qemu/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, unsigned short af)
 218 {
 219         struct ip save_ip, *ip;
 220         struct ip6 save_ip6, *ip6;
 221         register struct tcpiphdr *ti;
 222         caddr_t optp = NULL;
 223         int optlen = 0;
 224         int len, tlen, off;
 225         register struct tcpcb *tp = NULL;
 226         register int tiflags;
 227         struct socket *so = NULL;
 228         int todrop, acked, ourfinisacked, needoutput = 0;
 229         int iss = 0;
 230         u_long tiwin;
 231         int ret;
 232         struct sockaddr_storage lhost, fhost;
 233         struct sockaddr_in *lhost4, *fhost4;
 234         struct sockaddr_in6 *lhost6, *fhost6;
 235     struct ex_list *ex_ptr;
 236     Slirp *slirp;
 237
 238         DEBUG_CALL("tcp_input");
 239         DEBUG_ARGS((dfd, " m = %p  iphlen = %2d  inso = %p\n",
 240                     m, iphlen, inso));
 241
 242         /*
 243          * If called with m == 0, then we're continuing the connect
 244          */
 245         if (m == NULL) {
 246                 so = inso;
 247                 slirp = so->slirp;
 248
 249                 /* Re-set a few variables */
 250                 tp = sototcpcb(so);
 251                 m = so->so_m;
 252                 so->so_m = NULL;
 253                 ti = so->so_ti;
 254                 tiwin = ti->ti_win;
 255                 tiflags = ti->ti_flags;
 256
 257                 goto cont_conn;
 258         }
 259         slirp = m->slirp;
 260
 261         ip = mtod(m, struct ip *);
 262         ip6 = mtod(m, struct ip6 *);
 263
 264         switch (af) {
 265         case AF_INET:
 266             if (iphlen > sizeof(struct ip)) {
 267                 ip_stripoptions(m, (struct mbuf *)0);
 268                 iphlen = sizeof(struct ip);
 269             }
 270             /* XXX Check if too short */
 271
 272
 273             /*
 274              * Save a copy of the IP header in case we want restore it
 275              * for sending an ICMP error message in response.
 276              */
 277             save_ip = *ip;
 278             save_ip.ip_len += iphlen;
 279
 280             /*
 281              * Get IP and TCP header together in first mbuf.
 282              * Note: IP leaves IP header in first mbuf.
 283              */
 284             m->m_data -= sizeof(struct tcpiphdr) - sizeof(struct ip)
 285                                                  - sizeof(struct tcphdr);
 286             m->m_len += sizeof(struct tcpiphdr) - sizeof(struct ip)
 287                                                 - sizeof(struct tcphdr);
 288             ti = mtod(m, struct tcpiphdr *);
 289
 290             /*
 291              * Checksum extended TCP header and data.
 292              */
 293             tlen = ip->ip_len;
 294             tcpiphdr2qlink(ti)->next = tcpiphdr2qlink(ti)->prev = NULL;
 295             memset(&ti->ih_mbuf, 0 , sizeof(struct mbuf_ptr));
 296             memset(&ti->ti, 0, sizeof(ti->ti));
 297             ti->ti_x0 = 0;
 298             ti->ti_src = save_ip.ip_src;
 299             ti->ti_dst = save_ip.ip_dst;
 300             ti->ti_pr = save_ip.ip_p;
 301             ti->ti_len = htons((uint16_t)tlen);
 302             break;
 303
 304         case AF_INET6:
 305             /*
 306              * Save a copy of the IP header in case we want restore it
 307              * for sending an ICMP error message in response.
 308              */
 309             save_ip6 = *ip6;
 310             /*
 311              * Get IP and TCP header together in first mbuf.
 312              * Note: IP leaves IP header in first mbuf.
 313              */
 314             m->m_data -= sizeof(struct tcpiphdr) - (sizeof(struct ip6)
 315                                                  + sizeof(struct tcphdr));
 316             m->m_len  += sizeof(struct tcpiphdr) - (sizeof(struct ip6)
 317                                                  + sizeof(struct tcphdr));
 318             ti = mtod(m, struct tcpiphdr *);
 319
 320             tlen = ip6->ip_pl;
 321             tcpiphdr2qlink(ti)->next = tcpiphdr2qlink(ti)->prev = NULL;
 322             memset(&ti->ih_mbuf, 0 , sizeof(struct mbuf_ptr));
 323             memset(&ti->ti, 0, sizeof(ti->ti));
 324             ti->ti_x0 = 0;
 325             ti->ti_src6 = save_ip6.ip_src;
 326             ti->ti_dst6 = save_ip6.ip_dst;
 327             ti->ti_nh6 = save_ip6.ip_nh;
 328             ti->ti_len = htons((uint16_t)tlen);
 329             break;
 330
 331         default:
 332             g_assert_not_reached();
 333         }
 334
 335         len = ((sizeof(struct tcpiphdr) - sizeof(struct tcphdr)) + tlen);
 336         if (cksum(m, len)) {
 337             goto drop;
 338         }
 339
 340         /*
 341          * Check that TCP offset makes sense,
 342          * pull out TCP options and adjust length.              XXX
 343          */
 344         off = ti->ti_off << 2;
 345         if (off < sizeof (struct tcphdr) || off > tlen) {
 346           goto drop;
 347         }
 348         tlen -= off;
 349         ti->ti_len = tlen;
 350         if (off > sizeof (struct tcphdr)) {
 351           optlen = off - sizeof (struct tcphdr);
 352           optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
 353         }
 354         tiflags = ti->ti_flags;
 355
 356         /*
 357          * Convert TCP protocol specific fields to host format.
 358          */
 359         NTOHL(ti->ti_seq);
 360         NTOHL(ti->ti_ack);
 361         NTOHS(ti->ti_win);
 362         NTOHS(ti->ti_urp);
 363
 364         /*
 365          * Drop TCP, IP headers and TCP options.
 366          */
 367         m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
 368         m->m_len  -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
 369
 370         /*
 371          * Locate pcb for segment.
 372          */
 373 findso:
 374         lhost.ss_family = af;
 375         fhost.ss_family = af;
 376         switch (af) {
 377         case AF_INET:
 378             lhost4 = (struct sockaddr_in *) &lhost;
 379             lhost4->sin_addr = ti->ti_src;
 380             lhost4->sin_port = ti->ti_sport;
 381             fhost4 = (struct sockaddr_in *) &fhost;
 382             fhost4->sin_addr = ti->ti_dst;
 383             fhost4->sin_port = ti->ti_dport;
 384             break;
 385         case AF_INET6:
 386             lhost6 = (struct sockaddr_in6 *) &lhost;
 387             lhost6->sin6_addr = ti->ti_src6;
 388             lhost6->sin6_port = ti->ti_sport;
 389             fhost6 = (struct sockaddr_in6 *) &fhost;
 390             fhost6->sin6_addr = ti->ti_dst6;
 391             fhost6->sin6_port = ti->ti_dport;
 392             break;
 393         default:
 394             g_assert_not_reached();
 395         }
 396
 397         so = solookup(&slirp->tcp_last_so, &slirp->tcb, &lhost, &fhost);
 398
 399         /*
 400          * If the state is CLOSED (i.e., TCB does not exist) then
 401          * all data in the incoming segment is discarded.
 402          * If the TCB exists but is in CLOSED state, it is embryonic,
 403          * but should either do a listen or a connect soon.
 404          *
 405          * state == CLOSED means we've done socreate() but haven't
 406          * attached it to a protocol yet...
 407          *
 408          * XXX If a TCB does not exist, and the TH_SYN flag is
 409          * the only flag set, then create a session, mark it
 410          * as if it was LISTENING, and continue...
 411          */
 412         if (so == NULL) {
 413           if (slirp->restricted) {
 414             /* Any hostfwds will have an existing socket, so we only get here
 415              * for non-hostfwd connections. These should be dropped, unless it
 416              * happens to be a guestfwd.
 417              */
 418             for (ex_ptr = slirp->exec_list; ex_ptr; ex_ptr = ex_ptr->ex_next) {
 419                 if (ex_ptr->ex_fport == ti->ti_dport &&
 420                     ti->ti_dst.s_addr == ex_ptr->ex_addr.s_addr) {
 421                     break;
 422                 }
 423             }
 424             if (!ex_ptr) {
 425                 goto dropwithreset;
 426             }
 427           }
 428
 429           if ((tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) != TH_SYN)
 430             goto dropwithreset;
 431
 432           if ((so = socreate(slirp)) == NULL)
 433             goto dropwithreset;
 434           if (tcp_attach(so) < 0) {
 435             free(so); /* Not sofree (if it failed, it's not insqued) */
 436             goto dropwithreset;
 437           }
 438
 439           sbreserve(&so->so_snd, TCP_SNDSPACE);
 440           sbreserve(&so->so_rcv, TCP_RCVSPACE);
 441
 442           so->lhost.ss = lhost;
 443           so->fhost.ss = fhost;
 444
 445           so->so_iptos = tcp_tos(so);
 446           if (so->so_iptos == 0) {
 447               switch (af) {
 448               case AF_INET:
 449                   so->so_iptos = ((struct ip *)ti)->ip_tos;
 450                   break;
 451               case AF_INET6:
 452                   break;
 453               default:
 454                   g_assert_not_reached();
 455               }
 456           }
 457
 458           tp = sototcpcb(so);
 459           tp->t_state = TCPS_LISTEN;
 460         }
 461
 462         /*
 463          * If this is a still-connecting socket, this probably
 464          * a retransmit of the SYN.  Whether it's a retransmit SYN
 465          * or something else, we nuke it.
 466          */
 467         if (so->so_state & SS_ISFCONNECTING)
 468                 goto drop;
 469
 470         tp = sototcpcb(so);
 471
 472         /* XXX Should never fail */
 473         if (tp == NULL)
 474                 goto dropwithreset;
 475         if (tp->t_state == TCPS_CLOSED)
 476                 goto drop;
 477
 478         tiwin = ti->ti_win;
 479
 480         /*
 481          * Segment received on connection.
 482          * Reset idle time and keep-alive timer.
 483          */
 484         tp->t_idle = 0;
 485         if (SO_OPTIONS)
 486            tp->t_timer[TCPT_KEEP] = TCPTV_KEEPINTVL;
 487         else
 488            tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_IDLE;
 489
 490         /*
 491          * Process options if not in LISTEN state,
 492          * else do it below (after getting remote address).
 493          */
 494         if (optp && tp->t_state != TCPS_LISTEN)
 495                 tcp_dooptions(tp, (u_char *)optp, optlen, ti);
 496
 497         /*
 498          * Header prediction: check for the two common cases
 499          * of a uni-directional data xfer.  If the packet has
 500          * no control flags, is in-sequence, the window didn't
 501          * change and we're not retransmitting, it's a
 502          * candidate.  If the length is zero and the ack moved
 503          * forward, we're the sender side of the xfer.  Just
 504          * free the data acked & wake any higher level process
 505          * that was blocked waiting for space.  If the length
 506          * is non-zero and the ack didn't move, we're the
 507          * receiver side.  If we're getting packets in-order
 508          * (the reassembly queue is empty), add the data to
 509          * the socket buffer and note that we need a delayed ack.
 510          *
 511          * XXX Some of these tests are not needed
 512          * eg: the tiwin == tp->snd_wnd prevents many more
 513          * predictions.. with no *real* advantage..
 514          */
 515         if (tp->t_state == TCPS_ESTABLISHED &&
 516             (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
 517             ti->ti_seq == tp->rcv_nxt &&
 518             tiwin && tiwin == tp->snd_wnd &&
 519             tp->snd_nxt == tp->snd_max) {
 520                 if (ti->ti_len == 0) {
 521                         if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
 522                             SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
 523                             tp->snd_cwnd >= tp->snd_wnd) {
 524                                 /*
 525                                  * this is a pure ack for outstanding data.
 526                                  */
 527                                 if (tp->t_rtt &&
 528                                     SEQ_GT(ti->ti_ack, tp->t_rtseq))
 529                                         tcp_xmit_timer(tp, tp->t_rtt);
 530                                 acked = ti->ti_ack - tp->snd_una;
 531                                 sbdrop(&so->so_snd, acked);
 532                                 tp->snd_una = ti->ti_ack;
 533                                 m_free(m);
 534
 535                                 /*
 536                                  * If all outstanding data are acked, stop
 537                                  * retransmit timer, otherwise restart timer
 538                                  * using current (possibly backed-off) value.
 539                                  * If process is waiting for space,
 540                                  * wakeup/selwakeup/signal.  If data
 541                                  * are ready to send, let tcp_output
 542                                  * decide between more output or persist.
 543                                  */
 544                                 if (tp->snd_una == tp->snd_max)
 545                                         tp->t_timer[TCPT_REXMT] = 0;
 546                                 else if (tp->t_timer[TCPT_PERSIST] == 0)
 547                                         tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
 548
 549                                 /*
 550                                  * This is called because sowwakeup might have
 551                                  * put data into so_snd.  Since we don't so sowwakeup,
 552                                  * we don't need this.. XXX???
 553                                  */
 554                                 if (so->so_snd.sb_cc)
 555                                         (void) tcp_output(tp);
 556
 557                                 return;
 558                         }
 559                 } else if (ti->ti_ack == tp->snd_una &&
 560                     tcpfrag_list_empty(tp) &&
 561                     ti->ti_len <= sbspace(&so->so_rcv)) {
 562                         /*
 563                          * this is a pure, in-sequence data packet
 564                          * with nothing on the reassembly queue and
 565                          * we have enough buffer space to take it.
 566                          */
 567                         tp->rcv_nxt += ti->ti_len;
 568                         /*
 569                          * Add data to socket buffer.
 570                          */
 571                         if (so->so_emu) {
 572                                 if (tcp_emu(so,m)) sbappend(so, m);
 573                         } else
 574                                 sbappend(so, m);
 575
 576                         /*
 577                          * If this is a short packet, then ACK now - with Nagel
 578                          *      congestion avoidance sender won't send more until
 579                          *      he gets an ACK.
 580                          *
 581                          * It is better to not delay acks at all to maximize
 582                          * TCP throughput.  See RFC 2581.
 583                          */
 584                         tp->t_flags |= TF_ACKNOW;
 585                         tcp_output(tp);
 586                         return;
 587                 }
 588         } /* header prediction */
 589         /*
 590          * Calculate amount of space in receive window,
 591          * and then do TCP input processing.
 592          * Receive window is amount of space in rcv queue,
 593          * but not less than advertised window.
 594          */
 595         { int win;
 596           win = sbspace(&so->so_rcv);
 597           if (win < 0)
 598             win = 0;
 599           tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
 600         }
 601
 602         switch (tp->t_state) {
 603
 604         /*
 605          * If the state is LISTEN then ignore segment if it contains an RST.
 606          * If the segment contains an ACK then it is bad and send a RST.
 607          * If it does not contain a SYN then it is not interesting; drop it.
 608          * Don't bother responding if the destination was a broadcast.
 609          * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
 610          * tp->iss, and send a segment:
 611          *     <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
 612          * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
 613          * Fill in remote peer address fields if not previously specified.
 614          * Enter SYN_RECEIVED state, and process any other fields of this
 615          * segment in this state.
 616          */
 617         case TCPS_LISTEN: {
 618
 619           if (tiflags & TH_RST)
 620             goto drop;
 621           if (tiflags & TH_ACK)
 622             goto dropwithreset;
 623           if ((tiflags & TH_SYN) == 0)
 624             goto drop;
 625
 626           /*
 627            * This has way too many gotos...
 628            * But a bit of spaghetti code never hurt anybody :)
 629            */
 630
 631           /*
 632            * If this is destined for the control address, then flag to
 633            * tcp_ctl once connected, otherwise connect
 634            */
 635           if (af == AF_INET &&
 636                  (so->so_faddr.s_addr & slirp->vnetwork_mask.s_addr) ==
 637                  slirp->vnetwork_addr.s_addr) {
 638             if (so->so_faddr.s_addr != slirp->vhost_addr.s_addr &&
 639                 so->so_faddr.s_addr != slirp->vnameserver_addr.s_addr) {
 640                 /* May be an add exec */
 641                 for (ex_ptr = slirp->exec_list; ex_ptr;
 642                      ex_ptr = ex_ptr->ex_next) {
 643                   if(ex_ptr->ex_fport == so->so_fport &&
 644                      so->so_faddr.s_addr == ex_ptr->ex_addr.s_addr) {
 645                     so->so_state |= SS_CTL;
 646                     break;
 647                   }
 648                 }
 649                 if (so->so_state & SS_CTL) {
 650                     goto cont_input;
 651                 }
 652             }
 653             /* CTL_ALIAS: Do nothing, tcp_fconnect will be called on it */
 654           }
 655
 656           if (so->so_emu & EMU_NOCONNECT) {
 657             so->so_emu &= ~EMU_NOCONNECT;
 658             goto cont_input;
 659           }
 660
 661           if ((tcp_fconnect(so, so->so_ffamily) == -1) &&
 662               (errno != EAGAIN) &&
 663               (errno != EINPROGRESS) && (errno != EWOULDBLOCK)
 664           ) {
 665             uint8_t code;
 666             DEBUG_MISC((dfd, " tcp fconnect errno = %d-%s\n",
 667                         errno,strerror(errno)));
 668             if(errno == ECONNREFUSED) {
 669               /* ACK the SYN, send RST to refuse the connection */
 670               tcp_respond(tp, ti, m, ti->ti_seq + 1, (tcp_seq) 0,
 671                           TH_RST | TH_ACK, af);
 672             } else {
 673               switch (af) {
 674               case AF_INET:
 675                 code = ICMP_UNREACH_NET;
 676                 if (errno == EHOSTUNREACH) {
 677                   code = ICMP_UNREACH_HOST;
 678                 }
 679                 break;
 680               case AF_INET6:
 681                 code = ICMP6_UNREACH_NO_ROUTE;
 682                 if (errno == EHOSTUNREACH) {
 683                   code = ICMP6_UNREACH_ADDRESS;
 684                 }
 685                 break;
 686               default:
 687                 g_assert_not_reached();
 688               }
 689               HTONL(ti->ti_seq);             /* restore tcp header */
 690               HTONL(ti->ti_ack);
 691               HTONS(ti->ti_win);
 692               HTONS(ti->ti_urp);
 693               m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
 694               m->m_len  += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
 695               switch (af) {
 696               case AF_INET:
 697                 m->m_data += sizeof(struct tcpiphdr) - sizeof(struct ip)
 698                                                      - sizeof(struct tcphdr);
 699                 m->m_len  -= sizeof(struct tcpiphdr) - sizeof(struct ip)
 700                                                      - sizeof(struct tcphdr);
 701                 *ip = save_ip;
 702                 icmp_send_error(m, ICMP_UNREACH, code, 0, strerror(errno));
 703                 break;
 704               case AF_INET6:
 705                 m->m_data += sizeof(struct tcpiphdr) - (sizeof(struct ip6)
 706                                                      + sizeof(struct tcphdr));
 707                 m->m_len  -= sizeof(struct tcpiphdr) - (sizeof(struct ip6)
 708                                                      + sizeof(struct tcphdr));
 709                 *ip6 = save_ip6;
 710                 icmp6_send_error(m, ICMP6_UNREACH, code);
 711                 break;
 712               default:
 713                 g_assert_not_reached();
 714               }
 715             }
 716             tcp_close(tp);
 717             m_free(m);
 718           } else {
 719             /*
 720              * Haven't connected yet, save the current mbuf
 721              * and ti, and return
 722              * XXX Some OS's don't tell us whether the connect()
 723              * succeeded or not.  So we must time it out.
 724              */
 725             so->so_m = m;
 726             so->so_ti = ti;
 727             tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
 728             tp->t_state = TCPS_SYN_RECEIVED;
 729             /*
 730              * Initialize receive sequence numbers now so that we can send a
 731              * valid RST if the remote end rejects our connection.
 732              */
 733             tp->irs = ti->ti_seq;
 734             tcp_rcvseqinit(tp);
 735             tcp_template(tp);
 736           }
 737           return;
 738
 739         cont_conn:
 740           /* m==NULL
 741            * Check if the connect succeeded
 742            */
 743           if (so->so_state & SS_NOFDREF) {
 744             tp = tcp_close(tp);
 745             goto dropwithreset;
 746           }
 747         cont_input:
 748           tcp_template(tp);
 749
 750           if (optp)
 751             tcp_dooptions(tp, (u_char *)optp, optlen, ti);
 752
 753           if (iss)
 754             tp->iss = iss;
 755           else
 756             tp->iss = slirp->tcp_iss;
 757           slirp->tcp_iss += TCP_ISSINCR/2;
 758           tp->irs = ti->ti_seq;
 759           tcp_sendseqinit(tp);
 760           tcp_rcvseqinit(tp);
 761           tp->t_flags |= TF_ACKNOW;
 762           tp->t_state = TCPS_SYN_RECEIVED;
 763           tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
 764           goto trimthenstep6;
 765         } /* case TCPS_LISTEN */
 766
 767         /*
 768          * If the state is SYN_SENT:
 769          *      if seg contains an ACK, but not for our SYN, drop the input.
 770          *      if seg contains a RST, then drop the connection.
 771          *      if seg does not contain SYN, then drop it.
 772          * Otherwise this is an acceptable SYN segment
 773          *      initialize tp->rcv_nxt and tp->irs
 774          *      if seg contains ack then advance tp->snd_una
 775          *      if SYN has been acked change to ESTABLISHED else SYN_RCVD state
 776          *      arrange for segment to be acked (eventually)
 777          *      continue processing rest of data/controls, beginning with URG
 778          */
 779         case TCPS_SYN_SENT:
 780                 if ((tiflags & TH_ACK) &&
 781                     (SEQ_LEQ(ti->ti_ack, tp->iss) ||
 782                      SEQ_GT(ti->ti_ack, tp->snd_max)))
 783                         goto dropwithreset;
 784
 785                 if (tiflags & TH_RST) {
 786                         if (tiflags & TH_ACK) {
 787                                 tcp_drop(tp, 0); /* XXX Check t_softerror! */
 788                         }
 789                         goto drop;
 790                 }
 791
 792                 if ((tiflags & TH_SYN) == 0)
 793                         goto drop;
 794                 if (tiflags & TH_ACK) {
 795                         tp->snd_una = ti->ti_ack;
 796                         if (SEQ_LT(tp->snd_nxt, tp->snd_una))
 797                                 tp->snd_nxt = tp->snd_una;
 798                 }
 799
 800                 tp->t_timer[TCPT_REXMT] = 0;
 801                 tp->irs = ti->ti_seq;
 802                 tcp_rcvseqinit(tp);
 803                 tp->t_flags |= TF_ACKNOW;
 804                 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
 805                         soisfconnected(so);
 806                         tp->t_state = TCPS_ESTABLISHED;
 807
 808                         (void) tcp_reass(tp, (struct tcpiphdr *)0,
 809                                 (struct mbuf *)0);
 810                         /*
 811                          * if we didn't have to retransmit the SYN,
 812                          * use its rtt as our initial srtt & rtt var.
 813                          */
 814                         if (tp->t_rtt)
 815                                 tcp_xmit_timer(tp, tp->t_rtt);
 816                 } else
 817                         tp->t_state = TCPS_SYN_RECEIVED;
 818
 819 trimthenstep6:
 820                 /*
 821                  * Advance ti->ti_seq to correspond to first data byte.
 822                  * If data, trim to stay within window,
 823                  * dropping FIN if necessary.
 824                  */
 825                 ti->ti_seq++;
 826                 if (ti->ti_len > tp->rcv_wnd) {
 827                         todrop = ti->ti_len - tp->rcv_wnd;
 828                         m_adj(m, -todrop);
 829                         ti->ti_len = tp->rcv_wnd;
 830                         tiflags &= ~TH_FIN;
 831                 }
 832                 tp->snd_wl1 = ti->ti_seq - 1;
 833                 tp->rcv_up = ti->ti_seq;
 834                 goto step6;
 835         } /* switch tp->t_state */
 836         /*
 837          * States other than LISTEN or SYN_SENT.
 838          * Check that at least some bytes of segment are within
 839          * receive window.  If segment begins before rcv_nxt,
 840          * drop leading data (and SYN); if nothing left, just ack.
 841          */
 842         todrop = tp->rcv_nxt - ti->ti_seq;
 843         if (todrop > 0) {
 844                 if (tiflags & TH_SYN) {
 845                         tiflags &= ~TH_SYN;
 846                         ti->ti_seq++;
 847                         if (ti->ti_urp > 1)
 848                                 ti->ti_urp--;
 849                         else
 850                                 tiflags &= ~TH_URG;
 851                         todrop--;
 852                 }
 853                 /*
 854                  * Following if statement from Stevens, vol. 2, p. 960.
 855                  */
 856                 if (todrop > ti->ti_len
 857                     || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) {
 858                         /*
 859                          * Any valid FIN must be to the left of the window.
 860                          * At this point the FIN must be a duplicate or out
 861                          * of sequence; drop it.
 862                          */
 863                         tiflags &= ~TH_FIN;
 864
 865                         /*
 866                          * Send an ACK to resynchronize and drop any data.
 867                          * But keep on processing for RST or ACK.
 868                          */
 869                         tp->t_flags |= TF_ACKNOW;
 870                         todrop = ti->ti_len;
 871                 }
 872                 m_adj(m, todrop);
 873                 ti->ti_seq += todrop;
 874                 ti->ti_len -= todrop;
 875                 if (ti->ti_urp > todrop)
 876                         ti->ti_urp -= todrop;
 877                 else {
 878                         tiflags &= ~TH_URG;
 879                         ti->ti_urp = 0;
 880                 }
 881         }
 882         /*
 883          * If new data are received on a connection after the
 884          * user processes are gone, then RST the other end.
 885          */
 886         if ((so->so_state & SS_NOFDREF) &&
 887             tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
 888                 tp = tcp_close(tp);
 889                 goto dropwithreset;
 890         }
 891
 892         /*
 893          * If segment ends after window, drop trailing data
 894          * (and PUSH and FIN); if nothing left, just ACK.
 895          */
 896         todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
 897         if (todrop > 0) {
 898                 if (todrop >= ti->ti_len) {
 899                         /*
 900                          * If a new connection request is received
 901                          * while in TIME_WAIT, drop the old connection
 902                          * and start over if the sequence numbers
 903                          * are above the previous ones.
 904                          */
 905                         if (tiflags & TH_SYN &&
 906                             tp->t_state == TCPS_TIME_WAIT &&
 907                             SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
 908                                 iss = tp->rcv_nxt + TCP_ISSINCR;
 909                                 tp = tcp_close(tp);
 910                                 goto findso;
 911                         }
 912                         /*
 913                          * If window is closed can only take segments at
 914                          * window edge, and have to drop data and PUSH from
 915                          * incoming segments.  Continue processing, but
 916                          * remember to ack.  Otherwise, drop segment
 917                          * and ack.
 918                          */
 919                         if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
 920                                 tp->t_flags |= TF_ACKNOW;
 921                         } else {
 922                                 goto dropafterack;
 923                         }
 924                 }
 925                 m_adj(m, -todrop);
 926                 ti->ti_len -= todrop;
 927                 tiflags &= ~(TH_PUSH|TH_FIN);
 928         }
 929
 930         /*
 931          * If the RST bit is set examine the state:
 932          *    SYN_RECEIVED STATE:
 933          *      If passive open, return to LISTEN state.
 934          *      If active open, inform user that connection was refused.
 935          *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
 936          *      Inform user that connection was reset, and close tcb.
 937          *    CLOSING, LAST_ACK, TIME_WAIT STATES
 938          *      Close the tcb.
 939          */
 940         if (tiflags&TH_RST) switch (tp->t_state) {
 941
 942         case TCPS_SYN_RECEIVED:
 943         case TCPS_ESTABLISHED:
 944         case TCPS_FIN_WAIT_1:
 945         case TCPS_FIN_WAIT_2:
 946         case TCPS_CLOSE_WAIT:
 947                 tp->t_state = TCPS_CLOSED;
 948                 tcp_close(tp);
 949                 goto drop;
 950
 951         case TCPS_CLOSING:
 952         case TCPS_LAST_ACK:
 953         case TCPS_TIME_WAIT:
 954                 tcp_close(tp);
 955                 goto drop;
 956         }
 957
 958         /*
 959          * If a SYN is in the window, then this is an
 960          * error and we send an RST and drop the connection.
 961          */
 962         if (tiflags & TH_SYN) {
 963                 tp = tcp_drop(tp,0);
 964                 goto dropwithreset;
 965         }
 966
 967         /*
 968          * If the ACK bit is off we drop the segment and return.
 969          */
 970         if ((tiflags & TH_ACK) == 0) goto drop;
 971
 972         /*
 973          * Ack processing.
 974          */
 975         switch (tp->t_state) {
 976         /*
 977          * In SYN_RECEIVED state if the ack ACKs our SYN then enter
 978          * ESTABLISHED state and continue processing, otherwise
 979          * send an RST.  una<=ack<=max
 980          */
 981         case TCPS_SYN_RECEIVED:
 982
 983                 if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
 984                     SEQ_GT(ti->ti_ack, tp->snd_max))
 985                         goto dropwithreset;
 986                 tp->t_state = TCPS_ESTABLISHED;
 987                 /*
 988                  * The sent SYN is ack'ed with our sequence number +1
 989                  * The first data byte already in the buffer will get
 990                  * lost if no correction is made.  This is only needed for
 991                  * SS_CTL since the buffer is empty otherwise.
 992                  * tp->snd_una++; or:
 993                  */
 994                 tp->snd_una=ti->ti_ack;
 995                 if (so->so_state & SS_CTL) {
 996                   /* So tcp_ctl reports the right state */
 997                   ret = tcp_ctl(so);
 998                   if (ret == 1) {
 999                     soisfconnected(so);
1000                     so->so_state &= ~SS_CTL;   /* success XXX */
1001                   } else if (ret == 2) {
1002                     so->so_state &= SS_PERSISTENT_MASK;
1003                     so->so_state |= SS_NOFDREF; /* CTL_CMD */
1004                   } else {
1005                     needoutput = 1;
1006                     tp->t_state = TCPS_FIN_WAIT_1;
1007                   }
1008                 } else {
1009                   soisfconnected(so);
1010                 }
1011
1012                 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0);
1013                 tp->snd_wl1 = ti->ti_seq - 1;
1014                 /* Avoid ack processing; snd_una==ti_ack  =>  dup ack */
1015                 goto synrx_to_est;
1016                 /* fall into ... */
1017
1018         /*
1019          * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1020          * ACKs.  If the ack is in the range
1021          *      tp->snd_una < ti->ti_ack <= tp->snd_max
1022          * then advance tp->snd_una to ti->ti_ack and drop
1023          * data from the retransmission queue.  If this ACK reflects
1024          * more up to date window information we update our window information.
1025          */
1026         case TCPS_ESTABLISHED:
1027         case TCPS_FIN_WAIT_1:
1028         case TCPS_FIN_WAIT_2:
1029         case TCPS_CLOSE_WAIT:
1030         case TCPS_CLOSING:
1031         case TCPS_LAST_ACK:
1032         case TCPS_TIME_WAIT:
1033
1034                 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
1035                         if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
1036                           DEBUG_MISC((dfd, " dup ack  m = %p  so = %p\n",
1037                                       m, so));
1038                                 /*
1039                                  * If we have outstanding data (other than
1040                                  * a window probe), this is a completely
1041                                  * duplicate ack (ie, window info didn't
1042                                  * change), the ack is the biggest we've
1043                                  * seen and we've seen exactly our rexmt
1044                                  * threshold of them, assume a packet
1045                                  * has been dropped and retransmit it.
1046                                  * Kludge snd_nxt & the congestion
1047                                  * window so we send only this one
1048                                  * packet.
1049                                  *
1050                                  * We know we're losing at the current
1051                                  * window size so do congestion avoidance
1052                                  * (set ssthresh to half the current window
1053                                  * and pull our congestion window back to
1054                                  * the new ssthresh).
1055                                  *
1056                                  * Dup acks mean that packets have left the
1057                                  * network (they're now cached at the receiver)
1058                                  * so bump cwnd by the amount in the receiver
1059                                  * to keep a constant cwnd packets in the
1060                                  * network.
1061                                  */
1062                                 if (tp->t_timer[TCPT_REXMT] == 0 ||
1063                                     ti->ti_ack != tp->snd_una)
1064                                         tp->t_dupacks = 0;
1065                                 else if (++tp->t_dupacks == TCPREXMTTHRESH) {
1066                                         tcp_seq onxt = tp->snd_nxt;
1067                                         u_int win =
1068                                             min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1069                                                 tp->t_maxseg;
1070
1071                                         if (win < 2)
1072                                                 win = 2;
1073                                         tp->snd_ssthresh = win * tp->t_maxseg;
1074                                         tp->t_timer[TCPT_REXMT] = 0;
1075                                         tp->t_rtt = 0;
1076                                         tp->snd_nxt = ti->ti_ack;
1077                                         tp->snd_cwnd = tp->t_maxseg;
1078                                         (void) tcp_output(tp);
1079                                         tp->snd_cwnd = tp->snd_ssthresh +
1080                                                tp->t_maxseg * tp->t_dupacks;
1081                                         if (SEQ_GT(onxt, tp->snd_nxt))
1082                                                 tp->snd_nxt = onxt;
1083                                         goto drop;
1084                                 } else if (tp->t_dupacks > TCPREXMTTHRESH) {
1085                                         tp->snd_cwnd += tp->t_maxseg;
1086                                         (void) tcp_output(tp);
1087                                         goto drop;
1088                                 }
1089                         } else
1090                                 tp->t_dupacks = 0;
1091                         break;
1092                 }
1093         synrx_to_est:
1094                 /*
1095                  * If the congestion window was inflated to account
1096                  * for the other side's cached packets, retract it.
1097                  */
1098                 if (tp->t_dupacks > TCPREXMTTHRESH &&
1099                     tp->snd_cwnd > tp->snd_ssthresh)
1100                         tp->snd_cwnd = tp->snd_ssthresh;
1101                 tp->t_dupacks = 0;
1102                 if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
1103                         goto dropafterack;
1104                 }
1105                 acked = ti->ti_ack - tp->snd_una;
1106
1107                 /*
1108                  * If transmit timer is running and timed sequence
1109                  * number was acked, update smoothed round trip time.
1110                  * Since we now have an rtt measurement, cancel the
1111                  * timer backoff (cf., Phil Karn's retransmit alg.).
1112                  * Recompute the initial retransmit timer.
1113                  */
1114                 if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1115                         tcp_xmit_timer(tp,tp->t_rtt);
1116
1117                 /*
1118                  * If all outstanding data is acked, stop retransmit
1119                  * timer and remember to restart (more output or persist).
1120                  * If there is more data to be acked, restart retransmit
1121                  * timer, using current (possibly backed-off) value.
1122                  */
1123                 if (ti->ti_ack == tp->snd_max) {
1124                         tp->t_timer[TCPT_REXMT] = 0;
1125                         needoutput = 1;
1126                 } else if (tp->t_timer[TCPT_PERSIST] == 0)
1127                         tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1128                 /*
1129                  * When new data is acked, open the congestion window.
1130                  * If the window gives us less than ssthresh packets
1131                  * in flight, open exponentially (maxseg per packet).
1132                  * Otherwise open linearly: maxseg per window
1133                  * (maxseg^2 / cwnd per packet).
1134                  */
1135                 {
1136                   register u_int cw = tp->snd_cwnd;
1137                   register u_int incr = tp->t_maxseg;
1138
1139                   if (cw > tp->snd_ssthresh)
1140                     incr = incr * incr / cw;
1141                   tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1142                 }
1143                 if (acked > so->so_snd.sb_cc) {
1144                         tp->snd_wnd -= so->so_snd.sb_cc;
1145                         sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1146                         ourfinisacked = 1;
1147                 } else {
1148                         sbdrop(&so->so_snd, acked);
1149                         tp->snd_wnd -= acked;
1150                         ourfinisacked = 0;
1151                 }
1152                 tp->snd_una = ti->ti_ack;
1153                 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1154                         tp->snd_nxt = tp->snd_una;
1155
1156                 switch (tp->t_state) {
1157
1158                 /*
1159                  * In FIN_WAIT_1 STATE in addition to the processing
1160                  * for the ESTABLISHED state if our FIN is now acknowledged
1161                  * then enter FIN_WAIT_2.
1162                  */
1163                 case TCPS_FIN_WAIT_1:
1164                         if (ourfinisacked) {
1165                                 /*
1166                                  * If we can't receive any more
1167                                  * data, then closing user can proceed.
1168                                  * Starting the timer is contrary to the
1169                                  * specification, but if we don't get a FIN
1170                                  * we'll hang forever.
1171                                  */
1172                                 if (so->so_state & SS_FCANTRCVMORE) {
1173                                         tp->t_timer[TCPT_2MSL] = TCP_MAXIDLE;
1174                                 }
1175                                 tp->t_state = TCPS_FIN_WAIT_2;
1176                         }
1177                         break;
1178
1179                 /*
1180                  * In CLOSING STATE in addition to the processing for
1181                  * the ESTABLISHED state if the ACK acknowledges our FIN
1182                  * then enter the TIME-WAIT state, otherwise ignore
1183                  * the segment.
1184                  */
1185                 case TCPS_CLOSING:
1186                         if (ourfinisacked) {
1187                                 tp->t_state = TCPS_TIME_WAIT;
1188                                 tcp_canceltimers(tp);
1189                                 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1190                         }
1191                         break;
1192
1193                 /*
1194                  * In LAST_ACK, we may still be waiting for data to drain
1195                  * and/or to be acked, as well as for the ack of our FIN.
1196                  * If our FIN is now acknowledged, delete the TCB,
1197                  * enter the closed state and return.
1198                  */
1199                 case TCPS_LAST_ACK:
1200                         if (ourfinisacked) {
1201                                 tcp_close(tp);
1202                                 goto drop;
1203                         }
1204                         break;
1205
1206                 /*
1207                  * In TIME_WAIT state the only thing that should arrive
1208                  * is a retransmission of the remote FIN.  Acknowledge
1209                  * it and restart the finack timer.
1210                  */
1211                 case TCPS_TIME_WAIT:
1212                         tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1213                         goto dropafterack;
1214                 }
1215         } /* switch(tp->t_state) */
1216
1217 step6:
1218         /*
1219          * Update window information.
1220          * Don't look at window if no ACK: TAC's send garbage on first SYN.
1221          */
1222         if ((tiflags & TH_ACK) &&
1223             (SEQ_LT(tp->snd_wl1, ti->ti_seq) ||
1224             (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
1225             (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) {
1226                 tp->snd_wnd = tiwin;
1227                 tp->snd_wl1 = ti->ti_seq;
1228                 tp->snd_wl2 = ti->ti_ack;
1229                 if (tp->snd_wnd > tp->max_sndwnd)
1230                         tp->max_sndwnd = tp->snd_wnd;
1231                 needoutput = 1;
1232         }
1233
1234         /*
1235          * Process segments with URG.
1236          */
1237         if ((tiflags & TH_URG) && ti->ti_urp &&
1238             TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1239                 /*
1240                  * This is a kludge, but if we receive and accept
1241                  * random urgent pointers, we'll crash in
1242                  * soreceive.  It's hard to imagine someone
1243                  * actually wanting to send this much urgent data.
1244                  */
1245                 if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen) {
1246                         ti->ti_urp = 0;
1247                         tiflags &= ~TH_URG;
1248                         goto dodata;
1249                 }
1250                 /*
1251                  * If this segment advances the known urgent pointer,
1252                  * then mark the data stream.  This should not happen
1253                  * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1254                  * a FIN has been received from the remote side.
1255                  * In these states we ignore the URG.
1256                  *
1257                  * According to RFC961 (Assigned Protocols),
1258                  * the urgent pointer points to the last octet
1259                  * of urgent data.  We continue, however,
1260                  * to consider it to indicate the first octet
1261                  * of data past the urgent section as the original
1262                  * spec states (in one of two places).
1263                  */
1264                 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1265                         tp->rcv_up = ti->ti_seq + ti->ti_urp;
1266                         so->so_urgc =  so->so_rcv.sb_cc +
1267                                 (tp->rcv_up - tp->rcv_nxt); /* -1; */
1268                         tp->rcv_up = ti->ti_seq + ti->ti_urp;
1269
1270                 }
1271         } else
1272                 /*
1273                  * If no out of band data is expected,
1274                  * pull receive urgent pointer along
1275                  * with the receive window.
1276                  */
1277                 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1278                         tp->rcv_up = tp->rcv_nxt;
1279 dodata:
1280
1281         /*
1282          * If this is a small packet, then ACK now - with Nagel
1283          *      congestion avoidance sender won't send more until
1284          *      he gets an ACK.
1285          */
1286         if (ti->ti_len && (unsigned)ti->ti_len <= 5 &&
1287             ((struct tcpiphdr_2 *)ti)->first_char == (char)27) {
1288                 tp->t_flags |= TF_ACKNOW;
1289         }
1290
1291         /*
1292          * Process the segment text, merging it into the TCP sequencing queue,
1293          * and arranging for acknowledgment of receipt if necessary.
1294          * This process logically involves adjusting tp->rcv_wnd as data
1295          * is presented to the user (this happens in tcp_usrreq.c,
1296          * case PRU_RCVD).  If a FIN has already been received on this
1297          * connection then we just ignore the text.
1298          */
1299         if ((ti->ti_len || (tiflags&TH_FIN)) &&
1300             TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1301                 TCP_REASS(tp, ti, m, so, tiflags);
1302         } else {
1303                 m_free(m);
1304                 tiflags &= ~TH_FIN;
1305         }
1306
1307         /*
1308          * If FIN is received ACK the FIN and let the user know
1309          * that the connection is closing.
1310          */
1311         if (tiflags & TH_FIN) {
1312                 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1313                         /*
1314                          * If we receive a FIN we can't send more data,
1315                          * set it SS_FDRAIN
1316                          * Shutdown the socket if there is no rx data in the
1317                          * buffer.
1318                          * soread() is called on completion of shutdown() and
1319                          * will got to TCPS_LAST_ACK, and use tcp_output()
1320                          * to send the FIN.
1321                          */
1322                         sofwdrain(so);
1323
1324                         tp->t_flags |= TF_ACKNOW;
1325                         tp->rcv_nxt++;
1326                 }
1327                 switch (tp->t_state) {
1328
1329                 /*
1330                  * In SYN_RECEIVED and ESTABLISHED STATES
1331                  * enter the CLOSE_WAIT state.
1332                  */
1333                 case TCPS_SYN_RECEIVED:
1334                 case TCPS_ESTABLISHED:
1335                   if(so->so_emu == EMU_CTL)        /* no shutdown on socket */
1336                     tp->t_state = TCPS_LAST_ACK;
1337                   else
1338                     tp->t_state = TCPS_CLOSE_WAIT;
1339                   break;
1340
1341                 /*
1342                  * If still in FIN_WAIT_1 STATE FIN has not been acked so
1343                  * enter the CLOSING state.
1344                  */
1345                 case TCPS_FIN_WAIT_1:
1346                         tp->t_state = TCPS_CLOSING;
1347                         break;
1348
1349                 /*
1350                  * In FIN_WAIT_2 state enter the TIME_WAIT state,
1351                  * starting the time-wait timer, turning off the other
1352                  * standard timers.
1353                  */
1354                 case TCPS_FIN_WAIT_2:
1355                         tp->t_state = TCPS_TIME_WAIT;
1356                         tcp_canceltimers(tp);
1357                         tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1358                         break;
1359
1360                 /*
1361                  * In TIME_WAIT state restart the 2 MSL time_wait timer.
1362                  */
1363                 case TCPS_TIME_WAIT:
1364                         tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1365                         break;
1366                 }
1367         }
1368
1369         /*
1370          * Return any desired output.
1371          */
1372         if (needoutput || (tp->t_flags & TF_ACKNOW)) {
1373                 (void) tcp_output(tp);
1374         }
1375         return;
1376
1377 dropafterack:
1378         /*
1379          * Generate an ACK dropping incoming segment if it occupies
1380          * sequence space, where the ACK reflects our state.
1381          */
1382         if (tiflags & TH_RST)
1383                 goto drop;
1384         m_free(m);
1385         tp->t_flags |= TF_ACKNOW;
1386         (void) tcp_output(tp);
1387         return;
1388
1389 dropwithreset:
1390         /* reuses m if m!=NULL, m_free() unnecessary */
1391         if (tiflags & TH_ACK)
1392                 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST, af);
1393         else {
1394                 if (tiflags & TH_SYN) ti->ti_len++;
1395                 tcp_respond(tp, ti, m, ti->ti_seq + ti->ti_len, (tcp_seq) 0,
1396                     TH_RST | TH_ACK, af);
1397         }
1398
1399         return;
1400
1401 drop:
1402         /*
1403          * Drop space held by incoming segment and return.
1404          */
1405         m_free(m);
1406 }
1407
1408 static void
1409 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcpiphdr *ti)
1410 {
1411         uint16_t mss;
1412         int opt, optlen;
1413
1414         DEBUG_CALL("tcp_dooptions");
1415         DEBUG_ARGS((dfd, " tp = %p  cnt=%i\n", tp, cnt));
1416
1417         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1418                 opt = cp[0];
1419                 if (opt == TCPOPT_EOL)
1420                         break;
1421                 if (opt == TCPOPT_NOP)
1422                         optlen = 1;
1423                 else {
1424                         optlen = cp[1];
1425                         if (optlen <= 0)
1426                                 break;
1427                 }
1428                 switch (opt) {
1429
1430                 default:
1431                         continue;
1432
1433                 case TCPOPT_MAXSEG:
1434                         if (optlen != TCPOLEN_MAXSEG)
1435                                 continue;
1436                         if (!(ti->ti_flags & TH_SYN))
1437                                 continue;
1438                         memcpy((char *) &mss, (char *) cp + 2, sizeof(mss));
1439                         NTOHS(mss);
1440                         (void) tcp_mss(tp, mss);        /* sets t_maxseg */
1441                         break;
1442                 }
1443         }
1444 }
1445
1446
1447 /*
1448  * Pull out of band byte out of a segment so
1449  * it doesn't appear in the user's data queue.
1450  * It is still reflected in the segment length for
1451  * sequencing purposes.
1452  */
1453
1454 #ifdef notdef
1455
1456 void
1457 tcp_pulloutofband(so, ti, m)
1458         struct socket *so;
1459         struct tcpiphdr *ti;
1460         register struct mbuf *m;
1461 {
1462         int cnt = ti->ti_urp - 1;
1463
1464         while (cnt >= 0) {
1465                 if (m->m_len > cnt) {
1466                         char *cp = mtod(m, caddr_t) + cnt;
1467                         struct tcpcb *tp = sototcpcb(so);
1468
1469                         tp->t_iobc = *cp;
1470                         tp->t_oobflags |= TCPOOB_HAVEDATA;
1471                         memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1472                         m->m_len--;
1473                         return;
1474                 }
1475                 cnt -= m->m_len;
1476                 m = m->m_next; /* XXX WRONG! Fix it! */
1477                 if (m == 0)
1478                         break;
1479         }
1480         panic("tcp_pulloutofband");
1481 }
1482
1483 #endif /* notdef */
1484
1485 /*
1486  * Collect new round-trip time estimate
1487  * and update averages and current timeout.
1488  */
1489
1490 static void
1491 tcp_xmit_timer(register struct tcpcb *tp, int rtt)
1492 {
1493         register short delta;
1494
1495         DEBUG_CALL("tcp_xmit_timer");
1496         DEBUG_ARG("tp = %p", tp);
1497         DEBUG_ARG("rtt = %d", rtt);
1498
1499         if (tp->t_srtt != 0) {
1500                 /*
1501                  * srtt is stored as fixed point with 3 bits after the
1502                  * binary point (i.e., scaled by 8).  The following magic
1503                  * is equivalent to the smoothing algorithm in rfc793 with
1504                  * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1505                  * point).  Adjust rtt to origin 0.
1506                  */
1507                 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1508                 if ((tp->t_srtt += delta) <= 0)
1509                         tp->t_srtt = 1;
1510                 /*
1511                  * We accumulate a smoothed rtt variance (actually, a
1512                  * smoothed mean difference), then set the retransmit
1513                  * timer to smoothed rtt + 4 times the smoothed variance.
1514                  * rttvar is stored as fixed point with 2 bits after the
1515                  * binary point (scaled by 4).  The following is
1516                  * equivalent to rfc793 smoothing with an alpha of .75
1517                  * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
1518                  * rfc793's wired-in beta.
1519                  */
1520                 if (delta < 0)
1521                         delta = -delta;
1522                 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1523                 if ((tp->t_rttvar += delta) <= 0)
1524                         tp->t_rttvar = 1;
1525         } else {
1526                 /*
1527                  * No rtt measurement yet - use the unsmoothed rtt.
1528                  * Set the variance to half the rtt (so our first
1529                  * retransmit happens at 3*rtt).
1530                  */
1531                 tp->t_srtt = rtt << TCP_RTT_SHIFT;
1532                 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1533         }
1534         tp->t_rtt = 0;
1535         tp->t_rxtshift = 0;
1536
1537         /*
1538          * the retransmit should happen at rtt + 4 * rttvar.
1539          * Because of the way we do the smoothing, srtt and rttvar
1540          * will each average +1/2 tick of bias.  When we compute
1541          * the retransmit timer, we want 1/2 tick of rounding and
1542          * 1 extra tick because of +-1/2 tick uncertainty in the
1543          * firing of the timer.  The bias will give us exactly the
1544          * 1.5 tick we need.  But, because the bias is
1545          * statistical, we have to test that we don't drop below
1546          * the minimum feasible timer (which is 2 ticks).
1547          */
1548         TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1549             (short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1550
1551         /*
1552          * We received an ack for a packet that wasn't retransmitted;
1553          * it is probably safe to discard any error indications we've
1554          * received recently.  This isn't quite right, but close enough
1555          * for now (a route might have failed after we sent a segment,
1556          * and the return path might not be symmetrical).
1557          */
1558         tp->t_softerror = 0;
1559 }
1560
1561 /*
1562  * Determine a reasonable value for maxseg size.
1563  * If the route is known, check route for mtu.
1564  * If none, use an mss that can be handled on the outgoing
1565  * interface without forcing IP to fragment; if bigger than
1566  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1567  * to utilize large mbufs.  If no route is found, route has no mtu,
1568  * or the destination isn't local, use a default, hopefully conservative
1569  * size (usually 512 or the default IP max size, but no more than the mtu
1570  * of the interface), as we can't discover anything about intervening
1571  * gateways or networks.  We also initialize the congestion/slow start
1572  * window to be a single segment if the destination isn't local.
1573  * While looking at the routing entry, we also initialize other path-dependent
1574  * parameters from pre-set or cached values in the routing entry.
1575  */
1576
1577 int
1578 tcp_mss(struct tcpcb *tp, u_int offer)
1579 {
1580         struct socket *so = tp->t_socket;
1581         int mss;
1582
1583         DEBUG_CALL("tcp_mss");
1584         DEBUG_ARG("tp = %p", tp);
1585         DEBUG_ARG("offer = %d", offer);
1586
1587         switch (so->so_ffamily) {
1588         case AF_INET:
1589             mss = min(IF_MTU, IF_MRU) - sizeof(struct tcphdr)
1590                                       + sizeof(struct ip);
1591             break;
1592         case AF_INET6:
1593             mss = min(IF_MTU, IF_MRU) - sizeof(struct tcphdr)
1594                                       + sizeof(struct ip6);
1595             break;
1596         default:
1597             g_assert_not_reached();
1598         }
1599
1600         if (offer)
1601                 mss = min(mss, offer);
1602         mss = max(mss, 32);
1603         if (mss < tp->t_maxseg || offer != 0)
1604            tp->t_maxseg = mss;
1605
1606         tp->snd_cwnd = mss;
1607
1608         sbreserve(&so->so_snd, TCP_SNDSPACE + ((TCP_SNDSPACE % mss) ?
1609                                                (mss - (TCP_SNDSPACE % mss)) :
1610                                                0));
1611         sbreserve(&so->so_rcv, TCP_RCVSPACE + ((TCP_RCVSPACE % mss) ?
1612                                                (mss - (TCP_RCVSPACE % mss)) :
1613                                                0));
1614
1615         DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1616
1617         return mss;
1618 }