nfs_socket.c revision 1.170.2.3
1 /* $NetBSD: nfs_socket.c,v 1.170.2.3 2009/05/04 10:28:53 yamt Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1991, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 35 */ 36 37 /* 38 * Socket operations for use by nfs 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: nfs_socket.c,v 1.170.2.3 2009/05/04 10:28:53 yamt Exp $"); 43 44 #ifdef _KERNEL_OPT 45 #include "fs_nfs.h" 46 #include "opt_nfs.h" 47 #include "opt_mbuftrace.h" 48 #endif 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 #ifdef DEBUG 118 /* 119 * Avoid spamming the console with debugging messages. We only print 120 * the nfs timer and reply error debugs every 10 seconds. 121 */ 122 static const struct timeval nfs_err_interval = { 10, 0 }; 123 static struct timeval nfs_reply_last_err_time; 124 static struct timeval nfs_timer_last_err_time; 125 #endif 126 127 /* 128 * Defines which timer to use for the procnum. 129 * 0 - default 130 * 1 - getattr 131 * 2 - lookup 132 * 3 - read 133 * 4 - write 134 */ 135 static const int proct[NFS_NPROCS] = { 136 [NFSPROC_NULL] = 0, 137 [NFSPROC_GETATTR] = 1, 138 [NFSPROC_SETATTR] = 0, 139 [NFSPROC_LOOKUP] = 2, 140 [NFSPROC_ACCESS] = 1, 141 [NFSPROC_READLINK] = 3, 142 [NFSPROC_READ] = 3, 143 [NFSPROC_WRITE] = 4, 144 [NFSPROC_CREATE] = 0, 145 [NFSPROC_MKDIR] = 0, 146 [NFSPROC_SYMLINK] = 0, 147 [NFSPROC_MKNOD] = 0, 148 [NFSPROC_REMOVE] = 0, 149 [NFSPROC_RMDIR] = 0, 150 [NFSPROC_RENAME] = 0, 151 [NFSPROC_LINK] = 0, 152 [NFSPROC_READDIR] = 3, 153 [NFSPROC_READDIRPLUS] = 3, 154 [NFSPROC_FSSTAT] = 0, 155 [NFSPROC_FSINFO] = 0, 156 [NFSPROC_PATHCONF] = 0, 157 [NFSPROC_COMMIT] = 0, 158 [NFSPROC_NOOP] = 0, 159 }; 160 161 /* 162 * There is a congestion window for outstanding rpcs maintained per mount 163 * point. The cwnd size is adjusted in roughly the way that: 164 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 165 * SIGCOMM '88". ACM, August 1988. 166 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 167 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 168 * of rpcs is in progress. 169 * (The sent count and cwnd are scaled for integer arith.) 170 * Variants of "slow start" were tried and were found to be too much of a 171 * performance hit (ave. rtt 3 times larger), 172 * I suspect due to the large rtt that nfs rpcs have. 173 */ 174 #define NFS_CWNDSCALE 256 175 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 176 static const int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 177 int nfsrtton = 0; 178 struct nfsrtt nfsrtt; 179 kmutex_t nfs_reqq_lock; 180 struct nfsreqhead nfs_reqq; 181 static callout_t nfs_timer_ch; 182 static struct evcnt nfs_timer_ev; 183 static struct evcnt nfs_timer_start_ev; 184 static struct evcnt nfs_timer_stop_ev; 185 static kmutex_t nfs_timer_lock; 186 static bool (*nfs_timer_srvvec)(void); 187 188 #ifdef NFS 189 static int nfs_sndlock(struct nfsmount *, struct nfsreq *); 190 static void nfs_sndunlock(struct nfsmount *); 191 #endif 192 static int nfs_rcvlock(struct nfsmount *, struct nfsreq *); 193 static void nfs_rcvunlock(struct nfsmount *); 194 195 /* 196 * Initialize sockets and congestion for a new NFS connection. 197 * We do not free the sockaddr if error. 198 */ 199 int 200 nfs_connect(struct nfsmount *nmp, struct nfsreq *rep, struct lwp *l) 201 { 202 struct socket *so; 203 int error, rcvreserve, sndreserve; 204 struct sockaddr *saddr; 205 struct sockaddr_in *sin; 206 struct sockaddr_in6 *sin6; 207 struct mbuf *m; 208 int val; 209 210 KASSERT(rw_write_held(&nmp->nm_solock)); 211 KASSERT(nmp->nm_so == NULL); 212 213 saddr = mtod(nmp->nm_nam, struct sockaddr *); 214 error = socreate(saddr->sa_family, &nmp->nm_so, 215 nmp->nm_sotype, nmp->nm_soproto, l, NULL); 216 if (error) 217 goto bad; 218 so = nmp->nm_so; 219 #ifdef MBUFTRACE 220 so->so_mowner = &nfs_mowner; 221 so->so_rcv.sb_mowner = &nfs_mowner; 222 so->so_snd.sb_mowner = &nfs_mowner; 223 #endif 224 nmp->nm_soflags = so->so_proto->pr_flags; 225 226 /* 227 * Some servers require that the client port be a reserved port number. 228 */ 229 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 230 val = IP_PORTRANGE_LOW; 231 232 if ((error = so_setsockopt(NULL, so, IPPROTO_IP, IP_PORTRANGE, 233 &val, sizeof(val)))) 234 goto bad; 235 m = m_get(M_WAIT, MT_SONAME); 236 MCLAIM(m, so->so_mowner); 237 sin = mtod(m, struct sockaddr_in *); 238 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 239 sin->sin_family = AF_INET; 240 sin->sin_addr.s_addr = INADDR_ANY; 241 sin->sin_port = 0; 242 error = sobind(so, m, &lwp0); 243 m_freem(m); 244 if (error) 245 goto bad; 246 } 247 if (saddr->sa_family == AF_INET6 && (nmp->nm_flag & NFSMNT_RESVPORT)) { 248 val = IPV6_PORTRANGE_LOW; 249 250 if ((error = so_setsockopt(NULL, so, IPPROTO_IPV6, 251 IPV6_PORTRANGE, &val, sizeof(val)))) 252 goto bad; 253 m = m_get(M_WAIT, MT_SONAME); 254 MCLAIM(m, so->so_mowner); 255 sin6 = mtod(m, struct sockaddr_in6 *); 256 memset(sin6, 0, sizeof(*sin6)); 257 sin6->sin6_len = m->m_len = sizeof (struct sockaddr_in6); 258 sin6->sin6_family = AF_INET6; 259 error = sobind(so, m, &lwp0); 260 m_freem(m); 261 if (error) 262 goto bad; 263 } 264 265 /* 266 * Protocols that do not require connections may be optionally left 267 * unconnected for servers that reply from a port other than NFS_PORT. 268 */ 269 solock(so); 270 if (nmp->nm_flag & NFSMNT_NOCONN) { 271 if (nmp->nm_soflags & PR_CONNREQUIRED) { 272 sounlock(so); 273 error = ENOTCONN; 274 goto bad; 275 } 276 } else { 277 error = soconnect(so, nmp->nm_nam, l); 278 if (error) { 279 sounlock(so); 280 goto bad; 281 } 282 283 /* 284 * Wait for the connection to complete. Cribbed from the 285 * connect system call but with the wait timing out so 286 * that interruptible mounts don't hang here for a long time. 287 */ 288 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 289 (void)sowait(so, false, 2 * hz); 290 if ((so->so_state & SS_ISCONNECTING) && 291 so->so_error == 0 && rep && 292 (error = nfs_sigintr(nmp, rep, rep->r_lwp)) != 0){ 293 so->so_state &= ~SS_ISCONNECTING; 294 sounlock(so); 295 goto bad; 296 } 297 } 298 if (so->so_error) { 299 error = so->so_error; 300 so->so_error = 0; 301 sounlock(so); 302 goto bad; 303 } 304 } 305 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 306 so->so_rcv.sb_timeo = (5 * hz); 307 so->so_snd.sb_timeo = (5 * hz); 308 } else { 309 /* 310 * enable receive timeout to detect server crash and reconnect. 311 * otherwise, we can be stuck in soreceive forever. 312 */ 313 so->so_rcv.sb_timeo = (5 * hz); 314 so->so_snd.sb_timeo = 0; 315 } 316 if (nmp->nm_sotype == SOCK_DGRAM) { 317 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 318 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 319 NFS_MAXPKTHDR) * 2; 320 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 321 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 322 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 323 NFS_MAXPKTHDR) * 2; 324 } else { 325 sounlock(so); 326 if (nmp->nm_sotype != SOCK_STREAM) 327 panic("nfscon sotype"); 328 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 329 val = 1; 330 so_setsockopt(NULL, so, SOL_SOCKET, SO_KEEPALIVE, &val, 331 sizeof(val)); 332 } 333 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 334 val = 1; 335 so_setsockopt(NULL, so, IPPROTO_TCP, TCP_NODELAY, &val, 336 sizeof(val)); 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 mutex_enter(&nfs_reqq_lock); 355 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 356 NFS_TIMEO << 3; 357 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 358 nmp->nm_sdrtt[3] = 0; 359 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 360 nmp->nm_timeouts = 0; 361 mutex_exit(&nfs_reqq_lock); 362 return (0); 363 364 bad: 365 nfs_disconnect(nmp); 366 return (error); 367 } 368 369 /* 370 * Reconnect routine: 371 * Called when a connection is broken on a reliable protocol. 372 * - clean up the old socket 373 * - nfs_connect() again 374 * - set R_MUSTRESEND for all outstanding requests on mount point 375 * If this fails the mount point is DEAD! 376 * nb: Must be called with the nfs_sndlock() set on the mount point. 377 */ 378 int 379 nfs_reconnect(struct nfsreq *rep) 380 { 381 struct nfsmount *nmp = rep->r_nmp; 382 int error; 383 384 KASSERT(rep->r_nmp->nm_sndlwp == curlwp); 385 KASSERT(rep->r_nmp->nm_rcvlwp == curlwp); 386 if (!rw_tryupgrade(&nmp->nm_solock)) { 387 printf("%s: nmp=%p: upgrade failed\n", __func__, nmp); 388 return EAGAIN; 389 } 390 printf("%s: nmp=%p: upgrade succeeded\n", __func__, nmp); 391 nfs_disconnect(nmp); 392 while ((error = nfs_connect(nmp, rep, &lwp0)) != 0) { 393 if (error == EINTR || error == ERESTART) { 394 rw_downgrade(&nmp->nm_solock); 395 return EINTR; 396 } 397 kpause("nfscn2", false, hz, NULL); 398 } 399 400 rw_downgrade(&nmp->nm_solock); 401 return 0; 402 } 403 404 /* 405 * NFS disconnect. Clean up and unlink. 406 */ 407 void 408 nfs_disconnect(struct nfsmount *nmp) 409 { 410 struct nfsreq *rp; 411 struct socket *so; 412 int drain = 0; 413 414 KASSERT(rw_write_held(&nmp->nm_solock)); 415 if (nmp->nm_so) { 416 so = nmp->nm_so; 417 nmp->nm_so = NULL; 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 } 440 441 #ifdef DIAGNOSTIC 442 if (drain && (nmp->nm_waiters > 0)) 443 panic("nfs_disconnect: waiters left after drain?"); 444 #endif 445 446 /* 447 * Loop through outstanding request list and fix up all requests 448 * on old socket. 449 */ 450 mutex_enter(&nfs_reqq_lock); 451 TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { 452 if (rp->r_nmp == nmp) { 453 if ((rp->r_flags & R_MUSTRESEND) == 0) 454 rp->r_flags |= R_MUSTRESEND | R_REXMITTED; 455 rp->r_rexmit = 0; 456 if ((rp->r_rflags & RR_SENT) != 0) { 457 KASSERT(nmp->nm_sent >= NFS_CWNDSCALE); 458 rp->r_rflags &= ~RR_SENT; 459 nmp->nm_sent -= NFS_CWNDSCALE; 460 } 461 } 462 } 463 KASSERT(nmp->nm_sent == 0); 464 mutex_exit(&nfs_reqq_lock); 465 } 466 467 void 468 nfs_safedisconnect(struct nfsmount *nmp) 469 { 470 471 KASSERT(rw_write_held(&nmp->nm_solock)); 472 473 nfs_rcvlock(nmp, NULL); /* XXX ignored error return */ 474 nfs_disconnect(nmp); 475 nfs_rcvunlock(nmp); 476 } 477 478 /* 479 * This is the nfs send routine. For connection based socket types, it 480 * must be called with an nfs_sndlock() on the socket. 481 * "rep == NULL" indicates that it has been called from a server. 482 * For the client side: 483 * - return EINTR if the RPC is terminated, 0 otherwise 484 * - set R_MUSTRESEND if the send fails for any reason 485 * - do any cleanup required by recoverable socket errors (? ? ?) 486 * For the server side: 487 * - return EINTR or ERESTART if interrupted by a signal 488 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 489 * - do any cleanup required by recoverable socket errors (? ? ?) 490 */ 491 int 492 nfs_send(struct socket *so, struct mbuf *nam, struct mbuf *top, 493 struct nfsreq *rep) 494 { 495 struct mbuf *sendnam; 496 int error, soflags, flags; 497 498 if (rep) { 499 KASSERT(rw_read_held(&rep->r_nmp->nm_solock)); 500 KASSERT((rep->r_nmp->nm_soflags & PR_CONNREQUIRED) == 0 || 501 rep->r_nmp->nm_sndlwp == curlwp); 502 if (rep->r_flags & R_SOFTTERM) { 503 m_freem(top); 504 return (EINTR); 505 } 506 if ((so = rep->r_nmp->nm_so) == NULL) { 507 rep->r_flags |= R_MUSTRESEND; 508 m_freem(top); 509 return (0); 510 } 511 rep->r_flags &= ~R_MUSTRESEND; 512 soflags = rep->r_nmp->nm_soflags; 513 } else 514 soflags = so->so_proto->pr_flags; 515 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 516 sendnam = NULL; 517 else 518 sendnam = nam; 519 if (so->so_type == SOCK_SEQPACKET) 520 flags = MSG_EOR; 521 else 522 flags = 0; 523 524 error = (*so->so_send)(so, sendnam, NULL, top, NULL, flags, curlwp); 525 if (error) { 526 if (rep) { 527 if (error == ENOBUFS && so->so_type == SOCK_DGRAM) { 528 /* 529 * We're too fast for the network/driver, 530 * and UDP isn't flowcontrolled. 531 * We need to resend. This is not fatal, 532 * just try again. 533 * 534 * Could be smarter here by doing some sort 535 * of a backoff, but this is rare. 536 */ 537 rep->r_flags |= R_MUSTRESEND; 538 } else { 539 if (error != EPIPE) 540 log(LOG_INFO, 541 "nfs send error %d for %s\n", 542 error, 543 rep->r_nmp->nm_mountp-> 544 mnt_stat.f_mntfromname); 545 /* 546 * Deal with errors for the client side. 547 */ 548 if (rep->r_flags & R_SOFTTERM) 549 error = EINTR; 550 else 551 rep->r_flags |= R_MUSTRESEND; 552 } 553 } else { 554 /* 555 * See above. This error can happen under normal 556 * circumstances and the log is too noisy. 557 * The error will still show up in nfsstat. 558 */ 559 if (error != ENOBUFS || so->so_type != SOCK_DGRAM) 560 log(LOG_INFO, "nfsd send error %d\n", error); 561 } 562 563 /* 564 * Handle any recoverable (soft) socket errors here. (? ? ?) 565 */ 566 if (error != EINTR && error != ERESTART && 567 error != EWOULDBLOCK && error != EPIPE) 568 error = 0; 569 } 570 return (error); 571 } 572 573 #ifdef NFS 574 /* 575 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 576 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 577 * Mark and consolidate the data into a new mbuf list. 578 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 579 * small mbufs. 580 * For SOCK_STREAM we must be very careful to read an entire record once 581 * we have read any of it, even if the system call has been interrupted. 582 */ 583 static int 584 nfs_receive(struct nfsreq *rep, struct mbuf **aname, struct mbuf **mp, 585 struct lwp *l) 586 { 587 struct socket *so; 588 struct uio auio; 589 struct iovec aio; 590 struct mbuf *m; 591 struct mbuf *control; 592 u_int32_t len; 593 struct mbuf **getnam; 594 int error, sotype, rcvflg; 595 struct nfsmount *nmp = rep->r_nmp; 596 597 KASSERT(rw_read_held(&nmp->nm_solock)); 598 KASSERT(nmp->nm_rcvlwp == curlwp); 599 600 /* 601 * Set up arguments for soreceive() 602 */ 603 *mp = NULL; 604 *aname = NULL; 605 sotype = nmp->nm_sotype; 606 607 /* 608 * For reliable protocols, lock against other senders/receivers 609 * in case a reconnect is necessary. 610 * For SOCK_STREAM, first get the Record Mark to find out how much 611 * more there is to get. 612 * We must lock the socket against other receivers 613 * until we have an entire rpc request/reply. 614 */ 615 if (sotype != SOCK_DGRAM) { 616 error = nfs_sndlock(nmp, rep); 617 if (error) 618 return (error); 619 tryagain: 620 /* 621 * Check for fatal errors and resending request. 622 */ 623 /* 624 * Ugh: If a reconnect attempt just happened, nm_so 625 * would have changed. NULL indicates a failed 626 * attempt that has essentially shut down this 627 * mount point. 628 */ 629 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 630 nfs_sndunlock(nmp); 631 return (EINTR); 632 } 633 so = nmp->nm_so; 634 if (!so) { 635 reconnect: 636 error = nfs_reconnect(rep); 637 if (error) { 638 nfs_sndunlock(nmp); 639 return error; 640 } 641 goto tryagain; 642 } 643 while (rep->r_flags & R_MUSTRESEND) { 644 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 645 nfsstats.rpcretries++; 646 rep->r_rtt = 0; 647 rep->r_flags &= ~R_TIMING; 648 mutex_enter(&nfs_reqq_lock); 649 if ((rep->r_rflags & RR_SENT) == 0) { 650 rep->r_rflags |= RR_SENT; 651 nmp->nm_sent += NFS_CWNDSCALE; 652 } 653 mutex_exit(&nfs_reqq_lock); 654 error = nfs_send(so, nmp->nm_nam, m, rep); 655 if ((rep->r_flags & R_MUSTRESEND) != 0) { 656 mutex_enter(&nfs_reqq_lock); 657 if ((rep->r_rflags & RR_SENT) != 0) { 658 KASSERT(nmp->nm_sent >= NFS_CWNDSCALE); 659 rep->r_rflags &= ~RR_SENT; 660 nmp->nm_sent -= NFS_CWNDSCALE; 661 } 662 mutex_exit(&nfs_reqq_lock); 663 } 664 if (error) { 665 if (error == EINTR || error == ERESTART) { 666 nfs_sndunlock(nmp); 667 return error; 668 } 669 goto reconnect; 670 } 671 } 672 nfs_sndunlock(nmp); 673 if (sotype == SOCK_STREAM) { 674 aio.iov_base = (void *) &len; 675 aio.iov_len = sizeof(u_int32_t); 676 auio.uio_iov = &aio; 677 auio.uio_iovcnt = 1; 678 auio.uio_rw = UIO_READ; 679 auio.uio_offset = 0; 680 auio.uio_resid = sizeof(u_int32_t); 681 UIO_SETUP_SYSSPACE(&auio); 682 do { 683 rcvflg = MSG_WAITALL; 684 error = (*so->so_receive)(so, NULL, &auio, 685 NULL, NULL, &rcvflg); 686 if (error == EWOULDBLOCK && rep) { 687 if (rep->r_flags & R_SOFTTERM) 688 return (EINTR); 689 /* 690 * if it seems that the server died after it 691 * received our request, set EPIPE so that 692 * we'll reconnect and retransmit requests. 693 */ 694 if (rep->r_rexmit >= nmp->nm_retry) { 695 nfsstats.rpctimeouts++; 696 error = EPIPE; 697 } 698 } 699 } while (error == EWOULDBLOCK); 700 if (!error && auio.uio_resid > 0) { 701 /* 702 * Don't log a 0 byte receive; it means 703 * that the socket has been closed, and 704 * can happen during normal operation 705 * (forcible unmount or Solaris server). 706 */ 707 if (auio.uio_resid != sizeof (u_int32_t)) 708 log(LOG_INFO, 709 "short receive (%lu/%lu) from nfs server %s\n", 710 (u_long)sizeof(u_int32_t) - auio.uio_resid, 711 (u_long)sizeof(u_int32_t), 712 nmp->nm_mountp->mnt_stat.f_mntfromname); 713 error = EPIPE; 714 } 715 if (error) 716 goto errout; 717 len = ntohl(len) & ~0x80000000; 718 /* 719 * This is SERIOUS! We are out of sync with the sender 720 * and forcing a disconnect/reconnect is all I can do. 721 */ 722 if (len > NFS_MAXPACKET) { 723 log(LOG_ERR, "%s (%d) from nfs server %s\n", 724 "impossible packet length", 725 len, nmp->nm_mountp->mnt_stat.f_mntfromname); 726 error = EFBIG; 727 goto errout; 728 } 729 auio.uio_resid = len; 730 do { 731 rcvflg = MSG_WAITALL; 732 error = (*so->so_receive)(so, NULL, &auio, mp, 733 NULL, &rcvflg); 734 } while (error == EWOULDBLOCK || error == EINTR || 735 error == ERESTART); 736 if (!error && auio.uio_resid > 0) { 737 if (len != auio.uio_resid) 738 log(LOG_INFO, 739 "short receive (%lu/%d) from nfs server %s\n", 740 (u_long)len - auio.uio_resid, len, 741 nmp->nm_mountp->mnt_stat.f_mntfromname); 742 error = EPIPE; 743 } 744 } else { 745 /* SEQPACKET */ 746 /* 747 * NB: Since uio_resid is big, MSG_WAITALL is ignored 748 * and soreceive() will return when it has either a 749 * control msg or a data msg. 750 * We have no use for control msg., but must grab them 751 * and then throw them away so we know what is going 752 * on. 753 */ 754 auio.uio_resid = len = 100000000; /* Anything Big */ 755 /* not need to setup uio_vmspace */ 756 do { 757 rcvflg = 0; 758 error = (*so->so_receive)(so, NULL, &auio, mp, 759 &control, &rcvflg); 760 if (control) 761 m_freem(control); 762 if (error == EWOULDBLOCK && rep) { 763 if (rep->r_flags & R_SOFTTERM) 764 return (EINTR); 765 } 766 } while (error == EWOULDBLOCK || 767 (!error && *mp == NULL && control)); 768 if ((rcvflg & MSG_EOR) == 0) 769 printf("Egad!!\n"); 770 if (!error && *mp == NULL) 771 error = EPIPE; 772 len -= auio.uio_resid; 773 } 774 errout: 775 if (error && error != EINTR && error != ERESTART) { 776 m_freem(*mp); 777 *mp = NULL; 778 if (error != EPIPE) 779 log(LOG_INFO, 780 "receive error %d from nfs server %s\n", 781 error, 782 nmp->nm_mountp->mnt_stat.f_mntfromname); 783 error = nfs_sndlock(nmp, rep); 784 if (error == 0) { 785 goto reconnect; 786 } 787 nfs_sndunlock(nmp); 788 } 789 } else { 790 if ((so = nmp->nm_so) == NULL) 791 return (EACCES); 792 if (so->so_state & SS_ISCONNECTED) 793 getnam = NULL; 794 else 795 getnam = aname; 796 auio.uio_resid = len = 1000000; 797 /* not need to setup uio_vmspace */ 798 do { 799 rcvflg = 0; 800 error = (*so->so_receive)(so, getnam, &auio, mp, NULL, 801 &rcvflg); 802 if (error == EWOULDBLOCK && (rep->r_flags & R_SOFTTERM)) 803 return (EINTR); 804 } while (error == EWOULDBLOCK); 805 len -= auio.uio_resid; 806 if (!error && *mp == NULL) 807 error = EPIPE; 808 } 809 if (error) { 810 m_freem(*mp); 811 *mp = NULL; 812 } 813 return (error); 814 } 815 816 /* 817 * Implement receipt of reply on a socket. 818 * We must search through the list of received datagrams matching them 819 * with outstanding requests using the xid, until ours is found. 820 */ 821 /* ARGSUSED */ 822 static int 823 nfs_reply(struct nfsreq *myrep, struct lwp *lwp) 824 { 825 struct nfsreq *rep; 826 struct nfsmount *nmp = myrep->r_nmp; 827 int32_t t1; 828 struct mbuf *mrep, *nam, *md; 829 u_int32_t rxid, *tl; 830 char *dpos, *cp2; 831 int error; 832 833 /* 834 * Loop around until we get our own reply 835 */ 836 for (;;) { 837 KASSERT(rw_read_held(&nmp->nm_solock)); 838 839 /* 840 * Lock against other receivers so that I don't get stuck in 841 * sbwait() after someone else has received my reply for me. 842 * Also necessary for connection based protocols to avoid 843 * race conditions during a reconnect. 844 */ 845 error = nfs_rcvlock(nmp, myrep); 846 if (error == EALREADY) 847 return (0); 848 if (error) 849 return (error); 850 /* 851 * Get the next Rpc reply off the socket 852 */ 853 854 mutex_enter(&nmp->nm_lock); 855 nmp->nm_waiters++; 856 mutex_exit(&nmp->nm_lock); 857 858 error = nfs_receive(myrep, &nam, &mrep, lwp); 859 860 mutex_enter(&nmp->nm_lock); 861 nmp->nm_waiters--; 862 cv_signal(&nmp->nm_disconcv); 863 mutex_exit(&nmp->nm_lock); 864 865 if (error) { 866 nfs_rcvunlock(nmp); 867 if (error == EAGAIN) { /* from nfs_reconnect */ 868 printf("%s: nmp=%p: draining\n", __func__, nmp); 869 rw_exit(&nmp->nm_solock); 870 rw_enter(&nmp->nm_solock, RW_WRITER); 871 rw_exit(&nmp->nm_solock); 872 rw_enter(&nmp->nm_solock, RW_READER); 873 continue; 874 } 875 876 if (nmp->nm_iflag & NFSMNT_DISMNT) { 877 /* 878 * Oops, we're going away now.. 879 */ 880 return error; 881 } 882 /* 883 * Ignore routing errors on connectionless protocols? ? 884 */ 885 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 886 nmp->nm_so->so_error = 0; 887 #ifdef DEBUG 888 if (ratecheck(&nfs_reply_last_err_time, 889 &nfs_err_interval)) 890 printf("%s: ignoring error %d\n", 891 __func__, error); 892 #endif 893 continue; 894 } 895 return (error); 896 } 897 if (nam) 898 m_freem(nam); 899 900 /* 901 * Get the xid and check that it is an rpc reply 902 */ 903 md = mrep; 904 dpos = mtod(md, void *); 905 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 906 rxid = *tl++; 907 if (*tl != rpc_reply) { 908 nfsstats.rpcinvalid++; 909 m_freem(mrep); 910 nfsmout: 911 nfs_rcvunlock(nmp); 912 continue; 913 } 914 915 /* 916 * Loop through the request list to match up the reply 917 * Iff no match, just drop the datagram 918 */ 919 mutex_enter(&nfs_reqq_lock); 920 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 921 if (rep->r_nmp != nmp) { 922 continue; 923 } 924 if (rxid != rep->r_xid) { 925 continue; 926 } 927 if (rep->r_mrep != NULL) { 928 continue; 929 } 930 if (/* nmp->nm_sotype == SOCK_DGRAM && */ 931 (rep->r_rflags & RR_SENT) == 0) { 932 printf("%s: got a reply for unsent req:" 933 " xid=0x%x\n", 934 __func__, rep->r_xid); 935 continue; 936 } 937 break; 938 } 939 if (rep != NULL) { 940 /* Found it.. */ 941 rep->r_mrep = mrep; 942 rep->r_md = md; 943 rep->r_dpos = dpos; 944 if (nfsrtton) { 945 struct rttl *rt; 946 947 rt = &nfsrtt.rttl[nfsrtt.pos]; 948 rt->proc = rep->r_procnum; 949 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 950 rt->sent = nmp->nm_sent; 951 rt->cwnd = nmp->nm_cwnd; 952 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 953 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 954 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx; 955 getmicrotime(&rt->tstamp); 956 if (rep->r_flags & R_TIMING) 957 rt->rtt = rep->r_rtt; 958 else 959 rt->rtt = 1000000; 960 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 961 } 962 /* 963 * Update congestion window. 964 * Do the additive increase of 965 * one rpc/rtt. 966 */ 967 if (nmp->nm_cwnd <= nmp->nm_sent) { 968 nmp->nm_cwnd += 969 (NFS_CWNDSCALE * NFS_CWNDSCALE + 970 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 971 if (nmp->nm_cwnd > NFS_MAXCWND) 972 nmp->nm_cwnd = NFS_MAXCWND; 973 } 974 if ((rep->r_rflags & RR_SENT) != 0) { 975 KASSERT(nmp->nm_sent >= NFS_CWNDSCALE); 976 rep->r_rflags &= ~RR_SENT; 977 nmp->nm_sent -= NFS_CWNDSCALE; 978 } 979 /* 980 * Update rtt using a gain of 0.125 on the mean 981 * and a gain of 0.25 on the deviation. 982 */ 983 if (rep->r_flags & R_TIMING) { 984 /* 985 * Since the timer resolution of 986 * NFS_HZ is so course, it can often 987 * result in r_rtt == 0. Since 988 * r_rtt == N means that the actual 989 * rtt is between N+dt and N+2-dt ticks, 990 * add 1. 991 */ 992 t1 = rep->r_rtt + 1; 993 t1 -= (NFS_SRTT(rep) >> 3); 994 NFS_SRTT(rep) += t1; 995 if (t1 < 0) 996 t1 = -t1; 997 t1 -= (NFS_SDRTT(rep) >> 2); 998 NFS_SDRTT(rep) += t1; 999 } 1000 nmp->nm_timeouts = 0; 1001 } 1002 mutex_exit(&nfs_reqq_lock); 1003 nfs_rcvunlock(nmp); 1004 /* 1005 * If not matched to a request, drop it. 1006 * If it's mine, get out. 1007 */ 1008 if (rep == NULL) { 1009 printf("%s: unexpected reply: xid=0x%" PRIx32 1010 ", sotype=%d\n", 1011 __func__, rxid, nmp->nm_sotype); 1012 nfsstats.rpcunexpected++; 1013 m_freem(mrep); 1014 } else if (rep == myrep) { 1015 if (rep->r_mrep == NULL) 1016 panic("nfsreply nil"); 1017 return (0); 1018 } 1019 } 1020 } 1021 1022 /* 1023 * nfs_request - goes something like this 1024 * - fill in request struct 1025 * - links it into list 1026 * - calls nfs_send() for first transmit 1027 * - calls nfs_receive() to get reply 1028 * - break down rpc header and return with nfs reply pointed to 1029 * by mrep or error 1030 * nb: always frees up mreq mbuf list 1031 */ 1032 int 1033 nfs_request(struct nfsnode *np, struct mbuf *mrest, int procnum, struct lwp *lwp, kauth_cred_t cred, struct mbuf **mrp, struct mbuf **mdp, char **dposp, int *rexmitp) 1034 { 1035 struct mbuf *m, *mrep; 1036 struct nfsreq *rep; 1037 u_int32_t *tl; 1038 int i; 1039 struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount); 1040 struct mbuf *md, *mheadend; 1041 char nickv[RPCX_NICKVERF]; 1042 time_t waituntil; 1043 char *dpos, *cp2; 1044 int t1, error = 0, mrest_len, auth_len, auth_type; 1045 int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0; 1046 int verf_len, verf_type; 1047 u_int32_t xid; 1048 char *auth_str, *verf_str; 1049 NFSKERBKEY_T key; /* save session key */ 1050 kauth_cred_t acred; 1051 struct mbuf *mrest_backup = NULL; 1052 kauth_cred_t origcred = NULL; /* XXX: gcc */ 1053 bool retry_cred = true; 1054 bool use_opencred = (np->n_flag & NUSEOPENCRED) != 0; 1055 1056 if (rexmitp != NULL) 1057 *rexmitp = 0; 1058 1059 acred = kauth_cred_alloc(); 1060 1061 tryagain_cred: 1062 KASSERT(cred != NULL); 1063 rep = kmem_alloc(sizeof(*rep), KM_SLEEP); 1064 rep->r_nmp = nmp; 1065 KASSERT(lwp == NULL || lwp == curlwp); 1066 rep->r_lwp = lwp; 1067 rep->r_procnum = procnum; 1068 i = 0; 1069 m = mrest; 1070 while (m) { 1071 i += m->m_len; 1072 m = m->m_next; 1073 } 1074 mrest_len = i; 1075 1076 /* 1077 * Get the RPC header with authorization. 1078 */ 1079 kerbauth: 1080 verf_str = auth_str = (char *)0; 1081 if (nmp->nm_flag & NFSMNT_KERB) { 1082 verf_str = nickv; 1083 verf_len = sizeof (nickv); 1084 auth_type = RPCAUTH_KERB4; 1085 memset((void *)key, 0, sizeof (key)); 1086 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 1087 &auth_len, verf_str, verf_len)) { 1088 error = nfs_getauth(nmp, rep, cred, &auth_str, 1089 &auth_len, verf_str, &verf_len, key); 1090 if (error) { 1091 kmem_free(rep, sizeof(*rep)); 1092 m_freem(mrest); 1093 KASSERT(kauth_cred_getrefcnt(acred) == 1); 1094 kauth_cred_free(acred); 1095 return (error); 1096 } 1097 } 1098 retry_cred = false; 1099 } else { 1100 /* AUTH_UNIX */ 1101 uid_t uid; 1102 gid_t gid; 1103 1104 /* 1105 * on the most unix filesystems, permission checks are 1106 * done when the file is open(2)'ed. 1107 * ie. once a file is successfully open'ed, 1108 * following i/o operations never fail with EACCES. 1109 * we try to follow the semantics as far as possible. 1110 * 1111 * note that we expect that the nfs server always grant 1112 * accesses by the file's owner. 1113 */ 1114 origcred = cred; 1115 switch (procnum) { 1116 case NFSPROC_READ: 1117 case NFSPROC_WRITE: 1118 case NFSPROC_COMMIT: 1119 uid = np->n_vattr->va_uid; 1120 gid = np->n_vattr->va_gid; 1121 if (kauth_cred_geteuid(cred) == uid && 1122 kauth_cred_getegid(cred) == gid) { 1123 retry_cred = false; 1124 break; 1125 } 1126 if (use_opencred) 1127 break; 1128 kauth_cred_setuid(acred, uid); 1129 kauth_cred_seteuid(acred, uid); 1130 kauth_cred_setsvuid(acred, uid); 1131 kauth_cred_setgid(acred, gid); 1132 kauth_cred_setegid(acred, gid); 1133 kauth_cred_setsvgid(acred, gid); 1134 cred = acred; 1135 break; 1136 default: 1137 retry_cred = false; 1138 break; 1139 } 1140 /* 1141 * backup mbuf chain if we can need it later to retry. 1142 * 1143 * XXX maybe we can keep a direct reference to 1144 * mrest without doing m_copym, but it's ...ugly. 1145 */ 1146 if (retry_cred) 1147 mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT); 1148 auth_type = RPCAUTH_UNIX; 1149 /* XXX elad - ngroups */ 1150 auth_len = (((kauth_cred_ngroups(cred) > nmp->nm_numgrps) ? 1151 nmp->nm_numgrps : kauth_cred_ngroups(cred)) << 2) + 1152 5 * NFSX_UNSIGNED; 1153 } 1154 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 1155 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 1156 if (auth_str) 1157 free(auth_str, M_TEMP); 1158 1159 /* 1160 * For stream protocols, insert a Sun RPC Record Mark. 1161 */ 1162 if (nmp->nm_sotype == SOCK_STREAM) { 1163 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 1164 *mtod(m, u_int32_t *) = htonl(0x80000000 | 1165 (m->m_pkthdr.len - NFSX_UNSIGNED)); 1166 } 1167 rep->r_mreq = m; 1168 rep->r_xid = xid; 1169 tryagain: 1170 if (nmp->nm_flag & NFSMNT_SOFT) 1171 rep->r_retry = nmp->nm_retry; 1172 else 1173 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 1174 rep->r_rtt = rep->r_rexmit = 0; 1175 if (proct[procnum] > 0) 1176 rep->r_flags = R_TIMING; 1177 else 1178 rep->r_flags = 0; 1179 rep->r_rflags = 0; 1180 rep->r_mrep = NULL; 1181 1182 /* 1183 * Do the client side RPC. 1184 */ 1185 nfsstats.rpcrequests++; 1186 /* 1187 * Chain request into list of outstanding requests. Be sure 1188 * to put it LAST so timer finds oldest requests first. 1189 */ 1190 1191 rw_enter(&nmp->nm_solock, RW_READER); 1192 mutex_enter(&nfs_reqq_lock); 1193 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 1194 1195 /* 1196 * If backing off another request or avoiding congestion, don't 1197 * send this one now but let timer do it. If not timing a request, 1198 * do it now. 1199 */ 1200 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 1201 (nmp->nm_flag & NFSMNT_DUMBTIMR) || nmp->nm_sent < nmp->nm_cwnd)) { 1202 nmp->nm_sent += NFS_CWNDSCALE; 1203 rep->r_rflags |= RR_SENT; 1204 mutex_exit(&nfs_reqq_lock); 1205 if (nmp->nm_soflags & PR_CONNREQUIRED) 1206 error = nfs_sndlock(nmp, rep); 1207 if (!error) { 1208 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 1209 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 1210 if (nmp->nm_soflags & PR_CONNREQUIRED) 1211 nfs_sndunlock(nmp); 1212 } 1213 mutex_enter(&nfs_reqq_lock); 1214 /* 1215 * note that we might have gotten a reply already. 1216 */ 1217 if (error != 0 || (rep->r_flags & R_MUSTRESEND) != 0) { 1218 if ((rep->r_rflags & RR_SENT) != 0) { 1219 KASSERT(nmp->nm_sent >= NFS_CWNDSCALE); 1220 rep->r_rflags &= ~RR_SENT; 1221 nmp->nm_sent -= NFS_CWNDSCALE; 1222 } 1223 } 1224 } else { 1225 rep->r_rtt = -1; 1226 } 1227 mutex_exit(&nfs_reqq_lock); 1228 1229 nfs_timer_start(); 1230 1231 /* 1232 * Wait for the reply from our send or the timer's. 1233 */ 1234 if (!error || error == EPIPE) { 1235 error = nfs_reply(rep, lwp); 1236 } 1237 1238 /* 1239 * RPC done, unlink the request. 1240 */ 1241 mutex_enter(&nfs_reqq_lock); 1242 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 1243 1244 /* 1245 * Decrement the outstanding request count. 1246 */ 1247 if (rep->r_rflags & RR_SENT) { 1248 KASSERT(nmp->nm_sent >= NFS_CWNDSCALE); 1249 rep->r_rflags &= ~RR_SENT; /* paranoia */ 1250 nmp->nm_sent -= NFS_CWNDSCALE; 1251 } 1252 mutex_exit(&nfs_reqq_lock); 1253 1254 if (rexmitp != NULL) { 1255 int rexmit; 1256 1257 if (nmp->nm_sotype != SOCK_DGRAM) 1258 rexmit = (rep->r_flags & R_REXMITTED) != 0; 1259 else 1260 rexmit = rep->r_rexmit; 1261 *rexmitp = rexmit; 1262 } 1263 1264 rw_exit(&nmp->nm_solock); 1265 1266 /* 1267 * If there was a successful reply and a tprintf msg. 1268 * tprintf a response. 1269 */ 1270 if (!error && (rep->r_flags & R_TPRINTFMSG)) 1271 nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname, 1272 "is alive again"); 1273 mrep = rep->r_mrep; 1274 md = rep->r_md; 1275 dpos = rep->r_dpos; 1276 if (error) 1277 goto nfsmout; 1278 1279 /* 1280 * break down the rpc header and check if ok 1281 */ 1282 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1283 if (*tl++ == rpc_msgdenied) { 1284 if (*tl == rpc_mismatch) 1285 error = EOPNOTSUPP; 1286 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1287 if (!failed_auth) { 1288 failed_auth++; 1289 mheadend->m_next = (struct mbuf *)0; 1290 m_freem(mrep); 1291 m_freem(rep->r_mreq); 1292 goto kerbauth; 1293 } else 1294 error = EAUTH; 1295 } else 1296 error = EACCES; 1297 m_freem(mrep); 1298 goto nfsmout; 1299 } 1300 1301 /* 1302 * Grab any Kerberos verifier, otherwise just throw it away. 1303 */ 1304 verf_type = fxdr_unsigned(int, *tl++); 1305 i = fxdr_unsigned(int32_t, *tl); 1306 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 1307 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 1308 if (error) 1309 goto nfsmout; 1310 } else if (i > 0) 1311 nfsm_adv(nfsm_rndup(i)); 1312 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1313 /* 0 == ok */ 1314 if (*tl == 0) { 1315 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1316 if (*tl != 0) { 1317 error = fxdr_unsigned(int, *tl); 1318 switch (error) { 1319 case NFSERR_PERM: 1320 error = EPERM; 1321 break; 1322 1323 case NFSERR_NOENT: 1324 error = ENOENT; 1325 break; 1326 1327 case NFSERR_IO: 1328 error = EIO; 1329 break; 1330 1331 case NFSERR_NXIO: 1332 error = ENXIO; 1333 break; 1334 1335 case NFSERR_ACCES: 1336 error = EACCES; 1337 if (!retry_cred) 1338 break; 1339 m_freem(mrep); 1340 m_freem(rep->r_mreq); 1341 kmem_free(rep, sizeof(*rep)); 1342 use_opencred = !use_opencred; 1343 if (mrest_backup == NULL) { 1344 /* m_copym failure */ 1345 KASSERT( 1346 kauth_cred_getrefcnt(acred) == 1); 1347 kauth_cred_free(acred); 1348 return ENOMEM; 1349 } 1350 mrest = mrest_backup; 1351 mrest_backup = NULL; 1352 cred = origcred; 1353 error = 0; 1354 retry_cred = false; 1355 goto tryagain_cred; 1356 1357 case NFSERR_EXIST: 1358 error = EEXIST; 1359 break; 1360 1361 case NFSERR_XDEV: 1362 error = EXDEV; 1363 break; 1364 1365 case NFSERR_NODEV: 1366 error = ENODEV; 1367 break; 1368 1369 case NFSERR_NOTDIR: 1370 error = ENOTDIR; 1371 break; 1372 1373 case NFSERR_ISDIR: 1374 error = EISDIR; 1375 break; 1376 1377 case NFSERR_INVAL: 1378 error = EINVAL; 1379 break; 1380 1381 case NFSERR_FBIG: 1382 error = EFBIG; 1383 break; 1384 1385 case NFSERR_NOSPC: 1386 error = ENOSPC; 1387 break; 1388 1389 case NFSERR_ROFS: 1390 error = EROFS; 1391 break; 1392 1393 case NFSERR_MLINK: 1394 error = EMLINK; 1395 break; 1396 1397 case NFSERR_TIMEDOUT: 1398 error = ETIMEDOUT; 1399 break; 1400 1401 case NFSERR_NAMETOL: 1402 error = ENAMETOOLONG; 1403 break; 1404 1405 case NFSERR_NOTEMPTY: 1406 error = ENOTEMPTY; 1407 break; 1408 1409 case NFSERR_DQUOT: 1410 error = EDQUOT; 1411 break; 1412 1413 case NFSERR_STALE: 1414 /* 1415 * If the File Handle was stale, invalidate the 1416 * lookup cache, just in case. 1417 */ 1418 error = ESTALE; 1419 cache_purge(NFSTOV(np)); 1420 break; 1421 1422 case NFSERR_REMOTE: 1423 error = EREMOTE; 1424 break; 1425 1426 case NFSERR_WFLUSH: 1427 case NFSERR_BADHANDLE: 1428 case NFSERR_NOT_SYNC: 1429 case NFSERR_BAD_COOKIE: 1430 error = EINVAL; 1431 break; 1432 1433 case NFSERR_NOTSUPP: 1434 error = ENOTSUP; 1435 break; 1436 1437 case NFSERR_TOOSMALL: 1438 case NFSERR_SERVERFAULT: 1439 case NFSERR_BADTYPE: 1440 error = EINVAL; 1441 break; 1442 1443 case NFSERR_TRYLATER: 1444 if ((nmp->nm_flag & NFSMNT_NFSV3) == 0) 1445 break; 1446 m_freem(mrep); 1447 error = 0; 1448 waituntil = time_second + trylater_delay; 1449 while (time_second < waituntil) { 1450 kpause("nfstrylater", false, hz, NULL); 1451 } 1452 trylater_delay *= NFS_TRYLATERDELMUL; 1453 if (trylater_delay > NFS_TRYLATERDELMAX) 1454 trylater_delay = NFS_TRYLATERDELMAX; 1455 /* 1456 * RFC1813: 1457 * The client should wait and then try 1458 * the request with a new RPC transaction ID. 1459 */ 1460 nfs_renewxid(rep); 1461 goto tryagain; 1462 1463 default: 1464 #ifdef DIAGNOSTIC 1465 printf("Invalid rpc error code %d\n", error); 1466 #endif 1467 error = EINVAL; 1468 break; 1469 } 1470 1471 if (nmp->nm_flag & NFSMNT_NFSV3) { 1472 *mrp = mrep; 1473 *mdp = md; 1474 *dposp = dpos; 1475 error |= NFSERR_RETERR; 1476 } else 1477 m_freem(mrep); 1478 goto nfsmout; 1479 } 1480 1481 /* 1482 * note which credential worked to minimize number of retries. 1483 */ 1484 if (use_opencred) 1485 np->n_flag |= NUSEOPENCRED; 1486 else 1487 np->n_flag &= ~NUSEOPENCRED; 1488 1489 *mrp = mrep; 1490 *mdp = md; 1491 *dposp = dpos; 1492 1493 KASSERT(error == 0); 1494 goto nfsmout; 1495 } 1496 m_freem(mrep); 1497 error = EPROTONOSUPPORT; 1498 nfsmout: 1499 KASSERT(kauth_cred_getrefcnt(acred) == 1); 1500 kauth_cred_free(acred); 1501 m_freem(rep->r_mreq); 1502 kmem_free(rep, sizeof(*rep)); 1503 m_freem(mrest_backup); 1504 return (error); 1505 } 1506 #endif /* NFS */ 1507 1508 /* 1509 * Generate the rpc reply header 1510 * siz arg. is used to decide if adding a cluster is worthwhile 1511 */ 1512 int 1513 nfs_rephead(int siz, struct nfsrv_descript *nd, struct nfssvc_sock *slp, int err, int cache, u_quad_t *frev, struct mbuf **mrq, struct mbuf **mbp, char **bposp) 1514 { 1515 u_int32_t *tl; 1516 struct mbuf *mreq; 1517 char *bpos; 1518 struct mbuf *mb; 1519 1520 mreq = m_gethdr(M_WAIT, MT_DATA); 1521 MCLAIM(mreq, &nfs_mowner); 1522 mb = mreq; 1523 /* 1524 * If this is a big reply, use a cluster else 1525 * try and leave leading space for the lower level headers. 1526 */ 1527 siz += RPC_REPLYSIZ; 1528 if (siz >= max_datalen) { 1529 m_clget(mreq, M_WAIT); 1530 } else 1531 mreq->m_data += max_hdr; 1532 tl = mtod(mreq, u_int32_t *); 1533 mreq->m_len = 6 * NFSX_UNSIGNED; 1534 bpos = ((char *)tl) + mreq->m_len; 1535 *tl++ = txdr_unsigned(nd->nd_retxid); 1536 *tl++ = rpc_reply; 1537 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { 1538 *tl++ = rpc_msgdenied; 1539 if (err & NFSERR_AUTHERR) { 1540 *tl++ = rpc_autherr; 1541 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); 1542 mreq->m_len -= NFSX_UNSIGNED; 1543 bpos -= NFSX_UNSIGNED; 1544 } else { 1545 *tl++ = rpc_mismatch; 1546 *tl++ = txdr_unsigned(RPC_VER2); 1547 *tl = txdr_unsigned(RPC_VER2); 1548 } 1549 } else { 1550 *tl++ = rpc_msgaccepted; 1551 1552 /* 1553 * For Kerberos authentication, we must send the nickname 1554 * verifier back, otherwise just RPCAUTH_NULL. 1555 */ 1556 if (nd->nd_flag & ND_KERBFULL) { 1557 struct nfsuid *nuidp; 1558 struct timeval ktvin, ktvout; 1559 1560 memset(&ktvout, 0, sizeof ktvout); /* XXX gcc */ 1561 1562 LIST_FOREACH(nuidp, 1563 NUIDHASH(slp, kauth_cred_geteuid(nd->nd_cr)), 1564 nu_hash) { 1565 if (kauth_cred_geteuid(nuidp->nu_cr) == 1566 kauth_cred_geteuid(nd->nd_cr) && 1567 (!nd->nd_nam2 || netaddr_match( 1568 NU_NETFAM(nuidp), &nuidp->nu_haddr, 1569 nd->nd_nam2))) 1570 break; 1571 } 1572 if (nuidp) { 1573 ktvin.tv_sec = 1574 txdr_unsigned(nuidp->nu_timestamp.tv_sec 1575 - 1); 1576 ktvin.tv_usec = 1577 txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1578 1579 /* 1580 * Encrypt the timestamp in ecb mode using the 1581 * session key. 1582 */ 1583 #ifdef NFSKERB 1584 XXX 1585 #endif 1586 1587 *tl++ = rpc_auth_kerb; 1588 *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); 1589 *tl = ktvout.tv_sec; 1590 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1591 *tl++ = ktvout.tv_usec; 1592 *tl++ = txdr_unsigned( 1593 kauth_cred_geteuid(nuidp->nu_cr)); 1594 } else { 1595 *tl++ = 0; 1596 *tl++ = 0; 1597 } 1598 } else { 1599 *tl++ = 0; 1600 *tl++ = 0; 1601 } 1602 switch (err) { 1603 case EPROGUNAVAIL: 1604 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1605 break; 1606 case EPROGMISMATCH: 1607 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1608 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1609 *tl++ = txdr_unsigned(2); 1610 *tl = txdr_unsigned(3); 1611 break; 1612 case EPROCUNAVAIL: 1613 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1614 break; 1615 case EBADRPC: 1616 *tl = txdr_unsigned(RPC_GARBAGE); 1617 break; 1618 default: 1619 *tl = 0; 1620 if (err != NFSERR_RETVOID) { 1621 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1622 if (err) 1623 *tl = txdr_unsigned(nfsrv_errmap(nd, err)); 1624 else 1625 *tl = 0; 1626 } 1627 break; 1628 }; 1629 } 1630 1631 if (mrq != NULL) 1632 *mrq = mreq; 1633 *mbp = mb; 1634 *bposp = bpos; 1635 if (err != 0 && err != NFSERR_RETVOID) 1636 nfsstats.srvrpc_errs++; 1637 return (0); 1638 } 1639 1640 static void 1641 nfs_timer_schedule(void) 1642 { 1643 1644 callout_schedule(&nfs_timer_ch, nfs_ticks); 1645 } 1646 1647 void 1648 nfs_timer_start(void) 1649 { 1650 1651 if (callout_pending(&nfs_timer_ch)) 1652 return; 1653 1654 nfs_timer_start_ev.ev_count++; 1655 nfs_timer_schedule(); 1656 } 1657 1658 void 1659 nfs_timer_init(void) 1660 { 1661 1662 mutex_init(&nfs_timer_lock, MUTEX_DEFAULT, IPL_NONE); 1663 callout_init(&nfs_timer_ch, CALLOUT_MPSAFE); 1664 callout_setfunc(&nfs_timer_ch, nfs_timer, NULL); 1665 evcnt_attach_dynamic(&nfs_timer_ev, EVCNT_TYPE_MISC, NULL, 1666 "nfs", "timer"); 1667 evcnt_attach_dynamic(&nfs_timer_start_ev, EVCNT_TYPE_MISC, NULL, 1668 "nfs", "timer start"); 1669 evcnt_attach_dynamic(&nfs_timer_stop_ev, EVCNT_TYPE_MISC, NULL, 1670 "nfs", "timer stop"); 1671 } 1672 1673 void 1674 nfs_timer_fini(void) 1675 { 1676 1677 callout_halt(&nfs_timer_ch, NULL); 1678 callout_destroy(&nfs_timer_ch); 1679 mutex_destroy(&nfs_timer_lock); 1680 evcnt_detach(&nfs_timer_ev); 1681 evcnt_detach(&nfs_timer_start_ev); 1682 evcnt_detach(&nfs_timer_stop_ev); 1683 } 1684 1685 void 1686 nfs_timer_srvinit(bool (*func)(void)) 1687 { 1688 1689 nfs_timer_srvvec = func; 1690 } 1691 1692 void 1693 nfs_timer_srvfini(void) 1694 { 1695 1696 mutex_enter(&nfs_timer_lock); 1697 nfs_timer_srvvec = NULL; 1698 mutex_exit(&nfs_timer_lock); 1699 } 1700 1701 1702 /* 1703 * Nfs timer routine 1704 * Scan the nfsreq list and retranmit any requests that have timed out 1705 * To avoid retransmission attempts on STREAM sockets (in the future) make 1706 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1707 */ 1708 void 1709 nfs_timer(void *arg) 1710 { 1711 struct nfsreq *rep; 1712 struct mbuf *m; 1713 struct socket *so; 1714 struct nfsmount *nmp; 1715 int timeo; 1716 int error; 1717 bool more = false; 1718 1719 nfs_timer_ev.ev_count++; 1720 1721 mutex_enter(&nfs_reqq_lock); 1722 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 1723 more = true; 1724 nmp = rep->r_nmp; 1725 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1726 continue; 1727 if (nfs_sigintr(nmp, rep, rep->r_lwp)) { 1728 rep->r_flags |= R_SOFTTERM; 1729 continue; 1730 } 1731 if (rep->r_rtt >= 0) { 1732 rep->r_rtt++; 1733 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1734 timeo = nmp->nm_timeo; 1735 else 1736 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1737 if (nmp->nm_timeouts > 0) 1738 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1739 if (timeo > NFS_MAXTIMEO) 1740 timeo = NFS_MAXTIMEO; 1741 if (rep->r_rtt <= timeo) 1742 continue; 1743 if (nmp->nm_timeouts < 1744 (sizeof(nfs_backoff) / sizeof(nfs_backoff[0]))) 1745 nmp->nm_timeouts++; 1746 } 1747 /* 1748 * Check for server not responding 1749 */ 1750 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1751 rep->r_rexmit > nmp->nm_deadthresh) { 1752 nfs_msg(rep->r_lwp, 1753 nmp->nm_mountp->mnt_stat.f_mntfromname, 1754 "not responding"); 1755 rep->r_flags |= R_TPRINTFMSG; 1756 } 1757 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1758 nfsstats.rpctimeouts++; 1759 rep->r_flags |= R_SOFTTERM; 1760 continue; 1761 } 1762 if (nmp->nm_sotype != SOCK_DGRAM) { 1763 if (++rep->r_rexmit > NFS_MAXREXMIT) 1764 rep->r_rexmit = NFS_MAXREXMIT; 1765 continue; 1766 } 1767 if (!rw_tryenter(&nmp->nm_solock, RW_READER)) { 1768 printf("%s: rw_trylock failed\n", __func__); 1769 continue; 1770 } 1771 if ((so = nmp->nm_so) == NULL) { 1772 rw_exit(&nmp->nm_solock); 1773 continue; 1774 } 1775 1776 /* 1777 * If there is enough space and the window allows.. 1778 * Resend it 1779 * Set r_rtt to -1 in case we fail to send it now. 1780 */ 1781 /* solock(so); XXX PR 40491 */ 1782 rep->r_rtt = -1; 1783 /* XXX kernel_lock for sbspace? */ 1784 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1785 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1786 (rep->r_rflags & RR_SENT) || 1787 nmp->nm_sent < nmp->nm_cwnd) && 1788 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))) { 1789 if (so->so_state & SS_ISCONNECTED) 1790 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, 1791 m, NULL, NULL, NULL); 1792 else 1793 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, 1794 m, nmp->nm_nam, NULL, NULL); 1795 if (error) { 1796 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 1797 #ifdef DEBUG 1798 if (ratecheck(&nfs_timer_last_err_time, 1799 &nfs_err_interval)) 1800 printf("%s: ignoring error " 1801 "%d\n", __func__, error); 1802 #endif 1803 so->so_error = 0; 1804 } 1805 } else { 1806 /* 1807 * Iff first send, start timing 1808 * else turn timing off, backoff timer 1809 * and divide congestion window by 2. 1810 */ 1811 if (rep->r_rflags & RR_SENT) { 1812 rep->r_flags &= ~R_TIMING; 1813 if (++rep->r_rexmit > NFS_MAXREXMIT) 1814 rep->r_rexmit = NFS_MAXREXMIT; 1815 nmp->nm_cwnd >>= 1; 1816 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1817 nmp->nm_cwnd = NFS_CWNDSCALE; 1818 nfsstats.rpcretries++; 1819 } else { 1820 rep->r_rflags |= RR_SENT; 1821 nmp->nm_sent += NFS_CWNDSCALE; 1822 } 1823 rep->r_rtt = 0; 1824 } 1825 } 1826 sounlock(so); 1827 rw_exit(&nmp->nm_solock); 1828 } 1829 mutex_exit(&nfs_reqq_lock); 1830 1831 mutex_enter(&nfs_timer_lock); 1832 if (nfs_timer_srvvec != NULL) { 1833 more |= (*nfs_timer_srvvec)(); 1834 } 1835 mutex_exit(&nfs_timer_lock); 1836 1837 if (more) { 1838 nfs_timer_schedule(); 1839 } else { 1840 nfs_timer_stop_ev.ev_count++; 1841 } 1842 } 1843 1844 /* 1845 * Test for a termination condition pending on the process. 1846 * This is used for NFSMNT_INT mounts. 1847 */ 1848 int 1849 nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct lwp *l) 1850 { 1851 sigset_t ss; 1852 1853 if (rep && (rep->r_flags & R_SOFTTERM)) 1854 return (EINTR); 1855 if (!(nmp->nm_flag & NFSMNT_INT)) 1856 return (0); 1857 if (l) { 1858 sigpending1(l, &ss); 1859 #if 0 1860 sigminusset(&l->l_proc->p_sigctx.ps_sigignore, &ss); 1861 #endif 1862 if (sigismember(&ss, SIGINT) || sigismember(&ss, SIGTERM) || 1863 sigismember(&ss, SIGKILL) || sigismember(&ss, SIGHUP) || 1864 sigismember(&ss, SIGQUIT)) 1865 return (EINTR); 1866 } 1867 return (0); 1868 } 1869 1870 #ifdef NFS 1871 /* 1872 * Lock a socket against others. 1873 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1874 * and also to avoid race conditions between the processes with nfs requests 1875 * in progress when a reconnect is necessary. 1876 */ 1877 static int 1878 nfs_sndlock(struct nfsmount *nmp, struct nfsreq *rep) 1879 { 1880 struct lwp *l; 1881 int timeo = 0; 1882 bool catch; 1883 int error = 0; 1884 1885 KASSERT(nmp == rep->r_nmp); 1886 1887 l = rep->r_lwp; 1888 catch = (nmp->nm_flag & NFSMNT_INT) != 0; 1889 mutex_enter(&nmp->nm_lock); 1890 while (nmp->nm_sndlwp != NULL) { 1891 KASSERT(nmp->nm_sndlwp != curlwp); 1892 if (rep && nfs_sigintr(rep->r_nmp, rep, l)) { 1893 error = EINTR; 1894 goto quit; 1895 } 1896 if (catch) { 1897 cv_timedwait_sig(&nmp->nm_sndcv, &nmp->nm_lock, timeo); 1898 } else { 1899 cv_timedwait(&nmp->nm_sndcv, &nmp->nm_lock, timeo); 1900 } 1901 if (catch) { 1902 catch = false; 1903 timeo = 2 * hz; 1904 } 1905 } 1906 nmp->nm_sndlwp = curlwp; 1907 quit: 1908 mutex_exit(&nmp->nm_lock); 1909 return error; 1910 } 1911 1912 /* 1913 * Unlock the stream socket for others. 1914 */ 1915 static void 1916 nfs_sndunlock(struct nfsmount *nmp) 1917 { 1918 1919 mutex_enter(&nmp->nm_lock); 1920 if (nmp->nm_sndlwp != curlwp) 1921 panic("nfs sndunlock"); 1922 nmp->nm_sndlwp = NULL; 1923 cv_signal(&nmp->nm_sndcv); 1924 mutex_exit(&nmp->nm_lock); 1925 } 1926 #endif /* NFS */ 1927 1928 static int 1929 nfs_rcvlock(struct nfsmount *nmp, struct nfsreq *rep) 1930 { 1931 int slptimeo = 0; 1932 bool catch; 1933 int error = 0; 1934 1935 KASSERT(rep == NULL || nmp == rep->r_nmp); 1936 1937 catch = (nmp->nm_flag & NFSMNT_INT) != 0; 1938 mutex_enter(&nmp->nm_lock); 1939 while (/* CONSTCOND */ true) { 1940 KASSERT(nmp->nm_rcvlwp != curlwp); 1941 if (nmp->nm_iflag & NFSMNT_DISMNT) { 1942 cv_signal(&nmp->nm_disconcv); 1943 error = EIO; 1944 break; 1945 } 1946 /* If our reply was received while we were sleeping, 1947 * then just return without taking the lock to avoid a 1948 * situation where a single iod could 'capture' the 1949 * receive lock. 1950 */ 1951 if (rep != NULL) { 1952 if (rep->r_mrep != NULL) { 1953 error = EALREADY; 1954 break; 1955 } 1956 if (nfs_sigintr(rep->r_nmp, rep, rep->r_lwp)) { 1957 error = EINTR; 1958 break; 1959 } 1960 } 1961 if (nmp->nm_rcvlwp == NULL) { 1962 nmp->nm_rcvlwp = curlwp; 1963 break; 1964 } 1965 if (catch) { 1966 cv_timedwait_sig(&nmp->nm_rcvcv, &nmp->nm_lock, 1967 slptimeo); 1968 } else { 1969 cv_timedwait(&nmp->nm_rcvcv, &nmp->nm_lock, 1970 slptimeo); 1971 } 1972 if (catch) { 1973 catch = false; 1974 slptimeo = 2 * hz; 1975 } 1976 } 1977 mutex_exit(&nmp->nm_lock); 1978 return error; 1979 } 1980 1981 /* 1982 * Unlock the stream socket for others. 1983 */ 1984 static void 1985 nfs_rcvunlock(struct nfsmount *nmp) 1986 { 1987 1988 mutex_enter(&nmp->nm_lock); 1989 if (nmp->nm_rcvlwp != curlwp) 1990 panic("nfs rcvunlock"); 1991 nmp->nm_rcvlwp = NULL; 1992 cv_broadcast(&nmp->nm_rcvcv); 1993 mutex_exit(&nmp->nm_lock); 1994 } 1995 1996 /* 1997 * Parse an RPC request 1998 * - verify it 1999 * - allocate and fill in the cred. 2000 */ 2001 int 2002 nfs_getreq(struct nfsrv_descript *nd, struct nfsd *nfsd, int has_header) 2003 { 2004 int len, i; 2005 u_int32_t *tl; 2006 int32_t t1; 2007 struct uio uio; 2008 struct iovec iov; 2009 char *dpos, *cp2, *cp; 2010 u_int32_t nfsvers, auth_type; 2011 uid_t nickuid; 2012 int error = 0, ticklen; 2013 struct mbuf *mrep, *md; 2014 struct nfsuid *nuidp; 2015 struct timeval tvin, tvout; 2016 2017 memset(&tvout, 0, sizeof tvout); /* XXX gcc */ 2018 2019 KASSERT(nd->nd_cr == NULL); 2020 mrep = nd->nd_mrep; 2021 md = nd->nd_md; 2022 dpos = nd->nd_dpos; 2023 if (has_header) { 2024 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); 2025 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); 2026 if (*tl++ != rpc_call) { 2027 m_freem(mrep); 2028 return (EBADRPC); 2029 } 2030 } else 2031 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 2032 nd->nd_repstat = 0; 2033 nd->nd_flag = 0; 2034 if (*tl++ != rpc_vers) { 2035 nd->nd_repstat = ERPCMISMATCH; 2036 nd->nd_procnum = NFSPROC_NOOP; 2037 return (0); 2038 } 2039 if (*tl != nfs_prog) { 2040 nd->nd_repstat = EPROGUNAVAIL; 2041 nd->nd_procnum = NFSPROC_NOOP; 2042 return (0); 2043 } 2044 tl++; 2045 nfsvers = fxdr_unsigned(u_int32_t, *tl++); 2046 if (nfsvers < NFS_VER2 || nfsvers > NFS_VER3) { 2047 nd->nd_repstat = EPROGMISMATCH; 2048 nd->nd_procnum = NFSPROC_NOOP; 2049 return (0); 2050 } 2051 if (nfsvers == NFS_VER3) 2052 nd->nd_flag = ND_NFSV3; 2053 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); 2054 if (nd->nd_procnum == NFSPROC_NULL) 2055 return (0); 2056 if (nd->nd_procnum > NFSPROC_COMMIT || 2057 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { 2058 nd->nd_repstat = EPROCUNAVAIL; 2059 nd->nd_procnum = NFSPROC_NOOP; 2060 return (0); 2061 } 2062 if ((nd->nd_flag & ND_NFSV3) == 0) 2063 nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; 2064 auth_type = *tl++; 2065 len = fxdr_unsigned(int, *tl++); 2066 if (len < 0 || len > RPCAUTH_MAXSIZ) { 2067 m_freem(mrep); 2068 return (EBADRPC); 2069 } 2070 2071 nd->nd_flag &= ~ND_KERBAUTH; 2072 /* 2073 * Handle auth_unix or auth_kerb. 2074 */ 2075 if (auth_type == rpc_auth_unix) { 2076 uid_t uid; 2077 gid_t gid; 2078 2079 nd->nd_cr = kauth_cred_alloc(); 2080 len = fxdr_unsigned(int, *++tl); 2081 if (len < 0 || len > NFS_MAXNAMLEN) { 2082 m_freem(mrep); 2083 error = EBADRPC; 2084 goto errout; 2085 } 2086 nfsm_adv(nfsm_rndup(len)); 2087 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2088 2089 uid = fxdr_unsigned(uid_t, *tl++); 2090 gid = fxdr_unsigned(gid_t, *tl++); 2091 kauth_cred_setuid(nd->nd_cr, uid); 2092 kauth_cred_seteuid(nd->nd_cr, uid); 2093 kauth_cred_setsvuid(nd->nd_cr, uid); 2094 kauth_cred_setgid(nd->nd_cr, gid); 2095 kauth_cred_setegid(nd->nd_cr, gid); 2096 kauth_cred_setsvgid(nd->nd_cr, gid); 2097 2098 len = fxdr_unsigned(int, *tl); 2099 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 2100 m_freem(mrep); 2101 error = EBADRPC; 2102 goto errout; 2103 } 2104 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); 2105 2106 if (len > 0) { 2107 size_t grbuf_size = min(len, NGROUPS) * sizeof(gid_t); 2108 gid_t *grbuf = kmem_alloc(grbuf_size, KM_SLEEP); 2109 2110 for (i = 0; i < len; i++) { 2111 if (i < NGROUPS) /* XXX elad */ 2112 grbuf[i] = fxdr_unsigned(gid_t, *tl++); 2113 else 2114 tl++; 2115 } 2116 kauth_cred_setgroups(nd->nd_cr, grbuf, 2117 min(len, NGROUPS), -1, UIO_SYSSPACE); 2118 kmem_free(grbuf, grbuf_size); 2119 } 2120 2121 len = fxdr_unsigned(int, *++tl); 2122 if (len < 0 || len > RPCAUTH_MAXSIZ) { 2123 m_freem(mrep); 2124 error = EBADRPC; 2125 goto errout; 2126 } 2127 if (len > 0) 2128 nfsm_adv(nfsm_rndup(len)); 2129 } else if (auth_type == rpc_auth_kerb) { 2130 switch (fxdr_unsigned(int, *tl++)) { 2131 case RPCAKN_FULLNAME: 2132 ticklen = fxdr_unsigned(int, *tl); 2133 *((u_int32_t *)nfsd->nfsd_authstr) = *tl; 2134 uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; 2135 nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; 2136 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 2137 m_freem(mrep); 2138 error = EBADRPC; 2139 goto errout; 2140 } 2141 uio.uio_offset = 0; 2142 uio.uio_iov = &iov; 2143 uio.uio_iovcnt = 1; 2144 UIO_SETUP_SYSSPACE(&uio); 2145 iov.iov_base = (void *)&nfsd->nfsd_authstr[4]; 2146 iov.iov_len = RPCAUTH_MAXSIZ - 4; 2147 nfsm_mtouio(&uio, uio.uio_resid); 2148 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2149 if (*tl++ != rpc_auth_kerb || 2150 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { 2151 printf("Bad kerb verifier\n"); 2152 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2153 nd->nd_procnum = NFSPROC_NOOP; 2154 return (0); 2155 } 2156 nfsm_dissect(cp, void *, 4 * NFSX_UNSIGNED); 2157 tl = (u_int32_t *)cp; 2158 if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { 2159 printf("Not fullname kerb verifier\n"); 2160 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2161 nd->nd_procnum = NFSPROC_NOOP; 2162 return (0); 2163 } 2164 cp += NFSX_UNSIGNED; 2165 memcpy(nfsd->nfsd_verfstr, cp, 3 * NFSX_UNSIGNED); 2166 nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; 2167 nd->nd_flag |= ND_KERBFULL; 2168 nfsd->nfsd_flag |= NFSD_NEEDAUTH; 2169 break; 2170 case RPCAKN_NICKNAME: 2171 if (len != 2 * NFSX_UNSIGNED) { 2172 printf("Kerb nickname short\n"); 2173 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); 2174 nd->nd_procnum = NFSPROC_NOOP; 2175 return (0); 2176 } 2177 nickuid = fxdr_unsigned(uid_t, *tl); 2178 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2179 if (*tl++ != rpc_auth_kerb || 2180 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { 2181 printf("Kerb nick verifier bad\n"); 2182 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); 2183 nd->nd_procnum = NFSPROC_NOOP; 2184 return (0); 2185 } 2186 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2187 tvin.tv_sec = *tl++; 2188 tvin.tv_usec = *tl; 2189 2190 LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), 2191 nu_hash) { 2192 if (kauth_cred_geteuid(nuidp->nu_cr) == nickuid && 2193 (!nd->nd_nam2 || 2194 netaddr_match(NU_NETFAM(nuidp), 2195 &nuidp->nu_haddr, nd->nd_nam2))) 2196 break; 2197 } 2198 if (!nuidp) { 2199 nd->nd_repstat = 2200 (NFSERR_AUTHERR|AUTH_REJECTCRED); 2201 nd->nd_procnum = NFSPROC_NOOP; 2202 return (0); 2203 } 2204 2205 /* 2206 * Now, decrypt the timestamp using the session key 2207 * and validate it. 2208 */ 2209 #ifdef NFSKERB 2210 XXX 2211 #endif 2212 2213 tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); 2214 tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); 2215 if (nuidp->nu_expire < time_second || 2216 nuidp->nu_timestamp.tv_sec > tvout.tv_sec || 2217 (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && 2218 nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { 2219 nuidp->nu_expire = 0; 2220 nd->nd_repstat = 2221 (NFSERR_AUTHERR|AUTH_REJECTVERF); 2222 nd->nd_procnum = NFSPROC_NOOP; 2223 return (0); 2224 } 2225 kauth_cred_hold(nuidp->nu_cr); 2226 nd->nd_cr = nuidp->nu_cr; 2227 nd->nd_flag |= ND_KERBNICK; 2228 } 2229 } else { 2230 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); 2231 nd->nd_procnum = NFSPROC_NOOP; 2232 return (0); 2233 } 2234 2235 nd->nd_md = md; 2236 nd->nd_dpos = dpos; 2237 KASSERT((nd->nd_cr == NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) != 0) 2238 || (nd->nd_cr != NULL && (nfsd->nfsd_flag & NFSD_NEEDAUTH) == 0)); 2239 return (0); 2240 nfsmout: 2241 errout: 2242 KASSERT(error != 0); 2243 if (nd->nd_cr != NULL) { 2244 kauth_cred_free(nd->nd_cr); 2245 nd->nd_cr = NULL; 2246 } 2247 return (error); 2248 } 2249 2250 int 2251 nfs_msg(struct lwp *l, const char *server, const char *msg) 2252 { 2253 tpr_t tpr; 2254 2255 if (l) 2256 tpr = tprintf_open(l->l_proc); 2257 else 2258 tpr = NULL; 2259 tprintf(tpr, "nfs server %s: %s\n", server, msg); 2260 tprintf_close(tpr); 2261 return (0); 2262 } 2263 2264 static struct pool nfs_srvdesc_pool; 2265 2266 void 2267 nfsdreq_init(void) 2268 { 2269 2270 pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript), 2271 0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr, IPL_NONE); 2272 } 2273 2274 void 2275 nfsdreq_fini(void) 2276 { 2277 2278 pool_destroy(&nfs_srvdesc_pool); 2279 } 2280 2281 struct nfsrv_descript * 2282 nfsdreq_alloc(void) 2283 { 2284 struct nfsrv_descript *nd; 2285 2286 nd = pool_get(&nfs_srvdesc_pool, PR_WAITOK); 2287 nd->nd_cr = NULL; 2288 return nd; 2289 } 2290 2291 void 2292 nfsdreq_free(struct nfsrv_descript *nd) 2293 { 2294 kauth_cred_t cr; 2295 2296 cr = nd->nd_cr; 2297 if (cr != NULL) { 2298 kauth_cred_free(cr); 2299 } 2300 pool_put(&nfs_srvdesc_pool, nd); 2301 } 2302 2303 #if defined(NFS) 2304 void nfs_reqq_dump(void); 2305 2306 void 2307 nfs_reqq_dump(void) 2308 { 2309 struct nfsreq *r; 2310 2311 TAILQ_FOREACH(r, &nfs_reqq, r_chain) { 2312 printf("%p: proc=%" PRId32 ", lwp=%p, nmp=%p\n" 2313 "\tflags=0x%x, rflags=0x%x\n" 2314 "\treq=%p, rep=%p, xid=0x%" PRIx32 "\n" 2315 "\tretry=%d, rexmit=%d, timer=%d, rtt=%d\n", 2316 r, r->r_procnum, r->r_lwp, r->r_nmp, 2317 r->r_flags, r->r_rflags, 2318 r->r_mreq, r->r_mrep, r->r_xid, 2319 r->r_retry, r->r_rexmit, r->r_timer, r->r_rtt); 2320 } 2321 } 2322 #endif /* defined(NFS) */ 2323
Indexes created Tue Sep 30 11:09:46 GMT 2025