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 "qemu/osdep.h"
  42 #include "slirp.h"
  43
  44 static const u_char  tcp_outflags[TCP_NSTATES] = {
  45         TH_RST|TH_ACK, 0,      TH_SYN,        TH_SYN|TH_ACK,
  46         TH_ACK,        TH_ACK, TH_FIN|TH_ACK, TH_FIN|TH_ACK,
  47         TH_FIN|TH_ACK, TH_ACK, TH_ACK,
  48 };
  49
  50
  51 #undef MAX_TCPOPTLEN
  52 #define MAX_TCPOPTLEN   32      /* max # bytes that go in options */
  53
  54 /*
  55  * Tcp output routine: figure out what should be sent and send it.
  56  */
  57 int
  58 tcp_output(struct tcpcb *tp)
  59 {
  60         register struct socket *so = tp->t_socket;
  61         register long len, win;
  62         int off, flags, error;
  63         register struct mbuf *m;
  64         register struct tcpiphdr *ti, tcpiph_save;
  65         struct ip *ip;
  66         struct ip6 *ip6;
  67         u_char opt[MAX_TCPOPTLEN];
  68         unsigned optlen, hdrlen;
  69         int idle, sendalot;
  70
  71         DEBUG_CALL("tcp_output");
  72         DEBUG_ARG("tp = %p", tp);
  73
  74         /*
  75          * Determine length of data that should be transmitted,
  76          * and flags that will be used.
  77          * If there is some data or critical controls (SYN, RST)
  78          * to send, then transmit; otherwise, investigate further.
  79          */
  80         idle = (tp->snd_max == tp->snd_una);
  81         if (idle && tp->t_idle >= tp->t_rxtcur)
  82                 /*
  83                  * We have been idle for "a while" and no acks are
  84                  * expected to clock out any data we send --
  85                  * slow start to get ack "clock" running again.
  86                  */
  87                 tp->snd_cwnd = tp->t_maxseg;
  88 again:
  89         sendalot = 0;
  90         off = tp->snd_nxt - tp->snd_una;
  91         win = MIN(tp->snd_wnd, tp->snd_cwnd);
  92
  93         flags = tcp_outflags[tp->t_state];
  94
  95         DEBUG_MISC(" --- tcp_output flags = 0x%x", flags);
  96
  97         /*
  98          * If in persist timeout with window of 0, send 1 byte.
  99          * Otherwise, if window is small but nonzero
 100          * and timer expired, we will send what we can
 101          * and go to transmit state.
 102          */
 103         if (tp->t_force) {
 104                 if (win == 0) {
 105                         /*
 106                          * If we still have some data to send, then
 107                          * clear the FIN bit.  Usually this would
 108                          * happen below when it realizes that we
 109                          * aren't sending all the data.  However,
 110                          * if we have exactly 1 byte of unset data,
 111                          * then it won't clear the FIN bit below,
 112                          * and if we are in persist state, we wind
 113                          * up sending the packet without recording
 114                          * that we sent the FIN bit.
 115                          *
 116                          * We can't just blindly clear the FIN bit,
 117                          * because if we don't have any more data
 118                          * to send then the probe will be the FIN
 119                          * itself.
 120                          */
 121                         if (off < so->so_snd.sb_cc)
 122                                 flags &= ~TH_FIN;
 123                         win = 1;
 124                 } else {
 125                         tp->t_timer[TCPT_PERSIST] = 0;
 126                         tp->t_rxtshift = 0;
 127                 }
 128         }
 129
 130         len = MIN(so->so_snd.sb_cc, win) - off;
 131
 132         if (len < 0) {
 133                 /*
 134                  * If FIN has been sent but not acked,
 135                  * but we haven't been called to retransmit,
 136                  * len will be -1.  Otherwise, window shrank
 137                  * after we sent into it.  If window shrank to 0,
 138                  * cancel pending retransmit and pull snd_nxt
 139                  * back to (closed) window.  We will enter persist
 140                  * state below.  If the window didn't close completely,
 141                  * just wait for an ACK.
 142                  */
 143                 len = 0;
 144                 if (win == 0) {
 145                         tp->t_timer[TCPT_REXMT] = 0;
 146                         tp->snd_nxt = tp->snd_una;
 147                 }
 148         }
 149
 150         if (len > tp->t_maxseg) {
 151                 len = tp->t_maxseg;
 152                 sendalot = 1;
 153         }
 154         if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
 155                 flags &= ~TH_FIN;
 156
 157         win = sbspace(&so->so_rcv);
 158
 159         /*
 160          * Sender silly window avoidance.  If connection is idle
 161          * and can send all data, a maximum segment,
 162          * at least a maximum default-size segment do it,
 163          * or are forced, do it; otherwise don't bother.
 164          * If peer's buffer is tiny, then send
 165          * when window is at least half open.
 166          * If retransmitting (possibly after persist timer forced us
 167          * to send into a small window), then must resend.
 168          */
 169         if (len) {
 170                 if (len == tp->t_maxseg)
 171                         goto send;
 172                 if ((1 || idle || tp->t_flags & TF_NODELAY) &&
 173                     len + off >= so->so_snd.sb_cc)
 174                         goto send;
 175                 if (tp->t_force)
 176                         goto send;
 177                 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
 178                         goto send;
 179                 if (SEQ_LT(tp->snd_nxt, tp->snd_max))
 180                         goto send;
 181         }
 182
 183         /*
 184          * Compare available window to amount of window
 185          * known to peer (as advertised window less
 186          * next expected input).  If the difference is at least two
 187          * max size segments, or at least 50% of the maximum possible
 188          * window, then want to send a window update to peer.
 189          */
 190         if (win > 0) {
 191                 /*
 192                  * "adv" is the amount we can increase the window,
 193                  * taking into account that we are limited by
 194                  * TCP_MAXWIN << tp->rcv_scale.
 195                  */
 196                 long adv = MIN(win, (long)TCP_MAXWIN << tp->rcv_scale) -
 197                         (tp->rcv_adv - tp->rcv_nxt);
 198
 199                 if (adv >= (long) (2 * tp->t_maxseg))
 200                         goto send;
 201                 if (2 * adv >= (long) so->so_rcv.sb_datalen)
 202                         goto send;
 203         }
 204
 205         /*
 206          * Send if we owe peer an ACK.
 207          */
 208         if (tp->t_flags & TF_ACKNOW)
 209                 goto send;
 210         if (flags & (TH_SYN|TH_RST))
 211                 goto send;
 212         if (SEQ_GT(tp->snd_up, tp->snd_una))
 213                 goto send;
 214         /*
 215          * If our state indicates that FIN should be sent
 216          * and we have not yet done so, or we're retransmitting the FIN,
 217          * then we need to send.
 218          */
 219         if (flags & TH_FIN &&
 220             ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
 221                 goto send;
 222
 223         /*
 224          * TCP window updates are not reliable, rather a polling protocol
 225          * using ``persist'' packets is used to insure receipt of window
 226          * updates.  The three ``states'' for the output side are:
 227          *      idle                    not doing retransmits or persists
 228          *      persisting              to move a small or zero window
 229          *      (re)transmitting        and thereby not persisting
 230          *
 231          * tp->t_timer[TCPT_PERSIST]
 232          *      is set when we are in persist state.
 233          * tp->t_force
 234          *      is set when we are called to send a persist packet.
 235          * tp->t_timer[TCPT_REXMT]
 236          *      is set when we are retransmitting
 237          * The output side is idle when both timers are zero.
 238          *
 239          * If send window is too small, there is data to transmit, and no
 240          * retransmit or persist is pending, then go to persist state.
 241          * If nothing happens soon, send when timer expires:
 242          * if window is nonzero, transmit what we can,
 243          * otherwise force out a byte.
 244          */
 245         if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 &&
 246             tp->t_timer[TCPT_PERSIST] == 0) {
 247                 tp->t_rxtshift = 0;
 248                 tcp_setpersist(tp);
 249         }
 250
 251         /*
 252          * No reason to send a segment, just return.
 253          */
 254         return (0);
 255
 256 send:
 257         /*
 258          * Before ESTABLISHED, force sending of initial options
 259          * unless TCP set not to do any options.
 260          * NOTE: we assume that the IP/TCP header plus TCP options
 261          * always fit in a single mbuf, leaving room for a maximum
 262          * link header, i.e.
 263          *      max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MHLEN
 264          */
 265         optlen = 0;
 266         hdrlen = sizeof (struct tcpiphdr);
 267         if (flags & TH_SYN) {
 268                 tp->snd_nxt = tp->iss;
 269                 if ((tp->t_flags & TF_NOOPT) == 0) {
 270                         uint16_t mss;
 271
 272                         opt[0] = TCPOPT_MAXSEG;
 273                         opt[1] = 4;
 274                         mss = htons((uint16_t) tcp_mss(tp, 0));
 275                         memcpy((caddr_t)(opt + 2), (caddr_t)&mss, sizeof(mss));
 276                         optlen = 4;
 277                 }
 278         }
 279
 280         hdrlen += optlen;
 281
 282         /*
 283          * Adjust data length if insertion of options will
 284          * bump the packet length beyond the t_maxseg length.
 285          */
 286          if (len > tp->t_maxseg - optlen) {
 287                 len = tp->t_maxseg - optlen;
 288                 sendalot = 1;
 289          }
 290
 291         /*
 292          * Grab a header mbuf, attaching a copy of data to
 293          * be transmitted, and initialize the header from
 294          * the template for sends on this connection.
 295          */
 296         if (len) {
 297                 m = m_get(so->slirp);
 298                 if (m == NULL) {
 299                         error = 1;
 300                         goto out;
 301                 }
 302                 m->m_data += IF_MAXLINKHDR;
 303                 m->m_len = hdrlen;
 304
 305                 sbcopy(&so->so_snd, off, (int) len, mtod(m, caddr_t) + hdrlen);
 306                 m->m_len += len;
 307
 308                 /*
 309                  * If we're sending everything we've got, set PUSH.
 310                  * (This will keep happy those implementations which only
 311                  * give data to the user when a buffer fills or
 312                  * a PUSH comes in.)
 313                  */
 314                 if (off + len == so->so_snd.sb_cc)
 315                         flags |= TH_PUSH;
 316         } else {
 317                 m = m_get(so->slirp);
 318                 if (m == NULL) {
 319                         error = 1;
 320                         goto out;
 321                 }
 322                 m->m_data += IF_MAXLINKHDR;
 323                 m->m_len = hdrlen;
 324         }
 325
 326         ti = mtod(m, struct tcpiphdr *);
 327
 328         memcpy((caddr_t)ti, &tp->t_template, sizeof (struct tcpiphdr));
 329
 330         /*
 331          * Fill in fields, remembering maximum advertised
 332          * window for use in delaying messages about window sizes.
 333          * If resending a FIN, be sure not to use a new sequence number.
 334          */
 335         if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
 336             tp->snd_nxt == tp->snd_max)
 337                 tp->snd_nxt--;
 338         /*
 339          * If we are doing retransmissions, then snd_nxt will
 340          * not reflect the first unsent octet.  For ACK only
 341          * packets, we do not want the sequence number of the
 342          * retransmitted packet, we want the sequence number
 343          * of the next unsent octet.  So, if there is no data
 344          * (and no SYN or FIN), use snd_max instead of snd_nxt
 345          * when filling in ti_seq.  But if we are in persist
 346          * state, snd_max might reflect one byte beyond the
 347          * right edge of the window, so use snd_nxt in that
 348          * case, since we know we aren't doing a retransmission.
 349          * (retransmit and persist are mutually exclusive...)
 350          */
 351         if (len || (flags & (TH_SYN|TH_FIN)) || tp->t_timer[TCPT_PERSIST])
 352                 ti->ti_seq = htonl(tp->snd_nxt);
 353         else
 354                 ti->ti_seq = htonl(tp->snd_max);
 355         ti->ti_ack = htonl(tp->rcv_nxt);
 356         if (optlen) {
 357                 memcpy((caddr_t)(ti + 1), (caddr_t)opt, optlen);
 358                 ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2;
 359         }
 360         ti->ti_flags = flags;
 361         /*
 362          * Calculate receive window.  Don't shrink window,
 363          * but avoid silly window syndrome.
 364          */
 365         if (win < (long)(so->so_rcv.sb_datalen / 4) && win < (long)tp->t_maxseg)
 366                 win = 0;
 367         if (win > (long)TCP_MAXWIN << tp->rcv_scale)
 368                 win = (long)TCP_MAXWIN << tp->rcv_scale;
 369         if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
 370                 win = (long)(tp->rcv_adv - tp->rcv_nxt);
 371         ti->ti_win = htons((uint16_t) (win>>tp->rcv_scale));
 372
 373         if (SEQ_GT(tp->snd_up, tp->snd_una)) {
 374                 ti->ti_urp = htons((uint16_t)(tp->snd_up - ntohl(ti->ti_seq)));
 375                 ti->ti_flags |= TH_URG;
 376         } else
 377                 /*
 378                  * If no urgent pointer to send, then we pull
 379                  * the urgent pointer to the left edge of the send window
 380                  * so that it doesn't drift into the send window on sequence
 381                  * number wraparound.
 382                  */
 383                 tp->snd_up = tp->snd_una;               /* drag it along */
 384
 385         /*
 386          * Put TCP length in extended header, and then
 387          * checksum extended header and data.
 388          */
 389         if (len + optlen)
 390                 ti->ti_len = htons((uint16_t)(sizeof (struct tcphdr) +
 391                     optlen + len));
 392         ti->ti_sum = cksum(m, (int)(hdrlen + len));
 393
 394         /*
 395          * In transmit state, time the transmission and arrange for
 396          * the retransmit.  In persist state, just set snd_max.
 397          */
 398         if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) {
 399                 tcp_seq startseq = tp->snd_nxt;
 400
 401                 /*
 402                  * Advance snd_nxt over sequence space of this segment.
 403                  */
 404                 if (flags & (TH_SYN|TH_FIN)) {
 405                         if (flags & TH_SYN)
 406                                 tp->snd_nxt++;
 407                         if (flags & TH_FIN) {
 408                                 tp->snd_nxt++;
 409                                 tp->t_flags |= TF_SENTFIN;
 410                         }
 411                 }
 412                 tp->snd_nxt += len;
 413                 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
 414                         tp->snd_max = tp->snd_nxt;
 415                         /*
 416                          * Time this transmission if not a retransmission and
 417                          * not currently timing anything.
 418                          */
 419                         if (tp->t_rtt == 0) {
 420                                 tp->t_rtt = 1;
 421                                 tp->t_rtseq = startseq;
 422                         }
 423                 }
 424
 425                 /*
 426                  * Set retransmit timer if not currently set,
 427                  * and not doing an ack or a keep-alive probe.
 428                  * Initial value for retransmit timer is smoothed
 429                  * round-trip time + 2 * round-trip time variance.
 430                  * Initialize shift counter which is used for backoff
 431                  * of retransmit time.
 432                  */
 433                 if (tp->t_timer[TCPT_REXMT] == 0 &&
 434                     tp->snd_nxt != tp->snd_una) {
 435                         tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
 436                         if (tp->t_timer[TCPT_PERSIST]) {
 437                                 tp->t_timer[TCPT_PERSIST] = 0;
 438                                 tp->t_rxtshift = 0;
 439                         }
 440                 }
 441         } else
 442                 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max))
 443                         tp->snd_max = tp->snd_nxt + len;
 444
 445         /*
 446          * Fill in IP length and desired time to live and
 447          * send to IP level.  There should be a better way
 448          * to handle ttl and tos; we could keep them in
 449          * the template, but need a way to checksum without them.
 450          */
 451         m->m_len = hdrlen + len; /* XXX Needed? m_len should be correct */
 452         tcpiph_save = *mtod(m, struct tcpiphdr *);
 453
 454         switch (so->so_ffamily) {
 455         case AF_INET:
 456             m->m_data += sizeof(struct tcpiphdr) - sizeof(struct tcphdr)
 457                                                  - sizeof(struct ip);
 458             m->m_len  -= sizeof(struct tcpiphdr) - sizeof(struct tcphdr)
 459                                                  - sizeof(struct ip);
 460             ip = mtod(m, struct ip *);
 461
 462             ip->ip_len = m->m_len;
 463             ip->ip_dst = tcpiph_save.ti_dst;
 464             ip->ip_src = tcpiph_save.ti_src;
 465             ip->ip_p = tcpiph_save.ti_pr;
 466
 467             ip->ip_ttl = IPDEFTTL;
 468             ip->ip_tos = so->so_iptos;
 469             error = ip_output(so, m);
 470             break;
 471
 472         case AF_INET6:
 473             m->m_data += sizeof(struct tcpiphdr) - sizeof(struct tcphdr)
 474                                                  - sizeof(struct ip6);
 475             m->m_len  -= sizeof(struct tcpiphdr) - sizeof(struct tcphdr)
 476                                                  - sizeof(struct ip6);
 477             ip6 = mtod(m, struct ip6 *);
 478
 479             ip6->ip_pl = tcpiph_save.ti_len;
 480             ip6->ip_dst = tcpiph_save.ti_dst6;
 481             ip6->ip_src = tcpiph_save.ti_src6;
 482             ip6->ip_nh = tcpiph_save.ti_nh6;
 483
 484             error = ip6_output(so, m, 0);
 485             break;
 486
 487         default:
 488             g_assert_not_reached();
 489         }
 490
 491         if (error) {
 492 out:
 493                 return (error);
 494         }
 495
 496         /*
 497          * Data sent (as far as we can tell).
 498          * If this advertises a larger window than any other segment,
 499          * then remember the size of the advertised window.
 500          * Any pending ACK has now been sent.
 501          */
 502         if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
 503                 tp->rcv_adv = tp->rcv_nxt + win;
 504         tp->last_ack_sent = tp->rcv_nxt;
 505         tp->t_flags &= ~(TF_ACKNOW|TF_DELACK);
 506         if (sendalot)
 507                 goto again;
 508
 509         return (0);
 510 }
 511
 512 void
 513 tcp_setpersist(struct tcpcb *tp)
 514 {
 515     int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
 516
 517         /*
 518          * Start/restart persistence timer.
 519          */
 520         TCPT_RANGESET(tp->t_timer[TCPT_PERSIST],
 521             t * tcp_backoff[tp->t_rxtshift],
 522             TCPTV_PERSMIN, TCPTV_PERSMAX);
 523         if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
 524                 tp->t_rxtshift++;
 525 }