp) { 386 if ((rp->r_flags & R_MUSTRESEND) == 0) 387 rp->r_flags |= R_MUSTRESEND | R_REXMITTED; 388 rp->r_rexmit = 0; 389 } 390 } 391 return (0); 392 } 393 394 /* 395 * NFS disconnect. Clean up and unlink. 396 */ 397 void 398 nfs_disconnect(nmp) 399 struct nfsmount *nmp; 400 { 401 struct socket *so; 402 int drain = 0; 403 404 if (nmp->nm_so) { 405 so = nmp->nm_so; 406 nmp->nm_so = (struct socket *)0; 407 soshutdown(so, 2); 408 drain = (nmp->nm_iflag & NFSMNT_DISMNT) != 0; 409 if (drain) { 410 /* 411 * soshutdown() above should wake up the current 412 * listener. 413 * Now wake up those waiting for the receive lock, and 414 * wait for them to go away unhappy, to prevent *nmp 415 * from evaporating while they're sleeping. 416 */ 417 while (nmp->nm_waiters > 0) { 418 wakeup (&nmp->nm_iflag); 419 (void) tsleep(&nmp->nm_waiters, PVFS, 420 "nfsdis", 0); 421 } 422 } 423 soclose(so); 424 } 425 #ifdef DIAGNOSTIC 426 if (drain && (nmp->nm_waiters > 0)) 427 panic("nfs_disconnect: waiters left after drain?"); 428 #endif 429 } 430 431 void 432 nfs_safedisconnect(nmp) 433 struct nfsmount *nmp; 434 { 435 struct nfsreq dummyreq; 436 437 memset(&dummyreq, 0, sizeof(dummyreq)); 438 dummyreq.r_nmp = nmp; 439 nfs_rcvlock(&dummyreq); /* XXX ignored error return */ 440 nfs_disconnect(nmp); 441 nfs_rcvunlock(nmp); 442 } 443 444 /* 445 * This is the nfs send routine. For connection based socket types, it 446 * must be called with an nfs_sndlock() on the socket. 447 * "rep == NULL" indicates that it has been called from a server. 448 * For the client side: 449 * - return EINTR if the RPC is terminated, 0 otherwise 450 * - set R_MUSTRESEND if the send fails for any reason 451 * - do any cleanup required by recoverable socket errors (? ? ?) 452 * For the server side: 453 * - return EINTR or ERESTART if interrupted by a signal 454 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 455 * - do any cleanup required by recoverable socket errors (? ? ?) 456 */ 457 int 458 nfs_send(so, nam, top, rep, l) 459 struct socket *so; 460 struct mbuf *nam; 461 struct mbuf *top; 462 struct nfsreq *rep; 463 struct lwp *l; 464 { 465 struct mbuf *sendnam; 466 int error, soflags, flags; 467 468 /* XXX nfs_doio()/nfs_request() calls with rep->r_lwp == NULL */ 469 if (l == NULL && rep->r_lwp == NULL) 470 l = curlwp; 471 472 if (rep) { 473 if (rep->r_flags & R_SOFTTERM) { 474 m_freem(top); 475 return (EINTR); 476 } 477 if ((so = rep->r_nmp->nm_so) == NULL) { 478 rep->r_flags |= R_MUSTRESEND; 479 m_freem(top); 480 return (0); 481 } 482 rep->r_flags &= ~R_MUSTRESEND; 483 soflags = rep->r_nmp->nm_soflags; 484 } else 485 soflags = so->so_proto->pr_flags; 486 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 487 sendnam = (struct mbuf *)0; 488 else 489 sendnam = nam; 490 if (so->so_type == SOCK_SEQPACKET) 491 flags = MSG_EOR; 492 else 493 flags = 0; 494 495 error = (*so->so_send)(so, sendnam, (struct uio *)0, top, 496 (struct mbuf *)0, flags, l); 497 if (error) { 498 if (rep) { 499 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 500 /* 501 * We're too fast for the network/driver, 502 * and UDP isn't flowcontrolled. 503 * We need to resend. This is not fatal, 504 * just try again. 505 * 506 * Could be smarter here by doing some sort 507 * of a backoff, but this is rare. 508 */ 509 rep->r_flags |= R_MUSTRESEND; 510 } else { 511 if (error != EPIPE) 512 log(LOG_INFO, 513 "nfs send error %d for %s\n", 514 error, 515 rep->r_nmp->nm_mountp-> 516 mnt_stat.f_mntfromname); 517 /* 518 * Deal with errors for the client side. 519 */ 520 if (rep->r_flags & R_SOFTTERM) 521 error = EINTR; 522 else 523 rep->r_flags |= R_MUSTRESEND; 524 } 525 } else { 526 /* 527 * See above. This error can happen under normal 528 * circumstances and the log is too noisy. 529 * The error will still show up in nfsstat. 530 */ 531 if (error != ENOBUFS || so->so_type != SOCK_DGRAM) 532 log(LOG_INFO, "nfsd send error %d\n", error); 533 } 534 535 /* 536 * Handle any recoverable (soft) socket errors here. (? ? ?) 537 */ 538 if (error != EINTR && error != ERESTART && 539 error != EWOULDBLOCK && error != EPIPE) 540 error = 0; 541 } 542 return (error); 543 } 544 545 #ifdef NFS 546 /* 547 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 548 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 549 * Mark and consolidate the data into a new mbuf list. 550 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 551 * small mbufs. 552 * For SOCK_STREAM we must be very careful to read an entire record once 553 * we have read any of it, even if the system call has been interrupted. 554 */ 555 int 556 nfs_receive(rep, aname, mp, l) 557 struct nfsreq *rep; 558 struct mbuf **aname; 559 struct mbuf **mp; 560 struct lwp *l; 561 { 562 struct socket *so; 563 struct uio auio; 564 struct iovec aio; 565 struct mbuf *m; 566 struct mbuf *control; 567 u_int32_t len; 568 struct mbuf **getnam; 569 int error, sotype, rcvflg; 570 571 /* 572 * Set up arguments for soreceive() 573 */ 574 *mp = (struct mbuf *)0; 575 *aname = (struct mbuf *)0; 576 sotype = rep->r_nmp->nm_sotype; 577 578 /* 579 * For reliable protocols, lock against other senders/receivers 580 * in case a reconnect is necessary. 581 * For SOCK_STREAM, first get the Record Mark to find out how much 582 * more there is to get. 583 * We must lock the socket against other receivers 584 * until we have an entire rpc request/reply. 585 */ 586 if (sotype != SOCK_DGRAM) { 587 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 588 if (error) 589 return (error); 590 tryagain: 591 /* 592 * Check for fatal errors and resending request. 593 */ 594 /* 595 * Ugh: If a reconnect attempt just happened, nm_so 596 * would have changed. NULL indicates a failed 597 * attempt that has essentially shut down this 598 * mount point. 599 */ 600 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 601 nfs_sndunlock(&rep->r_nmp->nm_iflag); 602 return (EINTR); 603 } 604 so = rep->r_nmp->nm_so; 605 if (!so) { 606 error = nfs_reconnect(rep, l); 607 if (error) { 608 nfs_sndunlock(&rep->r_nmp->nm_iflag); 609 return (error); 610 } 611 goto tryagain; 612 } 613 while (rep->r_flags & R_MUSTRESEND) { 614 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 615 nfsstats.rpcretries++; 616 rep->r_rtt = 0; 617 rep->r_flags &= ~R_TIMING; 618 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, l); 619 if (error) { 620 if (error == EINTR || error == ERESTART || 621 (error = nfs_reconnect(rep, l)) != 0) { 622 nfs_sndunlock(&rep->r_nmp->nm_iflag); 623 return (error); 624 } 625 goto tryagain; 626 } 627 } 628 nfs_sndunlock(&rep->r_nmp->nm_iflag); 629 if (sotype == SOCK_STREAM) { 630 aio.iov_base = (caddr_t) &len; 631 aio.iov_len = sizeof(u_int32_t); 632 auio.uio_iov = &aio; 633 auio.uio_iovcnt = 1; 634 auio.uio_rw = UIO_READ; 635 auio.uio_offset = 0; 636 auio.uio_resid = sizeof(u_int32_t); 637 UIO_SETUP_SYSSPACE(&auio); 638 do { 639 rcvflg = MSG_WAITALL; 640 error = (*so->so_receive)(so, (struct mbuf **)0, &auio, 641 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 642 if (error == EWOULDBLOCK && rep) { 643 if (rep->r_flags & R_SOFTTERM) 644 return (EINTR); 645 /* 646 * if it seems that the server died after it 647 * received our request, set EPIPE so that 648 * we'll reconnect and retransmit requests. 649 */ 650 if (rep->r_rexmit >= rep->r_nmp->nm_retry) { 651 nfsstats.rpctimeouts++; 652 error = EPIPE; 653 } 654 } 655 } while (error == EWOULDBLOCK); 656 if (!error && auio.uio_resid > 0) { 657 /* 658 * Don't log a 0 byte receive; it means 659 * that the socket has been closed, and 660 * can happen during normal operation 661 * (forcible unmount or Solaris server). 662 */ 663 if (auio.uio_resid != sizeof (u_int32_t)) 664 log(LOG_INFO, 665 "short receive (%lu/%lu) from nfs server %s\n", 666 (u_long)sizeof(u_int32_t) - auio.uio_resid, 667 (u_long)sizeof(u_int32_t), 668 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 669 error = EPIPE; 670 } 671 if (error) 672 goto errout; 673 len = ntohl(len) & ~0x80000000; 674 /* 675 * This is SERIOUS! We are out of sync with the sender 676 * and forcing a disconnect/reconnect is all I can do. 677 */ 678 if (len > NFS_MAXPACKET) { 679 log(LOG_ERR, "%s (%d) from nfs server %s\n", 680 "impossible packet length", 681 len, 682 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 683 error = EFBIG; 684 goto errout; 685 } 686 auio.uio_resid = len; 687 do { 688 rcvflg = MSG_WAITALL; 689 error = (*so->so_receive)(so, (struct mbuf **)0, 690 &auio, mp, (struct mbuf **)0, &rcvflg); 691 } while (error == EWOULDBLOCK || error == EINTR || 692 error == ERESTART); 693 if (!error && auio.uio_resid > 0) { 694 if (len != auio.uio_resid) 695 log(LOG_INFO, 696 "short receive (%lu/%d) from nfs server %s\n", 697 (u_long)len - auio.uio_resid, len, 698 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 699 error = EPIPE; 700 } 701 } else { 702 /* 703 * NB: Since uio_resid is big, MSG_WAITALL is ignored 704 * and soreceive() will return when it has either a 705 * control msg or a data msg. 706 * We have no use for control msg., but must grab them 707 * and then throw them away so we know what is going 708 * on. 709 */ 710 auio.uio_resid = len = 100000000; /* Anything Big */ 711 /* not need to setup uio_vmspace */ 712 do { 713 rcvflg = 0; 714 error = (*so->so_receive)(so, (struct mbuf **)0, 715 &auio, mp, &control, &rcvflg); 716 if (control) 717 m_freem(control); 718 if (error == EWOULDBLOCK && rep) { 719 if (rep->r_flags & R_SOFTTERM) 720 return (EINTR); 721 } 722 } while (error == EWOULDBLOCK || 723 (!error && *mp == NULL && control)); 724 if ((rcvflg & MSG_EOR) == 0) 725 printf("Egad!!\n"); 726 if (!error && *mp == NULL) 727 error = EPIPE; 728 len -= auio.uio_resid; 729 } 730 errout: 731 if (error && error != EINTR && error != ERESTART) { 732 m_freem(*mp); 733 *mp = (struct mbuf *)0; 734 if (error != EPIPE) 735 log(LOG_INFO, 736 "receive error %d from nfs server %s\n", 737 error, 738 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 739 error = nfs_sndlock(&rep->r_nmp->nm_iflag, rep); 740 if (!error) 741 error = nfs_reconnect(rep, l); 742 if (!error) 743 goto tryagain; 744 else 745 nfs_sndunlock(&rep->r_nmp->nm_iflag); 746 } 747 } else { 748 if ((so = rep->r_nmp->nm_so) == NULL) 749 return (EACCES); 750 if (so->so_state & SS_ISCONNECTED) 751 getnam = (struct mbuf **)0; 752 else 753 getnam = aname; 754 auio.uio_resid = len = 1000000; 755 /* not need to setup uio_vmspace */ 756 do { 757 rcvflg = 0; 758 error = (*so->so_receive)(so, getnam, &auio, mp, 759 (struct mbuf **)0, &rcvflg); 760 if (error == EWOULDBLOCK && 761 (rep->r_flags & R_SOFTTERM)) 762 return (EINTR); 763 } while (error == EWOULDBLOCK); 764 len -= auio.uio_resid; 765 if (!error && *mp == NULL) 766 error = EPIPE; 767 } 768 if (error) { 769 m_freem(*mp); 770 *mp = (struct mbuf *)0; 771 } 772 return (error); 773 } 774 775 /* 776 * Implement receipt of reply on a socket. 777 * We must search through the list of received datagrams matching them 778 * with outstanding requests using the xid, until ours is found. 779 */ 780 /* ARGSUSED */ 781 int 782 nfs_reply(myrep, lwp) 783 struct nfsreq *myrep; 784 struct lwp *lwp; 785 { 786 struct nfsreq *rep; 787 struct nfsmount *nmp = myrep->r_nmp; 788 int32_t t1; 789 struct mbuf *mrep, *nam, *md; 790 u_int32_t rxid, *tl; 791 caddr_t dpos, cp2; 792 int error; 793 794 /* 795 * Loop around until we get our own reply 796 */ 797 for (;;) { 798 /* 799 * Lock against other receivers so that I don't get stuck in 800 * sbwait() after someone else has received my reply for me. 801 * Also necessary for connection based protocols to avoid 802 * race conditions during a reconnect. 803 */ 804 error = nfs_rcvlock(myrep); 805 if (error == EALREADY) 806 return (0); 807 if (error) 808 return (error); 809 /* 810 * Get the next Rpc reply off the socket 811 */ 812 nmp->nm_waiters++; 813 error = nfs_receive(myrep, &nam, &mrep, lwp); 814 nfs_rcvunlock(nmp); 815 if (error) { 816 817 if (nmp->nm_iflag & NFSMNT_DISMNT) { 818 /* 819 * Oops, we're going away now.. 820 */ 821 nmp->nm_waiters--; 822 wakeup (&nmp->nm_waiters); 823 return error; 824 } 825 nmp->nm_waiters--; 826 /* 827 * Ignore routing errors on connectionless protocols? ? 828 */ 829 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 830 nmp->nm_so->so_error = 0; 831 #ifdef DEBUG 832 printf("nfs_reply: ignoring error %d\n", error); 833 #endif 834 if (myrep->r_flags & R_GETONEREP) 835 return (0); 836 continue; 837 } 838 return (error); 839 } 840 nmp->nm_waiters--; 841 if (nam) 842 m_freem(nam); 843 844 /* 845 * Get the xid and check that it is an rpc reply 846 */ 847 md = mrep; 848 dpos = mtod(md, caddr_t); 849 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 850 rxid = *tl++; 851 if (*tl != rpc_reply) { 852 #ifndef NFS_V2_ONLY 853 if (nmp->nm_flag & NFSMNT_NQNFS) { 854 if (nqnfs_callback(nmp, mrep, md, dpos, 855 myrep->r_lwp)) 856 nfsstats.rpcinvalid++; 857 } else 858 #endif 859 { 860 nfsstats.rpcinvalid++; 861 m_freem(mrep); 862 } 863 nfsmout: 864 if (myrep->r_flags & R_GETONEREP) 865 return (0); 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 rt->tstamp = time; 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 /* 937 * If not matched to a request, drop it. 938 * If it's mine, get out. 939 */ 940 if (rep == 0) { 941 nfsstats.rpcunexpected++; 942 m_freem(mrep); 943 } else if (rep == myrep) { 944 if (rep->r_mrep == NULL) 945 panic("nfsreply nil"); 946 return (0); 947 } 948 if (myrep->r_flags & R_GETONEREP) 949 return (0); 950 } 951 } 952 953 /* 954 * nfs_request - goes something like this 955 * - fill in request struct 956 * - links it into list 957 * - calls nfs_send() for first transmit 958 * - calls nfs_receive() to get reply 959 * - break down rpc header and return with nfs reply pointed to 960 * by mrep or error 961 * nb: always frees up mreq mbuf list 962 */ 963 int 964 nfs_request(np, mrest, procnum, lwp, cred, mrp, mdp, dposp, rexmitp) 965 struct nfsnode *np; 966 struct mbuf *mrest; 967 int procnum; 968 struct lwp *lwp; 969 kauth_cred_t cred; 970 struct mbuf **mrp; 971 struct mbuf **mdp; 972 caddr_t *dposp; 973 int *rexmitp; 974 { 975 struct mbuf *m, *mrep; 976 struct nfsreq *rep; 977 u_int32_t *tl; 978 int i; 979 struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount); 980 struct mbuf *md, *mheadend; 981 char nickv[RPCX_NICKVERF]; 982 time_t reqtime, waituntil; 983 caddr_t dpos, cp2; 984 int t1, s, error = 0, mrest_len, auth_len, auth_type; 985 int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0; 986 int verf_len, verf_type; 987 u_int32_t xid; 988 char *auth_str, *verf_str; 989 NFSKERBKEY_T key; /* save session key */ 990 kauth_cred_t acred; 991 #ifndef NFS_V2_ONLY 992 int nqlflag, cachable; 993 u_quad_t frev; 994 #endif 995 struct mbuf *mrest_backup = NULL; 996 kauth_cred_t origcred = NULL; /* XXX: gcc */ 997 boolean_t retry_cred = TRUE; 998 boolean_t use_opencred = (np->n_flag & NUSEOPENCRED) != 0; 999 1000 if (rexmitp != NULL) 1001 *rexmitp = 0; 1002 1003 acred = kauth_cred_alloc(); 1004 1005 tryagain_cred: 1006 KASSERT(cred != NULL); 1007 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 1008 rep->r_nmp = nmp; 1009 rep->r_lwp = lwp; 1010 rep->r_procnum = procnum; 1011 i = 0; 1012 m = mrest; 1013 while (m) { 1014 i += m->m_len; 1015 m = m->m_next; 1016 } 1017 mrest_len = i; 1018 1019 /* 1020 * Get the RPC header with authorization. 1021 */ 1022 kerbauth: 1023 verf_str = auth_str = (char *)0; 1024 if (nmp->nm_flag & NFSMNT_KERB) { 1025 verf_str = nickv; 1026 verf_len = sizeof (nickv); 1027 auth_type = RPCAUTH_KERB4; 1028 memset((caddr_t)key, 0, sizeof (key)); 1029 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 1030 &auth_len, verf_str, verf_len)) { 1031 error = nfs_getauth(nmp, rep, cred, &auth_str, 1032 &auth_len, verf_str, &verf_len, key); 1033 if (error) { 1034 free((caddr_t)rep, M_NFSREQ); 1035 m_freem(mrest); 1036 KASSERT(kauth_cred_getrefcnt(acred) == 1); 1037 kauth_cred_free(acred); 1038 return (error); 1039 } 1040 } 1041 retry_cred = FALSE; 1042 } else { 1043 /* AUTH_UNIX */ 1044 uid_t uid; 1045 gid_t gid; 1046 1047 /* 1048 * on the most unix filesystems, permission checks are 1049 * done when the file is open(2)'ed. 1050 * ie. once a file is successfully open'ed, 1051 * following i/o operations never fail with EACCES. 1052 * we try to follow the semantics as far as possible. 1053 * 1054 * note that we expect that the nfs server always grant 1055 * accesses by the file's owner. 1056 */ 1057 origcred = cred; 1058 switch (procnum) { 1059 case NFSPROC_READ: 1060 case NFSPROC_WRITE: 1061 case NFSPROC_COMMIT: 1062 uid = np->n_vattr->va_uid; 1063 gid = np->n_vattr->va_gid; 1064 if (kauth_cred_geteuid(cred) == uid && 1065 kauth_cred_getegid(cred) == gid) { 1066 retry_cred = FALSE; 1067 break; 1068 } 1069 if (use_opencred) 1070 break; 1071 kauth_cred_setuid(acred, uid); 1072 kauth_cred_seteuid(acred, uid); 1073 kauth_cred_setsvuid(acred, uid); 1074 kauth_cred_setgid(acred, gid); 1075 kauth_cred_setegid(acred, gid); 1076 kauth_cred_setsvgid(acred, gid); 1077 cred = acred; 1078 break; 1079 default: 1080 retry_cred = FALSE; 1081 break; 1082 } 1083 /* 1084 * backup mbuf chain if we can need it later to retry. 1085 * 1086 * XXX maybe we can keep a direct reference to 1087 * mrest without doing m_copym, but it's ...ugly. 1088 */ 1089 if (retry_cred) 1090 mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT); 1091 auth_type = RPCAUTH_UNIX; 1092 /* XXX elad - ngroups */ 1093 auth_len = (((kauth_cred_ngroups(cred) > nmp->nm_numgrps) ? 1094 nmp->nm_numgrps : kauth_cred_ngroups(cred)) << 2) + 1095 5 * NFSX_UNSIGNED; 1096 } 1097 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 1098 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 1099 if (auth_str) 1100 free(auth_str, M_TEMP); 1101 1102 /* 1103 * For stream protocols, insert a Sun RPC Record Mark. 1104 */ 1105 if (nmp->nm_sotype == SOCK_STREAM) { 1106 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 1107 *mtod(m, u_int32_t *) = htonl(0x80000000 | 1108 (m->m_pkthdr.len - NFSX_UNSIGNED)); 1109 } 1110 rep->r_mreq = m; 1111 rep->r_xid = xid; 1112 tryagain: 1113 if (nmp->nm_flag & NFSMNT_SOFT) 1114 rep->r_retry = nmp->nm_retry; 1115 else 1116 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1117 rep->r_rtt = rep->r_rexmit = 0; 1118 if (proct[procnum] > 0) 1119 rep->r_flags = R_TIMING; 1120 else 1121 rep->r_flags = 0; 1122 rep->r_mrep = NULL; 1123 1124 /* 1125 * Do the client side RPC. 1126 */ 1127 nfsstats.rpcrequests++; 1128 /* 1129 * Chain request into list of outstanding requests. Be sure 1130 * to put it LAST so timer finds oldest requests first. 1131 */ 1132 s = splsoftnet(); 1133 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1134 1135 /* Get send time for nqnfs */ 1136 reqtime = time.tv_sec; 1137 1138 /* 1139 * If backing off another request or avoiding congestion, don't 1140 * send this one now but let timer do it. If not timing a request, 1141 * do it now. 1142 */ 1143 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1144 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 1145 nmp->nm_sent < nmp->nm_cwnd)) { 1146 splx(s); 1147 if (nmp->nm_soflags & PR_CONNREQUIRED) 1148 error = nfs_sndlock(&nmp->nm_iflag, rep); 1149 if (!error) { 1150 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1151 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp); 1152 if (nmp->nm_soflags & PR_CONNREQUIRED) 1153 nfs_sndunlock(&nmp->nm_iflag); 1154 } 1155 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 1156 nmp->nm_sent += NFS_CWNDSCALE; 1157 rep->r_flags |= R_SENT; 1158 } 1159 } else { 1160 splx(s); 1161 rep->r_rtt = -1; 1162 } 1163 1164 /* 1165 * Wait for the reply from our send or the timer's. 1166 */ 1167 if (!error || error == EPIPE) 1168 error = nfs_reply(rep, lwp); 1169 1170 /* 1171 * RPC done, unlink the request. 1172 */ 1173 s = splsoftnet(); 1174 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1175 splx(s); 1176 1177 /* 1178 * Decrement the outstanding request count. 1179 */ 1180 if (rep->r_flags & R_SENT) { 1181 rep->r_flags &= ~R_SENT; /* paranoia */ 1182 nmp->nm_sent -= NFS_CWNDSCALE; 1183 } 1184 1185 if (rexmitp != NULL) { 1186 int rexmit; 1187 1188 if (nmp->nm_sotype != SOCK_DGRAM) 1189 rexmit = (rep->r_flags & R_REXMITTED) != 0; 1190 else 1191 rexmit = rep->r_rexmit; 1192 *rexmitp = rexmit; 1193 } 1194 1195 /* 1196 * If there was a successful reply and a tprintf msg. 1197 * tprintf a response. 1198 */ 1199 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1200 nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1201 "is alive again"); 1202 mrep = rep->r_mrep; 1203 md = rep->r_md; 1204 dpos = rep->r_dpos; 1205 if (error) 1206 goto nfsmout; 1207 1208 /* 1209 * break down the rpc header and check if ok 1210 */ 1211 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1212 if (*tl++ == rpc_msgdenied) { 1213 if (*tl == rpc_mismatch) 1214 error = EOPNOTSUPP; 1215 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1216 if (!failed_auth) { 1217 failed_auth++; 1218 mheadend->m_next = (struct mbuf *)0; 1219 m_freem(mrep); 1220 m_freem(rep->r_mreq); 1221 goto kerbauth; 1222 } else 1223 error = EAUTH; 1224 } else 1225 error = EACCES; 1226 m_freem(mrep); 1227 goto nfsmout; 1228 } 1229 1230 /* 1231 * Grab any Kerberos verifier, otherwise just throw it away. 1232 */ 1233 verf_type = fxdr_unsigned(int, *tl++); 1234 i = fxdr_unsigned(int32_t, *tl); 1235 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1236 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1237 if (error) 1238 goto nfsmout; 1239 } else if (i > 0) 1240 nfsm_adv(nfsm_rndup(i)); 1241 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1242 /* 0 == ok */ 1243 if (*tl == 0) { 1244 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1245 if (*tl != 0) { 1246 error = fxdr_unsigned(int, *tl); 1247 switch (error) { 1248 case NFSERR_PERM: 1249 error = EPERM; 1250 break; 1251 1252 case NFSERR_NOENT: 1253 error = ENOENT; 1254 break; 1255 1256 case NFSERR_IO: 1257 error = EIO; 1258 break; 1259 1260 case NFSERR_NXIO: 1261 error = ENXIO; 1262 break; 1263 1264 case NFSERR_ACCES: 1265 error = EACCES; 1266 if (!retry_cred) 1267 break; 1268 m_freem(mrep); 1269 m_freem(rep->r_mreq); 1270 FREE(rep, M_NFSREQ); 1271 use_opencred = !use_opencred; 1272 if (mrest_backup == NULL) { 1273 /* m_copym failure */ 1274 KASSERT( 1275 kauth_cred_getrefcnt(acred) == 1); 1276 kauth_cred_free(acred); 1277 return ENOMEM; 1278 } 1279 mrest = mrest_backup; 1280 mrest_backup = NULL; 1281 cred = origcred; 1282 error = 0; 1283 retry_cred = FALSE; 1284 goto tryagain_cred; 1285 1286 case NFSERR_EXIST: 1287 error = EEXIST; 1288 break; 1289 1290 case NFSERR_XDEV: 1291 error = EXDEV; 1292 break; 1293 1294 case NFSERR_NODEV: 1295 error = ENODEV; 1296 break; 1297 1298 case NFSERR_NOTDIR: 1299 error = ENOTDIR; 1300 break; 1301 1302 case NFSERR_ISDIR: 1303 error = EISDIR; 1304 break; 1305 1306 case NFSERR_INVAL: 1307 error = EINVAL; 1308 break; 1309 1310 case NFSERR_FBIG: 1311 error = EFBIG; 1312 break; 1313 1314 case NFSERR_NOSPC: 1315 error = ENOSPC; 1316 break; 1317 1318 case NFSERR_ROFS: 1319 error = EROFS; 1320 break; 1321 1322 case NFSERR_MLINK: 1323 error = EMLINK; 1324 break; 1325 1326 case NFSERR_TIMEDOUT: 1327 error = ETIMEDOUT; 1328 break; 1329 1330 case NFSERR_NAMETOL: 1331 error = ENAMETOOLONG; 1332 break; 1333 1334 case NFSERR_NOTEMPTY: 1335 error = ENOTEMPTY; 1336 break; 1337 1338 case NFSERR_DQUOT: 1339 error = EDQUOT; 1340 break; 1341 1342 case NFSERR_STALE: 1343 /* 1344 * If the File Handle was stale, invalidate the 1345 * lookup cache, just in case. 1346 */ 1347 error = ESTALE; 1348 cache_purge(NFSTOV(np)); 1349 break; 1350 1351 case NFSERR_REMOTE: 1352 error = EREMOTE; 1353 break; 1354 1355 case NFSERR_WFLUSH: 1356 case NFSERR_BADHANDLE: 1357 case NFSERR_NOT_SYNC: 1358 case NFSERR_BAD_COOKIE: 1359 error = EINVAL; 1360 break; 1361 1362 case NFSERR_NOTSUPP: 1363 error = ENOTSUP; 1364 break; 1365 1366 case NFSERR_TOOSMALL: 1367 case NFSERR_SERVERFAULT: 1368 case NFSERR_BADTYPE: 1369 error = EINVAL; 1370 break; 1371 1372 case NFSERR_TRYLATER: 1373 if ((nmp->nm_flag & NFSMNT_NFSV3) == 0) 1374 break; 1375 m_freem(mrep); 1376 error = 0; 1377 waituntil = time.tv_sec + trylater_delay; 1378 while (time.tv_sec < waituntil) 1379 (void) tsleep((caddr_t)&lbolt, 1380 PSOCK, "nqnfstry", 0); 1381 trylater_delay *= NFS_TRYLATERDELMUL; 1382 if (trylater_delay > NFS_TRYLATERDELMAX) 1383 trylater_delay = NFS_TRYLATERDELMAX; 1384 /* 1385 * RFC1813: 1386 * The client should wait and then try 1387 * the request with a new RPC transaction ID. 1388 */ 1389 nfs_renewxid(rep); 1390 goto tryagain; 1391 1392 case NFSERR_STALEWRITEVERF: 1393 error = EINVAL; 1394 break; 1395 1396 default: 1397 #ifdef DIAGNOSTIC 1398 printf("Invalid rpc error code %d\n", error); 1399 #endif 1400 error = EINVAL; 1401 break; 1402 } 1403 1404 if (nmp->nm_flag & NFSMNT_NFSV3) { 1405 *mrp = mrep; 1406 *mdp = md; 1407 *dposp = dpos; 1408 error |= NFSERR_RETERR; 1409 } else 1410 m_freem(mrep); 1411 goto nfsmout; 1412 } 1413 1414 /* 1415 * note which credential worked to minimize number of retries. 1416 */ 1417 if (use_opencred) 1418 np->n_flag |= NUSEOPENCRED; 1419 else 1420 np->n_flag &= ~NUSEOPENCRED; 1421 1422 #ifndef NFS_V2_ONLY 1423 /* 1424 * For nqnfs, get any lease in reply 1425 */ 1426 if (nmp->nm_flag & NFSMNT_NQNFS) { 1427 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1428 if (*tl) { 1429 nqlflag = fxdr_unsigned(int, *tl); 1430 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED); 1431 cachable = fxdr_unsigned(int, *tl++); 1432 reqtime += fxdr_unsigned(int, *tl++); 1433 if (reqtime > time.tv_sec) { 1434 frev = fxdr_hyper(tl); 1435 nqnfs_clientlease(nmp, np, nqlflag, 1436 cachable, reqtime, frev); 1437 } 1438 } 1439 } 1440 #endif 1441 *mrp = mrep; 1442 *mdp = md; 1443 *dposp = dpos; 1444 1445 KASSERT(error == 0); 1446 goto nfsmout; 1447 } 1448 m_freem(mrep); 1449 error = EPROTONOSUPPORT; 1450 nfsmout: 1451 KASSERT(kauth_cred_getrefcnt(acred) == 1); 1452 kauth_cred_free(acred); 1453 m_freem(rep->r_mreq); 1454 free((caddr_t)rep, M_NFSREQ); 1455 m_freem(mrest_backup); 1456 return (error); 1457 } 1458 #endif /* NFS */ 1459 1460 /* 1461 * Generate the rpc reply header 1462 * siz arg. is used to decide if adding a cluster is worthwhile 1463 */ 1464 int 1465 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) 1466 int siz; 1467 struct nfsrv_descript *nd; 1468 struct nfssvc_sock *slp; 1469 int err; 1470 int cache; 1471 u_quad_t *frev; 1472 struct mbuf **mrq; 1473 struct mbuf **mbp; 1474 caddr_t *bposp; 1475 { 1476 u_int32_t *tl; 1477 struct mbuf *mreq; 1478 caddr_t bpos; 1479 struct mbuf *mb; 1480 1481 mreq = m_gethdr(M_WAIT, MT_DATA); 1482 MCLAIM(mreq, &nfs_mowner); 1483 mb = mreq; 1484 /* 1485 * If this is a big reply, use a cluster else 1486 * try and leave leading space for the lower level headers. 1487 */ 1488 siz += RPC_REPLYSIZ; 1489 if (siz >= max_datalen) { 1490 m_clget(mreq, M_WAIT); 1491 } else 1492 mreq->m_data += max_hdr; 1493 tl = mtod(mreq, u_int32_t *); 1494 mreq->m_len = 6 * NFSX_UNSIGNED; 1495 bpos = ((caddr_t)tl) + mreq->m_len; 1496 *tl++ = txdr_unsigned(nd->nd_retxid); 1497 *tl++ = rpc_reply; 1498 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1499 *tl++ = rpc_msgdenied; 1500 if (err & NFSERR_AUTHERR) { 1501 *tl++ = rpc_autherr; 1502 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1503 mreq->m_len -= NFSX_UNSIGNED; 1504 bpos -= NFSX_UNSIGNED; 1505 } else { 1506 *tl++ = rpc_mismatch; 1507 *tl++ = txdr_unsigned(RPC_VER2); 1508 *tl = txdr_unsigned(RPC_VER2); 1509 } 1510 } else { 1511 *tl++ = rpc_msgaccepted; 1512 1513 /* 1514 * For Kerberos authentication, we must send the nickname 1515 * verifier back, otherwise just RPCAUTH_NULL. 1516 */ 1517 if (nd->nd_flag & ND_KERBFULL) { 1518 struct nfsuid *nuidp; 1519 struct timeval ktvin, ktvout; 1520 1521 memset(&ktvout, 0, sizeof ktvout); /* XXX gcc */ 1522 1523 LIST_FOREACH(nuidp, 1524 NUIDHASH(slp, kauth_cred_geteuid(nd->nd_cr)), 1525 nu_hash) { 1526 if (kauth_cred_geteuid(nuidp->nu_cr) == 1527 kauth_cred_geteuid(nd->nd_cr) && 1528 (!nd->nd_nam2 || netaddr_match( 1529 NU_NETFAM(nuidp), &nuidp->nu_haddr, 1530 nd->nd_nam2))) 1531 break; 1532 } 1533 if (nuidp) { 1534 ktvin.tv_sec = 1535 txdr_unsigned(nuidp->nu_timestamp.tv_sec 1536 - 1); 1537 ktvin.tv_usec = 1538 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1539 1540 /* 1541 * Encrypt the timestamp in ecb mode using the 1542 * session key. 1543 */ 1544 #ifdef NFSKERB 1545 XXX 1546 #endif 1547 1548 *tl++ = rpc_auth_kerb; 1549 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1550 *tl = ktvout.tv_sec; 1551 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1552 *tl++ = ktvout.tv_usec; 1553 *tl++ = txdr_unsigned( 1554 kauth_cred_geteuid(nuidp->nu_cr)); 1555 } else { 1556 *tl++ = 0; 1557 *tl++ = 0; 1558 } 1559 } else { 1560 *tl++ = 0; 1561 *tl++ = 0; 1562 } 1563 switch (err) { 1564 case EPROGUNAVAIL: 1565 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1566 break; 1567 case EPROGMISMATCH: 1568 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1569 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1570 if (nd->nd_flag & ND_NQNFS) { 1571 *tl++ = txdr_unsigned(3); 1572 *tl = txdr_unsigned(3); 1573 } else { 1574 *tl++ = txdr_unsigned(2); 1575 *tl = txdr_unsigned(3); 1576 } 1577 break; 1578 case EPROCUNAVAIL: 1579 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1580 break; 1581 case EBADRPC: 1582 *tl = txdr_unsigned(RPC_GARBAGE); 1583 break; 1584 default: 1585 *tl = 0; 1586 if (err != NFSERR_RETVOID) { 1587 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1588 if (err) 1589 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1590 else 1591 *tl = 0; 1592 } 1593 break; 1594 }; 1595 } 1596 1597 /* 1598 * For nqnfs, piggyback lease as requested. 1599 */ 1600 if ((nd->nd_flag & ND_NQNFS) && err == 0) { 1601 if (nd->nd_flag & ND_LEASE) { 1602 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1603 *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); 1604 *tl++ = txdr_unsigned(cache); 1605 *tl++ = txdr_unsigned(nd->nd_duration); 1606 txdr_hyper(*frev, tl); 1607 } else { 1608 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1609 *tl = 0; 1610 } 1611 } 1612 if (mrq != NULL) 1613 *mrq = mreq; 1614 *mbp = mb; 1615 *bposp = bpos; 1616 if (err != 0 && err != NFSERR_RETVOID) 1617 nfsstats.srvrpc_errs++; 1618 return (0); 1619 } 1620 1621 /* 1622 * Nfs timer routine 1623 * Scan the nfsreq list and retranmit any requests that have timed out 1624 * To avoid retransmission attempts on STREAM sockets (in the future) make 1625 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1626 */ 1627 void 1628 nfs_timer(arg) 1629 void *arg; /* never used */ 1630 { 1631 struct nfsreq *rep; 1632 struct mbuf *m; 1633 struct socket *so; 1634 struct nfsmount *nmp; 1635 int timeo; 1636 int s, error; 1637 #ifdef NFSSERVER 1638 struct nfssvc_sock *slp; 1639 static long lasttime = 0; 1640 u_quad_t cur_usec; 1641 #endif 1642 1643 s = splsoftnet(); 1644 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1645 nmp = rep->r_nmp; 1646 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1647 continue; 1648 if (nfs_sigintr(nmp, rep, rep->r_lwp)) { 1649 rep->r_flags |= R_SOFTTERM; 1650 continue; 1651 } 1652 if (rep->r_rtt >= 0) { 1653 rep->r_rtt++; 1654 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1655 timeo = nmp->nm_timeo; 1656 else 1657 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1658 if (nmp->nm_timeouts > 0) 1659 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1660 if (rep->r_rtt <= timeo) 1661 continue; 1662 if (nmp->nm_timeouts < 1663 (sizeof(nfs_backoff) / sizeof(nfs_backoff[0]))) 1664 nmp->nm_timeouts++; 1665 } 1666 /* 1667 * Check for server not responding 1668 */ 1669 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1670 rep->r_rexmit > nmp->nm_deadthresh) { 1671 nfs_msg(rep->r_lwp, 1672 nmp->nm_mountp->mnt_stat.f_mntfromname, 1673 "not responding"); 1674 rep->r_flags |= R_TPRINTFMSG; 1675 } 1676 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1677 nfsstats.rpctimeouts++; 1678 rep->r_flags |= R_SOFTTERM; 1679 continue; 1680 } 1681 if (nmp->nm_sotype != SOCK_DGRAM) { 1682 if (++rep->r_rexmit > NFS_MAXREXMIT) 1683 rep->r_rexmit = NFS_MAXREXMIT; 1684 continue; 1685 } 1686 if ((so = nmp->nm_so) == NULL) 1687 continue; 1688 1689 /* 1690 * If there is enough space and the window allows.. 1691 * Resend it 1692 * Set r_rtt to -1 in case we fail to send it now. 1693 */ 1694 rep->r_rtt = -1; 1695 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1696 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1697 (rep->r_flags & R_SENT) || 1698 nmp->nm_sent < nmp->nm_cwnd) && 1699 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1700 if (so->so_state & SS_ISCONNECTED) 1701 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1702 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1703 else 1704 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1705 nmp->nm_nam, (struct mbuf *)0, (struct lwp *)0); 1706 if (error) { 1707 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1708 #ifdef DEBUG 1709 printf("nfs_timer: ignoring error %d\n", 1710 error); 1711 #endif 1712 so->so_error = 0; 1713 } 1714 } else { 1715 /* 1716 * Iff first send, start timing 1717 * else turn timing off, backoff timer 1718 * and divide congestion window by 2. 1719 */ 1720 if (rep->r_flags & R_SENT) { 1721 rep->r_flags &= ~R_TIMING; 1722 if (++rep->r_rexmit > NFS_MAXREXMIT) 1723 rep->r_rexmit = NFS_MAXREXMIT; 1724 nmp->nm_cwnd >>= 1; 1725 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1726 nmp->nm_cwnd = NFS_CWNDSCALE; 1727 nfsstats.rpcretries++; 1728 } else { 1729 rep->r_flags |= R_SENT; 1730 nmp->nm_sent += NFS_CWNDSCALE; 1731 } 1732 rep->r_rtt = 0; 1733 } 1734 } 1735 } 1736 1737 #ifdef NFSSERVER 1738 /* 1739 * Call the nqnfs server timer once a second to handle leases. 1740 */ 1741 if (lasttime != time.tv_sec) { 1742 lasttime = time.tv_sec; 1743 nqnfs_serverd(); 1744 } 1745 1746 /* 1747 * Scan the write gathering queues for writes that need to be 1748 * completed now. 1749 */ 1750 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1751 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 1752 if (LIST_FIRST(&slp->ns_tq) && 1753 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) 1754 nfsrv_wakenfsd(slp); 1755 } 1756 #endif /* NFSSERVER */ 1757 splx(s); 1758 callout_schedule(&nfs_timer_ch, nfs_ticks); 1759 } 1760 1761 /*ARGSUSED*/ 1762 void 1763 nfs_exit(p, v) 1764 struct proc *p; 1765 void *v; 1766 { 1767 struct nfsreq *rp; 1768 int s = splsoftnet(); 1769 1770 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 1771 if (rp->r_lwp && rp->r_lwp->l_proc == p) 1772 TAILQ_REMOVE(&nfs_reqq, rp, r_chain); 1773 } 1774 splx(s); 1775 } 1776 1777 /* 1778 * Test for a termination condition pending on the process. 1779 * This is used for NFSMNT_INT mounts. 1780 */ 1781 int 1782 nfs_sigintr(nmp, rep, l) 1783 struct nfsmount *nmp; 1784 struct nfsreq *rep; 1785 struct lwp *l; 1786 { 1787 sigset_t ss; 1788 1789 if (rep && (rep->r_flags & R_SOFTTERM)) 1790 return (EINTR); 1791 if (!(nmp->nm_flag & NFSMNT_INT)) 1792 return (0); 1793 if (l) { 1794 sigpending1(l->l_proc, &ss); 1795 #if 0 1796 sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss); 1797 #endif 1798 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1799 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1800 sigismember(&ss, SIGQUIT)) 1801 return (EINTR); 1802 } 1803 return (0); 1804 } 1805 1806 /* 1807 * Lock a socket against others. 1808 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1809 * and also to avoid race conditions between the processes with nfs requests 1810 * in progress when a reconnect is necessary. 1811 */ 1812 int 1813 nfs_sndlock(flagp, rep) 1814 int *flagp; 1815 struct nfsreq *rep; 1816 { 1817 struct lwp *l; 1818 int slpflag = 0, slptimeo = 0; 1819 1820 if (rep) { 1821 l = rep->r_lwp; 1822 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1823 slpflag = PCATCH; 1824 } else 1825 l = (struct lwp *)0; 1826 while (*flagp & NFSMNT_SNDLOCK) { 1827 if (rep && nfs_sigintr(rep->r_nmp, rep, l)) 1828 return (EINTR); 1829 *flagp |= NFSMNT_WANTSND; 1830 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1831 slptimeo); 1832 if (slpflag == PCATCH) { 1833 slpflag = 0; 1834 slptimeo = 2 * hz; 1835 } 1836 } 1837 *flagp |= NFSMNT_SNDLOCK; 1838 return (0); 1839 } 1840 1841 /* 1842 * Unlock the stream socket for others. 1843 */ 1844 void 1845 nfs_sndunlock(flagp) 1846 int *flagp; 1847 { 1848 1849 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1850 panic("nfs sndunlock"); 1851 *flagp &= ~NFSMNT_SNDLOCK; 1852 if (*flagp & NFSMNT_WANTSND) { 1853 *flagp &= ~NFSMNT_WANTSND; 1854 wakeup((caddr_t)flagp); 1855 } 1856 } 1857 1858 int 1859 nfs_rcvlock(rep) 1860 struct nfsreq *rep; 1861 { 1862 struct nfsmount *nmp = rep->r_nmp; 1863 int *flagp = &nmp->nm_iflag; 1864 int slpflag, slptimeo = 0; 1865 int error = 0; 1866 1867 if (*flagp & NFSMNT_DISMNT) 1868 return EIO; 1869 1870 if (*flagp & NFSMNT_INT) 1871 slpflag = PCATCH; 1872 else 1873 slpflag = 0; 1874 simple_lock(&nmp->nm_slock); 1875 while (*flagp & NFSMNT_RCVLOCK) { 1876 if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) { 1877 error = EINTR; 1878 goto quit; 1879 } 1880 *flagp |= NFSMNT_WANTRCV; 1881 nmp->nm_waiters++; 1882 (void) ltsleep(flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1883 slptimeo, &nmp->nm_slock); 1884 nmp->nm_waiters--; 1885 if (*flagp & NFSMNT_DISMNT) { 1886 wakeup(&nmp->nm_waiters); 1887 error = EIO; 1888 goto quit; 1889 } 1890 /* If our reply was received while we were sleeping, 1891 * then just return without taking the lock to avoid a 1892 * situation where a single iod could 'capture' the 1893 * receive lock. 1894 */ 1895 if (rep->r_mrep != NULL) { 1896 error = EALREADY; 1897 goto quit; 1898 } 1899 if (slpflag == PCATCH) { 1900 slpflag = 0; 1901 slptimeo = 2 * hz; 1902 } 1903 } 1904 *flagp |= NFSMNT_RCVLOCK; 1905 quit: 1906 simple_unlock(&nmp->nm_slock); 1907 return error; 1908 } 1909 1910 /* 1911 * Unlock the stream socket for others. 1912 */ 1913 void 1914 nfs_rcvunlock(nmp) 1915 struct nfsmount *nmp; 1916 { 1917 int *flagp = &nmp->nm_iflag; 1918 1919 simple_lock(&nmp->nm_slock); 1920 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1921 panic("nfs rcvunlock"); 1922 *flagp &= ~NFSMNT_RCVLOCK; 1923 if (*flagp & NFSMNT_WANTRCV) { 1924 *flagp &= ~NFSMNT_WANTRCV; 1925 wakeup((caddr_t)flagp); 1926 } 1927 simple_unlock(&nmp->nm_slock); 1928 } 1929 1930 /* 1931 * Parse an RPC request 1932 * - verify it 1933 * - allocate and fill in the cred. 1934 */ 1935 int 1936 nfs_getreq(nd, nfsd, has_header) 1937 struct nfsrv_descript *nd; 1938 struct nfsd *nfsd; 1939 int has_header; 1940 { 1941 int len, i; 1942 u_int32_t *tl; 1943 int32_t t1; 1944 struct uio uio; 1945 struct iovec iov; 1946 caddr_t dpos, cp2, cp; 1947 u_int32_t nfsvers, auth_type; 1948 uid_t nickuid; 1949 int error = 0, nqnfs = 0, ticklen; 1950 struct mbuf *mrep, *md; 1951 struct nfsuid *nuidp; 1952 struct timeval tvin, tvout; 1953 1954 memset(&tvout, 0, sizeof tvout); /* XXX gcc */ 1955 1956 KASSERT(nd->nd_cr == NULL); 1957 mrep = nd->nd_mrep; 1958 md = nd->nd_md; 1959 dpos = nd->nd_dpos; 1960 if (has_header) { 1961 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 1962 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 1963 if (*tl++ != rpc_call) { 1964 m_freem(mrep); 1965 return (EBADRPC); 1966 } 1967 } else 1968 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 1969 nd->nd_repstat = 0; 1970 nd->nd_flag = 0; 1971 if (*tl++ != rpc_vers) { 1972 nd->nd_repstat = ERPCMISMATCH; 1973 nd->nd_procnum = NFSPROC_NOOP; 1974 return (0); 1975 } 1976 if (*tl != nfs_prog) { 1977 if (*tl == nqnfs_prog) 1978 nqnfs++; 1979 else { 1980 nd->nd_repstat = EPROGUNAVAIL; 1981 nd->nd_procnum = NFSPROC_NOOP; 1982 return (0); 1983 } 1984 } 1985 tl++; 1986 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 1987 if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || 1988 (nfsvers != NQNFS_VER3 && nqnfs)) { 1989 nd->nd_repstat = EPROGMISMATCH; 1990 nd->nd_procnum = NFSPROC_NOOP; 1991 return (0); 1992 } 1993 if (nqnfs) 1994 nd->nd_flag = (ND_NFSV3 | ND_NQNFS); 1995 else if (nfsvers == NFS_VER3) 1996 nd->nd_flag = ND_NFSV3; 1997 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 1998 if (nd->nd_procnum == NFSPROC_NULL) 1999 return (0); 2000 if (nd->nd_procnum >= NFS_NPROCS || 2001 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || 2002 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 2003 nd->nd_repstat = EPROCUNAVAIL; 2004 nd->nd_procnum = NFSPROC_NOOP; 2005 return (0); 2006 } 2007 if ((nd->nd_flag & ND_NFSV3) == 0) 2008 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 2009 auth_type = *tl++; 2010 len = fxdr_unsigned(int, *tl++); 2011 if (len < 0 || len > RPCAUTH_MAXSIZ) { 2012 m_freem(mrep); 2013 return (EBADRPC); 2014 } 2015 2016 nd->nd_flag &= ~ND_KERBAUTH; 2017 /* 2018 * Handle auth_unix or auth_kerb. 2019 */ 2020 if (auth_type == rpc_auth_unix) { 2021 uid_t uid; 2022 gid_t gid, *grbuf; 2023 2024 nd->nd_cr = kauth_cred_alloc(); 2025 len = fxdr_unsigned(int, *++tl); 2026 if (len < 0 || len > NFS_MAXNAMLEN) { 2027 m_freem(mrep); 2028 error = EBADRPC; 2029 goto errout; 2030 } 2031 nfsm_adv(nfsm_rndup(len)); 2032 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2033 2034 uid = fxdr_unsigned(uid_t, *tl++); 2035 gid = fxdr_unsigned(gid_t, *tl++); 2036 kauth_cred_setuid(nd->nd_cr, uid); 2037 kauth_cred_setgid(nd->nd_cr, gid); 2038 kauth_cred_seteuid(nd->nd_cr, uid); 2039 kauth_cred_setsvuid(nd->nd_cr, gid); 2040 kauth_cred_setegid(nd->nd_cr, uid); 2041 kauth_cred_setsvgid(nd->nd_cr, gid); 2042 2043 len = fxdr_unsigned(int, *tl); 2044 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 2045 m_freem(mrep); 2046 error = EBADRPC; 2047 goto errout; 2048 } 2049 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 2050 2051 grbuf = malloc(len * sizeof(gid_t), M_TEMP, M_WAITOK); 2052 for (i = 0; i < len; i++) { 2053 if (i < NGROUPS) /* XXX elad */ 2054 grbuf[i] = fxdr_unsigned(gid_t, *tl++); 2055 else 2056 tl++; 2057 } 2058 kauth_cred_setgroups(nd->nd_cr, grbuf, min(len, NGROUPS), -1); 2059 free(grbuf, M_TEMP); 2060 2061 len = fxdr_unsigned(int, *++tl); 2062 if (len < 0 || len > RPCAUTH_MAXSIZ) { 2063 m_freem(mrep); 2064 error = EBADRPC; 2065 goto errout; 2066 } 2067 if (len > 0) 2068 nfsm_adv(nfsm_rndup(len)); 2069 } else if (auth_type == rpc_auth_kerb) { 2070 switch (fxdr_unsigned(int, *tl++)) { 2071 case RPCAKN_FULLNAME: 2072 ticklen = fxdr_unsigned(int, *tl); 2073 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 2074 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 2075 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 2076 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 2077 m_freem(mrep); 2078 error = EBADRPC; 2079 goto errout; 2080 } 2081 uio.uio_offset = 0; 2082 uio.uio_iov = &iov; 2083 uio.uio_iovcnt = 1; 2084 UIO_SETUP_SYSSPACE(&uio); 2085 iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; 2086 iov.iov_len = RPCAUTH_MAXSIZ - 4; 2087 nfsm_mtouio(&uio, uio.uio_resid); 2088 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2089 if (*tl++ != rpc_auth_kerb || 2090 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 2091 printf("Bad kerb verifier\n"); 2092 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2093 nd->nd_procnum = NFSPROC_NOOP; 2094 return (0); 2095 } 2096 nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); 2097 tl = (u_int32_t *)cp; 2098 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 2099 printf("Not fullname kerb verifier\n"); 2100 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2101 nd->nd_procnum = NFSPROC_NOOP; 2102 return (0); 2103 } 2104 cp += NFSX_UNSIGNED; 2105 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 2106 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 2107 nd->nd_flag |= ND_KERBFULL; 2108 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 2109 break; 2110 case RPCAKN_NICKNAME: 2111 if (len != 2 * NFSX_UNSIGNED) { 2112 printf("Kerb nickname short\n"); 2113 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 2114 nd->nd_procnum = NFSPROC_NOOP; 2115 return (0); 2116 } 2117 nickuid = fxdr_unsigned(uid_t, *tl); 2118 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2119 if (*tl++ != rpc_auth_kerb || 2120 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 2121 printf("Kerb nick verifier bad\n"); 2122 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2123 nd->nd_procnum = NFSPROC_NOOP; 2124 return (0); 2125 } 2126 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2127 tvin.tv_sec = *tl++; 2128 tvin.tv_usec = *tl; 2129 2130 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), 2131 nu_hash) { 2132 if (kauth_cred_geteuid(nuidp->nu_cr) == nickuid && 2133 (!nd->nd_nam2 || 2134 netaddr_match(NU_NETFAM(nuidp), 2135 &nuidp->nu_haddr, nd->nd_nam2))) 2136 break; 2137 } 2138 if (!nuidp) { 2139 nd->nd_repstat = 2140 (NFSERR_AUTHERR|AUTH_REJECTCRED); 2141 nd->nd_procnum = NFSPROC_NOOP; 2142 return (0); 2143 } 2144 2145 /* 2146 * Now, decrypt the timestamp using the session key 2147 * and validate it. 2148 */ 2149 #ifdef NFSKERB 2150 XXX 2151 #endif 2152 2153 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 2154 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 2155 if (nuidp->nu_expire < time.tv_sec || 2156 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 2157 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 2158 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 2159 nuidp->nu_expire = 0; 2160 nd->nd_repstat = 2161 (NFSERR_AUTHERR|AUTH_REJECTVERF); 2162 nd->nd_procnum = NFSPROC_NOOP; 2163 return (0); 2164 } 2165 kauth_cred_hold(nuidp->nu_cr); 2166 nd->nd_cr = nuidp->nu_cr; 2167 nd->nd_flag |= ND_KERBNICK; 2168 } 2169 } else { 2170 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 2171 nd->nd_procnum = NFSPROC_NOOP; 2172 return (0); 2173 } 2174 2175 /* 2176 * For nqnfs, get piggybacked lease request. 2177 */ 2178 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 2179 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2180 nd->nd_flag |= fxdr_unsigned(int, *tl); 2181 if (nd->nd_flag & ND_LEASE) { 2182 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2183 nd->nd_duration = fxdr_unsigned(u_int32_t, *tl); 2184 } else 2185 nd->nd_duration = NQ_MINLEASE; 2186 } else 2187 nd->nd_duration = NQ_MINLEASE; 2188 nd->nd_md = md; 2189 nd->nd_dpos = dpos; 2190 KASSERT((nd->nd_cr == NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) != 0) 2191 || (nd->nd_cr != NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) == 0)); 2192 return (0); 2193 nfsmout: 2194 errout: 2195 KASSERT(error != 0); 2196 if (nd->nd_cr != NULL) { 2197 kauth_cred_free(nd->nd_cr); 2198 nd->nd_cr = NULL; 2199 } 2200 return (error); 2201 } 2202 2203 int 2204 nfs_msg(l, server, msg) 2205 struct lwp *l; 2206 const char *server, *msg; 2207 { 2208 tpr_t tpr; 2209 2210 if (l) 2211 tpr = tprintf_open(l->l_proc); 2212 else 2213 tpr = NULL; 2214 tprintf(tpr, "nfs server %s: %s\n", server, msg); 2215 tprintf_close(tpr); 2216 return (0); 2217 } 2218 2219 #ifdef NFSSERVER 2220 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 2221 struct nfssvc_sock *, struct lwp *, 2222 struct mbuf **)) = { 2223 nfsrv_null, 2224 nfsrv_getattr, 2225 nfsrv_setattr, 2226 nfsrv_lookup, 2227 nfsrv3_access, 2228 nfsrv_readlink, 2229 nfsrv_read, 2230 nfsrv_write, 2231 nfsrv_create, 2232 nfsrv_mkdir, 2233 nfsrv_symlink, 2234 nfsrv_mknod, 2235 nfsrv_remove, 2236 nfsrv_rmdir, 2237 nfsrv_rename, 2238 nfsrv_link, 2239 nfsrv_readdir, 2240 nfsrv_readdirplus, 2241 nfsrv_statfs, 2242 nfsrv_fsinfo, 2243 nfsrv_pathconf, 2244 nfsrv_commit, 2245 nqnfsrv_getlease, 2246 nqnfsrv_vacated, 2247 nfsrv_noop, 2248 nfsrv_noop 2249 }; 2250 2251 /* 2252 * Socket upcall routine for the nfsd sockets. 2253 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 2254 * Essentially do as much as possible non-blocking, else punt and it will 2255 * be called with M_WAIT from an nfsd. 2256 */ 2257 void 2258 nfsrv_rcv(so, arg, waitflag) 2259 struct socket *so; 2260 caddr_t arg; 2261 int waitflag; 2262 { 2263 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 2264 struct mbuf *m; 2265 struct mbuf *mp, *nam; 2266 struct uio auio; 2267 int flags, error; 2268 int setflags = 0; 2269 2270 error = nfsdsock_lock(slp, (waitflag != M_DONTWAIT)); 2271 if (error) { 2272 setflags |= SLP_NEEDQ; 2273 goto dorecs_unlocked; 2274 } 2275 2276 KASSERT(so == slp->ns_so); 2277 #define NFS_TEST_HEAVY 2278 #ifdef NFS_TEST_HEAVY 2279 /* 2280 * Define this to test for nfsds handling this under heavy load. 2281 * 2282 * XXX it isn't safe to call so_receive from so_upcall context. 2283 */ 2284 if (waitflag == M_DONTWAIT) { 2285 setflags |= SLP_NEEDQ; 2286 goto dorecs; 2287 } 2288 #endif 2289 simple_lock(&slp->ns_lock); 2290 slp->ns_flag &= ~SLP_NEEDQ; 2291 simple_unlock(&slp->ns_lock); 2292 if (so->so_type == SOCK_STREAM) { 2293 #ifndef NFS_TEST_HEAVY 2294 /* 2295 * If there are already records on the queue, defer soreceive() 2296 * to an nfsd so that there is feedback to the TCP layer that 2297 * the nfs servers are heavily loaded. 2298 */ 2299 if (slp->ns_rec && waitflag == M_DONTWAIT) { 2300 setflags |= SLP_NEEDQ; 2301 goto dorecs; 2302 } 2303 #endif 2304 2305 /* 2306 * Do soreceive(). 2307 */ 2308 auio.uio_resid = 1000000000; 2309 /* not need to setup uio_vmspace */ 2310 flags = MSG_DONTWAIT; 2311 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, &flags); 2312 if (error || mp == NULL) { 2313 if (error == EWOULDBLOCK) 2314 setflags |= SLP_NEEDQ; 2315 else 2316 setflags |= SLP_DISCONN; 2317 goto dorecs; 2318 } 2319 m = mp; 2320 if (slp->ns_rawend) { 2321 slp->ns_rawend->m_next = m; 2322 slp->ns_cc += 1000000000 - auio.uio_resid; 2323 } else { 2324 slp->ns_raw = m; 2325 slp->ns_cc = 1000000000 - auio.uio_resid; 2326 } 2327 while (m->m_next) 2328 m = m->m_next; 2329 slp->ns_rawend = m; 2330 2331 /* 2332 * Now try and parse record(s) out of the raw stream data. 2333 */ 2334 error = nfsrv_getstream(slp, waitflag); 2335 if (error) { 2336 if (error == EPERM) 2337 setflags |= SLP_DISCONN; 2338 else 2339 setflags |= SLP_NEEDQ; 2340 } 2341 } else { 2342 do { 2343 auio.uio_resid = 1000000000; 2344 /* not need to setup uio_vmspace */ 2345 flags = MSG_DONTWAIT; 2346 error = (*so->so_receive)(so, &nam, &auio, &mp, NULL, 2347 &flags); 2348 if (mp) { 2349 if (nam) { 2350 m = nam; 2351 m->m_next = mp; 2352 } else 2353 m = mp; 2354 if (slp->ns_recend) 2355 slp->ns_recend->m_nextpkt = m; 2356 else 2357 slp->ns_rec = m; 2358 slp->ns_recend = m; 2359 m->m_nextpkt = (struct mbuf *)0; 2360 } 2361 if (error) { 2362 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 2363 && error != EWOULDBLOCK) { 2364 setflags |= SLP_DISCONN; 2365 goto dorecs; 2366 } 2367 } 2368 } while (mp); 2369 } 2370 dorecs: 2371 nfsdsock_unlock(slp); 2372 2373 dorecs_unlocked: 2374 /* 2375 * Now try and process the request records, non-blocking. 2376 */ 2377 if (setflags) { 2378 simple_lock(&slp->ns_lock); 2379 slp->ns_flag |= setflags; 2380 simple_unlock(&slp->ns_lock); 2381 } 2382 if (waitflag == M_DONTWAIT && 2383 (slp->ns_rec || (slp->ns_flag & (SLP_DISCONN | SLP_NEEDQ)) != 0)) { 2384 nfsrv_wakenfsd(slp); 2385 } 2386 } 2387 2388 int 2389 nfsdsock_lock(struct nfssvc_sock *slp, boolean_t waitok) 2390 { 2391 2392 simple_lock(&slp->ns_lock); 2393 while ((slp->ns_flag & (SLP_BUSY|SLP_VALID)) == SLP_BUSY) { 2394 if (!waitok) { 2395 simple_unlock(&slp->ns_lock); 2396 return EWOULDBLOCK; 2397 } 2398 slp->ns_flag |= SLP_WANT; 2399 ltsleep(&slp->ns_flag, PSOCK, "nslock", 0, &slp->ns_lock); 2400 } 2401 if ((slp->ns_flag & SLP_VALID) == 0) { 2402 simple_unlock(&slp->ns_lock); 2403 return EINVAL; 2404 } 2405 slp->ns_flag |= SLP_BUSY; 2406 simple_unlock(&slp->ns_lock); 2407 2408 return 0; 2409 } 2410 2411 void 2412 nfsdsock_unlock(struct nfssvc_sock *slp) 2413 { 2414 2415 KASSERT((slp->ns_flag & SLP_BUSY) != 0); 2416 2417 simple_lock(&slp->ns_lock); 2418 if ((slp->ns_flag & SLP_WANT) != 0) { 2419 wakeup(&slp->ns_flag); 2420 } 2421 slp->ns_flag &= ~(SLP_BUSY|SLP_WANT); 2422 simple_unlock(&slp->ns_lock); 2423 } 2424 2425 int 2426 nfsdsock_drain(struct nfssvc_sock *slp) 2427 { 2428 int error = 0; 2429 2430 simple_lock(&slp->ns_lock); 2431 if ((slp->ns_flag & SLP_VALID) == 0) { 2432 error = EINVAL; 2433 goto done; 2434 } 2435 slp->ns_flag &= ~SLP_VALID; 2436 while ((slp->ns_flag & SLP_BUSY) != 0) { 2437 slp->ns_flag |= SLP_WANT; 2438 ltsleep(&slp->ns_flag, PSOCK, "nsdrain", 0, &slp->ns_lock); 2439 } 2440 done: 2441 simple_unlock(&slp->ns_lock); 2442 2443 return error; 2444 } 2445 2446 /* 2447 * Try and extract an RPC request from the mbuf data list received on a 2448 * stream socket. The "waitflag" argument indicates whether or not it 2449 * can sleep. 2450 */ 2451 int 2452 nfsrv_getstream(slp, waitflag) 2453 struct nfssvc_sock *slp; 2454 int waitflag; 2455 { 2456 struct mbuf *m, **mpp; 2457 struct mbuf *recm; 2458 u_int32_t recmark; 2459 int error = 0; 2460 2461 for (;;) { 2462 if (slp->ns_reclen == 0) { 2463 if (slp->ns_cc < NFSX_UNSIGNED) { 2464 break; 2465 } 2466 m = slp->ns_raw; 2467 m_copydata(m, 0, NFSX_UNSIGNED, (caddr_t)&recmark); 2468 m_adj(m, NFSX_UNSIGNED); 2469 slp->ns_cc -= NFSX_UNSIGNED; 2470 recmark = ntohl(recmark); 2471 slp->ns_reclen = recmark & ~0x80000000; 2472 if (recmark & 0x80000000) 2473 slp->ns_flag |= SLP_LASTFRAG; 2474 else 2475 slp->ns_flag &= ~SLP_LASTFRAG; 2476 if (slp->ns_reclen > NFS_MAXPACKET) { 2477 error = EPERM; 2478 break; 2479 } 2480 } 2481 2482 /* 2483 * Now get the record part. 2484 * 2485 * Note that slp->ns_reclen may be 0. Linux sometimes 2486 * generates 0-length records. 2487 */ 2488 if (slp->ns_cc == slp->ns_reclen) { 2489 recm = slp->ns_raw; 2490 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 2491 slp->ns_cc = slp->ns_reclen = 0; 2492 } else if (slp->ns_cc > slp->ns_reclen) { 2493 recm = slp->ns_raw; 2494 m = m_split(recm, slp->ns_reclen, waitflag); 2495 if (m == NULL) { 2496 error = EWOULDBLOCK; 2497 break; 2498 } 2499 m_claimm(recm, &nfs_mowner); 2500 slp->ns_raw = m; 2501 if (m->m_next == NULL) 2502 slp->ns_rawend = m; 2503 slp->ns_cc -= slp->ns_reclen; 2504 slp->ns_reclen = 0; 2505 } else { 2506 break; 2507 } 2508 2509 /* 2510 * Accumulate the fragments into a record. 2511 */ 2512 mpp = &slp->ns_frag; 2513 while (*mpp) 2514 mpp = &((*mpp)->m_next); 2515 *mpp = recm; 2516 if (slp->ns_flag & SLP_LASTFRAG) { 2517 if (slp->ns_recend) 2518 slp->ns_recend->m_nextpkt = slp->ns_frag; 2519 else 2520 slp->ns_rec = slp->ns_frag; 2521 slp->ns_recend = slp->ns_frag; 2522 slp->ns_frag = (struct mbuf *)0; 2523 } 2524 } 2525 2526 return error; 2527 } 2528 2529 /* 2530 * Parse an RPC header. 2531 */ 2532 int 2533 nfsrv_dorec(slp, nfsd, ndp) 2534 struct nfssvc_sock *slp; 2535 struct nfsd *nfsd; 2536 struct nfsrv_descript **ndp; 2537 { 2538 struct mbuf *m, *nam; 2539 struct nfsrv_descript *nd; 2540 int error; 2541 2542 *ndp = NULL; 2543 2544 if (nfsdsock_lock(slp, TRUE)) { 2545 return ENOBUFS; 2546 } 2547 m = slp->ns_rec; 2548 if (m == NULL) { 2549 nfsdsock_unlock(slp); 2550 return ENOBUFS; 2551 } 2552 slp->ns_rec = m->m_nextpkt; 2553 if (slp->ns_rec) 2554 m->m_nextpkt = NULL; 2555 else 2556 slp->ns_recend = NULL; 2557 nfsdsock_unlock(slp); 2558 2559 if (m->m_type == MT_SONAME) { 2560 nam = m; 2561 m = m->m_next; 2562 nam->m_next = NULL; 2563 } else 2564 nam = NULL; 2565 nd = nfsdreq_alloc(); 2566 nd->nd_md = nd->nd_mrep = m; 2567 nd->nd_nam2 = nam; 2568 nd->nd_dpos = mtod(m, caddr_t); 2569 error = nfs_getreq(nd, nfsd, TRUE); 2570 if (error) { 2571 m_freem(nam); 2572 nfsdreq_free(nd); 2573 return (error); 2574 } 2575 *ndp = nd; 2576 nfsd->nfsd_nd = nd; 2577 return (0); 2578 } 2579 2580 /* 2581 * Search for a sleeping nfsd and wake it up. 2582 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 2583 * running nfsds will go look for the work in the nfssvc_sock list. 2584 */ 2585 void 2586 nfsrv_wakenfsd(slp) 2587 struct nfssvc_sock *slp; 2588 { 2589 struct nfsd *nd; 2590 2591 if ((slp->ns_flag & SLP_VALID) == 0) 2592 return; 2593 simple_lock(&nfsd_slock); 2594 if (slp->ns_flag & SLP_DOREC) { 2595 simple_unlock(&nfsd_slock); 2596 return; 2597 } 2598 nd = SLIST_FIRST(&nfsd_idle_head); 2599 if (nd) { 2600 SLIST_REMOVE_HEAD(&nfsd_idle_head, nfsd_idle); 2601 simple_unlock(&nfsd_slock); 2602 2603 if (nd->nfsd_slp) 2604 panic("nfsd wakeup"); 2605 slp->ns_sref++; 2606 nd->nfsd_slp = slp; 2607 wakeup(nd); 2608 return; 2609 } 2610 slp->ns_flag |= SLP_DOREC; 2611 nfsd_head_flag |= NFSD_CHECKSLP; 2612 TAILQ_INSERT_TAIL(&nfssvc_sockpending, slp, ns_pending); 2613 simple_unlock(&nfsd_slock); 2614 } 2615 2616 int 2617 nfsdsock_sendreply(struct nfssvc_sock *slp, struct nfsrv_descript *nd) 2618 { 2619 int error; 2620 2621 if (nd->nd_mrep != NULL) { 2622 m_freem(nd->nd_mrep); 2623 nd->nd_mrep = NULL; 2624 } 2625 2626 simple_lock(&slp->ns_lock); 2627 if ((slp->ns_flag & SLP_SENDING) != 0) { 2628 SIMPLEQ_INSERT_TAIL(&slp->ns_sendq, nd, nd_sendq); 2629 simple_unlock(&slp->ns_lock); 2630 return 0; 2631 } 2632 KASSERT(SIMPLEQ_EMPTY(&slp->ns_sendq)); 2633 slp->ns_flag |= SLP_SENDING; 2634 simple_unlock(&slp->ns_lock); 2635 2636 again: 2637 error = nfs_send(slp->ns_so, nd->nd_nam2, nd->nd_mreq, NULL, curlwp); 2638 if (nd->nd_nam2) { 2639 m_free(nd->nd_nam2); 2640 } 2641 nfsdreq_free(nd); 2642 2643 simple_lock(&slp->ns_lock); 2644 KASSERT((slp->ns_flag & SLP_SENDING) != 0); 2645 nd = SIMPLEQ_FIRST(&slp->ns_sendq); 2646 if (nd != NULL) { 2647 SIMPLEQ_REMOVE_HEAD(&slp->ns_sendq, nd_sendq); 2648 simple_unlock(&slp->ns_lock); 2649 goto again; 2650 } 2651 slp->ns_flag &= ~SLP_SENDING; 2652 simple_unlock(&slp->ns_lock); 2653 2654 return error; 2655 } 2656 #endif /* NFSSERVER */ 2657 2658 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) 2659 static struct pool nfs_srvdesc_pool; 2660 2661 void 2662 nfsdreq_init(void) 2663 { 2664 2665 pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript), 2666 0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr); 2667 } 2668 2669 struct nfsrv_descript * 2670 nfsdreq_alloc(void) 2671 { 2672 struct nfsrv_descript *nd; 2673 2674 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK); 2675 nd->nd_cr = NULL; 2676 return nd; 2677 } 2678 2679 void 2680 nfsdreq_free(struct nfsrv_descript *nd) 2681 { 2682 kauth_cred_t cr; 2683 2684 cr = nd->nd_cr; 2685 if (cr != NULL) { 2686 KASSERT(kauth_cred_getrefcnt(cr) == 1); 2687 kauth_cred_free(cr); 2688 } 2689 pool_put(&nfs_srvdesc_pool, nd); 2690 } 2691 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */ 2692
Indexes created Sun Sep 21 20:09:37 GMT 2025