Context Navigation

source: rtems/cpukit/libnetworking/netinet/tcp_input.c @ 33679ec4

4.104.114.84.95

Last change on this file since 33679ec4 was 33679ec4, checked in by Joel Sherrill <joel.sherrill@…>, on 08/21/98 at 13:04:55
All warnings removed.
Property mode set to `100644`
File size: 60.0 KB

Line
1	/*
2	* Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
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 8.12 (Berkeley) 5/24/95
34	* $Id$
35	*/
36
37	#include "opt_tcpdebug.h"
38
39	#ifndef TUBA_INCLUDE
40	#include <sys/param.h>
41	#include <sys/queue.h>
42	#include <sys/systm.h>
43	#include <sys/kernel.h>
44	#include <sys/sysctl.h>
45	#include <sys/malloc.h>
46	#include <sys/mbuf.h>
47	#include <sys/protosw.h>
48	#include <sys/socket.h>
49	#include <sys/socketvar.h>
50	#include <sys/errno.h>
51	#include <sys/syslog.h>
52
53	#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
54
55	#include <net/if.h>
56	#include <net/route.h>
57
58	#include <netinet/in.h>
59	#include <netinet/in_systm.h>
60	#include <netinet/ip.h>
61	#include <netinet/in_pcb.h>
62	#include <netinet/ip_var.h>
63	#include <netinet/tcp.h>
64	#include <netinet/tcp_fsm.h>
65	#include <netinet/tcp_seq.h>
66	#include <netinet/tcp_timer.h>
67	#include <netinet/tcp_var.h>
68	#include <netinet/tcpip.h>
69	#ifdef TCPDEBUG
70	#include <netinet/tcp_debug.h>
71	static struct tcpiphdr tcp_saveti;
72	#endif
73
74	static int tcprexmtthresh = 3;
75	tcp_seq tcp_iss;
76	tcp_cc tcp_ccgen;
77
78	struct tcpstat tcpstat;
79	SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats,
80	CTLFLAG_RD, &tcpstat , tcpstat, "");
81
82	static int log_in_vain = 0;
83	SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
84	&log_in_vain, 0, "");
85
86	u_long tcp_now;
87	struct inpcbhead tcb;
88	struct inpcbinfo tcbinfo;
89
90	static void tcp_dooptions __P((struct tcpcb *,
91	u_char , int, struct tcpiphdr , struct tcpopt *));
92	static void tcp_pulloutofband __P((struct socket *,
93	struct tcpiphdr , struct mbuf ));
94	static int tcp_reass __P((struct tcpcb , struct tcpiphdr , struct mbuf *));
95	static void tcp_xmit_timer __P((struct tcpcb *, int));
96
97	#endif /* TUBA_INCLUDE */
98
99	/*
100	* Insert segment ti into reassembly queue of tcp with
101	* control block tp. Return TH_FIN if reassembly now includes
102	* a segment with FIN. The macro form does the common case inline
103	* (segment is the next to be received on an established connection,
104	* and the queue is empty), avoiding linkage into and removal
105	* from the queue and repetition of various conversions.
106	* Set DELACK for segments received in order, but ack immediately
107	* when segments are out of order (so fast retransmit can work).
108	*/
109	#ifdef TCP_ACK_HACK
110	#define TCP_REASS(tp, ti, m, so, flags) { \
111	if ((ti)->ti_seq == (tp)->rcv_nxt && \
112	(tp)->seg_next == (struct tcpiphdr *)(tp) && \
113	(tp)->t_state == TCPS_ESTABLISHED) { \
114	if (ti->ti_flags & TH_PUSH) \
115	tp->t_flags \|= TF_ACKNOW; \
116	else \
117	tp->t_flags \|= TF_DELACK; \
118	(tp)->rcv_nxt += (ti)->ti_len; \
119	flags = (ti)->ti_flags & TH_FIN; \
120	tcpstat.tcps_rcvpack++;\
121	tcpstat.tcps_rcvbyte += (ti)->ti_len;\
122	sbappend(&(so)->so_rcv, (m)); \
123	sorwakeup(so); \
124	} else { \
125	(flags) = tcp_reass((tp), (ti), (m)); \
126	tp->t_flags \|= TF_ACKNOW; \
127	} \
128	}
129	#else
130	#define TCP_REASS(tp, ti, m, so, flags) { \
131	if ((ti)->ti_seq == (tp)->rcv_nxt && \
132	(tp)->seg_next == (struct tcpiphdr *)(tp) && \
133	(tp)->t_state == TCPS_ESTABLISHED) { \
134	tp->t_flags \|= TF_DELACK; \
135	(tp)->rcv_nxt += (ti)->ti_len; \
136	flags = (ti)->ti_flags & TH_FIN; \
137	tcpstat.tcps_rcvpack++;\
138	tcpstat.tcps_rcvbyte += (ti)->ti_len;\
139	sbappend(&(so)->so_rcv, (m)); \
140	sorwakeup(so); \
141	} else { \
142	(flags) = tcp_reass((tp), (ti), (m)); \
143	tp->t_flags \|= TF_ACKNOW; \
144	} \
145	}
146	#endif
147	#ifndef TUBA_INCLUDE
148
149	static int
150	tcp_reass(tp, ti, m)
151	register struct tcpcb *tp;
152	register struct tcpiphdr *ti;
153	struct mbuf *m;
154	{
155	register struct tcpiphdr *q;
156	struct socket *so = tp->t_inpcb->inp_socket;
157	int flags;
158
159	/*
160	* Call with ti==0 after become established to
161	* force pre-ESTABLISHED data up to user socket.
162	*/
163	if (ti == 0)
164	goto present;
165
166	/*
167	* Find a segment which begins after this one does.
168	*/
169	for (q = tp->seg_next; q != (struct tcpiphdr *)tp;
170	q = (struct tcpiphdr *)q->ti_next)
171	if (SEQ_GT(q->ti_seq, ti->ti_seq))
172	break;
173
174	/*
175	* If there is a preceding segment, it may provide some of
176	* our data already. If so, drop the data from the incoming
177	* segment. If it provides all of our data, drop us.
178	*/
179	if ((struct tcpiphdr )q->ti_prev != (struct tcpiphdr )tp) {
180	register int i;
181	q = (struct tcpiphdr *)q->ti_prev;
182	/* conversion to int (in i) handles seq wraparound */
183	i = q->ti_seq + q->ti_len - ti->ti_seq;
184	if (i > 0) {
185	if (i >= ti->ti_len) {
186	tcpstat.tcps_rcvduppack++;
187	tcpstat.tcps_rcvdupbyte += ti->ti_len;
188	m_freem(m);
189	/*
190	* Try to present any queued data
191	* at the left window edge to the user.
192	* This is needed after the 3-WHS
193	* completes.
194	*/
195	goto present; /* ??? */
196	}
197	m_adj(m, i);
198	ti->ti_len -= i;
199	ti->ti_seq += i;
200	}
201	q = (struct tcpiphdr *)(q->ti_next);
202	}
203	tcpstat.tcps_rcvoopack++;
204	tcpstat.tcps_rcvoobyte += ti->ti_len;
205	REASS_MBUF(ti) = m; /* XXX */
206
207	/*
208	* While we overlap succeeding segments trim them or,
209	* if they are completely covered, dequeue them.
210	*/
211	while (q != (struct tcpiphdr *)tp) {
212	register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq;
213	if (i <= 0)
214	break;
215	if (i < q->ti_len) {
216	q->ti_seq += i;
217	q->ti_len -= i;
218	m_adj(REASS_MBUF(q), i);
219	break;
220	}
221	q = (struct tcpiphdr *)q->ti_next;
222	m = REASS_MBUF((struct tcpiphdr *)q->ti_prev);
223	remque(q->ti_prev);
224	m_freem(m);
225	}
226
227	/*
228	* Stick new segment in its place.
229	*/
230	insque(ti, q->ti_prev);
231
232	present:
233	/*
234	* Present data to user, advancing rcv_nxt through
235	* completed sequence space.
236	*/
237	if (!TCPS_HAVEESTABLISHED(tp->t_state))
238	return (0);
239	ti = tp->seg_next;
240	if (ti == (struct tcpiphdr *)tp \|\| ti->ti_seq != tp->rcv_nxt)
241	return (0);
242	do {
243	tp->rcv_nxt += ti->ti_len;
244	flags = ti->ti_flags & TH_FIN;
245	remque(ti);
246	m = REASS_MBUF(ti);
247	ti = (struct tcpiphdr *)ti->ti_next;
248	if (so->so_state & SS_CANTRCVMORE)
249	m_freem(m);
250	else
251	sbappend(&so->so_rcv, m);
252	} while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
253	sorwakeup(so);
254	return (flags);
255	}
256
257	/*
258	* TCP input routine, follows pages 65-76 of the
259	* protocol specification dated September, 1981 very closely.
260	*/
261	void
262	tcp_input(m, iphlen)
263	register struct mbuf *m;
264	int iphlen;
265	{
266	register struct tcpiphdr *ti;
267	register struct inpcb *inp;
268	u_char *optp = NULL;
269	int optlen = 0;
270	int len, tlen, off;
271	register struct tcpcb *tp = 0;
272	register int tiflags;
273	struct socket *so = 0;
274	int todrop, acked, ourfinisacked, needoutput = 0;
275	struct in_addr laddr;
276	int dropsocket = 0;
277	int iss = 0;
278	u_long tiwin;
279	struct tcpopt to; /* options in this segment */
280	struct rmxp_tao taop; / pointer to our TAO cache entry */
281	struct rmxp_tao tao_noncached; /* in case there's no cached entry */
282	#ifdef TCPDEBUG
283	short ostate = 0;
284	#endif
285
286	bzero((char *)&to, sizeof(to));
287
288	tcpstat.tcps_rcvtotal++;
289	/*
290	* Get IP and TCP header together in first mbuf.
291	* Note: IP leaves IP header in first mbuf.
292	*/
293	ti = mtod(m, struct tcpiphdr *);
294	if (iphlen > sizeof (struct ip))
295	ip_stripoptions(m, (struct mbuf *)0);
296	if (m->m_len < sizeof (struct tcpiphdr)) {
297	if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
298	tcpstat.tcps_rcvshort++;
299	return;
300	}
301	ti = mtod(m, struct tcpiphdr *);
302	}
303
304	/*
305	* Checksum extended TCP header and data.
306	*/
307	tlen = ((struct ip *)ti)->ip_len;
308	len = sizeof (struct ip) + tlen;
309	ti->ti_next = ti->ti_prev = 0;
310	ti->ti_x1 = 0;
311	ti->ti_len = (u_short)tlen;
312	HTONS(ti->ti_len);
313	ti->ti_sum = in_cksum(m, len);
314	if (ti->ti_sum) {
315	tcpstat.tcps_rcvbadsum++;
316	goto drop;
317	}
318	#endif /* TUBA_INCLUDE */
319
320	/*
321	* Check that TCP offset makes sense,
322	* pull out TCP options and adjust length. XXX
323	*/
324	off = ti->ti_off << 2;
325	if (off < sizeof (struct tcphdr) \|\| off > tlen) {
326	tcpstat.tcps_rcvbadoff++;
327	goto drop;
328	}
329	tlen -= off;
330	ti->ti_len = tlen;
331	if (off > sizeof (struct tcphdr)) {
332	if (m->m_len < sizeof(struct ip) + off) {
333	if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) {
334	tcpstat.tcps_rcvshort++;
335	return;
336	}
337	ti = mtod(m, struct tcpiphdr *);
338	}
339	optlen = off - sizeof (struct tcphdr);
340	optp = mtod(m, u_char *) + sizeof (struct tcpiphdr);
341	}
342	tiflags = ti->ti_flags;
343
344	/*
345	* Convert TCP protocol specific fields to host format.
346	*/
347	NTOHL(ti->ti_seq);
348	NTOHL(ti->ti_ack);
349	NTOHS(ti->ti_win);
350	NTOHS(ti->ti_urp);
351
352	/*
353	* Drop TCP, IP headers and TCP options.
354	*/
355	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
356	m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
357
358	/*
359	* Locate pcb for segment.
360	*/
361	findpcb:
362	inp = in_pcblookuphash(&tcbinfo, ti->ti_src, ti->ti_sport,
363	ti->ti_dst, ti->ti_dport, 1);
364
365	/*
366	* If the state is CLOSED (i.e., TCB does not exist) then
367	* all data in the incoming segment is discarded.
368	* If the TCB exists but is in CLOSED state, it is embryonic,
369	* but should either do a listen or a connect soon.
370	*/
371	if (inp == NULL) {
372	if (log_in_vain && tiflags & TH_SYN) {
373	char buf[4*sizeof "123"];
374
375	strcpy(buf, inet_ntoa(ti->ti_dst));
376	log(LOG_INFO, "Connection attempt to TCP %s:%d"
377	" from %s:%d\n",
378	buf, ntohs(ti->ti_dport),
379	inet_ntoa(ti->ti_src), ntohs(ti->ti_sport));
380	}
381	goto dropwithreset;
382	}
383	tp = intotcpcb(inp);
384	if (tp == 0)
385	goto dropwithreset;
386	if (tp->t_state == TCPS_CLOSED)
387	goto drop;
388
389	/* Unscale the window into a 32-bit value. */
390	if ((tiflags & TH_SYN) == 0)
391	tiwin = ti->ti_win << tp->snd_scale;
392	else
393	tiwin = ti->ti_win;
394
395	so = inp->inp_socket;
396	if (so->so_options & (SO_DEBUG\|SO_ACCEPTCONN)) {
397	#ifdef TCPDEBUG
398	if (so->so_options & SO_DEBUG) {
399	ostate = tp->t_state;
400	tcp_saveti = *ti;
401	}
402	#endif
403	if (so->so_options & SO_ACCEPTCONN) {
404	register struct tcpcb *tp0 = tp;
405	struct socket *so2;
406	if ((tiflags & (TH_RST\|TH_ACK\|TH_SYN)) != TH_SYN) {
407	/*
408	* Note: dropwithreset makes sure we don't
409	* send a RST in response to a RST.
410	*/
411	if (tiflags & TH_ACK) {
412	tcpstat.tcps_badsyn++;
413	goto dropwithreset;
414	}
415	goto drop;
416	}
417	so2 = sonewconn(so, 0);
418	if (so2 == 0) {
419	tcpstat.tcps_listendrop++;
420	so2 = sodropablereq(so);
421	if (so2) {
422	tcp_drop(sototcpcb(so2), ETIMEDOUT);
423	so2 = sonewconn(so, 0);
424	}
425	if (!so2)
426	goto drop;
427	}
428	so = so2;
429	/*
430	* This is ugly, but ....
431	*
432	* Mark socket as temporary until we're
433	* committed to keeping it. The code at
434	* ``drop'' and ``dropwithreset'' check the
435	* flag dropsocket to see if the temporary
436	* socket created here should be discarded.
437	* We mark the socket as discardable until
438	* we're committed to it below in TCPS_LISTEN.
439	*/
440	dropsocket++;
441	inp = (struct inpcb *)so->so_pcb;
442	inp->inp_laddr = ti->ti_dst;
443	inp->inp_lport = ti->ti_dport;
444	in_pcbrehash(inp);
445	#if BSD>=43
446	inp->inp_options = ip_srcroute();
447	#endif
448	tp = intotcpcb(inp);
449	tp->t_state = TCPS_LISTEN;
450	tp->t_flags \|= tp0->t_flags & (TF_NOPUSH\|TF_NOOPT);
451
452	/* Compute proper scaling value from buffer space */
453	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
454	TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat)
455	tp->request_r_scale++;
456	}
457	}
458
459	/*
460	* Segment received on connection.
461	* Reset idle time and keep-alive timer.
462	*/
463	tp->t_idle = 0;
464	if (TCPS_HAVEESTABLISHED(tp->t_state))
465	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
466
467	/*
468	* Process options if not in LISTEN state,
469	* else do it below (after getting remote address).
470	*/
471	if (tp->t_state != TCPS_LISTEN)
472	tcp_dooptions(tp, optp, optlen, ti, &to);
473
474	/*
475	* Header prediction: check for the two common cases
476	* of a uni-directional data xfer. If the packet has
477	* no control flags, is in-sequence, the window didn't
478	* change and we're not retransmitting, it's a
479	* candidate. If the length is zero and the ack moved
480	* forward, we're the sender side of the xfer. Just
481	* free the data acked & wake any higher level process
482	* that was blocked waiting for space. If the length
483	* is non-zero and the ack didn't move, we're the
484	* receiver side. If we're getting packets in-order
485	* (the reassembly queue is empty), add the data to
486	* the socket buffer and note that we need a delayed ack.
487	* Make sure that the hidden state-flags are also off.
488	* Since we check for TCPS_ESTABLISHED above, it can only
489	* be TH_NEEDSYN.
490	*/
491	if (tp->t_state == TCPS_ESTABLISHED &&
492	(tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK &&
493	((tp->t_flags & (TF_NEEDSYN\|TF_NEEDFIN)) == 0) &&
494	((to.to_flag & TOF_TS) == 0 \|\|
495	TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
496	/*
497	* Using the CC option is compulsory if once started:
498	* the segment is OK if no T/TCP was negotiated or
499	* if the segment has a CC option equal to CCrecv
500	*/
501	((tp->t_flags & (TF_REQ_CC\|TF_RCVD_CC)) != (TF_REQ_CC\|TF_RCVD_CC) \|\|
502	((to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
503	ti->ti_seq == tp->rcv_nxt &&
504	tiwin && tiwin == tp->snd_wnd &&
505	tp->snd_nxt == tp->snd_max) {
506
507	/*
508	* If last ACK falls within this segment's sequence numbers,
509	* record the timestamp.
510	* NOTE that the test is modified according to the latest
511	* proposal of the tcplw@cray.com list (Braden 1993/04/26).
512	*/
513	if ((to.to_flag & TOF_TS) != 0 &&
514	SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) {
515	tp->ts_recent_age = tcp_now;
516	tp->ts_recent = to.to_tsval;
517	}
518
519	if (ti->ti_len == 0) {
520	if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
521	SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
522	tp->snd_cwnd >= tp->snd_wnd &&
523	tp->t_dupacks < tcprexmtthresh) {
524	/*
525	* this is a pure ack for outstanding data.
526	*/
527	++tcpstat.tcps_predack;
528	if ((to.to_flag & TOF_TS) != 0)
529	tcp_xmit_timer(tp,
530	tcp_now - to.to_tsecr + 1);
531	else if (tp->t_rtt &&
532	SEQ_GT(ti->ti_ack, tp->t_rtseq))
533	tcp_xmit_timer(tp, tp->t_rtt);
534	acked = ti->ti_ack - tp->snd_una;
535	tcpstat.tcps_rcvackpack++;
536	tcpstat.tcps_rcvackbyte += acked;
537	sbdrop(&so->so_snd, acked);
538	tp->snd_una = ti->ti_ack;
539	m_freem(m);
540
541	/*
542	* If all outstanding data are acked, stop
543	* retransmit timer, otherwise restart timer
544	* using current (possibly backed-off) value.
545	* If process is waiting for space,
546	* wakeup/selwakeup/signal. If data
547	* are ready to send, let tcp_output
548	* decide between more output or persist.
549	*/
550	if (tp->snd_una == tp->snd_max)
551	tp->t_timer[TCPT_REXMT] = 0;
552	else if (tp->t_timer[TCPT_PERSIST] == 0)
553	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
554
555	if (so->so_snd.sb_flags & SB_NOTIFY)
556	sowwakeup(so);
557	if (so->so_snd.sb_cc)
558	(void) tcp_output(tp);
559	return;
560	}
561	} else if (ti->ti_ack == tp->snd_una &&
562	tp->seg_next == (struct tcpiphdr *)tp &&
563	ti->ti_len <= sbspace(&so->so_rcv)) {
564	/*
565	* this is a pure, in-sequence data packet
566	* with nothing on the reassembly queue and
567	* we have enough buffer space to take it.
568	*/
569	++tcpstat.tcps_preddat;
570	tp->rcv_nxt += ti->ti_len;
571	tcpstat.tcps_rcvpack++;
572	tcpstat.tcps_rcvbyte += ti->ti_len;
573	/*
574	* Add data to socket buffer.
575	*/
576	sbappend(&so->so_rcv, m);
577	sorwakeup(so);
578	#ifdef TCP_ACK_HACK
579	/*
580	* If this is a short packet, then ACK now - with Nagel
581	* congestion avoidance sender won't send more until
582	* he gets an ACK.
583	*/
584	if (tiflags & TH_PUSH) {
585	tp->t_flags \|= TF_ACKNOW;
586	tcp_output(tp);
587	} else {
588	tp->t_flags \|= TF_DELACK;
589	}
590	#else
591	tp->t_flags \|= TF_DELACK;
592	#endif
593	return;
594	}
595	}
596
597	/*
598	* Calculate amount of space in receive window,
599	* and then do TCP input processing.
600	* Receive window is amount of space in rcv queue,
601	* but not less than advertised window.
602	*/
603	{ int win;
604
605	win = sbspace(&so->so_rcv);
606	if (win < 0)
607	win = 0;
608	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
609	}
610
611	switch (tp->t_state) {
612
613	/*
614	* If the state is LISTEN then ignore segment if it contains an RST.
615	* If the segment contains an ACK then it is bad and send a RST.
616	* If it does not contain a SYN then it is not interesting; drop it.
617	* If it is from this socket, drop it, it must be forged.
618	* Don't bother responding if the destination was a broadcast.
619	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
620	* tp->iss, and send a segment:
621	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
622	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
623	* Fill in remote peer address fields if not previously specified.
624	* Enter SYN_RECEIVED state, and process any other fields of this
625	* segment in this state.
626	*/
627	case TCPS_LISTEN: {
628	struct mbuf *am;
629	register struct sockaddr_in *sin;
630
631	if (tiflags & TH_RST)
632	goto drop;
633	if (tiflags & TH_ACK)
634	goto dropwithreset;
635	if ((tiflags & TH_SYN) == 0)
636	goto drop;
637	if ((ti->ti_dport == ti->ti_sport) &&
638	(ti->ti_dst.s_addr == ti->ti_src.s_addr))
639	goto drop;
640	/*
641	* RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
642	* in_broadcast() should never return true on a received
643	* packet with M_BCAST not set.
644	*/
645	if (m->m_flags & (M_BCAST\|M_MCAST) \|\|
646	IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
647	goto drop;
648	am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */
649	if (am == NULL)
650	goto drop;
651	am->m_len = sizeof (struct sockaddr_in);
652	sin = mtod(am, struct sockaddr_in *);
653	sin->sin_family = AF_INET;
654	sin->sin_len = sizeof(*sin);
655	sin->sin_addr = ti->ti_src;
656	sin->sin_port = ti->ti_sport;
657	bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
658	laddr = inp->inp_laddr;
659	if (inp->inp_laddr.s_addr == INADDR_ANY)
660	inp->inp_laddr = ti->ti_dst;
661	if (in_pcbconnect(inp, am)) {
662	inp->inp_laddr = laddr;
663	(void) m_free(am);
664	goto drop;
665	}
666	(void) m_free(am);
667	tp->t_template = tcp_template(tp);
668	if (tp->t_template == 0) {
669	tp = tcp_drop(tp, ENOBUFS);
670	dropsocket = 0; /* socket is already gone */
671	goto drop;
672	}
673	if ((taop = tcp_gettaocache(inp)) == NULL) {
674	taop = &tao_noncached;
675	bzero(taop, sizeof(*taop));
676	}
677	tcp_dooptions(tp, optp, optlen, ti, &to);
678	if (iss)
679	tp->iss = iss;
680	else
681	tp->iss = tcp_iss;
682	tcp_iss += TCP_ISSINCR/4;
683	tp->irs = ti->ti_seq;
684	tcp_sendseqinit(tp);
685	tcp_rcvseqinit(tp);
686	/*
687	* Initialization of the tcpcb for transaction;
688	* set SND.WND = SEG.WND,
689	* initialize CCsend and CCrecv.
690	*/
691	tp->snd_wnd = tiwin; /* initial send-window */
692	tp->cc_send = CC_INC(tcp_ccgen);
693	tp->cc_recv = to.to_cc;
694	/*
695	* Perform TAO test on incoming CC (SEG.CC) option, if any.
696	* - compare SEG.CC against cached CC from the same host,
697	* if any.
698	* - if SEG.CC > chached value, SYN must be new and is accepted
699	* immediately: save new CC in the cache, mark the socket
700	* connected, enter ESTABLISHED state, turn on flag to
701	* send a SYN in the next segment.
702	* A virtual advertised window is set in rcv_adv to
703	* initialize SWS prevention. Then enter normal segment
704	* processing: drop SYN, process data and FIN.
705	* - otherwise do a normal 3-way handshake.
706	*/
707	if ((to.to_flag & TOF_CC) != 0) {
708	if (taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) {
709	taop->tao_cc = to.to_cc;
710	tp->t_state = TCPS_ESTABLISHED;
711
712	/*
713	* If there is a FIN, or if there is data and the
714	* connection is local, then delay SYN,ACK(SYN) in
715	* the hope of piggy-backing it on a response
716	* segment. Otherwise must send ACK now in case
717	* the other side is slow starting.
718	*/
719	if ((tiflags & TH_FIN) \|\| (ti->ti_len != 0 &&
720	in_localaddr(inp->inp_faddr)))
721	tp->t_flags \|= (TF_DELACK \| TF_NEEDSYN);
722	else
723	tp->t_flags \|= (TF_ACKNOW \| TF_NEEDSYN);
724
725	/*
726	* Limit the `virtual advertised window' to TCP_MAXWIN
727	* here. Even if we requested window scaling, it will
728	* become effective only later when our SYN is acked.
729	*/
730	tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN);
731	tcpstat.tcps_connects++;
732	soisconnected(so);
733	tp->t_timer[TCPT_KEEP] = tcp_keepinit;
734	dropsocket = 0; /* committed to socket */
735	tcpstat.tcps_accepts++;
736	goto trimthenstep6;
737	}
738	/* else do standard 3-way handshake */
739	} else {
740	/*
741	* No CC option, but maybe CC.NEW:
742	* invalidate cached value.
743	*/
744	taop->tao_cc = 0;
745	}
746	/*
747	* TAO test failed or there was no CC option,
748	* do a standard 3-way handshake.
749	*/
750	tp->t_flags \|= TF_ACKNOW;
751	tp->t_state = TCPS_SYN_RECEIVED;
752	tp->t_timer[TCPT_KEEP] = tcp_keepinit;
753	dropsocket = 0; /* committed to socket */
754	tcpstat.tcps_accepts++;
755	goto trimthenstep6;
756	}
757
758	/*
759	* If the state is SYN_RECEIVED:
760	* if seg contains SYN/ACK, send a RST.
761	* if seg contains an ACK, but not for our SYN/ACK, send a RST.
762	*/
763	case TCPS_SYN_RECEIVED:
764	if (tiflags & TH_ACK) {
765	if (tiflags & TH_SYN) {
766	tcpstat.tcps_badsyn++;
767	goto dropwithreset;
768	}
769	if (SEQ_LEQ(ti->ti_ack, tp->snd_una) \|\|
770	SEQ_GT(ti->ti_ack, tp->snd_max))
771	goto dropwithreset;
772	}
773	break;
774
775	/*
776	* If the state is SYN_SENT:
777	* if seg contains an ACK, but not for our SYN, drop the input.
778	* if seg contains a RST, then drop the connection.
779	* if seg does not contain SYN, then drop it.
780	* Otherwise this is an acceptable SYN segment
781	* initialize tp->rcv_nxt and tp->irs
782	* if seg contains ack then advance tp->snd_una
783	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
784	* arrange for segment to be acked (eventually)
785	* continue processing rest of data/controls, beginning with URG
786	*/
787	case TCPS_SYN_SENT:
788	if ((taop = tcp_gettaocache(inp)) == NULL) {
789	taop = &tao_noncached;
790	bzero(taop, sizeof(*taop));
791	}
792
793	if ((tiflags & TH_ACK) &&
794	(SEQ_LEQ(ti->ti_ack, tp->iss) \|\|
795	SEQ_GT(ti->ti_ack, tp->snd_max))) {
796	/*
797	* If we have a cached CCsent for the remote host,
798	* hence we haven't just crashed and restarted,
799	* do not send a RST. This may be a retransmission
800	* from the other side after our earlier ACK was lost.
801	* Our new SYN, when it arrives, will serve as the
802	* needed ACK.
803	*/
804	if (taop->tao_ccsent != 0)
805	goto drop;
806	else
807	goto dropwithreset;
808	}
809	if (tiflags & TH_RST) {
810	if (tiflags & TH_ACK)
811	tp = tcp_drop(tp, ECONNREFUSED);
812	goto drop;
813	}
814	if ((tiflags & TH_SYN) == 0)
815	goto drop;
816	tp->snd_wnd = ti->ti_win; /* initial send window */
817	tp->cc_recv = to.to_cc; /* foreign CC */
818
819	tp->irs = ti->ti_seq;
820	tcp_rcvseqinit(tp);
821	if (tiflags & TH_ACK) {
822	/*
823	* Our SYN was acked. If segment contains CC.ECHO
824	* option, check it to make sure this segment really
825	* matches our SYN. If not, just drop it as old
826	* duplicate, but send an RST if we're still playing
827	* by the old rules. If no CC.ECHO option, make sure
828	* we don't get fooled into using T/TCP.
829	*/
830	if (to.to_flag & TOF_CCECHO) {
831	if (tp->cc_send != to.to_ccecho)
832	if (taop->tao_ccsent != 0)
833	goto drop;
834	else
835	goto dropwithreset;
836	} else
837	tp->t_flags &= ~TF_RCVD_CC;
838	tcpstat.tcps_connects++;
839	soisconnected(so);
840	/* Do window scaling on this connection? */
841	if ((tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE)) ==
842	(TF_RCVD_SCALE\|TF_REQ_SCALE)) {
843	tp->snd_scale = tp->requested_s_scale;
844	tp->rcv_scale = tp->request_r_scale;
845	}
846	/* Segment is acceptable, update cache if undefined. */
847	if (taop->tao_ccsent == 0)
848	taop->tao_ccsent = to.to_ccecho;
849
850	tp->rcv_adv += tp->rcv_wnd;
851	tp->snd_una++; /* SYN is acked */
852	/*
853	* If there's data, delay ACK; if there's also a FIN
854	* ACKNOW will be turned on later.
855	*/
856	if (ti->ti_len != 0)
857	tp->t_flags \|= TF_DELACK;
858	else
859	tp->t_flags \|= TF_ACKNOW;
860	/*
861	* Received <SYN,ACK> in SYN_SENT[*] state.
862	* Transitions:
863	* SYN_SENT --> ESTABLISHED
864	* SYN_SENT* --> FIN_WAIT_1
865	*/
866	if (tp->t_flags & TF_NEEDFIN) {
867	tp->t_state = TCPS_FIN_WAIT_1;
868	tp->t_flags &= ~TF_NEEDFIN;
869	tiflags &= ~TH_SYN;
870	} else {
871	tp->t_state = TCPS_ESTABLISHED;
872	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
873	}
874	} else {
875	/*
876	* Received initial SYN in SYN-SENT[*] state => simul-
877	* taneous open. If segment contains CC option and there is
878	* a cached CC, apply TAO test; if it succeeds, connection is
879	* half-synchronized. Otherwise, do 3-way handshake:
880	* SYN-SENT -> SYN-RECEIVED
881	* SYN-SENT* -> SYN-RECEIVED*
882	* If there was no CC option, clear cached CC value.
883	*/
884	tp->t_flags \|= TF_ACKNOW;
885	tp->t_timer[TCPT_REXMT] = 0;
886	if (to.to_flag & TOF_CC) {
887	if (taop->tao_cc != 0 &&
888	CC_GT(to.to_cc, taop->tao_cc)) {
889	/*
890	* update cache and make transition:
891	* SYN-SENT -> ESTABLISHED*
892	* SYN-SENT* -> FIN-WAIT-1*
893	*/
894	taop->tao_cc = to.to_cc;
895	if (tp->t_flags & TF_NEEDFIN) {
896	tp->t_state = TCPS_FIN_WAIT_1;
897	tp->t_flags &= ~TF_NEEDFIN;
898	} else {
899	tp->t_state = TCPS_ESTABLISHED;
900	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
901	}
902	tp->t_flags \|= TF_NEEDSYN;
903	} else
904	tp->t_state = TCPS_SYN_RECEIVED;
905	} else {
906	/* CC.NEW or no option => invalidate cache */
907	taop->tao_cc = 0;
908	tp->t_state = TCPS_SYN_RECEIVED;
909	}
910	}
911
912	trimthenstep6:
913	/*
914	* Advance ti->ti_seq to correspond to first data byte.
915	* If data, trim to stay within window,
916	* dropping FIN if necessary.
917	*/
918	ti->ti_seq++;
919	if (ti->ti_len > tp->rcv_wnd) {
920	todrop = ti->ti_len - tp->rcv_wnd;
921	m_adj(m, -todrop);
922	ti->ti_len = tp->rcv_wnd;
923	tiflags &= ~TH_FIN;
924	tcpstat.tcps_rcvpackafterwin++;
925	tcpstat.tcps_rcvbyteafterwin += todrop;
926	}
927	tp->snd_wl1 = ti->ti_seq - 1;
928	tp->rcv_up = ti->ti_seq;
929	/*
930	* Client side of transaction: already sent SYN and data.
931	* If the remote host used T/TCP to validate the SYN,
932	* our data will be ACK'd; if so, enter normal data segment
933	* processing in the middle of step 5, ack processing.
934	* Otherwise, goto step 6.
935	*/
936	if (tiflags & TH_ACK)
937	goto process_ACK;
938	goto step6;
939	/*
940	* If the state is LAST_ACK or CLOSING or TIME_WAIT:
941	* if segment contains a SYN and CC [not CC.NEW] option:
942	* if state == TIME_WAIT and connection duration > MSL,
943	* drop packet and send RST;
944	*
945	* if SEG.CC > CCrecv then is new SYN, and can implicitly
946	* ack the FIN (and data) in retransmission queue.
947	* Complete close and delete TCPCB. Then reprocess
948	* segment, hoping to find new TCPCB in LISTEN state;
949	*
950	* else must be old SYN; drop it.
951	* else do normal processing.
952	*/
953	case TCPS_LAST_ACK:
954	case TCPS_CLOSING:
955	case TCPS_TIME_WAIT:
956	if ((tiflags & TH_SYN) &&
957	(to.to_flag & TOF_CC) && tp->cc_recv != 0) {
958	if (tp->t_state == TCPS_TIME_WAIT &&
959	tp->t_duration > TCPTV_MSL)
960	goto dropwithreset;
961	if (CC_GT(to.to_cc, tp->cc_recv)) {
962	tp = tcp_close(tp);
963	goto findpcb;
964	}
965	else
966	goto drop;
967	}
968	break; /* continue normal processing */
969	}
970
971	/*
972	* States other than LISTEN or SYN_SENT.
973	* First check timestamp, if present.
974	* Then check the connection count, if present.
975	* Then check that at least some bytes of segment are within
976	* receive window. If segment begins before rcv_nxt,
977	* drop leading data (and SYN); if nothing left, just ack.
978	*
979	* RFC 1323 PAWS: If we have a timestamp reply on this segment
980	* and it's less than ts_recent, drop it.
981	*/
982	if ((to.to_flag & TOF_TS) != 0 && (tiflags & TH_RST) == 0 &&
983	tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) {
984
985	/* Check to see if ts_recent is over 24 days old. */
986	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
987	/*
988	* Invalidate ts_recent. If this segment updates
989	* ts_recent, the age will be reset later and ts_recent
990	* will get a valid value. If it does not, setting
991	* ts_recent to zero will at least satisfy the
992	* requirement that zero be placed in the timestamp
993	* echo reply when ts_recent isn't valid. The
994	* age isn't reset until we get a valid ts_recent
995	* because we don't want out-of-order segments to be
996	* dropped when ts_recent is old.
997	*/
998	tp->ts_recent = 0;
999	} else {
1000	tcpstat.tcps_rcvduppack++;
1001	tcpstat.tcps_rcvdupbyte += ti->ti_len;
1002	tcpstat.tcps_pawsdrop++;
1003	goto dropafterack;
1004	}
1005	}
1006
1007	/*
1008	* T/TCP mechanism
1009	* If T/TCP was negotiated and the segment doesn't have CC,
1010	* or if it's CC is wrong then drop the segment.
1011	* RST segments do not have to comply with this.
1012	*/
1013	if ((tp->t_flags & (TF_REQ_CC\|TF_RCVD_CC)) == (TF_REQ_CC\|TF_RCVD_CC) &&
1014	((to.to_flag & TOF_CC) == 0 \|\| tp->cc_recv != to.to_cc) &&
1015	(tiflags & TH_RST) == 0)
1016	goto dropafterack;
1017
1018	todrop = tp->rcv_nxt - ti->ti_seq;
1019	if (todrop > 0) {
1020	if (tiflags & TH_SYN) {
1021	tiflags &= ~TH_SYN;
1022	ti->ti_seq++;
1023	if (ti->ti_urp > 1)
1024	ti->ti_urp--;
1025	else
1026	tiflags &= ~TH_URG;
1027	todrop--;
1028	}
1029	/*
1030	* Following if statement from Stevens, vol. 2, p. 960.
1031	*/
1032	if (todrop > ti->ti_len
1033	\|\| (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) {
1034	/*
1035	* Any valid FIN must be to the left of the window.
1036	* At this point the FIN must be a duplicate or out
1037	* of sequence; drop it.
1038	*/
1039	tiflags &= ~TH_FIN;
1040
1041	/*
1042	* Send an ACK to resynchronize and drop any data.
1043	* But keep on processing for RST or ACK.
1044	*/
1045	tp->t_flags \|= TF_ACKNOW;
1046	todrop = ti->ti_len;
1047	tcpstat.tcps_rcvduppack++;
1048	tcpstat.tcps_rcvdupbyte += todrop;
1049	} else {
1050	tcpstat.tcps_rcvpartduppack++;
1051	tcpstat.tcps_rcvpartdupbyte += todrop;
1052	}
1053	m_adj(m, todrop);
1054	ti->ti_seq += todrop;
1055	ti->ti_len -= todrop;
1056	if (ti->ti_urp > todrop)
1057	ti->ti_urp -= todrop;
1058	else {
1059	tiflags &= ~TH_URG;
1060	ti->ti_urp = 0;
1061	}
1062	}
1063
1064	/*
1065	* If new data are received on a connection after the
1066	* user processes are gone, then RST the other end.
1067	*/
1068	if ((so->so_state & SS_NOFDREF) &&
1069	tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
1070	tp = tcp_close(tp);
1071	tcpstat.tcps_rcvafterclose++;
1072	goto dropwithreset;
1073	}
1074
1075	/*
1076	* If segment ends after window, drop trailing data
1077	* (and PUSH and FIN); if nothing left, just ACK.
1078	*/
1079	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1080	if (todrop > 0) {
1081	tcpstat.tcps_rcvpackafterwin++;
1082	if (todrop >= ti->ti_len) {
1083	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1084	/*
1085	* If a new connection request is received
1086	* while in TIME_WAIT, drop the old connection
1087	* and start over if the sequence numbers
1088	* are above the previous ones.
1089	*/
1090	if (tiflags & TH_SYN &&
1091	tp->t_state == TCPS_TIME_WAIT &&
1092	SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
1093	iss = tp->rcv_nxt + TCP_ISSINCR;
1094	tp = tcp_close(tp);
1095	goto findpcb;
1096	}
1097	/*
1098	* If window is closed can only take segments at
1099	* window edge, and have to drop data and PUSH from
1100	* incoming segments. Continue processing, but
1101	* remember to ack. Otherwise, drop segment
1102	* and ack.
1103	*/
1104	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
1105	tp->t_flags \|= TF_ACKNOW;
1106	tcpstat.tcps_rcvwinprobe++;
1107	} else
1108	goto dropafterack;
1109	} else
1110	tcpstat.tcps_rcvbyteafterwin += todrop;
1111	m_adj(m, -todrop);
1112	ti->ti_len -= todrop;
1113	tiflags &= ~(TH_PUSH\|TH_FIN);
1114	}
1115
1116	/*
1117	* If last ACK falls within this segment's sequence numbers,
1118	* record its timestamp.
1119	* NOTE that the test is modified according to the latest
1120	* proposal of the tcplw@cray.com list (Braden 1993/04/26).
1121	*/
1122	if ((to.to_flag & TOF_TS) != 0 &&
1123	SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) {
1124	tp->ts_recent_age = tcp_now;
1125	tp->ts_recent = to.to_tsval;
1126	}
1127
1128	/*
1129	* If the RST bit is set examine the state:
1130	* SYN_RECEIVED STATE:
1131	* If passive open, return to LISTEN state.
1132	* If active open, inform user that connection was refused.
1133	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1134	* Inform user that connection was reset, and close tcb.
1135	* CLOSING, LAST_ACK, TIME_WAIT STATES
1136	* Close the tcb.
1137	*/
1138	if (tiflags&TH_RST) switch (tp->t_state) {
1139
1140	case TCPS_SYN_RECEIVED:
1141	so->so_error = ECONNREFUSED;
1142	goto close;
1143
1144	case TCPS_ESTABLISHED:
1145	case TCPS_FIN_WAIT_1:
1146	case TCPS_FIN_WAIT_2:
1147	case TCPS_CLOSE_WAIT:
1148	so->so_error = ECONNRESET;
1149	close:
1150	tp->t_state = TCPS_CLOSED;
1151	tcpstat.tcps_drops++;
1152	tp = tcp_close(tp);
1153	goto drop;
1154
1155	case TCPS_CLOSING:
1156	case TCPS_LAST_ACK:
1157	case TCPS_TIME_WAIT:
1158	tp = tcp_close(tp);
1159	goto drop;
1160	}
1161
1162	/*
1163	* If a SYN is in the window, then this is an
1164	* error and we send an RST and drop the connection.
1165	*/
1166	if (tiflags & TH_SYN) {
1167	tp = tcp_drop(tp, ECONNRESET);
1168	goto dropwithreset;
1169	}
1170
1171	/*
1172	* If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1173	* flag is on (half-synchronized state), then queue data for
1174	* later processing; else drop segment and return.
1175	*/
1176	if ((tiflags & TH_ACK) == 0) {
1177	if (tp->t_state == TCPS_SYN_RECEIVED \|\|
1178	(tp->t_flags & TF_NEEDSYN))
1179	goto step6;
1180	else
1181	goto drop;
1182	}
1183
1184	/*
1185	* Ack processing.
1186	*/
1187	switch (tp->t_state) {
1188
1189	/*
1190	* In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1191	* ESTABLISHED state and continue processing.
1192	* The ACK was checked above.
1193	*/
1194	case TCPS_SYN_RECEIVED:
1195
1196	tcpstat.tcps_connects++;
1197	soisconnected(so);
1198	/* Do window scaling? */
1199	if ((tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE)) ==
1200	(TF_RCVD_SCALE\|TF_REQ_SCALE)) {
1201	tp->snd_scale = tp->requested_s_scale;
1202	tp->rcv_scale = tp->request_r_scale;
1203	}
1204	/*
1205	* Upon successful completion of 3-way handshake,
1206	* update cache.CC if it was undefined, pass any queued
1207	* data to the user, and advance state appropriately.
1208	*/
1209	if ((taop = tcp_gettaocache(inp)) != NULL &&
1210	taop->tao_cc == 0)
1211	taop->tao_cc = tp->cc_recv;
1212
1213	/*
1214	* Make transitions:
1215	* SYN-RECEIVED -> ESTABLISHED
1216	* SYN-RECEIVED* -> FIN-WAIT-1
1217	*/
1218	if (tp->t_flags & TF_NEEDFIN) {
1219	tp->t_state = TCPS_FIN_WAIT_1;
1220	tp->t_flags &= ~TF_NEEDFIN;
1221	} else {
1222	tp->t_state = TCPS_ESTABLISHED;
1223	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
1224	}
1225	/*
1226	* If segment contains data or ACK, will call tcp_reass()
1227	* later; if not, do so now to pass queued data to user.
1228	*/
1229	if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0)
1230	(void) tcp_reass(tp, (struct tcpiphdr *)0,
1231	(struct mbuf *)0);
1232	tp->snd_wl1 = ti->ti_seq - 1;
1233	/* fall into ... */
1234
1235	/*
1236	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1237	* ACKs. If the ack is in the range
1238	* tp->snd_una < ti->ti_ack <= tp->snd_max
1239	* then advance tp->snd_una to ti->ti_ack and drop
1240	* data from the retransmission queue. If this ACK reflects
1241	* more up to date window information we update our window information.
1242	*/
1243	case TCPS_ESTABLISHED:
1244	case TCPS_FIN_WAIT_1:
1245	case TCPS_FIN_WAIT_2:
1246	case TCPS_CLOSE_WAIT:
1247	case TCPS_CLOSING:
1248	case TCPS_LAST_ACK:
1249	case TCPS_TIME_WAIT:
1250
1251	if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
1252	if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
1253	tcpstat.tcps_rcvdupack++;
1254	/*
1255	* If we have outstanding data (other than
1256	* a window probe), this is a completely
1257	* duplicate ack (ie, window info didn't
1258	* change), the ack is the biggest we've
1259	* seen and we've seen exactly our rexmt
1260	* threshhold of them, assume a packet
1261	* has been dropped and retransmit it.
1262	* Kludge snd_nxt & the congestion
1263	* window so we send only this one
1264	* packet.
1265	*
1266	* We know we're losing at the current
1267	* window size so do congestion avoidance
1268	* (set ssthresh to half the current window
1269	* and pull our congestion window back to
1270	* the new ssthresh).
1271	*
1272	* Dup acks mean that packets have left the
1273	* network (they're now cached at the receiver)
1274	* so bump cwnd by the amount in the receiver
1275	* to keep a constant cwnd packets in the
1276	* network.
1277	*/
1278	if (tp->t_timer[TCPT_REXMT] == 0 \|\|
1279	ti->ti_ack != tp->snd_una)
1280	tp->t_dupacks = 0;
1281	else if (++tp->t_dupacks == tcprexmtthresh) {
1282	tcp_seq onxt = tp->snd_nxt;
1283	u_int win =
1284	min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1285	tp->t_maxseg;
1286
1287	if (win < 2)
1288	win = 2;
1289	tp->snd_ssthresh = win * tp->t_maxseg;
1290	tp->t_timer[TCPT_REXMT] = 0;
1291	tp->t_rtt = 0;
1292	tp->snd_nxt = ti->ti_ack;
1293	tp->snd_cwnd = tp->t_maxseg;
1294	(void) tcp_output(tp);
1295	tp->snd_cwnd = tp->snd_ssthresh +
1296	tp->t_maxseg * tp->t_dupacks;
1297	if (SEQ_GT(onxt, tp->snd_nxt))
1298	tp->snd_nxt = onxt;
1299	goto drop;
1300	} else if (tp->t_dupacks > tcprexmtthresh) {
1301	tp->snd_cwnd += tp->t_maxseg;
1302	(void) tcp_output(tp);
1303	goto drop;
1304	}
1305	} else
1306	tp->t_dupacks = 0;
1307	break;
1308	}
1309	/*
1310	* If the congestion window was inflated to account
1311	* for the other side's cached packets, retract it.
1312	*/
1313	if (tp->t_dupacks >= tcprexmtthresh &&
1314	tp->snd_cwnd > tp->snd_ssthresh)
1315	tp->snd_cwnd = tp->snd_ssthresh;
1316	tp->t_dupacks = 0;
1317	if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
1318	tcpstat.tcps_rcvacktoomuch++;
1319	goto dropafterack;
1320	}
1321	/*
1322	* If we reach this point, ACK is not a duplicate,
1323	* i.e., it ACKs something we sent.
1324	*/
1325	if (tp->t_flags & TF_NEEDSYN) {
1326	/*
1327	* T/TCP: Connection was half-synchronized, and our
1328	* SYN has been ACK'd (so connection is now fully
1329	* synchronized). Go to non-starred state,
1330	* increment snd_una for ACK of SYN, and check if
1331	* we can do window scaling.
1332	*/
1333	tp->t_flags &= ~TF_NEEDSYN;
1334	tp->snd_una++;
1335	/* Do window scaling? */
1336	if ((tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE)) ==
1337	(TF_RCVD_SCALE\|TF_REQ_SCALE)) {
1338	tp->snd_scale = tp->requested_s_scale;
1339	tp->rcv_scale = tp->request_r_scale;
1340	}
1341	}
1342
1343	process_ACK:
1344	acked = ti->ti_ack - tp->snd_una;
1345	tcpstat.tcps_rcvackpack++;
1346	tcpstat.tcps_rcvackbyte += acked;
1347
1348	/*
1349	* If we have a timestamp reply, update smoothed
1350	* round trip time. If no timestamp is present but
1351	* transmit timer is running and timed sequence
1352	* number was acked, update smoothed round trip time.
1353	* Since we now have an rtt measurement, cancel the
1354	* timer backoff (cf., Phil Karn's retransmit alg.).
1355	* Recompute the initial retransmit timer.
1356	*/
1357	if (to.to_flag & TOF_TS)
1358	tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1);
1359	else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1360	tcp_xmit_timer(tp,tp->t_rtt);
1361
1362	/*
1363	* If all outstanding data is acked, stop retransmit
1364	* timer and remember to restart (more output or persist).
1365	* If there is more data to be acked, restart retransmit
1366	* timer, using current (possibly backed-off) value.
1367	*/
1368	if (ti->ti_ack == tp->snd_max) {
1369	tp->t_timer[TCPT_REXMT] = 0;
1370	needoutput = 1;
1371	} else if (tp->t_timer[TCPT_PERSIST] == 0)
1372	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1373
1374	/*
1375	* If no data (only SYN) was ACK'd,
1376	* skip rest of ACK processing.
1377	*/
1378	if (acked == 0)
1379	goto step6;
1380
1381	/*
1382	* When new data is acked, open the congestion window.
1383	* If the window gives us less than ssthresh packets
1384	* in flight, open exponentially (maxseg per packet).
1385	* Otherwise open linearly: maxseg per window
1386	* (maxseg^2 / cwnd per packet).
1387	*/
1388	{
1389	register u_int cw = tp->snd_cwnd;
1390	register u_int incr = tp->t_maxseg;
1391
1392	if (cw > tp->snd_ssthresh)
1393	incr = incr * incr / cw;
1394	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1395	}
1396	if (acked > so->so_snd.sb_cc) {
1397	tp->snd_wnd -= so->so_snd.sb_cc;
1398	sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1399	ourfinisacked = 1;
1400	} else {
1401	sbdrop(&so->so_snd, acked);
1402	tp->snd_wnd -= acked;
1403	ourfinisacked = 0;
1404	}
1405	if (so->so_snd.sb_flags & SB_NOTIFY)
1406	sowwakeup(so);
1407	tp->snd_una = ti->ti_ack;
1408	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1409	tp->snd_nxt = tp->snd_una;
1410
1411	switch (tp->t_state) {
1412
1413	/*
1414	* In FIN_WAIT_1 STATE in addition to the processing
1415	* for the ESTABLISHED state if our FIN is now acknowledged
1416	* then enter FIN_WAIT_2.
1417	*/
1418	case TCPS_FIN_WAIT_1:
1419	if (ourfinisacked) {
1420	/*
1421	* If we can't receive any more
1422	* data, then closing user can proceed.
1423	* Starting the timer is contrary to the
1424	* specification, but if we don't get a FIN
1425	* we'll hang forever.
1426	*/
1427	if (so->so_state & SS_CANTRCVMORE) {
1428	soisdisconnected(so);
1429	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1430	}
1431	tp->t_state = TCPS_FIN_WAIT_2;
1432	}
1433	break;
1434
1435	/*
1436	* In CLOSING STATE in addition to the processing for
1437	* the ESTABLISHED state if the ACK acknowledges our FIN
1438	* then enter the TIME-WAIT state, otherwise ignore
1439	* the segment.
1440	*/
1441	case TCPS_CLOSING:
1442	if (ourfinisacked) {
1443	tp->t_state = TCPS_TIME_WAIT;
1444	tcp_canceltimers(tp);
1445	/* Shorten TIME_WAIT [RFC-1644, p.28] */
1446	if (tp->cc_recv != 0 &&
1447	tp->t_duration < TCPTV_MSL)
1448	tp->t_timer[TCPT_2MSL] =
1449	tp->t_rxtcur * TCPTV_TWTRUNC;
1450	else
1451	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1452	soisdisconnected(so);
1453	}
1454	break;
1455
1456	/*
1457	* In LAST_ACK, we may still be waiting for data to drain
1458	* and/or to be acked, as well as for the ack of our FIN.
1459	* If our FIN is now acknowledged, delete the TCB,
1460	* enter the closed state and return.
1461	*/
1462	case TCPS_LAST_ACK:
1463	if (ourfinisacked) {
1464	tp = tcp_close(tp);
1465	goto drop;
1466	}
1467	break;
1468
1469	/*
1470	* In TIME_WAIT state the only thing that should arrive
1471	* is a retransmission of the remote FIN. Acknowledge
1472	* it and restart the finack timer.
1473	*/
1474	case TCPS_TIME_WAIT:
1475	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1476	goto dropafterack;
1477	}
1478	}
1479
1480	step6:
1481	/*
1482	* Update window information.
1483	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1484	*/
1485	if ((tiflags & TH_ACK) &&
1486	(SEQ_LT(tp->snd_wl1, ti->ti_seq) \|\|
1487	(tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) \|\|
1488	(tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) {
1489	/* keep track of pure window updates */
1490	if (ti->ti_len == 0 &&
1491	tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)
1492	tcpstat.tcps_rcvwinupd++;
1493	tp->snd_wnd = tiwin;
1494	tp->snd_wl1 = ti->ti_seq;
1495	tp->snd_wl2 = ti->ti_ack;
1496	if (tp->snd_wnd > tp->max_sndwnd)
1497	tp->max_sndwnd = tp->snd_wnd;
1498	needoutput = 1;
1499	}
1500
1501	/*
1502	* Process segments with URG.
1503	*/
1504	if ((tiflags & TH_URG) && ti->ti_urp &&
1505	TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1506	/*
1507	* This is a kludge, but if we receive and accept
1508	* random urgent pointers, we'll crash in
1509	* soreceive. It's hard to imagine someone
1510	* actually wanting to send this much urgent data.
1511	*/
1512	if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) {
1513	ti->ti_urp = 0; /* XXX */
1514	tiflags &= ~TH_URG; /* XXX */
1515	goto dodata; /* XXX */
1516	}
1517	/*
1518	* If this segment advances the known urgent pointer,
1519	* then mark the data stream. This should not happen
1520	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1521	* a FIN has been received from the remote side.
1522	* In these states we ignore the URG.
1523	*
1524	* According to RFC961 (Assigned Protocols),
1525	* the urgent pointer points to the last octet
1526	* of urgent data. We continue, however,
1527	* to consider it to indicate the first octet
1528	* of data past the urgent section as the original
1529	* spec states (in one of two places).
1530	*/
1531	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1532	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1533	so->so_oobmark = so->so_rcv.sb_cc +
1534	(tp->rcv_up - tp->rcv_nxt) - 1;
1535	if (so->so_oobmark == 0)
1536	so->so_state \|= SS_RCVATMARK;
1537	sohasoutofband(so);
1538	tp->t_oobflags &= ~(TCPOOB_HAVEDATA \| TCPOOB_HADDATA);
1539	}
1540	/*
1541	* Remove out of band data so doesn't get presented to user.
1542	* This can happen independent of advancing the URG pointer,
1543	* but if two URG's are pending at once, some out-of-band
1544	* data may creep in... ick.
1545	*/
1546	if (ti->ti_urp <= (u_long)ti->ti_len
1547	#ifdef SO_OOBINLINE
1548	&& (so->so_options & SO_OOBINLINE) == 0
1549	#endif
1550	)
1551	tcp_pulloutofband(so, ti, m);
1552	} else
1553	/*
1554	* If no out of band data is expected,
1555	* pull receive urgent pointer along
1556	* with the receive window.
1557	*/
1558	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1559	tp->rcv_up = tp->rcv_nxt;
1560	dodata: /* XXX */
1561
1562	/*
1563	* Process the segment text, merging it into the TCP sequencing queue,
1564	* and arranging for acknowledgment of receipt if necessary.
1565	* This process logically involves adjusting tp->rcv_wnd as data
1566	* is presented to the user (this happens in tcp_usrreq.c,
1567	* case PRU_RCVD). If a FIN has already been received on this
1568	* connection then we just ignore the text.
1569	*/
1570	if ((ti->ti_len \|\| (tiflags&TH_FIN)) &&
1571	TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1572	TCP_REASS(tp, ti, m, so, tiflags);
1573	/*
1574	* Note the amount of data that peer has sent into
1575	* our window, in order to estimate the sender's
1576	* buffer size.
1577	*/
1578	len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
1579	} else {
1580	m_freem(m);
1581	tiflags &= ~TH_FIN;
1582	}
1583
1584	/*
1585	* If FIN is received ACK the FIN and let the user know
1586	* that the connection is closing.
1587	*/
1588	if (tiflags & TH_FIN) {
1589	if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1590	socantrcvmore(so);
1591	/*
1592	* If connection is half-synchronized
1593	* (ie NEEDSYN flag on) then delay ACK,
1594	* so it may be piggybacked when SYN is sent.
1595	* Otherwise, since we received a FIN then no
1596	* more input can be expected, send ACK now.
1597	*/
1598	if (tp->t_flags & TF_NEEDSYN)
1599	tp->t_flags \|= TF_DELACK;
1600	else
1601	tp->t_flags \|= TF_ACKNOW;
1602	tp->rcv_nxt++;
1603	}
1604	switch (tp->t_state) {
1605
1606	/*
1607	* In SYN_RECEIVED and ESTABLISHED STATES
1608	* enter the CLOSE_WAIT state.
1609	*/
1610	case TCPS_SYN_RECEIVED:
1611	case TCPS_ESTABLISHED:
1612	tp->t_state = TCPS_CLOSE_WAIT;
1613	break;
1614
1615	/*
1616	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1617	* enter the CLOSING state.
1618	*/
1619	case TCPS_FIN_WAIT_1:
1620	tp->t_state = TCPS_CLOSING;
1621	break;
1622
1623	/*
1624	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1625	* starting the time-wait timer, turning off the other
1626	* standard timers.
1627	*/
1628	case TCPS_FIN_WAIT_2:
1629	tp->t_state = TCPS_TIME_WAIT;
1630	tcp_canceltimers(tp);
1631	/* Shorten TIME_WAIT [RFC-1644, p.28] */
1632	if (tp->cc_recv != 0 &&
1633	tp->t_duration < TCPTV_MSL) {
1634	tp->t_timer[TCPT_2MSL] =
1635	tp->t_rxtcur * TCPTV_TWTRUNC;
1636	/* For transaction client, force ACK now. */
1637	tp->t_flags \|= TF_ACKNOW;
1638	}
1639	else
1640	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1641	soisdisconnected(so);
1642	break;
1643
1644	/*
1645	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1646	*/
1647	case TCPS_TIME_WAIT:
1648	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1649	break;
1650	}
1651	}
1652	#ifdef TCPDEBUG
1653	if (so->so_options & SO_DEBUG)
1654	tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0);
1655	#endif
1656
1657	/*
1658	* Return any desired output.
1659	*/
1660	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1661	(void) tcp_output(tp);
1662	return;
1663
1664	dropafterack:
1665	/*
1666	* Generate an ACK dropping incoming segment if it occupies
1667	* sequence space, where the ACK reflects our state.
1668	*/
1669	if (tiflags & TH_RST)
1670	goto drop;
1671	#ifdef TCPDEBUG
1672	if (so->so_options & SO_DEBUG)
1673	tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1674	#endif
1675	m_freem(m);
1676	tp->t_flags \|= TF_ACKNOW;
1677	(void) tcp_output(tp);
1678	return;
1679
1680	dropwithreset:
1681	/*
1682	* Generate a RST, dropping incoming segment.
1683	* Make ACK acceptable to originator of segment.
1684	* Don't bother to respond if destination was broadcast/multicast.
1685	*/
1686	if ((tiflags & TH_RST) \|\| m->m_flags & (M_BCAST\|M_MCAST) \|\|
1687	IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
1688	goto drop;
1689	#ifdef TCPDEBUG
1690	if (tp == 0 \|\| (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1691	tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1692	#endif
1693	if (tiflags & TH_ACK)
1694	tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1695	else {
1696	if (tiflags & TH_SYN)
1697	ti->ti_len++;
1698	tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1699	TH_RST\|TH_ACK);
1700	}
1701	/* destroy temporarily created socket */
1702	if (dropsocket)
1703	(void) soabort(so);
1704	return;
1705
1706	drop:
1707	/*
1708	* Drop space held by incoming segment and return.
1709	*/
1710	#ifdef TCPDEBUG
1711	if (tp == 0 \|\| (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1712	tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1713	#endif
1714	m_freem(m);
1715	/* destroy temporarily created socket */
1716	if (dropsocket)
1717	(void) soabort(so);
1718	return;
1719	#ifndef TUBA_INCLUDE
1720	}
1721
1722	static void
1723	tcp_dooptions(tp, cp, cnt, ti, to)
1724	struct tcpcb *tp;
1725	u_char *cp;
1726	int cnt;
1727	struct tcpiphdr *ti;
1728	struct tcpopt *to;
1729	{
1730	u_short mss = 0;
1731	int opt, optlen;
1732
1733	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1734	opt = cp[0];
1735	if (opt == TCPOPT_EOL)
1736	break;
1737	if (opt == TCPOPT_NOP)
1738	optlen = 1;
1739	else {
1740	optlen = cp[1];
1741	if (optlen <= 0)
1742	break;
1743	}
1744	switch (opt) {
1745
1746	default:
1747	continue;
1748
1749	case TCPOPT_MAXSEG:
1750	if (optlen != TCPOLEN_MAXSEG)
1751	continue;
1752	if (!(ti->ti_flags & TH_SYN))
1753	continue;
1754	bcopy((char ) cp + 2, (char ) &mss, sizeof(mss));
1755	NTOHS(mss);
1756	break;
1757
1758	case TCPOPT_WINDOW:
1759	if (optlen != TCPOLEN_WINDOW)
1760	continue;
1761	if (!(ti->ti_flags & TH_SYN))
1762	continue;
1763	tp->t_flags \|= TF_RCVD_SCALE;
1764	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1765	break;
1766
1767	case TCPOPT_TIMESTAMP:
1768	if (optlen != TCPOLEN_TIMESTAMP)
1769	continue;
1770	to->to_flag \|= TOF_TS;
1771	bcopy((char *)cp + 2,
1772	(char *)&to->to_tsval, sizeof(to->to_tsval));
1773	NTOHL(to->to_tsval);
1774	bcopy((char *)cp + 6,
1775	(char *)&to->to_tsecr, sizeof(to->to_tsecr));
1776	NTOHL(to->to_tsecr);
1777
1778	/*
1779	* A timestamp received in a SYN makes
1780	* it ok to send timestamp requests and replies.
1781	*/
1782	if (ti->ti_flags & TH_SYN) {
1783	tp->t_flags \|= TF_RCVD_TSTMP;
1784	tp->ts_recent = to->to_tsval;
1785	tp->ts_recent_age = tcp_now;
1786	}
1787	break;
1788	case TCPOPT_CC:
1789	if (optlen != TCPOLEN_CC)
1790	continue;
1791	to->to_flag \|= TOF_CC;
1792	bcopy((char *)cp + 2,
1793	(char *)&to->to_cc, sizeof(to->to_cc));
1794	NTOHL(to->to_cc);
1795	/*
1796	* A CC or CC.new option received in a SYN makes
1797	* it ok to send CC in subsequent segments.
1798	*/
1799	if (ti->ti_flags & TH_SYN)
1800	tp->t_flags \|= TF_RCVD_CC;
1801	break;
1802	case TCPOPT_CCNEW:
1803	if (optlen != TCPOLEN_CC)
1804	continue;
1805	if (!(ti->ti_flags & TH_SYN))
1806	continue;
1807	to->to_flag \|= TOF_CCNEW;
1808	bcopy((char *)cp + 2,
1809	(char *)&to->to_cc, sizeof(to->to_cc));
1810	NTOHL(to->to_cc);
1811	/*
1812	* A CC or CC.new option received in a SYN makes
1813	* it ok to send CC in subsequent segments.
1814	*/
1815	tp->t_flags \|= TF_RCVD_CC;
1816	break;
1817	case TCPOPT_CCECHO:
1818	if (optlen != TCPOLEN_CC)
1819	continue;
1820	if (!(ti->ti_flags & TH_SYN))
1821	continue;
1822	to->to_flag \|= TOF_CCECHO;
1823	bcopy((char *)cp + 2,
1824	(char *)&to->to_ccecho, sizeof(to->to_ccecho));
1825	NTOHL(to->to_ccecho);
1826	break;
1827	}
1828	}
1829	if (ti->ti_flags & TH_SYN)
1830	tcp_mss(tp, mss); /* sets t_maxseg */
1831	}
1832
1833	/*
1834	* Pull out of band byte out of a segment so
1835	* it doesn't appear in the user's data queue.
1836	* It is still reflected in the segment length for
1837	* sequencing purposes.
1838	*/
1839	static void
1840	tcp_pulloutofband(so, ti, m)
1841	struct socket *so;
1842	struct tcpiphdr *ti;
1843	register struct mbuf *m;
1844	{
1845	int cnt = ti->ti_urp - 1;
1846
1847	while (cnt >= 0) {
1848	if (m->m_len > cnt) {
1849	char *cp = mtod(m, caddr_t) + cnt;
1850	struct tcpcb *tp = sototcpcb(so);
1851
1852	tp->t_iobc = *cp;
1853	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1854	bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
1855	m->m_len--;
1856	return;
1857	}
1858	cnt -= m->m_len;
1859	m = m->m_next;
1860	if (m == 0)
1861	break;
1862	}
1863	panic("tcp_pulloutofband");
1864	}
1865
1866	/*
1867	* Collect new round-trip time estimate
1868	* and update averages and current timeout.
1869	*/
1870	static void
1871	tcp_xmit_timer(tp, rtt)
1872	register struct tcpcb *tp;
1873	short rtt;
1874	{
1875	register int delta;
1876
1877	tcpstat.tcps_rttupdated++;
1878	tp->t_rttupdated++;
1879	if (tp->t_srtt != 0) {
1880	/*
1881	* srtt is stored as fixed point with 5 bits after the
1882	* binary point (i.e., scaled by 8). The following magic
1883	* is equivalent to the smoothing algorithm in rfc793 with
1884	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1885	* point). Adjust rtt to origin 0.
1886	*/
1887	delta = ((rtt - 1) << TCP_DELTA_SHIFT)
1888	- (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
1889
1890	if ((tp->t_srtt += delta) <= 0)
1891	tp->t_srtt = 1;
1892
1893	/*
1894	* We accumulate a smoothed rtt variance (actually, a
1895	* smoothed mean difference), then set the retransmit
1896	* timer to smoothed rtt + 4 times the smoothed variance.
1897	* rttvar is stored as fixed point with 4 bits after the
1898	* binary point (scaled by 16). The following is
1899	* equivalent to rfc793 smoothing with an alpha of .75
1900	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1901	* rfc793's wired-in beta.
1902	*/
1903	if (delta < 0)
1904	delta = -delta;
1905	delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
1906	if ((tp->t_rttvar += delta) <= 0)
1907	tp->t_rttvar = 1;
1908	} else {
1909	/*
1910	* No rtt measurement yet - use the unsmoothed rtt.
1911	* Set the variance to half the rtt (so our first
1912	* retransmit happens at 3*rtt).
1913	*/
1914	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1915	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1916	}
1917	tp->t_rtt = 0;
1918	tp->t_rxtshift = 0;
1919
1920	/*
1921	* the retransmit should happen at rtt + 4 * rttvar.
1922	* Because of the way we do the smoothing, srtt and rttvar
1923	* will each average +1/2 tick of bias. When we compute
1924	* the retransmit timer, we want 1/2 tick of rounding and
1925	* 1 extra tick because of +-1/2 tick uncertainty in the
1926	* firing of the timer. The bias will give us exactly the
1927	* 1.5 tick we need. But, because the bias is
1928	* statistical, we have to test that we don't drop below
1929	* the minimum feasible timer (which is 2 ticks).
1930	*/
1931	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1932	max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
1933
1934	/*
1935	* We received an ack for a packet that wasn't retransmitted;
1936	* it is probably safe to discard any error indications we've
1937	* received recently. This isn't quite right, but close enough
1938	* for now (a route might have failed after we sent a segment,
1939	* and the return path might not be symmetrical).
1940	*/
1941	tp->t_softerror = 0;
1942	}
1943
1944	/*
1945	* Determine a reasonable value for maxseg size.
1946	* If the route is known, check route for mtu.
1947	* If none, use an mss that can be handled on the outgoing
1948	* interface without forcing IP to fragment; if bigger than
1949	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1950	* to utilize large mbufs. If no route is found, route has no mtu,
1951	* or the destination isn't local, use a default, hopefully conservative
1952	* size (usually 512 or the default IP max size, but no more than the mtu
1953	* of the interface), as we can't discover anything about intervening
1954	* gateways or networks. We also initialize the congestion/slow start
1955	* window to be a single segment if the destination isn't local.
1956	* While looking at the routing entry, we also initialize other path-dependent
1957	* parameters from pre-set or cached values in the routing entry.
1958	*
1959	* Also take into account the space needed for options that we
1960	* send regularly. Make maxseg shorter by that amount to assure
1961	* that we can send maxseg amount of data even when the options
1962	* are present. Store the upper limit of the length of options plus
1963	* data in maxopd.
1964	*
1965	* NOTE that this routine is only called when we process an incoming
1966	* segment, for outgoing segments only tcp_mssopt is called.
1967	*
1968	* In case of T/TCP, we call this routine during implicit connection
1969	* setup as well (offer = -1), to initialize maxseg from the cached
1970	* MSS of our peer.
1971	*/
1972	void
1973	tcp_mss(tp, offer)
1974	struct tcpcb *tp;
1975	int offer;
1976	{
1977	register struct rtentry *rt;
1978	struct ifnet *ifp;
1979	register int rtt, mss;
1980	u_long bufsize;
1981	struct inpcb *inp;
1982	struct socket *so;
1983	struct rmxp_tao *taop;
1984	int origoffer = offer;
1985
1986	inp = tp->t_inpcb;
1987	if ((rt = tcp_rtlookup(inp)) == NULL) {
1988	tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
1989	return;
1990	}
1991	ifp = rt->rt_ifp;
1992	so = inp->inp_socket;
1993
1994	taop = rmx_taop(rt->rt_rmx);
1995	/*
1996	* Offer == -1 means that we didn't receive SYN yet,
1997	* use cached value in that case;
1998	*/
1999	if (offer == -1)
2000	offer = taop->tao_mssopt;
2001	/*
2002	* Offer == 0 means that there was no MSS on the SYN segment,
2003	* in this case we use tcp_mssdflt.
2004	*/
2005	if (offer == 0)
2006	offer = tcp_mssdflt;
2007	else
2008	/*
2009	* Sanity check: make sure that maxopd will be large
2010	* enough to allow some data on segments even is the
2011	* all the option space is used (40bytes). Otherwise
2012	* funny things may happen in tcp_output.
2013	*/
2014	offer = max(offer, 64);
2015	taop->tao_mssopt = offer;
2016
2017	/*
2018	* While we're here, check if there's an initial rtt
2019	* or rttvar. Convert from the route-table units
2020	* to scaled multiples of the slow timeout timer.
2021	*/
2022	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2023	/*
2024	* XXX the lock bit for RTT indicates that the value
2025	* is also a minimum value; this is subject to time.
2026	*/
2027	if (rt->rt_rmx.rmx_locks & RTV_RTT)
2028	tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
2029	tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
2030	tcpstat.tcps_usedrtt++;
2031	if (rt->rt_rmx.rmx_rttvar) {
2032	tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2033	(RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
2034	tcpstat.tcps_usedrttvar++;
2035	} else {
2036	/* default variation is +- 1 rtt */
2037	tp->t_rttvar =
2038	tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2039	}
2040	TCPT_RANGESET(tp->t_rxtcur,
2041	((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2042	tp->t_rttmin, TCPTV_REXMTMAX);
2043	}
2044	/*
2045	* if there's an mtu associated with the route, use it
2046	*/
2047	if (rt->rt_rmx.rmx_mtu)
2048	mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
2049	else
2050	{
2051	mss = ifp->if_mtu - sizeof(struct tcpiphdr);
2052	if (!in_localaddr(inp->inp_faddr))
2053	mss = min(mss, tcp_mssdflt);
2054	}
2055	mss = min(mss, offer);
2056	/*
2057	* maxopd stores the maximum length of data AND options
2058	* in a segment; maxseg is the amount of data in a normal
2059	* segment. We need to store this value (maxopd) apart
2060	* from maxseg, because now every segment carries options
2061	* and thus we normally have somewhat less data in segments.
2062	*/
2063	tp->t_maxopd = mss;
2064
2065	/*
2066	* In case of T/TCP, origoffer==-1 indicates, that no segments
2067	* were received yet. In this case we just guess, otherwise
2068	* we do the same as before T/TCP.
2069	*/
2070	if ((tp->t_flags & (TF_REQ_TSTMP\|TF_NOOPT)) == TF_REQ_TSTMP &&
2071	(origoffer == -1 \|\|
2072	(tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2073	mss -= TCPOLEN_TSTAMP_APPA;
2074	if ((tp->t_flags & (TF_REQ_CC\|TF_NOOPT)) == TF_REQ_CC &&
2075	(origoffer == -1 \|\|
2076	(tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2077	mss -= TCPOLEN_CC_APPA;
2078
2079	#if (MCLBYTES & (MCLBYTES - 1)) == 0
2080	if (mss > MCLBYTES)
2081	mss &= ~(MCLBYTES-1);
2082	#else
2083	if (mss > MCLBYTES)
2084	mss = mss / MCLBYTES * MCLBYTES;
2085	#endif
2086	/*
2087	* If there's a pipesize, change the socket buffer
2088	* to that size. Make the socket buffers an integral
2089	* number of mss units; if the mss is larger than
2090	* the socket buffer, decrease the mss.
2091	*/
2092	#ifdef RTV_SPIPE
2093	if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2094	#endif
2095	bufsize = so->so_snd.sb_hiwat;
2096	if (bufsize < mss)
2097	mss = bufsize;
2098	else {
2099	bufsize = roundup(bufsize, mss);
2100	if (bufsize > sb_max)
2101	bufsize = sb_max;
2102	(void)sbreserve(&so->so_snd, bufsize);
2103	}
2104	tp->t_maxseg = mss;
2105
2106	#ifdef RTV_RPIPE
2107	if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2108	#endif
2109	bufsize = so->so_rcv.sb_hiwat;
2110	if (bufsize > mss) {
2111	bufsize = roundup(bufsize, mss);
2112	if (bufsize > sb_max)
2113	bufsize = sb_max;
2114	(void)sbreserve(&so->so_rcv, bufsize);
2115	}
2116	/*
2117	* Don't force slow-start on local network.
2118	*/
2119	if (!in_localaddr(inp->inp_faddr))
2120	tp->snd_cwnd = mss;
2121
2122	if (rt->rt_rmx.rmx_ssthresh) {
2123	/*
2124	* There's some sort of gateway or interface
2125	* buffer limit on the path. Use this to set
2126	* the slow start threshhold, but set the
2127	* threshold to no less than 2*mss.
2128	*/
2129	tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2130	tcpstat.tcps_usedssthresh++;
2131	}
2132	}
2133
2134	/*
2135	* Determine the MSS option to send on an outgoing SYN.
2136	*/
2137	int
2138	tcp_mssopt(tp)
2139	struct tcpcb *tp;
2140	{
2141	struct rtentry *rt;
2142
2143	rt = tcp_rtlookup(tp->t_inpcb);
2144	if (rt == NULL)
2145	return tcp_mssdflt;
2146
2147	return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr);
2148	}
2149	#endif /* TUBA_INCLUDE */

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

Download in other formats: