sys/netipx/spx_usrreq.c

   1 /*-
   2  * Copyright (c) 1984, 1985, 1986, 1987, 1993
   3  *      The Regents of the University of California.
   4  * Copyright (c) 2004-2006 Robert N. M. Watson
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 4. Neither the name of the University nor the names of its contributors
  16  *    may be used to endorse or promote products derived from this software
  17  *    without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  *
  31  * Copyright (c) 1995, Mike Mitchell
  32  * All rights reserved.
  33  *
  34  * Redistribution and use in source and binary forms, with or without
  35  * modification, are permitted provided that the following conditions
  36  * are met:
  37  * 1. Redistributions of source code must retain the above copyright
  38  *    notice, this list of conditions and the following disclaimer.
  39  * 2. Redistributions in binary form must reproduce the above copyright
  40  *    notice, this list of conditions and the following disclaimer in the
  41  *    documentation and/or other materials provided with the distribution.
  42  * 3. All advertising materials mentioning features or use of this software
  43  *    must display the following acknowledgement:
  44  *      This product includes software developed by the University of
  45  *      California, Berkeley and its contributors.
  46  * 4. Neither the name of the University nor the names of its contributors
  47  *    may be used to endorse or promote products derived from this software
  48  *    without specific prior written permission.
  49  *
  50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  60  * SUCH DAMAGE.
  61  *
  62  *      @(#)spx_usrreq.h
  63  */
  64
  65 #include <sys/cdefs.h>
  66 __FBSDID("$FreeBSD$");
  67
  68 #include <sys/param.h>
  69 #include <sys/lock.h>
  70 #include <sys/malloc.h>
  71 #include <sys/mbuf.h>
  72 #include <sys/mutex.h>
  73 #include <sys/proc.h>
  74 #include <sys/protosw.h>
  75 #include <sys/signalvar.h>
  76 #include <sys/socket.h>
  77 #include <sys/socketvar.h>
  78 #include <sys/sx.h>
  79 #include <sys/systm.h>
  80
  81 #include <net/route.h>
  82 #include <netinet/tcp_fsm.h>
  83
  84 #include <netipx/ipx.h>
  85 #include <netipx/ipx_pcb.h>
  86 #include <netipx/ipx_var.h>
  87 #include <netipx/spx.h>
  88 #include <netipx/spx_debug.h>
  89 #include <netipx/spx_timer.h>
  90 #include <netipx/spx_var.h>
  91
  92 /*
  93  * SPX protocol implementation.
  94  */
  95 static struct   mtx spx_mtx;                    /* Protects only spx_iss. */
  96 static u_short  spx_iss;
  97 static u_short  spx_newchecks[50];
  98 static int      spx_hardnosed;
  99 static int      spx_use_delack = 0;
 100 static int      traceallspxs = 0;
 101 static struct   spx_istat spx_istat;
 102 static int      spxrexmtthresh = 3;
 103
 104 #define SPX_LOCK_INIT() mtx_init(&spx_mtx, "spx_mtx", NULL, MTX_DEF)
 105 #define SPX_LOCK()      mtx_lock(&spx_mtx)
 106 #define SPX_UNLOCK()    mtx_unlock(&spx_mtx)
 107
 108 /* Following was struct spxstat spxstat; */
 109 #ifndef spxstat
 110 #define spxstat spx_istat.newstats
 111 #endif
 112
 113 static const int spx_backoff[SPX_MAXRXTSHIFT+1] =
 114     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
 115
 116 static  void spx_close(struct spxpcb *cb);
 117 static  void spx_disconnect(struct spxpcb *cb);
 118 static  void spx_drop(struct spxpcb *cb, int errno);
 119 static  int spx_output(struct spxpcb *cb, struct mbuf *m0);
 120 static  int spx_reass(struct spxpcb *cb, struct spx *si);
 121 static  void spx_setpersist(struct spxpcb *cb);
 122 static  void spx_template(struct spxpcb *cb);
 123 static  void spx_timers(struct spxpcb *cb, int timer);
 124 static  void spx_usrclosed(struct spxpcb *cb);
 125
 126 static  void spx_usr_abort(struct socket *so);
 127 static  int spx_accept(struct socket *so, struct sockaddr **nam);
 128 static  int spx_attach(struct socket *so, int proto, struct thread *td);
 129 static  int spx_bind(struct socket *so, struct sockaddr *nam, struct thread *td);
 130 static  void spx_usr_close(struct socket *so);
 131 static  int spx_connect(struct socket *so, struct sockaddr *nam,
 132                         struct thread *td);
 133 static  void spx_detach(struct socket *so);
 134 static  void spx_pcbdetach(struct ipxpcb *ipxp);
 135 static  int spx_usr_disconnect(struct socket *so);
 136 static  int spx_listen(struct socket *so, int backlog, struct thread *td);
 137 static  int spx_rcvd(struct socket *so, int flags);
 138 static  int spx_rcvoob(struct socket *so, struct mbuf *m, int flags);
 139 static  int spx_send(struct socket *so, int flags, struct mbuf *m,
 140                      struct sockaddr *addr, struct mbuf *control,
 141                      struct thread *td);
 142 static  int spx_shutdown(struct socket *so);
 143 static  int spx_sp_attach(struct socket *so, int proto, struct thread *td);
 144
 145 struct  pr_usrreqs spx_usrreqs = {
 146         .pru_abort =            spx_usr_abort,
 147         .pru_accept =           spx_accept,
 148         .pru_attach =           spx_attach,
 149         .pru_bind =             spx_bind,
 150         .pru_connect =          spx_connect,
 151         .pru_control =          ipx_control,
 152         .pru_detach =           spx_detach,
 153         .pru_disconnect =       spx_usr_disconnect,
 154         .pru_listen =           spx_listen,
 155         .pru_peeraddr =         ipx_peeraddr,
 156         .pru_rcvd =             spx_rcvd,
 157         .pru_rcvoob =           spx_rcvoob,
 158         .pru_send =             spx_send,
 159         .pru_shutdown =         spx_shutdown,
 160         .pru_sockaddr =         ipx_sockaddr,
 161         .pru_close =            spx_usr_close,
 162 };
 163
 164 struct  pr_usrreqs spx_usrreq_sps = {
 165         .pru_abort =            spx_usr_abort,
 166         .pru_accept =           spx_accept,
 167         .pru_attach =           spx_sp_attach,
 168         .pru_bind =             spx_bind,
 169         .pru_connect =          spx_connect,
 170         .pru_control =          ipx_control,
 171         .pru_detach =           spx_detach,
 172         .pru_disconnect =       spx_usr_disconnect,
 173         .pru_listen =           spx_listen,
 174         .pru_peeraddr =         ipx_peeraddr,
 175         .pru_rcvd =             spx_rcvd,
 176         .pru_rcvoob =           spx_rcvoob,
 177         .pru_send =             spx_send,
 178         .pru_shutdown =         spx_shutdown,
 179         .pru_sockaddr =         ipx_sockaddr,
 180         .pru_close =            spx_usr_close,
 181 };
 182
 183 void
 184 spx_init(void)
 185 {
 186
 187         SPX_LOCK_INIT();
 188         spx_iss = 1; /* WRONG !! should fish it out of TODR */
 189 }
 190
 191 void
 192 spx_input(struct mbuf *m, struct ipxpcb *ipxp)
 193 {
 194         struct spxpcb *cb;
 195         struct spx *si = mtod(m, struct spx *);
 196         struct socket *so;
 197         struct spx spx_savesi;
 198         int dropsocket = 0;
 199         short ostate = 0;
 200
 201         spxstat.spxs_rcvtotal++;
 202         KASSERT(ipxp != NULL, ("spx_input: ipxpcb == NULL"));
 203
 204         /*
 205          * spx_input() assumes that the caller will hold both the pcb list
 206          * lock and also the ipxp lock.  spx_input() will release both before
 207          * returning, and may in fact trade in the ipxp lock for another pcb
 208          * lock following sonewconn().
 209          */
 210         IPX_LIST_LOCK_ASSERT();
 211         IPX_LOCK_ASSERT(ipxp);
 212
 213         cb = ipxtospxpcb(ipxp);
 214         KASSERT(cb != NULL, ("spx_input: cb == NULL"));
 215
 216         if (ipxp->ipxp_flags & IPXP_DROPPED)
 217                 goto drop;
 218
 219         if (m->m_len < sizeof(*si)) {
 220                 if ((m = m_pullup(m, sizeof(*si))) == NULL) {
 221                         IPX_UNLOCK(ipxp);
 222                         IPX_LIST_UNLOCK();
 223                         spxstat.spxs_rcvshort++;
 224                         return;
 225                 }
 226                 si = mtod(m, struct spx *);
 227         }
 228         si->si_seq = ntohs(si->si_seq);
 229         si->si_ack = ntohs(si->si_ack);
 230         si->si_alo = ntohs(si->si_alo);
 231
 232         so = ipxp->ipxp_socket;
 233         KASSERT(so != NULL, ("spx_input: so == NULL"));
 234
 235         if (so->so_options & SO_DEBUG || traceallspxs) {
 236                 ostate = cb->s_state;
 237                 spx_savesi = *si;
 238         }
 239         if (so->so_options & SO_ACCEPTCONN) {
 240                 struct spxpcb *ocb = cb;
 241
 242                 so = sonewconn(so, 0);
 243                 if (so == NULL)
 244                         goto drop;
 245
 246                 /*
 247                  * This is ugly, but ....
 248                  *
 249                  * Mark socket as temporary until we're committed to keeping
 250                  * it.  The code at ``drop'' and ``dropwithreset'' check the
 251                  * flag dropsocket to see if the temporary socket created
 252                  * here should be discarded.  We mark the socket as
 253                  * discardable until we're committed to it below in
 254                  * TCPS_LISTEN.
 255                  *
 256                  * XXXRW: In the new world order of real kernel parallelism,
 257                  * temporarily allocating the socket when we're "not sure"
 258                  * seems like a bad idea, as we might race to remove it if
 259                  * the listen socket is closed...?
 260                  *
 261                  * We drop the lock of the listen socket ipxp, and acquire
 262                  * the lock of the new socket ippx.
 263                  */
 264                 dropsocket++;
 265                 IPX_UNLOCK(ipxp);
 266                 ipxp = (struct ipxpcb *)so->so_pcb;
 267                 IPX_LOCK(ipxp);
 268                 ipxp->ipxp_laddr = si->si_dna;
 269                 cb = ipxtospxpcb(ipxp);
 270                 cb->s_mtu = ocb->s_mtu;         /* preserve sockopts */
 271                 cb->s_flags = ocb->s_flags;     /* preserve sockopts */
 272                 cb->s_flags2 = ocb->s_flags2;   /* preserve sockopts */
 273                 cb->s_state = TCPS_LISTEN;
 274         }
 275         IPX_LOCK_ASSERT(ipxp);
 276
 277         /*
 278          * Packet received on connection.  Reset idle time and keep-alive
 279          * timer.
 280          */
 281         cb->s_idle = 0;
 282         cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
 283
 284         switch (cb->s_state) {
 285         case TCPS_LISTEN:{
 286                 struct sockaddr_ipx *sipx, ssipx;
 287                 struct ipx_addr laddr;
 288
 289                 /*
 290                  * If somebody here was carying on a conversation and went
 291                  * away, and his pen pal thinks he can still talk, we get the
 292                  * misdirected packet.
 293                  */
 294                 if (spx_hardnosed && (si->si_did != 0 || si->si_seq != 0)) {
 295                         spx_istat.gonawy++;
 296                         goto dropwithreset;
 297                 }
 298                 sipx = &ssipx;
 299                 bzero(sipx, sizeof *sipx);
 300                 sipx->sipx_len = sizeof(*sipx);
 301                 sipx->sipx_family = AF_IPX;
 302                 sipx->sipx_addr = si->si_sna;
 303                 laddr = ipxp->ipxp_laddr;
 304                 if (ipx_nullhost(laddr))
 305                         ipxp->ipxp_laddr = si->si_dna;
 306                 if (ipx_pcbconnect(ipxp, (struct sockaddr *)sipx, &thread0)) {
 307                         ipxp->ipxp_laddr = laddr;
 308                         spx_istat.noconn++;
 309                         goto drop;
 310                 }
 311                 spx_template(cb);
 312                 dropsocket = 0;         /* committed to socket */
 313                 cb->s_did = si->si_sid;
 314                 cb->s_rack = si->si_ack;
 315                 cb->s_ralo = si->si_alo;
 316 #define THREEWAYSHAKE
 317 #ifdef THREEWAYSHAKE
 318                 cb->s_state = TCPS_SYN_RECEIVED;
 319                 cb->s_force = 1 + SPXT_KEEP;
 320                 spxstat.spxs_accepts++;
 321                 cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
 322                 }
 323                 break;
 324
 325          case TCPS_SYN_RECEIVED: {
 326                 /*
 327                  * This state means that we have heard a response to our
 328                  * acceptance of their connection.  It is probably logically
 329                  * unnecessary in this implementation.
 330                  */
 331                 if (si->si_did != cb->s_sid) {
 332                         spx_istat.wrncon++;
 333                         goto drop;
 334                 }
 335 #endif
 336                 ipxp->ipxp_fport =  si->si_sport;
 337                 cb->s_timer[SPXT_REXMT] = 0;
 338                 cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
 339                 soisconnected(so);
 340                 cb->s_state = TCPS_ESTABLISHED;
 341                 spxstat.spxs_accepts++;
 342                 }
 343                 break;
 344
 345         case TCPS_SYN_SENT:
 346                 /*
 347                  * This state means that we have gotten a response to our
 348                  * attempt to establish a connection.  We fill in the data
 349                  * from the other side, telling us which port to respond to,
 350                  * instead of the well-known one we might have sent to in the
 351                  * first place.  We also require that this is a response to
 352                  * our connection id.
 353                  */
 354                 if (si->si_did != cb->s_sid) {
 355                         spx_istat.notme++;
 356                         goto drop;
 357                 }
 358                 spxstat.spxs_connects++;
 359                 cb->s_did = si->si_sid;
 360                 cb->s_rack = si->si_ack;
 361                 cb->s_ralo = si->si_alo;
 362                 cb->s_dport = ipxp->ipxp_fport =  si->si_sport;
 363                 cb->s_timer[SPXT_REXMT] = 0;
 364                 cb->s_flags |= SF_ACKNOW;
 365                 soisconnected(so);
 366                 cb->s_state = TCPS_ESTABLISHED;
 367                 /*
 368                  * Use roundtrip time of connection request for initial rtt.
 369                  */
 370                 if (cb->s_rtt) {
 371                         cb->s_srtt = cb->s_rtt << 3;
 372                         cb->s_rttvar = cb->s_rtt << 1;
 373                         SPXT_RANGESET(cb->s_rxtcur,
 374                             ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1,
 375                             SPXTV_MIN, SPXTV_REXMTMAX);
 376                             cb->s_rtt = 0;
 377                 }
 378         }
 379
 380         if (so->so_options & SO_DEBUG || traceallspxs)
 381                 spx_trace(SA_INPUT, (u_char)ostate, cb, &spx_savesi, 0);
 382
 383         m->m_len -= sizeof(struct ipx);
 384         m->m_pkthdr.len -= sizeof(struct ipx);
 385         m->m_data += sizeof(struct ipx);
 386
 387         if (spx_reass(cb, si))
 388                 m_freem(m);
 389         if (cb->s_force || (cb->s_flags & (SF_ACKNOW|SF_WIN|SF_RXT)))
 390                 spx_output(cb, NULL);
 391         cb->s_flags &= ~(SF_WIN|SF_RXT);
 392         IPX_UNLOCK(ipxp);
 393         IPX_LIST_UNLOCK();
 394         return;
 395
 396 dropwithreset:
 397         IPX_LOCK_ASSERT(ipxp);
 398         if (cb == NULL || (cb->s_ipxpcb->ipxp_socket->so_options & SO_DEBUG ||
 399             traceallspxs))
 400                 spx_trace(SA_DROP, (u_char)ostate, cb, &spx_savesi, 0);
 401         IPX_UNLOCK(ipxp);
 402         if (dropsocket) {
 403                 struct socket *head;
 404                 ACCEPT_LOCK();
 405                 KASSERT((so->so_qstate & SQ_INCOMP) != 0,
 406                     ("spx_input: nascent socket not SQ_INCOMP on soabort()"));
 407                 head = so->so_head;
 408                 TAILQ_REMOVE(&head->so_incomp, so, so_list);
 409                 head->so_incqlen--;
 410                 so->so_qstate &= ~SQ_INCOMP;
 411                 so->so_head = NULL;
 412                 ACCEPT_UNLOCK();
 413                 soabort(so);
 414         }
 415         IPX_LIST_UNLOCK();
 416         m_freem(dtom(si));
 417         return;
 418
 419 drop:
 420         IPX_LOCK_ASSERT(ipxp);
 421         if (cb->s_ipxpcb->ipxp_socket->so_options & SO_DEBUG || traceallspxs)
 422                 spx_trace(SA_DROP, (u_char)ostate, cb, &spx_savesi, 0);
 423         IPX_UNLOCK(ipxp);
 424         IPX_LIST_UNLOCK();
 425         m_freem(m);
 426 }
 427
 428 /*
 429  * This is structurally similar to the tcp reassembly routine but its
 430  * function is somewhat different:  It merely queues packets up, and
 431  * suppresses duplicates.
 432  */
 433 static int
 434 spx_reass(struct spxpcb *cb, struct spx *si)
 435 {
 436         struct spx_q *q;
 437         struct mbuf *m;
 438         struct socket *so = cb->s_ipxpcb->ipxp_socket;
 439         char packetp = cb->s_flags & SF_HI;
 440         int incr;
 441         char wakeup = 0;
 442
 443         IPX_LOCK_ASSERT(cb->s_ipxpcb);
 444
 445         if (si == SI(0))
 446                 goto present;
 447         /*
 448          * Update our news from them.
 449          */
 450         if (si->si_cc & SPX_SA)
 451                 cb->s_flags |= (spx_use_delack ? SF_DELACK : SF_ACKNOW);
 452         if (SSEQ_GT(si->si_alo, cb->s_ralo))
 453                 cb->s_flags |= SF_WIN;
 454         if (SSEQ_LEQ(si->si_ack, cb->s_rack)) {
 455                 if ((si->si_cc & SPX_SP) && cb->s_rack != (cb->s_smax + 1)) {
 456                         spxstat.spxs_rcvdupack++;
 457                         /*
 458                          * If this is a completely duplicate ack and other
 459                          * conditions hold, we assume a packet has been
 460                          * dropped and retransmit it exactly as in
 461                          * tcp_input().
 462                          */
 463                         if (si->si_ack != cb->s_rack ||
 464                             si->si_alo != cb->s_ralo)
 465                                 cb->s_dupacks = 0;
 466                         else if (++cb->s_dupacks == spxrexmtthresh) {
 467                                 u_short onxt = cb->s_snxt;
 468                                 int cwnd = cb->s_cwnd;
 469
 470                                 cb->s_snxt = si->si_ack;
 471                                 cb->s_cwnd = CUNIT;
 472                                 cb->s_force = 1 + SPXT_REXMT;
 473                                 spx_output(cb, NULL);
 474                                 cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
 475                                 cb->s_rtt = 0;
 476                                 if (cwnd >= 4 * CUNIT)
 477                                         cb->s_cwnd = cwnd / 2;
 478                                 if (SSEQ_GT(onxt, cb->s_snxt))
 479                                         cb->s_snxt = onxt;
 480                                 return (1);
 481                         }
 482                 } else
 483                         cb->s_dupacks = 0;
 484                 goto update_window;
 485         }
 486         cb->s_dupacks = 0;
 487
 488         /*
 489          * If our correspondent acknowledges data we haven't sent TCP would
 490          * drop the packet after acking.  We'll be a little more permissive.
 491          */
 492         if (SSEQ_GT(si->si_ack, (cb->s_smax + 1))) {
 493                 spxstat.spxs_rcvacktoomuch++;
 494                 si->si_ack = cb->s_smax + 1;
 495         }
 496         spxstat.spxs_rcvackpack++;
 497
 498         /*
 499          * If transmit timer is running and timed sequence number was acked,
 500          * update smoothed round trip time.  See discussion of algorithm in
 501          * tcp_input.c
 502          */
 503         if (cb->s_rtt && SSEQ_GT(si->si_ack, cb->s_rtseq)) {
 504                 spxstat.spxs_rttupdated++;
 505                 if (cb->s_srtt != 0) {
 506                         short delta;
 507                         delta = cb->s_rtt - (cb->s_srtt >> 3);
 508                         if ((cb->s_srtt += delta) <= 0)
 509                                 cb->s_srtt = 1;
 510                         if (delta < 0)
 511                                 delta = -delta;
 512                         delta -= (cb->s_rttvar >> 2);
 513                         if ((cb->s_rttvar += delta) <= 0)
 514                                 cb->s_rttvar = 1;
 515                 } else {
 516                         /*
 517                          * No rtt measurement yet.
 518                          */
 519                         cb->s_srtt = cb->s_rtt << 3;
 520                         cb->s_rttvar = cb->s_rtt << 1;
 521                 }
 522                 cb->s_rtt = 0;
 523                 cb->s_rxtshift = 0;
 524                 SPXT_RANGESET(cb->s_rxtcur,
 525                         ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1,
 526                         SPXTV_MIN, SPXTV_REXMTMAX);
 527         }
 528
 529         /*
 530          * If all outstanding data is acked, stop retransmit timer and
 531          * remember to restart (more output or persist).  If there is more
 532          * data to be acked, restart retransmit timer, using current
 533          * (possibly backed-off) value;
 534          */
 535         if (si->si_ack == cb->s_smax + 1) {
 536                 cb->s_timer[SPXT_REXMT] = 0;
 537                 cb->s_flags |= SF_RXT;
 538         } else if (cb->s_timer[SPXT_PERSIST] == 0)
 539                 cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
 540
 541         /*
 542          * When new data is acked, open the congestion window.  If the window
 543          * gives us less than ssthresh packets in flight, open exponentially
 544          * (maxseg at a time).  Otherwise open linearly (maxseg^2 / cwnd at a
 545          * time).
 546          */
 547         incr = CUNIT;
 548         if (cb->s_cwnd > cb->s_ssthresh)
 549                 incr = max(incr * incr / cb->s_cwnd, 1);
 550         cb->s_cwnd = min(cb->s_cwnd + incr, cb->s_cwmx);
 551
 552         /*
 553          * Trim Acked data from output queue.
 554          */
 555         SOCKBUF_LOCK(&so->so_snd);
 556         while ((m = so->so_snd.sb_mb) != NULL) {
 557                 if (SSEQ_LT((mtod(m, struct spx *))->si_seq, si->si_ack))
 558                         sbdroprecord_locked(&so->so_snd);
 559                 else
 560                         break;
 561         }
 562         sowwakeup_locked(so);
 563         cb->s_rack = si->si_ack;
 564 update_window:
 565         if (SSEQ_LT(cb->s_snxt, cb->s_rack))
 566                 cb->s_snxt = cb->s_rack;
 567         if (SSEQ_LT(cb->s_swl1, si->si_seq) || ((cb->s_swl1 == si->si_seq &&
 568             (SSEQ_LT(cb->s_swl2, si->si_ack))) ||
 569              (cb->s_swl2 == si->si_ack && SSEQ_LT(cb->s_ralo, si->si_alo)))) {
 570                 /* keep track of pure window updates */
 571                 if ((si->si_cc & SPX_SP) && cb->s_swl2 == si->si_ack
 572                     && SSEQ_LT(cb->s_ralo, si->si_alo)) {
 573                         spxstat.spxs_rcvwinupd++;
 574                         spxstat.spxs_rcvdupack--;
 575                 }
 576                 cb->s_ralo = si->si_alo;
 577                 cb->s_swl1 = si->si_seq;
 578                 cb->s_swl2 = si->si_ack;
 579                 cb->s_swnd = (1 + si->si_alo - si->si_ack);
 580                 if (cb->s_swnd > cb->s_smxw)
 581                         cb->s_smxw = cb->s_swnd;
 582                 cb->s_flags |= SF_WIN;
 583         }
 584
 585         /*
 586          * If this packet number is higher than that which we have allocated
 587          * refuse it, unless urgent.
 588          */
 589         if (SSEQ_GT(si->si_seq, cb->s_alo)) {
 590                 if (si->si_cc & SPX_SP) {
 591                         spxstat.spxs_rcvwinprobe++;
 592                         return (1);
 593                 } else
 594                         spxstat.spxs_rcvpackafterwin++;
 595                 if (si->si_cc & SPX_OB) {
 596                         if (SSEQ_GT(si->si_seq, cb->s_alo + 60)) {
 597                                 m_freem(dtom(si));
 598                                 return (0);
 599                         } /* else queue this packet; */
 600                 } else {
 601 #ifdef BROKEN
 602                         /*
 603                          * XXXRW: This is broken on at least one count:
 604                          * spx_close() will free the ipxp and related parts,
 605                          * which are then touched by spx_input() after the
 606                          * return from spx_reass().
 607                          */
 608                         /*struct socket *so = cb->s_ipxpcb->ipxp_socket;
 609                         if (so->so_state && SS_NOFDREF) {
 610                                 spx_close(cb);
 611                         } else
 612                                        would crash system*/
 613 #endif
 614                         spx_istat.notyet++;
 615                         m_freem(dtom(si));
 616                         return (0);
 617                 }
 618         }
 619
 620         /*
 621          * If this is a system packet, we don't need to queue it up, and
 622          * won't update acknowledge #.
 623          */
 624         if (si->si_cc & SPX_SP)
 625                 return (1);
 626
 627         /*
 628          * We have already seen this packet, so drop.
 629          */
 630         if (SSEQ_LT(si->si_seq, cb->s_ack)) {
 631                 spx_istat.bdreas++;
 632                 spxstat.spxs_rcvduppack++;
 633                 if (si->si_seq == cb->s_ack - 1)
 634                         spx_istat.lstdup++;
 635                 return (1);
 636         }
 637
 638         /*
 639          * Loop through all packets queued up to insert in appropriate
 640          * sequence.
 641          */
 642         for (q = cb->s_q.si_next; q != &cb->s_q; q = q->si_next) {
 643                 if (si->si_seq == SI(q)->si_seq) {
 644                         spxstat.spxs_rcvduppack++;
 645                         return (1);
 646                 }
 647                 if (SSEQ_LT(si->si_seq, SI(q)->si_seq)) {
 648                         spxstat.spxs_rcvoopack++;
 649                         break;
 650                 }
 651         }
 652         insque(si, q->si_prev);
 653         /*
 654          * If this packet is urgent, inform process
 655          */
 656         if (si->si_cc & SPX_OB) {
 657                 cb->s_iobc = ((char *)si)[1 + sizeof(*si)];
 658                 sohasoutofband(so);
 659                 cb->s_oobflags |= SF_IOOB;
 660         }
 661 present:
 662 #define SPINC sizeof(struct spxhdr)
 663         SOCKBUF_LOCK(&so->so_rcv);
 664
 665         /*
 666          * Loop through all packets queued up to update acknowledge number,
 667          * and present all acknowledged data to user; if in packet interface
 668          * mode, show packet headers.
 669          */
 670         for (q = cb->s_q.si_next; q != &cb->s_q; q = q->si_next) {
 671                   if (SI(q)->si_seq == cb->s_ack) {
 672                         cb->s_ack++;
 673                         m = dtom(q);
 674                         if (SI(q)->si_cc & SPX_OB) {
 675                                 cb->s_oobflags &= ~SF_IOOB;
 676                                 if (so->so_rcv.sb_cc)
 677                                         so->so_oobmark = so->so_rcv.sb_cc;
 678                                 else
 679                                         so->so_rcv.sb_state |= SBS_RCVATMARK;
 680                         }
 681                         q = q->si_prev;
 682                         remque(q->si_next);
 683                         wakeup = 1;
 684                         spxstat.spxs_rcvpack++;
 685 #ifdef SF_NEWCALL
 686                         if (cb->s_flags2 & SF_NEWCALL) {
 687                                 struct spxhdr *sp = mtod(m, struct spxhdr *);
 688                                 u_char dt = sp->spx_dt;
 689                                 spx_newchecks[4]++;
 690                                 if (dt != cb->s_rhdr.spx_dt) {
 691                                         struct mbuf *mm =
 692                                            m_getclr(M_DONTWAIT, MT_CONTROL);
 693                                         spx_newchecks[0]++;
 694                                         if (mm != NULL) {
 695                                                 u_short *s =
 696                                                         mtod(mm, u_short *);
 697                                                 cb->s_rhdr.spx_dt = dt;
 698                                                 mm->m_len = 5; /*XXX*/
 699                                                 s[0] = 5;
 700                                                 s[1] = 1;
 701                                                 *(u_char *)(&s[2]) = dt;
 702                                                 sbappend_locked(&so->so_rcv, mm);
 703                                         }
 704                                 }
 705                                 if (sp->spx_cc & SPX_OB) {
 706                                         MCHTYPE(m, MT_OOBDATA);
 707                                         spx_newchecks[1]++;
 708                                         so->so_oobmark = 0;
 709                                         so->so_rcv.sb_state &= ~SBS_RCVATMARK;
 710                                 }
 711                                 if (packetp == 0) {
 712                                         m->m_data += SPINC;
 713                                         m->m_len -= SPINC;
 714                                         m->m_pkthdr.len -= SPINC;
 715                                 }
 716                                 if ((sp->spx_cc & SPX_EM) || packetp) {
 717                                         sbappendrecord_locked(&so->so_rcv, m);
 718                                         spx_newchecks[9]++;
 719                                 } else
 720                                         sbappend_locked(&so->so_rcv, m);
 721                         } else
 722 #endif
 723                         if (packetp)
 724                                 sbappendrecord_locked(&so->so_rcv, m);
 725                         else {
 726                                 cb->s_rhdr = *mtod(m, struct spxhdr *);
 727                                 m->m_data += SPINC;
 728                                 m->m_len -= SPINC;
 729                                 m->m_pkthdr.len -= SPINC;
 730                                 sbappend_locked(&so->so_rcv, m);
 731                         }
 732                   } else
 733                         break;
 734         }
 735         if (wakeup)
 736                 sorwakeup_locked(so);
 737         else
 738                 SOCKBUF_UNLOCK(&so->so_rcv);
 739         return (0);
 740 }
 741
 742 void
 743 spx_ctlinput(int cmd, struct sockaddr *arg_as_sa, void *dummy)
 744 {
 745
 746         /* Currently, nothing. */
 747 }
 748
 749 static int
 750 spx_output(struct spxpcb *cb, struct mbuf *m0)
 751 {
 752         struct socket *so = cb->s_ipxpcb->ipxp_socket;
 753         struct mbuf *m;
 754         struct spx *si = NULL;
 755         struct sockbuf *sb = &so->so_snd;
 756         int len = 0, win, rcv_win;
 757         short span, off, recordp = 0;
 758         u_short alo;
 759         int error = 0, sendalot;
 760 #ifdef notdef
 761         int idle;
 762 #endif
 763         struct mbuf *mprev;
 764
 765         IPX_LOCK_ASSERT(cb->s_ipxpcb);
 766
 767         if (m0 != NULL) {
 768                 int mtu = cb->s_mtu;
 769                 int datalen;
 770
 771                 /*
 772                  * Make sure that packet isn't too big.
 773                  */
 774                 for (m = m0; m != NULL; m = m->m_next) {
 775                         mprev = m;
 776                         len += m->m_len;
 777                         if (m->m_flags & M_EOR)
 778                                 recordp = 1;
 779                 }
 780                 datalen = (cb->s_flags & SF_HO) ?
 781                                 len - sizeof(struct spxhdr) : len;
 782                 if (datalen > mtu) {
 783                         if (cb->s_flags & SF_PI) {
 784                                 m_freem(m0);
 785                                 return (EMSGSIZE);
 786                         } else {
 787                                 int oldEM = cb->s_cc & SPX_EM;
 788
 789                                 cb->s_cc &= ~SPX_EM;
 790                                 while (len > mtu) {
 791                                         m = m_copym(m0, 0, mtu, M_DONTWAIT);
 792                                         if (m == NULL) {
 793                                             cb->s_cc |= oldEM;
 794                                             m_freem(m0);
 795                                             return (ENOBUFS);
 796                                         }
 797                                         if (cb->s_flags & SF_NEWCALL) {
 798                                             struct mbuf *mm = m;
 799                                             spx_newchecks[7]++;
 800                                             while (mm != NULL) {
 801                                                 mm->m_flags &= ~M_EOR;
 802                                                 mm = mm->m_next;
 803                                             }
 804                                         }
 805                                         error = spx_output(cb, m);
 806                                         if (error) {
 807                                                 cb->s_cc |= oldEM;
 808                                                 m_freem(m0);
 809                                                 return (error);
 810                                         }
 811                                         m_adj(m0, mtu);
 812                                         len -= mtu;
 813                                 }
 814                                 cb->s_cc |= oldEM;
 815                         }
 816                 }
 817
 818                 /*
 819                  * Force length even, by adding a "garbage byte" if
 820                  * necessary.
 821                  */
 822                 if (len & 1) {
 823                         m = mprev;
 824                         if (M_TRAILINGSPACE(m) >= 1)
 825                                 m->m_len++;
 826                         else {
 827                                 struct mbuf *m1 = m_get(M_DONTWAIT, MT_DATA);
 828
 829                                 if (m1 == NULL) {
 830                                         m_freem(m0);
 831                                         return (ENOBUFS);
 832                                 }
 833                                 m1->m_len = 1;
 834                                 *(mtod(m1, u_char *)) = 0;
 835                                 m->m_next = m1;
 836                         }
 837                 }
 838                 m = m_gethdr(M_DONTWAIT, MT_DATA);
 839                 if (m == NULL) {
 840                         m_freem(m0);
 841                         return (ENOBUFS);
 842                 }
 843
 844                 /*
 845                  * Fill in mbuf with extended SP header and addresses and
 846                  * length put into network format.
 847                  */
 848                 MH_ALIGN(m, sizeof(struct spx));
 849                 m->m_len = sizeof(struct spx);
 850                 m->m_next = m0;
 851                 si = mtod(m, struct spx *);
 852                 si->si_i = *cb->s_ipx;
 853                 si->si_s = cb->s_shdr;
 854                 if ((cb->s_flags & SF_PI) && (cb->s_flags & SF_HO)) {
 855                         struct spxhdr *sh;
 856                         if (m0->m_len < sizeof(*sh)) {
 857                                 if((m0 = m_pullup(m0, sizeof(*sh))) == NULL) {
 858                                         m_free(m);
 859                                         m_freem(m0);
 860                                         return (EINVAL);
 861                                 }
 862                                 m->m_next = m0;
 863                         }
 864                         sh = mtod(m0, struct spxhdr *);
 865                         si->si_dt = sh->spx_dt;
 866                         si->si_cc |= sh->spx_cc & SPX_EM;
 867                         m0->m_len -= sizeof(*sh);
 868                         m0->m_data += sizeof(*sh);
 869                         len -= sizeof(*sh);
 870                 }
 871                 len += sizeof(*si);
 872                 if ((cb->s_flags2 & SF_NEWCALL) && recordp) {
 873                         si->si_cc |= SPX_EM;
 874                         spx_newchecks[8]++;
 875                 }
 876                 if (cb->s_oobflags & SF_SOOB) {
 877                         /*
 878                          * Per jqj@cornell: Make sure OB packets convey
 879                          * exactly 1 byte.  If the packet is 1 byte or
 880                          * larger, we have already guaranted there to be at
 881                          * least one garbage byte for the checksum, and extra
 882                          * bytes shouldn't hurt!
 883                          */
 884                         if (len > sizeof(*si)) {
 885                                 si->si_cc |= SPX_OB;
 886                                 len = (1 + sizeof(*si));
 887                         }
 888                 }
 889                 si->si_len = htons((u_short)len);
 890                 m->m_pkthdr.len = ((len - 1) | 1) + 1;
 891
 892                 /*
 893                  * Queue stuff up for output.
 894                  */
 895                 sbappendrecord(sb, m);
 896                 cb->s_seq++;
 897         }
 898 #ifdef notdef
 899         idle = (cb->s_smax == (cb->s_rack - 1));
 900 #endif
 901 again:
 902         sendalot = 0;
 903         off = cb->s_snxt - cb->s_rack;
 904         win = min(cb->s_swnd, (cb->s_cwnd / CUNIT));
 905
 906         /*
 907          * If in persist timeout with window of 0, send a probe.  Otherwise,
 908          * if window is small but nonzero and timer expired, send what we can
 909          * and go into transmit state.
 910          */
 911         if (cb->s_force == 1 + SPXT_PERSIST) {
 912                 if (win != 0) {
 913                         cb->s_timer[SPXT_PERSIST] = 0;
 914                         cb->s_rxtshift = 0;
 915                 }
 916         }
 917         span = cb->s_seq - cb->s_rack;
 918         len = min(span, win) - off;
 919
 920         if (len < 0) {
 921                 /*
 922                  * Window shrank after we went into it.  If window shrank to
 923                  * 0, cancel pending restransmission and pull s_snxt back to
 924                  * (closed) window.  We will enter persist state below.  If
 925                  * the widndow didn't close completely, just wait for an ACK.
 926                  */
 927                 len = 0;
 928                 if (win == 0) {
 929                         cb->s_timer[SPXT_REXMT] = 0;
 930                         cb->s_snxt = cb->s_rack;
 931                 }
 932         }
 933         if (len > 1)
 934                 sendalot = 1;
 935         rcv_win = sbspace(&so->so_rcv);
 936
 937         /*
 938          * Send if we owe peer an ACK.
 939          */
 940         if (cb->s_oobflags & SF_SOOB) {
 941                 /*
 942                  * Must transmit this out of band packet.
 943                  */
 944                 cb->s_oobflags &= ~ SF_SOOB;
 945                 sendalot = 1;
 946                 spxstat.spxs_sndurg++;
 947                 goto found;
 948         }
 949         if (cb->s_flags & SF_ACKNOW)
 950                 goto send;
 951         if (cb->s_state < TCPS_ESTABLISHED)
 952                 goto send;
 953
 954         /*
 955          * Silly window can't happen in spx.  Code from TCP deleted.
 956          */
 957         if (len)
 958                 goto send;
 959
 960         /*
 961          * Compare available window to amount of window known to peer (as
 962          * advertised window less next expected input.)  If the difference is
 963          * at least two packets or at least 35% of the mximum possible
 964          * window, then want to send a window update to peer.
 965          */
 966         if (rcv_win > 0) {
 967                 u_short delta =  1 + cb->s_alo - cb->s_ack;
 968                 int adv = rcv_win - (delta * cb->s_mtu);
 969
 970                 if ((so->so_rcv.sb_cc == 0 && adv >= (2 * cb->s_mtu)) ||
 971                     (100 * adv / so->so_rcv.sb_hiwat >= 35)) {
 972                         spxstat.spxs_sndwinup++;
 973                         cb->s_flags |= SF_ACKNOW;
 974                         goto send;
 975                 }
 976
 977         }
 978
 979         /*
 980          * Many comments from tcp_output.c are appropriate here including ...
 981          * If send window is too small, there is data to transmit, and no
 982          * retransmit or persist is pending, then go to persist state.  If
 983          * nothing happens soon, send when timer expires: if window is
 984          * nonzero, transmit what we can, otherwise send a probe.
 985          */
 986         if (so->so_snd.sb_cc && cb->s_timer[SPXT_REXMT] == 0 &&
 987             cb->s_timer[SPXT_PERSIST] == 0) {
 988                 cb->s_rxtshift = 0;
 989                 spx_setpersist(cb);
 990         }
 991
 992         /*
 993          * No reason to send a packet, just return.
 994          */
 995         cb->s_outx = 1;
 996         return (0);
 997
 998 send:
 999         /*
1000          * Find requested packet.
1001          */
1002         si = 0;
1003         if (len > 0) {
1004                 cb->s_want = cb->s_snxt;
1005                 for (m = sb->sb_mb; m != NULL; m = m->m_act) {
1006                         si = mtod(m, struct spx *);
1007                         if (SSEQ_LEQ(cb->s_snxt, si->si_seq))
1008                                 break;
1009                 }
1010         found:
1011                 if (si != NULL) {
1012                         if (si->si_seq == cb->s_snxt)
1013                                         cb->s_snxt++;
1014                                 else
1015                                         spxstat.spxs_sndvoid++, si = 0;
1016                 }
1017         }
1018
1019         /*
1020          * Update window.
1021          */
1022         if (rcv_win < 0)
1023                 rcv_win = 0;
1024         alo = cb->s_ack - 1 + (rcv_win / ((short)cb->s_mtu));
1025         if (SSEQ_LT(alo, cb->s_alo))
1026                 alo = cb->s_alo;
1027
1028         if (si != NULL) {
1029                 /*
1030                  * Must make a copy of this packet for ipx_output to monkey
1031                  * with.
1032                  */
1033                 m = m_copy(dtom(si), 0, (int)M_COPYALL);
1034                 if (m == NULL)
1035                         return (ENOBUFS);
1036                 si = mtod(m, struct spx *);
1037                 if (SSEQ_LT(si->si_seq, cb->s_smax))
1038                         spxstat.spxs_sndrexmitpack++;
1039                 else
1040                         spxstat.spxs_sndpack++;
1041         } else if (cb->s_force || cb->s_flags & SF_ACKNOW) {
1042                 /*
1043                  * Must send an acknowledgement or a probe.
1044                  */
1045                 if (cb->s_force)
1046                         spxstat.spxs_sndprobe++;
1047                 if (cb->s_flags & SF_ACKNOW)
1048                         spxstat.spxs_sndacks++;
1049                 m = m_gethdr(M_DONTWAIT, MT_DATA);
1050                 if (m == NULL)
1051                         return (ENOBUFS);
1052
1053                 /*
1054                  * Fill in mbuf with extended SP header and addresses and
1055                  * length put into network format.
1056                  */
1057                 MH_ALIGN(m, sizeof(struct spx));
1058                 m->m_len = sizeof(*si);
1059                 m->m_pkthdr.len = sizeof(*si);
1060                 si = mtod(m, struct spx *);
1061                 si->si_i = *cb->s_ipx;
1062                 si->si_s = cb->s_shdr;
1063                 si->si_seq = cb->s_smax + 1;
1064                 si->si_len = htons(sizeof(*si));
1065                 si->si_cc |= SPX_SP;
1066         } else {
1067                 cb->s_outx = 3;
1068                 if (so->so_options & SO_DEBUG || traceallspxs)
1069                         spx_trace(SA_OUTPUT, cb->s_state, cb, si, 0);
1070                 return (0);
1071         }
1072         /*
1073          * Stuff checksum and output datagram.
1074          */
1075         if ((si->si_cc & SPX_SP) == 0) {
1076                 if (cb->s_force != (1 + SPXT_PERSIST) ||
1077                     cb->s_timer[SPXT_PERSIST] == 0) {
1078                         /*
1079                          * If this is a new packet and we are not currently
1080                          * timing anything, time this one.
1081                          */
1082                         if (SSEQ_LT(cb->s_smax, si->si_seq)) {
1083                                 cb->s_smax = si->si_seq;
1084                                 if (cb->s_rtt == 0) {
1085                                         spxstat.spxs_segstimed++;
1086                                         cb->s_rtseq = si->si_seq;
1087                                         cb->s_rtt = 1;
1088                                 }
1089                         }
1090
1091                         /*
1092                          * Set rexmt timer if not currently set, initial
1093                          * value for retransmit timer is smoothed round-trip
1094                          * time + 2 * round-trip time variance.  Initialize
1095                          * shift counter which is used for backoff of
1096                          * retransmit time.
1097                          */
1098                         if (cb->s_timer[SPXT_REXMT] == 0 &&
1099                             cb->s_snxt != cb->s_rack) {
1100                                 cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
1101                                 if (cb->s_timer[SPXT_PERSIST]) {
1102                                         cb->s_timer[SPXT_PERSIST] = 0;
1103                                         cb->s_rxtshift = 0;
1104                                 }
1105                         }
1106                 } else if (SSEQ_LT(cb->s_smax, si->si_seq))
1107                         cb->s_smax = si->si_seq;
1108         } else if (cb->s_state < TCPS_ESTABLISHED) {
1109                 if (cb->s_rtt == 0)
1110                         cb->s_rtt = 1; /* Time initial handshake */
1111                 if (cb->s_timer[SPXT_REXMT] == 0)
1112                         cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
1113         }
1114
1115         /*
1116          * Do not request acks when we ack their data packets or when we do a
1117          * gratuitous window update.
1118          */
1119         if (((si->si_cc & SPX_SP) == 0) || cb->s_force)
1120                 si->si_cc |= SPX_SA;
1121         si->si_seq = htons(si->si_seq);
1122         si->si_alo = htons(alo);
1123         si->si_ack = htons(cb->s_ack);
1124
1125         if (ipxcksum)
1126                 si->si_sum = ipx_cksum(m, ntohs(si->si_len));
1127         else
1128                 si->si_sum = 0xffff;
1129
1130         cb->s_outx = 4;
1131         if (so->so_options & SO_DEBUG || traceallspxs)
1132                 spx_trace(SA_OUTPUT, cb->s_state, cb, si, 0);
1133
1134         if (so->so_options & SO_DONTROUTE)
1135                 error = ipx_outputfl(m, NULL, IPX_ROUTETOIF);
1136         else
1137                 error = ipx_outputfl(m, &cb->s_ipxpcb->ipxp_route, 0);
1138         if (error)
1139                 return (error);
1140         spxstat.spxs_sndtotal++;
1141
1142         /*
1143          * Data sent (as far as we can tell).  If this advertises a larger
1144          * window than any other segment, then remember the size of the
1145          * advertized window.  Any pending ACK has now been sent.
1146          */
1147         cb->s_force = 0;
1148         cb->s_flags &= ~(SF_ACKNOW|SF_DELACK);
1149         if (SSEQ_GT(alo, cb->s_alo))
1150                 cb->s_alo = alo;
1151         if (sendalot)
1152                 goto again;
1153         cb->s_outx = 5;
1154         return (0);
1155 }
1156
1157 static int spx_do_persist_panics = 0;
1158
1159 static void
1160 spx_setpersist(struct spxpcb *cb)
1161 {
1162         int t = ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1;
1163
1164         IPX_LOCK_ASSERT(cb->s_ipxpcb);
1165
1166         if (cb->s_timer[SPXT_REXMT] && spx_do_persist_panics)
1167                 panic("spx_output REXMT");
1168
1169         /*
1170          * Start/restart persistance timer.
1171          */
1172         SPXT_RANGESET(cb->s_timer[SPXT_PERSIST],
1173             t*spx_backoff[cb->s_rxtshift],
1174             SPXTV_PERSMIN, SPXTV_PERSMAX);
1175         if (cb->s_rxtshift < SPX_MAXRXTSHIFT)
1176                 cb->s_rxtshift++;
1177 }
1178
1179 int
1180 spx_ctloutput(struct socket *so, struct sockopt *sopt)
1181 {
1182         struct spxhdr spxhdr;
1183         struct ipxpcb *ipxp;
1184         struct spxpcb *cb;
1185         int mask, error;
1186         short soptval;
1187         u_short usoptval;
1188         int optval;
1189
1190         ipxp = sotoipxpcb(so);
1191         KASSERT(ipxp != NULL, ("spx_ctloutput: ipxp == NULL"));
1192
1193         /*
1194          * This will have to be changed when we do more general stacking of
1195          * protocols.
1196          */
1197         if (sopt->sopt_level != IPXPROTO_SPX)
1198                 return (ipx_ctloutput(so, sopt));
1199
1200         IPX_LOCK(ipxp);
1201         if (ipxp->ipxp_flags & IPXP_DROPPED) {
1202                 IPX_UNLOCK(ipxp);
1203                 return (ECONNRESET);
1204         }
1205
1206         IPX_LOCK(ipxp);
1207         cb = ipxtospxpcb(ipxp);
1208         KASSERT(cb != NULL, ("spx_ctloutput: cb == NULL"));
1209
1210         error = 0;
1211         switch (sopt->sopt_dir) {
1212         case SOPT_GET:
1213                 switch (sopt->sopt_name) {
1214                 case SO_HEADERS_ON_INPUT:
1215                         mask = SF_HI;
1216                         goto get_flags;
1217
1218                 case SO_HEADERS_ON_OUTPUT:
1219                         mask = SF_HO;
1220                 get_flags:
1221                         soptval = cb->s_flags & mask;
1222                         IPX_UNLOCK(ipxp);
1223                         error = sooptcopyout(sopt, &soptval,
1224                             sizeof(soptval));
1225                         break;
1226
1227                 case SO_MTU:
1228                         usoptval = cb->s_mtu;
1229                         IPX_UNLOCK(ipxp);
1230                         error = sooptcopyout(sopt, &usoptval,
1231                             sizeof(usoptval));
1232                         break;
1233
1234                 case SO_LAST_HEADER:
1235                         spxhdr = cb->s_rhdr;
1236                         IPX_UNLOCK(ipxp);
1237                         error = sooptcopyout(sopt, &spxhdr, sizeof(spxhdr));
1238                         break;
1239
1240                 case SO_DEFAULT_HEADERS:
1241                         spxhdr = cb->s_shdr;
1242                         IPX_UNLOCK(ipxp);
1243                         error = sooptcopyout(sopt, &spxhdr, sizeof(spxhdr));
1244                         break;
1245
1246                 default:
1247                         IPX_UNLOCK(ipxp);
1248                         error = ENOPROTOOPT;
1249                 }
1250                 break;
1251
1252         case SOPT_SET:
1253                 /*
1254                  * XXX Why are these shorts on get and ints on set?  That
1255                  * doesn't make any sense...
1256                  *
1257                  * XXXRW: Note, when we re-acquire the ipxp lock, we should
1258                  * re-check that it's not dropped.
1259                  */
1260                 IPX_UNLOCK(ipxp);
1261                 switch (sopt->sopt_name) {
1262                 case SO_HEADERS_ON_INPUT:
1263                         mask = SF_HI;
1264                         goto set_head;
1265
1266                 case SO_HEADERS_ON_OUTPUT:
1267                         mask = SF_HO;
1268                 set_head:
1269                         error = sooptcopyin(sopt, &optval, sizeof optval,
1270                                             sizeof optval);
1271                         if (error)
1272                                 break;
1273
1274                         IPX_LOCK(ipxp);
1275                         if (cb->s_flags & SF_PI) {
1276                                 if (optval)
1277                                         cb->s_flags |= mask;
1278                                 else
1279                                         cb->s_flags &= ~mask;
1280                         } else error = EINVAL;
1281                         IPX_UNLOCK(ipxp);
1282                         break;
1283
1284                 case SO_MTU:
1285                         error = sooptcopyin(sopt, &usoptval, sizeof usoptval,
1286                                             sizeof usoptval);
1287                         if (error)
1288                                 break;
1289                         /* Unlocked write. */
1290                         cb->s_mtu = usoptval;
1291                         break;
1292
1293 #ifdef SF_NEWCALL
1294                 case SO_NEWCALL:
1295                         error = sooptcopyin(sopt, &optval, sizeof optval,
1296                                             sizeof optval);
1297                         if (error)
1298                                 break;
1299                         IPX_LOCK(ipxp);
1300                         if (optval) {
1301                                 cb->s_flags2 |= SF_NEWCALL;
1302                                 spx_newchecks[5]++;
1303                         } else {
1304                                 cb->s_flags2 &= ~SF_NEWCALL;
1305                                 spx_newchecks[6]++;
1306                         }
1307                         IPX_UNLOCK(ipxp);
1308                         break;
1309 #endif
1310
1311                 case SO_DEFAULT_HEADERS:
1312                         {
1313                                 struct spxhdr sp;
1314
1315                                 error = sooptcopyin(sopt, &sp, sizeof sp,
1316                                                     sizeof sp);
1317                                 if (error)
1318                                         break;
1319                                 IPX_LOCK(ipxp);
1320                                 cb->s_dt = sp.spx_dt;
1321                                 cb->s_cc = sp.spx_cc & SPX_EM;
1322                                 IPX_UNLOCK(ipxp);
1323                         }
1324                         break;
1325
1326                 default:
1327                         error = ENOPROTOOPT;
1328                 }
1329                 break;
1330
1331         default:
1332                 panic("spx_ctloutput: bad socket option direction");
1333         }
1334         return (error);
1335 }
1336
1337 static void
1338 spx_usr_abort(struct socket *so)
1339 {
1340         struct ipxpcb *ipxp;
1341         struct spxpcb *cb;
1342
1343         ipxp = sotoipxpcb(so);
1344         KASSERT(ipxp != NULL, ("spx_usr_abort: ipxp == NULL"));
1345
1346         cb = ipxtospxpcb(ipxp);
1347         KASSERT(cb != NULL, ("spx_usr_abort: cb == NULL"));
1348
1349         IPX_LIST_LOCK();
1350         IPX_LOCK(ipxp);
1351         spx_drop(cb, ECONNABORTED);
1352         IPX_UNLOCK(ipxp);
1353         IPX_LIST_UNLOCK();
1354 }
1355
1356 /*
1357  * Accept a connection.  Essentially all the work is done at higher levels;
1358  * just return the address of the peer, storing through addr.
1359  */
1360 static int
1361 spx_accept(struct socket *so, struct sockaddr **nam)
1362 {
1363         struct ipxpcb *ipxp;
1364         struct sockaddr_ipx *sipx, ssipx;
1365
1366         ipxp = sotoipxpcb(so);
1367         KASSERT(ipxp != NULL, ("spx_accept: ipxp == NULL"));
1368
1369         sipx = &ssipx;
1370         bzero(sipx, sizeof *sipx);
1371         sipx->sipx_len = sizeof *sipx;
1372         sipx->sipx_family = AF_IPX;
1373         IPX_LOCK(ipxp);
1374         sipx->sipx_addr = ipxp->ipxp_faddr;
1375         IPX_UNLOCK(ipxp);
1376         *nam = sodupsockaddr((struct sockaddr *)sipx, M_WAITOK);
1377         return (0);
1378 }
1379
1380 static int
1381 spx_attach(struct socket *so, int proto, struct thread *td)
1382 {
1383         struct ipxpcb *ipxp;
1384         struct spxpcb *cb;
1385         struct mbuf *mm;
1386         struct sockbuf *sb;
1387         int error;
1388
1389         ipxp = sotoipxpcb(so);
1390         KASSERT(ipxp == NULL, ("spx_attach: ipxp != NULL"));
1391
1392         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1393                 error = soreserve(so, (u_long) 3072, (u_long) 3072);
1394                 if (error)
1395                         return (error);
1396         }
1397
1398         MALLOC(cb, struct spxpcb *, sizeof *cb, M_PCB, M_NOWAIT | M_ZERO);
1399         if (cb == NULL)
1400                 return (ENOBUFS);
1401         mm = m_getclr(M_DONTWAIT, MT_DATA);
1402         if (mm == NULL) {
1403                 FREE(cb, M_PCB);
1404                 return (ENOBUFS);
1405         }
1406
1407         IPX_LIST_LOCK();
1408         error = ipx_pcballoc(so, &ipxpcb_list, td);
1409         if (error) {
1410                 IPX_LIST_UNLOCK();
1411                 m_free(mm);
1412                 FREE(cb, M_PCB);
1413                 return (error);
1414         }
1415         ipxp = sotoipxpcb(so);
1416         ipxp->ipxp_flags |= IPXP_SPX;
1417
1418         cb->s_ipx = mtod(mm, struct ipx *);
1419         cb->s_state = TCPS_LISTEN;
1420         cb->s_smax = -1;
1421         cb->s_swl1 = -1;
1422         cb->s_q.si_next = cb->s_q.si_prev = &cb->s_q;
1423         cb->s_ipxpcb = ipxp;
1424         cb->s_mtu = 576 - sizeof(struct spx);
1425         sb = &so->so_snd;
1426         cb->s_cwnd = sbspace(sb) * CUNIT / cb->s_mtu;
1427         cb->s_ssthresh = cb->s_cwnd;
1428         cb->s_cwmx = sbspace(sb) * CUNIT / (2 * sizeof(struct spx));
1429         /*
1430          * Above is recomputed when connecting to account for changed
1431          * buffering or mtu's.
1432          */
1433         cb->s_rtt = SPXTV_SRTTBASE;
1434         cb->s_rttvar = SPXTV_SRTTDFLT << 2;
1435         SPXT_RANGESET(cb->s_rxtcur,
1436             ((SPXTV_SRTTBASE >> 2) + (SPXTV_SRTTDFLT << 2)) >> 1,
1437             SPXTV_MIN, SPXTV_REXMTMAX);
1438         ipxp->ipxp_pcb = (caddr_t)cb;
1439         IPX_LIST_UNLOCK();
1440         return (0);
1441 }
1442
1443 static void
1444 spx_pcbdetach(struct ipxpcb *ipxp)
1445 {
1446         struct spxpcb *cb;
1447         struct spx_q *s;
1448         struct mbuf *m;
1449
1450         IPX_LOCK_ASSERT(ipxp);
1451
1452         cb = ipxtospxpcb(ipxp);
1453         KASSERT(cb != NULL, ("spx_pcbdetach: cb == NULL"));
1454
1455         s = cb->s_q.si_next;
1456         while (s != &(cb->s_q)) {
1457                 s = s->si_next;
1458                 remque(s);
1459                 m = dtom(s);
1460                 m_freem(m);
1461         }
1462         m_free(dtom(cb->s_ipx));
1463         FREE(cb, M_PCB);
1464         ipxp->ipxp_pcb = NULL;
1465 }
1466
1467 static int
1468 spx_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1469 {
1470         struct ipxpcb *ipxp;
1471         int error;
1472
1473         ipxp = sotoipxpcb(so);
1474         KASSERT(ipxp != NULL, ("spx_bind: ipxp == NULL"));
1475
1476         IPX_LIST_LOCK();
1477         IPX_LOCK(ipxp);
1478         if (ipxp->ipxp_flags & IPXP_DROPPED) {
1479                 error = EINVAL;
1480                 goto out;
1481         }
1482         error = ipx_pcbbind(ipxp, nam, td);
1483 out:
1484         IPX_UNLOCK(ipxp);
1485         IPX_LIST_UNLOCK();
1486         return (error);
1487 }
1488
1489 static void
1490 spx_usr_close(struct socket *so)
1491 {
1492         struct ipxpcb *ipxp;
1493         struct spxpcb *cb;
1494
1495         ipxp = sotoipxpcb(so);
1496         KASSERT(ipxp != NULL, ("spx_usr_close: ipxp == NULL"));
1497
1498         cb = ipxtospxpcb(ipxp);
1499         KASSERT(cb != NULL, ("spx_usr_close: cb == NULL"));
1500
1501         IPX_LIST_LOCK();
1502         IPX_LOCK(ipxp);
1503         if (cb->s_state > TCPS_LISTEN)
1504                 spx_disconnect(cb);
1505         else
1506                 spx_close(cb);
1507         IPX_UNLOCK(ipxp);
1508         IPX_LIST_UNLOCK();
1509 }
1510
1511 /*
1512  * Initiate connection to peer.  Enter SYN_SENT state, and mark socket as
1513  * connecting.  Start keep-alive timer, setup prototype header, send initial
1514  * system packet requesting connection.
1515  */
1516 static int
1517 spx_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1518 {
1519         struct ipxpcb *ipxp;
1520         struct spxpcb *cb;
1521         int error;
1522
1523         ipxp = sotoipxpcb(so);
1524         KASSERT(ipxp != NULL, ("spx_connect: ipxp == NULL"));
1525
1526         cb = ipxtospxpcb(ipxp);
1527         KASSERT(cb != NULL, ("spx_connect: cb == NULL"));
1528
1529         IPX_LIST_LOCK();
1530         IPX_LOCK(ipxp);
1531         if (ipxp->ipxp_flags & IPXP_DROPPED) {
1532                 error = EINVAL;
1533                 goto spx_connect_end;
1534         }
1535         if (ipxp->ipxp_lport == 0) {
1536                 error = ipx_pcbbind(ipxp, NULL, td);
1537                 if (error)
1538                         goto spx_connect_end;
1539         }
1540         error = ipx_pcbconnect(ipxp, nam, td);
1541         if (error)
1542                 goto spx_connect_end;
1543         soisconnecting(so);
1544         spxstat.spxs_connattempt++;
1545         cb->s_state = TCPS_SYN_SENT;
1546         cb->s_did = 0;
1547         spx_template(cb);
1548         cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
1549         cb->s_force = 1 + SPXTV_KEEP;
1550         /*
1551          * Other party is required to respond to the port I send from, but he
1552          * is not required to answer from where I am sending to, so allow
1553          * wildcarding.  Original port I am sending to is still saved in
1554          * cb->s_dport.
1555          */
1556         ipxp->ipxp_fport = 0;
1557         error = spx_output(cb, NULL);
1558 spx_connect_end:
1559         IPX_UNLOCK(ipxp);
1560         IPX_LIST_UNLOCK();
1561         return (error);
1562 }
1563
1564 static void
1565 spx_detach(struct socket *so)
1566 {
1567         struct ipxpcb *ipxp;
1568         struct spxpcb *cb;
1569
1570         /*
1571          * XXXRW: Should assert appropriately detached.
1572          */
1573         ipxp = sotoipxpcb(so);
1574         KASSERT(ipxp != NULL, ("spx_detach: ipxp == NULL"));
1575
1576         cb = ipxtospxpcb(ipxp);
1577         KASSERT(cb != NULL, ("spx_detach: cb == NULL"));
1578
1579         IPX_LIST_LOCK();
1580         IPX_LOCK(ipxp);
1581         spx_pcbdetach(ipxp);
1582         ipx_pcbfree(ipxp);
1583         IPX_LIST_UNLOCK();
1584 }
1585
1586 /*
1587  * We may decide later to implement connection closing handshaking at the spx
1588  * level optionally.  Here is the hook to do it:
1589  */
1590 static int
1591 spx_usr_disconnect(struct socket *so)
1592 {
1593         struct ipxpcb *ipxp;
1594         struct spxpcb *cb;
1595         int error;
1596
1597         ipxp = sotoipxpcb(so);
1598         KASSERT(ipxp != NULL, ("spx_usr_disconnect: ipxp == NULL"));
1599
1600         cb = ipxtospxpcb(ipxp);
1601         KASSERT(cb != NULL, ("spx_usr_disconnect: cb == NULL"));
1602
1603         IPX_LIST_LOCK();
1604         IPX_LOCK(ipxp);
1605         if (ipxp->ipxp_flags & IPXP_DROPPED) {
1606                 error = EINVAL;
1607                 goto out;
1608         }
1609         spx_disconnect(cb);
1610         error = 0;
1611 out:
1612         IPX_UNLOCK(ipxp);
1613         IPX_LIST_UNLOCK();
1614         return (error);
1615 }
1616
1617 static int
1618 spx_listen(struct socket *so, int backlog, struct thread *td)
1619 {
1620         int error;
1621         struct ipxpcb *ipxp;
1622         struct spxpcb *cb;
1623
1624         error = 0;
1625         ipxp = sotoipxpcb(so);
1626         KASSERT(ipxp != NULL, ("spx_listen: ipxp == NULL"));
1627
1628         cb = ipxtospxpcb(ipxp);
1629         KASSERT(cb != NULL, ("spx_listen: cb == NULL"));
1630
1631         IPX_LIST_LOCK();
1632         IPX_LOCK(ipxp);
1633         if (ipxp->ipxp_flags & IPXP_DROPPED) {
1634                 error = EINVAL;
1635                 goto out;
1636         }
1637         SOCK_LOCK(so);
1638         error = solisten_proto_check(so);
1639         if (error == 0 && ipxp->ipxp_lport == 0)
1640                 error = ipx_pcbbind(ipxp, NULL, td);
1641         if (error == 0) {
1642                 cb->s_state = TCPS_LISTEN;
1643                 solisten_proto(so, backlog);
1644         }
1645         SOCK_UNLOCK(so);
1646 out:
1647         IPX_UNLOCK(ipxp);
1648         IPX_LIST_UNLOCK();
1649         return (error);
1650 }
1651
1652 /*
1653  * After a receive, possibly send acknowledgment updating allocation.
1654  */
1655 static int
1656 spx_rcvd(struct socket *so, int flags)
1657 {
1658         struct ipxpcb *ipxp;
1659         struct spxpcb *cb;
1660         int error;
1661
1662         ipxp = sotoipxpcb(so);
1663         KASSERT(ipxp != NULL, ("spx_rcvd: ipxp == NULL"));
1664
1665         cb = ipxtospxpcb(ipxp);
1666         KASSERT(cb != NULL, ("spx_rcvd: cb == NULL"));
1667
1668         IPX_LOCK(ipxp);
1669         if (ipxp->ipxp_flags & IPXP_DROPPED) {
1670                 error = EINVAL;
1671                 goto out;
1672         }
1673         cb->s_flags |= SF_RVD;
1674         spx_output(cb, NULL);
1675         cb->s_flags &= ~SF_RVD;
1676         error = 0;
1677 out:
1678         IPX_UNLOCK(ipxp);
1679         return (error);
1680 }
1681
1682 static int
1683 spx_rcvoob(struct socket *so, struct mbuf *m, int flags)
1684 {
1685         struct ipxpcb *ipxp;
1686         struct spxpcb *cb;
1687         int error;
1688
1689         ipxp = sotoipxpcb(so);
1690         KASSERT(ipxp != NULL, ("spx_rcvoob: ipxp == NULL"));
1691
1692         cb = ipxtospxpcb(ipxp);
1693         KASSERT(cb != NULL, ("spx_rcvoob: cb == NULL"));
1694
1695         IPX_LOCK(ipxp);
1696         if (ipxp->ipxp_flags & IPXP_DROPPED) {
1697                 error = EINVAL;
1698                 goto out;
1699         }
1700         SOCKBUF_LOCK(&so->so_rcv);
1701         if ((cb->s_oobflags & SF_IOOB) || so->so_oobmark ||
1702             (so->so_rcv.sb_state & SBS_RCVATMARK)) {
1703                 SOCKBUF_UNLOCK(&so->so_rcv);
1704                 m->m_len = 1;
1705                 *mtod(m, caddr_t) = cb->s_iobc;
1706                 error = 0;
1707                 goto out;
1708         }
1709         SOCKBUF_UNLOCK(&so->so_rcv);
1710         error = EINVAL;
1711 out:
1712         IPX_UNLOCK(ipxp);
1713         return (error);
1714 }
1715
1716 static int
1717 spx_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1718     struct mbuf *controlp, struct thread *td)
1719 {
1720         struct ipxpcb *ipxp;
1721         struct spxpcb *cb;
1722         int error;
1723
1724         ipxp = sotoipxpcb(so);
1725         KASSERT(ipxp != NULL, ("spx_send: ipxp == NULL"));
1726
1727         cb = ipxtospxpcb(ipxp);
1728         KASSERT(cb != NULL, ("spx_send: cb == NULL"));
1729
1730         error = 0;
1731         IPX_LOCK(ipxp);
1732         if (ipxp->ipxp_flags & IPXP_DROPPED) {
1733                 error = ECONNRESET;
1734                 goto spx_send_end;
1735         }
1736         if (flags & PRUS_OOB) {
1737                 if (sbspace(&so->so_snd) < -512) {
1738                         error = ENOBUFS;
1739                         goto spx_send_end;
1740                 }
1741                 cb->s_oobflags |= SF_SOOB;
1742         }
1743         if (controlp != NULL) {
1744                 u_short *p = mtod(controlp, u_short *);
1745                 spx_newchecks[2]++;
1746                 if ((p[0] == 5) && (p[1] == 1)) { /* XXXX, for testing */
1747                         cb->s_shdr.spx_dt = *(u_char *)(&p[2]);
1748                         spx_newchecks[3]++;
1749                 }
1750                 m_freem(controlp);
1751         }
1752         controlp = NULL;
1753         error = spx_output(cb, m);
1754         m = NULL;
1755 spx_send_end:
1756         IPX_UNLOCK(ipxp);
1757         if (controlp != NULL)
1758                 m_freem(controlp);
1759         if (m != NULL)
1760                 m_freem(m);
1761         return (error);
1762 }
1763
1764 static int
1765 spx_shutdown(struct socket *so)
1766 {
1767         struct ipxpcb *ipxp;
1768         struct spxpcb *cb;
1769         int error;
1770
1771         ipxp = sotoipxpcb(so);
1772         KASSERT(ipxp != NULL, ("spx_shutdown: ipxp == NULL"));
1773
1774         cb = ipxtospxpcb(ipxp);
1775         KASSERT(cb != NULL, ("spx_shutdown: cb == NULL"));
1776
1777         socantsendmore(so);
1778         IPX_LIST_LOCK();
1779         IPX_LOCK(ipxp);
1780         if (ipxp->ipxp_flags & IPXP_DROPPED) {
1781                 error = EINVAL;
1782                 goto out;
1783         }
1784         spx_usrclosed(cb);
1785         error = 0;
1786 out:
1787         IPX_UNLOCK(ipxp);
1788         IPX_LIST_UNLOCK();
1789         return (error);
1790 }
1791
1792 static int
1793 spx_sp_attach(struct socket *so, int proto, struct thread *td)
1794 {
1795         struct ipxpcb *ipxp;
1796         struct spxpcb *cb;
1797         int error;
1798
1799         KASSERT(so->so_pcb == NULL, ("spx_sp_attach: so_pcb != NULL"));
1800
1801         error = spx_attach(so, proto, td);
1802         if (error)
1803                 return (error);
1804
1805         ipxp = sotoipxpcb(so);
1806         KASSERT(ipxp != NULL, ("spx_sp_attach: ipxp == NULL"));
1807
1808         cb = ipxtospxpcb(ipxp);
1809         KASSERT(cb != NULL, ("spx_sp_attach: cb == NULL"));
1810
1811         IPX_LOCK(ipxp);
1812         cb->s_flags |= (SF_HI | SF_HO | SF_PI);
1813         IPX_UNLOCK(ipxp);
1814         return (0);
1815 }
1816
1817 /*
1818  * Create template to be used to send spx packets on a connection.  Called
1819  * after host entry created, fills in a skeletal spx header (choosing
1820  * connection id), minimizing the amount of work necessary when the
1821  * connection is used.
1822  */
1823 static void
1824 spx_template(struct spxpcb *cb)
1825 {
1826         struct ipxpcb *ipxp = cb->s_ipxpcb;
1827         struct ipx *ipx = cb->s_ipx;
1828         struct sockbuf *sb = &(ipxp->ipxp_socket->so_snd);
1829
1830         IPX_LOCK_ASSERT(ipxp);
1831
1832         ipx->ipx_pt = IPXPROTO_SPX;
1833         ipx->ipx_sna = ipxp->ipxp_laddr;
1834         ipx->ipx_dna = ipxp->ipxp_faddr;
1835         SPX_LOCK();
1836         cb->s_sid = htons(spx_iss);
1837         spx_iss += SPX_ISSINCR/2;
1838         SPX_UNLOCK();
1839         cb->s_alo = 1;
1840         cb->s_cwnd = (sbspace(sb) * CUNIT) / cb->s_mtu;
1841         /* Try to expand fast to full complement of large packets. */
1842         cb->s_ssthresh = cb->s_cwnd;
1843         cb->s_cwmx = (sbspace(sb) * CUNIT) / (2 * sizeof(struct spx));
1844         /* But allow for lots of little packets as well. */
1845         cb->s_cwmx = max(cb->s_cwmx, cb->s_cwnd);
1846 }
1847
1848 /*
1849  * Close a SPIP control block.  Wake up any sleepers.  We used to free any
1850  * queued packets and cb->s_ipx here, but now we defer that until the pcb is
1851  * discarded.
1852  */
1853 void
1854 spx_close(struct spxpcb *cb)
1855 {
1856         struct ipxpcb *ipxp = cb->s_ipxpcb;
1857         struct socket *so = ipxp->ipxp_socket;
1858
1859         KASSERT(ipxp != NULL, ("spx_close: ipxp == NULL"));
1860         IPX_LIST_LOCK_ASSERT();
1861         IPX_LOCK_ASSERT(ipxp);
1862
1863         ipxp->ipxp_flags |= IPXP_DROPPED;
1864         soisdisconnected(so);
1865         spxstat.spxs_closed++;
1866 }
1867
1868 /*
1869  * Someday we may do level 3 handshaking to close a connection or send a
1870  * xerox style error.  For now, just close.  cb will always be invalid after
1871  * this call.
1872  */
1873 static void
1874 spx_usrclosed(struct spxpcb *cb)
1875 {
1876
1877         IPX_LIST_LOCK_ASSERT();
1878         IPX_LOCK_ASSERT(cb->s_ipxpcb);
1879
1880         spx_close(cb);
1881 }
1882
1883 /*
1884  * cb will always be invalid after this call.
1885  */
1886 static void
1887 spx_disconnect(struct spxpcb *cb)
1888 {
1889
1890         IPX_LIST_LOCK_ASSERT();
1891         IPX_LOCK_ASSERT(cb->s_ipxpcb);
1892
1893         spx_close(cb);
1894 }
1895
1896 /*
1897  * Drop connection, reporting the specified error.  cb will always be invalid
1898  * after this call.
1899  */
1900 static void
1901 spx_drop(struct spxpcb *cb, int errno)
1902 {
1903         struct socket *so = cb->s_ipxpcb->ipxp_socket;
1904
1905         IPX_LIST_LOCK_ASSERT();
1906         IPX_LOCK_ASSERT(cb->s_ipxpcb);
1907
1908         /*
1909          * Someday, in the xerox world we will generate error protocol
1910          * packets announcing that the socket has gone away.
1911          */
1912         if (TCPS_HAVERCVDSYN(cb->s_state)) {
1913                 spxstat.spxs_drops++;
1914                 cb->s_state = TCPS_CLOSED;
1915                 /*tcp_output(cb);*/
1916         } else
1917                 spxstat.spxs_conndrops++;
1918         so->so_error = errno;
1919         spx_close(cb);
1920 }
1921
1922 /*
1923  * Fast timeout routine for processing delayed acks.
1924  */
1925 void
1926 spx_fasttimo(void)
1927 {
1928         struct ipxpcb *ipxp;
1929         struct spxpcb *cb;
1930
1931         IPX_LIST_LOCK();
1932         LIST_FOREACH(ipxp, &ipxpcb_list, ipxp_list) {
1933                 IPX_LOCK(ipxp);
1934                 if (!(ipxp->ipxp_flags & IPXP_SPX) ||
1935                     (ipxp->ipxp_flags & IPXP_DROPPED)) {
1936                         IPX_UNLOCK(ipxp);
1937                         continue;
1938                 }
1939                 cb = ipxtospxpcb(ipxp);
1940                 if (cb->s_flags & SF_DELACK) {
1941                         cb->s_flags &= ~SF_DELACK;
1942                         cb->s_flags |= SF_ACKNOW;
1943                         spxstat.spxs_delack++;
1944                         spx_output(cb, NULL);
1945                 }
1946                 IPX_UNLOCK(ipxp);
1947         }
1948         IPX_LIST_UNLOCK();
1949 }
1950
1951 /*
1952  * spx protocol timeout routine called every 500 ms.  Updates the timers in
1953  * all active pcb's and causes finite state machine actions if timers expire.
1954  */
1955 void
1956 spx_slowtimo(void)
1957 {
1958         struct ipxpcb *ipxp;
1959         struct spxpcb *cb;
1960         int i;
1961
1962         /*
1963          * Search through tcb's and update active timers.  Once, timers could
1964          * free ipxp's, but now we do that only when detaching a socket.
1965          */
1966         IPX_LIST_LOCK();
1967         LIST_FOREACH(ipxp, &ipxpcb_list, ipxp_list) {
1968                 IPX_LOCK(ipxp);
1969                 if (!(ipxp->ipxp_flags & IPXP_SPX) ||
1970                     (ipxp->ipxp_flags & IPXP_DROPPED)) {
1971                         IPX_UNLOCK(ipxp);
1972                         continue;
1973                 }
1974
1975                 cb = (struct spxpcb *)ipxp->ipxp_pcb;
1976                 KASSERT(cb != NULL, ("spx_slowtimo: cb == NULL"));
1977                 for (i = 0; i < SPXT_NTIMERS; i++) {
1978                         if (cb->s_timer[i] && --cb->s_timer[i] == 0) {
1979                                 spx_timers(cb, i);
1980                                 if (ipxp->ipxp_flags & IPXP_DROPPED)
1981                                         break;
1982                         }
1983                 }
1984                 if (!(ipxp->ipxp_flags & IPXP_DROPPED)) {
1985                         cb->s_idle++;
1986                         if (cb->s_rtt)
1987                                 cb->s_rtt++;
1988                 }
1989                 IPX_UNLOCK(ipxp);
1990         }
1991         IPX_LIST_UNLOCK();
1992         SPX_LOCK();
1993         spx_iss += SPX_ISSINCR/PR_SLOWHZ;               /* increment iss */
1994         SPX_UNLOCK();
1995 }
1996
1997 /*
1998  * SPX timer processing.
1999  */
2000 static void
2001 spx_timers(struct spxpcb *cb, int timer)
2002 {
2003         long rexmt;
2004         int win;
2005
2006         IPX_LIST_LOCK_ASSERT();
2007         IPX_LOCK_ASSERT(cb->s_ipxpcb);
2008
2009         cb->s_force = 1 + timer;
2010         switch (timer) {
2011         case SPXT_2MSL:
2012                 /*
2013                  * 2 MSL timeout in shutdown went off.  TCP deletes
2014                  * connection control block.
2015                  */
2016                 printf("spx: SPXT_2MSL went off for no reason\n");
2017                 cb->s_timer[timer] = 0;
2018                 break;
2019
2020         case SPXT_REXMT:
2021                 /*
2022                  * Retransmission timer went off.  Message has not been acked
2023                  * within retransmit interval.  Back off to a longer
2024                  * retransmit interval and retransmit one packet.
2025                  */
2026                 if (++cb->s_rxtshift > SPX_MAXRXTSHIFT) {
2027                         cb->s_rxtshift = SPX_MAXRXTSHIFT;
2028                         spxstat.spxs_timeoutdrop++;
2029                         spx_drop(cb, ETIMEDOUT);
2030                         break;
2031                 }
2032                 spxstat.spxs_rexmttimeo++;
2033                 rexmt = ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1;
2034                 rexmt *= spx_backoff[cb->s_rxtshift];
2035                 SPXT_RANGESET(cb->s_rxtcur, rexmt, SPXTV_MIN, SPXTV_REXMTMAX);
2036                 cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
2037
2038                 /*
2039                  * If we have backed off fairly far, our srtt estimate is
2040                  * probably bogus.  Clobber it so we'll take the next rtt
2041                  * measurement as our srtt; move the current srtt into rttvar
2042                  * to keep the current retransmit times until then.
2043                  */
2044                 if (cb->s_rxtshift > SPX_MAXRXTSHIFT / 4 ) {
2045                         cb->s_rttvar += (cb->s_srtt >> 2);
2046                         cb->s_srtt = 0;
2047                 }
2048                 cb->s_snxt = cb->s_rack;
2049
2050                 /*
2051                  * If timing a packet, stop the timer.
2052                  */
2053                 cb->s_rtt = 0;
2054
2055                 /*
2056                  * See very long discussion in tcp_timer.c about congestion
2057                  * window and sstrhesh.
2058                  */
2059                 win = min(cb->s_swnd, (cb->s_cwnd/CUNIT)) / 2;
2060                 if (win < 2)
2061                         win = 2;
2062                 cb->s_cwnd = CUNIT;
2063                 cb->s_ssthresh = win * CUNIT;
2064                 spx_output(cb, NULL);
2065                 break;
2066
2067         case SPXT_PERSIST:
2068                 /*
2069                  * Persistance timer into zero window.  Force a probe to be
2070                  * sent.
2071                  */
2072                 spxstat.spxs_persisttimeo++;
2073                 spx_setpersist(cb);
2074                 spx_output(cb, NULL);
2075                 break;
2076
2077         case SPXT_KEEP:
2078                 /*
2079                  * Keep-alive timer went off; send something or drop
2080                  * connection if idle for too long.
2081                  */
2082                 spxstat.spxs_keeptimeo++;
2083                 if (cb->s_state < TCPS_ESTABLISHED)
2084                         goto dropit;
2085                 if (cb->s_ipxpcb->ipxp_socket->so_options & SO_KEEPALIVE) {
2086                         if (cb->s_idle >= SPXTV_MAXIDLE)
2087                                 goto dropit;
2088                         spxstat.spxs_keepprobe++;
2089                         spx_output(cb, NULL);
2090                 } else
2091                         cb->s_idle = 0;
2092                 cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
2093                 break;
2094
2095         dropit:
2096                 spxstat.spxs_keepdrops++;
2097                 spx_drop(cb, ETIMEDOUT);
2098                 break;
2099
2100         default:
2101                 panic("spx_timers: unknown timer %d", timer);
2102         }
2103 }