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  * $DragonFly: src/sys/vfs/nfs/nfs_socket.c,v 1.11 2004/02/10 07:28:41 dillon Exp $
  39  */
  40
  41 /*
  42  * Socket operations for use by nfs
  43  */
  44
  45 #include <sys/param.h>
  46 #include <sys/systm.h>
  47 #include <sys/proc.h>
  48 #include <sys/malloc.h>
  49 #include <sys/mount.h>
  50 #include <sys/kernel.h>
  51 #include <sys/mbuf.h>
  52 #include <sys/vnode.h>
  53 #include <sys/protosw.h>
  54 #include <sys/socket.h>
  55 #include <sys/socketvar.h>
  56 #include <sys/syslog.h>
  57 #include <sys/tprintf.h>
  58 #include <sys/sysctl.h>
  59 #include <sys/signalvar.h>
  60
  61 #include <netinet/in.h>
  62 #include <netinet/tcp.h>
  63
  64 #include "rpcv2.h"
  65 #include "nfsproto.h"
  66 #include "nfs.h"
  67 #include "xdr_subs.h"
  68 #include "nfsm_subs.h"
  69 #include "nfsmount.h"
  70 #include "nfsnode.h"
  71 #include "nfsrtt.h"
  72 #include "nqnfs.h"
  73
  74 #define TRUE    1
  75 #define FALSE   0
  76
  77 /*
  78  * Estimate rto for an nfs rpc sent via. an unreliable datagram.
  79  * Use the mean and mean deviation of rtt for the appropriate type of rpc
  80  * for the frequent rpcs and a default for the others.
  81  * The justification for doing "other" this way is that these rpcs
  82  * happen so infrequently that timer est. would probably be stale.
  83  * Also, since many of these rpcs are
  84  * non-idempotent, a conservative timeout is desired.
  85  * getattr, lookup - A+2D
  86  * read, write     - A+4D
  87  * other           - nm_timeo
  88  */
  89 #define NFS_RTO(n, t) \
  90         ((t) == 0 ? (n)->nm_timeo : \
  91          ((t) < 3 ? \
  92           (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
  93           ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
  94 #define NFS_SRTT(r)     (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
  95 #define NFS_SDRTT(r)    (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
  96 /*
  97  * External data, mostly RPC constants in XDR form
  98  */
  99 extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers,
 100         rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr,
 101         rpc_auth_kerb;
 102 extern u_int32_t nfs_prog, nqnfs_prog;
 103 extern time_t nqnfsstarttime;
 104 extern struct nfsstats nfsstats;
 105 extern int nfsv3_procid[NFS_NPROCS];
 106 extern int nfs_ticks;
 107
 108 /*
 109  * Defines which timer to use for the procnum.
 110  * 0 - default
 111  * 1 - getattr
 112  * 2 - lookup
 113  * 3 - read
 114  * 4 - write
 115  */
 116 static int proct[NFS_NPROCS] = {
 117         0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0,
 118         0, 0, 0,
 119 };
 120
 121 static int nfs_realign_test;
 122 static int nfs_realign_count;
 123 static int nfs_bufpackets = 4;
 124
 125 SYSCTL_DECL(_vfs_nfs);
 126
 127 SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_test, CTLFLAG_RW, &nfs_realign_test, 0, "");
 128 SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_count, CTLFLAG_RW, &nfs_realign_count, 0, "");
 129 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0, "");
 130
 131
 132 /*
 133  * There is a congestion window for outstanding rpcs maintained per mount
 134  * point. The cwnd size is adjusted in roughly the way that:
 135  * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
 136  * SIGCOMM '88". ACM, August 1988.
 137  * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
 138  * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
 139  * of rpcs is in progress.
 140  * (The sent count and cwnd are scaled for integer arith.)
 141  * Variants of "slow start" were tried and were found to be too much of a
 142  * performance hit (ave. rtt 3 times larger),
 143  * I suspect due to the large rtt that nfs rpcs have.
 144  */
 145 #define NFS_CWNDSCALE   256
 146 #define NFS_MAXCWND     (NFS_CWNDSCALE * 32)
 147 static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
 148 int nfsrtton = 0;
 149 struct nfsrtt nfsrtt;
 150 struct callout_handle   nfs_timer_handle;
 151
 152 static int      nfs_msg (struct thread *,char *,char *);
 153 static int      nfs_rcvlock (struct nfsreq *);
 154 static void     nfs_rcvunlock (struct nfsreq *);
 155 static void     nfs_realign (struct mbuf **pm, int hsiz);
 156 static int      nfs_receive (struct nfsreq *rep, struct sockaddr **aname,
 157                                  struct mbuf **mp);
 158 static void     nfs_softterm (struct nfsreq *rep);
 159 static int      nfs_reconnect (struct nfsreq *rep);
 160 #ifndef NFS_NOSERVER
 161 static int      nfsrv_getstream (struct nfssvc_sock *,int);
 162
 163 int (*nfsrv3_procs[NFS_NPROCS]) (struct nfsrv_descript *nd,
 164                                     struct nfssvc_sock *slp,
 165                                     struct thread *td,
 166                                     struct mbuf **mreqp) = {
 167         nfsrv_null,
 168         nfsrv_getattr,
 169         nfsrv_setattr,
 170         nfsrv_lookup,
 171         nfsrv3_access,
 172         nfsrv_readlink,
 173         nfsrv_read,
 174         nfsrv_write,
 175         nfsrv_create,
 176         nfsrv_mkdir,
 177         nfsrv_symlink,
 178         nfsrv_mknod,
 179         nfsrv_remove,
 180         nfsrv_rmdir,
 181         nfsrv_rename,
 182         nfsrv_link,
 183         nfsrv_readdir,
 184         nfsrv_readdirplus,
 185         nfsrv_statfs,
 186         nfsrv_fsinfo,
 187         nfsrv_pathconf,
 188         nfsrv_commit,
 189         nqnfsrv_getlease,
 190         nqnfsrv_vacated,
 191         nfsrv_noop,
 192         nfsrv_noop
 193 };
 194 #endif /* NFS_NOSERVER */
 195
 196 /*
 197  * Initialize sockets and congestion for a new NFS connection.
 198  * We do not free the sockaddr if error.
 199  */
 200 int
 201 nfs_connect(struct nfsmount *nmp, struct nfsreq *rep)
 202 {
 203         struct socket *so;
 204         int s, error, rcvreserve, sndreserve;
 205         int pktscale;
 206         struct sockaddr *saddr;
 207         struct sockaddr_in *sin;
 208         struct thread *td = &thread0; /* only used for socreate and sobind */
 209
 210         nmp->nm_so = (struct socket *)0;
 211         saddr = nmp->nm_nam;
 212         error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype,
 213                 nmp->nm_soproto, td);
 214         if (error)
 215                 goto bad;
 216         so = nmp->nm_so;
 217         nmp->nm_soflags = so->so_proto->pr_flags;
 218
 219         /*
 220          * Some servers require that the client port be a reserved port number.
 221          */
 222         if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
 223                 struct sockopt sopt;
 224                 int ip;
 225                 struct sockaddr_in ssin;
 226
 227                 bzero(&sopt, sizeof sopt);
 228                 ip = IP_PORTRANGE_LOW;
 229                 sopt.sopt_dir = SOPT_SET;
 230                 sopt.sopt_level = IPPROTO_IP;
 231                 sopt.sopt_name = IP_PORTRANGE;
 232                 sopt.sopt_val = (void *)&ip;
 233                 sopt.sopt_valsize = sizeof(ip);
 234                 sopt.sopt_td = NULL;
 235                 error = sosetopt(so, &sopt);
 236                 if (error)
 237                         goto bad;
 238                 bzero(&ssin, sizeof ssin);
 239                 sin = &ssin;
 240                 sin->sin_len = sizeof (struct sockaddr_in);
 241                 sin->sin_family = AF_INET;
 242                 sin->sin_addr.s_addr = INADDR_ANY;
 243                 sin->sin_port = htons(0);
 244                 error = sobind(so, (struct sockaddr *)sin, td);
 245                 if (error)
 246                         goto bad;
 247                 bzero(&sopt, sizeof sopt);
 248                 ip = IP_PORTRANGE_DEFAULT;
 249                 sopt.sopt_dir = SOPT_SET;
 250                 sopt.sopt_level = IPPROTO_IP;
 251                 sopt.sopt_name = IP_PORTRANGE;
 252                 sopt.sopt_val = (void *)&ip;
 253                 sopt.sopt_valsize = sizeof(ip);
 254                 sopt.sopt_td = NULL;
 255                 error = sosetopt(so, &sopt);
 256                 if (error)
 257                         goto bad;
 258         }
 259
 260         /*
 261          * Protocols that do not require connections may be optionally left
 262          * unconnected for servers that reply from a port other than NFS_PORT.
 263          */
 264         if (nmp->nm_flag & NFSMNT_NOCONN) {
 265                 if (nmp->nm_soflags & PR_CONNREQUIRED) {
 266                         error = ENOTCONN;
 267                         goto bad;
 268                 }
 269         } else {
 270                 error = soconnect(so, nmp->nm_nam, td);
 271                 if (error)
 272                         goto bad;
 273
 274                 /*
 275                  * Wait for the connection to complete. Cribbed from the
 276                  * connect system call but with the wait timing out so
 277                  * that interruptible mounts don't hang here for a long time.
 278                  */
 279                 s = splnet();
 280                 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
 281                         (void) tsleep((caddr_t)&so->so_timeo, 0,
 282                                 "nfscon", 2 * hz);
 283                         if ((so->so_state & SS_ISCONNECTING) &&
 284                             so->so_error == 0 && rep &&
 285                             (error = nfs_sigintr(nmp, rep, rep->r_td)) != 0){
 286                                 so->so_state &= ~SS_ISCONNECTING;
 287                                 splx(s);
 288                                 goto bad;
 289                         }
 290                 }
 291                 if (so->so_error) {
 292                         error = so->so_error;
 293                         so->so_error = 0;
 294                         splx(s);
 295                         goto bad;
 296                 }
 297                 splx(s);
 298         }
 299         so->so_rcv.sb_timeo = (5 * hz);
 300         so->so_snd.sb_timeo = (5 * hz);
 301
 302         /*
 303          * Get buffer reservation size from sysctl, but impose reasonable
 304          * limits.
 305          */
 306         pktscale = nfs_bufpackets;
 307         if (pktscale < 2)
 308                 pktscale = 2;
 309         if (pktscale > 64)
 310                 pktscale = 64;
 311
 312         if (nmp->nm_sotype == SOCK_DGRAM) {
 313                 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
 314                 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
 315                     NFS_MAXPKTHDR) * pktscale;
 316         } else if (nmp->nm_sotype == SOCK_SEQPACKET) {
 317                 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
 318                 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
 319                     NFS_MAXPKTHDR) * pktscale;
 320         } else {
 321                 if (nmp->nm_sotype != SOCK_STREAM)
 322                         panic("nfscon sotype");
 323                 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
 324                         struct sockopt sopt;
 325                         int val;
 326
 327                         bzero(&sopt, sizeof sopt);
 328                         sopt.sopt_level = SOL_SOCKET;
 329                         sopt.sopt_name = SO_KEEPALIVE;
 330                         sopt.sopt_val = &val;
 331                         sopt.sopt_valsize = sizeof val;
 332                         val = 1;
 333                         sosetopt(so, &sopt);
 334                 }
 335                 if (so->so_proto->pr_protocol == IPPROTO_TCP) {
 336                         struct sockopt sopt;
 337                         int val;
 338
 339                         bzero(&sopt, sizeof sopt);
 340                         sopt.sopt_level = IPPROTO_TCP;
 341                         sopt.sopt_name = TCP_NODELAY;
 342                         sopt.sopt_val = &val;
 343                         sopt.sopt_valsize = sizeof val;
 344                         val = 1;
 345                         sosetopt(so, &sopt);
 346                 }
 347                 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
 348                     sizeof (u_int32_t)) * pktscale;
 349                 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
 350                     sizeof (u_int32_t)) * pktscale;
 351         }
 352         error = soreserve(so, sndreserve, rcvreserve);
 353         if (error)
 354                 goto bad;
 355         so->so_rcv.sb_flags |= SB_NOINTR;
 356         so->so_snd.sb_flags |= SB_NOINTR;
 357
 358         /* Initialize other non-zero congestion variables */
 359         nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] =
 360                 nmp->nm_srtt[3] = (NFS_TIMEO << 3);
 361         nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
 362                 nmp->nm_sdrtt[3] = 0;
 363         nmp->nm_cwnd = NFS_MAXCWND / 2;     /* Initial send window */
 364         nmp->nm_sent = 0;
 365         nmp->nm_timeouts = 0;
 366         return (0);
 367
 368 bad:
 369         nfs_disconnect(nmp);
 370         return (error);
 371 }
 372
 373 /*
 374  * Reconnect routine:
 375  * Called when a connection is broken on a reliable protocol.
 376  * - clean up the old socket
 377  * - nfs_connect() again
 378  * - set R_MUSTRESEND for all outstanding requests on mount point
 379  * If this fails the mount point is DEAD!
 380  * nb: Must be called with the nfs_sndlock() set on the mount point.
 381  */
 382 static int
 383 nfs_reconnect(rep)
 384         struct nfsreq *rep;
 385 {
 386         struct nfsreq *rp;
 387         struct nfsmount *nmp = rep->r_nmp;
 388         int error;
 389
 390         nfs_disconnect(nmp);
 391         while ((error = nfs_connect(nmp, rep)) != 0) {
 392                 if (error == EINTR || error == ERESTART)
 393                         return (EINTR);
 394                 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
 395         }
 396
 397         /*
 398          * Loop through outstanding request list and fix up all requests
 399          * on old socket.
 400          */
 401         for (rp = nfs_reqq.tqh_first; rp != 0; rp = rp->r_chain.tqe_next) {
 402                 if (rp->r_nmp == nmp)
 403                         rp->r_flags |= R_MUSTRESEND;
 404         }
 405         return (0);
 406 }
 407
 408 /*
 409  * NFS disconnect. Clean up and unlink.
 410  */
 411 void
 412 nfs_disconnect(nmp)
 413         struct nfsmount *nmp;
 414 {
 415         struct socket *so;
 416
 417         if (nmp->nm_so) {
 418                 so = nmp->nm_so;
 419                 nmp->nm_so = (struct socket *)0;
 420                 soshutdown(so, 2);
 421                 soclose(so);
 422         }
 423 }
 424
 425 void
 426 nfs_safedisconnect(nmp)
 427         struct nfsmount *nmp;
 428 {
 429         struct nfsreq dummyreq;
 430
 431         bzero(&dummyreq, sizeof(dummyreq));
 432         dummyreq.r_nmp = nmp;
 433         dummyreq.r_td = NULL;
 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         struct socket *so;
 455         struct sockaddr *nam;
 456         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
 486                     (so, sendnam, 0, top, 0, flags, curthread /*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(struct nfsreq *rep, struct sockaddr **aname, struct mbuf **mp)
 532 {
 533         struct socket *so;
 534         struct uio auio;
 535         struct iovec aio;
 536         struct mbuf *m;
 537         struct mbuf *control;
 538         u_int32_t len;
 539         struct sockaddr **getnam;
 540         int error, sotype, rcvflg;
 541         struct thread *td = curthread;  /* XXX */
 542
 543         /*
 544          * Set up arguments for soreceive()
 545          */
 546         *mp = (struct mbuf *)0;
 547         *aname = (struct sockaddr *)0;
 548         sotype = rep->r_nmp->nm_sotype;
 549
 550         /*
 551          * For reliable protocols, lock against other senders/receivers
 552          * in case a reconnect is necessary.
 553          * For SOCK_STREAM, first get the Record Mark to find out how much
 554          * more there is to get.
 555          * We must lock the socket against other receivers
 556          * until we have an entire rpc request/reply.
 557          */
 558         if (sotype != SOCK_DGRAM) {
 559                 error = nfs_sndlock(rep);
 560                 if (error)
 561                         return (error);
 562 tryagain:
 563                 /*
 564                  * Check for fatal errors and resending request.
 565                  */
 566                 /*
 567                  * Ugh: If a reconnect attempt just happened, nm_so
 568                  * would have changed. NULL indicates a failed
 569                  * attempt that has essentially shut down this
 570                  * mount point.
 571                  */
 572                 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
 573                         nfs_sndunlock(rep);
 574                         return (EINTR);
 575                 }
 576                 so = rep->r_nmp->nm_so;
 577                 if (!so) {
 578                         error = nfs_reconnect(rep);
 579                         if (error) {
 580                                 nfs_sndunlock(rep);
 581                                 return (error);
 582                         }
 583                         goto tryagain;
 584                 }
 585                 while (rep->r_flags & R_MUSTRESEND) {
 586                         m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
 587                         nfsstats.rpcretries++;
 588                         error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
 589                         if (error) {
 590                                 if (error == EINTR || error == ERESTART ||
 591                                     (error = nfs_reconnect(rep)) != 0) {
 592                                         nfs_sndunlock(rep);
 593                                         return (error);
 594                                 }
 595                                 goto tryagain;
 596                         }
 597                 }
 598                 nfs_sndunlock(rep);
 599                 if (sotype == SOCK_STREAM) {
 600                         aio.iov_base = (caddr_t) &len;
 601                         aio.iov_len = sizeof(u_int32_t);
 602                         auio.uio_iov = &aio;
 603                         auio.uio_iovcnt = 1;
 604                         auio.uio_segflg = UIO_SYSSPACE;
 605                         auio.uio_rw = UIO_READ;
 606                         auio.uio_offset = 0;
 607                         auio.uio_resid = sizeof(u_int32_t);
 608                         auio.uio_td = td;
 609                         do {
 610                            rcvflg = MSG_WAITALL;
 611                            error = so->so_proto->pr_usrreqs->pru_soreceive
 612                                    (so, (struct sockaddr **)0, &auio,
 613                                     (struct mbuf **)0, (struct mbuf **)0,
 614                                     &rcvflg);
 615                            if (error == EWOULDBLOCK && rep) {
 616                                 if (rep->r_flags & R_SOFTTERM)
 617                                         return (EINTR);
 618                            }
 619                         } while (error == EWOULDBLOCK);
 620                         if (!error && auio.uio_resid > 0) {
 621                             /*
 622                              * Don't log a 0 byte receive; it means
 623                              * that the socket has been closed, and
 624                              * can happen during normal operation
 625                              * (forcible unmount or Solaris server).
 626                              */
 627                             if (auio.uio_resid != sizeof (u_int32_t))
 628                             log(LOG_INFO,
 629                                  "short receive (%d/%d) from nfs server %s\n",
 630                                  (int)(sizeof(u_int32_t) - auio.uio_resid),
 631                                  (int)sizeof(u_int32_t),
 632                                  rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
 633                             error = EPIPE;
 634                         }
 635                         if (error)
 636                                 goto errout;
 637                         len = ntohl(len) & ~0x80000000;
 638                         /*
 639                          * This is SERIOUS! We are out of sync with the sender
 640                          * and forcing a disconnect/reconnect is all I can do.
 641                          */
 642                         if (len > NFS_MAXPACKET) {
 643                             log(LOG_ERR, "%s (%d) from nfs server %s\n",
 644                                 "impossible packet length",
 645                                 len,
 646                                 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
 647                             error = EFBIG;
 648                             goto errout;
 649                         }
 650                         auio.uio_resid = len;
 651                         do {
 652                             rcvflg = MSG_WAITALL;
 653                             error =  so->so_proto->pr_usrreqs->pru_soreceive
 654                                     (so, (struct sockaddr **)0,
 655                                      &auio, mp, (struct mbuf **)0, &rcvflg);
 656                         } while (error == EWOULDBLOCK || error == EINTR ||
 657                                  error == ERESTART);
 658                         if (!error && auio.uio_resid > 0) {
 659                             if (len != auio.uio_resid)
 660                             log(LOG_INFO,
 661                                 "short receive (%d/%d) from nfs server %s\n",
 662                                 len - auio.uio_resid, len,
 663                                 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
 664                             error = EPIPE;
 665                         }
 666                 } else {
 667                         /*
 668                          * NB: Since uio_resid is big, MSG_WAITALL is ignored
 669                          * and soreceive() will return when it has either a
 670                          * control msg or a data msg.
 671                          * We have no use for control msg., but must grab them
 672                          * and then throw them away so we know what is going
 673                          * on.
 674                          */
 675                         auio.uio_resid = len = 100000000; /* Anything Big */
 676                         auio.uio_td = td;
 677                         do {
 678                             rcvflg = 0;
 679                             error =  so->so_proto->pr_usrreqs->pru_soreceive
 680                                     (so, (struct sockaddr **)0,
 681                                 &auio, mp, &control, &rcvflg);
 682                             if (control)
 683                                 m_freem(control);
 684                             if (error == EWOULDBLOCK && rep) {
 685                                 if (rep->r_flags & R_SOFTTERM)
 686                                         return (EINTR);
 687                             }
 688                         } while (error == EWOULDBLOCK ||
 689                                  (!error && *mp == NULL && control));
 690                         if ((rcvflg & MSG_EOR) == 0)
 691                                 printf("Egad!!\n");
 692                         if (!error && *mp == NULL)
 693                                 error = EPIPE;
 694                         len -= auio.uio_resid;
 695                 }
 696 errout:
 697                 if (error && error != EINTR && error != ERESTART) {
 698                         m_freem(*mp);
 699                         *mp = (struct mbuf *)0;
 700                         if (error != EPIPE)
 701                                 log(LOG_INFO,
 702                                     "receive error %d from nfs server %s\n",
 703                                     error,
 704                                  rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
 705                         error = nfs_sndlock(rep);
 706                         if (!error) {
 707                                 error = nfs_reconnect(rep);
 708                                 if (!error)
 709                                         goto tryagain;
 710                                 else
 711                                         nfs_sndunlock(rep);
 712                         }
 713                 }
 714         } else {
 715                 if ((so = rep->r_nmp->nm_so) == NULL)
 716                         return (EACCES);
 717                 if (so->so_state & SS_ISCONNECTED)
 718                         getnam = (struct sockaddr **)0;
 719                 else
 720                         getnam = aname;
 721                 auio.uio_resid = len = 1000000;
 722                 auio.uio_td = td;
 723                 do {
 724                         rcvflg = 0;
 725                         error =  so->so_proto->pr_usrreqs->pru_soreceive
 726                                 (so, getnam, &auio, mp,
 727                                 (struct mbuf **)0, &rcvflg);
 728                         if (error == EWOULDBLOCK &&
 729                             (rep->r_flags & R_SOFTTERM))
 730                                 return (EINTR);
 731                 } while (error == EWOULDBLOCK);
 732                 len -= auio.uio_resid;
 733         }
 734         if (error) {
 735                 m_freem(*mp);
 736                 *mp = (struct mbuf *)0;
 737         }
 738         /*
 739          * Search for any mbufs that are not a multiple of 4 bytes long
 740          * or with m_data not longword aligned.
 741          * These could cause pointer alignment problems, so copy them to
 742          * well aligned mbufs.
 743          */
 744         nfs_realign(mp, 5 * NFSX_UNSIGNED);
 745         return (error);
 746 }
 747
 748 /*
 749  * Implement receipt of reply on a socket.
 750  * We must search through the list of received datagrams matching them
 751  * with outstanding requests using the xid, until ours is found.
 752  */
 753 /* ARGSUSED */
 754 int
 755 nfs_reply(myrep)
 756         struct nfsreq *myrep;
 757 {
 758         struct nfsreq *rep;
 759         struct nfsmount *nmp = myrep->r_nmp;
 760         int32_t t1;
 761         struct mbuf *mrep, *md;
 762         struct sockaddr *nam;
 763         u_int32_t rxid, *tl;
 764         caddr_t dpos, cp2;
 765         int error;
 766
 767         /*
 768          * Loop around until we get our own reply
 769          */
 770         for (;;) {
 771                 /*
 772                  * Lock against other receivers so that I don't get stuck in
 773                  * sbwait() after someone else has received my reply for me.
 774                  * Also necessary for connection based protocols to avoid
 775                  * race conditions during a reconnect.
 776                  * If nfs_rcvlock() returns EALREADY, that means that
 777                  * the reply has already been recieved by another
 778                  * process and we can return immediately.  In this
 779                  * case, the lock is not taken to avoid races with
 780                  * other processes.
 781                  */
 782                 error = nfs_rcvlock(myrep);
 783                 if (error == EALREADY)
 784                         return (0);
 785                 if (error)
 786                         return (error);
 787                 /*
 788                  * Get the next Rpc reply off the socket
 789                  */
 790                 error = nfs_receive(myrep, &nam, &mrep);
 791                 nfs_rcvunlock(myrep);
 792                 if (error) {
 793
 794                         /*
 795                          * Ignore routing errors on connectionless protocols??
 796                          */
 797                         if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) {
 798                                 nmp->nm_so->so_error = 0;
 799                                 if (myrep->r_flags & R_GETONEREP)
 800                                         return (0);
 801                                 continue;
 802                         }
 803                         return (error);
 804                 }
 805                 if (nam)
 806                         FREE(nam, M_SONAME);
 807
 808                 /*
 809                  * Get the xid and check that it is an rpc reply
 810                  */
 811                 md = mrep;
 812                 dpos = mtod(md, caddr_t);
 813                 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED);
 814                 rxid = *tl++;
 815                 if (*tl != rpc_reply) {
 816 #ifndef NFS_NOSERVER
 817                         if (nmp->nm_flag & NFSMNT_NQNFS) {
 818                                 if (nqnfs_callback(nmp, mrep, md, dpos))
 819                                         nfsstats.rpcinvalid++;
 820                         } else {
 821                                 nfsstats.rpcinvalid++;
 822                                 m_freem(mrep);
 823                         }
 824 #else
 825                         nfsstats.rpcinvalid++;
 826                         m_freem(mrep);
 827 #endif
 828 nfsmout:
 829                         if (myrep->r_flags & R_GETONEREP)
 830                                 return (0);
 831                         continue;
 832                 }
 833
 834                 /*
 835                  * Loop through the request list to match up the reply
 836                  * Iff no match, just drop the datagram
 837                  */
 838                 for (rep = nfs_reqq.tqh_first; rep != 0;
 839                     rep = rep->r_chain.tqe_next) {
 840                         if (rep->r_mrep == NULL && rxid == rep->r_xid) {
 841                                 /* Found it.. */
 842                                 rep->r_mrep = mrep;
 843                                 rep->r_md = md;
 844                                 rep->r_dpos = dpos;
 845                                 if (nfsrtton) {
 846                                         struct rttl *rt;
 847
 848                                         rt = &nfsrtt.rttl[nfsrtt.pos];
 849                                         rt->proc = rep->r_procnum;
 850                                         rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]);
 851                                         rt->sent = nmp->nm_sent;
 852                                         rt->cwnd = nmp->nm_cwnd;
 853                                         rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1];
 854                                         rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1];
 855                                         rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid;
 856                                         getmicrotime(&rt->tstamp);
 857                                         if (rep->r_flags & R_TIMING)
 858                                                 rt->rtt = rep->r_rtt;
 859                                         else
 860                                                 rt->rtt = 1000000;
 861                                         nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ;
 862                                 }
 863                                 /*
 864                                  * Update congestion window.
 865                                  * Do the additive increase of
 866                                  * one rpc/rtt.
 867                                  */
 868                                 if (nmp->nm_cwnd <= nmp->nm_sent) {
 869                                         nmp->nm_cwnd +=
 870                                            (NFS_CWNDSCALE * NFS_CWNDSCALE +
 871                                            (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
 872                                         if (nmp->nm_cwnd > NFS_MAXCWND)
 873                                                 nmp->nm_cwnd = NFS_MAXCWND;
 874                                 }
 875                                 if (rep->r_flags & R_SENT) {
 876                                         rep->r_flags &= ~R_SENT;
 877                                         nmp->nm_sent -= NFS_CWNDSCALE;
 878                                 }
 879                                 /*
 880                                  * Update rtt using a gain of 0.125 on the mean
 881                                  * and a gain of 0.25 on the deviation.
 882                                  */
 883                                 if (rep->r_flags & R_TIMING) {
 884                                         /*
 885                                          * Since the timer resolution of
 886                                          * NFS_HZ is so course, it can often
 887                                          * result in r_rtt == 0. Since
 888                                          * r_rtt == N means that the actual
 889                                          * rtt is between N+dt and N+2-dt ticks,
 890                                          * add 1.
 891                                          */
 892                                         t1 = rep->r_rtt + 1;
 893                                         t1 -= (NFS_SRTT(rep) >> 3);
 894                                         NFS_SRTT(rep) += t1;
 895                                         if (t1 < 0)
 896                                                 t1 = -t1;
 897                                         t1 -= (NFS_SDRTT(rep) >> 2);
 898                                         NFS_SDRTT(rep) += t1;
 899                                 }
 900                                 nmp->nm_timeouts = 0;
 901                                 break;
 902                         }
 903                 }
 904                 /*
 905                  * If not matched to a request, drop it.
 906                  * If it's mine, get out.
 907                  */
 908                 if (rep == 0) {
 909                         nfsstats.rpcunexpected++;
 910                         m_freem(mrep);
 911                 } else if (rep == myrep) {
 912                         if (rep->r_mrep == NULL)
 913                                 panic("nfsreply nil");
 914                         return (0);
 915                 }
 916                 if (myrep->r_flags & R_GETONEREP)
 917                         return (0);
 918         }
 919 }
 920
 921 /*
 922  * nfs_request - goes something like this
 923  *      - fill in request struct
 924  *      - links it into list
 925  *      - calls nfs_send() for first transmit
 926  *      - calls nfs_receive() to get reply
 927  *      - break down rpc header and return with nfs reply pointed to
 928  *        by mrep or error
 929  * nb: always frees up mreq mbuf list
 930  */
 931 int
 932 nfs_request(vp, mrest, procnum, td, cred, mrp, mdp, dposp)
 933         struct vnode *vp;
 934         struct mbuf *mrest;
 935         int procnum;
 936         struct thread *td;
 937         struct ucred *cred;
 938         struct mbuf **mrp;
 939         struct mbuf **mdp;
 940         caddr_t *dposp;
 941 {
 942         struct mbuf *mrep, *m2;
 943         struct nfsreq *rep;
 944         u_int32_t *tl;
 945         int i;
 946         struct nfsmount *nmp;
 947         struct mbuf *m, *md, *mheadend;
 948         struct nfsnode *np;
 949         char nickv[RPCX_NICKVERF];
 950         time_t reqtime, waituntil;
 951         caddr_t dpos, cp2;
 952         int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type;
 953         int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0;
 954         int verf_len, verf_type;
 955         u_int32_t xid;
 956         u_quad_t frev;
 957         char *auth_str, *verf_str;
 958         NFSKERBKEY_T key;               /* save session key */
 959
 960         /* Reject requests while attempting a forced unmount. */
 961         if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) {
 962                 m_freem(mrest);
 963                 return (ESTALE);
 964         }
 965         nmp = VFSTONFS(vp->v_mount);
 966         MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
 967         rep->r_nmp = nmp;
 968         rep->r_vp = vp;
 969         rep->r_td = td;
 970         rep->r_procnum = procnum;
 971         i = 0;
 972         m = mrest;
 973         while (m) {
 974                 i += m->m_len;
 975                 m = m->m_next;
 976         }
 977         mrest_len = i;
 978
 979         /*
 980          * Get the RPC header with authorization.
 981          */
 982 kerbauth:
 983         verf_str = auth_str = (char *)0;
 984         if (nmp->nm_flag & NFSMNT_KERB) {
 985                 verf_str = nickv;
 986                 verf_len = sizeof (nickv);
 987                 auth_type = RPCAUTH_KERB4;
 988                 bzero((caddr_t)key, sizeof (key));
 989                 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str,
 990                         &auth_len, verf_str, verf_len)) {
 991                         error = nfs_getauth(nmp, rep, cred, &auth_str,
 992                                 &auth_len, verf_str, &verf_len, key);
 993                         if (error) {
 994                                 free((caddr_t)rep, M_NFSREQ);
 995                                 m_freem(mrest);
 996                                 return (error);
 997                         }
 998                 }
 999         } else {
1000                 auth_type = RPCAUTH_UNIX;
1001                 if (cred->cr_ngroups < 1)
1002                         panic("nfsreq nogrps");
1003                 auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ?
1004                         nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) +
1005                         5 * NFSX_UNSIGNED;
1006         }
1007         m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len,
1008              auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid);
1009         if (auth_str)
1010                 free(auth_str, M_TEMP);
1011
1012         /*
1013          * For stream protocols, insert a Sun RPC Record Mark.
1014          */
1015         if (nmp->nm_sotype == SOCK_STREAM) {
1016                 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
1017                 if (m == NULL)
1018                         return (ENOBUFS);
1019                 *mtod(m, u_int32_t *) = htonl(0x80000000 |
1020                          (m->m_pkthdr.len - NFSX_UNSIGNED));
1021         }
1022         rep->r_mreq = m;
1023         rep->r_xid = xid;
1024 tryagain:
1025         if (nmp->nm_flag & NFSMNT_SOFT)
1026                 rep->r_retry = nmp->nm_retry;
1027         else
1028                 rep->r_retry = NFS_MAXREXMIT + 1;       /* past clip limit */
1029         rep->r_rtt = rep->r_rexmit = 0;
1030         if (proct[procnum] > 0)
1031                 rep->r_flags = R_TIMING;
1032         else
1033                 rep->r_flags = 0;
1034         rep->r_mrep = NULL;
1035
1036         /*
1037          * Do the client side RPC.
1038          */
1039         nfsstats.rpcrequests++;
1040         /*
1041          * Chain request into list of outstanding requests. Be sure
1042          * to put it LAST so timer finds oldest requests first.
1043          */
1044         s = splsoftclock();
1045         TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain);
1046
1047         /* Get send time for nqnfs */
1048         reqtime = time_second;
1049
1050         /*
1051          * If backing off another request or avoiding congestion, don't
1052          * send this one now but let timer do it. If not timing a request,
1053          * do it now.
1054          */
1055         if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
1056                 (nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1057                 nmp->nm_sent < nmp->nm_cwnd)) {
1058                 splx(s);
1059                 if (nmp->nm_soflags & PR_CONNREQUIRED)
1060                         error = nfs_sndlock(rep);
1061                 if (!error) {
1062                         m2 = m_copym(m, 0, M_COPYALL, M_WAIT);
1063                         error = nfs_send(nmp->nm_so, nmp->nm_nam, m2, rep);
1064                         if (nmp->nm_soflags & PR_CONNREQUIRED)
1065                                 nfs_sndunlock(rep);
1066                 }
1067                 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) {
1068                         nmp->nm_sent += NFS_CWNDSCALE;
1069                         rep->r_flags |= R_SENT;
1070                 }
1071         } else {
1072                 splx(s);
1073                 rep->r_rtt = -1;
1074         }
1075
1076         /*
1077          * Wait for the reply from our send or the timer's.
1078          */
1079         if (!error || error == EPIPE)
1080                 error = nfs_reply(rep);
1081
1082         /*
1083          * RPC done, unlink the request.
1084          */
1085         s = splsoftclock();
1086         TAILQ_REMOVE(&nfs_reqq, rep, r_chain);
1087         splx(s);
1088
1089         /*
1090          * Decrement the outstanding request count.
1091          */
1092         if (rep->r_flags & R_SENT) {
1093                 rep->r_flags &= ~R_SENT;        /* paranoia */
1094                 nmp->nm_sent -= NFS_CWNDSCALE;
1095         }
1096
1097         /*
1098          * If there was a successful reply and a tprintf msg.
1099          * tprintf a response.
1100          */
1101         if (!error && (rep->r_flags & R_TPRINTFMSG))
1102                 nfs_msg(rep->r_td, nmp->nm_mountp->mnt_stat.f_mntfromname,
1103                     "is alive again");
1104         mrep = rep->r_mrep;
1105         md = rep->r_md;
1106         dpos = rep->r_dpos;
1107         if (error) {
1108                 m_freem(rep->r_mreq);
1109                 free((caddr_t)rep, M_NFSREQ);
1110                 return (error);
1111         }
1112
1113         /*
1114          * break down the rpc header and check if ok
1115          */
1116         nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1117         if (*tl++ == rpc_msgdenied) {
1118                 if (*tl == rpc_mismatch)
1119                         error = EOPNOTSUPP;
1120                 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) {
1121                         if (!failed_auth) {
1122                                 failed_auth++;
1123                                 mheadend->m_next = (struct mbuf *)0;
1124                                 m_freem(mrep);
1125                                 m_freem(rep->r_mreq);
1126                                 goto kerbauth;
1127                         } else
1128                                 error = EAUTH;
1129                 } else
1130                         error = EACCES;
1131                 m_freem(mrep);
1132                 m_freem(rep->r_mreq);
1133                 free((caddr_t)rep, M_NFSREQ);
1134                 return (error);
1135         }
1136
1137         /*
1138          * Grab any Kerberos verifier, otherwise just throw it away.
1139          */
1140         verf_type = fxdr_unsigned(int, *tl++);
1141         i = fxdr_unsigned(int32_t, *tl);
1142         if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) {
1143                 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep);
1144                 if (error)
1145                         goto nfsmout;
1146         } else if (i > 0)
1147                 nfsm_adv(nfsm_rndup(i));
1148         nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1149         /* 0 == ok */
1150         if (*tl == 0) {
1151                 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1152                 if (*tl != 0) {
1153                         error = fxdr_unsigned(int, *tl);
1154                         if ((nmp->nm_flag & NFSMNT_NFSV3) &&
1155                                 error == NFSERR_TRYLATER) {
1156                                 m_freem(mrep);
1157                                 error = 0;
1158                                 waituntil = time_second + trylater_delay;
1159                                 while (time_second < waituntil)
1160                                         (void) tsleep((caddr_t)&lbolt,
1161                                                 0, "nqnfstry", 0);
1162                                 trylater_delay *= nfs_backoff[trylater_cnt];
1163                                 if (trylater_cnt < 7)
1164                                         trylater_cnt++;
1165                                 goto tryagain;
1166                         }
1167
1168                         /*
1169                          * If the File Handle was stale, invalidate the
1170                          * lookup cache, just in case.
1171                          */
1172                         if (error == ESTALE)
1173                                 cache_purge(vp);
1174                         if (nmp->nm_flag & NFSMNT_NFSV3) {
1175                                 *mrp = mrep;
1176                                 *mdp = md;
1177                                 *dposp = dpos;
1178                                 error |= NFSERR_RETERR;
1179                         } else
1180                                 m_freem(mrep);
1181                         m_freem(rep->r_mreq);
1182                         free((caddr_t)rep, M_NFSREQ);
1183                         return (error);
1184                 }
1185
1186                 /*
1187                  * For nqnfs, get any lease in reply
1188                  */
1189                 if (nmp->nm_flag & NFSMNT_NQNFS) {
1190                         nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1191                         if (*tl) {
1192                                 np = VTONFS(vp);
1193                                 nqlflag = fxdr_unsigned(int, *tl);
1194                                 nfsm_dissect(tl, u_int32_t *, 4*NFSX_UNSIGNED);
1195                                 cachable = fxdr_unsigned(int, *tl++);
1196                                 reqtime += fxdr_unsigned(int, *tl++);
1197                                 if (reqtime > time_second) {
1198                                     frev = fxdr_hyper(tl);
1199                                     nqnfs_clientlease(nmp, np, nqlflag,
1200                                         cachable, reqtime, frev);
1201                                 }
1202                         }
1203                 }
1204                 *mrp = mrep;
1205                 *mdp = md;
1206                 *dposp = dpos;
1207                 m_freem(rep->r_mreq);
1208                 FREE((caddr_t)rep, M_NFSREQ);
1209                 return (0);
1210         }
1211         m_freem(mrep);
1212         error = EPROTONOSUPPORT;
1213 nfsmout:
1214         m_freem(rep->r_mreq);
1215         free((caddr_t)rep, M_NFSREQ);
1216         return (error);
1217 }
1218
1219 #ifndef NFS_NOSERVER
1220 /*
1221  * Generate the rpc reply header
1222  * siz arg. is used to decide if adding a cluster is worthwhile
1223  */
1224 int
1225 nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp)
1226         int siz;
1227         struct nfsrv_descript *nd;
1228         struct nfssvc_sock *slp;
1229         int err;
1230         int cache;
1231         u_quad_t *frev;
1232         struct mbuf **mrq;
1233         struct mbuf **mbp;
1234         caddr_t *bposp;
1235 {
1236         u_int32_t *tl;
1237         struct mbuf *mreq;
1238         caddr_t bpos;
1239         struct mbuf *mb, *mb2;
1240
1241         MGETHDR(mreq, M_WAIT, MT_DATA);
1242         mb = mreq;
1243         /*
1244          * If this is a big reply, use a cluster else
1245          * try and leave leading space for the lower level headers.
1246          */
1247         siz += RPC_REPLYSIZ;
1248         if ((max_hdr + siz) >= MINCLSIZE) {
1249                 MCLGET(mreq, M_WAIT);
1250         } else
1251                 mreq->m_data += max_hdr;
1252         tl = mtod(mreq, u_int32_t *);
1253         mreq->m_len = 6 * NFSX_UNSIGNED;
1254         bpos = ((caddr_t)tl) + mreq->m_len;
1255         *tl++ = txdr_unsigned(nd->nd_retxid);
1256         *tl++ = rpc_reply;
1257         if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
1258                 *tl++ = rpc_msgdenied;
1259                 if (err & NFSERR_AUTHERR) {
1260                         *tl++ = rpc_autherr;
1261                         *tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
1262                         mreq->m_len -= NFSX_UNSIGNED;
1263                         bpos -= NFSX_UNSIGNED;
1264                 } else {
1265                         *tl++ = rpc_mismatch;
1266                         *tl++ = txdr_unsigned(RPC_VER2);
1267                         *tl = txdr_unsigned(RPC_VER2);
1268                 }
1269         } else {
1270                 *tl++ = rpc_msgaccepted;
1271
1272                 /*
1273                  * For Kerberos authentication, we must send the nickname
1274                  * verifier back, otherwise just RPCAUTH_NULL.
1275                  */
1276                 if (nd->nd_flag & ND_KERBFULL) {
1277                     struct nfsuid *nuidp;
1278                     struct timeval ktvin, ktvout;
1279
1280                     for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first;
1281                         nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1282                         if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid &&
1283                             (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp),
1284                              &nuidp->nu_haddr, nd->nd_nam2)))
1285                             break;
1286                     }
1287                     if (nuidp) {
1288                         ktvin.tv_sec =
1289                             txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1);
1290                         ktvin.tv_usec =
1291                             txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1292
1293                         /*
1294                          * Encrypt the timestamp in ecb mode using the
1295                          * session key.
1296                          */
1297 #ifdef NFSKERB
1298                         XXX
1299 #endif
1300
1301                         *tl++ = rpc_auth_kerb;
1302                         *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED);
1303                         *tl = ktvout.tv_sec;
1304                         nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1305                         *tl++ = ktvout.tv_usec;
1306                         *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid);
1307                     } else {
1308                         *tl++ = 0;
1309                         *tl++ = 0;
1310                     }
1311                 } else {
1312                         *tl++ = 0;
1313                         *tl++ = 0;
1314                 }
1315                 switch (err) {
1316                 case EPROGUNAVAIL:
1317                         *tl = txdr_unsigned(RPC_PROGUNAVAIL);
1318                         break;
1319                 case EPROGMISMATCH:
1320                         *tl = txdr_unsigned(RPC_PROGMISMATCH);
1321                         nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1322                         if (nd->nd_flag & ND_NQNFS) {
1323                                 *tl++ = txdr_unsigned(3);
1324                                 *tl = txdr_unsigned(3);
1325                         } else {
1326                                 *tl++ = txdr_unsigned(2);
1327                                 *tl = txdr_unsigned(3);
1328                         }
1329                         break;
1330                 case EPROCUNAVAIL:
1331                         *tl = txdr_unsigned(RPC_PROCUNAVAIL);
1332                         break;
1333                 case EBADRPC:
1334                         *tl = txdr_unsigned(RPC_GARBAGE);
1335                         break;
1336                 default:
1337                         *tl = 0;
1338                         if (err != NFSERR_RETVOID) {
1339                                 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1340                                 if (err)
1341                                     *tl = txdr_unsigned(nfsrv_errmap(nd, err));
1342                                 else
1343                                     *tl = 0;
1344                         }
1345                         break;
1346                 };
1347         }
1348
1349         /*
1350          * For nqnfs, piggyback lease as requested.
1351          */
1352         if ((nd->nd_flag & ND_NQNFS) && err == 0) {
1353                 if (nd->nd_flag & ND_LEASE) {
1354                         nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1355                         *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE);
1356                         *tl++ = txdr_unsigned(cache);
1357                         *tl++ = txdr_unsigned(nd->nd_duration);
1358                         txdr_hyper(*frev, tl);
1359                 } else {
1360                         nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1361                         *tl = 0;
1362                 }
1363         }
1364         if (mrq != NULL)
1365             *mrq = mreq;
1366         *mbp = mb;
1367         *bposp = bpos;
1368         if (err != 0 && err != NFSERR_RETVOID)
1369                 nfsstats.srvrpc_errs++;
1370         return (0);
1371 }
1372
1373
1374 #endif /* NFS_NOSERVER */
1375 /*
1376  * Nfs timer routine
1377  * Scan the nfsreq list and retranmit any requests that have timed out
1378  * To avoid retransmission attempts on STREAM sockets (in the future) make
1379  * sure to set the r_retry field to 0 (implies nm_retry == 0).
1380  */
1381 void
1382 nfs_timer(arg)
1383         void *arg;      /* never used */
1384 {
1385         struct nfsreq *rep;
1386         struct mbuf *m;
1387         struct socket *so;
1388         struct nfsmount *nmp;
1389         int timeo;
1390         int s, error;
1391 #ifndef NFS_NOSERVER
1392         static long lasttime = 0;
1393         struct nfssvc_sock *slp;
1394         u_quad_t cur_usec;
1395 #endif /* NFS_NOSERVER */
1396         struct thread *td = &thread0; /* XXX for credentials, will break if sleep */
1397
1398         s = splnet();
1399         for (rep = nfs_reqq.tqh_first; rep != 0; rep = rep->r_chain.tqe_next) {
1400                 nmp = rep->r_nmp;
1401                 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
1402                         continue;
1403                 if (nfs_sigintr(nmp, rep, rep->r_td)) {
1404                         nfs_softterm(rep);
1405                         continue;
1406                 }
1407                 if (rep->r_rtt >= 0) {
1408                         rep->r_rtt++;
1409                         if (nmp->nm_flag & NFSMNT_DUMBTIMR)
1410                                 timeo = nmp->nm_timeo;
1411                         else
1412                                 timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
1413                         if (nmp->nm_timeouts > 0)
1414                                 timeo *= nfs_backoff[nmp->nm_timeouts - 1];
1415                         if (rep->r_rtt <= timeo)
1416                                 continue;
1417                         if (nmp->nm_timeouts < 8)
1418                                 nmp->nm_timeouts++;
1419                 }
1420                 /*
1421                  * Check for server not responding
1422                  */
1423                 if ((rep->r_flags & R_TPRINTFMSG) == 0 &&
1424                      rep->r_rexmit > nmp->nm_deadthresh) {
1425                         nfs_msg(rep->r_td,
1426                             nmp->nm_mountp->mnt_stat.f_mntfromname,
1427                             "not responding");
1428                         rep->r_flags |= R_TPRINTFMSG;
1429                 }
1430                 if (rep->r_rexmit >= rep->r_retry) {    /* too many */
1431                         nfsstats.rpctimeouts++;
1432                         nfs_softterm(rep);
1433                         continue;
1434                 }
1435                 if (nmp->nm_sotype != SOCK_DGRAM) {
1436                         if (++rep->r_rexmit > NFS_MAXREXMIT)
1437                                 rep->r_rexmit = NFS_MAXREXMIT;
1438                         continue;
1439                 }
1440                 if ((so = nmp->nm_so) == NULL)
1441                         continue;
1442
1443                 /*
1444                  * If there is enough space and the window allows..
1445                  *      Resend it
1446                  * Set r_rtt to -1 in case we fail to send it now.
1447                  */
1448                 rep->r_rtt = -1;
1449                 if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
1450                    ((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
1451                     (rep->r_flags & R_SENT) ||
1452                     nmp->nm_sent < nmp->nm_cwnd) &&
1453                    (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){
1454                         if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
1455                             error = (*so->so_proto->pr_usrreqs->pru_send)
1456                                     (so, 0, m, (struct sockaddr *)0,
1457                                      (struct mbuf *)0, td);
1458                         else
1459                             error = (*so->so_proto->pr_usrreqs->pru_send)
1460                                     (so, 0, m, nmp->nm_nam, (struct mbuf *)0,
1461                                      td);
1462                         if (error) {
1463                                 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
1464                                         so->so_error = 0;
1465                         } else {
1466                                 /*
1467                                  * Iff first send, start timing
1468                                  * else turn timing off, backoff timer
1469                                  * and divide congestion window by 2.
1470                                  */
1471                                 if (rep->r_flags & R_SENT) {
1472                                         rep->r_flags &= ~R_TIMING;
1473                                         if (++rep->r_rexmit > NFS_MAXREXMIT)
1474                                                 rep->r_rexmit = NFS_MAXREXMIT;
1475                                         nmp->nm_cwnd >>= 1;
1476                                         if (nmp->nm_cwnd < NFS_CWNDSCALE)
1477                                                 nmp->nm_cwnd = NFS_CWNDSCALE;
1478                                         nfsstats.rpcretries++;
1479                                 } else {
1480                                         rep->r_flags |= R_SENT;
1481                                         nmp->nm_sent += NFS_CWNDSCALE;
1482                                 }
1483                                 rep->r_rtt = 0;
1484                         }
1485                 }
1486         }
1487 #ifndef NFS_NOSERVER
1488         /*
1489          * Call the nqnfs server timer once a second to handle leases.
1490          */
1491         if (lasttime != time_second) {
1492                 lasttime = time_second;
1493                 nqnfs_serverd();
1494         }
1495
1496         /*
1497          * Scan the write gathering queues for writes that need to be
1498          * completed now.
1499          */
1500         cur_usec = nfs_curusec();
1501         for (slp = nfssvc_sockhead.tqh_first; slp != 0;
1502             slp = slp->ns_chain.tqe_next) {
1503             if (slp->ns_tq.lh_first && slp->ns_tq.lh_first->nd_time<=cur_usec)
1504                 nfsrv_wakenfsd(slp);
1505         }
1506 #endif /* NFS_NOSERVER */
1507         splx(s);
1508         nfs_timer_handle = timeout(nfs_timer, (void *)0, nfs_ticks);
1509 }
1510
1511 /*
1512  * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1513  * wait for all requests to complete. This is used by forced unmounts
1514  * to terminate any outstanding RPCs.
1515  */
1516 int
1517 nfs_nmcancelreqs(nmp)
1518         struct nfsmount *nmp;
1519 {
1520         struct nfsreq *req;
1521         int i, s;
1522
1523         s = splnet();
1524         TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
1525                 if (nmp != req->r_nmp || req->r_mrep != NULL ||
1526                     (req->r_flags & R_SOFTTERM))
1527                         continue;
1528                 nfs_softterm(req);
1529         }
1530         splx(s);
1531
1532         for (i = 0; i < 30; i++) {
1533                 s = splnet();
1534                 TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
1535                         if (nmp == req->r_nmp)
1536                                 break;
1537                 }
1538                 splx(s);
1539                 if (req == NULL)
1540                         return (0);
1541                 tsleep(&lbolt, 0, "nfscancel", 0);
1542         }
1543         return (EBUSY);
1544 }
1545
1546 /*
1547  * Flag a request as being about to terminate (due to NFSMNT_INT/NFSMNT_SOFT).
1548  * The nm_send count is decremented now to avoid deadlocks when the process in
1549  * soreceive() hasn't yet managed to send its own request.
1550  */
1551
1552 static void
1553 nfs_softterm(rep)
1554         struct nfsreq *rep;
1555 {
1556         rep->r_flags |= R_SOFTTERM;
1557
1558         if (rep->r_flags & R_SENT) {
1559                 rep->r_nmp->nm_sent -= NFS_CWNDSCALE;
1560                 rep->r_flags &= ~R_SENT;
1561         }
1562 }
1563
1564 /*
1565  * Test for a termination condition pending on the process.
1566  * This is used for NFSMNT_INT mounts.
1567  */
1568 int
1569 nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct thread *td)
1570 {
1571         sigset_t tmpset;
1572         struct proc *p;
1573
1574         if (rep && (rep->r_flags & R_SOFTTERM))
1575                 return (EINTR);
1576         /* Terminate all requests while attempting a forced unmount. */
1577         if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1578                 return (EINTR);
1579         if (!(nmp->nm_flag & NFSMNT_INT))
1580                 return (0);
1581         /* td might be NULL YYY */
1582         if (td == NULL || (p = td->td_proc) == NULL)
1583                 return (0);
1584
1585         tmpset = p->p_siglist;
1586         SIGSETNAND(tmpset, p->p_sigmask);
1587         SIGSETNAND(tmpset, p->p_sigignore);
1588         if (SIGNOTEMPTY(p->p_siglist) && NFSINT_SIGMASK(tmpset))
1589                 return (EINTR);
1590
1591         return (0);
1592 }
1593
1594 /*
1595  * Lock a socket against others.
1596  * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
1597  * and also to avoid race conditions between the processes with nfs requests
1598  * in progress when a reconnect is necessary.
1599  */
1600 int
1601 nfs_sndlock(struct nfsreq *rep)
1602 {
1603         int *statep = &rep->r_nmp->nm_state;
1604         struct thread *td;
1605         int slpflag = 0, slptimeo = 0;
1606
1607         td = rep->r_td;
1608         if (rep->r_nmp->nm_flag & NFSMNT_INT)
1609                 slpflag = PCATCH;
1610         while (*statep & NFSSTA_SNDLOCK) {
1611                 if (nfs_sigintr(rep->r_nmp, rep, td))
1612                         return (EINTR);
1613                 *statep |= NFSSTA_WANTSND;
1614                 (void) tsleep((caddr_t)statep, slpflag,
1615                         "nfsndlck", slptimeo);
1616                 if (slpflag == PCATCH) {
1617                         slpflag = 0;
1618                         slptimeo = 2 * hz;
1619                 }
1620         }
1621         /* Always fail if our request has been cancelled. */
1622         if ((rep->r_flags & R_SOFTTERM))
1623                 return (EINTR);
1624         *statep |= NFSSTA_SNDLOCK;
1625         return (0);
1626 }
1627
1628 /*
1629  * Unlock the stream socket for others.
1630  */
1631 void
1632 nfs_sndunlock(rep)
1633         struct nfsreq *rep;
1634 {
1635         int *statep = &rep->r_nmp->nm_state;
1636
1637         if ((*statep & NFSSTA_SNDLOCK) == 0)
1638                 panic("nfs sndunlock");
1639         *statep &= ~NFSSTA_SNDLOCK;
1640         if (*statep & NFSSTA_WANTSND) {
1641                 *statep &= ~NFSSTA_WANTSND;
1642                 wakeup((caddr_t)statep);
1643         }
1644 }
1645
1646 static int
1647 nfs_rcvlock(rep)
1648         struct nfsreq *rep;
1649 {
1650         int *statep = &rep->r_nmp->nm_state;
1651         int slpflag, slptimeo = 0;
1652
1653         if (rep->r_nmp->nm_flag & NFSMNT_INT)
1654                 slpflag = PCATCH;
1655         else
1656                 slpflag = 0;
1657         while (*statep & NFSSTA_RCVLOCK) {
1658                 if (nfs_sigintr(rep->r_nmp, rep, rep->r_td))
1659                         return (EINTR);
1660                 *statep |= NFSSTA_WANTRCV;
1661                 (void) tsleep((caddr_t)statep, slpflag, "nfsrcvlk", slptimeo);
1662                 /*
1663                  * If our reply was recieved while we were sleeping,
1664                  * then just return without taking the lock to avoid a
1665                  * situation where a single iod could 'capture' the
1666                  * recieve lock.
1667                  */
1668                 if (rep->r_mrep != NULL)
1669                         return (EALREADY);
1670                 if (slpflag == PCATCH) {
1671                         slpflag = 0;
1672                         slptimeo = 2 * hz;
1673                 }
1674         }
1675         *statep |= NFSSTA_RCVLOCK;
1676         return (0);
1677 }
1678
1679 /*
1680  * Unlock the stream socket for others.
1681  */
1682 static void
1683 nfs_rcvunlock(rep)
1684         struct nfsreq *rep;
1685 {
1686         int *statep = &rep->r_nmp->nm_state;
1687
1688         if ((*statep & NFSSTA_RCVLOCK) == 0)
1689                 panic("nfs rcvunlock");
1690         *statep &= ~NFSSTA_RCVLOCK;
1691         if (*statep & NFSSTA_WANTRCV) {
1692                 *statep &= ~NFSSTA_WANTRCV;
1693                 wakeup((caddr_t)statep);
1694         }
1695 }
1696
1697 /*
1698  *      nfs_realign:
1699  *
1700  *      Check for badly aligned mbuf data and realign by copying the unaligned
1701  *      portion of the data into a new mbuf chain and freeing the portions
1702  *      of the old chain that were replaced.
1703  *
1704  *      We cannot simply realign the data within the existing mbuf chain
1705  *      because the underlying buffers may contain other rpc commands and
1706  *      we cannot afford to overwrite them.
1707  *
1708  *      We would prefer to avoid this situation entirely.  The situation does
1709  *      not occur with NFS/UDP and is supposed to only occassionally occur
1710  *      with TCP.  Use vfs.nfs.realign_count and realign_test to check this.
1711  */
1712 static void
1713 nfs_realign(pm, hsiz)
1714         struct mbuf **pm;
1715         int hsiz;
1716 {
1717         struct mbuf *m;
1718         struct mbuf *n = NULL;
1719         int off = 0;
1720
1721         ++nfs_realign_test;
1722
1723         while ((m = *pm) != NULL) {
1724                 if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) {
1725                         MGET(n, M_WAIT, MT_DATA);
1726                         if (m->m_len >= MINCLSIZE) {
1727                                 MCLGET(n, M_WAIT);
1728                         }
1729                         n->m_len = 0;
1730                         break;
1731                 }
1732                 pm = &m->m_next;
1733         }
1734
1735         /*
1736          * If n is non-NULL, loop on m copying data, then replace the
1737          * portion of the chain that had to be realigned.
1738          */
1739         if (n != NULL) {
1740                 ++nfs_realign_count;
1741                 while (m) {
1742                         m_copyback(n, off, m->m_len, mtod(m, caddr_t));
1743                         off += m->m_len;
1744                         m = m->m_next;
1745                 }
1746                 m_freem(*pm);
1747                 *pm = n;
1748         }
1749 }
1750
1751 #ifndef NFS_NOSERVER
1752
1753 /*
1754  * Parse an RPC request
1755  * - verify it
1756  * - fill in the cred struct.
1757  */
1758 int
1759 nfs_getreq(nd, nfsd, has_header)
1760         struct nfsrv_descript *nd;
1761         struct nfsd *nfsd;
1762         int has_header;
1763 {
1764         int len, i;
1765         u_int32_t *tl;
1766         int32_t t1;
1767         struct uio uio;
1768         struct iovec iov;
1769         caddr_t dpos, cp2, cp;
1770         u_int32_t nfsvers, auth_type;
1771         uid_t nickuid;
1772         int error = 0, nqnfs = 0, ticklen;
1773         struct mbuf *mrep, *md;
1774         struct nfsuid *nuidp;
1775         struct timeval tvin, tvout;
1776 #if 0                           /* until encrypted keys are implemented */
1777         NFSKERBKEYSCHED_T keys; /* stores key schedule */
1778 #endif
1779
1780         mrep = nd->nd_mrep;
1781         md = nd->nd_md;
1782         dpos = nd->nd_dpos;
1783         if (has_header) {
1784                 nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED);
1785                 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
1786                 if (*tl++ != rpc_call) {
1787                         m_freem(mrep);
1788                         return (EBADRPC);
1789                 }
1790         } else
1791                 nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
1792         nd->nd_repstat = 0;
1793         nd->nd_flag = 0;
1794         if (*tl++ != rpc_vers) {
1795                 nd->nd_repstat = ERPCMISMATCH;
1796                 nd->nd_procnum = NFSPROC_NOOP;
1797                 return (0);
1798         }
1799         if (*tl != nfs_prog) {
1800                 if (*tl == nqnfs_prog)
1801                         nqnfs++;
1802                 else {
1803                         nd->nd_repstat = EPROGUNAVAIL;
1804                         nd->nd_procnum = NFSPROC_NOOP;
1805                         return (0);
1806                 }
1807         }
1808         tl++;
1809         nfsvers = fxdr_unsigned(u_int32_t, *tl++);
1810         if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) ||
1811                 (nfsvers != NQNFS_VER3 && nqnfs)) {
1812                 nd->nd_repstat = EPROGMISMATCH;
1813                 nd->nd_procnum = NFSPROC_NOOP;
1814                 return (0);
1815         }
1816         if (nqnfs)
1817                 nd->nd_flag = (ND_NFSV3 | ND_NQNFS);
1818         else if (nfsvers == NFS_VER3)
1819                 nd->nd_flag = ND_NFSV3;
1820         nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
1821         if (nd->nd_procnum == NFSPROC_NULL)
1822                 return (0);
1823         if (nd->nd_procnum >= NFS_NPROCS ||
1824                 (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) ||
1825                 (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) {
1826                 nd->nd_repstat = EPROCUNAVAIL;
1827                 nd->nd_procnum = NFSPROC_NOOP;
1828                 return (0);
1829         }
1830         if ((nd->nd_flag & ND_NFSV3) == 0)
1831                 nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
1832         auth_type = *tl++;
1833         len = fxdr_unsigned(int, *tl++);
1834         if (len < 0 || len > RPCAUTH_MAXSIZ) {
1835                 m_freem(mrep);
1836                 return (EBADRPC);
1837         }
1838
1839         nd->nd_flag &= ~ND_KERBAUTH;
1840         /*
1841          * Handle auth_unix or auth_kerb.
1842          */
1843         if (auth_type == rpc_auth_unix) {
1844                 len = fxdr_unsigned(int, *++tl);
1845                 if (len < 0 || len > NFS_MAXNAMLEN) {
1846                         m_freem(mrep);
1847                         return (EBADRPC);
1848                 }
1849                 nfsm_adv(nfsm_rndup(len));
1850                 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1851                 bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred));
1852                 nd->nd_cr.cr_ref = 1;
1853                 nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
1854                 nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
1855                 len = fxdr_unsigned(int, *tl);
1856                 if (len < 0 || len > RPCAUTH_UNIXGIDS) {
1857                         m_freem(mrep);
1858                         return (EBADRPC);
1859                 }
1860                 nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED);
1861                 for (i = 1; i <= len; i++)
1862                     if (i < NGROUPS)
1863                         nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
1864                     else
1865                         tl++;
1866                 nd->nd_cr.cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1);
1867                 if (nd->nd_cr.cr_ngroups > 1)
1868                     nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
1869                 len = fxdr_unsigned(int, *++tl);
1870                 if (len < 0 || len > RPCAUTH_MAXSIZ) {
1871                         m_freem(mrep);
1872                         return (EBADRPC);
1873                 }
1874                 if (len > 0)
1875                         nfsm_adv(nfsm_rndup(len));
1876         } else if (auth_type == rpc_auth_kerb) {
1877                 switch (fxdr_unsigned(int, *tl++)) {
1878                 case RPCAKN_FULLNAME:
1879                         ticklen = fxdr_unsigned(int, *tl);
1880                         *((u_int32_t *)nfsd->nfsd_authstr) = *tl;
1881                         uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED;
1882                         nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED;
1883                         if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
1884                                 m_freem(mrep);
1885                                 return (EBADRPC);
1886                         }
1887                         uio.uio_offset = 0;
1888                         uio.uio_iov = &iov;
1889                         uio.uio_iovcnt = 1;
1890                         uio.uio_segflg = UIO_SYSSPACE;
1891                         iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4];
1892                         iov.iov_len = RPCAUTH_MAXSIZ - 4;
1893                         nfsm_mtouio(&uio, uio.uio_resid);
1894                         nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1895                         if (*tl++ != rpc_auth_kerb ||
1896                                 fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) {
1897                                 printf("Bad kerb verifier\n");
1898                                 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1899                                 nd->nd_procnum = NFSPROC_NOOP;
1900                                 return (0);
1901                         }
1902                         nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED);
1903                         tl = (u_int32_t *)cp;
1904                         if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) {
1905                                 printf("Not fullname kerb verifier\n");
1906                                 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1907                                 nd->nd_procnum = NFSPROC_NOOP;
1908                                 return (0);
1909                         }
1910                         cp += NFSX_UNSIGNED;
1911                         bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED);
1912                         nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED;
1913                         nd->nd_flag |= ND_KERBFULL;
1914                         nfsd->nfsd_flag |= NFSD_NEEDAUTH;
1915                         break;
1916                 case RPCAKN_NICKNAME:
1917                         if (len != 2 * NFSX_UNSIGNED) {
1918                                 printf("Kerb nickname short\n");
1919                                 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED);
1920                                 nd->nd_procnum = NFSPROC_NOOP;
1921                                 return (0);
1922                         }
1923                         nickuid = fxdr_unsigned(uid_t, *tl);
1924                         nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1925                         if (*tl++ != rpc_auth_kerb ||
1926                                 fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) {
1927                                 printf("Kerb nick verifier bad\n");
1928                                 nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF);
1929                                 nd->nd_procnum = NFSPROC_NOOP;
1930                                 return (0);
1931                         }
1932                         nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1933                         tvin.tv_sec = *tl++;
1934                         tvin.tv_usec = *tl;
1935
1936                         for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first;
1937                             nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1938                                 if (nuidp->nu_cr.cr_uid == nickuid &&
1939                                     (!nd->nd_nam2 ||
1940                                      netaddr_match(NU_NETFAM(nuidp),
1941                                       &nuidp->nu_haddr, nd->nd_nam2)))
1942                                         break;
1943                         }
1944                         if (!nuidp) {
1945                                 nd->nd_repstat =
1946                                         (NFSERR_AUTHERR|AUTH_REJECTCRED);
1947                                 nd->nd_procnum = NFSPROC_NOOP;
1948                                 return (0);
1949                         }
1950
1951                         /*
1952                          * Now, decrypt the timestamp using the session key
1953                          * and validate it.
1954                          */
1955 #ifdef NFSKERB
1956                         XXX
1957 #endif
1958
1959                         tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec);
1960                         tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec);
1961                         if (nuidp->nu_expire < time_second ||
1962                             nuidp->nu_timestamp.tv_sec > tvout.tv_sec ||
1963                             (nuidp->nu_timestamp.tv_sec == tvout.tv_sec &&
1964                              nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) {
1965                                 nuidp->nu_expire = 0;
1966                                 nd->nd_repstat =
1967                                     (NFSERR_AUTHERR|AUTH_REJECTVERF);
1968                                 nd->nd_procnum = NFSPROC_NOOP;
1969                                 return (0);
1970                         }
1971                         nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr);
1972                         nd->nd_flag |= ND_KERBNICK;
1973                 };
1974         } else {
1975                 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
1976                 nd->nd_procnum = NFSPROC_NOOP;
1977                 return (0);
1978         }
1979
1980         /*
1981          * For nqnfs, get piggybacked lease request.
1982          */
1983         if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
1984                 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1985                 nd->nd_flag |= fxdr_unsigned(int, *tl);
1986                 if (nd->nd_flag & ND_LEASE) {
1987                         nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1988                         nd->nd_duration = fxdr_unsigned(int32_t, *tl);
1989                 } else
1990                         nd->nd_duration = NQ_MINLEASE;
1991         } else
1992                 nd->nd_duration = NQ_MINLEASE;
1993         nd->nd_md = md;
1994         nd->nd_dpos = dpos;
1995         return (0);
1996 nfsmout:
1997         return (error);
1998 }
1999
2000 #endif
2001
2002 /*
2003  * Send a message to the originating process's terminal.  The thread and/or
2004  * process may be NULL.  YYY the thread should not be NULL but there may
2005  * still be some uio_td's that are still being passed as NULL through to
2006  * nfsm_request().
2007  */
2008 static int
2009 nfs_msg(struct thread *td, char *server, char *msg)
2010 {
2011         tpr_t tpr;
2012
2013         if (td && td->td_proc)
2014                 tpr = tprintf_open(td->td_proc);
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         struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
2036         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_td = 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                                 int mf = (waitflag & M_DONTWAIT) ?
2110                                             M_NOWAIT : M_WAITOK;
2111                                 rec = malloc(sizeof(struct nfsrv_rec),
2112                                              M_NFSRVDESC, mf);
2113                                 if (!rec) {
2114                                         if (nam)
2115                                                 FREE(nam, M_SONAME);
2116                                         m_freem(mp);
2117                                         continue;
2118                                 }
2119                                 nfs_realign(&mp, 10 * NFSX_UNSIGNED);
2120                                 rec->nr_address = nam;
2121                                 rec->nr_packet = mp;
2122                                 STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
2123                         }
2124                         if (error) {
2125                                 if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
2126                                         && error != EWOULDBLOCK) {
2127                                         slp->ns_flag |= SLP_DISCONN;
2128                                         goto dorecs;
2129                                 }
2130                         }
2131                 } while (mp);
2132         }
2133
2134         /*
2135          * Now try and process the request records, non-blocking.
2136          */
2137 dorecs:
2138         if (waitflag == M_DONTWAIT &&
2139                 (STAILQ_FIRST(&slp->ns_rec)
2140                  || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
2141                 nfsrv_wakenfsd(slp);
2142 }
2143
2144 /*
2145  * Try and extract an RPC request from the mbuf data list received on a
2146  * stream socket. The "waitflag" argument indicates whether or not it
2147  * can sleep.
2148  */
2149 static int
2150 nfsrv_getstream(slp, waitflag)
2151         struct nfssvc_sock *slp;
2152         int waitflag;
2153 {
2154         struct mbuf *m, **mpp;
2155         char *cp1, *cp2;
2156         int len;
2157         struct mbuf *om, *m2, *recm;
2158         u_int32_t recmark;
2159
2160         if (slp->ns_flag & SLP_GETSTREAM)
2161                 panic("nfs getstream");
2162         slp->ns_flag |= SLP_GETSTREAM;
2163         for (;;) {
2164             if (slp->ns_reclen == 0) {
2165                 if (slp->ns_cc < NFSX_UNSIGNED) {
2166                         slp->ns_flag &= ~SLP_GETSTREAM;
2167                         return (0);
2168                 }
2169                 m = slp->ns_raw;
2170                 if (m->m_len >= NFSX_UNSIGNED) {
2171                         bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
2172                         m->m_data += NFSX_UNSIGNED;
2173                         m->m_len -= NFSX_UNSIGNED;
2174                 } else {
2175                         cp1 = (caddr_t)&recmark;
2176                         cp2 = mtod(m, caddr_t);
2177                         while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
2178                                 while (m->m_len == 0) {
2179                                         m = m->m_next;
2180                                         cp2 = mtod(m, caddr_t);
2181                                 }
2182                                 *cp1++ = *cp2++;
2183                                 m->m_data++;
2184                                 m->m_len--;
2185                         }
2186                 }
2187                 slp->ns_cc -= NFSX_UNSIGNED;
2188                 recmark = ntohl(recmark);
2189                 slp->ns_reclen = recmark & ~0x80000000;
2190                 if (recmark & 0x80000000)
2191                         slp->ns_flag |= SLP_LASTFRAG;
2192                 else
2193                         slp->ns_flag &= ~SLP_LASTFRAG;
2194                 if (slp->ns_reclen > NFS_MAXPACKET) {
2195                         slp->ns_flag &= ~SLP_GETSTREAM;
2196                         return (EPERM);
2197                 }
2198             }
2199
2200             /*
2201              * Now get the record part.
2202              *
2203              * Note that slp->ns_reclen may be 0.  Linux sometimes
2204              * generates 0-length RPCs
2205              */
2206             recm = NULL;
2207             if (slp->ns_cc == slp->ns_reclen) {
2208                 recm = slp->ns_raw;
2209                 slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
2210                 slp->ns_cc = slp->ns_reclen = 0;
2211             } else if (slp->ns_cc > slp->ns_reclen) {
2212                 len = 0;
2213                 m = slp->ns_raw;
2214                 om = (struct mbuf *)0;
2215
2216                 while (len < slp->ns_reclen) {
2217                         if ((len + m->m_len) > slp->ns_reclen) {
2218                                 m2 = m_copym(m, 0, slp->ns_reclen - len,
2219                                         waitflag);
2220                                 if (m2) {
2221                                         if (om) {
2222                                                 om->m_next = m2;
2223                                                 recm = slp->ns_raw;
2224                                         } else
2225                                                 recm = m2;
2226                                         m->m_data += slp->ns_reclen - len;
2227                                         m->m_len -= slp->ns_reclen - len;
2228                                         len = slp->ns_reclen;
2229                                 } else {
2230                                         slp->ns_flag &= ~SLP_GETSTREAM;
2231                                         return (EWOULDBLOCK);
2232                                 }
2233                         } else if ((len + m->m_len) == slp->ns_reclen) {
2234                                 om = m;
2235                                 len += m->m_len;
2236                                 m = m->m_next;
2237                                 recm = slp->ns_raw;
2238                                 om->m_next = (struct mbuf *)0;
2239                         } else {
2240                                 om = m;
2241                                 len += m->m_len;
2242                                 m = m->m_next;
2243                         }
2244                 }
2245                 slp->ns_raw = m;
2246                 slp->ns_cc -= len;
2247                 slp->ns_reclen = 0;
2248             } else {
2249                 slp->ns_flag &= ~SLP_GETSTREAM;
2250                 return (0);
2251             }
2252
2253             /*
2254              * Accumulate the fragments into a record.
2255              */
2256             mpp = &slp->ns_frag;
2257             while (*mpp)
2258                 mpp = &((*mpp)->m_next);
2259             *mpp = recm;
2260             if (slp->ns_flag & SLP_LASTFRAG) {
2261                 struct nfsrv_rec *rec;
2262                 int mf = (waitflag & M_DONTWAIT) ? M_NOWAIT : M_WAITOK;
2263                 rec = malloc(sizeof(struct nfsrv_rec), M_NFSRVDESC, mf);
2264                 if (!rec) {
2265                     m_freem(slp->ns_frag);
2266                 } else {
2267                     nfs_realign(&slp->ns_frag, 10 * NFSX_UNSIGNED);
2268                     rec->nr_address = (struct sockaddr *)0;
2269                     rec->nr_packet = slp->ns_frag;
2270                     STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
2271                 }
2272                 slp->ns_frag = (struct mbuf *)0;
2273             }
2274         }
2275 }
2276
2277 /*
2278  * Parse an RPC header.
2279  */
2280 int
2281 nfsrv_dorec(slp, nfsd, ndp)
2282         struct nfssvc_sock *slp;
2283         struct nfsd *nfsd;
2284         struct nfsrv_descript **ndp;
2285 {
2286         struct nfsrv_rec *rec;
2287         struct mbuf *m;
2288         struct sockaddr *nam;
2289         struct nfsrv_descript *nd;
2290         int error;
2291
2292         *ndp = NULL;
2293         if ((slp->ns_flag & SLP_VALID) == 0 || !STAILQ_FIRST(&slp->ns_rec))
2294                 return (ENOBUFS);
2295         rec = STAILQ_FIRST(&slp->ns_rec);
2296         STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
2297         nam = rec->nr_address;
2298         m = rec->nr_packet;
2299         free(rec, M_NFSRVDESC);
2300         MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
2301                 M_NFSRVDESC, M_WAITOK);
2302         nd->nd_md = nd->nd_mrep = m;
2303         nd->nd_nam2 = nam;
2304         nd->nd_dpos = mtod(m, caddr_t);
2305         error = nfs_getreq(nd, nfsd, TRUE);
2306         if (error) {
2307                 if (nam) {
2308                         FREE(nam, M_SONAME);
2309                 }
2310                 free((caddr_t)nd, M_NFSRVDESC);
2311                 return (error);
2312         }
2313         *ndp = nd;
2314         nfsd->nfsd_nd = nd;
2315         return (0);
2316 }
2317
2318 /*
2319  * Search for a sleeping nfsd and wake it up.
2320  * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
2321  * running nfsds will go look for the work in the nfssvc_sock list.
2322  */
2323 void
2324 nfsrv_wakenfsd(slp)
2325         struct nfssvc_sock *slp;
2326 {
2327         struct nfsd *nd;
2328
2329         if ((slp->ns_flag & SLP_VALID) == 0)
2330                 return;
2331         for (nd = nfsd_head.tqh_first; nd != 0; nd = nd->nfsd_chain.tqe_next) {
2332                 if (nd->nfsd_flag & NFSD_WAITING) {
2333                         nd->nfsd_flag &= ~NFSD_WAITING;
2334                         if (nd->nfsd_slp)
2335                                 panic("nfsd wakeup");
2336                         slp->ns_sref++;
2337                         nd->nfsd_slp = slp;
2338                         wakeup((caddr_t)nd);
2339                         return;
2340                 }
2341         }
2342         slp->ns_flag |= SLP_DOREC;
2343         nfsd_head_flag |= NFSD_CHECKSLP;
2344 }
2345 #endif /* NFS_NOSERVER */