tcp_output.c@ 16684

Last change on this file since 16684 was 15890, checked in by vboxsync, 16 years ago

NAT: 1. wo sync enhancement branch is still functional (was corrupted with using ICMP file handler in select(1))

after sending send queue doesn't need to synchronize with NAT thread to free mbuf instead NAT queue used to call freeing slirp routine.
no more copying on slirp to guest sent.

Property svn:eol-style set to native

File size: 20.4 KB

Line
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. All advertising materials mentioning features or use of this software
14	* must display the following acknowledgement:
15	* This product includes software developed by the University of
16	* California, Berkeley and its contributors.
17	* 4. Neither the name of the University nor the names of its contributors
18	* may be used to endorse or promote products derived from this software
19	* without specific prior written permission.
20	*
21	* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22	* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23	* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24	* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25	* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26	* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27	* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28	* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29	* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30	* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31	* SUCH DAMAGE.
32	*
33	* @(#)tcp_output.c 8.3 (Berkeley) 12/30/93
34	* tcp_output.c,v 1.3 1994/09/15 10:36:55 davidg Exp
35	*/
36
37	/*
38	* Changes and additions relating to SLiRP
39	* Copyright (c) 1995 Danny Gasparovski.
40	*
41	* Please read the file COPYRIGHT for the
42	* terms and conditions of the copyright.
43	*/
44
45	#include <slirp.h>
46
47	/*
48	* Since this is only used in "stats socket", we give meaning
49	* names instead of the REAL names
50	*/
51	const char * const tcpstates[] =
52	{
53	/* "CLOSED", "LISTEN", "SYN_SENT", "SYN_RCVD", */
54	"REDIRECT", "LISTEN", "SYN_SENT", "SYN_RCVD",
55	"ESTABLISHED", "CLOSE_WAIT", "FIN_WAIT_1", "CLOSING",
56	"LAST_ACK", "FIN_WAIT_2", "TIME_WAIT",
57	};
58
59	static const u_char tcp_outflags[TCP_NSTATES] =
60	{
61	TH_RST\|TH_ACK, 0, TH_SYN, TH_SYN\|TH_ACK,
62	TH_ACK, TH_ACK, TH_FIN\|TH_ACK, TH_FIN\|TH_ACK,
63	TH_FIN\|TH_ACK, TH_ACK, TH_ACK,
64	};
65
66
67	#define MAX_TCPOPTLEN 32 /* max # bytes that go in options */
68
69	/*
70	* Tcp output routine: figure out what should be sent and send it.
71	*/
72	int
73	tcp_output(PNATState pData, register struct tcpcb *tp)
74	{
75	register struct socket *so = tp->t_socket;
76	register long len, win;
77	int off, flags, error;
78	register struct mbuf *m;
79	register struct tcpiphdr *ti;
80	u_char opt[MAX_TCPOPTLEN];
81	unsigned optlen, hdrlen;
82	int idle, sendalot;
83
84	DEBUG_CALL("tcp_output");
85	DEBUG_ARG("tp = %lx", (long )tp);
86
87	/*
88	* Determine length of data that should be transmitted,
89	* and flags that will be used.
90	* If there is some data or critical controls (SYN, RST)
91	* to send, then transmit; otherwise, investigate further.
92	*/
93	idle = (tp->snd_max == tp->snd_una);
94	if (idle && tp->t_idle >= tp->t_rxtcur)
95	/*
96	* We have been idle for "a while" and no acks are
97	* expected to clock out any data we send --
98	* slow start to get ack "clock" running again.
99	*/
100	tp->snd_cwnd = tp->t_maxseg;
101
102	again:
103	sendalot = 0;
104	off = tp->snd_nxt - tp->snd_una;
105	win = min(tp->snd_wnd, tp->snd_cwnd);
106
107	flags = tcp_outflags[tp->t_state];
108
109	DEBUG_MISC((dfd, " --- tcp_output flags = 0x%x\n",flags));
110
111	/*
112	* If in persist timeout with window of 0, send 1 byte.
113	* Otherwise, if window is small but nonzero
114	* and timer expired, we will send what we can
115	* and go to transmit state.
116	*/
117	if (tp->t_force)
118	{
119	if (win == 0)
120	{
121	/*
122	* If we still have some data to send, then
123	* clear the FIN bit. Usually this would
124	* happen below when it realizes that we
125	* aren't sending all the data. However,
126	* if we have exactly 1 byte of unset data,
127	* then it won't clear the FIN bit below,
128	* and if we are in persist state, we wind
129	* up sending the packet without recording
130	* that we sent the FIN bit.
131	*
132	* We can't just blindly clear the FIN bit,
133	* because if we don't have any more data
134	* to send then the probe will be the FIN
135	* itself.
136	*/
137	if (off < so->so_snd.sb_cc)
138	flags &= ~TH_FIN;
139	win = 1;
140	}
141	else
142	{
143	tp->t_timer[TCPT_PERSIST] = 0;
144	tp->t_rxtshift = 0;
145	}
146	}
147
148	len = min(so->so_snd.sb_cc, win) - off;
149	if (len < 0)
150	{
151	/*
152	* If FIN has been sent but not acked,
153	* but we haven't been called to retransmit,
154	* len will be -1. Otherwise, window shrank
155	* after we sent into it. If window shrank to 0,
156	* cancel pending retransmit and pull snd_nxt
157	* back to (closed) window. We will enter persist
158	* state below. If the window didn't close completely,
159	* just wait for an ACK.
160	*/
161	len = 0;
162	if (win == 0)
163	{
164	tp->t_timer[TCPT_REXMT] = 0;
165	tp->snd_nxt = tp->snd_una;
166	}
167	}
168	if (len > tp->t_maxseg)
169	{
170	len = tp->t_maxseg;
171	sendalot = 1;
172	}
173	if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
174	flags &= ~TH_FIN;
175
176	win = sbspace(&so->so_rcv);
177
178	/*
179	* Sender silly window avoidance. If connection is idle
180	* and can send all data, a maximum segment,
181	* at least a maximum default-size segment do it,
182	* or are forced, do it; otherwise don't bother.
183	* If peer's buffer is tiny, then send
184	* when window is at least half open.
185	* If retransmitting (possibly after persist timer forced us
186	* to send into a small window), then must resend.
187	*/
188	if (len)
189	{
190	if (len == tp->t_maxseg)
191	goto send;
192	if ((1 \|\| idle \|\| tp->t_flags & TF_NODELAY) &&
193	len + off >= so->so_snd.sb_cc)
194	goto send;
195	if (tp->t_force)
196	goto send;
197	if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
198	goto send;
199	if (SEQ_LT(tp->snd_nxt, tp->snd_max))
200	goto send;
201	}
202
203	/*
204	* Compare available window to amount of window
205	* known to peer (as advertised window less
206	* next expected input). If the difference is at least two
207	* max size segments, or at least 50% of the maximum possible
208	* window, then want to send a window update to peer.
209	*/
210	if (win > 0)
211	{
212	/*
213	* "adv" is the amount we can increase the window,
214	* taking into account that we are limited by
215	* TCP_MAXWIN << tp->rcv_scale.
216	*/
217	long adv = min(win,
218	(long)TCP_MAXWIN << tp->rcv_scale) -
219	(tp->rcv_adv - tp->rcv_nxt);
220
221	if (adv >= (long) (2 * tp->t_maxseg))
222	goto send;
223	if (2 * adv >= (long) so->so_rcv.sb_datalen)
224	goto send;
225	}
226
227	/*
228	* Send if we owe peer an ACK.
229	*/
230	if (tp->t_flags & TF_ACKNOW)
231	goto send;
232	if (flags & (TH_SYN\|TH_RST))
233	goto send;
234	if (SEQ_GT(tp->snd_up, tp->snd_una))
235	goto send;
236	/*
237	* If our state indicates that FIN should be sent
238	* and we have not yet done so, or we're retransmitting the FIN,
239	* then we need to send.
240	*/
241	if ( flags & TH_FIN
242	&& ((tp->t_flags & TF_SENTFIN) == 0 \|\| tp->snd_nxt == tp->snd_una))
243	goto send;
244
245	/*
246	* TCP window updates are not reliable, rather a polling protocol
247	* using ``persist'' packets is used to insure receipt of window
248	* updates. The three ``states'' for the output side are:
249	* idle not doing retransmits or persists
250	* persisting to move a small or zero window
251	* (re)transmitting and thereby not persisting
252	*
253	* tp->t_timer[TCPT_PERSIST]
254	* is set when we are in persist state.
255	* tp->t_force
256	* is set when we are called to send a persist packet.
257	* tp->t_timer[TCPT_REXMT]
258	* is set when we are retransmitting
259	* The output side is idle when both timers are zero.
260	*
261	* If send window is too small, there is data to transmit, and no
262	* retransmit or persist is pending, then go to persist state.
263	* If nothing happens soon, send when timer expires:
264	* if window is nonzero, transmit what we can,
265	* otherwise force out a byte.
266	*/
267	if ( so->so_snd.sb_cc
268	&& tp->t_timer[TCPT_REXMT] == 0
269	&& tp->t_timer[TCPT_PERSIST] == 0)
270	{
271	tp->t_rxtshift = 0;
272	tcp_setpersist(tp);
273	}
274
275	/*
276	* No reason to send a segment, just return.
277	*/
278	tcpstat.tcps_didnuttin++;
279
280	return (0);
281
282	send:
283	/*
284	* Before ESTABLISHED, force sending of initial options
285	* unless TCP set not to do any options.
286	* NOTE: we assume that the IP/TCP header plus TCP options
287	* always fit in a single mbuf, leaving room for a maximum
288	* link header, i.e.
289	* max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MHLEN
290	*/
291	optlen = 0;
292	hdrlen = sizeof (struct tcpiphdr);
293	if (flags & TH_SYN)
294	{
295	tp->snd_nxt = tp->iss;
296	if ((tp->t_flags & TF_NOOPT) == 0)
297	{
298	u_int16_t mss;
299
300	opt[0] = TCPOPT_MAXSEG;
301	opt[1] = 4;
302	mss = htons((u_int16_t) tcp_mss(pData, tp, 0));
303	memcpy((caddr_t)(opt + 2), (caddr_t)&mss, sizeof(mss));
304	optlen = 4;
305
306	#if 0
307	if ( (tp->t_flags & TF_REQ_SCALE)
308	&& ( (flags & TH_ACK) == 0
309	\|\| (tp->t_flags & TF_RCVD_SCALE)))
310	{
311	((u_int32_t ) (opt + optlen)) = htonl( TCPOPT_NOP << 24
312	\| TCPOPT_WINDOW << 16
313	\| TCPOLEN_WINDOW << 8
314	\| tp->request_r_scale);
315	optlen += 4;
316	}
317	#endif
318	}
319	}
320
321	/*
322	* Send a timestamp and echo-reply if this is a SYN and our side
323	* wants to use timestamps (TF_REQ_TSTMP is set) or both our side
324	* and our peer have sent timestamps in our SYN's.
325	*/
326	#if 0
327	if ( (tp->t_flags & (TF_REQ_TSTMP\|TF_NOOPT)) == TF_REQ_TSTMP
328	&& (flags & TH_RST) == 0
329	&& ( (flags & (TH_SYN\|TH_ACK)) == TH_SYN
330	\|\| (tp->t_flags & TF_RCVD_TSTMP)))
331	{
332	u_int32_t lp = (u_int32_t )(opt + optlen);
333
334	/* Form timestamp option as shown in appendix A of RFC 1323. */
335	*lp++ = htonl(TCPOPT_TSTAMP_HDR);
336	*lp++ = htonl(tcp_now);
337	*lp = htonl(tp->ts_recent);
338	optlen += TCPOLEN_TSTAMP_APPA;
339	}
340	#endif
341	hdrlen += optlen;
342
343	/*
344	* Adjust data length if insertion of options will
345	* bump the packet length beyond the t_maxseg length.
346	*/
347	if (len > tp->t_maxseg - optlen)
348	{
349	len = tp->t_maxseg - optlen;
350	sendalot = 1;
351	}
352
353	/*
354	* Grab a header mbuf, attaching a copy of data to
355	* be transmitted, and initialize the header from
356	* the template for sends on this connection.
357	*/
358	if (len)
359	{
360	if (tp->t_force && len == 1)
361	tcpstat.tcps_sndprobe++;
362	else if (SEQ_LT(tp->snd_nxt, tp->snd_max))
363	{
364	tcpstat.tcps_sndrexmitpack++;
365	tcpstat.tcps_sndrexmitbyte += len;
366	}
367	else
368	{
369	tcpstat.tcps_sndpack++;
370	tcpstat.tcps_sndbyte += len;
371	}
372
373	m = m_get(pData);
374	if (m == NULL)
375	{
376	/* error = ENOBUFS; */
377	error = 1;
378	goto out;
379	}
380	m->m_data += if_maxlinkhdr;
381	m->m_len = hdrlen;
382
383	/*
384	* This will always succeed, since we make sure our mbufs
385	* are big enough to hold one MSS packet + header + ... etc.
386	*/
387	#if 0
388	if (len <= MHLEN - hdrlen - max_linkhdr)
389	{
390	#endif
391	sbcopy(&so->so_snd, off, (int) len, mtod(m, caddr_t) + hdrlen);
392	m->m_len += len;
393	#if 0
394	}
395	else
396	{
397	m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
398	if (m->m_next == 0)
399	len = 0;
400	}
401	#endif
402	/*
403	* If we're sending everything we've got, set PUSH.
404	* (This will keep happy those implementations which only
405	* give data to the user when a buffer fills or
406	* a PUSH comes in.)
407	*/
408	if (off + len == so->so_snd.sb_cc)
409	flags \|= TH_PUSH;
410	}
411	else
412	{
413	if (tp->t_flags & TF_ACKNOW)
414	tcpstat.tcps_sndacks++;
415	else if (flags & (TH_SYN\|TH_FIN\|TH_RST))
416	tcpstat.tcps_sndctrl++;
417	else if (SEQ_GT(tp->snd_up, tp->snd_una))
418	tcpstat.tcps_sndurg++;
419	else
420	tcpstat.tcps_sndwinup++;
421
422	m = m_get(pData);
423	if (m == NULL)
424	{
425	/* error = ENOBUFS; */
426	error = 1;
427	goto out;
428	}
429	m->m_data += if_maxlinkhdr;
430	#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
431	m->m_data += sizeof(struct ip)
432	+ sizeof(struct tcphdr);
433	#endif
434	m->m_len = hdrlen;
435	}
436
437	ti = mtod(m, struct tcpiphdr *);
438
439	memcpy((caddr_t)ti, &tp->t_template, sizeof (struct tcpiphdr));
440
441	/*
442	* Fill in fields, remembering maximum advertised
443	* window for use in delaying messages about window sizes.
444	* If resending a FIN, be sure not to use a new sequence number.
445	*/
446	if ( flags & TH_FIN
447	&& tp->t_flags & TF_SENTFIN
448	&& tp->snd_nxt == tp->snd_max)
449	tp->snd_nxt--;
450	/*
451	* If we are doing retransmissions, then snd_nxt will
452	* not reflect the first unsent octet. For ACK only
453	* packets, we do not want the sequence number of the
454	* retransmitted packet, we want the sequence number
455	* of the next unsent octet. So, if there is no data
456	* (and no SYN or FIN), use snd_max instead of snd_nxt
457	* when filling in ti_seq. But if we are in persist
458	* state, snd_max might reflect one byte beyond the
459	* right edge of the window, so use snd_nxt in that
460	* case, since we know we aren't doing a retransmission.
461	* (retransmit and persist are mutually exclusive...)
462	*/
463	if (len \|\| (flags & (TH_SYN\|TH_FIN)) \|\| tp->t_timer[TCPT_PERSIST])
464	ti->ti_seq = htonl(tp->snd_nxt);
465	else
466	ti->ti_seq = htonl(tp->snd_max);
467	ti->ti_ack = htonl(tp->rcv_nxt);
468	if (optlen)
469	{
470	memcpy((caddr_t)(ti + 1), (caddr_t)opt, optlen);
471	ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2;
472	}
473	ti->ti_flags = flags;
474	/*
475	* Calculate receive window. Don't shrink window,
476	* but avoid silly window syndrome.
477	*/
478	if (win < (long)(so->so_rcv.sb_datalen / 4) && win < (long)tp->t_maxseg)
479	win = 0;
480	if (win > (long)TCP_MAXWIN << tp->rcv_scale)
481	win = (long)TCP_MAXWIN << tp->rcv_scale;
482	if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
483	win = (long)(tp->rcv_adv - tp->rcv_nxt);
484	ti->ti_win = htons((u_int16_t) (win>>tp->rcv_scale));
485
486	#if 0
487	if (SEQ_GT(tp->snd_up, tp->snd_nxt))
488	{
489	ti->ti_urp = htons((u_int16_t)(tp->snd_up - tp->snd_nxt));
490	#else
491	if (SEQ_GT(tp->snd_up, tp->snd_una))
492	{
493	ti->ti_urp = htons((u_int16_t)(tp->snd_up - ntohl(ti->ti_seq)));
494	#endif
495	ti->ti_flags \|= TH_URG;
496	}
497	else
498	/*
499	* If no urgent pointer to send, then we pull
500	* the urgent pointer to the left edge of the send window
501	* so that it doesn't drift into the send window on sequence
502	* number wraparound.
503	*/
504	tp->snd_up = tp->snd_una; /* drag it along */
505
506	/*
507	* Put TCP length in extended header, and then
508	* checksum extended header and data.
509	*/
510	if (len + optlen)
511	ti->ti_len = htons((u_int16_t)(sizeof (struct tcphdr)
512	+ optlen + len));
513	ti->ti_sum = cksum(m, (int)(hdrlen + len));
514
515	/*
516	* In transmit state, time the transmission and arrange for
517	* the retransmit. In persist state, just set snd_max.
518	*/
519	if (tp->t_force == 0 \|\| tp->t_timer[TCPT_PERSIST] == 0)
520	{
521	tcp_seq startseq = tp->snd_nxt;
522
523	/*
524	* Advance snd_nxt over sequence space of this segment.
525	*/
526	if (flags & (TH_SYN\|TH_FIN))
527	{
528	if (flags & TH_SYN)
529	tp->snd_nxt++;
530	if (flags & TH_FIN)
531	{
532	tp->snd_nxt++;
533	tp->t_flags \|= TF_SENTFIN;
534	}
535	}
536	tp->snd_nxt += len;
537	if (SEQ_GT(tp->snd_nxt, tp->snd_max))
538	{
539	tp->snd_max = tp->snd_nxt;
540	/*
541	* Time this transmission if not a retransmission and
542	* not currently timing anything.
543	*/
544	if (tp->t_rtt == 0)
545	{
546	tp->t_rtt = 1;
547	tp->t_rtseq = startseq;
548	tcpstat.tcps_segstimed++;
549	}
550	}
551
552	/*
553	* Set retransmit timer if not currently set,
554	* and not doing an ack or a keep-alive probe.
555	* Initial value for retransmit timer is smoothed
556	* round-trip time + 2 * round-trip time variance.
557	* Initialize shift counter which is used for backoff
558	* of retransmit time.
559	*/
560	if ( tp->t_timer[TCPT_REXMT] == 0
561	&& tp->snd_nxt != tp->snd_una)
562	{
563	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
564	if (tp->t_timer[TCPT_PERSIST])
565	{
566	tp->t_timer[TCPT_PERSIST] = 0;
567	tp->t_rxtshift = 0;
568	}
569	}
570	}
571	else
572	if (SEQ_GT(tp->snd_nxt + len, tp->snd_max))
573	tp->snd_max = tp->snd_nxt + len;
574
575	/*
576	* Fill in IP length and desired time to live and
577	* send to IP level. There should be a better way
578	* to handle ttl and tos; we could keep them in
579	* the template, but need a way to checksum without them.
580	*/
581	m->m_len = hdrlen + len; /* XXX Needed? m_len should be correct */
582
583	{
584	((struct ip *)ti)->ip_len = m->m_len;
585	((struct ip *)ti)->ip_ttl = ip_defttl;
586	((struct ip *)ti)->ip_tos = so->so_iptos;
587
588	/* #if BSD >= 43 */
589	/* Don't do IP options... */
590	#if 0
591	error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
592	so->so_options & SO_DONTROUTE, 0);
593	#endif
594	error = ip_output(pData, so, m);
595
596	#if 0
597	/* #else */
598	error = ip_output(m, (struct mbuf *)0, &tp->t_inpcb->inp_route,
599	so->so_options & SO_DONTROUTE);
600	/* #endif */
601	#endif
602	}
603	if (error)
604	{
605	out:
606	#if 0
607	if (error == ENOBUFS)
608	{
609	tcp_quench(tp->t_inpcb, 0);
610	return (0);
611	}
612
613	if ( ( error == EHOSTUNREACH
614	\|\| error == ENETDOWN)
615	&& TCPS_HAVERCVDSYN(tp->t_state))
616	{
617	tp->t_softerror = error;
618	return (0);
619	}
620	#endif
621	return (error);
622	}
623	tcpstat.tcps_sndtotal++;
624
625	/*
626	* Data sent (as far as we can tell).
627	* If this advertises a larger window than any other segment,
628	* then remember the size of the advertised window.
629	* Any pending ACK has now been sent.
630	*/
631	if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
632	tp->rcv_adv = tp->rcv_nxt + win;
633	tp->last_ack_sent = tp->rcv_nxt;
634	tp->t_flags &= ~(TF_ACKNOW\|TF_DELACK);
635	if (sendalot)
636	goto again;
637
638	return (0);
639	}
640
641	void
642	tcp_setpersist(struct tcpcb *tp)
643	{
644	int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
645
646	#if 0
647	if (tp->t_timer[TCPT_REXMT])
648	panic("tcp_output REXMT");
649	#endif
650	/*
651	* Start/restart persistence timer.
652	*/
653	TCPT_RANGESET(tp->t_timer[TCPT_PERSIST],
654	t * tcp_backoff[tp->t_rxtshift],
655	TCPTV_PERSMIN, TCPTV_PERSMAX);
656	if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
657	tp->t_rxtshift++;
658	}

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_output.c@ 16684

Download in other formats: