1 /* 2 * Copyright (c) 1989, 1991 The Regents of the University of California. 3 * All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * %sccs.include.redist.c% 9 * 10 * @(#)nfs_socket.c 7.42 (Berkeley) 02/02/93 11 */ 12 13 /* 14 * Socket operations for use by nfs 15 */ 16 17 #include <sys/param.h> 18 #include <sys/systm.h> 19 #include <sys/proc.h> 20 #include <sys/mount.h> 21 #include <sys/kernel.h> 22 #include <sys/mbuf.h> 23 #include <sys/vnode.h> 24 #include <sys/domain.h> 25 #include <sys/protosw.h> 26 #include <sys/socket.h> 27 #include <sys/socketvar.h> 28 #include <sys/syslog.h> 29 #include <sys/tprintf.h> 30 31 #include <netinet/in.h> 32 #include <netinet/tcp.h> 33 #include <nfs/rpcv2.h> 34 #include <nfs/nfsv2.h> 35 #include <nfs/nfs.h> 36 #include <nfs/xdr_subs.h> 37 #include <nfs/nfsm_subs.h> 38 #include <nfs/nfsmount.h> 39 #include <nfs/nfsnode.h> 40 #include <nfs/nfsrtt.h> 41 #include <nfs/nqnfs.h> 42 43 #define TRUE 1 44 #define FALSE 0 45 46 /* 47 * Estimate rto for an nfs rpc sent via. an unreliable datagram. 48 * Use the mean and mean deviation of rtt for the appropriate type of rpc 49 * for the frequent rpcs and a default for the others. 50 * The justification for doing "other" this way is that these rpcs 51 * happen so infrequently that timer est. would probably be stale. 52 * Also, since many of these rpcs are 53 * non-idempotent, a conservative timeout is desired. 54 * getattr, lookup - A+2D 55 * read, write - A+4D 56 * other - nm_timeo 57 */ 58 #define NFS_RTO(n, t) \ 59 ((t) == 0 ? (n)->nm_timeo : \ 60 ((t) < 3 ? \ 61 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ 62 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) 63 #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] 64 #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] 65 /* 66 * External data, mostly RPC constants in XDR form 67 */ 68 extern u_long rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, rpc_auth_unix, 69 rpc_msgaccepted, rpc_call, rpc_autherr, rpc_rejectedcred, 70 rpc_auth_kerb; 71 extern u_long nfs_prog, nfs_vers, nqnfs_prog, nqnfs_vers; 72 extern time_t nqnfsstarttime; 73 extern int nonidempotent[NFS_NPROCS]; 74 75 /* 76 * Maps errno values to nfs error numbers. 77 * Use NFSERR_IO as the catch all for ones not specifically defined in 78 * RFC 1094. 79 */ 80 static int nfsrv_errmap[ELAST] = { 81 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, 82 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 83 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, 84 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR, 85 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 86 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, 87 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 88 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 89 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 90 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 91 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 92 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 93 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO, 94 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, 95 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 96 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 97 NFSERR_IO, 98 }; 99 100 /* 101 * Defines which timer to use for the procnum. 102 * 0 - default 103 * 1 - getattr 104 * 2 - lookup 105 * 3 - read 106 * 4 - write 107 */ 108 static int proct[NFS_NPROCS] = { 109 0, 1, 0, 0, 2, 3, 3, 0, 4, 0, 0, 0, 0, 0, 0, 0, 3, 0, 3, 0, 0, 0, 0, 110 }; 111 112 /* 113 * There is a congestion window for outstanding rpcs maintained per mount 114 * point. The cwnd size is adjusted in roughly the way that: 115 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of 116 * SIGCOMM '88". ACM, August 1988. 117 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout 118 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd 119 * of rpcs is in progress. 120 * (The sent count and cwnd are scaled for integer arith.) 121 * Variants of "slow start" were tried and were found to be too much of a 122 * performance hit (ave. rtt 3 times larger), 123 * I suspect due to the large rtt that nfs rpcs have. 124 */ 125 #define NFS_CWNDSCALE 256 126 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) 127 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; 128 int nfs_sbwait(); 129 void nfs_disconnect(), nfs_realign(), nfsrv_wakenfsd(), nfs_sndunlock(); 130 void nfs_rcvunlock(), nqnfs_serverd(), nqnfs_clientlease(); 131 struct mbuf *nfsm_rpchead(); 132 int nfsrtton = 0; 133 struct nfsrtt nfsrtt; 134 struct nfsd nfsd_head; 135 136 int nfsrv_null(), 137 nfsrv_getattr(), 138 nfsrv_setattr(), 139 nfsrv_lookup(), 140 nfsrv_readlink(), 141 nfsrv_read(), 142 nfsrv_write(), 143 nfsrv_create(), 144 nfsrv_remove(), 145 nfsrv_rename(), 146 nfsrv_link(), 147 nfsrv_symlink(), 148 nfsrv_mkdir(), 149 nfsrv_rmdir(), 150 nfsrv_readdir(), 151 nfsrv_statfs(), 152 nfsrv_noop(), 153 nqnfsrv_readdirlook(), 154 nqnfsrv_getlease(), 155 nqnfsrv_vacated(), 156 nqnfsrv_access(); 157 158 int (*nfsrv_procs[NFS_NPROCS])() = { 159 nfsrv_null, 160 nfsrv_getattr, 161 nfsrv_setattr, 162 nfsrv_noop, 163 nfsrv_lookup, 164 nfsrv_readlink, 165 nfsrv_read, 166 nfsrv_noop, 167 nfsrv_write, 168 nfsrv_create, 169 nfsrv_remove, 170 nfsrv_rename, 171 nfsrv_link, 172 nfsrv_symlink, 173 nfsrv_mkdir, 174 nfsrv_rmdir, 175 nfsrv_readdir, 176 nfsrv_statfs, 177 nqnfsrv_readdirlook, 178 nqnfsrv_getlease, 179 nqnfsrv_vacated, 180 nfsrv_noop, 181 nqnfsrv_access, 182 }; 183 184 struct nfsreq nfsreqh; 185 186 /* 187 * Initialize sockets and congestion for a new NFS connection. 188 * We do not free the sockaddr if error. 189 */ 190 nfs_connect(nmp, rep) 191 register struct nfsmount *nmp; 192 struct nfsreq *rep; 193 { 194 register struct socket *so; 195 int s, error, rcvreserve, sndreserve; 196 struct sockaddr *saddr; 197 struct sockaddr_in *sin; 198 struct mbuf *m; 199 u_short tport; 200 201 nmp->nm_so = (struct socket *)0; 202 saddr = mtod(nmp->nm_nam, struct sockaddr *); 203 if (error = socreate(saddr->sa_family, 204 &nmp->nm_so, nmp->nm_sotype, nmp->nm_soproto)) 205 goto bad; 206 so = nmp->nm_so; 207 nmp->nm_soflags = so->so_proto->pr_flags; 208 209 /* 210 * Some servers require that the client port be a reserved port number. 211 */ 212 if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { 213 MGET(m, M_WAIT, MT_SONAME); 214 sin = mtod(m, struct sockaddr_in *); 215 sin->sin_len = m->m_len = sizeof (struct sockaddr_in); 216 sin->sin_family = AF_INET; 217 sin->sin_addr.s_addr = INADDR_ANY; 218 tport = IPPORT_RESERVED - 1; 219 sin->sin_port = htons(tport); 220 while ((error = sobind(so, m)) == EADDRINUSE && 221 --tport > IPPORT_RESERVED / 2) 222 sin->sin_port = htons(tport); 223 m_freem(m); 224 if (error) 225 goto bad; 226 } 227 228 /* 229 * Protocols that do not require connections may be optionally left 230 * unconnected for servers that reply from a port other than NFS_PORT. 231 */ 232 if (nmp->nm_flag & NFSMNT_NOCONN) { 233 if (nmp->nm_soflags & PR_CONNREQUIRED) { 234 error = ENOTCONN; 235 goto bad; 236 } 237 } else { 238 if (error = soconnect(so, nmp->nm_nam)) 239 goto bad; 240 241 /* 242 * Wait for the connection to complete. Cribbed from the 243 * connect system call but with the wait timing out so 244 * that interruptible mounts don't hang here for a long time. 245 */ 246 s = splnet(); 247 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 248 (void) tsleep((caddr_t)&so->so_timeo, PSOCK, 249 "nfscon", 2 * hz); 250 if ((so->so_state & SS_ISCONNECTING) && 251 so->so_error == 0 && rep && 252 (error = nfs_sigintr(nmp, rep, rep->r_procp))) { 253 so->so_state &= ~SS_ISCONNECTING; 254 splx(s); 255 goto bad; 256 } 257 } 258 if (so->so_error) { 259 error = so->so_error; 260 so->so_error = 0; 261 splx(s); 262 goto bad; 263 } 264 splx(s); 265 } 266 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 267 so->so_rcv.sb_timeo = (5 * hz); 268 so->so_snd.sb_timeo = (5 * hz); 269 } else { 270 so->so_rcv.sb_timeo = 0; 271 so->so_snd.sb_timeo = 0; 272 } 273 if (nmp->nm_sotype == SOCK_DGRAM) { 274 sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR; 275 rcvreserve = nmp->nm_rsize + NFS_MAXPKTHDR; 276 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 277 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 278 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR) * 2; 279 } else { 280 if (nmp->nm_sotype != SOCK_STREAM) 281 panic("nfscon sotype"); 282 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 283 MGET(m, M_WAIT, MT_SOOPTS); 284 *mtod(m, int *) = 1; 285 m->m_len = sizeof(int); 286 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 287 } 288 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 289 MGET(m, M_WAIT, MT_SOOPTS); 290 *mtod(m, int *) = 1; 291 m->m_len = sizeof(int); 292 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 293 } 294 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + sizeof (u_long)) 295 * 2; 296 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + sizeof (u_long)) 297 * 2; 298 } 299 if (error = soreserve(so, sndreserve, rcvreserve)) 300 goto bad; 301 so->so_rcv.sb_flags |= SB_NOINTR; 302 so->so_snd.sb_flags |= SB_NOINTR; 303 304 /* Initialize other non-zero congestion variables */ 305 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 306 nmp->nm_srtt[4] = (NFS_TIMEO << 3); 307 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = 308 nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0; 309 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ 310 nmp->nm_sent = 0; 311 nmp->nm_timeouts = 0; 312 return (0); 313 314 bad: 315 nfs_disconnect(nmp); 316 return (error); 317 } 318 319 /* 320 * Reconnect routine: 321 * Called when a connection is broken on a reliable protocol. 322 * - clean up the old socket 323 * - nfs_connect() again 324 * - set R_MUSTRESEND for all outstanding requests on mount point 325 * If this fails the mount point is DEAD! 326 * nb: Must be called with the nfs_sndlock() set on the mount point. 327 */ 328 nfs_reconnect(rep) 329 register struct nfsreq *rep; 330 { 331 register struct nfsreq *rp; 332 register struct nfsmount *nmp = rep->r_nmp; 333 int error; 334 335 nfs_disconnect(nmp); 336 while (error = nfs_connect(nmp, rep)) { 337 if (error == EINTR || error == ERESTART) 338 return (EINTR); 339 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); 340 } 341 342 /* 343 * Loop through outstanding request list and fix up all requests 344 * on old socket. 345 */ 346 rp = nfsreqh.r_next; 347 while (rp != &nfsreqh) { 348 if (rp->r_nmp == nmp) 349 rp->r_flags |= R_MUSTRESEND; 350 rp = rp->r_next; 351 } 352 return (0); 353 } 354 355 /* 356 * NFS disconnect. Clean up and unlink. 357 */ 358 void 359 nfs_disconnect(nmp) 360 register struct nfsmount *nmp; 361 { 362 register struct socket *so; 363 364 if (nmp->nm_so) { 365 so = nmp->nm_so; 366 nmp->nm_so = (struct socket *)0; 367 soshutdown(so, 2); 368 soclose(so); 369 } 370 } 371 372 /* 373 * This is the nfs send routine. For connection based socket types, it 374 * must be called with an nfs_sndlock() on the socket. 375 * "rep == NULL" indicates that it has been called from a server. 376 * For the client side: 377 * - return EINTR if the RPC is terminated, 0 otherwise 378 * - set R_MUSTRESEND if the send fails for any reason 379 * - do any cleanup required by recoverable socket errors (???) 380 * For the server side: 381 * - return EINTR or ERESTART if interrupted by a signal 382 * - return EPIPE if a connection is lost for connection based sockets (TCP...) 383 * - do any cleanup required by recoverable socket errors (???) 384 */ 385 nfs_send(so, nam, top, rep) 386 register struct socket *so; 387 struct mbuf *nam; 388 register struct mbuf *top; 389 struct nfsreq *rep; 390 { 391 struct mbuf *sendnam; 392 int error, soflags, flags; 393 394 if (rep) { 395 if (rep->r_flags & R_SOFTTERM) { 396 m_freem(top); 397 return (EINTR); 398 } 399 if ((so = rep->r_nmp->nm_so) == NULL) { 400 rep->r_flags |= R_MUSTRESEND; 401 m_freem(top); 402 return (0); 403 } 404 rep->r_flags &= ~R_MUSTRESEND; 405 soflags = rep->r_nmp->nm_soflags; 406 } else 407 soflags = so->so_proto->pr_flags; 408 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) 409 sendnam = (struct mbuf *)0; 410 else 411 sendnam = nam; 412 if (so->so_type == SOCK_SEQPACKET) 413 flags = MSG_EOR; 414 else 415 flags = 0; 416 417 error = sosend(so, sendnam, (struct uio *)0, top, 418 (struct mbuf *)0, flags); 419 if (error) { 420 if (rep) { 421 log(LOG_INFO, "nfs send error %d for server %s\n",error, 422 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 423 /* 424 * Deal with errors for the client side. 425 */ 426 if (rep->r_flags & R_SOFTTERM) 427 error = EINTR; 428 else 429 rep->r_flags |= R_MUSTRESEND; 430 } else 431 log(LOG_INFO, "nfsd send error %d\n", error); 432 433 /* 434 * Handle any recoverable (soft) socket errors here. (???) 435 */ 436 if (error != EINTR && error != ERESTART && 437 error != EWOULDBLOCK && error != EPIPE) 438 error = 0; 439 } 440 return (error); 441 } 442 443 /* 444 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 445 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 446 * Mark and consolidate the data into a new mbuf list. 447 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 448 * small mbufs. 449 * For SOCK_STREAM we must be very careful to read an entire record once 450 * we have read any of it, even if the system call has been interrupted. 451 */ 452 nfs_receive(rep, aname, mp) 453 register struct nfsreq *rep; 454 struct mbuf **aname; 455 struct mbuf **mp; 456 { 457 register struct socket *so; 458 struct uio auio; 459 struct iovec aio; 460 register struct mbuf *m; 461 struct mbuf *control; 462 u_long len; 463 struct mbuf **getnam; 464 int error, sotype, rcvflg; 465 struct proc *p = curproc; /* XXX */ 466 467 /* 468 * Set up arguments for soreceive() 469 */ 470 *mp = (struct mbuf *)0; 471 *aname = (struct mbuf *)0; 472 sotype = rep->r_nmp->nm_sotype; 473 474 /* 475 * For reliable protocols, lock against other senders/receivers 476 * in case a reconnect is necessary. 477 * For SOCK_STREAM, first get the Record Mark to find out how much 478 * more there is to get. 479 * We must lock the socket against other receivers 480 * until we have an entire rpc request/reply. 481 */ 482 if (sotype != SOCK_DGRAM) { 483 if (error = nfs_sndlock(&rep->r_nmp->nm_flag, rep)) 484 return (error); 485 tryagain: 486 /* 487 * Check for fatal errors and resending request. 488 */ 489 /* 490 * Ugh: If a reconnect attempt just happened, nm_so 491 * would have changed. NULL indicates a failed 492 * attempt that has essentially shut down this 493 * mount point. 494 */ 495 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 496 nfs_sndunlock(&rep->r_nmp->nm_flag); 497 return (EINTR); 498 } 499 if ((so = rep->r_nmp->nm_so) == NULL) { 500 if (error = nfs_reconnect(rep)) { 501 nfs_sndunlock(&rep->r_nmp->nm_flag); 502 return (error); 503 } 504 goto tryagain; 505 } 506 while (rep->r_flags & R_MUSTRESEND) { 507 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 508 nfsstats.rpcretries++; 509 if (error = nfs_send(so, rep->r_nmp->nm_nam, m, rep)) { 510 if (error == EINTR || error == ERESTART || 511 (error = nfs_reconnect(rep))) { 512 nfs_sndunlock(&rep->r_nmp->nm_flag); 513 return (error); 514 } 515 goto tryagain; 516 } 517 } 518 nfs_sndunlock(&rep->r_nmp->nm_flag); 519 if (sotype == SOCK_STREAM) { 520 aio.iov_base = (caddr_t) &len; 521 aio.iov_len = sizeof(u_long); 522 auio.uio_iov = &aio; 523 auio.uio_iovcnt = 1; 524 auio.uio_segflg = UIO_SYSSPACE; 525 auio.uio_rw = UIO_READ; 526 auio.uio_offset = 0; 527 auio.uio_resid = sizeof(u_long); 528 auio.uio_procp = p; 529 do { 530 rcvflg = MSG_WAITALL; 531 error = soreceive(so, (struct mbuf **)0, &auio, 532 (struct mbuf **)0, (struct mbuf **)0, &rcvflg); 533 if (error == EWOULDBLOCK && rep) { 534 if (rep->r_flags & R_SOFTTERM) 535 return (EINTR); 536 } 537 } while (error == EWOULDBLOCK); 538 if (!error && auio.uio_resid > 0) { 539 log(LOG_INFO, 540 "short receive (%d/%d) from nfs server %s\n", 541 sizeof(u_long) - auio.uio_resid, 542 sizeof(u_long), 543 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 544 error = EPIPE; 545 } 546 if (error) 547 goto errout; 548 len = ntohl(len) & ~0x80000000; 549 /* 550 * This is SERIOUS! We are out of sync with the sender 551 * and forcing a disconnect/reconnect is all I can do. 552 */ 553 if (len > NFS_MAXPACKET) { 554 log(LOG_ERR, "%s (%d) from nfs server %s\n", 555 "impossible packet length", 556 len, 557 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 558 error = EFBIG; 559 goto errout; 560 } 561 auio.uio_resid = len; 562 do { 563 rcvflg = MSG_WAITALL; 564 error = soreceive(so, (struct mbuf **)0, 565 &auio, mp, (struct mbuf **)0, &rcvflg); 566 } while (error == EWOULDBLOCK || error == EINTR || 567 error == ERESTART); 568 if (!error && auio.uio_resid > 0) { 569 log(LOG_INFO, 570 "short receive (%d/%d) from nfs server %s\n", 571 len - auio.uio_resid, len, 572 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 573 error = EPIPE; 574 } 575 } else { 576 /* 577 * NB: Since uio_resid is big, MSG_WAITALL is ignored 578 * and soreceive() will return when it has either a 579 * control msg or a data msg. 580 * We have no use for control msg., but must grab them 581 * and then throw them away so we know what is going 582 * on. 583 */ 584 auio.uio_resid = len = 100000000; /* Anything Big */ 585 auio.uio_procp = p; 586 do { 587 rcvflg = 0; 588 error = soreceive(so, (struct mbuf **)0, 589 &auio, mp, &control, &rcvflg); 590 if (control) 591 m_freem(control); 592 if (error == EWOULDBLOCK && rep) { 593 if (rep->r_flags & R_SOFTTERM) 594 return (EINTR); 595 } 596 } while (error == EWOULDBLOCK || 597 (!error && *mp == NULL && control)); 598 if ((rcvflg & MSG_EOR) == 0) 599 printf("Egad!!\n"); 600 if (!error && *mp == NULL) 601 error = EPIPE; 602 len -= auio.uio_resid; 603 } 604 errout: 605 if (error && error != EINTR && error != ERESTART) { 606 m_freem(*mp); 607 *mp = (struct mbuf *)0; 608 if (error != EPIPE) 609 log(LOG_INFO, 610 "receive error %d from nfs server %s\n", 611 error, 612 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 613 error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); 614 if (!error) 615 error = nfs_reconnect(rep); 616 if (!error) 617 goto tryagain; 618 } 619 } else { 620 if ((so = rep->r_nmp->nm_so) == NULL) 621 return (EACCES); 622 if (so->so_state & SS_ISCONNECTED) 623 getnam = (struct mbuf **)0; 624 else 625 getnam = aname; 626 auio.uio_resid = len = 1000000; 627 auio.uio_procp = p; 628 do { 629 rcvflg = 0; 630 error = soreceive(so, getnam, &auio, mp, 631 (struct mbuf **)0, &rcvflg); 632 if (error == EWOULDBLOCK && 633 (rep->r_flags & R_SOFTTERM)) 634 return (EINTR); 635 } while (error == EWOULDBLOCK); 636 len -= auio.uio_resid; 637 } 638 if (error) { 639 m_freem(*mp); 640 *mp = (struct mbuf *)0; 641 } 642 /* 643 * Search for any mbufs that are not a multiple of 4 bytes long 644 * or with m_data not longword aligned. 645 * These could cause pointer alignment problems, so copy them to 646 * well aligned mbufs. 647 */ 648 nfs_realign(*mp, 5 * NFSX_UNSIGNED); 649 return (error); 650 } 651 652 /* 653 * Implement receipt of reply on a socket. 654 * We must search through the list of received datagrams matching them 655 * with outstanding requests using the xid, until ours is found. 656 */ 657 /* ARGSUSED */ 658 nfs_reply(myrep) 659 struct nfsreq *myrep; 660 { 661 register struct nfsreq *rep; 662 register struct nfsmount *nmp = myrep->r_nmp; 663 register long t1; 664 struct mbuf *mrep, *nam, *md; 665 u_long rxid, *tl; 666 caddr_t dpos, cp2; 667 int error; 668 669 /* 670 * Loop around until we get our own reply 671 */ 672 for (;;) { 673 /* 674 * Lock against other receivers so that I don't get stuck in 675 * sbwait() after someone else has received my reply for me. 676 * Also necessary for connection based protocols to avoid 677 * race conditions during a reconnect. 678 */ 679 if (error = nfs_rcvlock(myrep)) 680 return (error); 681 /* Already received, bye bye */ 682 if (myrep->r_mrep != NULL) { 683 nfs_rcvunlock(&nmp->nm_flag); 684 return (0); 685 } 686 /* 687 * Get the next Rpc reply off the socket 688 */ 689 error = nfs_receive(myrep, &nam, &mrep); 690 nfs_rcvunlock(&nmp->nm_flag); 691 if (error) { 692 693 /* 694 * Ignore routing errors on connectionless protocols?? 695 */ 696 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 697 nmp->nm_so->so_error = 0; 698 continue; 699 } 700 return (error); 701 } 702 if (nam) 703 m_freem(nam); 704 705 /* 706 * Get the xid and check that it is an rpc reply 707 */ 708 md = mrep; 709 dpos = mtod(md, caddr_t); 710 nfsm_dissect(tl, u_long *, 2*NFSX_UNSIGNED); 711 rxid = *tl++; 712 if (*tl != rpc_reply) { 713 if (nmp->nm_flag & NFSMNT_NQNFS) { 714 if (nqnfs_callback(nmp, mrep, md, dpos)) 715 nfsstats.rpcinvalid++; 716 } else { 717 nfsstats.rpcinvalid++; 718 m_freem(mrep); 719 } 720 nfsmout: 721 continue; 722 } 723 724 /* 725 * Loop through the request list to match up the reply 726 * Iff no match, just drop the datagram 727 */ 728 rep = nfsreqh.r_next; 729 while (rep != &nfsreqh) { 730 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 731 /* Found it.. */ 732 rep->r_mrep = mrep; 733 rep->r_md = md; 734 rep->r_dpos = dpos; 735 if (nfsrtton) { 736 struct rttl *rt; 737 738 rt = &nfsrtt.rttl[nfsrtt.pos]; 739 rt->proc = rep->r_procnum; 740 rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); 741 rt->sent = nmp->nm_sent; 742 rt->cwnd = nmp->nm_cwnd; 743 rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; 744 rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; 745 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; 746 rt->tstamp = time; 747 if (rep->r_flags & R_TIMING) 748 rt->rtt = rep->r_rtt; 749 else 750 rt->rtt = 1000000; 751 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 752 } 753 /* 754 * Update congestion window. 755 * Do the additive increase of 756 * one rpc/rtt. 757 */ 758 if (nmp->nm_cwnd <= nmp->nm_sent) { 759 nmp->nm_cwnd += 760 (NFS_CWNDSCALE * NFS_CWNDSCALE + 761 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 762 if (nmp->nm_cwnd > NFS_MAXCWND) 763 nmp->nm_cwnd = NFS_MAXCWND; 764 } 765 rep->r_flags &= ~R_SENT; 766 nmp->nm_sent -= NFS_CWNDSCALE; 767 /* 768 * Update rtt using a gain of 0.125 on the mean 769 * and a gain of 0.25 on the deviation. 770 */ 771 if (rep->r_flags & R_TIMING) { 772 /* 773 * Since the timer resolution of 774 * NFS_HZ is so course, it can often 775 * result in r_rtt == 0. Since 776 * r_rtt == N means that the actual 777 * rtt is between N+dt and N+2-dt ticks, 778 * add 1. 779 */ 780 t1 = rep->r_rtt + 1; 781 t1 -= (NFS_SRTT(rep) >> 3); 782 NFS_SRTT(rep) += t1; 783 if (t1 < 0) 784 t1 = -t1; 785 t1 -= (NFS_SDRTT(rep) >> 2); 786 NFS_SDRTT(rep) += t1; 787 } 788 nmp->nm_timeouts = 0; 789 break; 790 } 791 rep = rep->r_next; 792 } 793 /* 794 * If not matched to a request, drop it. 795 * If it's mine, get out. 796 */ 797 if (rep == &nfsreqh) { 798 nfsstats.rpcunexpected++; 799 m_freem(mrep); 800 } else if (rep == myrep) { 801 if (rep->r_mrep == NULL) 802 panic("nfsreply nil"); 803 return (0); 804 } 805 } 806 } 807 808 /* 809 * nfs_request - goes something like this 810 * - fill in request struct 811 * - links it into list 812 * - calls nfs_send() for first transmit 813 * - calls nfs_receive() to get reply 814 * - break down rpc header and return with nfs reply pointed to 815 * by mrep or error 816 * nb: always frees up mreq mbuf list 817 */ 818 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) 819 struct vnode *vp; 820 struct mbuf *mrest; 821 int procnum; 822 struct proc *procp; 823 struct ucred *cred; 824 struct mbuf **mrp; 825 struct mbuf **mdp; 826 caddr_t *dposp; 827 { 828 register struct mbuf *m, *mrep; 829 register struct nfsreq *rep; 830 register u_long *tl; 831 register int i; 832 struct nfsmount *nmp; 833 struct mbuf *md, *mheadend; 834 struct nfsreq *reph; 835 struct nfsnode *tp, *np; 836 time_t reqtime, waituntil; 837 caddr_t dpos, cp2; 838 int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; 839 int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; 840 u_long xid; 841 u_quad_t frev; 842 char *auth_str; 843 844 nmp = VFSTONFS(vp->v_mount); 845 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); 846 rep->r_nmp = nmp; 847 rep->r_vp = vp; 848 rep->r_procp = procp; 849 rep->r_procnum = procnum; 850 i = 0; 851 m = mrest; 852 while (m) { 853 i += m->m_len; 854 m = m->m_next; 855 } 856 mrest_len = i; 857 858 /* 859 * Get the RPC header with authorization. 860 */ 861 kerbauth: 862 auth_str = (char *)0; 863 if (nmp->nm_flag & NFSMNT_KERB) { 864 if (failed_auth) { 865 error = nfs_getauth(nmp, rep, cred, &auth_type, 866 &auth_str, &auth_len); 867 if (error) { 868 free((caddr_t)rep, M_NFSREQ); 869 m_freem(mrest); 870 return (error); 871 } 872 } else { 873 auth_type = RPCAUTH_UNIX; 874 auth_len = 5 * NFSX_UNSIGNED; 875 } 876 } else { 877 auth_type = RPCAUTH_UNIX; 878 if (cred->cr_ngroups < 1) 879 panic("nfsreq nogrps"); 880 auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ? 881 nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) + 882 5 * NFSX_UNSIGNED; 883 } 884 m = nfsm_rpchead(cred, (nmp->nm_flag & NFSMNT_NQNFS), procnum, 885 auth_type, auth_len, auth_str, mrest, mrest_len, &mheadend, &xid); 886 if (auth_str) 887 free(auth_str, M_TEMP); 888 889 /* 890 * For stream protocols, insert a Sun RPC Record Mark. 891 */ 892 if (nmp->nm_sotype == SOCK_STREAM) { 893 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 894 *mtod(m, u_long *) = htonl(0x80000000 | 895 (m->m_pkthdr.len - NFSX_UNSIGNED)); 896 } 897 rep->r_mreq = m; 898 rep->r_xid = xid; 899 tryagain: 900 if (nmp->nm_flag & NFSMNT_SOFT) 901 rep->r_retry = nmp->nm_retry; 902 else 903 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 904 rep->r_rtt = rep->r_rexmit = 0; 905 if (proct[procnum] > 0) 906 rep->r_flags = R_TIMING; 907 else 908 rep->r_flags = 0; 909 rep->r_mrep = NULL; 910 911 /* 912 * Do the client side RPC. 913 */ 914 nfsstats.rpcrequests++; 915 /* 916 * Chain request into list of outstanding requests. Be sure 917 * to put it LAST so timer finds oldest requests first. 918 */ 919 s = splsoftclock(); 920 reph = &nfsreqh; 921 reph->r_prev->r_next = rep; 922 rep->r_prev = reph->r_prev; 923 reph->r_prev = rep; 924 rep->r_next = reph; 925 926 /* Get send time for nqnfs */ 927 reqtime = time.tv_sec; 928 929 /* 930 * If backing off another request or avoiding congestion, don't 931 * send this one now but let timer do it. If not timing a request, 932 * do it now. 933 */ 934 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 935 (nmp->nm_flag & NFSMNT_DUMBTIMR) || 936 nmp->nm_sent < nmp->nm_cwnd)) { 937 splx(s); 938 if (nmp->nm_soflags & PR_CONNREQUIRED) 939 error = nfs_sndlock(&nmp->nm_flag, rep); 940 if (!error) { 941 m = m_copym(m, 0, M_COPYALL, M_WAIT); 942 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep); 943 if (nmp->nm_soflags & PR_CONNREQUIRED) 944 nfs_sndunlock(&nmp->nm_flag); 945 } 946 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 947 nmp->nm_sent += NFS_CWNDSCALE; 948 rep->r_flags |= R_SENT; 949 } 950 } else { 951 splx(s); 952 rep->r_rtt = -1; 953 } 954 955 /* 956 * Wait for the reply from our send or the timer's. 957 */ 958 if (!error || error == EPIPE) 959 error = nfs_reply(rep); 960 961 /* 962 * RPC done, unlink the request. 963 */ 964 s = splsoftclock(); 965 rep->r_prev->r_next = rep->r_next; 966 rep->r_next->r_prev = rep->r_prev; 967 splx(s); 968 969 /* 970 * Decrement the outstanding request count. 971 */ 972 if (rep->r_flags & R_SENT) { 973 rep->r_flags &= ~R_SENT; /* paranoia */ 974 nmp->nm_sent -= NFS_CWNDSCALE; 975 } 976 977 /* 978 * If there was a successful reply and a tprintf msg. 979 * tprintf a response. 980 */ 981 if (!error && (rep->r_flags & R_TPRINTFMSG)) 982 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, 983 "is alive again"); 984 mrep = rep->r_mrep; 985 md = rep->r_md; 986 dpos = rep->r_dpos; 987 if (error) { 988 m_freem(rep->r_mreq); 989 free((caddr_t)rep, M_NFSREQ); 990 return (error); 991 } 992 993 /* 994 * break down the rpc header and check if ok 995 */ 996 nfsm_dissect(tl, u_long *, 3*NFSX_UNSIGNED); 997 if (*tl++ == rpc_msgdenied) { 998 if (*tl == rpc_mismatch) 999 error = EOPNOTSUPP; 1000 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 1001 if (*tl == rpc_rejectedcred && failed_auth == 0) { 1002 failed_auth++; 1003 mheadend->m_next = (struct mbuf *)0; 1004 m_freem(mrep); 1005 m_freem(rep->r_mreq); 1006 goto kerbauth; 1007 } else 1008 error = EAUTH; 1009 } else 1010 error = EACCES; 1011 m_freem(mrep); 1012 m_freem(rep->r_mreq); 1013 free((caddr_t)rep, M_NFSREQ); 1014 return (error); 1015 } 1016 1017 /* 1018 * skip over the auth_verf, someday we may want to cache auth_short's 1019 * for nfs_reqhead(), but for now just dump it 1020 */ 1021 if (*++tl != 0) { 1022 i = nfsm_rndup(fxdr_unsigned(long, *tl)); 1023 nfsm_adv(i); 1024 } 1025 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1026 /* 0 == ok */ 1027 if (*tl == 0) { 1028 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1029 if (*tl != 0) { 1030 error = fxdr_unsigned(int, *tl); 1031 m_freem(mrep); 1032 if ((nmp->nm_flag & NFSMNT_NQNFS) && 1033 error == NQNFS_TRYLATER) { 1034 error = 0; 1035 waituntil = time.tv_sec + trylater_delay; 1036 while (time.tv_sec < waituntil) 1037 (void) tsleep((caddr_t)&lbolt, 1038 PSOCK, "nqnfstry", 0); 1039 trylater_delay *= nfs_backoff[trylater_cnt]; 1040 if (trylater_cnt < 7) 1041 trylater_cnt++; 1042 goto tryagain; 1043 } 1044 1045 /* 1046 * If the File Handle was stale, invalidate the 1047 * lookup cache, just in case. 1048 */ 1049 if (error == ESTALE) 1050 cache_purge(vp); 1051 m_freem(rep->r_mreq); 1052 free((caddr_t)rep, M_NFSREQ); 1053 return (error); 1054 } 1055 1056 /* 1057 * For nqnfs, get any lease in reply 1058 */ 1059 if (nmp->nm_flag & NFSMNT_NQNFS) { 1060 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1061 if (*tl) { 1062 np = VTONFS(vp); 1063 nqlflag = fxdr_unsigned(int, *tl); 1064 nfsm_dissect(tl, u_long *, 4*NFSX_UNSIGNED); 1065 cachable = fxdr_unsigned(int, *tl++); 1066 reqtime += fxdr_unsigned(int, *tl++); 1067 if (reqtime > time.tv_sec) { 1068 fxdr_hyper(tl, &frev); 1069 nqnfs_clientlease(nmp, np, nqlflag, 1070 cachable, reqtime, frev); 1071 } 1072 } 1073 } 1074 *mrp = mrep; 1075 *mdp = md; 1076 *dposp = dpos; 1077 m_freem(rep->r_mreq); 1078 FREE((caddr_t)rep, M_NFSREQ); 1079 return (0); 1080 } 1081 m_freem(mrep); 1082 m_freem(rep->r_mreq); 1083 free((caddr_t)rep, M_NFSREQ); 1084 error = EPROTONOSUPPORT; 1085 nfsmout: 1086 return (error); 1087 } 1088 1089 /* 1090 * Generate the rpc reply header 1091 * siz arg. is used to decide if adding a cluster is worthwhile 1092 */ 1093 nfs_rephead(siz, nd, err, cache, frev, mrq, mbp, bposp) 1094 int siz; 1095 struct nfsd *nd; 1096 int err; 1097 int cache; 1098 u_quad_t *frev; 1099 struct mbuf **mrq; 1100 struct mbuf **mbp; 1101 caddr_t *bposp; 1102 { 1103 register u_long *tl; 1104 register struct mbuf *mreq; 1105 caddr_t bpos; 1106 struct mbuf *mb, *mb2; 1107 1108 MGETHDR(mreq, M_WAIT, MT_DATA); 1109 mb = mreq; 1110 /* 1111 * If this is a big reply, use a cluster else 1112 * try and leave leading space for the lower level headers. 1113 */ 1114 siz += RPC_REPLYSIZ; 1115 if (siz >= MINCLSIZE) { 1116 MCLGET(mreq, M_WAIT); 1117 } else 1118 mreq->m_data += max_hdr; 1119 tl = mtod(mreq, u_long *); 1120 mreq->m_len = 6*NFSX_UNSIGNED; 1121 bpos = ((caddr_t)tl)+mreq->m_len; 1122 *tl++ = nd->nd_retxid; 1123 *tl++ = rpc_reply; 1124 if (err == ERPCMISMATCH || err == NQNFS_AUTHERR) { 1125 *tl++ = rpc_msgdenied; 1126 if (err == NQNFS_AUTHERR) { 1127 *tl++ = rpc_autherr; 1128 *tl = rpc_rejectedcred; 1129 mreq->m_len -= NFSX_UNSIGNED; 1130 bpos -= NFSX_UNSIGNED; 1131 } else { 1132 *tl++ = rpc_mismatch; 1133 *tl++ = txdr_unsigned(2); 1134 *tl = txdr_unsigned(2); 1135 } 1136 } else { 1137 *tl++ = rpc_msgaccepted; 1138 *tl++ = 0; 1139 *tl++ = 0; 1140 switch (err) { 1141 case EPROGUNAVAIL: 1142 *tl = txdr_unsigned(RPC_PROGUNAVAIL); 1143 break; 1144 case EPROGMISMATCH: 1145 *tl = txdr_unsigned(RPC_PROGMISMATCH); 1146 nfsm_build(tl, u_long *, 2*NFSX_UNSIGNED); 1147 *tl++ = txdr_unsigned(2); 1148 *tl = txdr_unsigned(2); /* someday 3 */ 1149 break; 1150 case EPROCUNAVAIL: 1151 *tl = txdr_unsigned(RPC_PROCUNAVAIL); 1152 break; 1153 default: 1154 *tl = 0; 1155 if (err != VNOVAL) { 1156 nfsm_build(tl, u_long *, NFSX_UNSIGNED); 1157 if (err) 1158 *tl = txdr_unsigned(nfsrv_errmap[err - 1]); 1159 else 1160 *tl = 0; 1161 } 1162 break; 1163 }; 1164 } 1165 1166 /* 1167 * For nqnfs, piggyback lease as requested. 1168 */ 1169 if (nd->nd_nqlflag != NQL_NOVAL && err == 0) { 1170 if (nd->nd_nqlflag) { 1171 nfsm_build(tl, u_long *, 5*NFSX_UNSIGNED); 1172 *tl++ = txdr_unsigned(nd->nd_nqlflag); 1173 *tl++ = txdr_unsigned(cache); 1174 *tl++ = txdr_unsigned(nd->nd_duration); 1175 txdr_hyper(frev, tl); 1176 } else { 1177 if (nd->nd_nqlflag != 0) 1178 panic("nqreph"); 1179 nfsm_build(tl, u_long *, NFSX_UNSIGNED); 1180 *tl = 0; 1181 } 1182 } 1183 *mrq = mreq; 1184 *mbp = mb; 1185 *bposp = bpos; 1186 if (err != 0 && err != VNOVAL) 1187 nfsstats.srvrpc_errs++; 1188 return (0); 1189 } 1190 1191 /* 1192 * Nfs timer routine 1193 * Scan the nfsreq list and retranmit any requests that have timed out 1194 * To avoid retransmission attempts on STREAM sockets (in the future) make 1195 * sure to set the r_retry field to 0 (implies nm_retry == 0). 1196 */ 1197 void 1198 nfs_timer(arg) 1199 void *arg; 1200 { 1201 register struct nfsreq *rep; 1202 register struct mbuf *m; 1203 register struct socket *so; 1204 register struct nfsmount *nmp; 1205 register int timeo; 1206 static long lasttime = 0; 1207 int s, error; 1208 1209 s = splnet(); 1210 for (rep = nfsreqh.r_next; rep != &nfsreqh; rep = rep->r_next) { 1211 nmp = rep->r_nmp; 1212 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) 1213 continue; 1214 if (nfs_sigintr(nmp, rep, rep->r_procp)) { 1215 rep->r_flags |= R_SOFTTERM; 1216 continue; 1217 } 1218 if (rep->r_rtt >= 0) { 1219 rep->r_rtt++; 1220 if (nmp->nm_flag & NFSMNT_DUMBTIMR) 1221 timeo = nmp->nm_timeo; 1222 else 1223 timeo = NFS_RTO(nmp, proct[rep->r_procnum]); 1224 if (nmp->nm_timeouts > 0) 1225 timeo *= nfs_backoff[nmp->nm_timeouts - 1]; 1226 if (rep->r_rtt <= timeo) 1227 continue; 1228 if (nmp->nm_timeouts < 8) 1229 nmp->nm_timeouts++; 1230 } 1231 /* 1232 * Check for server not responding 1233 */ 1234 if ((rep->r_flags & R_TPRINTFMSG) == 0 && 1235 rep->r_rexmit > nmp->nm_deadthresh) { 1236 nfs_msg(rep->r_procp, 1237 nmp->nm_mountp->mnt_stat.f_mntfromname, 1238 "not responding"); 1239 rep->r_flags |= R_TPRINTFMSG; 1240 } 1241 if (rep->r_rexmit >= rep->r_retry) { /* too many */ 1242 nfsstats.rpctimeouts++; 1243 rep->r_flags |= R_SOFTTERM; 1244 continue; 1245 } 1246 if (nmp->nm_sotype != SOCK_DGRAM) { 1247 if (++rep->r_rexmit > NFS_MAXREXMIT) 1248 rep->r_rexmit = NFS_MAXREXMIT; 1249 continue; 1250 } 1251 if ((so = nmp->nm_so) == NULL) 1252 continue; 1253 1254 /* 1255 * If there is enough space and the window allows.. 1256 * Resend it 1257 * Set r_rtt to -1 in case we fail to send it now. 1258 */ 1259 rep->r_rtt = -1; 1260 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && 1261 ((nmp->nm_flag & NFSMNT_DUMBTIMR) || 1262 (rep->r_flags & R_SENT) || 1263 nmp->nm_sent < nmp->nm_cwnd) && 1264 (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ 1265 if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) 1266 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1267 (struct mbuf *)0, (struct mbuf *)0); 1268 else 1269 error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, 1270 nmp->nm_nam, (struct mbuf *)0); 1271 if (error) { 1272 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) 1273 so->so_error = 0; 1274 } else { 1275 /* 1276 * Iff first send, start timing 1277 * else turn timing off, backoff timer 1278 * and divide congestion window by 2. 1279 */ 1280 if (rep->r_flags & R_SENT) { 1281 rep->r_flags &= ~R_TIMING; 1282 if (++rep->r_rexmit > NFS_MAXREXMIT) 1283 rep->r_rexmit = NFS_MAXREXMIT; 1284 nmp->nm_cwnd >>= 1; 1285 if (nmp->nm_cwnd < NFS_CWNDSCALE) 1286 nmp->nm_cwnd = NFS_CWNDSCALE; 1287 nfsstats.rpcretries++; 1288 } else { 1289 rep->r_flags |= R_SENT; 1290 nmp->nm_sent += NFS_CWNDSCALE; 1291 } 1292 rep->r_rtt = 0; 1293 } 1294 } 1295 } 1296 1297 /* 1298 * Call the nqnfs server timer once a second to handle leases. 1299 */ 1300 if (lasttime != time.tv_sec) { 1301 lasttime = time.tv_sec; 1302 nqnfs_serverd(); 1303 } 1304 splx(s); 1305 timeout(nfs_timer, (caddr_t)0, hz/NFS_HZ); 1306 } 1307 1308 /* 1309 * Test for a termination condition pending on the process. 1310 * This is used for NFSMNT_INT mounts. 1311 */ 1312 nfs_sigintr(nmp, rep, p) 1313 struct nfsmount *nmp; 1314 struct nfsreq *rep; 1315 register struct proc *p; 1316 { 1317 1318 if (rep && (rep->r_flags & R_SOFTTERM)) 1319 return (EINTR); 1320 if (!(nmp->nm_flag & NFSMNT_INT)) 1321 return (0); 1322 if (p && p->p_sig && (((p->p_sig &~ p->p_sigmask) &~ p->p_sigignore) & 1323 NFSINT_SIGMASK)) 1324 return (EINTR); 1325 return (0); 1326 } 1327 1328 /* 1329 * Lock a socket against others. 1330 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 1331 * and also to avoid race conditions between the processes with nfs requests 1332 * in progress when a reconnect is necessary. 1333 */ 1334 nfs_sndlock(flagp, rep) 1335 register int *flagp; 1336 struct nfsreq *rep; 1337 { 1338 struct proc *p; 1339 int slpflag = 0, slptimeo = 0; 1340 1341 if (rep) { 1342 p = rep->r_procp; 1343 if (rep->r_nmp->nm_flag & NFSMNT_INT) 1344 slpflag = PCATCH; 1345 } else 1346 p = (struct proc *)0; 1347 while (*flagp & NFSMNT_SNDLOCK) { 1348 if (nfs_sigintr(rep->r_nmp, rep, p)) 1349 return (EINTR); 1350 *flagp |= NFSMNT_WANTSND; 1351 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", 1352 slptimeo); 1353 if (slpflag == PCATCH) { 1354 slpflag = 0; 1355 slptimeo = 2 * hz; 1356 } 1357 } 1358 *flagp |= NFSMNT_SNDLOCK; 1359 return (0); 1360 } 1361 1362 /* 1363 * Unlock the stream socket for others. 1364 */ 1365 void 1366 nfs_sndunlock(flagp) 1367 register int *flagp; 1368 { 1369 1370 if ((*flagp & NFSMNT_SNDLOCK) == 0) 1371 panic("nfs sndunlock"); 1372 *flagp &= ~NFSMNT_SNDLOCK; 1373 if (*flagp & NFSMNT_WANTSND) { 1374 *flagp &= ~NFSMNT_WANTSND; 1375 wakeup((caddr_t)flagp); 1376 } 1377 } 1378 1379 nfs_rcvlock(rep) 1380 register struct nfsreq *rep; 1381 { 1382 register int *flagp = &rep->r_nmp->nm_flag; 1383 int slpflag, slptimeo = 0; 1384 1385 if (*flagp & NFSMNT_INT) 1386 slpflag = PCATCH; 1387 else 1388 slpflag = 0; 1389 while (*flagp & NFSMNT_RCVLOCK) { 1390 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) 1391 return (EINTR); 1392 *flagp |= NFSMNT_WANTRCV; 1393 (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", 1394 slptimeo); 1395 if (slpflag == PCATCH) { 1396 slpflag = 0; 1397 slptimeo = 2 * hz; 1398 } 1399 } 1400 *flagp |= NFSMNT_RCVLOCK; 1401 return (0); 1402 } 1403 1404 /* 1405 * Unlock the stream socket for others. 1406 */ 1407 void 1408 nfs_rcvunlock(flagp) 1409 register int *flagp; 1410 { 1411 1412 if ((*flagp & NFSMNT_RCVLOCK) == 0) 1413 panic("nfs rcvunlock"); 1414 *flagp &= ~NFSMNT_RCVLOCK; 1415 if (*flagp & NFSMNT_WANTRCV) { 1416 *flagp &= ~NFSMNT_WANTRCV; 1417 wakeup((caddr_t)flagp); 1418 } 1419 } 1420 1421 /* 1422 * Check for badly aligned mbuf data areas and 1423 * realign data in an mbuf list by copying the data areas up, as required. 1424 */ 1425 void 1426 nfs_realign(m, hsiz) 1427 register struct mbuf *m; 1428 int hsiz; 1429 { 1430 register struct mbuf *m2; 1431 register int siz, mlen, olen; 1432 register caddr_t tcp, fcp; 1433 struct mbuf *mnew; 1434 1435 while (m) { 1436 /* 1437 * This never happens for UDP, rarely happens for TCP 1438 * but frequently happens for iso transport. 1439 */ 1440 if ((m->m_len & 0x3) || (mtod(m, int) & 0x3)) { 1441 olen = m->m_len; 1442 fcp = mtod(m, caddr_t); 1443 m->m_flags &= ~M_PKTHDR; 1444 if (m->m_flags & M_EXT) 1445 m->m_data = m->m_ext.ext_buf; 1446 else 1447 m->m_data = m->m_dat; 1448 m->m_len = 0; 1449 tcp = mtod(m, caddr_t); 1450 mnew = m; 1451 m2 = m->m_next; 1452 1453 /* 1454 * If possible, only put the first invariant part 1455 * of the RPC header in the first mbuf. 1456 */ 1457 if (olen <= hsiz) 1458 mlen = hsiz; 1459 else 1460 mlen = M_TRAILINGSPACE(m); 1461 1462 /* 1463 * Loop through the mbuf list consolidating data. 1464 */ 1465 while (m) { 1466 while (olen > 0) { 1467 if (mlen == 0) { 1468 m2->m_flags &= ~M_PKTHDR; 1469 if (m2->m_flags & M_EXT) 1470 m2->m_data = m2->m_ext.ext_buf; 1471 else 1472 m2->m_data = m2->m_dat; 1473 m2->m_len = 0; 1474 mlen = M_TRAILINGSPACE(m2); 1475 tcp = mtod(m2, caddr_t); 1476 mnew = m2; 1477 m2 = m2->m_next; 1478 } 1479 siz = min(mlen, olen); 1480 if (tcp != fcp) 1481 bcopy(fcp, tcp, siz); 1482 mnew->m_len += siz; 1483 mlen -= siz; 1484 olen -= siz; 1485 tcp += siz; 1486 fcp += siz; 1487 } 1488 m = m->m_next; 1489 if (m) { 1490 olen = m->m_len; 1491 fcp = mtod(m, caddr_t); 1492 } 1493 } 1494 1495 /* 1496 * Finally, set m_len == 0 for any trailing mbufs that have 1497 * been copied out of. 1498 */ 1499 while (m2) { 1500 m2->m_len = 0; 1501 m2 = m2->m_next; 1502 } 1503 return; 1504 } 1505 m = m->m_next; 1506 } 1507 } 1508 1509 /* 1510 * Socket upcall routine for the nfsd sockets. 1511 * The caddr_t arg is a pointer to the "struct nfssvc_sock". 1512 * Essentially do as much as possible non-blocking, else punt and it will 1513 * be called with M_WAIT from an nfsd. 1514 */ 1515 void 1516 nfsrv_rcv(so, arg, waitflag) 1517 struct socket *so; 1518 caddr_t arg; 1519 int waitflag; 1520 { 1521 register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; 1522 register struct mbuf *m; 1523 struct mbuf *mp, *nam; 1524 struct uio auio; 1525 int flags, error; 1526 1527 if ((slp->ns_flag & SLP_VALID) == 0) 1528 return; 1529 #ifdef notdef 1530 /* 1531 * Define this to test for nfsds handling this under heavy load. 1532 */ 1533 if (waitflag == M_DONTWAIT) { 1534 slp->ns_flag |= SLP_NEEDQ; goto dorecs; 1535 } 1536 #endif 1537 auio.uio_procp = NULL; 1538 if (so->so_type == SOCK_STREAM) { 1539 /* 1540 * If there are already records on the queue, defer soreceive() 1541 * to an nfsd so that there is feedback to the TCP layer that 1542 * the nfs servers are heavily loaded. 1543 */ 1544 if (slp->ns_rec && waitflag == M_DONTWAIT) { 1545 slp->ns_flag |= SLP_NEEDQ; 1546 goto dorecs; 1547 } 1548 1549 /* 1550 * Do soreceive(). 1551 */ 1552 auio.uio_resid = 1000000000; 1553 flags = MSG_DONTWAIT; 1554 error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); 1555 if (error || mp == (struct mbuf *)0) { 1556 if (error == EWOULDBLOCK) 1557 slp->ns_flag |= SLP_NEEDQ; 1558 else 1559 slp->ns_flag |= SLP_DISCONN; 1560 goto dorecs; 1561 } 1562 m = mp; 1563 if (slp->ns_rawend) { 1564 slp->ns_rawend->m_next = m; 1565 slp->ns_cc += 1000000000 - auio.uio_resid; 1566 } else { 1567 slp->ns_raw = m; 1568 slp->ns_cc = 1000000000 - auio.uio_resid; 1569 } 1570 while (m->m_next) 1571 m = m->m_next; 1572 slp->ns_rawend = m; 1573 1574 /* 1575 * Now try and parse record(s) out of the raw stream data. 1576 */ 1577 if (error = nfsrv_getstream(slp, waitflag)) { 1578 if (error == EPERM) 1579 slp->ns_flag |= SLP_DISCONN; 1580 else 1581 slp->ns_flag |= SLP_NEEDQ; 1582 } 1583 } else { 1584 do { 1585 auio.uio_resid = 1000000000; 1586 flags = MSG_DONTWAIT; 1587 error = soreceive(so, &nam, &auio, &mp, 1588 (struct mbuf **)0, &flags); 1589 if (mp) { 1590 nfs_realign(mp, 10 * NFSX_UNSIGNED); 1591 if (nam) { 1592 m = nam; 1593 m->m_next = mp; 1594 } else 1595 m = mp; 1596 if (slp->ns_recend) 1597 slp->ns_recend->m_nextpkt = m; 1598 else 1599 slp->ns_rec = m; 1600 slp->ns_recend = m; 1601 m->m_nextpkt = (struct mbuf *)0; 1602 } 1603 if (error) { 1604 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) 1605 && error != EWOULDBLOCK) { 1606 slp->ns_flag |= SLP_DISCONN; 1607 goto dorecs; 1608 } 1609 } 1610 } while (mp); 1611 } 1612 1613 /* 1614 * Now try and process the request records, non-blocking. 1615 */ 1616 dorecs: 1617 if (waitflag == M_DONTWAIT && 1618 (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) 1619 nfsrv_wakenfsd(slp); 1620 } 1621 1622 /* 1623 * Try and extract an RPC request from the mbuf data list received on a 1624 * stream socket. The "waitflag" argument indicates whether or not it 1625 * can sleep. 1626 */ 1627 nfsrv_getstream(slp, waitflag) 1628 register struct nfssvc_sock *slp; 1629 int waitflag; 1630 { 1631 register struct mbuf *m; 1632 register char *cp1, *cp2; 1633 register int len; 1634 struct mbuf *om, *m2, *recm; 1635 u_long recmark; 1636 1637 if (slp->ns_flag & SLP_GETSTREAM) 1638 panic("nfs getstream"); 1639 slp->ns_flag |= SLP_GETSTREAM; 1640 for (;;) { 1641 if (slp->ns_reclen == 0) { 1642 if (slp->ns_cc < NFSX_UNSIGNED) { 1643 slp->ns_flag &= ~SLP_GETSTREAM; 1644 return (0); 1645 } 1646 m = slp->ns_raw; 1647 if (m->m_len >= NFSX_UNSIGNED) { 1648 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); 1649 m->m_data += NFSX_UNSIGNED; 1650 m->m_len -= NFSX_UNSIGNED; 1651 } else { 1652 cp1 = (caddr_t)&recmark; 1653 cp2 = mtod(m, caddr_t); 1654 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { 1655 while (m->m_len == 0) { 1656 m = m->m_next; 1657 cp2 = mtod(m, caddr_t); 1658 } 1659 *cp1++ = *cp2++; 1660 m->m_data++; 1661 m->m_len--; 1662 } 1663 } 1664 slp->ns_cc -= NFSX_UNSIGNED; 1665 slp->ns_reclen = ntohl(recmark) & ~0x80000000; 1666 if (slp->ns_reclen < NFS_MINPACKET || slp->ns_reclen > NFS_MAXPACKET) { 1667 slp->ns_flag &= ~SLP_GETSTREAM; 1668 return (EPERM); 1669 } 1670 } 1671 1672 /* 1673 * Now get the record part. 1674 */ 1675 if (slp->ns_cc == slp->ns_reclen) { 1676 recm = slp->ns_raw; 1677 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; 1678 slp->ns_cc = slp->ns_reclen = 0; 1679 } else if (slp->ns_cc > slp->ns_reclen) { 1680 len = 0; 1681 m = slp->ns_raw; 1682 om = (struct mbuf *)0; 1683 while (len < slp->ns_reclen) { 1684 if ((len + m->m_len) > slp->ns_reclen) { 1685 m2 = m_copym(m, 0, slp->ns_reclen - len, 1686 waitflag); 1687 if (m2) { 1688 if (om) { 1689 om->m_next = m2; 1690 recm = slp->ns_raw; 1691 } else 1692 recm = m2; 1693 m->m_data += slp->ns_reclen - len; 1694 m->m_len -= slp->ns_reclen - len; 1695 len = slp->ns_reclen; 1696 } else { 1697 slp->ns_flag &= ~SLP_GETSTREAM; 1698 return (EWOULDBLOCK); 1699 } 1700 } else if ((len + m->m_len) == slp->ns_reclen) { 1701 om = m; 1702 len += m->m_len; 1703 m = m->m_next; 1704 recm = slp->ns_raw; 1705 om->m_next = (struct mbuf *)0; 1706 } else { 1707 om = m; 1708 len += m->m_len; 1709 m = m->m_next; 1710 } 1711 } 1712 slp->ns_raw = m; 1713 slp->ns_cc -= len; 1714 slp->ns_reclen = 0; 1715 } else { 1716 slp->ns_flag &= ~SLP_GETSTREAM; 1717 return (0); 1718 } 1719 nfs_realign(recm, 10 * NFSX_UNSIGNED); 1720 if (slp->ns_recend) 1721 slp->ns_recend->m_nextpkt = recm; 1722 else 1723 slp->ns_rec = recm; 1724 slp->ns_recend = recm; 1725 } 1726 } 1727 1728 /* 1729 * Parse an RPC header. 1730 */ 1731 nfsrv_dorec(slp, nd) 1732 register struct nfssvc_sock *slp; 1733 register struct nfsd *nd; 1734 { 1735 register struct mbuf *m; 1736 int error; 1737 1738 if ((slp->ns_flag & SLP_VALID) == 0 || 1739 (m = slp->ns_rec) == (struct mbuf *)0) 1740 return (ENOBUFS); 1741 if (slp->ns_rec = m->m_nextpkt) 1742 m->m_nextpkt = (struct mbuf *)0; 1743 else 1744 slp->ns_recend = (struct mbuf *)0; 1745 if (m->m_type == MT_SONAME) { 1746 nd->nd_nam = m; 1747 nd->nd_md = nd->nd_mrep = m->m_next; 1748 m->m_next = (struct mbuf *)0; 1749 } else { 1750 nd->nd_nam = (struct mbuf *)0; 1751 nd->nd_md = nd->nd_mrep = m; 1752 } 1753 nd->nd_dpos = mtod(nd->nd_md, caddr_t); 1754 if (error = nfs_getreq(nd, TRUE)) { 1755 m_freem(nd->nd_nam); 1756 return (error); 1757 } 1758 return (0); 1759 } 1760 1761 /* 1762 * Parse an RPC request 1763 * - verify it 1764 * - fill in the cred struct. 1765 */ 1766 nfs_getreq(nd, has_header) 1767 register struct nfsd *nd; 1768 int has_header; 1769 { 1770 register int len, i; 1771 register u_long *tl; 1772 register long t1; 1773 struct uio uio; 1774 struct iovec iov; 1775 caddr_t dpos, cp2; 1776 u_long nfsvers, auth_type; 1777 int error = 0, nqnfs = 0; 1778 struct mbuf *mrep, *md; 1779 1780 mrep = nd->nd_mrep; 1781 md = nd->nd_md; 1782 dpos = nd->nd_dpos; 1783 if (has_header) { 1784 nfsm_dissect(tl, u_long *, 10*NFSX_UNSIGNED); 1785 nd->nd_retxid = *tl++; 1786 if (*tl++ != rpc_call) { 1787 m_freem(mrep); 1788 return (EBADRPC); 1789 } 1790 } else { 1791 nfsm_dissect(tl, u_long *, 8*NFSX_UNSIGNED); 1792 } 1793 nd->nd_repstat = 0; 1794 if (*tl++ != rpc_vers) { 1795 nd->nd_repstat = ERPCMISMATCH; 1796 nd->nd_procnum = NFSPROC_NOOP; 1797 return (0); 1798 } 1799 nfsvers = nfs_vers; 1800 if (*tl != nfs_prog) { 1801 if (*tl == nqnfs_prog) { 1802 nqnfs++; 1803 nfsvers = nqnfs_vers; 1804 } else { 1805 nd->nd_repstat = EPROGUNAVAIL; 1806 nd->nd_procnum = NFSPROC_NOOP; 1807 return (0); 1808 } 1809 } 1810 tl++; 1811 if (*tl++ != nfsvers) { 1812 nd->nd_repstat = EPROGMISMATCH; 1813 nd->nd_procnum = NFSPROC_NOOP; 1814 return (0); 1815 } 1816 nd->nd_procnum = fxdr_unsigned(u_long, *tl++); 1817 if (nd->nd_procnum == NFSPROC_NULL) 1818 return (0); 1819 if (nd->nd_procnum >= NFS_NPROCS || 1820 (!nqnfs && nd->nd_procnum > NFSPROC_STATFS) || 1821 (*tl != rpc_auth_unix && *tl != rpc_auth_kerb)) { 1822 nd->nd_repstat = EPROCUNAVAIL; 1823 nd->nd_procnum = NFSPROC_NOOP; 1824 return (0); 1825 } 1826 auth_type = *tl++; 1827 len = fxdr_unsigned(int, *tl++); 1828 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1829 m_freem(mrep); 1830 return (EBADRPC); 1831 } 1832 1833 /* 1834 * Handle auth_unix or auth_kerb. 1835 */ 1836 if (auth_type == rpc_auth_unix) { 1837 len = fxdr_unsigned(int, *++tl); 1838 if (len < 0 || len > NFS_MAXNAMLEN) { 1839 m_freem(mrep); 1840 return (EBADRPC); 1841 } 1842 nfsm_adv(nfsm_rndup(len)); 1843 nfsm_dissect(tl, u_long *, 3*NFSX_UNSIGNED); 1844 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1845 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); 1846 len = fxdr_unsigned(int, *tl); 1847 if (len < 0 || len > RPCAUTH_UNIXGIDS) { 1848 m_freem(mrep); 1849 return (EBADRPC); 1850 } 1851 nfsm_dissect(tl, u_long *, (len + 2)*NFSX_UNSIGNED); 1852 for (i = 1; i <= len; i++) 1853 if (i < NGROUPS) 1854 nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); 1855 else 1856 tl++; 1857 nd->nd_cr.cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1); 1858 } else if (auth_type == rpc_auth_kerb) { 1859 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); 1860 nd->nd_authlen = fxdr_unsigned(int, *tl); 1861 uio.uio_resid = nfsm_rndup(nd->nd_authlen); 1862 if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { 1863 m_freem(mrep); 1864 return (EBADRPC); 1865 } 1866 uio.uio_offset = 0; 1867 uio.uio_iov = &iov; 1868 uio.uio_iovcnt = 1; 1869 uio.uio_segflg = UIO_SYSSPACE; 1870 iov.iov_base = (caddr_t)nd->nd_authstr; 1871 iov.iov_len = RPCAUTH_MAXSIZ; 1872 nfsm_mtouio(&uio, uio.uio_resid); 1873 nfsm_dissect(tl, u_long *, 2 * NFSX_UNSIGNED); 1874 nd->nd_flag |= NFSD_NEEDAUTH; 1875 } 1876 1877 /* 1878 * Do we have any use for the verifier. 1879 * According to the "Remote Procedure Call Protocol Spec." it 1880 * should be AUTH_NULL, but some clients make it AUTH_UNIX? 1881 * For now, just skip over it 1882 */ 1883 len = fxdr_unsigned(int, *++tl); 1884 if (len < 0 || len > RPCAUTH_MAXSIZ) { 1885 m_freem(mrep); 1886 return (EBADRPC); 1887 } 1888 if (len > 0) { 1889 nfsm_adv(nfsm_rndup(len)); 1890 } 1891 1892 /* 1893 * For nqnfs, get piggybacked lease request. 1894 */ 1895 if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { 1896 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1897 nd->nd_nqlflag = fxdr_unsigned(int, *tl); 1898 if (nd->nd_nqlflag) { 1899 nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); 1900 nd->nd_duration = fxdr_unsigned(int, *tl); 1901 } else 1902 nd->nd_duration = NQ_MINLEASE; 1903 } else { 1904 nd->nd_nqlflag = NQL_NOVAL; 1905 nd->nd_duration = NQ_MINLEASE; 1906 } 1907 nd->nd_md = md; 1908 nd->nd_dpos = dpos; 1909 return (0); 1910 nfsmout: 1911 return (error); 1912 } 1913 1914 /* 1915 * Search for a sleeping nfsd and wake it up. 1916 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the 1917 * running nfsds will go look for the work in the nfssvc_sock list. 1918 */ 1919 void 1920 nfsrv_wakenfsd(slp) 1921 struct nfssvc_sock *slp; 1922 { 1923 register struct nfsd *nd = nfsd_head.nd_next; 1924 1925 if ((slp->ns_flag & SLP_VALID) == 0) 1926 return; 1927 while (nd != (struct nfsd *)&nfsd_head) { 1928 if (nd->nd_flag & NFSD_WAITING) { 1929 nd->nd_flag &= ~NFSD_WAITING; 1930 if (nd->nd_slp) 1931 panic("nfsd wakeup"); 1932 slp->ns_sref++; 1933 nd->nd_slp = slp; 1934 wakeup((caddr_t)nd); 1935 return; 1936 } 1937 nd = nd->nd_next; 1938 } 1939 slp->ns_flag |= SLP_DOREC; 1940 nfsd_head.nd_flag |= NFSD_CHECKSLP; 1941 } 1942 1943 nfs_msg(p, server, msg) 1944 struct proc *p; 1945 char *server, *msg; 1946 { 1947 tpr_t tpr; 1948 1949 if (p) 1950 tpr = tprintf_open(p); 1951 else 1952 tpr = NULL; 1953 tprintf(tpr, "nfs server %s: %s\n", server, msg); 1954 tprintf_close(tpr); 1955 } 1956