tcp_input.c@ 15522

Last change on this file since 15522 was 15472, checked in by vboxsync, 16 years ago
NAT: readability
Property svn:eol-style set to `native`
File size: 58.1 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 = malloc(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	free(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	free(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	sbappend(pData, so, q->tqe_m);
237	free(q);
238	tp->t_segqlen--;
239	tcp_reass_qsize--;
240	q = nq;
241	}
242	while (q && q->tqe_th->th_seq == tp->rcv_nxt);
243
244	return flags;
245	}
246
247	/*
248	* TCP input routine, follows pages 65-76 of the
249	* protocol specification dated September, 1981 very closely.
250	*/
251	void
252	tcp_input(PNATState pData, register struct mbuf m, int iphlen, struct socket inso)
253	{
254	struct ip save_ip, *ip;
255	register struct tcpiphdr *ti;
256	caddr_t optp = NULL;
257	int optlen = 0;
258	int len, tlen, off;
259	register struct tcpcb *tp = 0;
260	register int tiflags;
261	struct socket *so = 0;
262	int todrop, acked, ourfinisacked, needoutput = 0;
263	/* int dropsocket = 0; */
264	int iss = 0;
265	u_long tiwin;
266	/* int ts_present = 0; */
267
268	DEBUG_CALL("tcp_input");
269	DEBUG_ARGS((dfd," m = %8lx iphlen = %2d inso = %lx\n",
270	(long )m, iphlen, (long )inso ));
271
272	/*
273	* If called with m == 0, then we're continuing the connect
274	*/
275	if (m == NULL)
276	{
277	so = inso;
278
279	/* Re-set a few variables */
280	tp = sototcpcb(so);
281	m = so->so_m;
282	so->so_m = 0;
283	ti = so->so_ti;
284	tiwin = ti->ti_win;
285	tiflags = ti->ti_flags;
286
287	goto cont_conn;
288	}
289
290	tcpstat.tcps_rcvtotal++;
291	/*
292	* Get IP and TCP header together in first mbuf.
293	* Note: IP leaves IP header in first mbuf.
294	*/
295	ti = mtod(m, struct tcpiphdr *);
296	if (iphlen > sizeof(struct ip ))
297	{
298	ip_stripoptions(m, (struct mbuf *)0);
299	iphlen = sizeof(struct ip );
300	}
301	/* XXX Check if too short */
302
303
304	/*
305	* Save a copy of the IP header in case we want restore it
306	* for sending an ICMP error message in response.
307	*/
308	ip = mtod(m, struct ip *);
309	save_ip = *ip;
310	save_ip.ip_len+= iphlen;
311
312	/*
313	* Checksum extended TCP header and data.
314	*/
315	tlen = ((struct ip *)ti)->ip_len;
316	memset(ti->ti_x1, 0, 9);
317	ti->ti_len = htons((u_int16_t)tlen);
318	len = sizeof(struct ip ) + tlen;
319	/* keep checksum for ICMP reply
320	* ti->ti_sum = cksum(m, len);
321	* if (ti->ti_sum) { */
322	if (cksum(m, len))
323	{
324	tcpstat.tcps_rcvbadsum++;
325	goto drop;
326	}
327
328	/*
329	* Check that TCP offset makes sense,
330	* pull out TCP options and adjust length. XXX
331	*/
332	off = ti->ti_off << 2;
333	if ( off < sizeof (struct tcphdr)
334	\|\| off > tlen)
335	{
336	tcpstat.tcps_rcvbadoff++;
337	goto drop;
338	}
339	tlen -= off;
340	ti->ti_len = tlen;
341	if (off > sizeof (struct tcphdr))
342	{
343	optlen = off - sizeof (struct tcphdr);
344	optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
345
346	/*
347	* Do quick retrieval of timestamp options ("options
348	* prediction?"). If timestamp is the only option and it's
349	* formatted as recommended in RFC 1323 appendix A, we
350	* quickly get the values now and not bother calling
351	* tcp_dooptions(), etc.
352	*/
353	#if 0
354	if (( optlen == TCPOLEN_TSTAMP_APPA
355	\|\| ( optlen > TCPOLEN_TSTAMP_APPA
356	&& optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
357	(u_int32_t )optp == htonl(TCPOPT_TSTAMP_HDR) &&
358	(ti->ti_flags & TH_SYN) == 0)
359	{
360	ts_present = 1;
361	ts_val = ntohl((u_int32_t )(optp + 4));
362	ts_ecr = ntohl((u_int32_t )(optp + 8));
363	optp = NULL; / * we have parsed the options * /
364	}
365	#endif
366	}
367	tiflags = ti->ti_flags;
368
369	/*
370	* Convert TCP protocol specific fields to host format.
371	*/
372	NTOHL(ti->ti_seq);
373	NTOHL(ti->ti_ack);
374	NTOHS(ti->ti_win);
375	NTOHS(ti->ti_urp);
376
377	/*
378	* Drop TCP, IP headers and TCP options.
379	*/
380	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
381	m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
382
383	/*
384	* Locate pcb for segment.
385	*/
386	findso:
387	so = tcp_last_so;
388	if ( so->so_fport != ti->ti_dport
389	\|\| so->so_lport != ti->ti_sport
390	\|\| so->so_laddr.s_addr != ti->ti_src.s_addr
391	\|\| so->so_faddr.s_addr != ti->ti_dst.s_addr)
392	{
393	so = solookup(&tcb, ti->ti_src, ti->ti_sport,
394	ti->ti_dst, ti->ti_dport);
395	if (so)
396	tcp_last_so = so;
397	++tcpstat.tcps_socachemiss;
398	}
399
400	/*
401	* If the state is CLOSED (i.e., TCB does not exist) then
402	* all data in the incoming segment is discarded.
403	* If the TCB exists but is in CLOSED state, it is embryonic,
404	* but should either do a listen or a connect soon.
405	*
406	* state == CLOSED means we've done socreate() but haven't
407	* attached it to a protocol yet...
408	*
409	* XXX If a TCB does not exist, and the TH_SYN flag is
410	* the only flag set, then create a session, mark it
411	* as if it was LISTENING, and continue...
412	*/
413	if (so == 0)
414	{
415	if ((tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) != TH_SYN)
416	goto dropwithreset;
417
418	if ((so = socreate()) == NULL)
419	goto dropwithreset;
420	if (tcp_attach(pData, so) < 0)
421	{
422	free(so); /* Not sofree (if it failed, it's not insqued) */
423	goto dropwithreset;
424	}
425
426	sbreserve(&so->so_snd, tcp_sndspace);
427	sbreserve(&so->so_rcv, tcp_rcvspace);
428
429	/* tcp_last_so = so; / / XXX ? */
430	/* tp = sototcpcb(so); */
431
432	so->so_laddr = ti->ti_src;
433	so->so_lport = ti->ti_sport;
434	so->so_faddr = ti->ti_dst;
435	so->so_fport = ti->ti_dport;
436
437	if ((so->so_iptos = tcp_tos(so)) == 0)
438	so->so_iptos = ((struct ip *)ti)->ip_tos;
439
440	tp = sototcpcb(so);
441	tp->t_state = TCPS_LISTEN;
442	}
443
444	/*
445	* If this is a still-connecting socket, this probably
446	* a retransmit of the SYN. Whether it's a retransmit SYN
447	* or something else, we nuke it.
448	*/
449	if (so->so_state & SS_ISFCONNECTING)
450	goto drop;
451
452	tp = sototcpcb(so);
453
454	/* XXX Should never fail */
455	if (tp == 0)
456	goto dropwithreset;
457	if (tp->t_state == TCPS_CLOSED)
458	goto drop;
459
460	/* Unscale the window into a 32-bit value. */
461	/* if ((tiflags & TH_SYN) == 0)
462	* tiwin = ti->ti_win << tp->snd_scale;
463	* else
464	*/
465	tiwin = ti->ti_win;
466
467	/*
468	* Segment received on connection.
469	* Reset idle time and keep-alive timer.
470	*/
471	tp->t_idle = 0;
472	if (so_options)
473	tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
474	else
475	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
476
477	/*
478	* Process options if not in LISTEN state,
479	* else do it below (after getting remote address).
480	*/
481	if (optp && tp->t_state != TCPS_LISTEN)
482	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
483	/* , */
484	/* &ts_present, &ts_val, &ts_ecr); */
485
486	/*
487	* Header prediction: check for the two common cases
488	* of a uni-directional data xfer. If the packet has
489	* no control flags, is in-sequence, the window didn't
490	* change and we're not retransmitting, it's a
491	* candidate. If the length is zero and the ack moved
492	* forward, we're the sender side of the xfer. Just
493	* free the data acked & wake any higher level process
494	* that was blocked waiting for space. If the length
495	* is non-zero and the ack didn't move, we're the
496	* receiver side. If we're getting packets in-order
497	* (the reassembly queue is empty), add the data to
498	* the socket buffer and note that we need a delayed ack.
499	*
500	* XXX Some of these tests are not needed
501	* eg: the tiwin == tp->snd_wnd prevents many more
502	* predictions.. with no real advantage..
503	*/
504	if ( tp->t_state == TCPS_ESTABLISHED
505	&& (tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK
506	/* && (!ts_present \|\| TSTMP_GEQ(ts_val, tp->ts_recent)) */
507	&& ti->ti_seq == tp->rcv_nxt
508	&& tiwin && tiwin == tp->snd_wnd
509	&& tp->snd_nxt == tp->snd_max)
510	{
511	/*
512	* If last ACK falls within this segment's sequence numbers,
513	* record the timestamp.
514	*/
515	#if 0
516	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
517	SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) {
518	tp->ts_recent_age = tcp_now;
519	tp->ts_recent = ts_val;
520	}
521	#endif
522
523	if (ti->ti_len == 0)
524	{
525	if ( SEQ_GT(ti->ti_ack, tp->snd_una)
526	&& SEQ_LEQ(ti->ti_ack, tp->snd_max)
527	&& tp->snd_cwnd >= tp->snd_wnd)
528	{
529	/*
530	* this is a pure ack for outstanding data.
531	*/
532	++tcpstat.tcps_predack;
533	#if 0
534	if (ts_present)
535	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
536	else
537	#endif
538	if ( tp->t_rtt
539	&& SEQ_GT(ti->ti_ack, tp->t_rtseq))
540	tcp_xmit_timer(pData, tp, tp->t_rtt);
541	acked = ti->ti_ack - tp->snd_una;
542	tcpstat.tcps_rcvackpack++;
543	tcpstat.tcps_rcvackbyte += acked;
544	sbdrop(&so->so_snd, acked);
545	tp->snd_una = ti->ti_ack;
546	m_freem(pData, m);
547
548	/*
549	* If all outstanding data are acked, stop
550	* retransmit timer, otherwise restart timer
551	* using current (possibly backed-off) value.
552	* If process is waiting for space,
553	* wakeup/selwakeup/signal. If data
554	* are ready to send, let tcp_output
555	* decide between more output or persist.
556	*/
557	if (tp->snd_una == tp->snd_max)
558	tp->t_timer[TCPT_REXMT] = 0;
559	else if (tp->t_timer[TCPT_PERSIST] == 0)
560	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
561
562	/*
563	* There's room in so_snd, sowwakup will read()
564	* from the socket if we can
565	*/
566	#if 0
567	if (so->so_snd.sb_flags & SB_NOTIFY)
568	sowwakeup(so);
569	#endif
570	/*
571	* This is called because sowwakeup might have
572	* put data into so_snd. Since we don't so sowwakeup,
573	* we don't need this.. XXX???
574	*/
575	if (so->so_snd.sb_cc)
576	(void) tcp_output(pData, tp);
577
578	return;
579	}
580	}
581	else if ( ti->ti_ack == tp->snd_una
582	&& LIST_FIRST(&tp->t_segq)
583	&& ti->ti_len <= sbspace(&so->so_rcv))
584	{
585	/*
586	* this is a pure, in-sequence data packet
587	* with nothing on the reassembly queue and
588	* we have enough buffer space to take it.
589	*/
590	++tcpstat.tcps_preddat;
591	tp->rcv_nxt += ti->ti_len;
592	tcpstat.tcps_rcvpack++;
593	tcpstat.tcps_rcvbyte += ti->ti_len;
594	/*
595	* Add data to socket buffer.
596	*/
597	if (so->so_emu)
598	{
599	if (tcp_emu(pData, so,m)) sbappend(pData, so, m);
600	}
601	else
602	sbappend(pData, so, m);
603
604	/*
605	* XXX This is called when data arrives. Later, check
606	* if we can actually write() to the socket
607	* XXX Need to check? It's be NON_BLOCKING
608	*/
609	/* sorwakeup(so); */
610
611	/*
612	* If this is a short packet, then ACK now - with Nagel
613	* congestion avoidance sender won't send more until
614	* he gets an ACK.
615	*
616	* It is better to not delay acks at all to maximize
617	* TCP throughput. See RFC 2581.
618	*/
619	tp->t_flags \|= TF_ACKNOW;
620	tcp_output(pData, tp);
621	return;
622	}
623	} /* header prediction */
624	/*
625	* Calculate amount of space in receive window,
626	* and then do TCP input processing.
627	* Receive window is amount of space in rcv queue,
628	* but not less than advertised window.
629	*/
630	{
631	int win;
632	win = sbspace(&so->so_rcv);
633	if (win < 0)
634	win = 0;
635	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
636	}
637
638	switch (tp->t_state)
639	{
640	/*
641	* If the state is LISTEN then ignore segment if it contains an RST.
642	* If the segment contains an ACK then it is bad and send a RST.
643	* If it does not contain a SYN then it is not interesting; drop it.
644	* Don't bother responding if the destination was a broadcast.
645	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
646	* tp->iss, and send a segment:
647	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
648	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
649	* Fill in remote peer address fields if not previously specified.
650	* Enter SYN_RECEIVED state, and process any other fields of this
651	* segment in this state.
652	*/
653	case TCPS_LISTEN:
654	{
655	if (tiflags & TH_RST)
656	goto drop;
657	if (tiflags & TH_ACK)
658	goto dropwithreset;
659	if ((tiflags & TH_SYN) == 0)
660	goto drop;
661
662	/*
663	* This has way too many gotos...
664	* But a bit of spaghetti code never hurt anybody :)
665	*/
666
667	if (so->so_emu & EMU_NOCONNECT)
668	{
669	so->so_emu &= ~EMU_NOCONNECT;
670	goto cont_input;
671	}
672
673	if ( (tcp_fconnect(pData, so) == -1)
674	&& errno != EINPROGRESS
675	&& errno != EWOULDBLOCK)
676	{
677	u_char code = ICMP_UNREACH_NET;
678	DEBUG_MISC((dfd," tcp fconnect errno = %d-%s\n",
679	errno,strerror(errno)));
680	if (errno == ECONNREFUSED)
681	{
682	/* ACK the SYN, send RST to refuse the connection */
683	tcp_respond(pData, tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
684	TH_RST\|TH_ACK);
685	}
686	else
687	{
688	if (errno == EHOSTUNREACH)
689	code = ICMP_UNREACH_HOST;
690	HTONL(ti->ti_seq); /* restore tcp header */
691	HTONL(ti->ti_ack);
692	HTONS(ti->ti_win);
693	HTONS(ti->ti_urp);
694	m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
695	m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
696	*ip = save_ip;
697	icmp_error(pData, m, ICMP_UNREACH,code, 0,strerror(errno));
698	}
699	tp = tcp_close(pData, tp);
700	m_free(pData, m);
701	}
702	else
703	{
704	/*
705	* Haven't connected yet, save the current mbuf
706	* and ti, and return
707	* XXX Some OS's don't tell us whether the connect()
708	* succeeded or not. So we must time it out.
709	*/
710	so->so_m = m;
711	so->so_ti = ti;
712	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
713	tp->t_state = TCPS_SYN_RECEIVED;
714	}
715	return;
716
717	cont_conn:
718	/* m==NULL
719	* Check if the connect succeeded
720	*/
721	if (so->so_state & SS_NOFDREF)
722	{
723	tp = tcp_close(pData, tp);
724	goto dropwithreset;
725	}
726	cont_input:
727	tcp_template(tp);
728
729	if (optp)
730	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
731
732	if (iss)
733	tp->iss = iss;
734	else
735	tp->iss = tcp_iss;
736	tcp_iss += TCP_ISSINCR/2;
737	tp->irs = ti->ti_seq;
738	tcp_sendseqinit(tp);
739	tcp_rcvseqinit(tp);
740	tp->t_flags \|= TF_ACKNOW;
741	tp->t_state = TCPS_SYN_RECEIVED;
742	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
743	tcpstat.tcps_accepts++;
744	goto trimthenstep6;
745	} /* case TCPS_LISTEN */
746
747	/*
748	* If the state is SYN_SENT:
749	* if seg contains an ACK, but not for our SYN, drop the input.
750	* if seg contains a RST, then drop the connection.
751	* if seg does not contain SYN, then drop it.
752	* Otherwise this is an acceptable SYN segment
753	* initialize tp->rcv_nxt and tp->irs
754	* if seg contains ack then advance tp->snd_una
755	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
756	* arrange for segment to be acked (eventually)
757	* continue processing rest of data/controls, beginning with URG
758	*/
759	case TCPS_SYN_SENT:
760	if ( (tiflags & TH_ACK)
761	&& ( SEQ_LEQ(ti->ti_ack, tp->iss)
762	\|\| SEQ_GT(ti->ti_ack, tp->snd_max)))
763	goto dropwithreset;
764
765	if (tiflags & TH_RST)
766	{
767	if (tiflags & TH_ACK)
768	tp = tcp_drop(pData, tp,0); /* XXX Check t_softerror! */
769	goto drop;
770	}
771
772	if ((tiflags & TH_SYN) == 0)
773	goto drop;
774	if (tiflags & TH_ACK)
775	{
776	tp->snd_una = ti->ti_ack;
777	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
778	tp->snd_nxt = tp->snd_una;
779	}
780
781	tp->t_timer[TCPT_REXMT] = 0;
782	tp->irs = ti->ti_seq;
783	tcp_rcvseqinit(tp);
784	tp->t_flags \|= TF_ACKNOW;
785	if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss))
786	{
787	tcpstat.tcps_connects++;
788	soisfconnected(so);
789	tp->t_state = TCPS_ESTABLISHED;
790
791	/* Do window scaling on this connection? */
792	#if 0
793	if (( tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
794	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
795	{
796	tp->snd_scale = tp->requested_s_scale;
797	tp->rcv_scale = tp->request_r_scale;
798	}
799	#endif
800	(void) tcp_reass(pData, tp, (struct tcphdr )0, NULL, (struct mbuf )0);
801	/*
802	* if we didn't have to retransmit the SYN,
803	* use its rtt as our initial srtt & rtt var.
804	*/
805	if (tp->t_rtt)
806	tcp_xmit_timer(pData, tp, tp->t_rtt);
807	}
808	else
809	tp->t_state = TCPS_SYN_RECEIVED;
810
811	trimthenstep6:
812	/*
813	* Advance ti->ti_seq to correspond to first data byte.
814	* If data, trim to stay within window,
815	* dropping FIN if necessary.
816	*/
817	ti->ti_seq++;
818	if (ti->ti_len > tp->rcv_wnd)
819	{
820	todrop = ti->ti_len - tp->rcv_wnd;
821	m_adj(m, -todrop);
822	ti->ti_len = tp->rcv_wnd;
823	tiflags &= ~TH_FIN;
824	tcpstat.tcps_rcvpackafterwin++;
825	tcpstat.tcps_rcvbyteafterwin += todrop;
826	}
827	tp->snd_wl1 = ti->ti_seq - 1;
828	tp->rcv_up = ti->ti_seq;
829	goto step6;
830	} /* switch tp->t_state */
831	/*
832	* States other than LISTEN or SYN_SENT.
833	* First check timestamp, if present.
834	* Then check that at least some bytes of segment are within
835	* receive window. If segment begins before rcv_nxt,
836	* drop leading data (and SYN); if nothing left, just ack.
837	*
838	* RFC 1323 PAWS: If we have a timestamp reply on this segment
839	* and it's less than ts_recent, drop it.
840	*/
841	#if 0
842	if ( ts_present
843	&& (tiflags & TH_RST) == 0
844	&& tp->ts_recent
845	&& TSTMP_LT(ts_val, tp->ts_recent))
846	{
847	/* Check to see if ts_recent is over 24 days old. */
848	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE)
849	{
850	/*
851	* Invalidate ts_recent. If this segment updates
852	* ts_recent, the age will be reset later and ts_recent
853	* will get a valid value. If it does not, setting
854	* ts_recent to zero will at least satisfy the
855	* requirement that zero be placed in the timestamp
856	* echo reply when ts_recent isn't valid. The
857	* age isn't reset until we get a valid ts_recent
858	* because we don't want out-of-order segments to be
859	* dropped when ts_recent is old.
860	*/
861	tp->ts_recent = 0;
862	}
863	else
864	{
865	tcpstat.tcps_rcvduppack++;
866	tcpstat.tcps_rcvdupbyte += ti->ti_len;
867	tcpstat.tcps_pawsdrop++;
868	goto dropafterack;
869	}
870	}
871	#endif
872
873	todrop = tp->rcv_nxt - ti->ti_seq;
874	if (todrop > 0)
875	{
876	if (tiflags & TH_SYN)
877	{
878	tiflags &= ~TH_SYN;
879	ti->ti_seq++;
880	if (ti->ti_urp > 1)
881	ti->ti_urp--;
882	else
883	tiflags &= ~TH_URG;
884	todrop--;
885	}
886	/*
887	* Following if statement from Stevens, vol. 2, p. 960.
888	*/
889	if ( todrop > ti->ti_len
890	\|\| ( todrop == ti->ti_len
891	&& (tiflags & TH_FIN) == 0))
892	{
893	/*
894	* Any valid FIN must be to the left of the window.
895	* At this point the FIN must be a duplicate or out
896	* of sequence; drop it.
897	*/
898	tiflags &= ~TH_FIN;
899
900	/*
901	* Send an ACK to resynchronize and drop any data.
902	* But keep on processing for RST or ACK.
903	*/
904	tp->t_flags \|= TF_ACKNOW;
905	todrop = ti->ti_len;
906	tcpstat.tcps_rcvduppack++;
907	tcpstat.tcps_rcvdupbyte += todrop;
908	}
909	else
910	{
911	tcpstat.tcps_rcvpartduppack++;
912	tcpstat.tcps_rcvpartdupbyte += todrop;
913	}
914	m_adj(m, todrop);
915	ti->ti_seq += todrop;
916	ti->ti_len -= todrop;
917	if (ti->ti_urp > todrop)
918	ti->ti_urp -= todrop;
919	else
920	{
921	tiflags &= ~TH_URG;
922	ti->ti_urp = 0;
923	}
924	}
925	/*
926	* If new data are received on a connection after the
927	* user processes are gone, then RST the other end.
928	*/
929	if ( (so->so_state & SS_NOFDREF)
930	&& tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len)
931	{
932	tp = tcp_close(pData, tp);
933	tcpstat.tcps_rcvafterclose++;
934	goto dropwithreset;
935	}
936
937	/*
938	* If segment ends after window, drop trailing data
939	* (and PUSH and FIN); if nothing left, just ACK.
940	*/
941	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
942	if (todrop > 0)
943	{
944	tcpstat.tcps_rcvpackafterwin++;
945	if (todrop >= ti->ti_len)
946	{
947	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
948	/*
949	* If a new connection request is received
950	* while in TIME_WAIT, drop the old connection
951	* and start over if the sequence numbers
952	* are above the previous ones.
953	*/
954	if ( tiflags & TH_SYN
955	&& tp->t_state == TCPS_TIME_WAIT
956	&& SEQ_GT(ti->ti_seq, tp->rcv_nxt))
957	{
958	iss = tp->rcv_nxt + TCP_ISSINCR;
959	tp = tcp_close(pData, tp);
960	goto findso;
961	}
962	/*
963	* If window is closed can only take segments at
964	* window edge, and have to drop data and PUSH from
965	* incoming segments. Continue processing, but
966	* remember to ack. Otherwise, drop segment
967	* and ack.
968	*/
969	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt)
970	{
971	tp->t_flags \|= TF_ACKNOW;
972	tcpstat.tcps_rcvwinprobe++;
973	}
974	else
975	goto dropafterack;
976	}
977	else
978	tcpstat.tcps_rcvbyteafterwin += todrop;
979	m_adj(m, -todrop);
980	ti->ti_len -= todrop;
981	tiflags &= ~(TH_PUSH\|TH_FIN);
982	}
983
984	/*
985	* If last ACK falls within this segment's sequence numbers,
986	* record its timestamp.
987	*/
988	#if 0
989	if ( ts_present
990	&& SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)
991	&& SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + ((tiflags & (TH_SYN\|TH_FIN)) != 0)))
992	{
993	tp->ts_recent_age = tcp_now;
994	tp->ts_recent = ts_val;
995	}
996	#endif
997
998	/*
999	* If the RST bit is set examine the state:
1000	* SYN_RECEIVED STATE:
1001	* If passive open, return to LISTEN state.
1002	* If active open, inform user that connection was refused.
1003	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1004	* Inform user that connection was reset, and close tcb.
1005	* CLOSING, LAST_ACK, TIME_WAIT STATES
1006	* Close the tcb.
1007	*/
1008	if (tiflags&TH_RST)
1009	switch (tp->t_state)
1010	{
1011	case TCPS_SYN_RECEIVED:
1012	/* so->so_error = ECONNREFUSED; */
1013	goto close;
1014
1015	case TCPS_ESTABLISHED:
1016	case TCPS_FIN_WAIT_1:
1017	case TCPS_FIN_WAIT_2:
1018	case TCPS_CLOSE_WAIT:
1019	/* so->so_error = ECONNRESET; */
1020	close:
1021	tp->t_state = TCPS_CLOSED;
1022	tcpstat.tcps_drops++;
1023	tp = tcp_close(pData, tp);
1024	goto drop;
1025
1026	case TCPS_CLOSING:
1027	case TCPS_LAST_ACK:
1028	case TCPS_TIME_WAIT:
1029	tp = tcp_close(pData, tp);
1030	goto drop;
1031	}
1032
1033	/*
1034	* If a SYN is in the window, then this is an
1035	* error and we send an RST and drop the connection.
1036	*/
1037	if (tiflags & TH_SYN)
1038	{
1039	tp = tcp_drop(pData, tp,0);
1040	goto dropwithreset;
1041	}
1042
1043	/*
1044	* If the ACK bit is off we drop the segment and return.
1045	*/
1046	if ((tiflags & TH_ACK) == 0)
1047	goto drop;
1048
1049	/*
1050	* Ack processing.
1051	*/
1052	switch (tp->t_state)
1053	{
1054	/*
1055	* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1056	* ESTABLISHED state and continue processing, otherwise
1057	* send an RST. una<=ack<=max
1058	*/
1059	case TCPS_SYN_RECEIVED:
1060	if ( SEQ_GT(tp->snd_una, ti->ti_ack)
1061	\|\| SEQ_GT(ti->ti_ack, tp->snd_max))
1062	goto dropwithreset;
1063	tcpstat.tcps_connects++;
1064	tp->t_state = TCPS_ESTABLISHED;
1065	/*
1066	* The sent SYN is ack'ed with our sequence number +1
1067	* The first data byte already in the buffer will get
1068	* lost if no correction is made. This is only needed for
1069	* SS_CTL since the buffer is empty otherwise.
1070	* tp->snd_una++; or:
1071	*/
1072	tp->snd_una = ti->ti_ack;
1073	soisfconnected(so);
1074
1075	/* Do window scaling? */
1076	#if 0
1077	if ( (tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
1078	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
1079	{
1080	tp->snd_scale = tp->requested_s_scale;
1081	tp->rcv_scale = tp->request_r_scale;
1082	}
1083	#endif
1084	(void) tcp_reass(pData, tp, (struct tcphdr )0, (int )0, (struct mbuf *)0);
1085	tp->snd_wl1 = ti->ti_seq - 1;
1086	/* Avoid ack processing; snd_una==ti_ack => dup ack */
1087	goto synrx_to_est;
1088	/* fall into ... */
1089
1090	/*
1091	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1092	* ACKs. If the ack is in the range
1093	* tp->snd_una < ti->ti_ack <= tp->snd_max
1094	* then advance tp->snd_una to ti->ti_ack and drop
1095	* data from the retransmission queue. If this ACK reflects
1096	* more up to date window information we update our window information.
1097	*/
1098	case TCPS_ESTABLISHED:
1099	case TCPS_FIN_WAIT_1:
1100	case TCPS_FIN_WAIT_2:
1101	case TCPS_CLOSE_WAIT:
1102	case TCPS_CLOSING:
1103	case TCPS_LAST_ACK:
1104	case TCPS_TIME_WAIT:
1105	if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
1106	{
1107	if (ti->ti_len == 0 && tiwin == tp->snd_wnd)
1108	{
1109	tcpstat.tcps_rcvdupack++;
1110	DEBUG_MISC((dfd," dup ack m = %lx so = %lx \n",
1111	(long )m, (long )so));
1112	/*
1113	* If we have outstanding data (other than
1114	* a window probe), this is a completely
1115	* duplicate ack (ie, window info didn't
1116	* change), the ack is the biggest we've
1117	* seen and we've seen exactly our rexmt
1118	* threshold of them, assume a packet
1119	* has been dropped and retransmit it.
1120	* Kludge snd_nxt & the congestion
1121	* window so we send only this one
1122	* packet.
1123	*
1124	* We know we're losing at the current
1125	* window size so do congestion avoidance
1126	* (set ssthresh to half the current window
1127	* and pull our congestion window back to
1128	* the new ssthresh).
1129	*
1130	* Dup acks mean that packets have left the
1131	* network (they're now cached at the receiver)
1132	* so bump cwnd by the amount in the receiver
1133	* to keep a constant cwnd packets in the
1134	* network.
1135	*/
1136	if ( tp->t_timer[TCPT_REXMT] == 0
1137	\|\| ti->ti_ack != tp->snd_una)
1138	tp->t_dupacks = 0;
1139	else if (++tp->t_dupacks == tcprexmtthresh)
1140	{
1141	tcp_seq onxt = tp->snd_nxt;
1142	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
1143	if (win < 2)
1144	win = 2;
1145	tp->snd_ssthresh = win * tp->t_maxseg;
1146	tp->t_timer[TCPT_REXMT] = 0;
1147	tp->t_rtt = 0;
1148	tp->snd_nxt = ti->ti_ack;
1149	tp->snd_cwnd = tp->t_maxseg;
1150	(void) tcp_output(pData, tp);
1151	tp->snd_cwnd = tp->snd_ssthresh +
1152	tp->t_maxseg * tp->t_dupacks;
1153	if (SEQ_GT(onxt, tp->snd_nxt))
1154	tp->snd_nxt = onxt;
1155	goto drop;
1156	}
1157	else if (tp->t_dupacks > tcprexmtthresh)
1158	{
1159	tp->snd_cwnd += tp->t_maxseg;
1160	(void) tcp_output(pData, tp);
1161	goto drop;
1162	}
1163	}
1164	else
1165	tp->t_dupacks = 0;
1166	break;
1167	}
1168	synrx_to_est:
1169	/*
1170	* If the congestion window was inflated to account
1171	* for the other side's cached packets, retract it.
1172	*/
1173	if ( tp->t_dupacks > tcprexmtthresh
1174	&& tp->snd_cwnd > tp->snd_ssthresh)
1175	tp->snd_cwnd = tp->snd_ssthresh;
1176	tp->t_dupacks = 0;
1177	if (SEQ_GT(ti->ti_ack, tp->snd_max))
1178	{
1179	tcpstat.tcps_rcvacktoomuch++;
1180	goto dropafterack;
1181	}
1182	acked = ti->ti_ack - tp->snd_una;
1183	tcpstat.tcps_rcvackpack++;
1184	tcpstat.tcps_rcvackbyte += acked;
1185
1186	/*
1187	* If we have a timestamp reply, update smoothed
1188	* round trip time. If no timestamp is present but
1189	* transmit timer is running and timed sequence
1190	* number was acked, update smoothed round trip time.
1191	* Since we now have an rtt measurement, cancel the
1192	* timer backoff (cf., Phil Karn's retransmit alg.).
1193	* Recompute the initial retransmit timer.
1194	*/
1195	#if 0
1196	if (ts_present)
1197	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1198	else
1199	#endif
1200	if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1201	tcp_xmit_timer(pData, tp,tp->t_rtt);
1202
1203	/*
1204	* If all outstanding data is acked, stop retransmit
1205	* timer and remember to restart (more output or persist).
1206	* If there is more data to be acked, restart retransmit
1207	* timer, using current (possibly backed-off) value.
1208	*/
1209	if (ti->ti_ack == tp->snd_max)
1210	{
1211	tp->t_timer[TCPT_REXMT] = 0;
1212	needoutput = 1;
1213	}
1214	else if (tp->t_timer[TCPT_PERSIST] == 0)
1215	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1216	/*
1217	* When new data is acked, open the congestion window.
1218	* If the window gives us less than ssthresh packets
1219	* in flight, open exponentially (maxseg per packet).
1220	* Otherwise open linearly: maxseg per window
1221	* (maxseg^2 / cwnd per packet).
1222	*/
1223	{
1224	register u_int cw = tp->snd_cwnd;
1225	register u_int incr = tp->t_maxseg;
1226
1227	if (cw > tp->snd_ssthresh)
1228	incr = incr * incr / cw;
1229	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1230	}
1231	if (acked > so->so_snd.sb_cc)
1232	{
1233	tp->snd_wnd -= so->so_snd.sb_cc;
1234	sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1235	ourfinisacked = 1;
1236	}
1237	else
1238	{
1239	sbdrop(&so->so_snd, acked);
1240	tp->snd_wnd -= acked;
1241	ourfinisacked = 0;
1242	}
1243	/*
1244	* XXX sowwakup is called when data is acked and there's room for
1245	* for more data... it should read() the socket
1246	*/
1247	#if 0
1248	if (so->so_snd.sb_flags & SB_NOTIFY)
1249	sowwakeup(so);
1250	#endif
1251	tp->snd_una = ti->ti_ack;
1252	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1253	tp->snd_nxt = tp->snd_una;
1254
1255	switch (tp->t_state)
1256	{
1257	/*
1258	* In FIN_WAIT_1 STATE in addition to the processing
1259	* for the ESTABLISHED state if our FIN is now acknowledged
1260	* then enter FIN_WAIT_2.
1261	*/
1262	case TCPS_FIN_WAIT_1:
1263	if (ourfinisacked)
1264	{
1265	/*
1266	* If we can't receive any more
1267	* data, then closing user can proceed.
1268	* Starting the timer is contrary to the
1269	* specification, but if we don't get a FIN
1270	* we'll hang forever.
1271	*/
1272	if (so->so_state & SS_FCANTRCVMORE)
1273	{
1274	soisfdisconnected(so);
1275	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1276	}
1277	tp->t_state = TCPS_FIN_WAIT_2;
1278	}
1279	break;
1280
1281	/*
1282	* In CLOSING STATE in addition to the processing for
1283	* the ESTABLISHED state if the ACK acknowledges our FIN
1284	* then enter the TIME-WAIT state, otherwise ignore
1285	* the segment.
1286	*/
1287	case TCPS_CLOSING:
1288	if (ourfinisacked)
1289	{
1290	tp->t_state = TCPS_TIME_WAIT;
1291	tcp_canceltimers(tp);
1292	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1293	soisfdisconnected(so);
1294	}
1295	break;
1296
1297	/*
1298	* In LAST_ACK, we may still be waiting for data to drain
1299	* and/or to be acked, as well as for the ack of our FIN.
1300	* If our FIN is now acknowledged, delete the TCB,
1301	* enter the closed state and return.
1302	*/
1303	case TCPS_LAST_ACK:
1304	if (ourfinisacked)
1305	{
1306	tp = tcp_close(pData, tp);
1307	goto drop;
1308	}
1309	break;
1310
1311	/*
1312	* In TIME_WAIT state the only thing that should arrive
1313	* is a retransmission of the remote FIN. Acknowledge
1314	* it and restart the finack timer.
1315	*/
1316	case TCPS_TIME_WAIT:
1317	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1318	goto dropafterack;
1319	}
1320	} /* switch(tp->t_state) */
1321
1322	step6:
1323	/*
1324	* Update window information.
1325	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1326	*/
1327	if ( (tiflags & TH_ACK)
1328	&& ( SEQ_LT(tp->snd_wl1, ti->ti_seq)
1329	\|\| ( tp->snd_wl1 == ti->ti_seq
1330	&& ( SEQ_LT(tp->snd_wl2, ti->ti_ack)
1331	\|\| ( tp->snd_wl2 == ti->ti_ack
1332	&& tiwin > tp->snd_wnd)))))
1333	{
1334	/* keep track of pure window updates */
1335	if ( ti->ti_len == 0
1336	&& tp->snd_wl2 == ti->ti_ack
1337	&& tiwin > tp->snd_wnd)
1338	tcpstat.tcps_rcvwinupd++;
1339	tp->snd_wnd = tiwin;
1340	tp->snd_wl1 = ti->ti_seq;
1341	tp->snd_wl2 = ti->ti_ack;
1342	if (tp->snd_wnd > tp->max_sndwnd)
1343	tp->max_sndwnd = tp->snd_wnd;
1344	needoutput = 1;
1345	}
1346
1347	/*
1348	* Process segments with URG.
1349	*/
1350	if ((tiflags & TH_URG) && ti->ti_urp &&
1351	TCPS_HAVERCVDFIN(tp->t_state) == 0)
1352	{
1353	/*
1354	* This is a kludge, but if we receive and accept
1355	* random urgent pointers, we'll crash in
1356	* soreceive. It's hard to imagine someone
1357	* actually wanting to send this much urgent data.
1358	*/
1359	if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen)
1360	{
1361	ti->ti_urp = 0;
1362	tiflags &= ~TH_URG;
1363	goto dodata;
1364	}
1365	/*
1366	* If this segment advances the known urgent pointer,
1367	* then mark the data stream. This should not happen
1368	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1369	* a FIN has been received from the remote side.
1370	* In these states we ignore the URG.
1371	*
1372	* According to RFC961 (Assigned Protocols),
1373	* the urgent pointer points to the last octet
1374	* of urgent data. We continue, however,
1375	* to consider it to indicate the first octet
1376	* of data past the urgent section as the original
1377	* spec states (in one of two places).
1378	*/
1379	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up))
1380	{
1381	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1382	so->so_urgc = so->so_rcv.sb_cc +
1383	(tp->rcv_up - tp->rcv_nxt); /* -1; */
1384	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1385	}
1386	}
1387	else
1388	/*
1389	* If no out of band data is expected,
1390	* pull receive urgent pointer along
1391	* with the receive window.
1392	*/
1393	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1394	tp->rcv_up = tp->rcv_nxt;
1395	dodata:
1396
1397	/*
1398	* If this is a small packet, then ACK now - with Nagel
1399	* congestion avoidance sender won't send more until
1400	* he gets an ACK.
1401	*
1402	* See above.
1403	*/
1404	if ( ti->ti_len
1405	&& (unsigned)ti->ti_len <= 5
1406	&& ((struct tcpiphdr_2 *)ti)->first_char == (char)27)
1407	{
1408	tp->t_flags \|= TF_ACKNOW;
1409	}
1410
1411	/*
1412	* Process the segment text, merging it into the TCP sequencing queue,
1413	* and arranging for acknowledgment of receipt if necessary.
1414	* This process logically involves adjusting tp->rcv_wnd as data
1415	* is presented to the user (this happens in tcp_usrreq.c,
1416	* case PRU_RCVD). If a FIN has already been received on this
1417	* connection then we just ignore the text.
1418	*/
1419	if ( (ti->ti_len \|\| (tiflags&TH_FIN))
1420	&& TCPS_HAVERCVDFIN(tp->t_state) == 0)
1421	{
1422	if ( ti->ti_seq == tp->rcv_nxt
1423	&& LIST_EMPTY(&tp->t_segq)
1424	&& tp->t_state == TCPS_ESTABLISHED)
1425	{
1426	DELAY_ACK(tp, ti); /* little bit different from BSD declaration see netinet/tcp_input.c */
1427	tp->rcv_nxt += tlen;
1428	tiflags = ti->ti_t.th_flags & TH_FIN;
1429	tcpstat.tcps_rcvpack++;
1430	tcpstat.tcps_rcvbyte += tlen;
1431	if (so->so_state & SS_FCANTRCVMORE)
1432	m_freem(pData, m);
1433	else
1434	sbappend(pData, so, m);
1435	}
1436	else
1437	{
1438	tiflags = tcp_reass(pData, tp, &ti->ti_t, &tlen, m);
1439	tiflags \|= TF_ACKNOW;
1440	}
1441	/*
1442	* Note the amount of data that peer has sent into
1443	* our window, in order to estimate the sender's
1444	* buffer size.
1445	*/
1446	len = so->so_rcv.sb_datalen - (tp->rcv_adv - tp->rcv_nxt);
1447	}
1448	else
1449	{
1450	m_free(pData, m);
1451	tiflags &= ~TH_FIN;
1452	}
1453
1454	/*
1455	* If FIN is received ACK the FIN and let the user know
1456	* that the connection is closing.
1457	*/
1458	if (tiflags & TH_FIN)
1459	{
1460	if (TCPS_HAVERCVDFIN(tp->t_state) == 0)
1461	{
1462	/*
1463	* If we receive a FIN we can't send more data,
1464	* set it SS_FDRAIN
1465	* Shutdown the socket if there is no rx data in the
1466	* buffer.
1467	* soread() is called on completion of shutdown() and
1468	* will got to TCPS_LAST_ACK, and use tcp_output()
1469	* to send the FIN.
1470	*/
1471	/* sofcantrcvmore(so); */
1472	sofwdrain(so);
1473
1474	tp->t_flags \|= TF_ACKNOW;
1475	tp->rcv_nxt++;
1476	}
1477	switch (tp->t_state)
1478	{
1479	/*
1480	* In SYN_RECEIVED and ESTABLISHED STATES
1481	* enter the CLOSE_WAIT state.
1482	*/
1483	case TCPS_SYN_RECEIVED:
1484	case TCPS_ESTABLISHED:
1485	if(so->so_emu == EMU_CTL) /* no shutdown on socket */
1486	tp->t_state = TCPS_LAST_ACK;
1487	else
1488	tp->t_state = TCPS_CLOSE_WAIT;
1489	break;
1490
1491	/*
1492	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1493	* enter the CLOSING state.
1494	*/
1495	case TCPS_FIN_WAIT_1:
1496	tp->t_state = TCPS_CLOSING;
1497	break;
1498
1499	/*
1500	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1501	* starting the time-wait timer, turning off the other
1502	* standard timers.
1503	*/
1504	case TCPS_FIN_WAIT_2:
1505	tp->t_state = TCPS_TIME_WAIT;
1506	tcp_canceltimers(tp);
1507	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1508	soisfdisconnected(so);
1509	break;
1510
1511	/*
1512	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1513	*/
1514	case TCPS_TIME_WAIT:
1515	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1516	break;
1517	}
1518	}
1519
1520	/*
1521	* Return any desired output.
1522	*/
1523	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1524	tcp_output(pData, tp);
1525
1526	return;
1527
1528	dropafterack:
1529	/*
1530	* Generate an ACK dropping incoming segment if it occupies
1531	* sequence space, where the ACK reflects our state.
1532	*/
1533	if (tiflags & TH_RST)
1534	goto drop;
1535	m_freem(pData, m);
1536	tp->t_flags \|= TF_ACKNOW;
1537	(void) tcp_output(pData, tp);
1538	return;
1539
1540	dropwithreset:
1541	/* reuses m if m!=NULL, m_free() unnecessary */
1542	if (tiflags & TH_ACK)
1543	tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1544	else
1545	{
1546	if (tiflags & TH_SYN) ti->ti_len++;
1547	tcp_respond(pData, tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1548	TH_RST\|TH_ACK);
1549	}
1550
1551	return;
1552
1553	drop:
1554	/*
1555	* Drop space held by incoming segment and return.
1556	*/
1557	m_free(pData, m);
1558
1559	return;
1560	}
1561
1562	void
1563	tcp_dooptions(PNATState pData, struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti)
1564	{
1565	u_int16_t mss;
1566	int opt, optlen;
1567
1568	DEBUG_CALL("tcp_dooptions");
1569	DEBUG_ARGS((dfd," tp = %lx cnt=%i \n", (long )tp, cnt));
1570
1571	for (; cnt > 0; cnt -= optlen, cp += optlen)
1572	{
1573	opt = cp[0];
1574	if (opt == TCPOPT_EOL)
1575	break;
1576	if (opt == TCPOPT_NOP)
1577	optlen = 1;
1578	else
1579	{
1580	optlen = cp[1];
1581	if (optlen <= 0)
1582	break;
1583	}
1584	switch (opt)
1585	{
1586	default:
1587	continue;
1588
1589	case TCPOPT_MAXSEG:
1590	if (optlen != TCPOLEN_MAXSEG)
1591	continue;
1592	if (!(ti->ti_flags & TH_SYN))
1593	continue;
1594	memcpy((char ) &mss, (char ) cp + 2, sizeof(mss));
1595	NTOHS(mss);
1596	(void) tcp_mss(pData, tp, mss); /* sets t_maxseg */
1597	break;
1598
1599	#if 0
1600	case TCPOPT_WINDOW:
1601	if (optlen != TCPOLEN_WINDOW)
1602	continue;
1603	if (!(ti->ti_flags & TH_SYN))
1604	continue;
1605	tp->t_flags \|= TF_RCVD_SCALE;
1606	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1607	break;
1608
1609	case TCPOPT_TIMESTAMP:
1610	if (optlen != TCPOLEN_TIMESTAMP)
1611	continue;
1612	*ts_present = 1;
1613	memcpy((char ) ts_val, (char )cp + 2, sizeof(*ts_val));
1614	NTOHL(*ts_val);
1615	memcpy((char ) ts_ecr, (char )cp + 6, sizeof(*ts_ecr));
1616	NTOHL(*ts_ecr);
1617
1618	/*
1619	* A timestamp received in a SYN makes
1620	* it ok to send timestamp requests and replies.
1621	*/
1622	if (ti->ti_flags & TH_SYN)
1623	{
1624	tp->t_flags \|= TF_RCVD_TSTMP;
1625	tp->ts_recent = *ts_val;
1626	tp->ts_recent_age = tcp_now;
1627	}
1628	break;
1629	#endif
1630	}
1631	}
1632	}
1633
1634
1635	/*
1636	* Pull out of band byte out of a segment so
1637	* it doesn't appear in the user's data queue.
1638	* It is still reflected in the segment length for
1639	* sequencing purposes.
1640	*/
1641
1642	#if 0
1643	void
1644	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1645	{
1646	int cnt = ti->ti_urp - 1;
1647
1648	while (cnt >= 0)
1649	{
1650	if (m->m_len > cnt)
1651	{
1652	char *cp = mtod(m, caddr_t) + cnt;
1653	struct tcpcb *tp = sototcpcb(so);
1654
1655	tp->t_iobc = *cp;
1656	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1657	memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1658	m->m_len--;
1659	return;
1660	}
1661	cnt -= m->m_len;
1662	m = m->m_next; /* XXX WRONG! Fix it! */
1663	if (m == 0)
1664	break;
1665	}
1666	panic("tcp_pulloutofband");
1667	}
1668	#endif
1669
1670	/*
1671	* Collect new round-trip time estimate
1672	* and update averages and current timeout.
1673	*/
1674
1675	void
1676	tcp_xmit_timer(PNATState pData, register struct tcpcb *tp, int rtt)
1677	{
1678	register short delta;
1679
1680	DEBUG_CALL("tcp_xmit_timer");
1681	DEBUG_ARG("tp = %lx", (long)tp);
1682	DEBUG_ARG("rtt = %d", rtt);
1683
1684	tcpstat.tcps_rttupdated++;
1685	if (tp->t_srtt != 0)
1686	{
1687	/*
1688	* srtt is stored as fixed point with 3 bits after the
1689	* binary point (i.e., scaled by 8). The following magic
1690	* is equivalent to the smoothing algorithm in rfc793 with
1691	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1692	* point). Adjust rtt to origin 0.
1693	*/
1694	delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1695	if ((tp->t_srtt += delta) <= 0)
1696	tp->t_srtt = 1;
1697	/*
1698	* We accumulate a smoothed rtt variance (actually, a
1699	* smoothed mean difference), then set the retransmit
1700	* timer to smoothed rtt + 4 times the smoothed variance.
1701	* rttvar is stored as fixed point with 2 bits after the
1702	* binary point (scaled by 4). The following is
1703	* equivalent to rfc793 smoothing with an alpha of .75
1704	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1705	* rfc793's wired-in beta.
1706	*/
1707	if (delta < 0)
1708	delta = -delta;
1709	delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1710	if ((tp->t_rttvar += delta) <= 0)
1711	tp->t_rttvar = 1;
1712	}
1713	else
1714	{
1715	/*
1716	* No rtt measurement yet - use the unsmoothed rtt.
1717	* Set the variance to half the rtt (so our first
1718	* retransmit happens at 3*rtt).
1719	*/
1720	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1721	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1722	}
1723	tp->t_rtt = 0;
1724	tp->t_rxtshift = 0;
1725
1726	/*
1727	* the retransmit should happen at rtt + 4 * rttvar.
1728	* Because of the way we do the smoothing, srtt and rttvar
1729	* will each average +1/2 tick of bias. When we compute
1730	* the retransmit timer, we want 1/2 tick of rounding and
1731	* 1 extra tick because of +-1/2 tick uncertainty in the
1732	* firing of the timer. The bias will give us exactly the
1733	* 1.5 tick we need. But, because the bias is
1734	* statistical, we have to test that we don't drop below
1735	* the minimum feasible timer (which is 2 ticks).
1736	*/
1737	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1738	(short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1739
1740	/*
1741	* We received an ack for a packet that wasn't retransmitted;
1742	* it is probably safe to discard any error indications we've
1743	* received recently. This isn't quite right, but close enough
1744	* for now (a route might have failed after we sent a segment,
1745	* and the return path might not be symmetrical).
1746	*/
1747	tp->t_softerror = 0;
1748	}
1749
1750	/*
1751	* Determine a reasonable value for maxseg size.
1752	* If the route is known, check route for mtu.
1753	* If none, use an mss that can be handled on the outgoing
1754	* interface without forcing IP to fragment; if bigger than
1755	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1756	* to utilize large mbufs. If no route is found, route has no mtu,
1757	* or the destination isn't local, use a default, hopefully conservative
1758	* size (usually 512 or the default IP max size, but no more than the mtu
1759	* of the interface), as we can't discover anything about intervening
1760	* gateways or networks. We also initialize the congestion/slow start
1761	* window to be a single segment if the destination isn't local.
1762	* While looking at the routing entry, we also initialize other path-dependent
1763	* parameters from pre-set or cached values in the routing entry.
1764	*/
1765
1766	int
1767	tcp_mss(PNATState pData, register struct tcpcb *tp, u_int offer)
1768	{
1769	struct socket *so = tp->t_socket;
1770	int mss;
1771
1772	DEBUG_CALL("tcp_mss");
1773	DEBUG_ARG("tp = %lx", (long)tp);
1774	DEBUG_ARG("offer = %d", offer);
1775
1776	mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
1777	if (offer)
1778	mss = min(mss, offer);
1779	mss = max(mss, 32);
1780	if (mss < tp->t_maxseg \|\| offer != 0)
1781	tp->t_maxseg = mss;
1782
1783	tp->snd_cwnd = mss;
1784
1785	sbreserve(&so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
1786	sbreserve(&so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
1787
1788	DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1789
1790	return mss;
1791	}

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 15522

Download in other formats: