tcp_input.c@ 20383

Last change on this file since 20383 was 20377, checked in by vboxsync, 16 years ago
NAT: sbuf perfomance counters
Property svn:eol-style set to `native`
File size: 60.4 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	Log4(("NAT: tcp_input: %R[natsock]\n", so));
293	/* Re-set a few variables */
294	tp = sototcpcb(so);
295	m = so->so_m;
296
297	so->so_m = 0;
298	ti = so->so_ti;
299	/* @todo (r -vvl) clarify why it might happens */
300	if (ti == NULL)
301	{
302	LogRel(("NAT: ti is null. can't do any reseting connection actions\n"));
303	/* mbuf should be cleared in sofree called from tcp_close */
304	tcp_close(pData, tp);
305	return;
306	}
307	tiwin = ti->ti_win;
308	tiflags = ti->ti_flags;
309
310	goto cont_conn;
311	}
312
313	tcpstat.tcps_rcvtotal++;
314	/*
315	* Get IP and TCP header together in first mbuf.
316	* Note: IP leaves IP header in first mbuf.
317	*/
318	ti = mtod(m, struct tcpiphdr *);
319	if (iphlen > sizeof(struct ip ))
320	{
321	ip_stripoptions(m, (struct mbuf *)0);
322	iphlen = sizeof(struct ip );
323	}
324	/* XXX Check if too short */
325
326
327	/*
328	* Save a copy of the IP header in case we want restore it
329	* for sending an ICMP error message in response.
330	*/
331	ip = mtod(m, struct ip *);
332	save_ip = *ip;
333	save_ip.ip_len+= iphlen;
334
335	/*
336	* Checksum extended TCP header and data.
337	*/
338	tlen = ((struct ip *)ti)->ip_len;
339	memset(ti->ti_x1, 0, 9);
340	ti->ti_len = htons((u_int16_t)tlen);
341	len = sizeof(struct ip ) + tlen;
342	/* keep checksum for ICMP reply
343	* ti->ti_sum = cksum(m, len);
344	* if (ti->ti_sum) { */
345	if (cksum(m, len))
346	{
347	tcpstat.tcps_rcvbadsum++;
348	goto drop;
349	}
350
351	/*
352	* Check that TCP offset makes sense,
353	* pull out TCP options and adjust length. XXX
354	*/
355	off = ti->ti_off << 2;
356	if ( off < sizeof (struct tcphdr)
357	\|\| off > tlen)
358	{
359	tcpstat.tcps_rcvbadoff++;
360	goto drop;
361	}
362	tlen -= off;
363	ti->ti_len = tlen;
364	if (off > sizeof (struct tcphdr))
365	{
366	optlen = off - sizeof (struct tcphdr);
367	optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
368
369	/*
370	* Do quick retrieval of timestamp options ("options
371	* prediction?"). If timestamp is the only option and it's
372	* formatted as recommended in RFC 1323 appendix A, we
373	* quickly get the values now and not bother calling
374	* tcp_dooptions(), etc.
375	*/
376	#if 0
377	if (( optlen == TCPOLEN_TSTAMP_APPA
378	\|\| ( optlen > TCPOLEN_TSTAMP_APPA
379	&& optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
380	(u_int32_t )optp == htonl(TCPOPT_TSTAMP_HDR) &&
381	(ti->ti_flags & TH_SYN) == 0)
382	{
383	ts_present = 1;
384	ts_val = ntohl((u_int32_t )(optp + 4));
385	ts_ecr = ntohl((u_int32_t )(optp + 8));
386	optp = NULL; / * we have parsed the options * /
387	}
388	#endif
389	}
390	tiflags = ti->ti_flags;
391
392	/*
393	* Convert TCP protocol specific fields to host format.
394	*/
395	NTOHL(ti->ti_seq);
396	NTOHL(ti->ti_ack);
397	NTOHS(ti->ti_win);
398	NTOHS(ti->ti_urp);
399
400	/*
401	* Drop TCP, IP headers and TCP options.
402	*/
403	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
404	m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
405
406	/*
407	* Locate pcb for segment.
408	*/
409	findso:
410	if (so != NULL && so != &tcb)
411	SOCKET_UNLOCK(so);
412	QSOCKET_LOCK(tcb);
413	so = tcp_last_so;
414	if ( so->so_fport != ti->ti_dport
415	\|\| so->so_lport != ti->ti_sport
416	\|\| so->so_laddr.s_addr != ti->ti_src.s_addr
417	\|\| so->so_faddr.s_addr != ti->ti_dst.s_addr)
418	{
419	struct socket *sonxt;
420	QSOCKET_UNLOCK(tcb);
421	/* @todo fix SOLOOKUP macrodefinition to be usable here */
422	#ifndef VBOX_WITH_SLIRP_MT
423	so = solookup(&tcb, ti->ti_src, ti->ti_sport,
424	ti->ti_dst, ti->ti_dport);
425	#else
426	so = NULL;
427	QSOCKET_FOREACH(so, sonxt, tcp)
428	/* { */
429	if ( 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	&& so->so_fport == ti->ti_dport
433	&& so->so_deleted != 1)
434	{
435	break; /* so is locked here */
436	}
437	LOOP_LABEL(tcp, so, sonxt);
438	}
439	if (so == &tcb) {
440	so = NULL;
441	}
442	#endif
443	if (so)
444	{
445	tcp_last_so = so;
446	}
447	++tcpstat.tcps_socachemiss;
448	}
449	else
450	{
451	SOCKET_LOCK(so);
452	QSOCKET_UNLOCK(tcb);
453	}
454
455	/*
456	* If the state is CLOSED (i.e., TCB does not exist) then
457	* all data in the incoming segment is discarded.
458	* If the TCB exists but is in CLOSED state, it is embryonic,
459	* but should either do a listen or a connect soon.
460	*
461	* state == CLOSED means we've done socreate() but haven't
462	* attached it to a protocol yet...
463	*
464	* XXX If a TCB does not exist, and the TH_SYN flag is
465	* the only flag set, then create a session, mark it
466	* as if it was LISTENING, and continue...
467	*/
468	if (so == 0)
469	{
470	if ((tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) != TH_SYN)
471	goto dropwithreset;
472
473	if ((so = socreate()) == NULL)
474	goto dropwithreset;
475	if (tcp_attach(pData, so) < 0)
476	{
477	RTMemFree(so); /* Not sofree (if it failed, it's not insqued) */
478	goto dropwithreset;
479	}
480	SOCKET_LOCK(so);
481	sbreserve(pData, &so->so_snd, tcp_sndspace);
482	sbreserve(pData, &so->so_rcv, tcp_rcvspace);
483
484	/* tcp_last_so = so; / / XXX ? */
485	/* tp = sototcpcb(so); */
486
487	so->so_laddr = ti->ti_src;
488	so->so_lport = ti->ti_sport;
489	so->so_faddr = ti->ti_dst;
490	so->so_fport = ti->ti_dport;
491
492	if ((so->so_iptos = tcp_tos(so)) == 0)
493	so->so_iptos = ((struct ip *)ti)->ip_tos;
494
495	tp = sototcpcb(so);
496	tp->t_state = TCPS_LISTEN;
497	}
498
499	/*
500	* If this is a still-connecting socket, this probably
501	* a retransmit of the SYN. Whether it's a retransmit SYN
502	* or something else, we nuke it.
503	*/
504	if (so->so_state & SS_ISFCONNECTING)
505	{
506	goto drop;
507	}
508
509	tp = sototcpcb(so);
510
511	/* XXX Should never fail */
512	if (tp == 0)
513	goto dropwithreset;
514	if (tp->t_state == TCPS_CLOSED)
515	{
516	goto drop;
517	}
518
519	/* Unscale the window into a 32-bit value. */
520	/* if ((tiflags & TH_SYN) == 0)
521	* tiwin = ti->ti_win << tp->snd_scale;
522	* else
523	*/
524	tiwin = ti->ti_win;
525
526	/*
527	* Segment received on connection.
528	* Reset idle time and keep-alive timer.
529	*/
530	tp->t_idle = 0;
531	if (so_options)
532	tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
533	else
534	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
535
536	/*
537	* Process options if not in LISTEN state,
538	* else do it below (after getting remote address).
539	*/
540	if (optp && tp->t_state != TCPS_LISTEN)
541	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
542	/* , */
543	/* &ts_present, &ts_val, &ts_ecr); */
544
545	/*
546	* Header prediction: check for the two common cases
547	* of a uni-directional data xfer. If the packet has
548	* no control flags, is in-sequence, the window didn't
549	* change and we're not retransmitting, it's a
550	* candidate. If the length is zero and the ack moved
551	* forward, we're the sender side of the xfer. Just
552	* free the data acked & wake any higher level process
553	* that was blocked waiting for space. If the length
554	* is non-zero and the ack didn't move, we're the
555	* receiver side. If we're getting packets in-order
556	* (the reassembly queue is empty), add the data to
557	* the socket buffer and note that we need a delayed ack.
558	*
559	* XXX Some of these tests are not needed
560	* eg: the tiwin == tp->snd_wnd prevents many more
561	* predictions.. with no real advantage..
562	*/
563	if ( tp->t_state == TCPS_ESTABLISHED
564	&& (tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK
565	/* && (!ts_present \|\| TSTMP_GEQ(ts_val, tp->ts_recent)) */
566	&& ti->ti_seq == tp->rcv_nxt
567	&& tiwin && tiwin == tp->snd_wnd
568	&& tp->snd_nxt == tp->snd_max)
569	{
570	/*
571	* If last ACK falls within this segment's sequence numbers,
572	* record the timestamp.
573	*/
574	#if 0
575	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
576	SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len))
577	{
578	tp->ts_recent_age = tcp_now;
579	tp->ts_recent = ts_val;
580	}
581	#endif
582
583	if (ti->ti_len == 0)
584	{
585	if ( SEQ_GT(ti->ti_ack, tp->snd_una)
586	&& SEQ_LEQ(ti->ti_ack, tp->snd_max)
587	&& tp->snd_cwnd >= tp->snd_wnd)
588	{
589	/*
590	* this is a pure ack for outstanding data.
591	*/
592	++tcpstat.tcps_predack;
593	#if 0
594	if (ts_present)
595	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
596	else
597	#endif
598	if ( tp->t_rtt
599	&& SEQ_GT(ti->ti_ack, tp->t_rtseq))
600	tcp_xmit_timer(pData, tp, tp->t_rtt);
601	acked = ti->ti_ack - tp->snd_una;
602	tcpstat.tcps_rcvackpack++;
603	tcpstat.tcps_rcvackbyte += acked;
604	sbdrop(&so->so_snd, acked);
605	tp->snd_una = ti->ti_ack;
606	m_freem(pData, m);
607
608	/*
609	* If all outstanding data are acked, stop
610	* retransmit timer, otherwise restart timer
611	* using current (possibly backed-off) value.
612	* If process is waiting for space,
613	* wakeup/selwakeup/signal. If data
614	* are ready to send, let tcp_output
615	* decide between more output or persist.
616	*/
617	if (tp->snd_una == tp->snd_max)
618	tp->t_timer[TCPT_REXMT] = 0;
619	else if (tp->t_timer[TCPT_PERSIST] == 0)
620	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
621
622	/*
623	* There's room in so_snd, sowwakup will read()
624	* from the socket if we can
625	*/
626	#if 0
627	if (so->so_snd.sb_flags & SB_NOTIFY)
628	sowwakeup(so);
629	#endif
630	/*
631	* This is called because sowwakeup might have
632	* put data into so_snd. Since we don't so sowwakeup,
633	* we don't need this.. XXX???
634	*/
635	if (so->so_snd.sb_cc)
636	(void) tcp_output(pData, tp);
637
638	SOCKET_UNLOCK(so);
639	return;
640	}
641	}
642	else if ( ti->ti_ack == tp->snd_una
643	&& LIST_FIRST(&tp->t_segq)
644	&& ti->ti_len <= sbspace(&so->so_rcv))
645	{
646	/*
647	* this is a pure, in-sequence data packet
648	* with nothing on the reassembly queue and
649	* we have enough buffer space to take it.
650	*/
651	++tcpstat.tcps_preddat;
652	tp->rcv_nxt += ti->ti_len;
653	tcpstat.tcps_rcvpack++;
654	tcpstat.tcps_rcvbyte += ti->ti_len;
655	/*
656	* Add data to socket buffer.
657	*/
658	if (so->so_emu)
659	{
660	if (tcp_emu(pData, so, m))
661	sbappend(pData, so, m);
662	}
663	else
664	sbappend(pData, so, m);
665
666	/*
667	* XXX This is called when data arrives. Later, check
668	* if we can actually write() to the socket
669	* XXX Need to check? It's be NON_BLOCKING
670	*/
671	/* sorwakeup(so); */
672
673	/*
674	* If this is a short packet, then ACK now - with Nagel
675	* congestion avoidance sender won't send more until
676	* he gets an ACK.
677	*
678	* It is better to not delay acks at all to maximize
679	* TCP throughput. See RFC 2581.
680	*/
681	tp->t_flags \|= TF_ACKNOW;
682	tcp_output(pData, tp);
683	SOCKET_UNLOCK(so);
684	return;
685	}
686	} /* header prediction */
687	/*
688	* Calculate amount of space in receive window,
689	* and then do TCP input processing.
690	* Receive window is amount of space in rcv queue,
691	* but not less than advertised window.
692	*/
693	{
694	int win;
695	win = sbspace(&so->so_rcv);
696	if (win < 0)
697	win = 0;
698	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
699	}
700
701	switch (tp->t_state)
702	{
703	/*
704	* If the state is LISTEN then ignore segment if it contains an RST.
705	* If the segment contains an ACK then it is bad and send a RST.
706	* If it does not contain a SYN then it is not interesting; drop it.
707	* Don't bother responding if the destination was a broadcast.
708	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
709	* tp->iss, and send a segment:
710	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
711	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
712	* Fill in remote peer address fields if not previously specified.
713	* Enter SYN_RECEIVED state, and process any other fields of this
714	* segment in this state.
715	*/
716	case TCPS_LISTEN:
717	{
718	if (tiflags & TH_RST) {
719	goto drop;
720	}
721	if (tiflags & TH_ACK)
722	goto dropwithreset;
723	if ((tiflags & TH_SYN) == 0)
724	{
725	goto drop;
726	}
727
728	/*
729	* This has way too many gotos...
730	* But a bit of spaghetti code never hurt anybody :)
731	*/
732
733	if (so->so_emu & EMU_NOCONNECT)
734	{
735	so->so_emu &= ~EMU_NOCONNECT;
736	goto cont_input;
737	}
738
739	if ( (tcp_fconnect(pData, so) == -1)
740	&& errno != EINPROGRESS
741	&& errno != EWOULDBLOCK)
742	{
743	u_char code = ICMP_UNREACH_NET;
744	DEBUG_MISC((dfd," tcp fconnect errno = %d-%s\n",
745	errno, strerror(errno)));
746	if (errno == ECONNREFUSED)
747	{
748	/* ACK the SYN, send RST to refuse the connection */
749	tcp_respond(pData, tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
750	TH_RST\|TH_ACK);
751	}
752	else
753	{
754	if (errno == EHOSTUNREACH)
755	code = ICMP_UNREACH_HOST;
756	HTONL(ti->ti_seq); /* restore tcp header */
757	HTONL(ti->ti_ack);
758	HTONS(ti->ti_win);
759	HTONS(ti->ti_urp);
760	m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
761	m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
762	*ip = save_ip;
763	icmp_error(pData, m, ICMP_UNREACH, code, 0, strerror(errno));
764	tp->t_socket->so_m = NULL;
765	}
766	tp = tcp_close(pData, tp);
767	m_free(pData, m);
768	}
769	else
770	{
771	/*
772	* Haven't connected yet, save the current mbuf
773	* and ti, and return
774	* XXX Some OS's don't tell us whether the connect()
775	* succeeded or not. So we must time it out.
776	*/
777	so->so_m = m;
778	so->so_ti = ti;
779	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
780	tp->t_state = TCPS_SYN_RECEIVED;
781	}
782	SOCKET_UNLOCK(so);
783	return;
784
785	cont_conn:
786	/* m==NULL
787	* Check if the connect succeeded
788	*/
789	if (so->so_state & SS_NOFDREF)
790	{
791	tp = tcp_close(pData, tp);
792	goto dropwithreset;
793	}
794	cont_input:
795	tcp_template(tp);
796
797	if (optp)
798	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
799
800	if (iss)
801	tp->iss = iss;
802	else
803	tp->iss = tcp_iss;
804	tcp_iss += TCP_ISSINCR/2;
805	tp->irs = ti->ti_seq;
806	tcp_sendseqinit(tp);
807	tcp_rcvseqinit(tp);
808	tp->t_flags \|= TF_ACKNOW;
809	tp->t_state = TCPS_SYN_RECEIVED;
810	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
811	tcpstat.tcps_accepts++;
812	goto trimthenstep6;
813	} /* case TCPS_LISTEN */
814
815	/*
816	* If the state is SYN_SENT:
817	* if seg contains an ACK, but not for our SYN, drop the input.
818	* if seg contains a RST, then drop the connection.
819	* if seg does not contain SYN, then drop it.
820	* Otherwise this is an acceptable SYN segment
821	* initialize tp->rcv_nxt and tp->irs
822	* if seg contains ack then advance tp->snd_una
823	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
824	* arrange for segment to be acked (eventually)
825	* continue processing rest of data/controls, beginning with URG
826	*/
827	case TCPS_SYN_SENT:
828	if ( (tiflags & TH_ACK)
829	&& ( SEQ_LEQ(ti->ti_ack, tp->iss)
830	\|\| SEQ_GT(ti->ti_ack, tp->snd_max)))
831	goto dropwithreset;
832
833	if (tiflags & TH_RST)
834	{
835	if (tiflags & TH_ACK)
836	tp = tcp_drop(pData, tp, 0); /* XXX Check t_softerror! */
837	goto drop;
838	}
839
840	if ((tiflags & TH_SYN) == 0)
841	{
842	goto drop;
843	}
844	if (tiflags & TH_ACK)
845	{
846	tp->snd_una = ti->ti_ack;
847	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
848	tp->snd_nxt = tp->snd_una;
849	}
850
851	tp->t_timer[TCPT_REXMT] = 0;
852	tp->irs = ti->ti_seq;
853	tcp_rcvseqinit(tp);
854	tp->t_flags \|= TF_ACKNOW;
855	if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss))
856	{
857	tcpstat.tcps_connects++;
858	soisfconnected(so);
859	tp->t_state = TCPS_ESTABLISHED;
860
861	/* Do window scaling on this connection? */
862	#if 0
863	if (( tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
864	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
865	{
866	tp->snd_scale = tp->requested_s_scale;
867	tp->rcv_scale = tp->request_r_scale;
868	}
869	#endif
870	(void) tcp_reass(pData, tp, (struct tcphdr )0, NULL, (struct mbuf )0);
871	/*
872	* if we didn't have to retransmit the SYN,
873	* use its rtt as our initial srtt & rtt var.
874	*/
875	if (tp->t_rtt)
876	tcp_xmit_timer(pData, tp, tp->t_rtt);
877	}
878	else
879	tp->t_state = TCPS_SYN_RECEIVED;
880
881	trimthenstep6:
882	/*
883	* Advance ti->ti_seq to correspond to first data byte.
884	* If data, trim to stay within window,
885	* dropping FIN if necessary.
886	*/
887	ti->ti_seq++;
888	if (ti->ti_len > tp->rcv_wnd)
889	{
890	todrop = ti->ti_len - tp->rcv_wnd;
891	m_adj(m, -todrop);
892	ti->ti_len = tp->rcv_wnd;
893	tiflags &= ~TH_FIN;
894	tcpstat.tcps_rcvpackafterwin++;
895	tcpstat.tcps_rcvbyteafterwin += todrop;
896	}
897	tp->snd_wl1 = ti->ti_seq - 1;
898	tp->rcv_up = ti->ti_seq;
899	Log2(("hit6"));
900	goto step6;
901	} /* switch tp->t_state */
902	/*
903	* States other than LISTEN or SYN_SENT.
904	* First check timestamp, if present.
905	* Then check that at least some bytes of segment are within
906	* receive window. If segment begins before rcv_nxt,
907	* drop leading data (and SYN); if nothing left, just ack.
908	*
909	* RFC 1323 PAWS: If we have a timestamp reply on this segment
910	* and it's less than ts_recent, drop it.
911	*/
912	#if 0
913	if ( ts_present
914	&& (tiflags & TH_RST) == 0
915	&& tp->ts_recent
916	&& TSTMP_LT(ts_val, tp->ts_recent))
917	{
918	/* Check to see if ts_recent is over 24 days old. */
919	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE)
920	{
921	/*
922	* Invalidate ts_recent. If this segment updates
923	* ts_recent, the age will be reset later and ts_recent
924	* will get a valid value. If it does not, setting
925	* ts_recent to zero will at least satisfy the
926	* requirement that zero be placed in the timestamp
927	* echo reply when ts_recent isn't valid. The
928	* age isn't reset until we get a valid ts_recent
929	* because we don't want out-of-order segments to be
930	* dropped when ts_recent is old.
931	*/
932	tp->ts_recent = 0;
933	}
934	else
935	{
936	tcpstat.tcps_rcvduppack++;
937	tcpstat.tcps_rcvdupbyte += ti->ti_len;
938	tcpstat.tcps_pawsdrop++;
939	goto dropafterack;
940	}
941	}
942	#endif
943
944	todrop = tp->rcv_nxt - ti->ti_seq;
945	if (todrop > 0)
946	{
947	if (tiflags & TH_SYN)
948	{
949	tiflags &= ~TH_SYN;
950	ti->ti_seq++;
951	if (ti->ti_urp > 1)
952	ti->ti_urp--;
953	else
954	tiflags &= ~TH_URG;
955	todrop--;
956	}
957	/*
958	* Following if statement from Stevens, vol. 2, p. 960.
959	*/
960	if ( todrop > ti->ti_len
961	\|\| ( todrop == ti->ti_len
962	&& (tiflags & TH_FIN) == 0))
963	{
964	/*
965	* Any valid FIN must be to the left of the window.
966	* At this point the FIN must be a duplicate or out
967	* of sequence; drop it.
968	*/
969	tiflags &= ~TH_FIN;
970
971	/*
972	* Send an ACK to resynchronize and drop any data.
973	* But keep on processing for RST or ACK.
974	*/
975	tp->t_flags \|= TF_ACKNOW;
976	todrop = ti->ti_len;
977	tcpstat.tcps_rcvduppack++;
978	tcpstat.tcps_rcvdupbyte += todrop;
979	}
980	else
981	{
982	tcpstat.tcps_rcvpartduppack++;
983	tcpstat.tcps_rcvpartdupbyte += todrop;
984	}
985	m_adj(m, todrop);
986	ti->ti_seq += todrop;
987	ti->ti_len -= todrop;
988	if (ti->ti_urp > todrop)
989	ti->ti_urp -= todrop;
990	else
991	{
992	tiflags &= ~TH_URG;
993	ti->ti_urp = 0;
994	}
995	}
996	/*
997	* If new data are received on a connection after the
998	* user processes are gone, then RST the other end.
999	*/
1000	if ( (so->so_state & SS_NOFDREF)
1001	&& tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len)
1002	{
1003	tp = tcp_close(pData, tp);
1004	tcpstat.tcps_rcvafterclose++;
1005	goto dropwithreset;
1006	}
1007
1008	/*
1009	* If segment ends after window, drop trailing data
1010	* (and PUSH and FIN); if nothing left, just ACK.
1011	*/
1012	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1013	if (todrop > 0)
1014	{
1015	tcpstat.tcps_rcvpackafterwin++;
1016	if (todrop >= ti->ti_len)
1017	{
1018	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1019	/*
1020	* If a new connection request is received
1021	* while in TIME_WAIT, drop the old connection
1022	* and start over if the sequence numbers
1023	* are above the previous ones.
1024	*/
1025	if ( tiflags & TH_SYN
1026	&& tp->t_state == TCPS_TIME_WAIT
1027	&& SEQ_GT(ti->ti_seq, tp->rcv_nxt))
1028	{
1029	iss = tp->rcv_nxt + TCP_ISSINCR;
1030	tp = tcp_close(pData, tp);
1031	SOCKET_UNLOCK(tp->t_socket);
1032	goto findso;
1033	}
1034	/*
1035	* If window is closed can only take segments at
1036	* window edge, and have to drop data and PUSH from
1037	* incoming segments. Continue processing, but
1038	* remember to ack. Otherwise, drop segment
1039	* and ack.
1040	*/
1041	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt)
1042	{
1043	tp->t_flags \|= TF_ACKNOW;
1044	tcpstat.tcps_rcvwinprobe++;
1045	}
1046	else
1047	goto dropafterack;
1048	}
1049	else
1050	tcpstat.tcps_rcvbyteafterwin += todrop;
1051	m_adj(m, -todrop);
1052	ti->ti_len -= todrop;
1053	tiflags &= ~(TH_PUSH\|TH_FIN);
1054	}
1055
1056	/*
1057	* If last ACK falls within this segment's sequence numbers,
1058	* record its timestamp.
1059	*/
1060	#if 0
1061	if ( ts_present
1062	&& SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)
1063	&& SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + ((tiflags & (TH_SYN\|TH_FIN)) != 0)))
1064	{
1065	tp->ts_recent_age = tcp_now;
1066	tp->ts_recent = ts_val;
1067	}
1068	#endif
1069
1070	/*
1071	* If the RST bit is set examine the state:
1072	* SYN_RECEIVED STATE:
1073	* If passive open, return to LISTEN state.
1074	* If active open, inform user that connection was refused.
1075	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1076	* Inform user that connection was reset, and close tcb.
1077	* CLOSING, LAST_ACK, TIME_WAIT STATES
1078	* Close the tcb.
1079	*/
1080	if (tiflags&TH_RST)
1081	switch (tp->t_state)
1082	{
1083	case TCPS_SYN_RECEIVED:
1084	/* so->so_error = ECONNREFUSED; */
1085	goto close;
1086
1087	case TCPS_ESTABLISHED:
1088	case TCPS_FIN_WAIT_1:
1089	case TCPS_FIN_WAIT_2:
1090	case TCPS_CLOSE_WAIT:
1091	/* so->so_error = ECONNRESET; */
1092	close:
1093	tp->t_state = TCPS_CLOSED;
1094	tcpstat.tcps_drops++;
1095	tp = tcp_close(pData, tp);
1096	goto drop;
1097
1098	case TCPS_CLOSING:
1099	case TCPS_LAST_ACK:
1100	case TCPS_TIME_WAIT:
1101	tp = tcp_close(pData, tp);
1102	goto drop;
1103	}
1104
1105	/*
1106	* If a SYN is in the window, then this is an
1107	* error and we send an RST and drop the connection.
1108	*/
1109	if (tiflags & TH_SYN)
1110	{
1111	tp = tcp_drop(pData, tp, 0);
1112	goto dropwithreset;
1113	}
1114
1115	/*
1116	* If the ACK bit is off we drop the segment and return.
1117	*/
1118	if ((tiflags & TH_ACK) == 0)
1119	{
1120	goto drop;
1121	}
1122
1123	/*
1124	* Ack processing.
1125	*/
1126	switch (tp->t_state)
1127	{
1128	/*
1129	* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1130	* ESTABLISHED state and continue processing, otherwise
1131	* send an RST. una<=ack<=max
1132	*/
1133	case TCPS_SYN_RECEIVED:
1134	if ( SEQ_GT(tp->snd_una, ti->ti_ack)
1135	\|\| SEQ_GT(ti->ti_ack, tp->snd_max))
1136	goto dropwithreset;
1137	tcpstat.tcps_connects++;
1138	tp->t_state = TCPS_ESTABLISHED;
1139	/*
1140	* The sent SYN is ack'ed with our sequence number +1
1141	* The first data byte already in the buffer will get
1142	* lost if no correction is made. This is only needed for
1143	* SS_CTL since the buffer is empty otherwise.
1144	* tp->snd_una++; or:
1145	*/
1146	tp->snd_una = ti->ti_ack;
1147	soisfconnected(so);
1148
1149	/* Do window scaling? */
1150	#if 0
1151	if ( (tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
1152	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
1153	{
1154	tp->snd_scale = tp->requested_s_scale;
1155	tp->rcv_scale = tp->request_r_scale;
1156	}
1157	#endif
1158	(void) tcp_reass(pData, tp, (struct tcphdr )0, (int )0, (struct mbuf *)0);
1159	tp->snd_wl1 = ti->ti_seq - 1;
1160	/* Avoid ack processing; snd_una==ti_ack => dup ack */
1161	goto synrx_to_est;
1162	/* fall into ... */
1163
1164	/*
1165	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1166	* ACKs. If the ack is in the range
1167	* tp->snd_una < ti->ti_ack <= tp->snd_max
1168	* then advance tp->snd_una to ti->ti_ack and drop
1169	* data from the retransmission queue. If this ACK reflects
1170	* more up to date window information we update our window information.
1171	*/
1172	case TCPS_ESTABLISHED:
1173	case TCPS_FIN_WAIT_1:
1174	case TCPS_FIN_WAIT_2:
1175	case TCPS_CLOSE_WAIT:
1176	case TCPS_CLOSING:
1177	case TCPS_LAST_ACK:
1178	case TCPS_TIME_WAIT:
1179	if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
1180	{
1181	if (ti->ti_len == 0 && tiwin == tp->snd_wnd)
1182	{
1183	tcpstat.tcps_rcvdupack++;
1184	DEBUG_MISC((dfd," dup ack m = %lx so = %lx \n",
1185	(long )m, (long )so));
1186	/*
1187	* If we have outstanding data (other than
1188	* a window probe), this is a completely
1189	* duplicate ack (ie, window info didn't
1190	* change), the ack is the biggest we've
1191	* seen and we've seen exactly our rexmt
1192	* threshold of them, assume a packet
1193	* has been dropped and retransmit it.
1194	* Kludge snd_nxt & the congestion
1195	* window so we send only this one
1196	* packet.
1197	*
1198	* We know we're losing at the current
1199	* window size so do congestion avoidance
1200	* (set ssthresh to half the current window
1201	* and pull our congestion window back to
1202	* the new ssthresh).
1203	*
1204	* Dup acks mean that packets have left the
1205	* network (they're now cached at the receiver)
1206	* so bump cwnd by the amount in the receiver
1207	* to keep a constant cwnd packets in the
1208	* network.
1209	*/
1210	if ( tp->t_timer[TCPT_REXMT] == 0
1211	\|\| ti->ti_ack != tp->snd_una)
1212	tp->t_dupacks = 0;
1213	else if (++tp->t_dupacks == tcprexmtthresh)
1214	{
1215	tcp_seq onxt = tp->snd_nxt;
1216	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
1217	if (win < 2)
1218	win = 2;
1219	tp->snd_ssthresh = win * tp->t_maxseg;
1220	tp->t_timer[TCPT_REXMT] = 0;
1221	tp->t_rtt = 0;
1222	tp->snd_nxt = ti->ti_ack;
1223	tp->snd_cwnd = tp->t_maxseg;
1224	(void) tcp_output(pData, tp);
1225	tp->snd_cwnd = tp->snd_ssthresh +
1226	tp->t_maxseg * tp->t_dupacks;
1227	if (SEQ_GT(onxt, tp->snd_nxt))
1228	tp->snd_nxt = onxt;
1229	goto drop;
1230	}
1231	else if (tp->t_dupacks > tcprexmtthresh)
1232	{
1233	tp->snd_cwnd += tp->t_maxseg;
1234	(void) tcp_output(pData, tp);
1235	goto drop;
1236	}
1237	}
1238	else
1239	tp->t_dupacks = 0;
1240	break;
1241	}
1242	synrx_to_est:
1243	/*
1244	* If the congestion window was inflated to account
1245	* for the other side's cached packets, retract it.
1246	*/
1247	if ( tp->t_dupacks > tcprexmtthresh
1248	&& tp->snd_cwnd > tp->snd_ssthresh)
1249	tp->snd_cwnd = tp->snd_ssthresh;
1250	tp->t_dupacks = 0;
1251	if (SEQ_GT(ti->ti_ack, tp->snd_max))
1252	{
1253	tcpstat.tcps_rcvacktoomuch++;
1254	goto dropafterack;
1255	}
1256	acked = ti->ti_ack - tp->snd_una;
1257	tcpstat.tcps_rcvackpack++;
1258	tcpstat.tcps_rcvackbyte += acked;
1259
1260	/*
1261	* If we have a timestamp reply, update smoothed
1262	* round trip time. If no timestamp is present but
1263	* transmit timer is running and timed sequence
1264	* number was acked, update smoothed round trip time.
1265	* Since we now have an rtt measurement, cancel the
1266	* timer backoff (cf., Phil Karn's retransmit alg.).
1267	* Recompute the initial retransmit timer.
1268	*/
1269	#if 0
1270	if (ts_present)
1271	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1272	else
1273	#endif
1274	if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1275	tcp_xmit_timer(pData, tp, tp->t_rtt);
1276
1277	/*
1278	* If all outstanding data is acked, stop retransmit
1279	* timer and remember to restart (more output or persist).
1280	* If there is more data to be acked, restart retransmit
1281	* timer, using current (possibly backed-off) value.
1282	*/
1283	if (ti->ti_ack == tp->snd_max)
1284	{
1285	tp->t_timer[TCPT_REXMT] = 0;
1286	needoutput = 1;
1287	}
1288	else if (tp->t_timer[TCPT_PERSIST] == 0)
1289	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1290	/*
1291	* When new data is acked, open the congestion window.
1292	* If the window gives us less than ssthresh packets
1293	* in flight, open exponentially (maxseg per packet).
1294	* Otherwise open linearly: maxseg per window
1295	* (maxseg^2 / cwnd per packet).
1296	*/
1297	{
1298	register u_int cw = tp->snd_cwnd;
1299	register u_int incr = tp->t_maxseg;
1300
1301	if (cw > tp->snd_ssthresh)
1302	incr = incr * incr / cw;
1303	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1304	}
1305	if (acked > so->so_snd.sb_cc)
1306	{
1307	tp->snd_wnd -= so->so_snd.sb_cc;
1308	sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1309	ourfinisacked = 1;
1310	}
1311	else
1312	{
1313	sbdrop(&so->so_snd, acked);
1314	tp->snd_wnd -= acked;
1315	ourfinisacked = 0;
1316	}
1317	/*
1318	* XXX sowwakup is called when data is acked and there's room for
1319	* for more data... it should read() the socket
1320	*/
1321	#if 0
1322	if (so->so_snd.sb_flags & SB_NOTIFY)
1323	sowwakeup(so);
1324	#endif
1325	tp->snd_una = ti->ti_ack;
1326	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1327	tp->snd_nxt = tp->snd_una;
1328
1329	switch (tp->t_state)
1330	{
1331	/*
1332	* In FIN_WAIT_1 STATE in addition to the processing
1333	* for the ESTABLISHED state if our FIN is now acknowledged
1334	* then enter FIN_WAIT_2.
1335	*/
1336	case TCPS_FIN_WAIT_1:
1337	if (ourfinisacked)
1338	{
1339	/*
1340	* If we can't receive any more
1341	* data, then closing user can proceed.
1342	* Starting the timer is contrary to the
1343	* specification, but if we don't get a FIN
1344	* we'll hang forever.
1345	*/
1346	if (so->so_state & SS_FCANTRCVMORE)
1347	{
1348	soisfdisconnected(so);
1349	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1350	}
1351	tp->t_state = TCPS_FIN_WAIT_2;
1352	}
1353	break;
1354
1355	/*
1356	* In CLOSING STATE in addition to the processing for
1357	* the ESTABLISHED state if the ACK acknowledges our FIN
1358	* then enter the TIME-WAIT state, otherwise ignore
1359	* the segment.
1360	*/
1361	case TCPS_CLOSING:
1362	if (ourfinisacked)
1363	{
1364	tp->t_state = TCPS_TIME_WAIT;
1365	tcp_canceltimers(tp);
1366	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1367	soisfdisconnected(so);
1368	}
1369	break;
1370
1371	/*
1372	* In LAST_ACK, we may still be waiting for data to drain
1373	* and/or to be acked, as well as for the ack of our FIN.
1374	* If our FIN is now acknowledged, delete the TCB,
1375	* enter the closed state and return.
1376	*/
1377	case TCPS_LAST_ACK:
1378	if (ourfinisacked)
1379	{
1380	tp = tcp_close(pData, tp);
1381	goto drop;
1382	}
1383	break;
1384
1385	/*
1386	* In TIME_WAIT state the only thing that should arrive
1387	* is a retransmission of the remote FIN. Acknowledge
1388	* it and restart the finack timer.
1389	*/
1390	case TCPS_TIME_WAIT:
1391	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1392	goto dropafterack;
1393	}
1394	} /* switch(tp->t_state) */
1395
1396	step6:
1397	/*
1398	* Update window information.
1399	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1400	*/
1401	if ( (tiflags & TH_ACK)
1402	&& ( SEQ_LT(tp->snd_wl1, ti->ti_seq)
1403	\|\| ( tp->snd_wl1 == ti->ti_seq
1404	&& ( SEQ_LT(tp->snd_wl2, ti->ti_ack)
1405	\|\| ( tp->snd_wl2 == ti->ti_ack
1406	&& tiwin > tp->snd_wnd)))))
1407	{
1408	/* keep track of pure window updates */
1409	if ( ti->ti_len == 0
1410	&& tp->snd_wl2 == ti->ti_ack
1411	&& tiwin > tp->snd_wnd)
1412	tcpstat.tcps_rcvwinupd++;
1413	tp->snd_wnd = tiwin;
1414	tp->snd_wl1 = ti->ti_seq;
1415	tp->snd_wl2 = ti->ti_ack;
1416	if (tp->snd_wnd > tp->max_sndwnd)
1417	tp->max_sndwnd = tp->snd_wnd;
1418	needoutput = 1;
1419	}
1420
1421	/*
1422	* Process segments with URG.
1423	*/
1424	if ((tiflags & TH_URG) && ti->ti_urp &&
1425	TCPS_HAVERCVDFIN(tp->t_state) == 0)
1426	{
1427	/*
1428	* This is a kludge, but if we receive and accept
1429	* random urgent pointers, we'll crash in
1430	* soreceive. It's hard to imagine someone
1431	* actually wanting to send this much urgent data.
1432	*/
1433	if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen)
1434	{
1435	ti->ti_urp = 0;
1436	tiflags &= ~TH_URG;
1437	goto dodata;
1438	}
1439	/*
1440	* If this segment advances the known urgent pointer,
1441	* then mark the data stream. This should not happen
1442	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1443	* a FIN has been received from the remote side.
1444	* In these states we ignore the URG.
1445	*
1446	* According to RFC961 (Assigned Protocols),
1447	* the urgent pointer points to the last octet
1448	* of urgent data. We continue, however,
1449	* to consider it to indicate the first octet
1450	* of data past the urgent section as the original
1451	* spec states (in one of two places).
1452	*/
1453	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up))
1454	{
1455	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1456	so->so_urgc = so->so_rcv.sb_cc +
1457	(tp->rcv_up - tp->rcv_nxt); /* -1; */
1458	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1459	}
1460	}
1461	else
1462	/*
1463	* If no out of band data is expected,
1464	* pull receive urgent pointer along
1465	* with the receive window.
1466	*/
1467	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1468	tp->rcv_up = tp->rcv_nxt;
1469	dodata:
1470
1471	/*
1472	* If this is a small packet, then ACK now - with Nagel
1473	* congestion avoidance sender won't send more until
1474	* he gets an ACK.
1475	*
1476	* See above.
1477	*/
1478	if ( ti->ti_len
1479	&& (unsigned)ti->ti_len <= 5
1480	&& ((struct tcpiphdr_2 *)ti)->first_char == (char)27)
1481	{
1482	tp->t_flags \|= TF_ACKNOW;
1483	}
1484
1485	/*
1486	* Process the segment text, merging it into the TCP sequencing queue,
1487	* and arranging for acknowledgment of receipt if necessary.
1488	* This process logically involves adjusting tp->rcv_wnd as data
1489	* is presented to the user (this happens in tcp_usrreq.c,
1490	* case PRU_RCVD). If a FIN has already been received on this
1491	* connection then we just ignore the text.
1492	*/
1493	if ( (ti->ti_len \|\| (tiflags&TH_FIN))
1494	&& TCPS_HAVERCVDFIN(tp->t_state) == 0)
1495	{
1496	if ( ti->ti_seq == tp->rcv_nxt
1497	&& LIST_EMPTY(&tp->t_segq)
1498	&& tp->t_state == TCPS_ESTABLISHED)
1499	{
1500	DELAY_ACK(tp, ti); /* little bit different from BSD declaration see netinet/tcp_input.c */
1501	tp->rcv_nxt += tlen;
1502	tiflags = ti->ti_t.th_flags & TH_FIN;
1503	tcpstat.tcps_rcvpack++;
1504	tcpstat.tcps_rcvbyte += tlen;
1505	if (so->so_state & SS_FCANTRCVMORE)
1506	m_freem(pData, m);
1507	else
1508	{
1509	if (so->so_emu)
1510	{
1511	if (tcp_emu(pData, so, m))
1512	sbappend(pData, so, m);
1513	}
1514	else
1515	sbappend(pData, so, m);
1516	}
1517	}
1518	else
1519	{
1520	tiflags = tcp_reass(pData, tp, &ti->ti_t, &tlen, m);
1521	tiflags \|= TF_ACKNOW;
1522	}
1523	/*
1524	* Note the amount of data that peer has sent into
1525	* our window, in order to estimate the sender's
1526	* buffer size.
1527	*/
1528	len = so->so_rcv.sb_datalen - (tp->rcv_adv - tp->rcv_nxt);
1529	}
1530	else
1531	{
1532	m_free(pData, m);
1533	tiflags &= ~TH_FIN;
1534	}
1535
1536	/*
1537	* If FIN is received ACK the FIN and let the user know
1538	* that the connection is closing.
1539	*/
1540	if (tiflags & TH_FIN)
1541	{
1542	if (TCPS_HAVERCVDFIN(tp->t_state) == 0)
1543	{
1544	/*
1545	* If we receive a FIN we can't send more data,
1546	* set it SS_FDRAIN
1547	* Shutdown the socket if there is no rx data in the
1548	* buffer.
1549	* soread() is called on completion of shutdown() and
1550	* will got to TCPS_LAST_ACK, and use tcp_output()
1551	* to send the FIN.
1552	*/
1553	/* sofcantrcvmore(so); */
1554	sofwdrain(so);
1555
1556	tp->t_flags \|= TF_ACKNOW;
1557	tp->rcv_nxt++;
1558	}
1559	switch (tp->t_state)
1560	{
1561	/*
1562	* In SYN_RECEIVED and ESTABLISHED STATES
1563	* enter the CLOSE_WAIT state.
1564	*/
1565	case TCPS_SYN_RECEIVED:
1566	case TCPS_ESTABLISHED:
1567	if(so->so_emu == EMU_CTL) /* no shutdown on socket */
1568	tp->t_state = TCPS_LAST_ACK;
1569	else
1570	tp->t_state = TCPS_CLOSE_WAIT;
1571	break;
1572
1573	/*
1574	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1575	* enter the CLOSING state.
1576	*/
1577	case TCPS_FIN_WAIT_1:
1578	tp->t_state = TCPS_CLOSING;
1579	break;
1580
1581	/*
1582	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1583	* starting the time-wait timer, turning off the other
1584	* standard timers.
1585	*/
1586	case TCPS_FIN_WAIT_2:
1587	tp->t_state = TCPS_TIME_WAIT;
1588	tcp_canceltimers(tp);
1589	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1590	soisfdisconnected(so);
1591	break;
1592
1593	/*
1594	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1595	*/
1596	case TCPS_TIME_WAIT:
1597	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1598	break;
1599	}
1600	}
1601
1602	/*
1603	* Return any desired output.
1604	*/
1605	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1606	tcp_output(pData, tp);
1607
1608	SOCKET_UNLOCK(so);
1609	return;
1610
1611	dropafterack:
1612	Log2(("drop after ack\n"));
1613	/*
1614	* Generate an ACK dropping incoming segment if it occupies
1615	* sequence space, where the ACK reflects our state.
1616	*/
1617	if (tiflags & TH_RST)
1618	goto drop;
1619	m_freem(pData, m);
1620	tp->t_flags \|= TF_ACKNOW;
1621	(void) tcp_output(pData, tp);
1622	SOCKET_UNLOCK(so);
1623	return;
1624
1625	dropwithreset:
1626	/* reuses m if m!=NULL, m_free() unnecessary */
1627	if (tiflags & TH_ACK)
1628	tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1629	else
1630	{
1631	if (tiflags & TH_SYN) ti->ti_len++;
1632	tcp_respond(pData, tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1633	TH_RST\|TH_ACK);
1634	}
1635
1636	if (so != &tcb)
1637	SOCKET_UNLOCK(so);
1638	return;
1639
1640	drop:
1641	/*
1642	* Drop space held by incoming segment and return.
1643	*/
1644	m_free(pData, m);
1645
1646	#ifdef VBOX_WITH_SLIRP_MT
1647	if (RTCritSectIsOwned(&so->so_mutex))
1648	{
1649	SOCKET_UNLOCK(so);
1650	}
1651	#endif
1652
1653	return;
1654	}
1655
1656	void
1657	tcp_dooptions(PNATState pData, struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti)
1658	{
1659	u_int16_t mss;
1660	int opt, optlen;
1661
1662	DEBUG_CALL("tcp_dooptions");
1663	DEBUG_ARGS((dfd," tp = %lx cnt=%i \n", (long )tp, cnt));
1664
1665	for (; cnt > 0; cnt -= optlen, cp += optlen)
1666	{
1667	opt = cp[0];
1668	if (opt == TCPOPT_EOL)
1669	break;
1670	if (opt == TCPOPT_NOP)
1671	optlen = 1;
1672	else
1673	{
1674	optlen = cp[1];
1675	if (optlen <= 0)
1676	break;
1677	}
1678	switch (opt)
1679	{
1680	default:
1681	continue;
1682
1683	case TCPOPT_MAXSEG:
1684	if (optlen != TCPOLEN_MAXSEG)
1685	continue;
1686	if (!(ti->ti_flags & TH_SYN))
1687	continue;
1688	memcpy((char ) &mss, (char ) cp + 2, sizeof(mss));
1689	NTOHS(mss);
1690	(void) tcp_mss(pData, tp, mss); /* sets t_maxseg */
1691	break;
1692
1693	#if 0
1694	case TCPOPT_WINDOW:
1695	if (optlen != TCPOLEN_WINDOW)
1696	continue;
1697	if (!(ti->ti_flags & TH_SYN))
1698	continue;
1699	tp->t_flags \|= TF_RCVD_SCALE;
1700	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1701	break;
1702
1703	case TCPOPT_TIMESTAMP:
1704	if (optlen != TCPOLEN_TIMESTAMP)
1705	continue;
1706	*ts_present = 1;
1707	memcpy((char ) ts_val, (char )cp + 2, sizeof(*ts_val));
1708	NTOHL(*ts_val);
1709	memcpy((char ) ts_ecr, (char )cp + 6, sizeof(*ts_ecr));
1710	NTOHL(*ts_ecr);
1711
1712	/*
1713	* A timestamp received in a SYN makes
1714	* it ok to send timestamp requests and replies.
1715	*/
1716	if (ti->ti_flags & TH_SYN)
1717	{
1718	tp->t_flags \|= TF_RCVD_TSTMP;
1719	tp->ts_recent = *ts_val;
1720	tp->ts_recent_age = tcp_now;
1721	}
1722	break;
1723	#endif
1724	}
1725	}
1726	}
1727
1728
1729	/*
1730	* Pull out of band byte out of a segment so
1731	* it doesn't appear in the user's data queue.
1732	* It is still reflected in the segment length for
1733	* sequencing purposes.
1734	*/
1735
1736	#if 0
1737	void
1738	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1739	{
1740	int cnt = ti->ti_urp - 1;
1741
1742	while (cnt >= 0)
1743	{
1744	if (m->m_len > cnt)
1745	{
1746	char *cp = mtod(m, caddr_t) + cnt;
1747	struct tcpcb *tp = sototcpcb(so);
1748
1749	tp->t_iobc = *cp;
1750	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1751	memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1752	m->m_len--;
1753	return;
1754	}
1755	cnt -= m->m_len;
1756	m = m->m_next; /* XXX WRONG! Fix it! */
1757	if (m == 0)
1758	break;
1759	}
1760	panic("tcp_pulloutofband");
1761	}
1762	#endif
1763
1764	/*
1765	* Collect new round-trip time estimate
1766	* and update averages and current timeout.
1767	*/
1768
1769	void
1770	tcp_xmit_timer(PNATState pData, register struct tcpcb *tp, int rtt)
1771	{
1772	register short delta;
1773
1774	DEBUG_CALL("tcp_xmit_timer");
1775	DEBUG_ARG("tp = %lx", (long)tp);
1776	DEBUG_ARG("rtt = %d", rtt);
1777
1778	tcpstat.tcps_rttupdated++;
1779	if (tp->t_srtt != 0)
1780	{
1781	/*
1782	* srtt is stored as fixed point with 3 bits after the
1783	* binary point (i.e., scaled by 8). The following magic
1784	* is equivalent to the smoothing algorithm in rfc793 with
1785	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1786	* point). Adjust rtt to origin 0.
1787	*/
1788	delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1789	if ((tp->t_srtt += delta) <= 0)
1790	tp->t_srtt = 1;
1791	/*
1792	* We accumulate a smoothed rtt variance (actually, a
1793	* smoothed mean difference), then set the retransmit
1794	* timer to smoothed rtt + 4 times the smoothed variance.
1795	* rttvar is stored as fixed point with 2 bits after the
1796	* binary point (scaled by 4). The following is
1797	* equivalent to rfc793 smoothing with an alpha of .75
1798	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1799	* rfc793's wired-in beta.
1800	*/
1801	if (delta < 0)
1802	delta = -delta;
1803	delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1804	if ((tp->t_rttvar += delta) <= 0)
1805	tp->t_rttvar = 1;
1806	}
1807	else
1808	{
1809	/*
1810	* No rtt measurement yet - use the unsmoothed rtt.
1811	* Set the variance to half the rtt (so our first
1812	* retransmit happens at 3*rtt).
1813	*/
1814	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1815	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1816	}
1817	tp->t_rtt = 0;
1818	tp->t_rxtshift = 0;
1819
1820	/*
1821	* the retransmit should happen at rtt + 4 * rttvar.
1822	* Because of the way we do the smoothing, srtt and rttvar
1823	* will each average +1/2 tick of bias. When we compute
1824	* the retransmit timer, we want 1/2 tick of rounding and
1825	* 1 extra tick because of +-1/2 tick uncertainty in the
1826	* firing of the timer. The bias will give us exactly the
1827	* 1.5 tick we need. But, because the bias is
1828	* statistical, we have to test that we don't drop below
1829	* the minimum feasible timer (which is 2 ticks).
1830	*/
1831	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1832	(short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1833
1834	/*
1835	* We received an ack for a packet that wasn't retransmitted;
1836	* it is probably safe to discard any error indications we've
1837	* received recently. This isn't quite right, but close enough
1838	* for now (a route might have failed after we sent a segment,
1839	* and the return path might not be symmetrical).
1840	*/
1841	tp->t_softerror = 0;
1842	}
1843
1844	/*
1845	* Determine a reasonable value for maxseg size.
1846	* If the route is known, check route for mtu.
1847	* If none, use an mss that can be handled on the outgoing
1848	* interface without forcing IP to fragment; if bigger than
1849	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1850	* to utilize large mbufs. If no route is found, route has no mtu,
1851	* or the destination isn't local, use a default, hopefully conservative
1852	* size (usually 512 or the default IP max size, but no more than the mtu
1853	* of the interface), as we can't discover anything about intervening
1854	* gateways or networks. We also initialize the congestion/slow start
1855	* window to be a single segment if the destination isn't local.
1856	* While looking at the routing entry, we also initialize other path-dependent
1857	* parameters from pre-set or cached values in the routing entry.
1858	*/
1859
1860	int
1861	tcp_mss(PNATState pData, register struct tcpcb *tp, u_int offer)
1862	{
1863	struct socket *so = tp->t_socket;
1864	int mss;
1865
1866	DEBUG_CALL("tcp_mss");
1867	DEBUG_ARG("tp = %lx", (long)tp);
1868	DEBUG_ARG("offer = %d", offer);
1869
1870	mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
1871	if (offer)
1872	mss = min(mss, offer);
1873	mss = max(mss, 32);
1874	if (mss < tp->t_maxseg \|\| offer != 0)
1875	tp->t_maxseg = mss;
1876
1877	tp->snd_cwnd = mss;
1878
1879	sbreserve(pData, &so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
1880	sbreserve(pData, &so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
1881
1882	DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1883
1884	return mss;
1885	}

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 20383

Download in other formats: