tcp_input.c@ 22584

最後變更在這個檔案從22584是 22584,由 vboxsync 提交於 15 年前
Devices: tab
屬性 svn:eol-style 設為 `native`
檔案大小: 61.5 KB

行
1	/*
2	* Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994
3	* The Regents of the University of California. All rights reserved.
4	*
5	* Redistribution and use in source and binary forms, with or without
6	* modification, are permitted provided that the following conditions
7	* are met:
8	* 1. Redistributions of source code must retain the above copyright
9	* notice, this list of conditions and the following disclaimer.
10	* 2. Redistributions in binary form must reproduce the above copyright
11	* notice, this list of conditions and the following disclaimer in the
12	* documentation and/or other materials provided with the distribution.
13	* 3. All advertising materials mentioning features or use of this software
14	* must display the following acknowledgement:
15	* This product includes software developed by the University of
16	* California, Berkeley and its contributors.
17	* 4. Neither the name of the University nor the names of its contributors
18	* may be used to endorse or promote products derived from this software
19	* without specific prior written permission.
20	*
21	* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22	* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23	* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24	* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25	* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26	* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27	* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28	* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29	* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30	* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31	* SUCH DAMAGE.
32	*
33	* @(#)tcp_input.c 8.5 (Berkeley) 4/10/94
34	* tcp_input.c,v 1.10 1994/10/13 18:36:32 wollman Exp
35	*/
36
37	/*
38	* Changes and additions relating to SLiRP
39	* Copyright (c) 1995 Danny Gasparovski.
40	*
41	* Please read the file COPYRIGHT for the
42	* terms and conditions of the copyright.
43	*/
44
45	#include <slirp.h>
46	#include "ip_icmp.h"
47
48
49	#define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ)
50
51	/* for modulo comparisons of timestamps */
52	#define TSTMP_LT(a, b) ((int)((a)-(b)) < 0)
53	#define TSTMP_GEQ(a, b) ((int)((a)-(b)) >= 0)
54
55	#ifndef TCP_ACK_HACK
56	#define DELAY_ACK(tp, ti) \
57	if (ti->ti_flags & TH_PUSH) \
58	tp->t_flags \|= TF_ACKNOW; \
59	else \
60	tp->t_flags \|= TF_DELACK;
61	#else /* !TCP_ACK_HACK */
62	#define DELAY_ACK(tp, ign) \
63	tp->t_flags \|= TF_DELACK;
64	#endif /* TCP_ACK_HACK */
65
66
67	/*
68	* deps: netinet/tcp_reass.c
69	* tcp_reass_maxqlen = 48 (deafault)
70	* tcp_reass_maxseg = nmbclusters/16 (nmbclusters = 1024 + maxusers * 64 from kern/kern_mbuf.c let's say 256)
71	*/
72	int
73	tcp_reass(PNATState pData, struct tcpcb tp, struct tcphdr th, int tlenp, struct mbuf m)
74	{
75	struct tseg_qent *q;
76	struct tseg_qent *p = NULL;
77	struct tseg_qent *nq;
78	struct tseg_qent *te = NULL;
79	struct socket *so = tp->t_socket;
80	int flags;
81	STAM_PROFILE_START(&pData->StatTCP_reassamble, tcp_reassamble);
82
83	/*
84	* XXX: tcp_reass() is rather inefficient with its data structures
85	* and should be rewritten (see NetBSD for optimizations). While
86	* doing that it should move to its own file tcp_reass.c.
87	*/
88
89	/*
90	* Call with th==NULL after become established to
91	* force pre-ESTABLISHED data up to user socket.
92	*/
93	if (th == NULL)
94	goto present;
95
96	/*
97	* Limit the number of segments in the reassembly queue to prevent
98	* holding on to too many segments (and thus running out of mbufs).
99	* Make sure to let the missing segment through which caused this
100	* queue. Always keep one global queue entry spare to be able to
101	* process the missing segment.
102	*/
103	if ( th->th_seq != tp->rcv_nxt
104	&& ( tcp_reass_qsize + 1 >= tcp_reass_maxseg
105	\|\| tp->t_segqlen >= tcp_reass_maxqlen))
106	{
107	tcp_reass_overflows++;
108	tcpstat.tcps_rcvmemdrop++;
109	m_freem(pData, m);
110	*tlenp = 0;
111	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
112	return (0);
113	}
114
115	/*
116	* Allocate a new queue entry. If we can't, or hit the zone limit
117	* just drop the pkt.
118	*/
119	te = RTMemAlloc(sizeof(struct tseg_qent));
120	if (te == NULL)
121	{
122	tcpstat.tcps_rcvmemdrop++;
123	m_freem(pData, m);
124	*tlenp = 0;
125	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
126	return (0);
127	}
128	tp->t_segqlen++;
129	tcp_reass_qsize++;
130
131	/*
132	* Find a segment which begins after this one does.
133	*/
134	LIST_FOREACH(q, &tp->t_segq, tqe_q)
135	{
136	if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
137	break;
138	p = q;
139	}
140
141	/*
142	* If there is a preceding segment, it may provide some of
143	* our data already. If so, drop the data from the incoming
144	* segment. If it provides all of our data, drop us.
145	*/
146	if (p != NULL)
147	{
148	int i;
149	/* conversion to int (in i) handles seq wraparound */
150	i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
151	if (i > 0)
152	{
153	if (i >= *tlenp)
154	{
155	tcpstat.tcps_rcvduppack++;
156	tcpstat.tcps_rcvdupbyte += *tlenp;
157	m_freem(pData, m);
158	RTMemFree(te);
159	tp->t_segqlen--;
160	tcp_reass_qsize--;
161	/*
162	* Try to present any queued data
163	* at the left window edge to the user.
164	* This is needed after the 3-WHS
165	* completes.
166	*/
167	goto present; /* ??? */
168	}
169	m_adj(m, i);
170	*tlenp -= i;
171	th->th_seq += i;
172	}
173	}
174	tcpstat.tcps_rcvoopack++;
175	tcpstat.tcps_rcvoobyte += *tlenp;
176
177	/*
178	* While we overlap succeeding segments trim them or,
179	* if they are completely covered, dequeue them.
180	*/
181	while (q)
182	{
183	int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
184	if (i <= 0)
185	break;
186	if (i < q->tqe_len)
187	{
188	q->tqe_th->th_seq += i;
189	q->tqe_len -= i;
190	m_adj(q->tqe_m, i);
191	break;
192	}
193
194	nq = LIST_NEXT(q, tqe_q);
195	LIST_REMOVE(q, tqe_q);
196	m_freem(pData, q->tqe_m);
197	RTMemFree(q);
198	tp->t_segqlen--;
199	tcp_reass_qsize--;
200	q = nq;
201	}
202
203	/* Insert the new segment queue entry into place. */
204	te->tqe_m = m;
205	te->tqe_th = th;
206	te->tqe_len = *tlenp;
207
208	if (p == NULL)
209	{
210	LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
211	}
212	else
213	{
214	LIST_INSERT_AFTER(p, te, tqe_q);
215	}
216
217	present:
218	/*
219	* Present data to user, advancing rcv_nxt through
220	* completed sequence space.
221	*/
222	if (!TCPS_HAVEESTABLISHED(tp->t_state))
223	{
224	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
225	return (0);
226	}
227	q = LIST_FIRST(&tp->t_segq);
228	if (!q \|\| q->tqe_th->th_seq != tp->rcv_nxt)
229	{
230	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
231	return (0);
232	}
233	do
234	{
235	tp->rcv_nxt += q->tqe_len;
236	flags = q->tqe_th->th_flags & TH_FIN;
237	nq = LIST_NEXT(q, tqe_q);
238	LIST_REMOVE(q, tqe_q);
239	/* XXX: This place should be checked for the same code in
240	* original BSD code for Slirp and current BSD used SS_FCANTRCVMORE
241	*/
242	if (so->so_state & SS_FCANTSENDMORE)
243	m_freem(pData, q->tqe_m);
244	else
245	{
246	if (so->so_emu)
247	{
248	if (tcp_emu(pData, so, q->tqe_m))
249	sbappend(pData, so, q->tqe_m);
250	}
251	else
252	sbappend(pData, so, q->tqe_m);
253	}
254	RTMemFree(q);
255	tp->t_segqlen--;
256	tcp_reass_qsize--;
257	q = nq;
258	}
259	while (q && q->tqe_th->th_seq == tp->rcv_nxt);
260
261	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
262	return flags;
263	}
264
265	/*
266	* TCP input routine, follows pages 65-76 of the
267	* protocol specification dated September, 1981 very closely.
268	*/
269	void
270	tcp_input(PNATState pData, register struct mbuf m, int iphlen, struct socket inso)
271	{
272	struct ip save_ip, *ip;
273	register struct tcpiphdr *ti;
274	caddr_t optp = NULL;
275	int optlen = 0;
276	int len, tlen, off;
277	register struct tcpcb *tp = 0;
278	register int tiflags;
279	struct socket *so = 0;
280	int todrop, acked, ourfinisacked, needoutput = 0;
281	/* int dropsocket = 0; */
282	int iss = 0;
283	u_long tiwin;
284	/* int ts_present = 0; */
285	STAM_PROFILE_START(&pData->StatTCP_input, counter_input);
286
287	DEBUG_CALL("tcp_input");
288	DEBUG_ARGS((dfd," m = %8lx iphlen = %2d inso = %lx\n",
289	(long )m, iphlen, (long )inso ));
290
291	if (inso != NULL)
292	{
293	QSOCKET_LOCK(tcb);
294	SOCKET_LOCK(inso);
295	QSOCKET_UNLOCK(tcb);
296	}
297	/*
298	* If called with m == 0, then we're continuing the connect
299	*/
300	if (m == NULL)
301	{
302	so = inso;
303	Log4(("NAT: tcp_input: %R[natsock]\n", so));
304	/* Re-set a few variables */
305	tp = sototcpcb(so);
306	m = so->so_m;
307
308	so->so_m = 0;
309	ti = so->so_ti;
310
311	/** @todo (vvl) clarify why it might happens */
312	if (ti == NULL)
313	{
314	LogRel(("NAT: ti is null. can't do any reseting connection actions\n"));
315	/* mbuf should be cleared in sofree called from tcp_close */
316	tcp_close(pData, tp);
317	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
318	return;
319	}
320
321	tiwin = ti->ti_win;
322	tiflags = ti->ti_flags;
323
324	goto cont_conn;
325	}
326
327	tcpstat.tcps_rcvtotal++;
328	/*
329	* Get IP and TCP header together in first mbuf.
330	* Note: IP leaves IP header in first mbuf.
331	*/
332	ti = mtod(m, struct tcpiphdr *);
333	if (iphlen > sizeof(struct ip ))
334	{
335	ip_stripoptions(m, (struct mbuf *)0);
336	iphlen = sizeof(struct ip );
337	}
338	/* XXX Check if too short */
339
340
341	/*
342	* Save a copy of the IP header in case we want restore it
343	* for sending an ICMP error message in response.
344	*/
345	ip = mtod(m, struct ip *);
346	save_ip = *ip;
347	save_ip.ip_len+= iphlen;
348
349	/*
350	* Checksum extended TCP header and data.
351	*/
352	tlen = ((struct ip *)ti)->ip_len;
353	memset(ti->ti_x1, 0, 9);
354	ti->ti_len = htons((u_int16_t)tlen);
355	len = sizeof(struct ip ) + tlen;
356	/* keep checksum for ICMP reply
357	* ti->ti_sum = cksum(m, len);
358	* if (ti->ti_sum) { */
359	if (cksum(m, len))
360	{
361	tcpstat.tcps_rcvbadsum++;
362	goto drop;
363	}
364
365	/*
366	* Check that TCP offset makes sense,
367	* pull out TCP options and adjust length. XXX
368	*/
369	off = ti->ti_off << 2;
370	if ( off < sizeof (struct tcphdr)
371	\|\| off > tlen)
372	{
373	tcpstat.tcps_rcvbadoff++;
374	goto drop;
375	}
376	tlen -= off;
377	ti->ti_len = tlen;
378	if (off > sizeof (struct tcphdr))
379	{
380	optlen = off - sizeof (struct tcphdr);
381	optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
382
383	/*
384	* Do quick retrieval of timestamp options ("options
385	* prediction?"). If timestamp is the only option and it's
386	* formatted as recommended in RFC 1323 appendix A, we
387	* quickly get the values now and not bother calling
388	* tcp_dooptions(), etc.
389	*/
390	#if 0
391	if (( optlen == TCPOLEN_TSTAMP_APPA
392	\|\| ( optlen > TCPOLEN_TSTAMP_APPA
393	&& optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
394	(u_int32_t )optp == htonl(TCPOPT_TSTAMP_HDR) &&
395	(ti->ti_flags & TH_SYN) == 0)
396	{
397	ts_present = 1;
398	ts_val = ntohl((u_int32_t )(optp + 4));
399	ts_ecr = ntohl((u_int32_t )(optp + 8));
400	optp = NULL; / * we have parsed the options * /
401	}
402	#endif
403	}
404	tiflags = ti->ti_flags;
405
406	/*
407	* Convert TCP protocol specific fields to host format.
408	*/
409	NTOHL(ti->ti_seq);
410	NTOHL(ti->ti_ack);
411	NTOHS(ti->ti_win);
412	NTOHS(ti->ti_urp);
413
414	/*
415	* Drop TCP, IP headers and TCP options.
416	*/
417	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
418	m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
419
420	/*
421	* Locate pcb for segment.
422	*/
423	findso:
424	if (so != NULL && so != &tcb)
425	SOCKET_UNLOCK(so);
426	QSOCKET_LOCK(tcb);
427	so = tcp_last_so;
428	if ( so->so_fport != ti->ti_dport
429	\|\| so->so_lport != ti->ti_sport
430	\|\| so->so_laddr.s_addr != ti->ti_src.s_addr
431	\|\| so->so_faddr.s_addr != ti->ti_dst.s_addr)
432	{
433	struct socket *sonxt;
434	QSOCKET_UNLOCK(tcb);
435	/* @todo fix SOLOOKUP macrodefinition to be usable here */
436	#ifndef VBOX_WITH_SLIRP_MT
437	so = solookup(&tcb, ti->ti_src, ti->ti_sport,
438	ti->ti_dst, ti->ti_dport);
439	#else
440	so = NULL;
441	QSOCKET_FOREACH(so, sonxt, tcp)
442	/* { */
443	if ( so->so_lport == ti->ti_sport
444	&& so->so_laddr.s_addr == ti->ti_src.s_addr
445	&& so->so_faddr.s_addr == ti->ti_dst.s_addr
446	&& so->so_fport == ti->ti_dport
447	&& so->so_deleted != 1)
448	{
449	break; /* so is locked here */
450	}
451	LOOP_LABEL(tcp, so, sonxt);
452	}
453	if (so == &tcb) {
454	so = NULL;
455	}
456	#endif
457	if (so)
458	{
459	tcp_last_so = so;
460	}
461	++tcpstat.tcps_socachemiss;
462	}
463	else
464	{
465	SOCKET_LOCK(so);
466	QSOCKET_UNLOCK(tcb);
467	}
468
469	/*
470	* If the state is CLOSED (i.e., TCB does not exist) then
471	* all data in the incoming segment is discarded.
472	* If the TCB exists but is in CLOSED state, it is embryonic,
473	* but should either do a listen or a connect soon.
474	*
475	* state == CLOSED means we've done socreate() but haven't
476	* attached it to a protocol yet...
477	*
478	* XXX If a TCB does not exist, and the TH_SYN flag is
479	* the only flag set, then create a session, mark it
480	* as if it was LISTENING, and continue...
481	*/
482	if (so == 0)
483	{
484	if ((tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) != TH_SYN)
485	goto dropwithreset;
486
487	if ((so = socreate()) == NULL)
488	goto dropwithreset;
489	if (tcp_attach(pData, so) < 0)
490	{
491	RTMemFree(so); /* Not sofree (if it failed, it's not insqued) */
492	goto dropwithreset;
493	}
494	SOCKET_LOCK(so);
495	sbreserve(pData, &so->so_snd, tcp_sndspace);
496	sbreserve(pData, &so->so_rcv, tcp_rcvspace);
497
498	/* tcp_last_so = so; / / XXX ? */
499	/* tp = sototcpcb(so); */
500
501	so->so_laddr = ti->ti_src;
502	so->so_lport = ti->ti_sport;
503	so->so_faddr = ti->ti_dst;
504	so->so_fport = ti->ti_dport;
505
506	if ((so->so_iptos = tcp_tos(so)) == 0)
507	so->so_iptos = ((struct ip *)ti)->ip_tos;
508
509	tp = sototcpcb(so);
510	tp->t_state = TCPS_LISTEN;
511	}
512
513	/*
514	* If this is a still-connecting socket, this probably
515	* a retransmit of the SYN. Whether it's a retransmit SYN
516	* or something else, we nuke it.
517	*/
518	if (so->so_state & SS_ISFCONNECTING)
519	{
520	goto drop;
521	}
522
523	tp = sototcpcb(so);
524
525	/* XXX Should never fail */
526	if (tp == 0)
527	goto dropwithreset;
528	if (tp->t_state == TCPS_CLOSED)
529	{
530	goto drop;
531	}
532
533	/* Unscale the window into a 32-bit value. */
534	/* if ((tiflags & TH_SYN) == 0)
535	* tiwin = ti->ti_win << tp->snd_scale;
536	* else
537	*/
538	tiwin = ti->ti_win;
539
540	/*
541	* Segment received on connection.
542	* Reset idle time and keep-alive timer.
543	*/
544	tp->t_idle = 0;
545	if (so_options)
546	tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
547	else
548	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
549
550	/*
551	* Process options if not in LISTEN state,
552	* else do it below (after getting remote address).
553	*/
554	if (optp && tp->t_state != TCPS_LISTEN)
555	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
556	/* , */
557	/* &ts_present, &ts_val, &ts_ecr); */
558
559	/*
560	* Header prediction: check for the two common cases
561	* of a uni-directional data xfer. If the packet has
562	* no control flags, is in-sequence, the window didn't
563	* change and we're not retransmitting, it's a
564	* candidate. If the length is zero and the ack moved
565	* forward, we're the sender side of the xfer. Just
566	* free the data acked & wake any higher level process
567	* that was blocked waiting for space. If the length
568	* is non-zero and the ack didn't move, we're the
569	* receiver side. If we're getting packets in-order
570	* (the reassembly queue is empty), add the data to
571	* the socket buffer and note that we need a delayed ack.
572	*
573	* XXX Some of these tests are not needed
574	* eg: the tiwin == tp->snd_wnd prevents many more
575	* predictions.. with no real advantage..
576	*/
577	if ( tp->t_state == TCPS_ESTABLISHED
578	&& (tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK
579	/* && (!ts_present \|\| TSTMP_GEQ(ts_val, tp->ts_recent)) */
580	&& ti->ti_seq == tp->rcv_nxt
581	&& tiwin && tiwin == tp->snd_wnd
582	&& tp->snd_nxt == tp->snd_max)
583	{
584	/*
585	* If last ACK falls within this segment's sequence numbers,
586	* record the timestamp.
587	*/
588	#if 0
589	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
590	SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len))
591	{
592	tp->ts_recent_age = tcp_now;
593	tp->ts_recent = ts_val;
594	}
595	#endif
596
597	if (ti->ti_len == 0)
598	{
599	if ( SEQ_GT(ti->ti_ack, tp->snd_una)
600	&& SEQ_LEQ(ti->ti_ack, tp->snd_max)
601	&& tp->snd_cwnd >= tp->snd_wnd)
602	{
603	/*
604	* this is a pure ack for outstanding data.
605	*/
606	++tcpstat.tcps_predack;
607	#if 0
608	if (ts_present)
609	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
610	else
611	#endif
612	if ( tp->t_rtt
613	&& SEQ_GT(ti->ti_ack, tp->t_rtseq))
614	tcp_xmit_timer(pData, tp, tp->t_rtt);
615	acked = ti->ti_ack - tp->snd_una;
616	tcpstat.tcps_rcvackpack++;
617	tcpstat.tcps_rcvackbyte += acked;
618	sbdrop(&so->so_snd, acked);
619	tp->snd_una = ti->ti_ack;
620	m_freem(pData, m);
621
622	/*
623	* If all outstanding data are acked, stop
624	* retransmit timer, otherwise restart timer
625	* using current (possibly backed-off) value.
626	* If process is waiting for space,
627	* wakeup/selwakeup/signal. If data
628	* are ready to send, let tcp_output
629	* decide between more output or persist.
630	*/
631	if (tp->snd_una == tp->snd_max)
632	tp->t_timer[TCPT_REXMT] = 0;
633	else if (tp->t_timer[TCPT_PERSIST] == 0)
634	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
635
636	/*
637	* There's room in so_snd, sowwakup will read()
638	* from the socket if we can
639	*/
640	#if 0
641	if (so->so_snd.sb_flags & SB_NOTIFY)
642	sowwakeup(so);
643	#endif
644	/*
645	* This is called because sowwakeup might have
646	* put data into so_snd. Since we don't so sowwakeup,
647	* we don't need this.. XXX???
648	*/
649	if (so->so_snd.sb_cc)
650	(void) tcp_output(pData, tp);
651
652	SOCKET_UNLOCK(so);
653	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
654	return;
655	}
656	}
657	else if ( ti->ti_ack == tp->snd_una
658	&& LIST_FIRST(&tp->t_segq)
659	&& ti->ti_len <= sbspace(&so->so_rcv))
660	{
661	/*
662	* this is a pure, in-sequence data packet
663	* with nothing on the reassembly queue and
664	* we have enough buffer space to take it.
665	*/
666	++tcpstat.tcps_preddat;
667	tp->rcv_nxt += ti->ti_len;
668	tcpstat.tcps_rcvpack++;
669	tcpstat.tcps_rcvbyte += ti->ti_len;
670	/*
671	* Add data to socket buffer.
672	*/
673	if (so->so_emu)
674	{
675	if (tcp_emu(pData, so, m))
676	sbappend(pData, so, m);
677	}
678	else
679	sbappend(pData, so, m);
680
681	/*
682	* XXX This is called when data arrives. Later, check
683	* if we can actually write() to the socket
684	* XXX Need to check? It's be NON_BLOCKING
685	*/
686	/* sorwakeup(so); */
687
688	/*
689	* If this is a short packet, then ACK now - with Nagel
690	* congestion avoidance sender won't send more until
691	* he gets an ACK.
692	*
693	* It is better to not delay acks at all to maximize
694	* TCP throughput. See RFC 2581.
695	*/
696	tp->t_flags \|= TF_ACKNOW;
697	tcp_output(pData, tp);
698	SOCKET_UNLOCK(so);
699	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
700	return;
701	}
702	} /* header prediction */
703	/*
704	* Calculate amount of space in receive window,
705	* and then do TCP input processing.
706	* Receive window is amount of space in rcv queue,
707	* but not less than advertised window.
708	*/
709	{
710	int win;
711	win = sbspace(&so->so_rcv);
712	if (win < 0)
713	win = 0;
714	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
715	}
716
717	switch (tp->t_state)
718	{
719	/*
720	* If the state is LISTEN then ignore segment if it contains an RST.
721	* If the segment contains an ACK then it is bad and send a RST.
722	* If it does not contain a SYN then it is not interesting; drop it.
723	* Don't bother responding if the destination was a broadcast.
724	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
725	* tp->iss, and send a segment:
726	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
727	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
728	* Fill in remote peer address fields if not previously specified.
729	* Enter SYN_RECEIVED state, and process any other fields of this
730	* segment in this state.
731	*/
732	case TCPS_LISTEN:
733	{
734	if (tiflags & TH_RST) {
735	goto drop;
736	}
737	if (tiflags & TH_ACK)
738	goto dropwithreset;
739	if ((tiflags & TH_SYN) == 0)
740	{
741	goto drop;
742	}
743
744	/*
745	* This has way too many gotos...
746	* But a bit of spaghetti code never hurt anybody :)
747	*/
748
749	if (so->so_emu & EMU_NOCONNECT)
750	{
751	so->so_emu &= ~EMU_NOCONNECT;
752	goto cont_input;
753	}
754
755	if ( (tcp_fconnect(pData, so) == -1)
756	&& errno != EINPROGRESS
757	&& errno != EWOULDBLOCK)
758	{
759	u_char code = ICMP_UNREACH_NET;
760	DEBUG_MISC((dfd," tcp fconnect errno = %d-%s\n",
761	errno, strerror(errno)));
762	if (errno == ECONNREFUSED)
763	{
764	/* ACK the SYN, send RST to refuse the connection */
765	tcp_respond(pData, tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
766	TH_RST\|TH_ACK);
767	}
768	else
769	{
770	if (errno == EHOSTUNREACH)
771	code = ICMP_UNREACH_HOST;
772	HTONL(ti->ti_seq); /* restore tcp header */
773	HTONL(ti->ti_ack);
774	HTONS(ti->ti_win);
775	HTONS(ti->ti_urp);
776	m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
777	m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
778	*ip = save_ip;
779	icmp_error(pData, m, ICMP_UNREACH, code, 0, strerror(errno));
780	tp->t_socket->so_m = NULL;
781	}
782	tp = tcp_close(pData, tp);
783	m_free(pData, m);
784	}
785	else
786	{
787	/*
788	* Haven't connected yet, save the current mbuf
789	* and ti, and return
790	* XXX Some OS's don't tell us whether the connect()
791	* succeeded or not. So we must time it out.
792	*/
793	so->so_m = m;
794	so->so_ti = ti;
795	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
796	tp->t_state = TCPS_SYN_RECEIVED;
797	}
798	SOCKET_UNLOCK(so);
799	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
800	return;
801
802	cont_conn:
803	/* m==NULL
804	* Check if the connect succeeded
805	*/
806	if (so->so_state & SS_NOFDREF)
807	{
808	tp = tcp_close(pData, tp);
809	goto dropwithreset;
810	}
811	cont_input:
812	tcp_template(tp);
813
814	if (optp)
815	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
816
817	if (iss)
818	tp->iss = iss;
819	else
820	tp->iss = tcp_iss;
821	tcp_iss += TCP_ISSINCR/2;
822	tp->irs = ti->ti_seq;
823	tcp_sendseqinit(tp);
824	tcp_rcvseqinit(tp);
825	tp->t_flags \|= TF_ACKNOW;
826	tp->t_state = TCPS_SYN_RECEIVED;
827	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
828	tcpstat.tcps_accepts++;
829	goto trimthenstep6;
830	} /* case TCPS_LISTEN */
831
832	/*
833	* If the state is SYN_SENT:
834	* if seg contains an ACK, but not for our SYN, drop the input.
835	* if seg contains a RST, then drop the connection.
836	* if seg does not contain SYN, then drop it.
837	* Otherwise this is an acceptable SYN segment
838	* initialize tp->rcv_nxt and tp->irs
839	* if seg contains ack then advance tp->snd_una
840	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
841	* arrange for segment to be acked (eventually)
842	* continue processing rest of data/controls, beginning with URG
843	*/
844	case TCPS_SYN_SENT:
845	if ( (tiflags & TH_ACK)
846	&& ( SEQ_LEQ(ti->ti_ack, tp->iss)
847	\|\| SEQ_GT(ti->ti_ack, tp->snd_max)))
848	goto dropwithreset;
849
850	if (tiflags & TH_RST)
851	{
852	if (tiflags & TH_ACK)
853	tp = tcp_drop(pData, tp, 0); /* XXX Check t_softerror! */
854	goto drop;
855	}
856
857	if ((tiflags & TH_SYN) == 0)
858	{
859	goto drop;
860	}
861	if (tiflags & TH_ACK)
862	{
863	tp->snd_una = ti->ti_ack;
864	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
865	tp->snd_nxt = tp->snd_una;
866	}
867
868	tp->t_timer[TCPT_REXMT] = 0;
869	tp->irs = ti->ti_seq;
870	tcp_rcvseqinit(tp);
871	tp->t_flags \|= TF_ACKNOW;
872	if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss))
873	{
874	tcpstat.tcps_connects++;
875	soisfconnected(so);
876	tp->t_state = TCPS_ESTABLISHED;
877
878	/* Do window scaling on this connection? */
879	#if 0
880	if (( tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
881	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
882	{
883	tp->snd_scale = tp->requested_s_scale;
884	tp->rcv_scale = tp->request_r_scale;
885	}
886	#endif
887	(void) tcp_reass(pData, tp, (struct tcphdr )0, NULL, (struct mbuf )0);
888	/*
889	* if we didn't have to retransmit the SYN,
890	* use its rtt as our initial srtt & rtt var.
891	*/
892	if (tp->t_rtt)
893	tcp_xmit_timer(pData, tp, tp->t_rtt);
894	}
895	else
896	tp->t_state = TCPS_SYN_RECEIVED;
897
898	trimthenstep6:
899	/*
900	* Advance ti->ti_seq to correspond to first data byte.
901	* If data, trim to stay within window,
902	* dropping FIN if necessary.
903	*/
904	ti->ti_seq++;
905	if (ti->ti_len > tp->rcv_wnd)
906	{
907	todrop = ti->ti_len - tp->rcv_wnd;
908	m_adj(m, -todrop);
909	ti->ti_len = tp->rcv_wnd;
910	tiflags &= ~TH_FIN;
911	tcpstat.tcps_rcvpackafterwin++;
912	tcpstat.tcps_rcvbyteafterwin += todrop;
913	}
914	tp->snd_wl1 = ti->ti_seq - 1;
915	tp->rcv_up = ti->ti_seq;
916	Log2(("hit6"));
917	goto step6;
918	} /* switch tp->t_state */
919	/*
920	* States other than LISTEN or SYN_SENT.
921	* First check timestamp, if present.
922	* Then check that at least some bytes of segment are within
923	* receive window. If segment begins before rcv_nxt,
924	* drop leading data (and SYN); if nothing left, just ack.
925	*
926	* RFC 1323 PAWS: If we have a timestamp reply on this segment
927	* and it's less than ts_recent, drop it.
928	*/
929	#if 0
930	if ( ts_present
931	&& (tiflags & TH_RST) == 0
932	&& tp->ts_recent
933	&& TSTMP_LT(ts_val, tp->ts_recent))
934	{
935	/* Check to see if ts_recent is over 24 days old. */
936	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE)
937	{
938	/*
939	* Invalidate ts_recent. If this segment updates
940	* ts_recent, the age will be reset later and ts_recent
941	* will get a valid value. If it does not, setting
942	* ts_recent to zero will at least satisfy the
943	* requirement that zero be placed in the timestamp
944	* echo reply when ts_recent isn't valid. The
945	* age isn't reset until we get a valid ts_recent
946	* because we don't want out-of-order segments to be
947	* dropped when ts_recent is old.
948	*/
949	tp->ts_recent = 0;
950	}
951	else
952	{
953	tcpstat.tcps_rcvduppack++;
954	tcpstat.tcps_rcvdupbyte += ti->ti_len;
955	tcpstat.tcps_pawsdrop++;
956	goto dropafterack;
957	}
958	}
959	#endif
960
961	todrop = tp->rcv_nxt - ti->ti_seq;
962	if (todrop > 0)
963	{
964	if (tiflags & TH_SYN)
965	{
966	tiflags &= ~TH_SYN;
967	ti->ti_seq++;
968	if (ti->ti_urp > 1)
969	ti->ti_urp--;
970	else
971	tiflags &= ~TH_URG;
972	todrop--;
973	}
974	/*
975	* Following if statement from Stevens, vol. 2, p. 960.
976	*/
977	if ( todrop > ti->ti_len
978	\|\| ( todrop == ti->ti_len
979	&& (tiflags & TH_FIN) == 0))
980	{
981	/*
982	* Any valid FIN must be to the left of the window.
983	* At this point the FIN must be a duplicate or out
984	* of sequence; drop it.
985	*/
986	tiflags &= ~TH_FIN;
987
988	/*
989	* Send an ACK to resynchronize and drop any data.
990	* But keep on processing for RST or ACK.
991	*/
992	tp->t_flags \|= TF_ACKNOW;
993	todrop = ti->ti_len;
994	tcpstat.tcps_rcvduppack++;
995	tcpstat.tcps_rcvdupbyte += todrop;
996	}
997	else
998	{
999	tcpstat.tcps_rcvpartduppack++;
1000	tcpstat.tcps_rcvpartdupbyte += todrop;
1001	}
1002	m_adj(m, todrop);
1003	ti->ti_seq += todrop;
1004	ti->ti_len -= todrop;
1005	if (ti->ti_urp > todrop)
1006	ti->ti_urp -= todrop;
1007	else
1008	{
1009	tiflags &= ~TH_URG;
1010	ti->ti_urp = 0;
1011	}
1012	}
1013	/*
1014	* If new data are received on a connection after the
1015	* user processes are gone, then RST the other end.
1016	*/
1017	if ( (so->so_state & SS_NOFDREF)
1018	&& tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len)
1019	{
1020	tp = tcp_close(pData, tp);
1021	tcpstat.tcps_rcvafterclose++;
1022	goto dropwithreset;
1023	}
1024
1025	/*
1026	* If segment ends after window, drop trailing data
1027	* (and PUSH and FIN); if nothing left, just ACK.
1028	*/
1029	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1030	if (todrop > 0)
1031	{
1032	tcpstat.tcps_rcvpackafterwin++;
1033	if (todrop >= ti->ti_len)
1034	{
1035	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1036	/*
1037	* If a new connection request is received
1038	* while in TIME_WAIT, drop the old connection
1039	* and start over if the sequence numbers
1040	* are above the previous ones.
1041	*/
1042	if ( tiflags & TH_SYN
1043	&& tp->t_state == TCPS_TIME_WAIT
1044	&& SEQ_GT(ti->ti_seq, tp->rcv_nxt))
1045	{
1046	iss = tp->rcv_nxt + TCP_ISSINCR;
1047	tp = tcp_close(pData, tp);
1048	SOCKET_UNLOCK(tp->t_socket);
1049	goto findso;
1050	}
1051	/*
1052	* If window is closed can only take segments at
1053	* window edge, and have to drop data and PUSH from
1054	* incoming segments. Continue processing, but
1055	* remember to ack. Otherwise, drop segment
1056	* and ack.
1057	*/
1058	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt)
1059	{
1060	tp->t_flags \|= TF_ACKNOW;
1061	tcpstat.tcps_rcvwinprobe++;
1062	}
1063	else
1064	goto dropafterack;
1065	}
1066	else
1067	tcpstat.tcps_rcvbyteafterwin += todrop;
1068	m_adj(m, -todrop);
1069	ti->ti_len -= todrop;
1070	tiflags &= ~(TH_PUSH\|TH_FIN);
1071	}
1072
1073	/*
1074	* If last ACK falls within this segment's sequence numbers,
1075	* record its timestamp.
1076	*/
1077	#if 0
1078	if ( ts_present
1079	&& SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)
1080	&& SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + ((tiflags & (TH_SYN\|TH_FIN)) != 0)))
1081	{
1082	tp->ts_recent_age = tcp_now;
1083	tp->ts_recent = ts_val;
1084	}
1085	#endif
1086
1087	/*
1088	* If the RST bit is set examine the state:
1089	* SYN_RECEIVED STATE:
1090	* If passive open, return to LISTEN state.
1091	* If active open, inform user that connection was refused.
1092	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1093	* Inform user that connection was reset, and close tcb.
1094	* CLOSING, LAST_ACK, TIME_WAIT STATES
1095	* Close the tcb.
1096	*/
1097	if (tiflags&TH_RST)
1098	switch (tp->t_state)
1099	{
1100	case TCPS_SYN_RECEIVED:
1101	/* so->so_error = ECONNREFUSED; */
1102	goto close;
1103
1104	case TCPS_ESTABLISHED:
1105	case TCPS_FIN_WAIT_1:
1106	case TCPS_FIN_WAIT_2:
1107	case TCPS_CLOSE_WAIT:
1108	/* so->so_error = ECONNRESET; */
1109	close:
1110	tp->t_state = TCPS_CLOSED;
1111	tcpstat.tcps_drops++;
1112	tp = tcp_close(pData, tp);
1113	goto drop;
1114
1115	case TCPS_CLOSING:
1116	case TCPS_LAST_ACK:
1117	case TCPS_TIME_WAIT:
1118	tp = tcp_close(pData, tp);
1119	goto drop;
1120	}
1121
1122	/*
1123	* If a SYN is in the window, then this is an
1124	* error and we send an RST and drop the connection.
1125	*/
1126	if (tiflags & TH_SYN)
1127	{
1128	tp = tcp_drop(pData, tp, 0);
1129	goto dropwithreset;
1130	}
1131
1132	/*
1133	* If the ACK bit is off we drop the segment and return.
1134	*/
1135	if ((tiflags & TH_ACK) == 0)
1136	{
1137	goto drop;
1138	}
1139
1140	/*
1141	* Ack processing.
1142	*/
1143	switch (tp->t_state)
1144	{
1145	/*
1146	* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1147	* ESTABLISHED state and continue processing, otherwise
1148	* send an RST. una<=ack<=max
1149	*/
1150	case TCPS_SYN_RECEIVED:
1151	if ( SEQ_GT(tp->snd_una, ti->ti_ack)
1152	\|\| SEQ_GT(ti->ti_ack, tp->snd_max))
1153	goto dropwithreset;
1154	tcpstat.tcps_connects++;
1155	tp->t_state = TCPS_ESTABLISHED;
1156	/*
1157	* The sent SYN is ack'ed with our sequence number +1
1158	* The first data byte already in the buffer will get
1159	* lost if no correction is made. This is only needed for
1160	* SS_CTL since the buffer is empty otherwise.
1161	* tp->snd_una++; or:
1162	*/
1163	tp->snd_una = ti->ti_ack;
1164	soisfconnected(so);
1165
1166	/* Do window scaling? */
1167	#if 0
1168	if ( (tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
1169	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
1170	{
1171	tp->snd_scale = tp->requested_s_scale;
1172	tp->rcv_scale = tp->request_r_scale;
1173	}
1174	#endif
1175	(void) tcp_reass(pData, tp, (struct tcphdr )0, (int )0, (struct mbuf *)0);
1176	tp->snd_wl1 = ti->ti_seq - 1;
1177	/* Avoid ack processing; snd_una==ti_ack => dup ack */
1178	goto synrx_to_est;
1179	/* fall into ... */
1180
1181	/*
1182	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1183	* ACKs. If the ack is in the range
1184	* tp->snd_una < ti->ti_ack <= tp->snd_max
1185	* then advance tp->snd_una to ti->ti_ack and drop
1186	* data from the retransmission queue. If this ACK reflects
1187	* more up to date window information we update our window information.
1188	*/
1189	case TCPS_ESTABLISHED:
1190	case TCPS_FIN_WAIT_1:
1191	case TCPS_FIN_WAIT_2:
1192	case TCPS_CLOSE_WAIT:
1193	case TCPS_CLOSING:
1194	case TCPS_LAST_ACK:
1195	case TCPS_TIME_WAIT:
1196	if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
1197	{
1198	if (ti->ti_len == 0 && tiwin == tp->snd_wnd)
1199	{
1200	tcpstat.tcps_rcvdupack++;
1201	DEBUG_MISC((dfd," dup ack m = %lx so = %lx \n",
1202	(long )m, (long )so));
1203	/*
1204	* If we have outstanding data (other than
1205	* a window probe), this is a completely
1206	* duplicate ack (ie, window info didn't
1207	* change), the ack is the biggest we've
1208	* seen and we've seen exactly our rexmt
1209	* threshold of them, assume a packet
1210	* has been dropped and retransmit it.
1211	* Kludge snd_nxt & the congestion
1212	* window so we send only this one
1213	* packet.
1214	*
1215	* We know we're losing at the current
1216	* window size so do congestion avoidance
1217	* (set ssthresh to half the current window
1218	* and pull our congestion window back to
1219	* the new ssthresh).
1220	*
1221	* Dup acks mean that packets have left the
1222	* network (they're now cached at the receiver)
1223	* so bump cwnd by the amount in the receiver
1224	* to keep a constant cwnd packets in the
1225	* network.
1226	*/
1227	if ( tp->t_timer[TCPT_REXMT] == 0
1228	\|\| ti->ti_ack != tp->snd_una)
1229	tp->t_dupacks = 0;
1230	else if (++tp->t_dupacks == tcprexmtthresh)
1231	{
1232	tcp_seq onxt = tp->snd_nxt;
1233	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
1234	if (win < 2)
1235	win = 2;
1236	tp->snd_ssthresh = win * tp->t_maxseg;
1237	tp->t_timer[TCPT_REXMT] = 0;
1238	tp->t_rtt = 0;
1239	tp->snd_nxt = ti->ti_ack;
1240	tp->snd_cwnd = tp->t_maxseg;
1241	(void) tcp_output(pData, tp);
1242	tp->snd_cwnd = tp->snd_ssthresh +
1243	tp->t_maxseg * tp->t_dupacks;
1244	if (SEQ_GT(onxt, tp->snd_nxt))
1245	tp->snd_nxt = onxt;
1246	goto drop;
1247	}
1248	else if (tp->t_dupacks > tcprexmtthresh)
1249	{
1250	tp->snd_cwnd += tp->t_maxseg;
1251	(void) tcp_output(pData, tp);
1252	goto drop;
1253	}
1254	}
1255	else
1256	tp->t_dupacks = 0;
1257	break;
1258	}
1259	synrx_to_est:
1260	/*
1261	* If the congestion window was inflated to account
1262	* for the other side's cached packets, retract it.
1263	*/
1264	if ( tp->t_dupacks > tcprexmtthresh
1265	&& tp->snd_cwnd > tp->snd_ssthresh)
1266	tp->snd_cwnd = tp->snd_ssthresh;
1267	tp->t_dupacks = 0;
1268	if (SEQ_GT(ti->ti_ack, tp->snd_max))
1269	{
1270	tcpstat.tcps_rcvacktoomuch++;
1271	goto dropafterack;
1272	}
1273	acked = ti->ti_ack - tp->snd_una;
1274	tcpstat.tcps_rcvackpack++;
1275	tcpstat.tcps_rcvackbyte += acked;
1276
1277	/*
1278	* If we have a timestamp reply, update smoothed
1279	* round trip time. If no timestamp is present but
1280	* transmit timer is running and timed sequence
1281	* number was acked, update smoothed round trip time.
1282	* Since we now have an rtt measurement, cancel the
1283	* timer backoff (cf., Phil Karn's retransmit alg.).
1284	* Recompute the initial retransmit timer.
1285	*/
1286	#if 0
1287	if (ts_present)
1288	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1289	else
1290	#endif
1291	if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1292	tcp_xmit_timer(pData, tp, tp->t_rtt);
1293
1294	/*
1295	* If all outstanding data is acked, stop retransmit
1296	* timer and remember to restart (more output or persist).
1297	* If there is more data to be acked, restart retransmit
1298	* timer, using current (possibly backed-off) value.
1299	*/
1300	if (ti->ti_ack == tp->snd_max)
1301	{
1302	tp->t_timer[TCPT_REXMT] = 0;
1303	needoutput = 1;
1304	}
1305	else if (tp->t_timer[TCPT_PERSIST] == 0)
1306	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1307	/*
1308	* When new data is acked, open the congestion window.
1309	* If the window gives us less than ssthresh packets
1310	* in flight, open exponentially (maxseg per packet).
1311	* Otherwise open linearly: maxseg per window
1312	* (maxseg^2 / cwnd per packet).
1313	*/
1314	{
1315	register u_int cw = tp->snd_cwnd;
1316	register u_int incr = tp->t_maxseg;
1317
1318	if (cw > tp->snd_ssthresh)
1319	incr = incr * incr / cw;
1320	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1321	}
1322	if (acked > so->so_snd.sb_cc)
1323	{
1324	tp->snd_wnd -= so->so_snd.sb_cc;
1325	sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1326	ourfinisacked = 1;
1327	}
1328	else
1329	{
1330	sbdrop(&so->so_snd, acked);
1331	tp->snd_wnd -= acked;
1332	ourfinisacked = 0;
1333	}
1334	/*
1335	* XXX sowwakup is called when data is acked and there's room for
1336	* for more data... it should read() the socket
1337	*/
1338	#if 0
1339	if (so->so_snd.sb_flags & SB_NOTIFY)
1340	sowwakeup(so);
1341	#endif
1342	tp->snd_una = ti->ti_ack;
1343	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1344	tp->snd_nxt = tp->snd_una;
1345
1346	switch (tp->t_state)
1347	{
1348	/*
1349	* In FIN_WAIT_1 STATE in addition to the processing
1350	* for the ESTABLISHED state if our FIN is now acknowledged
1351	* then enter FIN_WAIT_2.
1352	*/
1353	case TCPS_FIN_WAIT_1:
1354	if (ourfinisacked)
1355	{
1356	/*
1357	* If we can't receive any more
1358	* data, then closing user can proceed.
1359	* Starting the timer is contrary to the
1360	* specification, but if we don't get a FIN
1361	* we'll hang forever.
1362	*/
1363	if (so->so_state & SS_FCANTRCVMORE)
1364	{
1365	soisfdisconnected(so);
1366	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1367	}
1368	tp->t_state = TCPS_FIN_WAIT_2;
1369	}
1370	break;
1371
1372	/*
1373	* In CLOSING STATE in addition to the processing for
1374	* the ESTABLISHED state if the ACK acknowledges our FIN
1375	* then enter the TIME-WAIT state, otherwise ignore
1376	* the segment.
1377	*/
1378	case TCPS_CLOSING:
1379	if (ourfinisacked)
1380	{
1381	tp->t_state = TCPS_TIME_WAIT;
1382	tcp_canceltimers(tp);
1383	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1384	soisfdisconnected(so);
1385	}
1386	break;
1387
1388	/*
1389	* In LAST_ACK, we may still be waiting for data to drain
1390	* and/or to be acked, as well as for the ack of our FIN.
1391	* If our FIN is now acknowledged, delete the TCB,
1392	* enter the closed state and return.
1393	*/
1394	case TCPS_LAST_ACK:
1395	if (ourfinisacked)
1396	{
1397	tp = tcp_close(pData, tp);
1398	goto drop;
1399	}
1400	break;
1401
1402	/*
1403	* In TIME_WAIT state the only thing that should arrive
1404	* is a retransmission of the remote FIN. Acknowledge
1405	* it and restart the finack timer.
1406	*/
1407	case TCPS_TIME_WAIT:
1408	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1409	goto dropafterack;
1410	}
1411	} /* switch(tp->t_state) */
1412
1413	step6:
1414	/*
1415	* Update window information.
1416	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1417	*/
1418	if ( (tiflags & TH_ACK)
1419	&& ( SEQ_LT(tp->snd_wl1, ti->ti_seq)
1420	\|\| ( tp->snd_wl1 == ti->ti_seq
1421	&& ( SEQ_LT(tp->snd_wl2, ti->ti_ack)
1422	\|\| ( tp->snd_wl2 == ti->ti_ack
1423	&& tiwin > tp->snd_wnd)))))
1424	{
1425	/* keep track of pure window updates */
1426	if ( ti->ti_len == 0
1427	&& tp->snd_wl2 == ti->ti_ack
1428	&& tiwin > tp->snd_wnd)
1429	tcpstat.tcps_rcvwinupd++;
1430	tp->snd_wnd = tiwin;
1431	tp->snd_wl1 = ti->ti_seq;
1432	tp->snd_wl2 = ti->ti_ack;
1433	if (tp->snd_wnd > tp->max_sndwnd)
1434	tp->max_sndwnd = tp->snd_wnd;
1435	needoutput = 1;
1436	}
1437
1438	/*
1439	* Process segments with URG.
1440	*/
1441	if ((tiflags & TH_URG) && ti->ti_urp &&
1442	TCPS_HAVERCVDFIN(tp->t_state) == 0)
1443	{
1444	/*
1445	* This is a kludge, but if we receive and accept
1446	* random urgent pointers, we'll crash in
1447	* soreceive. It's hard to imagine someone
1448	* actually wanting to send this much urgent data.
1449	*/
1450	if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen)
1451	{
1452	ti->ti_urp = 0;
1453	tiflags &= ~TH_URG;
1454	goto dodata;
1455	}
1456	/*
1457	* If this segment advances the known urgent pointer,
1458	* then mark the data stream. This should not happen
1459	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1460	* a FIN has been received from the remote side.
1461	* In these states we ignore the URG.
1462	*
1463	* According to RFC961 (Assigned Protocols),
1464	* the urgent pointer points to the last octet
1465	* of urgent data. We continue, however,
1466	* to consider it to indicate the first octet
1467	* of data past the urgent section as the original
1468	* spec states (in one of two places).
1469	*/
1470	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up))
1471	{
1472	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1473	so->so_urgc = so->so_rcv.sb_cc +
1474	(tp->rcv_up - tp->rcv_nxt); /* -1; */
1475	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1476	}
1477	}
1478	else
1479	/*
1480	* If no out of band data is expected,
1481	* pull receive urgent pointer along
1482	* with the receive window.
1483	*/
1484	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1485	tp->rcv_up = tp->rcv_nxt;
1486	dodata:
1487
1488	/*
1489	* If this is a small packet, then ACK now - with Nagel
1490	* congestion avoidance sender won't send more until
1491	* he gets an ACK.
1492	*
1493	* See above.
1494	*/
1495	if ( ti->ti_len
1496	&& (unsigned)ti->ti_len <= 5
1497	&& ((struct tcpiphdr_2 *)ti)->first_char == (char)27)
1498	{
1499	tp->t_flags \|= TF_ACKNOW;
1500	}
1501
1502	/*
1503	* Process the segment text, merging it into the TCP sequencing queue,
1504	* and arranging for acknowledgment of receipt if necessary.
1505	* This process logically involves adjusting tp->rcv_wnd as data
1506	* is presented to the user (this happens in tcp_usrreq.c,
1507	* case PRU_RCVD). If a FIN has already been received on this
1508	* connection then we just ignore the text.
1509	*/
1510	if ( (ti->ti_len \|\| (tiflags&TH_FIN))
1511	&& TCPS_HAVERCVDFIN(tp->t_state) == 0)
1512	{
1513	if ( ti->ti_seq == tp->rcv_nxt
1514	&& LIST_EMPTY(&tp->t_segq)
1515	&& tp->t_state == TCPS_ESTABLISHED)
1516	{
1517	DELAY_ACK(tp, ti); /* little bit different from BSD declaration see netinet/tcp_input.c */
1518	tp->rcv_nxt += tlen;
1519	tiflags = ti->ti_t.th_flags & TH_FIN;
1520	tcpstat.tcps_rcvpack++;
1521	tcpstat.tcps_rcvbyte += tlen;
1522	if (so->so_state & SS_FCANTRCVMORE)
1523	m_freem(pData, m);
1524	else
1525	{
1526	if (so->so_emu)
1527	{
1528	if (tcp_emu(pData, so, m))
1529	sbappend(pData, so, m);
1530	}
1531	else
1532	sbappend(pData, so, m);
1533	}
1534	}
1535	else
1536	{
1537	tiflags = tcp_reass(pData, tp, &ti->ti_t, &tlen, m);
1538	tiflags \|= TF_ACKNOW;
1539	}
1540	/*
1541	* Note the amount of data that peer has sent into
1542	* our window, in order to estimate the sender's
1543	* buffer size.
1544	*/
1545	len = so->so_rcv.sb_datalen - (tp->rcv_adv - tp->rcv_nxt);
1546	}
1547	else
1548	{
1549	m_free(pData, m);
1550	tiflags &= ~TH_FIN;
1551	}
1552
1553	/*
1554	* If FIN is received ACK the FIN and let the user know
1555	* that the connection is closing.
1556	*/
1557	if (tiflags & TH_FIN)
1558	{
1559	if (TCPS_HAVERCVDFIN(tp->t_state) == 0)
1560	{
1561	/*
1562	* If we receive a FIN we can't send more data,
1563	* set it SS_FDRAIN
1564	* Shutdown the socket if there is no rx data in the
1565	* buffer.
1566	* soread() is called on completion of shutdown() and
1567	* will got to TCPS_LAST_ACK, and use tcp_output()
1568	* to send the FIN.
1569	*/
1570	/* sofcantrcvmore(so); */
1571	sofwdrain(so);
1572
1573	tp->t_flags \|= TF_ACKNOW;
1574	tp->rcv_nxt++;
1575	}
1576	switch (tp->t_state)
1577	{
1578	/*
1579	* In SYN_RECEIVED and ESTABLISHED STATES
1580	* enter the CLOSE_WAIT state.
1581	*/
1582	case TCPS_SYN_RECEIVED:
1583	case TCPS_ESTABLISHED:
1584	if(so->so_emu == EMU_CTL) /* no shutdown on socket */
1585	tp->t_state = TCPS_LAST_ACK;
1586	else
1587	tp->t_state = TCPS_CLOSE_WAIT;
1588	break;
1589
1590	/*
1591	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1592	* enter the CLOSING state.
1593	*/
1594	case TCPS_FIN_WAIT_1:
1595	tp->t_state = TCPS_CLOSING;
1596	break;
1597
1598	/*
1599	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1600	* starting the time-wait timer, turning off the other
1601	* standard timers.
1602	*/
1603	case TCPS_FIN_WAIT_2:
1604	tp->t_state = TCPS_TIME_WAIT;
1605	tcp_canceltimers(tp);
1606	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1607	soisfdisconnected(so);
1608	break;
1609
1610	/*
1611	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1612	*/
1613	case TCPS_TIME_WAIT:
1614	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1615	break;
1616	}
1617	}
1618
1619	/*
1620	* Return any desired output.
1621	*/
1622	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1623	tcp_output(pData, tp);
1624
1625	SOCKET_UNLOCK(so);
1626	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1627	return;
1628
1629	dropafterack:
1630	Log2(("drop after ack\n"));
1631	/*
1632	* Generate an ACK dropping incoming segment if it occupies
1633	* sequence space, where the ACK reflects our state.
1634	*/
1635	if (tiflags & TH_RST)
1636	goto drop;
1637	m_freem(pData, m);
1638	tp->t_flags \|= TF_ACKNOW;
1639	(void) tcp_output(pData, tp);
1640	SOCKET_UNLOCK(so);
1641	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1642	return;
1643
1644	dropwithreset:
1645	/* reuses m if m!=NULL, m_free() unnecessary */
1646	if (tiflags & TH_ACK)
1647	tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1648	else
1649	{
1650	if (tiflags & TH_SYN) ti->ti_len++;
1651	tcp_respond(pData, tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1652	TH_RST\|TH_ACK);
1653	}
1654
1655	if (so != &tcb)
1656	SOCKET_UNLOCK(so);
1657	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1658	return;
1659
1660	drop:
1661	/*
1662	* Drop space held by incoming segment and return.
1663	*/
1664	m_free(pData, m);
1665
1666	#ifdef VBOX_WITH_SLIRP_MT
1667	if (RTCritSectIsOwned(&so->so_mutex))
1668	{
1669	SOCKET_UNLOCK(so);
1670	}
1671	#endif
1672
1673	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1674	return;
1675	}
1676
1677	void
1678	tcp_dooptions(PNATState pData, struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti)
1679	{
1680	u_int16_t mss;
1681	int opt, optlen;
1682
1683	DEBUG_CALL("tcp_dooptions");
1684	DEBUG_ARGS((dfd," tp = %lx cnt=%i \n", (long )tp, cnt));
1685
1686	for (; cnt > 0; cnt -= optlen, cp += optlen)
1687	{
1688	opt = cp[0];
1689	if (opt == TCPOPT_EOL)
1690	break;
1691	if (opt == TCPOPT_NOP)
1692	optlen = 1;
1693	else
1694	{
1695	optlen = cp[1];
1696	if (optlen <= 0)
1697	break;
1698	}
1699	switch (opt)
1700	{
1701	default:
1702	continue;
1703
1704	case TCPOPT_MAXSEG:
1705	if (optlen != TCPOLEN_MAXSEG)
1706	continue;
1707	if (!(ti->ti_flags & TH_SYN))
1708	continue;
1709	memcpy((char ) &mss, (char ) cp + 2, sizeof(mss));
1710	NTOHS(mss);
1711	(void) tcp_mss(pData, tp, mss); /* sets t_maxseg */
1712	break;
1713
1714	#if 0
1715	case TCPOPT_WINDOW:
1716	if (optlen != TCPOLEN_WINDOW)
1717	continue;
1718	if (!(ti->ti_flags & TH_SYN))
1719	continue;
1720	tp->t_flags \|= TF_RCVD_SCALE;
1721	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1722	break;
1723
1724	case TCPOPT_TIMESTAMP:
1725	if (optlen != TCPOLEN_TIMESTAMP)
1726	continue;
1727	*ts_present = 1;
1728	memcpy((char ) ts_val, (char )cp + 2, sizeof(*ts_val));
1729	NTOHL(*ts_val);
1730	memcpy((char ) ts_ecr, (char )cp + 6, sizeof(*ts_ecr));
1731	NTOHL(*ts_ecr);
1732
1733	/*
1734	* A timestamp received in a SYN makes
1735	* it ok to send timestamp requests and replies.
1736	*/
1737	if (ti->ti_flags & TH_SYN)
1738	{
1739	tp->t_flags \|= TF_RCVD_TSTMP;
1740	tp->ts_recent = *ts_val;
1741	tp->ts_recent_age = tcp_now;
1742	}
1743	break;
1744	#endif
1745	}
1746	}
1747	}
1748
1749
1750	/*
1751	* Pull out of band byte out of a segment so
1752	* it doesn't appear in the user's data queue.
1753	* It is still reflected in the segment length for
1754	* sequencing purposes.
1755	*/
1756
1757	#if 0
1758	void
1759	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1760	{
1761	int cnt = ti->ti_urp - 1;
1762
1763	while (cnt >= 0)
1764	{
1765	if (m->m_len > cnt)
1766	{
1767	char *cp = mtod(m, caddr_t) + cnt;
1768	struct tcpcb *tp = sototcpcb(so);
1769
1770	tp->t_iobc = *cp;
1771	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1772	memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1773	m->m_len--;
1774	return;
1775	}
1776	cnt -= m->m_len;
1777	m = m->m_next; /* XXX WRONG! Fix it! */
1778	if (m == 0)
1779	break;
1780	}
1781	panic("tcp_pulloutofband");
1782	}
1783	#endif
1784
1785	/*
1786	* Collect new round-trip time estimate
1787	* and update averages and current timeout.
1788	*/
1789
1790	void
1791	tcp_xmit_timer(PNATState pData, register struct tcpcb *tp, int rtt)
1792	{
1793	register short delta;
1794
1795	DEBUG_CALL("tcp_xmit_timer");
1796	DEBUG_ARG("tp = %lx", (long)tp);
1797	DEBUG_ARG("rtt = %d", rtt);
1798
1799	tcpstat.tcps_rttupdated++;
1800	if (tp->t_srtt != 0)
1801	{
1802	/*
1803	* srtt is stored as fixed point with 3 bits after the
1804	* binary point (i.e., scaled by 8). The following magic
1805	* is equivalent to the smoothing algorithm in rfc793 with
1806	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1807	* point). Adjust rtt to origin 0.
1808	*/
1809	delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1810	if ((tp->t_srtt += delta) <= 0)
1811	tp->t_srtt = 1;
1812	/*
1813	* We accumulate a smoothed rtt variance (actually, a
1814	* smoothed mean difference), then set the retransmit
1815	* timer to smoothed rtt + 4 times the smoothed variance.
1816	* rttvar is stored as fixed point with 2 bits after the
1817	* binary point (scaled by 4). The following is
1818	* equivalent to rfc793 smoothing with an alpha of .75
1819	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1820	* rfc793's wired-in beta.
1821	*/
1822	if (delta < 0)
1823	delta = -delta;
1824	delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1825	if ((tp->t_rttvar += delta) <= 0)
1826	tp->t_rttvar = 1;
1827	}
1828	else
1829	{
1830	/*
1831	* No rtt measurement yet - use the unsmoothed rtt.
1832	* Set the variance to half the rtt (so our first
1833	* retransmit happens at 3*rtt).
1834	*/
1835	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1836	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1837	}
1838	tp->t_rtt = 0;
1839	tp->t_rxtshift = 0;
1840
1841	/*
1842	* the retransmit should happen at rtt + 4 * rttvar.
1843	* Because of the way we do the smoothing, srtt and rttvar
1844	* will each average +1/2 tick of bias. When we compute
1845	* the retransmit timer, we want 1/2 tick of rounding and
1846	* 1 extra tick because of +-1/2 tick uncertainty in the
1847	* firing of the timer. The bias will give us exactly the
1848	* 1.5 tick we need. But, because the bias is
1849	* statistical, we have to test that we don't drop below
1850	* the minimum feasible timer (which is 2 ticks).
1851	*/
1852	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1853	(short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1854
1855	/*
1856	* We received an ack for a packet that wasn't retransmitted;
1857	* it is probably safe to discard any error indications we've
1858	* received recently. This isn't quite right, but close enough
1859	* for now (a route might have failed after we sent a segment,
1860	* and the return path might not be symmetrical).
1861	*/
1862	tp->t_softerror = 0;
1863	}
1864
1865	/*
1866	* Determine a reasonable value for maxseg size.
1867	* If the route is known, check route for mtu.
1868	* If none, use an mss that can be handled on the outgoing
1869	* interface without forcing IP to fragment; if bigger than
1870	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1871	* to utilize large mbufs. If no route is found, route has no mtu,
1872	* or the destination isn't local, use a default, hopefully conservative
1873	* size (usually 512 or the default IP max size, but no more than the mtu
1874	* of the interface), as we can't discover anything about intervening
1875	* gateways or networks. We also initialize the congestion/slow start
1876	* window to be a single segment if the destination isn't local.
1877	* While looking at the routing entry, we also initialize other path-dependent
1878	* parameters from pre-set or cached values in the routing entry.
1879	*/
1880
1881	int
1882	tcp_mss(PNATState pData, register struct tcpcb *tp, u_int offer)
1883	{
1884	struct socket *so = tp->t_socket;
1885	int mss;
1886
1887	DEBUG_CALL("tcp_mss");
1888	DEBUG_ARG("tp = %lx", (long)tp);
1889	DEBUG_ARG("offer = %d", offer);
1890
1891	mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
1892	if (offer)
1893	mss = min(mss, offer);
1894	mss = max(mss, 32);
1895	if (mss < tp->t_maxseg \|\| offer != 0)
1896	tp->t_maxseg = mss;
1897
1898	tp->snd_cwnd = mss;
1899
1900	sbreserve(pData, &so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
1901	sbreserve(pData, &so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
1902
1903	DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1904
1905	return mss;
1906	}

注意: 瀏覽 TracBrowser 來幫助您使用儲存庫瀏覽器

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 22584

以其他格式下載: