tcp_input.c@ 27848

Last change on this file since 27848 was 27797, checked in by vboxsync, 15 years ago
misc compiler warning fixes, comment typos and other minor cleanups
Property svn:eol-style set to `native`
File size: 61.5 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	STAM_PROFILE_START(&pData->StatTCP_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	STAM_PROFILE_STOP(&pData->StatTCP_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	STAM_PROFILE_STOP(&pData->StatTCP_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	STAM_PROFILE_STOP(&pData->StatTCP_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	STAM_PROFILE_STOP(&pData->StatTCP_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	STAM_PROFILE_STOP(&pData->StatTCP_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	STAM_PROFILE_START(&pData->StatTCP_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
311	/** @todo (vvl) clarify why it might happens */
312	if (ti == NULL)
313	{
314	LogRel(("NAT: ti is null. can't do any reseting connection actions\n"));
315	/* mbuf should be cleared in sofree called from tcp_close */
316	tcp_close(pData, tp);
317	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
318	return;
319	}
320
321	tiwin = ti->ti_win;
322	tiflags = ti->ti_flags;
323
324	goto cont_conn;
325	}
326
327	tcpstat.tcps_rcvtotal++;
328	/*
329	* Get IP and TCP header together in first mbuf.
330	* Note: IP leaves IP header in first mbuf.
331	*/
332	ti = mtod(m, struct tcpiphdr *);
333	if (iphlen > sizeof(struct ip ))
334	{
335	ip_stripoptions(m, (struct mbuf *)0);
336	iphlen = sizeof(struct ip );
337	}
338	/* XXX Check if too short */
339
340
341	/*
342	* Save a copy of the IP header in case we want restore it
343	* for sending an ICMP error message in response.
344	*/
345	ip = mtod(m, struct ip *);
346	save_ip = *ip;
347	save_ip.ip_len+= iphlen;
348
349	/*
350	* Checksum extended TCP header and data.
351	*/
352	tlen = ((struct ip *)ti)->ip_len;
353	memset(ti->ti_x1, 0, 9);
354	ti->ti_len = RT_H2N_U16((u_int16_t)tlen);
355	len = sizeof(struct ip ) + tlen;
356	/* keep checksum for ICMP reply
357	* ti->ti_sum = cksum(m, len);
358	* if (ti->ti_sum) { */
359	if (cksum(m, len))
360	{
361	tcpstat.tcps_rcvbadsum++;
362	goto drop;
363	}
364
365	/*
366	* Check that TCP offset makes sense,
367	* pull out TCP options and adjust length. XXX
368	*/
369	off = ti->ti_off << 2;
370	if ( off < sizeof (struct tcphdr)
371	\|\| off > tlen)
372	{
373	tcpstat.tcps_rcvbadoff++;
374	goto drop;
375	}
376	tlen -= off;
377	ti->ti_len = tlen;
378	if (off > sizeof (struct tcphdr))
379	{
380	optlen = off - sizeof (struct tcphdr);
381	optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
382
383	/*
384	* Do quick retrieval of timestamp options ("options
385	* prediction?"). If timestamp is the only option and it's
386	* formatted as recommended in RFC 1323 appendix A, we
387	* quickly get the values now and not bother calling
388	* tcp_dooptions(), etc.
389	*/
390	#if 0
391	if (( optlen == TCPOLEN_TSTAMP_APPA
392	\|\| ( optlen > TCPOLEN_TSTAMP_APPA
393	&& optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
394	(u_int32_t )optp == RT_H2N_U32_C(TCPOPT_TSTAMP_HDR) &&
395	(ti->ti_flags & TH_SYN) == 0)
396	{
397	ts_present = 1;
398	ts_val = RT_N2H_U32((u_int32_t )(optp + 4));
399	ts_ecr = RT_N2H_U32((u_int32_t )(optp + 8));
400	optp = NULL; / * we have parsed the options * /
401	}
402	#endif
403	}
404	tiflags = ti->ti_flags;
405
406	/*
407	* Convert TCP protocol specific fields to host format.
408	*/
409	NTOHL(ti->ti_seq);
410	NTOHL(ti->ti_ack);
411	NTOHS(ti->ti_win);
412	NTOHS(ti->ti_urp);
413
414	/*
415	* Drop TCP, IP headers and TCP options.
416	*/
417	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
418	m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
419
420	/*
421	* Locate pcb for segment.
422	*/
423	findso:
424	if (so != NULL && so != &tcb)
425	SOCKET_UNLOCK(so);
426	QSOCKET_LOCK(tcb);
427	so = tcp_last_so;
428	if ( so->so_fport != ti->ti_dport
429	\|\| so->so_lport != ti->ti_sport
430	\|\| so->so_laddr.s_addr != ti->ti_src.s_addr
431	\|\| so->so_faddr.s_addr != ti->ti_dst.s_addr)
432	{
433	#ifdef VBOX_WITH_SLIRP_MT
434	struct socket *sonxt;
435	#endif
436	QSOCKET_UNLOCK(tcb);
437	/* @todo fix SOLOOKUP macrodefinition to be usable here */
438	#ifndef VBOX_WITH_SLIRP_MT
439	so = solookup(&tcb, ti->ti_src, ti->ti_sport,
440	ti->ti_dst, ti->ti_dport);
441	#else
442	so = NULL;
443	QSOCKET_FOREACH(so, sonxt, tcp)
444	/* { */
445	if ( so->so_lport == ti->ti_sport
446	&& so->so_laddr.s_addr == ti->ti_src.s_addr
447	&& so->so_faddr.s_addr == ti->ti_dst.s_addr
448	&& so->so_fport == ti->ti_dport
449	&& so->so_deleted != 1)
450	{
451	break; /* so is locked here */
452	}
453	LOOP_LABEL(tcp, so, sonxt);
454	}
455	if (so == &tcb) {
456	so = NULL;
457	}
458	#endif
459	if (so)
460	{
461	tcp_last_so = so;
462	}
463	++tcpstat.tcps_socachemiss;
464	}
465	else
466	{
467	SOCKET_LOCK(so);
468	QSOCKET_UNLOCK(tcb);
469	}
470
471	/*
472	* If the state is CLOSED (i.e., TCB does not exist) then
473	* all data in the incoming segment is discarded.
474	* If the TCB exists but is in CLOSED state, it is embryonic,
475	* but should either do a listen or a connect soon.
476	*
477	* state == CLOSED means we've done socreate() but haven't
478	* attached it to a protocol yet...
479	*
480	* XXX If a TCB does not exist, and the TH_SYN flag is
481	* the only flag set, then create a session, mark it
482	* as if it was LISTENING, and continue...
483	*/
484	if (so == 0)
485	{
486	if ((tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) != TH_SYN)
487	goto dropwithreset;
488
489	if ((so = socreate()) == NULL)
490	goto dropwithreset;
491	if (tcp_attach(pData, so) < 0)
492	{
493	RTMemFree(so); /* Not sofree (if it failed, it's not insqued) */
494	goto dropwithreset;
495	}
496	SOCKET_LOCK(so);
497	sbreserve(pData, &so->so_snd, tcp_sndspace);
498	sbreserve(pData, &so->so_rcv, tcp_rcvspace);
499
500	/* tcp_last_so = so; / / XXX ? */
501	/* tp = sototcpcb(so); */
502
503	so->so_laddr = ti->ti_src;
504	so->so_lport = ti->ti_sport;
505	so->so_faddr = ti->ti_dst;
506	so->so_fport = ti->ti_dport;
507
508	if ((so->so_iptos = tcp_tos(so)) == 0)
509	so->so_iptos = ((struct ip *)ti)->ip_tos;
510
511	tp = sototcpcb(so);
512	tp->t_state = TCPS_LISTEN;
513	}
514
515	/*
516	* If this is a still-connecting socket, this probably
517	* a retransmit of the SYN. Whether it's a retransmit SYN
518	* or something else, we nuke it.
519	*/
520	if (so->so_state & SS_ISFCONNECTING)
521	{
522	goto drop;
523	}
524
525	tp = sototcpcb(so);
526
527	/* XXX Should never fail */
528	if (tp == 0)
529	goto dropwithreset;
530	if (tp->t_state == TCPS_CLOSED)
531	{
532	goto drop;
533	}
534
535	/* Unscale the window into a 32-bit value. */
536	/* if ((tiflags & TH_SYN) == 0)
537	* tiwin = ti->ti_win << tp->snd_scale;
538	* else
539	*/
540	tiwin = ti->ti_win;
541
542	/*
543	* Segment received on connection.
544	* Reset idle time and keep-alive timer.
545	*/
546	tp->t_idle = 0;
547	if (so_options)
548	tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
549	else
550	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
551
552	/*
553	* Process options if not in LISTEN state,
554	* else do it below (after getting remote address).
555	*/
556	if (optp && tp->t_state != TCPS_LISTEN)
557	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
558	/* , */
559	/* &ts_present, &ts_val, &ts_ecr); */
560
561	/*
562	* Header prediction: check for the two common cases
563	* of a uni-directional data xfer. If the packet has
564	* no control flags, is in-sequence, the window didn't
565	* change and we're not retransmitting, it's a
566	* candidate. If the length is zero and the ack moved
567	* forward, we're the sender side of the xfer. Just
568	* free the data acked & wake any higher level process
569	* that was blocked waiting for space. If the length
570	* is non-zero and the ack didn't move, we're the
571	* receiver side. If we're getting packets in-order
572	* (the reassembly queue is empty), add the data to
573	* the socket buffer and note that we need a delayed ack.
574	*
575	* XXX Some of these tests are not needed
576	* eg: the tiwin == tp->snd_wnd prevents many more
577	* predictions.. with no real advantage..
578	*/
579	if ( tp->t_state == TCPS_ESTABLISHED
580	&& (tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK
581	/* && (!ts_present \|\| TSTMP_GEQ(ts_val, tp->ts_recent)) */
582	&& ti->ti_seq == tp->rcv_nxt
583	&& tiwin && tiwin == tp->snd_wnd
584	&& tp->snd_nxt == tp->snd_max)
585	{
586	/*
587	* If last ACK falls within this segment's sequence numbers,
588	* record the timestamp.
589	*/
590	#if 0
591	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
592	SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len))
593	{
594	tp->ts_recent_age = tcp_now;
595	tp->ts_recent = ts_val;
596	}
597	#endif
598
599	if (ti->ti_len == 0)
600	{
601	if ( SEQ_GT(ti->ti_ack, tp->snd_una)
602	&& SEQ_LEQ(ti->ti_ack, tp->snd_max)
603	&& tp->snd_cwnd >= tp->snd_wnd)
604	{
605	/*
606	* this is a pure ack for outstanding data.
607	*/
608	++tcpstat.tcps_predack;
609	#if 0
610	if (ts_present)
611	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
612	else
613	#endif
614	if ( tp->t_rtt
615	&& SEQ_GT(ti->ti_ack, tp->t_rtseq))
616	tcp_xmit_timer(pData, tp, tp->t_rtt);
617	acked = ti->ti_ack - tp->snd_una;
618	tcpstat.tcps_rcvackpack++;
619	tcpstat.tcps_rcvackbyte += acked;
620	sbdrop(&so->so_snd, acked);
621	tp->snd_una = ti->ti_ack;
622	m_freem(pData, m);
623
624	/*
625	* If all outstanding data are acked, stop
626	* retransmit timer, otherwise restart timer
627	* using current (possibly backed-off) value.
628	* If process is waiting for space,
629	* wakeup/selwakeup/signal. If data
630	* are ready to send, let tcp_output
631	* decide between more output or persist.
632	*/
633	if (tp->snd_una == tp->snd_max)
634	tp->t_timer[TCPT_REXMT] = 0;
635	else if (tp->t_timer[TCPT_PERSIST] == 0)
636	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
637
638	/*
639	* There's room in so_snd, sowwakup will read()
640	* from the socket if we can
641	*/
642	#if 0
643	if (so->so_snd.sb_flags & SB_NOTIFY)
644	sowwakeup(so);
645	#endif
646	/*
647	* This is called because sowwakeup might have
648	* put data into so_snd. Since we don't so sowwakeup,
649	* we don't need this.. XXX???
650	*/
651	if (so->so_snd.sb_cc)
652	(void) tcp_output(pData, tp);
653
654	SOCKET_UNLOCK(so);
655	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
656	return;
657	}
658	}
659	else if ( ti->ti_ack == tp->snd_una
660	&& LIST_FIRST(&tp->t_segq)
661	&& ti->ti_len <= sbspace(&so->so_rcv))
662	{
663	/*
664	* this is a pure, in-sequence data packet
665	* with nothing on the reassembly queue and
666	* we have enough buffer space to take it.
667	*/
668	++tcpstat.tcps_preddat;
669	tp->rcv_nxt += ti->ti_len;
670	tcpstat.tcps_rcvpack++;
671	tcpstat.tcps_rcvbyte += ti->ti_len;
672	/*
673	* Add data to socket buffer.
674	*/
675	if (so->so_emu)
676	{
677	if (tcp_emu(pData, so, m))
678	sbappend(pData, so, m);
679	}
680	else
681	sbappend(pData, so, m);
682
683	/*
684	* XXX This is called when data arrives. Later, check
685	* if we can actually write() to the socket
686	* XXX Need to check? It's be NON_BLOCKING
687	*/
688	/* sorwakeup(so); */
689
690	/*
691	* If this is a short packet, then ACK now - with Nagel
692	* congestion avoidance sender won't send more until
693	* he gets an ACK.
694	*
695	* It is better to not delay acks at all to maximize
696	* TCP throughput. See RFC 2581.
697	*/
698	tp->t_flags \|= TF_ACKNOW;
699	tcp_output(pData, tp);
700	SOCKET_UNLOCK(so);
701	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
702	return;
703	}
704	} /* header prediction */
705	/*
706	* Calculate amount of space in receive window,
707	* and then do TCP input processing.
708	* Receive window is amount of space in rcv queue,
709	* but not less than advertised window.
710	*/
711	{
712	int win;
713	win = sbspace(&so->so_rcv);
714	if (win < 0)
715	win = 0;
716	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
717	}
718
719	switch (tp->t_state)
720	{
721	/*
722	* If the state is LISTEN then ignore segment if it contains an RST.
723	* If the segment contains an ACK then it is bad and send a RST.
724	* If it does not contain a SYN then it is not interesting; drop it.
725	* Don't bother responding if the destination was a broadcast.
726	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
727	* tp->iss, and send a segment:
728	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
729	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
730	* Fill in remote peer address fields if not previously specified.
731	* Enter SYN_RECEIVED state, and process any other fields of this
732	* segment in this state.
733	*/
734	case TCPS_LISTEN:
735	{
736	if (tiflags & TH_RST) {
737	goto drop;
738	}
739	if (tiflags & TH_ACK)
740	goto dropwithreset;
741	if ((tiflags & TH_SYN) == 0)
742	{
743	goto drop;
744	}
745
746	/*
747	* This has way too many gotos...
748	* But a bit of spaghetti code never hurt anybody :)
749	*/
750
751	if (so->so_emu & EMU_NOCONNECT)
752	{
753	so->so_emu &= ~EMU_NOCONNECT;
754	goto cont_input;
755	}
756
757	if ( (tcp_fconnect(pData, so) == -1)
758	&& errno != EINPROGRESS
759	&& errno != EWOULDBLOCK)
760	{
761	u_char code = ICMP_UNREACH_NET;
762	DEBUG_MISC((dfd," tcp fconnect errno = %d-%s\n",
763	errno, strerror(errno)));
764	if (errno == ECONNREFUSED)
765	{
766	/* ACK the SYN, send RST to refuse the connection */
767	tcp_respond(pData, tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
768	TH_RST\|TH_ACK);
769	}
770	else
771	{
772	if (errno == EHOSTUNREACH)
773	code = ICMP_UNREACH_HOST;
774	HTONL(ti->ti_seq); /* restore tcp header */
775	HTONL(ti->ti_ack);
776	HTONS(ti->ti_win);
777	HTONS(ti->ti_urp);
778	m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
779	m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
780	*ip = save_ip;
781	icmp_error(pData, m, ICMP_UNREACH, code, 0, strerror(errno));
782	tp->t_socket->so_m = NULL;
783	}
784	tp = tcp_close(pData, tp);
785	m_free(pData, m);
786	}
787	else
788	{
789	/*
790	* Haven't connected yet, save the current mbuf
791	* and ti, and return
792	* XXX Some OS's don't tell us whether the connect()
793	* succeeded or not. So we must time it out.
794	*/
795	so->so_m = m;
796	so->so_ti = ti;
797	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
798	tp->t_state = TCPS_SYN_RECEIVED;
799	}
800	SOCKET_UNLOCK(so);
801	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
802	return;
803
804	cont_conn:
805	/* m==NULL
806	* Check if the connect succeeded
807	*/
808	if (so->so_state & SS_NOFDREF)
809	{
810	tp = tcp_close(pData, tp);
811	goto dropwithreset;
812	}
813	cont_input:
814	tcp_template(tp);
815
816	if (optp)
817	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
818
819	if (iss)
820	tp->iss = iss;
821	else
822	tp->iss = tcp_iss;
823	tcp_iss += TCP_ISSINCR/2;
824	tp->irs = ti->ti_seq;
825	tcp_sendseqinit(tp);
826	tcp_rcvseqinit(tp);
827	tp->t_flags \|= TF_ACKNOW;
828	tp->t_state = TCPS_SYN_RECEIVED;
829	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
830	tcpstat.tcps_accepts++;
831	goto trimthenstep6;
832	} /* case TCPS_LISTEN */
833
834	/*
835	* If the state is SYN_SENT:
836	* if seg contains an ACK, but not for our SYN, drop the input.
837	* if seg contains a RST, then drop the connection.
838	* if seg does not contain SYN, then drop it.
839	* Otherwise this is an acceptable SYN segment
840	* initialize tp->rcv_nxt and tp->irs
841	* if seg contains ack then advance tp->snd_una
842	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
843	* arrange for segment to be acked (eventually)
844	* continue processing rest of data/controls, beginning with URG
845	*/
846	case TCPS_SYN_SENT:
847	if ( (tiflags & TH_ACK)
848	&& ( SEQ_LEQ(ti->ti_ack, tp->iss)
849	\|\| SEQ_GT(ti->ti_ack, tp->snd_max)))
850	goto dropwithreset;
851
852	if (tiflags & TH_RST)
853	{
854	if (tiflags & TH_ACK)
855	tp = tcp_drop(pData, tp, 0); /* XXX Check t_softerror! */
856	goto drop;
857	}
858
859	if ((tiflags & TH_SYN) == 0)
860	{
861	goto drop;
862	}
863	if (tiflags & TH_ACK)
864	{
865	tp->snd_una = ti->ti_ack;
866	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
867	tp->snd_nxt = tp->snd_una;
868	}
869
870	tp->t_timer[TCPT_REXMT] = 0;
871	tp->irs = ti->ti_seq;
872	tcp_rcvseqinit(tp);
873	tp->t_flags \|= TF_ACKNOW;
874	if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss))
875	{
876	tcpstat.tcps_connects++;
877	soisfconnected(so);
878	tp->t_state = TCPS_ESTABLISHED;
879
880	/* Do window scaling on this connection? */
881	#if 0
882	if (( tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
883	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
884	{
885	tp->snd_scale = tp->requested_s_scale;
886	tp->rcv_scale = tp->request_r_scale;
887	}
888	#endif
889	(void) tcp_reass(pData, tp, (struct tcphdr )0, NULL, (struct mbuf )0);
890	/*
891	* if we didn't have to retransmit the SYN,
892	* use its rtt as our initial srtt & rtt var.
893	*/
894	if (tp->t_rtt)
895	tcp_xmit_timer(pData, tp, tp->t_rtt);
896	}
897	else
898	tp->t_state = TCPS_SYN_RECEIVED;
899
900	trimthenstep6:
901	/*
902	* Advance ti->ti_seq to correspond to first data byte.
903	* If data, trim to stay within window,
904	* dropping FIN if necessary.
905	*/
906	ti->ti_seq++;
907	if (ti->ti_len > tp->rcv_wnd)
908	{
909	todrop = ti->ti_len - tp->rcv_wnd;
910	m_adj(m, -todrop);
911	ti->ti_len = tp->rcv_wnd;
912	tiflags &= ~TH_FIN;
913	tcpstat.tcps_rcvpackafterwin++;
914	tcpstat.tcps_rcvbyteafterwin += todrop;
915	}
916	tp->snd_wl1 = ti->ti_seq - 1;
917	tp->rcv_up = ti->ti_seq;
918	Log2(("hit6"));
919	goto step6;
920	} /* switch tp->t_state */
921	/*
922	* States other than LISTEN or SYN_SENT.
923	* First check timestamp, if present.
924	* Then check that at least some bytes of segment are within
925	* receive window. If segment begins before rcv_nxt,
926	* drop leading data (and SYN); if nothing left, just ack.
927	*
928	* RFC 1323 PAWS: If we have a timestamp reply on this segment
929	* and it's less than ts_recent, drop it.
930	*/
931	#if 0
932	if ( ts_present
933	&& (tiflags & TH_RST) == 0
934	&& tp->ts_recent
935	&& TSTMP_LT(ts_val, tp->ts_recent))
936	{
937	/* Check to see if ts_recent is over 24 days old. */
938	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE)
939	{
940	/*
941	* Invalidate ts_recent. If this segment updates
942	* ts_recent, the age will be reset later and ts_recent
943	* will get a valid value. If it does not, setting
944	* ts_recent to zero will at least satisfy the
945	* requirement that zero be placed in the timestamp
946	* echo reply when ts_recent isn't valid. The
947	* age isn't reset until we get a valid ts_recent
948	* because we don't want out-of-order segments to be
949	* dropped when ts_recent is old.
950	*/
951	tp->ts_recent = 0;
952	}
953	else
954	{
955	tcpstat.tcps_rcvduppack++;
956	tcpstat.tcps_rcvdupbyte += ti->ti_len;
957	tcpstat.tcps_pawsdrop++;
958	goto dropafterack;
959	}
960	}
961	#endif
962
963	todrop = tp->rcv_nxt - ti->ti_seq;
964	if (todrop > 0)
965	{
966	if (tiflags & TH_SYN)
967	{
968	tiflags &= ~TH_SYN;
969	ti->ti_seq++;
970	if (ti->ti_urp > 1)
971	ti->ti_urp--;
972	else
973	tiflags &= ~TH_URG;
974	todrop--;
975	}
976	/*
977	* Following if statement from Stevens, vol. 2, p. 960.
978	*/
979	if ( todrop > ti->ti_len
980	\|\| ( todrop == ti->ti_len
981	&& (tiflags & TH_FIN) == 0))
982	{
983	/*
984	* Any valid FIN must be to the left of the window.
985	* At this point the FIN must be a duplicate or out
986	* of sequence; drop it.
987	*/
988	tiflags &= ~TH_FIN;
989
990	/*
991	* Send an ACK to resynchronize and drop any data.
992	* But keep on processing for RST or ACK.
993	*/
994	tp->t_flags \|= TF_ACKNOW;
995	todrop = ti->ti_len;
996	tcpstat.tcps_rcvduppack++;
997	tcpstat.tcps_rcvdupbyte += todrop;
998	}
999	else
1000	{
1001	tcpstat.tcps_rcvpartduppack++;
1002	tcpstat.tcps_rcvpartdupbyte += todrop;
1003	}
1004	m_adj(m, todrop);
1005	ti->ti_seq += todrop;
1006	ti->ti_len -= todrop;
1007	if (ti->ti_urp > todrop)
1008	ti->ti_urp -= todrop;
1009	else
1010	{
1011	tiflags &= ~TH_URG;
1012	ti->ti_urp = 0;
1013	}
1014	}
1015	/*
1016	* If new data are received on a connection after the
1017	* user processes are gone, then RST the other end.
1018	*/
1019	if ( (so->so_state & SS_NOFDREF)
1020	&& tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len)
1021	{
1022	tp = tcp_close(pData, tp);
1023	tcpstat.tcps_rcvafterclose++;
1024	goto dropwithreset;
1025	}
1026
1027	/*
1028	* If segment ends after window, drop trailing data
1029	* (and PUSH and FIN); if nothing left, just ACK.
1030	*/
1031	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1032	if (todrop > 0)
1033	{
1034	tcpstat.tcps_rcvpackafterwin++;
1035	if (todrop >= ti->ti_len)
1036	{
1037	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1038	/*
1039	* If a new connection request is received
1040	* while in TIME_WAIT, drop the old connection
1041	* and start over if the sequence numbers
1042	* are above the previous ones.
1043	*/
1044	if ( tiflags & TH_SYN
1045	&& tp->t_state == TCPS_TIME_WAIT
1046	&& SEQ_GT(ti->ti_seq, tp->rcv_nxt))
1047	{
1048	iss = tp->rcv_nxt + TCP_ISSINCR;
1049	tp = tcp_close(pData, tp);
1050	SOCKET_UNLOCK(tp->t_socket);
1051	goto findso;
1052	}
1053	/*
1054	* If window is closed can only take segments at
1055	* window edge, and have to drop data and PUSH from
1056	* incoming segments. Continue processing, but
1057	* remember to ack. Otherwise, drop segment
1058	* and ack.
1059	*/
1060	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt)
1061	{
1062	tp->t_flags \|= TF_ACKNOW;
1063	tcpstat.tcps_rcvwinprobe++;
1064	}
1065	else
1066	goto dropafterack;
1067	}
1068	else
1069	tcpstat.tcps_rcvbyteafterwin += todrop;
1070	m_adj(m, -todrop);
1071	ti->ti_len -= todrop;
1072	tiflags &= ~(TH_PUSH\|TH_FIN);
1073	}
1074
1075	/*
1076	* If last ACK falls within this segment's sequence numbers,
1077	* record its timestamp.
1078	*/
1079	#if 0
1080	if ( ts_present
1081	&& SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)
1082	&& SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + ((tiflags & (TH_SYN\|TH_FIN)) != 0)))
1083	{
1084	tp->ts_recent_age = tcp_now;
1085	tp->ts_recent = ts_val;
1086	}
1087	#endif
1088
1089	/*
1090	* If the RST bit is set examine the state:
1091	* SYN_RECEIVED STATE:
1092	* If passive open, return to LISTEN state.
1093	* If active open, inform user that connection was refused.
1094	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1095	* Inform user that connection was reset, and close tcb.
1096	* CLOSING, LAST_ACK, TIME_WAIT STATES
1097	* Close the tcb.
1098	*/
1099	if (tiflags&TH_RST)
1100	switch (tp->t_state)
1101	{
1102	case TCPS_SYN_RECEIVED:
1103	/* so->so_error = ECONNREFUSED; */
1104	goto close;
1105
1106	case TCPS_ESTABLISHED:
1107	case TCPS_FIN_WAIT_1:
1108	case TCPS_FIN_WAIT_2:
1109	case TCPS_CLOSE_WAIT:
1110	/* so->so_error = ECONNRESET; */
1111	close:
1112	tp->t_state = TCPS_CLOSED;
1113	tcpstat.tcps_drops++;
1114	tp = tcp_close(pData, tp);
1115	goto drop;
1116
1117	case TCPS_CLOSING:
1118	case TCPS_LAST_ACK:
1119	case TCPS_TIME_WAIT:
1120	tp = tcp_close(pData, tp);
1121	goto drop;
1122	}
1123
1124	/*
1125	* If a SYN is in the window, then this is an
1126	* error and we send an RST and drop the connection.
1127	*/
1128	if (tiflags & TH_SYN)
1129	{
1130	tp = tcp_drop(pData, tp, 0);
1131	goto dropwithreset;
1132	}
1133
1134	/*
1135	* If the ACK bit is off we drop the segment and return.
1136	*/
1137	if ((tiflags & TH_ACK) == 0)
1138	{
1139	goto drop;
1140	}
1141
1142	/*
1143	* Ack processing.
1144	*/
1145	switch (tp->t_state)
1146	{
1147	/*
1148	* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1149	* ESTABLISHED state and continue processing, otherwise
1150	* send an RST. una<=ack<=max
1151	*/
1152	case TCPS_SYN_RECEIVED:
1153	if ( SEQ_GT(tp->snd_una, ti->ti_ack)
1154	\|\| SEQ_GT(ti->ti_ack, tp->snd_max))
1155	goto dropwithreset;
1156	tcpstat.tcps_connects++;
1157	tp->t_state = TCPS_ESTABLISHED;
1158	/*
1159	* The sent SYN is ack'ed with our sequence number +1
1160	* The first data byte already in the buffer will get
1161	* lost if no correction is made. This is only needed for
1162	* SS_CTL since the buffer is empty otherwise.
1163	* tp->snd_una++; or:
1164	*/
1165	tp->snd_una = ti->ti_ack;
1166	soisfconnected(so);
1167
1168	/* Do window scaling? */
1169	#if 0
1170	if ( (tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
1171	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
1172	{
1173	tp->snd_scale = tp->requested_s_scale;
1174	tp->rcv_scale = tp->request_r_scale;
1175	}
1176	#endif
1177	(void) tcp_reass(pData, tp, (struct tcphdr )0, (int )0, (struct mbuf *)0);
1178	tp->snd_wl1 = ti->ti_seq - 1;
1179	/* Avoid ack processing; snd_una==ti_ack => dup ack */
1180	goto synrx_to_est;
1181	/* fall into ... */
1182
1183	/*
1184	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1185	* ACKs. If the ack is in the range
1186	* tp->snd_una < ti->ti_ack <= tp->snd_max
1187	* then advance tp->snd_una to ti->ti_ack and drop
1188	* data from the retransmission queue. If this ACK reflects
1189	* more up to date window information we update our window information.
1190	*/
1191	case TCPS_ESTABLISHED:
1192	case TCPS_FIN_WAIT_1:
1193	case TCPS_FIN_WAIT_2:
1194	case TCPS_CLOSE_WAIT:
1195	case TCPS_CLOSING:
1196	case TCPS_LAST_ACK:
1197	case TCPS_TIME_WAIT:
1198	if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
1199	{
1200	if (ti->ti_len == 0 && tiwin == tp->snd_wnd)
1201	{
1202	tcpstat.tcps_rcvdupack++;
1203	DEBUG_MISC((dfd," dup ack m = %lx so = %lx \n",
1204	(long )m, (long )so));
1205	/*
1206	* If we have outstanding data (other than
1207	* a window probe), this is a completely
1208	* duplicate ack (ie, window info didn't
1209	* change), the ack is the biggest we've
1210	* seen and we've seen exactly our rexmt
1211	* threshold of them, assume a packet
1212	* has been dropped and retransmit it.
1213	* Kludge snd_nxt & the congestion
1214	* window so we send only this one
1215	* packet.
1216	*
1217	* We know we're losing at the current
1218	* window size so do congestion avoidance
1219	* (set ssthresh to half the current window
1220	* and pull our congestion window back to
1221	* the new ssthresh).
1222	*
1223	* Dup acks mean that packets have left the
1224	* network (they're now cached at the receiver)
1225	* so bump cwnd by the amount in the receiver
1226	* to keep a constant cwnd packets in the
1227	* network.
1228	*/
1229	if ( tp->t_timer[TCPT_REXMT] == 0
1230	\|\| ti->ti_ack != tp->snd_una)
1231	tp->t_dupacks = 0;
1232	else if (++tp->t_dupacks == tcprexmtthresh)
1233	{
1234	tcp_seq onxt = tp->snd_nxt;
1235	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
1236	if (win < 2)
1237	win = 2;
1238	tp->snd_ssthresh = win * tp->t_maxseg;
1239	tp->t_timer[TCPT_REXMT] = 0;
1240	tp->t_rtt = 0;
1241	tp->snd_nxt = ti->ti_ack;
1242	tp->snd_cwnd = tp->t_maxseg;
1243	(void) tcp_output(pData, tp);
1244	tp->snd_cwnd = tp->snd_ssthresh +
1245	tp->t_maxseg * tp->t_dupacks;
1246	if (SEQ_GT(onxt, tp->snd_nxt))
1247	tp->snd_nxt = onxt;
1248	goto drop;
1249	}
1250	else if (tp->t_dupacks > tcprexmtthresh)
1251	{
1252	tp->snd_cwnd += tp->t_maxseg;
1253	(void) tcp_output(pData, tp);
1254	goto drop;
1255	}
1256	}
1257	else
1258	tp->t_dupacks = 0;
1259	break;
1260	}
1261	synrx_to_est:
1262	/*
1263	* If the congestion window was inflated to account
1264	* for the other side's cached packets, retract it.
1265	*/
1266	if ( tp->t_dupacks > tcprexmtthresh
1267	&& tp->snd_cwnd > tp->snd_ssthresh)
1268	tp->snd_cwnd = tp->snd_ssthresh;
1269	tp->t_dupacks = 0;
1270	if (SEQ_GT(ti->ti_ack, tp->snd_max))
1271	{
1272	tcpstat.tcps_rcvacktoomuch++;
1273	goto dropafterack;
1274	}
1275	acked = ti->ti_ack - tp->snd_una;
1276	tcpstat.tcps_rcvackpack++;
1277	tcpstat.tcps_rcvackbyte += acked;
1278
1279	/*
1280	* If we have a timestamp reply, update smoothed
1281	* round trip time. If no timestamp is present but
1282	* transmit timer is running and timed sequence
1283	* number was acked, update smoothed round trip time.
1284	* Since we now have an rtt measurement, cancel the
1285	* timer backoff (cf., Phil Karn's retransmit alg.).
1286	* Recompute the initial retransmit timer.
1287	*/
1288	#if 0
1289	if (ts_present)
1290	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1291	else
1292	#endif
1293	if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1294	tcp_xmit_timer(pData, tp, tp->t_rtt);
1295
1296	/*
1297	* If all outstanding data is acked, stop retransmit
1298	* timer and remember to restart (more output or persist).
1299	* If there is more data to be acked, restart retransmit
1300	* timer, using current (possibly backed-off) value.
1301	*/
1302	if (ti->ti_ack == tp->snd_max)
1303	{
1304	tp->t_timer[TCPT_REXMT] = 0;
1305	needoutput = 1;
1306	}
1307	else if (tp->t_timer[TCPT_PERSIST] == 0)
1308	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1309	/*
1310	* When new data is acked, open the congestion window.
1311	* If the window gives us less than ssthresh packets
1312	* in flight, open exponentially (maxseg per packet).
1313	* Otherwise open linearly: maxseg per window
1314	* (maxseg^2 / cwnd per packet).
1315	*/
1316	{
1317	register u_int cw = tp->snd_cwnd;
1318	register u_int incr = tp->t_maxseg;
1319
1320	if (cw > tp->snd_ssthresh)
1321	incr = incr * incr / cw;
1322	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1323	}
1324	if (acked > so->so_snd.sb_cc)
1325	{
1326	tp->snd_wnd -= so->so_snd.sb_cc;
1327	sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1328	ourfinisacked = 1;
1329	}
1330	else
1331	{
1332	sbdrop(&so->so_snd, acked);
1333	tp->snd_wnd -= acked;
1334	ourfinisacked = 0;
1335	}
1336	/*
1337	* XXX sowwakup is called when data is acked and there's room for
1338	* for more data... it should read() the socket
1339	*/
1340	#if 0
1341	if (so->so_snd.sb_flags & SB_NOTIFY)
1342	sowwakeup(so);
1343	#endif
1344	tp->snd_una = ti->ti_ack;
1345	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1346	tp->snd_nxt = tp->snd_una;
1347
1348	switch (tp->t_state)
1349	{
1350	/*
1351	* In FIN_WAIT_1 STATE in addition to the processing
1352	* for the ESTABLISHED state if our FIN is now acknowledged
1353	* then enter FIN_WAIT_2.
1354	*/
1355	case TCPS_FIN_WAIT_1:
1356	if (ourfinisacked)
1357	{
1358	/*
1359	* If we can't receive any more
1360	* data, then closing user can proceed.
1361	* Starting the timer is contrary to the
1362	* specification, but if we don't get a FIN
1363	* we'll hang forever.
1364	*/
1365	if (so->so_state & SS_FCANTRCVMORE)
1366	{
1367	soisfdisconnected(so);
1368	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1369	}
1370	tp->t_state = TCPS_FIN_WAIT_2;
1371	}
1372	break;
1373
1374	/*
1375	* In CLOSING STATE in addition to the processing for
1376	* the ESTABLISHED state if the ACK acknowledges our FIN
1377	* then enter the TIME-WAIT state, otherwise ignore
1378	* the segment.
1379	*/
1380	case TCPS_CLOSING:
1381	if (ourfinisacked)
1382	{
1383	tp->t_state = TCPS_TIME_WAIT;
1384	tcp_canceltimers(tp);
1385	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1386	soisfdisconnected(so);
1387	}
1388	break;
1389
1390	/*
1391	* In LAST_ACK, we may still be waiting for data to drain
1392	* and/or to be acked, as well as for the ack of our FIN.
1393	* If our FIN is now acknowledged, delete the TCB,
1394	* enter the closed state and return.
1395	*/
1396	case TCPS_LAST_ACK:
1397	if (ourfinisacked)
1398	{
1399	tp = tcp_close(pData, tp);
1400	goto drop;
1401	}
1402	break;
1403
1404	/*
1405	* In TIME_WAIT state the only thing that should arrive
1406	* is a retransmission of the remote FIN. Acknowledge
1407	* it and restart the finack timer.
1408	*/
1409	case TCPS_TIME_WAIT:
1410	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1411	goto dropafterack;
1412	}
1413	} /* switch(tp->t_state) */
1414
1415	step6:
1416	/*
1417	* Update window information.
1418	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1419	*/
1420	if ( (tiflags & TH_ACK)
1421	&& ( SEQ_LT(tp->snd_wl1, ti->ti_seq)
1422	\|\| ( tp->snd_wl1 == ti->ti_seq
1423	&& ( SEQ_LT(tp->snd_wl2, ti->ti_ack)
1424	\|\| ( tp->snd_wl2 == ti->ti_ack
1425	&& tiwin > tp->snd_wnd)))))
1426	{
1427	/* keep track of pure window updates */
1428	if ( ti->ti_len == 0
1429	&& tp->snd_wl2 == ti->ti_ack
1430	&& tiwin > tp->snd_wnd)
1431	tcpstat.tcps_rcvwinupd++;
1432	tp->snd_wnd = tiwin;
1433	tp->snd_wl1 = ti->ti_seq;
1434	tp->snd_wl2 = ti->ti_ack;
1435	if (tp->snd_wnd > tp->max_sndwnd)
1436	tp->max_sndwnd = tp->snd_wnd;
1437	needoutput = 1;
1438	}
1439
1440	/*
1441	* Process segments with URG.
1442	*/
1443	if ((tiflags & TH_URG) && ti->ti_urp &&
1444	TCPS_HAVERCVDFIN(tp->t_state) == 0)
1445	{
1446	/*
1447	* This is a kludge, but if we receive and accept
1448	* random urgent pointers, we'll crash in
1449	* soreceive. It's hard to imagine someone
1450	* actually wanting to send this much urgent data.
1451	*/
1452	if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen)
1453	{
1454	ti->ti_urp = 0;
1455	tiflags &= ~TH_URG;
1456	goto dodata;
1457	}
1458	/*
1459	* If this segment advances the known urgent pointer,
1460	* then mark the data stream. This should not happen
1461	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1462	* a FIN has been received from the remote side.
1463	* In these states we ignore the URG.
1464	*
1465	* According to RFC961 (Assigned Protocols),
1466	* the urgent pointer points to the last octet
1467	* of urgent data. We continue, however,
1468	* to consider it to indicate the first octet
1469	* of data past the urgent section as the original
1470	* spec states (in one of two places).
1471	*/
1472	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up))
1473	{
1474	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1475	so->so_urgc = so->so_rcv.sb_cc +
1476	(tp->rcv_up - tp->rcv_nxt); /* -1; */
1477	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1478	}
1479	}
1480	else
1481	/*
1482	* If no out of band data is expected,
1483	* pull receive urgent pointer along
1484	* with the receive window.
1485	*/
1486	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1487	tp->rcv_up = tp->rcv_nxt;
1488	dodata:
1489
1490	/*
1491	* If this is a small packet, then ACK now - with Nagel
1492	* congestion avoidance sender won't send more until
1493	* he gets an ACK.
1494	*
1495	* See above.
1496	*/
1497	if ( ti->ti_len
1498	&& (unsigned)ti->ti_len <= 5
1499	&& ((struct tcpiphdr_2 *)ti)->first_char == (char)27)
1500	{
1501	tp->t_flags \|= TF_ACKNOW;
1502	}
1503
1504	/*
1505	* Process the segment text, merging it into the TCP sequencing queue,
1506	* and arranging for acknowledgment of receipt if necessary.
1507	* This process logically involves adjusting tp->rcv_wnd as data
1508	* is presented to the user (this happens in tcp_usrreq.c,
1509	* case PRU_RCVD). If a FIN has already been received on this
1510	* connection then we just ignore the text.
1511	*/
1512	if ( (ti->ti_len \|\| (tiflags&TH_FIN))
1513	&& TCPS_HAVERCVDFIN(tp->t_state) == 0)
1514	{
1515	if ( ti->ti_seq == tp->rcv_nxt
1516	&& LIST_EMPTY(&tp->t_segq)
1517	&& tp->t_state == TCPS_ESTABLISHED)
1518	{
1519	DELAY_ACK(tp, ti); /* little bit different from BSD declaration see netinet/tcp_input.c */
1520	tp->rcv_nxt += tlen;
1521	tiflags = ti->ti_t.th_flags & TH_FIN;
1522	tcpstat.tcps_rcvpack++;
1523	tcpstat.tcps_rcvbyte += tlen;
1524	if (so->so_state & SS_FCANTRCVMORE)
1525	m_freem(pData, m);
1526	else
1527	{
1528	if (so->so_emu)
1529	{
1530	if (tcp_emu(pData, so, m))
1531	sbappend(pData, so, m);
1532	}
1533	else
1534	sbappend(pData, so, m);
1535	}
1536	}
1537	else
1538	{
1539	tiflags = tcp_reass(pData, tp, &ti->ti_t, &tlen, m);
1540	tiflags \|= TF_ACKNOW;
1541	}
1542	/*
1543	* Note the amount of data that peer has sent into
1544	* our window, in order to estimate the sender's
1545	* buffer size.
1546	*/
1547	len = so->so_rcv.sb_datalen - (tp->rcv_adv - tp->rcv_nxt);
1548	}
1549	else
1550	{
1551	m_free(pData, m);
1552	tiflags &= ~TH_FIN;
1553	}
1554
1555	/*
1556	* If FIN is received ACK the FIN and let the user know
1557	* that the connection is closing.
1558	*/
1559	if (tiflags & TH_FIN)
1560	{
1561	if (TCPS_HAVERCVDFIN(tp->t_state) == 0)
1562	{
1563	/*
1564	* If we receive a FIN we can't send more data,
1565	* set it SS_FDRAIN
1566	* Shutdown the socket if there is no rx data in the
1567	* buffer.
1568	* soread() is called on completion of shutdown() and
1569	* will got to TCPS_LAST_ACK, and use tcp_output()
1570	* to send the FIN.
1571	*/
1572	/* sofcantrcvmore(so); */
1573	sofwdrain(so);
1574
1575	tp->t_flags \|= TF_ACKNOW;
1576	tp->rcv_nxt++;
1577	}
1578	switch (tp->t_state)
1579	{
1580	/*
1581	* In SYN_RECEIVED and ESTABLISHED STATES
1582	* enter the CLOSE_WAIT state.
1583	*/
1584	case TCPS_SYN_RECEIVED:
1585	case TCPS_ESTABLISHED:
1586	if(so->so_emu == EMU_CTL) /* no shutdown on socket */
1587	tp->t_state = TCPS_LAST_ACK;
1588	else
1589	tp->t_state = TCPS_CLOSE_WAIT;
1590	break;
1591
1592	/*
1593	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1594	* enter the CLOSING state.
1595	*/
1596	case TCPS_FIN_WAIT_1:
1597	tp->t_state = TCPS_CLOSING;
1598	break;
1599
1600	/*
1601	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1602	* starting the time-wait timer, turning off the other
1603	* standard timers.
1604	*/
1605	case TCPS_FIN_WAIT_2:
1606	tp->t_state = TCPS_TIME_WAIT;
1607	tcp_canceltimers(tp);
1608	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1609	soisfdisconnected(so);
1610	break;
1611
1612	/*
1613	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1614	*/
1615	case TCPS_TIME_WAIT:
1616	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1617	break;
1618	}
1619	}
1620
1621	/*
1622	* Return any desired output.
1623	*/
1624	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1625	tcp_output(pData, tp);
1626
1627	SOCKET_UNLOCK(so);
1628	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1629	return;
1630
1631	dropafterack:
1632	Log2(("drop after ack\n"));
1633	/*
1634	* Generate an ACK dropping incoming segment if it occupies
1635	* sequence space, where the ACK reflects our state.
1636	*/
1637	if (tiflags & TH_RST)
1638	goto drop;
1639	m_freem(pData, m);
1640	tp->t_flags \|= TF_ACKNOW;
1641	(void) tcp_output(pData, tp);
1642	SOCKET_UNLOCK(so);
1643	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1644	return;
1645
1646	dropwithreset:
1647	/* reuses m if m!=NULL, m_free() unnecessary */
1648	if (tiflags & TH_ACK)
1649	tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1650	else
1651	{
1652	if (tiflags & TH_SYN) ti->ti_len++;
1653	tcp_respond(pData, tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1654	TH_RST\|TH_ACK);
1655	}
1656
1657	if (so != &tcb)
1658	SOCKET_UNLOCK(so);
1659	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1660	return;
1661
1662	drop:
1663	/*
1664	* Drop space held by incoming segment and return.
1665	*/
1666	m_free(pData, m);
1667
1668	#ifdef VBOX_WITH_SLIRP_MT
1669	if (RTCritSectIsOwned(&so->so_mutex))
1670	{
1671	SOCKET_UNLOCK(so);
1672	}
1673	#endif
1674
1675	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1676	return;
1677	}
1678
1679	void
1680	tcp_dooptions(PNATState pData, struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti)
1681	{
1682	u_int16_t mss;
1683	int opt, optlen;
1684
1685	DEBUG_CALL("tcp_dooptions");
1686	DEBUG_ARGS((dfd," tp = %lx cnt=%i \n", (long )tp, cnt));
1687
1688	for (; cnt > 0; cnt -= optlen, cp += optlen)
1689	{
1690	opt = cp[0];
1691	if (opt == TCPOPT_EOL)
1692	break;
1693	if (opt == TCPOPT_NOP)
1694	optlen = 1;
1695	else
1696	{
1697	optlen = cp[1];
1698	if (optlen <= 0)
1699	break;
1700	}
1701	switch (opt)
1702	{
1703	default:
1704	continue;
1705
1706	case TCPOPT_MAXSEG:
1707	if (optlen != TCPOLEN_MAXSEG)
1708	continue;
1709	if (!(ti->ti_flags & TH_SYN))
1710	continue;
1711	memcpy((char ) &mss, (char ) cp + 2, sizeof(mss));
1712	NTOHS(mss);
1713	(void) tcp_mss(pData, tp, mss); /* sets t_maxseg */
1714	break;
1715
1716	#if 0
1717	case TCPOPT_WINDOW:
1718	if (optlen != TCPOLEN_WINDOW)
1719	continue;
1720	if (!(ti->ti_flags & TH_SYN))
1721	continue;
1722	tp->t_flags \|= TF_RCVD_SCALE;
1723	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1724	break;
1725
1726	case TCPOPT_TIMESTAMP:
1727	if (optlen != TCPOLEN_TIMESTAMP)
1728	continue;
1729	*ts_present = 1;
1730	memcpy((char ) ts_val, (char )cp + 2, sizeof(*ts_val));
1731	NTOHL(*ts_val);
1732	memcpy((char ) ts_ecr, (char )cp + 6, sizeof(*ts_ecr));
1733	NTOHL(*ts_ecr);
1734
1735	/*
1736	* A timestamp received in a SYN makes
1737	* it ok to send timestamp requests and replies.
1738	*/
1739	if (ti->ti_flags & TH_SYN)
1740	{
1741	tp->t_flags \|= TF_RCVD_TSTMP;
1742	tp->ts_recent = *ts_val;
1743	tp->ts_recent_age = tcp_now;
1744	}
1745	break;
1746	#endif
1747	}
1748	}
1749	}
1750
1751
1752	/*
1753	* Pull out of band byte out of a segment so
1754	* it doesn't appear in the user's data queue.
1755	* It is still reflected in the segment length for
1756	* sequencing purposes.
1757	*/
1758
1759	#if 0
1760	void
1761	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1762	{
1763	int cnt = ti->ti_urp - 1;
1764
1765	while (cnt >= 0)
1766	{
1767	if (m->m_len > cnt)
1768	{
1769	char *cp = mtod(m, caddr_t) + cnt;
1770	struct tcpcb *tp = sototcpcb(so);
1771
1772	tp->t_iobc = *cp;
1773	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1774	memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1775	m->m_len--;
1776	return;
1777	}
1778	cnt -= m->m_len;
1779	m = m->m_next; /* XXX WRONG! Fix it! */
1780	if (m == 0)
1781	break;
1782	}
1783	panic("tcp_pulloutofband");
1784	}
1785	#endif
1786
1787	/*
1788	* Collect new round-trip time estimate
1789	* and update averages and current timeout.
1790	*/
1791
1792	void
1793	tcp_xmit_timer(PNATState pData, register struct tcpcb *tp, int rtt)
1794	{
1795	register short delta;
1796
1797	DEBUG_CALL("tcp_xmit_timer");
1798	DEBUG_ARG("tp = %lx", (long)tp);
1799	DEBUG_ARG("rtt = %d", rtt);
1800
1801	tcpstat.tcps_rttupdated++;
1802	if (tp->t_srtt != 0)
1803	{
1804	/*
1805	* srtt is stored as fixed point with 3 bits after the
1806	* binary point (i.e., scaled by 8). The following magic
1807	* is equivalent to the smoothing algorithm in rfc793 with
1808	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1809	* point). Adjust rtt to origin 0.
1810	*/
1811	delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1812	if ((tp->t_srtt += delta) <= 0)
1813	tp->t_srtt = 1;
1814	/*
1815	* We accumulate a smoothed rtt variance (actually, a
1816	* smoothed mean difference), then set the retransmit
1817	* timer to smoothed rtt + 4 times the smoothed variance.
1818	* rttvar is stored as fixed point with 2 bits after the
1819	* binary point (scaled by 4). The following is
1820	* equivalent to rfc793 smoothing with an alpha of .75
1821	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1822	* rfc793's wired-in beta.
1823	*/
1824	if (delta < 0)
1825	delta = -delta;
1826	delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1827	if ((tp->t_rttvar += delta) <= 0)
1828	tp->t_rttvar = 1;
1829	}
1830	else
1831	{
1832	/*
1833	* No rtt measurement yet - use the unsmoothed rtt.
1834	* Set the variance to half the rtt (so our first
1835	* retransmit happens at 3*rtt).
1836	*/
1837	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1838	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1839	}
1840	tp->t_rtt = 0;
1841	tp->t_rxtshift = 0;
1842
1843	/*
1844	* the retransmit should happen at rtt + 4 * rttvar.
1845	* Because of the way we do the smoothing, srtt and rttvar
1846	* will each average +1/2 tick of bias. When we compute
1847	* the retransmit timer, we want 1/2 tick of rounding and
1848	* 1 extra tick because of +-1/2 tick uncertainty in the
1849	* firing of the timer. The bias will give us exactly the
1850	* 1.5 tick we need. But, because the bias is
1851	* statistical, we have to test that we don't drop below
1852	* the minimum feasible timer (which is 2 ticks).
1853	*/
1854	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1855	(short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1856
1857	/*
1858	* We received an ack for a packet that wasn't retransmitted;
1859	* it is probably safe to discard any error indications we've
1860	* received recently. This isn't quite right, but close enough
1861	* for now (a route might have failed after we sent a segment,
1862	* and the return path might not be symmetrical).
1863	*/
1864	tp->t_softerror = 0;
1865	}
1866
1867	/*
1868	* Determine a reasonable value for maxseg size.
1869	* If the route is known, check route for mtu.
1870	* If none, use an mss that can be handled on the outgoing
1871	* interface without forcing IP to fragment; if bigger than
1872	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1873	* to utilize large mbufs. If no route is found, route has no mtu,
1874	* or the destination isn't local, use a default, hopefully conservative
1875	* size (usually 512 or the default IP max size, but no more than the mtu
1876	* of the interface), as we can't discover anything about intervening
1877	* gateways or networks. We also initialize the congestion/slow start
1878	* window to be a single segment if the destination isn't local.
1879	* While looking at the routing entry, we also initialize other path-dependent
1880	* parameters from pre-set or cached values in the routing entry.
1881	*/
1882
1883	int
1884	tcp_mss(PNATState pData, register struct tcpcb *tp, u_int offer)
1885	{
1886	struct socket *so = tp->t_socket;
1887	int mss;
1888
1889	DEBUG_CALL("tcp_mss");
1890	DEBUG_ARG("tp = %lx", (long)tp);
1891	DEBUG_ARG("offer = %d", offer);
1892
1893	mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
1894	if (offer)
1895	mss = min(mss, offer);
1896	mss = max(mss, 32);
1897	if (mss < tp->t_maxseg \|\| offer != 0)
1898	tp->t_maxseg = mss;
1899
1900	tp->snd_cwnd = mss;
1901
1902	sbreserve(pData, &so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
1903	sbreserve(pData, &so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
1904
1905	DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1906
1907	return mss;
1908	}

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 27848

Download in other formats: