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