tcp_input.c@ 17389

Last change on this file since 17389 was 17191, checked in by vboxsync, 16 years ago
spaces
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
82	/*
83	* XXX: tcp_reass() is rather inefficient with its data structures
84	* and should be rewritten (see NetBSD for optimizations). While
85	* doing that it should move to its own file tcp_reass.c.
86	*/
87
88	/*
89	* Call with th==NULL after become established to
90	* force pre-ESTABLISHED data up to user socket.
91	*/
92	if (th == NULL)
93	goto present;
94
95	/*
96	* Limit the number of segments in the reassembly queue to prevent
97	* holding on to too many segments (and thus running out of mbufs).
98	* Make sure to let the missing segment through which caused this
99	* queue. Always keep one global queue entry spare to be able to
100	* process the missing segment.
101	*/
102	if ( th->th_seq != tp->rcv_nxt
103	&& ( tcp_reass_qsize + 1 >= tcp_reass_maxseg
104	\|\| tp->t_segqlen >= tcp_reass_maxqlen))
105	{
106	tcp_reass_overflows++;
107	tcpstat.tcps_rcvmemdrop++;
108	m_freem(pData, m);
109	*tlenp = 0;
110	return (0);
111	}
112
113	/*
114	* Allocate a new queue entry. If we can't, or hit the zone limit
115	* just drop the pkt.
116	*/
117	te = RTMemAlloc(sizeof(struct tseg_qent));
118	if (te == NULL)
119	{
120	tcpstat.tcps_rcvmemdrop++;
121	m_freem(pData, m);
122	*tlenp = 0;
123	return (0);
124	}
125	tp->t_segqlen++;
126	tcp_reass_qsize++;
127
128	/*
129	* Find a segment which begins after this one does.
130	*/
131	LIST_FOREACH(q, &tp->t_segq, tqe_q)
132	{
133	if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
134	break;
135	p = q;
136	}
137
138	/*
139	* If there is a preceding segment, it may provide some of
140	* our data already. If so, drop the data from the incoming
141	* segment. If it provides all of our data, drop us.
142	*/
143	if (p != NULL)
144	{
145	int i;
146	/* conversion to int (in i) handles seq wraparound */
147	i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
148	if (i > 0)
149	{
150	if (i >= *tlenp)
151	{
152	tcpstat.tcps_rcvduppack++;
153	tcpstat.tcps_rcvdupbyte += *tlenp;
154	m_freem(pData, m);
155	RTMemFree(te);
156	tp->t_segqlen--;
157	tcp_reass_qsize--;
158	/*
159	* Try to present any queued data
160	* at the left window edge to the user.
161	* This is needed after the 3-WHS
162	* completes.
163	*/
164	goto present; /* ??? */
165	}
166	m_adj(m, i);
167	*tlenp -= i;
168	th->th_seq += i;
169	}
170	}
171	tcpstat.tcps_rcvoopack++;
172	tcpstat.tcps_rcvoobyte += *tlenp;
173
174	/*
175	* While we overlap succeeding segments trim them or,
176	* if they are completely covered, dequeue them.
177	*/
178	while (q)
179	{
180	int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
181	if (i <= 0)
182	break;
183	if (i < q->tqe_len)
184	{
185	q->tqe_th->th_seq += i;
186	q->tqe_len -= i;
187	m_adj(q->tqe_m, i);
188	break;
189	}
190
191	nq = LIST_NEXT(q, tqe_q);
192	LIST_REMOVE(q, tqe_q);
193	m_freem(pData, q->tqe_m);
194	RTMemFree(q);
195	tp->t_segqlen--;
196	tcp_reass_qsize--;
197	q = nq;
198	}
199
200	/* Insert the new segment queue entry into place. */
201	te->tqe_m = m;
202	te->tqe_th = th;
203	te->tqe_len = *tlenp;
204
205	if (p == NULL)
206	{
207	LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
208	}
209	else
210	{
211	LIST_INSERT_AFTER(p, te, tqe_q);
212	}
213
214	present:
215	/*
216	* Present data to user, advancing rcv_nxt through
217	* completed sequence space.
218	*/
219	if (!TCPS_HAVEESTABLISHED(tp->t_state))
220	return (0);
221	q = LIST_FIRST(&tp->t_segq);
222	if (!q \|\| q->tqe_th->th_seq != tp->rcv_nxt)
223	return (0);
224	do
225	{
226	tp->rcv_nxt += q->tqe_len;
227	flags = q->tqe_th->th_flags & TH_FIN;
228	nq = LIST_NEXT(q, tqe_q);
229	LIST_REMOVE(q, tqe_q);
230	/* XXX: This place should be checked for the same code in
231	* original BSD code for Slirp and current BSD used SS_FCANTRCVMORE
232	*/
233	if (so->so_state & SS_FCANTSENDMORE)
234	m_freem(pData, q->tqe_m);
235	else
236	{
237	if (so->so_emu)
238	{
239	if (tcp_emu(pData, so, q->tqe_m))
240	sbappend(pData, so, q->tqe_m);
241	}
242	else
243	sbappend(pData, so, q->tqe_m);
244	}
245	RTMemFree(q);
246	tp->t_segqlen--;
247	tcp_reass_qsize--;
248	q = nq;
249	}
250	while (q && q->tqe_th->th_seq == tp->rcv_nxt);
251
252	return flags;
253	}
254
255	/*
256	* TCP input routine, follows pages 65-76 of the
257	* protocol specification dated September, 1981 very closely.
258	*/
259	void
260	tcp_input(PNATState pData, register struct mbuf m, int iphlen, struct socket inso)
261	{
262	struct ip save_ip, *ip;
263	register struct tcpiphdr *ti;
264	caddr_t optp = NULL;
265	int optlen = 0;
266	int len, tlen, off;
267	register struct tcpcb *tp = 0;
268	register int tiflags;
269	struct socket *so = 0;
270	int todrop, acked, ourfinisacked, needoutput = 0;
271	/* int dropsocket = 0; */
272	int iss = 0;
273	u_long tiwin;
274	/* int ts_present = 0; */
275
276	DEBUG_CALL("tcp_input");
277	DEBUG_ARGS((dfd," m = %8lx iphlen = %2d inso = %lx\n",
278	(long )m, iphlen, (long )inso ));
279
280	if (inso != NULL)
281	{
282	QSOCKET_LOCK(tcb);
283	SOCKET_LOCK(inso);
284	QSOCKET_UNLOCK(tcb);
285	}
286	/*
287	* If called with m == 0, then we're continuing the connect
288	*/
289	if (m == NULL)
290	{
291	so = inso;
292
293	/* Re-set a few variables */
294	tp = sototcpcb(so);
295	m = so->so_m;
296	so->so_m = 0;
297	ti = so->so_ti;
298	tiwin = ti->ti_win;
299	tiflags = ti->ti_flags;
300
301	goto cont_conn;
302	}
303
304	tcpstat.tcps_rcvtotal++;
305	/*
306	* Get IP and TCP header together in first mbuf.
307	* Note: IP leaves IP header in first mbuf.
308	*/
309	ti = mtod(m, struct tcpiphdr *);
310	if (iphlen > sizeof(struct ip ))
311	{
312	ip_stripoptions(m, (struct mbuf *)0);
313	iphlen = sizeof(struct ip );
314	}
315	/* XXX Check if too short */
316
317
318	/*
319	* Save a copy of the IP header in case we want restore it
320	* for sending an ICMP error message in response.
321	*/
322	ip = mtod(m, struct ip *);
323	save_ip = *ip;
324	save_ip.ip_len+= iphlen;
325
326	/*
327	* Checksum extended TCP header and data.
328	*/
329	tlen = ((struct ip *)ti)->ip_len;
330	memset(ti->ti_x1, 0, 9);
331	ti->ti_len = htons((u_int16_t)tlen);
332	len = sizeof(struct ip ) + tlen;
333	/* keep checksum for ICMP reply
334	* ti->ti_sum = cksum(m, len);
335	* if (ti->ti_sum) { */
336	if (cksum(m, len))
337	{
338	tcpstat.tcps_rcvbadsum++;
339	Log2(("checksum is invalid => drop\n"));
340	goto drop;
341	}
342
343	/*
344	* Check that TCP offset makes sense,
345	* pull out TCP options and adjust length. XXX
346	*/
347	off = ti->ti_off << 2;
348	if ( off < sizeof (struct tcphdr)
349	\|\| off > tlen)
350	{
351	tcpstat.tcps_rcvbadoff++;
352	Log2(("ti_off(tlen(%d)<%d<(tcphdr(%d))) is invalid =>drop\n", tlen, off, sizeof(struct tcphdr)));
353	goto drop;
354	}
355	tlen -= off;
356	ti->ti_len = tlen;
357	if (off > sizeof (struct tcphdr))
358	{
359	optlen = off - sizeof (struct tcphdr);
360	optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
361
362	/*
363	* Do quick retrieval of timestamp options ("options
364	* prediction?"). If timestamp is the only option and it's
365	* formatted as recommended in RFC 1323 appendix A, we
366	* quickly get the values now and not bother calling
367	* tcp_dooptions(), etc.
368	*/
369	#if 0
370	if (( optlen == TCPOLEN_TSTAMP_APPA
371	\|\| ( optlen > TCPOLEN_TSTAMP_APPA
372	&& optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
373	(u_int32_t )optp == htonl(TCPOPT_TSTAMP_HDR) &&
374	(ti->ti_flags & TH_SYN) == 0)
375	{
376	ts_present = 1;
377	ts_val = ntohl((u_int32_t )(optp + 4));
378	ts_ecr = ntohl((u_int32_t )(optp + 8));
379	optp = NULL; / * we have parsed the options * /
380	}
381	#endif
382	}
383	tiflags = ti->ti_flags;
384
385	/*
386	* Convert TCP protocol specific fields to host format.
387	*/
388	NTOHL(ti->ti_seq);
389	NTOHL(ti->ti_ack);
390	NTOHS(ti->ti_win);
391	NTOHS(ti->ti_urp);
392
393	/*
394	* Drop TCP, IP headers and TCP options.
395	*/
396	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
397	m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
398
399	/*
400	* Locate pcb for segment.
401	*/
402	findso:
403	if (so != NULL && so != &tcb)
404	SOCKET_UNLOCK(so);
405	QSOCKET_LOCK(tcb);
406	so = tcp_last_so;
407	if ( so->so_fport != ti->ti_dport
408	\|\| so->so_lport != ti->ti_sport
409	\|\| so->so_laddr.s_addr != ti->ti_src.s_addr
410	\|\| so->so_faddr.s_addr != ti->ti_dst.s_addr)
411	{
412	struct socket *sonxt;
413	QSOCKET_UNLOCK(tcb);
414	/* @todo fix SOLOOKUP macrodefinition to be usable here */
415	#ifndef VBOX_WITH_SLIRP_MT
416	so = solookup(&tcb, ti->ti_src, ti->ti_sport,
417	ti->ti_dst, ti->ti_dport);
418	#else
419	so = NULL;
420	QSOCKET_FOREACH(so, sonxt, tcp)
421	/* { */
422	if ( so->so_lport == ti->ti_sport
423	&& so->so_laddr.s_addr == ti->ti_src.s_addr
424	&& so->so_faddr.s_addr == ti->ti_dst.s_addr
425	&& so->so_fport == ti->ti_dport
426	&& so->so_deleted != 1)
427	{
428	Log2(("lock: %s:%d We found socket %R[natsock]\n", __FUNCTION__, __LINE__, so));
429	break; /* so is locked here */
430	}
431	LOOP_LABEL(tcp, so, sonxt);
432	}
433	if (so == &tcb) {
434	Log2(("lock: %s:%d Haven't find anything \n", __FUNCTION__, __LINE__));
435	so = NULL;
436	}
437	#endif
438	if (so)
439	{
440	tcp_last_so = so;
441	}
442	++tcpstat.tcps_socachemiss;
443	}
444	else
445	{
446	SOCKET_LOCK(so);
447	QSOCKET_UNLOCK(tcb);
448	}
449
450	/*
451	* If the state is CLOSED (i.e., TCB does not exist) then
452	* all data in the incoming segment is discarded.
453	* If the TCB exists but is in CLOSED state, it is embryonic,
454	* but should either do a listen or a connect soon.
455	*
456	* state == CLOSED means we've done socreate() but haven't
457	* attached it to a protocol yet...
458	*
459	* XXX If a TCB does not exist, and the TH_SYN flag is
460	* the only flag set, then create a session, mark it
461	* as if it was LISTENING, and continue...
462	*/
463	Log2(("so = %R[natsock]\n", so));
464	if (so == 0)
465	{
466	if ((tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) != TH_SYN)
467	goto dropwithreset;
468
469	if ((so = socreate()) == NULL)
470	goto dropwithreset;
471	if (tcp_attach(pData, so) < 0)
472	{
473	RTMemFree(so); /* Not sofree (if it failed, it's not insqued) */
474	goto dropwithreset;
475	}
476
477	SOCKET_LOCK(so);
478	sbreserve(&so->so_snd, tcp_sndspace);
479	sbreserve(&so->so_rcv, tcp_rcvspace);
480
481	/* tcp_last_so = so; / / XXX ? */
482	/* tp = sototcpcb(so); */
483
484	so->so_laddr = ti->ti_src;
485	so->so_lport = ti->ti_sport;
486	so->so_faddr = ti->ti_dst;
487	so->so_fport = ti->ti_dport;
488
489	if ((so->so_iptos = tcp_tos(so)) == 0)
490	so->so_iptos = ((struct ip *)ti)->ip_tos;
491
492	tp = sototcpcb(so);
493	tp->t_state = TCPS_LISTEN;
494	}
495
496	/*
497	* If this is a still-connecting socket, this probably
498	* a retransmit of the SYN. Whether it's a retransmit SYN
499	* or something else, we nuke it.
500	*/
501	if (so->so_state & SS_ISFCONNECTING)
502	{
503	Log2(("so_state(%x) of %R[natsock] is still connecting =>drop\n", so->so_state, so));
504	goto drop;
505	}
506
507	tp = sototcpcb(so);
508
509	/* XXX Should never fail */
510	if (tp == 0)
511	goto dropwithreset;
512	if (tp->t_state == TCPS_CLOSED)
513	{
514	Log2(("t_state(%x) is closed =>drop\n", tp->t_state));
515	goto drop;
516	}
517
518	/* Unscale the window into a 32-bit value. */
519	/* if ((tiflags & TH_SYN) == 0)
520	* tiwin = ti->ti_win << tp->snd_scale;
521	* else
522	*/
523	tiwin = ti->ti_win;
524
525	/*
526	* Segment received on connection.
527	* Reset idle time and keep-alive timer.
528	*/
529	tp->t_idle = 0;
530	if (so_options)
531	tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
532	else
533	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
534
535	/*
536	* Process options if not in LISTEN state,
537	* else do it below (after getting remote address).
538	*/
539	if (optp && tp->t_state != TCPS_LISTEN)
540	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
541	/* , */
542	/* &ts_present, &ts_val, &ts_ecr); */
543
544	/*
545	* Header prediction: check for the two common cases
546	* of a uni-directional data xfer. If the packet has
547	* no control flags, is in-sequence, the window didn't
548	* change and we're not retransmitting, it's a
549	* candidate. If the length is zero and the ack moved
550	* forward, we're the sender side of the xfer. Just
551	* free the data acked & wake any higher level process
552	* that was blocked waiting for space. If the length
553	* is non-zero and the ack didn't move, we're the
554	* receiver side. If we're getting packets in-order
555	* (the reassembly queue is empty), add the data to
556	* the socket buffer and note that we need a delayed ack.
557	*
558	* XXX Some of these tests are not needed
559	* eg: the tiwin == tp->snd_wnd prevents many more
560	* predictions.. with no real advantage..
561	*/
562	if ( tp->t_state == TCPS_ESTABLISHED
563	&& (tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK
564	/* && (!ts_present \|\| TSTMP_GEQ(ts_val, tp->ts_recent)) */
565	&& ti->ti_seq == tp->rcv_nxt
566	&& tiwin && tiwin == tp->snd_wnd
567	&& tp->snd_nxt == tp->snd_max)
568	{
569	/*
570	* If last ACK falls within this segment's sequence numbers,
571	* record the timestamp.
572	*/
573	#if 0
574	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
575	SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len))
576	{
577	tp->ts_recent_age = tcp_now;
578	tp->ts_recent = ts_val;
579	}
580	#endif
581
582	if (ti->ti_len == 0)
583	{
584	if ( SEQ_GT(ti->ti_ack, tp->snd_una)
585	&& SEQ_LEQ(ti->ti_ack, tp->snd_max)
586	&& tp->snd_cwnd >= tp->snd_wnd)
587	{
588	/*
589	* this is a pure ack for outstanding data.
590	*/
591	++tcpstat.tcps_predack;
592	#if 0
593	if (ts_present)
594	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
595	else
596	#endif
597	if ( tp->t_rtt
598	&& SEQ_GT(ti->ti_ack, tp->t_rtseq))
599	tcp_xmit_timer(pData, tp, tp->t_rtt);
600	acked = ti->ti_ack - tp->snd_una;
601	tcpstat.tcps_rcvackpack++;
602	tcpstat.tcps_rcvackbyte += acked;
603	sbdrop(&so->so_snd, acked);
604	tp->snd_una = ti->ti_ack;
605	m_freem(pData, m);
606
607	/*
608	* If all outstanding data are acked, stop
609	* retransmit timer, otherwise restart timer
610	* using current (possibly backed-off) value.
611	* If process is waiting for space,
612	* wakeup/selwakeup/signal. If data
613	* are ready to send, let tcp_output
614	* decide between more output or persist.
615	*/
616	if (tp->snd_una == tp->snd_max)
617	tp->t_timer[TCPT_REXMT] = 0;
618	else if (tp->t_timer[TCPT_PERSIST] == 0)
619	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
620
621	/*
622	* There's room in so_snd, sowwakup will read()
623	* from the socket if we can
624	*/
625	#if 0
626	if (so->so_snd.sb_flags & SB_NOTIFY)
627	sowwakeup(so);
628	#endif
629	/*
630	* This is called because sowwakeup might have
631	* put data into so_snd. Since we don't so sowwakeup,
632	* we don't need this.. XXX???
633	*/
634	if (so->so_snd.sb_cc)
635	(void) tcp_output(pData, tp);
636
637	SOCKET_UNLOCK(so);
638	return;
639	}
640	}
641	else if ( ti->ti_ack == tp->snd_una
642	&& LIST_FIRST(&tp->t_segq)
643	&& ti->ti_len <= sbspace(&so->so_rcv))
644	{
645	/*
646	* this is a pure, in-sequence data packet
647	* with nothing on the reassembly queue and
648	* we have enough buffer space to take it.
649	*/
650	++tcpstat.tcps_preddat;
651	tp->rcv_nxt += ti->ti_len;
652	tcpstat.tcps_rcvpack++;
653	tcpstat.tcps_rcvbyte += ti->ti_len;
654	/*
655	* Add data to socket buffer.
656	*/
657	if (so->so_emu)
658	{
659	if (tcp_emu(pData, so, m))
660	sbappend(pData, so, m);
661	}
662	else
663	sbappend(pData, so, m);
664
665	/*
666	* XXX This is called when data arrives. Later, check
667	* if we can actually write() to the socket
668	* XXX Need to check? It's be NON_BLOCKING
669	*/
670	/* sorwakeup(so); */
671
672	/*
673	* If this is a short packet, then ACK now - with Nagel
674	* congestion avoidance sender won't send more until
675	* he gets an ACK.
676	*
677	* It is better to not delay acks at all to maximize
678	* TCP throughput. See RFC 2581.
679	*/
680	tp->t_flags \|= TF_ACKNOW;
681	tcp_output(pData, tp);
682	SOCKET_UNLOCK(so);
683	return;
684	}
685	} /* header prediction */
686	/*
687	* Calculate amount of space in receive window,
688	* and then do TCP input processing.
689	* Receive window is amount of space in rcv queue,
690	* but not less than advertised window.
691	*/
692	{
693	int win;
694	win = sbspace(&so->so_rcv);
695	if (win < 0)
696	win = 0;
697	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
698	}
699
700	switch (tp->t_state)
701	{
702	/*
703	* If the state is LISTEN then ignore segment if it contains an RST.
704	* If the segment contains an ACK then it is bad and send a RST.
705	* If it does not contain a SYN then it is not interesting; drop it.
706	* Don't bother responding if the destination was a broadcast.
707	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
708	* tp->iss, and send a segment:
709	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
710	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
711	* Fill in remote peer address fields if not previously specified.
712	* Enter SYN_RECEIVED state, and process any other fields of this
713	* segment in this state.
714	*/
715	case TCPS_LISTEN:
716	{
717	if (tiflags & TH_RST) {
718	Log2(("RST(%x) is on listen =>drop\n", tiflags));
719	goto drop;
720	}
721	if (tiflags & TH_ACK)
722	goto dropwithreset;
723	if ((tiflags & TH_SYN) == 0)
724	{
725	Log2(("SYN(%x) is off on listen =>drop\n", tiflags));
726	goto drop;
727	}
728
729	/*
730	* This has way too many gotos...
731	* But a bit of spaghetti code never hurt anybody :)
732	*/
733
734	if (so->so_emu & EMU_NOCONNECT)
735	{
736	so->so_emu &= ~EMU_NOCONNECT;
737	goto cont_input;
738	}
739
740	if ( (tcp_fconnect(pData, so) == -1)
741	&& errno != EINPROGRESS
742	&& errno != EWOULDBLOCK)
743	{
744	u_char code = ICMP_UNREACH_NET;
745	DEBUG_MISC((dfd," tcp fconnect errno = %d-%s\n",
746	errno, strerror(errno)));
747	if (errno == ECONNREFUSED)
748	{
749	/* ACK the SYN, send RST to refuse the connection */
750	tcp_respond(pData, tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
751	TH_RST\|TH_ACK);
752	}
753	else
754	{
755	if (errno == EHOSTUNREACH)
756	code = ICMP_UNREACH_HOST;
757	HTONL(ti->ti_seq); /* restore tcp header */
758	HTONL(ti->ti_ack);
759	HTONS(ti->ti_win);
760	HTONS(ti->ti_urp);
761	m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
762	m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
763	*ip = save_ip;
764	icmp_error(pData, m, ICMP_UNREACH, code, 0, strerror(errno));
765	tp->t_socket->so_m = NULL;
766	}
767	tp = tcp_close(pData, tp);
768	m_free(pData, m);
769	}
770	else
771	{
772	/*
773	* Haven't connected yet, save the current mbuf
774	* and ti, and return
775	* XXX Some OS's don't tell us whether the connect()
776	* succeeded or not. So we must time it out.
777	*/
778	so->so_m = m;
779	so->so_ti = ti;
780	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
781	tp->t_state = TCPS_SYN_RECEIVED;
782	}
783	SOCKET_UNLOCK(so);
784	return;
785
786	cont_conn:
787	/* m==NULL
788	* Check if the connect succeeded
789	*/
790	if (so->so_state & SS_NOFDREF)
791	{
792	tp = tcp_close(pData, tp);
793	goto dropwithreset;
794	}
795	cont_input:
796	tcp_template(tp);
797
798	if (optp)
799	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
800
801	if (iss)
802	tp->iss = iss;
803	else
804	tp->iss = tcp_iss;
805	tcp_iss += TCP_ISSINCR/2;
806	tp->irs = ti->ti_seq;
807	tcp_sendseqinit(tp);
808	tcp_rcvseqinit(tp);
809	tp->t_flags \|= TF_ACKNOW;
810	tp->t_state = TCPS_SYN_RECEIVED;
811	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
812	tcpstat.tcps_accepts++;
813	Log2(("hit trimthenstep6\n"));
814	goto trimthenstep6;
815	} /* case TCPS_LISTEN */
816
817	/*
818	* If the state is SYN_SENT:
819	* if seg contains an ACK, but not for our SYN, drop the input.
820	* if seg contains a RST, then drop the connection.
821	* if seg does not contain SYN, then drop it.
822	* Otherwise this is an acceptable SYN segment
823	* initialize tp->rcv_nxt and tp->irs
824	* if seg contains ack then advance tp->snd_una
825	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
826	* arrange for segment to be acked (eventually)
827	* continue processing rest of data/controls, beginning with URG
828	*/
829	case TCPS_SYN_SENT:
830	if ( (tiflags & TH_ACK)
831	&& ( SEQ_LEQ(ti->ti_ack, tp->iss)
832	\|\| SEQ_GT(ti->ti_ack, tp->snd_max)))
833	goto dropwithreset;
834
835	if (tiflags & TH_RST)
836	{
837	if (tiflags & TH_ACK)
838	tp = tcp_drop(pData, tp, 0); /* XXX Check t_softerror! */
839	Log2(("RST(%x) is on SYN_SENT =>drop\n", tiflags));
840	goto drop;
841	}
842
843	if ((tiflags & TH_SYN) == 0)
844	{
845	Log2(("SYN(%x) bit is off on SYN_SENT =>drop\n", tiflags));
846	goto drop;
847	}
848	if (tiflags & TH_ACK)
849	{
850	tp->snd_una = ti->ti_ack;
851	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
852	tp->snd_nxt = tp->snd_una;
853	}
854
855	tp->t_timer[TCPT_REXMT] = 0;
856	tp->irs = ti->ti_seq;
857	tcp_rcvseqinit(tp);
858	tp->t_flags \|= TF_ACKNOW;
859	if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss))
860	{
861	tcpstat.tcps_connects++;
862	soisfconnected(so);
863	tp->t_state = TCPS_ESTABLISHED;
864
865	/* Do window scaling on this connection? */
866	#if 0
867	if (( tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
868	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
869	{
870	tp->snd_scale = tp->requested_s_scale;
871	tp->rcv_scale = tp->request_r_scale;
872	}
873	#endif
874	(void) tcp_reass(pData, tp, (struct tcphdr )0, NULL, (struct mbuf )0);
875	/*
876	* if we didn't have to retransmit the SYN,
877	* use its rtt as our initial srtt & rtt var.
878	*/
879	if (tp->t_rtt)
880	tcp_xmit_timer(pData, tp, tp->t_rtt);
881	}
882	else
883	tp->t_state = TCPS_SYN_RECEIVED;
884
885	trimthenstep6:
886	/*
887	* Advance ti->ti_seq to correspond to first data byte.
888	* If data, trim to stay within window,
889	* dropping FIN if necessary.
890	*/
891	ti->ti_seq++;
892	if (ti->ti_len > tp->rcv_wnd)
893	{
894	todrop = ti->ti_len - tp->rcv_wnd;
895	m_adj(m, -todrop);
896	ti->ti_len = tp->rcv_wnd;
897	tiflags &= ~TH_FIN;
898	tcpstat.tcps_rcvpackafterwin++;
899	tcpstat.tcps_rcvbyteafterwin += todrop;
900	}
901	tp->snd_wl1 = ti->ti_seq - 1;
902	tp->rcv_up = ti->ti_seq;
903	Log2(("hit6"));
904	goto step6;
905	} /* switch tp->t_state */
906	/*
907	* States other than LISTEN or SYN_SENT.
908	* First check timestamp, if present.
909	* Then check that at least some bytes of segment are within
910	* receive window. If segment begins before rcv_nxt,
911	* drop leading data (and SYN); if nothing left, just ack.
912	*
913	* RFC 1323 PAWS: If we have a timestamp reply on this segment
914	* and it's less than ts_recent, drop it.
915	*/
916	#if 0
917	if ( ts_present
918	&& (tiflags & TH_RST) == 0
919	&& tp->ts_recent
920	&& TSTMP_LT(ts_val, tp->ts_recent))
921	{
922	/* Check to see if ts_recent is over 24 days old. */
923	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE)
924	{
925	/*
926	* Invalidate ts_recent. If this segment updates
927	* ts_recent, the age will be reset later and ts_recent
928	* will get a valid value. If it does not, setting
929	* ts_recent to zero will at least satisfy the
930	* requirement that zero be placed in the timestamp
931	* echo reply when ts_recent isn't valid. The
932	* age isn't reset until we get a valid ts_recent
933	* because we don't want out-of-order segments to be
934	* dropped when ts_recent is old.
935	*/
936	tp->ts_recent = 0;
937	}
938	else
939	{
940	tcpstat.tcps_rcvduppack++;
941	tcpstat.tcps_rcvdupbyte += ti->ti_len;
942	tcpstat.tcps_pawsdrop++;
943	goto dropafterack;
944	}
945	}
946	#endif
947
948	todrop = tp->rcv_nxt - ti->ti_seq;
949	if (todrop > 0)
950	{
951	if (tiflags & TH_SYN)
952	{
953	tiflags &= ~TH_SYN;
954	ti->ti_seq++;
955	if (ti->ti_urp > 1)
956	ti->ti_urp--;
957	else
958	tiflags &= ~TH_URG;
959	todrop--;
960	}
961	/*
962	* Following if statement from Stevens, vol. 2, p. 960.
963	*/
964	if ( todrop > ti->ti_len
965	\|\| ( todrop == ti->ti_len
966	&& (tiflags & TH_FIN) == 0))
967	{
968	/*
969	* Any valid FIN must be to the left of the window.
970	* At this point the FIN must be a duplicate or out
971	* of sequence; drop it.
972	*/
973	tiflags &= ~TH_FIN;
974
975	/*
976	* Send an ACK to resynchronize and drop any data.
977	* But keep on processing for RST or ACK.
978	*/
979	tp->t_flags \|= TF_ACKNOW;
980	todrop = ti->ti_len;
981	tcpstat.tcps_rcvduppack++;
982	tcpstat.tcps_rcvdupbyte += todrop;
983	}
984	else
985	{
986	tcpstat.tcps_rcvpartduppack++;
987	tcpstat.tcps_rcvpartdupbyte += todrop;
988	}
989	m_adj(m, todrop);
990	ti->ti_seq += todrop;
991	ti->ti_len -= todrop;
992	if (ti->ti_urp > todrop)
993	ti->ti_urp -= todrop;
994	else
995	{
996	tiflags &= ~TH_URG;
997	ti->ti_urp = 0;
998	}
999	}
1000	/*
1001	* If new data are received on a connection after the
1002	* user processes are gone, then RST the other end.
1003	*/
1004	if ( (so->so_state & SS_NOFDREF)
1005	&& tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len)
1006	{
1007	tp = tcp_close(pData, tp);
1008	tcpstat.tcps_rcvafterclose++;
1009	goto dropwithreset;
1010	}
1011
1012	/*
1013	* If segment ends after window, drop trailing data
1014	* (and PUSH and FIN); if nothing left, just ACK.
1015	*/
1016	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1017	if (todrop > 0)
1018	{
1019	tcpstat.tcps_rcvpackafterwin++;
1020	if (todrop >= ti->ti_len)
1021	{
1022	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1023	/*
1024	* If a new connection request is received
1025	* while in TIME_WAIT, drop the old connection
1026	* and start over if the sequence numbers
1027	* are above the previous ones.
1028	*/
1029	if ( tiflags & TH_SYN
1030	&& tp->t_state == TCPS_TIME_WAIT
1031	&& SEQ_GT(ti->ti_seq, tp->rcv_nxt))
1032	{
1033	iss = tp->rcv_nxt + TCP_ISSINCR;
1034	tp = tcp_close(pData, tp);
1035	SOCKET_UNLOCK(tp->t_socket);
1036	goto findso;
1037	}
1038	/*
1039	* If window is closed can only take segments at
1040	* window edge, and have to drop data and PUSH from
1041	* incoming segments. Continue processing, but
1042	* remember to ack. Otherwise, drop segment
1043	* and ack.
1044	*/
1045	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt)
1046	{
1047	tp->t_flags \|= TF_ACKNOW;
1048	tcpstat.tcps_rcvwinprobe++;
1049	}
1050	else
1051	goto dropafterack;
1052	}
1053	else
1054	tcpstat.tcps_rcvbyteafterwin += todrop;
1055	m_adj(m, -todrop);
1056	ti->ti_len -= todrop;
1057	tiflags &= ~(TH_PUSH\|TH_FIN);
1058	}
1059
1060	/*
1061	* If last ACK falls within this segment's sequence numbers,
1062	* record its timestamp.
1063	*/
1064	#if 0
1065	if ( ts_present
1066	&& SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)
1067	&& SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + ((tiflags & (TH_SYN\|TH_FIN)) != 0)))
1068	{
1069	tp->ts_recent_age = tcp_now;
1070	tp->ts_recent = ts_val;
1071	}
1072	#endif
1073
1074	/*
1075	* If the RST bit is set examine the state:
1076	* SYN_RECEIVED STATE:
1077	* If passive open, return to LISTEN state.
1078	* If active open, inform user that connection was refused.
1079	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1080	* Inform user that connection was reset, and close tcb.
1081	* CLOSING, LAST_ACK, TIME_WAIT STATES
1082	* Close the tcb.
1083	*/
1084	if (tiflags&TH_RST)
1085	switch (tp->t_state)
1086	{
1087	case TCPS_SYN_RECEIVED:
1088	/* so->so_error = ECONNREFUSED; */
1089	goto close;
1090
1091	case TCPS_ESTABLISHED:
1092	case TCPS_FIN_WAIT_1:
1093	case TCPS_FIN_WAIT_2:
1094	case TCPS_CLOSE_WAIT:
1095	/* so->so_error = ECONNRESET; */
1096	close:
1097	Log2(("closing...=>drop\n", tp->t_state));
1098	tp->t_state = TCPS_CLOSED;
1099	tcpstat.tcps_drops++;
1100	tp = tcp_close(pData, tp);
1101	goto drop;
1102
1103	case TCPS_CLOSING:
1104	case TCPS_LAST_ACK:
1105	case TCPS_TIME_WAIT:
1106	Log2(("t_state is (%x) sort of close =>drop\n", tp->t_state));
1107	tp = tcp_close(pData, tp);
1108	goto drop;
1109	}
1110
1111	/*
1112	* If a SYN is in the window, then this is an
1113	* error and we send an RST and drop the connection.
1114	*/
1115	if (tiflags & TH_SYN)
1116	{
1117	tp = tcp_drop(pData, tp, 0);
1118	goto dropwithreset;
1119	}
1120
1121	/*
1122	* If the ACK bit is off we drop the segment and return.
1123	*/
1124	if ((tiflags & TH_ACK) == 0)
1125	{
1126	Log2(("ACK(%x) bit is off =>drop\n", tiflags));
1127	goto drop;
1128	}
1129
1130	/*
1131	* Ack processing.
1132	*/
1133	switch (tp->t_state)
1134	{
1135	/*
1136	* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1137	* ESTABLISHED state and continue processing, otherwise
1138	* send an RST. una<=ack<=max
1139	*/
1140	case TCPS_SYN_RECEIVED:
1141	if ( SEQ_GT(tp->snd_una, ti->ti_ack)
1142	\|\| SEQ_GT(ti->ti_ack, tp->snd_max))
1143	goto dropwithreset;
1144	tcpstat.tcps_connects++;
1145	tp->t_state = TCPS_ESTABLISHED;
1146	/*
1147	* The sent SYN is ack'ed with our sequence number +1
1148	* The first data byte already in the buffer will get
1149	* lost if no correction is made. This is only needed for
1150	* SS_CTL since the buffer is empty otherwise.
1151	* tp->snd_una++; or:
1152	*/
1153	tp->snd_una = ti->ti_ack;
1154	soisfconnected(so);
1155
1156	/* Do window scaling? */
1157	#if 0
1158	if ( (tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
1159	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
1160	{
1161	tp->snd_scale = tp->requested_s_scale;
1162	tp->rcv_scale = tp->request_r_scale;
1163	}
1164	#endif
1165	(void) tcp_reass(pData, tp, (struct tcphdr )0, (int )0, (struct mbuf *)0);
1166	tp->snd_wl1 = ti->ti_seq - 1;
1167	/* Avoid ack processing; snd_una==ti_ack => dup ack */
1168	Log2(("hit synrx_to_est\n"));
1169	goto synrx_to_est;
1170	/* fall into ... */
1171
1172	/*
1173	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1174	* ACKs. If the ack is in the range
1175	* tp->snd_una < ti->ti_ack <= tp->snd_max
1176	* then advance tp->snd_una to ti->ti_ack and drop
1177	* data from the retransmission queue. If this ACK reflects
1178	* more up to date window information we update our window information.
1179	*/
1180	case TCPS_ESTABLISHED:
1181	case TCPS_FIN_WAIT_1:
1182	case TCPS_FIN_WAIT_2:
1183	case TCPS_CLOSE_WAIT:
1184	case TCPS_CLOSING:
1185	case TCPS_LAST_ACK:
1186	case TCPS_TIME_WAIT:
1187	if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
1188	{
1189	if (ti->ti_len == 0 && tiwin == tp->snd_wnd)
1190	{
1191	tcpstat.tcps_rcvdupack++;
1192	DEBUG_MISC((dfd," dup ack m = %lx so = %lx \n",
1193	(long )m, (long )so));
1194	/*
1195	* If we have outstanding data (other than
1196	* a window probe), this is a completely
1197	* duplicate ack (ie, window info didn't
1198	* change), the ack is the biggest we've
1199	* seen and we've seen exactly our rexmt
1200	* threshold of them, assume a packet
1201	* has been dropped and retransmit it.
1202	* Kludge snd_nxt & the congestion
1203	* window so we send only this one
1204	* packet.
1205	*
1206	* We know we're losing at the current
1207	* window size so do congestion avoidance
1208	* (set ssthresh to half the current window
1209	* and pull our congestion window back to
1210	* the new ssthresh).
1211	*
1212	* Dup acks mean that packets have left the
1213	* network (they're now cached at the receiver)
1214	* so bump cwnd by the amount in the receiver
1215	* to keep a constant cwnd packets in the
1216	* network.
1217	*/
1218	if ( tp->t_timer[TCPT_REXMT] == 0
1219	\|\| ti->ti_ack != tp->snd_una)
1220	tp->t_dupacks = 0;
1221	else if (++tp->t_dupacks == tcprexmtthresh)
1222	{
1223	tcp_seq onxt = tp->snd_nxt;
1224	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
1225	if (win < 2)
1226	win = 2;
1227	tp->snd_ssthresh = win * tp->t_maxseg;
1228	tp->t_timer[TCPT_REXMT] = 0;
1229	tp->t_rtt = 0;
1230	tp->snd_nxt = ti->ti_ack;
1231	tp->snd_cwnd = tp->t_maxseg;
1232	(void) tcp_output(pData, tp);
1233	tp->snd_cwnd = tp->snd_ssthresh +
1234	tp->t_maxseg * tp->t_dupacks;
1235	if (SEQ_GT(onxt, tp->snd_nxt))
1236	tp->snd_nxt = onxt;
1237	Log2(("t_dupacks(%d) == tcprexmtthresh(%d)=>drop\n", tp->t_dupacks, tcprexmtthresh));
1238	goto drop;
1239	}
1240	else if (tp->t_dupacks > tcprexmtthresh)
1241	{
1242	tp->snd_cwnd += tp->t_maxseg;
1243	(void) tcp_output(pData, tp);
1244	Log2(("t_dupacks(%d) > tcprexmtthresh(%d)=>drop\n", tp->t_dupacks, tcprexmtthresh));
1245	goto drop;
1246	}
1247	}
1248	else
1249	tp->t_dupacks = 0;
1250	break;
1251	}
1252	synrx_to_est:
1253	Log2(("enter synrx_to_est\n"));
1254	/*
1255	* If the congestion window was inflated to account
1256	* for the other side's cached packets, retract it.
1257	*/
1258	if ( tp->t_dupacks > tcprexmtthresh
1259	&& tp->snd_cwnd > tp->snd_ssthresh)
1260	tp->snd_cwnd = tp->snd_ssthresh;
1261	tp->t_dupacks = 0;
1262	if (SEQ_GT(ti->ti_ack, tp->snd_max))
1263	{
1264	tcpstat.tcps_rcvacktoomuch++;
1265	goto dropafterack;
1266	}
1267	acked = ti->ti_ack - tp->snd_una;
1268	tcpstat.tcps_rcvackpack++;
1269	tcpstat.tcps_rcvackbyte += acked;
1270
1271	/*
1272	* If we have a timestamp reply, update smoothed
1273	* round trip time. If no timestamp is present but
1274	* transmit timer is running and timed sequence
1275	* number was acked, update smoothed round trip time.
1276	* Since we now have an rtt measurement, cancel the
1277	* timer backoff (cf., Phil Karn's retransmit alg.).
1278	* Recompute the initial retransmit timer.
1279	*/
1280	#if 0
1281	if (ts_present)
1282	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1283	else
1284	#endif
1285	if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1286	tcp_xmit_timer(pData, tp, tp->t_rtt);
1287
1288	/*
1289	* If all outstanding data is acked, stop retransmit
1290	* timer and remember to restart (more output or persist).
1291	* If there is more data to be acked, restart retransmit
1292	* timer, using current (possibly backed-off) value.
1293	*/
1294	if (ti->ti_ack == tp->snd_max)
1295	{
1296	tp->t_timer[TCPT_REXMT] = 0;
1297	needoutput = 1;
1298	}
1299	else if (tp->t_timer[TCPT_PERSIST] == 0)
1300	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1301	/*
1302	* When new data is acked, open the congestion window.
1303	* If the window gives us less than ssthresh packets
1304	* in flight, open exponentially (maxseg per packet).
1305	* Otherwise open linearly: maxseg per window
1306	* (maxseg^2 / cwnd per packet).
1307	*/
1308	{
1309	register u_int cw = tp->snd_cwnd;
1310	register u_int incr = tp->t_maxseg;
1311
1312	if (cw > tp->snd_ssthresh)
1313	incr = incr * incr / cw;
1314	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1315	}
1316	if (acked > so->so_snd.sb_cc)
1317	{
1318	tp->snd_wnd -= so->so_snd.sb_cc;
1319	sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1320	ourfinisacked = 1;
1321	}
1322	else
1323	{
1324	sbdrop(&so->so_snd, acked);
1325	tp->snd_wnd -= acked;
1326	ourfinisacked = 0;
1327	}
1328	/*
1329	* XXX sowwakup is called when data is acked and there's room for
1330	* for more data... it should read() the socket
1331	*/
1332	#if 0
1333	if (so->so_snd.sb_flags & SB_NOTIFY)
1334	sowwakeup(so);
1335	#endif
1336	tp->snd_una = ti->ti_ack;
1337	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1338	tp->snd_nxt = tp->snd_una;
1339
1340	switch (tp->t_state)
1341	{
1342	/*
1343	* In FIN_WAIT_1 STATE in addition to the processing
1344	* for the ESTABLISHED state if our FIN is now acknowledged
1345	* then enter FIN_WAIT_2.
1346	*/
1347	case TCPS_FIN_WAIT_1:
1348	if (ourfinisacked)
1349	{
1350	/*
1351	* If we can't receive any more
1352	* data, then closing user can proceed.
1353	* Starting the timer is contrary to the
1354	* specification, but if we don't get a FIN
1355	* we'll hang forever.
1356	*/
1357	if (so->so_state & SS_FCANTRCVMORE)
1358	{
1359	soisfdisconnected(so);
1360	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1361	}
1362	tp->t_state = TCPS_FIN_WAIT_2;
1363	}
1364	break;
1365
1366	/*
1367	* In CLOSING STATE in addition to the processing for
1368	* the ESTABLISHED state if the ACK acknowledges our FIN
1369	* then enter the TIME-WAIT state, otherwise ignore
1370	* the segment.
1371	*/
1372	case TCPS_CLOSING:
1373	if (ourfinisacked)
1374	{
1375	tp->t_state = TCPS_TIME_WAIT;
1376	tcp_canceltimers(tp);
1377	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1378	soisfdisconnected(so);
1379	}
1380	break;
1381
1382	/*
1383	* In LAST_ACK, we may still be waiting for data to drain
1384	* and/or to be acked, as well as for the ack of our FIN.
1385	* If our FIN is now acknowledged, delete the TCB,
1386	* enter the closed state and return.
1387	*/
1388	case TCPS_LAST_ACK:
1389	if (ourfinisacked)
1390	{
1391	Log2(("ourfinisacked=>drop\n"));
1392	tp = tcp_close(pData, tp);
1393	goto drop;
1394	}
1395	break;
1396
1397	/*
1398	* In TIME_WAIT state the only thing that should arrive
1399	* is a retransmission of the remote FIN. Acknowledge
1400	* it and restart the finack timer.
1401	*/
1402	case TCPS_TIME_WAIT:
1403	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1404	goto dropafterack;
1405	}
1406	} /* switch(tp->t_state) */
1407
1408	step6:
1409	/*
1410	* Update window information.
1411	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1412	*/
1413	if ( (tiflags & TH_ACK)
1414	&& ( SEQ_LT(tp->snd_wl1, ti->ti_seq)
1415	\|\| ( tp->snd_wl1 == ti->ti_seq
1416	&& ( SEQ_LT(tp->snd_wl2, ti->ti_ack)
1417	\|\| ( tp->snd_wl2 == ti->ti_ack
1418	&& tiwin > tp->snd_wnd)))))
1419	{
1420	/* keep track of pure window updates */
1421	if ( ti->ti_len == 0
1422	&& tp->snd_wl2 == ti->ti_ack
1423	&& tiwin > tp->snd_wnd)
1424	tcpstat.tcps_rcvwinupd++;
1425	tp->snd_wnd = tiwin;
1426	tp->snd_wl1 = ti->ti_seq;
1427	tp->snd_wl2 = ti->ti_ack;
1428	if (tp->snd_wnd > tp->max_sndwnd)
1429	tp->max_sndwnd = tp->snd_wnd;
1430	needoutput = 1;
1431	}
1432
1433	/*
1434	* Process segments with URG.
1435	*/
1436	if ((tiflags & TH_URG) && ti->ti_urp &&
1437	TCPS_HAVERCVDFIN(tp->t_state) == 0)
1438	{
1439	/*
1440	* This is a kludge, but if we receive and accept
1441	* random urgent pointers, we'll crash in
1442	* soreceive. It's hard to imagine someone
1443	* actually wanting to send this much urgent data.
1444	*/
1445	if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen)
1446	{
1447	ti->ti_urp = 0;
1448	tiflags &= ~TH_URG;
1449	goto dodata;
1450	}
1451	/*
1452	* If this segment advances the known urgent pointer,
1453	* then mark the data stream. This should not happen
1454	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1455	* a FIN has been received from the remote side.
1456	* In these states we ignore the URG.
1457	*
1458	* According to RFC961 (Assigned Protocols),
1459	* the urgent pointer points to the last octet
1460	* of urgent data. We continue, however,
1461	* to consider it to indicate the first octet
1462	* of data past the urgent section as the original
1463	* spec states (in one of two places).
1464	*/
1465	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up))
1466	{
1467	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1468	so->so_urgc = so->so_rcv.sb_cc +
1469	(tp->rcv_up - tp->rcv_nxt); /* -1; */
1470	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1471	}
1472	}
1473	else
1474	/*
1475	* If no out of band data is expected,
1476	* pull receive urgent pointer along
1477	* with the receive window.
1478	*/
1479	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1480	tp->rcv_up = tp->rcv_nxt;
1481	dodata:
1482	Log2(("do data hit!\n"));
1483
1484	/*
1485	* If this is a small packet, then ACK now - with Nagel
1486	* congestion avoidance sender won't send more until
1487	* he gets an ACK.
1488	*
1489	* See above.
1490	*/
1491	if ( ti->ti_len
1492	&& (unsigned)ti->ti_len <= 5
1493	&& ((struct tcpiphdr_2 *)ti)->first_char == (char)27)
1494	{
1495	tp->t_flags \|= TF_ACKNOW;
1496	}
1497
1498	/*
1499	* Process the segment text, merging it into the TCP sequencing queue,
1500	* and arranging for acknowledgment of receipt if necessary.
1501	* This process logically involves adjusting tp->rcv_wnd as data
1502	* is presented to the user (this happens in tcp_usrreq.c,
1503	* case PRU_RCVD). If a FIN has already been received on this
1504	* connection then we just ignore the text.
1505	*/
1506	if ( (ti->ti_len \|\| (tiflags&TH_FIN))
1507	&& TCPS_HAVERCVDFIN(tp->t_state) == 0)
1508	{
1509	if ( ti->ti_seq == tp->rcv_nxt
1510	&& LIST_EMPTY(&tp->t_segq)
1511	&& tp->t_state == TCPS_ESTABLISHED)
1512	{
1513	DELAY_ACK(tp, ti); /* little bit different from BSD declaration see netinet/tcp_input.c */
1514	tp->rcv_nxt += tlen;
1515	tiflags = ti->ti_t.th_flags & TH_FIN;
1516	tcpstat.tcps_rcvpack++;
1517	tcpstat.tcps_rcvbyte += tlen;
1518	if (so->so_state & SS_FCANTRCVMORE)
1519	m_freem(pData, m);
1520	else
1521	{
1522	if (so->so_emu)
1523	{
1524	if (tcp_emu(pData, so, m))
1525	sbappend(pData, so, m);
1526	}
1527	else
1528	sbappend(pData, so, m);
1529	}
1530	}
1531	else
1532	{
1533	tiflags = tcp_reass(pData, tp, &ti->ti_t, &tlen, m);
1534	tiflags \|= TF_ACKNOW;
1535	}
1536	/*
1537	* Note the amount of data that peer has sent into
1538	* our window, in order to estimate the sender's
1539	* buffer size.
1540	*/
1541	len = so->so_rcv.sb_datalen - (tp->rcv_adv - tp->rcv_nxt);
1542	}
1543	else
1544	{
1545	m_free(pData, m);
1546	tiflags &= ~TH_FIN;
1547	}
1548
1549	/*
1550	* If FIN is received ACK the FIN and let the user know
1551	* that the connection is closing.
1552	*/
1553	if (tiflags & TH_FIN)
1554	{
1555	if (TCPS_HAVERCVDFIN(tp->t_state) == 0)
1556	{
1557	/*
1558	* If we receive a FIN we can't send more data,
1559	* set it SS_FDRAIN
1560	* Shutdown the socket if there is no rx data in the
1561	* buffer.
1562	* soread() is called on completion of shutdown() and
1563	* will got to TCPS_LAST_ACK, and use tcp_output()
1564	* to send the FIN.
1565	*/
1566	/* sofcantrcvmore(so); */
1567	sofwdrain(so);
1568
1569	tp->t_flags \|= TF_ACKNOW;
1570	tp->rcv_nxt++;
1571	}
1572	switch (tp->t_state)
1573	{
1574	/*
1575	* In SYN_RECEIVED and ESTABLISHED STATES
1576	* enter the CLOSE_WAIT state.
1577	*/
1578	case TCPS_SYN_RECEIVED:
1579	case TCPS_ESTABLISHED:
1580	if(so->so_emu == EMU_CTL) /* no shutdown on socket */
1581	tp->t_state = TCPS_LAST_ACK;
1582	else
1583	tp->t_state = TCPS_CLOSE_WAIT;
1584	break;
1585
1586	/*
1587	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1588	* enter the CLOSING state.
1589	*/
1590	case TCPS_FIN_WAIT_1:
1591	tp->t_state = TCPS_CLOSING;
1592	break;
1593
1594	/*
1595	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1596	* starting the time-wait timer, turning off the other
1597	* standard timers.
1598	*/
1599	case TCPS_FIN_WAIT_2:
1600	tp->t_state = TCPS_TIME_WAIT;
1601	tcp_canceltimers(tp);
1602	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1603	soisfdisconnected(so);
1604	break;
1605
1606	/*
1607	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1608	*/
1609	case TCPS_TIME_WAIT:
1610	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1611	break;
1612	}
1613	}
1614
1615	/*
1616	* Return any desired output.
1617	*/
1618	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1619	tcp_output(pData, tp);
1620
1621	SOCKET_UNLOCK(so);
1622	return;
1623
1624	dropafterack:
1625	Log2(("drop after ack\n"));
1626	/*
1627	* Generate an ACK dropping incoming segment if it occupies
1628	* sequence space, where the ACK reflects our state.
1629	*/
1630	if (tiflags & TH_RST)
1631	goto drop;
1632	m_freem(pData, m);
1633	tp->t_flags \|= TF_ACKNOW;
1634	(void) tcp_output(pData, tp);
1635	SOCKET_UNLOCK(so);
1636	return;
1637
1638	dropwithreset:
1639	/* reuses m if m!=NULL, m_free() unnecessary */
1640	if (tiflags & TH_ACK)
1641	tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1642	else
1643	{
1644	if (tiflags & TH_SYN) ti->ti_len++;
1645	tcp_respond(pData, tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1646	TH_RST\|TH_ACK);
1647	}
1648
1649	if (so != &tcb)
1650	SOCKET_UNLOCK(so);
1651	return;
1652
1653	drop:
1654	/*
1655	* Drop space held by incoming segment and return.
1656	*/
1657	m_free(pData, m);
1658
1659	#ifdef VBOX_WITH_SLIRP_MT
1660	if (RTCritSectIsOwned(&so->so_mutex))
1661	{
1662	SOCKET_UNLOCK(so);
1663	}
1664	#endif
1665
1666	return;
1667	}
1668
1669	void
1670	tcp_dooptions(PNATState pData, struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti)
1671	{
1672	u_int16_t mss;
1673	int opt, optlen;
1674
1675	DEBUG_CALL("tcp_dooptions");
1676	DEBUG_ARGS((dfd," tp = %lx cnt=%i \n", (long )tp, cnt));
1677
1678	for (; cnt > 0; cnt -= optlen, cp += optlen)
1679	{
1680	opt = cp[0];
1681	if (opt == TCPOPT_EOL)
1682	break;
1683	if (opt == TCPOPT_NOP)
1684	optlen = 1;
1685	else
1686	{
1687	optlen = cp[1];
1688	if (optlen <= 0)
1689	break;
1690	}
1691	switch (opt)
1692	{
1693	default:
1694	continue;
1695
1696	case TCPOPT_MAXSEG:
1697	if (optlen != TCPOLEN_MAXSEG)
1698	continue;
1699	if (!(ti->ti_flags & TH_SYN))
1700	continue;
1701	memcpy((char ) &mss, (char ) cp + 2, sizeof(mss));
1702	NTOHS(mss);
1703	(void) tcp_mss(pData, tp, mss); /* sets t_maxseg */
1704	break;
1705
1706	#if 0
1707	case TCPOPT_WINDOW:
1708	if (optlen != TCPOLEN_WINDOW)
1709	continue;
1710	if (!(ti->ti_flags & TH_SYN))
1711	continue;
1712	tp->t_flags \|= TF_RCVD_SCALE;
1713	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1714	break;
1715
1716	case TCPOPT_TIMESTAMP:
1717	if (optlen != TCPOLEN_TIMESTAMP)
1718	continue;
1719	*ts_present = 1;
1720	memcpy((char ) ts_val, (char )cp + 2, sizeof(*ts_val));
1721	NTOHL(*ts_val);
1722	memcpy((char ) ts_ecr, (char )cp + 6, sizeof(*ts_ecr));
1723	NTOHL(*ts_ecr);
1724
1725	/*
1726	* A timestamp received in a SYN makes
1727	* it ok to send timestamp requests and replies.
1728	*/
1729	if (ti->ti_flags & TH_SYN)
1730	{
1731	tp->t_flags \|= TF_RCVD_TSTMP;
1732	tp->ts_recent = *ts_val;
1733	tp->ts_recent_age = tcp_now;
1734	}
1735	break;
1736	#endif
1737	}
1738	}
1739	}
1740
1741
1742	/*
1743	* Pull out of band byte out of a segment so
1744	* it doesn't appear in the user's data queue.
1745	* It is still reflected in the segment length for
1746	* sequencing purposes.
1747	*/
1748
1749	#if 0
1750	void
1751	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1752	{
1753	int cnt = ti->ti_urp - 1;
1754
1755	while (cnt >= 0)
1756	{
1757	if (m->m_len > cnt)
1758	{
1759	char *cp = mtod(m, caddr_t) + cnt;
1760	struct tcpcb *tp = sototcpcb(so);
1761
1762	tp->t_iobc = *cp;
1763	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1764	memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1765	m->m_len--;
1766	return;
1767	}
1768	cnt -= m->m_len;
1769	m = m->m_next; /* XXX WRONG! Fix it! */
1770	if (m == 0)
1771	break;
1772	}
1773	panic("tcp_pulloutofband");
1774	}
1775	#endif
1776
1777	/*
1778	* Collect new round-trip time estimate
1779	* and update averages and current timeout.
1780	*/
1781
1782	void
1783	tcp_xmit_timer(PNATState pData, register struct tcpcb *tp, int rtt)
1784	{
1785	register short delta;
1786
1787	DEBUG_CALL("tcp_xmit_timer");
1788	DEBUG_ARG("tp = %lx", (long)tp);
1789	DEBUG_ARG("rtt = %d", rtt);
1790
1791	tcpstat.tcps_rttupdated++;
1792	if (tp->t_srtt != 0)
1793	{
1794	/*
1795	* srtt is stored as fixed point with 3 bits after the
1796	* binary point (i.e., scaled by 8). The following magic
1797	* is equivalent to the smoothing algorithm in rfc793 with
1798	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1799	* point). Adjust rtt to origin 0.
1800	*/
1801	delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1802	if ((tp->t_srtt += delta) <= 0)
1803	tp->t_srtt = 1;
1804	/*
1805	* We accumulate a smoothed rtt variance (actually, a
1806	* smoothed mean difference), then set the retransmit
1807	* timer to smoothed rtt + 4 times the smoothed variance.
1808	* rttvar is stored as fixed point with 2 bits after the
1809	* binary point (scaled by 4). The following is
1810	* equivalent to rfc793 smoothing with an alpha of .75
1811	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1812	* rfc793's wired-in beta.
1813	*/
1814	if (delta < 0)
1815	delta = -delta;
1816	delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1817	if ((tp->t_rttvar += delta) <= 0)
1818	tp->t_rttvar = 1;
1819	}
1820	else
1821	{
1822	/*
1823	* No rtt measurement yet - use the unsmoothed rtt.
1824	* Set the variance to half the rtt (so our first
1825	* retransmit happens at 3*rtt).
1826	*/
1827	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1828	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1829	}
1830	tp->t_rtt = 0;
1831	tp->t_rxtshift = 0;
1832
1833	/*
1834	* the retransmit should happen at rtt + 4 * rttvar.
1835	* Because of the way we do the smoothing, srtt and rttvar
1836	* will each average +1/2 tick of bias. When we compute
1837	* the retransmit timer, we want 1/2 tick of rounding and
1838	* 1 extra tick because of +-1/2 tick uncertainty in the
1839	* firing of the timer. The bias will give us exactly the
1840	* 1.5 tick we need. But, because the bias is
1841	* statistical, we have to test that we don't drop below
1842	* the minimum feasible timer (which is 2 ticks).
1843	*/
1844	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1845	(short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1846
1847	/*
1848	* We received an ack for a packet that wasn't retransmitted;
1849	* it is probably safe to discard any error indications we've
1850	* received recently. This isn't quite right, but close enough
1851	* for now (a route might have failed after we sent a segment,
1852	* and the return path might not be symmetrical).
1853	*/
1854	tp->t_softerror = 0;
1855	}
1856
1857	/*
1858	* Determine a reasonable value for maxseg size.
1859	* If the route is known, check route for mtu.
1860	* If none, use an mss that can be handled on the outgoing
1861	* interface without forcing IP to fragment; if bigger than
1862	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1863	* to utilize large mbufs. If no route is found, route has no mtu,
1864	* or the destination isn't local, use a default, hopefully conservative
1865	* size (usually 512 or the default IP max size, but no more than the mtu
1866	* of the interface), as we can't discover anything about intervening
1867	* gateways or networks. We also initialize the congestion/slow start
1868	* window to be a single segment if the destination isn't local.
1869	* While looking at the routing entry, we also initialize other path-dependent
1870	* parameters from pre-set or cached values in the routing entry.
1871	*/
1872
1873	int
1874	tcp_mss(PNATState pData, register struct tcpcb *tp, u_int offer)
1875	{
1876	struct socket *so = tp->t_socket;
1877	int mss;
1878
1879	DEBUG_CALL("tcp_mss");
1880	DEBUG_ARG("tp = %lx", (long)tp);
1881	DEBUG_ARG("offer = %d", offer);
1882
1883	mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
1884	if (offer)
1885	mss = min(mss, offer);
1886	mss = max(mss, 32);
1887	if (mss < tp->t_maxseg \|\| offer != 0)
1888	tp->t_maxseg = mss;
1889
1890	tp->snd_cwnd = mss;
1891
1892	sbreserve(&so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
1893	sbreserve(&so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
1894
1895	DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1896
1897	return mss;
1898	}

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 17389

Download in other formats: