tcp_input.c@ 29603

Last change on this file since 29603 was 28800, checked in by vboxsync, 15 years ago
Automated rebranding to Oracle copyright/license strings via filemuncher
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 61.5 KB

Line
1	/* $Id: tcp_input.c 28800 2010-04-27 08:22:32Z vboxsync $ */
2	/** @file
3	* NAT - TCP input.
4	*/
5
6	/*
7	* Copyright (C) 2006-2010 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	#define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ)
69
70	/* for modulo comparisons of timestamps */
71	#define TSTMP_LT(a, b) ((int)((a)-(b)) < 0)
72	#define TSTMP_GEQ(a, b) ((int)((a)-(b)) >= 0)
73
74	#ifndef TCP_ACK_HACK
75	#define DELAY_ACK(tp, ti) \
76	if (ti->ti_flags & TH_PUSH) \
77	tp->t_flags \|= TF_ACKNOW; \
78	else \
79	tp->t_flags \|= TF_DELACK;
80	#else /* !TCP_ACK_HACK */
81	#define DELAY_ACK(tp, ign) \
82	tp->t_flags \|= TF_DELACK;
83	#endif /* TCP_ACK_HACK */
84
85
86	/*
87	* deps: netinet/tcp_reass.c
88	* tcp_reass_maxqlen = 48 (deafault)
89	* tcp_reass_maxseg = nmbclusters/16 (nmbclusters = 1024 + maxusers * 64 from kern/kern_mbuf.c let's say 256)
90	*/
91	int
92	tcp_reass(PNATState pData, struct tcpcb tp, struct tcphdr th, int tlenp, struct mbuf m)
93	{
94	struct tseg_qent *q;
95	struct tseg_qent *p = NULL;
96	struct tseg_qent *nq;
97	struct tseg_qent *te = NULL;
98	struct socket *so = tp->t_socket;
99	int flags;
100	STAM_PROFILE_START(&pData->StatTCP_reassamble, tcp_reassamble);
101
102	/*
103	* XXX: tcp_reass() is rather inefficient with its data structures
104	* and should be rewritten (see NetBSD for optimizations). While
105	* doing that it should move to its own file tcp_reass.c.
106	*/
107
108	/*
109	* Call with th==NULL after become established to
110	* force pre-ESTABLISHED data up to user socket.
111	*/
112	if (th == NULL)
113	goto present;
114
115	/*
116	* Limit the number of segments in the reassembly queue to prevent
117	* holding on to too many segments (and thus running out of mbufs).
118	* Make sure to let the missing segment through which caused this
119	* queue. Always keep one global queue entry spare to be able to
120	* process the missing segment.
121	*/
122	if ( th->th_seq != tp->rcv_nxt
123	&& ( tcp_reass_qsize + 1 >= tcp_reass_maxseg
124	\|\| tp->t_segqlen >= tcp_reass_maxqlen))
125	{
126	tcp_reass_overflows++;
127	tcpstat.tcps_rcvmemdrop++;
128	m_freem(pData, m);
129	*tlenp = 0;
130	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
131	return (0);
132	}
133
134	/*
135	* Allocate a new queue entry. If we can't, or hit the zone limit
136	* just drop the pkt.
137	*/
138	te = RTMemAlloc(sizeof(struct tseg_qent));
139	if (te == NULL)
140	{
141	tcpstat.tcps_rcvmemdrop++;
142	m_freem(pData, m);
143	*tlenp = 0;
144	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
145	return (0);
146	}
147	tp->t_segqlen++;
148	tcp_reass_qsize++;
149
150	/*
151	* Find a segment which begins after this one does.
152	*/
153	LIST_FOREACH(q, &tp->t_segq, tqe_q)
154	{
155	if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
156	break;
157	p = q;
158	}
159
160	/*
161	* If there is a preceding segment, it may provide some of
162	* our data already. If so, drop the data from the incoming
163	* segment. If it provides all of our data, drop us.
164	*/
165	if (p != NULL)
166	{
167	int i;
168	/* conversion to int (in i) handles seq wraparound */
169	i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
170	if (i > 0)
171	{
172	if (i >= *tlenp)
173	{
174	tcpstat.tcps_rcvduppack++;
175	tcpstat.tcps_rcvdupbyte += *tlenp;
176	m_freem(pData, m);
177	RTMemFree(te);
178	tp->t_segqlen--;
179	tcp_reass_qsize--;
180	/*
181	* Try to present any queued data
182	* at the left window edge to the user.
183	* This is needed after the 3-WHS
184	* completes.
185	*/
186	goto present; /* ??? */
187	}
188	m_adj(m, i);
189	*tlenp -= i;
190	th->th_seq += i;
191	}
192	}
193	tcpstat.tcps_rcvoopack++;
194	tcpstat.tcps_rcvoobyte += *tlenp;
195
196	/*
197	* While we overlap succeeding segments trim them or,
198	* if they are completely covered, dequeue them.
199	*/
200	while (q)
201	{
202	int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
203	if (i <= 0)
204	break;
205	if (i < q->tqe_len)
206	{
207	q->tqe_th->th_seq += i;
208	q->tqe_len -= i;
209	m_adj(q->tqe_m, i);
210	break;
211	}
212
213	nq = LIST_NEXT(q, tqe_q);
214	LIST_REMOVE(q, tqe_q);
215	m_freem(pData, q->tqe_m);
216	RTMemFree(q);
217	tp->t_segqlen--;
218	tcp_reass_qsize--;
219	q = nq;
220	}
221
222	/* Insert the new segment queue entry into place. */
223	te->tqe_m = m;
224	te->tqe_th = th;
225	te->tqe_len = *tlenp;
226
227	if (p == NULL)
228	{
229	LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
230	}
231	else
232	{
233	LIST_INSERT_AFTER(p, te, tqe_q);
234	}
235
236	present:
237	/*
238	* Present data to user, advancing rcv_nxt through
239	* completed sequence space.
240	*/
241	if (!TCPS_HAVEESTABLISHED(tp->t_state))
242	{
243	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
244	return (0);
245	}
246	q = LIST_FIRST(&tp->t_segq);
247	if (!q \|\| q->tqe_th->th_seq != tp->rcv_nxt)
248	{
249	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
250	return (0);
251	}
252	do
253	{
254	tp->rcv_nxt += q->tqe_len;
255	flags = q->tqe_th->th_flags & TH_FIN;
256	nq = LIST_NEXT(q, tqe_q);
257	LIST_REMOVE(q, tqe_q);
258	/* XXX: This place should be checked for the same code in
259	* original BSD code for Slirp and current BSD used SS_FCANTRCVMORE
260	*/
261	if (so->so_state & SS_FCANTSENDMORE)
262	m_freem(pData, q->tqe_m);
263	else
264	sbappend(pData, so, q->tqe_m);
265	RTMemFree(q);
266	tp->t_segqlen--;
267	tcp_reass_qsize--;
268	q = nq;
269	}
270	while (q && q->tqe_th->th_seq == tp->rcv_nxt);
271
272	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
273	return flags;
274	}
275
276	/*
277	* TCP input routine, follows pages 65-76 of the
278	* protocol specification dated September, 1981 very closely.
279	*/
280	void
281	tcp_input(PNATState pData, register struct mbuf m, int iphlen, struct socket inso)
282	{
283	struct ip save_ip, *ip;
284	register struct tcpiphdr *ti;
285	caddr_t optp = NULL;
286	int optlen = 0;
287	int len, tlen, off;
288	register struct tcpcb *tp = 0;
289	register int tiflags;
290	struct socket *so = 0;
291	int todrop, acked, ourfinisacked, needoutput = 0;
292	/* int dropsocket = 0; */
293	int iss = 0;
294	u_long tiwin;
295	/* int ts_present = 0; */
296	STAM_PROFILE_START(&pData->StatTCP_input, counter_input);
297
298	DEBUG_CALL("tcp_input");
299	DEBUG_ARGS((dfd," m = %8lx iphlen = %2d inso = %lx\n",
300	(long )m, iphlen, (long )inso ));
301
302	if (inso != NULL)
303	{
304	QSOCKET_LOCK(tcb);
305	SOCKET_LOCK(inso);
306	QSOCKET_UNLOCK(tcb);
307	}
308	/*
309	* If called with m == 0, then we're continuing the connect
310	*/
311	if (m == NULL)
312	{
313	so = inso;
314	Log4(("NAT: tcp_input: %R[natsock]\n", so));
315	/* Re-set a few variables */
316	tp = sototcpcb(so);
317	m = so->so_m;
318
319	so->so_m = 0;
320	ti = so->so_ti;
321
322	/** @todo (vvl) clarify why it might happens */
323	if (ti == NULL)
324	{
325	LogRel(("NAT: ti is null. can't do any reseting connection actions\n"));
326	/* mbuf should be cleared in sofree called from tcp_close */
327	tcp_close(pData, tp);
328	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
329	return;
330	}
331
332	tiwin = ti->ti_win;
333	tiflags = ti->ti_flags;
334
335	goto cont_conn;
336	}
337
338	tcpstat.tcps_rcvtotal++;
339	/*
340	* Get IP and TCP header together in first mbuf.
341	* Note: IP leaves IP header in first mbuf.
342	*/
343	ti = mtod(m, struct tcpiphdr *);
344	if (iphlen > sizeof(struct ip ))
345	{
346	ip_stripoptions(m, (struct mbuf *)0);
347	iphlen = sizeof(struct ip );
348	}
349	/* XXX Check if too short */
350
351
352	/*
353	* Save a copy of the IP header in case we want restore it
354	* for sending an ICMP error message in response.
355	*/
356	ip = mtod(m, struct ip *);
357	/*
358	* (vvl) ip_input substracts IP header length from ip->ip_len value.
359	* here we do the test the same as input method of UDP protocol.
360	*/
361	#ifdef VBOX_WITH_SLIRP_BSD_MBUF
362	Assert((ip->ip_len + iphlen == m_length(m, NULL)));
363	#else
364	Assert((ip->ip_len + iphlen == m->m_len));
365	#endif
366	save_ip = *ip;
367	save_ip.ip_len+= iphlen;
368
369	/*
370	* Checksum extended TCP header and data.
371	*/
372	tlen = ((struct ip *)ti)->ip_len;
373	memset(ti->ti_x1, 0, 9);
374	ti->ti_len = RT_H2N_U16((u_int16_t)tlen);
375	len = sizeof(struct ip ) + tlen;
376	/* keep checksum for ICMP reply
377	* ti->ti_sum = cksum(m, len);
378	* if (ti->ti_sum) { */
379	if (cksum(m, len))
380	{
381	tcpstat.tcps_rcvbadsum++;
382	goto drop;
383	}
384
385	/*
386	* Check that TCP offset makes sense,
387	* pull out TCP options and adjust length. XXX
388	*/
389	off = ti->ti_off << 2;
390	if ( off < sizeof (struct tcphdr)
391	\|\| off > tlen)
392	{
393	tcpstat.tcps_rcvbadoff++;
394	goto drop;
395	}
396	tlen -= off;
397	ti->ti_len = tlen;
398	if (off > sizeof (struct tcphdr))
399	{
400	optlen = off - sizeof (struct tcphdr);
401	optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
402
403	/*
404	* Do quick retrieval of timestamp options ("options
405	* prediction?"). If timestamp is the only option and it's
406	* formatted as recommended in RFC 1323 appendix A, we
407	* quickly get the values now and not bother calling
408	* tcp_dooptions(), etc.
409	*/
410	#if 0
411	if (( optlen == TCPOLEN_TSTAMP_APPA
412	\|\| ( optlen > TCPOLEN_TSTAMP_APPA
413	&& optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
414	(u_int32_t )optp == RT_H2N_U32_C(TCPOPT_TSTAMP_HDR) &&
415	(ti->ti_flags & TH_SYN) == 0)
416	{
417	ts_present = 1;
418	ts_val = RT_N2H_U32((u_int32_t )(optp + 4));
419	ts_ecr = RT_N2H_U32((u_int32_t )(optp + 8));
420	optp = NULL; / * we have parsed the options * /
421	}
422	#endif
423	}
424	tiflags = ti->ti_flags;
425
426	/*
427	* Convert TCP protocol specific fields to host format.
428	*/
429	NTOHL(ti->ti_seq);
430	NTOHL(ti->ti_ack);
431	NTOHS(ti->ti_win);
432	NTOHS(ti->ti_urp);
433
434	/*
435	* Drop TCP, IP headers and TCP options.
436	*/
437	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
438	m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
439
440	/*
441	* Locate pcb for segment.
442	*/
443	findso:
444	if (so != NULL && so != &tcb)
445	SOCKET_UNLOCK(so);
446	QSOCKET_LOCK(tcb);
447	so = tcp_last_so;
448	if ( so->so_fport != ti->ti_dport
449	\|\| so->so_lport != ti->ti_sport
450	\|\| so->so_laddr.s_addr != ti->ti_src.s_addr
451	\|\| so->so_faddr.s_addr != ti->ti_dst.s_addr)
452	{
453	#ifdef VBOX_WITH_SLIRP_MT
454	struct socket *sonxt;
455	#endif
456	QSOCKET_UNLOCK(tcb);
457	/* @todo fix SOLOOKUP macrodefinition to be usable here */
458	#ifndef VBOX_WITH_SLIRP_MT
459	so = solookup(&tcb, ti->ti_src, ti->ti_sport,
460	ti->ti_dst, ti->ti_dport);
461	#else
462	so = NULL;
463	QSOCKET_FOREACH(so, sonxt, tcp)
464	/* { */
465	if ( so->so_lport == ti->ti_sport
466	&& so->so_laddr.s_addr == ti->ti_src.s_addr
467	&& so->so_faddr.s_addr == ti->ti_dst.s_addr
468	&& so->so_fport == ti->ti_dport
469	&& so->so_deleted != 1)
470	{
471	break; /* so is locked here */
472	}
473	LOOP_LABEL(tcp, so, sonxt);
474	}
475	if (so == &tcb) {
476	so = NULL;
477	}
478	#endif
479	if (so)
480	{
481	tcp_last_so = so;
482	}
483	++tcpstat.tcps_socachemiss;
484	}
485	else
486	{
487	SOCKET_LOCK(so);
488	QSOCKET_UNLOCK(tcb);
489	}
490
491	/*
492	* If the state is CLOSED (i.e., TCB does not exist) then
493	* all data in the incoming segment is discarded.
494	* If the TCB exists but is in CLOSED state, it is embryonic,
495	* but should either do a listen or a connect soon.
496	*
497	* state == CLOSED means we've done socreate() but haven't
498	* attached it to a protocol yet...
499	*
500	* XXX If a TCB does not exist, and the TH_SYN flag is
501	* the only flag set, then create a session, mark it
502	* as if it was LISTENING, and continue...
503	*/
504	if (so == 0)
505	{
506	if ((tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) != TH_SYN)
507	goto dropwithreset;
508
509	if ((so = socreate()) == NULL)
510	goto dropwithreset;
511	if (tcp_attach(pData, so) < 0)
512	{
513	RTMemFree(so); /* Not sofree (if it failed, it's not insqued) */
514	goto dropwithreset;
515	}
516	SOCKET_LOCK(so);
517	sbreserve(pData, &so->so_snd, tcp_sndspace);
518	sbreserve(pData, &so->so_rcv, tcp_rcvspace);
519
520	/* tcp_last_so = so; / / XXX ? */
521	/* tp = sototcpcb(so); */
522
523	so->so_laddr = ti->ti_src;
524	so->so_lport = ti->ti_sport;
525	so->so_faddr = ti->ti_dst;
526	so->so_fport = ti->ti_dport;
527
528	so->so_iptos = ((struct ip *)ti)->ip_tos;
529
530	tp = sototcpcb(so);
531	tp->t_state = TCPS_LISTEN;
532	}
533
534	/*
535	* If this is a still-connecting socket, this probably
536	* a retransmit of the SYN. Whether it's a retransmit SYN
537	* or something else, we nuke it.
538	*/
539	if (so->so_state & SS_ISFCONNECTING)
540	{
541	goto drop;
542	}
543
544	tp = sototcpcb(so);
545
546	/* XXX Should never fail */
547	if (tp == 0)
548	goto dropwithreset;
549	if (tp->t_state == TCPS_CLOSED)
550	{
551	goto drop;
552	}
553
554	/* Unscale the window into a 32-bit value. */
555	/* if ((tiflags & TH_SYN) == 0)
556	* tiwin = ti->ti_win << tp->snd_scale;
557	* else
558	*/
559	tiwin = ti->ti_win;
560
561	/*
562	* Segment received on connection.
563	* Reset idle time and keep-alive timer.
564	*/
565	tp->t_idle = 0;
566	if (so_options)
567	tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
568	else
569	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
570
571	/*
572	* Process options if not in LISTEN state,
573	* else do it below (after getting remote address).
574	*/
575	if (optp && tp->t_state != TCPS_LISTEN)
576	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
577	/* , */
578	/* &ts_present, &ts_val, &ts_ecr); */
579
580	/*
581	* Header prediction: check for the two common cases
582	* of a uni-directional data xfer. If the packet has
583	* no control flags, is in-sequence, the window didn't
584	* change and we're not retransmitting, it's a
585	* candidate. If the length is zero and the ack moved
586	* forward, we're the sender side of the xfer. Just
587	* free the data acked & wake any higher level process
588	* that was blocked waiting for space. If the length
589	* is non-zero and the ack didn't move, we're the
590	* receiver side. If we're getting packets in-order
591	* (the reassembly queue is empty), add the data to
592	* the socket buffer and note that we need a delayed ack.
593	*
594	* XXX Some of these tests are not needed
595	* eg: the tiwin == tp->snd_wnd prevents many more
596	* predictions.. with no real advantage..
597	*/
598	if ( tp->t_state == TCPS_ESTABLISHED
599	&& (tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK
600	/* && (!ts_present \|\| TSTMP_GEQ(ts_val, tp->ts_recent)) */
601	&& ti->ti_seq == tp->rcv_nxt
602	&& tiwin && tiwin == tp->snd_wnd
603	&& tp->snd_nxt == tp->snd_max)
604	{
605	/*
606	* If last ACK falls within this segment's sequence numbers,
607	* record the timestamp.
608	*/
609	#if 0
610	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
611	SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len))
612	{
613	tp->ts_recent_age = tcp_now;
614	tp->ts_recent = ts_val;
615	}
616	#endif
617
618	if (ti->ti_len == 0)
619	{
620	if ( SEQ_GT(ti->ti_ack, tp->snd_una)
621	&& SEQ_LEQ(ti->ti_ack, tp->snd_max)
622	&& tp->snd_cwnd >= tp->snd_wnd)
623	{
624	/*
625	* this is a pure ack for outstanding data.
626	*/
627	++tcpstat.tcps_predack;
628	#if 0
629	if (ts_present)
630	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
631	else
632	#endif
633	if ( tp->t_rtt
634	&& SEQ_GT(ti->ti_ack, tp->t_rtseq))
635	tcp_xmit_timer(pData, tp, tp->t_rtt);
636	acked = ti->ti_ack - tp->snd_una;
637	tcpstat.tcps_rcvackpack++;
638	tcpstat.tcps_rcvackbyte += acked;
639	sbdrop(&so->so_snd, acked);
640	tp->snd_una = ti->ti_ack;
641	m_freem(pData, m);
642
643	/*
644	* If all outstanding data are acked, stop
645	* retransmit timer, otherwise restart timer
646	* using current (possibly backed-off) value.
647	* If process is waiting for space,
648	* wakeup/selwakeup/signal. If data
649	* are ready to send, let tcp_output
650	* decide between more output or persist.
651	*/
652	if (tp->snd_una == tp->snd_max)
653	tp->t_timer[TCPT_REXMT] = 0;
654	else if (tp->t_timer[TCPT_PERSIST] == 0)
655	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
656
657	/*
658	* There's room in so_snd, sowwakup will read()
659	* from the socket if we can
660	*/
661	#if 0
662	if (so->so_snd.sb_flags & SB_NOTIFY)
663	sowwakeup(so);
664	#endif
665	/*
666	* This is called because sowwakeup might have
667	* put data into so_snd. Since we don't so sowwakeup,
668	* we don't need this.. XXX???
669	*/
670	if (so->so_snd.sb_cc)
671	(void) tcp_output(pData, tp);
672
673	SOCKET_UNLOCK(so);
674	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
675	return;
676	}
677	}
678	else if ( ti->ti_ack == tp->snd_una
679	&& LIST_FIRST(&tp->t_segq)
680	&& ti->ti_len <= sbspace(&so->so_rcv))
681	{
682	/*
683	* this is a pure, in-sequence data packet
684	* with nothing on the reassembly queue and
685	* we have enough buffer space to take it.
686	*/
687	++tcpstat.tcps_preddat;
688	tp->rcv_nxt += ti->ti_len;
689	tcpstat.tcps_rcvpack++;
690	tcpstat.tcps_rcvbyte += ti->ti_len;
691	/*
692	* Add data to socket buffer.
693	*/
694	sbappend(pData, so, m);
695
696	/*
697	* XXX This is called when data arrives. Later, check
698	* if we can actually write() to the socket
699	* XXX Need to check? It's be NON_BLOCKING
700	*/
701	/* sorwakeup(so); */
702
703	/*
704	* If this is a short packet, then ACK now - with Nagel
705	* congestion avoidance sender won't send more until
706	* he gets an ACK.
707	*
708	* It is better to not delay acks at all to maximize
709	* TCP throughput. See RFC 2581.
710	*/
711	tp->t_flags \|= TF_ACKNOW;
712	tcp_output(pData, tp);
713	SOCKET_UNLOCK(so);
714	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
715	return;
716	}
717	} /* header prediction */
718	/*
719	* Calculate amount of space in receive window,
720	* and then do TCP input processing.
721	* Receive window is amount of space in rcv queue,
722	* but not less than advertised window.
723	*/
724	{
725	int win;
726	win = sbspace(&so->so_rcv);
727	if (win < 0)
728	win = 0;
729	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
730	}
731
732	switch (tp->t_state)
733	{
734	/*
735	* If the state is LISTEN then ignore segment if it contains an RST.
736	* If the segment contains an ACK then it is bad and send a RST.
737	* If it does not contain a SYN then it is not interesting; drop it.
738	* Don't bother responding if the destination was a broadcast.
739	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
740	* tp->iss, and send a segment:
741	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
742	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
743	* Fill in remote peer address fields if not previously specified.
744	* Enter SYN_RECEIVED state, and process any other fields of this
745	* segment in this state.
746	*/
747	case TCPS_LISTEN:
748	{
749	if (tiflags & TH_RST) {
750	goto drop;
751	}
752	if (tiflags & TH_ACK)
753	goto dropwithreset;
754	if ((tiflags & TH_SYN) == 0)
755	{
756	goto drop;
757	}
758
759	/*
760	* This has way too many gotos...
761	* But a bit of spaghetti code never hurt anybody :)
762	*/
763	if ( (tcp_fconnect(pData, so) == -1)
764	&& errno != EINPROGRESS
765	&& errno != EWOULDBLOCK)
766	{
767	u_char code = ICMP_UNREACH_NET;
768	DEBUG_MISC((dfd," tcp fconnect errno = %d-%s\n",
769	errno, strerror(errno)));
770	if (errno == ECONNREFUSED)
771	{
772	/* ACK the SYN, send RST to refuse the connection */
773	tcp_respond(pData, tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
774	TH_RST\|TH_ACK);
775	}
776	else
777	{
778	if (errno == EHOSTUNREACH)
779	code = ICMP_UNREACH_HOST;
780	HTONL(ti->ti_seq); /* restore tcp header */
781	HTONL(ti->ti_ack);
782	HTONS(ti->ti_win);
783	HTONS(ti->ti_urp);
784	m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
785	m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
786	*ip = save_ip;
787	icmp_error(pData, m, ICMP_UNREACH, code, 0, strerror(errno));
788	tp->t_socket->so_m = NULL;
789	}
790	tp = tcp_close(pData, tp);
791	m_freem(pData, m);
792	}
793	else
794	{
795	/*
796	* Haven't connected yet, save the current mbuf
797	* and ti, and return
798	* XXX Some OS's don't tell us whether the connect()
799	* succeeded or not. So we must time it out.
800	*/
801	so->so_m = m;
802	so->so_ti = ti;
803	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
804	tp->t_state = TCPS_SYN_RECEIVED;
805	}
806	SOCKET_UNLOCK(so);
807	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
808	return;
809
810	cont_conn:
811	/* m==NULL
812	* Check if the connect succeeded
813	*/
814	if (so->so_state & SS_NOFDREF)
815	{
816	tp = tcp_close(pData, tp);
817	goto dropwithreset;
818	}
819	cont_input:
820	tcp_template(tp);
821
822	if (optp)
823	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
824
825	if (iss)
826	tp->iss = iss;
827	else
828	tp->iss = tcp_iss;
829	tcp_iss += TCP_ISSINCR/2;
830	tp->irs = ti->ti_seq;
831	tcp_sendseqinit(tp);
832	tcp_rcvseqinit(tp);
833	tp->t_flags \|= TF_ACKNOW;
834	tp->t_state = TCPS_SYN_RECEIVED;
835	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
836	tcpstat.tcps_accepts++;
837	goto trimthenstep6;
838	} /* case TCPS_LISTEN */
839
840	/*
841	* If the state is SYN_SENT:
842	* if seg contains an ACK, but not for our SYN, drop the input.
843	* if seg contains a RST, then drop the connection.
844	* if seg does not contain SYN, then drop it.
845	* Otherwise this is an acceptable SYN segment
846	* initialize tp->rcv_nxt and tp->irs
847	* if seg contains ack then advance tp->snd_una
848	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
849	* arrange for segment to be acked (eventually)
850	* continue processing rest of data/controls, beginning with URG
851	*/
852	case TCPS_SYN_SENT:
853	if ( (tiflags & TH_ACK)
854	&& ( SEQ_LEQ(ti->ti_ack, tp->iss)
855	\|\| SEQ_GT(ti->ti_ack, tp->snd_max)))
856	goto dropwithreset;
857
858	if (tiflags & TH_RST)
859	{
860	if (tiflags & TH_ACK)
861	tp = tcp_drop(pData, tp, 0); /* XXX Check t_softerror! */
862	goto drop;
863	}
864
865	if ((tiflags & TH_SYN) == 0)
866	{
867	goto drop;
868	}
869	if (tiflags & TH_ACK)
870	{
871	tp->snd_una = ti->ti_ack;
872	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
873	tp->snd_nxt = tp->snd_una;
874	}
875
876	tp->t_timer[TCPT_REXMT] = 0;
877	tp->irs = ti->ti_seq;
878	tcp_rcvseqinit(tp);
879	tp->t_flags \|= TF_ACKNOW;
880	if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss))
881	{
882	tcpstat.tcps_connects++;
883	soisfconnected(so);
884	tp->t_state = TCPS_ESTABLISHED;
885
886	/* Do window scaling on this connection? */
887	#if 0
888	if (( tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
889	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
890	{
891	tp->snd_scale = tp->requested_s_scale;
892	tp->rcv_scale = tp->request_r_scale;
893	}
894	#endif
895	(void) tcp_reass(pData, tp, (struct tcphdr )0, NULL, (struct mbuf )0);
896	/*
897	* if we didn't have to retransmit the SYN,
898	* use its rtt as our initial srtt & rtt var.
899	*/
900	if (tp->t_rtt)
901	tcp_xmit_timer(pData, tp, tp->t_rtt);
902	}
903	else
904	tp->t_state = TCPS_SYN_RECEIVED;
905
906	trimthenstep6:
907	/*
908	* Advance ti->ti_seq to correspond to first data byte.
909	* If data, trim to stay within window,
910	* dropping FIN if necessary.
911	*/
912	ti->ti_seq++;
913	if (ti->ti_len > tp->rcv_wnd)
914	{
915	todrop = ti->ti_len - tp->rcv_wnd;
916	m_adj(m, -todrop);
917	ti->ti_len = tp->rcv_wnd;
918	tiflags &= ~TH_FIN;
919	tcpstat.tcps_rcvpackafterwin++;
920	tcpstat.tcps_rcvbyteafterwin += todrop;
921	}
922	tp->snd_wl1 = ti->ti_seq - 1;
923	tp->rcv_up = ti->ti_seq;
924	Log2(("hit6"));
925	goto step6;
926	} /* switch tp->t_state */
927	/*
928	* States other than LISTEN or SYN_SENT.
929	* First check timestamp, if present.
930	* Then check that at least some bytes of segment are within
931	* receive window. If segment begins before rcv_nxt,
932	* drop leading data (and SYN); if nothing left, just ack.
933	*
934	* RFC 1323 PAWS: If we have a timestamp reply on this segment
935	* and it's less than ts_recent, drop it.
936	*/
937	#if 0
938	if ( ts_present
939	&& (tiflags & TH_RST) == 0
940	&& tp->ts_recent
941	&& TSTMP_LT(ts_val, tp->ts_recent))
942	{
943	/* Check to see if ts_recent is over 24 days old. */
944	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE)
945	{
946	/*
947	* Invalidate ts_recent. If this segment updates
948	* ts_recent, the age will be reset later and ts_recent
949	* will get a valid value. If it does not, setting
950	* ts_recent to zero will at least satisfy the
951	* requirement that zero be placed in the timestamp
952	* echo reply when ts_recent isn't valid. The
953	* age isn't reset until we get a valid ts_recent
954	* because we don't want out-of-order segments to be
955	* dropped when ts_recent is old.
956	*/
957	tp->ts_recent = 0;
958	}
959	else
960	{
961	tcpstat.tcps_rcvduppack++;
962	tcpstat.tcps_rcvdupbyte += ti->ti_len;
963	tcpstat.tcps_pawsdrop++;
964	goto dropafterack;
965	}
966	}
967	#endif
968
969	todrop = tp->rcv_nxt - ti->ti_seq;
970	if (todrop > 0)
971	{
972	if (tiflags & TH_SYN)
973	{
974	tiflags &= ~TH_SYN;
975	ti->ti_seq++;
976	if (ti->ti_urp > 1)
977	ti->ti_urp--;
978	else
979	tiflags &= ~TH_URG;
980	todrop--;
981	}
982	/*
983	* Following if statement from Stevens, vol. 2, p. 960.
984	*/
985	if ( todrop > ti->ti_len
986	\|\| ( todrop == ti->ti_len
987	&& (tiflags & TH_FIN) == 0))
988	{
989	/*
990	* Any valid FIN must be to the left of the window.
991	* At this point the FIN must be a duplicate or out
992	* of sequence; drop it.
993	*/
994	tiflags &= ~TH_FIN;
995
996	/*
997	* Send an ACK to resynchronize and drop any data.
998	* But keep on processing for RST or ACK.
999	*/
1000	tp->t_flags \|= TF_ACKNOW;
1001	todrop = ti->ti_len;
1002	tcpstat.tcps_rcvduppack++;
1003	tcpstat.tcps_rcvdupbyte += todrop;
1004	}
1005	else
1006	{
1007	tcpstat.tcps_rcvpartduppack++;
1008	tcpstat.tcps_rcvpartdupbyte += todrop;
1009	}
1010	m_adj(m, todrop);
1011	ti->ti_seq += todrop;
1012	ti->ti_len -= todrop;
1013	if (ti->ti_urp > todrop)
1014	ti->ti_urp -= todrop;
1015	else
1016	{
1017	tiflags &= ~TH_URG;
1018	ti->ti_urp = 0;
1019	}
1020	}
1021	/*
1022	* If new data are received on a connection after the
1023	* user processes are gone, then RST the other end.
1024	*/
1025	if ( (so->so_state & SS_NOFDREF)
1026	&& tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len)
1027	{
1028	tp = tcp_close(pData, tp);
1029	tcpstat.tcps_rcvafterclose++;
1030	goto dropwithreset;
1031	}
1032
1033	/*
1034	* If segment ends after window, drop trailing data
1035	* (and PUSH and FIN); if nothing left, just ACK.
1036	*/
1037	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1038	if (todrop > 0)
1039	{
1040	tcpstat.tcps_rcvpackafterwin++;
1041	if (todrop >= ti->ti_len)
1042	{
1043	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1044	/*
1045	* If a new connection request is received
1046	* while in TIME_WAIT, drop the old connection
1047	* and start over if the sequence numbers
1048	* are above the previous ones.
1049	*/
1050	if ( tiflags & TH_SYN
1051	&& tp->t_state == TCPS_TIME_WAIT
1052	&& SEQ_GT(ti->ti_seq, tp->rcv_nxt))
1053	{
1054	iss = tp->rcv_nxt + TCP_ISSINCR;
1055	tp = tcp_close(pData, tp);
1056	SOCKET_UNLOCK(tp->t_socket);
1057	goto findso;
1058	}
1059	/*
1060	* If window is closed can only take segments at
1061	* window edge, and have to drop data and PUSH from
1062	* incoming segments. Continue processing, but
1063	* remember to ack. Otherwise, drop segment
1064	* and ack.
1065	*/
1066	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt)
1067	{
1068	tp->t_flags \|= TF_ACKNOW;
1069	tcpstat.tcps_rcvwinprobe++;
1070	}
1071	else
1072	goto dropafterack;
1073	}
1074	else
1075	tcpstat.tcps_rcvbyteafterwin += todrop;
1076	m_adj(m, -todrop);
1077	ti->ti_len -= todrop;
1078	tiflags &= ~(TH_PUSH\|TH_FIN);
1079	}
1080
1081	/*
1082	* If last ACK falls within this segment's sequence numbers,
1083	* record its timestamp.
1084	*/
1085	#if 0
1086	if ( ts_present
1087	&& SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)
1088	&& SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + ((tiflags & (TH_SYN\|TH_FIN)) != 0)))
1089	{
1090	tp->ts_recent_age = tcp_now;
1091	tp->ts_recent = ts_val;
1092	}
1093	#endif
1094
1095	/*
1096	* If the RST bit is set examine the state:
1097	* SYN_RECEIVED STATE:
1098	* If passive open, return to LISTEN state.
1099	* If active open, inform user that connection was refused.
1100	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1101	* Inform user that connection was reset, and close tcb.
1102	* CLOSING, LAST_ACK, TIME_WAIT STATES
1103	* Close the tcb.
1104	*/
1105	if (tiflags&TH_RST)
1106	switch (tp->t_state)
1107	{
1108	case TCPS_SYN_RECEIVED:
1109	/* so->so_error = ECONNREFUSED; */
1110	goto close;
1111
1112	case TCPS_ESTABLISHED:
1113	case TCPS_FIN_WAIT_1:
1114	case TCPS_FIN_WAIT_2:
1115	case TCPS_CLOSE_WAIT:
1116	/* so->so_error = ECONNRESET; */
1117	close:
1118	tp->t_state = TCPS_CLOSED;
1119	tcpstat.tcps_drops++;
1120	tp = tcp_close(pData, tp);
1121	goto drop;
1122
1123	case TCPS_CLOSING:
1124	case TCPS_LAST_ACK:
1125	case TCPS_TIME_WAIT:
1126	tp = tcp_close(pData, tp);
1127	goto drop;
1128	}
1129
1130	/*
1131	* If a SYN is in the window, then this is an
1132	* error and we send an RST and drop the connection.
1133	*/
1134	if (tiflags & TH_SYN)
1135	{
1136	tp = tcp_drop(pData, tp, 0);
1137	goto dropwithreset;
1138	}
1139
1140	/*
1141	* If the ACK bit is off we drop the segment and return.
1142	*/
1143	if ((tiflags & TH_ACK) == 0)
1144	{
1145	goto drop;
1146	}
1147
1148	/*
1149	* Ack processing.
1150	*/
1151	switch (tp->t_state)
1152	{
1153	/*
1154	* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1155	* ESTABLISHED state and continue processing, otherwise
1156	* send an RST. una<=ack<=max
1157	*/
1158	case TCPS_SYN_RECEIVED:
1159	if ( SEQ_GT(tp->snd_una, ti->ti_ack)
1160	\|\| SEQ_GT(ti->ti_ack, tp->snd_max))
1161	goto dropwithreset;
1162	tcpstat.tcps_connects++;
1163	tp->t_state = TCPS_ESTABLISHED;
1164	/*
1165	* The sent SYN is ack'ed with our sequence number +1
1166	* The first data byte already in the buffer will get
1167	* lost if no correction is made. This is only needed for
1168	* SS_CTL since the buffer is empty otherwise.
1169	* tp->snd_una++; or:
1170	*/
1171	tp->snd_una = ti->ti_ack;
1172	soisfconnected(so);
1173
1174	/* Do window scaling? */
1175	#if 0
1176	if ( (tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
1177	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
1178	{
1179	tp->snd_scale = tp->requested_s_scale;
1180	tp->rcv_scale = tp->request_r_scale;
1181	}
1182	#endif
1183	(void) tcp_reass(pData, tp, (struct tcphdr )0, (int )0, (struct mbuf *)0);
1184	tp->snd_wl1 = ti->ti_seq - 1;
1185	/* Avoid ack processing; snd_una==ti_ack => dup ack */
1186	goto synrx_to_est;
1187	/* fall into ... */
1188
1189	/*
1190	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1191	* ACKs. If the ack is in the range
1192	* tp->snd_una < ti->ti_ack <= tp->snd_max
1193	* then advance tp->snd_una to ti->ti_ack and drop
1194	* data from the retransmission queue. If this ACK reflects
1195	* more up to date window information we update our window information.
1196	*/
1197	case TCPS_ESTABLISHED:
1198	case TCPS_FIN_WAIT_1:
1199	case TCPS_FIN_WAIT_2:
1200	case TCPS_CLOSE_WAIT:
1201	case TCPS_CLOSING:
1202	case TCPS_LAST_ACK:
1203	case TCPS_TIME_WAIT:
1204	if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
1205	{
1206	if (ti->ti_len == 0 && tiwin == tp->snd_wnd)
1207	{
1208	tcpstat.tcps_rcvdupack++;
1209	DEBUG_MISC((dfd," dup ack m = %lx so = %lx \n",
1210	(long )m, (long )so));
1211	/*
1212	* If we have outstanding data (other than
1213	* a window probe), this is a completely
1214	* duplicate ack (ie, window info didn't
1215	* change), the ack is the biggest we've
1216	* seen and we've seen exactly our rexmt
1217	* threshold of them, assume a packet
1218	* has been dropped and retransmit it.
1219	* Kludge snd_nxt & the congestion
1220	* window so we send only this one
1221	* packet.
1222	*
1223	* We know we're losing at the current
1224	* window size so do congestion avoidance
1225	* (set ssthresh to half the current window
1226	* and pull our congestion window back to
1227	* the new ssthresh).
1228	*
1229	* Dup acks mean that packets have left the
1230	* network (they're now cached at the receiver)
1231	* so bump cwnd by the amount in the receiver
1232	* to keep a constant cwnd packets in the
1233	* network.
1234	*/
1235	if ( tp->t_timer[TCPT_REXMT] == 0
1236	\|\| ti->ti_ack != tp->snd_una)
1237	tp->t_dupacks = 0;
1238	else if (++tp->t_dupacks == tcprexmtthresh)
1239	{
1240	tcp_seq onxt = tp->snd_nxt;
1241	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
1242	if (win < 2)
1243	win = 2;
1244	tp->snd_ssthresh = win * tp->t_maxseg;
1245	tp->t_timer[TCPT_REXMT] = 0;
1246	tp->t_rtt = 0;
1247	tp->snd_nxt = ti->ti_ack;
1248	tp->snd_cwnd = tp->t_maxseg;
1249	(void) tcp_output(pData, tp);
1250	tp->snd_cwnd = tp->snd_ssthresh +
1251	tp->t_maxseg * tp->t_dupacks;
1252	if (SEQ_GT(onxt, tp->snd_nxt))
1253	tp->snd_nxt = onxt;
1254	goto drop;
1255	}
1256	else if (tp->t_dupacks > tcprexmtthresh)
1257	{
1258	tp->snd_cwnd += tp->t_maxseg;
1259	(void) tcp_output(pData, tp);
1260	goto drop;
1261	}
1262	}
1263	else
1264	tp->t_dupacks = 0;
1265	break;
1266	}
1267	synrx_to_est:
1268	/*
1269	* If the congestion window was inflated to account
1270	* for the other side's cached packets, retract it.
1271	*/
1272	if ( tp->t_dupacks > tcprexmtthresh
1273	&& tp->snd_cwnd > tp->snd_ssthresh)
1274	tp->snd_cwnd = tp->snd_ssthresh;
1275	tp->t_dupacks = 0;
1276	if (SEQ_GT(ti->ti_ack, tp->snd_max))
1277	{
1278	tcpstat.tcps_rcvacktoomuch++;
1279	goto dropafterack;
1280	}
1281	acked = ti->ti_ack - tp->snd_una;
1282	tcpstat.tcps_rcvackpack++;
1283	tcpstat.tcps_rcvackbyte += acked;
1284
1285	/*
1286	* If we have a timestamp reply, update smoothed
1287	* round trip time. If no timestamp is present but
1288	* transmit timer is running and timed sequence
1289	* number was acked, update smoothed round trip time.
1290	* Since we now have an rtt measurement, cancel the
1291	* timer backoff (cf., Phil Karn's retransmit alg.).
1292	* Recompute the initial retransmit timer.
1293	*/
1294	#if 0
1295	if (ts_present)
1296	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1297	else
1298	#endif
1299	if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1300	tcp_xmit_timer(pData, tp, tp->t_rtt);
1301
1302	/*
1303	* If all outstanding data is acked, stop retransmit
1304	* timer and remember to restart (more output or persist).
1305	* If there is more data to be acked, restart retransmit
1306	* timer, using current (possibly backed-off) value.
1307	*/
1308	if (ti->ti_ack == tp->snd_max)
1309	{
1310	tp->t_timer[TCPT_REXMT] = 0;
1311	needoutput = 1;
1312	}
1313	else if (tp->t_timer[TCPT_PERSIST] == 0)
1314	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1315	/*
1316	* When new data is acked, open the congestion window.
1317	* If the window gives us less than ssthresh packets
1318	* in flight, open exponentially (maxseg per packet).
1319	* Otherwise open linearly: maxseg per window
1320	* (maxseg^2 / cwnd per packet).
1321	*/
1322	{
1323	register u_int cw = tp->snd_cwnd;
1324	register u_int incr = tp->t_maxseg;
1325
1326	if (cw > tp->snd_ssthresh)
1327	incr = incr * incr / cw;
1328	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1329	}
1330	if (acked > so->so_snd.sb_cc)
1331	{
1332	tp->snd_wnd -= so->so_snd.sb_cc;
1333	sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1334	ourfinisacked = 1;
1335	}
1336	else
1337	{
1338	sbdrop(&so->so_snd, acked);
1339	tp->snd_wnd -= acked;
1340	ourfinisacked = 0;
1341	}
1342	/*
1343	* XXX sowwakup is called when data is acked and there's room for
1344	* for more data... it should read() the socket
1345	*/
1346	#if 0
1347	if (so->so_snd.sb_flags & SB_NOTIFY)
1348	sowwakeup(so);
1349	#endif
1350	tp->snd_una = ti->ti_ack;
1351	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1352	tp->snd_nxt = tp->snd_una;
1353
1354	switch (tp->t_state)
1355	{
1356	/*
1357	* In FIN_WAIT_1 STATE in addition to the processing
1358	* for the ESTABLISHED state if our FIN is now acknowledged
1359	* then enter FIN_WAIT_2.
1360	*/
1361	case TCPS_FIN_WAIT_1:
1362	if (ourfinisacked)
1363	{
1364	/*
1365	* If we can't receive any more
1366	* data, then closing user can proceed.
1367	* Starting the timer is contrary to the
1368	* specification, but if we don't get a FIN
1369	* we'll hang forever.
1370	*/
1371	if (so->so_state & SS_FCANTRCVMORE)
1372	{
1373	soisfdisconnected(so);
1374	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1375	}
1376	tp->t_state = TCPS_FIN_WAIT_2;
1377	}
1378	break;
1379
1380	/*
1381	* In CLOSING STATE in addition to the processing for
1382	* the ESTABLISHED state if the ACK acknowledges our FIN
1383	* then enter the TIME-WAIT state, otherwise ignore
1384	* the segment.
1385	*/
1386	case TCPS_CLOSING:
1387	if (ourfinisacked)
1388	{
1389	tp->t_state = TCPS_TIME_WAIT;
1390	tcp_canceltimers(tp);
1391	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1392	soisfdisconnected(so);
1393	}
1394	break;
1395
1396	/*
1397	* In LAST_ACK, we may still be waiting for data to drain
1398	* and/or to be acked, as well as for the ack of our FIN.
1399	* If our FIN is now acknowledged, delete the TCB,
1400	* enter the closed state and return.
1401	*/
1402	case TCPS_LAST_ACK:
1403	if (ourfinisacked)
1404	{
1405	tp = tcp_close(pData, tp);
1406	goto drop;
1407	}
1408	break;
1409
1410	/*
1411	* In TIME_WAIT state the only thing that should arrive
1412	* is a retransmission of the remote FIN. Acknowledge
1413	* it and restart the finack timer.
1414	*/
1415	case TCPS_TIME_WAIT:
1416	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1417	goto dropafterack;
1418	}
1419	} /* switch(tp->t_state) */
1420
1421	step6:
1422	/*
1423	* Update window information.
1424	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1425	*/
1426	if ( (tiflags & TH_ACK)
1427	&& ( SEQ_LT(tp->snd_wl1, ti->ti_seq)
1428	\|\| ( tp->snd_wl1 == ti->ti_seq
1429	&& ( SEQ_LT(tp->snd_wl2, ti->ti_ack)
1430	\|\| ( tp->snd_wl2 == ti->ti_ack
1431	&& tiwin > tp->snd_wnd)))))
1432	{
1433	/* keep track of pure window updates */
1434	if ( ti->ti_len == 0
1435	&& tp->snd_wl2 == ti->ti_ack
1436	&& tiwin > tp->snd_wnd)
1437	tcpstat.tcps_rcvwinupd++;
1438	tp->snd_wnd = tiwin;
1439	tp->snd_wl1 = ti->ti_seq;
1440	tp->snd_wl2 = ti->ti_ack;
1441	if (tp->snd_wnd > tp->max_sndwnd)
1442	tp->max_sndwnd = tp->snd_wnd;
1443	needoutput = 1;
1444	}
1445
1446	/*
1447	* Process segments with URG.
1448	*/
1449	if ((tiflags & TH_URG) && ti->ti_urp &&
1450	TCPS_HAVERCVDFIN(tp->t_state) == 0)
1451	{
1452	/*
1453	* This is a kludge, but if we receive and accept
1454	* random urgent pointers, we'll crash in
1455	* soreceive. It's hard to imagine someone
1456	* actually wanting to send this much urgent data.
1457	*/
1458	if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen)
1459	{
1460	ti->ti_urp = 0;
1461	tiflags &= ~TH_URG;
1462	goto dodata;
1463	}
1464	/*
1465	* If this segment advances the known urgent pointer,
1466	* then mark the data stream. This should not happen
1467	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1468	* a FIN has been received from the remote side.
1469	* In these states we ignore the URG.
1470	*
1471	* According to RFC961 (Assigned Protocols),
1472	* the urgent pointer points to the last octet
1473	* of urgent data. We continue, however,
1474	* to consider it to indicate the first octet
1475	* of data past the urgent section as the original
1476	* spec states (in one of two places).
1477	*/
1478	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up))
1479	{
1480	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1481	so->so_urgc = so->so_rcv.sb_cc +
1482	(tp->rcv_up - tp->rcv_nxt); /* -1; */
1483	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1484	}
1485	}
1486	else
1487	/*
1488	* If no out of band data is expected,
1489	* pull receive urgent pointer along
1490	* with the receive window.
1491	*/
1492	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1493	tp->rcv_up = tp->rcv_nxt;
1494	dodata:
1495
1496	/*
1497	* If this is a small packet, then ACK now - with Nagel
1498	* congestion avoidance sender won't send more until
1499	* he gets an ACK.
1500	*
1501	* See above.
1502	*/
1503	if ( ti->ti_len
1504	&& (unsigned)ti->ti_len <= 5
1505	&& ((struct tcpiphdr_2 *)ti)->first_char == (char)27)
1506	{
1507	tp->t_flags \|= TF_ACKNOW;
1508	}
1509
1510	/*
1511	* Process the segment text, merging it into the TCP sequencing queue,
1512	* and arranging for acknowledgment of receipt if necessary.
1513	* This process logically involves adjusting tp->rcv_wnd as data
1514	* is presented to the user (this happens in tcp_usrreq.c,
1515	* case PRU_RCVD). If a FIN has already been received on this
1516	* connection then we just ignore the text.
1517	*/
1518	if ( (ti->ti_len \|\| (tiflags&TH_FIN))
1519	&& TCPS_HAVERCVDFIN(tp->t_state) == 0)
1520	{
1521	if ( ti->ti_seq == tp->rcv_nxt
1522	&& LIST_EMPTY(&tp->t_segq)
1523	&& tp->t_state == TCPS_ESTABLISHED)
1524	{
1525	DELAY_ACK(tp, ti); /* little bit different from BSD declaration see netinet/tcp_input.c */
1526	tp->rcv_nxt += tlen;
1527	tiflags = ti->ti_t.th_flags & TH_FIN;
1528	tcpstat.tcps_rcvpack++;
1529	tcpstat.tcps_rcvbyte += tlen;
1530	if (so->so_state & SS_FCANTRCVMORE)
1531	m_freem(pData, m);
1532	else
1533	sbappend(pData, so, m);
1534	}
1535	else
1536	{
1537	tiflags = tcp_reass(pData, tp, &ti->ti_t, &tlen, m);
1538	tiflags \|= TF_ACKNOW;
1539	}
1540	/*
1541	* Note the amount of data that peer has sent into
1542	* our window, in order to estimate the sender's
1543	* buffer size.
1544	*/
1545	len = so->so_rcv.sb_datalen - (tp->rcv_adv - tp->rcv_nxt);
1546	}
1547	else
1548	{
1549	m_freem(pData, m);
1550	tiflags &= ~TH_FIN;
1551	}
1552
1553	/*
1554	* If FIN is received ACK the FIN and let the user know
1555	* that the connection is closing.
1556	*/
1557	if (tiflags & TH_FIN)
1558	{
1559	if (TCPS_HAVERCVDFIN(tp->t_state) == 0)
1560	{
1561	/*
1562	* If we receive a FIN we can't send more data,
1563	* set it SS_FDRAIN
1564	* Shutdown the socket if there is no rx data in the
1565	* buffer.
1566	* soread() is called on completion of shutdown() and
1567	* will got to TCPS_LAST_ACK, and use tcp_output()
1568	* to send the FIN.
1569	*/
1570	/* sofcantrcvmore(so); */
1571	sofwdrain(so);
1572
1573	tp->t_flags \|= TF_ACKNOW;
1574	tp->rcv_nxt++;
1575	}
1576	switch (tp->t_state)
1577	{
1578	/*
1579	* In SYN_RECEIVED and ESTABLISHED STATES
1580	* enter the CLOSE_WAIT state.
1581	*/
1582	case TCPS_SYN_RECEIVED:
1583	case TCPS_ESTABLISHED:
1584	tp->t_state = TCPS_CLOSE_WAIT;
1585	break;
1586
1587	/*
1588	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1589	* enter the CLOSING state.
1590	*/
1591	case TCPS_FIN_WAIT_1:
1592	tp->t_state = TCPS_CLOSING;
1593	break;
1594
1595	/*
1596	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1597	* starting the time-wait timer, turning off the other
1598	* standard timers.
1599	*/
1600	case TCPS_FIN_WAIT_2:
1601	tp->t_state = TCPS_TIME_WAIT;
1602	tcp_canceltimers(tp);
1603	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1604	soisfdisconnected(so);
1605	break;
1606
1607	/*
1608	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1609	*/
1610	case TCPS_TIME_WAIT:
1611	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1612	break;
1613	}
1614	}
1615
1616	/*
1617	* Return any desired output.
1618	*/
1619	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1620	tcp_output(pData, tp);
1621
1622	SOCKET_UNLOCK(so);
1623	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1624	return;
1625
1626	dropafterack:
1627	Log2(("drop after ack\n"));
1628	/*
1629	* Generate an ACK dropping incoming segment if it occupies
1630	* sequence space, where the ACK reflects our state.
1631	*/
1632	if (tiflags & TH_RST)
1633	goto drop;
1634	m_freem(pData, m);
1635	tp->t_flags \|= TF_ACKNOW;
1636	(void) tcp_output(pData, tp);
1637	SOCKET_UNLOCK(so);
1638	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1639	return;
1640
1641	dropwithreset:
1642	/* reuses m if m!=NULL, m_free() unnecessary */
1643	if (tiflags & TH_ACK)
1644	tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1645	else
1646	{
1647	if (tiflags & TH_SYN) ti->ti_len++;
1648	tcp_respond(pData, tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1649	TH_RST\|TH_ACK);
1650	}
1651
1652	if (so != &tcb)
1653	SOCKET_UNLOCK(so);
1654	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1655	return;
1656
1657	drop:
1658	/*
1659	* Drop space held by incoming segment and return.
1660	*/
1661	m_freem(pData, m);
1662
1663	#ifdef VBOX_WITH_SLIRP_MT
1664	if (RTCritSectIsOwned(&so->so_mutex))
1665	{
1666	SOCKET_UNLOCK(so);
1667	}
1668	#endif
1669
1670	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1671	return;
1672	}
1673
1674	void
1675	tcp_dooptions(PNATState pData, struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti)
1676	{
1677	u_int16_t mss;
1678	int opt, optlen;
1679
1680	DEBUG_CALL("tcp_dooptions");
1681	DEBUG_ARGS((dfd," tp = %lx cnt=%i \n", (long )tp, cnt));
1682
1683	for (; cnt > 0; cnt -= optlen, cp += optlen)
1684	{
1685	opt = cp[0];
1686	if (opt == TCPOPT_EOL)
1687	break;
1688	if (opt == TCPOPT_NOP)
1689	optlen = 1;
1690	else
1691	{
1692	optlen = cp[1];
1693	if (optlen <= 0)
1694	break;
1695	}
1696	switch (opt)
1697	{
1698	default:
1699	continue;
1700
1701	case TCPOPT_MAXSEG:
1702	if (optlen != TCPOLEN_MAXSEG)
1703	continue;
1704	if (!(ti->ti_flags & TH_SYN))
1705	continue;
1706	memcpy((char ) &mss, (char ) cp + 2, sizeof(mss));
1707	NTOHS(mss);
1708	(void) tcp_mss(pData, tp, mss); /* sets t_maxseg */
1709	break;
1710
1711	#if 0
1712	case TCPOPT_WINDOW:
1713	if (optlen != TCPOLEN_WINDOW)
1714	continue;
1715	if (!(ti->ti_flags & TH_SYN))
1716	continue;
1717	tp->t_flags \|= TF_RCVD_SCALE;
1718	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1719	break;
1720
1721	case TCPOPT_TIMESTAMP:
1722	if (optlen != TCPOLEN_TIMESTAMP)
1723	continue;
1724	*ts_present = 1;
1725	memcpy((char ) ts_val, (char )cp + 2, sizeof(*ts_val));
1726	NTOHL(*ts_val);
1727	memcpy((char ) ts_ecr, (char )cp + 6, sizeof(*ts_ecr));
1728	NTOHL(*ts_ecr);
1729
1730	/*
1731	* A timestamp received in a SYN makes
1732	* it ok to send timestamp requests and replies.
1733	*/
1734	if (ti->ti_flags & TH_SYN)
1735	{
1736	tp->t_flags \|= TF_RCVD_TSTMP;
1737	tp->ts_recent = *ts_val;
1738	tp->ts_recent_age = tcp_now;
1739	}
1740	break;
1741	#endif
1742	}
1743	}
1744	}
1745
1746
1747	/*
1748	* Pull out of band byte out of a segment so
1749	* it doesn't appear in the user's data queue.
1750	* It is still reflected in the segment length for
1751	* sequencing purposes.
1752	*/
1753
1754	#if 0
1755	void
1756	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1757	{
1758	int cnt = ti->ti_urp - 1;
1759
1760	while (cnt >= 0)
1761	{
1762	if (m->m_len > cnt)
1763	{
1764	char *cp = mtod(m, caddr_t) + cnt;
1765	struct tcpcb *tp = sototcpcb(so);
1766
1767	tp->t_iobc = *cp;
1768	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1769	memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1770	m->m_len--;
1771	return;
1772	}
1773	cnt -= m->m_len;
1774	m = m->m_next; /* XXX WRONG! Fix it! */
1775	if (m == 0)
1776	break;
1777	}
1778	panic("tcp_pulloutofband");
1779	}
1780	#endif
1781
1782	/*
1783	* Collect new round-trip time estimate
1784	* and update averages and current timeout.
1785	*/
1786
1787	void
1788	tcp_xmit_timer(PNATState pData, register struct tcpcb *tp, int rtt)
1789	{
1790	register short delta;
1791
1792	DEBUG_CALL("tcp_xmit_timer");
1793	DEBUG_ARG("tp = %lx", (long)tp);
1794	DEBUG_ARG("rtt = %d", rtt);
1795
1796	tcpstat.tcps_rttupdated++;
1797	if (tp->t_srtt != 0)
1798	{
1799	/*
1800	* srtt is stored as fixed point with 3 bits after the
1801	* binary point (i.e., scaled by 8). The following magic
1802	* is equivalent to the smoothing algorithm in rfc793 with
1803	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1804	* point). Adjust rtt to origin 0.
1805	*/
1806	delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1807	if ((tp->t_srtt += delta) <= 0)
1808	tp->t_srtt = 1;
1809	/*
1810	* We accumulate a smoothed rtt variance (actually, a
1811	* smoothed mean difference), then set the retransmit
1812	* timer to smoothed rtt + 4 times the smoothed variance.
1813	* rttvar is stored as fixed point with 2 bits after the
1814	* binary point (scaled by 4). The following is
1815	* equivalent to rfc793 smoothing with an alpha of .75
1816	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1817	* rfc793's wired-in beta.
1818	*/
1819	if (delta < 0)
1820	delta = -delta;
1821	delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1822	if ((tp->t_rttvar += delta) <= 0)
1823	tp->t_rttvar = 1;
1824	}
1825	else
1826	{
1827	/*
1828	* No rtt measurement yet - use the unsmoothed rtt.
1829	* Set the variance to half the rtt (so our first
1830	* retransmit happens at 3*rtt).
1831	*/
1832	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1833	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1834	}
1835	tp->t_rtt = 0;
1836	tp->t_rxtshift = 0;
1837
1838	/*
1839	* the retransmit should happen at rtt + 4 * rttvar.
1840	* Because of the way we do the smoothing, srtt and rttvar
1841	* will each average +1/2 tick of bias. When we compute
1842	* the retransmit timer, we want 1/2 tick of rounding and
1843	* 1 extra tick because of +-1/2 tick uncertainty in the
1844	* firing of the timer. The bias will give us exactly the
1845	* 1.5 tick we need. But, because the bias is
1846	* statistical, we have to test that we don't drop below
1847	* the minimum feasible timer (which is 2 ticks).
1848	*/
1849	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1850	(short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1851
1852	/*
1853	* We received an ack for a packet that wasn't retransmitted;
1854	* it is probably safe to discard any error indications we've
1855	* received recently. This isn't quite right, but close enough
1856	* for now (a route might have failed after we sent a segment,
1857	* and the return path might not be symmetrical).
1858	*/
1859	tp->t_softerror = 0;
1860	}
1861
1862	/*
1863	* Determine a reasonable value for maxseg size.
1864	* If the route is known, check route for mtu.
1865	* If none, use an mss that can be handled on the outgoing
1866	* interface without forcing IP to fragment; if bigger than
1867	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1868	* to utilize large mbufs. If no route is found, route has no mtu,
1869	* or the destination isn't local, use a default, hopefully conservative
1870	* size (usually 512 or the default IP max size, but no more than the mtu
1871	* of the interface), as we can't discover anything about intervening
1872	* gateways or networks. We also initialize the congestion/slow start
1873	* window to be a single segment if the destination isn't local.
1874	* While looking at the routing entry, we also initialize other path-dependent
1875	* parameters from pre-set or cached values in the routing entry.
1876	*/
1877
1878	int
1879	tcp_mss(PNATState pData, register struct tcpcb *tp, u_int offer)
1880	{
1881	struct socket *so = tp->t_socket;
1882	int mss;
1883
1884	DEBUG_CALL("tcp_mss");
1885	DEBUG_ARG("tp = %lx", (long)tp);
1886	DEBUG_ARG("offer = %d", offer);
1887
1888	mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
1889	if (offer)
1890	mss = min(mss, offer);
1891	mss = max(mss, 32);
1892	if (mss < tp->t_maxseg \|\| offer != 0)
1893	tp->t_maxseg = mss;
1894
1895	tp->snd_cwnd = mss;
1896
1897	sbreserve(pData, &so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
1898	sbreserve(pData, &so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
1899
1900	DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1901
1902	return mss;
1903	}

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 29603

Download in other formats: