slirp/tcp_output.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
   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_output.c        8.3 (Berkeley) 12/30/93
  30  * tcp_output.c,v 1.3 1994/09/15 10:36:55 davidg 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 "slirp.h"
  42
  43 static const uint8_t  tcp_outflags[TCP_NSTATES] = {
  44         TH_RST|TH_ACK, 0,      TH_SYN,        TH_SYN|TH_ACK,
  45         TH_ACK,        TH_ACK, TH_FIN|TH_ACK, TH_FIN|TH_ACK,
  46         TH_FIN|TH_ACK, TH_ACK, TH_ACK,
  47 };
  48
  49
  50 #undef MAX_TCPOPTLEN
  51 #define MAX_TCPOPTLEN   32      /* max # bytes that go in options */
  52
  53 /*
  54  * Tcp output routine: figure out what should be sent and send it.
  55  */
  56 int
  57 tcp_output(struct tcpcb *tp)
  58 {
  59         register struct socket *so = tp->t_socket;
  60         register long len, win;
  61         int off, flags, error;
  62         register struct mbuf *m;
  63         register struct tcpiphdr *ti, tcpiph_save;
  64         struct ip *ip;
  65         struct ip6 *ip6;
  66         uint8_t opt[MAX_TCPOPTLEN];
  67         unsigned optlen, hdrlen;
  68         int idle, sendalot;
  69
  70         DEBUG_CALL("tcp_output");
  71         DEBUG_ARG("tp = %p", tp);
  72
  73         /*
  74          * Determine length of data that should be transmitted,
  75          * and flags that will be used.
  76          * If there is some data or critical controls (SYN, RST)
  77          * to send, then transmit; otherwise, investigate further.
  78          */
  79         idle = (tp->snd_max == tp->snd_una);
  80         if (idle && tp->t_idle >= tp->t_rxtcur)
  81                 /*
  82                  * We have been idle for "a while" and no acks are
  83                  * expected to clock out any data we send --
  84                  * slow start to get ack "clock" running again.
  85                  */
  86                 tp->snd_cwnd = tp->t_maxseg;
  87 again:
  88         sendalot = 0;
  89         off = tp->snd_nxt - tp->snd_una;
  90         win = MIN(tp->snd_wnd, tp->snd_cwnd);
  91
  92         flags = tcp_outflags[tp->t_state];
  93
  94         DEBUG_MISC(" --- tcp_output flags = 0x%x", flags);
  95
  96         /*
  97          * If in persist timeout with window of 0, send 1 byte.
  98          * Otherwise, if window is small but nonzero
  99          * and timer expired, we will send what we can
 100          * and go to transmit state.
 101          */
 102         if (tp->t_force) {
 103                 if (win == 0) {
 104                         /*
 105                          * If we still have some data to send, then
 106                          * clear the FIN bit.  Usually this would
 107                          * happen below when it realizes that we
 108                          * aren't sending all the data.  However,
 109                          * if we have exactly 1 byte of unset data,
 110                          * then it won't clear the FIN bit below,
 111                          * and if we are in persist state, we wind
 112                          * up sending the packet without recording
 113                          * that we sent the FIN bit.
 114                          *
 115                          * We can't just blindly clear the FIN bit,
 116                          * because if we don't have any more data
 117                          * to send then the probe will be the FIN
 118                          * itself.
 119                          */
 120                         if (off < so->so_snd.sb_cc)
 121                                 flags &= ~TH_FIN;
 122                         win = 1;
 123                 } else {
 124                         tp->t_timer[TCPT_PERSIST] = 0;
 125                         tp->t_rxtshift = 0;
 126                 }
 127         }
 128
 129         len = MIN(so->so_snd.sb_cc, win) - off;
 130
 131         if (len < 0) {
 132                 /*
 133                  * If FIN has been sent but not acked,
 134                  * but we haven't been called to retransmit,
 135                  * len will be -1.  Otherwise, window shrank
 136                  * after we sent into it.  If window shrank to 0,
 137                  * cancel pending retransmit and pull snd_nxt
 138                  * back to (closed) window.  We will enter persist
 139                  * state below.  If the window didn't close completely,
 140                  * just wait for an ACK.
 141                  */
 142                 len = 0;
 143                 if (win == 0) {
 144                         tp->t_timer[TCPT_REXMT] = 0;
 145                         tp->snd_nxt = tp->snd_una;
 146                 }
 147         }
 148
 149         if (len > tp->t_maxseg) {
 150                 len = tp->t_maxseg;
 151                 sendalot = 1;
 152         }
 153         if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
 154                 flags &= ~TH_FIN;
 155
 156         win = sbspace(&so->so_rcv);
 157
 158         /*
 159          * Sender silly window avoidance.  If connection is idle
 160          * and can send all data, a maximum segment,
 161          * at least a maximum default-size segment do it,
 162          * or are forced, do it; otherwise don't bother.
 163          * If peer's buffer is tiny, then send
 164          * when window is at least half open.
 165          * If retransmitting (possibly after persist timer forced us
 166          * to send into a small window), then must resend.
 167          */
 168         if (len) {
 169                 if (len == tp->t_maxseg)
 170                         goto send;
 171                 if ((1 || idle || tp->t_flags & TF_NODELAY) &&
 172                     len + off >= so->so_snd.sb_cc)
 173                         goto send;
 174                 if (tp->t_force)
 175                         goto send;
 176                 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
 177                         goto send;
 178                 if (SEQ_LT(tp->snd_nxt, tp->snd_max))
 179                         goto send;
 180         }
 181
 182         /*
 183          * Compare available window to amount of window
 184          * known to peer (as advertised window less
 185          * next expected input).  If the difference is at least two
 186          * max size segments, or at least 50% of the maximum possible
 187          * window, then want to send a window update to peer.
 188          */
 189         if (win > 0) {
 190                 /*
 191                  * "adv" is the amount we can increase the window,
 192                  * taking into account that we are limited by
 193                  * TCP_MAXWIN << tp->rcv_scale.
 194                  */
 195                 long adv = MIN(win, (long)TCP_MAXWIN << tp->rcv_scale) -
 196                         (tp->rcv_adv - tp->rcv_nxt);
 197
 198                 if (adv >= (long) (2 * tp->t_maxseg))
 199                         goto send;
 200                 if (2 * adv >= (long) so->so_rcv.sb_datalen)
 201                         goto send;
 202         }
 203
 204         /*
 205          * Send if we owe peer an ACK.
 206          */
 207         if (tp->t_flags & TF_ACKNOW)
 208                 goto send;
 209         if (flags & (TH_SYN|TH_RST))
 210                 goto send;
 211         if (SEQ_GT(tp->snd_up, tp->snd_una))
 212                 goto send;
 213         /*
 214          * If our state indicates that FIN should be sent
 215          * and we have not yet done so, or we're retransmitting the FIN,
 216          * then we need to send.
 217          */
 218         if (flags & TH_FIN &&
 219             ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
 220                 goto send;
 221
 222         /*
 223          * TCP window updates are not reliable, rather a polling protocol
 224          * using ``persist'' packets is used to insure receipt of window
 225          * updates.  The three ``states'' for the output side are:
 226          *      idle                    not doing retransmits or persists
 227          *      persisting              to move a small or zero window
 228          *      (re)transmitting        and thereby not persisting
 229          *
 230          * tp->t_timer[TCPT_PERSIST]
 231          *      is set when we are in persist state.
 232          * tp->t_force
 233          *      is set when we are called to send a persist packet.
 234          * tp->t_timer[TCPT_REXMT]
 235          *      is set when we are retransmitting
 236          * The output side is idle when both timers are zero.
 237          *
 238          * If send window is too small, there is data to transmit, and no
 239          * retransmit or persist is pending, then go to persist state.
 240          * If nothing happens soon, send when timer expires:
 241          * if window is nonzero, transmit what we can,
 242          * otherwise force out a byte.
 243          */
 244         if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 &&
 245             tp->t_timer[TCPT_PERSIST] == 0) {
 246                 tp->t_rxtshift = 0;
 247                 tcp_setpersist(tp);
 248         }
 249
 250         /*
 251          * No reason to send a segment, just return.
 252          */
 253         return (0);
 254
 255 send:
 256         /*
 257          * Before ESTABLISHED, force sending of initial options
 258          * unless TCP set not to do any options.
 259          * NOTE: we assume that the IP/TCP header plus TCP options
 260          * always fit in a single mbuf, leaving room for a maximum
 261          * link header, i.e.
 262          *      max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MHLEN
 263          */
 264         optlen = 0;
 265         hdrlen = sizeof (struct tcpiphdr);
 266         if (flags & TH_SYN) {
 267                 tp->snd_nxt = tp->iss;
 268                 if ((tp->t_flags & TF_NOOPT) == 0) {
 269                         uint16_t mss;
 270
 271                         opt[0] = TCPOPT_MAXSEG;
 272                         opt[1] = 4;
 273                         mss = htons((uint16_t) tcp_mss(tp, 0));
 274                         memcpy((char *)(opt + 2), (char *)&mss, sizeof(mss));
 275                         optlen = 4;
 276                 }
 277         }
 278
 279         hdrlen += optlen;
 280
 281         /*
 282          * Adjust data length if insertion of options will
 283          * bump the packet length beyond the t_maxseg length.
 284          */
 285          if (len > tp->t_maxseg - optlen) {
 286                 len = tp->t_maxseg - optlen;
 287                 sendalot = 1;
 288          }
 289
 290         /*
 291          * Grab a header mbuf, attaching a copy of data to
 292          * be transmitted, and initialize the header from
 293          * the template for sends on this connection.
 294          */
 295         if (len) {
 296                 m = m_get(so->slirp);
 297                 if (m == NULL) {
 298                         error = 1;
 299                         goto out;
 300                 }
 301                 m->m_data += IF_MAXLINKHDR;
 302                 m->m_len = hdrlen;
 303
 304                 sbcopy(&so->so_snd, off, (int) len, mtod(m, char *) + hdrlen);
 305                 m->m_len += len;
 306
 307                 /*
 308                  * If we're sending everything we've got, set PUSH.
 309                  * (This will keep happy those implementations which only
 310                  * give data to the user when a buffer fills or
 311                  * a PUSH comes in.)
 312                  */
 313                 if (off + len == so->so_snd.sb_cc)
 314                         flags |= TH_PUSH;
 315         } else {
 316                 m = m_get(so->slirp);
 317                 if (m == NULL) {
 318                         error = 1;
 319                         goto out;
 320                 }
 321                 m->m_data += IF_MAXLINKHDR;
 322                 m->m_len = hdrlen;
 323         }
 324
 325         ti = mtod(m, struct tcpiphdr *);
 326
 327         memcpy((char *)ti, &tp->t_template, sizeof (struct tcpiphdr));
 328
 329         /*
 330          * Fill in fields, remembering maximum advertised
 331          * window for use in delaying messages about window sizes.
 332          * If resending a FIN, be sure not to use a new sequence number.
 333          */
 334         if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
 335             tp->snd_nxt == tp->snd_max)
 336                 tp->snd_nxt--;
 337         /*
 338          * If we are doing retransmissions, then snd_nxt will
 339          * not reflect the first unsent octet.  For ACK only
 340          * packets, we do not want the sequence number of the
 341          * retransmitted packet, we want the sequence number
 342          * of the next unsent octet.  So, if there is no data
 343          * (and no SYN or FIN), use snd_max instead of snd_nxt
 344          * when filling in ti_seq.  But if we are in persist
 345          * state, snd_max might reflect one byte beyond the
 346          * right edge of the window, so use snd_nxt in that
 347          * case, since we know we aren't doing a retransmission.
 348          * (retransmit and persist are mutually exclusive...)
 349          */
 350         if (len || (flags & (TH_SYN|TH_FIN)) || tp->t_timer[TCPT_PERSIST])
 351                 ti->ti_seq = htonl(tp->snd_nxt);
 352         else
 353                 ti->ti_seq = htonl(tp->snd_max);
 354         ti->ti_ack = htonl(tp->rcv_nxt);
 355         if (optlen) {
 356                 memcpy((char *)(ti + 1), (char *)opt, optlen);
 357                 ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2;
 358         }
 359         ti->ti_flags = flags;
 360         /*
 361          * Calculate receive window.  Don't shrink window,
 362          * but avoid silly window syndrome.
 363          */
 364         if (win < (long)(so->so_rcv.sb_datalen / 4) && win < (long)tp->t_maxseg)
 365                 win = 0;
 366         if (win > (long)TCP_MAXWIN << tp->rcv_scale)
 367                 win = (long)TCP_MAXWIN << tp->rcv_scale;
 368         if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
 369                 win = (long)(tp->rcv_adv - tp->rcv_nxt);
 370         ti->ti_win = htons((uint16_t) (win>>tp->rcv_scale));
 371
 372         if (SEQ_GT(tp->snd_up, tp->snd_una)) {
 373                 ti->ti_urp = htons((uint16_t)(tp->snd_up - ntohl(ti->ti_seq)));
 374                 ti->ti_flags |= TH_URG;
 375         } else
 376                 /*
 377                  * If no urgent pointer to send, then we pull
 378                  * the urgent pointer to the left edge of the send window
 379                  * so that it doesn't drift into the send window on sequence
 380                  * number wraparound.
 381                  */
 382                 tp->snd_up = tp->snd_una;               /* drag it along */
 383
 384         /*
 385          * Put TCP length in extended header, and then
 386          * checksum extended header and data.
 387          */
 388         if (len + optlen)
 389                 ti->ti_len = htons((uint16_t)(sizeof (struct tcphdr) +
 390                     optlen + len));
 391         ti->ti_sum = cksum(m, (int)(hdrlen + len));
 392
 393         /*
 394          * In transmit state, time the transmission and arrange for
 395          * the retransmit.  In persist state, just set snd_max.
 396          */
 397         if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) {
 398                 tcp_seq startseq = tp->snd_nxt;
 399
 400                 /*
 401                  * Advance snd_nxt over sequence space of this segment.
 402                  */
 403                 if (flags & (TH_SYN|TH_FIN)) {
 404                         if (flags & TH_SYN)
 405                                 tp->snd_nxt++;
 406                         if (flags & TH_FIN) {
 407                                 tp->snd_nxt++;
 408                                 tp->t_flags |= TF_SENTFIN;
 409                         }
 410                 }
 411                 tp->snd_nxt += len;
 412                 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
 413                         tp->snd_max = tp->snd_nxt;
 414                         /*
 415                          * Time this transmission if not a retransmission and
 416                          * not currently timing anything.
 417                          */
 418                         if (tp->t_rtt == 0) {
 419                                 tp->t_rtt = 1;
 420                                 tp->t_rtseq = startseq;
 421                         }
 422                 }
 423
 424                 /*
 425                  * Set retransmit timer if not currently set,
 426                  * and not doing an ack or a keep-alive probe.
 427                  * Initial value for retransmit timer is smoothed
 428                  * round-trip time + 2 * round-trip time variance.
 429                  * Initialize shift counter which is used for backoff
 430                  * of retransmit time.
 431                  */
 432                 if (tp->t_timer[TCPT_REXMT] == 0 &&
 433                     tp->snd_nxt != tp->snd_una) {
 434                         tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
 435                         if (tp->t_timer[TCPT_PERSIST]) {
 436                                 tp->t_timer[TCPT_PERSIST] = 0;
 437                                 tp->t_rxtshift = 0;
 438                         }
 439                 }
 440         } else
 441                 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max))
 442                         tp->snd_max = tp->snd_nxt + len;
 443
 444         /*
 445          * Fill in IP length and desired time to live and
 446          * send to IP level.  There should be a better way
 447          * to handle ttl and tos; we could keep them in
 448          * the template, but need a way to checksum without them.
 449          */
 450         m->m_len = hdrlen + len; /* XXX Needed? m_len should be correct */
 451         tcpiph_save = *mtod(m, struct tcpiphdr *);
 452
 453         switch (so->so_ffamily) {
 454         case AF_INET:
 455             m->m_data += sizeof(struct tcpiphdr) - sizeof(struct tcphdr)
 456                                                  - sizeof(struct ip);
 457             m->m_len  -= sizeof(struct tcpiphdr) - sizeof(struct tcphdr)
 458                                                  - sizeof(struct ip);
 459             ip = mtod(m, struct ip *);
 460
 461             ip->ip_len = m->m_len;
 462             ip->ip_dst = tcpiph_save.ti_dst;
 463             ip->ip_src = tcpiph_save.ti_src;
 464             ip->ip_p = tcpiph_save.ti_pr;
 465
 466             ip->ip_ttl = IPDEFTTL;
 467             ip->ip_tos = so->so_iptos;
 468             error = ip_output(so, m);
 469             break;
 470
 471         case AF_INET6:
 472             m->m_data += sizeof(struct tcpiphdr) - sizeof(struct tcphdr)
 473                                                  - sizeof(struct ip6);
 474             m->m_len  -= sizeof(struct tcpiphdr) - sizeof(struct tcphdr)
 475                                                  - sizeof(struct ip6);
 476             ip6 = mtod(m, struct ip6 *);
 477
 478             ip6->ip_pl = tcpiph_save.ti_len;
 479             ip6->ip_dst = tcpiph_save.ti_dst6;
 480             ip6->ip_src = tcpiph_save.ti_src6;
 481             ip6->ip_nh = tcpiph_save.ti_nh6;
 482
 483             error = ip6_output(so, m, 0);
 484             break;
 485
 486         default:
 487             g_assert_not_reached();
 488         }
 489
 490         if (error) {
 491 out:
 492                 return (error);
 493         }
 494
 495         /*
 496          * Data sent (as far as we can tell).
 497          * If this advertises a larger window than any other segment,
 498          * then remember the size of the advertised window.
 499          * Any pending ACK has now been sent.
 500          */
 501         if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
 502                 tp->rcv_adv = tp->rcv_nxt + win;
 503         tp->last_ack_sent = tp->rcv_nxt;
 504         tp->t_flags &= ~(TF_ACKNOW|TF_DELACK);
 505         if (sendalot)
 506                 goto again;
 507
 508         return (0);
 509 }
 510
 511 void
 512 tcp_setpersist(struct tcpcb *tp)
 513 {
 514     int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
 515
 516         /*
 517          * Start/restart persistence timer.
 518          */
 519         TCPT_RANGESET(tp->t_timer[TCPT_PERSIST],
 520             t * tcp_backoff[tp->t_rxtshift],
 521             TCPTV_PERSMIN, TCPTV_PERSMAX);
 522         if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
 523                 tp->t_rxtshift++;
 524 }