tcp_input.c@ 17145

Last change on this file since 17145 was 17145, checked in by vboxsync, 16 years ago
NAT: preventing attempt of double ICMP packet sends, causing crash.
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)) sbappend(pData, so, m);
660	}
661	else
662	sbappend(pData, so, m);
663
664	/*
665	* XXX This is called when data arrives. Later, check
666	* if we can actually write() to the socket
667	* XXX Need to check? It's be NON_BLOCKING
668	*/
669	/* sorwakeup(so); */
670
671	/*
672	* If this is a short packet, then ACK now - with Nagel
673	* congestion avoidance sender won't send more until
674	* he gets an ACK.
675	*
676	* It is better to not delay acks at all to maximize
677	* TCP throughput. See RFC 2581.
678	*/
679	tp->t_flags \|= TF_ACKNOW;
680	tcp_output(pData, tp);
681	SOCKET_UNLOCK(so);
682	return;
683	}
684	} /* header prediction */
685	/*
686	* Calculate amount of space in receive window,
687	* and then do TCP input processing.
688	* Receive window is amount of space in rcv queue,
689	* but not less than advertised window.
690	*/
691	{
692	int win;
693	win = sbspace(&so->so_rcv);
694	if (win < 0)
695	win = 0;
696	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
697	}
698
699	switch (tp->t_state)
700	{
701	/*
702	* If the state is LISTEN then ignore segment if it contains an RST.
703	* If the segment contains an ACK then it is bad and send a RST.
704	* If it does not contain a SYN then it is not interesting; drop it.
705	* Don't bother responding if the destination was a broadcast.
706	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
707	* tp->iss, and send a segment:
708	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
709	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
710	* Fill in remote peer address fields if not previously specified.
711	* Enter SYN_RECEIVED state, and process any other fields of this
712	* segment in this state.
713	*/
714	case TCPS_LISTEN:
715	{
716	if (tiflags & TH_RST) {
717	Log2(("RST(%x) is on listen =>drop\n", tiflags));
718	goto drop;
719	}
720	if (tiflags & TH_ACK)
721	goto dropwithreset;
722	if ((tiflags & TH_SYN) == 0)
723	{
724	Log2(("SYN(%x) is off on listen =>drop\n", tiflags));
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	Log2(("hit trimthenstep6\n"));
813	goto trimthenstep6;
814	} /* case TCPS_LISTEN */
815
816	/*
817	* If the state is SYN_SENT:
818	* if seg contains an ACK, but not for our SYN, drop the input.
819	* if seg contains a RST, then drop the connection.
820	* if seg does not contain SYN, then drop it.
821	* Otherwise this is an acceptable SYN segment
822	* initialize tp->rcv_nxt and tp->irs
823	* if seg contains ack then advance tp->snd_una
824	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
825	* arrange for segment to be acked (eventually)
826	* continue processing rest of data/controls, beginning with URG
827	*/
828	case TCPS_SYN_SENT:
829	if ( (tiflags & TH_ACK)
830	&& ( SEQ_LEQ(ti->ti_ack, tp->iss)
831	\|\| SEQ_GT(ti->ti_ack, tp->snd_max)))
832	goto dropwithreset;
833
834	if (tiflags & TH_RST)
835	{
836	if (tiflags & TH_ACK)
837	tp = tcp_drop(pData, tp,0); /* XXX Check t_softerror! */
838	Log2(("RST(%x) is on SYN_SENT =>drop\n", tiflags));
839	goto drop;
840	}
841
842	if ((tiflags & TH_SYN) == 0)
843	{
844	Log2(("SYN(%x) bit is off on SYN_SENT =>drop\n", tiflags));
845	goto drop;
846	}
847	if (tiflags & TH_ACK)
848	{
849	tp->snd_una = ti->ti_ack;
850	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
851	tp->snd_nxt = tp->snd_una;
852	}
853
854	tp->t_timer[TCPT_REXMT] = 0;
855	tp->irs = ti->ti_seq;
856	tcp_rcvseqinit(tp);
857	tp->t_flags \|= TF_ACKNOW;
858	if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss))
859	{
860	tcpstat.tcps_connects++;
861	soisfconnected(so);
862	tp->t_state = TCPS_ESTABLISHED;
863
864	/* Do window scaling on this connection? */
865	#if 0
866	if (( tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
867	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
868	{
869	tp->snd_scale = tp->requested_s_scale;
870	tp->rcv_scale = tp->request_r_scale;
871	}
872	#endif
873	(void) tcp_reass(pData, tp, (struct tcphdr )0, NULL, (struct mbuf )0);
874	/*
875	* if we didn't have to retransmit the SYN,
876	* use its rtt as our initial srtt & rtt var.
877	*/
878	if (tp->t_rtt)
879	tcp_xmit_timer(pData, tp, tp->t_rtt);
880	}
881	else
882	tp->t_state = TCPS_SYN_RECEIVED;
883
884	trimthenstep6:
885	/*
886	* Advance ti->ti_seq to correspond to first data byte.
887	* If data, trim to stay within window,
888	* dropping FIN if necessary.
889	*/
890	ti->ti_seq++;
891	if (ti->ti_len > tp->rcv_wnd)
892	{
893	todrop = ti->ti_len - tp->rcv_wnd;
894	m_adj(m, -todrop);
895	ti->ti_len = tp->rcv_wnd;
896	tiflags &= ~TH_FIN;
897	tcpstat.tcps_rcvpackafterwin++;
898	tcpstat.tcps_rcvbyteafterwin += todrop;
899	}
900	tp->snd_wl1 = ti->ti_seq - 1;
901	tp->rcv_up = ti->ti_seq;
902	Log2(("hit6"));
903	goto step6;
904	} /* switch tp->t_state */
905	/*
906	* States other than LISTEN or SYN_SENT.
907	* First check timestamp, if present.
908	* Then check that at least some bytes of segment are within
909	* receive window. If segment begins before rcv_nxt,
910	* drop leading data (and SYN); if nothing left, just ack.
911	*
912	* RFC 1323 PAWS: If we have a timestamp reply on this segment
913	* and it's less than ts_recent, drop it.
914	*/
915	#if 0
916	if ( ts_present
917	&& (tiflags & TH_RST) == 0
918	&& tp->ts_recent
919	&& TSTMP_LT(ts_val, tp->ts_recent))
920	{
921	/* Check to see if ts_recent is over 24 days old. */
922	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE)
923	{
924	/*
925	* Invalidate ts_recent. If this segment updates
926	* ts_recent, the age will be reset later and ts_recent
927	* will get a valid value. If it does not, setting
928	* ts_recent to zero will at least satisfy the
929	* requirement that zero be placed in the timestamp
930	* echo reply when ts_recent isn't valid. The
931	* age isn't reset until we get a valid ts_recent
932	* because we don't want out-of-order segments to be
933	* dropped when ts_recent is old.
934	*/
935	tp->ts_recent = 0;
936	}
937	else
938	{
939	tcpstat.tcps_rcvduppack++;
940	tcpstat.tcps_rcvdupbyte += ti->ti_len;
941	tcpstat.tcps_pawsdrop++;
942	goto dropafterack;
943	}
944	}
945	#endif
946
947	todrop = tp->rcv_nxt - ti->ti_seq;
948	if (todrop > 0)
949	{
950	if (tiflags & TH_SYN)
951	{
952	tiflags &= ~TH_SYN;
953	ti->ti_seq++;
954	if (ti->ti_urp > 1)
955	ti->ti_urp--;
956	else
957	tiflags &= ~TH_URG;
958	todrop--;
959	}
960	/*
961	* Following if statement from Stevens, vol. 2, p. 960.
962	*/
963	if ( todrop > ti->ti_len
964	\|\| ( todrop == ti->ti_len
965	&& (tiflags & TH_FIN) == 0))
966	{
967	/*
968	* Any valid FIN must be to the left of the window.
969	* At this point the FIN must be a duplicate or out
970	* of sequence; drop it.
971	*/
972	tiflags &= ~TH_FIN;
973
974	/*
975	* Send an ACK to resynchronize and drop any data.
976	* But keep on processing for RST or ACK.
977	*/
978	tp->t_flags \|= TF_ACKNOW;
979	todrop = ti->ti_len;
980	tcpstat.tcps_rcvduppack++;
981	tcpstat.tcps_rcvdupbyte += todrop;
982	}
983	else
984	{
985	tcpstat.tcps_rcvpartduppack++;
986	tcpstat.tcps_rcvpartdupbyte += todrop;
987	}
988	m_adj(m, todrop);
989	ti->ti_seq += todrop;
990	ti->ti_len -= todrop;
991	if (ti->ti_urp > todrop)
992	ti->ti_urp -= todrop;
993	else
994	{
995	tiflags &= ~TH_URG;
996	ti->ti_urp = 0;
997	}
998	}
999	/*
1000	* If new data are received on a connection after the
1001	* user processes are gone, then RST the other end.
1002	*/
1003	if ( (so->so_state & SS_NOFDREF)
1004	&& tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len)
1005	{
1006	tp = tcp_close(pData, tp);
1007	tcpstat.tcps_rcvafterclose++;
1008	goto dropwithreset;
1009	}
1010
1011	/*
1012	* If segment ends after window, drop trailing data
1013	* (and PUSH and FIN); if nothing left, just ACK.
1014	*/
1015	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1016	if (todrop > 0)
1017	{
1018	tcpstat.tcps_rcvpackafterwin++;
1019	if (todrop >= ti->ti_len)
1020	{
1021	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1022	/*
1023	* If a new connection request is received
1024	* while in TIME_WAIT, drop the old connection
1025	* and start over if the sequence numbers
1026	* are above the previous ones.
1027	*/
1028	if ( tiflags & TH_SYN
1029	&& tp->t_state == TCPS_TIME_WAIT
1030	&& SEQ_GT(ti->ti_seq, tp->rcv_nxt))
1031	{
1032	iss = tp->rcv_nxt + TCP_ISSINCR;
1033	tp = tcp_close(pData, tp);
1034	SOCKET_UNLOCK(tp->t_socket);
1035	goto findso;
1036	}
1037	/*
1038	* If window is closed can only take segments at
1039	* window edge, and have to drop data and PUSH from
1040	* incoming segments. Continue processing, but
1041	* remember to ack. Otherwise, drop segment
1042	* and ack.
1043	*/
1044	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt)
1045	{
1046	tp->t_flags \|= TF_ACKNOW;
1047	tcpstat.tcps_rcvwinprobe++;
1048	}
1049	else
1050	goto dropafterack;
1051	}
1052	else
1053	tcpstat.tcps_rcvbyteafterwin += todrop;
1054	m_adj(m, -todrop);
1055	ti->ti_len -= todrop;
1056	tiflags &= ~(TH_PUSH\|TH_FIN);
1057	}
1058
1059	/*
1060	* If last ACK falls within this segment's sequence numbers,
1061	* record its timestamp.
1062	*/
1063	#if 0
1064	if ( ts_present
1065	&& SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)
1066	&& SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + ((tiflags & (TH_SYN\|TH_FIN)) != 0)))
1067	{
1068	tp->ts_recent_age = tcp_now;
1069	tp->ts_recent = ts_val;
1070	}
1071	#endif
1072
1073	/*
1074	* If the RST bit is set examine the state:
1075	* SYN_RECEIVED STATE:
1076	* If passive open, return to LISTEN state.
1077	* If active open, inform user that connection was refused.
1078	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1079	* Inform user that connection was reset, and close tcb.
1080	* CLOSING, LAST_ACK, TIME_WAIT STATES
1081	* Close the tcb.
1082	*/
1083	if (tiflags&TH_RST)
1084	switch (tp->t_state)
1085	{
1086	case TCPS_SYN_RECEIVED:
1087	/* so->so_error = ECONNREFUSED; */
1088	goto close;
1089
1090	case TCPS_ESTABLISHED:
1091	case TCPS_FIN_WAIT_1:
1092	case TCPS_FIN_WAIT_2:
1093	case TCPS_CLOSE_WAIT:
1094	/* so->so_error = ECONNRESET; */
1095	close:
1096	Log2(("closing...=>drop\n", tp->t_state));
1097	tp->t_state = TCPS_CLOSED;
1098	tcpstat.tcps_drops++;
1099	tp = tcp_close(pData, tp);
1100	goto drop;
1101
1102	case TCPS_CLOSING:
1103	case TCPS_LAST_ACK:
1104	case TCPS_TIME_WAIT:
1105	Log2(("t_state is (%x) sort of close =>drop\n", tp->t_state));
1106	tp = tcp_close(pData, tp);
1107	goto drop;
1108	}
1109
1110	/*
1111	* If a SYN is in the window, then this is an
1112	* error and we send an RST and drop the connection.
1113	*/
1114	if (tiflags & TH_SYN)
1115	{
1116	tp = tcp_drop(pData, tp,0);
1117	goto dropwithreset;
1118	}
1119
1120	/*
1121	* If the ACK bit is off we drop the segment and return.
1122	*/
1123	if ((tiflags & TH_ACK) == 0)
1124	{
1125	Log2(("ACK(%x) bit is off =>drop\n", tiflags));
1126	goto drop;
1127	}
1128
1129	/*
1130	* Ack processing.
1131	*/
1132	switch (tp->t_state)
1133	{
1134	/*
1135	* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1136	* ESTABLISHED state and continue processing, otherwise
1137	* send an RST. una<=ack<=max
1138	*/
1139	case TCPS_SYN_RECEIVED:
1140	if ( SEQ_GT(tp->snd_una, ti->ti_ack)
1141	\|\| SEQ_GT(ti->ti_ack, tp->snd_max))
1142	goto dropwithreset;
1143	tcpstat.tcps_connects++;
1144	tp->t_state = TCPS_ESTABLISHED;
1145	/*
1146	* The sent SYN is ack'ed with our sequence number +1
1147	* The first data byte already in the buffer will get
1148	* lost if no correction is made. This is only needed for
1149	* SS_CTL since the buffer is empty otherwise.
1150	* tp->snd_una++; or:
1151	*/
1152	tp->snd_una = ti->ti_ack;
1153	soisfconnected(so);
1154
1155	/* Do window scaling? */
1156	#if 0
1157	if ( (tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
1158	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
1159	{
1160	tp->snd_scale = tp->requested_s_scale;
1161	tp->rcv_scale = tp->request_r_scale;
1162	}
1163	#endif
1164	(void) tcp_reass(pData, tp, (struct tcphdr )0, (int )0, (struct mbuf *)0);
1165	tp->snd_wl1 = ti->ti_seq - 1;
1166	/* Avoid ack processing; snd_una==ti_ack => dup ack */
1167	Log2(("hit synrx_to_est\n"));
1168	goto synrx_to_est;
1169	/* fall into ... */
1170
1171	/*
1172	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1173	* ACKs. If the ack is in the range
1174	* tp->snd_una < ti->ti_ack <= tp->snd_max
1175	* then advance tp->snd_una to ti->ti_ack and drop
1176	* data from the retransmission queue. If this ACK reflects
1177	* more up to date window information we update our window information.
1178	*/
1179	case TCPS_ESTABLISHED:
1180	case TCPS_FIN_WAIT_1:
1181	case TCPS_FIN_WAIT_2:
1182	case TCPS_CLOSE_WAIT:
1183	case TCPS_CLOSING:
1184	case TCPS_LAST_ACK:
1185	case TCPS_TIME_WAIT:
1186	if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
1187	{
1188	if (ti->ti_len == 0 && tiwin == tp->snd_wnd)
1189	{
1190	tcpstat.tcps_rcvdupack++;
1191	DEBUG_MISC((dfd," dup ack m = %lx so = %lx \n",
1192	(long )m, (long )so));
1193	/*
1194	* If we have outstanding data (other than
1195	* a window probe), this is a completely
1196	* duplicate ack (ie, window info didn't
1197	* change), the ack is the biggest we've
1198	* seen and we've seen exactly our rexmt
1199	* threshold of them, assume a packet
1200	* has been dropped and retransmit it.
1201	* Kludge snd_nxt & the congestion
1202	* window so we send only this one
1203	* packet.
1204	*
1205	* We know we're losing at the current
1206	* window size so do congestion avoidance
1207	* (set ssthresh to half the current window
1208	* and pull our congestion window back to
1209	* the new ssthresh).
1210	*
1211	* Dup acks mean that packets have left the
1212	* network (they're now cached at the receiver)
1213	* so bump cwnd by the amount in the receiver
1214	* to keep a constant cwnd packets in the
1215	* network.
1216	*/
1217	if ( tp->t_timer[TCPT_REXMT] == 0
1218	\|\| ti->ti_ack != tp->snd_una)
1219	tp->t_dupacks = 0;
1220	else if (++tp->t_dupacks == tcprexmtthresh)
1221	{
1222	tcp_seq onxt = tp->snd_nxt;
1223	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
1224	if (win < 2)
1225	win = 2;
1226	tp->snd_ssthresh = win * tp->t_maxseg;
1227	tp->t_timer[TCPT_REXMT] = 0;
1228	tp->t_rtt = 0;
1229	tp->snd_nxt = ti->ti_ack;
1230	tp->snd_cwnd = tp->t_maxseg;
1231	(void) tcp_output(pData, tp);
1232	tp->snd_cwnd = tp->snd_ssthresh +
1233	tp->t_maxseg * tp->t_dupacks;
1234	if (SEQ_GT(onxt, tp->snd_nxt))
1235	tp->snd_nxt = onxt;
1236	Log2(("t_dupacks(%d) == tcprexmtthresh(%d)=>drop\n", tp->t_dupacks, tcprexmtthresh));
1237	goto drop;
1238	}
1239	else if (tp->t_dupacks > tcprexmtthresh)
1240	{
1241	tp->snd_cwnd += tp->t_maxseg;
1242	(void) tcp_output(pData, tp);
1243	Log2(("t_dupacks(%d) > tcprexmtthresh(%d)=>drop\n", tp->t_dupacks, tcprexmtthresh));
1244	goto drop;
1245	}
1246	}
1247	else
1248	tp->t_dupacks = 0;
1249	break;
1250	}
1251	synrx_to_est:
1252	Log2(("enter synrx_to_est\n"));
1253	/*
1254	* If the congestion window was inflated to account
1255	* for the other side's cached packets, retract it.
1256	*/
1257	if ( tp->t_dupacks > tcprexmtthresh
1258	&& tp->snd_cwnd > tp->snd_ssthresh)
1259	tp->snd_cwnd = tp->snd_ssthresh;
1260	tp->t_dupacks = 0;
1261	if (SEQ_GT(ti->ti_ack, tp->snd_max))
1262	{
1263	tcpstat.tcps_rcvacktoomuch++;
1264	goto dropafterack;
1265	}
1266	acked = ti->ti_ack - tp->snd_una;
1267	tcpstat.tcps_rcvackpack++;
1268	tcpstat.tcps_rcvackbyte += acked;
1269
1270	/*
1271	* If we have a timestamp reply, update smoothed
1272	* round trip time. If no timestamp is present but
1273	* transmit timer is running and timed sequence
1274	* number was acked, update smoothed round trip time.
1275	* Since we now have an rtt measurement, cancel the
1276	* timer backoff (cf., Phil Karn's retransmit alg.).
1277	* Recompute the initial retransmit timer.
1278	*/
1279	#if 0
1280	if (ts_present)
1281	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1282	else
1283	#endif
1284	if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1285	tcp_xmit_timer(pData, tp,tp->t_rtt);
1286
1287	/*
1288	* If all outstanding data is acked, stop retransmit
1289	* timer and remember to restart (more output or persist).
1290	* If there is more data to be acked, restart retransmit
1291	* timer, using current (possibly backed-off) value.
1292	*/
1293	if (ti->ti_ack == tp->snd_max)
1294	{
1295	tp->t_timer[TCPT_REXMT] = 0;
1296	needoutput = 1;
1297	}
1298	else if (tp->t_timer[TCPT_PERSIST] == 0)
1299	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1300	/*
1301	* When new data is acked, open the congestion window.
1302	* If the window gives us less than ssthresh packets
1303	* in flight, open exponentially (maxseg per packet).
1304	* Otherwise open linearly: maxseg per window
1305	* (maxseg^2 / cwnd per packet).
1306	*/
1307	{
1308	register u_int cw = tp->snd_cwnd;
1309	register u_int incr = tp->t_maxseg;
1310
1311	if (cw > tp->snd_ssthresh)
1312	incr = incr * incr / cw;
1313	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1314	}
1315	if (acked > so->so_snd.sb_cc)
1316	{
1317	tp->snd_wnd -= so->so_snd.sb_cc;
1318	sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1319	ourfinisacked = 1;
1320	}
1321	else
1322	{
1323	sbdrop(&so->so_snd, acked);
1324	tp->snd_wnd -= acked;
1325	ourfinisacked = 0;
1326	}
1327	/*
1328	* XXX sowwakup is called when data is acked and there's room for
1329	* for more data... it should read() the socket
1330	*/
1331	#if 0
1332	if (so->so_snd.sb_flags & SB_NOTIFY)
1333	sowwakeup(so);
1334	#endif
1335	tp->snd_una = ti->ti_ack;
1336	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1337	tp->snd_nxt = tp->snd_una;
1338
1339	switch (tp->t_state)
1340	{
1341	/*
1342	* In FIN_WAIT_1 STATE in addition to the processing
1343	* for the ESTABLISHED state if our FIN is now acknowledged
1344	* then enter FIN_WAIT_2.
1345	*/
1346	case TCPS_FIN_WAIT_1:
1347	if (ourfinisacked)
1348	{
1349	/*
1350	* If we can't receive any more
1351	* data, then closing user can proceed.
1352	* Starting the timer is contrary to the
1353	* specification, but if we don't get a FIN
1354	* we'll hang forever.
1355	*/
1356	if (so->so_state & SS_FCANTRCVMORE)
1357	{
1358	soisfdisconnected(so);
1359	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1360	}
1361	tp->t_state = TCPS_FIN_WAIT_2;
1362	}
1363	break;
1364
1365	/*
1366	* In CLOSING STATE in addition to the processing for
1367	* the ESTABLISHED state if the ACK acknowledges our FIN
1368	* then enter the TIME-WAIT state, otherwise ignore
1369	* the segment.
1370	*/
1371	case TCPS_CLOSING:
1372	if (ourfinisacked)
1373	{
1374	tp->t_state = TCPS_TIME_WAIT;
1375	tcp_canceltimers(tp);
1376	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1377	soisfdisconnected(so);
1378	}
1379	break;
1380
1381	/*
1382	* In LAST_ACK, we may still be waiting for data to drain
1383	* and/or to be acked, as well as for the ack of our FIN.
1384	* If our FIN is now acknowledged, delete the TCB,
1385	* enter the closed state and return.
1386	*/
1387	case TCPS_LAST_ACK:
1388	if (ourfinisacked)
1389	{
1390	Log2(("ourfinisacked=>drop\n"));
1391	tp = tcp_close(pData, tp);
1392	goto drop;
1393	}
1394	break;
1395
1396	/*
1397	* In TIME_WAIT state the only thing that should arrive
1398	* is a retransmission of the remote FIN. Acknowledge
1399	* it and restart the finack timer.
1400	*/
1401	case TCPS_TIME_WAIT:
1402	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1403	goto dropafterack;
1404	}
1405	} /* switch(tp->t_state) */
1406
1407	step6:
1408	/*
1409	* Update window information.
1410	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1411	*/
1412	if ( (tiflags & TH_ACK)
1413	&& ( SEQ_LT(tp->snd_wl1, ti->ti_seq)
1414	\|\| ( tp->snd_wl1 == ti->ti_seq
1415	&& ( SEQ_LT(tp->snd_wl2, ti->ti_ack)
1416	\|\| ( tp->snd_wl2 == ti->ti_ack
1417	&& tiwin > tp->snd_wnd)))))
1418	{
1419	/* keep track of pure window updates */
1420	if ( ti->ti_len == 0
1421	&& tp->snd_wl2 == ti->ti_ack
1422	&& tiwin > tp->snd_wnd)
1423	tcpstat.tcps_rcvwinupd++;
1424	tp->snd_wnd = tiwin;
1425	tp->snd_wl1 = ti->ti_seq;
1426	tp->snd_wl2 = ti->ti_ack;
1427	if (tp->snd_wnd > tp->max_sndwnd)
1428	tp->max_sndwnd = tp->snd_wnd;
1429	needoutput = 1;
1430	}
1431
1432	/*
1433	* Process segments with URG.
1434	*/
1435	if ((tiflags & TH_URG) && ti->ti_urp &&
1436	TCPS_HAVERCVDFIN(tp->t_state) == 0)
1437	{
1438	/*
1439	* This is a kludge, but if we receive and accept
1440	* random urgent pointers, we'll crash in
1441	* soreceive. It's hard to imagine someone
1442	* actually wanting to send this much urgent data.
1443	*/
1444	if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen)
1445	{
1446	ti->ti_urp = 0;
1447	tiflags &= ~TH_URG;
1448	goto dodata;
1449	}
1450	/*
1451	* If this segment advances the known urgent pointer,
1452	* then mark the data stream. This should not happen
1453	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1454	* a FIN has been received from the remote side.
1455	* In these states we ignore the URG.
1456	*
1457	* According to RFC961 (Assigned Protocols),
1458	* the urgent pointer points to the last octet
1459	* of urgent data. We continue, however,
1460	* to consider it to indicate the first octet
1461	* of data past the urgent section as the original
1462	* spec states (in one of two places).
1463	*/
1464	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up))
1465	{
1466	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1467	so->so_urgc = so->so_rcv.sb_cc +
1468	(tp->rcv_up - tp->rcv_nxt); /* -1; */
1469	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1470	}
1471	}
1472	else
1473	/*
1474	* If no out of band data is expected,
1475	* pull receive urgent pointer along
1476	* with the receive window.
1477	*/
1478	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1479	tp->rcv_up = tp->rcv_nxt;
1480	dodata:
1481	Log2(("do data hit!\n"));
1482
1483	/*
1484	* If this is a small packet, then ACK now - with Nagel
1485	* congestion avoidance sender won't send more until
1486	* he gets an ACK.
1487	*
1488	* See above.
1489	*/
1490	if ( ti->ti_len
1491	&& (unsigned)ti->ti_len <= 5
1492	&& ((struct tcpiphdr_2 *)ti)->first_char == (char)27)
1493	{
1494	tp->t_flags \|= TF_ACKNOW;
1495	}
1496
1497	/*
1498	* Process the segment text, merging it into the TCP sequencing queue,
1499	* and arranging for acknowledgment of receipt if necessary.
1500	* This process logically involves adjusting tp->rcv_wnd as data
1501	* is presented to the user (this happens in tcp_usrreq.c,
1502	* case PRU_RCVD). If a FIN has already been received on this
1503	* connection then we just ignore the text.
1504	*/
1505	if ( (ti->ti_len \|\| (tiflags&TH_FIN))
1506	&& TCPS_HAVERCVDFIN(tp->t_state) == 0)
1507	{
1508	if ( ti->ti_seq == tp->rcv_nxt
1509	&& LIST_EMPTY(&tp->t_segq)
1510	&& tp->t_state == TCPS_ESTABLISHED)
1511	{
1512	DELAY_ACK(tp, ti); /* little bit different from BSD declaration see netinet/tcp_input.c */
1513	tp->rcv_nxt += tlen;
1514	tiflags = ti->ti_t.th_flags & TH_FIN;
1515	tcpstat.tcps_rcvpack++;
1516	tcpstat.tcps_rcvbyte += tlen;
1517	if (so->so_state & SS_FCANTRCVMORE)
1518	m_freem(pData, m);
1519	else
1520	{
1521	if (so->so_emu)
1522	{
1523	if (tcp_emu(pData, so,m))
1524	sbappend(pData, so, m);
1525	}
1526	else
1527	sbappend(pData, so, m);
1528	}
1529	}
1530	else
1531	{
1532	tiflags = tcp_reass(pData, tp, &ti->ti_t, &tlen, m);
1533	tiflags \|= TF_ACKNOW;
1534	}
1535	/*
1536	* Note the amount of data that peer has sent into
1537	* our window, in order to estimate the sender's
1538	* buffer size.
1539	*/
1540	len = so->so_rcv.sb_datalen - (tp->rcv_adv - tp->rcv_nxt);
1541	}
1542	else
1543	{
1544	m_free(pData, m);
1545	tiflags &= ~TH_FIN;
1546	}
1547
1548	/*
1549	* If FIN is received ACK the FIN and let the user know
1550	* that the connection is closing.
1551	*/
1552	if (tiflags & TH_FIN)
1553	{
1554	if (TCPS_HAVERCVDFIN(tp->t_state) == 0)
1555	{
1556	/*
1557	* If we receive a FIN we can't send more data,
1558	* set it SS_FDRAIN
1559	* Shutdown the socket if there is no rx data in the
1560	* buffer.
1561	* soread() is called on completion of shutdown() and
1562	* will got to TCPS_LAST_ACK, and use tcp_output()
1563	* to send the FIN.
1564	*/
1565	/* sofcantrcvmore(so); */
1566	sofwdrain(so);
1567
1568	tp->t_flags \|= TF_ACKNOW;
1569	tp->rcv_nxt++;
1570	}
1571	switch (tp->t_state)
1572	{
1573	/*
1574	* In SYN_RECEIVED and ESTABLISHED STATES
1575	* enter the CLOSE_WAIT state.
1576	*/
1577	case TCPS_SYN_RECEIVED:
1578	case TCPS_ESTABLISHED:
1579	if(so->so_emu == EMU_CTL) /* no shutdown on socket */
1580	tp->t_state = TCPS_LAST_ACK;
1581	else
1582	tp->t_state = TCPS_CLOSE_WAIT;
1583	break;
1584
1585	/*
1586	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1587	* enter the CLOSING state.
1588	*/
1589	case TCPS_FIN_WAIT_1:
1590	tp->t_state = TCPS_CLOSING;
1591	break;
1592
1593	/*
1594	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1595	* starting the time-wait timer, turning off the other
1596	* standard timers.
1597	*/
1598	case TCPS_FIN_WAIT_2:
1599	tp->t_state = TCPS_TIME_WAIT;
1600	tcp_canceltimers(tp);
1601	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1602	soisfdisconnected(so);
1603	break;
1604
1605	/*
1606	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1607	*/
1608	case TCPS_TIME_WAIT:
1609	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1610	break;
1611	}
1612	}
1613
1614	/*
1615	* Return any desired output.
1616	*/
1617	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1618	tcp_output(pData, tp);
1619
1620	SOCKET_UNLOCK(so);
1621	return;
1622
1623	dropafterack:
1624	Log2(("drop after ack\n"));
1625	/*
1626	* Generate an ACK dropping incoming segment if it occupies
1627	* sequence space, where the ACK reflects our state.
1628	*/
1629	if (tiflags & TH_RST)
1630	goto drop;
1631	m_freem(pData, m);
1632	tp->t_flags \|= TF_ACKNOW;
1633	(void) tcp_output(pData, tp);
1634	SOCKET_UNLOCK(so);
1635	return;
1636
1637	dropwithreset:
1638	/* reuses m if m!=NULL, m_free() unnecessary */
1639	if (tiflags & TH_ACK)
1640	tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1641	else
1642	{
1643	if (tiflags & TH_SYN) ti->ti_len++;
1644	tcp_respond(pData, tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1645	TH_RST\|TH_ACK);
1646	}
1647
1648	if (so != &tcb)
1649	SOCKET_UNLOCK(so);
1650	return;
1651
1652	drop:
1653	/*
1654	* Drop space held by incoming segment and return.
1655	*/
1656	m_free(pData, m);
1657
1658	#ifdef VBOX_WITH_SLIRP_MT
1659	if (RTCritSectIsOwned(&so->so_mutex))
1660	{
1661	SOCKET_UNLOCK(so);
1662	}
1663	#endif
1664
1665	return;
1666	}
1667
1668	void
1669	tcp_dooptions(PNATState pData, struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti)
1670	{
1671	u_int16_t mss;
1672	int opt, optlen;
1673
1674	DEBUG_CALL("tcp_dooptions");
1675	DEBUG_ARGS((dfd," tp = %lx cnt=%i \n", (long )tp, cnt));
1676
1677	for (; cnt > 0; cnt -= optlen, cp += optlen)
1678	{
1679	opt = cp[0];
1680	if (opt == TCPOPT_EOL)
1681	break;
1682	if (opt == TCPOPT_NOP)
1683	optlen = 1;
1684	else
1685	{
1686	optlen = cp[1];
1687	if (optlen <= 0)
1688	break;
1689	}
1690	switch (opt)
1691	{
1692	default:
1693	continue;
1694
1695	case TCPOPT_MAXSEG:
1696	if (optlen != TCPOLEN_MAXSEG)
1697	continue;
1698	if (!(ti->ti_flags & TH_SYN))
1699	continue;
1700	memcpy((char ) &mss, (char ) cp + 2, sizeof(mss));
1701	NTOHS(mss);
1702	(void) tcp_mss(pData, tp, mss); /* sets t_maxseg */
1703	break;
1704
1705	#if 0
1706	case TCPOPT_WINDOW:
1707	if (optlen != TCPOLEN_WINDOW)
1708	continue;
1709	if (!(ti->ti_flags & TH_SYN))
1710	continue;
1711	tp->t_flags \|= TF_RCVD_SCALE;
1712	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1713	break;
1714
1715	case TCPOPT_TIMESTAMP:
1716	if (optlen != TCPOLEN_TIMESTAMP)
1717	continue;
1718	*ts_present = 1;
1719	memcpy((char ) ts_val, (char )cp + 2, sizeof(*ts_val));
1720	NTOHL(*ts_val);
1721	memcpy((char ) ts_ecr, (char )cp + 6, sizeof(*ts_ecr));
1722	NTOHL(*ts_ecr);
1723
1724	/*
1725	* A timestamp received in a SYN makes
1726	* it ok to send timestamp requests and replies.
1727	*/
1728	if (ti->ti_flags & TH_SYN)
1729	{
1730	tp->t_flags \|= TF_RCVD_TSTMP;
1731	tp->ts_recent = *ts_val;
1732	tp->ts_recent_age = tcp_now;
1733	}
1734	break;
1735	#endif
1736	}
1737	}
1738	}
1739
1740
1741	/*
1742	* Pull out of band byte out of a segment so
1743	* it doesn't appear in the user's data queue.
1744	* It is still reflected in the segment length for
1745	* sequencing purposes.
1746	*/
1747
1748	#if 0
1749	void
1750	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1751	{
1752	int cnt = ti->ti_urp - 1;
1753
1754	while (cnt >= 0)
1755	{
1756	if (m->m_len > cnt)
1757	{
1758	char *cp = mtod(m, caddr_t) + cnt;
1759	struct tcpcb *tp = sototcpcb(so);
1760
1761	tp->t_iobc = *cp;
1762	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1763	memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1764	m->m_len--;
1765	return;
1766	}
1767	cnt -= m->m_len;
1768	m = m->m_next; /* XXX WRONG! Fix it! */
1769	if (m == 0)
1770	break;
1771	}
1772	panic("tcp_pulloutofband");
1773	}
1774	#endif
1775
1776	/*
1777	* Collect new round-trip time estimate
1778	* and update averages and current timeout.
1779	*/
1780
1781	void
1782	tcp_xmit_timer(PNATState pData, register struct tcpcb *tp, int rtt)
1783	{
1784	register short delta;
1785
1786	DEBUG_CALL("tcp_xmit_timer");
1787	DEBUG_ARG("tp = %lx", (long)tp);
1788	DEBUG_ARG("rtt = %d", rtt);
1789
1790	tcpstat.tcps_rttupdated++;
1791	if (tp->t_srtt != 0)
1792	{
1793	/*
1794	* srtt is stored as fixed point with 3 bits after the
1795	* binary point (i.e., scaled by 8). The following magic
1796	* is equivalent to the smoothing algorithm in rfc793 with
1797	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1798	* point). Adjust rtt to origin 0.
1799	*/
1800	delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1801	if ((tp->t_srtt += delta) <= 0)
1802	tp->t_srtt = 1;
1803	/*
1804	* We accumulate a smoothed rtt variance (actually, a
1805	* smoothed mean difference), then set the retransmit
1806	* timer to smoothed rtt + 4 times the smoothed variance.
1807	* rttvar is stored as fixed point with 2 bits after the
1808	* binary point (scaled by 4). The following is
1809	* equivalent to rfc793 smoothing with an alpha of .75
1810	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1811	* rfc793's wired-in beta.
1812	*/
1813	if (delta < 0)
1814	delta = -delta;
1815	delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1816	if ((tp->t_rttvar += delta) <= 0)
1817	tp->t_rttvar = 1;
1818	}
1819	else
1820	{
1821	/*
1822	* No rtt measurement yet - use the unsmoothed rtt.
1823	* Set the variance to half the rtt (so our first
1824	* retransmit happens at 3*rtt).
1825	*/
1826	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1827	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1828	}
1829	tp->t_rtt = 0;
1830	tp->t_rxtshift = 0;
1831
1832	/*
1833	* the retransmit should happen at rtt + 4 * rttvar.
1834	* Because of the way we do the smoothing, srtt and rttvar
1835	* will each average +1/2 tick of bias. When we compute
1836	* the retransmit timer, we want 1/2 tick of rounding and
1837	* 1 extra tick because of +-1/2 tick uncertainty in the
1838	* firing of the timer. The bias will give us exactly the
1839	* 1.5 tick we need. But, because the bias is
1840	* statistical, we have to test that we don't drop below
1841	* the minimum feasible timer (which is 2 ticks).
1842	*/
1843	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1844	(short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1845
1846	/*
1847	* We received an ack for a packet that wasn't retransmitted;
1848	* it is probably safe to discard any error indications we've
1849	* received recently. This isn't quite right, but close enough
1850	* for now (a route might have failed after we sent a segment,
1851	* and the return path might not be symmetrical).
1852	*/
1853	tp->t_softerror = 0;
1854	}
1855
1856	/*
1857	* Determine a reasonable value for maxseg size.
1858	* If the route is known, check route for mtu.
1859	* If none, use an mss that can be handled on the outgoing
1860	* interface without forcing IP to fragment; if bigger than
1861	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1862	* to utilize large mbufs. If no route is found, route has no mtu,
1863	* or the destination isn't local, use a default, hopefully conservative
1864	* size (usually 512 or the default IP max size, but no more than the mtu
1865	* of the interface), as we can't discover anything about intervening
1866	* gateways or networks. We also initialize the congestion/slow start
1867	* window to be a single segment if the destination isn't local.
1868	* While looking at the routing entry, we also initialize other path-dependent
1869	* parameters from pre-set or cached values in the routing entry.
1870	*/
1871
1872	int
1873	tcp_mss(PNATState pData, register struct tcpcb *tp, u_int offer)
1874	{
1875	struct socket *so = tp->t_socket;
1876	int mss;
1877
1878	DEBUG_CALL("tcp_mss");
1879	DEBUG_ARG("tp = %lx", (long)tp);
1880	DEBUG_ARG("offer = %d", offer);
1881
1882	mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
1883	if (offer)
1884	mss = min(mss, offer);
1885	mss = max(mss, 32);
1886	if (mss < tp->t_maxseg \|\| offer != 0)
1887	tp->t_maxseg = mss;
1888
1889	tp->snd_cwnd = mss;
1890
1891	sbreserve(&so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
1892	sbreserve(&so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
1893
1894	DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1895
1896	return mss;
1897	}

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 17145

Download in other formats: