tcp_input.c@ 16292

Last change on this file since 16292 was 16291, checked in by vboxsync, 16 years ago
NAT: multi threading. Introduces set of macroces to use locking/unlocking/creating/destroing mutexes Also every socket enqueueing is accomponement with lock creation (it doen't work yet, compilable on Windows don't know about Linux)
Property svn:eol-style set to `native`
File size: 59.8 KB

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

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 16292

Download in other formats: