tcp_input.c@ 16482

Last change on this file since 16482 was 16450, checked in by vboxsync, 16 years ago
NAT:MT: workaround the lack of unlock call. should be fixed.
Property svn:eol-style set to `native`
File size: 61.4 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
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	}
765	tp = tcp_close(pData, tp);
766	m_free(pData, m);
767	}
768	else
769	{
770	/*
771	* Haven't connected yet, save the current mbuf
772	* and ti, and return
773	* XXX Some OS's don't tell us whether the connect()
774	* succeeded or not. So we must time it out.
775	*/
776	so->so_m = m;
777	so->so_ti = ti;
778	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
779	tp->t_state = TCPS_SYN_RECEIVED;
780	}
781	SOCKET_UNLOCK(so);
782	return;
783
784	cont_conn:
785	/* m==NULL
786	* Check if the connect succeeded
787	*/
788	if (so->so_state & SS_NOFDREF)
789	{
790	tp = tcp_close(pData, tp);
791	goto dropwithreset;
792	}
793	cont_input:
794	tcp_template(tp);
795
796	if (optp)
797	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
798
799	if (iss)
800	tp->iss = iss;
801	else
802	tp->iss = tcp_iss;
803	tcp_iss += TCP_ISSINCR/2;
804	tp->irs = ti->ti_seq;
805	tcp_sendseqinit(tp);
806	tcp_rcvseqinit(tp);
807	tp->t_flags \|= TF_ACKNOW;
808	tp->t_state = TCPS_SYN_RECEIVED;
809	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
810	tcpstat.tcps_accepts++;
811	Log2(("hit trimthenstep6\n"));
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	Log2(("RST(%x) is on SYN_SENT =>drop\n", tiflags));
838	goto drop;
839	}
840
841	if ((tiflags & TH_SYN) == 0)
842	{
843	Log2(("SYN(%x) bit is off on SYN_SENT =>drop\n", tiflags));
844	goto drop;
845	}
846	if (tiflags & TH_ACK)
847	{
848	tp->snd_una = ti->ti_ack;
849	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
850	tp->snd_nxt = tp->snd_una;
851	}
852
853	tp->t_timer[TCPT_REXMT] = 0;
854	tp->irs = ti->ti_seq;
855	tcp_rcvseqinit(tp);
856	tp->t_flags \|= TF_ACKNOW;
857	if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss))
858	{
859	tcpstat.tcps_connects++;
860	soisfconnected(so);
861	tp->t_state = TCPS_ESTABLISHED;
862
863	/* Do window scaling on this connection? */
864	#if 0
865	if (( tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
866	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
867	{
868	tp->snd_scale = tp->requested_s_scale;
869	tp->rcv_scale = tp->request_r_scale;
870	}
871	#endif
872	(void) tcp_reass(pData, tp, (struct tcphdr )0, NULL, (struct mbuf )0);
873	/*
874	* if we didn't have to retransmit the SYN,
875	* use its rtt as our initial srtt & rtt var.
876	*/
877	if (tp->t_rtt)
878	tcp_xmit_timer(pData, tp, tp->t_rtt);
879	}
880	else
881	tp->t_state = TCPS_SYN_RECEIVED;
882
883	trimthenstep6:
884	/*
885	* Advance ti->ti_seq to correspond to first data byte.
886	* If data, trim to stay within window,
887	* dropping FIN if necessary.
888	*/
889	ti->ti_seq++;
890	if (ti->ti_len > tp->rcv_wnd)
891	{
892	todrop = ti->ti_len - tp->rcv_wnd;
893	m_adj(m, -todrop);
894	ti->ti_len = tp->rcv_wnd;
895	tiflags &= ~TH_FIN;
896	tcpstat.tcps_rcvpackafterwin++;
897	tcpstat.tcps_rcvbyteafterwin += todrop;
898	}
899	tp->snd_wl1 = ti->ti_seq - 1;
900	tp->rcv_up = ti->ti_seq;
901	Log2(("hit6"));
902	goto step6;
903	} /* switch tp->t_state */
904	/*
905	* States other than LISTEN or SYN_SENT.
906	* First check timestamp, if present.
907	* Then check that at least some bytes of segment are within
908	* receive window. If segment begins before rcv_nxt,
909	* drop leading data (and SYN); if nothing left, just ack.
910	*
911	* RFC 1323 PAWS: If we have a timestamp reply on this segment
912	* and it's less than ts_recent, drop it.
913	*/
914	#if 0
915	if ( ts_present
916	&& (tiflags & TH_RST) == 0
917	&& tp->ts_recent
918	&& TSTMP_LT(ts_val, tp->ts_recent))
919	{
920	/* Check to see if ts_recent is over 24 days old. */
921	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE)
922	{
923	/*
924	* Invalidate ts_recent. If this segment updates
925	* ts_recent, the age will be reset later and ts_recent
926	* will get a valid value. If it does not, setting
927	* ts_recent to zero will at least satisfy the
928	* requirement that zero be placed in the timestamp
929	* echo reply when ts_recent isn't valid. The
930	* age isn't reset until we get a valid ts_recent
931	* because we don't want out-of-order segments to be
932	* dropped when ts_recent is old.
933	*/
934	tp->ts_recent = 0;
935	}
936	else
937	{
938	tcpstat.tcps_rcvduppack++;
939	tcpstat.tcps_rcvdupbyte += ti->ti_len;
940	tcpstat.tcps_pawsdrop++;
941	goto dropafterack;
942	}
943	}
944	#endif
945
946	todrop = tp->rcv_nxt - ti->ti_seq;
947	if (todrop > 0)
948	{
949	if (tiflags & TH_SYN)
950	{
951	tiflags &= ~TH_SYN;
952	ti->ti_seq++;
953	if (ti->ti_urp > 1)
954	ti->ti_urp--;
955	else
956	tiflags &= ~TH_URG;
957	todrop--;
958	}
959	/*
960	* Following if statement from Stevens, vol. 2, p. 960.
961	*/
962	if ( todrop > ti->ti_len
963	\|\| ( todrop == ti->ti_len
964	&& (tiflags & TH_FIN) == 0))
965	{
966	/*
967	* Any valid FIN must be to the left of the window.
968	* At this point the FIN must be a duplicate or out
969	* of sequence; drop it.
970	*/
971	tiflags &= ~TH_FIN;
972
973	/*
974	* Send an ACK to resynchronize and drop any data.
975	* But keep on processing for RST or ACK.
976	*/
977	tp->t_flags \|= TF_ACKNOW;
978	todrop = ti->ti_len;
979	tcpstat.tcps_rcvduppack++;
980	tcpstat.tcps_rcvdupbyte += todrop;
981	}
982	else
983	{
984	tcpstat.tcps_rcvpartduppack++;
985	tcpstat.tcps_rcvpartdupbyte += todrop;
986	}
987	m_adj(m, todrop);
988	ti->ti_seq += todrop;
989	ti->ti_len -= todrop;
990	if (ti->ti_urp > todrop)
991	ti->ti_urp -= todrop;
992	else
993	{
994	tiflags &= ~TH_URG;
995	ti->ti_urp = 0;
996	}
997	}
998	/*
999	* If new data are received on a connection after the
1000	* user processes are gone, then RST the other end.
1001	*/
1002	if ( (so->so_state & SS_NOFDREF)
1003	&& tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len)
1004	{
1005	tp = tcp_close(pData, tp);
1006	tcpstat.tcps_rcvafterclose++;
1007	goto dropwithreset;
1008	}
1009
1010	/*
1011	* If segment ends after window, drop trailing data
1012	* (and PUSH and FIN); if nothing left, just ACK.
1013	*/
1014	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1015	if (todrop > 0)
1016	{
1017	tcpstat.tcps_rcvpackafterwin++;
1018	if (todrop >= ti->ti_len)
1019	{
1020	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1021	/*
1022	* If a new connection request is received
1023	* while in TIME_WAIT, drop the old connection
1024	* and start over if the sequence numbers
1025	* are above the previous ones.
1026	*/
1027	if ( tiflags & TH_SYN
1028	&& tp->t_state == TCPS_TIME_WAIT
1029	&& SEQ_GT(ti->ti_seq, tp->rcv_nxt))
1030	{
1031	iss = tp->rcv_nxt + TCP_ISSINCR;
1032	tp = tcp_close(pData, tp);
1033	SOCKET_UNLOCK(tp->t_socket);
1034	goto findso;
1035	}
1036	/*
1037	* If window is closed can only take segments at
1038	* window edge, and have to drop data and PUSH from
1039	* incoming segments. Continue processing, but
1040	* remember to ack. Otherwise, drop segment
1041	* and ack.
1042	*/
1043	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt)
1044	{
1045	tp->t_flags \|= TF_ACKNOW;
1046	tcpstat.tcps_rcvwinprobe++;
1047	}
1048	else
1049	goto dropafterack;
1050	}
1051	else
1052	tcpstat.tcps_rcvbyteafterwin += todrop;
1053	m_adj(m, -todrop);
1054	ti->ti_len -= todrop;
1055	tiflags &= ~(TH_PUSH\|TH_FIN);
1056	}
1057
1058	/*
1059	* If last ACK falls within this segment's sequence numbers,
1060	* record its timestamp.
1061	*/
1062	#if 0
1063	if ( ts_present
1064	&& SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)
1065	&& SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + ((tiflags & (TH_SYN\|TH_FIN)) != 0)))
1066	{
1067	tp->ts_recent_age = tcp_now;
1068	tp->ts_recent = ts_val;
1069	}
1070	#endif
1071
1072	/*
1073	* If the RST bit is set examine the state:
1074	* SYN_RECEIVED STATE:
1075	* If passive open, return to LISTEN state.
1076	* If active open, inform user that connection was refused.
1077	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1078	* Inform user that connection was reset, and close tcb.
1079	* CLOSING, LAST_ACK, TIME_WAIT STATES
1080	* Close the tcb.
1081	*/
1082	if (tiflags&TH_RST)
1083	switch (tp->t_state)
1084	{
1085	case TCPS_SYN_RECEIVED:
1086	/* so->so_error = ECONNREFUSED; */
1087	goto close;
1088
1089	case TCPS_ESTABLISHED:
1090	case TCPS_FIN_WAIT_1:
1091	case TCPS_FIN_WAIT_2:
1092	case TCPS_CLOSE_WAIT:
1093	/* so->so_error = ECONNRESET; */
1094	close:
1095	Log2(("closing...=>drop\n", tp->t_state));
1096	tp->t_state = TCPS_CLOSED;
1097	tcpstat.tcps_drops++;
1098	tp = tcp_close(pData, tp);
1099	goto drop;
1100
1101	case TCPS_CLOSING:
1102	case TCPS_LAST_ACK:
1103	case TCPS_TIME_WAIT:
1104	Log2(("t_state is (%x) sort of close =>drop\n", tp->t_state));
1105	tp = tcp_close(pData, tp);
1106	goto drop;
1107	}
1108
1109	/*
1110	* If a SYN is in the window, then this is an
1111	* error and we send an RST and drop the connection.
1112	*/
1113	if (tiflags & TH_SYN)
1114	{
1115	tp = tcp_drop(pData, tp,0);
1116	goto dropwithreset;
1117	}
1118
1119	/*
1120	* If the ACK bit is off we drop the segment and return.
1121	*/
1122	if ((tiflags & TH_ACK) == 0)
1123	{
1124	Log2(("ACK(%x) bit is off =>drop\n", tiflags));
1125	goto drop;
1126	}
1127
1128	/*
1129	* Ack processing.
1130	*/
1131	switch (tp->t_state)
1132	{
1133	/*
1134	* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1135	* ESTABLISHED state and continue processing, otherwise
1136	* send an RST. una<=ack<=max
1137	*/
1138	case TCPS_SYN_RECEIVED:
1139	if ( SEQ_GT(tp->snd_una, ti->ti_ack)
1140	\|\| SEQ_GT(ti->ti_ack, tp->snd_max))
1141	goto dropwithreset;
1142	tcpstat.tcps_connects++;
1143	tp->t_state = TCPS_ESTABLISHED;
1144	/*
1145	* The sent SYN is ack'ed with our sequence number +1
1146	* The first data byte already in the buffer will get
1147	* lost if no correction is made. This is only needed for
1148	* SS_CTL since the buffer is empty otherwise.
1149	* tp->snd_una++; or:
1150	*/
1151	tp->snd_una = ti->ti_ack;
1152	soisfconnected(so);
1153
1154	/* Do window scaling? */
1155	#if 0
1156	if ( (tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
1157	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
1158	{
1159	tp->snd_scale = tp->requested_s_scale;
1160	tp->rcv_scale = tp->request_r_scale;
1161	}
1162	#endif
1163	(void) tcp_reass(pData, tp, (struct tcphdr )0, (int )0, (struct mbuf *)0);
1164	tp->snd_wl1 = ti->ti_seq - 1;
1165	/* Avoid ack processing; snd_una==ti_ack => dup ack */
1166	Log2(("hit synrx_to_est\n"));
1167	goto synrx_to_est;
1168	/* fall into ... */
1169
1170	/*
1171	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1172	* ACKs. If the ack is in the range
1173	* tp->snd_una < ti->ti_ack <= tp->snd_max
1174	* then advance tp->snd_una to ti->ti_ack and drop
1175	* data from the retransmission queue. If this ACK reflects
1176	* more up to date window information we update our window information.
1177	*/
1178	case TCPS_ESTABLISHED:
1179	case TCPS_FIN_WAIT_1:
1180	case TCPS_FIN_WAIT_2:
1181	case TCPS_CLOSE_WAIT:
1182	case TCPS_CLOSING:
1183	case TCPS_LAST_ACK:
1184	case TCPS_TIME_WAIT:
1185	if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
1186	{
1187	if (ti->ti_len == 0 && tiwin == tp->snd_wnd)
1188	{
1189	tcpstat.tcps_rcvdupack++;
1190	DEBUG_MISC((dfd," dup ack m = %lx so = %lx \n",
1191	(long )m, (long )so));
1192	/*
1193	* If we have outstanding data (other than
1194	* a window probe), this is a completely
1195	* duplicate ack (ie, window info didn't
1196	* change), the ack is the biggest we've
1197	* seen and we've seen exactly our rexmt
1198	* threshold of them, assume a packet
1199	* has been dropped and retransmit it.
1200	* Kludge snd_nxt & the congestion
1201	* window so we send only this one
1202	* packet.
1203	*
1204	* We know we're losing at the current
1205	* window size so do congestion avoidance
1206	* (set ssthresh to half the current window
1207	* and pull our congestion window back to
1208	* the new ssthresh).
1209	*
1210	* Dup acks mean that packets have left the
1211	* network (they're now cached at the receiver)
1212	* so bump cwnd by the amount in the receiver
1213	* to keep a constant cwnd packets in the
1214	* network.
1215	*/
1216	if ( tp->t_timer[TCPT_REXMT] == 0
1217	\|\| ti->ti_ack != tp->snd_una)
1218	tp->t_dupacks = 0;
1219	else if (++tp->t_dupacks == tcprexmtthresh)
1220	{
1221	tcp_seq onxt = tp->snd_nxt;
1222	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
1223	if (win < 2)
1224	win = 2;
1225	tp->snd_ssthresh = win * tp->t_maxseg;
1226	tp->t_timer[TCPT_REXMT] = 0;
1227	tp->t_rtt = 0;
1228	tp->snd_nxt = ti->ti_ack;
1229	tp->snd_cwnd = tp->t_maxseg;
1230	(void) tcp_output(pData, tp);
1231	tp->snd_cwnd = tp->snd_ssthresh +
1232	tp->t_maxseg * tp->t_dupacks;
1233	if (SEQ_GT(onxt, tp->snd_nxt))
1234	tp->snd_nxt = onxt;
1235	Log2(("t_dupacks(%d) == tcprexmtthresh(%d)=>drop\n", tp->t_dupacks, tcprexmtthresh));
1236	goto drop;
1237	}
1238	else if (tp->t_dupacks > tcprexmtthresh)
1239	{
1240	tp->snd_cwnd += tp->t_maxseg;
1241	(void) tcp_output(pData, tp);
1242	Log2(("t_dupacks(%d) > tcprexmtthresh(%d)=>drop\n", tp->t_dupacks, tcprexmtthresh));
1243	goto drop;
1244	}
1245	}
1246	else
1247	tp->t_dupacks = 0;
1248	break;
1249	}
1250	synrx_to_est:
1251	Log2(("enter synrx_to_est\n"));
1252	/*
1253	* If the congestion window was inflated to account
1254	* for the other side's cached packets, retract it.
1255	*/
1256	if ( tp->t_dupacks > tcprexmtthresh
1257	&& tp->snd_cwnd > tp->snd_ssthresh)
1258	tp->snd_cwnd = tp->snd_ssthresh;
1259	tp->t_dupacks = 0;
1260	if (SEQ_GT(ti->ti_ack, tp->snd_max))
1261	{
1262	tcpstat.tcps_rcvacktoomuch++;
1263	goto dropafterack;
1264	}
1265	acked = ti->ti_ack - tp->snd_una;
1266	tcpstat.tcps_rcvackpack++;
1267	tcpstat.tcps_rcvackbyte += acked;
1268
1269	/*
1270	* If we have a timestamp reply, update smoothed
1271	* round trip time. If no timestamp is present but
1272	* transmit timer is running and timed sequence
1273	* number was acked, update smoothed round trip time.
1274	* Since we now have an rtt measurement, cancel the
1275	* timer backoff (cf., Phil Karn's retransmit alg.).
1276	* Recompute the initial retransmit timer.
1277	*/
1278	#if 0
1279	if (ts_present)
1280	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1281	else
1282	#endif
1283	if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1284	tcp_xmit_timer(pData, tp,tp->t_rtt);
1285
1286	/*
1287	* If all outstanding data is acked, stop retransmit
1288	* timer and remember to restart (more output or persist).
1289	* If there is more data to be acked, restart retransmit
1290	* timer, using current (possibly backed-off) value.
1291	*/
1292	if (ti->ti_ack == tp->snd_max)
1293	{
1294	tp->t_timer[TCPT_REXMT] = 0;
1295	needoutput = 1;
1296	}
1297	else if (tp->t_timer[TCPT_PERSIST] == 0)
1298	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1299	/*
1300	* When new data is acked, open the congestion window.
1301	* If the window gives us less than ssthresh packets
1302	* in flight, open exponentially (maxseg per packet).
1303	* Otherwise open linearly: maxseg per window
1304	* (maxseg^2 / cwnd per packet).
1305	*/
1306	{
1307	register u_int cw = tp->snd_cwnd;
1308	register u_int incr = tp->t_maxseg;
1309
1310	if (cw > tp->snd_ssthresh)
1311	incr = incr * incr / cw;
1312	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1313	}
1314	if (acked > so->so_snd.sb_cc)
1315	{
1316	tp->snd_wnd -= so->so_snd.sb_cc;
1317	sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1318	ourfinisacked = 1;
1319	}
1320	else
1321	{
1322	sbdrop(&so->so_snd, acked);
1323	tp->snd_wnd -= acked;
1324	ourfinisacked = 0;
1325	}
1326	/*
1327	* XXX sowwakup is called when data is acked and there's room for
1328	* for more data... it should read() the socket
1329	*/
1330	#if 0
1331	if (so->so_snd.sb_flags & SB_NOTIFY)
1332	sowwakeup(so);
1333	#endif
1334	tp->snd_una = ti->ti_ack;
1335	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1336	tp->snd_nxt = tp->snd_una;
1337
1338	switch (tp->t_state)
1339	{
1340	/*
1341	* In FIN_WAIT_1 STATE in addition to the processing
1342	* for the ESTABLISHED state if our FIN is now acknowledged
1343	* then enter FIN_WAIT_2.
1344	*/
1345	case TCPS_FIN_WAIT_1:
1346	if (ourfinisacked)
1347	{
1348	/*
1349	* If we can't receive any more
1350	* data, then closing user can proceed.
1351	* Starting the timer is contrary to the
1352	* specification, but if we don't get a FIN
1353	* we'll hang forever.
1354	*/
1355	if (so->so_state & SS_FCANTRCVMORE)
1356	{
1357	soisfdisconnected(so);
1358	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1359	}
1360	tp->t_state = TCPS_FIN_WAIT_2;
1361	}
1362	break;
1363
1364	/*
1365	* In CLOSING STATE in addition to the processing for
1366	* the ESTABLISHED state if the ACK acknowledges our FIN
1367	* then enter the TIME-WAIT state, otherwise ignore
1368	* the segment.
1369	*/
1370	case TCPS_CLOSING:
1371	if (ourfinisacked)
1372	{
1373	tp->t_state = TCPS_TIME_WAIT;
1374	tcp_canceltimers(tp);
1375	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1376	soisfdisconnected(so);
1377	}
1378	break;
1379
1380	/*
1381	* In LAST_ACK, we may still be waiting for data to drain
1382	* and/or to be acked, as well as for the ack of our FIN.
1383	* If our FIN is now acknowledged, delete the TCB,
1384	* enter the closed state and return.
1385	*/
1386	case TCPS_LAST_ACK:
1387	if (ourfinisacked)
1388	{
1389	Log2(("ourfinisacked=>drop\n"));
1390	tp = tcp_close(pData, tp);
1391	goto drop;
1392	}
1393	break;
1394
1395	/*
1396	* In TIME_WAIT state the only thing that should arrive
1397	* is a retransmission of the remote FIN. Acknowledge
1398	* it and restart the finack timer.
1399	*/
1400	case TCPS_TIME_WAIT:
1401	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1402	goto dropafterack;
1403	}
1404	} /* switch(tp->t_state) */
1405
1406	step6:
1407	/*
1408	* Update window information.
1409	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1410	*/
1411	if ( (tiflags & TH_ACK)
1412	&& ( SEQ_LT(tp->snd_wl1, ti->ti_seq)
1413	\|\| ( tp->snd_wl1 == ti->ti_seq
1414	&& ( SEQ_LT(tp->snd_wl2, ti->ti_ack)
1415	\|\| ( tp->snd_wl2 == ti->ti_ack
1416	&& tiwin > tp->snd_wnd)))))
1417	{
1418	/* keep track of pure window updates */
1419	if ( ti->ti_len == 0
1420	&& tp->snd_wl2 == ti->ti_ack
1421	&& tiwin > tp->snd_wnd)
1422	tcpstat.tcps_rcvwinupd++;
1423	tp->snd_wnd = tiwin;
1424	tp->snd_wl1 = ti->ti_seq;
1425	tp->snd_wl2 = ti->ti_ack;
1426	if (tp->snd_wnd > tp->max_sndwnd)
1427	tp->max_sndwnd = tp->snd_wnd;
1428	needoutput = 1;
1429	}
1430
1431	/*
1432	* Process segments with URG.
1433	*/
1434	if ((tiflags & TH_URG) && ti->ti_urp &&
1435	TCPS_HAVERCVDFIN(tp->t_state) == 0)
1436	{
1437	/*
1438	* This is a kludge, but if we receive and accept
1439	* random urgent pointers, we'll crash in
1440	* soreceive. It's hard to imagine someone
1441	* actually wanting to send this much urgent data.
1442	*/
1443	if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen)
1444	{
1445	ti->ti_urp = 0;
1446	tiflags &= ~TH_URG;
1447	goto dodata;
1448	}
1449	/*
1450	* If this segment advances the known urgent pointer,
1451	* then mark the data stream. This should not happen
1452	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1453	* a FIN has been received from the remote side.
1454	* In these states we ignore the URG.
1455	*
1456	* According to RFC961 (Assigned Protocols),
1457	* the urgent pointer points to the last octet
1458	* of urgent data. We continue, however,
1459	* to consider it to indicate the first octet
1460	* of data past the urgent section as the original
1461	* spec states (in one of two places).
1462	*/
1463	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up))
1464	{
1465	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1466	so->so_urgc = so->so_rcv.sb_cc +
1467	(tp->rcv_up - tp->rcv_nxt); /* -1; */
1468	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1469	}
1470	}
1471	else
1472	/*
1473	* If no out of band data is expected,
1474	* pull receive urgent pointer along
1475	* with the receive window.
1476	*/
1477	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1478	tp->rcv_up = tp->rcv_nxt;
1479	dodata:
1480	Log2(("do data hit!\n"));
1481
1482	/*
1483	* If this is a small packet, then ACK now - with Nagel
1484	* congestion avoidance sender won't send more until
1485	* he gets an ACK.
1486	*
1487	* See above.
1488	*/
1489	if ( ti->ti_len
1490	&& (unsigned)ti->ti_len <= 5
1491	&& ((struct tcpiphdr_2 *)ti)->first_char == (char)27)
1492	{
1493	tp->t_flags \|= TF_ACKNOW;
1494	}
1495
1496	/*
1497	* Process the segment text, merging it into the TCP sequencing queue,
1498	* and arranging for acknowledgment of receipt if necessary.
1499	* This process logically involves adjusting tp->rcv_wnd as data
1500	* is presented to the user (this happens in tcp_usrreq.c,
1501	* case PRU_RCVD). If a FIN has already been received on this
1502	* connection then we just ignore the text.
1503	*/
1504	if ( (ti->ti_len \|\| (tiflags&TH_FIN))
1505	&& TCPS_HAVERCVDFIN(tp->t_state) == 0)
1506	{
1507	if ( ti->ti_seq == tp->rcv_nxt
1508	&& LIST_EMPTY(&tp->t_segq)
1509	&& tp->t_state == TCPS_ESTABLISHED)
1510	{
1511	DELAY_ACK(tp, ti); /* little bit different from BSD declaration see netinet/tcp_input.c */
1512	tp->rcv_nxt += tlen;
1513	tiflags = ti->ti_t.th_flags & TH_FIN;
1514	tcpstat.tcps_rcvpack++;
1515	tcpstat.tcps_rcvbyte += tlen;
1516	if (so->so_state & SS_FCANTRCVMORE)
1517	m_freem(pData, m);
1518	else
1519	{
1520	if (so->so_emu)
1521	{
1522	if (tcp_emu(pData, so,m))
1523	sbappend(pData, so, m);
1524	}
1525	else
1526	sbappend(pData, so, m);
1527	}
1528	}
1529	else
1530	{
1531	tiflags = tcp_reass(pData, tp, &ti->ti_t, &tlen, m);
1532	tiflags \|= TF_ACKNOW;
1533	}
1534	/*
1535	* Note the amount of data that peer has sent into
1536	* our window, in order to estimate the sender's
1537	* buffer size.
1538	*/
1539	len = so->so_rcv.sb_datalen - (tp->rcv_adv - tp->rcv_nxt);
1540	}
1541	else
1542	{
1543	m_free(pData, m);
1544	tiflags &= ~TH_FIN;
1545	}
1546
1547	/*
1548	* If FIN is received ACK the FIN and let the user know
1549	* that the connection is closing.
1550	*/
1551	if (tiflags & TH_FIN)
1552	{
1553	if (TCPS_HAVERCVDFIN(tp->t_state) == 0)
1554	{
1555	/*
1556	* If we receive a FIN we can't send more data,
1557	* set it SS_FDRAIN
1558	* Shutdown the socket if there is no rx data in the
1559	* buffer.
1560	* soread() is called on completion of shutdown() and
1561	* will got to TCPS_LAST_ACK, and use tcp_output()
1562	* to send the FIN.
1563	*/
1564	/* sofcantrcvmore(so); */
1565	sofwdrain(so);
1566
1567	tp->t_flags \|= TF_ACKNOW;
1568	tp->rcv_nxt++;
1569	}
1570	switch (tp->t_state)
1571	{
1572	/*
1573	* In SYN_RECEIVED and ESTABLISHED STATES
1574	* enter the CLOSE_WAIT state.
1575	*/
1576	case TCPS_SYN_RECEIVED:
1577	case TCPS_ESTABLISHED:
1578	if(so->so_emu == EMU_CTL) /* no shutdown on socket */
1579	tp->t_state = TCPS_LAST_ACK;
1580	else
1581	tp->t_state = TCPS_CLOSE_WAIT;
1582	break;
1583
1584	/*
1585	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1586	* enter the CLOSING state.
1587	*/
1588	case TCPS_FIN_WAIT_1:
1589	tp->t_state = TCPS_CLOSING;
1590	break;
1591
1592	/*
1593	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1594	* starting the time-wait timer, turning off the other
1595	* standard timers.
1596	*/
1597	case TCPS_FIN_WAIT_2:
1598	tp->t_state = TCPS_TIME_WAIT;
1599	tcp_canceltimers(tp);
1600	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1601	soisfdisconnected(so);
1602	break;
1603
1604	/*
1605	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1606	*/
1607	case TCPS_TIME_WAIT:
1608	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1609	break;
1610	}
1611	}
1612
1613	/*
1614	* Return any desired output.
1615	*/
1616	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1617	tcp_output(pData, tp);
1618
1619	SOCKET_UNLOCK(so);
1620	return;
1621
1622	dropafterack:
1623	Log2(("drop after ack\n"));
1624	/*
1625	* Generate an ACK dropping incoming segment if it occupies
1626	* sequence space, where the ACK reflects our state.
1627	*/
1628	if (tiflags & TH_RST)
1629	goto drop;
1630	m_freem(pData, m);
1631	tp->t_flags \|= TF_ACKNOW;
1632	(void) tcp_output(pData, tp);
1633	SOCKET_UNLOCK(so);
1634	return;
1635
1636	dropwithreset:
1637	/* reuses m if m!=NULL, m_free() unnecessary */
1638	if (tiflags & TH_ACK)
1639	tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1640	else
1641	{
1642	if (tiflags & TH_SYN) ti->ti_len++;
1643	tcp_respond(pData, tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1644	TH_RST\|TH_ACK);
1645	}
1646
1647	if (so != &tcb)
1648	SOCKET_UNLOCK(so);
1649	return;
1650
1651	drop:
1652	/*
1653	* Drop space held by incoming segment and return.
1654	*/
1655	m_free(pData, m);
1656
1657	#ifdef VBOX_WITH_SLIRP_MT
1658	if (RTCritSectIsOwned(&so->so_mutex))
1659	{
1660	SOCKET_UNLOCK(so);
1661	}
1662	#endif
1663
1664	return;
1665	}
1666
1667	void
1668	tcp_dooptions(PNATState pData, struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti)
1669	{
1670	u_int16_t mss;
1671	int opt, optlen;
1672
1673	DEBUG_CALL("tcp_dooptions");
1674	DEBUG_ARGS((dfd," tp = %lx cnt=%i \n", (long )tp, cnt));
1675
1676	for (; cnt > 0; cnt -= optlen, cp += optlen)
1677	{
1678	opt = cp[0];
1679	if (opt == TCPOPT_EOL)
1680	break;
1681	if (opt == TCPOPT_NOP)
1682	optlen = 1;
1683	else
1684	{
1685	optlen = cp[1];
1686	if (optlen <= 0)
1687	break;
1688	}
1689	switch (opt)
1690	{
1691	default:
1692	continue;
1693
1694	case TCPOPT_MAXSEG:
1695	if (optlen != TCPOLEN_MAXSEG)
1696	continue;
1697	if (!(ti->ti_flags & TH_SYN))
1698	continue;
1699	memcpy((char ) &mss, (char ) cp + 2, sizeof(mss));
1700	NTOHS(mss);
1701	(void) tcp_mss(pData, tp, mss); /* sets t_maxseg */
1702	break;
1703
1704	#if 0
1705	case TCPOPT_WINDOW:
1706	if (optlen != TCPOLEN_WINDOW)
1707	continue;
1708	if (!(ti->ti_flags & TH_SYN))
1709	continue;
1710	tp->t_flags \|= TF_RCVD_SCALE;
1711	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1712	break;
1713
1714	case TCPOPT_TIMESTAMP:
1715	if (optlen != TCPOLEN_TIMESTAMP)
1716	continue;
1717	*ts_present = 1;
1718	memcpy((char ) ts_val, (char )cp + 2, sizeof(*ts_val));
1719	NTOHL(*ts_val);
1720	memcpy((char ) ts_ecr, (char )cp + 6, sizeof(*ts_ecr));
1721	NTOHL(*ts_ecr);
1722
1723	/*
1724	* A timestamp received in a SYN makes
1725	* it ok to send timestamp requests and replies.
1726	*/
1727	if (ti->ti_flags & TH_SYN)
1728	{
1729	tp->t_flags \|= TF_RCVD_TSTMP;
1730	tp->ts_recent = *ts_val;
1731	tp->ts_recent_age = tcp_now;
1732	}
1733	break;
1734	#endif
1735	}
1736	}
1737	}
1738
1739
1740	/*
1741	* Pull out of band byte out of a segment so
1742	* it doesn't appear in the user's data queue.
1743	* It is still reflected in the segment length for
1744	* sequencing purposes.
1745	*/
1746
1747	#if 0
1748	void
1749	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1750	{
1751	int cnt = ti->ti_urp - 1;
1752
1753	while (cnt >= 0)
1754	{
1755	if (m->m_len > cnt)
1756	{
1757	char *cp = mtod(m, caddr_t) + cnt;
1758	struct tcpcb *tp = sototcpcb(so);
1759
1760	tp->t_iobc = *cp;
1761	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1762	memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1763	m->m_len--;
1764	return;
1765	}
1766	cnt -= m->m_len;
1767	m = m->m_next; /* XXX WRONG! Fix it! */
1768	if (m == 0)
1769	break;
1770	}
1771	panic("tcp_pulloutofband");
1772	}
1773	#endif
1774
1775	/*
1776	* Collect new round-trip time estimate
1777	* and update averages and current timeout.
1778	*/
1779
1780	void
1781	tcp_xmit_timer(PNATState pData, register struct tcpcb *tp, int rtt)
1782	{
1783	register short delta;
1784
1785	DEBUG_CALL("tcp_xmit_timer");
1786	DEBUG_ARG("tp = %lx", (long)tp);
1787	DEBUG_ARG("rtt = %d", rtt);
1788
1789	tcpstat.tcps_rttupdated++;
1790	if (tp->t_srtt != 0)
1791	{
1792	/*
1793	* srtt is stored as fixed point with 3 bits after the
1794	* binary point (i.e., scaled by 8). The following magic
1795	* is equivalent to the smoothing algorithm in rfc793 with
1796	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1797	* point). Adjust rtt to origin 0.
1798	*/
1799	delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1800	if ((tp->t_srtt += delta) <= 0)
1801	tp->t_srtt = 1;
1802	/*
1803	* We accumulate a smoothed rtt variance (actually, a
1804	* smoothed mean difference), then set the retransmit
1805	* timer to smoothed rtt + 4 times the smoothed variance.
1806	* rttvar is stored as fixed point with 2 bits after the
1807	* binary point (scaled by 4). The following is
1808	* equivalent to rfc793 smoothing with an alpha of .75
1809	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1810	* rfc793's wired-in beta.
1811	*/
1812	if (delta < 0)
1813	delta = -delta;
1814	delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1815	if ((tp->t_rttvar += delta) <= 0)
1816	tp->t_rttvar = 1;
1817	}
1818	else
1819	{
1820	/*
1821	* No rtt measurement yet - use the unsmoothed rtt.
1822	* Set the variance to half the rtt (so our first
1823	* retransmit happens at 3*rtt).
1824	*/
1825	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1826	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1827	}
1828	tp->t_rtt = 0;
1829	tp->t_rxtshift = 0;
1830
1831	/*
1832	* the retransmit should happen at rtt + 4 * rttvar.
1833	* Because of the way we do the smoothing, srtt and rttvar
1834	* will each average +1/2 tick of bias. When we compute
1835	* the retransmit timer, we want 1/2 tick of rounding and
1836	* 1 extra tick because of +-1/2 tick uncertainty in the
1837	* firing of the timer. The bias will give us exactly the
1838	* 1.5 tick we need. But, because the bias is
1839	* statistical, we have to test that we don't drop below
1840	* the minimum feasible timer (which is 2 ticks).
1841	*/
1842	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1843	(short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1844
1845	/*
1846	* We received an ack for a packet that wasn't retransmitted;
1847	* it is probably safe to discard any error indications we've
1848	* received recently. This isn't quite right, but close enough
1849	* for now (a route might have failed after we sent a segment,
1850	* and the return path might not be symmetrical).
1851	*/
1852	tp->t_softerror = 0;
1853	}
1854
1855	/*
1856	* Determine a reasonable value for maxseg size.
1857	* If the route is known, check route for mtu.
1858	* If none, use an mss that can be handled on the outgoing
1859	* interface without forcing IP to fragment; if bigger than
1860	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1861	* to utilize large mbufs. If no route is found, route has no mtu,
1862	* or the destination isn't local, use a default, hopefully conservative
1863	* size (usually 512 or the default IP max size, but no more than the mtu
1864	* of the interface), as we can't discover anything about intervening
1865	* gateways or networks. We also initialize the congestion/slow start
1866	* window to be a single segment if the destination isn't local.
1867	* While looking at the routing entry, we also initialize other path-dependent
1868	* parameters from pre-set or cached values in the routing entry.
1869	*/
1870
1871	int
1872	tcp_mss(PNATState pData, register struct tcpcb *tp, u_int offer)
1873	{
1874	struct socket *so = tp->t_socket;
1875	int mss;
1876
1877	DEBUG_CALL("tcp_mss");
1878	DEBUG_ARG("tp = %lx", (long)tp);
1879	DEBUG_ARG("offer = %d", offer);
1880
1881	mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
1882	if (offer)
1883	mss = min(mss, offer);
1884	mss = max(mss, 32);
1885	if (mss < tp->t_maxseg \|\| offer != 0)
1886	tp->t_maxseg = mss;
1887
1888	tp->snd_cwnd = mss;
1889
1890	sbreserve(&so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
1891	sbreserve(&so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
1892
1893	DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1894
1895	return mss;
1896	}

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 16482

Download in other formats: