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 /*
44 * Since this is only used in "stats socket", we give meaning
45 * names instead of the REAL names
46 */
47 const char *tcpstates[] = {
48 /* "CLOSED", "LISTEN", "SYN_SENT", "SYN_RCVD", */
49 "REDIRECT", "LISTEN", "SYN_SENT", "SYN_RCVD",
50 "ESTABLISHED", "CLOSE_WAIT", "FIN_WAIT_1", "CLOSING",
51 "LAST_ACK", "FIN_WAIT_2", "TIME_WAIT",
52 };
53
54 u_char tcp_outflags[TCP_NSTATES] = {
55 TH_RST|TH_ACK, 0, TH_SYN, TH_SYN|TH_ACK,
56 TH_ACK, TH_ACK, TH_FIN|TH_ACK, TH_FIN|TH_ACK,
57 TH_FIN|TH_ACK, TH_ACK, TH_ACK,
58 };
59
60
61 #define MAX_TCPOPTLEN 32 /* max # bytes that go in options */
62
63 /*
64 * Tcp output routine: figure out what should be sent and send it.
65 */
tcp_output(struct tcpcb * tp)66 int tcp_output(struct tcpcb *tp)
67 {
68 struct socket *so = tp->t_socket;
69 long len, win;
70 int off, flags, error;
71 struct mbuf *m;
72 struct tcpiphdr *ti;
73 u_char opt[MAX_TCPOPTLEN];
74 unsigned optlen, hdrlen;
75 int idle, sendalot;
76
77 DEBUG_CALL("tcp_output");
78 DEBUG_ARG("tp = %p", tp);
79
80 /*
81 * Determine length of data that should be transmitted,
82 * and flags that will be used.
83 * If there is some data or critical controls (SYN, RST)
84 * to send, then transmit; otherwise, investigate further.
85 */
86 idle = (tp->snd_max == tp->snd_una);
87 if (idle && tp->t_idle >= tp->t_rxtcur)
88 /*
89 * We have been idle for "a while" and no acks are
90 * expected to clock out any data we send --
91 * slow start to get ack "clock" running again.
92 */
93 tp->snd_cwnd = tp->t_maxseg;
94 again:
95 sendalot = 0;
96 off = tp->snd_nxt - tp->snd_una;
97 win = min(tp->snd_wnd, tp->snd_cwnd);
98
99 flags = tcp_outflags[tp->t_state];
100
101 DEBUG_MISC((" --- tcp_output flags = 0x%x\n",flags));
102
103 /*
104 * If in persist timeout with window of 0, send 1 byte.
105 * Otherwise, if window is small but nonzero
106 * and timer expired, we will send what we can
107 * and go to transmit state.
108 */
109 if (tp->t_force) {
110 if (win == 0) {
111 /*
112 * If we still have some data to send, then
113 * clear the FIN bit. Usually this would
114 * happen below when it realizes that we
115 * aren't sending all the data. However,
116 * if we have exactly 1 byte of unset data,
117 * then it won't clear the FIN bit below,
118 * and if we are in persist state, we wind
119 * up sending the packet without recording
120 * that we sent the FIN bit.
121 *
122 * We can't just blindly clear the FIN bit,
123 * because if we don't have any more data
124 * to send then the probe will be the FIN
125 * itself.
126 */
127 if (off < so->so_snd.sb_cc)
128 flags &= ~TH_FIN;
129 win = 1;
130 } else {
131 tp->t_timer[TCPT_PERSIST] = 0;
132 tp->t_rxtshift = 0;
133 }
134 }
135
136 len = min(so->so_snd.sb_cc, win) - off;
137
138 if (len < 0) {
139 /*
140 * If FIN has been sent but not acked,
141 * but we haven't been called to retransmit,
142 * len will be -1. Otherwise, window shrank
143 * after we sent into it. If window shrank to 0,
144 * cancel pending retransmit and pull snd_nxt
145 * back to (closed) window. We will enter persist
146 * state below. If the window didn't close completely,
147 * just wait for an ACK.
148 */
149 len = 0;
150 if (win == 0) {
151 tp->t_timer[TCPT_REXMT] = 0;
152 tp->snd_nxt = tp->snd_una;
153 }
154 }
155
156 if (len > tp->t_maxseg) {
157 len = tp->t_maxseg;
158 sendalot = 1;
159 }
160 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
161 flags &= ~TH_FIN;
162
163 win = sbspace(&so->so_rcv);
164
165 /*
166 * Sender silly window avoidance. If connection is idle
167 * and can send all data, a maximum segment,
168 * at least a maximum default-size segment do it,
169 * or are forced, do it; otherwise don't bother.
170 * If peer's buffer is tiny, then send
171 * when window is at least half open.
172 * If retransmitting (possibly after persist timer forced us
173 * to send into a small window), then must resend.
174 */
175 if (len) {
176 if (len == tp->t_maxseg)
177 goto send;
178 if ((1 || idle || tp->t_flags & TF_NODELAY) &&
179 len + off >= so->so_snd.sb_cc)
180 goto send;
181 if (tp->t_force)
182 goto send;
183 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
184 goto send;
185 if (SEQ_LT(tp->snd_nxt, tp->snd_max))
186 goto send;
187 }
188
189 /*
190 * Compare available window to amount of window
191 * known to peer (as advertised window less
192 * next expected input). If the difference is at least two
193 * max size segments, or at least 50% of the maximum possible
194 * window, then want to send a window update to peer.
195 */
196 if (win > 0) {
197 /*
198 * "adv" is the amount we can increase the window,
199 * taking into account that we are limited by
200 * TCP_MAXWIN << tp->rcv_scale.
201 */
202 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) -
203 (tp->rcv_adv - tp->rcv_nxt);
204
205 if (adv >= (long) (2 * tp->t_maxseg))
206 goto send;
207 if (2 * adv >= (long) so->so_rcv.sb_datalen)
208 goto send;
209 }
210
211 /*
212 * Send if we owe peer an ACK.
213 */
214 if (tp->t_flags & TF_ACKNOW)
215 goto send;
216 if (flags & (TH_SYN|TH_RST))
217 goto send;
218 if (SEQ_GT(tp->snd_up, tp->snd_una))
219 goto send;
220 /*
221 * If our state indicates that FIN should be sent
222 * and we have not yet done so, or we're retransmitting the FIN,
223 * then we need to send.
224 */
225 if (flags & TH_FIN &&
226 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
227 goto send;
228
229 /*
230 * TCP window updates are not reliable, rather a polling protocol
231 * using ``persist'' packets is used to insure receipt of window
232 * updates. The three ``states'' for the output side are:
233 * idle not doing retransmits or persists
234 * persisting to move a small or zero window
235 * (re)transmitting and thereby not persisting
236 *
237 * tp->t_timer[TCPT_PERSIST]
238 * is set when we are in persist state.
239 * tp->t_force
240 * is set when we are called to send a persist packet.
241 * tp->t_timer[TCPT_REXMT]
242 * is set when we are retransmitting
243 * The output side is idle when both timers are zero.
244 *
245 * If send window is too small, there is data to transmit, and no
246 * retransmit or persist is pending, then go to persist state.
247 * If nothing happens soon, send when timer expires:
248 * if window is nonzero, transmit what we can,
249 * otherwise force out a byte.
250 */
251 if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 &&
252 tp->t_timer[TCPT_PERSIST] == 0) {
253 tp->t_rxtshift = 0;
254 tcp_setpersist(tp);
255 }
256
257 /*
258 * No reason to send a segment, just return.
259 */
260 tcpstat.tcps_didnuttin++;
261
262 return (0);
263
264 send:
265 /*
266 * Before ESTABLISHED, force sending of initial options
267 * unless TCP set not to do any options.
268 * NOTE: we assume that the IP/TCP header plus TCP options
269 * always fit in a single mbuf, leaving room for a maximum
270 * link header, i.e.
271 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MHLEN
272 */
273 optlen = 0;
274 hdrlen = sizeof (struct tcpiphdr);
275 if (flags & TH_SYN) {
276 tp->snd_nxt = tp->iss;
277 if ((tp->t_flags & TF_NOOPT) == 0) {
278 u_int16_t mss;
279
280 opt[0] = TCPOPT_MAXSEG;
281 opt[1] = 4;
282 mss = htons((u_int16_t) tcp_mss(tp, 0));
283 memcpy((caddr_t)(opt + 2), (caddr_t)&mss, sizeof(mss));
284 optlen = 4;
285
286 /* if ((tp->t_flags & TF_REQ_SCALE) &&
287 * ((flags & TH_ACK) == 0 ||
288 * (tp->t_flags & TF_RCVD_SCALE))) {
289 * *((u_int32_t *) (opt + optlen)) = htonl(
290 * TCPOPT_NOP << 24 |
291 * TCPOPT_WINDOW << 16 |
292 * TCPOLEN_WINDOW << 8 |
293 * tp->request_r_scale);
294 * optlen += 4;
295 * }
296 */
297 }
298 }
299
300 /*
301 * Send a timestamp and echo-reply if this is a SYN and our side
302 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side
303 * and our peer have sent timestamps in our SYN's.
304 */
305 /* if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
306 * (flags & TH_RST) == 0 &&
307 * ((flags & (TH_SYN|TH_ACK)) == TH_SYN ||
308 * (tp->t_flags & TF_RCVD_TSTMP))) {
309 * u_int32_t *lp = (u_int32_t *)(opt + optlen);
310 *
311 * / * Form timestamp option as shown in appendix A of RFC 1323. * /
312 * *lp++ = htonl(TCPOPT_TSTAMP_HDR);
313 * *lp++ = htonl(tcp_now);
314 * *lp = htonl(tp->ts_recent);
315 * optlen += TCPOLEN_TSTAMP_APPA;
316 * }
317 */
318 hdrlen += optlen;
319
320 /*
321 * Adjust data length if insertion of options will
322 * bump the packet length beyond the t_maxseg length.
323 */
324 if (len > tp->t_maxseg - optlen) {
325 len = tp->t_maxseg - optlen;
326 sendalot = 1;
327 }
328
329 /*
330 * Grab a header mbuf, attaching a copy of data to
331 * be transmitted, and initialize the header from
332 * the template for sends on this connection.
333 */
334 if (len) {
335 if (tp->t_force && len == 1)
336 tcpstat.tcps_sndprobe++;
337 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) {
338 tcpstat.tcps_sndrexmitpack++;
339 tcpstat.tcps_sndrexmitbyte += len;
340 } else {
341 tcpstat.tcps_sndpack++;
342 tcpstat.tcps_sndbyte += len;
343 }
344
345 m = m_get();
346 if (m == NULL) {
347 /* error = ENOBUFS; */
348 error = 1;
349 goto out;
350 }
351 m->m_data += if_maxlinkhdr;
352 m->m_len = hdrlen;
353
354 /*
355 * This will always succeed, since we make sure our mbufs
356 * are big enough to hold one MSS packet + header + ... etc.
357 */
358 /* if (len <= MHLEN - hdrlen - max_linkhdr) { */
359
360 sbcopy(&so->so_snd, off, (int) len, mtod(m, caddr_t) + hdrlen);
361 m->m_len += len;
362
363 /* } else {
364 * m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
365 * if (m->m_next == 0)
366 * len = 0;
367 * }
368 */
369 /*
370 * If we're sending everything we've got, set PUSH.
371 * (This will keep happy those implementations which only
372 * give data to the user when a buffer fills or
373 * a PUSH comes in.)
374 */
375 if (off + len == so->so_snd.sb_cc)
376 flags |= TH_PUSH;
377 } else {
378 if (tp->t_flags & TF_ACKNOW)
379 tcpstat.tcps_sndacks++;
380 else if (flags & (TH_SYN|TH_FIN|TH_RST))
381 tcpstat.tcps_sndctrl++;
382 else if (SEQ_GT(tp->snd_up, tp->snd_una))
383 tcpstat.tcps_sndurg++;
384 else
385 tcpstat.tcps_sndwinup++;
386
387 m = m_get();
388 if (m == NULL) {
389 /* error = ENOBUFS; */
390 error = 1;
391 goto out;
392 }
393 m->m_data += if_maxlinkhdr;
394 m->m_len = hdrlen;
395 }
396
397 ti = mtod(m, struct tcpiphdr *);
398
399 memcpy((caddr_t)ti, &tp->t_template, sizeof (struct tcpiphdr));
400
401 /*
402 * Fill in fields, remembering maximum advertised
403 * window for use in delaying messages about window sizes.
404 * If resending a FIN, be sure not to use a new sequence number.
405 */
406 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
407 tp->snd_nxt == tp->snd_max)
408 tp->snd_nxt--;
409 /*
410 * If we are doing retransmissions, then snd_nxt will
411 * not reflect the first unsent octet. For ACK only
412 * packets, we do not want the sequence number of the
413 * retransmitted packet, we want the sequence number
414 * of the next unsent octet. So, if there is no data
415 * (and no SYN or FIN), use snd_max instead of snd_nxt
416 * when filling in ti_seq. But if we are in persist
417 * state, snd_max might reflect one byte beyond the
418 * right edge of the window, so use snd_nxt in that
419 * case, since we know we aren't doing a retransmission.
420 * (retransmit and persist are mutually exclusive...)
421 */
422 if (len || (flags & (TH_SYN|TH_FIN)) || tp->t_timer[TCPT_PERSIST])
423 ti->ti_seq = htonl(tp->snd_nxt);
424 else
425 ti->ti_seq = htonl(tp->snd_max);
426 ti->ti_ack = htonl(tp->rcv_nxt);
427 if (optlen) {
428 memcpy((caddr_t)(ti + 1), (caddr_t)opt, optlen);
429 ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2;
430 }
431 ti->ti_flags = flags;
432 /*
433 * Calculate receive window. Don't shrink window,
434 * but avoid silly window syndrome.
435 */
436 if (win < (long)(so->so_rcv.sb_datalen / 4) && win < (long)tp->t_maxseg)
437 win = 0;
438 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
439 win = (long)TCP_MAXWIN << tp->rcv_scale;
440 if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
441 win = (long)(tp->rcv_adv - tp->rcv_nxt);
442 ti->ti_win = htons((u_int16_t) (win>>tp->rcv_scale));
443
444 if (SEQ_GT(tp->snd_up, tp->snd_una)) {
445 ti->ti_urp = htons((u_int16_t)(tp->snd_up - ntohl(ti->ti_seq)));
446 #ifdef notdef
447 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
448 ti->ti_urp = htons((u_int16_t)(tp->snd_up - tp->snd_nxt));
449 #endif
450 ti->ti_flags |= TH_URG;
451 } else
452 /*
453 * If no urgent pointer to send, then we pull
454 * the urgent pointer to the left edge of the send window
455 * so that it doesn't drift into the send window on sequence
456 * number wraparound.
457 */
458 tp->snd_up = tp->snd_una; /* drag it along */
459
460 /*
461 * Put TCP length in extended header, and then
462 * checksum extended header and data.
463 */
464 if (len + optlen)
465 ti->ti_len = htons((u_int16_t)(sizeof (struct tcphdr) +
466 optlen + len));
467 ti->ti_sum = cksum(m, (int)(hdrlen + len));
468
469 /*
470 * In transmit state, time the transmission and arrange for
471 * the retransmit. In persist state, just set snd_max.
472 */
473 if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) {
474 tcp_seq startseq = tp->snd_nxt;
475
476 /*
477 * Advance snd_nxt over sequence space of this segment.
478 */
479 if (flags & (TH_SYN|TH_FIN)) {
480 if (flags & TH_SYN)
481 tp->snd_nxt++;
482 if (flags & TH_FIN) {
483 tp->snd_nxt++;
484 tp->t_flags |= TF_SENTFIN;
485 }
486 }
487 tp->snd_nxt += len;
488 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
489 tp->snd_max = tp->snd_nxt;
490 /*
491 * Time this transmission if not a retransmission and
492 * not currently timing anything.
493 */
494 if (tp->t_rtt == 0) {
495 tp->t_rtt = 1;
496 tp->t_rtseq = startseq;
497 tcpstat.tcps_segstimed++;
498 }
499 }
500
501 /*
502 * Set retransmit timer if not currently set,
503 * and not doing an ack or a keep-alive probe.
504 * Initial value for retransmit timer is smoothed
505 * round-trip time + 2 * round-trip time variance.
506 * Initialize shift counter which is used for backoff
507 * of retransmit time.
508 */
509 if (tp->t_timer[TCPT_REXMT] == 0 &&
510 tp->snd_nxt != tp->snd_una) {
511 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
512 if (tp->t_timer[TCPT_PERSIST]) {
513 tp->t_timer[TCPT_PERSIST] = 0;
514 tp->t_rxtshift = 0;
515 }
516 }
517 } else
518 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max))
519 tp->snd_max = tp->snd_nxt + len;
520
521 /*
522 * Fill in IP length and desired time to live and
523 * send to IP level. There should be a better way
524 * to handle ttl and tos; we could keep them in
525 * the template, but need a way to checksum without them.
526 */
527 m->m_len = hdrlen + len; /* XXX Needed? m_len should be correct */
528
529 {
530
531 ((struct ip *)ti)->ip_len = m->m_len;
532
533 ((struct ip *)ti)->ip_ttl = ip_defttl;
534 ((struct ip *)ti)->ip_tos = so->so_iptos;
535
536 /* #if BSD >= 43 */
537 /* Don't do IP options... */
538 /* error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
539 * so->so_options & SO_DONTROUTE, 0);
540 */
541 error = ip_output(so, m);
542
543 /* #else
544 * error = ip_output(m, (struct mbuf *)0, &tp->t_inpcb->inp_route,
545 * so->so_options & SO_DONTROUTE);
546 * #endif
547 */
548 }
549 if (error) {
550 out:
551 /* if (error == ENOBUFS) {
552 * tcp_quench(tp->t_inpcb, 0);
553 * return (0);
554 * }
555 */
556 /* if ((error == EHOSTUNREACH || error == ENETDOWN)
557 * && TCPS_HAVERCVDSYN(tp->t_state)) {
558 * tp->t_softerror = error;
559 * return (0);
560 * }
561 */
562 return (error);
563 }
564 tcpstat.tcps_sndtotal++;
565
566 /*
567 * Data sent (as far as we can tell).
568 * If this advertises a larger window than any other segment,
569 * then remember the size of the advertised window.
570 * Any pending ACK has now been sent.
571 */
572 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
573 tp->rcv_adv = tp->rcv_nxt + win;
574 tp->last_ack_sent = tp->rcv_nxt;
575 tp->t_flags &= ~(TF_ACKNOW|TF_DELACK);
576 if (sendalot)
577 goto again;
578
579 return (0);
580 }
581
582 void tcp_setpersist(struct tcpcb *tp)
583 {
584 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
585
586 /* if (tp->t_timer[TCPT_REXMT])
587 * panic("tcp_output REXMT");
588 */
589 /*
590 * Start/restart persistence timer.
591 */
592 TCPT_RANGESET(tp->t_timer[TCPT_PERSIST],
593 t * tcp_backoff[tp->t_rxtshift],
594 TCPTV_PERSMIN, TCPTV_PERSMAX);
595 if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
596 tp->t_rxtshift++;
597 }
598