tcp_input.c@ 28482

Last change on this file since 28482 was 28482, 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: 61.7 KB

Line
1	/* $Id: tcp_input.c 28482 2010-04-19 16:44:22Z 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	so->so_iptos = ((struct ip *)ti)->ip_tos;
533
534	tp = sototcpcb(so);
535	tp->t_state = TCPS_LISTEN;
536	}
537
538	/*
539	* If this is a still-connecting socket, this probably
540	* a retransmit of the SYN. Whether it's a retransmit SYN
541	* or something else, we nuke it.
542	*/
543	if (so->so_state & SS_ISFCONNECTING)
544	{
545	goto drop;
546	}
547
548	tp = sototcpcb(so);
549
550	/* XXX Should never fail */
551	if (tp == 0)
552	goto dropwithreset;
553	if (tp->t_state == TCPS_CLOSED)
554	{
555	goto drop;
556	}
557
558	/* Unscale the window into a 32-bit value. */
559	/* if ((tiflags & TH_SYN) == 0)
560	* tiwin = ti->ti_win << tp->snd_scale;
561	* else
562	*/
563	tiwin = ti->ti_win;
564
565	/*
566	* Segment received on connection.
567	* Reset idle time and keep-alive timer.
568	*/
569	tp->t_idle = 0;
570	if (so_options)
571	tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
572	else
573	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
574
575	/*
576	* Process options if not in LISTEN state,
577	* else do it below (after getting remote address).
578	*/
579	if (optp && tp->t_state != TCPS_LISTEN)
580	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
581	/* , */
582	/* &ts_present, &ts_val, &ts_ecr); */
583
584	/*
585	* Header prediction: check for the two common cases
586	* of a uni-directional data xfer. If the packet has
587	* no control flags, is in-sequence, the window didn't
588	* change and we're not retransmitting, it's a
589	* candidate. If the length is zero and the ack moved
590	* forward, we're the sender side of the xfer. Just
591	* free the data acked & wake any higher level process
592	* that was blocked waiting for space. If the length
593	* is non-zero and the ack didn't move, we're the
594	* receiver side. If we're getting packets in-order
595	* (the reassembly queue is empty), add the data to
596	* the socket buffer and note that we need a delayed ack.
597	*
598	* XXX Some of these tests are not needed
599	* eg: the tiwin == tp->snd_wnd prevents many more
600	* predictions.. with no real advantage..
601	*/
602	if ( tp->t_state == TCPS_ESTABLISHED
603	&& (tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK
604	/* && (!ts_present \|\| TSTMP_GEQ(ts_val, tp->ts_recent)) */
605	&& ti->ti_seq == tp->rcv_nxt
606	&& tiwin && tiwin == tp->snd_wnd
607	&& tp->snd_nxt == tp->snd_max)
608	{
609	/*
610	* If last ACK falls within this segment's sequence numbers,
611	* record the timestamp.
612	*/
613	#if 0
614	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
615	SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len))
616	{
617	tp->ts_recent_age = tcp_now;
618	tp->ts_recent = ts_val;
619	}
620	#endif
621
622	if (ti->ti_len == 0)
623	{
624	if ( SEQ_GT(ti->ti_ack, tp->snd_una)
625	&& SEQ_LEQ(ti->ti_ack, tp->snd_max)
626	&& tp->snd_cwnd >= tp->snd_wnd)
627	{
628	/*
629	* this is a pure ack for outstanding data.
630	*/
631	++tcpstat.tcps_predack;
632	#if 0
633	if (ts_present)
634	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
635	else
636	#endif
637	if ( tp->t_rtt
638	&& SEQ_GT(ti->ti_ack, tp->t_rtseq))
639	tcp_xmit_timer(pData, tp, tp->t_rtt);
640	acked = ti->ti_ack - tp->snd_una;
641	tcpstat.tcps_rcvackpack++;
642	tcpstat.tcps_rcvackbyte += acked;
643	sbdrop(&so->so_snd, acked);
644	tp->snd_una = ti->ti_ack;
645	m_freem(pData, m);
646
647	/*
648	* If all outstanding data are acked, stop
649	* retransmit timer, otherwise restart timer
650	* using current (possibly backed-off) value.
651	* If process is waiting for space,
652	* wakeup/selwakeup/signal. If data
653	* are ready to send, let tcp_output
654	* decide between more output or persist.
655	*/
656	if (tp->snd_una == tp->snd_max)
657	tp->t_timer[TCPT_REXMT] = 0;
658	else if (tp->t_timer[TCPT_PERSIST] == 0)
659	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
660
661	/*
662	* There's room in so_snd, sowwakup will read()
663	* from the socket if we can
664	*/
665	#if 0
666	if (so->so_snd.sb_flags & SB_NOTIFY)
667	sowwakeup(so);
668	#endif
669	/*
670	* This is called because sowwakeup might have
671	* put data into so_snd. Since we don't so sowwakeup,
672	* we don't need this.. XXX???
673	*/
674	if (so->so_snd.sb_cc)
675	(void) tcp_output(pData, tp);
676
677	SOCKET_UNLOCK(so);
678	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
679	return;
680	}
681	}
682	else if ( ti->ti_ack == tp->snd_una
683	&& LIST_FIRST(&tp->t_segq)
684	&& ti->ti_len <= sbspace(&so->so_rcv))
685	{
686	/*
687	* this is a pure, in-sequence data packet
688	* with nothing on the reassembly queue and
689	* we have enough buffer space to take it.
690	*/
691	++tcpstat.tcps_preddat;
692	tp->rcv_nxt += ti->ti_len;
693	tcpstat.tcps_rcvpack++;
694	tcpstat.tcps_rcvbyte += ti->ti_len;
695	/*
696	* Add data to socket buffer.
697	*/
698	sbappend(pData, so, m);
699
700	/*
701	* XXX This is called when data arrives. Later, check
702	* if we can actually write() to the socket
703	* XXX Need to check? It's be NON_BLOCKING
704	*/
705	/* sorwakeup(so); */
706
707	/*
708	* If this is a short packet, then ACK now - with Nagel
709	* congestion avoidance sender won't send more until
710	* he gets an ACK.
711	*
712	* It is better to not delay acks at all to maximize
713	* TCP throughput. See RFC 2581.
714	*/
715	tp->t_flags \|= TF_ACKNOW;
716	tcp_output(pData, tp);
717	SOCKET_UNLOCK(so);
718	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
719	return;
720	}
721	} /* header prediction */
722	/*
723	* Calculate amount of space in receive window,
724	* and then do TCP input processing.
725	* Receive window is amount of space in rcv queue,
726	* but not less than advertised window.
727	*/
728	{
729	int win;
730	win = sbspace(&so->so_rcv);
731	if (win < 0)
732	win = 0;
733	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
734	}
735
736	switch (tp->t_state)
737	{
738	/*
739	* If the state is LISTEN then ignore segment if it contains an RST.
740	* If the segment contains an ACK then it is bad and send a RST.
741	* If it does not contain a SYN then it is not interesting; drop it.
742	* Don't bother responding if the destination was a broadcast.
743	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
744	* tp->iss, and send a segment:
745	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
746	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
747	* Fill in remote peer address fields if not previously specified.
748	* Enter SYN_RECEIVED state, and process any other fields of this
749	* segment in this state.
750	*/
751	case TCPS_LISTEN:
752	{
753	if (tiflags & TH_RST) {
754	goto drop;
755	}
756	if (tiflags & TH_ACK)
757	goto dropwithreset;
758	if ((tiflags & TH_SYN) == 0)
759	{
760	goto drop;
761	}
762
763	/*
764	* This has way too many gotos...
765	* But a bit of spaghetti code never hurt anybody :)
766	*/
767	if ( (tcp_fconnect(pData, so) == -1)
768	&& errno != EINPROGRESS
769	&& errno != EWOULDBLOCK)
770	{
771	u_char code = ICMP_UNREACH_NET;
772	DEBUG_MISC((dfd," tcp fconnect errno = %d-%s\n",
773	errno, strerror(errno)));
774	if (errno == ECONNREFUSED)
775	{
776	/* ACK the SYN, send RST to refuse the connection */
777	tcp_respond(pData, tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
778	TH_RST\|TH_ACK);
779	}
780	else
781	{
782	if (errno == EHOSTUNREACH)
783	code = ICMP_UNREACH_HOST;
784	HTONL(ti->ti_seq); /* restore tcp header */
785	HTONL(ti->ti_ack);
786	HTONS(ti->ti_win);
787	HTONS(ti->ti_urp);
788	m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
789	m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
790	*ip = save_ip;
791	icmp_error(pData, m, ICMP_UNREACH, code, 0, strerror(errno));
792	tp->t_socket->so_m = NULL;
793	}
794	tp = tcp_close(pData, tp);
795	m_freem(pData, m);
796	}
797	else
798	{
799	/*
800	* Haven't connected yet, save the current mbuf
801	* and ti, and return
802	* XXX Some OS's don't tell us whether the connect()
803	* succeeded or not. So we must time it out.
804	*/
805	so->so_m = m;
806	so->so_ti = ti;
807	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
808	tp->t_state = TCPS_SYN_RECEIVED;
809	}
810	SOCKET_UNLOCK(so);
811	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
812	return;
813
814	cont_conn:
815	/* m==NULL
816	* Check if the connect succeeded
817	*/
818	if (so->so_state & SS_NOFDREF)
819	{
820	tp = tcp_close(pData, tp);
821	goto dropwithreset;
822	}
823	cont_input:
824	tcp_template(tp);
825
826	if (optp)
827	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
828
829	if (iss)
830	tp->iss = iss;
831	else
832	tp->iss = tcp_iss;
833	tcp_iss += TCP_ISSINCR/2;
834	tp->irs = ti->ti_seq;
835	tcp_sendseqinit(tp);
836	tcp_rcvseqinit(tp);
837	tp->t_flags \|= TF_ACKNOW;
838	tp->t_state = TCPS_SYN_RECEIVED;
839	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
840	tcpstat.tcps_accepts++;
841	goto trimthenstep6;
842	} /* case TCPS_LISTEN */
843
844	/*
845	* If the state is SYN_SENT:
846	* if seg contains an ACK, but not for our SYN, drop the input.
847	* if seg contains a RST, then drop the connection.
848	* if seg does not contain SYN, then drop it.
849	* Otherwise this is an acceptable SYN segment
850	* initialize tp->rcv_nxt and tp->irs
851	* if seg contains ack then advance tp->snd_una
852	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
853	* arrange for segment to be acked (eventually)
854	* continue processing rest of data/controls, beginning with URG
855	*/
856	case TCPS_SYN_SENT:
857	if ( (tiflags & TH_ACK)
858	&& ( SEQ_LEQ(ti->ti_ack, tp->iss)
859	\|\| SEQ_GT(ti->ti_ack, tp->snd_max)))
860	goto dropwithreset;
861
862	if (tiflags & TH_RST)
863	{
864	if (tiflags & TH_ACK)
865	tp = tcp_drop(pData, tp, 0); /* XXX Check t_softerror! */
866	goto drop;
867	}
868
869	if ((tiflags & TH_SYN) == 0)
870	{
871	goto drop;
872	}
873	if (tiflags & TH_ACK)
874	{
875	tp->snd_una = ti->ti_ack;
876	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
877	tp->snd_nxt = tp->snd_una;
878	}
879
880	tp->t_timer[TCPT_REXMT] = 0;
881	tp->irs = ti->ti_seq;
882	tcp_rcvseqinit(tp);
883	tp->t_flags \|= TF_ACKNOW;
884	if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss))
885	{
886	tcpstat.tcps_connects++;
887	soisfconnected(so);
888	tp->t_state = TCPS_ESTABLISHED;
889
890	/* Do window scaling on this connection? */
891	#if 0
892	if (( tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
893	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
894	{
895	tp->snd_scale = tp->requested_s_scale;
896	tp->rcv_scale = tp->request_r_scale;
897	}
898	#endif
899	(void) tcp_reass(pData, tp, (struct tcphdr )0, NULL, (struct mbuf )0);
900	/*
901	* if we didn't have to retransmit the SYN,
902	* use its rtt as our initial srtt & rtt var.
903	*/
904	if (tp->t_rtt)
905	tcp_xmit_timer(pData, tp, tp->t_rtt);
906	}
907	else
908	tp->t_state = TCPS_SYN_RECEIVED;
909
910	trimthenstep6:
911	/*
912	* Advance ti->ti_seq to correspond to first data byte.
913	* If data, trim to stay within window,
914	* dropping FIN if necessary.
915	*/
916	ti->ti_seq++;
917	if (ti->ti_len > tp->rcv_wnd)
918	{
919	todrop = ti->ti_len - tp->rcv_wnd;
920	m_adj(m, -todrop);
921	ti->ti_len = tp->rcv_wnd;
922	tiflags &= ~TH_FIN;
923	tcpstat.tcps_rcvpackafterwin++;
924	tcpstat.tcps_rcvbyteafterwin += todrop;
925	}
926	tp->snd_wl1 = ti->ti_seq - 1;
927	tp->rcv_up = ti->ti_seq;
928	Log2(("hit6"));
929	goto step6;
930	} /* switch tp->t_state */
931	/*
932	* States other than LISTEN or SYN_SENT.
933	* First check timestamp, if present.
934	* Then check that at least some bytes of segment are within
935	* receive window. If segment begins before rcv_nxt,
936	* drop leading data (and SYN); if nothing left, just ack.
937	*
938	* RFC 1323 PAWS: If we have a timestamp reply on this segment
939	* and it's less than ts_recent, drop it.
940	*/
941	#if 0
942	if ( ts_present
943	&& (tiflags & TH_RST) == 0
944	&& tp->ts_recent
945	&& TSTMP_LT(ts_val, tp->ts_recent))
946	{
947	/* Check to see if ts_recent is over 24 days old. */
948	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE)
949	{
950	/*
951	* Invalidate ts_recent. If this segment updates
952	* ts_recent, the age will be reset later and ts_recent
953	* will get a valid value. If it does not, setting
954	* ts_recent to zero will at least satisfy the
955	* requirement that zero be placed in the timestamp
956	* echo reply when ts_recent isn't valid. The
957	* age isn't reset until we get a valid ts_recent
958	* because we don't want out-of-order segments to be
959	* dropped when ts_recent is old.
960	*/
961	tp->ts_recent = 0;
962	}
963	else
964	{
965	tcpstat.tcps_rcvduppack++;
966	tcpstat.tcps_rcvdupbyte += ti->ti_len;
967	tcpstat.tcps_pawsdrop++;
968	goto dropafterack;
969	}
970	}
971	#endif
972
973	todrop = tp->rcv_nxt - ti->ti_seq;
974	if (todrop > 0)
975	{
976	if (tiflags & TH_SYN)
977	{
978	tiflags &= ~TH_SYN;
979	ti->ti_seq++;
980	if (ti->ti_urp > 1)
981	ti->ti_urp--;
982	else
983	tiflags &= ~TH_URG;
984	todrop--;
985	}
986	/*
987	* Following if statement from Stevens, vol. 2, p. 960.
988	*/
989	if ( todrop > ti->ti_len
990	\|\| ( todrop == ti->ti_len
991	&& (tiflags & TH_FIN) == 0))
992	{
993	/*
994	* Any valid FIN must be to the left of the window.
995	* At this point the FIN must be a duplicate or out
996	* of sequence; drop it.
997	*/
998	tiflags &= ~TH_FIN;
999
1000	/*
1001	* Send an ACK to resynchronize and drop any data.
1002	* But keep on processing for RST or ACK.
1003	*/
1004	tp->t_flags \|= TF_ACKNOW;
1005	todrop = ti->ti_len;
1006	tcpstat.tcps_rcvduppack++;
1007	tcpstat.tcps_rcvdupbyte += todrop;
1008	}
1009	else
1010	{
1011	tcpstat.tcps_rcvpartduppack++;
1012	tcpstat.tcps_rcvpartdupbyte += todrop;
1013	}
1014	m_adj(m, todrop);
1015	ti->ti_seq += todrop;
1016	ti->ti_len -= todrop;
1017	if (ti->ti_urp > todrop)
1018	ti->ti_urp -= todrop;
1019	else
1020	{
1021	tiflags &= ~TH_URG;
1022	ti->ti_urp = 0;
1023	}
1024	}
1025	/*
1026	* If new data are received on a connection after the
1027	* user processes are gone, then RST the other end.
1028	*/
1029	if ( (so->so_state & SS_NOFDREF)
1030	&& tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len)
1031	{
1032	tp = tcp_close(pData, tp);
1033	tcpstat.tcps_rcvafterclose++;
1034	goto dropwithreset;
1035	}
1036
1037	/*
1038	* If segment ends after window, drop trailing data
1039	* (and PUSH and FIN); if nothing left, just ACK.
1040	*/
1041	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1042	if (todrop > 0)
1043	{
1044	tcpstat.tcps_rcvpackafterwin++;
1045	if (todrop >= ti->ti_len)
1046	{
1047	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1048	/*
1049	* If a new connection request is received
1050	* while in TIME_WAIT, drop the old connection
1051	* and start over if the sequence numbers
1052	* are above the previous ones.
1053	*/
1054	if ( tiflags & TH_SYN
1055	&& tp->t_state == TCPS_TIME_WAIT
1056	&& SEQ_GT(ti->ti_seq, tp->rcv_nxt))
1057	{
1058	iss = tp->rcv_nxt + TCP_ISSINCR;
1059	tp = tcp_close(pData, tp);
1060	SOCKET_UNLOCK(tp->t_socket);
1061	goto findso;
1062	}
1063	/*
1064	* If window is closed can only take segments at
1065	* window edge, and have to drop data and PUSH from
1066	* incoming segments. Continue processing, but
1067	* remember to ack. Otherwise, drop segment
1068	* and ack.
1069	*/
1070	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt)
1071	{
1072	tp->t_flags \|= TF_ACKNOW;
1073	tcpstat.tcps_rcvwinprobe++;
1074	}
1075	else
1076	goto dropafterack;
1077	}
1078	else
1079	tcpstat.tcps_rcvbyteafterwin += todrop;
1080	m_adj(m, -todrop);
1081	ti->ti_len -= todrop;
1082	tiflags &= ~(TH_PUSH\|TH_FIN);
1083	}
1084
1085	/*
1086	* If last ACK falls within this segment's sequence numbers,
1087	* record its timestamp.
1088	*/
1089	#if 0
1090	if ( ts_present
1091	&& SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)
1092	&& SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + ((tiflags & (TH_SYN\|TH_FIN)) != 0)))
1093	{
1094	tp->ts_recent_age = tcp_now;
1095	tp->ts_recent = ts_val;
1096	}
1097	#endif
1098
1099	/*
1100	* If the RST bit is set examine the state:
1101	* SYN_RECEIVED STATE:
1102	* If passive open, return to LISTEN state.
1103	* If active open, inform user that connection was refused.
1104	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1105	* Inform user that connection was reset, and close tcb.
1106	* CLOSING, LAST_ACK, TIME_WAIT STATES
1107	* Close the tcb.
1108	*/
1109	if (tiflags&TH_RST)
1110	switch (tp->t_state)
1111	{
1112	case TCPS_SYN_RECEIVED:
1113	/* so->so_error = ECONNREFUSED; */
1114	goto close;
1115
1116	case TCPS_ESTABLISHED:
1117	case TCPS_FIN_WAIT_1:
1118	case TCPS_FIN_WAIT_2:
1119	case TCPS_CLOSE_WAIT:
1120	/* so->so_error = ECONNRESET; */
1121	close:
1122	tp->t_state = TCPS_CLOSED;
1123	tcpstat.tcps_drops++;
1124	tp = tcp_close(pData, tp);
1125	goto drop;
1126
1127	case TCPS_CLOSING:
1128	case TCPS_LAST_ACK:
1129	case TCPS_TIME_WAIT:
1130	tp = tcp_close(pData, tp);
1131	goto drop;
1132	}
1133
1134	/*
1135	* If a SYN is in the window, then this is an
1136	* error and we send an RST and drop the connection.
1137	*/
1138	if (tiflags & TH_SYN)
1139	{
1140	tp = tcp_drop(pData, tp, 0);
1141	goto dropwithreset;
1142	}
1143
1144	/*
1145	* If the ACK bit is off we drop the segment and return.
1146	*/
1147	if ((tiflags & TH_ACK) == 0)
1148	{
1149	goto drop;
1150	}
1151
1152	/*
1153	* Ack processing.
1154	*/
1155	switch (tp->t_state)
1156	{
1157	/*
1158	* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1159	* ESTABLISHED state and continue processing, otherwise
1160	* send an RST. una<=ack<=max
1161	*/
1162	case TCPS_SYN_RECEIVED:
1163	if ( SEQ_GT(tp->snd_una, ti->ti_ack)
1164	\|\| SEQ_GT(ti->ti_ack, tp->snd_max))
1165	goto dropwithreset;
1166	tcpstat.tcps_connects++;
1167	tp->t_state = TCPS_ESTABLISHED;
1168	/*
1169	* The sent SYN is ack'ed with our sequence number +1
1170	* The first data byte already in the buffer will get
1171	* lost if no correction is made. This is only needed for
1172	* SS_CTL since the buffer is empty otherwise.
1173	* tp->snd_una++; or:
1174	*/
1175	tp->snd_una = ti->ti_ack;
1176	soisfconnected(so);
1177
1178	/* Do window scaling? */
1179	#if 0
1180	if ( (tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
1181	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
1182	{
1183	tp->snd_scale = tp->requested_s_scale;
1184	tp->rcv_scale = tp->request_r_scale;
1185	}
1186	#endif
1187	(void) tcp_reass(pData, tp, (struct tcphdr )0, (int )0, (struct mbuf *)0);
1188	tp->snd_wl1 = ti->ti_seq - 1;
1189	/* Avoid ack processing; snd_una==ti_ack => dup ack */
1190	goto synrx_to_est;
1191	/* fall into ... */
1192
1193	/*
1194	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1195	* ACKs. If the ack is in the range
1196	* tp->snd_una < ti->ti_ack <= tp->snd_max
1197	* then advance tp->snd_una to ti->ti_ack and drop
1198	* data from the retransmission queue. If this ACK reflects
1199	* more up to date window information we update our window information.
1200	*/
1201	case TCPS_ESTABLISHED:
1202	case TCPS_FIN_WAIT_1:
1203	case TCPS_FIN_WAIT_2:
1204	case TCPS_CLOSE_WAIT:
1205	case TCPS_CLOSING:
1206	case TCPS_LAST_ACK:
1207	case TCPS_TIME_WAIT:
1208	if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
1209	{
1210	if (ti->ti_len == 0 && tiwin == tp->snd_wnd)
1211	{
1212	tcpstat.tcps_rcvdupack++;
1213	DEBUG_MISC((dfd," dup ack m = %lx so = %lx \n",
1214	(long )m, (long )so));
1215	/*
1216	* If we have outstanding data (other than
1217	* a window probe), this is a completely
1218	* duplicate ack (ie, window info didn't
1219	* change), the ack is the biggest we've
1220	* seen and we've seen exactly our rexmt
1221	* threshold of them, assume a packet
1222	* has been dropped and retransmit it.
1223	* Kludge snd_nxt & the congestion
1224	* window so we send only this one
1225	* packet.
1226	*
1227	* We know we're losing at the current
1228	* window size so do congestion avoidance
1229	* (set ssthresh to half the current window
1230	* and pull our congestion window back to
1231	* the new ssthresh).
1232	*
1233	* Dup acks mean that packets have left the
1234	* network (they're now cached at the receiver)
1235	* so bump cwnd by the amount in the receiver
1236	* to keep a constant cwnd packets in the
1237	* network.
1238	*/
1239	if ( tp->t_timer[TCPT_REXMT] == 0
1240	\|\| ti->ti_ack != tp->snd_una)
1241	tp->t_dupacks = 0;
1242	else if (++tp->t_dupacks == tcprexmtthresh)
1243	{
1244	tcp_seq onxt = tp->snd_nxt;
1245	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
1246	if (win < 2)
1247	win = 2;
1248	tp->snd_ssthresh = win * tp->t_maxseg;
1249	tp->t_timer[TCPT_REXMT] = 0;
1250	tp->t_rtt = 0;
1251	tp->snd_nxt = ti->ti_ack;
1252	tp->snd_cwnd = tp->t_maxseg;
1253	(void) tcp_output(pData, tp);
1254	tp->snd_cwnd = tp->snd_ssthresh +
1255	tp->t_maxseg * tp->t_dupacks;
1256	if (SEQ_GT(onxt, tp->snd_nxt))
1257	tp->snd_nxt = onxt;
1258	goto drop;
1259	}
1260	else if (tp->t_dupacks > tcprexmtthresh)
1261	{
1262	tp->snd_cwnd += tp->t_maxseg;
1263	(void) tcp_output(pData, tp);
1264	goto drop;
1265	}
1266	}
1267	else
1268	tp->t_dupacks = 0;
1269	break;
1270	}
1271	synrx_to_est:
1272	/*
1273	* If the congestion window was inflated to account
1274	* for the other side's cached packets, retract it.
1275	*/
1276	if ( tp->t_dupacks > tcprexmtthresh
1277	&& tp->snd_cwnd > tp->snd_ssthresh)
1278	tp->snd_cwnd = tp->snd_ssthresh;
1279	tp->t_dupacks = 0;
1280	if (SEQ_GT(ti->ti_ack, tp->snd_max))
1281	{
1282	tcpstat.tcps_rcvacktoomuch++;
1283	goto dropafterack;
1284	}
1285	acked = ti->ti_ack - tp->snd_una;
1286	tcpstat.tcps_rcvackpack++;
1287	tcpstat.tcps_rcvackbyte += acked;
1288
1289	/*
1290	* If we have a timestamp reply, update smoothed
1291	* round trip time. If no timestamp is present but
1292	* transmit timer is running and timed sequence
1293	* number was acked, update smoothed round trip time.
1294	* Since we now have an rtt measurement, cancel the
1295	* timer backoff (cf., Phil Karn's retransmit alg.).
1296	* Recompute the initial retransmit timer.
1297	*/
1298	#if 0
1299	if (ts_present)
1300	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1301	else
1302	#endif
1303	if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1304	tcp_xmit_timer(pData, tp, tp->t_rtt);
1305
1306	/*
1307	* If all outstanding data is acked, stop retransmit
1308	* timer and remember to restart (more output or persist).
1309	* If there is more data to be acked, restart retransmit
1310	* timer, using current (possibly backed-off) value.
1311	*/
1312	if (ti->ti_ack == tp->snd_max)
1313	{
1314	tp->t_timer[TCPT_REXMT] = 0;
1315	needoutput = 1;
1316	}
1317	else if (tp->t_timer[TCPT_PERSIST] == 0)
1318	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1319	/*
1320	* When new data is acked, open the congestion window.
1321	* If the window gives us less than ssthresh packets
1322	* in flight, open exponentially (maxseg per packet).
1323	* Otherwise open linearly: maxseg per window
1324	* (maxseg^2 / cwnd per packet).
1325	*/
1326	{
1327	register u_int cw = tp->snd_cwnd;
1328	register u_int incr = tp->t_maxseg;
1329
1330	if (cw > tp->snd_ssthresh)
1331	incr = incr * incr / cw;
1332	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1333	}
1334	if (acked > so->so_snd.sb_cc)
1335	{
1336	tp->snd_wnd -= so->so_snd.sb_cc;
1337	sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1338	ourfinisacked = 1;
1339	}
1340	else
1341	{
1342	sbdrop(&so->so_snd, acked);
1343	tp->snd_wnd -= acked;
1344	ourfinisacked = 0;
1345	}
1346	/*
1347	* XXX sowwakup is called when data is acked and there's room for
1348	* for more data... it should read() the socket
1349	*/
1350	#if 0
1351	if (so->so_snd.sb_flags & SB_NOTIFY)
1352	sowwakeup(so);
1353	#endif
1354	tp->snd_una = ti->ti_ack;
1355	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1356	tp->snd_nxt = tp->snd_una;
1357
1358	switch (tp->t_state)
1359	{
1360	/*
1361	* In FIN_WAIT_1 STATE in addition to the processing
1362	* for the ESTABLISHED state if our FIN is now acknowledged
1363	* then enter FIN_WAIT_2.
1364	*/
1365	case TCPS_FIN_WAIT_1:
1366	if (ourfinisacked)
1367	{
1368	/*
1369	* If we can't receive any more
1370	* data, then closing user can proceed.
1371	* Starting the timer is contrary to the
1372	* specification, but if we don't get a FIN
1373	* we'll hang forever.
1374	*/
1375	if (so->so_state & SS_FCANTRCVMORE)
1376	{
1377	soisfdisconnected(so);
1378	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1379	}
1380	tp->t_state = TCPS_FIN_WAIT_2;
1381	}
1382	break;
1383
1384	/*
1385	* In CLOSING STATE in addition to the processing for
1386	* the ESTABLISHED state if the ACK acknowledges our FIN
1387	* then enter the TIME-WAIT state, otherwise ignore
1388	* the segment.
1389	*/
1390	case TCPS_CLOSING:
1391	if (ourfinisacked)
1392	{
1393	tp->t_state = TCPS_TIME_WAIT;
1394	tcp_canceltimers(tp);
1395	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1396	soisfdisconnected(so);
1397	}
1398	break;
1399
1400	/*
1401	* In LAST_ACK, we may still be waiting for data to drain
1402	* and/or to be acked, as well as for the ack of our FIN.
1403	* If our FIN is now acknowledged, delete the TCB,
1404	* enter the closed state and return.
1405	*/
1406	case TCPS_LAST_ACK:
1407	if (ourfinisacked)
1408	{
1409	tp = tcp_close(pData, tp);
1410	goto drop;
1411	}
1412	break;
1413
1414	/*
1415	* In TIME_WAIT state the only thing that should arrive
1416	* is a retransmission of the remote FIN. Acknowledge
1417	* it and restart the finack timer.
1418	*/
1419	case TCPS_TIME_WAIT:
1420	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1421	goto dropafterack;
1422	}
1423	} /* switch(tp->t_state) */
1424
1425	step6:
1426	/*
1427	* Update window information.
1428	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1429	*/
1430	if ( (tiflags & TH_ACK)
1431	&& ( SEQ_LT(tp->snd_wl1, ti->ti_seq)
1432	\|\| ( tp->snd_wl1 == ti->ti_seq
1433	&& ( SEQ_LT(tp->snd_wl2, ti->ti_ack)
1434	\|\| ( tp->snd_wl2 == ti->ti_ack
1435	&& tiwin > tp->snd_wnd)))))
1436	{
1437	/* keep track of pure window updates */
1438	if ( ti->ti_len == 0
1439	&& tp->snd_wl2 == ti->ti_ack
1440	&& tiwin > tp->snd_wnd)
1441	tcpstat.tcps_rcvwinupd++;
1442	tp->snd_wnd = tiwin;
1443	tp->snd_wl1 = ti->ti_seq;
1444	tp->snd_wl2 = ti->ti_ack;
1445	if (tp->snd_wnd > tp->max_sndwnd)
1446	tp->max_sndwnd = tp->snd_wnd;
1447	needoutput = 1;
1448	}
1449
1450	/*
1451	* Process segments with URG.
1452	*/
1453	if ((tiflags & TH_URG) && ti->ti_urp &&
1454	TCPS_HAVERCVDFIN(tp->t_state) == 0)
1455	{
1456	/*
1457	* This is a kludge, but if we receive and accept
1458	* random urgent pointers, we'll crash in
1459	* soreceive. It's hard to imagine someone
1460	* actually wanting to send this much urgent data.
1461	*/
1462	if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen)
1463	{
1464	ti->ti_urp = 0;
1465	tiflags &= ~TH_URG;
1466	goto dodata;
1467	}
1468	/*
1469	* If this segment advances the known urgent pointer,
1470	* then mark the data stream. This should not happen
1471	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1472	* a FIN has been received from the remote side.
1473	* In these states we ignore the URG.
1474	*
1475	* According to RFC961 (Assigned Protocols),
1476	* the urgent pointer points to the last octet
1477	* of urgent data. We continue, however,
1478	* to consider it to indicate the first octet
1479	* of data past the urgent section as the original
1480	* spec states (in one of two places).
1481	*/
1482	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up))
1483	{
1484	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1485	so->so_urgc = so->so_rcv.sb_cc +
1486	(tp->rcv_up - tp->rcv_nxt); /* -1; */
1487	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1488	}
1489	}
1490	else
1491	/*
1492	* If no out of band data is expected,
1493	* pull receive urgent pointer along
1494	* with the receive window.
1495	*/
1496	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1497	tp->rcv_up = tp->rcv_nxt;
1498	dodata:
1499
1500	/*
1501	* If this is a small packet, then ACK now - with Nagel
1502	* congestion avoidance sender won't send more until
1503	* he gets an ACK.
1504	*
1505	* See above.
1506	*/
1507	if ( ti->ti_len
1508	&& (unsigned)ti->ti_len <= 5
1509	&& ((struct tcpiphdr_2 *)ti)->first_char == (char)27)
1510	{
1511	tp->t_flags \|= TF_ACKNOW;
1512	}
1513
1514	/*
1515	* Process the segment text, merging it into the TCP sequencing queue,
1516	* and arranging for acknowledgment of receipt if necessary.
1517	* This process logically involves adjusting tp->rcv_wnd as data
1518	* is presented to the user (this happens in tcp_usrreq.c,
1519	* case PRU_RCVD). If a FIN has already been received on this
1520	* connection then we just ignore the text.
1521	*/
1522	if ( (ti->ti_len \|\| (tiflags&TH_FIN))
1523	&& TCPS_HAVERCVDFIN(tp->t_state) == 0)
1524	{
1525	if ( ti->ti_seq == tp->rcv_nxt
1526	&& LIST_EMPTY(&tp->t_segq)
1527	&& tp->t_state == TCPS_ESTABLISHED)
1528	{
1529	DELAY_ACK(tp, ti); /* little bit different from BSD declaration see netinet/tcp_input.c */
1530	tp->rcv_nxt += tlen;
1531	tiflags = ti->ti_t.th_flags & TH_FIN;
1532	tcpstat.tcps_rcvpack++;
1533	tcpstat.tcps_rcvbyte += tlen;
1534	if (so->so_state & SS_FCANTRCVMORE)
1535	m_freem(pData, m);
1536	else
1537	sbappend(pData, so, m);
1538	}
1539	else
1540	{
1541	tiflags = tcp_reass(pData, tp, &ti->ti_t, &tlen, m);
1542	tiflags \|= TF_ACKNOW;
1543	}
1544	/*
1545	* Note the amount of data that peer has sent into
1546	* our window, in order to estimate the sender's
1547	* buffer size.
1548	*/
1549	len = so->so_rcv.sb_datalen - (tp->rcv_adv - tp->rcv_nxt);
1550	}
1551	else
1552	{
1553	m_freem(pData, m);
1554	tiflags &= ~TH_FIN;
1555	}
1556
1557	/*
1558	* If FIN is received ACK the FIN and let the user know
1559	* that the connection is closing.
1560	*/
1561	if (tiflags & TH_FIN)
1562	{
1563	if (TCPS_HAVERCVDFIN(tp->t_state) == 0)
1564	{
1565	/*
1566	* If we receive a FIN we can't send more data,
1567	* set it SS_FDRAIN
1568	* Shutdown the socket if there is no rx data in the
1569	* buffer.
1570	* soread() is called on completion of shutdown() and
1571	* will got to TCPS_LAST_ACK, and use tcp_output()
1572	* to send the FIN.
1573	*/
1574	/* sofcantrcvmore(so); */
1575	sofwdrain(so);
1576
1577	tp->t_flags \|= TF_ACKNOW;
1578	tp->rcv_nxt++;
1579	}
1580	switch (tp->t_state)
1581	{
1582	/*
1583	* In SYN_RECEIVED and ESTABLISHED STATES
1584	* enter the CLOSE_WAIT state.
1585	*/
1586	case TCPS_SYN_RECEIVED:
1587	case TCPS_ESTABLISHED:
1588	tp->t_state = TCPS_CLOSE_WAIT;
1589	break;
1590
1591	/*
1592	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1593	* enter the CLOSING state.
1594	*/
1595	case TCPS_FIN_WAIT_1:
1596	tp->t_state = TCPS_CLOSING;
1597	break;
1598
1599	/*
1600	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1601	* starting the time-wait timer, turning off the other
1602	* standard timers.
1603	*/
1604	case TCPS_FIN_WAIT_2:
1605	tp->t_state = TCPS_TIME_WAIT;
1606	tcp_canceltimers(tp);
1607	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1608	soisfdisconnected(so);
1609	break;
1610
1611	/*
1612	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1613	*/
1614	case TCPS_TIME_WAIT:
1615	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1616	break;
1617	}
1618	}
1619
1620	/*
1621	* Return any desired output.
1622	*/
1623	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1624	tcp_output(pData, tp);
1625
1626	SOCKET_UNLOCK(so);
1627	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1628	return;
1629
1630	dropafterack:
1631	Log2(("drop after ack\n"));
1632	/*
1633	* Generate an ACK dropping incoming segment if it occupies
1634	* sequence space, where the ACK reflects our state.
1635	*/
1636	if (tiflags & TH_RST)
1637	goto drop;
1638	m_freem(pData, m);
1639	tp->t_flags \|= TF_ACKNOW;
1640	(void) tcp_output(pData, tp);
1641	SOCKET_UNLOCK(so);
1642	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1643	return;
1644
1645	dropwithreset:
1646	/* reuses m if m!=NULL, m_free() unnecessary */
1647	if (tiflags & TH_ACK)
1648	tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1649	else
1650	{
1651	if (tiflags & TH_SYN) ti->ti_len++;
1652	tcp_respond(pData, tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1653	TH_RST\|TH_ACK);
1654	}
1655
1656	if (so != &tcb)
1657	SOCKET_UNLOCK(so);
1658	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1659	return;
1660
1661	drop:
1662	/*
1663	* Drop space held by incoming segment and return.
1664	*/
1665	m_freem(pData, m);
1666
1667	#ifdef VBOX_WITH_SLIRP_MT
1668	if (RTCritSectIsOwned(&so->so_mutex))
1669	{
1670	SOCKET_UNLOCK(so);
1671	}
1672	#endif
1673
1674	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1675	return;
1676	}
1677
1678	void
1679	tcp_dooptions(PNATState pData, struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti)
1680	{
1681	u_int16_t mss;
1682	int opt, optlen;
1683
1684	DEBUG_CALL("tcp_dooptions");
1685	DEBUG_ARGS((dfd," tp = %lx cnt=%i \n", (long )tp, cnt));
1686
1687	for (; cnt > 0; cnt -= optlen, cp += optlen)
1688	{
1689	opt = cp[0];
1690	if (opt == TCPOPT_EOL)
1691	break;
1692	if (opt == TCPOPT_NOP)
1693	optlen = 1;
1694	else
1695	{
1696	optlen = cp[1];
1697	if (optlen <= 0)
1698	break;
1699	}
1700	switch (opt)
1701	{
1702	default:
1703	continue;
1704
1705	case TCPOPT_MAXSEG:
1706	if (optlen != TCPOLEN_MAXSEG)
1707	continue;
1708	if (!(ti->ti_flags & TH_SYN))
1709	continue;
1710	memcpy((char ) &mss, (char ) cp + 2, sizeof(mss));
1711	NTOHS(mss);
1712	(void) tcp_mss(pData, tp, mss); /* sets t_maxseg */
1713	break;
1714
1715	#if 0
1716	case TCPOPT_WINDOW:
1717	if (optlen != TCPOLEN_WINDOW)
1718	continue;
1719	if (!(ti->ti_flags & TH_SYN))
1720	continue;
1721	tp->t_flags \|= TF_RCVD_SCALE;
1722	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1723	break;
1724
1725	case TCPOPT_TIMESTAMP:
1726	if (optlen != TCPOLEN_TIMESTAMP)
1727	continue;
1728	*ts_present = 1;
1729	memcpy((char ) ts_val, (char )cp + 2, sizeof(*ts_val));
1730	NTOHL(*ts_val);
1731	memcpy((char ) ts_ecr, (char )cp + 6, sizeof(*ts_ecr));
1732	NTOHL(*ts_ecr);
1733
1734	/*
1735	* A timestamp received in a SYN makes
1736	* it ok to send timestamp requests and replies.
1737	*/
1738	if (ti->ti_flags & TH_SYN)
1739	{
1740	tp->t_flags \|= TF_RCVD_TSTMP;
1741	tp->ts_recent = *ts_val;
1742	tp->ts_recent_age = tcp_now;
1743	}
1744	break;
1745	#endif
1746	}
1747	}
1748	}
1749
1750
1751	/*
1752	* Pull out of band byte out of a segment so
1753	* it doesn't appear in the user's data queue.
1754	* It is still reflected in the segment length for
1755	* sequencing purposes.
1756	*/
1757
1758	#if 0
1759	void
1760	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1761	{
1762	int cnt = ti->ti_urp - 1;
1763
1764	while (cnt >= 0)
1765	{
1766	if (m->m_len > cnt)
1767	{
1768	char *cp = mtod(m, caddr_t) + cnt;
1769	struct tcpcb *tp = sototcpcb(so);
1770
1771	tp->t_iobc = *cp;
1772	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1773	memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1774	m->m_len--;
1775	return;
1776	}
1777	cnt -= m->m_len;
1778	m = m->m_next; /* XXX WRONG! Fix it! */
1779	if (m == 0)
1780	break;
1781	}
1782	panic("tcp_pulloutofband");
1783	}
1784	#endif
1785
1786	/*
1787	* Collect new round-trip time estimate
1788	* and update averages and current timeout.
1789	*/
1790
1791	void
1792	tcp_xmit_timer(PNATState pData, register struct tcpcb *tp, int rtt)
1793	{
1794	register short delta;
1795
1796	DEBUG_CALL("tcp_xmit_timer");
1797	DEBUG_ARG("tp = %lx", (long)tp);
1798	DEBUG_ARG("rtt = %d", rtt);
1799
1800	tcpstat.tcps_rttupdated++;
1801	if (tp->t_srtt != 0)
1802	{
1803	/*
1804	* srtt is stored as fixed point with 3 bits after the
1805	* binary point (i.e., scaled by 8). The following magic
1806	* is equivalent to the smoothing algorithm in rfc793 with
1807	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1808	* point). Adjust rtt to origin 0.
1809	*/
1810	delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1811	if ((tp->t_srtt += delta) <= 0)
1812	tp->t_srtt = 1;
1813	/*
1814	* We accumulate a smoothed rtt variance (actually, a
1815	* smoothed mean difference), then set the retransmit
1816	* timer to smoothed rtt + 4 times the smoothed variance.
1817	* rttvar is stored as fixed point with 2 bits after the
1818	* binary point (scaled by 4). The following is
1819	* equivalent to rfc793 smoothing with an alpha of .75
1820	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1821	* rfc793's wired-in beta.
1822	*/
1823	if (delta < 0)
1824	delta = -delta;
1825	delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1826	if ((tp->t_rttvar += delta) <= 0)
1827	tp->t_rttvar = 1;
1828	}
1829	else
1830	{
1831	/*
1832	* No rtt measurement yet - use the unsmoothed rtt.
1833	* Set the variance to half the rtt (so our first
1834	* retransmit happens at 3*rtt).
1835	*/
1836	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1837	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1838	}
1839	tp->t_rtt = 0;
1840	tp->t_rxtshift = 0;
1841
1842	/*
1843	* the retransmit should happen at rtt + 4 * rttvar.
1844	* Because of the way we do the smoothing, srtt and rttvar
1845	* will each average +1/2 tick of bias. When we compute
1846	* the retransmit timer, we want 1/2 tick of rounding and
1847	* 1 extra tick because of +-1/2 tick uncertainty in the
1848	* firing of the timer. The bias will give us exactly the
1849	* 1.5 tick we need. But, because the bias is
1850	* statistical, we have to test that we don't drop below
1851	* the minimum feasible timer (which is 2 ticks).
1852	*/
1853	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1854	(short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1855
1856	/*
1857	* We received an ack for a packet that wasn't retransmitted;
1858	* it is probably safe to discard any error indications we've
1859	* received recently. This isn't quite right, but close enough
1860	* for now (a route might have failed after we sent a segment,
1861	* and the return path might not be symmetrical).
1862	*/
1863	tp->t_softerror = 0;
1864	}
1865
1866	/*
1867	* Determine a reasonable value for maxseg size.
1868	* If the route is known, check route for mtu.
1869	* If none, use an mss that can be handled on the outgoing
1870	* interface without forcing IP to fragment; if bigger than
1871	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1872	* to utilize large mbufs. If no route is found, route has no mtu,
1873	* or the destination isn't local, use a default, hopefully conservative
1874	* size (usually 512 or the default IP max size, but no more than the mtu
1875	* of the interface), as we can't discover anything about intervening
1876	* gateways or networks. We also initialize the congestion/slow start
1877	* window to be a single segment if the destination isn't local.
1878	* While looking at the routing entry, we also initialize other path-dependent
1879	* parameters from pre-set or cached values in the routing entry.
1880	*/
1881
1882	int
1883	tcp_mss(PNATState pData, register struct tcpcb *tp, u_int offer)
1884	{
1885	struct socket *so = tp->t_socket;
1886	int mss;
1887
1888	DEBUG_CALL("tcp_mss");
1889	DEBUG_ARG("tp = %lx", (long)tp);
1890	DEBUG_ARG("offer = %d", offer);
1891
1892	mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
1893	if (offer)
1894	mss = min(mss, offer);
1895	mss = max(mss, 32);
1896	if (mss < tp->t_maxseg \|\| offer != 0)
1897	tp->t_maxseg = mss;
1898
1899	tp->snd_cwnd = mss;
1900
1901	sbreserve(pData, &so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
1902	sbreserve(pData, &so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
1903
1904	DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1905
1906	return mss;
1907	}

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 28482

Download in other formats: