Make sure to zero out memory before calling madvise to increase robustness
[qemu-kvm/fedora.git] / slirp / tcp_output.c
blob9ed50f5002664d3f9681fd0f2e69a8d17f4c6bcf
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(struct tcpcb *tp)
69 register struct socket *so = tp->t_socket;
70 register long len, win;
71 int off, flags, error;
72 register struct mbuf *m;
73 register struct tcpiphdr *ti;
74 u_char opt[MAX_TCPOPTLEN];
75 unsigned optlen, hdrlen;
76 int idle, sendalot;
78 DEBUG_CALL("tcp_output");
79 DEBUG_ARG("tp = %lx", (long )tp);
82 * Determine length of data that should be transmitted,
83 * and flags that will be used.
84 * If there is some data or critical controls (SYN, RST)
85 * to send, then transmit; otherwise, investigate further.
87 idle = (tp->snd_max == tp->snd_una);
88 if (idle && tp->t_idle >= tp->t_rxtcur)
90 * We have been idle for "a while" and no acks are
91 * expected to clock out any data we send --
92 * slow start to get ack "clock" running again.
94 tp->snd_cwnd = tp->t_maxseg;
95 again:
96 sendalot = 0;
97 off = tp->snd_nxt - tp->snd_una;
98 win = min(tp->snd_wnd, tp->snd_cwnd);
100 flags = tcp_outflags[tp->t_state];
102 DEBUG_MISC((dfd, " --- tcp_output flags = 0x%x\n",flags));
105 * If in persist timeout with window of 0, send 1 byte.
106 * Otherwise, if window is small but nonzero
107 * and timer expired, we will send what we can
108 * and go to transmit state.
110 if (tp->t_force) {
111 if (win == 0) {
113 * If we still have some data to send, then
114 * clear the FIN bit. Usually this would
115 * happen below when it realizes that we
116 * aren't sending all the data. However,
117 * if we have exactly 1 byte of unset data,
118 * then it won't clear the FIN bit below,
119 * and if we are in persist state, we wind
120 * up sending the packet without recording
121 * that we sent the FIN bit.
123 * We can't just blindly clear the FIN bit,
124 * because if we don't have any more data
125 * to send then the probe will be the FIN
126 * itself.
128 if (off < so->so_snd.sb_cc)
129 flags &= ~TH_FIN;
130 win = 1;
131 } else {
132 tp->t_timer[TCPT_PERSIST] = 0;
133 tp->t_rxtshift = 0;
137 len = min(so->so_snd.sb_cc, win) - off;
139 if (len < 0) {
141 * If FIN has been sent but not acked,
142 * but we haven't been called to retransmit,
143 * len will be -1. Otherwise, window shrank
144 * after we sent into it. If window shrank to 0,
145 * cancel pending retransmit and pull snd_nxt
146 * back to (closed) window. We will enter persist
147 * state below. If the window didn't close completely,
148 * just wait for an ACK.
150 len = 0;
151 if (win == 0) {
152 tp->t_timer[TCPT_REXMT] = 0;
153 tp->snd_nxt = tp->snd_una;
157 if (len > tp->t_maxseg) {
158 len = tp->t_maxseg;
159 sendalot = 1;
161 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
162 flags &= ~TH_FIN;
164 win = sbspace(&so->so_rcv);
167 * Sender silly window avoidance. If connection is idle
168 * and can send all data, a maximum segment,
169 * at least a maximum default-size segment do it,
170 * or are forced, do it; otherwise don't bother.
171 * If peer's buffer is tiny, then send
172 * when window is at least half open.
173 * If retransmitting (possibly after persist timer forced us
174 * to send into a small window), then must resend.
176 if (len) {
177 if (len == tp->t_maxseg)
178 goto send;
179 if ((1 || idle || tp->t_flags & TF_NODELAY) &&
180 len + off >= so->so_snd.sb_cc)
181 goto send;
182 if (tp->t_force)
183 goto send;
184 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
185 goto send;
186 if (SEQ_LT(tp->snd_nxt, tp->snd_max))
187 goto send;
191 * Compare available window to amount of window
192 * known to peer (as advertised window less
193 * next expected input). If the difference is at least two
194 * max size segments, or at least 50% of the maximum possible
195 * window, then want to send a window update to peer.
197 if (win > 0) {
199 * "adv" is the amount we can increase the window,
200 * taking into account that we are limited by
201 * TCP_MAXWIN << tp->rcv_scale.
203 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) -
204 (tp->rcv_adv - tp->rcv_nxt);
206 if (adv >= (long) (2 * tp->t_maxseg))
207 goto send;
208 if (2 * adv >= (long) so->so_rcv.sb_datalen)
209 goto send;
213 * Send if we owe peer an ACK.
215 if (tp->t_flags & TF_ACKNOW)
216 goto send;
217 if (flags & (TH_SYN|TH_RST))
218 goto send;
219 if (SEQ_GT(tp->snd_up, tp->snd_una))
220 goto send;
222 * If our state indicates that FIN should be sent
223 * and we have not yet done so, or we're retransmitting the FIN,
224 * then we need to send.
226 if (flags & TH_FIN &&
227 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
228 goto send;
231 * TCP window updates are not reliable, rather a polling protocol
232 * using ``persist'' packets is used to insure receipt of window
233 * updates. The three ``states'' for the output side are:
234 * idle not doing retransmits or persists
235 * persisting to move a small or zero window
236 * (re)transmitting and thereby not persisting
238 * tp->t_timer[TCPT_PERSIST]
239 * is set when we are in persist state.
240 * tp->t_force
241 * is set when we are called to send a persist packet.
242 * tp->t_timer[TCPT_REXMT]
243 * is set when we are retransmitting
244 * The output side is idle when both timers are zero.
246 * If send window is too small, there is data to transmit, and no
247 * retransmit or persist is pending, then go to persist state.
248 * If nothing happens soon, send when timer expires:
249 * if window is nonzero, transmit what we can,
250 * otherwise force out a byte.
252 if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 &&
253 tp->t_timer[TCPT_PERSIST] == 0) {
254 tp->t_rxtshift = 0;
255 tcp_setpersist(tp);
259 * No reason to send a segment, just return.
261 STAT(tcpstat.tcps_didnuttin++);
263 return (0);
265 send:
267 * Before ESTABLISHED, force sending of initial options
268 * unless TCP set not to do any options.
269 * NOTE: we assume that the IP/TCP header plus TCP options
270 * always fit in a single mbuf, leaving room for a maximum
271 * link header, i.e.
272 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MHLEN
274 optlen = 0;
275 hdrlen = sizeof (struct tcpiphdr);
276 if (flags & TH_SYN) {
277 tp->snd_nxt = tp->iss;
278 if ((tp->t_flags & TF_NOOPT) == 0) {
279 u_int16_t mss;
281 opt[0] = TCPOPT_MAXSEG;
282 opt[1] = 4;
283 mss = htons((u_int16_t) tcp_mss(tp, 0));
284 memcpy((caddr_t)(opt + 2), (caddr_t)&mss, sizeof(mss));
285 optlen = 4;
287 /* if ((tp->t_flags & TF_REQ_SCALE) &&
288 * ((flags & TH_ACK) == 0 ||
289 * (tp->t_flags & TF_RCVD_SCALE))) {
290 * *((u_int32_t *) (opt + optlen)) = htonl(
291 * TCPOPT_NOP << 24 |
292 * TCPOPT_WINDOW << 16 |
293 * TCPOLEN_WINDOW << 8 |
294 * tp->request_r_scale);
295 * optlen += 4;
302 * Send a timestamp and echo-reply if this is a SYN and our side
303 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side
304 * and our peer have sent timestamps in our SYN's.
306 /* if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
307 * (flags & TH_RST) == 0 &&
308 * ((flags & (TH_SYN|TH_ACK)) == TH_SYN ||
309 * (tp->t_flags & TF_RCVD_TSTMP))) {
310 * u_int32_t *lp = (u_int32_t *)(opt + optlen);
312 * / * Form timestamp option as shown in appendix A of RFC 1323. * /
313 * *lp++ = htonl(TCPOPT_TSTAMP_HDR);
314 * *lp++ = htonl(tcp_now);
315 * *lp = htonl(tp->ts_recent);
316 * optlen += TCPOLEN_TSTAMP_APPA;
319 hdrlen += optlen;
322 * Adjust data length if insertion of options will
323 * bump the packet length beyond the t_maxseg length.
325 if (len > tp->t_maxseg - optlen) {
326 len = tp->t_maxseg - optlen;
327 sendalot = 1;
331 * Grab a header mbuf, attaching a copy of data to
332 * be transmitted, and initialize the header from
333 * the template for sends on this connection.
335 if (len) {
336 if (tp->t_force && len == 1)
337 STAT(tcpstat.tcps_sndprobe++);
338 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) {
339 STAT(tcpstat.tcps_sndrexmitpack++);
340 STAT(tcpstat.tcps_sndrexmitbyte += len);
341 } else {
342 STAT(tcpstat.tcps_sndpack++);
343 STAT(tcpstat.tcps_sndbyte += len);
346 m = m_get();
347 if (m == NULL) {
348 /* error = ENOBUFS; */
349 error = 1;
350 goto out;
352 m->m_data += IF_MAXLINKHDR;
353 m->m_len = hdrlen;
356 * This will always succeed, since we make sure our mbufs
357 * are big enough to hold one MSS packet + header + ... etc.
359 /* if (len <= MHLEN - hdrlen - max_linkhdr) { */
361 sbcopy(&so->so_snd, off, (int) len, mtod(m, caddr_t) + hdrlen);
362 m->m_len += len;
364 /* } else {
365 * m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
366 * if (m->m_next == 0)
367 * len = 0;
371 * If we're sending everything we've got, set PUSH.
372 * (This will keep happy those implementations which only
373 * give data to the user when a buffer fills or
374 * a PUSH comes in.)
376 if (off + len == so->so_snd.sb_cc)
377 flags |= TH_PUSH;
378 } else {
379 if (tp->t_flags & TF_ACKNOW)
380 STAT(tcpstat.tcps_sndacks++);
381 else if (flags & (TH_SYN|TH_FIN|TH_RST))
382 STAT(tcpstat.tcps_sndctrl++);
383 else if (SEQ_GT(tp->snd_up, tp->snd_una))
384 STAT(tcpstat.tcps_sndurg++);
385 else
386 STAT(tcpstat.tcps_sndwinup++);
388 m = m_get();
389 if (m == NULL) {
390 /* error = ENOBUFS; */
391 error = 1;
392 goto out;
394 m->m_data += IF_MAXLINKHDR;
395 m->m_len = hdrlen;
398 ti = mtod(m, struct tcpiphdr *);
400 memcpy((caddr_t)ti, &tp->t_template, sizeof (struct tcpiphdr));
403 * Fill in fields, remembering maximum advertised
404 * window for use in delaying messages about window sizes.
405 * If resending a FIN, be sure not to use a new sequence number.
407 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
408 tp->snd_nxt == tp->snd_max)
409 tp->snd_nxt--;
411 * If we are doing retransmissions, then snd_nxt will
412 * not reflect the first unsent octet. For ACK only
413 * packets, we do not want the sequence number of the
414 * retransmitted packet, we want the sequence number
415 * of the next unsent octet. So, if there is no data
416 * (and no SYN or FIN), use snd_max instead of snd_nxt
417 * when filling in ti_seq. But if we are in persist
418 * state, snd_max might reflect one byte beyond the
419 * right edge of the window, so use snd_nxt in that
420 * case, since we know we aren't doing a retransmission.
421 * (retransmit and persist are mutually exclusive...)
423 if (len || (flags & (TH_SYN|TH_FIN)) || tp->t_timer[TCPT_PERSIST])
424 ti->ti_seq = htonl(tp->snd_nxt);
425 else
426 ti->ti_seq = htonl(tp->snd_max);
427 ti->ti_ack = htonl(tp->rcv_nxt);
428 if (optlen) {
429 memcpy((caddr_t)(ti + 1), (caddr_t)opt, optlen);
430 ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2;
432 ti->ti_flags = flags;
434 * Calculate receive window. Don't shrink window,
435 * but avoid silly window syndrome.
437 if (win < (long)(so->so_rcv.sb_datalen / 4) && win < (long)tp->t_maxseg)
438 win = 0;
439 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
440 win = (long)TCP_MAXWIN << tp->rcv_scale;
441 if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
442 win = (long)(tp->rcv_adv - tp->rcv_nxt);
443 ti->ti_win = htons((u_int16_t) (win>>tp->rcv_scale));
445 if (SEQ_GT(tp->snd_up, tp->snd_una)) {
446 ti->ti_urp = htons((u_int16_t)(tp->snd_up - ntohl(ti->ti_seq)));
447 #ifdef notdef
448 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
449 ti->ti_urp = htons((u_int16_t)(tp->snd_up - tp->snd_nxt));
450 #endif
451 ti->ti_flags |= TH_URG;
452 } else
454 * If no urgent pointer to send, then we pull
455 * the urgent pointer to the left edge of the send window
456 * so that it doesn't drift into the send window on sequence
457 * number wraparound.
459 tp->snd_up = tp->snd_una; /* drag it along */
462 * Put TCP length in extended header, and then
463 * checksum extended header and data.
465 if (len + optlen)
466 ti->ti_len = htons((u_int16_t)(sizeof (struct tcphdr) +
467 optlen + len));
468 ti->ti_sum = cksum(m, (int)(hdrlen + len));
471 * In transmit state, time the transmission and arrange for
472 * the retransmit. In persist state, just set snd_max.
474 if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) {
475 tcp_seq startseq = tp->snd_nxt;
478 * Advance snd_nxt over sequence space of this segment.
480 if (flags & (TH_SYN|TH_FIN)) {
481 if (flags & TH_SYN)
482 tp->snd_nxt++;
483 if (flags & TH_FIN) {
484 tp->snd_nxt++;
485 tp->t_flags |= TF_SENTFIN;
488 tp->snd_nxt += len;
489 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
490 tp->snd_max = tp->snd_nxt;
492 * Time this transmission if not a retransmission and
493 * not currently timing anything.
495 if (tp->t_rtt == 0) {
496 tp->t_rtt = 1;
497 tp->t_rtseq = startseq;
498 STAT(tcpstat.tcps_segstimed++);
503 * Set retransmit timer if not currently set,
504 * and not doing an ack or a keep-alive probe.
505 * Initial value for retransmit timer is smoothed
506 * round-trip time + 2 * round-trip time variance.
507 * Initialize shift counter which is used for backoff
508 * of retransmit time.
510 if (tp->t_timer[TCPT_REXMT] == 0 &&
511 tp->snd_nxt != tp->snd_una) {
512 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
513 if (tp->t_timer[TCPT_PERSIST]) {
514 tp->t_timer[TCPT_PERSIST] = 0;
515 tp->t_rxtshift = 0;
518 } else
519 if (SEQ_GT(tp->snd_nxt + len, tp->snd_max))
520 tp->snd_max = tp->snd_nxt + len;
523 * Fill in IP length and desired time to live and
524 * send to IP level. There should be a better way
525 * to handle ttl and tos; we could keep them in
526 * the template, but need a way to checksum without them.
528 m->m_len = hdrlen + len; /* XXX Needed? m_len should be correct */
532 ((struct ip *)ti)->ip_len = m->m_len;
534 ((struct ip *)ti)->ip_ttl = IPDEFTTL;
535 ((struct ip *)ti)->ip_tos = so->so_iptos;
537 /* #if BSD >= 43 */
538 /* Don't do IP options... */
539 /* error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
540 * so->so_options & SO_DONTROUTE, 0);
542 error = ip_output(so, m);
544 /* #else
545 * error = ip_output(m, (struct mbuf *)0, &tp->t_inpcb->inp_route,
546 * so->so_options & SO_DONTROUTE);
547 * #endif
550 if (error) {
551 out:
552 /* if (error == ENOBUFS) {
553 * tcp_quench(tp->t_inpcb, 0);
554 * return (0);
557 /* if ((error == EHOSTUNREACH || error == ENETDOWN)
558 * && TCPS_HAVERCVDSYN(tp->t_state)) {
559 * tp->t_softerror = error;
560 * return (0);
563 return (error);
565 STAT(tcpstat.tcps_sndtotal++);
568 * Data sent (as far as we can tell).
569 * If this advertises a larger window than any other segment,
570 * then remember the size of the advertised window.
571 * Any pending ACK has now been sent.
573 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
574 tp->rcv_adv = tp->rcv_nxt + win;
575 tp->last_ack_sent = tp->rcv_nxt;
576 tp->t_flags &= ~(TF_ACKNOW|TF_DELACK);
577 if (sendalot)
578 goto again;
580 return (0);
583 void
584 tcp_setpersist(struct tcpcb *tp)
586 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
588 /* if (tp->t_timer[TCPT_REXMT])
589 * panic("tcp_output REXMT");
592 * Start/restart persistence timer.
594 TCPT_RANGESET(tp->t_timer[TCPT_PERSIST],
595 t * tcp_backoff[tp->t_rxtshift],
596 TCPTV_PERSMIN, TCPTV_PERSMAX);
597 if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
598 tp->t_rxtshift++;