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 u_char  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 #define MAX_TCPOPTLEN   32      /* max # bytes that go in options */
  51
  52 /*
  53  * Tcp output routine: figure out what should be sent and send it.
  54  */
  55 int
  56 tcp_output(struct tcpcb *tp)
  57 {
  58         register struct socket *so = tp->t_socket;
  59         register long len, win;
  60         int off, flags, error;
  61         register struct mbuf *m;
  62         register struct tcpiphdr *ti;
  63         u_char opt[MAX_TCPOPTLEN];
  64         unsigned optlen, hdrlen;
  65         int idle, sendalot;
  66
  67         DEBUG_CALL("tcp_output");
  68         DEBUG_ARG("tp = %lx", (long )tp);
  69
  70         /*
  71          * Determine length of data that should be transmitted,
  72          * and flags that will be used.
  73          * If there is some data or critical controls (SYN, RST)
  74          * to send, then transmit; otherwise, investigate further.
  75          */
  76         idle = (tp->snd_max == tp->snd_una);
  77         if (idle && tp->t_idle >= tp->t_rxtcur)
  78                 /*
  79                  * We have been idle for "a while" and no acks are
  80                  * expected to clock out any data we send --
  81                  * slow start to get ack "clock" running again.
  82                  */
  83                 tp->snd_cwnd = tp->t_maxseg;
  84 again:
  85         sendalot = 0;
  86         off = tp->snd_nxt - tp->snd_una;
  87         win = min(tp->snd_wnd, tp->snd_cwnd);
  88
  89         flags = tcp_outflags[tp->t_state];
  90
  91         DEBUG_MISC((dfd, " --- tcp_output flags = 0x%x\n",flags));
  92
  93         /*
  94          * If in persist timeout with window of 0, send 1 byte.
  95          * Otherwise, if window is small but nonzero
  96          * and timer expired, we will send what we can
  97          * and go to transmit state.
  98          */
  99         if (tp->t_force) {
 100                 if (win == 0) {
 101                         /*
 102                          * If we still have some data to send, then
 103                          * clear the FIN bit.  Usually this would
 104                          * happen below when it realizes that we
 105                          * aren't sending all the data.  However,
 106                          * if we have exactly 1 byte of unset data,
 107                          * then it won't clear the FIN bit below,
 108                          * and if we are in persist state, we wind
 109                          * up sending the packet without recording
 110                          * that we sent the FIN bit.
 111                          *
 112                          * We can't just blindly clear the FIN bit,
 113                          * because if we don't have any more data
 114                          * to send then the probe will be the FIN
 115                          * itself.
 116                          */
 117                         if (off < so->so_snd.sb_cc)
 118                                 flags &= ~TH_FIN;
 119                         win = 1;
 120                 } else {
 121                         tp->t_timer[TCPT_PERSIST] = 0;
 122                         tp->t_rxtshift = 0;
 123                 }
 124         }
 125
 126         len = min(so->so_snd.sb_cc, win) - off;
 127
 128         if (len < 0) {
 129                 /*
 130                  * If FIN has been sent but not acked,
 131                  * but we haven't been called to retransmit,
 132                  * len will be -1.  Otherwise, window shrank
 133                  * after we sent into it.  If window shrank to 0,
 134                  * cancel pending retransmit and pull snd_nxt
 135                  * back to (closed) window.  We will enter persist
 136                  * state below.  If the window didn't close completely,
 137                  * just wait for an ACK.
 138                  */
 139                 len = 0;
 140                 if (win == 0) {
 141                         tp->t_timer[TCPT_REXMT] = 0;
 142                         tp->snd_nxt = tp->snd_una;
 143                 }
 144         }
 145
 146         if (len > tp->t_maxseg) {
 147                 len = tp->t_maxseg;
 148                 sendalot = 1;
 149         }
 150         if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
 151                 flags &= ~TH_FIN;
 152
 153         win = sbspace(&so->so_rcv);
 154
 155         /*
 156          * Sender silly window avoidance.  If connection is idle
 157          * and can send all data, a maximum segment,
 158          * at least a maximum default-size segment do it,
 159          * or are forced, do it; otherwise don't bother.
 160          * If peer's buffer is tiny, then send
 161          * when window is at least half open.
 162          * If retransmitting (possibly after persist timer forced us
 163          * to send into a small window), then must resend.
 164          */
 165         if (len) {
 166                 if (len == tp->t_maxseg)
 167                         goto send;
 168                 if ((1 || idle || tp->t_flags & TF_NODELAY) &&
 169                     len + off >= so->so_snd.sb_cc)
 170                         goto send;
 171                 if (tp->t_force)
 172                         goto send;
 173                 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
 174                         goto send;
 175                 if (SEQ_LT(tp->snd_nxt, tp->snd_max))
 176                         goto send;
 177         }
 178
 179         /*
 180          * Compare available window to amount of window
 181          * known to peer (as advertised window less
 182          * next expected input).  If the difference is at least two
 183          * max size segments, or at least 50% of the maximum possible
 184          * window, then want to send a window update to peer.
 185          */
 186         if (win > 0) {
 187                 /*
 188                  * "adv" is the amount we can increase the window,
 189                  * taking into account that we are limited by
 190                  * TCP_MAXWIN << tp->rcv_scale.
 191                  */
 192                 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) -
 193                         (tp->rcv_adv - tp->rcv_nxt);
 194
 195                 if (adv >= (long) (2 * tp->t_maxseg))
 196                         goto send;
 197                 if (2 * adv >= (long) so->so_rcv.sb_datalen)
 198                         goto send;
 199         }
 200
 201         /*
 202          * Send if we owe peer an ACK.
 203          */
 204         if (tp->t_flags & TF_ACKNOW)
 205                 goto send;
 206         if (flags & (TH_SYN|TH_RST))
 207                 goto send;
 208         if (SEQ_GT(tp->snd_up, tp->snd_una))
 209                 goto send;
 210         /*
 211          * If our state indicates that FIN should be sent
 212          * and we have not yet done so, or we're retransmitting the FIN,
 213          * then we need to send.
 214          */
 215         if (flags & TH_FIN &&
 216             ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
 217                 goto send;
 218
 219         /*
 220          * TCP window updates are not reliable, rather a polling protocol
 221          * using ``persist'' packets is used to insure receipt of window
 222          * updates.  The three ``states'' for the output side are:
 223          *      idle                    not doing retransmits or persists
 224          *      persisting              to move a small or zero window
 225          *      (re)transmitting        and thereby not persisting
 226          *
 227          * tp->t_timer[TCPT_PERSIST]
 228          *      is set when we are in persist state.
 229          * tp->t_force
 230          *      is set when we are called to send a persist packet.
 231          * tp->t_timer[TCPT_REXMT]
 232          *      is set when we are retransmitting
 233          * The output side is idle when both timers are zero.
 234          *
 235          * If send window is too small, there is data to transmit, and no
 236          * retransmit or persist is pending, then go to persist state.
 237          * If nothing happens soon, send when timer expires:
 238          * if window is nonzero, transmit what we can,
 239          * otherwise force out a byte.
 240          */
 241         if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 &&
 242             tp->t_timer[TCPT_PERSIST] == 0) {
 243                 tp->t_rxtshift = 0;
 244                 tcp_setpersist(tp);
 245         }
 246
 247         /*
 248          * No reason to send a segment, just return.
 249          */
 250         STAT(tcpstat.tcps_didnuttin++);
 251
 252         return (0);
 253
 254 send:
 255         /*
 256          * Before ESTABLISHED, force sending of initial options
 257          * unless TCP set not to do any options.
 258          * NOTE: we assume that the IP/TCP header plus TCP options
 259          * always fit in a single mbuf, leaving room for a maximum
 260          * link header, i.e.
 261          *      max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MHLEN
 262          */
 263         optlen = 0;
 264         hdrlen = sizeof (struct tcpiphdr);
 265         if (flags & TH_SYN) {
 266                 tp->snd_nxt = tp->iss;
 267                 if ((tp->t_flags & TF_NOOPT) == 0) {
 268                         u_int16_t mss;
 269
 270                         opt[0] = TCPOPT_MAXSEG;
 271                         opt[1] = 4;
 272                         mss = htons((u_int16_t) tcp_mss(tp, 0));
 273                         memcpy((caddr_t)(opt + 2), (caddr_t)&mss, sizeof(mss));
 274                         optlen = 4;
 275                 }
 276         }
 277
 278         hdrlen += optlen;
 279
 280         /*
 281          * Adjust data length if insertion of options will
 282          * bump the packet length beyond the t_maxseg length.
 283          */
 284          if (len > tp->t_maxseg - optlen) {
 285                 len = tp->t_maxseg - optlen;
 286                 sendalot = 1;
 287          }
 288
 289         /*
 290          * Grab a header mbuf, attaching a copy of data to
 291          * be transmitted, and initialize the header from
 292          * the template for sends on this connection.
 293          */
 294         if (len) {
 295                 if (tp->t_force && len == 1)
 296                         STAT(tcpstat.tcps_sndprobe++);
 297                 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) {
 298                         STAT(tcpstat.tcps_sndrexmitpack++);
 299                         STAT(tcpstat.tcps_sndrexmitbyte += len);
 300                 } else {
 301                         STAT(tcpstat.tcps_sndpack++);
 302                         STAT(tcpstat.tcps_sndbyte += len);
 303                 }
 304
 305                 m = m_get();
 306                 if (m == NULL) {
 307                         error = 1;
 308                         goto out;
 309                 }
 310                 m->m_data += IF_MAXLINKHDR;
 311                 m->m_len = hdrlen;
 312
 313                 sbcopy(&so->so_snd, off, (int) len, mtod(m, caddr_t) + hdrlen);
 314                 m->m_len += len;
 315
 316                 /*
 317                  * If we're sending everything we've got, set PUSH.
 318                  * (This will keep happy those implementations which only
 319                  * give data to the user when a buffer fills or
 320                  * a PUSH comes in.)
 321                  */
 322                 if (off + len == so->so_snd.sb_cc)
 323                         flags |= TH_PUSH;
 324         } else {
 325                 if (tp->t_flags & TF_ACKNOW)
 326                         STAT(tcpstat.tcps_sndacks++);
 327                 else if (flags & (TH_SYN|TH_FIN|TH_RST))
 328                         STAT(tcpstat.tcps_sndctrl++);
 329                 else if (SEQ_GT(tp->snd_up, tp->snd_una))
 330                         STAT(tcpstat.tcps_sndurg++);
 331                 else
 332                         STAT(tcpstat.tcps_sndwinup++);
 333
 334                 m = m_get();
 335                 if (m == NULL) {
 336                         error = 1;
 337                         goto out;
 338                 }
 339                 m->m_data += IF_MAXLINKHDR;
 340                 m->m_len = hdrlen;
 341         }
 342
 343         ti = mtod(m, struct tcpiphdr *);
 344
 345         memcpy((caddr_t)ti, &tp->t_template, sizeof (struct tcpiphdr));
 346
 347         /*
 348          * Fill in fields, remembering maximum advertised
 349          * window for use in delaying messages about window sizes.
 350          * If resending a FIN, be sure not to use a new sequence number.
 351          */
 352         if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
 353             tp->snd_nxt == tp->snd_max)
 354                 tp->snd_nxt--;
 355         /*
 356          * If we are doing retransmissions, then snd_nxt will
 357          * not reflect the first unsent octet.  For ACK only
 358          * packets, we do not want the sequence number of the
 359          * retransmitted packet, we want the sequence number
 360          * of the next unsent octet.  So, if there is no data
 361          * (and no SYN or FIN), use snd_max instead of snd_nxt
 362          * when filling in ti_seq.  But if we are in persist
 363          * state, snd_max might reflect one byte beyond the
 364          * right edge of the window, so use snd_nxt in that
 365          * case, since we know we aren't doing a retransmission.
 366          * (retransmit and persist are mutually exclusive...)
 367          */
 368         if (len || (flags & (TH_SYN|TH_FIN)) || tp->t_timer[TCPT_PERSIST])
 369                 ti->ti_seq = htonl(tp->snd_nxt);
 370         else
 371                 ti->ti_seq = htonl(tp->snd_max);
 372         ti->ti_ack = htonl(tp->rcv_nxt);
 373         if (optlen) {
 374                 memcpy((caddr_t)(ti + 1), (caddr_t)opt, optlen);
 375                 ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2;
 376         }
 377         ti->ti_flags = flags;
 378         /*
 379          * Calculate receive window.  Don't shrink window,
 380          * but avoid silly window syndrome.
 381          */
 382         if (win < (long)(so->so_rcv.sb_datalen / 4) && win < (long)tp->t_maxseg)
 383                 win = 0;
 384         if (win > (long)TCP_MAXWIN << tp->rcv_scale)
 385                 win = (long)TCP_MAXWIN << tp->rcv_scale;
 386         if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
 387                 win = (long)(tp->rcv_adv - tp->rcv_nxt);
 388         ti->ti_win = htons((u_int16_t) (win>>tp->rcv_scale));
 389
 390         if (SEQ_GT(tp->snd_up, tp->snd_una)) {
 391                 ti->ti_urp = htons((u_int16_t)(tp->snd_up - ntohl(ti->ti_seq)));
 392                 ti->ti_flags |= TH_URG;
 393         } else
 394                 /*
 395                  * If no urgent pointer to send, then we pull
 396                  * the urgent pointer to the left edge of the send window
 397                  * so that it doesn't drift into the send window on sequence
 398                  * number wraparound.
 399                  */
 400                 tp->snd_up = tp->snd_una;               /* drag it along */
 401
 402         /*
 403          * Put TCP length in extended header, and then
 404          * checksum extended header and data.
 405          */
 406         if (len + optlen)
 407                 ti->ti_len = htons((u_int16_t)(sizeof (struct tcphdr) +
 408                     optlen + len));
 409         ti->ti_sum = cksum(m, (int)(hdrlen + len));
 410
 411         /*
 412          * In transmit state, time the transmission and arrange for
 413          * the retransmit.  In persist state, just set snd_max.
 414          */
 415         if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) {
 416                 tcp_seq startseq = tp->snd_nxt;
 417
 418                 /*
 419                  * Advance snd_nxt over sequence space of this segment.
 420                  */
 421                 if (flags & (TH_SYN|TH_FIN)) {
 422                         if (flags & TH_SYN)
 423                                 tp->snd_nxt++;
 424                         if (flags & TH_FIN) {
 425                                 tp->snd_nxt++;
 426                                 tp->t_flags |= TF_SENTFIN;
 427                         }
 428                 }
 429                 tp->snd_nxt += len;
 430                 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
 431                         tp->snd_max = tp->snd_nxt;
 432                         /*
 433                          * Time this transmission if not a retransmission and
 434                          * not currently timing anything.
 435                          */
 436                         if (tp->t_rtt == 0) {
 437                                 tp->t_rtt = 1;
 438                                 tp->t_rtseq = startseq;
 439                                 STAT(tcpstat.tcps_segstimed++);
 440                         }
 441                 }
 442
 443                 /*
 444                  * Set retransmit timer if not currently set,
 445                  * and not doing an ack or a keep-alive probe.
 446                  * Initial value for retransmit timer is smoothed
 447                  * round-trip time + 2 * round-trip time variance.
 448                  * Initialize shift counter which is used for backoff
 449                  * of retransmit time.
 450                  */
 451                 if (tp->t_timer[TCPT_REXMT] == 0 &&
 452                     tp->snd_nxt != tp->snd_una) {
 453                         tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
 454                         if (tp->t_timer[TCPT_PERSIST]) {
 455                                 tp->t_timer[TCPT_PERSIST] = 0;
 456                                 tp->t_rxtshift = 0;
 457                         }
 458                 }
 459         } else
 460                 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max))
 461                         tp->snd_max = tp->snd_nxt + len;
 462
 463         /*
 464          * Fill in IP length and desired time to live and
 465          * send to IP level.  There should be a better way
 466          * to handle ttl and tos; we could keep them in
 467          * the template, but need a way to checksum without them.
 468          */
 469         m->m_len = hdrlen + len; /* XXX Needed? m_len should be correct */
 470
 471     {
 472
 473         ((struct ip *)ti)->ip_len = m->m_len;
 474
 475         ((struct ip *)ti)->ip_ttl = IPDEFTTL;
 476         ((struct ip *)ti)->ip_tos = so->so_iptos;
 477
 478         error = ip_output(so, m);
 479     }
 480         if (error) {
 481 out:
 482                 return (error);
 483         }
 484         STAT(tcpstat.tcps_sndtotal++);
 485
 486         /*
 487          * Data sent (as far as we can tell).
 488          * If this advertises a larger window than any other segment,
 489          * then remember the size of the advertised window.
 490          * Any pending ACK has now been sent.
 491          */
 492         if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
 493                 tp->rcv_adv = tp->rcv_nxt + win;
 494         tp->last_ack_sent = tp->rcv_nxt;
 495         tp->t_flags &= ~(TF_ACKNOW|TF_DELACK);
 496         if (sendalot)
 497                 goto again;
 498
 499         return (0);
 500 }
 501
 502 void
 503 tcp_setpersist(struct tcpcb *tp)
 504 {
 505     int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
 506
 507         /*
 508          * Start/restart persistence timer.
 509          */
 510         TCPT_RANGESET(tp->t_timer[TCPT_PERSIST],
 511             t * tcp_backoff[tp->t_rxtshift],
 512             TCPTV_PERSMIN, TCPTV_PERSMAX);
 513         if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
 514                 tp->t_rxtshift++;
 515 }