tcp_input.c@ 28480

Last change on this file since 28480 was 28480, checked in by vboxsync, 15 years ago
NAT: nuke dead code.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 62.0 KB

Line
1	/* $Id: tcp_input.c 28480 2010-04-19 16:25:07Z vboxsync $ */
2	/** @file
3	* NAT - TCP input.
4	*/
5
6	/*
7	* Copyright (C) 2006-2010 Sun Microsystems, Inc.
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	* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa
18	* Clara, CA 95054 USA or visit http://www.sun.com if you need
19	* additional information or have any questions.
20	*/
21
22	/*
23	* This code is based on:
24	*
25	* Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994
26	* The Regents of the University of California. All rights reserved.
27	*
28	* Redistribution and use in source and binary forms, with or without
29	* modification, are permitted provided that the following conditions
30	* are met:
31	* 1. Redistributions of source code must retain the above copyright
32	* notice, this list of conditions and the following disclaimer.
33	* 2. Redistributions in binary form must reproduce the above copyright
34	* notice, this list of conditions and the following disclaimer in the
35	* documentation and/or other materials provided with the distribution.
36	* 3. All advertising materials mentioning features or use of this software
37	* must display the following acknowledgement:
38	* This product includes software developed by the University of
39	* California, Berkeley and its contributors.
40	* 4. Neither the name of the University nor the names of its contributors
41	* may be used to endorse or promote products derived from this software
42	* without specific prior written permission.
43	*
44	* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
45	* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
46	* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
47	* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
48	* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
49	* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
50	* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
51	* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
52	* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
53	* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
54	* SUCH DAMAGE.
55	*
56	* @(#)tcp_input.c 8.5 (Berkeley) 4/10/94
57	* tcp_input.c,v 1.10 1994/10/13 18:36:32 wollman Exp
58	*/
59
60	/*
61	* Changes and additions relating to SLiRP
62	* Copyright (c) 1995 Danny Gasparovski.
63	*
64	* Please read the file COPYRIGHT for the
65	* terms and conditions of the copyright.
66	*/
67
68	#include <slirp.h>
69	#include "ip_icmp.h"
70
71
72	#define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ)
73
74	/* for modulo comparisons of timestamps */
75	#define TSTMP_LT(a, b) ((int)((a)-(b)) < 0)
76	#define TSTMP_GEQ(a, b) ((int)((a)-(b)) >= 0)
77
78	#ifndef TCP_ACK_HACK
79	#define DELAY_ACK(tp, ti) \
80	if (ti->ti_flags & TH_PUSH) \
81	tp->t_flags \|= TF_ACKNOW; \
82	else \
83	tp->t_flags \|= TF_DELACK;
84	#else /* !TCP_ACK_HACK */
85	#define DELAY_ACK(tp, ign) \
86	tp->t_flags \|= TF_DELACK;
87	#endif /* TCP_ACK_HACK */
88
89
90	/*
91	* deps: netinet/tcp_reass.c
92	* tcp_reass_maxqlen = 48 (deafault)
93	* tcp_reass_maxseg = nmbclusters/16 (nmbclusters = 1024 + maxusers * 64 from kern/kern_mbuf.c let's say 256)
94	*/
95	int
96	tcp_reass(PNATState pData, struct tcpcb tp, struct tcphdr th, int tlenp, struct mbuf m)
97	{
98	struct tseg_qent *q;
99	struct tseg_qent *p = NULL;
100	struct tseg_qent *nq;
101	struct tseg_qent *te = NULL;
102	struct socket *so = tp->t_socket;
103	int flags;
104	STAM_PROFILE_START(&pData->StatTCP_reassamble, tcp_reassamble);
105
106	/*
107	* XXX: tcp_reass() is rather inefficient with its data structures
108	* and should be rewritten (see NetBSD for optimizations). While
109	* doing that it should move to its own file tcp_reass.c.
110	*/
111
112	/*
113	* Call with th==NULL after become established to
114	* force pre-ESTABLISHED data up to user socket.
115	*/
116	if (th == NULL)
117	goto present;
118
119	/*
120	* Limit the number of segments in the reassembly queue to prevent
121	* holding on to too many segments (and thus running out of mbufs).
122	* Make sure to let the missing segment through which caused this
123	* queue. Always keep one global queue entry spare to be able to
124	* process the missing segment.
125	*/
126	if ( th->th_seq != tp->rcv_nxt
127	&& ( tcp_reass_qsize + 1 >= tcp_reass_maxseg
128	\|\| tp->t_segqlen >= tcp_reass_maxqlen))
129	{
130	tcp_reass_overflows++;
131	tcpstat.tcps_rcvmemdrop++;
132	m_freem(pData, m);
133	*tlenp = 0;
134	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
135	return (0);
136	}
137
138	/*
139	* Allocate a new queue entry. If we can't, or hit the zone limit
140	* just drop the pkt.
141	*/
142	te = RTMemAlloc(sizeof(struct tseg_qent));
143	if (te == NULL)
144	{
145	tcpstat.tcps_rcvmemdrop++;
146	m_freem(pData, m);
147	*tlenp = 0;
148	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
149	return (0);
150	}
151	tp->t_segqlen++;
152	tcp_reass_qsize++;
153
154	/*
155	* Find a segment which begins after this one does.
156	*/
157	LIST_FOREACH(q, &tp->t_segq, tqe_q)
158	{
159	if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
160	break;
161	p = q;
162	}
163
164	/*
165	* If there is a preceding segment, it may provide some of
166	* our data already. If so, drop the data from the incoming
167	* segment. If it provides all of our data, drop us.
168	*/
169	if (p != NULL)
170	{
171	int i;
172	/* conversion to int (in i) handles seq wraparound */
173	i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
174	if (i > 0)
175	{
176	if (i >= *tlenp)
177	{
178	tcpstat.tcps_rcvduppack++;
179	tcpstat.tcps_rcvdupbyte += *tlenp;
180	m_freem(pData, m);
181	RTMemFree(te);
182	tp->t_segqlen--;
183	tcp_reass_qsize--;
184	/*
185	* Try to present any queued data
186	* at the left window edge to the user.
187	* This is needed after the 3-WHS
188	* completes.
189	*/
190	goto present; /* ??? */
191	}
192	m_adj(m, i);
193	*tlenp -= i;
194	th->th_seq += i;
195	}
196	}
197	tcpstat.tcps_rcvoopack++;
198	tcpstat.tcps_rcvoobyte += *tlenp;
199
200	/*
201	* While we overlap succeeding segments trim them or,
202	* if they are completely covered, dequeue them.
203	*/
204	while (q)
205	{
206	int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
207	if (i <= 0)
208	break;
209	if (i < q->tqe_len)
210	{
211	q->tqe_th->th_seq += i;
212	q->tqe_len -= i;
213	m_adj(q->tqe_m, i);
214	break;
215	}
216
217	nq = LIST_NEXT(q, tqe_q);
218	LIST_REMOVE(q, tqe_q);
219	m_freem(pData, q->tqe_m);
220	RTMemFree(q);
221	tp->t_segqlen--;
222	tcp_reass_qsize--;
223	q = nq;
224	}
225
226	/* Insert the new segment queue entry into place. */
227	te->tqe_m = m;
228	te->tqe_th = th;
229	te->tqe_len = *tlenp;
230
231	if (p == NULL)
232	{
233	LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
234	}
235	else
236	{
237	LIST_INSERT_AFTER(p, te, tqe_q);
238	}
239
240	present:
241	/*
242	* Present data to user, advancing rcv_nxt through
243	* completed sequence space.
244	*/
245	if (!TCPS_HAVEESTABLISHED(tp->t_state))
246	{
247	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
248	return (0);
249	}
250	q = LIST_FIRST(&tp->t_segq);
251	if (!q \|\| q->tqe_th->th_seq != tp->rcv_nxt)
252	{
253	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
254	return (0);
255	}
256	do
257	{
258	tp->rcv_nxt += q->tqe_len;
259	flags = q->tqe_th->th_flags & TH_FIN;
260	nq = LIST_NEXT(q, tqe_q);
261	LIST_REMOVE(q, tqe_q);
262	/* XXX: This place should be checked for the same code in
263	* original BSD code for Slirp and current BSD used SS_FCANTRCVMORE
264	*/
265	if (so->so_state & SS_FCANTSENDMORE)
266	m_freem(pData, q->tqe_m);
267	else
268	sbappend(pData, so, q->tqe_m);
269	RTMemFree(q);
270	tp->t_segqlen--;
271	tcp_reass_qsize--;
272	q = nq;
273	}
274	while (q && q->tqe_th->th_seq == tp->rcv_nxt);
275
276	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
277	return flags;
278	}
279
280	/*
281	* TCP input routine, follows pages 65-76 of the
282	* protocol specification dated September, 1981 very closely.
283	*/
284	void
285	tcp_input(PNATState pData, register struct mbuf m, int iphlen, struct socket inso)
286	{
287	struct ip save_ip, *ip;
288	register struct tcpiphdr *ti;
289	caddr_t optp = NULL;
290	int optlen = 0;
291	int len, tlen, off;
292	register struct tcpcb *tp = 0;
293	register int tiflags;
294	struct socket *so = 0;
295	int todrop, acked, ourfinisacked, needoutput = 0;
296	/* int dropsocket = 0; */
297	int iss = 0;
298	u_long tiwin;
299	/* int ts_present = 0; */
300	STAM_PROFILE_START(&pData->StatTCP_input, counter_input);
301
302	DEBUG_CALL("tcp_input");
303	DEBUG_ARGS((dfd," m = %8lx iphlen = %2d inso = %lx\n",
304	(long )m, iphlen, (long )inso ));
305
306	if (inso != NULL)
307	{
308	QSOCKET_LOCK(tcb);
309	SOCKET_LOCK(inso);
310	QSOCKET_UNLOCK(tcb);
311	}
312	/*
313	* If called with m == 0, then we're continuing the connect
314	*/
315	if (m == NULL)
316	{
317	so = inso;
318	Log4(("NAT: tcp_input: %R[natsock]\n", so));
319	/* Re-set a few variables */
320	tp = sototcpcb(so);
321	m = so->so_m;
322
323	so->so_m = 0;
324	ti = so->so_ti;
325
326	/** @todo (vvl) clarify why it might happens */
327	if (ti == NULL)
328	{
329	LogRel(("NAT: ti is null. can't do any reseting connection actions\n"));
330	/* mbuf should be cleared in sofree called from tcp_close */
331	tcp_close(pData, tp);
332	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
333	return;
334	}
335
336	tiwin = ti->ti_win;
337	tiflags = ti->ti_flags;
338
339	goto cont_conn;
340	}
341
342	tcpstat.tcps_rcvtotal++;
343	/*
344	* Get IP and TCP header together in first mbuf.
345	* Note: IP leaves IP header in first mbuf.
346	*/
347	ti = mtod(m, struct tcpiphdr *);
348	if (iphlen > sizeof(struct ip ))
349	{
350	ip_stripoptions(m, (struct mbuf *)0);
351	iphlen = sizeof(struct ip );
352	}
353	/* XXX Check if too short */
354
355
356	/*
357	* Save a copy of the IP header in case we want restore it
358	* for sending an ICMP error message in response.
359	*/
360	ip = mtod(m, struct ip *);
361	/*
362	* (vvl) ip_input substracts IP header length from ip->ip_len value.
363	* here we do the test the same as input method of UDP protocol.
364	*/
365	#ifdef VBOX_WITH_SLIRP_BSD_MBUF
366	Assert((ip->ip_len + iphlen == m_length(m, NULL)));
367	#else
368	Assert((ip->ip_len + iphlen == m->m_len));
369	#endif
370	save_ip = *ip;
371	save_ip.ip_len+= iphlen;
372
373	/*
374	* Checksum extended TCP header and data.
375	*/
376	tlen = ((struct ip *)ti)->ip_len;
377	memset(ti->ti_x1, 0, 9);
378	ti->ti_len = RT_H2N_U16((u_int16_t)tlen);
379	len = sizeof(struct ip ) + tlen;
380	/* keep checksum for ICMP reply
381	* ti->ti_sum = cksum(m, len);
382	* if (ti->ti_sum) { */
383	if (cksum(m, len))
384	{
385	tcpstat.tcps_rcvbadsum++;
386	goto drop;
387	}
388
389	/*
390	* Check that TCP offset makes sense,
391	* pull out TCP options and adjust length. XXX
392	*/
393	off = ti->ti_off << 2;
394	if ( off < sizeof (struct tcphdr)
395	\|\| off > tlen)
396	{
397	tcpstat.tcps_rcvbadoff++;
398	goto drop;
399	}
400	tlen -= off;
401	ti->ti_len = tlen;
402	if (off > sizeof (struct tcphdr))
403	{
404	optlen = off - sizeof (struct tcphdr);
405	optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
406
407	/*
408	* Do quick retrieval of timestamp options ("options
409	* prediction?"). If timestamp is the only option and it's
410	* formatted as recommended in RFC 1323 appendix A, we
411	* quickly get the values now and not bother calling
412	* tcp_dooptions(), etc.
413	*/
414	#if 0
415	if (( optlen == TCPOLEN_TSTAMP_APPA
416	\|\| ( optlen > TCPOLEN_TSTAMP_APPA
417	&& optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
418	(u_int32_t )optp == RT_H2N_U32_C(TCPOPT_TSTAMP_HDR) &&
419	(ti->ti_flags & TH_SYN) == 0)
420	{
421	ts_present = 1;
422	ts_val = RT_N2H_U32((u_int32_t )(optp + 4));
423	ts_ecr = RT_N2H_U32((u_int32_t )(optp + 8));
424	optp = NULL; / * we have parsed the options * /
425	}
426	#endif
427	}
428	tiflags = ti->ti_flags;
429
430	/*
431	* Convert TCP protocol specific fields to host format.
432	*/
433	NTOHL(ti->ti_seq);
434	NTOHL(ti->ti_ack);
435	NTOHS(ti->ti_win);
436	NTOHS(ti->ti_urp);
437
438	/*
439	* Drop TCP, IP headers and TCP options.
440	*/
441	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
442	m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
443
444	/*
445	* Locate pcb for segment.
446	*/
447	findso:
448	if (so != NULL && so != &tcb)
449	SOCKET_UNLOCK(so);
450	QSOCKET_LOCK(tcb);
451	so = tcp_last_so;
452	if ( so->so_fport != ti->ti_dport
453	\|\| so->so_lport != ti->ti_sport
454	\|\| so->so_laddr.s_addr != ti->ti_src.s_addr
455	\|\| so->so_faddr.s_addr != ti->ti_dst.s_addr)
456	{
457	#ifdef VBOX_WITH_SLIRP_MT
458	struct socket *sonxt;
459	#endif
460	QSOCKET_UNLOCK(tcb);
461	/* @todo fix SOLOOKUP macrodefinition to be usable here */
462	#ifndef VBOX_WITH_SLIRP_MT
463	so = solookup(&tcb, ti->ti_src, ti->ti_sport,
464	ti->ti_dst, ti->ti_dport);
465	#else
466	so = NULL;
467	QSOCKET_FOREACH(so, sonxt, tcp)
468	/* { */
469	if ( so->so_lport == ti->ti_sport
470	&& so->so_laddr.s_addr == ti->ti_src.s_addr
471	&& so->so_faddr.s_addr == ti->ti_dst.s_addr
472	&& so->so_fport == ti->ti_dport
473	&& so->so_deleted != 1)
474	{
475	break; /* so is locked here */
476	}
477	LOOP_LABEL(tcp, so, sonxt);
478	}
479	if (so == &tcb) {
480	so = NULL;
481	}
482	#endif
483	if (so)
484	{
485	tcp_last_so = so;
486	}
487	++tcpstat.tcps_socachemiss;
488	}
489	else
490	{
491	SOCKET_LOCK(so);
492	QSOCKET_UNLOCK(tcb);
493	}
494
495	/*
496	* If the state is CLOSED (i.e., TCB does not exist) then
497	* all data in the incoming segment is discarded.
498	* If the TCB exists but is in CLOSED state, it is embryonic,
499	* but should either do a listen or a connect soon.
500	*
501	* state == CLOSED means we've done socreate() but haven't
502	* attached it to a protocol yet...
503	*
504	* XXX If a TCB does not exist, and the TH_SYN flag is
505	* the only flag set, then create a session, mark it
506	* as if it was LISTENING, and continue...
507	*/
508	if (so == 0)
509	{
510	if ((tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) != TH_SYN)
511	goto dropwithreset;
512
513	if ((so = socreate()) == NULL)
514	goto dropwithreset;
515	if (tcp_attach(pData, so) < 0)
516	{
517	RTMemFree(so); /* Not sofree (if it failed, it's not insqued) */
518	goto dropwithreset;
519	}
520	SOCKET_LOCK(so);
521	sbreserve(pData, &so->so_snd, tcp_sndspace);
522	sbreserve(pData, &so->so_rcv, tcp_rcvspace);
523
524	/* tcp_last_so = so; / / XXX ? */
525	/* tp = sototcpcb(so); */
526
527	so->so_laddr = ti->ti_src;
528	so->so_lport = ti->ti_sport;
529	so->so_faddr = ti->ti_dst;
530	so->so_fport = ti->ti_dport;
531
532	if ((so->so_iptos = tcp_tos(so)) == 0)
533	so->so_iptos = ((struct ip *)ti)->ip_tos;
534
535	tp = sototcpcb(so);
536	tp->t_state = TCPS_LISTEN;
537	}
538
539	/*
540	* If this is a still-connecting socket, this probably
541	* a retransmit of the SYN. Whether it's a retransmit SYN
542	* or something else, we nuke it.
543	*/
544	if (so->so_state & SS_ISFCONNECTING)
545	{
546	goto drop;
547	}
548
549	tp = sototcpcb(so);
550
551	/* XXX Should never fail */
552	if (tp == 0)
553	goto dropwithreset;
554	if (tp->t_state == TCPS_CLOSED)
555	{
556	goto drop;
557	}
558
559	/* Unscale the window into a 32-bit value. */
560	/* if ((tiflags & TH_SYN) == 0)
561	* tiwin = ti->ti_win << tp->snd_scale;
562	* else
563	*/
564	tiwin = ti->ti_win;
565
566	/*
567	* Segment received on connection.
568	* Reset idle time and keep-alive timer.
569	*/
570	tp->t_idle = 0;
571	if (so_options)
572	tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
573	else
574	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
575
576	/*
577	* Process options if not in LISTEN state,
578	* else do it below (after getting remote address).
579	*/
580	if (optp && tp->t_state != TCPS_LISTEN)
581	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
582	/* , */
583	/* &ts_present, &ts_val, &ts_ecr); */
584
585	/*
586	* Header prediction: check for the two common cases
587	* of a uni-directional data xfer. If the packet has
588	* no control flags, is in-sequence, the window didn't
589	* change and we're not retransmitting, it's a
590	* candidate. If the length is zero and the ack moved
591	* forward, we're the sender side of the xfer. Just
592	* free the data acked & wake any higher level process
593	* that was blocked waiting for space. If the length
594	* is non-zero and the ack didn't move, we're the
595	* receiver side. If we're getting packets in-order
596	* (the reassembly queue is empty), add the data to
597	* the socket buffer and note that we need a delayed ack.
598	*
599	* XXX Some of these tests are not needed
600	* eg: the tiwin == tp->snd_wnd prevents many more
601	* predictions.. with no real advantage..
602	*/
603	if ( tp->t_state == TCPS_ESTABLISHED
604	&& (tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK
605	/* && (!ts_present \|\| TSTMP_GEQ(ts_val, tp->ts_recent)) */
606	&& ti->ti_seq == tp->rcv_nxt
607	&& tiwin && tiwin == tp->snd_wnd
608	&& tp->snd_nxt == tp->snd_max)
609	{
610	/*
611	* If last ACK falls within this segment's sequence numbers,
612	* record the timestamp.
613	*/
614	#if 0
615	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
616	SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len))
617	{
618	tp->ts_recent_age = tcp_now;
619	tp->ts_recent = ts_val;
620	}
621	#endif
622
623	if (ti->ti_len == 0)
624	{
625	if ( SEQ_GT(ti->ti_ack, tp->snd_una)
626	&& SEQ_LEQ(ti->ti_ack, tp->snd_max)
627	&& tp->snd_cwnd >= tp->snd_wnd)
628	{
629	/*
630	* this is a pure ack for outstanding data.
631	*/
632	++tcpstat.tcps_predack;
633	#if 0
634	if (ts_present)
635	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
636	else
637	#endif
638	if ( tp->t_rtt
639	&& SEQ_GT(ti->ti_ack, tp->t_rtseq))
640	tcp_xmit_timer(pData, tp, tp->t_rtt);
641	acked = ti->ti_ack - tp->snd_una;
642	tcpstat.tcps_rcvackpack++;
643	tcpstat.tcps_rcvackbyte += acked;
644	sbdrop(&so->so_snd, acked);
645	tp->snd_una = ti->ti_ack;
646	m_freem(pData, m);
647
648	/*
649	* If all outstanding data are acked, stop
650	* retransmit timer, otherwise restart timer
651	* using current (possibly backed-off) value.
652	* If process is waiting for space,
653	* wakeup/selwakeup/signal. If data
654	* are ready to send, let tcp_output
655	* decide between more output or persist.
656	*/
657	if (tp->snd_una == tp->snd_max)
658	tp->t_timer[TCPT_REXMT] = 0;
659	else if (tp->t_timer[TCPT_PERSIST] == 0)
660	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
661
662	/*
663	* There's room in so_snd, sowwakup will read()
664	* from the socket if we can
665	*/
666	#if 0
667	if (so->so_snd.sb_flags & SB_NOTIFY)
668	sowwakeup(so);
669	#endif
670	/*
671	* This is called because sowwakeup might have
672	* put data into so_snd. Since we don't so sowwakeup,
673	* we don't need this.. XXX???
674	*/
675	if (so->so_snd.sb_cc)
676	(void) tcp_output(pData, tp);
677
678	SOCKET_UNLOCK(so);
679	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
680	return;
681	}
682	}
683	else if ( ti->ti_ack == tp->snd_una
684	&& LIST_FIRST(&tp->t_segq)
685	&& ti->ti_len <= sbspace(&so->so_rcv))
686	{
687	/*
688	* this is a pure, in-sequence data packet
689	* with nothing on the reassembly queue and
690	* we have enough buffer space to take it.
691	*/
692	++tcpstat.tcps_preddat;
693	tp->rcv_nxt += ti->ti_len;
694	tcpstat.tcps_rcvpack++;
695	tcpstat.tcps_rcvbyte += ti->ti_len;
696	/*
697	* Add data to socket buffer.
698	*/
699	sbappend(pData, so, m);
700
701	/*
702	* XXX This is called when data arrives. Later, check
703	* if we can actually write() to the socket
704	* XXX Need to check? It's be NON_BLOCKING
705	*/
706	/* sorwakeup(so); */
707
708	/*
709	* If this is a short packet, then ACK now - with Nagel
710	* congestion avoidance sender won't send more until
711	* he gets an ACK.
712	*
713	* It is better to not delay acks at all to maximize
714	* TCP throughput. See RFC 2581.
715	*/
716	tp->t_flags \|= TF_ACKNOW;
717	tcp_output(pData, tp);
718	SOCKET_UNLOCK(so);
719	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
720	return;
721	}
722	} /* header prediction */
723	/*
724	* Calculate amount of space in receive window,
725	* and then do TCP input processing.
726	* Receive window is amount of space in rcv queue,
727	* but not less than advertised window.
728	*/
729	{
730	int win;
731	win = sbspace(&so->so_rcv);
732	if (win < 0)
733	win = 0;
734	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
735	}
736
737	switch (tp->t_state)
738	{
739	/*
740	* If the state is LISTEN then ignore segment if it contains an RST.
741	* If the segment contains an ACK then it is bad and send a RST.
742	* If it does not contain a SYN then it is not interesting; drop it.
743	* Don't bother responding if the destination was a broadcast.
744	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
745	* tp->iss, and send a segment:
746	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
747	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
748	* Fill in remote peer address fields if not previously specified.
749	* Enter SYN_RECEIVED state, and process any other fields of this
750	* segment in this state.
751	*/
752	case TCPS_LISTEN:
753	{
754	if (tiflags & TH_RST) {
755	goto drop;
756	}
757	if (tiflags & TH_ACK)
758	goto dropwithreset;
759	if ((tiflags & TH_SYN) == 0)
760	{
761	goto drop;
762	}
763
764	/*
765	* This has way too many gotos...
766	* But a bit of spaghetti code never hurt anybody :)
767	*/
768
769	if (so->so_emu & EMU_NOCONNECT)
770	{
771	so->so_emu &= ~EMU_NOCONNECT;
772	goto cont_input;
773	}
774
775	if ( (tcp_fconnect(pData, so) == -1)
776	&& errno != EINPROGRESS
777	&& errno != EWOULDBLOCK)
778	{
779	u_char code = ICMP_UNREACH_NET;
780	DEBUG_MISC((dfd," tcp fconnect errno = %d-%s\n",
781	errno, strerror(errno)));
782	if (errno == ECONNREFUSED)
783	{
784	/* ACK the SYN, send RST to refuse the connection */
785	tcp_respond(pData, tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
786	TH_RST\|TH_ACK);
787	}
788	else
789	{
790	if (errno == EHOSTUNREACH)
791	code = ICMP_UNREACH_HOST;
792	HTONL(ti->ti_seq); /* restore tcp header */
793	HTONL(ti->ti_ack);
794	HTONS(ti->ti_win);
795	HTONS(ti->ti_urp);
796	m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
797	m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
798	*ip = save_ip;
799	icmp_error(pData, m, ICMP_UNREACH, code, 0, strerror(errno));
800	tp->t_socket->so_m = NULL;
801	}
802	tp = tcp_close(pData, tp);
803	m_freem(pData, m);
804	}
805	else
806	{
807	/*
808	* Haven't connected yet, save the current mbuf
809	* and ti, and return
810	* XXX Some OS's don't tell us whether the connect()
811	* succeeded or not. So we must time it out.
812	*/
813	so->so_m = m;
814	so->so_ti = ti;
815	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
816	tp->t_state = TCPS_SYN_RECEIVED;
817	}
818	SOCKET_UNLOCK(so);
819	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
820	return;
821
822	cont_conn:
823	/* m==NULL
824	* Check if the connect succeeded
825	*/
826	if (so->so_state & SS_NOFDREF)
827	{
828	tp = tcp_close(pData, tp);
829	goto dropwithreset;
830	}
831	cont_input:
832	tcp_template(tp);
833
834	if (optp)
835	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
836
837	if (iss)
838	tp->iss = iss;
839	else
840	tp->iss = tcp_iss;
841	tcp_iss += TCP_ISSINCR/2;
842	tp->irs = ti->ti_seq;
843	tcp_sendseqinit(tp);
844	tcp_rcvseqinit(tp);
845	tp->t_flags \|= TF_ACKNOW;
846	tp->t_state = TCPS_SYN_RECEIVED;
847	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
848	tcpstat.tcps_accepts++;
849	goto trimthenstep6;
850	} /* case TCPS_LISTEN */
851
852	/*
853	* If the state is SYN_SENT:
854	* if seg contains an ACK, but not for our SYN, drop the input.
855	* if seg contains a RST, then drop the connection.
856	* if seg does not contain SYN, then drop it.
857	* Otherwise this is an acceptable SYN segment
858	* initialize tp->rcv_nxt and tp->irs
859	* if seg contains ack then advance tp->snd_una
860	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
861	* arrange for segment to be acked (eventually)
862	* continue processing rest of data/controls, beginning with URG
863	*/
864	case TCPS_SYN_SENT:
865	if ( (tiflags & TH_ACK)
866	&& ( SEQ_LEQ(ti->ti_ack, tp->iss)
867	\|\| SEQ_GT(ti->ti_ack, tp->snd_max)))
868	goto dropwithreset;
869
870	if (tiflags & TH_RST)
871	{
872	if (tiflags & TH_ACK)
873	tp = tcp_drop(pData, tp, 0); /* XXX Check t_softerror! */
874	goto drop;
875	}
876
877	if ((tiflags & TH_SYN) == 0)
878	{
879	goto drop;
880	}
881	if (tiflags & TH_ACK)
882	{
883	tp->snd_una = ti->ti_ack;
884	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
885	tp->snd_nxt = tp->snd_una;
886	}
887
888	tp->t_timer[TCPT_REXMT] = 0;
889	tp->irs = ti->ti_seq;
890	tcp_rcvseqinit(tp);
891	tp->t_flags \|= TF_ACKNOW;
892	if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss))
893	{
894	tcpstat.tcps_connects++;
895	soisfconnected(so);
896	tp->t_state = TCPS_ESTABLISHED;
897
898	/* Do window scaling on this connection? */
899	#if 0
900	if (( tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
901	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
902	{
903	tp->snd_scale = tp->requested_s_scale;
904	tp->rcv_scale = tp->request_r_scale;
905	}
906	#endif
907	(void) tcp_reass(pData, tp, (struct tcphdr )0, NULL, (struct mbuf )0);
908	/*
909	* if we didn't have to retransmit the SYN,
910	* use its rtt as our initial srtt & rtt var.
911	*/
912	if (tp->t_rtt)
913	tcp_xmit_timer(pData, tp, tp->t_rtt);
914	}
915	else
916	tp->t_state = TCPS_SYN_RECEIVED;
917
918	trimthenstep6:
919	/*
920	* Advance ti->ti_seq to correspond to first data byte.
921	* If data, trim to stay within window,
922	* dropping FIN if necessary.
923	*/
924	ti->ti_seq++;
925	if (ti->ti_len > tp->rcv_wnd)
926	{
927	todrop = ti->ti_len - tp->rcv_wnd;
928	m_adj(m, -todrop);
929	ti->ti_len = tp->rcv_wnd;
930	tiflags &= ~TH_FIN;
931	tcpstat.tcps_rcvpackafterwin++;
932	tcpstat.tcps_rcvbyteafterwin += todrop;
933	}
934	tp->snd_wl1 = ti->ti_seq - 1;
935	tp->rcv_up = ti->ti_seq;
936	Log2(("hit6"));
937	goto step6;
938	} /* switch tp->t_state */
939	/*
940	* States other than LISTEN or SYN_SENT.
941	* First check timestamp, if present.
942	* Then check that at least some bytes of segment are within
943	* receive window. If segment begins before rcv_nxt,
944	* drop leading data (and SYN); if nothing left, just ack.
945	*
946	* RFC 1323 PAWS: If we have a timestamp reply on this segment
947	* and it's less than ts_recent, drop it.
948	*/
949	#if 0
950	if ( ts_present
951	&& (tiflags & TH_RST) == 0
952	&& tp->ts_recent
953	&& TSTMP_LT(ts_val, tp->ts_recent))
954	{
955	/* Check to see if ts_recent is over 24 days old. */
956	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE)
957	{
958	/*
959	* Invalidate ts_recent. If this segment updates
960	* ts_recent, the age will be reset later and ts_recent
961	* will get a valid value. If it does not, setting
962	* ts_recent to zero will at least satisfy the
963	* requirement that zero be placed in the timestamp
964	* echo reply when ts_recent isn't valid. The
965	* age isn't reset until we get a valid ts_recent
966	* because we don't want out-of-order segments to be
967	* dropped when ts_recent is old.
968	*/
969	tp->ts_recent = 0;
970	}
971	else
972	{
973	tcpstat.tcps_rcvduppack++;
974	tcpstat.tcps_rcvdupbyte += ti->ti_len;
975	tcpstat.tcps_pawsdrop++;
976	goto dropafterack;
977	}
978	}
979	#endif
980
981	todrop = tp->rcv_nxt - ti->ti_seq;
982	if (todrop > 0)
983	{
984	if (tiflags & TH_SYN)
985	{
986	tiflags &= ~TH_SYN;
987	ti->ti_seq++;
988	if (ti->ti_urp > 1)
989	ti->ti_urp--;
990	else
991	tiflags &= ~TH_URG;
992	todrop--;
993	}
994	/*
995	* Following if statement from Stevens, vol. 2, p. 960.
996	*/
997	if ( todrop > ti->ti_len
998	\|\| ( todrop == ti->ti_len
999	&& (tiflags & TH_FIN) == 0))
1000	{
1001	/*
1002	* Any valid FIN must be to the left of the window.
1003	* At this point the FIN must be a duplicate or out
1004	* of sequence; drop it.
1005	*/
1006	tiflags &= ~TH_FIN;
1007
1008	/*
1009	* Send an ACK to resynchronize and drop any data.
1010	* But keep on processing for RST or ACK.
1011	*/
1012	tp->t_flags \|= TF_ACKNOW;
1013	todrop = ti->ti_len;
1014	tcpstat.tcps_rcvduppack++;
1015	tcpstat.tcps_rcvdupbyte += todrop;
1016	}
1017	else
1018	{
1019	tcpstat.tcps_rcvpartduppack++;
1020	tcpstat.tcps_rcvpartdupbyte += todrop;
1021	}
1022	m_adj(m, todrop);
1023	ti->ti_seq += todrop;
1024	ti->ti_len -= todrop;
1025	if (ti->ti_urp > todrop)
1026	ti->ti_urp -= todrop;
1027	else
1028	{
1029	tiflags &= ~TH_URG;
1030	ti->ti_urp = 0;
1031	}
1032	}
1033	/*
1034	* If new data are received on a connection after the
1035	* user processes are gone, then RST the other end.
1036	*/
1037	if ( (so->so_state & SS_NOFDREF)
1038	&& tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len)
1039	{
1040	tp = tcp_close(pData, tp);
1041	tcpstat.tcps_rcvafterclose++;
1042	goto dropwithreset;
1043	}
1044
1045	/*
1046	* If segment ends after window, drop trailing data
1047	* (and PUSH and FIN); if nothing left, just ACK.
1048	*/
1049	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1050	if (todrop > 0)
1051	{
1052	tcpstat.tcps_rcvpackafterwin++;
1053	if (todrop >= ti->ti_len)
1054	{
1055	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1056	/*
1057	* If a new connection request is received
1058	* while in TIME_WAIT, drop the old connection
1059	* and start over if the sequence numbers
1060	* are above the previous ones.
1061	*/
1062	if ( tiflags & TH_SYN
1063	&& tp->t_state == TCPS_TIME_WAIT
1064	&& SEQ_GT(ti->ti_seq, tp->rcv_nxt))
1065	{
1066	iss = tp->rcv_nxt + TCP_ISSINCR;
1067	tp = tcp_close(pData, tp);
1068	SOCKET_UNLOCK(tp->t_socket);
1069	goto findso;
1070	}
1071	/*
1072	* If window is closed can only take segments at
1073	* window edge, and have to drop data and PUSH from
1074	* incoming segments. Continue processing, but
1075	* remember to ack. Otherwise, drop segment
1076	* and ack.
1077	*/
1078	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt)
1079	{
1080	tp->t_flags \|= TF_ACKNOW;
1081	tcpstat.tcps_rcvwinprobe++;
1082	}
1083	else
1084	goto dropafterack;
1085	}
1086	else
1087	tcpstat.tcps_rcvbyteafterwin += todrop;
1088	m_adj(m, -todrop);
1089	ti->ti_len -= todrop;
1090	tiflags &= ~(TH_PUSH\|TH_FIN);
1091	}
1092
1093	/*
1094	* If last ACK falls within this segment's sequence numbers,
1095	* record its timestamp.
1096	*/
1097	#if 0
1098	if ( ts_present
1099	&& SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)
1100	&& SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + ((tiflags & (TH_SYN\|TH_FIN)) != 0)))
1101	{
1102	tp->ts_recent_age = tcp_now;
1103	tp->ts_recent = ts_val;
1104	}
1105	#endif
1106
1107	/*
1108	* If the RST bit is set examine the state:
1109	* SYN_RECEIVED STATE:
1110	* If passive open, return to LISTEN state.
1111	* If active open, inform user that connection was refused.
1112	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1113	* Inform user that connection was reset, and close tcb.
1114	* CLOSING, LAST_ACK, TIME_WAIT STATES
1115	* Close the tcb.
1116	*/
1117	if (tiflags&TH_RST)
1118	switch (tp->t_state)
1119	{
1120	case TCPS_SYN_RECEIVED:
1121	/* so->so_error = ECONNREFUSED; */
1122	goto close;
1123
1124	case TCPS_ESTABLISHED:
1125	case TCPS_FIN_WAIT_1:
1126	case TCPS_FIN_WAIT_2:
1127	case TCPS_CLOSE_WAIT:
1128	/* so->so_error = ECONNRESET; */
1129	close:
1130	tp->t_state = TCPS_CLOSED;
1131	tcpstat.tcps_drops++;
1132	tp = tcp_close(pData, tp);
1133	goto drop;
1134
1135	case TCPS_CLOSING:
1136	case TCPS_LAST_ACK:
1137	case TCPS_TIME_WAIT:
1138	tp = tcp_close(pData, tp);
1139	goto drop;
1140	}
1141
1142	/*
1143	* If a SYN is in the window, then this is an
1144	* error and we send an RST and drop the connection.
1145	*/
1146	if (tiflags & TH_SYN)
1147	{
1148	tp = tcp_drop(pData, tp, 0);
1149	goto dropwithreset;
1150	}
1151
1152	/*
1153	* If the ACK bit is off we drop the segment and return.
1154	*/
1155	if ((tiflags & TH_ACK) == 0)
1156	{
1157	goto drop;
1158	}
1159
1160	/*
1161	* Ack processing.
1162	*/
1163	switch (tp->t_state)
1164	{
1165	/*
1166	* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1167	* ESTABLISHED state and continue processing, otherwise
1168	* send an RST. una<=ack<=max
1169	*/
1170	case TCPS_SYN_RECEIVED:
1171	if ( SEQ_GT(tp->snd_una, ti->ti_ack)
1172	\|\| SEQ_GT(ti->ti_ack, tp->snd_max))
1173	goto dropwithreset;
1174	tcpstat.tcps_connects++;
1175	tp->t_state = TCPS_ESTABLISHED;
1176	/*
1177	* The sent SYN is ack'ed with our sequence number +1
1178	* The first data byte already in the buffer will get
1179	* lost if no correction is made. This is only needed for
1180	* SS_CTL since the buffer is empty otherwise.
1181	* tp->snd_una++; or:
1182	*/
1183	tp->snd_una = ti->ti_ack;
1184	soisfconnected(so);
1185
1186	/* Do window scaling? */
1187	#if 0
1188	if ( (tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
1189	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
1190	{
1191	tp->snd_scale = tp->requested_s_scale;
1192	tp->rcv_scale = tp->request_r_scale;
1193	}
1194	#endif
1195	(void) tcp_reass(pData, tp, (struct tcphdr )0, (int )0, (struct mbuf *)0);
1196	tp->snd_wl1 = ti->ti_seq - 1;
1197	/* Avoid ack processing; snd_una==ti_ack => dup ack */
1198	goto synrx_to_est;
1199	/* fall into ... */
1200
1201	/*
1202	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1203	* ACKs. If the ack is in the range
1204	* tp->snd_una < ti->ti_ack <= tp->snd_max
1205	* then advance tp->snd_una to ti->ti_ack and drop
1206	* data from the retransmission queue. If this ACK reflects
1207	* more up to date window information we update our window information.
1208	*/
1209	case TCPS_ESTABLISHED:
1210	case TCPS_FIN_WAIT_1:
1211	case TCPS_FIN_WAIT_2:
1212	case TCPS_CLOSE_WAIT:
1213	case TCPS_CLOSING:
1214	case TCPS_LAST_ACK:
1215	case TCPS_TIME_WAIT:
1216	if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
1217	{
1218	if (ti->ti_len == 0 && tiwin == tp->snd_wnd)
1219	{
1220	tcpstat.tcps_rcvdupack++;
1221	DEBUG_MISC((dfd," dup ack m = %lx so = %lx \n",
1222	(long )m, (long )so));
1223	/*
1224	* If we have outstanding data (other than
1225	* a window probe), this is a completely
1226	* duplicate ack (ie, window info didn't
1227	* change), the ack is the biggest we've
1228	* seen and we've seen exactly our rexmt
1229	* threshold of them, assume a packet
1230	* has been dropped and retransmit it.
1231	* Kludge snd_nxt & the congestion
1232	* window so we send only this one
1233	* packet.
1234	*
1235	* We know we're losing at the current
1236	* window size so do congestion avoidance
1237	* (set ssthresh to half the current window
1238	* and pull our congestion window back to
1239	* the new ssthresh).
1240	*
1241	* Dup acks mean that packets have left the
1242	* network (they're now cached at the receiver)
1243	* so bump cwnd by the amount in the receiver
1244	* to keep a constant cwnd packets in the
1245	* network.
1246	*/
1247	if ( tp->t_timer[TCPT_REXMT] == 0
1248	\|\| ti->ti_ack != tp->snd_una)
1249	tp->t_dupacks = 0;
1250	else if (++tp->t_dupacks == tcprexmtthresh)
1251	{
1252	tcp_seq onxt = tp->snd_nxt;
1253	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
1254	if (win < 2)
1255	win = 2;
1256	tp->snd_ssthresh = win * tp->t_maxseg;
1257	tp->t_timer[TCPT_REXMT] = 0;
1258	tp->t_rtt = 0;
1259	tp->snd_nxt = ti->ti_ack;
1260	tp->snd_cwnd = tp->t_maxseg;
1261	(void) tcp_output(pData, tp);
1262	tp->snd_cwnd = tp->snd_ssthresh +
1263	tp->t_maxseg * tp->t_dupacks;
1264	if (SEQ_GT(onxt, tp->snd_nxt))
1265	tp->snd_nxt = onxt;
1266	goto drop;
1267	}
1268	else if (tp->t_dupacks > tcprexmtthresh)
1269	{
1270	tp->snd_cwnd += tp->t_maxseg;
1271	(void) tcp_output(pData, tp);
1272	goto drop;
1273	}
1274	}
1275	else
1276	tp->t_dupacks = 0;
1277	break;
1278	}
1279	synrx_to_est:
1280	/*
1281	* If the congestion window was inflated to account
1282	* for the other side's cached packets, retract it.
1283	*/
1284	if ( tp->t_dupacks > tcprexmtthresh
1285	&& tp->snd_cwnd > tp->snd_ssthresh)
1286	tp->snd_cwnd = tp->snd_ssthresh;
1287	tp->t_dupacks = 0;
1288	if (SEQ_GT(ti->ti_ack, tp->snd_max))
1289	{
1290	tcpstat.tcps_rcvacktoomuch++;
1291	goto dropafterack;
1292	}
1293	acked = ti->ti_ack - tp->snd_una;
1294	tcpstat.tcps_rcvackpack++;
1295	tcpstat.tcps_rcvackbyte += acked;
1296
1297	/*
1298	* If we have a timestamp reply, update smoothed
1299	* round trip time. If no timestamp is present but
1300	* transmit timer is running and timed sequence
1301	* number was acked, update smoothed round trip time.
1302	* Since we now have an rtt measurement, cancel the
1303	* timer backoff (cf., Phil Karn's retransmit alg.).
1304	* Recompute the initial retransmit timer.
1305	*/
1306	#if 0
1307	if (ts_present)
1308	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1309	else
1310	#endif
1311	if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1312	tcp_xmit_timer(pData, tp, tp->t_rtt);
1313
1314	/*
1315	* If all outstanding data is acked, stop retransmit
1316	* timer and remember to restart (more output or persist).
1317	* If there is more data to be acked, restart retransmit
1318	* timer, using current (possibly backed-off) value.
1319	*/
1320	if (ti->ti_ack == tp->snd_max)
1321	{
1322	tp->t_timer[TCPT_REXMT] = 0;
1323	needoutput = 1;
1324	}
1325	else if (tp->t_timer[TCPT_PERSIST] == 0)
1326	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1327	/*
1328	* When new data is acked, open the congestion window.
1329	* If the window gives us less than ssthresh packets
1330	* in flight, open exponentially (maxseg per packet).
1331	* Otherwise open linearly: maxseg per window
1332	* (maxseg^2 / cwnd per packet).
1333	*/
1334	{
1335	register u_int cw = tp->snd_cwnd;
1336	register u_int incr = tp->t_maxseg;
1337
1338	if (cw > tp->snd_ssthresh)
1339	incr = incr * incr / cw;
1340	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1341	}
1342	if (acked > so->so_snd.sb_cc)
1343	{
1344	tp->snd_wnd -= so->so_snd.sb_cc;
1345	sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1346	ourfinisacked = 1;
1347	}
1348	else
1349	{
1350	sbdrop(&so->so_snd, acked);
1351	tp->snd_wnd -= acked;
1352	ourfinisacked = 0;
1353	}
1354	/*
1355	* XXX sowwakup is called when data is acked and there's room for
1356	* for more data... it should read() the socket
1357	*/
1358	#if 0
1359	if (so->so_snd.sb_flags & SB_NOTIFY)
1360	sowwakeup(so);
1361	#endif
1362	tp->snd_una = ti->ti_ack;
1363	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1364	tp->snd_nxt = tp->snd_una;
1365
1366	switch (tp->t_state)
1367	{
1368	/*
1369	* In FIN_WAIT_1 STATE in addition to the processing
1370	* for the ESTABLISHED state if our FIN is now acknowledged
1371	* then enter FIN_WAIT_2.
1372	*/
1373	case TCPS_FIN_WAIT_1:
1374	if (ourfinisacked)
1375	{
1376	/*
1377	* If we can't receive any more
1378	* data, then closing user can proceed.
1379	* Starting the timer is contrary to the
1380	* specification, but if we don't get a FIN
1381	* we'll hang forever.
1382	*/
1383	if (so->so_state & SS_FCANTRCVMORE)
1384	{
1385	soisfdisconnected(so);
1386	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1387	}
1388	tp->t_state = TCPS_FIN_WAIT_2;
1389	}
1390	break;
1391
1392	/*
1393	* In CLOSING STATE in addition to the processing for
1394	* the ESTABLISHED state if the ACK acknowledges our FIN
1395	* then enter the TIME-WAIT state, otherwise ignore
1396	* the segment.
1397	*/
1398	case TCPS_CLOSING:
1399	if (ourfinisacked)
1400	{
1401	tp->t_state = TCPS_TIME_WAIT;
1402	tcp_canceltimers(tp);
1403	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1404	soisfdisconnected(so);
1405	}
1406	break;
1407
1408	/*
1409	* In LAST_ACK, we may still be waiting for data to drain
1410	* and/or to be acked, as well as for the ack of our FIN.
1411	* If our FIN is now acknowledged, delete the TCB,
1412	* enter the closed state and return.
1413	*/
1414	case TCPS_LAST_ACK:
1415	if (ourfinisacked)
1416	{
1417	tp = tcp_close(pData, tp);
1418	goto drop;
1419	}
1420	break;
1421
1422	/*
1423	* In TIME_WAIT state the only thing that should arrive
1424	* is a retransmission of the remote FIN. Acknowledge
1425	* it and restart the finack timer.
1426	*/
1427	case TCPS_TIME_WAIT:
1428	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1429	goto dropafterack;
1430	}
1431	} /* switch(tp->t_state) */
1432
1433	step6:
1434	/*
1435	* Update window information.
1436	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1437	*/
1438	if ( (tiflags & TH_ACK)
1439	&& ( SEQ_LT(tp->snd_wl1, ti->ti_seq)
1440	\|\| ( tp->snd_wl1 == ti->ti_seq
1441	&& ( SEQ_LT(tp->snd_wl2, ti->ti_ack)
1442	\|\| ( tp->snd_wl2 == ti->ti_ack
1443	&& tiwin > tp->snd_wnd)))))
1444	{
1445	/* keep track of pure window updates */
1446	if ( ti->ti_len == 0
1447	&& tp->snd_wl2 == ti->ti_ack
1448	&& tiwin > tp->snd_wnd)
1449	tcpstat.tcps_rcvwinupd++;
1450	tp->snd_wnd = tiwin;
1451	tp->snd_wl1 = ti->ti_seq;
1452	tp->snd_wl2 = ti->ti_ack;
1453	if (tp->snd_wnd > tp->max_sndwnd)
1454	tp->max_sndwnd = tp->snd_wnd;
1455	needoutput = 1;
1456	}
1457
1458	/*
1459	* Process segments with URG.
1460	*/
1461	if ((tiflags & TH_URG) && ti->ti_urp &&
1462	TCPS_HAVERCVDFIN(tp->t_state) == 0)
1463	{
1464	/*
1465	* This is a kludge, but if we receive and accept
1466	* random urgent pointers, we'll crash in
1467	* soreceive. It's hard to imagine someone
1468	* actually wanting to send this much urgent data.
1469	*/
1470	if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen)
1471	{
1472	ti->ti_urp = 0;
1473	tiflags &= ~TH_URG;
1474	goto dodata;
1475	}
1476	/*
1477	* If this segment advances the known urgent pointer,
1478	* then mark the data stream. This should not happen
1479	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1480	* a FIN has been received from the remote side.
1481	* In these states we ignore the URG.
1482	*
1483	* According to RFC961 (Assigned Protocols),
1484	* the urgent pointer points to the last octet
1485	* of urgent data. We continue, however,
1486	* to consider it to indicate the first octet
1487	* of data past the urgent section as the original
1488	* spec states (in one of two places).
1489	*/
1490	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up))
1491	{
1492	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1493	so->so_urgc = so->so_rcv.sb_cc +
1494	(tp->rcv_up - tp->rcv_nxt); /* -1; */
1495	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1496	}
1497	}
1498	else
1499	/*
1500	* If no out of band data is expected,
1501	* pull receive urgent pointer along
1502	* with the receive window.
1503	*/
1504	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1505	tp->rcv_up = tp->rcv_nxt;
1506	dodata:
1507
1508	/*
1509	* If this is a small packet, then ACK now - with Nagel
1510	* congestion avoidance sender won't send more until
1511	* he gets an ACK.
1512	*
1513	* See above.
1514	*/
1515	if ( ti->ti_len
1516	&& (unsigned)ti->ti_len <= 5
1517	&& ((struct tcpiphdr_2 *)ti)->first_char == (char)27)
1518	{
1519	tp->t_flags \|= TF_ACKNOW;
1520	}
1521
1522	/*
1523	* Process the segment text, merging it into the TCP sequencing queue,
1524	* and arranging for acknowledgment of receipt if necessary.
1525	* This process logically involves adjusting tp->rcv_wnd as data
1526	* is presented to the user (this happens in tcp_usrreq.c,
1527	* case PRU_RCVD). If a FIN has already been received on this
1528	* connection then we just ignore the text.
1529	*/
1530	if ( (ti->ti_len \|\| (tiflags&TH_FIN))
1531	&& TCPS_HAVERCVDFIN(tp->t_state) == 0)
1532	{
1533	if ( ti->ti_seq == tp->rcv_nxt
1534	&& LIST_EMPTY(&tp->t_segq)
1535	&& tp->t_state == TCPS_ESTABLISHED)
1536	{
1537	DELAY_ACK(tp, ti); /* little bit different from BSD declaration see netinet/tcp_input.c */
1538	tp->rcv_nxt += tlen;
1539	tiflags = ti->ti_t.th_flags & TH_FIN;
1540	tcpstat.tcps_rcvpack++;
1541	tcpstat.tcps_rcvbyte += tlen;
1542	if (so->so_state & SS_FCANTRCVMORE)
1543	m_freem(pData, m);
1544	else
1545	sbappend(pData, so, m);
1546	}
1547	else
1548	{
1549	tiflags = tcp_reass(pData, tp, &ti->ti_t, &tlen, m);
1550	tiflags \|= TF_ACKNOW;
1551	}
1552	/*
1553	* Note the amount of data that peer has sent into
1554	* our window, in order to estimate the sender's
1555	* buffer size.
1556	*/
1557	len = so->so_rcv.sb_datalen - (tp->rcv_adv - tp->rcv_nxt);
1558	}
1559	else
1560	{
1561	m_freem(pData, m);
1562	tiflags &= ~TH_FIN;
1563	}
1564
1565	/*
1566	* If FIN is received ACK the FIN and let the user know
1567	* that the connection is closing.
1568	*/
1569	if (tiflags & TH_FIN)
1570	{
1571	if (TCPS_HAVERCVDFIN(tp->t_state) == 0)
1572	{
1573	/*
1574	* If we receive a FIN we can't send more data,
1575	* set it SS_FDRAIN
1576	* Shutdown the socket if there is no rx data in the
1577	* buffer.
1578	* soread() is called on completion of shutdown() and
1579	* will got to TCPS_LAST_ACK, and use tcp_output()
1580	* to send the FIN.
1581	*/
1582	/* sofcantrcvmore(so); */
1583	sofwdrain(so);
1584
1585	tp->t_flags \|= TF_ACKNOW;
1586	tp->rcv_nxt++;
1587	}
1588	switch (tp->t_state)
1589	{
1590	/*
1591	* In SYN_RECEIVED and ESTABLISHED STATES
1592	* enter the CLOSE_WAIT state.
1593	*/
1594	case TCPS_SYN_RECEIVED:
1595	case TCPS_ESTABLISHED:
1596	if(so->so_emu == EMU_CTL) /* no shutdown on socket */
1597	tp->t_state = TCPS_LAST_ACK;
1598	else
1599	tp->t_state = TCPS_CLOSE_WAIT;
1600	break;
1601
1602	/*
1603	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1604	* enter the CLOSING state.
1605	*/
1606	case TCPS_FIN_WAIT_1:
1607	tp->t_state = TCPS_CLOSING;
1608	break;
1609
1610	/*
1611	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1612	* starting the time-wait timer, turning off the other
1613	* standard timers.
1614	*/
1615	case TCPS_FIN_WAIT_2:
1616	tp->t_state = TCPS_TIME_WAIT;
1617	tcp_canceltimers(tp);
1618	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1619	soisfdisconnected(so);
1620	break;
1621
1622	/*
1623	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1624	*/
1625	case TCPS_TIME_WAIT:
1626	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1627	break;
1628	}
1629	}
1630
1631	/*
1632	* Return any desired output.
1633	*/
1634	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1635	tcp_output(pData, tp);
1636
1637	SOCKET_UNLOCK(so);
1638	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1639	return;
1640
1641	dropafterack:
1642	Log2(("drop after ack\n"));
1643	/*
1644	* Generate an ACK dropping incoming segment if it occupies
1645	* sequence space, where the ACK reflects our state.
1646	*/
1647	if (tiflags & TH_RST)
1648	goto drop;
1649	m_freem(pData, m);
1650	tp->t_flags \|= TF_ACKNOW;
1651	(void) tcp_output(pData, tp);
1652	SOCKET_UNLOCK(so);
1653	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1654	return;
1655
1656	dropwithreset:
1657	/* reuses m if m!=NULL, m_free() unnecessary */
1658	if (tiflags & TH_ACK)
1659	tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1660	else
1661	{
1662	if (tiflags & TH_SYN) ti->ti_len++;
1663	tcp_respond(pData, tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1664	TH_RST\|TH_ACK);
1665	}
1666
1667	if (so != &tcb)
1668	SOCKET_UNLOCK(so);
1669	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1670	return;
1671
1672	drop:
1673	/*
1674	* Drop space held by incoming segment and return.
1675	*/
1676	m_freem(pData, m);
1677
1678	#ifdef VBOX_WITH_SLIRP_MT
1679	if (RTCritSectIsOwned(&so->so_mutex))
1680	{
1681	SOCKET_UNLOCK(so);
1682	}
1683	#endif
1684
1685	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1686	return;
1687	}
1688
1689	void
1690	tcp_dooptions(PNATState pData, struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti)
1691	{
1692	u_int16_t mss;
1693	int opt, optlen;
1694
1695	DEBUG_CALL("tcp_dooptions");
1696	DEBUG_ARGS((dfd," tp = %lx cnt=%i \n", (long )tp, cnt));
1697
1698	for (; cnt > 0; cnt -= optlen, cp += optlen)
1699	{
1700	opt = cp[0];
1701	if (opt == TCPOPT_EOL)
1702	break;
1703	if (opt == TCPOPT_NOP)
1704	optlen = 1;
1705	else
1706	{
1707	optlen = cp[1];
1708	if (optlen <= 0)
1709	break;
1710	}
1711	switch (opt)
1712	{
1713	default:
1714	continue;
1715
1716	case TCPOPT_MAXSEG:
1717	if (optlen != TCPOLEN_MAXSEG)
1718	continue;
1719	if (!(ti->ti_flags & TH_SYN))
1720	continue;
1721	memcpy((char ) &mss, (char ) cp + 2, sizeof(mss));
1722	NTOHS(mss);
1723	(void) tcp_mss(pData, tp, mss); /* sets t_maxseg */
1724	break;
1725
1726	#if 0
1727	case TCPOPT_WINDOW:
1728	if (optlen != TCPOLEN_WINDOW)
1729	continue;
1730	if (!(ti->ti_flags & TH_SYN))
1731	continue;
1732	tp->t_flags \|= TF_RCVD_SCALE;
1733	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1734	break;
1735
1736	case TCPOPT_TIMESTAMP:
1737	if (optlen != TCPOLEN_TIMESTAMP)
1738	continue;
1739	*ts_present = 1;
1740	memcpy((char ) ts_val, (char )cp + 2, sizeof(*ts_val));
1741	NTOHL(*ts_val);
1742	memcpy((char ) ts_ecr, (char )cp + 6, sizeof(*ts_ecr));
1743	NTOHL(*ts_ecr);
1744
1745	/*
1746	* A timestamp received in a SYN makes
1747	* it ok to send timestamp requests and replies.
1748	*/
1749	if (ti->ti_flags & TH_SYN)
1750	{
1751	tp->t_flags \|= TF_RCVD_TSTMP;
1752	tp->ts_recent = *ts_val;
1753	tp->ts_recent_age = tcp_now;
1754	}
1755	break;
1756	#endif
1757	}
1758	}
1759	}
1760
1761
1762	/*
1763	* Pull out of band byte out of a segment so
1764	* it doesn't appear in the user's data queue.
1765	* It is still reflected in the segment length for
1766	* sequencing purposes.
1767	*/
1768
1769	#if 0
1770	void
1771	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1772	{
1773	int cnt = ti->ti_urp - 1;
1774
1775	while (cnt >= 0)
1776	{
1777	if (m->m_len > cnt)
1778	{
1779	char *cp = mtod(m, caddr_t) + cnt;
1780	struct tcpcb *tp = sototcpcb(so);
1781
1782	tp->t_iobc = *cp;
1783	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1784	memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1785	m->m_len--;
1786	return;
1787	}
1788	cnt -= m->m_len;
1789	m = m->m_next; /* XXX WRONG! Fix it! */
1790	if (m == 0)
1791	break;
1792	}
1793	panic("tcp_pulloutofband");
1794	}
1795	#endif
1796
1797	/*
1798	* Collect new round-trip time estimate
1799	* and update averages and current timeout.
1800	*/
1801
1802	void
1803	tcp_xmit_timer(PNATState pData, register struct tcpcb *tp, int rtt)
1804	{
1805	register short delta;
1806
1807	DEBUG_CALL("tcp_xmit_timer");
1808	DEBUG_ARG("tp = %lx", (long)tp);
1809	DEBUG_ARG("rtt = %d", rtt);
1810
1811	tcpstat.tcps_rttupdated++;
1812	if (tp->t_srtt != 0)
1813	{
1814	/*
1815	* srtt is stored as fixed point with 3 bits after the
1816	* binary point (i.e., scaled by 8). The following magic
1817	* is equivalent to the smoothing algorithm in rfc793 with
1818	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1819	* point). Adjust rtt to origin 0.
1820	*/
1821	delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1822	if ((tp->t_srtt += delta) <= 0)
1823	tp->t_srtt = 1;
1824	/*
1825	* We accumulate a smoothed rtt variance (actually, a
1826	* smoothed mean difference), then set the retransmit
1827	* timer to smoothed rtt + 4 times the smoothed variance.
1828	* rttvar is stored as fixed point with 2 bits after the
1829	* binary point (scaled by 4). The following is
1830	* equivalent to rfc793 smoothing with an alpha of .75
1831	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1832	* rfc793's wired-in beta.
1833	*/
1834	if (delta < 0)
1835	delta = -delta;
1836	delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1837	if ((tp->t_rttvar += delta) <= 0)
1838	tp->t_rttvar = 1;
1839	}
1840	else
1841	{
1842	/*
1843	* No rtt measurement yet - use the unsmoothed rtt.
1844	* Set the variance to half the rtt (so our first
1845	* retransmit happens at 3*rtt).
1846	*/
1847	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1848	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1849	}
1850	tp->t_rtt = 0;
1851	tp->t_rxtshift = 0;
1852
1853	/*
1854	* the retransmit should happen at rtt + 4 * rttvar.
1855	* Because of the way we do the smoothing, srtt and rttvar
1856	* will each average +1/2 tick of bias. When we compute
1857	* the retransmit timer, we want 1/2 tick of rounding and
1858	* 1 extra tick because of +-1/2 tick uncertainty in the
1859	* firing of the timer. The bias will give us exactly the
1860	* 1.5 tick we need. But, because the bias is
1861	* statistical, we have to test that we don't drop below
1862	* the minimum feasible timer (which is 2 ticks).
1863	*/
1864	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1865	(short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1866
1867	/*
1868	* We received an ack for a packet that wasn't retransmitted;
1869	* it is probably safe to discard any error indications we've
1870	* received recently. This isn't quite right, but close enough
1871	* for now (a route might have failed after we sent a segment,
1872	* and the return path might not be symmetrical).
1873	*/
1874	tp->t_softerror = 0;
1875	}
1876
1877	/*
1878	* Determine a reasonable value for maxseg size.
1879	* If the route is known, check route for mtu.
1880	* If none, use an mss that can be handled on the outgoing
1881	* interface without forcing IP to fragment; if bigger than
1882	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1883	* to utilize large mbufs. If no route is found, route has no mtu,
1884	* or the destination isn't local, use a default, hopefully conservative
1885	* size (usually 512 or the default IP max size, but no more than the mtu
1886	* of the interface), as we can't discover anything about intervening
1887	* gateways or networks. We also initialize the congestion/slow start
1888	* window to be a single segment if the destination isn't local.
1889	* While looking at the routing entry, we also initialize other path-dependent
1890	* parameters from pre-set or cached values in the routing entry.
1891	*/
1892
1893	int
1894	tcp_mss(PNATState pData, register struct tcpcb *tp, u_int offer)
1895	{
1896	struct socket *so = tp->t_socket;
1897	int mss;
1898
1899	DEBUG_CALL("tcp_mss");
1900	DEBUG_ARG("tp = %lx", (long)tp);
1901	DEBUG_ARG("offer = %d", offer);
1902
1903	mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
1904	if (offer)
1905	mss = min(mss, offer);
1906	mss = max(mss, 32);
1907	if (mss < tp->t_maxseg \|\| offer != 0)
1908	tp->t_maxseg = mss;
1909
1910	tp->snd_cwnd = mss;
1911
1912	sbreserve(pData, &so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
1913	sbreserve(pData, &so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
1914
1915	DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1916
1917	return mss;
1918	}

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 28480

Download in other formats: