tcp_input.c@ 16445

Last change on this file since 16445 was 16445, checked in by vboxsync, 16 years ago

NAT:MT

foreach/label sequences enhanced
debug printing for sockets corretctly resolves NULL and SS_NOFDREF sockets.

Property svn:eol-style set to native

File size: 61.3 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	{
427	Log2(("lock: %s:%d We found socket %R[natsock]\n", __FUNCTION__, __LINE__, so));
428	break; /* so is locked here */
429	}
430	LOOP_LABEL(tcp, so, sonxt);
431	}
432	if (so == &tcb) {
433	Log2(("lock: %s:%d Haven't find anything \n", __FUNCTION__, __LINE__));
434	so = NULL;
435	}
436	#endif
437	if (so)
438	{
439	tcp_last_so = so;
440	}
441	++tcpstat.tcps_socachemiss;
442	}
443	else
444	{
445	SOCKET_LOCK(so);
446	QSOCKET_UNLOCK(tcb);
447	}
448
449	/*
450	* If the state is CLOSED (i.e., TCB does not exist) then
451	* all data in the incoming segment is discarded.
452	* If the TCB exists but is in CLOSED state, it is embryonic,
453	* but should either do a listen or a connect soon.
454	*
455	* state == CLOSED means we've done socreate() but haven't
456	* attached it to a protocol yet...
457	*
458	* XXX If a TCB does not exist, and the TH_SYN flag is
459	* the only flag set, then create a session, mark it
460	* as if it was LISTENING, and continue...
461	*/
462	Log2(("so = %R[natsock]\n", so));
463	if (so == 0)
464	{
465	if ((tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) != TH_SYN)
466	goto dropwithreset;
467
468	if ((so = socreate()) == NULL)
469	goto dropwithreset;
470	if (tcp_attach(pData, so) < 0)
471	{
472	RTMemFree(so); /* Not sofree (if it failed, it's not insqued) */
473	goto dropwithreset;
474	}
475
476	SOCKET_LOCK(so);
477	sbreserve(&so->so_snd, tcp_sndspace);
478	sbreserve(&so->so_rcv, tcp_rcvspace);
479
480	/* tcp_last_so = so; / / XXX ? */
481	/* tp = sototcpcb(so); */
482
483	so->so_laddr = ti->ti_src;
484	so->so_lport = ti->ti_sport;
485	so->so_faddr = ti->ti_dst;
486	so->so_fport = ti->ti_dport;
487
488	if ((so->so_iptos = tcp_tos(so)) == 0)
489	so->so_iptos = ((struct ip *)ti)->ip_tos;
490
491	tp = sototcpcb(so);
492	tp->t_state = TCPS_LISTEN;
493	}
494
495	/*
496	* If this is a still-connecting socket, this probably
497	* a retransmit of the SYN. Whether it's a retransmit SYN
498	* or something else, we nuke it.
499	*/
500	if (so->so_state & SS_ISFCONNECTING)
501	{
502	Log2(("so_state(%x) of %R[natsock] is still connecting =>drop\n", so->so_state, so));
503	goto drop;
504	}
505
506	tp = sototcpcb(so);
507
508	/* XXX Should never fail */
509	if (tp == 0)
510	goto dropwithreset;
511	if (tp->t_state == TCPS_CLOSED)
512	{
513	Log2(("t_state(%x) is closed =>drop\n", tp->t_state));
514	goto drop;
515	}
516
517	/* Unscale the window into a 32-bit value. */
518	/* if ((tiflags & TH_SYN) == 0)
519	* tiwin = ti->ti_win << tp->snd_scale;
520	* else
521	*/
522	tiwin = ti->ti_win;
523
524	/*
525	* Segment received on connection.
526	* Reset idle time and keep-alive timer.
527	*/
528	tp->t_idle = 0;
529	if (so_options)
530	tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
531	else
532	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
533
534	/*
535	* Process options if not in LISTEN state,
536	* else do it below (after getting remote address).
537	*/
538	if (optp && tp->t_state != TCPS_LISTEN)
539	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
540	/* , */
541	/* &ts_present, &ts_val, &ts_ecr); */
542
543	/*
544	* Header prediction: check for the two common cases
545	* of a uni-directional data xfer. If the packet has
546	* no control flags, is in-sequence, the window didn't
547	* change and we're not retransmitting, it's a
548	* candidate. If the length is zero and the ack moved
549	* forward, we're the sender side of the xfer. Just
550	* free the data acked & wake any higher level process
551	* that was blocked waiting for space. If the length
552	* is non-zero and the ack didn't move, we're the
553	* receiver side. If we're getting packets in-order
554	* (the reassembly queue is empty), add the data to
555	* the socket buffer and note that we need a delayed ack.
556	*
557	* XXX Some of these tests are not needed
558	* eg: the tiwin == tp->snd_wnd prevents many more
559	* predictions.. with no real advantage..
560	*/
561	if ( tp->t_state == TCPS_ESTABLISHED
562	&& (tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK
563	/* && (!ts_present \|\| TSTMP_GEQ(ts_val, tp->ts_recent)) */
564	&& ti->ti_seq == tp->rcv_nxt
565	&& tiwin && tiwin == tp->snd_wnd
566	&& tp->snd_nxt == tp->snd_max)
567	{
568	/*
569	* If last ACK falls within this segment's sequence numbers,
570	* record the timestamp.
571	*/
572	#if 0
573	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
574	SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len))
575	{
576	tp->ts_recent_age = tcp_now;
577	tp->ts_recent = ts_val;
578	}
579	#endif
580
581	if (ti->ti_len == 0)
582	{
583	if ( SEQ_GT(ti->ti_ack, tp->snd_una)
584	&& SEQ_LEQ(ti->ti_ack, tp->snd_max)
585	&& tp->snd_cwnd >= tp->snd_wnd)
586	{
587	/*
588	* this is a pure ack for outstanding data.
589	*/
590	++tcpstat.tcps_predack;
591	#if 0
592	if (ts_present)
593	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
594	else
595	#endif
596	if ( tp->t_rtt
597	&& SEQ_GT(ti->ti_ack, tp->t_rtseq))
598	tcp_xmit_timer(pData, tp, tp->t_rtt);
599	acked = ti->ti_ack - tp->snd_una;
600	tcpstat.tcps_rcvackpack++;
601	tcpstat.tcps_rcvackbyte += acked;
602	sbdrop(&so->so_snd, acked);
603	tp->snd_una = ti->ti_ack;
604	m_freem(pData, m);
605
606	/*
607	* If all outstanding data are acked, stop
608	* retransmit timer, otherwise restart timer
609	* using current (possibly backed-off) value.
610	* If process is waiting for space,
611	* wakeup/selwakeup/signal. If data
612	* are ready to send, let tcp_output
613	* decide between more output or persist.
614	*/
615	if (tp->snd_una == tp->snd_max)
616	tp->t_timer[TCPT_REXMT] = 0;
617	else if (tp->t_timer[TCPT_PERSIST] == 0)
618	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
619
620	/*
621	* There's room in so_snd, sowwakup will read()
622	* from the socket if we can
623	*/
624	#if 0
625	if (so->so_snd.sb_flags & SB_NOTIFY)
626	sowwakeup(so);
627	#endif
628	/*
629	* This is called because sowwakeup might have
630	* put data into so_snd. Since we don't so sowwakeup,
631	* we don't need this.. XXX???
632	*/
633	if (so->so_snd.sb_cc)
634	(void) tcp_output(pData, tp);
635
636	SOCKET_UNLOCK(so);
637	return;
638	}
639	}
640	else if ( ti->ti_ack == tp->snd_una
641	&& LIST_FIRST(&tp->t_segq)
642	&& ti->ti_len <= sbspace(&so->so_rcv))
643	{
644	/*
645	* this is a pure, in-sequence data packet
646	* with nothing on the reassembly queue and
647	* we have enough buffer space to take it.
648	*/
649	++tcpstat.tcps_preddat;
650	tp->rcv_nxt += ti->ti_len;
651	tcpstat.tcps_rcvpack++;
652	tcpstat.tcps_rcvbyte += ti->ti_len;
653	/*
654	* Add data to socket buffer.
655	*/
656	if (so->so_emu)
657	{
658	if (tcp_emu(pData, so,m)) sbappend(pData, so, m);
659	}
660	else
661	sbappend(pData, so, m);
662
663	/*
664	* XXX This is called when data arrives. Later, check
665	* if we can actually write() to the socket
666	* XXX Need to check? It's be NON_BLOCKING
667	*/
668	/* sorwakeup(so); */
669
670	/*
671	* If this is a short packet, then ACK now - with Nagel
672	* congestion avoidance sender won't send more until
673	* he gets an ACK.
674	*
675	* It is better to not delay acks at all to maximize
676	* TCP throughput. See RFC 2581.
677	*/
678	tp->t_flags \|= TF_ACKNOW;
679	tcp_output(pData, tp);
680	SOCKET_UNLOCK(so);
681	return;
682	}
683	} /* header prediction */
684	/*
685	* Calculate amount of space in receive window,
686	* and then do TCP input processing.
687	* Receive window is amount of space in rcv queue,
688	* but not less than advertised window.
689	*/
690	{
691	int win;
692	win = sbspace(&so->so_rcv);
693	if (win < 0)
694	win = 0;
695	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
696	}
697
698	switch (tp->t_state)
699	{
700	/*
701	* If the state is LISTEN then ignore segment if it contains an RST.
702	* If the segment contains an ACK then it is bad and send a RST.
703	* If it does not contain a SYN then it is not interesting; drop it.
704	* Don't bother responding if the destination was a broadcast.
705	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
706	* tp->iss, and send a segment:
707	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
708	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
709	* Fill in remote peer address fields if not previously specified.
710	* Enter SYN_RECEIVED state, and process any other fields of this
711	* segment in this state.
712	*/
713	case TCPS_LISTEN:
714	{
715	if (tiflags & TH_RST) {
716	Log2(("RST(%x) is on listen =>drop\n", tiflags));
717	goto drop;
718	}
719	if (tiflags & TH_ACK)
720	goto dropwithreset;
721	if ((tiflags & TH_SYN) == 0)
722	{
723	Log2(("SYN(%x) is off on listen =>drop\n", tiflags));
724	goto drop;
725	}
726
727	/*
728	* This has way too many gotos...
729	* But a bit of spaghetti code never hurt anybody :)
730	*/
731
732	if (so->so_emu & EMU_NOCONNECT)
733	{
734	so->so_emu &= ~EMU_NOCONNECT;
735	goto cont_input;
736	}
737
738	if ( (tcp_fconnect(pData, so) == -1)
739	&& errno != EINPROGRESS
740	&& errno != EWOULDBLOCK)
741	{
742	u_char code = ICMP_UNREACH_NET;
743	DEBUG_MISC((dfd," tcp fconnect errno = %d-%s\n",
744	errno,strerror(errno)));
745	if (errno == ECONNREFUSED)
746	{
747	/* ACK the SYN, send RST to refuse the connection */
748	tcp_respond(pData, tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
749	TH_RST\|TH_ACK);
750	}
751	else
752	{
753	if (errno == EHOSTUNREACH)
754	code = ICMP_UNREACH_HOST;
755	HTONL(ti->ti_seq); /* restore tcp header */
756	HTONL(ti->ti_ack);
757	HTONS(ti->ti_win);
758	HTONS(ti->ti_urp);
759	m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
760	m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
761	*ip = save_ip;
762	icmp_error(pData, m, ICMP_UNREACH,code, 0,strerror(errno));
763	}
764	tp = tcp_close(pData, tp);
765	m_free(pData, m);
766	}
767	else
768	{
769	/*
770	* Haven't connected yet, save the current mbuf
771	* and ti, and return
772	* XXX Some OS's don't tell us whether the connect()
773	* succeeded or not. So we must time it out.
774	*/
775	so->so_m = m;
776	so->so_ti = ti;
777	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
778	tp->t_state = TCPS_SYN_RECEIVED;
779	}
780	SOCKET_UNLOCK(so);
781	return;
782
783	cont_conn:
784	/* m==NULL
785	* Check if the connect succeeded
786	*/
787	if (so->so_state & SS_NOFDREF)
788	{
789	tp = tcp_close(pData, tp);
790	goto dropwithreset;
791	}
792	cont_input:
793	tcp_template(tp);
794
795	if (optp)
796	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
797
798	if (iss)
799	tp->iss = iss;
800	else
801	tp->iss = tcp_iss;
802	tcp_iss += TCP_ISSINCR/2;
803	tp->irs = ti->ti_seq;
804	tcp_sendseqinit(tp);
805	tcp_rcvseqinit(tp);
806	tp->t_flags \|= TF_ACKNOW;
807	tp->t_state = TCPS_SYN_RECEIVED;
808	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
809	tcpstat.tcps_accepts++;
810	Log2(("hit trimthenstep6\n"));
811	goto trimthenstep6;
812	} /* case TCPS_LISTEN */
813
814	/*
815	* If the state is SYN_SENT:
816	* if seg contains an ACK, but not for our SYN, drop the input.
817	* if seg contains a RST, then drop the connection.
818	* if seg does not contain SYN, then drop it.
819	* Otherwise this is an acceptable SYN segment
820	* initialize tp->rcv_nxt and tp->irs
821	* if seg contains ack then advance tp->snd_una
822	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
823	* arrange for segment to be acked (eventually)
824	* continue processing rest of data/controls, beginning with URG
825	*/
826	case TCPS_SYN_SENT:
827	if ( (tiflags & TH_ACK)
828	&& ( SEQ_LEQ(ti->ti_ack, tp->iss)
829	\|\| SEQ_GT(ti->ti_ack, tp->snd_max)))
830	goto dropwithreset;
831
832	if (tiflags & TH_RST)
833	{
834	if (tiflags & TH_ACK)
835	tp = tcp_drop(pData, tp,0); /* XXX Check t_softerror! */
836	Log2(("RST(%x) is on SYN_SENT =>drop\n", tiflags));
837	goto drop;
838	}
839
840	if ((tiflags & TH_SYN) == 0)
841	{
842	Log2(("SYN(%x) bit is off on SYN_SENT =>drop\n", tiflags));
843	goto drop;
844	}
845	if (tiflags & TH_ACK)
846	{
847	tp->snd_una = ti->ti_ack;
848	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
849	tp->snd_nxt = tp->snd_una;
850	}
851
852	tp->t_timer[TCPT_REXMT] = 0;
853	tp->irs = ti->ti_seq;
854	tcp_rcvseqinit(tp);
855	tp->t_flags \|= TF_ACKNOW;
856	if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss))
857	{
858	tcpstat.tcps_connects++;
859	soisfconnected(so);
860	tp->t_state = TCPS_ESTABLISHED;
861
862	/* Do window scaling on this connection? */
863	#if 0
864	if (( tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
865	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
866	{
867	tp->snd_scale = tp->requested_s_scale;
868	tp->rcv_scale = tp->request_r_scale;
869	}
870	#endif
871	(void) tcp_reass(pData, tp, (struct tcphdr )0, NULL, (struct mbuf )0);
872	/*
873	* if we didn't have to retransmit the SYN,
874	* use its rtt as our initial srtt & rtt var.
875	*/
876	if (tp->t_rtt)
877	tcp_xmit_timer(pData, tp, tp->t_rtt);
878	}
879	else
880	tp->t_state = TCPS_SYN_RECEIVED;
881
882	trimthenstep6:
883	/*
884	* Advance ti->ti_seq to correspond to first data byte.
885	* If data, trim to stay within window,
886	* dropping FIN if necessary.
887	*/
888	ti->ti_seq++;
889	if (ti->ti_len > tp->rcv_wnd)
890	{
891	todrop = ti->ti_len - tp->rcv_wnd;
892	m_adj(m, -todrop);
893	ti->ti_len = tp->rcv_wnd;
894	tiflags &= ~TH_FIN;
895	tcpstat.tcps_rcvpackafterwin++;
896	tcpstat.tcps_rcvbyteafterwin += todrop;
897	}
898	tp->snd_wl1 = ti->ti_seq - 1;
899	tp->rcv_up = ti->ti_seq;
900	Log2(("hit6"));
901	goto step6;
902	} /* switch tp->t_state */
903	/*
904	* States other than LISTEN or SYN_SENT.
905	* First check timestamp, if present.
906	* Then check that at least some bytes of segment are within
907	* receive window. If segment begins before rcv_nxt,
908	* drop leading data (and SYN); if nothing left, just ack.
909	*
910	* RFC 1323 PAWS: If we have a timestamp reply on this segment
911	* and it's less than ts_recent, drop it.
912	*/
913	#if 0
914	if ( ts_present
915	&& (tiflags & TH_RST) == 0
916	&& tp->ts_recent
917	&& TSTMP_LT(ts_val, tp->ts_recent))
918	{
919	/* Check to see if ts_recent is over 24 days old. */
920	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE)
921	{
922	/*
923	* Invalidate ts_recent. If this segment updates
924	* ts_recent, the age will be reset later and ts_recent
925	* will get a valid value. If it does not, setting
926	* ts_recent to zero will at least satisfy the
927	* requirement that zero be placed in the timestamp
928	* echo reply when ts_recent isn't valid. The
929	* age isn't reset until we get a valid ts_recent
930	* because we don't want out-of-order segments to be
931	* dropped when ts_recent is old.
932	*/
933	tp->ts_recent = 0;
934	}
935	else
936	{
937	tcpstat.tcps_rcvduppack++;
938	tcpstat.tcps_rcvdupbyte += ti->ti_len;
939	tcpstat.tcps_pawsdrop++;
940	goto dropafterack;
941	}
942	}
943	#endif
944
945	todrop = tp->rcv_nxt - ti->ti_seq;
946	if (todrop > 0)
947	{
948	if (tiflags & TH_SYN)
949	{
950	tiflags &= ~TH_SYN;
951	ti->ti_seq++;
952	if (ti->ti_urp > 1)
953	ti->ti_urp--;
954	else
955	tiflags &= ~TH_URG;
956	todrop--;
957	}
958	/*
959	* Following if statement from Stevens, vol. 2, p. 960.
960	*/
961	if ( todrop > ti->ti_len
962	\|\| ( todrop == ti->ti_len
963	&& (tiflags & TH_FIN) == 0))
964	{
965	/*
966	* Any valid FIN must be to the left of the window.
967	* At this point the FIN must be a duplicate or out
968	* of sequence; drop it.
969	*/
970	tiflags &= ~TH_FIN;
971
972	/*
973	* Send an ACK to resynchronize and drop any data.
974	* But keep on processing for RST or ACK.
975	*/
976	tp->t_flags \|= TF_ACKNOW;
977	todrop = ti->ti_len;
978	tcpstat.tcps_rcvduppack++;
979	tcpstat.tcps_rcvdupbyte += todrop;
980	}
981	else
982	{
983	tcpstat.tcps_rcvpartduppack++;
984	tcpstat.tcps_rcvpartdupbyte += todrop;
985	}
986	m_adj(m, todrop);
987	ti->ti_seq += todrop;
988	ti->ti_len -= todrop;
989	if (ti->ti_urp > todrop)
990	ti->ti_urp -= todrop;
991	else
992	{
993	tiflags &= ~TH_URG;
994	ti->ti_urp = 0;
995	}
996	}
997	/*
998	* If new data are received on a connection after the
999	* user processes are gone, then RST the other end.
1000	*/
1001	if ( (so->so_state & SS_NOFDREF)
1002	&& tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len)
1003	{
1004	tp = tcp_close(pData, tp);
1005	tcpstat.tcps_rcvafterclose++;
1006	goto dropwithreset;
1007	}
1008
1009	/*
1010	* If segment ends after window, drop trailing data
1011	* (and PUSH and FIN); if nothing left, just ACK.
1012	*/
1013	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1014	if (todrop > 0)
1015	{
1016	tcpstat.tcps_rcvpackafterwin++;
1017	if (todrop >= ti->ti_len)
1018	{
1019	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1020	/*
1021	* If a new connection request is received
1022	* while in TIME_WAIT, drop the old connection
1023	* and start over if the sequence numbers
1024	* are above the previous ones.
1025	*/
1026	if ( tiflags & TH_SYN
1027	&& tp->t_state == TCPS_TIME_WAIT
1028	&& SEQ_GT(ti->ti_seq, tp->rcv_nxt))
1029	{
1030	iss = tp->rcv_nxt + TCP_ISSINCR;
1031	tp = tcp_close(pData, tp);
1032	SOCKET_UNLOCK(tp->t_socket);
1033	goto findso;
1034	}
1035	/*
1036	* If window is closed can only take segments at
1037	* window edge, and have to drop data and PUSH from
1038	* incoming segments. Continue processing, but
1039	* remember to ack. Otherwise, drop segment
1040	* and ack.
1041	*/
1042	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt)
1043	{
1044	tp->t_flags \|= TF_ACKNOW;
1045	tcpstat.tcps_rcvwinprobe++;
1046	}
1047	else
1048	goto dropafterack;
1049	}
1050	else
1051	tcpstat.tcps_rcvbyteafterwin += todrop;
1052	m_adj(m, -todrop);
1053	ti->ti_len -= todrop;
1054	tiflags &= ~(TH_PUSH\|TH_FIN);
1055	}
1056
1057	/*
1058	* If last ACK falls within this segment's sequence numbers,
1059	* record its timestamp.
1060	*/
1061	#if 0
1062	if ( ts_present
1063	&& SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)
1064	&& SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + ((tiflags & (TH_SYN\|TH_FIN)) != 0)))
1065	{
1066	tp->ts_recent_age = tcp_now;
1067	tp->ts_recent = ts_val;
1068	}
1069	#endif
1070
1071	/*
1072	* If the RST bit is set examine the state:
1073	* SYN_RECEIVED STATE:
1074	* If passive open, return to LISTEN state.
1075	* If active open, inform user that connection was refused.
1076	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1077	* Inform user that connection was reset, and close tcb.
1078	* CLOSING, LAST_ACK, TIME_WAIT STATES
1079	* Close the tcb.
1080	*/
1081	if (tiflags&TH_RST)
1082	switch (tp->t_state)
1083	{
1084	case TCPS_SYN_RECEIVED:
1085	/* so->so_error = ECONNREFUSED; */
1086	goto close;
1087
1088	case TCPS_ESTABLISHED:
1089	case TCPS_FIN_WAIT_1:
1090	case TCPS_FIN_WAIT_2:
1091	case TCPS_CLOSE_WAIT:
1092	/* so->so_error = ECONNRESET; */
1093	close:
1094	Log2(("closing...=>drop\n", tp->t_state));
1095	tp->t_state = TCPS_CLOSED;
1096	tcpstat.tcps_drops++;
1097	tp = tcp_close(pData, tp);
1098	goto drop;
1099
1100	case TCPS_CLOSING:
1101	case TCPS_LAST_ACK:
1102	case TCPS_TIME_WAIT:
1103	Log2(("t_state is (%x) sort of close =>drop\n", tp->t_state));
1104	tp = tcp_close(pData, tp);
1105	goto drop;
1106	}
1107
1108	/*
1109	* If a SYN is in the window, then this is an
1110	* error and we send an RST and drop the connection.
1111	*/
1112	if (tiflags & TH_SYN)
1113	{
1114	tp = tcp_drop(pData, tp,0);
1115	goto dropwithreset;
1116	}
1117
1118	/*
1119	* If the ACK bit is off we drop the segment and return.
1120	*/
1121	if ((tiflags & TH_ACK) == 0)
1122	{
1123	Log2(("ACK(%x) bit is off =>drop\n", tiflags));
1124	goto drop;
1125	}
1126
1127	/*
1128	* Ack processing.
1129	*/
1130	switch (tp->t_state)
1131	{
1132	/*
1133	* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1134	* ESTABLISHED state and continue processing, otherwise
1135	* send an RST. una<=ack<=max
1136	*/
1137	case TCPS_SYN_RECEIVED:
1138	if ( SEQ_GT(tp->snd_una, ti->ti_ack)
1139	\|\| SEQ_GT(ti->ti_ack, tp->snd_max))
1140	goto dropwithreset;
1141	tcpstat.tcps_connects++;
1142	tp->t_state = TCPS_ESTABLISHED;
1143	/*
1144	* The sent SYN is ack'ed with our sequence number +1
1145	* The first data byte already in the buffer will get
1146	* lost if no correction is made. This is only needed for
1147	* SS_CTL since the buffer is empty otherwise.
1148	* tp->snd_una++; or:
1149	*/
1150	tp->snd_una = ti->ti_ack;
1151	soisfconnected(so);
1152
1153	/* Do window scaling? */
1154	#if 0
1155	if ( (tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
1156	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
1157	{
1158	tp->snd_scale = tp->requested_s_scale;
1159	tp->rcv_scale = tp->request_r_scale;
1160	}
1161	#endif
1162	(void) tcp_reass(pData, tp, (struct tcphdr )0, (int )0, (struct mbuf *)0);
1163	tp->snd_wl1 = ti->ti_seq - 1;
1164	/* Avoid ack processing; snd_una==ti_ack => dup ack */
1165	Log2(("hit synrx_to_est\n"));
1166	goto synrx_to_est;
1167	/* fall into ... */
1168
1169	/*
1170	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1171	* ACKs. If the ack is in the range
1172	* tp->snd_una < ti->ti_ack <= tp->snd_max
1173	* then advance tp->snd_una to ti->ti_ack and drop
1174	* data from the retransmission queue. If this ACK reflects
1175	* more up to date window information we update our window information.
1176	*/
1177	case TCPS_ESTABLISHED:
1178	case TCPS_FIN_WAIT_1:
1179	case TCPS_FIN_WAIT_2:
1180	case TCPS_CLOSE_WAIT:
1181	case TCPS_CLOSING:
1182	case TCPS_LAST_ACK:
1183	case TCPS_TIME_WAIT:
1184	if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
1185	{
1186	if (ti->ti_len == 0 && tiwin == tp->snd_wnd)
1187	{
1188	tcpstat.tcps_rcvdupack++;
1189	DEBUG_MISC((dfd," dup ack m = %lx so = %lx \n",
1190	(long )m, (long )so));
1191	/*
1192	* If we have outstanding data (other than
1193	* a window probe), this is a completely
1194	* duplicate ack (ie, window info didn't
1195	* change), the ack is the biggest we've
1196	* seen and we've seen exactly our rexmt
1197	* threshold of them, assume a packet
1198	* has been dropped and retransmit it.
1199	* Kludge snd_nxt & the congestion
1200	* window so we send only this one
1201	* packet.
1202	*
1203	* We know we're losing at the current
1204	* window size so do congestion avoidance
1205	* (set ssthresh to half the current window
1206	* and pull our congestion window back to
1207	* the new ssthresh).
1208	*
1209	* Dup acks mean that packets have left the
1210	* network (they're now cached at the receiver)
1211	* so bump cwnd by the amount in the receiver
1212	* to keep a constant cwnd packets in the
1213	* network.
1214	*/
1215	if ( tp->t_timer[TCPT_REXMT] == 0
1216	\|\| ti->ti_ack != tp->snd_una)
1217	tp->t_dupacks = 0;
1218	else if (++tp->t_dupacks == tcprexmtthresh)
1219	{
1220	tcp_seq onxt = tp->snd_nxt;
1221	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
1222	if (win < 2)
1223	win = 2;
1224	tp->snd_ssthresh = win * tp->t_maxseg;
1225	tp->t_timer[TCPT_REXMT] = 0;
1226	tp->t_rtt = 0;
1227	tp->snd_nxt = ti->ti_ack;
1228	tp->snd_cwnd = tp->t_maxseg;
1229	(void) tcp_output(pData, tp);
1230	tp->snd_cwnd = tp->snd_ssthresh +
1231	tp->t_maxseg * tp->t_dupacks;
1232	if (SEQ_GT(onxt, tp->snd_nxt))
1233	tp->snd_nxt = onxt;
1234	Log2(("t_dupacks(%d) == tcprexmtthresh(%d)=>drop\n", tp->t_dupacks, tcprexmtthresh));
1235	goto drop;
1236	}
1237	else if (tp->t_dupacks > tcprexmtthresh)
1238	{
1239	tp->snd_cwnd += tp->t_maxseg;
1240	(void) tcp_output(pData, tp);
1241	Log2(("t_dupacks(%d) > tcprexmtthresh(%d)=>drop\n", tp->t_dupacks, tcprexmtthresh));
1242	goto drop;
1243	}
1244	}
1245	else
1246	tp->t_dupacks = 0;
1247	break;
1248	}
1249	synrx_to_est:
1250	Log2(("enter synrx_to_est\n"));
1251	/*
1252	* If the congestion window was inflated to account
1253	* for the other side's cached packets, retract it.
1254	*/
1255	if ( tp->t_dupacks > tcprexmtthresh
1256	&& tp->snd_cwnd > tp->snd_ssthresh)
1257	tp->snd_cwnd = tp->snd_ssthresh;
1258	tp->t_dupacks = 0;
1259	if (SEQ_GT(ti->ti_ack, tp->snd_max))
1260	{
1261	tcpstat.tcps_rcvacktoomuch++;
1262	goto dropafterack;
1263	}
1264	acked = ti->ti_ack - tp->snd_una;
1265	tcpstat.tcps_rcvackpack++;
1266	tcpstat.tcps_rcvackbyte += acked;
1267
1268	/*
1269	* If we have a timestamp reply, update smoothed
1270	* round trip time. If no timestamp is present but
1271	* transmit timer is running and timed sequence
1272	* number was acked, update smoothed round trip time.
1273	* Since we now have an rtt measurement, cancel the
1274	* timer backoff (cf., Phil Karn's retransmit alg.).
1275	* Recompute the initial retransmit timer.
1276	*/
1277	#if 0
1278	if (ts_present)
1279	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1280	else
1281	#endif
1282	if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1283	tcp_xmit_timer(pData, tp,tp->t_rtt);
1284
1285	/*
1286	* If all outstanding data is acked, stop retransmit
1287	* timer and remember to restart (more output or persist).
1288	* If there is more data to be acked, restart retransmit
1289	* timer, using current (possibly backed-off) value.
1290	*/
1291	if (ti->ti_ack == tp->snd_max)
1292	{
1293	tp->t_timer[TCPT_REXMT] = 0;
1294	needoutput = 1;
1295	}
1296	else if (tp->t_timer[TCPT_PERSIST] == 0)
1297	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1298	/*
1299	* When new data is acked, open the congestion window.
1300	* If the window gives us less than ssthresh packets
1301	* in flight, open exponentially (maxseg per packet).
1302	* Otherwise open linearly: maxseg per window
1303	* (maxseg^2 / cwnd per packet).
1304	*/
1305	{
1306	register u_int cw = tp->snd_cwnd;
1307	register u_int incr = tp->t_maxseg;
1308
1309	if (cw > tp->snd_ssthresh)
1310	incr = incr * incr / cw;
1311	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1312	}
1313	if (acked > so->so_snd.sb_cc)
1314	{
1315	tp->snd_wnd -= so->so_snd.sb_cc;
1316	sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1317	ourfinisacked = 1;
1318	}
1319	else
1320	{
1321	sbdrop(&so->so_snd, acked);
1322	tp->snd_wnd -= acked;
1323	ourfinisacked = 0;
1324	}
1325	/*
1326	* XXX sowwakup is called when data is acked and there's room for
1327	* for more data... it should read() the socket
1328	*/
1329	#if 0
1330	if (so->so_snd.sb_flags & SB_NOTIFY)
1331	sowwakeup(so);
1332	#endif
1333	tp->snd_una = ti->ti_ack;
1334	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1335	tp->snd_nxt = tp->snd_una;
1336
1337	switch (tp->t_state)
1338	{
1339	/*
1340	* In FIN_WAIT_1 STATE in addition to the processing
1341	* for the ESTABLISHED state if our FIN is now acknowledged
1342	* then enter FIN_WAIT_2.
1343	*/
1344	case TCPS_FIN_WAIT_1:
1345	if (ourfinisacked)
1346	{
1347	/*
1348	* If we can't receive any more
1349	* data, then closing user can proceed.
1350	* Starting the timer is contrary to the
1351	* specification, but if we don't get a FIN
1352	* we'll hang forever.
1353	*/
1354	if (so->so_state & SS_FCANTRCVMORE)
1355	{
1356	soisfdisconnected(so);
1357	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1358	}
1359	tp->t_state = TCPS_FIN_WAIT_2;
1360	}
1361	break;
1362
1363	/*
1364	* In CLOSING STATE in addition to the processing for
1365	* the ESTABLISHED state if the ACK acknowledges our FIN
1366	* then enter the TIME-WAIT state, otherwise ignore
1367	* the segment.
1368	*/
1369	case TCPS_CLOSING:
1370	if (ourfinisacked)
1371	{
1372	tp->t_state = TCPS_TIME_WAIT;
1373	tcp_canceltimers(tp);
1374	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1375	soisfdisconnected(so);
1376	}
1377	break;
1378
1379	/*
1380	* In LAST_ACK, we may still be waiting for data to drain
1381	* and/or to be acked, as well as for the ack of our FIN.
1382	* If our FIN is now acknowledged, delete the TCB,
1383	* enter the closed state and return.
1384	*/
1385	case TCPS_LAST_ACK:
1386	if (ourfinisacked)
1387	{
1388	Log2(("ourfinisacked=>drop\n"));
1389	tp = tcp_close(pData, tp);
1390	goto drop;
1391	}
1392	break;
1393
1394	/*
1395	* In TIME_WAIT state the only thing that should arrive
1396	* is a retransmission of the remote FIN. Acknowledge
1397	* it and restart the finack timer.
1398	*/
1399	case TCPS_TIME_WAIT:
1400	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1401	goto dropafterack;
1402	}
1403	} /* switch(tp->t_state) */
1404
1405	step6:
1406	/*
1407	* Update window information.
1408	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1409	*/
1410	if ( (tiflags & TH_ACK)
1411	&& ( SEQ_LT(tp->snd_wl1, ti->ti_seq)
1412	\|\| ( tp->snd_wl1 == ti->ti_seq
1413	&& ( SEQ_LT(tp->snd_wl2, ti->ti_ack)
1414	\|\| ( tp->snd_wl2 == ti->ti_ack
1415	&& tiwin > tp->snd_wnd)))))
1416	{
1417	/* keep track of pure window updates */
1418	if ( ti->ti_len == 0
1419	&& tp->snd_wl2 == ti->ti_ack
1420	&& tiwin > tp->snd_wnd)
1421	tcpstat.tcps_rcvwinupd++;
1422	tp->snd_wnd = tiwin;
1423	tp->snd_wl1 = ti->ti_seq;
1424	tp->snd_wl2 = ti->ti_ack;
1425	if (tp->snd_wnd > tp->max_sndwnd)
1426	tp->max_sndwnd = tp->snd_wnd;
1427	needoutput = 1;
1428	}
1429
1430	/*
1431	* Process segments with URG.
1432	*/
1433	if ((tiflags & TH_URG) && ti->ti_urp &&
1434	TCPS_HAVERCVDFIN(tp->t_state) == 0)
1435	{
1436	/*
1437	* This is a kludge, but if we receive and accept
1438	* random urgent pointers, we'll crash in
1439	* soreceive. It's hard to imagine someone
1440	* actually wanting to send this much urgent data.
1441	*/
1442	if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen)
1443	{
1444	ti->ti_urp = 0;
1445	tiflags &= ~TH_URG;
1446	goto dodata;
1447	}
1448	/*
1449	* If this segment advances the known urgent pointer,
1450	* then mark the data stream. This should not happen
1451	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1452	* a FIN has been received from the remote side.
1453	* In these states we ignore the URG.
1454	*
1455	* According to RFC961 (Assigned Protocols),
1456	* the urgent pointer points to the last octet
1457	* of urgent data. We continue, however,
1458	* to consider it to indicate the first octet
1459	* of data past the urgent section as the original
1460	* spec states (in one of two places).
1461	*/
1462	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up))
1463	{
1464	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1465	so->so_urgc = so->so_rcv.sb_cc +
1466	(tp->rcv_up - tp->rcv_nxt); /* -1; */
1467	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1468	}
1469	}
1470	else
1471	/*
1472	* If no out of band data is expected,
1473	* pull receive urgent pointer along
1474	* with the receive window.
1475	*/
1476	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1477	tp->rcv_up = tp->rcv_nxt;
1478	dodata:
1479	Log2(("do data hit!\n"));
1480
1481	/*
1482	* If this is a small packet, then ACK now - with Nagel
1483	* congestion avoidance sender won't send more until
1484	* he gets an ACK.
1485	*
1486	* See above.
1487	*/
1488	if ( ti->ti_len
1489	&& (unsigned)ti->ti_len <= 5
1490	&& ((struct tcpiphdr_2 *)ti)->first_char == (char)27)
1491	{
1492	tp->t_flags \|= TF_ACKNOW;
1493	}
1494
1495	/*
1496	* Process the segment text, merging it into the TCP sequencing queue,
1497	* and arranging for acknowledgment of receipt if necessary.
1498	* This process logically involves adjusting tp->rcv_wnd as data
1499	* is presented to the user (this happens in tcp_usrreq.c,
1500	* case PRU_RCVD). If a FIN has already been received on this
1501	* connection then we just ignore the text.
1502	*/
1503	if ( (ti->ti_len \|\| (tiflags&TH_FIN))
1504	&& TCPS_HAVERCVDFIN(tp->t_state) == 0)
1505	{
1506	if ( ti->ti_seq == tp->rcv_nxt
1507	&& LIST_EMPTY(&tp->t_segq)
1508	&& tp->t_state == TCPS_ESTABLISHED)
1509	{
1510	DELAY_ACK(tp, ti); /* little bit different from BSD declaration see netinet/tcp_input.c */
1511	tp->rcv_nxt += tlen;
1512	tiflags = ti->ti_t.th_flags & TH_FIN;
1513	tcpstat.tcps_rcvpack++;
1514	tcpstat.tcps_rcvbyte += tlen;
1515	if (so->so_state & SS_FCANTRCVMORE)
1516	m_freem(pData, m);
1517	else
1518	{
1519	if (so->so_emu)
1520	{
1521	if (tcp_emu(pData, so,m))
1522	sbappend(pData, so, m);
1523	}
1524	else
1525	sbappend(pData, so, m);
1526	}
1527	}
1528	else
1529	{
1530	tiflags = tcp_reass(pData, tp, &ti->ti_t, &tlen, m);
1531	tiflags \|= TF_ACKNOW;
1532	}
1533	/*
1534	* Note the amount of data that peer has sent into
1535	* our window, in order to estimate the sender's
1536	* buffer size.
1537	*/
1538	len = so->so_rcv.sb_datalen - (tp->rcv_adv - tp->rcv_nxt);
1539	}
1540	else
1541	{
1542	m_free(pData, m);
1543	tiflags &= ~TH_FIN;
1544	}
1545
1546	/*
1547	* If FIN is received ACK the FIN and let the user know
1548	* that the connection is closing.
1549	*/
1550	if (tiflags & TH_FIN)
1551	{
1552	if (TCPS_HAVERCVDFIN(tp->t_state) == 0)
1553	{
1554	/*
1555	* If we receive a FIN we can't send more data,
1556	* set it SS_FDRAIN
1557	* Shutdown the socket if there is no rx data in the
1558	* buffer.
1559	* soread() is called on completion of shutdown() and
1560	* will got to TCPS_LAST_ACK, and use tcp_output()
1561	* to send the FIN.
1562	*/
1563	/* sofcantrcvmore(so); */
1564	sofwdrain(so);
1565
1566	tp->t_flags \|= TF_ACKNOW;
1567	tp->rcv_nxt++;
1568	}
1569	switch (tp->t_state)
1570	{
1571	/*
1572	* In SYN_RECEIVED and ESTABLISHED STATES
1573	* enter the CLOSE_WAIT state.
1574	*/
1575	case TCPS_SYN_RECEIVED:
1576	case TCPS_ESTABLISHED:
1577	if(so->so_emu == EMU_CTL) /* no shutdown on socket */
1578	tp->t_state = TCPS_LAST_ACK;
1579	else
1580	tp->t_state = TCPS_CLOSE_WAIT;
1581	break;
1582
1583	/*
1584	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1585	* enter the CLOSING state.
1586	*/
1587	case TCPS_FIN_WAIT_1:
1588	tp->t_state = TCPS_CLOSING;
1589	break;
1590
1591	/*
1592	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1593	* starting the time-wait timer, turning off the other
1594	* standard timers.
1595	*/
1596	case TCPS_FIN_WAIT_2:
1597	tp->t_state = TCPS_TIME_WAIT;
1598	tcp_canceltimers(tp);
1599	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1600	soisfdisconnected(so);
1601	break;
1602
1603	/*
1604	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1605	*/
1606	case TCPS_TIME_WAIT:
1607	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1608	break;
1609	}
1610	}
1611
1612	/*
1613	* Return any desired output.
1614	*/
1615	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1616	tcp_output(pData, tp);
1617
1618	SOCKET_UNLOCK(so);
1619	return;
1620
1621	dropafterack:
1622	Log2(("drop after ack\n"));
1623	/*
1624	* Generate an ACK dropping incoming segment if it occupies
1625	* sequence space, where the ACK reflects our state.
1626	*/
1627	if (tiflags & TH_RST)
1628	goto drop;
1629	m_freem(pData, m);
1630	tp->t_flags \|= TF_ACKNOW;
1631	(void) tcp_output(pData, tp);
1632	SOCKET_UNLOCK(so);
1633	return;
1634
1635	dropwithreset:
1636	/* reuses m if m!=NULL, m_free() unnecessary */
1637	if (tiflags & TH_ACK)
1638	tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1639	else
1640	{
1641	if (tiflags & TH_SYN) ti->ti_len++;
1642	tcp_respond(pData, tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1643	TH_RST\|TH_ACK);
1644	}
1645
1646	if (so != &tcb)
1647	SOCKET_UNLOCK(so);
1648	return;
1649
1650	drop:
1651	/*
1652	* Drop space held by incoming segment and return.
1653	*/
1654	m_free(pData, m);
1655
1656	SOCKET_UNLOCK(so);
1657	return;
1658	}
1659
1660	void
1661	tcp_dooptions(PNATState pData, struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti)
1662	{
1663	u_int16_t mss;
1664	int opt, optlen;
1665
1666	DEBUG_CALL("tcp_dooptions");
1667	DEBUG_ARGS((dfd," tp = %lx cnt=%i \n", (long )tp, cnt));
1668
1669	for (; cnt > 0; cnt -= optlen, cp += optlen)
1670	{
1671	opt = cp[0];
1672	if (opt == TCPOPT_EOL)
1673	break;
1674	if (opt == TCPOPT_NOP)
1675	optlen = 1;
1676	else
1677	{
1678	optlen = cp[1];
1679	if (optlen <= 0)
1680	break;
1681	}
1682	switch (opt)
1683	{
1684	default:
1685	continue;
1686
1687	case TCPOPT_MAXSEG:
1688	if (optlen != TCPOLEN_MAXSEG)
1689	continue;
1690	if (!(ti->ti_flags & TH_SYN))
1691	continue;
1692	memcpy((char ) &mss, (char ) cp + 2, sizeof(mss));
1693	NTOHS(mss);
1694	(void) tcp_mss(pData, tp, mss); /* sets t_maxseg */
1695	break;
1696
1697	#if 0
1698	case TCPOPT_WINDOW:
1699	if (optlen != TCPOLEN_WINDOW)
1700	continue;
1701	if (!(ti->ti_flags & TH_SYN))
1702	continue;
1703	tp->t_flags \|= TF_RCVD_SCALE;
1704	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1705	break;
1706
1707	case TCPOPT_TIMESTAMP:
1708	if (optlen != TCPOLEN_TIMESTAMP)
1709	continue;
1710	*ts_present = 1;
1711	memcpy((char ) ts_val, (char )cp + 2, sizeof(*ts_val));
1712	NTOHL(*ts_val);
1713	memcpy((char ) ts_ecr, (char )cp + 6, sizeof(*ts_ecr));
1714	NTOHL(*ts_ecr);
1715
1716	/*
1717	* A timestamp received in a SYN makes
1718	* it ok to send timestamp requests and replies.
1719	*/
1720	if (ti->ti_flags & TH_SYN)
1721	{
1722	tp->t_flags \|= TF_RCVD_TSTMP;
1723	tp->ts_recent = *ts_val;
1724	tp->ts_recent_age = tcp_now;
1725	}
1726	break;
1727	#endif
1728	}
1729	}
1730	}
1731
1732
1733	/*
1734	* Pull out of band byte out of a segment so
1735	* it doesn't appear in the user's data queue.
1736	* It is still reflected in the segment length for
1737	* sequencing purposes.
1738	*/
1739
1740	#if 0
1741	void
1742	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1743	{
1744	int cnt = ti->ti_urp - 1;
1745
1746	while (cnt >= 0)
1747	{
1748	if (m->m_len > cnt)
1749	{
1750	char *cp = mtod(m, caddr_t) + cnt;
1751	struct tcpcb *tp = sototcpcb(so);
1752
1753	tp->t_iobc = *cp;
1754	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1755	memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1756	m->m_len--;
1757	return;
1758	}
1759	cnt -= m->m_len;
1760	m = m->m_next; /* XXX WRONG! Fix it! */
1761	if (m == 0)
1762	break;
1763	}
1764	panic("tcp_pulloutofband");
1765	}
1766	#endif
1767
1768	/*
1769	* Collect new round-trip time estimate
1770	* and update averages and current timeout.
1771	*/
1772
1773	void
1774	tcp_xmit_timer(PNATState pData, register struct tcpcb *tp, int rtt)
1775	{
1776	register short delta;
1777
1778	DEBUG_CALL("tcp_xmit_timer");
1779	DEBUG_ARG("tp = %lx", (long)tp);
1780	DEBUG_ARG("rtt = %d", rtt);
1781
1782	tcpstat.tcps_rttupdated++;
1783	if (tp->t_srtt != 0)
1784	{
1785	/*
1786	* srtt is stored as fixed point with 3 bits after the
1787	* binary point (i.e., scaled by 8). The following magic
1788	* is equivalent to the smoothing algorithm in rfc793 with
1789	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1790	* point). Adjust rtt to origin 0.
1791	*/
1792	delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1793	if ((tp->t_srtt += delta) <= 0)
1794	tp->t_srtt = 1;
1795	/*
1796	* We accumulate a smoothed rtt variance (actually, a
1797	* smoothed mean difference), then set the retransmit
1798	* timer to smoothed rtt + 4 times the smoothed variance.
1799	* rttvar is stored as fixed point with 2 bits after the
1800	* binary point (scaled by 4). The following is
1801	* equivalent to rfc793 smoothing with an alpha of .75
1802	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1803	* rfc793's wired-in beta.
1804	*/
1805	if (delta < 0)
1806	delta = -delta;
1807	delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1808	if ((tp->t_rttvar += delta) <= 0)
1809	tp->t_rttvar = 1;
1810	}
1811	else
1812	{
1813	/*
1814	* No rtt measurement yet - use the unsmoothed rtt.
1815	* Set the variance to half the rtt (so our first
1816	* retransmit happens at 3*rtt).
1817	*/
1818	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1819	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1820	}
1821	tp->t_rtt = 0;
1822	tp->t_rxtshift = 0;
1823
1824	/*
1825	* the retransmit should happen at rtt + 4 * rttvar.
1826	* Because of the way we do the smoothing, srtt and rttvar
1827	* will each average +1/2 tick of bias. When we compute
1828	* the retransmit timer, we want 1/2 tick of rounding and
1829	* 1 extra tick because of +-1/2 tick uncertainty in the
1830	* firing of the timer. The bias will give us exactly the
1831	* 1.5 tick we need. But, because the bias is
1832	* statistical, we have to test that we don't drop below
1833	* the minimum feasible timer (which is 2 ticks).
1834	*/
1835	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1836	(short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1837
1838	/*
1839	* We received an ack for a packet that wasn't retransmitted;
1840	* it is probably safe to discard any error indications we've
1841	* received recently. This isn't quite right, but close enough
1842	* for now (a route might have failed after we sent a segment,
1843	* and the return path might not be symmetrical).
1844	*/
1845	tp->t_softerror = 0;
1846	}
1847
1848	/*
1849	* Determine a reasonable value for maxseg size.
1850	* If the route is known, check route for mtu.
1851	* If none, use an mss that can be handled on the outgoing
1852	* interface without forcing IP to fragment; if bigger than
1853	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1854	* to utilize large mbufs. If no route is found, route has no mtu,
1855	* or the destination isn't local, use a default, hopefully conservative
1856	* size (usually 512 or the default IP max size, but no more than the mtu
1857	* of the interface), as we can't discover anything about intervening
1858	* gateways or networks. We also initialize the congestion/slow start
1859	* window to be a single segment if the destination isn't local.
1860	* While looking at the routing entry, we also initialize other path-dependent
1861	* parameters from pre-set or cached values in the routing entry.
1862	*/
1863
1864	int
1865	tcp_mss(PNATState pData, register struct tcpcb *tp, u_int offer)
1866	{
1867	struct socket *so = tp->t_socket;
1868	int mss;
1869
1870	DEBUG_CALL("tcp_mss");
1871	DEBUG_ARG("tp = %lx", (long)tp);
1872	DEBUG_ARG("offer = %d", offer);
1873
1874	mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
1875	if (offer)
1876	mss = min(mss, offer);
1877	mss = max(mss, 32);
1878	if (mss < tp->t_maxseg \|\| offer != 0)
1879	tp->t_maxseg = mss;
1880
1881	tp->snd_cwnd = mss;
1882
1883	sbreserve(&so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
1884	sbreserve(&so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
1885
1886	DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1887
1888	return mss;
1889	}

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 16445

Download in other formats: