tcp_input.c@ 20453

Last change on this file since 20453 was 20453, checked in by vboxsync, 15 years ago
NAT: adds performance counters to sbuf and tcp_input
Property svn:eol-style set to `native`
File size: 61.3 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	SLIRP_PROFILE_START(TCP_reassamble, tcp_reassamble);
82
83	/*
84	* XXX: tcp_reass() is rather inefficient with its data structures
85	* and should be rewritten (see NetBSD for optimizations). While
86	* doing that it should move to its own file tcp_reass.c.
87	*/
88
89	/*
90	* Call with th==NULL after become established to
91	* force pre-ESTABLISHED data up to user socket.
92	*/
93	if (th == NULL)
94	goto present;
95
96	/*
97	* Limit the number of segments in the reassembly queue to prevent
98	* holding on to too many segments (and thus running out of mbufs).
99	* Make sure to let the missing segment through which caused this
100	* queue. Always keep one global queue entry spare to be able to
101	* process the missing segment.
102	*/
103	if ( th->th_seq != tp->rcv_nxt
104	&& ( tcp_reass_qsize + 1 >= tcp_reass_maxseg
105	\|\| tp->t_segqlen >= tcp_reass_maxqlen))
106	{
107	tcp_reass_overflows++;
108	tcpstat.tcps_rcvmemdrop++;
109	m_freem(pData, m);
110	*tlenp = 0;
111	SLIRP_PROFILE_STOP(TCP_reassamble, tcp_reassamble);
112	return (0);
113	}
114
115	/*
116	* Allocate a new queue entry. If we can't, or hit the zone limit
117	* just drop the pkt.
118	*/
119	te = RTMemAlloc(sizeof(struct tseg_qent));
120	if (te == NULL)
121	{
122	tcpstat.tcps_rcvmemdrop++;
123	m_freem(pData, m);
124	*tlenp = 0;
125	SLIRP_PROFILE_STOP(TCP_reassamble, tcp_reassamble);
126	return (0);
127	}
128	tp->t_segqlen++;
129	tcp_reass_qsize++;
130
131	/*
132	* Find a segment which begins after this one does.
133	*/
134	LIST_FOREACH(q, &tp->t_segq, tqe_q)
135	{
136	if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
137	break;
138	p = q;
139	}
140
141	/*
142	* If there is a preceding segment, it may provide some of
143	* our data already. If so, drop the data from the incoming
144	* segment. If it provides all of our data, drop us.
145	*/
146	if (p != NULL)
147	{
148	int i;
149	/* conversion to int (in i) handles seq wraparound */
150	i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
151	if (i > 0)
152	{
153	if (i >= *tlenp)
154	{
155	tcpstat.tcps_rcvduppack++;
156	tcpstat.tcps_rcvdupbyte += *tlenp;
157	m_freem(pData, m);
158	RTMemFree(te);
159	tp->t_segqlen--;
160	tcp_reass_qsize--;
161	/*
162	* Try to present any queued data
163	* at the left window edge to the user.
164	* This is needed after the 3-WHS
165	* completes.
166	*/
167	goto present; /* ??? */
168	}
169	m_adj(m, i);
170	*tlenp -= i;
171	th->th_seq += i;
172	}
173	}
174	tcpstat.tcps_rcvoopack++;
175	tcpstat.tcps_rcvoobyte += *tlenp;
176
177	/*
178	* While we overlap succeeding segments trim them or,
179	* if they are completely covered, dequeue them.
180	*/
181	while (q)
182	{
183	int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
184	if (i <= 0)
185	break;
186	if (i < q->tqe_len)
187	{
188	q->tqe_th->th_seq += i;
189	q->tqe_len -= i;
190	m_adj(q->tqe_m, i);
191	break;
192	}
193
194	nq = LIST_NEXT(q, tqe_q);
195	LIST_REMOVE(q, tqe_q);
196	m_freem(pData, q->tqe_m);
197	RTMemFree(q);
198	tp->t_segqlen--;
199	tcp_reass_qsize--;
200	q = nq;
201	}
202
203	/* Insert the new segment queue entry into place. */
204	te->tqe_m = m;
205	te->tqe_th = th;
206	te->tqe_len = *tlenp;
207
208	if (p == NULL)
209	{
210	LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
211	}
212	else
213	{
214	LIST_INSERT_AFTER(p, te, tqe_q);
215	}
216
217	present:
218	/*
219	* Present data to user, advancing rcv_nxt through
220	* completed sequence space.
221	*/
222	if (!TCPS_HAVEESTABLISHED(tp->t_state))
223	{
224	SLIRP_PROFILE_STOP(TCP_reassamble, tcp_reassamble);
225	return (0);
226	}
227	q = LIST_FIRST(&tp->t_segq);
228	if (!q \|\| q->tqe_th->th_seq != tp->rcv_nxt)
229	{
230	SLIRP_PROFILE_STOP(TCP_reassamble, tcp_reassamble);
231	return (0);
232	}
233	do
234	{
235	tp->rcv_nxt += q->tqe_len;
236	flags = q->tqe_th->th_flags & TH_FIN;
237	nq = LIST_NEXT(q, tqe_q);
238	LIST_REMOVE(q, tqe_q);
239	/* XXX: This place should be checked for the same code in
240	* original BSD code for Slirp and current BSD used SS_FCANTRCVMORE
241	*/
242	if (so->so_state & SS_FCANTSENDMORE)
243	m_freem(pData, q->tqe_m);
244	else
245	{
246	if (so->so_emu)
247	{
248	if (tcp_emu(pData, so, q->tqe_m))
249	sbappend(pData, so, q->tqe_m);
250	}
251	else
252	sbappend(pData, so, q->tqe_m);
253	}
254	RTMemFree(q);
255	tp->t_segqlen--;
256	tcp_reass_qsize--;
257	q = nq;
258	}
259	while (q && q->tqe_th->th_seq == tp->rcv_nxt);
260
261	SLIRP_PROFILE_STOP(TCP_reassamble, tcp_reassamble);
262	return flags;
263	}
264
265	/*
266	* TCP input routine, follows pages 65-76 of the
267	* protocol specification dated September, 1981 very closely.
268	*/
269	void
270	tcp_input(PNATState pData, register struct mbuf m, int iphlen, struct socket inso)
271	{
272	struct ip save_ip, *ip;
273	register struct tcpiphdr *ti;
274	caddr_t optp = NULL;
275	int optlen = 0;
276	int len, tlen, off;
277	register struct tcpcb *tp = 0;
278	register int tiflags;
279	struct socket *so = 0;
280	int todrop, acked, ourfinisacked, needoutput = 0;
281	/* int dropsocket = 0; */
282	int iss = 0;
283	u_long tiwin;
284	/* int ts_present = 0; */
285	SLIRP_PROFILE_START(TCP_input, counter_input);
286
287	DEBUG_CALL("tcp_input");
288	DEBUG_ARGS((dfd," m = %8lx iphlen = %2d inso = %lx\n",
289	(long )m, iphlen, (long )inso ));
290
291	if (inso != NULL)
292	{
293	QSOCKET_LOCK(tcb);
294	SOCKET_LOCK(inso);
295	QSOCKET_UNLOCK(tcb);
296	}
297	/*
298	* If called with m == 0, then we're continuing the connect
299	*/
300	if (m == NULL)
301	{
302	so = inso;
303	Log4(("NAT: tcp_input: %R[natsock]\n", so));
304	/* Re-set a few variables */
305	tp = sototcpcb(so);
306	m = so->so_m;
307
308	so->so_m = 0;
309	ti = so->so_ti;
310	/* @todo (r -vvl) clarify why it might happens */
311	if (ti == NULL)
312	{
313	LogRel(("NAT: ti is null. can't do any reseting connection actions\n"));
314	/* mbuf should be cleared in sofree called from tcp_close */
315	tcp_close(pData, tp);
316	SLIRP_PROFILE_STOP(TCP_input, counter_input);
317	return;
318	}
319	tiwin = ti->ti_win;
320	tiflags = ti->ti_flags;
321
322	goto cont_conn;
323	}
324
325	tcpstat.tcps_rcvtotal++;
326	/*
327	* Get IP and TCP header together in first mbuf.
328	* Note: IP leaves IP header in first mbuf.
329	*/
330	ti = mtod(m, struct tcpiphdr *);
331	if (iphlen > sizeof(struct ip ))
332	{
333	ip_stripoptions(m, (struct mbuf *)0);
334	iphlen = sizeof(struct ip );
335	}
336	/* XXX Check if too short */
337
338
339	/*
340	* Save a copy of the IP header in case we want restore it
341	* for sending an ICMP error message in response.
342	*/
343	ip = mtod(m, struct ip *);
344	save_ip = *ip;
345	save_ip.ip_len+= iphlen;
346
347	/*
348	* Checksum extended TCP header and data.
349	*/
350	tlen = ((struct ip *)ti)->ip_len;
351	memset(ti->ti_x1, 0, 9);
352	ti->ti_len = htons((u_int16_t)tlen);
353	len = sizeof(struct ip ) + tlen;
354	/* keep checksum for ICMP reply
355	* ti->ti_sum = cksum(m, len);
356	* if (ti->ti_sum) { */
357	if (cksum(m, len))
358	{
359	tcpstat.tcps_rcvbadsum++;
360	goto drop;
361	}
362
363	/*
364	* Check that TCP offset makes sense,
365	* pull out TCP options and adjust length. XXX
366	*/
367	off = ti->ti_off << 2;
368	if ( off < sizeof (struct tcphdr)
369	\|\| off > tlen)
370	{
371	tcpstat.tcps_rcvbadoff++;
372	goto drop;
373	}
374	tlen -= off;
375	ti->ti_len = tlen;
376	if (off > sizeof (struct tcphdr))
377	{
378	optlen = off - sizeof (struct tcphdr);
379	optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
380
381	/*
382	* Do quick retrieval of timestamp options ("options
383	* prediction?"). If timestamp is the only option and it's
384	* formatted as recommended in RFC 1323 appendix A, we
385	* quickly get the values now and not bother calling
386	* tcp_dooptions(), etc.
387	*/
388	#if 0
389	if (( optlen == TCPOLEN_TSTAMP_APPA
390	\|\| ( optlen > TCPOLEN_TSTAMP_APPA
391	&& optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
392	(u_int32_t )optp == htonl(TCPOPT_TSTAMP_HDR) &&
393	(ti->ti_flags & TH_SYN) == 0)
394	{
395	ts_present = 1;
396	ts_val = ntohl((u_int32_t )(optp + 4));
397	ts_ecr = ntohl((u_int32_t )(optp + 8));
398	optp = NULL; / * we have parsed the options * /
399	}
400	#endif
401	}
402	tiflags = ti->ti_flags;
403
404	/*
405	* Convert TCP protocol specific fields to host format.
406	*/
407	NTOHL(ti->ti_seq);
408	NTOHL(ti->ti_ack);
409	NTOHS(ti->ti_win);
410	NTOHS(ti->ti_urp);
411
412	/*
413	* Drop TCP, IP headers and TCP options.
414	*/
415	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
416	m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
417
418	/*
419	* Locate pcb for segment.
420	*/
421	findso:
422	if (so != NULL && so != &tcb)
423	SOCKET_UNLOCK(so);
424	QSOCKET_LOCK(tcb);
425	so = tcp_last_so;
426	if ( so->so_fport != ti->ti_dport
427	\|\| so->so_lport != ti->ti_sport
428	\|\| so->so_laddr.s_addr != ti->ti_src.s_addr
429	\|\| so->so_faddr.s_addr != ti->ti_dst.s_addr)
430	{
431	struct socket *sonxt;
432	QSOCKET_UNLOCK(tcb);
433	/* @todo fix SOLOOKUP macrodefinition to be usable here */
434	#ifndef VBOX_WITH_SLIRP_MT
435	so = solookup(&tcb, ti->ti_src, ti->ti_sport,
436	ti->ti_dst, ti->ti_dport);
437	#else
438	so = NULL;
439	QSOCKET_FOREACH(so, sonxt, tcp)
440	/* { */
441	if ( so->so_lport == ti->ti_sport
442	&& so->so_laddr.s_addr == ti->ti_src.s_addr
443	&& so->so_faddr.s_addr == ti->ti_dst.s_addr
444	&& so->so_fport == ti->ti_dport
445	&& so->so_deleted != 1)
446	{
447	break; /* so is locked here */
448	}
449	LOOP_LABEL(tcp, so, sonxt);
450	}
451	if (so == &tcb) {
452	so = NULL;
453	}
454	#endif
455	if (so)
456	{
457	tcp_last_so = so;
458	}
459	++tcpstat.tcps_socachemiss;
460	}
461	else
462	{
463	SOCKET_LOCK(so);
464	QSOCKET_UNLOCK(tcb);
465	}
466
467	/*
468	* If the state is CLOSED (i.e., TCB does not exist) then
469	* all data in the incoming segment is discarded.
470	* If the TCB exists but is in CLOSED state, it is embryonic,
471	* but should either do a listen or a connect soon.
472	*
473	* state == CLOSED means we've done socreate() but haven't
474	* attached it to a protocol yet...
475	*
476	* XXX If a TCB does not exist, and the TH_SYN flag is
477	* the only flag set, then create a session, mark it
478	* as if it was LISTENING, and continue...
479	*/
480	if (so == 0)
481	{
482	if ((tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) != TH_SYN)
483	goto dropwithreset;
484
485	if ((so = socreate()) == NULL)
486	goto dropwithreset;
487	if (tcp_attach(pData, so) < 0)
488	{
489	RTMemFree(so); /* Not sofree (if it failed, it's not insqued) */
490	goto dropwithreset;
491	}
492	SOCKET_LOCK(so);
493	sbreserve(pData, &so->so_snd, tcp_sndspace);
494	sbreserve(pData, &so->so_rcv, tcp_rcvspace);
495
496	/* tcp_last_so = so; / / XXX ? */
497	/* tp = sototcpcb(so); */
498
499	so->so_laddr = ti->ti_src;
500	so->so_lport = ti->ti_sport;
501	so->so_faddr = ti->ti_dst;
502	so->so_fport = ti->ti_dport;
503
504	if ((so->so_iptos = tcp_tos(so)) == 0)
505	so->so_iptos = ((struct ip *)ti)->ip_tos;
506
507	tp = sototcpcb(so);
508	tp->t_state = TCPS_LISTEN;
509	}
510
511	/*
512	* If this is a still-connecting socket, this probably
513	* a retransmit of the SYN. Whether it's a retransmit SYN
514	* or something else, we nuke it.
515	*/
516	if (so->so_state & SS_ISFCONNECTING)
517	{
518	goto drop;
519	}
520
521	tp = sototcpcb(so);
522
523	/* XXX Should never fail */
524	if (tp == 0)
525	goto dropwithreset;
526	if (tp->t_state == TCPS_CLOSED)
527	{
528	goto drop;
529	}
530
531	/* Unscale the window into a 32-bit value. */
532	/* if ((tiflags & TH_SYN) == 0)
533	* tiwin = ti->ti_win << tp->snd_scale;
534	* else
535	*/
536	tiwin = ti->ti_win;
537
538	/*
539	* Segment received on connection.
540	* Reset idle time and keep-alive timer.
541	*/
542	tp->t_idle = 0;
543	if (so_options)
544	tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
545	else
546	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
547
548	/*
549	* Process options if not in LISTEN state,
550	* else do it below (after getting remote address).
551	*/
552	if (optp && tp->t_state != TCPS_LISTEN)
553	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
554	/* , */
555	/* &ts_present, &ts_val, &ts_ecr); */
556
557	/*
558	* Header prediction: check for the two common cases
559	* of a uni-directional data xfer. If the packet has
560	* no control flags, is in-sequence, the window didn't
561	* change and we're not retransmitting, it's a
562	* candidate. If the length is zero and the ack moved
563	* forward, we're the sender side of the xfer. Just
564	* free the data acked & wake any higher level process
565	* that was blocked waiting for space. If the length
566	* is non-zero and the ack didn't move, we're the
567	* receiver side. If we're getting packets in-order
568	* (the reassembly queue is empty), add the data to
569	* the socket buffer and note that we need a delayed ack.
570	*
571	* XXX Some of these tests are not needed
572	* eg: the tiwin == tp->snd_wnd prevents many more
573	* predictions.. with no real advantage..
574	*/
575	if ( tp->t_state == TCPS_ESTABLISHED
576	&& (tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK
577	/* && (!ts_present \|\| TSTMP_GEQ(ts_val, tp->ts_recent)) */
578	&& ti->ti_seq == tp->rcv_nxt
579	&& tiwin && tiwin == tp->snd_wnd
580	&& tp->snd_nxt == tp->snd_max)
581	{
582	/*
583	* If last ACK falls within this segment's sequence numbers,
584	* record the timestamp.
585	*/
586	#if 0
587	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
588	SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len))
589	{
590	tp->ts_recent_age = tcp_now;
591	tp->ts_recent = ts_val;
592	}
593	#endif
594
595	if (ti->ti_len == 0)
596	{
597	if ( SEQ_GT(ti->ti_ack, tp->snd_una)
598	&& SEQ_LEQ(ti->ti_ack, tp->snd_max)
599	&& tp->snd_cwnd >= tp->snd_wnd)
600	{
601	/*
602	* this is a pure ack for outstanding data.
603	*/
604	++tcpstat.tcps_predack;
605	#if 0
606	if (ts_present)
607	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
608	else
609	#endif
610	if ( tp->t_rtt
611	&& SEQ_GT(ti->ti_ack, tp->t_rtseq))
612	tcp_xmit_timer(pData, tp, tp->t_rtt);
613	acked = ti->ti_ack - tp->snd_una;
614	tcpstat.tcps_rcvackpack++;
615	tcpstat.tcps_rcvackbyte += acked;
616	sbdrop(&so->so_snd, acked);
617	tp->snd_una = ti->ti_ack;
618	m_freem(pData, m);
619
620	/*
621	* If all outstanding data are acked, stop
622	* retransmit timer, otherwise restart timer
623	* using current (possibly backed-off) value.
624	* If process is waiting for space,
625	* wakeup/selwakeup/signal. If data
626	* are ready to send, let tcp_output
627	* decide between more output or persist.
628	*/
629	if (tp->snd_una == tp->snd_max)
630	tp->t_timer[TCPT_REXMT] = 0;
631	else if (tp->t_timer[TCPT_PERSIST] == 0)
632	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
633
634	/*
635	* There's room in so_snd, sowwakup will read()
636	* from the socket if we can
637	*/
638	#if 0
639	if (so->so_snd.sb_flags & SB_NOTIFY)
640	sowwakeup(so);
641	#endif
642	/*
643	* This is called because sowwakeup might have
644	* put data into so_snd. Since we don't so sowwakeup,
645	* we don't need this.. XXX???
646	*/
647	if (so->so_snd.sb_cc)
648	(void) tcp_output(pData, tp);
649
650	SOCKET_UNLOCK(so);
651	SLIRP_PROFILE_STOP(TCP_input, counter_input);
652	return;
653	}
654	}
655	else if ( ti->ti_ack == tp->snd_una
656	&& LIST_FIRST(&tp->t_segq)
657	&& ti->ti_len <= sbspace(&so->so_rcv))
658	{
659	/*
660	* this is a pure, in-sequence data packet
661	* with nothing on the reassembly queue and
662	* we have enough buffer space to take it.
663	*/
664	++tcpstat.tcps_preddat;
665	tp->rcv_nxt += ti->ti_len;
666	tcpstat.tcps_rcvpack++;
667	tcpstat.tcps_rcvbyte += ti->ti_len;
668	/*
669	* Add data to socket buffer.
670	*/
671	if (so->so_emu)
672	{
673	if (tcp_emu(pData, so, m))
674	sbappend(pData, so, m);
675	}
676	else
677	sbappend(pData, so, m);
678
679	/*
680	* XXX This is called when data arrives. Later, check
681	* if we can actually write() to the socket
682	* XXX Need to check? It's be NON_BLOCKING
683	*/
684	/* sorwakeup(so); */
685
686	/*
687	* If this is a short packet, then ACK now - with Nagel
688	* congestion avoidance sender won't send more until
689	* he gets an ACK.
690	*
691	* It is better to not delay acks at all to maximize
692	* TCP throughput. See RFC 2581.
693	*/
694	tp->t_flags \|= TF_ACKNOW;
695	tcp_output(pData, tp);
696	SOCKET_UNLOCK(so);
697	SLIRP_PROFILE_STOP(TCP_input, counter_input);
698	return;
699	}
700	} /* header prediction */
701	/*
702	* Calculate amount of space in receive window,
703	* and then do TCP input processing.
704	* Receive window is amount of space in rcv queue,
705	* but not less than advertised window.
706	*/
707	{
708	int win;
709	win = sbspace(&so->so_rcv);
710	if (win < 0)
711	win = 0;
712	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
713	}
714
715	switch (tp->t_state)
716	{
717	/*
718	* If the state is LISTEN then ignore segment if it contains an RST.
719	* If the segment contains an ACK then it is bad and send a RST.
720	* If it does not contain a SYN then it is not interesting; drop it.
721	* Don't bother responding if the destination was a broadcast.
722	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
723	* tp->iss, and send a segment:
724	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
725	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
726	* Fill in remote peer address fields if not previously specified.
727	* Enter SYN_RECEIVED state, and process any other fields of this
728	* segment in this state.
729	*/
730	case TCPS_LISTEN:
731	{
732	if (tiflags & TH_RST) {
733	goto drop;
734	}
735	if (tiflags & TH_ACK)
736	goto dropwithreset;
737	if ((tiflags & TH_SYN) == 0)
738	{
739	goto drop;
740	}
741
742	/*
743	* This has way too many gotos...
744	* But a bit of spaghetti code never hurt anybody :)
745	*/
746
747	if (so->so_emu & EMU_NOCONNECT)
748	{
749	so->so_emu &= ~EMU_NOCONNECT;
750	goto cont_input;
751	}
752
753	if ( (tcp_fconnect(pData, so) == -1)
754	&& errno != EINPROGRESS
755	&& errno != EWOULDBLOCK)
756	{
757	u_char code = ICMP_UNREACH_NET;
758	DEBUG_MISC((dfd," tcp fconnect errno = %d-%s\n",
759	errno, strerror(errno)));
760	if (errno == ECONNREFUSED)
761	{
762	/* ACK the SYN, send RST to refuse the connection */
763	tcp_respond(pData, tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
764	TH_RST\|TH_ACK);
765	}
766	else
767	{
768	if (errno == EHOSTUNREACH)
769	code = ICMP_UNREACH_HOST;
770	HTONL(ti->ti_seq); /* restore tcp header */
771	HTONL(ti->ti_ack);
772	HTONS(ti->ti_win);
773	HTONS(ti->ti_urp);
774	m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
775	m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
776	*ip = save_ip;
777	icmp_error(pData, m, ICMP_UNREACH, code, 0, strerror(errno));
778	tp->t_socket->so_m = NULL;
779	}
780	tp = tcp_close(pData, tp);
781	m_free(pData, m);
782	}
783	else
784	{
785	/*
786	* Haven't connected yet, save the current mbuf
787	* and ti, and return
788	* XXX Some OS's don't tell us whether the connect()
789	* succeeded or not. So we must time it out.
790	*/
791	so->so_m = m;
792	so->so_ti = ti;
793	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
794	tp->t_state = TCPS_SYN_RECEIVED;
795	}
796	SOCKET_UNLOCK(so);
797	SLIRP_PROFILE_STOP(TCP_input, counter_input);
798	return;
799
800	cont_conn:
801	/* m==NULL
802	* Check if the connect succeeded
803	*/
804	if (so->so_state & SS_NOFDREF)
805	{
806	tp = tcp_close(pData, tp);
807	goto dropwithreset;
808	}
809	cont_input:
810	tcp_template(tp);
811
812	if (optp)
813	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
814
815	if (iss)
816	tp->iss = iss;
817	else
818	tp->iss = tcp_iss;
819	tcp_iss += TCP_ISSINCR/2;
820	tp->irs = ti->ti_seq;
821	tcp_sendseqinit(tp);
822	tcp_rcvseqinit(tp);
823	tp->t_flags \|= TF_ACKNOW;
824	tp->t_state = TCPS_SYN_RECEIVED;
825	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
826	tcpstat.tcps_accepts++;
827	goto trimthenstep6;
828	} /* case TCPS_LISTEN */
829
830	/*
831	* If the state is SYN_SENT:
832	* if seg contains an ACK, but not for our SYN, drop the input.
833	* if seg contains a RST, then drop the connection.
834	* if seg does not contain SYN, then drop it.
835	* Otherwise this is an acceptable SYN segment
836	* initialize tp->rcv_nxt and tp->irs
837	* if seg contains ack then advance tp->snd_una
838	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
839	* arrange for segment to be acked (eventually)
840	* continue processing rest of data/controls, beginning with URG
841	*/
842	case TCPS_SYN_SENT:
843	if ( (tiflags & TH_ACK)
844	&& ( SEQ_LEQ(ti->ti_ack, tp->iss)
845	\|\| SEQ_GT(ti->ti_ack, tp->snd_max)))
846	goto dropwithreset;
847
848	if (tiflags & TH_RST)
849	{
850	if (tiflags & TH_ACK)
851	tp = tcp_drop(pData, tp, 0); /* XXX Check t_softerror! */
852	goto drop;
853	}
854
855	if ((tiflags & TH_SYN) == 0)
856	{
857	goto drop;
858	}
859	if (tiflags & TH_ACK)
860	{
861	tp->snd_una = ti->ti_ack;
862	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
863	tp->snd_nxt = tp->snd_una;
864	}
865
866	tp->t_timer[TCPT_REXMT] = 0;
867	tp->irs = ti->ti_seq;
868	tcp_rcvseqinit(tp);
869	tp->t_flags \|= TF_ACKNOW;
870	if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss))
871	{
872	tcpstat.tcps_connects++;
873	soisfconnected(so);
874	tp->t_state = TCPS_ESTABLISHED;
875
876	/* Do window scaling on this connection? */
877	#if 0
878	if (( tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
879	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
880	{
881	tp->snd_scale = tp->requested_s_scale;
882	tp->rcv_scale = tp->request_r_scale;
883	}
884	#endif
885	(void) tcp_reass(pData, tp, (struct tcphdr )0, NULL, (struct mbuf )0);
886	/*
887	* if we didn't have to retransmit the SYN,
888	* use its rtt as our initial srtt & rtt var.
889	*/
890	if (tp->t_rtt)
891	tcp_xmit_timer(pData, tp, tp->t_rtt);
892	}
893	else
894	tp->t_state = TCPS_SYN_RECEIVED;
895
896	trimthenstep6:
897	/*
898	* Advance ti->ti_seq to correspond to first data byte.
899	* If data, trim to stay within window,
900	* dropping FIN if necessary.
901	*/
902	ti->ti_seq++;
903	if (ti->ti_len > tp->rcv_wnd)
904	{
905	todrop = ti->ti_len - tp->rcv_wnd;
906	m_adj(m, -todrop);
907	ti->ti_len = tp->rcv_wnd;
908	tiflags &= ~TH_FIN;
909	tcpstat.tcps_rcvpackafterwin++;
910	tcpstat.tcps_rcvbyteafterwin += todrop;
911	}
912	tp->snd_wl1 = ti->ti_seq - 1;
913	tp->rcv_up = ti->ti_seq;
914	Log2(("hit6"));
915	goto step6;
916	} /* switch tp->t_state */
917	/*
918	* States other than LISTEN or SYN_SENT.
919	* First check timestamp, if present.
920	* Then check that at least some bytes of segment are within
921	* receive window. If segment begins before rcv_nxt,
922	* drop leading data (and SYN); if nothing left, just ack.
923	*
924	* RFC 1323 PAWS: If we have a timestamp reply on this segment
925	* and it's less than ts_recent, drop it.
926	*/
927	#if 0
928	if ( ts_present
929	&& (tiflags & TH_RST) == 0
930	&& tp->ts_recent
931	&& TSTMP_LT(ts_val, tp->ts_recent))
932	{
933	/* Check to see if ts_recent is over 24 days old. */
934	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE)
935	{
936	/*
937	* Invalidate ts_recent. If this segment updates
938	* ts_recent, the age will be reset later and ts_recent
939	* will get a valid value. If it does not, setting
940	* ts_recent to zero will at least satisfy the
941	* requirement that zero be placed in the timestamp
942	* echo reply when ts_recent isn't valid. The
943	* age isn't reset until we get a valid ts_recent
944	* because we don't want out-of-order segments to be
945	* dropped when ts_recent is old.
946	*/
947	tp->ts_recent = 0;
948	}
949	else
950	{
951	tcpstat.tcps_rcvduppack++;
952	tcpstat.tcps_rcvdupbyte += ti->ti_len;
953	tcpstat.tcps_pawsdrop++;
954	goto dropafterack;
955	}
956	}
957	#endif
958
959	todrop = tp->rcv_nxt - ti->ti_seq;
960	if (todrop > 0)
961	{
962	if (tiflags & TH_SYN)
963	{
964	tiflags &= ~TH_SYN;
965	ti->ti_seq++;
966	if (ti->ti_urp > 1)
967	ti->ti_urp--;
968	else
969	tiflags &= ~TH_URG;
970	todrop--;
971	}
972	/*
973	* Following if statement from Stevens, vol. 2, p. 960.
974	*/
975	if ( todrop > ti->ti_len
976	\|\| ( todrop == ti->ti_len
977	&& (tiflags & TH_FIN) == 0))
978	{
979	/*
980	* Any valid FIN must be to the left of the window.
981	* At this point the FIN must be a duplicate or out
982	* of sequence; drop it.
983	*/
984	tiflags &= ~TH_FIN;
985
986	/*
987	* Send an ACK to resynchronize and drop any data.
988	* But keep on processing for RST or ACK.
989	*/
990	tp->t_flags \|= TF_ACKNOW;
991	todrop = ti->ti_len;
992	tcpstat.tcps_rcvduppack++;
993	tcpstat.tcps_rcvdupbyte += todrop;
994	}
995	else
996	{
997	tcpstat.tcps_rcvpartduppack++;
998	tcpstat.tcps_rcvpartdupbyte += todrop;
999	}
1000	m_adj(m, todrop);
1001	ti->ti_seq += todrop;
1002	ti->ti_len -= todrop;
1003	if (ti->ti_urp > todrop)
1004	ti->ti_urp -= todrop;
1005	else
1006	{
1007	tiflags &= ~TH_URG;
1008	ti->ti_urp = 0;
1009	}
1010	}
1011	/*
1012	* If new data are received on a connection after the
1013	* user processes are gone, then RST the other end.
1014	*/
1015	if ( (so->so_state & SS_NOFDREF)
1016	&& tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len)
1017	{
1018	tp = tcp_close(pData, tp);
1019	tcpstat.tcps_rcvafterclose++;
1020	goto dropwithreset;
1021	}
1022
1023	/*
1024	* If segment ends after window, drop trailing data
1025	* (and PUSH and FIN); if nothing left, just ACK.
1026	*/
1027	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1028	if (todrop > 0)
1029	{
1030	tcpstat.tcps_rcvpackafterwin++;
1031	if (todrop >= ti->ti_len)
1032	{
1033	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1034	/*
1035	* If a new connection request is received
1036	* while in TIME_WAIT, drop the old connection
1037	* and start over if the sequence numbers
1038	* are above the previous ones.
1039	*/
1040	if ( tiflags & TH_SYN
1041	&& tp->t_state == TCPS_TIME_WAIT
1042	&& SEQ_GT(ti->ti_seq, tp->rcv_nxt))
1043	{
1044	iss = tp->rcv_nxt + TCP_ISSINCR;
1045	tp = tcp_close(pData, tp);
1046	SOCKET_UNLOCK(tp->t_socket);
1047	goto findso;
1048	}
1049	/*
1050	* If window is closed can only take segments at
1051	* window edge, and have to drop data and PUSH from
1052	* incoming segments. Continue processing, but
1053	* remember to ack. Otherwise, drop segment
1054	* and ack.
1055	*/
1056	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt)
1057	{
1058	tp->t_flags \|= TF_ACKNOW;
1059	tcpstat.tcps_rcvwinprobe++;
1060	}
1061	else
1062	goto dropafterack;
1063	}
1064	else
1065	tcpstat.tcps_rcvbyteafterwin += todrop;
1066	m_adj(m, -todrop);
1067	ti->ti_len -= todrop;
1068	tiflags &= ~(TH_PUSH\|TH_FIN);
1069	}
1070
1071	/*
1072	* If last ACK falls within this segment's sequence numbers,
1073	* record its timestamp.
1074	*/
1075	#if 0
1076	if ( ts_present
1077	&& SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)
1078	&& SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + ((tiflags & (TH_SYN\|TH_FIN)) != 0)))
1079	{
1080	tp->ts_recent_age = tcp_now;
1081	tp->ts_recent = ts_val;
1082	}
1083	#endif
1084
1085	/*
1086	* If the RST bit is set examine the state:
1087	* SYN_RECEIVED STATE:
1088	* If passive open, return to LISTEN state.
1089	* If active open, inform user that connection was refused.
1090	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1091	* Inform user that connection was reset, and close tcb.
1092	* CLOSING, LAST_ACK, TIME_WAIT STATES
1093	* Close the tcb.
1094	*/
1095	if (tiflags&TH_RST)
1096	switch (tp->t_state)
1097	{
1098	case TCPS_SYN_RECEIVED:
1099	/* so->so_error = ECONNREFUSED; */
1100	goto close;
1101
1102	case TCPS_ESTABLISHED:
1103	case TCPS_FIN_WAIT_1:
1104	case TCPS_FIN_WAIT_2:
1105	case TCPS_CLOSE_WAIT:
1106	/* so->so_error = ECONNRESET; */
1107	close:
1108	tp->t_state = TCPS_CLOSED;
1109	tcpstat.tcps_drops++;
1110	tp = tcp_close(pData, tp);
1111	goto drop;
1112
1113	case TCPS_CLOSING:
1114	case TCPS_LAST_ACK:
1115	case TCPS_TIME_WAIT:
1116	tp = tcp_close(pData, tp);
1117	goto drop;
1118	}
1119
1120	/*
1121	* If a SYN is in the window, then this is an
1122	* error and we send an RST and drop the connection.
1123	*/
1124	if (tiflags & TH_SYN)
1125	{
1126	tp = tcp_drop(pData, tp, 0);
1127	goto dropwithreset;
1128	}
1129
1130	/*
1131	* If the ACK bit is off we drop the segment and return.
1132	*/
1133	if ((tiflags & TH_ACK) == 0)
1134	{
1135	goto drop;
1136	}
1137
1138	/*
1139	* Ack processing.
1140	*/
1141	switch (tp->t_state)
1142	{
1143	/*
1144	* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1145	* ESTABLISHED state and continue processing, otherwise
1146	* send an RST. una<=ack<=max
1147	*/
1148	case TCPS_SYN_RECEIVED:
1149	if ( SEQ_GT(tp->snd_una, ti->ti_ack)
1150	\|\| SEQ_GT(ti->ti_ack, tp->snd_max))
1151	goto dropwithreset;
1152	tcpstat.tcps_connects++;
1153	tp->t_state = TCPS_ESTABLISHED;
1154	/*
1155	* The sent SYN is ack'ed with our sequence number +1
1156	* The first data byte already in the buffer will get
1157	* lost if no correction is made. This is only needed for
1158	* SS_CTL since the buffer is empty otherwise.
1159	* tp->snd_una++; or:
1160	*/
1161	tp->snd_una = ti->ti_ack;
1162	soisfconnected(so);
1163
1164	/* Do window scaling? */
1165	#if 0
1166	if ( (tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
1167	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
1168	{
1169	tp->snd_scale = tp->requested_s_scale;
1170	tp->rcv_scale = tp->request_r_scale;
1171	}
1172	#endif
1173	(void) tcp_reass(pData, tp, (struct tcphdr )0, (int )0, (struct mbuf *)0);
1174	tp->snd_wl1 = ti->ti_seq - 1;
1175	/* Avoid ack processing; snd_una==ti_ack => dup ack */
1176	goto synrx_to_est;
1177	/* fall into ... */
1178
1179	/*
1180	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1181	* ACKs. If the ack is in the range
1182	* tp->snd_una < ti->ti_ack <= tp->snd_max
1183	* then advance tp->snd_una to ti->ti_ack and drop
1184	* data from the retransmission queue. If this ACK reflects
1185	* more up to date window information we update our window information.
1186	*/
1187	case TCPS_ESTABLISHED:
1188	case TCPS_FIN_WAIT_1:
1189	case TCPS_FIN_WAIT_2:
1190	case TCPS_CLOSE_WAIT:
1191	case TCPS_CLOSING:
1192	case TCPS_LAST_ACK:
1193	case TCPS_TIME_WAIT:
1194	if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
1195	{
1196	if (ti->ti_len == 0 && tiwin == tp->snd_wnd)
1197	{
1198	tcpstat.tcps_rcvdupack++;
1199	DEBUG_MISC((dfd," dup ack m = %lx so = %lx \n",
1200	(long )m, (long )so));
1201	/*
1202	* If we have outstanding data (other than
1203	* a window probe), this is a completely
1204	* duplicate ack (ie, window info didn't
1205	* change), the ack is the biggest we've
1206	* seen and we've seen exactly our rexmt
1207	* threshold of them, assume a packet
1208	* has been dropped and retransmit it.
1209	* Kludge snd_nxt & the congestion
1210	* window so we send only this one
1211	* packet.
1212	*
1213	* We know we're losing at the current
1214	* window size so do congestion avoidance
1215	* (set ssthresh to half the current window
1216	* and pull our congestion window back to
1217	* the new ssthresh).
1218	*
1219	* Dup acks mean that packets have left the
1220	* network (they're now cached at the receiver)
1221	* so bump cwnd by the amount in the receiver
1222	* to keep a constant cwnd packets in the
1223	* network.
1224	*/
1225	if ( tp->t_timer[TCPT_REXMT] == 0
1226	\|\| ti->ti_ack != tp->snd_una)
1227	tp->t_dupacks = 0;
1228	else if (++tp->t_dupacks == tcprexmtthresh)
1229	{
1230	tcp_seq onxt = tp->snd_nxt;
1231	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
1232	if (win < 2)
1233	win = 2;
1234	tp->snd_ssthresh = win * tp->t_maxseg;
1235	tp->t_timer[TCPT_REXMT] = 0;
1236	tp->t_rtt = 0;
1237	tp->snd_nxt = ti->ti_ack;
1238	tp->snd_cwnd = tp->t_maxseg;
1239	(void) tcp_output(pData, tp);
1240	tp->snd_cwnd = tp->snd_ssthresh +
1241	tp->t_maxseg * tp->t_dupacks;
1242	if (SEQ_GT(onxt, tp->snd_nxt))
1243	tp->snd_nxt = onxt;
1244	goto drop;
1245	}
1246	else if (tp->t_dupacks > tcprexmtthresh)
1247	{
1248	tp->snd_cwnd += tp->t_maxseg;
1249	(void) tcp_output(pData, tp);
1250	goto drop;
1251	}
1252	}
1253	else
1254	tp->t_dupacks = 0;
1255	break;
1256	}
1257	synrx_to_est:
1258	/*
1259	* If the congestion window was inflated to account
1260	* for the other side's cached packets, retract it.
1261	*/
1262	if ( tp->t_dupacks > tcprexmtthresh
1263	&& tp->snd_cwnd > tp->snd_ssthresh)
1264	tp->snd_cwnd = tp->snd_ssthresh;
1265	tp->t_dupacks = 0;
1266	if (SEQ_GT(ti->ti_ack, tp->snd_max))
1267	{
1268	tcpstat.tcps_rcvacktoomuch++;
1269	goto dropafterack;
1270	}
1271	acked = ti->ti_ack - tp->snd_una;
1272	tcpstat.tcps_rcvackpack++;
1273	tcpstat.tcps_rcvackbyte += acked;
1274
1275	/*
1276	* If we have a timestamp reply, update smoothed
1277	* round trip time. If no timestamp is present but
1278	* transmit timer is running and timed sequence
1279	* number was acked, update smoothed round trip time.
1280	* Since we now have an rtt measurement, cancel the
1281	* timer backoff (cf., Phil Karn's retransmit alg.).
1282	* Recompute the initial retransmit timer.
1283	*/
1284	#if 0
1285	if (ts_present)
1286	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1287	else
1288	#endif
1289	if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1290	tcp_xmit_timer(pData, tp, tp->t_rtt);
1291
1292	/*
1293	* If all outstanding data is acked, stop retransmit
1294	* timer and remember to restart (more output or persist).
1295	* If there is more data to be acked, restart retransmit
1296	* timer, using current (possibly backed-off) value.
1297	*/
1298	if (ti->ti_ack == tp->snd_max)
1299	{
1300	tp->t_timer[TCPT_REXMT] = 0;
1301	needoutput = 1;
1302	}
1303	else if (tp->t_timer[TCPT_PERSIST] == 0)
1304	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1305	/*
1306	* When new data is acked, open the congestion window.
1307	* If the window gives us less than ssthresh packets
1308	* in flight, open exponentially (maxseg per packet).
1309	* Otherwise open linearly: maxseg per window
1310	* (maxseg^2 / cwnd per packet).
1311	*/
1312	{
1313	register u_int cw = tp->snd_cwnd;
1314	register u_int incr = tp->t_maxseg;
1315
1316	if (cw > tp->snd_ssthresh)
1317	incr = incr * incr / cw;
1318	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1319	}
1320	if (acked > so->so_snd.sb_cc)
1321	{
1322	tp->snd_wnd -= so->so_snd.sb_cc;
1323	sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1324	ourfinisacked = 1;
1325	}
1326	else
1327	{
1328	sbdrop(&so->so_snd, acked);
1329	tp->snd_wnd -= acked;
1330	ourfinisacked = 0;
1331	}
1332	/*
1333	* XXX sowwakup is called when data is acked and there's room for
1334	* for more data... it should read() the socket
1335	*/
1336	#if 0
1337	if (so->so_snd.sb_flags & SB_NOTIFY)
1338	sowwakeup(so);
1339	#endif
1340	tp->snd_una = ti->ti_ack;
1341	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1342	tp->snd_nxt = tp->snd_una;
1343
1344	switch (tp->t_state)
1345	{
1346	/*
1347	* In FIN_WAIT_1 STATE in addition to the processing
1348	* for the ESTABLISHED state if our FIN is now acknowledged
1349	* then enter FIN_WAIT_2.
1350	*/
1351	case TCPS_FIN_WAIT_1:
1352	if (ourfinisacked)
1353	{
1354	/*
1355	* If we can't receive any more
1356	* data, then closing user can proceed.
1357	* Starting the timer is contrary to the
1358	* specification, but if we don't get a FIN
1359	* we'll hang forever.
1360	*/
1361	if (so->so_state & SS_FCANTRCVMORE)
1362	{
1363	soisfdisconnected(so);
1364	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1365	}
1366	tp->t_state = TCPS_FIN_WAIT_2;
1367	}
1368	break;
1369
1370	/*
1371	* In CLOSING STATE in addition to the processing for
1372	* the ESTABLISHED state if the ACK acknowledges our FIN
1373	* then enter the TIME-WAIT state, otherwise ignore
1374	* the segment.
1375	*/
1376	case TCPS_CLOSING:
1377	if (ourfinisacked)
1378	{
1379	tp->t_state = TCPS_TIME_WAIT;
1380	tcp_canceltimers(tp);
1381	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1382	soisfdisconnected(so);
1383	}
1384	break;
1385
1386	/*
1387	* In LAST_ACK, we may still be waiting for data to drain
1388	* and/or to be acked, as well as for the ack of our FIN.
1389	* If our FIN is now acknowledged, delete the TCB,
1390	* enter the closed state and return.
1391	*/
1392	case TCPS_LAST_ACK:
1393	if (ourfinisacked)
1394	{
1395	tp = tcp_close(pData, tp);
1396	goto drop;
1397	}
1398	break;
1399
1400	/*
1401	* In TIME_WAIT state the only thing that should arrive
1402	* is a retransmission of the remote FIN. Acknowledge
1403	* it and restart the finack timer.
1404	*/
1405	case TCPS_TIME_WAIT:
1406	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1407	goto dropafterack;
1408	}
1409	} /* switch(tp->t_state) */
1410
1411	step6:
1412	/*
1413	* Update window information.
1414	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1415	*/
1416	if ( (tiflags & TH_ACK)
1417	&& ( SEQ_LT(tp->snd_wl1, ti->ti_seq)
1418	\|\| ( tp->snd_wl1 == ti->ti_seq
1419	&& ( SEQ_LT(tp->snd_wl2, ti->ti_ack)
1420	\|\| ( tp->snd_wl2 == ti->ti_ack
1421	&& tiwin > tp->snd_wnd)))))
1422	{
1423	/* keep track of pure window updates */
1424	if ( ti->ti_len == 0
1425	&& tp->snd_wl2 == ti->ti_ack
1426	&& tiwin > tp->snd_wnd)
1427	tcpstat.tcps_rcvwinupd++;
1428	tp->snd_wnd = tiwin;
1429	tp->snd_wl1 = ti->ti_seq;
1430	tp->snd_wl2 = ti->ti_ack;
1431	if (tp->snd_wnd > tp->max_sndwnd)
1432	tp->max_sndwnd = tp->snd_wnd;
1433	needoutput = 1;
1434	}
1435
1436	/*
1437	* Process segments with URG.
1438	*/
1439	if ((tiflags & TH_URG) && ti->ti_urp &&
1440	TCPS_HAVERCVDFIN(tp->t_state) == 0)
1441	{
1442	/*
1443	* This is a kludge, but if we receive and accept
1444	* random urgent pointers, we'll crash in
1445	* soreceive. It's hard to imagine someone
1446	* actually wanting to send this much urgent data.
1447	*/
1448	if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen)
1449	{
1450	ti->ti_urp = 0;
1451	tiflags &= ~TH_URG;
1452	goto dodata;
1453	}
1454	/*
1455	* If this segment advances the known urgent pointer,
1456	* then mark the data stream. This should not happen
1457	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1458	* a FIN has been received from the remote side.
1459	* In these states we ignore the URG.
1460	*
1461	* According to RFC961 (Assigned Protocols),
1462	* the urgent pointer points to the last octet
1463	* of urgent data. We continue, however,
1464	* to consider it to indicate the first octet
1465	* of data past the urgent section as the original
1466	* spec states (in one of two places).
1467	*/
1468	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up))
1469	{
1470	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1471	so->so_urgc = so->so_rcv.sb_cc +
1472	(tp->rcv_up - tp->rcv_nxt); /* -1; */
1473	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1474	}
1475	}
1476	else
1477	/*
1478	* If no out of band data is expected,
1479	* pull receive urgent pointer along
1480	* with the receive window.
1481	*/
1482	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1483	tp->rcv_up = tp->rcv_nxt;
1484	dodata:
1485
1486	/*
1487	* If this is a small packet, then ACK now - with Nagel
1488	* congestion avoidance sender won't send more until
1489	* he gets an ACK.
1490	*
1491	* See above.
1492	*/
1493	if ( ti->ti_len
1494	&& (unsigned)ti->ti_len <= 5
1495	&& ((struct tcpiphdr_2 *)ti)->first_char == (char)27)
1496	{
1497	tp->t_flags \|= TF_ACKNOW;
1498	}
1499
1500	/*
1501	* Process the segment text, merging it into the TCP sequencing queue,
1502	* and arranging for acknowledgment of receipt if necessary.
1503	* This process logically involves adjusting tp->rcv_wnd as data
1504	* is presented to the user (this happens in tcp_usrreq.c,
1505	* case PRU_RCVD). If a FIN has already been received on this
1506	* connection then we just ignore the text.
1507	*/
1508	if ( (ti->ti_len \|\| (tiflags&TH_FIN))
1509	&& TCPS_HAVERCVDFIN(tp->t_state) == 0)
1510	{
1511	if ( ti->ti_seq == tp->rcv_nxt
1512	&& LIST_EMPTY(&tp->t_segq)
1513	&& tp->t_state == TCPS_ESTABLISHED)
1514	{
1515	DELAY_ACK(tp, ti); /* little bit different from BSD declaration see netinet/tcp_input.c */
1516	tp->rcv_nxt += tlen;
1517	tiflags = ti->ti_t.th_flags & TH_FIN;
1518	tcpstat.tcps_rcvpack++;
1519	tcpstat.tcps_rcvbyte += tlen;
1520	if (so->so_state & SS_FCANTRCVMORE)
1521	m_freem(pData, m);
1522	else
1523	{
1524	if (so->so_emu)
1525	{
1526	if (tcp_emu(pData, so, m))
1527	sbappend(pData, so, m);
1528	}
1529	else
1530	sbappend(pData, so, m);
1531	}
1532	}
1533	else
1534	{
1535	tiflags = tcp_reass(pData, tp, &ti->ti_t, &tlen, m);
1536	tiflags \|= TF_ACKNOW;
1537	}
1538	/*
1539	* Note the amount of data that peer has sent into
1540	* our window, in order to estimate the sender's
1541	* buffer size.
1542	*/
1543	len = so->so_rcv.sb_datalen - (tp->rcv_adv - tp->rcv_nxt);
1544	}
1545	else
1546	{
1547	m_free(pData, m);
1548	tiflags &= ~TH_FIN;
1549	}
1550
1551	/*
1552	* If FIN is received ACK the FIN and let the user know
1553	* that the connection is closing.
1554	*/
1555	if (tiflags & TH_FIN)
1556	{
1557	if (TCPS_HAVERCVDFIN(tp->t_state) == 0)
1558	{
1559	/*
1560	* If we receive a FIN we can't send more data,
1561	* set it SS_FDRAIN
1562	* Shutdown the socket if there is no rx data in the
1563	* buffer.
1564	* soread() is called on completion of shutdown() and
1565	* will got to TCPS_LAST_ACK, and use tcp_output()
1566	* to send the FIN.
1567	*/
1568	/* sofcantrcvmore(so); */
1569	sofwdrain(so);
1570
1571	tp->t_flags \|= TF_ACKNOW;
1572	tp->rcv_nxt++;
1573	}
1574	switch (tp->t_state)
1575	{
1576	/*
1577	* In SYN_RECEIVED and ESTABLISHED STATES
1578	* enter the CLOSE_WAIT state.
1579	*/
1580	case TCPS_SYN_RECEIVED:
1581	case TCPS_ESTABLISHED:
1582	if(so->so_emu == EMU_CTL) /* no shutdown on socket */
1583	tp->t_state = TCPS_LAST_ACK;
1584	else
1585	tp->t_state = TCPS_CLOSE_WAIT;
1586	break;
1587
1588	/*
1589	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1590	* enter the CLOSING state.
1591	*/
1592	case TCPS_FIN_WAIT_1:
1593	tp->t_state = TCPS_CLOSING;
1594	break;
1595
1596	/*
1597	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1598	* starting the time-wait timer, turning off the other
1599	* standard timers.
1600	*/
1601	case TCPS_FIN_WAIT_2:
1602	tp->t_state = TCPS_TIME_WAIT;
1603	tcp_canceltimers(tp);
1604	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1605	soisfdisconnected(so);
1606	break;
1607
1608	/*
1609	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1610	*/
1611	case TCPS_TIME_WAIT:
1612	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1613	break;
1614	}
1615	}
1616
1617	/*
1618	* Return any desired output.
1619	*/
1620	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1621	tcp_output(pData, tp);
1622
1623	SOCKET_UNLOCK(so);
1624	SLIRP_PROFILE_STOP(TCP_input, counter_input);
1625	return;
1626
1627	dropafterack:
1628	Log2(("drop after ack\n"));
1629	/*
1630	* Generate an ACK dropping incoming segment if it occupies
1631	* sequence space, where the ACK reflects our state.
1632	*/
1633	if (tiflags & TH_RST)
1634	goto drop;
1635	m_freem(pData, m);
1636	tp->t_flags \|= TF_ACKNOW;
1637	(void) tcp_output(pData, tp);
1638	SOCKET_UNLOCK(so);
1639	SLIRP_PROFILE_STOP(TCP_input, counter_input);
1640	return;
1641
1642	dropwithreset:
1643	/* reuses m if m!=NULL, m_free() unnecessary */
1644	if (tiflags & TH_ACK)
1645	tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1646	else
1647	{
1648	if (tiflags & TH_SYN) ti->ti_len++;
1649	tcp_respond(pData, tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1650	TH_RST\|TH_ACK);
1651	}
1652
1653	if (so != &tcb)
1654	SOCKET_UNLOCK(so);
1655	SLIRP_PROFILE_STOP(TCP_input, counter_input);
1656	return;
1657
1658	drop:
1659	/*
1660	* Drop space held by incoming segment and return.
1661	*/
1662	m_free(pData, m);
1663
1664	#ifdef VBOX_WITH_SLIRP_MT
1665	if (RTCritSectIsOwned(&so->so_mutex))
1666	{
1667	SOCKET_UNLOCK(so);
1668	}
1669	#endif
1670
1671	SLIRP_PROFILE_STOP(TCP_input, counter_input);
1672	return;
1673	}
1674
1675	void
1676	tcp_dooptions(PNATState pData, struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti)
1677	{
1678	u_int16_t mss;
1679	int opt, optlen;
1680
1681	DEBUG_CALL("tcp_dooptions");
1682	DEBUG_ARGS((dfd," tp = %lx cnt=%i \n", (long )tp, cnt));
1683
1684	for (; cnt > 0; cnt -= optlen, cp += optlen)
1685	{
1686	opt = cp[0];
1687	if (opt == TCPOPT_EOL)
1688	break;
1689	if (opt == TCPOPT_NOP)
1690	optlen = 1;
1691	else
1692	{
1693	optlen = cp[1];
1694	if (optlen <= 0)
1695	break;
1696	}
1697	switch (opt)
1698	{
1699	default:
1700	continue;
1701
1702	case TCPOPT_MAXSEG:
1703	if (optlen != TCPOLEN_MAXSEG)
1704	continue;
1705	if (!(ti->ti_flags & TH_SYN))
1706	continue;
1707	memcpy((char ) &mss, (char ) cp + 2, sizeof(mss));
1708	NTOHS(mss);
1709	(void) tcp_mss(pData, tp, mss); /* sets t_maxseg */
1710	break;
1711
1712	#if 0
1713	case TCPOPT_WINDOW:
1714	if (optlen != TCPOLEN_WINDOW)
1715	continue;
1716	if (!(ti->ti_flags & TH_SYN))
1717	continue;
1718	tp->t_flags \|= TF_RCVD_SCALE;
1719	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1720	break;
1721
1722	case TCPOPT_TIMESTAMP:
1723	if (optlen != TCPOLEN_TIMESTAMP)
1724	continue;
1725	*ts_present = 1;
1726	memcpy((char ) ts_val, (char )cp + 2, sizeof(*ts_val));
1727	NTOHL(*ts_val);
1728	memcpy((char ) ts_ecr, (char )cp + 6, sizeof(*ts_ecr));
1729	NTOHL(*ts_ecr);
1730
1731	/*
1732	* A timestamp received in a SYN makes
1733	* it ok to send timestamp requests and replies.
1734	*/
1735	if (ti->ti_flags & TH_SYN)
1736	{
1737	tp->t_flags \|= TF_RCVD_TSTMP;
1738	tp->ts_recent = *ts_val;
1739	tp->ts_recent_age = tcp_now;
1740	}
1741	break;
1742	#endif
1743	}
1744	}
1745	}
1746
1747
1748	/*
1749	* Pull out of band byte out of a segment so
1750	* it doesn't appear in the user's data queue.
1751	* It is still reflected in the segment length for
1752	* sequencing purposes.
1753	*/
1754
1755	#if 0
1756	void
1757	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1758	{
1759	int cnt = ti->ti_urp - 1;
1760
1761	while (cnt >= 0)
1762	{
1763	if (m->m_len > cnt)
1764	{
1765	char *cp = mtod(m, caddr_t) + cnt;
1766	struct tcpcb *tp = sototcpcb(so);
1767
1768	tp->t_iobc = *cp;
1769	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1770	memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1771	m->m_len--;
1772	return;
1773	}
1774	cnt -= m->m_len;
1775	m = m->m_next; /* XXX WRONG! Fix it! */
1776	if (m == 0)
1777	break;
1778	}
1779	panic("tcp_pulloutofband");
1780	}
1781	#endif
1782
1783	/*
1784	* Collect new round-trip time estimate
1785	* and update averages and current timeout.
1786	*/
1787
1788	void
1789	tcp_xmit_timer(PNATState pData, register struct tcpcb *tp, int rtt)
1790	{
1791	register short delta;
1792
1793	DEBUG_CALL("tcp_xmit_timer");
1794	DEBUG_ARG("tp = %lx", (long)tp);
1795	DEBUG_ARG("rtt = %d", rtt);
1796
1797	tcpstat.tcps_rttupdated++;
1798	if (tp->t_srtt != 0)
1799	{
1800	/*
1801	* srtt is stored as fixed point with 3 bits after the
1802	* binary point (i.e., scaled by 8). The following magic
1803	* is equivalent to the smoothing algorithm in rfc793 with
1804	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1805	* point). Adjust rtt to origin 0.
1806	*/
1807	delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1808	if ((tp->t_srtt += delta) <= 0)
1809	tp->t_srtt = 1;
1810	/*
1811	* We accumulate a smoothed rtt variance (actually, a
1812	* smoothed mean difference), then set the retransmit
1813	* timer to smoothed rtt + 4 times the smoothed variance.
1814	* rttvar is stored as fixed point with 2 bits after the
1815	* binary point (scaled by 4). The following is
1816	* equivalent to rfc793 smoothing with an alpha of .75
1817	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1818	* rfc793's wired-in beta.
1819	*/
1820	if (delta < 0)
1821	delta = -delta;
1822	delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1823	if ((tp->t_rttvar += delta) <= 0)
1824	tp->t_rttvar = 1;
1825	}
1826	else
1827	{
1828	/*
1829	* No rtt measurement yet - use the unsmoothed rtt.
1830	* Set the variance to half the rtt (so our first
1831	* retransmit happens at 3*rtt).
1832	*/
1833	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1834	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1835	}
1836	tp->t_rtt = 0;
1837	tp->t_rxtshift = 0;
1838
1839	/*
1840	* the retransmit should happen at rtt + 4 * rttvar.
1841	* Because of the way we do the smoothing, srtt and rttvar
1842	* will each average +1/2 tick of bias. When we compute
1843	* the retransmit timer, we want 1/2 tick of rounding and
1844	* 1 extra tick because of +-1/2 tick uncertainty in the
1845	* firing of the timer. The bias will give us exactly the
1846	* 1.5 tick we need. But, because the bias is
1847	* statistical, we have to test that we don't drop below
1848	* the minimum feasible timer (which is 2 ticks).
1849	*/
1850	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1851	(short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1852
1853	/*
1854	* We received an ack for a packet that wasn't retransmitted;
1855	* it is probably safe to discard any error indications we've
1856	* received recently. This isn't quite right, but close enough
1857	* for now (a route might have failed after we sent a segment,
1858	* and the return path might not be symmetrical).
1859	*/
1860	tp->t_softerror = 0;
1861	}
1862
1863	/*
1864	* Determine a reasonable value for maxseg size.
1865	* If the route is known, check route for mtu.
1866	* If none, use an mss that can be handled on the outgoing
1867	* interface without forcing IP to fragment; if bigger than
1868	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1869	* to utilize large mbufs. If no route is found, route has no mtu,
1870	* or the destination isn't local, use a default, hopefully conservative
1871	* size (usually 512 or the default IP max size, but no more than the mtu
1872	* of the interface), as we can't discover anything about intervening
1873	* gateways or networks. We also initialize the congestion/slow start
1874	* window to be a single segment if the destination isn't local.
1875	* While looking at the routing entry, we also initialize other path-dependent
1876	* parameters from pre-set or cached values in the routing entry.
1877	*/
1878
1879	int
1880	tcp_mss(PNATState pData, register struct tcpcb *tp, u_int offer)
1881	{
1882	struct socket *so = tp->t_socket;
1883	int mss;
1884
1885	DEBUG_CALL("tcp_mss");
1886	DEBUG_ARG("tp = %lx", (long)tp);
1887	DEBUG_ARG("offer = %d", offer);
1888
1889	mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
1890	if (offer)
1891	mss = min(mss, offer);
1892	mss = max(mss, 32);
1893	if (mss < tp->t_maxseg \|\| offer != 0)
1894	tp->t_maxseg = mss;
1895
1896	tp->snd_cwnd = mss;
1897
1898	sbreserve(pData, &so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
1899	sbreserve(pData, &so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
1900
1901	DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1902
1903	return mss;
1904	}

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 20453

Download in other formats: