sys/vfs/nfs/nfs_socket.c

   1 /*
   2  * Copyright (c) 1989, 1991, 1993, 1995
   3  *      The Regents of the University of California.  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  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by the University of
  19  *      California, Berkeley and its contributors.
  20  * 4. Neither the name of the University nor the names of its contributors
  21  *    may be used to endorse or promote products derived from this software
  22  *    without specific prior written permission.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  *
  36  *      @(#)nfs_socket.c        8.5 (Berkeley) 3/30/95
  37  * $FreeBSD: src/sys/nfs/nfs_socket.c,v 1.60.2.6 2003/03/26 01:44:46 alfred Exp $
  38  */
  39
  40 /*
  41  * Socket operations for use by nfs
  42  */
  43
  44 #include <sys/param.h>
  45 #include <sys/systm.h>
  46 #include <sys/proc.h>
  47 #include <sys/malloc.h>
  48 #include <sys/mount.h>
  49 #include <sys/kernel.h>
  50 #include <sys/mbuf.h>
  51 #include <sys/vnode.h>
  52 #include <sys/protosw.h>
  53 #include <sys/socket.h>
  54 #include <sys/socketvar.h>
  55 #include <sys/syslog.h>
  56 #include <sys/tprintf.h>
  57 #include <sys/sysctl.h>
  58 #include <sys/signalvar.h>
  59
  60 #include <netinet/in.h>
  61 #include <netinet/tcp.h>
  62
  63 #include <nfs/rpcv2.h>
  64 #include <nfs/nfsproto.h>
  65 #include <nfs/nfs.h>
  66 #include <nfs/xdr_subs.h>
  67 #include <nfs/nfsm_subs.h>
  68 #include <nfs/nfsmount.h>
  69 #include <nfs/nfsnode.h>
  70 #include <nfs/nfsrtt.h>
  71 #include <nfs/nqnfs.h>
  72
  73 #define TRUE    1
  74 #define FALSE   0
  75
  76 /*
  77  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
  78  * Use the mean and mean deviation of rtt for the appropriate type of rpc
  79  * for the frequent rpcs and a default for the others.
  80  * The justification for doing "other" this way is that these rpcs
  81  * happen so infrequently that timer est. would probably be stale.
  82  * Also, since many of these rpcs are
  83  * non-idempotent, a conservative timeout is desired.
  84  * getattr, lookup - A+2D
  85  * read, write     - A+4D
  86  * other           - nm_timeo
  87  */
  88 #define NFS_RTO(n, t) \
  89         ((t) == 0 ? (n)->nm_timeo : \
  90          ((t) < 3 ? \
  91           (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
  92           ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
  93 #define NFS_SRTT(r)     (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
  94 #define NFS_SDRTT(r)    (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
  95 /*
  96  * External data, mostly RPC constants in XDR form
  97  */
  98 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
  99         rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
 100         rpc_auth_kerb;
 101 extern u_int32_t nfs_prog, nqnfs_prog;
 102 extern time_t nqnfsstarttime;
 103 extern struct nfsstats nfsstats;
 104 extern int nfsv3_procid[NFS_NPROCS];
 105 extern int nfs_ticks;
 106
 107 /*
 108  * Defines which timer to use for the procnum.
 109  * 0 - default
 110  * 1 - getattr
 111  * 2 - lookup
 112  * 3 - read
 113  * 4 - write
 114  */
 115 static int proct[NFS_NPROCS] = {
 116         0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
 117         0, 0, 0,
 118 };
 119
 120 static int nfs_realign_test;
 121 static int nfs_realign_count;
 122 static int nfs_bufpackets = 4;
 123
 124 SYSCTL_DECL(_vfs_nfs);
 125
 126 SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_test, CTLFLAG_RW, &nfs_realign_test, 0, "");
 127 SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_count, CTLFLAG_RW, &nfs_realign_count, 0, "");
 128 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0, "");
 129
 130
 131 /*
 132  * There is a congestion window for outstanding rpcs maintained per mount
 133  * point. The cwnd size is adjusted in roughly the way that:
 134  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
 135  * SIGCOMM '88". ACM, August 1988.
 136  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
 137  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
 138  * of rpcs is in progress.
 139  * (The sent count and cwnd are scaled for integer arith.)
 140  * Variants of "slow start" were tried and were found to be too much of a
 141  * performance hit (ave. rtt 3 times larger),
 142  * I suspect due to the large rtt that nfs rpcs have.
 143  */
 144 #define NFS_CWNDSCALE   256
 145 #define NFS_MAXCWND     (NFS_CWNDSCALE * 32)
 146 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
 147 int nfsrtton = 0;
 148 struct nfsrtt nfsrtt;
 149 struct callout_handle   nfs_timer_handle;
 150
 151 static int      nfs_msg __P((struct proc *,char *,char *));
 152 static int      nfs_rcvlock __P((struct nfsreq *));
 153 static void     nfs_rcvunlock __P((struct nfsreq *));
 154 static void     nfs_realign __P((struct mbuf **pm, int hsiz));
 155 static int      nfs_receive __P((struct nfsreq *rep, struct sockaddr **aname,
 156                                  struct mbuf **mp));
 157 static void     nfs_softterm __P((struct nfsreq *rep));
 158 static int      nfs_reconnect __P((struct nfsreq *rep));
 159 #ifndef NFS_NOSERVER
 160 static int      nfsrv_getstream __P((struct nfssvc_sock *,int));
 161
 162 int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *nd,
 163                                     struct nfssvc_sock *slp,
 164                                     struct proc *procp,
 165                                     struct mbuf **mreqp)) = {
 166         nfsrv_null,
 167         nfsrv_getattr,
 168         nfsrv_setattr,
 169         nfsrv_lookup,
 170         nfsrv3_access,
 171         nfsrv_readlink,
 172         nfsrv_read,
 173         nfsrv_write,
 174         nfsrv_create,
 175         nfsrv_mkdir,
 176         nfsrv_symlink,
 177         nfsrv_mknod,
 178         nfsrv_remove,
 179         nfsrv_rmdir,
 180         nfsrv_rename,
 181         nfsrv_link,
 182         nfsrv_readdir,
 183         nfsrv_readdirplus,
 184         nfsrv_statfs,
 185         nfsrv_fsinfo,
 186         nfsrv_pathconf,
 187         nfsrv_commit,
 188         nqnfsrv_getlease,
 189         nqnfsrv_vacated,
 190         nfsrv_noop,
 191         nfsrv_noop
 192 };
 193 #endif /* NFS_NOSERVER */
 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(nmp, rep)
 201         register struct nfsmount *nmp;
 202         struct nfsreq *rep;
 203 {
 204         register struct socket *so;
 205         int s, error, rcvreserve, sndreserve;
 206         int pktscale;
 207         struct sockaddr *saddr;
 208         struct sockaddr_in *sin;
 209         struct proc *p = &proc0; /* only used for socreate and sobind */
 210
 211         nmp->nm_so = (struct socket *)0;
 212         saddr = nmp->nm_nam;
 213         error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
 214                 nmp->nm_soproto, p);
 215         if (error)
 216                 goto bad;
 217         so = nmp->nm_so;
 218         nmp->nm_soflags = so->so_proto->pr_flags;
 219
 220         /*
 221          * Some servers require that the client port be a reserved port number.
 222          */
 223         if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
 224                 struct sockopt sopt;
 225                 int ip;
 226                 struct sockaddr_in ssin;
 227
 228                 bzero(&sopt, sizeof sopt);
 229                 ip = IP_PORTRANGE_LOW;
 230                 sopt.sopt_dir = SOPT_SET;
 231                 sopt.sopt_level = IPPROTO_IP;
 232                 sopt.sopt_name = IP_PORTRANGE;
 233                 sopt.sopt_val = (void *)&ip;
 234                 sopt.sopt_valsize = sizeof(ip);
 235                 sopt.sopt_p = NULL;
 236                 error = sosetopt(so, &sopt);
 237                 if (error)
 238                         goto bad;
 239                 bzero(&ssin, sizeof ssin);
 240                 sin = &ssin;
 241                 sin->sin_len = sizeof (struct sockaddr_in);
 242                 sin->sin_family = AF_INET;
 243                 sin->sin_addr.s_addr = INADDR_ANY;
 244                 sin->sin_port = htons(0);
 245                 error = sobind(so, (struct sockaddr *)sin, p);
 246                 if (error)
 247                         goto bad;
 248                 bzero(&sopt, sizeof sopt);
 249                 ip = IP_PORTRANGE_DEFAULT;
 250                 sopt.sopt_dir = SOPT_SET;
 251                 sopt.sopt_level = IPPROTO_IP;
 252                 sopt.sopt_name = IP_PORTRANGE;
 253                 sopt.sopt_val = (void *)&ip;
 254                 sopt.sopt_valsize = sizeof(ip);
 255                 sopt.sopt_p = NULL;
 256                 error = sosetopt(so, &sopt);
 257                 if (error)
 258                         goto bad;
 259         }
 260
 261         /*
 262          * Protocols that do not require connections may be optionally left
 263          * unconnected for servers that reply from a port other than NFS_PORT.
 264          */
 265         if (nmp->nm_flag & NFSMNT_NOCONN) {
 266                 if (nmp->nm_soflags & PR_CONNREQUIRED) {
 267                         error = ENOTCONN;
 268                         goto bad;
 269                 }
 270         } else {
 271                 error = soconnect(so, nmp->nm_nam, p);
 272                 if (error)
 273                         goto bad;
 274
 275                 /*
 276                  * Wait for the connection to complete. Cribbed from the
 277                  * connect system call but with the wait timing out so
 278                  * that interruptible mounts don't hang here for a long time.
 279                  */
 280                 s = splnet();
 281                 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
 282                         (void) tsleep((caddr_t)&so->so_timeo, PSOCK,
 283                                 "nfscon", 2 * hz);
 284                         if ((so->so_state & SS_ISCONNECTING) &&
 285                             so->so_error == 0 && rep &&
 286                             (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){
 287                                 so->so_state &= ~SS_ISCONNECTING;
 288                                 splx(s);
 289                                 goto bad;
 290                         }
 291                 }
 292                 if (so->so_error) {
 293                         error = so->so_error;
 294                         so->so_error = 0;
 295                         splx(s);
 296                         goto bad;
 297                 }
 298                 splx(s);
 299         }
 300         so->so_rcv.sb_timeo = (5 * hz);
 301         so->so_snd.sb_timeo = (5 * hz);
 302
 303         /*
 304          * Get buffer reservation size from sysctl, but impose reasonable
 305          * limits.
 306          */
 307         pktscale = nfs_bufpackets;
 308         if (pktscale < 2)
 309                 pktscale = 2;
 310         if (pktscale > 64)
 311                 pktscale = 64;
 312
 313         if (nmp->nm_sotype == SOCK_DGRAM) {
 314                 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
 315                 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
 316                     NFS_MAXPKTHDR) * pktscale;
 317         } else if (nmp->nm_sotype == SOCK_SEQPACKET) {
 318                 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
 319                 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
 320                     NFS_MAXPKTHDR) * pktscale;
 321         } else {
 322                 if (nmp->nm_sotype != SOCK_STREAM)
 323                         panic("nfscon sotype");
 324                 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
 325                         struct sockopt sopt;
 326                         int val;
 327
 328                         bzero(&sopt, sizeof sopt);
 329                         sopt.sopt_level = SOL_SOCKET;
 330                         sopt.sopt_name = SO_KEEPALIVE;
 331                         sopt.sopt_val = &val;
 332                         sopt.sopt_valsize = sizeof val;
 333                         val = 1;
 334                         sosetopt(so, &sopt);
 335                 }
 336                 if (so->so_proto->pr_protocol == IPPROTO_TCP) {
 337                         struct sockopt sopt;
 338                         int val;
 339
 340                         bzero(&sopt, sizeof sopt);
 341                         sopt.sopt_level = IPPROTO_TCP;
 342                         sopt.sopt_name = TCP_NODELAY;
 343                         sopt.sopt_val = &val;
 344                         sopt.sopt_valsize = sizeof val;
 345                         val = 1;
 346                         sosetopt(so, &sopt);
 347                 }
 348                 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
 349                     sizeof (u_int32_t)) * pktscale;
 350                 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
 351                     sizeof (u_int32_t)) * pktscale;
 352         }
 353         error = soreserve(so, sndreserve, rcvreserve);
 354         if (error)
 355                 goto bad;
 356         so->so_rcv.sb_flags |= SB_NOINTR;
 357         so->so_snd.sb_flags |= SB_NOINTR;
 358
 359         /* Initialize other non-zero congestion variables */
 360         nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] =
 361                 nmp->nm_srtt[3] = (NFS_TIMEO << 3);
 362         nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
 363                 nmp->nm_sdrtt[3] = 0;
 364         nmp->nm_cwnd = NFS_MAXCWND / 2;     /* Initial send window */
 365         nmp->nm_sent = 0;
 366         nmp->nm_timeouts = 0;
 367         return (0);
 368
 369 bad:
 370         nfs_disconnect(nmp);
 371         return (error);
 372 }
 373
 374 /*
 375  * Reconnect routine:
 376  * Called when a connection is broken on a reliable protocol.
 377  * - clean up the old socket
 378  * - nfs_connect() again
 379  * - set R_MUSTRESEND for all outstanding requests on mount point
 380  * If this fails the mount point is DEAD!
 381  * nb: Must be called with the nfs_sndlock() set on the mount point.
 382  */
 383 static int
 384 nfs_reconnect(rep)
 385         register struct nfsreq *rep;
 386 {
 387         register struct nfsreq *rp;
 388         register struct nfsmount *nmp = rep->r_nmp;
 389         int error;
 390
 391         nfs_disconnect(nmp);
 392         while ((error = nfs_connect(nmp, rep)) != 0) {
 393                 if (error == EINTR || error == ERESTART)
 394                         return (EINTR);
 395                 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0);
 396         }
 397
 398         /*
 399          * Loop through outstanding request list and fix up all requests
 400          * on old socket.
 401          */
 402         for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) {
 403                 if (rp->r_nmp == nmp)
 404                         rp->r_flags |= R_MUSTRESEND;
 405         }
 406         return (0);
 407 }
 408
 409 /*
 410  * NFS disconnect. Clean up and unlink.
 411  */
 412 void
 413 nfs_disconnect(nmp)
 414         register struct nfsmount *nmp;
 415 {
 416         register struct socket *so;
 417
 418         if (nmp->nm_so) {
 419                 so = nmp->nm_so;
 420                 nmp->nm_so = (struct socket *)0;
 421                 soshutdown(so, 2);
 422                 soclose(so);
 423         }
 424 }
 425
 426 void
 427 nfs_safedisconnect(nmp)
 428         struct nfsmount *nmp;
 429 {
 430         struct nfsreq dummyreq;
 431
 432         bzero(&dummyreq, sizeof(dummyreq));
 433         dummyreq.r_nmp = nmp;
 434         nfs_rcvlock(&dummyreq);
 435         nfs_disconnect(nmp);
 436         nfs_rcvunlock(&dummyreq);
 437 }
 438
 439 /*
 440  * This is the nfs send routine. For connection based socket types, it
 441  * must be called with an nfs_sndlock() on the socket.
 442  * "rep == NULL" indicates that it has been called from a server.
 443  * For the client side:
 444  * - return EINTR if the RPC is terminated, 0 otherwise
 445  * - set R_MUSTRESEND if the send fails for any reason
 446  * - do any cleanup required by recoverable socket errors (?)
 447  * For the server side:
 448  * - return EINTR or ERESTART if interrupted by a signal
 449  * - return EPIPE if a connection is lost for connection based sockets (TCP...)
 450  * - do any cleanup required by recoverable socket errors (?)
 451  */
 452 int
 453 nfs_send(so, nam, top, rep)
 454         register struct socket *so;
 455         struct sockaddr *nam;
 456         register struct mbuf *top;
 457         struct nfsreq *rep;
 458 {
 459         struct sockaddr *sendnam;
 460         int error, soflags, flags;
 461
 462         if (rep) {
 463                 if (rep->r_flags & R_SOFTTERM) {
 464                         m_freem(top);
 465                         return (EINTR);
 466                 }
 467                 if ((so = rep->r_nmp->nm_so) == NULL) {
 468                         rep->r_flags |= R_MUSTRESEND;
 469                         m_freem(top);
 470                         return (0);
 471                 }
 472                 rep->r_flags &= ~R_MUSTRESEND;
 473                 soflags = rep->r_nmp->nm_soflags;
 474         } else
 475                 soflags = so->so_proto->pr_flags;
 476         if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
 477                 sendnam = (struct sockaddr *)0;
 478         else
 479                 sendnam = nam;
 480         if (so->so_type == SOCK_SEQPACKET)
 481                 flags = MSG_EOR;
 482         else
 483                 flags = 0;
 484
 485         error = so->so_proto->pr_usrreqs->pru_sosend(so, sendnam, 0, top, 0,
 486                                                      flags, curproc /*XXX*/);
 487         /*
 488          * ENOBUFS for dgram sockets is transient and non fatal.
 489          * No need to log, and no need to break a soft mount.
 490          */
 491         if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
 492                 error = 0;
 493                 if (rep)                /* do backoff retransmit on client */
 494                         rep->r_flags |= R_MUSTRESEND;
 495         }
 496
 497         if (error) {
 498                 if (rep) {
 499                         log(LOG_INFO, "nfs send error %d for server %s\n",error,
 500                             rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
 501                         /*
 502                          * Deal with errors for the client side.
 503                          */
 504                         if (rep->r_flags & R_SOFTTERM)
 505                                 error = EINTR;
 506                         else
 507                                 rep->r_flags |= R_MUSTRESEND;
 508                 } else
 509                         log(LOG_INFO, "nfsd send error %d\n", error);
 510
 511                 /*
 512                  * Handle any recoverable (soft) socket errors here. (?)
 513                  */
 514                 if (error != EINTR && error != ERESTART &&
 515                         error != EWOULDBLOCK && error != EPIPE)
 516                         error = 0;
 517         }
 518         return (error);
 519 }
 520
 521 /*
 522  * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
 523  * done by soreceive(), but for SOCK_STREAM we must deal with the Record
 524  * Mark and consolidate the data into a new mbuf list.
 525  * nb: Sometimes TCP passes the data up to soreceive() in long lists of
 526  *     small mbufs.
 527  * For SOCK_STREAM we must be very careful to read an entire record once
 528  * we have read any of it, even if the system call has been interrupted.
 529  */
 530 static int
 531 nfs_receive(rep, aname, mp)
 532         register struct nfsreq *rep;
 533         struct sockaddr **aname;
 534         struct mbuf **mp;
 535 {
 536         register struct socket *so;
 537         struct uio auio;
 538         struct iovec aio;
 539         register struct mbuf *m;
 540         struct mbuf *control;
 541         u_int32_t len;
 542         struct sockaddr **getnam;
 543         int error, sotype, rcvflg;
 544         struct proc *p = curproc;       /* XXX */
 545
 546         /*
 547          * Set up arguments for soreceive()
 548          */
 549         *mp = (struct mbuf *)0;
 550         *aname = (struct sockaddr *)0;
 551         sotype = rep->r_nmp->nm_sotype;
 552
 553         /*
 554          * For reliable protocols, lock against other senders/receivers
 555          * in case a reconnect is necessary.
 556          * For SOCK_STREAM, first get the Record Mark to find out how much
 557          * more there is to get.
 558          * We must lock the socket against other receivers
 559          * until we have an entire rpc request/reply.
 560          */
 561         if (sotype != SOCK_DGRAM) {
 562                 error = nfs_sndlock(rep);
 563                 if (error)
 564                         return (error);
 565 tryagain:
 566                 /*
 567                  * Check for fatal errors and resending request.
 568                  */
 569                 /*
 570                  * Ugh: If a reconnect attempt just happened, nm_so
 571                  * would have changed. NULL indicates a failed
 572                  * attempt that has essentially shut down this
 573                  * mount point.
 574                  */
 575                 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
 576                         nfs_sndunlock(rep);
 577                         return (EINTR);
 578                 }
 579                 so = rep->r_nmp->nm_so;
 580                 if (!so) {
 581                         error = nfs_reconnect(rep);
 582                         if (error) {
 583                                 nfs_sndunlock(rep);
 584                                 return (error);
 585                         }
 586                         goto tryagain;
 587                 }
 588                 while (rep->r_flags & R_MUSTRESEND) {
 589                         m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
 590                         nfsstats.rpcretries++;
 591                         error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
 592                         if (error) {
 593                                 if (error == EINTR || error == ERESTART ||
 594                                     (error = nfs_reconnect(rep)) != 0) {
 595                                         nfs_sndunlock(rep);
 596                                         return (error);
 597                                 }
 598                                 goto tryagain;
 599                         }
 600                 }
 601                 nfs_sndunlock(rep);
 602                 if (sotype == SOCK_STREAM) {
 603                         aio.iov_base = (caddr_t) &len;
 604                         aio.iov_len = sizeof(u_int32_t);
 605                         auio.uio_iov = &aio;
 606                         auio.uio_iovcnt = 1;
 607                         auio.uio_segflg = UIO_SYSSPACE;
 608                         auio.uio_rw = UIO_READ;
 609                         auio.uio_offset = 0;
 610                         auio.uio_resid = sizeof(u_int32_t);
 611                         auio.uio_procp = p;
 612                         do {
 613                            rcvflg = MSG_WAITALL;
 614                            error = so->so_proto->pr_usrreqs->pru_soreceive
 615                                    (so, (struct sockaddr **)0, &auio,
 616                                     (struct mbuf **)0, (struct mbuf **)0,
 617                                     &rcvflg);
 618                            if (error == EWOULDBLOCK && rep) {
 619                                 if (rep->r_flags & R_SOFTTERM)
 620                                         return (EINTR);
 621                            }
 622                         } while (error == EWOULDBLOCK);
 623                         if (!error && auio.uio_resid > 0) {
 624                             /*
 625                              * Don't log a 0 byte receive; it means
 626                              * that the socket has been closed, and
 627                              * can happen during normal operation
 628                              * (forcible unmount or Solaris server).
 629                              */
 630                             if (auio.uio_resid != sizeof (u_int32_t))
 631                             log(LOG_INFO,
 632                                  "short receive (%d/%d) from nfs server %s\n",
 633                                  (int)(sizeof(u_int32_t) - auio.uio_resid),
 634                                  (int)sizeof(u_int32_t),
 635                                  rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
 636                             error = EPIPE;
 637                         }
 638                         if (error)
 639                                 goto errout;
 640                         len = ntohl(len) & ~0x80000000;
 641                         /*
 642                          * This is SERIOUS! We are out of sync with the sender
 643                          * and forcing a disconnect/reconnect is all I can do.
 644                          */
 645                         if (len > NFS_MAXPACKET) {
 646                             log(LOG_ERR, "%s (%d) from nfs server %s\n",
 647                                 "impossible packet length",
 648                                 len,
 649                                 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
 650                             error = EFBIG;
 651                             goto errout;
 652                         }
 653                         auio.uio_resid = len;
 654                         do {
 655                             rcvflg = MSG_WAITALL;
 656                             error =  so->so_proto->pr_usrreqs->pru_soreceive
 657                                     (so, (struct sockaddr **)0,
 658                                      &auio, mp, (struct mbuf **)0, &rcvflg);
 659                         } while (error == EWOULDBLOCK || error == EINTR ||
 660                                  error == ERESTART);
 661                         if (!error && auio.uio_resid > 0) {
 662                             if (len != auio.uio_resid)
 663                             log(LOG_INFO,
 664                                 "short receive (%d/%d) from nfs server %s\n",
 665                                 len - auio.uio_resid, len,
 666                                 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
 667                             error = EPIPE;
 668                         }
 669                 } else {
 670                         /*
 671                          * NB: Since uio_resid is big, MSG_WAITALL is ignored
 672                          * and soreceive() will return when it has either a
 673                          * control msg or a data msg.
 674                          * We have no use for control msg., but must grab them
 675                          * and then throw them away so we know what is going
 676                          * on.
 677                          */
 678                         auio.uio_resid = len = 100000000; /* Anything Big */
 679                         auio.uio_procp = p;
 680                         do {
 681                             rcvflg = 0;
 682                             error =  so->so_proto->pr_usrreqs->pru_soreceive
 683                                     (so, (struct sockaddr **)0,
 684                                 &auio, mp, &control, &rcvflg);
 685                             if (control)
 686                                 m_freem(control);
 687                             if (error == EWOULDBLOCK && rep) {
 688                                 if (rep->r_flags & R_SOFTTERM)
 689                                         return (EINTR);
 690                             }
 691                         } while (error == EWOULDBLOCK ||
 692                                  (!error && *mp == NULL && control));
 693                         if ((rcvflg & MSG_EOR) == 0)
 694                                 printf("Egad!!\n");
 695                         if (!error && *mp == NULL)
 696                                 error = EPIPE;
 697                         len -= auio.uio_resid;
 698                 }
 699 errout:
 700                 if (error && error != EINTR && error != ERESTART) {
 701                         m_freem(*mp);
 702                         *mp = (struct mbuf *)0;
 703                         if (error != EPIPE)
 704                                 log(LOG_INFO,
 705                                     "receive error %d from nfs server %s\n",
 706                                     error,
 707                                  rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
 708                         error = nfs_sndlock(rep);
 709                         if (!error) {
 710                                 error = nfs_reconnect(rep);
 711                                 if (!error)
 712                                         goto tryagain;
 713                                 else
 714                                         nfs_sndunlock(rep);
 715                         }
 716                 }
 717         } else {
 718                 if ((so = rep->r_nmp->nm_so) == NULL)
 719                         return (EACCES);
 720                 if (so->so_state & SS_ISCONNECTED)
 721                         getnam = (struct sockaddr **)0;
 722                 else
 723                         getnam = aname;
 724                 auio.uio_resid = len = 1000000;
 725                 auio.uio_procp = p;
 726                 do {
 727                         rcvflg = 0;
 728                         error =  so->so_proto->pr_usrreqs->pru_soreceive
 729                                 (so, getnam, &auio, mp,
 730                                 (struct mbuf **)0, &rcvflg);
 731                         if (error == EWOULDBLOCK &&
 732                             (rep->r_flags & R_SOFTTERM))
 733                                 return (EINTR);
 734                 } while (error == EWOULDBLOCK);
 735                 len -= auio.uio_resid;
 736         }
 737         if (error) {
 738                 m_freem(*mp);
 739                 *mp = (struct mbuf *)0;
 740         }
 741         /*
 742          * Search for any mbufs that are not a multiple of 4 bytes long
 743          * or with m_data not longword aligned.
 744          * These could cause pointer alignment problems, so copy them to
 745          * well aligned mbufs.
 746          */
 747         nfs_realign(mp, 5 * NFSX_UNSIGNED);
 748         return (error);
 749 }
 750
 751 /*
 752  * Implement receipt of reply on a socket.
 753  * We must search through the list of received datagrams matching them
 754  * with outstanding requests using the xid, until ours is found.
 755  */
 756 /* ARGSUSED */
 757 int
 758 nfs_reply(myrep)
 759         struct nfsreq *myrep;
 760 {
 761         register struct nfsreq *rep;
 762         register struct nfsmount *nmp = myrep->r_nmp;
 763         register int32_t t1;
 764         struct mbuf *mrep, *md;
 765         struct sockaddr *nam;
 766         u_int32_t rxid, *tl;
 767         caddr_t dpos, cp2;
 768         int error;
 769
 770         /*
 771          * Loop around until we get our own reply
 772          */
 773         for (;;) {
 774                 /*
 775                  * Lock against other receivers so that I don't get stuck in
 776                  * sbwait() after someone else has received my reply for me.
 777                  * Also necessary for connection based protocols to avoid
 778                  * race conditions during a reconnect.
 779                  * If nfs_rcvlock() returns EALREADY, that means that
 780                  * the reply has already been recieved by another
 781                  * process and we can return immediately.  In this
 782                  * case, the lock is not taken to avoid races with
 783                  * other processes.
 784                  */
 785                 error = nfs_rcvlock(myrep);
 786                 if (error == EALREADY)
 787                         return (0);
 788                 if (error)
 789                         return (error);
 790                 /*
 791                  * Get the next Rpc reply off the socket
 792                  */
 793                 error = nfs_receive(myrep, &nam, &mrep);
 794                 nfs_rcvunlock(myrep);
 795                 if (error) {
 796
 797                         /*
 798                          * Ignore routing errors on connectionless protocols??
 799                          */
 800                         if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
 801                                 nmp->nm_so->so_error = 0;
 802                                 if (myrep->r_flags & R_GETONEREP)
 803                                         return (0);
 804                                 continue;
 805                         }
 806                         return (error);
 807                 }
 808                 if (nam)
 809                         FREE(nam, M_SONAME);
 810
 811                 /*
 812                  * Get the xid and check that it is an rpc reply
 813                  */
 814                 md = mrep;
 815                 dpos = mtod(md, caddr_t);
 816                 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
 817                 rxid = *tl++;
 818                 if (*tl != rpc_reply) {
 819 #ifndef NFS_NOSERVER
 820                         if (nmp->nm_flag & NFSMNT_NQNFS) {
 821                                 if (nqnfs_callback(nmp, mrep, md, dpos))
 822                                         nfsstats.rpcinvalid++;
 823                         } else {
 824                                 nfsstats.rpcinvalid++;
 825                                 m_freem(mrep);
 826                         }
 827 #else
 828                         nfsstats.rpcinvalid++;
 829                         m_freem(mrep);
 830 #endif
 831 nfsmout:
 832                         if (myrep->r_flags & R_GETONEREP)
 833                                 return (0);
 834                         continue;
 835                 }
 836
 837                 /*
 838                  * Loop through the request list to match up the reply
 839                  * Iff no match, just drop the datagram
 840                  */
 841                 for (rep = nfs_reqq.tqh_first; rep != 0;
 842                     rep = rep->r_chain.tqe_next) {
 843                         if (rep->r_mrep == NULL && rxid == rep->r_xid) {
 844                                 /* Found it.. */
 845                                 rep->r_mrep = mrep;
 846                                 rep->r_md = md;
 847                                 rep->r_dpos = dpos;
 848                                 if (nfsrtton) {
 849                                         struct rttl *rt;
 850
 851                                         rt = &nfsrtt.rttl[nfsrtt.pos];
 852                                         rt->proc = rep->r_procnum;
 853                                         rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
 854                                         rt->sent = nmp->nm_sent;
 855                                         rt->cwnd = nmp->nm_cwnd;
 856                                         rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
 857                                         rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
 858                                         rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
 859                                         getmicrotime(&rt->tstamp);
 860                                         if (rep->r_flags & R_TIMING)
 861                                                 rt->rtt = rep->r_rtt;
 862                                         else
 863                                                 rt->rtt = 1000000;
 864                                         nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
 865                                 }
 866                                 /*
 867                                  * Update congestion window.
 868                                  * Do the additive increase of
 869                                  * one rpc/rtt.
 870                                  */
 871                                 if (nmp->nm_cwnd <= nmp->nm_sent) {
 872                                         nmp->nm_cwnd +=
 873                                            (NFS_CWNDSCALE * NFS_CWNDSCALE +
 874                                            (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
 875                                         if (nmp->nm_cwnd > NFS_MAXCWND)
 876                                                 nmp->nm_cwnd = NFS_MAXCWND;
 877                                 }
 878                                 if (rep->r_flags & R_SENT) {
 879                                         rep->r_flags &= ~R_SENT;
 880                                         nmp->nm_sent -= NFS_CWNDSCALE;
 881                                 }
 882                                 /*
 883                                  * Update rtt using a gain of 0.125 on the mean
 884                                  * and a gain of 0.25 on the deviation.
 885                                  */
 886                                 if (rep->r_flags & R_TIMING) {
 887                                         /*
 888                                          * Since the timer resolution of
 889                                          * NFS_HZ is so course, it can often
 890                                          * result in r_rtt == 0. Since
 891                                          * r_rtt == N means that the actual
 892                                          * rtt is between N+dt and N+2-dt ticks,
 893                                          * add 1.
 894                                          */
 895                                         t1 = rep->r_rtt + 1;
 896                                         t1 -= (NFS_SRTT(rep) >> 3);
 897                                         NFS_SRTT(rep) += t1;
 898                                         if (t1 < 0)
 899                                                 t1 = -t1;
 900                                         t1 -= (NFS_SDRTT(rep) >> 2);
 901                                         NFS_SDRTT(rep) += t1;
 902                                 }
 903                                 nmp->nm_timeouts = 0;
 904                                 break;
 905                         }
 906                 }
 907                 /*
 908                  * If not matched to a request, drop it.
 909                  * If it's mine, get out.
 910                  */
 911                 if (rep == 0) {
 912                         nfsstats.rpcunexpected++;
 913                         m_freem(mrep);
 914                 } else if (rep == myrep) {
 915                         if (rep->r_mrep == NULL)
 916                                 panic("nfsreply nil");
 917                         return (0);
 918                 }
 919                 if (myrep->r_flags & R_GETONEREP)
 920                         return (0);
 921         }
 922 }
 923
 924 /*
 925  * nfs_request - goes something like this
 926  *      - fill in request struct
 927  *      - links it into list
 928  *      - calls nfs_send() for first transmit
 929  *      - calls nfs_receive() to get reply
 930  *      - break down rpc header and return with nfs reply pointed to
 931  *        by mrep or error
 932  * nb: always frees up mreq mbuf list
 933  */
 934 int
 935 nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp)
 936         struct vnode *vp;
 937         struct mbuf *mrest;
 938         int procnum;
 939         struct proc *procp;
 940         struct ucred *cred;
 941         struct mbuf **mrp;
 942         struct mbuf **mdp;
 943         caddr_t *dposp;
 944 {
 945         register struct mbuf *mrep, *m2;
 946         register struct nfsreq *rep;
 947         register u_int32_t *tl;
 948         register int i;
 949         struct nfsmount *nmp;
 950         struct mbuf *m, *md, *mheadend;
 951         struct nfsnode *np;
 952         char nickv[RPCX_NICKVERF];
 953         time_t reqtime, waituntil;
 954         caddr_t dpos, cp2;
 955         int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type;
 956         int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0;
 957         int verf_len, verf_type;
 958         u_int32_t xid;
 959         u_quad_t frev;
 960         char *auth_str, *verf_str;
 961         NFSKERBKEY_T key;               /* save session key */
 962
 963         /* Reject requests while attempting a forced unmount. */
 964         if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) {
 965                 m_freem(mrest);
 966                 return (ESTALE);
 967         }
 968         nmp = VFSTONFS(vp->v_mount);
 969         MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
 970         rep->r_nmp = nmp;
 971         rep->r_vp = vp;
 972         rep->r_procp = procp;
 973         rep->r_procnum = procnum;
 974         i = 0;
 975         m = mrest;
 976         while (m) {
 977                 i += m->m_len;
 978                 m = m->m_next;
 979         }
 980         mrest_len = i;
 981
 982         /*
 983          * Get the RPC header with authorization.
 984          */
 985 kerbauth:
 986         verf_str = auth_str = (char *)0;
 987         if (nmp->nm_flag & NFSMNT_KERB) {
 988                 verf_str = nickv;
 989                 verf_len = sizeof (nickv);
 990                 auth_type = RPCAUTH_KERB4;
 991                 bzero((caddr_t)key, sizeof (key));
 992                 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
 993                         &auth_len, verf_str, verf_len)) {
 994                         error = nfs_getauth(nmp, rep, cred, &auth_str,
 995                                 &auth_len, verf_str, &verf_len, key);
 996                         if (error) {
 997                                 free((caddr_t)rep, M_NFSREQ);
 998                                 m_freem(mrest);
 999                                 return (error);
1000                         }
1001                 }
1002         } else {
1003                 auth_type = RPCAUTH_UNIX;
1004                 if (cred->cr_ngroups < 1)
1005                         panic("nfsreq nogrps");
1006                 auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ?
1007                         nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) +
1008                         5 * NFSX_UNSIGNED;
1009         }
1010         m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
1011              auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
1012         if (auth_str)
1013                 free(auth_str, M_TEMP);
1014
1015         /*
1016          * For stream protocols, insert a Sun RPC Record Mark.
1017          */
1018         if (nmp->nm_sotype == SOCK_STREAM) {
1019                 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
1020                 *mtod(m, u_int32_t *) = htonl(0x80000000 |
1021                          (m->m_pkthdr.len - NFSX_UNSIGNED));
1022         }
1023         rep->r_mreq = m;
1024         rep->r_xid = xid;
1025 tryagain:
1026         if (nmp->nm_flag & NFSMNT_SOFT)
1027                 rep->r_retry = nmp->nm_retry;
1028         else
1029                 rep->r_retry = NFS_MAXREXMIT + 1;       /* past clip limit */
1030         rep->r_rtt = rep->r_rexmit = 0;
1031         if (proct[procnum] > 0)
1032                 rep->r_flags = R_TIMING;
1033         else
1034                 rep->r_flags = 0;
1035         rep->r_mrep = NULL;
1036
1037         /*
1038          * Do the client side RPC.
1039          */
1040         nfsstats.rpcrequests++;
1041         /*
1042          * Chain request into list of outstanding requests. Be sure
1043          * to put it LAST so timer finds oldest requests first.
1044          */
1045         s = splsoftclock();
1046         TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1047
1048         /* Get send time for nqnfs */
1049         reqtime = time_second;
1050
1051         /*
1052          * If backing off another request or avoiding congestion, don't
1053          * send this one now but let timer do it. If not timing a request,
1054          * do it now.
1055          */
1056         if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1057                 (nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1058                 nmp->nm_sent < nmp->nm_cwnd)) {
1059                 splx(s);
1060                 if (nmp->nm_soflags & PR_CONNREQUIRED)
1061                         error = nfs_sndlock(rep);
1062                 if (!error) {
1063                         m2 = m_copym(m, 0, M_COPYALL, M_WAIT);
1064                         error = nfs_send(nmp->nm_so, nmp->nm_nam, m2, rep);
1065                         if (nmp->nm_soflags & PR_CONNREQUIRED)
1066                                 nfs_sndunlock(rep);
1067                 }
1068                 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1069                         nmp->nm_sent += NFS_CWNDSCALE;
1070                         rep->r_flags |= R_SENT;
1071                 }
1072         } else {
1073                 splx(s);
1074                 rep->r_rtt = -1;
1075         }
1076
1077         /*
1078          * Wait for the reply from our send or the timer's.
1079          */
1080         if (!error || error == EPIPE)
1081                 error = nfs_reply(rep);
1082
1083         /*
1084          * RPC done, unlink the request.
1085          */
1086         s = splsoftclock();
1087         TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1088         splx(s);
1089
1090         /*
1091          * Decrement the outstanding request count.
1092          */
1093         if (rep->r_flags & R_SENT) {
1094                 rep->r_flags &= ~R_SENT;        /* paranoia */
1095                 nmp->nm_sent -= NFS_CWNDSCALE;
1096         }
1097
1098         /*
1099          * If there was a successful reply and a tprintf msg.
1100          * tprintf a response.
1101          */
1102         if (!error && (rep->r_flags & R_TPRINTFMSG))
1103                 nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname,
1104                     "is alive again");
1105         mrep = rep->r_mrep;
1106         md = rep->r_md;
1107         dpos = rep->r_dpos;
1108         if (error) {
1109                 m_freem(rep->r_mreq);
1110                 free((caddr_t)rep, M_NFSREQ);
1111                 return (error);
1112         }
1113
1114         /*
1115          * break down the rpc header and check if ok
1116          */
1117         nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1118         if (*tl++ == rpc_msgdenied) {
1119                 if (*tl == rpc_mismatch)
1120                         error = EOPNOTSUPP;
1121                 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1122                         if (!failed_auth) {
1123                                 failed_auth++;
1124                                 mheadend->m_next = (struct mbuf *)0;
1125                                 m_freem(mrep);
1126                                 m_freem(rep->r_mreq);
1127                                 goto kerbauth;
1128                         } else
1129                                 error = EAUTH;
1130                 } else
1131                         error = EACCES;
1132                 m_freem(mrep);
1133                 m_freem(rep->r_mreq);
1134                 free((caddr_t)rep, M_NFSREQ);
1135                 return (error);
1136         }
1137
1138         /*
1139          * Grab any Kerberos verifier, otherwise just throw it away.
1140          */
1141         verf_type = fxdr_unsigned(int, *tl++);
1142         i = fxdr_unsigned(int32_t, *tl);
1143         if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1144                 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1145                 if (error)
1146                         goto nfsmout;
1147         } else if (i > 0)
1148                 nfsm_adv(nfsm_rndup(i));
1149         nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1150         /* 0 == ok */
1151         if (*tl == 0) {
1152                 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1153                 if (*tl != 0) {
1154                         error = fxdr_unsigned(int, *tl);
1155                         if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1156                                 error == NFSERR_TRYLATER) {
1157                                 m_freem(mrep);
1158                                 error = 0;
1159                                 waituntil = time_second + trylater_delay;
1160                                 while (time_second < waituntil)
1161                                         (void) tsleep((caddr_t)&lbolt,
1162                                                 PSOCK, "nqnfstry", 0);
1163                                 trylater_delay *= nfs_backoff[trylater_cnt];
1164                                 if (trylater_cnt < 7)
1165                                         trylater_cnt++;
1166                                 goto tryagain;
1167                         }
1168
1169                         /*
1170                          * If the File Handle was stale, invalidate the
1171                          * lookup cache, just in case.
1172                          */
1173                         if (error == ESTALE)
1174                                 cache_purge(vp);
1175                         if (nmp->nm_flag & NFSMNT_NFSV3) {
1176                                 *mrp = mrep;
1177                                 *mdp = md;
1178                                 *dposp = dpos;
1179                                 error |= NFSERR_RETERR;
1180                         } else
1181                                 m_freem(mrep);
1182                         m_freem(rep->r_mreq);
1183                         free((caddr_t)rep, M_NFSREQ);
1184                         return (error);
1185                 }
1186
1187                 /*
1188                  * For nqnfs, get any lease in reply
1189                  */
1190                 if (nmp->nm_flag & NFSMNT_NQNFS) {
1191                         nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1192                         if (*tl) {
1193                                 np = VTONFS(vp);
1194                                 nqlflag = fxdr_unsigned(int, *tl);
1195                                 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1196                                 cachable = fxdr_unsigned(int, *tl++);
1197                                 reqtime += fxdr_unsigned(int, *tl++);
1198                                 if (reqtime > time_second) {
1199                                     frev = fxdr_hyper(tl);
1200                                     nqnfs_clientlease(nmp, np, nqlflag,
1201                                         cachable, reqtime, frev);
1202                                 }
1203                         }
1204                 }
1205                 *mrp = mrep;
1206                 *mdp = md;
1207                 *dposp = dpos;
1208                 m_freem(rep->r_mreq);
1209                 FREE((caddr_t)rep, M_NFSREQ);
1210                 return (0);
1211         }
1212         m_freem(mrep);
1213         error = EPROTONOSUPPORT;
1214 nfsmout:
1215         m_freem(rep->r_mreq);
1216         free((caddr_t)rep, M_NFSREQ);
1217         return (error);
1218 }
1219
1220 #ifndef NFS_NOSERVER
1221 /*
1222  * Generate the rpc reply header
1223  * siz arg. is used to decide if adding a cluster is worthwhile
1224  */
1225 int
1226 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1227         int siz;
1228         struct nfsrv_descript *nd;
1229         struct nfssvc_sock *slp;
1230         int err;
1231         int cache;
1232         u_quad_t *frev;
1233         struct mbuf **mrq;
1234         struct mbuf **mbp;
1235         caddr_t *bposp;
1236 {
1237         register u_int32_t *tl;
1238         register struct mbuf *mreq;
1239         caddr_t bpos;
1240         struct mbuf *mb, *mb2;
1241
1242         MGETHDR(mreq, M_WAIT, MT_DATA);
1243         mb = mreq;
1244         /*
1245          * If this is a big reply, use a cluster else
1246          * try and leave leading space for the lower level headers.
1247          */
1248         siz += RPC_REPLYSIZ;
1249         if ((max_hdr + siz) >= MINCLSIZE) {
1250                 MCLGET(mreq, M_WAIT);
1251         } else
1252                 mreq->m_data += max_hdr;
1253         tl = mtod(mreq, u_int32_t *);
1254         mreq->m_len = 6 * NFSX_UNSIGNED;
1255         bpos = ((caddr_t)tl) + mreq->m_len;
1256         *tl++ = txdr_unsigned(nd->nd_retxid);
1257         *tl++ = rpc_reply;
1258         if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1259                 *tl++ = rpc_msgdenied;
1260                 if (err & NFSERR_AUTHERR) {
1261                         *tl++ = rpc_autherr;
1262                         *tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1263                         mreq->m_len -= NFSX_UNSIGNED;
1264                         bpos -= NFSX_UNSIGNED;
1265                 } else {
1266                         *tl++ = rpc_mismatch;
1267                         *tl++ = txdr_unsigned(RPC_VER2);
1268                         *tl = txdr_unsigned(RPC_VER2);
1269                 }
1270         } else {
1271                 *tl++ = rpc_msgaccepted;
1272
1273                 /*
1274                  * For Kerberos authentication, we must send the nickname
1275                  * verifier back, otherwise just RPCAUTH_NULL.
1276                  */
1277                 if (nd->nd_flag & ND_KERBFULL) {
1278                     register struct nfsuid *nuidp;
1279                     struct timeval ktvin, ktvout;
1280
1281                     for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1282                         nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1283                         if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1284                             (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1285                              &nuidp->nu_haddr, nd->nd_nam2)))
1286                             break;
1287                     }
1288                     if (nuidp) {
1289                         ktvin.tv_sec =
1290                             txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1291                         ktvin.tv_usec =
1292                             txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1293
1294                         /*
1295                          * Encrypt the timestamp in ecb mode using the
1296                          * session key.
1297                          */
1298 #ifdef NFSKERB
1299                         XXX
1300 #endif
1301
1302                         *tl++ = rpc_auth_kerb;
1303                         *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1304                         *tl = ktvout.tv_sec;
1305                         nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1306                         *tl++ = ktvout.tv_usec;
1307                         *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1308                     } else {
1309                         *tl++ = 0;
1310                         *tl++ = 0;
1311                     }
1312                 } else {
1313                         *tl++ = 0;
1314                         *tl++ = 0;
1315                 }
1316                 switch (err) {
1317                 case EPROGUNAVAIL:
1318                         *tl = txdr_unsigned(RPC_PROGUNAVAIL);
1319                         break;
1320                 case EPROGMISMATCH:
1321                         *tl = txdr_unsigned(RPC_PROGMISMATCH);
1322                         nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1323                         if (nd->nd_flag & ND_NQNFS) {
1324                                 *tl++ = txdr_unsigned(3);
1325                                 *tl = txdr_unsigned(3);
1326                         } else {
1327                                 *tl++ = txdr_unsigned(2);
1328                                 *tl = txdr_unsigned(3);
1329                         }
1330                         break;
1331                 case EPROCUNAVAIL:
1332                         *tl = txdr_unsigned(RPC_PROCUNAVAIL);
1333                         break;
1334                 case EBADRPC:
1335                         *tl = txdr_unsigned(RPC_GARBAGE);
1336                         break;
1337                 default:
1338                         *tl = 0;
1339                         if (err != NFSERR_RETVOID) {
1340                                 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1341                                 if (err)
1342                                     *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1343                                 else
1344                                     *tl = 0;
1345                         }
1346                         break;
1347                 };
1348         }
1349
1350         /*
1351          * For nqnfs, piggyback lease as requested.
1352          */
1353         if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1354                 if (nd->nd_flag & ND_LEASE) {
1355                         nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1356                         *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1357                         *tl++ = txdr_unsigned(cache);
1358                         *tl++ = txdr_unsigned(nd->nd_duration);
1359                         txdr_hyper(*frev, tl);
1360                 } else {
1361                         nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1362                         *tl = 0;
1363                 }
1364         }
1365         if (mrq != NULL)
1366             *mrq = mreq;
1367         *mbp = mb;
1368         *bposp = bpos;
1369         if (err != 0 && err != NFSERR_RETVOID)
1370                 nfsstats.srvrpc_errs++;
1371         return (0);
1372 }
1373
1374
1375 #endif /* NFS_NOSERVER */
1376 /*
1377  * Nfs timer routine
1378  * Scan the nfsreq list and retranmit any requests that have timed out
1379  * To avoid retransmission attempts on STREAM sockets (in the future) make
1380  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1381  */
1382 void
1383 nfs_timer(arg)
1384         void *arg;      /* never used */
1385 {
1386         register struct nfsreq *rep;
1387         register struct mbuf *m;
1388         register struct socket *so;
1389         register struct nfsmount *nmp;
1390         register int timeo;
1391         int s, error;
1392 #ifndef NFS_NOSERVER
1393         static long lasttime = 0;
1394         register struct nfssvc_sock *slp;
1395         u_quad_t cur_usec;
1396 #endif /* NFS_NOSERVER */
1397         struct proc *p = &proc0; /* XXX for credentials, will break if sleep */
1398
1399         s = splnet();
1400         for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) {
1401                 nmp = rep->r_nmp;
1402                 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1403                         continue;
1404                 if (nfs_sigintr(nmp, rep, rep->r_procp)) {
1405                         nfs_softterm(rep);
1406                         continue;
1407                 }
1408                 if (rep->r_rtt >= 0) {
1409                         rep->r_rtt++;
1410                         if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1411                                 timeo = nmp->nm_timeo;
1412                         else
1413                                 timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1414                         if (nmp->nm_timeouts > 0)
1415                                 timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1416                         if (rep->r_rtt <= timeo)
1417                                 continue;
1418                         if (nmp->nm_timeouts < 8)
1419                                 nmp->nm_timeouts++;
1420                 }
1421                 /*
1422                  * Check for server not responding
1423                  */
1424                 if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1425                      rep->r_rexmit > nmp->nm_deadthresh) {
1426                         nfs_msg(rep->r_procp,
1427                             nmp->nm_mountp->mnt_stat.f_mntfromname,
1428                             "not responding");
1429                         rep->r_flags |= R_TPRINTFMSG;
1430                 }
1431                 if (rep->r_rexmit >= rep->r_retry) {    /* too many */
1432                         nfsstats.rpctimeouts++;
1433                         nfs_softterm(rep);
1434                         continue;
1435                 }
1436                 if (nmp->nm_sotype != SOCK_DGRAM) {
1437                         if (++rep->r_rexmit > NFS_MAXREXMIT)
1438                                 rep->r_rexmit = NFS_MAXREXMIT;
1439                         continue;
1440                 }
1441                 if ((so = nmp->nm_so) == NULL)
1442                         continue;
1443
1444                 /*
1445                  * If there is enough space and the window allows..
1446                  *      Resend it
1447                  * Set r_rtt to -1 in case we fail to send it now.
1448                  */
1449                 rep->r_rtt = -1;
1450                 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1451                    ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1452                     (rep->r_flags & R_SENT) ||
1453                     nmp->nm_sent < nmp->nm_cwnd) &&
1454                    (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1455                         if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
1456                             error = (*so->so_proto->pr_usrreqs->pru_send)
1457                                     (so, 0, m, (struct sockaddr *)0,
1458                                      (struct mbuf *)0, p);
1459                         else
1460                             error = (*so->so_proto->pr_usrreqs->pru_send)
1461                                     (so, 0, m, nmp->nm_nam, (struct mbuf *)0,
1462                                      p);
1463                         if (error) {
1464                                 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
1465                                         so->so_error = 0;
1466                         } else {
1467                                 /*
1468                                  * Iff first send, start timing
1469                                  * else turn timing off, backoff timer
1470                                  * and divide congestion window by 2.
1471                                  */
1472                                 if (rep->r_flags & R_SENT) {
1473                                         rep->r_flags &= ~R_TIMING;
1474                                         if (++rep->r_rexmit > NFS_MAXREXMIT)
1475                                                 rep->r_rexmit = NFS_MAXREXMIT;
1476                                         nmp->nm_cwnd >>= 1;
1477                                         if (nmp->nm_cwnd < NFS_CWNDSCALE)
1478                                                 nmp->nm_cwnd = NFS_CWNDSCALE;
1479                                         nfsstats.rpcretries++;
1480                                 } else {
1481                                         rep->r_flags |= R_SENT;
1482                                         nmp->nm_sent += NFS_CWNDSCALE;
1483                                 }
1484                                 rep->r_rtt = 0;
1485                         }
1486                 }
1487         }
1488 #ifndef NFS_NOSERVER
1489         /*
1490          * Call the nqnfs server timer once a second to handle leases.
1491          */
1492         if (lasttime != time_second) {
1493                 lasttime = time_second;
1494                 nqnfs_serverd();
1495         }
1496
1497         /*
1498          * Scan the write gathering queues for writes that need to be
1499          * completed now.
1500          */
1501         cur_usec = nfs_curusec();
1502         for (slp = nfssvc_sockhead.tqh_first; slp != 0;
1503             slp = slp->ns_chain.tqe_next) {
1504             if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1505                 nfsrv_wakenfsd(slp);
1506         }
1507 #endif /* NFS_NOSERVER */
1508         splx(s);
1509         nfs_timer_handle = timeout(nfs_timer, (void *)0, nfs_ticks);
1510 }
1511
1512 /*
1513  * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1514  * wait for all requests to complete. This is used by forced unmounts
1515  * to terminate any outstanding RPCs.
1516  */
1517 int
1518 nfs_nmcancelreqs(nmp)
1519         struct nfsmount *nmp;
1520 {
1521         struct nfsreq *req;
1522         int i, s;
1523
1524         s = splnet();
1525         TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
1526                 if (nmp != req->r_nmp || req->r_mrep != NULL ||
1527                     (req->r_flags & R_SOFTTERM))
1528                         continue;
1529                 nfs_softterm(req);
1530         }
1531         splx(s);
1532
1533         for (i = 0; i < 30; i++) {
1534                 s = splnet();
1535                 TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
1536                         if (nmp == req->r_nmp)
1537                                 break;
1538                 }
1539                 splx(s);
1540                 if (req == NULL)
1541                         return (0);
1542                 tsleep(&lbolt, PSOCK, "nfscancel", 0);
1543         }
1544         return (EBUSY);
1545 }
1546
1547 /*
1548  * Flag a request as being about to terminate (due to NFSMNT_INT/NFSMNT_SOFT).
1549  * The nm_send count is decremented now to avoid deadlocks when the process in
1550  * soreceive() hasn't yet managed to send its own request.
1551  */
1552
1553 static void
1554 nfs_softterm(rep)
1555         struct nfsreq *rep;
1556 {
1557         rep->r_flags |= R_SOFTTERM;
1558
1559         if (rep->r_flags & R_SENT) {
1560                 rep->r_nmp->nm_sent -= NFS_CWNDSCALE;
1561                 rep->r_flags &= ~R_SENT;
1562         }
1563 }
1564
1565 /*
1566  * Test for a termination condition pending on the process.
1567  * This is used for NFSMNT_INT mounts.
1568  */
1569 int
1570 nfs_sigintr(nmp, rep, p)
1571         struct nfsmount *nmp;
1572         struct nfsreq *rep;
1573         register struct proc *p;
1574 {
1575         sigset_t tmpset;
1576
1577         if (rep && (rep->r_flags & R_SOFTTERM))
1578                 return (EINTR);
1579         /* Terminate all requests while attempting a forced unmount. */
1580         if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1581                 return (EINTR);
1582         if (!(nmp->nm_flag & NFSMNT_INT))
1583                 return (0);
1584         if (p == NULL)
1585                 return (0);
1586
1587         tmpset = p->p_siglist;
1588         SIGSETNAND(tmpset, p->p_sigmask);
1589         SIGSETNAND(tmpset, p->p_sigignore);
1590         if (SIGNOTEMPTY(p->p_siglist) && NFSINT_SIGMASK(tmpset))
1591                 return (EINTR);
1592
1593         return (0);
1594 }
1595
1596 /*
1597  * Lock a socket against others.
1598  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1599  * and also to avoid race conditions between the processes with nfs requests
1600  * in progress when a reconnect is necessary.
1601  */
1602 int
1603 nfs_sndlock(rep)
1604         struct nfsreq *rep;
1605 {
1606         register int *statep = &rep->r_nmp->nm_state;
1607         struct proc *p;
1608         int slpflag = 0, slptimeo = 0;
1609
1610         p = rep->r_procp;
1611         if (rep->r_nmp->nm_flag & NFSMNT_INT)
1612                 slpflag = PCATCH;
1613         while (*statep & NFSSTA_SNDLOCK) {
1614                 if (nfs_sigintr(rep->r_nmp, rep, p))
1615                         return (EINTR);
1616                 *statep |= NFSSTA_WANTSND;
1617                 (void) tsleep((caddr_t)statep, slpflag | (PZERO - 1),
1618                         "nfsndlck", slptimeo);
1619                 if (slpflag == PCATCH) {
1620                         slpflag = 0;
1621                         slptimeo = 2 * hz;
1622                 }
1623         }
1624         /* Always fail if our request has been cancelled. */
1625         if ((rep->r_flags & R_SOFTTERM))
1626                 return (EINTR);
1627         *statep |= NFSSTA_SNDLOCK;
1628         return (0);
1629 }
1630
1631 /*
1632  * Unlock the stream socket for others.
1633  */
1634 void
1635 nfs_sndunlock(rep)
1636         struct nfsreq *rep;
1637 {
1638         register int *statep = &rep->r_nmp->nm_state;
1639
1640         if ((*statep & NFSSTA_SNDLOCK) == 0)
1641                 panic("nfs sndunlock");
1642         *statep &= ~NFSSTA_SNDLOCK;
1643         if (*statep & NFSSTA_WANTSND) {
1644                 *statep &= ~NFSSTA_WANTSND;
1645                 wakeup((caddr_t)statep);
1646         }
1647 }
1648
1649 static int
1650 nfs_rcvlock(rep)
1651         register struct nfsreq *rep;
1652 {
1653         register int *statep = &rep->r_nmp->nm_state;
1654         int slpflag, slptimeo = 0;
1655
1656         if (rep->r_nmp->nm_flag & NFSMNT_INT)
1657                 slpflag = PCATCH;
1658         else
1659                 slpflag = 0;
1660         while (*statep & NFSSTA_RCVLOCK) {
1661                 if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp))
1662                         return (EINTR);
1663                 *statep |= NFSSTA_WANTRCV;
1664                 (void) tsleep((caddr_t)statep, slpflag | (PZERO - 1), "nfsrcvlk",
1665                         slptimeo);
1666                 /*
1667                  * If our reply was recieved while we were sleeping,
1668                  * then just return without taking the lock to avoid a
1669                  * situation where a single iod could 'capture' the
1670                  * recieve lock.
1671                  */
1672                 if (rep->r_mrep != NULL)
1673                         return (EALREADY);
1674                 if (slpflag == PCATCH) {
1675                         slpflag = 0;
1676                         slptimeo = 2 * hz;
1677                 }
1678         }
1679         *statep |= NFSSTA_RCVLOCK;
1680         return (0);
1681 }
1682
1683 /*
1684  * Unlock the stream socket for others.
1685  */
1686 static void
1687 nfs_rcvunlock(rep)
1688         register struct nfsreq *rep;
1689 {
1690         register int *statep = &rep->r_nmp->nm_state;
1691
1692         if ((*statep & NFSSTA_RCVLOCK) == 0)
1693                 panic("nfs rcvunlock");
1694         *statep &= ~NFSSTA_RCVLOCK;
1695         if (*statep & NFSSTA_WANTRCV) {
1696                 *statep &= ~NFSSTA_WANTRCV;
1697                 wakeup((caddr_t)statep);
1698         }
1699 }
1700
1701 /*
1702  *      nfs_realign:
1703  *
1704  *      Check for badly aligned mbuf data and realign by copying the unaligned
1705  *      portion of the data into a new mbuf chain and freeing the portions
1706  *      of the old chain that were replaced.
1707  *
1708  *      We cannot simply realign the data within the existing mbuf chain
1709  *      because the underlying buffers may contain other rpc commands and
1710  *      we cannot afford to overwrite them.
1711  *
1712  *      We would prefer to avoid this situation entirely.  The situation does
1713  *      not occur with NFS/UDP and is supposed to only occassionally occur
1714  *      with TCP.  Use vfs.nfs.realign_count and realign_test to check this.
1715  */
1716 static void
1717 nfs_realign(pm, hsiz)
1718         register struct mbuf **pm;
1719         int hsiz;
1720 {
1721         struct mbuf *m;
1722         struct mbuf *n = NULL;
1723         int off = 0;
1724
1725         ++nfs_realign_test;
1726
1727         while ((m = *pm) != NULL) {
1728                 if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) {
1729                         MGET(n, M_WAIT, MT_DATA);
1730                         if (m->m_len >= MINCLSIZE) {
1731                                 MCLGET(n, M_WAIT);
1732                         }
1733                         n->m_len = 0;
1734                         break;
1735                 }
1736                 pm = &m->m_next;
1737         }
1738
1739         /*
1740          * If n is non-NULL, loop on m copying data, then replace the
1741          * portion of the chain that had to be realigned.
1742          */
1743         if (n != NULL) {
1744                 ++nfs_realign_count;
1745                 while (m) {
1746                         m_copyback(n, off, m->m_len, mtod(m, caddr_t));
1747                         off += m->m_len;
1748                         m = m->m_next;
1749                 }
1750                 m_freem(*pm);
1751                 *pm = n;
1752         }
1753 }
1754
1755 #ifndef NFS_NOSERVER
1756
1757 /*
1758  * Parse an RPC request
1759  * - verify it
1760  * - fill in the cred struct.
1761  */
1762 int
1763 nfs_getreq(nd, nfsd, has_header)
1764         register struct nfsrv_descript *nd;
1765         struct nfsd *nfsd;
1766         int has_header;
1767 {
1768         register int len, i;
1769         register u_int32_t *tl;
1770         register int32_t t1;
1771         struct uio uio;
1772         struct iovec iov;
1773         caddr_t dpos, cp2, cp;
1774         u_int32_t nfsvers, auth_type;
1775         uid_t nickuid;
1776         int error = 0, nqnfs = 0, ticklen;
1777         struct mbuf *mrep, *md;
1778         register struct nfsuid *nuidp;
1779         struct timeval tvin, tvout;
1780 #if 0                           /* until encrypted keys are implemented */
1781         NFSKERBKEYSCHED_T keys; /* stores key schedule */
1782 #endif
1783
1784         mrep = nd->nd_mrep;
1785         md = nd->nd_md;
1786         dpos = nd->nd_dpos;
1787         if (has_header) {
1788                 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1789                 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1790                 if (*tl++ != rpc_call) {
1791                         m_freem(mrep);
1792                         return (EBADRPC);
1793                 }
1794         } else
1795                 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1796         nd->nd_repstat = 0;
1797         nd->nd_flag = 0;
1798         if (*tl++ != rpc_vers) {
1799                 nd->nd_repstat = ERPCMISMATCH;
1800                 nd->nd_procnum = NFSPROC_NOOP;
1801                 return (0);
1802         }
1803         if (*tl != nfs_prog) {
1804                 if (*tl == nqnfs_prog)
1805                         nqnfs++;
1806                 else {
1807                         nd->nd_repstat = EPROGUNAVAIL;
1808                         nd->nd_procnum = NFSPROC_NOOP;
1809                         return (0);
1810                 }
1811         }
1812         tl++;
1813         nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1814         if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1815                 (nfsvers != NQNFS_VER3 && nqnfs)) {
1816                 nd->nd_repstat = EPROGMISMATCH;
1817                 nd->nd_procnum = NFSPROC_NOOP;
1818                 return (0);
1819         }
1820         if (nqnfs)
1821                 nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1822         else if (nfsvers == NFS_VER3)
1823                 nd->nd_flag = ND_NFSV3;
1824         nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1825         if (nd->nd_procnum == NFSPROC_NULL)
1826                 return (0);
1827         if (nd->nd_procnum >= NFS_NPROCS ||
1828                 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1829                 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1830                 nd->nd_repstat = EPROCUNAVAIL;
1831                 nd->nd_procnum = NFSPROC_NOOP;
1832                 return (0);
1833         }
1834         if ((nd->nd_flag & ND_NFSV3) == 0)
1835                 nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1836         auth_type = *tl++;
1837         len = fxdr_unsigned(int, *tl++);
1838         if (len < 0 || len > RPCAUTH_MAXSIZ) {
1839                 m_freem(mrep);
1840                 return (EBADRPC);
1841         }
1842
1843         nd->nd_flag &= ~ND_KERBAUTH;
1844         /*
1845          * Handle auth_unix or auth_kerb.
1846          */
1847         if (auth_type == rpc_auth_unix) {
1848                 len = fxdr_unsigned(int, *++tl);
1849                 if (len < 0 || len > NFS_MAXNAMLEN) {
1850                         m_freem(mrep);
1851                         return (EBADRPC);
1852                 }
1853                 nfsm_adv(nfsm_rndup(len));
1854                 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1855                 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred));
1856                 nd->nd_cr.cr_ref = 1;
1857                 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1858                 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1859                 len = fxdr_unsigned(int, *tl);
1860                 if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1861                         m_freem(mrep);
1862                         return (EBADRPC);
1863                 }
1864                 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1865                 for (i = 1; i <= len; i++)
1866                     if (i < NGROUPS)
1867                         nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1868                     else
1869                         tl++;
1870                 nd->nd_cr.cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1);
1871                 if (nd->nd_cr.cr_ngroups > 1)
1872                     nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1873                 len = fxdr_unsigned(int, *++tl);
1874                 if (len < 0 || len > RPCAUTH_MAXSIZ) {
1875                         m_freem(mrep);
1876                         return (EBADRPC);
1877                 }
1878                 if (len > 0)
1879                         nfsm_adv(nfsm_rndup(len));
1880         } else if (auth_type == rpc_auth_kerb) {
1881                 switch (fxdr_unsigned(int, *tl++)) {
1882                 case RPCAKN_FULLNAME:
1883                         ticklen = fxdr_unsigned(int, *tl);
1884                         *((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1885                         uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1886                         nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1887                         if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1888                                 m_freem(mrep);
1889                                 return (EBADRPC);
1890                         }
1891                         uio.uio_offset = 0;
1892                         uio.uio_iov = &iov;
1893                         uio.uio_iovcnt = 1;
1894                         uio.uio_segflg = UIO_SYSSPACE;
1895                         iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1896                         iov.iov_len = RPCAUTH_MAXSIZ - 4;
1897                         nfsm_mtouio(&uio, uio.uio_resid);
1898                         nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1899                         if (*tl++ != rpc_auth_kerb ||
1900                                 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1901                                 printf("Bad kerb verifier\n");
1902                                 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1903                                 nd->nd_procnum = NFSPROC_NOOP;
1904                                 return (0);
1905                         }
1906                         nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1907                         tl = (u_int32_t *)cp;
1908                         if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1909                                 printf("Not fullname kerb verifier\n");
1910                                 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1911                                 nd->nd_procnum = NFSPROC_NOOP;
1912                                 return (0);
1913                         }
1914                         cp += NFSX_UNSIGNED;
1915                         bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED);
1916                         nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1917                         nd->nd_flag |= ND_KERBFULL;
1918                         nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1919                         break;
1920                 case RPCAKN_NICKNAME:
1921                         if (len != 2 * NFSX_UNSIGNED) {
1922                                 printf("Kerb nickname short\n");
1923                                 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1924                                 nd->nd_procnum = NFSPROC_NOOP;
1925                                 return (0);
1926                         }
1927                         nickuid = fxdr_unsigned(uid_t, *tl);
1928                         nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1929                         if (*tl++ != rpc_auth_kerb ||
1930                                 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1931                                 printf("Kerb nick verifier bad\n");
1932                                 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1933                                 nd->nd_procnum = NFSPROC_NOOP;
1934                                 return (0);
1935                         }
1936                         nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1937                         tvin.tv_sec = *tl++;
1938                         tvin.tv_usec = *tl;
1939
1940                         for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1941                             nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1942                                 if (nuidp->nu_cr.cr_uid == nickuid &&
1943                                     (!nd->nd_nam2 ||
1944                                      netaddr_match(NU_NETFAM(nuidp),
1945                                       &nuidp->nu_haddr, nd->nd_nam2)))
1946                                         break;
1947                         }
1948                         if (!nuidp) {
1949                                 nd->nd_repstat =
1950                                         (NFSERR_AUTHERR|AUTH_REJECTCRED);
1951                                 nd->nd_procnum = NFSPROC_NOOP;
1952                                 return (0);
1953                         }
1954
1955                         /*
1956                          * Now, decrypt the timestamp using the session key
1957                          * and validate it.
1958                          */
1959 #ifdef NFSKERB
1960                         XXX
1961 #endif
1962
1963                         tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1964                         tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1965                         if (nuidp->nu_expire < time_second ||
1966                             nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1967                             (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1968                              nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1969                                 nuidp->nu_expire = 0;
1970                                 nd->nd_repstat =
1971                                     (NFSERR_AUTHERR|AUTH_REJECTVERF);
1972                                 nd->nd_procnum = NFSPROC_NOOP;
1973                                 return (0);
1974                         }
1975                         nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1976                         nd->nd_flag |= ND_KERBNICK;
1977                 };
1978         } else {
1979                 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1980                 nd->nd_procnum = NFSPROC_NOOP;
1981                 return (0);
1982         }
1983
1984         /*
1985          * For nqnfs, get piggybacked lease request.
1986          */
1987         if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
1988                 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1989                 nd->nd_flag |= fxdr_unsigned(int, *tl);
1990                 if (nd->nd_flag & ND_LEASE) {
1991                         nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1992                         nd->nd_duration = fxdr_unsigned(int32_t, *tl);
1993                 } else
1994                         nd->nd_duration = NQ_MINLEASE;
1995         } else
1996                 nd->nd_duration = NQ_MINLEASE;
1997         nd->nd_md = md;
1998         nd->nd_dpos = dpos;
1999         return (0);
2000 nfsmout:
2001         return (error);
2002 }
2003
2004 #endif
2005
2006 static int
2007 nfs_msg(p, server, msg)
2008         struct proc *p;
2009         char *server, *msg;
2010 {
2011         tpr_t tpr;
2012
2013         if (p)
2014                 tpr = tprintf_open(p);
2015         else
2016                 tpr = NULL;
2017         tprintf(tpr, "nfs server %s: %s\n", server, msg);
2018         tprintf_close(tpr);
2019         return (0);
2020 }
2021
2022 #ifndef NFS_NOSERVER
2023 /*
2024  * Socket upcall routine for the nfsd sockets.
2025  * The caddr_t arg is a pointer to the "struct nfssvc_sock".
2026  * Essentially do as much as possible non-blocking, else punt and it will
2027  * be called with M_WAIT from an nfsd.
2028  */
2029 void
2030 nfsrv_rcv(so, arg, waitflag)
2031         struct socket *so;
2032         void *arg;
2033         int waitflag;
2034 {
2035         register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
2036         register struct mbuf *m;
2037         struct mbuf *mp;
2038         struct sockaddr *nam;
2039         struct uio auio;
2040         int flags, error;
2041
2042         if ((slp->ns_flag & SLP_VALID) == 0)
2043                 return;
2044 #ifdef notdef
2045         /*
2046          * Define this to test for nfsds handling this under heavy load.
2047          */
2048         if (waitflag == M_DONTWAIT) {
2049                 slp->ns_flag |= SLP_NEEDQ; goto dorecs;
2050         }
2051 #endif
2052         auio.uio_procp = NULL;
2053         if (so->so_type == SOCK_STREAM) {
2054                 /*
2055                  * If there are already records on the queue, defer soreceive()
2056                  * to an nfsd so that there is feedback to the TCP layer that
2057                  * the nfs servers are heavily loaded.
2058                  */
2059                 if (STAILQ_FIRST(&slp->ns_rec) && waitflag == M_DONTWAIT) {
2060                         slp->ns_flag |= SLP_NEEDQ;
2061                         goto dorecs;
2062                 }
2063
2064                 /*
2065                  * Do soreceive().
2066                  */
2067                 auio.uio_resid = 1000000000;
2068                 flags = MSG_DONTWAIT;
2069                 error = so->so_proto->pr_usrreqs->pru_soreceive
2070                         (so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
2071                 if (error || mp == (struct mbuf *)0) {
2072                         if (error == EWOULDBLOCK)
2073                                 slp->ns_flag |= SLP_NEEDQ;
2074                         else
2075                                 slp->ns_flag |= SLP_DISCONN;
2076                         goto dorecs;
2077                 }
2078                 m = mp;
2079                 if (slp->ns_rawend) {
2080                         slp->ns_rawend->m_next = m;
2081                         slp->ns_cc += 1000000000 - auio.uio_resid;
2082                 } else {
2083                         slp->ns_raw = m;
2084                         slp->ns_cc = 1000000000 - auio.uio_resid;
2085                 }
2086                 while (m->m_next)
2087                         m = m->m_next;
2088                 slp->ns_rawend = m;
2089
2090                 /*
2091                  * Now try and parse record(s) out of the raw stream data.
2092                  */
2093                 error = nfsrv_getstream(slp, waitflag);
2094                 if (error) {
2095                         if (error == EPERM)
2096                                 slp->ns_flag |= SLP_DISCONN;
2097                         else
2098                                 slp->ns_flag |= SLP_NEEDQ;
2099                 }
2100         } else {
2101                 do {
2102                         auio.uio_resid = 1000000000;
2103                         flags = MSG_DONTWAIT;
2104                         error = so->so_proto->pr_usrreqs->pru_soreceive
2105                                 (so, &nam, &auio, &mp,
2106                                                 (struct mbuf **)0, &flags);
2107                         if (mp) {
2108                                 struct nfsrv_rec *rec;
2109                                 rec = malloc(sizeof(struct nfsrv_rec),
2110                                              M_NFSRVDESC, waitflag);
2111                                 if (!rec) {
2112                                         if (nam)
2113                                                 FREE(nam, M_SONAME);
2114                                         m_freem(mp);
2115                                         continue;
2116                                 }
2117                                 nfs_realign(&mp, 10 * NFSX_UNSIGNED);
2118                                 rec->nr_address = nam;
2119                                 rec->nr_packet = mp;
2120                                 STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
2121                         }
2122                         if (error) {
2123                                 if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2124                                         && error != EWOULDBLOCK) {
2125                                         slp->ns_flag |= SLP_DISCONN;
2126                                         goto dorecs;
2127                                 }
2128                         }
2129                 } while (mp);
2130         }
2131
2132         /*
2133          * Now try and process the request records, non-blocking.
2134          */
2135 dorecs:
2136         if (waitflag == M_DONTWAIT &&
2137                 (STAILQ_FIRST(&slp->ns_rec)
2138                  || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
2139                 nfsrv_wakenfsd(slp);
2140 }
2141
2142 /*
2143  * Try and extract an RPC request from the mbuf data list received on a
2144  * stream socket. The "waitflag" argument indicates whether or not it
2145  * can sleep.
2146  */
2147 static int
2148 nfsrv_getstream(slp, waitflag)
2149         register struct nfssvc_sock *slp;
2150         int waitflag;
2151 {
2152         register struct mbuf *m, **mpp;
2153         register char *cp1, *cp2;
2154         register int len;
2155         struct mbuf *om, *m2, *recm;
2156         u_int32_t recmark;
2157
2158         if (slp->ns_flag & SLP_GETSTREAM)
2159                 panic("nfs getstream");
2160         slp->ns_flag |= SLP_GETSTREAM;
2161         for (;;) {
2162             if (slp->ns_reclen == 0) {
2163                 if (slp->ns_cc < NFSX_UNSIGNED) {
2164                         slp->ns_flag &= ~SLP_GETSTREAM;
2165                         return (0);
2166                 }
2167                 m = slp->ns_raw;
2168                 if (m->m_len >= NFSX_UNSIGNED) {
2169                         bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
2170                         m->m_data += NFSX_UNSIGNED;
2171                         m->m_len -= NFSX_UNSIGNED;
2172                 } else {
2173                         cp1 = (caddr_t)&recmark;
2174                         cp2 = mtod(m, caddr_t);
2175                         while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
2176                                 while (m->m_len == 0) {
2177                                         m = m->m_next;
2178                                         cp2 = mtod(m, caddr_t);
2179                                 }
2180                                 *cp1++ = *cp2++;
2181                                 m->m_data++;
2182                                 m->m_len--;
2183                         }
2184                 }
2185                 slp->ns_cc -= NFSX_UNSIGNED;
2186                 recmark = ntohl(recmark);
2187                 slp->ns_reclen = recmark & ~0x80000000;
2188                 if (recmark & 0x80000000)
2189                         slp->ns_flag |= SLP_LASTFRAG;
2190                 else
2191                         slp->ns_flag &= ~SLP_LASTFRAG;
2192                 if (slp->ns_reclen > NFS_MAXPACKET) {
2193                         slp->ns_flag &= ~SLP_GETSTREAM;
2194                         return (EPERM);
2195                 }
2196             }
2197
2198             /*
2199              * Now get the record part.
2200              *
2201              * Note that slp->ns_reclen may be 0.  Linux sometimes
2202              * generates 0-length RPCs
2203              */
2204             recm = NULL;
2205             if (slp->ns_cc == slp->ns_reclen) {
2206                 recm = slp->ns_raw;
2207                 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2208                 slp->ns_cc = slp->ns_reclen = 0;
2209             } else if (slp->ns_cc > slp->ns_reclen) {
2210                 len = 0;
2211                 m = slp->ns_raw;
2212                 om = (struct mbuf *)0;
2213
2214                 while (len < slp->ns_reclen) {
2215                         if ((len + m->m_len) > slp->ns_reclen) {
2216                                 m2 = m_copym(m, 0, slp->ns_reclen - len,
2217                                         waitflag);
2218                                 if (m2) {
2219                                         if (om) {
2220                                                 om->m_next = m2;
2221                                                 recm = slp->ns_raw;
2222                                         } else
2223                                                 recm = m2;
2224                                         m->m_data += slp->ns_reclen - len;
2225                                         m->m_len -= slp->ns_reclen - len;
2226                                         len = slp->ns_reclen;
2227                                 } else {
2228                                         slp->ns_flag &= ~SLP_GETSTREAM;
2229                                         return (EWOULDBLOCK);
2230                                 }
2231                         } else if ((len + m->m_len) == slp->ns_reclen) {
2232                                 om = m;
2233                                 len += m->m_len;
2234                                 m = m->m_next;
2235                                 recm = slp->ns_raw;
2236                                 om->m_next = (struct mbuf *)0;
2237                         } else {
2238                                 om = m;
2239                                 len += m->m_len;
2240                                 m = m->m_next;
2241                         }
2242                 }
2243                 slp->ns_raw = m;
2244                 slp->ns_cc -= len;
2245                 slp->ns_reclen = 0;
2246             } else {
2247                 slp->ns_flag &= ~SLP_GETSTREAM;
2248                 return (0);
2249             }
2250
2251             /*
2252              * Accumulate the fragments into a record.
2253              */
2254             mpp = &slp->ns_frag;
2255             while (*mpp)
2256                 mpp = &((*mpp)->m_next);
2257             *mpp = recm;
2258             if (slp->ns_flag & SLP_LASTFRAG) {
2259                 struct nfsrv_rec *rec;
2260                 rec = malloc(sizeof(struct nfsrv_rec), M_NFSRVDESC, waitflag);
2261                 if (!rec) {
2262                     m_freem(slp->ns_frag);
2263                 } else {
2264                     nfs_realign(&slp->ns_frag, 10 * NFSX_UNSIGNED);
2265                     rec->nr_address = (struct sockaddr *)0;
2266                     rec->nr_packet = slp->ns_frag;
2267                     STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
2268                 }
2269                 slp->ns_frag = (struct mbuf *)0;
2270             }
2271         }
2272 }
2273
2274 /*
2275  * Parse an RPC header.
2276  */
2277 int
2278 nfsrv_dorec(slp, nfsd, ndp)
2279         register struct nfssvc_sock *slp;
2280         struct nfsd *nfsd;
2281         struct nfsrv_descript **ndp;
2282 {
2283         struct nfsrv_rec *rec;
2284         register struct mbuf *m;
2285         struct sockaddr *nam;
2286         register struct nfsrv_descript *nd;
2287         int error;
2288
2289         *ndp = NULL;
2290         if ((slp->ns_flag & SLP_VALID) == 0 || !STAILQ_FIRST(&slp->ns_rec))
2291                 return (ENOBUFS);
2292         rec = STAILQ_FIRST(&slp->ns_rec);
2293         STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
2294         nam = rec->nr_address;
2295         m = rec->nr_packet;
2296         free(rec, M_NFSRVDESC);
2297         MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2298                 M_NFSRVDESC, M_WAITOK);
2299         nd->nd_md = nd->nd_mrep = m;
2300         nd->nd_nam2 = nam;
2301         nd->nd_dpos = mtod(m, caddr_t);
2302         error = nfs_getreq(nd, nfsd, TRUE);
2303         if (error) {
2304                 if (nam) {
2305                         FREE(nam, M_SONAME);
2306                 }
2307                 free((caddr_t)nd, M_NFSRVDESC);
2308                 return (error);
2309         }
2310         *ndp = nd;
2311         nfsd->nfsd_nd = nd;
2312         return (0);
2313 }
2314
2315 /*
2316  * Search for a sleeping nfsd and wake it up.
2317  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2318  * running nfsds will go look for the work in the nfssvc_sock list.
2319  */
2320 void
2321 nfsrv_wakenfsd(slp)
2322         struct nfssvc_sock *slp;
2323 {
2324         register struct nfsd *nd;
2325
2326         if ((slp->ns_flag & SLP_VALID) == 0)
2327                 return;
2328         for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) {
2329                 if (nd->nfsd_flag & NFSD_WAITING) {
2330                         nd->nfsd_flag &= ~NFSD_WAITING;
2331                         if (nd->nfsd_slp)
2332                                 panic("nfsd wakeup");
2333                         slp->ns_sref++;
2334                         nd->nfsd_slp = slp;
2335                         wakeup((caddr_t)nd);
2336                         return;
2337                 }
2338         }
2339         slp->ns_flag |= SLP_DOREC;
2340         nfsd_head_flag |= NFSD_CHECKSLP;
2341 }
2342 #endif /* NFS_NOSERVER */