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
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
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 "qemu/osdep.h"
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
,
52 #define MAX_TCPOPTLEN 32 /* max # bytes that go in options */
55 * Tcp output routine: figure out what should be sent and send it.
58 tcp_output(struct tcpcb
*tp
)
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
;
65 u_char opt
[MAX_TCPOPTLEN
];
66 unsigned optlen
, hdrlen
;
69 DEBUG_CALL("tcp_output");
70 DEBUG_ARG("tp = %p", tp
);
73 * Determine length of data that should be transmitted,
74 * and flags that will be used.
75 * If there is some data or critical controls (SYN, RST)
76 * to send, then transmit; otherwise, investigate further.
78 idle
= (tp
->snd_max
== tp
->snd_una
);
79 if (idle
&& tp
->t_idle
>= tp
->t_rxtcur
)
81 * We have been idle for "a while" and no acks are
82 * expected to clock out any data we send --
83 * slow start to get ack "clock" running again.
85 tp
->snd_cwnd
= tp
->t_maxseg
;
88 off
= tp
->snd_nxt
- tp
->snd_una
;
89 win
= min(tp
->snd_wnd
, tp
->snd_cwnd
);
91 flags
= tcp_outflags
[tp
->t_state
];
93 DEBUG_MISC((dfd
, " --- tcp_output flags = 0x%x\n",flags
));
96 * If in persist timeout with window of 0, send 1 byte.
97 * Otherwise, if window is small but nonzero
98 * and timer expired, we will send what we can
99 * and go to transmit state.
104 * If we still have some data to send, then
105 * clear the FIN bit. Usually this would
106 * happen below when it realizes that we
107 * aren't sending all the data. However,
108 * if we have exactly 1 byte of unset data,
109 * then it won't clear the FIN bit below,
110 * and if we are in persist state, we wind
111 * up sending the packet without recording
112 * that we sent the FIN bit.
114 * We can't just blindly clear the FIN bit,
115 * because if we don't have any more data
116 * to send then the probe will be the FIN
119 if (off
< so
->so_snd
.sb_cc
)
123 tp
->t_timer
[TCPT_PERSIST
] = 0;
128 len
= min(so
->so_snd
.sb_cc
, win
) - off
;
132 * If FIN has been sent but not acked,
133 * but we haven't been called to retransmit,
134 * len will be -1. Otherwise, window shrank
135 * after we sent into it. If window shrank to 0,
136 * cancel pending retransmit and pull snd_nxt
137 * back to (closed) window. We will enter persist
138 * state below. If the window didn't close completely,
139 * just wait for an ACK.
143 tp
->t_timer
[TCPT_REXMT
] = 0;
144 tp
->snd_nxt
= tp
->snd_una
;
148 if (len
> tp
->t_maxseg
) {
152 if (SEQ_LT(tp
->snd_nxt
+ len
, tp
->snd_una
+ so
->so_snd
.sb_cc
))
155 win
= sbspace(&so
->so_rcv
);
158 * Sender silly window avoidance. If connection is idle
159 * and can send all data, a maximum segment,
160 * at least a maximum default-size segment do it,
161 * or are forced, do it; otherwise don't bother.
162 * If peer's buffer is tiny, then send
163 * when window is at least half open.
164 * If retransmitting (possibly after persist timer forced us
165 * to send into a small window), then must resend.
168 if (len
== tp
->t_maxseg
)
170 if ((1 || idle
|| tp
->t_flags
& TF_NODELAY
) &&
171 len
+ off
>= so
->so_snd
.sb_cc
)
175 if (len
>= tp
->max_sndwnd
/ 2 && tp
->max_sndwnd
> 0)
177 if (SEQ_LT(tp
->snd_nxt
, tp
->snd_max
))
182 * Compare available window to amount of window
183 * known to peer (as advertised window less
184 * next expected input). If the difference is at least two
185 * max size segments, or at least 50% of the maximum possible
186 * window, then want to send a window update to peer.
190 * "adv" is the amount we can increase the window,
191 * taking into account that we are limited by
192 * TCP_MAXWIN << tp->rcv_scale.
194 long adv
= min(win
, (long)TCP_MAXWIN
<< tp
->rcv_scale
) -
195 (tp
->rcv_adv
- tp
->rcv_nxt
);
197 if (adv
>= (long) (2 * tp
->t_maxseg
))
199 if (2 * adv
>= (long) so
->so_rcv
.sb_datalen
)
204 * Send if we owe peer an ACK.
206 if (tp
->t_flags
& TF_ACKNOW
)
208 if (flags
& (TH_SYN
|TH_RST
))
210 if (SEQ_GT(tp
->snd_up
, tp
->snd_una
))
213 * If our state indicates that FIN should be sent
214 * and we have not yet done so, or we're retransmitting the FIN,
215 * then we need to send.
217 if (flags
& TH_FIN
&&
218 ((tp
->t_flags
& TF_SENTFIN
) == 0 || tp
->snd_nxt
== tp
->snd_una
))
222 * TCP window updates are not reliable, rather a polling protocol
223 * using ``persist'' packets is used to insure receipt of window
224 * updates. The three ``states'' for the output side are:
225 * idle not doing retransmits or persists
226 * persisting to move a small or zero window
227 * (re)transmitting and thereby not persisting
229 * tp->t_timer[TCPT_PERSIST]
230 * is set when we are in persist state.
232 * is set when we are called to send a persist packet.
233 * tp->t_timer[TCPT_REXMT]
234 * is set when we are retransmitting
235 * The output side is idle when both timers are zero.
237 * If send window is too small, there is data to transmit, and no
238 * retransmit or persist is pending, then go to persist state.
239 * If nothing happens soon, send when timer expires:
240 * if window is nonzero, transmit what we can,
241 * otherwise force out a byte.
243 if (so
->so_snd
.sb_cc
&& tp
->t_timer
[TCPT_REXMT
] == 0 &&
244 tp
->t_timer
[TCPT_PERSIST
] == 0) {
250 * No reason to send a segment, just return.
256 * Before ESTABLISHED, force sending of initial options
257 * unless TCP set not to do any options.
258 * NOTE: we assume that the IP/TCP header plus TCP options
259 * always fit in a single mbuf, leaving room for a maximum
261 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MHLEN
264 hdrlen
= sizeof (struct tcpiphdr
);
265 if (flags
& TH_SYN
) {
266 tp
->snd_nxt
= tp
->iss
;
267 if ((tp
->t_flags
& TF_NOOPT
) == 0) {
270 opt
[0] = TCPOPT_MAXSEG
;
272 mss
= htons((uint16_t) tcp_mss(tp
, 0));
273 memcpy((caddr_t
)(opt
+ 2), (caddr_t
)&mss
, sizeof(mss
));
281 * Adjust data length if insertion of options will
282 * bump the packet length beyond the t_maxseg length.
284 if (len
> tp
->t_maxseg
- optlen
) {
285 len
= tp
->t_maxseg
- optlen
;
290 * Grab a header mbuf, attaching a copy of data to
291 * be transmitted, and initialize the header from
292 * the template for sends on this connection.
295 m
= m_get(so
->slirp
);
300 m
->m_data
+= IF_MAXLINKHDR
;
303 sbcopy(&so
->so_snd
, off
, (int) len
, mtod(m
, caddr_t
) + hdrlen
);
307 * If we're sending everything we've got, set PUSH.
308 * (This will keep happy those implementations which only
309 * give data to the user when a buffer fills or
312 if (off
+ len
== so
->so_snd
.sb_cc
)
315 m
= m_get(so
->slirp
);
320 m
->m_data
+= IF_MAXLINKHDR
;
324 ti
= mtod(m
, struct tcpiphdr
*);
326 memcpy((caddr_t
)ti
, &tp
->t_template
, sizeof (struct tcpiphdr
));
329 * Fill in fields, remembering maximum advertised
330 * window for use in delaying messages about window sizes.
331 * If resending a FIN, be sure not to use a new sequence number.
333 if (flags
& TH_FIN
&& tp
->t_flags
& TF_SENTFIN
&&
334 tp
->snd_nxt
== tp
->snd_max
)
337 * If we are doing retransmissions, then snd_nxt will
338 * not reflect the first unsent octet. For ACK only
339 * packets, we do not want the sequence number of the
340 * retransmitted packet, we want the sequence number
341 * of the next unsent octet. So, if there is no data
342 * (and no SYN or FIN), use snd_max instead of snd_nxt
343 * when filling in ti_seq. But if we are in persist
344 * state, snd_max might reflect one byte beyond the
345 * right edge of the window, so use snd_nxt in that
346 * case, since we know we aren't doing a retransmission.
347 * (retransmit and persist are mutually exclusive...)
349 if (len
|| (flags
& (TH_SYN
|TH_FIN
)) || tp
->t_timer
[TCPT_PERSIST
])
350 ti
->ti_seq
= htonl(tp
->snd_nxt
);
352 ti
->ti_seq
= htonl(tp
->snd_max
);
353 ti
->ti_ack
= htonl(tp
->rcv_nxt
);
355 memcpy((caddr_t
)(ti
+ 1), (caddr_t
)opt
, optlen
);
356 ti
->ti_off
= (sizeof (struct tcphdr
) + optlen
) >> 2;
358 ti
->ti_flags
= flags
;
360 * Calculate receive window. Don't shrink window,
361 * but avoid silly window syndrome.
363 if (win
< (long)(so
->so_rcv
.sb_datalen
/ 4) && win
< (long)tp
->t_maxseg
)
365 if (win
> (long)TCP_MAXWIN
<< tp
->rcv_scale
)
366 win
= (long)TCP_MAXWIN
<< tp
->rcv_scale
;
367 if (win
< (long)(tp
->rcv_adv
- tp
->rcv_nxt
))
368 win
= (long)(tp
->rcv_adv
- tp
->rcv_nxt
);
369 ti
->ti_win
= htons((uint16_t) (win
>>tp
->rcv_scale
));
371 if (SEQ_GT(tp
->snd_up
, tp
->snd_una
)) {
372 ti
->ti_urp
= htons((uint16_t)(tp
->snd_up
- ntohl(ti
->ti_seq
)));
373 ti
->ti_flags
|= TH_URG
;
376 * If no urgent pointer to send, then we pull
377 * the urgent pointer to the left edge of the send window
378 * so that it doesn't drift into the send window on sequence
381 tp
->snd_up
= tp
->snd_una
; /* drag it along */
384 * Put TCP length in extended header, and then
385 * checksum extended header and data.
388 ti
->ti_len
= htons((uint16_t)(sizeof (struct tcphdr
) +
390 ti
->ti_sum
= cksum(m
, (int)(hdrlen
+ len
));
393 * In transmit state, time the transmission and arrange for
394 * the retransmit. In persist state, just set snd_max.
396 if (tp
->t_force
== 0 || tp
->t_timer
[TCPT_PERSIST
] == 0) {
397 tcp_seq startseq
= tp
->snd_nxt
;
400 * Advance snd_nxt over sequence space of this segment.
402 if (flags
& (TH_SYN
|TH_FIN
)) {
405 if (flags
& TH_FIN
) {
407 tp
->t_flags
|= TF_SENTFIN
;
411 if (SEQ_GT(tp
->snd_nxt
, tp
->snd_max
)) {
412 tp
->snd_max
= tp
->snd_nxt
;
414 * Time this transmission if not a retransmission and
415 * not currently timing anything.
417 if (tp
->t_rtt
== 0) {
419 tp
->t_rtseq
= startseq
;
424 * Set retransmit timer if not currently set,
425 * and not doing an ack or a keep-alive probe.
426 * Initial value for retransmit timer is smoothed
427 * round-trip time + 2 * round-trip time variance.
428 * Initialize shift counter which is used for backoff
429 * of retransmit time.
431 if (tp
->t_timer
[TCPT_REXMT
] == 0 &&
432 tp
->snd_nxt
!= tp
->snd_una
) {
433 tp
->t_timer
[TCPT_REXMT
] = tp
->t_rxtcur
;
434 if (tp
->t_timer
[TCPT_PERSIST
]) {
435 tp
->t_timer
[TCPT_PERSIST
] = 0;
440 if (SEQ_GT(tp
->snd_nxt
+ len
, tp
->snd_max
))
441 tp
->snd_max
= tp
->snd_nxt
+ len
;
444 * Fill in IP length and desired time to live and
445 * send to IP level. There should be a better way
446 * to handle ttl and tos; we could keep them in
447 * the template, but need a way to checksum without them.
449 m
->m_len
= hdrlen
+ len
; /* XXX Needed? m_len should be correct */
453 ((struct ip
*)ti
)->ip_len
= m
->m_len
;
455 ((struct ip
*)ti
)->ip_ttl
= IPDEFTTL
;
456 ((struct ip
*)ti
)->ip_tos
= so
->so_iptos
;
458 error
= ip_output(so
, m
);
466 * Data sent (as far as we can tell).
467 * If this advertises a larger window than any other segment,
468 * then remember the size of the advertised window.
469 * Any pending ACK has now been sent.
471 if (win
> 0 && SEQ_GT(tp
->rcv_nxt
+win
, tp
->rcv_adv
))
472 tp
->rcv_adv
= tp
->rcv_nxt
+ win
;
473 tp
->last_ack_sent
= tp
->rcv_nxt
;
474 tp
->t_flags
&= ~(TF_ACKNOW
|TF_DELACK
);
482 tcp_setpersist(struct tcpcb
*tp
)
484 int t
= ((tp
->t_srtt
>> 2) + tp
->t_rttvar
) >> 1;
487 * Start/restart persistence timer.
489 TCPT_RANGESET(tp
->t_timer
[TCPT_PERSIST
],
490 t
* tcp_backoff
[tp
->t_rxtshift
],
491 TCPTV_PERSMIN
, TCPTV_PERSMAX
);
492 if (tp
->t_rxtshift
< TCP_MAXRXTSHIFT
)