qemu:virtio-net: Enable filtering based on MAC, promisc, broadcast and allmulti ...
[qemu.git] / slirp / tcp_output.c
blob4a7bdbceea655ee43b00a8722a81d9e8520c81c8
1 /*
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
3 * The Regents of the University of California. All rights reserved.
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.
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.
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
34 * Changes and additions relating to SLiRP
35 * Copyright (c) 1995 Danny Gasparovski.
37 * Please read the file COPYRIGHT for the
38 * terms and conditions of the copyright.
41 #include <slirp.h>
44 * Since this is only used in "stats socket", we give meaning
45 * names instead of the REAL names
47 const char * const 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",
54 static const 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,
61 #define MAX_TCPOPTLEN 32 /* max # bytes that go in options */
64 * Tcp output routine: figure out what should be sent and send it.
66 int
67 tcp_output(tp)
68 register struct tcpcb *tp;
70 register struct socket *so = tp->t_socket;
71 register long len, win;
72 int off, flags, error;
73 register struct mbuf *m;
74 register struct tcpiphdr *ti;
75 u_char opt[MAX_TCPOPTLEN];
76 unsigned optlen, hdrlen;
77 int idle, sendalot;
79 DEBUG_CALL("tcp_output");
80 DEBUG_ARG("tp = %lx", (long )tp);
83 * Determine length of data that should be transmitted,
84 * and flags that will be used.
85 * If there is some data or critical controls (SYN, RST)
86 * to send, then transmit; otherwise, investigate further.
88 idle = (tp->snd_max == tp->snd_una);
89 if (idle && tp->t_idle >= tp->t_rxtcur)
91 * We have been idle for "a while" and no acks are
92 * expected to clock out any data we send --
93 * slow start to get ack "clock" running again.
95 tp->snd_cwnd = tp->t_maxseg;
96 again:
97 sendalot = 0;
98 off = tp->snd_nxt - tp->snd_una;
99 win = min(tp->snd_wnd, tp->snd_cwnd);
101 flags = tcp_outflags[tp->t_state];
103 DEBUG_MISC((dfd, " --- tcp_output flags = 0x%x\n",flags));
106 * If in persist timeout with window of 0, send 1 byte.
107 * Otherwise, if window is small but nonzero
108 * and timer expired, we will send what we can
109 * and go to transmit state.
111 if (tp->t_force) {
112 if (win == 0) {
114 * If we still have some data to send, then
115 * clear the FIN bit. Usually this would
116 * happen below when it realizes that we
117 * aren't sending all the data. However,
118 * if we have exactly 1 byte of unset data,
119 * then it won't clear the FIN bit below,
120 * and if we are in persist state, we wind
121 * up sending the packet without recording
122 * that we sent the FIN bit.
124 * We can't just blindly clear the FIN bit,
125 * because if we don't have any more data
126 * to send then the probe will be the FIN
127 * itself.
129 if (off < so->so_snd.sb_cc)
130 flags &= ~TH_FIN;
131 win = 1;
132 } else {
133 tp->t_timer[TCPT_PERSIST] = 0;
134 tp->t_rxtshift = 0;
138 len = min(so->so_snd.sb_cc, win) - off;
140 if (len < 0) {
142 * If FIN has been sent but not acked,
143 * but we haven't been called to retransmit,
144 * len will be -1. Otherwise, window shrank
145 * after we sent into it. If window shrank to 0,
146 * cancel pending retransmit and pull snd_nxt
147 * back to (closed) window. We will enter persist
148 * state below. If the window didn't close completely,
149 * just wait for an ACK.
151 len = 0;
152 if (win == 0) {
153 tp->t_timer[TCPT_REXMT] = 0;
154 tp->snd_nxt = tp->snd_una;
158 if (len > tp->t_maxseg) {
159 len = tp->t_maxseg;
160 sendalot = 1;
162 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
163 flags &= ~TH_FIN;
165 win = sbspace(&so->so_rcv);
168 * Sender silly window avoidance. If connection is idle
169 * and can send all data, a maximum segment,
170 * at least a maximum default-size segment do it,
171 * or are forced, do it; otherwise don't bother.
172 * If peer's buffer is tiny, then send
173 * when window is at least half open.
174 * If retransmitting (possibly after persist timer forced us
175 * to send into a small window), then must resend.
177 if (len) {
178 if (len == tp->t_maxseg)
179 goto send;
180 if ((1 || idle || tp->t_flags & TF_NODELAY) &&
181 len + off >= so->so_snd.sb_cc)
182 goto send;
183 if (tp->t_force)
184 goto send;
185 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
186 goto send;
187 if (SEQ_LT(tp->snd_nxt, tp->snd_max))
188 goto send;
192 * Compare available window to amount of window
193 * known to peer (as advertised window less
194 * next expected input). If the difference is at least two
195 * max size segments, or at least 50% of the maximum possible
196 * window, then want to send a window update to peer.
198 if (win > 0) {
200 * "adv" is the amount we can increase the window,
201 * taking into account that we are limited by
202 * TCP_MAXWIN << tp->rcv_scale.
204 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) -
205 (tp->rcv_adv - tp->rcv_nxt);
207 if (adv >= (long) (2 * tp->t_maxseg))
208 goto send;
209 if (2 * adv >= (long) so->so_rcv.sb_datalen)
210 goto send;
214 * Send if we owe peer an ACK.
216 if (tp->t_flags & TF_ACKNOW)
217 goto send;
218 if (flags & (TH_SYN|TH_RST))
219 goto send;
220 if (SEQ_GT(tp->snd_up, tp->snd_una))
221 goto send;
223 * If our state indicates that FIN should be sent
224 * and we have not yet done so, or we're retransmitting the FIN,
225 * then we need to send.
227 if (flags & TH_FIN &&
228 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
229 goto send;
232 * TCP window updates are not reliable, rather a polling protocol
233 * using ``persist'' packets is used to insure receipt of window
234 * updates. The three ``states'' for the output side are:
235 * idle not doing retransmits or persists
236 * persisting to move a small or zero window
237 * (re)transmitting and thereby not persisting
239 * tp->t_timer[TCPT_PERSIST]
240 * is set when we are in persist state.
241 * tp->t_force
242 * is set when we are called to send a persist packet.
243 * tp->t_timer[TCPT_REXMT]
244 * is set when we are retransmitting
245 * The output side is idle when both timers are zero.
247 * If send window is too small, there is data to transmit, and no
248 * retransmit or persist is pending, then go to persist state.
249 * If nothing happens soon, send when timer expires:
250 * if window is nonzero, transmit what we can,
251 * otherwise force out a byte.
253 if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 &&
254 tp->t_timer[TCPT_PERSIST] == 0) {
255 tp->t_rxtshift = 0;
256 tcp_setpersist(tp);
260 * No reason to send a segment, just return.
262 STAT(tcpstat.tcps_didnuttin++);
264 return (0);
266 send:
268 * Before ESTABLISHED, force sending of initial options
269 * unless TCP set not to do any options.
270 * NOTE: we assume that the IP/TCP header plus TCP options
271 * always fit in a single mbuf, leaving room for a maximum
272 * link header, i.e.
273 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MHLEN
275 optlen = 0;
276 hdrlen = sizeof (struct tcpiphdr);
277 if (flags & TH_SYN) {
278 tp->snd_nxt = tp->iss;
279 if ((tp->t_flags & TF_NOOPT) == 0) {
280 u_int16_t mss;
282 opt[0] = TCPOPT_MAXSEG;
283 opt[1] = 4;
284 mss = htons((u_int16_t) tcp_mss(tp, 0));
285 memcpy((caddr_t)(opt + 2), (caddr_t)&mss, sizeof(mss));
286 optlen = 4;
288 /* if ((tp->t_flags & TF_REQ_SCALE) &&
289 * ((flags & TH_ACK) == 0 ||
290 * (tp->t_flags & TF_RCVD_SCALE))) {
291 * *((u_int32_t *) (opt + optlen)) = htonl(
292 * TCPOPT_NOP << 24 |
293 * TCPOPT_WINDOW << 16 |
294 * TCPOLEN_WINDOW << 8 |
295 * tp->request_r_scale);
296 * optlen += 4;
303 * Send a timestamp and echo-reply if this is a SYN and our side
304 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side
305 * and our peer have sent timestamps in our SYN's.
307 /* if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
308 * (flags & TH_RST) == 0 &&
309 * ((flags & (TH_SYN|TH_ACK)) == TH_SYN ||
310 * (tp->t_flags & TF_RCVD_TSTMP))) {
311 * u_int32_t *lp = (u_int32_t *)(opt + optlen);
313 * / * Form timestamp option as shown in appendix A of RFC 1323. * /
314 * *lp++ = htonl(TCPOPT_TSTAMP_HDR);
315 * *lp++ = htonl(tcp_now);
316 * *lp = htonl(tp->ts_recent);
317 * optlen += TCPOLEN_TSTAMP_APPA;
320 hdrlen += optlen;
323 * Adjust data length if insertion of options will
324 * bump the packet length beyond the t_maxseg length.
326 if (len > tp->t_maxseg - optlen) {
327 len = tp->t_maxseg - optlen;
328 sendalot = 1;
332 * Grab a header mbuf, attaching a copy of data to
333 * be transmitted, and initialize the header from
334 * the template for sends on this connection.
336 if (len) {
337 if (tp->t_force && len == 1)
338 STAT(tcpstat.tcps_sndprobe++);
339 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) {
340 STAT(tcpstat.tcps_sndrexmitpack++);
341 STAT(tcpstat.tcps_sndrexmitbyte += len);
342 } else {
343 STAT(tcpstat.tcps_sndpack++);
344 STAT(tcpstat.tcps_sndbyte += len);
347 m = m_get();
348 if (m == NULL) {
349 /* error = ENOBUFS; */
350 error = 1;
351 goto out;
353 m->m_data += IF_MAXLINKHDR;
354 m->m_len = hdrlen;
357 * This will always succeed, since we make sure our mbufs
358 * are big enough to hold one MSS packet + header + ... etc.
360 /* if (len <= MHLEN - hdrlen - max_linkhdr) { */
362 sbcopy(&so->so_snd, off, (int) len, mtod(m, caddr_t) + hdrlen);
363 m->m_len += len;
365 /* } else {
366 * m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
367 * if (m->m_next == 0)
368 * len = 0;
372 * If we're sending everything we've got, set PUSH.
373 * (This will keep happy those implementations which only
374 * give data to the user when a buffer fills or
375 * a PUSH comes in.)
377 if (off + len == so->so_snd.sb_cc)
378 flags |= TH_PUSH;
379 } else {
380 if (tp->t_flags & TF_ACKNOW)
381 STAT(tcpstat.tcps_sndacks++);
382 else if (flags & (TH_SYN|TH_FIN|TH_RST))
383 STAT(tcpstat.tcps_sndctrl++);
384 else if (SEQ_GT(tp->snd_up, tp->snd_una))
385 STAT(tcpstat.tcps_sndurg++);
386 else
387 STAT(tcpstat.tcps_sndwinup++);
389 m = m_get();
390 if (m == NULL) {
391 /* error = ENOBUFS; */
392 error = 1;
393 goto out;
395 m->m_data += IF_MAXLINKHDR;
396 m->m_len = hdrlen;
399 ti = mtod(m, struct tcpiphdr *);
401 memcpy((caddr_t)ti, &tp->t_template, sizeof (struct tcpiphdr));
404 * Fill in fields, remembering maximum advertised
405 * window for use in delaying messages about window sizes.
406 * If resending a FIN, be sure not to use a new sequence number.
408 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
409 tp->snd_nxt == tp->snd_max)
410 tp->snd_nxt--;
412 * If we are doing retransmissions, then snd_nxt will
413 * not reflect the first unsent octet. For ACK only
414 * packets, we do not want the sequence number of the
415 * retransmitted packet, we want the sequence number
416 * of the next unsent octet. So, if there is no data
417 * (and no SYN or FIN), use snd_max instead of snd_nxt
418 * when filling in ti_seq. But if we are in persist
419 * state, snd_max might reflect one byte beyond the
420 * right edge of the window, so use snd_nxt in that
421 * case, since we know we aren't doing a retransmission.
422 * (retransmit and persist are mutually exclusive...)
424 if (len || (flags & (TH_SYN|TH_FIN)) || tp->t_timer[TCPT_PERSIST])
425 ti->ti_seq = htonl(tp->snd_nxt);
426 else
427 ti->ti_seq = htonl(tp->snd_max);
428 ti->ti_ack = htonl(tp->rcv_nxt);
429 if (optlen) {
430 memcpy((caddr_t)(ti + 1), (caddr_t)opt, optlen);
431 ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2;
433 ti->ti_flags = flags;
435 * Calculate receive window. Don't shrink window,
436 * but avoid silly window syndrome.
438 if (win < (long)(so->so_rcv.sb_datalen / 4) && win < (long)tp->t_maxseg)
439 win = 0;
440 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
441 win = (long)TCP_MAXWIN << tp->rcv_scale;
442 if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
443 win = (long)(tp->rcv_adv - tp->rcv_nxt);
444 ti->ti_win = htons((u_int16_t) (win>>tp->rcv_scale));
446 if (SEQ_GT(tp->snd_up, tp->snd_una)) {
447 ti->ti_urp = htons((u_int16_t)(tp->snd_up - ntohl(ti->ti_seq)));
448 #ifdef notdef
449 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
450 ti->ti_urp = htons((u_int16_t)(tp->snd_up - tp->snd_nxt));
451 #endif
452 ti->ti_flags |= TH_URG;
453 } else
455 * If no urgent pointer to send, then we pull
456 * the urgent pointer to the left edge of the send window
457 * so that it doesn't drift into the send window on sequence
458 * number wraparound.
460 tp->snd_up = tp->snd_una; /* drag it along */
463 * Put TCP length in extended header, and then
464 * checksum extended header and data.
466 if (len + optlen)
467 ti->ti_len = htons((u_int16_t)(sizeof (struct tcphdr) +
468 optlen + len));
469 ti->ti_sum = cksum(m, (int)(hdrlen + len));
472 * In transmit state, time the transmission and arrange for
473 * the retransmit. In persist state, just set snd_max.
475 if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) {
476 tcp_seq startseq = tp->snd_nxt;
479 * Advance snd_nxt over sequence space of this segment.
481 if (flags & (TH_SYN|TH_FIN)) {
482 if (flags & TH_SYN)
483 tp->snd_nxt++;
484 if (flags & TH_FIN) {
485 tp->snd_nxt++;
486 tp->t_flags |= TF_SENTFIN;
489 tp->snd_nxt += len;
490 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
491 tp->snd_max = tp->snd_nxt;
493 * Time this transmission if not a retransmission and
494 * not currently timing anything.
496 if (tp->t_rtt == 0) {
497 tp->t_rtt = 1;
498 tp->t_rtseq = startseq;
499 STAT(tcpstat.tcps_segstimed++);
504 * Set retransmit timer if not currently set,
505 * and not doing an ack or a keep-alive probe.
506 * Initial value for retransmit timer is smoothed
507 * round-trip time + 2 * round-trip time variance.
508 * Initialize shift counter which is used for backoff
509 * of retransmit time.
511 if (tp->t_timer[TCPT_REXMT] == 0 &&
512 tp->snd_nxt != tp->snd_una) {
513 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
514 if (tp->t_timer[TCPT_PERSIST]) {
515 tp->t_timer[TCPT_PERSIST] = 0;
516 tp->t_rxtshift = 0;
519 } else
520 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max))
521 tp->snd_max = tp->snd_nxt + len;
524 * Fill in IP length and desired time to live and
525 * send to IP level. There should be a better way
526 * to handle ttl and tos; we could keep them in
527 * the template, but need a way to checksum without them.
529 m->m_len = hdrlen + len; /* XXX Needed? m_len should be correct */
533 ((struct ip *)ti)->ip_len = m->m_len;
535 ((struct ip *)ti)->ip_ttl = IPDEFTTL;
536 ((struct ip *)ti)->ip_tos = so->so_iptos;
538 /* #if BSD >= 43 */
539 /* Don't do IP options... */
540 /* error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
541 * so->so_options & SO_DONTROUTE, 0);
543 error = ip_output(so, m);
545 /* #else
546 * error = ip_output(m, (struct mbuf *)0, &tp->t_inpcb->inp_route,
547 * so->so_options & SO_DONTROUTE);
548 * #endif
551 if (error) {
552 out:
553 /* if (error == ENOBUFS) {
554 * tcp_quench(tp->t_inpcb, 0);
555 * return (0);
558 /* if ((error == EHOSTUNREACH || error == ENETDOWN)
559 * && TCPS_HAVERCVDSYN(tp->t_state)) {
560 * tp->t_softerror = error;
561 * return (0);
564 return (error);
566 STAT(tcpstat.tcps_sndtotal++);
569 * Data sent (as far as we can tell).
570 * If this advertises a larger window than any other segment,
571 * then remember the size of the advertised window.
572 * Any pending ACK has now been sent.
574 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
575 tp->rcv_adv = tp->rcv_nxt + win;
576 tp->last_ack_sent = tp->rcv_nxt;
577 tp->t_flags &= ~(TF_ACKNOW|TF_DELACK);
578 if (sendalot)
579 goto again;
581 return (0);
584 void
585 tcp_setpersist(tp)
586 register struct tcpcb *tp;
588 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
590 /* if (tp->t_timer[TCPT_REXMT])
591 * panic("tcp_output REXMT");
594 * Start/restart persistence timer.
596 TCPT_RANGESET(tp->t_timer[TCPT_PERSIST],
597 t * tcp_backoff[tp->t_rxtshift],
598 TCPTV_PERSMIN, TCPTV_PERSMAX);
599 if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
600 tp->t_rxtshift++;