slirp/tcp_input.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)tcp_input.c 8.5 (Berkeley) 4/10/94
  30  * tcp_input.c,v 1.10 1994/10/13 18:36:32 wollman Exp
  31  */
  32
  33 /*
  34  * Changes and additions relating to SLiRP
  35  * Copyright (c) 1995 Danny Gasparovski.
  36  *
  37  * Please read the file COPYRIGHT for the
  38  * terms and conditions of the copyright.
  39  */
  40
  41 #include "qemu/osdep.h"
  42 #include <slirp.h>
  43 #include "ip_icmp.h"
  44
  45 #define TCPREXMTTHRESH 3
  46
  47 #define TCP_PAWS_IDLE   (24 * 24 * 60 * 60 * PR_SLOWHZ)
  48
  49 /* for modulo comparisons of timestamps */
  50 #define TSTMP_LT(a,b)   ((int)((a)-(b)) < 0)
  51 #define TSTMP_GEQ(a,b)  ((int)((a)-(b)) >= 0)
  52
  53 /*
  54  * Insert segment ti into reassembly queue of tcp with
  55  * control block tp.  Return TH_FIN if reassembly now includes
  56  * a segment with FIN.  The macro form does the common case inline
  57  * (segment is the next to be received on an established connection,
  58  * and the queue is empty), avoiding linkage into and removal
  59  * from the queue and repetition of various conversions.
  60  * Set DELACK for segments received in order, but ack immediately
  61  * when segments are out of order (so fast retransmit can work).
  62  */
  63 #ifdef TCP_ACK_HACK
  64 #define TCP_REASS(tp, ti, m, so, flags) {\
  65        if ((ti)->ti_seq == (tp)->rcv_nxt && \
  66            tcpfrag_list_empty(tp) && \
  67            (tp)->t_state == TCPS_ESTABLISHED) {\
  68                if (ti->ti_flags & TH_PUSH) \
  69                        tp->t_flags |= TF_ACKNOW; \
  70                else \
  71                        tp->t_flags |= TF_DELACK; \
  72                (tp)->rcv_nxt += (ti)->ti_len; \
  73                flags = (ti)->ti_flags & TH_FIN; \
  74                if (so->so_emu) { \
  75                        if (tcp_emu((so),(m))) sbappend((so), (m)); \
  76                } else \
  77                        sbappend((so), (m)); \
  78         } else {\
  79                (flags) = tcp_reass((tp), (ti), (m)); \
  80                tp->t_flags |= TF_ACKNOW; \
  81        } \
  82 }
  83 #else
  84 #define TCP_REASS(tp, ti, m, so, flags) { \
  85         if ((ti)->ti_seq == (tp)->rcv_nxt && \
  86         tcpfrag_list_empty(tp) && \
  87             (tp)->t_state == TCPS_ESTABLISHED) { \
  88                 tp->t_flags |= TF_DELACK; \
  89                 (tp)->rcv_nxt += (ti)->ti_len; \
  90                 flags = (ti)->ti_flags & TH_FIN; \
  91                 if (so->so_emu) { \
  92                         if (tcp_emu((so),(m))) sbappend(so, (m)); \
  93                 } else \
  94                         sbappend((so), (m)); \
  95         } else { \
  96                 (flags) = tcp_reass((tp), (ti), (m)); \
  97                 tp->t_flags |= TF_ACKNOW; \
  98         } \
  99 }
 100 #endif
 101 static void tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt,
 102                           struct tcpiphdr *ti);
 103 static void tcp_xmit_timer(register struct tcpcb *tp, int rtt);
 104
 105 static int
 106 tcp_reass(register struct tcpcb *tp, register struct tcpiphdr *ti,
 107           struct mbuf *m)
 108 {
 109         register struct tcpiphdr *q;
 110         struct socket *so = tp->t_socket;
 111         int flags;
 112
 113         /*
 114          * Call with ti==NULL after become established to
 115          * force pre-ESTABLISHED data up to user socket.
 116          */
 117         if (ti == NULL)
 118                 goto present;
 119
 120         /*
 121          * Find a segment which begins after this one does.
 122          */
 123         for (q = tcpfrag_list_first(tp); !tcpfrag_list_end(q, tp);
 124             q = tcpiphdr_next(q))
 125                 if (SEQ_GT(q->ti_seq, ti->ti_seq))
 126                         break;
 127
 128         /*
 129          * If there is a preceding segment, it may provide some of
 130          * our data already.  If so, drop the data from the incoming
 131          * segment.  If it provides all of our data, drop us.
 132          */
 133         if (!tcpfrag_list_end(tcpiphdr_prev(q), tp)) {
 134                 register int i;
 135                 q = tcpiphdr_prev(q);
 136                 /* conversion to int (in i) handles seq wraparound */
 137                 i = q->ti_seq + q->ti_len - ti->ti_seq;
 138                 if (i > 0) {
 139                         if (i >= ti->ti_len) {
 140                                 m_free(m);
 141                                 /*
 142                                  * Try to present any queued data
 143                                  * at the left window edge to the user.
 144                                  * This is needed after the 3-WHS
 145                                  * completes.
 146                                  */
 147                                 goto present;   /* ??? */
 148                         }
 149                         m_adj(m, i);
 150                         ti->ti_len -= i;
 151                         ti->ti_seq += i;
 152                 }
 153                 q = tcpiphdr_next(q);
 154         }
 155         ti->ti_mbuf = m;
 156
 157         /*
 158          * While we overlap succeeding segments trim them or,
 159          * if they are completely covered, dequeue them.
 160          */
 161         while (!tcpfrag_list_end(q, tp)) {
 162                 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq;
 163                 if (i <= 0)
 164                         break;
 165                 if (i < q->ti_len) {
 166                         q->ti_seq += i;
 167                         q->ti_len -= i;
 168                         m_adj(q->ti_mbuf, i);
 169                         break;
 170                 }
 171                 q = tcpiphdr_next(q);
 172                 m = tcpiphdr_prev(q)->ti_mbuf;
 173                 remque(tcpiphdr2qlink(tcpiphdr_prev(q)));
 174                 m_free(m);
 175         }
 176
 177         /*
 178          * Stick new segment in its place.
 179          */
 180         insque(tcpiphdr2qlink(ti), tcpiphdr2qlink(tcpiphdr_prev(q)));
 181
 182 present:
 183         /*
 184          * Present data to user, advancing rcv_nxt through
 185          * completed sequence space.
 186          */
 187         if (!TCPS_HAVEESTABLISHED(tp->t_state))
 188                 return (0);
 189         ti = tcpfrag_list_first(tp);
 190         if (tcpfrag_list_end(ti, tp) || ti->ti_seq != tp->rcv_nxt)
 191                 return (0);
 192         if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len)
 193                 return (0);
 194         do {
 195                 tp->rcv_nxt += ti->ti_len;
 196                 flags = ti->ti_flags & TH_FIN;
 197                 remque(tcpiphdr2qlink(ti));
 198                 m = ti->ti_mbuf;
 199                 ti = tcpiphdr_next(ti);
 200                 if (so->so_state & SS_FCANTSENDMORE)
 201                         m_free(m);
 202                 else {
 203                         if (so->so_emu) {
 204                                 if (tcp_emu(so,m)) sbappend(so, m);
 205                         } else
 206                                 sbappend(so, m);
 207                 }
 208         } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
 209         return (flags);
 210 }
 211
 212 /*
 213  * TCP input routine, follows pages 65-76 of the
 214  * protocol specification dated September, 1981 very closely.
 215  */
 216 void
 217 tcp_input(struct mbuf *m, int iphlen, struct socket *inso)
 218 {
 219         struct ip save_ip, *ip;
 220         register struct tcpiphdr *ti;
 221         caddr_t optp = NULL;
 222         int optlen = 0;
 223         int len, tlen, off;
 224         register struct tcpcb *tp = NULL;
 225         register int tiflags;
 226         struct socket *so = NULL;
 227         int todrop, acked, ourfinisacked, needoutput = 0;
 228         int iss = 0;
 229         u_long tiwin;
 230         int ret;
 231         struct sockaddr_storage lhost, fhost;
 232         struct sockaddr_in *lhost4, *fhost4;
 233     struct ex_list *ex_ptr;
 234     Slirp *slirp;
 235
 236         DEBUG_CALL("tcp_input");
 237         DEBUG_ARGS((dfd, " m = %p  iphlen = %2d  inso = %p\n",
 238                     m, iphlen, inso));
 239
 240         /*
 241          * If called with m == 0, then we're continuing the connect
 242          */
 243         if (m == NULL) {
 244                 so = inso;
 245                 slirp = so->slirp;
 246
 247                 /* Re-set a few variables */
 248                 tp = sototcpcb(so);
 249                 m = so->so_m;
 250                 so->so_m = NULL;
 251                 ti = so->so_ti;
 252                 tiwin = ti->ti_win;
 253                 tiflags = ti->ti_flags;
 254
 255                 goto cont_conn;
 256         }
 257         slirp = m->slirp;
 258
 259         /*
 260          * Get IP and TCP header together in first mbuf.
 261          * Note: IP leaves IP header in first mbuf.
 262          */
 263         ti = mtod(m, struct tcpiphdr *);
 264         if (iphlen > sizeof(struct ip )) {
 265           ip_stripoptions(m, (struct mbuf *)0);
 266           iphlen=sizeof(struct ip );
 267         }
 268         /* XXX Check if too short */
 269
 270
 271         /*
 272          * Save a copy of the IP header in case we want restore it
 273          * for sending an ICMP error message in response.
 274          */
 275         ip=mtod(m, struct ip *);
 276         save_ip = *ip;
 277         save_ip.ip_len+= iphlen;
 278
 279         /*
 280          * Checksum extended TCP header and data.
 281          */
 282         tlen = ((struct ip *)ti)->ip_len;
 283         tcpiphdr2qlink(ti)->next = tcpiphdr2qlink(ti)->prev = NULL;
 284         memset(&ti->ti_i.ih_mbuf, 0 , sizeof(struct mbuf_ptr));
 285         ti->ti_x1 = 0;
 286         ti->ti_len = htons((uint16_t)tlen);
 287         len = sizeof(struct ip ) + tlen;
 288         if(cksum(m, len)) {
 289           goto drop;
 290         }
 291
 292         /*
 293          * Check that TCP offset makes sense,
 294          * pull out TCP options and adjust length.              XXX
 295          */
 296         off = ti->ti_off << 2;
 297         if (off < sizeof (struct tcphdr) || off > tlen) {
 298           goto drop;
 299         }
 300         tlen -= off;
 301         ti->ti_len = tlen;
 302         if (off > sizeof (struct tcphdr)) {
 303           optlen = off - sizeof (struct tcphdr);
 304           optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
 305         }
 306         tiflags = ti->ti_flags;
 307
 308         /*
 309          * Convert TCP protocol specific fields to host format.
 310          */
 311         NTOHL(ti->ti_seq);
 312         NTOHL(ti->ti_ack);
 313         NTOHS(ti->ti_win);
 314         NTOHS(ti->ti_urp);
 315
 316         /*
 317          * Drop TCP, IP headers and TCP options.
 318          */
 319         m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
 320         m->m_len  -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
 321
 322         /*
 323          * Locate pcb for segment.
 324          */
 325 findso:
 326         lhost.ss_family = AF_INET;
 327         lhost4 = (struct sockaddr_in *) &lhost;
 328         lhost4->sin_addr = ti->ti_src;
 329         lhost4->sin_port = ti->ti_sport;
 330         fhost.ss_family = AF_INET;
 331         fhost4 = (struct sockaddr_in *) &fhost;
 332         fhost4->sin_addr = ti->ti_dst;
 333         fhost4->sin_port = ti->ti_dport;
 334
 335         so = solookup(&slirp->tcp_last_so, &slirp->tcb, &lhost, &fhost);
 336
 337         /*
 338          * If the state is CLOSED (i.e., TCB does not exist) then
 339          * all data in the incoming segment is discarded.
 340          * If the TCB exists but is in CLOSED state, it is embryonic,
 341          * but should either do a listen or a connect soon.
 342          *
 343          * state == CLOSED means we've done socreate() but haven't
 344          * attached it to a protocol yet...
 345          *
 346          * XXX If a TCB does not exist, and the TH_SYN flag is
 347          * the only flag set, then create a session, mark it
 348          * as if it was LISTENING, and continue...
 349          */
 350         if (so == NULL) {
 351           if (slirp->restricted) {
 352             /* Any hostfwds will have an existing socket, so we only get here
 353              * for non-hostfwd connections. These should be dropped, unless it
 354              * happens to be a guestfwd.
 355              */
 356             for (ex_ptr = slirp->exec_list; ex_ptr; ex_ptr = ex_ptr->ex_next) {
 357                 if (ex_ptr->ex_fport == ti->ti_dport &&
 358                     ti->ti_dst.s_addr == ex_ptr->ex_addr.s_addr) {
 359                     break;
 360                 }
 361             }
 362             if (!ex_ptr) {
 363                 goto dropwithreset;
 364             }
 365           }
 366
 367           if ((tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) != TH_SYN)
 368             goto dropwithreset;
 369
 370           if ((so = socreate(slirp)) == NULL)
 371             goto dropwithreset;
 372           if (tcp_attach(so) < 0) {
 373             free(so); /* Not sofree (if it failed, it's not insqued) */
 374             goto dropwithreset;
 375           }
 376
 377           sbreserve(&so->so_snd, TCP_SNDSPACE);
 378           sbreserve(&so->so_rcv, TCP_RCVSPACE);
 379
 380           so->lhost.ss = lhost;
 381           so->fhost.ss = fhost;
 382
 383           if ((so->so_iptos = tcp_tos(so)) == 0)
 384             so->so_iptos = ((struct ip *)ti)->ip_tos;
 385
 386           tp = sototcpcb(so);
 387           tp->t_state = TCPS_LISTEN;
 388         }
 389
 390         /*
 391          * If this is a still-connecting socket, this probably
 392          * a retransmit of the SYN.  Whether it's a retransmit SYN
 393          * or something else, we nuke it.
 394          */
 395         if (so->so_state & SS_ISFCONNECTING)
 396                 goto drop;
 397
 398         tp = sototcpcb(so);
 399
 400         /* XXX Should never fail */
 401         if (tp == NULL)
 402                 goto dropwithreset;
 403         if (tp->t_state == TCPS_CLOSED)
 404                 goto drop;
 405
 406         tiwin = ti->ti_win;
 407
 408         /*
 409          * Segment received on connection.
 410          * Reset idle time and keep-alive timer.
 411          */
 412         tp->t_idle = 0;
 413         if (SO_OPTIONS)
 414            tp->t_timer[TCPT_KEEP] = TCPTV_KEEPINTVL;
 415         else
 416            tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_IDLE;
 417
 418         /*
 419          * Process options if not in LISTEN state,
 420          * else do it below (after getting remote address).
 421          */
 422         if (optp && tp->t_state != TCPS_LISTEN)
 423                 tcp_dooptions(tp, (u_char *)optp, optlen, ti);
 424
 425         /*
 426          * Header prediction: check for the two common cases
 427          * of a uni-directional data xfer.  If the packet has
 428          * no control flags, is in-sequence, the window didn't
 429          * change and we're not retransmitting, it's a
 430          * candidate.  If the length is zero and the ack moved
 431          * forward, we're the sender side of the xfer.  Just
 432          * free the data acked & wake any higher level process
 433          * that was blocked waiting for space.  If the length
 434          * is non-zero and the ack didn't move, we're the
 435          * receiver side.  If we're getting packets in-order
 436          * (the reassembly queue is empty), add the data to
 437          * the socket buffer and note that we need a delayed ack.
 438          *
 439          * XXX Some of these tests are not needed
 440          * eg: the tiwin == tp->snd_wnd prevents many more
 441          * predictions.. with no *real* advantage..
 442          */
 443         if (tp->t_state == TCPS_ESTABLISHED &&
 444             (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
 445             ti->ti_seq == tp->rcv_nxt &&
 446             tiwin && tiwin == tp->snd_wnd &&
 447             tp->snd_nxt == tp->snd_max) {
 448                 if (ti->ti_len == 0) {
 449                         if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
 450                             SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
 451                             tp->snd_cwnd >= tp->snd_wnd) {
 452                                 /*
 453                                  * this is a pure ack for outstanding data.
 454                                  */
 455                                 if (tp->t_rtt &&
 456                                     SEQ_GT(ti->ti_ack, tp->t_rtseq))
 457                                         tcp_xmit_timer(tp, tp->t_rtt);
 458                                 acked = ti->ti_ack - tp->snd_una;
 459                                 sbdrop(&so->so_snd, acked);
 460                                 tp->snd_una = ti->ti_ack;
 461                                 m_free(m);
 462
 463                                 /*
 464                                  * If all outstanding data are acked, stop
 465                                  * retransmit timer, otherwise restart timer
 466                                  * using current (possibly backed-off) value.
 467                                  * If process is waiting for space,
 468                                  * wakeup/selwakeup/signal.  If data
 469                                  * are ready to send, let tcp_output
 470                                  * decide between more output or persist.
 471                                  */
 472                                 if (tp->snd_una == tp->snd_max)
 473                                         tp->t_timer[TCPT_REXMT] = 0;
 474                                 else if (tp->t_timer[TCPT_PERSIST] == 0)
 475                                         tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
 476
 477                                 /*
 478                                  * This is called because sowwakeup might have
 479                                  * put data into so_snd.  Since we don't so sowwakeup,
 480                                  * we don't need this.. XXX???
 481                                  */
 482                                 if (so->so_snd.sb_cc)
 483                                         (void) tcp_output(tp);
 484
 485                                 return;
 486                         }
 487                 } else if (ti->ti_ack == tp->snd_una &&
 488                     tcpfrag_list_empty(tp) &&
 489                     ti->ti_len <= sbspace(&so->so_rcv)) {
 490                         /*
 491                          * this is a pure, in-sequence data packet
 492                          * with nothing on the reassembly queue and
 493                          * we have enough buffer space to take it.
 494                          */
 495                         tp->rcv_nxt += ti->ti_len;
 496                         /*
 497                          * Add data to socket buffer.
 498                          */
 499                         if (so->so_emu) {
 500                                 if (tcp_emu(so,m)) sbappend(so, m);
 501                         } else
 502                                 sbappend(so, m);
 503
 504                         /*
 505                          * If this is a short packet, then ACK now - with Nagel
 506                          *      congestion avoidance sender won't send more until
 507                          *      he gets an ACK.
 508                          *
 509                          * It is better to not delay acks at all to maximize
 510                          * TCP throughput.  See RFC 2581.
 511                          */
 512                         tp->t_flags |= TF_ACKNOW;
 513                         tcp_output(tp);
 514                         return;
 515                 }
 516         } /* header prediction */
 517         /*
 518          * Calculate amount of space in receive window,
 519          * and then do TCP input processing.
 520          * Receive window is amount of space in rcv queue,
 521          * but not less than advertised window.
 522          */
 523         { int win;
 524           win = sbspace(&so->so_rcv);
 525           if (win < 0)
 526             win = 0;
 527           tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
 528         }
 529
 530         switch (tp->t_state) {
 531
 532         /*
 533          * If the state is LISTEN then ignore segment if it contains an RST.
 534          * If the segment contains an ACK then it is bad and send a RST.
 535          * If it does not contain a SYN then it is not interesting; drop it.
 536          * Don't bother responding if the destination was a broadcast.
 537          * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
 538          * tp->iss, and send a segment:
 539          *     <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
 540          * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
 541          * Fill in remote peer address fields if not previously specified.
 542          * Enter SYN_RECEIVED state, and process any other fields of this
 543          * segment in this state.
 544          */
 545         case TCPS_LISTEN: {
 546
 547           if (tiflags & TH_RST)
 548             goto drop;
 549           if (tiflags & TH_ACK)
 550             goto dropwithreset;
 551           if ((tiflags & TH_SYN) == 0)
 552             goto drop;
 553
 554           /*
 555            * This has way too many gotos...
 556            * But a bit of spaghetti code never hurt anybody :)
 557            */
 558
 559           /*
 560            * If this is destined for the control address, then flag to
 561            * tcp_ctl once connected, otherwise connect
 562            */
 563           if ((so->so_faddr.s_addr & slirp->vnetwork_mask.s_addr) ==
 564               slirp->vnetwork_addr.s_addr) {
 565             if (so->so_faddr.s_addr != slirp->vhost_addr.s_addr &&
 566                 so->so_faddr.s_addr != slirp->vnameserver_addr.s_addr) {
 567                 /* May be an add exec */
 568                 for (ex_ptr = slirp->exec_list; ex_ptr;
 569                      ex_ptr = ex_ptr->ex_next) {
 570                   if(ex_ptr->ex_fport == so->so_fport &&
 571                      so->so_faddr.s_addr == ex_ptr->ex_addr.s_addr) {
 572                     so->so_state |= SS_CTL;
 573                     break;
 574                   }
 575                 }
 576                 if (so->so_state & SS_CTL) {
 577                     goto cont_input;
 578                 }
 579             }
 580             /* CTL_ALIAS: Do nothing, tcp_fconnect will be called on it */
 581           }
 582
 583           if (so->so_emu & EMU_NOCONNECT) {
 584             so->so_emu &= ~EMU_NOCONNECT;
 585             goto cont_input;
 586           }
 587
 588           if ((tcp_fconnect(so, so->so_ffamily) == -1) &&
 589 #if defined(_WIN32)
 590               socket_error() != WSAEWOULDBLOCK
 591 #else
 592               (errno != EINPROGRESS) && (errno != EWOULDBLOCK)
 593 #endif
 594           ) {
 595             u_char code=ICMP_UNREACH_NET;
 596             DEBUG_MISC((dfd, " tcp fconnect errno = %d-%s\n",
 597                         errno,strerror(errno)));
 598             if(errno == ECONNREFUSED) {
 599               /* ACK the SYN, send RST to refuse the connection */
 600               tcp_respond(tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
 601                           TH_RST|TH_ACK);
 602             } else {
 603               if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
 604               HTONL(ti->ti_seq);             /* restore tcp header */
 605               HTONL(ti->ti_ack);
 606               HTONS(ti->ti_win);
 607               HTONS(ti->ti_urp);
 608               m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
 609               m->m_len  += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
 610               *ip=save_ip;
 611               icmp_error(m, ICMP_UNREACH,code, 0,strerror(errno));
 612             }
 613             tcp_close(tp);
 614             m_free(m);
 615           } else {
 616             /*
 617              * Haven't connected yet, save the current mbuf
 618              * and ti, and return
 619              * XXX Some OS's don't tell us whether the connect()
 620              * succeeded or not.  So we must time it out.
 621              */
 622             so->so_m = m;
 623             so->so_ti = ti;
 624             tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
 625             tp->t_state = TCPS_SYN_RECEIVED;
 626             tcp_template(tp);
 627           }
 628           return;
 629
 630         cont_conn:
 631           /* m==NULL
 632            * Check if the connect succeeded
 633            */
 634           if (so->so_state & SS_NOFDREF) {
 635             tp = tcp_close(tp);
 636             goto dropwithreset;
 637           }
 638         cont_input:
 639           tcp_template(tp);
 640
 641           if (optp)
 642             tcp_dooptions(tp, (u_char *)optp, optlen, ti);
 643
 644           if (iss)
 645             tp->iss = iss;
 646           else
 647             tp->iss = slirp->tcp_iss;
 648           slirp->tcp_iss += TCP_ISSINCR/2;
 649           tp->irs = ti->ti_seq;
 650           tcp_sendseqinit(tp);
 651           tcp_rcvseqinit(tp);
 652           tp->t_flags |= TF_ACKNOW;
 653           tp->t_state = TCPS_SYN_RECEIVED;
 654           tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
 655           goto trimthenstep6;
 656         } /* case TCPS_LISTEN */
 657
 658         /*
 659          * If the state is SYN_SENT:
 660          *      if seg contains an ACK, but not for our SYN, drop the input.
 661          *      if seg contains a RST, then drop the connection.
 662          *      if seg does not contain SYN, then drop it.
 663          * Otherwise this is an acceptable SYN segment
 664          *      initialize tp->rcv_nxt and tp->irs
 665          *      if seg contains ack then advance tp->snd_una
 666          *      if SYN has been acked change to ESTABLISHED else SYN_RCVD state
 667          *      arrange for segment to be acked (eventually)
 668          *      continue processing rest of data/controls, beginning with URG
 669          */
 670         case TCPS_SYN_SENT:
 671                 if ((tiflags & TH_ACK) &&
 672                     (SEQ_LEQ(ti->ti_ack, tp->iss) ||
 673                      SEQ_GT(ti->ti_ack, tp->snd_max)))
 674                         goto dropwithreset;
 675
 676                 if (tiflags & TH_RST) {
 677                         if (tiflags & TH_ACK) {
 678                                 tcp_drop(tp, 0); /* XXX Check t_softerror! */
 679                         }
 680                         goto drop;
 681                 }
 682
 683                 if ((tiflags & TH_SYN) == 0)
 684                         goto drop;
 685                 if (tiflags & TH_ACK) {
 686                         tp->snd_una = ti->ti_ack;
 687                         if (SEQ_LT(tp->snd_nxt, tp->snd_una))
 688                                 tp->snd_nxt = tp->snd_una;
 689                 }
 690
 691                 tp->t_timer[TCPT_REXMT] = 0;
 692                 tp->irs = ti->ti_seq;
 693                 tcp_rcvseqinit(tp);
 694                 tp->t_flags |= TF_ACKNOW;
 695                 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
 696                         soisfconnected(so);
 697                         tp->t_state = TCPS_ESTABLISHED;
 698
 699                         (void) tcp_reass(tp, (struct tcpiphdr *)0,
 700                                 (struct mbuf *)0);
 701                         /*
 702                          * if we didn't have to retransmit the SYN,
 703                          * use its rtt as our initial srtt & rtt var.
 704                          */
 705                         if (tp->t_rtt)
 706                                 tcp_xmit_timer(tp, tp->t_rtt);
 707                 } else
 708                         tp->t_state = TCPS_SYN_RECEIVED;
 709
 710 trimthenstep6:
 711                 /*
 712                  * Advance ti->ti_seq to correspond to first data byte.
 713                  * If data, trim to stay within window,
 714                  * dropping FIN if necessary.
 715                  */
 716                 ti->ti_seq++;
 717                 if (ti->ti_len > tp->rcv_wnd) {
 718                         todrop = ti->ti_len - tp->rcv_wnd;
 719                         m_adj(m, -todrop);
 720                         ti->ti_len = tp->rcv_wnd;
 721                         tiflags &= ~TH_FIN;
 722                 }
 723                 tp->snd_wl1 = ti->ti_seq - 1;
 724                 tp->rcv_up = ti->ti_seq;
 725                 goto step6;
 726         } /* switch tp->t_state */
 727         /*
 728          * States other than LISTEN or SYN_SENT.
 729          * Check that at least some bytes of segment are within
 730          * receive window.  If segment begins before rcv_nxt,
 731          * drop leading data (and SYN); if nothing left, just ack.
 732          */
 733         todrop = tp->rcv_nxt - ti->ti_seq;
 734         if (todrop > 0) {
 735                 if (tiflags & TH_SYN) {
 736                         tiflags &= ~TH_SYN;
 737                         ti->ti_seq++;
 738                         if (ti->ti_urp > 1)
 739                                 ti->ti_urp--;
 740                         else
 741                                 tiflags &= ~TH_URG;
 742                         todrop--;
 743                 }
 744                 /*
 745                  * Following if statement from Stevens, vol. 2, p. 960.
 746                  */
 747                 if (todrop > ti->ti_len
 748                     || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) {
 749                         /*
 750                          * Any valid FIN must be to the left of the window.
 751                          * At this point the FIN must be a duplicate or out
 752                          * of sequence; drop it.
 753                          */
 754                         tiflags &= ~TH_FIN;
 755
 756                         /*
 757                          * Send an ACK to resynchronize and drop any data.
 758                          * But keep on processing for RST or ACK.
 759                          */
 760                         tp->t_flags |= TF_ACKNOW;
 761                         todrop = ti->ti_len;
 762                 }
 763                 m_adj(m, todrop);
 764                 ti->ti_seq += todrop;
 765                 ti->ti_len -= todrop;
 766                 if (ti->ti_urp > todrop)
 767                         ti->ti_urp -= todrop;
 768                 else {
 769                         tiflags &= ~TH_URG;
 770                         ti->ti_urp = 0;
 771                 }
 772         }
 773         /*
 774          * If new data are received on a connection after the
 775          * user processes are gone, then RST the other end.
 776          */
 777         if ((so->so_state & SS_NOFDREF) &&
 778             tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
 779                 tp = tcp_close(tp);
 780                 goto dropwithreset;
 781         }
 782
 783         /*
 784          * If segment ends after window, drop trailing data
 785          * (and PUSH and FIN); if nothing left, just ACK.
 786          */
 787         todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
 788         if (todrop > 0) {
 789                 if (todrop >= ti->ti_len) {
 790                         /*
 791                          * If a new connection request is received
 792                          * while in TIME_WAIT, drop the old connection
 793                          * and start over if the sequence numbers
 794                          * are above the previous ones.
 795                          */
 796                         if (tiflags & TH_SYN &&
 797                             tp->t_state == TCPS_TIME_WAIT &&
 798                             SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
 799                                 iss = tp->rcv_nxt + TCP_ISSINCR;
 800                                 tp = tcp_close(tp);
 801                                 goto findso;
 802                         }
 803                         /*
 804                          * If window is closed can only take segments at
 805                          * window edge, and have to drop data and PUSH from
 806                          * incoming segments.  Continue processing, but
 807                          * remember to ack.  Otherwise, drop segment
 808                          * and ack.
 809                          */
 810                         if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
 811                                 tp->t_flags |= TF_ACKNOW;
 812                         } else {
 813                                 goto dropafterack;
 814                         }
 815                 }
 816                 m_adj(m, -todrop);
 817                 ti->ti_len -= todrop;
 818                 tiflags &= ~(TH_PUSH|TH_FIN);
 819         }
 820
 821         /*
 822          * If the RST bit is set examine the state:
 823          *    SYN_RECEIVED STATE:
 824          *      If passive open, return to LISTEN state.
 825          *      If active open, inform user that connection was refused.
 826          *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
 827          *      Inform user that connection was reset, and close tcb.
 828          *    CLOSING, LAST_ACK, TIME_WAIT STATES
 829          *      Close the tcb.
 830          */
 831         if (tiflags&TH_RST) switch (tp->t_state) {
 832
 833         case TCPS_SYN_RECEIVED:
 834         case TCPS_ESTABLISHED:
 835         case TCPS_FIN_WAIT_1:
 836         case TCPS_FIN_WAIT_2:
 837         case TCPS_CLOSE_WAIT:
 838                 tp->t_state = TCPS_CLOSED;
 839                 tcp_close(tp);
 840                 goto drop;
 841
 842         case TCPS_CLOSING:
 843         case TCPS_LAST_ACK:
 844         case TCPS_TIME_WAIT:
 845                 tcp_close(tp);
 846                 goto drop;
 847         }
 848
 849         /*
 850          * If a SYN is in the window, then this is an
 851          * error and we send an RST and drop the connection.
 852          */
 853         if (tiflags & TH_SYN) {
 854                 tp = tcp_drop(tp,0);
 855                 goto dropwithreset;
 856         }
 857
 858         /*
 859          * If the ACK bit is off we drop the segment and return.
 860          */
 861         if ((tiflags & TH_ACK) == 0) goto drop;
 862
 863         /*
 864          * Ack processing.
 865          */
 866         switch (tp->t_state) {
 867         /*
 868          * In SYN_RECEIVED state if the ack ACKs our SYN then enter
 869          * ESTABLISHED state and continue processing, otherwise
 870          * send an RST.  una<=ack<=max
 871          */
 872         case TCPS_SYN_RECEIVED:
 873
 874                 if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
 875                     SEQ_GT(ti->ti_ack, tp->snd_max))
 876                         goto dropwithreset;
 877                 tp->t_state = TCPS_ESTABLISHED;
 878                 /*
 879                  * The sent SYN is ack'ed with our sequence number +1
 880                  * The first data byte already in the buffer will get
 881                  * lost if no correction is made.  This is only needed for
 882                  * SS_CTL since the buffer is empty otherwise.
 883                  * tp->snd_una++; or:
 884                  */
 885                 tp->snd_una=ti->ti_ack;
 886                 if (so->so_state & SS_CTL) {
 887                   /* So tcp_ctl reports the right state */
 888                   ret = tcp_ctl(so);
 889                   if (ret == 1) {
 890                     soisfconnected(so);
 891                     so->so_state &= ~SS_CTL;   /* success XXX */
 892                   } else if (ret == 2) {
 893                     so->so_state &= SS_PERSISTENT_MASK;
 894                     so->so_state |= SS_NOFDREF; /* CTL_CMD */
 895                   } else {
 896                     needoutput = 1;
 897                     tp->t_state = TCPS_FIN_WAIT_1;
 898                   }
 899                 } else {
 900                   soisfconnected(so);
 901                 }
 902
 903                 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0);
 904                 tp->snd_wl1 = ti->ti_seq - 1;
 905                 /* Avoid ack processing; snd_una==ti_ack  =>  dup ack */
 906                 goto synrx_to_est;
 907                 /* fall into ... */
 908
 909         /*
 910          * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
 911          * ACKs.  If the ack is in the range
 912          *      tp->snd_una < ti->ti_ack <= tp->snd_max
 913          * then advance tp->snd_una to ti->ti_ack and drop
 914          * data from the retransmission queue.  If this ACK reflects
 915          * more up to date window information we update our window information.
 916          */
 917         case TCPS_ESTABLISHED:
 918         case TCPS_FIN_WAIT_1:
 919         case TCPS_FIN_WAIT_2:
 920         case TCPS_CLOSE_WAIT:
 921         case TCPS_CLOSING:
 922         case TCPS_LAST_ACK:
 923         case TCPS_TIME_WAIT:
 924
 925                 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
 926                         if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
 927                           DEBUG_MISC((dfd, " dup ack  m = %p  so = %p\n",
 928                                       m, so));
 929                                 /*
 930                                  * If we have outstanding data (other than
 931                                  * a window probe), this is a completely
 932                                  * duplicate ack (ie, window info didn't
 933                                  * change), the ack is the biggest we've
 934                                  * seen and we've seen exactly our rexmt
 935                                  * threshold of them, assume a packet
 936                                  * has been dropped and retransmit it.
 937                                  * Kludge snd_nxt & the congestion
 938                                  * window so we send only this one
 939                                  * packet.
 940                                  *
 941                                  * We know we're losing at the current
 942                                  * window size so do congestion avoidance
 943                                  * (set ssthresh to half the current window
 944                                  * and pull our congestion window back to
 945                                  * the new ssthresh).
 946                                  *
 947                                  * Dup acks mean that packets have left the
 948                                  * network (they're now cached at the receiver)
 949                                  * so bump cwnd by the amount in the receiver
 950                                  * to keep a constant cwnd packets in the
 951                                  * network.
 952                                  */
 953                                 if (tp->t_timer[TCPT_REXMT] == 0 ||
 954                                     ti->ti_ack != tp->snd_una)
 955                                         tp->t_dupacks = 0;
 956                                 else if (++tp->t_dupacks == TCPREXMTTHRESH) {
 957                                         tcp_seq onxt = tp->snd_nxt;
 958                                         u_int win =
 959                                             min(tp->snd_wnd, tp->snd_cwnd) / 2 /
 960                                                 tp->t_maxseg;
 961
 962                                         if (win < 2)
 963                                                 win = 2;
 964                                         tp->snd_ssthresh = win * tp->t_maxseg;
 965                                         tp->t_timer[TCPT_REXMT] = 0;
 966                                         tp->t_rtt = 0;
 967                                         tp->snd_nxt = ti->ti_ack;
 968                                         tp->snd_cwnd = tp->t_maxseg;
 969                                         (void) tcp_output(tp);
 970                                         tp->snd_cwnd = tp->snd_ssthresh +
 971                                                tp->t_maxseg * tp->t_dupacks;
 972                                         if (SEQ_GT(onxt, tp->snd_nxt))
 973                                                 tp->snd_nxt = onxt;
 974                                         goto drop;
 975                                 } else if (tp->t_dupacks > TCPREXMTTHRESH) {
 976                                         tp->snd_cwnd += tp->t_maxseg;
 977                                         (void) tcp_output(tp);
 978                                         goto drop;
 979                                 }
 980                         } else
 981                                 tp->t_dupacks = 0;
 982                         break;
 983                 }
 984         synrx_to_est:
 985                 /*
 986                  * If the congestion window was inflated to account
 987                  * for the other side's cached packets, retract it.
 988                  */
 989                 if (tp->t_dupacks > TCPREXMTTHRESH &&
 990                     tp->snd_cwnd > tp->snd_ssthresh)
 991                         tp->snd_cwnd = tp->snd_ssthresh;
 992                 tp->t_dupacks = 0;
 993                 if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
 994                         goto dropafterack;
 995                 }
 996                 acked = ti->ti_ack - tp->snd_una;
 997
 998                 /*
 999                  * If transmit timer is running and timed sequence
1000                  * number was acked, update smoothed round trip time.
1001                  * Since we now have an rtt measurement, cancel the
1002                  * timer backoff (cf., Phil Karn's retransmit alg.).
1003                  * Recompute the initial retransmit timer.
1004                  */
1005                 if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1006                         tcp_xmit_timer(tp,tp->t_rtt);
1007
1008                 /*
1009                  * If all outstanding data is acked, stop retransmit
1010                  * timer and remember to restart (more output or persist).
1011                  * If there is more data to be acked, restart retransmit
1012                  * timer, using current (possibly backed-off) value.
1013                  */
1014                 if (ti->ti_ack == tp->snd_max) {
1015                         tp->t_timer[TCPT_REXMT] = 0;
1016                         needoutput = 1;
1017                 } else if (tp->t_timer[TCPT_PERSIST] == 0)
1018                         tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1019                 /*
1020                  * When new data is acked, open the congestion window.
1021                  * If the window gives us less than ssthresh packets
1022                  * in flight, open exponentially (maxseg per packet).
1023                  * Otherwise open linearly: maxseg per window
1024                  * (maxseg^2 / cwnd per packet).
1025                  */
1026                 {
1027                   register u_int cw = tp->snd_cwnd;
1028                   register u_int incr = tp->t_maxseg;
1029
1030                   if (cw > tp->snd_ssthresh)
1031                     incr = incr * incr / cw;
1032                   tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1033                 }
1034                 if (acked > so->so_snd.sb_cc) {
1035                         tp->snd_wnd -= so->so_snd.sb_cc;
1036                         sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1037                         ourfinisacked = 1;
1038                 } else {
1039                         sbdrop(&so->so_snd, acked);
1040                         tp->snd_wnd -= acked;
1041                         ourfinisacked = 0;
1042                 }
1043                 tp->snd_una = ti->ti_ack;
1044                 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1045                         tp->snd_nxt = tp->snd_una;
1046
1047                 switch (tp->t_state) {
1048
1049                 /*
1050                  * In FIN_WAIT_1 STATE in addition to the processing
1051                  * for the ESTABLISHED state if our FIN is now acknowledged
1052                  * then enter FIN_WAIT_2.
1053                  */
1054                 case TCPS_FIN_WAIT_1:
1055                         if (ourfinisacked) {
1056                                 /*
1057                                  * If we can't receive any more
1058                                  * data, then closing user can proceed.
1059                                  * Starting the timer is contrary to the
1060                                  * specification, but if we don't get a FIN
1061                                  * we'll hang forever.
1062                                  */
1063                                 if (so->so_state & SS_FCANTRCVMORE) {
1064                                         tp->t_timer[TCPT_2MSL] = TCP_MAXIDLE;
1065                                 }
1066                                 tp->t_state = TCPS_FIN_WAIT_2;
1067                         }
1068                         break;
1069
1070                 /*
1071                  * In CLOSING STATE in addition to the processing for
1072                  * the ESTABLISHED state if the ACK acknowledges our FIN
1073                  * then enter the TIME-WAIT state, otherwise ignore
1074                  * the segment.
1075                  */
1076                 case TCPS_CLOSING:
1077                         if (ourfinisacked) {
1078                                 tp->t_state = TCPS_TIME_WAIT;
1079                                 tcp_canceltimers(tp);
1080                                 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1081                         }
1082                         break;
1083
1084                 /*
1085                  * In LAST_ACK, we may still be waiting for data to drain
1086                  * and/or to be acked, as well as for the ack of our FIN.
1087                  * If our FIN is now acknowledged, delete the TCB,
1088                  * enter the closed state and return.
1089                  */
1090                 case TCPS_LAST_ACK:
1091                         if (ourfinisacked) {
1092                                 tcp_close(tp);
1093                                 goto drop;
1094                         }
1095                         break;
1096
1097                 /*
1098                  * In TIME_WAIT state the only thing that should arrive
1099                  * is a retransmission of the remote FIN.  Acknowledge
1100                  * it and restart the finack timer.
1101                  */
1102                 case TCPS_TIME_WAIT:
1103                         tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1104                         goto dropafterack;
1105                 }
1106         } /* switch(tp->t_state) */
1107
1108 step6:
1109         /*
1110          * Update window information.
1111          * Don't look at window if no ACK: TAC's send garbage on first SYN.
1112          */
1113         if ((tiflags & TH_ACK) &&
1114             (SEQ_LT(tp->snd_wl1, ti->ti_seq) ||
1115             (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
1116             (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) {
1117                 tp->snd_wnd = tiwin;
1118                 tp->snd_wl1 = ti->ti_seq;
1119                 tp->snd_wl2 = ti->ti_ack;
1120                 if (tp->snd_wnd > tp->max_sndwnd)
1121                         tp->max_sndwnd = tp->snd_wnd;
1122                 needoutput = 1;
1123         }
1124
1125         /*
1126          * Process segments with URG.
1127          */
1128         if ((tiflags & TH_URG) && ti->ti_urp &&
1129             TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1130                 /*
1131                  * This is a kludge, but if we receive and accept
1132                  * random urgent pointers, we'll crash in
1133                  * soreceive.  It's hard to imagine someone
1134                  * actually wanting to send this much urgent data.
1135                  */
1136                 if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen) {
1137                         ti->ti_urp = 0;
1138                         tiflags &= ~TH_URG;
1139                         goto dodata;
1140                 }
1141                 /*
1142                  * If this segment advances the known urgent pointer,
1143                  * then mark the data stream.  This should not happen
1144                  * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1145                  * a FIN has been received from the remote side.
1146                  * In these states we ignore the URG.
1147                  *
1148                  * According to RFC961 (Assigned Protocols),
1149                  * the urgent pointer points to the last octet
1150                  * of urgent data.  We continue, however,
1151                  * to consider it to indicate the first octet
1152                  * of data past the urgent section as the original
1153                  * spec states (in one of two places).
1154                  */
1155                 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1156                         tp->rcv_up = ti->ti_seq + ti->ti_urp;
1157                         so->so_urgc =  so->so_rcv.sb_cc +
1158                                 (tp->rcv_up - tp->rcv_nxt); /* -1; */
1159                         tp->rcv_up = ti->ti_seq + ti->ti_urp;
1160
1161                 }
1162         } else
1163                 /*
1164                  * If no out of band data is expected,
1165                  * pull receive urgent pointer along
1166                  * with the receive window.
1167                  */
1168                 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1169                         tp->rcv_up = tp->rcv_nxt;
1170 dodata:
1171
1172         /*
1173          * If this is a small packet, then ACK now - with Nagel
1174          *      congestion avoidance sender won't send more until
1175          *      he gets an ACK.
1176          */
1177         if (ti->ti_len && (unsigned)ti->ti_len <= 5 &&
1178             ((struct tcpiphdr_2 *)ti)->first_char == (char)27) {
1179                 tp->t_flags |= TF_ACKNOW;
1180         }
1181
1182         /*
1183          * Process the segment text, merging it into the TCP sequencing queue,
1184          * and arranging for acknowledgment of receipt if necessary.
1185          * This process logically involves adjusting tp->rcv_wnd as data
1186          * is presented to the user (this happens in tcp_usrreq.c,
1187          * case PRU_RCVD).  If a FIN has already been received on this
1188          * connection then we just ignore the text.
1189          */
1190         if ((ti->ti_len || (tiflags&TH_FIN)) &&
1191             TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1192                 TCP_REASS(tp, ti, m, so, tiflags);
1193         } else {
1194                 m_free(m);
1195                 tiflags &= ~TH_FIN;
1196         }
1197
1198         /*
1199          * If FIN is received ACK the FIN and let the user know
1200          * that the connection is closing.
1201          */
1202         if (tiflags & TH_FIN) {
1203                 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1204                         /*
1205                          * If we receive a FIN we can't send more data,
1206                          * set it SS_FDRAIN
1207                          * Shutdown the socket if there is no rx data in the
1208                          * buffer.
1209                          * soread() is called on completion of shutdown() and
1210                          * will got to TCPS_LAST_ACK, and use tcp_output()
1211                          * to send the FIN.
1212                          */
1213                         sofwdrain(so);
1214
1215                         tp->t_flags |= TF_ACKNOW;
1216                         tp->rcv_nxt++;
1217                 }
1218                 switch (tp->t_state) {
1219
1220                 /*
1221                  * In SYN_RECEIVED and ESTABLISHED STATES
1222                  * enter the CLOSE_WAIT state.
1223                  */
1224                 case TCPS_SYN_RECEIVED:
1225                 case TCPS_ESTABLISHED:
1226                   if(so->so_emu == EMU_CTL)        /* no shutdown on socket */
1227                     tp->t_state = TCPS_LAST_ACK;
1228                   else
1229                     tp->t_state = TCPS_CLOSE_WAIT;
1230                   break;
1231
1232                 /*
1233                  * If still in FIN_WAIT_1 STATE FIN has not been acked so
1234                  * enter the CLOSING state.
1235                  */
1236                 case TCPS_FIN_WAIT_1:
1237                         tp->t_state = TCPS_CLOSING;
1238                         break;
1239
1240                 /*
1241                  * In FIN_WAIT_2 state enter the TIME_WAIT state,
1242                  * starting the time-wait timer, turning off the other
1243                  * standard timers.
1244                  */
1245                 case TCPS_FIN_WAIT_2:
1246                         tp->t_state = TCPS_TIME_WAIT;
1247                         tcp_canceltimers(tp);
1248                         tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1249                         break;
1250
1251                 /*
1252                  * In TIME_WAIT state restart the 2 MSL time_wait timer.
1253                  */
1254                 case TCPS_TIME_WAIT:
1255                         tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1256                         break;
1257                 }
1258         }
1259
1260         /*
1261          * Return any desired output.
1262          */
1263         if (needoutput || (tp->t_flags & TF_ACKNOW)) {
1264                 (void) tcp_output(tp);
1265         }
1266         return;
1267
1268 dropafterack:
1269         /*
1270          * Generate an ACK dropping incoming segment if it occupies
1271          * sequence space, where the ACK reflects our state.
1272          */
1273         if (tiflags & TH_RST)
1274                 goto drop;
1275         m_free(m);
1276         tp->t_flags |= TF_ACKNOW;
1277         (void) tcp_output(tp);
1278         return;
1279
1280 dropwithreset:
1281         /* reuses m if m!=NULL, m_free() unnecessary */
1282         if (tiflags & TH_ACK)
1283                 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1284         else {
1285                 if (tiflags & TH_SYN) ti->ti_len++;
1286                 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1287                     TH_RST|TH_ACK);
1288         }
1289
1290         return;
1291
1292 drop:
1293         /*
1294          * Drop space held by incoming segment and return.
1295          */
1296         m_free(m);
1297 }
1298
1299 static void
1300 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcpiphdr *ti)
1301 {
1302         uint16_t mss;
1303         int opt, optlen;
1304
1305         DEBUG_CALL("tcp_dooptions");
1306         DEBUG_ARGS((dfd, " tp = %p  cnt=%i\n", tp, cnt));
1307
1308         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1309                 opt = cp[0];
1310                 if (opt == TCPOPT_EOL)
1311                         break;
1312                 if (opt == TCPOPT_NOP)
1313                         optlen = 1;
1314                 else {
1315                         optlen = cp[1];
1316                         if (optlen <= 0)
1317                                 break;
1318                 }
1319                 switch (opt) {
1320
1321                 default:
1322                         continue;
1323
1324                 case TCPOPT_MAXSEG:
1325                         if (optlen != TCPOLEN_MAXSEG)
1326                                 continue;
1327                         if (!(ti->ti_flags & TH_SYN))
1328                                 continue;
1329                         memcpy((char *) &mss, (char *) cp + 2, sizeof(mss));
1330                         NTOHS(mss);
1331                         (void) tcp_mss(tp, mss);        /* sets t_maxseg */
1332                         break;
1333                 }
1334         }
1335 }
1336
1337
1338 /*
1339  * Pull out of band byte out of a segment so
1340  * it doesn't appear in the user's data queue.
1341  * It is still reflected in the segment length for
1342  * sequencing purposes.
1343  */
1344
1345 #ifdef notdef
1346
1347 void
1348 tcp_pulloutofband(so, ti, m)
1349         struct socket *so;
1350         struct tcpiphdr *ti;
1351         register struct mbuf *m;
1352 {
1353         int cnt = ti->ti_urp - 1;
1354
1355         while (cnt >= 0) {
1356                 if (m->m_len > cnt) {
1357                         char *cp = mtod(m, caddr_t) + cnt;
1358                         struct tcpcb *tp = sototcpcb(so);
1359
1360                         tp->t_iobc = *cp;
1361                         tp->t_oobflags |= TCPOOB_HAVEDATA;
1362                         memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1363                         m->m_len--;
1364                         return;
1365                 }
1366                 cnt -= m->m_len;
1367                 m = m->m_next; /* XXX WRONG! Fix it! */
1368                 if (m == 0)
1369                         break;
1370         }
1371         panic("tcp_pulloutofband");
1372 }
1373
1374 #endif /* notdef */
1375
1376 /*
1377  * Collect new round-trip time estimate
1378  * and update averages and current timeout.
1379  */
1380
1381 static void
1382 tcp_xmit_timer(register struct tcpcb *tp, int rtt)
1383 {
1384         register short delta;
1385
1386         DEBUG_CALL("tcp_xmit_timer");
1387         DEBUG_ARG("tp = %p", tp);
1388         DEBUG_ARG("rtt = %d", rtt);
1389
1390         if (tp->t_srtt != 0) {
1391                 /*
1392                  * srtt is stored as fixed point with 3 bits after the
1393                  * binary point (i.e., scaled by 8).  The following magic
1394                  * is equivalent to the smoothing algorithm in rfc793 with
1395                  * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1396                  * point).  Adjust rtt to origin 0.
1397                  */
1398                 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1399                 if ((tp->t_srtt += delta) <= 0)
1400                         tp->t_srtt = 1;
1401                 /*
1402                  * We accumulate a smoothed rtt variance (actually, a
1403                  * smoothed mean difference), then set the retransmit
1404                  * timer to smoothed rtt + 4 times the smoothed variance.
1405                  * rttvar is stored as fixed point with 2 bits after the
1406                  * binary point (scaled by 4).  The following is
1407                  * equivalent to rfc793 smoothing with an alpha of .75
1408                  * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
1409                  * rfc793's wired-in beta.
1410                  */
1411                 if (delta < 0)
1412                         delta = -delta;
1413                 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1414                 if ((tp->t_rttvar += delta) <= 0)
1415                         tp->t_rttvar = 1;
1416         } else {
1417                 /*
1418                  * No rtt measurement yet - use the unsmoothed rtt.
1419                  * Set the variance to half the rtt (so our first
1420                  * retransmit happens at 3*rtt).
1421                  */
1422                 tp->t_srtt = rtt << TCP_RTT_SHIFT;
1423                 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1424         }
1425         tp->t_rtt = 0;
1426         tp->t_rxtshift = 0;
1427
1428         /*
1429          * the retransmit should happen at rtt + 4 * rttvar.
1430          * Because of the way we do the smoothing, srtt and rttvar
1431          * will each average +1/2 tick of bias.  When we compute
1432          * the retransmit timer, we want 1/2 tick of rounding and
1433          * 1 extra tick because of +-1/2 tick uncertainty in the
1434          * firing of the timer.  The bias will give us exactly the
1435          * 1.5 tick we need.  But, because the bias is
1436          * statistical, we have to test that we don't drop below
1437          * the minimum feasible timer (which is 2 ticks).
1438          */
1439         TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1440             (short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1441
1442         /*
1443          * We received an ack for a packet that wasn't retransmitted;
1444          * it is probably safe to discard any error indications we've
1445          * received recently.  This isn't quite right, but close enough
1446          * for now (a route might have failed after we sent a segment,
1447          * and the return path might not be symmetrical).
1448          */
1449         tp->t_softerror = 0;
1450 }
1451
1452 /*
1453  * Determine a reasonable value for maxseg size.
1454  * If the route is known, check route for mtu.
1455  * If none, use an mss that can be handled on the outgoing
1456  * interface without forcing IP to fragment; if bigger than
1457  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1458  * to utilize large mbufs.  If no route is found, route has no mtu,
1459  * or the destination isn't local, use a default, hopefully conservative
1460  * size (usually 512 or the default IP max size, but no more than the mtu
1461  * of the interface), as we can't discover anything about intervening
1462  * gateways or networks.  We also initialize the congestion/slow start
1463  * window to be a single segment if the destination isn't local.
1464  * While looking at the routing entry, we also initialize other path-dependent
1465  * parameters from pre-set or cached values in the routing entry.
1466  */
1467
1468 int
1469 tcp_mss(struct tcpcb *tp, u_int offer)
1470 {
1471         struct socket *so = tp->t_socket;
1472         int mss;
1473
1474         DEBUG_CALL("tcp_mss");
1475         DEBUG_ARG("tp = %p", tp);
1476         DEBUG_ARG("offer = %d", offer);
1477
1478         mss = min(IF_MTU, IF_MRU) - sizeof(struct tcpiphdr);
1479         if (offer)
1480                 mss = min(mss, offer);
1481         mss = max(mss, 32);
1482         if (mss < tp->t_maxseg || offer != 0)
1483            tp->t_maxseg = mss;
1484
1485         tp->snd_cwnd = mss;
1486
1487         sbreserve(&so->so_snd, TCP_SNDSPACE + ((TCP_SNDSPACE % mss) ?
1488                                                (mss - (TCP_SNDSPACE % mss)) :
1489                                                0));
1490         sbreserve(&so->so_rcv, TCP_RCVSPACE + ((TCP_RCVSPACE % mss) ?
1491                                                (mss - (TCP_RCVSPACE % mss)) :
1492                                                0));
1493
1494         DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1495
1496         return mss;
1497 }