tcp_input.c@ 63215

Last change on this file since 63215 was 63016, checked in by vboxsync, 8 years ago
Devices: warnings (debug builds)
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 66.4 KB

Line
1	/* $Id: tcp_input.c 63016 2016-08-04 22:47:52Z vboxsync $ */
2	/** @file
3	* NAT - TCP input.
4	*/
5
6	/*
7	* Copyright (C) 2006-2016 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18	/*
19	* This code is based on:
20	*
21	* Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994
22	* The Regents of the University of California. All rights reserved.
23	*
24	* Redistribution and use in source and binary forms, with or without
25	* modification, are permitted provided that the following conditions
26	* are met:
27	* 1. Redistributions of source code must retain the above copyright
28	* notice, this list of conditions and the following disclaimer.
29	* 2. Redistributions in binary form must reproduce the above copyright
30	* notice, this list of conditions and the following disclaimer in the
31	* documentation and/or other materials provided with the distribution.
32	* 3. All advertising materials mentioning features or use of this software
33	* must display the following acknowledgement:
34	* This product includes software developed by the University of
35	* California, Berkeley and its contributors.
36	* 4. Neither the name of the University nor the names of its contributors
37	* may be used to endorse or promote products derived from this software
38	* without specific prior written permission.
39	*
40	* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
41	* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
42	* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
43	* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
44	* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
45	* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
46	* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
47	* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
48	* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
49	* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
50	* SUCH DAMAGE.
51	*
52	* @(#)tcp_input.c 8.5 (Berkeley) 4/10/94
53	* tcp_input.c,v 1.10 1994/10/13 18:36:32 wollman Exp
54	*/
55
56	/*
57	* Changes and additions relating to SLiRP
58	* Copyright (c) 1995 Danny Gasparovski.
59	*
60	* Please read the file COPYRIGHT for the
61	* terms and conditions of the copyright.
62	*/
63
64	#include <slirp.h>
65	#include "ip_icmp.h"
66
67
68	#if 0 /* code using this macroses is commented out */
69	# define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ)
70
71	/* for modulo comparisons of timestamps */
72	# define TSTMP_LT(a, b) ((int)((a)-(b)) < 0)
73	# define TSTMP_GEQ(a, b) ((int)((a)-(b)) >= 0)
74	#endif
75
76	#ifndef TCP_ACK_HACK
77	#define DELAY_ACK(tp, ti) \
78	if (ti->ti_flags & TH_PUSH) \
79	tp->t_flags \|= TF_ACKNOW; \
80	else \
81	tp->t_flags \|= TF_DELACK;
82	#else /* !TCP_ACK_HACK */
83	#define DELAY_ACK(tp, ign) \
84	tp->t_flags \|= TF_DELACK;
85	#endif /* TCP_ACK_HACK */
86
87
88	/*
89	* deps: netinet/tcp_reass.c
90	* tcp_reass_maxqlen = 48 (deafault)
91	* tcp_reass_maxseg = nmbclusters/16 (nmbclusters = 1024 + maxusers * 64 from kern/kern_mbuf.c let's say 256)
92	*/
93	int
94	tcp_reass(PNATState pData, struct tcpcb tp, struct tcphdr th, int tlenp, struct mbuf m)
95	{
96	struct tseg_qent *q;
97	struct tseg_qent *p = NULL;
98	struct tseg_qent *nq;
99	struct tseg_qent *te = NULL;
100	struct socket *so = tp->t_socket;
101	int flags;
102	STAM_PROFILE_START(&pData->StatTCP_reassamble, tcp_reassamble);
103	LogFlowFunc(("ENTER: pData:%p, tp:%R[tcpcb793], th:%p, tlenp:%p, m:%p\n", pData, tp, th, tlenp, m));
104
105	/*
106	* XXX: tcp_reass() is rather inefficient with its data structures
107	* and should be rewritten (see NetBSD for optimizations). While
108	* doing that it should move to its own file tcp_reass.c.
109	*/
110
111	/*
112	* Call with th==NULL after become established to
113	* force pre-ESTABLISHED data up to user socket.
114	*/
115	if (th == NULL)
116	{
117	LogFlowFunc(("%d -> present\n", __LINE__));
118	goto present;
119	}
120
121	/*
122	* Limit the number of segments in the reassembly queue to prevent
123	* holding on to too many segments (and thus running out of mbufs).
124	* Make sure to let the missing segment through which caused this
125	* queue. Always keep one global queue entry spare to be able to
126	* process the missing segment.
127	*/
128	if ( th->th_seq != tp->rcv_nxt
129	&& ( tcp_reass_qsize + 1 >= tcp_reass_maxseg
130	\|\| tp->t_segqlen >= tcp_reass_maxqlen))
131	{
132	tcp_reass_overflows++;
133	tcpstat.tcps_rcvmemdrop++;
134	m_freem(pData, m);
135	*tlenp = 0;
136	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
137	LogFlowFuncLeave();
138	return (0);
139	}
140
141	/*
142	* Allocate a new queue entry. If we can't, or hit the zone limit
143	* just drop the pkt.
144	*/
145	te = RTMemAlloc(sizeof(struct tseg_qent));
146	if (te == NULL)
147	{
148	tcpstat.tcps_rcvmemdrop++;
149	m_freem(pData, m);
150	*tlenp = 0;
151	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
152	LogFlowFuncLeave();
153	return (0);
154	}
155	tp->t_segqlen++;
156	tcp_reass_qsize++;
157
158	/*
159	* Find a segment which begins after this one does.
160	*/
161	LIST_FOREACH(q, &tp->t_segq, tqe_q)
162	{
163	if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
164	break;
165	p = q;
166	}
167
168	/*
169	* If there is a preceding segment, it may provide some of
170	* our data already. If so, drop the data from the incoming
171	* segment. If it provides all of our data, drop us.
172	*/
173	if (p != NULL)
174	{
175	int i;
176	/* conversion to int (in i) handles seq wraparound */
177	i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
178	if (i > 0)
179	{
180	if (i >= *tlenp)
181	{
182	tcpstat.tcps_rcvduppack++;
183	tcpstat.tcps_rcvdupbyte += *tlenp;
184	m_freem(pData, m);
185	RTMemFree(te);
186	tp->t_segqlen--;
187	tcp_reass_qsize--;
188	/*
189	* Try to present any queued data
190	* at the left window edge to the user.
191	* This is needed after the 3-WHS
192	* completes.
193	*/
194	LogFlowFunc(("%d -> present\n", __LINE__));
195	goto present; /* ??? */
196	}
197	m_adj(m, i);
198	*tlenp -= i;
199	th->th_seq += i;
200	}
201	}
202	tcpstat.tcps_rcvoopack++;
203	tcpstat.tcps_rcvoobyte += *tlenp;
204
205	/*
206	* While we overlap succeeding segments trim them or,
207	* if they are completely covered, dequeue them.
208	*/
209	while (q)
210	{
211	int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
212	if (i <= 0)
213	break;
214	if (i < q->tqe_len)
215	{
216	q->tqe_th->th_seq += i;
217	q->tqe_len -= i;
218	m_adj(q->tqe_m, i);
219	break;
220	}
221
222	nq = LIST_NEXT(q, tqe_q);
223	LIST_REMOVE(q, tqe_q);
224	m_freem(pData, q->tqe_m);
225	RTMemFree(q);
226	tp->t_segqlen--;
227	tcp_reass_qsize--;
228	q = nq;
229	}
230
231	/* Insert the new segment queue entry into place. */
232	te->tqe_m = m;
233	te->tqe_th = th;
234	te->tqe_len = *tlenp;
235
236	if (p == NULL)
237	{
238	LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
239	}
240	else
241	{
242	LIST_INSERT_AFTER(p, te, tqe_q);
243	}
244
245	present:
246	/*
247	* Present data to user, advancing rcv_nxt through
248	* completed sequence space.
249	*/
250	if (!TCPS_HAVEESTABLISHED(tp->t_state))
251	{
252	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
253	return (0);
254	}
255	q = LIST_FIRST(&tp->t_segq);
256	if (!q \|\| q->tqe_th->th_seq != tp->rcv_nxt)
257	{
258	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
259	return (0);
260	}
261	do
262	{
263	tp->rcv_nxt += q->tqe_len;
264	flags = q->tqe_th->th_flags & TH_FIN;
265	nq = LIST_NEXT(q, tqe_q);
266	LIST_REMOVE(q, tqe_q);
267	/* XXX: This place should be checked for the same code in
268	* original BSD code for Slirp and current BSD used SS_FCANTRCVMORE
269	*/
270	if (so->so_state & SS_FCANTSENDMORE)
271	m_freem(pData, q->tqe_m);
272	else
273	sbappend(pData, so, q->tqe_m);
274	RTMemFree(q);
275	tp->t_segqlen--;
276	tcp_reass_qsize--;
277	q = nq;
278	}
279	while (q && q->tqe_th->th_seq == tp->rcv_nxt);
280
281	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
282	return flags;
283	}
284
285	/*
286	* TCP input routine, follows pages 65-76 of the
287	* protocol specification dated September, 1981 very closely.
288	*/
289	void
290	tcp_input(PNATState pData, register struct mbuf m, int iphlen, struct socket inso)
291	{
292	struct ip ip, save_ip;
293	register struct tcpiphdr *ti;
294	caddr_t optp = NULL;
295	int optlen = 0;
296	int len, off;
297	int tlen = 0; /* Shut up MSC (didn't check whether MSC was right). */
298	register struct tcpcb *tp = 0;
299	register int tiflags;
300	struct socket *so = 0;
301	int todrop, acked, ourfinisacked, needoutput = 0;
302	/* int dropsocket = 0; */
303	int iss = 0;
304	u_long tiwin;
305	/* int ts_present = 0; */
306	unsigned ohdrlen;
307	uint8_t ohdr[60 + 8]; /* max IP header plus 8 bytes of payload for icmp */
308
309	STAM_PROFILE_START(&pData->StatTCP_input, counter_input);
310
311	LogFlow(("tcp_input: m = %p, iphlen = %2d, inso = %R[natsock]\n", m, iphlen, inso));
312
313	if (inso != NULL)
314	{
315	QSOCKET_LOCK(tcb);
316	SOCKET_LOCK(inso);
317	QSOCKET_UNLOCK(tcb);
318	}
319	/*
320	* If called with m == 0, then we're continuing the connect
321	*/
322	if (m == NULL)
323	{
324	so = inso;
325	Log4(("NAT: tcp_input: %R[natsock]\n", so));
326	/* Re-set a few variables */
327	tp = sototcpcb(so);
328	m = so->so_m;
329	so->so_m = 0;
330
331	if (RT_LIKELY(so->so_ohdr != NULL))
332	{
333	RTMemFree(so->so_ohdr);
334	so->so_ohdr = NULL;
335	}
336
337	ti = so->so_ti;
338
339	/** @todo (vvl) clarify why it might happens */
340	if (ti == NULL)
341	{
342	LogRel(("NAT: ti is null. can't do any reseting connection actions\n"));
343	/* mbuf should be cleared in sofree called from tcp_close */
344	tcp_close(pData, tp);
345	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
346	LogFlowFuncLeave();
347	return;
348	}
349
350	tiwin = ti->ti_win;
351	tiflags = ti->ti_flags;
352
353	LogFlowFunc(("%d -> cont_conn\n", __LINE__));
354	goto cont_conn;
355	}
356
357	tcpstat.tcps_rcvtotal++;
358
359	ip = mtod(m, struct ip *);
360
361	/* ip_input() subtracts iphlen from ip::ip_len */
362	AssertStmt(ip->ip_len + iphlen == (ssize_t)m_length(m, NULL), goto drop);
363	if (RT_UNLIKELY(ip->ip_len < sizeof(struct tcphdr)))
364	{
365	/* tcps_rcvshort++; */
366	goto drop;
367	}
368
369	/*
370	* Save a copy of the IP header in case we want to restore it for
371	* sending an ICMP error message in response.
372	*
373	* XXX: This function should really be fixed to not strip IP
374	* options, to not overwrite IP header and to use "tlen" local
375	* variable (instead of ti->ti_len), then "m" could be passed to
376	* icmp_error() directly.
377	*/
378	ohdrlen = iphlen + 8;
379	m_copydata(m, 0, ohdrlen, (caddr_t)ohdr);
380	save_ip = (struct ip *)ohdr;
381	save_ip->ip_len += iphlen; /* undo change by ip_input() */
382
383
384	/*
385	* Get IP and TCP header together in first mbuf.
386	* Note: IP leaves IP header in first mbuf.
387	*/
388	ti = mtod(m, struct tcpiphdr *);
389	if (iphlen > sizeof(struct ip))
390	{
391	ip_stripoptions(m, (struct mbuf *)0);
392	iphlen = sizeof(struct ip);
393	}
394
395	/*
396	* Checksum extended TCP header and data.
397	*/
398	tlen = ((struct ip *)ti)->ip_len;
399	memset(ti->ti_x1, 0, 9);
400	ti->ti_len = RT_H2N_U16((u_int16_t)tlen);
401	len = sizeof(struct ip) + tlen;
402	/* keep checksum for ICMP reply
403	* ti->ti_sum = cksum(m, len);
404	* if (ti->ti_sum) { */
405	if (cksum(m, len))
406	{
407	tcpstat.tcps_rcvbadsum++;
408	LogFlowFunc(("%d -> drop\n", __LINE__));
409	goto drop;
410	}
411
412	/*
413	* Check that TCP offset makes sense,
414	* pull out TCP options and adjust length. XXX
415	*/
416	off = ti->ti_off << 2;
417	if ( off < sizeof (struct tcphdr)
418	\|\| off > tlen)
419	{
420	tcpstat.tcps_rcvbadoff++;
421	LogFlowFunc(("%d -> drop\n", __LINE__));
422	goto drop;
423	}
424	tlen -= off;
425	ti->ti_len = tlen;
426	if (off > sizeof (struct tcphdr))
427	{
428	optlen = off - sizeof (struct tcphdr);
429	optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
430
431	/*
432	* Do quick retrieval of timestamp options ("options
433	* prediction?"). If timestamp is the only option and it's
434	* formatted as recommended in RFC 1323 appendix A, we
435	* quickly get the values now and not bother calling
436	* tcp_dooptions(), etc.
437	*/
438	#if 0
439	if (( optlen == TCPOLEN_TSTAMP_APPA
440	\|\| ( optlen > TCPOLEN_TSTAMP_APPA
441	&& optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
442	(u_int32_t )optp == RT_H2N_U32_C(TCPOPT_TSTAMP_HDR) &&
443	(ti->ti_flags & TH_SYN) == 0)
444	{
445	ts_present = 1;
446	ts_val = RT_N2H_U32((u_int32_t )(optp + 4));
447	ts_ecr = RT_N2H_U32((u_int32_t )(optp + 8));
448	optp = NULL; / * we have parsed the options * /
449	}
450	#endif
451	}
452	tiflags = ti->ti_flags;
453
454	/*
455	* Convert TCP protocol specific fields to host format.
456	*/
457	NTOHL(ti->ti_seq);
458	NTOHL(ti->ti_ack);
459	NTOHS(ti->ti_win);
460	NTOHS(ti->ti_urp);
461
462	/*
463	* Drop TCP, IP headers and TCP options.
464	*/
465	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
466	m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
467
468	/*
469	* Locate pcb for segment.
470	*/
471	findso:
472	LogFlowFunc(("(enter) findso: %R[natsock]\n", so));
473	if (so != NULL && so != &tcb)
474	SOCKET_UNLOCK(so);
475	QSOCKET_LOCK(tcb);
476	so = tcp_last_so;
477	if ( so->so_fport != ti->ti_dport
478	\|\| so->so_lport != ti->ti_sport
479	\|\| so->so_laddr.s_addr != ti->ti_src.s_addr
480	\|\| so->so_faddr.s_addr != ti->ti_dst.s_addr)
481	{
482	QSOCKET_UNLOCK(tcb);
483	/* @todo fix SOLOOKUP macrodefinition to be usable here */
484	so = solookup(&tcb, ti->ti_src, ti->ti_sport,
485	ti->ti_dst, ti->ti_dport);
486	if (so)
487	{
488	tcp_last_so = so;
489	}
490	++tcpstat.tcps_socachemiss;
491	}
492	else
493	{
494	SOCKET_LOCK(so);
495	QSOCKET_UNLOCK(tcb);
496	}
497	LogFlowFunc(("(leave) findso: %R[natsock]\n", so));
498
499	/*
500	* If the state is CLOSED (i.e., TCB does not exist) then
501	* all data in the incoming segment is discarded.
502	* If the TCB exists but is in CLOSED state, it is embryonic,
503	* but should either do a listen or a connect soon.
504	*
505	* state == CLOSED means we've done socreate() but haven't
506	* attached it to a protocol yet...
507	*
508	* XXX If a TCB does not exist, and the TH_SYN flag is
509	* the only flag set, then create a session, mark it
510	* as if it was LISTENING, and continue...
511	*/
512	if (so == 0)
513	{
514	if ((tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) != TH_SYN)
515	{
516	LogFlowFunc(("%d -> dropwithreset\n", __LINE__));
517	goto dropwithreset;
518	}
519
520	if ((so = socreate()) == NULL)
521	{
522	LogFlowFunc(("%d -> dropwithreset\n", __LINE__));
523	goto dropwithreset;
524	}
525	if (tcp_attach(pData, so) < 0)
526	{
527	RTMemFree(so); /* Not sofree (if it failed, it's not insqued) */
528	LogFlowFunc(("%d -> dropwithreset\n", __LINE__));
529	goto dropwithreset;
530	}
531	SOCKET_LOCK(so);
532	sbreserve(pData, &so->so_snd, tcp_sndspace);
533	sbreserve(pData, &so->so_rcv, tcp_rcvspace);
534
535	/* tcp_last_so = so; / / XXX ? */
536	/* tp = sototcpcb(so); */
537
538	so->so_laddr = ti->ti_src;
539	so->so_lport = ti->ti_sport;
540	so->so_faddr = ti->ti_dst;
541	so->so_fport = ti->ti_dport;
542
543	so->so_iptos = ((struct ip *)ti)->ip_tos;
544
545	tp = sototcpcb(so);
546	TCP_STATE_SWITCH_TO(tp, TCPS_LISTEN);
547	}
548
549	/*
550	* If this is a still-connecting socket, this probably
551	* a retransmit of the SYN. Whether it's a retransmit SYN
552	* or something else, we nuke it.
553	*/
554	if (so->so_state & SS_ISFCONNECTING)
555	{
556	LogFlowFunc(("%d -> drop\n", __LINE__));
557	goto drop;
558	}
559
560	tp = sototcpcb(so);
561
562	/* XXX Should never fail */
563	if (tp == 0)
564	{
565	LogFlowFunc(("%d -> dropwithreset\n", __LINE__));
566	goto dropwithreset;
567	}
568	if (tp->t_state == TCPS_CLOSED)
569	{
570	LogFlowFunc(("%d -> drop\n", __LINE__));
571	goto drop;
572	}
573
574	/* Unscale the window into a 32-bit value. */
575	/* if ((tiflags & TH_SYN) == 0)
576	* tiwin = ti->ti_win << tp->snd_scale;
577	* else
578	*/
579	tiwin = ti->ti_win;
580
581	/*
582	* Segment received on connection.
583	* Reset idle time and keep-alive timer.
584	*/
585	tp->t_idle = 0;
586	if (so_options)
587	tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
588	else
589	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
590
591	/*
592	* Process options if not in LISTEN state,
593	* else do it below (after getting remote address).
594	*/
595	if (optp && tp->t_state != TCPS_LISTEN)
596	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
597	/* , */
598	/* &ts_present, &ts_val, &ts_ecr); */
599
600	/*
601	* Header prediction: check for the two common cases
602	* of a uni-directional data xfer. If the packet has
603	* no control flags, is in-sequence, the window didn't
604	* change and we're not retransmitting, it's a
605	* candidate. If the length is zero and the ack moved
606	* forward, we're the sender side of the xfer. Just
607	* free the data acked & wake any higher level process
608	* that was blocked waiting for space. If the length
609	* is non-zero and the ack didn't move, we're the
610	* receiver side. If we're getting packets in-order
611	* (the reassembly queue is empty), add the data to
612	* the socket buffer and note that we need a delayed ack.
613	*
614	* XXX Some of these tests are not needed
615	* eg: the tiwin == tp->snd_wnd prevents many more
616	* predictions.. with no real advantage..
617	*/
618	if ( tp->t_state == TCPS_ESTABLISHED
619	&& (tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK
620	/* && (!ts_present \|\| TSTMP_GEQ(ts_val, tp->ts_recent)) */
621	&& ti->ti_seq == tp->rcv_nxt
622	&& tiwin && tiwin == tp->snd_wnd
623	&& tp->snd_nxt == tp->snd_max)
624	{
625	/*
626	* If last ACK falls within this segment's sequence numbers,
627	* record the timestamp.
628	*/
629	#if 0
630	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
631	SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len))
632	{
633	tp->ts_recent_age = tcp_now;
634	tp->ts_recent = ts_val;
635	}
636	#endif
637
638	if (ti->ti_len == 0)
639	{
640	if ( SEQ_GT(ti->ti_ack, tp->snd_una)
641	&& SEQ_LEQ(ti->ti_ack, tp->snd_max)
642	&& tp->snd_cwnd >= tp->snd_wnd)
643	{
644	/*
645	* this is a pure ack for outstanding data.
646	*/
647	++tcpstat.tcps_predack;
648	#if 0
649	if (ts_present)
650	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
651	else
652	#endif
653	if ( tp->t_rtt
654	&& SEQ_GT(ti->ti_ack, tp->t_rtseq))
655	tcp_xmit_timer(pData, tp, tp->t_rtt);
656	acked = ti->ti_ack - tp->snd_una;
657	tcpstat.tcps_rcvackpack++;
658	tcpstat.tcps_rcvackbyte += acked;
659	sbdrop(&so->so_snd, acked);
660	tp->snd_una = ti->ti_ack;
661	m_freem(pData, m);
662
663	/*
664	* If all outstanding data are acked, stop
665	* retransmit timer, otherwise restart timer
666	* using current (possibly backed-off) value.
667	* If process is waiting for space,
668	* wakeup/selwakeup/signal. If data
669	* are ready to send, let tcp_output
670	* decide between more output or persist.
671	*/
672	if (tp->snd_una == tp->snd_max)
673	tp->t_timer[TCPT_REXMT] = 0;
674	else if (tp->t_timer[TCPT_PERSIST] == 0)
675	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
676
677	/*
678	* There's room in so_snd, sowwakup will read()
679	* from the socket if we can
680	*/
681	#if 0
682	if (so->so_snd.sb_flags & SB_NOTIFY)
683	sowwakeup(so);
684	#endif
685	/*
686	* This is called because sowwakeup might have
687	* put data into so_snd. Since we don't so sowwakeup,
688	* we don't need this.. XXX???
689	*/
690	if (SBUF_LEN(&so->so_snd))
691	(void) tcp_output(pData, tp);
692
693	SOCKET_UNLOCK(so);
694	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
695	return;
696	}
697	}
698	else if ( ti->ti_ack == tp->snd_una
699	&& LIST_EMPTY(&tp->t_segq)
700	&& ti->ti_len <= sbspace(&so->so_rcv))
701	{
702	/*
703	* this is a pure, in-sequence data packet
704	* with nothing on the reassembly queue and
705	* we have enough buffer space to take it.
706	*/
707	++tcpstat.tcps_preddat;
708	tp->rcv_nxt += ti->ti_len;
709	tcpstat.tcps_rcvpack++;
710	tcpstat.tcps_rcvbyte += ti->ti_len;
711	/*
712	* Add data to socket buffer.
713	*/
714	sbappend(pData, so, m);
715
716	/*
717	* XXX This is called when data arrives. Later, check
718	* if we can actually write() to the socket
719	* XXX Need to check? It's be NON_BLOCKING
720	*/
721	/* sorwakeup(so); */
722
723	/*
724	* If this is a short packet, then ACK now - with Nagel
725	* congestion avoidance sender won't send more until
726	* he gets an ACK.
727	*
728	* It is better to not delay acks at all to maximize
729	* TCP throughput. See RFC 2581.
730	*/
731	tp->t_flags \|= TF_ACKNOW;
732	tcp_output(pData, tp);
733	SOCKET_UNLOCK(so);
734	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
735	return;
736	}
737	} /* header prediction */
738	/*
739	* Calculate amount of space in receive window,
740	* and then do TCP input processing.
741	* Receive window is amount of space in rcv queue,
742	* but not less than advertised window.
743	*/
744	{
745	int win;
746	win = sbspace(&so->so_rcv);
747	if (win < 0)
748	win = 0;
749	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
750	}
751
752	switch (tp->t_state)
753	{
754	/*
755	* If the state is LISTEN then ignore segment if it contains an RST.
756	* If the segment contains an ACK then it is bad and send a RST.
757	* If it does not contain a SYN then it is not interesting; drop it.
758	* Don't bother responding if the destination was a broadcast.
759	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
760	* tp->iss, and send a segment:
761	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
762	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
763	* Fill in remote peer address fields if not previously specified.
764	* Enter SYN_RECEIVED state, and process any other fields of this
765	* segment in this state.
766	*/
767	case TCPS_LISTEN:
768	{
769	if (tiflags & TH_RST)
770	{
771	LogFlowFunc(("%d -> drop\n", __LINE__));
772	goto drop;
773	}
774	if (tiflags & TH_ACK)
775	{
776	LogFlowFunc(("%d -> dropwithreset\n", __LINE__));
777	goto dropwithreset;
778	}
779	if ((tiflags & TH_SYN) == 0)
780	{
781	LogFlowFunc(("%d -> drop\n", __LINE__));
782	goto drop;
783	}
784
785	/*
786	* This has way too many gotos...
787	* But a bit of spaghetti code never hurt anybody :)
788	*/
789	if ( (tcp_fconnect(pData, so) == -1)
790	&& errno != EINPROGRESS
791	&& errno != EWOULDBLOCK)
792	{
793	u_char code = ICMP_UNREACH_NET;
794	Log2((" tcp fconnect errno = %d (%s)\n", errno, strerror(errno)));
795	if (errno == ECONNREFUSED)
796	{
797	/* ACK the SYN, send RST to refuse the connection */
798	tcp_respond(pData, tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
799	TH_RST\|TH_ACK);
800	}
801	else
802	{
803	if (errno == EHOSTUNREACH)
804	code = ICMP_UNREACH_HOST;
805	HTONL(ti->ti_seq); /* restore tcp header */
806	HTONL(ti->ti_ack);
807	HTONS(ti->ti_win);
808	HTONS(ti->ti_urp);
809	m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
810	m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
811	ip = save_ip;
812	icmp_error(pData, m, ICMP_UNREACH, code, 0, strerror(errno));
813	tp->t_socket->so_m = NULL;
814	}
815	tp = tcp_close(pData, tp);
816	}
817	else
818	{
819	/*
820	* Haven't connected yet, save the current mbuf
821	* and ti, and return
822	* XXX Some OS's don't tell us whether the connect()
823	* succeeded or not. So we must time it out.
824	*/
825	so->so_m = m;
826	so->so_ti = ti;
827	so->so_ohdr = RTMemDup(ohdr, ohdrlen);
828	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
829	TCP_STATE_SWITCH_TO(tp, TCPS_SYN_RECEIVED);
830	}
831	SOCKET_UNLOCK(so);
832	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
833	LogFlowFuncLeave();
834	return;
835
836	cont_conn:
837	/* m==NULL
838	* Check if the connect succeeded
839	*/
840	LogFlowFunc(("cont_conn:\n"));
841	if (so->so_state & SS_NOFDREF)
842	{
843	tp = tcp_close(pData, tp);
844	LogFlowFunc(("%d -> dropwithreset\n", __LINE__));
845	goto dropwithreset;
846	}
847
848	tcp_template(tp);
849
850	if (optp)
851	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
852
853	if (iss)
854	tp->iss = iss;
855	else
856	tp->iss = tcp_iss;
857	tcp_iss += TCP_ISSINCR/2;
858	tp->irs = ti->ti_seq;
859	tcp_sendseqinit(tp);
860	tcp_rcvseqinit(tp);
861	tp->t_flags \|= TF_ACKNOW;
862	TCP_STATE_SWITCH_TO(tp, TCPS_SYN_RECEIVED);
863	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
864	tcpstat.tcps_accepts++;
865	LogFlowFunc(("%d -> trimthenstep6\n", __LINE__));
866	goto trimthenstep6;
867	} /* case TCPS_LISTEN */
868
869	/*
870	* If the state is SYN_SENT:
871	* if seg contains an ACK, but not for our SYN, drop the input.
872	* if seg contains a RST, then drop the connection.
873	* if seg does not contain SYN, then drop it.
874	* Otherwise this is an acceptable SYN segment
875	* initialize tp->rcv_nxt and tp->irs
876	* if seg contains ack then advance tp->snd_una
877	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
878	* arrange for segment to be acked (eventually)
879	* continue processing rest of data/controls, beginning with URG
880	*/
881	case TCPS_SYN_SENT:
882	if ( (tiflags & TH_ACK)
883	&& ( SEQ_LEQ(ti->ti_ack, tp->iss)
884	\|\| SEQ_GT(ti->ti_ack, tp->snd_max)))
885	{
886	LogFlowFunc(("%d -> dropwithreset\n", __LINE__));
887	goto dropwithreset;
888	}
889
890	if (tiflags & TH_RST)
891	{
892	if (tiflags & TH_ACK)
893	tp = tcp_drop(pData, tp, 0); /* XXX Check t_softerror! */
894	LogFlowFunc(("%d -> drop\n", __LINE__));
895	goto drop;
896	}
897
898	if ((tiflags & TH_SYN) == 0)
899	{
900	LogFlowFunc(("%d -> drop\n", __LINE__));
901	goto drop;
902	}
903	if (tiflags & TH_ACK)
904	{
905	tp->snd_una = ti->ti_ack;
906	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
907	tp->snd_nxt = tp->snd_una;
908	}
909
910	tp->t_timer[TCPT_REXMT] = 0;
911	tp->irs = ti->ti_seq;
912	tcp_rcvseqinit(tp);
913	tp->t_flags \|= TF_ACKNOW;
914	if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss))
915	{
916	tcpstat.tcps_connects++;
917	soisfconnected(so);
918	TCP_STATE_SWITCH_TO(tp, TCPS_ESTABLISHED);
919
920	/* Do window scaling on this connection? */
921	#if 0
922	if (( tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
923	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
924	{
925	tp->snd_scale = tp->requested_s_scale;
926	tp->rcv_scale = tp->request_r_scale;
927	}
928	#endif
929	(void) tcp_reass(pData, tp, (struct tcphdr )0, NULL, (struct mbuf )0);
930	/*
931	* if we didn't have to retransmit the SYN,
932	* use its rtt as our initial srtt & rtt var.
933	*/
934	if (tp->t_rtt)
935	tcp_xmit_timer(pData, tp, tp->t_rtt);
936	}
937	else
938	TCP_STATE_SWITCH_TO(tp, TCPS_SYN_RECEIVED);
939
940	trimthenstep6:
941	LogFlowFunc(("trimthenstep6:\n"));
942	/*
943	* Advance ti->ti_seq to correspond to first data byte.
944	* If data, trim to stay within window,
945	* dropping FIN if necessary.
946	*/
947	ti->ti_seq++;
948	if (ti->ti_len > tp->rcv_wnd)
949	{
950	todrop = ti->ti_len - tp->rcv_wnd;
951	m_adj(m, -todrop);
952	ti->ti_len = tp->rcv_wnd;
953	tiflags &= ~TH_FIN;
954	tcpstat.tcps_rcvpackafterwin++;
955	tcpstat.tcps_rcvbyteafterwin += todrop;
956	}
957	tp->snd_wl1 = ti->ti_seq - 1;
958	tp->rcv_up = ti->ti_seq;
959	LogFlowFunc(("%d -> step6\n", __LINE__));
960	goto step6;
961	} /* switch tp->t_state */
962	/*
963	* States other than LISTEN or SYN_SENT.
964	* First check timestamp, if present.
965	* Then check that at least some bytes of segment are within
966	* receive window. If segment begins before rcv_nxt,
967	* drop leading data (and SYN); if nothing left, just ack.
968	*
969	* RFC 1323 PAWS: If we have a timestamp reply on this segment
970	* and it's less than ts_recent, drop it.
971	*/
972	#if 0
973	if ( ts_present
974	&& (tiflags & TH_RST) == 0
975	&& tp->ts_recent
976	&& TSTMP_LT(ts_val, tp->ts_recent))
977	{
978	/* Check to see if ts_recent is over 24 days old. */
979	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE)
980	{
981	/*
982	* Invalidate ts_recent. If this segment updates
983	* ts_recent, the age will be reset later and ts_recent
984	* will get a valid value. If it does not, setting
985	* ts_recent to zero will at least satisfy the
986	* requirement that zero be placed in the timestamp
987	* echo reply when ts_recent isn't valid. The
988	* age isn't reset until we get a valid ts_recent
989	* because we don't want out-of-order segments to be
990	* dropped when ts_recent is old.
991	*/
992	tp->ts_recent = 0;
993	}
994	else
995	{
996	tcpstat.tcps_rcvduppack++;
997	tcpstat.tcps_rcvdupbyte += ti->ti_len;
998	tcpstat.tcps_pawsdrop++;
999	goto dropafterack;
1000	}
1001	}
1002	#endif
1003
1004	todrop = tp->rcv_nxt - ti->ti_seq;
1005	if (todrop > 0)
1006	{
1007	if (tiflags & TH_SYN)
1008	{
1009	tiflags &= ~TH_SYN;
1010	ti->ti_seq++;
1011	if (ti->ti_urp > 1)
1012	ti->ti_urp--;
1013	else
1014	tiflags &= ~TH_URG;
1015	todrop--;
1016	}
1017	/*
1018	* Following if statement from Stevens, vol. 2, p. 960.
1019	*/
1020	if ( todrop > ti->ti_len
1021	\|\| ( todrop == ti->ti_len
1022	&& (tiflags & TH_FIN) == 0))
1023	{
1024	/*
1025	* Any valid FIN must be to the left of the window.
1026	* At this point the FIN must be a duplicate or out
1027	* of sequence; drop it.
1028	*/
1029	tiflags &= ~TH_FIN;
1030
1031	/*
1032	* Send an ACK to resynchronize and drop any data.
1033	* But keep on processing for RST or ACK.
1034	*/
1035	tp->t_flags \|= TF_ACKNOW;
1036	todrop = ti->ti_len;
1037	tcpstat.tcps_rcvduppack++;
1038	tcpstat.tcps_rcvdupbyte += todrop;
1039	}
1040	else
1041	{
1042	tcpstat.tcps_rcvpartduppack++;
1043	tcpstat.tcps_rcvpartdupbyte += todrop;
1044	}
1045	m_adj(m, todrop);
1046	ti->ti_seq += todrop;
1047	ti->ti_len -= todrop;
1048	if (ti->ti_urp > todrop)
1049	ti->ti_urp -= todrop;
1050	else
1051	{
1052	tiflags &= ~TH_URG;
1053	ti->ti_urp = 0;
1054	}
1055	}
1056	/*
1057	* If new data are received on a connection after the
1058	* user processes are gone, then RST the other end.
1059	*/
1060	if ( (so->so_state & SS_NOFDREF)
1061	&& tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len)
1062	{
1063	tp = tcp_close(pData, tp);
1064	tcpstat.tcps_rcvafterclose++;
1065	LogFlowFunc(("%d -> dropwithreset\n", __LINE__));
1066	goto dropwithreset;
1067	}
1068
1069	/*
1070	* If segment ends after window, drop trailing data
1071	* (and PUSH and FIN); if nothing left, just ACK.
1072	*/
1073	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1074	if (todrop > 0)
1075	{
1076	tcpstat.tcps_rcvpackafterwin++;
1077	if (todrop >= ti->ti_len)
1078	{
1079	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1080	/*
1081	* If a new connection request is received
1082	* while in TIME_WAIT, drop the old connection
1083	* and start over if the sequence numbers
1084	* are above the previous ones.
1085	*/
1086	if ( tiflags & TH_SYN
1087	&& tp->t_state == TCPS_TIME_WAIT
1088	&& SEQ_GT(ti->ti_seq, tp->rcv_nxt))
1089	{
1090	iss = tp->rcv_nxt + TCP_ISSINCR;
1091	tp = tcp_close(pData, tp);
1092	SOCKET_UNLOCK(tp->t_socket);
1093	LogFlowFunc(("%d -> findso\n", __LINE__));
1094	goto findso;
1095	}
1096	/*
1097	* If window is closed can only take segments at
1098	* window edge, and have to drop data and PUSH from
1099	* incoming segments. Continue processing, but
1100	* remember to ack. Otherwise, drop segment
1101	* and ack.
1102	*/
1103	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt)
1104	{
1105	tp->t_flags \|= TF_ACKNOW;
1106	tcpstat.tcps_rcvwinprobe++;
1107	}
1108	else
1109	{
1110	LogFlowFunc(("%d -> dropafterack\n", __LINE__));
1111	goto dropafterack;
1112	}
1113	}
1114	else
1115	tcpstat.tcps_rcvbyteafterwin += todrop;
1116	m_adj(m, -todrop);
1117	ti->ti_len -= todrop;
1118	tiflags &= ~(TH_PUSH\|TH_FIN);
1119	}
1120
1121	/*
1122	* If last ACK falls within this segment's sequence numbers,
1123	* record its timestamp.
1124	*/
1125	#if 0
1126	if ( ts_present
1127	&& SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)
1128	&& SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + ((tiflags & (TH_SYN\|TH_FIN)) != 0)))
1129	{
1130	tp->ts_recent_age = tcp_now;
1131	tp->ts_recent = ts_val;
1132	}
1133	#endif
1134
1135	/*
1136	* If the RST bit is set examine the state:
1137	* SYN_RECEIVED STATE:
1138	* If passive open, return to LISTEN state.
1139	* If active open, inform user that connection was refused.
1140	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1141	* Inform user that connection was reset, and close tcb.
1142	* CLOSING, LAST_ACK, TIME_WAIT STATES
1143	* Close the tcb.
1144	*/
1145	if (tiflags&TH_RST)
1146	switch (tp->t_state)
1147	{
1148	case TCPS_SYN_RECEIVED:
1149	/* so->so_error = ECONNREFUSED; */
1150	LogFlowFunc(("%d -> close\n", __LINE__));
1151	goto close;
1152
1153	case TCPS_ESTABLISHED:
1154	case TCPS_FIN_WAIT_1:
1155	case TCPS_FIN_WAIT_2:
1156	case TCPS_CLOSE_WAIT:
1157	/* so->so_error = ECONNRESET; */
1158	close:
1159	LogFlowFunc(("close:\n"));
1160	TCP_STATE_SWITCH_TO(tp, TCPS_CLOSED);
1161	tcpstat.tcps_drops++;
1162	tp = tcp_close(pData, tp);
1163	LogFlowFunc(("%d -> drop\n", __LINE__));
1164	goto drop;
1165
1166	case TCPS_CLOSING:
1167	case TCPS_LAST_ACK:
1168	case TCPS_TIME_WAIT:
1169	tp = tcp_close(pData, tp);
1170	LogFlowFunc(("%d -> drop\n", __LINE__));
1171	goto drop;
1172	}
1173
1174	/*
1175	* If a SYN is in the window, then this is an
1176	* error and we send an RST and drop the connection.
1177	*/
1178	if (tiflags & TH_SYN)
1179	{
1180	tp = tcp_drop(pData, tp, 0);
1181	LogFlowFunc(("%d -> dropwithreset\n", __LINE__));
1182	goto dropwithreset;
1183	}
1184
1185	/*
1186	* If the ACK bit is off we drop the segment and return.
1187	*/
1188	if ((tiflags & TH_ACK) == 0)
1189	{
1190	LogFlowFunc(("%d -> drop\n", __LINE__));
1191	goto drop;
1192	}
1193
1194	/*
1195	* Ack processing.
1196	*/
1197	switch (tp->t_state)
1198	{
1199	/*
1200	* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1201	* ESTABLISHED state and continue processing, otherwise
1202	* send an RST. una<=ack<=max
1203	*/
1204	case TCPS_SYN_RECEIVED:
1205	LogFlowFunc(("%d -> TCPS_SYN_RECEIVED\n", __LINE__));
1206	if ( SEQ_GT(tp->snd_una, ti->ti_ack)
1207	\|\| SEQ_GT(ti->ti_ack, tp->snd_max))
1208	goto dropwithreset;
1209	tcpstat.tcps_connects++;
1210	TCP_STATE_SWITCH_TO(tp, TCPS_ESTABLISHED);
1211	/*
1212	* The sent SYN is ack'ed with our sequence number +1
1213	* The first data byte already in the buffer will get
1214	* lost if no correction is made. This is only needed for
1215	* SS_CTL since the buffer is empty otherwise.
1216	* tp->snd_una++; or:
1217	*/
1218	tp->snd_una = ti->ti_ack;
1219	soisfconnected(so);
1220
1221	/* Do window scaling? */
1222	#if 0
1223	if ( (tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
1224	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
1225	{
1226	tp->snd_scale = tp->requested_s_scale;
1227	tp->rcv_scale = tp->request_r_scale;
1228	}
1229	#endif
1230	(void) tcp_reass(pData, tp, (struct tcphdr )0, (int )0, (struct mbuf *)0);
1231	tp->snd_wl1 = ti->ti_seq - 1;
1232	/* Avoid ack processing; snd_una==ti_ack => dup ack */
1233	LogFlowFunc(("%d -> synrx_to_est\n", __LINE__));
1234	goto synrx_to_est;
1235	/* fall into ... */
1236
1237	/*
1238	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1239	* ACKs. If the ack is in the range
1240	* tp->snd_una < ti->ti_ack <= tp->snd_max
1241	* then advance tp->snd_una to ti->ti_ack and drop
1242	* data from the retransmission queue. If this ACK reflects
1243	* more up to date window information we update our window information.
1244	*/
1245	case TCPS_ESTABLISHED:
1246	case TCPS_FIN_WAIT_1:
1247	case TCPS_FIN_WAIT_2:
1248	case TCPS_CLOSE_WAIT:
1249	case TCPS_CLOSING:
1250	case TCPS_LAST_ACK:
1251	case TCPS_TIME_WAIT:
1252	LogFlowFunc(("%d -> TCPS_ESTABLISHED\|TCPS_FIN_WAIT_1\|TCPS_FIN_WAIT_2\|TCPS_CLOSE_WAIT\|"
1253	"TCPS_CLOSING\|TCPS_LAST_ACK\|TCPS_TIME_WAIT\n", __LINE__));
1254	if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
1255	{
1256	if (ti->ti_len == 0 && tiwin == tp->snd_wnd)
1257	{
1258	tcpstat.tcps_rcvdupack++;
1259	Log2((" dup ack m = %p, so = %p\n", m, so));
1260	/*
1261	* If we have outstanding data (other than
1262	* a window probe), this is a completely
1263	* duplicate ack (ie, window info didn't
1264	* change), the ack is the biggest we've
1265	* seen and we've seen exactly our rexmt
1266	* threshold of them, assume a packet
1267	* has been dropped and retransmit it.
1268	* Kludge snd_nxt & the congestion
1269	* window so we send only this one
1270	* packet.
1271	*
1272	* We know we're losing at the current
1273	* window size so do congestion avoidance
1274	* (set ssthresh to half the current window
1275	* and pull our congestion window back to
1276	* the new ssthresh).
1277	*
1278	* Dup acks mean that packets have left the
1279	* network (they're now cached at the receiver)
1280	* so bump cwnd by the amount in the receiver
1281	* to keep a constant cwnd packets in the
1282	* network.
1283	*/
1284	if ( tp->t_timer[TCPT_REXMT] == 0
1285	\|\| ti->ti_ack != tp->snd_una)
1286	tp->t_dupacks = 0;
1287	else if (++tp->t_dupacks == tcprexmtthresh)
1288	{
1289	tcp_seq onxt = tp->snd_nxt;
1290	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
1291	if (win < 2)
1292	win = 2;
1293	tp->snd_ssthresh = win * tp->t_maxseg;
1294	tp->t_timer[TCPT_REXMT] = 0;
1295	tp->t_rtt = 0;
1296	tp->snd_nxt = ti->ti_ack;
1297	tp->snd_cwnd = tp->t_maxseg;
1298	(void) tcp_output(pData, tp);
1299	tp->snd_cwnd = tp->snd_ssthresh +
1300	tp->t_maxseg * tp->t_dupacks;
1301	if (SEQ_GT(onxt, tp->snd_nxt))
1302	tp->snd_nxt = onxt;
1303	LogFlowFunc(("%d -> drop\n", __LINE__));
1304	goto drop;
1305	}
1306	else if (tp->t_dupacks > tcprexmtthresh)
1307	{
1308	tp->snd_cwnd += tp->t_maxseg;
1309	(void) tcp_output(pData, tp);
1310	LogFlowFunc(("%d -> drop\n", __LINE__));
1311	goto drop;
1312	}
1313	}
1314	else
1315	tp->t_dupacks = 0;
1316	break;
1317	}
1318	synrx_to_est:
1319	LogFlowFunc(("synrx_to_est:\n"));
1320	/*
1321	* If the congestion window was inflated to account
1322	* for the other side's cached packets, retract it.
1323	*/
1324	if ( tp->t_dupacks > tcprexmtthresh
1325	&& tp->snd_cwnd > tp->snd_ssthresh)
1326	tp->snd_cwnd = tp->snd_ssthresh;
1327	tp->t_dupacks = 0;
1328	if (SEQ_GT(ti->ti_ack, tp->snd_max))
1329	{
1330	tcpstat.tcps_rcvacktoomuch++;
1331	LogFlowFunc(("%d -> dropafterack\n", __LINE__));
1332	goto dropafterack;
1333	}
1334	acked = ti->ti_ack - tp->snd_una;
1335	tcpstat.tcps_rcvackpack++;
1336	tcpstat.tcps_rcvackbyte += acked;
1337
1338	/*
1339	* If we have a timestamp reply, update smoothed
1340	* round trip time. If no timestamp is present but
1341	* transmit timer is running and timed sequence
1342	* number was acked, update smoothed round trip time.
1343	* Since we now have an rtt measurement, cancel the
1344	* timer backoff (cf., Phil Karn's retransmit alg.).
1345	* Recompute the initial retransmit timer.
1346	*/
1347	#if 0
1348	if (ts_present)
1349	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1350	else
1351	#endif
1352	if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1353	tcp_xmit_timer(pData, tp, tp->t_rtt);
1354
1355	/*
1356	* If all outstanding data is acked, stop retransmit
1357	* timer and remember to restart (more output or persist).
1358	* If there is more data to be acked, restart retransmit
1359	* timer, using current (possibly backed-off) value.
1360	*/
1361	if (ti->ti_ack == tp->snd_max)
1362	{
1363	tp->t_timer[TCPT_REXMT] = 0;
1364	needoutput = 1;
1365	}
1366	else if (tp->t_timer[TCPT_PERSIST] == 0)
1367	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1368	/*
1369	* When new data is acked, open the congestion window.
1370	* If the window gives us less than ssthresh packets
1371	* in flight, open exponentially (maxseg per packet).
1372	* Otherwise open linearly: maxseg per window
1373	* (maxseg^2 / cwnd per packet).
1374	*/
1375	{
1376	register u_int cw = tp->snd_cwnd;
1377	register u_int incr = tp->t_maxseg;
1378
1379	if (cw > tp->snd_ssthresh)
1380	incr = incr * incr / cw;
1381	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1382	}
1383	if (acked > SBUF_LEN(&so->so_snd))
1384	{
1385	tp->snd_wnd -= SBUF_LEN(&so->so_snd);
1386	sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1387	ourfinisacked = 1;
1388	}
1389	else
1390	{
1391	sbdrop(&so->so_snd, acked);
1392	tp->snd_wnd -= acked;
1393	ourfinisacked = 0;
1394	}
1395	/*
1396	* XXX sowwakup is called when data is acked and there's room for
1397	* for more data... it should read() the socket
1398	*/
1399	#if 0
1400	if (so->so_snd.sb_flags & SB_NOTIFY)
1401	sowwakeup(so);
1402	#endif
1403	tp->snd_una = ti->ti_ack;
1404	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1405	tp->snd_nxt = tp->snd_una;
1406
1407	switch (tp->t_state)
1408	{
1409	/*
1410	* In FIN_WAIT_1 STATE in addition to the processing
1411	* for the ESTABLISHED state if our FIN is now acknowledged
1412	* then enter FIN_WAIT_2.
1413	*/
1414	case TCPS_FIN_WAIT_1:
1415	if (ourfinisacked)
1416	{
1417	/*
1418	* If we can't receive any more
1419	* data, then closing user can proceed.
1420	* Starting the timer is contrary to the
1421	* specification, but if we don't get a FIN
1422	* we'll hang forever.
1423	*/
1424	if (so->so_state & SS_FCANTRCVMORE)
1425	{
1426	soisfdisconnected(so);
1427	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1428	}
1429	TCP_STATE_SWITCH_TO(tp, TCPS_FIN_WAIT_2);
1430	}
1431	break;
1432
1433	/*
1434	* In CLOSING STATE in addition to the processing for
1435	* the ESTABLISHED state if the ACK acknowledges our FIN
1436	* then enter the TIME-WAIT state, otherwise ignore
1437	* the segment.
1438	*/
1439	case TCPS_CLOSING:
1440	if (ourfinisacked)
1441	{
1442	TCP_STATE_SWITCH_TO(tp, TCPS_TIME_WAIT);
1443	tcp_canceltimers(tp);
1444	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1445	soisfdisconnected(so);
1446	}
1447	break;
1448
1449	/*
1450	* In LAST_ACK, we may still be waiting for data to drain
1451	* and/or to be acked, as well as for the ack of our FIN.
1452	* If our FIN is now acknowledged, delete the TCB,
1453	* enter the closed state and return.
1454	*/
1455	case TCPS_LAST_ACK:
1456	if (ourfinisacked)
1457	{
1458	tp = tcp_close(pData, tp);
1459	LogFlowFunc(("%d -> drop\n", __LINE__));
1460	goto drop;
1461	}
1462	break;
1463
1464	/*
1465	* In TIME_WAIT state the only thing that should arrive
1466	* is a retransmission of the remote FIN. Acknowledge
1467	* it and restart the finack timer.
1468	*/
1469	case TCPS_TIME_WAIT:
1470	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1471	LogFlowFunc(("%d -> dropafterack\n", __LINE__));
1472	goto dropafterack;
1473	}
1474	} /* switch(tp->t_state) */
1475
1476	step6:
1477	LogFlowFunc(("step6:\n"));
1478	/*
1479	* Update window information.
1480	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1481	*/
1482	if ( (tiflags & TH_ACK)
1483	&& ( SEQ_LT(tp->snd_wl1, ti->ti_seq)
1484	\|\| ( tp->snd_wl1 == ti->ti_seq
1485	&& ( SEQ_LT(tp->snd_wl2, ti->ti_ack)
1486	\|\| ( tp->snd_wl2 == ti->ti_ack
1487	&& tiwin > tp->snd_wnd)))))
1488	{
1489	/* keep track of pure window updates */
1490	if ( ti->ti_len == 0
1491	&& tp->snd_wl2 == ti->ti_ack
1492	&& tiwin > tp->snd_wnd)
1493	tcpstat.tcps_rcvwinupd++;
1494	tp->snd_wnd = tiwin;
1495	tp->snd_wl1 = ti->ti_seq;
1496	tp->snd_wl2 = ti->ti_ack;
1497	if (tp->snd_wnd > tp->max_sndwnd)
1498	tp->max_sndwnd = tp->snd_wnd;
1499	needoutput = 1;
1500	}
1501
1502	/*
1503	* Process segments with URG.
1504	*/
1505	if ((tiflags & TH_URG) && ti->ti_urp &&
1506	TCPS_HAVERCVDFIN(tp->t_state) == 0)
1507	{
1508	/*
1509	* This is a kludge, but if we receive and accept
1510	* random urgent pointers, we'll crash in
1511	* soreceive. It's hard to imagine someone
1512	* actually wanting to send this much urgent data.
1513	*/
1514	if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen)
1515	{
1516	ti->ti_urp = 0;
1517	tiflags &= ~TH_URG;
1518	LogFlowFunc(("%d -> dodata\n", __LINE__));
1519	goto dodata;
1520	}
1521
1522	/*
1523	* If this segment advances the known urgent pointer,
1524	* then mark the data stream. This should not happen
1525	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1526	* a FIN has been received from the remote side.
1527	* In these states we ignore the URG.
1528	*
1529	* According to RFC961 (Assigned Protocols),
1530	* the urgent pointer points to the last octet
1531	* of urgent data. We continue, however,
1532	* to consider it to indicate the first octet
1533	* of data past the urgent section as the original
1534	* spec states (in one of two places).
1535	*/
1536	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up))
1537	{
1538	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1539	so->so_urgc = SBUF_LEN(&so->so_rcv) +
1540	(tp->rcv_up - tp->rcv_nxt); /* -1; */
1541	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1542	}
1543	}
1544	else
1545	/*
1546	* If no out of band data is expected,
1547	* pull receive urgent pointer along
1548	* with the receive window.
1549	*/
1550	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1551	tp->rcv_up = tp->rcv_nxt;
1552	dodata:
1553	LogFlowFunc(("dodata:\n"));
1554
1555	/*
1556	* If this is a small packet, then ACK now - with Nagel
1557	* congestion avoidance sender won't send more until
1558	* he gets an ACK.
1559	*
1560	* XXX: In case you wonder... The magic "27" below is ESC that
1561	* presumably starts a terminal escape-sequence and that we want
1562	* to ACK ASAP. [Original slirp code had three different
1563	* heuristics to chose from here and in the header prediction case
1564	* above, but the commented out alternatives were lost and the
1565	* header prediction case that had an expanded comment about this
1566	* has been modified to always send an ACK].
1567	*/
1568	if ( ti->ti_len
1569	&& (unsigned)ti->ti_len <= 5
1570	&& ((struct tcpiphdr_2 *)ti)->first_char == (char)27)
1571	{
1572	tp->t_flags \|= TF_ACKNOW;
1573	}
1574
1575	/*
1576	* Process the segment text, merging it into the TCP sequencing queue,
1577	* and arranging for acknowledgment of receipt if necessary.
1578	* This process logically involves adjusting tp->rcv_wnd as data
1579	* is presented to the user (this happens in tcp_usrreq.c,
1580	* case PRU_RCVD). If a FIN has already been received on this
1581	* connection then we just ignore the text.
1582	*/
1583	if ( (ti->ti_len \|\| (tiflags&TH_FIN))
1584	&& TCPS_HAVERCVDFIN(tp->t_state) == 0)
1585	{
1586	if ( ti->ti_seq == tp->rcv_nxt
1587	&& LIST_EMPTY(&tp->t_segq)
1588	&& tp->t_state == TCPS_ESTABLISHED)
1589	{
1590	DELAY_ACK(tp, ti); /* little bit different from BSD declaration see netinet/tcp_input.c */
1591	tp->rcv_nxt += tlen;
1592	tiflags = ti->ti_t.th_flags & TH_FIN;
1593	tcpstat.tcps_rcvpack++;
1594	tcpstat.tcps_rcvbyte += tlen;
1595	if (so->so_state & SS_FCANTRCVMORE)
1596	m_freem(pData, m);
1597	else
1598	sbappend(pData, so, m);
1599	}
1600	else
1601	{
1602	tiflags = tcp_reass(pData, tp, &ti->ti_t, &tlen, m);
1603	tp->t_flags \|= TF_ACKNOW;
1604	}
1605	/*
1606	* Note the amount of data that peer has sent into
1607	* our window, in order to estimate the sender's
1608	* buffer size.
1609	*/
1610	len = SBUF_SIZE(&so->so_rcv) - (tp->rcv_adv - tp->rcv_nxt);
1611	}
1612	else
1613	{
1614	m_freem(pData, m);
1615	tiflags &= ~TH_FIN;
1616	}
1617
1618	/*
1619	* If FIN is received ACK the FIN and let the user know
1620	* that the connection is closing.
1621	*/
1622	if (tiflags & TH_FIN)
1623	{
1624	if (TCPS_HAVERCVDFIN(tp->t_state) == 0)
1625	{
1626	/*
1627	* If we receive a FIN we can't send more data,
1628	* set it SS_FDRAIN
1629	* Shutdown the socket if there is no rx data in the
1630	* buffer.
1631	* soread() is called on completion of shutdown() and
1632	* will got to TCPS_LAST_ACK, and use tcp_output()
1633	* to send the FIN.
1634	*/
1635	/* sofcantrcvmore(so); */
1636	sofwdrain(so);
1637
1638	tp->t_flags \|= TF_ACKNOW;
1639	tp->rcv_nxt++;
1640	}
1641	switch (tp->t_state)
1642	{
1643	/*
1644	* In SYN_RECEIVED and ESTABLISHED STATES
1645	* enter the CLOSE_WAIT state.
1646	*/
1647	case TCPS_SYN_RECEIVED:
1648	case TCPS_ESTABLISHED:
1649	TCP_STATE_SWITCH_TO(tp, TCPS_CLOSE_WAIT);
1650	break;
1651
1652	/*
1653	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1654	* enter the CLOSING state.
1655	*/
1656	case TCPS_FIN_WAIT_1:
1657	TCP_STATE_SWITCH_TO(tp, TCPS_CLOSING);
1658	break;
1659
1660	/*
1661	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1662	* starting the time-wait timer, turning off the other
1663	* standard timers.
1664	*/
1665	case TCPS_FIN_WAIT_2:
1666	TCP_STATE_SWITCH_TO(tp, TCPS_TIME_WAIT);
1667	tcp_canceltimers(tp);
1668	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1669	soisfdisconnected(so);
1670	break;
1671
1672	/*
1673	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1674	*/
1675	case TCPS_TIME_WAIT:
1676	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1677	break;
1678	}
1679	}
1680
1681	/*
1682	* Return any desired output.
1683	*/
1684	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1685	tcp_output(pData, tp);
1686
1687	SOCKET_UNLOCK(so);
1688	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1689	LogFlowFuncLeave();
1690	return;
1691
1692	dropafterack:
1693	LogFlowFunc(("dropafterack:\n"));
1694	/*
1695	* Generate an ACK dropping incoming segment if it occupies
1696	* sequence space, where the ACK reflects our state.
1697	*/
1698	if (tiflags & TH_RST)
1699	{
1700	LogFlowFunc(("%d -> drop\n", __LINE__));
1701	goto drop;
1702	}
1703	m_freem(pData, m);
1704	tp->t_flags \|= TF_ACKNOW;
1705	(void) tcp_output(pData, tp);
1706	SOCKET_UNLOCK(so);
1707	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1708	LogFlowFuncLeave();
1709	return;
1710
1711	dropwithreset:
1712	LogFlowFunc(("dropwithreset:\n"));
1713	/* reuses m if m!=NULL, m_free() unnecessary */
1714	if (tiflags & TH_ACK)
1715	tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1716	else
1717	{
1718	if (tiflags & TH_SYN)
1719	ti->ti_len++;
1720	tcp_respond(pData, tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1721	TH_RST\|TH_ACK);
1722	}
1723
1724	if (so != &tcb)
1725	SOCKET_UNLOCK(so);
1726	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1727	LogFlowFuncLeave();
1728	return;
1729
1730	drop:
1731	LogFlowFunc(("drop:\n"));
1732	/*
1733	* Drop space held by incoming segment and return.
1734	*/
1735	m_freem(pData, m);
1736
1737	#ifdef VBOX_WITH_SLIRP_MT
1738	if (RTCritSectIsOwned(&so->so_mutex))
1739	{
1740	SOCKET_UNLOCK(so);
1741	}
1742	#endif
1743
1744	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1745	LogFlowFuncLeave();
1746	return;
1747	}
1748
1749
1750	void
1751	tcp_fconnect_failed(PNATState pData, struct socket *so, int sockerr)
1752	{
1753	struct tcpcb *tp;
1754	int code;
1755
1756	Log2(("NAT: connect error %d %R[natsock]\n", sockerr, so));
1757
1758	Assert(so->so_state & SS_ISFCONNECTING);
1759	so->so_state = SS_NOFDREF;
1760
1761	if (sockerr == ECONNREFUSED \|\| sockerr == ECONNRESET)
1762	{
1763	/* hand off to tcp_input():cont_conn to send RST */
1764	TCP_INPUT(pData, NULL, 0, so);
1765	return;
1766	}
1767
1768	tp = sototcpcb(so);
1769	if (RT_UNLIKELY(tp == NULL)) /* should never happen */
1770	{
1771	LogRel(("NAT: tp == NULL %R[natsock]\n", so));
1772	sofree(pData, so);
1773	return;
1774	}
1775
1776	if (sockerr == ENETUNREACH \|\| sockerr == ENETDOWN)
1777	code = ICMP_UNREACH_NET;
1778	else if (sockerr == EHOSTUNREACH \|\| sockerr == EHOSTDOWN)
1779	code = ICMP_UNREACH_HOST;
1780	else
1781	code = -1;
1782
1783	if (code >= 0)
1784	{
1785	struct ip *oip;
1786	unsigned ohdrlen;
1787	struct mbuf *m;
1788
1789	if (RT_UNLIKELY(so->so_ohdr == NULL))
1790	goto out;
1791
1792	oip = (struct ip *)so->so_ohdr;
1793	ohdrlen = oip->ip_hl * 4 + 8;
1794
1795	m = m_gethdr(pData, M_NOWAIT, MT_HEADER);
1796	if (RT_UNLIKELY(m == NULL))
1797	goto out;
1798
1799	m_copyback(pData, m, 0, ohdrlen, (caddr_t)so->so_ohdr);
1800	m->m_pkthdr.header = mtod(m, void *);
1801
1802	icmp_error(pData, m, ICMP_UNREACH, code, 0, NULL);
1803	}
1804
1805	out:
1806	tcp_close(pData, tp);
1807	}
1808
1809
1810	void
1811	tcp_dooptions(PNATState pData, struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti)
1812	{
1813	u_int16_t mss;
1814	int opt, optlen;
1815
1816	LogFlowFunc(("tcp_dooptions: tp = %R[tcpcb793], cnt=%i\n", tp, cnt));
1817
1818	for (; cnt > 0; cnt -= optlen, cp += optlen)
1819	{
1820	opt = cp[0];
1821	if (opt == TCPOPT_EOL)
1822	break;
1823	if (opt == TCPOPT_NOP)
1824	optlen = 1;
1825	else
1826	{
1827	optlen = cp[1];
1828	if (optlen <= 0)
1829	break;
1830	}
1831	switch (opt)
1832	{
1833	default:
1834	continue;
1835
1836	case TCPOPT_MAXSEG:
1837	if (optlen != TCPOLEN_MAXSEG)
1838	continue;
1839	if (!(ti->ti_flags & TH_SYN))
1840	continue;
1841	memcpy((char ) &mss, (char ) cp + 2, sizeof(mss));
1842	NTOHS(mss);
1843	(void) tcp_mss(pData, tp, mss); /* sets t_maxseg */
1844	break;
1845
1846	#if 0
1847	case TCPOPT_WINDOW:
1848	if (optlen != TCPOLEN_WINDOW)
1849	continue;
1850	if (!(ti->ti_flags & TH_SYN))
1851	continue;
1852	tp->t_flags \|= TF_RCVD_SCALE;
1853	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1854	break;
1855
1856	case TCPOPT_TIMESTAMP:
1857	if (optlen != TCPOLEN_TIMESTAMP)
1858	continue;
1859	*ts_present = 1;
1860	memcpy((char ) ts_val, (char )cp + 2, sizeof(*ts_val));
1861	NTOHL(*ts_val);
1862	memcpy((char ) ts_ecr, (char )cp + 6, sizeof(*ts_ecr));
1863	NTOHL(*ts_ecr);
1864
1865	/*
1866	* A timestamp received in a SYN makes
1867	* it ok to send timestamp requests and replies.
1868	*/
1869	if (ti->ti_flags & TH_SYN)
1870	{
1871	tp->t_flags \|= TF_RCVD_TSTMP;
1872	tp->ts_recent = *ts_val;
1873	tp->ts_recent_age = tcp_now;
1874	}
1875	break;
1876	#endif
1877	}
1878	}
1879	}
1880
1881
1882	/*
1883	* Pull out of band byte out of a segment so
1884	* it doesn't appear in the user's data queue.
1885	* It is still reflected in the segment length for
1886	* sequencing purposes.
1887	*/
1888
1889	#if 0
1890	void
1891	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1892	{
1893	int cnt = ti->ti_urp - 1;
1894
1895	while (cnt >= 0)
1896	{
1897	if (m->m_len > cnt)
1898	{
1899	char *cp = mtod(m, caddr_t) + cnt;
1900	struct tcpcb *tp = sototcpcb(so);
1901
1902	tp->t_iobc = *cp;
1903	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1904	memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1905	m->m_len--;
1906	return;
1907	}
1908	cnt -= m->m_len;
1909	m = m->m_next; /* XXX WRONG! Fix it! */
1910	if (m == 0)
1911	break;
1912	}
1913	panic("tcp_pulloutofband");
1914	}
1915	#endif
1916
1917	/*
1918	* Collect new round-trip time estimate
1919	* and update averages and current timeout.
1920	*/
1921
1922	void
1923	tcp_xmit_timer(PNATState pData, register struct tcpcb *tp, int rtt)
1924	{
1925	register short delta;
1926
1927	LogFlowFunc(("ENTER: tcp_xmit_timer: tp = %R[tcpcb793] rtt = %d\n", tp, rtt));
1928
1929	tcpstat.tcps_rttupdated++;
1930	if (tp->t_srtt != 0)
1931	{
1932	/*
1933	* srtt is stored as fixed point with 3 bits after the
1934	* binary point (i.e., scaled by 8). The following magic
1935	* is equivalent to the smoothing algorithm in rfc793 with
1936	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1937	* point). Adjust rtt to origin 0.
1938	*/
1939	delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1940	if ((tp->t_srtt += delta) <= 0)
1941	tp->t_srtt = 1;
1942	/*
1943	* We accumulate a smoothed rtt variance (actually, a
1944	* smoothed mean difference), then set the retransmit
1945	* timer to smoothed rtt + 4 times the smoothed variance.
1946	* rttvar is stored as fixed point with 2 bits after the
1947	* binary point (scaled by 4). The following is
1948	* equivalent to rfc793 smoothing with an alpha of .75
1949	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1950	* rfc793's wired-in beta.
1951	*/
1952	if (delta < 0)
1953	delta = -delta;
1954	delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1955	if ((tp->t_rttvar += delta) <= 0)
1956	tp->t_rttvar = 1;
1957	}
1958	else
1959	{
1960	/*
1961	* No rtt measurement yet - use the unsmoothed rtt.
1962	* Set the variance to half the rtt (so our first
1963	* retransmit happens at 3*rtt).
1964	*/
1965	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1966	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1967	}
1968	tp->t_rtt = 0;
1969	tp->t_rxtshift = 0;
1970
1971	/*
1972	* the retransmit should happen at rtt + 4 * rttvar.
1973	* Because of the way we do the smoothing, srtt and rttvar
1974	* will each average +1/2 tick of bias. When we compute
1975	* the retransmit timer, we want 1/2 tick of rounding and
1976	* 1 extra tick because of +-1/2 tick uncertainty in the
1977	* firing of the timer. The bias will give us exactly the
1978	* 1.5 tick we need. But, because the bias is
1979	* statistical, we have to test that we don't drop below
1980	* the minimum feasible timer (which is 2 ticks).
1981	*/
1982	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1983	(short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1984
1985	/*
1986	* We received an ack for a packet that wasn't retransmitted;
1987	* it is probably safe to discard any error indications we've
1988	* received recently. This isn't quite right, but close enough
1989	* for now (a route might have failed after we sent a segment,
1990	* and the return path might not be symmetrical).
1991	*/
1992	tp->t_softerror = 0;
1993	}
1994
1995	/*
1996	* Determine a reasonable value for maxseg size.
1997	* If the route is known, check route for mtu.
1998	* If none, use an mss that can be handled on the outgoing
1999	* interface without forcing IP to fragment; if bigger than
2000	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2001	* to utilize large mbufs. If no route is found, route has no mtu,
2002	* or the destination isn't local, use a default, hopefully conservative
2003	* size (usually 512 or the default IP max size, but no more than the mtu
2004	* of the interface), as we can't discover anything about intervening
2005	* gateways or networks. We also initialize the congestion/slow start
2006	* window to be a single segment if the destination isn't local.
2007	* While looking at the routing entry, we also initialize other path-dependent
2008	* parameters from pre-set or cached values in the routing entry.
2009	*/
2010
2011	int
2012	tcp_mss(PNATState pData, register struct tcpcb *tp, u_int offer)
2013	{
2014	struct socket *so = tp->t_socket;
2015	int mss;
2016
2017	LogFlowFunc(("ENTER: tcp_mss: tp = %R[tcpcb793], offer = %d\n", tp, offer));
2018
2019	mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
2020	if (offer)
2021	mss = min(mss, offer);
2022	mss = max(mss, 32);
2023	if (mss < tp->t_maxseg \|\| offer != 0)
2024	tp->t_maxseg = mss;
2025
2026	tp->snd_cwnd = mss;
2027
2028	sbreserve(pData, &so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
2029	sbreserve(pData, &so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
2030
2031	Log2((" returning mss = %d\n", mss));
2032
2033	return mss;
2034	}

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 63215

Download in other formats: